Algebraic Geometry (Lei Fu)

Algebraic Geometry
Lei Fu
Nankai Institute of Mathematics
Tianjin, P. R. China
Tsinghua University Press

Preface
In this book we study the cohomology of coherent sheaves on schemes. An ex-

cellent textbook on this topic is [Hartshorne]. But in Hartshorn’s book, many
important theorems which hold for proper morphisms are proved only for pro-
jective morphisms. Moreover, Hartshorne doesn’t say much about the technique
of spectral sequences. The main purpose of this book is to fulfil this need. Of
course, everything in this book is contained in Grothendieck’s [EGA]. I hope
this book can make the wonderful ideas of Grothendieck which are hidden in
the voluminous [EGA] more clear. To make the book self-contained, I include
many materials which are treated nicely in [Hartshorne]. So there are some
overlaps with [Hartshorne] in this book, especially in sections 1.2 and 1.4.
This book roughly covers the main materials in [EGA] I, II and III. A few
years ago, I was invited to give a series of talks on l-adic cohomolgy theory
at the Morningside Center of Mathematics (MCM) at the Chinese Academy of
Science. To prepare the talk, I wrote a book (unpublished) on étale cohomology
theory which covered the main materials in [SGA] 1, 4, 4 21 , 5, and 7. The
current book together with [Matsumura] covers all the prerequisites for reading
my manuscripts. So I think it should provide adequate preparation for learning
étale cohomology theory.
I only assume the reader is familiar with Chapter 1-8 of [Atiyah-Macdonald].
All other results on algebra used in this book are either proved in this book, or
can be proved by the reader without much difficulty. The reason why I include
materials on algebra is my belief that the best way to learn algebra is to learn it
simultaneously with geometry so that one can get geometric intuition of abstract
algebraic concepts.
This book is by no means a complete treatise on algebraic geometry. Nothing
is said on how to apply the results obtained by cohomological method in this
book to study the geometry of algebraic varieties. Serre duality is also omitted.
The reader should consult [Hartshorne] and references there for these topics.
I thank heartily Prof. Keqin Feng. Ever since he knew my existence, he
has never stopped encouraging me. He invited me to MCM to lecture on l-adic
cohomology theory and other topics on algebraic geometry. Part of this book is
based on some of my lecture notes. Prof. Feng also makes it possible to publish
this book. Without his help, this book will never come into existence.
iii
iv PREFACE
During the preparation of this book, I was supported by the Qiu Shi Science
& Technologies Foundation, by Project 973, by IHES, and by MCM.
Lei Fu
Nankai Institute of Mathematics
Contents
Preface iii
1 Schemes and Coherent Sheaves 1

1.1 Presheaves and Sheaves . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Schemes and Morphisms . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Properties of Schemes and Morphisms . . . . . . . . . . . . . . . 24
1.4 Coherent Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.5 Formal Completions of Schemes and Sheaves . . . . . . . . . . . 76
2 Cohomology 99
2.1 Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
2.2 Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.3 Čech Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . 144
2.4 Cohomology of Affine and Projective Schemes . . . . . . . . . . . 155
2.5 Cohomological Study of Proper Morphisms . . . . . . . . . . . . 167
2.6 Local Freeness of Higher Direct Images . . . . . . . . . . . . . . . 180
2.7 Grothendieck’s Existence Theorem . . . . . . . . . . . . . . . . . 194
Bibliography 217
v
Chapter 1
Schemes and Coherent

Sheaves
1.1 Presheaves and Sheaves
Let X be a topological space. A presheaf F of sets on X consists of the following

data:
(a) For every nonempty open subset U of X, we have a set F(U ) whose
elements are called sections of F over U .
(b) For every inclusion V ⊂ U of nonempty open subsets of X, we have a
map ρU V : F(U ) → F(V ), called the restriction.
These data satisfy
(i) ρU U = idF (U ) ,
(ii) if W ⊂ V ⊂ U are open subsets of X, then ρU W = ρV W ρU V .
For any section s ∈ F(U ) and V ⊂ U , we often denote ρU V (s) by s|V .
Elements in F(U ) are often denoted by (s, U ) in order to make the open set U
explicit in the notation. We make the convention that F(∅) = ∅ for any presheaf
of sets F.
Define the category of open subsets of X so that its objects are nonempty
open subsets of X, and for any two objects U and V , define

∅ if V 6⊂ U,
Hom(V, U ) =
{V ,→ U } if V ⊂ U.
Then a presheaf F of sets on X is just a contravariant functor from the category

of open subsets of X to the category of sets.
Similarly we define a presheaf F of abelian groups (resp. rings) on X to be a
contravariant functor F from the category of open subsets of X to the category
of abelian groups (resp. rings). For any sheaf of abelian groups or rings, we
make the convention that F(∅) = {0}.
Here are some examples of presheaves:
1
2 CHAPTER 1. SCHEMES AND COHERENT SHEAVES
1. Let X be a topological space and A an abelian group. For every nonempty

open subset U of X, define F(U ) = A, and for every inclusion V ⊂ U of
nonempty open subsets, define ρU V = idA . Then F is a presheaf of abelian
groups, called the constant presheaf associated to A.
2. Let X be a topological space. For every open subset U of X, define
C(U ) to be the ring of complex valued continuous functions on U , and for every
inclusion V ⊂ U of nonempty open subsets, define ρU V : C(U ) → C(V ) to be
the restriction of functions. Then C is a presheaf of rings.
3. Let π : X 0 → X be a continuous map of topological spaces. For every
nonempty open subset U of X, define S(U ) to be the set of continuous sections
of π over U :
S(U ) = {s : U → π −1 (U )|πs = id, and s is continuous},
and for every inclusion V ⊂ U of nonempty open subsets, define ρU V : S(U ) →

S(V ) to be the restriction of sections. Then S is a presheaf of sets.
We say a presheaf F of sets (resp. abelian groups, resp. rings) is a sheaf if

it satisfies the following conditions:
(i) Let s, t ∈ F(U ) be two sections. If there exists an open covering {Ui }i∈I
of U such that s|Ui = t|Ui for any i, then s = t.
(ii) Suppose {Ui }i∈I is an open covering of U and si ∈ F(Ui ) are some
sections satisfying si |Ui ∩Uj = sj |Ui ∩Uj for any i, j ∈ I. Then there exists a
section s ∈ F(U ) such that s|Ui = si for any i ∈ I. (By (i), such s is unique.)
Note that a presheaf F of abelian groups is a sheaf if and only if for any
open covering {Ui }i∈I of any open subset U , the sequence
Y Y
0 → F(U ) → F(Ui ) → F(Ui ∩ Uj )
i∈I i,j∈I
is exact, where the second arrow is

Y
F(U ) → F(Ui ), s 7→ (s|Ui )
i∈I
and the third arrow is

Y Y
F(Ui ) → F(Ui ∩ Uj ), (si ) 7→ (sj |Ui ∩Uj − si |Ui ∩Uj ).
i∈I i,j∈I
The last two examples of presheaves given above are sheaves.
A direct set is a partially ordered set (I, ≤) such that for any i, j ∈ I, there
exists a k ∈ I such that i, j ≤ k. A direct system (Ai , φij )i∈I of sets consists of
a family of sets Ai (i ∈ I) and maps φij : Ai → Aj for pairs i ≤ j such that
φii = idAi and φjk φij = φik whenever i ≤ j ≤ k. For any xi ∈ Ai and xj ∈ Aj ,
we say xi is equivalent to xj if there exists a k ≥ i, j such that
`φik (xi ) = φjk (xj ).
This defines an equivalence relation on the disjoint union i Ai of Ai (i ∈ I).
1.1. PRESHEAVES AND SHEAVES 3
The direct limit inv. limi Ai of (Ai , φij )i∈I is defined to be the set of equivalence
classes.
Let X be a topological space and P a point in X. For any two neighborhoods
U and V of P , we say V ≤ U if U ⊂ V . Then the family of neighborhoods of
P becomes a direct set with respect to this order. For any presheaf F on X,
define the stalk FP of F at P by
FP = dir. lim F(U ),

P ∈U
where the direct limit is taken over the family of neighborhoods of P . So

elements of FP can be represented by sections of F over some neighborhoods of
P . Two sections s ∈ F(U ) and t ∈ F(V ) define the same element in FP if and
only if there exists a neighborhood W of P such that W ⊂ U ∩V and s|W = t|W .
For any neighborhood U of P , we have a canonical map F(U ) → FP . The image
of a section s ∈ F(U ) in FP is called the germ of s at P and is denoted by sP .
Let F and G be presheaves of abelian groups on X. A morphism of presheaves

φ : F → G consists of a homomorphism of abelian groups φ(U ) : F(U ) → G(U )
for every open subset U such that for every inclusion V ⊂ U of open subsets,
the following diagram commutes:
φ(U )
F(U ) → G(U )
ρU V ↓ ↓ ρU V
φ(V )
F(V ) → G(V ).
For any point P ∈ X, φ induces a homomorphism on stalks φP : FP → GP . If we

regard presheaves of abelian groups as contravariant functors from the category
of open subsets on X to the category of abelian groups, then a morphism of
presheaves is just a natural transformation. Similarly we can define morphisms
between presheaves of sets or rings. For any presheaf F, we have the identity
morphism idF . Given two morphisms of presheaves φ : F → G and ψ : G → H,
we can define their composition ψφ : F → H in the obvious way. We thus get
the category of presheaves. A morphism of presheaves φ : F → G is called an
isomorphism if it has a two-sided inverse, that is, there exists a morphism of
presheaves ψ : G → F such that ψφ = idF and φψ = idG . This is equivalent to
saying that φ(U ) : F(U ) → G(U ) is an isomorphism for every open subset U .
We define morphisms of sheaves as morphisms of presheaves. We thus get the
category of sheaves which is a full subcategory of the category of presheaves.
Proposition 1.1 Let φ : F → G be a morphism of sheaves on a topological

space X. Then φ is an isomorphism if and only if the induced map on stalks
φP : FP → GP is an isomorphism for every P ∈ X.
Proof. The “only if” part is obvious. Let’s prove the “if” part. Suppose φP :
FP → GP is bijective for every P ∈ X. We need to show φ(U ) : F(U ) → G(U )
is bijective for every open subset U of X.
Let s, s0 ∈ F(U ) be two sections such that φ(s) = φ(s0 ). Then φP (sP ) =
φP (s0P ) for any P ∈ U . Since φP is injective, we have sP = s0P . So there exists
a neighborhood UP of P contained in U such that s|UP = s0 |UP . Note that
{UP }P ∈U is an open covering of U . Since F is a sheaf, we must have s = s0 . So
φ(U ) is injective.
Let (t, U ) be a section in G(U ). For any P ∈ U , since φP : FP → GP is
surjective, we may find sP ∈ FP such that φP (sP ) = tP . We may assume sP is
the germ of a section (s, UP ) ∈ F(UP ) for some neighborhood UP of P . Note
that φ(s, UP ) and (t, U ) have the same germ at P . Choosing UP sufficiently
small, we may assume UP ⊂ U and φ(s, UP ) = (t, U )|UP . Then for any two
points P, Q ∈ U , we have
φ(s, UP )|UP ∩UQ = (t, U )|UP ∩UQ = φ(s, UQ )|UP ∩UQ .
By the injectivity of φ(UP ∩ UQ ) that we have proved above, we must have

(s, UP )|UP ∩UQ = (s, UQ )|UP ∩UQ . Note that {UP }P ∈U form an open covering
of U . Since F is a sheaf, we may find a section (s, U ) ∈ F(U ) such that
(s, U )|Up = (s, UP ) for any P ∈ U . We have (φ(s, U ))|UP = (t, U )|UP . Since G
is a sheaf, we must have φ(s, U ) = (t, U ). So φ(U ) is surjective.
Before going on, we introduce some concepts from the theory of categories.
Let C be a category. A morphism f : A → B in C is called a monomorphism
or injective if for any two morphisms α, β : C → A satisfying f α = f β, we
have α = β. An epimorphism is defined similarly by reversing the directions
of arrows. More precisely, f : A → B is called an epimorphism or surjective if
for any two morphisms α, β : B → C satisfying αf = βf , we have α = β. If a
morphism is both injective and surjective, we say it is bijective. An isomorphism
is a morphism with a two-sided inverse. Any isomorphism is bijective. But a
bijective morphism may not be an isomorphism.
Let Ai (i Q∈ I) be a family of objects in C. The direct productQof Ai (i ∈ I)
is an object i∈I Ai together with a family of morphisms pi : i∈I Ai → Ai
(i ∈ I) called projections with the following universal property: For any object
C and any family of morphisms
Q fi : C → Ai (i ∈ I), there exists one and only
one morphism f : C → i∈I Ai such that pi f = fi for any i. If the direct
product of Ai (i ∈ I) exists, it is unique up to unique isomorphism, that is, any
two direct product of Ai (i ∈ I) are isomorphic and the isomorphism between
them is unique.
The direct sum of Ai (i ∈ I) is defined similarly as above by reversing the
directions of arrows. More precisely, the direct sum of Ai (i ∈ I) is an object
⊕i∈I Ai together with a family of morphisms ki : Ai → ⊕i∈I Ai (i ∈ I) with the
following universal property: For any object C and any family of morphisms
fi : Ai → C (i ∈ I), there exists one and only one morphism f : ⊕i∈I Ai → C
such that f ki = fi for any i. If the direct product of Ai (i ∈ I) exists, it is
unique up to unique isomorphism.
Let (I, ≤) be a directed set. A direct system (Ai , φij )i∈I consists of a family
of objects Ai (i ∈ I) and morphisms φij : Ai → Aj for pairs i ≤ j such that
φii = idAi for any i and φjk φij = φik whenever i ≤ j ≤ k. The direct limit
of a direct system (Ai , φij ) is an object dir. limi Ai together with morphisms
φi : Ai → dir. limi Ai (i ∈ I) satisfying φj φij = φi whenever i ≤ j and having
the following universal property: For any object C and any morphisms ψi :
Ai → C (i ∈ I) satisfying ψj φij = ψi (i ≤ j), there exists a unique morphism
ψ : dir. limi Ai → C such that ψφi = ψi for any i. If the direct limit exists, it
is unique up to unique isomorphism. Let (A0i , φij )i∈I be another direct system.
A morphism from (Ai , φij ) to (A0i , φij ) is a family of morphisms ui : Ai → A0i
(i ∈ I) such that for any i ≤ j, the following diagram commutes:
u
Ai →i A0i
φij ↓ ↓ φ0ij
uj
Aj → A0j .
It induces a morphism dir. limi ui : dir. limi Ai → dir. limi A0i .

An inverse system (Ai , φji )i∈I consists of a family of objects Ai (i ∈ I)
and morphisms φji : Aj → Ai for pairs i ≤ j such that φii = idAi for any i
and φji φkj = φki whenever i ≤ j ≤ k. The inverse limit of an inverse system
(Ai , φij ) is an object inv. limi Ai together with morphisms φi : inv. limi Ai → Ai
(i ∈ I) satisfying φji φj = φi whenever i ≤ j and having the following universal
property: For any object C and any morphisms ψi : C → Ai (i ∈ I) satisfying
ψji ψj = ψi (i ≤ j), there exists a unique morphism ψ : C → inv. limi Ai such
that φi ψ = ψi for any i. A morphism from an inverse system (Ai , φji )i∈I to an
inverse system (A0i , φji )i∈I is a family of morphisms ui : Ai → A0i (i ∈ I) such
that for any i ≤ j, the following diagram commutes:
u
Aj →i A0j
φji ↓ ↓ φ0ji
uj
Ai → A0i .
0
It induces a morphism inv. limi ui : inv. limi Ai → inv. limi AQi . If (Ai , φji )i∈I is
an inverse system of sets,
Q then inv. limi A i is the subset of i Ai consisting of
those elements (xi ) ∈ i Ai satisfying φji (xj ) = xi for any i ≤ j.
A category C is called an additive category if for any objects A, B and C in

C, the direct product of A and B exists, Hom(A, B) is an abelian group, and
the map
Hom(A, B) × Hom(B, C) → Hom(A, C), (f, g) 7→ gf
is a homomorphism. We call 0 ∈ Hom(A, B) the zero morphism.
Proposition 1.2. Let C be an additive category and let A and B be two objects
in C.
(i) Let p1 : A × B → A and p2 : A × B → B be the projections. Define
k1 : A → A × B to be the unique morphism satisfying p1 k1 = idA and p2 k1 = 0,
and define k2 : B → A × B to be the unique morphism satisfying p1 k2 = 0 and
p2 k2 = idB . Then we have k1 p1 + k2 p2 = idA×B .
(ii) Suppose we have an object P and morphisms p1 : P → A, p2 : P → B,

k1 : A → P , k2 : B → P such that
p1 k1 = idA ,
p2 k2 = idB ,
k1 p1 + k2 p2 = idP .
Then P together with the morphisms p1 : P → A and p2 : P → B is the

direct product of A and B, P together with the morphisms k1 : A → P and
k2 : B → P is the direct sum of A and B.
Proof. (i) It is easy to verify that
p1 (k1 p1 + k2 p2 ) = p1 idA×B ,
p2 (k1 p1 + k2 p2 ) = p2 idA×B .
By the universal property of the direct product, we have k1 p1 + k2 p2 = idA×B .

(ii) We have
p1 k2 = p1 idP k2 = p1 (k1 p1 + k2 p2 )k2

= (p1 k1 )(p1 k2 ) + (p1 k2 )(p2 k2 ) = p1 k2 + p1 k2
= 2p1 k2 .
So p1 k2 = 0. Similarly p2 k1 = 0.
Let’s prove (P, k1 , k2 ) is the direct sum of A and B and leave to the reader
to prove (P, p1 , p2 ) is the direct product of A and B. Given any object C and
any morphisms f1 : A → C and f2 : B → C, define f = f1 p1 + f2 p2 . It is easy
to verify that f k1 = f1 and f k2 = f2 . If f 0 : P → C is a morphism such that
f 0 k1 = f1 and f 0 k2 = f2 , then we have
f0 = f 0 idP = f 0 (k1 p1 + k2 p2 )
= (f 0 k1 )p1 + (f 0 k2 )p2 = f1 p1 + f2 p2 .
This proves (P, k1 , k2 ) has the required universal property.
Let C be an additive category and f : A → B a morphism in C. We say a

monomorphism K → A is the kernel of f if the composition K → A → B is 0,
and for any morphism K 0 → A such that the composition K 0 → A → B is 0,
there exists a unique morphism K 0 → K such that the diagram
K0
↓ &
K → A
commutes. We often denote K by kerf and call it the kernel of f . Similarly we

define the cokernel of f to be an epimorphism B → C such that the composition
A → B → C is 0, and for any morphism B → C 0 such that the composition
A → B → C 0 is 0, there exists a unique morphism C → C 0 such that the

diagram
B → C
& ↓
C0
commutes. We often denote C by cokerf and call it the cokernel of f . We define
the image of f to be the kernel of the cokernel of f , and define the coimage of
f to be the cokernel of the kernel of f . There exists a canonical morphism
coimf → imf from the coimage to the image such that the diagram
A → B
↓ ↑
coimf → imf
commutes. For example, when f : A → B is a morphism in the category of
abelian groups, then
kerf = {a ∈ A|f (a) = 0},
imf = {b ∈ B|b = f (a) for some a ∈ A},
cokerf = B/imf,
coimf = A/kerf,
and the canonical morphism from the coimage to the image is the canonical
homomorphism A/kerf → imf (which is an isomorphism).
Let C be an additive category. A zero object 0 in C is an object such that
Hom(0, 0) = {0}. This is equivalent to saying that the identity morphism of 0 is
equal to the zero morphism. For any object X in C, we have Hom(X, 0) = {0}
and Hom(0, X) = {0}. Zero objects in C are isomorphic to each other.
An abelian category C is an additive category with zero objects such that
for any morphism f in C, the kernel and cokernel of f exist (and hence the
image and coimage of f exist), and the canonical morphism coimf → imf is an
isomorphism.
In an abelian category, a bijective morphism is an isomorphism. Indeed, if
f : A → B is injective, then the kernel of f is 0 → A and the coimage of f
is idA : A → A. If f : A → B is surjective, then the cokernel of f is B → 0
and the image of f is idB : B → B. If f : A → B is bijective, then the
canonical morphism coimf → imf is just f : A → B. Since coimf → imf is an
isomorphism, f : A → B is an isomorphism.
Suppose u : A → B is a monomorphism in an abelian category. We often
say A is a sub-object of B. Let B → C be the cokernel of u. We call C the
quotient of B by A and denote it by B/A.
In an abelian category, a sequence of morphisms
u v
A→B→C
is called exact if vu = 0 and the canonical morphism coimu → kerv is an
isomorphism. An exact sequence of the form
0→A→B→C→0
is called a short exact sequence. This short exact sequence is called split it is
isomorphic to
0 → A → A ⊕ C → C → 0,
where A → A ⊕ C and A ⊕ C = A × C → C are the canonical morphisms.
Proposition 1.3. Let

1 i p2
0 → A1 → A → A2 → 0
be a short exact sequence in an abelian category. The following conditions are
equivalent.
(i) The above short exact sequence is split.
(ii) There exists a morphism p1 : A → A1 such that p1 i1 = idA1 .
(iii) There exists a morphism i2 : A2 → A such that p2 i2 = idA2 .
Proof. (i)⇒(ii) and (i)⇒ (iii) are obvious.

