Toric Varieties Cox PDF

Toric Varieties
David Cox
John Little
Hal Schenck
D EPARTMENT
01002
OF
M ATHEMATICS , A MHERST C OLLEGE , A MHERST, MA
E-mail address: dac@math.amherst.edu

D EPARTMENT OF M ATHEMATICS AND C OMPUTER S CIENCE , C OLLEGE
THE H OLY C ROSS , W ORCESTER , MA 01610
OF
E-mail address: little@mathcs.holycross.edu

D EPARTMENT OF M ATHEMATICS , U NIVERSITY
C HAMPAIGN , U RBANA , IL 61801
E-mail address: schenck@math.uiuc.edu
OF I LLINOIS AT
U RBANA -
c
2010,
David Cox, John Little and Hal Schenck
Contents
Preface
vii
Notation
xiii
Part I. Basic Theory of Toric Varieties
Chapter 1.
Affine Toric Varieties
1.0.
Background: Affine Varieties
1.1.
Introduction to Affine Toric Varieties
10
1.2.
Cones and Affine Toric Varieties
23
1.3.
Properties of Affine Toric Varieties
35
Appendix: Tensor Products of Coordinate Rings

Chapter 2.
Projective Toric Varieties
48
49
2.0.
Background: Projective Varieties
49
2.1.
Lattice Points and Projective Toric Varieties
54
2.2.
Lattice Points and Polytopes
62
2.3.
Polytopes and Projective Toric Varieties
74
2.4.
Properties of Projective Toric Varieties
86
Chapter 3.
Normal Toric Varieties
93
3.0.
Background: Abstract Varieties
93
3.1.
Fans and Normal Toric Varieties
105
3.2.
The Orbit-Cone Correspondence
114
3.3.
Toric Morphisms
125
3.4.
Complete and Proper
139
iii
Contents
iv
Appendix: Nonnormal Toric Varieties

Chapter 4.
Divisors on Toric Varieties
150
155
4.0.
Background: Valuations, Divisors and Sheaves
155
4.1.
Weil Divisors on Toric Varieties
170
4.2.
Cartier Divisors on Toric Varieties
176
4.3.
The Sheaf of a Torus-Invariant Divisor
189
Chapter 5.
Homogeneous Coordinates on Toric Varieties
195
5.0.
Background: Quotients in Algebraic Geometry
195
5.1.
Quotient Constructions of Toric Varieties
205
5.2.
The Total Coordinate Ring
218
5.3.
Sheaves on Toric Varieties
226
5.4.
Homogenization and Polytopes
231
Chapter 6.
Line Bundles on Toric Varieties
245
6.0.
Background: Sheaves and Line Bundles
245
6.1.
Ample and Basepoint Free Divisors on Complete Toric Varieties
262
6.2.
Polytopes and Projective Toric Varieties
277
6.3.
The Nef and Mori Cones
286
6.4.
The Simplicial Case
298
Appendix: Quasicoherent Sheaves on Toric Varieties

Chapter 7.
Projective Toric Morphisms
309
313
7.0.
Background: Quasiprojective Varieties and Projective Morphisms
313
7.1.
Polyhedra and Toric Varieties
318
7.2.
Projective Morphisms and Toric Varieties
328
7.3.
Projective Bundles and Toric Varieties
335
Appendix: More on Projective Morphisms

Chapter 8.
The Canonical Divisor of a Toric Variety
345
347
8.0.
Background: Reflexive Sheaves and Differential Forms
347
8.1.
One-Forms on Toric Varieties
358
8.2.
Differential Forms on Toric Varieties
365
8.3.
Fano Toric Varieties
379
Chapter 9.
Sheaf Cohomology of Toric Varieties
387
9.0.
Background: Sheaf Cohomology
387
9.1.
Cohomology of Toric Divisors
398
Contents
9.2.
Vanishing Theorems I
409
9.3.
Vanishing Theorems II
420
9.4.
Applications to Lattice Polytopes
430
9.5.
Local Cohomology and the Total Coordinate Ring
443
Part II. Topics in Toric Geometry
457
Chapter 10.
459
Toric Surfaces
10.1.
Singularities of Toric Surfaces and Their Resolutions
459
10.2.
467
10.3.
Continued Fractions and Toric Surfaces

Grobner Fans and McKay Correspondences
485
10.4.
Smooth Toric Surfaces
495
10.5.
Riemann-Roch and Lattice Polygons
502
Chapter 11.
Toric Resolutions and Toric Singularities
513
11.1.
Resolution of Singularities
513
11.2.
Other Types of Resolutions
525
11.3.
Rees Algebras and Multiplier Ideals
534
11.4.
Toric Singularities
546
Chapter 12.
The Topology of Toric Varieties
561
12.1.
The Fundamental Group
561
12.2.
The Moment Map
568
12.3.
Singular Cohomology of Toric Varieties
577
12.4.
The Cohomology Ring
592
12.5.
The Chow Ring and Intersection Cohomology
612
Chapter 13.
Toric Hirzebruch-Riemann-Roch
623
13.1.
Chern Characters, Todd Classes, and HRR
624
13.2.
Brions Equalities
632
13.3.
Toric Equivariant Riemann-Roch
641
13.4.
The Volume Polynomial
654
13.5.
The Khovanskii-Pukhlikov Theorem
663
Appendix: Generalized Gysin Maps

Chapter 14.
Toric GIT and the Secondary Fan
672
677
14.1.
Introduction to Toric GIT
677
14.2.
Toric GIT and Polyhedra
685
14.3.
Toric GIT and Gale Duality
699
Contents
vi
14.4.
The Secondary Fan
Chapter 15.
Geometry of the Secondary Fan
712
725
15.1.
The Nef and Moving Cones
725
15.2.
Gale Duality and Triangulations
734
15.3.
Crossing a Wall
747
15.4.
Extremal Contractions and Flips
762
15.5.
The Toric Minimal Model Program
772
Appendix A. The History of Toric Varieties
787
A.1.
The First Ten Years
787
A.2.
The Story Since 1980
794
Appendix B. Computational Methods
797
B.1.
The Rational Quartic
798
B.2.
Polyhedral Computations
799
B.3.
Normalization and Normaliz
803
B.4.
Sheaf Cohomology and Resolutions
805
B.5.
Sheaf Cohomology on the Hirzebruch Surface H2
806
B.6.
Resolving Singularities
808
B.7.
Intersection Theory and Hirzebruch-Riemann-Roch
809
B.8.
Anticanonical Embedding of a Fano Toric Variety
810
Appendix C. Spectral Sequences
811
C.1.
Definitions and Basic Properties
811
C.2.
Spectral Sequences Appearing in the Text
814
Bibliography
817
Index
831
Preface
The study of toric varieties is a wonderful part of algebraic geometry. There are
elegant theorems and deep connections with polytopes, polyhedra, combinatorics,
commutative algebra, symplectic geometry, and topology. Toric varieties also have
unexpected applications in areas as diverse as physics, coding theory, algebraic
statistics, and geometric modeling. Moreover, as noted by Fulton [105], toric
varieties have provided a remarkably fertile testing ground for general theories.
At the same time, the concreteness of toric varieties provides an excellent context
for someone encountering the powerful techniques of modern algebraic geometry
for the first time. Our book is an introduction to this rich subject that assumes only
a modest background yet leads to the frontier of this active area of research.
Brief Summary. The text covers standard material on toric varieties, including:
(a) Convex polyhedral cones, polytopes, and fans.
(b) Affine, projective, and abstract toric varieties.
(c) Complete toric varieties and proper toric morphisms.
(d) Weil and Cartier divisors on toric varieties.
(e) Cohomology of sheaves on toric varieties.
(f) The classical theory of toric surfaces.
(g) The topology of toric varieties.
(h) Intersection theory on toric varieties.
These topics are discussed in earlier texts on the subject, such as [93], [105] and
[218]. One difference is that we provide more details, with numerous examples,
figures, and exercises to illustrate the concepts being discussed. We also provide
background material when needed. In addition, we cover a large number of topics
previously available only in the research literature.
vii
viii
Preface
The Fifteen Chapters. To give you a better idea of what is in the book, we now
highlight a few topics from each chapter.
Chapters 1, 2 and 3 cover the basic material mentioned in items (a)(c) above.
The toric varieties encountered include:
The affine toric variety YA of a finite set A M Zn (Chapter 1).
The affine toric variety U of a polyhedral cone NR Rn (Chapter 1).
The projective toric variety XA of a finite set A M Zn (Chapter 2).
The projective toric variety XP of a lattice polytope P MR Rn (Chapter 2).

The abstract toric variety X of a fan in NR Rn (Chapter 3).
Chapter 4 introduces Weil and Cartier divisors on toric varieties. We compute

the class group and Picard group of a toric variety and define the sheaf OX (D)
associated to a Weil divisor D on a toric variety X .

Chapter 5 shows that the classical construction P n = Cn+1 \ {0} /C can be
generalized to any toric variety X . The homogeneous coordinate ring C[x0 , . . . , xn ]
of P n also has a toric generalization, called the total coordinate ring of X .
Chapter 6 relates Cartier divisors to invertible sheaves on X . We introduce
ample, basepoint free, and nef divisors and discuss their relation to convexity. The
stucture of the nef cone and its dual, the Mori cone, are described in detail, as is
the intersection pairing between divisors and curves.
Chapter 7 extends the relation between polytopes and projective toric varieties
to a relation between polyhedra and projective toric morphisms : X U . We
also discuss projective bundles over a toric variety and use these to classify smooth
projective toric varieties of Picard number 2.
Chapter 8 relates Weil divisors to reflexive sheaves of rank one and defines
Zariski p-forms. For p = dim X , this gives the canonical sheaf X and canonical
divisor KX . In the toric case we describe these explicitly and study the relation between reflexive polytopes and Gorenstein Fano toric varieties, meaning that KX
is ample. We find the 16 reflexive polygons in R2 (up to equivalence) and note the
relation |P M| + |P N| = 12 for a reflexive polygon P and its dual P .
Chapter 9 is about sheaf cohomology. We give two methods for computing
sheaf cohomology on a toric variety and prove a dizzying array of cohomology
vanishing theorems. Applications range from showing that normal toric varieties
are Cohen-Macaulay to the Dehn-Sommerville equations for a simple polytope and
counting lattice points in multiples of a polytope via the Ehrhart polynomial.
Chapter 10 studies toric surfaces, where we add a few twists to this classical
subject. After using Hirzebruch-Jung continued fractions to compute the minimal
resolution of a toric surface singularity, we discuss the toric meaning of ordinary
continued fractions. We then describe unexpected connections with Grobner fans
and the McKay correspondence. Finally, we use the Riemann-Roch theorem on a
Preface
ix
smooth complete toric surface to explain the mysterious appearance of the number
12 in Chapter 8 when counting lattice points in reflexive polygons.
Chapter 11 proves the existence of toric resolutions of singularities for toric
varieties of all dimensions. This is more complicated than for surfaces because of
the existence of toric flips and flops. We consider simple normal crossing, crepant,
log, and embedded resolutions and study how Rees algebras and multiplier ideals
can be applied in the resolution problem. We also discuss toric singularities and
show that a fan is simplicial if and only if X has at worst finite quotient singularities and hence is rationally smooth. We also explain what canonical and terminal
singularities mean in the toric context.
Chapters 12 and 13 describe the singular and equivariant cohomology of a
complete simplicial toric variety X and prove the Hirzebruch-Riemann-Roch and
equivariant Riemann-Roch theorems when X is smooth. We compute the fundamental group of X and study the moment map, with a brief mention of topological
models of toric varieties and connections with symplectic geometry. We describe
the Chow ring and intersection cohomology of a complete simplicial toric variety.
After proving Riemann-Roch, we give applications to the volume polynomial and
lattice point enumeration in polytopes.
Chapters 14 and 15 explore the rich connections that link geometric invariant
theory, the secondary fan, the nef and moving cones, Gale duality, triangulations,
wall crossings, flips, extremal contractions, and the toric minimal model program.
Appendices. The book ends with three appendices:
Appendix A: The History of Toric Varieties.
Appendix B: Computational Methods.
Appendix C: Spectral Sequences.
Appendix A surveys the history of toric geometry since its origins in the early
1970s. It is fun to see how the concepts and terminology evolved. Appendix B
discusses some of the software packages for toric geometry and gives examples
to illustrate what they can do. Appendix C gives a brief introduction to spectral
sequences and describes the spectral sequences used in Chapters 9 and 12.
Prerequisites. We assume that the reader is familiar with the material covered in
basic graduate courses in algebra and topology, and to a somewhat lesser degree,
complex analysis. In addition, we assume that the reader has had some previous
experience with algebraic geometry, at the level of any of the following texts:
Ideals, Varieties and Algorithms by Cox, Little and OShea [69].
Introduction to Algebraic Geometry by Hassett [133].
Elementary Algebraic Geometry by Hulek [151].
Undergraduate Algebraic Geometry by Reid [238].
Preface
Computational Algebraic Geometry by Schenck [246].
An Invitation to Algebraic Geometry by Smith, Kahanpaa , Kekalainen and

Traves [253].
Chapters 9 and 12 assume knowledge of some basic algebraic topology. The books
by Hatcher [135] and Munkres [210] are useful references here.
Readers who have studied more sophisticated algebraic geometry texts such as
Harris [130], Hartshorne [131], or Shafarevich [245] certainly have the background
needed to read our book. For readers with a more modest background, an important
prerequisite is a willingness to absorb a lot of algebraic geometry.
Background Sections. Since we do not assume a complete knowledge of algebraic
geometry, Chapters 19 each begin with a background section that introduces the
definitions and theorems from algebraic geometry that are needed to understand
the chapter. References where proofs can be found are provided. The remaining
chapters do not have background sections. For some of those chapters, no further
background is necessary, while for others, the material is more sophisticated and
the requisite background is given by careful references to the literature.
What Is Omitted. We work exclusively with varieties defined over the complex
numbers C. This means that we do not consider toric varieties over arbitrary fields
(see [92] for a treatment of this topic), nor do we consider toric stacks (see [39] for
an introduction). Moreover, our viewpoint is primarily algebro-geometric. Thus,
while we hint at some of the connections with symplectic geometry and topology in
Chapter 12, we do not do justice to this side of the story. Even within the algebraic
geometry of toric varieties, there are many topics we have had to omit, though we
provide some references that should help readers who want to explore these areas.
We have also omitted any discussion of how toric varieties are used in physics and
applied mathematics. Some pointers to the literature are given in our discussion of
the recent history of toric varieties in A.2 of Appendix A.
The Structure of the Text. We number theorems, propositions, and equations
based on the chapter and the section. Thus 3.2 refers to Section 2 of Chapter 3,
and Theorem 3.2.6, equation (3.2.6) and Exercise 3.2.6 all appear in this section.
Definitions, theorems, propositions, lemmas, remarks, and examples are numbered
together in one sequence within each section.
Some individual chapters have appendices. Within a chapter appendix the
same numbering system is used, except that the section number is a capital A.
This means that Theorem 3.A.3 is in the appendix to Chapter 3. On the other hand,
the three appendices at the end of the book are treated in the numbering system as
chapters A, B, and C. Thus Definition C.1.1 is in the first section of Appendix C.
The end (or absence) of a proof is indicated by , and the end of an example
is indicated by .
Preface
xi
For the Instructor. There is much more material here than you can cover in any
one-semester graduate course, probably more than you can cover in a full year
in most cases. So choices will be necessary depending on the background and
the interests of the student audience. We think it is reasonable to expect to cover
most of Chapters 16, 8 and 9 in a one-semester course where the students have
a minimal background in algebraic geometry. More material can be covered, of
course, if the students know more algebraic geometry. If time permits, you can
use toric surfaces (Chapter 10) to illustrate the power of the basic material and
introduce more advanced topics such as the resolution of singularities (Chapter 11)
and the Riemann-Roch theorem (Chapter 13).
Finally, we emphasize that the exercises are extremely important. We have
found that when the students work in groups and present their solutions, their engagement with the material increases. We encourage instructors to consider using
this strategy.
For the Student. The book assumes that you will be an active reader. This means
in particular that you should do tons of exercisesthis is the best way to learn
about toric varieties. If you have a modest background in algebraic geometry, then
reading the book requires a commitment to learn both toric varieties and algebraic
geometry. It will be a lot of work but is worth the effort. This is a great subject.
Send Us Feedback. We greatly appreciate hearing from instructors, students, or
general readers about what worked and what didnt. Please notify one or all of us
about any typographical or mathematical errors you might find.
Acknowledgements. We would like to thank to Dan Abramovich, Matt Baker,
Matthias Beck, Tom Braden, Volker Braun, Sandra Di Rocco, Dan Edidin, Matthias
Franz, Dan Grayson, Paul Hacking, Jurgen Hausen, Al Kasprzyk, Diane Maclagan,
Anvar Mavlyutov, Uwe Nagel, Andrey Novoseltsev, Sam Payne, Matthieu Rambaud, Raman Sanyal, Thorsten Rahn, Greg Smith, Bernd Sturmfels, Zach Teitler,
Michele Vergne, Mark Walker, Gunter Ziegler, and Dylan Zwick for many helpful
conversations and emails as we worked on the book.
We would also like to thank all the participants of the 2009 MSRI Summer
Graduate Student Workshop on Toric Varieties and especially Dustin Cartwright
and Daniel Erman, our able assistants.
Finally we thank Ina Mette for her support and advice and Barbara Beeton for
her help with LATEX.
September 2010
David Cox
John Little
Hal Schenck
Notation
The notation used in the book is organized by topic. The number in parentheses at
the end of an entry indicates the chapter in which the notation first appears.
Basic Sets
Z, Q, R, C
N
integers, rational numbers, real numbers, complex numbers

semigroup of nonnegative integers {0, 1, 2, . . . }
The Torus
C
(C )n
M, m
N, u
TN
MR , MQ
NR , NQ
hm, ui
multiplicative group of nonzero complex numbers C \ {0}

standard n-dimensional torus
character lattice of a torus and character of m M
lattice of one-parameter subgroups of a torus and
one-parameter subgroup of u N
torus N Z C = HomZ (M, C ) associated to N and M
vector spaces M Z R, M Z Q built from M
vector spaces N Z R, N Z Q built from N
pairing of m M or MR with u N or NR
(1)
(1)
(1)
(1)
(1)
(1)
(1)
(1)
Hyperplanes and Half-Spaces

Hm
Hm+
Hu,b
+
Hu,b
hyperplane in NR defined by hm, i = 0, m MR \ {0}

half-space in NR defined by hm, i 0, m MR \ {0}
hyperplane in MR defined by h, ui = b, u NR \ {0}
half-space in MR defined by h, ui b, u NR \ {0}
(1)
(1)
(2)
(2)
xiii
Notation
xiv
Cones
Cone(S)
Span()
dim
Relint()
Int()
,
convex cone generated by S

rational convex polyhedral cone in NR
subspace spanned by
dimension of
dual cone of
relative interior of
interior of when Span() = NR
set of m MR with hm, i = 0
is a face or proper face of
face of dual to , equals
(1)
(1)
(1)
(1)
(1)
(1)
(1)
(1)
(1)
(1)
1-dimensional strongly convex cone (a ray) in NR

minimal generator of N, a rational ray in NR
rays of a strongly convex cone in NR
(1)
(1)
(1)
lattice generated by A
P
P
elements si=1 ai mi ZA with si=1 ai = 0
sublattice Z( N) = Span() N
quotient lattice N/N
dual lattice of N(), equals M
(1)
(2)
(3)
(3)
(3)
Rays
u
(1)
Lattices
ZA
ZA
N
N()
M()
Fans
(r)
max
Star()
()
(v)
fan in NR
r-dimensional cones of
maximal cones of
star of , a fan in N()
star subdivision of for
star subdivision of for v || N primitive
(2,3)
(3)
(3)
(3)
(3)
(11)
Polytopes and Polyhedra

n
Conv(S)
dim P
Q P, Q P
P
A+B
kP
standard n-simplex in Rn
convex hull of S
dimension of a polyhedron P
Q is a face or proper face of P
dual or polar of a polytope
Minkowski sum
multiple of a polytope or polyhedron
(2)
(1)
(2)
(2)
(2)
(2)
(2)
Notation
xv
Cones Built From Polyhedra

Cv
Q
P
C(P)
SP
Cone(P v) for a vertex v of a polytope or polyhedron

cone of a face Q P in the normal fan P
normal fan of a polytope or polyhedron P
cone over a polytope or polyhedron
semigroup algebra of C(P) (M Z)
(2)
(2)
(2)
(1)
(7)
Combinatorics and Lattice Points of Polytopes

fi
hp
L(P)
L (P)
EhrP ()
EhrPp ()
number of i-dimesional faces of P

Pn
ip i
i=p (1)
p f i , equals Betti number b2p (XP ) when P simple
number of lattice points of a lattice polytope
number of interior lattice points of a lattice polytope
Ehrhart polynomial of a lattice polytope
p-Ehrhart polynomial of a lattice polytope
(9)
(9)
(9)
(9)
(9)
(9)
affine semigroup and its semigroup algebra

affine semigroup generated by A
affine semigroup M
Hilbert basis of S
(1)
(1)
(1)
(1)
Semigroups
S, C[S]
NA
S = S,N
H
Rings
R f , RS , Rp
R
Rb
R C S
RG
R[a]
R[]
localization of R at f, a multiplicative set S, a prime ideal p

integral closure of the integral domain R
completion of local ring R
tensor product of rings over C
ring of invariants of G acting on R
Rees algebra of an ideal a R
Veronese subring of a graded ring R
(1)
(1)
(1)
(1)
(1,5)
(11)
(14)
polynomial ring in n variables

formal power series ring in n variables
ring of Laurent polynomials
ideal of an affine or projective variety
coordinate ring of an affine or projective variety
graded piece in degree d when V is projective
field of rational functions when V is irreducible
local ring of a variety at a point and its maximal ideal
(1)
(1)
(1)
(1,2)
(1,2)
(2)
(1)
(1)
Specific Rings
C[x1 , . . . , xn ]
C[[x1 , . . . , xn ]]
C[x11 , . . . , xn1 ]
I(V )
C[V ]
C[V ]d
C(V )
OV ,p , mV ,p
Notation
xvi
Varieties
V(I)
Vf
S
Tp (X)
dim X, dim p X
Spec(R)
Proj(S)
X Y
X S Y
Xb
affine or projective variety of an ideal

subset of an affine variety V where f 6= 0
Zariski closure of S in a variety
Zariski tangent space of a variety at a point
dimension of a variety and dimension at a point
affine variety of coordinate ring R
projective variety of graded ring S
product of varieties
fiber product of varieties
affine cone of a projective variety X
(1,2)
(1)
(1,3)
(1,3)
(1,3)
(1)
(7)
(1,3)
(3)
(2)
YA , XA
U = U,N
X = X,N
XP
O()
V () = O()
UP
U
affine and projective toric variety of A M

affine toric variety of a cone NR
toric variety of a fan in NR
projective toric variety of a lattice polytope or polyhedron
lattice homomorphism of a toric morphism : X1 X2
real extension of
distinguished point of U
torus orbit corresponding to
closure of orbit of , toric variety of Star()
affine toric variety of recession cone of a polyhedron
affine toric variety of a fan with convex support
(1,2)
(1)
(3)
(2.7)
(1,3)
(1)
(3)
(3)
(3)
(7)
(7)
Toric Varieties
Specific Varieties
Cn , P n
P(q0 , . . . , qn )
Cbd , Cd
Bl0 (Cn )
BlV () (X )
Hr
Sa,b
affine and projective n-dimensional space

weighted projective space
(1,2)
(2)
rational normal cone and curve

blowup of Cn at the origin
blowup of X along V (), toric variety of ()
Hirzebruch surface
rational normal scroll
(1,2)
(3)
(3)
(3)
(3)
Total Coordinate Ring

S
x
S
deg(x )
total coordinate ring of X

variable in S corresponding to (1)
graded piece of S in degree Cl(X )
degree in Cl(X ) of a monomial in S
(5)
(5)
(5)
(5)
Notation
x
B()
xhmi
xhm,Di
xF
xhm,Pi
xhv,Pi
M
M()
xvii
Q
x for
monomial (1)
/
irrelevant ideal of S, generated by the x
Q hm,u i
Laurent monomial x , m M
homogenization of m , m PD M
facet variable of a facet F P
P-monomial associated to m P M
vertex monomial associated to vertex v P M
graded S-module
shift of M by Cl(X )
(5)
(5)
(5)
(5)
(5)
(5)
(5)
(5)
(5)
Quotient Construction
X/G
X//G
Z()
G
good geometric quotient

good categorical quotient
exceptional set in quotient construction, equals V(B())
group in quotient construction, equals HomZ (Cl(X ), C )
(5)
(5)
(5)
(5)
local ring of a variety at a prime divisor

discrete valuation of a prime divisor D
principal divisor of a rational function
linear equivalence of divisors
effective divisor
group of principal divisors on X
group of Weil divisors on X
group of Cartier divisors on X
divisor class group of a normal variety X
Picard group of a normal variety X
Pic(X) Z R
support of a divisor
restriction of a divisor to an open set
local data of a Cartier divisor on X
complete linear system of D
round down and round up of a Q-divisor
(4)
(4)
(4)
(4)
(4)
(4)
(4)
(4)
(4)
(4)
(6)
(4)
(4)
(4)
(6)
(9)
Divisors
OX ,D
D
div( f )
DE
D0
Div0 (X)
Div(X)
CDiv(X)
Cl(X)
Pic(X)
Pic(X)R
Supp(D)
D|U
{(Ui , fi )}
|D|
D, D
Torus-Invariant Divisors
D = O()
DF
DivTN (X )
CDivTN (X )
{m }
torus-invariant prime divisor on X of ray (1)

torus-invariant prime divisor on XP of facet F P
group of torus-invariant Weil divisors on X
group of torus-invariant Cartier divisors on X
Cartier data of a torus-invariant Cartier divisor on X
(4)
(4)
(4)
(4)
(4)
Notation
xviii
DP
PD
XD
D
D
Cartier divisor of a polytope or polyhedron

polyhedron of a torus-invariant divisor
toric variety of a basepoint free divisor
fan of XD
pullback of a Cartier divisor
(4,7)
(4)
(6)
(6)
(6)
Support Functions
D
P
SF()
SF(, N)
support function of a Cartier divisor

support function of a polytope or polyhedron
support functions for
support functions for integral with respect to N
(4)
(4)
(4)
(4)
sections of a sheaf over an open set

restriction of a sheaf to an open set
stalk of a sheaf at a point
tensor product of sheaves of OX -modules
sheaf of homomorphisms
dual sheaf of F , equals H omOX (F , OX )
direct image sheaf
(4)
(4)
(6)
(6)
(6)
(6)
structure sheaf of a variety X

sheaf of invertible elements of OX
constant sheaf of rational functions for X irreducible
sheaf of a Weil divisor D on X
ideal sheaf of a subvariety Y X
sheaf on Spec(R) of an R-module M
(3)
(4)
(6)
(4)
(3)
(4)
sheaf on X of the graded S-modules M

sheaf of the S-module S()
(5)
(5)
Sheaves
(U, F )
F |U
Fp
F OX G
H omOX (F , G )
F
f F
Specific Sheaves
OX
OX
KX
OX (D)
IY
e
M
e
M
OX ()
Vector Bundles and Locally Free Sheaves

L, E
:V X
: VL X
f L
L ,W
P(V ), P(E )
D
invertible sheaf (line bundle) and locally free sheaf

vector bundle
rank 1 vector bundle of an invertible sheaf L
pullback of an invertible sheaf
map to projective space determined by W (X, L )
projective bundle of vector bundle or locally free sheaf
fan for rank 1 vector bundle VL for L = OX (D)
(6)
(6)
(6)
(6)
(6)
(7)
(7)
Notation
xix
Intersection Theory
deg(D)
D C
D D, C C
N 1 (X), N1 (X)
Nef(X)
Mov(X)
Eff(X)
NE(X)
NE(X)
degree of a divisor on a smooth complete curve

intersection product of Cartier divisor and complete curve
numerically equivalent Cartier divisors and complete curves
(CDiv(X)/ ) Z R and ({proper 1-cycles on X}/ ) Z R
cone in N 1 (X) generated by nef divisors
moving cone of a variety X in N 1 (X)
pseudoeffective cone of a variety X in N 1 (X)
cone in N1 (X) generated by complete curves
Mori cone, equals the closure of NE(X)
(6)
(6)
(6)
(6)
(6)
(15)
(15)
(6)
(6)
Differential Forms and Sheaves

R/C
1X , TX
IY /IY2 , NY /X
bp
p,
X
KX , X
1X (log D)
module of Kahler differentials of a C-algebra R

cotangent and tangent sheaves of a variety X
conormal and normal sheaves of Y X
sheaves of p-forms and Zariski p-forms on X

b n , n = dim X
canonical divisor and canonical sheaf
X
sheaf of 1-forms with logarithmic poles on D
(8)
(8)
(8)
(8)
(8)
(8)
Sheaf Cohomology
H 0 (X, F )
H p (X, F )
R p f F
p
(G , F )
ExtO
X
C (U , F )
(F )
global sections (X, F ) of a sheaf F on X

p-th sheaf cohomology group of a sheaf F on X
higher direct image sheaf
Ext groups of sheaves of OX -modules G , F
Cech
complex for sheaf cohomology
P
Euler characteristic of F , equals p (1) p dim H p (X, F )
(9)
(9)
(9)
(9)
(9)
(9)
graded piece of sheaf cohomology of L = OX (D) for m M

subsets of || used to compute H p (X , L )m
(9)
(9)
Sheaf Cohomology of a Toric Variety

H p (X , L )m
supp
VD,m , VD,m
Local Cohomology
HIp (M)
C (f, M)
ExtRp (N, M)
p-th local cohomology of an R-module M for the ideal I R
Cech
complex for local cohomology when I = hfi
Ext groups of R-modules N, M
(9)
(9)
(9)
Resolution of Singularities
Xsing
Exc()
singular locus of a variety

exceptional locus of a resolution of singularities
(11)
(11)
Notation
xx
J (c I )
(X, D)
multiplier ideal sheaf

P
log pair, D = i ai Di , ai [0, 1] Q
(11)
(11)
Singularities of Toric Varieties

mult()
P
can
multiplicity of a simplicial cone, equals [N : Zu1 + + Zud ] (6,11)

P
parallelotope of a simplicial cone, equals { i i ui | 0 i < 1} (11)
polytope related to canonical and terminal singularities of U
(11)
the polyhedron Conv( N \ {0})
(10, 11)
fan over bounded faces of , reduces to canonical singularities (11)
Topology of a Toric Variety

N
1 (X )
SN
(X )0
f,
sublattice of N generated by || N
fundamental group of X , isomorphic to N/N
real torus N Z S1 = HomZ (M, S1 ) (S1 )n
nonnegative real points of a toric variety
algebraic and symplectic moment maps XP MR
symplectic moment map C(1) Cl(X )R
(12)
(12)
(12)
(12)
(12)
(12)
Singular Homology and Cohomology

H i (X, R)
e i (X, R)
H
Hci (X, R)
Hi (X, R)
HiBM (X, R)
bi (X)
e(X)
`, a
H (X, R)
[W ]
[W ]r
f
R!
ith singular cohomology of X with coefficients in a ring R

(9)
i-th reduced cohomology of X
(9)
ith cohomology of X with compact supports
(12)
ith singular homology of X
(12)
ith Borel-Moore homology of X
(13)
ith Betti number of X, equals dim Hi (X, Q)
(12)
P
(9,10,12)
Euler characteristic of X, equals i (1)i bi (X)
cap and cup products
(12)
L p
(12)
cohomology ring p H (X, R) under cup product
cohomology class of a subvariety W in H 2n2k (X, Q)
(12,13)
refined cohomology class of W in H 2n2k (X, X \ W , Q)
(12,13)
generalized Gysin map
(13)
R
integral X : H (X, Q) Q, equals Gysin map of X {pt} (12,13)
EG
BG
EG G X
HG (X, R)
G , (G )Q
XG
a contractible space on which G acts freely

the quotient EG/G
quotient of EG X modulo relation (e g, x) (e, g x)
equivariant cohomology ring, equals H (EG G X, R)
integral and rational equivariant cohomology ring of a point
fixed point set for action of G on X
Equivariant Cohomology for a Group Action
(12)
(12)
(12)
(12)
(12)
(12)
Notation
xxi
Equivariant Cohomology for a Torus Action

SymZ (M)
SymQ (M)
s
[D]T
R
X eq
bT (X, Q)
H
b
symmetric algebra of M over Z

rational symmetric algebra on M, equals SymZ (M) Z Q
isomorphism s : SymQ (M) (T )Q
equivariant cohomology class of a T -invariant divisor D
R
equivariant integral X eq : HT (X, Q) (T )Q
Q
completion k=0 HTk (X, Q) of equivariant cohomology of X
completion of the equivariant cohomology of a point
(12)
(12)
(12)
(12)
(13)
(13)
(13)
Chow Groups and the Chow Ring

Ak (X)
Ak (X)
A (X)
A (X)Q
Chow group of k-cycles modulo rational equivalence

codimension k cycles modulo rational equivalence
integral Chow ring of X smooth and complete
rational Chow ring of X quasismooth and complete
(12)
(12)
(12)
(12)
Intersection Cohomology
IHip (X)
IH i (X)
IH i (X)Q
ith intersection homology of X for perversity p

ith intersection cohomology of X for middle perversity
ith rational intersection cohomology of X
(12)
(12)
(12)
Cohomology Ring of a Complete Simplicial Toric Variety

I
J
RQ ()
SRQ ()
Stanley-Reisner ideal of the fan , ideal in Q[x1 , . . . , xr ]

Pr
ideal h i=1 hm, ui ixi | m Mi Q[x1 , . . . , xr ]
Jurkiewicz-Danilov ring Q[x1 , . . . , xr ]/(I + J ) H (X , Q)
Stanley-Reisner ring Q[x1 , . . . , xr ]/I HT (X , Q)
(12)
(12)
(12)
(12)
Hirzebruch-Riemann-Roch
ci (E )
ch(L )
Td(X)
Bk
ci = ci (TX )
Ti
K(X)
T (L )
T (L )
chT(L )
Td T(X)
Todd(x)
ith Chern class of a locally free sheaf E

Chern character of a line bundle L
Todd class of the variety X
kth Bernoulli number
ith Chern class of the tangent bundle
ith Todd polynomial in the ci
Grothendieck group of classes of coherent sheaves on X
equivariant Euler characteristic
local contribution of (n) to T (L )
equivariant Chern character of L
equivariant Todd class of X
formal Todd differential operator for the variable x
(13)
(13)
(13)
(13)
(13)
(13)
(13)
(13)
(13)
(13)
(13)
(13)
Notation
xxii
Brions Equalities
Z[M]
Z[[M]]
Z[[M]]Sum
S( f )
e(L )
integral semigroup algebra of M

P
formal semigroup module of M, formal sums mM am m
summable elements in Z[[M]]
sum of an element f Z[[M]]Sum
P
Pn
i
m
mM
i=0 dim H (X, L )m Z[[M]]
(13)
(13)
(13)
(13)
(13)
b
G
character group of algebraic subgroup G (C )r
(14)
Geometric Invariant Theory
L
r s
(Cr )ss
, (C )
R
Cr // G
B()
Z()
P , Pa
Fi, , Fi,a
b
sheaf of sections of rank 1 line bundle on C for character G
semistable and stable points for
L
r
G
graded ring
d=0 (C , L d )
r
GIT quotient of C by G for , equals Proj(R ) = (Cr )ss
//G
irrelevant ideal of
exceptional set of , equals V(B())
polyhedra in Rr and MR for = a
ith virtual facet of P , Pa
r
(14)
(14)
(14)
(14)
(14)
(14)
(14)
(14)
The Secondary Fan

,
C , C
e ,I , ,I
B(, I )
GKZ
MovGKZ
PGKZ
bR and NR
lists of r vectors in G
cones generated by and
GKZ cones determined by , I

irrelevant ideal determined by , I
secondary fan of
moving cone of the secondary fan
secondary polytope, normal fan is GKZ
(14)
(14)
(14)
(14)
(14)
(15)
(15)
Toric Minimal Model Program

R
DR
f D
J , J+
, +
, +
extremal ray of the Mori cone

intersection product D C for [C] R \ {0}
birational transform of a divisor by a birational map
index sets determined by a wall relation
fans determined by a wall relation
toric morphisms determined by a wall relation
(15)
(15)
(15)
(15)
(15)
(15)
multiplicative group of dth roots of unity in C

ordinary continued fraction of a rational number
Hirzebruch-Jung continued fraction of a rational number
(1)
(10)
(10)
Miscellaneous
d
[a1 , . . . , as ]
[[b1 , . . . , br ]]
Part I. Basic Theory of

Toric Varieties
Chapters 1 to 9 introduce the theory of toric varieties. This part of the

book assumes only a minimal amount of algebraic geometry, at the level
of Ideals, Varieties and Algorithms [69]. Each chapter begins with a background section that develops the necessary algebraic geometry.
Chapter 1
Affine Toric Varieties
1.0. Background: Affine Varieties

We begin with the algebraic geometry needed for our study of affine toric varieties.
Our discussion assumes Chapters 15 and 9 of [69].
Coordinate Rings. An ideal I S = C[x1 , . . . , xn ] gives an affine variety
V(I) = {p Cn | f (p) = 0 for all f I}
and an affine variety V Cn gives the ideal
I(V ) = { f S | f (p) = 0 for all p V }.
By the Hilbert basis theorem, an affine variety V is defined by the vanishing of
finitely many
polynomials in S, and for any ideal I, the Nullstellensatz tells us that
I(V(I)) = I = { f S | f I for some 1} since C is algebraically closed.
The most important algebraic object associated to V is its coordinate ring
C[V ] = S/I(V ).
Elements of C[V ] can be interpreted as the C-valued polynomial functions on V .
Note that C[V ] is a C-algebra, meaning that its vector space structure is compatible
with its ring structure. Here are some basic facts about coordinate rings:
C[V ] is an integral domain I(V ) is a prime ideal V is irreducible.
Polynomial maps (also called morphisms) : V1 V2 between affine varieties

correspond to C-algebra homomorphisms : C[V2 ] C[V1 ], where (g) =
g for g C[V2 ].
Two affine varieties are isomorphic if and only if their coordinate rings are
isomorphic C-algebras.
Chapter 1. Affine Toric Varieties
A point p of an affine variety V gives the maximal ideal

{ f C[V ] | f (p) = 0} C[V ],
and all maximal ideals of C[V ] arise this way.
Coordinate rings of affine varieties can be characterized as follows (Exercise 1.0.1).

Lemma 1.0.1. A C-algebra R is isomorphic to the coordinate ring of an affine
variety if and only if R is a finitely generated C-algebra with no nonzero nilpotents,
i.e., if f R satisfies f = 0 for some 1, then f = 0.

To emphasize the close relation between V and C[V ], we sometimes write
(1.0.1)
V = Spec(C[V ]).
This can be made canonical by identifying V with the set of maximal ideals of
C[V ] via the fourth bullet above. More generally, one can take any commutative
ring R and define the affine scheme Spec(R). The general definition of Spec uses
all prime ideals of R, not just the maximal ideals as we have done. Thus some
authors would write (1.0.1) as V = Specm(C[V ]), the maximal spectrum of C[V ].
Readers wishing to learn about affine schemes should consult [90] and [131].
The Zariski Topology. An affine variety V Cn has two topologies we will use.
The first is the classical topology, induced from the usual topology on Cn . The
second is the Zariski topology, where the Zariski closed sets are subvarieties of V
(meaning affine varieties of Cn contained in V ) and the Zariski open sets are their
complements. Since subvarieties are closed in the classical topology (polynomials
are continuous), Zariski open subsets are open in the classical topology.
Given a subset S V , its closure S in the Zariski topology is the smallest
subvariety of V containing S. We call S the Zariski closure of S. It is easy to give
examples where this differs from the closure in the classical topology.
Affine Open Subsets and Localization. Some Zariski open subsets of an affine
variety V are themselves affine varieties. Given f C[V ] \ {0}, let
V f = {p V | f (p) 6= 0} V .
Then V f is Zariski open in V and is also an affine variety, as we now explain.

Let V Cn have I(V ) = h f1 , . . . , fs i and pick g C[x1 , . . . , xn ] representing f .
Then V f = V \ V(g) is Zariski open in V . Now consider a new variable y and let
W = V( f1 , . . . , fs , 1 gy) Cn C. Since the projection map Cn C Cn maps
W bijectively onto V f , we can identify V f with the affine variety W Cn C.
When V is irreducible, the coordinate ring of V f is easy to describe. Let C(V )

be the field of fractions of the integral domain C[V ]. Recall that elements of C(V )
give rational functions on V . Then let
(1.0.2)
C[V ] f = {g/ f C(V ) | g C[V ], 0}.
In Exercise 1.0.3 you will prove that Spec(C[V ] f ) is the affine variety V f .
Example 1.0.2. The n-dimensional torus is the affine open subset
(C )n = Cn \ V(x1 xn ) Cn ,
with coordinate ring
C[x1 , . . . , xn ]x1 xn = C[x11 , . . . , xn1 ].
Elements of this ring are called Laurent polynomials.
The ring C[V ] f from (1.0.2) is an example of localization. In Exercises 1.0.2

and 1.0.3 you will show how to construct this ring for all affine varieties, not just
irreducible ones. The general concept of localization is discussed in standard texts
in commutative algebra such as [10, Ch. 3] and [89, Ch. 2].
Normal Affine Varieties. Let R be an integral domain with field of fractions K.
Then R is normal, or integrally closed, if every element of K which is integral over
R (meaning that it is a root of a monic polynomial in R[x]) actually lies in R. For
example, any UFD is normal (Exercise 1.0.5).
Definition 1.0.3. An irreducible affine variety V is normal if its coordinate ring
C[V ] is normal.
For example, Cn is normal since its coordinate ring C[x1 , . . . , xn ] is a UFD and
hence normal. Here is an example of a nonnormal affine variety.
Example 1.0.4. Let C = V(x 3 y 2 ) C2 . This is an irreducible plane curve with a
cusp at the origin. It is easy to see that C[C] = C[x, y]/hx 3 y 2 i. Now let x and y be
the cosets of x and y in C[C] respectively. This gives y/
x C(C). A computation
shows that y/
x
/ C[C] and that (
y/
x)2 = x. Consequently C[C] and hence C are not
normal. We will see below that C is an affine toric variety.
An irreducible affine variety V has a normalization defined as follows. Let

C[V ] = { C(V ) : is integral over C[V ]}.
We call C[V ] the integral closure of C[V ]. One can show that C[V ] is normal and
(with more work) finitely generated as a C-algebra (see [89, Cor. 13.13]). This
gives the normal affine variety
V = Spec(C[V ] )
We call V the normalization of V . The natural inclusion C[V ] C[V ] = C[V ]
corresponds to a map V V . This is the normalization map.
Example 1.0.5. We saw in Example 1.0.4 that the curve C C2 defined by x 3 = y 2

has elements x, y C[C] such that y/
x
/ C[C] is integral over C[C]. In Exercise 1.0.6 you will show that C[
y/
x] C(C) is the integral closure of C[C] and
that the normalization map is the map C C defined by t 7 (t 2 ,t 3 ).
At first glance, the definition of normal does not seem very intuitive. Once we
enter the world of toric varieties, however, we will see that normality has a very
nice combinatorial interpretation and that the nicest toric varieties are the normal
ones. We will also see that normality leads to a nice theory of divisors.
In Exercise 1.0.7 you will prove some properties of normal domains that will
be used in 1.3 when we study normal affine toric varieties.
Smooth Points of Affine Varieties. In order to define a smooth point of an affine
variety V , we first need to define local rings and Zariski tangent spaces. When V
is irreducible, the local ring of V at p is
OV ,p = { f /g C(V ) | f , g C[V ] and g(p) 6= 0}.
Thus OV ,p consists of all rational functions on V that are defined at p. Inside of
OV ,p we have the maximal ideal
mV ,p = { OV ,p | (p) = 0}.
In fact, mV ,p is the unique maximal ideal of OV ,p , so that OV ,p is a local ring.
Exercises 1.0.2 and 1.0.4 explain how to define OV ,p when V is not irreducible.
The Zariski tangent space of V at p is defined to be
Tp (V ) = HomC (mV ,p /mV2 ,p , C).
In Exercise 1.0.8 you will verify that dim Tp (Cn ) = n for every p Cn . According
to [131, p. 32], we can compute the Zariski tangent space of a point in an affine
variety as follows.
Lemma 1.0.6. Let V Cn be an affine variety and let p V . Also assume that
I(V ) = h f1 , . . . , fs i C[x1 , . . . , xn ]. For each i, let
d p ( fi ) =
fi
fi
(p) x1 + +
(p) xn .
x1
xn
Then the Zariski tangent space Tp (V ) is isomorphic to the subspace of Cn defined

by the equations d p ( f1 ) = = d p ( fs ) = 0. In particular, dim Tp (V ) n.

Definition 1.0.7. A point p of an affine variety V is smooth or nonsingular if
dim Tp (V ) = dim p V , where dim p V is the maximum of the dimensions of the irreducible components of V containing p. The point p is singular if it is not smooth.
Finally, V is smooth if every point of V is smooth.
Points lying in the intersection of two or more irreducible components of V are
always singular (see [69, Thm. 8 of Ch. 9, 6]).
Since dim Tp (Cn ) = n for every p Cn , we see that Cn is smooth. For an
irreducible affine variety V Cn of dimension d, fix p V and write I(V ) =
h f1 , . . . , fs i. Using Lemma 1.0.6, it is straightforward to show that V is smooth

at p if and only if the Jacobian matrix
f

i
(1.0.3)
J p ( f1 , . . . , fs ) =
(p)
x j
1is,1 jn
has rank n d (Exercise 1.0.9). Here is a simple example.
Example 1.0.8. As noted in Example 1.0.4, the plane curve C defined by x 3 = y 2

has I(C) = hx 3 y 2 i C[x, y]. A point p = (a, b) C has Jacobian
Jp = (3a2 , 2b),
so the origin is the only singular point of C.
Since Tp (V ) = HomC (mV ,p /mV2 ,p , C), we see that V is smooth at p when dim V
equals the dimension of mV ,p /mV2 ,p as a vector space over OV ,p /mV ,p . In terms of
commutative algebra, this means that p V is smooth if and only if OV ,p is a
regular local ring. See [10, p. 123] or [89, 10.3].
We can relate smoothness and normality as follows.
Proposition 1.0.9. A smooth irreducible affine variety V is normal.
T
Proof. In 3.0 we will see that C[V ] = pV OV ,p . By Exercise 1.0.7, C[V ] is
normal once we prove that OV ,p is normal for all p V . Hence it suffices to show
that OV ,p is normal whenever p is smooth.
This follows from some powerful results in commutative algebra: OV ,p is a
regular local ring when p is a smooth point of V (see above), and every regular
local ring is a UFD (see [89, Thm. 19.19]). Then we are done since every UFD is
normal. A direct proof that OV ,p is normal at a smooth point p V is sketched in
Exercise 1.0.10.

The converse of Propostion 1.0.9 can fail. We will see in 1.3 that the affine
variety V(xy zw) C4 is normal, yet V(xy zw) is singular at the origin.
Products of Affine Varieties. Given affine varieties V1 and V2 , there are several
ways to show that the cartesian product V1 V2 is an affine variety. The most direct
way is to proceed as follows. Let V1 Cm = Spec(C[x1 , . . . , xm ]) and V2 Cn =
Spec(C[y1 , . . . , yn ]). Take I(V1 ) = h f1 , . . . , fs i and I(V2 ) = hg1 , . . . , gt i. Since the fi
and g j depend on separate sets of variables, it follows that
V1 V2 = V( f1 , . . . , fs , g1 , . . . , gt ) Cm+n
is an affine variety.
A fancier method is to use the mapping properties of the product. This will
also give an intrinsic description of its coordinate ring. Given V1 and V2 as above,
V1 V2 should be an affine variety with projections i : V1 V2 Vi such that

whenever we have a diagram
W
V1 V2
"
1
/ V1
$
V2
where i : W Vi are morphisms from an affine variety W , there should be a unique

morphism (the dotted arrow) that makes the diagram commute, i.e., i = i .
For the coordinate rings, this means that whenever we have a diagram
C[V2 ]
2
C[V1 ]
/ C[V1 V2 ]
%
. C[W ]
with C-algebra homomorphisms i : C[Vi ] C[W ], there should be a unique Calgebra homomorphism (the dotted arrow) that makes the diagram commute. By
the universal mapping property of the tensor product of C-algebras, C[V1 ]C C[V2 ]
has the mapping properties we want. Since C[V1 ] C C[V2 ] is a finitely generated
C-algebra with no nilpotents (see the appendix to this chapter), it is the coordinate
ring C[V1 V2 ]. For more on tensor products, see [10, pp. 2427] or [89, A2.2].
Example 1.0.10. Let V be an affine variety. Since Cn = Spec(C[y1 , . . . , yn ]), the
product V Cn has coordinate ring
C[V ] C C[y1 , . . . , yn ] = C[V ][y1 , . . . , yn ].
If V is contained in Cm with I(V ) = h f1 , . . . , fs i C[x1 , . . . , xm ], it follows that

I(V Cn ) = h f1 , . . . , fs i C[x1 , . . . , xm , y1 , . . . , yn ].
For later purposes, we also note that the coordinate ring of V (C )n is

C[V ] C C[y11 , . . . , yn1 ] = C[V ][y11 , . . . , yn1 ].
Given affine varieties V1 and V2 , we note that the Zariski topology on V1 V2

is usually not the product of the Zariski topologies on V1 and V2 .
Example 1.0.11. Consider C2 = C C. By definition, a basis for the product of
the Zariski topologies consists of sets U1 U2 where Ui are Zariski open in C. Such
a set is the complement of a union of collections of horizontal and vertical lines
in C2 . This makes it easy to see that Zariski closed sets in C2 such as V(y x 2 )
cannot be closed in the product topology.
Exercises for 1.0.

1.0.1. Prove Lemma 1.0.1. Hint: You will need the Nullstellensatz.
1.0.2. Let R be a commutative C-algebra. A subset S R is a multipliciative subset provided 1 S, 0
/ S, and S is closed under multiplication. The localization RS consists of all
formal expressions g/s, g R, s S, modulo the equivalence relation
g/s h/t u(tg sh) = 0 for some u S.
(a) Show that the usual formulas for adding and multiplying fractions induce well-defined
binary operations that make RS into C-algebra.
(b) If R has no nonzero nilpotents, then prove that the same is true for RS .
For more on localization, see [10, Ch. 3] or [89, Ch. 2].
1.0.3. Let R be a finitely generated C-algebra without nilpotents as in Lemma 1.0.1 and
let f R be nonzero. Then S = {1, f , f 2 , . . . } is a multiplicative set. The localization RS is
denoted R f and is called the localization of R at f .
(a) Show that R f is a finitely generated C-algebra without nilpotents.
(b) Show that R f satisfies Spec(R f ) = Spec(R) f .
(c) Show that R f is given by (1.0.2) when R is an integral domain.
1.0.4. Let V be an affine variety with coordinate ring C[V ]. Given a point p V , let
S = {g C[V ] | g(p) 6= 0}.
(a) Show that S is a multiplicative set. The localization C[V ]S is denoted OV ,p and is
called the local ring of V at p.
(b) Show that every OV ,p has a well-defined value (p) and that
mV ,p = { OV ,p | (p) = 0}
is the unique maximal ideal of OV ,p .
(c) When V is irreducible, show that OV ,p agrees with the definition given in the text.
1.0.5. Prove that a UFD is normal.
1.0.6. In the setting of Example 1.0.5, show that C[
y/
x] C(C) is the integral closure of
C[C] and that the normalization C C is defined by t 7 (t 2 ,t 3 ).
1.0.7. In this exercise, you will prove some properties of normal domains needed for 1.3.
(a) Let R be a normal domain with field of fractions K and let S R be a multiplicative
subset. Prove that the localization RS is normal.
(b) Let R , T
A, be normal domains with the same field of fractions K. Prove that the
intersection A R is normal.
1.0.8. Prove that dim Tp (Cn ) = n for all p Cn .
1.0.9. Use Lemma 1.0.6 to prove the claim made in the text that smoothness is determined
by the rank of the Jacobian matrix (1.0.3).
10
1.0.10. Let V be irreducible and suppose that p V is smooth. The goal of this exercise
is to prove that OV ,p is normal using standard results from commutative algebra. Set n =
dim V and consider the ring of formal power series C[[x1 , . . . , xn ]]. This is a local ring with
maximal ideal m = hx1 , . . . , xn i. We will use three facts:
C[[x1 , . . . , xn ]] is a UFD by [280, p. 148] and hence normal by Exercise 1.0.5.
Since p V is smooth, [207, 1C] proves the existence of a C-algebra homomorphism

OV ,p C[[x1 , . . . , xn ]] that induces isomorphisms
OV ,p /mV ,p C[[x1 , . . . , xn ]]/m
for all 0. This implies that the completion (see [10, Ch. 10])
ObV ,p = lim OV ,p /mV ,p
is isomorphic to a formal power series ring, i.e., ObV ,p C[[x1 , . . . , xn ]]. Such an isomorphism captures the intuitive idea that at a smooth point, functions should have
power series expansions in local coordinates x1 , . . . , xn .
If I OV ,p is an ideal, then
I=
=1 (I + mV ,p ).
This theorem of Krull holds for any ideal I in a Noetherian local ring A and follows
from [10, Cor. 10.19] with M = A/I.
Now assume that p V is smooth.
(a) Use the third bullet to show that OV ,p C[[x1 , . . . , xn ]] is injective.
(b) Suppose that a, b OV ,p satisfy b|a in C[[x1 , . . . , xn ]]. Prove that b|a in OV ,p . Hint:
Use the second bullet to show a bOV ,p + mV ,p and then use the third bullet.
(c) Prove that OV ,p is normal. Hint: Use part (b) and the first bullet.
This argument can be continued to show that OV ,p is a UFD. See [207, (1.28)]
1.0.11. Let V and W be affine varieties and let S V be a subset. Prove that S W = S W .
1.0.12. Let V and W be irreducible affine varieties. Prove that V W is irreducible. Hint:
Suppose V W = Z1 Z2 , where Z1 , Z2 are closed. Let Vi = {v V | {v} W Zi }. Prove
that V = V1 V2 and that Vi is closed in V . Exercise 1.0.11 will be useful.
1.1. Introduction to Affine Toric Varieties

We first discuss what we mean by torus and then explore various constructions
of affine toric varieties.
The Torus. The affine variety (C )n is a group under component-wise multiplication. A torus T is an affine variety isomorphic to (C )n , where T inherits a group
structure from the isomorphism.
The term torus is taken from the language of linear algebraic groups. We
will use (without proof) basic results about tori that can be found in standard texts
on algebraic groups such as [37], [152], and [256]. See also [36, Ch. 3] for a
self-contained treatment of tori.
11
We begin with characters and one-parameter subgroups.

A character of a torus T is a morphism : T C that is a group homomorphism. For example, m = (a1 , . . . , an ) Zn gives a character m : (C )n C
defined by
m (t1 , . . . ,tn ) = t1a1 tnan .
(1.1.1)
One can show that all characters of (C )n arise this way (see [152, 16]). Thus the
characters of (C )n form a group isomorphic to Zn .
For an arbitrary torus T , its characters form a free abelian group M of rank
equal to the dimension of T . It is customary to say that m M gives the character
m : T C .
We will need the following result concerning tori (see [152, 16] for a proof).
Proposition 1.1.1.
(a) Let T1 and T2 be tori and let : T1 T2 be a morphism that is a group homomorphism. Then the image of is a torus and is closed in T2 .
(b) Let T be a torus and let H T be an irreducible subvariety of T that is a
subgroup. Then H is a torus.

Assume that a torus T acts linearly on a finite dimensional vector space W over
C, where the action of t T on w W is denoted t w. A basic result is that the
linear maps w 7 t w are diagonalizable and can be simultaneously diagonalized.
We describe this as follows. Given m M, define the eigenspace
Wm = {w W | t w = m (t)w for all t T }.
If Wm 6= {0}, then every w Wm \ {0} is a simultaneous eigenvector for all t T ,

with eigenvalue given by m (t). See [256, Thm. 3.2.3] for a proof of the following.
L
Proposition 1.1.2. In the above situation, we have W = mM Wm .

A one-parameter subgroup of a torus T is a morphism : C T that is a
group homomorphism. For example, u = (b1 , . . . , bn ) Zn gives a one-parameter
subgroup u : C (C )n defined by
(1.1.2)
u (t) = (t b1 , . . . ,t bn ).
All one-parameter subgroups of (C )n arise this way (see [152, 16]). It follows
that the group of one-parameter subgroups of (C )n is naturally isomorphic to Zn .
For an arbitrary torus T , the one-parameter subgroups form a free abelian group N
of rank equal to the dimension of T . As with the character group, an element u N
gives the one-parameter subgroup u : C T .
There is a natural bilinear pairing h , i : M N Z defined as follows.
(Intrinsic) Given a character m and a one-parameter subgroup u , the composition m u : C C is a character of C , which is given by t 7 t for
some Z. Then hm, ui = .
12
(Concrete) If T = (C )n with m = (a1 , . . . , an ) Zn , u = (b1 , . . . , bn ) Zn , then

one computes that
n
X
ai bi ,
(1.1.3)
hm, ui =
i=1
i.e., the pairing is the usual dot product.
It follows that the characters and one-parameter subgroups of a torus T form

free abelian groups M and N of finite rank with a pairing h , i : M N Z that
identifies N with HomZ (M, Z) and M with HomZ (N, Z). In terms of tensor products, one obtains a canonical isomorphism N Z C T via u t 7 u (t). Hence
it is customary to write a torus as TN .
From this point of view, picking an isomorphism TN (C )n induces dual
bases of M and N, i.e., isomorphisms M Zn and N Zn that turn characters
into Laurent monomials (1.1.1), one-parameter subgroups into monomial curves
(1.1.2), and the pairing into dot product (1.1.3).
The Definition of Affine Toric Variety. We now define the main object of study of
this chapter.
Definition 1.1.3. An affine toric variety is an irreducible affine variety V containing a torus TN (C )n as a Zariski open subset such that the action of TN on itself
extends to an algebraic action of TN on V . (By algebraic action, we mean an action
TN V V given by a morphism.)
Obvious examples of affine toric varieties are (C )n and Cn . Here are some
less trivial examples.
Example 1.1.4. The plane curve C = V(x 3 y 2 ) C2 has a cusp at the origin.
This is an affine toric variety with torus
C \ {0} = C (C )2 = {(t 2 ,t 3 ) | t C } C ,
where the isomorphism is t 7 (t 2 ,t 3 ). Example 1.0.4 shows that C is a nonnormal

toric variety.
Example 1.1.5. The variety V = V(xy zw) C4 is a toric variety with torus
V (C )4 = {(t1 ,t2 ,t3 ,t1t2 t31 ) | ti C } (C )3 ,
where the isomorphism is (t1 ,t2 ,t3 ) 7 (t1 ,t2 ,t3 ,t1t2 t31 ). We will see later that V is
normal.
Example 1.1.6. Consider the surface in Cd+1 parametrized by the map

: C2 Cd+1
defined by (s,t) 7 (s d , s d1 t, . . . , st d1 ,t d ). Thus is defined using all degree d

monomials in s,t.
13
Let the coordinates of Cd+1 be x0 , . . . , xd and let I C[x0 , . . . , xd ] be the ideal

generated by the 2 2 minors of the matrix

x0 x1 xd2 xd1
,
x1 x2 xd1 xd
so I = hxi x j+1 xi+1 x j | 0 i < j d 1i. In Exercise 1.1.1 you will verify that
(C2 ) = V(I), so that Cbd = (C2 ) is an affine variety. You will also prove that
I(Cbd ) = I, so that I is the ideal of all polynomials vanishing on Cbd . It follows that I
is prime since V(I) is irreducible by Proposition 1.1.8 below. The affine surface Cbd
is called the rational normal cone of degree d and is an example of a determinantal
variety. We will see below that I is a toric ideal.
It is straightforward to show that Cbd is a toric variety with torus
((C )2 ) = Cbd (C )d+1 (C )2 .
We will study this example from the projective point of view in Chapter 2.
We next explore three equivalent ways of constructing affine toric varieties.

Lattice Points. In this book, a lattice is a free abelian group of finite rank. Thus
a lattice of rank n is isomorphic to Zn . For example, a torus TN has lattices M (of
characters) and N (of one-parameter subgroups).
Given a torus TN with character lattice M, a set A = {m1 , . . . , ms } M gives
characters mi : TN C . Then consider the map
(1.1.4)
defined by
A : TN Cs

A (t) = m1 (t), . . . , ms (t) Cs .
Definition 1.1.7. Given a finite set A M, the affine toric variety YA is defined
to be the Zariski closure of the image of the map A from (1.1.4).
This definition is justified by the following proposition.
Proposition 1.1.8. Given A M as above, let ZA M be the sublattice generated by A . Then YA is an affine toric variety whose torus has character lattice
ZA . In particular, the dimension of YA is the rank of ZA .
Proof. The map (1.1.4) can be regarded as a map
A : TN (C )s
of tori. By Proposition 1.1.1, the image T = A (TN ) is a torus that is closed in
(C )s . The latter implies that YA (C )s = T since YA is the Zariski closure of
the image. It follows that the image is Zariski open in YA . Furthermore, T is
irreducible (it is a torus), so the same is true for its Zariski closure YA .
14
We next consider the action of T . Since T (C )s , an element t T acts on

C and takes varieties to varieties. Then
s
T = t T t YA
shows that t YA is a variety containing T . Hence YA t YA by the definition of

Zariski closure. Replacing t with t 1 leads to YA = t YA , so that the action of T
induces an action on YA . We conclude that YA is an affine toric variety.
It remains to compute the character lattice of T , which we will temporarily
denote by M . Since T = A (TN ), the map A gives the commutative diagram
A
TN E
E
/ (C )s
O
EE
EE
EE
E" "
?
where denotes a surjective map and an injective map. This diagram of tori
induces a commutative diagram of character lattices
M aBo
bA
Zs
BB
BB
BB
B0 P
M.
b A : Zs M takes the standard basis e1 , . . . , es to m1 , . . . , ms , the image of

Since
b A is ZA . By the diagram, we obtain M ZA . Then we are done since the
dimension of a torus equals the rank of its character lattice.

In concrete terms, fix a basis of M, so that we may assume M = Zn . Then the s
vectors in A Zn can be regarded as the columns of an n s matrix A with integer
entries. In this case, the dimension of YA is simply the rank of the matrix A.
We will see below that every affine toric variety is isomorphic to YA for some
finite subset A of a lattice.
Toric Ideals. Let YA Cs = Spec(C[x1 , . . . , xs ]) be the affine toric variety coming from a finite set A = {m1 , . . . , ms } M. We can describe the ideal I(YA )
C[x1 , . . . , xs ] as follows. As in the proof of Proposition 1.1.8, A induces a map of
character lattices
b A : Zs M
that sends the standard basis e1 , . . . , es to m1 , . . . , ms . Let L be the kernel of this

map, so that we have an exact sequence
0 L Zs M.
In down to earth terms, elements = (1 , . . . , s ) of L satisfy
hence record the linear relations among the mi .
Ps
i=1 i mi
= 0 and
Given = (1 , . . . , s ) L, set
X
+ =
i ei
and
i >0
15
i ei .
i <0
Note that = + and that + , Ns . It follows easily that the binomial

Q
Q
x + x = i >0 xii i <0 xii
vanishes on the image of A and hence on YA since YA is the Zariski closure of

the image.
Proposition 1.1.9. The ideal of the affine toric variety YA Cs is

I(YA ) = x + x | L = x x | , N s and L .
Proof. We leave it to the reader to prove equality of the two ideals on the right
(Exercise 1.1.2). Let IL denote this ideal and note that IL I(YA ). We prove
the opposite inclusion following [264, Lem. 4.1]. Pick a monomial order > on
C[x1 , . . . , xs ] and an isomorphism TN (C )n . Thus we may assume M = Zn and
the map : (C )n Cs is given by Laurent monomials t mi in variables t1 , . . . ,tn .
If IL 6=
QI(YA ), then we can pick f I(YA ) \ IL with minimal leading monomial
x = si=1 xiai . Rescaling if necessary, x becomes the leading term of f .
Since f (t m1 , . . . ,t ms ) is identically zero as a polynomial in t1 , . . . ,tn , there must
be cancellation involving the term coming from x . In other words, f must contain
Q
a monomial x = si=1 xibi < x such that
s
Y
(t mi )ai =
i=1
i=1
This implies that
s
X
i=1
s
Y
(t mi )bi .
ai mi =
s
X
bi mi ,
i=1
P
so that = si=1 (ai bi )ei L. Then x x IL by the second description
of IL . It follows that f x + x also lies in I(YA ) \ IL and has strictly smaller
leading term. This contradiction completes the proof.

Given A M, there are several ways to compute the ideal I(YA ) = IL of
Proposition 1.1.9. In simple cases, the rational implicitization algorithm of [69,
Ch. 3, 3] can be used. One can also compute IL using a basis of L and ideal
quotients (Exercise 1.1.3). For more on computing IL , see [264, Ch. 12].
Inspired by Proposition 1.1.9, we make the following definition.
Definition 1.1.10. Let L Zs be a sublattice.

(a) The ideal IL = x x | , Ns and L is called a lattice ideal.
(b) A prime lattice ideal is called a toric ideal.
16
Since toric varieties are irreducible, the ideals appearing in Proposition 1.1.9
are toric ideals. Examples of toric ideals include:
Example 1.1.4 : hx 3 y 2 i C[x, y]
Example 1.1.5 : hxz ywi C[x, y, z, w]
Example 1.1.6 : hxi x j+1 xi+1 x j | 0 i < j d 1i C[x0 , . . . , xd ].

(The latter is the ideal of the rational normal cone Cbd Cd+1 .) In each example,
we have a prime ideal generated by binomials. As we now show, such ideals are
automatically toric.
Proposition 1.1.11. An ideal I C[x1 , . . . , xs ] is toric if and only if it is prime and
generated by binomials.
Proof. One direction is obvious. So suppose that I is prime and generated by binomials x i x i . Then observe that V(I) (C )s is nonempty (it contains (1, . . . , 1))
and is a subgroup of (C )s (easy to check). Since V(I) Cs is irreducible, it follows that V(I) (C )s is an irreducible subvariety of (C )s that is also a subgroup.
By Proposition 1.1.1, we see that T = V(I) (C )s is a torus.
Projecting on the ith coordinate of (C )s gives a character T (C )s C ,

which by our usual convention we write as mi : T C for mi M. It follows
easily that V(I) = YA for A = {m1 , . . . , ms }, and since I is prime, we have I =
I(YA ) by the Nullstellensatz. Then I is toric by Proposition 1.1.9.

We will later see that all affine toric varieties arise from toric ideals. For more
on toric ideals and lattice ideals, the reader should consult [204, Ch. 7].
Affine Semigroups. A semigroup is a set S with an associative binary operation
and an identity element. To be an affine semigroup, we further require that:
The binary operation on S is commutative. We will write the operation as +
and the identity element as 0. Thus a finite set A S gives

P
NA =
mA am m | am N S.
The semigroup is finitely generated, meaning that there is a finite set A S

such that NA = S.
The semigroup can be embedded in a lattice M.
The simplest example of an affine semigroup is Nn Zn . More generally, given

a lattice M and a finite set A M, we get the affine semigroup NA M. Up to
isomorphism, all affine semigroups are of this form.
Given an affine semigroup S M, the semigroup algebra C[S] is the vector
space over C with S as basis and multiplication induced by the semigroup structure
17
of S. To make this precise, we think of M as the character lattice of a torus TN , so

that m M gives the character m . Then
nX
o
C[S] =
cm m | cm C and cm = 0 for all but finitely many m ,
mS
with multiplication induced by
m m = m+m .
If S = NA for A = {m1 , . . . , ms }, then C[S] = C[ m1 , . . . , ms ].

Here are two basic examples.
Example 1.1.12. The affine semigroup N n Zn gives the polynomial ring

C[N n ] = C[x1 , . . . , xn ],
where xi = ei and e1 , . . . , en is the standard basis of Zn .
Example 1.1.13. If e1 , . . . , en is a basis of a lattice M, then M is generated by

A = {e1 , . . . , en } as an affine semigroup. Setting ti = ei gives the Laurent
polynomial ring
C[M] = C[t11 , . . . ,tn1 ].
Using Example 1.0.2, one sees that C[M] is the coordinate ring of the torus TN .
Affine semigroup rings give rise to affine toric varieties as follows.

Proposition 1.1.14. Let S M be an affine semigroup. Then:
(a) C[S] is an integral domain and finitely generated as a C-algebra.

(b) Spec(C[S]) is an affine toric variety whose torus has character lattice ZS, and
if S = NA for a finite set A M, then Spec(C[S]) = YA .
Proof. As noted above, A = {m1 , . . . , ms } implies C[S] = C[ m1 , . . . , ms ], so C[S]

is finitely generated. Since C[S] C[M] follows from S M, we see that C[S] is
an integral domain by Example 1.1.13.
Using A = {m1 , . . . , ms }, we get the C-algebra homomorphism
: C[x1 , . . . , xs ] C[M]
where xi 7 mi C[M]. This corresponds to the morphism

A : TN Cs
from (1.1.4), i.e., = (A ) in the notation of 1.0. One checks that the kernel
of is the toric ideal I(YA ) (Exercise 1.1.4). The image of is C[ m1 , . . . , ms ] =
C[S], and then the coordinate ring of YA is
(1.1.5)
C[YA ] = C[x1 , . . . , xs ]/I(YA )

= C[x1 , . . . , xs ]/Ker() Im() = C[S].
18
This proves that Spec(C[S]) = YA . Since S = NA implies ZS = ZA , the torus of

YA = Spec(C[S]) has the desired character lattice by Proposition 1.1.8.

Here is an example of this proposition.
Example 1.1.15. Consider the affine semigroup S Z generated by 2 and 3, so
that S = {0, 2, 3, . . . }. To study the semigroup algebra C[S], we use (1.1.5). If we
set A = {2, 3}, then A (t) = (t 2 ,t 3 ) and the toric ideal is I(YA ) = hx 3 y 2 i by
Example 1.1.4. Hence
C[S] = C[t 2 ,t 3 ] C[x, y]/hx 3 y 2 i
and the affine toric variety YA is the curve x 3 = y 2 from Example 1.1.4.
Equivalence of Constructions. Before stating our main result, we need to study

the action of the torus TN on the semigroup algebra C[M]. The action of TN on
itself given by multiplication induces an action on C[M] as follows: if t TN and
f C[M], then t f C[M] is defined by p 7 f (t 1 p) for p TN . The minus
sign will be explained in 5.0.
The following lemma will be used several times in the text.
Lemma 1.1.16. Let A C[M] be a subspace stable under the action of TN . Then
M
A=
C m.
m A
L
Proof. Let A = m A C m and note that A A. For the opposite inclusion,
pick f 6= 0 in A. Since A C[M], we can write
X
f=
cm m ,
mB
where B M is finite and cm 6= 0 for all m B. Then f B A, where

B = Span( m | m B) C[M].
An easy computation shows that t m = m (t 1 ) m . It follows that B and

hence B A are stable under the action of TN . Since B A is finite-dimensional,
Proposition 1.1.2 implies that B A is spanned by simultaneous eigenvectors of TN .
This is taking place in C[M], where simultaneous eigenvectors are characters. It
follows that BA is spanned by characters. Then the above expression for f BA
implies that m A for m B. Hence f A , as desired.

We can now state the main result of this section, which asserts that our various
approaches to affine toric varieties all give the same class of objects.
Theorem 1.1.17. Let V be an affine variety. The following are equivalent:
(a) V is an affine toric variety according to Definition 1.1.3.
(b) V = YA for a finite set A in a lattice.
19
(c) V is an affine variety defined by a toric ideal.

(d) V = Spec(C[S]) for an affine semigroup S.
Proof. The implications (b) (c) (d) (a) follow from Propositions 1.1.8,
1.1.9 and 1.1.14. For (a) (d), let V be an affine toric variety containing the
torus TN with character lattice M. Since the coordinate ring of TN is the semigroup
algebra C[M], the inclusion TN V induces a map of coordinate rings
C[V ] C[M].
This map is injective since TN is Zariski dense in V , so that we can regard C[V ] as
a subalgebra of C[M].
Since the action of TN on V is given by a morphism TN V V , we see that
if t TN and f C[V ], then p 7 f (t 1 p) is a morphism on V . It follows that
C[V ] C[M] is stable under the action of TN . By Lemma 1.1.16, we obtain
M
C[V ] =
C m.
m C[V ]
Hence C[V ] = C[S] for the semigroup S = {m M | m C[V ]}.
Finally, since C[V ] is finitely generated, we can find f1 , . . . , fs C[V ] with

C[V ] = C[ f1 , . . . , fs ]. Expressing each fi in terms of characters as above gives a
finite generating set of S. It follows that S is an affine semigroup.

Here is one way to think about the above proof. When an irreducible affine
variety V contains a torus TN as a Zariski open subset, we have the inclusion
C[V ] C[M].
Thus C[V ] consists of those functions on the torus TN that extend to polynomial
functions on V . Then the key insight is that V is a toric variety precisely when the
functions that extend are determined by the characters that extend.
Example 1.1.18. We have seen that V = V(xy zw) C4 is a toric variety with
toric ideal hxy zwi C[x, y, z, w]. Also, the torus is (C )3 via the map (t1 ,t2 ,t3 ) 7
(t1 ,t2 ,t3 ,t1t2 t31 ). The lattice points used in this map can be represented as the
columns of the matrix
1 0 0
1
0 1 0
(1.1.6)
1 .
0 0 1 1
The corresponding semigroup S Z3 consists of the N-linear combinations of the
column vectors. Hence the elements of S are lattice points lying in the polyhedral region in R3 pictured in Figure 1 on the next page. In this figure, the four
vectors generating S are shown in bold, and the boundary of the polyhedral region
is partially shaded. In the terminology of 1.2, this polyhedral region is a rational
20
(0,0,1)
(0,1,0)
(1,0,0)
(1,1,1)
Figure 1. Cone containing the lattice points corresponding to V = V(xy zw)
polyhedral cone. In Exercise 1.1.5 you will show that S consists of all lattice points
lying in the cone in Figure 1. We will use this in 1.3 to prove that V is normal.
Exercises for 1.1.
1.1.1. As in Example 1.1.6, let
I = hxi x j+1 xi+1 x j | 0 i < j d 1i C[x0 , . . . , xd ]
and let Cbd be the surface parametrized by
(s,t) = (s d , s d1 t, . . . , st d1 ,t d ) Cd+1 .
(a) Prove that V(I) = (C2 ) Cd+1 . Thus Cbd = V(I).

(b) Prove that I(Cbd ) is homogeneous.
(c) Consider lex monomial order with x0 > x1 > > xd . Let f I(Cbd ) be homogeneous
of degree and let r be the remainder of f on division by the generators of I. Prove
that r can be written
r = h0 (x0 , x1 ) + h1 (x1 , x2 ) + + hd1 (xd1 , xd )
where hi is homogeneous of degree . Also explain why we may assume that the
coefficient of xi in hi is zero for 1 i d 1.
(d) Use part (c) and r(s d , s d1 t, . . . , st d1 ,t d ) = 0 to show that r = 0.

(e) Use parts (b), (c) and (d) to prove that I = I(Cbd ). Also explain why the generators of
I are a Grobner basis for the above lex order.
1.1.2. Let L Zs be a sublattice. Prove that
hx + x | Li = hx x | , N s , Li.
Note that when L, the vectors + , N s have disjoint support (i.e., no coordinate is
positive in both), while this may fail for arbitrary , Ns with L.
21
1.1.3. Let IL be a toric ideal and let 1 , . . . , r be a basis of the sublattice L Zs . Define
i
i
IL = hx + x | i = 1, . . . , ri.
Prove that IL = IL : hx1 xs i . Hint: Given , N s with L, write =

P
r
i
i=1 ai , ai Z. This implies
!ai
!ai
i
Y x
Y x +i
1.
x
1 =
i
i
+
ai <0 x
ai >0 x
Show that multiplying this by (x1 xs )k gives an element of IL for k 0. (By being more
careful, one can show that this result holds for lattice ideals. See [204, Lem. 7.6].)
1.1.4. Fix an affine variety V . Then f1 , . . . , fs C[V ] give a polynomial map : V Cs ,
which on coordinate rings is given by
s
: C[x1 , . . . , xs ] C[V ],
Let Y C be the Zariski closure of the image of .
xi 7 fi .
(a) Prove that I(Y ) = Ker( ).
(b) Explain how this applies to the proof of Proposition 1.1.14.

1.1.5. Let m1 = (1, 0, 0), m2 = (0, 1, 0), m3 = (0, 0, 1), m4 = (1, 1, 1) be the columns of
the matrix in Example 1.1.18 and let
C=
4
nX
i=1
o
i mi | i R0 R3
be the cone in Figure 1. Prove that C Z3 is a semigroup generated by m1 , m2 , m3 , m4 .

1.1.6. An interesting observation is that different sets of lattice points can parametrize the
same affine toric variety, even though these parametrizations behave slightly differently. In
this exercise you will consider the parametrizations
1 (s,t) = (s2 , st, st 3 )
and 2 (s,t) = (s3 , st,t 3 ).
(a) Prove that 1 and 2 both give the affine toric variety Y = V(xz y 3 ) C3 .
(b) We can regard 1 and 2 as maps
1 : C2 Y
and 2 : C2 Y .
Prove that 2 is surjective and that 1 is not.

In general, a finite subset A Zn gives a rational map A : Cn 99K YA . The image of
A in Cs is called a toric set in the literature. Thus 1 (C2 ) and 2 (C2 ) are toric sets. The
papers [169] and [239] study when a toric set equals the corresponding affine toric variety.
1.1.7. In Example 1.1.6 and Exercise 1.1.1 we constructed the rational normal cone Cbd
using all monomials of degree d in s,t. If we drop some of the monomials, things become
more complicated. For example, consider the surface parametrized by
(s,t) = (s4 , s3 t, st 3 ,t 4 ) C4 .
This gives a toric variety Y C4 . Show that the toric ideal of Y is given by
I(Y ) = hxw yz, yw 2 z 3 , xz 2 y 2 w, x 2 z y 3 i C[x, y, z, w].
22
The ideal Cb4 has quadratic generators; by removing s2t 2 , we now get cubic generators. See
Example B.1.1 for a computational approach to this exercise. See also Example 2.1.10,
where we will study the paramaterization from the projective point of view.
1.1.8. Instead of working over C, we will work over an algebraically closed field k of
characteristic 2. Consider the affine toric variety V k 5 parametrized by
(s,t, u) = (s4 ,t 4 , u4 , s8 u,t 12 u3 ) k 5 .
(a) Find generators for the toric ideal I = I(V ) k[x1 , x2 , x3 , x4 , x5 ].
(b) Show that dim V = 3. You may assume that Proposition 1.1.8 holds over k.
q

x44 + x18 x3 , x54 + x212 x33 .
(c) Show that I =
It follows that V k 5 has codimension two and can be defined by two equations, i.e., V
is a set-theoretic complete intersection. The paper [12] shows that if we replace k with an
algebraically closed field of characteristic p > 2, then the above parametrization is never a
set-theoretic complete intersection.
1.1.9. Prove that a lattice ideal IL for L Zs is a toric ideal if and only if Zs /L is torsionfree. Hint: When Zs /L is torsion-free, it can be regarded as the character lattice of a torus.
The other direction of the proof is more challenging. If you get stuck, see [204, Thm. 7.4].
1.1.10. Prove that I = hx 2 1, xy 1, yz 1i is the lattice ideal for the lattice
L = {(a, b, c) Z3 | a + b + c 0 mod 2} Z3 .
Also compute the primary decomposition of I to show that I is not prime.

1.1.11. Let TN be a torus with character lattice M. Then every point t TN gives an evaluation map t : M C defined by t (m) = m (t). Prove that t is a group homomorphism
and that the map t 7 t induces a group isomorphism
TN HomZ (M, C ).
1.1.12. Consider tori T1 and T2 with character lattices M1 and M2 . By Example 1.1.13, the
coordinate rings of T1 and T2 are C[M1 ] and C[M2 ]. Let : T1 T2 be a morphism that is
a group homomorphism. Then induces maps
b : M2 M1 and : C[M2 ] C[M1 ]
b of
by composition. Prove that is the map of semigroup algebras induced by the map
affine semigroups.
1.1.13. A commutative semigroup S is cancellative if u + v = u + w implies v = w for all

u, v, w S and torsion-free if nu = nv implies u = v for all n N \ {0} and u, v S. Prove
that S is affine if and only if it is finitely generated, cancellative, and torsion-free.
1.1.14. The requirement that an affine semigroup be finitely generated is important since
lattices contain semigroups that are not finitely generated. For example, let > 0 be irrational and consider the semigroup
S = {(a, b) N 2 | b a} Z2 .
Prove that S is not finitely generated.

(The generators of S are related to continued fractions. For example, when = (1 + 5)/2 is the golden ratio, the minimal generators of
S are (0, 1) and (F2n , F2n+1 ) for n = 1, 2, . . . , where Fn is the nth Fibonacci number. See
[231] and [259]. Continued fractions will play an important role in Chapter 10.
1.2. Cones and Affine Toric Varieties
23
1.1.15. Suppose that : M M is a group isomorphism. Fix a finite set A M and

let B = (A ). Prove that the toric varieties YA and YB are equivariantly isomorphic
(meaning that the isomorphism respects the torus action).

We begin with a brief discussion of rational polyhedral cones and then explain how
they relate to affine toric varieties.
Convex Polyhedral Cones. Fix a pair of dual vector spaces MR and NR . Our discussion of cones will omit most proofswe refer the reader to [105] for more
details and [218, App. A.1] for careful statements. See also [51, 128, 241].
Definition 1.2.1. A convex polyhedral cone in NR is a set of the form
o
nX
u u | u 0 NR ,
= Cone(S) =
uS
where S NR is finite. We say that is generated by S. Also set Cone() = {0}.

A convex polyhedral cone is in fact convex, meaning that x, y implies
x + (1 )y for all 0 1, and is a cone, meaning that x implies
x for all 0. Since we will only consider convex cones, the cones satisfying
Definition 1.2.1 will be called simply polyhedral cones.
Examples of polyhedral cones include the first quadrant in R2 or first octant in
R3 . For another example, the cone Cone(e1 , e2 , e1 + e3 , e2 + e3 ) R3 is pictured
in Figure 2. It is also possible to have cones that contain entire lines. For example,
Figure 2. Cone in R3 generated by e1 , e2 , e1 + e3 , e2 + e3
24
Cone(e1 , e1 ) R2 is the x-axis, while Cone(e1 , e1 , e2 ) is the closed upper halfplane {(x, y) R2 | y 0}. As we will see below, these last two examples are not
strongly convex.
We can also create cones using polytopes, which are defined as follows.
Definition 1.2.2. A polytope in NR is a set of the form
o
nX
X
u u | u 0,
u = 1 NR ,
P = Conv(S) =
uS
uS
where S NR is finite. We say that P is the convex hull of S.
Polytopes include all polygons in R2 and bounded polyhedra in R3 . As we will

see in later chapters, polytopes play a prominent role in the theory of toric varieties.
Here, however, we simply observe that a polytope P NR gives a polyhedral cone
C(P) NR R, called the cone of P and defined by
C(P) = { (u, 1) NR R | u P, 0}.
If P = Conv(S), then we can also describe this as C(P) = Cone(S {1}). Figure 3
shows what this looks when P is a pentagon in the plane.
Figure 3. The cone C(P) of a pentagon P R2
The dimension dim of a polyhedral cone is the dimension of the smallest

subspace W = Span() of NR containing . We call Span() the span of .
Dual Cones and Faces. As usual, the pairing between MR and NR is denoted h , i.
Definition 1.2.3. Given a polyhedral cone NR , its dual cone is defined by
= {m MR | hm, ui 0 for all u }.
Duality has the following important properties.

Proposition 1.2.4. Let NR be a polyhedral cone. Then is a polyhedral cone
in MR and ( ) = .
25
Given m 6= 0 in MR , we get the hyperplane

Hm = {u NR | hm, ui = 0} NR
and the closed half-space
Hm+ = {u NR | hm, ui 0} NR .
Then Hm is a supporting hyperplane of a polyhedral cone NR if Hm+ ,

and Hm+ is a supporting half-space. Note that Hm is a supporting hyperplane of
if and only if m \ {0}. Furthermore, if m1 , . . . , ms generate , then it is
straightforward to check that
(1.2.1)
= Hm+1 Hm+s .
Thus every polyhedral cone is an intersection of finitely many closed half-spaces.

We can use supporting hyperplanes and half-spaces to define faces of a cone.
Definition 1.2.5. A face of a cone of the polyhedral cone is = Hm for some
m , written . Using m = 0 shows that is a face of itself, i.e., .
Faces 6= are called proper faces, written .
The faces of a polyhedral cone have the following obvious properties.
Lemma 1.2.6. Let = Cone(S) be a polyhedral cone. Then:
(a) Every face of is a polyhedral cone.
(b) An intersection of two faces of is again a face of .
(c) A face of a face of is again a face of .
You will prove the following useful property of faces in Exercise 1.2.1.
Lemma 1.2.7. Let be a face of a polyhedral cone . If v, w and v + w ,
then v, w .

A facet of is a face of codimension 1, i.e., dim = dim 1. An edge of
is a face of dimension 1. In Figure 4 on the next page we illustrate a 3-dimensional
cone with shaded facets and a supporting hyperplane (a plane in this case) that cuts
out the vertical edge of the cone.
Here are some properties of facets.
Proposition 1.2.8. Let NR Rn be a polyhedral cone. Then:
(a) If = Hm+1 Hm+s for mi , 1 i s, then = Cone(m1 , . . . , ms ).
(b) If dim = n, then in (a) we can assume that the facets of are i = Hmi .
(c) Every proper face is the intersection of the facets of containing .
Note how part (b) of the proposition refines (1.2.1) when dim = dim NR .
26
supporting
hyperplane
Figure 4. A cone R3 and a hyperplane H supporting an edge H
When working in Rn , dot product allows us to identify the dual with Rn . From
this point of view, the vectors m1 , . . . , ms in part (a) of the proposition are facet
normals, i.e., perpendicular to the facets. This makes it easy to compute examples.
Example 1.2.9. It is easy to see that the facet normals to the cone R3 in
Figure 2 are m1 = e1 , m2 = e2 , m3 = e3 , m4 = e1 + e2 e3 . Hence
= Cone(e1 , e2 , e3 , e1 + e2 e3 ) R3 .
This is the cone of Figure 1 at the end of 1.1 whose lattice points describe the
semigroup of the affine toric variety V(xy zw) (see Example 1.1.18). As we will
see, this is part of how cones describe normal affine toric varieties.
Now consider , which is the cone in Figure 1. The reader can check that the
facet normals of this cone are e1 , e2 , e1 + e3 , e2 + e3 . Using duality and part (a) of
Proposition 1.2.8, we obtain
= ( ) = Cone(e1 , e2 , e1 + e3 , e2 + e3 ).
Hence we recover our original description of . See also Example B.2.1.
In this example, facets of the cone correspond to edges of its dual. More generally, given a face NR , we define
= {m MR | hm, ui = 0 for all u }
= {m | hm, ui = 0 for all u }
= .
We call the dual face of because of the following proposition.

Proposition 1.2.10. If is a face of a polyhedral cone and = , then:
(a) is a face of .
27
(b) The map 7 is a bijective inclusion-reversing correspondence between the

faces of and the faces of .
(c) dim + dim = n.
Here is an example of Proposition 1.2.10 when dim < dim NR .

Example 1.2.11. Let = Cone(e1 , e2 ) R3 . Figure 5 shows and . You
Figure 5. A 2-dimensional cone R3 and its dual R3
should check that the maximal face of , namely itself, gives the minimal face
of , namely the z-axis. Note also that
dim + dim = 3
even though has dimension 2.
Relative Interiors. As already noted, the span of a cone NR is the smallest

subspace of NR containing . Then the relative interior of , denoted Relint(), is
the interior of in its span. Exercise 1.2.2 will characterize Relint() as follows:
u Relint() hm, ui > 0 for all m \ .
When the span equals NR , the relative interior is just the interior, denoted Int().
For an example of how relative interiors arise naturally, suppose that .
This gives the dual face = of . Furthermore, if m , then one
easily sees that
m Hm .
In Exercise 1.2.2, you will show that if m , then
m Relint( ) = Hm .
Thus the relative interior Relint( ) tells us exactly which supporting hyperplanes
of cut out the face .
28
Strong Convexity. Of the cones shown in Figures 15, all but in Figure 5 have
the nice property that the origin is a face. Such cones are called strongly convex.
This condition can be stated several ways.
Proposition 1.2.12. Let NR Rn be a polyhedral cone. Then:
is strongly convex {0} is a face of
contains no positive-dimensional subspace of NR
() = {0}
dim = n.
You will prove Proposition 1.2.12 in Exercise 1.2.3. One corollary is that if a
polyhedral cone is strongly convex of maximal dimension, then so is . The
cones pictured in Figures 14 satisfy this condition.
In general, a polyhedral cone always has a minimal face that is the largest
subspace W contained in . Furthermore:
W = ().
W = Hm whenever m Relint( ).
= /W NR /W is a strongly convex polyhedral cone.
See Exercise 1.2.4.
Separation. When two cones intersect in a face of each, we can separate the cones
with the following result, often called the separation lemma.
Lemma 1.2.13 (Separation Lemma). Let 1 , 2 be polyhedral cones in NR that
meet along a common face = 1 2 . Then
= Hm 1 = Hm 2
for any m Relint(1 (2 ) ).
Proof. Given A, B NR , we set A B = {a b | a A, b B}. A standard result

from cone theory tells us that
1 (2 ) = (1 2 ) .
Now fix m Relint(1 (2 ) ). The above equation and Exercise 1.2.4 imply
that Hm cuts out the minimal face of 1 2 , i.e.,
Hm (1 2 ) = (1 2 ) (2 1 ).
However, we also have

(1 2 ) (2 1 ) = .
One inclusion is obvious since = 1 2 . For the other inclusion, write u

(1 2 ) (2 1 ) as
u = a1 a2 = b2 b1 ,
a1 , b1 1 , a2 , b2 2 .
29
Then a1 + b1 = a2 + b2 implies that this element lies in = 1 2 . Since a1 , b1

1 , we have a1 , b1 by Lemma 1.2.7, and a2 , b2 follows similarly. Hence
u = a1 a2 , as desired.
We conclude that Hm (1 2 ) = . Intersecting with 1 , we obtain
Hm 1 = ( ) 1 = ,
where the last equality again uses Lemma 1.2.7 (Exercise 1.2.5). If instead we
intersect with 2 , we obtain
Hm (2 ) = ( ) (2 ) = ,
and Hm 2 = follows.
In the situation of Lemma 1.2.13 we call Hm a separating hyperplane.

Rational Polyhedral Cones. Let N and M be dual lattices with associated vector
spaces NR = N Z R and MR = M Z R. For Rn we usually use the lattice Zn .
Definition 1.2.14. A polyhedral cone NR is rational if = Cone(S) for some
finite set S N.
The cones appearing in Figures 1, 2 and 5 are rational. We note without proof
that faces and duals of rational polyhedral cones are rational. Furthermore, if =
Cone(S) for S N finite and NQ = N Z Q, then
P

(1.2.2)
NQ =
uS u u | u 0 in Q .
One new feature is that a strongly convex rational polyedral cone has a
canonical generating set, constructed as follows. Let be an edge of . Since is
strongly convex, is a ray, i.e., a half-line, and since is rational, the semigroup
N is generated by a unique element u N. We call u the ray generator of
. Figure 6 shows the ray generator of a rational ray in the plane. The dots are
the lattice N = Z2 and the white ones are N.
ray generator u
Figure 6. A rational ray R2 and its unique ray generator u
Lemma 1.2.15. A strongly convex rational polyhedral cone is generated by the ray
generators of its edges.
30
It is customary to call the ray generators of the edges the minimal generators
of a strongly convex rational polyhedral cone. Figures 1 and 2 show 3-dimensional
strongly convex rational polyhedral cones and their ray generators.
In a similar way, a rational polyhedral cone of maximal dimension has unique
facet normals, which are the ray generators of the dual , which is strongly convex by Proposition 1.2.12.
Here are some especially important strongly convex cones.
Definition 1.2.16. Let NR be a strongly convex rational polyhedral cone.
(a) is smooth or regular if its minimal generators form part of a Z-basis of N,
(b) is simplicial if its minimal generators are linearly independent over R.

The cone pictured in Figure 5 is smooth, while those in Figures 1 and 2 are
not even simplicial. Note also that the dual of a smooth (resp. simplicial) cone of
maximal dimension is again smooth (resp. simplicial). Later in the section we will
give examples of simplicial cones that are not smooth.
Semigroup Algebras and Affine Toric Varieties. Given a rational polyhedral cone
NR , the lattice points
S = M M
form a semigroup. A key fact is that this semigroup is finitely generated.
Proposition 1.2.17 (Gordans Lemma). S = M is finitely generated and
hence is an affine semigroup.
Proof.PSince is rational polyhedral, = Cone(T ) for a finite set T M. Then
K = { mT m m | 0 m < 1} is a bounded region of MR Rn , so that K M is
finite since M Zn . Note that T (K M) S .
We claim TP
(K M) generates S as a semigroup. To prove this, take w S
and write w = mT m m where m 0. Then m = m + m with m N
and 0 m < 1, so that
X
X
w=
m m +
m m.
mT
mT
The second sum is in K M (remember w M). It follows that w is a nonnegative

integer combination of elements of T (K M).

Since affine semigroups give affine toric varieties, we get the following.
Theorem 1.2.18. Let NR Rn be a rational polyhedral cone with semigroup
S = M. Then
U = Spec(C[S ]) = Spec(C[ M])
is an affine toric variety. Furthermore,

dim U = n the torus of U is TN = N Z C is strongly convex.
31
Proof. By Gordans Lemma and Proposition 1.1.14, U is an affine toric variety

whose torus has character lattice ZS M. To study ZS , note that
ZS = S S = {m1 m2 | m1 , m2 S }.
Now suppose that km ZS for some k > 1 and m M. Then km = m1 m2 for

m1 , m2 S = M. Since m1 and m2 lie in the convex set , we have
m + m2 = 1k m1 + k1
k m2 .
It follows that m = (m + m2 ) m2 ZS , so that M/ZS is torsion-free. Hence

(1.2.3)
the torus of U is TN ZS = M rank ZS = n.
Since is strongly convex if and only if dim = n (Proposition 1.2.12), it remains

to show that
dim U = n rank ZS = n dim = n.
The first equivalence follows since the dimension of an affine toric variety is the
dimension of its torus, which is the rank of its character lattice. We leave the proof
of the second equivalence to the reader (Exercise 1.2.6).

Remark 1.2.19.
(a) For the rest of the book, we will always assume that NR is strongly convex
since we want TN to be the torus of the affine toric variety U .
(b) The reader may ask why we focus on NR since U = Spec(C[ M])
makes MR seem more important. The answer will become clear once we
understand how normal toric varieties are constructed from affine pieces. The
discussion following Proposition 1.3.16 gives a first hint of how this works.
Example 1.2.20. Let = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) NR = R3 with N = Z3 .
This is the cone pictured in Figure 2. By Example 1.2.9, is the cone pictured
in Figure 1, and by Example 1.1.18, the lattice points in this cone are generated
by columns of matrix (1.1.6). It follows from Example 1.1.18 that U is the affine
toric variety V(xy zw).
Example 1.2.21. Fix 0 r n and set = Cone(e1 , . . . , er ) Rn . Then

= Cone(e1 , . . . , er , er+1 , . . . , en )
and the corresponding affine toric variety is

1
, . . . , xn1 ]) = Cr (C )nr
U = Spec(C[x1 , . . . , xr , xr+1
(Exercise 1.2.7). This implies that if NR Rn is a smooth cone of dimension r,

then U Cr (C )nr . Figure 5 from Example 1.2.11 shows r = 2 and n = 3.
Example 1.2.22. Fix a positive integer d and let = Cone(de1 e2 , e2 ) R2 .
This has dual cone = Cone(e1 , e1 + de2 ). Figure 7 on the next page shows
when d = 4. The affine semigroup S = Z2 is generated by the lattice points
32
(1, i) for 0 i d. When d = 4, these are the white dots in Figure 7. (You will
prove these assertions in Exercise 1.2.8.)
Figure 7. The cone when d = 4
By 1.1, the affine toric variety U is the Zariski closure of the image of the
map : (C )2 Cd+1 defined by
(s,t) = (s, st, st 2 , . . . , st d ).
This map has the same image as the map (s,t) 7 (s d , s d1 t, . . . , st d1 ,t d ) used in
Example 1.1.6. Thus U is isomorphic to the rational normal cone Cbd Cd+1
whose ideal is generated by the 2 2 minors of the matrix

x0 x1 xd2 xd1
.
x1 x2 xd1 xd
Note that the cones and are simplicial but not smooth.
We will return to this example often. One thing evident in Example 1.1.6 is the
difference between cone generators and semigroup generators: the cone has
two generators but the semigroup S = Z2 has d + 1.
When NR has maximal dimension, the semigroup S = M has a

unique minimal generating set constructed as follows. Define an element m 6= 0 of
S to be irreducible if m = m + m for m , m S implies m = 0 or m = 0.
Proposition 1.2.23. Let NR be a strongly convex rational polyhedral cone of
maximal dimension and let S = M. Then
H = {m S | m is irreducible}
has the following properties:

(a) H is finite and generates S .
(b) H contains the ray generators of the edges of .
(c) H is the minimal generating set of S with respect to inclusion.
33
Proof. Proposition 1.2.12 implies that is strongly convex, so we can find an

element u N \ {0} such that hm, ui N for all m S and hm, ui = 0 if and
only if m = 0.
Now suppose that m S is not irreducible. Then m = m + m where m and
m are nonzero elements of S . It follows that
hm, ui = hm , ui + hm , ui
with hm , ui, hm , ui N \ {0}, so that

hm , ui < hm, ui
and
hm , ui < hm, ui.
Using induction on hm, ui, we conclude that every element of S is a sum of irreducible elements, so that H generates S . Furthermore, using a finite generating
set of S , one easily sees that H is finite. This proves part (a). The remaining
parts of the proof are covered in Exercise 1.2.9.

The set H S is called the Hilbert basis of S and its elements are the
minimal generators of S . More generally, Proposition 1.2.23 holds for any affine
semigroup S satisfying S (S) = {0}. Algorithms for computing Hilbert bases
are discussed in [204, 7.3] and [264, Ch. 13], and Hilbert bases can be computed
using the computer program Normaliz [57]. See Examples B.3.1 and B.3.2.
Exercises for 1.2.
1.2.1. Prove Lemma 1.2.7. Hint: Write = Hm for m .
1.2.2. Here are some properties of relative interiors. Let NR be a cone.
(a) Show that if u , then u Relint() if and only if hm, ui > 0 for all m \ .
(b) Let and fix m . Prove that
m Hm
m Relint( ) = Hm .
1.2.3. Prove Proposition 1.2.12.
1.2.4. Let NR be a polyhedral cone.
(a) Use Proposition 1.2.10 to show that has a unique minimal face with respect to .
Let W denote this minimal face.
(b) Prove that W = ( ) .
(c) Prove that W is the largest subspace contained in .
(d) Prove that W = ().
(e) Fix m . Prove that m Relint( ) if and only if W = Hm .
(f) Prove that = /W NR /W is a strongly convex polyhedral cone.
1.2.5. Let NR and let be defined as in the proof of Lemma 1.2.13. Prove
that = ( ) . Also show that = Span( ), i.e., is the smallest subspace
of NR containing .
34
1.2.6. Fix a lattice M and let Span(S) denote the span over R of a subset S MR .
(a) Let S M be finite. Prove that rank ZS = dim Span(S).
(b) Let S MR be finite. Prove that dim Cone(S) = dim Span(S).
(c) Use parts (a) and (b) to complete the proof of Theorem 1.2.18.
1.2.7. Prove the assertions made in Example 1.2.21.

1.2.8. Prove the assertions made in Example 1.2.22. Hint: First show that when a cone is
smooth, the ray generators of the cone also generate the corresponding semigroup. Then
write the cone of Example 1.2.22 as a union of such cones.
1.2.9. Complete the proof of Proposition 1.2.23. Hint for part (b): Show that the ray
generators of the edges of are irreducible in S . Given an edge of , it will help to
pick u N \ {0} such that = Hu .
1.2.10. Let NR be a cone generated by a set of linearly independent vectors in NR .
Show that is strongly convex and simplicial.
1.2.11. Explain the picture illustrated in Figure 8 in terms of Proposition 1.2.8.
Figure 8. A cone in the plane and its dual
1.2.12. Let P NR be a polytope lying in an affine hyperplane (= translate of a hyperplane)

not containing the origin. Generalize Figure 3 by showing that P gives a strongly convex
polyhedral cone in NR . Draw a picture.
1.2.13. Consider the cone = Cone(3e1 2e2 , e2 ) R2 .
(a) Describe and find generators of Z2 . Draw a picture similar to Figure 7.
(b) Compute the toric ideal of the affine toric variety U and explain how this exercise
relates to Exercise 1.1.6.
1.2.14. Consider the simplicial cone = Cone(e1 , e2 , e1 + e2 + 2e3 ) R3 .

(a) Describe and find generators of Z3 .
(b) Compute the toric ideal of the affine toric variety U .
1.2.15. Let be a strongly convex polyhedral cone of maximal dimension. Here is an

example taken from [105, p. 132] to show that and need not have the same number
of edges. Let R4 be the cone generated by 2ei + e j for all 1 i, j 4, i 6= j.
(a) Show that has 12 edges.
P
(b) Show that is generated by ei and ei + 2 j6=i e j , 1 i 4 and has 8 edges.
1.3. Properties of Affine Toric Varieties
35

The final task of this chapter is to explore the properties of affine toric varieties.
We will also study maps between affine toric varieties.
Points of Affine Toric Varieties. We first consider various ways to describe the
points of an affine toric variety.
Proposition 1.3.1. Let V = Spec(C[S]) be the affine toric variety of the affine
semigroup S. Then there are bijective correspondences between the following:
(a) Points p V .
(b) Maximal ideals m C[S].
(c) Semigroup homomorphisms S C, where C is considered as a semigroup

under multiplication.
Proof. The correspondence between (a) and (b) is standard (see [69, Thm. 5 of Ch.
5, 4]). The correspondence between (a) and (c) is special to the toric case.
Given a point p V , define S C by sending m S to m (p) C. This
makes sense since m C[S] = C[V ]. One easily checks that S C is a semigroup
homomorphism.
Going the other way, let : S C be a semigroup homomorphism. Since
is a basis of C[S], induces a surjective linear map C[S] C which is
a C-algebra homomorphism. The kernel of the map C[S] C is a maximal ideal
and thus gives a point p V by the correspondence between (a) and (b).
{ m }mS
We construct p concretely as follows. Let A = {m1 , . . . , ms } generate S, so

that V = YA Cs . Let p = ((m1 ), . . . , (ms )) Cs . Let us prove that p V . By
Proposition 1.1.9, it suffices to show that x x vanishes at p for all exponent
vectors = (a1 , . . . , as ) and = (b1 , . . . , bs ) satisfying
s
X
i=1
ai mi =
s
X
bi mi .
i=1
This is easy, since being a semigroup homomorphism implies that

s
s
s
s
Y

X
X
Y
ai
(mi ) bi .
bi mi =
ai mi =
(mi ) =
i=1
i=1
i=1
i=1
It is straightforward to show that this point of V agrees with the one constructed in
the previous paragraph (Exercise 1.3.1).
As an application of this result, we describe the torus action on V . In terms of

the embedding V = YA Cs , the proof of Proposition 1.1.8 shows that the action
of TN on YA is induced by the usual action of (C )s on Cs . But how do we see the
action intrinsically, without embedding into affine space? This is where semigroup
36
homomorphisms prove their value. Fix t TN and p V , and let p correspond to

the semigroup homorphism m 7 (m). In Exercise 1.3.1 you will show that t p
is given by the semigroup homomorphism m 7 m (t)(m). This description will
prove useful in Chapter 3 when we study torus orbits.
From the point of view of group actions, the action of TN on V is given by a
map TN V V . Since both sides are affine varieties, this should be a morphism,
meaning that it should come from a C-algebra homomorphism
C[S] = C[V ] C[TN V ] = C[TN ] C C[V ] = C[M] C C[S].
This homomorphism is given by m 7 m m for m S (Exercise 1.3.2).
We next characterize when the torus action on an affine toric variety has a
fixed point. An affine semigroup S is pointed if S (S) = {0}, i.e., if 0 is the
only invertible element of S. This is the semigroup analog of strongly convex.
Proposition 1.3.2. Let V be an affine toric variety. Then:

(a) If we write V = Spec(C[S]), then the torus action has a fixed point if and only
if S is pointed, in which case the unique fixed point is given by the semigroup
homomorphism S C defined by
(
1 m=0
(1.3.1)
m 7
0 m 6= 0.
(b) If we write V = YA Cs for A S \ {0}, then the torus action has a fixed
point if and only if 0 YA , in which case the unique fixed point is 0.
Proof. For part (a), let p V be represented by the semigroup homomorphism

: S C. Then p is fixed by the torus action if and only if m (t)(m) = (m)
for all m S and t TN . This equation is satisfied for m = 0 since (0) = 1, and if
m 6= 0, then picking t with m (t) 6= 1 shows that (m) = 0. Thus, if a fixed point
exists, then it is unique and is given by (1.3.1). Then we are done since (1.3.1) is a
semigroup homomorphism if and only if S is pointed.
For part (b), first assume that V = YA Cs has a fixed point, in which case
S = NA is pointed and the unique point p is given by (1.3.1). Then A S \ {0}
and the proof of Proposition 1.3.1 imply that p is the origin in Cs , so that 0 YA .
The converse follows since 0 Cs is fixed by (C )s , hence by YA (C )s .

Here is a useful corollary of Proposition 1.3.2 (Exercise 1.3.3).
Corollary 1.3.3. Let U be the affine toric variety of a strongly convex rational
polyhedral cone NR . Then the torus action on U has a fixed point if and
only if dim = dim NR , in which case the fixed point is unique and is given by the
maximal ideal
h m | m S \ {0}i C[S ],
where as usual S = M.
37
We will see in Chapter 3 that this corollary is part of the correspondence between torus orbits of U and faces of .
Normality and Saturation. We next study the question of when an affine toric
variety V is normal. We need one definition before stating our normality criterion.
Definition 1.3.4. An affine semigroup S M is saturated if for all k N \ {0} and
m M, km S implies m S.
For example, if NR is a strongly convex rational polyhedral cone, then
S = M is easily seen to be saturated (Exercise 1.3.4).
Theorem 1.3.5. Let V be an affine toric variety with torus TN . Then the following
are equivalent:
(a) V is normal.
(b) V = Spec(C[S]), where S M is a saturated affine semigroup.
(c) V = Spec(C[S ]) (= U ), where S = M and NR is a strongly convex

rational polyhedral cone.
Proof. By Theorem 1.1.17, V = Spec(C[S]) for an affine semigroup S contained in
a lattice, and by Proposition 1.1.14, the torus of V has the character lattice M = ZS.
Also let n = dim V , so that M Zn .
(a) (b): If V is normal, then C[S] = C[V ] is integrally closed in its field of
fractions C(V ). Suppose that km S for some k N \ {0} and m M. Then m
is a polynomial function on TN and hence a rational function on V since TN V is
Zariski open. We also have km C[S] since km S. It follows that m is a root
of the monic polynomial X k km with coefficients in C[S]. By the definition of
normal, we obtain m C[S], i.e., m S. Thus S is saturated.
(b) (c): Let A S be a finite generating set of S. Then S lies in the rational
polyhedral cone Cone(A ) MR , and rank ZA = n implies dim Cone(A ) = n by
Exercise 1.2.6. It follows that = Cone(A ) NR is a strongly convex rational polyhedral cone such that S M. In Exercise 1.3.4 you will prove that
equality holds when S is saturated. Hence S = S .
(c) (a): We need to show that C[S ] = C[ M] is normal when NR is

a strongly convex rational polyhedral cone. Let 1 , . . . , r be the rays of . Since
is generated by its rays (Lemma 1.2.15), we have
r
\
=
i .
Intersecting with M gives S =
Tr
i=1
i=1 Si ,
C[S ] =
which easily implies
r
\
i=1
C[Si ].
38
By Exercise 1.0.7, C[S ] is normal if each C[Si ] is normal, so it suffices to prove

that C[S ] is normal when is a rational ray in NR . Let u N be the ray
/ N for all k > 1, we can find a
generator of . Since u is primitive, i.e, 1k u
basis e1 , . . . , en of N with u = e1 (Exercise 1.3.5). This allows us to assume that
= Cone(e1 ), so that
C[S ] = C[x1 , x21 , . . . , xn1 ]
by Example 1.2.21. But C[x1 , . . . , xn ] is normal (it is a UFD), so its localization
C[x1 , . . . , xn ]x2 xn = C[x1 , x21 , . . . , xn1 ]
is also normal by Exercise 1.0.7. This completes the proof.
Example 1.3.6. We saw in Example 1.2.20 that V = V(xy zw) is the affine toric
variety U of the cone = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) pictured in Figure 2. Then
Theorem 1.3.5 implies that V is normal, as claimed in Example 1.1.5.
Example 1.3.7. By Example 1.2.22, the rational normal cone Cbd Cd+1 is the
affine toric variety of a strongly convex rational polyhedral cone and hence is normal by Theorem 1.3.5.
It is instructive to view this example using the parametrization
A (s,t) = (s d , s d1 t, . . . , st d1 ,t d )
from Example 1.1.6. Plotting the lattice points in A for d = 2 gives the white
squares in Figure 9 (a) below. These generate the semigroup S = NA , and the
proof of Theorem 1.3.5 gives the cone = Cone(e1 , e2 ), which is the first quadrant in the figure. At first glance, something seems wrong. The affine variety Cb2
is normal, yet in Figure 9 (a) the semigroup generated by the white squares misses
some lattice points in . This semigroup does not look saturated. How can the
affine toric variety be normal?
(a)
(b)
Figure 9. Lattice points for the rational normal cone Cb2
The problem is that we are using the wrong lattice! Proposition 1.1.8 tells us
to use the lattice ZA , which gives the white dots and squares in Figure 9 (b). This
39
figure shows that the white squares generate the semigroup of lattice points in .
Hence S is saturated and everything is fine.
This example points out the importance of working with the correct lattice.
The Normalization of an Affine Toric Variety. The normalization of an affine toric
variety is easy to describe. Let V = Spec(C[S]) for an affine semigroup S, so that
the torus of V has character lattice M = ZS. Let Cone(S) denote the cone of any
finite generating set of S and set = Cone(S) NR . In Exercise 1.3.6 you will
prove the following.
Proposition 1.3.8. The above cone is a strongly convex rational polyhedral cone
in NR and the inclusion C[S] C[ M] induces a morphism U V that is the
normalization map of V .

The normalization of an affine toric variety of the form YA is constructed by
applying Proposition 1.3.8 to the affine semigroup NA and the lattice ZA .
Example 1.3.9. Let A = {(4, 0), (3, 1), (1, 3), (0, 4)} Z2 . Then
A (s,t) = (s4 , s3 t, st 3 ,t 4 )
parametrizes the surface YA C4 considered in Exercise 1.1.7. This is almost

the rational normal
cone Cb4 , except that we have omitted s2 t 2 . Using (2, 2) =

1
2 (4, 0) + (0, 4) , we see that NA is not saturated, so that YA is not normal.
Applying Proposition 1.3.8, one sees that the normalization of YA is Cb4 . You
can check this using Normaliz [57] as explained in Example B.3.2. Note also that
Cb4 is an affine variety in C5 , and the normalization map is induced by the obvious
projection C5 C4 .
Proposition 1.3.8 tells us that M is the saturation of the semigroup S. In

the appendix to Chapter 3 we will see that the normalization map U V constructed in Proposition 1.3.8 is onto but not necessarily one-to-one.
Smooth Affine Toric Varieties. Our next goal is to characterize when an affine toric
variety is smooth. Since smooth affine varieties are normal (Proposition 1.0.9),
we need only consider toric varieties U coming from strongly convex rational
polyhedral cones NR .
We first study U when has maximal dimension. Then is strongly convex,

so that S = M has a Hilbert basis H . Furthermore, Corollary 1.3.3 tells us
that the torus action on U has a unique fixed point, denoted here by p U . The
point p and the Hilbert basis H are related as follows.
Lemma 1.3.10. Let NR be a strongly convex rational polyhedral cone of maximal dimension and let Tp (U ) be the Zariski tangent space to the affine toric
variety U at the above point p . Then dim Tp (U ) = |H |.
40
Proof. By Corollary 1.3.3, the maximal ideal of C[S ] corresponding to p is m =

h m | m S \ {0}i. Since { m }mS is a basis of C[S ], we obtain
M

M
M
M
m=
C m =
C m
C m =
C m m2 .
m6=0
m irreducible
m reducible
mH
dim m/m2
It follows that
= |H |. To relate this to the maximal ideal mU ,p in the
local ring OU ,p , we use the natural map
m/m2 mU ,p /mU2 ,p
which is always an isomorphism (Exercise 1.3.7). Since Tp (U ) is the dual space

of mU ,p /mU2 ,p , we see that dim Tp (U ) = |H |.

The Hilbert basis H of S gives U = YH Cs , where s = |H |. This affine
embedding is especially nice. Given any affine embedding U C , we have
dim Tp (U ) by Lemma 1.0.6. In other words, dim Tp (U ) is a lower bound
on the dimension of an affine embedding. Then Lemma 1.3.10 shows that when
has maximal dimension, the Hilbert basis of S gives the most efficient affine
embedding of U .
Example 1.3.11. In Example 1.2.22, we saw that the rational normal cone Cbd
Cd+1 is the toric variety coming from = Cone(de1 e2 , e2 ) R2 and that S =
Z2 is generated by (1, i) for 0 i d. These generators form the Hilbert basis
of S , so that the Zariski tangent space of 0 Cbd has dimension d + 1. Hence Cd+1
in the smallest affine space in which we can embed Cbd .
We now come to our main result about smoothness. Recall from 1.2 that a
rational polyhedral cone is smooth if it can be generated by a subset of a basis of
the lattice.
Theorem 1.3.12. Let NR be a strongly convex rational polyhedral cone. Then

U is smooth if and only if is smooth. Furthermore, all smooth affine toric
varieties are of this form.
Proof. If an affine toric variety is smooth, then it is normal by Proposition 1.0.9
and hence of the form U . Also, Example 1.2.21 implies that if is smooth as
a cone, then U is smooth as a variety. It remains to prove the converse. So fix
NR such that U is smooth. Let n = dim U = dim NR .
First suppose that has dimension n and let p U be the point studied in
Lemma 1.3.10. Since p is smooth in U , the Zariski tangent space Tp (U ) has
dimension n by Definition 1.0.7. On the other hand, Lemma 1.3.10 implies that
dim Tp (U ) is the cardinality of the Hilbert basis H of S = M. Thus
n = |H | |{edges }| n,
where the first inequality holds by Proposition 1.2.23 (each edge contributes an element of H ) and the second holds since dim = n. It follows
41
that has n edges and H consists of the ray generators of these edges. Since
M = ZS by (1.2.3), the n edge generators of generate the lattice M Zn and
hence form a basis of M. Thus is smooth, and then = ( ) is smooth since
duality preserves smoothness.
Next suppose dim = r < n. We reduce to the previous case as follows. Let
N1 N be the smallest saturated sublattice containing the generators of . Then
N/N1 is torsion-free, which by Exercise 1.3.5 implies the existence of a sublattice
N2 N with N = N1 N2 . Note rank N1 = r and rank N2 = n r.
The cone lies in both (N1 )R and NR . This gives affine toric varieties U,N1
and U,N of dimensions r and n respectively. Furthermore, N = N1 N2 induces
M = M1 M2 , so that (N1 )R and NR give the affine semigroups S,N1 M1
and S,N M respectively. It is straighforward to show that
S,N = S,N1 M2 ,
which in terms of semigroup algebras can be written

C[S,N ] C[S,N1 ] C C[M2 ].
The right-hand side is the coordinate ring of U,N1 TN2 . Thus

U,N U,N1 TN2 ,
(1.3.2)
which in turn implies that
U,N U,N1 (C )nr U,N1 Cnr .
Since we are assuming that U,N is smooth, it follows that U,N1 Cnr is smooth
at any point (p, q) in U,N1 (C )nr . In Exercise 1.3.8 you will show that
U,N1 Cnr is smooth at (p, q) = U,N1 is smooth at p.
(1.3.3)
Letting p = p U,N1 , the previous case implies that is smooth in N1 since

dim = dim (N1 )R . Hence is clearly smooth in N = N1 N2 .

Equivariant Maps between Affine Toric Varieties. We next study maps V1 V2
between affine toric varieties that respect the torus actions on V1 and V2 .
Definition 1.3.13. Let Vi = Spec(C[Si ]) be the affine toric varieties coming from
the affine semigroups Si , i = 1, 2. Then a morphism : V1 V2 is toric if the corresponding map of coordinate rings : C[S2 ] C[S1 ] is induced by a semigroup
homomorphism b : S2 S1 .
Here is our first result concerning toric morphisms.
Proposition 1.3.14. Let TNi be the torus of the affine toric variety Vi , i = 1, 2. Then:
(a) A morphism : V1 V2 is toric if and only if
(TN1 ) TN2
and |TN : TN1 TN2 is a group homomorphism.

1
42
(b) A toric morphism : V1 V2 is equivariant, meaning that

(t p) = (t) (p)
for all t TN1 and p V1 .
Proof. Let Vi = Spec(C[Si ]), so that the character lattice of TNi is Mi = ZSi . If
comes from a semigroup homomorphism b : S2 S1 , then b extends to a group
homomorphism b : M2 M1 and hence gives a commutative diagram
C[S2 ] C[S1 ]
C[M2 ] C[M1 ].
Applying Spec, we see that (TN1 ) TN2 , and |TN : TN1 TN2 is a group homo1
morphism since TNi = HomZ (Mi , C ) by Exercise 1.1.11. Conversely, if satisfies these conditions, then |TN : TN1 TN2 induces a diagram as above where
1
the bottom map comes from a group homomorphism b : M2 M1 . This, combined with (C[S2 ]) C[S1 ], implies that b induces a semigroup homomorphism
b : S2 S1 . This proves part (a) of the proposition.
For part (b), suppose that we have a toric map : V1 V2 . The action of TNi
on Vi is given by a morphism i : TNi Vi Vi , and equivariance means that we
have a commutative diagram
TN1 V1
/ V1
|T
N1
TN2 V2
/ V2 .
If we replace Vi with TNi in the diagram, then it certainly commutes since |TN
1
is a group homomorphism. Then the whole diagram commutes since TN1 TN1 is
Zariski dense in TN1 V1 .

We can also characterize toric morphisms between affine toric varieties coming
from strongly convex rational polyhedral cones. First note that a homomorphism
: N1 N2 of lattices gives a group homomorphism : TN1 TN2 of tori. This
follows from TNi = Ni Z C , and one sees that is a morphism. Also, tensoring
with R gives R : (N1 )R (N2 )R .
Here is the result, whose proof we leave to the reader (Exercise 1.3.9).
Proposition 1.3.15. Suppose we have strongly convex rational polyhedral cones

i (Ni )R and a homomorphism : N1 N2 . Then : TN1 TN2 extends to a

map of affine toric varieties : U1 U2 if and only if R (1 ) 2 .
43
Faces and Affine Open Subsets. Let NR be a strongly convex rational polyhedral cone and let be a face. Then we can find m M such that
= Hm . This allows us to relate the semigroup algebras of and as follows.
Proposition 1.3.16. Let be a face of and as above write = Hm , where
m M. Then the semigroup algebra C[S ] = C[ M] is the localization of
C[S ] = C[ M] at m C[S ]. In other words,
C[S ] = C[S ]m .
Proof. The inclusion implies S S , and since hm, ui = 0 for all u , we

have m . It follows that
S + Z(m) S .
This inclusion is actually an equality, as we now prove. Fix a finite set S N with
= Cone(S) and pick m S . Set
C = max{|hm , ui|} N.
uS
It is straightforward to show that m +Cm S . This proves that

S + Z(m) = S ,
from which C[S ] = C[S ] m follows immediately.
This interprets nicely in terms of toric morphisms. By Proposition 1.3.15, the

identity map N N and the inclusion give the toric morphism U U that
corresponds to the inclusion C[S ] C[S ]. By Proposition 1.3.16,
(1.3.4) U = Spec(C[S ]) = Spec(C[S ]m ) = Spec(C[S ])m = (U )m U .
Thus U becomes an affine open subset of U when . In particular, if two

cones and intersect in a common face = , then we have inclusions
U U U .
We will use this in Chapters 2 and 3 when we glue together affine toric varieties to
create more general toric varieties.
The role of faces is the key reason why we describe affine toric varieties using
NR rather than MR . This answers the question raised in Remark 1.2.19.
Sublattices of Finite Index and Rings of Invariants. Another interesting class of
toric morphisms arises when we keep the same cone but change the lattice. Here is
an example we have already seen.
Example 1.3.17. In Example 1.3.7 the dual of = Cone(e1 , e2 ) R2 interacts
with the lattices shown in Figure 10 on the next page. To make this precise, let us
name the lattices involved: the lattices
N = Z2 N = {(a/2, b/2) | a, b Z, a + b 0 mod 2}
44
(a)
(b)
Figure 10. Lattice points of relative to two lattices
have NR NR , and the dual lattices
M = Z2 M = {(a, b) | a, b Z, a + b 0 mod 2}
have MR MR . Note that duality reverses inclusions and that M and N are
indeed dual under dot product. In Figure 10 (a), the black dots in the first quadrant
form the semigroup S,N = M , and in Figure 10 (b), the white dots in the
first quadrant form S,N = M.
This gives the affine toric varieties U,N and U,N . Clearly U,N = C2 since
is smooth for N , while Example 1.3.7 shows that U,N is the rational normal cone
Cb2 . The inclusion N N gives a toric morphism
C2 = U,N U,N = Cb2 .
Our next task is to find a nice description of this map.
In general, suppose we have lattices N N, where N has finite index in N,

and let NR = NR be a strongly convex rational polyhedral cone. Then the
inclusion N N gives the toric morphism
: U,N U,N .
The dual lattices satisfy M M, so that corresponds to the inclusion

C[ M ] C[ M]
of semigroup algebras. The idea is to realize C[ M] as a ring of invariants of a

group action on C[ M ].
Proposition 1.3.18. Let N have finite index in N with quotient G = N/N and let
NR = NR be a strongly convex rational polyhedral cone. Then:
(a) There are natural isomorphisms
G HomZ (M /M, C ) = ker(TN TN ).
45
(b) G acts on C[ M ] with ring of invariants
C[ M ]G = C[ M].
(c) G acts on U,N , and the morphism : U,N U,N is constant on G-orbits
and induces a bijection
U,N /G U,N .
Proof. Since TN = HomZ (M, C ) by Exercise 1.1.11, applying HomZ (, C ) to
0 M M M /M 0
gives the sequence
1 HomZ (M /M, C ) TN TN 1.
This is exact since HomZ (, C ) is left exact and C is divisible. Note also that
since N has finite index in N, we have inclusions
N N NQ
M M MQ .
and
The pairing between M and N induces a pairing MQ NQ Q. Hence the map

M /M N/N C
([m ], [u]) 7 e2ihm ,ui
is well-defined and induces G = N/N HomZ (M /M, C ) (Exercise 1.3.10).
The action of TN on U,N induces an action of G on U,N since G TN .

Using Exercise 1.3.1, one sees that if g G and U,N , then g is defined by
the semigroup homomorphism m 7 g([m ])(m ) for m M . It follows that
the corresponding action on the coordinate ring is given by
g m = g([m ])1 m ,
m M .
(Exercise 5.0.1 explains why we need the inverse.) Since m M lies in M if and
only if g([m ]) = 1 for all g G, the ring of invariants
C[ M ]G = { f C[ M ] | g f = f for all g G},
is precisely C[ M], i.e.,
C[ M ]G = C[ M].
This proves part (b).

When a finite group G acts algebraically on Cn , [69, Thm. 10 of Ch. 7, 4]
shows that the ring of invariants C[x1 , . . . , xn ]G C[x1 , . . . , xn ] gives a morphism of
affine varieties
Cn = Spec(C[x1 , . . . , xn ]) Spec(C[x1 , . . . , xn ]G )
that is constant on G-orbits and induces a bijection
Cn /G Spec(C[x1 , . . . , xn ]G ).
46
The proof extends without difficulty to the case when G acts algebraically on V =
Spec(R). Here, RG R gives a morphism of affine varieties
V = Spec(R) Spec(R G )
that is constant on G-orbits and induces a bijection

V /G Spec(R G ).
From here, part (c) follows immediately from part (b).
We will give a careful treatment of these ideas in 5.0, where we will show
that the map Spec(R) Spec(R G ) is a geometric quotient.
Here are some examples of Proposition 1.3.18.
Example 1.3.19. In the situation of Example 1.3.17, one computes that G is the
group 2 = {1} acting on U,N = Spec(C[s,t]) C2 by 1 (s,t) = (s, t).
Thus the rational normal cone Cb2 is the quotient
C2 /2 = U,N /2 U,N = Cb2 .
We can see this explicitly as follows. The invariant ring is easily seen to be
C[s,t]2 = C[s 2 , st,t 2 ] = C[Cb2 ] C[x0 , x1 , x2 ]/hx0 x2 x12 i,
where the last isomorphism follows from Example 1.1.6. From the point of view
of invariant theory, the generators s 2 , st,t 2 of the ring of invariants give a morphism
: C2 C3 ,
(s,t) (s 2 , st,t 2 )
that is constant on 2 -orbits. This map also separates orbits, so it induces

C2 /2 (C2 ) = Cb2 ,
where the last equality is by Example 1.1.6. But we can also think about this in
terms of semigroups, where the exponent vectors of s 2 , st,t 2 give the Hilbert basis
of the semigroup S,N pictured in Figure 10 (b). Everything fits together very
nicely.
In Exercise 1.3.11 you will generalize Example 1.3.19 to the case of the rational normal cone Cbd for arbitrary d.
Example 1.3.20. Let NR RnPbe a simplicial cone of dimension n with ray

generators u1 , . . . , un . Then N = ni=1 Zui is a sublattice of finite index in N.
Furthermore, is smooth relative to N , so that U,N = Cn . It follows that G =
N/N acts on Cn with quotient
Cn /G = U,N /G U,N .
Hence the affine toric variety of a simplicial cone is the quotient of affine space by
a finite abelian group. In the literature, varieties like U,N are called orbifolds and
are said to be Q-factorial.
47
Exercises for 1.3.

1.3.1. Consider the affine toric variety YA = Spec(C[S]), where A = {m1 , . . . , ms } and
S = NA . Let : S C be a semigroup homomorphism. In the proof of Proposition 1.3.1
we showed that p = ((m1 ), . . . , (ms )) lies in YA .
(a) Prove that the maximal ideal { f C[S] | f (p) = 0} is the kernel of the C-algebra
homomorphism C[S] C induced by .
(b) The torus TN of YA has character lattice M = ZA and fix t TN . As in the discussion
following Proposition 1.3.1, t p comes from the semigroup homomorphism m 7
m (t)(m). Prove that this corresponds to the point
( m1 (t), . . . , ms (t)) ((m1 ), . . . , (ms )) = ( m1 (t)(m1 ), . . . , ms (t)(ms )) Cs
coming from the action of t TN (C )s on p YA Cs .
1.3.2. Let V = Spec(C[S]) with TN = Spec(C[M]), M = ZS. The action TN V V

comes from a C-algebra homomorphism C[S] C[M] C C[S]. Prove that this homomorphism is given by m 7 m m . Hint: Show that this formula determines the C-algebra
homomorphism C[M] C[M] C C[M] that gives the group operation TN TN TN .
1.3.3. Prove Corollary 1.3.3.
1.3.4. Let A M be a finite set.
(a) Prove that the semigroup NA is saturated in M if and only if NA = Cone(A ) M.

Hint: Apply (1.2.2) to Cone(A ) MR .
(b) Complete the proof of (b) (c) from Theorem 1.3.5.
1.3.5. Let N be a lattice.

(a) Let N1 N be a sublattice such that N/N1 is torsion-free. Prove that there is a sublattice N2 N such that N = N1 N2 .
(b) Let u N be primitive as defined in the proof of Theorem 1.3.5. Prove that N has a
basis e1 , . . . , en such that e1 = u.
1.3.7. Let p be a point of an irreducible affine variety V . Then p gives the maximal ideal
m = { f C[V ] | f (p) = 0} as well as the maximal ideal mV ,p OV ,p defined in 1.0.
Prove that the natural map m/m2 mV ,p /mV2 ,p is an isomorphism of C-vector spaces.
1.3.8. Prove (1.3.3). Hint: Use Lemma 1.0.6 and Example 1.0.10.
1.3.10. Prove the assertions made in the proof of Proposition 1.3.18 concerning the pairing
M /M N/N C defined by ([m ], [u]) 7 e2ihm ,ui .
1.3.11. Let d = { C | d = 1} be the group of dth roots of unity. Then d acts on C2

by (x, y) = (x, y). Adapt Example 1.3.19 to show that C2 /d Cbd . Hint: Use lattices
N = Z2 N = {(a/d, b/d) | a, b Z, a + b 0 mod d}.
1.3.12. Prove that the normalization map in Proposition 1.3.8 is a toric morphism.
1.3.13. Let 1 (N1 )R and 2 (N2 )R be strongly convex rational polyhedral cones. This
gives the cone 1 2 (N1 N2 )R . Prove that U1 2 U1 U2 . Also explain how
this result applies to (1.3.2).
48
1.3.14. By Proposition 1.3.1, a point p of an affine toric variety V = Spec(C[S]) is represented by a semigroup homomorphism : S C. Prove that p lies in the torus of V if and
only if never vanishes, i.e., (m) 6= 0 for all m S.
Appendix: Tensor Products of Coordinate Rings

In this appendix, we will prove the following result used in 1.0 in our discussion of products of affine varieties.
Proposition 1.A.1. If R and S are finitely generated C-algebras without nilpotents, then
the same is true for R C S.
Proof. The tensor product is clearly a finitely generated C-algebra. Hence we need only
prove that R C S has no nilpotents. IfTwe write R C[x1 , . . . , xn ]/I, then I is radical and
s
thus has a primary decomposition I = i=1 Pi , where each Pi is prime (see [69, Ch. 4, 7]).
This gives
s
M
C[x1 , . . . , xn ]/Pi
R C[x1 , . . . , xn ]/I
i=1
where the map to the direct sum is injective. Each quotient C[x1 , . . . , xn ]/Pi is an integral
domain and hence injects into its field of fractions Ki . This yields an injection
s
M
R
Ki ,
i=1
and since tensoring over a field preserves exactness, we get an injection

s
M
R C S
Ki C S.
i=1
Hence it suffices to prove that K C S has no nilpotents when K is a finitely generated field
extension of C. A similar argument using S then reduces us to showing that K C L has no
nilpotents when K and L are finitely generated field extensions of C.
Since C has characteristic 0, the extension C L has a separating transcendence basis
(see [159, p. 519]). This means that we can find y1 , . . . , yt L such that y1 , . . . , yt are
algebraically independent over C and F = C(y1 , . . . , yt ) L is a finite separable extension.
Then
K C L K C (F F L) (K C F) F L.
But C = K C F = K C C(y1 , . . . , yt ) = K(y1 , . . . , yt ) is a field, so that we are reduced to
considering
C F L
where C and L are extensions of F and F L is finite and separable. The latter and the
theorem of the primitive element imply that L F[X]/h f (X)i, where f (X) has distinct
roots in some extension of F. Then
C F L C F F[X]/h f (X)i C[X]/h f (X)i.
Since f (X) has distinct roots, this quotient ring has no nilpotents. Our result follows.
A final remark is that we can replace C with any perfect field since finitely generated
extensions of perfect fields have separating transcendence bases (see [159, p. 519]).
Chapter 2
Projective Toric Varieties
2.0. Background: Projective Varieties

Our discussion assumes that the reader is familiar with the elementary theory of
projective varieties, at the level of [69, Ch. 8].
In Chapter 1, we introduced affine toric varieties. In general, a toric variety is
an irreducible variety X over C containing a torus TN (C )n as a Zariski open set
such that the action of (C )n on itself extends to an action on X . We will learn in
Chapter 3 that the concept of variety is somewhat subtle. Hence we will defer
the formal definition of toric variety until then and instead concentrate on toric
varieties that live in projective space P n , defined by
(2.0.1)
P n = (Cn+1 \ {0})/C ,
where C acts via homotheties, i.e., (a0 , . . . , an ) = (a0 , . . . , an ) for C and

(a0 , . . . , an ) Cn+1 . Thus (a0 , . . . , an ) are homogeneous coordinates of a point in
P n and are well-defined up to homothety.
The goal of this chapter is to use lattice points and polytopes to create toric
varieties that lie in P n . We will use the affine semigroups and polyhedral cones
introduced in Chapter 1 to describe the local structure of these varieties.
Homogeneous Coordinate Rings. A projective variety V P n is defined by the
vanishing of finitely many homogeneous polynomials in the polynomial ring S =
C[x0 , . . . , xn ]. The homogeneous coordinate ring of V is the quotient ring
C[V ] = S/I(V ),
where I(V ) is generated by all homogeneous polynomials that vanish on V .
49
50
Chapter 2. Projective Toric Varieties
The polynomial
ring S is graded by setting deg(xi ) = 1. This gives the decomL
position S =
S
d=0 d , where Sd is the vector space of homogeneous polynomials
of degree d. Homogeneous ideals decompose similarly, and the above coordinate
ring C[V ] inherits a grading where
C[V ]d = Sd /I(V )d .
The ideal I(V ) S = C[x0 , . . . , xn ] also defines an affine variety Vb Cn+1 , called
the affine cone of V . The variety Vb satisfies
(2.0.2)
V = (Vb \ {0})/C ,
and its coordinate ring is the homogeneous coordinate ring of V , i.e.,

C[Vb ] = C[V ].
Example 2.0.1. In Example 1.1.6 we encountered the ideal

I = hxi x j+1 xi+1 x j | 0 i < j d 1i C[x0 , . . . , xd ]
generated by the 2 2 minors of the matrix

x0 x1 xd2 xd1
.
x1 x2 xd1 xd
Since I is homogeneous, it defines a projective variety Cd P d that is the image of
the map
: P1 P d
defined in homogeneous coordinates by (s,t) 7 (s d , s d1 t, . . . , st d1 ,t d ) (see Exercise 1.1.1). This shows that Cd is a curve, called the rational normal curve of
degree d. Furthermore, the affine cone of Cd is the rational normal cone Cbd Cd+1
discussed in Example 1.1.6.
We know from Chapter 1 that Cbd is an affine toric surface; we will soon see
that Cd is a projective toric curve.
Example 2.0.2. The affine toric variety V(xy zw) C4 studied in Chapter 1 is
the affine cone of the projective surface V = V(xy zw) P 3 . Recall that this
surface is isomorphic to P1 P1 via the Segre embedding
P1 P1 P 3
given by (s,t; u, v) 7 (su,tv, sv,tu). We will see below that V P1 P1 is the

projective toric variety coming from the unit square in the plane.
As in the affine case, a projective variety V P n has the classical topology,

induced from the usual topology on P n , and the Zariski topology, where the Zariski
closed sets are subvarieties of V (meaning projective varieties of P n contained in V )
and the Zariski open sets are their complements.
51
Rational Functions on Irreducible Projective Varieties. A homogeneous polynomial f S of degree d > 0 does not give a function on P n since
f (x0 , . . . , xn ) = d f (x0 , . . . , xn ).
However, the quotient of two such polynomials f , g Sd gives the well-defined

function
f
: P n \ V(g) C.
g
provided g 6= 0. We write this as f /g : P n 99K C and say that f /g is a rational
function on P n .
More generally, suppose that V P n is irreducible, and let f , g C[V ] = C[Vb ]
be homogeneous of the same degree with g 6= 0. Then f and g give functions on
the affine cone Vb and hence an element f /g C(Vb ). By (2.0.2), this induces a
rational function f /g : V 99K C. Thus
C(V ) = { f /g C(Vb ) | f , g C[V ] homogeneous of the same degree, g 6= 0}
is the field of rational functions on V . It is customary to write the set on the left as
C(Vb )0 since it consists of the degree 0 elements of C(Vb ).
Affine Pieces of Projective Varieties. A projective variety V P n is a union of
Zariski open sets that are affine. To see why, let Ui = P n \ V(xi ). Then Ui Cn via
the map

a
ai+1
an
(2.0.3)
(a0 , . . . , an ) 7 aa0i , . . . , i1
ai , ai , . . . , ai ,
so that in the notation of Chapter 1, we have

x
xi+1
xn
Ui = Spec C xx0i , . . . , i1
xi , xi , . . . , xi .
Then V Ui is a Zariski open subset of V that maps via (2.0.3) to the affine variety
in Cn defined by the equations

x
xi+1
xn
(2.0.4)
f xx0i , . . . , i1
xi , 1, xi , . . . , xi = 0
as f varies over all homogeneous polynomials in I(V ).
We call V Ui an affine piece of V . These affine pieces cover V since the Ui

cover P n . Using localization, we can describe the coordinate rings of the affine
pieces as follows. The variable xi induces an element xi C[V ], so that we get the
localization
(2.0.5)
C[V ]xi = { f /
xik | f C[V ], k 0}
as in Exercises 1.0.2 and 1.0.3. Note that C[V ]xi has a well-defined Z-grading given
by deg( f /
xik ) = deg( f ) k when f is homogeneous. Then
(2.0.6)
(C[V ]xi )0 = { f /
xik C[V ]xi | f is homogeneous of degree k}
is the subring of C[V ]xi consisting of all elements of degree 0. This gives an affine
piece of V as follows.
52
Lemma 2.0.3. The affine piece V Ui of V has coordinate ring

C[V Ui] (C[V ]xi )0 .
Proof. We have an exact sequence

0 I(V ) C[x0 , . . . , xn ] C[V ] 0.
If we localize at xi , we get the exact sequence
(2.0.7)
0 I(V )xi C[x0 , . . . , xn ]xi C[V ]xi 0
since localization preserves exactness (Exercises 2.0.1 and 2.0.2). These sequences
preserve degrees, so that taking elements of degree 0 gives the exact sequence
0 (I(V )xi )0 (C[x0 , . . . , xn ]xi )0 (C[V ]xi )0 0.

x
xi+1
xn
,
,
.
.
.
,
Note that (C[x0 , . . . , xn ]xi )0 = C xx0i , . . . , i1
xi
xi
xi . If f I(V ) is homogeneous of degree k, then

x
xi+1
xn
f /xik = f xx0i , . . . , i1
xi , 1, xi , . . . , xi (I(V )xi )0 .
By (2.0.4), we conclude that (I(V )xi)0 maps to I(V Ui ). To show that this map
x
xi+1
xn
k
is onto, let g xx0i , . . . , i1
xi , xi , . . . , xi I(V Ui ). For k 0, xi g = f (x0 , . . . , xn )
is homogeneous of degree k. It then follows easily that xi f vanishes on V since
g = 0 on V Ui and xi = 0 on the complement of Ui . Thus xi f I(V ), and then
(xi f )/(xik+1 ) (I(V )xi )0 maps to g. The lemma follows immediately.
One can also explore what happens when we intersect affine pieces V Ui and
V U j for i 6= j. By Exercise 2.0.3, V Ui U j is affine with coordinate ring
(2.0.8)
C[V Ui U j ] (C[V ]xi xj )0 .
We will apply this to projective toric varieties in 2.2. We will also see later in
the book that Lemma 2.0.3 is related to the Proj construction, where Proj of a
graded ring gives a projective variety, just as Spec of an ordinary ring gives an
affine variety.
Products of Projective Spaces. One can study the product P n P m of projective
spaces using the bigraded ring C[x0 , . . . , xn , y0 , . . . , ym ], where xi has bidegree (1, 0)
and yi has bidegree (0, 1). Then a bihomogeneous polynomial f of bidegree (a, b)
gives a well-defined equation f = 0 in P n P m . This allows us to define varieties
in P n P m using bihomogeneous ideals. In particular, the ideal I(V ) of a variety
V P n P m is a bihomogeneous ideal.
Another way to study P n P m is via the Segre embedding
P n P m P nm+n+m
defined by mapping (a0 , . . . , an , b0 , . . . , bm ) to the point

(a0 b0 , a0 b1 , . . . , a0 bm , a1 b0 , . . . , a1 bm , . . . , an b0 , . . . , an bm ).
53
This map is studied in [69, Ex. 14 of Ch. 8, 4]. If P nm+n+m has homogeneous
coordinates xi j for 0 i n, 0 j m, then P n P m P nm+n+m is defined by the
vanishing of the 2 2 minors of the (n + 1) (m + 1) matrix
x00 x0m
..
.. .
.
.
xn0 xnm
This follows since an (n + 1) (m + 1) matrix has rank 1 if and only if it is a

product At B, where A and B are nonzero row matrices of lengths n + 1 and m + 1.
These approaches give the same notion of what it means to be a subvariety of

A homogeneous polynomial F(xi j ) of degree d gives the bihomogeneous
polynomial F(xi y j ) of bidegree (d, d). Hence any subvariety of P nm+n+m lying
in P n P m can be defined by a bihomogeneous ideal in C[x0 , . . . , xn , y0 , . . . , ym ].
Going the other way takes more thought and is discussed in Exercise 2.0.5.
P n P m.
We also have the following useful result proved in Exercise 2.0.6.

Proposition 2.0.4. Given subvarieties V P n and W P m , the product V W is
a subvariety of P n P m .

Weighted Projective Space. The graded ring associated to projective space P n is
C[x0 , . . . , xn ], where each variable xi has degree 1. More generally, let q0 , . . . , qn be
positive integers with gcd(q0 , . . . , qn ) = 1 and define
P(q0 , . . . , qn ) = (Cn+1 \ {0})/ ,
where is the equivalence relation
(a0 , . . . , an ) (b0 , . . . , bn ) ai = qi bi , i = 0, . . . , n for some C .
We call P(q0 , . . . , qn ) a weighted projective space. Note that P n = P(1, . . . , 1).

The ring corresponding to P(q0 , . . . , qn ) is the graded ring C[x0 , . . . , xn ], where
xi now has degree qi . A polynomial f is weighted homogeneous of degree d if every
monomial x appearing in f satisfies (q0 , . . . , qn ) = d. The equation f = 0 is
well-defined on P(q0 , . . . , qn ) when f is weighted homogeneous, and one can define
varieties in P(q0 , . . . , qn ) using weighted homogeneous ideals of C[x0 , . . . , xn ].
Example 2.0.5. We can embed the weighted projective plane P(1, 1, 2) in P 3 using
the monomials x02 , x0 x1 , x12 , x2 of weighted degree 2. In other words, the map
P(1, 1, 2) P 3
given by
(a0 , a1 , a2 ) 7 (a02 , a0 a1 , a12 , a2 )
is well-defined and injective. One can check that this map induces
P(1, 1, 2) V(y0 y2 y12 ) P 3 ,
where y0 , y1 , y2 , y3 are homogeneous coordinates on P 3 .
54
Later in the book we will use toric methods to construct projective embeddings
of arbitrary weighted projective spaces.
Exercises for 2.0.
2.0.1. Let R be a commutative C-algebra. Given f R \ {0} and an exact sequence of
R-modules 0 M1 M2 M3 0, prove that
0 M1 R R f M2 R R f M3 R R f 0
is also exact, where R f is the localization of R at f defined in Exercises 1.0.2 and 1.0.3.
2.0.2. Let V P n be a projective variety. If we set S = C[x0 , . . . , xn ], then V has coordinate
ring C[V ] = S/I(V ). Let xi be the image of xi in C[V ].
(a) Note that C[V ] is an S-module. Prove that C[V ]xi C[V ] S Sxi .
(b) Use part (a) and the previous exercise to prove that (2.0.7) is exact.
2.0.3. Prove the claim made in (2.0.8).

2.0.4. Let V P n be a projective variety. Take f0 , . . . , fm Sd such that the intersection
V V( f0 , . . . , fm ) is empty. Prove that the map
(a0 , . . . , an ) 7 ( f0 (a0 , . . . , an ), . . . , fm (a0 , . . . , an ))
induces a well-defined function : V P m .
2.0.5. Let V P n P m be defined by f (xi , y j ) = 0, where f (xi , y j ) is bihomogenous of

bidegree (a , b ), = 1, . . . , s. The goal of this exercise is to show that when we embed
P n P m in P nm+n+m via the Segre embedding described in the text, V becomes a subvariety
of P nm+n+m .
(a) For each , pick an integer d max{a , b } and consider the polynomials f,, =
x y f (xi , y j ) where = 1, . . . , s and || = d a , || = d b . Note that f,, is
bihomogenous of bidegree (d , d ). Prove that V P n P m is defined by the vanishing
of the f,, .
(b) Use part (a) to show that V is a subvariety of P nm+n+m under the Segre embedding.
2.0.6. Prove Proposition 2.0.4
2.0.7. Consider the Segre embedding P1 P1 P 3 . Show that after relabeling coordinates, the affine cone of P1 P1 in P 3 is the variety V(xy zw) C4 featured in many
examples in Chapter 1.
2.1. Lattice Points and Projective Toric Varieties

We first observe that P n is a toric variety with torus
TPn = P n \ V(x0 xn ) = {(a0 , . . . , an ) P n | a0 an 6= 0}
= {(1,t1 , . . . ,tn ) P n | t1 , . . . ,tn C } (C )n .
The action of TPn on itself clearly extends to an action on P n , making P n a toric

variety. To describe the lattices associated to TPn , we use the exact sequence of tori
1 C (C )n+1 TPn 1
55
coming from the definition (2.0.1) of P n . Hence the character lattice of TPn is
P
(2.1.1)
Mn = {(a0 , . . . , an ) Zn+1 | ni=0 ai = 0},
and the lattice of one-parameter subgroups Nn is the quotient
Nn = Zn+1 /Z(1, . . . , 1).
Lattice Points and Projective Toric Varieties. Let TN be a torus with lattices M
and N as usual. In Chapter 1, we used a finite set of lattice points of A =
{m1 , . . . , ms } M to create the affine toric variety YA as the Zariski closure of
the image of the map
A : TN Cs ,
t 7 ( m1 (t), . . . , ms (t)).
To get a projective toric variety, we regard A as a map to (C )s and compose

with the homomorphism : (C )s TPs1 to obtain
(2.1.2)
A
TN
Cs TPs1 P s1 .
Definition 2.1.1. Given a finite set A M, the projective toric variety XA is the
Zariski closure in P s1 of the image of the map A from (2.1.2).
Proposition 2.1.2. XA is a toric variety of dimension equal to the dimension of the
smallest affine subspace of MR containing A .
Proof. The proof that XA P s1 is a toric variety is similar to the proof given
in Proposition 1.1.8 of Chapter 1 that YA Cs is a toric variety. The assertion
concerning the dimension of XA will follow from Proposition 2.1.6 below.

More concretely, XA is the Zariski closure of the image of the map
TN P s1 ,
t 7 ( m1 (t), . . . , ms (t))
given by the characters coming from A = {m1 , . . . , ms } M. In particular, if

M = Zn , then mi is the Laurent monomial t mi and XA is the Zariski closure of the
image of
TN P s1 , t 7 (t m1 , . . . ,t ms ).
In the literature, A Zn is often given as an n s matrix A with integer entries, so
that the elements of A are the columns of A.
Here is an example where the lattice points themselves are matrices.
Example 2.1.3. Let M = Z33 be the lattice of 3 3 integer matrices and let
P3 = {3 3 permutation matrices} Z33 .
Write C[M] = C[t11 , . . . ,t91 ], where the variables give the generic 3 3 matrix
t1 t2 t3
t4 t5 t6
t7 t8 t9
56
5
with nonzero entries. Also
let P have homogeneous coordinates 5xi jk indexed by
123
triples such that i j k is a permutation in S3 . Then XP3 P is the Zariski
closure
of the image of the map TN P 5 given by the Laurent monomials tit j tk for

123
i j k S3 . The ideal of XP3 is
I(XP3 ) = hx123 x231 x312 x132 x321 x213 i C[xi jk ],
where the relation comes from the fact that the sum of the permutation matrices
corresponding to x123 , x231 , x312 equals the sum of the other three (Exercise 2.1.1).
Ideals of the toric varieties arising from permutation matrices have applications to
sampling problems in statistics (see [264, p. 148]).
The Affine Cone of a Projective Toric Variety. The projective variety XA P s1

has an affine cone XbA Cs . How does XbA relate to the affine toric variety YA Cs
constructed in Chapter 1?
Recall from Chapter 1 that when A = {m1 , . . . , ms } M, the map ei 7 mi

induces an exact sequence
0 L Zs M
(2.1.3)
and that the ideal of YA is the toric ideal

IL = x x | , Ns and L
(Proposition 1.1.9). Then we have the following result.
Proposition 2.1.4. Given YA , XA and IL as above, the following are equivalent:

(a) YA Cs is the affine cone XbA of XA P s1 .
(b) IL = I(XA ).
(c) IL is homogeneous.
(d) There is u N and k > 0 in N such that hmi , ui = k for i = 1, . . . , s.

Proof. The equivalence (a) (b) follows from the equalities I(XA ) = I(XbA ) and
IL = I(YA ), and the implication (b) (c) is obvious.
For (c) (d), assume that IL is a homogeneous ideal and take x x IL

for L. If x and x had different degrees, then x , x IL = I(YA ) would
vanish on YA . This is impossible since (1, . . . , 1) YA by (2.1.2). Hence x and
x have the same degree, which implies that (1, . . . , 1) = 0 for all L. Now
tensor (2.1.3) with Q and take duals to obtain an exact sequence
NQ Qs HomQ (LQ , Q) 0.
The above argument shows that (1, . . . , 1) Qs maps to zero in HomQ (LQ , Q) and
hence comes from an element u NQ . In other words, hmi , ui = 1 for all i. Clearing
denominators gives the desired u N and k > 0 in N.
57
Finally, we prove (d) (a). Since YA XbA and XbA is irreducible, it suffices
to show that
XbA (C )s YA .
Let p XbA (C )s . Since XA TPs1 is the torus of XA , it follows that
p = ( m1 (t), . . . , ms (t))
for some C and t TN . The element u N from part (d) gives a one-parameter
subgroup of TN , which we write as 7 u ( ) for C . Then u ( )t TN maps
to the point q YA given by

q = m1 (u ( )t), . . . , ms (u ( )t) = hm1 ,ui m1 (t), . . . , hms ,ui ms (t) ,
since m (u ( )) = hm,ui by the description of h , i given in 1.1. The hypothesis

of part (d) allows us to rewrite q as
q = k ( m1 (t), . . . , ms (t)).
Using k > 0, we can choose so that p = q YA . This completes the proof.
The condition hmi , ui = k, i = 1, . . . , s, for some u N and k > 0 in N means

that A lies in an affine hyperplane of MQ not containing the origin. When M = Zn
and A consists of the columns of an n s integer matrix A, this is equivalent to
(1, . . . , 1) lying in the row space of A (Exercise 2.1.2).
Example 2.1.5. We will examine the rational normal curve Cd P d using two
different sets of lattice points.
First let A consist of the columns of the 2 (d + 1) matrix

d d 1
1
0
.
A=
0
1
d 1 d
The columns give the Laurent monomials defining the rational normal curve Cd
in Example 2.0.1. It follows that Cd is a projective toric variety. The ideal of Cd
is the homogeneous ideal given in Example 2.0.1, and the corresponding affine
hyperplane of Z2 containing A (= the columns of A) consists of all points (a, b)
satisfying a + b = d. It is equally easy to see that (1, . . . , 1) is in the row space of
A. In particular, we have
XA = Cd
and YA = Cbd .
Now let B = {0, 1, . . . , d 1, d} Z. This gives the map

B : C P d ,
t 7 (1,t, . . . ,t d1 ,t d ).
The resulting projective variety is the rational normal curve, i.e., XB = Cd , but
the affine variety of B is not the rational normal cone, i.e., YB 6= Cbd . This is
because I(YB ) C[x0 , . . . , xd ] is not homogeneous. For example, x12 x2 vanishes
at (1,t, . . . ,t d1 ,t d ) Cd+1 for all t C . Thus x12 x2 I(YB ).
58
Given any A M, there is a standard way to modify A so that the conditions

of Proposition 2.1.4 are met: use A {1} M Z. This lattice corresponds to the
torus Tn C , and since
(2.1.4)
A {1} (t, ) = ( m1 (t) , . . . , ms (t) ) = ( m1 (t), . . . , ms (t)),
it follows immediately that XA {1} = XA P s1 . Since A {1} lies in an affine

hyperplane missing the origin, Proposition 2.1.4 implies that XA has affine cone
YA {1} = XbA . When M = Zn and A is represented by the columns of an n s
integer matrix A, we obtain A {1} by adding the row (1, . . . , 1) to A.
The Torus of a Projective Toric Variety. Our next task is to determine the torus
of XA . We will do so by identifying its character lattice. This will also tell us the
dimension of XA . Given A = {m1 , . . . , ms } M, we set
Ps

Ps
ZA =
i=1 ai mi | ai Z,
i=1 ai = 0 .
The rank of ZA is the dimension of the smallest affine subspace of MR containing

the set A (Exercise 2.1.3).
Proposition 2.1.6. Let XA be the projective toric variety of A M. Then:
(a) The lattice ZA is the character lattice of the torus of XA .
(b) The dimension of XA is the dimension of the smallest affine subspace of MR

containing A . In particular,
(
rank ZA 1 if A satisfies the conditions of Proposition 2.1.4
dim XA =
rank ZA
otherwise.
Proof. To prove part (a), let M be the character lattice of the torus TXA of XA . By
(2.1.2), we have the commutative diagram
TN D
D
/ TPs1
O
/ P s1
DD
DD
DD!
! ?
TXA
which induces the commutative diagram of character lattices

M bEo
Ms1
EE
EE
EE
EE
1Q
Ps
since Ms1 = {(a1 , . . . , as ) Zs | i=0 ai = 0} is the character lattice of TPs1 by

(2.1.1). The map Ms1 M is induced by the map Zs M that sends ei to mi .
Thus ZA is the image of Ms1 M and hence equals M by the above diagram.
The first assertion of part (b) follows from part (a) and the observation that
rank ZA is the dimension of the smallest affine subspace of MR containing A .
59
Furthermore, if YA equals the affine cone of XA , then P

there is u N with P
hmi , ui =
k > 0 for all i by Proposition 2.1.4. This implies that h si=1 ai mi , ui = k( si=1 ai ),
which gives the exact sequence
h ,ui
0 ZA ZA kZ 0.
Then k > 0 implies rank ZA 1 = rank ZA = dim XA . However, if YA 6= XbA ,

then the ideal IL is not homogeneous. Thus some generator x y is not homogeneous, so that ( ) (1, . . . , 1) 6= 0. But L, where L is defined by
0 L Zs ZA 0.
This implies that in the exact sequence

0 Ms1 Zs Z 0
(see (2.1.1)), the image of L Zs is Z Z for some > 0. This gives a diagram
0
0
0
/L

/ Z

/Z
/ ZA

/ Z/Z

/ L Ms1

/ Ms1

/ ZA

/ Zs

/0
/0
/0
with exact rows and columns. Hence rank ZA = rank ZA = dim XA .
Example 2.1.7. Let A = {e1 , e2 , e1 + 2e2 , 2e1 + e2 } Z2 . One computes that

ZA = Z2 but ZA = {(a, b) Z2 | a + b 0 mod 2}. Thus ZA has index 2 in
ZA . This means that YA 6= XbA and the map of tori
TYA TXA
is two-to-one, i.e., its kernel has order 2 (Exercise 2.1.4).
Affine Pieces of a Projective Toric Variety. So far, our treatment of projective toric
varieties has used lattice points and toric ideals. Where are the semigroups? There
are actually lots of semigroups, one for each affine piece of XA P s1 .
The affine open set Ui = P s1 \ V(xi ) contains the torus TPs1 . Thus
TXA = XA TPs1 XA Ui.
Since XA is the Zariski closure of TXA in P s1 , it follows that XA Ui is the Zariski

closure of TXA in Ui Cs1 . Thus XA Ui is an affine toric variety.
Given A = {m1 , . . . , ms } MR , the affine semigroup associated to XA Ui is

easy to determine. Recall that Ui Cs1 is given by
(a1 , . . . , as ) 7 (a1 /ai , . . . , ai1 /ai , ai+1 /ai , . . . , as /ai ).
60
Combining this and m j / mi = m j mi with the map (2.1.2), we see that XA Ui

is the Zariski closure of the image of the map
TN Cs1
given by
(2.1.5)

t 7 m1 mi (t), . . . , mi1 mi (t), mi+1 mi (t), . . . , ms mi (t) .
If we set Ai = A mi = {m j mi | j 6= i}, it follows that

XA Ui = YAi = Spec(C[Si ]),
where Si = NAi is the affine semigroup generated by Ai . We have thus proved the
following result.
Proposition 2.1.8. Let XA P s1 for A = {m1 , . . . , ms } MR . Then the affine
piece XA Ui is the affine toric variety
XA Ui = YAi = Spec(C[Si ])
where Ai = A mi and Si = NAi .
We also note that the results of Chapter 1 imply that the torus of YAi has character lattice ZAi . Yet the torus is TXA , which has character lattice ZA by Proposition 2.1.6. These are consistent since ZAi = ZA for all i.
Besides describing the affine pieces XA Ui of XA P s1 , we can also describe how they patch together. In other words, we can give a completely toric
description of the inclusions
XA Ui XA Ui U j XA U j
when i 6= j. For instance, Ui U j consists of all points of XA Ui where x j /xi 6= 0.

In terms of XA Ui = Spec(C[Si ]), this means those points where m j mi 6= 0.
Thus

(2.1.6)
XA Ui U j = Spec C[Si ] m j mi = Spec C[Si ] m j mi ,
so that if we set m = m j mi , then the inclusion XA Ui U j XA Ui can be

written
(2.1.7)
Spec(C[Si ]) m Spec(C[Si ]).
This looks very similar to the inclusion constructed in (1.3.4) using faces of cones.
We will see in 2.3 that this is no accident.
We next say a few words about how the polytope P = Conv(A ) MR relates
to XA . As we will learn in 2.2, the dimension of P is the dimension of the smallest affine subspace of MR containing P, which is the same as the smallest affine
subspace of MR containing A . It follows from Proposition 2.1.6 that
dim XA = dim P.
Furthermore, the vertices of P give an especially efficient affine covering of XA .
61
Proposition
2.1.9. Given A = {m1 , . . . , ms } M, let P = Conv(A ) MR and set

J = j {1, . . . , s} | m j is a vertex of P . Then
[
XA =
XA U j .
jJ
Proof. We will prove that if i {1, . . . , s}, then XA Ui XA U j for some j J.

The discussion of polytopes from 2.2 below implies that
P

P
P MQ =
jJ r j m j | r j Q0 ,
jJ r j = 1 .
Given i {1, . . . , s}, we have mi P M, so that mi is a convex Q-linear combination of the vertices m j . Clearing denominators, we get integers k > 0 and k j 0
such that
P
P
kmi = jJ k j m j ,
jJ k j = k.
P
Thus
jJ k j (m j mi ) = 0, which implies that mi m j Si when k j > 0. Fix
such a j. Then m j mi C[Si ] is invertible, so C[Si ] m j mi = C[Si ]. By (2.1.6),
XA Ui U j = Spec(C[Si ]) = XA Ui , giving XA Ui XA U j .

Projective Normality. An irreducible variety V P n is called projectively normal
if its affine cone Vb Cn+1 is normal. A projectively normal variety is always
normal (Exercise 2.1.5). Here is an example to show that the converse can fail.
Example 2.1.10. Let A Z2 consist of the columns of the matrix

4 3 1 0
,
0 1 3 4
giving the Laurent monomials s4 , s3 t, st 3 ,t 4 . The polytope P = Conv(A ) is the line

segment connecting (4, 0) and (0, 4), with vertices corresponding to s4 and t 4 . The
affine piece of XA corresponding to s4 has coordinate ring
C[s3 t/s4 , st 3 /s4 ,t 4 /s4 ] = C[t/s, (t/s)3 , (t/s)4 ] = C[t/s],
which is normal since it is a polynomial ring. Similarly, one sees that the coordinate
ring corresponding to t 4 is C[s/t], also normal. These affine pieces cover XA by
Proposition 2.1.9, so that XA is normal.
Since (1, 1, 1, 1) is in the row space of the matrix, YA is the affine cone of XA
by Proposition 2.1.4. The affine variety YA is not normal by Example 1.3.9, so that
XA is normal but not projectively normal. See Example B.1.2 for a sophisticated
proof of this fact that uses sheaf cohomology from Chapter 9.
The notion of normality used in this example is a bit ad-hoc since we have not
formally defined normality for projective varieties. Once we define normality for
abstract varieties in Chapter 3, we will see that Example 2.1.10 is fully rigorous.
We will say more about projective normality when we explore the connection
with polytopes suggested by the above results.
62
Exercises for 2.1.

2.1.1. Consider the set P3 Z33 of 3 3 permutation matrices defined in Example 2.1.3.
(a) Prove the claim made in Example 2.1.3 that three of the permutation matrices sum to
the other three and use this to explain why x123 x231 x312 x132 x321 x213 I(XP3 ).
(b) Show that dim XP3 = 4 by computing ZP3 .
(c) Conclude that I(XP3 ) = hx123 x231 x312 x132 x321 x213 i.
2.1.2. Let A Zn consist of the columns of an n s matrix A with integer entries. Prove
that the conditions of Proposition 2.1.4 are equivalent to the assertion that (1, . . . , 1) Zs
lies in the row space of A over R or Q.
2.1.3. Given a finite set A M, prove that the rank of ZA equals the dimension of the
smallest affine subspace (over Q or R) containing A .
2.1.4. Verify the claims made in Example 2.1.7. Also compute I(YA ) and check that it is
not homogeneous.
2.1.5. Let V P n be projectively normal. Use (2.0.6) to prove that the affine pieces V Ui
of V are normal.
2.1.6. Fix a finite subset A M. Given m M, let A + m = {m + m | m A }. This is
the translate of A by m.
(a) Prove that A and its translate A + m give the same projective toric variety, i.e., XA =
XA +m .
(b) Give an example to show that the affine toric varieties YA and YA +m can differ. Hint:
Pick A so that it lies in an affine hyperplane not containing the origin. Then translate
A to the origin.
2.1.7. In Proposition 2.1.4, give a direct proof that (d) (c).
2.1.8. In Example 2.1.5, the rational normal curve Cd P d was parametrized using the
homogeneous monomials s i t j , i + j = d. Here we will consider the curve parametrized by
a subset of these monomials corresponding to the exponent vectors
A = {(a0 , b0 ), . . . , (an , bn )}
where a0 > a1 > > an and ai + bi = d for every i. This gives the projective curve
XA P n . We assume n 2.
(a) If a0 < d or an > 0, explain why we can obtain the same projective curve using monomials of strictly smaller degree.
(b) Assume a0 = d and an = 0. Use Proposition 2.1.8 to show that C is smooth if and only
if a1 = d 1 and an1 = 1. Hint: For one direction, it helps to remember that smooth
varieties are normal.
2.2. Lattice Points and Polytopes

Before we can begin our exploration of the rich connections between toric varieties
and polytopes, we first need to study polytopes and their lattice points.
63
Polytopes. Recall from Chapter 1 that a polytope P MR is the convex hull of

a finite set S MR , i.e., P = Conv(S). Similar to what we did for cones, our
discussion of polytopes will omit proofs. Detailed treatments of polytopes can be
found in [51], [128] and [281].
The dimension of a polytope P MR is the dimension of the smallest affine
subspace of MR containing P. Given a nonzero vector u in the dual space NR and
+
defined by
b R, we get the affine hyperplane Hu,b and closed half-space Hu,b
Hu,b = {m MR | hm, ui = b}
and
+
= {m MR | hm, ui b}.
Hu,b
A subset Q P is a face of P, written Q P, if there are u NR \ {0}, b R with

+
.
Q = Hu,b P and P Hu,b
We say that Hu,b is a supporting affine hyperplane in this situation. Figure 1 shows
a polygon with the supporting lines of its 1-dimensional faces. The arrows in the
figure indicate the vectors u.
Figure 1. A polygon P and four of its supporting lines
We also regard P as a face of itself. Every face of P is again a polytope, and

if P = Conv(S) and Q = Hu,b P as above, then Q = Conv(S Hu,b ). Faces of
P of special interest are facets, edges and vertices, which are faces of dimension
dim P 1, 1 and 0 respectively. Facets will usually be denoted by the letter F.
Here are some properties of faces.
Proposition 2.2.1. Let P MR be a polytope. Then:

(a) P is the convex hull of its vertices.
(b) If P = Conv(S), then every vertex of P lies in S.

(c) If Q is a face of P, then the faces of Q are precisely the faces of P lying in Q.
(d) Every proper face Q P is the intersection of the facets F containing Q.
A polytope P MR can also be written as a finite intersection of closed halfspaces. The converse is true provided the intersection is bounded. In other words,
if an intersection
s
\
Hu+i ,bi
P=
i=1
64
is bounded, then P is a polytope. Here is a famous example.

Example 2.2.2. A d d matrix M Rdd is doubly-stochastic if it has nonnegative
entries and its row and column sums are all 1. If we regard Rdd as the affine
2
space Rd with coordinates ai j , then the set Md of all doubly-stochastic matrices
is defined by the inequalites
ai j 0
(all i, j)
Pd
Pd
i=1 ai j 1,
i=1 ai j 1 (all j)
Pd
Pd
j=1 ai j 1,
j=1 ai j 1 (all i).
(We use two inequalities to get one equality.) These inequalities easily imply that
0 ai j 1 for all i, j, so that Md is bounded and hence is a polytope.
Birkhoff and Von Neumann proved independently that the vertices of Md are
the d! permutation matrices and that dim Md = (d 1)2 . In the literature, Md has
various names, including the Birkhoff polytope and the transportation polytope.
See [281, p. 20] for more on this interesting polytope.
When P is full dimensional, i.e., dim P = dim MR , its presentation as an intersection of closed half-spaces has an especially nice form because each facet F has
a unique supporting affine hyperplane. We write the supporting affine hyperplane
and corresponding closed half-space as
HF = {m MR | hm, uF i = aF } and HF+ = {m MR | hm, uF i aF },
where (uF , aF ) NR R is unique up to multiplication by a positive real number.

We call uF an inward-pointing facet normal of the facet F. It follows that
\
(2.2.1)
P=
HF+ = {m MR | hm, uF i aF for all facets F P}.
F facet
In Figure 1, the supporting lines plus arrows determine the supporting half-planes
whose intersection is the polygon P. We write (2.2.1) with minus signs in order to
simplify formulas in later chapters.
Here are some important classes of polytopes.
Definition 2.2.3. Let P MR be a polytope of dimension d.
(a) P is a simplex or d-simplex if it has d + 1 vertices.
(b) P is simplicial if every facet of P is a simplex.

(c) P is simple if every vertex is the intersection of precisely d facets.
Examples include the Platonic solids in R3 :
A tetrahedron is a 3-simplex.
The octahedron and icosahedron are simplicial since their facets are triangles.
The cube and dodecahedron are simple since three facets meet at every vertex.
65
Polytopes P1 and P2 are combinatorially equivalent if there is a bijection

{faces of P1 } {faces of P2 }
that preserves dimensions, intersections, and the face relation . For example,
simplices of the same dimension are combinatorially equivalent, and in the plane,
the same holds for polygons with the same number of vertices.
Sums, Multiples, and Duals. Given a polytope P = Conv(S), its multiple rP =
Conv(rS) is again a polytope for any r 0. If P is defined by the inequalities
hm, ui i ai ,
1 i s,
then rP is given by
hm, ui i rai ,
1 i s.
In particular, when P is full dimensional, then P and rP have the same inwardpointing facet normals.
The Minkowski sum of subsets A1 , A2 MR is
A1 + A2 = {m1 + m2 | m1 A1 , m2 A2 }.
Given polytopes P1 = Conv(S1 ) and P2 = Conv(S2 ), their Minkowski sum P1 +P2 =

Conv(S1 + S2 ) is again a polytope. We also have the distributive law
rP + sP = (r + s)P.
When P MR is full dimensional and 0 is an interior point of P, we define the
dual or polar polytope
P = {u NR | hm, ui 1 for all m P} NR .
Figure 2 shows an example of this in the plane.
Figure 2. A polygon P and its dual P in the plane
When we write P = {m MR | hm, uF i aF , F facet} as in (2.2.1), we have

aF > 0 for all F since 0 is in the interior. Then P is the convex hull of the vectors
(1/aF )uF NR (Exercise 2.2.1). We also have (P ) = P in this situation.
66
Lattice Polytopes. Now let M and N be dual lattices with associated vector spaces
MR and NR . A lattice polytope P MR is the convex hull of a finite set S M. It
follows easily that a polytope in MR is a lattice polytope if and only if its vertices
lie in M (Exercise 2.2.2).
Example 2.2.4. The standard n-simplex in Rn is
n = Conv(0, e1 , . . . , en ).
Another simplex in R3 is P = Conv(0, e1 , e2 , e1 + e2 + 3e3 ), shown in Figure 3.
e1 + e2 + 3e3
e2
e1
Figure 3. The simplex P = Conv(0, e1 , e2 , e1 + e2 + 3e3 ) R3
The lattice polytopes 3 and P are combinatorially equivalent but will give very
different projective toric varieties.
Example 2.2.5. The Birkhoff polytope defined in Example 2.2.2 is a lattice polytope relative to the lattice of integer matrices Zdd since its vertices are the permutation matrices, whose entries are all 0 or 1.
One can show that faces of lattice polytopes are again lattice polytopes and that
Minkowski sums and integer multiples of lattice polytopes are lattice polytopes
(Exercise 2.2.2). Furthermore, everyTlattice polytope is an intersection of closed
half-spaces defined over M, i.e., P = si=1 Hu+i ,bi where ui N and bi Z.
When a lattice polytope P is full dimensional, the facet presentation given in

(2.2.1) has an especially nice form. If F is a facet of P, then the inward-pointing
facet normals of F lie on a rational ray in NR . Let uF denote the unique ray generator. The corresponding aF is integral since hm, uF i = aF when m is a vertex of
F. It follows that
(2.2.2)
P = {m MR | hm, uF i aF for all facets F P}
is the unique facet presentation of the lattice polytope P.
67
Example 2.2.6. Consider the square P = Conv(e1 e2 ) R2 . The facet normals

of P are e1 and e2 and the facet presentation of P is given by
hm, e1 i 1
hm, e2 i 1.
Since the aF are all equal to 1, it follows that P = Conv(e1 , e2 ) is also a lattice
polytope. The polytopes P and P are pictured in Figure 2.
It is rare that the dual of a lattice polytope is a lattice polytopethis is related
to the reflexive polytopes that will be studied later in the book.
Example 2.2.7. The 3-simplex P = Conv(0, e1 , e2 , e1 + e2 + 3e3 ) R3 pictured in
Example 2.2.4 has facet presentation
hm, e3 i 0
hm, 3e1 e3 i 0
hm, 3e2 e3 i 0
hm, 3e1 3e2 + e3 i 3

(Exercise 2.2.3). However, if we replace 3 with 1 in the last inequality, we get
integer inequalities that define (1/3)P, which is not a lattice polytope.
The combinatorial type of a polytope is an interesting object of study. This

leads to the question Is every polytope combinatorially equivalent to a lattice
polytope? If the given polytope is simplicial, the answer is yesjust wiggle
the vertices to make them rational and clear denominators to get a lattice polytope.
The same argument works for simple polytopes by wiggling the facet normals.
This will enable us to prove results about arbitrary simplicial or simple polytopes
using toric varieties. But in general, the answer is nothere exist polytopes in
every dimension 8 not combinatorially equivalent to any lattice polytope. An
example is described in [281, Ex. 6.21].
A First Guess for the Toric Variety of Polytope. A lattice polytope P gives a finite
set of lattice points P M, which in turn gives a projective toric variety XPM . This
is a natural candidate for the toric variety of P. However, XPM may fail to reflect
the properties of P when there are too few lattice points.
Example 2.2.8. In Example 2.2.4, we defined the standard 3-simplex 3 and the
3-simplex P = Conv(0, e1 , e2 , e1 + e2 + 3e3 ). Both have only four lattice points
(their vertices). Thus the toric varieties X3 Z3 and XPZ3 live in P 3 , and in fact
we have X3 Z3 = XPZ3 = P 3 (Exercise 2.2.3). Yet 3 and P are very different
lattice polytopes. For example, the nonzero vertices of 3 form a basis of Z3 , but
this is not true for P.
68
We will see below that the construction P 7 P M 7 XPM works fine when
the lattice polytope P has enough lattice points. Hence our first task is to explore
what this means.
Normal Polytopes. Here is one way to say that P has enough lattice points.
Definition 2.2.9. A lattice polytope P MR is normal if
for all k, N.
(kP) M + (P) M = ((k + )P) M
The inclusion (kP) M + (P) M ((k + )P) M is automatic. Thus normality means that all lattice points of (k + )P come from lattice points of kP and
P. In particular, a lattice polytope is normal if and only if
P M + + P M = (kP) M.
{z
}
|
k times
for all integers k 1. So normality means that P has enough lattice points to
generate the lattice points in all integer multiples of P.
Lattice polytopes of dimension 1 are normal (Exercise 2.2.4). Here is another
class of normal polytopes.
Definition 2.2.10. A simplex P MR is basic if P has a vertex m0 such that the
differences m m0, for vertices m 6= m0 of P, form a subset of a Z-basis of M.
This definition is independent of which vertex m0 P is chosen. The standard
simplex n Rn is basic, and any basic simplex is normal (Exercise 2.2.5). More
general simplices, however, need not be normal.
Example 2.2.11. We noted in Example 2.2.8 that the only lattice points of P =
Conv(0, e1 , e2 , e1 + e2 + 3e3 ) R3 are its vertices. It follows easily that
e1 + e2 + e3 = 61 (0) + 13 (2e1 ) + 13 (2e2 ) + 61 (2e1 + 2e2 + 6e3 ) 2P
is not the sum of lattice points of P. This shows that P is not normal. In particular,
P is not basic.
Here is an important result on normality.

Theorem 2.2.12. Let P MR be a full dimensional lattice polytope of dimension
n 2. Then kP is normal for all k n 1.
Proof. This result was first explicitly stated in [55], though as noted in [55], its
essential content follows from [94] and [187]. We will use ideas from [187] and
[223] to show that
(2.2.3)
(kP) M + P M = ((k + 1)P) M for k Z, k n 1.
69
In Exercise 2.2.6 you will show that (2.2.3) implies that kP is normal for all integers
k n 1. Note also that for (2.2.3), it suffices to prove that
((k + 1)P) M (kP) M + P M
since the other inclusion is obvious.

First consider the case where P is a simplex with no interior lattice points.
Let the vertices of P be m0 , . . . , mn and take k n 1. Then (k + 1)P has vertices
(k+1)m0 , . . . , (k+1)mn , so that a point m ((k+1)P)M is a convex combination
P
P
m = ni=0 i (k + 1)mi , where i 0, ni=0 i = 1.
If we set i = (k + 1)i , then
P
P
m = ni=0 i mi , where i 0, ni=0 i = k + 1.
If i 1 for some i, then one easily sees that m mi (kP) M. Hence m =

(m mi) + mi is the desired decomposition. On the other hand, if i < 1 for all i,
then
P
n = (n 1) + 1 k + 1 = ni=0 i < n + 1,
P
so that k = n 1 and ni=0 i = n. Now consider the lattice point
P
m
e = m0 + + mn ) m = ni=0 (1 i )mi .
P
The coefficients are positive since i < 1 for all i, and their sum is ni=0 (1 i ) =
n + 1 n = 1. Hence m
e is a lattice point in the interior of P since 1 i > 0 for all
i. This contradicts our assumption on P and completes the proof when P is a lattice
simplex containing no interior lattice points.
To prove (2.2.3) for the general case, it suffices to prove that P is a finite union
of n-dimensional lattice simplices with no interior lattice points (Exercise 2.2.7).
For this, we use Caratheodorys theorem (see [281, Prop. 1.15]), which asserts that
for a finite set A MR , we have
[
Conv(A ) = Conv(B),
where the union is over all subsets B A consisting of dim Conv(A ) + 1 affinely
independent elements. Thus each Conv(B) is a simplex. This enables us to write
our lattice polytope P as a finite union of n-dimensional lattice simplices.
If an n-dimensional lattice simplex Q = Conv(w0 , . . . , wn ) has an interior lattice
point v, then
n
[
bi , . . . , wn , v)
Q = Qi , Qi = Conv(w0 , . . . , w
i=0
is a finite union of n-dimensional lattice simplices, each of which has fewer interior
lattice points than Q since v becomes a vertex of each Qi . By repeating this process
on those Qi that still have interior lattice points, we can eventually write Q and
hence our original polytope P as a finite union of n-dimensional lattice simplices
with no interior lattice points. You will verify the details in Exercise 2.2.7.
70
This theorem shows that for the nonnormal 3-simplex P of Example 2.2.11, its
multiple 2P is normal. Here is another consequence of Theorem 2.2.12.
Corollary 2.2.13. Every lattice polygon P R2 is normal.
We can also interpret normality in terms of the cone of P, defined by

C(P) = Cone(P {1}) MR R.
This was introduced in Figure 3 of Chapter 1. The key feature of this cone is that
kP is the slice of C(P) at height k, as illustrated in Figure 4. It follows that lattice
points m kP correspond to points (m, k) C(P) (M Z).
2P
height = 2
C(P)
P
height = 1
Figure 4. The cone C(P) sliced at heights 1 and 2
In Exercise 2.2.8 you will show that the semigroup C(P) (M Z) of lattice
points in C(P) relates to normality as follows.
Lemma 2.2.14. Let P MR be a lattice polytope. Then P is normal if and only if
(P M) {1} generates the semigroup C(P) (M Z).

This lemma tells us that P MR is normal if and only if (P M) {1} is the
Hilbert basis of C(P) (M Z).
Example 2.2.15. In Example 2.2.11, the simplex P = Conv(0, e1 , e2 , e1 + e2 + 3e3 )
gives the cone C(P) R4 . The Hilbert basis of C(P) (M Z) is
(0, 1), (e1 , 1), (e2 , 1), (e1 + e2 + 3e3 , 1), (e1 + e2 + e3 , 2), (e1 + e2 + 2e3 , 2)
(Exercise 2.2.3). Since the Hilbert basis has generators of height greater than 1,
Lemma 2.2.14 gives another proof that P is not normal.
71
In Exercise 2.2.9, you will generalize Lemma 2.2.14 as follows.

Lemma 2.2.16. Let P MR Rn be a lattice polytope of dimension n 2 and let
k0 be the maximum height of an element of the Hilbert basis of C(P). Then:
(a) k0 n 1.
(b) kP is normal for any k k0 .
The Hilbert basis of the simplex P of Example 2.2.15 has maximum height 2.
Then Lemma 2.2.16 gives another proof that 2P is normal. The paper [187] gives
a version of Lemma 2.2.16 that applies to Hilbert bases of more general cones.
Very Ample Polytopes. Here is a slightly different notion of what it means for a
polytope to have enough lattice points.
Definition 2.2.17. A lattice polytope P MR is very ample if for every vertex
m P, the semigroup SP,m = N(P M m) generated by the set P M m =
{m m | m P M} is saturated in M.
This definition relates to normal polytopes as follows.
Proposition 2.2.18. A normal lattice polytope P is very ample.
Proof. Fix a vertex m0 P and take m M such that km SP,m0 for some integer
k 1. To prove that m SP,m0 , write km SP,m0 as
P
km = m PM am (m m0 ), am N.
P
Pick d N so that kd m PM am . Then

P
P
km + kdm0 = m PM am m + kd m PM am m0 kdP.
Dividing by k gives m + dm0 dP, which by normality implies that

m + dm0 =
We conclude that m =
Pd
d
X
mi ,
i=1
i=1 (mi m0 )
mi P M for all i.
SP,m0 , as desired.
Combining this with Theorem 2.2.12 and Corollary 2.2.13 gives the following.
Corollary 2.2.19. Let P MR Rn be a full dimensional lattice polytope. Then:
(a) If dim P 2, then kP is very ample for all k n 1.
(b) If dim P = 2, then P is very ample.
Part (a) was first proved in [94]. We will soon see that very ampleness is
precisely the property needed to define the toric variety of a lattice polytope.
The following example taken from [52, Ex. 5.1] shows that very ample polytopes need not be normal, i.e., the converse of Proposition 2.2.18 is false.
72
Example 2.2.20. Given 1 i < j < k 6, let [i jk] denote the vector in Z6 with 1
in positions i, j, k and 0 elsewhere. Thus [123] = (1, 1, 1, 0, 0, 0). Then let

A = [123], [124], [135], [146], [156], [236], [245], [256], [345], [346] Z6 .
The lattice polytope P = Conv(A ) lies in the affine hyperplane of R6 where the
coordinates sum to 3. As explained in [52], this configuration can be interpreted in
terms of a triangulation of the real projective plane.
The points of A are the only lattice points of P (Exercise 2.2.10), so that A is
the set of vertices of P. Number the points of A as m1 , . . . , m10 . Then
(1, 1, 1, 1, 1, 1) =
1
5
10
X
i=1
mi =
10
X
1
10 (2mi )
i=1
shows that v = (1, 1, 1, 1, 1, 1) 2P. Since v is not a sum of lattice points of P

(when [i jk] A , the vector v [i jk] is not in A ), we conclude that P is not a
normal polytope.
To show that P is very ample, we first prove that A {1}{(v, 2)} R6 R is
a Hilbert basis of the semigroup C(P) Z7 , where C(P) R6 R is the cone over
P {1}. We show how do this using Normaliz [57] in Example B.3.1. Another
approach would be to use 4ti2 [140].
Now fix i and let SP,mi be the semigroup generated by the m j mi . Take m
such that km SP,mi . As in the proof of Proposition 2.2.18, this implies that
m + dmi dP for some d N. Thus (m + dmi , d) C(P) Z7 . Expressing this in
terms of the above Hilbert basis easily implies that
Z6
m = a(v 2mi ) +
10
X
j=1
a j (m j mi),
a, a j N.
If we can show that v 2mi SP,mi , then m SP,mi follows immediately and proves
that SP,mi is saturated. When i = 1, one can check that
v + [123] = [124] + [135] + [236],
which implies that
v 2m1 = (m2 m1 ) + (m3 m1 ) + (m6 m1 ) SP,m1 .
One obtains similar formulas for i = 2, . . . , 10 (Exercise 2.2.10), which completes
the proof that P is very ample.
The polytope P has further interesting properties. For example, up to affine
equivalence, P can be described as the convex hull of the 10 points in Z5 given by
(0, 0, 0, 0, 0), (0, 0, 0, 0, 1), (0, 0, 1, 1, 0), (0, 1, 0, 1, 1), (0, 1, 1, 1, 0)
(1, 0, 1, 0, 1), (1, 0, 1, 1, 1), (1, 1, 0, 0, 0), (1, 1, 0, 1, 1), (1, 1, 1, 0, 0).
73
Of all 5-dimensional polytopes whose vertices lie in {0, 1}5 , this polytope
has the

most edges, namely 45 (see [4]). Since it has 10 vertices and 45 = 10
,
every
pair
2
of distinct vertices is joined by an edge. Such polytopes are 2-neighborly.
Exercises for 2.2.

2.2.1. Let P MR be a full dimensional polytope with the origin as an interior point.
(a) Write P = {m MR | hm, uF i aF for all facets F}. Prove that aF > 0 for all F and
that P = Conv((1/aF )uF | F a facet).
(b) Prove that the dual of a simplicial polytope is simple and vice versa.
(c) Prove that (rP) = (1/r)P for all r > 0.
(d) Use part (c) to construct an example of a lattice polytope whose dual is not a lattice
polytope.
2.2.2. Let P MR be a polytope.
(a) Prove that P is a lattice polytope if and only if the vertices of P lie in M.
(b) Prove that P is a lattice polytope if and only if P is the convex hull of its lattice points,
i.e., P = Conv(P M).
(c) Prove that every face of a lattice polytope is a lattice polytope.

(d) Prove that Minkowski sums and integer multiples of lattice polytopes are again lattice
polytopes.
2.2.3. Let P = Conv(0, e1 , e2 , e1 + e2 + 3e3 ) R3 be the simplex studied in Examples 2.2.4,

2.2.7, 2.2.8, 2.2.11 and 2.2.15.
(a) Verify the facet presentation of P given in Example 2.2.7.
(b) Show that the only lattice points of P are its vertices.
(c) Show that the toric variety XPZ3 is P 3 , as claimed in Example 2.2.8.
(d) Show that the vectors given in Example 2.2.15 form the Hilbert basis of the semigroup
C(P) (M Z).
2.2.4. Prove that every 1-dimensional lattice polytope is normal.
2.2.5. Recall the definition of basic simplex given in Definition 2.2.10.
(a) Show that if a simplex satisfies Definition 2.2.10 for one vertex, then it satisfies the
definition for all vertices.
(b) Show that the standard simplex n is basic.
(c) Prove that a basic simplex is normal.
2.2.6. Let P MR Rn be an n-dimensional lattice polytope.
(a) Prove that (2.2.3) implies that
(kP) M + (P) M = ((k + )P) M
for all integers k n 1 and 0. Hint: When = 2, we have
(kP) M + P M + P M (kP) M + (2P) M ((k + 2)P) M.
Apply (2.2.3) twice on the right.

(b) Use part (a) to prove that kP is normal when k n 1 and P satisifes (2.2.3).
74
2.2.7. Let P MR Rn be an n-dimensional lattice polytope.

(a) Follow the hints given in the text to give a careful proof that P is a finite union of
n-dimensional lattice simplices with no interior lattice points.
(b) In the text, we showed that (2.2.3) holds for an n-dimensional lattice simplex with no
interior lattice points. Use this and part (a) to show that (2.2.3) holds for P.
2.2.8. Prove Lemma 2.2.14.
2.2.9. In this exercise you will prove Lemma 2.2.16. As in the lemma, let k0 be the maximum height of a generator of the Hilbert basis of C(P) (M Z).
(a) Adapt the proof of Gordans Lemma (Proposition 1.2.17) to show that if H is the
Hilbert basis of the semigroup of lattice points in a strongly convex cone Cone(A ),
then the lattice points in the cone can be written as the union

S
NA mH m + NA .
(b) If the Hilbert basis of C(P) (M Z) is {(m1 , h1 ), . . . , (ms , hs )}, then conclude that

Ss
C(P) (M Z) = S i=1 (mi , hi ) + S ,
where S = N((P M) {1}).
(c) Use part (b) to show that (2.2.3) holds for k k0 .

2.2.10. Consider the polytope P = Conv(A ) from Example 2.2.20.
(a) Prove that A is the set of lattice points of P.
(b) Complete the proof begun in the text that P is very ample.
2.2.11. Prove that every proper face of a simplicial polytope is a simplex.
2.2.12. In Corollary 2.2.19 we proved that kP is very ample for k n 1 using Theorem 2.2.12 and Proposition 2.2.18. Give a direct proof of the weaker result that kP is very
ample for k sufficiently large. Hint: A vertex m P gives the cone CP,m generated by the
semigroup SP,m defined in Definition 2.2.17. The cone CP,m is strongly convex since m is
a vertex and hence CP,m M has a Hilbert basis. Furthermore, CP,m = CkP,km for all k N.
Now argue that when k is large, (kP) M km contains the Hilbert basis of CP,m M. A
picture will help.
2.2.13. Fix an integer a 1 and consider the 3-simplex P = Conv(0, ae1 , ae2 , e3 ) R3 .
(a) Work out the facet presentation of P and verify that the facet normals can be labeled
so that u0 + u1 + u2 + au3 = 0.
(b) Show that P is normal. Hint: Show that P Z3 + (kP) Z3 = ((k + 1)P) Z3 .
We will see later that the toric variety of P is the weighted projective space P(1, 1, 1, a).
2.3. Polytopes and Projective Toric Varieties

Our next task is to define the toric variety of a lattice polytope. As noted in 2.2,
we need to make sure that the polytope has enough lattice points. Hence we begin
with very ample polytopes. Strongly convex rational polyhedral cones will play an
important role in our development.
75
The Very Ample Case. Let P MR be a full dimensional very ample polytope
relative to the lattice M, and let dim P = n. If P M = {m1 , . . . , ms }, then XPM is
the Zariski closure of the image of the map TN P s1 given by

t 7 m1 (t), . . . , ms (t) P s1 .
Fix homogeneous coordinates x1 , . . . , xs for P s1 .
We examine the structure of XPM P s1 using Propositions 2.1.8 and 2.1.9.

For each mi P M consider the semigroup
Si = N(P M mi )
generated by m j mi for m j P M. In P s1 we have the affine open subset

Ui Cs1 consisting of those points where xi 6= 0. Proposition 2.1.8 showed that
the affine open piece XPM Ui of XPM is the affine toric variety
XPM Ui Spec(C[Si ]),
and Proposition 2.1.9 showed that
XPM =
mi vertex of P
XPM Ui.
Here is our first major result about XPM .

Theorem 2.3.1. Let XPM be the projective toric variety of the very ample polytope
P MR , and assume that P is full dimensional with dim P = n. Then:
(a) For each vertex mi P M, the affine piece XPM Ui is the affine toric variety
XPM Ui = Ui = Spec(C[i M])
where i NR is the strongly convex rational polyhedral cone dual to the cone
Cone(P M mi) MR . Furthermore, dim i = n.
(b) The torus of XPM has character lattice M and hence is the torus TN .
Proof. Let Ci = Cone(P M mi). Since mi is a vertex, it has a supporting hy+ and P H
perplane Hu,a such that P Hu,a
u,a = {mi }. It follows that Hu,0 is a
supporting hyperplane of 0 Ci (Exercise 2.3.1), so that Ci is strongly convex. It is
also easy to see that dim Ci = dim P (Exercise 2.3.1). It follows that Ci and i = Ci
are strongly convex rational polyhedral cones of dimension n.
We have Si Ci M = i M. By hypothesis, P is very ample, which means
that Si M is saturated. Since Si and Ci = i are both generated by P M mi ,
part (a) of Exercise 1.3.4 implies Si = i M. (This exercise was part of the proof
of the characterization of normal affine toric varieties given in Theorem 1.3.5.)
Part (a) of the theorem follows immediately.
For part (b), Theorem 1.2.18 implies that TN is the torus of Ui since i is
strongly convex. Then TN Ui = XPM Ui XPM shows that TN is also the
torus of XPM .
76
The affine pieces XPM Ui and XPM U j intersect in XPM Ui U j . In order

to describe this intersection carefully, we need to study how the cones i and j fit
together in NR . This leads to our next topic.
The Normal Fan. The cones i NR appearing in Theorem 2.3.1 fit together in a
remarkably nice way, giving a structure called the normal fan of P.
Let P MR be a full dimensional lattice polytope, not necessarily very ample.
Faces, facets and vertices of P will be denoted by Q, F and v respectively. Hence
we write the facet presentation of P as
P = {m MR | hm, uF i aF for all F}.
(2.3.1)
A vertex v P gives cones

Cv = Cone(P M v) MR
and
v = Cv NR .
(When v = mi , these are the cones Ci and i studied above.) Faces Q P containing
the vertex v correspond bijectively to faces Qv Cv via the maps
Q 7 Qv = Cone(Q M v)
(2.3.2)
Qv 7 Q = (Qv + v) P
which are inverses of each other. These maps preserve dimensions, inclusions, and
intersections (Exercise 2.3.2), as illustrated in Figure 5.
P
Cv
Q
Qv
Figure 5. The cone Cv of a vertex v P
In particular, all facets of Cv come from facets of P containing v, so that

Cv = {m MR | hm, uF i 0 for all F containing v}.
77
By the duality results of Chapter 1, it follows that the dual cone v is given by
v = Cone(uF | F contains v).
This construction generalizes to arbitrary faces Q P by setting

Q = Cone(uF | F contains Q).
Thus the cone F is the ray generated by uF , and P = {0} since {0} is the cone
generated by the empty set. Here is our main result about these cones.
Theorem 2.3.2. Let P MR be a full dimensional lattice polytope and set P =
{Q | Q P}. Then:
(a) For all Q P , each face of Q is also in P .
(b) The intersection Q Q of any two cones in P is a face of each.

Remark 2.3.3. A finite collection of strongly convex rational polyhedral cones
that satisfies (a) and (b) of Theorem 2.3.2 is called a fan. General fans will be
introduced in Chapter 3. Since the cones in the fan P are built from the inwardpointing normal vectors uF , we call P the normal fan or inner normal fan of P.
Theorem 2.3.2 will follow from the results proved below. But first, we give a
simple example of a polytope and its normal fan.
Example 2.3.4. The 2-simplex 2 R2 has vertices 0, e1 , e2 . Let P = k2 for an
integer k > 0. Figure 6 shows P and its normal fan P . At each vertex vi of P, we
v2
v0
0
2
v1
Figure 6. The triangle P = k2 R2 and its normal fan P
show the cone i = Cvi generated by the normal vectors of the facets containing vi .
The reassembled cones appear on the right as P .
Notice that the cones Cvi MR do not fit together nicely; rather, it is their duals
i NR that give the fan P . This explains why cones in NR are the key players in
toric geometry.
Here is the first of the results we need to prove Theorem 2.3.2.
78
Lemma 2.3.5. Let Q be a face of P and let Hu,b be a supporting affine hyperplane
of P. Then u Q if and only if Q Hu,b P.
P
Proof. First suppose that u Q and write u = QF F uF , F 0. Then setting
P
+
and Q Hu,b0 P. Recall that
b0 = QF F aF easily implies that P Hu,b
0
the integers aF come from the facet presentation (2.3.1). It follows that Hu,b0 is a
supporting hyperplane of P. Since Hu,b is a supporting hyperplane by hypothesis,
we must have b0 = b, hence Q Hu,b P.
Going the other way, suppose that Q Hu,b P. Take a vertex v Q. Then
+
+
. Hence u Cv = v , so that
and v Hu,b imply that Cv Hu,0
P Hu,b
P
u = vF F uF , F 0.
Let F0 be a facet of P containing v but not Q, and pick p Q with p

/ F0 . Then
p, v Q Hu,b imply that
P
b = hp, ui = vF F hp, uF i
P
P
b = hv, ui = vF F hv, uF i = vF F aF ,
where the last equality uses hv, uF i = aF for v F. These equations imply
P
P
vF F hp, uF i = vF F aF .
However, p
/ F0 gives hp, uF0 i > aF0 , and since hp, uF i aF for all F, it follows
that F0 = 0 whenever Q 6 F0 . This gives u Q and completes the proof of the
lemma.

Corollary 2.3.6. If Q P and F P is a facet, then uF Q if and only if Q F.
Proof. One direction is obvious by the definition of Q , and the other direction
follows from Lemma 2.3.5 since HuF ,aF is a supporting affine hyperplane of P
with HuF ,aF P = F.

Theorem 2.3.2 is an immediate consequence of the following proposition.
Proposition 2.3.7. Let Q and Q be faces of a full dimensional lattice polytope
P MR . Then:
(a) Q Q if and only if Q Q .
(b) If Q Q , then Q is a face of Q , and all faces of Q are of this form.
(c) Q Q = Q , where Q is the smallest face of P containing Q and Q .

Proof. To prove part (a), note that if Q Q , then any facet containing Q also
contains Q, which implies Q Q . The other direction follows easily from
Corollary 2.3.6 since every face is the intersection of the facets containing it by
Proposition 2.2.1.
79
For part (b), fix a vertex v Q and note that by (2.3.2), Q determines a face Qv
of Cv . Using the duality of Proposition 1.2.10, Qv gives the dual face
Qv = Cv Q
v = v Q v
of the cone v . Then using v = Cone(uF | v F) and Qv Cv = v , one obtains

Qv = Cone(uF | v F, Qv HuF ,0 ).
Since v Q, the inclusion Qv HuF ,0 is equivalent to Q HuF ,aF , which in turn

is equivalent to Q F. It follows that
(2.3.3)
Qv = Cone(uF | Q F) = Q ,
so that Q is a face of v , and all faces of v arise in this way.

In particular, Q Q means that Q is also a face of v , and since Q Q
by part (a), we see that Q is a face of Q . Furthermore, every face of Q is a face
of v by Proposition 1.2.6 and hence is of the form Q for some face Q . Using
part (a) again, we see that Q Q , and part (b) follows.
For part (c), let Q be the smallest face of P containing Q and Q . This exists
because a face is the intersection of the facets containing it, so that Q is the intersection of all facets containing Q and Q (if there are no such facets, then Q = P).
By part (b) Q is a face of both Q and Q . Thus Q Q Q .
It remains to prove the opposite inclusion. If Q Q = {0} = P , then Q = P

and we are done. If Q Q 6= {0}, any nonzero u in the intersection lies in both
Q and Q . The proof of Proposition 2.3.8 given below will show that Hu,b is a
supporting affine hyperplane of P for some b R. By Lemma 2.3.5, u Q and
u Q imply that Q and Q lie in Hu,b P. The latter is a face of P containing
Q and Q , so that Q Hu,b P since Q is the smallest such face. Applying
Lemma 2.3.5 again, we see that u Q .

Proposition 2.3.7 shows that there is a bijective correspondence between faces
of P and cones of the normal fan P . Here are some further properties of this
correspondence.
Proposition 2.3.8. Let P MR be a full dimensional lattice polytope of dimension
n and consider the cones Q in the normal fan P of P. Then:
(a) dim Q + dim Q = n for all faces Q P.
S
S
(b) NR = v vertex of P v = Q P Q .
Proof. Suppose Q P and take a vertex v of Q. By (2.3.2) this gives a face Qv of

the cone Cv , which has a dual face Qv of the dual cone Cv = v . Since Qv = Q
by (2.3.3), we have
dim Q + dim Q = dim Qv + dim Qv = n,
80
where the first equality uses Exercise 2.3.2 and the second follows from Proposition 1.2.10. This proves part (a). For part (b), let u NR be nonzero and set
+
and v Hu,b for at least one verb = min{hv, ui | v vertex of P}. Then P Hu,b
tex of P, so that u v by Lemma 2.3.5. The final equality of part (b) follows
immediately.

A fan satisfying the condition of part (b) of Proposition 2.3.8 is called complete. Thus the normal fan of a lattice polytope is always complete. We will learn
more about complete fans in Chapter 3.
In general, multiplying a polytope by a positive integer has no effect on its
normal fan, and the same is true for translations by lattice points. We record these
properties in the following proposition (Exercise 2.3.3).
Proposition 2.3.9. Let P MR be a full dimensional lattice polytope. Then for
any lattice point m M and any integer k 1, the polytopes m + P and kP have
the same normal fan as P.

Examples of Normal Fans. Here are some more examples of normal fans.
Example 2.3.10. Figure 7 shows a lattice hexagon P in the plane together with
its normal fan. The vertices of P are labeled v1 , . . . , v6 , with corresponding cone
1 , . . . , 6 in the normal fan. In the figure, P is shown on the left, and at each vertex
vi , we have drawn the normal vectors of the facets containing vi and shaded the
cone i they generate. On the right, these cones are assembled at the origin to give
the normal fan.
v5
v4
v6
v3
P
v1
2
4
1
5
v2
Figure 7. A lattice hexagon P and its normal fan P
Notice how one can read off the structure of P from the normal fan. For example, two cones i and j share a ray in P if and only if the vertices vi and v j lie
on an edge of P.
The hexagon P is an example of a zonotope since it is a Minkowski sum of
line segments (three in this case). Notice also that P is determined by three lines
81
through the origin. In 6.2 we will prove that the normal fan of any zonotope is
determined by an arrangement of hyperplanes containing the origin.
Example 2.3.11. Consider the cube P R3 with vertices (1, 1, 1). The facet
normals are e1 , e2 , e3 , and the facet presentation of P is
hm, ei i 1.
The origin is an interior point of P. By Exercise 2.2.1, the facet normals are the
vertices of the dual polytope P , the octahedron in Figure 8.
z
P
x
x
Figure 8. A cube P R3 and its dual octahedron P
However, the facet normals also give the normal fan of P, and one can check
that in the above figure, the maximal cones of the normal fan are the octants of R3 ,
which are just the cones over the facets of the dual polytope P .
As noted earlier, it is rare that both P and P are lattice polytopes. However,
whenever P MR is a lattice polytope containing 0 as an interior point, it is still
true that maximal cones of the normal fan P are the cones over the facets of
P NR (Exercise 2.3.4).
The special behavior of the polytopes P and P discussed in Examples 2.2.6

and 2.3.11 leads to the following definition.
Definition 2.3.12. A full dimensional lattice polytope P MR is reflexive if its
facet presentation is
P = {m MR | hm, uF i 1 for all facets F}.
If P is reflexive, then 0 is a lattice point of P and is the only interior lattice

point of P (Exercise 2.3.5). Since aF = 1 for all F, Exercise 2.2.1 implies that
P = Conv(uF | F facet of P).
82
Thus P is a lattice polytope and is in fact reflexive (Exercise 2.3.5).

We will see later that reflexive polytopes lead to some very interesting toric
varieties that are important for mirror symmetry.
Intersection of Affine Pieces. Let P MR be a full dimensional very ample polytope and set s = |P M|. This gives
XPM P s1 .
If XPM Uv is the affine piece corresponding to a vertex v P, then

XPM Uv = Uv = Spec(C[v M])
by Theorem 2.3.1. Thus the affine piece XPM Uv is the toric variety of the cone
v in the normal fan P of P.
Our next task is to describe the intersection of two of these affine pieces.
Proposition 2.3.13. Let P MR be full dimensional and very ample. If v 6= w are
vertices of P and Q is the smallest face of P containing v and w, then
XPM Uv Uw = UQ = Spec(C[Q M])
and the inclusions
XPM Uv XPM Uv Uw XPM Uw
can be written
(2.3.4)
Uv (Uv ) wv = UQ = (Uw ) vw Uw .
Proof. We analyzed the intersection of affine pieces of XPM in 2.1. Translated

to the notation being used here, (2.1.6) and (2.1.7) imply that
XPM Uv Uw = (Uv ) wv = (Uw ) vw .
Thus all we need to show is that
(Uv ) wv = UQ .
However, we have w v Cv = v , so that = Hwv v is a face of v . In this
situation, Proposition 1.3.16 and equation (1.3.4) imply that
(Uv ) wv = U .
Thus the proposition will follow once we prove = Q , i.e., Hwv v = Q .
Since Q = v w by Proposition 2.3.7, it suffices to prove that
Hwv v = v w .
Let u Hwv v . If u 6= 0, there is b R such Hu,b is a supporting affine hyperplane of P. Then u v implies v Hu,b by Lemma 2.3.5, so that w Hu,b since
u Hwv . Applying Lemma 2.3.5 again, we get u w . Going the other way, let
u v w . If u 6= 0, pick b R as above. Then u v w and Lemma 2.3.5
83
imply that v, w Hu,b , from which u Hwv follows easily. This completes the
proof.

This proposition and Theorem 2.3.1 have the remarkable result that the normal
fan P completely determines the internal structure of XPM : we build XPM from
local pieces given by the affine toric varieties Uv , glued together via (2.3.4). We
do not need the ambient projective space P s1 for any of thiseverything we need
to know is contained in the normal fan.
The Toric Variety of a Polytope. We can now give the general definition of the
toric variety of a polytope.
Definition 2.3.14. Let P MR be a full dimensional lattice polytope. Then we
define the toric variety of P to be
XP = X(kP)M
where k is any positive integer such that kP is very ample.
Such integers k exist by Corollary 2.2.19, and if k and are two such integers,
then kP and P have the same normal fan by Proposition 2.3.9, namely kP =
P = P . It follows that while X(kP)M and X(P)M lie in different projective
spaces, they are built from the affine toric varieties Uv glued together via (2.3.4).
Once we develop the language of abstract varieties in Chapter 3, we will see that
XP is well-defined as an abstract variety.
We will often speak of XP without regard to the projective embedding. When
we want to use a specific embedding, we will say XP is embedded using kP,
where we assume that kP is very ample. In Chapter 6 we will use the language of
divisors and line bundles to restate this in terms of a divisor DP on XP such that
kDP is very ample precisely when kP is.
Here is a simple example to illustrate the difference between XP as an abstract
variety and XP as sitting in a specific projective space.
Example 2.3.15. Consider the n-simplex n Rn . We can define Xn using kn
for any integer k 1 since n is normal
and hence very ample. The lattice points

monomials
of C[t1 , . . . ,tn ] of total degree k.
in kn correspond to the sk = n+k
k
s
1
k
This gives an embedding Xn P
. When k = 1, n Zn = {0, e1 , . . . , en }
implies that
Xn = P n .
The normal fan of n is described in Exercise 2.3.6. For an arbitrary k 1, we can
regard Xn P sk 1 as the image of the map
k : P n P sk 1
84
defined using all monomials of total degree k in C[x0 , . . . , xn ] (Exercise 2.3.6). It

follows that this map is an embedding, usually called the Veronese embedding. But
when we forget the embedding, the underlying toric variety is just P n .
The Veronese embedding allows us to construct some interesting affine open
subsets of P
P n . Let f C[x0 , . . . , xn ] be nonzero and homogeneous of degree k and
write f = ||=k c x . We write the homogeneous coordinates of P s1 as y for
P
|| = k. Then L = ||=k c y is a nonzero linear form in the variables y , so that
P sk 1 \ V(L) is a copy of Csk 1 (Exercise 2.3.6). If follows that

P n \ V( f ) k (P n ) P sk 1 \ V(L)
is an affine variety (usually not toric). This shows that P n has a richer supply of
affine open subsets than just the open sets Ui = P n \ V(xi ) considered earlier in the
chapter.
When we explain the Proj construction of P n later in the book, we will see the
intrinsic reason why P n \ V( f ) is an affine open subset of P n .
Example 2.3.16. The 2-dimensional analog of the rational normal curve Cd is the
rational normal scroll Sa,b , which is the toric variety of the polygon
Pa,b = Conv(0, ae1 , e2 , be1 + e2 ) R2 ,
where a, b N satisfy 1 a b. The polygon P = P2,4 and its normal fan are
pictured in Figure 9.
v2 = e2
v3 = 4e1 + e2
P = P2,4
3
2
v4 = 2e1
v1 = 0
Figure 9. The polygon of a rational normal scroll and its normal fan
In general, the polygon Pa,b has a + b + 2 lattice points and gives the map
(C )2 P a+b+1 ,
(s,t) 7 (1, s, s2 , . . . , s a ,t, st, s2 t, . . . , s b t)
such that Sa,b = XPa,b is the Zariski closure of the image. To describe the image, we
rewrite the map as
C P1 P a+b+1 ,
(s, , ) 7 (, s, s2 , . . . , s a , , s, s2 , . . . , s b ).
85
When (, ) = (1, 0), the map is s 7 (1, s, s2 , . . . , s a , 0, . . . , 0), which is the rational
normal curve Ca mapped to the first a + 1 coordinates of P a+b+1 . In the same
way, (, ) = (0, 1) gives the rational normal curve Cb mapped to the last b + 1
coordinates of P a+b+1 . If we think of these two curves as the edges of a scroll,
then fixing s gives a point on each edge, and letting (, ) P1 vary gives the line
of the scroll connecting the two points. So it really is a scroll!
An important observation is that the normal fan depends only on the difference
b a, since this determines the slope of the slanted edge of Pa,b . If we denote the
difference by r N, it follows that as abstract toric varieties, we have
XP1,r+1 = XP2,r+2 = XP3,r+3 =
since they are all constructed from the same normal fan. In Chapter 3, we will see
that this is the Hirzebruch surface Hr .
But if we think of the projective surface Sa,b P a+b+1 , then a and b have a
unique meaning. For example, they have a strong influence on the defining equations of Sa,b . Let the homogeneous coordinates of P a+b+1 be x0 , . . . , xa , y0 , . . . , yb
and consider the 2 (a + b) matrix

x0 x1 xa1 y0 y1 yb1
.
x1 x2
xa y1 y2
yb
One can show that I(Sa,b ) C[x0 , . . . , xa , y0 , . . . , yb ] is generated by the 22 minors

of this matrix (see [130, Ex. 9.11], for example).
Example 2.3.16 is another example of a determinantal variety, as is the rational

normal curve from Example 2.0.1. Note that the rational normal curve Cd comes
from the polytope [0, d] = d1 , where the underlying toric variety is just P1 .
Exercises for 2.3.
2.3.1. This exercise will use the same notation as the proof of Theorem 2.3.1.
(a) Let Hu,a be a supporting hyperplane of a vertex mi P. Prove that Hu,0 is a supporting
hyperplane of 0 Ci
(b) Prove that dim Ci = dim P.
2.3.2. Consider the maps defined in (2.3.2).

(a) Show that these maps are inverses of each other and define a bijection between the
faces of the cone Cv and the faces of P containing v.
(b) Prove that these maps preserve dimensions, inclusions, and intersections.
(c) Explain how this exercise relates to Exercise 2.3.1.
2.3.4. Let P MR be a full dimensional lattice polytope containing 0 as an interior point,
and let P NR be its dual polytope. Prove that the normal fan P consists of the cones
over the faces of P . Hint: Exercise 2.2.1 will be useful.
86
2.3.5. Let P MR be a reflexive polytope.

(a) Prove that 0 is the only interior lattice point of P.
(b) Prove that P NR is reflexive.
2.3.6. This exercise is concerned with Example 2.3.15.
(a) Let e1 , . . . , en be the standard basis of Rn . Prove that the normal fan of the standard
n-simplex consists
Pn of the cones Cone(S) for all proper subsets S {e0 , e1 , . . . , en },
where e0 = i=1 ei . Draw pictures of the normal fan for n = 1, 2, 3.
(b) For an integer k 1, show that the toric variety Xkn P sk 1 is given by the map
k : P n P sk 1 defined using all monomials of total degree k in C[x0 , . . . , xn ].
2.3.7. Let P MR Rn be an n-dimensional lattice polytope and let Q P be a face.

Prove the following intrinsic description of the cone Q P :
Q = {u NR | hm, ui hm , ui for all m Q, m P}.
2.3.8. Prove that all lattice rectangles in the plane with edges parallel to the coordinate
axes have the same normal fan.
2.4. Properties of Projective Toric Varieties

We conclude this chapter by studying when the projective toric variety XP of a
polytope P is smooth or normal.
Normality. Recall from 2.1 that a projective variety is projectively normal if its
affine cone is normal.
Theorem 2.4.1. Let P MR be a full dimensional lattice polytope. Then:
(a) XP is normal.
(b) XP is projectively normal under the embedding given by kP if and only if kP is

normal.
Proof. Part (a) is immediate since XP is the union of affine pieces Uv , v a vertex
of P, and Uv is normal by Theorem 1.3.5. In Chapter 3 we will give an intrinsic
definition of normality that will make this argument completely rigorous.
It remains to prove part (b). The discussion following (2.1.4) shows that the
projective embedding of XP given by X(kP)M has Y((kP)M){1} as its affine cone.

By Theorem 1.3.5, this is normal if and only if the semigroup N ((kP) M) {1}
is saturated in M Z. Since ((kP) M) {1} generates the cone C(P), this is
equivalent to the assertion that the semigroup C(P) (M Z) is generated by
((kP) M) {1}. Then we are done by Lemma 2.2.14.

Smoothness. Given the results of Chapter 1, the smoothness of XP is equally easy
to determine. We need one definition.
87
Definition 2.4.2. Let P MR be a lattice polytope.
(a) Given a vertex v of P and an edge E containing v, let wE be the first lattice
point of E different from v encountered as one tranverses E starting at v. In
other words, wE v is the ray generator of the ray Cone(E v).
(b) P is smooth if for every vertex v, the vectors wE v, where E is an edge of P

containing v, form a subset of a basis of M. In particular, if dim P = dim MR ,
then the vectors wE v form a basis of M.
We can now characterize when XP is smooth.
Theorem 2.4.3. Let P MR be a full dimensional lattice polytope. Then the following are equivalent:
(a) XP is a smooth projective variety.
(b) P is a smooth fan, meaning that every cone in P is smooth in the sense of
Definition 1.2.16.
(c) P is a smooth polytope.
Proof. Smoothness is a local condition, so that a variety is smooth if and only if its
local pieces are smooth. Thus XP is smooth if and only if Uv is smooth for every
vertex v of P, and Uv is smooth if and only if v is smooth by Theorem 1.3.12.
Since faces of smooth cones are smooth and P consists of the v and their faces,
the equivalence (a) (b) follows immediately.
For (b) (c), first observe that v is smooth if and only if its dual Cv = v is
smooth. The discussion following (2.3.2) makes it easy to see that the ray generators of Cv are the vectors wE v from Definition 2.4.2. It follows immediately that
P is smooth if and only if Cv is smooth for every vertex v, and we are done.
The theorem makes it easy to check the smoothness of simple examples such
as the toric variety of the hexagon in Example 2.3.10 or the rational normal scroll
Sa,b of Example 2.3.16 (Exercise 2.4.1).
We also note the following useful fact, which you will prove in Exercise 2.4.2.
Proposition 2.4.4. Every smooth full dimensional lattice polytope P MR is very
ample.

One can also ask whether every smooth lattice polytope is normal. This is an
important open problem in the study of lattice polytopes.
Here is an example of a smooth reflexive polytope whose dual is not smooth.
Example 2.4.5. Let P = (n + 1)n (1, . . . , 1) Rn , where n is the standard
n-simplex. Thus P is the translate of (n + 1)n by (1, . . . , 1). Proposition 2.3.9
implies that P and n have the same normal fan, so that P and XP are smooth. Note
also that XP = Xn = P n .
88
The polytope P has the following interesting properties (Exercise 2.4.3). First,
P has the facet presentation
xi 1,
i = 1, . . . , n,
x1 xn 1,
so that P is reflexive with dual
P = Conv(e0 , e1 , . . . , en ),
P
e0 = e1 en .
The normal fan of

consists of cones over the faces of P. In particular, the cone
of P corresponding to the vertex e0 P is the cone
e0 = Conv(v1 , . . . , vn ),
vi = e0 + (n + 1)ei .
Figure 10 shows P and the cone e0 when n = 2.

v2
e2
e0
P
e1
P
v1
e0
Figure 10. The cone e0 of the normal fan of P
For general n, observe that vi v j = (n + 1)(ei e j ). This makes it easy to

see that Zv1 + + Zvn has index (n + 1)n1 in Zn . Thus e0 is not smooth when
n 2. It follows that the dual toric variety XP is singular for n 2. Later we
will construct XP as the quotient of P n under the action of a finite group G
(Z/(n + 1)Z)n1 .
Example 2.4.6. Consider P = Conv(0, 2e1 , e2 ) R2 . Since P is very ample, the

lattice points P Z2 = {0, e1 , 2e1 , e2 } give the map (C )2 P 3 defined by
(s,t) 7 (1, s, s2 ,t)
such that XP is the Zariski closure of the image. If P 3 has homogeneous coordinates
y0 , y1 , y2 , y3 , then we have
XP = V(y0 y2 y12 ) P 3 .
Comparing this to Example 2.0.5, we see that XP is the weighted projective space
P(1, 1, 2). Later we will learn the systematic reason why this is true.
89
The variety XP is not smooth. By working on the affine piece XP U3 , one can
check directly that (0, 0, 0, 1) is a singular point of XP .
We can also use Theorem 2.4.3 and the normal fan of P, shown in Figure 11.
One can check that the cones 0 and 1 are smooth, but 2 is not, so that P
v2
P
v0
0
2
v1
Figure 11. The polygon giving P(1, 1, 2) and its normal fan
is not a smooth fan. In terms of P, note that the vectors from v2 to the first lattice
points along the edges containing v2 do not generate Z2 . Either way, Theorem 2.4.3
implies that XP is not smooth.
If you look carefully, you will see that 2 is the only nonsmooth cone of the
normal fan P . Once we study the correspondence between cones and orbits in
Chapter 3, we will see that the nonsmooth cone 2 corresponds to the singular
point (0, 0, 0, 1) of XP .
Products of Projective Toric Varieties. Our final task is to understand the toric
variety of a product of polytopes. Let Pi (Mi )R Rni be lattice polytopes with
dim Pi = ni for i = 1, 2. This gives a lattice polytope P1 P2 (M1 M2 )R of
dimension n1 + n2 .
Replacing P1 and P2 with suitable multiples, we can assume that P1 and P2 are
very ample. This gives projective embeddings
XPi P si 1 ,
si = |Pi Mi|,
so that by Proposition 2.0.4, XP1 XP2 is a subvariety of P s1 1 P s2 1 . Using the

Segre embedding
P s1 1 P s2 1 P s1 , s = s1 s2 ,
we get an embedding
(2.4.1)
XP1 XP2 P s1 .
We can understand this projective variety as follows.
90
Theorem 2.4.7. If P1 and P2 are very ample, then:

(a) P1 P2 (M1 M2 )R is a very ample polytope with lattice points
(P1 P2 ) (M1 M2 ) = (P1 M1 ) (P2 M2 ).
Thus the integer s defined above is s = |(P1 P2 ) (M1 M2 )|.
(b) The image of the embedding XP1 P2 P s1 determined by the very ample
polytope P1 P2 equals the image of (2.4.1).
(c) XP1 P2 XP1 XP2 .
Proof. For part (a), the assertions about lattice points are clear. The vertices of
P1 P2 consist of ordered pairs (v1 , v2 ) where vi is a vertex of Pi (Exercise 2.4.4).
Given such a vertex, we have
(P1 P2 ) (M1 M2 ) (v1 , v2 ) = (P1 M1 v1 ) (P2 M2 v2 ).
Since Pi is very ample, we know that N(Pi Mi vi ) is saturated in Mi . From here,

it follows easily that P1 P2 is very ample.
For part (b), let TNi be the torus of XPi . Since TNi is Zariski dense in XPi , it follows that TN1 TN2 is Zariski dense in XP1 XP2 (Exercise 2.4.4). When combined
with the Segre embedding, it follows that XP1 XP2 is the Zariski closure of the
image of the map
TN1 TN2 P s1 s2 1
given by the characters m m , where m ranges over the s1 elements of P1 M1

and m ranges over the s2 elements of P2 M2 . When we identify TN1 TN2 with
TN1 N2 , the product m m becomes the character (m,m ) , so that the above map
coincides with the map
TN1 N2 P s1
coming from the product polytope P1 P2 (M1 M2 )R . Part (b) follows, and
part (c) is an immediate consequence.

Here is an obvious example.
Example 2.4.8. Since P n is the toric variety of the standard n-simplex n , it follows that P n P m is the toric variety of n m .
This also works for more than two factors. Thus P1 P1 P1 is the toric
variety of the cube pictured in Figure 8.
To have a complete theory of products, we need to know what happens to the

normal fan. Here is the result, whose proof is left to the reader (Exercise 2.4.5).
Proposition 2.4.9. Let Pi (Mi )R be full dimensional lattice polytopes for i = 1, 2.
Then
P1 P2 = P1 P2 .
91
Here is an easy example.

Example 2.4.10. The normal fan of an interval [a, b] R, where a < b in Z, is
given by
s
The corresponding toric variety is P1 . The cartesian product of two such intervals
is a lattice rectangle whose toric variety is P1 P1 by Theorem 2.4.7. If we set
i j = i j , then Proposition 2.4.9 gives the normal fan given in Figure 12.
10
00
11
01
Figure 12. The normal fan of a lattice rectangle giving P1 P1
We will revisit this example in Chapter 3 when we construct toric varieties

directly from fans.
Proposition 2.4.9 suggests a different way to think about the product. Let vi
range over the vertices of Pi for i = 1, 2. Then the vi are the maximal cones in the
normal fan Pi , which implies that
S
(2.4.2)
XPi = vi Uvi , i = 1, 2.
Thus
XP1 XP2 =
=
=

S
U
v
v
v2
v1
2
1
(v1 ,v2 )Uv1 Uv2
(v1 ,v2 )Uv1v2
(v1 ,v2 )U(v1 ,v2 )
= XP1 P2 .
In this sequence of equalities, the first follows from (2.4.2), the second is obvious,
the third uses Exercise 1.3.13, the fourth uses Proposition 2.4.9, and the last follows
since (v1 , v2 ) ranges over all vertices of P1 P2 .
This argument shows that we can construct cartesian products of varieties using
affine open covers, which reduces to the cartesian product of affine varieties defined
in Chapter 1. We will use this idea in Chapter 3 to define the cartesian product of
abstract varieties.
92
Exercises for 2.4.

2.4.1. Show that the hexagon P = Conv(0, e1 , e2 , 2e1 + e2 , e1 + 2e2 , 2e1 + 2e2 ) pictured in
Figure 6 and the trapezoid Pa,b pictured in Figure 9 are smooth polygons. Also, of the
polytopes shown in Figure 8, determine which ones are smooth.
2.4.3. Consider the polytope P = (n + 1)n (1, . . . , 1) from Example 2.4.5.
(a) Verify the facet presentation of P given in the example.
(b) What is the facet presentation of P ? Hint: You know the vertices of P.
(c) Let vi = e0 + (n + 1)ei , where i = 1, . . . , n and e0 = e1 en , and then set L =
Zv1 + + Zvn . Use the hint given in the text to prove Zn /L (Z/(n + 1)Z)n1 . This
shows that the index of L in Zn is (n + 1)n1, as claimed in the text.
2.4.4. Let Pi (Mi )R Rni be lattice polytopes with dim Pi = ni for i = 1, 2. Also let Si
be the set of vertices of Pi .
(a) Use supporting hyperplanes to prove that every element of S1 S2 is a vertex of P1 P2 .
(b) Prove that P1 P2 = Conv(S1 S2 ) and conclude that S1 S2 is the set of vertices of
P1 P2 .
2.4.5. The goal of this exercise is to prove Proposition 2.4.9. We know from Exercise 2.4.4
that the vertices of P1 P2 are the ordered pairs (v1 , v2 ) where vi is a vertex of Pi .
(a) Adapt the argument of part (a) of Theorem 2.4.7 to show that C(v1 ,v2 ) = Cv1 Cv2 .
Taking duals, we see that the maximal cones of P1 P2 are (v1 ,v2 ) = v1 v2 .
(b) Given rational polyhedral cones i (Ni )R and faces i i , prove that 1 2 is a
face of 1 2 and that all faces of 1 2 arise this way.
(c) Prove that P1 P2 = P1 P2 .
Pn
2.4.6. Consider positive integers 1 = q0 q1 qn with the property that qi | j=0 q j

Pn
for i = 0, . . . , n. Set ki =
j=0 q j /qi for i = 1, . . . , n and let
Pq0 ,...,qn = Conv(0, k1 e1 , k2 e2 , . . . , kn en ) (1, . . . , 1).
Prove that Pq0 ,...,qn is reflexive and explain how it relates to Example 2.4.5. We will prove
later that the toric variety of this polytope is the weighted projective space P(q0 , . . . , qn ).
2.4.7. The Sylvester sequence is defined by a0 = 2 and ak+1 = 1 + a0a1 ak . It begins
2, 3, 7, 43, 1807, . . . and is described in [251, A000058]. Now fix a positive integer n 3
and define q0 , . . . , qn by q0 = q1 = 1 and qi = 2(an1 1)/ani for i = 2, . . . , n. For n = 3
and 4 this gives 1, 1, 4, 6 and 1, 1, 12, 28, 42. Prove that q0 , . . . , qn satisfies the conditions
of Exercise 2.4.6 and hence gives a reflexive simplex, denoted SQn in [215]. This paper
proves that when n 4, SQn has the largest volume of all n-dimensional reflexive simplices
and conjectures that it also has the largest number of lattice points.
Chapter 3
Normal Toric Varieties
3.0. Background: Abstract Varieties

The projective toric varieties studied in Chapter 2 are unions of Zariski open sets,
each of which is an affine variety. We begin with a general construction of abstract
varieties obtained by gluing together affine varieties in an analogous way. The
resulting varieties will be abstract in the sense that they do not come with any
given ambient affine or projective space. We will see that this is exactly the idea
needed to construct a toric variety using the combinatorial data contained in a fan.
Sheaf theory, while important for later chapters, will make only a modest appearance here. For a more general approach to the concept of abstract variety, we
recommend standard books such as [90], [131] or [245].
Regular Functions. Let V = Spec(R) be an affine variety. In 1.0, we defined the
Zariski open subset V f = V \ V( f ) V for f R and showed that V f = Spec(R f ),
where R f is the localization of R at f . The open sets V f form a basis for theSZariski
topology on V in the sense that every open set U is a (finite) union U = f S V f
for some S R (Exercise 3.0.1).
For an affine variety, a morphism V C is called a regular map, so that the

coordinate ring of V consists of all regular maps from V to C. We now define what
it means to be regular on an open subset of V .
Definition 3.0.1. Given an affine variety V = Spec(R) and a Zariski open U V ,
we say a function : U C is regular if for all p U , there exists f p R such
that p V f p U and |V f R f p . Then define
p
OV (U ) = { : U C | is regular}.
93
Chapter 3. Normal Toric Varieties
94
The condition p V f p means that f p (p) 6= 0, and saying |V f R f p means that

p
= a p / f p p for some a p R and n p 0.
Here are some cases where OV (U ) is easy to compute.
Proposition 3.0.2. Let V = Spec(R) be an affine variety. Then:

(a) OV (V ) = R.
(b) If f R, then OV (V f ) = R f .
Proof. It is clear from Definition 3.0.1 that elements of R define regular functions
on V , hence elements of OV (V ). Conversely, if OV (V ), then for all p V there
n
n
is f p R such that p V f p and = a p / f p p R f p . The ideal I = h f p p | p V i R
satisfies
V(I) = since f p (p) 6= 0 for all p V . Hence the Nullstellensatz implies
that I = I(V(I)) = R, so there exists a finite set S V and polynomials g p for
p S such that
X
n
1=
gp f p p .
Hence =
np
pS g p f p =
pS
pS g p a p
R, as desired.
For part (b), let U V f be Zariski open. Then U is Zariski open in V , and
whenever g R satisfies Vg U , we have Vg = V f g with coordinate ring
R f g = (R f )g/ f
for all 0. These observations easily imply that
OV (U ) = OV f (U ).
(3.0.1)
Then setting U = V f gives
OV (V f ) = OV f (V f ) = R f ,
where the last equality follows by applying part (a) to V f = Spec(R f ).
Local Rings. When V = Spec(R) is an irreducible affine variety, we can describe

regular functions using the local rings OV ,p introduced in 1.0. A rational function in C(V ) is contained in the local ring OV ,p precisely when it is regular in a
neighborhood of p. It follows that whenever U V is open, we have
\
OV ,p = OV (U ).
pU
Thus regular functions on U are rational functions on V that are defined everywhere
on U . In particular, when U = V , Proposition 3.0.2 implies that
\
(3.0.2)
OV ,p = OV (V ) = R = C[V ].
pV
95
The Structure Sheaf of an Affine Variety. Given an affine variety V , the mapping
U 7 OV (U ),
U V open,
has the following useful properties:

When U U , Definition 3.0.1 shows that there is an obvious restriction map
U,U : OV (U ) OV (U )
defined by U,U () = |U . It follows that U,U is the identity map and that
U ,U U,U = U,U whenever U U U .
If {U } is an open cover of U V , then the sequence

Y
Y
0 OV (U )
OV (U )
OV (U U )
is exact. Here, the second arrow is defined by the restrictions U,U and the
double arrow is defined by U ,U U and U ,U U . Exactness at OV (U )
means that regular functions are determined locally, i.e., two regular functions
on U are equal if their restrictions to all U are equal. For the middle
term,
Q
exactness means that we have an equalizer: an element ( f ) OV (U )
comes from f OV (U ) if and only if the restrictions f |U U = f |U U are
equal for all , . This is true because regular functions on the U agreeing on
the overlaps U U patch together to give a regular function on U .
In the language of sheaf theory, these properties imply that OV is a sheaf of Calgebras, called the structure sheaf of V . We call (V , OV ) a ringed space over C.
Also, since (3.0.1) holds for all open sets U V f , we write
OV |V f = OV f .
In terms of ringed spaces, this means (V f , OV |V f ) = (V f , OV f ).

Morphisms. By 1.0, a polynomial mapping : V1 V2 between affine varieties corresponds to the C-algebra homomorphism : C[V2 ] C[V1 ] defined by
() = for C[V2 ]. We now extend this to open sets of affine varieties.
Definition 3.0.3. Let Ui Vi be Zariski open subsets of affine varieties for i = 1, 2.
A function : U1 U2 is a morphism if 7 defines a map
: OV2 (U2 ) OV1 (U1 ).
Thus : U1 U2 is a morphism if composing with regular functions on U2

gives regular functions on U1 . Note also that is a C-algebra homomorphism
since it comes from composition of functions.
Example 3.0.4. Suppose that : V1 V2 is a morphism according to Definition 3.0.3. If Vi = Spec(Ri ), then the above map gives the C-algebra homomorphism
R2 = OV2 (V2 ) OV1 (V1 ) = R1 .
96
By Chapter 1, the C-algebra homomorphism R2 R1 gives a map of affine varieties V1 V2 . In Exercise 3.0.3 you will show that this is the original map
: V1 V2 we started with.
Example 3.0.4 shows that when we apply Definition 3.0.3 to maps between
affine varieties, we get the same morphisms as in Chapter 1. In Exercise 3.0.3 you
will verify the following properties of morphisms:
If U is open in an affine variety V , then
OV (U ) = { : U C | is a morphism}.
Hence regular functions on U are just morphisms from U to C.

A composition of morphisms is a morphism.
An inclusion of open sets W U of an affine variety V is a morphism.
Morphisms are continuous in the Zariski topology.
We say that a morphism : U1 U2 is an isomorphism if is bijective and its

inverse function 1 : U2 U1 is also a morphism.
Gluing Together Affine Varieties. We now are ready to define abstract varieties by
gluing together open subsets of affine varieties. The model is what happens for P n .
Recall from 2.0 of that P n is covered by open sets

x
xi+1
xn
Ui = P n \ V(xi ) = Spec C xx0i , . . . , i1
xi , xi , . . . , xi
for i = 0, . . . , n. Each Ui is a copy of Cn that uses a different set of variables. For
i 6= j, we glue together these copies as follows. We have open subsets
(Ui ) x j Ui
(3.0.3)
(U j ) xi U j ,
and
xj
xi
and we also have the isomorphism
g ji : (Ui ) x j (U j ) xi
(3.0.4)
xj
xi
since both give the same open set Ui U j in P n . The notation g ji was chosen so
that g ji (x) means x Ui since the index i is closest to x, hence g ji (x) U j . At the
level of coordinate rings, g ji comes from the isomorphism

x0
x
x j+1
xi1 xi+1
xn
xn x
gji : C xx0j , . . . , j1
x j , x j , . . . , x j xi C xi , . . . , xi , xi , . . . , xi j
xj
xi
defined by
xk
xj
xk x j
xi / xi
(k 6= j)
and

xi 1
xj
xj
xi .
We can turn this around and start from the affine varieties Ui Cn given above
and glue together the open sets in (3.0.3) using the isomorphisms g ji from (3.0.4).
This gluing is consistent since gi j = g1
ji and gki = gk j g ji wherever all three maps
are defined. The result of this gluing is the projective space P n .
97
To generalize this, suppose we have a finite collection {V } of affine varieties

and for all pairs , we have Zariski open sets V V and isomorphisms g :
V V satisfying the following compatibility conditions:
g = g1
for all pairs , .
g (V V ) = V V and g = g g on V V for all , , .
The notation g means that in the expression g (x), the point x lies in V since
is the index closest to x, and the result g (x) lies in V .
We are now ready to glue. Let Y be the disjoint union of the V and define
a relation on Y by a b if and only if a V , b V for some , with b =
g (a). The first compatibility condition shows that is reflexive and symmetric;
the second shows that it is transitive. Hence is an equivalence relation and we
can form the quotient space X = Y / with the quotient topology. For each , let
U = {[a] X | a V }.
Then U X is an open set and the map h (a) = [a] defines a homeomorphism
h : V U X . Thus X locally looks like an affine variety.
Definition 3.0.5. We call X the abstract variety determined by the above data.
An abstract variety X comes equipped with the Zariski topology whose open
sets are those sets that restrict to open sets in each U . The Zariski closed subsets
Y X are called subvarieties of X . We say that X is irreducible if it is not the union
of two proper subvarieties. One can show that X is a finite union of irreducible
subvarieties X = Y1 Ys such that Yi 6 Y j for i 6= j. We call the Yi the irreducible
components of X .
Here are some examples of Definition 3.0.5.
Example 3.0.6. We saw above that P n can be obtained by gluing together the
open sets (3.0.3) using the isomorphisms gi j from (3.0.4). This shows that P n
is an abstract variety with affine open subsets Ui P n . More generally, given a
projective variety V P n , we can cover V with affine open subsets V Ui , and the
gluing implicit in equation (2.0.8). We conclude that projective varieties are also
abstract varieties.
Example 3.0.7. In a similar way, P n Cm can be viewed as gluing affine spaces

Ui Cm Cn+m along suitable open subsets. Thus P n Cm is an abstract variety,
and the same is true for subvarieties V P n Cm .
Example 3.0.8. Let V0 = C2 = Spec(C[u, v]) and V1 = C2 = Spec(C[w, z]), with

V10 = V0 \ V(v) = Spec(C[u, v]v )
V01 = V1 \ V(z) = Spec(C[w, z]z )
98
and gluing data

g10 : V10 V01 coming from the C-algebra homomorphism
g10 : C[w, z]z C[u, v]v defined by w 7 uv and z 7 1/v
and
g01 : V01 V10 coming from the C-algebra homomorphism
g01 : C[u, v]v C[w, z]z defined by u 7 wz and v 7 1/z.
One checks that g01 = g1

10 , and the other compatibility condition is satisfied since
there are only two Vi . It follows that we get an abstract variety X .
The variety X has another description. Consider the product P1 C2 with
homogeneous coordinates (x0 , x1 ) on P1 and coordinates (x, y) on C2 . We will
identify X with the subvariety W = V(x0 y x1 x) P1 C2 , called the blowup of
C2 at the origin, and denoted Bl0 (C2 ). First note that P1 C2 is covered by
U0 C2 = Spec(C[x1 /x0 , x, y])
and U1 C2 = Spec(C[x0 /x1 , x, y]).
Then W is covered by W0 = W (U0 C2 ) and W1 = W (U1 C2 ). Also,

W0 = V(y (x1 /x0 )x) U0 C2 ,
which gives the coordinate ring

C[x1 /x0 , x, y]/hy (x1 /x0 )xi C[x, x1 /x0 ]
via y 7 (x1 /x0 )x.
C[x0 /x1 , x, y]/hx (x0 /x1 )yi C[y, x0 /x1 ]
via x 7 (x0 /x1 )y.
Similarly, W1 = V(x (x0 /x1 )y) U1 C has coordinate ring

2
You can check that these are glued together in W in exactly the same way V0 and
V1 are glued together in X . We will generalize this example in Exercise 3.0.8.
Morphisms Between Abstract
Varieties.
S Let X and Y be abstract varieties with
S
affine open covers X = U and Y = U . A morphism : X Y is a Zariski
continuous mapping such that the restrictions
|U 1 (U ) : U 1 (U ) U
are morphisms in the sense of Definition 3.0.3.

The Structure Sheaf of an Abstract Variety. Let U be an an open subset of an
abstract variety X and set W = h1
(U U ) V . Then a function : U C is
regular if
h |W : W C
is regular for all . The compatibility conditions ensure that this is well-defined,
so that one can define
OX (U ) = { : U C | is regular}.
99
This gives the structure sheaf OX of X . Thus an abstract variety is really a ringed
space (X , OX ) with a finite open covering {U } such that (U , OX |U ) is isomorphic to the ringed space (V , OV ) of the affine variety V . (We leave the definition
of isomorphism of ringed spaces to the reader.)
Open and Closed Subvarieties. Given an abstract variety X and an open subset
U , we note that U has a natural structure of an abstract variety. For an affine
open subset
SU X , U U is open in U and hence can be written as a union
U U = f S (U ) f for a finite subset S C[U ]. It follows that U is covered
by finitely many affine open subsets and thus is an abstract variety. The structure
sheaf OU is simply the restriction of OX to U , i.e., OU = OX |U . Note also that a
function : U C is regular if and only if is a morphism as defined above.
In a similar way, a closed subset Y X also gives an abstract variety. For an

affine open set U X , Y U is closed in U and hence is an affine variety. Thus
Y is covered by finitely many affine open subsets and thus is an abstract variety.
This justifies the term subvariety for closed subsets of an abstract variety. The
structure sheaf OY is related to OX as follows. The inclusion i : Y X is a morphism. Let i OY be the sheaf on X defined by i OY (U ) = OY (U Y ). Restricting
functions on X to functions on Y gives a map of sheaves OX i OY whose kernel
is the subsheaf IY OX of functions vanishing on Y , meaning
IY (U ) = { f OX (U ) | f (p) = 0 for all p Y U }.
In the language of Chapter 6, we have an exact sequence of sheaves
0 IY OX i OY 0.
All of the types of variety introduced so far can be subsumed under the concept of abstract variety. From now on, we will usually be thinking of abstract
varieties. Hence we will usually say variety rather than abstract variety.
Local Rings and Rational Functions. Let p be a point of an affine variety V .
Elements of the local ring OV ,p are quotients f /g in a suitable localization with
f , g C[V ] and g(p) 6= 0. It follows that Vg is a neighborhood of p in V and f /g is
a regular function on Vg . In this way, we can think of elements of OV ,p as regular
functions defined in a neighborhood of p.
This idea extends to the abstract case. Given a point p of a variety X and
neighborhoods U1 ,U2 of p, regular functions fi : Ui C are equivalent at p, written
f1 f2 , if there is a neighbhorhood p U U1 U2 such that f1 |U = f2 |U .
Definition 3.0.9. Let p be a point of a variety X . Then
OX,p = { f : U C | U is a neighborhood of p in X }/
is the local ring of X at p.
100
Every OX,p has a well-defined value (p). It is not difficult to see that
OX,p is a local ring with unique maximal ideal
mX,p = { OX,p | (p) = 0}.
The local ring OX,p can also be defined as the direct limit
OX,p = lim OX (U )
pU
over all neighborhoods of p in X (see Definition 6.0.1).

When X is irreducible, we can also define the field of rational functions C(X ).
A rational function on X is a regular function f : U C defined on a nonempty
Zariski open set U X , and two rational functions on X are equivalent if they
agree on a nonempty Zariski open subset. In Exercise 3.0.4 you will show that this
relation is an equivalence relation and that the set of equivalence classes is a field,
called the function field of X , denoted C(X ).
Normal Varieties. We return to the notion of normality introduced in Chapter 1.
Definition 3.0.10. A variety X is called normal if it is irreducible and the local
rings OX,p are normal for all p X .
At first glance, this looks different from the definition given for affine varieties
in Definition 1.0.3. In fact, the two notions are equivalent in the affine case.
Proposition 3.0.11. Let V be an irreducible affine variety. Then C[V ] is normal if
and only if the local rings OV ,p are normal for all p V .
Proof. If OV ,p is normal for all p, then (3.0.2) shows that C[V ] is an intersection
of normal domains, all of which have the same field of fractions. Since such an
intersection is normal by Exercise 1.0.7, it follows that C[V ] is normal.
For the converse, suppose that C[V ] is normal and let C(V ) satisfy
k + a1 k1 + + ak = 0,
ai OV ,p .
Write ai = gi / fi with gi , fi C[V ] and fi (p) 6= 0. The product f = f1 fk has

the properties that ai C[V ] f and f (p) 6= 0. The localization C[V ] f is normal by
Exercise 1.0.7 and is contained in OV ,p since f (p) 6= 0. Hence C[V ] f OV ,p .
This completes the proof.

Here is a consequence of Proposition 3.0.11 and Definition 3.0.10.
Proposition 3.0.12. Let X be an irreducible variety with a cover consisting of
affine open sets V . Then X is normal if and only if each V is normal.
101
Smooth Varieties. For an affine variety V , the definition of a smooth point p V

(Definition 1.0.7) used Tp (V ), the Zariski tangent space of V at p, and dim p V ,
the maximum dimension of an irreducible component of V containing p. You will
show in Exercise 3.0.2 that Tp (X ) and dim p X are well-defined for a point p X of
a general variety.
Definition 3.0.13. Let X be a variety. A point p X is smooth if dim Tp (X ) =
dim p X , and X is smooth if every point of X is smooth.
Products of Varieties. As another example of abstract varieties and gluing, we
indicate why the product X1 X2 of varieties X1 and X2 also has the structure of
a variety. In 1.0 we constructed the product of affine varieties. From here, it is
relatively routine to see that if X1 is obtained by gluing together affine varieties U
and X2 is obtained by gluing together affines U , then X1 X2 is obtained by gluing
together the U U in the corresponding fashion. Furthermore, X1 X2 has the
correct universal mapping property. Namely, given a diagram
W
X1 X2
"
/ X1
$
X2
where i : W Xi are morphisms, there is a unique morphism : W X1 X2

(the dotted arrow) that makes the diagram commute.
Example 3.0.14. Let us construct the product P1 C2 . Write P1 = V0 V1 where
V0 = Spec(C[u]) and V1 = Spec(C[v]), with the gluing given by
Then
P1 C2
C[v]v C[u]u ,
v 7 1/u.
is constructed from
U0 C2 = Spec(C[u] C C[x, y]) C3
U1 C2 = Spec(C[v] C C[x, y]) C3 ,

with gluing given by
(U0 C2 )u (U1 C2 )v
corresponding to the obvious isomorphism of coordinate rings.
Separated Varieties. From the point of view of the classical topology, arbitrary
gluings can lead to varieties with some strange properties.
Example 3.0.15. In Example 3.0.14 we saw how to construct P1 from affine varieties V0 = Spec(C[u]) C and V1 = Spec(C[v]) C with the gluing given by
v 7 1/u on open sets C (V0 )u V0 and C (V1 )v V1 . This expresses P1 as
102
consisting of C plus two additional points. But now consider the abstract variety
arising from the gluing map
(V0 )u (V1 )v
that corresponds to the map of C-algebras defined by v 7 u. As before, the glued

variety X consists of C together with two additional points. However here we
have a morphism : X C whose fiber 1 (a) over a C contains one point,
but whose fiber over 0 consists of two points, p1 corresponding to 0 V0 and p2
corresponding to 0 V1 . If U1 ,U2 are classical open sets in X with p1 U1 and
p2 U2 , then U1 U2 6= . So the classical topology on X is not Hausdorff.
Since varieties are rarely Hausdorff in the Zariski topology (Exercise 3.0.5), we
need a different way to think about Example 3.0.15. Consider the product X X
and the diagonal mapping : X X X defined by (p) = (p, p) for p X . For
X from Example 3.0.15, there is a morphism X X C whose fiber over over 0
consists of the four points (pi , p j ). Any Zariski closed subset of X X containing
one of these four points must contain all of them. The image of the diagonal
mapping contains (p1 , p1 ) and (p2 , p2 ), but not the other two, so the diagonal is
not Zariski closed. This example motivates the following definition.
Definition 3.0.16. We say a variety X is separated if the image of the diagonal
map : X X X is Zariski closed in X X .
For instance, Cn is separated because the image of the diagonal in Cn Cn =
Spec(C[x1 , . . . , xn , y1 , . . . , yn ]) is the affine variety V(x1 y1 , . . . , xn yn ). Similarly,
any affine variety is separated.
The connection between failure of separatedness and failure of the Hausdorff
property in the classical topology seen in Example 3.0.15 is a general phenomenon.
Theorem 3.0.17. A variety is separated if and only if it is Hausdorff in the classical
topology.

Here are some additional properties of separated varieties (Exercise 3.0.6).
Proposition 3.0.18. Let X be a separated variety. Then:
(a) If f , g : Y X are morphisms, then {y Y | f (y) = g(y)} is Zariski closed in Y .
(b) If U ,V are affine open subsets of X , then U V is also affine.
The requirement that X be separated is often included in the definition of an

abstract variety. When this is done, what we have called a variety is sometimes
called a pre-variety.
Fiber Products. Finally in this section, we will define fiber products of varieties,
a construction required for the discussion of proper morphisms in 3.4. First, if we
103
have mappings of sets f : X S and g : Y S, then the fiber product X S Y is

defined to be
(3.0.5)
X S Y = {(x, y) X Y | f (x) = g(y)}.
The fiber product construction gives a very flexible language for describing ordinary products, intersections of subsets, fibers of mappings, the set where two
mappings agree, and so forth:
If S is a point, then X S Y is the ordinary product X Y .
If X ,Y are subsets of S and f , g are the inclusions, then X S Y X Y .
If Y = {s} S, then X S Y f 1 (s).
The third property is the reason for the name. All are easy exercises that we leave
to the reader.
In analogy with the universal mapping property of the product discussed above,
the fiber product has the following universal property. Whenever we have mappings 1 : W X and 2 : W Y such that f 1 = g 2 , there is a unique
: W X S Y that makes the following diagram commute:
W
X S Y
!
1
/X
%
/ S.
Equation (3.0.5) defines X S Y as a set. To prove that X S Y is a variety, we

assume for simplicity that S is separated. Then f : X S and g : Y S give a
morphism ( f , g) : X Y S S, and one easily checks that
X S Y = ( f , g)1 ((S)),
where (S) S S is the diagonal. This is closed in S S since S is separated,
and it follows that X S Y is closed in X Y and hence has a natural structure as
a variety. From here, it is straightforward to show that X S Y has the desired universal mapping property. Proving that X S Y is a variety when S is not separated
takes more work and will not be discussed here.
In the affine case, we can also describe the coordinate ring of X S Y . Let
X = Spec(R1 ), Y = Spec(R2 ), and S = Spec(R). The morphisms f , g correspond
to ring homomorphisms f : R R1 , g : R R2 . Hence both R1 , R2 have the
structure of R-modules, and we have the tensor product R1 R R2 . This is also a
finitely generated C-algebra, though it may have nilpotents (Exercise 3.0.9). To
get a coordinate ring, we need to take the quotient by the ideal N of all nilpotents.
104
Then one can prove that

X S Y = Spec(R1 R R2 /N).
We can avoid worrying about nilpotents by constructing X S Y as the affine scheme
Spec(R1 R R2 ). Interested readers can learn about the construction of fiber products as schemes in [90, I.3.1] and [131, pp. 8789].
Exercises for 3.0.
3.0.1. Let V = Spec(R) be an affine variety.
(a) Show that every ideal I R can be written in the form I = h f1 , . . . , fs i, where fi R.
(This is the Hilbert basis theorem in R.)
(b) Let W V be a subvariety. Show that the complement of W in V can be written as a
union of a finite collection of open affine sets of the form V f .
(c) Deduce that every open cover of V (in the Zariski topology) has a finite subcover.
(This says that affine varieties are quasicompact in the Zariski topology.)
3.0.2. As in the affine case, we want to say a variety X is smooth at p if dim Tp (X) =
dim p X. In this exercise, you will show that this is a well-defined notion.
(a) Show that if p X is in the intersection of two affine open sets V V , then the
Zariski tangent spaces TV ,p and TV ,p are isomorphic as vector spaces over C.
(b) Show that dim p X is a well-defined integer.
(c) Deduce that the proposed notion of smoothness at p is well-defined.
3.0.3. This exercise explores some properties of the morphisms defined in Definition 3.0.3.
(a) Prove the claim made in Example 3.0.4. Hint: Take a point p V1 and define m p =
{ f R1 | f (p) = 0}. Then describe ( )1 (m p ) in terms of (p).
(b) Prove the properties of morphisms listed on page 96.

3.0.4. Let X be an irreducible abstract variety.
(a) Let f , g be rational functions on X. Show that f g if f |U = g|U for some nonempty
open set U X is an equivalence relation.
(b) Show that the set of equivalence classes of the relation in part (a) is a field.
(c) Show that if U X is a nonempty open subset of X, then C(U) C(X).
3.0.5. Show that a variety is Hausdorff in the Zariski topology if and only if it consists of
finitely many points.
3.0.6. Consider Proposition 3.0.18.
(a) Prove part (a) of the proposition. Hint: Show first that if F : Y X X is defined by
F(y) = ( f (y), g(y)), then Z = F 1 ((X)).
(b) Prove part (b) of the proposition. Hint: Show first that U V can be identified with
(X) (U V ) X X.
3.0.7. Let V = Spec(R) be an affine variety. The diagonal mapping : V V V corresponds to a C-algebra homomorphism R C R R. Which one? Hint: Consider the
universal mapping property of V V .
3.1. Fans and Normal Toric Varieties
105
3.0.8. In this exercise, we will study an important variety in P n1 Cn , the blowup of

Cn at the origin, denoted Bl0 (Cn ). This generalizes Bl0 (C2 ) from Example 3.0.8. Write
the homogeneous coordinates on P n1 as x0 , . . . , xn1 , and the affine coordinates on Cn as
y1 , . . . , yn . Let
(3.0.6)
W = Bl0 (Cn ) = V(xi1 y j x j1 yi | 1 i < j n) P n1 Cn .
Let Ui1 , i = 1, . . . , n, be the standard affine opens in P n1 :

Ui1 = P n1 \ V(xi1 ),
i = 1, . . . , n (note the slightly non-standard indexing). So the Ui1 Cn form a cover of
P n1 Cn .
(a) Show that for each i = 1, . . . , n, Wi1 = W (Ui1 Cn )

x0
xi2 xi
xn1
Spec C
,...,
,
,...,
, yi
xi1
xi1 xi1
xi1
using the equations (3.0.6) defining W .
(b) Give the gluing data for identifying the subsets Wi1 \ V(x j1 ) and W j1 \ V(xi1 ).
3.0.9. Let V = V(y 2 x) C2 and consider the morphism : V C given by projection
onto the x-axis. We will study the fibers of .
(a) As noted in the text, the fiber 1 (0) = {(0, 0)} can be represented as the fiber
product {0} C V . In terms of coordinate rings, we have {0} = Spec(C[x]/hxi),
C = Spec(C[x]) and V = Spec(C[x, y]/hy 2 xi. Prove that
C[x]/hxi C[x] C[x, y]/hy 2 xi C[y]/hy 2 i.
Thus, the coordinate rings C[x]/hxi, C[x] and C[x, y]/hy 2 xi lead to a tensor product
that has nilpotents and hence cannot be a coordinate ring.
(b) If a 6= 0 in C, then 1 (a) = {(a, a)}. Show that the analogous tensor product is
C[x]/hx ai C[x] C[x, y]/hy 2 xi C[y]/hy 2 ai
C[y]/hy ai C[y]/hy + ai.

This has no nilpotents and hence is the coordinate ring of 1 (a).
What happens in part (a) is that the two square roots coincide, so that we get only one
point with multiplicity 2. The multiplicity information is recorded in the affine scheme
Spec(C[y]/hy 2 i). This is an example of the power of schemes.

In this section we construct the toric variety X corresponding to a fan . We will
also relate the varieties X to many of the examples encountered previously, and
we will see how properties of the fan correspond to properties such as smoothness
and compactness of X .
106
The Toric Variety of a Fan. A toric variety continues to mean the same thing as in
Chapters 1 and 2, although we now allow abstract varieties as in 3.0.
Definition 3.1.1. A toric variety is an irreducible variety X containing a torus
TN (C )n as a Zariski open subset such that the action of TN on itself extends to an
algebraic action of TN on X . (By algebraic action, we mean an action TN X X
given by a morphism.)
The other ingredient in this section is a fan in the vector space NR .
Definition 3.1.2. A fan in NR is a finite collection of cones NR such that:
(a) Every is a strongly convex rational polyhedral cone.
(b) For all , each face of is also in .
(c) For all 1 , 2 , the intersection 1 2 is a face of each (hence also in ).
Furthermore, if is a fan, then:

S
The support of is || = NR .
(r) is the set of r-dimensional cones of .
We have already seen some examples of fans. Theorem 2.3.2 shows that the
normal fan P of a full dimensional lattice polytope P MR is a fan in the sense
of Definition 3.1.2. However, there exist fans that are not equal to the normal fan
of any lattice polytope. An example of such a fan will be given in Example 4.2.13.
We now show how the cones in any fan give the combinatorial data necessary
to glue a collection of affine toric varieties together to yield an abstract toric variety.
By Theorem 1.2.18, each cone in gives the affine toric variety
U = Spec(C[S ]) = Spec(C[ M]).
Recall from Definition 1.2.5 that a face is given by = Hm , where m

and Hm = {u NR | hm, ui = 0} is the hyperplane defined by m. In Chapter 1, we
proved two useful facts:
First, Proposition 1.3.16 used the equality
(3.1.1)
S = S + Z(m)
to show that C[S ] is the localization C[S ] m . Thus U = (U ) m when .

Second, if = 1 2 , then Lemma 1.2.13 implies that
(3.1.2)
1 Hm = = 2 Hm ,
for some m 1 (2 ) M. This shows that

(3.1.3)
U1 (U1 ) m = U = (U2 ) m U2 .
The following proposition gives an additional property of the S and their

semigroup rings that we will need.
107
Proposition 3.1.3. If 1 , 2 and = 1 2 , then

S = S1 + S2 .
Proof. The inclusion S1 + S2 S follows directly from the general fact that
1 + 2 = (1 2 ) = . For the reverse inclusion, take p S and assume
that m 1 (2 ) M satisfies (3.1.2). Then (3.1.1) applied to 1 gives p =
q + (m) for some q S1 and N. But m 2 implies m S2 , so that
p S1 + S2 .

This result is sometimes called the separation lemma and is a key ingredient in
showing that the toric varieties X are separated in the sense of Definition 3.0.16.
Example 3.1.4. Let 1 = Cone(e1 + e2 , e2 ) (as in Exercise 1.2.11), and let 2 =
Cone(e1 , e1 + e2 ) in NR = R2 . Then = 1 2 = Cone(e1 + e2 ). We show the
dual cones 1 = Cone(e1 , e1 + e2 ), 2 = Cone(e1 e2 , e2 ), and = 1 + 2
in Figure 1.
1
1
2
Figure 1. The cones 1 , 2 , and their duals
The dark shaded region on the right is 1 2 . Note = 1 Hm = 2 Hm ,

where m = e1 + e2 1 and m = e1 e2 2 . Since S is the set of all sums
m + m with m 1 M and m 2 M, we see that S = S1 + S2 .
Now consider the collection of affine toric varieties U = Spec(C[S ]), where
runs over all cones in a fan . Let 1 and 2 be any two of these cones and let
= 1 2 . By (3.1.3), we have an isomorphism
g2 ,1 : (U1 ) m (U2 ) m
which is the identity on U . By Exercise 3.1.1, the compatibility conditions as in

3.0 for gluing the affine varieties U along the subvarieties (U ) m are satisfied.
Hence we obtain an abstract variety X associated to the fan .
Theorem 3.1.5. Let be a fan in NR . The variety X is a normal separated toric
variety.
108
Proof. Since each cone in is strongly convex, {0} N is a face of all

. Hence we have TN = Spec(C[M]) (C )n U for all . These tori are all
identified by the gluing, so we have TN X . We know from Chapter 1 that each
U has an action of TN . The gluing isomorphism g2 ,1 reduces to the identity
mapping on C[S1 2 ]. Hence the actions are compatible on the intersections of
every pair of sets in the open affine cover, and patch together to give an algebraic
action of TN on X .
The variety X is irreducible because all of the U are irreducible affine toric
varieties containing the torus TN . Furthermore, U is a normal affine variety by
Theorem 1.3.5. Hence the variety X is normal by Proposition 3.0.12.
To see that X is separated it suffices to show that for each pair of cones 1 , 2
in , the image of the diagonal map
: U U1 U2 ,
= 1 2
is Zariski closed (Exercise 3.1.2). But comes from the C-algebra homomorphism
: C[S1 ] C C[S2 ] C[S ]
defined by m n 7 m+n . By Proposition 3.1.3, is surjective, so that
C[S ] (C[S1 ] C C[S2 ])/ker( ).
Hence the image of is a Zariski closed subset of U1 U2 .
Toric varieties were originally known as torus embeddings, and the variety X
would be written TN emb() in older references such as [218]. Other commonly
used notations are X (), or X (), if the fan is denoted by . When we want to
emphasize the dependence on the lattice N, we will write X as X,N .
Many of the toric varieties encountered in Chapters 1 and 2 come from fans.
For example, Theorem 1.3.5 implies that a normal affine toric variety comes from
a fan consisting of a single cone together with all of its faces. Furthermore, the
projective toric variety associated to a lattice polytope in Chapter 2 comes from a
fan. Here is the precise result.
Proposition 3.1.6. Let P MR be a full dimensional lattice polytope. Then the
projective toric variety XP XP , where P is the normal fan of P.
Proof. When P is very ample, this follows immediately from the description of the
intersections of the affine open pieces of XP in Proposition 2.3.13 and the definition
of the normal fan P . The general case follows since the normal fans of P and kP
are the same for all positive integers k.

In general, every separated normal toric variety comes from a fan. This is a
consequence of a theorem of Sumihiro from [265].
109
Theorem 3.1.7 (Sumihiro). Let the torus TN act on a normal separated variety X .
Then every point p X has a TN -invariant affine open neighborhood.

Corollary 3.1.8. Let X be a normal separated toric variety with torus TN . Then
there exists a fan in NR such that X X .
Proof. The proof will be sketched in Exercise 3.2.11 after we have developed the
properties of TN -orbits on toric varieties.

Examples. We now turn to some concrete examples. Many of these are toric varieties already encountered in previous chapters.
Example 3.1.9. Consider the fan in NR = R2 in Figure 2, where N = Z2 has
standard basis e1 , e2 . This is the normal fan of the simplex 2 as in Example 2.3.4.
Here we show all points in the cones inside a rectangular viewing box (all figures
of fans in the plane in this chapter will be drawn using the same convention.)
Figure 2. The fan for P 2
From the discussion in Chapter 2, we expect X P 2 , and we will show

this in detail. The fan has three 2-dimensional cones 0 = Cone(e1 , e2 ), 1 =
Cone(e1 e2 , e2 ), and 2 = Cone(e1 , e1 e2 ), together with the three rays
i j = i j for i 6= j, and the origin. The toric variety X is covered by the
affine opens
U0 = Spec(C[S0 ]) Spec(C[x, y])
U1 = Spec(C[S1 ]) Spec(C[x1 , x1 y])
U2 = Spec(C[S2 ]) Spec(C[xy1 , y1 ]).

Moreover, by Proposition 3.1.3, the gluing data on the coordinate rings is given by
g10 : C[x, y]x C[x1 , x1 y]x1
g20 : C[x, y]y C[xy1 , y1 ]y1
g21 : C[x1 , x1 y]x1 y C[xy1 , y1 ]xy1 .
110
It is easy to see that if we use the usual homogeneous coordinates (x0 , x1 , x2 ) on P 2 ,

then x 7 xx01 and y 7 xx02 identifies the standard affine open Ui P 2 with Ui X .
Hence we have recovered P 2 as the toric variety X .
Example 3.1.10. Generalizing Example 3.1.9, let NR = Rn , where N = Zn has

standard basis e1 , . . . , en . Set
e0 = e1 e2 en
and let be the fan in NR consisting of the cones generated by all proper subsets
of {e0 , . . . , en }. This is the normal fan of the n-simplex n , and X P n by Example 2.3.15 and Exercise 2.3.6. You will check the details to verify that this gives
the usual affine open cover of P n in Exercise 3.1.3.
Example 3.1.11. We classify all 1-dimensional normal toric varieties as follows.
We may assume N = Z and NR = R. The only cones are the intervals 0 = [0, )
and 1 = (, 0] and the trivial cone = {0}. It follows that there are only four
possible fans, which gives the following list of toric varieties:
{ }, which gives C
{0 , } and {1 , }, both of which give C
{0 , 1 , }, which gives P1 .
Here is a picture of the fan for P1 :
1
This is the fan of Example 3.1.10 when n = 1.
Example 3.1.12. By Example 2.4.8, P n P m is the toric variety of the polytope

n m . The normal fan of n m is the product of the normal fans of each
factor (Proposition 2.4.9). These normal fans are described in Example 3.1.10. It
follows that the product fan gives X P n P m .
When n = m = 1, we obtain the fan R2 NR pictured in Figure 3 on the

next page. Here, we can use an elementary gluing argument to show that this fan
gives P1 P1 . Label the 2-dimensional cones i j = i j as above. Then
Spec(C[S00 ]) C[x, y]
Spec(C[S10 ]) C[x1 , y]
Spec(C[S11 ]) C[x1 , y1 ]
Spec(C[S01 ]) C[x, y1 ].
We see that if U0 and U1 are the standard affine open sets in P1 , then Ui j Ui U j
and it is easy to check that the gluing makes X P1 P1 .
111
10
00
11
01
Figure 3. A fan with X P1 P1
Example 3.1.13. Let N = N1 N2 , with N1 = Zn and N2 = Zm . Let 1 in (N1 )R

be the fan giving P n , but let 2 be the fan consisting of the cone Cone(e1 , . . . , em )
together with all its faces. Then = 1 2 is a fan in NR and the the corresponding toric variety is X P n Cm . The case P1 C2 was studied in Example 3.0.14.
Examples 3.1.12 and 3.1.13 are special cases of the following general construction, whose proof will be left to the reader (Exercise 3.1.4).
Proposition 3.1.14. Suppose we have fans 1 in (N1 )R and 2 in (N2 )R . Then
1 2 = {1 2 | i i }
is a fan in (N1 )R (N2 )R = (N1 N2 )R and
X1 2 X1 X2 .
Example 3.1.15. The two cones 1 and 2 in NR = R2 from Example 3.1.4 (see
Figure 1), together with their faces, form a fan . By comparing the descriptions
of the coordinate rings of Vi given there with what we did in Example 3.0.8, it is
easy to check that X W , where W P1 C2 is the blowup of C2 at the origin,
defined as W = V(x0 y x1 x) (Exercise 3.1.5).
Generalizing this, let N = Zn with standard basis e1 , . . . , en and then set e0 =

e1 + + en . Let be the fan in NR consisting of the cones generated by all subsets
of {e0 , . . . , en } not containing {e1 , . . . , en }. Then the toric variety X is isomorphic
to the blowup of Cn at the origin (Exercise 3.0.8).
Example 3.1.16. Let r N and consider the fan r in NR = R2 consisting of the

four cones i shown in Figure 4 on the next page, together with all of their faces.
112
4
(1, r)
Figure 4. A fan r with Xr Hr
The corresponding toric variety Xr is covered by open affine subsets,

U1 = Spec(C[x, y]) C2
U2 = Spec(C[x, y1 ]) C2
U3 = Spec(C[x1 , xr y1 ]) C2
U4 = Spec(C[x1 , x r y]) C2 ,
and glued according to (3.1.3). We call Xr the Hirzebruch surface Hr .

Example 2.3.16 constructed the rational normal scroll Sa,b using the polygon
Pa,b with b a 1. The normal fan of Pa,b is the fan ba defined above, so that
as an abstract variety, Sa,b Hba . Note also that H0 P1 P1 .
The Hirzebruch surfaces Hr will play an important role in the classification of

smooth projective toric surfaces given in Chapter 10.
Example 3.1.17. Let q0 , . . . , qn Z>0 satisfy gcd(q0 , . . . , qn ) = 1. Consider the
weighted projective space P(q0 , . . . , qn ) introduced in Chapter 2. Define the lattice
N = Zn+1 /Z (q0 , . . . , qn ) and let ui , i = 0, . . . , n, be the images in N of the standard
basis vectors in Zn+1 , so the relation
q0 u0 + + qn un = 0
holds in N. Let be the fan made up of the cones generated by all the proper subsets of {u0 , . . . , un }. When qi = 1 for all i, we obtain X P n by Example 3.1.10.
And indeed, X P(q0 , . . . , qn ) in general. This will be proved in Chapter 5 using
the toric generalization of homogeneous coordinates in P n .
Here, we will consider the special case P(1, 1, 2), where u0 = u1 2u2 . The
fan in NR is pictured in Figure 5 on the next page, using the plane spanned by
u1 , u2 . This example is different from the ones we have seen so far. Consider 2 =
Cone(u0 , u1 ) = Cone(u1 2u2 , u1 ). Then 2 = Cone(u2 , 2u1 u2 ) M, so the
situation is similar to the case studied in Example 1.2.22. Indeed, there is a change
113
0
1
Figure 5. A fan with X P(1, 1, 2)
of coordinates defined by a matrix in GL(2, Z) that takes to the cone with d = 2

from that example. It follows that there is an isomorphism U2 V(xz y 2 ) C3
(Exercise 3.1.6). This is the rational normal cone Cb2 , hence has a singular point
at the origin. The toric variety X is singular because of the singular point in this
affine open subset.
In Example 2.4.6, we saw that the polytope P = Conv(0, 2e1 , e2 ) R2 gives
XP P(1, 1, 2) and that the normal fan P coincides with the fan shown above.
There is a dictionary between properties of X and properties of that generalizes Theorem 1.3.12 and Example 1.3.20. We begin with some terminology. The
first two items parallel Definition 1.2.16.
Definition 3.1.18. Let NR be a fan.
(a) We say is smooth (or regular) if every cone in is smooth (or regular).
(b) We say is simplicial if every cone in is simplicial.
S
(c) We say is complete if its support || = is all of NR .
Theorem 3.1.19. Let X be the toric variety defined by a fan NR . Then:

(a) X is a smooth variety if and only if the fan is smooth.
(b) X is an orbifold (that is, X has only finite quotient singularities) if and only
if the fan is simplicial.
(c) X is compact in the classical topology if and only if is complete.
Proof. Part (a) follows from the corresponding statement for affine toric varieties,
Theorem 1.3.12, because smoothness is a local property (Definition 3.0.13). In
part (b), Example 1.3.20 gives one implication. The other implication will be
proved in Chapter 11. A proof of part (c) will be given in 3.4.

The blowup of C2 at the origin (Example 3.1.15) is not compact, since the
support of the cones in the corresponding fan is not all of R2 . The Hirzebruch
114
surfaces Hr from Example 3.1.16 are smooth and compact because every cone in
the corresponding fan is smooth, and the union of the cones is R2 . The variety
P(1, 1, 2) from Example 3.1.17 is compact but not smooth. It is an orbifold (it has
only finite quotient singularities) since the corresponding fan is simplicial.
Exercises for 3.1.
3.1.1. Let be a fan in NR . Show that the isomorphisms g1 ,2 satisfy the compatibility
conditions from 0 for gluing the U together to create X .
3.1.2. Let X be a variety obtained by gluing affine open subsets {V } along open subsets
V V by isomorphisms g : V V . Show that X is separated when the image of
: V V V defined by (p) = (p, g (p)) is Zariski closed for all , .
3.1.3. Verify that if is the fan given in Example 3.1.10, then X P n .
3.1.5. Let N Zn , let e1 , . . . , en N be the standard basis and let e0 = e1 + + en . Let
be the set of cones generated by all subsets of {e0 , . . . , en } not containing {e1 , . . . , en }.
(a) Show that is a fan in NR .
(b) Construct the affine open subsets covering the corresponding toric variety X , and
give the gluing isomorphisms.
(c) Show that X is isomorphic to the blowup of Cn at the origin, described earlier in
Exercise 3.0.8. Hint: The blowup is the subvariety of P n1 Cn given by W =
V(xi y j x j yi | 1 i < j n). Cover W by affine open subsets Wi = Wxi and compare those affines with your answer to part (b).
3.1.6. In this exercise, you will verify the claims made in Example 3.1.17.
(a) Show that there is a matrix A GL(2, Z) defining a change of coordinates that takes
the cone in this example to the cone from Example 1.2.22, and find the mapping that
takes 2 to the dual cone.
(b) Show that Spec(C[S2 ]) V(xz y 2 ) C3 .
3.1.7. In NR = R2 , consider the fan with cones {0}, Cone(e1 ), and Cone(e1 ). Show
that X P1 C .
3.2. The Orbit-Cone Correspondence

In this section, we will study the orbits for the action of TN on the toric variety X .
Our main result will show that there is a bijective correspondence between cones
in and TN -orbits in X . The connection comes ultimately from looking at limit
points of the one-parameter subgroups of TN defined in 1.1.
A First Example. We introduce the key features of the correspondence between
orbits and cones by looking at a concrete example.
115
Example 3.2.1. Consider P 2 X for the fan from Figure 2 of 3.1. The torus
TN = (C )2 P 2 consists of points with homogeneous coordinates (1, s,t), s,t 6= 0.
For each u = (a, b) N = Z2 , we have the corresponding curve in P 2 :
u (t) = (1,t a ,t b ).
We are abusing notation slightly; strictly speaking, the one-parameter subgroup u
is a curve in (C )2 , but we view it as a curve in P 2 via the inclusion (C )2 P 2 .
We start by analyzing the limit of u (t) as t 0. The limit point in P 2 depends

on u = (a, b). It is easy to check that the pattern is as follows:
limit is (1,1,0)
limit is (1,0,0)
limit is (0,1,0)
limit is (1,0,1)
limit is (1,1,1)
limit is (0,0,1)
limit is (0,1,1)
Figure 6. limt0 u (t) for u = (a, b) Z2
For instance, suppose a, b > 0 in Z. These points lie in the first quadrant. Here,
it is obvious that limt0 (1,t a ,t b ) = (1, 0, 0). Next suppose that a = b < 0 in Z,
corresponding to points on the diagonal in the third quadrant. Note that
(1,t a ,t b ) = (1,t a ,t a ) (t a , 1, 1)
since we are using homogeneous coordinates in P 2 . Then a > 0 implies that
limt0 (t a , 1, 1) = (0, 1, 1). You will check the remaining cases in Exercise 3.2.1.
The regions of N described in Figure 6 correspond to cones of the fan . In
each case, the set of u giving one of the limit points equals N Relint(), where
Relint() is the relative interior of a cone . In other words, we have recovered
the structure of the fan by considering these limits!
Now we relate this to the TN -orbits in P 2 . By considering the description P 2
(C \ {0})/C , you will see in Exercise 3.2.1 that there are exactly seven TN -orbits
3
116
in P 2 :
O1 = {(x0 , x1 , x2 ) | xi 6= 0 for all i} (1, 1, 1)
O2 = {(x0 , x1 , x2 ) | x2 = 0, and x0 , x1 6= 0} (1, 1, 0)
O3 = {(x0 , x1 , x2 ) | x1 = 0, and x0 , x2 6= 0} (1, 0, 1)
O4 = {(x0 , x1 , x2 ) | x0 = 0, and x1 , x2 6= 0} (0, 1, 1)
O5 = {(x0 , x1 , x2 ) | x1 = x2 = 0, and x0 6= 0} = {(1, 0, 0)}
O6 = {(x0 , x1 , x2 ) | x0 = x2 = 0, and x1 6= 0} = {(0, 1, 0)}
O7 = {(x0 , x1 , x2 ) | x0 = x1 = 0, and x2 6= 0} = {(0, 0, 1)}.

This list shows that each orbit contains a unique limit point. Hence we obtain a
correspondence between cones and orbits O by
corresponds to O lim u (t) O for all u Relint().
t0
We will soon see that these observations generalize to all toric varieties X .
Points and Semigroup Homomorphisms. It will be convenient to use the intrinsic

description of the points of an affine toric variety U given in Proposition 1.3.1.
We recall how this works and make some additional observations:
Points of U are in bijective correspondence with semigroup homomorphisms
: S C. Recall that S = M and U = Spec(C[S ]).
For each cone we have a point of U defined by

(
1 m S = M
m S 7
0 otherwise.
This is a semigroup homomorphism since is a face of . Thus, if

m, m S and m + m S , then m, m S . We denote this point
by and call it the distinguished point corresponding to .
The point is fixed under the TN -action if and only if dim = dim NR (Corollary 1.3.3).
If is a face, then U . This follows since .
Limits of One-Parameter Subgroups. In Example 3.2.1, the limit points of oneparameter subgroups are exactly the distinguished points for the cones in the fan
of P 2 (Exercise 3.2.1). We now show that this is true for all affine toric varieties.
Proposition 3.2.2. Let NR be a strongly convex rational polyhedral cone and
let u N. Then
u lim u (t) exists in U .
t0
Moreover, if u Relint(), then limt0 u (t) = .
117
Proof. Given u N, we have
lim u (t) exists in U lim m (u (t)) exists in C for all m S
t0
t0
lim t hm,ui exists in C for all m S

t0
hm, ui 0 for all m M
u ( ) = ,
where the first equivalence is proved in Exercise 3.2.2 and the other equivalences
are clear. This proves the first assertion of the proposition.
In Exercise 3.2.2 you will also show that when u N, limt0 u (t) is the
point corresponding to the semigroup homomorphism S C defined by
m M 7 lim t hm,ui .
t0
If u Relint(), then hm, ui > 0 for all m S \ (Exercise 1.2.2), and hm, ui = 0
if m S . Hence the limit point is precisely the distinguished point .
Using this proposition, we can recover the fan from X cone by cone as in
Example 3.2.1. This is also the key observation needed for the proof of Corollary 3.1.8 from the previous section.
Let us apply Proposition 3.2.2 to a familiar example.
Example 3.2.3. Consider the affine toric variety V = V(xy zw) studied in a number of examples from Chapter 1. For instance, in Example 1.1.18, we showed that
V is the normal toric variety corresponding to a cone whose dual cone is
(3.2.1)
= Cone(e1 , e2 , e3 , e1 + e2 e3 ),
and V = Spec(C[ M]).
In Example 1.1.18, we introduced the torus T = (C )3 of V as the image of
(3.2.2)
(t1 ,t2 ,t3 ) 7 (t1 ,t2 ,t3 ,t1 t2t31 ).
Given u = (a, b, c) N = Z3 , we have the one-parameter subgroup

(3.2.3)
u (t) = (t a ,t b ,t c ,t a+bc )
contained in V , and we proceed to examine limit points using Proposition 3.2.2.

Clearly, limt0 u (t) exists in V if and only if a, b, c 0 and a + b c. These
conditions determine the cone NR given by
(3.2.4)
= Cone(e1 , e2 , e1 + e3 , e2 + e3 ).
One easily checks that (3.2.1) is the dual of this cone (Exercise 3.2.3). Note
also that u Relint() means a, b, c > 0 and a + b > c, in which case the limit
limt0 u (t) = (0, 0, 0, 0), which is the distinguished point .
118
The Torus Orbits. Now we turn to the TN -orbits in X . We saw above that each
cone has a distinguished point U X . This gives the torus orbit
O() = TN X .
In order to determine the structure of O(), we need the following lemma, which
you will prove in Exercise 3.2.4.
Lemma 3.2.4. Let be a strongly convex rational polyhedral cone in NR . Let N
be the sublattice of N spanned by the points in N, and let N() = N/N .
(a) There is a perfect pairing
h , i : M N() Z,
induced by the dual pairing h , i : M N Z.
(b) The pairing of part (a) induces a natural isomorphism

HomZ ( M, C ) TN() ,
where TN() = N() Z C is the torus associated to N().
To study O() U , we recall how t TN acts on semigroup homomorphisms.

If p U is represented by : S C, then by Exercise 1.3.1, the point t p is
represented by the semigroup homomorphism
(3.2.5)
t : m 7 m (t)(m).
Lemma 3.2.5. Let be a strongly convex rational polyhedral cone in NR . Then

O() = { : S C | (m) 6= 0 m M}
HomZ ( M, C ) TN() ,
where N() is the lattice defined in Lemma 3.2.4.

Proof. The set O = { : S C | (m) 6= 0 m M} contains and is
invariant under the action of TN described in (3.2.5).
Next observe that is the largest vector subspace of MR contained in .
Hence M is a subgroup of S = M. If O , then restricting to
m S = M yields a group homomorphism b
: M C (Exercise 3.2.5). Conversely, if
b : M C is a group homomorphism, we obtain
a semigroup homomorphism O by defining
(
b(m) if m M
(m) =
0
otherwise.
It follows that O HomZ ( M, C ).
Now consider the exact sequence
(3.2.6)
0 N N N() 0.
119
Tensoring with C and using Lemma 3.2.4, we obtain a surjection

TN = N Z C TN() = N() Z C HomZ ( M, C ).
The bijections
TN() HomZ ( M, C ) O
are compatible with the TN -action, so that TN acts transtively on O . Then O

implies that O = TN = O(), as desired.

The Orbit-Cone Correspondence. Our next theorem is the major result of this
section. Recall that the face relation holds when is a face of .
Theorem 3.2.6 (Orbit-Cone Correspondence). Let X be the toric variety of the
fan in NR . Then:
(a) There is a bijective correspondence
{cones in } {TN -orbits in X }
O() HomZ ( M, C ).
(b) Let n = dim NR . For each cone , dim O() = n dim .
(c) The affine open subset U is the union of orbits

[
U =
O( ).

(d) if and only if O() O( ), and

[
O( ) =
O(),

where O( ) denotes the closure in both the classical and Zariski topologies.
For instance, Example 3.2.1 tells us that for P 2 , there are three types of cones
and torus orbits:
The trivial cone = {(0, 0)} corresponds to the orbit O() = TN P 2 , which
satisfies dim O() = 2 = 2 dim . This is a face of all the other cones in
, and hence all the other orbits are contained in the closure of this one by
part (d). Note also that U = O() (C )2 by part (c), since there are no
cones properly contained in .
The three 1-dimensional cones give the torus orbits of dimension 1. Each is
isomorphic to C . The closures of these orbits are the coordinate axes V(xi ) in
P 2 , each a copy of P1 . Note that each is contained in two maximal cones.
The three maximal cones i in the fan correspond to the three fixed points
(1, 0, 0), (0, 1, 0), (0, 0, 1) of the torus action on P 2 . There are two of these in
the closure of each of the 1-dimensional torus orbits.
120
Proof of Theorem 3.2.6. Let O be a TN -orbit in X . Since X is covered by the

TN -invariant affine open subsets U X and U1 U2 = U1 2 , there is a unique
minimal cone with O U . We claim that O = O(). Note that part (a) will
follow immediately once we prove this claim.
To prove the claim, let O and consider those m S satisfying (m) 6= 0.
In Exercise 3.2.6, you will show that these ms lie on a face of . But faces of
are all of the form for some face by Proposition 1.2.10. In other
words, there is a face such that
{m S | (m) 6= 0} = M.
This easily implies U (Exercise 3.2.6), and then = by the minimality of .
Hence {m S | (m) 6= 0} = M, and then O() by Lemma 3.2.5. This
implies O = O() since two orbits are either equal or disjoint.
Part (b) follows from Lemma 3.2.5 and (3.2.6).
Next consider part (c). We know that U is a union of orbits. If is a face of ,
then O( ) U U implies that O( ) is an orbit contained in U . Furthermore,
the analysis of part (a) easily implies that any orbit contained in U must equal
O( ) for some face .
We now turn to part (d). We begin with the closure of O( ) in the classical
topology, which we denote O( ). This is invariant under TN (Exercise 3.2.6) and
hence is a union of orbits. Suppose that O() O( ). Then O( ) U , since
otherwise O( ) U = , which would imply O( ) U = since U is open in
the classical topology. Once we have O( ) U , it follows that by part (c).
Conversely, assume . To prove that O() O( ), it suffices to show that
O( ) O() 6= . We will do this by using limits of one-parameter subgroups as
in Proposition 3.2.2.
Let be the semigroup homomorphism corresponding to the distinguished
point of U , so (m) = 1 if m M, and 0 otherwise. Let u Relint(), and
for t C define (t) = u (t) . As a semigroup homomorphism, (t) is
m 7 m (u (t)) (m) = t hm,ui (m).
Note that (t) O( ) for all t C since the orbit of is O( ). Now let t 0.
Since u Relint(), hm, ui > 0 if m \ , and = 0 if m . It follows
that (0) = limt0 (t) exists as a point in U by Proposition 3.2.2, and represents
a point in O(). But it is also in the closure of O( ) by construction, so that
O() O( ) 6= . This establishes the first assertion of (d), and
[
O( ) =
O()

follows immediately for the classical topology.
121
It remains to show that this set is also the Zariski closure. If we intersect O( )
with an affine open subset U , parts (c) and (d) imply that
[
O( ) U =
O().

In Exercise 3.2.6, you will show that this is the subvariety V(I) U for the ideal
(3.2.7)
I = h m | m ( ) Mi C[( ) M] = S .
This easily implies that the classical closure O( ) is a subvariety of X and hence
is the Zariski closure of O( ).

Orbit Closures as Toric Varieties. In the example of P 2 , the orbit closures O( )
also have the structure of toric varieties. This holds in general. We use the notation
V ( ) = O( ).
By part (d) of Theorem 3.2.6, we have if and only if O() V ( ), and
[
V ( ) =
O().

The torus O( ) = TN( ) is an open subset of V ( ), where N( ) is defined in

Lemma 3.2.4. We will show that V ( ) is a normal toric variety by constructing
its fan. For each cone containing , let be the image cone in N( )R under
the quotient map
NR N( )R
in (3.2.6). Then
(3.2.8)
Star( ) = { N( )R | }
is a fan in N( )R (Exercise 3.2.7).

Proposition 3.2.7. For any , the orbit closure V ( ) = O( ) is isomorphic to
the toric variety XStar( ) .
Proof. This follows from parts (a) and (d) of Theorem 3.2.6 (Exercise 3.2.7).
Example 3.2.8. Consider the fan in NR = R3 shown in Figure 7 on the next page.
The support of is the cone in Figure 2 of Chapter 1, and is obtained from this
cone by adding a new 1-dimensional cone in the center and subdividing. The
orbit O( ) has dimension 2 by Theorem 3.2.6. By Proposition 3.2.7, the orbit
closure V ( ) is constructed from the cones of containing and then collapsing
to a point in N( )R = (N/N )R R2 . This clearly gives the fan for P1 P1 , so
that V ( ) P1 P1 .
122
Figure 7. The fan and its 1-dimensional cone in Example 3.2.8
A nice example of orbit closures comes from the toric variety XP of a full
dimensional lattice polytope P MR . Here, we use the normal fan P of P, which
by Theorem 2.3.2 consists of cones
(3.2.9)
Q = Cone(uF | F is a facet of P containing Q)
for each face Q P. Recall that uF is the facet normal of F.
The basic idea is that the orbit closure of V (Q ) is the toric variety of the lattice
polytope Q. Since Q need not be full dimensional in MR , we need to be careful.
The idea is to translate P by a vertex of Q so that the origin is a vertex of Q. This
affects neither P nor XP , but Q is now full dimensional in Span(Q) and is a lattice
polytope relative to Span(Q) M. This gives the toric variety XQ , which is easily
seen to be independent of how we translate to the origin. Here is our result.
Proposition 3.2.9. For each face Q P, we have V (Q ) XQ .
Proof. We sketch the proof and leave the details to reader (Exercise 3.2.8). Let
P = {m MR | hm, uF i aF for all facets F P}
be the facet presentation of P. The facets of P containing Q also contain the origin,
so that aF = 0 for these facets. This implies that
Q = Span(Q),
and then N(Q ) is dual to Span(Q) M. Note also that N(Q )R = NR /Span(Q ).
To keep track of which polytope we are using, we will write the cone (3.2.9)
associated to a face Q P as Q,P . Then XP and XQ are given by the normal fans
P = {Q ,P NR | Q P}
Q = {Q ,Q N(Q,P )R | Q Q}.
123
By Proposition 3.2.7, the toric variety V (Q ) = V (Q,P ) is determined by the fan

Star(Q,P ) = { | Q,P P }
= {Q ,P | Q,P Q ,P P } = {Q ,P | Q Q}.
Then the proposition follows once one proves that Q ,P = Q ,Q .
Final Comments. The technique of using limit points of one-parameter subgroups

to study a group action is also a major tool in Geometric Invariant Theory as in
[209], where the main problem is to construct varieties (or possibly more general
objects) representing orbit spaces for the actions of algebraic groups on varieties.
We will apply ideas from group actions and orbit spaces to the study of toric varieties in Chapters 5 and 14.
We also note the observation made in part (d) of Theorem 3.2.6 that torus orbits
have the same closure in the classical and Zariski topologies. For arbitrary subsets
of a variety, these closures may differ. A torus orbit is an example of a constructible
subset, and we will see in 3.4 that constructible subsets have the same classical
and Zariski closures since we are working over C.
Exercises for 3.2.
3.2.1. In this exercise, you will verify the claims made in Example 3.2.1 and the following
discussion.
(a) Show that the remaining limits of one-parameter subgroups P 2 are as claimed in the
example.
(b) Show that the (C )2 -orbits in P 2 are as claimed in the example.
(c) Show that the limit point equals the distinguished point of the corresponding cone
in each case.
3.2.2. Let NR be a strongly convex rational polyhedral cone. This exercise will consider limt0 f (t), where f : C TN is an arbitrary function.
(a) Prove that limt0 f (t) exists in U if and only if limt0 m ( f (t)) exists in C for all
m S . Hint: Consider a finite set of characters A such that S = NA .
(b) When limt0 f (t) exists in U , prove that the limit is given by the semigroup homomorphism that maps m S to limt0 m ( f (t)).
3.2.3. Consider the situation of Example 3.2.3.
(a) Show that the cones in (3.2.1) and (3.2.4) are dual.
(b) Identify the limits of all one-parameter subgroups in this example, and describe the
Orbit-Cone Correspondence in this case.
(c) Show that the matrix
1 1
A= 1 0
1 0
1
0
1
defines an automorphism of N Z3 and the corresponding linear map on NR maps the

cone to .
124
(d) Deduce that the affine toric varieties U and U are isomorphic. Hint: Use Proposition 1.3.15.
3.2.4. Prove Lemma 3.2.4.
3.2.5. Let O be as defined in the proof of Lemma 3.2.5. In this exercise, you will complete
the proof that O is a TN -orbit in U .
(a) Show that if O , then
b : M C is a group homomorphism.
(b) Deduce that O has the structure of a group.
(c) Verify carefully that we have an isomorphism of groups O HomZ ( M, C ).
3.2.6. This exercise is concerned with the proof of Theorem 3.2.6.

(a) Let : S C be a semigroup homomorphism giving a point of U . Prove that
{m S | (m) 6= 0} = M for some face .
(b) Show O( ) is invariant under the action of TN .
(c) Prove that O( ) U is the variety of the ideal I defined in (3.2.7).

3.2.7. Let be a cone in a fan , and let Star( ) be as defined in (3.2.8).
(a) Show that Star( ) is a fan in N( )R .
(b) Prove Proposition 3.2.7.
3.2.8. Supply the details omitted in the proof of Proposition 3.2.9.
3.2.9. Consider the action of TN on the affine toric variety U . Use parts (c) and (d) of
Theorem 3.2.6 to show that O() is the unique closed orbit of TN acting on U .
3.2.10. In Proposition 1.3.16, we saw that if is a face of the strongly convex rational polyhedral cone in NR then U = Spec(C[S ]) is an affine open subset of U = Spec(C[S ]).
In this exercise, you will prove the converse, i.e., that if and the induced map of
affine toric varieties : U U is an open immersion, then , i.e., is a face of .
(a) Let u, u N , and assume u + u . Show that
lim u (t) lim u (t) U .
t0
t0
(b) Show that limt0 u (t) and limt0 u (t) are each in U . Hint: Use the description of
points as semigroup homomorphisms.
(c) Deduce that u, u , so is a face of .
3.2.11. In this exercise, you will use Proposition 3.2.2 and Theorem 3.2.6 to deduce Corollary 3.1.8 from Theorem 3.1.7.
(a) By Theorem 3.1.7, and the results of Chapter 1, a separated toric variety has an open
cover consisting of affine toric varieties Ui = Ui for some collection of cones i . Show
that for all i, j, Ui U j is also affine. Hint: Use the fact that X is separated.
(b) Show that Ui U j is the affine toric variety corresponding to the cone = i j .
Hint: Exercise 3.2.2 will be useful.
(c) If = i j , then show that is a face of both i and j . Hint: Use Exercise 3.2.10.
(d) Deduce that X X for the fan consisting of the i and all their faces.
3.3. Toric Morphisms
125

S
Recall from
3.0
that
if
X
and
Y
are
varieties
with
affine
open
covers
X
=
U
S
and Y = U , then a morphism : X Y is a Zariski-continuous mapping such
that the restrictions
|U 1 (U ) : U 1 (U ) U
are morphisms in the sense of Definition 3.0.3 for all , .

In 1.3 we defined toric morphisms between affine toric varieties and studied
their properties. When applied to arbitrary normal toric varieties, these results yield
a class of morphisms whose construction comes directly from the combinatorics of
the associated fans. The goal of this section is to study these special morphisms.
Definition 3.3.1. Let N1 , N2 be two lattices with 1 a fan in (N1 )R and 2 a fan in
(N2 )R . A Z-linear mapping : N1 N2 is compatible with the fans 1 and 2 if
for every cone 1 1 , there exists a cone 2 2 such that R (1 ) 2 .
Example 3.3.2. Let N1 = Z2 with basis e1 , e2 and let r be the fan from Figure 4
in 3.1. By Example 3.1.16, Xr is the Hirzebruch surface Hr . Also let N2 = Z
and consider the fan giving P1 :
s
as in Example 3.1.11. The mapping

: N1 N2 ,
ae1 + be2 7 a
is compatible with the fans r and since each cone of r maps onto a cone of .
If r 6= 0, on the other hand, the mapping
: N1 N2 ,
ae1 + be2 7 b
is not compatible with these fans since 3 r does not map into a cone of .
The Definition of Toric Morphism. In 1.3, we defined a toric morphism in the

affine case and gave an equivalent condition in Proposition 1.3.14. For general
toric varieties, it more convenient to take the result of Proposition 1.3.14 as the
definition of toric morphism.
Definition 3.3.3. Let X1 , X2 be normal toric varieties, with 1 a fan in (N1 )R
and 2 a fan in (N2 )R . A morphism : X1 X2 is toric if maps the torus
TN1 X1 into TN2 X2 and |TN is a group homomorphism.
1
126
The proof of part (b) of Proposition 1.3.14 generalizes easily to show that any
toric morphism : X1 X2 is an equivariant mapping for the TN1 - and TN2 actions. That is, we have a commutative diagram
TN1 X1
(3.3.1)
/ X
1
|T
N1
TN2 X2
/ X
2
where the horizontal maps give the torus actions.

Our first result shows that toric morphisms : X1 X2 correspond to Zlinear mappings : N1 N2 that are compatible with the fans 1 and 2 .
Theorem 3.3.4. Let N1 , N2 be lattices and let i be a fan in (Ni )R , i = 1, 2.
(a) If : N1 N2 is a Z-linear map that is compatible with 1 and 2 , then there
is a toric morphism : X1 X2 such that |TN is the map
1
1 : N1 Z C N2 Z C .
(b) Conversely, if : X1 X2 is a toric morphism, then induces a Z-linear

map : N1 N2 that is compatible with the fans 1 and 2 .
Proof. To prove part (a), let 1 be a cone in 1 . Since is compatible with 1
and 2 , there is a cone 2 2 with R (1 ) 2 . Then Proposition 1.3.15 shows
that induces a toric morphism 1 : U1 U2 . Using the general criterion for
gluing morphisms from Exercise 3.3.1, you will show in Exercise 3.3.2 that the
1 glue together to give a morphism : X1 X2 . Moreover, is toric because taking 1 = {0} gives {0} : TN1 TN2 , which is easily seen to be the group
homomorphism induced by the Z-linear map : N1 N2 .
For part (b), we show first that the toric morphism induces a Z-linear map
: N1 N2 . This follows since |TN is a group homomorphism. Hence, given
1
u N1 , the one-parameter subgroup u : C TN1 can be composed with |TN to
1
give the one-parameter subgroup |TN u : C TN2 . This defines an element
1
(u) N2 . It is straightforward to show that : N1 N2 is Z-linear.
It remains to show that is compatible with 1 and 2 . Take 1 1 . By

the Orbit-Cone Correspondence (Theorem 3.2.6), this gives the TN1 -orbit O1 =
O(1 ) X1 . Because of the equivariance (3.3.1), there is a TN2 -orbit O2 X2
with (O1 ) O2 . Using Theorem 3.2.6 again, we have O2 = O(2 ) for some cone
2 2 . Thus (O(1 )) O(2 ). Furthermore, if 1 1 is a face, then by the
same reasoning, there is some cone 2 2 such that (O(1 )) O(2 ).
We claim that in this situation 2 must be a face of 2 . This follows since

O(1 ) O(
part (d) of Theorem 3.2.6. Since is continuous in the Zariski
1 ) by
topology, O(1 ) O(2 ). But the only orbits contained in the closure of O(2 )
127
are the orbits corresponding to cones which have 2 as a face. So 2 is a face of 2 .

It follows from part (c) of Theorem 3.2.6 that also maps the affine open subset
U1 X1 into U2 X2 , i.e.,
(3.3.2)
(U1 ) U2 .
Hence induces a toric morphism U1 U2 , which by Proposition 1.3.15 implies

that R (1 ) 2 . Hence is compatible with the fans 1 and 2 .

First Examples. Here are some examples of toric morphisms defined by mappings
compatible with the corresponding fans.
Example 3.3.5. Let N1 = Z2 and N2 = Z, and let
: N1 N2 ,
ae1 + be2 7 a,
be the first mapping in Example 3.3.2. We saw that is compatible with the fans
r of the Hirzebruch surface Hr and of P1 . Theorem 3.3.4 implies that there is
a corresponding toric morphism : Hr P1 . We will see later in the section that
this mapping gives Hr the structure of a P1 -bundle over P1 .
Example 3.3.6. Let N = Zn and be a fan in NR . For Z>0 , the multiplication

map
: N N,
a 7 a
is compatible with . By Theorem 3.3.4, there is a corresponding toric morphism

: X X whose restriction to TN X is the group endomorphism
|TN (t1 , . . . ,tn ) = (t1 , . . . ,tn ).
For a concrete example, let be the fan in NR = R2 from Figure 2 and take = 2.
Then we obtain the morphism 2 : P 2 P 2 defined in homogeneous coordinates
by 2 (x0 , x1 , x2 ) = (x02 , x12 , x22 ). We will use in Chapter 9.
Sublattices of Finite Index. We get an interesting toric morphism when a lattice

N has finite index in a larger lattice N. If is a fan in NR , then we can view as
a fan either in NR or in NR , and the inclusion N N is compatible with the fan
in NR and NR . As in Chapter 1, we obtain toric varieties X,N and X,N depending
on which lattice we consider, and the inclusion N N induces a toric morphism
: X,N X,N .
Proposition 3.3.7. Let N be a sublattice of finite index in N and let be a fan in
NR = NR . Let G = N/N . Then
: X,N X,N
induced by the inclusion N N presents X,N as the quotient X,N /G.
128
Proof. Since N has finite index in N, Proposition 1.3.18 shows that the finite group
G = N/N is the kernel of TN TN . It follows that G acts on X,N . This action is compatible with the inclusion U,N X,N for each cone . Using
Proposition 1.3.18 again, we see that U,N /G U,N , which easily implies that

X,N /G X,N .
We will revisit Proposition 3.3.7 in Chapter 5, where we will show that the
map : X,N X,N is a geometric quotient.
Example 3.3.8. Let N = Z2 , and be the fan shown in Figure 5, so X,N gives
the weighted projective space P(1, 1, 2). Let N be the sublattice of N given by
N = {(a, b) N | b 0 mod 2}, so N has index 2 in N. Note that N is generated
by u1 = e1 , u2 = 2e2 and that
u0 = e1 2e2 = u1 u2 N .
Let : N N be the inclusion map. It is not difficult to see that with respect to
the lattice N , X,N P 2 (Exercise 3.3.3). By Theorem 3.3.4, the Z-linear map
induces a toric morphism : P 2 P(1, 1, 2), and by Proposition 3.3.7, it follows
that P(1, 1, 2) P 2 /G for G = N/N Z/2Z.
Figure 8. The semigroups 2 M and 2 M
The cone 2 from Figure 5 has the dual cone 2 shown in Figure 8. It is
instructive to consider how 2 interacts with the lattice M dual to N . One checks
that M {(a, b/2) : a, b Z} and 2 = Cone(2e1 e2 , e2 ). In Figure 8, the
points in 2 M are shown in white, and the points in 2 M not in 2 M
are shown in black. Note that the picture in 2 M is the same (up to a change
of coordinates in GL(2, Z)) as Figure 10 from Chapter 1. This shows again that
P(1, 1, 2) contains the affine open subset U2 ,N isomorphic to the rational normal
cone Cb2 . On the other hand U2 ,N C2 is smooth. The other affine open subsets
corresponding to 1 and 0 are isomorphic to C2 in both P 2 and in P(1, 1, 2).
129
Torus Factors. A toric variety X has a torus factor if it is equivariantly isomorphic to the product of a nontrivial torus and a toric variety of smaller dimension.
Proposition 3.3.9. Let X be the toric variety of the fan . Then the following are
equivalent:
(a) X has a torus factor.
(b) There is a nonconstant morphism X C .
(c) The u , (1), do not span NR .
Recall that (1) consists of the 1-dimensional cones of , i.e., its rays, and
that u is the minimal generator of a ray (1).
Proof. If X X (C )r for r > 0 and some toric variety X , then a nontrivial
character of (C )r gives a nonconstant morphism X (C )r C .
If : X C is a nonconstant morphism, then Exercise 3.3.4 implies that

|TN = c m where c C and m M \ {0}. Multiplying by c1 , we may assume
that |TN = m . Then is a toric morphism coming from a nonzero homomorphism
: N Z. Since C comes from the trivial fan, R maps all cones of to the
origin. Hence u ker(R ) for all (1), so the u do not span NR .
Finally, suppose that the u , (1) span a proper subspace of NR . Then N =

Span(u | (1)) N is proper sublattice of N such that N/N is torsion-free, so
N has a complement N with N = N N . Furthermore, can be regarded as a
fan in NR , and then is the product fan = , where is the trivial
fan in NR . Then Proposition 3.1.14 gives an isomorphism
X X ,N TN X ,N (C )nk ,
where dim NR = n and dim NR = k.
In later chapters, toric varieties without torus factors will play an important
role. Hence we state the following corollary of Proposition 3.3.9.
Corollary 3.3.10. Let X be the toric variety of the fan . Then the following are
equivalent:
(a) X has no torus factors.
(b) Every morphism X C is constant, i.e., (X , OX ) = C .
(c) The u , (1), span NR .
We can also think about torus factors from the point of view of sublattices.
Proposition 3.3.11. Let N N be a sublattice with dim NR = n, dim NR = k. Let
be a fan in NR , which we can regard as a fan in NR . Then:
(a) If N is spanned by a subset of a basis of N, then we have an isomorphism
: X,N X,N TN/N X,N (C )nk .
130
(b) In general, a basis for N can be extended to a basis of a sublattice N N of

finite index. Then X,N is isomorphic to the quotient of
X,N X,N TN /N X,N (C )nk
by the finite abelian group N/N .
Proof. Part (a) follows from the proof of Proposition 3.3.9, and part (b) follows
from part (a) and Proposition 3.3.7.

Refinements of Fans and Blowups. Given a fan in NR , a fan refines if
every cone of is contained in a cone of and | | = ||. Hence every cone of
is a union of cones of . When refines , the identity mapping on N is clearly
compatible with and . This yields a toric morphism : X X .
Example 3.3.12. Consider the fan in N Z2 pictured in Figure 1 from 3.1.
This is a refinement of the fan consisting of Cone(e1 , e2 ) and its faces. The
corresponding toric varieties are X C2 and X W = V(x0 y x1 x) P1 C2 ,
the blowup of C2 at the origin (see Example 3.1.15). The identity map on N induces
a toric morphism : W C2 . This blowdown morphism maps P1 {0} W to
0 C2 and is injective outside of P1 {0} in W .
We can generalize this example and Example 3.1.5 as follows.

Definition 3.3.13. Let be a fan in NR Rn . Let = Cone(u1 , . . . , un ) be a
smooth cone in , so that u1 , . . . , un is a basis for N. Let u0 = u1 + + un and let
() be the set of all cones generated by subsets of {u0 , . . . , un } not containing
{u1 , . . . , un }. Then
() = ( \ {}) ()
is a fan in NR called the star subdivision of along .
Example 3.3.14. Let = Cone(u1 , u2 , u3 ) NR R3 be a smooth cone. Figure 9

on the next page shows the star subdivision of into three cones
Cone(u0 , u1 , u2 ), Cone(u0 , u1 , u3 ), Cone(u0 , u2 , u3 ).
The fan
()
consists of these cones, together with their faces.
Proposition 3.3.15. () is a refinement of , and the induced toric morphism

: X () X
makes X () the blowup of X at the distinguished point corresponding to the
cone .
Proof. Since and () are the same outside the cone , without loss of generality, we may reduce to the case that is the fan consisting of and all of its faces,
and X is the affine toric variety U Cn .
131
u0
u2
u1
u3
Figure 9. The star subdivision () in Example 3.3.14
Under the Orbit-Cone Correspondence (Theorem 3.2.6), corresponds to the

distinguished point , the origin (the unique fixed point of the torus action). By
Theorem 3.3.4, the identity map on N induces a toric morphism
: X () U Cn .
It is easy to check that the affine open sets covering X () are the same as for the
blowup of Cn at the origin from Exercise 3.0.8, and they are glued together in the
same way by Exercise 3.1.5.

The blowup X at is sometimes denoted Bl (X ). In this notation, the
blowup of Cn at the origin is written Bl0 (Cn ).
The point blown up in Proposition 3.3.15 is a fixed point of the torus action.
In some cases, torus-invariant subvarieties of larger dimension have equally nice
blowups. We begin with the affine case. The standard basis e1 , . . . , en of Zn gives
= Cone(e1 , . . . , en ) with U = Cn , and the face = Cone(e1 , . . . , er ), 2 r n,
gives the orbit closure
V ( ) = O( ) = {0} Cnr .
To construct the blowup of V ( ), let u0 = u1 + + ur and consider the fan
(3.3.3)
( ) = {Cone(A) | A {u0 , . . . , un }, {u1 , . . . , ur } 6 A}.
Example 3.3.16. Let = Cone(e1 , e2 , e3 ) NR R3 and = Cone(e1 , e2 ). The

star subdivision relative to subdivides into the cones
Cone(e0 , e1 , e3 ), Cone(e0 , e2 , e3 ),
as shown in Figure 10 on the next page. The fan ( ) consists of these two cones,
together with their faces.
132
e0
e2
e1
e3
Figure 10. The star subdivision ( ) in Example 3.3.16
For the fan (3.3.3), the toric variety X ( ) is the blowup of {0} Cnr Cn .
To see why, observe that ( ) is a product fan. Namely, Zn = Zr Znr , and
( ) = 1 2 ,
where 1 is the fan for Bl0 (Cr ) (coming from a refinement of Cone(u1 , . . . , ur ))
and 2 is the fan for Cnr (coming from Cone(ur+1 , . . . , un )). It follows that
X ( ) = Bl0 (Cr ) Cnr .
Since Bl0 (Cr ) is built by replacing 0 Cr with P r1 , it follows that X ( ) =
Bl0 (Cr ) Cnr is built by replacing {0} Cnr Cn with P r1 Cnr . The
intuitive idea is that Bl0 (Cr ) separates directions through the origin in Cr , while
the blowup Bl{0}Cnr (Cn ) = X ( ) separates normal directions to {0} Cnr in
Cn . One can also study Bl{0}Cnr (Cn ) by working on affine pieces given by the
maximal cones of ( )see [218, Prop. 1.26].
We generalize (3.3.3) as follows.
Definition 3.3.17. Let be a fan in NR Rn and assume
P has the property
that all cones of containing are smooth. Let u = (1) u and for each cone
containing , set
( ) = {Cone(A) | A {u } (1), (1) 6 A}.
Then the star subdivision of relative to is the fan

[
( ) = { | 6 }
( ).

The fan ( ) is a refinement of and hence induces a toric morphism

: X ( ) X .
133
Under the map , X ( ) becomes the blowup BlV ( ) (X ) of X along the orbit
closure V ( ).
In Chapters 10 and 11 we will use toric morphisms coming from a generalized
version of star subdivision to resolve the singularities of toric varieties.
Exact Sequences and Fibrations. Next, we consider a class of toric morphisms
that have a nice local structure. To begin, consider a surjective Z-linear mapping
: N N .
If in NR and in NR are compatible with , then we have a corresponding toric
morphism
: X X .
Now let N0 = ker(), so that we have an exact sequence
(3.3.4)
N 0.
0 N0 N
It is easy to check that

0 = { | (N0 )R }
is a subfan of whose cones lie in (N0 )R NR . By Proposition 3.3.11,

(3.3.5)
X0 ,N X0 ,N0 TN
since N/N0 N . Furthermore, is compatible with 0 in NR and the trivial fan

{0} in NR . This gives the toric morphism
|X
0 ,N
: X0 ,N TN .
In fact, by the reasoning to prove Proposition 3.3.4,

(3.3.6)
1 (TN ) = X0 ,N X0 ,N0 TN .
In other words, the part of X lying over TN X is identified with the product
of TN and the toric variety X0 ,N0 . We say this subset of X is a fiber bundle over
TN with fiber X0 ,N0 .
When the fan has a suitable structure relative to , we can make a similar
statement for every torus-invariant affine open subset of X .
Definition 3.3.18. In the situation of (3.3.4), we say is split by and 0 if there
b such that:
exists a subfan
b bijectively to a cone such that (b
(a) R maps each cone
b
N) =
b
N . Furthermore, the map
b 7 defines a bijection .
b and 0 0 , the sum
(b) Given cones
b
b + 0 lies in , and every cone of
arises this way.
134
Theorem 3.3.19. If is split by and 0 as in Definition 3.3.18, then X is a

locally trival fiber bundle over X with fiber X0 ,N0 , i.e., X has a cover by affine
open subsets U satisfying
1 (U ) X0 ,N0 U .
In particular, all fibers of X X are isomorphic to X0 ,N0 .

Proof. Fix in and let ( ) = { | () }. Then
1 (U ) = X( ) .
It remains to show that X( ) X0 ,N0 U . Since ( ) is split by 0 ( )

b ( ), we may assume X = U . In other words, we are reduced to the
and
case when consists of and its proper faces.
A Z-linear map : N N splits the exact sequence (3.3.4) provided is

the identity on N . A splitting induces an isomorphism
N0 N N.
b such that (b
N) = N and R
By Definition 3.3.18, there is a cone
b
maps
b bijectively to . Using
b, one can find a splitting with the property that
b (Exercise 3.3.5). Using Definition 3.3.18 again, we
R maps to b for all b
see that
(N0 )R NR NR
carries the product fan (0 , (N0 )R ) ( , NR ) to the fan (, NR ). By Proposition 3.1.14, we conclude that
X X0 ,N0 X X0 ,N0 U ,
and the theorem is proved.
Example 3.3.20. To complete the discussion begun in Examples 3.3.2 and 3.3.5,
consider the toric morphism : Hr P1 induced by the mapping
: Z2 Z,
ae1 + be2 7 a.
The fan r of Hr is split by the fan of P1 and 0 = { r | R () = {0}}

b of r consisting of the cones
because of the subfan
Cone(e1 + re2 ), {0}, Cone(e1 ).
These cones are mapped bijectively to the cones in under R . Note also that 0
consists of the cones
Cone(e2 ), {0}, Cone(e2 ).
b
The fans and 0 are shown in Figure 11 on the next page.
b and 0 0 , the sums

As we vary over all
b
b + 0 give all cones of Hr .
Hence Theorem 3.3.19 shows that Hr is a locally trivial fibration over P1 , with
fibers isomorphic to
X0 ,N0 P1 ,
135
(1, r)
Figure 11. The Splitting of the Fan r
where N0 = ker() gives the vertical axis in Figure 11. This fibration is not globally
trivial when r > 0, i.e., it is not true that Hr P1 P1 . There is some twisting
on the fibers involved when we try to glue together the 1 (U ) U P1 to
obtain Hr .
We will give another, more precise, description of these fiber bundles and the
twisting mentioned above using the language of sheaves in Chapter 7.
The definition of splitting fan in Definition 3.3.18 requires that (b
N) =
b
N when
b maps to . Exercise 3.3.6 will give an example of how
Theorem 3.3.19 can fail if this condition is not met, and Exercise 3.3.7 will explore
how to modify the theorem when this happens.
Images of Distinguished Points. Each orbit O() in a toric variety X contains

a distinguished point , and each orbit closure V () is a toric variety in its own
right. These structures are compatible with toric morphisms as follows.
Lemma 3.3.21. Let : X X be the toric morphism coming from a map :
N N that is compatible with and . Given , let be the minimal
cone of containing R (). Then:
(a) ( ) = , where O() and O( ) are the distinguished points.

(b) (O()) O( ) and V () V ( ).
(c) The induced map |V () : V () V ( ) is a toric morphism.
136
Proof. First observe that if 1 , 2 contain R (), then so does their intersection. Hence has a minimal cone containing R ().
To prove part (a), pick u Relint() and observe that (u) Relint( ) by the
minimality of . Then

( ) = limt0 u (t) = limt0 (u (t)) = limt0 (u) (t) = ,
where the first and last equalities use Proposition 3.2.2.
The first assertion of part (b) follows immediately from part (a) by the equivariance, and the second assertion follows by continuity (as usual, we get the same
closure in the classical and Zariski topologies).
For (c), observe that |O() : O() O( ) is a morphism that is also a group
homomorphismthis follows easily from equivariance. Since the orbit closures
are toric varieties by Proposition 3.2.7, the map |V () : V () V ( ) is a toric
morphism according to Definition 3.3.3.

Exercises for 3.3.
3.3.1. Let X be a variety with an affine open cover {Ui }, and let Y be a second variety. Let
i : Ui Y be a collection of morphisms. We say that a morphism : X Y is obtained
by gluing the i if |Ui = i for all i. Show that there exists such a : X Y if and only if
for every pair i, j,
i |Ui U j = j |Ui U j .
3.3.2. Let N1 , N2 be lattices, and let 1 in (N1 )R , 2 in (N2 )R be fans. Let : N1 N2 be
a Z-linear mapping that is compatible with the corresponding fans. Using Exercise 3.3.1
above, show that the toric morphisms 1 : U1 U2 constructed in the proof of Theorem 3.3.4 glue together to form a morphism : X1 X2 .
3.3.3. This exercise asks you to verify some of the claims made in Example 3.3.8.
(a) Verify that X,N P 2 with respect to the lattice N .
(b) Verify carefully that the affine open subset U2 ,N Cb2 , where Cb2 is the rational normal
cone Cbd with d = 2.
3.3.4. A character m , m M, gives a morphism TN C . Here you will determine all

morphisms TN C .
(a) Explain why morphisms TN C correspond to invertible elements in the coordinate

ring of TN .
(b) Let c C and Zn . Prove that ct is invertible in C[t11 , . . . ,tn1 ] and that all
invertible elements of C[t11 , . . . ,tn1 ] are of this form.
(c) Use part (a) to show that all morphisms TN C on TN are of the form c m for c C
and m M.
3.3.5. Let : N N be a surjective Z-linear mapping and let

b and be cones in NR
b bijectively onto . Prove that has

and NR respectively with the property that R maps
a splitting : N N such that maps to

b.
137
3.3.6. Let be the fan in R2 with minimal generators u1 = (0, 1), u2 = (2, 1) and u3 =
(0, 1) and maximal cones Cone(u1 , u2 ) and Cone(u2 , u3 ). Let : Z2 Z be projection onto
the first factor and let 0 be the subfan of defined in the discussion following (3.3.4).
Also let be the fan in R with maximal cone = R0 .
(a) Prove that is not split by and 0 . Hint: Show that X0 ,N0 and X are smooth and
then show that Theorem 3.3.19 must fail.
b be the subfan of with maximal cone
(b) Let
b = Cone(u2 ). Show that this satisfies all
parts of Definition 3.3.18 except for the requirement that (b
Z2 ) = Z. Draw a
picture similar to Figure 11 in Example 3.3.20.
b
3.3.7. In the situation of Definition 3.3.18, we say that is weakly split by and 0 if
satisfies Definition 3.3.18 except that we no longer require (b

N) = N .
(a) Explain why Exercise 3.3.6 is an example of a weak splitting that is not a splitting.
(b) In the situation of a weak splitting, prove that all fibers of : X X are isomorphic
to X0 ,N0 . Hint: First assume X = U . Prove that there is a sublattice of N N
of finite index such that splits when we use the lattices N and N . Then show that
there is a commutative diagram
X0 ,N0 U ,N
/ X

U ,N

/ U .
such that X0 ,N0 U ,N is the fiber product X U U ,N as defined in (3.0.5).
Thus, while Theorem 3.3.19 may fail for a weak splitting, at least part remains true.
3.3.8. Let be the fan obtained from the fan for P 2 in Example 3.1.9 by the following
process. Subdivide the cone 2 into two new cones 21 and 22 by inserting an edge
Cone(e2 ).
(a) Show that the resulting toric variety X is smooth.
(b) Show that X is the blowup of P 2 at the point V (2 ).
(c) Show that X is isomorphic to the Hirzebruch surface H1 .
3.3.9. Let X be the toric variety obtained from P 2 by blowing up the points V (1 ) and
V (2 ) (see Figure 2 in Example 3.1.9). Show that X is isomorphic to the blowup of
P1 P1 at the point V (11 ) (see Figure 3 in Example 3.1.12).
3.3.10. Let be the fan obtained from the fan for P(1, 1, 2) in Example 3.1.17 by the
following process. Subdivide the cone 2 into two new cones 21 and 22 by inserting an
edge Cone(u2 ).
(a) Show that the resulting toric variety X is smooth.
(b) Construct a morphism : X X and determine the fiber over the unique singular
point of X .
(c) One of our smooth examples is isomorphic to X . Which one is it?
3.3.11. Consider the action of the group G = {(, 3 ) | 5 = 1} (C )2 on C2 . We will
study the quotient C2 /G and its resolution of singularities using toric morphisms.
138
(a) Let N = Z2 and N = {(a/5, b/5) | a, b Z, b 3a mod 5}. Also let 5 = e2i/5 . Prove
that the map N (C )2 defined by (a/5, b/5) 7 (5a , 5b ) induces an exact sequence
0 N N G 0.
(b) Let = Cone(e1 , e2 ) NR = NR = R2 . The inclusion N N induces a toric morphism U,N U,N . Prove that this is the quotient map C2 C2 /G for the above
action of G on C2 .
(c) Find the Hilbert basis (i.e., the set of irreducible elements) of the semigroup N.
Hint: The Hilbert basis has four elements.
(d) Use the Hilbert basis from part (c) to subdivide . This gives a fan with || = .
Prove that is smooth relative to N and that the resulting toric morphism
X,N U,N = C2 /G
is a resolution of singularities. See Chapter 10 for more details.
(e) The group G gives the finite set G (C )2 C2 with ideal I(G) = hx 5 1, y x 3 i.
Read about the Grobner fan in [70, Ch. 8, 4] and compute the Grobner fan of I(G).
The answer will be identical to the fan described in part (d). This is no accident, as
shown in the paper [156] (see also 10.3). There is a lot of interesting mathematics
going on here, including the McKay correspondence and the G-Hilbert scheme. See
also [206] for the higher dimensional case.
3.3.12. Consider the fan in R3 shown in Figure 12. This fan has five 1-dimensional
y
x
Figure 12. The fan for Exercise 3.3.12
cones with four upward ray generators (1, 0, 1), (0, 1, 1) and one downward generator (0, 0, 1). There are also nine 2-dimensional cones. Figure 12 shows five of the
3.4. Complete and Proper
139
2-dimensional cones; the remaining four are generated by the combining the downward
generator with the four upward generators.
(a) Show that projection onto the y-axis induces a toric morphism X P1 .
(b) Show that X P1 is a locally trivial fiber bundle over P1 with fiber P(1, 1, 2). Hint:
Theorem 3.3.19 and (1, 0, 1) + (1, 0, 1) + 2(0, 0, 1) = 0. See Example 3.1.17.
(c) Explain how you can see the splitting (in the sense of Definition 3.3.18) in Figure 12.
Also explain why the figure makes it clear that the fiber is P(1, 1, 2).
3.3.13. Consider the fan in R2 with ray generators

u0 = e1 + e2 , u1 = e1 , u2 = e2 , u3 = e1
and 1-dimensional cones Cone(u0 , u1 ), Cone(u0 , u2 ), Cone(u2 , u3 ).

(a) Draw a picture of and prove that X is the blowup of P1 C at one point.
(b) Show that the map ae1 + be2 7 b induces a toric morphism : X C such that
1 () P1 for C and 1 (0) is a union of two copies of P1 meeting at a point.
Hint: Once you understand 1 (0), show that the fan for X \ 1 (0) gives P1 C .
(c) To get a better picture of X , consider the map : (C )2 P 3 C defined by
(s,t) = ((s 3 , s 2 , st,t 2 ),t).
Let X = ((C )2 ) P 3 C be the closure of the image. Prove that X X and

that the restriction of the projection P 3 C C to X gives the toric morphism of
part (b).
(d) Let x, y, z, w be coordinates on P 3 . Prove that X P 3 C is defined by the equations
yw z 2 = 0, xz ty 2 = 0, xw tyz = 0.
Also use these equations to describe the fibers 1 () for C, and explain how this
relates to part (b). Hint: The twisted cubic is relevant.
This is a semi-stable degeneration of toric varieties. See [148] for more details.

The Compactness Criterion. We begin by proving part (c) of Theorem 3.1.19.
Theorem 3.4.1. Let X be a toric variety. Then the following are equivalent:
(a) X is compact in the classical topology.
(b) The limit limt0 u (t) exists in X for all u N.
S
(c) is complete, i.e., || = = NR .
Proof. First observe that since X is separated (Theorem 3.1.5), it is Hausdorff in

the classical topology (Theorem 3.0.17). In fact, since the classical topology on
each affine open set U is a metric topology, X is compact if and only if every
sequence of points in X has a convergent subsequence.
For (a) (b), assume that X is compact and fix u N. Given a sequence
tk C converging to 0, we get the sequence u (tk ) X . By compactness, this sequence has a convergent subsequence. Passing to this subsequence, we can assume
140
that limk u (tk ) = X . Because X is the union of the affine open subsets
U for , we may assume U . Now take m M. The character m is
a regular function on U and hence is continuous in the classical topology. Thus
hm,ui
m () = lim m (u (tk )) = lim tk

k
Since tk 0, the exponent must be nonnegative, i.e., hm, ui 0 for all m M.

This implies hm, ui 0 for all m , so that u ( ) = . Then Proposition 3.2.2 implies that limt0 u (t) exists in U and hence in X .
To prove (b) (c), take u N and consider the limit limt0 u (t). This lies
in some affine open U , which implies u N by Proposition 3.2.2. Thus every
lattice point of NR is contained in a cone of . It follows that is complete.
We will prove (c) (a) by induction on n = dim NR . In the case n = 1, the
only complete fan is the fan in R pictured in Example 3.1.11. The corresponding
toric variety is X = P1 . This is homeomorphic to S2 , the 2-dimensional sphere,
and hence is compact.
Now assume the statement is true for all complete fans of dimension strictly
less than n, and consider a complete fan in NR Rn . Let k X be a sequence.
We will show that k has a convergent subsequence.
Since X is the union of finitely many orbits O( ), we may assume the sequence k lies entirely within an orbit O( ). If 6= {0}, then the closure of O( ) in
X is the toric variety V ( ) = XStar( ) of dimension n 1 by Proposition 3.2.7.
Since is complete, it is easy to check that the fan Star( ) is complete in N( )R
(Exercise 3.4.1). Then the induction hypothesis implies that there is a convergent
subsequence in V ( ). Hence, without loss of generality again, we may assume that
our sequence lies entirely in the torus TN X .
Recall from the discussion following Lemma 3.2.5 that
TN HomZ (M, C ).
Moreover, when we regard TN as a group homomorphism : M C , then for
any , restriction yields a semigroup homomorphism M C and hence
a point in U .
A key ingredient of the proof will be the logarithm map L : TN NR defined
as follows. Given a point : M C of TN , consider the map M R defined by
the formula
m 7 log|(m)|.
This is a homomorphism and hence gives an element L() HomZ (M, R) NR .

For more properties of this mapping, see Exercise 3.4.2 below.
For us, the most important property of L is the following. Suppose that a point
TN satisfies L() for some . If m M, then the definition of
141
L implies that
log |(m)| = hm, L()i,
(3.4.1)
which is 0 since m and L() . Hence |(m)| 1. Thus we have

proved that
(3.4.2)
L() = |(m)| 1 for all m M.
Now apply L to our sequence, which gives a sequence L(k ) NR . Since is

complete, the same is true for the fan consisting of the cones for . Hence,
by passing to a subsequence, we may assume that there is such that
L(k )
for all k. By (3.4.2), we conclude that |k (m)| 1 for all m M. It follows

that the k are a sequence of mappings to the closed unit disk in C. Since the closed
unit disk is compact, there is a subsequence k which converges to a point U .
You will check the details of this final assertion in Exercise 3.4.3.

Complete Varieties. The compactness criterion proved in Theorem 3.4.1 uses the
classical topology. It is natural to ask for an algebraic version of this theorem that
uses only the Zariski topology. The crucial idea is the notion of completeness.
To motivate the definition of completeness, we first reformate the topological
notion of compactness. You will prove the following in Exercise 3.4.4.
Proposition 3.4.2. Let X be a locally compact Hausdorff topological space. Then
the following are equivalent:
(a) X is compact.
(b) For every topological space Z, the projection map Z : X Z Z is closed,
i.e., Z (W ) Z is closed for all closed subsets W X Z.

We now define the algebraic analog of compactness.
Definition 3.4.3. A variety X is complete if for every variety Z, the projection map
Z : X Z Z is a closed mapping in the Zariski topology.
Here are two examples to illustrate this definition.
Example 3.4.4. Consider the affine variety C. We claim that C is not complete.
To see this, consider the projection map 2 : C C = C2 C. The closed subset
V(xy 1) C2 does not map to a Zariski-closed subset of C under 2 . Hence 2
is not a closed mapping, so that C is not complete.
Example 3.4.5. The Projective extension theorem [69, Thm. 6 of Ch. 8, 5] shows
that for X = P n , the mapping
Cm : P n Cm Cm
142
is closed in the Zariski topology for all m. It follows that if V Cm is any affine
variety, the projection
V : P n V V
is a closed mapping in the Zariski topology. Then the gluing construction shows
that Z : P n Z Z is closed for any variety Z, so P n is a complete variety. In fact,
one can think of P n as the prototypical complete variety. Moreover, any projective
variety is complete (Exercise 3.4.5). However, there are complete varieties that are
not projectivewe will give a toric example in Chapter 6.
Completeness is the algebraic version of compactness, and it can be shown

that a variety is complete if and only if it is compact in the classical topology. This
is proved in Serres famous paper Geometrie algebrique et geometrie analytique,
called GAGA for short. See [248, Prop. 6, p. 12]. As a consequence, we get the
following improved version of Theorem 3.4.1.
Theorem 3.4.6. Let X be a toric variety. Then the following are equivalent:
(a) X is compact in the classical topology.
(b) X is complete.
(c) The limit limt0 u (t) exists in X for all u N.
S
(d) is complete, i.e., || = = NR .
Proper Mappings. In algebraic geometry, many concepts that apply to varieties

have relative versions that apply to morphisms. To see how this works for complete
varieties, we will begin in the topological category with the relative version of
compactness.
Definition 3.4.7. A continuous mapping f : X Y is proper if the inverse image
f 1 (T ) is compact in X for every compact subset T Y .
It is immediate that X is compact if and only if the constant mapping from X
to the space Y = {pt} consisting of a single point is proper. This relative version
of compactness may also be reformulated, for reasonably nice topological spaces,
in the following way.
Proposition 3.4.8. Let f : X Y be a continuous mapping of locally compact first
countable Hausdorff spaces. Then the following are equivalent:
(a) f is proper.
(b) f is a closed mapping, i.e., f (W ) Y is closed for all closed subsets W X ,
and all fibers f 1 (y), y Y , are compact.
(c) Exery sequence xk X such that f (xk ) Y converges in Y has a subsequence

xk that converges in X .
143
Proof. A proof of (a) (b) can be found in [122, Ch. 9, 4]. See Exercise 3.4.6
for (a) (c).

Before we can give a definition of properness that works for morphisms, we
first need to reformulate the topological notion of properness. Recall from 3.0
that morphisms f : X S and g : Y S give the fiber product X S Y . Fiber
products can also be defined for continuous maps between topological spaces. You
will prove in Exercise 3.4.6 that properness can be described using fiber products
as follows.
Proposition 3.4.9. Let f : X Y be a continuous map between locally compact
Hausdorff spaces. Then f is proper if and only if f is universally closed, meaning
that for all spaces Z and all continuous mappings g : Z Y , the projection Z
defined by the commutative diagram
X Y Z
/X
/Y
is a closed mapping.
In algebraic geometry, we will use the following definition of properness for

morphisms between varieties.
Definition 3.4.10. A morphism of varieties : X Y is proper if it is universally
closed, in the sense that for all varieties Z and morphisms : Z Y , the projection
Z defined by the commutative diagram
X Y Z
/X
/Y
is a closed mapping in the Zariski topology.

It is easy to see that a variety X is complete if and only if the constant morphism
: X {pt} is proper. Furthermore, if X is complete, then the projection map
Z : X Z Z
is proper for any variety Z. You will prove these assertions in Exercise 3.4.7.
The Properness Criterion. Theorem 3.4.6 can be understood as a special case of
the following statement for toric morphisms.
144
Theorem 3.4.11. Let : X X be the toric morphism corresponding to a

homomorphism : N N that is compatible with fans in NR and in NR .
Then the following are equivalent:
(a) : X X is proper in the classical topology (Definition 3.4.7).
(b) : X X is a proper morphism (Definition 3.4.10).
(c) If u N and limt0 (u) (t) exists in X , then limt0 u (t) exists in X .
1
(d) R (| |) = ||.
Proof. The proof of (a) (b) uses two fundamental results in algebraic geometry.
First, given any morphism of varieties f : X Y and a Zariski closed subset

W X , a theorem of Chevalley tells us that the image f (W
S ) Y is constructible,
meaning that it can be written as a finite union f (W ) = i (Vi \Wi ), where Vi and
Wi are Zariski closed in Y . A proof appears in [131, Ex. II.3.19].
Second, given any constructible subset C of a variety Y , its closure in Y in the
classical topology equals its closure in the Zariski topology. When C is open in the
Zariski topology, a proof is given in [207, Thm. (2.33)], and when C is the image
of a morphism, a proof can be found in GAGA [248, Prop. 7, p. 12].
Now suppose that : X X is proper in the classical topology and let

: Z X be a morphism. This gives the commutative diagram
X X Z
/ X
/ X .
Let Y X X Z be Zariski closed. We need to prove that Z (Y ) is Zariski closed

in Z. First observe that Y is also closed in the classical topology, so that Z (Y )
is closed in Z in the classical topology by Proposition 3.4.9. However, Z (Y ) is
constructible by Chevalleys theorem, and then, being classically closed, it is also
Zariski closed by GAGA. Hence Z is a closed map in the Zariski topology for any
morphism : Z X . It follows that is a proper morphism.
To prove (b) (c), let u N and assume that = limt0 (u) (t) exists in X .
We first prove limt0 u (t) exists in X under the extra assumption that (u) 6= 0.
This means that (u) is a nontrivial one-parameter subgroup in X .
Let u (C ) X be the closure of u (C ) X in the classical topology.
Our earlier remarks imply that this equals the Zariski closure.
Since is proper,
u
it is closed in the Zariski topology, so that (C ) is closed in X in both
topologies. It follows that

(u) (C ) u (C ) .
145
Hence there is u (C ) mapping to . Thus there is a sequence of points tk C

such that u (tk ) . Then
= () = lim (u (tk )) = lim (u) (tk ).
k
This, together with = limt0 (u) (t) and (u) 6= 0, imply that tk 0. From
here, the arguments used to prove (a) (b) (c) of Theorem 3.4.1 easily imply
that limt0 u (t) exists in X .
For the general case when we no longer assume (u) 6= 0, consider the map
(, 1C ) : X C X C. This is proper since is proper (Exercise 3.4.8).
Furthermore, X C and X C are toric varieties by Proposition 3.1.14, and
the corresponding map on lattices is (, 1Z ) : N Z N Z. Then applying the
above argument to (u, 1) N Z shows that limt0 u (t) exists in X . We leave
the details to the reader (Exercise 3.4.8).
For (c) (d), first observe that the inclusion
1
|| R (| |)
is automatic since is compatible with and . For the opposite inclusion,
1
take u R (| |) N. Then (u) | |, which by Proposition 3.2.2 implies that
limt0 (u) (t) exists in X . By assumption, limt0 u (t) exists in X . Using
Proposition 3.2.2, we conclude that u N for some . Because all the
1
cones are rational, this immediately implies R (| |) ||.
Finally, we prove (d) (a). We begin with two special cases.
Special Case 1. Suppose that a toric morphism : X TN satisfies (d) and

has the additional property that : N N is onto. The fan of TN consists of the
trivial cone {0}, so that (d) implies
(3.4.3)
|| = R (0) = ker(R ).
When we think of as a fan in ker(R ) NR , (3.3.5) implies that

X X TN .
Then corresponds to the projection X TN TN . The fan is complete in
ker(R ) by (3.4.3), so that X is compact by Theorem 3.4.1. Thus X {pt} is
proper, which easily implies that X TN TN is proper. We conclude that is
proper in the classical topology.
Special Case 2. Suppose that a homomorphism of tori : TN TN has the
additional property that : N N is injective. Then (d) is obviously satisfied. An
elementary proof that is proper is given in Exercise 3.4.9.
Now consider a general toric morphism : X X satisfying (d). We will
prove that is proper in the classical topology using part (c) of Proposition 3.4.8.
146
Thus assume that k X is a sequence such that (k ) converges in X . We need

to prove that a subsequence of k converges in X .
Since X has only finitely many TN -orbits, we may assume that the sequence
lies in an orbit O(). As in Lemma 3.3.21, let be the minimal cone of
containing R (). The restriction
|V () : V () V ( )
is a toric morphism by Lemma 3.3.21, and the fans of V () and V ( ) are given by
Star() in N()R and Star( ) in N ( )R respectively. Furthermore, one can check
that since and satisfy (d), the same is true for the fans Star() and Star( )
(Exercise 3.4.10). Hence we may assume that k TN and (k ) TN for all k.
The limit = limk (k ) lies in an orbit O( ) for some . Thus the
sequence (k ) and its limit all lie in U . Note that { | () } is the
fan giving 1 (U ). Since (d) implies that
1
R ( ) =
R ()
1
we can assume that X = U , i.e., : X U and
( ) = ||.
If = {0}, then O( ) = U = TN . If we write as the composition

N (N) N ,
then : X TN factors as X T(N) TN . Special Cases 1 and 2 imply that

these maps are proper, and since the composition of proper maps is proper, we
conclude that is proper.
It remains to consider the case when 6= {0}. When we think of U as a
semigroup homomorphism : ( ) M C, Lemma 3.2.5 tells us that
(m ) = 0 for all m ( ) M \ ( ) M .
Since the (k ) : M C converge to in U , we see that
lim (k )(m ) = 0 for all m ( ) M \ ( ) M .
Since ( ) M is finitely generated, it follows that we may pass to a subsequence

and assume that
(3.4.4)
|(k )(m )| 1 for all k and all m ( ) M \ ( ) M .
147
The logarithm map from the proof of Theorem 3.4.1 gives maps LN : TN NR
and LN : TN NR linked by a commutative diagram:
TN
LN
/ NR
|T
TN
LN
/ N .
R
Let : M M be dual to : N N . Then m ( ) M \ ( ) M implies

that for all k, we have
h (m ), LN (k )i = hm , R (LN (k ))i
(3.4.5)
= hm , LN ((k ))i = log |(k )(m )| 0,
where the first equality is standard, the second follows from the above commutative
diagram, the third follows from (3.4.1), and the final inequality uses (3.4.4).
Now consider the following equivalences:
1
u R ( ) R (u)
hm , R (u)i 0 for all m ( ) M
h (m ), ui 0 for all m ( ) M ,
where the first and third equivalences are obvious and the second uses = ( )
and the rationality of . But we also know that 6= {0}, which means that ( )
is a cone whose maximal subspace ( ) is a proper subset. This implies that
1
u R ( ) h (m ), ui 0 for all m ( ) M \ ( ) M
1
(Exercise 3.4.11). Using (3.4.5), we conclude that LN (k ) R ( ) for all k.

1
But, as noted above, (d) means ( ) = ||. It follows that
LN (k ) ||
for all k. Passing to a subsequence, we may assume that there is such that
LN (k )
for all k. From here, the proof of (c) (a) in Theorem 3.4.1 implies that there is
a subsequence k which converges to a point U X . This proves that is
proper in the classical topology. The proof of the theorem is now complete.

We noted earlier that a variety is complete if and only if it is compact. In a
similar way, a morphism f : X Y of varieties is a proper morphism if and only if
it is proper in the classical topology. This is proved in [126, Prop. 3.2 of Exp. XII].
Thus the equivalence (a) (b) of Theorem 3.4.11 is a special case of this result.
148
Theorems 3.4.6 and 3.4.11 show that properness and completeness can be
tested using one-parameter subgroups. In the case of completeness, we can formulate this as follows. Given u N, the one-parameter subgroup gives a map
u : C \ {0} TN X , and saying that limt0 u (t) exists in X means that u
extends to a morphism u0 : C X . In other words, whenever we have a commutative diagram with solid arrows
u
C \ {0}
i
vv
/ X
v:
u0
(u)
0
/ {pt},
the dashed arrow u0 exists so that the diagram remains commutative. The existence
of u0 tells us that the variety X is not missing any points, which is where the
term complete comes from. In a similar way, the properness criterion given in
part (c) of Theorem 3.4.11 can be formulated as saying that whenever u N gives
a commutative diagram,
u
C \ {0}
i
vv
/ X
v;
u0
(u)
/ X ,
the dashed arrow u0 exists so that the diagram remains commutative.

For general varieties, there are similar criteria for completeness and properness
that replace u : C \ {0} X and u0 : C X with maps coming from discrete
valuation rings, to be discussed in Chapter 4. An example of a discrete valuation
ring is the ring of formal power series R = C[[t]], whose field of fractions is the
field of formal Laurent series K = C((t)). By replacing C with Spec(R) and C \{0}
with Spec(K) in the above diagrams, where R is now an arbitrary discrete valuation
ring, one gets the valuative criterion for properness (see [131, Ex. II.4.11 and Thm.
II.4.7]). This requires the full power of scheme theory since Spec(R) and Spec(K)
are not varieties as defined in this book. Using the valuative criterion of properness,
one can give a direct, purely algebraic proof of (d) (b) in Theorem 3.4.11 and
Corollary 3.4.6 (see [105, Sec. 2.4] or [218, Sec. 1.5]).
Example 3.4.12. An important class of proper morphisms are the toric morphisms
: X X induced by a refinement of . Condition (d) of Theorem 3.4.11 is
obviously fulfilled since : N N is the identity and every cone of is a union
of cones of . In particular, the blowups
: X () X
studied in Proposition 3.3.15 are always proper.
149
Exercises for 3.4.

3.4.1. Let be a complete fan in NR and let be a cone in . Show that the fan Star( )
defined in (3.2.8) is a complete fan in N( )R .
3.4.2. In this exercise, you will develop some additional properties of the logarithm mapping L : TN NR defined in the proof of Theorem 3.4.1.
(a) Let S1 be the unit circle in the complex plane, a subgroup of the multiplicative group
C . Show that there is an isomorphism of groups
: C S1 R
z 7 (|z|, log |z|),
where the operation in the second factor on the right is addition.

(b) Show that the compact real n-dimensional torus (S1 )n can be viewed as a subgroup of
TN and that L : TN NR induces an isomorphism TN /(S1 )n NR . Hint: Use from
part (a).
(c) Let be a fan in N. Show that the action of the compact real torus (S1 )n TN on TN
extends to an action on the toric variety X and that the quotient space
[
(X )/(S1 )n
= N()R ,
where
= denotes homeomorphism of topological spaces, and the union is over all
cones in the fan. Hint: Use the Orbit-Cone Correspondence (Theorem 3.2.6).
(d) Let in R2 be the fan from Example 3.1.9, so that X P 2 . Show that under the
action of (S1 )2 (C )2 as in part (c), P 2 /(S1 )2
= 2 , the 2-dimensional simplex.
We will say more about the topology of toric varieties in Chapter 12.
3.4.3. This exercise will complete the proof of Theorem 3.4.1. Let Hom( M, C) be
the set of semigroup homomorphisms M C. Assume that k Hom( M, C) is
a sequence such that |k (m)| 1 for all m M and all k. We want to show that there
is a subsequence k that converges to a point Hom( M, C).
(a) The semigroup S = M is generated by a finite set {m1 , . . . , ms }. Use this fact
and the compactness of {z C | |z| 1} to show that there exists a subsequence k
such that the sequences k (m j ) converge in C for all j.
(b) Deduce that the subsequence k converges to a Hom( M, C).
3.4.4. Prove Proposition 3.4.2. Hint: For (b) (a), let Z be the one-point compactification
of X and consider the projection of = {(x, x) | x X} X Z.
3.4.5. Show that any projective variety is complete according to Definition 3.4.10.
3.4.6. Here you will prove some characterizations of properness stated in the text.
(a) Prove (a) (c) from Proposition 3.4.8.
(b) Prove Proposition 3.4.9. Hint: First show that if f is proper, then so is Z : X Y Z Z
for any morphism Z Y . Then use (a) (b) of Proposition 3.4.8, which does not
require first countable. If f : X Y is universally closed, then prove that f 1 (y) {y}
is universally closed for any y Y . Then use Proposition 3.4.2 and (a) (b) of
Proposition 3.4.8.
150
3.4.7. Prove that X is complete if and only X {pt} is proper, and that if X is complete,
then Z : X Z Z is proper for any variety Z.
3.4.8. Complete the proof of (b) (c) of Theorem 3.4.11 begun in the text.
3.4.9. Let : TN TN be a map of tori corresponding to an injective homomorphism
: N N . Also let : M M be the dual map. Finally, let k TN be a sequence such

that (k ) converges to a point of TN .
(a) Prove that im( ) M has finite index. Hence we can pick an integer d > 0 such that
dM im( ).
(b) Show that m (k ) converges for all m im( ). Conclude that m (kd ) converges for
all m M, where d is as in part (a).
(c) Pick a basis of M so that TN (C )n and write k = (1,k , . . . , n,k ) (C )n . Show
d
d
that (1,k
, . . . , n,k
) converges to a point (
1 , . . . , n ) (C )n .
1/d
(d) Show that the dth roots i can be chosen so that a subsequence of the sequence
1/d
1/d
k = (1,k , . . . , n,k ) converges to a point = (
1 , . . . , n ) TN .
(e) Explain why this implies that TN TN is proper in the classical topology.
3.4.10. To finish the proof of (d) (a) of Theorem 3.4.11, suppose we have a toric morphism : X X and a cone . Let be the smallest cone containing R ().
(a) Prove that induces a homomorphism : N() N( ).
1
(b) Assume further that R (| |) = ||. Prove that ( )1

R (|Star( )|) = |Star()|.
3.4.11. Let 6= {0} be a strongly convex polyhedral cone in NR . Prove that

u hm , u i 0 for all m ( ) M \ ( ) M
and then apply this to u = R (u) to complete the argument in the text. Hint: To prove
, first show that the right hand side of the equivalence implies that hm , u i 0 for all
m ( ) MQ \ ( ) MQ . Then show that 6= {0} implies that any element of
( ) M is a limit of elements in ( ) MQ \ ( ) MQ .
3.4.12. Give a second argument for the implication
X compact complete
from part (c) of Theorem 3.1.19 using induction on the dimension n of N. Hint: If is not
complete and n > 1, then there is a 1-dimensional cone in the boundary of the support of
. Consider the fan Star( ) and the corresponding toric subvariety of X .
3.4.13. Let , be fans in NR compatible with the identity map N N. Prove that the
toric morphism : X X is proper if and only if is a refinement of .

In this appendix, we discuss toric varieties that are not necessarily normal. We begin with
an example to show that Sumihiros theorem (Theorem 3.1.7) on the existence of a torusinvariant affine open cover can fail in the nonnormal case.
151
Example 3.A.1. Consider the nodal cubic C P 2 defined by y 2 z = x 2 (x + z). The only
singularity of C is p = (0, 0, 1). We claim that C is a toric variety with C \ {p} C as
torus. Assuming this for the moment, consider a torus-invariant neighborhood of p. It
contains p and the torus and hence is the whole curve! We conclude that p has no torusinvariant affine open neighborhood. Thus Sumihiros theorem fails for C.
To see that C is a toric variety, we begin with the standard parametrization obtained
by intersecting lines y = t x with the affine curve y 2 = x 2 (x + 1). This easily leads to the
parametrization
x = t 2 1, y = t(t 2 1).
The values t = 1 map to the singular point p. To get a parametrization that looks more
t+1
to obtain
like a torus, we replace t with t1
x=
4t
,
(t 1)2
y=
4t(t + 1)
.
(t 1)3
Then t = 0, map to p and t C maps bijectively to C \ {p}.
Using this parametrization, we get C C, and the action of C on itself given by

multiplication extends to an action on C by making p a fixed point of the action. With
some work, one can show that this action is algebraic and hence gives a toric variety. (For
readers familiar with elliptic curves, the basic idea is that the description of the group law
in terms of lines connecting points on the curve reduces to multiplication in C C for
our curve C.)
In contrast, the projective toric varieties constructed in Chapter 2 satisfy Sumihiros

theorem by Proposition 2.1.8. Since these nonnormal toric varieties have a good local
structure, it is reasonable to expect that they share some of the nice properties of normal toric varieties. In particular, they satisfy a version of the Orbit-Cone Correspondence
(Theorem 3.2.6).
We begin with the affine case. Given M and a finite subset A = {m1 , . . . , ms } M,
we get the affine toric variety YA Cs whose torus has character group ZA (Proposition 1.1.8). Assume M = ZA and let NR be dual to Cone(A ) MR . By Proposition 1.3.8, the normalization of YA is the map
U YA
induced by the inclusion of semigroup algebras
C[NA ] C[ M].
Recall that C[ M] is the integral closure of C[NA ] in its field of fractions. We now
apply standard results in commutative algebra and algebraic geometry:
Since the integral closure C[ M] is a finitely generated C-algebra, it is a finitely
generated module over C[NA ] (see [10, Cor. 5.8]).
Thus the corresponding morphism U YA is finite as defined in [131, p. 84].
A finite morphism is proper with finite fibers (see [131, Ex. II.3.5 and II.4.1]).
Since U YA is the identity on the torus, the image of the normalization is Zariski dense
in YA . But the image is also closed since the normalization map is proper. This proves that
the normalization map is onto.
Here is an example of how the normalization map can fail to be one-to-one.
152
Example 3.A.2. The set A = {e1 , e1 + e2 , 2e2 } Z2 gives the parametrization A (s,t) =
(s, st,t 2 ), and one can check that
YA = V(y 2 x 2 z) C3 .
Furthermore, ZA = Z2 and = Cone(A ) = Cone(e1 , e2 ). It follows easily that the

normalization is given by
C2 YA
(s,t) 7 (s, st,t 2 ).

This map is one-to-one on the torus (the torus of YA is normal and hence is unchanged
under normalization) but not on the t-axis, since here the map is (0,t) 7 (0, 0,t 2 ). We will
soon see the intrinsic reason why this happens.
We now determine the orbit structure of YA .

Theorem 3.A.3. Let YA be an affine toric variety with M = ZA and let NR be as
above. Then:
{faces of } {TN -orbits in YA }
such that a face of of dimension k corresponds to an orbit of dimension dim YA k.
(b) If O YA is the orbit corresponding to a face of , then O is the torus with

character group Z( A ).
(c) The normalization U YA induces a bijection
{TN -orbits in U } {TN -orbits in YA }
such that if O U and O YA are the orbits corresponding to a face of , then the
induced map O O is the map of tori corresponding to the inclusion Z( A )
M of character groups.
Proof. We will sketch the main ideas and leave the details for the reader. The proof uses
the Orbit-Cone Correspondence (Theorem 3.2.6). We regard points of U and YA as semigroup homomorphisms, so that : M C in U maps to |NA : NA C in YA .
Note also that U YA is equivariant with respect to the action of TN .
By Lemma 3.2.5, the orbit O( ) U corresponding to a face of is the torus
consisting of homomorphisms : M C . Thus M is the character group of
O( ). The normalization maps this orbit onto an orbit O ( ) YA , where a point of
O( ) maps to its restriction to NA . Since
( M) ZA = ZA = Z( A ),
it follows that Z( A ) is the character group of O ( ). This proves part (b), and the
final assertion of part (c) follows easily.
Since M is the saturation of NA , it follows that there is an integer d > 0 such
that d M NA . It follows easily that Z( A ) has finite index in M, so that
dim O ( ) = dim O( ) = dim U dim = dim YA dim ,
proving the final assertion of part (a).
153
Finally, every orbit in YA comes from an orbit in U since U YA is onto. If orbits

O(1 ), O(2 ) map to the same orbit of YA , then
Z(1 A ) = Z(2 A ).
This implies 1 = 2 , so that 1 = 2 . The bijections in parts (a) and (c) now follow.
We leave it to the reader to work out other aspects of the Orbit-Cone Correspondence
(specifically, the analogs of parts (c) and (d) of Theorem 3.2.6) for YA .
Let us apply Theorem 3.A.3 to our previous example.
Example 3.A.4. Let A = {e1 , e1 + e2 , 2e2 } Z2 as in Example 3.A.2. The cone =
Cone(A ) = Cone(e1 , e2 ) has a face such that = Span(e2 ). Thus
Z( A ) = Z(2e2 )
M = Ze2 .
It follows that Z( A ) has index 2 in M, which explains why the normalization

map is two-to-one on the orbit corresponding to .
We now turn to the projective case. Here, A = {m1 , . . . , ms } M gives the projective
s1
toric varietyXP
whose torus
group ZA (Proposition 2.1.6). Recall
A P

Ps has character
s
that ZA =
a
m
|
a
Z,
a
=
0
.
i
i=1 i i
i=1 i
One observation is that translating A by m M leaves the corresponding projective

variety unchanged. In other words, Xm+A = XA (see part (a) of Exercise 2.1.6). Thus, by
translating an element of A to the origin, we may assume 0 A . Note that the torus of
XA has character lattice ZA = ZA when 0 A .
We defined the normalization of an affine variety in 1.0. Using a gluing construction,

one can define the normalization of any variety (see [131, Ex. II.3.8]). We can describe the
normalization of a projective toric variety XA as follows.
Theorem 3.A.5. Let XA be a projective toric variety where 0 A and M = ZA . If
P = Conv(A ) MR , then the normalization of XA is the toric variety XP of the normal
fan of P with respect to the lattice N = HomZ (M, Z).
Proof. Again, we sketch the proof and leave the details to the reader. We use the local
description of XA given in Propositions 2.1.8 and 2.1.9. There, we saw that XA has an
affine open covering given by the affine toric varieties YAv = Spec(NAv ), where v A is
a vertex of P = Conv(A ) and Av = A v = {m v | m A }.
For the moment, assume that P is very ample. Then Theorem 2.3.1 implies that XP has
an affine open cover given by the affine toric varieties Uv = Spec(v M), where v A
is a vertex of P and v = Cone(P M v). One can check that v M is the saturation of
NAv , so that Uv is the normalization of YAv . The gluings are also compatible by equations
(2.1.6), (2.1.7) and Proposition 2.3.13. It follows that we get a natural map XP XA that
is the normalization of XA .
In the general case, we note that k0 P is very ample for some integer k0 1 and that
P and k0 P have the same normal fan. Since v is a maximal cone of the normal fan, the
above argument now applies in general, and the theorem is proved.
154
Combining this result with the Orbit-Cone Correspondence and Theorem 3.A.3 gives
the following immediate corollary.
Corollary 3.A.6. With the same hypotheses as Theorem 3.A.5, we have:
{cones of P } {TN -orbits in XA }
such that a cone of dimension k corresponds to an orbit of dimension dim XA k.
(b) If O XA is the orbit corresponding to a cone of P , then O is the torus with

character group Z( A ).
(c) The normalization XP XA induces a bijection
{TN -orbits in XP } {TN -orbits in XA }
such that if O XP and O XA are the orbits corresponding to P , then the

induced map O O is the map of tori corresponding to the inclusion Z( A )
M of character groups.
We leave it to the reader to work out other aspects of the Orbit-Cone Correspondence
for XA . A different approach to the study of XA appears in [113, Ch. 5].
Chapter 4
Divisors on Toric Varieties
4.0. Background: Valuations, Divisors and Sheaves

Divisors are defined in terms of irreducible codimension one subvarieties. In this
chapter, we will consider Weil divisors and Cartier divisors. These classes coincide
on a smooth variety, but for a normal variety, the situation is more complicated. We
will also study divisor classes, which are defined using the order of vanishing of
a rational function on an irreducible divisor. We will see that normal varieties are
the natural setting to develop a theory of divisors and divisor classes.
First, we give a simple motivational example.
Example 4.0.1. If f (x) C(x) is nonzero, then there is a unique n Z such that
g(x)
f (x) = x n h(x)
, where g(x), h(x) C[x] are not divisible by x. This works because
C[x] is a UFD. The integer n describes the behavior of f (x) at 0: if n > 0, f (x)
vanishes to order n at 0, and if n < 0, f (x) has a pole of order |n| at 0. Furthermore,
the map from the multiplicative group C(x) to the additive group Z defined by
f (x) 7 n is easily seen to be a group homomorphism. This works in the same way
if we replace 0 with any point of C.
Discrete Valuation Rings. The simple construction given in Example 4.0.1 applies
in far greater generality. We begin by reviewing the algebraic machinery we will
need.
Definition 4.0.2. A discrete valuation on a field K is a group homomorphism
: K Z
that is onto and satisfies (x+ y) min((x), (y)) when x, y, x+ y K = K \{0}.

Note also that (xy) = (x) + (y). The corresponding discrete valuation ring is
R = {x K | (x) 0} {0}.
155
Chapter 4. Divisors on Toric Varieties
156
One can check that a DVR is indeed a ring. Here are some properties of DVRs.
Proposition 4.0.3. Let R be a DVR with valuation : K Z. Then:
(a) x R is invertible in R if and only if (x) = 0.
(b) R is a local ring with maximal ideal m = {x R | (x) > 0} {0}.
(c) R is normal.
(d) R is a principal ideal domain (PID).

(e) R is Noetherian.
(f) The only proper prime ideals of R are {0} and m.
Proof. First observe that since is a homomorphism, we have
(4.0.1)
(x1 ) = (x)
for all x K . If x R is a unit, then (x), (x1 ) 0 since x, x1 R. Thus

(x) = 0 by (4.0.1). Conversely, if (x) = 0, then (x1 ) = 0 by (4.0.1), so that
x1 R. This proves part (a).
For part (b), note that m = {x R | (x) > 0} {0} is an ideal of R (this
follows directly from Definition 4.0.2). Then part (a) easily implies that R is local
with maximal ideal m (Exercise 4.0.1).
To prove part (c), suppose x K = K \ {0} satisfies
x n + rn1 x n1 + + r0 = 0,
with ri R. If x R, we are done, so suppose x
/ R. Then n > 1 and (x) < 0.
1
1n
Using (4.0.1) again, we see that x R. So x
= (x1 )n1 R and hence
x1n (x n + rn1 x n1 + + r0 ) = 0,
showing that x = (rn1 + rn2 x1 + + r0 x1n ) R.
Let R satisfy () = 1 and let I 6= {0} be an ideal of R. Pick x I \ {0}

with k = (x) minimal. Then y = x k K satisfies (y) = (x) k() = 0, so
that y is invertible in R. From here, one proves without difficulty that I = h k i.
This proves part (d), and part (e) follows immediately.
For part (f), it is obvious that {0} and the maximal ideal m are prime. Note
also that m = hi. Now let P 6= {0} be a proper prime ideal. By the previous
paragraph, P = h k i for some k > 0. If k > 1, then k1 P and , k1
/P
give a contradiction.

This shows that every DVR is a Noetherian local domain of dimension one.
In general, the dimension dim R of a Noetherian ring R is one less than the length
of the longest chain P0 ( ( Pd of proper prime ideals contained in R. Among
Noetherian local domains of dimension one, DVRs are characterized as follows.
157
Theorem 4.0.4. If (R, m) is a Noetherian local domain of dimension one, then the
following are equivalent:
(a) R is a DVR.
(b) R is normal.
(c) m is principal.
(d) (R, m) is a regular local ring.
Proof. The implications (a) (b) and (a) (c) follow from Proposition 4.0.3, and
the equivalence (c) (d) is covered in Exercise 4.0.2. The remaining implications
can be found in [10, Prop. 9.2].

DVRs and Prime Divisors. DVRs have a natural geometric interpretation. Let X
be an irreducible variety. A prime divisor D X is an irreducible subvariety of
codimension one, meaning that dim D = dim X 1. Recall from 3.0 that X has
a field of rational functions C(X ). Our goal is to define a ring OX,D with field
of fractions C(X ) such that OX,D is a DVR when X is normal. This will give
a valuation D : C(X ) Z such that for f C(X ) , D ( f ) gives the order of
vanishing of f along D.
Definition 4.0.5. For a variety X and prime divisor D X , OX,D is the subring of
C(X ) defined by
OX,D = { C(X ) | is defined on U X open with U D 6= }.
We will see below that OX,D is a ring. Intuitively, this ring is built from rational
functions on X that are defined somewhere on D (and hence defined on most of D
since D is irreducible).
Since X is irreducible, Exercise 3.0.4 implies that C(X ) = C(U ) whenever
U X is open and nonempty. If we further assume that U D is nonempty, then
OX,D = OU,UD
(4.0.2)
follows easily (Exercise 4.0.3).
Hence we can reduce to the affine case X = Spec(R) for an integral domain
R. The codimension of a prime ideal p, also called its height, is defined to be
codim p = dim R dim V(p). It follows easily that p 7 V(p) induces a bijection
{codimension one prime ideals of R} {prime divisors of X}.
Given a prime divisor D = V(p), we can interpret OX,D in terms of R as follows.
The field of rational functions C(X ) is the field of fractions K of R, and a rational
function = f /g K, f , g R, is defined somewhere on D = V(p) precisely when
g
/ I(D) = p. It follows that
OX,D = { f /g K | f , g R, g
/ p},
158
which is the localization Rp of R at the multiplicative subset R \ p (note that R \ p

is closed under multiplication because p is prime). This localization is a local ring
with maximal ideal pRp (Exercise 4.0.3). It follows that
OX,D = Rp
(4.0.3)
when X = Spec(R) and p is a codimension one prime ideal of R.

Example 4.0.6. In Example 4.0.1, we constructed a discrete valuation on C(x) by
sending f (x) C(x) to n Z, provided
f (x) = x n
g(x)
,
h(x)
g(x), h(x) C[x], g(0) 6= 0, h(0) 6= 0.
The corresponding DVR is the localization C[x]hxi . It follows that the prime divisor
{0} = V(x) C = Spec(C[x]) has the local ring
OC,{0} = C[x]hxi
which is a DVR.
More generally, a normal ring or variety gives a DVR as follows.
Proposition 4.0.7.
(a) Let R be a normal domain and p R be a codimension one prime ideal. Then
the localization Rp is a DVR.
(b) Let X be a normal variety and D X a prime divisor. Then the local ring OX,D
is a DVR.
Proof. By Proposition 3.0.12, part (b) follows immediately from part (a) together
with (4.0.2) and (4.0.3).
It remains to prove part (a). The maximal ideal of Rp is the ideal mp = pRp
generated by p in Rp. The localization of a Noetherian ring is Noetherian (Exercise 4.0.4), and the same is true for normality by Exercise 1.0.7. It follows that the
local domain (Rp, mp) is Noetherian and normal.
We compute the dimension of Rp as follows. Since dim X = dim R (see [69,
Ex. 17 and 18 of Ch. 9, 4]), our hypothesis on D = V(p) implies that there are no
prime ideals strictly between {0} and p in R. By [10, Prop. 3.11], the same is true
for {0} and mp in Rp. It follows that Rp has dimension one. Then Rp is a DVR by
Theorem 4.0.4.

When D is a prime divisor on a normal variety X , the DVR OX,D means that
we have a discrete valuation
D : C(X ) Z,
where OX,D consists of 0 and those nonzero rational functions satisfying D ( f ) 0.

Given f OX,D \ {0}, we call D ( f ) the order of vanishing of f along the divisor
D. Thus the maximal ideal mX,D OX,D consists of 0 and those rational functions
159
that vanish on D. When f C(X ) satisfies D ( f ) = < 0, we say that f has a

pole of order || along D.
Weil Divisors. Recall that a prime divisor on an irreducible variety X is an irreducible subvariety of codimension one.
Definition 4.0.8. Div(X ) is the free abelian group generated by the prime divisors
on X . A Weil divisor is an element of Div(X ).
P
Thus a Weil divisor D Div(X ) is a finite sum D = i ai Di Div(X ) of prime
divisors Di with ai Z for all i. The divisor D is effective, written D 0, if the ai
are all nonnegative. The support of D is the union of the prime divisors appearing
in D:
[
Supp(D) =
Di .
ai 6=0
The Divisor of a Rational Function. An important class of Weil divisors comes

from rational functions. If X is normal, any prime divisor D on X corresponds to a
DVR OX,D with valuation D : C(X ) Z. Given f C(X ) , the integers D ( f )
tell us how f behaves on the prime divisors of X . Here is an important property of
these integers.
Lemma 4.0.9. If X is normal and f C(X ) , then D ( f ) is zero for all but a finite
number of prime divisors D X .
Proof. If f is constant, then it is a nonzero constant since f C(X ) . It follows
that D ( f ) = 0 for all D. On the other hand, if f is nonconstant, then we can find
a nonempty open subset U X such that f : U C is a nonconstant morphism.
Then V = f 1 (C ) is a nonempty open subset of X such that f |V : V C . The
complement X \ V is Zariski closed and hence is a union of irreducible components of dimension < n. Denote the irreducible components of codimension one
by D1 , . . . , Ds .
Now let D be prime divisor in X . If V D = , then D X \V , so that D is
contained in an irreducible component of X \V since D is irreducible. Dimension
considerations imply that D = Di for some i. On the other hand, if V D 6= , then
f is an invertible element of OX,D = OV ,V D , which implies that D ( f ) = 0.

Definition 4.0.10. Let X be a normal variety.
(a) The divisor of f C(X ) is
div( f ) =
D ( f ) D,
where the sum is over all prime divisors D X .
(b) div( f ) is called a principal divisor, and the set of all principal divisors is denoted Div0 (X ).
160
(c) Divisors D and E are linearly equivalent, written D E, if their difference is

a principal divisor, i.e., D E = div( f ) Div0 (X ) for some f C(X ) .
Lemma 4.0.9 implies that div( f ) Div(X ). If f , g C(X ) , then div( f g) =
div( f ) + div(g) and div( f 1 ) = div( f ) since valuations are group homomorphisms on C(X ) . It follows that Div0 (X ) is a subgroup of Div(X ).
Example 4.0.11. Let f = c(x a1 )m1 (x ar )mr C[x] be a polynomial of degree m > 0, where c C and a1 , . . . , ar C are distinct. Then:
P
When X = C, div( f ) = ri=1 mi {ai }.
P
When X = P1 = C {}, div( f ) = ri=1 mi {ai } m {}.

The divisor of f C(X ) can be written div( f ) = div0 ( f ) div ( f ), where
X
div0 ( f ) =
D ( f ) D
D ( f )>0
div ( f ) =
D ( f )<0
D ( f ) D.
We call div0 ( f ) the divisor of zeros of f and div ( f ) the divisor of poles of f .
Note that these are effective divisors.
Cartier Divisors. If D =
open subset, then
i ai Di
is a Weil divisor on X and U X is a nonempty
D|U =
UDi 6=
ai U Di
is a Weil divisor on U called the restriction of D to U .

We now define a special class of Weil divisors.
Definition 4.0.12. A Weil divisor D on a normal variety X is Cartier if it is locally
principal, meaning that X has an open cover {Ui }iI such that D|Ui is principal in
Ui for every i I. If D|Ui = div( fi )|Ui for i I, then we call {(Ui , fi )}iI the local
data for D.
A principal divisor is obviously locally principal. Thus div( f ) is Cartier for all
f C(X ) . One can also show that if D and E are Cartier divisors, then D + E and
D are Cartier (Exercise 4.0.5). It follows that the Cartier divisors on X form a
group CDiv(X ) satisfying
Div0 (X ) CDiv(X ) Div(X ).
Divisor Classes. For Weil and Cartier divisors, linear equivalence classes form the
following important groups.
161
Definition 4.0.13. Let X be a normal variety. Its class group is

Cl(X ) = Div(X )/Div0 (X ),
and its Picard group is
Pic(X ) = CDiv(X )/Div0 (X ).
We will give a more sophisticated definition of Pic(X ) in Chapter 6. Note that
since CDiv(X ) is a subgroup of Div(X ), we get a canonical injection
Pic(X ) Cl(X ).
In [131, II.6], Hartshorne writes The divisor class group of a scheme is a very
interesting invariant. In general it is not easy to calculate. Fortunately, divisor
class groups of normal toric varieties are easy to describe, as we will see in 4.1.
More Algebra. Before we can derive further properties of divisors, we need to
learn more about normal domains. Equation (3.0.2) shows that if X = Spec(R) is
irreducible, then
\
R =
OX,p .
pX
If a point p X corresponds to a maximal ideal m R, then the local ring OX,p is

the localization Rm. Hence the above equality can be written
\
R =
Rm.
m maximal
When R is normal, we get a similar result using codimension one prime ideals.
Theorem 4.0.14. If R is a Noetherian normal domain, then
\
R =
Rp.
codim p=1
Proof. Let K be the field of fractions of R and assume that a/b K, a, b R, lies
in Rp for all codimension one prime ideals p. It suffices to prove that a hbi. This
is obviously true when b is invertible in R, so we may assume that hbi is a proper
ideal of R. Then we have a primary decomposition (see [69, Ch. 4, 7])
hbi = q1 qs ,
and each prime ideal pi = qi is of the form pi = hbi : ci for some ci R. In the
terminology of [195, p. 38], the pi are the prime divisors of hbi.
(4.0.4)
Since R is Noetherian and normal, the Krull principal ideal theorem states that
every prime divisor of hbi has codimension one (see [195, Thm. 11.5] for a proof).
This implies that in the primary decomposition (4.0.4), the prime divisors pi have
codimension one and hence are distinct.
162
Note that a/b Rpi for all i by our assumption on a/b. This implies a bRpi .
Since (q j )pi = Rpi for j 6= i (Exercise 4.0.6), localizing (4.0.4) at pi shows that for
all i, we have
a bRpi = qi Rpi .
Since qi Rpi R = qi (Exercise 4.0.6), we obtain a
This result has the following useful corollary.
Ts
i=1 qi
= hbi.
Corollary 4.0.15. Let X be a normal variety and let f : U C be a morphism

defined on an open set U X . If X \U has codimension 2 in X , then f extends
to a morphism defined on all of X .
Proof. Since X has an affine open cover, we can assume that X = Spec(R), where
R is a Noetherian normal domain. If D X is a prime divisor, then U D 6= for
dimension reasons. It follows that f OU,UD = OX,D , so that
\
\
\
(4.0.5)
f OU,UD = OX,D =
Rp = R,
D
codim p=1
where the final equality is Theorem 4.0.14.
These results enable us to determine when the divisor of a rational function is

effective.
Proposition 4.0.16. Let X be a normal variety. If f C(X ) , then:
(a) div( f ) 0 if and only if f : X C is a morphism, i.e., f OX (X ).
(b) div( f ) = 0 if and only if f : X C is a morphism, i.e., f OX (X ).

In general, OX is the sheaf on X defined by
OX (U ) = {invertible elements of OX (U )}.
This is a sheaf of abelian groups under multiplication.
Proof. If f : X C is a morphism, then f OX,D for every prime divisor D, which
in turn implies D ( f ) 0. Hence div( f ) 0. Going the other way, suppose that
div( f ) 0. This remains true when we restrict to an affine open subset, so we may
assume that X is affine. Then div( f ) 0 implies
\
f OX,D ,
D
where the intersection is over all prime divisors. By (4.0.5), we conclude that f is
defined everywhere. This proves part (a), and part (b) follows immediately since
div( f ) = 0 if and only if div( f ) 0 and div( f 1 ) 0.
163
Singularities and Normality. The set of singular points of a variety X is denoted

Sing(X ) X .
We call Sing(X ) the singular locus of X . One can show that Sing(X ) is a proper
closed subvariety of X (see [131, Thm. I.5.3]). When X is normal, things are even
nicer.
Proposition 4.0.17. Let X be a normal variety. Then:
(a) Sing(X ) has codimension 2 in X .
(b) If X is a curve, then X is smooth.
Proof. You will prove part (b) in Exercise 4.0.7. A proof of part (a) can be found
in [245, Vol. 2, Thm. 3 of II.5].

Computing Divisor Classes. There are two results, one algebraic and one geometric, that enable us to compute class groups in some cases.
We begin with the algebraic result.
Theorem 4.0.18. Let R be a UFD and set X = Spec(R). Then:
(a) R is normal and every codimension one prime ideal is principal.
(b) Cl(X ) = 0.
Proof. For part (a), we know that a UFD is normal by Exercise 1.0.5. Let p be a
codimension one prime ideal of R and pick a p \ {0}. Since R is a UFD,
a=c
s
Y
piai ,
i=1
with the pi prime and c is invertible in R. Because p is prime, this means some
pi p, and since codim p = 1, this forces p = hpi i.
Turning to part (b), let D X be a prime divisor. Then p = I(D) is a codimension one prime ideal and hence is principal, say p = h f i. Then f generates the
maximal ideal of the DVR Rp, which implies D ( f ) = 1 (see the proof of Proposition 4.0.3). It follows easily that div( f ) = D. Then Cl(X ) = 0 since all prime
divisors are linearly equivalent to 0.

In fact, more is true: a normal Noetherian domain is a UFD if and only if every
codimension one prime ideal is principal (Exercise 4.0.8).
Example 4.0.19. C[x1 , . . . , xn ] is a UFD, so Cl(Cn ) = 0 by Theorem 4.0.18.
Before stating the geometric result, note that if U X is open and nonempty,
then restriction of divisors D 7 D|U induces a well-defined map Cl(X ) Cl(U )
(Exercise 4.0.9).
164
Theorem 4.0.20. Let U be a nonempty open subset of a normal variety X and let
D1 , . . . , Ds be the irreducible components of X \ U that are prime divisors. Then
the sequence
s
M
Z D j Cl(X ) Cl(U ) 0
j=1
P
is exact, where the first map sends sj=1 a j D j to its divisor class in Cl(X ) and the
second is induced by restriction to U .
P
Proof. Let D = i ai Di Div(U ) with Di a prime divisor in U . Then the Zariski
P
closure Di of Di in X is a prime divisor in X , and D = i ai Di satisfies D|U = D .
Hence Cl(X ) Cl(U ) is surjective.
Since each D j restricts to 0 in Div(U ), the composition of the two maps is

trivial. To finish the proof of exactness, suppose that [D] Cl(X ) restricts to 0 in
Cl(U ). This means that D|U is the divisor of some f C(U ) . Since C(U ) = C(X )
and the divisor of f in Div(X ) restricts to the divisor of f in Div(U ), it follows that
we have f C(X ) such that
D|U = div( f )|U .
This implies that L
the difference D div( f ) is supported on X \ U , which means
that D div( f ) sj=1 Z D j by the definition of the D j .
Example 4.0.21. Write P1 = C {} and note that {} is a prime divisor on P1 .

Then Theorem 4.0.20 and Example 4.0.19 give the exact sequence
Z{} Cl(P1 ) Cl(C) = 0.
Hence the map Z Cl(P1 ) defined by a 7 [a{}] is surjective. This map is
injective since a{} = div( f ) implies div( f )|C = 0, so that f (C, OC ) = C
by Proposition 4.0.16. Hence f is constant, which forces a = 0. If follows that
Cl(P1 ) Z.
Later in the chapter we will use similar methods to compute the class group of
an arbitrary normal toric variety.
Comparing Weil and Cartier Divisors. Once we understand Cartier divisors on
normal toric varieties, it will be easy to give examples of Weil divisors that are not
Cartier. On the other hand, there are varieties where every Weil divisor is Cartier.
Theorem 4.0.22. Let X be a normal variety. Then:
(a) If the local ring OX,p is a UFD for every p X , then every Weil divisor on X
is Cartier.
(b) If X is smooth, then every Weil divisor on X is Cartier.
165
Proof. If X is smooth, then OX,p is a regular local ring for all p X . Since every
regular local ring is a UFD (see 1.0), part (b) follows from part (a).
For part (a), it suffices to show that prime divisors are locally principal. This
condition is obviously local on X , so we may assume that X = Spec(R) is affine.
Let D = V(p) be a prime divisor on X , where p R is a codimension one prime
ideal. Note that D is obviously principal on U = X \ D since D|U = 0. It remains
to show that D is locally principal in a neighborhood of a point p D.
The point p corresponds to a maximal ideal m R. Thus p D implies p m.

Since p R has codimension one, it follows that the prime ideal pRm Rm also has
codimension one (this follows from [10, Prop. 3.11]). Then Theorem 4.0.18 implies that pRm is principal since Rm is a UFD by hypothesis. Thus pRm = (a/b)Rm
where a, b R and b
/ m. Since b is invertible in Rm, we in fact have pRm = aRm.
Now suppose p = ha1 , . . . , as i R. Then ai pRm = aRm, so that ai = (gi /hi )a,
where gi , hi R and hi
/ m, i.e., hi (p) 6= 0. If we set h = h1 hs , then pRh = aRh
follows easily. Then U = Spec(Rh ) is a neighborhood of p, and from here, it is
straightforward to see that D = div(a) on U .

Example 4.0.23. Since P1 is smooth, Theorem 4.0.22 and Example 4.0.21 imply
that Pic(P1 ) = Cl(P1 ) Z.
Sheaves of OX -modules. Weil and Cartier divisors on X lead to some important

sheaves on X . Hence we need a brief excursion into sheaf theory (we will go deeper
into the subject in Chapter 6). The sheaf OX was defined in 3.0. The definition
of a sheaf F of OX -modules is similar: for each open subset U X , there is an
OX (U )-module F (U ) with the following properties:
When U U , there is a restriction map
U,U : F (U ) F (U )
such that U,U is the identity and U ,U U,U = U,U when U U U .

Furthermore, U,U is compatible with the restriction map OX (U ) OX (U ).
If {U } is an open cover of U X , then the sequence

Y
Y
F (U U )
0 F (U )
F (U )
is exact, where the second arrow is defined by the restrictions U,U and the
double arrow is defined by U ,U U and U ,U U . Exactness means the
same as in 3.0.
When U 7 F (U ) satisfies just the first bullet, we say that F is a presheaf.
Given a sheaf of OX -modules F , elements of F (U ) are called sections of F

over U . The module of sections of F over U X is expressed in several ways:
F (U ) = (U , F ) = H 0 (U , F ).
166
We will use in this chapter and switch to H 0 in later chapters. Traditionally,

(X , F ) is called the module of global sections of F .
Example 4.0.24. Let f : X Y be a morphism of varieties and let F be a sheaf
of OX -modules on X . The direct image sheaf f F on Y is defined by
U 7 F ( f 1 (U ))
for U Y open. Then f F is a sheaf of OY -modules. For i : Y X , the direct
image i OY was mentioned in 3.0.
If F and G are sheaves of OX -modules, then a homomorphism of sheaves

: F G consists of OX (U )-module homomorphisms
U : F (U ) G (U ),
such that the diagram
F (U )
/ G (U )
U ,V
U ,V
F (V )
/ G (V )
commutes whenever V U . It should be clear what it means for sheaves F , G of

OX -modules to be isomorphic, written F G .
Example 4.0.25. Let f : X Y be a morphism of varieties. If U Y is open, then
composition with f induces a natural map
OY (U ) OX ( f 1 (U )) = f OX (U ).
This defines a sheaf homomorphism OY f OX .
Over an affine variety X = Spec(R), there is a standard way to get sheaves of

OX -modules. Recall that a nonzero element f R gives the localization R f such
that X f = Spec(R f ) is the open subset X \ V( f ). Given an R-module M, we get the
e of OX -modules such
R f -module M f = M R R f . Then there is a unique sheaf M
that
e f ) = Mf
M(X
for every nonzero f R (see [131, Prop. II.5.1]). This globalizes as follows.
Definition 4.0.26. Let F be a sheaf of OX -modules on a variety X .
(a) Let U X be open. Then the restriction F |U is the sheaf of OU -modules

defined by F |U (V ) = F (V ) for V U open.
(b) F is quasicoherent if X has an affine open cover {U }, U = Spec(R ), such

e .
that for each , there is an R -module M satisfying F |U M
(c) If in addition each M is a finitely generated R -module, then we say that F
is coherent.
167
The Sheaf of a Weil Divisor. Let D be a Weil divisor on a normal variety X . We

will show that D determines a sheaf OX (D) of OX -modules on X . Recall that if
U X is open, then OX (U ) consists of all morphisms U C. Proposition 4.0.16
tells us that an arbitrary element f C(X ) is a morphism on U if and only if
div( f )|U 0. It follows that the sheaf OX is defined by
U 7 OX (U ) = { f C(X ) | div( f )|U 0} {0}.
In a similar way, we define the sheaf OX (D) by

(4.0.6)
U 7 OX (D)(U ) = { f C(X ) | (div( f ) + D)|U 0} {0}.
Proposition 4.0.27. Let D be a Weil divisor on a normal variety X . Then the sheaf
OX (D) defined in (4.0.6) is a coherent sheaf of OX -modules on X .
Proof. In Exercise 4.0.10 you will show that OX (D) is a sheaf of OX -modules.
The proof is a nice application of the properties of valuations.
To show that OX (D) is coherent, we may assume that X = Spec(R). Let K be
the field of fractions of R. It suffices to prove the following two assertions:
M = (X , OX (D)) = { f K | div( f ) + D 0} {0} is a finitely generated
R-module.
(X f , OX (D)) = M f for all nonzero f R.
For the first bullet, we will prove the existence of an element h R \ {0} such
that h (X , OX (D)) R. This will imply that h (X , OX (D)) is an ideal of R and
hence has a finite basis since R is Noetherian. It will follow immediately that
(X , OX (D)) is a finitely generated R-module.
P
Write D = si=1 ai Di . Since supp(D) is a proper subvariety of X , we can find
g R \ {0} that vanishes on each Di . Then Di (g) > 0 for every i, so there is m N
with m Di (g) > ai for all i. Since div(g) 0, it follows that m div(g) D 0. Now
let f (X , OX (D)). Then div( f ) + D 0, so that
div(g m f ) = m div(g) + div( f ) = m div(g) D + div( f ) + D 0
since a sum of effective divisors is effective. By Proposition 4.0.16, we conclude

that g m f OX (X ) = R. Hence h = g m R has the desired property.
To prove the second bullet, observe that M K and f R \ 0 imply that

n g
o
Mf =
|
g
(X
,
O
(D)),
m
0
.
X
fm
It
Piss also easy to see that M f (X f , OX (D)). For the opposite inclusion, let D =
i=1 ai Di and write {1, . . . , s} = I J where Di X f 6= for i I and D j V( f )
for j J. Given h (X f , OX (D)), (div(h) + D)|X f 0 implies that Di (h) ai
for i I. There is no constraint on D j (h) for j J, but f vanishes on D j for j J,
so that D j ( f ) > 0. Hence we can pick m N sufficiently large such that
m D j ( f ) + D j (h) > 0 for j J.
168
Since div( f ) 0, it follows easily that div( f m h) + D 0 on X . Thus g = f m h

(X , OX (D)), and then h = g/ f m has the desired form.

The sheaves OX (D) are more than just coherent; they have the additional property of being reflexive. Furthermore, when D is Cartier, OX (D) is invertible. The
definitions of invertible and reflexive will be given in Chapters 6 and 8 respectively.
For now, we give two results about the sheaves OX (D). Here is the first.
Proposition 4.0.28. Distinct prime divisors D1 , . . . , Ds on a normal variety X give
the divisor D = D1 + + Ds and the subvariety Y = Supp(D) = D1 Ds .
Then OX (D) is the ideal sheaf IY of Y , i.e.,
(U , OX (D)) = { f OX (U ) | f vanishes on Y }
for all open subsets U X .
Proof. Since sheaves are local, we may assume that X = Spec(R). Then note that
f (X , OX (D)) implies div( f ) D 0, so div( f ) D 0 since D is effective.
Thus f R by Proposition 4.0.16 and hence (X , OX (D)) is an ideal of R.
Let pi = I(Di ) R be the prime ideal of Di . Then, for f R, we have
Di ( f ) > 0 f pi Rpi f pi ,
where the last equivalence uses the easy equality pi Rpi R = pi . Hence div( f ) D
if and only if f vanishes on D1 , . . . , Ds , and the proposition follows.

Linear equivalence of divisors tells us the following interesting fact about the
associated sheaves.
Proposition 4.0.29. If D E are linearly equivalent Weil divisors, then OX (D)
and OX (E) are isomorphic as sheaves of OX -modules.
Proof. By assumption, we have D = E + div(g) for some g C(X ) . Then
f (X , OX (D)) div( f ) + D 0
div( f ) + E + div(g) 0
div( f g) + E 0
f g (X , OX (E)).
Thus multiplication by g induces an isomorphism (X , OX (D)) (X , OX (E))
which is clearly an isomorphism of (X , OX )-modules.
The same argument works over any Zariski open set U , and the isomorphisms
are easily seen to be compatible with the restriction maps.

The converse of Proposition 4.0.29 is also true, i.e., an OX -module isomorphism OX (D) OX (E) implies that D E. The proof requires knowing more
about the sheaves OX (D) and hence will be postponed until Chapter 8.
169
Exercises for 4.0.

4.0.1. Complete the proof of part (b) of Proposition 4.0.3.
4.0.2. Prove (c) (d) in Theorem 4.0.4. Hint: Let m be the maximal ideal of R. Since R
has dimension one, it is regular if and only if m/m2 has dimension one as a vector space
over R/m. For (d) (c), use Nakayamas Lemma (see [10, Props. 2.6 and 2.8]).
4.0.3. This exercise will study the rings OX ,D and Rp .
(a) Prove (4.0.2).
(b) Let p be a prime ideal of a ring R and let Rp denote the localization of R with respect
to the multiplicative subset R \ p. Prove that Rp is a local ring and that its maximal
ideal is the ideal pRp Rp generated by p.
4.0.4. Let S be a multiplicative subset of a Noetherian ring R. Prove that the localization
RS is Noetherian.
4.0.5. Let D and E be Weil divisors on a normal variety.
(a) If D and E are Cartier, show that D + E and D are also Cartier.
(b) If D E, show that D is Cartier if and only if E is Cartier.
4.0.6. Complete the proof of Theorem 4.0.14.

4.0.7. Prove that a normal curve is smooth.
4.0.8. Let R be a Noetherian normal domain. Prove that the following are equivalent:
(a) R is a UFD.
(b) Cl(Spec(R)) = 0.
(c) Every codimension one prime ideal of R is principal.
Hint: For (b) (c), assume that D = div( f ) corresponds to p. Use Theorem 4.0.14 to
show f R and use the Krull principal ideal theorem to show h f i is primary in R. Then
pRp = f Rp and [10, Prop. 4.8] imply p = h f i. For (c) (a), let a R be noninvertible
and let D1 , . . . , Ds be the codimension one irreducible components of V(a). If I(Di ) = hai i,
Q
D (a)
compare the divisors of a and si=1 ai i using Proposition 4.0.16.
4.0.9. Prove that the restriction map D 7 D|U induces a well-defined homomorphism
Cl(X) Cl(U).
4.0.10. Let D be a Weil divisor on a normal variety X. Prove that (4.0.6) defines a sheaf
OX (D) of OX -modules.
4.0.11. For each of the following rings R, give a careful description of the field of fractions
K and show that the ring is a DVR by constructing an appropriate discrete valuation on K.
(a) R = {a/b Q | a, b Z, b 6= 0, gcd(b, p) = 1}, where p is a fixed prime number.
(b) R = C{{z}}, the ring consisting of all power series in z with coefficients in C that have
a positive radius of convergence.
4.0.12. The plane curve V(x 3 y 2 ) C2 has coordinate ring R = C[x, y]/hx 3 y 2 i. As
noted in Example 1.1.15, this is the coordinate ring of the affine toric variety given by
the affine semigroup S = {0, 2, 3, . . . }. This semigroup is not saturated, which means that
R C[S] = C[t 2 ,t 3 ] is not normal by Theorem 1.3.5. It follows that R is not a DVR by
170
Theorem 4.0.4. Give a direct proof of this fact using only the definition of DVR. Hint: The
field of fractions of C[t 2 ,t 3 ] is C(t). If C[t 2 ,t 3 ] comes from the discrete valuation , what
is (t)?
4.0.13. Let X be a normal variety. Use Proposition 4.0.16 to prove that there is an exact
sequence
1 OX (X) C(X) Div(X) Cl(X) 0,
where the map C(X) Div(X) is f 7 div( f ) and Div(X) Cl(X) is D 7 [D]. Similarly,
prove that there is an exact sequence
1 OX (X) C(X) CDiv(X) Pic(X) 0.

P
4.0.14. Let D = codim p=1 ap Dp be a Weil divisor on a normal affine variety X = Spec(R).
As usual, let K be the field of fractions of R. Here you give an algebraic description of
(X, OX (D)) in terms of the prime ideals p.
(a) Let p be a codimension one prime of R, so that Rp is a DVR. Hence the maximal ideal
pRp is principal. Use this to define pa Rp K for all a Z.
(b) Prove that
(X, OX (D)) =
pap Rp .
codim p=1
(c) Now assume that D is effective, i.e., ap 0 for all p. Prove that (X, OX (D)) is the
ideal of R given by
\
(X, OX (D)) =
pap Rp .
codim p=1
4.0.15. Let R be an integral domain with field of fractions K. A finitely generated Rsubmodule of K is called a fractional ideal. If R is normal and D is a Weil divisor on
X = Spec(R), explain why (X, OX (D)) K is a fractional ideal.
4.1. Weil Divisors on Toric Varieties

Let X be the toric variety of a fan in NR with dim NR = n. Then X is normal
of dimension n. We will use torus-invariant prime divisors and characters to give a
lovely description of the class group of X .
The Divisor of a Character. The order of vanishing of a character along a torusinvariant prime divisor is determined by the polyhedral geometry of the fan.
By the Orbit-Cone Correspondence (Theorem 3.2.6), k-dimensional cones
of correspond to (n k)-dimensional TN -orbits in X . As in Chapter 3, (1)
is the set of 1-dimensional cones (i.e., the rays) of . Thus (1) gives the
codimension 1 orbit O() whose closure O() is a TN -invariant prime divisor on
X . To emphasize that O() is a divisor we will denote it by D rather than V ().
Then D = O() gives the DVR OX ,D with valuation
= D : C(X ) Z.
171
Recall that the ray (1) has a minimal generator u N. Also note that
when m M, the character m : TN C is a rational function in C(X ) since TN
is Zariski open in X .
Proposition 4.1.1. Let X be the toric variety of a fan . If the ray (1) has
minimal generator u and m is character corresponding to m M, then
( m ) = hm, u i.
Proof. Since u N is primitive, we can extend u to a basis e1 = u , e2 , . . . , en of

N, then we can assume N = Zn and = Cone(e1 ) Rn . By Example 1.2.21, the
corresponding affine toric variety is
U = Spec(C[x1 , x21 , . . . , xn1 ]) = C (C )n1
and D U is defined by x1 = 0. It follows easily that the DVR is
OX ,D = OU ,U D = C[x1 , . . . , xn ]hx1 i .
Similar to Example 4.0.6, f C(x1 , . . . , xn ) has valuation ( f ) = Z when

g
f = x1 , g, h C[x1 , . . . , xn ] \ hx1 i.
h
To relate this to ( m ), note that x1 , . . . , xn are the characters of the dual basis
of e1 = u , e2 , . . . , en N. It follows that given any m M, we have
hm,e1 i hm,e2 i
xnhm,en i
x2
m = x1
hm,u i hm,e2 i
xnhm,en i .
x2
= x1
Comparing this to the previous equation implies that ( m ) = hm, u i.
We next compute the divisor of a character. As above, a ray (1) gives:
A minimal generator u N.
A prime TN -invariant divisor D = O() on X .
We will use this notation for the remainder of the chapter.

Proposition 4.1.2. For m M, the character m is a rational function on X , and
its divisor is given by
X
div( m ) =
hm, u i D .
(1)
Proof. The Orbit-Cone Correspondence (Theorem 3.2.6) implies that the D are
the irreducible components of X \ TNS
. Since m is defined and nonzero on TN , it
follows that div( m ) is supported on (1) D . Hence
X
div( m ) =
D ( m ) D .
(1)
Then we are done since D
( m ) =
hm, u i by Proposition 4.1.1.
172
P
Computing the Class Group. Divisors of the form (1) a D are precisely the
divisors invariant under the torus action on X (Exercise 4.1.1). Thus
M
DivTN (X ) =
Z D Div(X )
(1)
is the group of TN -invariant Weil divisors on X . Here is the main result of this
section.
Theorem 4.1.3. We have the exact sequence
M DivTN (X ) Cl(X ) 0,
where the first map is m 7 div( m ) and the second sends a TN -invariant divisor to
its divisor class in Cl(X ). Furthermore, we have a short exact sequence
0 M DivTN (X ) Cl(X ) 0
if and only if {u | (1)} spans NR , i.e., X has no torus factors.
Proof. Since the D are the irreducible components of X \ TN , Theorem 4.0.20
implies that we have an exact sequence
DivTN (X ) Cl(X ) Cl(TN ) 0.
Since C[x1 , . . . , xn ] is a UFD, the same is true for C[x11 , . . . , xn1 ]. This is the coordinate ring of the torus (C )n , which is isomorphic to the coordinate ring C[M]
of the torus TN . Hence C[M] is also a UFD, which implies Cl(TN ) = 0 by Theorem 4.0.18. We conclude that DivTN (X ) Cl(X ) is surjective.
The composition M DivTN (X ) Cl(X ) is obviously zero since the first
map is m 7 div( m ). Now suppose that D DivTN (X ) maps to 0 in Cl(X ). Then
D = div( f ) for some f C(X ) . Since the support of D misses TN , this implies
that div( f ) restricts to 0 on TN . When regarded as an element of C(TN ) , f has
zero divisor on TN , so that f C[M] by Proposition 4.0.16. Thus f = c m for
some c C and m M (Exercise 3.3.4). It follows that on X ,
D = div( f ) = div(c m ) = div( m ),
which proves exactness at DivTN (X ).
P
Finally, suppose that m M with div( m ) = (1) hm, u i D is the zero
divisor. Then hm, u i = 0 for all (1), which forces m = 0 when the u span
NR . This gives the desired exact sequence. Conversely, if the sequence is exact,
then one easily sees that the u span NR , which by Corollary 3.3.10 is equivalent
to X having no torus factors.

In particular, we see that Cl(X ) is a finitely generated abelian group.
173
Examples. It is easy to compute examples of class groups of toric varieties. In

practice, one usually picks a basis e1 , . . . , en of M, so that M Zn and (via the dual
basis) N Zn . Then the pairing hm, ui becomes dot product. We list the rays of
as 1 , . . . , r with corresponding ray generators u1 , . . . , ur Zn . We will think
of ui as the column vector (he1 , ui i, . . . , hen , ui i)T , where the superscript denotes
transpose.
With this setup, the map M DivTN (X ) in Theorem 4.1.3 is the map
A : Zn Zr
represented by the matrix whose rows are the ray generators u1 , . . . , ur . In other
words, A = (u1 , . . . , ur )T . By Theorem 4.1.3, the class group of X is the cokernel
of this map, which is easily computed from the Smith normal form of A.
When we want to think in terms of divisors, we let Di be the TN -invariant prime
divisor corresponding to i (1).
Example 4.1.4. The affine toric surface described in Example 1.2.22 comes from
the cone = Cone(de1 e2 , e2 ). For d = 3, is shown in Figure 1. The resulting
u2
u1
1
Figure 1. The cone when d = 3
toric variety U is the rational normal cone Cbd . Using the ray generators u1 =
de1 e2 = (d, 1) and u2 = e2 = (0, 1), we get the map Z2 Z2 given by the
matrix

d 1
A=
.
0
1
This makes it easy to compute that
Cl(Cbd ) Z/dZ.
We can also see this in terms of divisors as follows. The class group Cl(Cbd ) is
generated by the classes of the divisors D1 , D2 corresponding to 1 , 2 , subject to
174
the relations coming from the exact sequence of Theorem 4.1.3:

0 div( e1 ) = he1 , u1 i D1 + he1 , u2 i D2 = d D1
0 div( e2 ) = he2 , u1 i D1 + he2 , u2 i D2 = D1 + D2 .

Thus Cl(Cbd ) is generated by [D1 ] with d[D1 ] = 0, giving Cl(Cbd ) Z/dZ.
Example 4.1.5. In Example 3.1.4, we saw that the blowup of C2 at the origin is
the toric variety Bl0 (C2 ) given by the fan shown in Figure 2.
u2
u0
u1
Figure 2. The fan for the blowup of C2 at the origin
The ray generators are u1 = e1 , u2 = e2 , u0 = e1 + e2 corresponding to divisors

D1 , D2 , D0 . By Theorem 4.1.3, the class group is generated by the classes of the Di
subject to the relations
0 div( e1 ) = D1 + D0
0 div( e2 ) = D2 + D0 .
Thus Cl(Bl0 (C2 )) Z with generator [D1 ] = [D2 ] = [D0 ]. This calculation can
also be done using matrices as in the previous example.
Example 4.1.6. The fan of P n has ray generators given by u0 = e1 en and

u1 = e1 , . . . , un = en . Thus the map M DivTN (P n ) can be written as
Zn Zn+1
(a1 , . . . , an ) 7 (a1 an , a1 , . . . , an ).
Using the map
Zn+1 Z
(b0 , . . . , bn ) 7 b0 + + bn ,
one gets the exact sequence
0 Zn Zn+1 Z 0,
175
which proves that Cl(P n ) Z, generalizing Example 4.0.21. It is easy to redo this
calculation using divisors as in the previous example.
Example 4.1.7. The class group Cl(P n P m ) is isomorphic to Z2 . More generally,

Cl(X1 X2 ) Cl(X1 ) Cl(X2 ).
You will prove this in Exercise 4.1.2.
Example 4.1.8. The Hirzebruch surfaces Hr are described in Example 3.1.16.

The fan for Hr appears in Figure 3, along with the ray generators u1 = e1 + re2 ,
u2 = e2 , u3 = e1 , u4 = e2 .
u1 = (1, r)
u2
u3
u4
Figure 3. A fan r with Xr Hr
The class group is generated by the classes of D1 , D2 , D3 , D4 , with relations

0 div( e1 ) = D1 + D3
0 div( e2 ) = r D1 + D2 D4 .
It follows that Cl(Hr ) is the free abelian group generated by [D1 ] and [D2 ]. Thus
Cl(Hr ) Z2 .
In particular, r = 0 gives Cl(H0 ) = Cl(P1 P1 ) Z2 , which is a special case of

Example 4.1.7.
Exercises for 4.1.

4.1.1. This exercise will determine which divisors are invariant under the TN -action on
X . Given t TN and p X , the TN -action gives t p X . If D is a primeP
divisor,
the TN -action
gives
the
prime
divisor
t
D.
For
an
arbitrary
Weil
divisor
D
=
i a i Di ,
P
t D = i ai (t Di ). Then D is TN -invariant if t D = D for all t TN .
P
(a) Show that (1) a D is TN -invariant.
(b) Conversely, show that any TN -invariant Weil divisor can be written as in part (a). Hint:
Consider Supp(D) and use the Orbit-Cone Correspondence.
176
4.1.2. Given fans 1 in (N1 )R and 2 in (N2 )R , we get the product fan
1 2 = {1 2 | i i },
which by Proposition 3.1.14 is the fan of the toric variety X1 X2 . Prove that
Cl(X1 X2 ) Cl(X1 ) Cl(X2 ).
Hint: The product fan has rays 1 {0} and {0} 2 for 1 1 (1) and 2 2 (1).
4.1.3. Redo the divisor class group calculation given in Example 4.1.5 using matrices, and
redo the calculation given in Example 4.1.6 using divisors.
4.1.4. The blowup of Cn at the origin is the toric variety Bl0 (Cn ) of the fan described in
Example 3.1.15. Prove that Cl(Bl0 (Cn )) Z.
4.1.5. The weighted projective space P(q0 , . . . , qn ), gcd(q0 , . . . , qn ) = 1, is built from a fan
in N = Zn+1 /Z(q0 , . . . , qn ). The dual lattice is
M = {(a0 , . . . , an ) Zn+1 | a0 q0 + + an qn = 0}.
Let u0 , . . . , un N denote the images of the standard basis e0 , . . . , en Zn+1 . The ui are the
ray generators of the fan giving P(q0 , . . . , qn ). Define maps
M Zn+1 : m 7 (hm, u0 i, . . . , hm, un i)
Zn+1 Z : (a0 , . . . , an ) 7 a0 q0 + + an qn .
Show that these maps give an exact sequence

0 M Zn+1 Z 0
and conclude that Cl(P(q0 , . . . , qn )) Z.
4.2. Cartier Divisors on Toric Varieties

Let X be the toric variety of a fan . We will use the same notation as in 4.1,
where each (1) gives a minimal ray generator
u and a TN -invariant prime
P
divisor D X . In what follows, we write for a summation over the rays
(1) when there is no danger of confusion.
Computing the Picard Group. A Cartier divisor D on X is also a Weil divisor
and hence
X
a D , a Z,
D
by Theorem 4.1.3. Then
a D
is Cartier since D is (Exercise 4.0.5). Let
CDivTN (X ) DivTN (X )
denote the subgroup of DivTN (X ) consisting of TN -invariant Cartier divisors. Since

div( m ) CDivTN (X ) for all m M, we get the following immediate corollary of
Theorem 4.1.3.
177
Theorem 4.2.1. We have an exact sequence

M CDivTN (X ) Pic(X ) 0,
where the first map is defined above and the second sends a TN -invariant divisor to
its divisor class in Pic(X ). Furthermore, we have a short exact sequence
0 M CDivTN (X ) Pic(X ) 0
if and only if {u | (1)} spans NR .
Our next task is to determine the structure of CDivTN (X ). In other words,

which TN -invariant divisors are Cartier? We begin with the affine case.
Proposition 4.2.2. Let NR be a strongly convex polyhedral cone. Then:
(a) Every TN -invariant Cartier divisor on U is the divisor of a character.
(b) Pic(U ) = 0.
P
Proof. Let R = C[ M]. First suppose that D = a D is an effective TN invariant Cartier divisor. Using Proposition 4.0.16 as in the proof of Proposition 4.0.28, we see that
(U , OU (D)) = { f K | f = 0, or f 6= 0 and div( f ) D}
is an ideal I R. Furthermore, I is TN -invariant since D is. Hence
M
M
(4.2.1)
I=
C m =
C m
m I
div( m )D
by Lemma 1.1.16.
Under the Orbit-Cone Correspondence (Theorem 3.2.6), a ray (1) gives
an inclusion O() O() = D . Thus
\
O() D .
Now fix a point p O(). Since D is Cartier, it is locally principal, and in particular
is principal in a neighborhood U of p. Shrinking U if necessary, we may assume
that U = (U )h = Spec(Rh ), where h R satisfies h(p) 6= 0.
Thus D|U = div( f )|U for some f C(U ) . Since D is effective, f Rh by

Proposition 4.0.16, and since h is invertible on U , we may assume f R. Then
X
X
X
(4.2.2)
div( f ) =
D ( f ) D +
E ( f ) E
D ( f ) D = D.
E6=D
Here, E6=D denotes the sum over all prime divisors different from the D . The
first equality is the definition of div( f ), the second inequality follows since f R,
and the final equality follows from D|U = div( f )|U since p U D for all
(1). Then f I since div( f ) D by (4.2.2).
178
P
Using (4.2.1), we can write f = i ai mi with ai C and div( mi ) D.
Restricting to U , this becomes div( mi )|U div( f )|U , which implies that mi / f is
a morphism on U by Proposition 4.0.16. Then
P
ai mi X mi
1= i
=
ai
f
f
i
( mi / f )(p)
and p U imply that

6= 0 for some i. Hence mi / f is nonvanishing in
some open set V with p V U . It follows that
div( mi )|V = div( f )|V = D|V .
S
Since div( mi ) and D have support contained in D and every D meets V (this
follows from p V D ), we have div( mi ) = D.
To finish the proof of (a), let D be an arbitrary TN -invariant Cartier divisor on

U . Since dim = dim MR ( is strongly convex), we can find m M such
that hm, u i > 0 for all (1). Thus div( m ) is a positive linear combination of
the D , which implies that D = D + div( km ) 0 for k N sufficiently large. The
above argument implies that D is the divisor of a character, so that the same is true
for D. This completes the proof of part (a), and part (b) follows immediately using
Theorem 4.2.1.

Example 4.2.3. The rational normal cone Cbd is the affine toric variety of the cone
= Cone(de1 e2 , e2 ) R2 . We saw in Example 4.1.4 that Cl(U ) Z/dZ.
The edges 1 , 2 of give prime divisors D1 , D2 on Cbd , and the computations of
Example 4.1.4 show that [D1 ] = [D2 ] generates Cl(U ). Since Pic(U ) = 0 by
Proposition 4.2.2, it follows that the Weil divisors D1 , D2 are not Cartier if d > 1.
Next consider the fan 0 consisting of the cones 1 , 2 , {0}. This is a subfan of
the fan giving Cbd , and the corresponding toric variety is X0 Cbd \ { }, where
is the distinguished point that is the unique fixed point of the TN -action on
Cbd . The variety X0 is smooth since every cone in 0 is smooth (Theorem 3.1.19).
Since 0 and have the same 1-dimensional cones, they have the same class group
by Theorem 4.1.3. Thus
Pic(X0 ) = Cl(X0 ) = Cl(X ) = Cl(Cbd ) Z/dZ.
It follows that X0 is a smooth toric surface whose Picard group has torsion.
Example 4.2.4. One of our favorite examples is X = V(xy zw) C4 , which is the
toric variety of the cone = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) R3 . The ray generators
are
u1 = e1 , u2 = e2 , u3 = e1 + e3 , u4 = e2 + e3 .
Note that u1 + u4 = u2 + u3 . Let Di X be the divisor corresponding to ui . In
Exercise 4.2.1 you will verify that
a1 D1 + a2 D2 + a3 D3 + a4 D4 is Cartier a1 + a4 = a2 + a3
179
and that Cl(X ) Z. Since Pic(X ) = 0, we see that the Di are not Cartier, and in
fact no positive multiple of Di is Cartier.
Example 4.2.3 shows that the Picard group of a normal toric variety can have
torsion. However, if we assume that has a cone of maximal dimension, then the
torsion goes away. Here is the precise result.
Proposition 4.2.5. Let X be the toric variety of a fan in NR Rn . If contains
a cone of dimension n, then Pic(X ) is a free abelian group.
Proof. By the exact sequence in Theorem 4.2.1, it suffices to show that if D is a
TN -invariant Cartier divisor and k D is the divisor of a P
character for some k > 0,
then the same is true for D. To prove this, write D = a D and assume that
k D = div( m ), m M.
Let have dimension n. Since D is Cartier, its restriction to U is also Cartier.
Using the Orbit-Cone Correspondence, we have
X
D|U =
a D .
(1)
This is principal on U by Proposition 4.2.2, so that there is m M such that
D|U = div( m )|U . This implies that

a = hm , u i
for all (1).
On the other hand, k D = div( m ) implies that

k a = hm, u i for all (1).
Together, these equations imply
hk m , u i = k a = hm, u i for all (1).
The u span NR since dim = n. Then the above equation forces k m = m, and

D = div( m ) follows easily.
This proposition does not contradict the torsion Picard group in Example 4.2.3
since the fan 0 in that example has no maximal cone.
Comparing Weil and Cartier Divisors. Here is an application of Proposition 4.2.2.
equivalent:
(a) Every Weil divisor on X is Cartier.
(b) Pic(X ) = Cl(X ).
(c) X is smooth.
180
Proof. (a) (b) is obvious, and (c) (a) follows from Theorem 4.0.22. For
the converse, suppose that every Weil divisor on X is Cartier and let U X be
the affine open subset corresponding to . Since Cl(X ) Cl(U ) is onto
by Theorem 4.0.20, it follows that every Weil divisor on U is Cartier. Using
Pic(U ) = 0 from Proposition 4.2.2 and the exact sequence from Theorem 4.1.3,
we conclude that m 7 div( m ) induces a surjective map
M
M DivTN (U ) =
Z D .
(1)
Writing (1) = {1 , . . . , s }, this map becomes

M Zs
(4.2.3)
m 7 (hm, u1 i, . . . , hm, us i).

P
Now define : Zs N by (a1 , . . . , as ) = si=1 ai ui . The dual map
: M = HomZ (N, Z) HomZ (Zs , Z) = Zs
is easily seen to be (4.2.3). In Exercise 4.2.2 you will show that

(4.2.4)
is surjective is injective and N/(Zs ) is torsion-free.
u1 , . . . , us can be extended to a basis of N.
The first part of the proof shows that is surjective. Then (4.2.4) implies that the
u for (1) can be extended to a basis of N, which implies that is smooth.
Then X is smooth by Theorem 3.1.19.

Proposition 4.2.6 has a simplicial analog. Recall that X is simplicial when
every is simplicial, meaning that the minimal generators of are linearly
independent over R. You will prove the following result in Exercise 4.2.2.
equivalent:
(a) Every Weil divisor on X has a positive integer multiple that is Cartier.
(b) Pic(X ) has finite index in Cl(X ).
(c) X is simplicial.
In the literature, a Weil divisor is called Q-Cartier if some positive integer multiple is Cartier. Thus Proposition 4.2.7 characterizes those normal toric varieties for
which all Weil divisors are Q-Cartier.
Describing Cartier Divisors. We can use Proposition 4.2.2 to characterize TN invariant Cartier divisors as follows. Let max be the set of maximal cones
of , meaning cones in that are not proper subsets of another cone in .
Theorem 4.2.8. Let X be the toric variety of the fan and let D =
Then the following are equivalent:
(a) D is Cartier.
181
P
a D .
(b) D is principal on the affine open subset U for all .
(c) For each , there is m M with hm , u i = a for all (1).
(d) For each max , there is m M with hm , u i = a for all (1).
Furthermore, if D is Cartier and {m } is as in part (c), then:

(1) m is unique modulo M() = M.
(2) If is a face of , then m m mod M( ).

P
Proof. Since D|U = (1) a D , the equivalences (a) (b) (c) follow immediately from Proposition 4.2.2. The implication (c) (d) is clear, and (d) (c)
follows because every cone in is a face of some max and if m max
works for , it also works for all faces of .
For (1), suppose that m M satisfies hm, u i = a for all (1). Then,
given m M, we have
hm , u i = a for all (1) hm m , u i = 0 for all (1)
hm m , ui = 0 for all u
m m M = M().
It follows that m is unique modulo M(). Since m works for any face of ,
uniqueness implies that m m mod M( ), and (2) follows.

The m of part (c) of the theorem satisfy D|U = div( m )|U for all .
Thus {(U , m )} is local data for D in the sense of Definition 4.0.12. We
call {m } the Cartier data of D.
The minus signs in parts (c) and (d) of the theorem are related to the minus
signs in the facet presentation of a lattice polytope given in (2.2.2), namely
P = {m MR | hm, uF i aF for all facets F of P}.
We will say more about this below. The minus signs are also related to support
functions, to be discussed later in the section.
When is a complete fan in NR Rn , part (d) of Theorem 4.2.8 can be recast
as follows. Let (n)P= { | dim = n}. In Exercise 4.2.3 you will show that
a Weil divisor D = a D is Cartier if and only if:
(d) For each (n), there is m M with hm, u i = a for all (1).
Part (1) of Theorem 4.2.8 shows that these m s are uniquely determined.
In general, each m in Theorem 4.2.8 is only unique modulo M(). Hence we

can regard m as a uniquely determined element of M/M(). Furthermore, if is
a face of , then the canonical map M/M() M/M( ) sends m to m .
182
There are two ways to turn these observations into a complete description of
CDivTN (X ). For the first, write
max = {1 , . . . , r }
and consider the map
M
i
M/M(i )
M
i< j
M/M(i j )
(mi )i 7 (mi m j )i< j .
In Exercise 4.2.4 you will prove the following.

Proposition 4.2.9. There is a natural isomorphism

L
L
CDivTN (X ) ker
i M/M(i )
i< j M/M(i j ) .
For readers who know inverse limits (see [10, p. 103]), a more sophisticated
description of CDivTN (X ) comes from the directed set (, ), where is the face
relation. We get an inverse system where gives M/M() M/M( ), and
the inverse limit gives an isomorphism
(4.2.5)
CDivTN (X ) lim M/M().
The Toric Variety of a Polytope. In Chapter 2, we constructed the toric variety

XP of a full dimensional lattice polytope P MR . If MR Rn , this means that
dim P = n. As noted above, P has a canonical presentation
(4.2.6)
P = {m MR | hm, uF i aF for all facets F of P},
where aF Z and uF N is the inward-pointing facet normal that is the minimal

generator of the ray F = Cone(uF ). The normal fan P consists of cones Q
indexed by faces Q P, where
Q = Cone(uF | F contains Q).
Proposition 2.3.8 implies that the fan P is complete. Furthermore, the vertices of
P correspond to the maximal cones in P (n), and the facets of P correspond to the
rays in P (1).
The ray generators of the normal fan P are the facet normals uF . The corresponding prime divisors in XP will be denoted DF . Everything is now indexed by
the facets F of P. The normal fan tells us the facet normals uF in (4.2.6), but P
cannot give us the integers aF in (4.2.6). For these, we need the divisor
X
(4.2.7)
DP =
aF DF .
F
As we will see in later chapters, this divisor plays a central role in the study of
projective toric varieties. For now, we give the following useful result.
Proposition 4.2.10. DP is a Cartier divisor on XP and DP 6 0.
183
Proof. A vertex v P corresponds to a maximal cone v , and a ray F lies in v (1)

if and only if v F. But v F implies that hv, uF i = aF . Note also that v M
since P is a lattice polytope. Thus we have v M such that hv, uF i = aF for all
F v (1), so that DP is Cartier by Theorem 4.2.8. You will prove that DP 6 0 in
Exercise 4.2.5.

In the notation of Theorem 4.2.8, mv is the vertex v. Thus the Cartier data of
the Cartier divisor DP is the set
(4.2.8)
{mv }v P (n) = {v | v is a vertex of P}.
This is very satisfying and explains why the minus signs in (4.2.6) correspond to
the minus signs in Theorem 4.2.8.
The divisor class [DP ] Pic(XP ) also has a nice interpretation. If D DP , then
D = DP + div( m ) for some m M. In Proposition 2.3.9 we saw that P and its
translate P m have the same normal fan and hence give the same toric variety,
i.e., XP = Xm+P . We also have
D = DP + div( m ) = DPm
(Exercise 4.2.5), so that the divisor class of DP gives all translates of P.
The divisor DP has many more wonderful properties. We will get a glimpse
of this in 4.3 and learn the full power of DP in Chapter 6 when we study ample
divisors on toric varieties.
Support Functions. The Cartier data {m } that describes a torus-invariant
Cartier divisor can be cumbersome to work with. Here weSintroduce a more efficient computational tool. Recall that has support || = NR .
Definition 4.2.11. Let be a fan in NR .
(a) A support function is a function : || R that is linear on each cone of .

The set of all support functions is denoted SF().
(b) A support function is integral with respect to the lattice N if
(|| N) Z.
The set of all such support functions is denoted SF(, N).
P
Let D = a D be Cartier and let {m } be the Cartier data of D as in
Theorem 4.2.8. Thus
(4.2.9)
hm , u i = a for all (1).
We now describe Cartier divisors in terms of support functions.
184
Theorem 4.2.12. Let be a fan in NR . Then:

P
(a) Given D = a D with Cartier data {m } , the function
D : || R
u 7 D (u) = hm , ui when u
is a well-defined support function that is integral with respect to N.

(b) D (u ) = a for all (1), so that
X
D=
D (u ) D .
(c) The map D 7 D induces an isomorphism
CDivTN (X ) SF(, N).
Proof. Theorem 4.2.8 tells us that each m is unique modulo M and that
m m mod ( ) M. It follows easily that D is well-defined. Also, D
is linear on each since D | (u) = hm , ui for u , and it is integral with respect
to N since m M. This proves part (a), and part (b) follows from the definition of
D and (4.2.9).
It remains to prove part (c). First note that D SF(, N) by part (a). Since
D, E CDivTN (X ) and k Z imply that
D+E = D + E
k D = k D ,
the map CDivTN (X ) SF(, N) is a homomorphism, and injectivity follows from
part (b). To prove surjectivity, take SF(, N). Fix . Since is integral
with respect to N, it defines a N-linear map |N : N Z, which extends to
N-linear map : N Z, where N = Span() N. Since
HomZ (N , Z) M/M(),
it follows
that there is m M such that | (u) = hm , ui for u . Then D =
P

D (u ) D is a Cartier divisor that maps to .
In terms of support functions, the exact sequence of Theorem 4.2.1 becomes
(4.2.10)
M SF(, N) Pic(X ) 0,
where m M maps to the linear support P

function defined by u 7 hm, ui and
SF(, N) maps to the divisor class [ (u ) D ] Pic(X ). Be sure you
understand the minus signs.
Here is an example of how to compute with support functions.
Example 4.2.13. The eight points e1 e2 e3 are the vertices of a cube in R3 .

Taking the cones over the six faces gives a complete fan in R3 . Modify this fan by
replacing e1 + e2 + e3 with e1 + 2e2 + 3e3 . The resulting fan has the surprising
185
property that Pic(X ) = 0. In other words, X is a complete toric variety whose

Cartier divisors are all principal.
We will prove Pic(X ) = 0 by showing that all support functions for are
linear. Label the ray generators as follows, using coordinates for compactness:
u1 = (1, 2, 3), u2 = (1, 1, 1), u3 = (1, 1, 1), u4 = (1, 1, 1)
u5 = (1, 1, 1), u6 = (1, 1, 1), u7 = (1, 1, 1), u8 = (1, 1, 1).

The ray generators are shown in Figure 4. The figure also includes three maximal
cones of :
1 = Cone(u1 , u2 , u3 , u5 )
2 = Cone(u1 , u3 , u4 , u7 )
3 = Cone(u1 , u2 , u4 , u6 ).
The shading in Figure 4 indicates 1 2 , 1 3 , 2 3 . Besides 1 , 2 , 3 , the
fan has three other maximal cones, which we call left, down, and back. Thus
the cone left has ray generators u2 , u5 , u6 , u8 , and similarly for the other two.
u1
u4
u6
u2
u8
u7
1
u5
u3
Figure 4. A fan with Pic(X ) = 0
Take SF(, Z3 ). We show that is linear as follows. Since |1 is linear,

there is m1 Z3 such that (u) = hm1 , ui for u 1 . Hence the support function
u 7 (u) hm1 , ui
vanishes identically on 1 . Replacing with this support function, we may assume
that |1 = 0. Once we prove = 0 everywhere, it will follow that all support
functions are linear, and then Pic(X ) = 0 by (4.2.10).
Since u1 , u2 , u3 , u5 1 and vanishes on 1 , we have (u1 ) = (u2 ) =
(u3 ) = (u5 ) = 0. It suffices to prove (u4 ) = (u6 ) = (u7 ) = (u8 ) = 0.
186
To do this, we use the fact that each maximal cone has four generators, which
must satisfy a linear relation. Here are the cones and the corresponding relations:
cone
relation
1
2u1 + 5u5 = 4u2 + 3u3
2u1 + 4u7 = 3u3 + 5u4
2
3
2u1 + 3u6 = 4u2 + 5u4
left
u2 + u8 = u5 + u6
u3 + u8 = u5 + u7
down
back
u4 + u8 = u6 + u7
Since is linear on each cone and (u1 ) = (u2 ) = (u3 ) = (u5 ) = 0, the second,
third, fourth and fifth relations imply
4(u7 ) = 5(u4 )
3(u6 ) = 5(u4 )
(u8 ) = (u6 )
(u8 ) = (u7 ).
The last two equations give (u6 ) = (u7 ), and substituting these into the first two
shows that (u4 ) = (u6 ) = (u7 ) = (u8 ) = 0.
Since the toric variety of a polytope P has the non-principal Cartier divisor
DP , its follows that the fan of Example 4.2.13 is not the normal fan of any 3dimensional lattice polytope. As we will see later, this implies that X is complete
but not projective.
A full dimensional lattice polytope P MR leads to an interesting support
function on the normal fan P .
Proposition 4.2.14. Assume P MR is a full dimensional lattice polytope with
normal fan P . Then the function P : NR R defined by
P (u) = min(hm, ui | m P)
has the following properties:
(a) P is a support function for P and is integral with respect to N.
(b) The divisor corresponding to P is the divisor DP defined in (4.2.7).
Proof. First note that minimum used in the definition of P exists because P is
compact. Now write
P
P = {m MR | hm, uF i aF for all facets F of P}.
Then DP = F aF DF is Cartier by Proposition 4.2.10, and Theorem 4.2.12 shows

that the corresponding support function maps uF to aF .
187
It remains to show that P (u) SF(P ) and P (uF ) = aF . Recall that maximal cones of P correspond to vertices ofP
P, where the vertex v gives the maximal
cone v = Cone(uF | v F). Take u = vF F uF v , where F 0. Then
m P implies
X
X
(4.2.11)
hm, ui =
F hm, uF i
F aF .
vF
Thus P (u)
obtain
vF F aF .
vF
Since equality occurs in (4.2.11) when m = v, we
P (u) =
X
vF
F aF = hv, ui.
This shows that P SF(P , N). Furthermore, when v F, we have P (uF ) =

hv, uF i = aF , as desired.

We will return to support functions in Chapter 6, where we will use them to
give elegant criteria for a divisor to be ample or generated by its global sections.
Exercises for 4.2.
4.2.1. Prove the assertions made in Example 4.2.4.
4.2.2. Prove (4.2.4) and Proposition 4.2.7.
P
4.2.3. When is complete, prove that D = a D is Cartier if and only if it satisfies
condition (d) stated in the discussion following Theorem 4.2.8.
4.2.5. A lattice polytope P gives the toric variety XP and the divisor DP from (4.2.7).
(a) Prove that DP + div( m ) = DPm for any m M.
(b) Prove that DP 6 0. Hint: The normal fan of P is complete.
4.2.6. Let D be a TN -invariant Cartier divisor on X . By Theorem 4.2.8, D is determined

by its Cartier data {m } . Given any m M, show that D + div( m ) has Cartier data
{m m} . Be sure to explain where the minus sign comes from.
4.2.7. Let X be the toric variety of the fan . Prove the following consequences of the
Orbit-Cone Correspondence (Theorem 3.2.6).
T
(a) O() = (1) D .
(b) Rays u1 , . . . , ur (1) lie in a cone of if and only if D1 Dr 6= .
4.2.8. Let be a fan in NR Rn and assume that has a cone of dimension n.

(a) Fix a cone of dimension n. Prove that
Pic(X ) { SF(, N) | | = 0}.
(b) Explain how part (a) relates to Example 4.2.13.
(c) Use part (a) to give a different proof of Proposition 4.2.5.
188
4.2.9. Let be as in Example 4.2.4, but instead of using the lattice generated by e1 , e2 , e3 ,
1
1
e1 + Z 1b e2 + Z a1 e3 + Z 2b
(e1 + e2 + e3 ), where a, b are relatively
instead use N = Z 2b
prime positive integers with a > 1. Prove that no multiple of D1 + D2 + D3 + D4 is Cartier.
Hint: The first step will be to find the minimal generators (relative to N) of the edges of .
4.2.10. Let XP be the toric variety of the octahedron P = Conv(e1 , e2 , e3 ) R3 .
(a) Show that Cl(XP ) Z5 (Z/2Z)2 .
(b) Use support functions and the strategy of Example 4.2.13 to show that Pic(XP ) Z.
4.2.11. In Exercise 4.1.5, you showed that the weighted projective space P(q0 , . . . , qn ) has
class group Cl(P(q0 , . . . , qn )) Z. Prove that Pic(P(q0 , . . . , qn )) Cl(P(q
Pn 0 , . . . , qn )) maps
to the subgroup mZ Z,P
where m = lcm(q0 , . . . , qn ). Hint: Show that i=0 bi Di generates
n
the class group, where i=0 bi qi = 1. Also note that m MQ lies in M if and only if
hm, ui i Z for all i, where the ui are from Exercise 4.1.5.
4.2.12. Let X be a smooth toric variety and let have dimension 2. This gives the
orbit closure V ( ) = O( ) X . In 3.3 we defined the blowup BlV ( ) (X ). Prove that
Pic(BlV ( ) (X )) Pic(X ) Z.
P
4.2.13. A nonzero polynomial f = mZn cm x m C[x1 , . . . , xn ] has Newton polytope
P( f ) = Conv(m | cm 6= 0) Rn .
When P( f ) has dimension n, Proposition 4.2.14 tells us that the function P( f ) (u) =
min(hm, ui | m P( f )) is the support function of a divisor on XP( f ) . Here we interpret
P( f ) as the tropicalization of f .
The tropical semiring (R, , ) has operations
a b = min(a, b) (tropical addition)

ab = a+b
(tropical multiplication).
A tropical polynomial in real variables x1 , . . . , xn is a finite tropical sum

a
F = c1 x11,1 xn1,n cr x1r,1 xnr,n
where ci R and xia = xi xi (a times). For a more compact representation, define

a tropical monomial to be x m = x1a1 xnan for m = (a1 , . . . , an ) Nn . Then, using the
tropical analog of summation notation, the tropical polynomial F is
L
F = ri=1 ci x mi , mi = (ai,1 , . . . , ai,n ).
(a) Show that F = min1ir (ci + ai,1x1 + + ai,n xn ).
(b) The tropicalization of our original polynomial f is the tropical polynomial

L
Ff = cm 6=0 0 x m.
Prove that Ff = P( f ) . (The 0 is explained as follows. In tropical geometry, one

often works in a larger ring where the coefficients of f are Puiseux series, and the
tropicalization uses the order of vanishing of the coefficients. Here, we use a smaller
ring where the coefficients of f are nonzero constants, with order of vanishing 0.)
(c) The tropical variety of a tropical polynomial F is the set of points in Rn where F is
not linear. For f = x + 2y + 3x 2 xy 2 + 4x 2 y, compute the tropical variety of Ff and
show that it consists of the rays in the normal fan of P( f ).
A nice introduction to tropical algebraic geometry can be found in [240].
4.3. The Sheaf of a Torus-Invariant Divisor
189

P
If D = a D is a TN -invariant divisor on the normal toric variety X , we get the
sheaf OX (D) defined in 4.0. We will study these sheaves in detail in Chapters 6
and 8. In this section we will focus primarily on global sections.
We begin with a classic example of the sheaf OX (D).
Example 4.3.1. For P n , the divisors D0 , . . . , Dn correspond to the ray generators of
the usual fan for P n . The computation Cl(P n ) Z from Example 4.1.6 shows that
D0 D1 Dn . These linear equivalences give isomorphisms
OPn (D0 ) OPn (D1 ) OPn (Dn )
by Proposition 4.0.29. These sheaves are denoted OPn (1), and simiarly the sheaves
OPn (k Di ), k Z, are denoted OPn (k). We will see the intrinsic reason for this
notation in Example 5.3.8.
Global Sections. Let D be a TN -invariant divisor on a toric variety X . We will

give two descriptions of the global sections (X , OX (D)). Here is the first.
Proposition 4.3.2. If D is a TN -invariant Weil divisor on X , then
M
(X , OX (D)) =
C m.
div( m )+D0
Proof. If f (X , OX (D)), then div( f ) + D 0 implies div( f )|TN 0 since

D|TN = 0. Since C[M] is the coordinate ring of TN , Proposition 4.0.16 implies
f C[M]. Thus
(X , OX (D)) C[M].
Furthermore, (X , OX (D)) is invariant under the TN -action on C[M] since D is

TN -invariant. By Lemma 1.1.16, we obtain
M
(X , OX (D)) =
C m.
m (X ,OX (D))
Since m (X , OX (D)) if and only if div( m ) + D 0, we are done.

P
The Polyhedron of a Divisor. For D = a D and m M, div( m ) + D 0 is
equivalent to
hm, u i + a 0
for all (1),
which can be rewritten as

(4.3.1)
hm, u i a
for all (1).
This explains the minus signs! To emphasize the underlying geometry, we define
(4.3.2)
PD = {m MR | hm, u i a for all (1)}.
190
We say that PD is a polyhedron since it is an intersection of finitely many closed

half spaces. This looks very similar to the canonical presentation of a polytope
(see (4.2.6), for example). However, the reader should be aware that PD need not
be a polytope, and even when it is a polytope, it need not be a lattice polytope. All
of this will be explained in the examples given below.
For now, we simply note that (4.3.1) is equivalent to m PD M. This gives
our second description of the global sections.
Proposition 4.3.3. If D is a TN -invariant Weil divisor on X , then
M
(X , OX (D)) =
C m,
mPD M
where PD MR is the polyhedron defined in (4.3.2).
As noted above, a polyhedron is an intersection of finitely many closed half

spaces. A polytope is a bounded polyhedron.
Examples. Here are some examples to illustrate the kinds of polyhedra that can
occur in Proposition 4.3.3.
Example 4.3.4. The fan for the blowup Bl0 (C2 ) of C2 at the origin has ray
generators u0 = e1 + e2 , u1 = e1 , u2 = e2 and corresponding divisors D0 , D1 , D2 .
For the divisor D = D0 + D1 + D2 , a point m = (x, y) lies in PD if and only if
hm, u0 i 1 x + y 1
hm, u1 i 1 x 1
hm, u2 i 1 y 1.
u2
e2
u0
u1
PD
e1
Figure 5. The fan and the polyhedron PD
The fan and the polyhedron PD are shown in Figure 5. Note that PD is not
bounded. By Proposition 4.3.3, the lattice points of PD (the dots in Figure 5) give
characters that form a basis of (Bl0 (C2 ), OBl0 (C2 ) (D)).
191
Example 4.3.5. The fan 2 for the Hirzebruch surface H2 has ray generators
u1 = e1 + 2e2 , u2 = e2 , u3 = e1 , u4 = e2 . The corresponding divisors are
D1 , D2 , D3 , D4 , and Example 4.1.8 implies that the classes of D1 and D2 are a
basis of Cl(H2 ) Z2 .
Consider the divisor a D1 + D2 , a Z, and let Pa R2 be the corresponding
polyhedron, which is a polytope in this case. A point m = (x, y) lies in Pa if and
only if
hm, u1 i a y 21 x 2a
hm, u2 i 1 y 1
hm, u3 i 0 x 0.
hm, u4 i 0 y 0.
Figure 6 shows 2 , together with shaded areas marked A, B, C. These are related
u1
u2
a=1
a=2
a=3
A
B
u4
u3
e2
e1 e2
Figure 6. The fan 2 and the polyhedra Pa
to the polygons Pa for a = 1, 2, 3 by the equations

P1 = A
P2 = A B
P3 = A B C.
Notice that as we increase a, the line y = 21 x 2a corresponding to u1 moves to the

right and makes the polytope bigger. In fact, you can see that 2 is the normal fan
of the lattice polytope Pa for any a 3. For a = 2, we get a lattice polytope P2 , but
its normal fan is not 2 you can see how the facet with inward normal vector
u2 collapses to a point of P2 . For a = 1, P1 is not a lattice polytope since 21 e2 is a
vertex.
Chapters 6 and 7 will explain how the geometry of the polyhedron PD relates to
the properties of the divisor D. In particular, we will see that the divisor a D1 + D2
from Example 4.3.5 is ample if and only if a 3 since these are the only as for
which 2 is the normal fan Pa .
192
Example 4.3.6. By Example 4.3.1, the sheaf OPn (k) can be written OPn (k D0 ),
where the divisor D0 corresponds to the ray generator u0 from Example 4.1.6. It is
straightforward to show that the polyhedron of D = k D0 is
(
k<0
PD =
k n k 0,
where n Rn is the standard n-simplex. We can think of characters as Laurent

monomials t m = t1a1 tnan , where m = (a1 , . . . , an ). It follows that
(P n , OPn (k)) { f C[t1 , . . . ,tn ] | deg( f ) k}.
The homogenization of such a polynomial is

F = x0k f (x1 /x0 , . . . , xn /x0 ) C[x0 , . . . , xn ].
In this way, we get an isomorphism
(P n , OPn (k)) { f C[x0 , . . . , xn ] | f is homogeneous with deg( f ) = k}.
The toric interpretation of homogenization will be discussed in Chapter 5.
Example 4.3.7. Let XP be the toric variety of a full dimensional lattice polytope
P MR . The facet presentation of P gives the Cartier divisor DP defined in (4.2.7),
and one checks easily that the polyhedron PDP is the polytope P that we began with
(Exercise 4.3.1). It follows from Proposition 4.3.3 that
M
(XP , OXP (DP )) =
C m.
mPM
Recall from Chapter 2 that the characters m for m P M give the projective
toric variety XPM . The divisor k DP gives the polytope kP (Exercise 4.3.2), so that
M
(XP , OXP (k DP )) =
C m.
m(kP)M
In Chapter 2 we proved that kP is very ample for k sufficiently large, in which case
X(kP)M is the toric variety XP . So the characters m that realize XP as a projective
variety come from global sections of OXP (k DP ). In Chapter 6, we will pursue these
ideas when we study ample and very ample Cartier divisors.
Note also that dim (XP , OXP (k DP )) gives number of lattice points in multiples
of P. This will have important consequences in later chapters.
The operation sending a TN -invariant Weil divisor D X to the polyhedron

PD MR defined in (4.3.2) has the following properties:
Pk D = kPD for k 0.
PD+div( m ) = PD m.
PD + PE PD+E .
You will prove these in Exercise 4.3.2. The multiple kPD and Minkowski sum
PD + PE are defined in 2.2, and P m is translation.
193
Complete Fans. When the fan is complete, we have the following finiteness
result that you will prove in Exercise 4.3.3.
Proposition 4.3.8. Let X be the toric variety of a complete fan in NR . Then:
(a) (X , OX ) = C, so the only morphisms X C are the constant ones.
(b) PD is a polytope for any TN -invariant Weil divisor D on X .
(c) (X , OX (D)) has finite dimension as a vector space over C for any Weil
divisor on X .
The assertions of parts (a) and (c) are true in greater generality: if X is any
complete variety, then (X , OX ) = C, and if F is any coherent sheaf on X , then
dim (X , F ) < (see [245, Vol. 2, VI.1.1 and VI.3.4]).
Exercises for 4.3.
4.3.1. Prove the assertion PDP = P made in Example 4.3.7.
4.3.2. Prove the properties of D 7 PD listed above.
4.3.3. Prove Proposition 4.3.8. Hint: For part (a), use completeness to show that m = 0
when hm, u i 0 for all . For part (b), assume MR = Rn and suppose mi PD satisfy
mi
||mi || . Then consider the points ||m
on the sphere Sn1 Rn .
i ||
4.3.4. Let be a fan in NR with convex support. Then || NR is a convex polyhedral
cone with dual || MR .
(a) Prove that || is the polyhedron associated to the divisor D = 0 on X .
L
(b) Conclude that (X , OX ) = m|| M C m .
(c) Use part (b) to prove part (a) of Proposition 4.3.8.
4.3.5. Example 4.3.5 studied divisors on the Hirzebruch surface H2 . This exercise will
consider the divisors D = D4 and D = D + D2 = D2 + D4 .
(a) Show that D gives the same polygon as D , i.e., PD = PD .
(b) Since H2 is smooth, D and D are Cartier. Compute their respective Cartier data
{m }2 (2) and {m }2 (2) .
(c) Show that P = Conv(m | 2 (2)) and that P 6= Conv(m | 2 (2)).

Thus D and D give the same polygon but differ in how their Cartier data relates to the
polygon. In Chapter 6 we will use this to prove that OH2 (D) is generated by global sections
while OH2 (D ) has base points.
Chapter 5
Homogeneous Coordinates
on Toric Varieties
5.0. Background: Quotients in Algebraic Geometry

Projective space P n is usually defined as the quotient
P n = (Cn+1 \ {0})/C ,
where C acts on Cn+1 by scalar multiplication, i.e.,
(a0 , . . . , an ) = (a0 , . . . , an ).
The above representation defines P n as a set; making P n into a variety requires
the notion of abstract variety introduced in Chapter 3. The main goal of this chapter
is to prove that every toric variety has a similar quotient construction as a variety.
Group Actions. Let G be a group acting on a variety X . We always assume that
for every g G, the map g (x) = g x defines a morphism g : X X . When
X = Spec(R) is affine, g : X X comes from a homomorphism g : R R. We
define the induced action of G on R by
g f = g1 ( f )
for f R. In other words, (g f )(x) = f (g1 x) for all x X . You will check in
Exercise 5.0.1 this gives an action of G on R. Thus we have two objects:
The set G-orbits X /G = {G x | x X }.
The ring of invariants RG = { f R | g f = f for all g G}.
To make X /G into an affine variety, we need to define its coordinate ring, i.e., we
need to determine the polynomial functions on X /G. A key observation is that if
195
Chapter 5. Homogeneous Coordinates on Toric Varieties
196
f RG , then
f(G x) = f (x)
gives a well-defined function f : X /G C. Hence elements of G give obvious
polynomial functions on X /G, which suggests that
as an affine variety, X /G = Spec(RG ).
As shown by the following examples, this works in some cases but fails in others.
Example 5.0.1. Let 2 = {1} act on C2 = Spec(C[s,t]), where 1 2 acts
by multiplication by 1. Note that every orbit consists of two elements, with the
exception of the orbit of the origin, which is the unique fixed point of the action.
The ring of invariants C[s,t]2 = C[s 2 , st,t 2 ] is the coordinate ring of the affine
toric variety V(xz y 2 ). Hence we get a map
: C2 /2 Spec(C[s,t]2 ) = V(xz y 2 ) C3
where the orbit 2 (a, b) maps to (a 2 , ab, b 2 ). This is easily seen to be a bijection,
so that Spec(C[s,t]2 ) is the perfect way to make C2 /2 into an affine variety.
This is actually an example of the toric morphism induced by changing the
latticesee Examples 1.3.17 and 1.3.19.
Example 5.0.2. Let C act on C4 = Spec(C[x1 , x2 , x3 , x4 ]), where C acts via

(a1 , a2 , a3 , a4 ) = (a1 , a2 , 1 a3 , 1 a4 ).
In this case, the ring of invariants is
C[x1 , x2 , x3 , x4 ]C = C[x1 x3 , x2 x4 , x1 x4 , x2 x3 ],
which gives the map
: C4 /C Spec(C[x1 , x2 , x3 , x4 ]C ) = V(xy zw) C4
where the orbit C (a1 , a2 , a3 , a4 ) maps to (a1 a3 , a2 a4 , a1 a4 , a2 a3 ). Then we have

(Exercise 5.0.2):
is surjective.
If p V(xy zw) \ {0}, then 1 (p) consists of a single C -orbit which is
closed in C4 .
1 (0) consists of all C -orbits contained in C2 {(0, 0)} {(0, 0)} C2 .
Thus 1 (0) consists of infinitely many C -orbits.
This looks bad until we notice one further fact (Exercise 5.0.2):
The fixed point 0 C4 gives the unique closed orbit mapping to 0 under .
If (a, b) 6= (0, 0), then an example of a non-closed orbit is given by

C (a, b, 0, 0) = {(a, b, 0, 0) | C }
since lim0 (a, b, 0, 0) = 0. However, restricting to closed orbits gives

{closed C -orbits} V(xy zw).
We will see that this is the best we can do for this group action.
197
Example 5.0.3. Let C act on Cn+1 = Spec(C[x0 , . . . , xn ]) by scalar multiplication.

Then the ring of invariants consists of polynomials satsifying
f (x0 , . . . , xn ) = f (x0 , . . . , xn )
for all C . Such polynomials must be constant, so that
C[x0 , . . . , xn ]C = C.
It follows that the quotient is Spec(C), which is just a point. The reason for this
is that the only closed orbit is the orbit of the fixed point 0 Cn+1 .
Examples 5.0.2 and 5.0.3 show what happens when there are not enough invariant functions to separate G-orbits.
The Ring of Invariants. When G acts on an affine variety X = Spec(R), a natural
question concerns the structure of the ring of invariants. The coordinate ring R is a
finitely generated C-algebra without nilpotents. Is the same true for RG ? It clearly
has no nilpotents since RG R. But is RG finitely generated as a C-algebra? This
is related to Hilberts Fourteenth Problem, which was settled by a famous example
of Nagata that RG need not be a finitely generated C-algebra! An exposition of
Hilberts problem and Nagatas example can be found in [83, Ch. 4].
If we assume that RG is finitely generated, then Spec(RG ) is an affine variety
that is the best candidate for a quotient in the following sense.
Lemma 5.0.4. Let G act on X = Spec(R) such that RG is a finitely generated Calgebra, and let : X Y = Spec(RG ) be the morphism of affine varieties induced
by the inclusion RG R. Then:
(a) Given any diagram
X?
?
??
??
?
/
?Z
Y
where is a morphism of affine varieties such that (g x) = (x) for g G
and x X , there is a unique morphism making the diagram commute, i.e.,
= .
(b) If X is irreducible, then Y is irreducible.

(c) If X is normal, then Y is normal.
Proof. Suppose that Z = Spec(S) and that is induced by : S R. Then

(S) RG follows easily from (g x) = (x) for g G, x X . Thus factors uniquely as
S RG R.
198
The induced map : Y Z clearly has the desired properties.
Part (b) is immediate since RG is a subring of R. For part (c), let K be the field
of fractions of RG . If a K is integral over RG , then it is also integral over R and
hence lies in R since R is normal. It follows that a R K, which obviously equals
RG since G acts trivially on K. Thus RG is normal.

This shows that Y = Spec(RG ) has some good properties when RG is finitely
generated, but there are still some unanswered questions, such as:
Is : X Y surjective?
Does Y have the right topology? Ideally, we would like U Y to be open if

and only if 1 (U ) X is open. (Exercise 5.0.3 explores how this works for
group actions on topological spaces.)
While Y is the best affine approximation of the quotient X /G, could there be a
non-affine variety that is a better approximation?
We will see that the answers to these questions are all yes once we work with the
correct type of group action.
Good Categorical Quotients. In order to get the best properties of a quotient map,
we consider the general situation where G is a group acting on a variety X and
: X Y is a morphism that is constant on G-orbits. Then we have the following
definition.
Definition 5.0.5. Let G act on X and let : X Y be a morphism that is constant
on G-orbits. Then is a good categorical quotient if:
(a) If U Y is open, then the natural map OY (U ) OX ( 1 (U )) induces an
isomorphism
OY (U ) OX ( 1 (U ))G .
(b) If W X is closed and G-invariant, then (W ) Y is closed.
(c) If W1 ,W2 are closed, disjoint, and G-invariant in X , then (W1 ) and (W2 ) are
disjoint in Y .
We often write a good categorical quotient as : X X //G. Here are some
properties of good categorical quotients.
Theorem 5.0.6. Let : X X //G be a good categorical quotient. Then:
(a) Given any diagram
X D
D
DD
DD
DD
!
/
=Z
X //G
where is a morphism such that (g x) = (x) for g G and x X , there is

a unique morphism making the diagram commute, i.e., = .
199
(b) is surjective.
(c) A subset U X //G is open if and only if 1 (U ) X is open.
(d) If U X //G is open and nonempty, then |1 (U) : 1 (U ) U is a good

categorical quotient.
(e) Given points x, y X , we have
(x) = (y) G x G y 6= .
Proof. The proof of part (a) can be found in [83, Prop. 6.2]. The proofs of the
remaining parts are left to the reader (Exercise 5.0.4).

Algebraic Actions. So far, we have allowed G to be an arbitrary group acting on
X , assuming only that for every g G, the map x 7 g x is a morphism g : X X .
We now make the further assumption that G is an affine variety. To define this
carefully, we first note that the group GLn (C) of n n invertible matrices with
entries in C is the affine variety
2
GLn (C) = {A Cnn = Cn | det(A) 6= 0}.

A subgroup G GLn (C) is an affine algebraic group if it is also a subvariety of
GLn (C). Examples include GLn (C), SLn (C), (C )n , and finite groups.
If G is an affine algebraic group, then the connected component of the identity,
denoted G , has the following properties (see [152, 7.3]):
G is a normal subgroup of finite index in G.
G is an irreducible affine algebraic group.
An affine algebraic group G acts algebraically on a variety X if the G-action

(g, x) 7 g x defines a morphism
G X X.
Examples of algebraic actions include toric varieties since the torus TN X acts
algebraically on X . Examples 5.0.1, 5.0.2 and 5.0.3 are also algebraic actions.
Algebraic actions have the property that G-orbits are constructible sets in X .
This has the following nice consequence for good categorical quotients.
Proposition 5.0.7. Let an affine algebraic group G act algebraically on a variety
X , and assume that a good categorical quotient : X X //G exists. Then:
(a) If p X //G, then 1 (p) contains a unique closed G-orbit.
(b) induces a bijection
{closed G-orbits in X} X //G.

Proof. For part (a), first note that uniqueness follows immediately from part (e) of
Theorem 5.0.6. To prove the existence of a closed orbit in 1 (p), let G G be
200
the connected component of the identity. Then 1 (p) is stable under G , so we

can pick an orbit G x 1 (p) such that G x has minimal dimension.
Note that G x is irreducible since G is irreducible, and since G x is constructible, there is a nonempty Zariski open subset U of G x such that U G x.
If G x is not closed, then G x contains an orbit G y 6= G x. Thus
G y G x \ G x G x \U .
However, G x is irreducible, so that
dim (G x \U ) < dim G x.
Hence G y has strictly smaller dimension, a contradiction. Thus G x is closed.

If g1 , . . . , gt are coset representatives of G/G , then
Gx =
t
[
i=1
gi G x
shows that G x is also closed. This proves part (a) of the proposition, and part (b)
follows immediately from part (a) and the surjectivity of .

For the rest of the section, we will always assume that G is an affine algebraic
group acting algebraically on a variety X .
Geometric Quotients. The best quotients are those where points are orbits. For
good categorical quotients, this condition is captured by requiring that orbits be
closed. Here is the precise result.
Proposition 5.0.8. Let : X X //G be a good categorical quotient. Then the
(a) All G-orbits are closed in X .
(b) Given points x, y X , we have
(x) = (y) x and y lie in the same G-orbit.
(c) induces a bijection
{G-orbits in X} X //G.
(d) The image of the morphism G X X X defined by (g, x) 7 (g x, x) is the
fiber product X X//G X .
Proof. This follows easily from Theorem 5.0.6 and Proposition 5.0.7. We leave

In general, a good categorical quotient is called a geometric quotient if it
satisfies the condtions of Proposition 5.0.8. We write a geometric quotient as
: X X /G since points in X /G correspond bijectively to G-orbits in X .
201
We have yet to give an example of a good categorical or geometric quotient.

For instance, it is not clear that Examples 5.0.1, 5.0.2 and 5.0.3 satisfy Definition 5.0.5. Fortunately, once we restrict to the right kind of algebraic group, examples become abundant.
Reductive Groups. An affine algebraic group G is called reductive if its maximal
connected solvable subgroup is a torus. Examples of reductive groups include finite
groups, tori, and semisimple groups such as SLn (C).
For us, actions by reductive groups have the following key properties.
Proposition 5.0.9. Let G be a reductive group acting algebraically on an affine
variety X = Spec(R). Then:
(a) RG is a finitely generated C-algebra.
(b) The morphism : X Spec(RG ) induced by RG R is a good categorical
quotient.
Proof. See [83, Prop. 3.1] for part (a) and [83, Thm. 6.1] for part (b).
In the situation of Proposition 5.0.9, we can write Spec(R)//G = Spec(RG ).

Examples 5.0.1, 5.0.2 and 5.0.3 involve reductive groups acting on affine varieties.
Hence these are good categorical quotients that have all of the properties listed in
Theorem 5.0.6 and Proposition 5.0.7. Furthermore, Example 5.0.1 (the action of
2 on C2 ) is a geometric quotient. This last example generalizes as follows.
Example 5.0.10. Given a strongly convex rational polyhedral cone NR and
a sublattice N N of finite index, part (b) of Proposition 1.3.18 implies that the
finite group G = N/N acts on U,N such that the induced map on coordinate rings
is given by
C[ M]
C[ M ]G C[ M ].
It follows that : U,N U,N is a good categorical quotient. In fact, is a
geometric quotient since the G-orbits are finite and hence closed. This completes
the proof of part (c) of Proposition 1.3.18.
Almost Geometric Quotients. Let us examine Examples 5.0.2 and 5.0.3 more
closely. As noted above, both give good categorical quotients. However:
(Example 5.0.3) Here we have the quotient
Cn+1 //C = Spec(C[x0 , . . . , xn ]C ) = Spec(C) = {pt}.
So the good categorical quotient Cn+1 Cn+1 //C = {pt} is really bad.
(Example 5.0.2) In this case, the quotient is
: C4 C4 //C = V(xy zw).
Let U = V(xy zw) \ {0} and U0 = 1 (U ). Then |U0 : U0 U is a good categorical quotient by Theorem 5.0.6, and by Example 5.0.2, orbits of elements
202
in U0 are closed in C4 . Then |U0 is a geometric quotient by Proposition 5.0.8,

so that C4 //C = V(xy zw) is a geometric quotient outside of the origin.
The difference between these two examples is that the second is very close to being
a geometric quotient. Here is a result that describes this phenomenon in general.
Proposition 5.0.11. Let : X X //G be a good categorical quotient. Then the

(a) X has a G-invariant Zariski dense open subset U0 such that G x is closed in X
for all x U0 .
(b) X //G has a Zariski dense open subset U such that |1 (U) : 1 (U ) U is
a geometric quotient.
Proof. Given U0 satisfying (a), then W = X \ U0 is closed and G-invariant. For

x U0 , the orbit G x U0 is also closed and G-invariant. These are disjoint,
which implies (x)
/ (W ) since is a good categorical quotient. Since is onto,
we see that X //G = (U0 ) (W ) is a disjoint union. If we set U = (U0 ), then
U0 = 1 (U ). Note also that U is open since (W ) is closed and Zariski dense
since U0 is Zariski dense in X . Then |U0 : U0 U is a good categorical quotient
by Theorem 5.0.6, and by assumption, orbits of elements in U0 are closed in C4
and hence in U0 . It follows that |U0 is a geometric quotient by Proposition 5.0.8.
The proof going the other way is similar and is omitted (Exercise 5.0.6).
In general, a good categorical quotient is called an almost geometric quotient

if it satisfies the conditions of Proposition 5.0.11. Example 5.0.2 is an almost
geometric quotient while Example 5.0.3 is not.
Constructing Quotients. Now that we can handle affine quotients in the reductive
case, the next step is to handle more general quotients. Here is a useful result.
Proposition 5.0.12. Let G act on X and let : X Y beSa morphism of varieties
that is constant on G-orbits. If Y has an open cover Y = V such that
|1 (V ) : 1 (V ) V
is a good categorical quotient for every , then : X Y is a good categorical

quotient.
Proof. The key point is that the properties listed in Definition 5.0.5 can be checked
locally. We leave the details to the reader (Exercise 5.0.7).

Example 5.0.13. Consider a lattice N and a sublattice N N of finite index, and
let be a fan in NR = NR . This gives a toric morphism
: X,N X,N .
203
By Proposition 1.3.18, the finite group G = N/N is the kernel of TN TN , so that

G acts on X,N . Since
1 (U,N ) = U,N
for , Example 5.0.10 and Proposition 5.0.12 imply that is a geometric
quotient. This strengthens the result proved in Proposition 3.3.7.
It is sometimes possible to construct the quotient of X by G by taking rings

of invariants for a suitable affine open cover. If the local quotients patch together
to form a separated variety Y , then the resulting morphism : X Y is a good
categorical quotient by Proposition 5.0.12. Here are two examples that illustrate
this strategy.
Example 5.0.14. Let C act on C2 \ {0} by scalar multiplication, where C2 =
Spec(C[x0 , x1 ]). Then C2 \ {0} = U0 U1 , where
U0 = C2 \ V(x0 ) = Spec(C[x01 , x1 ])
U1 = C2 \ V(x1 ) = Spec(C[x0 , x11 ])
U0 U1 = C2 \ V(x0 x1 ) = Spec(C[x01 , x11 ]).

The rings of invariants are
C[x01 , x1 ]C = C[x1 /x0 ]
C[x0 , x11 ]C = C[x0 /x1 ]
C
= C[(x1 /x0 )1 ].
C[x01 , x1
1 ]
It follows that the Vi = Ui //C glue together in the usual way to create P1 . Since
C -orbits are closed in C2 \ {0}, it follows that
P1 = (C2 \ {0})/C
is a geometric quotient.
This example generalizes to show that
P n = (Cn+1 \ {0})/C
is a good geometric quotient when C acts on Cn+1 by scalar multiplication. At

the beginning of the section, we wrote this quotient as a set-theoretic construction.
It is now an algebro-geometric construction.
Our second example shows the importance of being separated.
Example 5.0.15. Let C act on C2 \ {0} by (a, b) = (a, 1 b). Then C2 \ {0} =
U0 U1 , where U0 , U1 and U0 U1 are as in Example 5.0.14. Here, the rings of
204
invariants are
C[x01 , x1 ]C = C[x0 x1 ]
C[x0 , x11 ]C = C[x0 x1 ]
C[x01 , x11 ]C = C[(x0 x1 )1 ].

Gluing together Vi = Ui //C along U0 U1 //C gives the variety obtained from
two copies of C by identifying all nonzero points. This is the non-separated variety
constructed in Example 3.0.15.
In Exercise 5.0.8 you will draw a picture of the C -orbits that explains why the
quotient cannot be separated in this example.
In this book, we usually use the word variety to mean separated variety.
For example, when we say that : X Y is a good categorical or geometric quotient, we always assume that X and Y are separated. So Example 5.0.15 is not a
good categorical quotient. In algebraic geometry, most operations on varieties preserve separatedness. Quotient constructions are one of the few exceptions where
care has to be taken to check that the resulting variety is separated.
Exercises for 5.0.
5.0.1. Let G act on an affine variety X = Spec(R) such that g (x) = g x is a morphism for
all g G.
(a) Show that g f = g1 ( f ) defines an action of G on R. Be sure you understand why
the inverse is necessary.
(b) The evaluation map R X C is defined by ( f , x) 7 f (x). Show that this map is
invariant under the action of G on R X given by g ( f , x) = (g f , g x).
5.0.2. Prove the claims made in Example 5.0.2.

5.0.3. Let G be a group acting on a Hausdorff topological space, and let X/G be the set of
G-orbits. Define : X X/G by (x) = G x. The quotient topology on X/G is defined
by saying that U X/G is open if and only if 1 (U) X is open.
(a) Prove that if X/G is Hausdorff, then the G-orbits are closed subsets of X.
(b) Prove that if W X is closed and G-invariant, then (W ) X/G is closed.
(c) Prove that if W1 ,W2 are closed, disjoint, and G-invariant in X, then (W1 ) and (W2 )
are disjoint in X/G.
5.0.4. Prove parts (b), (c), (d) and (e) of Theorem 5.0.6. Hint for part (b): Part (a) of Definition 5.0.5 implies that OX //G (U) injects into OX ( 1 (U)) for all open sets U X//G.
Use this to prove that (X) is Zariski dense in X//G. Then use part (b) of Definition 5.0.5.
5.0.6. Complete the proof of Proposition 5.0.11.
5.0.8. Consider the C action on C2 \ {0} described in Example 5.0.15.
5.1. Quotient Constructions of Toric Varieties
205
(a) Show that with two exceptions, the C -orbits are the hyperbolas x1 x2 = a, a 6= 0. Also
describe the two remaining C -orbits.
(b) Give an intuitive explanation, with picture, to show that the limit of the orbits x1 x2 =
a as a 0 consists of two distinct orbits.
(c) Explain how part (b) relates to the non-separated quotient constructed in the example.
5.0.9. Give an example of a reductive G-action on an affine variety X such that X has
a nonempty G-invariant affine open set U X with the property that the induced map
U//G X//G is not an inclusion.
5.0.10. Let a finite group G act on X. Then a good categorical quotient : X X//G
exists since finite groups are reductive. Explain why is a geometric quotient.

Let X be the toric variety of a fan in NR . The goal of this section is to construct
X as an almost geometric quotient
X (Cr \ Z)//G
for an appropriate choice of affine space Cr , exceptional set Z Cr , and reductive

group G. We will use our standard notation, where each (1) gives a minimal
generator u N and a TN -invariant prime divisor D X .
No Torus Factors. Toric varieties with no torus factors have the nicest quotient
constructions. Recall from Proposition 3.3.9 that X has no torus factors when NR
is spanned by u , (1), and when this happens, Theorem 4.1.3 gives the short
exact sequence
L
0 M Z D Cl(X ) 0,
P
where m M maps to div( m ) = hm, u i D and Cl(X ) is the class group
L P
defined in 4.0. We use the convention that in expressions such as , and
Q
, the index ranges over all (1).
We write the above sequence more compactly as
(5.1.1)
0 M Z(1) Cl(X ) 0.
Applying HomZ (, C ) gives
1 HomZ (Cl(X ), C ) HomZ (Z(1) , C ) HomZ (M, C ) 1,
which remains a short exact sequence since HomZ (, C ) is left exact and C is
divisible. We have natural isomorphisms
HomZ (Z(1) , C ) (C )(1)
HomZ (M, C ) TN ,
and we define the group G by

G = HomZ (Cl(X ), C ).
206
This gives the short exact sequence of affine algebraic groups

(5.1.2)
1 G (C )(1) TN 1.
The Group G. The group G defined above will appear in the quotient construction
of the toric variety X . For the time being, we assume that X has no torus factors.
The following result describes the structure of G and gives explicit equations
for G as a subgroup of the torus (C )(1) .
Lemma 5.1.1. Let G (C )(1) be as in (5.1.2). Then:
(a) Cl(X ) is the character group of G.
(b) G is isomorphic to a product of a torus and a finite abelian group. In particular,

G is reductive.
(c) Given a basis e1 , . . . , en of M, we have

Q hm,u i
G = (t ) (C )(1) | t = 1 for all m M

Q he ,u i
= (t ) (C )(1) | t i = 1 for 1 i n .
Proof. Since Cl(X ) is a finitely generated abelian group, Cl(X ) Z H, where

H is a finite abelian group. Then
G = HomZ (Cl(X ), C ) HomZ (Z H, C ) (C ) HomZ (H, C ).
This proves part (b) since HomZ (H, C ) is a finite abelian group. For part (a), note
that Cl(X ) gives the map that sends g G = HomZ (Cl(X ), C ) to g() C .
Thus elements of Cl(X ) give characters on G, and the above isomorphisms make
it easy to see that all characters of G arise this way.
For part (c), the first description of G follows from (5.1.2) since M Z(1)
is defined by m M 7 (hm, u i) Z(1) , and the second description is an easy
consequence of the first.