(ii)⇒(i) Consider the morphism idA − i1 p1 : A → A. We have
(idA − i1 p1 )i1 = i1 − i1 (p1 i1 ) = 0.
Since A2 is the cokernel of i1 : A1 → A, there exists a morphism i2 : A2 → A

so that idA − i1 p1 = i2 p2 . Our assertion then follows from Proposition 1.2 (ii).
Similarly one can prove (iii)⇒(i).
Let F : C → D be a covariant functor between abelian categories. We say F

is additive if for any objects A and B in C, the map
Hom(A, B) → Hom(F (A), F (B))
is a homomorphism. We then have
F (A ⊕ B) ∼
= F (A) ⊕ F (B).
Indeed, keeping the notations in Proposition 1.2 (i) and applying F to the
equalities there, we get
F (p1 )F (k1 ) = idF (A) ,

F (p2 )F (k2 ) = idF (B) ,
F (k1 )F (p1 ) + F (k2 )F (p2 ) = idF (A⊕B) .
So by Proposition 1.2 (ii), (F (A ⊕ B), F (k1 ), F (k2 )) is the direct sum of F (A)
and F (B). Hence if
0→A→B→C→0
is a split short exact sequence, then
0 → F (A) → F (B) → F (C) → 0
is also a split short exact sequence.

Note that the category of abelian groups is an abelian category. We leave

to the reader to prove the following proposition:
Proposition 1.4. Let X be a topological space. Then the category of presheaves

of abelian groups on X is an abelian category. Let φ : F → G be a morphism
of presheaves of abelian groups. Then the kernel, cokernel and image of φ are
the presheaves defined by
(kerφ)(U ) = ker(φ(U ) : F(U ) → G(U )),

(cokerφ)(U ) = coker(φ(U ) : F(U ) → G(U )),
(imφ)(U ) = im(φ(U ) : F(U ) → G(U ))
for every open subset U of X. The stalks of these presheaves at a point P ∈ X

are given by
(kerφ)P = ker(φP : FP → GP ),
(cokerφ)P = coker(φP : FP → GP ),
(imφ)P = im(φP : FP → GP ).
Proposition 1.5. Let F be a presheaf on a topological space X. There exists

a pair (F + , θ) consisting of a sheaf F + and a morphism θ : F → F + such that
for any sheaf G and any morphism φ : F → G, there exists a unique morphism
ψ : F + → G such that φ = ψθ. The pair (F + , θ) is unique up to unique
isomorphism. For any point P ∈ X, θP : FP → (F + )P is an isomorphism. We
call F + the sheaf associated to the presheaf F.
Proof. ` For any open subset U of X, define F + (U ) to be the set of functions

s : U → P ∈X FP satisfying the following two conditions:
(a) For any P ∈ U , we have s(P ) ∈ FP .
(b) For any P ∈ U , there exists a neighborhood UP of P contained in U and
a section t ∈ F(UP ) such that for any Q ∈ UP , s(Q) is the germ tQ of t at Q.
Then F + is a sheaf and we have a canonical morphism θ : F → F + . We leave
to the reader to verify the pair (F + , θ) has the required property.
Note that for any section s ∈ F + (U ), we may find an open covering {Ui }i∈I
of U and sections si ∈ F(Ui ) such that θ(si ) = s|Ui , that is, sections in F +
locally come from sections of F. Two sections s, t ∈ F(U ) have the same image
in F + (U ) if and only if there exists a covering {Ui }i∈I of U such that s|Ui = t|Ui
for any i, that is, two sections of F which are locally equal are identified in F + .
When F is a sheaf, θ : F → F + is an isomorphism.
Proposition 1.6. Let X be a topological space. Then the category of sheaves

of abelian groups on X is an abelian category. Let φ : F → G be a morphism
of sheaves of abelian groups. Then kerφ is the sheaf defined by
(kerφ)(U ) = ker(φ(U )),

cokerφ is the sheaf associated to the presheaf
U 7→ coker(φ(U )),
and imφ is the sheaf associated to the presheaf
U 7→ im(φ(U )).
Moreover, for any P ∈ X, we have
(kerφ)P = ker(φP ),
(cokerφ)P = coker(φP ),
(imφ)P = im(φP ).
Proof. It is easy to show that the presheaf defined by U 7→ ker(φ(U )) is a

sheaf and is the kernel of φ. Using Proposition 1.4 and 1.5, one can show the
cokernel of φ is the sheaf associated to the presheaf U 7→ coker(φ(U )). Denote
by C the presheaf defined by U 7→ coker(φ(U )), and by P the presheaf defined
by U 7→ im(φ(U )). Let’s prove P + is the image of φ, that is, the kernel of
the cokernel of φ. We have a canonical morphism of presheaves P → G. It
induces a morphism P + → G. Since the composition of P → G → C is 0, the
composition P → G → C + is also 0 and hence the composition P + → G → C +
is 0. So P + → G induces a morphism P + → ker(G → C + ). We claim it is an
isomorphism. By Proposition 1.1, it suffices to show that for any P ∈ X, the
homomorphism (P + )P → (ker(G → C + ))P is an isomorphism. By Proposition
1.4 and 1.5, we have
(P + )P = PP = im(φP ),
(ker(G → C + ))P = ker(GP → (C + )P ) = ker(GP → CP )
= ker(GP → coker(φP )).
Our claim follows immediately. We leave to the reader to prove the other asser-
tions in the proposition.
Lemma 1.7. Let f : A → B be a morphism in an abelian category. Then

(i) f is injective ⇔ kerf = 0.
(ii) f is surjective⇔ cokerf = 0 ⇔ imf = B.
Proof. We only prove the “⇐” part in (i) and leave to the reader to prove the
rest. Consider the commutative diagram
A → B
↓ ↑
coimf → imf.
If kerf = 0, then the left vertical arrow is an isomorphism. The bottom hori-
zontal arrow is an isomorphism by the axiom of abelian category, and the right
vertical arrow is a monomorphism. So the upper horizontal arrow is a monomor-

phism.
Corollary 1.8. Let X be a topological space.

(i) In the category of presheaves of abelian groups on X, a morphism φ : F →
G is injective (resp. surjective, resp. bijective) if and only if φ(U ) : F(U ) →
G(U ) is injective (resp. surjective, resp. bijective) for any open subset U of X.
(ii) In the category of sheaves of abelian groups on X, a morphism φ : F → G
is injective if and only if φ(U ) : F(U ) → G(U ) is injective for any open subset
U of X.
(iii) In the category of sheaves of abelian groups on X, a morphism φ : F → G
is injective (resp. surjective, resp. bijective) if and only if φP : FP → GP is
injective (resp. surjective, resp. bijective) for any P ∈ X.
(iv) In the category of sheaves of abelian groups on X, a morphism φ : F → G
is surjective if and only if for any open subset U and any section t ∈ G(U ),
there exists an open covering {Ui }i∈I of U and sections si ∈ F(Ui ) such that
φ(si ) = t|Ui for any i.
Proof. (i) follows from Proposition 1.4 and Lemma 1.7. (ii) follows from
Proposition 1.6 and Lemma 1.7. (iv) follows from (iii). Let’s prove the statement
about surjectivity in (iii) and leave to the reader to prove the rest. By Lemma
1.7, φ is surjective if and only if imφ ∼
= G. By Proposition 1.1, this is equivalent
to saying (imφ)P ∼ = GP for any P ∈ X. By Proposition 1.6, this is equivalent to
saying im(φP ) ∼ = GP , that is, φP is surjective for any P ∈ X.
We leave to the reader to prove the following corollary:
Corollary 1.9. Let X be topological space.

(i) In the category of presheaves of abelian groups on X, a sequence
φ ψ
F →G→H
is exact if and only if
φ(U ) ψ(U )
F(U ) → G(U ) → H(U )
is exact for any open subset U of X.
(ii) In the category of sheaves of abelian groups on X, a sequence
φ ψ
F →G→H
is exact if and only if
φP ψP
FP → GP → HP
is exact for any P ∈ X.
Let f : X → Y be a continuous map of topological spaces. For any sheaf F

on X, the direct image f∗ F of F is the sheaf on Y defined by
(f∗ F)(V ) = F(f −1 (V ))
for any open subset V of Y . Note that f∗ is an additive functor from the
category of sheaves of abelian groups on X to the category of sheaves of abelian
groups on Y . For any sheaf G on Y , the inverse image f −1 G of G is the sheaf
on X associated to the presheaf defined by
U 7→ dir. lim G(V )

f (U )⊂V
for every open subset U of X, where the limit is taken over the family of open
subsets V of Y containing f (U ). Note that f −1 is an additive functor from the
category of sheaves of abelian groups on Y to the category of sheaves of abelian
groups on X. When f is an imbedding, we often denote f −1 G by G|X and call
it the restriction of G to X.
For any sheaf F on X, define a canonical morphism
f −1 f∗ F → F
as follows: Since f −1 f∗ F is the sheaf associated to the presheaf
U 7→ dir. lim (f∗ F)(V ) = dir. lim F(f −1 (V )),

f (U )⊂V U ⊂f −1 (V )
by Proposition 1.5, it suffices to define a canonical morphism from the presheaf

U 7→ dir. limU ⊂f −1 (V ) F(f −1 (V )) to the sheaf F. We define
dir. lim F(f −1 (V )) → F(U )

U ⊂f −1 (V )
to be the map induced by the restrictions F(f −1 (V )) → F(U ).

For any sheaf G on Y , define a canonical morphism
G → f∗ f −1 G
as follows: For any open subset W of Y , since f (f −1 (W )) ⊂ W , we have a map

G(W ) → dir. limf (f −1 (W ))⊂V G(V ). Composing with the canonical homomor-
phism dir. limf (f −1 (W ))⊂V G(V ) → f −1 G(f −1 (W )), we get a homomorphism
G(W ) → f −1 G(f −1 (W )) = (f∗ f −1 G)(W ), and hence a morphism of sheaves
G → f∗ f −1 G.
For any sheaf F on X and G on Y , define
αF ,G : Hom(G, f∗ F) → Hom(f −1 G, F)
as follows: For any morphism φ : G → f∗ F, define αF ,G (φ) to be the composition

f −1 φ
f −1 G → f −1 f∗ F → F.
Define
βF ,G : Hom(f −1 G, F) → Hom(G, f∗ F)
as follows: For any morphism ψ : f −1 G → F, define βF ,G (ψ) to be the compo-
sition
f∗ ψ
G → f∗ f −1 G → f∗ F.
One can verify that αF ,G and βF ,G are inverse to each other, and they are
functorial with respect to F and G, that is, for any morphism F1 → F2 of
sheaves on X and any morphism G1 → G2 of sheaves on Y , the following diagram
commutes:
αF1 ,G2
Hom(G2 , f∗ F1 ) → Hom(f −1 G2 , F1 )
↓ ↓
αF2 ,G1
−1
Hom(G1 , f∗ F2 ) → Hom(f G1 , F2 ),
where the vertical arrows are induced by F1 → F2 and G1 → G2 . Moreover a
similar diagram for βF ,G commutes.
In general, let C and D be categories and let u : C → D and v : D → C be
functors. We say u is left adjoint to v or v is right adjoint to u if for any object
C in C and D in D, we have a bijection
αC,D : Hom(C, v(D)) ∼

= Hom(u(C), D)
which is functorial in C and D. We have seen that G 7→ f −1 G is left adjoint to

F 7→ f∗ F. Proposition 1.5 shows that the functor F 7→ F + from the category
of presheaves to the category of sheaves is left adjoint to the inclusion functor
from the category of sheaves to the category of presheaves.
If a functor left adjoint to v exists, then it is unique up to isomorphism.
Indeed, let u and u0 be two functors left adjoint to v. Then for any object C in
C, we have a functorial bijection
Hom(u(C), D) ∼
= Hom(u0 (C), D)
for any object D in D. Taking D = u(C), we see that there exists a morphism
φC ∈ Hom(u0 (C), u(C)) corresponding to idu(C) ∈ Hom(u(C), u(C)). Taking
D = u0 (C), we see that there exists a morphism ψC ∈ Hom(u(C), u0 (C)) cor-
responding to idu0 (C) ∈ Hom(u0 (C), u0 (C)). One can verify that φ and ψ are
natural transformations between u and u0 and they are inverse to each other.
Similarly, if a functor right adjoint to u exists, it is unique up to isomorphism.
Let f : X → Y and g : Y → Z be two continuous maps. Obviously we have

(gf )∗ = g∗ f∗ . So for any sheaf F on X and H on Z, we have
Hom((gf )−1 H, F) = Hom(H, (gf )∗ F) = Hom(H, g∗ f∗ F)

= Hom(g −1 H, f∗ F) = Hom(f −1 g −1 H, F),
that is,
Hom((gf )−1 H, F) = Hom(f −1 g −1 H, F).
Hence (gf )−1 = f −1 g −1 .
Let P be a point in X and let i : {P } → X be the inclusion. Then the
inverse image i−1 F of a sheaf F on X can be identified with the stalk FP . So
for any continuous map f : X → Y and any sheaf G on Y , we have
(f −1 G)P = i−1 f −1 G = (f i)−1 G = Gf (P ) ,

that is, (f −1 G)P = Gf (P ) .
We end this section with two lemmas which hold in any abelian categories.
We only need the special case where the abelian category is the category of
sheaves. In this case, the lemmas can be easily proved by diagram chasing.
Lemma 1.10. (The Snake Lemma) Let
0 → A → B → C → 0
u↓ v↓ w↓
0 → A0 → B 0 → C 0 → 0
be a commutative diagram in an abelian category such that the two rows are
exact. Then there exists a morphism δ : kerw → cokeru such that the following
sequence is exact:
δ
0 → keru → kerv → kerw → cokeru → cokerv → cokerw → 0.
Lemma 1.11. (The Five Lemma) Let
A1 → A2 → A3 → A4 → A5
f1 ↓ f2 ↓ f3 ↓ f4 ↓ f5 ↓
B1 → B2 → B3 → B4 → B5
be a commutative diagram in an abelian category such that the two rows are
exact. If f1 , f2 , f4 and f5 are isomorphisms, then f3 is also an isomorphism.
1.2 Schemes and Morphisms
Throughout this book, all rings are assumed to be commutative and have iden-
tity element 1, and all homomorphisms of rings are assumed to map 1 to 1.
For every ring A, let SpecA be the set of prime ideals of A. For every ideal
a of A, define
V (a) = {p ∈ SpecA|a ⊂ p}.
We leave to the reader to prove the following proposition:
Proposition 2.1.
(i) V (0) = SpecA and V (A) = ∅.
(ii) If a ⊂ b, then V (a)
P ⊃ V (b).
(iii) ∩i∈I V (ai ) = V ( i∈I ai ) for any family of ideals ai (i ∈ I) of A.
(iv) V (a) ∪ V (b) = V (ab) = V (a ∩ b).
This proposition shows that the family of subsets of SpecA of the form V (a)
is closed under the operations of intersection and finite union, and includes the
1.2. SCHEMES AND MORPHISMS 15
empty set and the total space SpecA. So SpecA is a topological space whose
closed sets are of the form V (a) for ideals a of A. This topology is called the
Zariski topology on SpecA.
Proposition 2.2.
(i) We have SpecA = ∅ if and only if 0 = 1 in A.
(ii) For any ideal a of A, define the nilpotent radical of a to be the ideal
√
a = {a ∈ A|an ∈ a for some natural number n}.
√
Then we have V (a)
√ = V ( Ta).
(iii) We have a = p.
p∈V (a) √
√
(iv) For any ideals a and b of A, we have V (a) ⊂ V (b) if and only if a ⊃ b.
Proof. (i) If 0 = 1, then A = 0 and SpecA = ∅. Suppose 0 6= 1. Let
S = {a|a is an ideal of A and 1 6∈ a}.
Then (0) ∈ S and hence S is nonempty. One can easily show that any totally
ordered subset of S with respect to the order defined by inclusion has an upper
bound in S. So by Zorn’s Lemma, S has a maximal element, that is, A has a
maximal ideal. Any maximal ideal is prime. So SpecA is nonempty.
(ii) Follows directly from
√ the definition
T of V (a). √
(iii) Obviously we have a ⊂ p. Suppose f 6∈ a. Then in (A/a)f , we
p∈V (a)
have 0 6= 1. By (i), we may find a prime ideal q of (A/a)f . Let p be the inverse
image of q under the canonical
T homomorphism A → (A/a)f . Then p ∈ V (a)
but f 6∈ p. Hence f 6∈ p.
p∈V (a)
(iv) Follows from (ii),(iii), and Proposition 2.1 (ii).
Proposition 2.3. For any f ∈ A, the subset
D(f ) = SpecA − V ((f )) = {p ∈ SpecA|f 6∈ p}
is open. Open subsets of this type form a basis for the Zariski topology of SpecA,
that is, any open subset of SpecA is a union of such open subsets. Moreover,
D(f ) is quasi-compact.
Proof. Since D(f ) is the complement of V ((f )), it is open. Let U be an open
subset of SpecA and p a point in U . Then U = SpecA − V (a) for some ideal a of
A. We have a 6⊂ p. So there exists an f ∈ a−p. Then D(f ) is a neighborhood of p
contained in U . So open subsets of the type D(f ) (f ∈ A) form a basis for SpecA.
Let’s prove D(f ) is quasi-compact. Suppose D(f ) ⊂ ∪i∈I D(fi ) for some P family
of elements fi (i ∈ I) in A. Then we have V ((f )) ⊃ ∩i∈I V ((fi )) = V ( i∈I (fi )).
p pP
By Proposition 2.2 (iv), we have (f ) ⊂ i∈I (fi ). So we may find some
natural number n such that f n = ai1 fi1 + · · · + aik fik for some i1 , . . . , ik ∈ I
and ai1 , . . . , aik ∈ A. This implies that D(f ) ⊂ D(fi1 ) ∪ · · · ∪ D(fik ). Hence
D(f ) is quasi-compact.
Define a sheaf of rings OSpecA on SpecA as follows:

` For any open subset U
of SpecA, OSpecA (U ) consists of functions s : U → p∈SpecA Ap satisfying the
following two conditions:
(a) For any p ∈ U , we have s(p) ∈ Ap .
(b) For any p ∈ U , there exist a neighborhood Up of p contained in U and
a, f ∈ A such that for any q ∈ Up , we have f 6∈ q and s(q) = fa in Aq .
For any inclusion of open subsets V ⊂ U , we define OSpecA (U ) → OSpecA (V ) to
be the restriction of functions. We call the pair (SpecA, OSpecA ) the spectrum
of A. We often denote OSpecA by O if this doesn’t cause any confusion.
Proposition 2.4.
(i) For any p ∈ SpecA, we have a canonical isomorphism Op ∼
= Ap .
(ii) For any f ∈ A, we have a canonical isomorphism O(D(f )) ∼
= Af . In
particular, taking f = 1, we get O(SpecA) ∼
= A.
Proof. (i) For any neighborhood U of p, we have a homomorphism O(U ) → Ap

defined by s 7→ s(p). These homomorphisms induce a homomorphism
Op = dir. lim O(U ) → Ap .

p∈U
One can easily show it is surjective. To show it is injective, assume (s, U ) ∈ O(U )
satisfies s(p) = 0. Let Up be a neighborhood of p contained in U and let a, f ∈ A
such that for any q ∈ Up , we have f 6∈ q and s(q) = fa in Aq . Since s(p) = 0,
we have fa = 0 in Ap . So there exists a t 6∈ p such that ta = 0. Then for any
q ∈ D(f ) ∩ D(t), we have fa = tf
ta
= 0 in Aq . Hence we have s|Up ∩D(t) = 0. Note
that Up ∩ D(t) is a neighborhood of p. This shows that Op → Ap is injective.
(ii) Let’s prove the homomorphism
Af → O(D(f ))
which sends fak ∈ Af to the section D(f ) → p∈SpecA Ap , p 7→ fak ∈ Ap is

`
bijective.
Injectivity: Suppose fak lies in the kernel of the above homomorphism. Then
for any p ∈ D(f ), we have fak = 0 in Ap . So for any p ∈ D(f ), there exists a
t 6∈ p such that ta = 0. Hence we have
p ∈ D(f ) ⇒ Ann(a) 6⊂ p,
where Ann(a) = {r ∈ A|ra = 0} is the annihilator of a. So
p 6∈ V ((f )) ⇒ p 6∈ V (Ann(a)).
Therefore
V (Ann(a)) ⊂ V ((f )).
p
By Proposition 2.2 (iv), we have f ∈ Ann(a), that is, f n a = 0 for some
natural number n. So fak = 0 in Af . This proves the injectivity.
`
Surjectivity: Let s : D(f ) → p∈SpecA Ap be a section in O(D(f )), let
{Ui }i∈I be an open covering of D(f ), and let ai , fi ∈ A such that for any
q ∈ Ui , we have fi 6∈ q and s(q) = afii in Aq . By Proposition 2.3, we may
assume Ui = D(gi ) for some gi ∈ A, and we may assume the covering is a finite
covering. Then for any q ∈ D(g p fi 6∈ q. Hence D(gi ) ⊂ D(fi ). By
pi ), we have
Proposition 2.2 (iv), we have (gi ) ⊂ (fi ). So giki = hi fi for some natural
number ki and hi ∈ A. Note that for any q ∈ D(gi ), we have hi fi 6∈ q and
ai hi ai hi ai
s(q) = = = ki
fi hi fi gi
in Aq . Moreover, we have D(gi ) = D(giki ). Replacing gi by giki and ai by hi ai ,

we may assume that we have a finite covering {D(gi )}i∈I of D(f ) and elements
ai ∈ A such that for any q ∈ D(gi ), we have s(q) = agii in Aq .
a
For any q ∈ D(gi ) ∩ D(gj ) = D(gi gj ), we have agii = gjj in Aq . So for
any q ∈ D(gi gj ), there exists a t 6∈ q such that t(ai gj − aj gi ) = 0. Hence
for any q ∈ D(gi gj ), we have Ann(ai gj − aj gi ) 6⊂ q, that is, if q 6∈ V ((gi gj )),
then q 6∈ V (Ann(ai gj − aj gi )). Therefore
p V (Ann(ai gj − aj gi )) ⊂ V ((gi gj )). By
Proposition 2.2 (iv), we have gi gj ∈ Ann(ai gj − aj gi ). Choose a large integer
k, we may assume (gi gj )k (ai gj − aj gi ) = 0 for every pair i, j. (Our covering is
ai gik
a finite covering.) We have ai
gi = gik+1
. Replacing gi by gik+1 and ai by ai gik , we
may assume {D(gi )}i∈I is a finite covering of D(f ), and for any q ∈ D(gi ), we
have s(q) = agii in Aq , and ai gj = aj gi for every pair i, j.
P
From D(f ) = ∪i D(gi ), we get V ((f )) = ∩i V ((gi )) = V (( i gi )). By Propo-
sition 2.2 (iv), we have f k = i bi gi for some natural number k and bi ∈ A.
P
For any j, we have
X X X X
aj f k = aj bi gi = bi aj gi = bi ai gj = ( bi ai )gj .
i i i i
So we have P
aj i bi ai
= .
gj fk
P
bi ai
Hence under the homomorphism Af → O(D(f )) defined at the beginning, fk
is mapped to the section s ∈ O(D(f )). This proves the surjectivity.
A ringed space is a pair (X, OX ) consisting of a topological space X and a

sheaf of rings OX on X. We say (X, OX ) is a locally ringed space if for any
P ∈ X, the stalk OX,P of OX at P is a local ring. (Recall that a ring A is called
local if it has only one maximal ideal. This is equivalent to saying that elements
in A which are not unit form an ideal of A.) By Proposition 2.4 (i), for any ring
A, the spectrum (SpecA, OSpecA ) is a locally ringed space. If X is a topological
space and CX is the sheaf of continuous function on X, then (X, CX ) is also a

locally ringed space.
Let (X, OX ) and (Y, OY ) be two ringed spaces. A morphism from (X, OX )
to (Y, OY ) is a pair (f, f ] ) consisting of a continuous map f : X → Y and
a morphism of sheaves f ] : OY → f∗ OX . For any point P ∈ X, we have a
homomorphism (f∗ OX )f (P ) → OX,P defined by
(f∗ OX )f (P ) = dir. lim (f∗ OX )(V ) = dir. lim OX (f −1 (V ))

f (P )∈V P ∈f −1 (V )
→ dir. lim OX (U ) = OX,P .

P ∈U
Composing this homomorphism with the homomorphism OY,f (P ) → (f∗ OX )f (P )

induced by f ] , we get a homomorphism
fP] : OY,f (P ) → OX,P .
Let (X, OX ) and (Y, OY ) be two locally ringed spaces. A morphism of locally
ringed spaces (f, f ] ) : (X, OX ) → (Y, OY ) is a morphism of ringed spaces such
that for any P ∈ X, fP] : OY,f (P ) → OX,P is a local homomorphism. (Recall
that a local homomorphism of local rings f : A → B is a ring homomorphism
such that f −1 (mB ) = mA , or equivalently, f (mA ) ⊂ mB , where mA and mB are
the maximal ideals of A and B, respectively.) We can define the composition
of morphisms of locally ringed spaces in the obvious way. An isomorphism of
locally ringed spaces (f, f ] ) : (X, OX ) → (Y, OY ) is a morphism with a two-sided
inverse. This is equivalent to saying that f : X → Y is a homeomorphism of
topological spaces and f ] : OY → f∗ OX is an isomorphism of sheaves.
Any locally ringed space that is isomorphic to (SpecA, OSpecA ) for some ring
A is called an affine scheme. A scheme (X, OX ) is a locally ringed space such
that there exists an open covering {Ui }i∈I of X such that each (Ui , OX |Ui ) is an
affine scheme. We call X the underlying topological space and OX the structure
sheaf. We often denote (X, OX ) by X. We define morphisms of schemes as
morphisms of locally ringed spaces. We often denote a morphism of schemes
(f, f ] ) : (X, OX ) → (Y, OY ) by f : X → Y .
Proposition 2.5.
(i) Let φ : A → B be a homomorphism of rings. Then φ induces canonically
a morphism of locally ringed spaces
(f, f ] ) : (SpecB, OSpecB ) → (SpecA, OSpecA ).

(ii) Any morphism (f, f ] ) : (SpecB, OSpecB ) → (SpecA, OSpecA ) of locally
ringed spaces is obtained this way.
Proof. (i) Given a homomorphism φ : A → B, define f : SpecB → SpecA by
f (q) = φ−1 (q)

for any q ∈ SpecB. For any ideal a of A, we have f −1 (V (a)) = V (aB), where
aB is the ideal of B generated by φ(a). Hence f is continuous. `
For any open subset V of SpecA and any section s : V → p∈SpecA Ap of
OSpecA (V ), define a section f ] (s) : f −1 (V ) → q∈SpecB Bq in OSpecB (f −1 (V ))
`
by
(f ] (s))(q) = φq (s(f (q)))
for any q ∈ f −1 (V ), where φq : Aφ−1 (q) → Bq is the homomorphism induced by
φ. In this way we get a morphism of sheaves f ] : OSpecA → f∗ (OSpecB ). One
can verify that fq] : OSpecA,q → OSpecB,f (q) coincides with the homomorphism
φq : Aφ−1 (q) → Bq through the identifications defined in Proposition 2.4 (i). So
fq] is a local homomorphism and hence (f, f ] ) is a morphism of schemes.
(ii) Suppose (f, f ] ) : (SpecB, OSpecB ) → (SpecA, OSpecA ) is a morphism
of locally ringed spaces. Define φ : A → B so that the following diagram
commutes:
φ
A → B
∼
=↓ ↓∼
=
f]
OSpecA (SpecA) → (f∗ OSpecB )(SpecA) = OSpecB (SpecB),
where the vertical arrows are defined by Proposition 2.4 (ii). For every q ∈
SpecB, define φ0q : Af (q) → Bq so that the following diagram commutes:
φ0q
Af (q) → Bq
∼
=↑ ↑∼
=
]
f
q
OSpecA,f (q) → OSpecB,q ,
where the vertical arrows are defined in Proposition 2.4 (i). Since fq] is a local
homomorphism, φ0q is also local. The following diagram commutes:
f]
OSpecA (SpecA) → OSpecB (SpecB)
↓ ↓
]
f
q
OSpecA,f (q) → OSpecB,q .
Through the identifications in Proposition 2.4, this diagram becomes
φ
A → B
pA ↓ ↓ pB
φ0q
Af (q) → Bq ,
where the vertical arrows pA : A → Af (q) and pB : B → Bq are the canonical
homomorphisms. So we have
φ−1 (q) = φ−1 p−1 −1 0 −1
B (qBq ) = pA (φq ) (qBq ) = p−1
A (f (q)Af (q) ) = f (q),
where the third equality follows from the fact that φ0q is a local homomorphism.
Therefore we have f (q) = φ−1 (q). Moreover the commutativity of the last
diagram shows that φ0q coincides with the homomorphism φq : Aφ−1 (q) → Bq
induced by φ. For any open subset V of SpecA and any q ∈ f −1 (V ), we have a
commutative diagram
f]
OSpecA (V ) → OSpecB (f −1 (V ))
↓ ↓
f]
q
OSpecA,f (q) → OSpecB,q
↓ ↓
φq
Aφ−1 (q) → Bq .
Note that the compositions of the vertical arrows in this diagram are the ho-
momorphisms which send a section to its value at φ−1 (q) and at q, respec-
tively. The
` commutativity of the above diagram shows that foràny section
s : V → p∈SpecA Ap in OSpecA (V ), the section f ] (s) : f −1 (V ) → q∈SpecB Bq
in OSpecB (f −1 (V )) is given by f ] (s)(q) = φq (s(φ−1 (q))). Hence (f, f ] ) is in-
duced by φ.
Proposition 2.6. For any f ∈ A, we have a canonical isomorphism of locally

ringed space
(D(f ), OSpecA |D(f ) ) ∼
= (SpecAf , OSpecAf ).
Proof. The canonical homomorphism A → Af induces a morphism (ϕ, ϕ] ) :

(SpecAf , OSpecAf ) → (SpecA, OSpecA ). It is not hard to verify that the map
ϕ : SpecAf → SpecA is one-to-one continuous and open with image D(f ).
So it induces a homeomorphism between SpecAf and D(f ). One then show
ϕ] induces an isomorphism of sheaves OSpecA |D(f ) ∼= (ϕ∗ OSpecAf )|D(f ) using
Proposition 1.1 and the fact that Ap ∼
= (Af )pf for any p ∈ D(f ).
Corollary 2.7. Let (X, OX ) be a scheme and U an open subset of X. Then

(U, OX |U ) is a scheme. We call this scheme an open subscheme of (X, OX ).
Proof. Cover X by affine open subschemes Ui = SpecAi (i ∈ I). Then U can

be covered by U ∩Ui (i ∈ I). Since each U ∩Ui is an open subset of U = SpecAi ,
we may cover each U ∩ Ui by D(fij ) (j ∈ Ji ) for some fij ∈ Ai . Then D(fij )
(i ∈ I, j ∈ Ji ) form an open covering of U , and each (D(fij ), OX |D(fij ) ) is affine
by Proposition 2.6. So (U, OX |U ) is a scheme.
The following proposition is a generalization of Proposition 2.5:
Proposition 2.8. Let X be a scheme and A a ring. Then there is a one-to-one

correspondence between the family of morphisms of schemes from X to SpecA
and the family of homomorphisms of rings from A to OX (X).
Proof. Define a map

α : Hom(X, SpecA) → Hom(A, OX (X))
as follows: For any morphism of schemes f : X → SpecA, define α(f ) to be the
homomorphism f ] (SpecA) : A = OSpecA (SpecA) → OX (X). Define a map
β : Hom(A, OX (X)) → Hom(X, SpecA)
as follows: Let φ : A → OX (X) be a homomorphism. Cover X by affine open
φ
subschemes Ui = SpecAi (i ∈ I). The composition A → O(X) → O(Ui ) = Ai
induces a morphism of affine schemes fi : Ui → SpecA for each i. We claim
that fi |Ui ∩Uj = fj |Ui ∩Uj so that we can glue fi together to get a morphism
f : X → SpecA. We then define β(φ) = f . Indeed, cover Ui ∩ Uj by open
affine subschemes Uijk = SpecAijk (k ∈ Kij ). Then both fi |Uijk : Uijk =
SpecAijk → SpecA and fj |Uijk : Uijk = SpecAijk → SpecA are induced by the
φ
composition A → O(X) → O(Uijk ) = Aijk . So fi |Uijk = fj |Uijk and hence
fi |Ui ∩Uj = fj |Ui ∩Uj . One can verify α and β are inverse to each other.
Recall that a graded ring is a ring S together with a decomposition S =

∞
L
Sd of the additive group S into a direct sum of abelian groups Sd (d ∈
d=0
N ∪ {0}) such that Sd Se ⊂ Sd+e for any nonnegative integers d and e. Elements
in Sd are called homogeneous of degree d. Note that the identity element 1 lies
in S0 . An ideal a of S is called homogeneous if it satisfies one of the following
equivalent L conditions:
(i) a = d (a ∩ Sd ). P
(ii) If a ∈ a and a = d ad with ad ∈ Sd , then each ad ∈ a.
(iii) a is generated by homogeneous elements as an additive subgroup of S.
(iv) a is generated by homogeneous elements as an ideal of S.
For any homogeneous ideal a of S, S/a is a graded ring and the canonical
homomorphism S → S/a preserves the gradings.
Let a be a homogeneous ideal. If for any homogeneous elements f and g
in S such that f g ∈ a, we have f ∈ a or g ∈ a, then a is a prime ideal. To
see this, suppose f and g are elements P
in S (not necessarily
P homogeneous) such
that f g ∈ a but f, g 6∈ a. Suppose f = i fi and g = i gi , where fi and gi are
homogeneous of degree i for each i . Let m be the smallest integer such that
fm 6∈ a and n the smallest integer such that gn 6∈ a. The degree m + n part in
the decomposition of f g is
m−1
X n−1
X
fi gm+n−i + fm gn + fm+n−j gj .
i=0 j=0
Since f g ∈ a, this sum lies in a. By the choice of m and n, we have fi ∈ a for

i = 0, . . . , m − 1 and gj ∈ a for j = 0, . . . , n − 1. So we must have fm gn ∈ a.
By our assumption, we then have fm ∈ a or gn ∈ a. But this contradicts to our
choice of m and n.
We leave to the reader to show that sums, products, intersections, and nilpo-
tent radicals of homogeneous ideals are homogeneous.
∞
L
Let S+ = Sd and let ProjS be the set of homogeneous prime ideals of S
d=1
not containing S+ . For any homogeneous ideal a of S, define
V+ (a) = {p ∈ ProjS|a ⊂ p}.
Proposition 2.9.
(i) V+ (0) = ProjS andPV+ (S) = ∅.
(ii) ∩i∈I V+ (ai ) = V+ ( i∈I ai ) for any family of homogeneous ideals ai (i ∈
I) of S.
(iii) V+ (a) ∪ V+ (b) = V+ (ab) = V+ (a ∩ b) for any homogeneous ideals a and
b of S.
By the above proposition, the family of subsets of ProjS of the form V+ (a)
is closed under the operations of intersection and finite union and includes the
empty set and the total space ProjS. So we may define a topology on ProjS
so that closed sets are of the form V+ (a) for homogeneous ideals a of S. This
topology is called the Zariski topology on ProjS.
Let T be a multiplicative subset of S, that is, T is a subset of S containing

1 and is closed under multiplication. Then
a
{ ∈ T −1 S|a ∈ S, t ∈ T, a and t are homogeneous of the same degree}
t
is a subring of T −1 S. When T = S − p for some homogeneous prime ideal p,
we denote this ring by S(p) . When T = {1, f, f 2 , . . . , } for some homogeneous
element f ∈ S, we denote this ring by S(f ) .
Define a sheaf of rings OProjS on ProjS as follows:
of ProjS, OProjS (U ) consists of functions s : U → p∈ProjS S(p) satisfying the
(a) For any p ∈ U , we have s(p) ∈ S(p) .
homogeneous elements a, f ∈ S of the same degree such that for any q ∈ Up ,
we have f 6∈ q and s(q) = fa in S(q) .
For any inclusion of open subsets V ⊂ U , define OProjS (U ) → OProjS (V ) to be
the restriction of functions. We often denote OProjS by O if this doesn’t cause
any confusion.
Proposition 2.10.
(i) For any p ∈ ProjS, we have a canonical isomorphism Op ∼= S( p ) .
(ii) For any homogeneous element f ∈ S+ of positive degree, let
D+ (f ) = ProjS − V+ ((f )) = {p ∈ ProjS|f 6∈ p}.

Then D+ (f ) is open in ProjS. Open subsets of this type form a basis for the
topology of ProjS. Moreover, we have an isomorphism of locally ringed spaces
(D+ (f ), OProjS |D+ (f ) ) ∼

= (SpecS(f ) , OSpecS(f ) ).
In particular, (ProjS, OProjS ) is a scheme.
Proof. We leave to the reader to prove (i). Let’s prove (ii). As the complement
of V+ ((f )), D+ (f ) is open. For any homogeneous ideal a of S and any point
p in the open subset ProjS − V+ (a), we have a 6⊂ p and S+ 6⊂ p and hence
aS+ 6⊂ p. Let f be a homogeneous element in aS+ but not in p. Then f has
positive degree and p ∈ D+ (f ) ⊂ ProjS − V+ (a). Hence open subsets of the
form D+ (f ) form a basis.
Consider the map
ϕ : D+ (f ) → SpecS(f ) , p 7→ pSf ∩ S(f ) .
We claim ϕ is bijective. Indeed, suppose p1 , p2 ∈ D+ (f ) and ϕ(p1 ) = ϕ(p2 ).