P
Example 5.1.2. The ray generators of the fan for P n are u0 = ni=1 ei , u1 =
e1 , . . . , un = en . By Lemma 5.1.1, (t0 , . . . ,tn ) (C )n+1 lies in G if and only if
hm,e1 en i hm,e1 i
t1
t0
tnhm,en i = 1
for all m M = Zn . Taking m equal to e1 , . . . , en , we see that G is defined by

t01 t1 = = t01 tn = 1.
Thus
G = {(, . . . , ) | C } C ,
which is the action of C on Cn+1 given by scalar multiplication.
207
Example 5.1.3. The fan for P1 P1 has ray generators u1 = e1 , u2 = e1 , u3 =

e2 , u4 = e2 in N = Z2 . By Lemma 5.1.1, (t1 ,t2 ,t3 ,t4 ) (C )4 lies in G if and
only if
hm,e i hm,e1 i hm,e2 i hm,e2 i
=1
t4
t3
t1 1 t2
for all m M = Z2 . Taking m equal to e1 , e2 , we obtain
t1t21 = t3t41 = 1.
Thus
G = {(, , , ) | , C } (C )2 .
Example 5.1.4. Let = Cone(de1 e2 , e2 ) R2 , which gives the rational normal

cone Cbd . Example 4.1.4 shows that Cl(Cbd ) Z/dZ, so that
G = HomZ (Z/dZ, C ) d ,
where d C is the group of dth roots of unity. To see how G acts on C2 , one
uses the ray generators u1 = de1 e2 and u2 = e2 to compute that
G = {(, ) | d = 1} d
(Exercise 5.1.1). This shows that G can have torsion.
The Exceptional Set. For the quotient representation of X , we have the group G
and the affine space C(1) . All that is missing is the exceptional set Z C(1) that
we remove from C(1) before taking the quotient by G.
One useful observation is that G and C(1) depend only on (1). In order to
get X , we need something that encodes the rest of the fan . We will do this using
a monomial ideal in the coordinate ring of C(1) . Introduce a variable x for each
(1) and let
S = C[x | (1)].
Then Spec(S) = C(1) . We call S the total coordinate ring of X .

For each cone , define the monomial
Y
x =
x .
(1)
/
Thus x is the product of the variables corresponding to rays not in . Then define
the irrelevant ideal
B() = hx | i S.
A useful observation is that x is a multiple of x whenever . Hence, if max
is the set of maximal cones of , then
B() = hx | max i.
Furthermore, one sees easily that the minimal generators of B() are precisely the
x for max . Hence, once we have (1), B() determines uniquely.
208
Now define
Z() = V(B()) C(1) .
The variety of a monomial ideal is a union of coordinate subspaces. For B(), the
coordinate subspaces can be described in terms of primitive collections, which are
defined as follows.
Definition 5.1.5. A subset C (1) is a primitive collection if:
(a) C 6 (1) for all .
(b) For every proper subset C ( C, there is with C (1).

Proposition 5.1.6. The Z() as a union of irreducible components is given by
[
Z() = V(x | C),
C
where the union is over all primitive collections C (1).
Proof. It suffices to determine the maximal coordinate subspaces contained in

Z(). Suppose that V(x1 , . . . , xs ) Z() is such a subspace and take .
Since x vanishes on Z() and hx1 , . . . , xs i is prime, the Nullstellensatz implies
x is divisible by some xi , i.e., i
/ (1). It follows that C = {1 , . . . , s } satisfies condition (a) of Definition 5.1.5, and condition (b) follows easily from the
maximality of V(x1 , . . . , xs ). Hence C is a primitive collection.
Conversely, every primitive collection C gives a maximal coordinate subspace
V(x | C) contained in Z(), and the proposition follows.

In Exercise 5.1.2 you will show that the algebraic analog of Proposition 5.1.6
is the primary decomposition
\
B() = hx | Ci.
C
Here are some easy examples.
Example 5.1.7. The fan for P n consists of cones generated by proper subsets of
{u0 , . . . , un }, where u0 , . . . , un are as in Example 5.1.2. Let ui generate i , 0 i n,
and let xi be the corresponding variable in the total coordinate ring. We compute
Z() in two ways:
The maximal cones of the fan are given by i = Cone(u0 , . . . , ubi , . . . , un ). Then
x i = xi , so that B() = hx0 , . . . , xn i. Hence Z() = {0}.
The only primitive collection is {0 , . . . , n }, so Z() = V(x0 , . . . , xn ) = {0}

by Proposition 5.1.6.
Example 5.1.8. The fan for P1 P1 has ray generators u1 = e1 , u2 = e1 , u3 =

e2 , u4 = e2 . See Example 3.1.12 for a picture of this fan. Each ui gives a ray i
and a variable xi . We compute Z() in two ways:
209
The maximal cone Cone(u1 , u3 ) gives the monomial x2 x4 , and similarly the
other maximal cones give the monomials x1 x4 , x1 x3 , x2 x3 . Thus
B() = hx2 x4 , x1 x4 , x1 x3 , x2 x3 i,
and one checks that Z() = {0} C2 C2 {0}.
The only primitive collections are {1 , 2 } and {3 , 4 }, so that

Z() = V(x1 , x2 ) V(x3 , x4 ) = {0} C2 C2 {0}
by Proposition 5.1.6. Note also that B() = hx1 , x2 i hx3 , x4 i.
A final observation is that (C )(1) acts on C(1) by diagonal matrices and

hence induces an action on C(1) \ Z(). It follows that G (C )(1) also acts on
C(1) \ Z(). We are now ready to take the quotient.
The Quotient Construction. To represent X as a quotient, we first construct a
toric morphism C(1) \ Z() X . Let {e | (1)} be the standard basis of
the lattice Z(1) . For each , define the cone
= Cone(e | (1)) R(1) .
It is easy to see that these cones and their faces form a fan
e = { |
for some }
in (Z(1) )R = R(1) . This fan has the following nice properties.

e be the fan defined above. Then:
Proposition 5.1.9. Let
e
(a) C(1) \ Z() is the toric variety of the fan .
(b) The map e 7 u defines a map of lattices Z(1) N that is compatible with
e in R(1) and in NR .
the fans
(c) The resulting toric morphism
: C(1) \ Z() X
is constant on G-orbits.
e 0 be the fan consisting of Cone(e | (1)) and its
Proof. For part (a), let
e is a subfan of
e 0 . Since
e 0 is the fan of C(1) , we get the
faces. Note that
e by taking C(1) and then removing the orbits corresponding to
toric variety of
e 0 \ .
e By the Orbit-Cone Correspondence (Theorem 3.2.6), this is
all cones in
e 0 \ .
e But
equivalent to removing the orbit closures of the minimal elements of
these minimal elements are precisely the primitive collections C (1). Since the
corresponding orbit closure is V(x | C), removing these orbit closures means
removing
[
Z() = V(x | C).
C
210
For part (b), define : Z(1) N by (e ) = u . Since the minimal generators

) = by the definition of
e.
of are given by u , (1), we have R (e
e
Hence is compatible with respect to the fans and .
The map of tori induced by is the map (C )(1) TN from the exact sequence (5.1.2) (you will check this in Exercise 5.1.3). Hence, if g G (C )(1)
and x C(1) \ Z(), then
(g x) = (g) (x) = (x),
where the first equality holds by equivariance and the second holds since G is the
kernel of (C )(1) TN . This proves part (c) of the proposition.

In Exercise 5.1.4 you will prove the following lemma, which will be used in
the proof of the quotient construction.
Lemma 5.1.10. Assume that V is an affine toric variety, not necessarily normal,
with torus T . Given a point p V , there is a point q T and a one-parameter
subgroup : C T such that p = limt0 (t)q.

We can now give the quotient construction of X .
Theorem 5.1.11. Let X be a toric variety without torus factors and consider the
toric morphism : C(1) \ Z() X from Proposition 5.1.9. Then:
(a) is an almost geometric quotient for the action of G on C(1) \ Z(). Thus

X C(1) \ Z() //G.
(b) is a geometric quotient if and only if is simplicial.
Proof. We begin by studying the map

(5.1.3)
|1 (U ) : 1 (U ) U
) is equivalent to .
for . First observe that if , , then R (e
It follows that 1 (U ) is the toric variety Ue of
e = Cone(e | (1)). This
shows that (5.1.3) is the toric morphism
: Ue U ,
where for simplicity we write instead of |1 (U ) .
Our first task is to show that is a good categorical quotient. Since G is

reductive, Proposition 5.0.9 reduces this to showing that the map on coordinate
rings induces an isomorphism
(5.1.4)
C[U ] C[Ue ]G .
The map can be described as follows:
211
For Ue , the cone

e gives the semigroup
e Z(1) = {(a ) Z(1) | a 0 for all (1)}.
Hence the coordinate ring of Ue is the semigroup algebra

Q a

C[Ue ] = C x | a 0 for all (1) = Sx ,
Q
x .
where Sx is the localization S = C[x | (1)] at x = (1)
/
For U , the coordinate ring is the usual semigroup algebra C[ M].
The map : Z(1) N dualizes to the map M Z(1) sending m M to

(hm, u i) Z(1) . It follows that : C[ M] Sx is given by
Q hm,u i
( m ) = x .
Note that hm, u i 0 for all (1), so that the expression on the right really
lies in Sx .
Thus can be written : C[ M] Sx , and since is constant on G-orbits,

factors as
G
C[ M] Sx Sx .
The map has Zariski dense image in U since ((C )(1) ) = TN by the
exact sequence (5.1.2). It follows that is injective. To show that its image is
(Sx )G , take f Sx and write it as
X
f=
ca x a
a
where each x a = x satisfies a 0 for all (1). Then f is G-invariant if

and only if for all t = (t ) G, we have
X
X
ca x a =
ca t a x a .
a
ta
Thus f is G-invariant if and only if = 1 for all t G whenever ca 6= 0. The

map t 7 t a is a character on G and hence is an element of its character group
Cl(X ) (Lemma 5.1.1). This character is trivial when ca 6= 0, so that by (5.1.1), the
exponent vector a = (a ) must come from an element m M, i.e., a = hm, u i for
all (1). But x a Sx , which implies that
hm, u i = a 0 for all (1).
Hence m M, which implies that f is in the image of . This proves (5.1.4).
We conclude that is a good categorical quotient.
We next follow ideas from [83, Prop. 12.1] to prove that
(5.1.5)
: U U is a geometric quotient is simplicial.
212
First suppose that is simplicial. By Proposition 5.0.8, it suffices to show that

G-orbits are closed in U . Let G G be the connected component of the identity.
Since G has finite index in G, it suffices to show that G -orbits are closed in U .
Take p U and p G p, where the closure is taken in U . Note that
is an affine toric variety, possibly nonnormal, with torus T G /G p . By
Lemma 5.1.10, there are : C T and q T such that p = limt0 (t)q p.
Lifting these to G gives a one-parameter subgroup : C G and a point q G
such that
G p
p = lim (t)q p.
(5.1.6)
t0
(C )(1) ,
Using
we can write (t) = (t a ) for exponents a Z. Since is

a one-parameter subgroup of G, we have
P
(5.1.7)
a u = 0.
This follows easily from Lemma 5.1.1 (Exercise 5.1.5). Write p = (p ), p = (p ),
and q = (q ). Then (5.1.6) implies
p = lim t a q p .
t0
G,
Since p, p Ue and q
their th coordinates are nonzero for
/ (1). Then
the above limit implies a = 0 for these s, so that (5.1.7) becomes
P
(1) a u = 0.
But is simplicial, which means that the u , (1), are linearly independent.
Hence a = 0 for all , so that is the trivial one-parameter subgroup. Then (5.1.6)
implies p = q p G p. We conclude that G p is closed in U .
For the other implication

of (5.1.5), suppose that is non-simplicial. Then
P
there is a relation (1) a u = 0 where a Z and a > 0 for at least one . If
we set a = 0 for
/ (1), then the one-parameter subgroup
(t) = (t a ) (C )(1)
is a one-parameter subgroup of G by Exercise 5.1.5. Define p = (p ) Ue by

(
1 a 0
p =
0 a < 0
and consider limt0 (t) p. The limit exists in C(1) since p = 0 for a < 0.
Furthermore, if
/ (1), the th coordinate of (t) p is 1 for all t, so that the limit
p = limt0 (t) p lies in Ue . Since there is 0 (1) with a0 > 0, we have:
Since the 0 th coordinate of p is nonzero, the same is true for all of G p.
Since a0 > 0, the 0 th coordinate of p = limt0 (t) p is zero.
Then G p is not closed in Ue since its Zariski closure contains p Ue \ G p. This

shows that is not a geometric quotient and completes the proof of (5.1.5).
213
We can now prove the theorem. Since the maps (5.1.3) are good categorical
quotients, the same is true for : C(1) \ Z() X by Proposition 5.0.12. To
prove part (a), let be the subfan of simplicial cones of . Then X is open
in X , and since (1) = (1), X and X have the same total coordinate ring S
and same group G. In Exercise 5.1.5, you will show that
[
(5.1.8)
1 (X ) = C(1) \ Z( ) =
Ue .
As above, |1 (X ) : 1 (X ) X is a good categorical quotient, and is
a geometric quotient for each by (5.1.5). It follows easily that |1 (X )
is a geometric quotient, and then Proposition 5.0.11 implies that is an almost

geometric quotient. This argument also implies that is a geometric quotient
when is simplicial, which proves half of part (b). The other half of part (b)
follows from (5.1.5). The proof of the theorem is now complete.

One nice feature of the quotient X = (C(1) \Z())//G is that it is compatible
with the tori, meaning that we have a commutative diagram
X (C(1) \ Z())//G
TN
(C )(1) /G
where the isomorphism on the bottom comes from (5.1.2) and the vertical arrows
are inclusions.
Examples. Here are some examples of the quotient construction.
Example 5.1.12. By Examples 5.1.2 and 5.1.7, P n has quotient representation
P n = (Cn+1 \ {0})/C ,
where C acts by scalar multiplication. This is a geometric quotient since is

smooth and hence simplicial.
Example 5.1.13. By Examples 5.1.3 and 5.1.8, P1 P1 has quotient representation

P1 P1 = C4 \ ({0} C2 C2 {0}) /(C )2 ,
where (C )2 acts via (, )(a, b, c, d) = (a, b, c, d). This is again a geometric

quotient.
Example 5.1.14. Fix positive integers q0 , . . . , qn with gcd(q0 , . . . , qn ) = 1 and let

N be the lattice Zn+1 /Z(q0 , . . . , qn ). The images of the standard basis in Zn+1 give
primitive elements u0 , . . . , un N satisfying q0 u0 + + qn un = 0. Let be the fan
consisting of all cones generated by proper subsets of {u0 , . . . , un }.
As in Example 3.1.17, the corresponding toric variety is denoted P(q0 , . . . , qn ).

Using the quotient construction, we can now explain why this is called a weighted
projective space.
214
We have C(1) = Cn+1 since has n+1 rays, and Z() = {0} by the argument
used in Example 5.1.7. It remains to compute G (C )n+1 . In Exercise 4.1.5, you
showed that the maps m M 7 (hm, u0 i, . . . , hm, un i) Zn+1 and (a0 , . . . , an )
Zn+1 7 a0 q0 + + an qn Z give the short exact sequence
(5.1.9)
0 M Zn+1 Z 0.
This shows that the class group is Z. Note also that ei Zn+1 maps to qi Z. In
Exercise 5.1.6 you will compute that
G = {(t q0 , . . . ,t qn ) | t C } C .
This is the action of C on Cn+1 given by
t (u0 , . . . , un ) = (t q0 u0 , . . . ,t qn un ).
Since is simplicial (every proper subset of {u0 , . . . , un } is linearly independent in
NR ), we get the geometric quotient
P(q0 , . . . , qn ) = (Cn+1 \ {0})/C .
This gives the set-theoretic definition of P(q0 , . . . , qn ) from 2.0 and also gives its
structure as a variety since we have a geometric quotient.
Example 5.1.15. Consider the cone = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) R3 . To

find the quotient representation of U , we label the ray generators as
u1 = e1 , u2 = e2 + e3 , u3 = e2 , u4 = e1 + e3 .
Then C(1) = C4 and Z() = since x = 1. To determine the group G (C )4 ,
note that the exact sequence (5.1.1) becomes
0 Z3 Z4 Z 0,
where (a1 , a2 , a3 , a4 ) Z4 7 a1 + a2 a3 a4 Z. This makes it straightforward

to show that
G = {(, , 1 , 1 ) | C } C .
Hence we get the quotient presentation
U = C4 //C .
In Example 5.0.2, we gave a naive argument that the quotient was V(xy zw). We
now see that the intrinsic meaning of Example 5.0.2 is the quotient construction of
U given by Theorem 5.1.11. This example is not a geometric quotient since is
not simplicial.
Example 5.1.16. Let Bl0 (C2 ) be the blowup of C2 at the origin, whose fan is
shown in Figure 1 on the next page. By Example 4.1.5, Cl(Bl0 (C2 )) Z with
generator [D1 ] = [D2 ] = [D0 ]. Hence G = C and the irrelevant ideal is B() =
hx, yi. This gives the geometric quotient

Bl0 (C2 ) C3 \ (C {0, 0}) /C ,
u2
215
u0
u1
Figure 1. The fan for the blowup of C2 at the origin
where the C -action is given by (t, x, y) = (1t, x, y).
We also have C[t, x, y]C = C[tx,ty]. Then the inclusion

C3 \ (C {0, 0}) C3
induces the map on quotients

: Bl0 (C2 ) C3 \ (C {0, 0}) /C C3 //C C2 ,
where the final isomorphism uses
C3 //C = Spec(C[t, x, y]C ) = Spec(C[tx,ty]).

In terms of homogeneous coordinates, (t, x, y) = (tx,ty). This map is the toric
morphism Bl0 (C2 ) C2 induced by the refinement of Cone(u1 , u2 ) given by .
The quotient representation makes it easy to see why Bl0 (C2 ) is the blowup of
C2 at the origin. Given a point of Bl0 (C2 ) with homogeneous coordinates (t, x, y),
there are two possibilities:
t 6= 0, in which case t (t, x, y) = (1,tx,ty). This maps to (tx,ty) C2 and
is nonzero since x, y cannot both be zero. It follows that the part of Bl0 (C2 )
where t 6= 0 looks like C2 \ {0, 0}.
t = 0, in which case (0, x, y) maps to the origin in C2 . Since (0, x, y) =

(x, y) and x, y cannot both be zero, it follows that the part of Bl0 (C2 ) where
t = 0 looks like P1 .
This shows that Bl0 (C2 ) is a built from C2 by replacing the origin with a copy
of P1 , which is called the exceptional locus E. Since E = 1 (0, 0), we see that
: X C2 induces an isomorphism
Bl0 (C2 ) \ E C2 \ {(0, 0)}.
Note also that E is the divisor D0 corresponding to the ray 0 . You should be able
to look at Figure 1 and see instantly that D0 P1 .
We can also check that lines through the origin behave properly. Consider the
line L defined by ax + by = 0, where (a, b) 6= (0, 0). When we pull this back to
216
Bl0 (C2 ), we get the subvariety defined by

a(tx) + b(ty) = 0.
This is the total transform of L. It factors as t(ax + by) = 0. Note that t = 0 defines
the exceptional locus, so that once we remove this, we get the curve in Bl0 (C2 )
defined by ax + by = 0. This is the proper transform of L, which meets the exceptional locus E at the point with homogeneous coordinates (0, b, a), corresponding
to (b, a) P1 . In this way, we see how blowing up separates tangent directions
through the origin.
The General Case. So far, we have assumed that X has no torus factors. When
torus factors are present, X still has a quotient construction, though it is no longer
canonical.
Let X be a toric variety with a torus factor. By Proposition 3.3.9, the ray
generators u , (1), span a proper subspace of NR . Let N be the intersection
of this subspace with N, and pick a complement N so that N = N N . The cones
of all lie in NR and hence give a fan in NR . As in the proof of Proposition 3.3.9.
we obtain
X X ,N (C )r
where N Zr . Theorem 5.1.11 applies to X,N since u , (1) = (1), span
NR by construction. Note also that B( ) = B() and Z( ) = Z(). Hence
X ,N (C(1) \ Z())//G.
It follows that
X X,N (C )r
(5.1.10)

(C(1) \ Z())//G (C )r

(C(1) (C )r ) \ (Z() (C )r ) //G,
In the last line, we use the trivial action of G on (C )r . You will verify the last
isomorphism in Exercise 5.1.7.
We can rewrite (5.1.10) as follows. Using (C )r = Cr \ V(x1 xr ), we obtain
(C(1) (C )r ) \ (Z() (C )r ) = C(1)+r \ Z (),
where C(1)+r = C(1) Cr and Z () = (Z() Cr ) (C(1) V(x1 xr )).
Hence the quotient presentation of X is the almost geometric quotient

(5.1.11)
X C(1)+r \ Z () //G.
This differs from Theorem 5.1.11 in two ways:
The representation (5.1.11) is non-canonical since it depends on the choice of

the complement N .
217
Z () contains V(x1 xr ) C(1) and hence has codimension 1 in C(1)+r .

In constrast, Z() always has codimension 2 in C(1) (this follows from
Proposition 5.1.6 since every primitive collection has at least two elements).
In practice, (5.1.11) is rarely used, while Theorem 5.1.11 is a common tool in toric
geometry.
Exercises for 5.1.
5.1.1. In Example 5.1.4, verify carefully that G = {(, ) | d }.
T
5.1.2. Prove that B() = C hx | Ci, where the intersection ranges over all primitive
collections C (1).
5.1.3. In Proposition 5.1.9, we defined : Z(1) N, and in the proof we use the map of
tori (C )(1) TN induced by . Show that this is the map appearing in (5.1.2).
5.1.4. This exercise will prove Lemma 5.1.10. In parts (a) and (b), we consider a normal
affine toric variety U and a point p U . By Theorem 3.2.6, there is a face such
that p O( ) U U . Also take u Relint( ) N.
(a) Prove limt0 u (t) = , where O( ) is the distinguished point defined in 3.2.
Hint: Proposition 3.2.2.
(b) Find q TN such that limt0 u (t)q = p. Hint: TN acts transitively on O( ).
(c) Prove Lemma 5.1.10. Hint: Let U V be the normalization map. Then apply
Theorem 3.A.3.
5.1.5. This exercise is concerned with the proof of Theorem 5.1.11.

(1)
a
takes values in G if and
(a) Prove that
P a one-parameter subgroup (t) = (t ) (C )
only if a u = 0. Hint: Use Lemma 5.1.1. You can give a more conceptual proof
by taking the dual of (5.1.1).
(b) Prove (5.1.8) and conclude that the quotient construction of X is the map |1 (X
(X ) X used in the proof of Theorem 5.1.11.
qn
5.1.6. Show that the group G in
5.1.14 is given by G = {(t q0 , . . . ,tQ
) | t C }.
PExample
n
n
Hint: Pick integers bi such that i=0 bi qi = 1. Given (t0 , . . . ,tn ) G, set t = i=0 tiai . Also
note that if e0 , . . . , en is the standard basis of Zn+1 , then qi e j q j ei Zn+1 maps to 0 Z
in (5.1.9).
5.1.7. Let X be a variety with trivial G action. Prove that (X Ue )//G X U and use
this to verify the final line of (5.1.10).
5.1.8. Consider the usual fan for P 2 with the lattice N = {(a, b) Z2 | a + b 0 mod d},
where d is a positive integer.
(a) Prove that the ray generators are u1 = (d, 0), u2 = (0, d) and
(
(d, d)
d odd
u0 =
(d/2, d/2) d even.
(b) Prove that the dual lattice is M = {(a/d, b/d) | a, b Z, a b 0 mod d}.
(c) Prove that Cl(X ) = Z Z/dZ (d odd) or Z Z/ d2 Z (d even).
218
(d) Compute the quotient representation of X .

5.1.9. Find the quotient representation of the Hirzeburch surface Hr in Example 3.1.16.
5.1.10. Prove that G acts freely on C(1) \ Z() when the fan is smooth. Hint: Let
and suppose that g = (t ) G fixes u = (u ) Ue . Show that t = 1 for
/ and then
use Lemma 5.1.1 to show that t = 1 for all .
5.1.11. Prove that G acts with finite isotropy subgroups on C(1) \ Z() when the fan is
simplicial. Hint: Adapt the argument used in Exercise 5.1.10.
5.1.12. Prove that 2 codim(Z()) |(1)|. When is a complete simplicial fan, a
stronger result states that either
(a) 2 codim(Z()) 12 dim X + 1, or
(b) |(1)| = dim X + 1 and Z() = {0}.
This is proved in [19, Prop. 2.8]. See the next exercise for more on part (b).
5.1.13. Let be a complete fan such that |(1)| = n + 1, where n = dim X . Prove that
there is a weighted projective space P(q0 , . . . , qn ) and a finite abelian group H acting on
P(q0 , . . . , qn ) such that
X P(q0 , . . . , qn )/H.
These are called fake weighted projective spaces in [60] and [167]. Also prove that the
(a) X is a weighted projective space.
(b) Cl(X ) Z.
(c) N is generated by u , (1).

Hint: Label the ray generators u0 , . . . , un . P
First show that is simplicial and that there
n
are positive integers q0 , . . . , qn satisfying i=0 qi ui = 0 and gcd(q0 , . . . , qn ) = 1. Then
consider the sublattice of N generated by the ui and use Example 5.1.14. You will also
need Proposition 3.3.7. If you get stuck, see [19, Lem. 2.11].
5.1.14. In the proof of Theorem 5.1.11, we showed that a non-simplicial cone leads to a
non-closed G-orbit. Show that the non-closed G-orbit exhibited in Example 5.0.2 is an
example of this construction. See also Example 5.1.15.
5.1.15. Example 5.1.16 gave the quotient construction of the blowup of 0 C2 and used
the quotient construction to describe the properties of the blowup. Give a similar treatment
for the blowup of Cr Cn using the star subdivision described in 3.3.
5.2. The Total Coordinate Ring

In this section we assume that X is a toric variety without torus factors. Its total
coordinate ring
S = C[x | (1)]
was defined in 5.1. This ring gives C(1) = Spec(S) and contains the irrelevant
ideal
B() = hx | i
used in the quotient construction of X . In this section we will explore how this
ring relates to the algebra and geometry of X .
219
The Grading. An important feature of the total coordinate ring S is its grading by
the class group Cl(X ). We have the exact sequence (5.1.1)
0 M Z(1) Cl(X ) 0,

P
where a = (a ) Z(1) maps to the divisor class
a D Cl(X ). Given a
Q a
monomial x a = x S, define its degree to be

P
deg(x a ) =
a D Cl(X ).
For Cl(X ), we let S denote the corresponding graded piece of S.
The grading on S is closely related to the group G = HomZ (Cl(X ), C ). Recall that Cl(X ) is the character group of G, where as usual Cl(X ) gives the
character : G C . The action of G on C(1) induces an action on S with the
property that given f S, we have
(5.2.1)
f S g f = (g1 ) f for all g G
f (g x) = (g) f (x) for all g G, x C(1)
(Exercise 5.2.1). Thus the graded pieces of S are the eigenspaces of the action of
G on S. We say that f S is homogeneous of degree .
Example 5.2.1. The total coordinate ring of P n is C[x0 , . . . , xn ]. By Example 4.1.6,
the map Zn+1 Z = Cl(P n ) is (a0 , . . . , an ) 7 a0 + + an . This gives the grading on C[x0 , . . . , xn ] where each variable xi has degree 1, so that homogeneous
polynomial has the usual meaning.
In Exercise 5.2.2 you will generalize this by showing that the total coordinate ring of the weighted projective space P(q0 , . . . , qn ) is C[x0 , . . . , xn ], where the
variable xi now has degree qi . Here, homogeneous polynomial means weighted
homogeneous polynomial.
Example 5.2.2. The fan for P n P m is the product of the fans of P n and P m , and
by Example 4.1.7, the class group is
Cl(P n P m ) Cl(P n ) Cl(P m ) Z2 .
The total coordinate ring is C[x0 , . . . , xn , y0 , . . . , ym ], where

deg(xi ) = (1, 0)
deg(yi ) = (0, 1)
(Exercise 5.2.3). For this ring, homogeneous polynomial means bihomogeneous

polynomial.
Example 5.2.3. Example 5.1.16 gave the quotient representation of the blowup
Bl0 (C2 ) of C2 at the origin. The fan of Bl0 (C2 ) is shown in Example 5.1.16 and
has ray generators u0 , u1 , u2 , corresponding to variables t, x, y in the total coordinate
ring S = C[t, x, y]. Since Cl(Bl0 (C2 )) Z, one can check that the grading on S is
given by
deg(t) = 1 and deg(x) = deg(y) = 1
220
(Exercise 5.2.4). Thus total coordinate rings can have some elements of positive
degree and other elements of negative degree.
The Toric Ideal-Variety Correspondence. For n-dimensional projective space P n ,

a homogeneous ideal I C[x0 , . . . , xn ] defines a projective variety V(I) P n . This
generalizes to more general toric varieties X as follows.
We first assume that is simplicial, so that we have a geometric quotient
: C(1) \ Z() X
by Theorem 5.1.11. Given p X , we say a point x 1 (p) gives homogeneous

coordinates for p. Since is a geometric quotient, we have 1 (p) = G x. Thus
all homogeneous coordinates for p are of the form g x for some g G.
Now let S be the total coordinate ring of X and let f S be homogeneous for
the Cl(X )-grading on S, say f S . Then
f (g x) = (g) f (x)
by (5.2.1), so that f (x) = 0 for one choice of homogeneous coordinates of p X

if and only if f (x) = 0 for all homogeneous coordinates of p. It follows that the
equation f = 0 is well-defined in X . We can use this to define subvarieties of X
as follows.
Proposition 5.2.4. Let S be the total coordinate ring of the simplicial toric variety
X . Then:
(a) If I S is a homogeneous ideal, then
V(I) = {(x) X | f (x) = 0 for all f I}
is a closed subvariety of X .
(b) All closed subvarieties of X arise this way.
Proof. Given I S as in part (a), notice that
W = {x C(1) \ Z() | f (x) = 0 for all f I}
is a closed G-invariant subset of C(1) \ Z(). By part (b) of the definition of good
categorical quotient (Definition 5.0.5), V(I) = (W ) is closed in X .
Conversely, given a closed subset Y X , its inverse image
1 (Y ) C(1) \ Z()
is closed and G-invariant. Then the same is true for the Zariski closure
1 (Y ) C(1) .
It follows without difficulty that I = I( 1 (Y )) S is a homogeneous ideal satisfying V(I) = Y .
221
Example 5.2.5. The equation x = 0 defines the TN -invariant closed subvariety

V(x ) X which is easily seen to be the prime divisor D . This shows that D
always has a global equation, though it fails to have local equations when D is not
Cartier (see Example 4.2.3).
Classically, the Weak Nullstellensatz gives a necessary and sufficient condition

for the variety of an ideal to be empty. This applies to Cn and P n as follows:
For Cn : Given an ideal I C[x1 , . . . , xn ],
V(I) = in Cn 1 I.
For P n : Given a homogeneous ideal I C[x0 , . . . , xn ],
V(I) = in P n hx0 , . . . , xn i I for some 0.
For a toric version of the weak Nullstellensatz, we use the irrelevant ideal B() =
hx | i S.
Proposition 5.2.6 (The Toric Weak Nullstellensatz). Let X be a simplicial toric
variety with total coordinate ring S and irrelevant ideal B() S. If I S is a
homogeneous ideal, then
V(I) = in X B() I for some 0.
Proof. Let Va (I) C(1) denote the affine variety defined by I S. Then:

V(I) = in X Va (I) C(1) \ Z() =
Va (I) Z() = Va (B())
B() I for some 0,

where the last equivalence uses the Nullstellensatz in C(1) .
For Cn and P n , the irrelevant ideal is h1i C[x1 , . . . , xn ] and hx0 , . . . , xn i

C[x0 , . . . , xn ] respectively. For Cn , the grading on C[x1 , . . . , xn ] by Cl(Cn ) = {0}
is trivial, so that every ideal is homogeneous. Thus the toric weak Nullstellensatz
implies the classical version of the weak Nullstellensatz for both Cn and P n .
The relation between ideals and varieties is not perfect because different ideals
can define the same subvariety. In Cn and P n , we avoid this by using radical ideals:
For Cn : There is a bijective correspondence
{closed subvarieties of Cn } {radical ideals I C[x1 , . . . , xn ]}.
For P n : There is a bijective correspondence

radical homogeneous ideals
n
.
{closed subvarieties of P }
I hx0 , . . . , xn i C[x0 , . . . , xn ]
Here is the toric version of this correspondence.
222
Proposition 5.2.7 (The Toric Ideal-Variety Correspondence). Let X be a simplicial toric variety. Then there is a bijective correspondence

radical homogeneous
.
{closed subvarieties of X }
ideals I B() S
Proof. Given a closed subvariety Y X , we
can find a homogeneous ideal I S
with
V(I)
=
Y
by
Proposition
5.2.4.
Then
I is also homogeneous and satisfies
V( I) = V(I) = Y , so we may assume that I is radical. Since

Va (I B()) = Va (I) Va (B()) = Va (I) Z()
in C(1) , we see that I B() B() is a radical homogeneous ideal satisfying
V(I B()) = Y . This proves surjectivity.
To prove injectivity, suppose that I, J B() are radical homogeneous ideals

with V(I) = V(J) in X . Then

Va (I) C(1) \ Z() = Va (J) C(1) \ Z() .
However, I, J B() implies that Z() is contained in Va (I) and Va (J). Hence
the above equality implies
Va (I) = Va (J),
so that I = J by the Nullstellensatz since I and J are radical.
For general ideals, another way to recover injectivity is to work with closed
subschemes rather than closed subvarieties. We will say more about this in the
appendix to Chapter 6.
When X is not simplicial, there is still a relation between ideals in the total
coordinate ring and closed subvarieties of X .
Proposition 5.2.8. Let S be the total coordinate ring of the toric variety X . Then:
(a) If I S is a homogeneous ideal, then
V(I) = {p X | there is x 1 (p) with f (x) = 0 for all f I}
is a closed subvariety of X .
(b) All closed subvarieties of X arise this way.
Proof. The proof is identical to the proof of Proposition 5.2.4.
The main difference between Propositions 5.2.4 and 5.2.8 is the phrase there
is x 1 (p). In the simplicial case, all such x are related by the group G, while
this may fail in the non-simplicial case. One consequence is that the ideal-variety
correspondence of Proposition 5.2.7 breaks down in the nonsimplicial case. Here
is a simple example.
223
Example 5.2.9. In Example 5.1.15 we described the quotient representation of

U = C4 //C for the cone = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) R3 , and in Example 5.0.2 we saw that the quotient map
: C4 U = V(xy zw) C4
is given by (a1 , a2 , a3 , a4 ) = (a1 a3 , a2 a4 , a1 a4 , a2 a3 ). Note that the irrelevant ideal
is B() = C[x1 , x2 , x3 , x4 ].
The ideals I1 = hx1 , x2 i and I2 = hx3 , x4 i are distinct radical homogeneous ideals
contained in B() that give the same subvariety in U :
V(I1 ) = (Va (I1 )) = (C2 {0}) = {0} U
V(I2 ) = (Va (I2 )) = ({0} C2 ) = {0} U .

Thus Proposition 5.2.7 fails to hold for this toric variety.
Local Coordinates. Let X be an n-dimensional toric variety. When contains a

smooth cone of dimension n, we get an affine open set
with U Cn
U X
The usual coordinates for Cn are compatible with the homogeneous coordinates
for X in the following sense. The cone gives the map : C(1) C(1) that
sends (a )(1) to the point (b )(1) defined by
(
a (1)
b =
1 otherwise.
Proposition 5.2.10. Let be a smooth cone of dimension n = dim X and let
: C(1) C(1) be defined as above. Then we have a commutative diagram
C(1)

/ C(1) \ Z()

/ X ,

U
where the vertical maps are the quotient maps from Theorem 5.1.11. Furthermore,
the vertical map on the left is an isomorphism.
Proof. We first show commutativity. In the proof of Theorem 5.1.11 we saw that
1 (U ) = Ue . Since the image of lies in Ue , we are reduced to the diagram

/ Ue

CC

CC

CC

C!

C(1)C
U .
224
Since everything is affine, we can consider the corresponding diagram of coordinate rings
C[x | (1)] o
dII
II
II
I
II
C[x | (1)]x
9
sss
s
s
ss
sss
C[ M],
Q
Q
hm,u i
hm,u i
where ( m ) = (1) x and ( m ) = (1) x for m M. It
is clear that = , and commutativity follows.
For the final assertion, note that is an isomorphism since the u , (1),
form a basis of N by our assumption on . This completes the proof.

It follows that if a closed subvariety Y X is defined by an ideal I S,
then the affine piece Y U U C(1) is defined by the dehomogenized ideal
I C[x | (1)] obtained by setting x = 1,
/ (1), in all polynomials of I.
We will give examples of this below, and in 5.4, we will explore the corresponding
notion of homogenization.
Proposition 5.2.10 can be generalized to any cone satisfying dim =
dim X (Exercise 5.2.5).
Example 5.2.11. In Example 5.1.16 we described the quotient construction of the
blowup of C2 at the origin. This variety can be expressed as the union Bl0 (C2 ) =
U1 U2 , where 1 , 2 are as in Example 5.1.16.
The map Bl0 (C2 ) C2 is given by (t, x, y) 7 (tx,ty) in homogeneous coordinates. Combining this with the local coordinate maps from Proposition 5.2.10, we
obtain
U1 X C2 : (t, x) 7 (t, x, 1) 7 (tx,t)
U2 X C2 : (t, y) 7 (t, 1, y) 7 (t,ty).
Consider the curve f (x, y) = 0 in the plane C2 , where f (x, y) = x 3 y 2 . We study

this on the blowup Bl0 (C2 ) using local coordinates as follows:
On U1 , we get f (tx,t) = 0, i.e., (tx)3 t 2 = t 2 (tx 3 1) = 0. Since t = 0
defines the exceptional locus, we get the proper transform tx 3 1 = 0.
On U2 , we get f (t,ty) = 0, i.e., t 3 (ty)2 = t 2 (t y 2 ) = 0, with proper transform t y 2 = 0.
Hence the proper transform is a smooth curve in Bl0 (C2 ). This method of studying
the blowup of a curve is explained in many elementary texts on algebraic geometry,
such as [236, p. 100].
We relate this to the homogeneous coordinates of Bl0 (C2 ) as follows. Using

the above map X C2 , we get the curve in X defined by f (tx,ty) = 0, i.e.,
(tx)3 (ty)2 = t 2 (tx 3 y 2 ) = 0. Hence the proper transform is tx 3 y 2 = 0. Then:
225
Setting y = 1 gives the proper transform tx 3 1 = 0 on U1 .

Setting x = 1 gives the proper transform t y 2 = 0 on U2 .
Hence the local proper transforms computed above are obtained from the homogeneous proper transform by setting appropriate coordinates equal to 1.
Exercises for 5.2.

5.2.1. Prove (5.2.1).
5.2.2. Show that the total coordinate ring of the weighted projective space P(q0 , . . . , qn ) is
C[x0 , . . . , xn ] where deg(xi ) = qi . Hint: See Example 5.1.14.
5.2.3. Prove the claims made about the total coordinate ring of the product P n P m made
in Example 5.2.2.
5.2.4. Prove the claims made about the class group and the total coordinate ring of the
blowup of P 2 at the origin made in Example 5.2.3.
5.2.5. Let X be the toric variety of the fan and assume as usual that X has no torus
factors. A subfan is full if = { | (1) (1)}. Consider a full subfan
with the property that X has no torus factors.
(a) Define the map : C (1) C(1) by sending (a ) (1) to the point (b )(1)
given by
(
a (1)
b =
1 otherwise.
Prove that there is a commutative diagram
/ (1)
C
\ Z()
C (1) \ Z( )

X

/ X ,
where the vertical maps are the quotient maps from Theorem 5.1.11.
(b) Explain how part (a) generalizes Proposition 5.2.10.
(c) Use part (a) to give a version of Proposition 5.2.10 that applies to any cone
satisfying dim = dim X .
5.2.6. The quintic y 2 = x 5 in C2 has a unique singular point at the origin. We will resolve
the singularity using successive blowups.
(a) Show that the proper transform of this curve in Bl0 (C2 ) is defined by y 2 t 3 x 5 = 0.
This uses the homogeneous coordinates t, x, y from Example 5.2.3.
(b) Show that the proper transform is smooth on U1 but singular on U2 .
(c) Subdivide 2 to obtain a smooth fan . The toric variety X has variables u,t, x, y,
where u corresponds to the ray that subdivides 2 . Show that Cl(X ) Z2 with
deg(u) = (0, 1), deg(t) = (1, 0), deg(x) = (1, 1), deg(y) = (1, 2).
(d) Show that (u,t, x, y) 7 (utx, u 2ty) defines a toric morphism X C2 and use this to
show that the proper transform of the quintic in C2 is defined by y 2 ut 3x 5 = 0.
(e) Show that the proper transform is smooth by inspecting it in local coordinates.
226
5.2.7. Adapt the method Exercise 5.2.6 to desingularize y 2 = x 2n+1 , n 1 an integer.

5.2.8. Given an ideal I in a commutative ring R, its Rees algebra is the graded ring
R[I] =
M
i=0
I it i R[t],
where t is a new variable and I = R. There is also the extended Rees algebra
M
R[I,t 1 ] =
I it i R[t,t 1 ],
iZ
where I = R for i 0. These rings are graded by letting deg(t) = 1, so that elements of R
have degree 0. See [56, 4.4] and 11.3 for more about Rees algebras.
(a) When I = hx, yi R = C[x, y], prove that the extended Rees algebra R[I,t 1 ] is the
polynomial ring C[xt, yt,t 1 ]
(b) Prove that the ring of part (a) is isomorphic to the total coordinate ring of the blowup
of C2 at the origin.
(c) Generalize parts (a) and (b) to the case of I = hx1 , . . . , xn i R = C[x1 , . . . , xn ].
5.3. Sheaves on Toric Varieties

Given a toric variety X , we show that graded modules over the total coordinate
ring S = C[x | (1)] give quasicoherent sheaves on X . We continue to assume
that X has no torus factors.
Graded Modules. The grading on S gives a direct sum decomposition
M
S=
S
Cl(X )
such that S S S+ for all , Cl(X ).

Definition 5.3.1. An S-module M is graded if it has a decomposition
M
M=
M
Cl(X )
such that S M M+ for all , Cl(X ). Given Cl(X ), the shift M()
is the graded S-module satisfying
M() = M+
for all Cl(X ).
The passage from a graded S-module to a quasicoherent sheaf on X requires
some tools from the proof of Theorem 5.1.11. A cone gives the monomial
Q
Q hm,u i
x = 6(1) x S, and by (5.1.4), the map m 7 xhmi = x induces an
isomorphism
: C[ M] (Sx )G Sx ,
227
where Sx is the localization of S at x . Since monomials are homogeneous, Sx is

also graded by Cl(X ), and its elements of degree 0 are precisely its G-invariants
(Exercise 5.3.1), i.e., (Sx )0 = (Sx )G . Hence the above isomorphism becomes
(5.3.1)
: C[ M] (Sx )0 .
These isomorphisms glue together just as we would hope.

Lemma 5.3.2. Let = m be a face of . Then (Sx )0 = ((Sx )0 ) ( m ) , and
there is a commutative diagram of isomorphisms
(Sx )0
/ ((Sx )0 ) ( m )

/ C[ M] m .
C[ M]
Proof. Since = m , we have hm, u i = 0 when (1) and hm, u i > 0 when
(1)\ (1). This means that Sx = (Sx ) ( m ) . Taking elements of degree zero
commutes with localization, hence (Sx )0 = ((Sx )0 ) ( m ) . The vertical maps
in the diagram come from (5.3.1), and the horizontal maps are localization. In
Exercise 5.3.2 you will chase the diagram to show that it commutes.

From Modules to Sheaves. We now construct the sheaf of a graded module.
Proposition 5.3.3. Let M be a graded S-module. Then there is a quasicoherent
e on X such that for every , the sections of M
e over U X are
sheaf M
e = (M )0 .
(U , M)
x
Proof. Since M is a graded S-module, it is immediate that Mx is a graded Sx module. Hence (Mx )0 is an (Sx )0 -module, which induces a sheaf (Mx )0 on
U = Spec(C[ M]) = Spec((Sx )0 ). The argument of Lemma 5.3.2 applies
verbatim to show that
(Mx )0 = ((Mx )0 ) ( m ) .
e on X which is quasicoherent by
Thus the sheaves (Mx )0 patch to give a sheaf M
construction.

Example 5.3.4. The total coordinate ring of P n is S = C[x0 , . . . , xn ] with the standard grading where every variable has degree 1. The quasicoherent sheaf on P n
associated to a graded S-module was first described by Serre in his foundational
paper Faisceaux algebriques coherents [247], called FAC for short.
An important special case is when M is a finitely generated graded S-module.

e
We will need the following finiteness result to understand the sheaf M.
Lemma 5.3.5. (Sx ) is finitely generated as a (Sx )0 -module for all and
Cl(X ).
228
P
Proof. Write = [ a D ] and consider the rational polyhedral cone
b = {(m, ) MR R | 0, hm, u i a for all } MR R.
By Gordans Lemma,
b (M Z) is a finitely generated semigroup. Let the generators with last coordinate equal to 1 be (m1 , 1), . . . , (mr , 1). Then you will prove
Q hm ,u i+a
in Exercise 5.3.3 that the monomials x i , i = 1, . . . , r, generate (Sx )

as a (Sx )0 -module.
Here are some coherent sheaves on X .
e on X is coherent when M is a finitely generated
Proposition 5.3.6. The sheaf M
graded S-module.
Proof. Because M is graded, we may assume its generators are homogeneous of

degrees 1 , . . . , r . Given , it follows immediately that Mx is finitely generated over Sx with generators in the same degrees. However, we need to be
careful when taking elements of degree 0. Multiply a generator of degree i by
the (Sx )0 -module generators of (Sx )i (finitely many by the previous lemma).
Doing this for all i gives finitely many elements in (Mx )0 that generate (Mx )0 as
e is coherent.

an (Sx )0 -module (Exercise 5.3.3). It follows that M
Given Cl(X ), the shifted S-module S() gives a coherent sheaf on X
denoted OX (). This is a sheaf we already know.
Proposition 5.3.7. Fix Cl(X ). Then:
(a) There is a natural isomorphism S (X , OX ()).

P
(b) If D = a D is a Weil divisor satisfying = [D], then
OX (D) OX ().
Proof. By definition, the sections of OX () over U are

(U , OX ()) = (S()x )0 = (Sx )
for . Since the open cover {U } of X satisfies U U = U , the
sheaf axiom gives the exact sequence
Y
Y
0 (X , OX ()) (Sx )
) .
(Sx [
Q b
The localization (Sx ) has a basis consisting of all Laurent monomials x of
degree such that b 0 for all (1). Then the exact sequence implies that
Q b
(X , OX ()) has a basis consisting of all Laurent monomials x of degree
such that b 0 for all (1). These are precisely the monomials in S of degree
, which gives the desired isomorphism S (X , OX ()).
P
We turn to part (b). Given a Weil divisor D = a D with = [D], we need
to construct a sheaf isomorphism OX (D) OX (). By the above description of
229
the sections over U , it suffices to prove that for every , we have isomorphisms
(U , OX (D)) (Sx ) .
(5.3.2)
compatible with inclusions U U induced by in .
To construct this isomorphism, we apply Proposition 4.3.3 to U to obtain

M
(U , OX (D)) =
C m.
mM
hm,u ia , (1)
A lattice point m M gives the Laurent monomial

Y hm,u i+a
(5.3.3)
xhm,Di =
x
.
When hm, u i a for (1), this lies in Sx , and in fact xhm,Di (Sx ) since
deg(xhm,Di ) =
P
(hm, u i + a )D

= div( m ) + D = [D] = .
We claim that map m 7 xhm,Di induces the desired isomorphism (5.3.2).
Suppose that m , m map to the same monomial. Then hm, u i = hm , u i

for all . This implies m = m since X has no torus factors. Furthermore, if
Q b
P
P
x b = x (Sx ) , then [ b D ] = = [ a D ], so that there is m M
such that b = hm, u i + a for all . Since x b is a monomial in Sx , b 0 for
(1), hence hm, u i a for (1). Then m (U , OX (D)) maps to
x b . This defines an isomorphism (5.3.2) which is easily seen to be compatible with
the inclusion of faces.

Example 5.3.8. For P n we have S = C[x0 , . . . , xn ] with the standard grading by
Z = Cl(P n ). Then OPn (k) is the sheaf associated to S(k) for k Z. The classes of
the toric divisors D0 Dn correspond to 1 Z, so that
OPn (k) OPn (kD0 ) OPn (kDn ).
Thus OPn (k) is a canonical model for the sheaf OPn (kDi ). This justifies what we
did in Example 4.3.1.
Also note that when k 0, we have
(P n , OPn (k)) = Sk .
Hence global sections of OPn (k) are homogeneous polynomials in x0 , . . . , xn of
degree k, which agrees with what we computed in Example 4.3.6.
230
Sheaves versus Modules. An important result is that all quasicoherent sheaves on

X come from graded modules.
Proposition 5.3.9. Let F be a quasicoherent sheaf on X . Then:
e F.
(a) There is a graded S-module M such that M
(b) If F is coherent, then M can be chosen to be finitely generated over S.
The proof will be given in the appendix to Chapter 6 since it involves tensor
products of sheaves from 6.0.
e is surjective (up to isomorphism), it is far from
Although the map M 7 M
injective. In particular, there are nontrivial graded modules that give the trivial
sheaf. This phenomenon is well-known for P n , where a finitely generated graded
module M over S = C[x0 , . . . , xn ] gives the trivial sheaf on P n if and only if M = 0
for 0 (see [131, Ex. II.5.9]). This is equivalent to
hx0 , . . . , xn i M = 0
for 0 (Exercise 5.3.4). Since hx0 , . . . , xn i is the irrelevant ideal for P n , this
suggests a toric generalization. In the smooth case, we have the following result.
Proposition 5.3.10. Let B() S be the irrelevant ideal of S for a smooth toric
e = 0 if and
variety X , and let M be a finitely generated graded S-module. Then M
only if B() M = 0 for 0.
e = 0 if and only if it vanishes on each affine open subset

Proof. First observe that M
U X . But on an affine variety, the correspondence between quasicoherent
e = 0 if and only
sheaves and modules is bijective (see [131, Cor. II.5.5]). Hence M
if (Mx )0 = 0 for all .
Next suppose that B() M = 0 for some 0. Then (x ) M = 0, which easily
e = 0 follows from the previous paragraph. This part
implies that Mx = 0. Then M
of the argument works for any toric variety.
For the converse, we have (Mx )0 = 0 for all . Given h M , we will
show that (x ) h = 0 for some 0, which will imply

P B() M = 0 for 0 since
M is finitely generated. Let = [D], where D = a D . Since is smooth, D is
Cartier, so there is m M such that hm , u i = a for all (1) (this is part
of thePCartier data for D). Replacing D with D + div( m ), we may assume that
D = (1)
a D . Now set k = max(0, a |
/ (1)) and observe that
/
Y a
Y ka
x =
x
S.
x b = (x )k
(1)
/
(1)
/
Furthermore,
Mx has degree 0. Hence x b h/(x )k = 0 in Mx , which
by the definition of localization implies that there is s 0 with
x b h/(x )k
(x )s x b h = 0 in M.
5.4. Homogenization and Polytopes
231
Since x b involves only x for

/ (1), we can find x a S such that x a x b is a
power of x . Hence multiplying the above equation by x a implies (x ) h = 0 for

some 0, as desired.

Unfortunately, the situation is more complicated when X is not smooth. Here
is an example to show what can go wrong when X is simplicial.
Example 5.3.11. The weighted projective space P(1, 1, 2) has total coordinate
ring S = C[x, y, z], where x, y have degree 1 and z has degree 2, and the irrelevant ideal is B() = hx, y, zi. The graded S-module M = S(1)/(xS(1) + yS(1))
has only elements of odd degree. Then (Mz )0 = 0 since z has degree 2, and it is
e = 0, yet one easily checks that
clear that (Mx )0 = (My )0 = 0. It follows that M
B() M = z M 6= 0 for all 0. Thus Proposition 5.3.10 fails for P(1, 1, 2).
Exercise 5.3.5 explores a version of Proposition 5.3.10 that applies to simplicial toric varieties. The condition that B() M = 0 is replaced with the weaker
condition that B() M = 0 for all Pic(X ).
We will say more about the relation between quasicoherent sheaves and graded
S-modules in the appendix to Chapter 6.
Exercises for 5.3.
5.3.1. As described in 5.0, the action of G on C(1) induces an action of G on the total
coordinate ring S. Also recall that g G is a homomorphism g : Cl(X ) C .
(a) Given x a S and g G, show that g x a = g1 () x a , where = deg(x a ).
(b) Show that SG = S0 and that a similar result holds for the localization Sx b .
5.3.2. Complete the proof of Lemma 5.3.2.

5.3.3. Complete the proofs of Lemma 5.3.5 and Proposition 5.3.6.
5.3.4. Let S = C[x0 , . . . , xn ] where deg(xi ) = 1 for all i, and let M be a finitely generated
graded S-module. Show that M = 0 for 0 if and only if hx0 , . . . , xn i M = 0 for 0.
5.3.5. Let X be a simplicial toric variety and let M be a finitely generated graded Se = 0 if and only if B() M = 0 for all 0 and Pic(X ).
module. Prove that M
5.3.6. Let X be a smooth toric variety. State and prove a version of Proposition 5.3.10
that applies to arbitrary graded S-modules M. Also explain what happens when X is
simplicial, as in Exercise 5.3.5.

The final section of the chapter will explore the relation between torus-invariant
divisors on a toric variety X and its total coordinate ring. We will also see that
when X comes from a polytope P, the quotient construction of X relates nicely
to the definition of projective toric variety given in Chapter 2.
232
Homogenization. When working with affine and projective space, one often needs
to homogenize polynomials. This process generalizes nicely to the toric context.
The full story involves characters, polyhedra, divisors, sheaves, and graded pieces
of the total coordinate ring.
P
A Weil divisor D = a D on X gives the polyhedron
Proposition 4.3.3 tells us that the global sections of the sheaf OX (D) are spanned
by characters coming from lattice points of PD , i.e.,
M
(X , OX (D)) =
C m.
mPD M
This relates to the total coordinate ring S = C[x | (1)] as follows. Given
m PD M, the D-homogenization of m is the monomial
Y hm,u i+a
xhm,Di =
x
defined in (5.3.3). The inequalities defining PD guarantee that xhm,Di lies in S. Here
are the basic properties of these monomials.
Proposition 5.4.1. Assume that X has no torus factors. If D and PD are as above
and = [D] Cl(X ) is the divisor class of D, then:
(a) For each m PD M, the monomial xhm,Di lies in S .
(b) The map sending the character m of m PD M to the monomial xhm,Di

induces an isomorphism
(X , OX (D)) S .
Proof. Part (a) follows from the proof of Proposition 5.3.7. As for part (b), we use
the same proposition to conclude that
(X , OX (D)) (X , OX ()) S .
One easily sees that this isomorphism is given by m 7 xhm,Di .
Here are some examples of homogenization.

P
Example 5.4.2. The fan for P n has ray generators u0 = ni=1 ei and ui = ei for
i = 1, . . . , n. This gives variables xi and divisors Di for i = 0, . . . , n. Since M = Zn ,
Q
the character of m = (b1 , . . . , bn ) Zn is the Laurent monomial t m = ni=1 tibi .
233
For a positive integer d, the divisor D = d D0 has polyhedron PD = dn , where

n is the standard n-simplex. Given m = (b1 , . . . , bn ) dn , its homogenization is
hm,u0 i+d hm,u1 i+0
n i+0
xhm,u
x1
n
= x0b1 bn +d x1b1 xnbn
x b1 x bn
n
1
,
= x0d
x0
x0
Q
which is the usual way to homogenize t m = ni=1 tibi with respect to x0 .
This monomial has degree d = [dD0 ] Cl(P n ) = Z, in agreement with Proposi-
xhm,Di = x0
tion 5.4.1. The proposition also implies the standard fact that monomials of degree
d in x0 , . . . , xn correspond to lattice points in dn .
Example 5.4.3. For P1 P1 , we have ray generators u1 = e1 , u2 = e1 , u3 =

e2 , u4 = e2 with corresponding variables xi and divisors Di . Given nonnegative
integers k, , we get the divisor D = k D2 + D4 . The polyhedron PD is the rectangle with vertices (0, 0), (k, 0), (0, ), (k, ), and given (a, b) PD Z2 , the Laurent
monomial t1a t2b homogenizes to
x a x b
1
3
x1a x2ka x3b x4b = x2k x4
,
x2
x4
which is the usual way of turning a two-variable monomial into a bihomogeneous monomial of degree (k, ) (remember that deg(x1 ) = deg(x2 ) = (1, 0) and
deg(x3 ) = deg(x4 ) = (0, 1)). Thus monomials of degree (k, ) correspond to lattice
points in the rectangle PD .
Example 5.4.4. The fan for Bl0 (C2 ) is shown in Example 5.1.16, and its total
coordinate ring S = C[t, x, y] is described in Example 5.2.3. If we pick D = 0, then
the polyhedron PD R2 is defined by the inequalities
hm, ui i 0,
i = 0, 1, 2.
Since u1 , u2 form a basis of N = Z2 and u0 = u1 + u2 , PD is the first quadrant in R2 .

Given m = (a, b) PD Z2 , the monomial t1a t2b homogenizes to
t hm,u0 i xhm,u1 i yhm,u2 i = t a+b x a y b = (tx)a (ty)b .
where the ray generators u0 , u1 , u2 correspond to the variables t, x, y.

For example, the singular cubic t13 t22 = 0 homogenizes to (tx)3 (ty)2 = 0,
which is the equation enountered in Example 5.2.11 when resolving the singularity
of this curve.
One thing to keep in mind when doing toric homogenization is that characters
m (in general) or Laurent monomials t m (in specific examples) are intrinsically
defined on the torus TN or (C )n . The homogenization process produces a global
object xhm,Di relative to a divisor D that lives in the total coordinate ring or, via
Proposition 5.4.1, in the global sections of OX (D).
234
We next study the isomorphisms S (X , OX (D)) from Proposition 5.4.1.

We will see that they are compatible with linear equivalence and multiplication.
First suppose that D and E are linearly equivalent torus-invariant divisors. This
means that D = E + div( m ) for some m M. Proposition 4.0.29 implies that
f 7 f m induces an isomorphism
(5.4.1)
(X , OX (D)) (X , OX (E)).
Turning to the associated polyhedra, we proved PE = PD + m in Exercise 4.3.2. An

easy calculation shows that if m PD , then
xhm ,Di = xhm +m,Ei
(Exercise 5.4.1). Hence (5.4.1) fits into a commutative diagram of isomorphisms

(X , OX (D))
(5.4.2)
OOO
O
OOO'
S .
/ (X , OX (E))
oo
o
o
wooo
Here, = [D] = [E] Cl(X ) and the diagonal maps are the isomorphisms from
Proposition 5.4.1. You will verify these claims in Exercise 5.4.1.
It follows that S gives a canonical model for (X , OX (D)), since the latter
depends on the particular choice of divisor D in the class . It is also possible to
give a canonical model for the polyhedron PD (Exercise 5.4.2).
Next consider multiplication. Let D and E be torus-invariant divisors on X
and set = [D], = [E] in Cl(X ). Then f g 7 f g induces a C-linear map
(X , OX (D)) C (X , OX (E)) (X , OX (D + E))
such that the isomorphisms of Proposition 5.4.1 give a commutative diagram
(5.4.3)
(X , OX (D)) C (X , OX (E))
/ (X , OX (D + E))

/ S+
S C S
where the bottom map is multiplication in the total coordinate ring (Exercise 5.4.3).
Thus homogenization turns multiplication of sections into ordinary multiplication.
Polytopes. A full dimensional lattice polytope P MR gives a toric variety XP .
Recall that XP can be constructed in two ways:
As the toric variety XP of the normal fan P of P (Chapter 3).
As the projective toric variety X(kP)M of the set of characters (kP) M for
k 0 (Chapter 2).
We will see that both descriptions relate nicely to homogeneous coordinates and
the total coordinate ring.
235
Given P as above, set n = dim P and let P(i) denote the set of i-dimensional
faces of P. Thus P(0) consists of vertices and P(n 1) consists of facets. The facet
presentation of P given in equation (2.2.2) can be written as
(5.4.4)
P = {m MR | hm, uF i aF for all F P(n 1)}.
In terms of the normal fan P , we have bijections

P(0) P (n)
P(n 1) P (1)
(vertices maximal cones)
(facets rays).
When dealing with polytopes we index everything by facets rather than rays. Thus
each facet F P(n 1) gives:
The facet normal uF , which is the ray generator of the corresponding cone.
The torus-invariant prime divisor DF XP .
The variable xF in the total coordinate ring S. We call xF a facet variable.
We also have the divisor
DP =
aF DF
from (4.2.7). The polytope PDP of this divisor is the polytope P we began with
(Exercise 4.3.1). Hence, if we set = [DP ] Cl(XP ), then we get isomorphisms
M
S (XP , OXP (DP ))
C m.
mPM
In this situation, we write the DP -homogenization of m as

Y hm,u i+a
xhm,Pi =
xF F F .
F
We call xhm,Pi a P-monomial.

The exponent of the variable xF in xhm,Pi gives the lattice distance from m
to the facet F. To see this, note that F lies in the supporting hyperplane defined
by hm, uF i + aF = 0. If the exponent of xF is a 0, then to get from the supporting hyperplane to m, we must pass through the a parallel hyperplanes, namely
hm, uF i + aF = j for j = 1, . . . , a. Here is an example.
Example 5.4.5. Consider the toric variety XP of the polygon P R2 with vertices
(1, 1), (1, 1), (1, 0), (0, 1), (1, 1), shown in Figure 2 on the next page. In
terms of (5.4.4), we have a1 = = a5 = 1, where the indices correspond to the
facet variables x1 , . . . , x5 indicated in Figure 2. The 8 points of P Z2 give the
P-monomials
x2 x32 x42
x1 x22 x32 x4
x12 x23 x32
2
x3 x4 x5 x1 x2 x3 x4 x5 x12 x22 x3 x5
x1 x4 x52
x12 x2 x52 ,
236
x5
x4
x1
@
x2 @
@
x3
Figure 2. A polygon with facets labeled by variables
where the position of each P-monomial xhm,Pi corresponds to the position of the
lattice point m P Z2 . The exponents are easy to understand if you think in
terms of lattice distances to facets.
The lattice-distance interpretation of the exponents in xhm,Pi shows that lattice

points
in the interior int(P) of P correspond to those P-monomials divisible by
Q
x
.
F F For example, the only P-monomial in Example 5.4.5 divisible by x1 x5
corresponds to the unique interior lattice point.
We next relate the constructions of toric varieties given in Chapter 2 and in

5.1. In Chapter 2, we wrote the lattice points of P as P M = {m1 , . . . , ms } and
considered the map
(5.4.5)
: TN P s1 ,
t 7 ( m1 (t), . . . , ms (t)).
The projective (possibly non-normal) toric variety XPM is the Zariski closure of
the image of .
On the other hand, we have the quotient construction of XP

XP Cr \ Z(P ) //G,
where we write Cr = CP (1) . Also, the exceptional set Z(P ) can be described in
terms of the P-monomials coming from the vertices of the polytope.
Lemma 5.4.6. The vertex monomials xhv,Pi , v a vertex of P, have the following
properties:
p
(a) hxhv,Pi | v P(0)i = B(P ), where B(P ) = hx | (n)i is the irrelevant
ideal of S.
(b) Z(P ) = V(xhv,Pi | v P(0)).
Proof. We saw above that vertices v P(0) correspond bijectively to cones v =

Cone(uF | v F) P (n). Then the lattice-distance interpretation of xhv,Pi shows
the facet variables xF appearing in xhv,Pi are precisely the variables appearing in

x v . This implies part (a), and part (b) follows immediately.
237
If we set = [DP ] as above, then the P-monomials xhmi ,Pi , i = 1, . . . , s, form a

basis of S and give a map
(5.4.6)
: Cr \ Z(P) P s1
p 7 (phm1 ,Pi , . . . , phms ,Pi ),
where phmi ,Pi is the evaluation of the monomial xhmi ,Pi at the point p Cr \ Z(P ).
This map is well-defined since for each p Cr \ Z(P ), Lemma 5.4.6 implies that
at least one P-monomial (in fact, at least one vertex monomial) must be nonzero.
The maps (5.4.5) and (5.4.6) fit into a diagram
(C )r
/ Cr \ Z(P )
::
:
::
::

::

/ XP
::
TN VVVV
VVVV
::
VVVV
:
VVVV
VVVV :::
VVV* &
P s1 .
Here, the map (C )r TN is from (5.1.2) and : Cr \ Z(P ) XP is the quotient

map. This diagram has the following properties.
Proposition 5.4.7. There is a morphism : XP P s1 represented by the dotted
arrow in the above diagram that makes the entire diagram commute. Furthermore,
the image of is precisely the projective toric variety XPM .
Proof. When we regard the xF as characters on (C )r = (C )P (1) , the exact sequence (5.1.1) tells us that
Y hm,u i
(5.4.7)
m =
xF F
F
for m M. Multiplying each side by F xFaF , we obtain

Y
xFaF m = xhm,Di .
F
If we let m = mi , i = 1, . . . , s and apply this to a point in p (C )r , we see that (p)

and
give the same point in projective space since the vector for (p) equals
Q (p)
aF
F pF times the vector for (p). It follows that, ignoring for the moment, the
rest of the above diagram commutes.
We next show that is constant on G-orbits. This holds since P-monomials

are homogeneous of the same degree. In more detail, fix points t = (tF ) G,
Q hm,u i+a
p = (pF ) Cr \Z(P ) and a P-monomial xhm,Di = F xF F F . Then evaluating
238
xhm,Di at t p gives
(t p)hm,Di =
=
Y
(tF pF )hm,uF i+aF
F
Y
F
hm,u i
tF F
Y
F
tFaF
hm,Di
Y
tFaF
phm,Di ,
where the last equality follows from the description of G given in Lemma 5.1.1.
Arguing as in the previous paragraph, it follows that (t p) and (p) give the
same point in P s1 . This proves the existence of since is a good categorical
quotient, and this choice of makes the entire diagram commute.
The final step is to show that the image of : XP P s1 is the Zariski closure
XPM of the image of : TN P s1 . First observe that
(XP ) = (TN ) (TN ) = (TN ) = XPM
since is continuous in the Zariski topology and |TN = by commutativity of the
diagram. However, (XP ) is Zariski closed in P s1 since XP is projective. You will
give two proofs of this in Exercise 5.4.4, one topological (using constructible sets
and compactness) and one algebraic (using completeness and properness). Once
we know that (XP ) is Zariski closed, (TN ) (XP ) implies
XPM = (TN ) (XP ),
and (XP ) = XPM follows.
In Chapter 2, we used the map , constructed from characters, to parametrize

a big chunk of the projective toric variety XPM . In contrast, Proposition 5.4.7 uses
the map , constructed from P-monomials, to parametrize all of XPM .
If the lattice polytope P is very ample, then the results of Chapter 2 imply that
XPM is the toric variety XP . So in the very ample case, the P-monomials give an
explicit construction of the quotient Cr \ Z(P ) //G by mapping Cr \ Z(P ) to
projective space via the P-monomials. It follows that we have two ways to take the
quotient of Cr by G:
At the beginning of the chapter, we took G-invariant polynomialselements
of S0 to construct an affine quotient.
Here, we use P-monomialselements of S to construct a projective quotient, after removing a set Z(P ) of bad points.
The P-monomials are not G-invariant but instead transform the same way under G.
This is why we map to projective space rather than affine space. We will explore
these ideas further in Chapter 14 when we discuss geometric invariant theory.
239
When P is very ample, we have a projective embedding XP P s1 given by

the P-monomials in S . If y1 , . . . , ys are homogeneous coordinates of P s1 , then
the homogeneous coordinate ring of XP P s1 is
C[XP ] = C[y1 , . . . , ys ]/I(XP )
as in 2.0. We also have the affine cone XbP Cs of XP , and C[XP ] is the ordinary
coordinate ring of XbP , i.e.,
C[XP ] = C[XbP ].
Recall that C[XP ] is an N-graded ring since I(XP ) is a homogeneous ideal.
L
Another N-graded ring is
k=0 Sk . This relates to C[XP ] as follows.
Theorem 5.4.8. Let P be a very ample lattice polytope with = [DP ] Cl(XP ).
Then:
L
(a)
k=0 Sk is normal.
L
L
(b) There is a natural inclusion C[XP ]
k=0 Sk such that
k=0 Sk is the normalization of C[XP ].
(c) The following are equivalent:
(1) XP P s1 is projectively normal.
(2) P is normal.
L
(3)
k=0 Sk = C[XP ].
L
(4)
k=0 Sk is generated as a C-algebra by its elements of degree 1.
Proof. Consider the cone

This cone is pictured in Figure 4 of 2.2. Recall that kP is the slice of C(P)
at height k. Since the divisor DkP associated to kP is kDP , homogenization with
respect to kP induces an isomorphism
M
Sk (XP , OXP (kDP ))
C m.
m(kP)M
Now consider the dual cone P = C(P) NR R. The semigroup algebra

C[C(P) (M Z)] is the coordinate ring of the affine toric variety UP . Given
(m, k) C(P) (M Z), we write the corresponding character as mt k .
The algebra C[C(P)(M Z)] is graded using the last coordinate, the height.
Since (m, k) C(P) (M Z) if and only if m kP (this is the slice observation
made above), we have
M
C[C(P) (M Z)]k =
C mt k .
m(kP)M
240
Using (5.4.3), we obtain a graded C-algebra isomorphism
This proves that
M
k=0
k=0 Sk
Sk C[C(P) (M Z)].
is normal.
We next claim that UP is the normalization of the affine cone XbP . For this,
we let A = (P M) {1} M Z. As noted in the proof of Theorem 2.4.1, the
affine cone of XP = XPM is XbP = YA . Since P is very ample, one easily checks
that A generates M Z, i.e., ZA = M Z (Exercise 5.4.5). It is also clear that A
generates the cone C(P) = P . Hence UP is the normalization of XbP by Proposition 1.3.8. This immediately implies part (b).
For part (c), we observe that (1) (2) follows from Theorem 2.4.1, and (1)
(3) follows from parts (a) and (b) since the projective normality of XP P s1
is equivalent to the normality of C[XP ]. Also (3) (4) is obvious since C[XP ] is
generated by the images of y1 , . . . , ys , which have degree 1. Finally, you will show
in Exercise 5.4.6 that (4) (2), completing the proof.

Further Examples. We begin with an example of that illustrates how there can be
many different polytopes that give the same toric variety.
Example 5.4.9. The toric surface in Example 5.4.5 was defined using the polygon
shown in Figure 2. In Figure 3 we see four polygons A, A B, A C, A D, all
B
@
@
@
@
@D @
@
@
@
Figure 3. Four polygons A, A B, A C, A D with the same normal fan
of which have the same normal fan and hence give the same toric variety. Since
we are in dimension 2, these polygons are very ample (in fact, normal), so that
Theorem 5.4.8 applies.
These four polygons give four different projective embeddings, each of which
has its own homogeneous coordinate ring as a projective variety. By Theorem 5.4.8,
these homogeneous coordinate rings all live in the total coordinate ring S. This explains the total in total coordinate ring.
241
Our next example involves torsion in the grading of the total coordinate ring.
P
Example 5.4.10. The fan for P 4 has ray generators u0 = 4i=1 ei and ui = ei
for i = 1, . . . , 4 in N = Z4 and is the normal fan of the standard simplex 4 R4 .
Another polytope with the same normal fan is
P = 54 (1, 1, 1, 1) MR = R4 ,
so that XP = P 4 . We saw that P is reflexive in Example 2.4.5. One checks that

DP = D0 + + D4 has degree 5 Z Cl(P 4 ). Since P is a translate of 54 ,
(5.4.2) implies that the P-monomials for m P Z4 coincide with the homogenizations coming from 54 , which are homogeneous polynomials of degree 5 in
S = C[x0 , . . . , x4 ].
Since P is reflexive, its dual P is also a lattice polytope. Furthermore,
P = Conv(u0 , . . . , u4 ) NR = R4
since the ray generators of the normal fan of P are the vertices of P by duality for
reflexive polytopes (be sure you understand thisExercise 5.4.7). The vertices of
P are
v0 = (1, 1, 1, 1), v1 = (4, 1, 1, 1), v2 = (1, 4, 1, 1)
(5.4.8)
v3 = (1, 1, 4, 1), v4 = (1, 1, 1, 4).
The vi generate a sublattice M1 M = Z4 . In Exercise 5.4.7 you will show that the
map M Z5 defined by
m M 7 (hm, u0 i, . . . , hm, u4 i) Z5

P
(5.4.9)
M/M1 (a0 , a1 , a2 , a3 , a4 ) (Z/5Z)5 : 4i=0 ai = 0 /(Z/5Z)
where Z/5Z (Z/5Z)5 is the diagonal subgroup. Then M/M1 (Z/5Z)3 , so that
M1 is a lattice of index 125 in M.
The dual toric variety XP is determined by the normal fan of P . The
ray generators of are the vectors v0 , . . . , v4 from (5.4.8). The only possible
complete fan in R4 with these ray generators is the fan whose cones are generated
by all proper subsets of {v0 , . . . , v4 }. Since v0 + + v4 = 0 and the vi generate M1 ,
the toric variety of relative to M1 is P 4 , i.e., X , M1 = P 4 . (Remember that
is a fan in (M1 )R = MR .) Since M1 M has index 125, Proposition 3.3.7 implies
XP = XP , M XP , M1 /(M/M1 ) = P 4 /(M/M1 ).
Hence the dual toric variety XP is the quotient of P 4 by a group of order 125.
The total coordinate ring S is the polynomial ring C[y0 , . . . , y4 ], graded by
Cl(XP ). The notation is challenging, since by duality N is the character lattice of
the torus of XP . Thus (5.1.1) becomes the short exact sequence
0 N Z5 Cl(XP ) 0,
242
where N Z5 is u 7 (hv0 , ui, . . . , hv4 , ui). If we let N1 = HomZ (M1 , Z), then
M1 M dualizes to N N1 of index 125. Now consider the diagram
0
0

/N
/ Z5
/ Cl(XP )
/0

/ N1

/ Z5

/Z
/0
N1 /N

0
with exact rows and columns. In the middle row, we use Cl(X ,M1 ) = Cl(P 4 ) = Z.
By the snake lemma, we obtain the exact sequence
0 N1 /N Cl(XP ) Z 0,
so Cl(XP ) Z N/N1. Thus the class group has torsion.
The polytope P has only six lattice points in N: the vertices u0 , . . . , u4 and the
origin (Exercise 5.4.7). When we homogenize these, we get six P -monomials
yh0,Di =
4
Y
hv j ,0i+1
= y0 y4
hv j ,ui i+1
= yi5 ,
yj
j=0
yhui ,Di =
4
Y
yj
i = 0, . . . , 4
j=0
since hv j , ui i = 5i j 1 (Exercise 5.4.7).
The equation
c0 y05 + + c4 y45 + c5 y0 y4 = 0
defines a hypersurface Y XP since it is built from P -monomials. If we want an
irreducible hypersurface, we must have c0 , . . . , c4 6= 0, in which case Y is isomorphic (via the torus action) to a hypersurface of the form
y05 + + y45 + y0 y4 = 0.
This is the quintic mirror family, which played a crucial role in the development of
mirror symmetry. See [68] for an introduction to this astonishing subject.
Exercises for 5.4.
5.4.1. Let D, E be linearly equivalent torus-invariant divisors with D = div( m ) + E.
(a) If m PD M, then prove that xhm ,Di = xhm +m,Ei .

(b) Prove (5.4.2).
243
P
5.4.2. Fix a torus-invariant divisor D = a D and consider its associated polyhedron
PD = {m MR | hm, u i a for all }. Define
D : MR R(1)
by D (m) = (hm, u i + a ) R(1) .
(a) Prove that D embeds MR as an affine subspace of R(1) . Hint: Remember that X
has no torus factors.
(b) Prove that D induces a bijection

(1)
D |PD : PD D (MR ) R0 .
(1)
This realizes PD as the polyhedron obtained by intersecting the positive orthant R0

of R(1) with an affine subspace.
(c) Let D = div( m ) + E. Prove that D (PD ) = E (PE ). Thus the polyhedron in R(1)
constructed in part (b) depends only on the divisor class of D. This is the canonical
model of PD .
5.4.3. Prove that the diagram (5.4.3) is commutative.
5.4.4. The proof of Proposition 5.4.7 claimed that the image of : XP P s1 was Zariski
closed. This follows from the general fact that if : X Y is a morphism of varieties and
X is complete, then (X) is Zariski closed in Y . You will prove this two ways.
(a) Give a topological proof that uses constructible sets and compactness. Hint: Remember that projective space is compact.
(b) Give an algebraic proof that uses completeness and properness from 3.4. Hint: Show
that X Y Y is proper and use the graph of .
5.4.5. Let P MR be a very ample lattice polytope and let A = (P M) {1} M Z.

Prove that ZA = M Z. Hint: First show that ZA = M {0}, where ZA is defined in
the discussion preceding Proposition 2.1.6.
5.4.6. Prove of (4) (2) in part (c) of Theorem 5.4.8. Hint: (4) implies that the map
S C Sk S(k+1) is onto for all k 0.
(a) Prove that if P Rn is reflexive, then the vertices of P are the ray generators of the
normal fan of P .
(b) Prove (5.4.9).
(c) Prove hv j , ui i = 5i j 1, where v j , ui are defined in Example 5.4.10.
(d) Let G = HomZ (Cl(XP ), C ) (C )5 . Use Proposition 1.3.18 to prove
G = {(0 , . . . , 4 ) | C , i 5 , 0 4 = 1} C M/M1 .
(e) Use part (e) and the quotient construction of XP to give another proof that XP =
P 4 /(M/M1 ). Also give an explicit description of the action of M/M1 on P 4 .
5.4.8. This exercise will give another way to think about homogenization. Let e1 , . . . , en
be a basis of M, so that ti = ei , i = 1, . . . , n, are coordinates for the torus TN .
Q he ,u i
(a) Adapt the proof of (5.4.7) to show that ti = x i when we think of the x as
characters on (C )(1) .
244
(b) Given m PD M, part (a) tells us that the Laurent monomial t m can be regarded as a
Laurent
Q amonomial in the x . Show that we can clear denominators by multiplying
by x to obtain a monomial in the polynomial ring S = C[x | (1)].
(c) Show that this monomial obtained in part (b) is the homogenization xhm,Di .
5.4.9. Consider the toric variety XP of Example 5.4.5.

(a) Compute Cl(XP ) and find the classes of the four polygons appearing in Figure 3.
(b) Show that XP is the blowup of P1 P1 at one point.
5.4.10. Consider the reflexive polytope P = 43 (1, 1, 1) R3 . Work out the analog of
Example 5.4.10 for P.
5.4.11. Fix an integer a 1 and consider the 3-simplex P = Conv(0, ae1 , ae2 , e3 ) R3 .
In Exercise 2.2.13, we claimed that the toric variety of P is the weighted projective space
P(1, 1, 1, a). Prove this.
Pn
5.4.12. Consider positive integers 1 = q0 q1 qn with the property that qi | j=0 q j

Pn
for i = 0, . . . , n. Set ki =
j=0 q j /qi for i = 1, . . . , n and let
Pq0 ,...,qn = Conv(0, k1 e1 , k2 e2 , . . . , kn en ) (1, . . . , 1) Rn .
This lattice polytope is reflexive by Exercise 2.4.6. Prove that the associated toric variety
is the weighted projective space P(q0 , q2 , . . . , qn ).
Chapter 6
Line Bundles on
Toric Varieties
6.0. Background: Sheaves and Line Bundles

Sheaves of OX -modules on a variety X were introduced in 4.0. Recall that for
e on V such that
an affine variety V = Spec(R), an R-module M gives a sheaf M
e f ) = M f for all f 6= 0 in R. Globalizing this leads to quasicoherent sheaves
M(V
on X . These include coherent sheaves, which locally come from finitely generated
modules. In this section we develop the language of sheaf theory and discuss vector
bundles and line bundles.
The Stalk of a Sheaf at a Point. Since sheaves are local in nature, we need a
method for inspecting a sheaf at a point p X . This is provided by the notion of
direct limit over a directed set.
Definition 6.0.1. A partially ordered set (I, ) is a directed set if
for all i, j I, there exists k I such that i k and j k.
If {Ri } is a family of rings indexed by a directed set (I, ) such that whenever i j
there is a homomorphism
ji : Ri R j
satisfying ii = 1Ri and
L k j ji = ki , then the Ri form a directed system. Let S
be the submodule of iI Ri generated by the relations r j ji (ri ), for ri Ri and
i j. Then the direct limit is defined as

L
Ri =
lim
iI Ri /S.
iI
245
Chapter 6. Line Bundles on Toric Varieties
246
For simplicity, we often write the direct limit as lim Ri . Note also that references
such as [10] write i j instead of ji .

For every i I, there is a natural map Ri lim Ri such that whenever i j,
the elements r Ri and ji (r) R j have the same image in lim Ri . More generally,
two elements ri Ri and r j R j are identified in lim Ri if there is a diagram
Ri QQQki
QQQ
(
Rj
6 Rk
mmm
mmmk j
such that ki (ri ) = k j (r j ).

Example 6.0.2. Given p X , the definition of sheaf shows that the rings OX (U ),
indexed by neighborhoods U of p, form a directed system under inclusion, so that
the ji are the restriction maps U,U for p U U . The direct limit is the local
ring OX,p . For a quasicoherent sheaf F , take an affine open subset V = Spec(R)
containing p so that F (V ) = M, where M is an R-module. If m p = I(p) R is the
corresponding maximal ideal, then OX,p is the localization Rmp , and
F (U ) = Mmp ,
lim
pU
where Mmp is the localization of M at the maximal ideal m p .
The term sheaf has agrarian origins: farmers harvesting their wheat tied a rope
around a big bundle, and left it standing to dry. Think of the footprint of the bundle
as an open set, so that increasingly smaller neighborhoods around a point on the
ground pick out smaller and smaller bits of the bundle, narrowing to a single stalk.
F (U ).
Definition 6.0.3. The stalk of a sheaf F at a point p X is F p = lim
pU
Injective and Surjective. A homomorphism : F G of OX -modules was defined in 4.0. We can also define what it means for to be injective or surjective.
The definition is a bit unexpected, since we need to take into account the fact that
sheaves are built to convey local data.
Definition 6.0.4. A sheaf homomorphism
: F G
is injective if for any point p X and open subset U X containing p, there exists
an open subset V U containing p, with V injective. Also, is surjective if for
any point p and open subset U containing p and any g G (U ), there is an open
subset V U containing p and f F (V ) such that V ( f ) = U,V (g).
247
In Exercise 6.0.1 you will prove that for a sheaf homomorphism : F G ,

U 7 ker(U : F (U ) G (U ))
defines a sheaf denoted ker(). You will also show that is injective exactly when
the naive idea works, i.e., ker() = 0. On the other hand, surjectivity of a sheaf
homomorphism need not mean that the maps U are surjective for all U . Here is
an example.
Example 6.0.5. On P1 = C {}, consider the Weil divisor D = {0} C P1 .
If we write of P1 = U0 U1 with U0 = Spec(C[t]) and U1 = Spec(C[t 1 ]), then
C(P1 ) = C(t). Since
(P1 , OP1 (D)) = { f C(t) | div( f ) + D 0} {0},
it follows easily that we have global sections
1,t 1 (P1 , OP1 (D)).
For any f (P1 , OP1 (D)), multiplication by f gives a sheaf homomorphism
OP1 (D) OP1 . Doing this for 1,t 1 (P1 , OP1 (D)) gives
OP1 (D) OP1 (D) OP1 .
(Direct sums of sheaves will be defined below.) In Exercise 6.0.2 you will check
that this sheaf homomorphism is surjective. However, taking global sections gives
0 0 = (P1 , OP1 (D)) (P1 , OP1 (D)) (P1 , OP1 ) = C,
which is clearly not surjective.

There is an additional point to make here. Given : F G , the presheaf
U 7 im(U : F (U ) G (U ))
need not be a sheaf. Fortunately, this can be rectified. Given a presheaf F , there is
an associated sheaf F + , the sheafification of F , which is defined by
Q
F + (U ) = { f : U pU F p | for all p U , f (p) F p and there is
p Vp U and t F (Vp ) with f (x) = t p for all x Vp }.
See [131, II.1] for a proof that F + is a sheaf with the same stalks as F p . Hence
U 7 im(U )
has a natural sheaf associated to it, denoted im().
248
Exactness. We define exact sequences of sheaves as follows.

Definition 6.0.6. A sequence of sheaves
d i1
di
F i1 F i F i+1
is exact at F i if there is an equality of sheaves
ker(d i ) = im(d i1 ).
The local nature of sheaves is again highlighted by the following result, whose
proof may be found in [131, II.1].
Proposition 6.0.7. The sequence in Definition 6.0.6 is exact if and only if
d pi1
d pi
F pi1 F pi F pi+1
is exact for all p X .
It follows from Example 6.0.5 that if

d1
d2
0 F 1 F 2 F 3 0
(6.0.1)
is a short exact sequence of sheaves, the corresponding sequence of global sections

may fail to be exact. However, we always have the following partial exactness,
which you will prove in Exercise 6.0.3.
Proposition 6.0.8. Given a short exact sequence of sheaves (6.0.1), taking global
sections gives the exact sequence
d1
d2
0 (X , F 1 ) (X , F 2 ) (X , F 3 ).
In Chapter 9 we will use sheaf cohomology to extend this exact sequence.
Example 6.0.9. For an affine variety V = Spec(R), an R-module M gives a quasie on V . This operation preserves exactness, i.e., an exact sequence
coherent sheaf M
of R-modules
0 M1 M2 M3 0
gives an exact sequence of sheaves
(see [131, Prop. II.5.2]).
e1 M
e2 M
e3 0
0 M
Here is a toric generalization of this example.

Example 6.0.10. Let S = C[x | (1)] be the total coordinate ring of a toric
variety X without torus factors. We saw in 5.3 that a graded S-module M gives
e on X .
the quasicoherent sheaf M
249
Then an exact sequence 0 M1 M2 M3 0 of graded S-modules gives

an exact sequence
e 1 M
e 2 M
e 3 0
0 M
e i to U X is the
on X . To see why, note that for , the restriction of M
sheaf associated to ((Mi )x )0 , the elements of degree 0 in the localization of Mi at
x S. Localization preserves exactness, as does taking elements of degree 0. The
desired exactness then follows from Example 6.0.9.
Another example is the following exact sequence of sheaves from 3.0.

Example 6.0.11. A closed subvariety i : Y X gives two sheaves:
The sheaf IY , defined by IY (U ) = { f OX (U ) | f (p) = 0 for p Y U }.
The direct image sheaf i OY , defined by i OY (U ) = OY (Y U ).
These are coherent sheaves on X and are related by the exact sequence
0 IY OX i OY 0.
Operations on Quasicoherent Sheaves of OX . Operations on modules over a ring

have natural analogs for quasicoherent sheaves. In particular, given quasicoherent
sheaves F , G , it is easy to show that U 7 F (U ) G (U ) defines the quasicoherent
sheaf F G . We can also define HomOX (F , G ) via
U 7 HomOX (U) (F (U ), G (U )).
In Exercise 6.0.4 you will show that HomOX (F , G ) is a quasicoherent sheaf.
On the other hand, U 7 F (U ) OX (U) G (U ) is only a presheaf, so the tensor

product F OX G is defined to be the sheaf associated to this presheaf. This sheaf
is again quasicoherent and satisfies
(U , F OX G ) = F (U ) OX (U) G (U )
whenever U X is an affine open set (see [131, Prop. II.5.2]).

Global Generation. For a module M over a ring, there is always a surjection from
a free module onto M. This is true for a sheaf F of OX -modules when (X , F ) is,
in a certain sense, large enough.
Definition 6.0.12. A sheaf F of OX -modules is generated by global sections if
there exists a set {si } (X , F ) such that at any point p X , the images of the si
generate the stalk F p .
Any global section s (X , F ) gives a sheaf homomorphism OX F . It
follows that if F is generated by {si }iI , there is a surjection of sheaves
M
OX F .
iI
In the next section we will see that when X is toric, there is a particularly nice way
of determining when the sheaves OX (D) are generated by global sections.
250
Locally Free Sheaves and Vector Bundles. We begin with locally free sheaves.
Definition 6.0.13. A sheaf F of OX -modules is locally free of rank r if there
exists an open cover {U } of X such that for all , F |U OUr .
Locally free sheaves are closely related to vector bundles.
Definition 6.0.14. A variety V is a vector bundle of rank r over a variety X if there
is a morphism
: V X
and an open cover {Ui } of X such that:
(a) For every i, there is an isomorphism
i : 1 (Ui ) Ui Cr
such that i followed by projection onto Ui is |1 (Ui ) .
(b) For every pair i, j, there is gi j GLr ((Ui U j , OX )) such that the diagram
i | 1
(Ui U j )
Ui U j Cr
k5
kkkk
kkkk
1 (Ui U j )
1gi j
SSSS
SSSS
S)
j | 1
(Ui U j )
Ui U j Cr
commutes.
Data {(Ui , i )} satisfying properties (a) and (b) is called a trivialization. The
map i : 1 (Ui ) Ui Cr gives a chart, where 1 (p) Cr for p Ui . We call
1 (p) the fiber over p. See Figure 1 on the next page.
For p Ui U j , the isomorphisms
Cr {p} Cr 1 (p) {p} Cr Cr
given by i and j are related by the linear map gi j (p). Hence the fiber 1 (p)
has a well-defined vector space structure. This shows that a vector bundle really is
a bundle of vector spaces.
On a vector bundle, the gi j are called transition functions and can be regarded
as a family of transition matrices that vary as p Ui U j varies. Just as there
is no preferred basis for a vector space, there is no canonical choice of basis for
a particular fiber. Note also that the transition functions satisfy the compatibility
conditions
gik = gi j g jk on Ui U j Uk
(6.0.2)
gi j = g1
on Ui U j .
ji
251
j : 1(p)
{p}Cr
i : 1(p)
{p}Cr
gij(p)
Uj Cr
Ui Cr
Ui
Uj
Figure 1. Visualizing a vector bundle
Definition 6.0.15. A section of a vector bundle V over U X open is a morphism

s : U V
such that ( s)(p) = p for all p U . A section s : X V is a global section.

A section s picks out a point s(p) in each fiber 1 (p), as shown in Figure 2.
1( p)
s(p)
s(x)
X
p
Figure 2. For a section s, s(p) 1 (p)
252
We can describe a vector bundle and its global sections purely in terms of the
transition functions gi j as follows.
Proposition 6.0.16. Let X be a variety with an affine open cover {Ui }, and assume
that for every i, j, we have gi j GLr ((Ui U j , OX )) satisfying the compatibility
conditions (6.0.2). Then:
(a) There is a vector bundle : V X of rank r, unique up to isomorphism, whose
transition functions are the gi j .
(b) A global section s : X V is uniquely determined by a collection of r-tuples
si OXr such that for all i, j,
si |Ui U j = gi j s j |Ui U j .
Proof. One easily checks that the g1
i j satisfy the gluing conditions from 3.0. It
follows that the affine varieties Ui Cr glue together to give a variety V . Furthermore, the projection maps Ui Cr Ui glue together to give a morphism
: V X . It follows easily that the open set of V corresponding to Ui Cr is
1 (Ui ), which gives an isomorphism i : 1 (Ui ) Ui Cr . Hence V is a vector
bundle with transition functions gi j .
Given a section s : X V , i s|Ui is a section of Ui Cr Ui . Thus
i s|Ui (p) = (p, si (p)) Ui Cr ,
where si OX (Ui )r . By Definition 6.0.14, the si satisfy the desired compatibility

condition, and since every global section arises this way, we are done.

Let F (U ) denote the set of all sections of V over U . One easily sees that F is
a sheaf on X and in fact is a sheaf of OX -modules since the fibers are vector spaces.
In fact, F is an especially nice sheaf.
Proposition 6.0.17. The sheaf of sections of a vector bundle is locally free.
Proof. For a trivial vector bundle U Cr U , the proof of Proposition 6.0.16
shows that a section is determined by a morphism U Cr , i.e., an element of
OU (U )r . Thus the sheaf associated to a trivial vector bundle over U is OUr .
For a general vector bundle : V X with trivialization {(Ui , i )}, each Ui
gives an isomorphism of vector bundles
1 (Ui )
99
99
99
| 1
(Ui ) 9
/ Ui Cr

Ui .
Since isomorphic vector bundles have isomorphic sheaves of sections, it follows

that if F is the sheaf of sections of : V X , then F |Ui OUr i .
253
Line Bundles and Cartier Divisors. Since a vector space of dimension one is a
line, a vector bundle of rank 1 is called a line bundle. Despite the new terminology,
line bundles are actually familiar objects when X is normal.
Theorem 6.0.18. The sheaf L = OX (D) of a Cartier divisor D on a normal variety
X is the sheaf of sections of a line bundle VL X .
Proof. Recall from Chapter 4 that a Cartier divisor is locally principal, so that
X has an affine open cover {Ui }iI with D|Ui = div( fi )|Ui , fi C(X ) . Thus
{(Ui , fi )}iI is local data for D. Note also that
div( fi )|Ui U j = div( f j )|Ui U j ,
which implies fi / f j OX (Ui U j ) by Proposition 4.0.16.
We use this data to construct a line bundle as follows. Since

GL1 (OX (Ui U j )) = OX (Ui U j ) ,
the quotients gi j = fi / f j may be regarded as transition functions. These satisfy the

hypotheses of Proposition 6.0.16 and hence give a line bundle : VL X .
A global section f (X , OX (D)) satisfies div( f ) + D 0, so that on Ui ,
div( f fi )|Ui = div( f )|Ui + div( fi )|Ui = (div( f ) + D)|Ui 0.
This shows that si = fi f OX (D)(Ui ). Then
gi j s j = fi / f j f j f = fi f = si ,
which by part (b) of Proposition 6.0.16 gives a global section of : VL X .

Conversely, the proposition shows that a global section of VL X gives functions
si OX (D)(Ui ) such that gi j s j = si . It follows that f = si / fi C(X ) is independent
of i. One easily checks that f (X , OX (D)). The same argument works when we
restrict to any open subset of X . It follows that L = OX (D) is the sheaf of sections
of : VL X .

We will see shortly that this process is reversible, i.e., there is a one-to-one
correspondence between line bundles and sheaves coming from Cartier divisors.
First, we give an important example.
Example 6.0.19. When we regard P n as the set of lines through the origin in Cn+1 ,
each point p P n corresponds to a line p Cn+1 . We assemble these lines into
a line bundle as follows. Let x0 , . . . , xn be homogeneous coordinates on P n and
y0 , . . . , yn be coordinates on Cn+1 . Define
V P n Cn+1
as the locus where the matrix
x0
y0
xn
yn
254
has rank one. Thus V is defined by the vanishing of xi y j x j yi . Then define the
map : V P n to be projection on the first factor of P n Cn+1 . To see that V
is a line bundle, consider the open subset Cn Ui P n where xi is invertible. On
1 (Ui ) the equations defining V become
xj
yi = y j , for all j 6= i.
xi
Thus (x0 , . . . , xn , y0 , . . . , yn ) 7 (x0 , . . . , , xn , yi ) defines an isomorphism
i : 1 (Ui ) Ui C.
In other words, yi is a local coordinate for the line C over Ui . Switching to the
coordinate system over U j , we have the local coordinate y j , which over Ui U j is
related to yi via
xi
y j = yi .
xj
Hence the the transition function from Ui U j C to Ui U j C is given by
xi
gi j = OPn (Ui U j ) .
xj
This bundle is called the tautological bundle on P n . In Example 6.0.21 below, we
will describe the sheaf of sections of this bundle.
Projective spaces are the simplest type of Grassmannian, and just as in this
example, the construction of the Grassmannian shows that it comes equipped with
a tautological vector bundle. In Exercise 6.0.5 you will determine the transition
functions for the Grassmannian G(1, 3).
Invertible Sheaves and the Picard Group. Propositions 6.0.17 and 6.0.18 imply
that the sheaf OX (D) of a Cartier divisor is locally free of rank 1. In general, a
locally free sheaf of rank 1 is called an invertible sheaf.
The relation between Cartier divisors, line bundles and invertible sheaves is
described in the following theorem.
Theorem 6.0.20. Let L be an invertible sheaf on a normal variety X . Then:
(a) There is a Cartier divisor D on X such that L OX (D).
(b) There is a line bundle VL X whose sheaf of sections is isomorphic to L .

Proof. The part (b) of the theorem follows from part (a) and Proposition 6.0.18. It
remains to prove part (a).
Since X is irreducible, any nonempty open U X gives a domain OX (U ) with
field of fractions C(U ). By Exercise 3.0.4, C(U ) = C(X ), so that U 7 C(U )
defines a constant sheaf on X , denoted KX . This sheaf is relevant since OX (D) is
defined as a subsheaf of KX .
255
First assume that L is a subsheaf of KX . Pick an open cover {Ui } of X such

that L |Ui OX |Ui for every i. Over Ui , this gives homomorphisms
OX (Ui ) L (Ui ) C(X ).
Let fi1 C(X ) be the image of 1 OX (Ui ). One can show without difficulty that
fi / f j OX (Ui U j ) . Then {(Ui , fi )} is local data for a Cartier divisor D on X
satisfying L = OX (D).
For the general case, observe that on an irreducible variety, every locally constant sheaf is globally constant (Exercise 6.0.6). Now let L be any invertible sheaf
on X . On a small enough open set U , L (U ) OX (U ), so that
L (U ) OX (U) KX (U ) OX (U ) OX (U) KX (U ) KX (U ) = C(X ).
Thus L OX KX is locally constant and hence constant. This easily implies that
L OX KX KX , and composing this with the inclusion
L L OX KX
expresses L as a subsheaf of KX .
We note without proof that the line bundle corresponding to an invertible sheaf
is unique up to isomorphism. Because of this result, algebraic geometers tend to
use the terms line bundle and invertible sheaf interchangeably, even though strictly
speaking the latter is the sheaf of sections of the former.
We next discuss some properties of invertible sheaves coming from Cartier
divisors. A first result is that if D and E are Cartier divisors on X , then
(6.0.3)
OX (D) OX OX (E) OX (D + E).
This follows because f g 7 f g induces a sheaf homomorphism

OX (D) OX OX (E) OX (D + E)
which is clearly an isomorphism on any open set where OX (D) is trivial.
By standard properties of tensor product, the isomorphism (6.0.3) induces an
isomorphism
OX (E) HomOX (OX (D), OX (D + E)).
In particular, when E = D, we obtain
OX (D) OX OX (D) OX
and
OX (D) OX (D) ,
where OX (D) = HomOX (OX (D), OX ) is the dual of OX (D).
More generally, the tensor product of invertible sheaves is again invertible, and
if L is invertible, then L = HomOX (L , OX ) is invertible and
L OX L OX .
This explains why locally free sheaves of rank 1 are called invertible.
256
Example 6.0.21. There is a nice relation between the tautological bundle on P n

and the invertible sheaf OPn (1) introduced in Example 4.3.1. Recall that the TN invariant divisors D0 , . . . , Dn on P n are all linearly equivalent, and so define isomorphic sheaves, usually denoted OPn (1). The local data for the Cartier divisor D0
is easily seen to be {(Ui , xx0i )}, where Ui P n is the open set where xi 6= 0. Thus
the transition functions for OX (D0 ) are given by
gi j =
x0
xi
x0
xj
xj
.
xi
These are the inverses of the transition functions for the tautological bundle from
Example 6.0.19. It follows that the sheaf of sections of the tautological bundle is
OPn (1) = OPn (1).

We can also explain when Cartier divisors give isomorphic invertible sheaves.
Proposition 6.0.22. Two Cartier divisors D, E give isomorphic invertible sheaves
OX (D) OX (E) if and only if D E.
Proof. By Proposition 4.0.29, linearly equivalent Cartier divisors give isomorphic
sheaves. For the converse, we first prove that OX (D) = OX implies D = 0.
Assume OX (D) = OX . Then 1 (X , OX ) = (X , OX (D)), so D 0. If D 6= 0,
then we can pick an irreducible divisor D0 that appears in D with positive coefficient. The local ring OX,D0 is a DVR, so we can find h OX,D0 with D0 (h) = 1.
Set U = X \ W , where W is the union of all irreducible divisors D 6= Z0 with
D (h) 6= 0. There are only finitely many such divisors, so that U is a nonempty
open subset of X with U D0 6= . Then h (U , OX ), and h1
/ (U , OX ) since
h vanishes on U D0 . However,
(D + div(h1 ))|U = (D div(h))|U = (D D0 )|U 0,
so that h1 (U , OX (D)) = (U , OX ). This contradiction proves D = 0.
Now suppose that Cartier divisors D, E satisfy OX (D) OX (E). Tensoring

each side with OX (E) and applying (6.0.3), we see that OX (D E) OX . If
1 (X , OX ) maps to g (X , OX (D E)) via this isomorphism, then
gOX = OX (D E)
as subsheaves of KX . Thus
OX = g1 OX (D E) = O(D E + div(g)),
where the last equality follows from the proof of Proposition 4.0.29. By the previous paragraph, we have D E + div(g) = 0, which implies that D E.

In Chapter 4, the Picard group was defined as the quotient
Pic(X ) = CDiv(X )/Div0 (X ).
257
We can interpret this in terms of invertible sheaves as follows. Given L invertible,

Theorem 6.0.20 tells us that L OX (D) for some Cartier divisor D, which is
unique up to linear equivalence by Proposition 6.0.22. Hence we have a bijection
Pic(X ) {isomorphism classes of invertible sheaves on X }.
The right-hand side has a group structure coming from tensor product of invertible
sheaves. By (6.0.3), the above bijection is a group isomorphism.
In more sophisticated treatments of algebraic geometry, the Picard group of an
arbitrary variety is defined using invertible sheaves. Also, Cartier divisors can be
defined on an irreducible variety in terms of local data, without assuming normality
(see [131, II.6]), though one loses the connection with Weil divisors. Since most
of our applications involve toric varieties coming from fans, we will continue to
assume normality when discussing Cartier divisors.
Stalks, Fibers, and Sections. From here on, we will think of a line bundle L on
X as the sheaf of sections of a rank 1 vector bundle : VL X . Given a section
s L (U ) and p U , we get the following:
Since VL is a vector bundle of rank 1, we have the fiber 1 (p) C. Then

s : U VL gives s(p) 1 (p).
Since L is a locally free sheaf of rank 1, we have the stalk L p OX,p . Then
s L (U ) gives s p L p .
In Exercise 6.0.7 you will show that these are related via the equivalences
(6.0.4)
s(p) 6= 0 in 1 (p) s p
/ m pLp
s p generates L p as an OX,p -module
A section s vanishes at p X if s(p) = 0 in 1 (p), i.e., if s p m p L p .

Basepoints. It can happen that many sections of a line bundle vanish at a point p.
This leads to the following definition.
Definition 6.0.23. A subspace W (X , L ) has no basepoints or is basepoint
free if for every p X , there is s W with s(p) 6= 0.
As noted earlier, a global section s (X , L ) gives a sheaf homomorphism
OX L . Thus a subspace W (X , L ) gives
W C OX L
defined by s h 7 h s. Then (6.0.4) and Proposition 6.0.7 imply the following.
Proposition 6.0.24. A subspace W (X , L ) has no basepoints if and only if
W C OX L is surjective.
258
For a line bundle L = OX (D) of a Cartier divisor D on a normal variety, the

vanishing locus of a global section has an especially nice interpretation. The local
data {(Ui , fi )} of D gives the rank 1 vector bundle : VL X with transition
functions gi j = fi / f j . Hence we can think of a nonzero global section of OX (D) in
two ways:
A rational function f C(X ) satisfying D + div( f ) 0.
A morphism s : X VL whose composition with is the identity on X .
The relation between s and f is given in the proof of Theorem 6.0.18: over Ui , the
section s looks like (p, si (p)) for si = fi f OX (Ui ). It follows that s = 0 exactly
when si = 0. Since D|Ui = div( fi )|Ui , the divisor of si on Ui is given by
div( fi f )|Ui = (D + div( f ))|Ui .
These patch together in the obvious way, so that the divisor of zeros of s is
div0 (s) = D + div( f ).
Thus the divisor of zeros of a global section is an effective divisor that is linearly
equivalent to D. It is also easy to see that any effective divisor linearly equivalent
to D is the divisor of zeros of a global section of OX (D) (Exercise 6.0.8).
In terms of Cartier divisors, Proposition 6.0.24 has the following corollary.
Corollary 6.0.25. The following are equivalent for a Cartier divisor D:
(a) OX (D) is generated by global sections in the sense of Definition 6.0.12.
(b) D is basepoint free, meaning that (X , OX (D)) is basepoint free.
(c) For every p X there is s (X , OX (D)) with p
/ Supp(div0 (s)).
The Pullback of a Line Bundle. Let L be a line bundle on X and VL X the

associated rank 1 vector bundle. A morphism f : Z X gives the fibered product
f VL = VL X Z from 3.0 that fits into the commutative diagram
f VL
/ VL
/ X.
It is easy to see that f VL is a rank 1 vector bundle over Z.

Definition 6.0.26. The pullback f L of the sheaf L is the sheaf of sections of
the rank 1 vector bundle f VL defined above.
Thus the pullback of a line bundle is again a line bundle. Furthermore, there is
a natural map on global sections
f : (X , L ) (Z, f L )
259
defined as follows. A global section s : X VL gives the commutative diagram:

f
/X
f (s)
"
f VL
/ VL
1Z
$
/ X.
The universal property of fibered products guarantees the existence and uniqueness
of the dotted arrow f (s) : Z f VL that makes the diagram commute. It follows
that f (s) (Z, f L ).
Example 6.0.27. Let X P n be a projective variety. If we write the inclusion as

i : X P n , then the line bundle OPn (1) gives the line bundle i OPn (1) on X . When
the projective embedding of X is fixed, this line bundle is often denoted OX (1).
Thus a projective variety always comes equipped with a line bundle. However,
it is not unique, since the same variety may have many projective embeddings. You
will work out an example of this in Exercise 6.0.9.
In general, given a sheaf F of OX -modules on X and a morphism f : Z X ,

one gets a sheaf f F of OZ -modules on Z. The definition is more complicated, so
we refer the reader to [131, II.5] for the details.
Line Bundles and Maps to Projective Space. We now reverse Example 6.0.27 by
using a line bundle L on X to create a map to projective space.
Fix a finite-dimensional subspace W (X , L ) with no basepoints and let
W = HomC (W , C) be its dual. The projective space of W is
P(W ) = (W \ {0})/C .
We define a map L ,W : X P(W ) as follows. Fix p X and pick a nonzero
element v p 1 (p) C, where : VL X is the rank 1 vector bundle associated
to L . For each s W , there is s C such that s(p) = s v p . Then the map defined
by p (s) = s is linear and nonzero since W has no basepoints. Thus p W , and
since v p is unique up to an element of C , the same is true for p . Then
L ,W (p) = p
defines the desired map L ,W : X P(W ).
Lemma 6.0.28. The map L ,W : X P(W ) is a morphism.

Proof. Let s0 , . . . , sm be a basis of W and let Ui = {p X | si (p) 6= 0}. These open
sets cover X since W has no basepoints. Furthermore, the natural map
Ui C 1 (Ui ),
(p, ) 7 si (p)
260
is easily seen to be an isomorphism. Since all sections of Ui C C are of the

form p 7 (p, h(p)) for h OX (Ui ), it follows that for all 0 j m, we can write
s j |Ui = hi j si |Ui , hi j OX (Ui ).
The definition of L ,W uses a nonzero vector v p 1 (p). Over Ui , we can
use si (p) 1 (p). Then s j (p) = hi j (p)si (p) implies p (s j (p)) = hi j (p). Since
7 ((s0 ), . . . , (sm )) gives an isomorphism P(W ) P m , L ,W |Ui can be written
Ui P m ,
(6.0.5)
which is a morphism since hii = 1.
p 7 (hi0 (p), . . . , him (p)),
When W has no basepoints and s0 , . . . , sm span W , L ,W is often written

X P m ,
(6.0.6)
p (s0 (p), . . . , sm (p)) P m
with the understanding that this means (6.0.5) on Ui = {p X | si (p) 6= 0}.
Furthermore, when L = OX (D), we can think of the global sections si as

rational functions gi such that D + div(gi ) 0. Then L ,W can be written
(6.0.7)
X P m ,
p (g0 (p), . . . , gm (p)) P m .
Since gi (p) may be undefined, this needs explanation. The local data {(U j , f j )} of
D implies that f j g0 , . . . , f j gm OX (U j ). Then (6.0.7) means that L ,W |U j is
U j P m ,
p ( f j g0 (p), . . . , f j gm (p)) P m .
This is a morphism on U j since the global sections corresponding to g0 , . . . , gm have

no base points.
Exercises for 6.0.
6.0.1. For a sheaf homomorphism : F G , show that
U 7 ker(U )
defines a sheaf. Also prove that the following are equivalent:

(a) The kernel sheaf is identically zero.
(b) U is injective for every open subset U.
(c) is injective as defined in Definition 6.0.4.
6.0.2. In Example 6.0.5, prove that OP1 (D) OP1 (D) OP1 is surjective.
6.0.4. Let F , G be quasicoherent sheaves on X. Prove that U 7 HomOX (U ) (F (U), G (U))
defines a quasicoherent sheaf HomOX (F , G ).
6.0.5. The Grassmannian G(1, 3) is defined as the space of lines in P 3 , or equivalently, of

2-dimensional subspaces of V = C4 . This exercise will construct the tautological bundle
on G(1, 3), which assembles these 2-dimensional subspaces into a rank 2 vector bundle
over G(1, 3). A point of G(1, 3) corresponds to a full rank matrix

0 1 2 3
p=
=
0 1 2 3
261
up to left multiplication by elements of GL2 (C). Then define

V G(1, 3) C4
to consist of all pairs (( ), v) such that v Span(, ).

(a) A pair (( ), v) gives the 3 4 matrix

v
v0 v1 v2
A = = 0 1 2
0 1 2
v3
3 .
3
Prove that (( ), v) is a point of V if and only if the maximal minors of A vanish. This
shows that V G(1, 3) C4 is a closed subvariety.
(b) Projection onto the first factor gives a morphism : V G(1, 3). Explain why the
fiber over p G(1, 3) is the 2-dimensional subspace of C4 corresponding to p.
(c) Given 0 i < j 3, define
Ui j = {( ) G(1, 3) | i j j i 6= 0}.
Prove that Ui j C4 and that the Ui j give an affine open cover of G(1, 3).
(d) Given 0 i < j 3, pick k < l such that {i, j, k, l} = {0, 1, 2, 3}. Prove that the map
(p, v) 7 (p, vk , vl ) gives an isomorphism
1 (Ui j ) Ui j C2 .
(e) By part (d), V is a vector bundle over G(1, 3). Determine its transition functions.
6.0.6. Prove that a locally constant sheaf on an irreducible variety is constant.
6.0.7. Prove (6.0.4).
6.0.8. Prove that an effective divisor linearly equivalent to a Cartier divisor D is the divisor
of zeros of a global section of OX (D).
6.0.9. Let d : P1 P d be the Veronese mapping defined in Example 2.3.15. Prove that
d OPd (1) = OP1 (d).
6.0.10. Let f : Z X be a morphism and let L be a line bundle on X that is generated by
global sections. Prove that the pullback line bundle f L is generated by global sections.
6.0.11. Let D be a Cartier divisor on a complete normal variety X.
(a) f , g (X, OX (D)) \ {0} give effective divisors D + div( f ), D + div(g) on X. Prove
that these divisors are equal if and only if f = g, C .
(b) The complete linear system of D is defined to be
|D| = {E CDiv(X) | E D, E 0}.
Thus the complete linear system of D consists of all effective Cartier divisors on X
linearly equivalent to D. Use part (a) to show that |D| can be identified with the
projective space of (X, OX (D)), i.e., there is a natural bijection
|D| = P((X, OX (D))) = ((X, OX (D)) \ {0})/C .
(c) Assume that D has no basepoints and set W = (X, OX (D)). Then we can identify
P(W ) with the set of hyperplanes in P(W ) = |D|. Prove that the morphism OX (D),W :
X P(W ) is given by
OX (D),W = {E |D| | p Supp(E)} |D|.
262
6.1. Ample and Basepoint Free Divisors on Complete Toric Varieties

In this section we will study two special classes of Cartier divisors on complete
toric varieties. We begin with the basepoint free case.
n
Basepoint
P Free Divisors. Consider the toric variety X of a fan in NR R and
let D = a D be a torus-invariant Cartier divisor on X . By Propositions 4.3.3
and 4.3.8, we have the global sections
M
(X , OX (D)) =
C m,
mPD M
where PD MR is the polyhedron defined by

P
Since D = a D is Cartier, there are m M for such that
(6.1.1)
(6.1.2)
hm , u i = a ,
(1).
Furthermore, when max = (n), D is uniquely determined by the Cartier data

{m }(n) . Then we have the following preliminary result.
Proposition 6.1.1. If max = (n), then the following are equivalent:
(a) D has no basepoints, i.e., OX (D) is generated by global sections.
(b) m PD for all (n).
Proof. First suppose that D is generated by global sections and take (n).
The TN -orbit corresponding to is a fixed point p of the TN -action, and by the
Orbit-Cone Correspondence,
\
{p} =
D .
(1)
By Corollary 6.0.25, there is a global section s such that p is not in the support
of the divisor of zeros div0 (s) of s. Since (X , OX (D)) is spanned by m for
m PD M, we can assume that s is given by m for some m PD M. The
discussion preceding Corollary 6.0.25 shows that the divisor of zeros of s is
X
div0 (s) = D + div( m ) =
(a + hm, u i)D .
The point p is not in the support of div0 (s) yet lies in D for every (1). This
forces a + hm, u i = 0 for (1). Since is n-dimensional, we conclude that
m = m PD .
For the converse, take (n). Since m PD , the character m gives a

global section s whose divisor of zeros is div0 (s) = D + div( m ). Using (6.1.2),
one sees that the support of div0 (s) misses U , so that s is nonvanishing on U .
Then we are done since the U cover X .
263
Here is an example to illustrate Proposition 6.1.1.

Example 6.1.2. The fan for the Hirzebruch surface H2 is shown in Figure 3. Let
u1 = (1,2)
u2
u3
u4
Figure 3. A fan 2 with X2 = H2
Di be the divisor corresponding to ui . We will study the divisors

D = D4
and
D = D2 + D4 .
Write the Cartier data for D and D with respect to 1 , . . . , 4 as {mi } and {mi }
respectively. Figure 4 shows PD and mi (left) and PD and mi (right) (see also
m2
m3
m2
PD
m1 = m4
m3
PD
2
2
m4
m1
Figure 4. PD and mi (left) and PD and mi (right)
Exercise 4.3.5). This figure and Proposition 6.1.1 make it clear that D is basepoint
free while D is not.
P
Support Functions and Their Graphs. Let D = a D be a Cartier divisor on a
toric variety X . As in Chapter 4, its support function D : || R is determined
by the following properties:
D is linear on each cone .
D (u ) = a for all (1).
264
This is where the {m } from (6.1.2) appear naturally, since the explicit formula
for D | is given by D (u) = hm , ui for all u .
When M = Z2 and is complete, it is easy to visualize the graph of D in

MR R = R3 : imagine a tent, with centerpole extending from (0, 0, 0) down the
z-axis, and tent stakes placed at positions (u , a ). Here is an example.
Example 6.1.3. Take P1 P1 and consider the divisor D = D1 + D2 + D3 + D4 .
This gives the support function where D (ui ) = 1 for the four ray generators
u1 , u2 , u3 , u4 of the fan of P1 P1 . The graph of D is shown in Figure 5. This
u2
u1
u3
u4
Figure 5. The graph of D
should be visualized as an infinite Egyptian pyramid, with apex at the origin and
edges going through (ui , 1) for 1 i 4.
Convex Functions. We now introduce the key concept of convexity.

Definition 6.1.4. Let S NR be convex. A function : S R is convex if
(tu + (1 t)v) t(u) + (1 t)(v),
for all u, v S and t [0, 1].
We caution the reader that some books define convexity with the inequality
going the other way.
Continuing with the tent analogy, a support function D is convex exactly if
there are unimpeded lines of sight inside the tent. It is clear that for Example 6.1.3,
the support function is convex.
265
Full Dimensional Convex Support. In this chapter, our main focus is on complete
fans. However, the natural setting for convexity is the class of fans in NR which
satisfy the following two conditions:
|| NR is convex.
dim || = n = dim NR .
We say that has convex support of full dimension. Such fans satisfy
[
(6.1.3)
|| = Cone(u | (1)) =
.
(n)
In particular, the maximal cones of have dimension n, so that we can focus on

the cones (n), just as in the complete case.
Support Functions and Convexity. The following lemma describes when a support function is convex. Given a fan in NR Rn , a cone (n 1) is called a
wall when it is the intersection of two n-dimensional cones , (n), i.e, when
= forms the wall separating and . If is complete, every (n 1)
is a wall.
Lemma 6.1.5. Let D be a Cartier divisor on a toric variety whose fan has convex
support of full dimension. Then the following are equivalent:
(a) The support function D : || R is convex.
(b) D (u) hm , ui for all u || and (n).

(c) D (u) = min(n) hm , ui for all u ||.
(d) For every wall = , there is u0 \ with D (u0 ) hm , u0 i.

Proof. First assume (a) and fix v in the interior of (n). Given u ||, we can
find t (0, 1) with tu + (1 t)v . By convexity, we have
hm ,tu + (1 t)vi = D (tu + (1 t)v)
tD (u) + (1 t)D (v) = tD (u) + (1 t)hm , vi.
This easily implies hm , ui D (u), proving (b). The implication (b) (c) is
immediate since D (u) = hm , ui for u , and (c) (a) follows because the
minimum of a finite set of linear functions is always convex (Exercise 6.1.1).
Since (b) (d) is obvious, it remains to prove the converse. Assume (d) and
fix a wall = . Then lies on one side of the wall. We claim that
(6.1.4)
hm , ui hm , ui,
when u, are on the same side of .
This is easy. The wall is defined by hm m , ui = 0. Then (d) implies that the
halfspace containing is defined by hm m , ui 0, and (6.1.4) follows.
Now take u || and (n). Since || is convex, we can pick v in the

interior of so that the line segment uv intersects every wall of in a single point,
as shown in Figure 6 on the next page. Using (6.1.4) repeatedly, we obtain
266
us
wall
wall
{z } |

{z
wall
} |
vs
{z
Figure 6. Crossing walls from u to v along uv
hm , ui hm , ui hm , ui .
When we arrive at the cone containing u, the pairing becomes D (u), so that
hm , ui D (u). This proves (b).

In terms of the tent analogy, part (b) of the lemma means that if we have a
convex support function and extend one side of the tent in all directions, the rest of
the tent lies below the resulting hyperplane. Then part (d) means that it suffices to
check this locally where two sides of the tent meet.
The proof of our main result about convexity will use the following lemma that
describes the polyhedron of a Cartier divisor in terms of its support function.
P
Lemma 6.1.6. Let be a fan and D = a D be a Cartier divisor on X . Then
PD = {m MR | D (u) hm, ui for all u ||}.
Proof. Assume D (u) hm, ui for all u ||. Applying this with u = u gives
a = D (u ) hm, u i,
so that m PD by the definition of PDP

. For the opposite inclusion, take m PD and
u ||. Thus u , so that u = (1) u , 0. Then
P
P
hm, ui = (1) hm, u i (1) (a )
P
= (1) D (u ) = D (u),
where the inequality follows from m PD , and the last two equalities follow from
the defining properties of D .
We now expand Proposition 6.1.1 to give a more complete characterization of

when a divisor is basepoint free. Recall that PD is a polytope when is complete.
Theorem 6.1.7. Assume || is convex of full dimension n and let D be the support
function of a Cartier divisor D on X . Then the following are equivalent:
(a) D is basepoint free.
(d) D : || R is convex.
If addition is complete, then (a)(d) are equivalent to the following:
267
(e) PD = Conv(m | (n)).
(f) {m | (n)} is the set of vertices of PD .
(g) D (u) = minmPD hm, ui for all u NR .
Proof. The equivalences (a) (b) and (c) (d) were proved in Proposition 6.1.1
and Lemma 6.1.5. Furthermore, Lemmas 6.1.5 and 6.1.6 imply that
D is convex D (u) hm , ui for all (n), u ||
m PD for all (n).
This proves (d) (b), so that (a), (b), (c) and (d) are equivalent.
Assume (b) and note that PD is a polytope since is complete. Then m PD

and D (u) = min(n) hm , ui. Combining these with Lemma 6.1.6, we obtain
D (u) min hm, ui min hm , ui = D (u),
mPD
(n)
proving (g). The implication (g) (d) follows since the minimum of a compact
set of linear functions is convex (Exercise 6.1.1). So (a) (b) (c) (d) (g).
The implications (f) (e) (b) are clear. It remains to prove (b) (f). Take
(n). Let u be in the interior of and set a = D (u). By Exercise 6.1.2,
Hu,a = {m MR | hm, ui = a} is a supporting hyperplane of PD and
(6.1.5)
Hu,a PD = {m }.
This implies that m is a vertex of PD . Conversely, let Hu,a be a supporting hyperplane of a vertex v PD . Thus hm, ui a for all m PD , with equality if and only if
m = v. Since (b) holds, we also have (c) and (g). By (g), D (u) = minmPD hm, ui =
hm, vi = a. Combining this with (c), we obtain
D (u) = min hm , ui = a.
(n)
Hence hm , ui = a must occur for some (n), which forces v = m .
Example 6.1.8. In Example 6.1.2 we showed that on the Hirzebruch surface H2 ,

D = D4 is basepoint free while D = D2 + D4 is not. Theorem 6.1.7 gives a different
proof using support functions. Figure 7 on the next page shows the graph of the
support function D . Notice that the portion of the roof containing the points
u1 , u2 , u3 and the origin lies in the plane z = 0, and it is clear that for D , there are
unimpeded lines of sight within the tent. In other words, D is convex.
The support function D is shown in Figure 8 on the next page. Here, the line
of sight from u1 to u3 lies in the plane z = 0, yet the ridgeline going from the origin
to the point (u2 , 1) on the tent lies below the plane z = 0. Hence this line of sight
does not lie inside the tent, so that D is not convex.
268
u3
u2
u4
u1
Figure 7. The graph of D = D4 in Example 6.1.8
not inside tent
u2
u3
u4
u1
Figure 8. The graph of D = D2 +D4 in Example 6.1.8
When D is basepoint free, Theorem 6.1.7 implies that the vertices of PD are
the lattice points m , (n). One caution is that in general, the m need not
be distinct, i.e., 6= can have m = m . An example is given by the divisor
D = D4 considered in Example 6.1.2see Figure 4. As we will see later, this
behavior illustrates the difference between basepoint free and ample.
It can also happen that PD has strictly smaller dimension than the dimension of
X . You will work out a simple example of this in Exercise 6.1.3.
Ample Divisors. We now introduce the second key concept of this section.
Definition 6.1.9. Let D be a Cartier divisor on a complete normal variety X . As
we noted in 4.3, W = (X , OX (D)) is finite-dimensional.
(a) The divisor D and the line bundle OX (D) are very ample when D has no basepoints and D = OX (D),W : X P(W ) is a closed embedding.
(b) D and OX (D) are ample when kD is very ample for some integer k > 0.
269
We will see that support functions give a simple, elegant characterization of

when a torus-invariant Cartier divisor is ample. But first, we explore how the very
ample polytopes from Definition 2.2.17 relate to Definition 6.1.9.
Very Ample Polytopes. Let P MR Rn a full dimensional lattice polytope with
facet presentation
P = {m MR | hm, uF i aF for all facets F}.
This gives the complete normal fan P and the toric variety XP . Write
P M = {m1 , . . . , ms }.
A vertex mi P corresponds to a maximal cone
i = Cone(P M mi) P (n).

P
Proposition 4.2.10 implies that DP = F aF DF is Cartier since hmi , uF i = aF
when mi F.
(6.1.6)
Recall from Definition 2.2.17 that P is very ample if for every vertex mi P,
the semigroup N(P M mi ) is saturated in M. The definition of XP given in
Chapter 2 used very ample polytopes. This is no accident.
Proposition 6.1.10. Let XP and DP be as above. Then:
(a) DP is ample and basepoint free.
(b) If n 2, then kDP is very ample for every k n 1.
(c) DP is very ample if and only if P is a very ample polytope.

Proof. First observe that the polytope of the divisor DP is the polytope P we began
with, i.e., PDP = P. This has two consequences:
DP is basepoint free by Proposition 6.1.1, which proves the final assertion of
part (a).
If P M = {m1 , . . . , ms }, then the characters mi span W = (XP , OXP (DP )).
Since DP is basepoint free, these global sections give the morphism

DP = OXP (DP ),W P s1
by Lemma 6.0.28. As explained in (6.0.7), D can be written
(6.1.7)
DP (p) = ( m1 (p), . . . , ms (p)).
It follows that DP factors as

XP XPM P s1 ,
where XPM is the projective toric variety of P M M from 2.1. We need to
understand when XP XPM is an isomorphism.
270
Fix coordinates x1 , . . . , xs of P s1 and let I {1, . . . , s} be the set of indices

such that mi is a vertex of P. Hence each i I gives a vertex mi and a corresponding
maximal cone i in the normal fan of P.
If i I, then hmi , uF i = aF for every facet F containing mi . For all other
facets F, hmi , uF i > aF . Hence, if si is the global section corresponding to mi ,
then the support of div(si )0 = DP + div( mi ) consists of those divisors missing the
affine open toric variety Ui XP of i . It follows that Ui is the nonvanishing
locus of si .
s1
Under DP , this nonvanishing
locus maps to the
S
S affine open subset Ui P
where xi 6= 0. Since XP = iI Ui , and XPM iI Ui by Proposition 2.1.9, it
suffices to study the maps
Ui XPM Ui
of affine toric varieties. By Proposition 2.1.8,

XPM Ui = Spec(C[N(P M mi)]).
Since i = Cone(P M mi ) by (6.1.6), we have an inclusion of semigroups
N(P M mi ) i M.
This is an equality precisely when N(P M mi ) is saturated in M. Since Ui =

Spec(C[i M]), we obtain the equivalences:
DP is very ample XP XPM is an isomorphism
Ui XPM Ui is an isomorphism for all i I
C[N(P M mi )] C[ M] is an
isomorphism for all i I
N(P M mi ) is saturated for all i I
P is very ample.
This proves part (c) of the proposition. For part (b), recall that if n 2 and P
is arbitrary, then kP is very ample when k n 1 by Corollary 2.2.19. Hence
kDP = DkP is very ample. This implies that DP is ample (the case n = 1 is trivial),
which completes the proof of part (a).

Example 6.1.11. In Example 2.2.11, we showed that the simplex
P = Conv(0, e1 , e2 , e1 + e2 + 3e3 ) R3
is not normal. We show that P is not very ample as follows. From Chapter 2
we know that the only lattice points of P are its vertices, so that DP : XP P 3 .
Since XP is singular (Exercise 6.1.4) of dimension 3, it follows that DP cannot be
a closed embedding. Hence P and DP are not very ample. However, 2P and 2DP
are very ample by Proposition 6.1.10.
271
Ampleness and Strict Convexity. We next determine when a Cartier divisor D =

P
a D on X is ample. Our criterion will involve the support function D of D.
Recall that the Cartier data {m }(n) of D satisfies
hm , ui = D (u),
for all u .
Definition 6.1.12. Assume that has full dimensional convex support. Then the
support function D of a Cartier divisor D on X is strictly convex if it is convex
and for every (n) satisfies
hm , ui = D (u) u .
The following lemma, which you will prove in Exercise 6.1.5, shows that there
are many ways to think about strict convexity.
Lemma 6.1.13. Let D Cartier divisor on a toric variety whose fan has convex
support of full dimension. Then the following are equivalent:
(a) The support function D : || R is strictly convex.
(b) D (u) < hm , ui for all u || \ and (n).
(c) For every wall = , there is u0 \ with D (u0 ) < hm , u0 i.
(d) D is convex and m 6= m when 6= in (n) and is a wall.
(e) D is convex and m 6= m when 6= in (n).
(f) hm , u i > a for all (1) \ (1) and (n).
(g) D (u + v) > D (u) + D (v) for all u, v || not in the same cone of .
We now relate strict convexity to ampleness.

Theorem 6.1.14. Assume that D is the support function of a Cartier divisor D =
P
a D on a complete toric variety X . Then
D is ample D is strictly convex.
Furthermore, if n 2 and D is ample, then kD is very ample for all k n 1.

Proof. First suppose that D is very ample. Very ample divisors have no basepoints,
so D is convex by Theorem 6.1.7. If strict convexity fails, then Lemma 6.1.13
implies that has a wall = with m = m . Let V ( ) = O( ) X .
Let PD be the polyhedron of D from (6.1.1), which is a polytope since is

complete. Let PD M = {m1 , . . . , ms }, so that D : X P s1 can be written
D (p) = ( m1 (p), . . . , ms (p))
as in (6.1.7). In this enumeration, m = m = mi0 for some i0 . We will study D
on the open subset U U X .
272
First consider U . Theorem 6.1.7 implies that m PD , so that the section

corresponding to m is nonvanishing on U by the proof of Proposition 6.1.1. It
follows that on U , D is given by
D (p) = ( m1 m (p), . . . , ms m (p)) Ui0 Cs1 ,
where Ui0 P s1 is the open subset where xi0 6= 0.
Since m = m , the same argument works on U . This gives a morphism

D |U U : U U Ui0 Cs1 .
The only n-dimensional cones of containing are , since is a wall. Hence

V ( ) U U
by the Orbit-Cone Correspondence. Note also V ( ) P1 since is a wall. Since

P1 is complete, Proposition 4.3.8 implies that all morphisms from P1 to affine space
are constant. Thus D maps V ( ) to a point, which is impossible since D is very
ample. Hence D is strictly convex when D is very ample.
If D is ample, then kD is very ample for k 0. Thus kD = kD must be
strictly convex, which implies that D is strictly convex.
For the converse, assume D is strictly convex. Let {m }(n) be the Cartier
data of D. Since is convex, Theorem 6.1.7 shows that the m are the vertices of
PD . Hence PD is a lattice polytope.
If PD is not full dimensional, then there are u 6= 0 in NR and k R such that
hm , ui = k for all (n). Then Theorem 6.1.7 implies
D (u) = hm , ui = k
for all (n). Using strict convexity and Definition 6.1.12, we conclude that
u for all (n). Hence u = 0 since is complete. This contradicts u 6= 0
and proves that PD is full dimensional.
Hence PD gives the toric variety XPD with normal fan PD . Furthermore, XPD
has the ample divisor DPD from Proposition 6.1.10. We studied the support function
of this divisor in Proposition 4.2.14, where we showed that it is the function
PD (u) = min hm, ui.
mPD
However, this is precisely D by Theorem 6.1.7. Hence PD = D is strictly convex

with respect to (by hypothesis) and PD (by the first part of the proof).
Definition 6.1.13 implies that the maximal cones of the fan are the maximal
subsets of NR on which a strictly convex support function is linear. This, combined
with the previous paragraph, implies that = PD . Thus
(6.1.8)
X = XPD .
Furthermore, we also have

(6.1.9)
D = DPD
273
since the divisors have the same support function. Since DPD is an ample divisor
by Proposition 6.1.10, it follows that D is also ample.
The final assertion of the theorem follows from Proposition 6.1.10.
The relation between polytopes and ample divisors given by (6.1.8) and (6.1.9)
will be explored in 6.2. These facts also give the following nice result.
Theorem 6.1.15. On a smooth complete toric variety X , a divisor D is ample if
and only if it is very ample.
Proof. If D is ample, then X is the toric variety of PD by (6.1.8). Since X is
smooth, PD is very ample by Theorem 2.4.3 and Proposition 2.4.4. Since D is the
divisor of PD by (6.1.9), D is very ample by Proposition 6.1.10.

Computing Ample Divisors. Given a wall P(n 1), write = and pick
(1) \ (1). Then a Cartier divisor D = a D gives the wall inequality
(6.1.10)
hm , u i > a .
Lemma 6.1.13 and Theorem 6.1.14 imply that D is ample if and only if it satisfies
the wall inequality (6.1.10) for every wall of .
In terms of divisor classes, recall the map CDivT (X ) Pic(X ) whose kernel
consists of divisors of characters. If we fix 0 (n), then we have an isomorphism

P
(6.1.11)
D = a D CDivT (X ) | a = 0 for all 0 (1) Pic(X )
(Exercise 6.1.6). Then (6.1.10) gives inequalities for determining when a divisor
class is ample. Here is a classic example.
Example 6.1.16. Let us determine the ample divisors on the Hirzebruch surface
Hr . The fan for H2 is shown in Figure 3 of Example 6.1.2, and this becomes
the fan for Hr by redefining u1 to be u1 = (1, r). Hence we have ray generators
u1 , u2 , u3 , u4 and maximal cones 1 , 2 , 3 , 4 .
In Examples 4.3.5 and 4.1.8, we used D1 and D2 to give a basis of Pic(Hr ) =
Cl(Hr ). Here, it is more convenient to use D3 and D4 . More precisely, applying
(6.1.11) for the cone 4 , we obtain
Pic(Hr ) {aD3 + bD4 | a, b Z}.
To determine when aD3 + bD4 is ample, we compute mi = mi to be
m1 = (a, 0), m2 = (a, b), m3 = (rb, b), m4 = (0, 0).
Then (6.1.10) gives four wall inequalities which reduce to a, b > 0. Thus
(6.1.12)
aD3 + bD4 is ample a, b > 0.
274
For an arbitrary divisor D =
P4
i=1 ai Di ,
the relations
0 div( e1 ) = D1 + D3
0 div( e2 ) = r D1 + D2 D4
show that D (a1 ra2 + a3 )D3 + (a2 + a4 )D4 . Hence

P4
i=1 ai Di is ample a1 + a3 > ra2 , a2 + a4 > 0.
Sometimes ampleness is easier to check if we think geometrically in terms of

support functions. For D = aD3 + bD4 , look back at Figure 7 and imagine moving
the vertex at u3 downwards. This gives the graph of D , which is strictly convex
when a, b > 0.
Here is an example of how to determine ampleness using support functions.

Example 6.1.17. The fan for P1 P1 P1 has the eight orthants of R3 as its maximal cones, and the ray generators are e1 , e2 , e3 . Take the positive orthant R30
and subdivide further by adding the new ray generators
a = (2, 1, 1), b = (1, 2, 1), c = (1, 1, 2), d = (1, 1, 1).
We obtain a complete fan by filling the first orthant with the cones in Figure 9,
which shows the intersection of R30 with the plane x + y + z = 1. You will check
that is smooth in Exercise 6.1.7.
e3
c
d
a
e1
e2
Figure 9. Cones of lying in R30
P
Let D = a D be a Cartier divisor on X . Replacing D with D + div( m )
for m = (ae1 , ae2 , ae3 ), we can assume that D satisfies
D (e1 ) = D (e2 ) = D (e3 ) = 0.
Now observe that e1 + b = (2, 2, 1) = e2 + a. Since e1 and b do not lie in a cone of
, part (g) of Lemma 6.1.13 implies that
D (e1 + b) > D (e1 ) + D (b) = D (b).
275
However, e2 and a generate a cone of , so that

D (a) = D (e2 ) + D (a) = D (e2 + a) = D (e1 + b).
Together, these imply D (a) > D (b). By symmetry, we obtain
D (a) > D (b) > D (c) > D (a),
an impossibility. Hence has no strictly convex support functions, which shows
that X is a smooth complete nonprojective variety. See also Example B.2.2 for a
computational approach using the Polyhedra package of Macaulay2 [123].
We will say more computing ample divisors later in the chapter.
The Toric Chow Lemma. Recall from Chapter 3 that a refinement of gives a
proper birational toric morphism X X . We will now use the methods of this
section to prove the toric Chow lemma, which asserts that any complete fan has a
refinement that gives a projective toric variety. Here is the precise result.
Theorem 6.1.18. A complete fan has a refinement such that X is projective.
Proof. Suppose is a fan in NR Rn . Let be obtained from by considering
the complete fan obtained from
[
Span( ).
(n1)
So for each wall , we take the entire hyperplane spanned by the wall. This yields
a subdivison with the property that
[
[
=
Span( ),
(n1)
(n1)
i.e., each hyperplane Span( ) is a union of walls of , and all walls of arise
this way.
Choosing m M so that
{u NR | hm , ui = 0} = Span( ),
define the map : NR R by
(u) =
(n1)
|hm , ui|.
Note that takes integer values on N and is convex by the triangle inequality (this
explains the minus sign).
Let us show that is piecewise linear with respect to . Fix (n 1) and
note that each cone of is contained in one of the closed half-spaces bounded by
Span( ). This implies that u 7 |hm , ui| is linear on each cone of . Hence the
same is true for .
276
Finally, we prove that is strictly convex. Suppose that = 1 2 is a wall

of . Then Span(0 ), 0 (n 1). We label 1 and 2 so that
P
| (u) = hm0 , ui 6=0 in (n1) |hm , ui|, u 1
1
P
| (u) = hm0 , ui 6=0 in (n1) |hm , ui|, u 2 .
2
P
The sum 6=0 in (n1) |hm , ui| is linear on 1 2 , so is represented by different linear functions on each side of the wall . Since is convex,
P it is strictly
convex by Lemma 6.1.13. Then X is projective since D = (u )D is
ample by Theorem 6.1.14.

Using the results of Chapter 11, one can improve this result by showing that
X can be chosen to be smooth and projective.
Exercises for 6.1.
6.1.1. Let S MR be a compact set and define : NR R by (u) = minmS hm, ui.
Explain carefully why the minimum exists and prove that is convex.
6.1.2. Let Hu,a be as in the proof of (b) (d) of Theorem 6.1.7.
P Prove that Hu,a is a
supporting hyperplane of PD that satisfies (6.1.5). Hint: Write u = (1) u , > 0.
P
Then show m PD implies hm, ui = (1) hm, u i D (u).
6.1.3. As noted in the text, the polytope PD of a basepoint free Cartier divisor on a complete
toric variety X can have dimension strictly less than dim X . Here are some examples.
(a) Let D be one of the four torus-invariant prime divisors on P1 P1 . Show that PD is a
line segment.
(b) Consider (P1 )n and fix an integer d with 0 < d < n. Find a basepoint free divisor D on
(P1 )n such that dim PD = d. Hint: See Exercise 6.1.9 below.
6.1.4. Show that the toric variety XP of the polytope P in Example 6.1.11 is singular.
6.1.5. This exercise is devoted to proving that the statements (a)(g) of Lemma 6.1.13 are
equivalent. Many of the implications use Lemma 6.1.5.
(a) Prove (a) (b) and (c) (d).
(b) Prove (b) (e) and (b) (f) (c).
(c) Prove (c) (b) by adapting the proof of (d) (b) from Lemma 6.1.5.
(d) Prove (b) (g) and use the obvious implication (e) (d) to complete the proof of
the lemma.
6.1.6. Let X be the toric variety of a fan in NR Rn and fix 0 (n). Prove that the
natural map CDivT (X ) Pic(X ) induces an isomorphism

P
D = a D CDivT (X ) | a = 0 for all 0 (1) Pic(X ).
6.1.7. Prove that the toric variety X of Example 6.1.17 is smooth.
6.1.8. For the following toric varieties X , compute Pic(X ) and describe which torusinvariant divisors are ample and which are basepoint free.
277
(a) X is the toric variety of the smooth complete fan in R2 with

(1) = {e1 , e2 , e1 + e2 }.
(b) X is the blowup Bl p (P ) of P n at a fixed point p of the torus action.

n
(c) X is the toric variety of the fan from Exercise 3.3.12. See Figure 12 from Chapter 3.
(d) X is the toric variety of the fan obtained from the fan of Figure 12 from Chapter 3 by
combining the two upward pointing cones.
6.1.9. The toric variety (P1 )n has ray generators e1 , . . . , en . Let D1 , . . . , Dn denote the
Pn
+

corresponding torus-invariant divisors. Consider D = i=1 (a+
i Di + ai Di ).
(a) Show that D is basepoint free if and only if a+

i + ai 0 for all i.
(b) Show that D is ample if and only if a+

i + ai > 0 for all i.
P
6.1.10. Let D = a D be an ample divisor on a complete toric variety X . Define
= Cone((u , a ) | (1)) NR R.
(a) Prove that is strongly convex.

(b) Prove that the boundary of is the graph of the support function D .
(c) Prove that is the set of cones obtained by projecting proper faces of onto MR .
6.1.11. Let be the fan from Example 4.2.13. Prove the X is not projective.

We begin with the set of polytopes

P MR | P is a full dimensional lattice polytope
and the set of pairs

(X , D) | a complete fan in NR , D a torus-invariant ample divisor on X .
These sets are related as follows.
Theorem 6.2.1. The maps P 7 (XP , DP ) and (X , D) 7 PD define bijections between the above sets that are inverses of each other.
Proof. The map P 7 (XP , DP ) comes from Proposition 6.1.10, where we showed
that DP is an ample divisor on XP . Also recall from Proposition 3.1.6 that XP is
the toric variety of the normal fan P , which is a fan in NR . For (X , D) 7 PD ,
we showed that PD MR is a full dimensional lattice polytope in the proof of
Theorem 6.1.14.
It remains to prove that these maps are inverses of each other. One direction is
easy, since P 7 (XP , DP ) 7 PDP = P, where the equality is Exercise 4.3.1. Going
the other way, we have (X , D) 7 PD 7 (XPD , DPD ) = (X , D), where the equality
follows from (6.1.8) and (6.1.9) in the proof of Theorem 6.1.14.

The goal of this section is to look more deeply into the above relationship. In
particular, we are interested in the following questions:
278
Suppose P and Q are full dimensional lattice polytopes with XP = XQ . How

are P and Q related?
Suppose D is a torus-invariant Cartier divisor on XP that is basepoint free. How
are P and PD related?
The answers to these questions will involve generalized fans, pullbacks of divisors,
and Minkowski sums of polytopes.
Generalized Fans. The polytope PD of a basepoint free Cartier divisor D is a lattice
polytope by Theorem 6.1.7, but need not be full dimensional (see Exercise 6.1.3).
If we want PD to have a normal fan, we need to allow for more general fans. Here
is the definition we will use.
Definition 6.2.2. A generalized fan in NR is a finite collection of cones NR
such that:
(a) Every is a rational polyhedral cone.
(b) For all , each face of is also in .
(c) For all 1 , 2 , the intersection 1 2 is a face of each (hence also in ).

This agrees with the definition of fan given in Definition 3.1.2, except that the
cones are no longer required be strongly convex. The definitions of support and
complete extend to generalized fans in the obvious way. A generalized fan that
is a ordinary fan is called nondegenerate; otherwise is degenerate. Generalized
fans will play an important role in Chapters 14 and 15.
T
Let be a generalized fan. Then 0 = is the minimal cone in . It
has no proper faces and hence must be a subspace of NR . Let N = N/(0 N) with
quotient map : N N. You will prove the following in Exercise 6.2.1:
is a fan if and only if 0 = {0}.
For , = /0 NR /0 = N R is a strongly convex rational polyhedral

cone such that = R1 ().
= { | } is a fan in N R .
The toric variety X of the generalized fan is defined to be the toric variety of
the usual fan , i.e., X = X .
The Normal Fan of a Lattice Polytope. Some of most interesting generalized fans
come from polyhedra. Let P MR be a lattice polytope. We do not assume that P
is full dimensional. A vertex v P gives the cone
Cv = Cone(P M v) MR .
Similar to 2.3, the dual cone v = Cv NR is a rational polyhedral cone, and
these cones give a generalized fan as follows (Exercise 6.2.2).
279
Proposition 6.2.3. Given a lattice polytope P MR , the set

P = { | v , v is a vertex of P}
is a complete generalized fan in NR . Furthermore:
(a) The minimal cone of P is the dual of Span(m m | m, m P M) MR .
(b) P is a fan if and only if P MR is full dimensional.
We call P the normal fan of P. The toric variety XP is then defined to be the
toric variety of the generalized fan XP , i.e., XP = XP .
Example 6.2.4. Let P MR be a line segment whose vertices are lattice points.
The cone Cv at each vertex is a ray, so that the normal fan P consists of two
closed half-spaces and the hyperplane where they intersect. Taking the quotient by
this hyperplane gives the usual fan for P1 , so that XP = P1 .
n
The Normal
P Fan of a Basepoint Free Divisor. If is a complete fan in NR R
and D = a D has no basepoints and Cartier data {m }(n) , then PD is a
lattice polytope with the m as vertices. We can describe the normal fan PD of PD
as follows.
P
Proposition 6.2.5. Let D = a D be a basepoint free Cartier divisor on X
with polytope PD . Then:
(a) If v PD is a vertex, then the corresponding cone v = Cv in the normal fan

PD is the union
[
v =
.
(n)
m =v
(b) is a refinement PD .
Proof. Part (b) follows immediately from part (a). Let v PD be a vertex. Since
Cv = Cone(PD M v) is strongly convex, its dual v = Cv has dimension n in
NR . It follows that part (a) is equivalent to the assertion
(6.2.1)
for all (n), Int() Int(v ) 6= implies m = v,
where Int denotes the interior (Exercise 6.2.3). Also note that any u v satisfies
hm v, ui 0,
for all m PD M.
In particular, m PD for (n) since D is basepoint free, so that

(6.2.2)
hm , ui hv, ui,
for all (n).
We now prove (6.2.1). Assume Int() Int(v ) 6= and let u be an element of

the intersection. Since v = m for some (n), we have
hv, ui D (u) = hm , ui
280
by convexity and part (b) of Lemma 6.1.5. Combining this with (6.2.2), we see that
hm , ui = hv, ui,
for all u Int() Int(v ).
Since Int() Int(v ) is open, this forces m = m , proving (6.2.1).
This proposition gives a nice way to think about the normal fan PD . One
begins with the Cartier data {m }(n) of D and then combines all cones (n)
whose m s give the same vertex of PD . These combined cones and their faces
satisfy the conditions for being a fan, except that strong convexity fails when PD is
not full dimensional. Here is an example of how this works.
Example 6.2.6. For the Hirzebruch surface H2 , consider the divisors D = D4 and
D = D1 . The polytope PD from Figure 4 of Example 6.1.2 is shown on the left in
Figure 10 on the next page. By Proposition 6.2.5, m1 = m4 tells us to combine 1
and 4 , as shown on the right in Figure 10. Thus the normal fan of PD is a fan with
three maximal cones.
m2
m3
PD
u1 = (1,2)
4
u2
m1 = m4
u3
2
u4
Figure 10. PD (left) and its normal fan (right)
The polytope PD is the line segment shown on the left in Figure 11. Here, we
combine 1 and 2 (since m1 = m2 ) and also combine 3 and 4 (since m3 = m4 ).
This gives the degenerate normal fan shown on the right in Figure 11. Thus the
toric variety of PD is P1 .
u1 = (1,2)
PD
m1 = m2
m3 = m4
4
u2
u3
1
u4
Figure 11. PD (left) and its degenerate normal fan (right)
281
Pulling Back via Toric Morphisms. In order to understand the full implications of
Proposition 6.2.5, we need the following description of pullbacks of torus-invariant
Cartier divisors by toric morphisms.
Proposition 6.2.7. Assume that : X1 X2 is the toric morphism induced by
: N1 N2 , and let D2 be a torus-invariant Cartier divisor with support function
D2 : |2 | R. Then there is a unique torus-invariant Cartier divisor D1 on X1
with the following properties:
(a) OX1(D1 ) OX2(D2 ).
(b) The support function D1 is the composition

D
2
|1 | |2 |
R.
Proof. Let the local data of D2 be {(U , m )}2 , where now refers to an
arbitrary cone of 2 . Recall that the minus sign comes from hm , u i = a when
(1). Then the proof of Theorem 6.0.18 shows that OX2(D2 ) is the sheaf of
sections of a rank 1 vector bundle V X2 with transition functions
g = m m .
Now take 1 and let 2 be the smallest cone satisfying R ( ) .
Using the dual map : M2 M1 , we set
m = (m ).
Since (U ) U , one can show without difficulty that
g = m m OX1(U U ) .
Then {(U , m )} 1 is the local data for a Cartier divisor D1 on X1 . It is

straightforward to verify that D1 has the required properties (Exercise 6.2.4).
In the situation of Proposition 6.2.7, we call D1 is the pullback of D2 via
since OX1(D1 ) is the pullback of OX2(D2 ) via . We denote this by D1 = D2 .
The Structure of Basepoint Free Divisors. Proposition 6.2.5 shows that refines
the normal fan PD . Hence we should have a toric morphism X XPD . This is
certainly true when PD is nondegenerate, and as we will see below, it remains
true when PD is degenerate. More importantly, D is (up to linear equivalence) the
pullback of an ample divisor on XPD via this morphism.
Theorem 6.2.8. Let D be a basepoint free Cartier divisor on a complete toric variety, and let XD be the toric variety of the polytope PD MR . Then the refinement
of PD induces a proper toric morphism
: X XPD .
Furthermore, D is linearly equivalent to the pullback via of the ample divisor on

XPD coming from PD .
282
Proof. The minimal cone 0 of PD is a subspace of NR . Let N = N/(0 N),

with quotient map : N N. Since refines PD and PD projects to a genuine
fan in N R , it follows that induces a toric morphism as claimed. Note also that
is proper since X and XPD are complete.
Let M M be dual to : N N. Part (a) of Proposition 6.2.3 implies that
M R = Span(m m | m, m PD M}. Translating PD by a lattice point, we may
assume that PD M R . This changes our original divisor D by a linear equivalence.
The polytope PD gives the ample divisor D = DPD on XPD . Since D is basepoint
free, Theorem 6.1.7 implies that
D (u) = min hm, ui.
mPD
Using PD M R , one sees that D factors through : N N, and in fact,

D = D R
(Exercise 6.2.5). By Proposition 6.2.7, D is the pullback of D = DPD via .
Theorem 6.2.8 implies that a Cartier divisor without basepoints on a complete

toric variety has a very nice structure: it is linearly equivalent to the pullback (via
a toric morphism) of an ample divisor on a projective toric variety of possibly
smaller dimension. This will be useful when we study the geometric invariant
theory of toric varieties in Chapters 14 and 15.
Here are two examples to illustrate what can happen in Theorem 6.2.8.
Example 6.2.9. The toric variety X of Example 6.1.17 has no ample divisors, but
it does have nontrivial basepoint free divisors. The ray generators of are
e1 , e2 , e3 , a, b, c, d,
with corresponding toric divisors
D1 , D2 , D3 , Da , Db , Dc , Dd .
Then one can show that
D = 2D
1 + 2D2 + 2D3 Da Db Dc Dd
is basepoint free (Exercise 6.2.6). Thus the support function D is convex.

Figure 9 in Example 6.1.17 shows that Cone(e1 , e2 , d) is a union of three cones
of . Using D (e1 ) = D (e2 ) = 0 and D (a) = D (b) = D (d) = 1, one sees that
these three cones all have m = e3 (Exercise 6.2.6). Hence we should combine
these three cones. The same thing happens in Cone(e1 , e3 , d) and Cone(e2 , e3 , d).
In the first orthant, the fan of XPD looks like Figure 12 on the next page when
intersected with x + y + z = 1. Hence XPD is the blowup of (P1 )3 at the point
corresponding to the first orthant (Exercise 6.2.6). Also, : X XPD is a proper
birational toric morphism since refines the (nondegenerate) normal fan PD .
283
e3
e1
e2
Figure 12. Combined cones of lying in R30 in Example 6.2.9
Example 6.2.10. Consider the divisor D = 3D1 + D2 D4 = D1 + div( e2 ) on

the Hirzebruch surface H2 . Then PD = Conv(e2 , e1 e2 ) = Conv(0, e1 ) e2 .
This gives the degenerate normal fan shown in Figure 11 of Example 6.2.6, and
: X XPD = P1 is the toric morphism from Example 3.3.5. Then D D1 ,
which is the pullback of an ample divisor on P1 .
N-Minkowski Summands. We now return to the questions asked at the beginning

of the section. In terms of normal fans, the answers are easy to give:
Full dimensional lattice polytopes P and Q in MR give the same toric variety if
and only if they have the same normal fan.
If D is a torus-invariant basepoint free Cartier divisor on XP , then the normal
fan of P refines the normal fan of PD by Proposition 6.2.5.
By rephrasing this in terms of Minkowski sums, we can state both of these purely
in the language of polytopes. Here is the definition.
Definition 6.2.11. Given lattice polytopes P and Q in MR , Q is an N-Minkowski
summand of P if
Q + Q = k P,
where k N is positive and Q MR is a lattice polytope.
Example 6.2.12. The rectangle Q = Conv(0, 2e1 , e2 , 2e1 + e2 ) is an N-Minkowski

summand of the hexagon P = Conv(0, e1 , e2 , 2e1 +e2 , e1 +2e2 , 2e1 +2e2 ), as shown
by Figure 13.
Q+e2
2P
Q
0
Figure 13. Q is an N-Minkowski summand of P since Q + Q = 2P
284
Minkowski sums are related to normal fans as follows [28, Prop. 1.2].
Proposition 6.2.13. Let P and Q be lattice polytopes in MR . Then:
(a) Q is an N-Minkowski summand of P if and only if P refines Q .
(b) P+Q is the coarsest common refinement of P and Q , i.e., any fan that
refines P and Q also refines P+Q .

Proposition 6.2.13 does not assume that the lattice polytopes P and Q have full
dimension, so the normal fans P and Q in the proposition may be degenerate.
Also note that P+Q is common refinement of P and Q by part (a). So the point
of part (b) is that P+Q is the most efficient common refinement.
We can now describe when two polytopes give the same toric variety.
Corollary 6.2.14. Full dimensional lattice polytopes in MR give the same toric
variety if and only if each is an N-Minkowski summand of the other.

Proof. This follows immediately from Proposition 6.2.13 since two fans are equal
if and only if each refines the other.

We also have the following lovely result about basepoint free divisors.
Corollary 6.2.15. Let P be a full dimensional lattice polytope in MR . Then a
polytope Q MR is an N-Minkowski summand of P if and only if there is a torusinvariant basepoint free Cartier divisor D on XP such that Q = PD .
Proof. If D is basepoint free on XP , then Propositions 6.2.5 and 6.2.13 imply that
PD is an N-Minkowski summand of P. For the converse, suppose that Q is an NMinkowski summand of P. Then P refines Q . We will write the maximal cones
of Q as v for v Q a vertex. Also let n = dim XP . We define D as follows. Each
P (n) is contained in v for some vertex v Q. Then D is the Cartier divisor
on XP whose Cartier data {m }P (n) is defined by m = v when v .
P
Thus D = P (1) a D , where a = hv, u i when u v . To prove that
PD = Q, take m PD , so that hm, u i a for all P (1). This implies
hm v, u i = hm, u i + a 0 for all u v .
These u s generate v since P refines Q , so that m v v = Cv . Hence

T
m v is a vertex of Q (Cv + v) = Q,
where the equality follows from Exercise 6.2.7. The opposite inclusion Q PD is
straightforward and hence is left to the reader. This proves PD = Q, and then D is
basepoint free by Proposition 6.1.1.

Example 6.2.16. Consider the rectangle Q and the hexagon P defined in Example 6.2.12. Since Q is an N-Minkowski summand of P, it gives a basepoint free
divisor D on XP . Let the ray generators u1 , . . . , u6 of P be arranged clockwise
285
around the origin, starting with u1 = e2 . Then the recipe for D given in the proof
of Corollary 6.2.15 makes it easy to show that
D = D3 + D4 + 2D5 + D6 ,
where Di is the toric divisor corresponding to ui (Exercise 6.2.8).
Zonotopes. Recall from Example 2.3.10 that a zonotope is a Minkowski sum of

line segments. Here we show that zonotopes have especially nice normal fans. A
central hyperplane arrangement in NR consists of finitely many rational hyperplanes H NR whose intersection is the origin. This determines a fan in NR whose
maximal cones are the closures of the connected components of the complement
of the arrangement.
Example 6.2.17. The hexagon P from Example 6.2.12 is a zonotope since P =
Conv(0, e1 ) + Conv(0, e2 ) + Conv(0, e1 + e2 ). Figure 14 reproduces Figure 7 from
v5
v4
v6
v3
P
v1
2
4
1
5
v2
Figure 14. A zonotope P and its normal fan P
Example 2.3.10. As you can see, the normal fan of P comes from an arrangement
of three lines through the origin in R2 .
Proposition 6.2.18. The normal fan of a full dimensional lattice zonotope P comes
from a central hyperplane arrangement.
Proof. First note that a Minkowski sum of parallel line segements is again a line
segment. Thus we can write P = L1 + + Ls as a Minkowski sum of line segments
where no two segments are parallel. Each normal fan Li is determined by the
hyperplane normal Hi to Li , as explained in Example 6.2.4. By Proposition 6.2.13,
P = L1 ++Ls is the coarsest common refinement of L1 ,. . . ,Ls . This is clearly
the fan determined by the central hyperplane arrangement H1 , . . . , Hs . Note that
that H1 Hs = {0} since P is a nondegenerate fan.

See [281, Thm. 7.16] for a different proof of Proposition 6.2.18 that uses linear
programming.
286
Exercises for 6.2.

6.2.1. Prove the properties of generalized fans stated in three bullets in the discussion
following Definition 6.2.2.
6.2.3. Prove that part (a) of Propostion 6.2.5 follows from (6.2.1).
6.2.5. Complete the proof of Theorem 6.2.8.
6.2.6. This exercise deals with Example 6.2.9.
(a) Let D = 2D
1 + 2D2 + 2D3 Da Db Dc Dd be the divisor from Example 6.2.9.
Prove that PD is the polytope with 10 vertices
e1 , e2 , e3 , 2e1 , 2e2 , 2e3 , 2e1 + 2e2 , 2e1 + 2e3 , 2e2 + 2e3 , 2e1 + 2e2 + 2e3
and conclude that D is basepoint free.
(b) In Example 6.2.9, we asserted that certain maximal cones of must be combined to
get the maximal cones of PD . Prove that this is correct.
(c) Show that XPD is the blowup of (P1 )3 at the point corresponding to the first orthant.
6.2.7. This exercise is concerned with the proof of Corollary 6.2.15.
(a) Given
T a lattice polytope Q MR , let Cv = Cone(Q M v) for v Q a vertex. Prove
that v is a vertex of Q (Cv + v) = Q.
(b) Complete the proof of the corollary by showing Q PD .
6.2.8. In Example 6.2.16, the rectangle Q is an N-Minkowski summand of the hexagon P.

(a) In the example, we claimed that D = D3 + D4 + 2D5 + D6 . Prove this.
(b) Let Q = Conv(0, e2 , 2e1 + 2e2 , 2e1 + 3e2 ). Prove carefully that Q + Q = 2P and compute the basepoint free divisor D determined by Q .
6.2.9. Suppose that full dimensional lattice polytopes P, Q MR give the same toric variety
X . Prove that P + Q also gives X .
6.2.10. Let P MR be a full dimensional lattice polytope. A face Q P determines a cone
Q in the normal fan of P. This gives the orbit closure V (Q ) XP , and V (Q ) XQ by
Proposition 3.2.9. This gives an inclusion i : XQ XP which is not a toric morphism when
Q P. Prove that i OXP (DP ) OXQ (DQ ).
6.3. The Nef and Mori Cones

In 6.1, we gave some nice criteria for when a Cartier divisor D is basepoint free
or ample. We now study the structure of the set of basepoint free divisors and the
set of ample divisors inside Pic(X )R = Pic(X ) Z R.
The main concept of this section is that of numerical effectivity. Roughly

speaking, the goal is to define a pairing between divisors and curves, such that
for a Cartier divisor D and complete curve C on a variety X , the number D C
counts the number of points of D C, with appropriate multiplicity.
287
Example 6.3.1. Suppose X = P 2 with homogeneous coordinates x, y, z, and let

D = V(y) and C = V(zy x 2 ). Then D and C meet at the single point p = (0, 0, 1),
where they share a common tangent. If we replace D with the linearly equivalent
divisor E = V(y z), then clearly E and C meet in two points. This suggests that
the point {p} = D C should be counted twice, since it is a tangent point. Hence
we should have D C = 2.
Despite this encouraging example, there are several technical hurdles to overcome in order to make this precise in a general setting. Note that in C2 , two lines
may or may not meet, so to get a reasonable theory, we will work with complete
curves C on a normal variety X . We also need to restrict to Cartier divisors D on X .
With these assumptions, the intersection product D C should possess the following
properties:
(D + E) C = D C + E C.
D C = E C when D E.
Let D be a prime divisor on X such that D C is finite. Assume each p D C

is smooth in C, D, X and that the tangent spaces Tp (C) Tp (X ) and Tp (D)
Tp (X ) meet transversely. Then D C = |D C|.
Using these properties, one can give a rigorous proof of the computation D C = 2
from Example 6.3.1.
The Degree of a Line Bundle. The key tool we will use is the notion of the degree
of a divisor on an irreducible
P smooth complete curve C. Such a divisor can be
written as a finite sum D = i ai pi where ai Z and pi C.
P
Definition 6.3.2. Let D = i ai pi be a divisor on an irreducible smooth complete
curve C. Then the degree of D is the integer
X
deg(D) =
ai Z.
i
Note the obvious property deg(D + E) = deg(D) + deg(E). The following key
result is proved in [131, Cor. II.6.10].
Theorem 6.3.3. Every principal divisor on an irreducible smooth complete curve
has degree zero.

In other words, deg(div( f )) = 0 for all nonzero rational functions f on an
irreducible smooth complete curve C. Thus
deg(D) = deg(E) when D E on C,
and the degree map induces a surjective homomorphism
deg : Pic(C) Z.
Note that all Weil divisors are Cartier since C is smooth.
288
In 6.0 we showed that Pic(C) is the set of isomorphism classes of line bundles
on C. Hence we can define the degree deg(L ) of a line bundle L on C. This leads
immediately to the following result.
Proposition 6.3.4. Let C be an irreducible smooth complete curve. Then a line
bundle L has a degree deg(L ) such that L 7 deg(L ) has the following properties:
(a) deg(L L ) = deg(L ) + deg(L ).
(b) deg(L ) = deg(L ) when L L .
(c) deg(L ) = deg(D) when L OC (D).
The Normalization of a Curve. We defined the normalization of an affine variety

in 1.0, and by gluing together the normalizations of affine pieces, one can define
the normalization of any variety (see [131, Ex. II.3.8]). In particular, an irreducible
curve C has a normalization map
: C C,
where C is an normal variety. Here are the key propeties of C.
Proposition 6.3.5. Let C be the normalization of an irreducible curve C. Then:
(a) C is smooth.
(b) C is complete whenever C is complete.
Proof. Since C is a curve, Proposition 4.0.17 implies that C is smooth. Part (b) is
covered by [131, Ex. II.5.8].

One can prove that every irreducible smooth complete curve is projective. See
[131, Ex. II.5.8].
The Intersection Product. We now have the tools needed to define the intersection product. Let X be a normal variety. Given a Cartier divisor D on X and an
irreducible complete curve C X , we have
The line bundle OX (D) on X .
The normalization : C C.
Then OX (D) is a line bundle on the irreducible smooth complete curve C.

Definition 6.3.6. The intersection product of D and C is D C = deg( OX (D)).
Here are some properties of the intersection product.
Proposition 6.3.7. Let C be an irreducible complete curve and D, E Cartier divisors on a normal variety X . Then:
(a) (D + E) C = D C + E C.
(b) D C = E C when D E.
289
Proof. The pullback of line bundles is compatible with tensor product, so that
part (a) follows from (6.0.3) and Proposition 6.3.4. Part (b) is an easy consequence
of Propositions 6.0.22 and 6.3.4.

The intersection product extends to Q-Cartier divisors as follows. Recall from
Chapter 4 that a Weil divisor D is Q-Cartier if D is Cartier for some integer > 0.
Given an irreducible complete curve C X , let
1
(6.3.1)
D C = (D) C Q.
In Exercise 6.3.1 you will show that this intersection product is well-defined and
satisfies Propostion 6.3.7.
Intersection Products on Toric Varieties. In the toric case, D C is easy to compute
when D and C are torus-invariant in X . In order for C to be torus-invariant and
complete, we must have C = V ( ) = O( ), where = (n 1) is the wall
separating cones , (n), n = dim X . We do not assume is complete.
In this situation, we have the sublattice N = Span( ) N N and the quotient

N( ) = N/N . Let and be the images of and in N( )R . Since is a wall,
N( ) Z and , are rays that correspond to the rays in the usual fan for P1 . In
particular, V ( ) P1 is smooth, so no normalization is needed when computing
the intersection product.
Proposition 6.3.8. Let C = V ( ) be the complete torus-invariant curve in X

coming from the wall = . Let D be a Cartier divisor with Cartier data
m , m M corresponding to , (n). Also pick u N that maps to the
minimal generator of N( )R . Then
D C = hm m , ui Z.
Proof. Since V ( ) U U , we can assume X = U U and is the fan

consisting of , and their faces. We also have
D|U = div( m )|U , D|U = div( m )|U .
The proof of Proposition 6.2.7 shows that the line bundle OX (D) is determined by
the transition function g = m m . Thus
D C = deg(i OX (D)),
where i : V ( ) X is the inclusion map. The pullback bundle is determined by

the restriction of g to
V ( ) U U = V ( ) U = O( ),
where O( ) is the TN -orbit corresponding to . This is also the torus of the toric
variety V ( ) = O( ). In Lemma 3.2.5, we showed that M is the dual of N( )
and that
O( ) HomZ (M , C ) TN( ) .
290
Now comes the key observation: since the linear functions given by m , m agree
on , we have m m M. Thus i OX (D) is the line bundle on V ( )
whose transition function is g = m m for m m M.
It follows that i OX (D) OV ( ) (D), where D is the divisor on V ( ) given by

the Cartier data
m = 0, m = m m .
Let p , p be the torus fixed points corresponding to , . Since u N maps

to the minimal generator u N( ), we have
D = hm , ui p + hm , ui p = hm m , ui p ,
where the second equality follows from m = m m M. Hence
D C = deg(i OX (D)) = deg(D) = hm m , ui.
Example 6.3.9. Consider the toric surface whose fan in R2 has ray generators
u1 = e1 , u2 = e2 , u0 = 2e1 + 3e2
and maximal cones
= Cone(u1 , u0 ), = Cone(u2 , u0 ).
The support of is the first quadrant and = = Cone(u0 ) gives the complete
torus-invariant curve C = V ( ) X .
If D1 , D2 , D0 are the divisors corresponding to u1 , u2 , u0 , then
D = aD1 + bD2 + cD0 is Cartier 2a + 3b c mod 6.

When this condition is satisfied, we have
m = ae1 +
3b c
2a c
e2 , m =
e1 be2 .
3
2
Also, u = e1 + 2e2 maps to the minimal generator of since u, u0 form a basis

of Z2 . (You will check these assertions in Exercise 6.3.2.) Thus
D C = hm m , ui =
2a + 3b c
6
by Proposition 6.3.8. Note that D is Q-Cartier since is simplicial. Then (6.3.1)

shows that the formula for D C holds for arbitrary integers a, b, c. In particular,
1
1
1
D1 C = , D2 C = , D0 C = .
3
2
6
In the next section we will see that these intersection products follow directly from
the relation u0 + 2u1 + 3u2 = 0 and the fact that Zu0 = Span( ) Z2 .
291
Nef Divisors. We now define an important class of Cartier divisors.

Definition 6.3.10. Let X be a normal variety. Then a Cartier divisor D on X is nef
(short for numerically effective) if
D C 0
for every irreducible complete curve C X .
A divisor linearly equivalent to a nef divisor is nef. Here is another result.
Proposition 6.3.11. Every basepoint free divisor is nef.
Proof. The pullback of a line bundle generated by global sections is generated by
global sections (Exercise 6.0.10). Thus, given : C C and D basepoint free,
the line bundle L = (OX (D)) is generated by global sections. This allows us to
write L = OC (D ) for a basepoint free divisor D on C. A nonzero global section
of OC (D ) gives an effective divisor E linearly equivalent to D . Then
D C = deg( (OX (D))) = deg(OC (D )) = deg(D ) = deg(E ) 0,
where the last inequality follows since E is effective.
In the toric case, nef divisors are especially easy to understand.

Theorem 6.3.12. Let D be a Cartier divisor on a toric variety X whose fan has
convex support of full dimension. The following are equivalent:
(a) D is basepoint free, i.e., OX (D) is generated by global sections.
(b) D is nef.
(c) D C 0 for all torus-invariant irreducible complete curves C X .
Proof. The first item implies the second by Proposition 6.3.11, and the second
item implies the third by the definition of nef. So suppose that D C 0 for all
torus-invariant irreducible curves C. We can replace D with a linearly equivalent
torus-invariant divisor. Then, by Theorem 6.1.7, it suffices to show that D is
convex.
Take a wall = of and set C = V ( ). If we pick u N as in
Proposition 6.3.8, then
hm m , ui = D C 0,
so that
hm , ui hm , ui = D (u).
Note that u
/ since the image of u is nonzero in N( ) = N/(Span( ) N). Then
Lemma 6.1.5 implies that D is convex.

A variant of the above proof leads to the following ampleness criterion, which
you will prove in Exercise 6.3.3.
292
Theorem 6.3.13 (Toric Kleiman Criterion). Let D be a Cartier divisor on a complete toric variety X . Then D is ample if and only if DC > 0 for all torus-invariant
irreducible curves C X .

Note that one direction of the proof follows from the general fact that on any
complete normal variety, an ample divisor D satisfies D C > 0 for all irreducible
curves C X (Exercise 6.3.4).
Theorems 6.3.12 and 6.3.13 were well-known in the smooth case and proved
more recently (and independently) in [185], [197] and [212] in the complete case.
Numerical Equivalence of Divisors. The intersection product leads to an important equivalence relation on Cartier divisors.
(a) A Cartier divisor D on X is numerically equivalent to zero if D C = 0 for all
irreducible complete curves C X .
(b) Cartier divisors D and E are numerically equivalent, written D E, if D E
is numerically equivalent to zero.
What does this say in the toric case?
Proposition 6.3.15. Let D be a Cartier divisor on a toric variety X whose fan
has convex support of full dimension. Then D 0 if and only if D 0.
Proof. Clearly if D is principal then D is numerically equivalent to zero. For the
converse, assume D 0 and let = be a wall of . If we pick u as in
Proposition 6.3.8, then
0 = D C = hm m , ui
for C = V ( ). This forces m = m since m m and u

/ . From here,
one sees that m = m for all , (n), and it follows that D is principal.
Numerical Equivalence of Curves. We also get an interesting equivalence relation

on curves. Let Z1 (X ) be the free abelian group generated by irreducible complete
curves C X . An element of Z1 (X ) is called a proper 1-cycle.
(a) A proper 1-cycle C on X is numerically equivalent to zero if D C = 0 for all
Cartier divisors D on X .
(b) Proper 1-cycles C and C are numerically equivalent, written C C , if C C
is numerically equivalent to zero.
The intersection product (D,C) 7 D C extends naturally to a pairing
CDiv(X ) Z1 (X ) Z.
293
between Cartier divisors and 1-cycles. In order to get a nondegenerate pairing, we

work over R and mod out by numerical equivalence.
Definition 6.3.17. For a normal variety X , define
N 1 (X ) = (CDiv(X )/ ) Z R and N1 (X ) = (Z1 (X )/ ) Z R.
It follows easily that we get a well-defined nondegenerate bilinear pairing
N 1 (X ) N1 (X ) R.
A deeper fact is that N 1 (X ) and N1 (X ) have finite dimension over R. Thus N 1 (X )

and N1 (X ) are dual vector spaces via intersection product.
The Nef and Mori Cones. The vector spaces N 1 (X ) and N1 (X ) contain some interesting cones.
(a) Nef(X ) is the cone in N 1 (X ) generated by classes of nef Cartier divisors. We
call Nef(X ) the nef cone.
(b) NE(X ) is the cone in N1 (X ) generated by classes of irreducible complete
curves.
(c) NE(X ) is the closure of NE(X ) in N1 (X ). We call NE(X ) the Mori cone.
Here are some easy observations about the nef and Mori cones.
Lemma 6.3.19.
(a) Nef(X ) and NE(X ) are closed convex cones and are dual to each other, i.e.,
Nef(X ) = NE(X )
and
NE(X ) = Nef(X ) .
(b) NE(X ) has full dimension in N1 (X ).

(c) Nef(X ) is strongly convex in N 1 (X ).
Proof. It is obvious that Nef(X ), NE(X ) and NE(X ) are convex cones, and Nef(X )
is closed since it is defined by inequalities of the form D C 0. In fact,
Nef(X ) = NE(X )
by the definition of nef. Then Nef(X ) = NE(X ) follows easily. In general, NE(X )
need not be closed. However, since the closure of a convex cone is its double dual,
we have
NE(X ) = NE(X ) = Nef(X ) .
Note that the cone NE(X ) has full dimension since N1 (X ) is spanned by the classes
of irreducible complete curves. Hence the same is true for its closure NE(X ). Then
Nef(X ) is strongly convex since its the dual has full dimension.
294
Let X be a toric variety whose fan has convex support of full dimension
and set Pic(X )R = Pic(X ) Z R. We have an inclusion
Pic(X ) Pic(X )R
since Pic(X ) is torsion-free by Proposition 4.2.5. Thus Pic(X ) is a lattice in the

vector space Pic(X )R . Note also that
(6.3.2)
Pic(X )R = N 1 (X )
since numerical and linear equivalence coincide by Proposition 6.3.15. In this case,
we will write Pic(X )R instead of N 1 (X ).
Theorem 6.3.20. Let X be a toric variety whose fan has convex support of full
dimension. Then:
(a) Nef(X ) is a rational polyhedral cone in Pic(X )R .
(b) NE(X ) = NE(X ) is a rational polyhedral cone in N1 (X ). Furthermore,
X
NE(X ) =
R0 [V ( )],
a wall of
where [V ( )] N1 (X ) is the class of V ( ).
Proof. Part (a) is an immediate consequence of part (b). For part (b), let =
P
a wall of R0 [V ( )] and note that is a rational polyhedral cone contained in
NE(X ). Furthermore, Theorem 6.3.12 easily implies Nef(X ) = . Then
NE(X ) = Nef(X ) = = NE(X ) NE(X ),
where the third equality follows since is polyhedral.
The formula from part (b) of Theorem 6.3.20, namely

X
NE(X ) =
R0 [V ( )],
a wall of
is called the Toric cone theorem. Although the Mori cone equals NE(X ) in this
case, we will continue to write NE(X ) for consistency with the literature. For the
same reason we write R0 [V ( )] instead of Cone([V ( )]).
The Mori cone of an arbitrary normal variety can have a complicated structure.
The cone theorem shows that some parts of the Mori cone are locally polyhedral.
See [179, Ch. 3] and [194, Ch. 7] for a discussion of this important result.
Since every irreducible complete curve C X gives a class in NE(X ), we
get the following corollary of the toric cone theorem.
Corollary 6.3.21. Assume the fan has convex support of full dimension. Then
any irreducible complete curve on X is numerically equivalent to a non-negative
linear combination of torus-invariant complete curves.

When X is projective we can say more about the nef and Mori cones.
295
Theorem 6.3.22. Let X be a projective toric variety. Then:

(a) Nef(X ) and NE(X ) are dual strongly convex rational polyhedral cones of
full dimension.
(b) A Cartier divisor D is ample if and only if its class in Pic(X )R lies in the
interior of Nef(X ).
Proof. By hypothesis, X has an ample divisor D. Then D C > 0 for every irreducible curve in X . This easily implies that the class of D lies in the interior of
Nef(X ). Thus Nef(X ) has full dimension and hence its dual NE(X ) is strongly
convex. When combined with Lemma 6.3.19, part (a) follows easily.
The strict inequality D C > 0 also shows that every irreducible curve gives a
nonzero class in N1 (X ). Now suppose that the class of D is in the interior of the
nef cone. Then [D] defines a supporting hyperplane of the origin of the dual cone
NE(X ). Since 0 6= [C] NE(X ) for every irreducible curve C X , we have
D C > 0 for all such C. Hence D is ample by Theorem 6.3.13.

It follows that NE(X ) is strongly convex in the projective case. The rays of
NE(X ) are called extremal rays, which by the toric cone theorem are of the form
R0 [V ( )]. The corresponding walls are called extremal walls.
Here is an example of the cones Nef(X ) and NE(X ).
Example 6.3.23. For the Hirzebruch surface Hr , we showed in Example 6.1.16
that Pic(Hr ) = {a[D3 ] + b[D4 ] | a, b Z}. Figure 15 shows Nef(Hr ) and NE(Hr ).
(0,1)
[D4]
(r,1)
[V(2)]
[D3]
(1,0)
nef cone
[V(4)]
[V(3)] = [V(1)]
(1,0)
Mori cone
Figure 15. The nef and Mori cones of Hr
Here, i = Cone(ui ), so that Di = V (i ). Using both notations helps distinguish

between Nef(Hr ) (built from divisors) and NE(Hr ) (built from curves).
The description of the nef cone follows from the characterization of ample
divisors on Hr given in Example 6.1.16. The Mori cone is generated by the classes
of the V (i ) by the toric cone theorem. Using the the basis given by D3 = V (3 ),
296
D4 = V (4 ) and the linear equivalences

D1 D3 ,
D2 rD3 + D4
from Example 6.1.16, we get the picture of NE(Hr ) shown in Figure 15. It follows
that [V (2 )] and [V (3 )] = [V (1 )] generate extremal rays, while [V (4 )] does not.
Thus 1 , 2 , 3 are extremal walls.
The explicit duality between the cones Nef(X ) and NE(X ) in Figure 15 will
be described in the next section.
Theorem 6.3.22 tells us that ample divisors correspond to lattice points in the
interior of Nef(Hr ). Thus lattice points on the boundary correspond to divisors
that are basepoint free but not ample. We can see this vividly by looking at the
polytopes PD associated to divisors D whose classes lie in Nef(Hr ).
PD =
PD =
PD =
nef cone
Figure 16. Polytopes PD associated to divisors D in nef cone of Hr
Figure 16 shows that when D is in the interior of the nef cone, PD is a polygon
whose normal fan is the fan of Hr . On the boundary of the nef cone, however,
things are different: PD is a triangle on the vertical ray and and a line segment on
the horizontal ray. This follows from Figures 10 and 11 in Example 6.2.6.
The Simplicial Case. When X is complete and simplicial, a result to be proved

in 6.4 gives the following criterion for X to be projective.
Proposition 6.3.24. A complete simplicial toric variety X is projective if and only
if its nef cone Nef(X ) Pic(X )R has full dimension in Pic(X )R .
Proof. One direction is an immediate consequence of Theorem 6.3.22. For the
converse, suppose that Nef(X ) has full dimension. Then we can find a Cartier
divisor D whose class lies in the interior of Nef(X ). Since Nef(X ) = NE(X ) , it
follows that D C > 0 for every curve C whose class in NE(X ) is nonzero. Hence,
if we can show that the torus-invariant curves V ( ), a wall, give nonzero
classes in NE(X ), then Theorem 6.3.13 will imply that D is ample, proving that
X is projective.
297
A wall is a facet of some maximal cone , and since is simplicial,

there is (1) such that (1) = (1) {}. Then Lemma 6.4.2 implies that
D V ( ) > 0. Hence the class of V ( ) in NE(X ) is nonzero.

Here is a nice application of this result.
Proposition 6.3.25. A complete toric surface X is projective.
Proof. Picking a basis of N, we may assume N = Z2 and NR = R2 . Let (1) =
{1 , . . . , r } and pick i i with ||i || = 1. Then define : R2 R such that
is linear on the cones of and satisfies (i ) = 1 for all i. The tent analogy (see
Figure 5 in Example 6.1.3) shows that is strictly convex with respect .
P
P
Let D = ri=1 (ui )Di = ri=1 ||ui ||Di , where ui Z2 is the minimal generator of i . Note that [D] Pic(X )R since is simplicial. Strict convexity and
the proof of Theorem 6.3.12 imply that D C > 0 for every torus-invariant curve
C X , so that [D] Nef(X ). The strict inequalities show that [D] is an interior
point, so that X is projective by Proposition 6.3.24.

When X is not simplicial, the criterion given in Proposition 6.3.24 can fail.
Here is an example due to Fujino [100].
Example 6.3.26. Consider the complete fan in R3 with six minimal generators
u1 = (1, 0, 1),
u2 = (0, 1, 1),
u3 = (1, 1, 1)
u4 = (1, 0, 1), u5 = (0, 1, 1), u6 = (1, 1, 1)

and six maximal cones
Cone(u1 , u2 , u3 ), Cone(u1 , u2 , u4 ), Cone(u2 , u4 , u5 )
Cone(u1 , u3 , u4 , u6 ), Cone(u2 , u3 , u5 , u6 ), Cone(u4 , u5 , u6 ).
You will draw a picture of this fan in Exercise 6.3.5 and show that the resulting
complete toric variety satisfies
Pic(X ) {a(D1 + D4 ) | a 3Z} Z.
The cones = Cone(u1 , u2 , u4 ) and = Cone(u2 , u4 , u5 ) meet along the wall
= = Cone(u2 , u4 ).
P
However, any Cartier divisor D = 6i=1 ai Di satisfies m = m (Exercise 6.3.5),
so that the irreducible complete curve C = V ( ) satisfies
D C = 0
by Proposition 6.3.8. This holds for all Cartier divisors on X , so C 0. Then X

has no ample divisors by the toric Kleiman criterion, so that X is nonprojective.
By Exercise 6.3.5, the nef cone of X is the half-line
Nef(X ) = {a[D1 + D4 ] | a 0}.
298
This has full dimension in Pic(X )R , yet X is not projective. Note also that the
Cartier divisor D = 3(D1 + D4 ) gives a class in the interior of the nef cone, yet D
is not ample. Hence part (b) of Theorem 6.3.22 also fails for X . The failure is due
to the existence of irreducible curves in X that are numerically equivalent to zero.
This shows that numerical equivalence of curves can be badly behaved in complete
toric varieties that are neither projective nor simplicial.
Exercises for 6.3.

6.3.1. Let X be a normal variety. Prove that (6.3.1) gives a well-defined pairing between
Q-Cartier divisors and irreducible complete curves. Also show that this pairing satisfies
Propostion 6.3.7.
6.3.2. Derive the formulas for m and m given in Example 6.3.9.
6.3.3. Prove Theorem 6.3.13.
6.3.4. Prove that on a complete normal variety, an ample divisor D satisfies D C > 0 for
all irreducible curves C X.
6.3.5. Consider the fan from Example 6.3.26.
(a) Draw a picture of this fan in R3 .
(b) Prove that Pic(X ) {a(D1 + D4 ) | a 3Z}.
(c) Prove that 3(D1 + D4 ) is nef.
6.4. The Simplicial Case

Here we assume that is a simplicial fan in NR Rn . Then Proposition 4.2.7
implies that every Weil divisor is Q-Cartier. Since we will be working in Pic(X )R ,
it follows that we can drop the adjective Cartier when discussing divisors.
Relations Among Minimal Generators. We begin our discussion of the simplicial
case with another way to think of elements of N1 (X ). Recall from Theorem 4.1.3
that we have an exact sequence
(6.4.1)
M Z(1) Cl(X ) 0
where (m) = (hm, u i)(1) and sends the standard basis element e Z(1)
to [D ] Cl(X ).
Proposition 6.4.1. Let be a simplicial fan in NR with convex support of full
dimension. Then there are dual exact sequences
0 MR R(1) Pic(X )R 0
and
0 N1 (X ) R(1) NR 0,
299
where
(e ) = u ,
e a standard basis vector of R(1)
([C]) = (D C)(1) ,
C X an irreducible complete curve.
Thus N1 (X ) can be interpreted as the spacePof linear relations among

Pthe minimal
generators of . Furthermore, given D = a D and a relation b u = 0,
the intersection pairing of [D] Pic(X )R and R = (b )(1) N1 (X ) is
X
[D] R =
a b .
Proof. Since is simplicial, all Weil divisors are Q-Cartier. Hence

Pic(X )R = Pic(X ) Z R = Cl(X ) Z R.
Tensoring with R preserves exactness, so exactness of the first sequence follows

from (6.4.1). Note also that Pic(X )R = N 1 (X ) by (6.3.2).
The dual of an exact sequence of finite-dimensional vector spaces is still exact.
Then the perfect pairings
MR NR R : (m, u) 7 hm, ui
Pic(X )R N1 (X ) R : ([D], [C]) 7 D C

easily imply that for m MR and [C] N1 (X ), we have
(m) = (hm, u i)(1) = (e ) = u
and
(e ) = [D ] = ([C]) = (D C)(1) .
Finally, the duality between the two exact sequences reduces to dot product on the
middle terms R(1) . This proves the final assertion of the proposition.

The map : N1 (X ) R(1) in Proposition 6.4.1 implies that an irreducible
complete curve C X gives the relation
P
(D C) u = 0 in NR .
This can be proved directly as follows. First observe that m M gives
P
m
hm, u iD = div( ) 0.
Taking the intersection product with C, we see that

P
hm, u i(D C) = 0
P

holds for all m MR . Writing this as m, (D C) u = 0, we obtain
P
(6.4.2)
(D C) u = 0 in NR .
This argument shows that (6.4.2) holds for any simplicial toric variety.
300
Intersection Products. Our next task is to compute D C when C is a torusinvariant complete curve in X . This means C = V ( ), where (n 1) is a
wall, i.e., the intersection of two cones in (n). Here, we only assume that is a
simplicial fan in NR Rn , with no hypotheses on its support.
We begin with an easy case. Fix a wall
= .
Since is simplicial, we can label the minimal generators of so that

= Cone(u1 , u2 , . . . , un )
= Cone(u2 , . . . , un ).
Thus is the facet of opposite to 1 . We will compute the intersection product D1 V ( ) in terms of the multiplicity (also called the index) of a simplicial
cone. This is defined as follows. If is a simplicial cone with minimal generators
u1 , . . . , uk , then mult() is the index of the sublattice
Zu1 + + Zuk N = Span() N = (Ru1 + + Ruk ) N.
Lemma 6.4.2. If , and 1 are as above, then
D1 V ( ) =
mult( )
.
mult()
Proof. Since {u1 , . . . , un } is a basis of NQ , we can find m MQ such that

(
1 i = 1
hm, ui i =
0
i = 2, . . . , n.
Pick a positive integer such that m M. On U U , D1 is the Cartier divisor
determined by m = m and m = 0. By (6.3.1) and Proposition 6.3.8,
1
1
D1 V ( ) = (D1 ) V ( ) = hm, ui = hm, ui,
where u maps to a generator of N( ). Recall that N( ) = N/N .
When we combine u with a basis of N , we get a basis of N. Thus there is a

positive integer such that u1 = u+ v, v N . The minus sign appears because
u and u1 lie on opposite sides of . By considering the sublattices
Zu1 + Zu2 + + Zun Zu1 + N Zu + N = N,
one sees that = mult()/mult( ). Thus

1
mult( )
u = u1 v =
u1 v .
mult()
Since m , it follows that
D1 V ( ) = hm, ui =
mult( )
mult( )
hm, u1 i =
.
mult()
mult()
301
Corollary 6.4.3. Let be a simplicial fan in NR Rn and (n 1) be a wall.

If (1) and generate a smooth cone of (n), then
D V ( ) = 1
Given a wall (n 1), our next task is to compute D V ( ) for the other
rays (1). Let = and write
= Cone(u1 , . . . , un )
(6.4.3)
= Cone(u2 , . . . , un+1 )
= Cone(u2 , . . . , un ).
This situation is pictured in Figure 17.
n+1
Figure 17. =
The n + 1 minimal generators u1 , . . . , un+1 are linearly dependent. Hence they

satisfy a linear relation, which we write as
n
X
bi ui + un+1 = 0.
(6.4.4)
u1 +
i=2
We may assume , > 0 since u1 and un+1 lie on opposite sides of the wall .
Then (6.4.4) is unique up to multiplication by a positive constant since u1 , . . . , un
are linearly independent. We call (6.4.4) a wall relation.
On the other hand, setting C = V ( ) in (6.4.2) gives the linear relation
X
(6.4.5)
(D V ( )) u = 0
302
As we now prove, the two relations are the same up to a positive constant.
Proposition 6.4.4. The relations (6.4.4) and (6.4.5) are equal after multiplication
by a positive constant. Furthermore:
(a) D V ( ) = 0 for all
/ {1 , . . . , n+1 }.
mult( )
mult( )
(b) D1 V ( ) =
and Dn+1 V ( ) =
.
mult()
mult( )
bi mult( )
bi mult( )
=
for i = 2, . . . , n.
(c) Di V ( ) =
mult() mult( )
Proof. Part (b) follows immediately from Lemma 6.4.2. Also observe that when
/ {1 , . . . , n+1 }, and never lie in the same cone of , so D V ( ) = by

the Orbit-Cone Correspondence. This easily implies D V ( ) = 0 (Exercise 6.4.1).
This proves part (a) and implies that (6.4.5) reduces to the equation
(D1 V ( )) u1
n
X
(Di V ( )) ui + (Dn+1 V ( )) un+1 = 0.
+
i=2
The coefficients of u1 and un+1 are positive by part (b), so up to a positive constant, this must be the wall relation (6.4.4). The first assertion of the lemma follows.
Since the above relation equals (6.4.4) up to a nonzero constant, we obtain
bi (D1 V ( )) = (Di V ( )),
bi (Dn+1 V ( )) = (Di V ( )),
for i = 2, . . . , n. Then the formulas for Di V ( ) in part (c) follow from part (b).
For a simplicial toric variety, Proposition 6.4.4 provides everything we need to

compute D V ( ) when is a wall of .
Example 6.4.5. Consider the fan in R2 from Example 6.3.9. We have the wall
= Cone(u0 ) = = Cone(u1 , u0 ) Cone(u2 , u0 ),
where u1 = e1 , u2 = e2 and u0 = 2e1 + 3e2 . Computing multiplicities gives
mult( ) = 1, mult() = 3, mult( ) = 2.
Then part (b) of Proposition 6.4.4 implies
1
1
D1 V ( ) = , D2 V ( ) = ,
3
2
and the relation
implies
2 u1 + (1) u0 + 3 u2 = 0
1
1 1 1 1
=
=
23
32
6
by part (c) of the proposition. Hence we recover the intersection products computed
in Example 6.3.9.
D0 V ( ) =
303
When X is smooth, all multiplicities are 1. Hence the wall relation (6.4.4) can
be written uniquely as
u1 +
(6.4.6)
n
X
bi ui + un+1 = 0,
i=2
bi Z,
and then the intersection formulas of Proposition 6.4.4 reduce to

(6.4.7)
D1 V ( ) = Dn+1 V ( ) = 1,
Di V ( ) = bi , i = 2, . . . , n.
Example 6.4.6. For the Hirzebruch surface Hr , the four curves coming from walls
are also divisors. Recall that the minimal generators are
u1 = e1 + re2 , u2 = e2 , u3 = e1 , u4 = e2 ,
arranged clockwise around the origin (see Figure 3 from Example 6.1.2 for the case
r = 2). Hence the wall generated by u1 gives the relation
u2 0 u3 + u4 = 0,
which implies
D1 D1 = 0
by (6.4.7). On the other hand, the wall generated by u2 gives the relation
u1 r u2 + u3 = 0.
Then (6.4.7) implies
Similarly, one can check that
D2 D2 = r.
D3 D3 = 0, D4 D4 = r,
and by Corollary 6.4.3 or (6.4.7) we also have
D1 D2 = D2 D3 = D3 D4 = D4 D1 = 1.
These computations give an explicit description of the duality between the nef and
Mori cones shown in Figure 15 of Example 6.3.23 (Exercise 6.4.2).
In general, a Q-Cartier divisor D on a complete surface has self-intersection

D D = D 2 . Self-intersections will play a prominent role in Chapter 10 when we
study toric surfaces.
Example 6.4.7. Let X be a complete toric surface. Write (1) = {1 , . . . , r },
where the i are arranged clockwise around the origin in NR R2 . Each i gives a
minimal generator ui and a toric divisor Di . Note also that Pic(X )R Rr2 .
Proposition 6.3.25 tells us that X is projective, so that its Mori cone NE(X )
is strongly convex of dimension r 2. Hence a minimal generating set has at
least r 2 elements. Since the r classes [Di ] = [V (i )] generate by the toric cone
theorem, we almost know the minimal generators.
304
Now suppose that X is smooth. Then the wall relation for Di = V (i ) is

2
uP
i1 + bi ui + ui+1 = 0 by (6.4.6), where bi = Di by (6.4.7). Given a divisor D =
r
i=1 ai Di , Proposition 6.4.1 implies that
D Di = ai1 + bi ai + ai+1 ,
so that D is nef (resp. ample) if and only if

ai1 + bi ai + ai+1 0 (resp. > 0)
for i = 1, . . . , r. This makes it easy to study nef and ample divisors on X .
Primitive Collections. In the projective case, there is a beautiful criterion for a

Cartier divisor to be nef or ample in terms of the primitive collections introduced
in Definition 5.1.5. Recall that
P = {1 , . . . , k } (1)
is a primitive collection if P is not contained in (1) for all but any proper
subset is. Since is simplicial, primitive means that P does not generate a cone of
but every proper subset does. This is the definition given by Batyrev in [14].
Example 6.4.8. Consider the complete fan in R3 shown in Figure 18. One can
check that
{1 , 3 }, {0 , 2 , 4 }
are the only primitive collections of .
3
2
y
x
0
Figure 18. A complete fan in R3 with two primitive collections
Here is the promised characterization, due to Batyrev [14] in the smooth case.
305
Theorem 6.4.9. Let X be a projective simplicial toric variety. Then:

(a) A Cartier divisor D is nef if and only if its support function D satisfies
D (u1 + + uk ) D (u1 ) + + D (uk )
for all primitive collections P = {1 , . . . , k } of .
(b) A Cartier divisor D is ample if and only if its support function D satisfies
D (u1 + + uk ) > D (u1 ) + + D (uk )
for all primitive collections P = {1 , . . . , k } of .
Before we discuss the proof of Theorem 6.4.9, we need to study the relations
that come from primitive collections.
Definition 6.4.10. Let P = {1 , . . . , k } (1) be a primitive collection for the
P
complete simplicial fan . Hence ki=1 ui lies in the relative interior of a cone
. Thus there is a unique expression
P
u1 + + uk = (1) c u , c Q>0 .
P
Then u1 + + uk (1) c u = 0 is the primitive relation of P.
The coefficient vector of this relation is r(P) = (b )(1) R(1) , where
1
P,
/ (1)
1 c P (1)
(6.4.8)
b =
c
(1),
/P
0
otherwise.
P
Then b u = 0, so that r(P) gives an element of N1 (X ) by Proposition 6.4.1.
In Exercise 6.4.3, you will prove that c < 1 when P (1). This means that P
is determined by the positive entries in the coefficient vector r(P).
The Mori cone for X has a nice description in terms of primitive relations.
Theorem 6.4.11. If X is a projective simplicial toric variety, then
X
NE(X ) =
R0 r(P),
P
where the sum is over all primitive collections P of .

Proof. Given a Cartier divisor D =
[D] r(P) =
a D ,
a b =
k
X
i=1
Proposition 6.4.1 shows that

ai
(1)
a c .
306
Since the support function of D satisfies D (u ) = a and is linear on , we can

rewrite this as
(6.4.9)
[D] r(P) = D (u1 ) D (uk ) + D (u1 + + uk ).
If D is nef, then it is basepoint free (Theorem 6.3.12), so D is convex. It follows

that [D] r(P) 0, which proves r(P) NE(X ). Note also that r(P) is nonzero.
To finish the proof, we need to show that NE(X ) is generated by the r(P).
In the discussion following the proof of Theorem 6.3.22, we noted that NE(X )
is generated by its extremal rays, each of which is of the form R0 [V ( )] for an
extremal wall . It suffices to show that [V ( )] is a positive multiple of r(P) for
some primitive collection P.
We first make a useful observation about nef divisors. Given a cone , we
claim that any nef divisor is linearly equivalent to a divisor of the form
P
(6.4.10)
D = a D , a = 0, (1) and a 0,
/ (1).
To prove this, first recallP

that any nef divisor is linearly equivalent to a torusinvariant nef divisor D = a D . Then we have m M with hm , u i = a
for (1). Since D is nef, it is also basepoint free, so that
hm , u i D (u ) = a ,
by Theorem 6.1.7. Replacing D with
(1),
D + div( m ),
we obtain (6.4.10).
Now assume we have an extremal wall and let C = V ( ). Consider the set
P = { | D C > 0}.
We will prove that P is a primitive collection whose primitive relation is the class
of C, up to a positive constant. Our argument is taken from [71], which is based on
ideas of Kresch [181].
We first prove by contradiction that P 6 (1) for all . Suppose P (1)
and take an ample divisor D (remember that X is projective). Then in particular D
is nef, so we may assume that D is of the form (6.4.10). Since a = 0 for (1),
we have
P
D C = (1)
a D C.
/
However, a 0 by (6.4.10), and P (1) implies D C 0 for

/ (1). It
follows that D C 0, which is impossible since D is ample. Thus no cone of
contains all rays in P.
It follows that some subset Q P is a primitive collection. This gives the
primitive relation with coefficient vector r(Q) N1 (X ), and we also have the
class [C] N1 (X ). Let
= [C] r(Q) N1 (X ),
where > 0. We claim that if is sufficiently small, then

(6.4.11)
{ | [D ] < 0} { | D C < 0}.
307
To prove this, first observe that the definition of implies

D C = [D ] r(Q) + [D ] .
Suppose [D ] < 0 and D C 0. This forces [D ] r(Q) > 0. Proposition 6.4.1
implies that [D ] r(Q) is the coefficent of u in the primitive relation of Q, which
by (6.4.8) is positive only when Q. Then Q P implies D C > 0 by the
definition of P. But we can clearly choose sufficiently small so that
D C > [D ] r(Q)
whenever D C > 0.
This inequality and the above equation imply [D ] > 0, which is a contradiction.
We next claim that NE(X ). By (6.4.11), we have
{ | [D ] < 0} { | D C < 0} (1),

where the second inclusion follows from C = V ( ) and Proposition 6.4.4. Now let
D be nef, and by (6.4.10) with = , we may assume that
X
D=
a D , a = 0, (1) and a 0,
/ (1).
Then
[D] =
/ (1)
a [D ] 0,
where the final inequality follows since a 0 and [D ] < 0 can happen only
when (1). This proves that NE(X ).
We showed earlier that r(Q) NE(X ). Thus the equation
[C] = r(Q) +
expresses [C] as a sum of elements of NE(X ). But [C] is extremal, i.e., it lies in
a 1-dimensional face of NE(X ). By Lemma 1.2.7, this forces r(Q) and to lie
in the face. Since r(Q) is nonzero, it generates the face, so that [C] is a positive
multiple of r(Q).
The relation corresponding to C has coefficients (D C)(1) , and P is the
set of s where D C > 0. But this relation is a positive multiple of r(Q), whose
positive entries correspond to Q. Thus P = Q and the proof is complete.

It is now straightforward to prove Theorem 6.4.9 using Theorem 6.4.11 and
(6.4.9) (Exercise 6.4.4). We should also mention that these results hold more generally for any projective toric variety (see [71]).
Example 6.4.12. Let be the fan shown in Figure 18 of Example 6.4.8. The
minimal generators of 0 , . . . , 4 are
u0 = (0, 0, 1), u1 = (0, 1, 1), u2 = (1, 0, 1), u3 = (0, 1, 1), u4 = (1, 0, 1).
308
The computations you did for part (c) of Exercise 6.1.8 imply that X is projective. Since the only primitive collections are {1 , 3 } and {0 , 2 , 4 }, Theorem 6.4.9 implies that a Cartier divisor D is nef if and only if
D (u1 + u3 ) D (u1 ) + D (u3 )
D (u0 + u2 + u4 ) D (u0 ) + D (u2 ) + D (u4 )

and ample if and only if these inequalities are strict. One can also check that
Pic(X ) {a[D1 ] + b[D2 ] | a, b 2Z}
and aD1 + bD2 is nef (resp. ample) if and only if a b 0 (resp. a > b > 0).
Exercise 6.4.5 will relate this example to the proof of Theorem 6.4.11.
Besides , the minimal generators u0 , . . . , u4 support two other complete fans
in R3 : first, the fan obtained by replacing Cone(u2 , u3 ) with Cone(u1 , u3 ) in
Figure 18, and second, the fan 0 obtained by removing this wall to create the cone
Cone(u1 , u2 , u3 , u4 ). Since (1) = (1) = 0 (1), the toric varieties X , X , X0
have the same class group, though X0 has strictly smaller Picard group since it is
not simplicial. Thus
Pic(X0 )R Pic(X )R = Pic(X )R R2 .
This allows us to draw all three nef cones in the same copy of R2 . In Exercise 6.4.5
you show that we get the cones shown in Figure 19. The ideas behind this figure
Nef(X0 )
Nef(X)
Nef(X )
Figure 19. The nef cones of X , X , X0
will be developed in Chapters 14 and 15 when we study geometric invariant theory

and the minimal model program for toric varieties.
Exercises for 6.4.

6.4.1. This exercise will describe a situation where D C is guaranteed to be zero.
(a) Let X be normal and assume that C is a complete irreducible curve disjoint from the
support of a Cartier divisor D. Prove that D C = 0. Hint: Use U = X \ Supp(D).
309
(b) Let be a wall of a fan and pick (1) such that and do not lie in the same
cone of . Use the Cone-Orbit Correspondence to prove that D V ( ) = , and
conclude that D V ( ) = 0.
6.4.2. As in Example 6.4.2, the classes [D3 ], [D4 ] give a basis of Pic(Hr )R . Since Hr is a
smooth complete surface, the intersection product Di V ( j ) is written Di D j .
(a) Give an explicit formula for (a[D3 ] + b[D4]) (a[D3 ] + b[D4]) using the computations
of Example 6.4.2.
(b) Use part (a) to show that the cones shown in Figure 15 in Example 6.3.23 are dual to
each other.
6.4.3. In the primitive relation defined in Definition 6.4.10, prove c < 1 when P
(1). Hint: If 1 (1) and c1 1, then cancel u1 and show that u2 , . . . , uk .
6.4.4. Prove Theorem 6.4.9 using Theorem 6.4.11 and (6.4.9).
6.4.5. Consider the fan from Examples 6.4.8 and 6.4.12. Every wall of is of the
form i j = Cone(ui , u j ) for suitable i < j. Let r(i j ) R5 denote the wall relation of i j .
Normalize by a positive constant so that the entries of r(i j ) are integers with gcd = 1.
(a) Show the nine walls give the three distinct wall relations r(02 ), r(03 ), r(24 ).
(b) Show that r(03 ) + r(24 ) = r(02 ) and conclude that 03 and 24 are extremal walls
whose classes generate the Mori cone of X .
(c) For each extremal wall of part (b), determine the corresponding primitive collection.
You should be able to read the primitive collection directly from the wall relation.
(d) Show that the nef cones of X , X , X0 give the cones shown in Figure 19.
6.4.6. Let X be the blowup of Pn at a fixed point of the torus action. Thus Pic(X ) Z2 .
(a) Compute the nef and Mori cones of X and draw pictures similar to Figure 15 in
Example 6.3.23.
(b) Determine the primitive relations and extremal walls of X .
6.4.7. Let Pr be the toric surface obtained by changing the ray u1 in the fan of the Hirzebruch surface Hr from (1, r) to (r, 1). Assume r > 1.
(a) Prove that Pr is singular. How many singular points are there?
(b) Determine which divisors a1 D1 + a2 D2 + a3 D3 + a4 D4 are Cartier and compute Di D j
for all i, j.
(c) Determine the primitive relations and extremal walls of Pr .

Now that we know more about sheaves (specifically, tensor products and exactness), we
can complete the discussion of quasicoherent sheaves on toric varieties begun in 5.3. In
this appendix, X will denote a toric variety with no torus factors. The total coordinate
ring of X is S = C[x | (1)], which is graded by Cl(X ).
Recall from 5.3 that for Cl(X ), the shifted S-module S() gives the sheaf
OX () satisfying OX () OX (D) for any Weil divisor with = [D]. In 6.0 we constructed a sheaf homomorphism OX (D) OX OX (E) OX (D + E). In a similar way, one
310
can define
(6.A.1)
OX () OX OX () OX ( + ).
for , Cl(X ) such that if = [D] and = [E], then the diagram
OX (D)) OX OX (E))
/ OX (D + E))

OX () OX OX ()

/ OX ( + )
commutes, where the vertical maps are isomorphisms.

From Sheaves to Modules. The main construction of 5.3 takes a graded S-module M and
e on X . We now go in the reverse direction and show that
produces a quasicoherent sheaf M
every quasicoherent sheaf on X arises in this way. We will use the following construction.
Definition 6.A.1. For a sheaf F of OX -modules on X and Cl(X ), define
F () = F OX OX ()
and then set

(F ) =
(X , F ()).
Cl(X )
For example, (OX ) = S since (X , OX ()) S by Proposition 5.3.7. Using

this and (6.A.1), we see that (F ) is a graded S-module.
We want to show that F is isomorphic to the sheaf associated to (F ) when F is
quasicoherent. We will need the following
lemma due to Mustata [212]. Recall that for
Q
, we have the monomial x = (1)
x S. Let = deg(x ) Cl(X ).
/
Lemma 6.A.2. Let F be a quasicoherent sheaf on X . Then:
(a) If v (U , F ), then there are 0 and u (X , F ( )) such that u restricts to

(x ) v (U , F ( )).
(b) If u (X , F ) restricts to 0 in (U , F ), then there is 0 such that (x ) u = 0 in

(X , F ( )).
Proof. For part (a), fix and take v (U , F ). Given , let v be the restriction
of v to U U . By (3.1.3), we can find m ( ) M such that U U = (U ) m =
Spec(C[ M] m ). In terms of the total coordinate ring S, we have C[ M] (Sx )0
by (5.3.1). Hence the coordinate ring of U U is the localization

(Sx )0 xhmi ,
Q hm,u i
where xhmi = x (Sx )0 since m M. This enables us to write
U U = (U )xhmi .
Since F is quasicoherent, F |U is determined by its sections G = (U , F ), and then

(U U , F ) is the localization Gxhmi .
It follows that v (U , F ) equals u /(xhmi )k , where k 0 and u (U , F ).

Hence u restricts to (xhmi )k v (U , F ). Since m ( ), we see that
(6.A.2)
x a = (x ) (xhmi )k S
311
for 0. This monomial has degree . Then u = x a u (U , F ( )) restricts

to (x ) v (U U , F ( )). By making sufficiently large, we can find one that
works for all .
To study whether the u patch to give a global section of F ( ), take 1 , 2 and

set = 1 2 . Thus U = U1 U2 , and
(6.A.3)
w = u1 |U u2 |U (U , F ( ))
restricts to 0 (U U , F ( )). Arguing as above, this group of sections is the localization (U , F ( ))xhmi , where m ( ) M such that U U = (U ) m .
Since w gives the zero element in this localization, there is k 0 with (xhmi )k w = 0 in
(U , F ( )). If we multiply by x b = (x ) (xhmi )k for 0, we obtain (x ) w = 0

in (U , F (( + ) )). Another way to think of this is that if we made in (6.A.2) big
enough to begin with, then in fact w = 0 in (U , F ( )) for all , . Given the definition (6.A.3) of w, it follows that the u patch to give a global section u (X , F ( ))
with the desired properties.
The proof of part (b) is similar and is left to the reader.
Proposition 6.A.3. Let F be a quasicoherent sheaf on X . Then F is isomorphic to the

sheaf associated to the graded S-module (F ).
e to
Proof. Let M = (F ) and recall from 5.3 that for every , the restriction of M
U is the sheaf associated to the (Sx )0 -module (Mx )0 .
e F . Elements of (Mx )0 are u/(x )
We first construct a sheaf homomorphism M

for u (X , F ( )). Since (x ) is a section of OX ( ) over U , the map
(U , OX ( )) C (U , F ( )) (U , F )
induces a homomorphism of (Sx )0 -modules

(Mx )0 (U , F ).
(6.A.4)
e | F | that patch to give M

e F.
This gives compatible sheaf homomorphisms M
U
U
Since F is quasicoherent, it suffices to show that (6.A.4) is an isomorphism for every

. First suppose that u/(x )k (Mx )0 maps to 0 (U , F ). It follows easily that
u restricts to zero in (U , F (k )). By Lemma 6.A.2 applied to F (k ), there is 0
such that (x ) u = 0 in (X , F (( + k) )). Then
u
(x )k
(x ) u
= 0 in (Mx )0 ,
(x )+k
which shows that (6.A.4) is injective. To prove surjectivity, take v (U , F ) and apply
Lemma 6.A.2 to find 0 and u (X , F ( )) such that u restricts to (x ) v. It
follows immediately that u/(x ) (Mx )0 maps to v.

This result proves part (a) of Proposition 5.3.9. We now turn our attention to part (b)
of the proposition, which applies to coherent sheaves.
Proposition 6.A.4. Every coherent sheaf F on X is isomorphic to the sheaf associated
to a finitely generated graded S-module.
312
Proof. On the affine open subset U , we can find finitely many sections fi, (U , F )
which generate F over U . By Lemma 6.A.2, we can find 0 such that (x ) fi, comes
from a global section gi, of F ( ). Now consider the graded S-module M (F )
generated by the gi, . Proposition 6.A.3 gives an isomorphism
^
(F ) F .
e F which is injective by the exactHence M (F ) gives a sheaf homomorphism M
ness proved in Example 6.0.10. Over U , we have fi, = gi, /(x ) (Mx )0 , and since
e F . Then we are done since M is
these sections generate F over U , it follows that M
clearly finitely generated.
The proof of Proposition 6.A.4 used a submodule of (F ) because the full module
need not be finitely generated when F is coherent. Here is an easy example.
Example 6.A.5. A point p P n gives a subvariety i : {p} P n . The sheaf F = i O{p}
can be thought of as a copy of C sitting over the point p. The line bundle OPn (a) is free in
a neighborhood of p, so that F (a) F for all a Z. Thus
M
M
(F ) =
(P n , F (a)) =
C.
aZ
aZ
This module is not finitely generated over S since it is nonzero in all negative degrees.
Subschemes and Homogeneous Ideals. For readers who know about schemes, we can
apply the above results to describe subschemes of a toric variety X with no torus factors.
Let I S be a homogeneous ideal. By Proposition 6.0.10, this gives a sheaf of ideals
I OX whose quotient is the structure sheaf OY of a closed subscheme Y X . This
differs from the subvarieties considered in the rest of the book since the structure sheaf OY
may have nilpotents.
Proposition 6.A.6. Every subscheme Y X is defined by a homogeneous ideal I S.
Proof. Given an ideal sheaf I OX , we get a homomorphism of S-modules
(I ) (OX ) = S.
If I S is the image of this map, then the map factors (I ) I S, where the first
arrow is surjective and the second injective. By Example 6.0.10 and Proposition 6.A.3, the

inclusion I OX factors as I I OX . It follows immediately that I = I.
In the case of P n , it is well-known that different graded ideals can give the same ideal
sheaf. The same happens in the toric situation, and as in 5.3, we get the best answer in
the smooth case. Not surprisingly, the irrelevant ideal B() S plays a key role in the
following result from [65, Cor. 3.8].
Proposition 6.A.7. Homogeneous ideals I, J S in the total coordinate of a smooth toric

variety X give the same ideal sheaf of OX if and only if I : B() = J : B() .
There is a less elegant version of this result that applies to simplicial toric varieties. See
[65] for a proof and more details about the relation between graded modules and sheaves.
See also [204] for more on multigraded commutative algebra.
Chapter 7
Projective Toric
Morphisms
7.0. Background: Quasiprojective Varieties and Projective

Morphisms
Many results of Chapter 6 can be generalized, but in order to do so, we need to
learn about quasiprojective varieties and projective morphisms.
Quasiprojective Varieties. Besides affine and projective varieties, we also have the
following important class of varieties.
Definition 7.0.1. A variety is quasiprojective if it is isomorphic to an open subset
of a projective variety.
Here are some easy properties of quasiprojective varieties.
Proposition 7.0.2.
(a) Affine varieties and projective varieties are quasiprojective.
(b) Every closed subvariety of a quasiprojective variety is quasiprojective.
(c) A product of quasiprojective varieties is quasiprojective.
Proof. You will prove this in Exercise 7.0.1.
Projective Morphisms. In algebraic geometry, concepts that apply to varieties

sometimes have relative versions that apply to morphisms between varieties. For
example, in 3.4, we defined completeness and properness, where the former applies to varieties and the latter applies to morphisms. Sometimes we say that the
313
Chapter 7. Projective Toric Morphisms
314
relative version of a complete variety is a proper morphism. In the same way, the
relative version of a projective variety is a projective morphism.
We begin with a special case. Let f : X Y be a morphism and L a line
bundle on X with a basepoint free finite-dimensional subspace W (X , L ). Then
combining f : X Y with the morphism L ,W : X P(W ) from 6.0 gives a
morphism X Y P(W ) that fits into a commutative diagram
f L ,W
(7.0.1)
/ Y P(W )
X OOO
OOO
OOO
p1
O
f OOOO

O'
Y.
If f L ,W is a closed embedding (meaning that its image Z Y P(W ) is

closed and the induced map X Z is an isomorphism), then you will show in
Exercise 7.0.2 that f has the following nice properties:
f is proper.
For every p Y , the fiber f 1 (p) is isomorphic to a closed subvariety of

P(W ) and hence is projective.
The general definition of projective morphism must include this special case.
In fact, going from the special case to the general case is not that hard.
Definition 7.0.3. A morphism f : X Y is projective if there is a line bundle L
on X and an affine open cover {Ui } of Y with the property that for each i, there is
a basepoint free finite-dimensional subspace Wi ( f 1 (Ui ), L ) such that
fi L ,W
i
f 1 (Ui ) i
Ui P(Wi )
is a closed embedding, where fi = f | f 1 (Ui ) and Li = L | f 1 (Ui ) . We say that

f : X Y is projective with respect to L .
The general case has the properties noted above in the special case.
Proposition 7.0.4. Let f : X Y be projective. Then:
(a) f is proper.
(b) For every p Y , the fiber f 1 (p) is a projective variety.

Here are some further properties.
Proposition 7.0.5.
(a) The composition of projective morphisms is projective.
(b) A closed embedding is a projective morphism.
(c) A variety X is projective if and only if X {pt} is a projective morphism.
7.0. Background: Quasiprojective Varieties and Projective Morphisms
315
Proof. Parts (a) and (b) are proved in [127, (5.5.5)]. For part (c), one direction
follows immediately from the previous proposition. Conversely, let i : X P n be
projective, and assume that X is nondegenerate, meaning that X is not contained in
any hyperplane of P n . Now let L = OX (1) = i OPn (1). Then
i : (P n , OPn (1)) (X , L )
is injective since X is nondegenerate. In Exercise 7.0.3 you will show that
(P n , OPn (1)) = Span(x0 , . . . , xn )
and that if W (X , L ) is the image of i , then L ,W is the embedding we began
with. Hence Definition 7.0.3 is satisfied for X {pt} and L .

When the domain is quasiprojective, the relation between proper and projective
is especially easy to understand.
Proposition 7.0.6. Let f : X Y be a morphism where X is quasiprojective. Then:
f is proper f is projective.
Proof. One direction is obvious since projective implies proper. For the opposite
direction, X is quasiprojective, which implies that there is a morphism
g : X Z
such that Z is projective, g(X ) Z is open, and X g(X ) via g. Then one can
prove without difficulty that the product map
(7.0.2)
f g : X Y Z
induces an isomorphism X ( f g)(X ).
Since f : X Y is proper, f g : X Y Z is also proper (Exercise 7.0.4).

Hence the image of f g is closed in X Z since proper morphisms are universally
closed. Thus X ( f g)(X ) and ( f g)(X ) is closed in Y Z. This proves that
(7.0.2) is a closed embedding.
Now take a closed embedding Z P s . Arguing as above, we get a closed
embedding of X into Y P s . From here, it is straightforward to show that f is
projective (Exercise 7.0.4).

To complicate matters, there are two definitions of projective morphism used
in the literature. In [131, II.4], a projective morphism is defined as the special
case considered in (7.0.1), while [127, (5.5.2)] and [273, 5.3] give a much more
general definition. Theorem 7.A.4 of the appendix to this chapter shows that the
more general definition is equivalent to Definition 7.0.3.
Projective Bundles. Vector bundles give rise to an interesting class of projective
morphisms.
316
Let : V X be a vector bundle of rank n 1. Recall from 6.0 that V has

a trivialization {(Ui , i )} with i : 1 (Ui ) Ui Cn . Furthermore, the transition
functions gi j GLn ((Ui U j , OX )) make the diagram
i | 1
(Ui U j )
Ui U j Cn
kk5
kkk
kkkk
1 (Ui U j )
1gi j
SSSS
SSSS
S)
j | 1
(Ui U j )
Ui U j Cn
commute. Note that 1 gi j induces an isomorphism
1 gi j : Ui U j P n1 Ui U j P n1 .
This gives gluing data for a variety P(V ). It is clear that induces a morphism
: P(V ) X and that i induces the trivialization
i : 1 (Ui ) Ui P n1 .
The discussion following Theorem 7.A.4 in the appendix to this chapter shows that
: P(V ) X is a projective morphism. We call P(V ) the projective bundle of V .
Example 7.0.7. Let W be a finite-dimensional vector space over C of positive
dimension. Then, for any variety X , the trivial bundle X W X gives the trivial
projective bundle X P(W ) X .
There is also a version of this construction for locally free sheaves. If E is

locally free of rank n, then E is the sheaf of sections of a vector bundle VE X of
rank n. When n = 1, we proved this in Theorem 6.0.20. Then define
(7.0.3)
P(E ) = P(VE ),
where VE is the dual vector bundle of VE . Here are some properties of P(E ).
Lemma 7.0.8.
(a) P(L ) = X when L is locally free of rank 1.
(b) P(E OX L ) = P(E ) when E is locally free and L is a line bundle.
(c) If a homomorphism E F of locally free sheaves is surjective, then the induced map P(F ) P(E ) of projective bundles is injective.
Proof. You will prove this in Exercise 7.0.5. The dual in (7.0.3) is the reason why
E F gives P(F ) P(E ).

The appearance of the dual in (7.0.3) can be explained as follows. Let L be a
line bundle with W (X , L ) basepoint free of finite dimension. As in 6.0, this
gives a morphism
L ,W : X P(W ).
7.0. Background: Quasiprojective Varieties and Projective Morphisms
317
Let E = W C OX . The corresponding vector bundle is VE = X W , so

(7.0.4)
P(E ) = P(VE ) = X P(W ).
By Proposition 6.0.24, the natural map E L is surjective since W has no basepoints. By Lemma 7.0.8, we get an injection of projective bundles
P(L ) P(E ).
The lemma also implies P(L ) = X . Using this and (7.0.4), we get an injection
X X P(W ).
Projection onto the second factor gives a morphism X P(W ), which is the
morphism L ,W from 6.0 (Exercise 7.0.6).
Proj of a Graded Ring. As described in [90, III.2] and [131, II.2], a graded ring
S=
Sd
d=0
gives the scheme Proj(S) such that for every non-nilpotent f Sd , d > 0, we have
the affine open subset D+ ( f ) Proj(S) with
D+ ( f ) Spec(S( f ) ),
where S( f ) is the homogeneous localization of S at f , i.e.,
ng
o
S( f ) =
|
g
S
,
N
.
d
f
Furthermore, if homogeneous elements f1 , . . . , fs S satisfy

M
p
Sd ,
h f1 , . . . , fs i = S+ =
d>0
then the affine open subsets D+ ( f1 ), . . . , D+ ( fs ) cover Proj(S). Thus we construct

Proj(S) by gluing together the affine varieties D+ ( fi ), just as we construct P n by
gluing together copies of Cn .
The inclusions S0 S( f ) for all f give a natural morphism Proj(S) Spec(S0 ).
We have the following important result from [131, Prop. II.7.10].
Theorem 7.0.9. The morphism Proj(S) Spec(S0 ) is projective.
Example 7.0.10. Let U = Spec(R) and consider the graded ring

S = R[x0 , . . . , xn ]
such that each xi has degree 1. Then
Proj(S) = U P n ,
where Proj(S) Spec(S0 ) = Spec(R) = U is projection onto the first factor.
318
In [131, II.7], the projective bundle P(E ) of a locally free sheaf E on X is

constructed via a relative version of the Proj construction. More generally, one can
define the projective bundle P(E ) for any coherent sheaf E on X .
Exercises for 7.0.
7.0.2. Prove Proposition 7.0.4. Hint: First prove the special case given by (7.0.1). Recall
from 3.4 that P n is complete, so that P n {pt} is proper.
7.0.4. Let : X Y and : Y Z be morphisms such that : X Z is proper. Prove

that : X Y is also proper. Hint: As noted in the comments following Corollary 3.4.6,
being proper is equivalent to being topologically proper (Definition 3.4.7). Also, T Y
implies 1 (T ) ( )1 ((T )).
7.0.5. Prove Lemma 7.0.8. Hint: Work on an open cover where the bundles are trivial.
7.0.6. In the discussion following (7.0.4), we constructed a morphism X P(W ) using

the surjection E = W C OX L . Prove that this coincides with the morphism L ,W .
7.0.7. Show that C2 \ {(0, 0)} is quasiprojective but neither affine nor projective.
7.1. Polyhedra and Toric Varieties

This section and the next will study quasiprojective toric varieties and projective
toric morphisms. Our starting point is the observation that just as polytopes give
projective toric varieties, polyhedra give projective toric morphisms.
Polyhedra. Recall that a polyhedron P MR is the intersection of finitely many
closed half-spaces
P = {m MR | hm, ui i ai , i = 1, . . . , s}.
A basic structure theorem tells us that P is a Minkowski sum
P = Q +C,
where Q is a polytope and C is a polyhedral cone (see [281, Thm. 1.2]). If P is
presented as above, then the cone part of P is
(7.1.1)
C = {m MR | hm, ui i 0, i = 1, . . . , s}.
(Exercise 7.1.1). Following [281], we call C the recession cone of P.

Similar to polytopes, polyhedra have supporting hyperplanes, faces, facets,
vertices, edges, etc. One difference is that some polyhedra have no vertices.
Lemma 7.1.1. Let P MR be a polyhedron with recession cone C.
(a) The set V = {v P | v is a vertex} is finite and is nonempty if and only if C is

strongly convex.
(b) If C is strongly convex, then P = Conv(V ) +C.
319
Proof. You will prove this in Exercises 7.1.27.1.5.

Example 7.1.2. The polyhedron P = {(a1 , . . . , an ) Rn | ai 0,
vertices e1 , . . . , en and recession cone C = Cone(e1 , . . . , en ).
Pn
i=1 ai
1} has
Lattice Polyhedra. We now generalize the notion of lattice polytope.