Then for any homogeneous element x ∈ p1 , we have
xdegf
∈ p1 Sf ∩ S(f ) = p2 Sf ∩ S(f ) .
f degx
(Here we use the fact that degf > 0.) So we have xdegf ∈ p2 and hence x ∈ p2 .
Therefore p1 ⊂ p2 . Similarly p2 ⊂ p1 . This proves the injectivity of ϕ. For any
q ∈ SpecS(f ) , let p be the nilpotent radical of the ideal generated by those a ∈ S
such that fak lie in q for some k. (Then a is necessarily homogeneous of degree
kdegf ). Note that p is homogeneous. Suppose a1 and a2 are homogeneous
such that a1 a2 ∈ p. Then there exists an n such that (a1 a2 )n lies in the ideal
generated by numerators of elements in q. Note that (a1 a2 )ndegf lies in the
(a1 a2 )ndegf
same ideal. This implies that f n(dega 1 +dega2 )
lies in q. Since q is prime, we have
andegf andegf
1
f ndega1
∈ q or f ndega
2
2
∈ q. Hence a1 ∈ p or a2 ∈ p. Therefore p is a prime ideal.
Obviously f 6∈ p and hence p ∈ D+ (f ). One can verify that ϕ(p) = q. So ϕ is
surjective.
For any homogeneous ideal a of S, set b = aSf ∩ S(f ) . One can verify
ϕ(D+ (f )∩V+ (a)) = V (b). On the other hand, for any ideal b of S(f ) , define a to
be the ideal of S generated by numerators of elements in b. Then b = aSf ∩S(f ) .
This shows that ϕ establish a one-to-one correspondence between the family of
closed subsets of D+ (f ) and the family of closed subsets of SpecS(f ) . So ϕ is a
homeomorphism.
We have an obvious homomorphism
(S(f ) )ϕ(p) → S(p)

a
for any p ∈ D+ (f ). Given any t ∈ S(p) , we have t 6∈ p and dega = degt. So
tdegf −1 a
degf degf −1
t t a f degt
f degt
∈ S(f ) − ϕ(p) and f degt
∈ S(f ) . It is easy to verify that tdegf
lies in
f degt
(S(f ) )ϕ(p) and is mapped to at in S(p) . This proves (S(f ) )ϕ(p) → S(p) is onto. It
is not hard to show it is also injective. Using this isomorphism, we can construct
an isomorphism of sheaves ϕ] : OSpecS(f ) → ϕ∗ (OProjS |D+ (f ) ).
1.3 Properties of Schemes and Morphisms
Proposition 3.1. Let A be a ring. The following statements are equivalent:

(i) SpecA is connected.
(ii) A has no idempotent element other than 0 and 1,
(iii) For any decomposition A = A1 × A2 , we have A1 = 0 or A2 = 0.
Proof. (i)⇒(ii) Suppose e is an idempotent element of A. We have e(1−e) = 0.

So
V ((e)) ∪ V ((1 − e)) = V (e(1 − e)) = SpecA,

V ((e)) ∩ V ((1 − e)) = V ((e + 1 − e)) = ∅.
Moreover V ((e)) and V ((1 − e)) are closed. If SpecA is connected, then we must
have V ((e)) = SpecA = V ((0)) or V ((1−e)) = SpecA = V ((0)). By Proposition
2.2 (iv), e or 1 − e is nilpotent. But both e and 1 − e are idempotent. So e = 0
or 1 − e = 0.
(ii)⇒(iii) Suppose A = A1 × A2 . Then (1, 0) is idempotent. So (1, 0) = (0, 0)
or (1, 0) = (1, 1), and hence A1 = 0 or A2 = 0.
(iii)⇒(i) Suppose SpecA is not connected. Then we can find two nonempty
disjoint open subsets U , V of SpecA such that U ∪ V = SpecA. Consider
the section e ∈ OSpecA (SpecA) defined by e|U = 1 and e|V = 0. Obviously
e is idempotent. Through the isomorphism OSpecA (SpecA) ∼ = A, e defines an
idempotent element in A different from 0 and 1. Let A1 = (e) and A2 = (1 − e).
It is easy to verify that A = A1 × A2 and A1 6= 0 and A2 6= 0.
A nonempty topological space X is called irreducible if it satisfies one of the

following equivalent conditions:
(i) X is not the union of two proper closed subsets, that is, if X = F1 ∪ F2
for two closed subsets F1 and F2 , then either F1 = X or F2 = X.
(ii) Any two nonempty open subsets of X has nonempty intersection.
(iii) Any nonempty open subset is dense.
Note that a Hausdorff topological space is irreducible if and only if it consists
of a single point. Irreducible topological spaces are connected. Let X be a
topological space and Y a subset of X. If Y is irreducible, then Y is irreducible.
Proposition 3.2. A closed subset of SpecA is irreducible if any only if it is of

the form V (p) for some prime ideal p of A.
1.3. PROPERTIES OF SCHEMES AND MORPHISMS 25
Proof. Suppose p is a prime ideal of SpecA. If V (p) = V (a) ∪ V (b) = V (ab),

√
then ab ⊂ p = p. So a ⊂ p or b ⊂ p and hence V (p) = V (a) or V (p) = V (a).
This proves V (p) is irreducible.
Given an irreducible closed subset Y of SpecA, we may find an p√ ideal p of
√
A such that Y = V (p). By Proposition 2.2 (ii) and the fact that p = p,
√
we may choose p so that p = p. Suppose ab ⊂ p. Then V (p) ⊂ V (ab) =
V (a) ∪ V (b). Since V (p) is irreducible, we have V (p) ⊂ V (a) or V (p) ⊂ V (b)
and hence a ⊂ p or b ⊂ p. So p is a prime ideal.
Proposition 3.3. Let X be a scheme. For any irreducible closed subset Y of

X, there exists a unique point y ∈ Y such that Y = {y}. We call y the generic
point of Y .
Proof. Let U = SpecA be an affine open subscheme of X such that U ∩ Y is

nonempty. Then U ∩ Y is an irreducible closed subset of U and hence is of the
form V (p) for some prime ideal p of A. It is easy to see that V (p) = {p}. Let y
be the point in U corresponding to the prime ideal p. Then the closure of {y}
in U is U ∩ Y . But U ∩ Y is a nonempty open subset in the irreducible space
Y , so it is dense in Y . Hence Y = {y}. Suppose y 0 is another point such that
Y = {y 0 }. Then y 0 ∈ Y ∩ U . Let p0 be the prime ideal of A corresponding to
the point y 0 . Then we have V (p0 ) = {p0 } = Y ∩ U = V (p). This implies p = p0 .
So y = y 0 .
A topological space X is called noetherian if it satisfies one of the following

equivalent conditions:
(i) The family of closed subsets of X satisfies the descending chain condition,
that is, for any descending chain
F1 ⊃ F2 ⊃ · · ·
of closed subsets in X, we have Fn = Fn+1 = · · · for large n.

(ii) The family of open subsets of X satisfies the ascending chain condition.
(iii) Every nonempty family of closed subsets of X has a minimal element.
(iv) Every nonempty family of open subsets of X has a maximal element.
One can show any subspace of a noetherian topological space is noetherian.
Any noetherian topological space X is quasi-compact. Indeed, given an open
covering {Ui }i∈I , consider the family {∪i∈J Ui |J ⊂ I and J is finite} of open
subsets in X. By (iv), this family has a maximal element. This element gives a
finite open covering of X.
If A is a noetherian ring, then SpecA is a noetherian topological space.
(But the converse may not be true.) Indeed, given a descending chain of closed
subsets in SpecA:
V (a1 ) ⊃ V (a2 ) ⊃ · · · ,
where ai are some ideals of A, we have an ascending chain of ideals in A:
√ √
a1 ⊂ a2 ⊂ · · · .
√ √
Since A is noetherian, we have an = an+1 = · · · for large n. But then
V (an ) = V (an+1 ) = · · ·.
Proposition 3.4. Suppose X is a noetherian topological space.

(i) For any closed subset Y of X, we have Y = Y1 ∪ · · · ∪ Yn for some closed
irreducible subsets Yi (i = 1, . . . , n) such that Yi 6⊂ Yj whenever i 6= j. Such
decomposition is unique. Yi are called irreducible components of Y .
(ii) An irreducible closed subset Y is an irreducible component of X if and
only if Y is maximal among the family of irreducible closed subsets of X.
Proof. (i) Let S be the family of closed subsets of X which cannot be written
as a union of finitely many irreducible closed subsets. If S is not empty, then
it has a minimal element Y because X is noetherian. Since Y ∈ S, Y is not
irreducible. So Y = F1 ∪ F2 for two proper closed subsets. By the minimality
of Y , we have F1 , F2 6∈ S. Hence F1 and F2 can be written as finite unions of
irreducible closed subsets. But then Y = F1 ∪ F2 can also be written as a finite
union of irreducible closed subsets. This contradicts to the fact that Y ∈ S. So
S is empty and hence every closed subset of X can be written as a finite union
of irreducible closed subsets.
Suppose
Y = Y1 ∪ · · · ∪ Ym = Y10 ∪ · · · ∪ Yn0
are two decompositions of Y into irreducible closed subsets with Yi 6⊂ Yj and
Yi0 6⊂ Yj0 for any i 6= j, Then Y1 = (Y1 ∩ Y10 ) ∪ · · · ∪ (Y1 ∩ Yn0 ). Since Y1 is
irreducible, we have Y1 = Y1 ∩ Yi0 and hence Y1 ⊂ Yi0 for some i. Without loss
of generality, we assume Y1 ⊂ Y10 . Similarly we have Y10 ⊂ Yj for some j. Since
Y1 6⊂ Yj for any j 6= 1, we must have j = 1 and hence Y1 = Y10 . Similarly, by
rearranging the order of Y10 , Y20 , . . . , we get m = n and Yi = Yi0 for every i.
(ii) Suppose X1 , . . . , Xn are the irreducible components of X. For any irre-
ducible closed subsets Y , we have Y = (Y ∩ X1 ) ∪ · · · ∪ (Y ∩ Xn ). So Y = Y ∩ Xi
and hence Y ⊂ Xi for some i. If Y is maximal among the family of irreducible
closed subsets, then we must have Y = Xi and hence Y is an irreducible compo-
nent. Let’s prove the irreducible component X1 is maximal among the family of
irreducible closed subsets. Suppose X1 ⊂ Y for some irreducible closed subset
Y . As we have seen above, there exists an irreducible component Xi such that
Y ⊂ Xi . But then we have X1 ⊂ Xi . This implies i = 1 and X1 = Y .
As a corollary of Proposition 3.4 (i), we see that a Hausdorff topological

space is noetherian if and only if it consists of finitely many points. Indeed,
the space has finitely many irreducible components and each component, being
irreducible and Hausdorff, has only one point.
Let A be a ring. A prime ideal of A is called a minimal prime ideal if it
contains no prime ideal other than itself. Using Zorn’s lemma, one can show
that any prime ideal of A contains a minimal prime ideal.
Corollary 3.5. Let A be a ring such that SpecA is a noetherian topological

space. There is a one-to-one correspondence between the family of irreducible
components of SpecA and the family of minimal prime ideals of A.
Proof. By Proposition 3.2 and 2.2 (iv), the map p 7→ V (p) establishes a one-
to-one correspondence between the family of prime ideals of A and the family
of irreducible closed subsets of SpecA. By Proposition 3.4 (ii), each irreducible
component of SpecA is a maximal irreducible closed subset. If we write it in
the form V (p) for some prime ideal p, then p is minimal prime ideal.
A scheme (X, OX ) is called connected (resp. irreducible) if X is connected

(resp. irreducible). It is called reduced (resp. integral) if for any open subset U
of X, OX (U ) is reduced (resp. an integral domain.)
Suppose X is an integral scheme. Let ξ be its generic point. Then OX,ξ is a
field. We call it the function field of X. Indeed, for any affine open subscheme
U = SpecA of X, we must have that A is an integral domain, ξ lies in U and
the prime ideal p of A corresponding to ξ is (0). So OX,ξ is isomorphic to the
fractional field Ap of A. This shows that the function field of X is the fraction
field of OX (U ) for any affine open subscheme U of X. For any P ∈ X and
any open neighborhood U of P , we have ξ ∈ U . The family of homomorphisms
OX (U ) → OX,ξ induces a homomorphism OX,P = dir. limP ∈U OX (U ) → OX,ξ .
This homomorphism is injective and it identifies OX,ξ with the fraction field of
OX,P for any point P ∈ X. For any inclusion of open subsets U ⊂ V and any
P in U , the canonical homomorphisms
OX (V ) → OX (U ) → OX,P → OX,ξ
are injective.
Proposition 3.6. (X, OX ) is integral if and only if it is irreducible and reduced.
Proof. Suppose (X, OX ) is integral. Then it is reduced by definition. If it is

not irreducible, then we can find two disjoint nonempty open subsets U1 and U2
of X. We have OX (U1 ∪ U2 ) = OX (U1 ) × OX (U2 ), which cannot be an integral
domain. (We have (1, 0)(0, 1) = (0, 0), but (1, 0), (0, 1) 6= (0, 0)). Contradiction.
So X is irreducible.
Suppose (X, OX ) is irreducible and reduced. Let U be an open subset of X
and let f1 , f2 ∈ OX (U ) such that f1 f2 = 0. For each i ∈ {1, 2}, let Ufi be the
subset of U consisting of those points P ∈ U such that the germ of fi at P is a
unit in the local ring OX,P . Cover U by open affine subschemes Uα = SpecAα
(α ∈ I) and denote the restriction fi |Uα ∈ OX (Uα ) = Aα by fiα . Then we
have Ufi ∩ Uα = D(fiα ) for any α. So each Ufi is open. Moreover, we have
Uf1 ∩ Uf2 = Uf1 f2 . Since f1 f2 = 0, we have Uf1 f2 = ∅. So Uf1 ∩ Uf2 = ∅.
Since U is irreducible, either Uf1 or Uf2 is empty, say Uf1 = ∅. Then D(f1α ) =
Uf1 ∩ Uα = ∅ for every α. This implies f1α ∈ Aα = OX (Uα ) is nilpotent for
every α. Since OX (Uα ) is reduced, we have f1α = 0 for every α. Hence f1 = 0.
This shows that OX (U ) is an integral domain.

Proposition 3.7. A scheme (X, OX ) is reduced if any only if OX,P is reduced

for any P ∈ X.
A scheme (X, OX ) is called locally integral if OX,P is an integral domain for

any P ∈ X. An integral scheme is locally integral. But the converse may not
be true.
Proposition 3.8. Let (X, OX ) be a scheme such that X is a noetherian topo-

logical space. Then (X, OX ) is locally integral if and only if it is reduced and its
irreducible components are disjoint. In particular, the connected components
of any locally integral scheme coincide with the irreducible components.
Proof. Suppose the irreducible components Xi (i = 1, . . . , n) of X are disjoint.

Then each Xi is open and closed. If (X, OX ) is reduced, then each (Xi , OXi ) is
reduced and irreducible. So each (Xi , OXi ) is integral. Hence for any P ∈ Xi ,
OX,P is an integral domain. Therefore (X, OX ) is locally integral.
Conversely suppose (X, OX ) is locally integral. Then it is reduced by Propo-
sition 3.7. Suppose X1 and X2 are two irreducible components of X and
P ∈ X1 ∩ X2 . Let U = SpecA be an affine open subscheme of X contain-
ing P . It is not hard to show each Xi ∩ U is an irreducible component of U . By
Corollary 3.5, we have Xi ∩ U = V (pi ) (i = 1, 2) for some minimal prime ideals
pi of A. Let p be the prime ideal of A corresponds to the point P ∈ X1 ∩ X2 ∩ U .
We have pi ⊂ p. So p1 Ap and p2 Ap are two minimal prime ideals of Ap . Since
Ap = OX,P is an integral domain, it has only one minimal prime ideal. So
p1 = p2 and hence V (p1 ) = V (p2 ), that is, X1 ∩ U = X2 ∩ U . But Xi ∩ U is
dense in Xi for each i. So we have X1 = X2 . This shows that the irreducible
components of X are disjoint.
Let (X, OX ) be a scheme. For any section f ∈ OX (X), define Xf to be the

subset of X consisting of those P ∈ X such that the germ of f at P is a unit in
OX,P . We summarize the properties of Xf in the following proposition:
Proposition 3.9. Let (X, OX ) be a scheme.

(i) For any f ∈ OX (X), Xf is open. It is empty if and only if there exists
an open covering {Ui }i∈I of X such that each f |Ui is nilpotent. For any f, g ∈
OX (X), we have Xf ∩ Xg = Xf g .
(ii) Let (ϕ, ϕ] ) : (X, OX ) → (Y, OY ) be a morphism of schemes, f ∈ OY (Y ),
and ϕ] (f ) the image of f under the homomorphism ϕ] (Y ) : OY (Y ) → OX (X).
Then ϕ−1 (Yf ) = Xϕ] (f ) .
(iii) Suppose X can be covered by finitely many affine open subschemes Ui
(i ∈ I) such that Ui ∩Uj can be covered by finitely many affine open subschemes
for any i, j ∈ I. Let A = OX (X). Then for any f ∈ A, we have OX (Xf ) ∼ = Af .
Proof. (i) Confer the proof of Proposition 3.6.

(ii) For any P ∈ X, the image of the germ fϕ(P ) of f at ϕ(P ) under the
homomorphism ϕ]P : OY,f (P ) → OX,P is the germ (ϕ] (f ))P of ϕ] (f ) at P . Since
this homomorphism is local, fϕ(P ) is a unit if and only if (ϕ] (f ))P is a unit,
that is, ϕ(P ) ∈ Yf if and only if P ∈ Xϕ] (f ) .
(iii) The restriction A = OX (X) → OX (Xf ) maps f to a unit. So it induces
a homomorphism Af → OX (Xf ). Let’s prove it is an isomorphism.
For each i, let Ui = SpecAi . Denote by fi the image of f under the restriction
OX (X) → OX (Ui ) = Ai . Then Xf ∩ Ui coincides with the open subset D(fi )
of Ui .
Suppose s ∈ OX (X) is a section such that s|Xf = 0. We will show there
exists a natural number n such that f n s = 0 in OX (X). This would prove the
homomorphism Af → OX (Xf ) is injective.
Note that s|Ui lies in the kernel of the restriction OX (Ui ) → OX (Ui ∩ Xf )
for each i. This restriction coincides with the canonical homomorphism Ai →
(Ai )fi . Since every element in the kernel of Ai → (Ai )fi is annihilated by some
power of fi , we have fini s|Ui = 0 in OX (Ui ) for some natural number ni . Since
{Ui }i∈I is a finite covering, we may find a large n such that fin s|Ui = 0 for every
i. We then have f n s = 0.
Next we show that for any section t ∈ OX (Xf ), there exists a natural number
n such that f n t can be extended to a section in OX (X) = A. This would prove
the homomorphism Af → OX (Xf ) is surjective.
Since the restriction OX (Ui ) → OX (Ui ∩ Xf ) coincides with the canoni-
cal homomorphism Ai → (Ai )fi , there exists a natural number n such that
fin t|Ui ∩Xf can be extended to a section ti in O(Ui ) for each i. Note that the
restriction of ti |Ui ∩ Uj − tj |Ui ∩ Uj to Ui ∩ Uj ∩ Xf vanishes for any i, j. By
what we have proved above, there exists a natural number m such that for any
i, j ∈ I, we have f m (ti |Ui ∩ Uj − tj |Ui ∩ Uj ) = 0. So we can glue fim ti together
to get a section in OX (X). This section is an extension of f m+n t.
Proposition 3.10. A scheme (X, OX ) is affine if and only if we can find finitely
many sections f1 , . . . , fn ∈ O(X) which generate the unit ideal of O(X) such
that each open subscheme (Xfi , OX |Xfi ) is affine.
Proof. The necessity is obvious. Let’s prove the sufficiency. Let A = O(X).
By Proposition 2.8, the identity homomorphism A → O(X) defines a morphism
ϕ : X → SpecA. Let’s prove ϕ is an isomorphism. Since fi (i = 1, . . . , n)
generate the unit ideal, D(fi ) (i = 1, . . . , n) form an open covering of SpecA. We
have ϕ−1 (D(fi )) = Xfi by Proposition 3.9 (ii). We will verify the conditions in
Proposition 3.9 (iii) hold so that we have OX (Xfi ) ∼ = Afi . Each Xfi is affine by
assumption. So by Proposition 2.5, ϕ induces an isomorphism ϕ−1 (D(fi )) ∼ = X fi
for each i and hence ϕ is an isomorphism.
Let’s verify the condition of Proposition 3.9 (iii) holds. X can be covered
by Xfi (i = 1, . . . , n) and each Xfi is affine. Suppose Xfi = SpecBi . For each
j, let fj0 be the image of fj under the restriction OX (X) → OX (Xfi ) = Bi . By
Proposition 3.9 (ii), Xfi ∩ Xfj is equal to the subset D(fj0 ) of SpecBi , which is
affine.
A scheme (X, OX ) is called locally noetherian if it can be covered by affine
open subschemes Ui = SpecAi (i ∈ I) such that each Ai is noetherian. It is
called noetherian if we can find a finite affine open covering Ui = SpecAi (i ∈ I)

of X such that each Ai is noetherian. (X, OX ) is noetherian if and only if it is
locally noetherian and quasi-compact. If (X, OX ) is a noetherian scheme, then
X is a noetherian topological space. But the converse may not be true.
Proposition 3.11. Let (X, OX ) be a locally noetherian scheme. Then for any
affine open subscheme U = SpecA of X, A is noetherian. In particular, an affine
scheme (Spec, OSpecA ) is locally noetherian if and only if A is noetherian.
Proof. Cover X by affine open subschemes Ui = SpecAi (i ∈ I) such that each

Ai is noetherian. Then U = SpecA can be covered by Ui ∩ U (i ∈ I). Since each
Ui ∩ U is open in Ui = SpecAi , we may cover Ui ∩ U by D(fij ) = Spec(Ai )fij
(j ∈ Ji ) for some fij ∈ Ai . Note that each (Ai )fij is noetherian. Changing
notations, we may cover U = SpecA by affine open subschemes Ui = SpecAi
(i ∈ I) such that each Ai is noetherian. Cover each Ui by D(fij ) (j ∈ Ji ) for
0
some fij ∈ A. For each j ∈ Ji , let fij be the image of fij under the restriction
O(U ) = A → O(Ui ) = Ai . Then the subset D(fij ) of SpecA coincides with
0
the subset D(fij ) of SpecAi by Proposition 3.9 (ii). So we have O(D(fij )) =
0
O(D(fij )), that is, Afij = (Ai )fij0 . Since Ai is noetherian, (Ai )fij0 is noetherian
and hence Afij is noetherian. Changing notations, we may cover U = SpecA
by open subsets D(fi ) (i ∈ I) for some fi ∈ A such that each Afi is noetherian.
Since SpecA is quasi-compact, we may assume this covering is finite. For any
a
ideal a of A, let { kijij } (j ∈ Ji ) be a finite family of generators of aAfi . We
fi
claim a is generated by the finite family {aij }. Given a ∈ a, for each i, we may
write
a X bij aij
= lij kij
1 j f f i i
in Afi . So for some large ki , we have

X
fiki a = cij aij
j
qP
in A. But ∪i D(fiki ) = SpecA. So V ( (fiki )) = V (A) and hence 1 ∈ (fiki ).
P
Suppose 1 = i bi fiki . Then we have

P
X
a= bi cij aij .
i,j
Hence a is generated by the finite family {aij }. So A is noetherian.
A morphism f : X → Y of schemes is called quasi-compact if there exists a

covering of Y by affine open subschemes Vi (i ∈ I) such that each f −1 (Vi ) is
quasi-compact. It is called affine if Y can be covered by affine open subschemes
Vi (i ∈ I) such that each f −1 (Vi ) is affine. It is called locally of finite type if
Y can be covered by affine open subschemes Vi = SpecBi (i ∈ I) such that
each f −1 (Vi ) can be covered by affine open subschemes Uij = SpecAij (j ∈ Ji )

for some finitely generated Bi -algebras Aij . It is called of finite type if it is
quasi-compact and locally of finite type. It is called finite if Y can be covered
by affine open subschemes Vi = SpecBi (i ∈ I) such that for each i, we have
f −1 (Vi ) = SpecAi for some Bi -algebra Ai which is finitely generated as a Bi -
module.
Proposition 3.12. Let f : X → Y be a morphism of schemes.

(i) f is quasi-compact if and only if for any open quasi-compact subset V of
Y , f −1 (V ) is quasi-compact.
(ii) f is affine if and only if for any affine open subscheme V of Y , f −1 (V )
is affine.
(iii) f is locally of finite type if and only if for any affine open subscheme
V = SpecB of Y , f −1 (V ) can be covered by affine open subschemes Uj = SpecAj
(j ∈ J) such that each Aj is a finitely generated B-algebra.
(iv) f is of finite type if and only if for any affine open subscheme V =
SpecB of Y , f −1 (V ) can be covered by finitely many affine open subschemes
Uj = SpecAj (j ∈ J) such that each Aj is a finitely generated B-algebra.
(v) f is finite if and only if for any affine open subscheme V = SpecB of
Y , f −1 (V ) = SpecA for some B-algebra A which is finitely generated as a
B-module.
Proof. We only prove (ii) and leave to the reader to prove the rest. Suppose
f : X → Y is an affine morphism. Cover Y by affine open subschemes Vi =
SpecAi (i ∈ I) such that f −1 (Vi ) = SpecBi for some Ai -algebras Bi . For any
affine open subscheme V = SpecA, we cover each Vi ∩ V by open subsets of the
form D(fij ) (j ∈ Ji ) for some fij ∈ Ai . Note that D(fij ) (i ∈ I, j ∈ Ji ) form an
open covering of V , each D(fij ) is affine and each f −1 (D(fij )) ∼ = Spec(Bi )fij
is also affine. Changing notations, we may assume V = SpecA can be covered
affine open subschemes Vi = SpecAi (i ∈ I) so that f −1 (Vi ) = SpecBi for some
Ai -algebras Bi . Cover each Vi by open subsets of the form D(fij ) (j ∈ Ji )
0
for some fij ∈ A. For each i and each j ∈ Ji , let fij be the image of fij
under the restriction A = OY (V ) → Ai = OY (Vi ). The subset D(fij ) of V =
SpecA coincides with the subset D(fij 0
) of Vi = SpecAi . Since f −1 (D(fij 0
)) ∼
=
−1
Spec(Bi )fij0 , each f (D(fij )) is affine. Changing notations, we may assume
V = SpecA can be covered by D(fi ) (i ∈ I) for some fi ∈ A such that each
f −1 (D(fi )) is affine. Since SpecA is quasi-compact, we may assume this is a
finite covering. One can verify fi (i ∈ I) generate the unit ideal of A. Denote
f −1 (V ) by U and let fi0 (i ∈ I) be the images of fi under the homomorphism
OY (V ) → OX (U ). Then fi0 (i ∈ I) generate the unit ideal of OX (U ). Moreover,
we have Ufi0 = f −1 (D(fi )) by Proposition 3.9 (ii). So each Ufi0 is affine. By
Proposition 3.10, U = f −1 (V ) is affine.
Proposition 3.13. Let X → S and Y → S be two morphisms and assume

Y → S is locally of finite type.
(i) Let f, g : X → Y be two morphisms making the following diagram

commute:
X → Y
& .
S
Let x ∈ X such that f (x) = g(x) = y and such that fx] , gx] : OY,y → OX,x are
the same. Then there exists a neighborhood U of x in X such that f |U = g|U .
(ii) Suppose S is locally noetherian. Let x ∈ X and y ∈ Y be two points
such that their images in S are the same point s ∈ S, and let φ : OY,y → OX,x
be a homomorphism making the following diagram commute:
φ
OX,x ← OY,y
- %
OS,s
Then there exist a neighborhood U of x in X and a morphism f : U → Y such

that f (x) = y, fx] = φ, and the following diagram commutes:
f
U → Y
& .
S
(iii) Suppose S is locally noetherian, X → S is also locally of finite type, and

f : X → Y is a morphism making the diagram in (i) commute. Assume f (x) = y
and fx] : OY,y → OX,x is an isomorphism. Then there exist neighborhoods U of
x in X and V of y in Y such that f induces an isomorphism from U to V .
Proof. (i) Let s be the image of y in S, let W = SpecC be an affine open

neighborhood of s in S, let V = SpecB an affine open neighborhood of y in Y
which is mapped into W under the morphism Y → S such that B is a finitely
generated C-algebra, and let U = SpecA be an affine open neighborhood of
x in X such that f (U ), g(U ) ⊂ V . Let p, q and r be the prime ideals of A,
B and C corresponding to the points x, y and s, respectively. Then f and g
define some C-algebra homomorphisms φ, ψ : B → A, respectively, such that
φ−1 (p) = ψ −1 (p) = q, and φ and ψ induce the same homomorphism Bq → Ap .
Let b1 , . . . , bn be a finite family of generators of the C-algebra B. Then we
have φ(b1 i ) = ψ(b1 i ) in Ap . So for each i, there exists an si ∈ A − p such that
si φ(bi ) = si ψ(bi ). Let s = s1 · · · sn . Then φ(b1 i ) = ψ(b1 i ) in As and hence φ and
ψ induce the same homomorphism B → As . This implies that the restrictions
of f and g to the open neighborhood D(s) of x are the same.
(ii) Let W = SpecC be an affine open neighborhood of s in S, V = SpecB an
affine open neighborhood of y in Y which is mapped into W under the morphism
Y → S such that B is a finitely generated C-algebra, and U = SpecA an affine
open neighborhood of x in X which is mapped into W under the morphism
X → S. Let p and q be the prime ideals of A and B corresponding to the
points x and y, respectively. Then φ : OY,y → OX,x corresponds to a C-algebra
homomorphism Bq → Ap , which we also denote by φ. Let b1 , . . . , bn be a finite

family of generators of the C-algebra B. Then we have an epimorphism of
C-algebras
C[x1 , . . . , xn ] → B, xi 7→ bi .
Let I be its kernel. It is finitely generated since C[x1 , . . . , xn ] is a noetherian
ring. Let f1 (x1 , . . . , xn ), . . . , fm (x1 , . . . , xn ) ∈ C[x1 , . . . , xn ] be a finite family
of generators for the ideal I. Choosing a sufficiently large d, we may write each
fi (x1 , . . . , xn ) in the form
X
fi (x1 , . . . , xn ) = cii1 ...in xi11 · · · xinn .
0≤i1 ,...,in ≤d
Denote by θ the composition

φ
C[x1 , . . . , xn ] → B → Bq → Ap .
We have φ( b1i ) = ai
si for some ai ∈ A and si ∈ A − p. We have
θ(fi (x1 , . . . , xn )) = 0
in Ap , that is,

fi (b1 , . . . , bn )
φ =0
1
in Ap . Since

fi (b1 , . . . , bn ) X a1 i1 an
φ = cii1 ...in ( ) · · · ( )in
1 s1 sn
0≤i1 ,...,in ≤d
cii1 ...in s1d−i1 ai11 · · · sd−i n in

P
n an
0≤i1 ,...,in ≤d
= ,
sd1 · · · sdn
there exists a ti ∈ A − p such that
X
ti cii1 ...in s1d−i1 ai11 · · · sd−i
n
n in
an = 0.
0≤i1 ,...,in ≤d
Let s = s1 · · · sn t1 · · · tm and consider the C-algebra homomorphism

ai ai s1 · · · ŝi · · · sn t1 · · · tm
C[x1 , . . . , xn ] → As , xi 7→ = .
si s
It maps each fi (x1 , . . . , xn ) to 0 and hence I is contained in its kernel. So it
induces a homomorphism B → As . This homomorphism induces a morphism
of schemes f : D(s) → Y with the required property.
(iii) By (ii), there exist a neighborhood V of y in Y and a morphism g : V →
−1
X such that g(y) = x, gy] = fx] and the following diagram commutes:
g
V → X.
& .
S
By (i), there exists a neighborhood U of x in f −1 (V ) such that g ◦ (f |U ) = idU ,

and there exists a neighborhood V 0 of y in V such that f ◦ (g|V 0 ) = idV 0 . We
claim f and g induce isomorphisms
f
U ∩ f −1 (V 0 ) → g −1 (U ) ∩ V 0 ,
g
g −1 (U ) ∩ V 0 → U ∩ f −1 (V 0 )
which are inverse to each other. Indeed, for any P ∈ U ∩ f −1 (V 0 ), we have

f (P ) ∈ V 0 . Moreover, since P ∈ U , we have gf (P ) = P and hence f (P ) ∈
g −1 (U ). So f (P ) ∈ g −1 (U ) ∩ V 0 . Thus f (U ∩ f −1 (V )) ⊂ g −1 (U ) ∩ V 0 and f
f
induces a morphism U ∩f −1 (V 0 ) → g −1 (U )∩V 0 . Similarly g(g −1 (U )∩V 0 ) ⊂ U ∩
g
f −1 (V 0 ) and hence g induces a morphism g −1 (U )∩V 0 → U ∩f −1 (V 0 ). Moreover,
we have (gf )|U ∩f −1 (V 0 ) = idU ∩f −1 (V 0 ) and (f g)|g−1 (U )∩V 0 = idg−1 (U )∩V 0 .
Let X and Y be two integral schemes. A morphism f : X → Y is called

dominant if f (X) = Y . Let ξ be the generic point of X and η the generic point
of Y . We have
{f (ξ)} = f ({ξ}) = f (X) = Y = {η}.
So we must have f (ξ) = η. We say f is birational if the homomorphism fξ] :

OY,η → OX,ξ on function fields is an isomorphism.
Corollary 3.14. Let S be a locally noetherian scheme and let X and Y be two
integral schemes. Suppose we have a commutative diagram
f
X → Y
& .
S
such that X → S and Y → S are locally of finite type and f is a birational

morphism. Then there exist nonempty open subsets U ⊂ X and V ⊂ Y such
that f induces an isomorphism from U to V .
Proof. Let ξ be the generic point of X and η the generic point of Y . We have
f (ξ) = η and fξ] : OY,η → OX,ξ is an isomorphism. Our assertion then follows
from Proposition 3.13 (iii).
Let (X, OX ) be a scheme. Recall that an open subscheme is of the form

(U, OX |U ) for some open subset U of X. A morphism (f, f ] ) : (Z, OZ ) →
(X, OX ) is called an open immersion if it induces an isomorphism of (Z, OZ )
with an open subscheme of (X, OX ).
A morphism (f, f ] ) : (Z, OZ ) → (X, OX ) is called a closed immersion if
f : Z → Y induces a homeomorphism of Z with a closed subset of X, and
f ] : OX → f∗ OZ is surjective. Two closed immersions Z1 → X and Z2 → X
are called equivalent if there exists an isomorphism (necessarily unique) Z1 → Z2

such that the diagram
Z1 → X
↓ %
Z2
commutes. The equivalent classes of closed immersions into X are called closed
subschemes of X.
A morphism is Z → X is called an immersion if it can be factorized as
Z → U → X such that U → X is an open immersion and Z → U is a closed
immersion. We leave to the reader to prove the following proposition:
Proposition 3.15. Let (f, f ] ) : (Z, OZ ) → (X, OX ) be a morphism.

(i) (f, f ] ) is an open immersion if and only if f induces a homeomorphism
of Z with an open subset of X and for any P ∈ Z, fP] : OX,f (P ) → OZ,P is an
isomorphism.
(ii) (f, f ] ) is an immersion if and only if f induces a homeomorphism of Z
with a locally closed subset of X and for every P ∈ Z, fP] : OX,f (P ) → OZ,P is
an epimorphism. (Recall that a subset Z of X is called locally closed if it is the
intersection of an open subset with a closed subset. This is equivalent to saying
that Z is open in Z.)
(iii) Immersions are monomorphisms in the category of schemes. Moreover,
the composition of immersions (resp. open immersions, resp. closed immersions)
is an immersion (resp. open immersion, resp. closed immersion.)
Proposition 3.16. Let A be a ring.

(i) For any ideal a of A, the morphism ϕ : SpecA/a → SpecA induced by
the canonical homomorphism φ : A → A/a is a closed immersion.
(ii) Any closed immersion into SpecA is isomorphic SpecA/a → SpecA for
some ideal a of A.
Proof. (i) We leave to the reader to verify that ϕ : SpecA/a → SpecA induces
a homeomorphism onto V (a) on the underlying topological spaces. To show
ϕ] : OSpecA → ϕ∗ OSpecA/a is surjective, we use Corollary 1.8 (iii) and the fact
that for any p ∈ SpecA, we have

(A/a)p/a if p ∈ V (a),
(ϕ∗ OSpecA/a )p =
0 otherwise.
(ii) Let ϕ : Z → SpecA be a closed immersion. We first prove Z is affine. For

any point P ∈ Z, we may find an open neighborhood UP of P in SpecA such that
UP ∩ Z is an affine open subscheme of Z. Choose fP ∈ A such that D(fP ) is a
neighborhood of P in SpecA contained in UP . Then D(fP ) ∩ Z is an affine open
subscheme of Z. In fact, if UP ∩ Z = SpecA0 and g is the image of fP under the
ϕ]
composition A = OSpecA (SpecA) → OSpecA (UP ) → OZ (UP ∩ Z) = A0 , then by
Proposition 3.9 (ii), we have D(fP ) ∩ Z = D(g) ∼
= SpecA0g . Cover SpecA − Z by
open subsets of the form D(f ). So we have a covering {D(fi )}i∈I of SpecA such
that for each i, D(fi ) ∩ Z is an affine open subscheme of Z. (Some of them may
be empty.) We may assume this covering is finite since SpecA is quasi-compact.
Since D(fi ) (i ∈ I) cover SpecA, fi (i ∈ I) generate the unit ideal. For each i,
denote the image of fi under ϕ] : A = OSpecA (SpecA) → OZ (Z) by f¯i . Then f¯i
(i ∈ I) generate the unit ideal OZ (Z). Moreover each Zf¯i = D(fi ) ∩ Z is affine.
So Z is affine by Proposition 3.10.
Suppose Z = SpecB. Then the closed immersion Z → SpecA is induced by
a homomorphism φ : A → B. Let a = kerφ. We have a commutative diagram
of morphism of schemes:
ϕ
SpecB → SpecA,
θ↓ %
SpecA/a
where the vertical arrow θ is induced by the monomorphism A/a → B. Using
the definition of closed immersions and the fact that SpecB → SpecA and
SpecA/a → SpecA are closed immersions, one can show θ : SpecB → SpecA/a
is a closed immersion. By Lemma 3.17 below, the image of ϕ : SpecB →
SpecA is V (a). Hence θ : SpecB → SpecA/a is a homeomorphism on the
underlying topological spaces. Since A/a → B is injective, for any f ∈ A,
the homomorphism (A/a)f¯ → Bφ(f ) is injective, where f¯ is the image of f in
A/a. So θ] (D(f )) : OSpecA/a (D(f¯)) → (θ∗ OSpecB )(D(f¯)) is injective. Hence
θ] : OSpecA/a → θ∗ OSpecB is injective. Since θ is a closed immersion, θ] must
be an isomorphism. Hence the closed immersion Z → SpecA is isomorphic to
SpecA/a → SpecA.
Lemma 3.17. Let φ : A → B be a homomorphism and let f : SpecB → SpecA

be the morphism induced by φ. Then we have f (SpecB) = V (kerφ).
Proof. We give three different proofs in order to show different techniques. T