Definition 7.1.3. A polyhedron P MR is a lattice polyhedron with respect to the
lattice M MR if
(a) The recession cone of P is a strongly convex rational polyhedral cone.
(b) The vertices of P lie in the lattice M.
A full dimensional lattice polyhedron has a unique facet presentation
(7.1.2)
P = {m MR | hm, uF i aF for all facets F},
where uF N is a primitive inward pointing facet normal. This was defined in

Chapter 2 for full dimensional lattice polytopes but applies equally well to full
dimensional lattice polyhedra. Then the cone of P is the cone C(P) MR R by
C(P) = {(m, ) MR R | hm, uF i aF for all F, 0}.
When P is a polytope, this reduces to the cone C(P) = Cone(P {1}) considered
in 2.2.
Example 7.1.4. The blowup of C2 at the origin is given by the fan in R2 with
minimal generators u0 = e1 + e2 , u1 = e1 , u2 = e2 and maximal cones Cone(u0 , u1 ),
Cone(u0 , u2 ). For the divisor D = D0 + D1 + D2 , we computed in Figure 5 from
Example 4.3.4 that the polyhedron PD is a 2-dimensional lattice polyhedron whose
recession cone C is the first quadrant.
Figure 1 on the next page shows the 3-dimensional cone C(PD ) with PD at
height 1. Notice how the cone C of PD appears naturally at height 0 in Figure 1.
Some of the properties suggested by Figure 1 hold in general.
Lemma 7.1.5. Let P be a full dimensional lattice polyhedron in MR with recession
cone C. Then C(P) is a strongly convex cone in MR R and
C(P) (MR {0}) = C.
Proof. The final assertion of the lemma follows from (7.1.1) and the definition of
C(P). For strong convexity, note that C(P) MR R0 implies
C(P) (C(P)) MR {0}.
Then we are done since C is strongly convex.
320
PD
y
C
x
Figure 1. The cone C(PD ) in Example 7.1.4
We say that a point (m, ) C(P) has height . Furthermore, when > 0, the
slice of C(P) at height is P. If we write P = Q + C, where Q is a polytope,
then for > 0,
P = Q +C
since C is a cone. It follows that as 0, the polytope shrinks to a point so that at
height 0, only the cone C remains, as in Lemma 7.1.5. You can see how this works
in Figure 1.
The Toric Variety of a Polyhedron. In Chapter 2, we constructed the normal fan
of a full dimensional lattice polytope. We now do the same for a full dimensional
lattice polyhedron P. Given a vertex v P, we get the cone
Cv = Cone(P M v) MR .
Note that v M since P is a lattice polyhedron. It follows easily that Cv is a strongly
convex rational polyhedral cone of maximal dimension, so that the same is true for
its dual
v = Cv = Cone(P M v) NR .
These cones fit together nicely.
Theorem 7.1.6. Given a full dimensional lattice polyhedron P MR with recession cone C, the set
P = { | v , v is a vertex of P}
is a fan in NR . Furthermore:
321
(a) The support of P is |P | = C .
(b) P has full dimensional convex support in NR .
Proof. The proof that we get a fan is similar to the proof for the polytope case (see
2.3) and hence is omitted. To prove part (a), we need to show
[
v = C ,
vV
where V is the set of vertices of P. Now take v V and m C M. Then m =

(v + m) v P M v, which easily implies C Cone(P M v). Taking duals,
we obtain v C . For the opposite inclusion, take u C and pick v V such
that hv, ui hw,P
ui for all w V . We show u P
v as follows. Any m P M can
be written m = wV w w + m where w 0, wV w = 1 and m C. Then

P
P
hm, ui = wV w hw, ui + hm , ui wV w hv, ui = hv, ui.
Thus hm v, ui 0 for all m v P M v, which proves u v .
Part (b) now follows since C is clearly convex, and has full dimension since
C is strongly convex.

The fan P of Theorem 7.1.6 is the normal fan of P. We define XP to be the
toric variety XP of P . Here is an example.
P
Example 7.1.7. The polyhedron P = {(a1 , . . . , an ) Rn | ai 0, P ni=1 ai 1} of
Example 7.1.2 has vertices e1 , . . . , en . The facet of P defined by ni=1 ai = 1 has
e1 + + en as inward normal. Then the vertex ei gives the cone
ei = Cone(e1 + + en , e1 , . . . , ebi , . . . , en ).
These cones form the fan of the blowup of Cn at the origin, so XP = Bl0 (Cn ).
Note that XP is not complete in this example. In general, the normal fan has
support |P | = C . We measure the deviation from completeness as follows.
The support |P | is a rational polyhedral cone but need not be strongly convex.
Recall that W = |P | (|P |) is the largest subspace contained in |P |. Hence
|P | gives the following:
The sublattice W N N and the quotient lattice NP = N/(W N).
The strongly convex cone P = |P |/W NR /W = (NP )R .
The affine toric variety UP of P .
The projection map : N NP is compatible with the fans of XP and UP since

R (|P |) = P . Hence we get a toric morphism
1
: XP UP .
Since |P | = R (P ) (Exercise 7.1.6), Theorem 3.4.11 implies that is proper.
322
The key result of this section is that : XP UP is a projective morphism. We

first give an elementary proof in Theorem 7.1.10. We will also give a more sophisticated proof that applies Proj construction from 7.0 to the semigroup algebra
(7.1.3)
SP = C[C(P) (M Z)].
The character associated to (m, k) C(P) (M Z) is written mt k , and SP is

graded by height, i.e, deg( mt k ) = k. In Proposition 7.1.13 we will prove that
XP Proj(SP ).
Then standard properties of Proj will imply that : XP UP is projective.

The Divisor of a Polyhedron. Let P be a full dimensional lattice polyhedron in
MR . As in the polytope case, facets of P correspond to rays in the normal fan P ,
so that each facet F gives a prime torus-invariant divisor DF XP . Thus the facet
presentation (7.1.2) of P gives the divisor
P
(7.1.4)
DP = F aF DF ,
where the sum is over all facets of P. Results from Chapter 4 (Proposition 4.2.10
and Example 4.3.7) easily adapt to the polyhedral case to show that DP is Cartier
(with mv = v for every vertex v) and the polyhedron of DP is P, i.e., P = PDP .
Then Proposition 4.3.3 implies that
L
(7.1.5)
(XP , OXP (DP )) = mPM C m .
The definition of projective morphism given in 7.0 involves a line bundle L

and a finite-dimensional subspace W of global sections. The line bundle will be
OXP (kDP ) for a suitably chosen integer k 1 and W will be determined by certain
carefully chosen lattice points of kP. The reason we need a multiple is that P might
not have enough lattice points.
Normal and Very Ample Polyhedra. In Chapter 2, we defined normal and very
ample polytopes, which are different ways of saying that there are enough lattice
points. For a lattice polyhedron P, the definitions are the same.
Definition 7.1.8. Let P MR be a lattice polyhedron. Then:
(a) P is normal if for all integers k 1, every lattice point of k P is a sum of k

lattice points of P.
(b) P is very ample if for every vertex v P, the semigroup N(PM v) generated
by P M v is saturated in M.
We have the following result about normal and very ample polyhedra.
Proposition 7.1.9. Let P MR be a lattice polyhedron. Then:

(a) If P is normal, then P is very ample.
(b) If dim P = n 2, then kP is normal and hence very ample for all k n 1.
323
Proof. Part (a) follows from the proof of Proposition 2.2.18. For part (b), write
P = Q + C, where Q is a lattice polytope and C is the recession cone of P. Let
C = Cone(m1 , . . . , ms ), mi M. In Exercise 7.1.7 you will show that
[
(7.1.6)
C=
Conv(0, sm1 , . . . , sms ) + m.
mCM
Since P = Q + C is a full dimensional polyhedron, Q + Conv(0, sm1 , . . . , sms ) is

a full dimensional lattice polytope. It follows that P = Q + C is a union of full
dimensional lattice polytopes. Then part (b) follows by applying Theorem 2.2.12
to each of these polytopes.

The Projective Morphism. Let P be a full dimensional lattice polyhedron in MR .
Assume P is very ample and pick a finite set A P M such that:
A contains the vertices of P.
For every vertex v P, A v generates Cone(P M v) M = v M.
We can always satsify the first condition, and the second is possible since P is very
ample. Using (7.1.5), we get the subspace
W = Span( m | m A ) (XP , OXP (DP )).
We claim that W has no basepoints since A contains the vertices of P. To prove
this, let v be a vertex. Recall that DP + div( v ) is the divisor of zeros of the global
section given by v . One computes that
P
DP + div( v ) = F (aF + hv, uF i)DF .
Since hv, uF i = aF for all facets containing v and hv, uF i > aF for all other
facets, the support of DP + div( v ) is the complement of the affine open subset
Uv XP , i.e., the nonvanishing set of the section is precisely Uv . Then we are
done since the Uv cover XP .
It follows that we get a morphism
L ,W : XP P(W )
for L = OXP (DP ). Here is our result.
Theorem 7.1.10. Let P be a full dimensional lattice polyhedron. Then:
(a) The toric variety XP is quasiprojective.
(b) : XP UP is a projective morphism.
Proof. First suppose that P is very ample. The proof of part (a) is similar to
the proof of Proposition 6.1.10. Let L , W and A be as above and write A =
{m1 , . . . , ms }. Consider the projective toric variety
XA P s1 = P(W ).
324
Let I {1, . . . , s} be the set of indices corresponding to vertices of P. So i I gives

a vertex mi and a corresponding cone i = mi in P . Also let Ui P s1 be the
affine open subset where the ith coordinate is nonzero. By our choice of A , the
proof of Proposition 6.1.10 shows that L ,W induces an isomorphism
Ui XA Ui.
Since XP is the union of the Ui for i I, it follows that

S
(7.1.7)
L ,W : XP XA iI Ui .
Since XA is projective, this shows that XP is quasiprojective. Part (b) now follows
immediately from Proposition 7.0.6 since : XP UP is proper.
When P is not very ample, we know that a positive multiple kP is. Since P
and kP have the same normal fan and same recession cone, the maps XP UP
and XkP UkP are identical. Hence the general case follows immediately from the
very ample case.

Here are two examples to illustrate Theorem 7.1.10.
Example 7.1.11. The polyhedron P from Example 7.1.7 is very ample (in fact, it
is normal), and the set A used in the proof of Theorem 7.1.10 can be chosen to
be A = {e1 , . . . , en , 2e1 , . . . , 2en } (Exercise 7.1.8). This gives XA P 2n1 , where
P 2n1 has variables x1 , . . . , xn , w1 , . . . , wn corresponding to the elements e1 , . . . , en ,
2e1 , . . . , 2en of A . Then XA P 2n1 is defined by the equations xi2 w j = x 2j wi
for 1 i < j n (Exercise 7.1.8). Since XP = Bl0 (Cn ) by Example 7.1.7, the
isomorphism (7.1.7) implies
Bl0 (Cn ) {(x1 , . . . , xn , w1 , . . . , wn ) P 2n1 | (x1 , . . . , xn ) 6= (0, . . . , 0)
and xi2 w j = x 2j wi for 1 i < j n}.
We get a better description of Bl0 (Cn ) using the vertices B = {e1 , . . . , en } of P.

This gives a map XP P n1 which, when combined with XP UP = Cn , gives
a morphism
: XP P n1 Cn .
Let P n1 and Cn have variables x1 , . . . , xn and y1 , . . . , yn respectively. Then is an
embedding onto the subvariety of P n1 Cn defined by xi y j = x j yi for 1 i < j n
(Exercise 7.1.8). Hence
Bl0 (Cn ) {(x1 , . . . , xn , y1 , . . . , yn ) P n1 Cn | xi y j = x j yi , 1 i < j n}.
This description of the blowup Bl0 (Cn ) can be found in many books on algebraic
geometry and appeared earlier in this book as Exercise 3.0.8. Note also that the
projective morphism of Theorem 7.1.10 is the blowdown map Bl0 (Cn ) Cn .
Example 7.1.12. Consider the full dimensional lattice polyhedron P R2 defined
by the inequalities
x 2, 0 y 2, y x + 1.
325
This polyhedron has vertices v1 = (1, 0), v2 = (2, 1), v3 = (2, 2) shown in Figure 2.
The left side of the figure also shows the recession cone C and the decompostion
P = Q +C, where Q is the convex hull of the vertices.
v3
1
2
v2
3
P
P
v1
(0,0)
P=Q+C
v3
v2
C
Q
(0,0)
v1
Figure 2. The polyhedron P = Q +C, the normal fan P , and the cone P
The normal vectors at each vertex vi are reassembled on the right to give the
maximal cones i of the normal fan P . Note also that |P | is not strictly convex,
so we mod out by its maximal subspace to get the strictly convex cone P . The
projection map on the right of Figure 2 gives the projective morphism XP UP ,
where UP C is the toric variety of P .
The Proj Construction. We conclude this section by explaining how construct XP

using Proj. Let P MR be a full dimensional lattice polyhedron. By (7.1.3), the
cone C(P) MR R gives the semigroup algebra
SP = C[C(P) (M Z)].
We use the height grading defined by deg( mt k ) = k. Then we can relate Proj(SP )
to XP as follows.
Theorem 7.1.13. There is a natural isomorphism XP Proj(SP ).
Proof. Let V be the set of vertices of P. In Exercise 7.1.9 you will prove:
326
h v t | v V i = (SP )+ =
If v V , then (SP )( v t) =
d>0 (SP )d .
C[v M], where
v = Cone(P M v) .
By the first bullet, Proj(SP ) is covered by the affine open subsets Spec((SP )( v t) ),
and the second shows that Spec((SP )( v t) ) is the affine toric variety of the cone v .
These patch together in the correct way to give XP Proj(SP ).

We can also interpret the morphism : XP UP in terms of Proj. The idea is to
compute (SP )0 , the degree 0 part of the graded ring SP = C[C(P) (M Z)]. The
slice of C(P) at height 0 is the recession cone C of P. Recall that NP = N/(W N),
where W C is the largest subspace contained in C and that UP is the affine
toric variety of P , which is the image of C in (NP )R . Then the inclusion MP M
dual to N NP gives
P = C (MP )R MR .
It follows that (SP )0 = C[C M] = C[P M]. This implies Spec((SP )0 ) = UP , so
that Proj(SP ) Spec((SP )0 ) becomes : XP UP . It follows that is projective
by Proposition 7.0.9. This gives a second proof of Theorem 7.1.10.
It is also possible to prove directly that Proj(SP ) Spec((SP )0 ) is projective
without using Proposition 7.0.9. See Exercise 7.1.10.
Exercises for 7.1.
7.1.1. Prove (7.1.1). Hint: Fix m0 P and take any m C. Show that m0 + m P for
> 0, so hm0 + m, ui i ai . Then divide by and let .
7.1.2. Let P = Q + C be a polyhedron in MR where Q is a polytope and C is a polyhedral

cone. Define P (u) = minmP hm, ui for u C .
(a) Show that P (u) = minmQ hm, ui for u C and conclude that P : C R is welldefined.
(b) Show that P (u) = minvVQ hv, ui for u C , where VQ is the set of vertices of Q.
(c) Show that P = {m MR | P (u) hm, ui for all u C }. Hint: For the non-obvious
direction, represent P as the intersection of closed half-spaces coming from supporting
hyperplanes.
7.1.3. Let P be a polyhedron in MR with recession cone C and let W = C (C) be the
largest subspace contained in C. Prove that every face of P contains a translate of W and
conclude that P has no vertices when C is not strongly convex.
7.1.4. Let P = Q + C be a polyhedron in MR where Q is a polytope and C is a strongly
convex polyhedral cone. Let VQ be the set of vertices of Q. Assume that there is v VQ
and u in the interior of C such that hv, ui < hw, ui for all w 6= v in VQ . Prove that v is a
vertex of P. Hint: Show that Hu,a , a = hv, ui, is a supporting hyperplane of P such that
Hu,a P = v. Also show if v and u satisfy the hypothesis of the problem, then so do v and
u for any u sufficiently close to u. Finally, Exercise 7.1.2 will be useful.
7.1.5. Let P = Q + C be a polyhedron in MR where Q is a polytope and C be a strongly
convex polyhedral cone. Let VQ be the set of vertices of Q and let
U0 = {u Int(C ) | hv, ui =
6 hw, ui whenever v 6= w in VQ }.
327
(a) Show that U0 is open and dense in C . Hint: dim C = dim NR .

(b) Use Exercise 7.1.4 to show that for every u U0 , there is a vertex v of P such that
P (u) = hv, ui. Conclude that the set VP of vertices of P is nonempty and finite.
(c) Show that P (u) = minvVP hv, ui for u C .
(d) Conclude that P = Conv(VP ) +C. Hint: The first step is to show that P = P , where
P = Conv(VP ) + C. Then use part (c) of Exercise 7.1.2.
7.1.6. Let C NR be a polyhedral cone. Set W = C (C) and let = (C) NR /W for
the quotient map : NR NR /W . Show that is strongly convex and C = 1 ().
P
P
7.1.7. Prove (7.1.6). Hint: Given si=1 i mi C, let m = si=1 i mi C M.
7.1.9. Supply proofs of the two bullets from the proof of Theorem 7.1.13.
7.1.10. Here you will give an elementary proof that Proj(SP ) Spec((SP )0 ) is projective,
where SP is the graded semigroup algebra from (7.1.3).
(a) Explain why we can assume that P is normal.
(b) Show that C(P)(M Z) is generated by its elements of height 1 when P is normal.
(c) Assume P is normal and let H be a Hilbert basis of C(P) (M Z). Then H =
H0 H1 , where elements of Hi have height i and write H1 = {(m1 , 1), . . . , (ms , 1)}.
Prove that SP is generated as an (SP )0 -algebra by m1 t, . . . , ms t and conclude that
there is a surjective homomorphism of graded rings
(SP )0 [x1 , . . . , xs ] SP ,
xi 7 mi t.
(d) Prove that when P is normal, there is a commutative diagram
/ UP P s1
XP OO
OOO
OOO
p1
OOO
OOO
'
UP
such that is a closed embedding and is a projective morphism.

7.1.11. In this exercise, you will prove a stronger version of part (b) of Theorem 7.1.10.
Let XA and W be as in the proof of the theorem. Prove that there is a commutative diagram
(7.1.8)
L ,W
/ UP P(W )
XP PP
PPP
PPP
PPP
p1
PPP
P'
UP
such that L ,W : XP UP P(W ) is a closed embedding. Hint: Proposition 7.0.6.

7.1.12. Let NR be a strongly convex rational polyhedral cone. This gives the semigroup algebra C[S ] = C[ M]. Given a monomial ideal a = h m1 , . . . , ms i C[S ],
we get the polyhedron
P = Conv(m M | m a),
Prove that P = Conv(m1 , . . . , ms ) + .
328
7.2. Projective Morphisms and Toric Varieties

We begin our study of projective toric morphisms with a toric variety X whose fan
has full dimensional convex support. We construct an affine toric variety from
|| as follows. The largest subspace contained in || is W = || (||). Similar
to 7.1, we have:
The sublattice W N N and the quotient lattice N = N/(W N).
The strongly convex cone = ||/W NR /W = (N )R .
The affine toric variety U = U .
The projection map : N N is compatible with the fans of X and U since

R (||) = . This gives a toric morphism
: X U .
(7.2.1)
which as in 7.1 is easily seen to be proper. The difference between here and
7.1 is that : X U may fail to be projective. Our first goal is to understand
when this happens. As you might suspect, the answer involves polyhedra, support
functions, and convexity.
The Polyhedron of a Divisor. A Weil divisor D =
hedron
a D
on X gives the poly-
PD = {m MR | hm, u i a for all }.

When is complete, this is a polytope, but in general we have
PD = Q +C,
where Q is a polytope and C is the recession cone of PD .
Lemma 7.2.1. Assume || is convex of full dimension and let D =
Weil divisor on X . If PD 6= , then:
(a) The recession cone of PD is || .
a D
be a
(b) The set V = {v PD | v is a vertex} is nonempty and finite.

(c) PD = Conv(V ) + || .
Proof. Combining (7.1.1) with the definition of PD , we see that the recession cone
of PD is
{m MR | hm, u i 0 for all } = ||
since || = Cone(u | (1)) by (6.1.3). This proves part (a). The recession
cone is strongly convex since || has full dimension, so that parts (b) and (c) follow
from Lemma 7.1.1.
329
Divisors and Convexity. Now that we know about recession cones, the convexity
result proved in Theorem 6.1.7 can be improved as follows.
Theorem 7.2.2. Assume || is convex of full dimension n and let D be the support
function of a Cartier divisor D on X . Then the following are equivalent:
(a) D is basepoint free.
(d) D : || R is convex.
(e) PD = Conv(m | (n)) + || .
(f) {m | (n)} is the set of vertices of PD .
(g) D (u) = minmPD hm, ui for all u ||.
In particular, PD is a lattice polyhedron when D is basepoint free.

Proof. Parts (a)(d) are equivalent by Theorem 6.1.7. Furthermore, (b) (f) and
(b) (g) follow as in the proof of Theorem 6.1.7, and (f) (e) follows from
Lemma 7.2.1. Also, (e) (b) is obvious. Finally, (g) (d) follows from part (b)
of Exercise 7.1.2.

Strict Convexity. Our next task is to show that : X U is projective if and only
if X has a Cartier divisor D with strictly convex support function. We continue
to assume that has full dimensional convex support. As in 6.1, D is strictly
convex if it is convex and for each (n), the equation D (u) = hm , ui holds
only on . The strict convexity criteria from Lemma 6.1.13 apply to this situation.
When D is strictly convex, the polyhedron PD has an especially nice relation
to the fan .
P
Proposition 7.2.3. Assume that || is convex of full dimension and D = a D
has a strictly convex support function. Then:
(a) PD is a full dimensional lattice polyhedron.
(b) is the normal fan of PD .

Proof. Theorem 7.2.2 and Lemma 6.1.13 imply that the m , (n), are distinct
and give the vertices of the polyhedron. As in 7.1, a vertex m PD gives the
cone Cm = Cone(PD M m ). We claim that
= Cm .
This easily implies that PD has full dimension and that is the normal fan of PD .
Fix (n) and let (1). Then m PD M implies
(7.2.2)
hm, u i D (u ) = hm , u i,
330
where the inequality holds by Lemma 6.1.6 and the equality holds since u .
Thus hm m , u i 0 for all m PD M, so that u Cm for all (1). Hence
Cm .
Since || is the recession cone of PD , the proof of Theorem 7.1.6 implies

S
Cm || = (n) .
Now take u Int(Cm ). Hence u for some (n). Then u Cm and

m m Cm imply
hm m , ui 0, so hm , ui hm , ui.
On the other hand, if we apply (7.2.2) to the cone and m = m , we obtain

hm , u i hm , u i. We conclude that
hm , ui = hm , ui,
and the same equality holds for all elements of Int(Cm ) . This easily implies

that m = m . Then = by strict convexity, so that u .
Here is the first major result of this section.
Theorem 7.2.4. Let : X U be the proper toric morphism where U is affine.
Then || is convex. Furthermore, the following are equivalent:
(a) X is quasiprojective.
(b) is a projective morphism.

(c) X has a torus-invariant Cartier divisor with strictly convex support function.
1
Proof. Since is proper, Theorem 3.4.11 implies that || = R (). Thus || is

convex. To prove (a) (b) (c), first assume that || has full dimension.
If (c) is true, then is the normal fan of the full dimensional lattice polyhedron
PD by Proposition 7.2.3. It follows that X = XPD , which is quasiprojective by
Theorem 7.1.10, proving (a). Furthermore, (a) (b) by Proposition 7.0.6.
If (b) is true, we will use the theory of ampleness developed in [127]. The
essential facts we need are summarized in the appendix to this chapter. Since
is projective, there is a line bundle L on X that satisfies Definition 7.0.3. Then,
since U is affine, Theorem 7.A.4 and Proposition 7.A.6 imply that
L k = L OX OX L (k times) is generated by global sections for some
integer k > 0.
The nonvanishing set of a global section of L is an affine open subset of X .
By 7.0, L OX (D) for some Cartier divisor on X , and since linearly equivalent Cartier divisors give isomorphic line bundles, we may assume that D is torusinvariant (this follows from Theorem 4.2.1). Then OX (kD) is generated by global
331
sections for some k > 0. This implies that kD = kD is convex by Theorem 7.2.2,
so that D is convex as well. We show that D is strictly convex by contradiction.
If strict convexity fails, then Lemma 6.1.13 implies that there is a wall =
in with m = m . Then m = m = m corresponds to a global section
s, which by the proof of Proposition 6.1.1 is nonvanishing on U (since m = m )
and on U (since m = m ). Thus the nonvanishing set contains U U , which
contains the complete curve V ( ) U U . But being affine, the nonvanishing
set cannot contain a complete curve (Exercise 7.2.1). This completes the proof of
the theorem when || has full dimension.
Finally, suppose that || fails to have full dimension. Let N1 = Span(||) N

and pick N0 N such that N = N0 N1 . The cones of lie in (N1 )R and hence
give a fan 1 in (N1 )R . If N0 has rank r, then Proposition 3.3.11 implies that
(7.2.3)
X (C )r X1 .
It follows that D : || = |1 | R is the support function of a Cartier divisor D1

on X1 . Note also that |1 | is convex of full dimension in (N1 )R . Since (C )r is
quasiprojective, this allows us to reduce to the case of full dimensional support.
You will supply the omitted details in Exercise 7.2.2.

f -Ample and f -Very Ample Divisors. The definitions of ample and very ample
from 6.1 generalize to the relative setting as follows. Recall from Definition 7.0.3
that a morphism f : X Y is projective with respect to the line bundle L when for
a suitable open cover {Ui } of Y , we can find global sections s0 , . . . , ski of L over
f 1 (Ui ) that give a closed embedding
f 1 (Ui ) Ui P ki .
Then we have the following definition.
Definition 7.2.5. Let D be a Cartier divisor on X and f : X Y be proper.
(a) The divisor D and the line bundle OX (D) are f -very ample if f is projective
with respect to the line bundle L = OX (D).
(b) D and OX (D) are f -ample when kD is f -very ample for some integer k > 0.
Hence f : X Y is projective if and only if X has an f -ample line bundle. In
the toric case, Proposition 7.1.9 and Theorem 7.2.4 give the following result.
Theorem 7.2.6.
P Let : X U be a proper toric morphism where U is affine,
and let D = a D be a Cartier divisor on X . Also let n = dim X . Then:
(a) D is -ample if and only if D is strictly convex.
(b) If n 2 and D is -ample, then kD is -very ample for all k n.
332
Here is are two examples of Theorem 7.2.6.

Example 7.2.7. Consider the blowdown morphism : Bl0 (Cn ) Cn . The fan for
Bl0 (Cn ) has minimal generators u0 = e1 + + en and ui = ei for 1 i n. Let
D0 be the divisor corresponding to u0 . The support function D0 of D0 is easily
seen to be strictly convex (Exercise 7.2.4). Thus D0 is -ample by Theorem 7.2.6.
Note also that the polyhedron PD0 is the polyhedron P from Example 7.1.7.
Example 7.2.8. The P be a full dimensional lattice polyhedron in MR . The map

: XP UP is projective by Theorem 7.1.10. We also have the Cartier divisor DP
on XP defined in (7.1.4). As noted in the discussion following (7.1.4), P = PDP
and the vertices of P give the Cartier data of DP , so that DP is strictly convex by
Theorem 7.2.2 and Lemma 6.1.13. Hence DP is -ample by Theorem 7.2.6.
Semiprojective Toric Varieties. Following [137], we say that X is semiprojective

if the natural map : X Spec((X , OX )) is projective and X has a torus
fixed point. We can characterize semiprojective toric varieties as follows.
Proposition 7.2.9. Given a toric variety X , the following are equivalent:
(a) X is semiprojective.
(b) X is quasiprojective and has full dimensional convex support in NR .
(c) X = XP is the toric variety of a full dimensional lattice polyhedron P MR .
Proof. By the Orbit-Cone Correspondence (Theorem 3.2.6), X has a torus fixed
point if and only if has a full dimensional cone, which is equivalent to having
full dimensional support. In Exercise 7.2.3 you will show that Spec((X , OX ))
is a normal affine toric variety U . Then (a) (b) follows from Theorem 7.2.4.
The equivalence (b) (c) follows from Proposition 7.2.3 and Theorem 7.2.4.
This completes the proof.
A semiprojective toric X variety comes equipped with a projective morphism

: X Spec((X , OX )), and a full dimensional lattice polyhedron P comes
with a projective morphism : XP UP by Theorem 7.1.10. These maps are the
same by Exercise 7.2.3.
We can also extend the relation between polytopes and ample divisors on complete toric varieties described in 6.2. Consider the set of polyhedra

P MR | P is a full dimensional lattice polytope
and the set of pairs

(X , D) | is a fan in NR , X is semiprojective, and

D is a torus-invariant -ample divisor on X .
These sets are related as follows.
333
Theorem 7.2.10. The maps P 7 (XP , DP ) and (X , D) 7 PD define bijections between the above sets that are inverses of each other.
Proof. First note that XP is semiprojective by Proposition 7.2.9, and DP is -ample
by Example 7.2.8. Going the other way, suppose that X is semiprojective and D
is -ample. Then Theorem 7.2.6 and Proposition 7.2.9 imply that DP has a strictly
convex support function, so that PD is a full dimensional lattice polyhedron by
Proposition 7.2.3.
Using Proposition 7.2.3 and P = PDP , it is easy to see that the two maps are
inverses of each other.

Projective Toric Morphisms. Suppose we have fans in NR and in NR . Recall
from 3.3 that a toric morphism
: X X
is induced from a map of lattices
: N N
compatible with and , i.e., for each there is with R () .
We first determine when a torus-invariant Cartier divisor on X is -ample.

Since projective morphisms are proper, we can assume that is proper, which by
Theorem 3.4.11 is equivalent to
(7.2.4)
|| = R (| |).
Here is our result.

Theorem
7.2.11. Let : X X be a proper toric morphism and let D =
P
a D be a Cartier divisor on X . Also let n = dim X . Then:
1
(a) D is -ample if and only if for every , D is strictly convex on R ( ).
(b) If n 2 and D is -ample, then kD is -very ample for all k n 1.
Proof. The idea is to study what happens over the affine open subsets U X
for . Observe that 1 (U ) is the toric variety corresponding to the fan
= { | R () }.
Thus 1 (U ) = X . Let = |1 (U
X
O
?
1 (U )
X
and consider the diagram

/ X
O
?
/U
/ U .
334
Also let D be the restriction of D to 1 (U ) = X .

By Proposition 7.A.5, D is -ample if and only if the restriction D|1 (U ) is
|1 (U ) -ample for all . Using the above notation, this becomes
D is -ample D is -ample for all .

However, Theorem 7.2.6 implies that
D is -ample D is strictly convex.
This completes the proof of the theorem.
It is now easy to characterize when a toric morphism is projective.

Theorem 7.2.12. Let : X X be a toric morphism. Then the following are
equivalent:
(a) is projective.
(b) is proper and X has a torus-invariant Cartier divisor D whose support
1
function D is strictly convex on R ( ) for all .

You will prove Theorem 7.2.12 in Exercise 7.2.5. The first proof of this theorem was given in [172, Thm. 13 of Ch. I]. In Chapter 11 we will use this result to
construct interesting examples of projective toric morphisms.
Exercises for 7.2.
7.2.1. Prove that an affine variety cannot contain a complete variety of positive dimension.
Hint: If X is complete and irreducible, then (X, OX ) = C.
7.2.2. This exercise will complete the proof of Theorem 7.2.4. Let : X U satisfy
the hypothesis of the theorem and write X as in (7.2.3). We also have the Cartier divisors
D on X and D1 on X1 as in the proof of the theorem.
(a) Assume that is projective. Prove that X is quasiprojective and conclude that X1 is
quasiprojective. Now use the first part of the proof to show that D is strictly convex.
Hint: See Exercise 7.0.1.
(b) Assume that D is strictly convex. Prove that X1 is quasiprojective and conclude that
X is quasiprojective. Then use Proposition 7.0.6.
7.2.3. Given a toric variety X , let C = {m MR | hm, u i 0 for all (1)}, and let
be the strongly convex cone obtained by taking the quotient of C by its minimal face.
Prove that U Spec((X , OX )).
7.2.4. Prove that the support function D0 in Example 7.2.7 is strictly convex. We will
generalize this result considerably in Chapter 11.
7.3. Projective Bundles and Toric Varieties
335

Given a vector bundle or projective bundle over a toric variety, the nicest case is
when the bundle is also a toric variety. This will lead to some lovely examples of
toric varieties.
P
Toric Vector Bundles and Cartier Divisors. A Cartier divisor D = a D on a
toric variety X gives the line bundle L = OX (D), which is the sheaf of sections
of the rank 1 vector bundle : VL X .
We will show that VL is a toric variety and is a toric morphism by constructing the fan of VL in terms of and D. To motivate our construction, recall that for
m M, we have
m (X , OX (D)) m PD
hm, ui D (u) for all u ||
the graph of u 7 hm, ui lies

above the graph of D .
The first equivalence follows from Proposition 4.3.3 and the second from Propostion 6.1.6. The key word is above: it tells us to focus on the part of NR R that
lies above the graph of D .
We define the fan D in NR R as follows. Given , set
e = {(u, ) | u , D (u)}
= Cone((0, 1), (u , a ) | (1)),
where the second equality follows since D (u ) = a and D is linear on . Note

that
e is a strongly convex rational polyhedral cone in NR R. Then let D be
the set consisting of the cones
e for and their faces. This is a fan in NR R,
and the projection : N Z N is clearly compatible with D and . Hence
we get a toric morphism
: XD X .
Proposition 7.3.1. : XD X is a rank 1 vector bundle whose sheaf of sections is OX (D).
Proof. We first show that is a toric fibration as in Theorem 3.3.19. The kernel
of : N Z N is N0 = {0} Z, and the fan 0 = { D | (N0 )R } has
0 = Cone((0, 1)) as its unique maximal cone. Also, for let
b = Cone((u , a ) | (1)).
This is the face of

e consisting of points (u, ) where D (u) = . Thus
b D
b = {b
and in fact
| } is a subfan of D. Since
e=
b + 0 and R
336
maps
b bijectively to , we see that D is split by and 0 in the sense of
Definition 3.3.18. Since X0 ,N0 = C, Theorem 3.3.19 implies that
1 (U ) U C.
To see that this gives the desired vector bundle, we study the transition functions. First note that 1 (U ) = Ue , so that the above isomorphism is
Ue U C,
which by projection induces a map Ue C. It is easy to check that this map is

(m ,1) , where D (u) = hm , ui for u (Exercise 7.3.1). Note that
(m , 1)
e (M Z),
follows directly from the definition of

e. Then, given another cone , the
transition map from U C U C to U C U C is given by (u,t) 7
(u, g (u)t), where g = m m (Exercise 7.3.1).
We are now done, since the proof of Proposition 6.2.7 shows that OX (D) is
the sheaf of sections of a rank 1 vector bundle over X whose transition functions

are g = m m .
This construction is easy but leads to some surprisingly rich examples.
Example 7.3.2. Consider P n with its usual fan and let D0 correspond to the minimal generator u0 = e1 en . Recall that OPn (D0 ) is denoted OPn (1).
This gives the rank 1 vector bundle V P n described in Proposition 7.3.1 whose
fan in Rn R = Rn+1 has minimal generators
e1 , . . . , en+1 , e1 en + en+1 .
You will check this in Exercise 7.3.2.
We can also describe this vector bundle geometrically as follows. Consider the
lattice polyhedron in Rn+1 given by
P = Conv(0, e1 , . . . , en ) + Cone(en+1 , e1 + en+1 , . . . , en + en+1 ).
The normal fan of P is the fan (Exercise 7.3.2), so that XP is the above vector
bundle V . Note also that || is dual to the recession cone of P.
It is easy to see that || is a smooth cone of dimension n + 1, so that the projective toric morphism XP UP constructed in 7.1 becomes XP Cn+1 . When
combined with the vector bundle map XP = V P n , we get a morphism
XP P n Cn+1 .
When the coordinates of P n and Cn+1 are ordered correctly, the image is precisely
the variety defined by xi y j = x j yi (Exercise 7.3.2). In this way, we recover the
description of V P n given in Example 6.0.19.
337
Proposition 7.3.1 extends easily P

to decomposable toric vector bundles. Suppose we have r Cartier divisors Di = ai D , i = 1, . . . , r. This gives the locally
free sheaf
(7.3.1)
OX (D1 ) OX (Dr )
of rank r. To construct the fan of the corresponding vector bundle, we work in

NR Rr . Let e1 , . . . , er be the standard basis of Rr and write elements of NR Rr
as u + 1 e1 + + r er . Then, given , we get the cone
e = {u + 1 e1 + + r er | u , i Di (u) for i = 1, . . . , r}
= Cone(u a1 e1 ar er | (1)) + Cone(e1 , . . . , er ).
One can show without difficulty that the set consisting of the cones
e for
and their faces is a fan in NR Rr such that the toric variety of this fan is the vector
bundle over X whose sheaf of sections is (7.3.1) (Exercise 7.3.3).
Besides decomposable vector bundles, one can also define a toric vector bundle
: V X . Here, rather than assume that V is a toric variety, one makes the weaker
assumption the torus of X acts on V such that the action is linear on the fibers and
is equivariant. Toric vector bundles have been classified by Klyachko [178] and
otherssee [225] for the historical background. Oda noted in [217, p. 41] that if a
toric vector bundle is a toric variety in its own right, then the bundle is a direct sum
of line bundles, as above. This can be proved using Klyachkos results.
Toric Projective Bundles. The decomposable toric vector bundles have associated
toric projective bundles. Cartier divisors D0 , . . . , Dr give the locally free sheaf
E = OX (D0 ) OX (Dr ),
of rank r + 1. Then P(E ) X is a projective bundle whose fibers look like P r .
To describe the fan of P(E ), we first give a new description of the fan of P r . In
Rr+1 , we use the standard basis e0 , . . . , er . The first orthant Cone(e0 , . . . , er ) has
r + 1 facets
Fi = Cone(e0 , . . . , ei , . . . , er ),
i = 0, . . . , r.
Now set N = Zr+1 /Z(e0 + + er ). Then the images ei of ei sum to zero in N and
the images F i of Fi give the fan for P r in N R .
The construction of P(E ) given in 7.0 involves taking the dual vector bundle.
Thus P(E ) = P(VE ), where VE is the vector bundle whose sheaf of sections is
OX (D0 ) OX (Dr ).
The fan of VE is built from cones
Cone(u + a0 e0 + + ar er | (1)) + Cone(e0 , . . . , er )
338
and their faces, as ranges over the cones . To get the fan for P(E ) = P(VE ),
take and let Fi be a facet of Cone(e0 , . . . , er ). This gives the cone
Cone(u + a0 e0 + + ar er | (1)) + Fi NR Rr+1 ,
and one sees that i NR N R is the image of this cone under the projection map
NR Rr+1 NR N R .
Proposition 7.3.3. The cones {i | , i = 0, . . . , r} and their faces form a fan
E in NR N R whose toric variety XE is the projective bundle P(E ).
Proof. Consider the fan 0 in N R given by the F i and their faces. Also, for ,
let
b be the image of Cone(u + a0 e0 + + ar er | (1)) in NR N R . Then
one easily adapts the proof of Proposition 7.3.1 to show that the toric variety XE
of E is a fibration over X with fiber P r . Furthermore, working over an affine
open subset of X , one sees that XE is obtained from VE by the process described
in 7.0. We leave the details as Exercise 7.3.4.

In practice, one usually replaces N = Zr+1 /Z(e0 + + er ) with Zr and the
basis e1 , . . . , er . Then set e0 = e1 er and we redefine Fi as
(7.3.2)
Fi = Cone(e0 , . . . ,ei , . . . , er ) Rr
and for a cone , redefine i as

(7.3.3)
i = Cone(u + (a1 a0 )e1 + + (ar a0 )er | (1)) + Fi
in NR Rr . This way, E is a fan in NR Rr . Here is a classic example.
Example 7.3.4. The fan for P1 has minimal generators u1 and u0 = u1 . Also
let OP1 (1) = OP1 (D0 ), where D0 is the divisor corresponding to u0 . Fix an integer
a 0 and consider
E = OP1 OP1 (a).
As above, we get a fan E in R R = R2 . The minimal

P generators u0 , u1 live in the
first factor. In the second factor, the vectors e0 = ri=1 ei , e1 , . . . , er in the above
construction reduce to e0 = e1 , e1 . Thus F0 = Cone(e1 ) and F1 = Cone(e0 ). We
will use u1 , e1 as the basis of R2 .
The maximal cones for the fan of P1 are = Cone(u1 ) and = Cone(u0 ).
Then E has four cones:
e0 = Cone(u1 + (0 0)e1 ) + F0 = Cone(u1 , e1 )
e1 = Cone(u1 + (0 0)e1 ) + F1 = Cone(u1 , e1 )
e0 = Cone(u0 + (a 0)e1 ) + F0 = Cone(u1 + ae1 , e1 )
e1 = Cone(u0 + (a 0)e1 ) + F1 = Cone(u1 + ae1 , e1 ).
This is the fan for the Hirzebruch surface Ha . Thus
Ha = P(OP1 OP1 (a)).
339
Note also that the toric morphism Ha P1 constructed earlier is the projection
map for the projective bundle.
This example generalizes as follows.

Example 7.3.5. Given integers s, r 1 and 0 a1 ar , consider the projective bundle
P(E ) = P(OPs OPs (a1 ) OPs (ar )).
s Rr , where Rs
The fan E of P(E ) has a nice description. We will work in RP
has basis u1 , . . . , us and Rr has basis e1 , . . . , er . Also set u0 = sj=1 u j and e0 =
P
si=1 ei . As usual, u0 corresponds to the divisor D0 of P s such that OPs (ai ) =
OPs (ai D0 ).
The description (7.3.3) of the cones in uses generators of the form

u + (a1 a0 )e1 + + (ar a0 )er ,
(7.3.4)
where the u are minimal generators of the fan of the base of the projective bundle.
Here, the u s are u0 , . . . , us . Since we are using the divisors 0, a1 D0 , . . . , ar D0 , the
formula (7.3.4) simplifies dramatically, giving minimal generators
u = u0 : v0 = u0 + a1 e1 + + ar er
u = u j : v j = u j , j = 1, . . . , s.
Since the maximal cones of P s are Cone(u0 , . . . , uj , . . . , us ), (7.3.2) and (7.3.3) imply that the maximal cones of are
Cone(v0 , . . . , v j , . . . , vs ) + Cone(e0 , . . . ,ei , . . . , er )
for all j = 0, . . . , s and i = 0, . . . , r. It is also easy to see that the minimal generators
v0 , . . . , vs , e0 , . . . , er have the following properties:
v1 , . . . , vs , e1 , . . . , er form a basis of Zs Zr .
e0 + + er = 0.
v0 + + vs = a1 e1 + + ar er .
The first two bullets are clear, and the third follows from
definition of the v j .
Ps
j=0 u j
= 0 and the
One also sees that XE = P(E ) is smooth of dimension s + r. Since E has

(s + 1) + (r + 1) = s + r + 2 minimal generators, the description of the Picard group
given in 4.2 implies that
Pic(P(E )) Z2 .
(Exercise 7.3.5). Also observe that {v0 , . . . , vs } and {e0 , . . . , er } give primitive collections of E . We will see below that these are the only primitive collections
of E . Furthermore, they are extremal in the sense of 6.4 and their primitive
relations generate the Mori cone of P(E ).
This is a very rich example!
340
A Classification Theorem. Kleinschmidt [177] classified all smooth projective

toric varieties with Picard number 2, i.e., with Pic(X ) Z2 . The rough idea is
that they are the toric projective bundles described in Example 7.3.5. Following
ideas of Batyrev [14], we will use primitive collections to obtain the classification.
We begin with some results from [14] about primitive collections. Recall from
6.4 that a primitive collection P = {1 , . . . , k } (1) gives the primitive relation
P
(7.3.5)
u1 + + uk (1) c u = 0, c Q>0 ,
where is the minimal cone containing u1 + + uk . When X is smooth
and projective, these primitive relations have some nice properties.
Proposition 7.3.6. Let X be a smooth projective toric variety. Then:
(a) In the primitive relation (7.3.5), P (1) = and c Z>0 for all (1).
(b) There is a primitive collection P with primitive relation u1 + + uk = 0.
Proof. The c are integral since is smooth. Let the minimal generators of be
u1 , . . . , u , so the primitive relation becomes
u1 + + uk = c1 u1 + + c u .
To prove part (a), suppose for example that u1 = u1 . Then
u2 + + uk = (c1 1)u1 + c2 u2 + c u .
Note that u2 , . . . , uk generate a cone of since P is a primitive collection. So the
above equation expresses an element of a cone of in terms of minimal generators
in two different ways. Since is smooth, these must coincide. To see what this
means, we consider two cases:
c1 > 1. Then {u2 , . . . , uk } = {u1 , u2 , . . . , u }, so that ui = u1 for some i > 1.
This is impossible since u1 = u1 .
c1 = 1. Then {u2 , . . . , uk } = {u2 , . . . , u }. Since u1 = u1 , we obtain P (1),
which is impossible since P is a primitive collection.
Since c1 must be positive, we conclude that u1 = u1 leads to a contradiction. From
here, it is easy to see that P (1) = .
Turning to part (b), let be the support function of an ample divisor on X .

Thus is strictly convex. Since is complete, we can find an expression
(7.3.6)
b1 u1 + + bs us = 0
such that b1 , . . . , bs are positive integers. Note that u1 , . . . , us cannot lie in a cone
of . By strict convexity and Lemma 6.1.13, it follows that
(7.3.7)
0 = (0) = (b1 u1 + + bs us ) > b1 (u1 ) + + bs (us ).
Pick a relation (7.3.6) so that the right-hand side is as big as possible.
341
The set {u1 , . . . , us } is not contained in a cone of and hence has a subset
that is a primitive collection. By relabeling, we may assume that {u1 , . . . , uk },
k s, is a primitive collection. Using (7.3.6) and the primitive relation (7.3.5), we
obtain the nonnegative relation
X
c u +
s
k
X
X
(bi 1)ui +
bi ui = 0.
i=1
(1)
i=k+1
Since is linear on and strictly convex,

k
k
X

X
X
X
(ui ),
ui >
c u =
c (u ) =
(1)
(1)
i=1
i=1
which implies that

s
k
X
X
X
bi (ui )
(bi 1)(ui ) +
c (u ) +
(1)
>
k
X
i=1
(ui ) +
i=1
k
X
i=1
(bi 1)(ui ) +
i=k+1
s
X
bi (ui ) =
i=k+1
s
X
bi (ui ).
i=1
This contradicts the maximality of the right-hand side of (7.3.7), unless k = s and
b1 = = bk = 1, in which case we get the desired primitive collection.

We now prove Kleinschmidts classification theorem.
Theorem 7.3.7. Let X be a smooth projective toric variety with Pic(X ) Z2 .
Then there are integers s, r 1, s + r = dim X and 0 a1 ar with

X P OPs OPs (a1 ) OPs (ar ) .
Proof. Let n = dim X . Then Pic(X ) Z2 and Theorem 4.2.1 imply that (1)
has n + 2 elements. We recall two facts about divisors D on X :
P
If D is nef and (n), then D a D where a = 0 for (1) and
a 0 for
/ (1).
If D 0 and D 0, then D = 0 since X is complete.
The first bullet was proved in (6.4.10), and the second is an easy consequence of
Propositions 4.0.16 and 4.3.8.
By assumption, X has an ample divisor D which lies in the interior of the
nef cone Nef(X ). Changing D if necessary, we can assume that D is effective
and [D] Pic(X )R is not a scalar multiple of any [D ] for (1). The line
determined by [D] divides Pic(X )R R2 into closed half-planes H + and H .
Then define
P = { (1) | [D ] H + }
Q = { (1) | [D ] H }.
342
Note that P Q = (1), and P Q = by our choice of D. We claim that

(7.3.8)
(n) = {, | P, Q}, where

, = Cone(u | (1) \ {, }).
To prove this, first take (n). Since |(1)| = n and |(1)| = n + 2, we have
(7.3.9)
(1) = (1) {, }.
Applying the first bullet above to D and , we get [D] = a[D ] + b[D ] where
a, b > 0 since [D] is a multiple of neither [D ] nor [D ]. It follows that [D ] and
[D ] lie on opposite sides of the line determined by [D]. We can relabel so that
P and Q, and then has the desired form by (7.3.9).
For the converse, take P and Q. Since Pic(X )R R2 , we can find a
linear dependence
a0 [D ] + b0 [D ] + c0 [D] = 0,
a0 , b0 , c0 Z not all 0.
We can assume that a0 , b0 0 since [D ] and [D ] lie on opposite sides of the

line determined by [D]. Note also that c0 < 0 by the second bullet above, and then
a0 , b0 > 0 by our choice of D. It follows that D = a0 D + b0 D is ample. In
Exercise 7.3.6 you will show that
X \ Supp(D ) = X \ (D D )
is the nonvanishing set of a global section of OX (D ) and hence is affine. This set
is also torus-invariant and hence is an affine toric variety. Thus it must be U for
some . In other words,
X = U D D .
Since U (D D ) = , the Orbit-Cone correspondence (Theorem 3.2.6) implies that satsisfies (7.3.9) and hence gives an element of (n). This completes
the proof of (7.3.8).
An immediate consequence of this description of (n) is that P and Q are
primitive collections. Be sure you understand why. It is also true that P and Q
are the only primitive collections of . To prove this, suppose that we had a third
primitive collection R. Then P 6 R, so there is P \ R, and similarly there is
Q \ R since Q 6 P. By (7.3.8), the rays of R all lie in , (n), which
contradicts the definition of primitive collection.
Since X is projective and smooth, Proposition 7.3.6 guarantees that has a
primitive collection whose elements sum to zero. We may assume that P is this
primitive collection. Let |P| = r + 1 and |Q| = s + 1, so r, s 1 since primitive
collections have at least two elements, and r + s = n since |P| + |Q| = n + 2.
Now rename the minimal generators of the rays in P as e0 , . . . , er . Thus
e0 + + er = 0.
343
The next step is to rename theP

minimal generators of the rays in P as v0 , . . . , vs .
Proposition 7.3.6 implies that sj=0 v j lies in a cone whose rays lie in the
complement of Q, which is P. Since P is a primitive collection, must omit at
least one element of P, which we may assume to be the ray generated by e0 . Then
the primitive relation of Q can be written
v0 + + vs = a1 e1 + + ar er ,
and by further relabeling, we may assume 0 a1 ar . Finally, observe that
v1 , . . . , vs , e1 , . . . , er generate a maximal cone of by (7.3.8). Since is smooth,
it follows that these r + s vectors form a basis of N. Comparing all of this to
Example 7.3.5, we conclude that
the toric variety of is the projective bundle

P OPs OPs (a1 ) OPs (ar ) .
The classification result proved in [177] is more general than the one given in
Theorem 7.3.7. By using a result from [191] on sphere triangulations with few
vertices, Kleinschmidt does not need to assume that X is projective. Another
proof of Theorem 7.3.7 that does not assume projective can be found in [14, Thm.
4.3]. We should also mention that (7.3.8) can be proved using the Gale transforms
discussed in [93, II.46] and [281, Ch. 6]. We will explain this is 15.2.
Exercises for 7.3.
7.3.1. Here you will supply some details needed to prove Theorem 7.3.1.
(a) In the proof we constructed a map Ue C. Show that this map is (m ,1) , where
D (u) = hm , ui for u .
(b) Given cones , , the transition map from U C U C to U C

U C is given by (u,t) 7 (u, g (u)t). Prove that g = m m .
7.3.2. In Example 7.3.2, we study the rank 1 vector bundle V P n whose sheaf of sections
is OPn (1). Let be the fan of V in Rn+1 .
(a) Prove that e1 , . . . , en+1 , e1 en + en+1 are the minimal generators of .
(b) Prove that is the normal fan of
P = Conv(0, e1 , . . . , en ) + Cone(en+1 , e1 + en+1 , . . . , en + en+1 ).

(c) The example constructs a morphism V P n Cn+1 . Prove that the image of this map
is defined by xi y j = x j yi and explain how this relates to Example 6.0.19.
7.3.3. Consider the locally free sheaf (7.3.1) and the cones
e NR Rr defined in the
discussion following (7.3.1). Prove that these cones and their faces give a fan in NR Rr
whose toric variety is the vector bundle with (7.3.1) as sheaf of sections.
7.3.5. Let P(E ) P s be the toric projective bundle constructed in Example 7.3.5. Prove
that Pic(P(E )) Z2 .
7.3.6. Let D be an ample effective divisor on a complete normal variety X. The goal of
this exercise is to prove that X \ Supp(D) is affine.
344
(a) Assume that D is very ample. Let s (X, OX (D)) be nonzero and consider the
nonvanishing set of s defined by U = {s X | s(x) 6= 0}. Prove that U is affine.
Hint: Show that a basis s = s0 , s1 , . . . , sm of (X, OX (D)) gives a closed embedding
X P m . Let P m have homogeneous coordinates x0 , . . . , xm and regard X as a subset
of P m . Prove that U = X U0 , where U0 P m is where x0 6= 0.
(b) Explain why part (a) remains true when D is ample but not necessarily very ample.
Hint: s k (X, OX (kD)).
(c) Since D is effective, 1 (X, OX (D)) is a global section. Prove that the nonvanishing set of this global section is X \ Supp(D). Hint: For s (X, OX (D)), recall the
definition of div0 (s) given in 4.0.
Parts (b) and (c) imply that X \ Supp(D) is affine when D is ample, as desired. Note also
that part (b) is a special case of Proposition 7.A.6.
7.3.7. By Example 2.3.16, the rational normal scroll Sa,b is the toric variety of
Pa,b = Conv(0, ae1 , e2 , be1 + e2 ) R2 ,
where a, b N satisfy 1 a b, and Sa,b Hba by Example 3.1.16. Thus rational
normals scroll are Hirzebruch surfaces. Here you will explore an n-dimensional analog.
Take integers 1 d0 d1 dn1 . Then Pd0 ,...,dn1 is the lattice polytope in Rn
having the 2n lattice points
0, d0 e1 , e2 , e2 + d1 e1 , e3 , e3 + d2 e1 , . . . , en , en + dn1 e1
as vertices. The toric variety of Pd0 ,...,dn1 is denoted Sd0 ,...,dn1 .
(a) Explain why Pd0 ,...,dn1 is a truncated prism whose base in {0} Rn1 is the standard
simplex n1 , and above the vertices of n1 there are edges of lengths d0 , . . . , dn1 .
Here, above means the e1 direction. Draw a picture when n = 3.
(b) Prove that Sd0 ,...,dn1 P(OP1 (d0 ) OP1 (dn1 )).
(c) Sd0 ,...,dn1 is smooth by part (b), so that Pd0 ,...,dn1 is very ample and hence gives a
projective embedding of Sd0 ,...,dn1 . Explain how this embedding consists of n embeddings of P1 such that for each point p P1 , the resulting n points in projective space
are connected by an (n 1)-dimensional plane that lies in Sd0 ,...,dn1 .
(d) Explain how part (c) relates to the scroll discussion in Example 2.3.16.
(e) Show that the (n 1)-dimensional plane associated to p P1 in part (c) is the fiber of
the projective bundle P(OP1 (d0 ) OP1 (dn1 )) P1 .
7.3.8. Consider the toric variety P(E ) constructed in Example 7.3.5.
(a) Prove that P(E ) is projective. Hint: Proposition 7.0.5.
(b) Show that P(E ) P(OPs (1) OPs (a1 + 1) OPs (ar + 1)). Hint: Part (b) of
Lemma 7.0.8.
(c) Find a lattice polytope in Rs Rr whose toric variety is P(E ). Hint: In the polytope
of Exercise 7.3.7, each vertex of {0} n1 R Rn1 is attached to a line segment
in the normal direction. Also observe that a line segment is a multiple of 1 . Adapt
this by using {0} r Rs Rr as base and then, at each vertex of r , attach a
positive mutliple of s in the normal direction.
345
7.3.9. Let X be a projective toric variety and let D0 , . . . , Dr be torus-invariant ample divisors on X . Each Di gives a lattice polytope Pi = DPi whose normal fan is . Prove that
the projective bundle P(OX (D0 ) OX (Dr )) is the toric variety of the polytope in
NR R
Conv(P0 {0} P1 {e1} Pr {er }).
Hint: If you get stuck, see [61, Sec. 3]. Do you see how this relates to Exercise 7.3.8?
7.3.10. Use primitive collections to show that P n is the only smooth projective toric variety
with Picard number 1.

In this appendix, we discuss some technical details related to projective morphisms.
ements
Ampleness. A comprehensive treatment of ampleness appears in Volume II of El
de geometrie algebrique (EGA) by Grothendieck and Dieudonne [127]. The results we
need from EGA are spread out over several sections. Here we collect the definitions and
theorems we will use in our discusion of ampleness.1
Definition 7.A.1. A line bundle L on a variety X is absolutely ample if for every coherent
sheaf F on X, there is an integer k0 such that F OX L k is generated by global sections
for all k k0 .
By [127, (4.5.5)], this is equivalent to what EGA calls ample in [127, (4.5.3)]. We
use the name absolutely ample to prevent confusion with Definition 6.1.9, where ample
is defined for line bundles on complete varieties. Here is another definition from EGA.
Definition 7.A.2. Let f : X Y be a morphism. A line bundle L on X is relatively ample
with respect to f if Y has an affine open cover {Ui } such that for every i, L | f 1 (Ui ) is
absolutely ample on f 1 (Ui ).
This is [127, (4.6.1)]. When mapping to an affine variety, relatively ample and absolutely ample coincide. More precisely, we have the following result from [127, (4.6.6)].
Proposition 7.A.3. Let f : X Y be a morphism, where Y is affine, and let L be a line
bundle on X. Then:
L is relatively ample with respect to f L is absolutely ample.
The reader should be warned that in EGA, relatively ample with respect to f and
f -ample are synonyms. In this text, they are slightly different, since relatively ample
with respect to f refers to Definition 7.A.2 while f -ample refers to Definition 7.2.5.
Fortunately, they coincide when the map f is proper.
Theorem 7.A.4. Let f : X Y be a proper morphism and L a line bundle on X. Then
the following are equivalent:
(a) L is relatively ample with respect to f in the sense of Definition 7.A.2.
1The theory developed in EGA applies to very general schemes. The varieties and morphisms appearing in
this book are nicely behavedthe varieties are quasi-compact and noetherian, the morphisms are of finite type,
and coherent is equivalent to quasicoherent of finite type. Hence most of the special hypotheses needed for some
of the results in [127] are automatically true in our situation.
346
(b) L is f -ample in the sense of Definition 7.2.5.

(c) There is an integer k > 0 such that f is projective with respect to L k in the sense of
Definition 7.0.3.
Proof. First observe that (b) and (c) are equivalent by Definition 7.2.5. Now suppose that
f is projective with respect to L k . Then there is an affine open covering {Ui } of Y such
that for each i, there is a finite-dimensional subspace W (Ui , L k ) that gives a closed
embedding of f 1 (Ui ) into Ui P(W ) for each i.
The locally free sheaf E = W C OUi is the sheaf of sections of the trivial vector
bundle Ui W Ui . This gives the projective bundle P(E ) = Ui P(W ), so that we
have a closed embedding
f 1 (Ui ) P(E ).
k
By definition [127, (4.4.2)], L | f 1 (Ui ) is very ample for f | f 1 (Ui ) . Then [127, (4.6.9)]
implies that L | f 1 (Ui ) is relatively ample with respect to f | f 1 (Ui ) , and hence absolutely
ample by Proposition 7.A.3. Then L is relatively ample with respect to f by Definition 7.A.2.
Finally, suppose that L is relatively ample with respect to f and let {Ui } be an affine
open covering of Y . Then [127, (4.6.4)] implies that L | f 1 (Ui ) is relatively ample with
respect to f | f 1 (Ui ) . By [127, (4.6.9)], L k | f 1 (Ui ) is very ample for f | f 1 (Ui ) , which by
definition [127, (4.4.2)] means that f 1 (Ui ) can be embedded in P(E ) for a coherent sheaf
E on Ui . Then the proof of [273, Thm. 5.44] shows how to find finitely many sections of
L k over f 1 (Ui ) that give a suitable embedding of f 1 (Ui ) into Ui P(W ).
In EGA [127, (5.5.2)], the definition of when a morphism f : X Y is projective

involves two equivalent conditions stated in [127, (5.5.1)]. The first condition uses the
projective bundle P(E ) of a coherent sheaf E on Y , and the second uses Proj(S ), where
S is a quasicoherent graded OY -algebra such that S1 is coherent and generates S . By
[127, (5.5.3)], projective is equivalent to proper and quasiprojective, and by the defintion
of quasiprojective [127, (5.5.1)], this means that X has a line bundle relatively ample with
respect to f . Hence Theorem 7.A.4 shows that the definition of projective morphism given
in EGA is equivalent to Definition 7.0.3.
We close with two further results about projective morphisms. Proofs can be found in
[127, (4.6.4)] and [127, (5.5.7)] respectively.
Proposition 7.A.5. Let f : X Y be a proper morphism and L a line bundle on X. Given
an affine open cover {Ui } of Y , the following are equivalent:
(a) L is f -ample.
(b) For every i, L | f 1 (Ui ) is f | f 1 (Ui ) -ample.
Proposition 7.A.6. Let f : X Y be a projective morphism with Y affine and let L be an

f -ample line bundle on X. Then:
(a) Given a global section s (X, L ), let Xs X be the open subset where s is nonvanishing. Then Xs is an affine open subset of X.
(b) There is an integer k0 such that L k is generated by global sections for all k k0 .
Chapter 8
The Canonical Divisor

of a Toric Variety
8.0. Background: Reflexive Sheaves and Differential Forms

This chapter will study the canonical divisor of a toric variety. The theory developed in Chapters 6 and 7 dealt with Cartier divisors and line bundles. As we will
see, the canonical divisor of a normal toric variety is a Weil divisor that is not
necessarily Cartier. We will also study the sheaves associated to Weil divisors.
Reflexive Sheaves. A Weil divisor D on a normal variety X gives the sheaf OX (D)
defined by
(U , OX (D)) = { f C(X ) | (div( f ) + D)|U 0} {0}.
Our first task is to characterize these sheaves.
Recall that the dual of a sheaf of OX -modules F is F = HomOX (F , OX ).
We say that F is reflexive if the natural map
F F
is an isomorphism. It is easy to see that locally free sheaves are reflexive. Here are
some properties of reflexive sheaves.
Proposition 8.0.1. Let F be a coherent sheaf on a normal variety X and consider
the inclusion j : U0 X where U0 is open with codim(X \U0 ) 2. Then:
(a) F and hence F are reflexive.
(b) If F is reflexive, then F j (F |U0 ).
(c) If F |U0 is locally free, then F j (F |U0 ).

347
Chapter 8. The Canonical Divisor of a Toric Variety
348
Proof. Recall from 4.0 that the direct image j G of a sheaf G on U0 is defined by
(U , j G ) = (U U0 , G ) for U X open.
Parts (a) and (b) of the proposition are proved in [132, Cor. 1.2 and Prop. 1.6].
For part (c), we first observe that restriction is compatible with taking the dual, i.e.,
(G )|U0 = (G |U0 ) for any coherent sheaf G on X . Then
F j ((F )|U0 ) = j ((F |U0 ) ) j (F |U0 ),
where the first isomorphism follows from parts (a) and (b), and the last follows
since F |U0 is locally free and hence reflexive.

Later in the section we will study the sheaf Xp of p-forms on X . This sheaf is
locally free when X is smooth. For X normal, however, Xp may be badly behaved,
though it is locally free on the smooth locus of X . Hence we can use part (c) of
Proposition 8.0.1 to create a reflexive version of Xp .
For more on reflexive sheaves, the reader should consult [132] and [235].
Reflexive Sheaves of Rank One. We first define the rank of a coherent sheaf on an
irreducible variety X . Recall that KX is the constant sheaf on X given by C(X ).
Definition 8.0.2. Given a F coherent sheaf on irreducible variety X , the global
sections of F OX KX form a finite-dimensional vector space over C(X ) whose
dimension is the rank of F .
For a locally free sheaf, the rank is just the rank of the associated vector bundle.
Other properties of the rank will be studied in Exercise 8.0.1.
In the smooth case, reflexive sheaves of rank 1 are easy to understand.
Proposition 8.0.3. On a smooth variety, a coherent sheaf of rank 1 is reflexive if
and only if it is a line bundle.

This is proved in [132, Prop. 1.9]. We now have all of the tools needed to
characterize which coherent sheaves on a normal variety come from Weil divisors.
Theorem 8.0.4. Let L be a coherent sheaf on a normal variety X . Then the
(a) L is reflexive of rank 1.
(b) There is an open subset j : U0 X such that codim(X \ U0 ) 2, L |U0 is a
line bundle on U0 , and L j (L |U0 ).
(c) L OX (D) for some Weil divisor D on X .
Proof. (a) (b) Since X is normal, its singular locus Y = Sing(X ) has codimension at least two in X by Proposition 4.0.17. Then U0 = X \ Y is smooth, which
implies that L |U0 is a line bundle by Proposition 8.0.3. Hence L j (L |U0 ) by
Proposition 8.0.1.
349
(b) (c) The line bundle L |U0 can be written as OU0 (E) for some Cartier
P
P
divisor E = i ai Ei on U0 . Consider the Weil divisor D = i ai Di , where Di is the
Zariski closure of Ei in X . Given f C(X ) , note that
div( f ) + D 0 (div( f ) + D)|U0 0
since codim(X \U0 ) 2, and the same holds over any open set of X . Combining
this with E = D|U0 , we obtain
OX (D) j OU0 (E) = j (L |U0 ) L .
(c) (a) The proof of (b) (c) shows that OX (D) j (OX (D)|U0 ). But
codim(X \U0 ) 2, and OX (D)|U0 = OU0 (D|U0 ) is locally free since U0 is smooth.
Thus j (OX (D)|U0 ) OX (D) by Proposition 8.0.1, so OX (D) OX (D) is
reflexive and has rank 1 since it is a line bundle on U0 .

Tensor Products and Duals. Given Weil divisors D, E on a normal variety X , the
map f g 7 f g defines a sheaf homomorphism
(8.0.1)
OX (D) OX OX (E) OX (D + E).
This is an isomorphism when D or E is Cartier but may fail to be an isomorphism

in general.
Example 8.0.5. Consider the affine quadric cone X = V(y 2 xz) C3 . From
examples in previous chapters, we know that this is a normal toric surface. The
line L = V(y, z) gives a Weil divisor that is not Cartier, though 2L is Cartier (this
follows from Example 4.2.3). The coordinate ring of X is R = C[x, y, z]/hy 2 xzi.
Let x, y, z denote the images of the variables in R. In Exercise 8.0.2 you will show
the following:
(X , OX (L)) is the ideal hy, zi R.
(X , OX (2L)) is the ideal hzi R (principal since 2L is Cartier).
On global sections, the image of the map
OX (L) OX OX (L) OX (2L)
is hy, zi2 , which is a proper subset of (X , OX (2L)) = hzi.
It follows that OX (L) OX OX (L) 6 OX (2L).
If we apply (8.0.1) when E = D, we get a map

OX (D) OX OX (D) OX ,
which in turn induces a map
(8.0.2)
OX (D) OX (D) .
As noted in 6.0, this is an isomorphism when D is Cartier. In general, we have the

following result about the maps (8.0.1) and (8.0.2).
350
Proposition 8.0.6. Let D, E be Weil divisors on a normal variety X . Then (8.0.1)

(OX (D) OX OX (E)) OX (D + E).
Furthermore, (8.0.2) is an isomorphism, i.e.,
OX (D) OX (D) .
Proof. The first isomorphism follows from Proposition 8.0.1 since OX (D + E) is
reflexive and (8.0.1) is an isomorphism on the smooth locus of X . The second
isomorphism follows similarly since both sheaves are reflexive and (8.0.2) is an
isomorphism on the smooth locus.

Divisor Classes. Recall that Weil divisors D and E on X are linearly equivalent,
written D E, if D = E + div( f ) for some f C(X ) .
Proposition 8.0.7. Let X be a normal variety.
(a) If D and E are Weil divisors on X , then
OX (D) OX (E) D E.
(b) If D is a Weil divisor on X , then
D is Cartier OX (D) is a line bundle.
Proof. Linearly equivalent divisors give isomorphic sheaves by Proposition 4.0.29.
Conversely, OX (D) OX (E) implies
OX (D) OX OX (E) OX (E) OX OX (E).
Taking the double dual and using Proposition 8.0.6, we get OX (DE) OX . From
here, showing that D E follows exactly as in the proof of Proposition 6.0.22.
One direction of part (b) was proved in Chapter 6 (see Proposition 6.0.17 and
Theorem 6.0.18). Conversely, if OX (D) is a line bundle on X , then Theorem 6.0.20
shows that OX (D) OX (E) for some Cartier divisor E. Thus D E by part (a).
Then we are done since any divisor linearly equivalent to a Cartier divisor is Cartier
by Exercise 4.0.5.

In Chapter 4 we defined the class group Cl(X ) and Picard group Pic(X ) in
terms of Weil and Cartier divisors. Then, in Chapter 6, we reinterpreted Pic(X )
as the group of isomorphism classes of line bundles, where the group operation
was tensor product and the inverse was the dual. We can now reinterpret Cl(X ) as
the group of isomorphism classes of reflexive sheaves of rank 1, where the group
operation is the double dual of the tensor product and the inverse is the dual. This
follows immediately from Propositions 8.0.6 and 8.0.7.
351
Kahler Differentials. In order to give an algebraic definition of differential forms

on a variety, we begin with the case of a C-algebra.
Definition 8.0.8. Let R be a C-algebra. The module of Kahler differentials of R
over C, denoted R/C , is the R-module generated by the formal symbols df for
f R, modulo the relations
(a) d(c f + g) = c df + dg for all c C, f , g R.
(b) d( f g) = f dg + g df for all f , g R.
Example 8.0.9. If R = C[x1 , . . . , xn ], then

R/C
n
M
R dxi .
i=1
This follows because the relations defining R/C imply df =

f R (Exercise 8.0.3).
Pn
f
i=1 xi
dxi for all
A C-algebra homomorphism R S induces a natural homomorphism

R/C S/C .
When we regard S/C as an R-module, we obtain a homomorphism of S-modules
S R R/C S/C .
Here is a case when this map is easy to understand. See [195, Thm. 25.2] for a
proof.
Proposition 8.0.10. Let R S be a surjection of C-algebras with kernel I. Then
there is an exact sequence of S-modules
I/I 2 S R R/C S/C 0,
where [ f ] I/I 2 maps to 1 df S R R/C .
Example 8.0.11. Let R = C[x1 , . . . , xn ] and S = R/I, where I = h f1 , . . . , fs i. The

generators of I give a surjection R s I and hence a surjection S s I/I 2 . Combining this with Proposition 8.0.10 and Example 8.0.9, we obtain an exact sequence
S s S n S/C 0,
where is given by the reduction of the n s Jacobian matrix
f1
fs
x1
x1
..
..
(8.0.3)
.
.
f1
xn
fs
xn
modulo the ideal I (Exercise 8.0.4). This presentation of S/C is very useful for
computing examples.
352
Kahler differentials also behave well under localization, as you will prove in
Exercise 8.0.5.
Proposition 8.0.12. Let R f be the localization of R at a non-nilpotent element
f R. Then R f /C R/C R f .

Cotangent and Tangent Sheaves. Now we globalize Definition 8.0.8.
Definition 8.0.13. Let X be a variety. The cotangent sheaf 1X is the sheaf of
OX -modules defined via
1X (U ) = OX (U)/C
on affine open sets U . The tangent sheaf TX is the dual sheaf
TX = (1X ) = HomOX (1X , OX ).
The reason for the superscript in the notation for the cotangent sheaf will become clear later in this chapter. In Exercise 8.0.6 you will use Example 8.0.11 and
Proposition 8.0.12 to show that 1X is a coherent sheaf. See [131, II.8] for a slightly
different approach to defining the sheaf 1X , and [131, II.8, Comment 8.9.2] for the
connection between these methods.
When U = Spec(R) is an affine open of X , the definition of the tangent sheaf
implies that
TX (U ) = HomR (R/C , R).
This can also be described in terms of derivationssee Exercises 8.0.7 and 8.0.8.
When X is smooth, these sheaves are nicely behaved, as shown by the following result from [131, Thm. II.8.15].
Theorem 8.0.14. A variety X is smooth if and only if 1X is locally free. When this
happens, 1X and TX are locally free sheaves of rank n, n = dim X .

In the smooth toric case, it is easy to see that the cotangent sheaf is locally free.
Example 8.0.15. A smooth cone NR Rn of dimension r gives the affine toric
variety
U Cr (C )nr Cn .
Then Example 8.0.9 and Proposition 8.0.12 imply that U1 is locally free of rank n.
It follows immediately that 1X is locally free for any smooth toric variety X .
We know from Chapter 6 that a locally free sheaf is the sheaf of sections of a
vector bundle. When X is smooth, the vector bundles corresponding to 1X and TX
are called the cotangent bundle and tangent bundle respectively.
353
Example 8.0.16. We construct the cotangent bundle for P 2 . Recall that P 2 has a
covering by the affine open sets
U0 Spec(C[x, y])
U1 Spec(C[yx1 , x1 ])
U2 Spec(C[xy1 , y1 ]).
where 0 , 1 , 2 are the maximal cones in the usual fan for P 2 .
Let C2 = Spec(R) for R = C[x, y]. The module R/C is a free R-module of rank
2 with generators dx, dy by Example 8.0.9. Thus a 1-form on C2 may be written
uniquely as f1 dx + f2 dy, where fi R. To generalize this to P 2 , we require that
after changing coordinates, dx and dy transform via the Jacobian matrix described
in Example 8.0.11. More precisely, the matrix for the transition function i j will
be the Jacobian of the map U j Ui .
On U2 , the coordinates (a1 , a2 ) are represented in terms of the (x, y) coordinates on U0 as (xy1 , y1 ), yielding

1/y x/y 2
20 =
.
0 1/y 2
Next, we compute 12 . Things get messy if we keep everything in (x, y) coordinates, so we first translate to coordinates (a1 , a2 ) on U2 , and then translate back.
On U2 we identify (a1 , a2 ) with (xy1 , y1 ). Then U1 has coordinates

1 a2
(yx1 , x1 ) =
,
.
a1 a1
So in terms of (a1 , a2 ), we have

1/a12
0
.
12 =
a2 /a12 1/a1

Rewriting this in terms of (x, y) yields

2 2
y /x
0
.
12 =
y/x 2 y/x
Finally, computing 10 directly, we obtain

y/x 2 1/x
.
10 =
1/x 2 0
A check shows that 10 = 12 20 . Similar computations show that the compatibility criteria are satisfied for all i, j, k, i.e.,
ik = i j jk .
Since det(i j ) is invertible on Ui U j , the same is true for i j . Hence we obtain
a rank 2 vector bundle on P 2 whose sheaf of sections is 1P2 .
354
Relation with the Zariski Tangent Space. The definition of the tangent sheaf
TX seems far removed from the definition of the Zariski tangent space Tp (X ) =
HomC (mX,p /m2X,p , C) given in Chapter 1. Here we explain (without proof) the
connection.
The stalk (TX ) p of the tangent sheaf at p X can be described as follows. The
stalk of 1X at p is the module of Kahler differentials
(1X ) p = OX,p /C ,
where OX,p is the local ring of X at p. Since OX,p /mX,p C and C/C = 0 (easy
to check), the exact sequence of Proposition 8.0.10 gives a surjection
mX,p /m2X,p OX,p /C OX,p C
which is an isomorphism of vector spaces over C by [131, Prop. II.8.7]. Since TX

is dual to 1X , taking the dual of the above isomorphism gives
(8.0.4)
(TX ) p OX,p C HomC (mX,p /m2X,p , C) = Tp (X ).
This omits many details but should help you understand why TX is the correct
definition of tangent sheaf.
Example 8.0.17. Let V Cn be defined by I = I(V ) = h f1 , . . . , fs i C[x1 , . . . , xn ].
The coordinate ring of V is S = C[x1 , . . . , xn ]/I, so that Example 8.0.11 gives the
exact sequence
S s S n S/C 0.
Now take p V and tensor with OV ,p to obtain the exact sequence

OVs ,p OVn,p OV ,p /C 0
(Exercise 8.0.9). If we tensor this with C and dualize, (8.0.4) and the isomorphism
OX,p /C OX,p C mX,p /m2X,p give the exact sequence
0 Tp (V ) Cn Cs ,

fi
where comes from the s n Jacobian matrix x
(p)
(Exercise 8.0.9). This
j
explains the description of Tp (V ) given in Lemma 1.0.6.
Conormal and Normal Sheaves. Given a closed subvariety i : Y X , it is natural

to ask how their cotangent sheaves relate. We begin with the exact sequence
0 IY OX i OY 0,
which we write more informally as
0 IY OX OY 0.
The quotient sheaf IY /IY2 has a natural structure as a sheaf of OY -modules, as

does F OX OY for any sheaf F of OX -modules. The following basic result is
proved in [131, Prop. II.8.12 and Thm. II.8.17].
355
Theorem 8.0.18. Let Y be a closed subvariety of a variety X . Then:

(a) There is an exact sequence of OY -modules
IY /IY2 1X OX OY Y1 0.
(b) If X and Y are smooth, then this sequence is also exact on the left and IY /IY2
is locally free of rank equal to the codimension of Y .

Note that part (a) of this theorem is a global version of Proposition 8.0.10. We
call IY /IY2 the conormal sheaf of Y in X and call its dual
NY /X = (IY /IY2 ) = HomOY (IY /IY2 , OY )
the normal sheaf of Y in X . When X and Y are smooth, we can dualize the sequence
appearing in Theorem 8.0.18 to obtain the exact sequence
0 TY TX OX OY NY /X 0.
The vector bundle associated to NY /X is the normal bundle of Y in X . Then the
above sequence says that when the tangent bundle of X is restricted to the subvariety Y , it contains the tangent bundle of Y with quotient given by the normal bundle.
This is the algebraic analog of what happens in differential geometry, where the
normal bundle is the orthogonal complement of the tangent bundle of Y .
Differential Forms. We call 1X the sheaf of 1-forms, and we define the sheaf of
p-forms to be the wedge product
Xp =
Vp
1X .
For any sheaf F of OX -modules, the exterior power p F is the sheaf

V associated
to the presheaf which to each open set U assigns the OX (U )-module p F (U ).
Example 8.0.19. For Cn , 1Cn is the sheaf associated to the free R-module R/C =
Ln
p
i=1 R dxi , R = C[x1 , . . . , xn ]. Then Cn is the sheaf associated to
M
Vp
R/C =
R dxi1 dxi p .
1i1 <<i p n
It follows that
Cp n
is free of rank
n
p .

More generally, Theorem 8.0.14 implies that Xp is locally free of rank np
when X is smooth of dimension n. In particular, Xn is a line bundle in this case.
Zariski p-Forms and the Canonical Sheaf . For a normal variety X , the sheaf of
p-forms Xp may fail to be locally free. However, this sheaf is locally free on the
smooth locus of X , and the complement of the smooth locus has codimension 2
since X is normal. Hence we can use Proposition 8.0.1 to define the sheaf of Zariski
p-forms
b p = ( p ) = j p ,
(8.0.5)
U0
356
b p is a
where j : U0 X is the inclusion
of the smooth locus of X . It follows that
X
n
reflexive sheaf of rank p , where n = dim X .
For later purposes, we note that by Proposition 8.0.1, (8.0.5) is valid for any
smooth open subset U0 X whose complement has codimension 2.
The case p = n is especially important.
Definition 8.0.20. The canonical sheaf of a normal variety X is

b Xn ,
X =
where n is the dimension of X . This is a reflexive sheaf of rank 1, so that

X OX (D)
for some Weil divisor D on X . We call this divisor a canonical divisor of X , often
denoted KX .
Proposition 8.0.7 shows that the canonical divisor KX is well-defined up to
linear equivalence and hence gives a unique divisor class in Cl(X ), known as the
canonical class of X . In the toric case, we will see later in the chapter that there is
a natural choice for the canonical divisor.
When X is smooth, we call X the canonical bundle since it is a line bundle. In
this case, the canonical divisor is Cartier. There are also singular varieties whose
canonical divisors are Cartierthese are the Gorenstein varieties to be studied later
in the chapter.
While it often suffices to know X up to isomorphism, there are
V situations
where a unique model of X is required. One such construction uses n C(X)/C ,
V
where C(X ) is the field of rational functions on X . We can regard n C(X)/C as
the constant sheaf of rational n-forms on X , similar to the way that C(X ) gives the
constant sheaf KX of rational functions on X . There is an obvious sheaf map
Xn
Vn
C(X)/C .
You will prove the following result in Exercises 8.0.10 and 8.0.11.
Proposition 8.0.21. When X is normal, image of the map Xn
X
Vn
C(X)/C .
Vn
C(X)/C is
The canonical sheaf can be defined for any irreducible variety X as a subsheaf
V
of n C(X)/C , though the definition is more sophisticated (see [184, 9]). When
X is projective, another approach is given in [131, III.7], where X is called the
dualizing sheaf. We will see in Chapter 9 that X plays a key role in Serre duality.
Exercises for 8.0.
8.0.1. The rank of a coherent sheaf on an irreducible variety was defined in Definition 8.0.2.
Here are some properties of the rank.
357
(a) Let an irreducible affine variety have coordinate ring R with field of fractions K. Let
M be a finitely generated R-module. Show that M R K is a finite-dimensional vector
e on Spec(R).
space over K whose dimension equals the rank of the coherent sheaf M
(b) Let F be a coherent sheaf on X let U X be an nonempty open subset. Prove that F
and F |U have the same rank.
(c) Let 0 F G H be an exact sequence of sheaves on X. Prove that rank(G ) =
rank(F ) + rank(H ).

8.0.3. Let R = C[x1 , . . . , xn ]. In Example 8.0.9 we claimed that df =
for all f R. Prove this.
Pn
f
i=1 xi
dxi in R/C
8.0.4. Prove that the map in the exact sequence from Example 8.0.11 comes from the
Jacobian matrix (8.0.3).
8.0.6. Prove that the cotangent sheaf 1X defined in Definition 8.0.13 is a coherent sheaf.
8.0.7. Given a C-algebra R and an R-module M, a C-derivation : R M is a C-linear
map that satisfies the Leibniz rule, i.e., ( f g) = f (g) + g( f ) for all f , g R.
(a) Show that f df defines a C-derivation d : R R/C .
(b) More generally, show that if : R/C M is an R-module homomorphism, then
d : R M is a C-derivation.
8.0.8. Continuing Exercise 8.0.7, we let DerC (R, M) denote the set of all C-derivations
: R M. This is an R-module where (r)( f ) = r( f ).
(a) Use part (b) of Exercise 8.0.7 to construct an R-module isomorphism DerC (R, M)
HomR (R/C , M). Explain why d : R R/C is called the universal derivation.
(b) Let TX be the tangent sheaf of a variety X and let U = Spec(R) be an affine open
subset of X. Prove that TX (U) = DerC (R, R).
8.0.9. Fill in the details omitted in Example 8.0.17.

8.0.10. Let j : U X be the inclusion of a nonempty open subset of a variety X.
(a) Show that there is a sheaf map F j (F |U ) for any sheaf F on X.
(b) Show that the map of part (a) is an isomorphism when X is irreducible and F is a
constant sheaf.
8.0.11. Prove Proposition 8.0.21. Hint: 1X is locally free when restricted to the smooth
locus of X. Exercise 8.0.10 will be useful.
8.0.12. Let 1 TN be the identity element of the torus TN and let m C[M] be the corresponding maximal ideal. Set NC = N Z C and MC = M Z C.
P
P
m
2
0 in C and
(a) Let
P f = m cm C[M]. Show that f m if and only if m cm e=
i
,
so
that f is
c
m
=
0
in
M
.
Hint:
Pick
a
basis
e
,
.
.
.
,
e
of
M
and
set
t
=
m
C
1
n
i
m
a Laurent monomial in t1 , . . . ,tn . Then show that f m2 if and only if f m and
f
ti (1) = 0 for all i.
(b) Use part (a) to construct an isomorphism m/m2 MC , and conclude that the Zariski
tangent space of TN at the identity is naturally isomorphic to NC .
358
8.0.13. Let F be a free module of rank n over a ring R and fix 0 p n. Recall that wedge
Vnp
Vp Vn
product induces an isomorphism
F HomR ( F, F).
(a) If X is smooth of dimension n, then show that Xnp H omOX (Xp , Xn ).
(b) Show that if X is a normal variety of dimension n, then there is an isomorphism

b np H omOX (
b p , ). Hint: If you get stuck, see [76, Prop. 4.7].
X
X
X
(c) In a similar vein, show that the tangent sheaf TX of a normal variety X satisfies TX
b 1 , OX ).
H omOX (
X
8.1. One-Forms on Toric Varieties
In this section we will describe two interesting exact sequences that involve the
b 1 on a normal toric variety X .
sheaves X1 and
X
The Torus. The coordinate ring of the torus TN is the semigroup algebra C[M].
Then the map
C[M]/C M Z C[M]
defined by d m 7 m m is easily seen to be an isomorphism. It follows that

(8.1.1)
1TN M Z OTN ,
and dualizing, we obtain

TTN N Z OTN = NC C OTN .
This makes intuitive sense since TN = N Z C as a complex Lie group. Thus its
tangent space at the identity is N Z C = NC via the exponential map. This is also
true algebraically, as shown in Exercise 8.0.12. The group action transports the
tangent space NC over the whole torus, which explains the above trivialization of
the tangent bundle TTN .
m
1
As a consequence, the 1-form d
m is a global section of TN that maps to m 1
in (8.1.1) and hence is invariant under the action of TN . See Exercise 8.1.1 for more
on invariant 1-forms on the torus.
The First Exact Sequence. Now consider the toric variety X of the fan . For
(1), the inclusion i : D X gives the sheaf i OD on X , which following
8.0 we write as OD . Using the map M Z given by m 7 hm, u i, we obtain the
composition
M Z OX Z Z OX = OX OD .
This gives a natural map
(8.1.2)
: M Z OX
OD ,
where the direct sum is over all (1). We also have a canonical map
(8.1.3)
: 1X M Z OX
359
constructed as follows. On the affine piece U = Spec(C[ M]), is defined by

d m C[ M]/C 7 m m M Z C[ M].
These C[ M]-module homomorphisms C[ M]/C M Z C[ M] patch

to give the desired map (8.1.3) (Exercise 8.1.2). Note that over the torus TN , the
map of (8.1.3) reduces to the isomorphism (8.1.1).
Theorem 8.1.1. For a smooth toric variety X , the sequence
L
0 1X M Z OX OD 0
formed using (8.1.2) and (8.1.3) is exact.
Proof. We first verify that is the zero map. On the affine piece U X , the
subvariety D U U is defined by the ideal I = I(D U ) C[ M]. By
Propositions 4.0.28 and 4.3.2, we have
M
M
C m =
C m.
(8.1.4) I = (U , OX (D )) =
m M,hm,u i>0
div( m )|U D |U
Over U , the composition 1X M Z OX OD takes a 1-form d m , m

M, to hm, u i m C[ M]/I . This is obviously zero if hm, u i = 0, and
if hm, u i =
6 0, it vanishes since m I in this case.
We now verify that the sequence is exact over U . Since is smooth, we

may assume = Cone(e1 , . . . , er ), where r n and e1 , . . . , en is a basis of N.
Then U = Cr (C )nr . Let x1 , . . . , xn denote the characters of the corresponding dual basis of M, also denoted e1 , . . . , en . The coordinate ring of U is R =
1
, . . . , xn1 ], and the 1-forms on U form the free R-module R/C =
C[x1 , . . . , xr , xr+1
Ln
i=1 R dxi by Example 8.0.9 and Proposition 8.0.12. Since takes dxi to ei xi ,
we see that can be regarded as the map
R/C =
that sends
Pn
n
M
i=1
i=1 fi dxi
R dxi M R R =
n
M
i=1
to ( f1 x1 , . . . , fn xn ). This gives the exact sequence
0 R/C
n
M
i=1
r
M
i=1
R/hxi i 0
since xr+1 , . . . , xn are units in R, and the theorem follows.
Logarithmic Forms. The exact sequence of Theorem 8.1.1 has a lovely interpretation in terms of residues of logarithmic 1-forms. The idea is that M Z OX can be
thought of as the sheaf X1 (log D) of 1-forms on X with logarithmic poles along
P
D = D . We begin with an example.
360
Example 8.1.2. The coordinate ring of Cn is R = C[x1 , . . . , xn ], and the L

divisor D is
the sum of the coordinate hyperplanes Di = V(xi ). As above, R/C = ni=1 R dxi .
Now introduce some denominators: a rational 1-form has logarithmic poles
along D if
n
X
dxi
fi
=
, fi R.
xi
i=1
L
i
These form the free R-module ni=1 R dx
xi , and the corresponding sheaf is defined
to be 1Cn (log D). The formal calculation
n
dxi
d m X
hm, ei i
=
m
xi
i=1
shows that the map
d m
m
7 m 1 induces an isomorphism of sheaves

1Cn (log D) M Cn OCn
such that the map : 1Cn M Cn OCn of (8.1.3) is induced by the inclusion of
1-forms 1Cn 1Cn (log D).

This construction works for any smooth affine toric variety U , and the sheaves
of logarithmic 1-forms on U patch to give the sheaf X1 (log D) for any smooth
toric variety X . Furthermore, we have a canonical isomorphism
(8.1.5)
X1 (log D) M Z OX
such that the map of (8.1.3) comes from the inclusion of 1-forms.
The construction of X1 (log D) can be done more generally. Let X be a smooth
P
variety. A divisor D = i Di on X has simple normal crossings if every Di is
smooth and irreducible, and for every p X , the divisors containing p meet nicely.
More precisely, if I p = {i | p Di }, we require that the tangent spaces Tp (Di )
Tp (X ) meet transversely, i.e.,
\

codim
Tp (Di ) = |I p |.
P
iIp
For example, the divisor D = D is a simple normal crossing divisor on any

smooth toric variety X . A nice discussion of 1X (log D) for complex manifolds
can be found in [125, p. 449].
The Poincare Residue Map. Let f (z) be an analytic (also called holomorphic)
function defined in a punctured neighborhood of a point p C. Take a counterclockwise loop C around p on X such that f (z) has no other poles inside C. Then
the residue of the 1-form = f (z) dz at p is the contour integral
Z
1
res p () =
.
2i C
361
In particular, if p = 0 and f (z) = g(z)

z , with g(z) analytic at zero, then the residue
theorem tells us that res p () is the coefficient of 1z in the Laurent series for f (z) at
0, and is equal to g(0). Note that the 1-form = f (z) dz = g(z) dzz has a logarithmic
pole at p = 0.
When there are several variables, we can do the same construction by working
one variable at a time. Here is an example.
Example 8.1.3. Given f = f (x1 , . . . , xn ) R = C[x1 , . . . , xn ], we get the logarith1
mic 1-form = f dx
x1 . In terms of the above discussion of residues, we can regard
f (0, x2 , . . . , xn ) as the residue of at V(x1 ). Note also that f (0, x2 , . . . , xn ) represents the class of f in R/hx1 i. Doing this for every variable shows that the map
1Cn (log D)
M Z OC
n
n
M
ODi
i=1
can be interpreted as a sum of residue maps.
P
More generally, if X is a smooth variety and D = i Di a simple normal crossing divisor, one can define the Poincare residue map
M
Pr : 1X (log D)
ODi
i
(see [227, p. 254]) such that we have an exact sequence

M
Pr
(8.1.6)
0 1X 1X (log D)
ODi 0.
i
P
When applied to a smooth toric variety X and the divisor D = D , this gives
the exact sequence of Theorem 8.1.1 via the isomorphism (8.1.5).
The Normal Case. When X is normal, we get an analog of Theorem 8.1.1 that
b 1 of Zariski 1-forms in place of 1 . Since M Z OX is reflexive,
uses the sheaf
X
X
taking the double dual of (8.1.3) gives a map

b 1X M Z OX .
Theorem 8.1.4. Let X be a normal toric variety. Then:

(a) The sequence
is exact.
b 1 M Z OX L OD
0
X
(b) If X is simplicial, then the map on the right is surjective, so that

b 1X M Z OX L OD 0
0
is exact.
362
S
Proof. Let j : U0 X be the inclusion map for U0 = U . Note that U0 is
a smooth toric variety whose fan has the same 1-dimensional cones as , and
codim(X \ U0 ) 2 by the Orbit-Cone Correspondence. By Theorem 8.1.1, we
have an exact sequence
L
0 U1 0 M Z OU0 OD U0 0,
so that applying j gives the exact sequence
0 j U1 0 j (M Z OU0 )
j OD U0
b 1 by the remarks
since j is left exact (Exercise 8.1.3). However, j U1 0 =
X
following (8.0.5), and j (M Z OU0 ) = M Z OX by Proposition 8.0.1. Hence we
get an exact sequence
L
b 1X M Z OX
0
j OD U0 .
In Exercise 8.1.4 you will show that the maps M Z OX j OD U0 factor as

M Z OX OD j OD U0 ,
where OD j OD U0 is injective. The exact sequence of part (a) then follows
immediately.
L
It remains to show that M Z OX OD is surjective when X is simplicial. Given , we need to show that
L
M Z OU (1) OD U
is surjective. Fix (1) and pick m M such that hm, u i =

6 0 and hm, u i = 0
for all 6= in (1). Such an m exists since is simplicial. Then m 1 maps

to a nonzero constant function on OD U and to the zero function on OD U for
6= . The desired surjectivity now follows easily.
When X has no torus factors, we learned in 5.3 that graded modules over the
total coordinate ring S = C[x | (1)] give quasicoherent sheaves on X . It is
b 1 . For each , there are two maps
easy to describe a graded S-module that gives
X
M Z S Z Z S = S S/hx i,
where the first map comes from m 7 hm, u i and the second map is obvious. This
L
b 1 to be the kernel
gives a homomorphism M Z S S/hx i, and we define
S
of this map. Hence we have an exact sequence of graded S-modules
b 1S M S L S/hx i.
(8.1.7)
0
Using Example 6.0.10 and Theorem 8.1.4, we obtain the following result.
b 1 is the sheaf associated to the

Corollary 8.1.5. When X has no torus factors,
X
1
b

graded S-module S .
363
The Euler Sequence. In [131, Thm. II.8.13], Hartshorne constructs an exact sequence
0 1Pn OPn (1)n+1 OPn 0,
(8.1.8)
called the Euler sequence of P n . He goes on to say This is a fundamental result,

upon which we will base all future calculations involving differentials on projective
varieties. Of course, P n is toric, and there is a toric generalization of this result,
due to Batyrev and Mel nikov [20] and Jaczewski [160] in the smooth case and
Batyrev and Cox [19, Thm. 12.1] in the simplicial case.
Theorem 8.1.6. Let X be a simplicial toric variety with no torus factors, i.e.,
{u | (1)} spans NR . Then there is an exact sequence
b 1X L OX (D ) Cl(X ) Z OX 0.
0
Furthermore, if X is smooth, then the sequence can be written

L
0 1X OX (D ) Pic(X ) Z OX 0.
Proof. Consider the following diagram:

0
0

/ M Z OX
0
/

/
b1

/ Cl(X ) Z OX

/ Cl(X ) Z OX
OX (D )
L
OX
0
L
OD
/
L
OD

/0
/0
/0
/0
The top row is from Theorem 8.1.4 and is exact since X is simplicial. Also, by
Proposition 4.0.28, each (1) gives an exact sequence
0 OX (D ) OX OD 0,
and the middle row is the direct sum of these exact sequences. The third row is the
obvious exact sequence that uses the identity map on Cl(X ) Z OX .
Since X has no torus factors, we have the exact sequence
L
0 M Z Cl(X ) 0
from Theorem 4.1.3, and tensoring this with OX gives the middle column. The
column on the right is the another obvious exact sequence, and one can check
without difficulty that the solid arrows in the diagram commute (Exercise 8.1.5).
364
Then commutativity and exactness imply the existence of the dotted arrows in the
diagram, which give the desired exact sequence by a standard diagram chase.
The exact sequence of sheaves in Theorem 8.1.6 is the (generalized) Euler
sequence of the toric variety X . We will use it in the next section to determine the
canonical sheaf of X . The Euler sequence also encodes relations generalizing the
classical Euler relation for homogeneous polynomials (see Exercise 8.1.8). Note
also that in [160], Jaczewski shows that smooth toric varieties can be characterized
as smooth varieties which admit a generalized Euler sequence.
Exercises for 8.1.
8.1.1. We will study invariant 1-forms and derivations on the torus. Since the torus TN =
Spec(C[M]) is affine, we know from Exercise 8.0.8 that the derivations DerC (C[M], C[M])
give the global sections of the tangent sheaf TTN .
(a) For u N, define u : C[M] C[M] by
u ( m ) = hm, ui m .
Prove that u DerC (C[M], C[M])
(b) Let x1 , . . . , xn be the characters corresponding to the elements of M dual to some par
ticular basis e1 , . . . , en of N. Thus C[M] = C[x11 , . . . , xn1 ]. Prove that ei = xi x
and
i
that (TN , TTN ) is the free C[M]-module generated by x1 x1 , . . . , xn xn .

(c) Dualizing, conclude that (TN , 1TN ) is the free C[M]-module generated by the TN dxn
1
invariant differentials dx
x1 , . . . , xn .
8.1.2. Consider the affine toric variety U = Spec(C[ M]).
(a) Prove that the map
d m C[ M]/C 7 m m M Z C[ M]
defines a C[ M]-module homomorphism.
(b) For a toric variety X , prove that these homomorphisms patch together to give the map
: 1X M Z OX in (8.1.3).
8.1.3. Let 0 F G H 0 be an exact sequence of sheaves on X and let f : X Y
be a morphism.
(a) Prove that 0 f F f G f H is exact on Y .
(b) Suppose that Y = {pt} and f : X Y is the obvious map. Use part (a) to give a new
proof of Proposition 6.0.8.
8.1.4. In the proof of Theorem 8.1.4, show that the map M Z OX j OD U0 factors as
M Z OX OD j OD U0 ,
where OD j OD U0 is injective.
8.2. Differential Forms on Toric Varieties
365
8.1.5. The proof of Theorem 8.1.6 contains the square

L
/ OD
M Z OX
L
OX
L
OD .
Describe the maps in this square carefully and prove that it commutes.
8.1.6. Show that the Euler sequence from Theorem 8.1.6 reduces to (8.1.8) when X = P n .
8.1.7. Sometimes the name Euler sequence is used to refer to an exact sequence for the
tangent sheaf TX of a smooth toric variety.
(a) Show that for P n , we have an exact sequence
0 OPn OPn (1)n+1 TPn 0.
Hint: Use (8.1.8).

(b) What is the corresponding sequence for a general smooth toric variety X for as in
Theorem 8.1.6?
8.1.8. Let f be a homogeneous polynomial of degree d in C[x1 , . . . , xn ]. The classical Euler
relation is the equation
n
X
f
= d f.
(8.1.9)
xi
xi
i=1
In this exercise, you will prove this relation and consider generalizations encoded by the
generalized Euler sequence from a toric variety.
(a) Prove (8.1.9). Hint: Differentiate the equation
f (tx1 , . . . ,txd ) = t d f (x1 , . . . , xn )
with respect to t.
(b) To see how the classical Euler relation generalizes, recall from Chapter 5 that given
a toric variety X with no torus factors (i.e., {u | (1)} spans NR ), we have the
total coordinate ring
S = C[x | (1)],
graded by Cl(X ). The graded pieces S for Cl(X ) consist of homogeneous
polynomials as described by (5.2.1) from Chapter 5. If HomZ (Cl(X ), Z) and
f S show that we have a generalized Euler relation
X
f
([D ]) x
= () f .
x
(1)
Hint: Follow what you did for part a, which is the case X = P n1 .
(c) When Cl(X ) has rank greater than 1, there will be several distinct generalized Euler
relations on homogeneous elements of S. Compute the Euler relations on X = P 1 P 1 .

For a toric variety X , we have the sheaf of p-forms Xp and the sheaf of Zariski
b n , n = dim X .
b p . By 8.0, the canonical sheaf of X is X =
p-forms
X
X
366
Properties of Wedge Products. We will need the following properties of wedge

products of free R-modules.
Proposition 8.2.1. Let F, G, H be free R-modules of finite rank.
(a) An R-module homomorphism : F G induces a homomorphism
Vp
Vp
Vp
G.
(b) Let 0 F G H 0 be an exact sequence with rank F = m and rank H =

n. Then rank G = m + n and there is a natural isomorphism
Vm+n
Vm
Vn
F R
H.
Proof.
Part (a) is straightforward (see Exercise 8.2.1 for an explicit description
V
of p ), as is the rank assertion in V
part (b). VFor theVisomorphism of part (b),
we assume n > 0 and define a map m F R n H m+n G as follows. If the
maps
: G H, then one checks that
Vm in
Vmthe exact
Vm sequence are : F
Vn G
Vnand V
: VF
G is V
injective and V
: G n HVis surjective. Then
Vm
F R n H maps to m () m+n G, where n = . This map is welldefined and gives the desired isomorphism (Exercise 8.2.1).

A corollary of this proposition is that if 0 F G H 0 is an exact
sequence of locally free sheaves on a variety X with rank F = m and rank H = n,
then rank G = m + n and there is a natural isomorphism
Vm+n
(8.2.1)
Vm
F OX
Vn
H.
Example 8.2.2. Suppose that Y X is a smooth subvariety of a smooth variety,

and let n = dim X , m = dim Y . Then we have the exact sequence
0 IY /IY2 1X OX OY Y1 0
from Theorem 8.0.18, where IY OX is the ideal sheaf of Y . By (8.2.1), we obtain

Vn
(1X OX OY )
V
Vnm
(IY /IY2 ) OY
Vm
Y1 .
n
One
can check that
(1 O ) ( n 1X ) OX OY . Now recall that X =
Vn 1
Vm 1 X OX Y
X and Y =
Y and that the normal sheaf of Y X is NY /X = (IY /IY2 ) .
Hence the above isomorphism implies
Y X OX
Vnm
NY /X .
This isomorphism is called the adjunction formula.
The Canonical Sheaf of a Toric Variety. Our first major result gives a formula for
the canonical sheaf of a toric variety.
Theorem 8.2.3. For a toric variety X , the canonical sheaf X is given by

P
X OX D .
P
Thus KX = D is a torus-invariant canonical divisor on X .
367
Proof. We first assume that X is smooth with no torus factors. Then we have the
Euler sequence
L
0 1X OX (D ) Pic(X ) Z OX 0
from Theorem 8.1.6. Each OX (D ) is a line bundle since X is smooth, and if

. Hence we
we set r = |(1)|, then one sees easily that Pic(X ) Z OX OXrn
can apply part (b) of Proposition 8.2.1 to obtain

Vn 1
Vr L
V
(8.2.2)
X OX rn OXrn
O
(D
)
.
X
It follows by induction from Proposition 8.2.1 that the right-hand side of (8.2.2) is
isomorphic to

P
N
D .
OX (D ) OX
Turning to the left-hand side of (8.2.2), note that

left-hand side is isomorphic to
Vn
Vrn
OXrn
OX , so that the
1X = Xn = X
since X is smooth. This proves the result when X is smooth without torus factors.
In Exercise 8.2.2 you will deduce the result for an arbitrary smooth toric variety.
Now suppose that X is normal
but not necessarily smooth. Let j : U0 X
S
be the inclusion map for U0 = U . We saw in the proof of Theorem 8.1.4 that
U0 is a smooth
toric variety satsifying codim(X \U0 ) 2. Now consider X and
P
OX ( D ). Since the fans for U0 and X have the same 1-dimensional cones,
these sheaves become isomorphic over U0 by the smooth case. Since these
P sheaves
are reflexive and codim(X \ U0 ) 2, we conclude that X OX ( D ) by
Proposition 8.0.1.

Here are some examples.
Example 8.2.4. Theorem 8.2.3 implies that the canonical bundle of P n is
Pn OPn (n 1)
for all n 1 since Cl(P n ) Z and D0 D1 Dn . In Exercise 8.2.3, you will
see another way to understand and derive this isomorphism.
Example 8.2.5. The previous example shows that P2 OP2 (3). We will compute this directly using
P2 = P2 2 =
V2
1P2
and the description of 1P2 as a rank 2 vector bundle given in Example 8.0.16.
Recall that the transition functions for this bundle are given by:

2 2
y/x 2 1/x
1/y x/y 2
y /x
0
.
, 10 =
20 =
, 12 =
1/x 2 0
y/x 2 y/x
0 1/y 2
368
By Exercise 8.2.1, the corresponding maps on

these 2 2 matrices:
V2
20 =
V2
are given by the determinants of
y 3 V2
1
1 V2
,
=
,
10 = 3 .
12
3
3
y
x
x
Note that each is a cube. It is also evident that

V2
10 =
V2
12
V2
20 ,
so that these give the transition functions for a line bundle on P 2 .

V
On the other hand, 2 1P2 OP2 (3) says the canonical bundle of P 2 is
the third tensor power of the tautological bundle described in Examples 6.0.19
and 6.0.21. To see this directly, we first need to calibrate the coordinate systems.
Example 8.0.16 used coordinates x, y from U0 , and Example 6.0.19 used homogeneous coordinates x0 , x1 , x2 for P 2 , with the standard open cover Ui = P 2 \ V(xi ).
Letting x = x0 /x2 and y = x1 /x2 gives an isomorphism U0 = U2 . Translating

coordinates for U1 , we have

1/x, y/x = 1/(x0 /x2 ), (x1 /x2 )/(x0 /x2 ) = x2 /x0 , x1 /x0 ,
hence U1 = U0 . A similar computation shows U2 = U1 . We are now set for the

final calculation. Keep in mind that the i j are in the coordinate system with charts
Ui . We will use i j to denote the same transition function, but using the Ui charts.
Thus we have
V2
= 1/y 3 = (x2 /x1 )3 = 12
V2 20
= y 3 /x 3 = (x1 /x0 )3 = 01
V2 12
10 = 1/x 3 = (x2 /x0 )3 = 02 .
Up to a sign, these are indeed the cubes of the transition functions that we computed
for the tautological bundle OPn (1) in Example 6.0.19. In Exercise 8.2.4, you will
work through the definition of the canonical bundle directly to find the transition
functions given in Example 8.0.16.
Example 8.2.6. When we computed the class group of the Hirzebruch surface Hr
in Example 4.1.8, we wrote the divisors D as D1 , D2 , D3 , D4 and showed that
D3 D1
D4 rD1 + D2 .
Thus Cl(Hr ) = Pic(Hr ) Z2 is freely generated by the classes of D1 and D2 . It
follows that the canonical bundle can be written
Hr OHr (D1 D2 D3 D4 ) OHr ((r + 2)D1 2D2 )
by Theorem 8.2.3.
369
The Canonical Module. For a toric variety X without torus factors, the canonical
sheaf X comes from a graded S-module, where S = C[x | (1)] is the total
coordinate ring of X . This module is easy to describe explicitly.
Each variable x S has degree deg(x ) = [D ] Cl(X ). Define

Q P
0 = deg x =
D Cl(X ).
Then S(0 ) is the graded S-module where S(0 ) = S0 for Cl(X ). As

in 5.3, the coherent sheaf associated to S(0 ) is denoted OX (0 ). We have
the following result.
Proposition 8.2.7. OX (0 ) X .
Proof. According to Proposition 5.3.7, OX (0 ) OX (D) for any WeilP
divisor
with 0 = [D] Cl(X ). The definition of 0 allows us to pick D = D .
Then Theorem 8.2.3 implies

P
OX (0 ) OX D X .

We call S(0 ) the canonical module of S.
Corollary 8.2.8. For any normal toric variety X we have an exact sequence
L
0 X OX OD .
Q
Proof. First suppose that X has no torus factors. Multiplication by x induces
an exact sequence of graded S-modules
L
0 S(0 ) S S/hx i
Q
since 0 = deg( x ). The sheaf associated to S/hx i is OD (Exercise 8.2.5), and
then we are done by Example 6.0.10 and Proposition 8.2.7. When X has a torus
factor, the result follows using the strategy described in Exercise 8.2.2.

We can also describe X in terms of n-forms in the x . Fix a basis e1 , . . . , en of
M. For each n-element subset I = {1 , . . . , n } (1), we get the nn determinant
det(uI ) = det(hei , u j i).
This depends on the ordering of the i , as does the n-form dx1 dxn , though
the product
det(uI ) dx1 dxn
depends only on e1 , . . . , en . It follows that the n-form
X

Q
(8.2.3)
0 =
det(uI )
I
/ x dx1 dxn
|I|=n
is well-defined up to 1. If we set deg(dx ) = deg(x ), then 0

homogeneous of degree 0 . This gives the submodule
S 0
Vn
S/C ,
Vn
S/C is
370
which is isomorphic to S(0 ) since deg(0 ) = 0 .

To see where 0 comes from, let L = C(x | (1)) be the field of fractions
of S. Then L is the function field of C(1) , and the surjective toric morphism
: C(1) \ Z() X
from Proposition 5.1.9 induces an injection on function fields
: C(X ) L.
Furthermore, the proof of Theorem 5.1.11 implies that for any m M, the character
m C(X ) maps to
Q hm,u i
( m ) = x .
Q hm,u i
We regard C(X ) as a subfield of L via , so that C(X ) L and m = x .
This induces an inclusion of Kahler n-forms
(8.2.4)
Vn
C(X )/C
Vn
L/C .
A basis e1 , . . . , en of M gives coordinates ti = ei for the torus TN , and we know

from 8.1 that dtti i is a TN -invariant section of 1TN . Then
dt1
dtn

t1
tn
is a TN -invariant section of TnN which pulls back to a rational n-form in n L/C

via (8.2.4). Note that
dx
dti X
=
hei , u i
ti
x
Q
Q he ,u i
since ti = x i . When we mutiply by x to clear denominators, we obtain
X
Q
dx
dx
dt1
dtn Q X
x
he
,
u
i
he
,
u
i
=
x
t1
tn
x
x
X

Q
=
det(uI )
I
/ x dx1 dxn = 0 .
|I|=n
Hence 0 arises in a completely natural way.
This can be interpreted in terms of sheaves as follows. We regard (8.2.4) as an

inclusion of constant sheaves on X . Then the inclusion
X
Vn
C(X )/C
from Proposition 8.0.21 induces an inclusion

X
Vn
L/C .
Vn
L/C .
On the other hand, it is easy to see that S 0

Sg
0
Vn
S/C induces an inclusion
371
Using the above derivation of 0 , one can prove that

Sg
0 = X
as subsheaves of the constant sheaf n L/C see [184, Prop. 14.14]. Note that this
is an equality of sheaves, not just an isomorphism. This shows that from the point
of view of differential forms, S 0 deserves to be called the canonical module. It is
isomorphic to the earlier version S(0 ) of the canonical module via the map that
takes 0 S 0 to 1 S(0 ).
The Affine Case. The canonical sheaf U of a normal affine toric variety U is
determined by its module of global
Psections, the canonical module. When we think
of U as the ideal sheaf OU ( D ), we get the ideal
(U , U ) (U , OU ) = C[ M].
Proposition 8.2.9. (U , U)C[ M] is the ideal generated by the characters

m for all m M in the interior of .
Proof. Corollary 8.2.8 gives the exact sequence
0 (U , U ) C[ M]
C[
M]/I ,
where I = I(D U ) is the ideal of D U U . Since

M
M
C m
I =
C m = C[ M]
m M,hm,u i>0
m M
by (8.1.4), it follows that (U , X ) is the direct sum of C m for all m M such

that hm, u i > 0 for all . Since the u generate , this is equivalent to saying that
m is in the interior of .

The Projective Case. A full dimensional lattice polytope P MR gives two interesting graded rings, each with its own canonical module. For the first, we use an
auxiliary variable t so that for every k N, a lattice point m (k P) M gives a
character mt k on TN C . These characters span the semigroup algebra
M
(8.2.5)
SP = C[C(P) (M Z)] =
C mt k ,
m(k P)M
where C(P) = Cone(P {1}) MR R is the cone of P. Recall that the slice
of C(P) at height k is k P (see Figure 4 is 2.2). The ring SP is graded by setting
deg( mt k ) = k. Then Proposition 8.2.9 tells us that the canonical sheaf of the
associated affine toric variety comes from the ideal
M
IP =
C mt k SP .
mInt(k P)M
This is the canonical module of SP .
372
The second ring associated to P uses the total coordinate ring S of the projective
toric variety XP and the corresponding ample divisor DP . Let = [DP ] Cl(XP )
be the divisor class of DP . The graded pieces Sk S, k N, form the graded ring
S =
Sk .
k=0
Q
The canonical module of S is S(0 ), which is isomorphic to the ideal
x S
Q
Q
via multiplication by x since 0 = deg
x ). When restricted to S , this
gives the ideal
M
I =
Ik S ,
k=0
where Ik Sk is generated by all monomials of degree k in which every variable

appears to a positive power.
It follows that the polytope P gives graded rings SP and S , each of which has
an ideal representing the canonical module. These are related as follows.
Theorem 8.2.10. The graded rings SP and S are naturally isomorphic via an
isomorphism that takes IP to I .
Proof. The proof of Theorem 5.4.8 used homogenization to construct an isomorphism SP S . It is straightforward to see that this isomorphism carries the ideal
IP SP to the ideal I S (Exercise 8.2.6).

Recall from Theorem 7.1.13 that XP Proj(SP ) via the Proj construction from
7.0. Furthemore, graded SP -modules give quasicoherent sheaves on Proj(SP ), as
described in [131, II.5] (this is similar to the construction given in 5.3). Then the
following is true (Exercise 8.2.7).
Proposition 8.2.11. The sheaf on XP associated to the ideal IP SP by the Proj
construction is the canonical sheaf of XP .

The situation becomes even nicer when P is a normal polytope. The ample
divisor DP is very ample and hence embeds XP into a projective space, which gives
the homogeneous coordinate ring C[XP ]. As we learned in 2.0, C[XP ] is also the
ordinary coordinate ring of its affine cone XbP , i.e.,
C[XP ] = C[XbP ].
Furthermore, since P is normal, Theorem 5.4.8 gives isomorphisms

SP S C[XP ]
and implies that XbP is the normal affine toric variety given by
XbP = Spec(SP ) = Spec(S ).
373
It follows that the canonical sheaf of the affine cone of XP comes from the ideal
IP SP , and when we use the grading on SP and IP , they give XP and its canonical
sheaf via XP Proj(SP ). Everything fits together very nicely.
There is a more general L

notion of canonical module that applies to any graded
Cohen-Macaulay ring S =
k=0 Sk where S0 is a fieldsee [56, Sec. 3.6]. The
canonical modules of SP S constructed above are canonical in this sense.
When thePCanonical Divisor is Cartier. For a toric variety X , the Weil divisor
KX = D is called the canonical divisor. Note that KX need not be a Cartier
divisor if X is not smooth. In fact, from Theorem 4.2.8, we have the following
characterization of the cases when the canonical divisor is Cartier.
Proposition 8.2.12. Let X be a normal toric variety. Then KX is Cartier if and
only if for each maximal cone , there exists m M such that
hm , u i = 1 for all (1).
Similarly, K is Q-Cartier if and only if for each maximal , there exists

m MQ such that hm , u i = 1 for all (1).
Example 8.2.13. Let = Cone(de1 e2 , e2 ) R2 . The affine toric variety U
is the rational normal cone Cbd . We computed Cl(Cbd ) Z/dZ in Example 4.1.4,
where the Weil divisors D1 , D2 coming from the rays satisfy D2 D1 , dD1 0.
The canonical divisor KCbd = D1 D2 has divisor class corresponding to

[2] Z/dZ. Since the Picard group of a normal affine toric variety is trivial
(Proposition 4.2.2), it follows that KCbd is Cartier if and only if d 2.
Another way to see this is via Proposition 8.2.12, where one easily computes
that m = (2/d)e1 + e2 satisfies hm , de1 e2 i = hm , e2 i = 1. This lies in M = Z2
if and only if d 2.
We will use the following terminology.

Definition 8.2.14. Let KX be a canonical divisor on a normal variety X . We say
that X is Gorenstein if KX is a Cartier divisor.
Since the canonical sheaf X = OX (KX ) is reflexive, we have
X is Gorenstein KX is Cartier X is a line bundle
by Proposition 8.0.7. All smooth varieties are Gorenstein, of course. We will study
further examples of singular Gorenstein varieties in the next section of the chapter.
Refinements. A refinement of a fan induces a proper birational toric morphism : X X . The canonical sheaves of X and X are related as follows.
374
Theorem 8.2.15. Let : X X be the toric morphism induced by a refinement

of a fan in NR Rn . Then
X X .
In particular, if is a smooth refinement of , then
Xn X .
P
Proof. First assume X = U , so that refines . Since K = (1) D ,
the description of global sections given in Proposition 4.3.3 implies that
M
C m,
(X , X ) =
mP M
where
P = {m MR | hm, u i 1 for all (1)}.
We clearly have P M Int( ) M since (1) (1). The opposite inclusion
also holds, as we now prove. Given m Int( ) M, we have hm, ui > 0 for all
u 6= 0 in . In particular, hm, u i > 0 for all (1). This is an integer, so that
hm, u i 1, which implies m P M, as desired. Since
M
C m
(U , U ) =
mInt( )M
by Proposition 8.2.9, we conclude that (X , X ) = (U , U ). Then we have

X U since U is affine.
In the general case, X is covered by affine open subsets U for , and

is the toric variety of the refinement of induced by . The above
paragraph gives isomorphisms
1 (U )
X |U X |U
which are compatible with the inclusion U U when is a face of . Hence

these isomorphisms patch to give the desired isomorphism X X .
In Chapter 11, we will prove the existence of smooth refinements. It follows
that for any n-dimensional toric variety X , there are two ways of constructing the
canonical sheaf X from the sheaf of n-forms of a smooth toric variety:
(Internal) X = j Un 0 , where j : U0 X is the inclusion of the smooth locus
of X .
(External) X = Xn , where : X X is the toric morphism coming
from a smooth refinement of .
375
b p for 1 < p < n = dim (X ) are also important

Sheaves of p-forms. The sheaves
X
for the geometry of X . We construct a sequence
p Vp
p L V
b p
M Z OX p1 ( M) Z OD
(8.2.6)
0
X
as follows. The map : 1X M Z OX from (8.1.3) induces
Vp
: Xp =
V p
Vp
1X
Vp
M Z OX ,
. To define p , recall that for u N, the contraction map

and then p =
Vp
V p1
iu : M
M has the property that
iu (m1 m p ) =
(8.2.7)
p
X
(1)i1 hmi , ui m1 m
bi m p
i=1
when m1 , . . . , m p M. Note also that im(iu )
V p1
(u M) (Exercise 8.2.8).
An element (1) gives u as usual. Using iu :

OX OD , we obtain the composition
Vp
M Z OX
V p1
( M) Z OX
This gives a natural map

p :
Vp
M Z OX
L V p1
Vp
V p1
V p1
( M) and
( M) Z OD .
( M) Z OD .
Theorem 8.2.16. The sequence (8.2.6) is exact for any normal toric variety X .
Proof. We begin with the smooth case. Since exactness of a sheaf sequence is
local, we can work over an affine toric variety U = Cr (C )nr , where is
generated by the first r elements of a basis of e1 , . . . , en of N. By abuse of notation,
e1 , . . . , en will denote the corresponding dual basis of M. Setting xi = ei , we get
U = Spec(R),
1
, . . . , xn1 ].
R = C[x1 , . . . , xr , xr+1
Then (8.2.6) comes from the exact sequence of R-modules

0
Vp
R/C
Vp
M Z R
r
M
V p1
i=1
(e
i M) Z R/hxi i,
where p maps dxi1 dxi p to ei1 ei p xi1 xi p by the description of

given in (8.1.3).
V
It is thus obvious that p p = 0. To prove exactness, we regard p M Z R

as
F, where F is the V
free R module with basis e1 , . . . , en . Now suppose that
p () = 0 for some p F. We can write uniquely as
X
=
fi1 ...i p ei1 ei p .
Vp
i1 <<i p
Exactness will follow once we prove xi1 xi p divides fi1 ...i p for all i1 < < i p .
376
If an index i > r appears in fi1 ...i p , then xi automatically divides fi1 ...i p since xi
is invertible in R. Now suppose that an index i r appears in fi1 ...i p . We can write
= ei 1 + 2 , where ei does not appear in 2 and all fi1 ...i p involving the index
i appear in 1 . Since ie1 () = 1 , the only way for p () to vanish is for fi1 ...i p to
be zero in R/hxi i, i.e., for xi to divide fi1 ...i p . This completes the proof of exactness
in the smooth case.
The proof for the general case follows from the smooth case by an argument
similar to what we did in the proof of Theorem 8.1.4.

Note that when p = 1, the exact sequence (8.2.6) reduces to the sequence appearing in Theorem 8.1.4, and when p = n, we get the sequence in Corollary 8.2.8
bn .
since X =
X
When X has no torus factors, it is easy to find a graded S-module whose

b p . Adapting the definition of p in (8.2.6) to the module
associated sheaf is
X
case, we get a homomorphism
Vp
L V
M Z S p1 ( M) Z S/hx i
b p . This gives an exact sequence of

of graded S-modules whose kernel we denote
S
graded S-modules
V
V
b p p M Z S L p1 ( M) Z S/hx i.
0
Using Example 6.0.10, we obtain the following corollary of Theorem 8.2.16.
b p is the sheaf associated to

Corollary 8.2.17. If X has no torus factors, then
X
b p.
the graded S-module

S
b p is determined by its
The Affine Case. For an affine toric variety U , the sheaf
U
global sections, which can be described using Theorem 8.2.16. We will need the
following notation. If m M and , let
V (m) = SpanC (m0 M | m0 the minimal face of containing m) MC ,
where MC = M Z C. Here is a result due to Danilov [76, Prop. 4.3].

Proposition 8.2.18. Let U be the toric variety of the cone . Then we have an
isomorphism
M Vp
bp )
(U ,
V (m) m .
U
m M
Proof. Using the inclusion M M and the exact sequence of Theorem 8.2.16
over U , we obtain the exact sequence
b p )
0 (U ,
U
Vp
M Z (U , OU )
L V p1
M Z (U , OD )
377
L
of modules over (U , OX ) = C[ M] = m M C m . Also note that
M
(8.2.8)
(U , OD ) =
C m
m M,hm,u i=0
by the analysis given in the proof of Theorem 8.1.1. It follows that we have a direct
L
bp
bp )=
sum decomposition (U ,
m M (U , U )m , where
U
(8.2.9)
b p )m
0 (U ,
U
Vp
MC
hm,u i=0
V p1
MC
is exact for m M and p is the sum of the contraction maps iu for all
satisfying hm, u i = 0.
V
Thus p MC is in the kernel of p if and only if iu () = 0 for all with

hm, u i = 0. You will show in Exercise 8.2.8 that iu () = 0 if and only if
V p
( )C . It follows that the kernel of p in (8.2.9) is the intersection

\ Vp
V \
( )C .
(8.2.10)
( )C = p
hm,u i=0
hm,u i=0
However, the intersection

F=
hm,u i=0
is the minimal face of
containing m, and one sees easily that

T
SpanR (m0 M | m0 F) = hm,u i=0 .
T
It follows that V (m) = hm,u i=0 ( )C . This plus (8.2.10) imply that the kernel
V
of p in (8.2.9) is pV (m), as claimed.
The Simplicial Case. When X is simplicial, the sequence (8.2.6) is exact on the
right when p = 1 by part (b) of Theorem 8.1.4. For p > 1, similar though longer
exact sequences exist in the simplicial case, as we now describe without proof.
Given a fan , consider the sheaf on X defined by
M V p j
K j (, p) =
( M) Z OV () ,
( j)
where V () = O() X is the orbit closure corresponding to . Thus

K 0 (, p) =
Vp
M Z OX
M V p1
M V p1
K 1 (, p) =
( M) Z OV () =
( M) Z OD ,
(1)
(1)
and the exact sequence (8.2.6) can be written

b p K 0 (, p) K 1 (, p).
0
X
When X is simplicial, one can extend this exact sequence as follows.
378
Theorem 8.2.19. When X is simplicial, there is an exact sequence

b p K 0 (, p) K 1 (, p) K p (, p) 0.
0
X
A proof of this result can be found in [218, Sec. 3.2], where K (, p) is called
the pth Ishida complex. We will use this exact sequence in Chapter 9 to prove a
vanishing theorem for sheaf cohomology on simplicial toric varieties.
Exercises for 8.2.
8.2.1. Given a map of free R-modules : F G, we can pick bases of F, G and represent
by a n m matrix with entries in V
R, where m
V p= rank(F), n = rank(G). These bases
p
induce
bases
of
the
wedge
products
F
and
G. Then prove that the induced map
Vp Vp
V
:
F p G is given by the p p minors of (with appropriate signs).
8.2.2. Complete the proof of Theorem 8.2.3 in the smooth case by showing how to reduce
to the case when X is smooth with no torus factors. Hint: First prove that any smooth
toric variety is equivariantly isomorphic to a product X (C )r , where X is a smooth
toric variety with no torus factors. Then consider X Cr .
8.2.3. Let TN be the torus of P n . If x0 , . . . , xn are the usual homogeneous coordinates, then
yi = xi /x0 for i = 1, . . . , n are affine coordinates for the open subset where x0 6= 0. The
Vn 1
dyn
n
1
TN ) and has poles of
differential n-form = dy
y1 yn spans (TN , TN ) = (TN ,
order 1 along each Di = V(xi ) for i = 1, . . . , n. Show that if we write z j = x j /xi for the
affine coordinates on the complement of V(xi ), and change coordinates to the z j , j 6= i,
n
then we can
Pnsee also has a pole of order 1 along V(x0 ). Hence on P , defines a section
of OPn ( i=0 Di ).
8.2.4. Using the open subsets Ui P 2 (see Figure 2 in Example 3.1.9), compute 1P2 (Ui ),
and write down the transition functions on Ui U j . Compare to Example 8.0.16. If the
result of your computation differs, describe how you can explain this via a change of basis.
8.2.5. Let S be the total coordinate ring of a toric variety X without torus factors. Prove
that OD is the sheaf associated to the graded S-module S/hx i. Hint: Consider the exact
sequence 0 hx i S S/hx i 0 and apply Proposition 5.3.7 and Example 6.0.10.
8.2.7. Prove that the sheaf associated to the ideal IP SP from Theorem 8.2.10 is the
canonical sheaf of XP = Proj(SP ). Hint: Let S be the total coordinate ring of XP . We saw
in 5.3 that a graded S-module such as S(0 ) gives a sheaf on XP . Compare this to how
I gives a sheaf on Proj(S ). See the proof of Theorem 7.1.13.
8.2.8. Let F be a free module of finite rank over a domain R. Given u F , we get the
Vp
V p1
kernel u F and the contraction map iu :
F
F. Assume u = re1 where r R
and e1 , . . . , en is a basis of F .
Vp1
u .
(a) Prove that the image of iu is contained in
Vp
Vp
(b) Given
F, prove that iu () = 0 if and only if
u .
b p () to be the sheaf
8.2.9. Assume that X has no torus factors. For Cl(X ), define
X
p
p
b (). Prove that (X ,
b ()) (
b p ) . Hint: If
associated to the graded S-module
X
S
S
you get stuck, see [19, Prop. 8.5].
8.3. Fano Toric Varieties
379
8.2.10. Compute the n-form 0 defined in (8.2.3) for P n , P(q0 , . . . , qn ), and P1 P1 .

8.2.11. Prove that our favorite affine toric variety V = V(xy zw) C4 is Gorenstein.
Hint: Example 1.2.20.
8.2.12. Prove that a product of two Gorenstein toric varieties is again Gorenstein. Hint:
Use Propositions 3.1.14 and 8.2.12.
8.2.13. We will see in Chapter 10 that every 2-dimensional rational cone in R2 is lattice
equivalent to a cone of the form = Cone(e2 , de1 ke2) for integers d > 0 and 0 k < d
with gcd(d, k) = 1. Prove that U is Gorenstein if and only if d = k + 1.
8.2.14. A strongly convex rational polyhedral cone is Gorenstein if its associated affine
toric variety is Gorenstein. In the discussion of the algebra (8.2.5), we saw that a full
dimensional lattice polytope P MR gives the cone
(a) Prove that C(P) is Gorenstein.

(b) Prove that the dual cone C(P) NR R is Gorenstein if and only if P is reflexive.
In general, a cone is reflexive Gorenstein if both the cone and its dual are Gorenstein.
Reflexive Gorenstein cones play an important role in mirror symmetrysee [18, 204].
L
8.2.15. Explain why S0 =
k=0 Sk0 gives another model for the canonical module
of the graded ring S discussed in the text.

We finish this chapter with a discussion of an interesting class of projective toric
varieties and their corresponding polytopes.
Definition 8.3.1. A complete normal variety X is said to be a Gorenstein Fano
variety if the anticanonical divisor KX is Cartier and ample.
Thus Gorenstein Fano varieties are projective. When X is smooth, we will
simply say that X is Fano.
Example 8.3.2. Example 8.2.5 shows that OP2 (KP2 ) OP2 (3) is ample, so P 2 is
a Fano variety. We continue this example to introduce the key ideas that will lead
to a classification of 2-dimensional Gorenstein Fano toric varieties.
The standard fan for P 2 = X has minimal generators u0 = e1 e2 , u1 = e1
and u2 = e2 . The polytope corresponding to the anticanonical divisor of P 2 is
P = {m R2 | hm, ui i 1, i = 0, 1, 2}.
In Exercise 8.3.1 you will check that
P = Conv(e1 e2 , 2e1 e2 , e1 + 2e2 ).
This lattice polygon is shown in Figure 1 on the next page. The open circle in the
figure represents the origin, which is the unique interior lattice point of P. Also,
380
t
@
t @t
@
t d @t
@
t t t @t
t
@
d @
t

t

Figure 1. The anticanonical polytope P and its dual P for P 2
by Exercise 8.3.1, the dual polytope P = {u NR | hm, ui 1 for all m P} is

given by
P = Conv(e1 , e2 , e1 e2 ),
which is the polytope generated by the ray generators of the fan of P 2 .
Example 8.3.3. In Exercise 8.3.2, you will generalize Example 8.3.2 by showing
that the weighted projective space P(q0 , . . . , qn ) is Gorenstein Fano if and only if
qi |q0 + + qn for all 0 i n.
The special features of Example 8.3.2 involve an unexpected relation between

Fano toric varieties and the reflexive polytopes introduced in Chapter 2.
Fano Toric Varieties and Reflexive Polytopes. Recall from Definition 2.3.12 that
a lattice polytope in MR is reflexive if its facet presentation is
(8.3.1)
P = {m MR | hm, uF i 1 for all facets F}.
It follows that if P is reflexive, the origin is the unique interior lattice point of P
(Exercise 2.3.5). Since aF = 1 for all facets F, the dual polytope is
(8.3.2)
P = Conv(uF | F is a facet of P)
(Exercise 2.2.1). Finally, P is a lattice polytope and is reflexive (Exercise 2.3.5).

The polytopes pictured in Figure 1 are reflexive.
The following result gives the connection between projective Gorenstein Fano
toric varieties and reflexive polytopes generalizing what we saw above for P 2 .
Theorem 8.3.4. Let X be a normal toric variety. If X is a projective Gorenstein
Fano
P variety, then the polytope associated to the anticanonical divisor KX =
D is reflexive. Conversely, if XP is the projective toric variety associated to a
reflexive polytope P, then XP is a Gorenstein Fano variety.
Proof. If X is Gorenstein Fano, then
P KX is Cartier and ample. This implies
that polytope associated to KX = D has facet presentation
P = {m MR | hm, u i 1 for all (1)}.
381
and hence is reflexive by (8.3.1). Conversely, let P be a reflexive polytope in MR .

The normal fan P of P defines the variety XP = XP . The facet presentation (8.3.1)
of P has aFP
= 1 for every facet F of P. Hence the Cartier divisor corresponding to
P is DP = F DF = KXP . We proved that DP is ample in Proposition 6.1.10, so
that KXP is ample. Hence XP is Gorenstein Fano.

Classification. By Theorem 8.3.4, classifying toric Gorenstein Fano varieties is
equivalent to classifying reflexive polytopes P in MR . Since reflexive polytopes
contain the origin as an interior point, classify means up to invertible linear maps
of MR induced by isomorphisms of M. This is called lattice equivalence. The
first interesting case is in dimension two, where toric Gorenstein Fano surfaces
correspond to 16 equivalence classes of reflexive polygons.
We begin with some general facts about a reflexive polytope P. First note that
the lattice points of P are the origin and the lattice points lying on the boundary.
Furthermore, any boundary lattice point is primitive. We will also need two less
obvious results from [214].
The first result concerns projections of reflexive polytopes. Given a reflexive
polygon P and a primitive element m M, there is a projection map
m : MR MR /Rm
whose image is a polytope whose vertices lie in M/Zm.
Lemma 8.3.5. Let P be a reflexive polytope in MR Rn and let m be a lattice point
m in the boundary of P. Then m (P) is a lattice polytope in MR /Rm containing the
origin as an interior lattice point, and
S

m (P) = m mF facet of P F .
Proof. For each p m (P), m1 (p) is a line parallel to the line Rm. By taking the
point in P that maps to p and is farthest along this line in the direction of m, we see
that the points in m (P) are in bijective correspondence with the points in
U = {x P | x + m
/ P for all > 0}.
If F is a facet of P containing m, one easily sees that F U . Conversely, let x U .
One can show without difficulty there exists some facet F of P, not parallel to m,
containing x and such that huF , mi < 0. Since P is reflexive and m S
M, huF , mi is
an integer 1. Hence huF , mi = 1, so that x, m F. Thus U mF F, and
from here the lemma follows easily.

The second result is the following fact about pairs of lattice points on the
boundary of a reflexive polytope. The vertices are primitive vectors in M, but
we can have other lattice points on the boundary of P as well.
382
Lemma 8.3.6. Let m, m be distinct lattice points on the boundary of a reflexive

polytope P. Then exactly one of the following holds:
(a) m and m lie in a common edge of P,
(b) m + m = 0, or
(c) m + m is also on the boundary of P.
The proof is left to the reader as Exercise 8.3.3.
The 2-Dimensional Case. The following theorem classifies reflexive polygons in
the plane MR R2 , up to lattice equivalence. Figure 2 shows 16 lattice polygons,
where the open circle in the center of each polygon is the origin and the solid circles
are the lattice points on the boundary. The numbers in the labels give the number
of boundary lattice points. Polygons 3 and 9 are the dual pair from Example 8.3.2.
t
@
d @
t

t

t
@
d @t
t
@
@t
4a
3
t
t
@
d @t
t
@
@t
t d
@

t
@
5a
t
t
A
A
d t

t
t
6c
t
t d
6d
t
6b
t
7a
8b
t d
@
@t

d

t

t

t 8c

d
A
At
t
6a
4c
t
@
d @t
t
@
@t
8a
A
At
t
A
A
4b
5b
t
t
@
d @t

d

t
7b
t
t 9
@
t @t
@
t d @t
@
t t t @t
Figure 2. The 16 equivalence classes of reflexive lattice polygons in R2
383
Theorem 8.3.7. There are exactly 16 equivalence classes of reflexive polygons in

the plane, shown in Figure 2.
Proof. We will sketch the proof following [215, Proposition 4.1], and leave the
details for the reader to verify (Exercise 8.3.4). We consider several cases.
Case A. First assume that P is a reflexive polygon such that each edge contains
exactly two lattice points, and fix one such pair. These lattice points must form a
basis for M since the triangle formed by these two vertices and the origin has lattice
points only at the vertices (part (a) of Exercise 8.3.4). Hence we can use a lattice
equivalence to place the two of them at e1 and e2 .
Subcase A.1. If P has exactly three vertices, the third is located at ae1 + be2
for some a, b Z. Since 0 is the only lattice point in the interior, we must have
a = b = 1 (part (b) of Exercise 8.3.4). This gives the polygon of type 3.
Subcase A.2. Next, still in Case A, assume that there are three distinct vertices
m, m , m such that m + m = m . There must be edges of P containing the pairs
m , m and m , m (part (c) of Exercise 8.3.4). Hence each pair must yield a basis
for M and we can place m = e1 , m2 = e1 + e2 , and then m = e2 . Project P from
m = e2 using Lemma 8.3.5. It follows that P can only contain points e1 , 0, e1 on
the line y = 0 and e2 , e1 e2 on the line y = 1. Moreover, P cannot contain any
points with y 2. It follows that the only other possible polygons in this case are
4b, 5a, 6a (part (d) of Exercise 8.3.4) up to lattice equivalence.
Subcase A.3. Finally, in Case A, if we are not in subcases A.1 or A.2, then by
Lemma 8.3.6, if m and m are not in the same edge, we must have m + m = 0. The
only possibility here is clearly polygon 4a.
Case B. Assume there are exactly three lattice points on some edge of P. By an
isomorphism of M, we can place these at e1 + e2 , e2 , e1 + e2 . Projecting from m =
e2 and using Lemma 8.3.5, we see that P must be contained in the strip 1 x 1.
Moreover, P cannot contain any points with y < 3, or else e2 6= 0 would be an
interior lattice point. Up to lattice equivalence, the possibilities are 4c, 5b, 6b, 6c,
6d, 7a, 7b, 8a, 8b, and 8c (part (e) of Exercise 8.3.4).
Case C. Assume no edge of P has exactly three lattice points and some edge has
four or more. Place the vertex of this edge at e1 e2 so that e1 , e1 + e2 and
e1 + 2e2 also lie on this edge. Projecting from e1 via Lemma 8.3.5 shows that
P lies above the line y = 1, and since the origin is an interior point, there must be
a lattice point m = ae1 + be2 with a > 0 and b 1.
Subcase C.1 If b = 1, then 2e1 e2 must lie in P. If e1 + 2e2 and 2e1 e2
do not lie on an edge of P, their sum e1 + e2 would lie in P by Lemma 8.3.6 and
force e1 , e2 to be interior points. This is impossible, so e1 + 2e2 and 2e1 e2 lie
on an edge of P. Hence P has type 9.
Subcase C.2 If b = 0, then m = e1 , for otherwise e1 would be an interior point.
Applying Lemma 8.3.6 to e1 and e1 + e2 shows that e2 P. If no edge connects
384
e1 , e2 , then Lemma 8.3.6 would imply e1 + e2 P. This point and e1 + 2e2 would
force e2 to be interior, again impossible. Hence e1 , e2 lie on an edge, which forces P
to be contained in the polygon of type 9. Then our hypotheses on P force equality.
Subcase C.3 If b 1, then e2 becomes an interior point of P (part (f) of Exercise 8.3.4). Hence this subcase cannot occur.

There are many other proofs of the classification given in Theorem 8.3.7, including [74, 214, 230]. References to further proofs are given in [74, Thm. 6.10].
The collection of 2-dimensional reflexive polygons also exhibits some very
interesting symmetries and regularities. For instance, we know that if P is reflexive,
then its dual is also reflexive. In Exercise 8.3.5 you will determine which are the
dual pairs in the list above. There is also a very interesting relation between the
numbers of boundary lattice points in P and P : in all cases, we have
(8.3.3)
|P M| + |P N| = 12.
We will see in Chapter 10 that there is an explanation for this coincidence coming
from the cohomology theory of sheaves on surfaces.
Higher Dimensions. Reflexive polytopes of dimension greater than two and their
associated toric varieties are being actively studied. See for instance [215] and the
references therein. One reason for the interest in these varieties is the relation with
mirror symmetry. It is known that there are a finite number of equivalence classes
of reflexive polytopes in all dimensions. Using their computer program PALP,
Kreuzer and Skarke [183] determined that there are 4319 classes of 3-dimensional
reflexive polytopes and 473800776 classes of 4-dimensional reflexive polytopes.
Since these numbers grow so quickly, most more recent work has focused on subclasses, for instance the polytopes giving smooth Fano toric varieties. These polytopes have a deceptively simple description (Exercise 8.3.6). There are 5 types of
such polytopes in dimension 2 (Exercise 8.3.7), and Batyrev [17] and Sato [244]
have shown that there are 18 types in dimension 3 and 124 types in dimension 4.
See [216] for a classification algorithm and further references.
Exercises for 8.3.
8.3.1. Verify the claims about the polygons P and P defined in Example 8.3.2.
8.3.2. Prove the claim about P(q0 , . . . , qn ) made in Example 8.3.3. Hint: Exercise 4.2.11.
8.3.3. Prove Lemma 8.3.6. Hint: Assume that (a) and (b) do not hold. Show that for any
facet F of P, huF , mi + huF , m i > 2, and use this to conclude that (c) must hold.
8.3.4. In this exercise, you will supply the details for the proof of Theorem 8.3.7.
(a) Let m, m M Z2 and assume that Conv(0, m, m ) has no lattice points other than
the vertices. Prove that m, m form a basis of M.
385
(b) Show that if P is a reflexive triangle with vertices at e1 , e2 , then the third vertex must
be e1 e2 . Hint: One way to show this succinctly is to use the projections from the
vertices e1 and e2 as in Lemma 8.3.5.
(c) In the case that each facet contains exactly two vertices, show that if there are vertices
m, m , m with m + m = m , then the pairs m, m and m , m must lie in edges.
(d) Complete the proof that the polygons of types 4b, 5a, 6a are the only possibilities in
Subcase A.2. Hint: Show that Conv(e2 , (e2 e1 ), e1 ) is lattice equivalent to 5a.
(e) Show that every reflexive polygon in Case B is equivalent to one of type 4c, 5b, 6b,
6c, 6d, 7a, 7b, 8a, 8b, or 8c.
(f) Prove the claim made in Subcase C.3.
8.3.5. Consider the 16 reflexive polygons in Figure 2. Since each polygon in the figure is
reflexive, its dual must also appear in the figure, up to lattice equivalence.
(a) For each polygon in the Figure 2, determine its dual polygon and where the dual fits
in the classification. In some cases, you will need to find an isomorphism of M that
takes the dual to a polygon in the figure. Also, some of the polygon are self-dual, up
to lattice equivalence.
(b) Show that the relation (8.3.3) holds for each polar pair.
8.3.6. A lattice polytope is called Fano if the origin is an interior point and the vertices of
every facet form a basis of the lattice. Let Q NR be a Fano polytope.
(a) Prove that Q is reflexive.

(b) Part (a) implies that Q MR is a lattice polytope. Prove that XQ is a smooth Fano
toric variety whose normal fan is formed by taking cones over proper faces of Q.
(c) Prove that every smooth Fano toric variety arises this way.
8.3.7. Of the 16 Gorenstein Fano varieties classified in Theorem 8.3.7, exactly five are
smooth. Find them.
8.3.8. Some of the polygons in Figure 2 give well-known toric varieties. For example,
type 9 gives P 2 and type 8a gives P 1 P 1 . This follows easily by computing the normal
fan. Here you will describe the toric surfaces coming from some of the other polygons in
Figure 2. This exercise is based on [74, Rem. 6.11].
(a) Show that type 8b corresponds to the blow-up of P 2 at one of the torus fixed points.
Also show that this is the Hirzebruch surface H1 . Hint: Remember the description of
blowing up given in 2.3.
(b) Similarly show that type 7a (resp. 6a) corresponds to the blow-up of P 2 at two (resp.
three) torus fixed points.
(c) Show that type 3 corresponds to a quotient P 2 /(Z/3Z) and is isomorphic to the surface
in P 3 defined by w3 = xyz. Hint: Compute the normal fan and show that its minimal
generators span a sublattice of index 3 in Z2 . Proposition 3.3.7 will be helpful. For
the final assertion, use the characters coming from the lattice points of the polygon.
(d) Show that type 8c gives the weighted projective space P(1, 1, 2) and type 4c gives the
quotient P(1, 1, 2)/(Z/2Z). Hint: See Example 3.1.17.
(e) Show that type 4a gives the quotient P1 P1 /(Z/2Z).
(f) Show that type 6d gives the weighted projective space P(1, 2, 3).
386
(g) Show that type 7b (resp. 6b) corresponds to the blow-up of P(1, 1, 2) at one (resp. two)
smooth torus fixed points.
(h) Show that type 5b gives the blow-up of P(1, 2, 3) at its unique smooth torus fixed point.
8.3.9. In this exercise you will consider a toric surface that is not quite Fano. The lattice
2
polygon
P P = Conv(3e1 , 3e2 , 2e1 2e2 ) MR = R gives a toric surface XP . Let
D = D be the anticanonical divisor of XP .
(a) Prove that the ample Cartier divisor DP associated to P is given by DP = 6D. Conclude
that D is not Cartier and that 6D is the smallest integer multiple of D that is Cartier.
(b) Show that the normal fan of P has minimal generators e1 2e2, 2e1 e2 and that
the minimal generators are the vertices of Q = Conv(e1 2e2, 2e1 e2 ) NR .
(c) Show that the dual of Q is Q = 61 P and conclude that 6 is the smallest integer multiple
of Q that is a lattice polytope.
8.3.10. A complete toric surface X is log del Pezzo if some integer multiple of its anticanonical divisor KX is an ample Cartier divisor. The index of X is the smallest positive
integer such that such that KX is Cartier. This exercise and the next will consider this
interesting class of toric surfaces.
(a) Prove that a complete toric surface is log del Pezzo of index 1 if and only if it is
Gorenstein Fano.
(b) Prove that the toric surface of Exercise 8.3.9 is log del Pezzo of index 6.
8.3.11. A lattice polygon Q NR is called LDP if the origin is an interior point of Q and
the vertices of Q are primitive vectors in N. The dual Q MR of a LDP polygon contains
the origin as an interior point but may fail to be a lattice polygon. The index of Q is the
smallest positive integer such that such that Q is a lattice polygon. This exercise will
explore the relation between LDP polygons and toric log del Pezzo surfaces.
(a) Show that the polygon Q of Exercise 8.3.9 is LDP of index 6.
(b) An LDP polygon Q NR gives a fan in NR by taking the cones over the faces of Q.
Show that the minimal generators of are the vertices of Q and that the toric surface
X is log del Pezzo of index equal to the index of Q.
(c) Conversely, let X be a toric log del Pezzo surface and let Q be the convex hull of
the minimal generators of . Note that Q is LDP and that is the fan obtained by
taking the cones over the faces of Q. Hint: The key point is to show that every minimal
generator of is a vertex of Q. This can be proved using the strict convexity of the
support function of KX .
This exercise shows that classifying toric log del Pezzo surfaces up to isomorphism is
equivalent to classifying LDP polygons up to lattice equivalence. This is an active area of
researchsee [168].
8.3.12. Let X be a smooth projective toric variety. As in Definition 6.4.10, a primitive
colection P = {1 , . . . , k } (1) gives a primitive relation
P
u1 + + uk = (1) c u , c Z>0 ,
Pk
where P
is the minimal cone containing i=1 ui . Following [17], the degree of P is
(P) = k (1) c . Prove that X is Fano if and only if (P) > 0 for all primitive
collections P. This is [17, Prop. 2.3.6]. Hint: Use (6.4.9) and Theorem 6.4.9.
Chapter 9
Sheaf Cohomology of
Toric Varieties
9.0. Background: Sheaf Cohomology

By Proposition 6.0.8, a short exact sequence of sheaves on X
(9.0.1)
0 F G H 0
gives rise to an exact sequence of global sections

(9.0.2)
0 (X , F ) (X , G ) (X , H ).
The failure of (X , G ) (X , H ) to be surjective is measured by a sheaf cohomology group. The main goal of this chapter is to understand sheaf cohomology
on a toric variety.
Sheaves and Cohomology. A sheaf F on a variety X has sheaf cohomology groups
H p (X , F ). The abstract definition uses an exact sequence of sheaves
d0
d1
d2
0 F I 0 I 1 I 2 ,
where I 0 , I 1 , . . . are injective. This term is from homological algebra: a sheaf I
is injective if given a sheaf homomorphism H I and an injection : H G ,

there exists a sheaf homomorphism making the diagram below commute:
IO a
/H
/ G.
387
Chapter 9. Sheaf Cohomology of Toric Varieties
388
We say that I is an injective resolution of F . From I we get the complex of

global sections
d0
d1
d2
(X , I ) : (X , I 0 ) (X , I 1 ) (X , I 2 ) .
The term complex refers to the fact that d p+1 d p = 0 for all p 0. Then the pth
sheaf cohomology group of F is defined to be
(9.0.3)
H p (X , F ) = H p ((X , I )) = ker(d p )/im(d p1 ),
where for p = 0, we define d 1 to be the zero map 0 (X , I 0 ). One can prove

that injective resolutions always exist and that two different injective resolutions of
F give the same sheaf cohomology groups.
This definition of H p (X , F ) has some very nice properties, including:
H 0 (X , F ) = (X , F ).
A sheaf homomorphism F G induces a homomorphism of cohomology

groups H i (X , F ) H i(X , G ) that is compatible with composition and takes
the identity to the identity, i.e., F 7 H i(X , F ) is a functor.
A short exact sequence of sheaves (9.0.1) gives a long exact sequence
0
0 H 0 (X , F ) H 0 (X , G ) H 0 (X , H )
p1
1
H 1 (X , F ) H 1 (X , G ) H 1 (X , H )

H p (X , F ) H p (X , G ) H p (X , H ) .
We call p : H p (X , H ) H p+1 (X , F ) a connecting homomorphism.
You will prove the first bullet in Exercise 9.0.1. For the second bullet, the key step
is to show that given a sheaf homomorphism : F G and injective resolutions
F A , G B , there are sheaf homomorphisms p : A p B p such that the
diagram
/A0
d0
/A0
d1
/ A1
d0

/ A1
d1
commutes. We say that : A B is a map of complexes. Then p induces the

desired map H p (X , F ) H p (X , G ). Finally, for the last bullet, an exact sequence
(9.0.1) lifts to an exact sequence of injective resolutions
0
/ A
O
/ B
O
/ C
O
/0
/F
/G
/H
/ 0,
and one can show that taking global sections gives an exact sequence of complexes
(9.0.4)
0 (X , A ) (X , B ) (X , C ) 0.
389
It is a general fact in homological algebra that any exact sequence of complexes

0 A B C 0
gives a long exact sequence
(9.0.5)
0
0 H 0 (A ) H 0 (B ) H 0 (C )
p1
1
H 1 (A ) H 1 (B ) H 1 (C )

H p (A ) H p (B ) H p (C ) .
Applied to (9.0.4), we get the desired long exact sequence in sheaf cohomology.
In the language of homological algebra, is left exact since (9.0.2) is exact.
Then the sheaf cohomology groups are the derived functors of . The texts [125,
131, 273] discuss sheaf cohomology and homological algebra in more detail. We
especially recommend Appendix 3 of [89] and Chapters 2 and 3 of [158].
Cech
Cohomology. While the abstract definition of sheaf cohomology has nice
properties, it is not useful for explicit computations. Fortunately, there is a down
to-earth way of viewing sheaf cohomology, in terms of the Cech
complex, which
we now describe.
Let U = {Ui }i=1 be an open cover of X . The definition of a sheaf F on X
shows that H 0 (X , F ) = (X , F ) is the kernel of the map
Y
Y
d0
(9.0.6)
F (Ui )
F (Ui U j ),
1i
1i< j
Q
where d 0 is defined as follows: if = (i ) 1i F (Ui ), then the component of
d 0 () in F (Ui U j ) for i < j is given by j |Ui U j i |Ui U j . Here, j 7 j |Ui U j
is the restriction map F (U j ) F (Ui U j ), and similarly for i 7 i |Ui U j .
To get higher information about how sections of F fit together, we extend
(9.0.6) to the Cech

complex. We will use the following notation. Let [] = {1, . . . , }
be the index set for the open cover and let [] p denote the set of all (p + 1)-tuples
(i0 , . . . , i p ) of elements of I satisfying i0 < < i p .
Definition 9.0.1. The group of pth Cech

cochains is
M
F (Ui0 Ui p ).
Cp (U , F ) =
(i0 ,...,i p )[] p
One can think of an element of Cp (U , F ) as a function that assigns an element of F (Ui0 Ui p ) to each (i0 , . . . , i p ) [] p . Then we define a differential
p
d
Cp (U , F ) Cp+1 (U , F )
390
by describing how d p () operates on elements of [] p+1 :

p+1
X
(1)k (i0 , . . . , ibk , . . . , i p+1 )|Ui
d ()(i0 , . . . , i p+1 ) =
p
k=0
Ui p+1 .
As above, (i0 , . . . , ibk , . . . , i p+1 )|Ui Ui is obtained by applying the restriction

0
p+1
map
bik Ui p+1 ) F (Ui0 Ui p+1 )
F (Ui0 U
to (i0 , . . . , ibk , . . . , i p+1 ). In Exercise 9.0.2 you will verify that d p d p1 = 0.
Definition 9.0.2. Given a sheaf F on X and an open cover U = {Ui }iI of X , the
Cech
complex is
0
d
d
d
C (U , F ) : 0 C0 (U , F ) C1 (U , F ) C2 (U , F ) ,
and the pth Cech

cohomology group is
p (U , F ) = H p (C (U , F )) = ker(d p )/im(d p1 ).
H
Notice that H 0 (U , F ) = H 0 (X , F ) = (X , F ) since F is a sheaf. However,
, F ) need not equal H p (X , F ) for p > 0. We will soon see that there is a
nice case where equality occurs for all p.
p (U
H
Cohomology of a Quasicoherent Sheaf . To compute the sheaf cohomology of a
quasicoherent sheaf F on a variety X using Cech

cohomology, the idea is to use an
p
open cover U of X such that H (U , F ) equals H p (X , F ) for all p. The following

vanishing theorem of Serre is very useful in this regard. Proofs can be found in
[125], [131] and [273].
Theorem 9.0.3 (Serre Vanishing for Affine Varieties). Let F be a quasicoherent
sheaf on an affine variety U . Then H p (U , F ) = 0 for all p > 0.

By Theorem 9.0.3, we can compute the cohomology of a quasicoherent sheaf
F using any affine open cover. Here is the rough intuition:
Consider an arbitrary open cover U = {Ui } of X . Since we can construct X by
gluing together the Ui , the cohomology of X should be obtained from the cohomologies of the Ui and their various intersections. In other words, H (X , F )
should be determined by the cohomology groups H (Ui0 Ui p+1 , F ) as we
vary over all p.
Now suppose that U = {Ui } is an affine open cover. Then Ui0 Ui p+1 is
affine and hence has vanishing higher cohomology by Serres theorem. So all
that is left is H 0 (Ui0 Ui p+1 , F ). This gives the Cech

complex, which
thus computes the sheaf cohomology of F .
This suggests the following result.
391
Theorem 9.0.4. Let U = {Ui } be an affine open cover of a variety X and let F
be a quasicoherent sheaf on X . Then there are natural isomorphisms
p (U , F ) H p (X , F )
H
for all p 0.
The proof will be given later in the section after we introduce a spectral sequence that makes the above intuition rigorous. A more elementary proof that does
not use spectral sequences can be found in [131, Thm. III.4.5].
Here is an application of Theorem 9.0.4.
Example 9.0.5. We compute the cohomology of OP1 , OP1 (1) and 1P1 on P1 .
Consider the affine open cover U = {U0 ,U1 } of P1 , where
U0 = Spec(C[x])
and U1 = Spec(C[x1 ]).
Note also that

U0 U1 = Spec(C[x, x1 ]).
For any quasicoherent sheaf F on P1 , the Cech

complex is
d0
F (U0 ) F (U1 ) F (U0 U1 ).
It follows that H p (P1 , F ) = 0 for p 2. Hence we need only consider H 0 (P1 , F )

and H 1 (P1 , F ). For simplicity, we write these sheaf cohomology groups as H 0 (F )
and H 1 (F ) respectively.
complex becomes
For F = OP1 , the Cech
(9.0.7)
d0
C[x] C[x1 ] C[x, x1 ]
where d 0 ( f (x), g(x1 )) = f (x) g(x1 ). Then
H 0 (OP1 ) = ker(d 0 ) = C
H 1 (OP1 ) = coker(d 0 ) = 0.
The assertion for H 0 is clear since f (x) g(x1 ) = 0 implies that f and g are the
same constant, and the assertion for H 1 follows since an element of C[x, x1 ] can
be written
am xm + + a1 x1 + a0 + a1 x + + an x n .
{z
}
|
{z
} |
f (x)
g(x1 )
We can represent OP1 (1) as the line bundle OP1 (D), where the divisor D is
one of the fixed points of the torus action on P1 . It follows that we have an exact
sequence of sheaves
0 OP1 (1) OP1 OD 0.
The long exact sequence in sheaf cohomology gives
0 H 0 (OP1 (1)) H 0 (OP1 ) H 0 (OD ) H 1 (OP1 (1)) H 1 (OP1 ) .
392
The map H 0 (OP1 ) H 0 (OD ) is the isomorphism that sends a constant function on
P1 to the same constant function on D. Since H 1 (OP1 ) = 0, the long exact sequence
implies that
(9.0.8)
H 0 (OP1 (1)) = H 1 (OP1 (1)) = 0.
Finally, for 1P1 , we use the Euler sequence

0 1P1 OP1 (1) OP1 (1) OP1 0
from (8.1.8). Since H p (X , F G ) H p (X , F ) H p (X , G ) (Exercise 9.0.3), the

vanishing (9.0.8) and the long exact sequence for cohomology imply that
H 0 (1P1 ) = 0
H 1 (1P1 ) H 0 (OP1 ) = C.
Earlier, in Example 6.0.5, we showed that the surjective sheaf homomorphism
OP1 (1) OP1 (1) OP1
is not surjective on global sections. This is what forces H 1 (1P1 ) to be nonzero.
A key part of Example 9.0.5 was the surjectivity of (9.0.7). This is the algebraic
analog of a Cousin problem in complex analysis. A discussion of Cousin problems
and their relation to sheaves and cohomology can be found in [180, Ch. 13].
Serre Vanishing for Projective Varieties. The Serre vanishing theorem for affine
varieties (Theorem 9.0.3) has a projective version.
Theorem 9.0.6 (Serre Vanishing for Projective Varieties). Let L be an ample line
bundle on a projective variety X . Then for any coherent sheaf F on X , we have
for all p > 0 and 0.
H p (X , F X L ) = 0
A proof can be found in [131, Prop. II.5.3]. We will use this result later in the
chapter to prove some interesting vanishing theorems for toric varieties.
Higher Direct Images. Given a morphism f : X Y of varieties and a sheaf F
of OX -modules on X , the direct image is the sheaf f F on Y defined by
U 7 F ( f 1 (U ))
for U Y open. We noted in Example 4.0.24 that f F is a sheaf of OY -modules.

The definition of f F implies in particular that
H 0 (Y , f F ) = H 0 (X , F )
since f 1 (Y ) = X . More generally, there are homomorphisms
(9.0.9)
H p (Y , f F ) H p (X , F ),
393
which need not be isomorphisms for p > 0.

In this situation, we also get the higher direct image R p f F , which is the sheaf
on Y associated to the presheaf defined by
U 7 H p ( f 1 (U ), F ).
Proposition 9.0.7. Let f : X Y be a morphism and F be a quasicoherent sheaf
on X . Then:
(a) The higher direct images R p f F are quasicoherent sheaves on Y .
(b) If U Y is affine, then R p f F |U is the sheaf associated to the OY (U )-module
H p ( f 1 (U ), F ).

A proof can be found in [131, Propositions II.5.8 and III.8.5]. One especially
nice case is when the higher direct images R p f F vanish for p > 0. We will
see below that the maps (9.0.9) are isomorphisms when this happens. The proof
involves our next topic, spectral sequences.
Spectral Sequences. Readers not familar with spectral sequences should glance at
Appendix C before proceeding farther. Here we discuss two spectral sequences
relevant to this section.
First suppose that f : X Y is a morphism and F is a quasicoherent sheaf
on X . As above, we get the higher direct images R p f F , which are sheaves on
Y . Proposition 9.0.7 shows that these sheaves compute the cohomology of F
over certain open subsets of X . So when we put these together, i.e., compute
H p (Y , R q f F ), we should get the cohomology of F on all of X . The precise form
of this intuition is the Leray spectral sequence:
E2p,q = H p (Y , R q f F ) H p+q (X , F ).
Furthermore, the map H p (Y , f F ) H p (X , F ) from (9.0.9) is the edge homomorphism E2p,0 H p (X , F ) (see Definition C.1.6). This spectral sequence is discussed in Theorem C.2.1 and in more detail in [115, II.4.17] and [125, p. 463].
Now assume R q f F = 0 for q > 0. Then
(
H p (Y , f F )
p,q
E2 =
0
q=0
q > 0.
Then Proposition C.1.7 implies that (9.0.9) is an isomorphism. Thus we have

proved the following.
Proposition 9.0.8. Suppose f : X Y is a morphism and F is a quasicoherent
sheaf on X such that R q f F = 0 for q > 0. Then the map (9.0.9) is an isomorphism
H p (Y , f F ) H p (X , F ).

For our second spectral sequence, let U = {Ui } be an open cover of X and F
be a sheaf on X . In the discussion leading up to Theorem 9.0.4, we asserted that
394
H (X , F ) is determined by H (Ui0 Ui p+1 , F ) as we vary over all p. The

precise meaning is given by the E1 spectral sequence
M
H q (Ui0 Ui p , F ) H p+q (X , F ),
(9.0.10)
E1p,q =
(i0 ,...,i p )[] p
where the differential d1p,q : E1p,q E1p+1,q is induced by inclusion, with signs sim
ilar to the differential in the Cech
complex. This spectral sequence is constructed
in [115, II.5.4]. See also Theorem C.2.2.
We now have the tools needed to prove Theorem 9.0.4.
Proof of Theorem 9.0.4. We are assuming that U = {Ui } is an affine open cover
of X . First observe that the q = 0 terms of (9.0.10) are given by
M
H 0 (Ui0 Ui p , F ) = Cp (U , F ),
E1p,0 =
(i0 ,...,i p )[] p
and the differentials d1p,0 are the differentials in the Cech

complex. Hence

E2p,0 = ker(d1 : E1p,0 E1p+1,0 ) im(d1 : E1p1,0 E1p,0 )
p (U , F ).
= H p (C (U , F )) = H
Since F is quasicoherent and the intersections Ui0 Ui p are affine for all p 0,
it follows from Theorem 9.0.3 that E1p,q = 0 for q > 0. This implies that E2p,q = 0 for
q > 0. Using Proposition C.1.7 again, we conclude that the edge homomorphism
E2p,0 = H p (U , F ) H p (X , F )
is an isomorphism for all p 0.
Cohen-Macaulay Varieties. We next discuss a class of varieties that play a crucial

role in duality theory and are interesting in their own right.
We first define what it means for a local ring (R, m) to be Cohen-Macaulay.
Elements f1 , . . . , fs m form a regular sequence if fi is not a zero divisor in
R/h f1 , . . . , fi1 i for all i, and the depth of R is the maximal length of a regular
sequence. Then R is Cohen-Macaulay if its depth equals its dimension. Examples
include regular local rings. A nice discussion of what Cohen-Macaulay means can
be found in [246, 10.2]. See also [179, pp. 153155].
A variety X is Cohen-Macaulay if its local rings (OX,p , mX,p ) are CohenMacaulay for all p X . Thus smooth varieties are Cohen-Macaulay. Later in
the chapter we will prove that normal toric varieties are Cohen-Macaulay.
Serre Duality. Cohen-Macaulay varieties provide the natural setting for a basic
duality theorem of Serre. Here is the simplest version.
395
Theorem 9.0.9 (Serre Duality I). Let X be the canonical sheaf of a complete
normal Cohen-Macaulay variety X of dimension n. Then for every locally free
sheaf F of finite rank on X , there are natural isomorphisms
H p (X , F ) H np (X , X OX F ).
In particular, when D is a Cartier divisor on X and KX is a canonical divisor, we
have isomorphisms
H p (X , OX (D)) H np (X , OX (KX D)).
A proof for the projective case can be found in [131, Thm. III.7.6]. The assertion for divisors follows from
X OX OX (D) OX (KX ) OX OX (D) OX (KX D),
where the last isomorphism holds since D is Cartier.
There is also a more general version of Serre duality that applies when F
is coherent but not necessarily locally free. The cohomology groups H p (X , F )
are the derived functors of the global section functor F 7 (X , F ), and in the
same way, the Ext groups ExtOp X(G , F ) are the derived functors of the Hom functor
F 7 HomOX(G , F ) for fixed G . Then we have the following result.
Theorem 9.0.10 (Serre Duality II). Let X be the canonical sheaf of a complete
normal Cohen-Macaulay variety X of dimension n. Then for every coherent sheaf
F on X , there are natural isomorphisms
H p (X , F ) ExtOnp
(F , X ).
X
When X is projective, this is proved in [131, Thm. III.7.6]. We will give a

version of Serre duality especially adapted to the toric case in 9.2.
When X fails to be Cohen-Macaulay, there is a more general duality theorem
where canonical sheaf X is replaced with the dualizing complex X . A discussion
of this version of duality can be found in [218, 3.2].
Singular Cohomology. Our discussion of the sheaf cohomology of a toric variety
in 9.1 will use some algebraic topology. Here we review the topological invariants
we will need, beginning with the singular cohomology groups
H p (Z, R),
where Z is a topological space and R is a commutative ring, usually Z, Q, R or C.
These are defined using continuous maps
: n Z,
where n is the standard n-simplex. There are several good introductions to singular cohomology, including [124], [135] and [210].
Here are some important properties of singular cohomology:
396
A continuous map f : Z W induces f : H p (W , R) H p (Z, R) such that

homotopic maps induce the same map on cohomology and the identity map
induces the identity on cohomology.
If i : A Z is a deformation retract (i.e., there is a continuous map r : Z A
such that r i = 1A and i r is homotopic to 1Z ), then i : H p (Z, R) H p (A, R)
is an isomorphism.
If Z is contractible (i.e, there is z Z such that {z} Z is a deformation
retract), then
(
R p=0
H p (Z, R) =
0 otherwise.
For n > 1, the singular cohomology of the (n 1)-sphere S n1 Rn is
(
R p = 0, n 1
H p (S n1 , R) =
0 otherwise.
We will always assume that Z is a locally contractible metric space. This allows
us to interpret the singular cohomology of Z in terms of sheaf cohomology as
follows. The ring R defines a presheaf on Z where R is the group of sections over
every nonempty open U Z. The corresponding sheaf is the constant sheaf of R.
By [45, III.1], the sheaf cohomology of the constant sheaf of R on Z is the singular
cohomology H p (Z, R).
The Cohomology Ring. For a commutative ring R, H (Z, R) is a graded R-algebra
with multiplication given by cup product. A continuous map induces a ring homomorphism on cohomology, so in particular, a deformation retract i : A Z induces
a ring isomorphism i : H (Z, R) H (A, R).
Example 9.0.11. The real torus (S1 )n has cohomology ring

H ((S1 )n , R) = R[1 , . . . , n ]
where the i lie in H 1 ((S1 )n , R) and satisfy the relations i j = j i and 2i = 0
(see Examples 3.11 and 3.15 of [135]). Thus
H ((S1 )n , R)
Rn,
i.e., the cohomology ring is the exterior algebra of a free R-module.
Example 9.0.12. The torus (C )n contains the real torus (S1 )n as a deformation
retract via (t1 , . . . ,tn ) 7 (t1 /|t1 |, . . . ,tn /|tn |). Hence
H ((C )n , R) H ((S1 )n , R)
Rn.
More canonically, the torus TN = N C has cohomology ring

H (TN , R)
M Z R,
397
where M is the dual of N. Thus a lattice homomorphism N N gives a map

TN TN of tori, and the induced map H (TN , R) H (TN , R) is the map
V
M Z R
M Z R
determined by the dual homomorphism M M.
Example 9.0.13. The cohomology ring of P n over Z is

H (P n , Z) Z[]/hn+1 i,
where H 2 (P n , Z) (see [135, Thm. 3.12]). In Theorem 12.4.4, we will give a

similar quotient description of the cohomology ring of any smooth complete toric
variety.
Reduced Cohomology. Given a topological space Z, the canonical map Z {pt}

sends all elements of Z to the single point denoted pt. This induces the map
R = H 0 ({pt}, R) H 0 (Z, R)
e 0 (Z, R). Thus, assuming R 6= 0,

whose cokernel is denoted H
e 0 (Z, R) = 0 Z is path connected.

H
The reduced cohomology of Z with coefficients in R is defined for p 0 by

(
e0
e p (Z, R) = H (Z, R) p = 0
H
H p (Z, R) p > 0,
and for p = 1 by
e 1
(
0 Z 6=
(Z, R) =
R Z = .
e 1 may look strange but means that the sequence

The definition of H
(9.0.11)
e 1 (Z, R) R H 0 (Z, R) H
e 0 (Z, R) 0
0 H
e 1 and (9.0.11)
is exact for all Z, including Z = (Exercise 9.0.4). We will use H
in 9.1 when we compute sheaf cohomology on a toric variety.
Exercises for 9.0.
9.0.1. Use (9.0.3) to show that H 0 (X, F ) = (X, F ).
9.0.2. Check that the map d p defined on the Cech

cochains satisfies d p d p1 = 0.
9.0.3. Let F , G be sheaves on X. Prove H p (X, F G ) H p (X, F ) H p (X, G ).
9.0.4. Prove the exactness of (9.0.11).
9.0.5. A morphism f : X Y is affine if Y has an affine open cover {Ui } such that f 1 (Ui )
is affine for all i. Now assume that f : X Y is affine and let F be a quasicoherent sheaf on
X. Use Theorem 9.0.3, Proposition 9.0.7 and Theorem 9.0.8 to prove that H p (Y , f F )
H p (X, F ) for all p 0.
398
9.0.6. Use Exercise 9.0.5 to prove the following isomorphisms in cohomology:

(a) H p (X, i F ) H p (Y , F ) when i : Y X is closed in X and F is quasicoherent.
(b) H p (X, F ) H p (V , F ) when : V X is a vector bundle and F is quasicoherent.
9.0.7. Let X be a variety that is a union of affine open subsets X = U1 Us .

(a) Let F be quasicoherent on X. Prove that H p (X, F ) = 0 for p > s 1.
(b) Let F be quasicoherent on P n . Prove that H p (P n , F ) = 0 for p > n.
9.0.8. Consider the (n 1)-sphere S n1 Rn . Show that

(
e p (S n1 , R) R p = n 1
H
0 otherwise.
Hint: Remember that S 0 consists of two points.
9.1. Cohomology of Toric Divisors

Demazure [82] gave a concrete description of the sheaf cohomology of the sheaf
OX (D) of a torus-invariant Cartier divisor D on a toric variety X . We generalize
this to torus-invariant Q-Cartier divisors, inspired by papers of Eisenbud, Mustata
and Stillman [91], Hering, Kuronya and Payne [141], and Perling [226].
The Toric Cech

Complex. When we compute sheaf cohomology using Cech
cohomology, the obvious choice of open cover is
U = {U }max ,
where max is the set of maximal cones in . We write these as i and order them
according to their indices.
P
Given a torus-invariant Cartier divisor D = a D on X , the Cech

complex
is given by
M
H 0 (Ui0 Ui p , OX (D)).
Cp (U , OX (D)) =
=(i0 ,...,i p )[] p
If we set = i0 i p for = (i0 , . . . , i p ) [] p , then we rewrite this as

M
(9.1.1)
Cp (U , OX (D)) =
H 0 (U , OX (D)).
[] p
The Grading on Cohomology. For an affine open subset U , Proposition 4.3.3

implies that the sections of OX (D) over U can be written
M
C m,
H 0 (U , OX (D)) =
m
where the direct sum is over all m M such that hm, u i a for all (1).
We can write this as
M
H 0 (U , OX (D)) =
H 0 (U , OX (D))m ,
mM
399
where for m M,
(9.1.2)
(
C m
H 0 (U , OX (D))m =
0
hm, u i a for all (1)

otherwise.
This induces a grading of the Cech

complex (9.1.1). The Cech
differential
is built from the restriction maps and hence respects the grading of the complex.
p (U , OX (D)), we obtain a natural decomposition of
Since H p (X , OX (D)) = H
sheaf cohomology
M
H p (X , OX (D)) =
H p (X , OX (D))m .
mM
When decomposing cohomology this way, we often refer to the m M as weights.

Here is an example of how weights can be used to compute sheaf cohomology.
Example 9.1.1. On P 2 , label the rays as usual: u0 = e1 e2 , u1 = e1 , u2 = e2 ,

and maximal cones i , starting with 0 in the first quadrant and going counterclockwise. We will compute H p (P 2 , OP2 (a)) for a Z, where OP2 (a) = OP2 (aD0 )
for the divisor D0 corresponding to u0 .
Let Ui be the affine open corresponding to i . Then Ui j is Ui U j and U012 is
the triple intersection. This allows us to write the Cech

complex as
1
(9.1.3)
1 1
0
C
B
C
B1
0
1
A
@
1
1
1
0
1
1
0 C0 (a) C1 (a) C2 (a) 0,
where
C0 (a) =
2
M
H 0 (Ui , OP2 (aD0 ))
i=0
C1 (a) =
H 0 (Ui j , OP2 (aD0 ))
i< j
C (a) = H 0 (U012 , OP2 (aD0 )).

We will compute the cohomology of this complex using the graded pieces for
m M = Z2 . For this purpose, let l0 , l1 , l2 be the lines in MR = R2 defined by
hm, u0 i = a, hm, u1 i = 0, hm, u2 i = 0. These lines divide the plane into various
regions called chambers. To get a disjoint decomposition, we define
C++ = {m MR | hm, u0 i < a, hm, u1 i 0, hm, u2 i 0}.
Note that a minus sign corresponds to strict inequality (< 0) while a plus sign
corresponds to a weak inequality ( 0). The other chambers C++ , C+ , etc. are
defined similarly.
Suppose a < 0. The corresponding chamber decomposition of MR is shown in
Figure 1 on the next page. The labels li are placed on the plus side of the lines, i.e.,
400
where hm, u0 i a, hm, u1 i 0, hm, u2 i 0. Each chamber is labeled with its sign
pattern, and the shading indicates which chamber the points on the lines belong to.
l1
++
+
l0
++
0
a
l2
++
Figure 1. The chamber decomposition for a < 0
The cone i is generated by u j , uk , where {i, j, k} = {0, 1, 2}. Thus

H 0 (U0 , OP2 (a))m 6= 0 m C++ M
(9.1.4)
H 0 (U1 , OP2 (a))m 6= 0 m C++ M
H 0 (U2 , OP2 (a))m 6= 0 m C++ M.
This follows from (9.1.2) (Exercise 9.1.1). Hence we know when C0 (a)m 6= 0.
The cone 1 2 is generated by u0 , so its dual is the half-plane of MR where

hm, u0 i a, i.e., on the plus side of l0 . Using Figure 1 and (9.1.2) again, we get
H 0 (U12 , OP2 (a))m 6= 0 m (C++ C+ C++ ) M.

Similar results hold for U01 and U02 , so we know when C1 (a)m 6= 0. Finally, since
U012 is the torus of P 2 , we have
C2 (a)m = H 0 (U012 , OP2 (a))m 6= 0 for all m M.
Putting everything together, we get Table 1 on the next page, which shows the
dimension of Cp (a) for m M and a < 0. For example, dim C1 (a)m = 2 when
m C++ since both U01 and U02 contribute 1-dimensional subspaces.
When a < 0, it is now easy to understand the Cech

complex
(9.1.5)
0 C0 (a)m C1 (a)m C2 (a)m 0.
The first line of Table 1 corresponds to m C++ C++ C++ . One can check
without difficulty that for these ms, the complex (9.1.5) is exact, and the same is
true for the second row as well.
401
m M is in
dim C0 (a)m dim C1 (a)m dim C2 (a)m
C++ C++ C++
1
2
1
C+ C+ C+
0
1
1
C
0
0
1
Table 1. Dimension of Cp (a)m for a < 0
These remarks imply that if m M is not in the interior of the triangle

a2 = Conv{(0, 0), (a, 0), (0, a)},
then
H p (P 2 , OP2 (a))m = 0 for all p.
However, if m is a lattice point in the interior of a2 , then H 2 (P 2 , OP2 (a))m is
nonzero. Summing up, we have:
(
0
p 6= 2
p 2

(9.1.6)
dim H (P , OP2 (a)) =
Int(a2 ) M = a1
p=2
2
when a < 0. In what follows, we write h p (a) = dim H p (P 2 , OP2 (a)) for a Z.
In Exercise 9.1.2 you will adapt the above methods to show that
(

a2 M = a+2
p=0
p
2
(9.1.7)
h (a) =
0
p>0
when a 0. Thus, for any line bundle on P 2 , we have completely determined the
dimensions of the cohomology groups H p (P 2 , OP2 (a)). The values for 7 a 4
are depicted in Table 2. Note the symmetry in the table. In Exercise 9.1.3 you will
explore how this symmetry relates to Serre duality.
a h0 (a) h1 (a) h2 (a)

7
0
0
15
6
0
0
10
5
0
0
6
4
0
0
3
3
0
0
1
2
0
0
0
1
0
0
0
0
1
0
0
1
3
0
0
2
6
0
0
3
10
0
0
4
15
0
0
Table 2. Sheaf cohomology of OP2 (a) on P 2
402
P
Computing Cohomology. Given a divisor D = a D on X and m M, define
two subsets of ||:
S
(When is general) VD,m = Conv(u | (1), hm, u i < a ).
supp
(When D is Q-Cartier) VD,m
= {u || | hm, ui < D (u)}, where D is the
support function of D (see Exercise 9.1.4).
Here is an example of these sets.

Example 9.1.2. Let X be the toric surface obtained by blowing up P 2 at the three
fixed points of the torus action. The minimal generators u1 , . . . , u6 of the fan give
divisors D1 , . . . , D6 on X . Let D = D3 + 2D5 D6 and m = e1 . Since
hm, u1 i 0, hm, u2 i 0, hm, u3 i < 1, hm, u4 i < 0, hm, u5 i 2, hm, u6 i < 1,
supp
we get the sets VD,m and VD,m shown in Figure 2 (Exercise 9.1.5). The open circle
u3
VD,m
u2
supp
VD,m
u1
u4
u3
u2
u4
u5
u6 VD,m
u5
u1
u6
supp
VD,m
supp
Figure 2. VD,m and VD,m
for D = D3 + 2D5 D6 and m = e1
supp
at the origin on the right side of the figure is a reminder that VD,m
does not contain
the origin.
supp
The sets VD,m and VD,m enable us to compute sheaf cohomology as follows.
P
Theorem 9.1.3. Let D = a D be a Weil divisor on X . Fix m M and p 0.
e p1 (VD,m , C).
(a) H p (X , OX (D))m H
e p1 (V supp , C).
(b) If D is Q-Cartier, then H p (X , OX (D))m H
D,m
Proof. For part (a), first note that if , then

(9.1.8)
VD,m = Conv(u | (1), hm, u i < a ).
403
One inclusion is trivial; the other follows easily using Lemma 1.2.7. Hence, for all
, we have
H 0 (U , OX (D))m 6= {0} hm, u i a for all (1)
(9.1.9)
VD,m = ,
where the first equivalence uses (9.1.2) and the second follows from (9.1.8).
The equation (9.1.8) shows that VD,m is convex and hence connected when
it is nonempty. It follows that H 0 (VD,m , C) = C when VD,m 6= . Since
H 0 (U , OX (D))m = C m when it is nonzero, the equivalences (9.1.9) give a
canonical exact sequence
0 H 0 (U , OX (D))m C H 0 (VD,m , C) 0
(9.1.10)
for all . It follows that for every p 0, we get an exact sequence

M
M
M
H 0 (VD,m , C) 0,
C
H 0 (U , OX (D))m
0
[] p
[] p
[] p
which we write as
0 Cp (U , OX (D))m B p C p 0.
The formula for the differential in the Cech

complex can be used to define differp
p+1
p
p+1
entials B B
and C C . Then we get an exact sequence of complexes
0 C (U , OX (D))m B C 0.
p (U , OX (D)) H p (X , OX (D)), the long exact sequence from (9.0.5)

Since H
becomes
0 H 0 (X , OX (D))m H 0 (B ) H 0 (C ) H 1 (X , OX (D))m .
In Exercises 9.1.6 and 9.1.7 you will use the theory of Koszul complexes to show
that the complex B has very simple cohomology:
(
C p=0
(9.1.11)
H p (B ) =
0 p > 0.
Thus our long exact sequence breaks up into an exact sequence
0 H 0 (X , OX (D))m C H 0 (C ) H 1 (X , OX (D))m 0
and isomorphisms
H p1 (C ) H p (X , OX (D))m ,
p 2.
We will show below that

(9.1.12)
H p (C ) H p (VD,m , C),
For p 2, this gives the desired isomorphism
p 0.
e p1 (VD,m , C) = H p1 (VD,m , C) H p (X , OX (D))m .

H
404
When p = 1, we obtain the exact sequence

0 H 0 (X , OX (D))m C H 0 (VD,m , C) H 1 (X , OX (D))m 0.
Since H 0 (X , OX (D))m 6= 0 implies VD,m = (Exercise 9.1.8), we obtain
(9.1.13)
e 0 (VD,m , C) H 1 (X , OX (D))m
H
e 1 (VD,m , C) H 0 (X , OX (D))m ,
H
where the last line uses the exact sequence (9.0.11).
It remains to prove (9.1.12). Since || is the union of the maximal cones and
VD,m ||, we get the closed cover C = {VD,m i } of VD,m , where the i are the
maximal cones of . Furthermore, C is the Cech

complex C (C , C) for the
constant sheaf C on VD,m , where we use quotation marks since C is a closed cover
rather than an open cover.
Similar to the spectral sequence of an open covering discussed in 9.0, there
is a spectral sequence for the closed covering C = {VD,m i } of VD,m . This E1
spectral sequence is
M
p,q
H q (VD,m , C) H p+q (VD,m , C).
E1 =
[] p
As explained in Theorem C.2.3 in Appendix C or [115, II.5.2], the differential d1p,q
is determined by the Cech

differential. In particular,
E1p,0 = Cp (C , C) = C p .
Also, we noted earlier in the proof that each VD,m is convex. It follows that
VD,m is contractible, so that
M
p,q
H q (VD,m , C) = 0, q > 0.
E1 =
[] p
As in the proof of Theorem 9.0.4, this implies that

(
H p (C (C , C)) = H p (C ) q = 0
E2p,q =
0
q > 0.
Similar to proof of Theorem 9.0.4, Proposition C.1.3 of Appendix C implies that
the edge homomorphisms are isomorphisms, i.e.,
H p (C ) H p (VD,m , C).
The proves (9.1.12) and completes the proof of part (a).
For part (b), we assume that D is Q-Cartier with support function D . Then
one can modify the proof of (9.1.9) to obtain the equivalence
supp
H 0 (U , OX (D))m 6= {0} VD,m =
405
(Exercise 9.1.9). This allows us to replace (9.1.10) with the exact sequence
supp
0 H 0 (U , OX (D))m C H 0 (VD,m , C) 0.
From here, the proof is identical to what we did in part (a) (Exercise 9.1.9).
Example 9.1.4. Consider the divisor D = D3 + 2D5 D6 on the surface X from

Example 9.1.2. Let us compute H 1 (X , OX (D))m for m = e1 . The sets VD,m and
supp
VD,m are shown in Figure 2 of Example 9.1.2. Since both sets have two connected
components, Theorem 9.1.3 implies that
e 0 (VD,m , C) H
e 0 (V supp , C) C.
H 1 (X , OX (D))m H
D,m
Each representation H p (X , OX (D))m in Theorem 9.1.3 has its advantages.

supp
For example, the vanishing theorems in 9.2 use VD,m because of its relation to
convexity, while for surfaces, we will see below that VD,m is easier to work with.
Other treatments of H p (X , OX (D))m can be found in [91], [141] and [226].
Also note that Theorem 9.1.3 can be stated using relative cohomology instead
of reduced cohomology. Consider, for example, the case when D is Q-Cartier.
supp
Then the inclusion VD,m
|| gives the long exact sequence
supp
supp
H p (||,VD,m , C) H p (||, C) H p (VD,m , C) .
Since || is contractible (it is a cone), this sequence and Theorem 9.1.3 imply
(9.1.14)
supp
H p (X , OX (D))m H p (||,VD,m , C),
p0
supp
(Exercise 9.1.11). Since VD,m
is open in ||, algebraic geometers often write this
relative cohomology group as HZpD,m (||, C), where ZD,m is the closed set
supp
ZD,m = || \VD,m
= {u || | hm, ui D (u)}.
Then part (b) of Theorem 9.1.3 can be restated as

H p (X , OX (D))m HZpD,m (||, C).
This is the version that appears in [105, p. 74].
The Surface Case. When X is a complete toric surface, the set VD,m appearing
in Theorem 9.1.3 has an especially simple topology that can be encoded by a sign
sequence. We illustrate this with an example.
Example 9.1.5. We return toP
the situation of Example 9.1.4. Write the divisor
D = D3 + 2D5 D6 as D = i ai Di , and label the ray generated by ui with + if
hm, ui i ai and with if hm, ui i < ai . Then the picture of VD,m from Figure 2
of Example 9.1.2 gives the sign pattern shown in Figure 3 on the next page.
If we start at u1 and go counterclockwise around the origin, we get the sign
pattern + + +, which we regard as cyclic. The positions of the s determine
VD,m , so that connected components of VD,m correspond to strings of consecutive
s in the sign pattern.
406
+
u3
VD,m
u1
u4
u6 VD,m
u5
u2
Figure 3. The sign pattern for D = D3 + 2D5 D6 and m = e1 in Example 9.1.5
These observations hold in general. Let X be a complete toric surface with

minimal generators u1 , . . . , ur arranged
P counterclockwise around the origin. Given
a torus-invariant Weil divisor D = i ai Di on X and m M Z2 , the sign pattern
signD (m) is the string of length r whose ith entry is + if hm, ui i ai and if
hm, ui i < ai . We regard signD (m) as cyclic. Thus, for example, + + + +
has only one string of consecutive s.
We have the following result from [142] and [154] (Exercise 9.1.12).
P
Proposition 9.1.6. Given a Weil divisor D = i ai Di on a complete toric surface
X , the dimension of H p (X , OX (D))m is given by
(
1 if signD (m) = + + ( VD,m = )
dim H 0 (X , OX (D))m =
0 otherwise
dim H 1 (X , OX (D))m = max(0, #connected components of VD,m 1)
= max(0, #strings of consecutive s in signD (m) 1)

(
1 if signD (m) = ( VD,m is a cycle)
2
dim H (X , OX (D))m =
0 otherwise.
Example 9.1.7. Let us revisit Example 9.1.1. For each m M = Z2 , the minimal
generators u0 , u1 , u2 give a sign pattern for the line bundle OP2 (a) = OP2 (aD0 ).
When a < 0, all possible sign patterns are recorded in Figure 1 of Example 9.1.1.
Using Proposition 9.1.6, we obtain:
H 0 (P 2 , OP2 (a)) = 0 since + + + does not appear.
H 1 (P 2 , OP2 (a)) = 0 since all patterns have one string of consecutive s.

L
H 2 (P 2 , OP2 (a)) = mInt(a2 ) C m since labels the interior of a2 .
Thus the computation of Example 9.1.1 follows immediately from Figure 1.
407
Example 9.1.8. Consider the surface X and divisor D = D3 + 2D5 D6 from

Example 9.1.4. The sign patterns signD (m) for all m M Z2 give the chamber
decomposition shown in Figure 4 (Exercise 9.1.10).
+++
++++
++++
+++
+++
+++++
++++
e1 + e2
l3
+++
++++
0
++
+++
l6
e1
2e1
+++
++
++
+++
+++
+++
++++
l4 = l1
++++
l2
l5
Figure 4. The chamber decompostion for D = D3 + 2D5 D6
P
As in Figure 1, we have lines li defined by hm, ui i = ai , where D = i ai Di .
We put the label li on the side where hm, ui i ai , and the shading indicates where
the boundary points lie. Each chamber is labeled with its sign pattern (some labels
are rather small), and we get five 1-dimensional chambers since l1 = l4 .
We now compute cohomology using Proposition 9.1.6. First,
H 0 (X , OX (D)) = H 2 (X , OX (D)) = 0
since neither + + + + ++ nor appear anywhere. Furthermore, of the
six chambers with sign patterns having more than one string of consecutive s,
three of them ( + + +, + +, plus the 1-dimensional + + +)
have no lattice points. Of the remaining three, the lattice points are:
m = 0 from + + + + (1-dimensional).
m = e1 , 2e1 from + + +.
m = e1 + e2 from + + + +.
These lattice points are indicated with white dots in Figure 4. Hence
H 1 (X , OX (D)) C 0 C e1 C 2e1 C e1 +e2 .
In particular, dim H 1 (X , OX (D)) = 4. Exercise B.8.2 explains how to check this

computation using Macaulay2 [123].
408
The paper [142] applies these methods to classify line bundles with vanishing
higher cohomology on the toric surface coming from three consecutive blowups of
the Hirzebruch surface H2 .
Exercises for 9.1.
9.1.1. In Example 9.1.1, verify the sign patterns in Figure 1.
9.1.2. Use the methods of Example 9.1.1 to prove (9.1.7).
9.1.3. The canonical bundle of P 2 is P2 = OP2 (3). Use this and Serre duality to explain
the symmetry in Table 2 in Example 9.1.1.
P
9.1.4. Recall that a Weil divisor D = a D on X is Q-Cartier if some positive integer
multiple of D is Cartier. Let D be the support function of D as in Theorem 4.2.12.
(a) Show that (1/)D : || R is a support function (Definition 4.2.11) and depends
only on D. We define the support function of D to be D = (1/)D .
(b) Show that D is Q-Cartier if and only if for every , there is m MQ such that
hm , u i = a for all (1). When this is satisfied, show that D (u) = hm , ui for
all u .
9.1.5. Verify Figure 2 in Example 9.1.2
9.1.6. Given a ring R and elements f1 , . . . , f R, define
Vp
Vp+1
d p : R
R
P
by d p () = ( i=1 fi ei ) , where e1 , . . . , e are the standard basis of R . Setting K p =

Vp
R , we get the complex
d0
d1
d 2
d 1
K : K 0 K 1 K 1 K ,
and for an R-module M, we get the Koszul complex

K ( f1 , . . . , f ; M) = K R M.
Thus K = K ( f1 , . . . , f ; R).
(a) Given : R R, show that there are maps s p : K p K p1 such that s1 = and
s p+q ( ) = s p () + (1) p s q () for all Kp and Kq .
(b) Show that for all K p , the maps s p satisfy
P
d p1 (s p ()) + s p+1 (d p ()) = ( i=1 fi ei ) .
(c) Now assume that h f1 , . . . , f i = R. Prove that K ( f1 , . . . , f ; M) is exact for every RP

module M. Hint: Define (ei ) = gi where i=1 gi fi = 1.
9.1.7. When f1 = = f = 1 and M is an R-module, the Koszul complex K (1, . . . , 1; M)

of Exercise 9.1.6 becomes
M
M
M
(9.1.15)
M
M
M
M ,
1i
1i< j
1i< j<k
where the entries of the matrices representing the differentials are 0 or 1.

(a) Use the previous exercise to prove that (9.1.15) is exact.
9.2. Vanishing Theorems I
409
(b) Prove that the complex B defined in the proof of Theorem 9.1.3 satisfies (9.1.11).
Hint: Show B is the Koszul complex (9.1.15) with M = C, minus its first term.
9.1.8. Prove that H 0 (X , OX (D))m 6= 0 if and only if Vm,D = . Then prove (9.1.13).
9.1.9. Complete the proof of part (b) of Theorem 9.1.3.
9.1.10. Verify Figure 4 in Example 9.1.8.
9.1.11. Prove (9.1.14).
9.1.13. Prove that H 0 (X , OX (D))m 6= 0 implies that H p (X , OX (D))m = 0 for all p > 0.
Hint: Exercise 9.1.8.

In this section we prove two basic vanishing theorems for Q-Cartier divisors on a
toric variety X . We then apply these results to show that normal toric varieties are
Cohen-Macaulay and hence satisfy Serre duality. We also give a version of Serre
duality that is special to the toric case.
Nef Q-Cartier Divisors. A Weil divisor D on a normal variety X is Q-Cartier if
some positive integer multiple is Cartier. Also recall that the intersection product
D C is defined whenever D is Q-Cartier and C is a complete irreducible curve in
X . Then D is nef if D C 0 for all complete irreducible curves C X . This
generalizes the definition of nef Cartier divisor given in 6.3.
In the toric case, we characterize Q-Cartier nef divisors as follows.
P
Lemma 9.2.1. Let D = a D be a Q-Cartier divisor on X . If has convex
support, then the following are equivalent:
(a) D is nef.
(b) For some integer > 0, D is Cartier and basepoint free, i.e., OX (D) is a
line bundle generated by its global sections.
(c) The support function D : || R is convex.
Proof. The proof combines ideas from Theorems 6.1.7, 6.3.12 and 7.2.2. We leave

In the Cartier case, we know that D is nef if and only if OX (D) is generated
by global sections (Theorems 6.1.7 and 6.3.12). This fails in the Q-Cartier case, as
shown by the following example.
Example 9.2.2. Let X = P(2, 3, 5) and let D0 , D1 , D2 be the divisors given by the
minimal generators u0 , u1 , u2 . Note that 2u0 + 3u1 + 5u2 = 0. Then Cl(X ) Z,
where the classes of D0 , D1 , D2 map to 2, 3, 5 respectively. Let D = D1 D0 and
note that D0 2D, D1 3D, D2 5D. It also easy to see that D is Q-Cartier and
410
nef. However, you will check in Exercise 9.2.2 that H 0 (X , OX (D)) = 0. Thus
OX (D) is not generated by its global sections.
Demazure Vanishing. The most basic vanishing theorem for toric varieties is the
following result, first proved by Demazure [82] for Cartier divisors.
Theorem 9.2.3 (Demazure Vanishing). Let D be a Q-Cartier divisor on X . If ||
is convex and D is nef, then
H p (X , OX (D)) = 0 for all p > 0.
Proof. The support function D : || R is convex by Lemma 9.2.1. Fix m M
supp
supp
and let VD,m = {u || | hm, ui < D (u)} as in Theorem 9.1.3. Let u, v VD,m ,
so that hm, ui < D (u) and hm, vi < D (u). Then, for 0 t 1, we have
hm, (1 t)u + tvi = (1 t)hm, ui + thm, vi
< (1 t)D (u) + tD (v)
D ((1 t)u + tv).
supp
The last line follows since D is convex. This implies that VD,m is convex and
hence contractible. Combining this with Theorem 9.1.3, we obtain
supp
e p1 (V
H p (X , OX (D))m H
D,m , C) = 0 for all p > 0.
In particular, if D is a basepoint free Cartier divisor on X , then D is nef, so

that Theorem 9.2.3 implies
H p (X , OX (D)) = 0 for all p > 0.
Example 9.2.4. For P n , aD0 is basepoint free for a 0. Hence the higher cohomology of OPn (a) = OPn (aD0 ) vanishes by Theorem 9.2.3, which agrees with
what we found in Example 9.1.1 when n = 2.
Vanishing of Higher Direct Images. Here is an easy application of Demazure

vanishing.
Theorem 9.2.5. Let : X1 X2 be a proper toric morphism and let L be a line
bundle on X2 Then:
(a) R p L = 0 for all p > 0.
(b) If the map : N1 N2 on the lattices of one-parameter subgroups is surjective,
then L L .
Proof. First assume L = OX2 , so that L = OX1 . Let U X2 be the affine
open corresponding to 2 . By Proposition 9.0.7, R p OX1 |U is the sheaf
associated to the module H p (1 (U ), OX1 ). Since is proper, Theorem 3.4.11
tells us that 1 (U ) is a toric variety whose fan is supported on the convex set
411
R (). Thus H p (1 (U ), OX1 ) = 0 for p > 0 by Theorem 9.2.3. It follows that

R p OX1 = 0 for p > 0 by Proposition 9.0.7.
The morphism induces a homomorphism of sheaves OX2 OX1 on X2
(Example 4.0.25). It suffices to show that this map is an isomorphism on each affine
open U X2 , 2 . Note also that the restriction 1 (U ) U is again a
1
proper toric morphism. Hence we may assume that X2 = U . Then |1 | = R ()
is a convex cone, so that
M
H 0 (X1 , OX1 ) =
C m
m|1 | M1
by Exercise 4.3.4. We analyze |1 | M1 as follows.
The surjection N1 N2 gives an injection M2 M1 on character lattices. Then

(N2 )R has dual (M2 )R (M1 )R and |1 | (N1 )R has dual |1 |
1
(M1 )R . Also, |1 | = R () implies = |1 | . It follows that
|1 | M1 = M1 = (M2 )R M1 = M2 ,
where the last equality follows since M1 is saturated in M2 by the surjectivity of

N1 N2 . Hence
M
H 0 (X1 , OX1 ) =
C m = H 0 (U , OU ).
m M2
This shows that OU OX1 induces an isomorphism on global sections, which

gives the desired sheaf isomorphism since we are working with quasicoherent
sheaves over an affine variety.
For a line bundle L on X2 , take U X2 . Then L |U OX2 |U , so that
R p L |U R p OX2 |U R p OX1 |U = 0
for p > 0. Furthermore, when : N1 N2 is surjective, we have

L OX1 OX L L ,
2
where the first isomorphism is part (a) of Exercise 9.2.3 and the second follows
from OX1 OX2 .

The Injectivity Lemma. We introduce a method that will be used several times in
this section and the next. The general framework uses a morphism f : Y X and
coherent sheaves G on Y and F on X such that:
f is an affine morphism, meaning f 1 (U ) X is affine when U Y is affine.
There is a split injection i : F f G of OX -modules. This means that there is

a homomorphism r : f G F such that r i = 1F .
By functoriality, a split injection i : F f G induces a split injection

H p (X , F ) H p (X , f G ).
412
However, since f is affine, we also have

H p (X , f G ) H p (Y , G )
by Exercise 9.0.5. Hence we get an injection
H p (X , F ) H p (Y , G ).
In particular, H p (Y , G ) = 0 implies H p (X , F ) = 0. This method is used in the
proofs of many vanishing theoremssee [186, Ch. 4].
In the toric context, Fujino [101] identified the best choice for the morphism
f . Given a fan in NR and a positive integer , let : N N be multiplication
by . This maps to itself and hence induces a toric morphism : X X . The
dual of is multiplication by on M, and the restriction of to TN = M Z C
is the group homomorphism given by raising to the th power. Furthermore, given
any , we have 1 (U ) = U since ( )1
R () = . This shows that is an
affine morphism.
Here is our first injectivity lemma.
Lemma 9.2.6. Let D be a Weil divisor on X and let be a positive integer. Then
there is an injection
H p (X , OX (D)) H p (X , OX (D)) for all p 0.
Proof. We will construct a split injection OX (D) OX (D). The lemma
then follows immediately from the above discussion.
P
For D = a D and , let P = {m MR | hm, ui a for all (1)}.
Over U , the sheaves OX (D) and OX (D) come from the C[ M]-modules
M
M
C m and
C m.
mPM
m(P)M
L
= U , OX (D) is determined by m(P)M C m with the
Since
module structure given by m a = m a. This implies that the map
M
M
(9.2.1)
C m
C m
1
(U )
mPM
m(P)M
that sends m to m is a homomorphism of C[ M]-modules (Exercise 9.2.4).

Furthermore, one can show that the map given by
(
0
m
/ M
(9.2.2)
r( m ) =
m
m = m , m M
defines a C[ M]-module splitting of (9.2.1) (Exercise 9.2.4).
The map (9.2.1) and the splitting r are easily seen to be compatible with the
inclusion U U when is a face of . It follows that the maps (9.2.1) patch to
give the split injection OX (D) OX (D).
413
Batyrev-Borisov Vanishing. In Example 9.1.1, we saw that OP2 (a) has nontrivial
H 2 when a 3. If we write a = b for b > 0, we can rewrite (9.1.6) as
(
0
p 6= 2
p
2

dim H (P , OP2 (b)) =
Int(b2 ) M p = 2.
This P
has been generalized by Batyrev and Borisov [18]. Recall that a Weil divisor
D = a D gives the polyhedron
PD = {m MR | hm, u i a for all (1)},
which is a polytope when is complete (Proposition 4.3.8).

P
Theorem 9.2.7 (Batyrev-Borisov Vanishing). Let D = a D be a Q-Cartier
divisor on a complete toric variety X . If D is nef, then
(a) H p (X , OX (D)) = 0 for all p 6= dim PD .
L
(b) When p = dim PD , H p (X , OX (D)) mRelint(PD )M C m .
Proof. First assume that D is Cartier and dim PD = dim NR = dim X . Then D is
basepoint free since it is nef (Theorem 6.3.12). By Proposition 6.2.5 and Theorem 6.2.8, combining cones of that share the same linear functional relative to
D gives a fan D such that refines D and D is strictly convex relative to D .
e p1 (V supp , C) by Theorem 9.1.3. Set
For m M, H p (X , OX (D))m H
D,m
ZD,m =
supp
NR \VD,m
= {u NR | hm, ui D (u)}.
Since D is strictly convex relative to D , the proof of Theorem 9.2.3 implies that
ZD,m is convex. In fact, it is strongly convex. To see why, note that for any nonzero
u NR , the strict convexity of D implies 0 = D (u + (u)) > D (u) + D (u)
since u and u do not lie in the same cone of D . Thus
D (u) = D (u) > D (u)
for all u 6= 0 in NR .
Now assume u ZD,m \ {0}. Then hm, ui D (u), so that

D (u) hm, ui.
Combining the displayed inequalities gives D (u) > hm, ui, which implies
supp
that u VD,m = NR \ ZD,m . Hence ZD,m is strongly convex.
We next prove that
(9.2.3)
ZD,m = {0} m Int(PD ) M.
If m Int(PD ) M, then m Int(PD ), so that

hm, u i > a = D (u ) for all (1).
Thus
(9.2.4)
hm, u i < D (u ) for all (1),
414
which easily implies that ZD,m = {0} (Exercise 9.2.5). On the other hand, if we
have m
/ Int(PD ) M, then one of the inequalities of (9.2.4) must fail, i.e.,
hm, u0 i D (u0 ) for some 0 (1).
Then u0 ZD,m , so that ZD,m 6= {0}. This proves (9.2.3).
We are now ready to compute H p (X , OX (D))m . If m Int(PD ) M, then

= NR \ {0} is homotopic to S n1 . Hence
(
supp
p
p1
p1
n1
e (V
e (S , C) = 0 p 6= n
H (X , OX (D))m H
D,m , C) H
C p = n.
supp
VD,m
supp
If m
/ Int(PD ) M, then VD,m = Rn \ ZD,m , where ZD,m is a closed strongly
convex cone of positive dimension. If u 6= 0 in ZD,m , we showed above that
supp
supp
u VD,m . It is easy to see that VD,m contracts to u (Exercise 9.2.6). Hence
e p1 (V supp , C) = 0
H p (X , OX (D))m H
D,m
for all p 0.
This completes the proof when D is Cartier and PD has dimension n.

Now suppose D is Cartier but dim PD < n. We will use Proposition 6.2.5 and
Theorem 6.2.8. After translation, PD spans (MD )R MR , where MD M is dual to
a surjection : N ND , and the normal fan of PD lies in (ND )R . If XD is the toric
variety of PD (MD )R , then Theorem 6.2.8 shows that induces a toric morphism
: X XD such that D is the pullback of the ample divisor D on XD determined
by PD . In particular, OX (D) OXD (D ).
Since is proper and : N ND is surjective, we have
R p OX (D) = 0, p > 0
OX (D) = OXD (D )
by Theorem 9.2.5. Then Proposition 9.0.8 implies that
(9.2.5)
H p (XD , OXD (D )) H p (X , OX (D)).
This proves the theorem, as we now explain. Since MD is saturated in M by the

surjectivity of N ND , we have
PD M = PD (MD )R M = PD MD ,
Since PD = PD is full dimensional in (MD )R , we are done by (9.2.5) and the full
dimensional case considered above.
Finally, we need to consider what happens when D is Q-Cartier. Pick an integer
> 0 such that D is Cartier. By Lemma 9.2.6, we have an injection
H p (X , OX (D)) H p (X , OX (D)).
415
Since D is Cartier and nef with PD = PD , the Cartier case proved above implies
that H p (X , OX (D)) = 0 for p 6= dim PD . Furthermore, when p = dim PD ,
M
M
e p1 (V supp , C) m =
H p (X , OX (D)) =
H
C m ,
D,m
mM
from which we conclude
supp
e p1 (V supp , C)
H
D,m
mRelint(PD )M
(
C
=
0
m Relint(PD ) M
otherwise.
supp
Since VD,m = VD,m and m Relint(PD ) M m Relint(PD ) M,

M
M
e p1 (V supp , C) m =
H p (X , OX (D)) =
C m .
H
D,m
when a < 0. This generalizes (9.1.6) from Example 9.1.1.
mM
mRelint(PD )M
Example 9.2.8. We saw in Example 9.2.4 that the higher cohomology of OPn (a)
vanishes when a 0. When a < 0, OPn (a) = OPn (|a|D0 ). Since |a|D0 is ample
with polytope |a|n , Theorem 9.2.7 implies
(
0
p 6= n

dim H p (P n , OPn (a)) =
Int(an ) M p = n
The Cohen-Macaulay Property. Here is a surprising consequence of BatyrevBorisov vanishing.
Theorem 9.2.9. A normal toric variety is Cohen-Macaulay.
Proof. First suppose that X is projective and let D be a torus-invariant ample

divisor on X . Corollary 5.72 of [179, p. 182] implies that X is Cohen-Macaulay
if and only if
(9.2.6)
H p (X , OX (D)) = 0 for all 0, p < n.
Now take any integer > 0. Since D is ample, the divisor D is nef and its
polytope PD has dimension n = dim X . Then (9.2.6) follows from Theorem 9.2.7.
We conclude that X is Cohen-Macaulay.
Next consider an affine toric variety U . We can find a projective toric variety
that contains U as an affine open subset (Exercise 9.2.7). Since being CohenMacaulay is a local property, it follows that U is Cohen-Macaulay, and then any
normal toric variety X is Cohen-Macaulay.

This result was originally proved by Hochster [146]. Another proof can be
found in [76, Thm. 3.4]. A projective variety is called arithmetically CohenMacaulay (aCM for short) if its affine cone is Cohen-Macaulay. In Exercise 9.2.8
you will show that normal lattice polytopes give aCM toric varieties.
416
Serre Duality. Theorem 9.2.9 shows that Serre duality holds for any normal toric
variety. However, the version stated in Theorem 9.0.9 only applies to locally free
sheaves, while the more general version Theorem 9.0.10 uses Ext groups. Many
b p , fail to be locally free when
of the sheaves we deal with, such as OX (D) and
X
X is not smooth. Fortunately, there is an Ext-free version of Serre duality that
holds for these sheaves.
Theorem 9.2.10 (Toric Serre Duality). Assume that X is a complete toric variety
of dimension n.
(a) If D is a Q-Cartier divisor and KX is the canonical divisor, then we have
natural isomorphisms
(b) If X is simplicial and F is locally free, then there are natural isomorphisms
b q O F ) H np (X ,
b nq O F ).
H p (X ,
X
X
X
X
Remark 9.2.11. Here are two interesting aspects of Theorem 9.2.10:

(a) OX (KX D) need not be isomorphic to
X OX OX (D) = OX (KX ) OX OX (D)
when D is Q-Cartier. So Theorem 9.2.10 does not follow from Theorem 9.0.9.
(b) Every Weil divisor is Q-Cartier on a simplicial toric variety. Theorem 9.2.10
holds for all Weil divisors in this case.
Before beginning the proof of Theorem 9.2.10, we introduce some tools from
commutative algebra. This is typical of algebraic geometryonce a variety ceases
to be smooth, the theory becomes more technically demanding.
In 9.0 we defined the depth of a local ring (R, m). More generally, the depth of
a finitely generated R-module F is the maximal length of a sequence f1 , . . . , fs m
such that fi is not a zero divisor in F/h f1 , . . . , fi1 iF for all i, and the dimension of
F is dim R/Ann(F), where Ann(F) = { f R | f F = 0}.
Then we define a coherent sheaf F on a variety X to be maximal CohenMacaulay (MCM for short) if for every point x X , the stalk Fx is an OX,x module whose depth and dimension both equal dim X . For example, if X is CohenMacaulay, then every locally free sheaf on X is MCM.
The following result from [226, Prop. 4.24] shows that MCM sheaves satisfy a
nice version of Serre duality. See also [179, Thm. 5.71].
Theorem 9.2.12 (Serre Duality III). Let F be a MCM sheaf on a complete normal
Cohen-Macaulay variety X of dimension n. There there are natural isomorphisms
H p (X , F ) H np (X , H omOX (F , X )).
417
Proof. We will use the sheaf version of Ext, where the sheaves E xtOq X (F , X ) are
the derived functors of F 7 H omOX (F , X ). The stalk at x X is
(9.2.7)
E xtOq X (F , X )x = ExtOq X,x (Fx , (X )x ).
The relation between the Ext groups ExtOq X (F , X ) and Ext sheaves E xt Oq X (F , X )
is described by the spectral sequence
E2p,q = H p (X , E xtOq X (F , X )) ExtOp+q
(F , X )
X
from Theorem C.2.4 of Appendix C.
q
Since F is MCM, [56, Cor. 3.5.11] implies that ExtOX,x (Fx , (X )x ) = 0 for
q > 0 and x X . By (9.2.7), we obtain E xt Oq X (F , X ) = 0 for q > 0. Hence
H p (X , H omOX (F , X )) ExtOp X (F , X )
by Proposition C.1.7 of Appendix C. Since H p (X , F ) ExtOnp
(F , X ) by the
X
version of Serre duality given in Theorem 9.0.10, we are done.

We can now prove the toric version of Serre duality.
Proof of Theorem 9.2.10. We first show that OX (D) is MCM when D is QCartier. Pick > 0 such that D is Cartier. We will use splitting methods, though
for clarity we replace : N N with the sublattice N N. The dual lattice
1 M M gives semigroup algebras S,N = C[ M] S,N = C[ 1 M].
Over an affine open subset U , we can pick m 1 M such that D (u) =

hm , ui for u . Now consider the S,N -module A = (U , OX (D)). If we set
P = m + , then one easily sees that
M
M
(9.2.8)
A =
C m B =
C m = m S,N ,
mPM
mP1 M
where the last equality uses m 1 M. Note that B = m S,N is free over S,N
and hence is MCM over S,N . However, S,N is finitely generated as a S,N module (Exercise 9.2.9), so that B is MCM over S,N by [56, Ex. 1.2.26].
Another property of MCM modules is that any nontrivial direct summand of an
MCM module is again MCM. This follows from [56, Thm. 3.5.7]. Hence it suffices
to split (9.2.8). In this situation, we use the map r : B A that sends m B to
(
m if m A
m
r( ) =
0
otherwise.
Similar to what we did in Lemma 9.2.6, r is a homomorphism of S,N -modules.
From here, it follows easily that OX (D) is MCM when D is Q-Cartier.
418
Now we prove duality. Since OX (D) is MCM, Theorem 9.2.12 implies that
H p (X , OX (D)) H np (X , H omOX (OX (D), X ))
= H np (X , H omOX (OX (D), OX (KX ))).
However, H omOX (OX (D), OX (E)) OX (E D) for any Weil divisors D, E on a

normal variety X (Exercise 9.2.10). This gives the desired duality for OX (D).
b q is MCM when X is simplicial. Since is
For part(b), we first show that
X
complete, we can work locally over U , dim = n. The minimal generators of
form a basis of a sublattice N N of finite index such that is smooth relative to
N . Let M be the dual of N . As above, we get semigroup algebras S,N S,N .
b q to U = U,N is determined by
By Proposition 8.2.18, the restriction of
X
the S,N -module
M Vq
A=
V (m) m ,
m M
where V (m) = SpanC (m0 M | m0 the minimal face of containing m). Now
consider the larger S,N -module defined by
M Vq
B=
V (m) m .
m M
Since V (m) is unaffected when M is replaced by M , we see that as a S,N -module,

bq
B = (U,N ,
U
,N
).
bq
However,
U,N is locally free since U,N is smooth by our choice of N . Hence
B is MCM over S,N . Arguing as in part (a), B is MCM over S,N and we have a
splitting map r defined by
(
m if m M
m
r( ) =
0
otherwise.
b is MCM, and then

b O F is MCM since F is locally free.
Thus
X
X
X
q
The proof is now easy to finish, since Theorem 9.2.12 implies
b O F ) H np (X , H omO (
b O F , X ))
H p (X ,
X
X
X
X
X
However, using the local freeness of F and Exercise 8.0.13, we have
b nq O F .
b q , X ) O F
b q O F , X ) H omO (
H omOX(
X
X
X
X
X
X
X
From here, the theorem follows easily.
The proof of part (a) of the theorem was inspired by [76, Lem. 3.4.2]. Other
proofs that Q-Cartier divisors give MCM sheaves can be found in [54, Cor. 4.2.2]
and [226, Prop. 4.22]. In part (b) we followed [76, Prop. 4.8].
419
We should also mention that besides Q-Cartier divisors, there can be other
Weil divisors that give MCM sheaves. For example, X is MCM on any normal
Cohen-Macaulay variety (see [56, Def. 3.3.1]), so that X is MCM on any toric
variety X , even when the canonical class KX fails to be Q-Cartier. You will give
a toric proof of this in Exercise 9.2.11.
In Exercise 9.2.12 you will use Alexander duality to give a purely toric proof
of Serre duality for simplicial toric varieties.
Exercises for 9.2.
9.2.2. Verify the claims made in Example 9.2.2.
9.2.3. We saw in Example 4.0.25 a morphism of varieties f : X Y induces a homomorphism OY f OX of OY -modules.
(a) Let L be a line bundle on Y . Construct an isomorphism f f L ( f OX ) OY L
of OY -modules. Hint: Construct a homomorphism and then study the homomorphism
over open subsets of Y where L is trivial.
(b) Generalize part (b) by showing that for any sheaf G of OX -modules on X and any line
bundle L on Y , there is an isomorphism f(G OX f L ) f G OY L . This is the
projection formula.
L
L
9.2.4. Consider the map mPM C m m(P)M C m from (9.2.1), where the
L
C[ M]-module structure on m(P)M C m is given by m a = m a. Prove that
(9.2.1) and (9.2.2) are C[ M]-module homomorphisms.
9.2.5. Prove that (9.2.4) implies ZD,m = {0}, as claimed in the proof of Theorem 9.2.7.
9.2.6. As in the proof of Theorem 9.2.7, assume that ZD,m is strongly convex and that u
supp
supp
ZD,m is nonzero. Thus u VD,m . Prove that for every v VD,m , the line segment uv
supp
supp
is contained in VD,m (this means that VD,m is star shaped with respect to u). Conclude
supp
that the constant map : VD,m
{u} is a contraction.
9.2.7. Prove that every affine toric variety U is contained in a projective toric variety X .
9.2.8. The lattice points of a normal lattice polytope P give a projective embedding of the
toric variety XP . Prove that XP is aCM in this embedding.
9.2.9. Assume that N1 N2 has finite index, with dual M2 M1 , and let be a cone in
(N1 )R = (N2 )R . This gives semigroup algebras S,N2 = C[ M2 ] S,N1 = C[ M1 ].
generated as a module over S,N2 . Hint: Let m1 , . . . , mr M2
Prove that S,N1 is finitely P
r
generate and consider { i=1 i mi | 0 i < 1} M1 .
9.2.10. Let D, E be Weil divisors on a normal variety X and let U X be the smooth locus.
Then (X, OX (D)) (U, OX (D)|U ), and the same holds for E.
(a) Construct a natural map X : (X, OX (E D)) HomOX (OX (D), OX (E)) and prove
that X is an isomorphism when X is smooth.
(b) Prove that H omOX (OX (D), OX (E)) is reflexive. Hint: Let U is the smooth locus and
study the restriction map HomOX (OX (D), OX (E)) HomOU(OX (D)|U , OX (D)|U ).
420
(c) Show that X is an isomorphism and that OX (E D) H omOX (OX (D), OX (E)).
9.2.11. Following [76, Cor. 3.5], you will show that X is MCM on a normal toric variety.
Fix a cone NR and pick a basis e1 , . . . , en of M such that en is in the interior of . Let
= lcm(hen , u i | (1)) and let M be the lattice withL
basis e1 , . . . , en1 , 1 en , with dual
N N. By Proposition 8.2.9, U comes from the ideal mInt( )M C m C[ M].
(a) Prove that u = (/hen , u i)u lies in N. Then use h1 en , u i = 1 to show that u is
the minimal generator of with respect to N.
(b) Conclude that the canonical divisor of U,N is Cartier.

L
L
(c) Construct a splitting of mInt( )M C m mInt( )M C m and conclude that
the canonical sheaf of U,N is MCM.
9.2.12. Alexander duality (see [210, 71]) states that if A S n1 is a closed subset such
that the pair (S n1 , A) is triangulable (see [210, p. 150]), then
e p1 (A, C) H
e np1 (S n1 \ A, C).
H
P
You will prove Serre duality when D = a D on a complete simplicial toric variety X
of dimension n. Let K = KX be the canonical divisor of X . Theorem 9.1.3 implies
e p1 (VD,m , C), m M.
H p (X , OX (D))m H
Set AD,m = VD,m and = Conv(u | (1)). Your goal is to prove that
np
(X , OX (K D))m .
H p (X , OX (D))
m H
S
(a) Explain why S = is homeomorphic to S n1 . Note that AD,m , AKD,m S.
(9.2.9)
(b) Prove that AD,m AKD,m = and that for , is contained in neither AD,m
nor AKD,m if and only if there are 1 , 2 (1) such that hm, u1 i < a1 and
hm, u2 i a2 . We say that is intermediate when this happens.
(c) Fix an intermediate cone . Let
be the face of generated by u s with
hm, u i < a and +
be
the
face
generated
by u s with hm, u i a . Show
that every u can be written uniquely as u = (1 t)u+ + tu where u+ +

,
+
u
and
0
1.
Then
show
that
is
a
deformation
retract.
(d) Prove that AKD,m S \ AD,m is a deformation retract.
(e) Finally, prove (9.2.9) by applying Alexander duality to AD,m S.
This exercise was inspired by [76, Prop. 7.7.1] and [105, Sec. 4.4].
9.3. Vanishing Theorems II

Vanishing theorems play an important role in algebraic geometry. A glance at the
index of Lazarsfelds two-volume treatise [186] lists vanishing theorems due to
Bogomolov, Demailly, Fujita, Grauert-Riemenschneider,
Griffiths, Kawamata-Viehweg, Kodaira, Kollar, Le Potier,
Manivel, Miyoka, Nadel, Nakano, and Serre.
We will explore toric versions of several of these results. In some cases, the toric
version is stronger, which is to be expected since toric varieties are so special.
421
Twisting. If D is a Cartier divisor and F a coherent sheaf on a normal variety X ,

then the twist of F by D is the sheaf
F (D) = F OX OX (D).
For example, if KX is a canonical divisor on X , then
X (D) = D OX OX (D) OX (KX ) OX OX (D) OX (KX + D).
This notation will be used some of the vanishing theorems stated below. However,
when we start working with non-Cartier divisors, we will drop the twist notation.
Kodaira and Nakano. Two of the earliest vanishing theorems are due to Kodaira
and Nakano. For an ample divisor D on a smooth projective variety X of dimension
n, Kodaira vanishing asserts that
H p (X , X (D)) = 0, p > 0,
and Nakano vanishing states that
H p (X , Xq (D)) = 0, p + q > n.
Nakanos theorem generalizes Kodairas since X = Xn in the smooth case.
In the toric case, we get more vanishing, in what has become known as the
Bott-Steenbrink-Danilov vanishing theorem.
Theorem 9.3.1 (Bott-Steenbrink-Danilov Vanishing). Let D be an ample divisor
on a projective toric variety X . Then for all p > 0 and q 0, we have
b q (D)) = 0.
H p (X ,
X
Proof. We give a proof due to Fujino [101] that uses the morphism : X X
from Lemma 9.2.6. We can assume that D is torus-invariant with support function
D : NR R. Let L = OX (D) be the associated line bundle.
The morphism has two key properties. The first concerns the pullback
OX (D). Proposition 6.2.7 implies that OX (D) comes from a divisor whose
support function is D . Since is multiplication by , D is the support
function of D. Hence
L = OX (D) OX (D) OX (D) = L .
The second key property of is that there is a split injection
(9.3.1)
We will assume this for now.
b q
bq .
X
X
Given these properties of , the theorem follows easily. Tensoring (9.3.1) with
L gives a split injection
b q X L
b q X L .
X
X
422
b q X L ) (this is the projection formula from

b q X L (
Since
X
X
Exercise 9.2.3) and L L , we obtain a split injection

b q X L ).
b q X L (
X
X
As in the discussion leading up to Lemma 9.2.6, this gives the injection

b q X L ) H p (X ,
b q X L ).
H p (X ,
X
X
When p > 0, the right-hand side vanishes for sufficiently large by Serre vanishing
(Theorem 9.0.6). Hence the left-hand side also vanishes for p > 0, which is what
we want.
It remains to prove (9.3.1). If we set D = 0 in the proof of Lemma 9.2.6, we
get a split injection i : OX OX . Recall that locally,
OX looks like OX , with module structure is given by m a = m a.
i sends m to m .
The splitting r : OX OX is defined by

(
0
m
/ M
r( m ) =
m
m = m , m M.
If we tensor i : OX OX with
Vq
(9.3.2)
M Z OX
together with a splitting map
(9.3.3)
Vq
Vq
M, we get a sheaf homomorphism

V
q M Z OX

V
M Z OX q M Z OX .
Thus (9.3.2) is a split injection.

V
b q is a subsheaf
b q sits inside q M Z OX , so that
By Theorem 8.2.16,
X
X

Vq
M Z OX . These subsheaves relate to (9.3.2) and (9.3.3) as follows.
of
Over U , Theorem 8.2.18 implies
M Vq
M Vq
V
m
bq )
(U ,
V
(m)
MC m = (U , q M Z OX ),
X
m M
m M
where V (m) MC is spanned by the lattice points lying in the same minimal face
of as m. Then over U , we have:
b q with module structure m a = m a.
b q looks like

X
The map (9.3.2) takes

Vq
Vq
V (m) m to
V (m) m =
Vq
bq )
V (m) m (U ,
X
since V (m) = V (m). Thus (9.3.2) induces a map (9.3.1).
The map (9.3.2) takes

Vq
Vq
423
V (m) m to 0 or, when m = m , to
V (m) m =
Vq
bq )
V (m ) m (U ,
X
bq .
bq
since V (m) = V (m ). Thus (9.3.3) induces a map
X
X
This gives the desired split injection (9.3.1).
Although Theorem 9.3.1 generalizes the vanishing theorems of Kodaira and

Nakano, its name Bott-Steenbrink-Danilov reflects the more special vanishing
that happens for projective spaces (Bott [40]), weighted projective spaces (Steenbrink [260]), and projective toric varieties (Danilov [76], stated without proof).
Theorem 9.3.1 was first proved by Batyrev and Cox [19] in the simplicial case and
by Buch, Thomsen, Lauritzen and Mehta [58] in general. Further proofs have been
given by Fujino [101] (noted above) and Mustata [212].
The Simplicial Case. When X is simplicial, there is another vanishing theorem
bq .
involving the sheaves
X
Theorem 9.3.2. If X is a complete simplicial toric variety, then
b q ) = 0 for all p 6= q.
H p (X ,
X
Proof. Since X is simplicial, we have the Ishida complex

(9.3.4)
b q K 0 (, q) K 1 (, q) K q (, q) 0
0
X
which is exact by Theorem 8.2.19. Recall that

M Vq j
K j (, q) =
( M) Z OV ()
( j)
and V () = O() X is the orbit closure corresponding to . Since V () is

a toric variety, its structure sheaf has vanishing higher cohomology by Demazure
vanishing, so that H p (X , K j (, q)) = 0 for p > 0. Since the above sequence has
q + 1 terms with vanishing higher cohomology, an easy argument using the long
b q ) = 0 for p > q (Exercise 9.3.1).
exact sequence in cohomology implies H p (X ,
X
Now suppose p < q. Since X is simplicial, the version of Serre duality given
in Theorem 9.2.10 implies
b nq )
b q ) H np (X ,
H p (X ,
X
X
The right-hand side vanishes since n p > n q, and the result follows.
When X complete but not necessarily simplicial, Danilov [76, Cor. 12.7]
b q ) = 0 when q > p.
proves that H p (X ,
X
b q (D), where D is a
In [198], Mavlyutov discusses a vanishing theorem for
X
nef Cartier divisor. His result goes as follows.
424
Theorem 9.3.3. Let X be a complete simplicial toric variety. If D is a nef Cartier

divisor on X , then
b q (D)) = 0
H p (X ,
X
whenever p > q or q > p + dim PD .
The proof for p > q is relatively easy (Exercise 9.3.2). Note that Theorem 9.3.2
b q (D))
is the case D = 0 of Theorem 9.3.3. The paper [198] computes H p (X ,
X
explicitly for all p, q.
Q-Weil Divisors. A Q-Weil divisor or D on a normal variety X is a formal Q-linear
combination of prime divisors. Thus a positive integer multiple of D is an ordinary
Weil divisor, often called integral in this context. A Q-Weil divisor is Q-Cartier if
some positive multiple is integral and Cartier. In the literature (see [186, 1.1.4]),
Q-Cartier Q-Weil divisors are often called Q-divisors. These divisors and their
close cousins, R-divisors, are essential tools in modern algebraic geometry.
For x R, x is the greatest
P integer x and x is the least integer x. Then,
given a Q-Weil divisor D = i ai Di , we get integral Weil divisors
P
D = i ai Di (the round down of D)
P
D = i ai Di (the round up of D).
We now prove an injectivity lemma for Q-Weil divisors due to Fujino [101].
Lemma 9.3.4. Let D be a Q-Weil divisor on a toric variety X and let > 0 be an
integer such that D is integral. Then for all p 0 there is an injection
H p (X , OX (D)) H p (X , OX (D)).
P
Proof. We adapt the proof of Lemma 9.2.6 to our situation. Write D = a D ,
P
where a = b + for b Z and 0 < 1. Thus D = b D . With as
above, we claim that : X X from Lemma 9.2.6 gives a split injection
OX (D) OX (D).
(9.3.5)
Assuming (9.3.5), the lemma follows from the remarks leading up to Lemma 9.2.6.
It remains to prove the existence of a split injection (9.3.5). Take an affine open
subset U X and consider
P = {m MR | hm, u i a for all (1)}.
Then
(U , OX (D)) =
mP1 M
C m , (U , OX (D)) =
mP M
C m,
425
where the first equality uses a = b + , 0 < 1. Since P M = (P1 M),

the map m 7 m induces an inclusion
M
M
C m
C m.
mP1 M
mP M
This is a C[ M]-module homomorphism, provided that the right-hand side has

module structure m a = m a, and the usual formula for the splitting map r is
also a C[ M]-module homomorphism. From here, we get the required split
injection (9.3.5) without difficulty.