The First Proof. For any subset Y of SpecA, we claim that Y = V ( p).
p∈Y
We then have
\ \ \
f (SpecB) = V ( p) = V ( φ−1 (q)) = V (φ−1 ( q))
p∈f (SpecB) q∈SpecB q∈SpecB
p p
= V (φ−1 ( (0))) = V ( kerφ) = V (kerφ).
T
Let’s prove the claim. Obviously we have Y ⊂ V ( p) and hence Y ⊂
p∈Y
T
V( p). Suppose Y = V (a) for some ideal a of A. Then for any p ∈ Y , we
p∈Y
T T
have a ⊂ p. So a ⊂ p and hence V ( p) ⊂ V (a) = Y .
p∈Y p∈Y
The Second Proof. Obviously we have f (SpecB) ⊂ V (kerφ) and hence
f (SpecB) ⊂ V (kerφ). For any p ∈ V (kerφ), applying Zorn’s lemma to the fam-
ily of prime ideals of A containing kerφ and contained in p, we can find a prime
ideal p0 of A which is minimal among those prime ideals containing kerφ and
contained in p. We claim that p0 ∈ f (SpecB). Since we have p ∈ V (p0 ) = {p0 },
this implies p ∈ f (SpecB). Hence V (kerφ) ⊂ f (SpecB).
Let’s prove the claim. p0 /kerφ is a minimal prime ideal of A/kerφ. Moreover,
the homomorphism A/kerφ → B is injective. Replacing A by A/kerφ, we are
reduced to prove the following assertion: For any monomorphism φ : A → B
and any minimal prime ideal p of A, there exists a prime ideal q of B such that
φ−1 (q) = p. Note that the homomorphism Ap → Bp is injective. We have 0 6= 1
in Ap by Proposition 2.2 (i). So 0 6= 1 in Bp and hence SpecBp is nonempty.
Therefore we may find a prime ideal q of B which is disjoint from φ(A − p), that
is, φ−1 (q) ⊂ p. By the minimality of p, we must have φ−1 (q) = p.
The Third Proof. Again it is obvious that f (SpecB) ⊂ V (kerφ). For any
p ∈ V (kerφ), to prove p ∈ f (SpecB), it suffices to show any neighborhood D(a)
of p contains an element in f (SpecB). So we need to show that for any prime
ideal p of A containing kerφ and for any a ∈ A − p, there exists a prime ideal
q of B such that a 6∈ φ−1 (q). It suffices to show SpecBφ(a) is not empty, or
equivalently, φ(a) is not nilpotent. Indeed, if φ(a) is nilpotent, then some power
an of a lies in kerφ and hence lies in p. This contradicts to the fact that a 6∈ p.
Proposition 3.18. Let (X, OX ) be a scheme and Y a closed subset. Then

there exists a unique reduced scheme structure (Y, OY ) on Y which makes Y a
closed subscheme of X. We call this subscheme structure the reduced induced
closed subscheme structure on Y .
Proof. First consider the case where X√= SpecA is affine. Then Y = V (a)
for a unique ideal a of A satisfying a = a. Note that SpecA/a is a reduced
closed subscheme of SpecA with Y being the underlying topological space. This
proves the existence of the reduced induced closed subscheme structure. By
Proposition 3.16 (ii), any closed subscheme of X with Y being the underlying
topological space is of the form SpecA/a for some ideal a √ satisfying Y = V (a).
If the closed subscheme is reduced, then we must have a = a. Such a is unique.
This proves the uniqueness of the reduced induced closed subscheme structure.
Now consider the general case. Cover X by affine open subschemes Ui =
SpecAi (i ∈ I). We have seen that for each i, the closed subset Y ∩Ui of the affine
scheme Ui has a reduced induced closed subscheme structure (Y ∩ Ui , OY ∩Ui ).
We claim that for any i, j, there exists a unique isomorphism between the closed
immersion
(Y ∩ Ui ∩ Uj , OY ∩Ui |Y ∩Ui ∩Uj ) → (Ui ∩ Uj , OUi ∩Uj )
and the closed immersion
(Y ∩ Ui ∩ Uj , OY ∩Uj |Y ∩Ui ∩Uj ) → (Ui ∩ Uj , OUi ∩Uj ).
Hence we may glue the closed subschemes (Y ∩Ui , OY ∩Ui ) (i ∈ I) together to get
a reduced closed subscheme structure on Y . (Confer Proposition 3.19 below.)
To prove our claim, we cover Ui ∩ Uj by affine open subschemes Uijk (k ∈ Kij ).
For any i, j, k, both (Y ∩ Uijk , OY ∩Ui |Y ∩Uijk ) and (Y ∩ Uijk , OY ∩Uj |Y ∩Uijk ) are
reduced and their underlying topological spaces are the closed subset Y ∩ Uijk
of the affine scheme Uijk . By the uniqueness of the reduced induced closed
subscheme structure in the affine case, there is a unique isomorphism between
the closed immersion
(Y ∩ Uijk , OY ∩Ui |Y ∩Uijk ) → (Uijk , OUijk )
and the closed immersion
(Y ∩ Uijk , OY ∩Uj |Y ∩Uijk ) → (Uijk , OUijk ).
Our claim follows since Uijk (k ∈ Kij ) cover Ui ∩ Uj .

The uniqueness of the reduced induced closed subscheme structure is proved
again by reducing to affine case. We leave details to the reader.
In the above proof, we obtain the reduced induced closed subscheme struc-
ture by glueing subscheme structures. We summarize the process of glueing in
the following proposition and leave the proof to the reader:
Proposition 3.19.
(i) Let Ui (i ∈ I) be a family of topological spaces. Suppose for any pair i 6= j,
we are given an open subset Uij of Ui and a homeomorphism ϕij : Uij → Uji
with the following properties:
(a) For any i 6= j, we have ϕij = ϕ−1 ji .
(b) For any distinct i, j, k, we have ϕij (Uij ∩ Uik ) = Uji ∩ Ujk and ϕjk ϕij =
ϕik when restricted to Uij ∩ Uik .
Then there exists a topological space X together with embeddings ϕi : Ui → X
such that ϕi (Ui ) are open and form a covering of X, ϕi (Uij ) = ϕi (Ui ) ∩ ϕj (Uj ),
and ϕj ϕij = ϕi on Uij . X is unique up to homeomorphism. We call {ϕij } the
glueing data and X is obtained by glueing {Ui } along {Uij } through {ϕij }.
(ii) Let X be a topological space and let {Ui }i∈I be an open covering of X.
Suppose we are given a sheaf Fi on Ui for each i and an isomorphism of sheaves
φij : Fi |Ui ∩Uj → Fj |Ui ∩Uj for each pair i, j with the following properties:
(a) For any i, we have φii = idFi .
(b) For any triple i, j, k, we have φik = φjk φij on Fi |Ui ∩Uj ∩Uk .
Then there exists a sheaf F on X together with isomorphisms φi : F|Ui → Fi
such that φj = φij φi on F|Ui ∩Uj . F is unique up to isomorphism. We call {φij }
the glueing data and F is obtained by glueing {Fi } through {φij }.
(iii) Let Ui (i ∈ I) be a family of schemes. Suppose for any pair i 6= j,
we are given an open subscheme Uij of Ui and an isomorphism of schemes
ϕij : Uij → Uji with the following properties:
(a) For any i 6= j, we have ϕij = ϕ−1 ji .
(b) For any distinct i, j, k, we have ϕij (Uij ∩ Uik ) = Uji ∩ Ujk and ϕjk ϕij =
ϕik when restricted to Uij ∩ Uik .
Then there exists a scheme X together with open immersions ϕi : Ui → X such
that ϕi (Ui ) form a covering of X, ϕi (Uij ) = ϕi (Ui ) ∩ ϕj (Uj ), and ϕj ϕij = ϕi
on Uij . X is unique up to isomorphism. We call {ϕij } the glueing data and X

is obtained by glueing {Ui } along {Uij } through {ϕij }.
Let S be a scheme. An S-scheme or a scheme over S is a scheme X together

with a morphism X → S. An S-morphism from an S-scheme X to an S-scheme
Y is a morphism X → Y such that the diagram
X → Y
& .
S
commutes. We define the composition of S-morphisms in the obvious way. We

thus get the category of S-schemes. Let X and Y be S-schemes. Their product
in the category of S-schemes is called the fibred product of X and Y over S. It
is an S-scheme X ×S Y , together with two S-morphisms p : X ×S Y → X and
q : X ×S Y → Y called projections with the following universal property: For
any S-scheme Z and any two S-morphisms f : Z → X and g : Z → Y , there
exists a unique S-morphism h : Z → X ×S Y such that ph = f and qh = g. We
have a commutative diagram
X ×S Y
. &
X Y
& .
S
A diagram is called cartesian if it is of the above form.

Note that by Proposition 2.8, for any scheme X, we have a unique morphism
X → SpecZ. So the category of schemes coincides with the category of SpecZ-
schemes. The product of two schemes X and Y in the category of schemes is
denoted by X × Y . It is the same as X ×SpecZ Y .
Proposition 3.20. For any S-schemes X and Y , their fibred product over S
exists and is unique up to unique isomorphism.
Sketch of the Proof. The uniqueness follows from the universal property of
the fibred product.
Step 1. Assume X = SpecA, Y = SpecB and S = SpecC are affine. Then
using Proposition 2.8, one can verify X ×S Y = Spec(A ⊗C B), and p and
q are induced by the canonical C-algebra homomorphisms A → A ⊗C B and
B → A ⊗C B, respectively.
Step 2. Let X and Y be S-schemes such that X ×S Y exists. Then for any
open subscheme U of X, p−1 (U ) is the fibred product of U and Y over S.
Step 3. Suppose X can be covered by open subschemes Ui (i ∈ I) such that
each Ui ×S Y exists. Then X ×S Y exists. Indeed, by the uniqueness of the
fibred product and Step 2, for any i, j ∈ I, we can consider (Ui ∩ Uj ) ×S Y as
open subschemes of Ui ×S Y and of Uj ×S Y . We can glue {Ui ×S Y } together

along {Uij ×S Y }. One then verifies the resulting scheme is X ×S Y .
Step 4. Covering X by affine open subschemes and applying Steps 1 and 3,
we see that when S and Y are affine, X ×S Y exists.
Step 5. Covering Y by affine open subschemes and applying Step 4 and Step
3 with X and Y interchanged, we see that X ×S Y exists when S is affine.
Step 6. In general, cover S by affine open subschemes Si (i ∈ I) and let
Xi and Yi be the inverse images of Si in X and Y , respectively. Then each
Xi ×Si Yi exists by Step 5. One can show Xi ×S Y ∼ = Xi ×Si Yi . Applying Step
3, we see X ×S Y exists.
Suppose we are given a morphism of schemes S 0 → S. For any S-scheme X,

the projection X ×S S 0 → S 0 define an S 0 -scheme structure on X ×S S 0 . For any
S-morphism f : X → Y between S-schemes, let f × idS 0 : X ×S S 0 → Y ×S S 0
be the unique S 0 -morphism making the following diagram commute:
X ×S S 0 → Y ×S S 0
↓ ↓
X → Y.
We thus get a functor from the category of S-schemes to the category of S 0 -

schemes. This functor is called the base change from S to S 0 .
Proposition 3.21. Let f : X → Y be a morphism, y a point in Y , and k(y) the

residue field of the local ring OY,y . The projection X ×Y Speck(y) → X induces
an embedding of X ×Y Speck(y) onto f −1 (y) on the underlying topological
spaces. We call the scheme X ×Y Speck(y) the (scheme theoretic) fiber of f over
y.
Proof. The problem is local. So we may assume X = SpecB and Y = SpecA

are affine and f : X → Y is induced by a homomorphism φ : A → B. Let p be
the prime ideal of A corresponding to the point y. Then we have
X ×Y Speck(y) = Spec(B ⊗A Ap /pAp ) = SpecBp /pBp .
Note that we have the following one-to-one correspondences:
{ prime ideals of Bp /pBp }

↔ { prime ideals of Bp containing pBp }
↔ { prime ideals q of B such that q ∩ φ(A − p) = ∅ and q ⊃ φ(p)}
↔ { prime ideals q of B such that φ−1 (q) = p}.
This show that
X ×Y Speck(y) = SpecBp /pBp → X = SpecB

induces a one-to-one correspondence between X ×Y Speck(y) and f −1 (y). To

show it is an embedding, we factorize it into the composition
SpecBp /pBp → SpecBp → SpecB.
Note that the first arrow is a closed immersion and the second arrow is an
embedding on the underlying topological spaces. (In fact, for any multiplicative
set S of B, the canonical morphism f : SpecS −1 B → SpecB is an embedding
on the underlying topological spaces.)
Let X be a topological space. Then X is Hausdorff if and only if the diag-

onal {(x, x)|x ∈ X} is closed in X × X. The concept of separated morphisms
corresponds to the concept of Hausdorff topological spaces.
Let f : X → Y be a morphism of schemes. Define the diagonal morphism
∆X/Y : X → X ×Y X to be the unique morphism satisfying
p∆X/Y = q∆X/Y = idX .
We often denote ∆X/Y by ∆. We say f : X → Y is a separated morphism or
X is separated over Y if ∆X/Y is a closed immersion. A scheme X is called a
separated scheme if the canonical morphism X → SpecZ is separated.
Proposition 3.22. Let f : SpecB → SpecA be a morphism between affine

schemes. Then f is separated.
Proof. It is not hard to see that ∆ : SpecB → Spec(B ⊗A B) corresponds to

the homomorphism
B ⊗A B → B, b1 ⊗ b2 7→ b1 b2 ,
which is surjective. So ∆ is a closed immersion by Proposition 3.16 (i).

(i) The diagonal morphism ∆ : X → X ×Y X is an immersion.
(ii) f : X → Y is separated if and only if ∆(X) is a closed subset of X ×Y X.
Proof. (i) Cover Y by affine open subschemes Vi (i ∈ I) and cover each f −1 (Vi )
by affine open subschemes Uij (j ∈ Ji ). Let U be the union of p−1 (Uij ) ∩
q −1 (Uij ) = Uij ×Vi Uij (i ∈ I, j ∈ Ji ). Obviously ∆(X) is contained in U .
So we can factorize ∆ as X → U → X ×Y X. The second arrow is an open
immersion. To prove ∆ is an immersion, it suffices to show the first arrow is a
closed immersion. Since U can be covered by p−1 (Uij ) ∩ q −1 (Uij ) (i ∈ I, j ∈ Ji ),
it suffices to show each
∆
∆−1 (p−1 (Uij ) ∩ q −1 (Uij )) → p−1 (Uij ) ∩ q −1 (Uij )
is a closed immersion. We have a commutative diagram
∆
∆−1 (p−1 (Uij ) ∩ q −1 (Uij )) → p−1 (Uij ) ∩ q −1 (Uij )
k ↓∼
=
∆Uij /Vi
Uij → Uij ×Vi Uij .
By Proposition 3.22, the bottom horizontal arrow is a closed immersion. Our

assertion follows.
(ii) Taking into account of (i), it suffices to show an immersion f : X → Y
is a closed immersion if and only if f (X) is a closed subset of Y . The necessity
is trivial. To prove the sufficiency, we factorize f : X → Y as X → V → Y
such that the first arrow is a closed immersion and the second arrow is an open
immersion. Since f (X) is closed, Y has an open covering {V, Y − f (X)}. Note
f f
that f −1 (V ) → V and f −1 (Y − f (X)) → Y − f (X) are closed immersions since
the first morphism coincides with X → V and f −1 (Y − f (X)) = ∅. So f is a
closed immersion.
We leave the proof of the following proposition to the reader:
Proposition 3.24. Let

X Y
& .
T
↓
S
be a family of morphisms. Then the diagram
X ×T Y → X ×S Y
↓ ↓
∆
T → T ×S T
is cartesian, where the top horizontal arrow is the unique morphism whose
compositions with the two projections of X ×S Y to its factors are the two
projections of X ×T Y to its factors, and the right vertical arrow is the unique
morphism whose compositions with the two projections of T ×S T to its factors
are the morphism X → T and Y → T , respectively.
Let X and Y be two S-schemes and f : X → Y an S-morphism. Applying

the above proposition to
X Y
f& . id
Y
↓
S,
and noting that X ×Y Y = X, we get a cartesian diagram
X → X ×S Y
↓ ↓
∆
Y → Y ×S Y.
The top horizontal arrow is called the graph of the S-morphism f : X → Y and
is denoted by Γf : X → X ×S Y . It is the unique morphism satisfying
pΓf = idX , qΓf = f.
It is obtained from ∆Y /S by base change X ×S Y → Y ×S Y . Using Proposition

3.16 (ii), one can show that the base changes of immersions (resp. open immer-
sions, resp. closed immersions) are immersions (resp. open immersions, resp.
closed immersions). So by Proposition 3.23 (i), Γf is an immersion. When Y is
separated over S, Γf is a closed immersion.
A morphism f : X → Y of schemes is called quasi-separated if the diagonal

morphism ∆ : X → X ×Y X is quasi-compact. We say X is quasi-separated if
the canonical morphism X → SpecZ is quasi-separated.
Proposition 3.25. Let f : X → SpecA be a morphism from X to an affine

scheme and let U and V be two affine open subschemes of X.
(i) If f is separated, then U ∩V is affine and the ring OX (U ∩V ) is generated
by the images of the restrictions OX (U ) → OX (U ∩ V ) and OX (V ) → OX (U ∩
V ).
(ii) If f is quasi-separated, then U ∩ V can be covered by finitely many affine
open subschemes.
Proof. (i) Since the diagonal morphism ∆ : X → X ×SpecA X is a closed

immersion, we have a closed immersion
∆
U ∩ V = ∆−1 (p−1 (U ) ∩ q −1 (V )) → p−1 (U ) ∩ q −1 (V ) = U ×SpecA V.
But U ×SpecA V is affine, so by Proposition 3.16 (ii), U ∩ V is also affine and

the homomorphism OU ×SpecA V (U ×SpecA V ) → OX (U ∩ V ) is surjective. This
surjective homomorphism can be identified with the homomorphism
OX (U ) ⊗A OX (V ) → OX (U ∩ V )
induced by the A-algebra homomorphisms OX (U ) → OX (U ∩V ) and OX (V ) →

OX (U ∩ V ). So OX (U ∩ V ) is generated by their images.
(ii) Note that p−1 (U ) ∩ q −1 (V ) is an affine open subscheme of X ×SpecA
X. Since ∆ : X → X ×SpecA X is quasi-compact, U ∩ V = ∆−1 (p−1 (U ) ∩
q −1 (V )) is quasi-compact and hence can be covered by finitely many affine
open subschemes.
We say a morphism f : X → Y is proper or X is proper over Y if f satisfies

the following properties:
(i) f is of finite type.
(ii) f is separated.
(iii) For any morphism Y 0 → Y , the base change f 0 : X ×Y Y 0 → Y 0 of f is
a closed map on the underlying topological spaces. (In particular, f is a closed
map on the underlying topological spaces.) We say f is universally closed if it

has this property.
Proposition 3.26.
(i) A morphism f : X → Y is separated (resp. proper) if and only if there
exists an open covering {Vi }i∈I of Y such that each f −1 (Vi ) → Vi is separated
(resp. proper).
(ii) Immersions are separated. Closed immersions are proper.
(iii) The composition of two separated morphisms (resp. proper morphisms)
is separated (resp. proper.)
(iv) Let f : X → Y and Y 0 → Y be morphisms and let f 0 : X ×Y Y 0 → Y 0
be the base change of f . If f is separated (resp. proper), then f 0 is separated
(resp. proper.)
f0
X ×Y Y 0 → Y0
↓ ↓
f
X → Y.
(v) Let f : X → Y and f 0 : X 0 → Y 0 be two separated (resp. proper) S-

morphisms between S-schemes. Then f × f 0 : X ×S X 0 → Y ×S Y 0 is separated
(resp. proper), where f × f 0 is the unique morphism making the following
diagrams commute:
f ×f 0 f ×f 0
X ×S Y → X 0 ×S Y 0 X ×S Y → X 0 ×S Y 0
↓ ↓ ↓ ↓
f f0
X → X 0, Y → Y 0.
(vi) Let f : X → Y and g : Y → Z be two morphisms. If gf is separated,

then f is separated. If gf is proper and g is separated, then f is proper.
Sketch of the Proof. We leave to the reader to prove (i) and the assertions
about properness.
(ii) Let X → Y be an immersion. Using the fact that it is a monomorphism
in the category of S-schemes, one can show X ×Y X ∼= X and ∆ : X → X ×Y X
can be identified with idX . Hence ∆ is a closed immersion and X is separated
over Y .
(iii) Let X → Y and Y → Z be two morphisms. By Proposition 3.24, we
have a cartesian diagram
X ×Y X → X ×Z X
↓ ↓
∆Y /Z
Y → Y ×Z Y.
If Y → Z is separated, then ∆Y /Z is a closed immersion and hence its base

change X ×Y X → X ×Z X is a closed immersion. We have a commutative
diagram
∆X/Y
X → X ×Y X
∆X/Z & ↓
X ×Z X
If X → Y is also separated, then ∆X/Z is a closed immersion since it is the
composition of two closed immersions ∆X/Y and X ×Y X → X ×Z X. Hence
X → Z is separated.
(iv) Let X 0 = X ×Y Y 0 . We have a cartesian diagram
∆X 0 /Y 0
X0 → X 0 ×Y 0 X 0
↓ ↓
∆X/Y
X → X ×Y X.
So if ∆X/Y is a closed immersion, then ∆X 0 /Y 0 is a closed immersion.