Here are Q-divisor versions of Demazure and Batyrev-Borisov vanishing.
Theorem 9.3.5. Let D be a Q-Cartier Q-Weil divisor on a toric variety X .
(a) If || is convex and D is nef, then
H p (X , OX (D)) = 0 for all p > 0.
(b) If is complete and D is nef, then

H p (X , OX (D)) = 0 for all p 6= dim PD .
Proof. Pick > 0 with D Cartier. For part (a), H p (X , OX (D)) = 0 for p > 0
by Theorem 9.2.3. The desired vanishing follows immediately from Lemma 9.3.4.
For part (b), replacing D with D in Lemma 9.3.4 gives
H p (X , OX (D)) = H p (X , OX (D)) H p (X , OX (D)),
and then we are done by Theorem 9.2.7.
Here is an example that uses part (a) of Theorem 9.3.5 to extend Demazure
vanishing beyond nef Q-Cartier divisors.
Example 9.3.6. The fan for the Hirzebruch surface Hr has minimal generators
u1 = (1, r), u2 = (0, 1), u3 = (1, 0), u4 = (0, 1), giving divisors D1 , . . . , D4 . By
Example 6.1.16,
Pic(Hr )R {aD3 + bD4 | a, b R}
and the nef cone is generated by D3 and D4 . Since D1 D3 and D2 D4 rD3 , it
follows that a Q-Weil divisor D = a1 D1 + + a4 D4 is nef if and only
(9.3.6)
a1 + a3 ra2
and
a2 + a4 0.
We will show that the divisors aD3 + bD4 , a, b 1, have vanishing higher
cohomology when r > 0. Given such a divisor, pick a rational number 0 < < 12
and consider the Q-Weil divisor
D = 2D1 + r D2 + (a + 1 )D3 + (b + 1 r )D4 .
This satisfies (9.3.6) and hence is nef. Then D = aD3 + bD4 has vanishing higher
cohomology by Theorem 9.3.5. Taking a = 1 or b = 1, we get non-nef divisors
whose higher cohomology vanishes.
426
Lemma 9.3.4 also leads to the following result due to Mustata [212].
Theorem 9.3.7. Let X be a projective toric variety and let 1 , . . . , r (1) be
distinct. Then for p > 0 and any ample Cartier divisor D on X , we have
H p (X , OX (D D1 Dr )) = 0.
Proof. Let B = D1 + + Dr . Since D is ample, Serre vanishing (Theorem 9.0.6)
implies that we can find > 0 such that H p (X , X , OX (D B)) = 0 for p > 0.
Now let E = D 1B. Since E = D B, the inclusion
H p (X , OX (E)) H p (X , OX (E))
from Lemma 9.3.4 implies
H p (X , OX (D B)) H p (X , OX (D B)) = 0, p > 0.
Here is a non-Q-Cartier divisor with vanishing higher cohomology.

Example 9.3.8. Consider the complete fan in R3 shown in Figure 5. It has
z
3
2
y
x
0
Figure 5. A complete fan in R3
divisors D0 , . . . , D4 corresponding to 0 , . . . , 4 . In Exercise 9.3.3 you will show

that for a Q-Weil divisor D = a0 D0 + + a4 D4 , we have:
D is Q-Cartier if and only a1 + a3 = a2 + a4 .
D is Q-Cartier and nef if and only a1 + a3 = a2 + a4 0.
In particular, D = D3 + D4 is Q-Cartier and nef, so that D4 = D D3 has

vanishing higher cohomology by Theorem 9.3.7. Yet D4 is not Q-Cartier.
427
Iitaka Dimension and Big Divisors. Given a nef Cartier divisor D on a complete
toric variety X and an integer > 0, the global sections W = H 0 (X , OX (D))
give a morphism W : X P(W ) as in Lemma 6.0.28. Also recall that since D
is nef, its polytope PD is a lattice polytope. The map W and the polytope PD are
related as follows.
Lemma 9.3.9. For 0, the image of W is isomorphic to the toric variety of
PD . In particular, dim W (X ) = dim PD for 0.
Proof. This follows from Proposition 6.2.8 (Exercise 9.3.4).
This situation is a special case of the definition of the Iitaka dimension (X , D)

(see [186, Def. 2.1.3]) of a Cartier divisor D on a complete irreducible variety X . In
this terminology, Lemma 9.3.9 implies that a nef Cartier divisor D on a complete
toric variety X has Iitaka dimension
(X , D) = dim PD .
In general, a Cartier divisor is big if it has maximal Iitaka dimension, which for a
nef Cartier divisor D on a complete toric variety X means
D is big dim PD = dim X .
It should be clear what it means for a Q-Cartier Q-Weil divisor to be big and nef.
Kawamata-Viehweg. The classic version of Kodaira vanishing can be stated as
H p (X , OX (KX + D)) = 0 for all p > 0
when D is an ample line bundle on a smooth projective variety X . In the 1982,
Kawamata and Viehweg independently weakened the hypotheses on D. Here is the
toric version of their result.
Theorem 9.3.10 (Toric Kawamata-Viehweg). Let X be a complete toric variety
and let D be a Q-Cartier Q-Weil divisor on X that is big and nef. Then
H p (X , OX (KX + D)) = 0 for all p > 0.
Proof. Let n = dim X . When D is Cartier, Serre duality implies
H p (X , OX (KX + D)) H np (X , OX (D)).
Since D is nef, the latter vanishes for n p 6= dim PD by Theorem 9.2.7, and since
D is big, the condition on p becomes n p 6= n, i.e, p 6= 0.
P
In general, write D = a D where a = b for b Z and 0 < 1.
Pick such that D is Cartier and 0 < + 1 < 1 for all . Let E = D + 1 KX .
Since b 1 = b 1, we have E = D + KX . Thus the inclusion
H p (X , OX (E)) H p (X , OX (E))
428
from Lemma 9.3.4 implies

H p (X , OX (KX + D)) H p (X , OX (KX + D)).
Then we are done by the Cartier case already proved.
Another approach to Theorem 9.3.10 is to prove a version of Serre duality that

implies H p (X , OX (KX + D)) H np (X , OX (D)) (Exercise 9.3.5).
Grauert-Riemenschneider. Our next vanishing result features the higher direct
images of the canonical sheaf. We will need the following preliminary result.
Proposition 9.3.11. Let be a simplicial fan whose support || is strongly convex.
Then H p (X , X ) = 0 for all p > 0.
P
Proof. Fix p 1 and m M. Theorem 9.1.3 with D = KX = D implies
where
e p1 (VD,m , C),
H p (X , X )m = H p (X , OX (D))m H
VD,m =
Conv(u
Conv(u
| (1), hm, u i < 1)
| (1), hm, u i 0).
The last equality follows since hm, u i Z. Now consider the set
W = {u || \ {0} | hm, ui 0} = (|| \ {0}) {u NR | hm, ui 0},
and observe that W is convex since || is strongly convex. Note also that
(9.3.7)
VD,m W .
The proposition will follow once we prove that (9.3.7) is a deformation retract.
Given with W 6= , we construct r : W VD,m as follows.
The assumption W 6= implies that
A = { (1) | hm, u i 0} =
6 .
P
Since is simplicial, u can be uniquely written u = (1) u , 0, and
P
P
P
when u W , define r (u) = ( A )1 A u . The sum A is
nonzero since u W , so that r (u) Conv(u | A) VD,m . It is also easy
to see that r is compatible with r whenever is a face of . Thus we get a map
r : W VD,m , which is the desired retraction by Exercise 9.3.6.

Here is a toric version of Grauert-Riemenschneider vanishing.
Theorem 9.3.12 (Toric Grauert-Riemenschneider Vanishing). Let : X X be
a surjective proper toric morphism between toric varieties of the same dimension.
If X is simplicial, then
R p X = 0 for all p > 0.
If in addition is birational, then X X .
429
Proof. Our hypothesis on implies that the associated lattice map : N N

induces an isomorphism R : NR NR such that is the inverse image of .
1
Thus, given , we see that R () is strongly convex. This is the support of
the fan of 1 (U ), so that
H p (1 (U ), X ) = 0
for p > 0 by Proposition 9.3.11. Then R p X = 0 for p > 0 by Proposition 9.0.7.
The final assertion of the theorem is an easy consequence of Theorem 8.2.15
(Exercise 9.3.7).

Most versions of Theorem 9.3.12 in the literature assume that X is smooth.
Other Vanishing Theorems. There are many more toric vanishing theorems. Both
Fujino [101] and Mustata [212] state general vanishing theorems that imply versions of Theorems 9.3.1, 9.3.5, 9.3.7 and 9.3.10. Further vanishing results can be
found in [102] and [226].
Exercises for 9.3.
9.3.1. A sheaf G on a variety X is acyclic if H p (X, G ) = 0 for all p > 0. Now suppose that
we have an exact sequence of sheaves
0 F G0 G1 Gr 0,
where G0 , . . . , Gr are acyclic. Prove H p (X, F ) = 0 for all p > r by induction on r.

9.3.2. Use (9.3.4) and the previous exercise to prove Theorem 9.3.3 when p > q.
9.3.3. Let D0 , . . . , D4 be the divisors from Example 9.3.8 and consider a Q-Weil divisor
D = a 0 D0 + + a 4 D4 .
(a) Show that D0 2D3 + 2D4 , D1 D3 , D2 D4 .
(b) Show that D is Q-Cartier if and only a1 + a3 = a2 + a4 .

(c) Show that D is Q-Cartier and nef if and only a1 + a3 = a2 + a4 0.
9.3.5. Let D be a Q-Cartier Q-Weil divisor on a complete toric variety X .
(a) Pick > 0 such that D is Cartier and consider OX (D) OX (D) from (9.3.5).
Use this to prove that OX (D) is MCM. Hint: Replace : N N with N N as
in the proof of Theorem 9.2.10.
(b) Adapt the proof of Theorem 9.2.10 to show that
(c) Prove H p (X , OX (D)) H np (X , OX (KX + D)). Hint: D = D.
(d) Prove Theorem 9.3.10 using part (c) and Theorem 9.3.5.
9.3.6. Consider the map r : W VD,m defined in the proof of Proposition 9.3.11.
(a) Prove that these maps patch to give a retraction r : W VD,m .
430
(b) Prove that when regarded as a map from W to itself, r is homotopic to the identity.
Then formulate and prove a similar result for r.
9.3.7. Let : X X be a proper birational toric morphism.
(a) Prove that : N N is an isomorphism.
(b) Let 0 be the fan in NR consisting of the cones R () for . Prove that is a
refinement of 0 .
(c) Complete the proof of Theorem 9.3.12.
P
9.3.8. Given a toric variety X , let D be a Weil divisor and E = a D a Q-Weil divisor
such that 0 a 1 for all and E is integral for some integer > 0. Prove that there is
an injection
H p (X , OX (D)) H p (X , OX (D + (D + E))) for all p 0.
This is a strengthened version of Theorem 0.1 from [212]. Hint: As suggested by [101],
apply Lemma 9.3.4 to the Q-Weil divisor D + +1

E.
9.4. Applications to Lattice Polytopes

In this section we use vanishing theorems to study lattice polytopes.
The Euler Characteristic of a Sheaf . Let F be a coherent sheaf on a complete
variety X . Its Euler characteristic (F ) is defined to be the alternating sum
X
(F ) =
(1) p dim H p (X , F ).
p0
Our hypotheses on X and F guarantee that dim H p (X , F ) < for all p, and
H p (X , F ) = 0 for p > dim X by [131, Thm. III.2.7]. Hence (F ) is a well-defined
integer. The Euler characteristic satisfies
(9.4.1)
(G ) = (F ) + (H )
whenever we have an exact sequence 0 F G H 0 of coherent sheaves.

This follows from the long exact sequence in cohomology (Exercise 9.4.1).
Given a line bundle L on X , define
OX OX L > 0
{z
}
|
times
L =
(L )()
<0
OX
= 0,
where as usual L = HomOX (L , OX ). A basic result [176] states that

(F OX L ),
Z,
431
is a polynomial in , called the Hilbert polynomial. Exercises 9.4.2 and 9.4.3 sketch
a proof in the ample case. When L = O(D) for a Cartier divisor D, the Hilbert
polynomial is written
(F (D)),
Z,
via the twisting convention introduced in 9.2.

Example 9.4.1. We will compute (OPn ()). When 0, Example 9.2.4 implies
that H p (P n , OPn ()) = 0 for p > 0, so that

(9.4.2)
(OPn ()) = dim H 0 (P n , OPn ()) = |n Zn | = +n
n .
The second equality follows from Proposition 4.3.3 since the polytope associated
to D0 is n , where n is the standard n-simplex from Example 4.3.6. You will
prove the last equality in Exercise 9.4.4. When < 0, Example 9.2.8 implies
(9.4.3)
(OPn ()) = (1)n dim H n (P n , OPn ()) = (1)n |Int(n ) Zn |

= (1)n 1
,
n
where the last equality uses Exercise 9.4.4.

Now consider the polynomial
(x + n)(x + n 1) (x + 1)
Q[x]
n!

and observe that p() = +n
when N.
n
p(x) =
We claim that p() = (OPn ()) for all Z. For 0 this follows easily
from (9.4.2). When < 0, note that

p() = (1)n 1
n
(Exercise 9.4.4). Then p() = (OPn ()) for < 0 by (9.4.3).
Note also that when > 0, (9.4.2) and (9.4.3) imply that
p() = |n Zn |
p() = (1)n |Int(n ) Zn |.

We will see below that this is a special case of Ehrhart reciprocity.
The Ehrhart Polynomial. Given a full dimensional lattice polytope P MR , the

functions
L(P) = |P M|
L (P) = |Int(P) M|
count lattice points in P or in its interior. For example, if > 0 is an integer, then
(9.4.2) and (9.4.3) imply that L(n ) = +n
and L (n ) = 1
n
n .
432
Theorem 9.4.2 (Ehrhart Reciprocity). Let P MR Rn be a full dimensional

lattice polytope. There is a polynomial EhrP (x) Q[x] such that if N, then
EhrP () = L(P).
Furthermore, if N is positive, then
EhrP () = (1)n L (P).
Proof. Let XP be the toric variety of P and DP the associated ample divisor. We
will show that the Hilbert polynomial
EhrP () = (OXP (DP ))
has the desired properties. The case when 0 is easy since DP is basepoint free
by Proposition 6.1.10. Thus
(OXP (DP )) = dim H 0 (XP , OXP (DP )) = |P M| = L(P)
by Demazure vanishing (Theorem 9.2.3) and Example 4.3.7.
Now assume > 0. Then PDP = P is full dimensional, hence
(OXP (DP )) = (1)n dim H n (XP , OXP (DP )) = (1)n |Int(P) M|
= (1)n L (P)
by Batyrev-Borisov vanishing (Theorem 9.2.7).
The polynomial EhrP in Theorem 9.4.2 is called the Ehrhart polynomial of P.

An elementary approach to Ehrhart polynomials and Ehrhart Reciprocity can be
found in [22].
When P is very ample, the Ehrhart polynomial has a nice interpretation. The
very ample divisor DP on XP gives a projective embedding i : XP P s1 , where
s = |P M|. Its homogeneous ideal I(XP ) C[x1 , . . . , xs ] gives the homogeneous
coordinate ring
C[XP ] = C[x1 , . . . , xs ]/I(XP ).
This graded ring has a Hilbert function
7 dim C[XP ] ,
0,
which is a polynomial for 0 (see [70, Ch. 6, 4]). This is the Hilbert polynomial
of XP P s1 ,
Proposition 9.4.3. If P is very ample, then the Ehrhart polynomial EhrP equals the
Hilbert polynomial of the toric variety XP under the projective embedding given by
the very ample divisor DP .
433
Proof. Consider the exact sequence

0 IXP OPs1 OXP 0.
Since OXP (DP ) is the restriction of OPs1 (1) to XP , tensoring with OPs1 () gives
an exact sequence
0 IXP () OPs1 () OXP (DP ) 0.
In Exercise 9.4.5 you will show that the resulting long exact sequence is
0 I(XP ) C[x1 , . . . , xs ] H 0 (XP , OXP (DP )) H 1 (P s1 , IXP ()) .
The H 1 term vanishes for 0 by Serre vanishing (Theorem 9.0.6). Hence for
large, we get an isomorphism
C[XP ] H 0 (XP , OXP (DP )).
This implies that the Hilbert polynomial of XP is the Ehrhart polynomial of P.
We can describe the degree and leading coefficient of the Ehrhart polynomial.
For the leading coefficient, we use the normalized volume in MR . Let e1 , . . . , en
be a basis of M and consider the simplex n = Conv(0, e1 , . . . , en ) MR . Then
the normalized volume is the usual n-dimensional Lebesgue
measure, scaled so
P
that n has volume equal to 1. Thus the unit cube { ni=1 i ei | 0 i 1} has
normalized volume n!.
Given a full dimensional lattice polytope P MR , its normalized volume is
denoted Vol(P) and is computed by the limit
(9.4.4)
Vol(P)
L(P)
= lim
.
n
n!
This is proved in many places, including [22, Lem. 3.19]. Since L(P) is given
by the polynomial EhrP (), it follows easily that EhrP has degree n and its leading
coefficient Vol(P)/n! (Exercise 9.4.7).
Here is a classic application of these ideas.
Example 9.4.4. Let P R2 be a lattice polygon with Ehrhart polynomial EhrP .
The leading coefficient of EhrP is 21 Vol(P), which is the usual Euclidean area
Area(P) since the normalized volume of the unit square is 2. The constant term is
also easy to compute, since EhrP (0) = L(0 P) = 1 by Theorem 9.4.2. Thus
EhrP (x) = Area(P) x2 + 12 B x + 1,
where B is yet to be determined. The reason for the
1
2
will soon become clear.
By Ehrhart reciprocity, we have

(9.4.5)
Area(P) + 12 B + 1 = EhrP (1) = L(P)

Area(P) 12 B + 1 = EhrP (1) = (1)2 L (P) = L (P).
434
Solving for B gives B = L(P) L (P) = |P M|, where P is the boundary of P.

Thus the Ehrhart polynomial of P is
EhrP (x) = Area(P) x2 + 21 |P M| x + 1.
Furthermore, solving the bottom equation of (9.4.5) for the area gives
Area(P) = L (P) + 12 |P M| 1.
This equation, called Picks formula, shows how to compute the area of a lattice
polygon in terms of its lattice points.
Example 9.4.5. Let P = Conv{(0, 0, 0), (1, 0, 0), (0, 1, 0), (1, 1, 3)} be the lattice
simplex shown in Figure 3 of Example 2.2.4. Since the normalized volume is 3 and
EhrP (0) = L(0 P) = 1, the Ehrhart polynomial is EhrP (x) = 12 x 3 + ax 2 + bx + 1.
We noted in Example 2.2.8 that the only lattice points of P are its vertices. By
Theorem 9.4.2, we obtain
EhrP (1) = 12 + a + b + 1 = L(P) = 4
EhrP (1) =
21
+ a b + 1 = (1) L (P) = 0
(4 lattice points)
(no interior lattice points)
Hence a = 1, b = 32 , so that EhrP (x) = 21 x 3 + x 2 + 23 x + 1. In Example B.3.3 we

will explain how to compute EhrP (x) using Normaliz [57].
Examples 9.4.4 and 9.4.5 used EhrP (0) = L(0 P) = 1. The latter equality is
obvious since 0 P = {0}. However, the equality EhrP (0) = L(0 P) (proved in
Theorem 9.4.2) is more subtle since it can fail when we replace the polytope P
with a polytopal complex (a collection of polytopes where the intersection of any
two is a face of each). Here is a simple example.
Example 9.4.6. Consider the boundary 2 of the 2-simplex 2 R2 . Thus 2
consists of three line segments. One easily computes that
L( 2 ) = |( 2 ) Z2 | = 3
when > 0 is an integer. This gives the Ehrhart polynomial Ehr2 (x) = 3x with
the property that Ehr2 () = L( 2 ) for integers > 0. However,
Ehr2 (0) = 3 0 = 0 6= L(0 2 ) = L({0}) = 1.
Further examples (and an explanation) will be given in Execise 9.4.8.
bp
The p-Ehrhart Polynomials. Following Materov [193], we use the sheaves
XP
to generalize the Ehrhart polynomial when the lattice polytope P is simple. Recall
that a full dimensional lattice polytope P MR Rn is simple if every vertex is the
intersection of exactly n facets. Hence each vertex has n facet normals that generate the corresponding cone in the normal fan P . It follows that P is simplicial
whenever P is simple. Hence the toric variety XP is simplicial.
435
We proved earlier that the Ehrhart polynomial EhrP (x) of P satisfies

EhrP () = (OXP (DP )), Z,
where DP is the ample divisor coming from P. More generally, given an integer
b p (DP )) is a polynomial function of .
0 p n, the Euler characteristic (
XP
Then the p-Ehrhart polynomial of P, denoted EhrPp (x), is the unique polynomial in
Q[x] that satisfies
b p (DP )), Z.
Ehr p () = (
P
XP
Note that EhrP0 (x) is the ordinary Ehrhart polynomial EhrP (x). To state the
properties of EhrPp (x), we will use the following notation:
P(i) = {Q | Q is an i-dimensional face of P}
fi = |P(i)| = # i-dimensional faces of P

n
X
ip i
(1)
hp =
fi .
p
i=p
The fi are the face numbers of P. Furthermore, if Q is a face of P, we define

L(Q) = |Q M|
L (Q) = |Relint(Q) M|.

These invariants are related to the p-Ehrhart polynomials by the following result of
Materov [193].
Theorem 9.4.7. Let XP be the toric variety of a full dimensional simple lattice
polytope P MR Rn .
(a) If > 0 is an integer, then
EhrPp () =
n X
X
i
i=p
L (Q).
QP(i)
(b) If 0 is an integer, then

EhrPp ()

p
X
i ni
(1)
=
n p
i=0
QP(ni)
(c) If 0 p n, then
EhrPp (x) = (1)n EhrPnp (x).
(d) If 0 p n, then
EhrPp (0) = (1) p h p .
L(Q).
436
We will defer the proof for now. Theorem 9.4.7 has some nice consequences.
First, setting x = 0 in part (c) and using part (d), we see that
(1) p h p = (1)n (1)np hnp .
This proves that
h p = hnp
(9.4.6)
for 0 p n. These are called the Dehn-Sommerville equations.
Also, if we write EhrPp () using part (b) of the theorem and EhrPnp () using
part (a), then part (c) gives the following formulas for > 0:

X
p
X
p
i ni
(1)
EhrP () =
L(Q),
n p
i=0
QP(ni)
EhrPp () = (1)n EhrPnp () = (1)n
X
n
X
i
L (Q).
n p
i=np
QP(i)
For p = 0 and > 0, these equations reduce to Ehrhart reciprocity:

EhrP0 () = L(P)
EhrP0 () = (1)n L (P).
Thus Theorem 9.4.7 simultaneously generalizes the Dehn-Sommerville equations
and Ehrhart reciprocity. The proof of the theorem will use the following lemma.
Lemma 9.4.8. Let P MR Rn be a full dimensional lattice polytope with toric
variety XP and ample divisor DP . Given an integer > 0 and m (P) M, let
VP (m) be the subspace of MC such that VP (m) + m is the smallest affine subspace
of MC containing the minimal face of P containing m. Then
M Vp
b p (DP )) =
H 0 (XP ,
VP (m) m .
XP
m(P)M
Proof. We adapt the strategy used in the proof of Proposition 8.2.18. To simplify
notation, set D = DP . Using M M and tensoring (8.2.6) with OXP (D), we
obtain the exact sequence
V
V
b p (D) p M Z OX (D) L p1 M Z OD (D).
0
P
XP
Take global sections

the graded piece for m M. Since
Lover XP and consider
m
0
H (XP , OXP (D)) = m(P)M C , we get the exact sequence
V
V
b p (D))m p MC L p1 M Z H 0 (XP , OD (D))m
0 H 0 (XP ,
XP
when m (P) M. To compute H 0 (XP , OD (D))m , recall from Proposition 4.0.28

that we have an exact sequence
0 OXP (D ) OXP OD 0.
437
Tensor this with OXP (D) and take global sections to obtain
0 H 0 (XP , OXP (D D )) H 0 (XP , OXP (D)) H 0 (XP , OD (D)) 0,
where the exactness on right follows from H 1 (XP , OXP (D DP

)) = 0 courtesy of
Theorem 9.3.7. Comparing the polytopes of D D and D = a D shows that
(
C hm, u i = a
0
H (XP , OD (D))m =
0 otherwise.
Let F be the facet of P corresponding to . For m P, hm, u i = a if and
only if m F . Hence we get an exact sequence
V
V p1
p L
b p (D))m p MC
0 H 0 (XP ,
MC ,
XP
mF
where p is a sum of contraction maps iu defined in (8.2.7).
V
Thus p MC is in the kernel of p if and only if iu () = 0 for all with

V
m F . We know from Exercise 8.2.8 that iu () = 0 if and only if p ( )C .
It follows that the kernel of p in (8.2.9) is the intersection

\ Vp
V T
(9.4.7)
( )C = p mF ( )C .
Since F =
mF
is the minimal face of P containing m, one sees easily that

T
SpanR (m0 m M | m0 F) = mF .
mF F
Thus VP (m) =
mF (
) .
C
This and (9.4.7) imply ker( p ) =
We are now ready to prove our main result.
Vp
VP (m).
Proof of Theorem 9.4.7. We begin with part (a). Lemma 9.4.8 implies that
X dim VP (m)
p
0
b
dim H (XP , XP (DP )) =
p
m(P)M
when > 0. Given a face Q of P and m (P) M, note that
m Relint(Q) M Q is the minimal face of P containing m.
When this happens, we have dim Q = dim VP (m). Since the i-dimensional faces
of P are Q for Q P(i), we obtain
n X
X
X i X
i
p
0
b
L (Q) =
L (Q).
dim H (XP , XP (DP )) =
p
p
i0
QP(i)
i=p
QP(i)
b p (DP )) = 0 for q > 0 by Theorem 9.3.1. Hence

We also have H q (XP ,
XP
n X
X
i
p
p
p
0
b (DP )) = dim H (XP ,
b (DP )) =
EhrP () = (
L (Q),
XP
XP
p
i=p
QP(i)
438
which proves part (a).

Now consider EhrPp (0). If > 0 and Q P(i), then L (Q) = (1)i EhrQ ()
by Ehrhart reciprocity. Since the previous display holds for all > 0, we obtain the
polynomial identity
n X
X
i
p
(9.4.8)
EhrP (x) =
(1)i EhrQ (x).
p
i=p
QP(i)
Setting x = 0 gives
EhrPp (0) =
n
X
i=p
(1)i
X
i
EhrQ (0) = (1) p h p
p
QP(i)
by the definition of h p , and part (d) follows.

For part (c), we use Serre duality as given in Theorem 9.2.10, which implies
b p (DP )) H nq (X ,
b np (DP ))
H q (X ,
XP
XP
(9.4.9)
since DP is Cartier for any Z. This easily implies
b p (DP )) = (1)n (
b np (DP ))
(
XP
XP
(Exercise 9.4.9). Thus EhrPp () = (1)n EhrPnp () for Z, proving part (d).
Finally, for part (b), take 0 and consider
EhrPp ()
= (1)
EhrPnp () =
X
n
X
i
(1)i EhrQ (),
(1)
n p
n
i=np
QP(i)
where the first equality uses part (d) and the second uses (9.4.8) with p replaced by
n p. Since 0 implies EhrQ (Q) = L(Q), we obtain

X
X
p
n
X
X
i
p
ni
j n j
(1)
(1)
L(Q) =
L(Q),
EhrP () =
n p
n p
i=np
QP(i)
j=0
QP(i)
where the last equality follows by setting j = n i. This completes the proof.
Identities equivalent to Theorem 9.4.7 were discovered by McMullen in 1977

in a very different context. See [22, Ch. 5] for details and references, including a
non-toric version of Theorem 9.4.7 in [22, Ex. 5.85.10].
Examples. We first give a classic example of the Dehn-Sommerville equations.
Example 9.4.9. Let P be a simple 3-dimensional lattice polytope in R3 . The DehnSommerville equations can be written
h3 = h0 , so f3 = f0 f1 + f2 f3
h2 = h1 , so f2 3 f3 = f1 2 f2 + 3 f3 .
439
Since f3 = 1 (P is the only 3-dimensional face of itself), we obtain

f0 f1 + f2 = 2
f1 = 3 f2 6.
The first equation Eulers celebrated formula V E + F = 2, which holds for any
3-dimensional polytope. The second seems more mysterious, but when combined
with the first reduces to 2 f1 = 3 f0 . This holds because is P is simpleevery vertex
meets three edges and every edge meets two vertices.
There is a version of the Dehn-Sommerville equations for the dual polytope

P NR , which is simplicial since P is simple (Exercise 9.4.10). More on the
Dehn-Sommerville equations and toric varieties can be found in [105, Sec. 5.6].
We also recommend [22, Ch. 5] and [281, Sec. 8.3].
We next give an application of Theorem 9.4.7.
Example 9.4.10. The standard n-simplex n has P n as its associated toric variety.
Given > 0, note that

X
n+1 1
L (n ) =
ni
i
Qn (i)
because
n+1
i+1
An n-simplex has
n+1
ni
faces of dimension i.
Each Q n (i) is lattice isomorphic to the standard i-simplex and hence has
1
interior lattice points by (9.4.3).
i
Then part (a) of Theorem 9.4.7 gives the formula

Ehrp n ()

n
X
i
n+1 1
i=p
Since
i
p
1
i
Ehrp n () =
1
p
p1
ip
ni
, this becomes

1 n+ p
1 X n+1 p1
=
,
ni
i p
p
p
i0
where the last equality uses the Vandermonde identity discussed in Exercise 9.4.11.
Hence

1 n+ p
p
0 n
dim H (P , Pn ()) =
.
p
This formula was first proved by Bott [40] in 1957.
440
b p has
Cohomology of p-Forms. Given a simple polytope P as above, the sheaf
XP
Euler characteristic
b p ) = Ehr p (0) = (1) p h p
(9.4.10)
(
XP
by Theorem 9.4.7. The factor of
(1) p
is explained as follows.
Theorem 9.4.11. Let XP be the toric variety of a full dimensional simple lattice
polytope in P MR Rn . Then
(
p
q
b ) = hp q = p
dim H (XP ,
XP
0 q 6= p.
Proof. The toric variety XP is simplicial, so that Theorem 9.3.2 applies to XP . This
b p ) = 0 for q 6= p. It follows that
shows that H q (XP ,
XP
b p ) = (1) p dim H p (XP ,
b p ).
(
XP
XP
The theorem follows by comparing this with (9.4.10).
b p ) = h p will be sketched in Exercise 9.4.12.

A different proof of dim H p (XP ,
XP
Theorem 9.4.11 has a nice application to the singular cohomology of XP because
of the the Hodge decomposition
M
b p ).
H q (XP ,
(9.4.11)
H k (XP , C)
XP
p+q=k
When XP is smooth, this is a classical factsee, for example, [125, p. 116]. In the
simplicial case, see [261] for a proof. Combined with Theorem 9.4.11, we obtain
(
h p k = 2p
k
(9.4.12)
dim H (XP , C) =
0 k odd.
We will give a topological proof of this formula in 12.4.
Exercises for 9.4.
9.4.1. Prove (9.4.1).
9.4.2. Here are some cases where it is easy to show that Euler characteristics give Hilbert
polynomials.
(a) Let F be a finitely generated graded module over the polynomial ring S = C[x0 , . . . , xn ].
By [70, Ch. 6, Thm. (3.8)], there is an exact sequence
0 Fr F1 F0 F 0,
where each Fi is a finite direct sum of modules for the form S(a). Now let F be a
coherent sheaf on P n . Prove that there is an exact sequence of sheaves
0 Fr F1 F0 F 0,
where each Fi is a finite direct sum of sheaves of the form OPn (a). Hint: Apply
Example 6.0.10 with X = P n .
441
(b) Use part (a) together with (9.4.1) and Example 9.4.1 to show that (F ()) is a polynomial in Z, where F () = F OPn OPn ().
(c) Let F be a coherent sheaf on a complete variety X and let D be a very ample divisor
on X. Use part (b) and Exercise 9.0.6 to show that (F (D)) is a polynomial in Z.
9.4.3. Let F be a coherent sheaf on a projective variety X, and let D be an ample divisor
on X. Thus there is k0 > 0 such that kD is very ample for all k k0 .
(a) Let a, k be integers with k k0 . Use Exercise 9.4.2 to show that there is a polynomial
pa,k (x) Q[x] such that pa,k () = (F ((a + k)D)) for all Z.
(b) Show that the polynomials pa,k (x/k) and pa,m (x/m) are equal when k, m k0 . Conclude that there is a polynomial pa (x) such that pa () = (F ((a + )D)) for all k0 .
(c) Show that the polynomials pa (x a) and pb (x b) are equal for any a, b Z and
conclude that there is a polynomial p(x) Q[x] such that p() = (F (D)) for all
Z.
9.4.4. In this exercise will always denote an integer.

(a) Prove that |(n ) Zn | = +n
for nonnegative. Hint: Lattice points in n give
n
monomials of degree in x0 , . . . , xn by Exercise 4.3.6. Here is a combinatorial proof.
Begin with a list of 1s of length + n. In n of the positions, convert the 1 to a 0. This
divides the remaining 1s into n + 1 groups. The number of elements in each group
gives the exponents of x0 , . . . , xn . See also [69, Ex. 13 of Ch. 9, 3].

(b) Prove that |Int(n ) Zn | = 1
for positive. Hint: Show that shifting the interior
n
lattice points by (1, . . . , 1) gives the lattice points in ( n 1)n.

(c) Prove that p(x) = (x + n)(x + n 1) (x + 1)/n! satisfies p() = (1)n 1
for
n
negative.
9.4.5. A projective variety X P n gives 0 IX OPn OX 0. After tensoring with

OPn (), we have an exact sequence 0 IX () OPn () OX () 0 whose long exact
sequence in cohomology begins
0 H 0 (P n , IX ()) H 0 (P n , OPn ()) H 0 (P n , OX ()) H 1 (P n , IX ()) .
We know from Example 4.3.6 that H 0 (P n , OPn ()) = S , where S = C[x0 , . . . , xn ] with the
standard grading. Also recall that as a sheaf on P n , OX () stands for i OX (), where
i : X P n is the inclusion map. It follows that H 0 (P n , OX ()) = H 0 (X, OX ()).
(a) Show that H 0 (P n , IX ()) = I(X) , where I(X) S is the ideal of X.
(b) Show that the Hilbert polynomial of the coordinate ring C[X] = S/I(X) is the Euler
characteristic (OX ()).
9.4.6. Recall that X P n is projectively normal if and only if its affine cone is normal.
By [131, Ex. II.5.14], this is equivalent to saying that X is normal and H 0 (P n , OPn ())
H 0 (X, OX ()) is surjective for all 0. Combine this with the previous exercise to show
that X is projectively normal if and only if X is normal and H 1 (P n , IX ()) = 0 for all 0.
See Example B.1.2 for a computational example.
9.4.7. Let p(x) be a polynomial such that lim p()/n exists and is nonzero.
(a) Prove that n = deg(p(x)) and that the above limit is the leading coefficient of p(x).
(b) Use part (a) to prove the assertion made following (9.4.4) concerning the degree and
leading coefficient of the Ehrhart polynomial.
442
9.4.8. Here we compute the Ehrhart polynomials of some simple polytopal complexes.
(a) As in Example 9.4.6, let 2 be the boundary of the standard simplex 2 R2 . Prove
that |( 2 ) Z2 | = 3, so that Ehr2 (x) = 3x, with constant term 0.
(b) Consider the butterfly B = Conv(0, e1 e2 ) Conv(0, e1 e2 ) with boundary B.
Prove that |( B) Z2 | = 8 1, so that EhrB (x) = 8x 1, with constant term 1.
(c) P
The Euler characteristic of a well-behaved topological space Z is defined to be e(Z) =
p
p
p (1) rankH (Z, Z). Show that e(2 ) = 0 and e(B) = 1.
The answer to part (c) is not a coincidencegiven any polytopal complex P whose vertices are lattice points, the constant term of its Ehrhart polynomial EhrP is e(P) (see
[189]). Since a polytope P is contractible, this gives a different way of seeing that the
constant term of EhrP is 1.
b p (DP )) = (1)n (
b np (DP )).
9.4.9. Use (9.4.9) to prove that (
XP
XP
9.4.10. Let P MR Rn be a full dimensional simplicial lattice polytope and define

n
X
ip i
simp
fni1 .
(1)
hp =
p
i=p
Rescaling and translating P does not affect the face numbers fi . So we may that assume the
origin is an interior point of P. Let P NR be the dual polytope of P. Use Exercise 2.3.4
and Proposition 2.3.8 to prove the following:
(a) P is simple.
P
(b) The face numbers fiP of P and fiP of P are related by fni1
= fiP .
simp
(c) h simp
= hnp
.
p
9.4.11. The goal of this exercise is to prove the identity used in Example 9.4.10.

P
(a) Prove the Vandermonde identity, which states that i+ j=k ai bj = a+b
for a, b N.
k
a
b
a+b
Hint: (1 + x) (1 + x) = (1 + x) .
p1

P
(b) Use part (a) to show that i0 n+1
= n+p
, as claimed in Example 9.4.10.
ni
ip
9.4.12. For a full dimensional simple polytope P MR Rn , Theorem 8.2.19 gives
b p K 0 (P , p) K 1 (P , p) K p (P , p) 0,
0
XP
L
V
i
where K (P , p) = P (i) pi ( M) Z OV () and V () XP is the closure of the
b p ).
orbit corresponding to P . You will use this to compute dim H p (XP ,
XP
Pp
p
j
j
b
(a) Show that (XP ) = j=0 (1) (K (P , p)).
(b) Use Theorem 9.2.3 to show that (K j (P , p)) = dim H 0 (XP , K j (P , p)).

n j
(c) Use Proposition 2.3.8 to show that dim H 0 (XP , K j (P , p)) = np
fn j . Hint: M
has rank n j for P ( j).
b p ) = hnp . Hint: Set j = n i.
(d) Use Theorem 9.3.2 to conclude that dim H p (XP ,
XP
9.4.13. When a polytope P has rational but not integral coordinates, the counting function
L(P) is almost a polynomial. Here you will study P = Conv(0, e1 , 21 e2 ) R2 .
9.5. Local Cohomology and the Total Coordinate Ring
(a) Given N, prove that

L(P) =
1 2
4 ++1
1 2
3
4 ++ 4
443
even
odd.
This is an example of a quasipolynomial. See [22, Sec. 3.7] for a discussion of the
Ehrhart quasipolynomial of a rational polytope.
(b) The weighted projective plane P(1, 1, 2) is given by the fan in R2 with minimal generators u0 = e1 2e2 , u1 = e1 , u2 = e2 . Show that X2P = P(1, 1, 2) with D2P = 2D0 ,
where D0 is the divisor corresponding to u0 . Also show that
(OP(1,1,2) (D0 )) = L(P).
The Euler characteristic is not a polynomial in . The reason is that D0 is not Cartier.
9.4.14. Let P MR Rn be a full dimensional lattice poltyope, not necessarily simple.
b p (DP )) for all Z.
Let EhrPp (x) be the unique polynomial satisfying EhrPp () = (
XP
(a) Prove that if > 0 in an integer, then
EhrPp () =
n X
X
i
i=p
L (Q).
QP(i)
Hint: Look at the hypotheses of Theorem 9.3.1 and Lemma 9.4.8.

(b) Prove that EhrP0 () = (1)n EhrPn (). Hint: Serre duality.
(c) Prove that EhrPp (0) = (1) p h p . Hint: Follow the proof of Theorem 9.4.7.
This shows that some parts of Theorem 9.4.7 hold for arbitrary lattice polytopes. In the
next exercise you will see that other parts can fail.
9.4.15. This exercise will show how things can go wrong for a non-simple polytope. Let
P = Conv(e1 , e2 , e3 ) R3 . Note that P is not simple, so that XP is not simplicial.
(a) Show that h1 6= h2 .
(b) Conclude that for this polytope, (9.4.9) cannot hold for all p, q, . So the version of
Serre duality stated in part (b) of Theorem 9.2.10 can fail for non-simplicial toric
varieties. Hint: The Dehn-Sommerville equations follow from Theorem 9.4.7.
9.4.16. Suppose that P R3 is a lattice simplex whose only lattice points are its vertices,
and let k be the normalized volume of P. Prove that 2P has k 1 interior lattice points.
Hint: Adapt Exercise 9.4.5.

In this section, we use local cohomology and Ext to compute the cohomology of a
coherent sheaf on a toric variety X . Our treatment is based on [91].
We first review the basics of local cohomology.
444
Local Cohomology. Let R be a finitely generated C-algebra, I R an ideal, and

M an R-module. First define
I (M) = {a M | I k a = 0 for some k N}.
This is the I-torsion submodule of M. One checks easily that M 7 I (M) is left
exact. Hence, just as we did for global sections of sheaves in 9.0, we get the
derived functors M 7 HIp (M) such that
HI0 (M) = I (M).
A short exact sequence 0 M N P 0 gives a long exact sequence

0 HI0 (M) HI0 (N) HI0 (P) HI1 (M) HI1 (N) HI1 (P) .
The Local Cech

Complex. As with sheaf cohomology, there is a Cech
complex
for local cohomology. The sheaf case used an affine open cover; here we use
generators of the ideal. More precisely, if I = h f1 , . . . , f i, let [] = {1, . . . , } be
the index set and as in 9.0, let [] p denote the set of all (p + 1)-tuples (i0 , . . . , i p )
of elements of I satisfying i0 < < i p . Also set f = ( f1 , . . . , f )
Given an R-module M, define
Cp (f, M) =
M fi0 fi p1 ,
(i0 ,...,i p1 )[] p1
where M fi0 fi p1 is the localization of M at fi0 fi p1 . An element of Cp (f, M) is

a function that assigns an element of M fi0 fi p1 to each (i0 , . . . , i p1 ) [] p1 .
Then define a differential
d p : Cp (f, M) Cp+1 (f, M)
by
p
X
(1)k (i0 , . . . , ibk , . . . , i p ),
d ()(i0 , . . . , i p ) =
p
k=0
where we regard (i0 , . . . , ibk , . . . , i p ) as an element of M fi0 fi p via the map

M fi
c
fik fi p
M fi0 fi p
given by localization at fik . Similar to Exercise 9.0.2, we have d p d p1 = 0.

Definition 9.5.1. Given an R-module M and generators f = ( f1 , . . . , f ) of I, the
local Cech
complex is
0
d
d
d
C (f, M) : 0 C0 (f, M) C1 (f, M) C2 (f, M) .
Just as the Cech

complex computes sheaf cohomology, the local Cech
complex
computes local cohomology. See [158, Thm. 7.13] for a proof of the following.
445
Theorem 9.5.2. Given an R-module M and generators f = ( f1 , . . . , f ) of I, there

are natural isomorphisms
HIp (M) H p (C (f, M))
for all p 0.
Example 9.5.3. For I = hx, yi S = C[x, y], the local Cech

complex is
0 S Sx Sy Sxy 0.
Theorem 9.5.2 implies that HI2 (S) is the cokernel of Sx Sy Sxy . Consider
x1 y1 C[x1 , y1 ] = SpanC (xa yb | a, b > 0) with S-module structure given by
(
x(ai) y(b j) i < a, j < b
i j a b
x y x y =
0
otherwise.
Then the map Sxy x1 y1 C[x1 , y1 ] defined by f (x, y)/(xy)k 7 f (x, y) xk yk
gives an exact sequence
(9.5.1)
Sx Sy Sxy x1 y1 C[x1 , y1 ] 0
(Exercise 9.5.1). Thus HI2 (S) x1 y1 C[x1 , y1 ].
It follows that if x and y have degree 1, then HI2 (S) is graded with
HI2 (S) = HI2 (S)4 HI2 (S)3 HI2 (S)2 0
and dim HI2 (S) = 1 for > 0 (Exercise 9.5.1).
The number 1 just computed is also the dimension of H 1 (P1 , OP1 ()) (see
Example 9.4.1). This is no accident, since we will prove below that
for all a Z.
HI2 (S)a H 1 (P1 , OP1 (a))
Relation with Ext. For us, an especially useful aspect of local cohomology is its
relation to Ext. In 9.0 we introduced Ext in the context of sheaves. There is
also a module version, where for a fixed R-module N, the derived functors of
M 7 HomR (N, M) are denoted ExtRp (N, M). They have the expected properties,
and more importantly, are easy to compute by computer algebra systems such as
Macaulay2.
To see the relation between Ext and local cohomology, we begin with the observation that an R-module homomorphism : R/I k M is determined by (1),
and choosing any a M gives a homomorphism, provided that I k a = 0. It follows
that HomR (R/I k , M) consists of those elements of M annihilated by I k . Comparing
this to the definition of I (M), we obtain
I (M) = lim
HomR (R/I k , M).
446
Since direct limit is an exact functor (Exercise 9.5.2), it follows without difficulty
that the derived functors are related the same way (see [158, Thm. 7.8] for a proof).
Theorem 9.5.4. HIp (M) = lim
ExtRp (R/I k , M).
We will need a variant of this result. For an ideal I = h f1 , . . . , f i, let

I [k] = h f1k , . . . , fk i.
Theorem 9.5.5. HIp (M) = lim

ExtRp (R/I [k] , M).
Proof. This follows from [158, Rem. 7.9] since I [k] I k and I k I [k] .
Here is a simple example.

Example 9.5.6. Let I = hx, yi S = C[x, y]. We know by Example 9.5.3 that
HI2 (S)5 has dimension 4. To compute this using Ext, let A = S/I [k] = S/hx k , y k i.
Since A and S are graded S-modules, the Ext group Ext2S (A, S) is also graded. As
described in Example B.4.1, we can compute the graded piece Ext2S (A, S)5 for
various values of k using Macaulay2. When k = 3, we find that
dim ExtS2 (S/I [3] , S)5 = 2.
Since dim HI2 (S)5 = 4, we see that k = 3 is not big enough. If we switch to k 4,
then the answer stabilizes at the number 4.
In this example, ExtS2 (S/I [k] , S)5 = HI2 (S)5 for k sufficiently large. We will
prove below that for any degree a Z, we have
ExtS2 (S/I [k] , S)a = HI2 (S)a
provided that k is sufficiently large. The problem is that ExtS2 (S/I [k] , S) is finitely
generated over S but HI2 (S) is not (Exercise 9.5.4). So we cannot use the same k
for all degrees a. Explicit bounds on k in terms of a are needed in order to turn the
method of Example 9.5.6 into an algorithm.
This concludes our overview of local cohomology; for more, see [89, App. 4]
or [158]. Our next task to is apply local cohomology to toric varieties.
The Toric Case. The total coordinate ring S = C[x | (1)]Qof a toric variety
a
X is graded by the class group Cl(X ), where a monomial x has degree
P
[ a D ] Cl(X ). We also have the irrelevant ideal
Q
B() = hx | max i, x = 6(1) x ,
introduced in Chapter 5. We assume that X has no torus factors.
We proved in Chapter 5 that a finitely generated graded S-module M and a

divisor class Cl(X ) give the following:
447
e on X . Every coherent sheaf on X arises in this

The coherent sheaf F = M
way (Proposition 5.3.9).
The shifted module M(), where M() = M+ for Cl(X ).
The sheaf associated to S() is denoted OX (), and Proposition 5.3.7 tells us that
OX () OX (D) when D is a Weil divisor with divisor class . More generally,
e then F () will denote the sheaf associated to M(). When is the
if F = M,
class of a Cartier divisor, one can prove that
(9.5.2)
F () F X OX ()
(Exercise 9.5.3). Our first main result is that the local cohomology for the irrelevant
ideal B() computes the sheaf cohomology of all twists of a coherent sheaf on X .
Theorem 9.5.7. Let M be a finitely generated graded S-module with associated
e on X . If p 2, then
coherent sheaf F = M
M
p
H p1 (X , F ()).
HB()
(M)
Cl(X )
Furthermore, we have an exact sequence

M
0
1
0 HB()
(M) M
H 0 (X , F ()) HB()
(M) 0.
Cl(X )
Proof. Let max = {1 , . . . , }, so that B() is generated by the monomials fi =
x i , i [] = {1, . . . , }. Then the terms of the local Cech

complex C (f, M) are
direct sums of localizations M at products of various fi s. These localizations are
Cl(X )-graded since M is graded and the fi are monomials. The differentials also
p
(M) has a natural
preserve the grading, which by Theorem 9.5.2 implies that HB()
Cl(X )-grading such that for all Cl(X ), we have
H p (M) = H p (C (f, M) ).
B()
We will relate Cp (f, M) to the Cech

complex for F () given in (9.1.1).
p
To compute C (f, M) , first observe that
M
M

M() fi0 fi p1 0
M fi0 fi p1 =
Cp (f, M) =
(i0 ,...,i p1 )[] p1
(i0 ,...,i p1 )[] p1
since shifting commutes with localization. For = (i0 , . . . , i p1 ) [] p1 , note that

Y a
(9.5.3)
fi0 fi p1 = x i0 x i p1 =
x ,
(1)
/
where a = |{i j |
/ i j }| > 0. If we set = i0 i p1 , then x gives
the same localization as (9.5.3). Since F () is the sheaf associated to M(),
Proposition 5.3.3 implies that

M() fi0 fi p1 0 = M()x 0 (U , F ()).
448
Comparing this with the Cech

complex (9.1.1) for the open cover U = {Ui }i[] ,
we obtain
M
Cp (f, M) =
(U , F ()) Cp1 (U , F ()).
[] p1
This isomorphism is compatible with the differentials. It follows that the local Cech
complex C (f, M) is obtained from the Cech complex C (U , F ()) by deleting

the first term and shifting the remaining terms. When p 2, this implies
p
HB()
(M) H p1 (X , F ()),
and with a little work (Exercise 9.5.5) we also get an exact sequence
0
1
0 HB()
(M) M H 0 (X , F ()) HB()
(M) 0.
Theorems 9.5.5 and 9.5.7 imply that when p 1,

H p (X , F ()) lim
ExtSp+1 (S/B()[k] , M)
e and Cl(X ). We can compute Ext p+1 (S/B()[k] , M) by the

when F = M
S
methods of Example 9.5.6. The problem is the direct limit. We tackle this next.
Stabilization of Ext. In the toric case, Ext has a nice relation to local cohomology.
Lemma 9.5.8. Let M be a finitely generated Cl(X )-graded S-module and fix
Cl(X ). If is a complete fan, then there exists k0 N such that for all k k0 , the
natural map S/B()[k+1] S/B()[k] induces an isomorphism
ExtSp (S/B()[k] , M) ExtSp (S/B()[k+1] , M) .
In particular, k k0 implies
p
ExtSp (S/B()[k] , M) HB()
(M) .
Proof. We give a proof only for M = S following Mustata [211, Thm. 1.1]. For the
general case, one replaces M with a free resolution and uses a spectral sequence
argument. See [91, Prop. 4.1] for the details.
Earlier we described Ext using the derived functors of M 7 HomR (N, M) for
fixed N. Ext also comes from the derived functors of N 7 HomR (N, M) for fixed M,
where one uses a projective resolution of N instead of an injective resolution of M
(see, for example, [89, A3.11]). In particular, we can compute ExtSp (S/B()[k] , S)
using any free resolution of S/B()[k] . An especially nice resolution is given by
the Taylor resolution, which is described as follows.
We begin with S/B(). The minimal generators of B() are fi = x i for i [].
Let Fs be a free S module with basis eI for all I [] with |I| = s. To define the
449
differential ds : Fs Fs1 , take I, J [] with |I| = |J| + 1 = s and list the elements
of I as i1 < < is . Then define
(
0
J 6 I
cIJ =
r
(1) fI / fJ I = J {ir },
where fI = lcm( fi | i I) and similarly for fJ . Then
P
ds (eI ) = J cIJ eJ .
Since F0 = S, we have an obvious map F0 S/B(). One can prove that

d
2
1
0 F

F1
F0 S/B() 0
is exact (see [89, Ex. 17.11]). This is the Taylor resolution (F , d ) of S/B().
This construction applies to any monomial ideal. In particular, it works for
[k] [k]
B()[k] = h f1k , . . . , fk i. Let (F , d ) denote the Taylor resolution of S/B()[k] .
[k]
It has the same modules Fs = Fs , and since the fi are square-free, we have fIk =
[k]
lcm( fik | i I). Thus the differentials ds in the Taylor resolution of S/B()[k] are
given by
P k
[k]
eJ , cIJ as above.
ds (eI ) = J cIJ
We now compare the Taylor resolutions of S/B()[k] and S/B()[k+1] . The
[k+1]
[k]
maps s : Fs
Fs defined by s (eI ) = fI eI induce a commutative diagram
s
[k+1]
Fs
[k]
Fs
[k+1]
ds
[k]
[k+1] s1
Fs1
ds
/ F [k] .
s1
These maps are compatible with the surjection S/B()[k+1] S/B()[k] . Hence
(9.5.4)
[k+1]
H p (HomS (F
[k]
, S)) H p (HomS (F , S))
induced by the s can be identified with the canonical map
ExtSp (S/B()[k] , S) ExtSp (S/B()[k+1] , S).

The next key idea involves the use of a finer grading than the Cl(X )-grading
used so far. Recall that this grading is induced by the map
Z(1) Cl(X )
Q a
P
where x a = x has degree [ a D ]. The ring S also has a Z(1) -grading
where deg(x a ) = a. Since B()[k] is a monomial ideal, the quotient ring S/B()[k]
is Z(1) -graded. Then ExtiS (S/B()[k] , S) inherits a natural Z(1) -grading.
(9.5.5)
[k]
[k]
Now grade the Taylor resolution (F , d ) by setting deg(eI ) = k deg( fI ). This

guarantees two things:
450
[k]
The differential ds has degree 0, so the isomorphism ExtSp (S/B()[k] , S)

[k]
H p (HomS (F , S) is Z(1) -graded.
The map s has degree 0, so (9.5.4) is Z(1) -graded.
[k]
[k]
It follows that (Fs ) = HomS (Fs , S) has dual basis eI with deg(eI ) = k deg( fI ).
Given a, b Z(1) , define
a b if and only if a b for all (1).

We claim that if a (k, . . . , k), then
(9.5.6)
[k]
[k+1]
)a
(
s )a : (Fs )a (Fs
is an isomorphism for all s.
To prove this, first observe that

s (eI ) = fI eI . It follows that (s )a is injective.
[k+1]
)a . Then b (k + 1)deg( fI ) = a, so
Now take x b eI (Fs
b = (k + 1)deg( fI ) + a (k + 1)deg( fI ) + (k, . . . , k).

Since fI is square-free, (k + 1)deg( fI ) is a vector with whose th entry is (k + 1) if
x divides fI and 0 otherwise. Then the above inequality implies that fI divides x b ,
so that x b eI is in the image of (
s )a . It follows that (9.5.6) is an isomorphism.
p
The local cohomology HB()
(S) has a Z(1) -grading which is compatible with
its Cl(X )-grading via (9.5.5). This follows easily from Theorem 9.5.5 and the
Taylor resolution. If Cl(X ) and p 2, then
p
HB()
(S) H p1 (X , OX ()).
by Theorem 9.5.7. The right-hand side is finite-dimensional since is complete.

p
This implies that when we decompose HB()
(S) into its nonzero Z(1) -graded
p
p
pieces HB()
(S)a , only finitely many can appear in HB()
(S) . For these finitely
many as, pick k0 such that they all satisfy a (k0 , . . . , k0 ). This k0 has the
required properties. The argument for p = 0, 1 is covered in Exercise 9.5.6.

Theorem 9.5.7 and Lemma 9.5.8 imply that for p 1 and k 0,
p+1
(S) ExtSp+1 (S/B()[k] , S).
H p (X , OX ()) HB()
Notice how these isomorphisms generalize Examples 9.5.3 and 9.5.6. Our final
task is to give an explicit method for deciding when k is big enough.
Bounds. For a complete fan , graded S-module M, and divisor class Cl(X ),
the paper [91] gives an explicit value for the number k0 appearing in Lemma 9.5.8.
We will state the results of [91] without proof, though the following example suggests some of the ideas involved.
451
Example 9.5.9. Let be a complete fan in MR R2 with minimal generators

u1 , . . P
. , ur arranged counterclockwise around the origin. Suppose we have a divisor
D = i ai Di on X and m M such that H 1 (X , OX (D))m 6= 0. It follows from
Proposition 9.1.6 that the sign pattern of m (determined by hm, ui i + ai 0 or < 0)
has at least two strings of consecutive s. The set I = {i [r] | hm, ui i + ai < 0}
records the locations of the s in the sign pattern of m.
Since H 1 (X , OX (D))m 6= 0 is finite-dimensional, there are only finitely many

ms with the same sign pattern. In other words, the inequalities
hm, ui i + ai < 0, i I
hm, ui i + ai 0, i [r] \ I
have only finitely many integer solutions, which means that the region of the plane
defined by these inequalities is bounded. You can see and example of this in Figure 4 from Example 9.1.8. Since this region determined by the ui and ai , it should
be possible to bound the size of the ms that appear in terms of the ui and ai .
To relate this to Lemma 9.5.8, let b = (hm, u1 i + a1 , . . . , hm, ur i + ar ). Then it
is easy to see that
2
H 1 (X , OX (D))m HB()
(S)b .
Thus bounding the ms with H 1 (X , OX (D))m 6= 0 is equivalent to bounding the

2
bs with HB()
(S)b 6= 0. And once we find this bound, we know how to pick k0 so
that b (k0 , . . . , k0 ) for all such bs. The proof of Lemma 9.5.8 shows that this
k0 works.
In the general case, we proceed as follows. Pick a basis e1 , . . . , en of M and

let A be the matrix whose rows give the coefficients of the u s with respect to the
chosen basis. Then A is an r n matrix, where r = |(1)|. For this matrix, define
qn = min(|nonzero n n minors of A|)
Q1 = max(|entries of A|)
(9.5.7)
Qn1 = max(|(n 1) (n 1) minors of A|).

The following bound is proved in [91, Cor. 3.3].
Theorem
9.5.10. Given a complete fan in NR Rn and a divisor class =
P
[ a D ] Cl(X ), we have
p
ExtSp+1 (S/B()[k] , S) HB()
(S)
for all k k0 , where

k0 = n2 max (|a |)
Q1 Qn1
.
qn
452
For a finitely generated graded S-module M, the formula for k0 involves the
minimal free resolution of M. Write the resolution as
F2 F1 F0 M 0
and write
Fj =
S()r j, .
Cl(X )
P
Finally, for each with r j, 6= 0, write = [ a, D ] Cl(X ). Then [91, Prop.
4.1] gives the following bound.
Theorem 9.5.11. Let and be as in Theorem 9.5.10, and let M be a finitely
generated graded S-module. Then
p
ExtSp+1 (S/B()[k] , M) HB()
(M)
for all k k0 , where

k0 = n2 max, j,r j, 6=0 (|a a, |)
Q1 Qn1
.
qn
The Cotangent Bundle. We end this section by using our methods to calculate the
cohomology of the cotangent bundle of a smooth complete toric variety X . The
e
first step is to find a graded S-module M with 1X M.
In (8.1.7) we constructed an exact sequence
0 1S M Z S
S/hx i
L
P
where M Z S S/hx i is defined by m f 7 hm, u i[ f ], and in Corollary 8.1.5 we showed that 1X is the sheaf of 1S . However, we need a description
of 1S that is easier to implement on a computer. We do this as follows.
Pick bases of M and Pic(X ) and order the elements of (1) as 1 , . . . , r .
Then the basic exact sequence
0 M Z(1) Pic(X ) 0
can be written
(9.5.8)
0 Zn Zr Zrn 0,
where A is an r n matrix whose ith row consists of the coefficients of ui in

the basis of M. This is the same matrix A that appears in Theorem 9.5.10. The
(r n) r matrix B is called the Gale dual of A.
Lemma 9.5.12. The (r n) r matrix
x1 0
0 x
2
= B .
..
.
.
.
0
0
..
.
0
0
..
.
xr
453
L
induces a graded homomorphism : ri=1 S(deg(xi )) S rn of degree 0 such
that 1X is the sheaf associated to ker().
Proof. Let x be the r r diagonal matrix of variables appearing in the statement of
the lemma. Similar to the proof of Theorem 8.1.6, we have a commutative diagram
0
0
0
Lr

/ 1
S

/ Z n Z S
i=1 S(deg(xi ))

/ Z r Z S
Zrn
ZS
Zrn
i=1 S/hxi i
Lr
/
Lr
i=1 S/hxi i
ZS

/0
/0
/0
The rows are exact, as are the center and right columns. By the diagram chase from
the proof of Theorem
L 8.1.6, the dotted arrows are exact, and by commutivity, the
dotted arrow from ri=1 S(deg(xi )) to Zrn Z S is given by = B x.
Example 9.5.13. For P 2 , the matrices A and B of (9.5.8) are given by
1
0

A= 0
1 , B = 1 1 1 .
1 1
Thus = (x y z), which gives the exact sequence
(x y z)
0 1S S(1)3 S.
Using Macaulay2 as in Example B.4.2, one computes the free resolution
0
(9.5.9)
1
z
@ x A
y
0 S(3) S(2)3 1S 0.
The numbers from (9.5.7) are q2 = Q1 = 1, so that for a Z, the formula for k0
from Theorem 9.5.11 is
k0 = 4 max(|a 2|, |a 3|).
For a in the range 4 a 4, we can use k0 = 28. This implies that for p 1,
H p (P 2 , 1P2 (a)) ExtSp+1 (S/hx 28 , y 28 , z 28 i, 1S )a
when 4 a 4. This can be computed by the methods of Example 9.5.6. We

can also compute H 0 (P 2 , 1P2 (a)) directly from 1S (Exercise 9.5.7). The results
are shown in Table 3 on the next page.
454
a\p 0
4 0
3 0
2 0
1 0
0
0
1
0
2
3
3
8
4 15
1 2
0 15
0 8
0 3
0 0
1 0
0 0
0 0
0 0
0 0
Table 3. dim H p (P 2 , 1P2 (a)) for 4 a 4
In Exercise 9.5.8 you will calculate this table by hand. Notice also that the symmetry of the table comes from Serre duality. See Example B.5.2 and Exercise B.5.1
for different way to compute Table 3.
Here is a slightly more complicated example.

Example 9.5.14. Let us compute the first cohomology group of various twists of
the cotangent sheaf of the Hirzebruch surface H2 . We will use the notation of
Example 9.3.6. The goal is to compute the cohomology groups of 1H2 (a, b) =
1H2 (aD3 + bD4 ) for all a, b.
One way to proceed would be to follow the approach of the previous example
using Macaulay2. This would involve the following steps:
Construct the total coordinate ring S = Q[x1 , x2 , x3 , x4 ] of H2 is with its Z2 grading and irrelevant ideal B = hx1 x3 , x2 x4 i.
Use the 2 4 matrix to define the S-module 1S as a kernel as in Lemma 9.5.8.
Use a free resolution of 1S to compute the bound k0 from Theorem 9.5.11.
Compute various graded pieces of ExtSp+1 (S/B[k0 ] , 1S ) as explained in B.4 of

Appendix B.
Alternatively, we could use the Macaulay2 package NormalToricVarieties
described in Examples B.5.1 and B.5.2. This package automates most of the above
steps and makes it easy to calculate H p (H2 , 1H2 (a, b)) directly and efficiently.
Table 4 records some computations of H 1 done this way (see Example B.5.2 for
the details). There are many things we can see from this table, including:
Theorem 9.4.11 implies that dim H 1 (H2 , 1H2 ) = h1 = f1 2 f2 = 4 2 1 = 2
since H2 comes from a quadrilateral (see Examples 2.3.16 and 3.1.16). This
explains the 2 when (a, b) = (0, 0).
Serre duality implies that H 1 (H2 , 1H2 (a, b)) H 1 (H2 , 1H2 (a, b)). This
explains the symmetry in the table.
455
a\b 2 1 0 1 2
2
0 0 3 4 3
1
0 0 2 2 2
0
1 1 2 1 1
1
2 2 2 0 0
2
3 4 3 0 0
Table 4. dim H 1 (H2 , 1H2 (a, b)) for 2 a, b 2
Bott-Danilov-Steenbrink vanishing implies that H 1 (H2 , 1H2 (a, b)) = 0 for

a, b > 0. This explains the 0s in the lower right corner of the table.
See [91, Sec. 5] for an algorithmic approach to these calculations. Using

[190, Cor. 3.4], one can describe an especially efficient method for computing the
sheaf cohomology of a toric variety. This method is implemented in Greg Smiths
Macaulay2 package NormalToricVarieites [252]. A equivalent method was
conjectured independently [35].
Exercises for 9.5.
9.5.1. Prove the exactness of (9.5.1).
9.5.2. Suppose that (Ai , i ), (Ai , i ), (Ai , i ) are directed systems, and that for each i,
there exist di and i commuting with the s, such that we have an exact sequence
i
i
Ai 0.
Ai
0 Ai
Prove the exactness of the sequence

0 lim
Ai lim
Ai lim
Ai 0.
9.5.3. Prove (9.5.2) when Cl(X ) is the class of a Cartier divisor. Hint: Look carefully
at Proposition 5.3.3 and remember that the restriction of OX () to U is trivial.
9.5.4. For I = hx, yi S = C[x, y], prove that HI2 (S) is not finitely generated as an S-module.
Hint: Use Example 9.5.3 to show that HI2 (S)a 6= 0 for all a 2.
9.5.5. Complete the proof of Theorem 9.5.7 by proving that there is an exact sequence
0
1
0 HB()
(M) M H 0 (X , F ()) HB()
(M) 0
for all Cl(X ).

1
0
9.5.6. Prove that HB()
(S) = HB()
(S) = 0. Hint: Proposition 5.3.7.
9.5.7. Here are some details to check from Example 9.5.13.

(a) Use the exact sequence from Lemma 9.5.12 to show that (1S ) H 0 (X , 1X ())
for all Pic(X ) when X is smooth and complete.
(b) Verify the first column of Table 3.
(c) Show that the first column agrees with the Bott formula from Example 9.4.10.
456
9.5.8. We can check Table 3 of Example 9.5.13 with a barehanded approach. Observe
that we have exact sequences involving 1P2 :
0 1P2 OP32 (1) OP2 0
0 OP2 (3) OP32 (2) 1P2 0.
The first comes from Lemma 9.5.12, and the second comes by sheafifying the free resolution of 1S computed in Example 9.5.13. Twist these sequences by a Z and consider the
resulting long exact sequences in cohomology. Using the vanishing theorems we know,
conclude that the nonzero values for H p (P 2 , 1P2 (a)) are:
dim H 0 (P 2 , 1P2 (a)) = a2 1 if a 2
dim H 1 (P 2 , 1P2 (a)) = 1
2
dim H (P
, 1P2 (a))
These formulas give the numbers in Table 3.
if a = 1
= a 1 if a 2.
Part II. Topics in

Toric Geometry
Chapters 10 to 15 explore further topics in theory of toric varieties. This

part of the book assumes a wider knowledge of algebraic geometry, though
we give careful references for all of the tools used in our study of toric
geometry. The topics presented here are just of few of many rich areas of
inquiry encountered in the study of toric varieties.
457
Chapter 10
Toric Surfaces
In this chapter, we will apply the theory developed so far to study the structure
of 2-dimensional normal toric varieties (toric surfaces). We will describe their
singularities, introduce the idea of a resolution of singularities, and also classify
smooth complete toric surfaces. Along the way, we will encounter two types of
continued fractions, Hilbert bases, the Grobner fan, the McKay correspondence,
the Riemann-Roch theorem, the sectional genus, and the number 12.
10.1. Singularities of Toric Surfaces and Their Resolutions

Singular Points of Toric Surfaces. If X is the toric surface of a fan in NR R2 ,
then minimal generators of the rays (1) are primitive and hence extend to a
basis of N. Then Theorem 3.1.19 implies that the toric surface obtained by removing the fixed points of the torus action (i.e., the points corresponding to the
2-dimensional cones under the Orbit-Cone Correspondence) is smooth. There are
only finitely many such points, so X has at most finitely many singular points.
Moreover, 2-dimensional cones are always simplicial, so from Example 1.3.20,
each of these singular points is a finite abelian quotient singularity (isomorphic to
the image of the origin in the quotient C2 /G where G is a finite abelian group).
All cones are assumed to be rational and polyhedral. A 2-dimensional strongly
convex cone in NR R2 has the following normal form that will facilitate our study
of the singularities of toric surfaces.
Proposition 10.1.1. Let NR R2 be a 2-dimensional strongly convex cone.
Then there exists a basis e1 , e2 for N such that
= Cone(e2 , de1 ke2 ),
where d > 0, 0 k < d, and gcd(d, k) = 1.
459
Chapter 10. Toric Surfaces
460
Proof. We will need the following modified division algorithm here and at several
other points in this chapter (Exercise 10.1.1):
(10.1.1)
Given integers l and d > 0, there are unique integers

s and k such that l = sd k and 0 k < d.
Say = Cone(u1 , u2 ), where ui are primitive vectors. Since u1 is primitive, we

can take it as part of a basis of N, and we let e2 = u1 . Since is strongly convex,
for any basis e1 , e2 for N, it will be true that
u2 = de1 + le2
for some d 6= 0. By replacing e1 by e1 if necessary, we can assume d > 0. By
(10.1.1), there are integers s, k such that l = sd k, where 0 k < d. Using this
integer s, let e1 = e1 + se2 . Then e1 , e2 is also a basis for N and
u2 = de1 + (l sd)e2 = de1 ke2 .
Hence = Cone(e2 , de1 ke2 ) as claimed, and gcd(d, k) = 1 follows since u2 is
primitive.

We will call the integers d, k in this statement the parameters of the cone ,
and {e1 , e2 } is called a normalized basis for N relative to . The uniqueness of d, k
will be studied in Proposition 10.1.3 below.
Using the normal form, we next describe the local structure of the point p
in the affine toric variety U . Recall from Example 1.3.20 that if N N is the
sublattice generated by the ray generators of , then U C2 /G, where G = N/N .
In our situation, N = Ze1 Ze2 , and
N = Ze2 Z(de1 ke2 ) = dZe1 Ze2 ,
so it follows easily that

(10.1.2)
G = N/N Z/dZ.
In particular, for singularities of toric surfaces, the finite group G is always cyclic.
The action of G on C2 is determined by the integers d, k as follows. We write
d = { C | d = 1}
for the group of dth roots of unity in C. Then a choice of a primitive dth root of
unity defines an isomorphism of groups d Z/dZ.
Proposition 10.1.2. Let M be the dual lattice of N and let m1 , m2 M be dual
to u1 , u2 in N . Using the coordinates x = m1 and y = m2 of C2 , the action of
d N/N on C2 is given by
(x, y) = ( x, k y).
Furthermore, U C2 /d with respect to this action.
461
Proof. The general discussion in 1.3 shows that the quotient N/N Z/dZ acts
on the coordinate ring of C2 via
(10.1.3)
(u + N ) m = e2ihm ,ui m ,
where m M and u = je1 for 0 j d 1.
An easy calculation shows that hm1 , e1 i = 1/d and hm2 , e1 i = k/d. Hence if
we set up the isomorphism d N/N by mapping e2i j/d 7 je1 + N , then for all
= e2i j/d d , we have
(x, y) = (e2i j/d x, e2i jk/d y) = ( x, k y)
by (10.1.3). This is what we wanted to show.
We next describe the slight but manageable ambiguity in the normal form for
2-dimensional cones. Two cones are lattice equivalent if there is a bijective Zlinear mapping : N N taking one cone to the other. After choice of basis for
N, such mappings are defined by matrices in GL(2, Z).
e e
Proposition 10.1.3. Let = Cone(e2 , de1 ke2 ) and
e = Cone(e2 , de
ke2 ) be
1
cones in normal form that are lattice equivalent. Then de = d and either e
k = k or
e
kk 1 mod d.
e for the sublattices

Proof. Since the cones are lattice equivalent, writing N and N
e .
as in (10.1.2), there is a bijective Z-linear mapping : N N such that (N ) = N
e N/N , so de= d. The statement about k and e
Hence N/N
k is left to the reader in
Exercise 10.1.2.

Here are two examples to illustrate Proposition 10.1.2.
Example 10.1.4. First consider a cone
= Cone(e2 , de1 e2 )
with parameters d > 1 (so the cone is not smooth) and k = 1. This is precisely
the cone considered in Example 1.2.22. The corresponding toric surface U is
the rational normal cone Cbd Cd+1 . The quotient Cbd C2 /d was studied in
the special case d = 2 in Example 1.3.19, and the general case was described in
Exercise 1.3.11. With the notation of Proposition 10.1.2, d acts on (x, y) C2
via (x, y) = (x, y) and the ring of invariants is
C[x, y]d = C[x d , x d1 y, . . . , xy d1 , y d ],
so
U C2 /d Spec(C[x d , x d1 y, . . . , xy d1 , y d ]).
On the other hand, from Example 1.2.22, we also have the description
U Spec(C[s, st, st 2 , . . . , st d ]).
462
Exercise 10.1.3 studies the relation between these representations of the coordinate
ring of U .
Example 10.1.5. Next consider a cone with parameters d and k = d 1, so

d = k + 1. We will express everything in terms of the parameter k in the following.
Unlike the previous example, this is a case we have not encountered previously.
Note that k 1 mod d. Hence by Proposition 10.1.2, the action of G = N/N on
C2 is given by
(x, y) = ( x, 1 y).
It is easy to check that the ring of invariants here is
C[x, y]k+1 = C[x k+1 , y k+1 , xy].

Moreover we have an isomorphism of rings
: C[X ,Y , Z]/hZ k+1 XY i C[x k+1 , y k+1 , xy]
X 7 x k+1
Y 7 y k+1
Z 7 xy,
so we may identify the toric surface U with the variety V(Z k+1 XY ) C3 .
The origin is the unique singular point of the affine variety of Example 10.1.5
and is called a rational double point (or Du Val singularity) of type Ak . Another
standard form of these singularities is given in Exercise 10.1.4. They are called
double points because the lowest degree nonzero term in the defining equation has
degree two (i.e., the multiplicity of the singularity is two). The rational double
points are the simplest singularities from a certain point of view. The exact definition, which we will give in 10.4, depends on the notion of a resolution of singularities, which will be introduced shortly. All rational double points appear as singularities of quotient surfaces C2 /G where G is a finite subgroup of SU(2, C). There
is a complete classification of such points in terms of the Dynkin diagrams of types
Ak , Dk , E6 , E7 , and E8 . The groups corresponding to the diagrams Dk , E6 , E7 , E8 are
not abelian, so by the comment after (10.1.2), such points do not appear on toric
surfaces. We will see one way that the Dynkin diagram Ak appears from the geometry of the toric surface U in Exercise 10.1.5, and we will return to this example
in 10.4. More details on these singularities can be found in [245, Ch. VI] and in
the article [85].
Here is another interesting aspect of Example 10.1.5. Recall that a normal
variety X is Gorenstein if its canonical divisor is Cartier (Definition 8.2.14). The
following result was proved in Exercise 8.2.13 of Chapter 8.
Proposition 10.1.6. For a cone = Cone(e2 , de1 ke2 ) in normal form, the affine
toric surface U is Gorenstein if and only if k = d 1.
463
Toric Resolution of Singularities. Let X be a normal toric surface, and denote by

Xsing the finite set of singular points of X (possibly empty).
Definition 10.1.7. A proper morphism : Y X is a resolution of singularities
of X if Y is a smooth surface and induces an isomorphism of varieties
(10.1.4)
Y \ 1 (Xsing ) X \ Xsing .
Such a mapping modifies X to produce a smooth variety without changing the

smooth locus X \ Xsing . One of the most appealing aspects of toric varieties is the
way that many questions that are difficult for general varieties admit simple and
concrete solutions in the toric case. The problem of finding resolutions of singularities is a perfect example. We illustrate this by constructing explicit resolutions
of singularities of the toric surfaces from Examples 10.1.4 and 10.1.5.
Example 10.1.8. Consider the rational normal cone of degree d, the affine toric
surface U for = Cone(e2 , de1 e2 ) studied in Example 10.1.4. Let be the fan
in Figure 1 obtained by inserting a new ray = Cone(e1 ) subdividing into two
2-dimensional cones:
1 = Cone(e2 , e1 )
2 = Cone(e1 , de1 e2 ).
Figure 1. The cone and the refinement given by 1 , 2 ,
We now use some results from Chapter 3. The identity mapping on the lattice
N is compatible with the fans and as in Definition 3.3.1. By Theorem 3.3.4,
we have a corresponding toric blowup morphism
(10.1.5)
: X U .
Note that both 1 and 2 (as well as all of their faces) are smooth cones. Hence Theorem 3.1.19 implies that X is a smooth surface. In addition, the toric morphism
is proper by Theorem 3.4.11 since is a refinement of . Finally, we claim that
satisfies (10.1.4). This follows from the Orbit-Cone Correspondence on the two
464
surfaces: if p is the distinguished point corresponding to the 2-dimensional cone

(the singular point of U at the origin), then restricts to an isomorphism
X \ 1 (p ) U \ {p } = (U )smooth .
The inverse image E = 1 (p ) is the curve on X given by the closure of the
TN -orbit O( ) corresponding to the ray . That is, the singular point blows up
to E P1 on the smooth surface. It follows that X and the morphism (10.1.5)
give a toric resolution of singularities of the rational normal cone. We call E the
exceptional divisor on the smooth surface. We will say more about how E sits
inside the surface X in 10.4.
Example 10.1.9. We consider the case d = 4 of Example 10.1.5, for which the
surface U has a rational double point of type A3 . We will leave the details, as well
as the generalization to all d 2, to the reader (Exercise 10.1.5). It is easy to find
subdivisions of
= Cone(e2 , 4e1 3e2 )
yielding collections of smooth cones. The most economical way to do this is to
insert three new rays 1 = Cone(e1 ), 2 = Cone(2e1 e2 ), 3 = Cone(3e1 2e2 )
to obtain a fan consisting of four 2-dimensional cones and their faces.
The fan produced by this subdivision is somewhat easier to visualize if we draw
the cones relative to a different basis u1 , u2 for N. For u1 = e2 and u2 = e1 e2 , the
cone = Cone(u1 , u1 + 4u2 ) and the fan with maximal cones
1 = Cone(u1 , u1 + u2 )
(10.1.6)
2 = Cone(u1 + u2 , u1 + 2u2 )
3 = Cone(u1 + 2u2 , u1 + 3u2 )
4 = Cone(u1 + 3u2 , u1 + 4u2 )
appear in Figure 2.
4 3
Figure 2. The cone and the refinement
465
You will check that each of these cones is smooth. Hence X is a smooth
surface. Since is a refinement of , we have a proper toric morphism
: X U .
As in the previous example, restricts to an isomorphism from X \ 1 (p ) to
X \ {p }. In this case, the exceptional divisor E = 1 (p ) is the union
E = V (1 ) V (2 ) V (3 )
on X . The curves V (i ) are isomorphic to P1 . The first two intersect transversely
at the fixed point of the TN -action on X corresponding to the cone 2 , while the
second two intersect transversely at the fixed point corresponding to 3 .
In these examples, we constructed toric resolutions of affine toric surfaces with

just one singular point. The same techniques can be applied to any normal toric
surface X .
Theorem 10.1.10. Let X be a normal toric surface. There exists a smooth fan
refining such that the associated toric morphism : X X is a toric
resolution of singularities.
Proof. It suffices to show the existence of the smooth fan refining . The
reasoning given in Example 10.1.8 applies to show that the corresponding toric
morphism is proper and birational, hence a resolution of singularities of X .
We will prove this by induction on an integer invariant of fans that measures
the complexity of the singularities on the corresponding surfaces. Let 1 , . . . ,
denote the 2-dimensional cones in a fan . For each i, we will write Ni for the
sublattice of N generated by the ray generators of i . Then we define
s() =
X
i=1
(mult(i ) 1) ,
where mult(i ) = [N : Ni ] as in 6.3. If s() = 0, then = 0 or mult(i ) = 1 for

all i. It is easy to see that this implies that is a smooth fan. Hence X is already
smooth and we take this as the base case for our induction.
For the induction step, we assume that the existence of smooth refinements has
been established for all fans with s() < s, and consider a fan with s() = s.
If s 1, then there exists some nonsmooth cone i in . By Proposition 10.1.1,
there is a basis e1 , e2 for N such that i = Cone(e2 , de1 ke2 ) with parameters
d > 0, 0 k < d, and gcd(d, k) = 1. Consider the refinement of obtained by
subdividing the cone i into two new cones
i = Cone(e2 , e1 )
i = Cone(e1 , de1 ke2 )
466
with a new 1-dimensional cone = Cone(e1 ). We must show that s( ) < s() to
invoke the induction hypothesis and conclude the proof.
In s(), the terms corresponding to the other cones j for j 6= i are unchanged.
The cone i is smooth since e1 , e2 is the normalized basis of N relative to i . So it
contributes a zero term in s( ). Now consider the cone i . In order to compute
its contribution to s( ), we must determine the parameters of i .
In terms of the basis e1 , e2 for N, the Z-linear mapping defined by the matrix

0 1
A=
1
0
(a 90-degree rotation) takes i to Cone(e2 , ke1 + de2 ). Since A GL(2, Z), it
defines an automorphism of N, and hence i will have the same parameters as
Cone(e2 , ke1 + de2 ). But now we apply (10.1.1) to write
(10.1.7)
d = sk l
where 0 l < k. Since gcd(d, k) = 1, we have gcd(k, l) = 1 as well. Hence the

cone i has parameters k and l obtained from (10.1.7). Since k < d, if Ni is the
sublattice generated by the ray generators of i , then by (10.1.2),
[N : Ni ] = k < [N : Ni ] = d.
It follows that s( ) < s(), and the proof is complete by induction.
We will see in the next section that in the affine case, the refinement that gives
the resolution of singularities of U has a very nice description. As a preview,
notice that in Examples 10.1.8 and 10.1.9, the refinement of the given cone was
produced by subdividing along the rays through the Hilbert basis (the irreducible
elements) of the semigroup N.
A resolution of a non-normal toric surface singularity can be constructed by

first saturating the associated semigroup as in Theorem 1.3.5, then applying the
results of this section. Toric resolutions of singularities for toric varieties of dimension three and larger also exist. However, we postpone the higher-dimensional
case until Chapter 11.
Exercises for 10.1.
10.1.1. Adapt the usual proof of the integer division algorithm to prove (10.1.1).
10.1.2. In this exercise, you will develop further properties of the parameters d, k in the
normal form for cones from Proposition 10.1.1 and prove part of Proposition 10.1.3.
(a) Show that if
e is obtained from a cone by parameters d, k by a Z-linear mapping of
N defined by a matrix in GL(2, Z), then the parameter e
k of
e satisfies either e
k = k, or
ke
k 1 mod d. Hint: There is a choice of orientation to be made in the normalization
ee
process. Recall that gcd(d, k) = 1, so there are integers d,
k such that d de+ ke
k = 1.
10.2. Continued Fractions and Toric Surfaces
467
(b) Show that if is a cone with parameters d, k, then the dual cone MR has parameters d, d k. Hint: Use the normal form for , write down in the corresponding
dual basis in M, then change bases in M to normalize .
10.1.3. With the notation in Example 10.1.4, show that
C[s, st, st 2 , . . . , st d ] C[x d , x d1 y, . . . , xy d1 , y d ]
under s 7 x d and t 7 y/x, and use Proposition 10.1.2 to explain where these identifications
come from in terms of the semigroup S . Hint: We have s = m1 and t = m2 where e1 , e2
is the normalized basis for N and m1 , m2 is the dual basis for M.
10.1.4. In Example 10.1.5, we gave one form of the rational double point of type Ak ,
namely the singular point at (0, 0, 0) on the surface V = V(Z k+1 XY ) C3 . Another
commonly used normal form for this type of singularity is the singular point at (0, 0, 0) on
the surface W = V(X k+1 +Y 2 + Z 2 ). Show that V and W are isomorphic as affine varieties,
hence the singularities at the origin are analytically equivalent. Hint: There is a linear
change of coordinates in C3 that does this.
10.1.5. In this exercise, you will check the claims made in Example 10.1.9 and show how
to extend the results there to the case = Cone(e2 , de1 (d 1)e2 ) for general d.
(a) Check that each of the four cones in (10.1.6) is smooth, so that the toric surface X is
smooth by Theorem 3.1.19.
(b) For general d, show how to insert new rays i to subdivide and obtain a fan whose
associated toric surface is smooth. Try to do this with as few new rays as possible.
Hence we obtain toric resolutions of singularities : X U for all d.
(c) Identify the inverse image C = 1 (p ) in general. For instance, how many irreducible components does C have? How are they connected? Hint: One way to represent the structure is to draw a graph with vertices corresponding to the components and
connect two vertices by an edge if and only if the components intersect on X . Do you
notice a relation between this graph and the Dynkin diagram Ak = Ad1 mentioned
before? We will discuss the relation in detail in 10.4.

To relate continued fractions to toric surfaces, we begin with the affine toric surface
U of a cone NR R2 in normal form with parameters d, k. We will always
assume d > k > 0, so that U has a unique singular point.
Hirzebruch-Jung Continued Fractions. When we construct a resolution of singularities of U by following the proof of Theorem 10.1.10, the first step is to refine
the cone = Cone(e2 , de1 ke2 ) to a fan containing the 2-dimensional cones
= Cone(e2 , e1 )
and
= Cone(e1 , de1 ke2 ).
The first is smooth, but the second may not be. However, we saw in the proof of
Theorem 10.1.10 that the cone has parameters k, k1 satsifying
d = b1 k k1 ,
468
where b1 2, 0 k1 < k as in (10.1.1). We used slightly different notation before,

writing s rather than b1 and l rather than k1 ; the new notation will help us keep
track of what happens as we continue the process and refine the cone .
Using the normalized basis for N relative to , we insert a new ray and obtain
a new smooth cone and a second, possibly nonsmooth cone with parameters k1 , k2 ,
where
k = b2 k1 k2
using (10.1.1). Doing this repeatedly yields a modified Euclidean algorithm
d = b1 k k1
(10.2.1)
k = b2 k1 k2
..
.
kr3 = br1 kr2 kr1

kr2 = br kr1
that computes the parameters of the new cones produced as we successively subdivide to produce the fan giving the resolution of singularities. The process terminates with kr = 0 for some r as shown, since as in the usual Euclidean algorithm,
the ki are a strictly decreasing sequence of nonnegative numbers. Also, by (10.1.1),
we have bi 2 for all i.
The equations (10.2.1) can be rearranged:
d/k = b1 k1 /k
(10.2.2)
k/k1 = b2 k2 /k1
..
.
kr3 /kr2 = br1 kr1 /kr2
kr2 /kr1 = br
and spliced together to give a type of continued fraction expansion for the rational
number d/k, with minus signs:
(10.2.3)
d/k = b1
1
b2
1
br
This is the Hirzebruch-Jung continued fraction expansion of d/k. For obvious

typographical reasons, it is desirable to have a more compact way to represent
these expressions. We will use the notation
d/k = [[b1 , b2 , . . . , br ]].
469
The integers bi are the partial quotients of the Hirzebruch-Jung continued fraction,
and the truncated Hirzebruch-Jung continued fractions
[[b1 , b2 , . . . , bi ]],
1 i r,
are the convergents.

Example 10.2.1. Consider the rational number 17/11. The Hirzebruch-Jung continued fraction expansion is
17/11 = [[2, 3, 2, 2, 2, 2]],
as may be verified directly using the modified Euclidean algorithm (10.2.1).
Proposition 10.2.2. Let d > k > 0 be integers with gcd(d, k) = 1 and let d/k =
[[b1 , . . . , br ]]. Define sequences Pi and Qi recursively as follows. Set
(10.2.4)
and for all 2 i r, let
(10.2.5)
P0 = 1,
Q0 = 0
P1 = b1 ,
Q1 = 1,
Pi = bi Pi1 Pi2
Qi = bi Qi1 Qi2 .
Then the Pi , Qi satisfy:

(a) The Pi and Qi are increasing sequences of integers.
(b) [[b1 , . . . , bi ]] = Pi /Qi for all 1 i r.
(c) Pi1 Qi PiQi1 = 1 for all 1 i r.
(d) The convergents form a strictly decreasing sequence:

d
Pr
Pr1
P1
=
<
< <
.
k
Qr Qr1
Q1
Proof. The proof of part (a) is left to the reader (Exercise 10.2.1).
To prove part (b), first observe that the expression on the right side of (10.2.3)
makes sense when the b j are any rational numbers (not just integers) such that all
denominators in (10.2.3) are nonzero. We will show that the sequences defined by
(10.2.5) satisfy
Ps
[[b1 , . . . , bs ]] =
Qs
for all such lists b1 , . . . , bs . The proof is by induction on the length s of the list.
When s = 1, we have [[b1 ]] = b1 = QP11 by (10.2.4). Now assume that the result has
been proved for all lists of of length t and consider the expression
[[b1 , . . . , bt+1 ]] = [[b1 , . . . , bt
1
bt+1
]],
470
where the right side comes from a list of length t. By the induction hypothesis, this
equals

1
Pt1 Pt2
bt bt+1

.
1
Qt1 Qt2
bt bt+1
By the recurrences (10.2.5), this equals

1
Pt bt+1
Pt1
1
Qt bt+1 Qt1
bt+1 Pt Pt1
Pt+1
=
,
bt+1 Qt Qt1 Qt+1
which is what we wanted to show.

Part (c) will be proved by induction on i. The base case i = 1 follows directly
from (10.2.4). Now assume that the result has been proved for i s, and consider
i = s + 1. Using the recurrences (10.2.5), we have
Ps Qs+1 Ps+1 Qs = Ps (bs Qs Qs1 ) (bs Ps Ps1)Qs
= Ps1 Qs Ps Qs1
=1
by the induction hypothesis.

Finally, from part (b), for each 1 i r 1, we have
Pi1
Pi
1
=
+
.
Qi1 Qi Qi1 Qi
Hence
Pi
Pi1
<
Qi Qi1
since Qi1 Qi > 0 by part (a). Hence part (d) follows.
Hirzebruch-Jung Continued Fractions and Resolutions. When is a cone with

parameters d > k > 0, the process of computing the Hirzebruch-Jung continued
fraction of d/k yields a convenient method for finding a refinement of such
that : X U is a toric resolution of singularities.
Theorem 10.2.3. Let = Cone(e2 , de1 ke2 ) be in normal form. Let u0 = e2 and
use the integers Pi and Qi from Proposition 10.2.2 to construct vectors
ui = Pi1 e1 Qi1 e2 ,
1 i r + 1.
i = Cone(ui1 , ui ),
1 i r + 1,
Then the cones

have the following properties:
(a) Each i is a smooth cone and ui1 , ui are its ray generators.
(b) For each i, i+1 i = Cone(ui ).
471
(c) 1 r+1 = , so the fan consisting of the i and their faces gives a
smooth refinement of .
(d) The toric morphism : X U is a resolution of singularities.
Proof. Both statements in part (a) follow easily from part (c) of Proposition 10.2.2.
For part (b), we note that the ratio Qi1 /Pi1 represents the slope of the line
through ui in the coordinate system relative to the normalized basis e1 , e2 for . By
part (d) of Proposition 10.2.2, these slopes form a strictly decreasing sequence for
i 0, which implies the statement in part (b).
Part (c) follows from part (b) by noting that u0 = e2 and Pr /Qr = d/k, so
ur+1 = de1 ke2 . Hence the cones i fill out .
Part (d) now follows by the reasoning used in Examples 10.1.8 and 10.1.5.
Example 10.2.4. Consider the cone = Cone(e2 , 7e1 5e2 ) in normal form. To
construct the resolution of singularities of the affine toric surface U , we simply
compute the Hirzebruch-Jung continued fraction expansion of the rational number
d/k = 7/5 using the modified Euclidean algorithm:
7 = 253
5 = 231
3 = 3 1.
Hence b0 = b1 = 2, b2 = 3, and
7/5 = [[2, 2, 3]].
(10.2.6)
Then from Proposition 10.2.2 we have

P0 = 1,
P1 = 2,
P2 = b2 P1 P0 = 3,
P3 = b3 P2 P1 = 7,
Q0 = 0
Q1 = 1
Q2 = b2 Q1 Q0 = 2
Q3 = b3 Q2 Q1 = 5.
Theorem 10.2.3 gives the vectors

u0 = e2 , u1 = e1 , u2 = 2e1 e2 , u3 = 3e1 2e2 , u4 = 7e1 5e2
and the cones
1 = Cone(e2 , e1 )
(10.2.7)
2 = Cone(e1 , 2e1 e2 )
3 = Cone(2e1 e2 , 3e1 2e2 )
4 = Cone(3e1 2e2 , 7e1 5e2 )

shown in Figure 3 on the next page. The cones i give a smooth refinement of .
You will see another example of this process in Exercise 10.2.2.
472
3
4
Figure 3. The refinement with open circles at ui = Pi1 e1 Qi1 e2 in Example 10.2.4
Next we show that the vectors ui from Theorem 10.2.3 determine the partial
quotients in the Hirzebruch-Jung continued fraction expansion of d/k.
Theorem 10.2.5. Let = Cone(e2 , de1 ke2 ) be in normal form, and let
d/k = [[b1 , b2 , . . . , br ]].
Then the vectors u0 , u1 , . . . , ur+1 constructed in Theorem 10.2.3 satisfy
(10.2.8)
ui1 + ui+1 = bi ui ,
bi 2,
for 1 i r.
Proof. By the recurrences (10.2.5),
ui1 + ui+1 = (Pi2 e1 Qi2 e2 ) + (Pi e1 Qi e2 )
= (Pi2 + Pi)e1 (Qi2 + Qi)e2
= bi (Pi1 e1 Qi1 e2 ) = bi ui .
Later in this chapter we will see several important consequences of (10.2.8)

connected with the geometry of smooth toric surfaces.
The nonuniqueness of Proposition 10.1.3 has a nice relation to Theorem 10.2.3.
For instance, Example 10.2.4 used the Hirzebruch-Jung expansion 7/5 = [[2, 2, 3]].
Since 53 1 mod 7, the cone of Example 10.2.4 also has parameters d = 7, k = 3.
We leave it to the reader to check that
7/3 = [[3, 2, 2]],
with the partial quotients the same as those in (10.2.6), but listed in reverse order.
This pattern holds for all Hirzebruch-Jung continued fractions. We give a proof
that uses the properties of the associated toric surfaces.
473
Proposition 10.2.6. Let 0 < k, e

k < d and assume ke
k 1 mod d. If the HirzebruchJung continued fraction expansion of d/k is
d/k = [[b1 , b2 , . . . , br ]],
then the Hirzebruch-Jung continued fraction expansion of d/ e

k is
d/ e
k = [[br , br1 , . . . , b1 ]].
Proof. Let = Cone(e2 , de1 ke2 ) and

e = Cone(e2 , de1 e
ke2 ) be the corree
sponding cones in normal form. Since kk 1 mod m, there is an integer de such
that d de+ ke
k = 1. The Z-linear mapping : N N defined with respect to the basis
e1 , e2 by the matrix
!
e
k
d
A= e
d k
is bijective, maps
e to , and is orientation-reversing. Thus (de1 e
ke2 ) = e2 and
(e2 ) = de1 ke2 . If we apply Theorem 10.2.3 to , then we obtain vectors ui
satisfying the equations
ui1 + ui+1 = bi ui
for all 1 i r. We claim that when we apply the mapping 1 defined by the
inverse of the matrix A above, then the vectors ui are taken to corresponding vectors uei for the cone
e. But since and 1 are orientation-reversing, the partial
quotients in the Hirzebruch-Jung continued fraction will be listed in the opposite
order. You will complete the proof of this assertion in Exercise 10.2.3.

Hilbert Bases and Convex Hulls. Our next result gives two alternative ways to
understand the vectors ui in Theorem 10.2.3. The idea is that gives two objects:
The semigroup N. Since is strongly convex, its irreducible elements
form the unique minimal generating set called the Hilbert basis of N. (See
Proposition 1.2.23.)
The convex hull = Conv( (N \ {0})). This is an unbounded polygon in
the plane whose bounded edges contain finitely many lattice points.
Example 10.2.7. Consider the cone = Cone(e2 , 7e1 5e2 ) from Example 10.2.4.
Figure 4 on the next page shows the convex hull , where the white circles represent the lattice points on the bounded edges. We will see below that these lattice
points give the Hilbert basis of N.
Here is the general result suggested by Example 10.2.7.

Theorem 10.2.8. Let = Cone(e2 , de1 ke2 ) be in normal form and let
S = {u0 , u1 , . . . , ur+1 }
be the set of vectors constructed in Theorem 10.2.3. Then:
474
Figure 4. Convex hull and lattice points on bounded edges in Example 10.2.7
(a) S is the Hilbert basis of the semigroup N.
(b) S is the set of lattice points on the bounded edges of .
Proof. For part (a), we use the notation of Theorem 10.2.3, where the cone i is
generated by ui1 , ui . Then i N is generated as a semigroup by ui1 , ui since i
is smooth. Using = 1 r+1, one sees easily that S generates N.
We claim next that all the ui are irreducible elements of N. This is clear for
u0 = e2 and ur+1 = de1 ke2 since they are the ray generators for . If 1 i r and
ui is not irreducible, then ui would have to be a linear combination of the vectors
in S \ {ui } with nonnegative integer coefficients, i.e.,
X

X

X
c j Pj1 e1
c j Q j1 e2
ui = Pi1 e1 Qi1 e2 =
c ju j =
j6=i
with c j 0 in Z. Hence
Pi1 =
X
j6=i
c j Pj1 ,
j6=i
Qi1 =
j6=i
c j Q j1 .
j6=i
Since the Pi and Qi are strictly increasing by part (a) of Proposition 10.2.2, we must
have c j = 0 for all j > i. But this would imply that ui is a linear combination with
nonnegative integer coefficients of the vectors in {u0 , . . . , ui1 }. This contradicts
the observation made in the proof of Theorem 10.2.3 that the slopes of the ui are
strictly decreasing. It follows that the ui are irreducible elements of N.
Finally, we must show that there are no other irreducible elements in N.

But this follows from what we have already said. Since = 1 r+1 , if u
is irreducible, then u i N for some i. But then u = ci1 ui1 + ci ui for some
ci1 , ci 0 in Z. Thus u is irreducible only if u = ui1 or ui .
475
For part (b), first observe that by Proposition 10.2.2, we have

Pi2 Qi PiQi2 = Pi2 (bi Qi1 Qi2 ) (bi Pi1 Pi2 )Qi2
= bi (Pi2 Qi1 Pi1 Qi2 ) = bi 2.
Combining this with part (c) of Proposition 10.2.2, one obtains the inequality
(Qi1 Qi2 ) (Qi Qi1 )
.
Pi1 Pi2
Pi Pi1
Since ui = Pi1 e1 Qi1 e2 , this inequality tells us that the slopes of the line segments ui1 ui and ui ui+1 are related by
slope of ui1 ui slope of ui ui+1 .
This implies that these line segments lie on boundary of . From here, it is easy
to see that the ui are the lattice points of the bounded edges of .

Ordinary Continued Fractions. The Hirzebruch-Jung continued fractions studied
above are less familiar than ordinary continued fraction expansions in which the
minus signs are replaced by plus signs. If d > k > 0 are integers, then the ordinary continued fraction expansion of d/k may be obtained by performing the
same sequence of integer divisions used in the usual Euclidean algorithm for the
gcd. Starting with k1 = d and k0 = k, we write ai for the quotient and ki for the
remainder at each step, so that the ith division is given by
(10.2.9)
ki2 = ai ki1 + ki ,
where 0 ki < ki1 .
Let ks1 be the final nonzero remainder (which equals gcd(d, k)). The resulting
equations splice together to form the ordinary continued fraction
1
d/k = a1 +
a2 +
1
+
1
as
To distinguish these from Hirzebruch-Jung continued fractions, we will use the

notation
(10.2.10)
d/k = [a1 , a2 , . . . , as ]
for the ordinary continued fraction. The ai are the (ordinary) partial quotients of
d/k, and the truncated continued fractions
[a1 , a2 , . . . , ai ],
are the (ordinary) convergents of d/k.
1 i s,
476
Example 10.2.9. By Example 10.2.1, the Hirzebruch-Jung continued fraction of

17/11 is
17/11 = [[2, 3, 2, 2, 2, 2]],
and the ordinary continued fraction is
17/11 = [1, 1, 1, 5].
The partial quotients and the lengths are different. However, each expansion determines the other, and there are methods for computing the Hirzebruch-Jung partial
quotients b j in terms of the ordinary partial quotients ai and vice versa. See [75,
Prop. 3.6], [145, p. 257], [231, Prop. 2.3], and Exercise 10.2.4.
The following result is mostly parallel to Proposition 10.2.2, but shows that
ordinary continued fractions are slightly more complicated than Hirzebruch-Jung
continued fractions. The proof is left to the reader (Exercise 10.2.5).
Proposition 10.2.10. Let d > k > 0 be integers with gcd(d, k) = 1, and let d/k =
[a1 , . . . , as ]. Define sequences pi and qi recursively as follows. First set
(10.2.11)
and for all 2 i s, let
(10.2.12)
p0 = 1,
q0 = 0
p1 = a1 ,
q1 = 1,
pi = ai pi1 + pi2
qi = ai qi1 + qi2 .
Then the pi , qi satisfy:

(a) [a1 , . . . , ai ] = pi /qi for all 1 i s.
(b) pi qi1 pi1 qi = (1)i for all 1 i s.
(c) The convergents converge to d/k, but in an oscillating fashion:

p1
p3
d
p4
p2
<
< <
< .
q1
q3
k
q4
q2
Ordinary Continued Fractions and Convex Hulls. Felix Klein discovered a lovely
o , u o of
geometric interpretation of ordinary continued fractions. Given a basis u1
0
N Z2 and relatively prime integers d > k > 0, compute the continued fraction
d/k = [a1 , . . . , as ] and set
(10.2.13)
o
+ pi u0o ,
uio = qi u1
1 i s.
o + du o ,
In this notation, the superscript o stands for ordinary. Then uso = ku1
0
and part (c) of Proposition 10.2.10 implies that the uio lie on one side of the ray
Cone(uso ) for even indices and on the other side for odd indices. To give a careful
description of what is happening, we introduce the cones
o
, uso ), 0 = Cone(u0o , uso )
1 = Cone(u1
and associated convex hulls i = i = Conv(i (N \ {0})), i = 1, 0.
477
0
uo3
0
1
u2o
1
u0o
u1o
o
u1
Figure 5. The cones 1 , 0 , the convex hulls 1 , 0 , and their vertices
Example 10.2.11. For d = 7, k = 5, the expansion 7/5 = [1, 2, 2] gives the vectors
o , . . . , u o shown in Figure 5. In this figure, it is clear that the u o are the vertices
u1
i
3
of the convex hulls 1 and 0 .
This example is a special case of the following general result.

o , . . . , u o and , as above, we have:
Theorem 10.2.12. For u1
1
0
s
(a) 1 has vertex set {u2o j1 | 1 j s/2} {uso }.
(b) 0 has vertex set {u2o j | 0 j s/2} {uso }.
o u o is an edge of
(c) For 1 i s, ui2
1 (resp. 0 ) for i odd (resp. even) with
i
ai + 1 lattice points.
Proof. First note that by Proposition 10.2.10, the vectors uio satisfy the recursion
(10.2.14)
o
o
,
+ ui2
uio = ai ui1
1 i s.
o is primitive (Proposition 10.2.10), part (c) follows from parts (a) and (b).
Since ui1
We now prove the theorem using induction on the length s of the continued
fraction expansion. Consider Figure 6 on the next page, which shows the first quado , u o , together with the vector u o . In the picture, (darker)
rant determined by u1
0
s
0
lies above the ray determined by uso and 1 (lighter) lies below. The figure also
o , u o ).
shows u1o and the smaller cone 1 = Cone(u1o , uso ) 1 = Cone(u1
s
o + du o , the ray starting from u o through u o passes through

Since uso = ku1
1
1
0
o + d/ku o and u o + (d/k + 1)u o .
the upper edge of 1 at a point between u1
1
1
1
But by the computation of the ordinary continued fraction, d/k = a1 . Therefore
o to u o is the first bounded edge of . It follows that
the segment from u1
1
1
o
, u1o )
1 Cone(u1
478
1
uso
1
u0o
o
u1o = a1 u0o + u1
o
u1
o
Figure 6. The cones 0 , 1 1 and vectors u1
, u0o , u1o , uso
o , u o . It remains to understand and

has vertices u1
0
1 1 . This is where
1
induction comes into play.
Apply the above construction to the new basis u0o , u1o and the continued fraction
k
=
d a1 k
d
k
1
= [a2 , . . . , as ]
a1
of length s 1. If we start numbering at 0 rather than at 1, then the vectors

u0o , u1o , . . . , uso are the same as before by the recursion (10.2.14). This gives the
cones 0 , 1 shown in Figure 6. By induction, the vectors u0o , u1o , . . . , uso give
the vertices of the corresponding convex hulls 0 = Conv(0 (N \ {0})) and
1 = Conv(1 (N \ {0})). It follows easily that the theorem holds for continued
fraction expansions of length s.

A discussion of Kleins formulation of Theorem 10.2.12 can be found in [231].
In Exercise 10.2.6 you will apply the theorem to the resolution of pairs of singular
points on certain toric surfaces. Geometric pictures similar to Figure 5 have also
appeared in recent work of McDuff on symplectic embeddings of 4-dimensional
ellipsoids (see [202]).
Ordinary Continued Fractions and the Supplementary Cone. To relate the above
theorem to toric geometry, we follow the approach of [231]. Given a cone =
Cone(e2 , de1 ke2 ) NR R2 in normal form, its supplement is the cone =
Cone(e2 , de1 ke2 ). Thus is the right half-plane, and the cones , give
the convex hulls
= Conv( (N \ {0}))
= Conv( (N \ {0})).
Example 10.2.13. When d = 7, k = 5, Figure 7 on the next page shows the cones
, and the convex hulls , . The open circles in the figure are the vertices
479
e2
e1
7e1 5e2
Figure 7. The cones , , the convex hulls , , and their vertices
of and . Also observe that the fourth quadrant portion of Figure 7 becomes
Figure 5 after a 90 counterclockwise rotation.
For = Cone(e2 , de1 ke2 ), the ordinary continued fraction expansion d/k =
[a1 , a2 , . . . , as ] gives the sequences pi , qi , 0 i s, defined in Proposition 10.2.10.
Then define the vectors
(10.2.15)
o
u1
= e2 , uio = pi e1 qi e2 , 0 i s.
These vectors enable us to describe the vertices of , as follows.

Theorem 10.2.14. Let = Cone(e2 , de1 ke2 ) be a cone in normal form with
supplement . Also let uio , 1 i s, be as defined in (10.2.15). Then:
(a) The set of vertices of is
{u2o j1 | 1 j s/2} {uso }.

(b) If a1 = 1, then the set of vertices of is
{e2 } {u2o j | 1 j s/2} {uso }.
(c) If a1 > 1, then the set of vertices of is
{e2 } {u2o j | 0 j s/2} {uso }.
o = e and u o = e , we can rewrite (10.2.15) as
Proof. First note that since u1
2
1
0
uio = pi e1 qi e2 = qi u1 + pi u0 ,
0 i s.
Thus we are in the situation of Theorem 10.2.12, where we have the cones 1 =
o , u o ) and = Cone(u o , u o ) and associated convex hulls
Cone(u1
1 = 1
0
s
0 s
o , u o , . . . , u o give
and 0 = 0 . Then Theorem 10.2.12 implies that the vectors u1
s
0
the vertices of 1 and 0 .
480
However, = Cone(e1 , e2 ) 0 and = 1 . In particular, = 1 , so

that part (a) of the theorem follows immediately. For parts (b) and (c), note that
= ( Cone(e1 , e2 )) 0 .
The intersection Cone(e1 , e2 ) has vertices e1 = u0o , e2 , while 0 has vertices

u0o , u2o , . . . , uso . If a1 = 1, then e2 , u0o , u2o are collinear, so that u0o is not a vertex.
This proves part (b) of the theorem. Finally, if a1 > 1, then one can prove without

difficulty that u0o is a vertex (Exercise 10.2.7), and part (c) follows.
Example 10.2.15. Figure 7 above illustrates Theorem 10.2.12 for the cone =
Cone(e2 , 7e1 5e2 ). Since 7/5 = [1, 2, 2], we use part (b) of the theorem.
Ordinary Continued Fractions and the Dual Cone. For a cone in normal form,
one surprise is that its supplement is essentially the dual of .
Lemma 10.2.16. Given a cone = Cone(e2 , de1 ke2 ) NR in normal form, its
supplementary cone NR is isomorphic to MR .
Proof. Let e1 , e2 be the basis of M dual to the basis e1 , e2 of N. Then the isomorphism defined by e1 7 e2 and e2 7 e1 takes = Cone(e2 , de1 ke2 ) to
Cone((e1 ), d(e2 ) k(e1 )) = Cone(e1 , ke1 + de2 ) = .
The isomorphism from this lemma leads to some nice results about
and the associated convex hull = Conv( (M \ {0})). Specifically, this
isomorphism takes the vectors
o
= e2 , uio = pi e1 qi e2 ,
u1
0is
from (10.2.15) to the dual vectors

m1 = e1 , mi = qi e1 + pi e2 ,
ke1 + de2 ,
0 i s.
In particular, ms =
so that
= Cone(m1 , ms ) in this notation. Then
Theorem 10.2.14 implies that the vertex set of the convex hull is
{m2 j1 | 1 j s/2} {ms }.
(10.2.16)
Hence, in the language of 7.1, MR is a full dimensional lattice polyhedron.

We can describe its recession cone and normal fan as follows.
Proposition 10.2.17. Let = Cone(e2 , de1 ke2 ) NR R2 be in normal form
and set P = = Conv( (M \ {0})). Then:
(a) P MR is a full dimensional lattice polyhedron with recession cone .
(b) The normal fan P of P is the refinement of obtained by adding the minimal
generators
o
u0o , u2o , u4o , . . . , us1
(s odd)
u0o ,
(s even).
u2o ,
u4o ,
...,
o
o
us2
, uso us1
481
(c) The toric morphism XP U is projective and XP is Gorenstein with at worst

rational double points.
o
Remark 10.2.18. When s is even, we can explain uso us1
as follows. By Theoo
o
rem 10.2.14, us2 and us give an edge . The vector determined by this edge is
o = a u o . Hence the edge has a + 1 lattice points since u o
uso us2
s s1
s
s1 is primitive.
Thus, if we start from uso , the next lattice point along the edge is
o
uso us1
.
Since d > k > 0 and d/k = [a1 , . . . , as ] is computed using the Euclidean algorithm,
o
is not a vertex of .
we have as 2 (Exercise 10.2.8). It follows that uso us1
Proof. Part (a) is straightforward (Exercise 10.2.9). For part (b), we will assume
that s is even and leave the case when s is odd to the reader (Exercise 10.2.9).
If we write s = 2, then the vertices (10.2.16) of P are m1 , m1 , . . . , m21 , m2 .
Thus the bounded edges of P = are
m1 m1 , m1 m3 , . . . , m23 m21 , m21 m2 .
The inward-pointing normals of these edges give the rays that refine in the normal
fan of P.
The mi s satisfy the same recursion mi = ai mi1 + mi2 , 1 i s, as the uio s
in (10.2.14). Hence, for the edges m2 j1 m2 j+1 , 1 j s/2, we have
m2 j+1 m2 j1 = a2 j+1 m2 j = a2 j+1 (q2 j e1 + p2 j e2 ).
Since u2o j = p2 j e1 q2 j e2 , one easily computes that hm2 j+1 m2 j1 , u2o j i = 0. It
follows that u2o j is the inward-pointing normal of this edge since it lies in and is
primitive. This takes care of all of the bounded edges except for m21 m2 . Here,
we compute
m2 m21 = (q2 q21 )e1 + (p2 p21 )e2 .
o uo
This is clearly normal to u2
21 = (p2 p21 )e1 (q2 q21 )e2 . The latter
vector is easily seen to be primitive by part (b) of Proposition 10.2.10. Furthermore,
o uo
u2
21 by Remark 10.2.18. Hence this is the inward-pointing normal of
the final bounded edge m21 m2 .
For part (c), note that XP U is projective by Theorem 7.1.10. To complete

the proof, we need to show that each maximal cone of P gives a Gorenstein affine
toric variety. For simplicity, we assume that s is odd (see Exercise 10.2.9 for the
even case). Since = Cone(e2 , de1 ke2 ) = Cone(e2 , uso ), the maximal cones of
o , u o ), plus the
P consist of two boundary cones Cone(e1 , u0o ) and Cone(us1
s
interior cones Cone(u2 j2 , u2 j ) P , 1 j s/2. The boundary cones are
easily seen to be smooth and hence Gorenstein (Exercise 10.2.9). For an interior
482
cone Cone(u2 j2 , u2 j ), we use part (b) of Proposition 10.2.10 to compute

hm2 j1 , u2o j2 i = hq2 j1 e1 + p2 j1 e2 , p2 j2 e1 q2 j2 e2 i
= (p2 j1 q2 j2 p2 j2 q2 j1 ) = (1)2 j1 = 1,
and a similar computation gives hm2 j1 , u2o j i = p2 j q2 j1 p2 j1 q2 j = (1)2 j = 1.

By Proposition 8.2.12, we conclude that the corresponding affine toric variety is
Gorenstein, as desired. Then the singular points of XP are rational double points
by Example 10.1.5 and Proposition 10.1.6.

In 11.3 we will revisit this result, where we will learn that the morphism
XP U from Proposition 10.2.17 is the blowup of the singular point of U .
Just as the morphism XP U is projective, one can show more generally

that the resolution of singularities X U from Theorem 10.2.3 is a projective
morphism. This requires finding a lattice polyhedron with the correct normal fan.
You will explore one way of doing this in Exercise 10.2.10.
We next consider the Hilbert basis H of M. Recall from Lemma 1.3.10
that |H | is the dimension of the Zariski tangent space at the singular point of U
and is the dimension of the most efficient embedding of U into affine space.
Theorem 10.2.8 tells us that the Hilbert basis of N is computed using the
Hirzebruch-Jung continued fraction expansion of d/k. Since has parameters
d, d k (part (b) of Exercise 10.1.2), it follows that we need the Hirzebruch-Jung
continued fraction expansion of d/(d k) to get the Hilbert basis of M. By
Exercise 10.2.4, the ordinary continued fraction
d/k = [a1 , . . . , as ]
gives the Hirzebruch-Jung continued fraction
(
[[(2)a1 1 , a2 + 2, (2)a3 1 , a4 + 2, . . . , (2)as1 1 , as + 1]] s even
d/(d k) =
[[(2)a1 1 , a2 + 2, (2)a3 1 , a4 + 2, . . . , as1 + 2, (2)as 1 ]] s odd.
Theorem 10.2.8, applied to this expansion, gives the Hilbert basis of M. To see
the underlying geometry, we need the following observation (Exercise 10.2.11):
In Theorem 10.2.5, three consecutive lattice points ui1 , ui , ui+1
(10.2.17)
are collinear if and only if ui1 + ui+1 = 2ui , i.e., bi = 2.
This means that a string of consecutive 2s in a Hirzebruch-Jung continued fraction
of d/(d k) gives a string of lattice points in the relative interior of a bounded
edge of the convex hull . For the vertices m1 , m1 , . . . of , this gives two
ways to think about lattice points on an edge connecting two adjacent vertices. For
example, the edge m1 m1 has a1 1 lattice points in its relative interior because:
The Hirzebruch-Jung expansion of d/(d k) starts with a1 1 consecutive 2s.
m1 m1 = a1 u0 , u0 primitive, gives a1 + 1 lattice points on the edge.
This pattern continues for the other bounded edges of .
483
Oda gives a different argument for this pattern in [218, Sec. 1.6] and uses it to
relate the Hirzebruch-Jung continued fractions for d/k and d/(d k). His relation
also follows from our approach (see part (d) of Exercise 10.2.4). The connection
between continued fractions and toric surfaces is surprisingly rich and varied and
is one of the reasons why toric varieties are so much fun to study. See [75] and
[231] for a further discussion of this wonderful topic.
Exercises for 10.2.
10.2.1. Prove part (a) of Proposition 10.2.2: Show that the sequences Pi and Qi from
(10.2.5) are increasing sequences of nonnegative numbers. Hint: Use bi 2 for all i.
10.2.2. In 10.1, we constructed several resolutions of singularities in a rather ad hoc way.
In this exercise, we will see that the resolutions given by Theorem 10.2.3 are the same as
what we saw before.
(a) When has parameters d, 1, show that the Hirzebruch-Jung continued fraction method
gives the same resolution of U as the one given in Example 10.1.8.
(b) Do the same for Example 10.1.9. Hint: First show that
d
= [[2, 2, . . . , 2]],
d1
where there are d 1 2s.
10.2.3. Verify the last claim in the proof of Proposition 10.2.6.
10.2.4. This exercise will consider some relations between ordinary continued fraction
expansions and Hirzebruch-Jung continued fraction expansions.
(a) Given integers a1 , a2 > 0 and a variable x, prove that
[a1 , a2 , x] = [[a1 + 1, (2)a2 1 , x + 1]],
where for any l 0, (2)l denotes a string of l 2s. Hint: Argue by induction on a2 .
(b) Use part (a) to prove the equality [1, 1, 1, 5] = [[2, 3, 2, 2, 2, 2]] from Example 10.2.9.
(c) Given d/k = [a1 , . . . , as ], prove that d/(d k) has the Hirzebruch-Jung expansion
given in the discussion leading up to (10.2.17). Hint: You will want to consider the
cases a1 = 1 and a1 > 1 separately, but the formula can be written as in (10.2.17) in
either case. If you get stuck, see [231].
(d) Starting from the ordinary continued fraction for d/k, useP
parts (a) and (c)P
to show that
r
s
if d/k = [[b1 , . . . , br ]] and d/(d k) = [[c1 , . . . , cs ]], then ( i=1 bi ) r = ( j=1 c j ) s.
10.2.5. In this exercise we will consider Proposition 10.2.10.

(a) Prove the proposition. Hint: For part (a), argue by induction on the length s 1 of
the expansion. The expression [a1 , a2 , . . . , ai ] is well-defined when the ai are positive
rational numbers, so we can write
[a1 , a2 , . . . , ai1 , ai ] = [a1 , a2 , . . . , ai1 + 1/ai].
Then use (10.2.12) and follow the reasoning from the proof of Proposition 10.2.2.
(b) Suppose we modify the initialization and the recurrences (10.2.12) as follows. Let
(r1 , s1 ) = (1, 0) and (r0 , s0 ) = (0, 1).
484
Then, for all 1 i s, compute
ri = ri2 ai ri1
si = si2 ai si1
(note the change in sign!). What is true about ri d + si k for all i? What do we get with
i = s? Hint: This fact is the basis for the extended Euclidean algorithm.
10.2.6. In this exercise, you will show that the ordinary continued fraction expansion of a
rational number d/k can be used to construct simultaneous resolutions of pairs of singularities of certain toric surfaces. Let 1 < k < d be relatively prime integers, and let P be the
triangle Conv(0, de1 , ke2 ) in R2 .
(a) Draw the normal fan P and show that XP has exactly two singular points. Note that
XP is isomorphic to the weighted projective plane P(1, k, d).
(b) Adapt Theorem 10.2.12 to produce a resolution of singularities of XP from the ordinary continued fraction expansion of d/k. Hint: First refine P by introducing 1dimensional cones Cone(e1 ) and Cone(e2 ). Then apply Theorem 10.2.12 to the
third quadrant of your drawing.
10.2.7. Complete the proof of Theorem 10.2.14 by showing that u0o is a vertex of if
and only if a1 > 1.
10.2.8. For relative prime integers d > k > 0, we used the Euclidean algorithm to construct
d/k = [a1 , . . . , as ]. Prove that as 2.
10.2.9. Prove part (a) of Proposition 10.2.17. Also prove part (b) for s odd and part (c) for
s even.
10.2.10. In this exercise, given a cone in normal form with parameters d, k, you will
see how to construct an unbounded polyhedron P in MR R2 whose recession cone is
= Cone(e1 , ke1 + de2 ), and whose normal fan defines the resolution of U from Theorem 10.2.3. Let Pi , Qi be the sequences constructed in Proposition 10.2.2. Let m be the
smallest positive integer such that md > P1 + + Pr1 and mk > Q1 + + Qr1 . Starting
from the point Ar = (mk, md) on the upper boundary ray of , construct the points
Ar1 = (mk Qr1 , md Pr1 )
Ar2 = (mk Qr1 Qr2 , md Pr1 Pr2)

..
.
A0 =
Pr1
i=1 Qi , md
Pr1
(mk i=1 Qi , 0).
A1 = (mk
Pr1
i=1
Pi )
Then let P be the polyhedron with one edge along the positive x-axis starting at A0 , edges
Ai Ai+1 for i = 0, . . . , r 1, and one edge along the upper edge of starting from Ar .
(a) Draw the polyhedron P for the cone with parameters (d, k) = (7, 5). What is the integer
m in this case?
(b) Show that is the recession cone of P.
(c) Show that the normal fan of P is the fan giving the resolution of U constructed in
Theorem 10.2.3.
10.2.11. Prove (10.2.11).
10.3. Grobner Fans and M cKay Correspondences
485
10.2.12. Let d/k be a rational number in lowest terms with 0 < k < d, and let
d/k = [[b1 , . . . , br ]] = [a1 , . . . , as ]
be its continued fraction expansions. You will show that the sequences Pi , Qi and pi , qi
considered in this section can be expressed in matrix form.

b 1
(a) Let M (b) =

. Show that for all 1 i r,
1
0

Pi Pi1
= M (b1 )M (b2 ) M (bi ).
Qi Qi1

a 1
(b) Let M + (a) =
. Show that for all 1 i s,
1 0

pi pi1
= M + (a1 )M + (a2 ) M + (ai ).
qi qi1
10.2.13. In this exercise, you will apply the results of this section to the weighted projective
plane P(q0 , q1 , q2 ) from 2.0 and Example 3.1.17.
(a) Construct a resolution of singularities for any P(1, 1, q2 ), where q2 2. What smooth
toric surface is obtained in this way? A complete classification of the smooth complete
toric surfaces will be developed in 10.4.
(b) Do the same for P(1, q1 , q2 ) in general. Hint: Exercise 10.2.6.
10.3. Grobner Fans and McKay Correspondences

The fans obtained by resolving the singularities of the affine toric toric surfaces U
have unexpected descriptions that involve Grobner bases and representation theory.
In this section we will present these ideas, following [156] and [157].
Grobner Bases and Grobner Fans. We assume the reader knows about Grobner
bases (see [69]) and Grobner fans (see [70, Ch. 8, 4] or [264]). A nonzero ideal
I C[x1 , . . . , xn ] has a unique reduced Grobner basis with respect to each monomial order > on the polynomial ring. However, the set of distinct reduced marked
Grobner bases for I (i.e., reduced Grobner bases with marked leading terms in each
polynomial) is finite. Hence the ideal I has a finite universal Grobner basis, i.e., a
finite subset U I that is a Grobner basis for all monomial orders simultaneously.
Let w Rn0 be a weight vector in the positive orthant (so w could be taken as
the first row of a weight matrix defining a monomial order). Let
G = {g1 , . . . , gt }
be one of the reduced marked Grobner bases for I, where

P
gi = x (i) + ci x ,
and x (i) is marked as the leading term of gi . If w (i) > w whenever ci 6= 0,

then I will have Grobner basis G with respect to any monomial order defined by a
486
weight matrix with first row w. The set

(10.3.1)
CG = {w Rn0 | w (i) w whenever ci 6= 0}
is the intersection of a finite collection of half-spaces, hence has the structure of a

closed convex polyhedral cone in Rn0 . The cones CG as G runs over all distinct
marked Grobner bases of I, together with all of their faces, have the structure of a
fan in Rn0 called the Grobner fan of I. In particular, for each pair G , G of marked
Grobner bases, the cones CG and CG intersect along a common face where the
w-weights of terms in some polynomials in G (and in G ) coincide.
A First Example. Let be a cone in normal form with parameters d, k, and recall
gcd(d, k) = 1 by hypothesis. By Proposition 10.1.2, the group
Gd,k = {(, k ) (C )2 | d = 1, 0 k d 1} d
acts on C2 by componentwise multiplication

(10.3.2)
(, k ) (x, y) = ( x, k y),
with quotient C2 /Gd,k U .
Let I(Gd,k ) be the ideal defining Gd,k as a variety in C2 . In the next extended
example, we will introduce the first main result of this section.
Example 10.3.1. Let d = 7, k = 5 and I = I(G7,5 ). It is easy to check that
I = hx7 1, y x 5 i.
Moreover, for lexicographic order with y > x, the set
G (1) = {x7 1, y x 5 }
is the reduced marked Grobner basis for I, where the underlines indicate the leading
terms. The corresponding cone in the Grobner fan of I is
CG (1) = {w = (a, b) R20 | b 5a} = Cone(e2 , e1 + 5e2 ).
There are three other marked reduced Grobner bases of I:
G (2) = {x 5 y, x 2 y 1, y 2 x 3 },
G (3) = {x 3 y 2 , x 2 y 1, y3 x},
G (4) = {y7 1, x y3 }.
It is easy to check that each of these sets is a Grobner basis for I using Buchbergers
criterion. The corresponding cones are
CG (2) = {(a, b) R20 | b 5a, 2b 3a} = Cone(e1 + 5e2 , 2e1 + 3e2 ),
CG (3) = {(a, b) R20 | 2b 3a, 3b a} = Cone(2e1 + 3e2 , 3e1 + e2 ),
CG (4) = {(a, b) R20 | 3b a} = Cone(3e1 + e2 , e1 ).
487
Since these three cones fill out rest of the first quadrant in R2 , the Grobner fan of
I consists of the four cones CG (i) and their faces, as shown in Figure 8. We will
denote this fan by in the following.
(1,5)
(2,3)
(3,1)
Figure 8. The Grobner fan
Next, let us consider the resolution of singularities

X U
for the cone with parameters d = 7, k = 5 computed in Example 10.2.4 in the
last section. The reader can check that the linear transformation T : NR NR with
matrix relative to the basis e1 , e2 given by

7 0
(10.3.3)
A=
5 1
maps the cones CG (i) in the Grobner fan to the corresponding cones i in the fan
. The matrix in (10.3.3) is invertible, but its inverse is not an integer matrix. The
image of the lattice N = Z2 under T is the proper sublattice 7Ze1 Ze2 , and T 1
maps N to the lattice

1
5
N = {(a/7, b/7) | a, b Z, b 5a mod 7} = N + Z
e1 + e2 .
7
7
There is an exact sequence
0 N N G 0
induced by the map N (C )2 defined by (a/7, b/7) 7 (e2ia/7 , e2ib/7 ). Letting

= Cone(e1 , e2 ), the corresponding toric morphism U,N U,N is the quotient
mapping C2 C2 /G.
It is easy to check that w0 = (0, 1) and w1 = (1/7, 5/7) form a basis of the
lattice N . In Figure 9 on the next page, the fan defined by the cones with ray
488
u0
(0,1) = w0
u1
(1/7,5/7) = w1
u2
u3
(2/7,3/7) = w2 = 2w1 - w0
(3/7,1/7) = w3 = 3w1 - 2w0
u4
(1,0) = w4 = 7w1 - 5w0

Figure 9. The toric variety X,N is the resolution of U in Example 10.3.1
generators on the left is the same as in Figure 8 above, and the fan on the right is
the same as in Figure 3 above. Note that T (wi ) = ui for 0 i 4 in Figure 9.
With respect to N , the cones in the Grobner fan are smooth cones, and it
follows from the discussion of toric morphisms in 3.3 that the toric surfaces X,N
and X are isomorphic. In other words, the Grobner fan of the ideal I encodes
the structure of the resolution of singularities of U .
Example B.2.4 shows how to compute this example using GFan [161].
A Tale of Two Fans. We next show that the last observation in Example 10.3.1
holds in general. As in the example, consider the ideal I = I(Gd,k ) and the action
of Gd,k given in (10.3.2). Each monomial in C[x, y] is equivalent modulo I to one
of the monomials x j , j = 0, . . . , d 1. This may be seen, for instance, from the
remainders on division by the lexicographic Grobner basis {x d 1, y x k }. As
a result, we have a direct sum decomposition of the coordinate ring of the variety
Gd,k as a C-vector space:
C[x, y]/I
(10.3.4)
d1
M
Vj,
j=0
where V j is the 1-dimensional subspace spanned by x j mod I.

The following result establishes a first connection between the ideal I and the
resolution of singularities of U described in Theorem 10.2.3.
Proposition 10.3.2. Let I = I(Gd,k ) where 0 < k < d and gcd(d, k) = 1. Consider
u0 = e2
ui = Pi1 e1 Qi1 e2 ,
i = 1, . . . , r + 1,
from Theorem 10.2.3. Let T : NR NR be the linear transformation with matrix

d 0
A=
k 1
489
and let wi = T 1 (ui ) for i = 0, . . . , r + 1. Write wi = d1 (ai e1 + bi e2 ) and define

gi = x bi y ai ,
0 i r + 1.
(a) The polynomials g0 , . . . , gr+1 are contained in the ideal I.

(b) S = {ae1 + be2 N | a, b 0, x b y a I} N is an additive semigroup.
(c) {dwi | i = 0, . . . , r + 1} is the Hilbert basis of the semigroup S of part (b).
Proof. It is an easy calculation to show T 1 maps = Cone(e2 , de1 ke2 ) to the

first quadrant R20 . Moreover, w0 = e2 , and for i = 1, . . . , r + 1,

1
wi = Pi1 e1 + (kPi1 dQi1 )e2 .
d
d
Therefore, g0 = x 1 and
gi = x kPi1 dQi1 y Pi1
for i = 1, . . . , r + 1. Since d = 1, these polynomials clearly vanish at (, k ) Gd,k .

Therefore gi I(Gd,k ) = I for all i.
The proof of part (b) is left to the reader as Exercise 10.3.3. For part (c), it
follows from parts (a) and (b) that dwi is contained in S. On the other hand, let
ae1 + be2 S. Then x b y a I, which implies that b ak mod d. Hence
1

b ak
T (ae1 + be2 ) = ae1 +
e2
d
d
must be an element of N. Since the ui are the Hilbert basis for the semigroup
N by Theorem 10.2.8, this vector is a nonnegative integer combination of the
ui . Hence ae1 + be2 is a nonnegative integer combination of the dwi . It follows that
S is generated by the dwi . The dwi are irreducible in S because the corresponding
ui = T (wi ) are irreducible in the semigroup N.

We also have a first result about reduced Grobner bases of the ideal I.
Lemma 10.3.3. Every element of a reduced Grobner basis of I = I(Gd,k ) is either
of the form x b y a or of the form x s yt 1 for s,t > 0.
Proof. Since I is generated by xd 1, y xk , the Buchberger algorithm implies
that a reduced Grobner basis G of I consists of binomials. By taking out common
factors, every g G can be written g = x i y j h, where h = x b y a or x s yt 1. Then
h I since it vanishes on Gd,k . Its leading term divisible by the leading term of an
element of G , which is impossible in a reduced Grobner basis unless g = h.

The Grobner bases in Example 10.3.1 give a nice illustration of Lemma 10.3.3.
Our next lemma relates the polynomials gi = x bi y ai from Proposition 10.3.2 to
the reduced Grobner bases of I.
Lemma 10.3.4. Let gi = x bi y ai be as in Proposition 10.3.2 and fix a monomial
order > on C[x, y].
490
(a) The ai are increasing and the bi are decreasing with i.

(b) There is some index i = i0 (depending on >) such that
LT > (gi ) = x
bi
for all i i0 , and LT > (gi ) = y ai for all i > i0 .
(c) If i = i0 is the index from part (b), then gi0 and gi0 +1 are elements of the reduced
Grobner basis of I = I(Gd,k ) with respect to >.
Proof. You will prove parts (a) and (b) in Exercise 10.3.4. For part (c), let G
be the reduced Grobner basis of I with respect to >. Since gi0 I by part (a) of
Proposition 10.3.2, there is g G whose leading term divides LT > (gi0 ) = x bi0 . By
Lemma 10.3.3, it follows that g = x b y a with LT > (g) = x b . In particular, b bi0
and ae1 + be2 S, where S is the semigroup from part (b) of Proposition 10.3.2.
Then part (c) of the same proposition implies that ae1 + be2 must be a nonnegative
integer combination
P
Pr+1
(10.3.5)
ae1 + be2 = r+1
i N.
i=0 i dwi =
i=0 i (ai e1 + bi e2 ),
Since b bi0 and the bi decrease with i, (10.3.5) can include only the dwi with
i i0 . Suppose that dwi appears in (10.3.5) with i > i0 . Then a ai and y ai > x bi ,
so that LT > (gi ) = y ai divides y a . Since gi I, LT > (gi ) is divisible by the leading
term of some h G . Hence LT > (h) divides y a , which is a term of g = x b y a G .
This is impossible in a reduced Grobner basis, hence i > i0 cannot occur in (10.3.5).
From here, it follows easily that g = gi0 , giving gi0 G as desired.
The statement for gi0 +1 follows by an argument parallel to the one above. The
details are left to the reader (Exercise 10.3.4).

We are now ready for the first major result of this section.
Theorem 10.3.5. Let be the fan in R2 with maximal cones
i = Cone(ai1 e1 + bi1 e2 , ai e1 + bi e2 ),
1 i r + 1,
for ai , bi as in Proposition 10.3.2. Then is the Grobner fan of the ideal I(Gd,k ).
Proof. The cones i fill out the first quadrant R20 . Take any w = ae1 + be2 lying
in the interior of some i and let > be a monomial order defined by a weight matrix
with w as first row. This gives a reduced Grobner basis G . The theorem will follow
once we prove that i is the Grobner cone CG of G .
First observe that for >, we must have must have i0 = i 1 in part (b) of
Lemma 10.3.4. It follows from part (c) of the lemma that gi1 and gi are elements
of G . Since a reduced Grobner basis has only one element with leading term a
power of x and only one with leading term a power of y, all other elements of G
will have the form x s yt 1, s,t > 0, by Lemma 10.3.3. Therefore, the Grobner
cone CG is exactly i and we are done.

As in Example 10.3.1, the following statement is an immediate consequence.
491
Corollary 10.3.6. Let N be the lattice

N = {(a/d, b/d) | a, b Z, b k mod d} = Z
k
1
d e1 + d e2
The toric surfaces X and X,N are isomorphic.
Ze2 .
In other words, the Grobner fan of I(Gd,k ) can be used to construct a resolution
of singularities of the affine toric surface U when has parameters d, k.
Connections with Representation Theory. We now consider the above results
from a different point of view. We assume the reader is familiar with the beginnings
of representation theory for finite abelian groups.
The group G = Gd,k from (10.3.2) acts on V = C2 by the 2-dimensional linear
representation of the group d of dth roots of unity defined by
(10.3.6)
: d GL(V ) = GL(2, C)

0
.
7
0 k
Since d is abelian, its irreducible representations are 1-dimensional over C, and

hence each is defined by a character
j : d C
7 j
for j = 0, . . . , d 1. The reason for the minus sign will soon become clear.
Via (10.3.2), we get the induced action of d on the polynomial ring C[x, y] by
x = 1 x,
y = j y,
as explained in 5.0. Each monomial x a y b spans an invariant subspace where

the action of d is given by the irreducible representation with character j for
j a + kb mod d. We call a + kb mod d the weight of the monomial x a y b with
respect to this action of d . Since the ideal of the group G (C )2 is invariant, the
action descends to the quotient C[x, y]/I(G), and we have a representation of d
on C[x, y]/I(G). The direct sum decomposition (10.3.4) shows that the irreducible
representation with character j appears exactly once in this representation, as
the subspace V j in (10.3.4). This means that the representation on C[x, y]/I(G) is
isomorphic to the regular representation of d (Exercise 10.3.5).
A 2-dimensional McKay Correspondence. In 1979, McKay pointed out that there
is a one-to-one correspondence between the irreducible representations of d and
the components of the exceptional divisor in the resolution : X U when
was a cone in normal form with parameters k = d 1, as in Example 10.1.5. In this
case, the singular point of U is a rational double point, and the image of the representation from (10.3.6) lies in SL(2, C). A great deal of research was devoted
to explaining the original McKay correspondence in representation-theoretic and
492
geometric terms (work of Gonzalez-Sprinberg, Artin, and Verdier). However, for

1 < k < d 1, (d ) is a subgroup of GL(2, C), not SL(2, C), and there are more
irreducible representations of d than components of the exceptional divisor. The
McKay correspondence can be extended to these cases by identifying certain special representations that correspond to the components of the exceptional divisor
(work of Wunram, Esnault, Ito and Nakamura, Kidoh, and others).
We will describe a generalized McKay correspondence that applies for all d, k.
Writing G = Gd,k as before, consider the ring of invariants C[x, y]G . You will prove
the following in Exercise 10.3.6.
Lemma 10.3.7. Let V j be an irreducible representation of d with character j ,
and consider the action of G = Gd,k d on C[x, y] C V j . Then the subspace of
invariants (C[x, y]C V j )G has the structure of a module over the ring C[x, y]G .
Example 10.3.8. Let G = G7,5 as in Example 10.3.1. The ring of invariants is
C[x, y]G = C[x7 , x 2 y, xy4 , y7 ] in this case. If v j is the basis of the representation V j ,
then it is easy to check that x a y b v j is invariant under G if and only if a + kb j
0 mod 7, or in other words if and only if x a y b has weight j under this action of 7 .
First consider the case j = 1 in Lemma 10.3.7. The monomials in the complement of the monomial ideal
M = hx7 , x 2 y, xy4 , y7 i
that have weight 1 are x and y3 . Then x v1 and y3 v1 generate the module
(C[x, y] V1 )G . Since x and y3 have the same weight with respect to this action
of 7 , the difference x y3 is an element of the ideal I(G), and this is one of the
polynomials gi as in the proof of Proposition 10.3.2.
On the other hand, if j = 2, then there are three monomials with weight 2 in
the complement of M, and these give three generators of (C[x, y] V2 )G , namely
x 2 v2 , xy3 v2 , and y6 v2 . It is still true that x 2 xy3 , x 2 y6 , and xy3 y6 are
elements of I(G), but these polynomials cannot appear in a reduced Grobner basis
for I(G). Moreover, no proper subset of the three generators generates the whole
module (C[x, y] V2 )G .
Definition 10.3.9. Let G = Gd,k d as above. We say that the representation V j

is special with respect to k if (C[x, y] V j )G is minimally generated as a module
over the invariant ring C[x, y]G by two elements.
Hence, in Example 10.3.8, V1 is special while V2 is not. According to our
definition, the trivial representation V0 is never special, since (C[x, y] V0 )G is
generated by the single monomial 1 over the invariant ring. Our next theorem
gives a rudimentary form of a McKay correspondence for the group Gd,k .
493
Theorem 10.3.10 (McKay Correspondence). Let be a cone with parameters

d, k, where 0 < k < d and gcd(d, k) = 1. Then there is a one-to-one correspondence between the representations of d that are special with respect to k and the
components of the exceptional divisor for the minimal resolution : X U .
Proof. Write G = Gd,k as above and consider the set B of monomials in the complement of the ideal M generated by the G-invariant monomials. This set contains
L = {1, x, x 2 , . . . , x d1 , y, y2 , . . . , y d1 }.
Since gcd(d, k) = 1, for each 1 j d 1, there is an integer 1 a j d 1 such

that x j and y a j have equal weight (equal to j) for the action of G. The representation V j is special with respect to k if and only if these are the only two monomials
of weight j in the set B, and nonspecial if and only if there is some monomial
x a y b with a, b > 0 in B which also has weight j. Since x j y a j I(G), saying
V j is special with respect to k is in turn equivalent to saying that the corresponding
vector a j e1 + je2 is an irreducible element in the semigroup from part (b) of Proposition 10.3.2 (Exercise 10.3.8). By Theorem 10.3.5, Cone(a j e1 + je2 ) is one of the
1-dimensional cones of the Grobner fan of I(G). Then V j corresponds to one of the
1-dimensional cones in the fan and hence to one of the irreducible components
of the exceptional divisor.

The original McKay correspondence is the following special case.
Corollary 10.3.11. When k = d 1, there is a one-to-one correspondence between the set of all irreducible representations of d and the components of the
exceptional divisor of the minimal resolution : X U .
Proof. In this case, the invariant ring is C[x d , xy, y d ], so the sets L and B in the
proof of the theorem coincide.

There has also been much work devoted to extend the McKay correspondence
to finite abelian subgroups G GL(n, C) for n 3, and several other ways to
understand these constructions have also been developed, including the theory of
G-Hilbert schemes. See Exercise 10.3.10 for the beginnings of this.
Exercises for 10.3.
10.3.1. In this exercise, you will verify the claims made in Example 10.3.1, and extend
some of the observations there.
(a) Show that each of the G (i) is a Grobner basis of I(G7,5 ).
(b) Show that G = {x 5 1, y x 3, x 2 y 1, x y 2, y 5 1} is a universal Grobner basis for
I(G7,5 ).
(c) Determine the cones CG (i) using (10.3.1).
(d) Verify the final claim that linear transformation defined by the matrix A from (10.3.3)
maps the Grobner cones CG (i) to the i for i = 1, 2, 3.
494
10.3.2. Verify the conclusions of Proposition 10.3.2 and Theorem 10.3.5 for the case d =
17, k = 11.
10.3.3. Show that the set S defined in part (b) of Proposition 10.3.2 is an additive semigroup. Hint: A direct proof starts from two general elements ae1 + be2 and a e1 + b e2 in
S. Consider (x b y a)(x b + y a ) and (x b + y a )(x b y a ).

10.3.4. In this exercise you will complete the proof of Lemma 10.3.4.
(a) Show that for each i, kPi dQi = ki , where the ki are produced by the modified Euclidean algorithm from (10.2.1).
(b) Prove part (a) of Lemma 10.3.4.
(c) Verify that there is an index i0 as in part (b) of Lemma 10.3.4.
(d) Verify that if i0 is as in part (c), then gi0 +1 is contained in the reduced Grobner basis.
10.3.5. If G is any finite group, the (left) regular representation of G is defined as follows.
Let W be a vector space over C of dimension |G| with a basis {eh | h G} indexed by the
elements of G. For each g G let (g) : W W be defined by (g)(eh ) = egh .
(a) Show that g 7 (g) is a group homomorphism from G to GL(W ).
(b) Now let G be the cyclic group d of order d. Show that W is the direct sum of 1dimensional invariant subspaces W j , j = 0, . . . , d 1 on which G acts by the character
j defined in the text, so that W decomposes as W d1
j=0 V j .
10.3.6. In this exercise you will consider the module structures from Lemma 10.3.7.
(a) Prove Lemma 10.3.7.
(b) Verify the claims made in Example 10.3.8.
10.3.7. In this exercise you will prove an alternate characterization of the special representations with respect to k from Definition 10.3.9. We write G = Gd,k d as usual.
(a) Let C2 2 = { f dx dy | f C[x, y]}. Show that x a y b dx dy = a+b+1+k x a y b dx dy
defines an action of G on C2 2 .
(b) Show that the spaces of G-invariants (C2 2 )G and (C2 2 V j )G have the structure of
modules over the invariant ring C[x, y]G .
(c) Show that V j is special with respect to k if and only if the multiplication map
is surjective.
(C2 2 )G (C[x, y] V j )G (C2 2 V j )G
10.3.8. In the proof of the McKay correspondence, show that a j e1 + je2 is irreducible in
the semigroup S from Proposition 10.3.2 if and only if the representation V j is special with
respect to k.
10.3.9. Let gi , 0 i r + 1, be the binomials constructed in Proposition 10.3.2. Show that
U = {g1 , . . . , gr } {x ay b 1 | x a y b is Gd,k -invariant}
is a universal Grobner basis for I(Gd,k ) (not always minimal, however).
10.3.10. Let G = Gd,k act on C2 as in (10.3.2). As a point set, G can be viewed as the orbit
of the point (1, 1) under this action. The ideal I = I(G) is invariant under the action of G
on C[x, y] and as we have seen, the corresponding representation on C[x, y]/I is isomorphic
to the regular representation of G.
10.4. Smooth Toric Surfaces
495
(a) Show that if p = (, ) is any point in C2 other than the origin, the ideal I of the orbit of
p is another G-invariant ideal and the corresponding representation of G on C[x, y]/I
is also isomorphic to the regular representation of G.
(b) The G-Hilbert scheme can be defined as the set of all G-invariant ideals in C[x, y] such
that the representation of G on C[x, y]/I is isomorphic to the regular representation of
G. Show that every such ideal has a set of generators of the form
{x a y c , y b x d , x ad y bc }
for some , C and where x a and y c (resp. y b and x d ) have equal weights for the
action of G. It can be seen from this result that the G-Hilbert scheme is also isomorphic
to the minimal resolution of singularities of U .

This section will use 10.1 and 10.2 to classify smooth complete toric surfaces
and study the relation between continued fractions and intersection products of
divisors on the resulting resolutions of singularities.
Classification of Smooth Toric Surfaces. We will show that smooth complete
toric surfaces are all obtained by toric blowups from either P 2 , P1 P1 , or one
of the Hirzebruch surfaces Hr with r 2 from Example 3.1.16. The proof will be
based on the following facts.
First, Proposition 3.3.15 implies that if = Cone(u1 , u2 ) is a smooth cone and
we refine by inserting the new 1-dimensional cone = Cone(u1 + u2 ), then on
the resulting toric surface, the smooth point p is blown up to a copy of P1 .
For the second ingredient of the proof, we introduce the following notation. If
is a smooth complete fan, then list the ray generators of the 2-dimensional cones
in as u0 , u1 , . . . , ur1 in clockwise order around the origin in NR , and we will
consider the indices as integers modulo r, so ur = u0 . Then we have the following
statement parallel to (10.2.8).
Lemma 10.4.1. Let u0 , . . . , ur be the ray generators for a smooth complete fan
in NR R2 . There exist integers bi , i = 0, . . . , r 1, such that
(10.4.1)
ui1 + ui+1 = bi ui .
Proof. This is a special case of the wall relation (6.4.4).
We also have the following result.

Lemma 10.4.2. Let be a smooth fan that refines a smooth cone . Then is
obtained from by a sequence of star subdivisions as in Definition 3.3.13.
Proof. Suppose has r cones of dimension 2, with ray generators u0 , . . . , ur , listed
clockwise starting from u0 . We argue by induction on r = |(2)|. If r = 1, there
is only one cone in and there is nothing to prove. Assume the result has been
496
proved for all with |(2)| = r, and consider a fan with |(2)| = r + 1. There
are r interior rays that by (6.4.4) give wall relations ui1 + ui+1 = bi ui , bi Z,
for 1 i r. Note that is strongly convex so bi > 0 for all i. We claim that there
exists some i such that bi = 1. If not, i.e., if bi 2 for all i, then as in 10.2, the
Hirzebruch-Jung continued fraction
[[b1 , b2 , . . . , br ]]
represents a rational number d/k and the cone has parameters d, k with d > k > 0.
But then d 2, which would contradict the assumption that is a smooth cone.
Hence there exists an i, 1 i r, such that
ui1 + ui+1 = ui .
In this situation, Cone(ui1 , ui+1 ) is also smooth (Exercise 10.4.1). Moreover,
Cone(ui1 , ui ) and Cone(ui , ui+1 ) are precisely the cones in the star subdivision
of Cone(ui1 , ui+1 ). Then we are done by induction.

We are now ready to state our classification theorem.
Theorem 10.4.3. Every smooth complete toric surface X is obtained from either
P 2 , P1 P1 , or Hr , r 2
by a finite sequence of blowups at fixed points of the torus action.
Proof. We follow the notation of Lemma 10.4.1. As in the proof of Lemma 10.4.2,
if bi = 1 in (10.4.1) for some i, then our surface is a blowup of the smooth surface
corresponding to the fan where ui is removed. Hence we only need to consider the
case in (10.4.1) where bi 6= 1 for all i.
Suppose first that u j = ui for some i < j. We relabel the vertices to make
u j = u1 for some j. Note that j > 2 since the cones must be strongly convex.
Then from (10.4.1),
u0 = u2 + b1 u1 .
Using the basis u1 , u2 of N, we get the picture shown in Figure 10 on the next page.
Comparing this with the fans for P 2 , P1 P1 and Ha (Figures 2, 3 and 4 from
3.1), we see that is a refinement of the fan of Hr if r = b1 > 2. The same
follows if b1 < 2 and r = |b1 | (see Exercise 10.4.2). Since b1 6= 1, the remaining
possibilities are b1 = 0 or 1, where we get a refinement of the fan of P1 P1 or
P 2 respectively. Then the theorem follows from Lemma 10.4.2 in this case.
We will complete the proof using primitive collections (Definition 5.1.5). The
first step is to note that X is projective by Proposition 6.3.25. This will allow
us to use Proposition 7.3.6, which asserts that has a primitive collection whose
minimal generators sum to 0.
If (1) has only three elements, it is easy to see that X = P 2 since is
smooth and complete (Exercise 10.4.3). If |(1)| > 3, every primitive collection
497
u0
u1
u2
uj
Figure 10. The ray generators u0 , u1 , u2 , u j when u j = u1
of has exactly two elements (Exercise 10.4.3). By Proposition 7.3.6, one of these
primitive collections must have minimal generators ui , u j that satisfy ui + u j = 0.
Hence u j = ui , and we are done by the earlier part of the proof.

Since there is also a Hirzebruch surface H1 , the statement of this theorem
might seem puzzling. The reason that H1 is not included is that this surface is
actually a blowup of P 2 (Exercise 10.4.4).
The problem of classifying smooth compete toric varieties of higher dimension
is much more difficult. We did this when rank Pic(X ) = 2 in Theorem 7.3.7. See
[14] for the case when rank Pic(X ) = 3.
Intersection Products on Smooth Surfaces. A fundamental feature of the theory
of smooth surfaces is the intersection product on divisors. In 6.3, we defined D C
when D is a Cartier divisor and C is a complete irreducible curve. On a smooth
complete surface, this means that the intersection product D C is defined for all
divisors D and C. In particular, taking D = C gives the self-intersection D D = D2 .
Here is a useful result about intersection numbers on a smooth toric surface.
Theorem 10.4.4. Let D be the divisor on a smooth toric surface X corresponding

to = Cone(u) which is the intersection of 2-dimensional cones Cone(u, u1 ) and
Cone(u, u2 ) in . Then:
(a) D D = b, where u1 + u2 = bu as in (10.4.1).
(b) For a divisor D 6= D , we have

(
1 = Cone(ui ), i = 1, 2
D D =
0 otherwise.
Proof. Since X is smooth, part (a) follows from Lemma 6.4.4 once you compare
(10.4.1) to (6.4.4). Part (b) follows from Corollary 6.4.3 and Lemma 6.4.4.
498
Example 10.4.5. Let = Cone(u1 , u2 ) be a cone in a smooth fan , and consider

the star subdivision, in which = Cone(u1 + u2 ) is inserted to subdivide into two
cones. Call the refined fan . Then the exceptional divisor E = D of the blowup
satisfies E E = 1 on X .
: X X
Complete curves with self-intersection number 1 on a smooth surface are

called exceptional curves of the first kind. They can always be contracted to a
smooth point on a birationally equivalent surface, as in the above example.
One of the foundational results in the theory of general algebraic surfaces is
that every smooth complete surface S has at least one relatively minimal model.
This means that there is a birational morphism S S, where S is a smooth surface
with the property that if : S S is a birational morphism to another smooth
surface S , then is necessarily an isomorphism. This is proved in [131, V.5.8].
Interestingly, the possible relatively minimal models for rational surfaces are precisely the surfaces P 2 , P1 P1 , and Hr , r 2, from Theorem 10.4.3.
On a smooth complete toric surface X , the intersection product can be regarded as a Z-valued symmetric bilinear form on Pic(X ). Here is an example.
Example 10.4.6. Consider the Hirzebruch surface Hr . Using the fan shown in
Figure 3 of Example 4.1.8, we get divisors D1 , . . . , D4 corresponding to minimal
generators u1 = e1 + re2 , u2 = e2 , u3 = e1 , u4 = e2 . By Theorem 10.4.4, we
have the self-intersections
D1 D1 = D3 D3 = 0, D2 D2 = r, D4 D4 = r.
The Picard group Pic(Hr ) is generated by the classes of D3 and D4 . Note also that
D3 D4 = D4 D3 = 1
by Theorem 10.4.4. The intersection product is described by the matrix

D3 D3 D3 D4
0 1
.
=
1 r
D4 D3 D4 D4
If D aD3 + bD4 and E cD3 + dD4 are any two divisors on the surface, then

0 1
c
= bc + ad + rbd.
(10.4.2)
DE = a b
d
1 r
For instance, with D = E = D2 rD3 + D4 , we obtain
D2 D2 = 1 (r) + (r) 1 + r 1 1 = r.
The self-intersection numbers D1 D1 = D3 D3 = 0 reflect the fibration structure

on Hr studied in Example 3.3.20. The divisors D1 and D3 are fibers of the mapping Hr P1 . Such curves always have self-intersection equal to zero. You will
compute several other intersection products on Hr in Exercise 10.4.5.
499
Resolution of Singularities Reconsidered. Another interesting class of smooth

toric surfaces consists of those that arise from a resolution of singularities of the
affine toric surface U of a 2-dimensional cone . Here is a simple example.
Example 10.4.7. Let = Cone(e2 , de1 e2 ) have parameters d, 1 where d > 1.
The resolution of singularities X U constructed in Example 10.1.8 uses the
smooth refinement of obtained by adding Cone(e1 ). This gives the exceptional
divisor E on X . Since
e2 + (de1 e2 ) = de1 ,
we see that
E E = d
is the self-intersection number of E.
More generally, suppose that the smooth toric surface X is obtained via a
resolution of singularities of U , where the 2-dimensional cone has parameters
d, k with d > 1. Let the Hirzebruch-Jung continued fraction expansion of d/k be
d/k = [[b1 , b2 , . . . , br ]].
Recall from Theorem 10.2.3 that is obtained from = Cone(u0 , ur+1 ) by adding
rays generated by u1 , . . . , ur , and by Theorem 10.2.5, we have
1 i r.
ui1 + ui+1 = bi ui ,
It follows that D1 , . . . , Dr are complete curves in X . They are the irreducible

components of the exceptional fiber, with self-intersections
Di Di = bi ,
1 i r,
by Theorem 10.4.4. Then the intersection matrix (Di D j )1i, jr is given by
bi if j = i
(10.4.3)
Di D j = 1
if |i j| = 1
0
otherwise.
In Exercise 10.4.6 you will show that the associated quadratic form is negative
definite. This condition is necessary for the contractibility of a complete curve C
on a smooth surface S, i.e., the existence of a proper birational morphism : S S,
where (C) is a (possibly singular) point on S.
The resolutions described here have another important property.
Definition 10.4.8. A resolution of singularities : Y X is minimal if for every
resolution of singularities : Z X , there is a morphism : Z Y such that
Y
?
/X
500
is a commutative diagram, i.e., = .

It is easy to see that if a minimal resolution of X exists, then it is unique up to
isomorphism. If X has a unique singular point p, then using the theory of birational
morphisms of surfaces it is not difficult to show that a resolution of singularities
: Y X is minimal if the exceptional fiber contains no irreducible components E
with E E = 1 (see Exercise 10.4.7). By Theorem 10.4.4 and the fact that bi 2
in Hirzebruch-Jung continued fractions, this holds for the resolutions constructed
in Theorem 10.2.3. Hence we have the following.
Corollary 10.4.9. The resolution of singularities of the affine toric surface U
constructed in Theorem 10.2.3 is minimal.

Rational Double Points Reconsidered. If has parameters d, d 1, then from
Exercise 10.2.2, the Hirzebruch-Jung continued fraction expansion of d/(d 1) is
given by
d/(d 1) = [[2, 2, . . . , 2]],
with d 1 terms. Hence bi = 2 for all i, and (10.4.3) gives the (d 1) (d 1)

matrix
2
1
0
0
1 2 1
0
.. . . . .
..
..
.
.
.
.
.
0
1 2
1
0
0
1 2
representing the intersection product on the subgroup of Pic(X ) generated by the
components of the exceptional divisor for the resolution of a rational double point
of type Ad1 . We can now fully explain the terminology for these singularities.
The problem of classifying lattices
Ze1 Zes
with negative definite bilinear forms B satisfying B(ei , ei ) = 2 for all i arises
in many areas within mathematics, most notably in the classification of complex
simple Lie algebras via root systems. The matrix above is the (negative of) the
Cartan matrix for the root system of type Ad1 , which is often represented by the
Dynkin diagram:
t
qqq
with d 1 vertices. The vertices represent the lattice basis vectors. The edges
connect the pairs with B(ei , e j ) 6= 0 and B(ei , ei ) = 2 for all i as above. In our
case, the vertices represent the components Di of the exceptional divisor, and the
bilinear form is the intersection product.
A precise definition of a surface rational double point follows.
501
Definition 10.4.10. A singular point p of a normal surface X is a rational double

point or Du Val singularity if X has a minimal resolution of singularities : Y X
such that if KY is a canonical divisor on Y , then every irreducible component Ei of
the exceptional divisor E over p satisfies
KY Ei = 0.
We can relate these concepts to the toric case as follows.
Proposition 10.4.11. Assume has parameters d > k > 0 and let : X U
be the resolution of singularities constructed in Theorem 10.2.3. Then the singular
point of U is a rational double point if and only if k = d 1.
P
Proof. The canonical divisor of X is KX = r+1
i=0 Di , and one computes that
KX Di = bi 2,
1 i r.
Thus the singular point is a rational double point if and only if bi = 2 for all i. This
easily implies k = d 1. You will verify these claims in Exercise 10.4.8.

There is much more to say about rational double points. For example, one can
show that E = E1 + + Er satisfies E E = 2 (you will prove this in the toric case
in Exercise 10.4.8). From a more sophisticated point of view, E E = 2 implies
that the canonical sheaf on Y is the pullback of the canonical sheaf on X under .
We will explore this in Proposition 11.2.8. See [85] for more on rational double
points.
Exercises for 10.4.
10.4.1. Here you will verify several statements made in the proof of Lemma 10.4.2.
(a) Show that if the cone is strictly convex, then the integers bi in (10.4.1) must be
strictly positive.
(b) Show that if ui1 + ui+1 = ui , then Cone(ui1 , ui+1 ) must also be smooth.
10.4.2. In the proof of Theorem 10.4.3, verify that if u j = u1 and u0 = u2 + b1 u1 with
b1 < 2, then is a refinement of a fan with X Hr , where r = |b1 |.
10.4.3. In this exercise, you will prove some facts used in the proof of Theorem 10.4.3.
Let be a smooth complete fan in NR R2 .
(a) If |(1)| = 3, prove that X P 2 .
(b) If |(1)| > 3, prove that every primitive collection of (1) has two elements.
10.4.4. In the statement of Theorem 10.4.3, you might have noticed the absence of the
Hirzebruch surface H1 . Show that this surface is isomorphic to the blowup of P 2 at one of
its torus-fixed points. See Exercise 3.3.8 for more details.
10.4.5. This exercise studies several further examples of the intersection product on Hr .
(a) Compute D1 D1 using (10.4.2) and also directly from Theorem 10.4.4.
(b) Compute K 2 = K K on Hr , where K = KHr is the canonical divisor.
502
10.4.6. Show that the matrix defined by (10.4.3) has a negative-definite associated quadratic form. Hint: Recall that if B(x, y) is a bilinear form, the associated quadratic form is
Q(x) = B(x, x).
10.4.7. Let X have a unique singular point p and let : Y X be a resolution of singularities such that no component E of the exceptional fiber 1 (p) has E E = 1. In this
exercise, you will show that Y is a minimal resolution of X according to Definition 10.4.8.
Let : Z X be another resolution of singularities and consider the possibly singular
surface S = Z X Y . Let R be a resolution of S. Then we have a commutative diagram of
morphisms
/
Y
R

Z

/ X.
(a) Explain why it suffices to show that must be an isomorphism.

(b) If not, apply [131, V.5.3] to show that factors as a sequence of blowups of points.
Hence R must contain curves L with L L = 1 in the exceptional fiber over p.
(c) Let L be an irreducible curve on R with L L = 1 and show that E = (L) satisfies
E E = 1.
(d) Deduce that is an isomorphism, hence : Y X is a minimal resolution.

10.4.8. This exercise deals with the proof of Proposition 10.4.11.
(a) Show that KX Di = bi 2 for 1 i r.
(b) Show that d/k = [[2, . . . , 2]] if and only if k = d 1. Hint: Exercise 10.2.2.
(c) Show that E = D1 + + Dr satisfies E E = 2.
10.4.9. Let have parmeters d, d 1, so that the singular point of U is a rational double
point. By Proposition 10.1.6, U is Gorenstein, so that its canonical sheaf U is a line
bundle. Let : X U be the resolution constructed in Theorem 10.2.3. Prove that
U is the canonical sheaf of X .
10.4.10. Another interesting numerical fact about the integers bi from (10.4.1) is the following. Suppose a smooth fan has 1-dimensional cones labeled as in Lemma 10.4.1.
Then
(10.4.4)
b0 + b1 + + br1 = 3r 12.
This exercise will sketch a proof of (10.4.4).

(a) Show that (10.4.4) holds for the standard fans of P 2 , P1 P1 and Hr , r 2.
(b) Show that if (10.4.4) holds for a smooth fan , then it holds for the fan obtained by
performing a star subdivision on one of the 2-dimensional cones of .
(c) Deduce that (10.4.4) holds for all smooth fans using Theorem 10.4.3.
10.5. Riemann-Roch and Lattice Polygons

Riemann-Roch theorems are a class of results about the dimensions of sheaf cohomology groups. The original statement along these lines was the theorem of
503
Riemann and Roch concerning sections of line bundles on algebraic curves. This
result and its generalizations to higher-dimensional varieties can be formulated
most conveniently in terms of the Euler characteristic of a sheaf, defined in 9.4.
Riemann-Roch for Curves. A modern form of the Riemann-Roch theorem for
curves states that if D is a divisor on a smooth projective curve C, then
(10.5.1)
(OC (D)) = deg(D) + (OC ),
where the degree deg(D) is defined in Definition 6.3.2. This equality can be rewritten using Serre duality as follows. Namely, if KC is a canonical divisor on C, then
we have
H 1 (C, OC (D)) H 0 (C, OC (KC D))
H 1 (C, OC ) H 0 (C, OC (KC )) .
The integer g = dim H 0 (C, OC (KC )) is the genus of the curve C. Then (10.5.1) can
be rewritten in the form commonly used in the theory of curves:
(10.5.2)
dim H 0 (C, OC (D)) dim H 0 (C, OC (KC D)) = deg(D) + 1 g.
A proof of this theorem and a number of its applications are given in [131, Ch. IV].
Also see Exercise 10.5.1 below. As a first consequence, note that if D = KC is a
canonical divisor, then
(10.5.3)
deg(KC ) = 2g 2.
We will need to use (10.5.2) most often in the simple case X P1 . Then g = 0
and the Riemann-Roch theorem for P1 is the statement for all divisors D on P1 ,
(10.5.4)
(OP1 (D)) = deg(D) + 1.
The Adjunction Formula. For a smooth curve C contained in a smooth surface X ,

the canonical sheaves C of the curve and X of the surface are related by
(10.5.5)
C X (C) OX OC .
This follows without difficulty from Example 8.2.2 (Exercise 10.5.2) and has the
following consequence for the intersection product on X .
Theorem 10.5.1 (Adjunction Formula). Let C be a smooth curve contained in a
smooth complete surface X . Then
KX C +C C = 2g 2,
where g is the genus of the curve C.
Proof. Let i : C X be the inclusion map. Then
C X (C) OX OC = i X (C) = i OX (KX +C),
so that
2g 2 = deg(C ) = deg(i OX (KX +C)) = (KX +C) C,
504
where the first equality is (10.5.3) and the last is the definition of (KX +C) C given
in 6.3.

Riemann-Roch for Surfaces. The statement for surfaces corresponding to (10.5.1)
is given next.
Theorem 10.5.2 (Riemann-Roch for Surfaces). Let D be a divisor on a smooth
projective surface X with canonical divisor KX . Then
D D D KX
(OX (D)) =
+ (OX ).
2
We will only prove this for X a smooth complete toric surface; there is a simple
and concrete proof in this case.
Proof. The theorem certainly holds for D = 0 since OX (D) = OX in this case. Our
proof will use the special properties of smooth complete toric surfaces. Recall that
if X = X , then Pic(X ) is generated by the classes of the divisors Di , i = 1, . . . , r,
corresponding to the 1-dimensional cones in . Hence, to prove the theorem, it
suffices to show that if the theorem holds for a divisor D, then it also holds for
D + Di and D Di for all i.
Assume the theorem holds for D. By Proposition 4.0.28, the sequence
0 OX (Di ) OX ODi 0
is exact. Tensoring this with OX (D + Di) gives the exact sequence
0 OX (D) OX (D + Di ) ODi (D + Di) 0.
By (9.4.1), it follows that
(OX (D + Di )) = (OX (D)) + (ODi (D + Di)).
By the induction hypothesis,
D D D KX
+ (OX ).
2
For (ODi (D + Di )), recall that Di P1 . Hence, by the Riemann-Roch theorem
for P1 given in (10.5.4), we have
(10.5.6)
(OX (D)) =
(10.5.7)
(ODi (D + Di)) = D Di + Di Di + 1.
Combining (10.5.6) and (10.5.7), we obtain

D D D KX
+ D Di + Di Di + 1 + (OX )
(OX (D + Di)) =
2
(D + Di) (D + Di) + Di Di D KX + 2
=
+ (OX )
2
However, using KX = (D1 + + Dr ) and Theorem 10.4.4, one computes that
(10.5.8)
Di KX = Di Di 2.
505
Substituting this into the above expression for (OX (D + Di )) and simplifying, we
obtain
(D + Di) (D + Di ) (D + Di) KX
(OX (D + Di)) =
+ (OX ),
2
which shows that the theorem holds for D + Di
The proof for D Di is similar and is left to the reader (Exercise 10.5.3).
The following statement is sometimes considered as the topological part of the

Riemann-Roch theorem for surfaces.
Theorem 10.5.3 (Noethers Theorem). Let X be a smooth projective surface with
canonical divisor KX . Then
KX KX + e(X )
,
12
where e(X ) is the topological Euler characteristic of X defined by
(OX ) =
e(X ) =
4
X
(1)k dim H k (X , C).
k=0
As before, we will give a proof only for a smooth complete toric surface. We
will also use the Hodge decomposition
M
H k (X , C)
H q (X , Xp )
p+q=k
from (9.4.11).
Proof. Demazure vanishing (Theorem 9.2.3) implies that for a smooth complete
toric surface X = X ,
(OX ) = dim H 0 (X , OX ) dim H 1 (X , OX ) + dim H 2 (X , OX )
(10.5.9)
= 1 0 + 0 = 1.
Thus Noethers theorem for a smooth complete toric surface is equivalent to

KX KX + e(X ) = 12.
(10.5.10)
We prove this as follows. Set r = |(1)| and let the minimal generators of the rays
P
be u0 , . . . , ur1 as in Lemma 10.4.1. Since KX = r1
i=0 Di , (10.5.8) implies
KX KX =
r1
X
i=0
Di KX =
r1
r1
X
X
Di Di .
(Di Di 2) = 2r +
i=0
i=0
If ui1 + uu+i = bi ui as in (10.4.1), then Di Di = bi by Theorem 10.4.4. Hence

KX KX = 2r
r1
X
i=0
bi = 2r (3r 12) = 12 r,
506
where the equality
Pr1
i=0
bi = 3r 12 is from Exercise 10.4.10.
We next compute e(X ). Proposition 11.2.8 shows that is the normal fan of
a polygon with r = |(1)| sides. Then the formula for dim H q (X , Xp ) given in
Theorem 9.4.11 implies
dim H 0 (X , C) = dim H 0 (X , OX ) = 1
dim H 1 (X , C) = dim H 0 (X , 1X ) + dim H 1 (X , OX ) = 0 + 0 = 0
dim H 2 (X , C) = dim H 0 (X , X2 ) + dim H 1 (X , 1X ) + dim H 2 (X , OX )
= 0 + f1 2 f2 = r 2
dim H 3 (X , C) = dim H 1 (X , X2 ) + dim H 2 (X , 1X ) = 0 + 0 = 0

dim H 4 (X , C) = dim H 2 (X , X2 ) = 1.
where f1 = r and f2 = 1 are the face numbers of a polygon with r sides. It follows
that e(X ) = 1 + (r 2) + 1 = r. Then (10.5.10) follows easily from the above
computation of KX KX .

We will give a topological proof of e(X ) = r in Chapter 12, and in Chapter 13,
we will interpret e(X ) in terms of the Chern classes of the tangent bundle.
The Riemann-Roch theorems for curves and surfaces have been vastly generalized by results of Hirzebruch and Grothendieck, and the precise relation of
Noethers theorem to the Riemann-Roch theorem for surfaces is a special case of
their approach. We will discuss Riemann-Roch theorems for higher-dimensional
toric varieties in Chapter 13.
Lattice Polygons. For the remainder of this section, we will explore the relation
between toric surfaces and the geometry and combinatorics of lattice polygons. We
will see that the results from 9.4 for lattice polytopes have an especially nice form
for lattice polygons.
Let X = X be a smooth complete toric surface. Since (OX ) = 1 by (10.5.9),
Riemann-Roch for a divisor D on X becomes
D D D KX
(10.5.11)
(OX (D)) =
+ 1,
2
Thus, for any Z,
(10.5.12)
(OX (D)) =
D D D KX
+ 1 = 21 (D D) 2 12 (D KX ) + 1.
2
The theory developed in 9.4 guarantees that (OX (D)) is a polynomial in ; the
above computation gives explicit formulas for the coefficients in terms of intersection products.
Here is an example of how this formula works.
507
Example 10.5.4. Let X be the Hirzebruch surface H2 . We will use the notation of
Example 10.4.6. The divisor D = D1 + D2 is clearly effective, but the inequalities
(9.3.6) defining the nef cone show that D is not nef. Using KX = D1 D4
and Example 10.4.6, one computes that
D D = 0,
D KX = 2.
Then (10.5.12) implies that

(OX (D)) = + 1.
An easy application of Serre duality gives H 2 (X , OX (D)) = 0 (Exercise 10.5.4).
Thus
dim H 0 (X , OX (D)) dim H 1 (X , OX (D)) = + 1.
Things get more surprising when we compute dim H 0 (X , OX (D)). Using the ray
generators u1 , . . . , u4 from Example 10.4.6, the polygon PD corresponding to D =
D1 + D2 is defined by the inequalities
hm, u1 i 1, hm, u2 i 1, hm, u3 i 0, hm, u4 i 0.
The polygon PD is shown in Figure 11.
u1
u2
u3
PD
u4
Figure 11. The polygon of the divisor D and the fan of H2
Even though PD is not a lattice polytope, Proposition 4.3.3 still applies. Thus
(
1 2
+ + 1 even
0
dim H (X , OX (D)) = |PD M| = |PD M| = 14 2
3
odd,
4 + + 4
where the final equality follows from Exercise 9.4.13. Combining this with the
above computation of (OX (D)), we obtain
(
1 2
even
dim H 1 (X , OX (D)) = 14 2 1
odd.
4 4
This is a vivid example how the Euler characteristic smooths out the complicated
behavior of the individual cohomology groups.
508
On the other hand, if D is nef, the higher cohomology is trivial by Demazure

vanishing, so that the Euler characteristic reduces to dim H 0 . We exploit this as
follows. Suppose that P MR R2 is a lattice polygon. By Theorem 9.4.2 and
Example 9.4.4, the Ehrhart polynomial EhrP (x) Q[x] of P satisfies
(10.5.13)
EhrP () = |(P) M| = Area(P) 2 + 12 |P M| + 1
for N. We next describe this polynomial in terms of intersection products.
By the results of 2.3, we get the projective toric surface XP coming from the
normal fan P of P. In general XP will not be smooth, so we compute a minimal
resolution of singularities
: X XP
using the methods of this chapter. Recall that XP has the ample divisor DP whose
associated polygon is P.
Proposition 10.5.5. There is unique torus-invariant nef divisor D on X such that
(a) The support function of D equals the support function of DP .
(b) (OX (D)) is the Ehrhart polynomial of P.
Proof. Proposition 6.2.7 implies that X has a divisor D that satisfies part (a). As
in 6.1, we call D the pullback of DP . Since DP has a convex support function, the
same is true for D, so that D is nef. Furthermore, P is the polytope associated to
DP and hence is the polytope associated to D since the polytope of a nef divisor is
determined by its support function (Theorem 6.1.7).
It follows that dim H 0 (X , OX (D)) = |PD M| = |(P) M| when 0,
so that H 0 (X , OX (D)) equals the Ehrhart polynomial of P when 0. However, D is nef when 0 and hence has trivial higher cohomology by Demazure
vanishing (Theorem 9.2.3). Thus 0 implies
(OX (D)) = dim H 0 (X , OX (D)) = |(P) M|.
Since (OX (D)) is a polynomial in , it must be the Ehrhart polynomial of P.

Theorem 10.5.5 and (10.5.12) imply that the Ehrhart polynomial of P is
EhrP () = 12 (D D) 2 21 (D KX ) + 1.
Comparing this to the formula (10.5.13) for EhrP (), we get the following result.
Proposition 10.5.6. Let P be a lattice polygon and let D be the pullback of DP
constructed in Proposition 10.5.5. Then
D D = 2 Area(P)
D KX = |P M|.
509
Example 10.5.7. Take the fan of H2 shown in Figure 11 from Example 10.5.4
and combine the two 2-dimensional cones containing u2 into a single cone. The
resulting fan has minimal generators u1 , u3 , u4 that satisfy u1 + u3 + 2u4 = 0, so the
resulting toric variety is P(1, 1, 2).
Let P = Conv(0, 2e1 , e2 ) MR , which is the double of the polytope shown
in Figure 11. The normal fan of P is the fan of P(1, 1, 2). The minimal generators
u1 , u3 , u4 of this fan give divisors D1 , D3 , D4 on P(1, 1, 2), and the divisor DP is
easily seen to be the ample divisor 2D1 .
Since the fan of H2 refines the fan of P(1, 1, 2), the resulting toric morphism
H2 P(1, 1, 2) is a resolution of singularities. By considering the support function
of DP , we find that the pullback of DP = 2D1 is D = 2D1 + 2D2 . We leave it as
Exercise 10.5.5 to compute D D and D KH2 and verify that they give the numbers
predicted by Proposition 10.5.6.
Sectional Genus. The divisor DP on XP is very ample since dim P = 2. Hence it

gives a projective embedding XP P s such that OPs (1) restricts to OX (DP ). In
geometric terms, this means that hyperplanes H P s give curves XP H XP that
are linearly equivalent to DP . For some hyperplanes, the intersection XP H can be
complicated. Since XP has only finitely many singular points, the Bertini theorem
(see [131, II 8.18 and III 7.9.1]) guarantees that when H is generic, C = XP H is
a smooth connected curve contained in the smooth locus of XP . The genus g of C
is called the sectional genus of the surface XP .
We will compute g in terms of the geometry of P using the adjunction formula.
Since we need a smooth surface for this, we use a resolution : X XP and note
that C can be regarded as a curve in X since is an isomorphism away from the
singular points of XP . Since C DP on XP , we have C D on X , where D is the
pullback of DP . Then the adjunction formula (Theorem 10.5.1) implies
2g 2 = KX C +C C = KX D + D D,
so that
(10.5.14)
g = 21 D (KX + D) + 1.
Then we have the following result.

Proposition 10.5.8. The sectional genus of XP is g = |Int(P) M|.
Proof. Picks formula from Example 9.4.4 can be written as
|Int(P) M| = Area(P) 12 |P M| + 1,
which by Proposition 10.5.6 becomes
|Int(P) M| = 21 D D + 12 D KX + 1.
The right-hand side is g by (10.5.14), completing the proof.
510
Example 10.5.9. Let P = d2 = Conv(0, de1 , de2 ). Then XP is the projective

plane P 2 in its dth Veronese embedding, and DP dL, where L P 2 is a line. The
hyperplane sections are the curves of degree d in P 2 , and the smooth ones have
genus
(d 1)(d 2)
.
g = |Int(d2 ) M| =
2
You will check this assertion and another example in Exercise 10.5.6.
The curves C X studied here can be generalized to the study of hypersurfaces in projective toric varieties coming from sections of a nef line bundle. The
geometry and topology of these hypersurfaces have been studied in many papers,
including [15], [19], [77] and [197].
Reflexive Polygons and The Number 12. Our final topic gives a way to understand
a somewhat mysterious formula we noted in the last section of Chapter 8. Recall
from Theorem 8.3.7 that there are exactly 16 equivalence classes of reflexive lattice
polytopes in R2 , shown in Figure 3 of 8.3. The article [230] gives four different
proofs of the following result.
Theorem 10.5.10. Let P be a reflexive lattice polygon in MR R2 . Then
|P M| + |P N| = 12.
One proof consists of a case-by-case verification of the statement for each of

16 equivalence classes. You proved the theorem this way in Exercise 8.3.5. The
argument was straightforward but not very enlightening! Here we will give another
proof using Noethers theorem.
Proof. Since Noethers theorem requires a smooth surface, we need to refine the
normal fan P of P MR . Since P is reflexive, we can do this using the dual
polygon P NR . We know from (8.3.2) that the vertices of P are the minimal
generators of P . Let be the refinement of P whose 1-dimensional cones are
generated by the rays through the lattice points on the boundary of P . This is
illustrated in Figure 12.
s
@
s @s

s

s
@
A
s A@s
A @
s As @s
HH@
H
s
s
s @
Hs
Figure 12. A reflexive polygon P and its dual P
511
The fan has the following properties:

For each cone of , its minimal generators and the origin form a triangle whose
only lattice points are the vertices. Thus is smooth by Exercise 8.3.4.
The minimal generators of are the lattice points of P lying on the boundary.
Thus |(1)| = |P N|.
From the first bullet, we get a resolution : X XP . Recall that DP = KXP since
P is reflexive. The wonderful fact is that its pullback via is again anticanonical,
i.e., D = KX . To prove this, recall that D and DP have the same support function
, which takes the value 1 at the vertices of P since DP = KXP . It follows that
= 1 on the boundary of P . Then D = KX because the minimal generators of
all lie on the boundary.
Now apply Noethers theorem to the toric surface X . By (10.5.10), we have
KX KX + e(X ) = 12.
We analyze each term on the left as follows. First, D = KX implies

KX KX = D KX = |P M|,
where the last equality follows from Proposition 10.5.6. Second, e(X ) is the number of minimal generators of by the proof of Theorem 10.5.3. In other words,
e(X ) = |(1)| = |P N|,
where the second equality follows from the above analysis of . Hence the theorem
is an immediate consequence of Noethers theorem.

A key step in the above proof was showing that the pullback of the canonical
divisor on XP was the canonical divisor on XP . This may fail for a general resolution
of singularities. We will say more about this when we study crepant resolutions in
Chapter 11.
Exercises for 10.5.
10.5.1. The Riemann-Roch theorem for curves, in the form (10.5.1), can be proved by
much the same method as used in the proof of Theorem 10.5.2. Namely, show that if
(10.5.1) holds for a divisor D then it also holds for the divisors D + P and D P, where P
is an arbitrary point on the curve.
10.5.2. Prove the adjunction formula (Theorem 10.5.1) using (10.5.3) and (10.5.5).
10.5.3. Complete the proof of Theorem 10.5.2 by showing that if the theorem holds for
D, then it also holds for D Di where Di is any one of the divisors corresponding to the
1-dimensional cones in .
P
10.5.4. Let D = a D be an effective Q-Cartier Weil divisor on a complete toric variety
X of dimension n. Use Serre duality (Theorem 9.2.10) to prove that H n (X , OX (D)) = 0.
10.5.5. In Example 10.5.7, compute D D and D KH2 and check that they agree with the
numbers given by Proposition 10.5.6.
512
10.5.6. This exercise studies the sectional genus of toric surfaces.

(a) Verify the formula given in Example 10.5.9 for the sectional genus of P 2 in its dth
Veronese embedding.
(b) Let P = Conv(0, ae1 , be2 , ae1 + be2 ). What is the smooth toric surface XP in this case?
Show that its sectional genus is (a 1)(b 1).
10.5.7. Let P be a reflexive polygon.
(a) Prove that the singularities (if any) of the toric surface XP are rational double points.
Hint: Proposition 10.1.6.
(b) Prove that XP has sectional genus g = 1. This means that smooth anticanonical curves
in XP are all elliptic curves.
(c) Explain how part (b) relates to Exercise 10.5.6.
10.5.8. According to Theorem 10.4.3, every smooth toric surface is a blowup of either P 2 ,
P1 P1 , or Hr for r 2. For each of the 16 reflexive polygons in Figure 3 of 8.3, the
process described in the proof of Theorem 10.5.10 produces a smooth toric surface X .
Where does X fit in this classification in each case? (This gives a classification of smooth
toric Del Pezzo surfaces.)
10.5.9. As in 9.3, the p-Ehrhart polynomials of a lattice polygon P MR are defined by
b p (DP )), p = 0, 1, 2.
Ehr p () = (
P
XP
EhrP0
We know that
is the usual Ehrhart polynomial EhrP , and then EhrP2 (x) = EhrP (x)
by Theorem 9.4.7. The remaining case is EhrP1 . Prove that
EhrP1 (x) = 2Area(P) x2 + f1 2,
where f1 is the number of edges of P. Hint: Use Theorem 9.4.11 for the constant term and
part (c) of Theorem 9.4.7 for the coefficient of x. For the leading coefficient, tensor the
exact sequence of Theorem 8.1.6 with OXP (DP ), take the Euler characteristic, and then let
.
Chapter 11
Toric Resolutions and

Toric Singularities
This chapter will study the singularities of a normal toric variety. We begin in
11.1 with the existence of toric resolutions of singularities. In 11.2 we consider
more special resolutions, including simple normal crossing resolutions, crepant
resolutions, log resolutions, and embedded resolutions. There are also relations
with ideal sheaves, Rees algebras, and multiplier ideals, to be studied in 11.3, and
finally, in 11.4 we consider some important classes of toric singularities.
11.1. Resolution of Singularities

The singular locus of an irreducible variety X , denoted Xsing , is the set of all singular points of X . One can prove that Xsing is a proper closed subvariety of X (see
[131, Thm. I.5.3]). We call the complement X \ Xsing the smooth locus of X .
Definition 11.1.1. Given an irreducible variety X , a resolution of singularities of
X is a morphism f : X X such that:
(a) X is smooth and irreducible.
(b) f is proper.
(c) f induces an isomorphism of varieties f 1 (X \ Xsing ) X \ Xsing .
Furthermore, f : X X is a projective resolution if f is a projective morphism.

A resolution of singularities modifies X to make it smooth without changing
the part that is already smooth. In particular, a resolution of singularities is a birational morphism. Hironaka [143] proved the existence of a resolution of singularities over an algebraically closed field of characteristic 0. An introduction to
Hironakas proof and more recent developments can be found in [138].
513
514
Chapter 11. Toric Resolutions and Toric Singularities
Fortunately, the toric case is much simpler, partly because toric varieties cannot
have arbitrarily bad singularities. Hence we will be able to give a complete proof
of resolution of singularities for toric varieties. The key idea is that given a fan
in NR , the resolution will be the toric morphism
: X X
coming from a suitable refinement of .
The Singular Locus. The first step is to determine the singular locus of X . This
has the following purely toric description.
Proposition 11.1.2. Let X be the toric variety of the fan . Then:
[
(X )sing =
V ()
not smooth
X \ (X )sing =
U ,
smooth
where V () = O() is the closure of the TN -orbit corresponding to .

Proof. Recall that is smooth if its minimal generators can be extended to a basis
of N and that is smooth if and only if U is smooth. Also observe that
if is smooth, then so is every face of , hence
if is not smooth, then so is every cone of containing .
The first bullet tells us that the smooth cones of form a fan whose toric variety is
S
S smooth U . This open set of X is clearly smooth, and its complement in X is
not smooth O() by the Orbit-Cone correspondence.
S
S
We also have not smooth V () = not smooth O() by the second bullet and
the Orbit-Cone correspondence. Hence the proof will be complete once we show
that every point of O() is singular in X when is not smooth. It suffices to work
in the affine open set U . Let N = Span() N and pick a complement N2 N
such that N = N N2 . By (1.3.2), we have
U,N U,N TN2 .
Since dim = rank N , the orbit ON () U,N consists of the unique fixed point
of the action of TN . Since is not smooth with respect to N , the proof of Theorem 1.3.12 shows that this fixed point is singular in U,N . Then (1.3.3) shows that
every point of ON () TN2 is singular in U,N TN2 . Since the above product
decomposition induces
ON () ON () TN2
(Exercise 11.1.1), we see that all points of O() = ON () are singular in X .
It follows that when constructing refinements of a fan , we need to be sure to
leave the smooth cones of unchanged.
515
Star Subdivisions. We will construct refinements using a generalization of the star

subdivision () defined in 3.3. Given a fan in NR and a primitive element
v || N, let (v) be the set of the following cones:
(a) , where v
/ .
(b) Cone(, v), where v

/ and {v} .
We call (v) the star subdivision of at v. This terminology is justified by the

following result.
Lemma 11.1.3. (v) is a refinement of .
Proof. Let Cone(, v) be a cone as in (b) above. Then observe that v
/ Span( )
since v
/ and = Span( ). It follows that dim Cone(, v) = dim + 1.
To show that (v) is a fan, first consider a face of a cone in (v). If the
cone arises from (a) above, then the face clearly lies in (v). If the cone arises
from (b), then we have a face of Cone(, v). Write = Hm Cone(, v) for
m Cone(, v) and observe that m and hm, vi 0. There are two cases:
hm, vi = 0, in which case = Cone(Hm , v) (v).
hm, vi > 0, in which case = Hm (v).
You will provide the details of these cases in Exercise 11.1.2.

Next consider two cones in (v). If neither contains v, then the intersection
is clearly in (v). Now suppose that one does not contain v and the other does.
This gives cones v
/ and Cone(, v) as in (a) and (b) above. We claim that
(11.1.1)
Cone(, v) = ,
which will imply that Cone(, v) (v). One inclusion of (11.1.1) is obvious;
for the other inclusion, let u Cone(, v) and write u = u0 + v for u0 and
0. Also let be a cone containing {v} . Then
u = u0 + v Cone(, v) .
Since u0 , v and is a face of , it follows that u0 , v . But v

/ ,
so that = 0 and hence u = u0 . Thus u , as desired.
The remaining case to consider is when both cones contain v. If the cones are
Cone(1 , v) and Cone(2 , v), then we claim that
(11.1.2)
Cone(1 , v) Cone(2 , v) = Cone(1 2 , v),
which will imply that Cone(1 , v) Cone(2 , v) (v). As above, one inclusion
is obvious, and for the other, take u Cone(1 , v) Cone(2 , v). Then
u = u1 + 1 v = u2 + 2 v,
where ui i and i 0. We may assume that 1 2 , in which case u2 = u1 +

(2 1 ) Cone(1 , v). Thus u2 2 Cone(1 , v), which by (11.1.1) implies
u2 2 Cone(1 , v) = 1 2 .
516
Thus u = u2 + 2 v Cone(1 2 , v), as desired.
Finally, we need to show that (v) is a refinement of . Since every cone of

is clearly contained in a cone of , we need only show that each is
a union of cones in (v). When v
/ , this is obvious since (v). On the
other hand, when v , pick cone generators mi , i = 1, . . . , r, and note that
v implies hmi , vi 0 for all i. Then set
(v)
(11.1.3)
= min
hmi ,vi>0
hmi , ui
.
hmi , vi
P
We claim that u0 = u v lies in . To prove this, take m = ri=1 i mi ,
i 0, and compute

Pr
hm, u0 i =
i=1 i mi , u v
hm , ui

P
P
i
= hmi ,vi=0 i hmi , ui + hmi ,vi>0 i hmi , vi
.
hmi , vi
The first sum is 0 since u , and the second is 0 by the definition of . Hence
u0 ( ) = , as claimed.
The definition (11.1.3) implies that we can pick i with hmi , vi > 0 and =
Then hmi , u0 i = 0, so that u0 lies in the face = Hmi of . This face does
not contain v since hmi , vi > 0, hence u Cone(, v) (v).

hmi ,ui
hmi ,vi .
Example 11.1.4. Let = Cone(u1 , . . . , ur ) be a smooth cone in NR Rn .

In 3.3, we defined the star subdivision () using u0 = u1 + + ur . It is easy
to check that this is the star subdivision (u0 ) since u0 N is primitive. See
Figure 9 in 3.3 for a picture when n = 3.
Example 11.1.5. Figure 1 shows a fan in R3 consisting of two 3-dimensional

cones (one simplicial, the other not) that meet along a 2-dimensional face . The
Figure 1. A fan and its star subdivision (v)
517
figure also shows the star subdivision induced by a primitive vector v Relint( ).
The fan (v) has five 3-dimensional cones.
The definition of (v) implies that v generates a 1-dimensional cone of (v)

and hence gives a divisor on X (v) . Recall that a Weil divisor is Q-Cartier if some
positive integer multiple is Cartier.
Proposition 11.1.6. The star subdivision (v) has the following properties:
(a) The 1-dimensional cones of (v) consist of the 1-dimesional cones of plus
the ray v = Cone(v) generated by v, i.e.,
(v)(1) = (1) {v }.
(b) The torus-invariant divisor prime Dv is Q-Cartier on X (v) .
(c) The induced toric morphism : X (v) X is projective.
Proof. The first assertion is Exercise 11.1.3. To show that Dv is Q-Cartier, we
construct a support function : | (v)| R as follows. For Cone(, v) (v),
consider the linear function defined on Span(Cone(, v)) = Span( )+Rv that maps
Span( ) to 0 and v to 1. This is well-defined since v
/ Span( ). Then define
|Cone(,v) to be the restriction of this linear map to Cone(, v). The remaining
cones in (v) are the cones not containing v. We set | = 0 for each of
these. By (11.1.1) and (11.1.2), we get a well defined map : | (v)| R that is
linear on each cone of (v). Furthermore, is clearly defined over NQ since
is rational and v N. It follows that some positive integer multiple k is integral
with respect to N, i.e., k SF( (v), N). Then by Theorem 4.2.12,
X
Dk =
k (u ) D
(v)(1)
is a Cartier divisor on X (v) . However, v is primitive and hence is the ray generator
of v . Since vanishes on all other ray generators, we have Dk = k (v) Dv =
k Dv . This proves that Dv and hence Dv are Q-Cartier.
For the final assertion, we use the projectivity criterion from Theorem 7.2.12.
Since (v) refines , it suffices to find a torus-invariant Cartier divisor on X (v)
whose support function is strictly convex with respect to the fan
(v) = { (v) | }
for every . The required Cartier divisor is easy to find: it is the divisor k Dv ,
where k > 0 is from the previous paragraph, with support function k . We can
ignore the factor of k when thinking about strict convexity. Hence the proof will be
complete once we prove that has the desired strict convexity.
When v
/ , this is obvious since (v) in this case. So it remains to consider what happens when v . By restricting to Span(), we can assume that is
518
full dimensional. Then, as noted in the discussion leading up to Proposition 7.2.3,

the strict convexity criteria from Lemma 6.1.13 apply to this situation.
S The analysis of (v) given in the proof of Lemma 11.1.3 shows that =
Cone(, v), where the union is over all facets of not containing v. Now fix
such a facet and consider the cone = Cone(, v) (v). Pick m M
such that Hm = . Since v \ , we have
hm, u i = 0, (1) and hm, vi > 0.
(11.1.4)
We claim that after rescaling m by a positive constant, we have (u) = hm, ui for
all u . To see why, note that is linear on , vanishes on , and takes the value
1 on v. Comparing this to (11.1.4), our claim follows immediately.
Given any (1) \ (1), the way we picked m implies that
hm, u i > 0.
However, m represents on , and
(1) \ (1) = (v) (1) \ (1)
by part (a) and the definitions of (v) and (v) . By (f) (a) of Lemma 6.1.13,
it follows that | is strictly convex with respect to (v) .

Simplicialization. As an application of star subdivisions, we prove that every fan
has an efficient simplicial refinement. Here is the precise result.
Proposition 11.1.7. Every fan has a refinement with the following properties:
(a) is simplicial.
(b) (1) = (1).
(c) contains every simplicial cone of .
(d) is obtained from by a sequence of star subdivisions.
(e) The induced toric morphism X X is projective.
Proof. First observe that part (c) follows from part (b). To see why, suppose that
is a refinement of satisfying (1) = (1) and let be simplicial. If
lies in , then (1) = (1) implies S

(1) (1), and hence is a face
of because the latter is simplicial. Since = , , it follows that some
in this union must equal , i.e., .
Note also that part (e) is a consequence of part (d). This follows because, first,
star subdivisions give projective morphisms by Proposition 11.1.6, and second,
compositions of projective morphisms are projective by Proposition 7.0.5.
Hence it suffices to find a sequence of star subdivisions that lead to a simplicial fan with the same 1-dimensional cones as . Our proof, based on an idea of
Thompson [270], uses complete induction on
r = |{ (1) | D is not Q-Cartier on X }|.
519
If r = 0, then every D is Q-Cartier. Since the D generate Cl(X ), every Weil

divisor on X is Q-Cartier. By Proposition 4.2.7, we conclude that is simplicial.
If r > 0, then we can pick (1) such that D is not Q-Cartier. Consider
the star subdivision (u ), where u is the ray generator of . Proposition 11.1.6
implies that
(u )(1) = (1) {} = (1)
and that D is Q-Cartier on X (u ) . Note also that if D is Q-Cartier on X , then

it remains Q-Cartier on X (u ) (think support functions). It follows easily that
{ (u )(1) | D is not Q-Cartier on X (u ) }
has strictly fewer than r elements. Our inductive hypothesis gives a refinement
of (u ) which is easily seen to be the desired refinement of .

In Exercise 11.1.4 you will show that the number of star subdivisions needed
to create the simplicial refinement of described in Proposition 11.1.7 is at
most the rank of Cl(X )/Pic(X ). A different proof of Proposition 11.1.7 can be
found in [99]. See also [93, Thm. V.4.2].
The Multiplicity of a Simplicial Cone. When we have a fan that is simplicial but
not smooth, we need to subdivide the non-smooth cones to make them smooth.
The elegant approach used in Chapter 10 does not generalize to higher dimensions.
Instead, we will use the multiplicity of a simplicial cone from 6.4 and show how a
carefully chosen star subdivision of a non-smooth cone can lower its multiplicity.
Let us recall the definition of multiplicity. Given a simplicial cone NR with
generators u1 , . . . , ud , let N = Span() N and note that Zu1 + + Zud N has
finite index. Then the multiplicity or index of is the index
(11.1.5)
mult() = [N : Zu1 + + Zud ].
The multiplicity of a simplicial cone has the following properties.

Proposition 11.1.8. Let NR be a simplicial cone and let u1 , . . . , ud and N be
as in (11.1.5). Then:
(a) is smooth if and only if mult() = 1.
(b) mult() is the number of points in P N, where
Pd
P =
i=1 i ui : 0 i < 1}.
P
(c) Let e1 , . . . , ed be a basis of N and write ui = dj=1 ai j e j . Then
mult() = |det(ai j )|.
(d) mult( ) mult() whenever .
520
Proof. The proof of part (a) is straightforward, and part (b) is equally easy since
the composition
P N N N /(Zu1 + + Zud )
is a bijection (Exercise 11.1.5). Part (c) is the standard fact that |det(ai j )| is the
index of Zu1 + + Zud in N (see [242, Corollary 9.63]). Finally, if , then
(1) (1). This implies P P , and then part (d) follows from part (b).

Some further properties of mult() are discussed in Exercise 11.1.5.
The Resolution. We can now construct a fan that resolves the singularities of a
normal toric variety X .
Theorem 11.1.9. Every fan has a refinement with the following properties:
(a) is smooth
(b) contains every smooth cone of .
(c) is obtained from by a sequence of star subdivisions.
(d) The toric morphism : X X is a projective resolution of singularities.
Proof. We first observe that part (d) follows from parts (a), (b) and (c). Since
the refinement comes from a sequence of star subdivisions, : X X is
projective by the proof of Proposition 11.1.7. Furthermore, is the identity on
the smooth locus of X by Proposition 11.1.2 and X is smooth Theorem 3.1.19.
Hence is a resolution of singularities.
To produce a refinement that satisfies (a), (b) and (c), we begin with Proposition 11.1.7, which uses star subdivisions to construct a simplicial refinement with
the same simplicial cones as . This refinement does not change the smooth cones
since smooth cones are simplicial.
Hence we may assume that is simplicial. To measure how close is to being
smooth, we define
mult() = max mult().
Note that mult() = 1 if and only if is a smooth fan. We will show that whenever
mult() > 1, we can find a star subdivision (v) that does not change the smooth
cones of and satisfies either
(11.1.6)
mult( (v)) < mult(), or

mult( (v)) = mult() and (v) has fewer cones of this multiplicity.
Once this is done, the theorem will follow immediately.

Suppose mult() > 1 and let 0 have maximal multiplicity. By part (b) of
Proposition 11.1.8, there is v P0 N \ {0}. The star subdivision (v) consists
521
of the cones of not containing v together with the cones Cone(, v) where v
/
and is a face of a cone containing v. We claim that
(11.1.7)
mult(Cone(, v)) < mult().
Assume (11.1.7) for the moment. Part (b) of Proposition 11.1.8 shows that v
lies in no smooth cone of , so that (v) contains all smooth cones of . Since all
cones of (v) either lie in or satisfy the inequality (11.1.7), it follows that when
we replace to (v), we lose at least one cone of maximum multiplicity (namely,
0 ) and create no new cones of this multiplicity. Thus (v) satisfies (11.1.6).
It remains to prove (11.1.7). Given v P0 N {0} as above, we can write
v = 1 u1 + + d ud ,
0 < i < 1,
u1 , . . . , ud , ud+1 , . . . , us ,
s = dim .
where u1 , . . . , ud are the ray generators of the minimal face 0 of 0 containing v.

We also have Cone(, v) . Since v implies that 0 is a face of , we can
write the ray generators of as
Also, v
/ implies that there is i {1, . . . , d} such that ui
/ . Then
mult(Cone(, v)) mult(Cone(u1 , . . . , ubi , . . . , ud , ud+1 , . . . , us , v)

= |det(u1 , . . . , ubi , . . . , ud , ud+1 , . . . , us , v)|
= |det(u1 , . . . , ubi , . . . , ud , ud+1 , . . . , us , i ui )|
= i |det(u1 , . . . , us )| = i mult() < mult().
The first line follows from part (d) of Proposition 11.1.8 and the second follows
P
from part (c). Here, det(w1 , . . . , wd ) = det(bi j ) where wi = dj=1 bi j e j . The last
two lines follow from 0 < i < 1 and standard properties of determinants. This
completes the proof of (11.1.7), and the theorem follows.

The Exceptional Locus. The resolution : X X from Theorem 11.1.9 is an
isomorphism above the smooth locus of X . It remains to consider what happens
above the singular locus. The exceptional locus of is Exc() = 1 ((X )sing ).
Proposition 11.1.2 implies that the irreducible components of (X )sing are given
by the orbit closures
(X )sing = V (1 ) V (s )
where 1 , . . . , s are the minimal non-smooth cones of . Hence, to understand the
exceptional locus Exc(), it suffices to describe 1 (V (i )) for 1 i s. We will
use the following more general result.
Proposition 11.1.10. Let be a refinement of with induced toric morphism
: X X , and fix . Then the irreducible components of 1 (V ()) are
1 (V ()) = V (1 ) V (r ),
where 1 , . . . , r are the minimal cones of that meet the relative interior of .
522
Proof. Since is equivariant, it maps TN -orbits into TN -orbits. More precisely, we

know from Lemma 3.3.21 that for , a cone satisfies (O( )) O()
if and only if is the minimal cone of containing . Furthermore, the latter
happens if and only if intersects the relative interior of . From here, the proof
is an easy application of the Orbit-Cone correspondence (Exercise 11.1.6).

Example 11.1.11. Our first example of a toric resolution was in Example 10.1.9,
where we considered the refinement of the cone shown in Figure 2. Here, U
Figure 2. A cone with smooth refinement
has a unique singular point, and the minimal cones of that meet the interior of
are the interior rays of . Hence the exceptional locus of X U is a divisor.
Example 11.1.12. Consider = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) R3 . Figure 3 on
the next page shows and three smooth refinements 1 , 2 , 3 that give
X3
X1
x
x| xx
FFF
F"
FFF
F# {xxxx
X2
For 1 , 2 , the smallest cone meeting the interior of is 2-dimensional, so the

exceptional locus is P1 in each case. On the other hand, the ray in Figure 3 is the
smallest cone of 3 meeting the interior of . Hence the exceptional locus is V ( ),
which by Example 3.2.8 is P1 P1 .
The exceptional loci in the above resolutions are related by the diagram
P PJ
JJJ
tt
t
J$
t
zt
_
_
_
_
_
_
_
/ P1
1
P II
u
III
u
$ zuuu
1
(11.1.8)
{pt}
where the nontrivial maps in (11.1.8) are the projections onto the factors of P1 P1
(Exercise 11.1.7).
Example B.6.1 shows how to find the refinement 1 of using Sage [262].
523
x
Figure 3. The cone with smooth refinements 1 , 2 , 3
524
Example 11.1.12 was first described by Atiyah [8] in 1958. The rational map
X1 99K X2 in (11.1.8) is an isomorphism outside of a set of codimension 2. In the
minimal model program, such rational maps are called flips or flops, depending on
the context. We will say more about this in Chapter 15.
In 10.4, we noted that when X is a toric surface, there is a unique minimal
toric resolution of singularities X X . Example 11.1.12 shows that unique
minimal resolutions do not exist in dimensions 3.
Exercises for 11.1.
11.1.1. The proof of Proposition 11.1.2 constructed an isomorphism U,N U,N TN2
and claimed that it induced an isomorphism of orbits
Prove this.
ON () ON () TN2 .
11.1.2. Let Cone(, v) be cone of the star subdivision (v) containing v, and let be the
face of Cone(, v) determined by m Cone(, v) . Thus = Hm Cone(, v). As noted in
the text, we have m and hm, vi 0.
(a) If hm, vi = 0, prove that = Cone(Hm , v) (v).
(b) If hm, vi > 0, in which case = Hm (v).
11.1.3. Let (v) be the star subdivision determined by v and let v = Cone(v). Prove that
(v)(1) = (1) {v }.
11.1.4. Recall that if G is a finitely generated abelian group and Gtor is the subgroup of
elements of finite order, then G/Gtor Zr , where r is called the rank of G. Prove that the
number of star subdivisions need to create the simplicial refinement of described in
Proposition 11.1.7 is at most the rank of Cl(X )/Pic(X ).
11.1.5. This exercise will study the multiplicity mult() of a simplicial cone .
(a) Prove part (a) of Proposition 11.1.8.
(b) Prove part (b) of Proposition 11.1.8.
(c) Prove that mult() is the normalized volume of P (N )R = Span(), where normalized means that the parallelotope determined by a basis of N has volume 1.
Hint: Use part (c) of Proposition 11.1.8 and remember how to compute the volume of
a parallelotope (see [263, Theorem 15.2.1]).
(d) Let be a face of and let u1 , . . . , ud be the minimal generators of . Prove that
mult() = mult( )[N : N + Zu1 + + Zud ]
and use this to give another proof of part (d) of Proposition 11.1.8.
11.1.7. Verify the claims made in Example 11.1.12.
n+1
11.1.8.
consisting of Cone(e0 , . . . , en ) Rn+1 and its faces. Let
Pn Let be the fan in R
v = i=0 qi ei , where the qi are positive integers satisfying gcd(q0 , . . . , qn ) = 1. The star
subdivision (v) gives a toric morphism : X (v) X = Cn+1 . Prove that 1 (0)
P(q0 , . . . , qn ).
11.2. Other Types of Resolutions
525
11.1.9. Consider the complete fan in R3 with six 3-dimensional cones and minimal
generators (1, 1, 1). Thus consists of the cones over the faces of a cube centered at
the origin in R3 .
(a) Show that the simplicialization process described in Theorem 11.1.7 requires exactly
two star subdivisions to produce a simplicial refinement of .
(b) Show that is smooth, so that X X is a projective resolution of singularities.
(c) Find a smooth refinement of such that (1) = (1) but X is not projective.
Conclude that X X is not a projective resolution of singularities.
11.1.10. The simplicial fan constructed in Proposition 11.1.7 is efficient (no new rays) but
noncanonical (no canonical choice for the star subdivisions used in construction). Here
you will explore a different method for simplicialization that is canonical but introduces
many new
Prays. The barycenter v of a nonzero cone NR is the minimal generator
of Cone( (1) u ) N. Now list the nonzero cones of the fan as 1 , . . . , r , where
dim 1 dim r . Then the barycentric subdivision of is the fan

() = (vr ) (vr1 ) (v1 ).
Thus () is obtained from by a sequence of star subdivisions where we use the barycenters of the cones, starting with the biggest cones and working down.
(a) Consider the fan in R3 from Exercise 11.1.9. Draw two pictures to illustrate the
construction of (). The first picture, drawn on the surface of the cube, should
illustrate the intermediate fan obtained by taking star subdivisions at the barycenters
of the 3-dimensional cones. The second picture should show how this gets further
subdivided to obtain ().
(b) When the cones are listed by increasing dimension, cones of the same dimension can
appear in any order. Hence the list is not unique. However, they all give the same
barycentric subdivision (). Prove this.
(c) Prove that () is a simplicial fan.
Part (a) of this exercise was inspired by [93, p. 74].

We next consider simple normal crossing, crepant, log, and embedded resolutions,
which are important refinements of what it means to be a resolution of singularities.
SNC Resolutions. Example 11.1.12 illustrates that the exceptional locus of a resolution of singularities need not always be a divisor. Yet in many situations in
algebraic geometry, one wants a divisor, and in fact, one often wants a simple
P
normal crossing divisor, as mentioned in 8.1. Recall that a divisor D = i Di
on a smooth variety X has simple normal crossings (SNC for short) if every Di is
smooth and irreducible, and for all p X , the divisors containing p meet as follows:
if I p = {i | p Di }, then the tangent spaces Tp (Di ) Tp (X ) meet transversely, i.e.,
\

codim
Tp (Di ) = |I p |.
iIp
526
For example,Pgiven a smooth toric variety and a nonempty subset A (1), the
divisor D = A D is SNC.
Definition 11.2.1. A resolution of singularities f : X X is called SNC if the

exceptional locus Exc( f ) is a divisor with simple normal crossings.
Theorem 11.2.2. Every fan has a refinement as in Theorem 11.1.9 such that
the toric morphism : X X is a projective SNC resolution of singularities.
Proof. By Theorem 11.1.9, has a smooth refinement 0 that gives a resolution

of singularities. Let V (i ) be an irreducible component of (X )sing and let 0
be a minimal cone that meets the relative interior
Pof i . If is not a ray, then
consider the star subdivision 0 (v ), where v = (1) u . Since 0 is smooth,
this is the fan denoted 0 () in Definition 3.3.17. In the discussion that followed
the definition in 3.3, we saw that corresponding toric morphism is the blowup
BlV () (X0 ) X0 . Since v lies in the relative interior of , the inverse image of
V () is the divisor corresponding to the ray of 0 (v ) generated by v .
The composed map BlV () (X0 ) X is a resolution of singularities whose
exceptional locus has one less non-divisorial component. Repeating this finitely
many times gives a projective resolution (star subdivisions give projective morphisms) whose exceptional locus is a divisor, automatically SNC by the remark
following Definition 11.2.1.

Example 11.2.3. The resolution X1 U constructed in Example 11.1.12 is not
SNC. Applying the process described in the proof of Theorem 11.2.2 gives the
resolution X3 U , which is SNC.
Log Resolutions. There are several types of log resolutions. Here we discuss the
one for Q-divisors.
P
Definition 11.2.4. Let D = i di Di be a Q-divisor on a normal variety X . Then
aS projective resolution of singularities f : X X is a log resolution of D if
1 (D ) Exc( f ) is a SNC divisor, where as usual Exc( f ) is the exceptional
i
i f
locus of f .
For
P a toric variety X , the natural choice for D is a torus-invariant Q-divisor
D = d D , a Q. In this case, a projective SNC resolution X X is automatically a log resolution of (X , D) since any collection of torus-invariant prime
divisors on X is SNC.
In 11.3 we will discuss a different type of log resolution, where the given data
consists of a normal variety X and a sheaf of ideals a OX . See [186, 9.1.B] for
more on log resolutions.
Crepant Resolutions. Given the importance of the canonical sheaf X , it is natural
to ask how X changes in a resolution of singularities f : X X . The nicest case is
527
when X is normal and Q-Gorenstein, i.e., X = OX (KX ) for a Q-Cartier canonical

divisor KX . (Note that if one canonical divisor is Q-Cartier, then they all are.)
Suppose that KX is Cartier, where > 0 is an integer. It follows that f (KX )
is a Cartier divisior, and then f KX is defined to be the Q-divisor
f KX =
Example 11.2.7 below will show that
1
f (KX ).
f KX need not
be integral.
Definition 11.2.5. Let X be a normal Q-Gorenstein variety. Then a resolution of

singularities f : X X is crepant if f (KX ) is integral and KX f (KX ).
Since KX is only unique up to linear equivalence, the definition of crepant is
often stated as KX = f KX . The term crepant is due to Reid [235] and signals
that there is no discrepancy between KX and f KX .
Example 11.2.6. The toric variety U of = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) R3
is Gorenstein since the minimal generators of lie in the plane hm, ui = 1, for
m = e1 e2 . Thus the canonical divisor is KU = div( m ).
For the fan 1 from Figure 3 of Example 11.1.12, the resolution X1 U is

crepant since
(KU ) = (divU ( m )) = divX ( m ) = KX .
The subscripts indicate where the divisor is being computed. You will verify this
computation in Exercise 11.2.1, where you will also show that the SNC resolution
X3 U from Example 11.1.12 is not crepant.
Example 11.2.7. Consider = Cone(e1 , e2 , e3 ) NR = R3 relative to the lattice

N = {(a, b, c) Z3 | a + b + c 0 mod 2}.
The minimal generators are ui = 2ei , 1 i 3, and the class group of U is Z/2Z,
where the torus-invariant divisors Di all map to the generator. It follows that the
canonical divisor KU = D1 D2 D3 is not Cartier, but 2KU = div( m ) for
m = e1 e2 e3 MR (Exercise 11.2.2).
The star subdivision with respect to u0 = e1 + e2 + e3 NR gives a resolution
of singularities : X U . The four minimal generators of give divisors
denoted D0 , D1 , D2 , D3 by abuse of notation. Then
(2KU ) = (divU,N ( m )) = divX ( m )
= 3D0 2D1 2D2 2D3 = 2KX D0 ,
so that KX = (KU ) + 21 D0 (Exercise 11.2.2). Thus is not crepant. Note also
that (KU ) is not an integral divisor even though KU is.
For a toric surface, we have the minimal resolution X X constructed in

10.2. It is easy to characterize when this resolution is crepant.
528
Proposition 11.2.8. Let X be a toric surface and consider the minimal resolution
of singularities : X X from 10.2. Then the following are equivalent:
(a) is crepant.
(b) X is Gorenstein.
(c) The singularties of X are at worst rational double points.
Proof. The equivalence (b) (c) follows from Proposition 10.1.6. For (a)
(c), one can reduce to the case of an affine surface U with ray generators u1 , u2 .
Here, the key observation is that (a) is equivalent to saying that the minimal smooth
refinement of has ray generators lying on the line segment connecting u1 and u2 ,
u1
u2
Figure 4. Cone with refinement ray generators on a line segement
as illustrated in Figure 4. Using the methods of 10.2, one sees that the rational
double points are the only toric surface singularities where is happens. We leave
the details as Exercise 11.2.3.

Here is an example that we have already seen.
Example 11.2.9. Let P be a reflexive polygon in R2 . In Theorem 10.5.10, we
proved that the minimal resolution X XP comes from the fan given by the lattice
points on the boundary of P . Figure 5 makes it clear that X XP is crepant, a
fact we used in the proof of Theorem 10.5.10.
s
@
s @
s

s

s
A
@
s A@s
A @
s AH
s @s
H@
H
Hs
s
s @
s
Figure 5. A reflexive polygon P and its dual P
529
More generally, crepant resolutions of toric varieties coming from reflexive

polytopes are important in mirror symmetry because they give rise to Calabi-Yau
varieties. A smooth projective variety Y is Calabi-Yau if Y is trivial and
(11.2.1)
H 1 (Y , OY ) = = H d1 (Y , OY ) = 0, d = dim Y .
Now fix a reflexive polyope P MR of dimension 2. We know from 8.3 that

XP is a Gorenstein Fano toric variety, so that KXP is Cartier and ample. Assume
also that : X XP is a crepant resolution. Then KX = (KXP ) is basepoint free.
By the Bertini theorem [131, Cor. III.10.9 and Ex. III.11.3], a generic section of
OX (KX ) gives a smooth connected hypersurface Y X such that Y KX .
We call Y an anticanonical hypersurface.
Proposition 11.2.10. With the above hypotheses, Y is Calabi-Yau.
Proof. Since IY = OX (Y ) by Proposition 4.0.28, we have IY OX (KX )
Xn , n = dim X . Then the desired vanishing (11.2.1) follows from the long exact
cohomology sequence associated to
0 IY OX OY 0
plus the vanishing of
H p (X , Xn )
for p < n from Theorem 9.3.2.
To compute the canonical bundle of Y , we use the adjunction formula from

Example 8.2.2:
Y X OX (IY /IY2 ) X (Y ) OX OY ,
where we use IY /IY2 IY OX (OX /IY ) = OX (Y ) OX OY . Since
X = OX (KX ) OX (Y ),
one sees immediately that Y = Yn1 is trivial.
Example 11.2.11. Continuing with Example 11.2.9, let P be a reflexive polygon

with minimal resolution X XP . In this case, a generic section of OX (KX ) is
a smooth connected curve C X . We have two ways to compute its genus g:
(a) Recall from 10.5 that g is the sectional genus of XP , which equals |Int(P) M|
by Proposition 10.5.8. Since P is reflexive, the origin is its unique interior
point. Hence g = 1.
(b) The canonical bundle of C is trivial by Proposition 11.2.10. The canonical
divisor has degree 0, so 0 = 2g 2 by (10.5.3), hence g = 1.
Thus C is an elliptic curve, which is a Calabi-Yau variety of dimension 1.
Unfortunately, for a reflexive polytope of dimension 3, XP need not have a

crepant resolution. Here, the best one can do is the maximal projective crepant
partial desingularization described by Batyrev in [16]. The resulting anticanonical
hypersurfaces are singular Calabi-Yau varieties (see [68, Def. 1.4.1]).
We will study toric singularities in more detail in 11.4.
530
Embedded Resolutions. Many treatments of the resolution of singularities use embedded resolutions. For an embedded resolution, we have an irreducible variety X
contained in a smooth variety W , which we think of as the ambient space. The
idea is to repeatedly blow up the ambient space along smooth subvarieties (thereby
changing the ambient space) until the proper transform of X becomes smooth.
Here is a more careful description. An irreducible smooth subvariety Z W
gives the blowup f : BlZ (W ) W , where BlZ (W ) is smooth and the exceptional
locus Exc( f ) (the subset of BlZ (W ) where f is not an isomorphism) is the divisor
given by the projective bundle of the normal bundle of Z W . We call Z the center
of the blowup. Then X W has a proper transform in BlZ (W ) defined as the
Zariski closure
f 1 (X \ Z) BlZ (W ).
Then one seeks a series of smooth centers Zi Wi , 0 i , such that W0 = W ,
Wi+1 = BlZi (Wi ), and Zi+1 is transverse to the exceptional locus of Wi+1 W .
Furthermore, the final map f : W W needs to have the following properties:
The induced map f |X : X X is a resolution of singularities, where X is the
proper transform of X in W .
X is transverse to the exceptional locus of f .
The last bullet implies that f |X : X X is a SNC resolution of singularities, and

one can show that f |X is a projective morphism. A careful explanation of what
this means can be found in [138], along with a discussion of further properties one
can impose on an embedded resolution.
In the toric version of this, we have an equivariant embedding X W of toric
varieties, with W smooth, and at each stage of the above process, Wi is toric, and
the center Zi Wi is an orbit closure. Then Zi is smooth, Wi+1 = BlZi (Wi ) is toric,
and Wi+1 Wi is a toric morphism. When such an embedded resolution exists, the
result is a toric SNC resolution of singularities X X achieved entirely through
blowups of torus-invariant centers in smooth toric ambient spaces. The papers [25],
[32] and [117] prove the existence of embedded toric resolution, and more importantly, they give algorithms for finding the centers Z0 , Z1 , . . . needed at each stage
of the resolution process. See also [268]. In contrast, the resolution constructed in
Theorem 11.1.9 depended on many choices and is far from unique.
The proof of toric embedded resolution, though much simpler than the general
case, is still not easy. Hence we will confine ourselves to some examples, together
with some remarks about nonnormal toric varieties.
Example 11.2.12. The toric variety V(xy zw) C4 is the affine toric variety U
of the cone from Example 11.1.12. We can resolve the singularity of U by
blowing up the origin in C4 . The proper transform of V(xy zw) in Bl0 (C4 ) is the
toric variety X3 from Example 11.1.12. There are two ways to see this.
531
First, the isomorphism U V(xy zw) induces the isomorphism (C )3

V(xy zw) (C )4 given by (t1 ,t2 ,t3 ) 7 (t1 ,t2 ,t3 ,t1 t2t31 ) (see Example 1.1.5). At
the level of one-parameter subgroups, this gives the embedding
: Z3 Z4 ,
(a, b, c) 7 (a, b, c, a + b c).
Since = R (R40 ) and R40 is the cone for C4 , the corresponding toric morphism
is the embedding : U C4 with image V(xy zw).
Let be the star subdivision of R40 at (1, 1, 1, 1). This gives the toric variety
X = Bl0 (C4 ). Via R , induces the refinement 3 of described in Example 11.1.12 (Exercise 11.2.4). This gives a toric morphism : X3 Bl0 (C4 )
whose image is the proper transform (we will say more about this below).
Second, the proper transform can studied by the method of Example 5.2.11.
The blowup has quotient representation
Bl0 (C4 ) = (C5 \ C {(0, 0, 0, 0)})/C ,
where (t, x, y, z, w) = (1 t, x, y, z, w), and the blowup map Bl0 (C4 ) C4
is given by
(t, x, y, z, w) 7 (tx,ty,tz,tw).
Via this map, 0 = xy zw becomes 0 = (tx)(ty) (tz)(tw) = t 2 (xy zw), so that

the proper transform is defined by xy zw = 0 in Bl0 (C4 ). There are two cases:
If t 6= 0, then t (t, x, y, z, w) = (1,tx,ty,tz,tw), so that this part of the proper
transform is isomorphic to V(xy zw) \ {0} C4 .
If t = 0, then this part of the proper transform is the fiber over 0 V(xy zw)
and is the subvariety of P 3 defined by xy = zw. This quadric surface is the
Segre embedding of P1 P1 .
One can also show that the proper transform is smooth by checking locally on the
affine pieces of Bl0 (C4 ), along the lines of Example 5.2.11 (Exercise 11.2.4).
The general situation has an interesting wrinkle. Given a smooth toric variety
X0 with torus TN0 , any subtorus T TN0 gives the toric variety
(11.2.2)
X = T X0
with torus T . However, X need not be normal, as shown by the simple example
where T = {(t 2 ,t 3 ) | t C } (C )2 and X = T = V(x 3 y 2 ) C2 .
For this reason, the embedded resolution of toric varieties described in [25],
[32] and [117] works with an equivariant embedding X W of toric varieties where
W is smooth but X is not assumed to be normal. These papers use cones and fans
for the ambient space W and its blowups, but use binomial equations for X and its
proper transforms, just as we did in the second half of Example 11.2.12.
532
Example 11.2.13. In Example 5.2.11 we showed that the proper transform of the
nonnormal toric variety V(x 3 y 2 ) C2 is smooth in Bl0 (C2 ). This not an embedded resolution since the proper transform is not transverse to the exceptional locus.
Two more blowups are needed to get an embedded resolution (Exercise 11.2.5).
While the toric variety X from (11.2.2) need not be normal, its normalization is
easy to construct. Suppose in general that X0 is a normal toric variety, where 0 is
a fan in (N0 )R . Then subtori of TN0 correspond bijectively to sublattices of N0 with
torsion-free quotient (Exercise 11.2.6). Let TN be the subtorus coming from N N0
and let X = TN X0 be its Zariski closure. Using the subspace NR (N0 )R , the
fan of the normalization of X is described as follows.
Proposition 11.2.14. Let X0 be the toric variety of a fan 0 in (N0 )R . Fix a
subtorus TN TN0 and let X = TN X0 . Then X is a (possibly nonnormal) toric
variety with torus TN whose normalization is the toric variety X of the fan
= { NR | 0 }
in NR . In particular, if X is normal, then X X .
Proof. The proof is similar to the proof of Theorem 3.A.5.
Example 11.2.15. As in Example 11.2.12, consider U V(xy zw) C4 , and let

X be the proper transform of V(xy zw) in Bl0 (C4 ). Then X is the Zariski closure
of its torus in Bl0 (C4 ), and we also know that X is smooth by the second half of
Example 11.2.12. Using Proposition 11.2.14, it is now easy to show that X is the
toric variety X3 , as claimed in Example 11.2.12 (Exercise 11.2.4).
Exercises for 11.2.

11.2.1. In Exercise 11.2.6,
(a) Prove that (divU ( m )) = divX ( m ) for any m M. Hint: Use Proposition 6.2.7.
(b) Show that the resolution X3 U from Example 11.1.12 is not crepant.
11.2.2. Fill in the details omitted in Example 11.2.7.