(v) Suppose f and f 0 are separated. We have f × f 0 = (f × idY 0 )(idX × f 0 ).
To prove f × f 0 is separated, it suffices to show f × idY 0 and idX × f 0 are
separated by (iii). It is not hard to verify that
X ×S Y 0 → X
f × idY 0 ↓ ↓f
Y ×S Y 0 → Y
is cartesian, (that is, X ×Y (Y ×S Y 0 ) = X ×S Y 0 ). So f × idY 0 is the base

change of f and hence is separated. Similarly idX × f 0 is the base change of f 0
and hence is separated.
(vi) We have f = qΓf , where Γf : X → X ×Z Y is the graph of the Z-
morphism f and q : X ×Z Y → Y is the projection. Note that q is the base
change of gf and hence is separated. Γf is an immersion and hence is separated.
So their composition f is separated.
Consider the ring Z[x0 , . . . , xn ] with the grading defined by the degrees of
polynomials. Denote ProjZ[x0 , . . . , xn ] by PnZ . For any scheme Y , define the
projective space over Y to be the Y -scheme PnY = PnZ × Y . We say a morphism
f : X → Y is projective or X is projective over Y if f can be factorized as
X → PnY → Y
for some closed immersion X → PnY . It is called quasi-projective if it can be

factorized as above with X → PnY being an immersion.
Proposition 3.27. Projective morphisms are proper.
Proof. Taking into account of Proposition 3.26 (ii), (iii) and (iv), it suffices to
show PnZ → SpecZ is of finite type, separated, and universally closed.
Cover PnZ by D+ (xi ) ∼

= SpecZ[x0 , . . . , xn ](xi ) (i = 0, . . . , n). We have
x0 x0
Z[x0 , . . . , xn ](xi ) ∼
= Z[ , . . . , ].
xn xn
So Z[x0 , . . . , xn ](xi ) are finitely generated Z-algebras. Hence PnZ → SpecZ is of
finite type.
Cover PnZ × PnZ by p−1 (D+ (xi )) ∩ q −1 (D+ (xj )) (i, j = 0, . . . , n). We have
∆−1 (p−1 (D+ (xi )) ∩ q −1 (D+ (xj ))) = D+ (xi ) ∩ D+ (xj ) = D+ (xi xj )
∼
= SpecZ[x0 , . . . , xn ](xi xj ) ,
p−1 (D+ (xi )) ∩ q −1 (D+ (xj )) ∼ = D+ (xi ) × D+ (xj )
∼
= Spec(Z[x 0 , . . . , x ]
n (x ) ⊗ Z Z[x0 , . . . , xn ](x ) ).
i j
Since the canonical homomorphisms

xk xl xk xl
Z[x0 , . . . , xn ](xi ) ⊗Z Z[x0 , . . . , xn ](xj ) → Z[x0 , . . . , xn ](xi xj ) , ⊗ 7→
xi xj xi xj
are surjective, the morphisms
∆
∆−1 (p−1 (D+ (xi )) ∩ q −1 (D+ (xj ))) → p−1 (D+ (xi )) ∩ q −1 (D+ (xj ))
are closed immersions. So ∆ : PnZ → PnZ × PnZ is a closed immersion. Hence

PnZ → SpecZ is separated.
To prove PnZ → SpecZ is universally closed, it suffices to show the canonical
morphism ProjA[x0 , . . . , xn ] → SpecA is a closed map on the underlying topo-
logical spaces for any ring A. For any homogeneous ideal a of A[x0 , . . . , xn ],
to show the image of V+ (a) is closed in SpecA, it suffices to show the image of
ProjA[x0 , . . . , xn ]/a → SpecA is closed. Let S = A[x0 , . . . , xn ]/a and let x̄i be
the image of xi in S. Denote the canonical morphism ProjS → SpecA by f .
Suppose p ∈ SpecA does not lie in the image of f , then f −1 (p)∩D+ (x̄i ) is empty
for each i. We have f −1 (p) ∩ D+ (x̄i ) ∼
= Spec(S(x̄i ) ⊗A Ap /pAp ). By Proposition
2.2 (i), we have 0 = 1 in S(x̄i ) ⊗A Ap /pAp . So x̄i ⊗1 is nilpotent in S ⊗A Ap /pAp
for any i. Hence we may find a natural number N such that Sk ⊗A Ap /pAp = 0
for any k ≥ N . By Nakayama’s lemma, we have Sk ⊗A Ap = 0 for any k ≥ N
and hence AnnA (Sk ) 6⊂ p by Lemma 3.28 below. Conversely, if there exists a
natural number N such that for any k ≥ N , we have AnnA (Sk ) 6⊂ p, then we
can show p does not lie in the image of f by reversing the above argument. So
we have p 6∈ imf if and only if there exists a natural number N such that for any
k ≥ N , we have AnnA (Sk ) 6⊂ p. Obviously we have AnnA (Sk ) ⊂ AnnA (Sk+1 ).
So we have the following equivalence:
p ∈ imf
⇔ for any natural number N, there exists a k ≥ N such that AnnA (Sk ) ⊂ p
⇔ for any natural number N, we have AnnA (SN ) ⊂ p
⇔ ∪N AnnA (SN ) ⊂ p.
So imf = V (∪N AnnA (SN )), which is closed.
Lemma 3.28. For any finitely generated A-module M , we have
{p ∈ SpecA|Mp = 0} = SpecA − V (AnnA (M )),
where AnnA (M ) = {r ∈ A|rx = 0 for any x ∈ M } is the annihilator of M .
Proof. Let x1 , . . . , xn be a finite family of generators of M . If Mp = 0, then

for each i, we have x1i = 0 in Mp and hence there exists an si 6∈ p such that
si xi = 0. Let s = s1 · · · sn . Then s 6∈ p and sxi = 0 for every i and hence
s ∈ AnnA (M ). So AnnA (M ) 6⊂ p.
Conversely, if AnnA (M ) 6⊂ p, then there exists an s ∈ AnnA (M ) such that
s 6∈ p. For any xt ∈ Mp , we have xt = sx st = 0. So Mp = 0.
Proposition 3.29.
(i) Closed immersions are projective.
(ii) The composition of projective morphisms is projective.
(iii) Let f : X → Y and Y 0 → Y be morphisms and let f 0 : X ×Y Y 0 → Y 0
be the base change of f . If f is projective, then f 0 is projective.
(iv) Let f : X → Y and f 0 : X 0 → Y 0 be projective S-morphisms between
S-schemes. Then f × f 0 : X ×S X 0 → Y ×S Y 0 is projective.
(v) Let f : X → Y and g : Y → Z be morphisms. If gf is projective and g
is separated, then f is projective.
Proof. We only give the proof of (iii) and leave the rest to the reader. Let
f : X → Y and g : Y → S be projective morphisms. Then we have factorizations
X → Pm
Y →Y
n
Y → PS → S
such that X → Pm n
Y and Y → PS are closed immersions. Consider the following
diagram
X → Pm Y → Pm n
S ×S PS
& ↓ ↓
Y → PnS
& ↓
S.
Note that the square in the above diagram is cartesian since
(Pm n
S ×S PS ) ×Pn
S
Y ∼
= Pm ∼ m
S ×S Y = PY .
Hence the composition X → Pm m n

Y → PS ×S PS is a closed immersion. Combining
with Lemma 3.30 below, this proves gf is projective.
(m+1)(n+1)−1
Lemma 3.30. There exists a closed immersion Pm n
S ×S PS → PS
which is an S-morphism.
Proof. We treat the case S = SpecZ. The general case follows from this case
by base change. We have
Pm
Z = ProjZ[xi ], (i = 0, . . . , m),
PnZ = ProjZ[yj ], (j = 0, . . . , n),
(m+1)(n+1)−1
PZ = ProjZ[zij ], (i = 0, . . . , m, j = 0, . . . , n).
For any k ∈ {0, . . . , m} and l ∈ {0, . . . , n}, we have an epimorphism

zij xi yj
Z[zij ](zkl ) → Z[xi ](xk ) ⊗Z Z[yj ](yl ) , 7→ ⊗ .
zkl xk yl
It induces a closed immersion
D+ (xk ) × D+ (yl ) → D+ (zkl ).
For any k, k 0 ∈ {0, . . . , m} and l, l0 ∈ {0, . . . , n}, we have a commutative diagram
Z[zij ](zkl ) → Z[xi ](xk ) ⊗Z Z[yj ](yl )

↓ ↓
Z[zij ](zkl zk0 l0 ) → Z[xi ](xk xk0 ) ⊗Z Z[yj ](yl yl0 )
↑ ↑
Z[zij ](zk0 l0 ) → Z[xi ](xk0 ) ⊗Z Z[yj ](yl0 ) ,
zij zi0 j 0 xi xi0 y j yj 0
where the middle horizontal arrow is defined by zkl zk0 l0 7→ xk xk0 ⊗ y l yl0 . Hence
we have a commutative diagram
D+ (xk ) × D+ (yl ) → D+ (zkl )

↑ ↑
D+ (xk xk0 ) × D+ (yl yl0 ) → D+ (zkl zk0 l0 )
↓ ↓
D+ (xk0 ) × D+ (yl0 ) → D+ (zk0 l0 ).
So we can glue the closed immersions D+ (xk ) × D+ (yl ) → D+ (zkl ) together to

(m+1)(n+1)−1
get a closed immersion Pm n
Z × PZ → P Z .
(m+1)(n+1)−1
The closed immersion Pm n
S ×S PS → PS constructed in the proof
of Lemma 3.30 is called the Segre embedding.
1.4 Coherent Sheaves
Let (X, OX ) be a ringed space. A sheaf of OX -modules or an OX -module is a

sheaf F such that for every open subset U of X, F(U ) is an OX (U )-module,
and for every inclusion of open subsets V ⊂ U , the restriction F(U ) → F(V ) is
1.4. COHERENT SHEAVES 49
compatible with the module structures and the ring homomorphism OX (U ) →

OX (V ), that is, for any a ∈ OX (U ) and s, t ∈ F(U ), we have
(s + t)|V = s|V + t|V , (as)|V = a|V · s|V .
Let F and G be OX -modules. A morphism φ : F → G is a morphism of

sheaves such that for every open subset U , φ(U ) : F(U ) → F(U ) is a homomor-
phism of OX (U )-modules. The set of all morphisms from F to G is denoted by
HomOX (F, G). It is an OX (X)-module.
If we let OX to be the sheaf associated to the constant presheaf Z, then a
sheaf of OX -modules is nothing but a sheaf of abelian groups.
Define HomOX (F, G) to be the OX -module
U 7→ HomOX |U (F|U , G|U ),
and define F ⊗OX G to be the OX -module associated to the presheaf
U 7→ F(U ) ⊗OX (U ) G(U ).
For any P ∈ X, we have
(F ⊗OX G)P = FP ⊗OX,P GP .
Given a family of OX -modules Fi (i ∈ I), their direct sum in the category

of OX -modules is the sheaf associated to the presheaf
M
U 7→ Fi (U ).
i∈I
When X is a noetherian topological space, this presheaf is a sheaf. The direct

product of Fi (i ∈ I) in the category of OX -modules is the sheaf
Y
U 7→ Fi (U ).
i∈I
Let (I, ≤) be a direct set. For any direct system (Fi , φij )i∈I of OX -modules,
its direct limit in the category of OX -modules is the sheaf associated to the
presheaf
U 7→ dir. lim Fi (U ).
i
If X is a noetherian topological space, then this presheaf is a sheaf. If (Fi , φji )i∈I
is an inverse system of OX -modules, then its inverse limit in the category of OX -
modules is the sheaf
U 7→ inv. lim Fi (U ).
i
An OX -module is called free if it is isomorphic to a direct sum of some

copies of OX . The number of copies is called the rank of the free OX -module.
An OX -module F is called locally free if there exists an open covering {Ui }i∈I
of X such that each F|Ui is a free OUi -module. For a locally free OX -module
F, we define its rank to be the function P 7→ rankOX,P FP on X. It is a locally

constant function. A locally free OX -module is called invertible if it has rank
one. Tensor products of invertible OX -modules are invertible. Moreover, if L is
an invertible OX -module, then HomOX (L, OX ) is also invertible, and we have
a canonical isomorphism
L ⊗OX HomOX (L, OX ) ∼

= OX .
So the isomorphic classes of invertible OX -modules form a group with respect

to tensor product. We call this group the Picard group and denote it by Pic(X).
Here is an example: Let X be a topological space and C the sheaf of complex

valued continuous functions. Then (X, C) is a ringed space. For any complex
vector bundle p : E → X over X, the sheaf of sections defined by
U 7→ {s : U → p−1 (U )|s is continuous and ps = id}
is a locally free sheaf of C-modules.
An OX -module F is called of finite presentation if we have an exact sequence

of the form
m n
OX → OX →F →0
m n
for some natural numbers m and n, where OX and OX are the free OX -modules
of ranks m and n, respectively. F is called locally of finite presentation if there
exists an open covering {Ui }i∈I of X such that each F|Ui has finite presentation.
An OX -module F is called of finite type if we have an exact sequence of the
form
n
OX →F →0
for some natural number n. F is called locally of finite type if there exists an
open covering {Ui }i∈I of X such that each F|Ui is of finite type.
Proposition 4.1. Let (X, OX ) be a ringed space and let F and G be two
OX -modules.
(i) If F is locally of finite presentation, then for any P ∈ X, we have
(HomOX (F, G))P ∼

= HomOX,P (FP , GP ).
(ii) Suppose F and G are locally of finite presentation. If there exists an

isomorphism φP : FP → GP for some P ∈ X, then there exists a neighborhood U
of P and an isomorphism φ : F|U → G|U such that φ induces the homomorphism
φP on stalks.
Proof. (i) We define (HomOX (F, G))P → HomOX,P (FP , GP ) in the obvious
way. To prove our assertion, we may assume F has finite presentation. Let
m n
OX → OX →F →0
be an exact sequence. Then we have the following commutative diagram:

n m
0 → (HomOX (F, G))P → (HomOX (OX , G))P → (HomOX (OX , G))P
↓ ↓ ↓
n m
0 → HomOX (FP , GP ) → HomOX (OX,P , GP ) → HomOX (OX,P , GP ).
The last two vertical arrows are obviously isomorphisms. So the first vertical
arrow is also an isomorphism.
(ii) φP is an element in HomOX,P (FP , GP ). By (i), there exists a neighbor-
hood U of P and a morphism φ : F|U → G|U which induces the homomorphism
φP on stalks. Let ψP : GP → FP be the inverse of φP . Shrinking U if necessary,
we may find a morphism ψ : G|U → G|U which induces the homomorphism ψP
on stalks. Since ψP φP = idFP and φP ψP = idGP , shrinking U if necessary, we
have ψφ = idF |U and φψ = idG|U . Hence φ is an isomorphism.
Let (f, f ] ) : (X, OX ) → (Y, OY ) be a morphism of ringed spaces and F an

OX -module. Then f∗ F is an f∗ OX -module. Through the morphism of sheaves
of rings f ] : OY → f∗ OX , f∗ F becomes an OX -module. We call it the direct
image of F. Let G be an OY -module. Then f −1 G is an f −1 OY -module. The
morphism f ] : OY → f∗ OX induces a morphism f −1 OY → OX . We define the
inverse image of G to be the OX -module
f ∗ G = OX ⊗f −1 OY f −1 G.
We leave to the reader to construct a bijection
HomOY (G, f∗ F) ∼
= HomOX (f ∗ G, F)
which is functorial with respect to F and G. This show that f ∗ is left adjoint
to f∗ . For any P ∈ X, we have
(f ∗ G)P = (OX ⊗f −1 OY f −1 G)P = OX,P ⊗(f −1 OY )P (f −1 G)P

= OX,P ⊗OY,f (P ) Gf (P ) ,
that is,
(f ∗ G)P = OX,P ⊗OY,f (P ) Gf (P ) .
Let A be a ring and M an A-module. Define a sheaf M ∼ of OSpecA -modules

∼
as follows:
` For any open subset U of SpecA, M (U ) consists of those functions
s : U → p∈SpecA Mp satisfying the following two conditions:
(a) For any p ∈ U , we have s(p) ∈ Mp .
m ∈ M and f ∈ A such that for any q ∈ Up , we have f 6∈ q and s(q) = m f in
Mq .
For any inclusion of open subsets V ⊂ U , we define M ∼ (U ) → M ∼ (V ) to be
the restriction of functions.
Obviously we have A∼ = OSpecA .
Proposition 4.2.
(i) (M ∼ )p ∼ = Mp for any p ∈ SpecA.
(ii) M ∼ (D(f )) ∼ = Mf for any f ∈ A. In particular, taking f = 1, we
get M ∼ (SpecA) ∼ = M . Moreover, through the isomorphism D(f ) ∼
= SpecAf ,
M ∼ |D(f ) is identified with Mf∼ .
(iii) A sequence of A-modules
M 0 → M → M 00
is exact if and only if the sequence of OSpecA -modules
M 0∼ → M ∼ → M 00∼
is exact.
(iv) For any A-modules M and N , we have
HomOSpecA (M ∼ , N ∼ ) ∼
= HomA (M, N ),
M ∼ ⊗OSpecA N ∼ ∼
= (M ⊗A N )∼ .
If M is an A-module with finite presentation, then we have
= (HomA (M, N ))∼ .
(v) For any family Mi (i ∈ I) of A-modules, we have
⊕i Mi∼ ∼
= (⊕i Mi )∼ .
(vi) For any direct system (Mi , φij )i∈I of A-modules, we have
dir. lim Mi∼ ∼

= (dir. lim Mi )∼ .
i i
Proof. We only prove (iv) and leave to the reader to prove the rest. Define
α : HomOSpecA (M ∼ , N ∼ ) → HomA (M, N )
as follows: For any morphism φ : M ∼ → N ∼ , we define α(φ) to be the compo-

sition
φ
M∼ = M ∼ (SpecA) → N ∼ (SpecA) ∼= N.
Define
β : HomA (M, N ) → HomOSpecA (M ∼ , N ∼ )
as follows: Given a homomorphism φ : M → N , for any p ∈ SpecA, let φp :
`p → Np be the ∼homomorphism induced by φ. Foràny section s : U
M
∼
→
p∈SpecA pM of M (U ), define a section β(φ)(s) : U → N
p∈SpecA p of N (U )
by
(β(φ)(s))(p) = φp (s(p))
for any p ∈ U . We thus get a morphism β(φ) : M ∼ → N ∼ . One can verify that
α and β are inverse to each other. So
= HomA (M, N ).
Define a morphism
γ : M ∼ ⊗OSpec A N ∼ → (M ⊗A N )∼
as follows: Since M ∼ ⊗OSpec A N ∼ is the sheaf associated to the presheaf
U 7→ M ∼ (U ) ⊗OSpecA (U ) N ∼ (U ),
∼
` a morphism from this presheaf to`(M ⊗A N ) . For any
it suffices to define
sections s : U 7→ p∈SpecA Mp in M ∼ (U ) and t : U 7→ p∈SpecA Np in N ∼ (U ),
define a section γ(s ⊗ t) : U 7→ p∈SpecA (M ⊗A N )p in (M ⊗A N )∼ (U ) so
`
that for any p ∈ U , γ(s ⊗ t)(p) is the image of s(p) ⊗ t(p) under the canonical
isomorphism
Mp ⊗Ap Np ∼ = (M ⊗A N )p .
One can verify that the homomorphism induced by γ on stalks at p is this
canonical isomorphism. So γ is an isomorphism.
For any p ∈ SpecA, we have a canonical homomorphism
(HomA (M, N ))p → HomAp (Mp , Np ).
Suppose M has finite presentation. Let
Am → An → M → 0
be an exact sequence. Then we have the following commutative diagram:
0 → (HomA (M, N ))p → (HomA (An , N ))p → (HomA (Am , N ))p

↓ ↓ ↓
0 → HomAp (Mp , Np ) → HomAp (Anp , Np ) → HomAp (Amp , Np ).
The last two vertical arrows are obviously isomorphisms. So we have
(HomA (M, N ))p ∼

= HomAp (Mp , Np ).
Similarly for any f ∈ A, we have a canonical isomorphism
(HomA (M, N ))f ∼