11.2.3. Complete the proof of Proposition 11.2.8. Hint: If has parameters d, k, then the
results of 10.2 give a smooth refinement of with minimal generators u0 , . . . , ur+1 such
that u0 = e2 , u1 = e1 , ur+1 = de1 ke2 . When are u0 , u1 , ur+1 collinear?
11.2.4. This exercise concerns Example 11.2.12.
(a) In the first half of the example, prove the claim that 3 is the refinement of induced
by the fan for Bl0 (C4 ) via the map R .
(b) Fill in the details omitted in the second half of the example.
(c) Use parts (a) and (b) together Proposition 11.2.14 to show that X3 is isomorphic to
the proper transform of V(xy zw) in Bl0 (C4 ).
11.2.5. Explain why the blowup Bl0 (C2 ) C2 does not give an embedded resolution of
singularities of V(x 3 y 2 ) C2 . Then construct a blowup of Bl0 (C2 ) that does.
533
11.2.6. Fix a lattice N0 . Prove that there is a bijection between sublattices of N0 with
torision-free quotient and subtori of TN0 .
11.2.7. Consider the complete fan in R3 consisting of the cones over the faces of the
cube with vertices (1, 1, 1). In Exercise 11.1.9 you constructed a refinement giving
a projective resolution of singularities.
(a) Explain why X X is not SNC.
(b) Find a nice refinement of that gives a projective SNC resolution of singularities.
11.2.8. The goal of this exercise and the next is to construct an embedded resolution of the
rational double point singularity Ak from Example 10.1.5. We will change notation slightly
and consider V(xz y k+1 ) C3 , k 1. The blowup Bl0 (C3 ) has homogeneous coordinates
t, x, y, z in the sense of Chapter 5 such that Bl0 (C3 ) C3 is given by (t, x, y, z) 7 (tx,ty,tz).
(a) Show that the proper transform of V(xz y k+1 ) is defined by xz = t k1 z k+1 .
(b) Bl0 (C3 ) is covered by three affine open subsets isomorphic to C3 . By the discussion
of local coordinates in 5.2, these affine charts are given by setting one of x, y, z equal
to 1. Show that the proper transform is smooth in the x = 1 and z = 1 charts and is
defined by xz = t k1 in the y = 1 chart. Conclude that the proper transform is normal.
(c) Show that the embedded resolution of V(xz y k+1 ) C3 can be done using k/2
blowups at smooth points of the ambient space. Hint: Use part (b) and recursion.
11.2.9. We continue the study of V(xz y k+1 ) C3 begun in Exercise 11.2.8, this time
focusing on the fan. First observe that V(xz y k+1 ) is the Zariski closure of the subtorus
{(t1 ,t2 ,t11t2k+1 ) | t1 ,t2 C } (C )3
corresponding to the sublattice
N = Z(1, 0, 1) + Z(0, 1, k + 1) Z3 .
Since V(xz y k+1 ) is normal (Example 10.1.5), Proposition 11.2.14 implies that it is the
affine toric variety of the cone = NR R30 since R30 is the cone for C3 .
(a) Show that = Cone((k + 1, 1, 0), (0, 1, k + 1)).
(b) Bl0 (C3 ) is the toric variety of the star subdivision of R30 at = (1, 1, 1). This induces
a refinement of , shown for k = 2 in Figure 6 on the next page. Prove that for
k 1, is the fan of the proper transform. Hint: Part (b) of the previous exercise.
(c) When k = 1, show that is smooth with exactly two 2-dimensional cones. Draw a
picture similar to Figure 6 and explain why it is simpler.
(d) When k 2, show that the fan has three 2-dimensional cones, where the outer two
cones are smooth and the middle cone gives Ak1 .
The methods of 10.2 resolve U by subdividing into k + 1 smooth cones (see Exercise 10.2.2). When k 2, the fan constructed in part (b) has the outermost cones of the
smooth refinement plus a middle cone that combines the remaining cones of the smooth
refinement. When k 3, blowing up Bl0 (C3 ) refines by adding two more cones from the
smooth refinement, and each successive blowup adds in two more cones from the smooth
refinement until the process is complete.
534
induced
refinement of
Figure 6. for k = 2 and refinement induced by blowing up 0 C3 in Exercise 11.2.9
11.3. Rees Algebras and Multiplier Ideals

So far, we have constructed resolution of singularities using repeated blowups. By
using the blowup of a sheaf of ideals, it is often possible to construct the resolution in a single blowup. From an abstract point of view, this is a standard fact.
Given any birational projective morphism f : X Y of irreducible varieties with Y
quasiprojective, there is an ideal sheaf I OY such that f : X Y is the blowup
map : BlI (Y ) Y . This is proved in [131, Thm. II.7.17]. In the toric case, we
will see that the blowup has an especially nice interpretation.
Blowups and Polyhedra. We begin with a full dimensional lattice polyhedron P
in MR . Thus P = Q + C, where Q is a lattice polytope and C is strongly convex.
In order to get a blowup, we assume in addition that C is full dimensional, so that
= C NR is also full dimensional and strongly convex. Then P = Q + and
the normal fan P is a refinement of . Hence we have a birational toric morphism
: XP U ,
which is projective by Theorem 7.1.10. Our goal is to represent this morphism as
the blowup of an ideal a in the coordinate ring R = C[ M] of U .
Translating P by an element of M has no effect on XP . For m Int( ) M,

one sees easily that P + m for integers 0. Hence we may assume that
P . This gives the ideal
M
M
(11.3.1)
a=
C m R = C[ M] =
C m,
m M
mPM
and from a R, we get the Rees algebra

R[a] =
M
k=0
ak t k ,
535
where t is a dummy variable that keeps track of the grading. This graded R-algebra
is generated by its elements of degree 1 as an R-algebra and satisfies R[a]0 = R.
Using the Proj construction described in the appendix to Chapter 7, we get a
projective morphism
(11.3.2)
X = Proj(R[a]) Spec(R) = U .
This is the blowup of Spec(R) with respect to a, denoted Bla(Spec(R)). For us, a
key property of this blowup involves the ideal sheaf aOX , defined as follows. On
an affine open subset U = Spec(A) X , the map U Spec(R) comes from a ring
homomorphism R A, and aOU is the sheaf associated to the ideal aA A. For
the blowup morphism (11.3.2) the sheaf aOX is a line bundlethis is [131, Prop.
II.7.13]. See [131, II.7] for a discussion of blowing up ideals and ideal sheaves.
Besides the Rees algebra, the polyhedron P gives a second graded algebra.
Recall from 7.1 that from P MR we get the cone C(P) MR R, where the
slice at height k > 0 is kP and the slice at height 0 is (see Figure 1 from 7.1).
The associated semigroup algebra SP = C[C(P) (M Z)] is graded by height,
and (SP )0 = R by the discussion following the proof of Theorem 7.1.13.
Proposition 11.3.1. With the above setup, there is an inclusion R[a] SP that
makes SP the integral closure of R[a]. Furthermore, this inclusion is an equality if
and only P is normal.
Proof. First observe that R[a] is a semigroup algebra. Let

A = ( M) {0} (P M) {1} M Z.
We leave it to reader to show that R[a] = C[NA ] (Exercise 11.3.1). Also note:
We have ZA = M Z since is full dimensional and P .
Cone(A ) = C(P).
Then Proposition 1.3.8 implies that SP = C[C(P) (M Z)] is the integral closure
of R[a] = C[NA ].
Since R[a] and SP are equal in degrees 1, the final assertion of the proposition
is an easy consequence of the definition of normal polyhedron.

To translate Proposition 11.3.1 into geometric terms, we will use the Cartier
divisor DP of the polyhedron P. Recall that the Cartier data of DP is given by the
vertices v P. This means that the support function of DP satisfies (u) = hv, ui
when u lies in the cone v P corresponding
to v. Thus (u) 0 since v P
P
and u v . It follows that D = (u )D is an effective divisor. Since
P
DP = (u )D , we conclude that DP = D, where D 0.
In particular, OXP (DP ) = OXP (D) is an ideal sheaf of OXP . This compares to
aOXP as follows.
536
Corollary 11.3.2. With the same hypotheses as Proposition 11.3.1, we have:

(a) The map XP U factors
XP Bla(U ) U ,
where the first morphism is normalization and the second is the blowup of a.
(b) aOXP = OXP (DP ), where DP is the Cartier divisor associated to P.
Proof. For part (a), recall from Theorem 7.1.13 that XP = Proj(SP ), where SP is the
integral closure of R[a] by Proposition 11.3.1. You will prove in Exercise 11.3.2
that the inclusion R[a] SP induces the normalization map
XP = Proj(SP ) Proj(R[a]).
For part (b), take a vertex v P. This gives the cone v P , and since
P = Conv(m | m a), one sees easily that v = Cone(m v | m a). Then any
m a can be written
m = v mv v C[v M],
from which we conclude that a C[v M] = v C[v M] since v a. Since XP
is covered by the affine open subsets Uv , it follows that a C[v M] is the line
bundle of the Cartier divisor whose Cartier data is given by the vertices of P. As
noted above, this divisor is DP , which completes the proof.

This corollary shows that the map XP U is a single blowup followed by a
normalization. Moreover, replacing P with a sufficiently high multiple, we may
assume that P is normal (Theorem 7.1.9). This does not change the normal fan, so
that in this case, the map XP U is the blowup of the ideal a.
By Theorem 7.2.4, it follows that any projective toric resolution X U is
the blowup of a suitable ideal a C[ M]. For the special case of blowing up
the origin 0 Cn or more generally a coordinate subspace {0} Cnr Cn , the
ideal a is easy to describe (Exercise 11.3.3).
More on the Ideal. If we fix the cone NR , then we can give a purely algebraic
description of the ideals that arise from (11.3.1). We first need a definition.
Definition 11.3.3. An ideal a in a ring R is integrally closed if whenever an element
r R satisfies an equation
r k + a1 r k1 + + ak1 r + ak = 0
with ai ai for 1 i k, we have r a.
537
Proposition 11.3.4. Fix a full dimensional strongly convex cone NR . Then the
maps
L
P 7 a = mPM C m
a 7 P = Conv(m | m a)
induce a bijection

nonzero integrally closed
lattice polyhedra P
ideals a C[ M]
.
with recession cone
generated by characters
Proof. Given a lattice polyhedron P with recession cone , we need to

show that a is integrally closed. Standard arguments (see [89, p. 137] or [266, Cor.
1.3.1]) imply that
a = {r C[ M] | r k + a1 r k1 + + ak = 0 for some ai ai }
is an integrally closed ideal containing a. Hence it suffices to prove a = a.

Since a is generated by characters, it is invariant under the TN -action on R,
which easily implies that a is TN -invariant. Then Lemma 1.1.16 implies that a is
also generated by characters (see also [266, Prop. 1.4.2]). Thus, to prove a = a, we
need to show that if a character m a satisfies
km + a1 (k1)m + + ak1 m + ak = 0
with ai ai , then m a. Solving for the first term of the above equation gives
km = a1 (k1)m ak1 m ak .
Writing ai as a linear combination of i-fold products of characters in a, it follows

that for some 1 i k, we must have an equality
km = m1 mi (ki)m
with m1 , . . . , mi a, i.e., m1 , . . . , mi P. This implies

m = 1i (m1 + + mi ),
which lies in P since P is convex. Hence m a, proving that a is integrally closed.
In the other direction, a nonzero ideal a = h m1 , . . . , ms i C[ M] gives

the convex set P = Conv(m | m I) . In Exercise 7.1.12 you showed that
P = Conv(m1 , . . . , ms ) + . Hence P is a lattice polyhedron contained in I with
recession cone .
It remains to show that these maps are inverses of each other. One direction is
easy, for if P 7 a, then a 7 Conv(m | m PM), which equals P since P is a lattice
polyhedron. The other direction takes more thought. A nonzero integrally closed
ideal a gives P = Conv(m | m a). We need to show that m P M implies
m a. To prove this, first note that m P M implies m Conv(m1 , . . . , ms )
where m1 , . . . , ms a. By Caratheodorys theorem (see [281, Prop. 1.15]), we
538
can assume that m1 , . . . , ms are affinely independent. Since m, m1 , . . . , ms M, it

follows that
P
P
m = si=1 i mi , where i 0, si=1 i = 1, i Q.
Ps
Alfter
clearing
denominators,
we
get
km
=
i=1 i mi where k, i , . . . , s N and
Ps
m
i=1 i = k > 0. Then r = satisfies the equation
Q
r k si=1 ( mi )i = 0.
Q
Since si=1 ( mi )i ak and a is integrally closed, we must have r = m a.
Example 11.3.5. For the cone R3 of Example 11.1.12, Figure 7 shows the
polyhedron P obtained by pushing one of the facets of one unit to the right.
translate
this facet
one unit
to the
right
Figure 7. The polyhedron P and the dual cone
The corresponding ideal a is generated by the vertices of P (a rare occurrence, but

true here), so that
a = h e2 , e1 +e2 e3 i C[ M].
Looking at P, one sees that its normal fan is the fan 2 from Figure 3 in Example 11.1.12. Since P is normal (Exercise 11.3.4), it follows that X2 = Bla(U ).
We can also think about this from the point of view of the simplicialization
process described in the proof of Proposition 11.1.7. Since is not simplicial,
we can take (1) such that D is not Cartier, say = Cone(e2 ). The star
subdivision at e2 gives the fan 2 such that D is becomes Cartier on X2 . Then
one can check in that P is the polyhedron of D , so that
aOX2 = OX2 (D )
(Exercise 11.3.4).
539
L
Example 11.3.6. The maximal ideal m = m (M\{0}) C m C[ M] is
clearly integrally closed. The polyhedron associated to m is the convex hull Pm =
Conv( (M \ {0})).
When has dimension 2, this convex hull was denoted in 10.2, i.e.,
Pm = . In Proposition 10.2.17, we described the properties of the associated
projective toric morphism XPm U . Since lattice polyhedra are normal in dimension 2, Proposition 11.3.1 and Corollary 11.3.2 imply that this morphism is the
blowup of the ideal m, i.e., XPm = Blm(U ).
Example 11.3.7. Consider the graph G in Figure 8, where the vertices are labeled
with variables x, y, z. This is gives the edge ideal a = hxy, yz, zxi C[x, y, z] whose
generators correspond to the edges of the graph. Note that a is radical since it is
z
xy, yz, zx
Ideal a
y
Graph G
e3
e1
e2
Fan
x
Polyhedron P
Figure 8. Graph, ideal, polyhedron, and fan
generated by square-free monomials and hence is integrally closed since radical

ideals are automatically integrally closed (Exercise 11.3.5).
The ideal a gives the polyhedron P shown in Figure 8. Since P sits inside the
first octant in R3 , the shaded facets form the back of P from our point of view.
The vertices of P correspond to the generators of a, though in general, the ideal of
a polyhedron usually has more generators than just the characters of the vertices.
540
One can show that P is normal (Exercise 11.3.5), so that

X = XP = Bla(C3 ) = Proj(R[a]),
R = C[x, y, z],
by Proposition 11.3.1 and Corollary 11.3.2. Another proof of normality comes

from [250, Thm. 6.3], which implies that R[a] is normal since G is an odd cycle.
Hence P is normal Proposition 11.3.1.
When a C[ M] is generated by characters but not integrally closed, we

have the following description of its integral closure.
Proposition 11.3.8. Let a C[ M] be an ideal generated by characters and
set P = Conv(m | m a). Then:
L
(a) a = mPM C m , so P a in the correspondence of Proposition 11.3.4.
(b) aOXP = aOXP = OXP (DP ).
Proof. We leave the proof of part (a) as Exercise 11.3.6. For part (b), applying
Corollary 11.3.2 to P and a gives aOXP = OXP (DP ). Furthermore, the proof of the
corollary shows that
a C[v M] = v C[v M]
when v is a vertex of P. However, P = Conv(m | m a) implies that the vertices
come from a, i.e., v a. Hence a C[v M] = a C[v M], and part (b) follows
immediately.

Part (a) of the proposition is well-known when a C[x1 , . . . , xn ] is a monomial
ideal. See, for example, [266, Prop. 1.4.6]. Another nice result of monomial
commutative algebra is the monomial Briancon-Skoda theorem, which asserts that
if a C[ M] is an ideal generated by characters, then
an+1 a
where n = rank M and 1 is arbitrary. A proof can be found in [268, 4]. Monomial ideals are an active area of research in commutative algebra (see [274] and the
references therein).
Global Aspect. Given a projective toric resolution of singularities X X , it is
natural to ask if X can be represented as the blowup BlI (X ) for some torusinvariant sheaf of ideals I OX . When X is quasiprojective, this can be done
by adapting the technique used in the proof of [131, Thm. II.7.17]. In general, it
might not be possible to find a sheaf of ideals, but one can always find a sheaf of
fractional ideals that works. This is explained in [172, Ch. I, Thms. 10 and 11].
541
Log Resolutions. Suppose that we have a sheaf of ideals I OX , where X is

smooth. Then I OY is a line bundle on Y = BlI (X ), though Y may be rather
singular. A log resolution for I asks for a smooth version of this. Here is the
precise definition.
Definition 11.3.9. Let I OX be an ideal sheaf on a smooth variety X . Then a
projective birational morphism f : X X is a log resolution of I if X is smooth,
I OX = OX (D) for a divisor D on X , and Supp(D) Exc( f ) is a SNC divisor.
This type of log resolution is often called a principalization. In the toric case,
we have the following principalization result of Howald [147] (see also [268]).
Theorem 11.3.10. Let a C[x1 , . . . , xn ] be a monomial ideal. Then there is a toric
morphism : X Cn that is a log resolution of a.
Proof. Let P = Conv(m | m a) as in Proposition 11.3.8. This gives XP Cn
with aOXP = OXP (DP ). Since XP may be singular, we take a projective resolution
of singularities X XP . The composed map : XP Cn is projective, toric, and
birational, though its exceptional locus Exc() need not be a divisor. This is easily
fixed by adapting the strategy used in the proof of Theorem 11.2.2. Hence we may
assume that Exc() is a divisor.
Then let D be the pullback of DP via X XP . One easily checks that
aOX = OX (D), and Supp(D) Exc() is SNC since it is a union of torusinvariant prime divisors on a smooth toric variety.

A stronger principalization result, proved by Goward in [121], asserts that the
log resolution : X Cn can be chosen to be a composition of blowups with
smooth torus-invariant centers. This is analogous to embedded resolution of a
toric variety, which also uses blowups with smooth torus-invariant centers. See
Exercise 11.3.7 for an example.
Multiplier Ideals. Given a proper birational morphism f : X X where X and
X are smooth, we have the relative canonical divisor
KX /X = KX f KX ,
As noted in [186, 9.1.B], this divisor is supported on the exceptional locus Exc( f )
and hence satisfies
(11.3.3)
f OX (KX /X ) OX .
In the toric context, these assertions are easy to understand and are covered in
Exercise 11.3.8.
One consequence of (11.3.3) is that if E is any effective divisor on X , then
OX (KX /X E) OX (KX /X ),
542
so that f OX (KX /X E) f OX (KX /X ) OX is an ideal sheaf. When applied

to the log resolution of an ideal sheaf, this gives the following definition.
Definition 11.3.11. Let I OX be a nonzero ideal sheaf on a smooth variety X
and let f : X X be a log resolution of I such that I OX = OX (D) for an
effective divisor D on X . Given any rational number c > 0, the multiplier ideal
J (c I ) is the ideal sheaf defined by
J (c I ) = f OX (KX /X cD).
One can show that this definition is independent of the log resolution [186,
Thm. 9.2.18]. See [34] for an introduction to multiplier ideals and [186, Ch. 910]
for a thorough discussion and many applications.
In the toric context, Howald [147] showed how to compute the multiplier ideal
J (c a) of a monomial ideal a C[x1 , . . . , xn ].
Theorem 11.3.12. Let P = Conv(m | m a) be the polyhedron associated to a
monomial ideal a C[x1 , . . . , xn ]. Then, for any c > 0 in Q, we have
M
J (c a) =
C m.
m+e0 Int(cP)
were e0 = (1, . . . , 1) Zn .
Proof. We use the log resolution : X Cn constructed in Theorem 11.3.10.
Thus aOX = OX (D), where D is the pullback of P
DP on XP , and it follows
that the polyhedron of D is PD = P. If we write D = a D , a 0, then
P = {m MR | hm, u i a for all (1)},
hence
(11.3.4)
cP = {m MR | hm, u i ca for all (1)}.
Also note that divCn ( e0 ) = KCn , so that KCn = divX ( e0 ). Hence

X
KX /Cn cD = KX KCn cD =
(1 + he0 , u i ca )D .
The multiplier ideal J (c a) is given by the global sections of the sheaf of this
divisor. By Proposition 4.3.3, we obtain
m J (c a) hm, u i 1 he0 , u i + ca for all (1)
hm + e0 , u i 1 + ca
hm + e0 , u i > ca
for all (1)
for all (1).
By (11.3.4), we conclude that m J (c a) if and only if m + e0 lies in the interior

of cP. This completes the proof.
543
Theorem 11.3.12 shows that the characters in the multiplier ideal J (ca) come
from translates by e0 = (1, . . . , 1) of interior points of c times the polyhedron.
This result is also discussed in [186, Thm. 9.3.27].
Example 11.3.13. Consider the monomial ideal a = hx 4 , x 2 y, y 3 i C[x, y]. The
multiplier ideal J (1 a) = J (a) is given by
J (a) = hx 2 , xy, y 2 i.
This follows from Figure 9, which shows the polyhedron P (light shading), its
interior
points of P
P
translated interior
points of P
Figure 9. The polyhedron P, its interior lattice points, and their translates
interior lattice points (in the dark shading), and the result of translating these lattice
points by (1, 1) (bounded by the dashed line).
Example 11.3.14. For the ideal a = hxy, yz, zxi C[x, y, z] of Example 11.3.7, one
sees easily that (1, 1, 1) lies in the interior of P. It follows immediately that the
multiplier ideal J (a) is trivial.
Exercises for 11.3.

11.3.1. Prove the claim made in the proof of Proposition 11.3.1 that R[a] = C[NA ].
11.3.2. In Proposition 11.3.1, we showed that SP = C[C(P) (M Z)] is the integral
closure of R[a] = C[NA ]. Prove that Proj(SP ) Proj(R[a]) is the normalization map.
Hint: Look at the proof of Theorem 7.1.13. A character m a gives an element mt of
degree 1 in R[a] and SP , which in turn gives open sets
U = Spec(R[a]( m t) ) and U = Spec((SP )( m t) )
of Proj(R[a]) and Proj(SP ) respectively. Show that these open sets cover Proj(R[a]) and
Proj(SP ) and that U is the normalization of U.
11.3.3. For the blowup of the origin 0 Cn , find an explicit ideal a R = C[x1 , . . . , xn ] such
that Bl0 (Cn ) = Proj(R[a]). Then do the same for the blowup of the coordinate subspace
{0} Cnr Cn .
11.3.4. Fill in the details omitted in the discussion of Example 11.3.5.
544
(a) Prove that an ideal in C[x1 , . . . , xn ] generated by square-free monomials is radical.

(b) Prove that every radical ideal is integrally closed.
(c) Prove that the polytope P pictured in Figure 8 of Example 11.3.7 is normal.
11.3.7. Let a = hx 3 , y 2 i C[x, y].
(a) Construct a log resolution : X C2 of a using the method described in the proof
of Theorem 11.3.10.
(b) Interpret the fan X constructed in part (a) as a sequence of three star subdivisions,
each of which blows up a fixed point of the torus action.
(c) Let Ei be the proper transform in X exceptional locus of the ith blowup. Show that
aOX = OX (2E1 3E2 6E3).
This exercise is taken from [34].
11.3.8. Let : X X be the toric morphism coming from a smooth refinement of a

smooth fan .
(a) Prove that KX KX is supported on the exceptional locus Exc() and is 0.
(b) Let D be an effective divisor supported on Exc(). Prove that OX (D) OX . Hint:
Reduce to the affine case.
11.3.9. Compute the multiplier ideal J (a) of the following monomial ideals:
(a) a = hx 8 , y6 i C[x, y] (see [147, Ex. 2]).
(b) a = hx7 , x 3 y, xy 2 , y6 i C[x, y] (see [186, Ex. 9.3.28]).
11.3.10. Let a = hx1a1 , . . . , xnan i C[x1 , . . . , xn ], where a1 , . . . , an are positive integers. Prove
Pn
that J (c a) is trivial if and only if c < i=1 a1i (see [147, Ex. 3]).
11.3.11. Consider the monomial ideal a = hxy, yz, zxi C[x, y, z] and polyhedron P from
Figure 8 in Example 11.3.7. The toric morphism XP C3 satisfies aOXP = OXP (DP ). To
get a log resolution of a as in Theorem 11.3.10, we need a smooth refinement of the normal
fan of P. Figure 10 shows and two smooth refinements 1 and 2 .
e3
e3
e1
e2
e1
e3
e2
e1
e2
2
Figure 10. The fan and smooth refinements 1 and 2
(a) Explain how 1 and 2 can be obtained from by a series of star subdivisions. Hence
X1 C3 and X2 C3 are birational projective toric morphisms.
545
(b) Show that X1 C3 is a composition of blowups at smooth torus-invariant centers.
(c) Show that X2 C3 is not a composition of blowups at smooth torus-invariant centers.
11.3.12. In this exercise and the next we follow [33] and explore the toric case of two
variants of multiplier ideals. Let U = Spec(C[ M]) be an affine toric variety. Fix an
effective torus-invariant Q-divisor D and an ideal a C[ M] generated by characters.
Also assume that KU + D is Q-Cartier. Thus there are m0 M and r > 0 in Z with
div( m0 ) = r(KU + D). Also let P be the polyhedron of a.
(a) Prove that there is a toric morphism : X U which is a log resolution for both D
and a. Write aOX = OX (A), where A is an effective divisor.
(b) Given c > 0 in Q, prove that we have a natural inclusion
OX (KX (KU + D) + cA) OU .
This gives a multiplier ideal J (D, c a) C[ M].
(c) Adapt the proof of Theorem 11.3.12 to show that

M
J (D, c a) =
C m.
1
m+ r m0 Int(cP)
Part (c) is [33, Thm. 1]. Note that Theorem 11.3.12 is a special case of this result. Earlier,
we defined the multiplier ideal J (c a) for Cn . This definition extends to Q-Gorenstein
affine toric varieties. However, when U fails to be Q-Gorenstein, then we use the log
version, since KU + D could be Q-Cartier for some Q-divisor D. See [186, 9.3.G].
11.3.13. To define a multiplier ideal that works for any affine toric variety U , we proceed
as follows. Let a C[ M] be generated by characters and let : X U be a log
resolution such that aOX = OX (A), A 0. Following [33], the multiplier module is
OX (KX cA) U , c > 0 in Q.
By Proposition 8.2.9, we can regard U as an ideal of C[ M] generated by characters

coming from lattice points in the interior of . Then the multiplier module gives an ideal
of C[ M] denoted J (c a). Prove that this ideal is given by
M
C m,
J (c a) =
mInt(cP)
where as usual P is the polyhedron of a. This is [33, Thm. 2].

11.3.14. Let a R = C[x1 , . . . , xn ] be a monomial ideal. In the text, we used a log resolution
to define the multiplier ideal J (c a). This exercise will present a different construction of
J (c a). Let S be the integral closure of the Rees algebra R[a]. This gives the projective
morphism : X = Proj(S) Cn . Prove that
J (c a) = OX (KX KCn cD),
where aOX = OX (D). Hint: Use Proposition 11.3.1. (For experts, this works because X
has rational singularities.)
11.3.15. This exercise is for readers who know
P about symbolic powers of ideal sheaves
(see, for example, [186, Def. 9.3.4]). Let D = i Di be a sum of distinct prime divisors on
a normal variety X. Then OX (D) OX is a sheaf of radical ideals. For an integer n 1,
show that OX (nD) equals the nth symbolic power OX (D)hni . Hint: Exercise 4.0.14.
546
11.4. Toric Singularities

We next discuss the singularities of normal toric varieties. This is an active area of
research, and our treatment will omit many important topics.
General Properties of Toric Singularities. Although a normal toric variety need
not be smooth, it is at least Cohen-Macaulay. Furthermore, its singularities are of
the following special type.
Definition 11.4.1. A normal variety X has rational singularities if for every resolution of singularities f : X X , we have
R p f OX = 0 for all p > 0.
One can show that if the above vanishing holds for one resolution of singularities, then it holds for all [179, Thm. 5.10]. Hence, in the toric case, we can use a
toric resolution of singularities.
Theorem 11.4.2. A normal toric variety X has rational singularities.
Proof. Let : X X be a toric resolution of singularities. In particular, is
proper, so that R p OX = 0 for p > 0 by Theorem 9.2.5. Then we are done by
the above remark.

The reader should consult [179, Sec. 5.1] for a careful discussion of rational
singularities.
The Simplicial Case. We next give several characterizations of the singularities of
a simplicial toric variety. The first is topological, based on the following idea. If X
is a smooth irreducible variety of dimension n, then any x X has a neighborhood
x U in the classical topology that is homeomorphic to an open ball in Cn R2n .
By excision and the long exact sequence for relative cohomology, one obtains
(
0 p 6= 2n
p
p
H (X , X \ {x}, Z) H (U ,U \ {x}, Z)
Z p = 2n.
When we weaken the coefficients from Z to Q, we get the following definition.
Definition 11.4.3. An irreducible variety X of dimension n is rationally smooth if
for every x X , we have
(
0 p 6= 2n
H p (X , X \ {x}, Q)
Q p = 2n.
Smooth varieties are obviously rationally smooth. Here is a example that will
be useful below.
Example 11.4.4. If G GL(n, C) is a finite group, then Cn /G is rationally smooth
by [49, Prop. A.1].
547
Before giving our second characterization, we need some terminology. The

structure sheaf OX of a variety X is a sheaf in the Zariski topology. When we
switch to the classical topology, the corresponding sheaf is the sheaf of analytic
functions on X , denoted OXan (see [131, App. B] for a brief description). Then
any classical open subset U X gives the analytic variety (U , OUan ) = (U , OXan |U ),
and varieties X1 and X2 are locally analytically equivalent at p1 X1 and p2 X2 if
there are classical neighborhoods p1 U1 X1 and p2 U2 X2 such that U1 U2
as analytic varieties, where the isomorphism take p1 to p2 .
Definition 11.4.5. Let X be an irreducible variety of dimension n.
(a) A point p X is a finite quotient singularity if there is a finite subgroup G
GL(n, C) such that p X is locally analytically equivalent to 0 Cn /G.
(b) X is an orbifold or is quasismooth or has finite quotient singularities if every

point of X is a finite quotient singularity.
(c) X has abelian finite quotient singularities if every point of X is a finite quotient
singularity such that the finite group G in part (a) is abelian.
Here is a lemma whose proof is almost obvious.
Lemma 11.4.6. Let G GL(n, C) be a finite subgroup.
(a) Cn /G is an orbifold.
(b) If G is abelian, then Cn /G has abelian finite quotient singularities.

Proof. Definition 11.4.5 guarantees that 0 Cn /G is a finite quotient singularity.
But what about the other points of Cn /G? For v Cn , let Gv = {g G | g v = v}
be its isotropy subgroup. We will show that 0 Cn /Gv is locally analytically
equivalent to v Cn /G.
Note that w 7 w + v is equivariant with respect to Gv and hence induces a

local analytic equivalence from 0 Cn /Gv to v Cn /Gv . To complete the proof,
we need to find a local analytic equivalence from v Cn /Gv to v Cn /G.
S
Pick coset representatives so that G = i gi Gv . The points gi v are distinct in
Cn /Gv , so there is a classical
neighborhood v U such that the gi U are disjoint,
T
and replacing U with gG g U , we can assume that U is Gv -invariant. Then V =
S
i gi U is a G-invariant neighborhood of v, and one sees easily that U /Gv V /G.
This gives the desired local analytic equivalence.

On Cn /G, the coordinates x1 , . . . , xn on Cn become local coordinates on the
quotient. The intuition is that anything G-invariant in the xi descends to Cn /G. For
b p n ) (Cn , p n )G , so that
b p n comes
example, one can show that (Cn /G,
C
C /G
C /G
from G-invariant p-forms in the local coordinates.
548
Example 11.4.7. If G SL(n, C), then dx1 dxn is G-invariant. It follows

b n n is trivial, i.e., Cn /G has trivial canonical divisor. This is why
that Cn /G
C /G
finite subgroups of SL(n, C) are so important.
In the analytic category, the existence of local coordinates enables one to do

analysis on orbifolds. See [68, App. A.3] for further discussion and references.
We now characterize simplicial toric varieties in terms of their singularities.
Theorem 11.4.8. Given a normal toric variety X , the following are equivalent:
(a) is simplicial.
(b) X has abelian finite quotient singularities.
(c) X is an orbifold.
(d) X is rationally smooth.
Proof. First assume that is simplicial and take . If N is the sublattice of
N Zn generated by the minimal generators of , then U,N Ck , where k =
dim . Then Proposition 3.3.11 implies that there is a finite abelian group G with
U = U,N (U,N (C )nk )/G (Ck (C )nk )/G Cn /G.
Since G is a subgroup of (C )n GL(n, C), Lemma 11.4.6 implies that U and
hence X have abelian finite quotient singularities.
This proves (a) (b). Then (b) (c) is obvious, and (c) (d) follows
from Example 11.4.4. It remains to prove (d) (a). For this, we can assume that
NR Rn is a cone such that U is rationally smooth. A face gives the
following objects:
The sublattice N = Span( ) N and quotient lattice N( ) = N/N .
The exact sequence 1 TN TN TN( ) 1 of tori.
The orbit closure V ( ) = O( ) U of dimension n dim .
Since O( ) TN( ) by Theorem 3.2.6, it follows easily that V ( ) is the fixed point
set of the action of TN on U (Exercise 11.4.1). In the standard notation for fixedT
point sets, this means UN = V ( ). Hence
(11.4.1)
dim UN = n dim .
Now let 1 , . . . , r be the facets of . Then each Ti = TNi is a subtorus of

T = TN of codimension 1 (Exercise 11.4.1). Furthermore, for any p V (),
(11.4.2)
dim p U dim p UT = n (n dim ) = dim
dim p UTi dim p UT = (n dim i ) (n dim ) = 1
549
by (11.4.1). Since U is rationally smooth at p, a result of Brion [49, Thm., p. 130]

implies that
X
dim p U dim p UT =
dim p UT dim p UT ,
T
where the sum is over all subtori T T of codimension 1. Using (11.4.2), we

obtain
r
X

dim p UTi dim p UT = r.
dim
i=1
For any cone, the number of facets is at least its dimension, with equality if and
only if the cone is simplicial (Exercise 11.4.1). Hence the above inequality implies
that is simplicial, and the proof is complete.

We note one other important characterization of simplicial toric varieties: a
normal toric variety is simplicial if and only if it is Q-factorial, meaning that every
Weil divisor is Q-Cartier. We proved this in Proposition 4.2.7.
Terminal and Canonical Singularities. Given a normal variety X , recall from
11.2 that a resolution of singularities f : X X is crepant if X is Q-Gorenstein
and KX = f KX . Having a crepant resolution is rare, so for most varieties, we
have KX 6= f KX no matter which resolution we use. Measuring the difference
between these divisorstheir discrepancyleads to some interesting classes of
singularities.
In general, let KX be a canonical divisor on a normal Q-Gorenstein variety X .
Given a proper birational morphism f : X X with X smooth, one can find a
canonical divisor KX on X such that
X
ai Ei ,
(11.4.3)
KX = f KX +
i
where ai Q and the Ei are the irreducible divisors lying in the exceptional locus
Exc( f ). This is proved in [194, Rem. 4-1-2].
Definition 11.4.9. Let X be a normal Q-Gorenstein variety.
(a) X has terminal singularities if there is a proper birational f : X X with X
smooth, such that the coefficients ai in (11.4.3) satisfy ai > 0 for all i.
(b) X has canonical singularities if there is a proper birational f : X X with X
smooth, such that the coefficients ai in (11.4.3) satisfy ai 0 for all i.
If X has terminal (resp. canonical) singularities, then the inequalities ai > 0

(resp. ai 0) hold for all proper birational morphisms f : X X with X smooth
[194, Rem. 4-1-2 and 4-2-2]. Note that f : X X may fail to be a resolution of
singularities since (for instance) some smooth points of X may get blown up by f .
However, it is often useful to be able to work with these more general morphisms.
550
If X is not smooth and has terminal singularities, then X cannot have a SNC
crepant resolution. Terminal singularities are very important in the minimal model
program, since a minimal model is a Q-factorial projective variety that has only
terminal singularities. Canonical singularities are also relevant, since (roughly
speaking) canonical singularities are the those that appear on canonical models
of varieties of general type. See [194, Ch. 4] for a general discussion of terminal
and canonical singularities. We will use terminal singularities in Chapter 15 when
we discuss the toric minimal model program.
Let us apply these ideas in the toric context. Given a normal Q-Gorenstein
toric variety X , we can find a toric resolution of singularities : X X . Here,
we have the toric canonical divisors KX and KX , and the relation (11.4.3) can be
described explicitly as follows.
Lemma 11.4.10. Let be the support function of the Q-Cartier divisor KX and
let : X X be the toric morphism coming from a refinement of . Then:
X
((u ) 1)D .
KX = KX +
(1)\(1)
Proof. The key observation is that P

KX and its pullback KX have the same
support function. Thus KX = (u )D . Then the desired equation

follows immediately since (u ) = 1 for all (1).

Here is an easy consequence of this lemma.
Proposition 11.4.11. Let X be a normal Gorenstein toric variety. Then X has
canonical singularities.
Proof. The support function of KX is integral with respect to N since X is
Gorenstein. Now let be a smooth refinement of . Given u (1), we have:
(u ) Z since u || N.
(u ) > 0 since u lies in a cone and > 0 on since takes the value
1 on every minimal generator of .
It follows that (u ) 1, and we are done by Lemma 11.4.10.
Our next result shows that we can determine when an affine toric variety U
has terminal or canonical singularities by studying the lattice points of the polytope
= Conv(0, u | (1)).
Proposition 11.4.12. Let U and be as above.
(a) The following are equivalent:
(i) U is Q-Gorenstein
(ii) There is m MQ such that hm, u i = 1 for all (1).
(iii) has a unique facet not containing the origin.
551
(b) If U is Q-Gorenstein, then:

(i) U has terminal singularities the only lattice points of are given
by its vertices.
(ii) U has canonical singularities the only nonzero lattice points of
lie in the facet not containing the origin.
Proof. We begin with part (a). On an affine toric variety, the canonical divisor is
Q-Cartier if and only some multiple is the divisor of a character. This easily gives
the equivalence of (i) and (ii). Furthermore, given m MQ as in (ii), lies on one
side of the hyperplane defined by hm, ui = 1. The face cut out by this hyperplane
contains all of the u and hence is the unique facet of not containing the origin,
proving (iii). We leave (iii) (ii) as Exercise 11.4.2.
For part (b), first note that the support function of KU is (u) = hm, ui for
m MQ is as in the previous paragraph. Then
= {u | hm, ui 1}.
Now take any primitive vector v N different from the u . By taking a smooth
refinement of the star subdivision given by v, we get a proper birational morphism : X U such that v is the minimal generator of = Cone(v) (1).
By Lemma 11.4.10, the coefficient of D in KX KU is
(11.4.4)
(v) 1 = hm, vi 1.
If U has terminal singularities, then (11.4.4) is positive. Hence hm, vi > 1, so

that v
/ . It follows that the only lattice points of are its vertices. Conversely,
suppose that satisfies this condition and let be a smooth refinement of .
Given any u (1) \ (1), we have u
/ , so that (11.4.4) is positive. Hence
U has terminal singularities.
The proof for canonical singularities is similar. The only difference is that the
coefficient (11.4.4) is allowed to be zero, which happens only for lattice points
lying on the facet of not containing the origin. You will supply the details in
Exercise 11.4.2.

Example 11.4.13. Consider the lattice N = {(a, b, c) Z3 | a + b + c 0 mod 2}
and cone = Cone(e1 , e2 , e3 ) from Example 11.2.7. There are two ways to see that
U has terminal singularities:
The resolution : X U constructed in Example 11.2.7 satisfies KX =
KU + 12 D0 . Since 21 > 0 and D0 is the exceptional locus, U has terminal
singularities by Definition 11.4.9.
The minimal generators of with respect to N are 2e1 , 2e2 , 2e3 , so that =
Conv(0, 2e1 , 2e2 , 2e3 ). One easily checks that N consists only of vertices,
so that U has terminal singularities by Proposition 11.4.12.
In the surface case, terminal and canonical singularities are easy to understand.
552
Theorem 11.4.14. Let X be a toric surface. Then:

(a) X has terminal singularities X is smooth.
(b) X has canonical singularities X is Gorenstein X has at worst rational

double points.
Proof. Part (a) follows because a 2-dimensional cone = Cone(u1 , u2 ) such that
N = Conv(0, u1 , u2 ) N = {0, u1 , u2 } is smooth (Exercise 8.3.4).
For part (b), Proposition 11.2.8 implies that being Gorenstein is equivalent
to having rational double points, and if X is Gorenstein, then it has canonical
singularities by Proposition 11.4.11. To complete the proof, we will show that an
affine toric surface U with canonical singularities has a rational double point. Let
d, k be the parameters of , so that has minimal generators e2 , de1 ke2 . Then
= {u | hm, ui 1},
m=
k+1
d e1 + e2 .
If k < d 1, then hm, e1 i = k+1

d < 1, so that e1 is an interior point of . This
is impossible since U has canonical singularities. The only remaining case is
k = d 1, which gives a rational double point by Example 10.1.5.

Reduction to Canonical and Terminal Singularities. We next discuss how any
toric singularity can be made first canonical and then terminal. We begin with the
canonical case.
Proposition 11.4.15. A strongly convex cone NR has a canonically determined
refinement can such that Xcan has canonical singularities and the induced toric
morphism : Xcan U is projective.
Proof. Let = Conv( (N \ {0})). In Exercise 11.4.3 you will show that
is a lattice polyhedron with as recession cone. You will also show that taking
cones over bounded faces of gives a fan can in NR such that the morphism
: Xcan U is projective.
Take can and suppose is the cone over the bounded face F of .
Then F is the unique facet of not containing the origin, so U is Q-Gorenstein
by Proposition 11.4.12. Since F is a facet of = Conv( (N \ {0})), the only
nonzero lattice points of lie in F, so that U has canonical singularities by
Proposition 11.4.12.

Example 11.4.16. When is 2-dimensional, the refinement can has an elegant
description as follows. Recall from Proposition 8.2.9 that the canonical sheaf U
comes from the ideal
M
C m.
(U , U ) =
mInt( )M
This ideal is integrally closed with associated polyhedron

P = Conv(Int() N)
553
as in Proposition 11.3.4. The reasoning in Example 11.3.6 shows that XP U is

the blowup of this ideal. We claim that
can is the normal fan of P ,
i.e., Xcan = XP . This will imply that XP has at worst rational double points by
Proposition 11.4.14 and Theorem 11.4.15.
In dimension 2, a refinement of is determined uniquely by its ray generators.
Thus it suffices to prove that
(11.4.5)
the vertices of are the ray generators of the normal fan of P .
We show this as follows. Recall that by Proposition 10.2.17, the ray generators of
the normal fan of Pm = = Conv( (M \ {0})) are almost the vertices of
. The difference arises from slight complications that can occur at the edges of
, as can be seen by looking at Theorem 10.2.12 and Proposition 10.2.17.
Let us compare Pm and P . Since the latter uses only interior lattice points,
these polyhedra differ along the edges of . To see how this affects their normal
fans, consider Figure 11, which shows what can happen at an edge . Let a1
be the lattice length (number of lattice points 1) of the bounded edge of Pm that
a1 > 1
a1 = 1
Figure 11. Two examples of Pm (outlined in bold) and P (shaded)
touches . As you can see, Pm and P have the same inner normal vectors near
when a1 > 1, but P has one less inner normal when a1 = 1. The same thing happens at the other edge of . Then (11.4.5) follows by comparing Theorem 10.2.12
and Proposition 10.2.17 to Figure 11 (Exercise 11.4.4).
We should also mention that P is the polyhedron denoted in [218, p. 28].

Before we can take a canonical singularity and make it terminal, we need to
extend the definition of crepant, which was originally defined only for resolutions
of singularities. Given normal varieties X ,Y where Y is Q-Gorenstein, a morphism
f : X Y is crepant if KX = f KY . Then we have the following result.
554
Proposition 11.4.17. If X has canonical singularities, then we can find a simplicial refinement of such that X has terminal singularities and the induced
toric morphism : X X is projective and crepant. Furthermore, X is Gorenstein if and only if X is Gorenstein.
Proof. Suppose that X has canonical singularities. To measure how far X is
from being terminal, set
X
|( N) \ {vertices}|.
t() =
max
By Proposition 11.4.12, X has terminal singularities if and only if t() = 0.

If t() > 0, pick a non-vertex N for some max . Then the star
subdivision = () satisfies the following:
X has canonical singularities with t( ) = t() 1.
X is Gorenstein if and only if X is.
X X is crepant.
You will verify these properties in Exercise 11.4.5. Figure 12 shows what happens
for the affine toric variety of a cone R3 with t() = 2. Here, is one of the
Figure 12. The polyedron and the star subdivision of coming from
two non-vertex lattice points of and the star subdivision has t( ) = 1.

By induction on t(), a sequence of these star subdivisions gives a refinement
of such that X has terminal singularities, and X is Gorenstein if and only
if X is. If is simplicial, then we are done. If not, then the simplicial refinement
constructed in Proposition 11.1.7 is easily seen to have the desired properties.
Terminal Singularities. Propositions 11.4.15 and 11.4.17 allow us to focus on

toric varieties with terminal singularities. We begin with the affine case.
555
Gorenstein affine toric varieties with terminal singularities can be classified in

terms of empty lattice polytopes, which are lattice polytopes whose only lattice
points are their vertices. Here is the precise result.
Proposition 11.4.18. Classifying Gorenstein affine toric varieties with terminal
singularities that come from n-dimensional cones NR Rn is equivalent to
classifying (n 1)-dimensional empty lattice polytopes.
Proof. First assume that U is Gorenstein with terminal singularities. Then there
is m M such that hm, u i = 1 for all (1). Extend m to a basis of M and
consider the dual basis of NR Rn . Since dim = n, it follows that = Cone(F),
where F Rn1 {1} is the facet of not containing the origin. Then F is a
lattice polytope of dimension n 1 and is empty since U has terminal singularities
(Proposition 11.4.12).
Conversely, given an empty lattice polytope P Rn1 of dimension n 1, we
get the n-dimensional cone = Cone(P {1}) Rn . In Exercise 11.4.6 you will
show that U is Gorenstein and has terminal singularities since P is empty.

Proposition 11.4.18 has some nice consequences in the 3-dimensional case.
Proposition 11.4.19. Let X be a 3-dimensional Gorenstein toric variety. Then:
(a) If X is simplicial, then X has terminal singularities X is smooth.
(b) X has a resolution of singularities : X X such that is projective and

crepant.
Proof. For part (a), assume that X is simplicial with terminal singularities and
take (3). Since 2-dimensional empty lattice simplices are lattice equivalent
to the standard 2-simplex (Exercise 8.3.4), Proposition 11.4.18 implies that is
smooth. When (2), Proposition 11.4.12 implies that is a 2-dimensional
empty lattice simplex, which as already noted is smooth. Hence X is smooth.
For part (b), recall that X has canonical singularities by Proposition 11.4.11.
Then Proposition 11.4.17 gives a simplicial refinement such that : X X is
projective and crepant, X has terminal singularities, and X is Gorenstein since
X is. Then X is smooth by part (a). One can also check that the refinement
constructed in Proposition 11.4.17 does not affect the smooth cones of (Exercise 11.4.7). Hence is the desired resolution of singularities.

Corollary 11.4.20. The toric variety of a 3-dimensional reflexive polytope has a
projective crepant resolution.
Proof. This follows immediately from Proposition 11.4.19 since the toric variety
of a reflexive polytope is a Gorenstein Fano variety by Theorem 8.3.4.

In the 3-dimensional case, there are also results about terminal singularities
that do not assume that the variety is Gorenstein. To state our result, we define the
556
index of a Q-Gorenstein variety X to be the smallest positive integer r such that

rKX is Cartier. The X is Gorenstein if and only if it is Q-Gorenstein of index 1.
Then we can classify all 3-dimensional cones that give terminal singularities
as follows.
Theorem 11.4.21. Let NR R3 be a 3-dimensional cone whose affine toric
variety U has terminal singularities.
(a) If is simplicial, then N has a basis u1 , u2 , u3 such that
= Cone(u1 , u2 , u1 + pu2 + qu3 ),
where 0 p < q are relatively prime. Furthermore, U is Q-Gorenstein of
index q, and
U C3 /q ,
where q = { C | q = 1} acts on C3 via (x, y, z) = ( 1 x, p y, z).
(b) If is not simplicial, then N has a basis u1 , u2 , u3 such that

= Cone(u1 , u2 , u1 + u3 , u2 + u3 ).
Furthermore, U is Gorenstein and U V(xy zw) C4 .

Proof. For part (a), first note that is a 3-dimensional empty lattice simplex.
By the terminal lemma [218, 1.6], N has a basis such that the vertices of
are 0, u1 , u2 , u1 + pu2 + qu3 , where 0 p < q are relatively prime. This gives the
desired formula for , and the quotient construction from Chapter 5 easily gives
the representation U C3 /q in the statement of the theorem (Exercise 11.4.8).
To compute the index, let m = (1, 1, p/q) relative to the dual basis of M. Then
hm, u1 i = hm, u2 i = hm, u3 i = 1.
Since gcd(p, q) = 1, qm is the smallest positive multiple of m that is a lattice point,

hence the index is q. See [103, Thm. 2.2] for part (b).

In [218, 1.6], Oda discusses the terminal lemma, with references. In the
literature, the action of q on C3 is often written differently. Given the action
(x, y, z) = ( 1 x, p y, z) as in the theorem, let 0 a < q be the multiplicative
inverse of p modulo q. Using the automorphism of q given by 7 a and
changing coordinates in C3 , the action becomes (x, y, z) = ( a x, a y, z). See
[218, 1.6] for references to the classification of 3-dimensional cones whose toric
varieties have canonical singularities. Also, [13] and the references therein discuss
the classification of empty lattice simplices in dimensions 3 and 4.
Part (b) of Theorem 11.4.21 is especially nice, since it shows that one of our
favorite examples, V(xy zw) C4 , is even more interesting than we realized.
The singularities of a normal variety have codimension 2. In Exercise 11.4.9

you will prove that the following result, which tells us that the codimension is
higher when the singularities are terminal.
557
Proposition 11.4.22. Assume that X has only terminal singularities.

(a) The singular locus (X )sing of X has codimension codim(X )sing 3.
(b) If in addition X is Gorenstein and simplicial, then codim(X )sing 4.
At the end of 8.3, we noted that in dimension 3, there are 4319 isomorphism
classes of Fano toric varieties, corresponding to the 4319 classes of 3-dimensional
reflexive polytopes. They all have canonical singularities by Proposition 11.4.11
since Fano varieties are Gorenstein. But only 100 of these have terminal singularities by [165], and as noted in 8.3, only 18 are smooth. This indicates the special
nature of terminal singularities. See also [166].
Log Singularities. Besides terminal and canonical singularities, there are other
classes of singularities that are important for the minimal model program. We
will consider two, log canonical and Kawamata log terminal. The latter is usually
abbreviated klt. These singularities are defined for a pair (X , D), where X is a
normal variety and D is a Q-divisor with coefficients in the interval [0, 1]. See
[179, p. 98] or [194, Ch. 11] for a nice discussion of why pairs are useful. A more
recent reference is [30]. Pairs are sometimes called log pairs.
P
Let (X , D), D = i di Di , be a pair. Also fix a proper birational morphism
S
f : X X such that X is smooth and Exc( f ) i f 1 (Di ) is a SNC divisor. Note
that f may fail be a log resolution of D. We define the birational transform D of
P
D to be the divisor on X defined by D = i di Di , where Di = f 1 (Di U ) and
U X is the largest open subset where f 1 is a morphism (see also 15.4).
Let (X , D) and f : X X be as above, and assume that KX + D is Q-Cartier.
By [179, Sec. 2.3], we can pick a canonical divisor KX such that
X
ai Ei ,
(11.4.6)
KX + D = f (KX + D) +
i
where ai Q and the Ei are the irreducible divisors lying in the exceptional locus.
This is the log version of (11.4.3).
Definition
11.4.23. Let (X , D) be a pair such that KX + D is Q-Cartier, and write
P
D = i di Di with di [0, 1] Q.
(a) (X , D) has log canonical singularities if there is f : X X as above such that
the coefficients ai in (11.4.6) satisfy ai 1 for all i.
(b) (X , D) has klt singularities if di [0, 1) for all i and there is f : X X as

above such that the coefficients ai in (11.4.6) satisfy ai > 1 for all i.
One can show that this definition is independent of the morphism f : X X .

In practice, one often says (X , D) is log canonical instead of (X , D) has log
canonical singularities, and (X , D) is klt has a similar meaning.
In the toric case, we consider pairs (X , D) where D is torus-invariant.
558
Proposition 11.4.24. Let X be a normal toric variety, and let D =

where d [0, 1] Q. If KX + D is Q-Cartier, then:
(a) (X , D) is log canonical.
d D ,
(b) If in addition d [0, 1) for all (1), then (X , D) is klt.

Proof. Let be the support function of KX + D, so that (u ) = 1 d for all
(1). Let : X X be a toric log resolution. For the rest of the proof,
D will denote the divisor
on X corresponding to (1). Thus the birational
P
transform of D is D = (1) d D .
To simplify notation, let (1) = A B, where A = (1) and B = (1) \ A.
Since is the support function of (KX + D), we obtain
X
X
X
(KX + D) =
(u )D =
(1 + d )D
(u )D
AB
D +
AB
KX
X
A
X
d D +
(1 (u ))D
X
+
(1 (u ))D .
B
Hence
(11.4.7)
KX + D = (KX + D) +
X
((u ) 1)D .
B
This is the log version of Lemma 11.4.10.

To analyze the coefficients, take P
B = (1) \ (1). Since refines , we
have u for some . Thus u = (1) u , where 0. Hence
X
X
(11.4.8)
(u ) =
(u ) =
(1 d ).
(1)
(1)
We have d 1 by assumption, so that 1 d 0. It follows that (u ) 0, and

thus the coefficients (u ) 1 in (11.4.7) are all 1. This proves that (X , D) is
log canonical.
Now assume in addition that d < 1 for all (1). Then 1 d > 0 in
(11.4.8), and since the are not all 0 (u 6= 0), we conclude that (u ) > 0 for all
B. This shows that the coefficients (u ) 1 in (11.4.7) are all > 1. Hence
(X , D) is klt, as claimed.

Here is an easy corollary that you will prove in Exercise 11.4.10.
Corollary 11.4.25. Let X be a normal toric variety. Then:
P
(a) (X , D ) is log canonical.
(b) If X is Q-Gorenstein, then (X , 0) is klt.
559
Here is an example based on an observation of Chen and Shokurov [62].

Example 11.4.26. Recall from 8.3 that a projective normal variety X is Gorenstein Fano if KX is ample. The log version says that a normal projective variety X
is of Fano type if there is Q-divisor D such that (KX + D) is Q-ample (meaning
that some positive multiple is Cartier and ample) and (X , D) is klt. Let us show
that every projective toric variety X is of Fano type.
Since X is projective, there is a lattice polytope P that gives an ample divisor
DP on X . Multiplying P by a suitably large integer, we can assume that P has
an interior lattice point m, and translating by m (which gives aP
linearly equivalent
divisor), we can assume that 0 is an interior point. Since DP = a D and
P = {m MR | hm, u i a },
it follows that a > 0 for all . Then set

X
D=
(1 a )D ,
where Q is positive and satisfies a < 1 for all . Then the coefficients of D
lie in [0, 1) for all .
By construction, D satisfies KX + D = DP , so that KX + D is Q-Cartier.
Then (X , D) is klt by Proposition 11.4.24. It follows that X is of Fano type since
(KX + D) = DP is Q-ample.
In the minimal model program, many results assume that (X , D) is a klt pair.
See [30, Sec. 4.3] for some examples. In Chapter 15 we will see that klt pairs arise
naturally in the toric minimal model program.
Exercises for 11.4.
11.4.1. In this exercise you will supply some details omitted from the proof of (d) (a)
in Theorem 11.4.8.
(a) Prove that V ( ) is the unique fixed point of the action of TN on U .
(b) Explain why TN is a subtorus of TN of codimension 1 when is a facet of .
(c) Prove that is simplicial if and only if its dimension equals the number of its facets.
11.4.2. This exercise is concerned with the proof of Proposition 11.4.12.
(a) Prove (iii) (ii) in part (a).
(b) Prove the characterization of canonical singularities given in part (b).
11.4.3. As in Proposition 11.4.15, let = Conv( N \ {0}).
(a) Prove that is a lattice polyhedron with as recession cone.

(b) Prove that taking cones over bounded faces of gives a fan can refining .
(c) Prove that can U is projective. Hint: Suppose can comes from a facet of
. If the facet is defined by hm, i = a, then consider the support function on
= |can | whose restriction to is given by 1a m.
560
11.4.4. Complete the proof of (11.4.5) sketched in Example 11.4.16.

11.4.5. Prove the properties of = () stated in the three bullets in the proof of Proposition 11.4.17.
11.4.6. In the proof of Proposition 11.4.18, we saw that an elementary lattice polytope
P gives a cone of one dimension higher. Prove that U is Gorenstein with terminal
singularities. Hint: P is the facet of lying at height 1.
11.4.7. Let : X X be the terminalization constructed in Proposition 11.4.17.
Prove that all terminal simplices of lie in and explain why this implies that is an
isomorphism above the Q-factorial terminal locus of X .
11.4.8. Complete the proof of Theorem 11.4.21 by using the methods of Chapter 5 to
describe the quotient representation of U . Hint: Lemma 5.1.1 will be useful.
11.4.9. Prove Proposition 11.4.22. Hint: Study the cones of dimension 3 using
the methods of the proof of Proposition 11.4.19.
11.4.10. Prove Corollary 11.4.25.
11.4.11. Here is a partial converse to Theorem 11.4.17. Assume that X is Q-Gorenstein.
Prove that if has a refinement such that X has terminal singularities and X X
is crepant, then X has canonical singularities.
11.4.12. Explain how Propositions 11.4.15 and 11.4.17 give the minimal resolution of
singularities of a toric surface X . Draw some pictures to illustrate what is happening.
11.4.13. Let X be a Q-Gorenstein toric variety with terminal singularities.
(a) Give an example where X has a crepant resolution of singularities. Hint: Try one of
our favorite examples.
(b) If X is simplicial but not smooth, then prove that all resolutions of singularities are
not crepant.
(c) If X is not smooth, then prove that all SNC resolutions of singularities are not crepant.
Chapter 12
The Topology of
Toric Varieties
In this chapter, we will study some topological invariants of a toric variety X ,

always for the classical topology. The fundamental group 1 (X ) is studied in
12.1, and 12.2 addresses the moment map and alternative topological models of X . Methods for computing the singular cohomology groups H k (X , Z) are
discussed in L
12.3. A complete description of the cup product ring structure on
H (X , Q) = k H k (X , Q) for simplicial complete X is developed in 12.4 using

the equivariant cohomology of X for the action of TN . The chapter concludes with
12.5, where we consider the Chow ring and intersection cohomology.
Our goal is to understand the information these invariants provide about the
topology and geometry of toric varieties and their applications to polytopes. We
will freely use the definitions and various properties of homotopy, homology, and
cohomology groups, referring to [135], [210], and [255] as our primary references.
12.1. The Fundamental Group

Topology of Tori. The most basic toric varieties are the tori TN , and their topology is correspondingly simple to understand. A choice of basis in N determines a
homeomorphism and isomorphism of groups TN (C )n . Let S1 denote the unit
circle in C . The usual polar coordinate system in each factor gives a homeomorphism and isomorphism of multiplicative groups
(C )n (R>0 )n (S1 )n .
In the following, we will use the notation SN for the compact real torus in TN .
561
Chapter 12. The Topology of Toric Varieties
562
Proposition 12.1.1. Let N be a lattice of rank n. Then:

(a) SN is a deformation retract of TN (C )n .
(b) There is an isomorphism
1 (TN ) Zn .
Proof. Since R>0 is contractible to the point 1 R>0 , we easily construct the
deformation retraction. Part (b) follows from the standard facts 1 (S1 ) Z and
1 (X Y ) 1 (X ) 1 (Y ).

There is also a more intrinsic way to understand this isomorphism. Recall that
each u N defines a one-parameter subgroup u : C TN . Restricting u to
S1 C gives a closed path in TN , whose homotopy class represents an element of
1 (TN ). In Exercise 12.1.1, you will show that the resulting map N 1 (TN ) is an
isomorphism. Hence we will usually use this intrinsic form of Proposition 12.1.1:
(12.1.1)
1 (TN ) N.
To determine the fundamental group of other toric varieties, we will use the
fact that the algebraic condition of normality has a very important, though quite
nontrivial, topological consequence.
Topological Implications of Normality. To see the pattern, we consider two toric
examples we have encountered before.
Example 12.1.2. In Example 3.A.2, we studied the nonnormal affine toric surface
YA = V(y 2 x 2 z) C3
with A = {e1 , e1 + e2 , 2e2 } Z2 . Note that the z-axis is clearly contained in YA .

In Exercise 12.1.2 you will show that the z-axis is precisely the singular locus of
YA . Letting p = (0, 0, z) with z 6= 0, you will also show that if U is any sufficiently
Figure 1. The surface V(y 2 x 2 z)
small open neighborhood of p in YA , then U \ (U Sing(YA )) has two connected

components, as suggested by the picture of the real points of the surface in Figure 1.
These are called the branches of YA at p.
563
Example 12.1.3. Consider the quadric Cb2 = V(xz y 2 ) C3 , the normal affine
toric surface from the cone = Cone(e2 , 2e1 e2 ). The only singularity of this
surface is the point p = (0, 0, 0). From our standard parametrization, Cb2 is the
image of the map
: C2 C3
(t1 ,t2 ) 7 (t1 ,t1 t2 ,t1 t22 ).

The singular point p is the image of the line V(t1 ) in C2 . If we remove that line,
then C2 \ V(t1 ) is still connected, so
Cb2 \ {p} = (C2 \ V(t1 ))
is the continuous image of a connected set, hence connected. The same will be true
if we intersect Cb2 \ {p} with any connected open neighborhood U of p in C3 .
These examples illustrate a general phenomenon. While a nonnormal variety

can have more than one branch at a singular point, at each point x of a normal
variety X , singular or nonsingular, X is locally irreducible or unibranch in the
following sense.
Proposition 12.1.4. Let X be a normal variety and take x X . Then there is a

basis {V | A} of open neighborhoods of x in X such that V \ (V Sing(X ))
is connected for all .

This is a topological version of the fundamental result known as Zariskis main
theorem. We refer to [208, III.9] for algebraic versions of the result and their
relation to Proposition 12.1.4.
Comparing 1 (X ) and 1 (TN ). Let be a fan in NR and consider the normal
toric variety X and the inclusion i : TN X . There is a general fact that applies
to the induced map i : 1 (TN ) 1 (X ) in this situation (see [109]).
Theorem 12.1.5. Let X be a normal variety and let i : U X be the inclusion of
an open subvariety. Then the induced map i : 1 (U ) 1 (X ) is surjective.
The proof of the theorem will consist of the following two lemmas. We recall
that any suitably nice topological space X has a universal covering space, a simplyconnected space Xe with an (unramified) covering map p : Xe X . The universal
cover of a variety over C is not always a variety, but it inherits the structure of a
complex analytic space from X , which will be sufficient for our purposes.
Lemma 12.1.6. If X is a normal variety and i : U X is the inclusion of an open
subvariety U , then U X Xe is path-connected.
564
Proof. Let U = X \ Y , where Y is Zariski closed in X . Then U is connected and

path-connected in the classical topology. By Proposition 12.1.4 and the construction of the universal covering space, Xe is also locally irreducible. Since it is also
connected, it is irreducible as an analytic space. Hence
Xe \ p1 (Y ) = p1 (X \Y ) = p1 (U )
is also connected and path-connected in the classical topology. The analytic space
U X Xe = {(u, xe) | p(e
x) = u}
is essentially the graph of p restricted to p1 (U ). Since p is continuous, this is also

connected and path-connected.

Lemma 12.1.7. Let X , Z be topological spaces and assume X has a universal covering space p : Xe X . Let f : Z X be continuous. If Z X Xe is path-connected,
then f : 1 (Z) 1 (X ) is a surjection.
Proof. We must show that for each homotopy class [] 1 (X ), there exists a
closed curve in Z such that f ([]) = [], or equivalently that f and are
homotopic in X . Pick a base point x0 on X with x0 = f (z0 ). Let xe0 p1 (x0 ) in
e and lift to e
X,
in Xe starting from the point xe0 . The final point of e
will be some
xe1 p1 (x0 ). Hence both (z0 , xe0 ) and (z0 , xe1 ) are points in
x)}.
Z X Xe = {(z, x) Z Xe | f (z) = p(e
By hypothesis, we can find a path : [0, 1] Z X Xe with (0) = (z0 , xe0 ) and
(1) = (z0 , xe1 ). Let p1 , p2 be the projections from Z Xe to the two factors. The
projection p2 is homotopic to
e since Xe is simply-connected. Hence p p2
is homotopic to . Moreover p1 = is a loop in Z such that f = p p2 .
This shows that f is surjective.

Because of Theorem 12.1.5, to determine 1 (X ) we must determine the kernel
of the homomorphism i : 1 (TN ) 1 (X ), and then
1 (X ) 1 (TN )/ker(i ) N/ker(i ).
In particular, the fundamental group of a normal toric variety is always a finitelygenerated abelian group. Here is one case where the kernel is easy to see.
Proposition 12.1.8. Let = Cone(u ) be a 1-dimensional cone with u the primitive ray generator in N. Then ker(i : 1 (TN ) 1 (U )) = N = Zu .
Proof. We use (12.1.1) and view i as a map from N to 1 (U ). Applying Proposition 3.2.2 to the cone , we see that for all u N,
(12.1.2)
u lim u (z) exists in U .

z0
565
Since u , the limit point limz0 u (z) = is in U . Restrict z to S1 C and

take t [0, 1]. Then the map
(t, z) : [0, 1] S1 X
(
u ((1 t)z) 0 t < 1
(t, z) 7
t=1
is continuous. Moreover gives a homotopy from a loop representing the homotopy class on TN corresponding to u N to a constant path at the distinguished
point U , where is defined on page 116. This shows N ker(i ). The
opposite inclusion follows since if ker(i ) contained any u N not in N , then
rank 1 (U ) = rank(N/ker(i ))
would be n 2 or smaller. But U
= (C )n1 C, so 1 (U ) Zn1 .
The Affine Case. The idea behind the homotopy constructed in the proof of the
last proposition can be generalized to prove the following fact.
Proposition 12.1.9. Let be a strongly convex rational polyhedral cone in NR .
(a) The torus orbit O() is a TN -equivariant deformation retract of U .
(b) There are isomorphisms
1 (U ) 1 (TN() ) N(),
where N() = N/N as usual.
Proof. Part (b) follows immediately from part (a) because O() is equal to the
torus TN() by Lemma 3.2.5. To prove part (a), view the points x U as semigroup
homomorphisms x : S C, where S = M. Choose any lattice point u in
Relint() N and define the map
: U [0, 1] U
by
(
m (u (1 t)) x(m)
(x,t)(m) =
(m) x(m)
if 0 t < 1
if t = 1,
where is the distinguished point corresponding to .

For all (x,t), it is not hard to see that (x,t) is a semigroup homomorphism
from S to C, hence a point of U (Exercise 12.1.3). Moreover, (x,t) is continuous
in x and t.
We have m (u (1 t)) = (1 t)hm,ui , so (x, 0) is the identity map on U .
Moreover if m M, m (u (1 t)) = 1 for all 0 t < 1. From Lemma 3.2.5,
O() = { : S C | (m) 6= 0 m M}.
Chasing the definitions, one sees that (x,t) = x for all t when x O().
566
Finally,
(
1
lim(1 t)hm,ui =
t1
0
if m S = M
otherwise,
which implies that (x, 1) O() for all x. Combining all of this, we see that is
a deformation retraction. You will show TN -equivariance in Exercise 12.1.3.

The General Case. For a general fan we will use the last proposition and the
Van Kampen theorem to complete the computation of 1 (X ).
Theorem 12.1.10. Let be a fan in NR and let N be the sublattice of N generated
by || N. Then 1 (X ) N/N .
Proof. We will prove the theorem by induction on the number of cones in . If
consists of a single cone {0}, then the claim follows from (12.1.1).
Now assume the result has been proved for all fans with k 1 cones or fewer
and let contain k cones. Pick any cone of maximal dimension in , and let
= \ {}. Then
X = X U ,
with X U = X for the fan consisting of the proper faces of . Since X ,

U , and X are all path connected, we can apply the Van Kampen theorem [135,
Thm. 1.20] to determine 1 (X ). From the diagram of inclusions
i1
u:
uuu
X I
I
i2
II
$
X H j
HH1
H
h
#
/ X
;
vv
vv j2
we obtain the corresponding diagram of fundamental groups

(X )
1
s9
ss
s
ss
i1
1 (X )
KK
KK
K
i2 K%
JJ j1
JJ
JJ
%
/ 1 (X ).
t9
tt
t
tt j2
1 (U )
The Van Kampen theorem identifies 1 (X ) as the free product of 1 (X ) and

1 (U ), modulo the normal subgroup generated by all elements of the form
( j1 i1 ) ([])(( j2 i2 ) ([]))1 ,
567
where [] 1 (X ). Our induction hypothesis gives isomorphisms

1 (X ) N/N
1 (X ) N/N
1 (U ) N/N .
By using presentations of these groups in terms of generators and relations, it is

easy to see that
1 (X ) N/(N + N ) = N/N .
You will complete the details in Exercise 12.1.4.
Theorem 12.1.10 implies that X is simply connected if and only if the support
of contains a basis for N. This is the case, for instance, if contains an ndimensional cone. It follows that a simply connected toric variety has no torus
factors. The converse of this assertion is not true, though, as the following example
shows.
Example 12.1.11. Let N have rank 2, and let be the fan consisting of the cones
{0}, Cone(e1 ), Cone(e1 + de2 ).
The corresponding toric variety X is the complement of the origin in the rational
normal cone Cbd . Clearly, N = Ze1 + dZe2 and 1 (X ) N/N Z/dZ.
Exercise 12.1.6 shows that for any finitely generated abelian group G, there
exists a normal toric variety X with 1 (X ) G. Some information on the higher
homotopy group 2 (X ) will be obtained later in Exercise 12.3.10.
Exercises for 12.1.
12.1.1. Let [] be the homotopy class of a closed path . Show that the map
N 1 (TN )
is an isomorphism of groups.
u 7 [u |S1 ]
12.1.2. In this exercise you will verify the claims made in Example 12.1.2 about the surface
YA = V(y 2 x 2 z) C3 , where A = {e1 , e1 + e2 , 2e2 }.
(a) Show that the singular locus of YA is the z-axis.
(b) Show that if p is any point of YA with z 6= 0 and V is a sufficiently small open neighborhood of p in YA , then V \ (V Sing(YA )) has two connected components.
12.1.3. In this exercise, you will complete the details of the proof of Proposition 12.1.9.
(a) Show that for all x U and t [0, 1], (x,t) : S C is a semigroup homomorphism,
hence a point of U .
(b) Verify the claim in the proof of Proposition 12.1.9 that (x,t) is continuous in x and t.
(c) Show that the deformation retraction is equivariant for the TN -action.
568
12.1.4. Complete the details of the argument using the Van Kampen theorem in the proof
of Theorem 12.1.10.
12.1.5. Let be the fan in NR R2 with cones
{0}, Cone(e1 + e2 ), Cone(e1 e2 ), Cone(e1 + e2 ), Cone(e1 e2 ).
Show that 1 (X ) Z/2Z.
12.1.6. Show that given any finitely generated abelian group G, there exists a normal toric
variety whose fundamental group is isomorphic to G.
12.1.7. Prove that 1 (X ) is finite if and only if X has no torus factors.
12.2. The Moment Map

The compact real torus SN (S1 )n is a subgroup of TN and hence also acts on the
toric variety X . We will show first that the quotient space X /SN can be identified
with a certain subset (X )0 of X .
The Nonnegative Part of a Toric Variety. Recall the interpretation of points of
an affine toric variety U as semigroup homomorphisms : S C. We can
define subsets of U by placing restrictions on the image of . For instance, the
nonnegative part of U is defined formally as
(U )0 = HomZ (S , R0 ).
The absolute value map z 7 |z| on C gives a retraction of C onto R0 . So there
is a corresponding retraction U (U )0 defined by mapping 7 ||. If X
is the toric variety of a fan , it is easy to check that the (U )0 for glue
together to form a closed subset (X )0 of X in the classical topology. Moreover
the retraction maps glue properly to give a retraction
X (X )0 .
By considering semigroup homomorphisms with images in R or R>0 , we can
define real or positive real points of toric varieties as well. The real points of toric
varieties are discussed for instance in [254].
If : X Y is a toric morphism, then the above discussion implies that
restricts to a map
0 : X0 Y0 .
Moreover, 0 fits in a commutative diagram

X
/Y
(12.2.1)

X0

/ Y0
569
where the vertical arrows are the above retractions (Exercise 12.2.1). This implies
that a character m of TN (C )n maps (TN )0 (R>0 )n to (C )0 = R>0 .
We can also apply (12.2.1) to the quotient construction of toric varieties from
Chapter 5. By Proposition 5.1.9, the quotient map
: C(1) \ Z() X .
is a toric morphism, so that 0 is defined.

Proposition 12.2.1. Let X be a normal toric variety without torus factors. Then
0 : (C(1) \ Z())0 (X )0
is surjective.
Proof. Theorem 5.1.11 implies that is a good categorical quotient. Hence is
surjective by Theorem 5.0.6. Surjectivity of 0 follows from the commutativity
of the diagram (12.2.1).

Here is one way to use Proposition 12.2.1.
Example 12.2.2. Consider X P1 P1 for the fan in NR R2 given in Example 3.1.12. By Example 5.1.8, we have
P1 P1 = U /(C )2 ,
where U = C4 \ (({(0, 0)} C 2 ) (C2 {(0, 0)})) and (C )2 acts via

(, ) (a, b, c, d) = (a, b, c, d).
By Proposition 12.2.1, a point in (P1 P1 )0 comes from (a, b, c, d) U R40 .

Since a, b 0 cannot both vanish, we can rescale by a positive number so that
a + b = 1. Doing the same for c, d, we see that (P1 P1 )0 can be written as
(P1 P1 )0 = {(a, b, c, d) R4 : a, b, c, d 0 and a + b = c + d = 1},
which is a square in a two-dimensional affine subspace in R4 . Note that this is

the same (combinatorially, at least) as a plane polygon P whose normal fan P
coincides with the original fan . We will see shortly that this is no coincidence.
In homogeneous coordinates, the retraction P1 P1 (P1 P1 )0 is given by

|a|
|b|
|c|
|d|
(a, b, c, d) 7
,
,
,
.
|a| + |b| |a| + |b| |c| + |d| |c| + |d|
Note the structure of the fibers of this map. Over the four corners of the square
(that is (a, b) = (1, 0) or (0, 1) and similarly for (c, d)), the fiber consists of a single
point. On an edge but not at a vertex, the fiber consists of a copy of S1 . Finally at
an interior point of the square, the fiber consists of a copy of S1 S1 = T 2 . Thus
P1 P1 has a stratification by fiber bundles with compact real torus fibers.
570
The Quotient of X by SN . Now we consider the quotient of X by the compact

real torus SN . To prepare for our next result, note that the isomorphism C
R>0 S1 used in 12.1, composed with the real logarithm map R>0 R on the
first factor, yields an isomorphism C R S1 . In Exercise 12.2.2, you will show
that SN can be identified with HomZ (M, S1 ) HomZ (M, C ), where S1 is the unit
circle in C . Putting all of this together, we obtain isomorphisms
(12.2.2)
TN = HomZ (M, C ) HomZ (M, R) HomZ (M, S1 ) NR SN .
The structure of the quotient X /SN is as follows.

Proposition 12.2.3. Let X be a normal toric variety.
(a) The retraction X (X )0 induces a homeomorphism X /SN
= (X )0 .
(b) For each cone in , the fiber of X (X )0 over a point in O()0 can
be identified with SN() , a compact real torus of dimension n dim .
Proof. For part (a), consider how SN acts on each torus orbit O(). Recall from
Lemma 3.2.5 that O() HomZ ( M, C ) TN() , where N() = N/N is the
dual lattice of M. Then (12.2.2) implies that
O() N()R SN() .
Using the real logarithm again, O() retracts to
O()0 = HomZ ( M, R>0) HomZ ( M, R) = N()R .
However, SN acts on O() via the compact real torus SN() in TN() . As a result,
O()/SN
= N()R
= O()0 . The assertion for X follows. Part (b) follows by
similar reasoning.

Example 12.2.4. The toric variety P1 is homeomorphic to the real 2-sphere S2 .
The torus SN in this case is S1 , a circle acting on P1 fixing two points shown as
the north and south poles in Figure 2. Every other point has a circle as orbit. The
quotient space is homeomorphic to a closed interval.
Figure 2. The quotient P1 /S1

= (P1 )0
571
Note also that the map P1 P1 (P1 P1 )0 from Example 12.2.2 can be
visualized as the Cartesian product of two copies of this picture.
The Moment Map. Recall that in Example 12.2.2 above we saw that (P1 P1 )0
was combinatorially the same as a polygon with normal fan equal to the standard
fan for P1 P1 . We now turn to a general connection between (XP )0 and the
polytope P.
Let P be a full dimensional lattice polytope in MR . Corresponding to P, we
have the normal fan P and the toric variety XP = XP . Replacing P by a multiple
P if necessary, we will assume that P M = {m0 , . . . , ms } defines an embedding
: XP P s
x 7 ( m0 (x), . . . , ms (x)).
Following the online version of [254], we define the algebraic moment map by
f : XP MR
(12.2.3)
x 7 X
1
m
mPM
| (x)| mPM
| m (x)|m.
The symplectic moment map corresponding to the action of the compact real torus
SN (as defined in symplectic geometry) is the closely related map
: XP MR
(12.2.4)
x 7 X
mPM
1
m
| (x)|
mPM
| m (x)|2 m.
Note that f and are invariant under the action of SN TN . The behavior we saw
in Example 12.2.2 is a special case of the next result.
Theorem 12.2.5. Let P be a full dimensional lattice polytope in MR . The restricted
algebraic moment map
f = f |(XP )0 : (XP )0 MR ,
is a homeomorphism from (XP )0 to P.
Proof. The characters m for m P M map x (XP )0 to nonnegative real
numbers, so that | m (x)| = m (x). Rescaling as in Example 12.2.2 gives (x) =
(a0 , . . . , as ) P s , where a0 + + as = 1 and ai 0 for all i. This implies that
f (x) = a0 m0 + + as ms .
It is clear from (12.2.3) that f is continuous. We will show that f is bijective and leave the verification that f 1 is continuous as an exercise. The proof
of bijectivity will be accomplished by showing that f maps each torus orbit in XP
572
bijectively to the relative interior of the corresponding face of P. We will give the
details for TN (C )n in XP and show that (TN )0 maps bijectively to the interior
of P. The result for the other orbits will then follow by similar reasoning.
First we establish surjectivity. Fixing an isomorphism M Zn , write mi =
(mi1 , . . . , min ) where mi j Z for all i, j. Given v in the interior of P, there are
in general many different ways to write v as a convex linear combination of the
mi . If a = (a0 , . . . , as ) is one vector of coefficients in such a combination and
a = (a0 , . . . , as ) is a second, then
v = a0 m0 + + as ms = a0 m0 + + as ms
with ai , ai 0 and a0 + + as = a0 + + as = 1. It follows that the bi = ai ai

satisfy the linear equations b0 m0 + + bs ms = 0 and b0 + + bs = 0. With the
notation
WP = {(b0 , . . . , bs ) Rs+1 | b0 m0 + + bs ms = 0 and b0 + + bs = 0},
P
the vectors a = (a0 , . . . , as ) satisfying v = si=0 ai mi are precisely the elements of
(a +WP ) Rs+1
0 .
Our proof will show, in fact, that v can be written as v = a0 m0 + + as ms ,

where ai = mi (x) where > 0 and x = (x1 , . . . , xn ) TN with xi > 0 for all i.
This will show the surjectivity of f as a map from the positive real points in TN in
XP to the interior of P. The particular representation of v we want will come from
minimizing a certain function on (a +WP ) Rs+1
0 .
First note that g(x) = xlog (x) x can be defined for all x 0 in R since
LHopitals rule implies that limx0+ g(x) = 0. Using g, we define
G : (R0 )s+1 R
(x0 , . . . , xs ) 7 g(x0 ) + + g(xs ).
Since g (x) = 1x , the Hessian (second derivative) of G is positive definite at any

point (x0 , . . . , xs ) with xi > 0 for all i. In other words, G is concave up at all such
points. You will show in Exercise 12.2.7 that G, restricted to (a +WP ) Rs+1
0 , has
a unique critical point a = (
a0 , . . . , as ), necessarily a minimum.
We claim that ai > 0 for all i. If not, then I = {i | ai = 0} is nonempty. In
Exercise 12.2.7, you will show that there exists b WP satisfying bi > 0 for all
i I. Let (t) = a + tb be the line through a with direction vector b, and let i (t)
be the components. Then 0 < i (t) < 1 for all i and all t in some sufficiently small
open interval (0, ). It follows that
s
X
d
lim
bi log(
ai + tbi ) = .
(G )(t) = lim
t0+ dt
t0+
i=0
But this contradicts the concavity properties of G. Hence a must have all nonzero
components.
573
For an arbitrary b in WP , let (t) = a + bt and consider the restriction of G to

this line. Since ai > 0 for all i, we have 0 < i (t) < 1 when t lies in some symmetric
interval about 0. Then (G )(t) has a local minimum at t = 0, so by the chain rule,
s
0=
X
d
bi log (
ai ).
(G )|t=0 =
dt
i=0
Since this is true for all b WP , linear algebra implies that the overdetermined
system of linear equations
(12.2.5)
log (
a0 ) = m01 y1 + + m0n yn + c
..
.
log(
as ) = ms1 y1 + + msn yn + c
in n + 1 variables (y1 , . . . , yn , c) has a real solution. Let yi = log(xi ) and c = log ()

mi
min
i1
where xi , are real and
Exponentiating, ai = xm
1 xn = (x) for
Ppositive.
s
all i = 0, . . . , s. Since i=0 ai = 1, the constant must be
1
.
mi
i=0 (x)
= Ps
This completes the proof of surjectivity of the restricted algebraic moment map.
The injectivity of
f : (XP )0 MR
is now a consequence of the constructions already made. Suppose v is in the interior of P and v = f (x) for x a positive real point in the TN in XP . Then the
min and
i1
definition of f gives v = f (x) = a0 m0 + + as ms with ai = xm
1 xn
Ps
= ( i=0 mi (x))1 . Writing yi = log (xi ) and c = log(), we have a system
log (a0 ) = m01 y1 + + m0n yn + c
..
.
log(as ) = ms1 y1 + + msn yn + c
of the same form as (12.2.5). Since the system is consistent, we must have
(12.2.6)
0=
s
X
bi log (ai )
i=0
for all b WP . However, by the same computations done before, the sum on the
right of (12.2.6) is the derivative at t = 0 of (G )(t), where (t) = a + bt. By the
concavity of G, this implies that a = a, the unique critical point of G on a+WP . But
then, since there are n linearly independent vectors among the mi , it follows that the
min
m
i1
ai = xm
1 xn are uniquely determined. Since we assume that for m P M
define an embedding of XP , this shows that x is also uniquely determined.
574
This proof basically follows the presentation in [93, VII.7]; [105] gives a different argument. The result of Theorem 12.2.5 remains true if we replace the algebraic
moment map by the symplectic moment map from (12.2.4) (Exercise 12.2.8).
Topological Models of Toric Varieties. Combining Theorems 12.2.3 and 12.2.5
gives a way to understand the underlying topological space of the projective toric
variety XP of a polytope P, similar to what we observed in Example 12.2.2. Indeed,
it is not difficult to see that there is an SN -equivariant homeomorphism
(12.2.7)
XP
= (SN P)/ ,
where two points (s1 , x1 ) and (s2 , x2 ) are identified if x1 = x2 , and s1 and s2 are
congruent modulo the subtorus SN of SN for the cone P corresponding to the
minimal face of P containing x1 (Exercise 12.2.9).
An analogous topological model can be given for any complete normal toric
variety X using the unit ball B n . Namely, a complete fan in NR determines
a spherical complex C on the unit sphere S n in NR by intersecting each cone
with the sphere. Then each determines a spherical dual
b in B n as follows.
By a process analogous to that described in Exercise 11.1.10, one constructs a
barycentric subdivision of C . If = {0} then
b = B n . Otherwise,
b is the union of
all spherical simplices in the barycentric subdivision whose vertices are barycenters
of S n with . Then
(12.2.8)
X
= (SN B n )/ ,
where two points (s1 , x1 ) and (s2 , x2 ) are identified if and only if x1 = x2 and s1 is
congruent to s2 modulo SN for the unique such that x1 is in the relative interior
of
b. See [162], where this construction of a space homeomorphic to X is attributed to MacPherson, and [97]. These references also discuss generalizations for
toric varieties associated to noncomplete fans. Some recent work in toric topology
essentially takes this topological construction as the definition of a toric variety.
The article [59] gives a nice overview of toric topology.
Symplectic Geometry and Toric Varieties. We conclude with a brief discussion

of the relations between moment maps, symplectic geometry, and toric varieties,
without proofs. In symplectic geometry, one studies Hamiltonian actions of a compact connected Lie group GR on a symplectic manifold X . Such an action has a
symplectic moment map
: X gR ,
where gR is the Lie algebra of GR . For example, the action of SN on the toric
variety XP gives the symplectic moment map (12.2.4). This follows because the
the dual of the Lie algebra of SN is NR = MR .
Another example comes from quotient construction of a projective simplicial
toric variety
X (C(1) \ Z())/G
575
from Chapter 5. The maximal compact subgroup of G = HomZ (Cl(X ), C ) is

GR = HomZ (Cl(X ), S1 ) and the Lie algebra of GR can be identified with Cl(X )R .
Here, the symplectic moment map
: C(1) Cl(X )R
factors as
The map is given by
C(1) R(1) Cl(X )R .
1
(z1 , . . . , zr ) = (|z1 |2 , . . . , |zr |2 ),
2
and the map comes from the exact sequence
0 MR R(1) Cl(X )R 0
obtained by tensoring (5.1.1) by R.
Let [D] be the class of an ample divisor on a complete simplicial toric variety
(1)
\ Z() and results of Guillemin imply that the natural
X . Then 1
([D]) C
map
(12.2.9)
1
([D])/GR X
is a diffeomorphism. The divisor D determines a symplectic structure on X . The

natural symplectic structure from C(1) descends to the quotient 1
([D])/GR and
the diffeomorphism in (12.2.9) preserves the cohomology classes of the symplectic
forms. Discussions of these results and references to detailed proofs can be found
in [66, 4].
Example 12.2.6. Let X = P n , and recall that Cl(P n )R R. The symplectic moment map above is given by
n+1
(z1 , . . . , zn+1 ) =
1X 2
|zi | .
2
i=1
The class of an ample [D] corresponds to a positive real value, so that 1

([D]) is
diffeomorphic to S2n+1 . The group GR S1 in this case, and the diffeomorphism
(12.2.9) is
S2n+1 /S1 P n .
We obtain the identification of P n as the base space of the Hopf fibration with total
space S2n+1 and fiber S1 [135, p. 337].
For symplectic geometers, (12.2.9) shows that toric varieties can be defined by
a construction known as symplectic reduction.
Results along the lines of Theorem 12.2.5 were originally developed in a more
general setting. A compact connected real 2n-dimensional symplectic manifold
(M, ) is said to be toric if it has an effective Hamiltonian (S1 )n action. Results
576
of Atiyah, Guillemin and Sternberg show that the image of the symplectic moment
map of such a manifold is a polytope P in Rn . Moreover, the polytopes that appear
have been characterized by Delzant in [81]. They are polytopes having n edges
incident at each vertex p, of the form p + tui for some ui Zn forming a Z-basis
of Zn . The vertices need not be lattice points, but if they are then P is a smooth
polytope. In any case, M is diffeomorphic to a smooth projective toric variety.
Exercises for 12.2.
12.2.1. This exercise supplies some details for the construction of the nonnegative part of
a toric variety and the behavior of a toric morphism on the nonnegative part.
(a) Let X be a normal toric variety. Show that the (U )0 for all in glue together to
form a closed subset (X )0 of X (in the classical topology).
(b) Show that the retraction maps U (U )0 for glue together properly to give
a retraction X (X )0 .
(c) Let : X Y be a toric morphism. Show that restricts to a map 0 : X0 Y0
commuting with the retractions on X, Y as in (12.2.1).
12.2.2. Let M, N be dual lattices Show that the compact torus SN TN is identified with
HomZ (M, S1 ) HomZ (M, C ).
12.2.3. A generating set m1 , . . . , ms of S gives an embedding U Cs . Prove that

(U )0 = U Rs0 . Hint: See the proof of Proposition 1.3.1.
12.2.4. Check the claims about the fibers of the retraction P1 P1 (P1 P1 )0 made in
Example 12.2.2.
12.2.5. Determine the image of the algebraic moment maps for each of the following toric
varieties directly from the associated polytope P without using Theorem 12.2.5.
(a) The projective plane P 2 , with P = Conv(0, e1 , e2 ). Generalize to P n .
(b) The rational normal scroll XP for P = Conv(0, 3e1 , e2 , e1 e2 ) (isomorphic to the
Hirzebruch surface H2 ).
12.2.6. Complete the proof that f in Theorem 12.2.5 is a homeomorphism by showing that
f 1 is continuous.
12.2.7.
This exercise concerns some details in the proof of Theorem 12.2.5. Let G =
Ps
s+1
x
log(x
defined in that proof. Recall that G has positive
i
i ) xi be the function on R
i=0
definite Hessian (second derivative) at all points with all positive coordinates.
(a) Show that if W is a translate of a linear subspace of Rs+1 , and W0 is the subset
of W consisting of points with all coordinates nonnegative, then G|W0 has a unique
minimum when W0 6= . Hint: Argue by contradiction. Let a and b be two minima
and consider G restricted to the line containing a and b.
(b) Show that if G attains its minimum on (a + WP ) Rs+1

such that I = {i | ai = 0}
0 at a
is nonempty, then there exists b WP with bi > 0 for all i I. Hint: Recall that the
point v is assumed to lie in the interior of P.
12.2.8. Show that the result of Theorem 12.2.5 is still true if we replace the algebraic
moment map by the symplectic moment map from (12.2.4).
12.2.9. Use Theorems 12.2.5 and 12.2.3 to construct a homeomorphism (12.2.7).
12.3. Singular Cohomology of Toric Varieties
577

In this section, we will study the singular cohomology groups of a toric variety
X . We first describe these groups using the singular cohomology of the affine
toric varieties U for max . We then give a different description that uses
the singular cohomology of the torus orbits O() for . In both cases, the
machinery of spectral sequences (see Appendix C) will establish the connection.
The Picard Group of a Toric Variety and H 2 . We begin with a lovely application
of spectral sequences that connects the Picard group of a toric variety to H 2 (X , Z).
Let be a fan in NR Rn . In Chapter 4, we constructed the map
M
M
M/M(i )
M/M(i j )
i max
i< j
(mi )i 7 (mi m j )i< j ,
where M() = M. Proposition 4.2.9 provides a natural isomorphism

L
L
CDivTN (X ) ker
i M/M(i )
i< j M/M(i j ) ,
where CDivTN (X ) is the group of torus-invariant Cartier divisors on X . Then

Theorem 4.2.1 relates this to Pic(X ) via the exact sequence
M CDivTN (X ) Pic(X ) 0.
The map M CDivTN (X ) is given by m 7 div( m ). Our goal is to relate Pic(X )

to the topological object H 2 (X , Z).
We begin by computing the singular cohomology of an affine toric variety.
Proposition 12.3.1. Let be a cone in NR . Then
H (U , Z) H (TN() , Z)
M().
Proof. By Proposition 12.1.9, TN() is a deformation retract of U , and this implies

the first isomorphism. The cohomology of a torus was given in Example 9.0.12 and
the second isomorphism follows from the duality of N N() and M() M.
The Picard group relates to the singular cohomology of X as follows.
Theorem 12.3.2. Let be a fan in NR Rn with all maximal cones n-dimensional.
Then
Pic(X ) H 2 (X , Z).
Proof. Consider the open cover of X given by
U = {U }max = {U }(n) .
As noted in 9.0, for the spaces we are considering, singular cohomology with
coefficients in Z is the sheaf cohomology of the constant sheaf (in the classical
578
topology) given by Z. Hence the spectral sequence of the covering U (see (9.0.10)
and Theorem C.2.2) becomes
M
H q (Ui0 Ui p , Z) H p+q (X , Z).
E1p,q =
=(i0 ,...,i p )Ip
p,q
Our strategy will be to compute E2 for small values of p, q.

The first observation is that
E1p,0 =
(i0 ,...,i p )Ip
since Ui0 Ui p = U , = i0 i p , is connected for all . Hence we

get the Koszul complex (9.1.15) with M = Z, minus its first term. Thus
(
0 p>0
E2p,0 =
Z p = 0.
Since the maximal cones in are n-dimensional, Proposition 12.3.1 implies
M
E10,q =
H q (Ui , Z) = 0, for all q > 0.
i (n)
It follows that
E20,q = 0 for all q > 0.
Thus the E2 sheet of the spectral sequence (with differentials shown only in the
p = 1 column) is:
0
0
Z
E21,2 QQQ E22,2

QQQ
QQQd21,2
E23,2
QQQ
QQQ
(
2,1
E23,1
E21,1 RRR E2
RRR
1,1
RRRd2
RRR
RRR
R)
0
0
0.
Then Er2,0 and Er0,2 must be zero for all r 2. Moreover, the differentials into and
out of Er1,1 for all r 2 must be zero and as a result,
1,1
H 2 (X , Z).
E21,1 = E
However, we also know that E21,1 is the kernel of the map

M
M
E11,1 =
H 1 (Ui U j , Z) E12,1 =
H 1 (Ui U j Uk , Z)
i< j
i< j<k
since E10,1 = 0. By Proposition 12.3.1,

H 1 (Ui U j , Z) M(i j ),
H 1 (Ui U j Uk , Z) M(i j k ).
579
Hence we get the commutative diagram:

0
0
0
M
/ H 2 (X , Z)
/
i< j M(i j )
/
i M RR
RRR
RRR
R
L R(
L
i< j M
i< j M/M(i j )
i< j<k M(i j k )
i< j<k M
i< j<k M/M(i j k )
The last two columns are obviously exact, and the first row is exact by the analysis
of H 2 (X , Z) given above. The second row is also exact since it is the Koszul
complex from (9.1.15). Then an easy diagram chase gives the exact sequence
M ker() H 2 (X , Z) 0.
However, i has dimension n for all i, so that M(i ) = 0 for all i. Thus

L
L
ker() = ker
i M/M(i )
i< j M/M(i j ) CDivTN (X ),
and it follows immediately that H 2 (X , Z) Pic(X ).
This is a wonderful illustration of how to use spectral sequences.

Computing the Other Cohomology Groups. The spectral sequence of the cover
U = {U }max can be used to study the H k (X , Z) for all k. However, if there
are many cones in it becomes somewhat unwieldy to derive detailed information
about H k for k > 2 this way. In remainder of this section, we will see how to
use the decomposition of X into torus orbits to compute the cohomology groups
H k (X , Z), and hence the additive structure of H (X , Z), more efficiently. The
ring structure of H (X , Z) will be discussed in 12.4.
A Family of Complexes. The method for computing the cohomology groups of
X that we present comes from [162]. To begin, we discuss a notion of orientation
for a pair of cones with dim = dim + 1. First, for each cone , we may
arbitrarily pick an orientation of the linear subspace (N )R spanned by , determined by a choice of basis. For instance, in the special case that is a simplicial
fan, we can number the one-dimensional cones i = Cone(ui ) in a fixed way and
if = i1 + + i p with i1 < < i p , then we can specify the orientation via the
element
V
ui1 ui p p N .
580
Now if dim = dim + 1, let v be any vector in not contained in . Then v

together with a basis for (N )R forms a basis for (N )R , and defines an orientation.
We define an orientation coefficient
+1 if the orientation of determined by agrees with the chosen one

c, = 1 if not
0 if is not a face of .
Given any fan , fix an integer q, 0 q n, and consider the abelian groups
and maps
V
V
C (, q ) = {(C p (, q ), p ) | p Z}
defined as follows. First, we take
C p (,
Vq
)=
(np)
Vq
M( ),
where M( ) = M as usual. This is free abelian with

Vq
p
p
rankC (, ) =
|(n p)|.
q
Then
p : C p (,
Vq
) C p+1 (,
Vq
is the map defined on the components corresponding to cones (, ) in the two

direct sums by the following rule. If is not a face of , that component of p is
defined to be zero. On the other hand, if , then p is defined by
q
c, i,
,
where c, are the orientation coefficients, and

q
i,
:
Vq
M( )
Vq
M()
is induced by the inclusion V

. In other words, the component of p in the
p+1
summand for the cone in C (, q ) is given by
X
q
c, i,
.
Note
that C p (,
Vq
) is nonzero only for 0 q p n.
Lemma 12.3.3. C (,
Vq
) is a complex, i.e., p+1 p = 0 for all p.
Proof. If is a codimension 2 face of a cone , then there are exactly two facets
of containing , and you will show in Exercise 12.3.1 that
X
(12.3.1)
c, c, = 0.
Now, for any cone (n p),
p+1 p |M( ) = p+1
(np2)
q
c, i,
(np1)

581
(np2)
(np1)

(np1)
q
q
c, c, i,
i,

q
c, c, i,
= 0,
using (12.3.1).
Example 12.3.4. Consider the fan defining P 2 , shown for instance in Figure 2
from Example
3.1.9. Denote i = Cone(ei ) for i = 0, 1, 2. We show the complexes
Vq
(C (, ), ) for q = 0, 1, 2 in the following diagram:

(12.3.2)
/Z
q=2:
/0
/0
q=1:
/0
/ Z3
/ Z3
q=0:
/ Z3
/ Z2
/Z
/0
/0
/0
/ .
For instance, on the row for q = 1, the group C1 (, 1 ) is

M V1
V
M() Z3 ,
C1 (, 1 ) =
(1)
since
V1
V1
Similarly,
V1
M(0 ) = Z(e1 e2 )
M(1 ) = Ze2
M(2 ) = Ze1 .
V
C2 (, 1 ) = 1 M({0}) = M Z2 .
Use orientation coefficients determined by the numbering of the one-dimensional
cones in the order 0 , 1 , 2 and the two-dimensional cones listed in the order
1 , 2 , 0 . The map denoted by C in (12.3.2) is defined by the matrix

1 0 1
.
C=
0 1 1
The maps denoted by A and B in (12.3.2) are
1 1
0
1 1 ,
A= 0
1
0
1

B= 1 1 1 .
582
Note that BA = 0 in agreement with Lemma 12.3.3.
The cohomology groups

H p (,
Vq
) = ker( p )/im( p1 )
of these complexes will appear shortly.

Spectral Sequence of a Filtered Topological Space. For each integer p, the toric
variety X contains the union of the closures of torus orbits of dimension p,
[
a
Xp =
V () =
O( ).
(np)
(), np
By definition, these subsets give an increasing filtration of X ,

(12.3.3)
= X1 X0 X1 Xn = X .
For technical reasons, when working with general , where X may not be
compact, we will consider cohomology with compact supports, denoted Hck (X , Z)
(see [135, p. 242]). When is complete the groups Hck (X , Z) and the ordinary
singular cohomology groups H k (X , Z) coincide. For orientable noncompact manifolds of real dimension d, it is the cohomology groups with compact supports that
appear in the statement of Poincare duality (see [135, Thm. 3.35]). Thus for instance, if X is homeomorphic to an open ball in Rd ,
(
Z if k = d
(12.3.4)
Hck (X , Z) =
0 otherwise.
Corresponding to the filtration (12.3.3), we have a first quadrant cohomology
spectral sequence Erp,q with
(12.3.5)
p,q
E1 = Hcp+q (X p , X p1 , Z) Hcp+q (X , Z).
See Theorem C.2.5 in Appendix C, and [136, Ch. 1] or [255, 9.4] for more details
on the construction.
Proposition 12.3.5. For p, q 0, the spectral sequence (12.3.5) has
M Vq
V
E1p,q
M( ) = C p (, q ).
(np)
Moreover, the differentials

d1p,q : E1p,q E1p+1,q agree with the coboundary maps
V
q
in the complex C (, ), so that
E2p,q = H p (,
Vq
).
Proof. By the excision property of cohomology with compact supports, we have

M
p,q
E1
Hcp+q (O( ), Z).
(np)
583
p
Furthermore, the homeomorphism O( )
= R>0 SN( ) and the Kunneth formula
imply that
M
p
Hcp+q (O( ), Z)
Hck (R>0
, Z) Z Hc (SN( ) , Z).
k+=p+q
To analyze the terms in this direct sum, note that by (12.3.4)

(
Z if k = p
p
Hck (R>0 , Z) =
0 otherwise.
Using Proposition 12.3.1, this shows that for each cone of dimension n p,
p,q
Hcp+q (O( ), Z) H q (SN( ) , Z)
Hence E1 C p (,
Vq
) as claimed.
Vq
M( ).
To complete
Vq the proof, we need to show that the coboundary maps in the com
plex C (, ) agree with the differentials in the spectral sequence. By the construction of the spectral sequence as in [255], the differential
d1p,q : E1p,q E1p+1,q
comes from the connecting homomorphism in the long exact cohomology sequence
of the triple (X p+1 , X p , X p1 ):
Hcp+q (X p , X p1 , Z) Hcp+q+1 (X p+1 , X p , Z).
By considering how this connecting homomorphism

arises, it can be seen that d1p,q
Vq
p
coincides with from the complex C (, ) (Exercise 12.3.2).

Since E1p,q is zero for all p, q outside the triangular region 0 q p n, it
follows easily that for r sufficiently large, all of the higher differentials
drp,q : Erp,q Erp+r,qr+1
will be zero and the spectral sequence will degenerate with
p,q
p,q
Erp,q = Er+1
= = E
p,q
for all p, q. When this happens, the E
with p + q = k are successive quotients in
k
a filtration of Hc (X , Z). In relatively small dimensions, and in many other good
cases as well, this information can be used to determine these cohomology groups
completely. We will illustrate this with the following examples.
Example 12.3.6. We continue from (12.3.2) to compute the E2 sheet of the spectral
sequence (12.3.5) arising from the fan for P 2 . By a direct computation, the q = 0
row is E20,0 = Z and E21,0 = E22,0 = 0 (see also Exercise 12.3.3).
On the second row, the kernel of C is 1-dimensional and the image of C is Z2 .
Hence E21,1 Z, and E22,1 = 0. Finally E22,2 = Z. Hence the E2 sheet of the spectral
584
sequence is reduced to:

0
E22,2 = Z
E21,1 = Z
E20,0 = Z
(12.3.6)
The E2 differentials are d2p,q : E2p,q E2p+2,q1 . It can be seen directly in this case
that the spectral sequence degenerates at E2 . Since there is at most one nonzero
group for each possible value of p + q, it follows that
H 0 (P 2 , Z) Z,
H 2 (P 2 , Z) Z,
H 4 (P 2 , Z) Z,
and all of the other H k (P 2 , Z) (including all the odd-numbered ones) are zero (see
Proposition C.1.5). The computations here can be generalized without difficulty to
the case of X = P n defined by the fan with one-dimensional cones i = Cone(ei ),
i = 0, . . . , n and e0 = e1 en . The result is that
H 2k (P n , Z) Z,
k = 0, . . . , n,
and the odd-numbered cohomology groups are all zero.
Our next examples show that the integral cohomology groups of a toric variety
can have torsion.
Example 12.3.7. Let in NR R2 be the complete fan with 1-dimensional cones
i and ray generators ui for i = 1, . . . , r, listed counterclockwise around the origin.
The 2-dimensional cones are i = i + i+1 , i = 1, . . . , r if we take indices modulo
r. The rows of the E1 sheet of the spectral sequence (12.3.5) for q = 0, 1, 2 are
0
(12.3.7)
/0
/0
/0
/ Zr
/ Zr
/ Zr
/Z
/0
/ Z2
/Z
/0
/0
/
/
/ .
By Exercise 12.3.3, E2p,0 = 0 for p = 1, 2 and E20,0 Z.
The map C on the q = 1 row is the defined by the 2 r matrix with columns
given by the vectors ui = (ai , bi ). Since each pair ui , ui+1 spans a two-dimensional
cone i , the kernel of C has rank r 2 and the image of C is a rank 2 sublattice of
M. The E2 sheet of the spectral sequence has the form
(12.3.8)
E22,2 = Z
E21,1 = Zr2
E22,1 = Z/mZ
E20,0 = Z
0.
585
The integer m is given by

ai a j
m = gcd det
bi b j

1i< jr .

Once again, because of the placement of the nonzero terms in E2 , all of the higher
p,q
differentials must be zero, so E2 = E and there is at most one nonzero E
for
each value of p + q. Hence
H 0 (X , Z) Z, H 1 (X , Z) = 0,
H 3 (X , Z) Z/mZ, H 4 (X , Z) Z.
H 2 (X , Z) Zr2 ,
By the universal coefficient theorem for cohomology [135, Thm. 3.2], the torsion
also appears in H2 (X , Z). Also, note that H1 (X , Z) = 0 as we expect from 12.1
since X is simply connected and H1 (X , Z) is the abelianization of 1 (X ). You
will study some special cases in Exercise 12.3.6.
We conclude with a nonsimplicial example that illustrates most of the general

behavior of these cohomology groups.
Example 12.3.8. Consider the fan in R3 whose maximal cones are the cones
over the faces of the cube with vertices (1, 1, 1). This cube is shown in Figure
8 of Example 2.3.11 and is the normal fan of the octahedron shown in the same
figure. The E1 sheet in the spectral sequence (12.3.5) is:
0
0
0
0
/0
/0
/0
/0
/0
/ Z6
/Z
/0
/ Z12
/ Z12
/ Z8
/ Z3
/ Z16
/ Z8
/ Z3
/Z
/0
/0
/0
/0
/
/ .
Suitable orientation coefficients can be obtained by placing any orientations on the

edges of the cube in R3 . We leave it to the reader to prove the following claims.
First, the matrices A, B,C on the q = 0 row of the E1 sheet make that row exact
except at the left and A has a rank 1 kernel, so E20,0 Z, while
E21,0 = E22,0 = E23,0 = 0.
For the q = 1 and q = 2 rows, we must determine the M() for the cones in .
When this is done, it is routine to show that
D has rank 11 and E has rank 3. The
V
image of E is a sublattice of index 2 in 1 M Z3 . Hence, E21,1 Z, E22,1 Z2 ,
and E23,1 Z/2Z. With respect to obvious choices of bases, the matrix F is
1 1
1
1
1 1
1 1
1
1 1
1 1 1
1 .
F = 1
1
1 1 1 1 1
1
1
586
It is easy to check that F has rank 3, but that the image is a sublattice of index 4 in
M Z3 . The quotient is isomorphic to Z/2Z Z/2Z. As a result the E2 sheet
of the spectral sequence is:
V2
(12.3.9)
0
E20,0 = Z
E22,2 = Z5 E23,2 = Z/2Z Z/2Z
E21,1 = Z E22,1 = Z2
0
E23,3 = Z
E23,1 = Z/2Z
0.
In this case too, all of the higher differentials are zero, so the spectral sequence
degenerates at this point, and the E sheet coincides with (12.3.9). You will verify
the details in Exercise 12.3.7 and show that
H 0 (X , Z) Z,
H 2 (X , Z) Z,
H 4 (X , Z) Z5 Z/2Z,
H 6 (X , Z) Z.
H 1 (X , Z) = 0
H 3 (X , Z) Z2
H 5 (X , Z) Z/2Z Z/2Z
Note that unlike the previous examples, some of the odd cohomology groups
would be nonzero in this example even if coefficients in Q were used. One can do
these computations using the Maple package torhom developed by Franz [98].
The Topological Euler Characteristic. The spectral sequence (12.3.5) can be used
to deduce several connections between the topology of the toric variety X and the
combinatorics of the fan . First, we consider the topological Euler characteristic
of X , which by [105, p. 95] and [162, Prop. 3.1.2] equals
(12.3.10)
e(X ) =
2n
2n
X
X
(1)k dim Hck (X , Q).
(1)k rank Hck (X , Z) =
k=0
k=0
Theorem 12.3.9. For an n-dimensional toric variety X , the Euler characteristic

is the number of n-dimensional cones in , i.e.,
e(X ) = |(n)|.
Proof. For each sheet of the spectral sequence (12.3.5), define
X
(1) p+q rank Erp,q .
e(Er ) =
p,qZ
Since Er+1 is the cohomology of Er with respect to drp,q , it follows that

e(Er+1 ) = e(Er )
587
p,q
for all r. The E
with p + q = k are the quotients of a filtration of H k (X , Z), and
by Proposition 12.3.5,

M Vq
X
X

p
M( ) =
(1) p+q
e(E1 ) =
(1) p+q rank
|(n p)|
q
p,qZ
since M( )
(np)
p,qZ
Zp
when (n p). Hence

p
n X
X
p
|(n p)| = |(n)|,
(1) p+q
e(X ) = e(E ) = e(E1 ) =
q
p=0
q=0
where the last equality follows since the inner sums vanish for p > 0 by properties
of binomial coefficients.

When X is complete and simplicial, the individual Betti numbers (the ranks
of the H k (X , Z)) can also be determined from the structure of .
Rational Coefficients. To suppress torsion in cohomology, we switch coefficients
from Z to Q. In this case, there is also a significant simplification in the behavior
of the cohomology spectral sequence
(12.3.11)
E1p,q = Hcp+q (X p , X p1 , Q) Hcp+q (X , Q).
The same argument given in Proposition 12.3.5 shows that

M Vq
(12.3.12)
E1p,q
M( )Q .
(np)
Proposition 12.3.10. The spectral sequence (12.3.11) degenerates at E2 .

p,q
Proof. We show that drp,q = 0 for all r 2 and all (p, q), so that E2p,q = E
. Recall
from Example 3.3.6 that for any positive integer , the multiplication map
: N N,
a 7 a
is compatible with so there is a corresponding toric morphism : X X

whose restriction to TN X is the group homomorphism
|TN (t1 , . . . ,tn ) = (t1 , . . . ,tn ),
and similarly on each torus orbit. Because respects the orbit decomposition of
X , it respects the filtration from (12.3.3) and induces homomorphisms
: Erp,q Erp,q
for each r. These commute with the differentials since (12.3.11) is functorial with
respect to maps that preserve the filtration.
L
In Exercise 12.3.8 you will use E1p,q (np) Hcq (O( ), Q) to show that
p,q
is
acts on E1p,q by multiplication by q . Then the same holds for all r since Er+1
p,q
a quotient of a subspace of Er .
588
Let Erp,q for r 2. Since drp,q () Erp+r,qr+1, we have

qr+1 drp,q () = (drp,q ())
= drp,q ( ())
= drp,q ( q )
= q drp,q ().
Since we use coefficients in Q, this implies drp,q () = 0 for all .
Proposition 12.3.10 shows that computing Hck (X , Q) using the cohomology

spectral sequence is especially simple. Franz has shown in [96] that if X is
smooth, the spectral sequence with integer coefficients also degenerates at E2 , and
no additional extension data is needed to determine Hc (X , Z).
A Vanishing Theorem for Singular Cohomology. We will now focus on complete
simplicial toric varieties. In this case, the following result of Oda [220, Thm. 4.1]
implies that the spectral sequence (12.3.11) simplifies even further.
Theorem 12.3.11. If X is complete and simplicial, then E2p,q = 0 when p 6= q in
the spectral sequence (12.3.11). Thus:
(a) H 2k+1 (X , Q) = 0 for all k.
(b) H 2k (X , Q) E2k,k for all k.
Proof. Our proof will use some results from Chapter 9. Theorem 9.3.2 tells us that
bq ) = 0
(12.3.13)
H p (X ,
X
when p 6= q, and the Hodge decomposition (9.4.11) states that

M
b q ).
(12.3.14)
H k (X , C)
H p (X ,
X
p+q=k
Here,
bq
is the sheaf of Zariski q-forms on X defined in 8.0.
It is now easy to show that E2p,q = 0 when p + q is odd, since (12.3.13) and
(12.3.14) imply that H k (X , C) = 0 for odd k. When we combine this with the
degeneration proved in Proposition 12.3.10, we obtain
P
0 = dim H k (X , Q) = p+q=k dim E2p,q .
When p + q is even, we change notation slightly and consider p + q = 2k. The

idea is to study how the maps : X X from the proof of Proposition 12.3.10
act on each side of (12.3.14). On the left-hand side, the degeneration implies that
H 2k (X , C) has a filtration 0 = F 2k F 0 = H 2k (X , C) such that
F p /F p1 (E2p,q )C = E2p,q Q C,
p,q
p + q = 2k.
p,q
By the proof of Proposition 12.3.10, : (E2 )C (E2 )C is multiplication by

q . Hence : H 2k (X , C) H 2k (X , C) is multiplication by q on F p /F p1 .
589
Now consider the right-hand side of (12.3.14) and assume for the moment that
X is smooth. Recall from 8.1 that
X1 M Z OX X1 (log D).
Since m = m , we have (d m / m ) = d m / m = d m / m . It follows
that over the affine openVsubset U , , acts on sections by multiplication
by . Hence, for Xq = q X1 , acts on sections over U by multiplication by
q . Computing cohomology via the Cech

complex, we conclude that acts on
q
p
q
H (X , X ) by multiplication by . In Exercise 12.3.9, you will show that in the
b q ).
general case, continues to be multiplication by q on H p (X ,
X
2k
k
k
b
Since H (X , C) H (X , ) by (12.3.13) and (12.3.14), it follows that
X
is multiplication by k on H 2k (X , C). Comparing this with our earlier analysis,

we see that the filtration {F p } must collapse, giving (E2p,q )C = 0 when p + q = 2k
and p 6= q. This proves the theorem.

You should check that Examples 12.3.6 and 12.3.7 illustrate the vanishing of
odd cohomology asserted by part (a) of Theorem 12.3.11. On the other hand, we
see from Example 12.3.8 that this can fail for nonsimplicial X . There is also a
more refined vanishing statement due to Brion [48, p. 5], which states that if X
is complete of dimension n and all cones of dimension m are simplicial,
then E2p,q = 0 unless 0 p q n m.
Some Combinatorial Consequences. There are interesting relations between the
numbers of cones of various dimensions in simplicial fans and the Betti numbers
of the corresponding toric varieties, which are the numbers
b2k (X ) = dim H 2k (X , Q).
Theorem 12.3.12. Let be a complete simplicial fan in NR Rn . Then the Betti
numbers of X are given by

n
X
ik i
(1)
b2k (X ) =
|(n i)|
k
i=k
and satisfy
b2k (X ) = b2n2k (X ).
Proof. The E2p,q terms of the spectral sequence (12.3.11) are the cohomology of
the E1p,q terms, and we also have E1p,q = 0 for p < q by (12.3.12). Since E2p,q = 0
unless p = q by Theorem 12.3.11, it follows that
0 E2k,k E1k,k E1k+1,k
590
is exact. Hence
b2k (X ) =
dim E2k,k
n
X
ik
(1)
dim E1i,k

n
X
ik i
(1)
=
|(n i)|,
k
i=k
i=k
where the last equality uses (12.3.12). The second assertion follows from Poincare
duality, which is discussed in more detail in 12.4.

If P is a full dimensional simple lattice polytope and XP is the corresponding
simplicial toric variety, then |(n i)| = fi , the number of i-dimensional faces of
P. An immediate corollary of the theorem is the formula

n
X
ik i
(1)
(12.3.15)
b2k (XP ) =
fi = hk ,
k
i=k
where the hk are the combinations of the face numbers introduced in 9.4. The
Dehn-Sommerville equations hk = hnk are a consequence of the symmetry of the
Betti numbers in Theorem 12.3.12.
A final comment is that as in the proof of Theorem 12.3.11, (12.3.13) and
(12.3.14) imply that
b Xk )
H 2k (XP , C) H k (XP ,
P
b k ) = hk by Theorem 9.4.11, we
for P as above. Since we also have dim H k (XP ,
XP
get another proof of (12.3.15). This was noted earlier in (9.4.12).
Exercises for 12.3.
12.3.1. Show that if is a face of of codimension 2, then

X
c, c, = 0,
where the sum has two terms corresponding to the two facets of containing .
12.3.2. Complete the proof of Proposition 12.3.5 by showing that
d1p,q : H p+q (Xp , Xp1 , Z) H p+q+1 (Xp+1 , Xp , Z)
Vq
Vq
coincides with p : C p (, ) C p+1 (, ) under the isomorphisms given in the first
part of the proof. Hint: If you get stuck, see [162].
12.3.3. This exercise will show that when X is complete, the q = 0 row of the E2 sheet of
the spectral sequence (12.3.5) is given by
(
Z if p = 0
p,0
E2 =
0 otherwise.
Let Sn1 be the unit sphere in NR and let
C = { Sn1 | , 6= {0}}
be the spherical cell complex determined by .

V
(a) Show that the complex (C (, 0 ), ) is isomorphic to the augmented cellular chain
complex of the spherical complex C (see [135, p. 139]).
591
enp1 (C, Z).

(b) Show that E2p,0 is isomorphic to the reduced homology group H
(c) If is complete, then C is a subdivision of Sn1 . Deduce the above formula for E2p,0 .
12.3.4. Let Bl0 (C2 ) denote the blowup of C2 at the origin. Compute the cohomology
groups H k (Bl0 (C2 ), Z) directly from the spectral sequence (12.3.5). Generalize this by
computing H k (Bl0 (Cn ), Z) for all n 2.
12.3.5. Let be the complete fan in NR R2 with four maximal cones Cone(e1 , e2 ),
so X = P1 P1 . Compute H k (X , Z) directly via the spectral sequence (12.3.5), and
compare with the result in Example 12.3.7. Generalize your computations to the complete
fans in NR Rn with maximal cones Cone(e1 , . . . , en ).
12.3.6. This exercise deals with Example 12.3.7.

(a) Show that if X is a smooth toric surface, then the cohomology groups H k (X , Z) are
torsion free.
(b) Let X be the toric variety of the complete fan in R2 whose minimal ray generators
are u = (1, 1). Show that H 3 (X , Z) Z/2Z.
12.3.7. In this exercise, you will verify details of the computation in Example 12.3.8.
(a) Show that the E sheet of the cohomology spectral sequence is given by (12.3.9).
(b) Show that the cohomology groups are as given in the example. Note that Proposition C.1.5 applies for all of the groups except H 4 (X , Z). Show that H 4 (X , Z)
p,q
Z5 Z/2Z directly by considering the filtration with quotients given by E
with
p + q = 4.
12.3.8. Show that is the th power map on O( ) and conclude that is multiplication
by q on Hcq (O( ), Q) H q (SN( ), Q).
12.3.9. Here are some details from the proof of Theorem 12.3.11
(a) Use the properties of reflexive sheaves from 8.0 to prove the assertion in the text that
b q by multiplication by q .
over U , acts on sections
X
(b) Explain carefully how computing cohomology via the Cech

complex implies that
q
p
q
b ) by multiplication by .
acts on H (X ,
X
12.3.10. In this exercise, you will show that if in NR Rn is smooth and complete, then
the higher homotopy group 2 (X ) can be identified with Pic(X ) .
(a) Show that 2 (X ) H2 (X , Z). Hint: Apply the Hurewicz theorem as stated in [135,
Thm. 4.32].
(b) Prove that H2 (X , Z) (H 2 (X , Z)) under these hypotheses. Hint: You will need to
show that H2 (X , Z) has no torsion.
(c) Conclude that 2 (X ) is isomorphic to Pic(X ) .
12.3.11. To what extent is the cohomology of a toric variety X determined by the combinatorics of the fan (the numbers of cones of each dimension, and their intersection
relations)? In this exercise, drawing on [200], you will see that for n 3, the Betti numbers of a toric variety are not necessarily determined by the combinatorial type of .
(a) To begin, deduce from Example 12.3.7 that the Betti numbers of a complete toric
surface are determined by the number of 2-dimensional cones in , and that this is the
only combinatorial invariant of .
592
(b) Now as in Example 12.3.8, consider the fan over the faces of the polytope P in R3
given by

P = Conv e1 , e2 , e3 , 21 e1 21 e2 12 e3 ,
where we take all possible choices of signs, so there are 14 points in all in the set.
Show that the Betti numbers of X are
1, 0, 2, 3, 11, 0, 1.
Note that X is not simplicial and Poincare duality does not hold.
(c) Now let P be the convex hull of the 14 points
e1 , e2 , 12 e1 + 12 e3 , e1 , e2 , e3 , 52 e1 + 35 e2 + 15 e3 , 21 e1 + 12 e2 12 e3 ,
3
1
1
1
1
2
1
1
1
1
1
2
5 e1 5 e2 + 5 e3 , 2 e1 2 e2 2 e3 , 3 e1 + 3 e2 + 3 e3 , 2 e1 + 2 e2 2 e3 ,
32 e1 13 e2 + 13 e3 , 21 e1 21 e2 12 e3 .
If is the fan over the faces of P , show that and are combinatorially equivalent,
but X has Betti numbers
1, 0, 1, 2, 11, 0, 1.
(d) Modify this example to produce similar examples in all dimensions n 3.
P2n
12.3.12. The Poincare polynomial PX (t) = k=0 bk (X)t k of a variety X of dimension n is
the generating function for P
its Betti numbers. Show that Theorem 12.3.12 is equivalent to
the assertion that PX (t) = (t 2 1)dim .
12.4. The Cohomology Ring
In this section, we will make the standing assumption that X is complete and
simplicial. Our goal is to prove a general theorem describing the ring structure
given by cup product on
H (X , Q) =
2n
M
H k (X , Q),
n = dim X .
k=0
Along the way, we will also describe the equivariant cohomology ring HTN (X , Q).
Rationally Smooth Varieties. By Theorem 11.4.8, simplicial toric varieties are
rationally smooth. The basic intuition is that rationally smooth varieties behave
like smooth varieties, provided that one works over Q.
We begin with some properties of the cohomology ring of an n-dimensional
complete rationally smooth variety X :
Poincare duality holds between cohomology and homology, so that
H k (X , Q) H2nk (X , Q).
The isomorphism H 2n (X , Q) H0 (X , Q) = Q induces a map

Z
: H (X , Q) Q,
X
where elements of H k (X , Q) map to zero when k < 2n.
593
Poincare duality implies that the cup product pairing
H k (X , Q) H 2nk (X , Q) Q
R
defined by (, ) 7 X ` is nondegenerate. Thus H 2nk (X , Q) is isomorphic to the dual vector space of H k (X , Q) and their dimensions are equal.
An irreducible subvariety W X of codimension k has a refined cohomology

class [W ]r H 2n2k (X , X \W , Q). By mapping this to H 2n2k (X , Q), W has a
cohomology class [W ] H 2n2k (X , Q).
P
P
The cohomology class of a divisor D = i ai Di is [D] = i ai [Di ] H 2 (X , Q),
and linearly equivalent divisors give the same cohomology class.
See [135, Sec. 3.3] for the manifold case. For rationally smooth varieties these
statements follow, for instance, from the properties of intersection cohomology
developed by Goresky and MacPherson in [119] since over Q, intersection cohomology coincides with ordinary cohomology for rationally smooth varieties. The
assertions about refined cohomology classes and linearly equivalent Cartier divisors can be found in [107, Ch. 19]. We also note that if X is smooth, then the above
properties hold over Z. See also the appendix to Chapter 13.
These properties enable us to generalize the intersection products defined in

6.3. Let V ,W X be irreducible subvarieties with dim V + dim W = n. The cup
product [V ] ` [W ] lies in H 2n (X , Q) and hence gives the intersection product
Z
(12.4.1)
V W = [V ] ` [W ] Q.
X
When D is an irreducible Cartier divisor and C is an irreducible curve, this agrees

with the intersection product D C from Definition 6.3.6. See Exercise 12.5.9 and
[107, 2.3].
Statement of the Main Theorem. Let X be a complete simplicial toric variety
and fix a numbering 1 , . . . , r for the rays in (1). Also let ui be the minimal
generator of i and introduce a variable xi for each i . In the ring Q[x1 , . . . , xr ], let
I be the monomial ideal with square-free generators as follows:
(12.4.2)
I = hxi1 xis | i j are distinct and i1 + + is is not a cone of i.
We call I the Stanley-Reisner ideal. Also let J be the ideal generated by the
linear forms
r
X
hm, ui ixi ,
(12.4.3)
i=1
where m ranges over M (or equivalently, over some basis for M). We define
RQ () = Q[x1 , . . . , xr ]/(I + J ).
Since I + J is homogeneous for the standard grading on Q[x1 , . . . , xr ], RQ () is

a graded ring. Let RQ ()d denote the graded piece in degree d.
594
Our next task is to show that xi [Di ] H 2 (X , Q) induces a well-defined

ring homomorphism
(12.4.4)
RQ () H (X , Q).
P
First note that ri=1 hm, ui iDi = div( m ) 0, which by the above properties of
P
cohomology classes implies that ri=1 hm, ui i[Di ] = 0 H 2 (X , Q). This shows
that the ideal J maps to zero in H (X , Q).
Moreover, if i1 + + is is not a cone of , then Di1 Dis = in X .

Since cup product in relative cohomology satisfies
H k (X , A, Q) and H (X , B, Q) = ` H k+ (X , A B, Q)
(see [210, Ch. 5]), the cup product of the refined classes [Di j ]r is
S
[Di1 ]r ` ` [Dis ]r H 2s (X , sj=1 (X \ Di j ), Q).
This relative cohomology group vanishes since Di1 Dis = . Then

[Di1 ] ` ` [Dis ] = 0 H 2s (X , Q)
since [Di j ]r maps to [Di j ]. Hence the Stanley-Reisner ideal I also maps to zero in
H (X , Q). This gives the ring homomorphism (12.4.4).
Our main theorem will show that RQ () and H (X , Q) are isomorphic.
Theorem 12.4.1. Let be complete and simplicial. Then the map (12.4.4) is an
isomorphism:
RQ () H (X , Q).
Thus, in even degrees, H 2k (X , Q) is isomorphic to RQ ()k , and in odd degrees,
H 2k+1 (X , Q) is zero.
The proof of this theorem will be given later in the section.
Example 12.4.2. As a toric variety, P n comes from the fan with ray generators
ui = ei for i = 1, . . . , n and u0 = e1 en . Then it is easy to check that
I = hx0 xn i,
and using the basis e1 , . . . , en for M to obtain generators for J , we have
J = hx1 x0 , . . . , xn x0 i.
Then Theorem 12.4.1 gives an isomorphism
H (P n , Q) Q[x0 , x1 , . . . , xn ]/hx0 xn , x1 x0 , . . . , xn x0 i
Q[x0 ]/hx0n+1 i.
This agrees with Example 9.0.13. See also Example 12.3.6.
595
Example 12.4.3. Consider the Hirzebruch surface Hr and label the cones in the
fan as in Example 10.4.6, so 1 = Cone(e1 + re2 ), 2 = Cone(e2 ), 3 = Cone(e1 ),
and 4 = Cone(e2 ). Theorem 12.4.1 gives
H (Hr , Q) Q[x1 , x2 , x3 , x4 ]/hx1 x3 , x2 x4 , x1 + x3 , r x1 + x2 x4 i.
The two linear relations from J can be used to eliminate x1 , x2 , giving

H (Hr , Q) Q[x3 , x4 ]/hx32 , x42 r x3 x4 i.
Note that 1, x3 , x4 , x3 x4 form a Q-basis of H (Hr , Q). Hence

H 0 (Hr , Q) Q,
H 2 (Hr , Q) Q2 ,
and
H 4 (Hr , Q) Q,
as we expect from 12.3. Also note that cup product in H (Hr , Q) is defined by
x32 = 0,
x42 = r x3 x4 .
By (12.4.1), we recover the intersection form on the divisor classes, described by

the matrix

0 1
(Di D j ) =
1 r
in Example 10.4.6.
Theorem 12.4.1 also holds over Z when X is smooth. The examples just given
show this for P n and the Hirzebruch surface Hr . Here is the precise result.
Theorem 12.4.4 (Jurkiewicz-Danilov). Let X be a smooth complete toric variety.
For the polynomial ring Z[x1 , . . . , xr ] with variables indexed by 1 , . . . , r (1),
let I and J be the ideals in Z[x1 , . . . , xr ] generated by the polynomials in (12.4.2)
and (12.4.3), and define
R() = Z[x1 , . . . , xr ]/(I + J ).
Then xi 7 [Di ] induces a ring isomorphism R() H (X , Z).
See [76, Thm. 10.8], [105, Sec. 5.2] (which gives a proof under an additional
hypothesis that is satisfied, for instance, if X is projective), and [218, Sec. 3.3].
Note that Theorem 12.4.1 is simultaneously less general (coefficients in Q instead
of Z) and more general (X is simplicial instead of smooth) than Theorem 12.4.4.
Let us make one final comment about the cohomology ring H (X , Q). Recall
from Theorem 12.3.12 that the Betti numbers bk (X ) = dim H k (X , Q) depend
only on the combinatorial structure of the fan in the simplicial case. Does this
extend to the ring structure on H (X , Q)? Sometimes the answer is yes.
Example 12.4.5. Given nonnegative integers r 6= s, the Hirzebruch surfaces Hr
and Hs have cohomology rings
H (Hr , Q) Q[x, y]/hx 2 , y 2 r xyi, H (Hs , Q) Q[x, y]/hx 2 , y 2 s xyi
by Example 12.4.3. One easily checks that (x, y) 7 (x, y + 21 (s r)x) induces a
ring isomorphism H (Hr , Q) H (Hs , Q).
596
In general, however, the ring structure on H (X , Q) is not a combinatorial

invariant, even when is simplicial (Exercise 12.4.1).
Equivariant Cohomology. We will prove Theorem 12.4.1 by first computing the
equivariant cohomology of X for the action of the torus TN and then passing from
equivariant cohomology to ordinary singular cohomology. This method of proof
comes from an extensive study of equivariant cohomology by many authors over
the past 30 years. Our presentation draws mostly on [47] and [108].
Let G be a Lie group acting on a topological space X on the left. When the
action of G on X is not free, the quotient X /G can be badly behaved. As a replacement for X /G when the action is not free, in [38], Borel introduced the following
idea. Let EG be a contractible space on which G acts freely on the right, and let
BG = EG/G. Then EG X is homotopy equivalent to X and G acts freely on this
space. Hence we can form the quotient
EG G X = EG X / , where (e g, x) (e, g x) for g G,
as replacement for X /G. Moreover, EG G X has the structure of a fiber bundle

over BG with fiber equal to X . It is a theorem that suitable, possibly infinitedimensional, EG always exist. The homotopy type of EG G X is also independent
of the EG used. Hence we have a well-defined new cohomology theory.
Definition 12.4.6. Let X be a topological space with a left action of the Lie group
G. The equivariant cohomology of X with respect to G with coefficients in a ring
R, denoted HG (X , R), is defined as
HG (X , R) = H (EG G X , R),
where the right-hand side is ordinary singular cohomology.
We note an important special case of this definition that has some useful consequences. If X = {pt} is a single point with the trivial action of G, then
EG G {pt}
= EG/G = BG.
Hence HG ({pt}, Z) = H (BG, Z) is an algebraic invariant depending only on the

group G. We will call this ring G . If X is any other space on which G acts, the
constant map X {pt} induces a map
(12.4.5)
G = HG ({pt}, Z) HG (X , Z)
that makes HG (X , Z) into a module over the ring G . Since the product in G is
the cup product, G may fail to be commutative if H k (BG, Z) 6= 0 for some odd k.
The Torus. For us, the group G will always be a torus T (C )n . Here, the spaces
EG and BG and the ring G can be described very concretely.
First consider the 1-dimensional torus C . Let C be the space of all vectors
(a0 , a1 , . . .) with ai C for i N, and ai = 0 for i 0. Note that C is the union
597
of C+1 for 1 if we embed C+1 as the set of vectors with zeros after the first
+ 1 entries. In Exercise 12.4.2 you will show the unexpected fact that
(12.4.6)
EC = C \ {0}
is a contractible space. The torus C acts freely on EC in the obvious way:

t (a0 , a1 , . . .) = (ta0 ,ta1 , . . .).
By analogy with the finite-dimensional construction, BC is denoted by P :

BC = EC /C = P .
Note that P =
=0 P . Thus, when computing cohomology in a fixed degree,
we can replace BC with the finite-dimensional approximation P for 0. To

see how this works, note that
H (P , Z) Z[t]/ht +1 i
since the computations in Example 12.4.2 work over Z. Letting , we obtain

(12.4.7)
C = H (BC , Z) = H (P , Z) Z[t],
where the generator t is in degree 2. Note that in this case C is a commutative

ring since H k (P , Z) = 0 when k is odd.
For the n-dimensional torus T = (C )n , we can take
ET = EC EC ,
which implies
BT = BC BC .
Therefore by the Kunneth formula,
(12.4.8)
HT ({pt}, Z) = H (BT , Z) = TN Z[t1 , . . . ,tn ].
Note that TN is commutative and the ti all have degree 2.
For a torus TN = HomZ (M, C ), we can describe (12.4.8) more intrinsically as

follows. Recall that M has the symmetric algebra
SymZ (M) = Z M Sym2 (M) .
To map this to H (BTN , Z), take m M. By the functorial properties of G 7 BG

(see [271]), m : TN C induces B m : BTN BC . Taking H 2 , we obtain
(B m )
Z t H 2 (BC , Z) H 2 (BTN , Z).
Then define s : M H 2 (BTN , Z) by m 7 (B m ) (t). This gives a canonical isomorphism
s : SymZ (M) H (BTN , Z).

To simplify notation, we will often write T instead of TN in what follows. Then,
when we switch from Z to Q, these isomorphisms become
(12.4.9)
s : SymQ (M) HT ({pt}, Q) = H (BT , Q) = (T )Q ,
598
where SymQ (M) is our simplified notation for SymZ (M) Z Q = SymQ (MQ ).
The Fixed Points. Now let T = TN be the torus of X and let XT denote the set
of fixed points for the torus action. The Orbit-Cone Correspondence implies that
there are finitely many fixed points x , one for each cone (n).
The inclusion XT X induces a ring homomorphism
(12.4.10)
Since XT is finite, we have
HT (X , Z) HT (XT , Z).
HT (XT , Q) =
(n)
HT ({x }, Q) =
so that (12.4.10) can be written

(12.4.11)
: HT (X , Q)
(T )Q ,
(n)
(T )Q .
(n)
The map is used in the localization theorem, which is a key tool in equivariant
cohomology. We will say more about this in Corollary 12.4.9.
Comparing Equivariant and Ordinary Cohomology. We next study the relation
between equivariant and singular cohomology for a toric variety. Let T = TN be
the torus of X and note that HT (X , Q) is a (T )Q -module by (12.4.5). This is
compatible with the ring structure, so that HT (X , Q) is a (T )Q -algebra.
The next proposition shows that the equivariant cohomology of a toric variety
can be obtained from its singular cohomology by a change of scalars.
Proposition 12.4.7. Let X be a complete simplicial toric variety and let T = TN .
Then there is a isomorphism of (T )Q -modules
HT (X , Q) (T )Q Q H (X , Q).
In particular, HT (X , Q) is a free (T )Q -module of finite rank.

Proof. By the Borel construction, ET T X is a fiber bundle over BT = (P )n
with fiber equal to X . The Serre spectral sequence for singular cohomology (see
[136, Ch. 1], [199, Thm. 5.2], [255, Sec. 9.4], and Theorem C.2.6) computes the
cohomology of the total space ET T X in terms of the cohomologies of the base
and the fiber. Since we use rational coefficients, the E2 sheet has the form
E2p,q = H p (BT , H q (X , Q)) H p (BT , Q) Q H q (X , Q)
by the universal coefficient theorem, and the spectral sequence converges to
H p+q (ET T X , Q) = HTp+q (X , Q).
However, H q (X , Q) = 0 for odd q by Theorem 12.3.11, and similarly H p (BT , Q)

vanishes for odd p by the comments following (12.4.8). Since the rth differential
is dr : Erp,q Erp+r,qr+1, it follows that all of the dr are zero for r 2. Hence
599
the Serre spectral sequence degenerates at E2 . Then the Leray-Hirsch theorem (see
[135, Thm. 4D.1]) implies that
H (ET T X , Q) = HT (X , Q) H (BT , Q) Q H (X , Q).
This is a (T )Q -module isomorphism since ET T X BT induces the edge

homomorphism E p,0 = H p (BT , Q) HTp (X , Q) by Theorem C.2.6.

Example 12.4.15 below will show that the isomorphism in Proposition 12.4.7
may fail to preserve the ring structure.
In general, the action of a torus T on a space X is called equivariantly formal
if the spectral sequence in Proposition 12.4.7 degenerates at E2 . Starting in [38],
equivariantly formal spaces have been studied extensively in topology and symplectic geometry. A good reference is the paper [118] by Goresky, Kottwitz and
MacPherson. For toric varieties, the paper [97] by Franz gives conditions under
which Proposition 12.4.7 remains true over Z.
We next describe two useful consequences of Proposition 12.4.7. The first
uses the ideal IT (T )Q generated by elements of positive degree. The quotient
(T )Q /IT Q gives Q the structure of a (T )Q -module. Here is the result.
Corollary 12.4.8. The ordinary cohomology ring H (X , Q) is determined by the
(T )Q -algebra structure of HT (X , Q) via the isomorphism
H (X , Q) HT (X , Q)/IT HT (X , Q) HT (X , Q) (T )Q Q.
Proof. We have to be careful since HT (X , Q) (T )Q Q H (X , Q) need not
be a ring isomorphism. When we combine this isomorphism with (T )Q Q, we
get a surjective homomorphism of graded Q-vector spaces
(12.4.12)
HT (X , Q) H (X , Q)
with kernel IT HT (X , Q). If we can show that (12.4.12) is a ring homomorphism,

then the corollary will follow.
The isomorphism of Proposition 12.4.7 comes from the Serre spectral sequence
used in the proof. Degeneration at E2 implies that
L
L
HTq (X , Q) = a+b=q E2a,b = a+b=q H a (BT , Q) Q H b (X , Q).
Since IT (T )Q = H (BT , Q) is generated by the elements of positive degree, the

map (12.4.12) in degree q is the edge homomorphism HTq (X , Q) E20,q , which
by Theorem C.2.6 is the map induced by the inclusion of the fiber
iX : X ET T X
in the fiber bundle given by the Borel construction. Hence (12.4.12) is the ring
homomorphism
(12.4.13)
iX : HT (X , Q) H (X , Q).
The map iX is often called taking the nonequivariant limit.
600
L
The second corollary concerns the map : HT (X , Q) (n) (T )Q that
we constructed in (12.4.10) using the fixed points of the torus action.
L
Corollary 12.4.9. : HT (X , Q) (n) (T )Q is injective.
Proof. By (12.4.10) and (12.4.11), is the map HT (X , Q) HT (XT , Q) induced

by the inclusion of the fixed point set XT X . Let K be the field of fractions of
(T )Q Q[t1 , . . . ,tn ]. The localization theorem in equivariant cohomology states
that becomes an isomorphism after tensoring with K (see [9]). This implies that
the kernel of is a (T )Q -torsion module. However, HT (X , Q) is free over (T )Q
by Proposition 12.4.7, so ker() is torsion-free. Hence the kernel vanishes.

Other proofs of this corollary can be found in [50, 3.4] and [118, Thm. 7.2].
Both papers explain how one can choose a single element f (T )Q such that
HT (X , Q) f HT (X T , Q) f is an isomorphism. See also Exercise 12.4.3.
The Stanley-Reisner Ring. Our strategy for proving Theorem 12.4.1 will be to
compute HT (X , Q) and then take its nonequivariant limit via Corollary 12.4.8.
For this purpose, we introduce a modification of the ring RQ () appearing in the
statement of Theorem 12.4.1.
Definition 12.4.10. The Stanley-Reisner ring of a fan in NR is
SRQ () = Q[x1 , . . . , xr ]/I ,
where (1) = {1 , . . . , r } and as in (12.4.2), I is the Stanley-Reisner ideal
I = hxi1 xis | i j distinct and i1 + + is is not a cone of i.
Example 12.4.11. Let r be the usual fan defining the Hirzebruch surface Hr as
in Example 12.4.3. Then SRQ (r ) = Q[x1 , x2 , x3 , x4 ]/hx1 x3 , x2 x4 i.
The Stanley-Reisner ring SRQ () also has a natural structure as an algebra

over SymQ (M). This comes from the ring homomorphism
Pr
SymQ (M) SRQ ()
induced by m 7 [ i=1 hm, ui ixi ] for m M, where [ f ] SymQ (M) is the image
of f Q[x1 , . . . , xr ]. As a SymQ (M)-module, an element m M acts on SRQ ()
P
by multiplication by the class of the linear form ri=1 hm, ui ixi . The unexpected
minus sign will become clear once we discuss equivariant cohomology classes at
the end of the section.
Here is a simple but illuminating example.
Example 12.4.12. Let us compute the SymQ (M)-module structure on SRQ (r ) =
Q[x1 , x2 , x3 , x4 ]/hx1 x3 , x2 x4 i from Example 12.4.11. Let e1 , e2 be the standard basis
of M = Z2 . Then
e1 acts by multiplication by
e2 acts by multiplication by
601
P4
i=1 he1 , ui ixi
= x1 x3
i=1 he2 , ui ixi
= r x1 x2 + x4 .
P4
For r 0, the surfaces Hr all have isomorphic Stanley-Reisner rings. But once
we introduce the SymQ (M)-module structure, then the action of e2 tells us which
Hirzebruch surface we have.
The Equivariant Cohomology Class of a Divisor. A divisor D on a toric variety

X has a cohomology class [D] H 2 (X , Q). If D is torus-invariant, then it also
has an equivariant cohomology class [D]T HT2 (X , Q) as follows.
Proposition 12.4.13. Let X be a simplicial toric variety. Then a torus-invariant
divisor has a class [D]T HT2 (X , Q) with the following properties:
(a) [D1 + D2 ]T = [D1 ]T + [D2 ]T .
(b) [div( m )]T = s(m) 1, where 1 HT0 (X , Q) and s(m) (T )Q by (12.4.9).
(c) i [D]T = [D|U ]T , where i : U X is the inclusion of a torus-invariant open

subset of X .
(d) iX [D]T = [D], where iX : HT (X , Q) H (X , Q) is the nonequivariant limit

map from (12.4.13).
We defer the proof until the the end of the section.
The Equivariant Cohomology Ring of a Toric Variety. Recall that HT (X , Q) is a
(T )Q -algebra. Then the isomorphism s : SymQ (M) (T )Q from (12.4.9) makes
HT2 (X , Q) into a SymQ (M)-algebra. Then we have the following theorem.
Theorem 12.4.14. The map xi 7 [Di ]T HT2 (X , Q) induces an isomorphism of
SymQ (M)-algebras SRQ () HT (X , Q).
Before starting the proof, we give two applications. The first is a proof of the
main result of this section.
Proof of Theorem 12.4.1. The ideal hMi SymQ (M) corresponds to IT (T )Q .
Thus Theorem 12.4.14 gives an isomorphism
SRQ ()/hMi HT (X , Q)/IT HT (X , Q).
The definition of SymQ (M)-module structure on SRQ () easily implies that

SRQ ()/hMi Q[x1 , . . . , xr ]/(I + J ) = RQ (),
HT (X , Q)/IT HT (X , Q)
and
is isomorphic to H (X , Q) by Corollary 12.4.8.
This gives a ring isomorphism RQ () H (X , Q) that takes the class of xi to
the nonequivariant limit of [Di ]T , which is [Di ] by Proposition 12.4.12.

We next give a concrete example of Theorem 12.4.14.
602
Example 12.4.15. Combining Example 12.4.12 and Theorem 12.4.14, we see that
the Hirzebruch surface Hr has equivariant cohomology ring
HT (Hr , Q) SRQ (r ) = Q[x1 , x2 , x3 , x4 ]/hx1 x3 , x2 x4 i,
where r is the usual fan for Hr . This has two interesting consequences:
HT (Hr , Q) (T )Q Q H (Hr , Q) by Proposition 12.4.7. This is not a ring
isomorphism, for it were, we would have an injective ring homomorphism
H (Hr , Q) HT (Hr , Q).
This is impossible since elements of H 2 (Hr , Q) are nilpotent in H (Hr , Q),

yet nonzero elements of HT2 (Hr , Q) are never nilpotent in HT (Hr , Q).
As a ring, HT (Hr , Q) Q[x1 , x2 , x3 , x4 ]/hx1 x3 , x2 x4 i does not depend on r.
However, r appears in SymQ (M)-module structure from Example 12.4.12.
In Exercise 12.4.4 you will show that r is an isomorphism invariant of the
SymQ (M)-algebra structure of HT (Hr , Q).
In contrast, we saw in Example 12.4.5 that r is not an isomorphism invariant of the
ring structure of H (Hr , Q).
Preparation for the Proof . We need two lemmas and a diagram before we can
prove Theorem 12.4.14. The first lemma embeds SRQ () into an exact sequence
of SRQ ()-modules. For a cone = i1 + + it in , let
Q[] = Q[xi1 , . . . , xit ] = Q[xi | i (1)] Q[x1 , . . . , xr ]/hx j | j
/ (1)i.
Thus we can view Q[] as a module over Q[x1 , . . . , xr ] where x j Q[ ] = 0 whenever

j
/ (1). Hence Q[] may also be viewed as a module over SRQ (), since each
generator of the ideal I acts by zero on Q[ ]. Then consider the sequence
L
L
(12.4.14)
0 SRQ () (n) Q[] (n1) Q[ ],
where and are defined as follows:

Since f 7 f |xi =0 for i (1)
sends the Stanley-Reisner ideal to zero, we
/
(n)
get a well-defined map

([ f ]) = f |xi =0 for i (1)
for [ f ] SRQ ().
/
(n)
Given f = ( f )(n) and (n 1), we define the -component (f) as
follows. Let the two cones of (n) containing be = + i and = + j
where i < j. Then
(f) = ( f )|xi =0 ( f )|x j =0 .
It is easy to see that and are SRQ ()-module homomorphisms.
Lemma 12.4.16. The sequence (12.4.14) is exact.
603
Proof. We leave it to the reader to show that is injective and that = 0

(Exercise 12.4.5). If we write the sequence as 0 SRQ () Cn Cn1 , then it
remains to prove exactness at Cn .
We will use the Nr -grading on Q[x1 , . . . , xr ] where the graded piece in degree
a Nr is Q x a . When we regard (12.4.14) as a sequence of Q[x1 , . . . , xr ]-modules,
the Nr -grading is preserved since the Stanley-Reisner ideal is generated by monomials. Hence it suffices to check exactness in degree a Nr . The case a = 0 is
easy, since an element of (Cn )0 is f = ( )(n) where Q. Then (f) = 0
forces the to be all equal since is a complete fan.
P
If a = (a1 , . . . , ar ) 6= 0, set = ai >0 i . If
/ , exactness holds trivially
since all graded pieces in degree a vanish. So we may assume . As in (3.2.8),
let = Star() be the fan in N( )R = (N/N )R consisting of the projections of
cones of containing . Note that is complete and simplicial since is.
By Exercise 12.4.5, SRQ ()a (Cn )a (Cn1 )a is isomorphic to
SRQ ()0
(nk) Q[]0
(nk1) Q[ ]0 ,
where k = dim . This is exact by our previous observation about degree 0.
We next consider the following large diagram that will be used in proof of
Theorem 12.4.14:
0
/ SRQ ()
(12.4.15)
0

/ H (X )
T
(n) Q[]
(n) HT (U )
(n1) Q[ ]
B
(n1) HT (U ).
All cohomology is computed over Q in this diagram. The dotted arrow is the
isomorphism we are trying to construct. The top row is exact by Lemma 12.4.14,
and we will prove below that the square on the right commutes and that A and B
are isomorphisms. As for the bottom row, exactness can be proved using [118,
(6.3)]. Our approach is different and will in particular give an elementary proof of
the exactness of the bottom row of (12.4.15).
Let us describe the maps in the diagram. We know and , and is induced
by the inclusions U X for (n). Similarly, if (n 1) is a facet of
(n), then on the and components in the bottom row of (12.4.15), is
induced by the inclusion U U , with the same sign as in the definition of .
For the vertical map A, let (n). For each i (1), the torus-invariant
divisor Di U U has a class [Di U ]T HT2 (U , Q). Then xi 7 [Di U ]T
gives a map Q[] HT (U ). This defines A, and B is defined similarly.
604
Lemma 12.4.17. In the diagram (12.4.15), we have:

(a) The square on the right commutes.
(b) = 0.
(c) is injective.
(d) The vertical maps A and B are isomorphisms.
Proof. Proposition 12.4.13 implies that [Di U ]T maps to [Di U ]T , and part (a)
follows immediately. For part (b), note that (n 1) is a face of exactly two
different maximal cones and since is complete. The commutative diagram
of inclusions
; X cH
H
xx
U cG
G
vv;
U
induces the commutative diagram of maps on equivariant cohomology
HT (X , Q)
rr
xrrr
MMM
MM&
LLL
LL&
qq
xqqq
HT (U , Q)
HT (U , Q)
HT (U , Q).
In Exercise 12.4.6, you will check that the component of mapping to Q[ ] is the
difference of these two restrictions. It follows that vanishes on the image of ,
For part (c), the inclusions {x } U of the fixed points for (n) give a
commutative diagram
/ L
HT (X , Q)
(n) HT (U , Q)
RRR
RRR
RRR
R
L R)
(n) HT ({x }, Q)
where is the injective map from Lemma 12.4.9. Hence is injective.

For part (d), take (n) and for simplicity assume (1) = {1 , . . . , n }.
Since is simplicial, Di U is Q-Cartier, so that i Di U is Cartier for some
integer i > 0. The Picard group of an affine toric variety is trivial, so i Di U =
div( mi ) for some mi M. By Proposition 12.4.13, we obtain
(12.4.16) i [Di U ]T = [i Di U ]T = [div( mi )]T = s(mi ) 1 HT2 (U , Q).
Note also that Di U = div( mi ) implies that hmi , u j i = i i j . The ui are a basis
1
of NQ since (n), and hence the 1
i mi are a basis of MQ . Thus xi 7 i mi
defines an isomorphism A : Q[] = Q[x1 , . . . , xn ] SymQ (M).
605
Let A be the -component of A and let i : HT2 (U , Q) HT2 ({x }, Q) =

(T )Q be induced by the inclusion of the fixed point i : {x } U . Using the
isomorphism s : SymQ (M) (T )Q from (12.4.9), we get a diagram
Q[]
xx
x| xx
EEA
EE
"
FFF
FF
i "
yy
|yy s
HT (U , Q)
SymQ (M)
(T )Q
which is commutative by (12.4.16). By Proposition 12.1.9, there is an equivariant

deformation retraction from U to x . Thus i is an isomorphism. Since A and s
are also isomorphisms, the same is true for A .
The proof that B is an isomorphism is similar (with some small differences)
and will be covered in Exercise 12.4.7.

We now have everything needed to compute the equivariant cohomology of X .
Proof of Theorem 12.4.14. Consider the map Q[x1 , . . . , xr ] HT (X , Q) defined
by xi 7 [Di ]T . To show that this map kills the Stanley-Reisner ideal, take distinct
indices i1 , . . . , is such that i1 + + is
/ . We need to prove that
[Di1 ]T ` ` [Dis ]T = 0 HT (X , Q).
For each (n), this cup product maps to [Di1 U ]T ` ` [Dis U ]T in

HT (U , Q) by Proposition 12.4.13. This vanishes in HT (U , Q) since i j
/ (1) for
some j {1, . . . , s}. Thus the map from (12.4.15) maps [Di1 ]T ` ` [Dis ]T to
zero. Since is injective by Lemma 12.4.17, the cup product is zero in HT (X , Q).
Thus we have a well-defined ring homomorphism : SRQ ()P HT (X , Q).
Also, m M acts on SRQ () by multiplication by the image of ri=1 hm, ui ixi in
SRQ (). This maps to
Pr

i=1 hm, ui iDi T = [div( m )]T = s(m) 1,
where the last equality follows from Proposition 12.4.13. It follows that is a
SymQ (M)-algebra homomorphism.
Now add to the diagram (12.4.15):
/ L
0 / SRQ ()
(n) Q[]
/ H (X )
T
(n) HT (U )
(n1) Q[ ]
(n1) HT (U ).
The argument used in in Proposition 12.4.13 shows that the square on the left
commutes. Hence we have a commutative diagram where the top row is exact
606
(Lemma 12.4.16), the maps A and B are isomorphisms (Lemma 12.4.17), and
is injective (Lemma 12.4.17). From here, an easy diagram chase shows that is
bijective, and the theorem is proved.

The above proof implies that the bottom row of our big diagram
M
M
0 HT (X , Q)
HT (U , Q)
HT (U , Q)
(n)
(n1)
is exact. There are equivariant deformation retractions U O() = {x } and

U O( ) by Proposition 12.1.9. Thus we can rewrite the exact sequence as
M
M
HT (O( ), Q).
(12.4.17) 0 HT (X , Q)
HT (O(), Q)
(n)
(n1)
This sequence, called a Chang-Skjelbred sequence, tells us that the structure of

HT (X , Q) (and hence that of H (X , Q)) is completely determined by three things:
The equivariant cohomology of the fixed point set
S
S
X0 = XT = (n) O() = (n) {x }.
The equivariant cohomology of the 1-dimensional orbits

S
X1 = (n1) O( ).
The inclusions X0 X1 X .
In [118, 7.2], Goresky, Kottwitz and MacPherson show that the same is true when
a variety X is equivariantly formal for the action of a torus and there are finitely
many fixed points and 1-dimensional orbits. If X is smooth, then the same arguments can be carried out with coefficients in Z rather than in Q, and this gives a
proof of the Jurkiewicz-Danilov theorem (Theorem 12.4.4).
Piecewise-Polynomial Functions. The exact sequence (12.4.17) has an interesting
interpretation. For (n), HT (O(), Q) HT (U , Q) SymQ (M), and for
(n 1), you will construct an isomorphism
HT (O( ), Q) SymQ (M )
in Exercise 12.4.7, where M = M/( M). It follows that the exact sequence
for equivariant cohomology can be written
M
M
(12.4.18)
0 HT (X , Q)
SymQ (M)
SymQ (M ).
(n)
(n1)
Now observe the following:

For each (n), SymQ (M) is naturally isomorphic to the ring of polynomial
functions on with coefficients in Q.
Similarly, for each (n 1), SymQ (M ) is naturally isomorphic to the ring
of polynomial functions on with coefficients in Q.
607
The exactness of (12.4.18) means that an element of HT (X , Q) can be thought

of as a collection { f }(n) , where f is a polynomial function on , such that
f | = f | whenever = (n 1).
Thus we can identify the equivariant cohomology of X with the ring of piecewise
polynomial functions with rational coefficients on the simplicial decomposition of
NR defined by with the usual pointwise sum and product. Such functions are also
known as polynomial splines and are an important tool in numerical analysis and
applications. This gives perhaps the most concrete way to compute and understand
the equivariant cohomology of X .
Example 12.4.18. The torus T (C )2 acts as usual on P 2 . In Exercise 12.4.8,
you will verify the following claims. We write SymQ (Z2 ) = Q[x, y]. First, by
(12.4.18), HT (P 2 , Q) is equal to the set of triples ( f0 , f1 , f2 ) Q[x, y]3 such that
f1 f0 hxi
f2 f0 hyi
f2 f1 hy xi.
Therefore,
HT (P 2 , Q) = {( f0 , f0 + xg, f0 + yg + y(y x)h) | f0 , g, h Q[x, y]}.
This shows that as a module over Q[x, y], HT (P 2 , Q) is generated by (1, 1, 1),
(0, x, y), and (0, 0, y(y x)). Since we have one generator in each degree 0, 1, 2,
remembering the doubling of degrees in Theorem 12.4.1, we obtain an explicit
description of the (T )Q -module isomorphism
HT (P 2 , Q) (T )Q Q H (P 2 , Q) SymQ (Z2 ) Q H (P 2 , Q)
from Proposition 12.4.7 in this case.
Properties of Equivariant Cohomology Classes. The final task of this section is to

define the equivariant cohomology class of a torus-invariant divisor on a simplicial
toric variety (not necessarily complete) and to prove that these classes have the
properties listed in Proposition 12.4.13. Our construction will use Chern classes.
In 13.1 we give an overview of the Chern classes of locally free sheaves on
varieties. Here, we combine the algebraic approach with a more topological version
of Chern classes that applies to complex vector bundles over topological spaces.
Let D be a torus-invariant Cartier divisor on a simplicial toric variety X . This
gives the invertible sheaf OX (D), which by Theorem 6.0.18 is the sheaf of sections
of a rank 1 vector bundle : VL X for L = OX (D). Here, we write VD instead
of VL . In Proposition 7.3.1 we constructed a fan D in NR R such that VD is
the toric variety of D. Hence the torus of VD is TNZ = TN C . In particular,
the torus T = TN of X acts on VD such that the projection map is equivariant.
This makes VD into an equivariant vector bundle over X .
608
In the classical topology, : VL X has a Chern class c1 (VD ) H 2 (X , Z),

which equals the algebraic Chern class c1 (OX (D)) discussed in 13.1. Thus
(12.4.19)
c1 (VD ) = c1 (OX (D)) = [D] H 2 (X , Z),
where the last equality uses the properties of Chern classes stated in 13.1.
To construct the cohomology class [D]T HT2 (X , Z), recall that HT (X , Z) =
T X , Z). Since : VD X is T -equivariant, the rank 1 vector bundle
H (ET
(VD )T = ET T VD ET T X

has Chern class [D]T = c1 (VD )T H 2 (ET T X , Z) = HT (X , Z).
Proof of Proposition 12.4.13. We need to show that the classes [D]T satisfy:
(a) [D1 + D2 ]T = [D1 ]T + [D2 ]T .
(b) [div( m )]T = s(m) 1, where 1 HT0 (X , Q) and s(m) (T )Q by (12.4.9).
(c) i [D]T = [D|U ]T , where i : U X is the inclusion of a torus-invariant open

subset of X .
(d) iX [D]T = [D], where iX : HT (X , Q) H (X , Q) is the nonequivariant limit
map from (12.4.13).
For part (a), note that VD1 +D2 = VD1 VD2 , so (VD1 +D2 )T = (VD1 )T (VD2 )T
as vector bundles over ET T X . Then we are done by the properties of Chern
classes of rank 1 vector bundles.
Part (c) is also easy. Given i : U X , one checks that VD|U is the pullback
of VD via i. Thus (VD|U )T is the pullback of (VD )T via iT : ET T U ET T X .
Then we are done since c1 is compatible with pullbacks.
For part (d), recall that the map ET T X BT is a fibration with fiber X . If
iX : X ET T X is the inclusion of the fiber, then VD is the pullback of (VD )T
via iX . Since c1 is compatible with pullbacks, we obtain

iX [D]T = iX c1 (VD )T = c1 (iX (VD )T ) = c1 (VD ) = [D],
where the last equality uses (12.4.19).
The above proofs work over Z since we are assuming that D is Cartier. If D
is an arbitrary divisor on X , then D is Cartier for some integer > 0 since X is
simplicial. Then
[D]T = 1 [D]T H (X , Q)
is well-defined. It follows that properties (a), (c) and (d) hold over Q.
The proof of part (b) takes a bit more work. Take m M and set D = div( m ).
Also let p : X {pt} be the map that takes every element of X to the same point,
and let V m = C {pt} be the vector bundle over {pt} where T acts on C via m .
609
In Exercise 12.4.9 you will show that there is a pullback diagram

/ V m
VD

/ {pt}
of T -equivariant vector bundles. In concrete terms, this says that VD is the trivial
bundle X C X where T acts on C via m .
It follows that (VD )T is the pullback of (V m )T via pT : ET T X BT . For

the moment, set = c1 (V m )T HT2 ({pt}, Z) = T . Then

[div( m )]T = c1 (VD )T = pT c1 (V m )T = pT = 1
The last equality follows from the definition of HT (X , Z) as a T -module.
Hence it suffices to prove that = s(m), where s is defined in the discussion

leading up to (12.4.9). First observe that m : T C induces B m : BT BC .
By Exercise 12.4.9, there is a pullback diagram
(V m )T

BT
/ (C)C
B m

/ BC ,
where C {pt} where C acts on C by multiplication. This implies that

(12.4.20)
= c1 (V m )T = (B m ) c1 (C)C .
that (C)C = EC C C,
So we need to compute c1 ((C)C ) H 2 (BC , Z). Recall
S
where C acts on C by multiplication. Also, EC = 0 (C+1 \ {0}) and BC =
S
0 P = P . When computing equivariant cohomology in a specific degree, one

can always restrict to a finite-dimensional approximation. Hence, for sufficiently
large, c1 ((C)C ) is the Chern class of the rank 1 vector bundle
V = (C+1 \ {0}) C C P .
We have seen this bundle before. Since C acts on (C+1 \ {0}) C via t (p, ) =
(t 1 p,t ), the map (p, ) 7 ([p], p) P C+1 is constant on C orbits. Here
[p] denotes the point in P whose homogeneous coordinates are p C+1 \ {0}.
Hence we get a map
V P C+1 .
In Exercise 12.4.9, you will prove that the image of this map is the vector bundle
V P from Example 6.0.19. Thus V V as vector bundles over P . Since the
sheaf of sections of V is OP (1) by Example 6.0.21, we have

(12.4.21)
c1 (C)C = c1 (V ) = c1 (V ) = c1 (OP (1))
in the cohomology group
Z t = H 2 (P , Z) = H 2 (P , Z) = H 2 (BC , Z).
610
Here we are using H (P , Z) = Z[t]/ht +1 i and H (P , Z) = Z[t] from (12.4.7).

A standard fact about Chern classes is that
c1 (OP (1)) = t Z t.
This is the normalization axiom for Chern classessee [144, I.4]. It follows that
c1 (OP (1)) = t. Combining this with (12.4.20) and (12.4.21), we have

= c1 (V m )T = (B m ) c1 (C)C = (B m ) c1 (OP (1))
= (B m ) (t) = (B m ) (t)
= s(m),
where the final equality is the definition of s(m) in (12.4.9). This completes the
proof of Proposition 12.4.13.

The proof of the proposition explains the minus sign in Proposition 12.4.13it
comes from the Chern class of the vector bundle over BC = P determined by
the equvariant bundle C {pt} where C acts on C by multiplication.
Exercises for 12.4.
12.4.1. The following example is taken from [95].
(a) Let denote the complete fan in NR R2 with minimal generators
u1 = e1 , u2 = e2 , u3 = e1 , u4 = e2
and the complete fan with minimal generators
u1 = e1 , u2 = e2 , u3 = e1 + e2 , u4 = e1 e2 .
Use Theorem 12.4.1 to show that
H (X , Q) Q[x, y]/hxy, x 2 + y 2 i
H (X , Q) Q[x, y]/hxy, x 2 + 2y 2i.
(b) Deduce that the cohomology rings H (X , Q) and H (X , Q) are not isomorphic.
Hint: Show that H (X , Q) has no elements , of degree 2 satisfing 2 = 2 6= 0
and = 0.
(c) Show that if we extend scalars to R, however, then the cohomology rings become
isomorphic:
H (X , R) = H (X , Q) Z R H (X , Q) Z R = H (X , R).
The paper [95] shows that the same is true for all complete toric surfacesthe real
cohomology algebra depends only on the number of rays in the fan. This result does
not extend to higher-dimensional toric varieties.
12.4.2. In this exercise, you will show that C \ {0} is contractible.
(a) Show that C \ {0} is homotopy equivalentPto the unit sphere S C , where S
is the set of all (a0 , a1 , . . . ) C such that i=0 |ai |2 = 1.
(b) Show that the shift map T : S S defined by T (a0 , a1 , . . . ) = (0, a0 , a1 , . . . ) is

homotopic to the identity map id : S S . Hint: Consider the 1-parameter family
of maps (1 t) id + t T , t [0, 1], and note that T has no eigenvalues or eigenvectors.
611
(c) Finally, show that T is homotopic to the constant map U(a0 , a1 , . . . ) = (1, 0, . . .). Hint:
Consider the 1-parameter family
cos(t/2) T (a0 , a1 , . . .) + sin(t/2)U(a0 , a1 , . . .).
12.4.3. For each

Q (n 1), let m be a nonzero element of M and let f (T )Q
correspond to (n1) m SymQ (M). Prove that the map from (12.4.11) becomes
an isomorphism after localizing at f . Hint: Use (12.4.18).
12.4.4. Consider SRQ (r ) = Q[x1 , x2 , x3 , x4 ]/hx1 x3 , x2 x4 i, which for M = Z2 becomes a

SymQ (M)-algebra where e1 acts as x1 x3 and e1 acts as r x1 x2 + x4 .
(a) SRQ (r ) has graded ring automorphisms given by rescaling the variables by elements
of Q , plus the maps induced by x1 x3 , by x2 x4 , and by (x1 , x3 ) (x2 , x4 ). Show
that these generate the group of all graded ring automorphisms of SRQ (r )
(b) Prove that if SRQ (r ) SRQ (s ) as graded SymQ (M)-algebras, then r = s.

12.4.6. In Lemma 12.4.17, show that if (n 1) is a face of , (n), then the
component of the map corresponding to can be identified with the difference of the
restriction maps HT (U , Q) HT (U , Q) and HT (U , Q) HT (U , Q).
12.4.7. We will use the notation of Lemma 12.4.17. Let (n 1). Define the map
B : Q[ ] HT (U , Q) by xi 7 [Di U ]T for i (1). Also let N = Span( ) N N,
with dual M M = M/( N).
(a) As in the proof of Lemma 12.4.17, there are i > 0 and mi M such that i Di U =
div( mi ). Explain why mi is not unique and show that xi 7 1
i [mi ] (M )Q defines
an isomorphism B : Q[ ] SymQ (M ).
(b) Use Proposition 12.1.9 to show that HT (U , Q) HT (O( ), Q).
(c) Let x be the identity element of O( ). Show that TNt is the isotropy subgroup of
x O( ) under the action of T = TN .
(d) Show that N = N N , where N Z. Conclude that T = TN C , where TNt acts

trivially on O( ) and C acts freely and transitively on O( ).
(e) Suppose a product group G H acts on a space X such that G acts trivially and H acts
freely and transitively. Prove that HGH

(X, Q) HG ({pt}, Q). Hint: Use E(G H) =
EG EH and regard G H as acting on {pt} X.
(f) Explain why HT (O( ), Q) HTN({pt}, Q) and conclude that B is an isomorphism.

Hint: (12.4.16) holds for .
12.4.8. Verify the claims in Example 12.4.18.

12.4.9. In proof of part (b) of Proposition 12.4.13
(a) Prove that VD is the pullback of V m via p : X {pt}.
(b) Prove that (V m )T is the pullback of (C)C via B m : BT BC .
(c) Let C act on C by multiplication. Prove that (C+1 \ {0}) C C P is the vector
bundle V P described in Example 6.0.19.
12.4.10. In the situation of Theorem 12.4.1, take and write (1) = {i1 , . . . , id },
where d = dim . Then let
x = [xi1 xid ] RQ ()d .
612
In the text, was often an element of (n); here, it represents an arbitrary element of .
In this exercise you will prove that RQ () is spanned over Q by the x for . (The x
may be linearly dependent, however.)
(a) Show that

x x
x+
0
if = {0} and +
if +
/ .
(b) Now consider a monomial x a = xia11 xias s with i1 < i2 < < is and a j 1. We may
assume i1 + + is since otherwise [x a ] = 0 in RQ (). If a1 > 1, then use the
relations from J to show that [xi1 ] is a linear combination with coefficients in Q of the
[x ] for
/ {i1 , . . . , is } Hint: ui1 , . . . , uis are part of a basis of NQ .
(c) Write x a = xi1 xia11 1 xias s and replace the first xi1 with the linear combination found in
part (b). Then repeat part (b) for each term the resulting linear combination. Show that
this process eventually expresses [x a ] as a linear combination
Ps of cosets of square-free
monomials. This will prove our claim. Hint: Induction on i=1 (ai 1).
12.5. The Chow Ring and Intersection Cohomology

Here we explore two alternative descriptions of the cohomology ring of a complete
simplicial toric variety. The first involves the Chow ring, while the second uses
intersection cohomology.
The Chow Ring. We will show that the description of the cohomology ring given
in Theorem 12.4.1 applies equally well to the rational Chow ring A (X )Q of a
complete simplicial toric variety X . We begin by sketching the construction of
the Chow ring. We refer the reader to [107] for the details.
Let Zk (X ) denote the group of k-dimensional algebraic cycles on a variety X .
In other words, Zk (X ) is the additive group of finite formal linear combinations
X
ni [Vi ]
i
of irreducible subvarieties Vi of dimension k with coefficients ni Z. A k-cycle

on X is defined to be rationally equivalent to zero if there are finitely many
(k + 1)-dimensional irreducible subvarieties Wi X and nonzero rational functions
fi C(Wi ) such that
X
=
[divWi ( fi )],
i
where divWi ( fi ) is the divisor of the rational function fi on Wi . Note that Wi may
fail to be normal, so this requires a more general definition of div( fi ) than the one
given in 4.0. The cycles rationally equivalent to zero form a subgroup Ratk (X ) of
Zk (X ) (Exercise 12.5.1). Then the group of k-cycles modulo rational equivalence
is defined to be
Ak (X ) = Zk (X )/Ratk (X ).
613
When X is normal of dimension n, An1 (X ) is the group Cl(X ) of Weil divisors

modulo linear equivalence defined in 4.0 (Exercise 12.5.1).
For a general n-dimensional variety X , the group of cycles of codimension k
modulo rational equivalence is defined to be Ak (X ) = Ank (X ). If in addition X is
smooth and projective, then with a substantial amount of work (see [107, Ch. 8]),
it can be shown that there is a product
Ak (X ) A(X ) Ak+ (X ),
which coincides with the geometric intersection of cycles in the case of transverse
intersections. We will write the product as for Ak (X ) and A (X ).
It can be shown that this
product is commutative and associative, and
Ln intersection
k
this makes A (X ) = k=0 A (X ) into a graded ring, called the Chow ring of X .
A (X ) can be viewed as a sort of algebraic version of the integral cohomology ring
H (X , Z). The intersection of cycles corresponds to cup product in cohomology,
so there is a ring homomorphism
(12.5.1)
A (X ) H (X , Z)
defined as follows. Let n = dim X . Then a cycle Ak (X ) = Ank (X ) gives a

homology class in H2n2k (X , Z) whose Poincare dual is in H 2k (X , Z). In particular,
the map (12.5.1) doubles degrees.
The Toric Case. If X is a complete simplicial toric variety of dimension n, then
the intersection product can be defined on rational cycles, making
n
M
Ak (X ) Z Q
A (X )Q = A (X ) Z Q =
k=0
into a graded ring. The key point here is that X has a covering by open affine
varieties of the form Cn /G for some finite group G. In this case,
A (Cn /G)Q A (Cn )G
Q,
the G-invariant subring. As in the smooth case, each subvariety of codimension
k in X defines a class in H2n2k (X , Q), hence by Poincare duality, a class in
H 2k (X , Q). Hence we also have a ring homomorphism
(12.5.2)
A (X )Q H (X , Q).
For each , the orbit closure V () is a subvariety of codimension dim .

Write [V ()] for its rational equivalence class in Adim (X ).
Lemma 12.5.1. The [V ()] for generate A (X ) as an abelian group.
Proof. We will show that the [V ()] for of dimension n k generate Ak (X ). In
general, when Y X is a closed subvariety of a variety X , there is a short exact
sequence of Chow groups
(12.5.3)
Ak (Y ) Ak (X ) Ak (X \Y ) 0,
614
where the first map comes from the inclusion Y X , and the second comes by
restriction. See [107, Prop. 1.8]) for a proof. If X is normal of dimension n, then
(12.5.3) for k = n 1 is precisely Theorem 4.0.20 (Exercise 12.5.2).
Consider the filtration of X as in (12.3.3),
= X1 X0 X1 Xn = X ,
where X p is the union of the closures V () of the torus orbits of dimension p. By
the Orbit-Cone Correspondence, we have
a
X p \ X p1 =
O().
(np)
Then (12.5.3) gives an exact sequence

Ak (X p1 ) Ak (X p )
(np)
Ak (O()) 0.
But O() is a p-dimensional torus when (n p) and hence can be viewed as

an affine open subset of C p . In Exercise 12.5.3 you will show that
(
Z[C p ] k = p
Ak (C p ) =
0
otherwise.
Using (12.5.3) for the inclusion O() C p , we see that
(
Z[O()] k = p
Ak (O()) =
0
otherwise.
When p > k, it follows that the above exact sequence simplifies to
Ak (X p1 ) Ak (X p ) 0.
Now fix k and let p increase from k to n. It follows that Ak (Xk ) surjects onto
Ak (X ). By Exercise 12.5.2, we also have
M
Ak (Xk ) =
Z[V ()]
S
(nk)
since Xk = (nk) V () and each V () is irreducible of dimension k. The

lemma now follows easily.

When (1), the orbit closure V () is the divisor denoted D . For m M,
recall from Proposition 4.1.2 that
P
div( m ) = hm, u iD .
Since A1(X ) = Cl(X ) by Exercise 12.5.2, we obtain
P
1
(12.5.4)
hm, u i[D ] = 0 in A (X ).
Here is an example of how to compute intersection products in the Chow ring.
615
Lemma 12.5.2. Assume that X is complete and simplicial. If 1 , . . . , d (1)

are distinct, then in A (X )Q , we have
(
1
[V ()] if = 1 + + d
[D1 ] [D2 ] [Dd ] = mult()
0
otherwise.
Proof. Let (1) and be a cone of not containing . We will show that
( mult()
[V ( )] if = +
(12.5.5)
[D ] [V ()] = mult( )
0
otherwise.
In Exercise 12.5.4 you will prove that the lemma follows from (12.5.5). You should
also check that (12.5.5) generalizes the intersection formulas from parts (a) and (b)
of Proposition 6.4.4.
We noted earlier that intersection products can be subtle. Life is easier when
intersecting with a divisor, but even here, some care is needed. The relevant theory
is developed in [107, 2.3]. We will use the following special case. Let D be a
Cartier divisor on X and let i : V X be a normal subvariety of dimension k.
The line bundle OX (D) pulls back to a line bundle i OX (D) on V , so that by
Theorem 6.0.20, i OX (D) OV (D ) for some Cartier divisor D on V . Then D is
a (k 1)-cycle on X and by [107, 2.3] gives the intersection product
(12.5.6)
[D] [V ] = [D ].
In Exercise 12.5.5 you will use this definition to show that [D] [V ] = 0 when
Supp(D) V = .
Since X is simplicial, D is Q-Cartier, so there is some positive multiple D =

D that is Cartier on X . If +
/ , then Supp(D) V () = D V () = by
the Orbit-Cone Correspondence. Then [D] [V ()] = 0 by the previous paragraph,
which implies [D ] [V ()] = 0. Hence we may assume that = + . We
need to prove that
(12.5.7)
[D] [V ()] =
mult()
[V ( )].
mult( )
The first step is to find a Cartier divisor D on V () such that i OX (D)

OV () (D ), where i : V () X is the inclusion. We will use the support function
D : NR R of D, which satisfies
(
if =
D (u ) =
0
otherwise.
Note in particular that D vanishes on Span() since
/ (1).
Recall from Proposition 3.2.7 that V () is the toric variety of the fan Star()
in N()R , where N() = N/(Span() N). The previous paragraph shows that D
616
factors through the quotient map NR N()R . By Exercise 12.5.6, the induced
map N()R R is the support function D of the divisor D we want.
The rays of Star() are images of rays of , so that D vanishes on all rays of
Star() except for = + = . Thus D is the divisor on V () given by
D = D (u )D ,
where u is the minimal generator of N(). In particular, the image of u in

N() is u = u for some integer . Thus
D (u ) =
1
1
1
D ( u ) =
D (u ) =
D (u ) = .
Since = mult( )/mult() by the proof of Lemma 6.4.2, we obtain

D =
mult()
D .
mult( )
However, since = + , the divisor D in V () is the subvariety V ( ) in V ().

From here, (12.5.7) follows easily, and the proof is complete.

Similar to what happened in 6.4, Lemma 12.5.2 shows that rational numbers
appear naturally when doing intersection theory on simplicial toric varieties.
The Chow Ring of a Toric Variety. As in 12.4, write (1) = {1 , . . . , r }. This
gives the ring
RQ () = Q[x1 , . . . , xr ]/(I + J )
for I and J as in (12.4.2) and (12.4.3). Then Lemma 12.5.2 and (12.5.4) imply
that [xi ] 7 [Di ] A1 (X )Q defines a ring homomorphism
(12.5.8)
RQ ()Q A (X )Q .
We also have the ring homomorphism A (X )Q H (X , Q) from (12.5.2).

Theorem 12.5.3. If X is complete and simplicial, then
RQ ()Q A (X )Q H (X , Q),
where the maps are given by (12.5.8) and (12.5.2).
Proof. The composition of the maps RQ ()Q A (X )Q H (X , Q) from
(12.5.8) and (12.5.2) is given by [xi ] 7 [Di ] H 2 (X , Q). Since this defines the
isomorphism RQ ()Q H (X , Q) from Theorem 12.4.1, it follows that the map
RQ ()Q A (X )Q is injective.
For surjectivity, take = i1 + + id . By Lemma 12.5.2, [xi1 xid ] =

[xi1 ] [xid ] maps to the product
[Di1 ] [Did ] =
1
[V ()].
mult()
617
Then surjectivity follows since the [V ()] span A (X )Q by Lemma 12.5.1. Hence
(12.5.8) is an isomorphism, and then the same must be true for (12.5.2).

The isomorphism here explains the connection between the cup product in
cohomology and the intersection formula from (12.4.1). We should also note that
there are isomorphisms
R() A (X ) H (X , Z)
(12.5.9)
when X is smooth, where R() is the ring from Theorem 12.4.4.

There is also an equivariant Chow ring AT (X ) associated to the action of
T = TN on X . If X is not simplicial, then Payne has shown in [224] that AT (X )
still coincides with the ring of continuous piecewise polynomial functions on the
polyhedral decomposition of NR defined by , even over Z. However, this ring
need not coincide with equivariant cohomology ring in this case.
An Exact Sequence
L for the Chow Group. The proof of Lemma 12.5.1 gives a
surjective map : (nk) Q[V ()] Ak (X )Q . In [111], Fulton and Sturmfels
describe generators of the kernel using the sequence
M
M
(12.5.10)
M( )Q
Q[V ()] Ak (X )Q 0.
(nk1)
(nk)
To define , take (n k 1), so dim V ( ) = k + 1, and note that M( ) is the

character group of the torus of the toric variety V ( ). If m M( ), then m is a
rational function on V ( ) and hence div( m ) is a divisor on V ( ). This gives the
k-cycle (m) = [div( m )] on X . In Exercise 12.5.7 you will show that
X
(12.5.11)
(m) =
hm, u, i[V ()],
(nk),
where u, generates N /N Z. Then the definition of rational equivalence

guarantees that (12.5.10) is a complex, i.e., = 0.
Theorem 12.5.4. The complex (12.5.10) is exact if X is complete and simplicial.
Proof. Since is surjective, (12.5.10) gives a surjective map coker() Ak (X )Q .
Now consider the dual of , which can be written
M V1
M V0

M( )Q
M()Q
(nk1)
(nk)
Since dim M()Q = k and dim M( )Q = k + 1, we have

V
V
V0
M()Q k M()Q Q k N()Q
V1
M( )Q )
Vk
M( )Q Q
Vk+1
N( )Q .
618
Picking bases of N() and N( ) and thinking in terms of orientation coefficients,

one sees that can be identified with the map
M Vk
M Vk
V
V
k
C k (, k )Q =
M()Q C k+1 (, k )Q =
M( )Q ,
(nk)
(nk1)
Vk
where C (, ) is the complex defined in the discussion preceding Lemma 12.3.3.

V
p,q
p,q
The spectral sequence E1 H p+q (X , Q) from (12.3.11) has E1 = C p (, q )Q
V
by (12.3.12). Since E1k1,k = C k1 (, k ) = 0 for dimension reasons, we obtain
E2k,k = ker(E1k,k E1k+1,k ) = ker( k ) = ker( ).
Theorem 12.3.11 implies that E2k,k H 2k (X , Q), which in turn is isomorphic to

H 2n2k (X , Q) Ank (X )Q by Poincare duality and Theorem 12.5.3. Thus
dim coker() = dim ker( ) = dim Ank (X )Q = dim Ak (X )Q .
It follows that the surjective map coker() Ak (X )Q is an isomorphism. This

proves that (12.5.10) is exact.

The result proved in [111, Prop. 2.1] is stronger than Theorem 12.5.4 since it
holds over Z and applies to arbitrary toric varieties. Note also that when k = n 1,
the exact sequence (12.5.10) becomes the familiar exact sequence
L
MQ Q[D ] An1 (X )Q = Cl(X )Q 0
from Chapter 4. For general k, you should check that (12.5.10) can be regarded as
the purely toric version of rational equivalence.
Intersection Cohomology. As we saw in 12.3, most of the nice properties of

the cohomology ring H (X , Q) for simplicial X (the vanishing of the odd-degree
groups, Poincare duality, and so forth) can fail for nonsimplicial toric varieties. The
theory of intersection homology and cohomology was first developed by Goresky
and MacPherson in the 1970s to study spaces with singularities more complicated
than finite quotient singularities, and many of the nice properties are recovered in
this cohomology theory. See [175] for a general introduction.
Intersection homology was originally defined for spaces with a special sort of
stratification (see [119] and [120]). In the following, cone(S) is the open topological cone over S, namely the quotient space of S [0, 1) modulo the relation that
identifies all (s, 0) for s S.
Definition 12.5.5. A topological pseudomanifold X is a paracompact Hausdorff
space with a filtration
X0 X1 Xn = X
such that:
619
(a) For each point p in Xi \ Xi1 , there is a neighborhood N of p in X , a compact

Hausdorff space L with an n i 1 dimensional topological stratification
= L1 L0 Lni1 = L,
and a homeomorphism : Ri cone(L) N that maps each Ri cone(L j )

homeomorphically to N Xi+ j+1 .
(b) Xn1 = Xn2 .
When the first condition holds, the second condition implies that X \ Xn2 is
an n-dimensional real manifold. It is known that all projective or quasiprojective
varieties over C are topological pseudomanifolds.
A perversity is a function p : {2, 3, . . .} N such that p(2) = 0 and both p(k)
and k 2 p(k) are nonnegative and nondecreasing functions of k. The complementary perversity of p is q(k) = k 2 p(k). The constant function min(k) = 0 is
the minimal perversity; its complement is the maximal perversity max(k) = k 2.
The middle perversity is p(k) = k2
2 . Perversities are used to specify allowable
intersections of cycles with the strata Xi in the definition of intersection homology.
Definition 12.5.6. The intersection homology groups IHip (X ) of a topological
pseudomanifold X with respect to a given perversity p are the homology groups
of a subcomplex ICp (X ) of the singular chains on X as follows. A singular isimplex : i X is said to be p-allowable if for every k, 1 (Xnk \ Xnk1 ) is
contained in the union of the faces of i of dimension i k + p(k). Then a singular
chain is said to be p-allowable if it is a linear combination of p-allowable singular
simplices. Finally, ICip (X ) consists of all p-allowable i-chains c such that c is
also p-allowable.
Following an idea of Deligne, this definition was later reformulated using the
derived category of the category of sheaves of Z-modules on X (see [120]). This
way of framing the theory has numerous technical advantages, but we will not
pursue this approach here. The definition of the IHip (X ) appears to depend on the
stratification, but Goresky and MacPherson proved that the IHip (X ) are actually
topological invariants of X .
Example 12.5.7. Let X be an irreducible variety of dimension n with a single
isolated singular point x. For the middle perversity, it is not difficult to see that
i>n
Hi (X )
IHi (X ) = im(Hi (X \ {x}) Hi (X )) i = n
Hi (X \ {x})
i < n.
You will prove this in Exercise 12.5.8.
The following theorem summarizes the properties of intersection homology

that we will discuss.
620
Theorem 12.5.8. Let X be a normal projective variety of dimension n.

(a) IHip (X ) is finitely generated for any i and any perversity p.
(b) For the minimal and maximal perversities, we have
IHimin (X ) H 2ni (X , Z),
IHimax (X ) Hi (X , Z).
Now let p be the middle perversity and define the intersection cohomology groups
p
as IH i (X )Q = IH2ni
(X ) Z Q. Then:
(c) Poincare duality holds for the intersection cohomology. In particular,
dim IH i (X )Q = dim IH 2ni (X )Q .
(d) IH (X )Q is a module over the cohomology ring H (X , Q).
(e) Let H 2 (X , Q) be the class of an ample divisor on X . Then for all 0 i n,
multiplication by ni defines an isomorphism
ni : IH i (X )Q IH 2ni (X )Q .
This is the hard Lefschetz theorem.
(f) Let be as in part (e). Then for 0 i < n, multiplication by
: IH i (X )Q IH i+2 (X )Q
is injective.
See [120] for parts (a)(d) of this theorem and [23] for the hard Lefschetz
theorem stated in part (e). Note that part (f) follows from part (e).
The intersection cohomology IH (X )Q does not have a natural ring structure.
However, when X is a complete simplicial toric variety, Comment 6.4 in [119]
implies that the odd-degree intersection cohomology groups are zero and that
(12.5.12)
IH 2i (X )Q H 2i (X , Q) Ai (X )Q ,
i = 0, . . . , n.
Applications to Polytopes. In 1980, Stanley [257] used toric varieties and their
cohomology to prove the McMullen conjecture for a simplicial polytope P. His
proof applied hard Lefschetz to the toric variety X , where is the fan consisting
of the cones over the proper faces of P. This is the normal fan of the dual polytope
P , which is simple. Hence X comes from a simplicial fan and we know hard
Lefschetz is true in this case by (12.5.12) and part (e) of Theorem 12.5.8. As
explained in Appendix A, Stanleys paper was an important event in the history of
toric geometry.
Let f0P , . . . , fnP denote the face numbers of an n-dimensional polytope P. If
P is simplicial, then (9.4.6), Exercise 9.4.10 and Theorem 12.3.12 imply that the
numbers

n
n
X
X
ik i
ik i
P
(1)
(1)
(12.5.13)
hk (P) =
fni1 =
fiP
k
k
i=k
i=k
621
are the even Betti numbers b2k = dim H 2k (X , Q) of X . Hence they satisfy the
Dehn-Sommerville equations hk (P) = hnk (P) by Poincare duality for the simplicial toric variety X . Also observe that
1 = h0 (P) h1 (P) hm (P),
m = n2
by the injectivity of part (f) of Theorem 12.5.8. This monotone property of the
hk (P) is used by Stanley in [257].
For a more general (i.e., nonsimplicial) polytope P, some of the hk (P) in
(12.5.13) can be negative, so a different approach is necessary if we want to relate
these numbers to cohomology groups. Following [258], we define polynomials
h(P,t) and g(P,t) recursively as follows. Let g(,t) = 1 and dim = 1. Now
assume that g(Q,t) has been defined for all polytopes Q of dimension < n. Then,
for a polytope P of dimension n, define
X
h(P,t) =
g(Q,t) (t 1)n1dim Q ,
QP
where is sum is over all proper faces Q of P, including Q = . Then write h(P,t) =
P
n
k
k=0 hk (P)t and define
n/2
g(P,t) =
X
k=0
(hk (P) hk1 (P))t k .
In particular, the numbers hk (P) are defined for any polytope P, and one can show
that they agree with (12.5.13) when P is simplicial. We call
h(P,t) =
n
X
hk (P)t k
k=0
the h-polynomial of P. Here is a result from [258].

Theorem 12.5.9. Let P be a full dimensional lattice polytope in Rn containing the
origin as an interior point and let be the fan whose cones are the cones over the
proper faces of P. Then:
(a) The hk (P) are the intersection cohomology Betti numbers of X , i.e.,
hk (P) = dim IH 2k (X )Q .
(b) hk (P) = hnk (P) for k = 0, . . . , n.
(c) Let m = n2 . Then
1 = h0 (P) h1 (P) hm (P).
The equalities in part (b) of the theorem generalize the Dehn-Sommerville

equations. They follow from part (a) and Poincare duality for intersection cohomology. Similarly, part (c) follows from part (f) of Theorem 12.5.8, which in turn
is a consequence of hard Lefschetz.
622
Motivated by this application to polytopes, a purely combinatorial version of

intersection cohomology for toric varieties has been studied by many researchers,
including Barthel, Brasselet, Bressler, Fieseler, Karu, Kaup and Lunts. See [42]
for a discussion of this topic and references to the original papers.
Exercises for 12.5.
12.5.1. This exercise will study rational equivalence on a variety X of dimension n.
(a) Show that Ratk (X) is a subgroup of Zk (X) for all k.
(b) When X is normal, prove that two (n 1) cycles on X are rationally equivalent if and
only if they are linearly equivalent as defined in 4.0. Conclude that An1 (X) = Cl(X).
(c) Explain why An1 (X) = Pic(X) when X is smooth.
12.5.2. Let X be a variety of dimension n.
(a) In general, X may be reducible and may have irreducible
L components of dimension
strictly smaller than n. Prove that An (X) = Zn (X) = Y Z[Y ], where the sum is over
all irreducible components Y X of dimension n.
(b) Assume in addition that X is normal. Then show that the exact sequence (12.5.3) is
Theorem 4.0.20 when k = n 1.
12.5.3. Show directly from the definitions that the Chow groups of affine space are
(
Z[C p ] if k = p
p
Ak (C ) =
0
otherwise.
12.5.4. Prove that Lemma 12.5.2 follows from (12.5.5).
12.5.5. Use (12.5.6) to prove that [D] [V ] = 0 when Supp(D) V = . Hint: Explain why
Supp(D) V = implies that i OX (D) is trivial on V .
12.5.6. In the proof of Lemma 12.5.2, we claimed that D : NR R is the composition of

NR N()R and D : N()R R. Prove this. Hint: Adapt the proof of Proposition 6.2.7.
12.5.7. Prove (12.5.11) when (n k 1) and m M( ).
12.5.8. Use the definition of intersection homology to prove the formula for IH i (X) given
in Example 12.5.7.
12.5.9. Let X be a complete smooth variety of dimension n. Given a Cartier divisor D and
a smooth curve C on X, we defined the intersection number D C in 6.3. In this exercise
you will use (12.5.6) to prove that
Z
D C = [D] ` [C],
X
where [D] and [C] are the corresponding cohomology classes in H (X, Z).
R
(a) Explain why a point p X gives a class [p] H 2n (X, Z) such that X [x] = 1.

Ps
(b) Let i : C X be the inclusion map and suppose that i OX (D) = OC
i=1 i pi for
Ps
Ps
i Z and pi C. Explain why D C = i=1 i and [D] [C] = i=1 i [pi ] in A (X).
R
(c) Use the ring homomorhpism A (X) H (X, Z) to prove D C = X [D] ` [C].
12.5.10. Explain how Exercise 12.4.10 relates to the results of this section.
Chapter 13
Toric HirzebruchRiemann-Roch
The Hirzebruch-Riemann-Roch theorem (called HRR) shows how to compute the

Euler characteristic of a coherent sheaf F on a smooth complete n-dimensional
variety X using intersection theory. The formula is
Z
(13.0.1)
(F ) = ch(F ) Td(X ).
X
The term on the left is the Euler characteristic of F from Chapter 9, defined by
(F ) = (X , F ) =
n
X
(1)i dim H i(X , F ).
i=0
The right-hand side of (13.0.1) introduces two objects we have not seen before, the
Chern character ch(F ) H (X , Q) and the Todd class Td(X ) H (X , Q). The
integrand on the right-hand side of (13.0.1) is the cup product ch(F ) ` Td(X ),
which is written ch(F )Td(X ) or ch(F ) Td(X ) in practice.
This chapter will prove HRR for a line bundle L = OX (D) on a smooth
complete toric variety X = X . Our strategy will be to first prove an equivariant
Riemann-Roch theorem for X and then derive HRR via the nonequivariant limit.
In 13.1, we begin by defining the terms involved in (13.0.1) and working

out some specific examples of HRR. To prepare for the equivariant version of
Riemann-Roch, we discuss some remarkable equalities due to Brion in 13.2. Then
13.3 proves equivariant Riemann-Roch in the smooth case, following the paper
[50] of Brion and Vergne. The final two sections 13.4 and 13.5 apply these ideas
to lattice points in polytopes, including the volume polynomial and the KhovanskiiPukhlikov theorem. We will also sketch a second proof of HRR.
623
Chapter 13. Toric Hirzebruch-Riemann-Roch
624
13.1. Chern Characters, Todd Classes, and HRR

In order to state our version of the toric HRR, we need to define Chern characters
and Todd classes. Both of these require knowledge of Chern classes, which we
now discuss.
Chern Classes. A locally free sheaf E of rank r on a variety X has Chern classes
ci (E ) H 2i (X , Z) for 0 i r. Its total Chern class is c(E ) = c0 (E )+ + cr (E ).
Here are the key properties we will need:
c0 (E ) = 1 for any E and c(OX ) = 1 for any X .
ci ( f E ) = f ci (E ) when f : X Y is a morphism and E is locally free on Y .
c(F ) = c(E ) c(G ) when 0 E F G 0 is an exact sequence.
ci (E ) = (1)i ci (E ), where E is the dual of E .

Vr
c1 (
E ) = c1 (E ) when E has rank r.
If D is a Cartier divisor on a smooth complete variety X , then

c1 (OX (D)) = [D] H 2 (X , Z).
We refer the reader to [107] or [144] for a further discussion of Chern classes. Later
in the section we will say a few words about Chern classes of coherent sheaves.
Here is a nice example that shows how to use the properties of Chern classes.
Example 13.1.1. Let X = X be a smooth complete toric variety. Its cotangent
bundle 1X fits into the generalized Euler sequence
L
(13.1.1)
0 1X OX (D ) Pic(X ) Z OX 0
from Theorem 8.1.6. Since Pic(X ) Z OX OXrn for r = |(1)|, the total Chern
class of Pic(X ) Z OX is 1. Hence
Q
Q
L
c(1X ) = c
(1 [D ]),
c(OX (D )) =
OX (D ) =
Q
P
where means cup product in H (X , Z). Thus c1 (1X ) = [ D ].
Let dim X = n. Then the canonical bundle X = Xn has first Chern class
V
P
c1 (X ) = c1 ( n 1X ) = c1 (1X ) = [ D ].
P
This gives a Chern class proof that X has canonical divisor KX = D .
The Chern classes of the tangent bundle TX of a smooth complete variety X

are called the Chern classes of X . We often write ci = ci (TX ). As we will soon
see, the Todd class of X is a special polynomial in the ci . The top Chern class is
cn = cn (TX ), n = dim X , which is usually called the Euler class of X . The name
comes from the well-known formula
Z
cn (TX ) = e(X ),
(13.1.2)
X
625
P
i
i
where e(X ) = 2n
i=0 (1) rank H (X , Z) is the topological Euler characteristic
of X . See, for example, [144, Thm. 4.10.1].
In the toric case, the Euler class is easy to compute. Recall from (12.5.9) that
H (X , Z) is isomorphic to the Chow ring A (X ) when X is a smooth complete toric
variety. In particular, a point of X gives a class [pt] An (X ) = H 2n (X , Z), where
n = dim X . Then we have the following result.
Proposition 13.1.2. Let X = X be a smooth complete n-dimensional toric variety
with tangent bundle TX . Then:
Q
P
(a) c(TX ) = (1 + [D ]) = [V ()].
P
(b) c1 = c1 (TX ) = [ D ] = [KX ].
(c) cn = cn (TX ) = [pt], where = |(n)|.
Proof. The first equality of part (a) follows easily by taking the dual of the exact
sequence (13.1.1) and applying the method of Example 13.1.1. Next observe that
Lemma 12.5.2 implies that if 1 , . . . , n (1) are distinct, then
(
[V ()] = 1 + + n (n)
[D1 ] [Dn ] =
0
otherwise,
Q
since X is smooth. Then multiplying out (1 + [D ]) gives the second equality.
Part (b) follows, and for part (c), note that if (n), then V () is a point, so
that [V ()] = [pt]. From here, the formula for cn follows immediately.
R
Since X [pt] = 1, (13.1.2) and Proposition 13.1.2 imply that the topological
Euler characteristic of X is
Z
Z
e(X ) = cn (TX ) = [pt] = = |(n)|.
X
This agrees with the result proved in Theorem 12.3.9.

Using Chern classes, it is easy to define the Chern character of a line bundle.
Definition 13.1.3. Let X be a variety of dimension n. The Chern character of a
line bundle L is
(13.1.3)
ch(L ) = 1 + c1 (L ) +
c1 (L )2 c1 (L )3
+
+ H (X , Q).
2!
3!
Truncating the Taylor series for e x in degrees greater than n and then substituting
c1 (L ) H 2 (X , Q) for x shows that we may write this more compactly as e c1 (L ) .
We will see later that Chern characters can also be defined for coherent sheaves.
626
The Todd Class. To define the Todd class Td(X ) of a smooth complete variety X
of dimension n, we begin by introducing symbolic objects 1 , . . . , n such that
c(TX ) =
n
Y
(1 + i ).
i=1
We regard the i as living in H 2 (X , Q). The i are the Chern roots of the tangent
bundle TX . In practical terms, this means that ci = ci (TX ) = i (1 , . . . , n ), where
i is the ith elementary symmetric polynomial.
Definition 13.1.4. Let X be a smooth complete n-dimensional variety and let
1 , . . . , n be the Chern roots of its tangent bundle. Then the Todd class of X is
Td(X ) =
n
Y
i=1
i
.
1 ei
When we expand the right-hand side of this formula in H (X , Q), we get a

symmetric polynomial in the i with rational coefficients. The theory of symmetric
polynomials implies that this is a polynomial expression in ci = i (1 , . . . , n ) for
1 i n. Hence the Todd class of X is a polynomial in its Chern classes c1 , . . . , cn .
The key fact is that the same polynomial is used for all smooth complete varieties
of dimension n.
To compute these polynomials, we use the well-known power series expansion
X
x
Bk 2k
1
1
1
1 4
(13.1.4)
=1+ x+ (1)k1
x = 1+ x+ x 2
x + ,
x
1 e
2
(2k)!
2
12
720
k=1
1
where B1 = 61 , B2 = 30
, B3 =
Here is a classic example.
1
42 , . . .
are the Bernoulli numbers (see [144, p. 13]).
Example 13.1.5. A smooth complete surface X has Chern classes c1 = 1 + 2 and

c2 = 1 2 , where 1 , 2 are the Chern roots. Since the Chern roots have degree 2
and all monomials in 1 , 2 of total degree 3 vanish (X is a surface), we obtain
1
1
1
1
Td(X ) = (1 + 1 + 12 )(1 + 2 + 22 )
2
12
2
12
1
1 2
= 1 + (1 + 2 ) + (1 + 3 1 2 + 22 )
2
12
1 2
1
= 1 + c1 + (c1 + c2 ).
2
12
When X has dimension n, one can show that Td(X ) = T0 + + Tn such that T
is a weighted homogeneous polynomial of weighted degree in c1 , . . . , c , provided
627
ci has weight i. We call Ti the ith Todd polynomial. The first few are
1
1
1
c1 c2
T0 = 1, T1 = c1 , T2 = (c12 + c2 ), T3 =
2
12
24
1
T4 =
(c14 + 4 c12 c2 + 3 c22 + c1 c3 c4 ).
720
You will compute T3 in Exercise 13.1.1. See [144, pp. 1314] for more on Todd
polynomials. At the end of the section we will say more about the intuition behind
Definition 13.1.4.
The Todd Class of a Smooth Toric Variety. In the toric setting, the cohomology
ring is generated by the classes [D ] for (1) by Theorems 12.4.1 and 12.4.4.
The Todd class can be expressed in terms of these classes as follows.
Theorem 13.1.6. The Todd class of a smooth complete toric variety X = X is
Y
[D ]
Td(X ) =
H (X , Q).
1 e[D ]
(1)
Proof. Recall that c(TX ) = (1 + [D ]) by Proposition 13.1.2. If we set (1) =

{1 , . . . , r } and Di = Di , then
ci = i ([D1 ], . . . , [Dr ]).
On the other hand, the Chern roots i satisfy ci = i (1 , . . . , n ), where n = dim X .

So [D1 ], . . . , [Dr ] behave like the Chern roots, except that r > n since r = |(1)|
and is a complete fan in NR Rn .
Q
Q
i]
purely
We can avoid this problem by computing ni=1 1eii and ri=1 1e[D[D
i]
symbolically. We first consider the formal power series ring Q[[x1 , . . . , xr ]] with
maximal ideal m = hx1 , . . . , xr i. Then there is a polynomial P of n variables such
that
r
Y

xi
(13.1.5)
P 1 (x1 , . . . , xr ), . . . , n (x1 , . . . , xr ) mod mn+1 .
x
i
1e
i=1
We can stop at n since i (x1 , . . . , xr ) mn+1 for i > n. Setting xi = [Di ], we obtain
r
Y
[Di ]
= P(c1 , . . . , cn ).
1 e[Di ]
i=1
Now let n = hx1 , . . . , xn i Q[[x1 , . . . , xn ]]. Since 1exixi = 1 + 12 xi + equals 1

when xi = 0, we see that setting xi = 0 for n + 1 i r in (13.1.5) gives the
identity
n
Y

xi
P 1 (x1 , . . . , xn , 0, . . . , 0), . . . , n (x1 , . . . , xn , 0, . . . , 0) mod nn+1
x
i
1e

i=1
P 1 (x1 , . . . , xn ), . . . , n (x1 , . . . , xn ) mod nn+1
628
in Q[x1 , . . . , xn ]. The substitution xi = i gives

n
Y
i=1
i
= P(c1 , . . . , cn ),
1 ei
and the theorem is proved.
The HRR Theorem. Now that we have defined Chern characters and the Todd
class, we can begin to play with HRR. If D is a divisor on a smooth complete
variety X , then HRR gives the equality
Z
(OX (D)) = ch(OX (D)) Td(X )
X
Here is what this looks like for a surface.

Example 13.1.7. Let X be a smooth complete surface, so that
1
1
Td(X ) = 1 + c1 + (c12 + c2 ).
2
12
Also recall that c1 = [KX ], where KX is the canonical divisor.
We first apply HRR to D = 0. Since ch(OX ) = 1, we obtain
Z
Z
Z

1
1
1
1 + c1 + (c12 + c2 ) =
(OX ) = Td(X ) =
c12 + c2
2
12
12
X
X
X
R
since X is trivial away from the top degree. It follows that the formula for Td(X )
can be written
1
(13.1.6)
Td(X ) = 1 [KX ] + (OX )[pt].
2
R 2 R
Furthermore,
since X c1 = X [KX ]2 = (KX ) (KX ) = KX KX by (12.4.1)
R
and X c2 = e(X ) by (13.1.2), the above formula for (OX ) simplifies to

1
KX KX + e(X ) ,
12
which is Noethers theorem (Theorem 10.5.3). This result expresses the topological
invariant e(X ) in terms of the algebro-geometric invariants (OX ) and KX KX .
(OX ) =
In Exercise 13.1.2 you will show more generally that HRR applied to OX (D)
gives the formula
1
(OX (D)) = (D D D KX ) + (OX ).
2
This is the version of Riemann-Roch stated in Theorem 10.5.2.
(13.1.7)
It should be no surprise that the Riemann-Roch theorem for smooth complete

curves is also a special case of HRR. See Exercise 13.1.3.
629
Example 13.1.8. The Hirzebruch surface X = H2 was studied in Example 4.1.8.

In particular, the Picard group Pic(X ) is generated by the classes of D3 and D4 .
We will use the Riemann-Roch theorem to compute (OX (D)) for the divisor D =
3D3 5D4 on X . Example 4.1.8 shows that D1 D3 and D2 2D3 + D4 . Thus
KX = (D1 + D2 + D3 + D4 ) = 2D4 .
The intersection pairing on X = H2 was determined in Example 10.4.6, which
showed that
D32 = 0, D3 D4 = 1, D42 = 2.
Also note that (OX ) = 1 by Demazure vanishing. Then applying Riemann-Roch
from (13.1.7) to the divisor D = 3D3 5D4 , we obtain
(OX (D)) = 12 D (D KX ) + 1
= 12 (3D3 5D4 ) (3D3 3D4 ) + 1
= 21 (9D32 24D3 D4 + 15D42 ) + 1 = 4.
In Example B.7.1 we show how to do this computation using Sage [262].
Grothendieck-Riemann-Roch. Hirzebruch proved his version of Riemann-Roch

for vector bundles on smooth projective varieties in 1954. This was generalized by
Grothendieck in 1957 to proper morphisms f : X Y between smooth quasiprojective varieties and coherent sheaves on X . His proof was published by Borel
and Serre in 1958. Botts wonderful review [41] of their paper points out that
Grothendiecks theorem explains why the Todd class from Definition 13.1.4 makes
sense. For this reason, we will briefly discuss Grothendieck-Riemann-Roch (called
GRR) in the special case that uses smooth complete varieties and replaces the Chow
ring used by Grothendieck with the cohomology ring.
For a smooth complete variety X , we first define the group K(X ) generated by
coherent sheaves on X , modulo the relations
[G ] = [F ] + [H ]
for all short exact sequences 0 F G H 0. One can also prove that a
coherent sheaf F on X has a Chern character ch(F ) H (X , Q) such that
ch(L ) = e c1 (L ) when L is a line bundle.
ch(G ) = ch(F ) + ch(H ) when 0 F G H 0 is exact.
It follows that the Chern character induces a group homomorphism

ch : K(X ) H (X , Q).
Now suppose that f : X Y is a proper morphism between smooth complete

varieties. If F is coherent on X , then one can prove that the higher direct images
630
Ri f F are coherent on Y . By the long exact sequence in cohomology, the map

X
f ! [F ] =
(1)i [Ri f F ] K(Y )
i0
defines a homomorphism f ! : K(X ) K(Y ). On the cohomology side, we also have

a map f ! : H (X , Q) H (Y , Q), which for f : X Y as above is the Poincare
dual of f : H (X , Q) H (Y , Q). We call f ! a generalized Gysin map. Here are
two important examples of these maps:
R
When f : X {pt}, f ! is the map X : H (X , Q) Q from 12.4.
When i : X Y is an inclusion map, then i ! : H k (X , Q) H k+2 codim X (Y , Q)

is the Gysin map, which satisfies i ! i = [X ] for H k (Y , Q).
We will say more about the maps f ! in the appendix at the end of the chapter.
These definitions show that a proper morphism f : X Y gives a diagram

K(X )
ch
/ H (X , Q)
ch
f!
f!
K(Y )
/ H (Y , Q)
which unfortunately does not commute. This is why Todd classes are needed: GRR
states that the modified diagram
K(X )
(13.1.8)
chTd(X)
/ H (X , Q)
chTd(Y )
f!
f!
K(Y )
/ H (Y , Q)
does commute. In Exercise 13.1.4 you will prove that (13.1.8) reduces to HRR
when Y = {pt}.
Following [41], we now specialize to the case where i : X Y is the inclusion
of a smooth hypersurface in a smooth complete variety, and we apply (13.1.8) to
[OX ] K(X ). First observe that i! [OX ] = [i OX ] since i is an affine morphism.
Then the exact sequence 0 OY (X ) OY i OX 0 and the properties of
the Chern character imply that

ch(i! [OX ]) Td(Y ) = ch([i OX ]) Td(Y ) = ch [OY ] [OY (X )] Td(Y )
= (1 e[X] ) Td(Y ).
This is encouraging since expressions like 1 e[X] appear in the denominator in

the formula for the Todd class given in Definition 13.1.4. We can do more since
S = [X ]/(1 e[X] ) is invertible in H (Y , Q), so that
(13.1.9)
(1 e[X] ) Td(Y ) = (S1 Td(Y )) [X ] = i! i (S1 Td(Y )),
631
where the last equality follows since i! is a Gysin map. In order for (13.1.8) to
commute, (13.1.9) must equal i! ch([OX ]) Td(X ) = i! (Td(X )). In other words,
we want Td(X ) = i (S1 Td(Y )), which is equivalent to the equation
(13.1.10)
i Td(Y ) = Td(X ) i
[X ]
.
1 e[X]
In Exercise 13.1.5, you will show that the formula of Definition 13.1.4 guarantees
that Todd classes satisfy this identity.
Exercises for 13.1.
13.1.1. Use the method of Example 13.1.5 to compute the Todd polynomial T3 .
13.1.2. Derive the Riemann-Roch theorem for surfaces (13.1.7) from (13.0.1).
13.1.3. Derive the Riemann-Roch theorem for curves (10.5.1) from (13.0.1).
13.1.4. Prove that HRR is the special case of the commutative diagram (13.1.8) when
Y = {pt}. Thus GRR implies HRR.
13.1.5. Let Y be a smooth complete variety and let i : X Y be a smooth hypersurface.
Then we have the exact sequence of locally free sheaves
0 TX i TY NX /Y 0
from Theorem 8.0.18, where NX /Y = (IX /IX2 ) .
(a) Prove that NX /Y i OY (X). Hint: The ideal sheaf of i : X Y is IX = OY (X)
since X has codimension 1.
(b) Prove the identity c(i TY ) = c(TX ) c(i OY (X)) of total Chern classes.
(c) Show that c(i OY (X)) = 1+i [X] and use part (b) to explain why i [X] can be regarded
as a Chern root of i TY .
(d) Finally, show that the Todd classes Td(X) and Td(Y ) satisfy (13.1.10).
13.1.6. Let E be a locally free sheaf of rank r on a smooth variety X. Similar to what
we did with
Qr the tangent bundle, the Chern roots of E are symbolic objects i such that
c(E ) = i=1 (1 + i ). Then the Chern character ch(E ) is defined as
ch(E ) =
X
1 k
( + + rk ).
k! 1
k=0
(a) Explain why ch(E ) is a polynomial in the Chern class of E .

(b) Let E = L1 Lr be a sum of line bundles. Use the definition of Chern character
to show that ch(E ) = ch(L1 ) + + ch(Lr ).
(c) If E has rank 2, show that
ch(E ) = 2 + c1(E ) +
1

1
c1 (E )2 2 c2 (E ) + c1 (E )3 3 c1 (E ) c2 (E ) + .
2
6
632
13.2. Brions Equalities

When we prove the toric version of equivariant Riemann-Roch in 13.3, we will
create equivariant versions of Euler characteristics, Chern characters and Todd
classes, and then decompose them into sums of local terms. In this section, we
will see how this works for the Euler characteristic of a line bundle.
When D is a torus-invariant divisor on a complete toric variety X = X of
dimension n, the sheaf L = OX (D) has Euler characteristic
(L ) = (X , L ) =
n
X
(1)i dim H i (X , L ).
i=0
Also recall from 9.1 that we used the decomposition

M
H i (X , L ) =
H i (X , L )m
mM
to compute sheaf cohomology of L . Combining these two formulas, we define
e(L ) =
e(X , L ) =
n
X
(1)i dim H i(X , L )m m .
i=0
This lives in the semigroup algebra Z[M]. The goal of this section is to write
e(X , L ) as a sum of local terms that live in a certain localization of Z[M].
The Formal Semigroup Module. The local terms will involve cohomology over
affine open subsets, which can lead to infinite sums. Following
[46], we introduce
P
the abelian group Z[[M]] consisting of all formal sums mM am m for am Z.
This is not a ring, but there is a multiplication map
Z[M] Z[[M]] Z[[M]]
that makes Z[[M]] into a Z[M]-module. We say that Z[[M]] is the formal semigroup
module of M.
Now let X = X be an arbitrary n-dimensional toric variety, which need not be
complete. Given a torus-invariant Weil divisor D on X , let L = OX (D) and define
(13.2.1)
n
XX
e(X , L ) =
(1)i dim H i (X , L )m m Z[[M]].
mM i=0
In Chapter 9, we computed H (X , L ) using the Cech

complex C (U , L ).
For the open cover U = {U }max , let max = {1 , . . . , } and define
M
H 0 (U , L ),
Cp (U , L ) =
[] p
where we refer to (9.1.1) for the precise meaning of [] p and . This gives a
Cech
formula for
e(X , L ) as follows.
633
Lemma 13.2.1.
e(X , L ) =
p0 (1)
e(U , L )
[] p
in Z[[M]].
Proof. Just as in 9.1, the M-grading on H (X , L ) and C (U , L ) implies that

H (X , L )m is the cohomology of the complex C (U , L )m . Since a complex and
its cohomology have the same Euler characteristic, it follows that
n
X
i=0
n
X
X
dim H 0 (U , L )m .
(1) p
(1) dim H (X , L )m =
i
p=0
[] p
Pn
However dim H 0 (U , L )m = i=0 (1)i dim H i(U , L )m since U is affine.

From here, the lemma follows easily.

Summable Series. For any cone , the vanishing of H i (U , L ) for i > 0
implies that
X
(13.2.2)
e(U , L ) =
dim H 0 (U , L )m m Z[[M]].
mM
We will soon see that these sums are reasonably behaved in the following sense.
Definition 13.2.2. An element f Z[[M]] is summable if there exists g Z[M]

and a finite set I M \ {0} such that in Z[[M]], we have
Y
f (1 m ) = g.
mI
Also let Z[[M]]Sum Z[[M]] be the subset of summable elements of Z[[M]].
By Exercise 13.2.1, Z[[M]]Sum is a Z[M]-submodule of Z[[M]]. Here are some

examples of summable elements.
P
Example 13.2.3. Let M = Z and let = 1 . Then

=0 is summable since
(1 )(1 + + 2 + ) = 1.
P
A more surprising example is

= , which is summable since
(1 )( + 2 + 1 + 1 + + 2 + ) = 0.
P
In Example 13.2.7 we will use this to show that

= sums to zero.
Here is a basic summability result.
Lemma 13.2.4. Let D be a torus-invariant Cartier divisor on X = X and set

L = OX (D). Then
e(U , L ) is summable for every . Furthermore:
(a) If dim < n, then
for some m0 M \ {0}.
e(U , L ) (1 m0 ) = 0
634
(b) If dim = n and is smooth, then
e(U , L )
n
Y
i=1
(1 m,i ) = m ,
where m M satisfies D|U = div( m )|U and m,1 , . . . , m,n M are the
minimal generators of .
Proof. If dim < n, then is not strongly convex, so that we can find a nonzero
m0 ( ) M. Then m0 C[ M] is invertible, which means that
multiplication by m0 gives an isomorphism H 0 (U , L ) H 0 (U , L ). Hence
H 0 (U , L )m H 0 (U , L )m+m0
e(U , L ) =
e(U , L ), and part (a) follows.
for all m M. This implies m0
For part (b), assume is smooth and let u1 , . . . , un be the minimal generators
of 1 , . . . , n (1). They form a basis of N since is smooth. Then the dual
basis {mP
of the minimal generators of . If we write
1 , . . . , mn } M consists P
n
D|U = i=1 ai Di , then m = ni=1 ai mi satisfies D|U = div( m )|U .
The polyhedron of L |U is PD, = {m M | hm, ui i ai , 1 i n}. Using
the above formula
for m , we obtain PD, M = m + M. We also have
L
H 0 (U , L ) = mPD, M C m by (4.3.3). Then (13.2.2) implies that
X
X
(13.2.3)
e(U , L ) =
m = m
m.
mm + M
m M
Since M = Nm1 + + Nmn , the sum on the right lives in the formal series
ring Q[[ m1 , . . . , mn ]]. In this ring, one computes that
X
m M
n X
Y
i=1
mi
=0
n
Y
i=1
(1 mi )1 .
Part (b) of the proposition now follows easily.

Finally, suppose that (n) is not smooth. Since D is Cartier, we have
D|U = div( m )|U for some m M. Then (13.2.3) still holds, so it suffices to
P
show that m M m is summable. The methods of 11.1 imply that MR
has a smooth refinement. Let the maximal cones of this refinement be C1 , . . . ,Cs .
By inclusion-exclusion, we obtain
X
X X
X X
(13.2.4)
m =
m
m + .
m M
mCi M
i< j mCi C j M
Since an intersection C = Ci1 Ci is smooth, the minimal generators

of C M
P
are a subset of a basis of M. Then the proof of part (b) implies that mCM m is
summable. We conclude that
e(U , L ) is summable.
635
The Simplicial Case. When (n) is smooth, part (b) of Lemma 13.2.4 shows
that
e(U , L ) satisfies a nice identity. With a little work, this can be generalized
to the simplicial case.
Let (n) be simplicial and let m,1 , . . . , m,n be the minimal generators of
MR . Recall from Proposition 11.1.8 that
P
(13.2.5)
mult( ) = |P |, P = { ni=1 i m,i | 0 i < 1},
where mult( ) = [M : Zm,1 + + Zm,n ]. You will explore the relation between
mult() and mult( ) in Exercise 13.2.3.
Lemma 13.2.5. If D is a torus-invariant Cartier divisor on X = X and (n)

is simplicial, then for L = OX (D), we have
e(U , L )
n
Y
X
(1 m,i ) = m
m,
mP M
i=1
where D|U = div( m )|U , and m,1 , . . . , m,n and P are as above.
P
Proof. For simplicity, write mi = m,i . Since
e(U , L ) = m m M m by
(13.2.3), it suffices to prove that
n
X
X
Y
(13.2.6)
m
m (1 mi ) =
m M
mP M
i=1
in Z[[M]]. Let A = Nm1 + + Nmn M. We claim that

(13.2.7)
m M m = m + m for m A and m P M.
One direction is obvious. For the

Pnother direction, take m M. Since the mi
generate , we can write m = i=1 ci mi for ci 0. Then writing ci = ai + bi for

ai N and 0 bi < 1 easily implies that m has the desired form, proving (13.2.7).
By Exercise 13.2.4, the decomposition m = m + m in (13.2.7) is unique when

it exists. It follows that
X
X
X

(13.2.8)
m .
m
m =
m M
mP M
mA
Arguing as in the proof of Theorem 13.2.8, we also have

n
X Y
m
(1 mi ) = 1.
mA
i=1
Hence multiplying each side of (13.2.8) by
Qn
i=1 (1
mi )
gives (13.2.6).
Example 13.2.6. Consider the cone = Cone(e1 , e1 2e2 ) NR , with dual

= Cone(e2 , 2e1 e2 ) MR . One computes easily that P = {0, e1 e2 }.
Setting ti = ei , Lemma 13.2.5 implies that
e(U , OU ) (1 t21 )(1 t12 t21 ) = 1 + t1 t21 .
636
In [50, 1.3], Brion and Vergne prove that

e(U , L ) is summable under the
weaker hypothesis that D is a Weil divisor, and in [50, 2.1], they give a version of
Lemma 13.2.5 that applies to Weil divisors and simplicial cones.
Brions Equalities. Consider
the multiplicative set S Z[M] consisting of all finite
Q
products of the form si=1 (1 mi ), where m1 , . . . , ms M \ {0}. Note that 1 S
because of the empty product. This gives the localization Z[M]S.
Given a summable element f Q[[M]]Sum , there are h S and g Z[M] such
that h f = g. Then set
g
S( f ) = Z[M]S.
h
In Exercise 13.2.1 you will prove that the map
S : Z[[M]]Sum Z[M]S
(13.2.9)
is a well-defined Z[M]-module homomorphism. We call S the sum function.

Example 13.2.7. Let M = Z and be as in Example 13.2.3. Then
P
1
S
=
since (1 )(1 + + 2 + ) = 1
=0
1

P
= 0
S
since (1 )( + 1 + 1 + + ) = 0.
=
This example shows that S is not injective. In fact, it has a rather large kernel,
which is key to proving the following remarkable equalities due to Brion.
Theorem 13.2.8. Let X = X be a complete toric variety of dimension n. If L =
OX (D) for a torus-invariant Cartier divisor D with Cartier data {m }(n) , then:
X
(a)
e(X , L ) =
S(e
(U , L )) in Q[M]S.
(n)
(b) If (n) is simplicial and m,1 , . . . , m,n M are the minimal generators of
MR , then
P
m mP M m
S(e
(U , L )) = Qn
m,i )
i=1 (1
for P as in (13.2.5). In particular, if is smooth, then
m
S(e
(U , L )) = Qn
m,i ) .
i=1 (1
Proof. First, S(e

(X , L )) =
e(X , L ) since X is complete. Then Lemma 13.2.1
implies that
X
X

S(e
(U , L )) .
e(X , L ) =
(1) p
p0
[] p
But we also know that S(e

(U , L )) = 0 when dim < n by Lemma 13.2.4.
Since X is complete, the open cover used in the Cech

complex is U = {U }(n) ,
637
and the only place n-dimensional cones appear is in the first term C 0 (U , L ) =
(n) H 0 (U , L ). Hence all terms in the above sum drop out except for those
for p = 0, which correspond to (n). This proves part (a). Part (b) now follows
immediately from Lemmas 13.2.4 and 13.2.5.

Example 13.2.9. For the cone = Cone(e1 , e1 2e2 ) NR of Example 13.2.6,
part (b) of Theorem 13.2.8 implies that
S(e
(U , OU )) =
1 + t1 t21
(1 t21 )(1 t12t21 )
This also follows from Example 13.2.6 by the definition of S.
Here is a wonderful application of Theorem 13.2.8.

Corollary 13.2.10 (Brions Equalities). Let P MR Rn be a full dimensional
lattice polytope, and for each vertex v P, let Cv = Cone(P M v) MR . Then:
X
X
X

m .
(a)
m =
v S
mPM
v vertex
mCv M
(b) If P is simple and v P is a vertex, let mv,1 , . . . , mv,n M be the minimal

generators of Cv . Then
P
X
X v mP M m
m
v
Qn
=
(1
mv,i )
i=1
v vertex
mPM
for Pv = PCv as in (13.2.5). In particular, if P is smooth, then

X
X
v
Qn
m =
mv,i ) .
(1
i=1
v vertex
mPM
Proof. We will apply Proposition 13.2.8 to the toric variety XP and the line bundle
L = OXP (DP ). By definition, the maximal cones of the normal fan P are v = Cv
for v P a vertex. Also recall from (4.2.8) that the vs are the Cartier data of DP .
The higher cohomology of L vanishes since DP is ample, so that
X
e(XP , L ) =
m.
mPM
Furthermore, v = Cv and (13.2.3) imply that

X
X
e(Uv , L ) = v
m = v
m.
mv M
mCv M
Then part (a) follows from Theorem 13.2.8, and the same is true for part (b).
Brions original proof [46] of Corollary 13.2.10 used equivariant RiemannRoch. In [155], Ishida gave an elementary proof of Brions equalities and used it
prove a special case of HRR. Brion and Vergne prove Theorem 13.2.8 in [50, 1.3]
638
as part of their toric proof of equivariant Riemann-Roch. A completely elementary

approach to Corollary 13.2.10 can be found in [22, Thm. 9.7].
Let us give a simple example of Brions equalities.
Example 13.2.11. Consider the interval [0, d] R with vertices v1 = 0 and v2 = d.
Note that Cv1 = [0, ) and Cv2 = (, 0]. Since
S
S
X

= S
=
=0
[0,)Z
X

=
= S
=0
(,0]Z
1
1
1
,
1 1
the right-hand side of Brions equality is

1
1
1 1 + d (1 )
+ d
=
1
1 1
(1 )(1 1 )
=
The final result is
(1 d+1 )(1 1 )
(1 )(1 1 )
= 1 + + 2 + + d.
[0,d]Z
, as predicted by Brions equality.
Here is a more substantial example that uses Theorem 13.2.8.

Example 13.2.12. For the Hirzebruch surface X = H2 and divisor D = 3D3 5D4 ,
we found that (L ) = 4 for L = OX (D) in Example 13.1.8. Here, M = Z2 , and
u1 = (1,2)
3
4
u2
u3
u4
Figure 1. The fan for X = H2
we set t1 = e1 and t2 = e2 . In Exercise 13.2.2 you will check that with labellings
639
on (2) shown in Figure 1, we have

S(e
(U1 , L )) =
S(e
(U2 , L )) =
S(e
(U3 , L )) =
t13
(1 t1 )(1 t2 )
t13 t25
(1 t1 )(1 t21 )
t110 t25
(1 t11 )(1 t12t21 )

1
S(e
(U4 , L )) =
.
1
(1 t1 )(1 t12t2 )
Adding these up and doing a lot of algebra yields
(13.2.10) t14 t23 + t15t23 + t14t24 + t15t24 + t16t24 + t17t24 t13t21 t12t21
In Figure 2, the six exponent vectors m = (a, b) where t1a t2b appears with a positive
sign in (13.2.10) are plotted as solid dots, and the two exponent vectors where t1a t2b
appears with a negative sign are plotted as hollow dots.
7 6 5 4 3 2 1
1
2
3
4
Figure 2. Exponent vectors of the monomials in (13.2.10)
In Exercise 13.2.2 you will show that PD = , so that H 0 (X , L ) = 0. Hence

X
X
e(X , L ) =
dim H 1 (X , L )m t m +
dim H 2 (X , L )m t m .
mZ2
mZ2
where t m = t1at2b for m = (a, b) Z2 . This equals (13.2.10) by Theorem 13.2.8.

The sign patterns from Proposition 9.1.6 show that the same m cant appear in both
H 1 (X , L ) and H 2 (X , L ). Thus solid dots correspond to m with H 2 (X , L )m 6= 0
and hollow dots correspond to m with H 1 (X , L )m 6= 0. We will confirm these
cohomology computations using Macaulay2 [123] in Example B.5.1 and using
Sage [262] in Example B.7.1.
640
When X is complete, m 7 1 takes

e(X , L ) to (X , L ). We will call this
taking the nonequivariant limit in 13.3. For instance, Example 13.2.12 gives
e(L ) = t14t23 + t15t23 + t14t24 + t15t24 + t16t24 + t17t24 t13t21 t12t21 ,
so that mapping ti = ei to 1 gives
(L ) = 1 + 1 + 1 + 1 + 1 + 1 1 1 = 4.
This agrees with the computation done in Example 13.1.8. The difference is that
previously, we used Riemann-Roch, while here, all we needed was the explicit
local decompostion of
e(L ). In the next section, this local decompostion will be
an important part of our proof of equivariant Riemann-Roch for toric varieties.
Exercises for 13.2.
13.2.1. Consider Q[[M]]Sum Q[[M]] from Definition 13.2.2.
(a) Prove that Q[[M]]Sum is a Q[M]-submodule of Q[[M]].

(b) Prove that (13.2.9) is well-defined and is a homomorphism of Q[M]-modules.
13.2.2. Consider X = H2 and L = OX (D), D = 3D3 5D4 , as in Example 13.2.12.
(a) Compute the polyhedron PD of D and conclude that H 0 (X, L ) = 0.

(b) This example computed the ms that occur in H 1 (X, L ) and H 2 (X, L ). Compute the
chamber decomposition for D on X (see Example 9.1.8) and then use Proposition 9.1.6
to show that we get the same ms as in Figure 2.
13.2.3. Let NR Rn be a simplicial cone of dimension n with minimal generators

u1 , . . . , un N, and let m1 , . . . , mn MR be the dual basis in the sense of linear algebra.
Also let i be the smallest positive integer such that mi = i mi M.
(a) Prove that m1 , . . . , mn are the minimal generators of .
Qn
(b) Prove that mult() mult( ) = i=1 i . Hint: Show that M Zm1 + + Zmn is dual
to Zu1 + + Zun N.
13.2.4. Prove that the decomposition m = m + m in (13.2.7) is unique when it exists.

13.2.5. Let L = OP1 (d) for d > 0.
(a) Compute
e(P1 , L ) using Batyrev-Borisov vanishing (Theorem 9.2.7).
(b) Compute
e(P1 , L ) using Theorem 13.2.8.
(c) Compute (P1 , L ) using the Riemann-Roch theorem for curves (10.5.1) and explain
how your answer relates to
e(P1 , L ).
13.2.6. For X = X and L as in Lemma 13.2.2, we have S(e

(U , L )) = 0 when
satisfies dim < n = dim X. Here you will prove that S(e
(U , L )) 6= 0 when dim = n.
To begin, assume that there are m1 , . . . , ms M \ {0} with
s
Y
(13.2.11)
e(U , OU ) (1 mi ) = 0.
i=1
(a) Show there is u Int( N) such that hmi , ui =

6 0 for all i.
(b) Show that we can assume hmi , ui > 0 for all i. Hint: If necessary, multiply (13.2.11)
by mi .
13.3. Toric Equivariant Riemann-Roch
641
(c) Show that the constant term 1 on the left-hand side of (13.2.11) cannot cancel. Then
conclude that S(e
(U , L )) 6= 0.

In 12.4 we described the cohomology ring of a complete simplicial toric variety
by first computing its equivariant cohomology and then taking the nonequivariant
limit. We use the same strategy here: we will deduce HRR for a smooth complete
toric variety from the following equivariant Riemann-Roch theorem.
Theorem 13.3.1. For a smooth complete toric variety X = X and a line bundle
L = OX (D) of a torus-invariant divisor D on X , we have
Z
chT(L ) Td T(X ).
T (L ) =
X eq
R
In this theorem, T (L ), X eq , chT(L ) and Td T(X ) are equivariant versions of
the corresponding objects appearing in (13.0.1). The equality of Theorem 13.3.1
takes place in the completion of the equivariant cohomology ring of a point, which
b All of this will be defined carefully in the course of the section.
we denote by .
We will follow [50], where Brion and Vergne prove equivariant Riemann-Roch
for complete simplicial varieties. We will discuss their simplicial version of the
theorem after completing the proof of the smooth case.
The Equivariant Euler Characteristic. As in 12.4, we set
(T )Q = HT ({pt}, Q),
where T = TN is the torus associated to M. Since
are infinite sums,
L exponentials
k ({pt}, Q), written
we will use the completion of HT ({pt}, Q) =
H
k=0 T
b =H
bT ({pt}, Q) =
HTk ({pt}, Q).
k=0
Recall from (12.4.9) that we have an isomorphism s : SymQ (M) (T )Q ,

L
k
where SymQ (M) is our notation for
k=0 SymQ (MQ ). Hence, if m M, then
b It follows that m M gives the exponential
s(m) is a degree 2 element of .
b
e s(m) = 1 + s(m) + 1 s(m)2 + 1 s(m)3 + .
2
We can now define the equivariant Euler characteristic that appears on the lefthand side of the equivariant Riemann-Roch theorem.
Definition 13.3.2. If X = X is complete and L = OX (D) is the line bundle of a

torus-invariant divisor D on X , then the equivariant Euler characteristic of L is
n
XX
T
b
(13.3.1)
(L ) =
(1)i dim H i (X , L )m e s(m) ,
mM i=0
642
where H i (X , L ) =
mM H
i (X , L )
m
is the decomposition from 9.1.
We will see below that T (L ) is closely related to

e(L ) from 13.2.
b Q be the map that sends elements of

The Nonequivariant Limit. Let ipt :
positive degree to zero. Then
ipt ( T (F )) = (F ) =
n
X
(1)i dim H i (X , F ),
i=0
i.e., ipt takes the equivariant Euler characteristic to the ordinary Euler characteristic.
Later in the section we will show that applying ipt to the equivariant RiemannRoch theorem gives HRR from (13.0.1). As mentioned in 12.4, the maps ipt and
iX : HT (X , Q) H (X , Q) are the nonequivariant limit that turns equivariant
cohomology into ordinary cohomology.
Equivariant Chern Characters and Todd Classes. To define equivariant Chern
characters, we replace the equivariant cohomology HT (X , Q) with its completion
bT (X , Q) =
H
The nonequivariant limit map

way to a ring homomorphism
iX
HTk (X , Q).
k=0
HT (X , Q)
H (X , Q) extends in the obvious
bT (X , Q) H (X , Q)
iX : H
since elements of degree > 2 dim X map to zero.
P
By Proposition (12.4.13), a torus-invariant divisor D = a D gives the
P
equivariant cohomology class [D]T = a [D ]T HT2 (X , Q). Then we define
the equivariant Chern character chT(L ) of L = OX (D) to be
bT (X , Q).
chT(L ) = e [D]T = 1 + [D]T + 2!1 [D]2T + H
Furthermore, since X is smooth, Theorem 13.1.6 implies that the ordinary Todd
class of X is
r
Y
[D ]
Td(X ) =
.
1 e[D ]
i=1
Hence it is natural to define the equivariant Todd class of X to be
Y [D ]T
Td T(X ) =
.
1 e[D ]T
bT (X , Q). Since iX [D ]T = [D ]
The power series (13.1.4) shows that Td T(X ) H
by Proposition 12.4.13, we have
(13.3.2)
iX chT (L ) = ch(L )
and
iX Td T(L ) = Td(L ).
643
R
The Equivariant Integral. We saw in 13.1 that the map X : H (X , Q) Q is
the generalized Gysin map of the constant function p : X {pt}. In the appendix
to this chapter, we will see that p also gives the equivariant Gysin map
p! : HT (X , Q) HT ({pt}, Q)
R
which maps HTk (X , Q) to HTk2 dim X ({pt}, Q). We write X eq instead of p! , so that
Z
b
bT (X , Q) H
bT ({pt}, Q) = .
:H
X eq
R
R
This is called the equivariant integral. We will prove later that X eq and X are
compatible with taking the nonequivariant limit. This will make it easy to derive
HRR from equivariant Riemann-Roch.
Strategy of the Proof . We have now defined everything needed to make sense of
the equivariant Riemann-Roch theorem for X = X in Theorem 13.3.1:
b
T (L ) lives in .
bT (X , Q).
chT(L ) and Td T(X ) live in the ring H
R
b
X eq maps HT (X , Q) to .
R
b
Thus equivariant RR is the equation T (L ) = X eq chT(L ) Td T(X ) in .
The next step is to say a few words about how we will prove the theorem. The
main idea is to use the map : HT (X , Q) HT (X T , Q) from (12.4.11) induced by
the inclusion of the fixed point set X T X . Recall that X T = {x | (n)},
where n = dim X . The completed version of is
M
M
b
bT ({x }, Q) =
bT (X , Q)
.
H
:H
(n)
(n)
This map is injective by Corollary 12.4.9, and the localization theorem (see [9])
implies that becomes an isomorphism after tensoring with the field of fractions of
b In our case, we can do better. Let S be the multiplicative set consisting
(T )Q .
b The empty product shows
of all finite products of nonzero degree 2 elements of .
that 1 S. Then Exercise 12.4.3 implies that the localized map
M
b S.
bT (X , Q)S
(13.3.3)
S : H
(n)
b S are the localizations in the sense of

bT (X , Q)S and
is an isomorphism. Here, H
commutative algebra. Thus we consider fractions whose denominators lie in S.
The strategy of the proof is to express each side of the desired equation as
b S indexed by (n). The proof will then reduce to
a sum of local terms in
b S . Many
checking that the local terms on each side of the equation are equal in
proofs in equivariant cohomology use this approach.
644
Local Euler Characteristics. Following the strategy outlined above, we first show
b decomposes as a sum of local
that the equivariant Euler characteristic T (L )
terms
X
b S.
T (L ) =
T (L ), T (L )
(n)
When X = X is complete, Brions equalities (Theorem 13.2.8) imply that

n
XX
(13.3.4)
e(X , L ) =
(1)i dim H i (X , L )m m Z[M]
mM i=0
from (13.2.1) as a sum of local terms

(13.3.5)
e(X , L ) =
(n)
S(e
(U , L )),
where each S(e

(U , L )) lies in the localization Z[M]S and the sum function S is
defined in (13.2.9).
b S using the ring homomorphism
We translate (13.3.5) into a decomposition in
b
: Z[M] ,
If m M is nonzero, then
m 7 e s(m) .
1 e s(m) s(m) 21 s(m)2

=
= 1 12 s(m)
s(m)
s(m)
b Thus
is invertible in .
(1 m ) = 1 e s(m) = s(m)
1 e s(m)
s(m)
b
= s(m) invertible element of .
b S inverts all s(m), it follows that extends to

Since Q[M]S inverts all 1 m and
a ring homomorphism
b S.
: Q[M]S
The definitions of T (L ) from (13.3.1) and

e(X , L ) from (13.3.4) make it
easy to see that
(e
(X , L )) = T (L )
when X = X is complete. Using , we now define the local version of T (L ).
Definition 13.3.3. Let X = X be complete of dimension n and set L = OX (D)

for a torus-invariant Cartier divisor D. Given (n), define
T (L ) = (S(e
(U , L ))).
Here is the desired decomposition.

Theorem 13.3.4. Let X = X be complete of dimension n and set L = OX (D) for
a torus-invariant Cartier divisor D with Cartier data {m }(n) . Then:
(a) T (L ) =
(n)
645
b S.
T (L ) in
(b) If (n) is simplicial and m,1 , . . . , m,n M are the minimal generators of
MR , then
P
e s(m ) mP M e s(m)
T (L ) =
Qn
s(m,i ) )
(1
e
i=1
for P as in (13.2.5). In particular, if is smooth, then
e s(m )
.
s(m,i ) )
i=1 (1 e
T (L ) = Qn
Proof. Apply to Theorem 13.2.8.
Fixed Points. After decomposing the equivariant Euler characteristic into local
factors, the next step is to decompose the equivariant integral. This requires that
we study the fixed points of the torus action on X = X . Here we will assume that
X is complete and simplicial.
Given (n), note that V () = {x } is a fixed point of the torus action. Let
i : V () = {x } X
be the inclusion map. We also have the constant map p : X {pt}. Since p i is
the canonical map that takes x to pt, we get a commutative diagram
(13.3.6)
bT (X , Q)
H
T
TTTTi
TTTT
T)
b /H
b ({x }, Q),
bT ({pt}, Q) =
H
p
where the isomorphism on the bottom is (pi ) = i p . Using this isomorphism

b we can write
b ({x }, Q) with H
bT ({pt}, Q) = ,
to identify H
bT (X , Q).
when H
b
i ()
Lemma 13.3.5. Let X = X be complete and simplicial of dimension n. Also let

(n) and (1). Then:
(a) If
/ (1), then i ([D ]T ) = 0.
(b) If (1), then i ([D ]T ) = 1

s(m), where be the smallest positive integer
such that D is Cartier and m M satisfies D U = div( m ).
j
Proof. Factor i : {x } X as {x } U
X . Proposition 12.4.13 implies that
for any (1),
j ([D ]T ) = [D U ]T .
646
If
/ (1), this is zero since D U = . On the other hand, if (1), then on
U , we have D U = div( m ). Using Proposition 12.4.13 again, we obtain
bT (U , Q).
j ([D ]T ) = [D U ]T = [div( m )]T = s(m) 1 H
b and everything is a module over ,

b mapping the above equation into
Since s(m)
bT ({x }, Q) implies i ([D ]T ) = s(m) 1. The desired formula follows.

H
bT ({pt}, Q) H
bT (X , Q) from
We also have the equivariant Gysin map i ! : H
the appendix to this chapter. Here is an important property of this map.
Proposition 13.3.6. Let X = X be complete and simplicial. If (n), then
Y
i ! (1) = mult()
[D ]T .
(1)
Proof. Since we have not developed the full theory of equivariant

cohomology
Q
classes, our argument will be somewhat ad-hoc. Note that (1) [D ]T lies in
HT2n (X , Q), and the same is true for i ! (1) by Proposition 13.A.9.
The first step is to show that
Z
Z
i ! (1) = 1 and
(13.3.7)
X eq
mult()
X eq
[D ]T = 1.
(1)
R
Since X eq = p! , the integral on the left is p! (i ! (1)), which by Proposition 13.A.9
is equal to (p i )! (1) = 1 since p i : {x } {pt}. For the other integral in
R Q
b=H
bT ({pt}, Q) has degree zero by
(13.3.7), observe that X eq (1) [D ]T
b 0 = Q, it suffices to consider
Proposition 13.A.9. Since
Z Y
Z
Z
Z Y
Y
Y
[D ],
iX [D ]T =
[D ]T =
[D ]T = iX
ipt
X eq (1)
X (1)
(1)
X (1)
where the first equality uses the commutative diagram from Proposition 13.A.11
and third uses
Then we are done by Lemma 12.5.2, which
Q Proposition 12.4.13. 1
implies that (1) [D ] = mult() [V ()] = mult()1 [{x }].
The second step is to show that

Y
(13.3.8) i (i ! (1)) = 0 and i
[D ]T = 0 when (n), 6= .
(1)
In this case, the second equality is easy, since some (1) is not contained in
(1), so that i [D ]T = 0 by Lemma 13.3.5. For the first, {x } {x } = and
Proposition 13.A.10 give a commutative diagram
0O
/H
b
bT ({x }, Q)
O
i
bT ({x }, Q)
H
i !
/H
b (X , Q).
647
It follows that i i ! = 0, and (13.3.8) is proved.
In Exercise 13.3.1 you will

Q give the easy argument that (13.3.7) and (13.3.8)
imply that i ! (1) and mult() (1) [D ]T are equal in HT2n (X , Q).

Decomposing the Equivariant Integral. When X is smooth and complete, we get
the following important formula for the equivariant integral.
Theorem 13.3.7. If X = X is an n-dimensional smooth complete toric variety and

bT (X , Q), then
H
Z
X
i ()
Qn
= (1)n
X eq
i=1 s(m,i )
(n)
b S , where m,1 , . . . , m,n are the minimal generators of MR for (n).

in
bT (X , Q)S . For (n), let

Proof. We will work in the localization H
n
b S.
bT (X , Q)S , where = Qn(1)
= i ! (1) H
i=1 s(m,i )
The two key properties of the are

Z
(13.3.9)
= i ()
X eq
(13.3.10)
(n)
bT (X , Q)S .
= 1 H
Note that the theorem follows immediately from these properties.

R
For (13.3.9), recall that X eq equals p! for the constant map p : X {pt}. Then
b
b
bT (X , Q) becomes a -module
H
via p and p! is a -module
homomorphism
since
R
p! (p () ) = p! () by Proposition 13.A.9. It follows that X eq extends to a

b S -module homomorphism between the localizations at S.
By Proposition 13.A.9, the equivariant Gysin map i ! satisfies i ! (1) =

i ! (i ()). Hence
= i ! (1) = i ! (i ()).
Then (13.3.9) follows from the equalities
Z
= p! ( i ! (i ()) = p! (i ! (i ()) = i (),
X eq
where the last equality follows because p! i ! = (p i )! is the inverse of the map
b
bT ({x }, Q) with H
bT ({pt}, Q) = .
(p i ) used in (13.3.6) to identify H
L
b S from (13.3.3).
bT (X , Q)S
For (13.3.10), we will use the map : H

(n)
The strategy is to show that for (n),

(13.3.11)
(i ! (1)) = (1)n
Qn
i=1 s(m,i ) e .
648
Once this is proved, then ( ) = e follows by the definition of . Thus

P
P
(n) =
(n) e = (1),
which implies (13.3.10) since is injective.
To prove (13.3.11), first note that i (i ! (1)) = 0 for 6= by (13.3.8). To

calculate i (i ! (1)), let ui be the minimal generator of i (1) for i i n.
Since is smooth, the dual basis mi = m,i gives the minimal generators of .
Furthermore, Di = Di is Cartier since X is smooth. Then
i (i ! (1)) =
n
Y
i ([Di ]T ) =
n
n
Y
Y
s(mi ),
(s(mi )) = (1)n
i=1
i=1
i=1
where the first equality uses Proposition 13.3.6 and the second uses Lemma 13.3.5
and Di U = div( mi ). This proves (13.3.11) and completes the proof.

Theorem 13.3.7 is a special case of a general
Q formula described in [9]. From
this more sophisticated view, the denominator ni=1 s(m,i ) is the equivariant Euler
class of the normal bundle of {x } X . See Exercise 13.3.2 for further details.
We should also mention that there is a simplicial version of Theorem 13.3.7,
which states that
Z
X mult( ) i ()
Qn
= (1)n
(13.3.12)
eq
s(m
,i )
X
i=1
(n)
when X is complete and simplicial. You will prove this in Exercise 13.3.3.
Proof of Equivariant RR. We finally have the tools needed to prove our version of
the equivariant Riemann-Roch theorem for smooth complete toric varieties, stated
earlier as Theorem 13.3.1.
Proof of Theorem 13.3.1. We need to show that
Z
T
, = chT (L ) Td T(X ),
(L ) =
X eq
when L = OX (D) and D is a torus-invariant divisor on X = X . Let the Cartier

data of D be {m }(n) . By Theorem 13.3.4, we have
X
e s(m )
.
(1 e s(m,i ) )
i=1
(n)
R
Comparing this to the decomposition of X eq given in Theorem 13.3.7, we see
that it suffices to prove that for (n), we have
T (L ) =
(13.3.13)
Qn
e s(m )
n Q i ()
,
=
(1)
n
s(mi ) )
i=1 s(mi )
i=1 (1 e
Qn
= chT (L ) Td T(X ).
where for simplicity we write mi = m,i . The mi are dual to the minimal generators
ni of . Let Di = Di be the divisor corresponding to i = Cone(ui ) (1).
649
Since i () = i (chT(L )) i (Td T(X )), we need to compute i (chT(L )) and
i (Td T(X )). We begin with the former. The Chern character is chT(L ) = e [D]T .
Adapting the proof of Lemma 13.3.5, one easily sees that
i ([D]T ) = s(m )
since the Cartier data of D satisfies D|U = div( m )|U . Hence
i (chT(L )) = i (e [D]T ) = e i ([D]T ) = e s(m ) .

We next compute i (Td T(X )). If
/ (1), then
[D ]

T
i
= i 1 + 21 [D ]T + = 1
[D
]
1e T
by Lemma 13.3.5. On the other hand, if = i for 1 i n, then Di U =
div( mi ) on U , so the same lemma implies that

[D ]
i [D ]T
s(mi )
T
.
=
i
[D ] =
[D
]
i
T
T
1 e s(mi )
1e
1e
Q
Since Td T(X ) = (1) [D ]T /(1 e[D ]T ), it follows that
i (Td T(X )) =
n
n
Y
Y
s(mi )
s(mi )
n
=
(1)
.
s(m
)
s(mi )
i
1
e
1
e
i=1
i=1
It is now easy to prove (13.3.13), since

n
Y
e s(m )
(1)n
s(mi )
s(m )
n
Qn
=
(1)
s(m
)
i )
1 e s(mi )
i=1 s(mi )
i=1 (1 e
i=1
Qn
(1)n
= Qn
i (chT (L )) i (Td T(X )),
i=1 s(mi )
where we have used the above computations of i (chT(L )) and i (Td T(X )). This
gives (13.3.13) since i is a ring homomorphism.

Passing from Equivariant RR to HRR. After a lot of hard work, our reward is an
easy proof of the main result of this chapter.
Theorem 13.3.8. For an invertible sheaf L on a smooth complete toric variety
X = X , we have
Z
(L ) = ch(L ) Td(X ).
X
Proof. We may assume that L = OX (D) for a torus-invariant divisor D on X . This

ensures that T (L ) and chT(L ) are defined. Then
Z
Z

T
T
ch (L ) Td (X ) = iX chT(L ) Td T(X ) ,
(L ) = ipt ( (L )) = ipt
X eq
650
where the first equality is (13.3.1), the second is equivariant Riemann-Roch, and
the third is the commutative diagram from Proposition 13.A.11. RSince iX is a ring
homomorphism, (13.3.2) implies that the integral on the right is X ch(L ) Td(X ).
We have proved the Hirzebruch-Riemann-Roch theorem!

To get a better sense of how HRR relates to equivariant RR, let us work out a
concrete example.
Example 13.3.9. Let L = OP1 (dD0 ), where the minimal generators of the fan of
P1 are u0 = e1 and u1 = e1 in N = Ze1 . The completed equivariant cohomology
ring of X = P1 is
bT (P1 , Q) = Q[[x0 , x1 ]]/hx0 x1 i
H
b = Q[[t]], where t = s(e1 ) for e1 M = Ze1 . As
by Theorem 12.4.14. We write
bT (P1 , Q) by
explained in the discussion following Example 12.4.11, t acts on H
multiplication by (he1 , u0 ix0 + he1 , u1 ix1 ) = x0 x1 .
Recall that
e(L ) = 1 + + + d Z[M] by Example 13.2.11. Since
s(e
)
maps to e 1 = et , we obtain

(13.3.14)
T (L ) = 1 + et + + e dt = d + 1 + d+1
2 t + .
Using xi = [Di ]T , we see that the equivariant Todd class of X = P1 is
x0
x1
1 2
1 2
Td T(P1 ) =
x0 )(1 + 12 x1 + 12
x1 )
= (1 + 12 x0 + 12
x
0
1e
1 ex1
1
(x02 + x12 ) +
= 1 + 21 (x0 + x1 ) + 12
since x0 x1 = 0 in Q[[x0 , x1 ]]/hx0 x1 i. The equivariant Chern character of L is
chT (L ) = e [dD0 ]T = e dx0 = 1 + dx0 + 21 (dx0 )2 + ,
and you will compute in Exercise 13.3.4 that

1
1 2
chT(L ) Td T(P1 ) = 1 + (d + 12 )x0 + 21 x1 + ( 12 d 2 + 21 d + 12
)x02 + 12
x1 + .
R
bT (P1 , Q) Q[[t]].
The next step is to describe the equivariant integral (P1 )eq : H
R
R
R
In Exercise 13.3.4 you will explain why (P1 )eq x0 = (P1 )eq x1 = 1. Since (P1 )eq is a
Q[[t]]-module homomorphism, we have
Z
Z
Z
Z
2
x0 = t.
t x0 = t
(x0 x1 )x0 =
x0 =
(P1 )eq
(P1 )eq
(P1 )eq
(P1 )eq
R
A similar computation shows that (P1 )eq x12 = t (be sure you see where the minus
R
sign comes form). Applying (P1 )eq to the above computation of chT(L ) Td T(P1 ),
we obtain
Z
1
1
chT(L ) Td T(P1 ) = 0 + d + 21 + 21 + ( 21 d 2 + 21 d + 12
)t + 12
(t) +
(P1 )eq
= d +1+
d+1
2 t +
651
The first two terms agree with what we computed in (13.3.14), and all terms agree
by the equivariant Riemann-Roch theorem. Note that HRR is the equality of the
constant terms.
The Simplicial Case. The equivariant Riemann-Roch theorem proved by Brion

and Vergne [50] applies when X = X is complete and simplicial. The problem is
that the formula
Y [D ]T
Td T(X ) =
1 e[D ]T
for the equivariant Todd class no longer holds in the simplicial case. Similarly, for
HRR, the formula
Y [D ]
(13.3.15)
Td(X ) =
1 e[D ]
needs to be modified when X is simplicial. Here is an example.

Example 13.3.10. Suppose that HRR holds for X = P(1, 1, 2). We will use the fan
shown in Figure 5 of Example 3.1.17, where the minimal generators are u1 , u2
and u0 = u1 2u2 . The relations in the class group are
D1 D0 and D2 2D0 ,
and the intersection pairing is determined by D1 D2 = 1 since u1 , u2 generate a
smooth cone. Note also that X = XP for P = Conv(0, 2e1 , e2 ) and that DP = 2D0 .
In Exercise 13.3.5 you will check the details of the following computations.
The cohomology ring of X is H (X , Q) Q Q[D0 ] Q[pt], where [D0 ]2 =
1
2 [pt]. If we apply the formula (13.3.15) to X = P(1, 1, 2), then one computes that
(13.3.16)
Td(X ) = 1 + 2[D0 ] + 87 [pt].
The actual Todd class will have the form a + b[D0 ] + c[pt]. To determine the
constants a, b, c, we apply HRR to L = OX (DP ), which gives
Z
(OX (DP )) = (1 + [DP ] + 21 [DP ]2 )(a + b[D0 ] + c[pt])
ZX
= (1 + [2D0 ] + 21 [2D0 ]2 )(a + b[D0 ] + c[pt])
ZX
= a + (b + 2a)[D0 ] + (a 2 + b + c)[pt]
X
2
= a + b + c.
On the other hand, (OX (kDP )) = |(P)M| is the Ehrhart polynomial of P, which
by (10.5.13) is
EhrP () = Area(P) 2 + 21 |P M| + 1 = 2 + 2 + 1.
652
It follows that the actual Todd class is

Td(X ) = 1 + 2[D0 ] + [pt].
This differs from Td(X ) by 81 [pt]. We will soon see the theoretical reason for this
discrepancy.
Returning to the equivariant case, let X = X be a complete simplicial toric

variety. To define its equivariant Todd class Td T(X ), consider the group G in the
quotient construction of X from 5.1, namely

Q hm,u i
G = (t ) (C )(1) | t = 1 for all m M .
Each (1) gives the character
: (C )(1) C
defined by projection on the th factor. Then for any cone , let
G = {g G | (g) = 1 for all
/ (1)}
= {(t ) (C )(1) | t = 1 for

/ (1),
One can show that

(13.3.17)
hm,u i
(1) t

= 1 for m M .
P
G N /( (1) Zu )
(Exercise 13.3.6), so that |G | = mult().

Then we set
G =
G G
and define the equivariant Todd class of X to be

X Y
[D ]T
(13.3.18)
Td T(X ) =
1 (g) e[D ]T
gG
(1)
when X = X is complete and simplicial. Here is the result proved in [50].

Theorem 13.3.11. Let X = X be a complete simplicial toric variety. If L =
OX (D) is the line bundle of a torus-invariant Cartier divisor D on X , then
Z
T
chT(L ) Td T(X ),
(L ) =
X eq
where Td T(X ) is defined in (13.3.18).
This result has also been proved by Edidin and Graham [87] using different
methods. Once we have a simplicial version of equivariant Riemann-Roch, the
proof of Theorem 13.3.8 gives HRR in the simplicial case as follows.
653
Corollary 13.3.12. Let X = X be a complete simplicial toric variety. If L is a

line bundle on X , then
Z
(L ) = ch(L ) Td(X ),
X
where
Td(X ) =
X Y
gG (1)
[D ]
1 (g) e[D ]
Let us apply this to the previous example.

Example 13.3.13. Consider P(1, 1, 2) as in Example 13.3.10. In Exercise 13.3.5
you will show that G = {(1, 1, 1), (1, 1, 1)} G = {(t,t,t 2 ) | t C }. Thus
Td(X ) is a sum of two terms. For g = (1, 1, 1), we gave (g) = 1 for all , so that
this term is what we get when we use the formula for the smooth case. In (13.3.16),
we computed this to be
Td(X ) = 1 + 2[D0 ] + 78 [pt].
For g = (1, 1, 1), we get the product
which becomes
[D1 ]
[D2 ]
[D0 ]
,
[D
]
[D
]
0
1
1+e
1+e
1 e[D2 ]
1
[D1 ]2 ) = 41 [D0 ][D1 ] = 18 [pt].
( 21 [D0 ] + 41 [D0 ]2 ) ( 21 [D1 ] + 14 [D1 ]2 ) (1 + 21 [D2 ] + 12
The terms for g = (1, 1, 1) and g = (1, 1, 1) sum to Td(X ) = 1 + 2[D0 ] + [pt], in
agreement with Example 13.3.10.
In [229], Pommersheim defines the mock Todd class of a complete simplicial

toric variety to be the class computed using the formula for the smooth case, i.e.,
Y [D ]
Td(X ) =
.
1 e[D ]
He shows that the difference between the actual and mock Todd classes depends
on the codimension of the singular locus of X . He also expresses the difference in
codimension 2 in terms of Dedekind sums.
We should also mention that a nice discussion of Todd classes of general toric
varieties and their enumerative applications can be found in [105, Sec. 5.3].
Exercises for 13.3.
13.3.1. Let X = X be complete and simplicial
R n, and for (n), let i be
R of dimension
as in (13.3.6). Let , HT2n (X, Q) satisfy X eq = 1 and X eq = 1 and assume there is
(n) such that i () = 0 and i () = 0 for all 6= . Prove that = . Hint: Let
b Then consider v u H 2n (X, Q).
u = i () and v = i () in .
T
654
13.3.2. The decomposition of the equivariant

integral given in Theorem 13.3.7 involves
Q
terms containing the product (1)n ni=1 s(m,i ) for (n). Here you will show that
this is the nth equivariant Chern class of the normal bundle of {x } X.
(a) Use {x } U X to show that the normal bundle is Cn where T acts on the ith
factor via m,i . Thus the normal bundle is the direct sum of the rank 1 equivariant
vector bundles given by the characters m,1 , . . . , m,n .
(b) If a rank n vector bundle is a direct sum of rank 1 bundles, show that its nth Chern
class is the product of the first Chern classes of the factors.
(c) Explain why the proof of Proposition 12.4.13 implies that the first equivariant Chern
of the ith factor of the normal bundle is s(m,i ).
13.3.3. Prove (13.3.12). Hint: Adapt the proof of Theorem 13.3.7 using Lemma 13.3.5
and Exercise 13.2.3.
13.3.4. Supply the details omitted in Example 13.3.9.
13.3.5. Supply the details omitted in Examples 13.3.10 and 13.3.13.
13.3.6. Prove (13.3.17). Also show that G acts on C(1) with quotient C(1) /G U .
b
13.3.7. Our discussion of the equivariant Euler characteristic used the map : Z[M]
defined by m 7 e s(m) . This exercise will explore the canonical meaning of . We first
replace Z[M] with Q[M] and note that Q[M] is the rational representation ring of T since
the m are the irreducible representations of T . The augmentation ideal of Q[M] is

P
P
m
I=
mM am = 0 .
mM am Q[M] |
Q
b with Sym
d = Symk (MQ ). Then we can think of as the map
We also replace
Q
k=0
d
: Q[M] Sym,
m 7 e m .
d and the I-adic completion of Q[M] at

Prove that induces an isomorphism between Sym
the augmentation ideal I. This is the canonical meaning of .
13.3.8. Let X = XP be a projective toric variety, where P MR is a full dimensional lattice
P
b =
b0
b2 .
polytope, and let L = OX (DP ). Then T (L ) = mPM e s(m)
(a) Show that the constant term of T (L ) is |P M|.

P
P
m . It is interesting to
(b) Show that the degree 2 term is mPM s(m) = s
mPM
P
1
note that |PM|
mPM m is the average lattice point of P.
In a similar way, the higher terms of T (L ) can be interpreted as various moments of the
lattice points of P. The power of equivariant Riemann-Roch is that when P is smooth, all
of these terms can be computed using equivariant intersection theory on X = XP .
13.4. The Volume Polynomial

In this section, we discuss the relation between cup product and the volume of a
polytope, leading up to the volume polynomial. This will give a new description of
the cohomology ring of a complete simplicial toric variety.
655
Cup Product and Normalized Volume. By Proposition 10.5.6, the Euclidean area
of a lattice polygon P in the plane is given by the intersection formula
2 Area(P) = D D,
where D is a torus-invariant nef divisor on a smooth complete toric surface X such
that P = PD .
Recall from 9.5 that a polytope P MR Rn has Euclidean volume vol(P),
where a fundamental parallelotope determined by a basis of M has volume 1. Then
we define the normalized volume of P by
Vol(P) = n! vol(P).
Thus the standard n-simplex n Rn has normalized volume Vol(n ) = 1.
Theorem 13.4.1. Let D be a torus-invariant divisor on a smooth complete toric
variety X of dimension n and set P = PD MR .
(a) If D is very ample and X P s , s + 1 = dim H 0 (X , OX (D)) = |P M|, is

the projective embedding given by global sections of OX (D), then
Z
s
[D]n = Vol(P).
deg(X P ) =
X
(b) If D is nef, then
[D]n = Vol(P).
X
Proof. The degree of X P s is the number of points in the intersection X L,

where L P s is a generic linear subspace of codimension n. Bertinis theorem
(see [131, II.8.18]) and (12.5.6) imply that if V P s is smooth and H P s is a
sufficiently generic hyperplane, then in the Chow ring A (P s ), we have [H] [V ] =
[H V ], where H V is smooth. Hence in H (P s , Z) we have
Z
s
[H1 ] [Hn ] [X ]
deg(X P ) =
Ps
when H1 , . . . , Hn are sufficiently generic. However, if i : X P s is the inclusion,

the Gysin map i! : H k (X , Z) H k+2n (P s , Z) satisfies i! i = ` [X ] for all
H (P s , Z). Setting = [H1 ] [Hn ], we get
Z
Z
Z
Z
Z
[D]n
i =
i! i =
` [X ] =
[H1 ] [Hn ] [X ] =
Ps
Ps
Ps
by properties of Gysin
R maps discussed in Theorem 13.A.6. Thus the degree is the
intersection number X [D]n .
To bring volume into the picture, we note that the degree is also n! times the
leading coefficient of the Hilbert polynomial of the homogeneous coordinate ring
C[X ] = C[x0 , . . . , xs ]/I(X ) of X P s . We proved in Proposition 9.4.3 that the
Hilbert polynomial is the Ehrhart polynomial of P, and by the discussion following
656
Proposition 9.4.3, we see that the leading coefficient of the Ehrhart polynomial is
Vol(P)/n!. Hence the degree is Vol(P).
For part (b), the nef divisor D need not give a projective embedding. In fact,
X need not be projective. So we instead use HRR. Let 1 be an integer. Then
|(P) M| = dim H 0 (X , OX (D)) = (OX (D)),
where the first equality comes from Example 4.3.7, and the second follows from
Demazure vanishing. By HRR,
Z
ch(OX (D)) Td(X ).
|(P) M| = (OX (D)) =
X
In ch(OX (D)), the term containing the highest power of is [D]n /n! = n [D]n /n!.
Since the degree zero term of Td(X ) is 1, we obtain
Z
Z
n
ch(OX (D)) Td(X ) =
[D]n + lower degree terms in .
n!
X
X
Therefore, by (9.4.4), in the limit as , the normalized volume is given by

Z
Z
|(P) M|
n!
Vol(P) = n! lim
[D]n ,
ch(OX (D)) Td(X ) =
= lim n

n
X
X
If P is not full dimensional, then Vol(P) = 0. As in the discussion following

Lemma 9.3.9, D is a big divisor if Vol(P) > 0. A different proof that the degree
equals in the volume in the very ample case can be found in [113, Thm. 6.2.3].
We next extend part (b) of Theorem 13.4.1 to arbitrary complete toric varieties.
We will need the following fact.
Lemma 13.4.2. Assume that f : X Y is a birational morphism between complete
irreducible varieties of dimension n. If D is a Cartier divisor on Y and D = f D,
then
Z
Z
n
[D ] = [D]n .
X
Proof. Set R= [D]n and let p : Y {pt} be the constant map. Example 13.A.3
explains that Y = p ( a [Y ]), where [Y ] H2n (Y , Q) is the fundamental class
and p : H (Y , Q) H ({pt}, Q) = Q. Then p = p f is the constant map for X ,
so that
Z
[D ]n = p ( f () a [X ]) = p f ( f () a [X ])
X
Z
= p ( a f [X ]) = p ( a [Y ]) = [D]n .
Y
Here, the second line uses Proposition 13.A.2, and in the third line, f [X ] = [Y ]
follows from Proposition 13.A.5.
657
We can now prove a more general version of Theorem 13.4.1.

Theorem 13.4.3. Let D be a torus-invariant nef Cartier divisor on a complete toric
variety X of dimension n. Then
Z
[D]n = Vol(PD ).
X
Proof. Let : X X be a toric resolution of singularities and set D = D.

The pullback D = D has the same support function as D by Proposition 6.2.7.
Then D and D have the same polytope by Lemma 6.1.6, i.e., PD = PD . Thus
Z
Z
n
[D]n ,
[D ] =
Vol(PD ) = Vol(PD ) =
X
where the second equality uses Theorem 13.4.1 and the third equality follows from
Lemma 13.4.2.

The Cartier hypothesis can also be relaxed slightly; Theorem 13.4.3 is also true
for any nef Q-Cartier divisor D (Exercise 13.4.1). Here is an example illustrating
the theorem.
Example 13.4.4. See Figure 11 in Example 2.4.6 and consider the triangle P =
Conv(2e1 + e2 , 2e1 e2 , 2e1 e2 ) in MR = R2 . The toric variety XP is the
weighted projective plane P(1, 1, 2). Label the ray generators of the normal fan of
P as u1 = e1 , u2 = e2 , u3 = e1 2e2 , and let i = Cone(ui ), Di = Di . Then it is
easy to see that P = PD for the divisor D = 2D1 + D2 . On XP P(1, 1, 2), we have
Z
[D]2 = Vol(P) = 8
XP
R
by Theorem 13.4.3. On the other hand, XP [D]2 = D 2 . Since D21 = 12 (D1 is not
Cartier), D1 D2 = 1, and D22 = 2, one computes directly that D 2 = 8.
The proof of Theorem 13.4.3 shows that we can also do this computation using
a resolution of singularities. If we refine the normal fan of XP by introducing a
new ray 0 = Cone(e2 ), we get a smooth fan . Indeed, the resulting surface
isomorphic to the Hirzebruch surface H2 . Let : H2 XP be the corresponding
morphism. If we let D = D = D0 + 2D1 + D2 , then one can verify directly that
PD = P, so that
Z
[D ]2 = Vol(P) = 8
H2
by Theorem 13.4.1. On H2 , D1 D3 , D2 D0 + 2D3 , so that D 2D0 + 4D3 .

Since D02 = 2, D32 = 0, and D0 D3 = 1, one computes that ( D)2 = 8. You will
check these assertions in Exercise 13.4.2.
658
The Volume Polynomial.

Let X be a complete simplicial toric variety of dimenP
sion n. The divisor t D is Q-Cartier when the t are all rational and hence has
a cohomology class in H 2 (X , Q). The resulting integral
Z
P
n
(13.4.1)
t D
X
P
is a homogeneous polynomial of degree n in t , (1). When D = t D is
nef and Cartier, the integral equals Vol(PD ) by Theorem 13.4.3. For this reason,
we call (13.4.1) the volume polynomial of X .
Example 13.4.5. Let D0 , . . . , Dn be the torus-invariant prime divisors on P n . Then
the volume polynomial is
Z
[t0 D0 + + tn Dn ]n = (t0 + + tn )n
Pn
R
since Di D0 for 1 i n and Pn [D0 ]n = 1. Note that D = t0 D0 + + tn Dn
is nef if and only if t0 + + tn 0. In this case, the above computation and
Theorem 13.4.1 imply that PD has normalized volume (t0 + + tn )n .
Example 13.4.6. Let be the usual fan for (P1 )n . Let the divisor D2i1 correspond
to the ray Cone(ei ) and D2i correspond to Cone(ei ) for 1 i n. Then the
polytope of a general divisor D = t1 D1 + + t2n D2n with t1 , . . . ,t2n 0 is the
rectangular solid
PD = [t1 ,t2 ] [t2n1 ,t2n ]
n
in MR = R . The Euclidean volume is (t1 + t2 ) (t2n1 + t2n ), which gives the
normalized volume
(13.4.2)
Vol(PD ) = n! (t1 + t2 ) (t2n1 + t2n ).
The reason for the n! becomes clear when we compute the volume polynomial
in H ((P 1 )n , Q). By the calculations in 12.3 and 12.4,
H ((P 1 )n , Q) Q[x1 , . . . , xn ]/hx12 , . . . , xn2 i,
with xi = [D2i1 ], 1 i n. The top-dimensional component of the cohomology

ring is H 2n ((P 1 )n , Q), generated by the product x1 xn . The cohomology class of
D is [D] = (t1 + t2 )x1 + + (t2n1 + t2n )xn H 2 ((P 1 )n , Q). Then
((t1 + t2 )x1 + + (t2n1 + t2n )xn )n = n! (t1 + t2 ) (t2n1 + t2n ) x1 xn ,
since there are n! waysRto order the factors in x1 xn , and all other terms in the nth
power are zero. Since (P1 )n x1 xn = 1, we get (13.4.2).
In general, as in these examples, if we start from a complete simplicial fan

with |(1)| = r and the divisors D1 , . . . , Dr on X corresponding to 1 , . . . , r in
(1), then the volume polynomial (13.4.1) becomes
Z
[t1 D1 + + tr Dr ]n .
V (t1 , . . . ,tr ) =
X
659
The classes [Di ] are not linearly independent in H 2 (X , Q). There is also a
more efficient reduced volume polynomial, which is constructed using divisors Di
whose classes give a basis of H 2 (X , Q) = Pic(X )Q . In Example 13.4.5, for
instance, Pic(P n )Q has dimension 1, and the reduced volume polynomial is just
V (t) = t n . In Example 13.4.6, on the other hand, Pic((P1 )n )Q has dimension n and
the reduced volume polynomial is V (t1 , . . . ,tn ) = n!t1 tn .
The Volume Polynomial and the Cohomology Ring. An alternative description of
the cohomology ring H (X , Q) for complete simplicial toric varieties was proved
by Khovanskii and Pukhlikov in [174]. We will sketch the ideas following the
presentation in [170] with some modifications.
The relation between the volume polynomial and the cohomology ring will
come from the following algebraic fact.
L
Lemma 13.4.7. Let A = m
i=0 Ai be a finite-dimensional commutative graded
algebra over Q satisfying:
(a) A0 Am Q.
(b) A is generated by A1 as a Q-algebra.

(c) For all i = 0, . . . , m, the bilinear map
Ai Ami Am Q
(u, v) 7 uv
is nondegenerate.
Let s1 , . . . , sr span A1 and define
P(t1 , . . . ,tr ) = (t1 s1 + + tr sr )m Am Q,
which we regard as a polynomial in Q[t1 , . . . ,tr ]. Finally, define

n
o

I = f (x1 , . . . , xr ) Q[x1 , . . . , xr ] f
P=0 .
,...,
t1
tr
Then the map xi 7 si induces an isomorphism of graded Q-algebras
Q[x1 , . . . , xr ]/I A.

Here f t1 , . . . , tr is the differential operator obtained from f by replacing
each xi with t i . We will postpone the proof of the lemma until we see how it
applies to H (X , Q).
The rings satisfying the hypotheses in Lemma 13.4.7 are a special class of
finite-dimensional graded Gorenstein rings. If X is complete and simplicial, then
hypothesis (b) is satisfied by the cohomology ring H (X , Q) by Theorem 12.4.1.
It is easy to see that hypothesis (a) holds from the cohomology spectral sequence
in 12.3. Then hypothesis (c) follows from Poincare duality on X . As a result,
660
the cohomology rings of complete simplicial toric varieties are Gorenstein rings of
this type. Here is the Khovanskii-Pukhlikov presentation of the cohomology ring.
Theorem 13.4.8. Let X be a complete simplicial toric variety. Let r = |(1)| and
s = dim H 2 (X , Q) = dim Pic(X )Q . Then:
(a) Let V (t1 , . . . ,tr ) be the volume polynomial of X . Then there is an isomorphism
of Q-algebras
H (X , Q) Q[x1 , . . . , xr ]/I,

where I = f (x1 , . . . , xr ) Q[x1 , . . . , xr ] f t1 , . . . , tr V (t1 , . . . ,tr ) = 0 . In
addition, I equals the ideal I + J from Theorem 12.4.1.
(b) Let V (t1 , . . . ,ts ) be a reduced volume polynomial for X . Then there is an
isomorphism of Q-algebras,
H (X , Q) Q[x1 , . . . , xs ]/J,

where J = f (x1 , . . . , xs ) Q[x1 , . . . , xs ] f t1 , . . . , ts V (t1 , . . . ,ts ) = 0 .

Proof. The isomorphisms in parts (a) and (b) follow from Lemma 13.4.7, and the
equality I = I + J then follows from Theorem 12.4.1 (Exercise 13.4.3).

Example 13.4.9. For instance, if X = P n , then from Example 13.4.5
V (t0 , . . . ,tn ) = (t0 + + tn )n .
It is easy to check directly that the ideal I from Theorem 13.4.8 is
I = hx1 x0 , . . . , xn x0 , x0 xn i = I + J
(Exercise 13.4.4). Using the reduced volume polynomial V (t) = t n , one computes
that J = hx n+1 i. Hence we get isomorphisms
Q[x0 , . . . , xn ]/I Q[x]/hx n+1 i H (P n , Q).
Both isomorphisms were known previously; the surprise is seeing how they arise
from volume polynomials.
Example 13.4.10. The volume polynomial computed in Example 13.4.6 is

V (t1 , . . . ,t2n ) = n! (t1 + t2 ) (t2n1 + t2n ).
Here, one computes that the ideal I from Theorem 13.4.8 is
I = hx1 x2 , . . . , x2n1 x2n , x1 x2 , . . . , x2n1 x2n i = I + J
(Exercise 13.4.4). Furthermore, the reduced volume polynomial V (t1 , . . . ,tn ) =

n!t1 tn gives the ideal
J = hx12 , . . . , xn2 i.
This gives the presentation of the cohomology ring used in Example 13.4.6.
We now turn to the proof of Lemma 13.4.7.
661
Proof of Lemma 13.4.7. It is easy to prove that

o
n

,...,
P=0
I = f (x1 , . . . , xr ) f
t1
tr
is an ideal in Q[x1 , . . . , xr ].
Using hypothesis (a) and spanning set {s1 , . . . , sr } for A1 , we have a surjection
: Q[x1 , . . . , xr ] A,
f 7 f (s1 , . . . , sr ),
and we need to show ker() = I. Since P(t1 , . . . ,tr ) is a homogeneous polynomial

of degree m, I will be generated by homogeneous polynomials, and I will contain
all homogeneous polynomials of degree strictly larger than m. Note also that we
have the expansion
X
m!
P(t1 , . . . ,tn ) = (t1 s1 + + tr sr )m =
t11 trr s1 1 sr r .
!
1
r
++ =m
1
If f (x1 , . . . , xr ) is homogeneous of degree m, then a direct calculation shows that

,...,
P(t1 , . . . ,tr ) = m! f (s1 , . . . , sr )
f
t1
tr
(Exercise 13.4.5). Hence
f I f (s1 , . . . , sr ) = 0 f ker()
in this case.
Now assume f is homogeneous of degree p < m. Suppose that f
/ ker(),
so f (s1 , . . . , sr ) 6= 0. By hypothesis (c) in the Proposition, there is a homogeneous
polynomial g of degree m p such that g(s1 , . . . , sr ) f (s1 , . . . , sr ) 6= 0 Am . Thus,
by the first part of the proof, g f
/ I so f
/ I. Thus f I implies f ker(). Finally,
suppose f ker(), so f (s1 , . . . , sr ) = 0. You will show in Exercise 13.4.5 that

f
,...,
P(t1 , . . . ,tr ) = f (s1 , . . . , sr ) h(t1 , . . . ,tr , s1 , . . . , sr ),
t1
tr
where h(t1 , . . . ,tr , s1 , . . . , sr ) is given by
X
m!
1 ++r =mp
1 ! r !
(t1 s1 )1 (tr sr )r .
Since we assume f (s1 , . . . , sr ) = 0, this shows that f I.
In [170], Kaveh shows that the results of this section can be generalized to
spherical varieties, which are varieties with an action of a reductive algebraic group
G such that some Borel subgroup of G has a dense orbit.
662
Exercises for 13.4.

13.4.1. Show that Theorem 13.4.3 is also true if the divisor D is nef and Q-Cartier.
13.4.2. Check the assertions made in Example 13.4.4.
13.4.3. Complete the proof of part (a) of Theorem 13.4.8 by showing that if X is a
complete simplicial toric variety, then the ideal I from the theorem is the ideal I + J
from Theorem 12.4.1. Hint: Think about the kernel of the homomorphism Q[x1 , . . . , xr ]
H (X , Q) that takes xi to [Di ].
13.4.4. In this exercise, you will check some claims made in Examples 13.4.9 and 13.4.10.
(a) In Example 13.4.9, verify that
I = hx1 x0 , . . . , xn x0 , x0 xn i.
Hint: One inclusion is clear. For the other, show that the ideal on the right-hand
side is P
hx1 x0 , . . . , xn x0 , x0n+1 i. Also note that f Q[x0 , . . . , xn ] can be written
n
as f = i=1 (xi x0 )Ai (x0 , . . . , xn ) + g(x0 ). Remember that I is homogeneous.
(b) In Example 13.4.10, verify that I is as claimed.
(c) In both examples, compute the ideal J coming from the reduced volume polynomial.
13.4.5. In this exercise, you will verify some details of the proof of Lemma 13.4.7.
(a) Show that if f (x1 , . . . , xr ) is homogeneous of degree m, then

f
P(t1 , . . . ,tr ) = m! f (s1 , . . . , sr ).
,...,
t1
tr
(b) Show that if f (x1 , . . . , xr ) is homogeneous of degree p < m, then

f
P(t1 , . . . ,tr ) = f (s1 , . . . , sr )h(t1 , . . . ,tr , s1 , . . . , sr ),
,...,
t1
tr
where h(t1 , . . . ,tr , s1 , . . . , sr ) is given by
X
m!
(t1 s1 )1 (tr sr )r .
1 ! r !
1 ++r =mp
13.4.6. Consider the Hirzebruch surface Hr .

(a) Determine the volume polynomial and verify the isomorphism from Theorem 13.4.8
in this case.
(b) In the notation of Example 12.4.3, compute the reduced volume polynomial V (x3 , x4 )
using the basis of H 2 (Hr , Q) = Pic(Hr )Q given by D3 , D4 . Then verify that this gives
the presentation of the cohomology ring constructed in Example 12.4.3.
13.4.7. Let be a complete simplicial fan in NR Rn . Following [170], we present
another way to think about the reduced volume polynomial when X is projective.
(a) Let S be the set of all lattice polytopes with normal fan , modulo the equivalence
relation P P if P is a translate of P. Show that S forms a semigroup under the
operation of Minkowski sum of polytopes. Hint: Use Proposition 6.2.13.
(b) Let V be the vector space generated by S over Q. Show that there is a natural
isomorphism
V H 2 (X , Q).
13.5. The Khovanskii-Pukhlikov Theorem
663
(c) The normalized volume function Vol : S Q is homogeneous of degree n. Show

that there is V Sym n (V ) such that V : V Q extends Vol, i.e., V |S = Vol. Hint:
You may find it useful to first do part (d) of the exercise.
(d) Let s = dim H 2 (X , Q) and consider a reduced volume polynomial V (t1 , . . . ,ts ) for
X , which we construct using prime toric divisors Di that give a basis of H 2 (X , Q).
Show that this basis gives an isomorphism V Q s that takes V to V (t1 , . . . ,ts ).

Given a positive integer n and a function f on the interval [0, n], the Euler-Maclaurin
summation
R n formula relates the sum of the values of f at the integers 0, . . . , n to the
integral 0 f (x) dx. In [174], Khovanskii and Pukhlikov use Brions equalities and
Todd differential operators to give an analogous formula relating the sum of the
values of a suitable function f at the lattice points in a polytope to integrals over
polytopes. Specializing to the constant function f = 1 gives a formula for the
number of lattice points in a polytope. We will see that the Khovanskii-Pukhlikov
theorem gives another proof of Hirzebruch-Riemann-Roch for a smooth projective
toric variety.
The Euler-Maclaurin Formula. In 13.1, we saw that the Bernoulli numbers Bk
give the power series
(13.5.1)
X
x
Bk 2k
1
(1)k1
=
1
+
x
+
x .
x
1e
2
(2k)!
k=1
This series converges for all x C with |x| < 2. Bernoulli numbers also play a
key role in the Euler-Maclaurin formula, which can be stated as follows.
Theorem 13.5.1 (Euler-Maclaurin Summation). If f is a C function on [0, n],
then
Z n
1
1
f (x) dx = f (0) + f (1) + f (2) + + f (n 1) + f (n)
2
2
0
X
Bk
( f (2k1) (n) f (2k1) (0)) + R2 ,
(1)k
+
(2k)!
k=1
where R2 is a remainder term. Furthermore, if there are positive constants C and

< 2 such that | f () (x)| C for all x [0, n] and all 1, then
Z n
1
1
f (x) dx = f (0) + f (1) + f (2) + + f (n 1) + f (n)
2
2
0
X
Bk
(1)k
+
( f (2k1) (n) f (2k1) (0)).

(2k)!
k=1
664
The explicit form of the remainder is not important for us here; it and a proof
of the summation formula are given in [5]. The Euler-Maclaurin formula is often
used as a generalization of the trapezoidal rule for approximating the integral of
f . We should mention that there are a number of different competing conventions
for writing Bernoulli numbers, so Euler-Maclaurin formulas in other sources may
look different from Theorem 13.5.1.
Todd Operators. We will also use formal Todd differential operators obtained from
the series expansion in (13.5.1). The definition is given by
(13.5.2)
Todd(x) = 1 +
1 X
Bk 2k
(1)k1
+
.
2 x
(2k)! x 2k
k=1
For notational convenience, we will write Todd operators as
/x
,
1 e /x
but an expression of this form should always be interpreted as the series (13.5.2).
Todd(x) =
A key property of the Todd operator for us will be the equation

Todd(x)(e xz ) =
(13.5.3)
which follows easily by a direct calculation:
ze xz
,
1 e z
X
1
Bk 2k xz
Todd(x)(e ) = e + ze xz +
z e
(1)k1
2
(2k)!
xz
xz
k=1

X
1
Bk 2k
= e xz 1 + z +
(1)k1
z
2
(2k)!
k=1
z
.
= e xz
1 e z
This computation is valid for all z C satisfying |z| < 2.
A Consequence of Brions Equalities. Let P be a simple lattice polytope in MR .

Recall from Brions equalities (Corollary 13.2.10) that
P
m
X
X
m
v Q mPv M
(13.5.4)
=
.
n
mv,i )
i=1 (1
v vertex
mPM
Here, the mv,i are the minimal generators of the cone Cv = Cone(P M v), and
P
Pv = { ni=1 i mv,i | 0 i < 1}
is the fundamental parallelotope of the simplicial cone Cv . Equation (13.5.4) is an

equality of two elements of the localization Z[M]S . In 13.3, we mapped equations
b S that arises in equivariant cohomology via the
of this form to the localized ring
665
map m 7 e s(m) , and we used the resulting equations in our proof of the equivariant
Riemann-Roch theorem.
In this section, on the other hand, P
we will consider a different consequence
of Brions equalities. Take a point z = i zi ui NC = N Z C, where zi C and
uQi N. Each ui gives the one-parameter subgroup ui : C TN . The point pz =
ui zi
i (e ) TN depends only on z. Here, e is the base of the natural logarithm.
Then observe that for any m M, we have
m (pz ) = ehm,zi C .
Choose z NC so that hmv,i , zi

/ 2iZ for all mv,i in (13.5.4). Then evaluating
m
(13.5.4) at the point pz maps to ehm,zi and hence gives the relation
P
X
X
ehm,zi
hm,zi
hv,zi
(13.5.5)
e
=
e
Qn mPv M hm ,zi
v,i
)
i=1 (1 e
v vertex
mPM
since none of the denominators vanish.
This shows the power of the what we did in 13.2. The version of Brions
equalities proved there yields different equalities depending on how we evaluate
the m , giving results that can be used in very different contexts.
From Discrete to Continuous. Brions equalities in the form (13.5.5) deal with
discrete sums over sets of lattice points in polytopes. Our next results deal with
continuous analogs of these sums, namely integrals over polytopes, and their relation with discrete sums. For the application to the Khovanskii-Pukhlikov theorem,
we need to consider a class of polytopes more general than lattice polytopes.
Theorem 13.5.2. Let P MR be a full dimensional simple polytope, and assume
the rays of the vertex cones Cv are spanned by primitive vectors mv,i M for all
vertices v. If z NC satisfies hmv,i , zi
/ 2iZ for all mv,i , then
Z
X ehv,zi mult(Cv )
Qn
ehx,zi dx = (1)n
.
P
i=1 hmv,i , zi
v vertex
Proof. The integral on the left is equal to the limit of a sum as follows. Scale the
lattice M by a factor of 1k , and consider the points m P 1k M. Approximate P
by a collection of small cubes of side 1k centered at all m P 1k M. Then evaluate
the exponential function e hm,ui at these points and multiply by the volumes of the
cubes to form a Riemann sum. In the limit as k we have
Z
1 X hm,zi
ehx,zi dx = lim n
(13.5.6)
e
.
k k
P
1
mP k M
The next step is to apply (13.5.5) to P and the scaled lattice 1k M. The original
fundamental parallelotope Pv = Pv,M depends on M. For 1k M, the primitive ray generators are 1k mv,i, and it follows that the 1k M-lattice points of the new fundamental
666
parallelotope are given by

Pv, 1 M 1k M = 1k (Pv M)
(13.5.7)
(Exercise 13.5.1). When we combine this with (13.5.5), we see that the right-hand
side of (13.5.6) can be written as
P
P
hm/k,zi
X ehv,zi mP M ehm/k,zi
1 X hv,zi
mPv M e
v
lim
= lim
.
e
Qn
Qn
hmv,i /k,zi )
k kn
k
(1
e
k(1
ehmv,i /k,zi )
i=1
i=1
v vertex
v vertex
For the term in the sum for the vertex v, the limit of the numerator is
X
ehm/k,zi = ehv,zi |Pv M| = ehv,zi mult(Cv )
lim ehv,zi
k
mPv M
by Proposition 11.1.8. For the denominator
Qn
hmv,i /k,zi ),
i=1 k(1 e
note that
1 ehmv,i ,zi/k
= hmv,i, zi
k
1/k
lim k(1 ehmv,i /k,zi ) = lim
by LHopitals Rule. Reassembling the different parts of the computation, we get

the desired expression for the integral.

The Khovanskii-Pukhlikov Theorem. Let P be a simple lattice polytope defined
by inequalities of the form
hm, uF i + aF 0,
where uF are the facet normals. Let h = (hF )F facet be a vector with real entries
indexed by the facets F of P. Consider the polytope P(h) with shifted facets defined
by the inequalities
(13.5.8)
hm, uF i + aF + hF 0.
Note that the shifting factors hF for the different facets are independent. It is not
difficult to see that if all entries of h are sufficiently small in absolute value, then
P(h) is still simple (Exercise 13.5.2). If some of the hF are not rational, then the
vertices might not be rational points. We want to consider what happens when we
apply differential operators with respect to the hF to integrals over the corresponding polytopes, and then set h = 0.
The differential operators alluded to above are the formal multivariate Todd
differential operators defined using (13.5.2):
Todd(h) =
/hF
.
/hF
1
e
F facet
We are now ready to state the Khovanskii-Pukhlikov theorem for smooth lattice
polytopes, the main result of this section.
667
Theorem 13.5.3. Let P be a smooth lattice polytope. Then

Z
X

ehx,zi dx
=
ehm,zi
Todd(h)
h=0
P(h)
mPM
provided z NC satisfies 0 < |hmv,i , zi| < 2 for all primitive ray generators mv,i
of the vertex cones Cv of P.
Before we give the proof, we want to indicate exactly how this result relates to
the Euler-Maclaurin formula from the start of the section.
Example 13.5.4. Let P = [0, n] R. The endpoints of P are its facets, so the facet
normals are 1, and the facet equations of P are
hm, 1i = m 0 and hm, 1i = m n.
Hence, the shifted polytope P(h) is the interval P(h) = [h0 , n + h1 ] where h0 , h1
are real-valued independent variables. Fix z C such that 0 < |z| < 2 and let
f (x) = e xz . In Exercise 13.5.3, you will show that for h = (h0 , h1 ),
Z n
Z n+h1

1

Todd(h)
f (x) dx
f (x) dx + ( f (n) + f (0))
=
2
h0 =h1 =0
0
h0
(13.5.9)
X
Bk
( f (2k1) (n) f (2k1) (0)).
(1)k1
+
(2k)!
k=1
However, Theorem 13.5.3 implies that

Z n+h1

e xz dx
Todd(h)
h0 =h1 =0
h0
Since f (x) = e xz , we can write this as

Z n+h1

Todd(h)
f (x) dx
=
h0 =h1 =0
h0
m[0,n]Z
e mz .
m[0,n]Z
f (m) = f (0) + + f (n).
Combining this with (13.5.9) gives the Euler-Maclaurin formula

Z n
1
1
f (x) dx = f (0) + f (1) + f (2) + + f (n 1) + f (n)
2
2
0
X
Bk
(1)k
+
( f (2k1) (n) f (2k1) (0))
(2k)!
k=1
from Theorem 13.5.1 for f (x) = e xz when 0 < |z| < 2.
From this point of view, we see that Theorem 13.5.3 is the polytope version
of the Euler-Maclaurin formula for exponential functions of the form f (x) = ehx,zi .
In [174], Khovanskii and Pukhlikov study more general functions (polynomials
times exponentials) and prove a related result in this case. We should also mention
668
that there is a large and growing literature on various forms of generalized EulerMaclaurin formulas on polytopes. See the notes at the end of [22, Ch. 10] for
references.
Now we turn to the proof of Theorem 13.5.3.
Proof of Theorem 13.5.3. Since P is smooth, each vertex cone Cv has multiplicity
mult(Cv ) = 1. Then Theorem 13.5.2 implies that
Z
X
e hv,zi
Qn
e hx,zi dx = (1)n
.
P
i=1 hmv,i , zi
v vertex
Our first goal is to determine the corresponding equation with P replaced by P(h).
For this, we need to determine how the vertices of P(h) are related to the vertices
of P. First we note that since each facet of P shifts to a parallel facet in P(h), the
mv,i do not change, and hence Theorem 13.5.2 will apply to all P(h) as well.
If v is the intersection of facets F1 (v), . . . , Fn (v), then the ray generators mv,i of
the vertex cone at v are the basis of M dual to the basis uF1 (v) , . . . , uFn (v) of N. It
follows that
P
hv ni=1 hFi (v) mv,i , uFj (v)i = aFj (v) hFj (v) .
P
So the vertex v of P shifts to v ni=1 hFi (v) mv,i in P(h). For simplicity, we will
write hFi (v) as hi (v) in the following. Hence, applying Theorem 13.5.2 to P(h),
Z
P
X
1
e hv i hi (v)mv,i ,ui Qn
e hx,zi dx = (1)n
(13.5.10)
.
hm
,
zi
v,i
P(h)
i=1
v vertex
Now, we apply the Todd operator to both sides of (13.5.10). Since each summand of the right hand side factors as
P
P
e hv,zi
e h i hi (v)mv,i ,zi = cv e h i hi (v)mv,i ,zi ,
i=1 hmv,i , zi
Z

where cv does not depend on h, we have Todd(h)
e hx,zi dx =
Qn
P(h)
= (1)n
v vertex
(13.5.11)
= (1)n
v vertex
cv Todd(h) e h
i hi (v)mv,i ,zi
hmv,i , zi
,
hmv,i ,zi )
(1
e
i=1
n
P
Y
cv e h i hi (v)mv,i ,zi
where the second line follows by applying (13.5.3) for each hF (Exercise 13.5.4).
Setting hF = 0 for all F and some algebraic simplification between cv and the other
factor in each term yields
X
v vertex
e hv,zi
,
hmv,i ,zi )
i=1 (1 e
Qn
669
which is exactly the sum on the right in (13.5.5). (Recall that we are assuming
the vertex cones of P are smooth, so Pv M = {0} for all v). This completes the
proof
theorem because the other side of the equality in (13.5.5) is the sum
P of thehm,zi
e
and this is what we wanted.

mPM
Theorem 13.5.3 has the following nice consequence.
Corollary 13.5.5. For a smooth lattice polytope P

Todd(h) vol(P(h))
= |P M|,
h=0
where vol is the usual Euclidean volume.
Proof. Theorem 13.5.3 applies to the exponential function f (x) = ehx,zi for z NC
satisfying 0 < |hmv,i , zi| < 2 for all mv,i. Then, taking the limit as z 0 in the
theorem implies that

Z
X

ehx,0i dx
=
ehm,0i ,
Todd(h)
h=0
P(h)
mPM
which gives the desired result. Taking the limit inside the Todd operator and then
inside the integral takes some care. We omit the details.

Relation to HRR. It turns out that Corollary 13.5.5 can be used to prove HRR for
smooth projective toric varieties.
Theorem 13.5.6. Let L be a line bundle on a smooth projective toric variety

X = X . Then
Z
(L ) = ch(L ) Td(X ).
X
Proof. We will sketch the proof, though several points in the discussion
Prequire
justification that we will omit. We begin with a very ample divisor D = a D
and set P = PD . Then part (a) of Theorem 13.4.1 implies that
Z
[D]n = Vol(P).
(13.5.12)
X
Unlike part (b) of Theorem 13.4.1, the proof

P of part (a) does not use HRR. Note
also that this equation holds when D = a D is ample since replacing D with
D multiplies each side by n .
P The next step is to let h = (h )(1) , where each h is real, and set D(h) =
(a + h )D . The cohomology class [D(h)] lives in H (X , R), and the polytope
associated to D(h) is defined by
hm, u i + a + h 0,
(1),
670
which as above is denoted by P(h). While P(h) is no longer a lattice polytope, its
normal fan is still when h is small. (Earlier, we indexed h by facets of P, which
correspond to the rays since = P .) For h sufficiently small, (13.5.12) implies
Z
[D(h)]n = Vol(P(h))
(13.5.13)
X
for h rational (by rescaling) and then for h arbitrary (by continuity).
Note that the Todd operators Todd(h ) act naturally on cohomology classes
in H (X , R) that depend polynomially on h. An example is e h [D ] , which is a
polynomial in h since [D ] has degree 2. The basic relation (13.5.3) implies that
Todd(h ) e h [D ] ) =
Then we compute
[D ] e h [D ]
.
1 e [D ]
Y

Todd(h) e [D(h)] h=0 =
Todd(h ) e (a +h )[D ]
h=0
Y [D ] e a [D ]
1 e [D ]
= ch(L ) Td(X )
Q
since ch(L ) = e [ a D ] and Td(X ) = [D ]/(1 e [D ] ) by Theorem 13.1.6.
R
R
The integral X : H (X , R) R is a linear map, which implies that X and
R
Todd(h) commute when applied to e [D(h)] . Since X kills everything in degree
different from 2n, we obtain
Z
Z

ch(L ) Td(X ) = Todd(h) e [D(h)]
h=0
X
X

Z

e [D(h)]
= Todd(h)
h=0
X
Z

n
1
= Todd(h)
D(h)

n!
h=0
X

1

= Todd(h) n! Vol(P(h)) h=0 = Todd(h) vol(P(h)) h=0 ,
P
where the fourth equality follows from (13.5.13).
Note that P = PD is a smooth lattice polytope since D is ample and X is smooth.

Thus we can bring Corollary 13.5.5 into the picture, which implies

= |P M| = (OX (D)) = (L ).
Todd(h) vol(P(h))
h=0
Combining this with the previous display gives the Hirzebruch-Riemann-Roch

equality
Z
ch(L ) Td(X ) = (L )
(13.5.14)
X
671
in this special case when L = OX (D) and D is ample.

NowPpick divisors D1 , . . . , Ds whose classes give a basis of Pic(X ) and let
D(a) = si=1 ai Di and L (a) = OX (D(a)). It suffices to prove that
Z
ch(L (a)) Td(X ) = (L (a))
(13.5.15)
X
) Zs .
for all a = (a1 , , as

The left-hand side of (13.5.15) is clearly a polynomial
in a, and the same is true for the Euler characterstic (L (a)) on the right-hand
side. When D1 is very ample, we proved in Exercise 9.4.2 that (OX (a1 D1 ))
is a polynomial in a1 , and a similar result for a1 D1 + + as Ds (for arbitrary
D1 , . . . , Ds ) is proved in [176].
If a Zs is chosen so that D(a) is ample, then equality holds for (13.5.15) by
(13.5.14). However, since X is projective, the ample divisor classes are the lattice
points in the interior of the nef cone in Pic(X )R . Since two polynomials on Pic(X )R
that agree on the lattice points in an open cone must be equal (Exercise 13.5.5), we
conclude that (13.5.15) is true for all a Zs . This completes our second proof of
the Hirzebruch-Riemann-Roch theorem for toric varieties.

The version of HRR proved here is not as general as Theorem 13.3.8, which
only assumes that X is complete. Nevertheless, it is amazing how the relatively
elementary techniques leading to the Khovanskii-Pukhlikov theorem are so closely
related to a deep theorem about the geometry of toric varieties.
Exercises for 13.5.
13.5.1. Prove (13.5.7).
13.5.2. Show that if |hF | is sufficiently small for all F in (13.5.8), then P is still simple.
13.5.3. In this exercise, you will give two proofs of (13.5.9). Assume 0 < |z| < 2.
(a) Prove (13.5.9) using the power series definition of Todd(h) = Todd(h0 )Todd(h1 ).
R n+h
(b) Compute h0 1 e xz dx explicitly and then use (13.5.3) to show that

Z n+h1
e nz
1

=
e xz dx
Todd(h)
+
.
z
1e
1 ez
h0 =h1 =0
h0
Then apply (13.5.1) to the right-hand side to derive (13.5.9).
(c) Use Example 13.2.11 to show that right-hand side of the equation from part (b) is
equal to 1 + e z + + e nz . Thus we get a direct proof of Theorem 13.5.3 in this case.
13.5.4. This exercise deals with some details in the proof of Theorem 13.5.3.
(a) Show that the second line of (13.5.11) equals the first by using (13.5.3) for each hF .
(b) Show that the second line in (13.5.11) simplifies to yield the right-hand side of (13.5.5).
13.5.5. Assume that f R[x1 , . . . , xs ] vanishes on all lattice points in the interior of a cone
in Rs of dimension s. Prove that f = 0.
672

Here we collect some facts we need about Borel-Moore homology and generalized Gysin
maps. This material is discussed briefly in [105, Ch. 19] and in [110] in more detail. There
is also a nice treatment of Borel-Moore homology in [106, App. B].
Borel-Moore Homology. Besides ordinary homology, a suitably nice topological space X
has Borel-Moore homology groups defined by
HiBM (X, Q) = H ni (Rn , Rn \ X, Q)
for a closed embedding X Rn . These groups are independent of the embedding and are
defined for any variety X in its classical topology.
Proposition 13.A.1 (Basic Properties).
(a) If X is compact, then HiBM (X, Q) = Hi (X, Q).
(b) There is a cap product operation
BM
H i (X, Q) Q H BM
j (X, Q) H ji (X, Q),
7 a
such that
( ` ) a = a ( a )
for all , H (X, Q) and HBM (X, Q).
The functorial properties of Borel-Moore homology are more complicated since a

continuous map f : X Y does not always induce a map f : HiBM (X, Q) HiBM (Y , Q).
Proposition 13.A.2 (Functorial Properties).
(a) A proper map f : X Y induces f : HiBM (X, Q) HiBM (Y , Q) such that
f ( f () a ) = a f ()
for all H (Y , Q) and HBM (X, Q).
(b) If f : X Y and g : Y Z are proper, then (g f ) = g f .
(c) An inclusion j : U Y of an open set induces j ! : HiBM (X, Q) HiBM (U, Q) such that
j ! ( a ) = j () a j ! ()
for all H (X, Q) and HBM (X, Q).
(d) If j : U U and j : U X are open inclusions, then ( j j )! = j! j ! .
(e) If f : X Y is proper and j : U X is an open inclusion, then the diagram

HBM (X, Q)
j!
/ H BM ( f 1 (U), Q)

HBM (Y , Q)

/ HBM (U, Q)
commutes, where f = f | f 1 (U ) and j : f 1 (U) X is inclusion.
673
Fundamental Classes and Refined Cohomology Classes. A variety X of dimension n has

BM
a canonically defined homology class [X] H2n
(X, Q) called the fundamental class of X.
Furthermore, if X is irreducible, then
BM
H2n
(X, Q) = Q[X].
Example 13.A.3. Let X be a complete variety

R of dimension n. Then the constant map
p : X {pt} is proper. Hence we can define X : H (X, Q) Q such that the diagram
H (X, Q)
PPP R
PPPX
PPP
a [X ]
PP'

p
BM
BM
/
H (X, Q)
H ({pt}, Q) = Q
commutes, i.e.,
= p ( a [X]). We use this map frequently in Chapters 12 and 13.
When Y X is a closed subset, the cap product in Proposition 13.A.1 generalizes to

BM
H i (X, X \ Y , Q) Q H BM
j (X, Q) H ji (Y , Q),
7 a .
Applied to the fundamental class [X], this gives a map

H i (X, X \ Y , Q) H BM
ji (Y , Q),
7 a [X].
When X is irreducible and rationally smooth, we get some classical duality theorems.
Proposition 13.A.4 (Duality). If X is irreducible and rationally smooth of dimension n
and Y X is closed, then we have isomorphisms
BM
(a) Poincare Duality: a [X] : H i (X, Q) H2ni
(X, Q).
BM
(b) Alexander Duality: a [X] : H i (X, X \ Y , Q) H2ni
(Y , Q).
This proposition explains why we are using coefficients in Q. If we want Poincare and
Alexander duality to hold over Z, then we need to assume that X is smooth.
For X as in Proposition 13.A.4, let i : Y X be a d-dimensional irreducible subvariety.
BM
Then the fundamental class [Y ] H2d
(Y , Q) gives the following classes:
BM
The refined cohomology class [Y ]r H 2n2d (X, X \ Y , Q) maps to [Y ] H2d
(Y , Q)
under Alexander duality.
BM
The cohomology class [Y ] H 2n2d (X, Q) maps to i [Y ] H2d
(X, Q) under Poincare
duality.
The natural map H 2n2d (X, X \ Y , Q) H 2n2d (X, Q) takes [Y ]r to [Y ]. We use refined
cohomology classes in 12.4.
We will also need the following property of fundamental classes.

Proposition 13.A.5. If f : X Y is a proper birational morphism between irreducible
varieties, then f [X] = [Y ].

Generalized Gysin Maps. Let f : X Y be proper map such that Y is irreducible and
rationally smooth. Besides the usual contravariant map f : H (Y , Q) H (X, Q), there
674
is also a unique covariant map f ! : H (X, Q) H (Y , Q) such that the diagram

H (X, Q)
f!
/ H (Y , Q)

/ HBM (Y , Q)
a [X ]
a [Y ]

HBM (X, Q)
commutes. This follows since the vertical map on the right is an isomorphism. We call f !
a generalized Gysin map. These maps behave nicely as follows.
Proposition 13.A.6. Assume that f : X Y is proper and Y is irreducible and rationally
smooth. Then:
(a) f ! : H k (X, Q) H k+2 dim Y 2 dim X (Y , Q).
(b) If g : Y Z is proper and Z is irreducible and rationally smooth, then (g f )! = g! f ! .
(c) f ! : H (X, Q) H (Y , Q) satisfies
f ! ( f () ` ) = ` f ! ()
for all H (Y , Q) and H (X, Q).
In particular, taking = 1 in part (c) of the proposition gives f ! ( f ()) = ` f ! (1).

Here are two cases where f ! (1) is explicitly known for f as in Proposition 13.A.6:
If f : X Y is proper and birational, then f ! (1) = 1. Thus f ! ( f ()) = for all
H (Y , Q).
If i : Y X is the inclusion of an irreducible subvariety, then i! (1) = [Y ]. Thus

i! (i ()) = ` [Y ] for all H (X, Q).
In the second bullet, the map i! : H (Y , Q) H (X, Q) is called the Gysin map.
Here is an example of a generalized Gysin map.
Example 13.A.7. Let X be complete, irreducible and rationally smooth of dimension n.RIf
p : X {pt} is the constant map, then p! : H (X, Q) H ({pt}, Q) RQ is the map X
from Example 13.A.3. This follows easily
R from the definitions of p! and X . Note also that
this agrees with the ad-hoc definition of X given in 12.4.
We will need the following compatibility result between pullbacks and generalized
Gysin maps.
Proposition 13.A.8. Suppose we have a commutative diagram of maps
X
(13.A.1)
/ Y

X

/Y
where f is proper, Y and Y are irreducible and rationally smooth, and X = X Y Y . Then
we have a commutative diagram in cohomology
H (X , Q)
O
f !
H (X, Q)
/ H (Y , Q)
O
f!
/ H (Y , Q).
675
The diagram (13.A.1) is called Cartesian when X = X Y Y . It is a standard fact that

when (13.A.1) is Cartesian, f is proper whenever f is.
The complicated behavior of Borel-Moore homology and Gysin maps were one of the
factors that inspired Fulton and MacPherson to develop the bivariant theories introduced
in [110]. This is related to the bivariant intersection theory discussed in [107, Ch. 17].
Equivariant Gysin Maps. Let T be a torus acting on varieties X and Y and let f : X Y
be T -equivariant and proper. Our goal is to define an equivariant Gysin map
(13.A.2)
f! : HT (X, Q) HT (Y , Q).
For simplicity, we will assume that Y is a simplicial toric variety where T acts on Y via a
homomorphism T TN = the torus of Y .
To define f! , we follow [203, App.]. Recall that HT (Y , Q) = H (ET T Y , Q), where

T (C )m implies EG (C \ {0})m . This follows from (12.4.6), where we showed that
EC = C \ {0}. Then define the finite-dimensional approximation
ET (C+1 \ {0})m
of ET . Taking the quotient by T , we get the approximation BT (P )m of BT .
When Y is a simplicial toric variety, Y = ET T Y is irreducible and rationally smooth

for all , so that the induced proper map f : X Y has a generalized Gysin map
f! : H (X , Q) H (Y , Q).
Furthermore, the inclusions X X+1 and Y Y+1 give a Cartesian diagram
X
_
/ Y
_

X+1
f+1

/ Y+1
It follows from Proposition 13.A.8 that the maps f! are compatible as . Hence we
get the desired map (13.A.2).
Here are the basic properties of equivariant Gysin maps.
Proposition 13.A.9. Assume that f : X Y is proper and T -equivariant, and assume that
Y is a simplicial toric variety. Then:
(a) f ! : HTk (X, Q) HTk+2 dim Y 2 dim X (Y , Q).
(b) If g : Y Z is proper and T -equivariant and Z is a simplicial toric variety, then

(g f )! = g! f ! .
(c) f ! : HT (X, Q) HT (Y , Q) satisfies
f ! ( f () ` ) = ` f ! ()
for all HT (Y , Q) and HT (X, Q).
Equivariant Gysin maps also work nicely in Cartesian squares.
676
Proposition 13.A.10. Suppose we have a commutative diagram of maps

f
/ Y

X

/Y
such that f is proper, f and g are equivariant, Y and Y are simplicial toric varieties, and
X = X Y Y . Then we have a commutative diagram in equivariant cohomology
f !
HT (X , Q)
O
/ H (Y , Q)
T O
g
g
f!
HT (X, Q)
/ HT (Y , Q).
For the final property we need, let X = X be a complete simplicial toric variety. Then
the constant map p : X {pt} is proper and equivariant under the action of T = TN . Hence
we get
Z
= p! : HT (X, Q) HT ({pt}, Q).

R
This is the equivariant integral. We need to show that X eq is compatible with the ordinary
integral
Z
X eq
= p! : H (X, Q) H ({pt}, Q) = Q.
Here is the precise result we will use in 13.4.
Proposition 13.A.11. In the above situation, we have a commutative diagram

R
H (X, Q)
O
i
X
HT (X, Q)
/Q
O
i
pt
eq
/ H ({pt}, Q).
T
Proof. We will prove this using the commutative diagram

X
_

X
/ {pt}
_

/ BT ,
where X is a complete simplicial toric variety. Here, p : X = ET T X BT is induced

by projection on the first factor and pt is any point of BT . This is a Cartesian diagram, so
that by Proposition 13.A.8, we get a commutative diagram
p!
H (X, Q)
O
i
pt
i
X
H (X , Q)
/Q
O
p !
/ H (BT , Q).
where iX : X X and ipt : {pt} BT are the inclusions. Letting , we get the
desired commutative diagram.
Chapter 14
Toric GIT and the

Secondary Fan
This chapter will explore a rich collection of ideas that give different ways to think
about toric varieties. We begin in 14.1 with the geometric invariant theory of a
closed subgroup G (C )r acting on Cr , which uses a character of G to lift the
G-action to a trivial line bundle over Cr . In 14.2 we show that the GIT quotient
is a semiprojective toric variety in this situation, and in 14.3 we use Gale duality
to help us understand how the quotient depends on the character . The full story
of what happens as varies is controlled by the secondary fan, which is the main
topic of 14.4. The geometry of the secondary fan will be explored in Chapter 15.
14.1. Introduction to Toric GIT

Geometric invariant theory was invented by Mumford [209] in 1965 to prove the
existence of suitable projective compactifications of the moduli spaces he was
studying. GIT, as it is now called, is a powerful tool in modern algebraic geometry.
See [83] for a nice introduction to the subject.
We will consider the special case of a closed subgroup G (C )r acting on
C = Spec(C[x1 , . . . , xr ]). Thus G (C ) H, where H is finite. Since G is
reductive, Proposition 5.0.9 gives the good categorical quotient
r
Cr //G = Spec(C[x1 , . . . , xr ]G ).
However, such quotients can behave badly, as shown by the diagonal action of C
on Cr . The only invariants are the constant polynomials, so that
Cr //C = Spec(C[x1 , . . . , xr ]C ) = Spec(C) = {pt}.

677
678
Chapter 14. Toric GIT and the Secondary Fan
A better model for the kind of quotient we want comes from the quotient construction of a toric variety X without torus factors. Here, Theorem 5.1.11 gives
the almost geometric quotient
X (C(1) \ Z())//G.
Thus, if we take C(1) and throw away some points, we get an almost geometric
quotient, which by definition means that removing further points gives a geometric
quotient. As we will see, this is similar to what happens in GIT, where (roughly
speaking) we first remove points that leave us with the set of semistable points, and
then we remove even more points to get the set of stable points.
Linearized Line Bundles. In GIT, deciding which points to remove is governed
by a lifting of the G-action on Cr to the rank 1 trivial vector bundle Cr C Cr .
Liftings are described by characters of G. Let the character group of G be
b = { : G C | is a homomorphism of algebraic groups}.
G
b gives the action of G on Cr C defined by

Then a character G
g (p,t) = (g p, (g)t),
g G, (p,t) Cr C.
This lifts the G-action on Cr , and all possible liftings arise this way.
Let L denote the sheaf of sections of Cr C with this G-action. We call L

the linearized line bundle with character . Note that for d Z, the tensor product
Ld is the linearized line bundle with character d , i.e., Ld = L d .
If we forget the G-action, then L OCr as a line bundle on Cr . It follows
that a global section s (Cr , L ) can be written
s(p) = (p, F (p)) Cr C,
p Cr ,
for a unique F C[x1 , . . . , xr ]. The group G acts on global sections as follows.
b
Lemma 14.1.1. Let L be the linearized line bundle on Cr with character G.
(a) If s is the global section of L given by F, then for any g G, g s is the global
section defined by
(g s)(p) = (p, (g) F (g1 p)),
p Cr .
(b) The G-invariant global sections are described by the isomorphism

(Cr , L )G {F C[x1 , . . . , xr ] | F(g p) = (g) F(p) for g G, p Cr }.
Proof. By Exercise 14.1.1, g G acts on a global section s by
(g s)(p) = g (s(g1 p)),
p Cr .
Since s(p) = (p, F(p)), part (a) follows immediately, and then we are done since
part (b) is an easy consequence of part (a).

Definition 14.1.2. The polynomials in part (b) of the lemma are (G, )-invariant.
679
c Z, where
Example 14.1.3. Consider the usual action of C on Cr . Then C
d Z gives the character (t) = t d for t C . One checks that (Cr , L )G is
isomorphic to the vector space of homogeneous polynomials of degree d.
Example 14.1.3 has a nice toric generalization.

Example 14.1.4. Let X be a toric variety with no torus factors. The class group
Cl(X ) gives the algebraic group G = HomZ (Cl(X ), C ), and using the map
Z(1) Cl(X ) from 4.1, we can regard G as a subgroup of C(1) .
The total coordinate ring of C(1) is S = C[x | (1)], which is graded by
b Cl(X ), where a divisor class Cl(X )
Cl(X ) as in 5.2. Note also that G
b defined by evaluation at . In Exercise 14.1.2 you will
gives the character G
construct an isomorphism
(C(1) , L )G S = {F S | deg(F) = }.
Thus S is the set of all (G, )-invariant polynomials.
Here is an example that will appear several times in this section and the next.
Example 14.1.5. Let G = {(t,t 1 , u) (C )3 | t C , u = 1} C 2 . One
easily sees that
C[x, y, z]G = C[xy, z 2 ],
so that C3 //G C2 .
b Z (Z/2Z) defined by (t,t 1 , u) = tu.
Now consider the character G
Then a monomial x a y b z c is (G, )-invariant if and only if
(tx)a (t 1 y)b (uz)c = tu x a y b z c t ab u c = tu a = b + 1, c 1 mod 2.
It follows easily that (C3 , L )G xz C[xy, z 2 ].
Semistable and Stable Points. Given a global section s of L , note that

(Cr )s = {p Cr | s(p) 6= 0}
is an affine open subset of Cr since s(p) = (p, F(p)) and s(p) 6= 0 means F(p) 6= 0.
Also observe that G acts on (Cr )s when s is G-invariant.
b with linearized line bundle L .
Definition 14.1.6. Fix G (C )r and G,
(a) p Cr is semistable if there are d > 0 and s (Cr , L d )G such that p (Cr )s .
(b) p Cr is stable if there are d > 0 and s (Cr , L d )G such that p (Cr )s ,
the isotropy subgroup G p is finite, and all G-orbits in (Cr )s are closed in (Cr )s .
r s
(c) The set of all semistable (resp. stable) points is denoted (Cr )ss
(resp. (C ) ).
Since a global section s (Cr , L d )G corresponds to a (G, d )-invariant

polynomial F, one can determine semistability and stability using (G, d )-invariant
polynomials. We will do this frequently in the remainder of the chapter.
680
In general treatments of GIT, where G acts on a variety X , one must also require
that the nonvanishing subset of the section s be affine in the definition of semistable
and stable point. This is automatic in our case. See [83, Ch. 8] and [209].
Example 14.1.7. Consider the usual action of C on Cr and let d , d Z, be as in
Example 14.1.3. Then one can check without difficulty that
r s
r
d > 0 :(Cr )ss
d = (C ) d = C \ {0}
r
r s
d = 0 :(Cr )ss
d = C , (C ) d =
r s
d < 0 :(Cr )ss
d = (C ) d = .
This shows that the notions of semistable and stable depend strongly on which
character we use.
Example 14.1.8. For G = {(t,t 1 , u) | t C , u = 1} and (t,t 1 , u) = tu from

Example 14.1.5, the (G, d )-invariant polynomials are
(
xd z C[xy, z 2 ] d odd
(14.1.1)
(C3 , L d )G d
x C[xy, z 2 ] d even.
In particular, x 2 is (G, 2 )-invariant, which implies that all points in C C2 are
semistable. With more work (Exercise 14.1.3), one can show that
2
(C3 )s = (C3 )ss
= C C .
For a toric variety X , the group G (C )(1) acts on C(1) as described in

b Cl(X ).
Example 14.1.4, and characters correspond to divisor classes since G
For the character of an ample class, semistable and stable have a nice meaning.
Proposition 14.1.9. Let X be a projective toric variety, and let be the

b comes from an ample
subfan consisting of all simplicial cones of . If G
divisor class Cl(X ), then
ss
C(1) = C(1) \ Z()
s
C(1) = C(1) \ Z( ).
P
Proof. Let = [D], where D = a D is ample. Then the polytope PD MR
has facet presentation
PD = {m MR | hm, u i a for all (1)},
and the vertices of PD give the Cartier data {m }(n) of D. Also recall from 5.4
that if S = C[x | (1)], then the graded piece S is spanned by the monomials
Q hm,u i+a
xhm,Di = x , m PD M.
(1)
\ Z(). Since Z() is defined by the vanishing of x =
Q Now take p C
x , (n), there must be (n) such that x does not vanish at p.
(1)
/
681
We claim that xhm ,Di S does not vanish at p. Since xhm ,Di is (G, )-invariant
by Example 14.1.4, this will imply that p is semistable. To prove our claim, recall
from the discussion before Example 5.4.5 that the exponent of x in xhm,Di is the
lattice distance from m to the facet of PD whose normal is u . For m , the lattice
distance is zero when (1) since these s give the facets containing the vertex
m , and all other lattice distances are positive. Thus xhm ,Di and x involve the
same variables, so that xhm ,Di does not vanish at p since the same is true for x .
Next assume p C(1) is semistable. Then there is d > 0 such that some
element of Sd does not vanish at p. It follows that xhm,dDi is nonzero at p for some
m (dPD ) M. Let Q dPD be the smallest face of dPD containing m and let dm
be a vertex of this face. We claim that x divides xhm,dDi . Since xhm,dDi does not
vanish at p, this will imply that the same is true for x , and p C(1) \ Z() will
follow. To prove our claim, take
/ (1) and consider the facet F dPD with facet
normal u . If m F, then Q F, which would give dm F. This contradicts
/ (1) by the definition of normal fan, so that m has positive lattice distance to
the facets corresponding to
/ (1). It follows that x divides xhm,dDi .
It remains to consider stable points. Since (1) = (1) (rays are simplicial),
X and X have the same class group and the same group G. The difference is that
implies an inclusion of irrelevant ideals B( ) B(), which in turn gives
the inclusion
C(1) \ Z( ) C(1) \ Z().
Since is simplicial, Theorem 5.1.11 implies that X (C(1) \ Z( ))//G is a

geometric quotient. It follows that G p is closed for all p C(1) \ Z( ). Furthermore, Exercise 5.1.11 implies that the isotropy subgroup G p is finite for these
s
ps since is simplicial. Thus p is stable, hence C(1) \ Z( ) C(1) .
s
For the opposite inclusion, let U = C(1) be the set of stable points. Using
the trivial character, the action of (C )(1) lifts to an action on L that is easily
seen to commute with the action of G on L . This induces an action of (C )(1)
on (C(1) , L d )G for every d N. It follows that if p C(1) is stable, so is
its (C )(1) -orbit. In other words, (C )(1) acts on U . Note that we also have
U C(1) \ Z() since stable points are semistable.
Now consider the quotient map
: C(1) \ Z() X = (C(1) \ Z())//G.

Because G-orbits in U are closed, Theorem 5.0.6 and Proposition 5.0.7 imply that
(U ) X is open. By the previous paragraph, (U ) is stable under the action of
TN = (C )(1) /G and hence is a toric variety. This means it comes from a subfan
of . Note also that |U : U (U ) is a geometric quotient since G-orbits are
closed in U . By Theorem 5.1.11, this subfan is simplicial and hence is contained
in . It follows easily that U C(1) \ Z( ), and we are done.
682
b our next task is to define the GIT

GIT Quotients. Given G (C )r and G,
ss
quotient Cr // G. The basic idea is to take the quotient of C(1)
under the
action of G. As shown by the non-separated quotient from Example 5.0.15, care
must be taken to ensure that the quotient is well-behaved.
The strategy will be to use the graded ring
(14.1.2)
R =
(Cr , L d )G .
d=0
We first study the structure of this ring.

Lemma 14.1.10. The graded ring R is a finitely generated C-algebra.
Proof. Consider the action of G on Cr C given by g (p,t) = (g p, 1 (g)t).
Then G acts on f C[x1 , . . . , xr , w] by
(g f )(p,t) = f (g1 (p,t)) = f (g1 p, 1 (g1 )t) = f (g1 p, (g)t).
In particular, if F C[x1 , . . . , xr ], then f = F w d is G-invariant if and only if for all

(p,t) Cr C and all g G, we have
f (g1 p, (g)t) = f (p,t) F(g1 p)((g)t)d = F(p)t d
F(g1 p) = ((g))d F(p) = d (g1 ) F(p).
Replacing g with g1 , we see that f = F w d is G-invariant if and only if F is

(G, d )-invariant. Using Lemma 14.1.1, we obtain an isomorphism
(14.1.3)
R C[x1 , . . . , xr , w]G .
Since G is reductive, R is a finitely generated C-algebra by Proposition 5.0.9
We now define the GIT quotient using the Proj construction described in 7.0.
b the GIT quotient Cr // G is
Definition 14.1.11. For G (C )r and G,
Cr // G = Proj(R ).
Here are some easy properties of GIT quotients.

b we have:
Proposition 14.1.12. For G (C )r and G,
(a) There is a projective morphism Cr // G Cr //G = Spec(C[x1 , . . . , xr ]G ).
(b) Cr // G 6= if and only if (Cr )ss

6= .
(c) The GIT quotient Cr // G is a good categorical quotient of (Cr )ss

under the
r
r ss
action of G, i.e., C // G (C ) //G.
(d) If (Cr )s 6= , then the action of G on (Cr )s has a geometric quotient (Cr )s /G
isomorphic to a nonempty open subset of Cr // G. Thus Cr // G (Cr )ss
//G
r
is an almost geometric quotient and dim C // G = r dim G.
683
Proof. The map Proj(R ) Spec((R )0 ) is projective by Proposition 7.0.9, and

then part (a) follows since (R )0 = C[x1 , . . . , xr ]G . For part (b), first note that
(Cr )ss
= (R )d = 0 for d > 0.
The proof of Lemma 14.1.10 shows that R is an integral domain, and then one
sees easily that Proj(R ) = if and only if (R )d = 0 for d > 0 (Exercise 14.1.5).
For part (c), let R = C[x1 , . . . , xr , w]G be the ring of invariants introduced in
Lemma 14.1.10, which we grade by degree in w. The isomorphism of graded
rings R R from (14.1.3) implies that Cr // G Proj(R). Recall from 7.0 that
Proj(R) is covered by open subsets D+ ( f ) = Spec(R( f ) ), where f Rd is nonzero
with d > 0 and
nh
o
(14.1.4)
R( f ) =
|
h
S
,
0
.
d
f
Now suppose that f = F w d , where F C[x1 , . . . , xr ] is (G, d )-invariant. Then

F corresponds to global section s (Cr , L d )G by Lemma 14.1.1. Furthermore,
the nonvanishing set (Cr )s Cr of s is
(Cr )s = (Cr )F = Spec(C[x1 , . . . , xr ]F ),
where C[x1 , . . . , xr ]F is localization at F. In Exercise 14.1.6 you will show that

(14.1.5)
H
H w d
H w d
=
F
(F w d )
f
induces an isomorphism (C[x1 , . . . , xr ]F )G R( f ) . This implies that
D+ ( f ) = Spec(R( f ) ) Spec((C[x1 , . . . , xr ]F )G ) = (Cr )s //G.
Then part (c) follows since Proj(R) is covered by the open subsets D+ ( f ) and
r
(Cr )ss
is covered by the affine open subsets (C )s . See [83, Sec. 8.2] for the details.
The first assertion of part (d) is a standard result in GIT (see [83, Thm. 8.1]).
r
Thus the quotient map : (Cr )ss
C // G of part (c) is an almost geometric
quotient as defined in 5.0. Finally, if p is a stable point, then 1 ((p)) = G p
G/G p , which has dimension dim G since p is stable. Then the dimension formula
r
dim (Cr )ss
= dim C // G + dim generic fiber of
(see [245, Thm. I.6.7]) shows that dim Cr // G = r dim G.
Remark 14.1.13. Part (d) of Proposition 14.1.12 has two useful consequences:
(a) If stable points exist, then Cr // G has the expected dimension r dim G. In
general, we always have dim Cr // G r dim G (Exercise 14.1.4).
(b) G-orbits of stable points are closed in (Cr )ss

(Exercise 14.1.4).
The properties of Cr // G stated in Proposition 14.1.12 apply to more general

GIT quotients, as explained in [83] and [209].
Here are two illustrations of Proposition 14.1.12.
684
Example 14.1.14. For G = {(t,t 1 , u) | t C , u = 1} and (t,t 1 u) = tu, we

have (C3 )ss = C C2 by Example 14.1.8. Since this is affine, Proposition 14.1.12
implies that
C3 // G (C C2 )//G = Spec(C[x1 , y, z]G ).
The ring of invariants is easily seen to be C[xy, z 2 ], so that
C3 // G Spec(C[xy, z 2 ]) C2 .
Working directly from C3 // G = Proj(R ) is harder since
R = C[xy, z 2 ] xz C[xy, z 2 ] x 2 C[xy, z 2 ] x 3 z C[xy, z 2 ] x 4 C[xy, z 2 ]
by (14.1.1). It is not obvious that Proj of this ring gives C2 . We will see in 14.2
that the complications of this example come from a polyhedron whose vertices are
not lattice points.
Example 14.1.15. Consider G = {(t,t, u, u) | t, u C } and (t,t, u, u) = t. Then

a polynomial F(x, y, z, w) is (G, d )-invariant if and only if
F(tx,ty, uz, uw) = t d F(x, y, z, w),
i.e., if and only if F has degree (d, 0) in the grading where x, y have degree (1, 0)
and z, w have degree (0, 1). It follows that as graded rings,
R C[x, y].
Thus C4 // G = Proj(C[x, y]) = P1 . This is disconcerting since C4 has dimension 4
and G has dimension 2, yet the quotient only has dimension 1. The reason is that
there are no stable points. In fact,
2
2
4 s
(C4 )ss
= C (C \ {0}), (C ) = .
Note that (C4 )s = follows from part (d) Proposition 14.1.12 since the quotient
does not have the expected dimension.
In contrast, if one uses the character defined by (t,t, u, u) = t a u b for a, b > 0,
then C4 // G = P1 P1 and (C4 )s 6= . Once we introduce the secondary fan in
14.4, this example will be easy to understand.
The GIT quotients in Examples 14.1.14 and 14.1.15 are toric varieties. This is
no accident, as we will prove in the next section.
More General Toric GIT Quotients. So far, we have studied the quotient of Cr
by a subgroup G (C )r . An obvious generalization would be to consider the
quotient of a toric variety X by a subgroup G TN . This question has been studied
extensively in the literature. For more details about these quotients, interested
readers should consult the papers [3] by ACampo-Neuen and Hausen, [149] by
Hu, and [164] by Kapranov, Sturmfels and Zelevinsky.
14.2. Toric GIT and Polyhedra
685
Exercises for 14.1.

14.1.1. Let : V X a vector bundle over a variety X. Suppose that an affine algebraic
group G acts algebraically on X and V such that is equivariant.
(a) Given a global section s : X V of , show that
(g s)(p) = g (s(g1 p))
defines a global section of .
(b) Show carefully that g (h s) = (gh) s for g, h G.
14.1.2. Prove the isomorphism (Cr , L )G S from Example 14.1.4.
2
14.1.3. Prove that (C3 )s = (C3 )ss
= C C in Example 14.1.8.
14.1.4. This exercise concerns Remark 14.1.13.

(a) Prove that dim Cr // G r dim G. Hint: Use [245, Thm. I.6.7].
(b) Prove that G-orbits of stable points are closed in (Cr )ss
. Hint: Use part (d) of Proposition 14.1.12 and remember that fibers are closed.
L
14.1.5. Let R = d=0 Rd be a graded integral domain. Prove that Proj(R) = if and only
if Rd = 0 for all d > 0.
14.1.6. Prove that the map (14.1.5) in the proof of Proposition 14.1.12 induces a ring
isomorphism (C[x1 , . . . , xr ]F )G R( f ) .
14.1.7. Prove that if C[x1 , . . . , xr ]G = C, then Cr // G is projective. The converse is more
subtle, as you will learn in the next two exercises.
14.1.8. Let G = {(t,t, 1) | t C } (C )3 and (t,t, 1) = t 1 .
(a) Show that (R )0 = C[x, y, z]G = C[z] and that (R )d = 0 for d > 0.
(b) Conclude that the converse of Exercise 14.1.7 is false. Hint: is a projective variety.
14.1.9. The converse of Exercise 14.1.7 is false by Exercise 14.1.8. Fortunately, once we
assume that Cr // G 6= , everything is fine.
(a) Let U V be a morphism of irreducible varieties such that the map C[V ] C[U] is
injective. Prove that the image of U is Zariski dense in V .
(b) Suppose (R )d 6= 0 for some d > 0. Use part (a) to prove that Proj(R ) Spec((R )0 )
has Zariski dense image. Hint: Construct a map (R )0 (R )( f ) for f 6= 0 in (R )d .
(c) Prove that if Cr // G is projective and nonempty, then C[x1 , . . . , xr ]G = C.
b
14.1.10. Prove that Cr // G Spec(C[x1 , . . . , xr ]G ) when is a torsion element of G.
b is a
We continue to assume that G (C )r is an algebraic subgroup and G
character. A full understanding of the GIT quotients introduced in 14.1 involves
the interesting polyhedra associated to characters.
686
The Group G. Before discussing polyhedra, we need to study G. Since Zr is

the character group of (C )r , the inclusion G (C )r induces a homomorphism
b whose kernel we denote by M. Hence we have an exact sequence
: Zr G
(14.2.1)
b
0 M Zr G.
b will be written (a) = a . The dual of is a map

The image of a Zr in G
Zr N, where N is the dual of M. The images of the standard basis e1 , . . . , er Zr
give elements 1 , . . . , r N, and the map in (14.2.1) can be written
(m) = (hm, 1 i, . . . , hm, r i),
m M.
Maps like this appear in the exact sequence (4.1.3) for the class group of a toric
variety, where the u , (1), play the role of the i . However, we will see in
Example 14.2.4 below that the i are not always as nice as the u .
Here is a basic result about G and its character group.
Lemma 14.2.1. Let TN = N Z C = HomZ (M, C ).
(a) The map in (14.2.1) is surjective, so that we have an exact sequence
(14.2.2)
b 0.
0 M Zr G
(b) (C )r /G TN , so that we have an exact sequence of algebraic groups
1 G (C )r TN 1.
Q
hm, i
(c) G = {(t1 , . . . ,tr ) (C )r | ri=1 ti i = 1 for all m M}.
Proof. We begin with part (b). Since G is reductive and G-orbits are closed in
(C )r , we have a geometric quotient (C )r /G = Spec(C[x11 , . . . , xr1 ]G ). One
b is the trivial
checks that Laurent monomial x b is G-invariant if and only (b) G
1
G
1
character. By (14.2.1), it follows that C[x1 , . . . , xr ] = C[M]. Part (b) follows
immediately since TN = Spec(C[M]).
For part (c), consider (C )r TN = HomZ (M, C ). In Exercise 14.2.1 you
will show that this map takes (t1 , . . . ,tr ) (C )r to HomZ (M, C ) defined by
Q
hm, i
(m) = ri=1 ti i . Then we are done since G is the kernel of this map by part (b).
Finally, for part (a), let n be the rank of M, which is also the rank of . Then
pick bases of M and Zr such that has Smith normal form
d1
0
...
.
.
= .
dn
.,
..
.
0
where d1 |d2 | |dn are positive and all other entries are zero. Using the dual basis
e1 , . . . , en of N, this means that is given by i = di ei for 1 i n and i = 0 for
687
i > n. Then applying part (c) gives a description of G which makes it obvious that
characters of G extend to characters of (C )r (Exercise 14.2.1).

The Polyhedron of a Character. We will give two models of the polyhedron of
b The first model is less intrinsic but has a nice relation to the treatment of
G.
polytopes and polyhedra given in Chapters 2 and 7.
b where a = (a1 , . . . , ar ) Zr , define the polyhedron
For a character = a G,
(14.2.3)
Pa = {m MR | hm, i i ai , 1 i r} MR .
For fixed , the various choices of a differ by elements of (M), so the polyhedra
Pa are translates of each other by elements of M.
Here is an important property of these polyhedra.
Lemma 14.2.2. If Pa 6= , then the recession cone of Pa is strongly convex.
Proof. Assume that Pa from (14.2.3) is nonempty. Then its recession cone is C =
{m MR | hm, i i 0, 1 i r}, so that C (C) is defined by hm, i i = 0 for
1 i r. Thus C (C) = {0} since the map in (14.2.2) is injective.

Recall from 7.1 that P MR is a lattice polyhedron when its recession cone
is strongly convex and its vertices are lattice points. Furthermore, when dim P =
dim MR , we constructed the toric variety XP with torus TN . Lemma 14.2.2 shows
that Pa has the right kind of recession cone. Unfortunately, it may have the wrong
dimension and its vertices may fail to be lattice points. Here are some examples.
Example 14.2.3. Consider G = {(t,t, u, u) | t, u C } and (t,t, u, u) = t. Then
= a for a = (1, 0, 0, 0), and the polyhedron Pa is the line segment
Pa = Conv(0, e1 ) MR = R2
(Exercise 14.2.2). Hence Pa is not full dimensional.
Example 14.2.4. For the group G = {(t,t 1 , u) | t C , u = 1} (C )3 , the

exact sequence (14.2.2) is
0 Z2 Z3 Z (Z/2Z) 0,
where (m) = (hm, e1 i, hm, e1 i, hm, 2e2 i) and (a, b, c) = (a b, c mod 2). Thus
1 = 2 = e1 and 3 = 2e2 , so these i are neither distinct nor primitive. This
contrasts with toric case, where the i are the ray generators of a fan and hence are
distinct and primitive.
The character (t,t 1 u) = tu from Example 14.1.14 is = a for a = (1, 0, 1).
Then Pa MR = R2 is defined by the inequalities
hm, e1 i 1, hm, e1 i 0, hm, 2e2 i 1.
Figure 1 on the next page shows Pa . Note that the vertex of Pa is not a lattice point.
We also see that the first inequality defining Pa is redundant.
688
Pa
(0,1/2)
Figure 1. The polyhedron Pa R2 for a = (1, 0, 1) in Example 14.2.4
Example 14.2.5. For the usual action of C on Cr , the exact sequence (14.2.2) is
0 Zr1 Zr Z 0,
where is defined using i = ei , 1 i r 1 and r = e1 er1 .

Let = a for a = (0, . . . , 0, d) Zr . Then:
dr1 d > 0
Pa = {0}
d=0
d < 0.
Theorem 14.2.13 below will explain how this relates to Example 14.1.7.
b has a second, more intrinsic model P described as

The polyhedron of G
follows. Tensoring (14.2.2) with R gives an exact sequence of vector spaces
R
R
bR 0,
(14.2.4)
0 MR
Rr
G
bR = G
b Z R. Also let 1 G
bR be the image of under the map G
bG
bR .
where G
b has torsion, which happens precisely
Note that this map is not injective when G
when G is not a torus.
Then we get the polyhedron P Rr defined by
(14.2.5)
P = {b Rr0 | R (b) = 1} = R1 ( 1) Rr0 .
It is straightforward to show that if = a for a Zr , then

P = R (Pa ) + a.
In other words, if a = (a1 , . . . , ar ), then

(14.2.6)
P = {(hm, 1 i + a1 , . . . hm, r i + ar ) | m Pa }
(Exercise 14.2.3). The key point is that the inequalities (14.2.3) defining Pa are
equivalent to saying that (hm, 1 i + a1 , . . . hm, r i + ar ) lies in Rr0 .
Via (14.2.6), lattice points of Pa correspond to the points
R (Pa M) + a = 1 () Nr R1 ( 1) Nr = P Zr .
689
One subtle feature of P is that the inclusion R (Pa M) + a P Zr is strict

b has torsion. Here is an example.
when G
Example 14.2.6. For the polyhedron Pa shown in Figure 1 of Example 14.2.4, we

have a = (1, 0, 1) and 1 = 2 = e1 , 3 = 2e2 . Thus
P = {(hm, e1 i + 1, hm, e1 i, hm, 2e2 i + 1) | m Pa }.
Figure 2 shows Pa and P . The vertex (1, 0, 0) P Z3 does not come from a
lattice point in Pa M.
Pa
(0,1/2)
(1,0,0)
Figure 2. The polyhedra Pa and P
In the literature, this difficulty is often avoided by assuming that G is a torus,

b is torsion-free (see, for example, [137]).
so that G
Our goal is to prove that Cr // G is a toric variety. The rough intuition is that
Cr // G is the toric variety of Pa MR . However, we cant apply the theory of
Chapter 7 directly since Pa may have the wrong kind of vertices and the wrong
dimension. Our next task is to develop some tools to address these problems.
Veronese Subrings. The polyhedron Pa MR need not be a lattice polyhedron
since its vertices may not be lattice points. However, the vertices are rational, so
some integer multiple will be a lattice polyhedron. We will see that when translated
into algebra, this gives a Veronese subring of a graded semigroup algebra.
Before defining Veronese subrings in general, we recall a classic example that
explains where the name comes from.
Example 14.2.7. For a positive integer , the line bundle OPn () on P n gives the
n+
th Veronese embedding P n X P( n )1 . If we let C[X ] be the homogeneous
n+
coordinate ring of X as a subvariety of P( n )1 , then
P n X Proj(C[X ]).
However, the total coordinate ring of P n as a toric variety is S = C[x0 , . . . , xn ], and
the th Veronese embedding is determined by the polytope P = n , where n is
690
the standard n-simplex. Since P is normal, Theorem 5.4.8 implies that

C[X ] = C[XP ]
M
d=0
Sd
Sd = S.
d=0
This is the th Veronese subring of S. When we combine the above two displays,
we see that P n is the Proj of not only S but also of its Veronese subrings.
L
More generally, let R =
d=0 Rd be a graded ring that is finitely generated as
a C-algebra. Given a positive integer , the th Veronese subring of R is
R[] =
M
d=0
so
(R [] )d
Rd
Rd = R,
d=0
= Rd . Then Example 14.2.7 generalizes as follows (Exercise 14.2.4).
Lemma 14.2.8. Given R and a Veronese subring R [] as above, there is a natural

isomorphism Proj(R[] ) Proj(R).

Here is an example of how to use Lemma 14.2.8.
Example 14.2.9. The coordinate ring A of an affine variety gives the graded ring
A[x0 , . . . , xn ], where all variables have degree 1. In Example 7.0.10, we noted that
Proj(A[x0 , . . . , xn ]) = Spec(A) P n .
In particular, when n = 0, we have
Proj(A[x0 ]) = Spec(A) P 0 = Spec(A).
Now consider the complicated graded ring R from Example 14.1.14. The
Veronese subring
R[2] = C[xy, z 2 ] x 2 C[xy, z 2 ] x 4 C[xy, z 2 ] x 6 C[xy, z 2 ]
[2]
is a polynomial ring in one variable over C[xy, z 2 ], i.e., R = C[xy, z 2 ][x 2 ]. Thus
C3 // G = Proj(R ) Proj(R[2] ) = Spec(C[xy, z 2 ]) C2 .
We now understand this example from the Proj point of view.
The Normal Fan of a Polyhedron. Previously, we have given three constructions

of the normal fan:
From 2.3: The normal fan of a full dimensional lattice polytope P MR is a
complete fan in NR .
From 6.2: The normal fan of a lattice polytope P MR is a complete generalized fan in NR . This fan is degenerate when P is not full dimensional.
From 7.1: The normal fan of a full dimensional lattice polyhedron P MR is
a fan in NR whose support is the dual of the recession cone of P.
691
It can happen that none of these constructions apply to the polyhedra Pa MR

considered here. Hence we need a fourth construction of a normal fan. You may
want to review the discussion of generalized fans given in 6.2.
Let P MR be a polyhedron with rational vertices such that the recession cone
of P is strongly convex and rational. A vertex v P gives the cone
Cv = Cone(P MQ v) MR .
This is similar to our earlier constructions, except that we use P MQ instead of

P M since P may have few lattice points (in fact, it could have none). The dual
cone Cv NR is a rational polyhedral cone, and these cones give a generalized fan
as follows (Exercise 14.2.5).
Proposition 14.2.10. Let C MR be the recession cone of P. Then
P = { | Cv , v is a vertex of P}
is a generalized fan in NR called the normal fan of P. Furthermore, |P | = C ,

and P is a fan if and only if P MR is full dimensional.

The toric variety XP is then defined to be the toric variety of the generalized
fan XP , i.e., XP = XP . Here is an example we studied in 6.2.
Example 14.2.11. Let D be a basepoint free torus-invariant Cartier divisor on a
complete toric variety X . From D we get PD MR , which is a lattice polytope by
Theorem 6.1.7, though it may fail to be full dimensional. The normal fan PD is a
generalized fan that gives the toric variety XPD featured in Theorem 6.2.8.
We note that generalized fans appear naturally in this situation. The vertices of
PD give the Cartier data {m }(n) of D, where n = dim X . The m are distinct
when D is ample, but some may coincide in the basepoint free case. Combining
those s for which m has a common value, say v PD , gives the union
[
v =
(n)
m =v
By Proposition 6.2.5, this union is a maximal cone in the normal fan of PD in NR .

You should reread the discussion following the proof of Proposition 6.2.5.
We will soon see that XPD has a natural interpretation as a GIT quotient.
Similar to 7.1, a polyhedron P MR gives a cone C(P) MR R whose slice

at height > 0 (resp. = 0) is P (resp. the recession cone of P). The associated
semigroup algebra
SP = C[C(P) (M Z)]
is graded as usual using the last coordinate. We now generalize Theorem 7.1.13.
Proposition 14.2.12. For a polyhedron P MR as above, we have XP Proj(SP ).
Furthermore, XP is semiprojective.
692
Proof. The polyhedron P gives the (possibly degenerate) normal fan P . First note
that multiplying P by a positive integer has no effect on the normal fan P . Also,
[]
Lemma 14.2.8 and the obvious isomorphism SP SP imply that
[]
Proj(SP ) Proj(SP ) Proj(SP ).

Hence we may assume that P is a lattice polyhedron. Note also that translating P
by m M has no effect on P . Since SP+m SP , we may assume that 0 P.
The minimal cone 0 P is the largest subspace of NR contained in every

cone of the normal fan. As we saw in 6.2, P projects to a genuine fan P in
N R = (N/0 )R , and XP is the toric variety of this fan. Since 0 P, it follows that:
0 is the smallest subspace of MR containing P.
M = 0 M M is dual to the map N N.
P is a full dimensional lattice polytope in M R whose normal fan in N R is given

by P .
The straightforward proofs are left to the reader as Exercise 14.2.6. The last bullet
and Theorem 7.1.13 imply that
XP Proj(C[C(P) (M Z)]).
Then the first bullet implies
C(P) (M Z) = C(P) (M Z),
which gives XP Proj(SP ).
Finally, we show that XP is semiprojective. As above, we may assume that

P MR is a lattice polyhedron containing the origin. Then P is a full dimensional

lattice polyhedron in M R , so XP is semiprojective by Proposition 7.2.9.
GIT Quotients are Toric. We can now prove that Cr // G is a toric variety.
b pick a Zr such that = a . Then:
Theorem 14.2.13. For G (C )r and G,
(a) The graded ring R from (14.1.2) satisfies R C[C(Pa ) (M Z)].
(b) The GIT quotient Cr // G = Proj(R ) is the toric variety of Pa MR .
Proof. We first note that R is the semigroup algebra of the semigroup given by
the disjoint union
`
1
d
r
S =
d=0 ( ) N ,
where is from (14.2.2). To see why, note that x b C[x1 , . . . , xr ] is (G, d )invariant if and only if (b) = d . Then (14.1.3) gives an isomorphism
L
d
R
d=0 {F C[x1 , . . . , xr ] | Fis (G, )-invariant} = C[S ].
Lemma 14.2.2 implies that Pa MR has a strongly convex rational recession
cone, and the description of Pa given in (14.2.3) shows that the vertices are rational.
693
Hence Propositions 14.2.10 and 14.2.12 apply to Pa . In particular, we get the cone
C(Pa ) MR R, and (m, k) 7 (m) + k a induces a semigroup isomorphism
(14.2.7)
C(Pa ) (M Z) S
(Exercise 14.2.7). It follows that R C[C(Pa ) (M Z)]. Thus

Cr // G = Proj(R ) Proj(C[C(Pa ) (M Z)]) XPa ,
where the last isomorphism follows from Proposition 14.2.12.
In [204, Def. 10.8], Miller and Sturmfels define a toric variety to be a GIT
quotient of Cr by G. They use the notation Cr //a G, which for us means Cr // a G.
Versions of Theorem 14.2.13 appear in [83, Ch. 12], [204, Ch. 10] and [232].
In Theorem 14.2.13, we used the polyhedron Pa rather than the more intrinsic
model P . The reason goes back to Example 14.2.4, where we learned that lattice
b
points in P Rr relative to Zr need not come from lattice points in Pa . When G
r
has no torsion (i.e., when G is a torus), this problem goes away and C // G is the
toric variety of the polyhedron P Rr .
Here are some examples that illustrate Theorem 14.2.13.
Example 14.2.14. Let X be a projective toric varietyP

with quotient construction
X = (C (1) \ Z())/G. Then a divisor class [D] = [ a D ] Cl(X ) gives a
b Note also that = a for a = (a ) Z(1) . Furthermore,
character G.
Pa = PD MR ,
where PD is from (4.3.2). There are two cases where we know the GIT quotient:
When D is ample, is the normal fan of PD = Pa . Hence
Cr // G X
by Theorem 14.2.13. This also follows from Propositions 14.1.9 and 14.1.12.
When D is nef, Theorem 14.2.13 implies that
Cr // G XPD ,
where XPD is the toric variety from Example 14.2.11 and Theorem 6.2.8.
We will pursue this example in Chapter 15 when we study the secondary fan.
Example 14.2.15. The affine space Sym3 (C) of 3 3 symmetric matrices contains
the torus G (C )3 consisting of the rank 1 matrices

2
t1
t1 t1t2 t1 t3

t2 t1 t2 t3 = t1t2 t22 t2 t3 , t1 ,t2 ,t3 C .
t1t3 t2t3 t32
t3
694
We will compute Sym3 (C)// G for the character that sends the above matrix to
t12t22 t32 . Write Sym3 (C) = Spec(C[x11 , x12 , x13 , x22 , x23 , x33 ]) C6 and note that
x11 x12 x13

2
2
2
det x12 x22 x23 = x11 x22 x33 + 2x12 x23 x13 x11 x23
x22 x13
x33 x12
.
x13 x23 x33
The monomials appearing in the determinant form a basis of the (G, )-invariant
polynomials. This explains our choice . Label these monomials as
2
2
2
X0 = x12 x23 x13 , X1 = x11 x23
, X2 = x22 x13
, X3 = x33 x12
, X4 = x11 x22 x33
and note that X02 X4 = X1 X2 X3 . Using coordinates Y0 , . . . ,Y4 on P 4 , we claim that

(14.2.8)
Sym3 (C)// G V(Y02Y4 Y1Y2Y3 ) P 4 .
We will sketch the proof, leaving the details as Exercise 14.2.8. The map
b Z3 is given by ei j 7 ei + e j , where e11 , e12 , e13 , e22 , e23 , e33 is the basis
G
6
of Z . Also observe that corresponds to (2, 2, 2) Z3 , so that = a for a =
(0, 1, 1, 0, 1, 0) Z6 . In Example 14.2.18 below we will see that the i s are
Z6
1 = (1, 0, 0), 2 = (1, 1, 1), 3 = (1, 1, 1),
4 = (0, 1, 0), 5 = (1, 1, 1), 6 = (0, 0, 1).
This gives the polytope Pa in Figure 3. The only lattice points of Pa are its vertices
m0 , . . . , m4 , which map to (m0 ) + a, . . . , (m4 ) + a, the exponent vectors of the
(G, )-invariant monomials X0 , . . . , X4 . Furthermore, since Pa is a normal polytope,
the monomials Xi generate R and give a projective embedding into P 4 . From here,
(14.2.8) follows easily. We thank Igor Dolgachev for showing us this example.
m3
m4
m0
m2
m1
Figure 3. Pa and its lattice points m0 , . . . , m4
Since Pa P , Proposition 14.2.12 and Theorem 14.2.13 have the following

immediate corollary.
695
b then the GIT quotient Cr // G is a

Corollary 14.2.16. If G (C )r and G,
semiprojective toric variety of dimension
dim Cr // G = dim P .
Multigraded Polynomial Rings. This section began with an algebraic subgroup G

of (C )r . Taking character groups, we obtained the exact sequence
b 0
0 M Zr G
from (14.2.2). This gives a multigrading on the polynomial ring S = C[x1 , . . . , xr ]

b An example is given by the grading on the total
by defining deg(x a ) = (a) G.
coordinate ring of a toric variety without torus factors.
b gives the graded piece S of S, and one checks
In particular, a character G
r
that (C , L ) S . It follows that the graded ring R used in the definition of
Cr // G is isomorphic to the ring
L
(14.2.9)
R
d=0 S d .
We will see below that determines an irrelevant ideal B() S and leads to a
quotient construction remarkably similar to what we did in Chapter 5.
Conversely, suppose that we start with a multigrading on C[x1 , . . . , xr ] given by
a surjective homomorphism : Zr A for a finitely generated abelian group A.
Applying HomZ (, C ), we get an inclusion
G = HomZ (A, C ) (C )r = HomZ (Zr , C )
b A. Hence the multigraded approach, featured in [190] and [204], is

such that G
fully equivalent to what we are doing here.
b Zs . This implies M Zrs , so that the
Matrices. When G is a torus, we have G
exact sequence (14.2.2) can be written
(14.2.10)
0 Zrs Zr Zs 0,
where B and A are integer matrices of respective sizes r (r s) and s r such

that AB = 0. Note that the r rows of B give 1 , . . . , r N = Zrs .
Example 14.2.17. Let r N and consider the exact sequence
B
0 Z2 Z4 Z2 0,
where
1
r

0 1
, A = 1 r 1 0 .
B=
1 0
0 1 0 1
0 1
696
This is the sequence 0 M Z4 Cl(Hr ) 0 that computes the class group

of the Hirzebruch surface Hr . The rows 1 , . . . , 4 of B are the minimal generators
ui of the fan of Hr shown in Figure 3 of Example 4.1.8.
Here the group is G = {(t,t r u,t, u) | t, u C }. Note that the first row of A
gives the t-exponent of elements of G and the second row gives the u-exponent.
Also, the last two columns of A show that the divisor classes [D3 ], [D4 ] corresponding to u3 , u4 give a basis of the class group, and the first two columns show how to
express [D1 ], [D2 ] in terms of this basis.
Example 14.2.18. In Example 14.2.15, the exact sequence (14.2.2) can be written
B
0 Z3 Z6 Z3 0,
where
1 0 0
1 1 1
2 1 1 0 0 0
1 1 1
B=
0 1 0 , A = 0 1 0 2 1 0 .
0
0
1
0
1
2
1 1 1
0 0 1
The matrix A gives the map ei j 7 ei + e j from Example 14.2.15, and given A, it
is easy to find a matrix B that makes the sequence exact. We call B the Gale dual
of A. Note that the i s are the rows of B. See 14.2 for more on Gale duality.
As shown by Examples 14.2.17 and 14.2.18, the matrices B and A make it easy
to determine the vectors i and the group G.
Virtual Facets. The facets of a polyhedron contain a lot of information about its
geometry and combinatorics. For a GIT quotient Cr // G, we have the polyhedra
P Pa for = a . However, to capture the full story of what is going on, we
need not only their facets but also their virtual facets.
This is easiest to see in Pa , which for a = (a1 , . . . , ar ) Zr is defined by
Pa = {m MR | hm, i i ai , 1 i r}
as in (14.2.3). For 1 i r, we call the set
Fi,a = {m Pa | hm, i i = ai }
a virtual facet of Pa . The facets of Pa all occur among the Fi,a , while other virtual
facets may be quite differentsome may be faces of smaller dimension, some may
be empty, and some may be all of Pa . We can also have Fi,a = Fj,a for i 6= j.
Example 14.2.19. In Example 14.2.4, we have M = Z3 , a = (1, 0, 1), and 1 =
2 = e1 , 3 = 2e2 . Thus
Pa = {m MR | hm, e1 i 1, hm, e1 i 0, hm, 2e2 i 1}.
697
It follows that F1,a = while F2,a and F3,a are the facets of Pa shown in Figure 1
from Example 14.2.4.
Example 14.2.20. Let G = {(t,t r u,t, u) | t, u C } as in Example 14.2.17. For

a = (0, 0, 0, 1), the polytope Pa R2 is shown in Figure 4. Here we have three genF4,a
Pa
F3,a
F1,a
F2,a
Figure 4. The polytope Pa R2 for a = (0, 0, 0, 1) and its virtual facets
uine facets F1,a , F3,a , F4,a and one virtual facet F2,a which is a vertex. We will see in
Example 14.3.7 below that Pa is the polytope of the divisor D4 on the Hirzebruch
surface Hr .
Virtual facets can also be defined for P . Using the description of P given
in (14.2.6), one sees that Fi,a Pa corresponds to the subset of P where the ith
coordinate vanishes. We will denote this set by Fi, , so that
Fi, = P V(xi ).
Semistable Points. Virtual facets determine the semistable points of Cr as follows.
r
Proposition 14.2.21. Let p = (p
T1 , . . . , pr ) C and set I(p) = {i | 1 i r, pi = 0}.
r ss
Then p (C ) if and only if iI(p) Fi, 6= .
Proof. Let p be semistable, so that some (G, d )-invariant polynomial F does not
vanish at p. Writing F as a linear combination of (G, d )-invariant monomials
shows that some (G, d )-invariant monomial x b that does not vanish at p. Then
b = (b1 , . . . , br ) lies in dP . For I(p) as above,
for all i I(p) since
T we have bi = 0 T
x b does not vanish at p. This implies b iI(p) dFi, , hence iI(p) Fi, 6= .
The converse
T is also easy, except that we have to be careful about torsion.
Assume that iI(p) Fi, 6= . Then the intersection contains a rational point, so
T
that there is b ( iI(p) dFi, ) Zr for some integer d > 0. This implies two
things:
x b does not vanish at p.
b
R (b) = d 1, so that (b) and d differ by a torsion element of G.
It follows that for some integer > 0, x b is (G, d )-invariant. Then p is semistable
since x b does not vanish at p.
698
When we think about semistable points algebraically, we get the ideal
Q

T
(14.2.11) B() =
iI
/ xi | I {1, . . . , r},
iI Fi, 6= S = C[x1 , . . . , xr ],
called the irrelevant ideal of . The vanishing locus of B() is the exceptional set
Z() = V(B()) Cr . This gives a nice description of the set of semistable points,
which you will prove in Exercise 14.2.9.
r
r
r
Corollary 14.2.22. (Cr )ss
= C \ Z(), so that C // G (C \ Z())//G.
Exercises for 14.2.

14.2.1. Supply the details omitted in the proof of Lemma 14.2.1.
14.2.3. Prove the description of P given in (14.2.6).
14.2.4. The exercise will prove Lemma 14.2.8. As in the proof of Proposition 14.1.12,
Proj(R) is covered by the affine open subsets D+ ( f ) = Spec(R( f ) ), where f Rd is nonnilpotent and R( f ) is the homogeneous localization (14.1.4).
(a) Note that f Rd implies f Rd = (R [] )d . Prove that
(R[] )( f ) = R( f ) .
p
L
(b) Pick f1 , . . . , fs R homogeneous such that h f1 , . . . , fs i = R+ = d>0 Rd . Prove that
q

f1 , . . . , fs =
f 1 , . . . , f s = R+ .
q

f1 , . . . , fs = (R [] )+ as ideals of R [] .
as ideals of R, and conclude that
(c) Prove that Proj(R [] ) Proj(R). Hint: For f1 , . . . , fs R as in part (b), {D+ ( fi )}si=1 is
an open cover of Proj(R) by 7.0.

14.2.6. Prove the assertions made in the three bullets in the proof of Proposition 14.2.12.
14.2.7. In (14.2.7) we claimed that the mapping (m, k) 7 (m) + ka induces a semigroup
isomorphism C(Pa ) (M Z) S . Prove this.
14.2.8. This exercise will fill in some of the details omitted in Example 14.2.15.
(a) Show that Pa R3 = MR is defined by x 0, y 0, z 0, x + y 1 + z, x + z 1 + y
and y + z 1 + x. Also show that the vertices are the only lattice points of Pa .
(b) Prove that Pa is normal. Hint: If (a, b, c) Pa , study abc = 0 and abc > 0 separately.
(c) Prove (14.2.8)
14.2.9. Prove Corollary 14.2.22. Hint: Use Proposition 14.1.12.

14.2.10. Consider the group G = {(t,t 1 , u) | t C , u = 1} and character (t,t 1 , u) =
tu from Example 14.1.5. Note that = a for a = (1, 0, 1). By Example 14.1.8, the graded
pieces of R have a complicated behavior, and by Example 14.2.4, multiples of Pa have a
complicated behavior since Pa is not a lattice polyhedron.
(a) Use the proof of Theorem 14.2.13 to explain how these complications are linked.
14.3. Toric GIT and Gale Duality
699
[2]
(b) In Example 14.2.9, we used the Veronese subring R to show that Proj(R ) C2 .
Explain how this relates to the lattice polyhedron 2Pa .
14.2.11. Consider the Grassmannian G(1, 3) of lines in P 3 defined in Exercise 6.0.5. We
4
use the Plucker embedding G(1, 3) P 5 = P (2)1 defined by

0 1 2 3
7 (pi j ) = (p01 , p02 , p03 , p12 , p13 , p23 ),
A=
0 1 2 3
where pi j is the 2 2 minor of A formed by columns i and j. An elementary treatment

of this Plucker embedding appears in [69, Ch. 8, 6]. The torus G = (C )4 acts on A
by right multiplication. This gives an action on G(1, 3) where g = (t0 ,t1 ,t2 ,t3 ) G acts
on Plucker coordinates via g (pi j ) = (ti t j pi j ). This lifts to an action on the affine cone
4
b 3) C6 = C(2) . The action of G extends to an action on C6 . For the character
G(1,
= a , a = (1, . . . , 1) Z6 , prove that C6 // G P 2 . The more general GIT quotient
b 3)// G is described in [83, Ex. 12.4].
G(1,
This section begins our study of what happens to the GIT quotient Cr // G as we
vary . A key player is Gale duality.
Gale Duality. A toric variety X comes with two finite sets indexed by (1):
The minimal generators u N.
The divisor classes [D ] Cl(X ).
When X has no torus factors, these are related by the exact sequence
0 M Z(1) Cl(X ) 0,
where the first map is defined by m 7 (hm, u i)(1) Z(1) and the second is
e 7 [D ] Cl(X ).
Gale duality generalizes this situation in the context of real vector spaces. Let
W be a vector space over R spanned by vectors 1 , . . . , r . Some of the i may be
zero, and repetitions are allowedwe do not assume that the i are distinct. We
will use to denote this list of vectors. Then we have an exact sequence
(14.3.1)
0 V Rr W 0,
where (ei ) = i and : V Rr is the inclusion of the kernel of . Dualizing, we

obtain the exact sequence
(14.3.2)
0 W Rr V 0,
where we use dot product to identify Rr with its dual. This gives vectors i =
(ei ) V . We will use to denote this list of vectors. As in 14.2, the map in
(14.3.1) is given by (v) = (hv, 1 i, . . . , hv, r i) Rr for v V .
Since (14.3.1) is the dual of (14.3.2), we could equally well start with the
vectors in V and then recover the vectors by duality. This implies in particular
700
that in (14.3.2) is given by (w) = (h1 , wi, . . . , hr , wi) Rr for w W . The

symmetry between and will be used often in our discussion of Gale duality.
Our first result describes what happens when a subset of or is a basis.
Lemma 14.3.1. Let I {1, . . . , r} and set J = {1, . . . , r} \ I. Then i , i I, form a
basis of W if and only if j , j J, form a basis of V .
Proof. Note that |I| = dim W implies |J| = dim V = dim V . Assume j , j J,
give P
a basis of V and let v j , j J, be the dual basis of V . Then we have (v j ) =
e j + iI hv j , i iei for j J. Applying , we see that j J implies that
P
j = iI hv j , i ii .
This proves that i , i I, span W , and since |I| = dim W , they form a basis of W .
The other direction of the proof follows immediately by duality.
Using and we get cones C = Cone() W and C = Cone() V .

Here is an example of how these cones are related.
Lemma 14.3.2. Consider the cones C W and C V defined above.
(a) If C = W , then C is strongly convex in V . Conversely, if consists of

nonzero vectors and C is strongly convex in V , then C = W .
(b) If C = V , then C is strongly convex in W . Conversely, if consists of

nonzero vectors and C is strongly convex in W , then C = V .
Proof. Assume C = W . Pick v C (C ) and write
v=
r
X
i=1
ai i =
r
X
bi i
i=1
Pr
with ai , bi 0. Then
i=1 (ai + bi )i = 0. Since (14.3.2) is exact, there is w W
P
such that (w) = ri=1 (ai + bi )ei , so that hi , wi = ai + bi by the description of
given in the discussion following (14.3.2). Since ai + bi 0 for all i, we conclude
that w Cone() = C = {0}, where the last equality follows from C = W .
Then ai + bi = 0 for all i, so that ai = bi = 0 for all i, hence v = 0.
For the converse, take w C W . Then hi , wi 0 for all i. Since (w) =
Pr
i=1 hi , wiei , we have
i=1 hi , wii = 0. Strong convexity implies that {0} C
is a face, and then hi , wii = 0 for all i by Lemma 1.2.7. Since the i are nonzero,
we must have hi , wi = 0 for all i, which implies w = 0 since the i span W . This
proves C = {0}. Taking duals gives C = W , completing the proof of part (a).
Pr
Part (b) of the lemma follow from part (a) by duality.
Gale duality also describes the faces of these cones. Here is the result for C ;
the corresponding statment for C follows immediately by duality.
701
Lemma 14.3.3. Let I {1, . . . , r} and set J = {1, . . . , r} \ I. Then the following
are equivalent:
(a) There is a face F C such that i F if and only if i I.
P
(b) There are positive numbers bi , i J, such that iJ bi i = 0.
Proof. Given (a), there is w C such that hi , wi = 0 for i I and hi , wi > 0

P
for i J. Then (b) holds since 0 = (w) = iJ hi , wii . The proof of the
converse is equally easy (Exercise 14.3.1).

The introduction to Gale duality given here follows [222]. There is much more
to say about this subjectthe full story involves ideas such as circuits and oriented
matroids. See [281, Lec. 6] for a more complete treatment of Gale duality. Circuits
will play an important role in Chapter 15.
Characters as Vectors. We return to the GIT quotient Cr // G, where G (C )r
b We will continue to regard G
b as a multiplicative group, but since the
and G.
b
b
tensor product GR = G Z R is a vector space over R of dimension dim G, we will
bR . Thus characters , G
b give
switch to additive notation when working in G
bR such that () 1 = 1 + 1.
vectors 1, 1 G
b defined by i (g) = ti for
The inclusion G (C )r gives characters i G
b
g = (t1 , . . . ,tr ) G. In terms of (14.2.2), this means i = (ei ). These generate G,
b
so that GR is spanned by the vectors
bR ,
i = i 1 = R (ei ) G
1 i r.
Q
Note that a = (a1 , . . . , ar ) Zr gives the character ri=1 ai i , which was written
(a) = a in the discussion following (14.2.2). It follows that
P
bR .
R (a) = a 1 = ri=1 ai i G
bR give a list we denote by , and the vectors

The Two Cones. The vectors i G
i = (ei ) N NR = MR give a list we denote by . Then the exact sequence

(14.3.3)
from (14.2.4) satisfies
R
R
bR 0
0 MR
Rr
G
R (ei ) = i for 1 i r.
R (m) = (hm, 1 i, . . . , hm, r i) for m MR .
Hence Gale duality applies to this situation. We have the additional structure given
b1 G
bR . Furthermore, the cones
by the lattices M MR and G
bR
C = Cone() = Cone(1 , . . . , r ) G
C = Cone() = Cone(1 , . . . , r ) NR
702
bR and NR and are rational polyhedral with respect to the

are full dimensional in G
b 1 and N. These cones will play an important role in this section.
lattices G
Example 14.3.4. Let XP be the toric variety of a full dimensional lattice polyhedron P MR . Then (14.3.3) is obtained from the standard exact sequence
0 M ZP (1) Cl(XP ) 0
by tensoring with R. Thus:
The i s come from the divisor classes [D ], so that the cone C is the cone of
torus-invariant effective R-divisor classes.
The i s are the minimal generators u , so that the cone C is the support of
the normal fan P .
Existence of Semistable and Stable Points. We first show the cone C controls
when the GIT quotient is nonempty, i.e., when semistable points exist.
b is a character, then the following are equivalent:
Proposition 14.3.5. If G
(a) Cr // G 6= .
(b) (Cr )ss

6= .
(c) 1 C .
Proof. We have (a) (b) by Proposition 14.1.12. Using dim Cr // G = dim P
(Corollary 14.2.16) and P = R1 ( 1) Rr0 , (a) (c) follows from
Cr // G 6= P 6= there is b = (b1 , . . . , br ) R1 ( 1) Rr0
1 =
r
X
i=1
bi i , bi 0 1 C .
Our next task is to determine when stable points exist. Here, the interior of C
is the key player.
Proposition 14.3.6. If 1 C , then the following are equivalent:
(a) (C )r (Cr )s .
(b) (Cr )s 6= .
(c) 1 is contained in the interior of C .
(d) Fi, is a proper face of P for all i.
Furthermore, the above conditions imply that

(e) dim Cr // G = r dim G,
and when consists of nonzero vectors, (e) is equivalent to (a)(d).

Proof. Throughout the proof we pick a = (a1 , . . . , ar ) Zr such that = a .
703
The implication (a) (b) is obvious. Assume (b) and suppose that 1 F,
where F is a proper face of C . Let I = {i | i F} and J = {1, . . . , r} \ I. Note
that J is nonempty since F is a proper face. If b = (b1 , . . . , br ) P , then
P
1 = ri=1 bi i .
Then Lemma 1.2.7 and 1 F imply that bi i F for all i. Since i
/ F for i J,
we must have bi = 0 for i J. Thus Fi, = P for i J. Relabeling if necessary, we
may assume 1 J. Then F1, = P and (14.2.11) imply that the minimal generators
of the ideal B() do not involve x1 . By Corollary 14.2.22, we have
(Cr )ss
= C U ,
(14.3.4)
U Cr1 open.
Translated to Pa , F1, = P means that

Pa {m MR | hm, 1 i = a1 }.
Since Pa has full dimension in MR by Proposition 14.1.12 and Theorem 14.2.13,
this inclusion forces 1 = 0. Using the description of G from Lemma 14.2.1, we
get a product decomposition
G = C G C (C )r1 = (C )r .
Thus (t) = (t, 1) C G = G is a one-parameter subgroup of G. Via (14.3.4),

pick p = (p1 , p ) C U = (Cr )ss
. Then (t) p G p, and the limit
lim (t) p = lim (t p1 , p ) = (0, p ) C U
t0
t0
exists as point of (Cr )ss

. But this limit is not in G p when p1 6= 0, so that G p
r ss
is not closed in (C ) when p1 6= 0. Thus p cannot be stable by Remark 14.1.13.
This is easily seen to contradict (Cr )s 6= , and (b) (c) follows.
Assume (c) and let J = {i | Fi, = P }. Suppose J 6= and set I S

= {1, . . . , r} \ J.
Since Fi, ( P for each i I, we can pick c = (c1 , . . . , cr ) P \ ( iI Fi, ). Then
the ith coordinate of c is positive for i I and is zero for i J. Hence
P
(14.3.5)
1 = iI ci i
Suppose that Cone(i | i J) NR is strongly convex. Then {0} is a face,

which implies that there is m MR such that hm, i i > 0 for all i J. This gives
c = (hm, 1 i, . . . , hm, r i) R1 (0) whose ith coordinate is positive for i J. Then
b = c + c R1 (0)
for all , and when > 0 is sufficiently small, the ith coordinate of b is positive for
all i. This follows easily by treating the cases i I and i J separately. Thus b P
has positive coordinates, which contradicts J = {i | Fi, = P } 6= . We conclude
that Cone(i | i J) is not strongly convex. This easily gives a nonempty subset
J J such that
P
iJ i i = 0, i > 0
704
(Exercise 14.3.2). By Gale duality (Lemma 14.3.3), it follows that there is a face
F C such that {i | i F} = {1, . . . , r}\J . This is a proper face since J 6= , and
it contains 1 by (14.3.5) and I = {1, . . . , r} \ J. This contradicts 1 Int(C )
and completes the proof of (c) (d).
Assume (d). Since Fi, 6= P for all i, P has a rational point with positive
coordinates. Arguing as in the proof of Proposition 14.2.21, there is d > 0 and
b dP Zr>0 such that x b is (G, d )-invariant. Since all exponents are positive,
the nonvanishing set of x b is (C )r . It follows easily that points of (C )r are stable.
This proves (d) (a).
The implication (b) (e) follows from Proposition 14.1.12. Now assume (e),
so that Pa is full dimensional in MR . This implies that if i 6= 0, then m 7 hm, i i
cannot be constant on Pa , and hence Fi,a 6= Pa . Thus (e) (d) when consists of
nonzero vectors, and the proof is complete.

Proposition 14.3.6 has many nice consequences concerning Cr // G when
consists of nonzero vectors, including:
Stable points exist if and only if the GIT quotient has the expected dimension.
The GIT quotient is nonempty of smaller than expected dimension if and only
if the character is on the boundary of C .
Here is an example to illustrate the last bullet.
Example 14.3.7. Let G = {(t,t r u,t, u) | t, u C } (C )4 . This group appeared
bR R2 is the shaded region in
in Example 14.2.17. For r = 1, the cone C G
Figure 5. The figure also shows 1 , 2 , 3 , 4 . Three of the P s in the figure include
dashed lines that represent virtual facets, two of which are empty and the other of
which is a vertex. Note that dim P = dim C4 // G drops dimension precisely on
the boundary of C , as predicted by Proposition 14.3.6.
P =
P =
P =
P =
2
4
P =
1 = 3
Figure 5. C and the polytopes P for all 1 C
705
Figure 5 is closely related to Figure 12 in Example 6.3.23. In that example, our

focus was on the Hirzebruch surface Hr , so there we saw only the first quadrant
portion of Figure 5. Looking at all of Figure 5, we see that away from the boundary,
Cr // G is either P 2 or H1 , depending on which quadrant we are in.
In many of the most interesting cases, consists of nonzero vectors. We should

nevertheless give an example where one of the i s vanishes.
Example 14.3.8. Consider the group G = {(t, u, u) | t, u C } and the character
(t, u, u) = u. We leave it as Exercise 14.3.3 to show that:
1 = 0, 2 = e1 , 3 = e1 , where e1 is a basis of NR Z.
C = R20 and 1 = (0, 1) C .
P = Conv((0, 1, 0), (0, 0, 1)) and F1, = P .
(C3 )ss = C (C2 \ {0}) and (C3 )s = .
Note that parts (a), (b), (c), (d) of Proposition 14.3.6 are false while part (e) is true
since C3 // G P1 has the expected dimension 3 2 = 1.
Projective and Birational GIT Quotients. By Proposition 14.1.12, the quotients

we are studying come with a projective morphism
Cr // G Cr //G = Spec(C[x1 , . . . , xr ]G ).
(14.3.6)
Here we explore the extreme cases where Cr //G is very small or very large. We
will need the following result about C = Cone() NR .
Lemma 14.3.9.
(a) C[x1 , . . . , xr ]G = C[C M], so that Cr //G = Spec(C[C M]).
(b) If 1 C , then C MR is the recession cone of Pa when = a .

Proof. For part (a), note that x b is G-invariant if and only if (b) is trivial, which
is equivalent to b = (m) for some m M. Since b Nr , the formula for (m)
shows that C[x1 , . . . , xr ]G is the semigroup algebra of lattice points in the cone
(14.3.7)
{m MR | hm, i i 0, 1 i r} = Cone(1 , . . . , r ) = C .
For part (b), let 1 C . Then Pa is nonempty by the proof of Proposition 14.3.5
and hence has recession cone (14.3.7) by the proof of Lemma 14.2.2.

Our first extreme case is when Cr //G = Spec(C[x1 , . . . , xr ]G ) is very small.
Proposition 14.3.10. The following are equivalent:
(a) Cr // G is projective for all 1 C .
(b) Cr // G is projective for some 1 C .
(c) C[x1 , . . . , xr ]G = C.
(d) consists of nonzero vectors and C is strongly convex.
706
Proof. First, (a) (b) is clear, and (b) (c) follows from Exercise 14.1.9 since
1 C implies Cr // G 6= . We also have (c) (a) by Exercise 14.1.7.
If (d) is true, then C = NR by Gale duality (Lemma 14.3.2), and then we have
= {0}. Assume 1 C . Then C = {0} is the recession cone of Pa by
Lemma 14.3.9, so that Pa is a polytope. It follows that Cr // G XPa is projective.
b
Finally, suppose that (c) is true. If i = 0, then i = i 1 implies that i G
has finite order, say . Since i : G C is projection onto the ith coordinate,
it follows that C[xi ] C[x1 , . . . , xr ]G , a contradiction. Hence the i are nonzero.
Also, (c) and Lemma 14.3.9 imply that C = {0}, so that C = NR . Hence C is
strongly convex by Gale duality (Lemma 14.3.2).
The other extreme is when Cr //G is large. Since dim Cr //G r dim G by
Remark 14.1.13, Cr //G is as large as possible when dim Cr //G = r dim G. Here
is our result.
Proposition 14.3.11. The following conditions are equivalent:
(a) dim Cr //G = r dim G.
(b) C NR is strongly convex.
(c) The map Cr // G Cr //G from (14.3.6) is birational for all 1 C .
Furthermore, when consists of nonzero vectors, we can replace (b) and (c) with:
bR .
(b ) C = G
(c ) Cr // G Cr //G is birational for all .
Proof. Lemma 14.3.9 implies dim Cr //G = dim C . Since r dim G = dim MR ,
we obtain
dim Cr //G = r dim G dim C = dim MR
C NR is strongly convex.
This proves (a) (b). Now assume (b). Then C is strongly convex, so that
Cr //G is the affine toric variety of C by Lemma 14.3.9. Furthermore, Pa is full
dimensional since its recession cone is C by Lemma 14.3.9. Hence the normal
fan of Pa refines C by Theorem 7.1.6. Then (c) follows since Cr // G Cr //G is
the birational map induced by this refinement. On the other hand, if (c) holds, then
Cr // G Cr //G is birational for 1 Int(C ), so that
dim Cr //G = dim Cr // G = r dim G,
where the last equality uses Proposition 14.3.6. Hence (c) (a).
Assume that consists of nonzero vectors. By Gale duality (Lemma 14.3.2),

bR , which gives (b) (b ). Then
C NR is strongly convex if and only if C = G
(c) (c ) follows immediately.
707
Example 14.3.12. For G = {(t,t,t 1 ) | t C }, one gets 1 = 2 = e1 , 3 = e1

bR R and 1 = e1 , 2 = e2 , 3 = e1 + e2 in NR R2 . Thus C = Cone(e1 , e2 )
in G
is strongly convex and i 6= 0 for all i. Hence all conditions of Proposition 14.3.11
b we have a birational morphism
apply, i.e., for all G,
C3 // G C3 //G = Spec(C[xz, yz]) = C2 .
To understand Cr // G, consider Figure 6, which shows the polyhedron P

for all . For positive , Cr // G is the blowup of C2 at the origin, while for
P =
P =
P =
1 = 2
Figure 6. The polytopes P for all
negative , the GIT quotient is C2 . The dashed lines in the figure indicate the
presence of a virtual facet, which is empty for negative and consists of the vertex
when is the origin. In Exercise 14.3.4 you will compare this to the quotient
construction of Bl0 (C2 ).
Generic Characters. We next study what it means for a character to be generic.

b is generic if 1 C and for every subset
Definition 14.3.13. A character G
/ Cone( ).
with dim Cone( ) < dim C = dim G, we have 1
We can determine when a character is generic as follows.
Theorem 14.3.14. For 1 C , the following are equivalent:

(a) is generic.
(b) Every vertex of P has precisely dim G nonzero coordinates.

(c) P is simple of dimension r dim G, every virtual facet Fi, P is either
empty or a genuine facet, and Fi, 6= Fj, if i 6= j and Fi, , Fj, are nonempty.
(d) (Cr )s = (Cr )ss
.
Proof. We will prove that (a), (c) and (d) are each equivalent to (b).
For (a) (b), assume 1 C is generic. Then it does not lie on the
boundary of C , which implies dim Cr // G = r dim G by Proposition 14.3.6. In
other words, dim P = r dim G.
Now take any vertex b P . Then dim P = r dim G implies that at least
this many facets meet at b. Since all facets of P occur among the virtual facets
Fi, = P V(xi ), it follows that at least r dim G coordinates of b vanish. Thus
b has at most dim G nonzero coordinates. Suppose that b has s < dim G nonzero
708
P
P
coordinates. Writing b = sj=1 j ei j , u j > 0, we obtain 1 = sj=1 j i j , which
is impossible since is generic and s < dim G. This proves (a) (b).
To prove
Ps (b) (a), assume that is not generic. By Carathedorys theorem,
1 = j=1 j i j , where u j > 0 and s < dim G (Exercise 14.3.5). Then b =
Ps
j=1 j ei j P has fewer than dim G nonzero coordinates. In Exercise 14.3.5 you
will prove that this gives a vertex of P with the same property, contradicting (b).
To prove (b) (c), assume (b) and take a vertex b P . Arguing as above, we
see that dim P = r dim G and that the actual facets of P make at least r dim G
coordinates of b vanish. But exactly r dim G vanish by (b), so that the only
virtual facets containing b correspond to actual facets. Then (c) follows easily.
The proof of (c) (b) is similar and is left to the reader (Exercise 14.3.5).
Before proving (b) (d),Twe first describe the semistable points. Given a
subset I {1, . . . , r}, note that iI Fi, 6= if and only if there is a vertex b P
with b Fi, for all i I. Hence the maximal such subsets are those given by
{i | b Fi, }, b P a vertex. It follows that the ideal B() defined in (14.2.11) is
given by
Q

(14.3.8)
B() =
bF
/ i, xi | b is a vertex of P .
By Corollary 14.2.22, it follows that
(Cr )ss
=
(14.3.9)
Crb .
b a vertex
where
Crb
= {(p1 , . . . , pr ) C | pi 6= 0 when b
/ Fi, }.
r
We now turn to (b) (d). The first step is to show that G-orbits are closed in
r
(Cr )ss
. By (14.3.9), we need only show that G-orbits of points in Cb are closed in
Crb for every vertex of b. By hypothesis, b has precisely r dim G coordinates that
vanish, i.e., precisely this many virtual facets Fi, contain b. Let I = {i | b Fi, }.
By the earlier part of the proof, we know that P is simple of dimension r dim G.
Hence the Fi, , i I, are the genuine facets containing b, so the facet normals i ,
i I, form a basis of MR .
Since the connected component of the identity G G has finite index, it suffices to show that G p is closed in Crb for all p Crb . Take p G p Crb .
Using Lemma 5.1.10 and arguing as in the proof of Theorem 5.1.11, we can find a
one-parameter subgroup : C G and a point g G such that
p = lim (t)g p.
(14.3.10)
Since
(14.3.11)
t0
(C )r ,
we can write (t) = (t a1 , . . . ,t ar ) for exponents ai Z, and

Pr
i=1 ai i = 0,
follows from (14.2.2) since is a one-parameter subgroup of G (Exercise 14.3.5).

The coordinates of p, p, g are nonzero for i
/P
I, hence the limit (14.3.10) implies
that ai = 0 for i
/ I. Thus (14.3.11) becomes iI ai i = 0, so that ai = 0 for all
709
i since the i , i I, are linearly independent. Hence is the trivial one-parameter

subgroup. which implies p = g p G p by (14.3.10). We conclude that G p
is closed in Crb .
In Exercise 14.3.5 you will show that the isotropy subgroup G p is finite for
p (Cr )ss
. This and the previous paragraphs imply that every semistable point is
stable, which completes the proof of (b) (d).
Finally, for (d) (b), first note that (Cr )s 6= , so that dim P = dim Cr // G =
r dim G = dim MR by Proposition 14.1.12. As noted earlier, this implies that
every vertex of P has at least r dim G coordinates that vanish. Now assume (b)
fails, so that some vertex b P has I = {i | b Fi, } with |I| > r dim G. Thus
the i MP
R , i I, are linearly dependent since dim MR = r dim G. Then there is
a relation iI ai i = 0 where ai Z and ai > 0 for at least one i. If we set ai = 0
for i
/ I, then
(t) = (t a1 , . . . ,t ar ) (C )r
is a one-parameter subgroup of G by Exercise 14.3.5. Now define the point p =

(p1 , . . . , pr ) Crb by
(
1 ai 0
pi =
0 ai < 0.
Consider limt0 (t) p. The limit exists in Cr since pi = 0 for ai < 0, and if i
/ I,
then the ith coordinate of (t) p is 1 for all t, so that the limit p = limt0 (t) p
lies in Crb . Since there is i0 I with ai0 > 0, the i0 th coordinate of p = limt0 (t) p
is zero. However, the i0 th coordinate of p is nonzero, so the same is true for G p.
Hence G p is not closed in Crb , and it follows that p is a point that is semistable
but not stable. This contradicts (d) and completes the proof of the theorem.

The equivalence (a) (b) in Theorem 14.3.14 appears in [137], while (c) is
taken from [275]. Our treatment of (d) was inspired by [83, Prop. 12.1].
Example 14.3.15. In Example 14.3.12, every nontrivial character is generic. When
is trivial, Figure 6 shows that part (c) of Theorem 14.3.14 fails because there is a
nonempty virtual facet that is not a facet.
Example 14.3.16. Figure 5 of Example 14.3.7 makes it clear that the non-generic
elements of C lie on the three rays generated by the i s. Hence the generic
elements form two chambers, a term that will be defined in 14.4. In the chamber
on the right in Figure 5, P is a quadrilateral, while in the chamber on the left, P
is a triangle with an empty virtual facet indicated by the dashed line. When
lies on the vertical axis, P is also a triangle but is non-generic because of the
nonempty virtual facet where the dashed line touches the triangle. This illustrates
the importance of the virtual facets, as described in part (c) of Theorem 14.3.14.
We will see in 14.4 that Figure 5 is an example of a secondary fan.
710
Example 14.3.17. The polytope in Figure 3 of Example 14.2.15 is not simple, so

that the character in that example is not generic. Since was chosen because
it gave the monomials appearing in the determinant, this shows that geometrically
interesting characters may fail to be generic.
Let us give one more detailed example of the theorem.

Example 14.3.18. Consider G = {(t,t,t 1 ) | t C } as in Example 14.3.12. If we
choose the trivial character = 1, then the GIT quotient is
C3 // G = C3 //G = Spec(C[x, y, z]G ) = Spec(C[xz, yz]) = C2 .
Every part of Theorem 14.3.14 fails in this example. It is instructive to see how
this happens:
(a) The trivial character is never generic, as follows from the definition.
(b) P is the middle polyhedron in Figure 6 of Example 14.3.12. All three virtual
facets meet at the vertex of P , which means that too many coordinates vanish.
(c) As shown by Figure 6, P has a nonempty virtual facet that is not a genuine
facet. Yet P is simple and has the correct dimension.
3
(d) = 1 easily implies that (C3 )ss

= C . Since G (1, 0, 0) = C {(0, 0)} is not
3 ss
3 s
closed in (C ) , it follows that (1, 0, 0)
/ (C ) .
For the quotient X (C(1) \ Z())//G considered in Proposition 14.1.9, all

semistable points were stable when the toric variety is simplicial. For general GIT
quotients such as the one considered here, things can be more complicated.
See also Exercise 14.3.6 for an example where the same genuine facet occurs
for two distinct indices.
The following immediate corollary of Theorem 14.3.14 summarizes the nice
properties of GIT quotients when the character is generic.
b is generic, then Cr // G is a simplicial semiprojective
Corollary 14.3.19. If G

toric variety of the expected dimension r dim G with (Cr )s = (Cr )ss
.
Generic characters are clearly very nice. In the next section we will see that
generic characters determine a refinement of C , the so-called secondary fan, that
tells the full story of Cr // G as we vary .
Exercises for 14.3.
14.3.2. Show that a cone Cone(A ) is not strongly
Pconvex if and only there is a nonempty
subset A A and v > 0 for v A such that vA v v = 0.
711
14.3.4. Explain how Example 14.3.12 relates to the quotient construction of Bl0 (C2 ) from
Example 5.1.16. See also Example 7.1.4.
14.3.5. This exercise is devoted to the proof of Theorem 14.3.14.
(a) In the proof of (b) (a), we claimed that Carathedorys theorem
the proof of
P(see
s
Theorem 2.2.12) implies that if is not generic, then 1 = j=1 j i j , where
u j > 0 and s < dim G. Prove this.
(b) Suppose that a point b P has fewer than dim G nonzero coordinates. Prove that
there is a vertex of P with the same property. Hint: Write b as a convex combination
of vertices plus an element of the recession cone.
ar
a1
(c) Given (a1 , . . . , ar ) Zr , prove
Pr that (t) = (t , . . . ,t ) defines a one-parameter subgroup of G if and only if i=1 ai i = 0.
(d) Complete the proof of (b) (d) by showing that the isotropy subgroup G p is finite
for every p Crb . Hint: Let I = {i | b Fi, }. Then show two things: first, that
g = (t1 , . . . ,tr ) G p has gi = 1 for i
/ I and second, that the i , i I, form a basis of
NR . Now use Exercise 14.2.1 to show that G has finite order. You may also want to
look at Exercise 5.1.11.
14.3.6. Consider G = {(t,t 1 ) | t C } with the trivial character = 1.
(a) Show that 1 = 2 = e1 and 1 = e1 , 2 = e1 .
(b) Show that P = Cone(e1 ) MR R with F1, = F2, = {0}.
(c) Show that C2 // G = C.
2
2 s
2
(d) Show that (C2 )ss
= C and (C ) = (C ) .
This is our first example where a character fails to be generic because the same genuine
facet occurs for two separate indices.
14.3.7. In the situation of Gale duality, let I {1, . . . , r} and J = {1, . . . , r} \ I. Prove that
the i , i I, are linearly independent in W if and only if the j , j J, span V . Hint:
Enlarge I to get a basis of W .
14.3.8. Suppose that W R with basis e1 . Then pick vectors 1 = e1 and 2 = 0 and
define V as in (14.3.1).
(a) Show that V has a basis e1 such that 1 = 0 and 2 = e1 .
(b) Use this example to explain why the nonzero vector hypothesis is needed in parts (b)
and (d) of Lemma 14.3.2.
(c) Adapt this situation to Z and give an example where part (e) of Proposition 14.3.6 is
true but the other parts of the proposition are false.
14.3.9. An augmented polyhedron consists of a polyhedron P and a list of faces Fi of P for
1 i r. We allow Fi = or Fi = P, and we can also have Fi = Fj for i 6= j. However,
we require that all actual facets of P occur among the Fi . An example of an augmented
polyhedron is given by P with its virtual facets Fi, , 1 i r.
(a) Define what it means for an augmented polyhedron to be simple in terms of the number
of virtual facets containing a vertex.
(b) Restate part (c) of Theorem 14.3.14 using augmented polyhedra. It will be elegant.
(c) Use one of the examples given in the text to show that there is a non-simple augmented
polyhedron whose underlying polyhedron is simple.
712
14.4. The Secondary Fan

b varies. The cones
For G (C )r , we study the GIT quotient Cr // G as G
bR
C = Cone() = Cone(1 , . . . , r ) G
C = Cone() = Cone(1 , . . . , r ) NR
bR and NR
introduced in 14.3 will play a key role in our discussion. Recall that G
b
have the lattices G 1 and N.
A first observation is that there are only finitely many distinct GIT quotients
up to isomorphism. To see why, recall from Corollary 14.2.22 that

Cr // G Cr \ Z() //G,
where Z() is the vanishing locus of the irrelevant ideal (14.2.11) defined by
Q

T
(14.4.1)
B() =
iI
/ xi | I {1, . . . , r},
iI Fi, 6= C[x1 , . . . , xr ].
There are only finitely many such ideals, hence only finitely many GIT quotients.
This decomposes the lattice points of C into finitely many disjoint subsets where
Cr // G is constant on each subset. The basic idea is that each subset lies in the
relative interior of a cone of the secondary fan, whose support is C .
A Shift in Focus. Giving a rigorous construction of the secondary fan will require
bR . Given
some new ideas, and we will also need to work in Rr rather than in G
r
a Z , the polyhedron Pa MR gives two objects of interest:
The generalized fan , which is the normal fan of Pa in NR .
The set I = {i | Fi,a = }, which determines the empty virtual facets of Pa .
We can characterize the pairs (, I ) that occur as follows.
Proposition 14.4.1. Given a generalized fan in NR and subset I {1, . . . , r},

then and I come from some a Zr if and only if
(a) || = C .
(b) X is semiprojective.
(c) = Cone(i | i , i
/ I ) for every .
Proof. First suppose that and I come from a Zr . Since is the normal fan
of Pa , Lemma 14.2.2 and Proposition 14.2.10 imply || = C . Furthermore, X is
the toric variety of Pa and hence is semiprojective by Proposition 14.2.12. Thus
satisfies conditions (a) and (b).
For (c), it suffices to consider maximal cones, which correspond to vertices
v Pa . In Exercise 14.4.1 you will show that v gives the maximal cone
Cone(i | hv, i i = ai ) = Cone(i | v Fi,a ).
Then (c) follows since v Fi,a implies i
/ I ,
713
Conversely, suppose that and I satisfy (a), (b) and (c). Let 0 be the minimal
cone of and set N = N/0 . Then gives a genuine fan in N R . Since X =
X is semiprojective, Theorem 7.2.4 and Proposition 7.2.9 imply that X has a
torus-invariant Cartier divisor whose support function is strictly convex on ||.
Composing this with the map NR N R gives a strictly convex support function
on || that takes integer values on N. Then define a = (a1 , . . . , ar ) Zr by
(
(i )
i
/ I
ai =
(i ) 1 i I .
In Exercise 14.4.1 you will show that and I come from a.
GKZ Cones. Following Gel fand, Kapranov and Zelevinsky [113, Ch. 7] (see also
[112]), we construct the secondary fan using cones defined in terms of support
functions. Our treatment is also influenced by [141] and [222].
Support functions with respect to a fan were defined in Definition 4.2.11. This
definition extends to generalized fans without change. Let SF() denote the set of
all support functions with respect the generalized fan . Since has full dimensional convex support by Proposition 14.4.1, we can define
CSF() = { SF() | is convex}.
(14.4.2)
The convexity criteria from 6.1 and 7.2 apply to our situation.
Definition 14.4.2. Let and I {1, . . . , r} be as in Proposition 14.4.1. Then the
GKZ cone of and I is the set

e ,I = (a1 , . . . , ar ) Rr | there is CSF() such that

(i ) = ai for i
/ I and (i ) ai for i I .
e ,I , the support
e,I is a cone. Note that when a
We will see below that
function is unique. This follows from Proposition 14.4.1 since is linear on the
cones of and satisfies (i ) = ai for i
/ I . Hence we write as a .
Here are some easy properties of GKZ cones.
Proposition 14.4.3. Let and I be as above. Then:

e ,I is a rational polyhedral cone in Rr .
(a)
e ,I is ker(R ) = im(R ) MR .
(b) The minimal face of
e ,I that will be useful later in the

Proof. We begin with a different model ofL
r
section. Consider the direct sum W = R max MR . We write points of W as
(a, {m }), where a = (a1 , . . . , ar ) and {m } = {m }max . Then let C,I W be
the subset consisting of all points (a, {m }) satisfying
(14.4.3)
hm , i i = ai , for i and i
/ I
hm , i i ai , for i
/ or i I .
714
Let : W Rr be the projection map. We claim that induces a bijection

e,I .
C,I
(14.4.4)
To prove this, take (a, {m }) C,I . Then (u) = minmax hm , ui is convex

by Lemma 6.1.5, and by (14.4.3), we have (i ) = ai , i
/ I , and (i ) ai ,
i I . If max , then (14.4.3) also implies that hm , i i = ai for i , i
/ I .
Since these i s generate by Proposition 14.4.1, we have CSF() and thus
e ,I . Conversely, let a
e ,I . For each max there is m MR such that
a
a (u) = hm , ui for u . By convexity and Lemma 6.1.5, hm , ui a (u) for all

u ||. From here, one sees easily that (a, {m }) C,I .
For part (a), observe that C,I is a rational polyhedral cone in W relative to the
L
e ,I is a
lattice Zr max M since i N for all i. Then (14.4.4) implies that
r
rational polyhedral cone in R .
For part (b), note that for m MR , the point R (m) = (hm, 1 i, . . . , hm, r i) lies
e ,I because of the convex support function defined by (u) = hm, ui. Now
in
e ,I (
e ,I ). Then a and a = a are convex, so that a is linear,
take a
say a (u) = hm, ui for u C . Since a (i ) ai and a (i ) ai for all i,

we obtain hm, i i = ai for all i, i.e., R (m) = a.

We will soon see that the GKZ cones form a generalized fan in Rr . The first
step is Proposition 14.4.3, which shows that GKZ cones are rational polyhedral
cones. The other properties we need will be covered in three lemmas that describe
the relative interiors, union, and faces of GKZ cones.
Three Lemmas. Our first lemma describes the relative interior of a GKZ cone.
e ,I is a GKZ cone, then
Lemma 14.4.4. If
e ,I ) = {a
e ,I | a is strictly convex and a (i ) > ai , i I }.
Relint(
e ,I ), then:
Furthermore, if a Relint(
(a) is the normal fan of Pa .
(b) I = {i | a (i ) > ai } = {i | Fi,a = }.
Proof. Let C,I W be the cone from the proof of Proposition 14.4.3. The first
line of (14.4.3) defines a subspace W0 W containing C,I , and then the second
line shows that C,I W0 is defined by the inequalities
(14.4.5)
hm , i i ai , for i
/ and i
/ I
hm , i i ai , for i I .
By assumption, and I come from a polytope Pa , a Zr . Let the vertices of

Pa be m for max . This gives the point (a, {m }) C,I that makes all of the
715
inequalities in (14.4.5) strict. It follows easily that the relative interior of C,I is
the subset of C,I defined by
hm , i i > ai , for i
/ and i
/ I
(14.4.6)
hm , i i > ai , for i I .
We claim that these strict inequalities correspond via (14.4.4) to the subset
e ,I | is strictly convex and (i ) > ai , i I }.
{a
To prove this, note that an arbitrary point (a, {m }) C,I gives a support function
a such that a (u) = hm , ui for u . Since max is built from the i for i
/ I ,
the first line of (14.4.6) is equivalent to the strict convexity of a by (a) (f) of
Lemma 6.1.13, and the second line is equivalent to a (i ) > ai for i I . This
e ,I ) follows.
proves our claim, and our description of Relint(
e
Now take a Relint(,I ) and suppose that (a, {m }) C,I maps to a. Then
m represents a on max , and the convexity of a implies that
a (u) = min hm , ui.
max
by Lemma 6.1.5. Then Theorem 7.2.2 shows that the m are the vertices of Pa . The
cone of the normal fan corresponding to this vertex is
Cone(Pa m ) ,
which is easily seen to be (Exercise 14.4.1). This proves part (a).
For part (b), take i I . Then Proposition 14.4.3 implies that a (i ) > ai
since a is in the relative interior. On the other hand, if i
/ I , then pick max
with i . The first line of (14.4.3) implies that a (i ) = ai , and then part (b)
follows easily.

A nice consequence of Lemma 14.4.4 is that the relative interiors of two GKZ
cones are either equal or disjoint.
Our second lemma describes the union of the GKZ cones.
S
e,I = 1 (C ), where the union is over all and I satLemma 14.4.5. ,I
R
isfying Proposition 14.4.1.
Proof. If a is in the union, then Pa is nonempty, so that R (a) C by the proof

of Proposition 14.3.5. Conversely, suppose that a R1 (C ) Qr . Then some
positive multiple a R1 (C ) Zr . By Proposition 14.4.1, a gives , I with
e ,I . Thus
a = (1/)(a)
[
e ,I ,
R1 (C ) Qr
,I
and then R1 (C )
,I ,I
follows by taking the closure.
716
The third lemma concerns the faces of a GKZ cone.

e ,I is a
Lemma 14.4.6. Every face of a GKZ cone is again a GKZ cone, and
e ,I if and only if refines and I I .

face of
Proof. We will use the cone C,I W from

L the proof of Proposition 14.4.3. Recall
that C,I W0 , where W0 W = Rr max MR is defined by the first line of
(14.4.3) and C,I W0 is defined by the inequalities from the second line, namely
(14.4.7)
hm , i i ai , for i
/ and i
/ I
hm , i i ai , for i I .
e ,I by (14.4.4).
The idea is to study the faces of C,I , which map to faces of
Recall that (a, {m }) C,I gives a polyhedron Pa with virtual facets Fi,a ,
normal fan a , and index set I,a = {i | Fi,a = }. Also note the following:
The m are the vertices of Pa by the proof of Lemma 14.4.4.
refines a since a SF().
I,a I . To see why, take i

/ I and pick max with i . This implies
hm , i i = ai , so that m Fi,a . Hence i
/ I,a .
Now suppose that F C,I is a proper face. Then F is defined by turning some
of the inequalities (14.4.7) into equalities. To keep track of which ones, define
IF = {(, i) | i
/ or i I , and hm , i i = ai for all (a, {m }) F}.
We claim that a and I,a are equal for all (a, {m }) Relint(F). For I,a , note
that since the m are the vertices of Pa , we have
Fi,a = hm , i i > ai for all max
i I and (, i)
/ IF for all max .
You will prove the second equivalence in Exercise 14.4.2. This shows that I,a
depends only on F when (a, {m }) Relint(F).
The argument for a will take more work. Recall from 6.1 that convexity
can be detected by looking at walls. We say that hm , i i ai is a (, , i )-wall
inequality if
, max , is a wall, i \ , i
/ I .
Every wall inequality appears among the inequalities in the first line of (14.4.7).
Furthermore, given (a, {m }) C and a (, , i )-wall inequality, we have
(14.4.8)
hm , i i = ai m = m .
This follows since is a wall and hm m , ui = 0 for all u .
The face F gives an equivalence relation on max as follows. Given cones

, max , we say that F if either = or there exists a chain of cones
= 0 , 1 , . . . , s1 , s = in max and i1 , . . . , is such that for 1 j s, we
717
have a ( j1 , j , i j )-wall inequality with ( j1 , i j ) IF . Thus F means

that and are connected by a chain of compatible wall inequalities that become
equalities in F. For (a, {m }) Relint(F), we will show that
m = m F .
One direction is easy: if F , then (14.4.8) and ( j1 , i j ) IF imply that
m j1 = m j for 1 j s, and m = m follows immediately. For the other
direction, suppose that m = m . Then m and m give the same vertex of Pa ,
which implies that and are contained in the same maximal cone
e a . Then
a suitably chosen line segment connecting interior points of , lies in
and
meets only walls of . This gives a compatible chain of wall inequalities that are
equalities. Since (a, {m }) Relint(F), inequalities indexed by (, i )
/ IF are
strict. Thus the wall inequalities in our chain come from IF , hence F .
It follows that each maximal cone of a is the union of s contained in an
equivalence class of F . This proves that a depends only on the face F when
(a, {m }) Relint(F). The same is true for I,a , so that via (14.4.4), we have
e a ,I
F
,a
when (a, {m }) Relint(F). Thus every face of a GKZ cone is a GKZ cone. From
here, Lemma 14.4.4 makes it straightforward to prove the final assertion of the
proposition. We leave the details to the reader (Exercise 14.4.2).

The Secondary Fan. The above results show that the set of GKZ cones
e GKZ = {
e ,I | , I as in Proposition 14.4.1}
e GKZ is a generalized fan, we need one further

has many nice properties. To see that
e of rational polyhedral cones is a generalized fan if and only if
observation: a set
e lies in .
e
Every face of an element of
e are pairwise disjoint.
The relative interiors of the elements of
You will prove this Exercise 14.4.3.

e GKZ is a generalized fan: GKZ cones are rational polyhedral by
It follows that
Proposition 14.4.3, their faces are again GKZ cones by Lemma 14.4.6, and their
e GKZ | =
relative interiors are pairwise disjoint by Lemma 14.4.4. Note also that |
1
R (C ) by Lemma 14.4.5.
We know from 6.2 that we can turn a generalized fan into an actual fan by
taking the quotient by its minimal face, which is the minimal face of every cone in
e GKZ , this minimal face is ker(R ) by Proposition 14.4.3. It follows
the fan. For
that the GKZ cones
e,I /ker(R ) Rr /ker(R ) = G
bR
,I =
718
form the secondary fan

GKZ = {,I | , I as in Proposition 14.4.1}.
Note that |GKZ | = C , so the maximal cones of GKZ have dimension dim G. The
maximal cones of GKZ are the chambers of the secondary fan. Another name for
the secondary fan is the GKZ decomposition, hence the notation GKZ .
Here are some properties of the secondary fan.
Theorem 14.4.7.
bR with |GKZ | = C .
(a) GKZ is a fan in G
(b) ,I ,I if and only if refines and I I .
(c) Cr // G X when 1 Relint(,I ).
Remark 14.4.8. Part (c) of the theorem tells us that the GIT quotient is constant
on the relative interiors of the GKZ cones.
Proof. We proved part (a) above, and part (b) follows from Lemma 14.4.6. For
part (c), assume that 1 Relint(,I ) and write = a for a Zr . The GIT
quotient Cr // G is the toric variety of Pa by Theorem 14.2.13. However, is the
normal fan of Pa by Lemma 14.4.4, and Cr // G X follows.

Our approach to the secondary fan is more general than what one finds in the
literature. The papers [141] and [222] assume that the i are nonzero and generate
distinct rays in NR ([141] also assumes that the i are primitive). We will study this
special case in 15.1, where we will see that the secondary fan has an especially
nice structure.
Generic Characters. Before giving examples of the secondary fan, we need to
relate GKZ cones to the generic characters introduced in 14.3.
Proposition 14.4.9. If 1 Relint(,I ), then the following are equivalent:
(a) is generic.
(b) is simplicial and i 7 Cone(i ) induces a bijection {1, . . . , r} \ I (1).
(c) ,I is a chamber of the secondary fan.
Remark 14.4.10. Using this proposition, one can show that the generic characters
determine the chambers of the secondary fan uniquely (Exercise 14.4.4).
Proof. Write = a for a Zr . By Theorem 14.3.14, is generic if and only if
Pa is simple of dimension r dim G and the nonempty virtual facets Fi,a , i
/ I , are
the actual facets of Pa , with no duplications. Then (a) (b) follows easily since
is the normal fan of Pa by Lemma 14.4.4.
719
e ,I = r.
To prove (b) (c), note that ,I is a chamber if and only if dim
Furthermore, the proof of Lemma 14.4.4 shows that

where W0 Rr
e ,I = dim C,I = dim W0 ,

dim
max MR
is defined by
hm , i i = ai , for i and i
/ I .
Hence it suffices to prove that (b) implies that dim W0 = r.

Assume (b) and take max . Then is simplicial of maximal dimension, so
that generators of its rays form a basis of NR . By assumption, the ray generators
can be chosen to be the i with i
/ I . It follows that for any b Rr and
max , the equations
hm , i i = bi , for i and i
/ I
have a unique solution m MR . When we do this for all max , we see that for
any b Rr , the equations defining W0 have a unique solution. Thus dim W0 = r.
Finally, assume (c). First observe that since (b) involves only and I , the
equivalence (a) (b) means that if (a) holds for one 1 Relint(,I ) =
Int(,I ), then it holds for all such 1. Hence it suffices to find one generic
character that gives an interior point of the chamber.

S
Let U = C \ Cone( ) , where the union is over all such that
b is generic if
dim Cone( ) < dim G. Then U is open and dense in C , and G
and only if 1 U . Since ,I is a chamber, its interior must have nonempty

intersection with U . This easily implies the existence of the required , and the
proof is complete.

We can finally give an example of a secondary fan.
Example 14.4.11. Consider Figure 5 from Example 14.3.7, which we reproduce
as Figure 7 on the next page. As we noted in Example 14.3.16, the non-generic
characters of C lie on the three rays generated by the i s. This shows that the
secondary fan has two chambers. You should compute and I for each of the six
GKZ cones in the secondary fan, and you should check that part (b) of Proposition 14.4.9 holds only in the interior of the chambers.
Another secondary fan is Figure 6 in 14.3, and Exercises 14.4.514.4.8 give

more secondary fans. Further examples will be given in Chapter 15.
Degenerate Fans. Recall that a generalized fan in NR is called degenerate when
it is not an actual fan. Equivalently,
(14.4.9)
is degenerate dim X < dim NR .
It is easy to determine where degenerate fans occur in the GKZ decomposition.
720
P =
P =
P =
P =
2
4
P =
1 = 3
Figure 7. C and the polytopes P for 1 C in Example 14.4.11
Proposition 14.4.12. If 1 Relint(,I ), then

(14.4.10)
is degenerate dim Cr // G < r dim G.
Furthermore:
(a) If is degenerate, then ,I is contained in the boundary of C .
(b) If = (1 , . . . , r ) consists of nonzero vectors, then
[
C =
,I .
degenerate
Proof. Take 1 Relint(,I ). Since dim NR = r dim G and Cr // G X

(Theorem 14.4.7), the equivalence (14.4.10) follows from (14.4.9).
Recall from (c) (e) of Proposition 14.3.6 that dim Cr // G has the expected
dimension r dim G when 1 lies in the interior of the cone C . Hence 1
must lie in the boundary when the dimension of dim Cr // G drops. Then when part
(a) follows from (14.4.10). Also recall that (c) (e) in Proposition 14.3.6 when
consists of nonzero vectors. In other words, when satisfies this condition, the
boundary of C is precisely where the dimension of the GIT quotient drops. Then
part (b) follows from (14.4.10).

Irrelevant Ideals. At the beginning of the section, we motivated the existence of
the secondary fan by noting that there are only finitely many possible irrelevant
ideals (14.4.1). Here we will show that the secondary fan can be described completely in terms of irrelevant ideals.
Recall that 1 C gives the polyhedron P Rr with virtual facets Fi, =
P V(xi ), and by (14.4.1), we have the irrelevant ideal
Q

T
B() =
iI
/ xi | I {1, . . . , r},
iI Fi, 6= .
By (14.3.8), the vertices of P determine the minimal generators of B(). Thus

Q
(14.4.11)
B() =
bF
/ i, xi | b is a vertex of P .
721
Example 14.4.13. Figure 8 shows an example of P and its virtual facets Fi, . This
picture and (14.4.11) make it easy to compute that B() = hx1 x2 , x2 x3 , x4 i. Note
that this is the polytope from Figure 7 when comes from the vertical axis.
F4,
F3,
P
F1,
F2,
Figure 8. A polytope P and its virtual facets
Given a GKZ cone ,I , define the ideal

Q
(14.4.12)
B(, I ) =
i
/ or iI xi | max .
The ideals B() and B(, I ) relate to GKZ cones as follows.

Proposition 14.4.14. If 1 C , then:
(a) 1 ,I if and only if B(, I ) B().
(b) 1 Relint(,I ) if and only if B(, I ) = B().
(c) is generic if and only if every minimal generator of B() has degree dim G.
Proof. We begin with two easy observations about the ideals B(, I ):
and I are easy to recover from B(, I ). To see why, observe first that
I = {i | B(, I ) hxi i} and second that the minimal generators of B(, I )
determine the maximal cones of by (14.4.12) and Proposition 14.4.1.
B(, I ) B( , I ) if and only if refines and I I . We leave the easy
proof to the reader.
Now suppose that 1 Relint(,I ) and write = a for a Zr . Then
is the normal fan of Pa and I = {i | Fi,a = } by Lemma 14.4.4, and it follows
easily from Proposition 14.4.1 that B(, I ) = B(). The converse follows easily
from the first bullet above. This proves part (a) of the proposition, and part (b)
follows easily from the second bullet and the description of the face relation given
in Lemma 14.4.6.
For part (c), suppose that is generic. Then Proposition 14.3.14 implies that
every vertex of P is contained in precisely r dim G virtual facets. By (14.4.11),
the corresponding minimal generator of B() has degree dim G. The converse is
equally straightforward.
722
Propositions 14.3.14, 14.4.9 and 14.4.14 give a robust understanding of what

it means for a character to be generic. Proposition 14.4.14 also gives an alternate
description of the face relation in the GKZ fan.
Corollary 14.4.15. Given GKZ cones ,I and ,I , we have
,I ,I B(, I ) B( , I ).
It follows that the ideals B(, I ) completely determine the secondary fan.
Here is what this looks like for a familiar example.
Example 14.4.16. Figure 7 from Example 14.4.11 shows P and its virtual facets
for five different s. Computing the ideals B() (going from left to right) gives
hx2 i hx1 x2 , x2 x3 , x2 x4 i hx1 x2 , x2 x3 , x4 i hx1 x2 , x2 x3 , x3 x4 , x1 x4 i hx1 , x3 i.
These five ideals are contained in the ideal of the trivial character, B(1) = h1i.
The inclusion relation of these ideals determines the face relation of the GKZ
fan, so that the chambers correspond to the minimal ideals hx1 x2 , x2 x3 , x2 x4 i and
hx1 x2 , x2 x3 , x3 x4 , x1 x4 i. The minimal generators of these two ideals all have degree
equal to 2. It follows that generic characters give elements in the interiors of the
corresponding cones, exactly as predicted by Proposition 14.4.14.
Exercises for 14.4.

14.4.1. This exercise is concerned with the proof of Proposition 14.4.1.
(a) Let a Rr and assume that Pa 6= . For a (possibly non-rational) vertex v Pa , define
the cone Cv = Cone(Pa v). Prove that
Cv = Cone(i | hv, i i = ai ) = Cone(i | v Fi,a ).
(b) Prove the final assertion made in the proof of Proposition 14.4.1.
14.4.2. This exercise is concerned with the proof of Lemma 14.4.6.
(a) As in the proof of the lemma, let (a, {m }) Relint(F). Prove that hm , i i > ai for
all max if and only if i I and (, i)
/ IF for all max .
(b) Complete the proof of the lemma.
e be a collection of polyhedral cones in a finite dimensional vector space W .

14.4.3. Let
e lies in
e and that the relative interiors of the cones
Assume that every face of a cone in
e
e is closed under
in are pairwise disjoint. The goal of this exercise is to prove that
e
intersection. For this purpose, take , .
(a) Given u , let (resp. ) be the minimal face of (resp. ) containing u.
Prove that = .
e for 1 i .
(b) Conclude that is a union 1 with i
(c) Show that there is some i such that = i . Hint: is closed under addition.
Take ui Relint(i ) and apply Lemma 1.2.7 to u1 + + u .
723
e is a fan in G
eR consisting of strongly convex rational polyhedral
14.4.4. Suppose that
e = C and that every generic 1 lies in the interior of a maximal
cones. Assume that ||
e
e = GKZ . This justifies the uniqueness claimed in Remark 14.4.8.
cone of . Prove that
14.4.5. Use the secondary fan to explain what is happening in Example 14.1.15.
b Z3 and 1 , . . . , 6 correspond to (2, 1, 1),

14.4.6. Assume that G (C )6 with G
3
(1, 0, 1) Z . Determine the number of chambers in the secondary fan. Hint: Think
about generic characters and draw a picture.
14.4.7. Let G = {(t,t r u,t, u) | t, u C } (C )4 . In Examples 14.3.7 and 14.4.11, we

drew the secondary fan when r = 1 and showed that for generic characters, the GIT quotient
was either H1 or P 2 . Draw the secondary fan for general r 2 and determine the generic
GIT quotients. (One will be Hr , while the other will be a weighted projective plane.) You
should also draw the generalized fan for each cone in the secondary fan.
14.4.8. The projective bundle X = P(OP2 OP2 (1)) is a 3-dimensional smooth projective
toric variety. The description given in Example 7.3.5 shows that has five minimal generators. Since Pic(X ) Z2 , the quotient construction of X uses a subgroup G (C )5
with G (C )2 . In this situation, the i are the minimal generators u .
(a) Determine the i . Hint: In terms of the matrices in the exact sequence (14.2.10), you
know the matrix B. Use this to find A. Three of the i are equal.
(b) Use Lemma 14.2.1 to give an explicit description of the group G.

(c) Determine the secondary fan and compute all possible GIT quotients C5 // G as we
vary . Hint: There are five distinct quotients.
(d) What happens when X = P(OP2 OP2 (a)) for a 2? Do you see the similarity to
Exercise 14.4.7?
(e) Generalize part (d) by computing the secondary fan coming from the quotient construction of X = P(OPs OPs (a)). Hint: Treat the cases a = 0 and a 1 separately.
Chapter 15
Geometry of the
Secondary Fan
b varies.
This chapter will continue our study of the GIT quotients Cr // G as G
Recall the key players introduced in Chapter 14:
bR and = (1 , . . . , r ), i NR . We
G (C )r gives = (1 , . . . , r ), i G
b
also have the cones C = Cone() GR and C = Cone() NR .
The secondary fan GKZ consists of GKZ cones ,I and has support |GKZ | =
C . A chamber is a full dimensional GKZ cone.
If 1 Relint(,I ), then Cr // G X , where X is semiprojective and

is a generalized fan with || = C . Furthermore, the cones of are generated
by the i for i
/ I , as described in Proposition 14.4.1.
The goal of this chapter is to understand the structure of the GKZ cones and what
happens to the associated toric varieties as we move around the secondary fan. We
will discuss a variety of topics, including nef cones, moving cones, Gale duality,
triangulations, wall crossings, and the toric minimal model program.
15.1. The Nef and Moving Cones

Before we can study GKZ cones, we first need to generalize the results of 6.3 and
6.4 concerning the nef and Mori cones of toric varieties.
The Nef Cone of a Toric Variety. Given a generalized fan in NR , the vector
space of support functions SF() contains the lattice
SF(, N) = { SF() | (|| N) Z}.
When || is convex, we also have CSF() = { SF() | is convex}.
725
Chapter 15. Geometry of the Secondary Fan
726
In 6.3 we studied the nef cone Nef(X ) for a fan with full dimensional
convex support. Our results apply to generalized fans as follows.
Theorem 15.1.1. Let be a generalized fan with full dimensional convex support
in NR . Then:
(a) There is an exact sequence
0 M SF(, N) Pic(X ) 0.
(b) A Cartier divisor on X is basepoint free if and only if it is nef.
(c) The nef cone Nef(X ) is contained in Pic(X )R , and its inverse image in SF()
is CSF().
(d) If X is semiprojective, then Nef(X ) Pic(X )R is full dimensional.
Proof. As explained in 6.2, the generalized fan in NR gives a genuine fan in
N R , where N = N/(0 N) and 0 is the minimal cone of . Let M M be the
dual of N N. Since X = X , the correspondence between Cartier divisors and
support functions (Theorem 4.2.12) gives an exact sequence
0 M SF(, N) Pic(X ) 0.
To relate this to the generalized fan , note that composition with NR N R
induces an injection SF(, N) SF(, N). This gives the commutative diagram
(ignore the dotted arrow for now):
0
/M
O
?
/ SF(, N)
O
?
/ Pic(X )
/0
/M
/ SF(, N)
/ Pic(X )
/ 0.
(15.1.1)
However, the vertical injections have isomorphic cokernels, because:

A element of SF(, N) comes from SF(, N) if and only if it vanishes on the
kernel of NR N R .
Every element of SF(, N) is linear on the kernel of NR N R since the kernel
is the minimal cone of .
It follows that the dotted arrow exists such that the above diagram commutes and
has exact rows. This proves part (a).
Part (b) follows from Theorem 6.3.12. For part (c), (6.3.2) implies
Nef(X ) N 1 (X ) = Pic(X )R .
In Exercise 15.1.1 you will show that the convexity criteria of Theorem 7.2.2 apply
to generalized fans. This plus part (b) complete the proof of part (c).
If X is semiprojective and is a fan, then Theorem 7.2.4 and Proposition 7.2.9
imply that CSF() contains a strictly convex support function. Strict convexity is
727
an open condition, so that CSF() is full dimensional in SF(). Then part (c)
implies that Nef(X ) is full dimensional Pic(X )R , as claimed in part (d).
To complete the proof, we need to consider the case of a generalized fan. Here,
the proof follows easily from (15.1.1) and the previous paragraph.

The Toric Cone Theorem. We can also describe the dual of Nef(X ) in terms of
walls when is a generalized fan with full dimensional convex support. A wall
of such a fan is a facet of full dimensional cones 6= . In Exercise 15.1.2
you will show that the orbit closure V ( ) is a complete torus-invariant curve in X
and that all complete torus-invariant curves arise this way.
The toric cone theorem stated in Theorem 6.3.20 generalizes as follows.
Theorem 15.1.2. If is a generalized fan with full dimensional convex support,
then we have
X
NE(X ) = NE(X ) =
R0 [V ( )].
a wall
Furthermore, NE(X ) is strongly convex when X is semiprojective.

Proof. Let be the genuine fan determined by . Since walls of correspond
bijectively to walls of , Theorem 6.3.20 gives the desired formula for NE(X ).
If X is semiprojective, then NE(X ) is strongly convex since its dual Nef(X ) is
full dimensional by Theorem 15.1.1.

When X is semiprojective, NE(X ) is minimally generated by its edges, called
extremal rays. If R is the extremal ray generated by the class of a curve C, then the
corresponding facet of Nef(X ) is defined by D C = 0 for [D] Nef(X ).
Also, in the situation of Theorem 15.1.2, we have a proper map : X U ,

where U is the affine toric variety associated to || as in (7.2.1). Then
N1 (X ) = N1 (X /U ).
Here, N1 (X /U ) is generated by irreducible curves lying in fibers of , modulo
numerical equivalence, which equals N1 (X ) since is proper and U is affine.
Hence we are in what is often called the relative case.
The Simplicial Case. When is simplicial with full dimensional convex support,
we have an exact sequence
(15.1.2)
0 MR R(1) Pic(X )R 0,
and since N1 (X ) is dual to Pic(X )R under intersection product, taking duals gives
the exact sequence
(15.1.3)
0 N1 (X ) R(1) NR 0.
728
Thus we can interpret classes in N1 (X ) as relations among the u s. In particular,

the class of an irreducible complete curve C X is represented by the relation
X
(15.1.4)
(D C)u = 0.
This follows from Proposition 6.4.1 and (6.4.2).
Nef Cones and the Secondary Fan. Now consider the secondary fan determined
by and as at the beginning of the chapter. A GKZ cone ,I gives the toric
variety
Cr // G X
for all 1 Relint(,I ). The nef cone Nef(X ) relates to ,I as follows.
Proposition 15.1.3. For any GKZ cone ,I , we have:
I
(a) ,I Nef(X ) R0
.
(b) dim ,I = dim Pic(X )R + |I |.

Proof. Since GIT quotients are semiprojective, part (b) follows from part (a) by
Theorem 15.1.1. For part (a), recall from Definition 14.4.2 that a ,I has convex support function a CSF(). Then we have an isomorphism of cones
e ,I CSF() RI ,
e,I /MR ,I and

where a ,I maps to (a , (a (i ) + ai )iI ). Since
CSF()/MR Nef(X ) by Theorem 15.1.1, part (a) follows immediately.

The Moving Cone of the Secondary Fan. The nicest case of Proposition 15.1.3 is
when I = , for here the GKZ cone is , Nef(X ). There may be many GKZ
cones with I = . Let
[
(15.1.5)
MovGKZ =
,
be the union of all GKZ cones with I = . This union has a nice structure.
bR .
Proposition 15.1.4. MovGKZ is a convex polyhedral cone in G
Proof. We need only prove convexity. Take 1 1 , and 2 2 , and consider

= t1 + (1 t)2 ,
Rr
t [0, 1].
Write j = R (a j ) for a j
and set a = ta1 + (1 t)a2 . By hypothesis, the
virtual facets Fi,a1 Pa1 and Fi,a2 Pa2 are nonempty. An easy calculation shows
that
tFi,a1 + (1 t)Fi,a2 Fi,a ,
which implies that the virtual facets of Pa are also nonempty. If Relint(,I ),
then I is the set of indices of empty virtual facets of Pa by Lemma 14.4.4. Hence
I = , so that MovGKZ .
729
We call MovGKZ the moving cone of the secondary fan. The name moving
cone will be explained later in the section.
Here is an example where the moving cone is small.
bR R2
Example 15.1.5. Let G = {(t, u,tu) | t, u C }. One easily computes that G
with basis 1 , 2 , and 3 = 1 + 2 . Furthermore 1 = 2 = e1 and 3 = e1 in N.
The secondary fan is illustrated in Figure 1. Virtual facets are indicated by dotted
lines. Hence the moving cone is the diagonal ray generated by 3 .
P =
F2
F1=F3
P =
F2
F1
F3
F1=F2 F3
P =
2
3
1
P =
F1
F2
F3
F2=F3
P =
F1
Figure 1. A secondary fan with small moving cone
The Geometric Case. Our next result characterizes when the secondary fan has a
chamber satisfying I = . We say that = (1 , . . . , r ) is geometric if every i is
nonzero and the i generate distinct rays in NR .
Proposition 15.1.6. The secondary fan has a chamber with I = if and only if
is geometric.
Proof. Suppose we have a chamber of the form , and pick 1 Int(, ).
Then is generic, so that by Proposition 14.4.9, i 7 Cone(i ) gives a bijection
{1, . . . , r} (1) since I = . It follows immediately that is geometric.
Conversely, assume that is geometric and for each 1 i r, let ui be the

minimal generator of Cone(i ) N. By hypothesis, this gives r primitive elements
of N. Now take any chamber ,I . By Proposition 14.4.9, we have a bijection
{1, . . . , r} \ I (1) defined by i 7 Cone(i ). If I = , then we are done. If not,

let be the refinement of obtained by successive star subdivisions (as defined
in 11.1) at ui for i I . The fan will depend on the order in which we perform
the star subdivisions, but all such have the following properties:
is simplicial since is simplicial and the star subdivision of a simplicial fan

is simplicial.
i 7 Cone(i ) is a bijection {1, . . . , r} (1) since the star subdivisions add the
rays Cone(i ), i I , to the original fan . This follows since is geometric.
730
X is semiprojective since X is semiprojective and X X is projective,

being a composition of projective morphisms.
It follows that and I = satisfy the conditions of Proposition 14.4.1 and hence
give the GKZ cone , . Furthermore, the first two bullets and Proposition 14.4.9
imply that this cone is a chamber.

Besides being geometric, a stronger condition is when is primitive geometric,
which means that is geometric and each i N is primitive. Here is an example.
Example 15.1.7. Let X be a projective toric variety. The quotient construction
X (C(1) \ Z())//G from 5.1 is a GIT quotient by Example 14.2.14. As
noted in Example 14.3.4, i s come from the divisor classes [D ] and the i s are
the minimal generators u . Thus is primitive geometric.
The Pseudoeffective and Moving Cones of a Normal Variety. Let X be a normal

variety. In addition to the nef cone Nef(X ), other interesting cones live in N 1 (X ),
in particular, the pseudoeffective cone Eff(X ) and the moving cone Mov(X ). These
cones are related to the nef cone by the inclusions
Nef(X ) Mov(X ) Eff(X ) N 1 (X ).
The pseudoeffective cone Eff(X ) is easy to define: it is the closure of the cone
generated by effective R-Cartier divisor classes. Here is one case where the pseudoeffective cone is easy to describe.
Lemma 15.1.8. If X is simplicial and semiprojective, then
Eff(X ) = Eff(X ) = Cone([D ] | (1)).
Proof. We know that N 1 (X ) = Pic(X )R since has full dimensional convex
support, and every D is Q-Cartier since X is simplicial. The classes [D ] are
clearly effective. Conversely, if D is effective, then H 0 (X , OX (D)) 6= 0. We may
assume that D is torus-invariant, and then Proposition 4.3.2 implies that there is
m M with D + div( m ) 0. Then [D] = [D + div( m )] Cone([D ]).

The moving cone Mov(X ) was introduced in [171]. An effective Weil divisor
D 0 on X has a fixed component if there is an effective divisor D0 6= 0 such that
every effective divisor D D satisfies D D0 . This implies that D = E + D0 ,
E 0, where every effective divisor D D of the form D = E + D0 , E 0. In
other words, as we vary D = E + D0 in its linear equivalence class, D0 is fixed.
We can formulate this in terms of sheaves as follows. If D D0 0 and
D0 6= 0, then D0 is a fixed component of D if and only if the natural inclusion of
sheaves OX (D D0 ) OX (D) induces an isomorphism
(15.1.6)
H 0 (X , OX (D D0 )) H 0 (X , OX (D))
(Exercise 15.1.3). Here is an example of a divisor with a fixed component.
731
Example 15.1.9. For the Hirzebruch surface H2 , we have the usual divisors D1 ,
D2 , D3 , D4 such that the classes of D3 , D4 generate the nef cone. Let D = D4
and D = D2 + D4 . Figure 4 from Example 6.1.2 shows that the divisors D and D
have the same polytopes and hence their sheaves have the same global sections by
Proposition 4.3.3. Since D D2 = D, the discussion leading up to (15.1.6) implies
that D2 is a fixed component of D .
A Weil divisor D on X is movable if it doesnt have a fixed component, and

then the moving cone of X is defined by
(15.1.7)
Mov(X ) = Cone([D] | D is Cartier and movable) N 1 (X ).
The Moving Cone in the Primitive Geometric Case. We now explain how the
moving cone MovGKZ defined in (15.1.5) relates to the moving cone (15.1.7).
Theorem 15.1.10. If is primitive geometric, then the moving cone MovGKZ is the
union of the GKZ chambers with I = . Furthermore, if , is a chamber, then:
(a) X is simplicial and semiprojective, with as minimal generators of (1).
(b) If 1 Relint(, ), then the GIT quotient

Cr // G = Cr \ Z() //G X
is precisely the quotient construction described in Theorem 5.1.11.

bR that takes the cones
(c) There is a natural isomorphism Pic(X )R G
Nef(X ) Mov(X ) Eff(X )
to the cones
MovGKZ C .
Proof. MovGKZ is the union of chambers with I = by Propositions 15.1.4 and

15.1.6. Now let , be a chamber. By Proposition 14.4.9, is simplicial and
i 7 i = Cone(i ) gives a bijection {1, . . . , r} (1). Thus (1) = {1 , . . . , r },
and the corresponding divisors will be denoted D1 , . . . , Dr . This proves part (a).
Since is primitive, the map : M Zr from (14.2.2) is the same as the
corresponding map in the exact sequence
0 M Z(1) = Zr Cl(X ) 0
b which
from (5.1.1) in Chapter 5. This gives a natural isomorphism Cl(X ) G,
shows that G is the group from the quotient construction in 5.1. Furthermore, if
B() is the irrelevant ideal from 5.1, then B() = B(, ) = B(), where the first
equality follows from Proposition 14.4.14 and the second from (14.4.12). From
here, is it straightforward to complete the proof of part (b).
b induces Pic(X )R G
bR since X is
For part (c), the isomorphism Cl(X ) G
simplicial. This takes Nef(X ) to , by Proposition 15.1.3. Since [Di ] 7 i , this
isomorphism takes Eff(X ) to C by Lemma 15.1.8.
732
Let Mov(X ) Pic(X )R be the cone generated by classes of movable divisors

(remember that every Weil divisor is Q-Cartier since X isP
simplicial). We prove
Nef(X ) Mov(X ) as follows. A nef Cartier divisor D = ri=1 ai Di has no basepoints by Theorem 15.1.1, so that its Cartier data {m }max lies in the polyhedron
PD by Theorem 7.2.2. This implies that the coefficient of Di in D + div( m ) 0
vanishes when i (1), hence D is movable.
We also note that X contains the toric variety X(1) whose fan consists of the
rays in (1), together with the trivial cone {0}. Since X(1) X is an open subset
whose complement has codimension 2, X and X(1) have the same Weil divisors
via the restriction D 7 D X(1). They also have the same principal divisors and
same class group. It follows easily that
(15.1.8)
X and X(1) have the same movable divisors.
For the remainder of the proof, fix a chamber , and identify Pic(X )R with
b
GR . If , is another chamber, then (1) = {1 , . . . , r } = (1), so that
X X(1) = X (1) X .
(15.1.9)
bR , these inclusions and (15.1.8)

When we identify Pic(X )R and Pic(X )R with G
imply that X and X have the same movable divisors, i.e., Mov(X ) = Mov(X ).
Since , = Nef(X ) Mov(X ), it follows , Mov(X ). This proves that
MovGKZ Mov(X ) since MovGKZ is the union of the chambers , .
P
For the opposite inclusion, take an effective Cartier divisor D = ri=1 ai Di and
assume that [D]
/ MovGKZ . Then [D] Relint( ,I ) with I 6= . We claim that
P
/ Mov(X ).
D0 = iI Di is a fixed component of D, which will imply [D]
By (15.1.6) and Proposition 4.3.3, the claim will follow once we prove that
PDD0 M = PD M . The inclusion PDD0 M PD M is trivial since D0 0.
For the other direction, take any m PD M. In the notation of 14.2, PD = Pa for
a = (a1 , . . . , ar ), so that the inequality hm, i i ai must be strict for i I since
I consists of the indices of empty virtual facets. Since hm, i i and ai are integers,
it follows that hm, i i (ai 1), hence m PDD0 M. This proves our claim.
We have now shown that MovGKZ = Mov(X ). Since MovGKZ is closed, the
same is true for Mov(X ). Thus MovGKZ = Mov(X ) and part (c) is proved.
The proof of Theorem 15.1.10 shows that Mov(X ) = Mov(X ), and we saw
that Eff(X ) = Eff(X ) in Lemma 15.1.8. Furthemore, these cones are polyhedral,
as is Nef(X ). This is typical of how life is simpler in the toric case. In general,
these cones can be much more complicated.
For a simplicial semiprojective toric variety X , Theorem 15.1.10 gives a vivid
description of the moving cone of X . Namely,
[
Mov(X ) =
Nef(X ),
733
where the union is over all simplicial fans in NR such that X is semiprojective and (1) = (1). Notice that X is isomorphic to X in codimension 1 by
(15.1.9). In 15.3 we will give a careful description of the birational isomorphism
X 99K X .
Here is an easy example of the moving cone.
Example 15.1.11. For the Hirzebruch surface H2 considered in Example 15.1.9,
we have the secondary fan pictured in Figure 2, where i = [Di ]. We showed in
Example 15.1.9 that D2 + D4 has a fixed component, so that the class [D2 + D4 ] is
not in the moving cone. This is also clear since I 6= in the left chamber. Thus
the moving cone is the right chamber, which is Nef(H2 ).
P =
P =
P =
P =
[D2]
[D4]
P =
[D1] = [D3]
Figure 2. The secondary fan of H2
In 15.2 we will give a more substantial example of a moving cone when we

study an alternate description of the moving cone and explore the relation between
the secondary fan and triangulations.
In the literature, the nef and moving cones are often defined in the relative
case of a projective morphism f : X S (see [171]). Here, numerical equivalence
is defined using complete curves in X that map to a point under f , and N 1 (X /S)
is defined using this notion of numerical equivalence. When the base S is affine,
every complete curve in X maps to a point, so that the definitions of numerical
equivalence and N 1 (X ) given in 6.2 agree with the relative versions. This applies
in particular to a semiprojective toric variety, since such a variety is projective over
an affine toric variety. See also [71, Sec. 1.4].
Exercises for 15.1.
15.1.1. Prove that the convexity criteria of Theorem 7.2.2 apply to generalized fans with
full dimensional convex support.
15.1.2. Let X be the toric variety of a generalized fan in NR .
(a) Prove that there is a bijection 7 O() between cones of and TN -orbits in X such
that dim O() = codim .
(b) Prove that if is a wall, then the orbit closure V ( ) is isomorphic to P1 .
734
15.1.3. Let D and D0 be effective Weil divisors on a normal variety X.

(a) Construct a natural inclusion of sheaves OX (D D0 ) OX (D).
(b) Prove that D0 is a fixed component of D if and only if the map of part (a) induces an
isomorphism of global sections as in (15.1.6).
B
15.1.4. Consider the exact sequence 0 Z2 Z4 Z2 0, where
0 1

1 1
, A = 1 0 1 1 .
B=
1 0
0 1 2 1
1 1
As in 14.2, the i s are the rows of B and the i s are the columns of A.
(a) Determine the group G (C )4 .
(b) Compute the secondary fan and, for each chamber, draw the corresponding fan that
gives the GIT quotient in that chamber.
(c) What is the moving cone in this case?
15.1.5. Suppose that = (1 , . . . , r ) consists of vectors in Z2 . Also assume that is
geometric and C R2 is strongly convex with C = Cone(1 , r ).
(a) Use Proposition 14.4.9 to prove that GKZ has 2r2 chambers. Hint: First show that if
,I is a chamber, then I {2, . . . , r 1} and that is uniquely determined by I .
(b) Explain why MovGKZ consists of a single chamber.
15.1.6. As in Example 15.1.7, a projective toric variety X gives a GIT quotient where
consists of the u , (1). The GKZ cone , equals Nef(X ) by Proposition 15.1.3.
Let D be a torus-invariant nef divisor on X and set = [D] Pic(X ). By Theorem 14.4.7,
1 Relint( , ), where refines . Prove that Cr // G X is the toric variety
XPD in Example 14.2.11 and Theorem 6.2.8. See also Example 14.2.14.
15.1.7. Let X be the toric variety of a generalized fan in NR . Parallel to the definition
of semiprojective, we say that X is semicomplete if the map X Spec(H 0 (X , OX )) is
proper and X has a torus fixed point. Prove that X is semicomplete if and only if has
full dimensional convex support. Hint: See the proof of Proposition 7.2.9.
15.1.8. Assume that is geometric. Prove that there is a bijective correspondence between
chambers of secondary fan and simplicial fans such that (1) {Cone(i ) | 1 i r},
|| = C , and X is semiprojective.
15.2. Gale Duality and Triangulations

In 15.1, we described various cones in C in terms of divisors on toric varieties.
In this section we will instead focus on the combinatorial structure of these cones.
We begin with the chambers of the secondary fan. Our goal is to show that they can
be constructed in a purely combinational way that makes nice use of Gale duality.
A subset J {1, . . . , r} is a -basis if |J| = dim G and the vectors i , i J, form
bR . Then Cone( J ) = Cone(i | i J) is a full dimensional simplicial
a basis of G
bR . These cones relate nicely to the chambers of GKZ .
cone in G
735
To see how this works, fix a chamber ,I . The lattice points in its interior
are all generic by Proposition 14.4.9. On the other hand, for a -basis J, the lattice
points in the boundary of Cone( J ) are nongeneric by definition. It follows easily
that for any -basis J, we have
(15.2.1)
either ,I Cone( J ) or Int(,I ) Cone( J ) = .
The surprise is that the -bases J that satisfy ,I Cone( J ) have a very nice
structure and that the chamber ,I is uniquely determined by these bases. Here
is the precise statement.
Proposition 15.2.1. Let ,I be a chamber of the secondary fan. Then:
(a) If max , then J = {i | i
/ or i I } is a -basis.
(b) if J is a -basis, then ,I Cone( J ) if and only if J = J for some max .

T
(c) ,I = max Cone( J ).
Proof. Our proof is based on [28, Lem. 4.2]. Since ,I is a chamber, we know
that is simplicial and that i 7 Cone(i ) induces a bijection {1, . . . , r}\I (1).
Thus, given a maximal cone max , the i with i
/ I form a basis of NR .
bR ,
By Gale duality (Lemma 14.3.1), the i with i
/ or i I form a basis of G
i.e., J = {i | i
/ or i I } is a -basis. This proves part (a).
For part (b), we first prove that if max , then
,I Cone( J ).
Take 1 Int(
P,I ) and let b P be the vertex corresponding to . This easily
implies that b = iJ i ei with i 0, so that 1 = R (b) Cone( J ). Then
the desired inclusion follows from (15.2.1).
J and take 1 Int(,I ).

Next assume
P,I Cone( J ) for a -basisP
Write 1 = iJ i ei with i > 0. Then b = iJ i ei Rr is a point of P
with precisely dim G nonzero coordinates. Since is generic, the vertices of P
have the same property by Proposition 14.3.14. It follows that b is a vertex and
hence corresponds to a maximal cone max . Since J is the set of indices of
nonzero coordinates of b, it follows easily that J = J .
T
Part (b) implies ,I max Cone(T
J ), and then part (c) will follow once
we prove the opposite inclusion. Take max Cone( J ) and pick a Rr
e ,I . This means finding a support
with = R (a). It suffices to prove that a
function CSF() such that (i ) ai for all i and (i ) = ai for i

/ I .
For
Pmax , our hypothesis on implies that Cone( J ), so that we can
write
=
iJ i i with i 0. Hence there is m MR such that a + R (m ) =
P
e
.
Thus
iJ i i
(15.2.2)
hm , i i ai for all i
hm , i i = ai for all i
/ J , i.e., for all i , i
/ I .
736
Now define
(u) = min hm , ui,
max
u C .
This function is clearly convex, and by (15.2.2) it satisfies

(i ) ai for all i
(i ) = hm , i i = ai for all i , i
/ I .
The second line implies that SF(), and it follows that has the required
properties. This completes the proof.

Here is an example that will appear several times in this section.
B
Example 15.2.2. Consider the exact sequence 0 Z3 Z5 Z2 0, where
2
1 1
2 1 1

1 2 4 1 0
0 1 , A =
.
B= 1
2 1 4 0 1
2
1 1
2 1 1
As in 14.2, the i s are the rows of B and the i s are the columns of A. The
secondary fan shown in Figure 3 has five chambers. The shaded chamber , is
1
2
4
Figure 3. The secondary fan and a selected chamber ,
contained in the cones

Cone(1 , 5 ), Cone(1 , 3 ), Cone(4 , 5 ), Cone(2 , 5 ).
By Proposition 15.2.1, the maximal cones of are the complementary cones
Cone(2 , 3 , 4 ), Cone(2 , 4 , 5 ), Cone(1 , 2 , 3 ), Cone(1 , 3 , 4 ).
737
These lie in R3 , but since the i s lie on the plane z = 1, we can intersect with
this plane to get the triangulation of the rectangle Conv() shown in Figure 4.
1
3
Figure 4. The fan intersected with the plane z = 1
Our second example is related to the classification theorem proved in 7.3.

Example 15.2.3. Let X be a smooth projective toric variety of dimension n whose
fan in NR Rn has n + 2 rays, with ray generators u1 , . . . , un+2 . Then we have a
bR =
geometric quotient X (Cn+2 \ Z())/G, where G (C )2 . Furthermore, G
2
Pic(X )R R . The i , 1 i n + 2, are the classes of the torus-invariant prime
divisors, and C = Cone() is the effective cone of X . The secondary fan refines
C , and the GKZ cone , is the nef cone Nef(X ) (Proposition 15.1.3). Note
also that C is strongly convex since is complete (Proposition 14.3.10).
Fix an ample class Nef(X ). Since Pic(X )R R2 , the line through
divides Pic(X )R into two half-planes H + and H . Let
P = {i | i H + }, |P| = r + 1
Q = { j | j H }, |Q| = s + 1,
as illustrated by Figure 5. Take any i P and j Q. Then i , j form a basis of

r + 1 is with i H +
z
}|
{
6
CO
C
C

H+

*

C

C

CX
XXX
XX
XX
z
s + 1 js with j H
Figure 5. The partition of induced by
Pic(X )R , so that {i, j} is a -basis. Note also that Cone(i , j ) since i and
738
j lie on opposite sides of the line determined by . By (15.2.1), it follows that

, Cone(i , j ). Using Proposition 15.2.1, we conclude that
i, j = Cone(uk | k 6= i, j)
is a maximal cone of and that all maximal cones of are of this form. This
proves (7.3.8), which is a key step in the proof of Kleinschmidts classification
theorem of smooth projective toric varieties of Picard number 2 (Theorem 7.3.7).
The theory developed here makes (7.3.8) easy to see.
The cones J , max , from Proposition 15.2.1 are an example of a bunch

of cones in the terminology of Berchtold and Hausen. Their paper [24] shows that
many aspects of toric geometry can be explained in the language of bunches. For
example, they give a bunch proof of Kleinschmidts classification theorem from
7.3 in [24, Prop. 10.1].
The Moving Cone. Proposition 15.2.1 shows how to represent the chambers of
the secondary fan as intersections of cones generated by certain subsets of . Our
next result, taken from [134], shows that the moving cone MovGKZ has a similar
representation.
Proposition 15.2.4. The moving cone of the secondary fan is the intersection
MovGKZ =
r
\
i=1
Cone(1 , . . . , bi , . . . , r ).
Proof. Take 1 C . The key observation is that
Fi, 6= 1 Cone(1 , . . . , bi , . . . , r ).
This follows from Fi, = P V(xi ) and P = R1 ( 1) Rr0 . Now suppose

that 1 Relint(,I ). Since I = {i | Fi, = } by Lemma 14.4.4, we obtain
T
I = 1 ri=1 Cone(1 , . . . , bi , . . . , r ).
Then we are done since MovGKZ is the union of all GKZ cones with I = .
In the primitive geometric case, Theorem 15.1.10 implies that the moving cone
is
Mov(X
) for any chamber , . In this case, i = [Di ], and a divisor D =
Pr
a
D
i=1 i i is movable if for
Peach i, we can move it away from Di , i.e., we can find
a linear equivalence D j6=i b j D j 0. You should check that this translates
exactly into the condition of Proposition 15.2.4.
Here are two examples to illustrate this proposition.
Example 15.2.5. In Figure 2 from Example 15.1.11, we have i = [Di ]. Looking

at the figure, we see that Cone(1 , 3 , 4 ) is the right chamber and
Cone(1 , 2 , 3 ) = Cone(1 , 3 , 4 ) = Cone(2 , 3 , 4 ) = C .
739
By Proposition 15.2.4, the moving cone is the intersection of these cones, which is
clearly the right chamber. This confirms the result of Example 15.1.11.
Example 15.2.6. In Figure 3 from Example 15.2.2, we see that Cone(1 , 2 , 4 , 5 )

is the first quadrant and that Cone(1 , . . . , bi , . . . , 5 ) = R2 for i = 1, 2, 4, 5. Then
Proposition 15.2.4 implies that the moving cone is the first quadrant, which by
Figure 3 is the union of three GKZ chambers.
Triangulations. The triangulation in Example 15.2.2 generalizes nicely. Before

giving the general theory, let us explore this example in more detail.
Example 15.2.7. Let and be as in Example 15.2.2, and recall that 1 , . . . , 5 lie
in the plane z = 1. The convex hull Q = Conv() is a rectangle. Figures 3 and 4
illustrate the triangulation of Q obtained from one of the GKZ chambers. Doing
this for every chamber gives the picture shown in Figure 6.
1
2
4
Figure 6. The secondary fan and its associated triangulations in Example 15.2.7
Note that each GKZ chamber gives a triangulation of Q such that the vertices
of the triangles lie in . Some elements of can be omitted provided that we
triangulate all of Q . The triangulations that use every element of correspond to
the chambers that make up the moving cone.
To generalize this construction, take distinct elements 1 , . . . , r N lying on

an integral affine hyperplane H NR with 0
/ H. Thus there are m M and
a Z \ {0} such that hm, i i = a for all i. This gives the lattice polytope
Q = Conv() = Conv(1 , . . . , r )
lying in the hyperplane H. We assume that Q has full dimension in H. Thus Q

has codimension 1 in NR and the cone C = Cone() has full dimension in NR .
Note also that C is strongly convex.
740
A triangulation T of is a collection of simplices satisfying:
Each simplex in T has codimension 1 in NR with vertices in .

The intersection of any two simplices in T is a face of each.
The union of the simplices in T is Q .
Figure 6 shows all triangulations of the vectors = (1 , . . . , 5 ) given by the rows

of the matrix B in Example 15.2.2. See [80] for an introduction to triangulations.
There is bijective correspondence between triangulations T of and simplicial
fans such that || = C and (1) {Cone(i ) | 1 i r}. The correspondence
is easy to describe:
Given T , the cones of are the cones over the simplices of T and their faces.
Given , the simplices of T are Q for max .
This correspondence, written 7 T = Q , will be used frequently in the our

discussion of triangulations.
We next define an especially nice class of triangulations, following [264].
Given nonnegative weights = (1 , . . . , r ) Rr0 , we get the cone
C, = Cone((1 , 1 ), . . . , (r , r )) NR R.
When consists of five equally spaced points on a line in R2 , Figure 7 shows C,

for one choice of the weights . In the figure, the lengths of the dashed lines are
Figure 7. The cone C , determined by and weights
determined by , and (4 , 4 ) is in the interior of the cone. Projection onto NR

maps C, onto C .
The lower hull of C, consists of all facets of C, whose inner normal has a
positive last coordinate. Projecting the facets in the lower hull and their faces gives
a fan in NR such that | | = C and (1) {Cone(i ) | 1 i r}.
Definition 15.2.8. A triangulation T of is regular if there are weights such
that is simplicial and T = Q .
741
To see how this relates to the secondary fan, note that gives an injection
M Zr defined by m 7 (hm, 1 i, . . . , hm, r i) as in 14.2. The cokernel of this
b of a subgroup G (C )r . Hence the theory developed
map is the character group G
bR , so that
in Chapter 14 applies. In particular, Lemma 14.3.2 implies that C = G
the secondary fan is complete in this situation.
Proposition 15.2.9. For as above, the map ,I 7 T = Q is a bijection
between chambers of the secondary fan and regular triangulations of .
Proof. If ,I is a chamber, then is a simplicial fan such that || = C and

(1) {Cone(i ) | 1 i r}. Thus T = Q is a triangulation of . If another
chamber ,I maps to the same triangulation T , then = since the maximal
cones of the fan are the cones over the simplices of T . Then Proposition 14.4.9
implies that I = I . Hence the map from chambers to triangulations is injective.
We prove that T = Q is regular as follows. Take Relint(,I ) and
pick a = (a1 , . . . , ar ) Rr0 that maps to . By Lemma 14.4.4, the support function
a is strictly convex with respect to . Furthermore,
a (i ) = ai for i
/ I , a (i ) > ai for i I .
For weights given by a, you will prove in Exercise 15.2.1 that
a convex the lower hull of C,a is the graph of a
a strictly convex facets of the lower hull map to maximal cones of .
Hence = a , so that T = Q is a regular triangulation of when ,I is a

chamber.
It remains to prove that every regular triangulation of arises in this way.
e ,I ) for
Suppose that T = Q . Then Rr0 implies that Relint(
e
some GKZ cone ,I . The associated support function is strictly convex with
respect to and satisfies
(i ) = i for i
/ I , (i ) > i for i I .
This implies = . However, ,I need not be a chamber, even though is

simplicial. Figure 8 on the next page gives an example where I = {2} and has
maximal cones Cone(1 , 3 ) and Cone(3 , 4 , 5 ). This is easy to fixfor any i
/ I
with Cone(i )
/ (1) (such as i = 4 in Figure 8), increase i so that (i ) > i .
When we do this, stays the same but we now have I = {i | Cone(i )
/ (1)}.
Since is also simplicial, Proposition 14.4.9 implies that ,I is a chamber.
The Secondary Polytope. In the situation of Proposition 15.2.9, the secondary fan
is complete. It is natural to ask if the secondary fan is the normal fan a polytope.
Gel fand, Kapranov and Zelevinsky [112] proved the existence of such a polytope,
called the secondary polytope. Other proofs have been given in [28] and [222]. We
will present a proof from [28] that uses Gale duality.
742
Figure 8. Weights where is simplicial but ,I is not a chamber in Proposition 15.2.9
A -basis J {1, . . . , r} gives the cone Cone( J ) = Cone(i | i J) featured

in Proposition 15.2.1. Given a real number > 0, consider the polytope
bR .
PJ = Conv(i , j | i J, j
/ J) G
We assume that is chosen sufficiently small so that Conv(J ) = Conv(i | i J)

bR , the origin is an interior point of PJ .
is a facet of PJ . Since C = Cone() = G
bR . Our
Hence taking cones over proper faces of PJ gives a complete fan J in G
choice of guarantees that Cone( J ) is a cone of J .
We will use the following notation. Given two complete fans 1 and 2 , the
set of all intersections 1 2 for i i , i = 1, 2, is a fan denoted 1 2 . Thus
1 2 = {1 2 | i i , i = 1, 2}
Any common refinement of 1 and 2 also refines 1 2 .
Lemma 15.2.10. As above, let J be the fan constructed from the -basis J. Then:
(a) The secondary fan GKZ refines J .
V
(b) GKZ = J is a -basis J .
Proof. Extending Definition 14.3.13, we say that C is generic if

/ Cone( )
for all subsets with dim Cone( ) < dim G.

Faces of PJ are of the form Conv(i , j | i A, j
/ B) for suitable sets A and
B. It follows that the cones of J are generated by subsets of . In particular, any
cone of J of dimension < dim G consists entirely of nongeneric elements.
Take any chamber ,I . The interior points of ,I are generic (this follows
easily from the proof of Proposition 14.4.9), so that Int(,I ) is disjoint from
any cone of J of dimension < dim G. This implies that ,I is contained in a
maximal cone of J , and part (a) follows without difficulty.
743
V
For part (b), first note that by part (a), GKZ refines J is a -basis J . Then
V
observe that every maximal cone of GKZ appears in J is a -basis J by Proposition 15.2.1. It follows that the two fans must be equal, which completes the proof
of part (b).

We now construct the secondary polytope.
b whose
Proposition 15.2.11. There is a full dimensional lattice polytope PGKZ G
R
bR is the secondary fan GKZ .
normal fan in G
bR contains the origin as an interior point,

Proof. Let J be a -basis. Since PJ G
b defined in 2.2, and J is the normal fan of

we have the dual polytope PJ G
R
PJ by Exercise 2.3.4. Recall from Proposition 6.2.13 that if polytopes P1 , P2 have

normal fans P1 , P2 , then the Minkowski sum P1 + P2 has normal fan P1 P2 .
Now consider the Minkowski sum
P
bR .
PGKZ = J is a -basis PJ G
V
It follows that PGKZ has normal fan J is a -basis J , which by Lemma 15.2.10 is
the secondary fan.

It is customary to call PGKZ the secondary polytope even though it is far
from unique. A specific choice for PGKZ , based on a different but quite elegant
construction, is given in [113] and [112] (see also [28]).
Since the vertices of a polytope correspond to maximal cones of the normal
fan, Propositions 15.2.9 and 15.2.11 have the following nice corollary.
Corollary 15.2.12. There is a bijective correspondence between vertices of the
b and regular triangulations of .
secondary polytope PGKZ G

R
The secondary polytope PGKZ has dimension dim G. For = (1 , . . . , r ), we

can compute this dimension as follows. First, the convex hull Q = Conv() has
dimension dim Q = dim NR 1 by assumption. Since dim G = r dim NR , we
get the dimension formula
dim PGKZ = || dim Q 1.
Here is a famous example of a secondary polytope.
Example 15.2.13. The associahedron is the secondary polytope for a configuration

where NR = R3 and consists of the vertices of a convex r-gon sitting in the
plane z = 1. This polytope has dimension is r 3 and is sometimes denoted K r3 .
Another name for the associahedron is the Stasheff polytope because of its origins
in algebraic topology.
The number of vertices of K r3 is the Catalan number Cr2 , where

1
2n
Cn =
.
n+1 n
744
Besides counting vertices of the associahedron, the Catalan number Cr2 counts
many things, including:
The number of triangulations of a convex r-gon that introduce no new vertices.
The number of rooted binary trees with r 1 leaves.
The number of ways pairs of parentheses can be introduced in a nonassociative

product x1 xr1 . For r 1 = 4, one gets
(((x1 x2 )x3 )x4 , (x1 (x2 x3 ))x4 , (x1 x2 )(x3 x4 ), x1 ((x2 x3 )x4 ), x1 (x2 (x3 x4 ))).
See Exercise 15.2.2 and [80, Thms. 1.1.2 and 1.1.3].
The associahedron K r3 is easy to work out when r = 4 or 5 (Exercise 15.2.3).
When r = 6, K 3 has C4 = 14 vertices. In Figure 9, we show K3 together with
the triangulation at each vertex. The figure uses Lodays construction [188] of the
associahedron, and the labeling of the vertices in Figure 9 was inspired by [80,
Fig. 1.15]. See Example B.2.3 for more on how to construct this figure.
Figure 9. The associahedron K 3 with triangulations at the vertices
Final comments. There is a lot more to say about the secondary fan. Here are
some highlights:
By Proposition 15.2.11, the secondary fan is the normal fan of the secondary
polytope when lies on an affine hyperplane not passing through the origin in
NR . One can ask more generally if the secondary fan is the normal fan of some
polyhedron for arbitrary . By[29], the answer is yes when is geometric.
The secondary polytope is related to Chow polytopes and the Newton polytope
of the principal A-determinant. See [113] and [112].
745
The secondary fan is also related to the Grobner fan defined in 10.3. See [80,
Sec. 9.4] and [264].
The secondary fan is used in mirror symmetry. See [66] and [68].
Chapter 5 of the book [80] is devoted entirely to regular triangulations and the
secondary fan.
Another important topic is what happens to the toric variety X of a chamber ,I
when we cross a wall in the secondary fan. This will be studied in the next section.
Exercises for 15.2.
15.2.1. Prove the claims about a made in the proof of Proposition 15.2.9.
15.2.2. The goal of this exercise is to relate triangulations, binary trees, and parentheses.
For r 3, fix an r-gon with vertices labeled 1, . . . , r. Also consider symbols x1 , . . . , xr1 ,
which we arrange in a diagram with an empty box before and after each symbol. The boxes
are numbered 1, . . . , r corresponding to the vertices of the r-gon:
1 2 3 r 1 r
(15.2.3)
x1 x2 xr1
A triangulation T of the r-gon is determined by adding certain diagonals. Each diagonal
is oriented so that we go from a smaller vertex to a bigger one. Also, at vertex i, we order
the diagonals containing i according to the label of the other vertex of the diagonal.
(a) Given T , fill in the ith box in the diagram (15.2.3) with open or close parentheses, one
for each diagonal containing i. The parentheses are placed in the same order as the
diagonals containing i, and we use an open parenthesis if the diagonal starts at i and a
close parenthesis if the diagonal ends at i. Prove that this defines a bijection between
triangulations and parenthesis placements in the symbols.
(b) Given a rooted binary tree with leaves x1 , . . . , xr1 , each internal node different from
the root is root of a subtree, and we put parentheses around the rightmost and leftmost leaves this subtree. Prove that this gives a bijection between binary trees and
parenthesis placements in the symbols.
Figure 10 shows an example of parts (a) and (b) of the exercise when r = 6. See also [80,
Thm. 1.1.3].
x1
x2
x3
2
1
x1((x2(x3x4))x5)
triangulation
parentheses
binary tree
Figure 10. The correspondences described in Exercise 15.2.2
15.2.3. Consider the associahedron K r3 from Example 15.2.13.
x4
x5
746
(a) Draw K 1 and K 2 . At each vertex, give the corresponding triangulation of the square
and pentagon, similar to Figure 9.
(b) Consider the top pentagonal facet of K 3 in Figure 9. All triangulations lying in this
facet contain a common diagonal that forms a triangle with two edges of the hexagon.
Use this and part (a) to explain why this facet is a pentagon. Then explain why K 3 has
six such facets.
(c) Pick one of the four-sided facets of Figure 9. Show that all triangulations lying in this
facet contain a common diagonal that bisects the hexagon. Use this to explain why the
facet is a quadrilateral and why there are three such facets.
15.2.4. The secondary fan computed in Exercise 15.1.4 has four chambers. Apply the
method of Example 15.2.2 to each of these chambers and determine the maximal cones of
the corresponding fan.
15.2.5. Consider the cone = Cone(e1 , e1 + de2 ) R2 and let be the smooth refinement
whose ray generators are i = e1 + ie2 for 0 i d. We assume d 2. By Exercise 15.1.5,
the secondary fan has 2d1 chambers corresponding to subsets I {1, . . . , d 1}. This
exercise will study the chambers where I = and I = {1, . . . , d 1}.
(a) The i s are the rows of the (d + 1) 2 matrix
1 0
1 1
B = . .
.. ..
1 d
B
that fits into an exact sequence 0 Z2 Zd+1 Zd1 0. Show that A can be
chosen to be the (d 1) (d + 1) matrix
1 2
1 0 0
0
1 2 1 0
A = .
..
..
..
..
..
.
.
. ..
0
2 1
Then the i s are the columns of A.
(b) Suppose that I = {1, . . . , d 1}. Show that U is the toric variety of this chamber and
use Proposition 15.2.1 to show that the chamber is the simplicial cone generated by
the columns 2, 3, . . . , d 1 of A.
(c) Suppose that I = . Show that X is the toric variety of this chamber. Proposition 15.2.1 represents this chamber as the intersection of d simplicial cones. The next
d1
parts of the exercise will show that this chamber is R0
.
Pd
(d) Show that the chamber is Nef(X ) and show that D = i=0 ai Di is nef if and only if
ai1 2ai + ai+1 0 for i = 1, . . . , d 1. Hint: Use Example 6.4.7.
Pd
d1
. Hint: [D] = i=0 ai i , where i is the ith column of A.
(e) Show Nef(X ) = R0
This exercise was inspired by [64, App. A], where the toric variety of the secondary fan
has a nice moduli interpretation.
15.2.6. The definition of regular triangulation involved choosing a weight vector in Rr0 .
15.3. Crossing a Wall
747
(a) Define what it means for a weight vector to be generic. Hint: The weights used in
Figure 7 are generic while those in Figure 8 are not.
(b) Explain how the definition given in part (a) relates to definition of generic given in the
proof of Lemma 15.2.10.
15.2.7. Let = (1 , . . . , 6 ) be the rows of the matrix
0 0 0 1
1 0 0 1
0 1 0 1
.
B=
0 0 1 1
1 0 1 1
0 1 1 1
(a) Show that Q = Conv() is a 3-dimensional prism in the hyperplane H R4 where

the last coordinate is 1. The secondary polytope has dimension r dim Q 1 = 2.
(b) Show that the secondary polytope is a hexagon and each triangulation has 3 simplices.
See [228, Fig. 11] for a splendid picture of the secondary polytope with the corresponding
triangulation of Q at each vertex.

The goal of this section is to understand what happens when we cross a wall in the
secondary fan. We begin with some examples.
Examples. In the general case when has elements that vanish or are repeated, it
is possible that very little happens when crossing a wall. Here is a simple example.
bR R2 , and 1 =
Example 15.3.1. In Example 15.1.5 we have 3 = 1 + 2 in G
2 = e1 , 3 = e1 in NR R. Figure 11 shows the secondary fan together with
P =
F2
F1
F3
F1=F2 F3
P =
2
P =
F1
F2
F3
1
Figure 11. A boring wall crossing
P and its virtual facets F1 , F2 , F3 for three choices of . These s were chosen to
illustrate that very little happens when crossing the wall between the two chambers:
748
the GIT quotient is P1 on either side of the wall and is also P1 on the wall itself.
The only thing that changes is which virtual facet is empty.
What allows Example 15.3.1 to be so boring is the equality 1 = 2 . For the

rest of this section, we will assume that is geometric, so that the i are nonzero
and generate distinct rays in NR . We will see that this assumption rules out the
behavior illustrated in Example 15.3.1.
Here are some more interesting wall crossings.
Example 15.3.2. Figure 6 from Example 15.2.7 shows a secondary fan with five
chambers. Here, consists of five vectors in R3 lying in the plane z = 1. Each
triangulation in Figure 6 is the intersection of a fan in R3 with z = 1.
We will study how these fans change as we go clockwise around the origin,
starting with the fan 0 in the large chamber to the left in in Figure 6. Figure 12
4
4
5
3
2
Figure 12. The fans to the left and right of the wall generated by 5
shows that as we cross the vertical wall generated by 5 , 0 changes to 1 , which

is the star division of 0 at the point 3 , as indicated by the shading.
Continuing our loop, we cross the wall generated by 1 . This takes us from 1
to the fan 2 shown in Figure 13. Here, the only change is that the shaded wall
in 1 flips to become the shaded wall in 2 in Figure 13. Weve seen this type
of flip beforethe large illustration given in Figure 3 of Example 11.1.12 has fans
(also called 1 and 2 ) that are related by the same type of flip.
4
5
Figure 13. The fans to the left and right of the wall generated by 1
749
To finish the clockwise loop, we have three more walls to cross in Figure 6:
Crossing the wall generated by 2 is another flip.
Crossing the wall generated by 4 undoes a star subdivision.
Crossing the wall generated by 3 is a flip.
Example 15.3.3. In the situation of Propositions 15.2.9 and 15.2.11, the secondary
fan is the normal fan of the secondary polytope. Thus a wall between two chambers
of GKZ corresponds to an edge of the secondary polytope PGKZ .
Now look at the associahedron K3 shown in Figure 9 of Example 15.2.13.
Every edge comes from a flip of the triangulations at the vertices of the edge.
Example 15.3.4. Figure 2 of Example 15.1.11 shows a secondary fan with exactly
two chambers. You should check that the corresponding fans are related by a star
subdivision.
With the exception of Example 15.3.1, the wall crossings in these examples all
come from flips or star subdivisions. We will soon see that this is no accident.
For the rest of this section we assume that is geometric.
Facets of GKZ Cones. Before we can understand wall crossings, we need to learn
more about the facets of a GKZ cone. Recall from Proposition 15.1.3 that
I
.
,I Nef(X ) R0
Using this and the face relation described in Theorem 14.4.7, one easily obtains
the following result (Exercise 15.3.1).
Lemma 15.3.5. The facets of ,I have one of two forms:
I
, where refines , or
(a) ,I (a facet of Nef(X )) R0
I \{i}
(b) ,I \{i} Nef(X ) R0
, where i I .
Part (b) of the lemma describes exactly what the facets look like, while part (a)
does not. Later in the section we will use the toric cone theorem to say more the
facets from part (a).
Note also that if ,I is a chamber of the secondary fan, then its facets are
either walls of the secondary fan or lie on the boundary of C . Since is geometric,
Proposition 14.4.12 implies that the latter happens only for facets of the form ,I
where is degenerate. This can only occur in part (a) of Lemma 15.3.5.
Star Subdivisions. There is one type of wall crossing in the secondary fan that is
very easy to describe.
Theorem 15.3.6. Assume that is geometric and let ,I be a chamber of the
secondary fan with I 6= . If i I , then:
750
(a) ,I \{i} is a facet of ,I and is a wall of the secondary fan.

(b) The chamber on the other side of the wall is ,I \{i} , where is the star
subdivision of at the minimal generator of Cone(i ) N.
(c) The exceptional locus of the toric morphism X X has codimension 1.
Proof. ,I \{i} is a facet of ,I by Lemma 15.3.5 and is not contained in the

boundary of C since is nondegenerate. Part (a) follows. Let be the star
subdivision of at the generator of Cone(i ) N, as defined in 11.1. The proof
of Proposition 15.1.6 shows that is simplicial with X semiprojective.
Since ,I is a chamber, j 7 Cone( j ) is a bijection {1, . . . , r} \ I (1) by
Proposition 14.4.9. This extends to a bijection {1, . . . , r} \ (I \ {i}) (1) since
is geometric, and it follows that ,I \{i} is a chamber by Proposition 14.4.9.
Furthermore, ,I \{i} is a face of this chamber by Theorem 14.4.7. Thus
,I ,I \{i} ,I \{i} .
The cone in the middle has codimension 1 (it is a facet of the chamber ,I ) and
hence must be the wall between the two chambers.
We get a birational toric morphism : X X since is a refinement of
. To describe the exceptional locus, let Di X be the divisor corresponding to
Cone(i ) (1) and let be the smallest cone containing i . Then (Di ) =
V () and induces an isomorphism X \ Di X \V () by the definition of star
subdivision. Since codimV () 2, Di is the exceptional locus.

Types of Walls. Theorem 15.3.6 describes the walls of the secondary fan that arise
from star subdivisions, which we call divisorial walls because of part (c) of the
proposition. The remaining walls are the intersection of two GKZ chambers that
satisfy part (a) of Lemma 15.3.5. We will denote such a wall by 0 ,I , and the
chambers on either side will be written
(15.3.1)
,I 0 ,I ,I .
These are the only GKZ cones containing 0 ,I since it is a wall. We call 0 ,I a
flipping wall. This terminology will be justified below. Here is a preliminary result
about flipping walls.
Lemma 15.3.7. If 0 ,I is a flipping wall between 0 ,I and 0 ,I , then:
(a) 0 is not simplicial.
(b) 0 (1) = (1) = (1).
(c) 0 is the coarsest common refinement of and .
(d) The exceptional loci of the toric morphisms X X0 and X X0 have
codimension 2.
751
Proof. We first prove part (b). Note 0 (1) (1) since refines 0 . Suppose
that Cone(i ) (1) \ 0 (1). Then 0 ,I 0 ,I {i} by Theorem 14.4.7. This
is impossible by (15.3.1), and part (b) follows easily.
For part (a), we have {1, . . . , r} \ I (1) by Proposition 14.4.9 since ,I
is a chamber. If 0 were simplicial, then 0 (1) = (1) and Proposition 14.4.9
would imply that 0 ,I is a chamber, which is clearly impossible. Hence 0 is not
simplicial.
Part (c) is straightforward and is left to the reader (Exercise 15.3.2).
Since is a refinement of 0 satsifying 0 (1) = (1), the birational toric
morphism X X0 is an isomorphism in codimension 1 (see (15.1.9)). Hence
the exceptional locus must have codimension 2. The same holds for X X0 ,
and part (d) follows.

Later we will give a careful description of the exceptional loci that occur in
part (d) of Lemma 15.3.7. The terms divisorial wall and flipping wall are
taken from the minimal model program. We will say more about this in 15.4.
The Geometry of Circuits. Our next task is to study a flipping wall given by
(15.3.1). The starting point is that 0 ,I ,I comes from a facet of the nef
cone Nef(X ) by Lemma 15.3.5. This corresponds to an edge of the Mori cone
NE(X ) N1 (X ), which by the toric cone theorem is generated by the class of an
orbit closure V ( ) of a wall = of .
Since is simplicial, we can describe using the notation of Figure 17 from
6.4. Let n = dim NR . After renumbering the i s, we can assume that
= Cone(1 , . . . , n )
(15.3.2)
= Cone(2 , . . . , n+1 )
= Cone(2 , . . . , n ).
The vectors 1 , . . . , n+1 are linearly dependent with a nontrivial linear relation
(15.3.3)
n+1
X
bi i = 0
i=1
that is unique up to multiplication by a nonzero constant. This relation played an

important role in 6.4, where it was called a wall relation (see (6.4.4)).
Here we will consider the more general situation where 1 , . . . , n+1 N Zn
are nonzero and span NR . A result of Reid [236] describes some simplicial fans
associated to 1 , . . . , n+1 . Our treatment is based on [80, Lem. 2.4.2]. Using the
relation (15.3.3), we define the sets
J = {i | bi < 0}, J0 = {i | bi = 0}, J+ = {i | bi > 0}.
The vectors i for i J J+ form a circuit since they are linearly dependent but
every proper subset is linearly independent. Its orientation is determined by J
752
and J+ . Multiplying (15.3.3) by a negative number switches J and J+ and hence

changes the orientation. Then define the following two sets of simplicial cones
(15.3.4)
= { | Cone(1 , . . . , bi , . . . , n+1 ), i J+ }
+ = { | Cone(1 , . . . , bi , . . . , n+1 ), i J }.
Let us give some examples of and + .
Example 15.3.8. The rows of the matrix B from Example 15.2.2 are the points
1 = (2, 1, 1), 2 = (2, 1, 1), 3 = (1, 0, 1), 4 = (2, 1, 1), 5 = (2, 1, 1) in
R3 . Then:
1 , 2 , 3 , 4 satisfy the relation 1 22 + 43 4 = 0, so that
J = {1, 2, 4}, J0 = , J+ = {3}.
Computing and + , we obtain:
has maximal cone Cone(1 , 2 , 4 )
(
Cone(2 , 3 , 4 ), Cone(1 , 3 , 4 ),
+ has maximal cones
Cone(1 , 2 , 3 ).
In Figure 12, is a subfan of 0 while + is a subfan of 1 .
2 , 3 , 4 , 5 satisfy the relation 32 + 43 24 + 5 = 0. Thus
J = {2, 4}, J0 = , J+ = {3, 5}.
Computing + and , we obtain:
has maximal cones Cone(2 , 4 , 5 ), Cone(2 , 3 , 4 )
+ has maximal cones Cone(3 , 4 , 5 ), Cone(2 , 3 , 5 ).
In Figure 13, is a subfan of 1 while + is a subfan of 2 .
Thus the fan changes observed in Figures 12 and 13 can be explained in terms of
replacing with + .
Here is a 4-dimensional example where things are more complicated.

Example 15.3.9. Consider the five points in R4 given by 1 = (0, 1, 0, 1), 2 =
(0, 1, 0, 1), 3 = (1, 0, 1, 1), 4 = (1, 0, 1, 1), 5 = (0, 0, 2, 1). The relation is
31 32 + 23 + 24 + 25 = 0, so
J = {1, 2}, J0 = , J+ = {3, 4, 5}.
These points lie in the hyperplane H R3 where the last coordinate is 1. If we

slice the fans and + with H, we get Figure 14 on the next page. For the fan
, the three maximal cones spin about the common face corresponding to J
(the thick line segment on the left), and for + , the two maximal cones meet along
the common face corresponding to J+ (the shaded triangle on the right).
753
3
1
2
H
3
1
2
+ H
Figure 14. The intersections H and + H
In Figure 14, going from to + is a flip where the three maximal cones
of that meet along Cone(1 , 2 ) are replaced with the two maximal cones of
+ that meet along Cone(3 , 4 , 5 ).
The fans and + defined in (15.3.4) have the following properties.

Lemma 15.3.10. Let 1 , . . . , n+1 N Zn be nonzero, and assume that 0 =
Cone(1 , . . . , n+1 ) is strongly convex of full dimension. Then:
(a) J and J+ are nonempty.
(b) and + are simplicial fans with support 0 such that (1) = + (1) =
{Cone(i ) | 1 i n + 1}. Furthermore, X and X+ are semiprojective.
(c) and + are the only fans with the properties listed in part (b).
Proof. Note that 1 , . . . , n+1 span NR since 0 has full dimension. We will first
Pn+1
compute the Gale dual of 1 , . . . , n+1 . The relation i=1
bi i = 0 gives the exact
sequence
R
R
0 MR
Rn+1
R 0,
where R (m) = (hm, 1 i, . . . , hm, n+1 i) and R (ei ) = bi . Thus the Gale dual is the
vector of coefficients = (b1 , . . . , bn+1 ).
Since Cone(1 , . . . , n+1 ) is strongly convex and the i are nonzero, C = R
by Lemma 14.3.2. Part (a) follows immediately. Also, C = R implies that the
secondary fan has two chambers, hence there are precisely two fans satisfying the
conditions of part (b). Thus part (c) will follow as soon as we prove part (b).
First take the GKZ cone R0 R. We will use Proposition 15.2.1 to determine
the corresponding fan. The -bases are the nonzero bi s, and R0 Cone(bi )
if and only if i J+ . By Proposition 15.2.1, the maximal cones of the fan are
Cone(1 , . . . , bi , . . . , n+1 ) for i J+ (you should check this carefully). This gives
, which proves that is a fan with the required properties. Similarly, applying
Proposition 15.2.1 to R0 shows that + also satisfies part (b).

The assumption in Lemma 15.3.10 that Cone(1 , . . . , n+1 ) is strongly convex
is needed to ensure that we have two fans. You will explore what happens when
strong convexity fails in Exercise 15.3.4.
754
The proof of Lemma 15.3.10 showed that the secondary fan of 1 , . . . , n+1 is
the complete fan in R. Thus the origin is a wall between the two chambers. To
describe the resulting wall crossing, we will use the following notation:
J = Cone(i | i J) for J {1, . . . , n}.
Then the fans and + from (15.3.4) can be written as

(15.3.5)
= {J | J+ 6 J}
+ = {J | J 6 J}
(Exercise 15.3.5). Here is our result.
Lemma 15.3.11. Let 1 , . . . , n+1 N Zn be nonzero, and assume that 0 =

Cone(1 , . . . , n+1 ) is strongly convex of full dimension. Then:
(a) If J = {i}, then the origin in the secondary fan of 1 , . . . , n+1 is a divisorial
wall. Furthermore, 0 is simplicial, + consists of 0 and its faces, and is
the star subdivision of 0 at the minimal generator of Cone(i ) N.
(b) If J+ = {i}, then part (a) holds with the roles of + and reversed.
(c) If J and J+ have at least two elements, then the origin is a flipping wall.
Furthermore, 0 is nonsimplicial, and the refinements and + of 0 give a
commutative diagram of surjective toric morphisms

V (J ) / X
X+ o ? _ V (J+ )
HH
HH
v
+
HH
v
HH
vv
HH
vv
HH
vv
v
v
$
v
z
HH
vv
HH
U0
vv
HH
v
O
HH
vv
HH
vv
# ?
{vv
V (J J+ )
such that is birational with exceptional locus V (J ). In addition,

dim V (J ) = |J0 | + |J+ | 1 = n |J |
dim V (J+ ) = |J0 | + |J | 1 = n |J+ |
dim V (J J+ ) = |J0 |.
Proof. If J = {i}, then i Cone(1 , . . . , bi , . . . , n+1 ), which implies that 0 =

Cone(1 , . . . , bi , . . . , n+1 ), so 0 is simplicial, and (15.3.4) shows that
P 0 is the
unique maximal cone of + . The relation (15.3.3) implies bi i = jJ+ b j j ,
where bi < 0 and b j > 0 for j J+ . Thus J+ is the minimal cone of + containing
i . Using the definition of star subdivison from 11.1, one can check that is the
star subdivison of + at the generator of Cone(i ) N (Exercise 15.3.6). Parts (a)
and (b) follow easily.
For part (c), suppose the origin is a divisorial wall. Then some i is a nonnegative combination of j for j 6= i. This implies J+ = {i} or J = {i}, a contradiction.
Hence the origin is a flipping wall and 0 is nonsimplicial by Lemma 15.3.7.
755
To analyze : U0 , first note that J X by (15.3.5). By the

Orbit-Cone Correspondence, X \V (J ) is the toric variety of the fan
(15.3.6)
\ { | J } = \ {J | J J}
= {J | J 6 J, J+ 6 J},
where the second equality uses (15.3.5). The next step is to describe the faces of
0 . If we set = (1 , . . . , n+1 ), then 0 = C , so that we can use Lemma 14.3.3
to describe its faces. Given J {1, . . . , n + 1}, the lemma implies that there is a
face
/ J such that
P 0 with J = {i | i } if and only if there are a j > 0 for j
jJ
/ a j j = 0. Using the partition {1, . . . , n + 1} = J J0 J+ , you will prove in
Exercise 15.3.6 that 0 has two types of faces:
J for J J+ J (all nonsimplicial)
(15.3.7)
J for J 6 J, J+ 6 J (all simplicial).
The cones in the first line of (15.3.7) are the ones we need to remove to get the fan
for U0 \V (J J+ ), while the cones in the second line are the cones of (15.3.6).
This proves that induces an isomorphism
(15.3.8)
X \V (J ) U0 \V (J J+ ).
Now we compute some dimensions. Since J is simplicial, we have

dim V (J ) = n dim J = n |J | = |J0 | + |J+ | 1
since n + 1 = |J+ | = |J0 | + |J |. The i , i J J+ , form a circuit, so that

dim V (J J+ ) = n dim J J+ = n (|J | + |J+ | 1) = |J0 |.
Then dim V (J ) > dim V (J+ J ) since |J+ | 2. This and (15.3.8) show that
V (J ) is the exceptional locus of . Note also that the exceptional locus has
codimension |J | 2.
We get the same picture for + with the roles of and + reversed. This gives
the diagram in the statement of the lemma and completes the proof of part (c).
The Flipping Theorem. Before stating our main result, we need an example of a
more complicated wall crossing.
Example 15.3.12. Consider the six points in R4 given by 1 = (0, 1, 0, 1), 2 =
(0, 1, 0, 1), 3 = (1, 0, 1, 1), 4 = (0, 0, 1, 1), 5 = (0, 0, 0, 1), 6 = (1, 0, 0, 1), which
lie in the hyperplane H R3 where the last coordinate is 1. Figure 15 on the next
page shows the secondary fan in R2 . There are four chambers, two divisorial walls,
and two flipping walls. In Exercise 15.3.7 you will compute this secondary fan and
the fan of each chamber.
We are most interested in the chambers 1 , and 2 , , which meet along the
flipping wall 0 , in Figure 15. Similar to what we did in Example 15.3.9, we
visualize the fans 1 , 2 , and 0 in R4 by slicing them with the hyperplane H to
get the picture shown in Figure 16 on the next page.
756
4 ,{5}
1 ,
s4 = 6
1 =s 2

5
s
s3

3 ,{5}
0 ,
2 ,
Figure 15. A secondary fan in R2 with two flipping walls
To understand Figure 16, first focus on the left-hand side of the intersections,
which involve 1 , 3 , 4 , 5 , 6 . Using the relation 3 + 4 5 + 6 = 0, we get
J = {3, 5} and J+ = {4, 6}. Then Lemma 15.3.11 explains the left half of the
fans we see in Figure 16.
4
1
1
4
1 H
1
6
2 H
2
6
0 H
Figure 16. The intersections 1 H, 2 H, and 0 H
Now shift your focus to the right-hand side of the intersections in the figure,
which involve 2 , 3 , 4 , 5 , 6 . Lemma 15.3.11 still applies, which explains the
right half of each fan. The key observation is that both sides use the same relation
3 + 4 5 + 6 = 0. In other words, the wall crossing is completely determined
by the single oriented circuit 3 + 4 5 + 6 = 0.
We now return to the general situation of two chambers in the secondary fan
separated by a flipping wall, which by (15.3.1) is written
,I 0 ,I ,I .
Recall that 0 is not simplicial by Lemma 15.3.7.
757
Similar to what we did in Lemma 15.3.11, given = (1 , . . . , r ), we set

J = Cone(i | i J) for J {1, . . . , r} \ I .
Theorem 15.3.13. Assume is geometric and let 0 ,I be a flipping wall between
chambers ,I and ,I of the secondary fan. Then there is an oriented circuit
X
X
(15.3.9)
bi i +
bi i = 0
iJ
iJ+
with J J+ {1, . . . , r} \ I , J J+ = , and bi < 0 for i J , bi > 0 for i J+ .

Furthermore:
(a) J , J+ , and J J+ 0 .
(b) X0 \V (J J+ ) is the simplicial locus of X0 .
(c) The refinements and of 0 give a commutative diagram of surjective toric

morphisms

V (J ) / X F
FF
FF
FF
FF
FF
F
X o ? _ V (J+ )
ww
ww
w
w{
w
ww
ww
#
FF
w
FF
ww
X0
ww
FF
O
w
FF
w
FF
ww
" ? {ww
V (J J+ )
such that and are birational with exceptional loci V (J ) and V (J+ ).
Also, codimV (J ) = |J | 2 and codimV (J+ ) = |J+ | 2.
(d) If 0 (0 )max is nonsimplicial, then there is J0 {1, . . . , r} \ I such that

the disjoint union J J0 J+ has cardinality dim 0 + 1 and satisfies 0 =
J J0 J+ . Furthermore, |0 = and |0 = + , so that restricting the
diagram of part (c) to U0 X0 gives the diagram of Lemma 15.3.11
Proof. As noted in the discussion leading up to (15.3.3), 0 ,I ,I comes
from a facet Nef(X0 ) Nef(X ) by Lemma 15.3.5. This facet is defined by an
extremal ray R of NE(X ) N1 (X ), which by the toric cone theorem is generated
by the class of V ( ) for a wall of .
Let n = dim NR and let 0 0 (n) be nonsimplicial.
Claim 1. If (n 1) is any wall that meets the interior of 0 , then the class
[V ( )] lies in the ray R.
P
To prove this, let D = a D be a Cartier divisor on X whose class lies in
the relative interior of Nef(X0 ) Nef(X ). Note that the support function of D
is linear on 0 . If = is a wall of that meets Int(0 ), then the Cartier
data of D satisfies m = m since and lie in 0 . Hence D V ( ) = 0 by
Proposition 6.3.8. Our hypothesis on D then implies that [V ( )] R.
758
To construct the oriented circuit (15.3.9), let = be a wall of that

meets the interior of 0 . We use the notation of (15.3.2), where the i are renumbered so that is generated by 2 , . . . , n , and (resp. ) is obtained from by
adding 1 (resp. n+1 ). The linear relation satisfied by 1 , . . . , n+1 can be written
(15.3.10)
n+1
X
bi i = 0,
b1 , bn+1 > 0.
i=1
P
Then (15.3.9) is defined to be the relation iJ J+ bi i = 0, where

J = i {1, . . . , n + 1} | bi < 0 , J+ = i {1, . . . , n + 1} | bi > 0 .
Let Di be the divisor corresponding to Cone(i ) (1) for i {1, . . . , r} \ I .

Then we can describe J and J+ in terms of the intersections Di V ( ) as follows:
(15.3.11)
J = {i | Di V ( ) < 0}
J+ = {i | Di V ( ) > 0}.
This follows from the intersection formulas of Proposition 6.4.4 since X is simplicial and b1 , bn+1 > 0 in (15.3.10). Here is an observation of Reid [236].
Claim 2. If i J+ , then i = Cone(1 , . . . , bi , . . . , n+1 ) .
First note that 1, n + 1 J+ since b1 , bn+1 > 0 in (15.3.10). For i = 1, n + 1,

the claim holds since 1 = and n+1 = , both of which lie in . Now take
i 6= 1, n + 1 in J+ . Then = Cone(2 , . . . , bi , . . . , n ) is a facet of and is in the
subfan |0 of consisting of cones of contained in 0 . Let N()R R2 be the
quotient of NR by Span(). By Proposition 3.2.7, V () is the toric variety of
Star() = { N()R | |0 },
where u NR gives u N()R . Since |0 has full dimensional convex support,

the same is true for Star(). Also, b1 1 + bi i + bn+1 n+1 = 0 in N()R . Since
b1 , bi , bn+1 > 0, we conclude that Star() is complete.
Consider the ray of generated by 1 . On one side of 1 we have =
Cone( 1 , i ) Star() as shown in Figure 17 on the next page. By completeness,
there must be |0 such that is on the other side of the wall generated
by 1 . Hence there is 6= i such that 1 r and = Cone( 1 , ). Thus
= + Cone(1 ) + Cone( ) = Cone(1 , 2 , . . . , bi , . . . , n , ) .
Note that = = Cone(1 , 2 , . . . , bi , . . . , n ) is a wall of |0.
If = n + 1, then = i and the claim follows. So suppose > n + 1. Then
(15.3.12)
D V ( ) > 0 by Proposition 6.4.4 since = + Cone( )
D V ( ) = 0 by Proposition 6.4.4 since

/ {1 , . . . , n+1 }.
This is impossible since [V ( )], [V ( )] R by Claim 1, and Claim 2 is proved.

Claim 3. 0 = Cone(1 , . . . , n+1 ) and |0 = .
759
AK
As 1
A
A
si
A

s
s n+1

?
Figure 17. The fan Star() in N()R R2
S
By Lemma 15.3.10, Cone(1 , . . . , n+1 ) = iJ+ i 0 , hence i |0 by
Claim 2. Suppose |0 contains a maximal cone different from the i . Since 0
is convex, there must be a wall = i where |0 (n) is different from
the i . Writing = + Cone( ), we have D V ( ) > 0 by the first line of
(15.3.12). Then the second line of (15.3.12) and Claim 1 imply that is one
of 1 , . . . , n+1 . It follows
S that Cone(1 , . . . , n+1 ). This is impossible since
Cone(1 , . . . , n+1 ) = iJ i and is different from the i . Claim 3 follows.
We are now ready to prove the theorem. First note that 0 is nonsimplicial and
hence has a nonsimplicial maximal cone 0 . By Claim 3 and Lemma 15.3.11, we
have J = |0 and J J+ 0 0 .
Also observe that R is the extremal ray defining Nef(X0 ) Nef(X ) since
the nef cones are on opposite sides of the wall defined by R. Hence, if 0 0 (n) is
nonsimplicial, then walls of that meet the interior of 0 satisfy [V ( )] R,
which is the version of Claim 1. The other claims modify in similar ways; in
particular, the version of Claim 3 states that |0 = + . Arguing as above, we
get J+ . This proves part (a).
For part (b), let be any nonsimplicial cone of 0 . It is contained in a nonsimplicial maximal cone 0 . The above claims imply that Lemma 15.3.11 applies
to 0 . Then (15.3.7) from the proof of the lemma show that J J+ . Thus
0 \ { 0 | J J+ } is the subfan of simplicial cones of 0 . It follows
from the Orbit-Cone Correspondence that X0 \V (J J+ ) is the simplicial locus
of X0 , proving part (b).
For part (d), let 0 0 (n) be nonsimplicial. By Claim 3, we have |0 =
and |0 = + . It follows that over U0 , the diagram of part (c) becomes the
diagram in Lemma 15.3.11. This proves part (d).
Finally, for (c), first observe that and induce isomorphisms
(15.3.13)
X \V (J ) X0 \V (J J+ ) X \V (J+ ).
760
To see why, recall that (1) = 0 (1) = (1) by Lemma 15.3.7. It follows that
any simplicial cone of 0 is a cone of since refines 0 (Exercise 15.3.8). The
same is true for , and the isomorphisms (15.3.13) follow.
Hence, to study and , it suffices to work over U0 for 0 0 (n) nonsimplicial. The key point is that all such 0 use the same relation (15.3.9). Then we
are done by part (c) and Lemma 15.3.11.

The maps and from Theorem 15.3.13 give a birational map
(15.3.14)
1 : X 99K X
called an elementary flip. This birational map is equivariant with respect to the
torus actions on X and X and is an isomorphism in codimension 1.
In 15.4 we will say more about the maps and when we discuss the toric
minimal model program. We will show that they are the extremal contractions
determined by the extremal rays R NE(X ) and R NE(X ).
An Application. Any two toric varieties X and X of dimension n are birational
since they both contain a copy of (C )n . However, if we require the birational map
to be equivariant and an isomorphism in codimension 1, then a much nicer picture
emerges.
Theorem 15.3.14. Let X and X be simplicial semiprojective toric varieties and
let : X 99K X be an equivariant birational map that is an isomorphism in codimension 1. Then is a composition of elementary flips and a toric isomorphism.
Proof. First note that induces an isomorphism : N N since is equivariant
and birational, and ((1)) = (1) since it is an isomorphism is codimension 1.
Changing X by a toric isomorphism, we may assume that is a fan in NR with
(1) = (1) and that is the identity on the torus TN .
Now consider the secondary fan where consists of the minimal generators
u for (1) = (1). The GKZ cones , and , are chambers of GKZ
by Proposition 14.4.9 and lie in the moving cone MovGKZ . This cone is convex by
Proposition 15.1.4, so that we can find a line segment connecting interior points of
, and , which meets all intermediate walls at points of their relative interior.
This gives chambers
, , 1 , , . . . , , , ,
such that consecutive chambers share a common wall. These walls are flipping
walls since I = on each side of the wall. It follows from Theorem 15.3.13 that
we get a composition of elementary flips
: X 99K X1 99K 99K X1 99K X .
Thus is an equivariant birational map that is an isomorphism in codimension 1

and is the identity on TN .
761
The composed birational map = 1 : X 99K X is the identity on TN

and hence is the identity as a birational map. It follows that the birational maps
and are equal.

When is primitive and lies in an affine hyperplane not containing the origin,
Theorem 15.3.6, Lemmas 15.3.10 and 15.3.11 and Theorem 15.3.13 can all be
interpreted in terms of triangulations. This is explained nicely in the book [80],
though one caution is that in [80], the term flip applies to both divisorial walls
and flipping walls.
The material in this section is based on [194, Ch. 14], [222], [236] and [269].
Exercises for 15.3.
15.3.1. Prove Lemma 15.3.5. Hint: The facets of a product cone are easy to describe, as
I
are the facets of R0
.
e be a common refinement of and
15.3.2. Prove part (c) of Lemma 15.3.7. Hint: Let
. Prove that ,I
e is a face of 0 ,I .
15.3.3. Prove part (d) of Proposition 15.3.10.
15.3.4. Let 1 , . . . , n+1 N span NR Rn and assume that 0 = Cone(1 , . . . , n+1 ) is not
strongly convex.
(a) Prove that J+ or J is empty. For the rest of the exercise, assume that J = .
(b) Prove that the secondary fan of 1 , . . . , n+1 consists has only one chamber.
(c) Conclude that 0 = Cone(1 , . . . , n+1 ) has a unique simplicial refinement with
| | = 0 and X semiprojective. Also explain why the maximal cones of are
given by Cone(1 , . . . , bi , . . . , n+1 ) for i J+ .
(d) Prove that J+ is the minimal face of 0 and that the quotient 0 /J+ is simplicial.
(e) Draw a picture of what happens when n = 2, J+ = {1, 3}, and J = .
15.3.5. Prove that (15.3.5) follows from (15.3.4).

15.3.6. This exercise will cover some details needed for the proof of Lemma 15.3.11.
(a) In part (a) of the lemma, show carefully that is the star division of + at the
generator of Cone(i ) N when J = {i}.
(b) Prove the description of the faces of 0 given in (15.3.7).
15.3.7. Compute the secondary fan shown in Figure 15.3.12. Also compute the fans
1 , . . . , 4 mentioned in the figure. Hint: Use Proposition 15.2.1.
15.3.8. Suppose that we have fans 1 and 2 in NR such that 1 (1) = 2 (1) and 1 refines
2 . Prove that every simplicial cone of 2 is contained in 1 .
15.3.9. In the situation of Lemma 15.3.11, prove that the following are equivalent:
(a) Cone(1 , . . . , n+1 ) is simplicial.
(b) i Cone(1 , . . . , bi , . . . , n+1 ) for some i.
(c) J or J+ consists of one element.
762
(d) and + are related by a star subdivision.

(e) (1) 6= + (1).
15.3.10. In the situation of Lemma 15.3.11, prove that the following are equivalent:
(a) Cone(1 , . . . , n+1 ) is not simplicial.
(b) J and J+ have at least two elements.
(c) (1) = + (1).
15.3.11. In the situation of Lemma 15.3.11, assume that J and J+ have at least two
elements.
(a) Prove that J+ is the unique primitive collection for .
(b) Prove that J is the unique primitive collection for + .
Hence the elementary flip X 99K X+ interchanges the primitive collections.
15.4. Extremal Contractions and Flips

The minimal model progam, called the MMP, is a first step toward the birational
classification of projective varieties. In dimension 1, birationally equivalent smooth
projective curves are isomorphic. This fails in dimension 2, where instead smooth
surfaces are successive blowups of smooth minimal surfaces. Minimal surfaces
play a key role in the Enriques-Kodaira classification of smooth projective surfaces.
In dimension 3, two complications arise. First, smooth minimal models need
not exist. Instead, a minimal model is a normal Q-Gorenstein projective variety
with only terminal singularities (Definition 11.4.9) and whose canonical divisor is
nef. The second complication is that constructing the minimal model of a smooth
projective variety may require successive blowdowns of divisors together with flips,
which are a type of birational map not present in dimension 2.
Brief Sketch of the MMP. The goal of the MMP is to show that if X is a Q-factorial
normal projective variety with only terminal singularities, then there is a sequence
of blowdowns of divisors and flips X 99K X 99K 99K Y such that either
Y is a minimal model, or
There is f : Y Z such that dim Z < dim Y and all curves C Y mapping to
a point in Z satisfy KY C < 0. We say that f is a Mori fibration.
To apply the MMP to a variety X as above, we can assume that KX is not nef. Then
one can show the following:
There is a curve C X such that KX C < 0 and C generates an edge of the
Mori cone of X , a so-called extremal ray.
An extremal ray gives an extremal contraction f : X X .
The contraction f is of one of three types: a Mori fibration, a contraction of a

divisor, or a small contraction, where the exceptional locus of f has codimension 2.
763
The type of f dictates the next step in the MMP. When f is a small contraction,
this leads to one of the flips mentioned above. Proving termination of this process
is a very difficult problem in general.
The MMP has been proved in dimension 3 by the work of Kawamata, Kollar,
Mori, Reid, Shokurov, and others. Recently, minimal models have been proved
to exist for varieties of general type of arbitrary dimension [31]. The MMP is an
active area of research in algebraic geometry. See [179] or [194] for an introduction
to the minimal model program.
The Toric MMP. When we apply the MMP to toric varieties and toric morphisms,
we get the toric minimal model progam. In one sense, this is uninteresting, since
n
a projective
P toric variety X is birationally equivalent to P and its canonical class
KX = D can never be made nef, so the toric MMP will always end with a
Mori fibration. Nevertheless, there are good reasons to study the toric MMP:
Many MMP constructions are easy to describe in the toric setting. For instance,
extremal contractions and flips are part of the rich geometry of wall crossings.
There are wonderful toric examples of the individual steps in the MMP.
When we switch from the MMP to the relative or log MMP in 15.5, we will
see that there are nontrivial toric examples of these versions of the MMP.
The goal of this section is to introduce the toric versions of extremal contractions and flips. Then, in 15.5, we will discuss the toric MMP.
Extremal Contractions. We begin with the following result that shows how an
extremal ray in the Mori cone gives an interesting toric morphism.
Proposition 15.4.1. Let be a simplicial fan in NR such that X is semiprojective
and let R NE(X ) be an extremal ray. Then there is a generalized fan 0 in NR
(a) refines 0 and X0 is semiprojective.
(b) The toric morphism : X X0 has the property that for every wall of ,
(V ( )) = {pt} [V ( )] R.
Proof. Consider the secondary fan when consists of the minimal generators u
for (1). Then , is chamber isomorphic to Nef(X ) by Proposition 15.1.3,
so that R defines a facet of , . By Proposition 14.4.6 this facet equals 0 ,
where the generalized fan 0 satisfies part (a) of the proposition.
Let be the support function of a Cartier divisor D on X whose class lies
in the relative interior of 0 , . Then is strictly convex with respect to 0 . If
{m }(n) , n = dim NR , is the Cartier data of D, then m = m if and only if
, lie in the same cone of 0 . This follows from the proof of Lemma 14.4.6.
764
Now suppose that = is a wall of . Then

(V ( )) = {pt} meets the interior of some 0 0 (n)
m = m
D V ( ) = 0
[V ( )] R.
The first equivalence is easy and the second follows from the previous paragraph.
The third equivalence follows from Proposition 6.3.8 plus standard arguments (see
the proof of Proposition 6.3.15), and the fourth follows since D lies in the relative
interior of the facet of Nef(X ) defined by R. Thus 0 satisfies part (b).

When is complete, one can give an elementary proof of Proposition 15.4.1
using Proposition 6.2.5 and Theorem 6.2.8 (Exercise 15.4.1).
The morphism : X X0 constructed in Proposition 15.4.1 is called an
extremal contraction. It is of one of three types:
Fibering: dim X0 < dim X .
Divisorial: is birational and its exceptional locus is a divisor.
Flipping: is birational and its exceptional locus codimension 2.
This follows from Theorem 15.3.6 and Lemma 15.3.7.
Structure of Extremal Contractions. Our next task is to describe the contraction

: X X0 from Proposition 15.4.1. Since is simplicial, the discussion of
(15.1.4) shows that elements of N1 (X ) come from relations among the u .
Let R NE(X ) be an extremal ray as in Proposition 15.4.1 and suppose that
R is generated by a relation
P
b u = 0.
We will construct a very simple toric variety XR that contains a lot of information
about the extremal contraction . The construction of XR will use only (1) and
P
b u = 0. The other cones of will play no role.
We begin by defining the sets
(15.4.1)
J = JR, = { (1) | b < 0}
J+ = JR,+ = { (1) | b > 0},
and for a subset J (1), let

J = Cone(u | J).
Then we have the sublattices
NJ = Span(J ) N NJ J+ = Span(J J+ ) N N,
and we set NR = NJ J+ /NJ . The map NJ J+ NR will be denoted u 7 u.
765
Lemma 15.4.2. R = { J | J ( J+ } is a complete simplicial fan in (NR )R whose

toric variety XR has the following properties:
(a) XR is Q-factorial and Q-Fano.
(b) XR has dimension |J+ | 1 and Picard number 1.
(c) XR is a finite abelian quotient of a weighted projective space.
Furthermore, we always have |J+ | 2.

Remark 15.4.3.
(a) A complete normal variety X is Q-Fano if KX is Q-Cartier and some positive integer multiple of KX is Cartier and ample. In the literature, Q-Fano
varieties with Picard number 1 are sometimes called unipolar [63].
(b) In the literature, the toric varieties of part (b) are often called fake weighted
projective spaces [60], [167].
Proof. The toric cone theorem tells us that has a wall = such that [V ( )]
generates R. This means that R is generated by the wall relation (15.1.4), which
we write as
X
b u = 0, b = D V ( ).
Then |J+ | 2 since the rays (1) and (1) opposite the common facet
( ) > 0 by Proposition 6.4.4. Also note that the u
satisfy D V ( ) > 0 and D VP
for J+ span NR and satisfy J+ b u = 0. Furthermore, dim NR = |J+ | 1
since the u , J J+ , form a circuit.
As in the proof of Lemma 15.3.10, the Gale transform of u , J+ , consists of
the positive numbers b > 0 for J+ . It follows that R0 is the unique chamber
of the secondary fan. Each b is a -basis in the sense of Proposition 15.2.1, so that
J+ \{} is a maximal cone of the fan corresponding to the chamber. Thus R is a
fan and XR is projective. Furthermore, since Pic(XR )R R, any effective divisor
(e.g., KR ) has a positive integer multiple that is ample. Hence XR is Q-Fano.
This proves parts (a) and (b), and then (c) follows from Exercise 5.1.13.
Example 15.4.4. Suppose that is a fan in R4 with minimal generators u1 = e1 ,

u2 = e2 , u3 = e3 , u4 = e1 + e2 + e3 + 2e4 , u5 = e2 + e3 2e4 and maximal cones
Cone(u2 , u3 , u4 , u5 ), Cone(u1 , u3 , u4 , u5 ), and Cone(u1 , u2 , u4 , u5 ). There are three
walls, all with the same relation
u1 + 2u2 + 2u3 u4 u5 = 0.
This relation generates an extremal wall R in NE(X ) where J+ consists of the rays
generated by u1 , u2 , u3 . Then NR Z2 and u1 , u2 , u3 NR generate a sublattice of
index 2 and satisfy u1 + 2u2 + 2u3 = 0 (Exercise 15.4.2). It follows that
XR P(1, 2, 2)/(Z/2Z).
766
Given an extremal ray R, we have the extremal contraction

: X X0
from Proposition 15.4.1. The behavior of is determined by J and J+ from

(15.4.1) and XR from Proposition 15.4.2 as follows.
Proposition 15.4.5. Let : X X0 come from the extremal ray R with J , J+
and XR as above, and recall that |J+ | 2 by Lemma 15.4.2. Then:
(a) is fibering 0 is a degenerate fan J = . Here, X0 is simplicial, and

the fibers of are isomorphic to XR .
(b) is divisorial 0 is a simplicial fan |J | = 1. Here, the exceptional

locus V (J ) is a divisor, and the fibers of |V (J ) are isomorphic to XR .
(c) is flipping 0 is a nonsimplicial fan |J | > 1. Here, the exceptional

locus V (J ) has codimension |J |, and the fibers of |V (J ) are isomorphic
to XR .
Proof. We begin with part (c). Pick a wall of so that [V ( )] generates R, which
defines a flipping wall of Nef(X ). Then R comes from the relation (15.1.4) for
V ( ), and J and J+ are defined in (15.4.1). On the other hand, Theorem 15.3.13
also uses sets denoted J and J+ , which are the same as here by (15.3.11).
This gives the equivalences of part (c). Also, Theorem 15.3.13 implies that
(15.4.2)
X \V (J ) X0 \V (J J+ )
and that the exceptional locus V (J ) has codimension |J+ |. It follows that |V (J )
is the map
(15.4.3)
= |V (J
: V (J ) V (J J+ ).
In the notation of Proposition 3.2.7, the fan Star(J ) of V (J ) lives in N(J )R =

(N/NJ )R , and the fan Star(J J+ ) of V (J J+ ) lives in N(J J+ )R , defined
similarly. Using the snake lemma, one easily obtains the exact sequence
0 NR N(J ) N(J J+ ) 0.
We will show that the Star(J ) is weakly split by R and Star(J J+ ) in the
sense of Exercise 3.3.7. This will imply that the fibers of are isomorphic to XR .
The proof of Theorem 15.3.13 shows that for J (1),
J J J+ 6 J and J J 0 (1) for some 0 0 .
When we project these into N(J )R , the elements in J map to zero. Thus
Star(J ) = { K | J+ 6 K and K 0 (1) \ J for some 0 0 }.
It follows that Star(J ) has two subfans:
767
The cones with K J+ give R = { K | K ( J+ }.

b = { K | K 0 (1) \ (J J+ ), 0 0 }.
The cones with K J+ = give
It is easy to see that every cone of Star(J ) can be written uniquely as a sum
b In Exercise 15.4.3 you will show that
K1 + K2 where K1 R and K2 .
b
R maps K bijectively to R ( K ) Star(J J+ ) such that K 7 R ( K )
b
defines a bijection
Star(J J+ ). This gives the desired weak splitting, and
part (c) follows.
For part (b), the equivalences follow easily. If J = {}, then the results of
15.3 imply that
/ 0 (1) and that is the star subdivision of 0 at u . A relation
P
b
u
=
0
generating
R can be chosen so that b = 1, hence
X
u =
b u , b > 0.
J+
Thus J+ is the minimal cone of 0 containing u , and J J+ = J+ . It follows

that as in part (c), induces an isomorphism (15.4.2) and |J is (15.4.3). Then
the analysis of part (c) shows that the fibers of |J are isomorphic to XR .
Finally, for part (a), suppose that J = . Then R is generated by a relation

X
(15.4.4)
b u = 0, b > 0.
J+
Let 0 be a maximal cone of the degenerate fan 0 . Then there is at least one wall
of whose interior meets 0 . The wall relation of must be (15.4.4). Then
Claims 1 and 2 from the proof of Theorem 15.3.13 remain true in this situation, as
does Claim 3, provided we interpret as the fan constructed in Exercise 15.3.4.
By Exercise 15.3.4, J+ is the minimal face of 0 . Since 0 was an arbitrary
maximal cone of 0 , it follows that J+ is the minimal cone of the degenerate fan
0 . Exercise 15.3.4 also implies that 0 /J+ is simplicial, so that the genuine fan
0 constructed from 0 is simplicial. Hence X0 = X0 is simplicial.
Since J = and J+ is the minimal cone of 0 , we have V (J J+ ) =
V (J+ ) = X0 , and V (J ) = X follows by a similar argument. Hence (15.4.3) is
the map : X X0 , and then the analysis of part (c) shows that the fibers of
are isomorphic to XR .

Example 15.4.6. Consider the complete fan in R3 pictured in Figure 18 on the
next page. The minimal generators of the rays 0 , . . . , 4 are
u0 = (0, 0, 1), u1 = (0, 1, 1), u2 = (1, 0, 1), u3 = (0, 1, 1), u4 = (1, 0, 1).
The wall = Cone(u2 , u4 ) gives an extremal ray R = R0 [V ( )], whose extremal
contraction is the toric morphism X P1 induced by projection onto the y-axis
in Figure 18. The fibers of are isomorphic to XR = P(1, 1, 2). In Exercise 15.4.4
768
3
2
y
x
0
Figure 18. The fan from Example 15.4.6
you will prove these claims and explain why Figure 19 from Example 6.4.12 is the
secondary fan of this example. See also Example 15.4.12 below.
Birational Transforms. Before discussing flips, we need to study how divisors

move between birationally equivalent varieties. Suppose that f : X 99K X is a
birational map between normal varieties and let U X is the largest open subset
of X where f is defined. A prime divisor D X gives a divisor on X defined by
(
f (D U ) if codim f (D U ) = 1
f D =
0
otherwise.
P
Then the birational transform by f of a Weil divisor D = i ai Di on X is the Weil
divisor on X defined by
P
f D = i ai f Di .
The birational transform of a Q-Weil divisor on X is defined similarly.

P
Example 15.4.7. Let D = a D be an effective torus-invariant divisor on a
simplicial semiprojective toric variety X . By Theorem 15.1.10, the u for
(1) give a secondary fan where Nef(X ) is the chamber , and the class of D
lies in the support C of the secondary fan. Hence there is a GKZ chamber with
[D] ,I . Under the isomorphism
,I Nef(X ) RI
from Proposition 15.1.3, [D] projects to the class [D ] of a nef divisor D on X .
Since X and X have the same torus TN , we get a birational map : X 99K X .
In Exercise 15.4.5 you will show that D is the birational transform of D by .
769
This example shows that an effective divisor D on X becomes nef after a

suitable birational transform. The toric MMP discussed in 15.5 will explain how
to do this using only elementary flips and divisorial extremal contractions.
One has to be careful with birational transforms since they are sometimes not
well-behaved. For example, Cartier divisors need not be preserved.
Example 15.4.8. Consider the toric varieties X and X+ from Example 15.3.9.
The fans and + are pictured in Figure 14, and Lemma 15.3.11 shows that we
have an elementary flip X 99K X+. In Exercise 15.4.6 you will show that
Cl(X ) = Cl(X+ ) Z (Z/2Z),
where
P
P
[ 5i=1 ai Di ] = [ 5i=1 ai Di ] 7 (3a1 + 3a2 2a3 2a4 2a5 , a3 + a4 mod 2).
Furthermore, you will show that the Picard groups do not have the same image in
Z (Z/2Z):
Pic(X ) = Z[2D5 ], [2D5 ] 7 (4, 0 mod 2)
Pic(X+ ) = Z[D1 ], [D1 ] 7 (3, 0 mod 2).
So 2D5 is Cartier on X but its birational transform 2D5 is not Cartier on X+.
Another problem with birational transforms is that they are not functorial.
Example 15.4.9. The blowup of a smooth point p of a variety Y of dimension n > 1
gives a birational morphism f : X Y whose exceptional locus E is a divisor. Then
f 1 : Y 99K X is also birational. However,
f1 f E = f1 0 = 0, yet ( f 1 f ) E = E,
since the largest open set where f 1 f is defined is all of X .
Things are much nicer if we work with special types of birational maps.
Lemma 15.4.10. Let f : X 99K X be a birational map between normal varieties
that is either an isomorphism in codimension 1 or a proper morphism defined on
all of X . Then:
(a) If D1 D2 on X , then f D1 f D2 on X .
(b) If g : X 99K X is birational of one of the same two types, then g f = (g f ) .

Proof. Part (a) is easy if f is an isomorphism in codimension 1 (Exercise 15.4.7).
So assume that f : X X is proper and birational. We claim that the birational
transform is the same as the push-forward defined in [107, Sec. 1.4]. Once we
prove this, then part (a) for f will follow from [107, Thm. 1.4].
770
If D X is a prime divisor, then f (D) is closed in X . The push-forward of D is

(
[C(D) : C( f (D))] f (D) if codim f (D) = 1
(15.4.5)
f D =
0
otherwise,
where [C(D) : C( f (D))] is the degree of the field extension C( f (D)) C(D).
However, f 1 : X 99K X is defined outside of a set of codimension 2. This
is proved in [131, Lem. V.5.1] when X , X are projective, but the same proof works
when f : X X is proper and birational. In particular, f 1 is defined on an open
subset of f (D), so that f |D : D f (D) is birational. Hence the degree is 1, which
shows that (15.4.5) agrees with the birational transform defined above.
Part (b) is straightforward and is left to the reader (Exercise 15.4.7).
Note that Lemma 15.4.10 applies to elementary flips and divisorial extremal
contractions. This will be important in our discussion of the toric MMP.
If : X 99K X is an isomorphism in codimension 1, the birational transform
gives an isomorphism Div(X ) Div(X ) that preserves linear equivalence and
hence induces an isomorphism Cl(X ) Cl(X ). It follows that if X and X are
simplicial semiprojective toric varieties and : X 99K X is equivariant and an
isomorphism in codimension 1, then induces isomorphisms
Pic(X )R Pic(X )R and N1 (X ) N1 (X ).
The left isomorphism was used implicitly in 15.1 and 15.3. Also note that a ray
R in N1 (X ) gives a ray in N1 (X ), which will be denoted by R.
Existence of Toric Flips. A flipping contraction is a birational extremal contractions whose exceptional locus has codimension 2. An important part of the MMP
is proving the existence of flips for flipping contractions. Our previous results make
this easy to do in the toric setting.
Theorem 15.4.11. Let X be a simplicial semiprojective toric variety and let R
be an extremal ray that gives a flipping contraction : X X0 . Then there is a
commutative diagram of toric maps
X @_ _ _ _ _/ X
@@
@@

X0
}
}}
~}}

(a) The birational map is an elementary flip in the sense of (15.3.14) and in
particular is equivariant and an isomorphism in codimension 1.
(b) X is simplicial and semiprojective.
(c) R = R is an extremal ray for X .
(d) is a flipping extremal contraction for R .
771
Proof. As in the proof of Proposition 15.4.1, gives a chamber , Nef(X )

in the secondary fan. The extremal ray R defines a flipping wall 0 , , which has
a chamber , Nef(X ) on the other side. Then the diagram of the theorem
follows from Theorem 15.3.13, and the proof of the theorem shows that is the

extremal contraction of X for the extremal ray R = R.
In the situation of Theorem 15.4.11, we say that : X X0 is the flip of
: X X0 . Here is an example.
Example 15.4.12. Consider the toric variety X from Example 15.4.6. The left
side of Figure 19 shows the secondary fan, whose chambers are the nef cones of
the toric varieties X and X . Here, i = [Di ], and is obtained from by
replacing Cone(u2 , u4 ) with Cone(u1 , u3 ) in Figure 18 of Example 15.4.6. The
right side of Figure 19 shows the corresponding Mori cones.
6Nef(X ) s
0
Nef(X )
2 = 4 s
1 =s 3
@ NE(X ) 6
I
@
R = R @
@
@
@
@
@ NE(X )
@
@
R@
R
@
Figure 19. Extremal rays R for X and R = R for X
The ray generated by 0 on the left side of Figure 19 is a flipping wall between
X and X . This flipping wall is defined by the extremal rays R for X and
R = R for X , as pictured on the right side of the figure.
Exercises for 15.4.

15.4.1. Here you will give an alternative proof of Proposition 15.4.1.
(a) Assume is complete and let R be an extremal ray. Pick a divisor class [D] in the
relative interior of the facet of Nef(X ) corresponding to R. Use Proposition 6.2.5
and Theorem 6.2.8 to give an elementary proof of Proposition 15.4.1 in this case.
(b) Show that Proposition 6.2.5 and Theorem 6.2.8 apply when has full dimensional
convex support. This gives an elementary proof of Proposition 15.4.1 in general.
(c) Explain how Proposition 15.4.1 relates to to Examples 14.2.11 and 14.2.14.
Theorem 4.1 of [104] gives a yet different approach to Proposition 15.4.1 that avoids fans.
15.4.2. Verify the details of Example 15.4.4.
772
b bijectively to
15.4.3. Consider part (c) of Proposition 15.4.5. Prove that R maps K
b Star(J J+ ).
R ( K ) Star(J J+ ) such that K 7 R ( K ) gives a bijection
15.4.4. In Example 15.4.6, show that XR P(1, 1, 2) and explain how the secondary fan
relates to Figure 19 of Example 6.4.12.
15.4.5. Prove the claim made in Example 15.4.7 that D = D.
15.4.6. Verify the details of Example 15.4.8.
15.4.7. In part (b) of Exercise 15.4.10, there are four cases to consider. When f and g are
both proper birational morphisms, g f = (g f ) follows from [107, Sec. 1.4]. Prove the
remaining three cases.
15.4.8. Let X be a simplicial semiprojective toric variety. Recall that P (1) is a
primitive collection if P is not contained in (1) for all but any proper subset is.
(a) Show that the definition of primitive relation given in Definition 6.4.10 applies to X .
This gives an element r(P) N1 (X ) as explained in (6.4.8)
(b) Prove that r(P) NE(X ). Hint: Use (6.4.9).
(c) Let R be a flipping extremal ray. Use Lemma 15.3.11 and Theorem 15.3.13 to show
that the rays generated by the ui for i J+ form a primitive collection whose primitive
relation generates R.
(d) Prove a similar result for fibering and divisorial extremal rays. Hint: See the proof of
Proposition 15.4.5.
P
(e) Conclude that NE(X ) = P R0 r(P), where the sum is over all primitive collections
P of .
This shows that Theorem 6.4.11 holds for simplicial semiprojective toric varieties.
15.5. The Toric Minimal Model Program

Our discussion of the toric MMP is based on the papers [104] and [236] and the
books [179] and [194]. As mentioned at the beginning of 15.4, the MMP starts
with a Q-factorial variety X with only terminal singularities and tries to make KX
nef using divisorial extremal contractions and flips. There are also relative, log,
and equivariant versions of the MMP described in [179, Sec. 2.2]. We will see that
all of these are relevant to the toric case.
Before giving the toric MMP, we first consider the extremal contractions and
flips constructed in 15.4 from the point of view of the MMP.
Toric Flips via Proj. We will use the following terminology:
Given a ray R in N1 (X ) and a Q-Cartier divisor D, we write D R to mean
D C for any generator [C] R. This is well-defined up to a positive constant.
A ray R is D-negative if D R < 0
A -ample Cartier divisor was defined in Definition 7.2.5. Then a Q-Cartier

divisor D is -ample if a positive integer multiple is a -ample Cartier divisor.
Here is a different way to think about a toric flip.
773
Proposition 15.5.1. Let : X X0 be the flip of : X X0 for the extremal

ray R of flipping type as in Theorem 15.4.11. If D is a divisor on X with birational
transform D on X , then
D is -ample D R > 0 D is -ample.
Furthermore, when D is -ample, we have isomorphisms

L
X Proj
=0 OX (D)

L
X Proj
=0 OX (D) .
Proof. By Theorem 7.2.11, a Cartier divisor on X is -ample if and only if for

every maximal cone 0 0 , its support function is strictly convex on the fan
|0 = { | 0 }. The proof generalizes to Q-Cartier divisors and hence
applies to D since is simplicial.
Assume D R > 0 and take 0 (0 )max . If 0 , then the support function
D is clearly strictly convex on |0 . When 0
/ , let = be a wall of
that meets the interior of 0 . Then maps V ( ) to a point, so [V ( )] R. Hence
D V ( ) > 0.
However, picking u N as in Proposition 6.3.8 implies
D V ( ) = hm m , ui
where m , m are the Cartier data of D for , . Since u , we obtain
D (u) = hm , ui < hm , ui.
Hence D is strictly convex on |0 by Lemma 6.1.13, proving that D is -ample.
The other direction is straightforward, so D is -ample if and only if D R > 0.
Since is the contraction associated to the extremal ray R , the equivalence

just proved implies that D R > 0 if and only if D is -ample. Since R =
R, the latter is equivalent to D R > 0. This gives the desired equivalences.
Now assume that
you will give a toric
LDis -ample on X . In Exercise 15.5.1
proof of X Proj( =0 OX (D)). Furthermore, D is -ample on X , so

L
X Proj
=0 OX (D ) .
However, : X X0 and : X X0 are isomorphisms in codimension 1.

This means that the divisors D on X and D on X have the same birational
transform D0 on X0 . Then it is straightforward to show that
OX (D) OX0 (D0 ) OX (D )
L
(Exercise 15.5.1), and X Proj(
=0 OX (D)) follows easily.
(15.5.1)
774
The Canonical Divisor and Terminal Singularities. Suppose we apply the MMP
to a toric variety X . Following strategy outlined at the beginning of 15.4, we
start with a KX -negative extremal ray R (i.e., KX R < 0). If R is of flipping
type, then applying Proposition 15.5.1 with D = KX gives a nice description of
the resulting flip.
Proposition 15.5.2. Suppose that : X X0 is a flipping contraction for the
extremal ray R satisfying KX R < 0. Then the flip of X is given by

L

X Proj
=0 OX (KX ) .
In the MMP, the flip of a flipping extremal contraction f : X Y satisfying

KX R < 0 is constructed via ProjL
as in Proposition 15.5.2. However, in order to
apply Proj, one has to know that
=0 OX (KX ) is finitely generated. This is
hard in general [31] but easy in the toric context (Exercise 15.5.1). Fujino and Sato
[104, Thm. 4.5] use the latter to give a quick proof of the existence of toric flips.
Another feature of the MMP is that it applies to Q-factorial varieties with only
terminal singularities. A key step is showing that divisorial extremal contractions
and flips preserves these singularities. In the toric case, we prove this as follows.
Proposition 15.5.3. Let : X X0 be the extremal contraction for the extremal
ray R. If R is KX -negative and X has only terminal singularities, then:
(a) If is divisorial, then X0 has only terminal singularities and KX = KX0.
(b) If is flipping with flip : X 99K X , then X has only terminal singularities
and KX = KX .
Proof. We studied terminal singularities in 11.4. Given a simplicial cone NR ,
recall from Proposition 11.4.12 that U has only terminal singularities if and only
if the only is terminal, meaning that lattice points of = Conv(0, u | (1))
are its vertices.
Now take a wall = of such that [V ( )] generates R and set n =
dim NR . Similar to (15.3.2), we label the minimal generators of and so that
= Cone(u1 , . . . , un )
= Cone(u2 , . . . , un+1 )
= Cone(u2 , . . . , un ).
Pn+1
bi ui = 0 with
Then [V ( )] is represented by the wall relation of the form i=1
b1 , bn+1 > 0. Define J and J+ as usual. Also let 0 = Cone(u1 , . . . , un+1 ) and
i = Cone(u1 , . . . , ubi , . . . , un+1 ) for 1 i n + 1.
P
Since KX = D , it follows from (15.1.2) and (15.1.3) that
Pn+1
(15.5.2)
KX V ( ) = i=1
bi
Pn+1
(Exercise 15.5.2). Thus KX R < 0 is equivalent to i=1
bi > 0.
775
Now suppose that is a flipping contraction. By Theorem 15.3.13, |0 has

maximal cones i for i J+ , and |0 has maximal cones i for i J . Since X
has only terminal singularities, i is terminal when i J+ . We need to prove that
the same is true for i when i J .
Pn+1
We begin with two consequences of i=1
bi > 0. First,
S
(15.5.3)
Conv(0, u1 , . . . , un+1 ) = iJ+ Pi .
Pn+1
Pn+1
For the nontrivial inclusion, suppose u = i=1
i ui with i 0 and i=1
i 1.
Pn+1
Pn+1
Set = miniJ+ i /bi . Since i=1 bi > 0 and u = i=1 (i bi )ui , one sees
easily that u Pi for any i J+ with = i /bi . This proves (15.5.3).
Pn+1
The second consequence of i=1
bi > 0 is
ui
/ Pi for i J .
(15.5.4)
P
To prove (15.5.4), assume otherwise, so that ui = j6=i j u j with j 0 and
P
P
j6=i j 1. This implies ui +
j6=i j u j = 0, where the coefficient of ui is
negative and the sum of the coefficients is 0. Since i J , this relation is a
Pn+1
Pn+1
positive multiple of i=1
bi ui = 0. Thus i=1
bi 0, a contradiction.
Using (15.5.3), we obtain
Conv(0, u1 , . . . , un+1 ) N =
=
iJ+ Pi
bi , . . . , un+1 }
iJ+ {0, u1 , . . . , u
= {0, u1 , . . . , un+1 }
since each i is terminal. This and (15.5.4) easily imply that i is terminal for
i J . This shows that X has only terminal singularities. We leave the proof of
KX = KX to the reader.
Finally, suppose that is a divisorial contraction, so that is a star subdivision
of 0 at a primitive element we will call ui . Suppose that ui 0 0 (n) and let the
minimal generators of 0 be u1 , . . . , ubi , . . . , un+1 , where things are labeled so that
P
P
i 6= 1, n + 1. Then ui = j6=i j u j , b j 0, so that we get a relation n+1
j=1 b j u j = 0
with J = {i}. Then one checks easily that (15.5.3) and (15.5.4) still hold and
that |0 = in the notation of Theorem 15.3.13. The argument of the previous
paragraph shows that 0 is terminal. It is also easy to see that KX = KX0 .
Example 15.5.4. Consider the smooth complete fan in R2 shown in Figure 20
on the next page. The walls 1 , . . . , 5 give curves Ci = V (i ) with Mori cone
NE(X ) = Cone([C2 ], [C3 ], [C4 ]) N1 (X ) R3
(Exercise 15.5.3). The resulting extremal rays R2 , R3 , R4 give three divisorial
extremal contractions, as you can see by removing 2 or 3 or 4 . Also,
KX C2 = 0, KX C3 = KX C4 = 1
776
4
6

3
2
1
-
5
Figure 20. A fan with three extremal contractions
(Exercise 15.5.3). By Proposition 15.5.3, the contractions for R3 , R4 map to

smooth varieties since terminal means smooth in dimension 2 (Theorem 11.4.14).
However, the extremal contraction for R2 maps to a singular variety. It still has
canonical singularities since KX C2 = 0, as you will prove in Exercise 15.5.4.
The Toric MMP. For a toric variety, it may be impossible to make the canonical
divisor nef. Hence we will modify the MMP for the toric case by replacing KX
with an arbitrary divisor D. The goal is to make D nef by using elementary flips
and divisorial contractions. We do this as follows.
Procedure 15.5.5. Let X be simplicial and semiprojective, and let D be a Weil
divisor on X . Then do the following steps:
(a) If D is nef, then stop.
(b) If D is not nef, then by the toric cone theorem, there is an extremal ray R with
D R < 0. Let : X X0 be the corresponding extremal contraction.
(c) If is a fibering contraction, then stop.
(d) If is a divisorial contraction, replace X and D with X0 and the birational

transform D. Note that X0 is simplicial and semiprojective with support
|0 | = ||, and as we saw in the proof of Lemma 15.4.10, D in the pushforward of D in the sense of [107, Sec. 1.4]. Return to step (a) and continue.
(e) If is a flipping contraction, then we have the flip
X B_ _ _ _ _/ X
BB
BB
{
{{
}{{
X0 .
Note that X is simplicial and semiprojective with | | = ||. Replace X and

D with X and the birational transform D. Return to step (a) and continue.
777
As we run this procedure on X , the toric varieties that appear all have the
same convex support, namely ||. If U is the affine toric variety of ||, then there
is a projective morphism X U . The other toric varieties involved also map
projectively to U , and the elementary flips and extremal contractions that occur
commute with the projective morphisms to U . Hence:
Procedure 15.5.5 is an example of the relative MMP, since everything is projective over the base U .
Procedure 15.5.5 is also an example of the equivariant MMP, since the torus
TN acts on all toric varieties that occur and all maps are TN -equivariant.
The procedure terminates with a composition : X 99K X 99K 99K X of
D-negative divisorial extremal contractions and elementary flips such that either
There is a D-negative fibering contraction from X to a toric variety of smaller
dimension, or
D is nef on X .
To explain what we mean by D-negative, let Xi 99K Xi+1 be an elementary flip

that arises from running Procedure 15.5.5 on X and D. Then the previous steps
of the MMP give a birational map i : X 99K Xi , and the flip Xi 99K Xi+1 is
associated to a flipping extremal ray of Xi that is negative with respect to the
birational transform i D. We say that Xi 99K Xi+1 is a D-negative elementary
flip in this situation, and the terms D-negative divisorial contraction and D-negative
fibering contraction have similar meanings. Note that we are making frequent use
of the functoriality proved in Lemma 15.4.10.
Termination of Flips. The big question is whether Procedure 15.5.5 terminates.
The secondary fan makes it easy to choose the extremal rays so that this happens.
Proposition 15.5.6. Let D be a Weil divisor on a simplicial semiprojective toric
variety X . Then the D-negative extremal rays in Procedure 15.5.5 for X and D
can be chosen so that the procedure stops after finitely many iterations.
Proof. We use induction on the rank of Pic(X ). The base case is Exercise 15.5.5.
Now assume that Pic(X ) has rank > 1. Consider the secondary fan where
consists of the u for (1). Then , = Nef(X ) is a chamber in the secondary
bR Pic(X )R . Assume [D]
fan in G
/ Nef(X ) and draw a line segment between
[D] and a generic point in the interior of Nef(X ) such that the segment always
crosses from one chamber to another at a relative interior point of a wall.
Consider the facet 0 , where the line segment leaves , . This gives an
extremal ray R such that D R < 0, and : X X0 is the extremal contraction
of R. There are four possibilities to consider:
If 0 , lies on the boundary the secondary fan, then 0 is degenerate by
Proposition 14.4.12, in which case we are done.
778
If 0 , is divisorial wall, then we replace X and D with X0 and D. Since

the rank of the Picard drops by one, we are done by induction.
If 0 , is flipping wall, then we have the flip : X X . If D is nef on
X , then we are done by Procedure 15.5.5.
The remaining case is when 0 , is flipping wall and D is not nef on X .
Then our chosen line segment leaves , = Nef(X ) at a facet defined by an
extremal ray R of X with ( D) R < 0. Then replace X , D, R with X ,
D, R and continue, using the same line segment as before.
The fourth bullet can occur only finitely many times since the line segment meets
only finitely many chambers. Hence the process must terminate.

We will next improve Proposition 15.5.6 by showing that Procedure 15.5.5
terminates no matter which D-negative extremal rays are used in step (b) of the
procedure. Since a divisorial extremal contraction lowers the rank of the Picard
group by 1, only finitely many of these steps can occur. Hence the problem reduces
to the termination of flips. We will need the following lemma.
Lemma 15.5.7.PLet : X 99K X be an elementary flip for an extremal ray R.

Pick a relation b u = 0 generating R and scale it so that
P
P
(15.5.5)
u0 = J+ b u = J b u
is a primitive element of N. Then:
(a) We have an equality of star subdivisions (u0 ) = 0 (u0 ) = (u0 ).

(b) Let be the fan of part (a). Then we have a commutative diagram
X
D
DD
{
DD
{{
{
!
}{
_
_
_
_
_/ X
X B
BB
{
BB
{{
}{{
X0 .
(c) Let D = D be the birational transform of a divisor D on X . Then

D = D (D C)D0 ,
where D0 is the toric divisor on X corresponding

P to u0 and C is a 1-cycle on
X whose class is represented by the relation b u = 0.
Proof. Part (a) is straightforward, and part (b) follows

P immediately from part (a)
(Exercise 15.5.6). For part (c), suppose that D = a D . The fan refines ,
which gives the proper birational map : X X . Since X is simplicial, D is
Q-Cartier and hence has a support function D . By Proposition 6.2.7, D and D
have the same support function D .
779
This makes D easy to compute. First note that (1) = (1) {0 }, where
0 = Cone(u0 ). Let D (resp. D0 ) be the divisor on X associated to (1)
(resp. 0 ). Then
P
P
D = (1) D (u )D D (u0 )D0 = (1) a D D (u0 )D0 ,
where the second equality uses D (u ) = a for (1). Theorem 15.3.13
implies that J P
= Cone(u | J ) . Since D is linear on J , (15.5.5)
yields D (u0 ) = J a b . Hence

P
P
D = (1) a D
J a b D0 .
Furthermore, J+ by Theorem 15.3.13, so repeating the above computation
with D on X gives

P
P
D = (1) a D +
J+ a b D0 .
We conclude that
D = D
(1) a b
D0 .
If C is a 1-cycle whose class is represented by the relation

quantity in parentheses is D C by Exercise 15.5.2.
b u
= 0, then the
Example 15.5.8. Figure 3 from Example 11.1.12 is a classic example of part (b)
of Lemma 15.5.7. In 11.1 we mentioned that Figure 3 was an example of a flip.
But the flip only concerns the bottom half of the figurethe common subdivision
is what explains the top half.
Here is an immediate consequence of Lemma 15.5.7.

Corollary 15.5.9. Assume that D R < 0 in the situation of Lemma 15.5.7. Then
there is a nonzero effective divisor D on X such that
D = D + D .
Corollary 15.5.9 is a special case of a result that applies more generally. See,
for example, [104, Lem. 4.10], [179, Lem. 3.38], or [194, Lem. 9-1-3].
We can now prove termination.
Theorem 15.5.10 (Termination of Toric Flips). Given a divisor D1 on a simplicial
semiprojective toric variety X1 , there is no infinite sequence of elementary flips
1
X1 99K X2 99K X3 99K

for flipping extremal rays Ri NE(Xi ) satisfying Di Ri < 0, where for i 2,
Di = (i1 1 ) D1 .
780
Proof. Assume that such an infinite sequence exists. Since 1 (1) = 2 (1) =
and there are only finitely many fans with a given set of rays, the same fan must
occur twice in the sequence. If we start numbering at this fan, then we obtain
1
X1 99K X2 99K 99K X = X1 .

Furthermore, 1 1 is the identity since it is the identity on TN X1 .
We will show that this is impossible by proving the existence of a commutative

diagram
Xb
i
>>
b
>> i

>

_
_
_
_
_
/ Xi
X1
bi
(15.5.6)
i1 1
such that
(15.5.7)
b Di + D
b D1 =
bi,
i
i
b i nonzero and effective.

D
Once this is proved, we get an easy contradiction. This is because when i = , the
b =
b since 1 1 is the identity. Then
diagram (15.5.6) implies that
b D1 + D
b D1 =
b D + D
b =
b
b = 0, contradicting (15.5.7).
since D = D1 . This forces D
We prove (15.5.6) and (15.5.7) by induction on i. The base case i = 2 follows

from Corollary 15.5.9. For the inductive step, we will do i = 3 for simplicity.
Consider the commutative diagram
b2
Xb 2

??
??
?
1
@@ 1 2 ~ 2 AA
@@
~~
~~ AAAA
~
@
2

~~
~~ 1
X1 _ _ _ _ _/ X2 _ _ _ _ _/ X3 ,
b 2 is any common
where the two lower triangles come from Lemma 15.5.7 and
refinement of 1 and 2 . Then Corollary 15.5.7 implies that
1 D1 = 1 D2 + D1 and 2 D2 = 1 D3 + D2 ,
where D1 and D2 are nonzero effective divisors. An easy diagram chase yields
b 2 D1 =
b 2 D3 + D1 + D2 .
Then we are done since the pullback by surjective map of a nonzero effective divisor is nonzero and effective.
781
In the general version of the MMP, termination of flips has been proved only
in dimension 3. However, a special case of termination is proved in [31], which
is sufficient to prove the existence of minimal models of n-dimensional projective
varieties of general type.
The Log Toric MMP. An importantP
technical tool in the MMP is provided by pairs
(X , D), where X is normal and D = i ai Di is a Q-divisor such that ai [0, 1] Q.
These log varieties were introduced in 11.4, where we gave the references [179,
p. 98] and [194, Ch. 11] that explain their usefulness.
We also defined log canonical and klt (kawamata log terminal) singularities of
a pair (X , D) in Definition
P 11.4.23, which are important in the MMP. For a toric
variety X , we use D = a D . When KX + D is Q-Cartier, Proposition 11.4.24
tells us that:
If a [0, 1] for all (1), then (X , D) is log canonical.
If a [0, 1) for all (1), then (X , D) is klt.
The standard MMP for X uses the divisor KX and requires that X be Q-factorial
with only terminal singularities. Hence it should not be surprising that the log
MMP for the pair (X , D) uses the divisor KX + D and requires that X be Q-factorial
with only klt singularities. In order to run the log MMP, one needs to show that
these properties are preserved by the divisorial contractions and flips associated to
(KX + D)-negative extremal rays.
In
Pthe toric case, let X be simplicial and semiprojective, and take a divisor
D = a D with a [0, 1) Q for all . Then KX + D is Q-Cartier and (X , D)
is klt by Proposition 11.4.24. Then the log toric MMP for the pair (X , D) is
described as follows.
Procedure 15.5.11. Let (X , D) be as above. Then do the following steps:
(a) If KX + D is nef, then stop.
(b) If KX + D is not nef, then there is an extremal ray R with (KX + D) R < 0.
Let : X X0 be the corresponding extremal contraction.
(c) If is fibering, then stop.
(d) If is divisorial, then replace (X , D) with (X0 , D). Return to step (a) and
continue.
(e) If is flipping with flip : X 99K X , then replace (X , D) with (X , D).
Return to step (a) and continue.
The log toric MMP for (X , D) is the toric MMP for X and KX + D since
(KX + D) = KX0 + D in step (d)
(KX + D) = KX + D in step (e)
782
by Proposition 15.5.3. Also note that klt singularities are preserved when we run
the log toric MMP since
(X0 , D) in step (d)
(X , D) in step (e)
are klt by Proposition 11.4.24. Finally, Proposition 15.5.6 (careful choice of R in
step (b)) and Theorem 15.5.10 (any (KX + D)-negative R in step (b)) guarantee
that the log toric MMP terminates.
Example 15.5.12. Suppose that R is any extremal ray for X . Let us show that
the associated extremal contraction : X X0 occurs in Procedure 15.5.11 for
the pair (X , D), provided that D is chosen correctly.
Pick a Cartier divisor D0 whose class lies in the interior of Nef(X ). Replacing
D0 with a positive integer multiple, we may assume that the polyhedron PD0 has
m
an interior
P point m, and then replacing D0 with D0 + div( ), we may assume that
D0 = a D , where a > 0 for all . Now consider the Q-divisor
X
D = KX D0 =
(1 a )D ,
where Q is positive and satisfies 0 < a < 1 for all . Then KX + D = D0 ,

which makes it easy to see that every extremal ray R is (KX + D)-negative.
Log Minimal Models and Flops. When we run the log toric MMP on (X , D), the
result is a sequence : X 99K X 99K 99K X of (KX + D)-negative divisorial
extremal contractions and elementary flips such that either
There is a (KX + D)-negative fibering extremal contraction from X to a toric
variety of smaller dimension, or
KX + D is nef on X, where D = D.
In the latter case, observe that (X, D ) is klt and KX + D is nef. We say that
(X, D ) is a log minimal model of (X , D).
A basic result of the MMP states that log minimal models are isomorphic
in codimension 1 [179, Thm. 3.52]. In our situation, two log minimal models
(X, D ) and (X, D ) of (X , D) satisfy
| | = | | = ||, (1) (1), (1) (1).
Since X and X are isomorphic in codimension 1, it follows that
(15.5.8)
(1) = (1).
This will enable us to show that the map connecting X and X is built from a
special type of flip called a flop. Here is the definition.
Definition 15.5.13. Let R be a flipping extremal ray with flip : X 99K X , and
let D be a Q-divisior on X . Then is a D-flop if D R = 0.
783
Here is a classic example.

Example 15.5.14. Figure 21 on the next page is Figure 3 of Example 11.1.12. For
now, focus on the bottom half of the figure, which involves the cone and two
smooth refinements 1 , 2 that give toric morphisms
X1 _ _ _ _/ X2
DDD
D!
U .
z
}zzz
The birational map is a flip (see, for instance, Figure 13 from Example 15.3.2).
One easily computes that KX1 0. Hence is a KX1-flop.
We can now describe the log minimal models of a toric pair (X , D).
Theorem
15.5.15. Assume that X is simplicial and semiprojective, and let D =
P
a D , where a [0, 1) Q for all . Then:
(a) Up to toric isomorphism, (X , D) has only finitely many log minimal models.
(b) Any two log minimal models of (X , D) are related by a finite sequence of
(KX + D)-flops.
Proof. Any two log minimal models share the same rays by (15.5.8), so finiteness
is immediate.
To relate the various log minimal models of (X , D), we start by fixing one,
say (X, D ). Consider the secondary fan where consists of u for (1).
Then the GKZ cone , = Nef(X ) is a chamber of the secondary fan. Now
let (X, D ) be a second log minimal model. Then (15.5.8) implies that (1) =
(1), so that , is also a chamber, which (via the birational transform) is
isomorphic to Nef(X ). Under this identification, we see that , and ,
both contain the class [KX + D ].
Hence the log minimal models of (X , D) correspond to certain chambers of
the secondary fan that contain [KX + D ]. Suppose for the moment that , and
, are chambers that contain [KX + D ] and share a common wall, regardless
of whether or not they are actual log minimal models. If R is an extremal ray
for X that defines this wall, then (KX + D ) R = 0 since KX + D is in both
chambers and hence is contained in the wall where they intersect. The resulting
flip X 99K X is a (KX + D )-flop. This is what we mean by (KX + D)-flop
in part (b) of the theorem.
Since any two chambers containing [KX + D ] can be connected by a chain
of such wall crossings, we see that any two log minimal models can be connected
by a finite composition of (KX + D)-flops.

Here is an example of the log toric MMP.
784
u3
u2
u4
u0
3
u1
x
Figure 21. The cone with smooth refinements 1 , 2 , 3 in Examples 15.5.14 and 15.5.16
785
Example 15.5.16. In Figure 21 from Example 15.5.14, the toric variety X3 on top
has five minimal generators
u0 = e1 + e2 + e3 , u1 = e1 , u2 = e2 , u3 = e2 + e3 , u4 = e1 + e3 .
We also have the four walls i = Cone(u0 , ui ) for 1 i 4. The walls 1 and 3
share the wall relation u2 u0 + u4 = 0, so that [V (1 )] = [V (3 )] in N1 (X3 ). One
similarly sees that [V (2 )] = [V (4 )]. This gives extremal rays
R1 = R0 [V (1 )] = R0 [V (3 )] and R2 = R0 [V (2 )] = R0 [V (4 )].
The intersection formulas from 6.4 make it easy to compute that
KX3 R1 = KX3 R2 = 1.
Notice also that X3 is smooth.
The log toric MMP for (X3 , 0) is the same as the toric MMP for X3 and KX3.
We have two ways to begin, since both extremal rays are KX3-negative. They are
also divisorial, and the resulting contractions give the commutative diagram
X3
(15.5.9)
EEE
{
E"
}{{{
_
_
_
_
/ X ,
X1
2
where the fans 1 , 2 are from Figure 21. You should check that X3 X1 is
contraction for R2 and X3 X2 is contraction for R1 .
We saw in Example 15.5.14 that KX1 0, and similarly KX2 0. It follows
that X1 and X2 are both minimal models of X3 (we drop the log in this case
since we are doing the MMP for the canonical divisor). In particular, X3 does not
have a unique minimal model. Also, Example 15.5.14 shows that the birational
map in (15.5.9) is the flop connecting the two minimal models.
Figure 21 tells the full story of this example. You can also see the two minimal
models of X3 by computing the secondary fan of X1.
Note that the cone at the bottom of Figure 21 gives the affine toric variety
V(xy zw) C4 from Example 1.1.5. It is amazing that this simple toric variety
is still relevant 750 pages after we first encountered it. This captures perfectly the
power of toric geometry to illustrate deep phenomena in algebraic geometry.
Exercises for 15.5.
15.5.1. This exercise will supply some details omitted in the proof of Proposition 15.5.1
L
(a) Prove X Proj(
=0 OX (D)) when : X X0 is -ample. Hint: You can
assume that X0 is affine. Use Proposition 7.2.3 and Theorems 7.2.4 and 7.1.13.
(b) Prove (15.5.1). Hint: and are isomorphisms in codimension 1.

L
(c) Let D be any divisor on a toric variety X . Show that =0 H 0 (X , OX (D)) is
a finitely generated C-algebra. Hint: Interpret the ring in terms of the graded ring
C[C(PD ) (M Z)].
786
15.5.2. Use (15.1.2) and (15.1.3) to prove (15.5.2).

15.5.3. This exercise concerns the fan from Example 15.5.4.
(a) Show that the Mori cone is generated by the classes of the cones C2 , C3 and C4 .
(b) Verify the intersection products computed in Example 15.5.4.
(c) Show that running the toric MMP with KX always leads to a fibering contraction.
Most of the fibering contractions map to P1 , though there is one that maps to a point.
15.5.4. Here we explore the relation between divisorial and flipping extremal contractions
and the canonical class.
(a) Assume that X has only terminal singularities and let : X X0 be a divisorial
extremal contraction for the extremal ray R. Show that X0 has only canonical singularities if and only if KX R 0.
(b) Let : X X0 be a flipping extremal contraction for the extremal ray R. Prove that
KX0 is Q-Cartier if and only if KX R = 0.
(c) As in part (b), let : X X0 be a flipping extremal contraction for the extremal
ray R, but now assume that KX R < 0. Prove that X0 is neither Q-factorial nor
Q-Gorenstein, i.e., 0 is not simplicial and KX0 is not Q-Cartier.
Part (c) shows that if R is a KX -negative extremal ray, then the associated contraction
maps to the badly behaved variety X0. This explains why flips are essential in the MMP.
15.5.5. Prove that the toric MMP always terminates when Pic(X ) has rank 1.
15.5.6. Prove parts (a) and (b) of Lemma 15.5.7.
15.5.7. Explain how Example 15.5.12 relates to Example 11.4.26.
15.5.8. Suppose that D is an effective divisor on a simplicial semiprojective toric variety
X . Show that applying Procedure 15.5.5 to X and D will never result in a D-negative
fibering contraction.
Appendix A
The History of
Toric Varieties
This appendix discusses the origins of toric geometry, with brief remarks about
recent developments. We will describe how the subject evolved and what its early
successes were. We make no claim as to completeness and omit many topics.
A.1. The First Ten Years

Specific examples of toric varieties, such as Cn , P n , P n P m , and the Hirzebruch
surfaces Hr , have been known for a long time. The idea of a general toric variety
is more recent.
Demazure. The first formal definition of toric variety came in 1970 in Demazures
paper Sous-groupes algebriques de rang maximum du groupe de Cremona [82].
The Cremona group is the (very large) group of birational automorphisms of P n .
Demazures main result is that automorphism groups of smooth toric varieties give
interesting algebraic subgroups of the Cremona group. For him, toric varieties are
certain Z-schemes with a cellular decomposition obtained by
adding certain points at infinity to a split torus.
Demazure worked with schemes over Spec(Z), and if M is a lattice, then the split
torus Spec(Z[M]) is a group scheme over Spec(Z). This makes M the character
group of the torus. So the notation for M was there from the beginning.
Here is Demazures definition of fan [82, Def. 4.2.1].
Definition A.1.1. Let M be a free abelian group of finite type. One calls a fan (in
French, e ventail) a finite set of subsets of M such that
787
788
(a) Every element of is a subset of a basis of M .

(b) Every subset of an element of belongs to .
(c) If K, L , one has NK NL = N(K L).
This looks odd until you realize that Demazure is only considering the smooth
case. His K is the set of minimal generators of a smooth cone . Also, M is the
dual of M, which is N in our notation. Here are some further definitions from [82]:
S
The support of is || = K K.
S
= K NK.
is complete if = M .
If || is the support of a fan in the modern sense, then || N is what Demazure

calls . Thus his notion of complete is equivalent to ours.
Given a fan as in Definition A.1.1, Demazure constructs its toric variety X
as a scheme over Spec(Z) by gluing together affine varieties, similar to what we
did in Chapter 3. Many basic results about smooth toric varieties appear in [82].
Here is a sample, with pointers to where the results appear in this book:
For a field k, Xk = X Spec(Z) Spec(k) is complete if and only if is complete
(Theorem 3.4.6).
The exact sequence M Z(1) Pic(X ) 0 (Theorem 4.2.1).
If X is smooth and complete, then every ample line bundle on X is very ample
(Theorem 6.1.15).
e i1 (VD,m , C) when m M (Theorem 9.1.3).
H i (X , OX (D))m H
When X is complete, Demazure applies the last bullet to prove that H p (X , OX ) = 0

for pP
> 0, the first toric vanishing theorem. He also gives criteria for a divisor
D = a D with Cartier data {m }max to be basepoint free or ample. Support
functions are not mentioned, so that, for example, his criterion for ampleness is
D is ample hm , u i > a when (1),
/ (1).
For us, this follows from Lemma 6.1.13 and Theorem 6.1.14.
Given Demazures focus on the Cremona group, it is not surprising that one of
the major results of [82] is an explicit description of the automorphism group of a
smooth complete toric variety (generalized to the simplicial case in [65]).
After Demazure. Besides [82], other papers touched on some of the same ideas,
though often in very different contexts. An example is Satakes 1973 Bulletin
article On the arithmetic of tube domains [243]. It did not take long for people to
flesh out the wonderful ideas implicit in these papers. Two particularly important
early works are the Springer lecture notes Toroidal Embeddings I by Kempf, Knudsen, Mumford and Saint-Donat, published in 1973 [172], and the paper Almost
homogeneous algebraic varieties under algebraic torus action by Miyake and Oda,
presented at a conference in 1973 and published in 1975 [205].
789
The spirit of the times is captured nicely in Mumfords introduction to [172],

which begins
The goal of these notes is to formalize and illustrate the power of
a technique which has cropped up independently in the work of
at least a dozen people, . . .
The aptness of the phrase cropped up independently is confirmed by a footnote
that Mumford added at the end of the introduction:
*) After this was written, I received a paper by K. Miyake and
T. Oda entitled Almost homogeneous algebraic varieties under
algebraic torus action also on this topic.
The introduction also shows that the elementary aspects of toric geometry were
already recognized:
When teaching algebraic geometry and illustrating simple singularities, varieties, and morphisms, one almost invariably tends to
choose examples of a monomial type type: i.e., varieties defined by equations
a
r+1
xnan
x1a1 xrar = xr+1
and morphisms f for which

f (yi ) = x1ai1 xnain .
In [172] we find the first appearance of M and N for the dual lattices used in
toric geometry. Here is their definition of fan.
Definition A.1.2. A finite rational partial polyhedral decomposition (we abbreviate this to f.r.p.p decomposition) of NR is a finite set { } of convex rational
polyhedral cones in NR such that:
(a) if is a face of , then = for some
(b) , , is a face of and .
This accidentally omits the hypothesis of strong convexity, which is expressed
elsewhere in [172] by saying that does not contain any linear subspace. Given
the unwieldy phrase f.r.p.p decomposition, we should be grateful to Demazure
for the elegant word fan now in use.
A key result of [172] and [205] is that if a normal variety is toric in the sense
of Definition 3.1.1, then it is the toric variety of a fan. This is our Corollary 3.1.8,
which follows from Sumihiros theorem (Theorem 3.1.7), just as in [172] and
[205]. We also find the Orbit-Cone Correspondence (Theorem 3.2.6) and the usual
criteria for completeness (Theorem 3.4.6) and smoothness (Theorem 3.1.19) in
[172] and [205].
790
The properness criterion from Theorem 3.4.11 is proved in [172]. Kempf et

al. also treat convexity in the general setting of sheaves of torus-invariant complete
fractional ideals. The support functions defined in Chapters 4 and 6 appear in part
III.c of Theorem 9 on [172, pp. 2829]. A major result is Theorem 13 on [172,
pp. 48], which gives the criterion of Theorem 7.2.12 for a toric morphism to be
projective. Our proof of this theorem uses results from EGA [127] described in the
appendix to Chapter 7. These results are not mentioned in [172] since the authors,
like many algebraic geometers of the time, knew much of EGA by heart.
Another important result in [172] is that any toric variety has a resolution of
singularities given by a projective toric morphism, our Theorem 11.1.9.
The material on toric varieties appears in the first chapter of [172]. The second
chapter introduces the more general toroidal varieties, which locally look like
toric varieties in a suitable sense. Toroidal varieties are used in the main result
of the book, the characteristic 0 semi-stable reduction theorem for surjective morphisms f : X C, where C is a smooth curve. More recently, toroidal varieties
have been used to study the structure of birational morphismssee [1] and [279].
Turning to [205], we find the first statement of the classification of smooth
complete toric surfaces (Theorem 10.4.3), along with preliminary classification results for toric threefolds. Miyake and Oda also give an example of a 3-dimensional
smooth complete nonprojective toric variety X that is simpler than the complicated
one given by Demazure. The fan for their example is Figure 9 in Example 6.1.17.
The nonprojective toric threefold X reappears at the end of [205], where the
authors use a series of smooth blowups followed by a series of smooth blowdowns
to convert X into P 3 . They also conjecture that this can be done for any smooth
complete toric threefold. This conjecture is still open, as is its generalization called
the strong Oda conjecture, which applies to birational toric maps between smooth
complete toric varieties of arbitrary dimension. However, if we allow the blowups
and blowdowns to be intermixed, then one gets the weak Oda conjecture, which
has been proved in [2] and [278].
We should mention two other notable papers from this period:
Hochsters 1971 paper Rings of invariants of tori, Cohen-Macaulay rings generated by monomials, and polytopes [146] proved that the semigroup algebra
of a saturated affine semigroup is Cohen-Macaulay. This implies that normal
toric varieties are Cohen-Macaulay (Theorem 9.2.9). Hochsters paper initiated
the important interaction between toric geometry and commutative algebra.
Ehlers 1975 paper Eine Klasse komplexer Mannigfaltigkeiten und die Auflosung einiger isolierter Singularitaten [88] considers toric varieties as complex
manifolds. Our proof of the compactness criterion in Theorem 3.4.1 uses his
argument. Ehlers used infinite fans to resolve some interesting singularities.
Similar infinite fans were also used to construct smooth toroidal compactifications
of bounded symmetric domains. See the books [6] from 1975 and [213] from 1980.
791
The Russian School. In the mid 1970s, Bernstein, Khovanskii and Kusnirenko
were studying subvarieties of (C )n defined by the vanishing of Laurent polynomials fi . The Newton polytopes of the fi play an important role their work. For
example, the number of solutions of a generic system f1 = = fn = 0 is given
by the mixed volume of the Newton polytopes of the fi . This is described in [70,
7.5] and [105, Sec. 5.5].
Drawing on [88], [146] and especially [172], Khovanskii studies toric varieties
in his 1977 paper Newton polyhedra and toroidal varieties. This paper is notable
for several reasons:
It introduced the term support function in the toric context.
It proved the Demazure vanishing theorem H p (X , OX (D)) = 0 for p > 0 when

D is basepoint free (Theorem 9.2.3).
It gives the first toric proof of the properties of the Ehrhart polynomial, similar
to what we did in Theorem 9.4.2.
Khovanskiis paper also makes it clear that there is a deep connection between toric
varieties and polytopes. We will soon see another important aspect of this paper.
Danilovs wonderful 1978 survey paper Geometry of toric varieties [76] is a
high point of the era. He drew on all of the references mentioned so far, with the
exception of [205], which was not known in Russia at the time. Danilov covers
an amazing amount of material in [76]. In addition to the basic facts about toric
varieties already mentioned, we also find:
The inverse limit formula (4.2.5) for Cartier divisors.
b n OX ( P D ) from Theorem 8.2.3.
The isomorphism X =
b p ) given in Proposition 8.2.18.

The formula for (U ,
U
The Demazure vanishing theorem (Theorem 9.2.3).

The toric proof of Serre duality (Exercise 9.2.12).
The Bott-Steenbrink-Danilov vanishing theorem (Theorem 9.3.1).
The fundamental group of a toric variety (Theorem 12.1.10).
The Chow and cohomology rings of a smooth toric variety (Theorem 12.5.3).
Riemann-Roch and lattice points in polytopes (Chapter 13).
Danilovs paper remains to this day one of the best introductions to toric geometry.
Toric Varieties. The name toric variety was not used until 1977. In earlier works,
we find a variety of names, such as:
In [82], Demazure says the scheme defined by the fan .
In [172], Kempf et al. say torus embedding.
In [205], Miyake and Oda say almost homogeneous algebraic variety under
torus action.
792
The 1977 article of Khovanskii mentioned earlier originally appeared in Russian as

Mnogogranniki Ntona i toriqeskie mnogoobrazi. This was translated as Newton polyhedra and toroidal varieties, but toroidal is not the right
word for toriqeskie (toricheskie), because the toroidal varieties defined in [172]
are slightly different from toric varieties. Something different was needed.
The same problem occurred with Danilovs 1978 survey, whose Russian title
is Geometri toriqeskih mnogoobrazi
i. One translation was Geometry
of toral varieties, but fortunately for us, toral did not stick. When Miles Reid
translated the paper into English for the Russian Math Surveys, he chose the title
Geometry of toric varieties. This is the origin of the name toric variety.
It took a while before toric variety became standard. For many years, the
term torus embedding introduced in [172] was more common, especially since
it was used in Odas wonderful books [217] from 1978 and [218] from 1988. The
switch to toric variety was confirmed by Odas use of the term in title of his
survey papers from 1991 [219] and 1994 [221].
Polytopes and Normal Fans. Polytopes have been objects of mathematical interest
for over 2000 years. The study of lattice points is more recent, dating from the 19th
century. For example, in 1844 Eisenstein gave a nice proof of quadratic reciprocity
that involved the interior lattice points in the triangle Conv(0, 2p e1 , 2p e1 + 2q e2 ) for
distinct odd primes p and q.
In toric geometry, the most interesting object associated to a lattice polytope
is its normal fan. For many years, normal fans were only implicit in the polytope
literature. Here are some examples:
Proposition 6.2.13 relates refinements of normal fans and Minkowski sums.
This result is implicit in the 1963 paper Decomposable convex polyhedra by
Shephard [249].
Given a polytope P, the cones Cv = Cone(P v) for v a vertex of P are studied
in Exercise 3.4.9 of Grunbaums classic 1967 book Convex Polytopes [128].
The duals of these cones are the maximal cones of the normal fan of P.
Proposition 6.2.18 relates normal fans of zonotopes to central hyperplane arrangements. This is implicit in McMullens 1971 paper On zonotopes [201].
In the toric literature, normal fans are implicit in Khovanskiis 1977 paper [173]
and are described clearly for the first time in Odas 1988 book [218], though the
term normal fan did not appear in print until the 1990s (see, for example, [28]).
Toric varieties came to the attention of the polytope community when people
started using them to prove theorems about polytopes, such as Khovanskiis 1977
toric proof of Ehrhart reciprocity [173]. Shortly thereafter, in 1979, Teissier [268]
and Khovanskii (unpublished) used toric varieties and the Hodge index theorem to
prove the Alexandrov-Fenchel inequality for mixed volumes, which states that
MV (P1 , P2 , P3 , . . . , Pn )2 MV (P1 , P1 , P3 . . . , Pn ) MV (P2 , P2 , P3 , . . . , Pn )
793
for rational polytopes P1 , . . . , Pn in Rn . See [105, Sec. 5.4] for a toric proof.
The McMullen Conjecture. The deeper connection between toric varieties and
polytopes appears in Stanleys 1980 paper [257], which completed the proof of the
McMullen conjecture on the face numbers of an n-dimensional simplicial polytope
P. If fi is the number of i-dimensional faces of P, then Stanley defined

i
n
X
X
i j n j
(ni)
(1)
(A.1.1)
hi =
f j1 =
(1)
fn1 ,
ni
ni
=ni
j=0
where f1 = 0. Note that hi equals the number hni (P) defined in (12.5.13). The
hi satisfy the Dehn-Sommerville equations
(A.1.2)
hi = hni ,
and in 1971 McMullen conjectured that

(A.1.3)
hi hi1 0,
1 i n2 ,
and that if

ni
ni1
nr
hi hi1 =
+
+ +
i
i1
r
n
with 1 i 2 1 and ni > ni1 > > nr r 1, then

ni + 1
ni1 + 1
nr + 1
(A.1.4)
hi+1 hi
+
+ +
.
i+1
i
r+1
Furthermore, McMullen conjectured that if positive integers f0 , . . . , fn1 satisfy

(A.1.2)(A.1.4), then they are the face numbers of an n-dimensional simplicial
polytope. In other words, he conjectured that (A.1.2)(A.1.4) are necessary and
sufficient conditions for the fi to come from a simplicial polytope.
The sufficiency of (A.1.2)(A.1.4) was proved by Billera and Lee [27]. It

thus remained to show necessity, specifically that (A.1.3) and (A.1.4) hold for any
simplicial polytope. Stanleys paper [257] is three pages long, where the first page
recalls the conjecture and the third is mostly references. It is a one-page proof!
Stanley first explains why one can assume that P NR is rational with the
origin as an interior point. Then cones over the faces of P give a complete fan in
NR . The toric variety X is projective, which for us is easy since is the normal
fan of the dual polytope P MR . Stanley had to work a little harder to prove this
since the theory of normal fans was not fully developed at the time. Note also that
X is an orbifold since P is simplicial. Hence:
(A.1.2) follows from Poincare duality.
(A.1.3) follows from the hard Lefschetz theorem.
Stanley then proves (A.1.4) using hard Lefschetz and known results about graded
algebras. A complication is that his arguments require a version of hard Lefschetz
that applies to projective orbifolds, yet Steenbrinks proof in [261] has a gap. To
794
get a complete proof of hard Lefschetz in this situation, one needs to use hard
Lefschetz for intersection cohomology, which was proved by Saito in 1990. We
discuss the intersection cohomology of toric varieties in 12.5 and give a careful
statement of hard Lefschetz in Theorem 12.5.8. See also [105, Sec. 5.2].
When we studied polytopes in 9.4, we focused on simple polytopes P MR
and used their face numbers to define the numbers hi in Theorem 9.4.7. These
differ from the hi defined in (A.1.1), but are closely related since the dual of a
simple polytope is simplicial. See Exercise 9.4.10.
A nice commentary on the first decade of toric geometry was written by Reid
in 1983 [237], where he notes that the construction of the toric variety of a fan
has been of considerable use within algebraic geometry in the
last 10 years . . . and has also been amazingly successful as a
tool of algebro-geometric imperialism, infiltrating areas of combinatorics. For example, the hard Lefschetz theorem on the cohomology of projective varieties has been translated into combinatorics to complete the proof of a long-standing conjecture of
P. McMullen giving an if and only if condition for the existence
of a simplicial polytope with a given number fi of i-dimensional
faces.
Final Comment. We end with an observation from the social history of modern
mathematics. The foundational works on toric varieties were written in a variety of
languages, including English, French, German and Russian. This is quite different
from the current era, where most mathematics is published in English.
A.2. The Story Since 1980

After 1980, the study of toric varieties expanded rapidly, especially during the
1990s. We will say some brief words about some of these developments, though
we caution the reader that many important topics will be omitted.
The 1980s. There was a steady stream of papers about toric varieties in the 1980s,
some of which are featured in this book. Here are a few examples:
In 1983, Reid published Decomposition of toric morphisms [236], cited in
Chapter 15.
In 1986, Danilov and Khovanskii published Newton polyhedra and an algorithm for calculating Hodge-Deligne numbers [77], whose results give a vast
generalization of the sectional genus formula proved in Proposition 10.5.8.
In 1988, Kleinschmidt published A classification of toric varieties with few
generators [177], discussed in 7.3.
A.2. The Story Since 1980
795
In 1989, Gel fand, Kapranov and Zelevinsky published Newton polyhedra of

principal A-determinants [112], which introduced the GKZ decomposition that
we studied in Chapters 14 and 15.
The titles of these papers give a good sense of the wide range of topics relevant to
toric geometry. A nice overview of the evolution of toric varieties in the 1980s can
be found in Odas 1989 survey paper [219].
One notable event of the decade was the 1988 publication of Odas Convex
Bodies and Algebraic Geometry [218]. This mature and polished book has had a
major influence on the field.
The 1990s. In 1989, the first author had the good fortune to attend a conference at
Washington University where Bill Fulton gave a series of lectures on toric varieties,
based on notes that eventually became his wonderful 1993 book Introduction to
Toric Varieties [105]. As a result, toric geometry began the 1990s with a rich body
of work and the superb expositions of Oda and Fulton. This, coupled with two
other events, led to an explosion of papers about toric varieties.
The first event was the quotient representation of toric varieties given in 5.1,
which appeared in Audins book The Topology of Torus Actions on Symplectic
Manifolds [11] in 1991. As before, this construction cropped up independently in
several places in the 1990s. The 1995 paper [65] notes five independent discoveries
of Theorem 5.1.11. The total coordinate ring from 5.2 was introduced in [65] and
has been used in non-toric situations, often under the name Cox ring (see [150]).
The second event was mirror symmetry in mathematical physics, which had a
huge impact on algebraic geometry and on the study of toric varieties in particular.
Here are three brief hints of the role played by toric geometry:
The 1993 paper Phases of N = 2 theories in two dimensions [277] by Witten
used the symplectic version of the quotient representation of a toric variety
(see the end of 12.2) to construct a quantum field theory called a gauged
linear sigma model.
The 1994 paper Dual polyhedra and mirror symmetry for Calabi-Yau hypersurfaces in toric varieties by Batyrev [16] showed that the duality of reflexive
polytopes (see 8.3 and Proposition 11.2.1) is related to mirror symmetry.
The 1994 paper Calabi-Yau moduli space, mirror manifolds and spacetime
topology change in string theory by Aspinwall, Greene and Morrison [7] used
ideas from the secondary fan (see Chapters 14 and 15) to study moduli spaces
associated to certain quantum field theories.
The full story is summarized in the paper [66, Sec. 10] and developed in more
detail in the book [68].
One way to understand the growth of the field during this decade is to consider
the four survey papers that have been written in recent years about toric varieties:
796
Odas 1989 Geometry of toric varieties [219], with 114 references.
Odas 1994 Recent topics on toric varieties [221], with 64 references.
Coxs 1997 Recent developments in toric geometry [66], with 157 references.
Coxs 2001 Update on toric geometry [67], with 240 references.
We should also mention three notable books published in the 1990s:
The 1993 book Discriminants, Resultants and Multidimensional Determinants

[113] by Gel fand, Kapranov and Zelevinsky included the first careful treatment of nonnormal toric varieties.
The 1996 book Combinatorial Convexity and Algebraic Geometry by Ewald
[93] was the first attempt to explain the classic theory of normal toric varieties
to a wider audience than just algebraic geometers.
The 1996 book Grobner Bases and Convex Polytopes by Sturmfels [264] covers the elementary theory of affine and projective toric varieties and explains
their relation to toric ideals, Grobner Bases, and graded algebras.
Applications of Toric Varieties. Besides the applications to physics coming from
mirror symmetry, toric varieties began to be applied to other contexts starting in
the mid 1990s. Here are five examples:
Relations between toric ideals and integer programming are discussed in the
1996 book Grobner Bases and Convex Polytopes [264] mentioned above.
Applications to solving systems of polynomial equations are described in the
1998 book Using Algebraic Geometry [70].
Applications to geometric modeling are explored in the 2003 book Topics in
Algebraic Geometry and Geometric Modeling [116].
Applications to coding theory appear in the 2008 book Advances in Algebraic
Geometry Codes [192].
Applications to algebraic statistics are discussed in the 2009 book Lectures on
Algebraic Statistics [84].
Toric Varieties in the 21st Century. Since 2001, research on toric varieties has
continued at a rapid pace. One can measure the growth of the field by looking
at the MathSciNet database. Starting in 1991, toric varieties have had their own
classification number, 14M25, and during the period between January 2001 and
August 2010, there were
229 papers that listed 14M25 as their primary classification, and
another 434 papers that listed 14M25 as their secondary classification.
It is our hope that this book will help you understand the reasons for this amazing activity and encourage some of you to make your own contributions to this
wonderful field of mathematics.
Appendix B
Computational Methods
There are a wide range of packages available for computations in toric geometry.
We focus on packages supported by two open source systems:
(a) Macaulay2 [123], by Dan Grayson, Mike Stillman and collaborators.
(b) Sage [262], by William Stein and collaborators.
Both Magma and GAP also have toric packages, and the open source algebra systems Singular [78] and CoCoA [72] are functionally similar to Macaulay2. The
Macaulay2 toric package NormalToricVarieties [252] is by Greg Smith, and
the Sage toric package ToricVarieties [44] is by Volker Braun and Andrey
Novoseltsev. There are many other programs which are useful for computations in
toric and polyhedral geometry; an incomplete list might include:
(a) Normaliz by Bruns, Ichim and Soger [57].
(b) LattE by De Loera [79].
(c) Polymake by Gawrilow and Joswig [114].
(d) Polyhedra(Sage) by Braun, Hampton, Novoseltsev and collaborators [43].
(e) 4ti2 by Hemmecke, Koppe, Malkin and Walter [140].
(f) Gfan by Jensen [161].
(g) TOPCOM by Rambau [233, 234].
Many of the programs listed above have interfaces to Macaulay2 and Sage, as
well as interfaces to each other. There are also packages for applications, such as
coding theory (toriccodes by Ilten [153], or toric by Joyner [163]) and physics
(PALP, by Kreuzer and Skarke [182, 183]). In what follows, we will illustrate how
to compute with toric varieties by working through a series of examples. To save
space, we will often suppress superfluous output.
797
798
B.1. The Rational Quartic

Example B.1.1. We begin with a barehanded example. In Exercise 1.1.7, we
considered the map : C2 C4 defined by
(s,t) = (s4 , s3 t, st 3 ,t 4 ).
Since the monomials are homogeneous, also defines a map P1 P 3 . Note that
these monomials correspond to the subset A Z2 given by the columns of

4 3 1 0
.
0 1 3 4
To find the image of via Macaulay2, we compute the kernel I of g = : S R:
i1: R = QQ[s,t]
i2: S = QQ[x,y,z,w]
i3 : g = map(R,S,{s^4,s^3*t,s*t^3,t^4})
o3 : RingMap R <--- S
bi4 : I = kernel g
3
2
2
2
3
2
o4 = ideal (y*z - x*w, z - y*w , x*z - y w, y - x z)
These are the equations appearing in Exercise 1.1.7.
Example B.1.2. The map of the previous example gives the projective curve
XA P 3 . Example 2.1.10 showed that XA is normal but not projectively normal.
To see this computationally, we use Exercise 9.4.6, which implies that XA P 3 is
projectively normal if and only if XA is normal and H 1 (P 3 , IXA ()) = 0 for all
0. By local duality [89, Thm. A4.2] for the curve XA P 3 ,
H 1 (P 3 , IXA ()) ExtS3 (S/IXA , S)4 .
Since XA is defined by the ideal I computed in Example B.1.1, we obtain:
i5
o5
i6
o6
:
=
:
=
E3 = Ext^3(coker gens I, S)
cokernel {-5} | w z y x |
hilbertFunction(-5,E3)
1
Since dim H 1 (P 3 , IXA (1)) = 1, projective normality fails. This reflects our failure
to use all lattice points of the polytope P = Conv(A ) to map to projective space.
Our computational check works in general: a normal variety X P n is projectively
normal exactly when ExtSn (S/IX , S) = 0, where S = C[x0 , . . . , xn ].
Exercises for B.1.

B.1.1. Use Chapter 9 to show that H i (P n , OPn ()) = 0 for all 1 i n 1 and all .
B.2. Polyhedral Computations
799

Example B.2.1. The cone = Cone(e1 , e2 , e1 + e3 , e2 + e3 ) is depicted in Figure 2 of Chapter 1, and Example 1.2.9 showed that the dual cone is given by
= Cone(e1 , e2 , e3 , e1 + e2 e3 ) R3 . The Macaulay2 package Polyhedra
is one option for computing with cones: for the example above, we have
i1
i2
i3
o3
:
:
:
=
i4 :
o4 =
i5
o5
i6
i7
o7
:
=
:
:
=
i8 :
o8 =
loadPackage "Polyhedra"
M= matrix{{1,0,0},{0,1,0},{1,0,1},{0,1,1}}
C=posHull transpose M
{ambient dimension => 3
}
dimension of lineality space => 0
dimension of the cone => 3
number of facets => 4
number of rays => 4
rays C
| 1 0 1 0 |
| 0 1 0 1 |
| 0 0 1 1 |
fVector C
{1, 4, 4, 1}
Cv = dualCone C
rays Cv
| 1 0 1 0 |
| 0 1 1 0 |
| 0 0 -1 1 |
hilbertBasis C
{| 1 |, | 0 |, | 1 |, | 0 |}
| 0 | | 1 | | 0 | | 1 |
| 0 | | 0 | | 1 | | 1 |
Example B.2.2. We now illustrate a fan computation. In Example 6.1.17 we studied the smooth fan F consisting of a subdivision of the positive orthant into cones
B1,..,B9, and the remaining 7 orthants C1,..,C7.
o9 :
i25 :
o25 =
i26 :
o26 =
C1=posHull transpose matrix{{0,-1,0},{0,0,1},{-1,0,0}};

(input cones C2..B9, supressed)
F=fan{C1,C2,C3,C4,C5,C6,C7,B1,B2,B3,B4,B5,B6,B7,B8,B9}
{ambient dimension => 3
}
number of generating cones => 16
number of rays => 10
top dimension of the cones => 3
isPolytopal F
false
800
Example B.2.3. In this example, we compute the secondary polytope for the convex hull of six points in the plane. The secondary polytope is an associahedron, and
appears in Example 15.2.13. One package which computes the secondary polytope
is TOPCOM by Rambau. Braun has created a Sage version of TOPCOM, which we use
for the computation (you may need to install this package separately).
sage: pointmatrix = matrix([
[ 1, 1, 0, -1, -1, 0 ],
[ 0, 1, 1, 0, -1, -1 ]])
sage: pc = PointConfiguration(pointmatrix.columns())
sage: Tris = pc.triangulations_list()
sage: Tri = Tris[7]
sage: list(Tri)
[[0, 1, 4], [0, 4, 5], [1, 2, 3], [1, 3, 4]]
sage: show(Tris[7].plot(axes=False))
Figure 1. The seventh triangulation
Next, we generate the secondary polytope and face lattice

sage: K3 = pc.secondary_polytope(); K3
A lattice polytope: 3-dimensional, 14 vertices.
sage: K3.faces()
[[0],[1], (vertices: output suppressed) [13]],
[[1,6],[5,6],[0,1],[0,5],[6,10],[1,4],[4,13],[10,13],
[5,7],[7,9],[9,10],[3,4],[2,3],[0,2],[3,12],[12,13],
[2,8],[8,11],[11,12],[7,8],[9,11]],
[[0,1,5,6],[7,8,9,11],[0,2,5,7,8],[0,1,2,3,4],[3,4,12,13],
[2,3,8,11,12],[5,6,7,9,10],[1,4,6,10,13],[9,10,11,12,13]]
801
sage: plotK3 = K3.plot3d() #view the secondary polytope

sage: plotK3.rotateZ(60*pi/180).show(viewer=tachyon)
Figure 2. The secondary polytope K3
Figure 2 is a simplified version of the Sage output optimized for viewing in black
and white. To see the 1-skeleton appearing in Figure 9 of Example 15.2.13, enter:
sage: Kp = Polyhedron(vertices=K3.vertices().columns())
sage: Egraph = Kp.graph()
sage: Egraph.show()
4
13
1
3
12
10
6
0
2
11
8
7
Figure 3. The edge graph of K3
802
Example B.2.4. In our final polyhedral example, we compute the Grobner fan of
the ideal hx7 1, y x5 i which appears in Example 10.3.1, using the package Gfan,
which has both Macaulay2 and Sage interfaces. In Macaulay2:
i1 : R=QQ[x,y]; I=ideal(x^7-1, y-x^5); gfan(I)
+-------------+----------------------------+
| 7
| 7
3
|
o1 = |{y , x}
|{y - 1, - y + x}
|
+-------------+----------------------------+
| 3
2
3 | 3
2
3
2
|
|{y , x y, x }|{y - x, x y - 1, x - y } |
+-------------+----------------------------+
| 2
2
5 |
3
2
2
5
|
|{y , x y, x }|{- x + y , x y - 1, x - y}|
+-------------+----------------------------+
|
7
|
5
7
|
|{y, x }
|{- x + y, x - 1}
|
+-------------+----------------------------+
And in Sage:
sage:
sage:
sage:
[[y^7
[-x^3
sage:
R.<x,y,z> = PolynomialRing(QQ)
I=R.ideal([x^7-1,y-x^5]).groebner_fan()
I.reduced_groebner_bases()
- 1, -y^3 + x], [y^3 - x, x^2*y - 1, x^3 - y^2],
+ y^2, x^2*y - 1, x^5 - y], [-x^5 + y, x^7 - 1]]
I.render()
10
10
12
Figure 4. Output of the render command
The render command expects input in three variables (hence the three variable
ring). The figure above is the intersection of the Grobner fan with the standard two
simplex, which after an affine transformation gives the fan of Example 10.3.1.
B.3. Normalization and Normaliz
803
B.3. Normalization and Normaliz

The Normaliz package is designed for computations with affine monoids, vector
configurations, lattice polytopes, and rational cones.
As input, Normaliz expects a file containing matrices; each matrix is preceded
by two lines, the first indicating the number of rows and the second the number of
columns.
Example B.3.1. We illustrate Normaliz for the very ample but not normal polytope of Example 2.2.20.
10
6
1 1 1 0 0 0
(8 additional lattice points, suppressed)
0 0 1 1 0 1
1
Save the file above as VeryampleNonNormal.in. Normaliz supports ten types
of input, specified by the number on the line following the matrix. In the example
above the 1 on the last line indicates we want to normalize the monoid generated by
the rows of the matrix. To execute the code, use the graphical interface jNormaliz,
or type directly from the command line:
% norm64 VeryampleNonNormal.in
The output is saved in the file VeryampleNonNormal.out, which reads:
11 Hilbert basis elements
10 height 1 Hilbert basis elements
10 extreme rays
22 support hyperplanes
rank = 6 (maximal)
index = 1
original monoid is not integrally closed
extreme rays are homogeneous via the linear form:
1 1 1 1 1 1
Hilbert basis elements are not homogeneous
multiplicity = 21
h-vector:
1 4 11 4 1 0
Hilbert polynomial:
1/1 157/60 25/8 53/24 7/8 7/40
****************************************************
11 Hilbert basis elements:
(output suppressed: original rays and 1 1 1 1 1 1)
804
10 extreme rays:
(output suppressed: original rays)
22 support hyperplanes:
-1 2 -1 2 2 -1
-1 -1 2 2 2 -1
(remaining 20 hyperplanes suppressed)
1 congruences:
1 1 1 1 1 1 3
10 height 1 Hilbert basis elements:
(output suppressed: original rays)
Normaliz has interfaces for both Macaulay2 and Singular. For example, start
Macaulay2 in the directory containing Normaliz, and enter the commands
i1 : installPackage "Normaliz"
i2 : A = matrix{{1,1,1,0,0,0},(suppressed),{0,0,1,1,0,1}}
i3 : normaliz(A,1)
Example B.3.2. Consider the binomial ideal I computed in Example B.1.1. To

find the associated monoid and normalization, we use the option 10 in Normaliz.
We encode the first generator yz xw of I as -1 1 1 -1, and similarly for the
other three generators. Hence our input file is
4
4
-1 1
0 -1
1 -2
-2 3
10
1 -1
3 -2
2 -1
-1 0
This yields output (with some items suppressed)

4 original generators:
0 4
3 1
1 3
4 0
5 Hilbert basis elements:
4 0
3 1
2 2
1 3
0 4
You should compare this to what we computed in Example 1.3.9.
B.4. Sheaf Cohomology and Resolutions
805
Example B.3.3. In Example 9.4.5 we found the Ehrhart polynomial for the lattice
simplex Conv{(0, 0, 0), (1, 0, 0), (0, 1, 0), (1, 1, 3)}. Create an input file B3-3.in as
below:
4
3
0
1
0
1
2
0
0
1
1
0
0
0
3
The command norm64 -h B3-3.in produces an output file B3-3.out as below.

The -h option stipulates that the Ehrhart polynomial is to be included in the output
data.
4 lattice points in polytope
4 extreme points of polytope
4 support hyperplanes
polytope is not integrally closed
dimension of the polytope = 3
normalized volume = 3
h-vector:
1 0 2 0
Ehrhart polynomial:
1/1 3/2 1/1 1/2
The translates to the polynomial 1 + 23 x + x 2 + 21 x 3 from Example 9.4.5.
B.4. Sheaf Cohomology and Resolutions

Example B.4.1. Example 9.5.6 computed H 1 (P1 , OP1 (5)) via Ext. Let I =
hx, yi S = C[x, y]. Then
H 1 (P1 , OP1 (5)) = HI2 (S)5 = Ext2S (A, S)5 ,
where A = S/I [k] = S/hx k , y k i for k sufficiently large. Since dim HI2 (S)2 = 4 by
Example 9.5.3, we can use Macaulay2 to find a k that works. Suppressing some
outputs, one computes:
i1
i2
i3
o3
:
:
:
=
S = QQ[x,y]
A = coker matrix{{x^3,y^3}}
hilbertFunction(-5,Ext^2(A,S))
2
So k = 3 is not big enough. Repeating this with k = 4 shows that k = 4 works.
806
Example B.4.2. In Example 9.5.13, we needed a resolution of the cotangent bundle of P1 . This is done using the commands:
i1
i2
i3
i4
:
:
:
:
S = QQ[x,y,z]
OM = ker matrix{{x,y,z}}
F = res OM
F.dd
3
1
o4 = 0 : S <-------------- S : 1
{2} | z |
{2} | x |
{2} | -y |
The last two commands compute the resolution and display the differentials. This
gives the resolution shown in (9.5.9).
B.5. Sheaf Cohomology on the Hirzebruch Surface H2

This example illustrates the NormalToricVarieties package for Macaulay2,
written by Greg Smith. Use viewHelp for documentation on the commands.
Example B.5.1. In Example 13.2.12 we studied the line bundle L = OX (D) for
D = 3D3 5D4 on X = H2 , computing that H 0 (X , L ) = 0, dim H 1 (X , L ) = 2,
and dim H 2 (X , L ) = 6. The command hirzebruchSurface(2) creates the toric
variety X = H2 :
i1
i2
i3
o3
i4
o4
i5
o5
:
:
:
=
:
=
:
=
installPackage "NormalToricVarieties"
H2 = hirzebruchSurface(2)
rays H2
{{1, 0}, {0, 1}, {-1, 2}, {0, -1}}
max H2
{{0, 1}, {0, 3}, {1, 2}, {2, 3}}
isFano H2
false
Now we bring D into the picture. In NormalToricVarieties a divisor is specified

by a tuple corresponding to the rays of the fan, so with the rays ordered as above,
D is 3D0 5D3 .
i6 : D = 3*H2_0 - 5*H2_3
o6 = 3*D - 5*D
0
3
i7 : isCartier D
o7 = true
i8 : isAmple D
o8 = false
B.5. Sheaf Cohomology on the Hirzebruch Surface H2
807
To compute cohomology, we turn D into the sheaf L = OX (D), and use a loop to
display the ranks of H i(X , L ):
i9 : L = OO D
1
o9 = OO (3,-5)
H2
i10 : for i to 2 list rank HHî(H2, L)
o10 = {0, 2, 6}
Example B.5.2. We continue to study cohomology on X = H2 , but now turn our

attention to 1X . In Table 4 of Example 9.5.14 we computed dim H 1 (X , 1X (a, b))
for (a, b) {(2, 2), . . . , (2, 2)}. We will build a script called cohomologyTable
which takes as input an integer k reflecting the H k to compute, a sheaf, and high and
low values for the range of degrees in the Picard group. Since cohomologyTable
builds on packages BoijSoederberg and BGG, we need to load these. Below, we
check the computations in Example 9.5.14
i11 : loadPackage "BoijSoederberg"
i12 : loadPackage "BGG"
i13 : OM = cotangentSheaf H2
o13 = cokernel {2, 0} | 2x_1x_3 |
{-1, 2} | x_0
|
{-1, 2} | -x_2
|
1
2
o13 : coherent sheaf on H2, quotient of OO (-2,0)++OO (1,-2)
H2
H2
i14 : cohomologyTable := (k,F,lo,hi) ->
(DegRange = toList(lo#0..hi#0);
new CohomologyTally from select(flatten apply(DegRange,
j -> apply(toList(lo#1..hi#1),
i -> {(j,i-j), rank HH^k(variety F, F(i,j))})),
p -> p#1 != 0));
i15 : cohomologyTable(1,OM,{-2,-2},{2,2})
-2 -1 0 1 2
o15 = 2: 3 2 1 . .
1: 4 2 1 . .
0: 3 2 2 2 3
-1: . . 1 2 4
-2: . . 1 2 3
Exercises for B.5.
B.5.1. Compute the cohomology table for H i (P 2 , 1P2 ( j)) appearing in Example 9.5.13.
808
B.6. Resolving Singularities

In this section we use the ToricVarieties package in Sage, written by Braun
and Novoseltsev, to resolve singularities. The package finds a simplicial resolution
automatically.
Example B.6.1. Let us revisit Example B.2.1. Consider the affine toric variety
defined by the cone = Cone(e1 , e2 , e1 + e3 , e2 + e3 ), which is the nonsimplicial
affine threefold appearing in Example 11.1.12. We first find a simplicial resolution:
sage: C = Cone([(1,0,0), (0,1,0), (1,0,1), (0,1,1)]); C
3-d cone in 3-d lattice N
sage: QuadCone = AffineToricVariety(C)
sage: Orbif = QuadCone.resolve_to_orbifold()
sage: Orbif.fan().ngenerating_cones()
2
sage: [ cone.ambient_ray_indices() for cone in Orbif.fan() ]
[(2, 3, 0), (1, 3, 0)]
This is the refinement 1 of appearing in Figure 3 of Example 11.1.12. Note that
the rays are indexed by 0..3 rather than 1..4.
Example B.6.2. In this example, we resolve weighted projective space P(1, 1, 2),
whose fan is pictured in Figure 11 of Example 2.4.6.
sage: s0 = Cone([(0,1), ( 1, 0)]);
sage: s1 = Cone([(0,1), (-1,-2)]);
sage: s2 = Cone([(1,0), (-1,-2)]);
sage: F = Fan([s0, s1, s2]);
sage: P112 = ToricVariety(F); P112
2-d toric variety covered by 3 affine patches
sage: P112.is_orbifold()
True
sage: P112.is_smooth()
False
sage: lp = LatticePolytope( matrix(F.rays()).transpose() )
sage: lp.points() # columns are all points in the convex hull
[ 0 1 -1 0 0]
[ 1 0 -2 -1 0]
sage: BlP112 = P112.resolve(new_rays=[(0,-1)]); BlP112
2-d toric variety covered by 4 affine patches
sage: BlP112.is_smooth()
True
We used this resolution of P(1, 1, 2) in Example 13.4.4.
B.7. Intersection Theory and Hirzebruch-Riemann-Roch
809
B.7. Intersection Theory and Hirzebruch-Riemann-Roch

Example B.7.1. In Example 13.2.12, we checked the Hirzebruch-Riemann-Roch
theorem for the divisor D = D3 5D4 on H2 . Using the ToricVarieties package
in Sage and observing that the variety BlP112 in the previous example is X = H2 ,
we calculate
Z
X
ch(L ) Td(X ),
L = OX (D).
sage: BlP112.fan().ray_matrix()
[ 0 1 -1 0]
[ 1 0 -2 -1]
sage: BlP112.cohomology_ring().gens()
([2*z2 + z3], [z2], [z2], [z3])
sage: d4 = BlP112.fan().cone_containing(0);d4
sage: D4 = d4.cohomology_class()
[2*z2 + z3]
sage: d3 = BlP112.fan().cone_containing(1);d3
sage: D3 = d3.cohomology_class(); D3
[z2]
sage: chL = (3*D3-5*D4).exp(); chL
[-5*z3^2 - 7*z2 - 5*z3 + 1]
sage: tdX = BlP112.Todd_class(); tdX
[-1/2*z3^2 + 2*z2 + z3 + 1]
sage: BlP112.integrate(tdX*chL)
4
This agrees with our computation of (L ) in B.5. We can also check by extracting the degree two piece of ch(L ) Td(X ):
sage: chD*tdX
[-2*z3^2 - 5*z2 - 4*z3 + 1]
sage: Deg2part=(chD*tdX).part_of_degree(2); Deg2part
[-2*z3^2]
With labelling as in Example 4.1.8, z3 corresponds to the ray u2 , and one checks
that the corresponding divisor D2 has self-intersection 2. Thus 2D22 = 4.
Example B.7.2. To compute sheaf cohomology using ToricVarieties, first we

create a divisor from an list indexed by the rays, where the order corresponds to
the ray matrix of the fan:
sage: Ddiv = BlP112.divisor([-5,3,0,0]);Ddiv
sage: Ddiv.cohomology()
(0, 2, 6)
sage: M = BlP112.fan().dual_lattice()
810
sage: Ddiv.cohomology(M(-2,1))
(0, 1, 0)
The last line gives dim H i(H2 , OH2 (D))(2,1) ; the weight (2, 1) corresponds to a
weight of (2, 1) in Figure 2 in Example 13.2.12 since the fans are flipped.
B.8. Anticanonical Embedding of a Fano Toric Variety

Example B.8.1. The toric variety X = X in Examples 9.1.2 and 9.1.8 is the
blowup of P 2 at the three torus-fixed points. We first show how to create this toric
variety from scratch using the Macaulay2 package NormalToricVarieties.
The input is a list of rays and a list of maximal cones. The rays below are ordered
as in Figure 2 of Example 9.1.2.
i2 : rayList = {{1,0},{1,1},{0,1},{-1,0},{-1,-1},{0,-1}}
i3 : coneList = {{0,1},{1,2},{2,3},{3,4},{4,5},{5,0}}
i4 : BlP2 = normalToricVariety(rayList,coneList)
Now let D = KX be the anticanonical divisor, which is very ample since X
is a smooth Fano surface. We analyze the equations that arise when we embed
X into P(H 0 (X , OX (D))). Below, we create a script which takes as input a toric
divisor D, finds the sections, and returns the ideal defining X P(H 0 (X , L )) for
L = OX (D). Make a file called Embed as below:
projEmb = (D)->(X = variety D;
L = OO D;
m = rank HH^0(X, L);
S = ring X;
R = QQ[y_0..y_(m-1)];
phi = map(S, R, basis(- first degrees L, S));
kernel phi)
Now we reap the benefits of our work:
i5
i6
i7
i8
o8
i9
:
:
:
:
=
:
D = BlP2_0+BlP2_1+BlP2_2+BlP2_3+BlP2_4+BlP2_5
load "Embed"
I = projEmb(D);
transpose mingens I
{-2} | u_4u_5-u_3u_6 | (more quadrics, output suppressed)
hilbertPolynomial(coker gens I, Projective=>false)
2
o9 = 3i + 3i + 1
Exercises for B.8.
B.8.1. Compute the Ehrhart polynomial of PKX using Normaliz as in Example B.3.3.
B.8.2. Compute H i (X, OX (D)) for the divisor D = D3 + 2D5 D6 from Example 9.1.8.
Appendix C
Spectral Sequences
Spectral sequences are important tools in modern algebraic geometry and algebraic
topology that are used several times in Chapters 9 and 12. Here we give some
background necessary for understanding these applications for readers who have
not encountered these objects before. Our discussion is very far from complete; we
refer the reader to [115], [136], [199], or [276] for more comprehensive treatments
of spectral sequences.
C.1. Definitions and Basic Properties

At first glance, spectral sequences may seem rather complicated. Fortunately, they
are often straightforward to use in practice.
The Definition. The spectral sequences encountered in this book will all be of the
following type and will be relatively simple to understand.
Definition C.1.1. A (cohomology) spectral sequence is a collection of abelian
groups Erp,q and homomorphisms drp,q with the following structure and properties:
(a) The groups are Erp,q , indexed by integers p, q, r. Fixing r, we obtain one sheet
of the spectral sequence, which is visualized as a diagram of groups indexed
by integer lattice points in the plane.
(b) In the rth sheet, there are homomorphisms
drp,q : Erp,q Erp+r,qr+1
such that drp+r,qr+1 drp,q = 0 for all p, q, r. In other words, the rth sheet
splits up into a collection of cochain complexes in which the differentials are
all mappings of bidegree (r, 1 r) for the indexing by p, q.
811
812
p,q
(c) The (r + 1)st sheet Er+1
is the cohomology of (Erp,q , drp,q ), i.e.,

p,q
Er+1
= ker(drp,q : Erp,q Erp+r,qr+1) im(drpr,q+r1 : Erpr,q+r1 Erp,q ).
In many of the examples in the text, the Erp,q are vector spaces over Q or C
and the drp,q are linear mappings. There are also homology spectral sequences that
appear in many applications; the only differences are that the groups are usually
r and the differentials d r in a homology spectral sequence have bidewritten E p,q
p,q
gree (r, r 1).
We will always work with first quadrant spectral sequences, for which Erp,q = 0
when p < 0 or q < 0. Thus in each sheet, the nonvanishing terms lie in the quadrant
where p, q 0. The minimum value of r is usually 1 or 2, in which case we say
that (Erp,q , drp,q ) is an E1 or E2 spectral sequence respectively.
Intuition and Meaning of Convergence. When working with an E1 spectral sequence, for instance, we often know the differentials d1p,q explicitly. In general,
however, the differentials drp,q for r 2 are much more difficult to describe. Fortunately, there are many cases where the structure of the nonzero terms in a spectral
sequence forces many differentials to be zero. For instance, in the first quadrant
spectral sequences we will encounter, the differentials mapping to and from Erp,q
for fixed p, q vanish when r is sufficiently large. It follows that for each p, q, there
exists some r (depending on p, q) such that
p,q
p,q
Erp,q = Er+1 = Er+2 = .
p,q
This common value is defined to be E
.
Definition C.1.2. A first quadrant spectral sequence (Erp,q , drp,q ) converges to a

sequence of abelian groups H k , k 0, if there is a filtration
0 = F k+1 H k F k H k F k1 H k F 1 H k F 0 H k = H k
of H k by subgroups such that
p,q
E
F p H p+q /F p+1 H p+q .
For an E1 or E2 spectral sequence, we write this as

p,q
E1 H p+q
or
p,q
E2 H p+q
respectively.
According to [199, Ch. 1], the generic theorem about spectral sequences
says something like: there is a spectral sequence with E1p,q or E2p,q something
computable and converging to H p+q , something desirable. Several specific examples of this are discussed later. The intuition behind a convergent E1 spectral sequence is that we can usually compute the E2 terms from the d1 differentials explicitly. But we are primarily interested in what the spectral sequence converges to. We
can think of E2p,q as a first approximation to the convergent, and then E3p,q , E4p,q , . . .
C.1. Definitions and Basic Properties
813
as better and better approximations. If the situation is favorable, enough of the

p,q
higher differentials drp,q for r 2 and enough of the E
may vanish so that the
p+q
H
can be determined.
Here is a simple first example where we can carry out this sort of analysis.
Proposition C.1.3. Suppose E2p,q H p+q is a first quadrant E2 spectral sequence
p,q
with the property that E2 = 0 for all q > 0. Then
E2p,0 H p
for all p 0.
Proof. First observe that all differentials drp,q vanish for r 2 since E2p,q = 0 for
p,q
q > 0. It follows that E2p,q = E
for all p, q. Then we have
E2p,0 F p H p /F p+1 H p = F p H p
and 0 = E2p1,1 = E2p2,2 = = E20,p implies
0 = F p1 H p /F p H p = F p2 H p /F p1 H p = = F 0 H p /F 1 H p .
Hence F p H p = F p1 H p = = F 1 H p = F 0 H p = H p . Thus E2p,0 H p , as claimed.

p,q
The hypothesis of Proposition C.1.3 implies that the differentials d2

for all p, q. This is an instance of the following general idea.
vanish
Definition C.1.4. We say that a spectral sequence degenerates at the Er sheet if

the differentials dsp,q = 0 for all p, q and all s r.
p,q
Note that degeneration at Er implies that Er

strong form of convergence in this case.
p,q
E for all p, q, so we have a

p,q
If we have a convergent spectral sequence, say E2 H p+q , it is important

p,q
to note that there are situations where knowing E
for all p, q is not quite enough
p,q
p+q
to determine H
. This can happen for instance if there are several E
with
p + q = k for some k. Definition C.1.2 shows that we know the quotients
p,q
E
F p H p+q /F p+1 H p+q
in the filtration of H k = H p+q . However, additional extension data may be needed

to determine the isomorphism type of the abelian group H k in general. This situation does not arise when the Erp,q are vector spaces over a field. In all cases, though,
the proof of Proposition C.1.3 shows the following.
p+q be a spectral sequence. If E p,q = 0 for all
Proposition C.1.5. Let Erp,q
0 H
p0 ,q0
p, q with p + q = k except possibly p, q = p0 , q0 , then H k E
.
814
Edge Homomorphisms. Another feature of a convergent first quadrant E2 spectral sequence is that all differentials starting from Erp,0 vanish, so that the spectral
sequence gives maps
p,0
E2p,0 E3p,0 E
F p H p /F p+1 H p = F p H p H p .
Definition C.1.6. The map E2p,0 H p is called an edge homomorphism.

This leads to a more precise version of Proposition C.1.3.
Proposition C.1.7. Suppose E2p,q H p+q is a first quadrant E2 spectral sequence
p,q
with the property that E2 = 0 for all q > 0. Then the edge homomorphism
E2p,0 H p
is an isomorphism for all p 0.
A convergent first quadrant E2 spectral sequence has second edge homomorphism since all differentials ending at Er0,q vanish, so that we have the map
0,q
E30,q E20,q .
H q H q /F 1 H q = F 0 H q /F 1 H q = E
Definition C.1.8. The map H q E20,q is called an edge homomorphism.

For many E2 spectral sequences, we know some of the edge homomorphisms.
For example, we know E2p,0 H p in the two applications of Proposition C.1.7
0,q
given in 9.0, and H q E2 is known when we apply the Serre spectral sequence
(Theorem C.2.6) in 12.4.
C.2. Spectral Sequences Appearing in the Text

In the remainder of this appendix we discuss the first quadrant spectral sequences
used in Chapters 9 and 12 to derive facts about singular and sheaf cohomology of
toric varieties.
The Leray Spectral Sequence. The Leray spectral sequence of a continuous map
is used in 9.0.
Theorem C.2.1. Let f : X Y be a continuous map of topological spaces, and let
F be a sheaf of abelian groups on X . There is a spectral sequence
E2p,q = H p (Y , R q f F ) H p+q (X , F ).
Furthermore, when q = 0, the map H p (Y , f F ) H p (X , F ) is the edge homomorphism E2p,0 H p (X , F ) (see Definition C.1.6 above).

This spectral sequence is covered in [115, II.4.17] and [125, p. 463], and its
intuitive meaning is explained in the discussion leading up to Proposition 9.0.8.
C.2. Spectral Sequences Appearing in the Text
815
The Spectral Sequence of a Covering. Next, let U = {Ui } be an open cover of X

and F be a sheaf on X . If U consists of two sets and F is a constant sheaf, then
the Mayer-Vietoris long exact sequence can be used to obtain information about
the cohomology of X from the cohomology of U1 ,U2 and U1 U2 . More generally
H (X , F ) is determined by H (Ui0 Ui p+1 , F ) as we vary over all p by way
of the following spectral sequence of an open cover.
Theorem C.2.2. Let U = {Ui } be an open cover of X and F be a sheaf on X .
Then there is a spectral sequence
M
H q (Ui0 Ui p , F ) H p+q (X , F ),
E1p,q =
(i0 ,...,i p )[] p
complex.

This spectral sequence is discussed in [115, II.5.4]. We use it in 9.0 and 12.3.
A variant ofS
Theorem C.2.2 applies to a locally finite closed cover C = {Ci }. This
means X = i Ci , each Ci is closed in X , and each point x X has a neighborhood
meeting only finitely many Ci . A finite cover is clearly locally finite.
Theorem C.2.3. Let C = {Ci } be a locally finite closed cover of X and F be a
sheaf on X . Then there is a spectral sequence
M
p,q
H q (Ci0 Ci p , F ) H p+q (X , F ),
E1 =
(i0 ,...,i p )[] p
complex.

This spectral sequence is constructed in [115, II.5.2]. We will use it in 9.1
when we compute the sheaf cohomology of a toric variety.
The Spectral Sequence for Ext. Let X be a variety. As in 9.0, the Ext groups
ExtOp X (F , G ) are the derived functors of the Hom functor G 7 HomOX (F , G ) for
fixed F . The Ext sheaves E xtOq X (F , G )) are defined similarly. These are related
by the following spectral sequence.
Theorem C.2.4. Let F and G be quasicoherent sheaves on a variety X . Then
there is a spectral sequence
E2p,q = H p (X , E xtOq X (F , G )) ExtOp+q
(F , G ).
X
We will use this in Theorem 9.2.12, which is part of our discussion of Serre
duality. The spectral sequences from Theorems C.2.1 and C.2.4 are special cases of
the Grothendieck spectral sequence of a composition of functors (see [276, 5.8]).
816
The Spectral Sequences of a Filtered Toplogical Space. First, given an increasing

filtration of a topological space X ,
(C.2.1)
= X1 X0 X1 Xn = X ,
suppose we have information about the singular cohomology of the pairs (X p , X p1 )

for all p. This information can be stitched together to produce the E1 sheet of the
spectral sequence of a filtration. See [136, Ch. 1] and [255, Sec. 9.4] for proofs.
Theorem C.2.5. Let X be a topological space with a filtration of the form (C.2.1)
and R be a ring. Then there is a spectral sequence
p,q
E1 = H p+q (X p , X p1 , R) H p+q (X , R),
where the filtration on H p+q (X , R) is given by
F p H p+q (X , R) = ker(H p+q (X , R) H p+q (X p , R))

for all p, q.
This is not always a first quadrant spectral sequence, though it will be when
we apply Theorem C.2.5 in 12.3 to the filtration of a toric variety, where X p is the
union of the closures of the TN -orbits of dimension p.
The Serre Spectral Sequence. The spectral sequence for a filtration can be used to
derive the following Serre spectral sequence of a fibration. We recall that a fibration is a continuous mapping p : E B satisfying the homotopy lifting property
with respect to all spaces Y (see [199, 4.3]). It follows that if B is path-connected,
then all of the fibers p1 (b) for b B are homotopy equivalent. Hence we can
speak of the fiber and denote it by F. Let i : F E be the inclusion of the fiber.
Theorem C.2.6. Let p : E B be an orientable fibration over a path-connected
and simply-connected base B with fiber F, and let R be a ring. Then there is a
spectral sequence
E2p,q = H p (B, H q (F, R)) H p+q (E, R)
for a suitable filtration on H p+q (E, R). Furthermore:
(a) The edge homomorphism E2p,0 H p (E, R) is p : H p (B, R) H p (E, R).

0,q
(b) The edge homomorphism H q (E, R) E2 is i : H q (E, R) H q (F, R).
If B is not simply connected, then a similar result is true, but the fundamental
group 1 (B) acts on the the cohomology of the fiber and the E2 sheet involves
the more complicated cohomology with local coefficients. This is discussed in
[136, Ch. 1], [199, Thm. 5.2], and [255, Sec. 9.4]. We use the Serre spectral
sequence in 12.4 toSstudy the equivariant cohomology of a toric variety. In our
n
case, B = (P )n = (
=1 P ) , so the hypotheses of Theorem C.2.6 are satisfied.
The Serre spectral sequence can be seen as a special case of the Leray spectral
sequence from Theorem C.2.6.
Bibliography
[1] D. Abramovich, K. Karu, K. Matsuki and J. Wodarczyk, Torification and factorization of

birational maps, J. Amer. Math. Soc. 15 (2002), 531572.
[2] D. Abramovich, K. Matsuk and S. Rashid, A note on the factorization theorem of toric birational maps after Morelli and its toroidal extension, Tohoku Math. J. 51 (1999), 489537.
[3] A. ACampo-Neuen and J. Hausen, Toric invariant theory, in Algebraic Group Actions and
Quotients, (J. Wisnewski, ed.), Hindawi Publ. Corp., Cairo, 2004 137.
[4] O. Aichholzer, Extremal properties of 0/1-polytopes of dimension 5, in Polytopes Combinatorics and Computation, (G. Kalai and G. Ziegler, eds.), DMV Seminar 29, Birkhauser Verlag,
Basel Boston Berlin, 2000, 111130.
[5] T. Apostol, An Elementary View of Eulers Summation Formula, Am. Math. Monthly 106
(1999) 409-418.
[6] A. Ash, D. Mumford, M. Rapoport and Y. Tai, Smooth Compactification of Locally Symmetric
Varieties, Math. Sci. Press, Brookline, MA, 1975.
[7] P. Aspinwall, B. Greene and D. Morrison, Calabi-Yau moduli space, mirror manifolds and
spacetime topology change in string theory, Nucl. Phys. B416 (1994), 414480.
[8] M. F. Atiyah, On analytic surfaces with double points, Proc. Roy. Soc. London Ser. A 247
(1958) 237-244.
[9] M. F. Atiyah and R. Bott, The moment map and equivariant cohomology, Topology 23 (1984),
128.
[10] M. F. Atiyah and I. G. MacDonald, Introduction to Commutative Algebra, Addison-Wesley,
Reading, MA, 1969.
[11] M. Audin, The Topology of Torus Actions on Symplectic Manifolds, Progress in Math. 93,
Birkhauser Verlag, Basel Boston Berlin, 1991.
[12] M. Barile, On simplicial toric varieties of codimension 2, Rend. Istit. Mat. Univ. Trieste 39
(2007), 942
[13] M. Barile, D. Bernardi, A. Borisov and J.-M. Kantor, On empty lattice simplicies in dimension
4, preprint, 2009. Available electronically at arXiv:0912.5310 [math.AG].
[14] V. Batyrev, On the classification of smooth projective toric varieties, Tohoku Math. J. 43
(1991), 569585.
817
818
Bibliography
[15] V. Batyrev, Variations of the mixed Hodge structure of affine hypersurfaces in algebraic tori,
Duke Math. J. 69 (1993), 349409.
[16] V. Batyrev, Dual polyhedra and mirror symmetry for Calabi-Yau hypersurfaces in toric varieties, J. Algebraic Geom. 3 (1994), 493535.
[17] V. Batyrev, On the classification of toric Fano 4-folds, J. Math. Sci. (New York) 94 (1999),
10211050.
[18] V. Batyrev and L. Borisov, On Calabi-Yau complete intersections in toric varieties, in Higherdimensional complex varieties (Trento, 1994), (M. Andreatta and T. Peternell, eds.), de
Gruyter, Berlin, 1996, 3965.
[19] V. Batyrev and D. Cox, On the Hodge structure of projective hypersurfaces in toric varieties,
Duke Math. J. 75 (1994), 293338.
[20] V. Batyrev and D. Mel nikov, A theorem on nonextendability of toric varieties, Vestnik
Moskov. Univ. Ser. I Mat. Mekh. 1986, 2024, 118; English translation, Moscow Univ. Bull.
41 (1986), 2327.
[21] M. Beck, C. Haase, and F. Sottile, Formulas of Brion, Lawrence, and Varchenko on rational
generating functions for cones, Math. Intelligencer 31 (2009), no. 1, 917.
[22] M. Beck and S. Robins, Computing the Continuous Discretely, Springer, New York, 2007.
[23] A. A. Beilinson, J. Bernstein and P. Deligne, Faisceaux pervers, in Analysis and topology on
singular spaces, I (Luminy, 1981), Asterisque 100 Soc. Math. France, Paris, 1982, 5171.
[24] F. Berchtold and J. Hausen, Bunches of cones in the divisor class groupa new combinatorial
language for toric varieties, Int. Math. Res. Not. (2004), 261302.
[25] E. Bierstone and P. Milman, Desingularization of toric and binomial varieties, J. Algebraic
Geom. 15 (2006), 443486.
[26] L. J. Billera, The algebra of continuous piecewise polynomials, Adv. Math. 76 (1989), 170
183.
[27] L. J. Billera and C. Lee, Sufficiency of McMullens conditions for f -vectors of simplicial polytopes, Bull. Amer. Math. Soc. 2 (1980), 181185.
[28] L. J. Billera, P. Filliman and B. Sturmfels, Constructions and complexity of secondary polytopes, Adv. Math. 83 (1990), 155179.
[29] L. J. Billera, I. M. Gel fand and B. Sturmfels, Duality and minors of secondary polyhedra, J.
Combin. Theory Ser. B 57 (1993), 258268.
[30] C. Birkar, Birational Geometry, lecture notes for a course taught at Cambridge University,
Winter, 2007. Available electronically at arXiv:0706.1794 [math.AG].
[31] C. Birkar, P. Cascini, C. D. Hacon and J. McKernan, Existence of minimal models for varieties
of log general type, J. Amer. Math. Soc. 23 (2010), 405468.
[32] R. Blanco and S. Encinas, Embedded desingularization of toric varieties, preprint, 2009.
Available electronically at arXiv:0901.2211 [math.AG].
[33] M. Blickle, Multiplier ideals and modules on toric varieties, Math. Z. 248 (2004), 113121.
[34] M. Blickle and R. Lazarsfeld, An informal introduction to multiplier ideals, in Trends in Commutative Algebra, (L. Avramov, M. Green and C. Huneke, eds.), Math. Sci. Res. Inst. Publ.
51, Cambridge Univ. Press, Cambridge, 2004, 87114.
[35] R. Blumenhagen, B. Jurke, T. Rahn and H. Roschy, Cohomology of Line Bundles: A Computational Algorithm, preprint, 2010. Available electronically at arXiv:1003.5217 [math.AG].
[36] E. Bombieri and W. Gubler, Heights in Diophantine Geometry, Cambridge Univ. Press, Cambridge, 2006.
[37] A. Borel, Linear Algebraic Groups, second edition, Graduate Texts in Math. 126, Springer,
New York, 1991.
Bibliography
819
[38] A. Borel, et al., Seminar on transformation groups, Annals of Math. Studies 46, Princeton
Univ. Press, Princeton, 1960.
[39] L. Borisov, L. Chen and G. Smith, The orbifold Chow ring of toric Deligne-Mumford stacks,
J. Amer. Math. Soc. 18 (2005), 193215.
[40] R. Bott, Homogeneous vectors bundles, Ann. Math. 66 (1957), 203248.
[41] R. Bott, Review of Le theorème de Riemann-Roch, Bull. Soc. Math. France 86 (1958), 97136
by A. Borel and J.-P. Serre, in Mathematical Reviews, MR0116022 (22 #6817).
[42] T. Braden, Remarks on the combinatorial intersection cohomology of fans, Pure Appl. Math.
Q. 2 (2006), 11491186.
[43] V. Braun, M. Hampton, A. Novoseltsev, et.al., Sage polyhedral package, available electronically at www.sagemath.org, 2010.
[44] V. Braun, A. Novoseltsev, Sage toric varieties package, available electronically at www.
sagemath.org, 2010.
[45] G. Bredon, Sheaf Theory, second edition, Springer, New York, 1997.
[46] M. Brion, Points entiers dans les polyedres convexes, Ann. Sci. Ecole
Norm. Sup. 21 (1988),
653663.
[47] M. Brion, Piecewise polynomial functions, convex polytopes and enumerative geometry, in
Parameter spaces (Warsaw, 1994), Banach Center Publ. 36, Polish Acad. Sci., Warsaw, 1996,
2544.
[48] M. Brion, The structure of the polytope algebra, Tohoku Math. J. 49 (1997), 132.
[49] M. Brion, Rational smoothness and fixed points of torus actions, Transform. Groups 4 (1999),
127156.
[50] M. Brion and M. Vergne, An equivariant Riemann-Roch theorem for complete, simplicial toric
varieties, J. Reine Angew. Math. 482 (1997), 6792.
[51] A. Brndsted, An Introduction to Convex Polytopes, Springer, New York, 1983.
[52] W. Bruns and J. Gubeladze, Polytopal linear groups, J. Algebra 218 (1999), 715737.
[53] W. Bruns and J. Gubeladze, Polytopes, Rings, and K-Theory, Monographs in Mathematics,
Springer, New York, 2009.
[54] W. Bruns and J. Gubeladze, Semigroup algebras and discrete geometry, in Geometry of toric
varieties, (L. Bonavero and M. Brion, eds.), Semin. Congr. 6, Soc. Math. France, Paris, 2002,
43127.
[55] W. Bruns, J. Gubeladze and N. V. Trung, Normal polytopes, triangulations, and Koszul algebras, J. Reine Angew. Math. 485 (1997), 123160.
[56] W. Brun and J. Herzog, Cohen-Macaulay Rings, Cambridge Univ. Press, Cambridge, 1993.
[57] W. Bruns, B. Ichim and C. Soger, NORMALIZ: Algorithms for rational cones and affine
monoids, available electronically at www.math.uos.de/normaliz/, 2010.
[58] A. Buch, J. F. Thomsen, N. Lauritzen and V. Mehta, The Frobenius morphism on a toric
variety, Tohoku Math. J. 49 (1997), 355366.
[59] V. Buchstaber and N. Ray, An invitation to toric topology: vertex four of a remarkable tetrahedron, in Toric topology, (M. Harada, Y. Karshon, M. Masuda and T. Panov, eds.), Contemp.
Math. 460 (2008), Amer. Math. Soc., Providence, RI, 127.
[60] W. Buczynska, Fake weighted projective spaces, English translation of Toryczne przestrzenie
rzutowe, Magister thesis, 2002. Available electronically at arXiv:0805.1211 [math.AG].
[61] E. Cattani, D. Cox and A. Dickenstein, Residues in toric varieties, Compositio Math. 108
(1997), 3576.
820
Bibliography
[62] Y. Chen and V. Shokurov, Strong rational connectedness of toric varieties, preprint, 2009.
Available electronically at arXiv:0905.1430 [math.AG].
[63] C. H. Clemens and Z. Ran, A new method in Fano geometry, Internat. Math. Res. Notices
(2000), 527549.
[64] T. Coates, A. Corti, H. Iritani and H.-H. Tseng, Computing genus-zero twisted Gromov-Witten
invariants, Duke Math. J. 147 (2009), 377438.
[65] D. Cox, The homogeneous coordinate ring of a toric variety, J. Algebraic Geom. 4 (1995),
1550.
[66] D. Cox, Recent developments in toric geometry, in Algebraic geometrySanta Cruz 1995, (J.
Kollar, R. Lazarsfeld and D. Morrison, eds.), Proc. Sympos. Pure Math. 62.2, Amer. Math.
Soc., Providence, RI, 1997, 389436.
[67] D. Cox, Update on toric geometry, in Geometry of toric varieties, (L. Bonavero and M. Brion,
eds.), Semin. Congr. 6, Soc. Math. France, Paris, 2002, 141.
[68] D. Cox and S. Katz, Mirror Symmetry and Algebraic Geometry, Mathematical Surveys and
Monographs 68, Amer. Math. Soc., Providence, RI, 1999.
[69] D. Cox, J. Little and D. OShea, Ideals, Varieties and Algorithms, third edition, Undergraduate
Texts in Math., Springer, New York, 2007.
[70] D. Cox, J. Little and D. OShea, Using Algebraic Geometry, second edition, Graduate Texts
in Math. 185, Springer, New York, 2005.
[71] D. Cox and C. von Renesse, Primitive collections and toric varieties, Tohoku Math. J. 61
(2009), 309332.
[72] CoCoATeam, CoCoA Version 4.7.5Computations in commutative algebra, available electronically at cocoa.dima.unige.it, 1987present.
[73] D. Dais, Resolving 3-dimensional toric singularities, in Geometry of toric varieties, (L.
Bonavero and M. Brion, eds.), Semin. Congr. 6, Soc. Math. France, Paris, 2002, 213225
[74] D. Dais, Geometric combinatorics in the study of compact toric surfaces, in Algebraic and
Geometric Combinatorics, (C. Athanasiadis, V. Batyrev, D. Dais, M. Henk and F. Santos,
eds.), Contemp. Math. 423, Amer. Math. Soc., Providence, RI, 2006, 71123.
[75] D. Dais, U.-U. Haus and M. Henk, On crepant resolutions of 2-parameter series of Gorenstein
cyclic quotient singularities, Results Math. 33 (1998), 208265.
[76] V. Danilov, The geometry of toric varieties, Uspekhi Mat. Nauk. 33 (1978), 85134. English
translation, Russian Math. Surveys 33 (1978), 97154.
[77] V. Danilov and A. Khovanskii, Newton polyhedra and an algorithm for calculating HodgeDeligne numbers, Izv. Akad. Nauk SSSR Ser. Mat. 50 (1986), 925945. English translation
Math. USSR-Izv. 29 (1987), 279298.
[78] W. Decker, G.-M. Greuel, G. Pfister, G. Schonemann, Singular Version 3.1.1Commutative
algebra and algebraic geometry, available electronically at www.singular.uni-kl.de,
1989present.
[79] J. De Loera, D. Haws, R. Hemmecke, P. Huggins, J. Tauzer and R. Yoshida, LattE Version
1.2Lattice points in convex polytopes and integer programming, available electronically at
www.math.ucdavis.edu/~ latte/software.htm, 2003present.
[80] J. De Loera, J. Rambau and F. Santos, Triangulations: Structures for Algorithms and Applications, Springer, Berlin, 2010.
[81] D. Delzant, Hamiltoniens periodiques et images convexes de lapplication moment, Bull. Soc.
Math. France 116 (1988), 315339.
[82] M. Demazure, Sous-groupes algebriques de rang maximum du groupe de Cremona, Ann. Sci.
Ecole
Norm. Sup. 3 (1970), 507588.
Bibliography
821
[83] I. Dolgachev, Lectures on Invariant Theory, Cambridge Univ. Press, Cambridge, 2003.
[84] M. Drton, B. Sturmfels and S. Sullivant, Lectures on Algebraic Statistics, Oberwolfach Seminars 39, Birkhauser Verlag, Basel Boston Berlin, 2009.
[85] A. Durfee, Fifteen characterizations of rational double points and simple critical points, Enseign. Math. 25 (1979), 131163.
[86] D. Edidin and W. Graham, Equivariant intersection theory, Invent. Math. 131 (1998), 595
634.
[87] D. Edidin and W. Graham, Riemann-Roch for quotients and Todd classes of simplicial toric
varieties, Comm. Algebra 31 (2003), 37353752.
[88] F. Ehlers, Eine Klasse komplexer Mannigfaltigkeiten und die Auflosung einiger isolierter Singularitaten, Math. Ann. 218 (1975), 127156.
[89] D. Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate Texts
in Math. 150, Springer, New York, 1995.
[90] D. Eisenbud and J. Harris, The Geometry of Schemes, Graduate Texts in Math. 197, Springer,
New York, 2000.
[91] D. Eisenbud, M. Mustata and M. Stillman, Cohomology on toric varieties and local cohomology with monomial supports, J. Symbolic Comput. 29 (2000), 583600.
[92] E. J. Elizondo, P. Lima-Filho, F. Sottile and Z. Teitler, Arithmetic toric varieties, preprint,
2010. Available electronically at arXiv:1003.5141 [math.AG].
[93] G. Ewald, Combinatorial Convexity and Algebraic Geometry, Graduate Texts in Math. 168,
Springer, New York, 1996.
[94] G. Ewald and U. Wessels, On the ampleness of invertible sheaves in complete projective toric
varieties, Results Math. 19 (1991), 275278.
[95] E. M. Feichtner, Rational versus real cohomology algebras of low-dimensional toric varieties,
Proc. Amer. Math. Soc. 131 (2003), 16951704.
[96] M. Franz, The integral cohomology of toric manifolds, Proc. Steklov Inst. Math. 252 (2006),
5362.
[97] M. Franz, Describing toric varieties and their equivariant cohomology, Colloq. Math. 121
(2010), 116.
[98] M. Franz, torhom, a Maple package for toric cohomology, available electronically at http:
//www.math.uwo.ca/~ mfranz/torhom/, 2004.
[99] O. Fujino, Notes on toric varieties from Mori theoretic viewpoint, Tohoku Math. J. 55 (2003),
551564.
[100] O. Fujino, On the Kleiman-Mori cone, Proc. Japan Acad. Ser. A Math. Sci. 81 (2005), 8084.
[101] O. Fujino, Multiplication maps and vanishing theorems for toric varieties, Math. Z. 257
(2007), 631641.
[102] O. Fujino, Vanishing theorems for toric polyhedra, in Higher dimensional algebraic varieties
and vector bundles, (S. Mukai, ed.), RIMS Lecture Notes B9, Res. Inst. Math. Sci. (RIMS),
Kyoto, 2008, 8195.
[103] O. Fujino, Three-dimensional terminal toric flips, Cent. Eur. J. Math. 7 (2009), 4653.
[104] O. Fujino and H. Sato, Introduction to the toric Mori theory, Michigan Math. J. 52 (2004),
649665.
[105] W. Fulton, Introduction to Toric Varieties, Annals of Math. Studies 131, Princeton Univ. Press,
Princeton, NJ, 1993.
[106] W. Fulton, Young Tableaux, with Applications to Representation Theory and Geometry, Cambridge Univ. Press, Cambridge, 1997.
822
Bibliography
[107] W. Fulton, Intersection theory, second edition, Ergebnisse der Mathematik und ihrer Grenzgebiete, 3. Folge, Band 2, Springer, Berlin, 1998.
[108] W. Fulton, Equivariant Cohomology in Algebraic Geometry, Eilenberg lectures, Columbia
University, Spring 2007, notes by D. Anderson. Available electronically at www.math.lsa.
umich.edu/~dandersn/eilenberg.
[109] W. Fulton and R. Lazarsfeld, Connectivity and its applications in algebraic geometry, in Algebraic Geometry: Proceedings University of Illinois at Chicago Circle, 1980, (A. Libgober
and P. Wagreich, eds), Lecture Notes in Math. 862, Springer, Berlin, 1981, 2692.
[110] W. Fulton and R. MacPherson, Categorical framework for the study of singular spaces, Mem.
Amer. Math. Soc. 31, (1981).
[111] W. Fulton and B. Sturmfels, Intersection theory on toric varieties, Topology 36 (1997), 335
353.
[112] I. M. Gel fand, M. M. Kapranov and A. V. Zelevinsky, Newton polyhedra of principal Adeterminants, Dokl. Akad. Nauk SSSR 308 (1989), 2023; English translation, Soviet Math.
Dokl. 40 (1990), 278281.
[113] I. M. Gel fand, M. M. Kapranov and A. V. Zelevinsky, Discriminants, Resultants and Multidimensional Determinants, Birkhauser Boston, Boston Basel Berlin, 1994.
[114] E. Gawrilow, M. Joswig, Polymake Version 2.9.8Convex polytopes, polyhedra, simplicial complexes, matroids, fans, and tropical objects, available electronically at wwwopt.
mathematik.tu-darmstadt.de/polymake/doku.php, 1997present.
[115] R. Godement, Theorie ds Faisceaux, Hermann, Paris, 1964.
[116] R. Goldman and R. Krasauskas, eds., Topics in Algebraic Geometry and Geometric Modeling
(Proceedings of AGGM, Vilnius, Lithuania), Contemp. Math. 334, Amer. Math. Soc., Providence, RI, 2003.
[117] P. Gonzalez Perez and B. Teissier, Embedded resolutions of non necessarily normal affine
toric varieties, C. R. Math. Acad. Sci. Paris 334 (2002), 379382.
[118] M. Goresky, R. Kottwitz and R. MacPherson, Equivariant cohomology, Koszul duality, and
the localization theorem, Invent. Math. 131 (1998), 2583.
[119] M. Goresky and R. MacPherson, Intersection homology theory, Topology 19 (1980), 135162.
[120] M. Goresky and R. MacPherson, Intersection homology theory II, Invent. Math. 72 (1983),
77129.
[121] R. Goward, A simple algorithm for principalization of monomial ideals, Trans. Amer. Math.
Soc. 357 (2005), 48054812.
[122] H. Grauert and R. Remmert, Coherent Analytic Sheaves, Grundlehren der mathematischen
Wissenschaften 265, Springer, Berlin, 1984.
[123] D. Grayson and M. Stillman, Macaulay2 Version 1.4Algebraic geometry and commutative
algebra, available electronically at www.math.uiuc.edu/Macaulay2, 1992present.
[124] M. Greenberg and J. Harper, Algebraic Topology: A First Course, Benjamin/Cummings, Reading, MA, 1981. Reprinted by Westview Press, Boulder, CO.
[125] P. Griffiths and J. Harris, Principles of Algebraic Geometry, Wiley, New York, 1978
[126] A. Grothendieck, Revetements e tales et groupe fondamental (SGA 1), Seminaire de geometrie
algebrique du Bois Marie 196061 (with two papers by M. Raynaud), Lecture Notes in
Math. 224, Springer, Berlin, 1971. Updated and annotated reprint, Documents Mathematiques
(Paris) 3, Societe Mathematique de France, Paris, 2003.
ements de geometrie algebrique (rediges avec la collaboration de Jean
[127] A. Grothendieck, El
Dieudonne), II. Etude globale e lementaire de quelques classes de morphismes, Inst. Hautes
tudes Sci. Publ. Math. 8 (1961), 5222.
Bibliography
823
[128] B. Grunbaum, Convex Polytopes, Wiley Interscience, New York, 1967. Second edition prepared by V. Kaibel, V. Klee and G. Ziegler, Springer, 2003.
[129] C. Haase and I. Melnikov, The reflexive dimension of a lattice polytope, Ann. Comb. 10 (2006),
211217.
[130] J. Harris, Algebraic Geometry: A First Course, Graduate Texts in Math. 133, Springer, New
York, 1992.
[131] R. Hartshorne, Algebraic Geometry, Graduate Texts in Math. 52, Springer, New York, 1977.
[132] R. Hartshorne, Stable reflexive sheaves, Math. Ann. 254 (1980), 121176.
[133] B. Hassett, Introduction to Algebraic Geometry, Cambridge Univ. Press, Cambridge, 2007.
[134] B. Hassett and Y. Tschinkel, Universal torsors and Cox rings, in Arithmetic of HigherDimensional Algebraic Varieties, (B. Poonen and Y. Tschinkel, eds), Progress in Math. 226,
Birkhauser Verlag, Basel Boston Berlin, 2004, 149173.
[135] A. Hatcher, Algebraic Topology, Cambridge Univ. Press, Cambridge, 2002.
[136] A. Hatcher, Spectral Sequences in Algebraic Topology, draft available electronically at www.
math.cornell.edu/~ hatcher/SSAT/SSATpage.html.
[137] T. Hausel and B. Sturmfels, Toric hyperKahler varieties, Doc. Math. 7 (2002), 495534.
[138] H. Hauser, The Hironaka Theorem on resolution of singularities (Or: a proof that we always
wanted to understand), Bull. Amer. Math. Soc. 40 (2003), 323403.
[139] D. Hensley, Continued Fractions, World Scientific, Singapore, 2006.
[140] R. Hemmecke, R. Hemmecke and P. Malkin, 4ti2 Version 1.2Computation of Hilbert bases,
Graver bases, toric Grobner bases, and more, available electronically at www.4ti2.de, 2005.
[141] M. Hering, A. Kuronya and S. Payne, Asymptotic cohomological functions of toric divisors,
Adv. Math. 207 (2006), 634645.
[142] L. Hille and M. Perling, A counterexample to Kings conjecture, Compos. Math. 142 (2006),
15071521.
[143] H. Hironaka, Resolution of singularities of an algebraic variety over a field of characteristic
zero I-II, Ann. Math. 79 (1964), 109326.
[144] F. Hirzebruch, New Methods in Algebraic Geometry, third enlarged edition, Die Grundlehren
der Mathematischen Wissenschaften, Band 131, Springer-Verlag, New York, 1966.
[145] F. Hirzebruch, Hilbert Modular Surfaces, Ens. Math. 19 (1973), 183281.
[146] M. Hochster, Rings of invariants of tori, Cohen-Macaulay rings generated by monomials, and
polytopes, Ann. Math. 96 (1972), 318337.
[147] J. Howald, Multiplier ideals of monomial ideals, Trans. Amer. Math. Soc. 353 (2001), 2665
2671.
[148] S. Hu, Semistable degeneration of toric varieties and their hypersurfaces, Comm. Anal.
Geom. 14 (2006), 5989.
[149] Y. Hu, Combinatorics and quotients of toric varieties, Discrete Comput. Geom. 28 (2002),
151174.
[150] Y. Hu and S. Keel, Mori dream spaces and GIT, Michigan Math. J. 48 (2000), 331348.
[151] K. Hulek, Elementary Algebraic Geometry, Amer. Math. Soc., Providence, RI, 2003.
[152] J. Humphreys, Linear Algebraic Groups, Graduate Texts in Math. 21, Springer, New York,
1975.
[153] N. Ilten, toriccodes.m2Macaulay2 package for divisorial polytopes, toric codes
and T-codes, available electronically at people.cs.uchicago.edu/~ nilten/files/
toriccodes.m2, 2008present.
824
Bibliography
[154] N. Ilten, Deformations of smooth toric surfaces, Manuscripta Math. to appear. Available electronically at arXiv:0902.0529 [math.AG].
[155] M.-N. Ishida, Polyhedral Laurent series and Brions equalities, Internat. J. Math. 1 (1990),
251265.
[156] Y. Ito, Minimal resolution via Grobner basis, in Algebraic Geometry in East Asia, (A.
Ohbuchi, K. Konno, S. Usui, A. Moriwaki and N. Nakayama, eds.), World Sci. Publ., River
Edge, NJ, 2002, 165174.
[157] Y. Ito, Special McKay Correspondence, in Geometry of toric varieties, (L. Bonavero and M.
Brion, eds.), Semin. Congr. 6, Soc. Math. France, Paris, 2002, 213225
[158] S. Iyengar, G. Leuschke, A. Leykin, C. Miller, E. Miller, A. Singh and U. Walther, Twentyfour Hours of Local Cohomology, Grad. Stud. Math. 87, Amer. Math. Soc., Providence, RI,
2007.
[159] N. Jacobson, Basic Algebra II, Freeman, San Francisco, 1980.
[160] K. Jaczewski, Generalized Euler sequences and toric varieties, in Classification of Algebraic
Varieties (LAquila, 1992), (C. Ciliberto, E. L. Livorni and A. Sommese, eds.), Contemp.
Math. 162, Amer. Math. Soc., Providence, RI, 1994, 227247.
[161] A. Jensen, Gfan Version 0.4plusGrobner fans and tropical varieties, available electronically
at www.math.tu-berlin.de/~ jensen/software/gfan/gfan.html, 2005present.
[162] A. Jordan, Homology and Cohomology of Toric Varieties, Ph.D. thesis, University of
Konstanz, 1998. Available electronically at citeseerx.ist.psu.edu/viewdoc/summary?
doi=10.1.1.45.7648.
[163] D. Joyner, GAP toric Version 1.6toric varieties and combinatorial geometry in GAP, available electronically at www.opensourcemath.org/toric, 2005present.
[164] M. Kapranov, B. Sturmfels and A. Zelevinsky, Quotients of toric varieties, Math. Ann. 290
(1991), 643655.
[165] A. Kasprzyk, Toric Fano 3-folds with terminal singularities, Tohoku Math. J. 58 (2006), 101
121.
[166] A. Kasprzyk, Canonical toric Fano threefolds, Canad. J. Math., to appear. Available electronically at arXiv:0806.2604 [math.AG].
[167] A. Kasprzyk, Bounds on fake weighted projective space, Kodai Math. J. 32 (2009), 197208.
[168] A. Kasprzyk, M. Kreuzer and B. Nill, On the combinatorial classification of toric log del
Pezzo surfaces, LMS J. Comput. Math. 13 (2010), 3346.
[169] A. Katsabekis and A. Thoma, Toric sets and orbits on toric varieties, J. Pure Appl. Algebra
181 (2003), 7583.
[170] K. Kaveh, Note on cohomology rings of spherical varieties and volume polynomial, preprint,
2010. Available electronically at arXiv:math/0312503 [math.AG].
[171] Y. Kawamata, Crepant blowing-up of 3-dimensional canonical singularities and its application to degeneration of surfaces, Ann. of Math. 127 (1988), 93163.
[172] G. Kempf, F. Knudsen, D. Mumford and B. Saint-Donat, Toroidal Embeddings I, Lecture
Notes in Math. 339, Springer, Berlin, 1973.
[173] A. G. Khovanskii, Newton polyhedra and toroidal varieties, Funktsional. Anal. i Prilozhen.
11 (1977), 5664; English translation, Functional Anal. Appl. 11 (1977), 289296.
[174] A. G. Khovanskii and A. V. Pukhlikov, The Riemann-Roch theorem for integrals and sums of
quasipolynomials on virtual polyhedra, Algebra i Analiz 4 (1992), 188216, English translation, St. Petersburg Math J. 4 (1993), 789812.
[175] F. Kirwan and J. Woolf, An Introduction to Intersection Homology Theory, second edition,
Chapman & Hall/CRC, Boca Raton, FL, 2006.
Bibliography
825
[176] S. Kleiman, Toward a numerical theory of ampleness, Ann. of Math. 84 (1966), 293344.
[177] P. Kleinschmidt, A classification of toric varieties with few generators, Aequationes Math. 35
(1988), 254266.
[178] A. Klyachko, Equivariant vector bundles over toric varieties, Izv. Akad. Nauk SSSR Ser. Mat.
53 (1989),10011039, 1135; English translation, Math. USSR-Izv. 35 (1990), 337375.
[179] J. Kollar and S. Mori, Birational Geometry of Algebraic Varieties, Cambridge Univ. Press,
Cambridge, 1998.
[180] S. Krantz, Geometric Function Theory: Explorations in Complex Analysis, Birkhauser
Boston, Boston Basel Berlin, 2005.
[181] A. Kresch, Gromov-Witten invariants of a class of toric varieties, Michigan Math. J. 48 (2000),
369391.
[182] M. Kreuzer and H. Skarke, PALP Version 1.1Lattice polytopes with a view towards physics,
available electronically at hep.itp.tuwien.ac.at/~ kreuzer/CY/CYpalp.html, 2002
present.
[183] M. Kreuzer and H. Skarke, PALP: A package for analyzing lattice polytopes with applications
to toric geometry, Comput. Phys. Comm. 157 (2004), 87106.
[184] E. Kunz (with contributions by D. Cox and A. Dickenstein), Residues and Duality for Projective Algebraic Varieties, University Lecture Series 47, Amer. Math. Soc., Providence, RI,
2008.
[185] R. Laterveer, Linear systems on toric varieties, Tohoku Math. J. 48 (1996), 451458.
[186] R. Lazarsfeld, Positivity in Algebraic Geometry, Vols. I and II, Springer, New York, 2004.
[187] Y. Liu, L. E. Trotter and G. Ziegler, On the height of the minimal Hilbert basis, Results Math.
23 (1993), 374376.
[188] J.-L. Loday, Realization of the Stasheff polytope, Arch. Math. 83 (2004), 267278.
[189] I. G. Macdonald, The volume of a lattice polyhedron, Proc. Camb. Phil. Soc. 59 (1963), 719
726.
[190] D. Maclagan and G. Smith, Multigraded Castelnuovo-Mumford regularity, J. Reine Angew.
Math. 571 (2004), 179212.
[191] P. Mani, Spheres with few vertices, Journal of Comb. Theory Ser. A 13 (1992), 346352.
[192] E. Martnez-Moro, C. Munuera and D. Ruano, eds., Advances in Algebraic Geometry Codes,
Ser. Coding Theory Cryptol. 5, World Sci. Publ., Hackensack, NJ, 2008.
[193] E. Materov, The Bott formula for toric varieties, Mosc. Math. J. 2 (2002), 161182, 200.
[194] K. Matsuki, Introduction to the Mori Program, Universitext, Springer, New York, 2002.
[195] H. Matsumura, Commutative Ring Theory, Cambridge Univ. Press, Cambridge, 1986.
[196] A. Mavlyutov, Cohomology of complete intersections in toric varieties, Pacific J. Math. 191
(1999), 133144.
[197] A. Mavlyutov, Semiample hypersurfaces in toric varieties, Duke Math. J. 101 (2000), 85116.
[198] A. Mavlyutov, Cohomology of rational forms and a vanishing theorem on toric varieties, J.
Reine Angew. Math. 615 (2008), 4558.
[199] J. McCleary, A Users Guide to Spectral Sequences, Cambridge Univ. Press, Cambridge, 2001.
[200] M. McConnell, The rational homology of toric varieties is not an combinatorial invariant,
Proc. Amer. Math. Soc. 104 (1989), 986-991.
[201] P. McMullen, On zonotopes, Trans. Amer. Math. Soc. 159 (1971), 91109.
[202] D. McDuff, Symplectic embeddings of 4-dimensional ellipsoids, J. Topol. 2 (2009), 122.
826
Bibliography
[203] L. Mihalcea, Positivity in equivariant quantum Schubert calculus, Am. J. Math. 128 (2006),
787803.
[204] E. Miller and B. Sturmfels, Combinatorial Commutative Algebra, Graduate Texts in Math.
227, Springer, New York, 2005.
[205] K. Miyake and T. Oda, Almost homogeneous algebraic varieties under algebraic torus action,
in ManifoldsTokyo 1973 (Proc. Internat. Conf., Tokyo, 1973), (A. Hattori, ed.), Univ. Tokyo
Press, Tokyo, 1975, 373381.
[206] T. Morita, Hilbert schemes of finite abelian group orbits and Grobner fans, Hokkaido Math.
J. 38 (2009), 249265.
[207] D. Mumford, Algebraic Geometry I: Complex Projective Varieties, Springer, Berlin, 1976.
[208] D. Mumford, The Red Book of Varieties and Schemes, second expanded edition, Lecture Notes
in Math. 1358, Springer, Berlin, 1999.
[209] D. Mumford, J. Fogarty and F. Kirwan, Geometric Invariant Theory, third edition, Ergebnisse
der Mathematik und ihrer Grenzgebiete 34, Springer, Berlin, 1994.
[210] J. Munkres, Elements of Algebraic Topology, Addison-Wesley, Reading, MA, 1984. Reprinted
by Westview Press, Boulder, CO.
[211] M. Mustata , Local cohomology at monomial ideals, J. Symbolic Comput. 29 (2000), 709720.
[212] M. Mustata , Vanishing theorems on toric varieties, Tohoku Math. J. 54 (2002), 451470.
[213] Y. Namikawa, Toroidal compactification of Siegel spaces, Lecture Notes in Math. 812,
Springer, Berlin, 1980.
[214] B. Nill, Gorenstein toric Fano varieties, Manuscripta Math. 116 (2005), 183210.
[215] B. Nill, Volume and lattice points of reflexive simplices, Discr. Comp. Geom. 37 (2007), 301
320.
[216] M. bro, An algorithm for the classification of smooth Fano polytopes, preprint, 2007. Available electronically at arXiv:0704.0049 [math.CO].
[217] T. Oda, Torus Embeddings and Applications. Based on joint work with Katsuya Miyake, Tata
Institute of Fundamental Research, Bombay; Springer, Berlin, 1978.
[218] T. Oda, Convex Bodies and Algebraic Geometry, Ergebnisse der Mathematik und ihrer Grenzgebiete, 3. Folge, Band 15, Springer, New York, 1988.
[219] T. Oda, Geometry of toric varieties, in Proceedings of the Hyderabad Conference on Algebraic
Groups, (S. Ramanan, C. Musili and N. Mohan Kumar, eds.), Manoj Prakashan, Madras, 1991,
407440.
[220] T. Oda, Simple convex polytopes and the strong Lefschetz theorem, J. Pure Appl. Algebra 71
(1991), 265286.
[221] T. Oda, Recent developments in the theory of toric varieties, Sugaku 46 (1994), 323335;
English translation, Recent topics on toric varieties, in Selected papers on number theory and
algebraic geometry, Amer. Math. Soc. Transl. Ser. 2, 172, Amer. Math. Soc., Providence, RI,
1996, 7791.
[222] T. Oda and H. S. Park, Linear Gale transforms and Gelfand-Kapranov-Zelevinskij decompositions, Tohoku Math. J. 43 (1991), 375399.
[223] S. Ogata and K. Nakagawa, On generators of ideals defining projective toric varieties,
Manuscripta Math. 108 (2002), 33-42.
[224] S. Payne, Equivariant Chow cohomology of toric varieties, Math. Res. Lett. 13 (2006), 2941.
[225] S. Payne, Moduli of toric vector bundles, Compos. Math. 144 (2008), 11991213.
[226] M. Perling, Divisorial cohomology vanishing on toric varieties, preprint, 2009. Available electronically at arXiv:0711.4836v2 [math.AG].
Bibliography
827
[227] C. Peters and J. Steenbrink, Mixed Hodge Structures, Springer, New York, 2008.
[228] J. Pfeifle and J. Rambau, Computing triangulations using oriented matroids, in Algebra, Geometry, and Sofware Systems, (M. Joswig and N. Takayma, eds.), Springer, Berlin, 2003,
4975.
[229] J. Pommersheim, Toric varieties, lattice points and Dedekind sums, Math. Ann. 295 (1993),
124.
[230] B. Poonen and F. Rodriguez-Villegas, Lattice polygons and the number 12, Am. Math.
Monthly 107 (2000), 238250.
[231] P. Popescu-Pampu, The geometry of continued fractions and the topology of surface singularities, in Singularities in geometry and topology 2004, (J.-P. Brasselet and T. Suwa, eds.), Adv.
Stud. Pure Math., 46, Math. Soc. Japan, Tokyo, 2007, 119195.
[232] N. Proudfoot, Geometric invariant theory and projective toric varieties, in Snowbird Lectures
in Algebraic Geometry, (R. Vakil, ed.), Contemp. Math. 388, Amer. Math Soc., Providence,
RI, 2005, 161167.
[233] J. Rambau, TOPCOM Version 0.16.0Triangulations of point configurations, available electronically at www.rambau.wm.uni-bayreuth.de/TOPCOM/, 1987present.
[234] J. Rambau, TOPCOM: Triangulations of point configurations and oriented matroids, in Mathematical Software - ICMS 2002, (A. Cohen, X. Gao and N. Takayama, eds.), World Scientific
2002, 330340.
[235] M. Reid, Canonical 3-folds, in Journees de Geometrie Algebrique dAngers, Juillet 1979/Algebraic Geometry, Angers, 1979, (A. Beauville, ed.), Sijthoff & Noordhoff, Alphen aan den
Rijn, 1980, 273310.
[236] M. Reid, Decomposition of Toric Morphisms, in Arithmetic and Geometry, (M. Artin and J.
Tate, eds.), Progress in Math. 36, Birkhauser Verlag, Basel Boston Berlin, 1983, 395418.
[237] M. Reid, Review of Geometry of convex bodies and algebraic geometry, Sugaku 33 (1981),
120133 by T. Oda, in Mathematical Reviews, MR0631487 (83f:14010).
[238] M. Reid, Undergraduate Algebraic Geometry, LMS Student Texts 12, Cambridge Univ. Press,
Cambridge, 1989.
[239] E. Reyes, R. Villarreal and L. Zarate, A note on affine toric varieties, Linear Algebra Appl.
318 (2000), 173179.
[240] J. Richter-Gebert, B. Sturmfels and T. Theobald, First steps in tropical geometry, in Idempotent Mathematics and Mathematical Physics, (G. Litvinov and V. Maslov, eds.), Contemp.
Math. 377, Amer. Math. Soc., Providence, RI, 2003.
[241] R. T. Rockafellar, Convex Analysis, Princeton Univ. Press, Princeton, NJ, 1970.
[242] J. J. Rotman, Advanced Modern Algebra, Prentice-Hall, Upper Saddle River, NJ, 2002.
[243] I. Satake, On the arithmetic of tube domains (blowing-up of the point at infinity), Bull. Amer.
Math. Soc. 79 (1973), 10761094.
[244] H. Sato, Toward the classification of higher-dimensional toric Fano varieties, Tohoku Math.
J. 52 (2000), 383413.
[245] I. Shafarevich, Basic Algebraic Geometry, Volumes 1 and 2, Springer, New York, 1994 and
1996.
[246] H. Schenck, Computational Algebraic Geometry, Cambridge Univ. Press, Cambridge, 2003.
[247] J.-P. Serre, Faisceaux algebriques coherents, Ann. Math. 61, (1955), 197278.
[248] J.-P. Serre, Geometrie algebrique et geometrie analytique, Ann. Inst. Fourier (Grenoble) 6
(1956), 142.
[249] G. Shephard, Decomposable convex polyhedra, Mathematika 10 (1963), 8995.
828
Bibliography
[250] A. Simis, W. Vasconcelos and R. Villarreal, On the ideal theory of graphs, J. Algebra 167
(1994), 389416.
[251] N. J. A. Sloane, On-line Encyclopedia of Integers Sequences, 2010. Published electronically
at the web site oeis.org.
[252] G. G. Smith, Macaulay2 toric varieties package, available electronically at www.math.uiuc.
edu/Macaulay2, 2010.
[253] K. Smith, L. Kahanpaa , P. Kekalainen and W. Traves, An Invitation to Algebraic Geometry,
Universitext, Springer, New York, 2000.
[254] F. Sottile, Toric ideals, real toric varieties, and the algebraic moment map, in Topics in Algebraic Geometry and Geometric Modeling (Proceedings of AGGM, Vilnius, Lithuania), (R.
Goldman and RKrasuaskas

eds), Contemp. Math. 334, Amer. Math. Soc., Providence, RI,
2003, 225240. Corrected version available electronically at arXiv:math/0212044 [math.AG].
[255] E. Spanier, Algebraic Topology, McGraw-Hill, New York, 1966. Reprinted by Springer, New
York, 1994.
[256] T. A. Springer, Linear Algebraic Groups, second edition, Progress in Math. 9, Birkhauser
Verlag, Basel Boston Berlin, 1998.
[257] R. Stanley, The number of faces of a simplicial convex polytope, Adv. Math. 35 (1980), 236
238.
[258] R. Stanley, Generalized H-vectors, intersection cohomology of toric varieties, and related
results, in Commutative algebra and combinatorics (Kyoto 1985), (M. Nagata and H. Matsumura, eds.), Adv. Stud. Pure Math. 11, North-Holland, Amsterdam, 1987, 187213.
[259] H. Stark, An Introduction to Number Theory, MIT Press, Cambridge, MA, 1978.
[260] J. Steenbrink, Intersection form for quasi-homogeneous singularities, Compositio Math. 34
(1977), 211223.
[261] J. Steenbrink, Mixed Hodge structure on the vanishing cohomology, in Real and complex
singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math., Oslo, 1976), Sijthoff
and Noordhoff, Alphen aan den Rijn, 1977, 525563.
[262] W. Stein et.al., Sage Mathematics Software, Version 2.6Free open-source mathematics software system, available electronically at www.sagemath.org, 2004present.
[263] R. S. Strichartz, The Way of Analysis, Jones and Bartlett, Boston, MA, 1995.
[264] B. Sturmfels, Grobner Bases and Convex Polytopes, University Lecture Series 8, Amer. Math.
Soc., Providence, RI, 1996.
[265] H. Sumihiro, Equivariant completion, I, II, J. Math. Kyoto Univ. 14 (1974), 128; 15 (1975),
573605.
[266] I. Swanson and C. Huneke, Integral Closure of Ideals, Rings, and Modules, LMS Lecture
Note Series 336, Cambridge Univ. Press, Cambridge, 2006.
[267] B. Tessier, Du theorème de lindex de Hodge aux inegalites isoperimetriques, C. R. Acad. Sci.
Paris Ser. A-B 288 (1979), A287A289.
[268] B. Tessier, Monomial ideals, binomial ideals, polynomial ideals, in Trends in Commutative
Algebra, (L. Avramov, M. Green and C. Huneke, eds.), Math. Sci. Res. Inst. Publ. 51, Cambridge Univ. Press, Cambridge, 2004, 211246.
[269] M. Thaddeus, Toric quotients and flips, in Topology, Geometry, and Field Theory, (K. Fukaya,
M. Furuta, T. Kohno and D. Kotschick, eds.), World Scientific Publ., River Edge, NJ, 1994,
193213.
[270] H. Thompson, Comments on toric varieties, preprint, 2003. Available electronically at
arXiv:math/0310336 [math.AG].
Bibliography
829
[271] T. tom Dieck, Transformation Groups, de Gruyter Studies in Mathematics 8, Walter de Gruyter
& Co., Berlin, 1987.
[272] B. Totaro, The Chow ring of a classifying space, in Algebraic K-theory (Seattle, WA, 1997),
(W. Raskind and C. Weibel, eds.), Proc. Sympos. Pure Math. 67, Amer. Math. Soc., Providence, RI, 1999, 249281.
[273] K. Ueno, Algebraic Geometry 2, Amer. Math. Soc., Providence, RI, 2001.
[274] R. Villarreal, Monomial Algebras, Marcel Dekker, Inc., New York, 2001.
[275] C. von Renesse, Combinatorial Aspects of Toric Varieties, Ph. D. Thesis, University of
Massachusetts, 2007. Available electronically at /www.wsc.ma.edu/renesse/research/
UMTHESIS.pdf.
[276] C. Weibel, An Introduction to Homological Algebra, Cambridge Univ. Press, Cambridge,
1994.
[277] E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B403 (1993), 159222.
[278] J. Wodarczyk, Decomposition of birational toric maps in blow-ups & blow-downs, Trans.
Amer. Math. Soc. 349 (1997), 373411.
[279] J. Wodarczyk, Toroidal varieties and the weak factorization theorem, Invent. Math. 154
(2003), 223331.
[280] O. Zariski and P. Samuel, Commutative Algebra, Volume II, Van Nostrand, Princeton, 1960.
Corrected reprint, Springer, New York, 1976.
[281] G. Ziegler, Lectures on Polytopes, Graduate Texts in Math. 152, Springer, 1995.
Index
absolutely ample, 345

abstract variety, 97, 99
adjunction formula, 366, 503
affine
piece of a variety, 51, 52
scheme, 4, 104
toric variety, 12, 18
of a cone, 30
variety, 3
affine cone
of a projective toric variety, 56
of a projective variety, 50
affine hyperplane, 63
affine morphism, 397
affine semigroup, 16
Alexander duality, 420, 673
Alexandrov-Fenchel inequality, 792
algebraic action, 12, 106, 199
algebraic cycle, 612
rationally equivalent to zero, 612
algebraic statistics, 796
algebro-geometric imperialism, 794
almost geometric quotient, 201, 202, 210, 682
almost homogeneous variety, 791
ample, see Cartier divisor, ample
absolutely, 345
relatively with respect to f , 345
analytic variety, 547
annihilator, 416
arithmetically Cohen-Macaulay (aCM), 415
associahedron, 743, 745, 749, 800
augmentation ideal, 654
-basis, 734
barycentric subdivision, 525
basepoint free, see also sheaf, generated by global

sections
Cartier divisor, see Cartier divisor, basepoint
free
subspace of global sections, 257
basic simplex, 68
Bernoulli numbers, 626, 663
Bertini theorem, 509
Betti numbers, 587, 589, 621
binomial, 16
birational transform of a divisor, 557, 768, 769
Birkhoff polytope, 64
bivariant intersection theory, 675
blowup, 98, 105, 111, 130, 131, 174, 176, 188,
190, 277, 286, 319, 591
of an ideal, 535
Borel construction, 596, 598, 599
for a torus, 596
Borel-Moore homology, 672
Bott formula, 439, 455
branches, 562
Brions equalities, 632, 636, 644, 664, 665
for a polytope, 637
bunch of cones, 738
Calabi-Yau variety, 529
cancellative semigroup, 22
canonical bundle, 356
canonical class, 356
canonical divisor, 356, 373
of a toric variety, 366, 791
relative, 541
canonical module
of an affine toric variety, 371
of the total coordinate ring, 369
831
832
canonical sheaf, 356, 365

of a toric variety, 366
canonical singularities, 549
Caratheodorys theorem, 69, 537
Cartan matrix, 500
Cartesian diagram, 675
Cartier data, 181, 193, 791
Cartier divisor, 160
ample, 268, 269, 271, 305, 788
basepoint free, 258, 262, 266, 269, 291, 329,
726
big, 427, 656
f -ample, 331, 772
f -very ample, 331
local data of, 160
nef, 291, 305, 726
numerical equivalence of, 292
numerically equivalent to zero, 292
of a polyhedron, 322
of a polytope, 182
on an irreducible variety, 257
torus-invariant, see torus-invariant, Cartier
divisor
very ample, 268, 269
Cech
cochains, 389
Cech
cohomology, 389391
Cech
Complex
toric, 398
Cech
complex, 389, 390
ceiling , 424
central hyperplane arrangement, 285
chamber, 718, 725
character, 11
generic, 707, 718
Chern character, 623, 625, 629, 631, 809
equivariant, 642
Chern class, 607, 624
of a locally free sheaf, 624
of a smooth variety, 624
total, 624
Chern roots, 626, 631
Chevalleys theorem, 144
Chow group, 612
Chow polytope, 744
Chow ring, 613
rational, 613
circuits, 701, 751
oriented, 751
class group, 161, 350
classical topology, 4, 50, 102, 113, 139, 142
closed half-space, 25, 63
CoCoA, 797
codimension of a prime ideal, 157
coding theory, 796
Index
Cohen-Macaulay variety, 394

toric, 415, 790
coherent sheaf, 166
cohomology class, 593, 673
equivariant, 601, 608
refined, 593, 673
combinatorially equivalent polytopes, 65
compact, 113, 139, 142
compatibility conditions, 97, 252
compatible map of lattices, 125
complete fan, 113, 139, 142, 193, 788
complete linear system, 261
complete variety, 141, 143
toric, 142
complex, 388
exact sequence of, 388, 389
Koszul, 403, 408
map of, 388
computer algebra
4ti2, see 4ti2
GAP, see GAP
Gfan, see Gfan
LattE, see LattE
Macaulay2, see Macaulay2
Magma, see Magma
Normaliz, see Normaliz
PALP, see PALP
Polymake, see Polymake
Sage, see Sage
Singular, see Singular
TOPCOM, see TOPCOM
cone
convex polyhedral, see polyhedral cone
Gorenstein, 379
of a polyhedron, 319, 325
of a polytope, 24, 70, 86
polyhedral, see polyhedral cone
rational polyhedral, see rational polyhedral cone
reflexive Gorenstein, 379
simplicial, see simplicial cone
smooth, see smooth cone
strongly convex, see strongly convex cone
connecting homomorphism, 388
conormal sheaf, 355
constant sheaf, 396
constructible set, 123
continued fraction
Felix Klein, 476
Hirzebruch-Jung, 468, 469, 473
ordinary, 475, 476
contractible, 396
convergents, 469, 475
convex
function, 264
hull, 24
polyhedral cone, see polyhedral cone
convex support of full dimension, 265
Index
coordinate ring, 3, 4
homogeneous, 49, 372
total, see total coordinate ring
cotangent bundle, 352
cotangent sheaf, 352
Cousin problem, 392
Cox ring, 795
Cremona group, 787
crepant morphism, 553
crepant resolution, 527, 529
cup product, 593
D-flop, 782, 783
Dedekind sums, 653
deformation retract, 396
degree
of a divisor on a curve, 287
of a line bundle on a curve, 288
Dehn-Sommerville equations, 436, 438, 439, 590,
621
depth, 394, 416
derivation, 357
torus-invariant, 364
universal, 357
derived functor, 389, 444, 445
determinantal variety, 13
diagonal map, 102
dimension at a point, 6, 101
dimension of a cone, 24
dimension of a module, 416
dimension of a polytope, 63
dimension of a ring, 156
direct image, 166, 392
higher, 393
left exact, 364
direct limit, 100, 245
directed set, 182, 245
directed system, 245
discrepancy, 549
discrete valuation, 155
discrete valuation ring (DVR), 155
distinguished point, 116, 118, 130, 135
division algorithm, 475
modified, 460, 467
divisor
anticanonical, 379
toric, 380
canonical, see canonical divisor
Cartier, see Cartier divisor
linear equivalence of, 160, 350
locally principal, see locally principal divisor
nef, see Cartier divisor, nef
numerically effective, see Cartier divisor, nef
of a character, 171
833
of a rational function, 159

of poles, 160
of zeros, 160
principal, 159
Weil, see Weil divisor
divisorial wall, 750
double dual, 347, 350
double point, 462
doubly-stochastic matrix, 64
Du Val singularity, 462, 501
dual cone, 24, 799
dual face, 26
dual polytope, 65
dual sheaf, 255, 347, 350
dualizing sheaf, 356
dualizing complex, 395
Dynkin diagram, 462, 467, 500
edge, 25, 63
edge ideal of a graph, 539
effective divisor, 159
Ehrhart polynomial, 432, 508, 655, 791, 805
of a polygon, 433, 434
of a polytopal complex, 434, 442
Ehrhart quasipolynomial, 443
Ehrhart reciprocity, 432, 436, 791
elementary flip, see flip, elementary
elementry symmetric polynomial, 626
Enriques-Kodaira classification, 762
equalizer, 95
equivariant cohomology, 596
equivariant Euler class, 648, 654
equivariant Gysin map, 646, 675, 676
equivariant integral, 643, 676
decomposition of, 647
equivariant map, 41, 126
equivariant Riemann-Roch, 623
for a toric variety, 641
proof of, 648
simplicial, 651, 652
equivariant Todd class, 642
simplicial, 652
equivariant vector bundle, 607
equivariantly formal, 599, 606
Euclidean algorithm, 468, 475
modifed, 468
Euler characteristic
decomposition in equivariant case, 643, 645
equivariant, 641
of a sheaf, 430, 440, 623
of a topological space, 442, 505, 586, 625
Euler class, 624
Euler relation, 365
generalized, 365
Euler sequence, 363, 365
Index
834
generalized, 364
toric, 363
Euler-Maclaurin formula, 663, 667
e ventail, 787
exact sequence
long, 389
of complexes, 388
of sheaves, 248
exceptional curve of the first kind, 498
exceptional divisor, 464
exceptional locus, 521, 762
exceptional set, 207
of a character, 698
expected dimension, 683, 704, 710, 720
Ext, 395, 445
exterior product, see wedge product
extremal wall, see wall, extremal
extremal contraction, 762, 763
divisorial, 764, 766, 774, 776
fibering, 764, 766, 776
flipping, 764, 766, 770, 773, 774, 776
structure of, 764, 766
extremal ray, 295, 306, 727, 762, 770
D-negative, 772, 777
extremal walls, 295
flip, 748, 761, 762

elementary, 760, 770
Proj construction, 772, 774
termination of, 777779, 781
toric, 770
flipping wall, 750, 757
floor , 424
flop, 782, 783
formal power series, 10
formal semigroup module, 632
summable elements of, 633
fractional ideal, 170
sheaf of, 540, 790
full dimensional
convex support, 265
polyhedron, 319
polytope, 64
function field, 100
functor, 388
derived, 389
fundamental class, 673
fundamental group
of a torus, 562
f -ample, see Cartier divisor, f -ample

f -very ample, see Cartier divisor, f -very ample
4ti2, 72, 797
f.r.p.p decomposition, 789
face, 25, 63
face numbers, 435, 620, 793
facet, 25, 63
normal, 26, 64
presentation, 66
virtual, 696, 697
fake weighted projective spaces, 765
fan, 77, 106, 787, 789
complete, see complete fan
degenerate, 278, 690, 719, 720, 749, 766, 777
generalized, 278, 690, 691, 712, 719, 725, 726
infinite, 790
nondegenerate, 278
normal, see normal fan
refinement of, see refinement of a fan
simplicial, see simplicial fan
smooth, see smooth fan
Fano type, 559
toric, 559
Fano variety, see variety, Fano
fiber bundle, 133
fiber product of varieties, 102
Fibonacci number, 22
finite quotient singularities, 113, 547
finitely generated semigroup, 16
fixed component, 730
fixed points, 598, 645
G-Hilbert scheme, 493, 495

(G, )-invariant polynomial, 678, 679
GAGA, 142, 144
Gale duality, 452, 699701, 711, 734, 741
GAP, 797
toric package, 797
gauged linear sigma model, 795
generalized Euler sequence, 624
generalized Gysin map, 630, 672674, see also
equivariant Gysin map
generated by global sections, see sheaf, generated
by global sections
geometric case, 729, 744, 748, 749
primtive, 731
geometric invariant theory, see GIT
geometric modeling, 796
geometric quotient, 200, 202, 210, 682
almost, see almost geometric quotient
Gfan, 488, 797, 802
GIT, 677, 678, see also expected dimension
general quotient, 684, 699
quotient, 682, 692, 725
GKZ cone, 713717, 725, 728
facet of, 749
irrelevant ideal of, 720
GKZ decomposition, 718
global sections, 189
of a sheaf, 166, 248
of a toric sheaf, 190, 193
of a vector bundle, 251
gluing data, 97, 252
Index
good categorical quotient, 198202, 682

Gordans Lemma, 30
Gorenstein Fano variety, see variety, Gorenstein
Fano
Gorenstein ring, 660
Grobner basis, 485, 796
marked, 485
reduced, 485
universal, 485
Grobner fan, 486, 745, 802
resolution of singularities, 490
graded module, 226, 248, 310
shift of, 226
graded polynomial ring, 219, 695
Grassmannian, 254, 260, 699
Grothendieck-Riemann-Roch, 629, 630
Gysin map, 674
Hom sheaf, 249, 255
h-polynomial, 621
hard Lefschetz theorem, 620, 621, 793
Hausdorff topological space, 102
height of a prime ideal, 157
higher direct image, 393
Hilbert basis, 32, 33, 473, 474, 799, 803
Hilbert basis theorem, 3
Hilbert polynomial, 655, 803
of a sheaf, 431, 440
Hirzebruch surface, 112, 175, 191, 263, 267, 273,
303, 338, 368, 408, 425, 454, 498, 509, 595,
600, 602, 611, 629, 638, 662, 696, 704, 731,
733, 787, 806, 807, 809
Hirzebruch-Riemann-Roch, 623, 628, 629, 656,
791, 809
proof of, 649
proof via Khovanskii-Pukhlikov, 669
simplicial, 653
Hodge decomposition, 440
Hodge index theorem, 792
homogeneous coordinate ring, see coordinate ring,
homogeneous
homogeneous coordinates, 49
homogenization, 192
homomorphism of sheaves, 166, 246
injective, 246
sujective, 246
Hopf fibration, 575
Iitaka dimension, 427
image sheaf, 247
index of a Q-Gorenstein toric variety, 555
index of a simplicial cone, see multiplicity of a
simplicial cone
injective resolution, 387, 388
injective sheaf, 387
injectivity lemma, 411, 412, 424, 430
835
inner normal fan, see normal fan

integer programming, 796
integral closure, 5
of an ideal, 537, 540
integral of a cohomology class, 592, 623
equivariant, see equivariant integral
integrally closed ideal, 536
integrally closed ring, 5
intersection cohomology, 620, 621, 793
Betti numbers, 621
intersection homology, 618, 619
intersection product, 288, 593
on a toric surface, 497
on a toric variety, 289, 300
inverse limit, 182
inverse system, 182
invertible sheaf, 168, 254, 255
inward-pointing facet normal, see facet, normal
irreducible components, 97
irreducible variety, 97
irrelevant ideal, 207
of a character, 698
of a GKZ cone, 720
Ishida complex, 378, 423
isomorphism of varieties, 96
Jacobian matrix, 7, 351
Jurkiewicz-Danilov
ring, 593, 595
theorem, 594, 595, 606
k-cycle, see algebraic cycle
Kahler differentials, 351, 352, 354
Kunneth formula, 597
Kawamata log terminal singularities, see klt
singularities
kernel sheaf, 247
Khovanskii-Pukhlikov theorem, 666, 667
proof of, 668
relation to HRR, 669
Kleinschmidts classification theorem, 341, 738,
794
klt singularities, 557, 781
Koszul complex, 403, 408
Krull principal ideal theorem, 161, 169
LattE, 797
lattice, 13
lattice polyhedron, 319
lattice equivalence, 381, 461
lattice ideal, 15
lattice polytope, 66
empty, 555
Laurent polynomial, 5
LDP polygon, 386
index of, 386
left exact, 389
836
Leray-Hirsch theorem, 599

limit of one-parameter subgroup, 115, 116, 139,
142, 144
line bundle, 253, 255, 348, see also Cartier divisor
ample, 268
f -ample, 331
f -very ample, 331
linearized, 678
pullback of, 258
very ample, 268
linear equivalence, see divisor, linear equivalence
of
linearized line bundle, see line bundle, linearized
linearly equivalent divisors, 256
local Cech
complex, 444
local analytic equivalence, 547
local cohomology, 444447
local data, 160, 253, 257
toric, 181
local duality, 798
local ring, 6, 156
at a point, 6, 9, 94, 99
at a prime divisor, 157
localization, 5, 9
homogeneous, 317
localization theorem, 600, 643
locally principal divisor, 160
locally irreducible, 563
locally trival fiber bundle, 134
log canonical singularities, 557, 781
log del Pezzo surface, 386
index of, 386
log minimal model, 782, 783
log pair, 557
log resolution, 526, 541
logarithmic 1-forms, 359, 360
logarithmic poles, 360
long exact sequence, 444
lower hull, 740
Macaulay2, 275, 407, 445, 639, 797, 798, 804,
805
BGG package, 807
BoijSoederberg package, 807
NormalToricVarieties package, 797, 806,
810
Polyhedra package, 275, 799
toriccodes package, 797
Magma, 797
maximal Cohen-Macaulay (MCM), 416
maximal projective crepant partial
desingularization, 529
maximal spectrum, 4
McKay correspondence, 491, 493
McMullen conjecture, 620, 793
minimal generator, 30
minimal model, 762
Index
log, 782, 783

minimal model program, see MMP
minimal surface, 762
Minkowski sum, 65, 192, 318, 792
mirror symmetry, 384, 745, 795
mixed volume, 791, 792
MMP, 762, 781
equivariant, 777
log toric, 781, 783
relative, 777
toric, 772, 776
moment map
algebraic, 571
symplectic, 571, 574
monomial Briancon-Skoda theorem, 540
monomial ideal, 540
Mori cone, 293
of a toric variety, 294, 305, 727
Mori fibration, 762
morphism of varieties, 3, 95
projective, see projective morphism
proper, see proper morphism
moving cone
of a variety, 730
of the secondary fan, 728, 729, 738
multipliciative subset, 9
multiplicity of a simplicial cone, 300, 465, 519,
520
properties of, 519
multiplier ideal, 542, 545
toric, 542, 545
multiplier module, 545
N-Minkowski summand, 283
Nakayamas Lemma, 169
nef cone, 293
of a toric variety, 294, 726, 728, 731
nef divisor, see Cartier divisor, nef and Q-Cartier
divisor, nef
Newton polytope, 188, 744, 791
nilpotents, 4, 8, 9, 48, 103, 105
Noethers theorem, 505, 511, 628
Noetherian, 156
nonassociative product, 744, 745
nonequivariant limit, 599, 601, 623, 642
nonnormal toric variety, 150
nonsingular point, 6
normal
affine toric variety, 37
polyhedron, 322
polytope, 68, 86, 372, 803
toric variety, 86, 107
variety, 5, 100
normal bundle, 355, 631, 648, 654
normal fan
of a polyhedron, 321, 690, 691
Index
of a polytope, 76, 77, 108, 279, 690, 792

normal ring, 5, 156, 157
normal sheaf, 355, 366
Normaliz, 33, 39, 72, 434, 797, 803805
normalization, 5, 151
of an affine toric variety, 39, 152
of an irreducible curve, 288
normalized basis, 460
normalized volume, 433, 655
Nullstellensatz, 3
numerically effective divisor, see Cartier divisor,
nef and Q-Cartier divisor, nef
numerically equivalent
divisors, see Cartier divisor, numerical
equivalence of
proper 1-cycles, see proper 1-cycle, numerical
equivalence of
to zero, see Cartier divisor, numerically
equivalent to zero and proper 1-cycle,
numerically equivalent to zero
one-parameter subgroup, 11
orbifold, 46, 113, 547
orbit closure, 121, 135
Orbit-Cone Correspondence, 119, 789
nonnormal case, 154
order of vanishing, 158
orientation coefficient, 580
oriented circuit, see circuits, oriented
oriented matroids, 701
p-allowable, 619
pair (X, D), 557, 781
PALP, 797
parameters of a cone, 460
partial quotients, 469, 475
p-Ehrhart polynomial, 435, 512
perfect field, 48
permutation matrix, 55
perversity, 619
complementary, 619
maximal, 619
middle, 619
minimal, 619
Picard group, 161, 256, 350
Picks formula, 434
piecewise polynomial functions, 606, 607
Poincare duality, 593, 620, 673
Poincare polynomial, 592
Poincare residue, 360, 361
pointed affine semigroup, 36
polar polytope, 65
pole, 159
polyhedral cone, 23
polyhedron, 190, 318
837
augmented, 711
full dimensional, see full dimensional,
polyhedron
lattice, see lattice polyhedron
normal, see normal, polyhedron
of a character, 687, 688
of a torus-invariant divisor, 190, 266, 328
very ample, see very ample, polyhedron
Polymake, 797
polynomial splines, 607
polytopal complex, 434
polytope, 24, 63
combinatorially equivalent, see combinatorially
equivalent polytopes
full dimensional, see full dimensional, polytope
lattice, see lattice polytope
normal, see normal, polytope
simple, see simple polytope
simplicial, see simplicial polytope
smooth, see smooth polytope
very ample, see very ample, polytope
pre-variety, 102
presheaf, 165, 247
primary decomposition, 161
prime divisor, 157
primitive collection, 304, 340, 772
primitive relation, 305, 340, 772
principal A-determinant, 744
principal divisor, see divisor, principal
principal ideal domain (PID), 156, 157
principalization, 541
product variety, 7, 53, 101, 313
class group of, 175
toric, 47, 90, 111
Proj, 317, 322, 372, 682, 690, 691
projective bundle
of a coherent sheaf, 318
toric, 337, 338
projective morphism, 314, 315, 346
projective space, 49
projective toric variety, 55
projective variety, 49
projective with respect to a line bundle, 314, 345
projectively normal variety, 61, 86, 441, 798
proper 1-cycle, 292
numerical equivalence of, 292
numerically equivalent to zero, 292
proper continuous map, 142, 144
proper face, 25
proper morphism, 143, 144, 315
pseudoeffective cone, 730
Puiseux series, 188
pullback, see line bundle, pullback of
pullback of a torus-invariant Cartier divisor, 281
Index
838
Q-Cartier divisor, 180

nef, 409
Q-divisor, 424
Q-factorial, 46
Q-Fano variety, see variety, Q-Fano
quadratic reciprocity, 792
quasicoherent sheaf, 166
quasicompact, 104
quasiprojective toric variety, 330
quasiprojective variety, 313, 315
quotient
almost geometric, see almost geometric quotient
geometric, see geometric quotient
good categorical, see good categorical quotient
Q-Weil divisor, 424
round down of, 424
round up of, 424
rank
of a coherent sheaf, 348
of a finitely generated abelian group, 524
of a module, 366
rational double point, 462, 464, 467, 500
rational function, 51, 100
rational normal cone, 13, 32, 38, 40, 46, 50, 178,
461, 463
rational normal curve, 50, 57
rational normal scroll, 84, 112
rational polyhedral cone, 29
rational singularities, 546
toric, 546
rationally smooth, 546, 592
ray generator, 29
R-divisor, 424
real projective plane, 72
real torus, 561, 570
recession cone, 318
reduced cohomology, 397
reductive group, 201
Rees algebra, 534
refined cohomology class, see cohomology class,
refined
refinement of a fan, 130, 373
reflexive polygon, 382, 510
reflexive polytope, 81, 87, 380, 381, 529, 795
classification, 384
reflexive sheaf, 168, 347
of rank 1, 348
regular cone, see smooth cone
regular fan, 113
regular local ring, 7, 157
regular map, 93, 98
regular representation, 491
regular sequence, 394
relative case, 727, 733
relative cohomology, 405
relative interior, 27
relatively ample with respect to f , 345

representation ring, 654
representations of finite groups, 491
special, 492
resolution of singularities, 463, 513
crepant, 511, 527
embedded, 530
Grobner fan, 490
Hironaka, 513
log, 526
minimal, 499
projective, 513
SNC, 526
toric, 520
toric surface, 465, 470
restriction of a divisor, 160
restriction of a sheaf, 166
Riemann-Roch, see also
Hirzebruch-Riemann-Roch
curves, 503, 628
surfaces, 504, 628
topological part, 505, 628
ring of invariants, 45
ringed space, 95, 99
rooted binary tree, 744, 745
Sage, 522, 629, 639, 797
PALP package, 797
Polyhedra package, 797
TOPCOM package, 800
ToricVarieties package, 797, 808, 809
saturated affine semigroup, 37
secondary fan, 712, 717, 725, 741, 744, 795
chamber of, see chamber
moving cone of, see moving cone, of the
secondary fan
properties of, 718
secondary polyhedron, 744
secondary polytope, 741, 743, 744, 749, 800
section
of a sheaf, 165
sectional genus, 509, 794
of a toric surface, 509
Segre embedding, 52
self-intersection, 303
semi-stable reduction, 790
semicomplete toric variety, 734
semigroup, 16
semigroup algebra, 16
of a cone, 30
semigroup homomorphism, 35, 116
semiprojective toric variety, 332, 691, 695, 712
semistable points, 679, 697, 698, 702, 707
separated variety, 102, 204
separating transcendence basis, 48
Index
separation lemma, 28, 107

Serre duality, 395, 420
for a toric variety, 416, 791
MCM sheaves, 416
set-theoretic complete intersection, 22
sheaf, 95
constant, 255
generated by global sections, 249, 258, 346
locally constant, 255
locally free, 250
of 1-forms, 352
of OX -modules, 165, 245
of p-forms, 355
of a graded module, 227, 248, 310
of a torus-invariant divisor, 189
global sections of, 189, 190, 193
of sections of a vector bundle, 252
of Zariski p-forms, 355
sheaf cohomology, 387, 388, 391
long exact sequence, 389
toric, 402, 447, 809
toric surface, 406
sheafification, 247
sheet, see spectral sequence, sheet of
simple normal crossings, see SNC divisor
simple polytope, 64, 434
simplex, 64
simplicial cone, 30
multiplicity of, see multiplicity of a simplicial
cone
simplicial fan, 113
simplicial polytope, 64
simplicial toric variety, 180
simplicialization, 518, 520
Singular, 797, 804
singular cohomology, 395
of a simplicial toric variety, 588
reduced, 397
relative, 405
ring, see singular cohomology ring
with compact supports, 582
singular cohomology ring, 396
of a toric variety, 594, 595, 660, 791
singular locus, 163, 513
of a normal variety, 163
singular point, 6
of a toric surface, 459
singularities, see also toric singularities
canonical, see canonical singularities
finite quotient, see finite quotient singularities
klt, see klt singularities
log canonical, see log canonical singularities
terminal, see terminal singularities
839
small contraction, 762

Smith normal form, 173
smooth cone, 30, 40
smooth fan, 113
smooth Fano polytope, 385
smooth locus, 513
smooth point, 6, 101
smooth polytope, 87
smooth toric variety, 40, 87, 113, 179
smooth variety, 6, 101
SNC divisor, 360, 525
span of a cone, 24
Spec, 4, 317
spectral sequence, 393, 811
convergent, 812
degenerates, 813
differentials, 811
E1 , 812
E2 , 812
edge homomorphism, 814
Ext, 417, 815
first quadrant, 812
Leray, 393, 814
of a closed cover, 404, 815
of a filtered space, 582, 816
of an open cover, 394, 578, 815
Serre, 598, 599, 816
sheet of, 811
singular cohomology of a toric variety, 582, 587
spherical dual, 574
spherical variety, 661
splitting fan, 133, see also weak splitting, 137
stable points, 679, 702, 707
stalk of a sheaf, 246, 257
standard n-simplex n , 66
Stanley-Reisner ideal, 593, 600
Stanley-Reisner ring, 600, 601
star subdivision, 130, 132, 515, 518, 520, 748, 749
properties of, 515, 517
Stasheff polytope, 743
strictly convex, see support function, of a Cartier
divisor, strictly convex
strong Oda conjecture, 790
strongly convex cone, 28
structure sheaf, 95, 99
analytic, 547
sublattice of finite index, 44
subvariety, 97
sum function, 636
Sumihiros theorem, 108, 151, 789
supplementary cone, 478, 479
isomorphic to dual cone, 480
support function, 183, 713, 725, 791
integral with respect to a lattice, 183
of a Q-Cartier divisor, 408
of a Cartier divisor, 184, 263
convex, 265, 266, 329, 725
Index
840
strictly convex, 271, 329331, 334, 518

of a polytope, 186
support of a divisor, 159
support of a fan, 106, 113, 788
supporting affine hyperplane, 63
supporting half-space, 25
supporting hyperplane, 25
Sylvester sequence, 92
symbolic powers, 545
symmetric algebra, 597
symplectic geometry, 574
symplectic reduction, 575
2-neighborly polytope, 73
tangent bundle, 352
tangent sheaf, 352
tautological bundle, 254, 256, 260
Taylor resolution, 448, 449
tensor product, 8, 48
of sheaves, 249, 255, 350
tent analogy, 264, 267
terminal lemma, 556
terminal singularities, 549, 762
Todd class, 623, 626, 630
equivariant, 642
mock, 653
of a simplicial toric variety, 653
of a smooth toric variety, 627, 809
Todd operators, 664
Todd polynomial, 627
TOPCOM, 797, 800
topological pseudomanifold, 618
torhom, 586
toric Chow lemma, 275
toric cone theorem, 294, 727
toric fibration, 134
toric ideal, 15, 796
toric Kleiman criterion, 291
toric minimal model progam, see MMP, toric
toric morphism, 41, 42, 125, 135
projective, 330, 334, 790
proper, 144, 790
toric set, 21
toric singularities
canonical, 550552, 554
finite quotient, 548
klt, 558, 781
log canonical, 558, 781
simplicial, 548
terminal, 551, 554, 555, 774
toric surface
resolution of singularities, 465, 470
singular points, 459
smooth classification, 495, 496, 790
toric topology, 574
toric variety, 106
affine, see affine, toric variety
complete, see complete variety, toric

name, 791, 792
nonnegative points of, 568
normal, see normal, toric variety
of a fan, 107
of a polytope, 83, 108
positive points of, 568
projective, see projective toric variety
quasiprojective, see quasiprojective toric variety
semicomplete, 734
semiprojective, see semiprojective toric variety
topological model of, 574
toroidal variety, 790
torsion submodule, 444
torsion-free semigroup, 22
torus, 5, 10
embedding, 108, 791, 792
orbit, 118
torus-invariant
Cartier divisor, 176
prime divisor, 170
Weil divisor, 172, 175
total coordinate ring, 207, 218, 248, 309, 362, 446,
795
transportation polytope, 64
triangulation, 739, 744, 745, 761
regular, 740, 743, 745
tropical polynomial, 188
tropical variety, 188
tropicalization, 188
twist of a sheaf, 421
unibranch, 563
unipolar variety, 765
unique factorization domain (UFD), 7
universal coefficient theorem, 598
universal covering space, 563
universally closed, 143
valuative criterion for properness, 148
Van Kampen theorem, 566
Vandermonde identity, 439, 442
vanishing theorem
Batyrev-Borisov, 413, 415, 425
Bott-Steenbrink-Danilov, 421, 423, 791
Demazure, 410, 425, 788, 791
Grauert-Riemenschneider, 428
higher direct images, 410
Kawamata-Viehweg, 427
Kodaira, 421, 423
Mavlyutov, 424
Mustata , 426
Nakano, 421, 423
Serre affine, 390
Index
Serre projective, 392

Zariski p-forms, 423, 440
variety
abstract, see abstract variety
affine, see affine, variety
complete, see complete variety, see complete
variety
Fano, 379
Gorenstein, 373
Gorenstein Fano, 379
toric, 380, 810
irreducible, see irreducible variety
normal, see normal, variety
of general type, 763
projective, see projective variety
projectively normal, see projectively normal
variety
Q-Fano, 765
quasiprojective, see quasiprojective variety
separated, see separated variety
toric, see toric variety
vector bundle, 250
chart of, 250
decomposable, 337
fiber above a point, 250, 257
sheaf of sections of, 252
toric, 335, 337
transition functions of, 250, 252
trivialization of, 250
Veronese embedding, 84, 689
Veronese subring, 689, 690
vertex, 63
very ample
divisor, see Cartier divisor, very ample
polyhedron, 322
polytope, 71, 75, 87, 269, 803
volume polynomial, 654, 658660
reduced, 659, 660
wall, 265, 301
extremal, 306
wall crossing, 747, 749, 750, 757
wall relation, 301, 751
weak Oda conjecture, 790
weak splitting, 137, 766
wedge product, 355, 366
weighted homogeneous polynomial, 53
weighted projective space, 53, 112, 176, 188, 651,
653, 657, 808
Weil divisor, 159, 348
torus-invariant, see torus-invariant, Weil divisor
Zariski 1-forms, 361
on a simplicial toric variety, 363
on a toric variety, 361
Zariski p-forms, 355, 416, 421, 423, 424, 435, 440
on a simplicial toric variety, 378
841
on a toric variety, 375

on an affine toric variety, 376
Zariski closure, 4
Zariski tangent space, 6, 101, 354
Zariski topology, 4, 50, 97
Zariskis main theorem, 563
zonotope, 81, 285, 792

Toric Varieties Cox PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Toric Varieties Cox PDF

Uploaded by

Copyright:

Available Formats

Toric Varieties

M ATHEMATICS , A MHERST C OLLEGE , A MHERST, MA

E-mail address: dac@math.amherst.edu

E-mail address: little@mathcs.holycross.edu

Part I. Basic Theory of Toric Varieties

Affine Toric Varieties

Background: Affine Varieties

Introduction to Affine Toric Varieties

Cones and Affine Toric Varieties

Properties of Affine Toric Varieties

Appendix: Tensor Products of Coordinate Rings

Projective Toric Varieties

Background: Projective Varieties

Lattice Points and Projective Toric Varieties

Lattice Points and Polytopes

Polytopes and Projective Toric Varieties

Properties of Projective Toric Varieties

Normal Toric Varieties

Background: Abstract Varieties

Fans and Normal Toric Varieties

The Orbit-Cone Correspondence

Complete and Proper

Appendix: Nonnormal Toric Varieties

Divisors on Toric Varieties

Background: Valuations, Divisors and Sheaves

Weil Divisors on Toric Varieties

Cartier Divisors on Toric Varieties

The Sheaf of a Torus-Invariant Divisor

Homogeneous Coordinates on Toric Varieties

Background: Quotients in Algebraic Geometry

Quotient Constructions of Toric Varieties

The Total Coordinate Ring

Sheaves on Toric Varieties

Homogenization and Polytopes

Line Bundles on Toric Varieties

Background: Sheaves and Line Bundles

Ample and Basepoint Free Divisors on Complete Toric Varieties

Polytopes and Projective Toric Varieties

The Nef and Mori Cones

The Simplicial Case

Appendix: Quasicoherent Sheaves on Toric Varieties

Projective Toric Morphisms

Background: Quasiprojective Varieties and Projective Morphisms

Polyhedra and Toric Varieties

Projective Morphisms and Toric Varieties

Projective Bundles and Toric Varieties

Appendix: More on Projective Morphisms

The Canonical Divisor of a Toric Variety

Background: Reflexive Sheaves and Differential Forms

One-Forms on Toric Varieties

Differential Forms on Toric Varieties

Fano Toric Varieties

Sheaf Cohomology of Toric Varieties

Background: Sheaf Cohomology

Cohomology of Toric Divisors

Applications to Lattice Polytopes

Local Cohomology and the Total Coordinate Ring

Part II. Topics in Toric Geometry

Singularities of Toric Surfaces and Their Resolutions

Continued Fractions and Toric Surfaces

Smooth Toric Surfaces

Riemann-Roch and Lattice Polygons

Toric Resolutions and Toric Singularities