= HomAf (Mf , Nf ).
Define a homomorphism
δ : HomOSpecA (M ∼ , N ∼ ) → (HomA (M, N ))∼
as follows: Given a section φ : M ∼ |U → N ∼ |U in HomOSpecA (M ∼ , N ∼ )(U ),

let φp : Mp → Np be the homomorphism induced by φ on stalks for any
`
p ∈ U . We define a map δ(φ) : U → p∈SpecA (HomA (M, N ))p so that for
any p ∈ U , (δ(φ))(p) is the image of φp : Mp → Np under the canonical
isomorphism HomAp (Mp , Np ) ∼= (HomA (M, N ))p . We claim δ(φ) is a section
∼
in (HomA (M, N )) (U ). To see this, we cover U by open subsets of the form
D(f ) for some f ∈ A. Then φ|D(f ) is a section in HomOSpecA (M ∼ , N ∼ )(D(f )).
But we have
HomOSpecA (M ∼ , N ∼ )(D(f )) = HomOSpecA |D(f ) (M ∼ |D(f ) , N ∼ |D(f ) )

∼
= HomOSpecAf (Mf∼ , Nf∼ ) ∼
= HomAf (Mf , Nf ) ∼
= (HomA (M, N ))f .
So φ|D(f ) corresponds to an element in (HomA (M, N ))f . Hence δ(φ) is a

section of (HomA (M, N ))∼ (U ). In this way, we get a morphism of sheaves
δ : HomOSpecA (M ∼ , N ∼ ) → (HomA (M, N ))∼ . One can verify it induces the
isomorphism HomAp (Mp , Np ) ∼ = (HomA (M, N ))p on stalks at p. So δ is an
isomorphism.
Proposition 4.3. Let φ : A → B be a homomorphism of rings and let f :

SpecB → SpecA be the corresponding morphism.
(i) For any B-module N , we have f∗ N ∼ ∼
= (A N )∼ , where on the right-hand
side, N is regarded as an A-module.
(ii) For any A-module M , we have f ∗ M ∼ ∼
= (B ⊗A M )∼ .
Proof. (i) Define

α : (A N )∼ → f∗ N ∼
as follows: For any section s : U → p∈SpecA (A N )p of (A N )∼ , we define a
`
section α(s) : f −1 (U ) → q∈SpecB Nq of (f∗ N ∼ )(U ) = N ∼ (f −1 (U )) so that

`
for any q ∈ f −1 (U ), (α(s))(q) is the image of s(φ−1 (q)) under the canonical
homomorphism
n n
(A N )φ−1 (q) → Nq , → .
s φ(s)
To prove α is an isomorphism, it suffices to show that for any a ∈ A, the
α
homomorphism (A N )∼ (D(a)) → (f∗ N ∼ )(D(a)) is an isomorphism. We have
(A N )∼ (D(a)) = (A N )a ,
(f∗ N ∼ )(D(a)) = N ∼ (f −1 (D(a)) = N ∼ (D(φ(a))) = Nφ(a) .
Our assertion follows from the fact that

n n
(A N )a → Nφ(a) , → k
ak φ(a)
is an isomorphism.
(ii) The A-module homomorphism
M →A (B ⊗A M ), m 7→ 1 ⊗ m
induces a morphism
M ∼ → (A (B ⊗A M ))∼ ∼
= f∗ (B ⊗A M )∼ .
Since f ∗ is left adjoint to f∗ , we get a morphism
f ∗ M ∼ → (B ⊗A M )∼ .
To prove it is an isomorphism, it suffices to prove it induces isomorphisms on

stalks. For any q ∈ SpecB, we have
(f ∗ M ∼ )q = OSpecB,q ⊗OSpecA,f (q) Mf∼(q) = Bq ⊗Aφ−1 (q) Mφ−1 (q) ,

(B ⊗A M )∼
q = (B ⊗A M )q .
Our assertion follows from the fact that
Bq ⊗Aφ−1 (q) Mφ−1 (q) ∼

= (B ⊗A M )q .
Let (X, OX ) be a scheme. A sheaf of OX -modules F is called quasi-coherent

if X can be covered by some affine open subschemes Ui = SpecAi (i ∈ I) such
that F|Ui ∼= Mi∼ for some Ai -modules Mi . When X is noetherian, we say F
is coherent if the above Mi can be chosen to be finitely generated. We leave to
the reader to verify that for any open subscheme U , F|U is quasi-coherent if F
is quasi-coherent, and when X is noetherian, F|U is coherent if F is coherent.
Lemma 4.4. Let F a quasi-coherent sheaf on an affine scheme SpecA and let
f ∈ A.
(i) For any section s ∈ F(SpecA) such that s|D(f ) = 0, there exists a natural
number n such that f n s = 0.
(ii) For any section t ∈ F(D(f )), there exists a natural number n such that
f n t can be extended to a section in F(SpecA).
(iii) Let M = F(X). Then Mf ∼ = F(D(f )).
Proof. Cover SpecA by finitely many open affine subschemes Ui = SpecAi

(i ∈ I) so that F|Ui = Mi∼ for some Ai -modules Mi . For each i, let fi ∈ Ai be
the image of f under the restriction O(SpecA) = A → O(SpecAi ) = Ai . Then
D(f ) ∩ Ui coincides with the open subset D(fi ) of Ui = SpecAi . So
F(D(f ) ∩ Ui ) = Mi∼ (D(fi )) = (Mi )fi .
Suppose s ∈ F(SpecA) such that s|D(f ) = 0. Then for each i, s|Ui lies in the
kernel of the restriction F(Ui ) → F(D(f ) ∩ Ui ). This restriction coincides with
the canonical homomorphism Mi → (Mi )fi . Since every element in the kernel
of Mi → (Mi )fi is annihilated by some power of fi , we have fini s|Ui = 0 for
some natural number ni . Since the covering {Ui }i∈I is a finite covering, there
exists a natural number n such that fin s|Ui = 0 for every i. Hence f n s = 0.
This proves (i).
Suppose t ∈ F(D(f )). Then t|D(f )∩Ui ∈ F(D(f ) ∩ Ui ). Since the restriction
F(Ui ) → F(D(f ) ∩ Ui ) coincides with the canonical homomorphism Mi →
(Mi )fi , there exists a natural number n such that fin t|D(f )∩Ui can be extended
to a section ti in F(Ui ) for any i. Note that
ti |D(f )∩Ui ∩Uj = tj |D(f )∩Ui ∩Uj
for any i, j. Moreover Ui ∩ Uj is affine by Proposition 3.25 (i). Applying (i) to

the quasi-coherent sheaf F|Ui ∩Uj , we get a natural number m such that
fim ti |Ui ∩Uj = fjm tj |Ui ∩Uj
for any i, j. So we may glue fim ti together to get a section in F(SpecA). It is

an extension of f m+n t. This proves (ii).
We have an obvious homomorphism Mf → F(D(f )). It is injective by (i)
and surjective by (ii). So it is an isomorphism.
Proposition 4.5. Let X be a scheme and F a quasi-coherent sheaf on X. Then

for any affine open subscheme U = SpecA of X, there exists an A-module M
such that F|U ∼ = M ∼ . If X is noetherian and F is coherent, then M is finitely
generated.
Proof. Replacing X by U and F by F|U , we may assume X = SpecA. Let

M = F(X). Define an isomorphism φ : M ∼ → F as follows: For any f ∈ A,
φ(D(f )) is given by
s 1
φ(D(f )) : M ∼ (D(f )) = Mf → F(D(f )), 7→ n s|D(f ) ,
fn f
where s ∈ M = O(SpecA). By Lemma 4.4 (iii), φ(D(f )) is an isomorphism.

This defines φ(V ) : M ∼ (V ) → F(V ) for any open subset of the form V = D(f ).
In general, we cover V by open subsets D(fi ) (i ∈ I). We have an exact sequence
Y Y
0 → F(V ) → F(D(fi )) → F(D(fi ) ∩ D(fj )),
i∈I i,j∈I
where the second arrow is defined by s 7→ (s|D(fi ) ) and the third arrow is defined
by (si ) 7→ (sj |D(fi )∩D(fj ) − si |D(fi )∩D(fj ) ). We have a similar exact sequence for
M ∼ . We define φ(V ) so that the following diagram commutes:
0 → M ∼ (V ) → ∼ ∼
Q Q
i∈I M (D(fi )) → i,j∈I M (D(fi ) ∩ D(fj ))
φ(V ) ↓ Q ↓ Q ↓
0 → F(V ) → i∈I F(D(fi )) → i,j∈I F(D(fi ) ∩ D(fj )),
where the second and the third vertical arrows are induced by φ(D(fi )) and
φ(D(fi fj )), respectively. One can verify that φ(V ) is independent of the choice
of the open covering of V . (If {D(fi )}i∈I and {D(fj )}j∈J are two coverings of
V , then the φ(V )’s defined by both coverings coincide with the φ(V ) defined by
the covering D(fi fj ) (i ∈ I, j ∈ J).)
Suppose A is noetherian and F is coherent. Then we may cover SpecA by
finitely many affine open subschemes Ui = SpecAi (i ∈ I) such that we have
F|Ui ∼ = Mi∼ for some finitely generated Ai -modules Mi . Cover each Ui by
finitely many open subsets D(fij ) (j ∈ Ji ) for some fij ∈ A. Denote the image
0
of each fij under the restriction OX (SpecA) = A → OX (Ui ) = Ai by fij . Then
we have
F(D(fij )) ∼
= Mfij ,
F(D(fij )) ∼ 0
= F|Ui (D(fij )) ∼
= Mi∼ (D(fij
0
)) ∼
= (Mi )fij0 .
So we have Mfij ∼ = (Mi )fij0 . Similarly, we have Afij ∼

= (Ai )fij0 . Since each Mi
is a finitely generated Ai -module, each (Mi )fij0 is a finitely generated (Ai )fij0 -
module. Hence each Mfij is a finitely generated Afij -module. Moreover, D(fij )
(i ∈ I, j ∈ Ji ) form an open covering of SpecA. These imply that M is a finitely
generated A-module. (Confer the proof of Proposition 3.11.)
Corollary 4.6. Let X be a scheme and let F and G be two quasi-coherent

OX -modules.
(i) F ⊗OX G is quasi-coherent. If F locally has finite presentation, then
HomOX (F, G) is quasi-coherent.
(ii) Let φ : F → G be a morphism. Then kerφ, cokerφ and imφ are quasi-
coherent.
If X is noetherian, the same statements hold for coherent sheaves.
Proof. We may assume X = SpecA is affine. Then F = M ∼ and G = N ∼ for

some A-modules M and N . By Proposition 4.2 (iv), we have
F ⊗OX G ∼
= (M ⊗A N )∼ .
So F ⊗OX G is quasi-coherent. If F locally has finite presentation, then using
Proposition 4.2 (iii), (iv), one can show M is an A-module with finite presenta-
tion. By Proposition 4.2 (iv), we have
HomOX (F, G) ∼
= (HomA (M, N ))∼ .
So HomOX (F, G) is quasi-coherent. Let φ : F → G be a morphism. By Propo-
sition 4.2 (iv), it is induced by some homomorphism ψ : M → N . By Proposi-
tion 4.2 (iii), kerφ, cokerφ and imφ are isomorphic to (kerψ)∼ , (cokerψ)∼ and
(imψ)∼ , respectively. So they are quasi-coherent. When X is noetherian and
F and G are coherent, all the modules above are finitely generated and hence
define coherent sheaves.
Recall that an A-module P is called projective if for any epimorphism M →

N of A-modules, any homomorphism P → N can be lifted to a homomorphism
P → M . Suppose
p
0 → M 0 → M → M 00 → 0
is an exact sequence of A-modules. If M 00 is projective, then the exact sequence

splits. Indeed, we can lift the homomorphism idM 00 to a homomorphism i :
M 00 → M . We have pi = idM 00 . Our assertion then follows from Proposition
1.3. For any projective module P , we may find epimorphism φ : F → P for
some free A-module F . The exact sequence
0 → kerφ → F → P → 0
splits. So P is a direct factor of a free module. Conversely, one can show any
direct factor of a free module is projective.
Proposition 4.7. Let M be an A-module with finite presentation. Then M is

a projective A-module if and only if M ∼ is a locally free OSpecA -module.
Proof. Suppose M ∼ is locally free. Let’s prove M is projective. Given any

exact sequence
N → N0 → 0
of A-modules, we have an exact sequence
∼
N∼ → N0 →0
of OSpecA -modules. Since M ∼ is locally free and its rank is necessarily finite,
the sequence
∼
HomOSpecA (M ∼ , N ∼ ) → HomOSpecA (M ∼ , N 0 ) → 0
is exact. (To see this, we may work locally and hence assume M ∼ is free with
finite rank. The assertion is then obvious.) So by Proposition 4.2 (iv), the
following sequence is exact:
(HomA (M, N ))∼ → (HomA (M, N 0 ))∼ → 0.
By Proposition 4.2 (iii), the sequence
HomA (M, N ) → HomA (M, N 0 ) → 0
is exact. Hence M is projective.

Conversely, suppose M is projective, that is, M is a direct factor of a free
A-module. Then for every p ∈ SpecA, Mp is also a direct factor of a free Ap -
module and hence is projective. By Lemma 4.8 below, Mp is free. So we have
an isomorphism Mp ∼ = Anp for some natural number n. By Proposition 4.1 (ii),
there exists a neighborhood U of p such that M ∼ |U ∼ = OSpecA |U . So M ∼ is
locally free.
Lemma 4.8. Let A be a local ring and M an A-module with finite presentation.
Then M is projective if and only if M is free.
Proof. The sufficiency is obvious. Let’s prove the necessity. Suppose M is

projective. Let m be the maximal ideal of A. Choose x1 , . . . , xn so that their
images in M/mM form a basis for the A/m-vector space M/mM . Consider the
homomorphism
An → M, ei 7→ xi (i = 1, . . . , n),
where for each i, ei is the element in An whose i-th component is 1 and whose
other components are 0. By Nakayama’s lemma, An → M is surjective. Let R
be its kernel. Since M is projective, the exact sequence
0 → R → An → M → 0
splits. This implies that
0 → R/mR → (A/mA)n → M/mM → 0
is split exact. By the choice of An → M , the homomorphism (A/m)n → M/mM

is an isomorphism. So R/mR = 0. Since M has finite presentation, R is finitely
generated. (We leave to the reader to verify this fact.) By Nakayama’s lemma,
we have R = 0. So An → M is an isomorphism and hence M is free.

(i) If G is a quasi-coherent OY -module, then f ∗ G is a quasi-coherent OX -
module.
(ii) If X and Y are noetherian and G is coherent, then f ∗ G is coherent.
(iii) If f is quasi-compact and quasi-separated, and F is a quasi-coherent
OX -module, then f∗ F is a quasi-coherent OY -module.
Proof. The problem is local with respect to Y . We may assume Y = SpecA

is affine. Cover X by affine open subschemes Ui = SpecBi (i ∈ I). For each i,
denote by fi the morphism Ui → Y induced by f .
(i) If G is a quasi-coherent OY -module, then G ∼= M ∼ for some A-module
∼ ∗ ∼
M . For each i, we have f G|Ui = fi G = (Bi ⊗A M )∼ by Proposition 4.3 (ii).
∗
So f ∗ G is quasi-coherent.
(ii) If X and Y are noetherian and G is coherent, then M is a finitely gener-
ated A-module and hence each Bi ⊗A Mi is a finitely generated Bi -module. So
f ∗ G is coherent.
(iii) If f is quasi-compact and quasi-separated, then we can cover X by
finitely many affine open subscheme Ui (i ∈ I), and by Proposition 3.25 (ii), each
Ui ∩ Uj can be covered by finitely many affine open subschemes Uijk (k ∈ Kij ).
Denote by fijk the morphism Uijk → Y induced by f . For any open subset V
of Y , we have an exact sequence,
Y Y
0 → F(f −1 (V )) → F(f −1 (V ) ∩ Ui ) → F(f −1 (V ) ∩ Uijk ),
i∈I i,j∈I,k∈Kij
where the second arrow is defined by s 7→ (s|f −1 (V )∩Ui ) and the third arrow
is defined by (si ) →
7 (sj |f −1 (V )∩Uijk − si |f −1 (V )∩Uijk ). So we have an exact
sequence Y Y
0 → f∗ F → fi ∗ (F|Ui ) → fijk ∗ (F|Uijk ).
i∈I i,j∈I,k∈Jij
Q Q
By Proposition 4.2 (v) and 4.3 (i), i∈I fi ∗ (F|Ui ) and i,j∈I,k∈Jij fijk ∗ (F|Uijk )
are quasi-coherent. Hence f∗ F is quasi-coherent.
Proposition 4.10. Let f : X → S be an affine morphism and let F and G be

quasi-coherent OX -modules.
(i) We have a bijection
HomOX (F, G) ∼
= Homf∗ OX (f∗ F, f∗ G),
where the right-hand side is the set of morphisms of f∗ OX -modules.
(ii) Let φ : F → G be a morphism. Then φ is an isomorphism if and only if
the morphism f∗ φ : f∗ F → f∗ G induced by φ is an isomorphism.
Proof. (i) Any morphism φ : F → G of OX -modules induces a morphism

f∗ φ : f∗ F → f∗ G of f∗ OX -modules. We define
F : HomOX (F, G) → Homf∗ OX (f∗ F, f∗ G)
by F (φ) = f∗ φ. Let’s prove it is bijective. First we consider the case where
S is affine. Then X is also affine. By Proposition 4.3 (i), f∗ F and f∗ G are
quasi-coherent. By Proposition 4.2 (iv), we have
HomOX (F, G) = HomOX (X) (F(X), G(X))
= Hom(f∗ OX )(S) ((f∗ F)(S), (f∗ G)(S))
= Homf∗ OX (f∗ F, f∗ G).
This proves our assertion in the affine case. In general, we cover S by affine
open subschemes Vi (i ∈ I) and cover each Vi ∩ Vj by affine open subschemes
Vijk (k ∈ Kij ). We have a commutative diagram
HomOX (F, G)(f −1 (Vi )) → HomOX (F, G)(f −1 (Vijk ))
Q Q
Q ↓ Q ↓
Homf∗ OX (f∗ F, f∗ G)(Vi ) → Homf∗ OX (f∗ F, f∗ G)(Vijk ),
where the horizontal arrows are defined by

(φi ) 7→ (φj |f −1 (Vijk ) − φi |f −1 (Vijk ) )
and
(φi ) 7→ (φj |Vijk − φi |Vijk ),
respectively. Both vertical arrows are isomorphisms by the affine case treated
above. Since HomOX (F, G) and Homf∗ OX (f∗ F, f∗ G) can be identified with the
kernel of the horizontal arrows, we have
HomOX (F, G) ∼
= Homf∗ OX (f∗ F, f∗ G).
(ii) The “only if” part is obvious. Let’s prove the “if” part. Suppose f∗ φ :
f∗ F → f∗ G is an isomorphism. By (i), there exists a morphism ψ : G → F such
that f∗ ψ is the inverse of f∗ φ. We have
f∗ (ψφ) = idf∗ F = f∗ (idF ).
So ψφ = idF by (i). Similarly, φψ = idG . Hence φ is an isomorphism.
Proposition 4.11. Let f : X → S and g : Y → S be morphisms of schemes.

Assume g is affine. Then we have a bijection
HomS (X, Y ) ∼
= HomOS (g∗ OY , f∗ OX ),
where the left-hand side is the set of S-morphisms from X to Y , and the right-
hand side is the set of morphisms of sheaves of OS -algebras from g∗ OY to f∗ OX .
Proof. Define
F : HomS (X, Y ) → HomOS (g∗ OY , f∗ OX )
so that for any S-morphism h : X → Y , F (h) is the composition
g∗ (h] )
g∗ OY → g∗ h∗ OX = f∗ OX .
Fix a covering {Vi }i∈I of S by affine open subschemes. Then each g −1 (Vi ) is
affine. So by Proposition 2.8, we have
HomVi (f −1 (Vi ), g −1 (Vi )) = HomOS (Vi ) (OY (g −1 (Vi )), OX (f −1 (Vi ))).
Define a map
G : HomOS (g∗ OY , f∗ OX ) → HomS (X, Y )
as follows: Given a morphism φ : g∗ OY → f∗ OX of OS -algebras, for each i,
define hi : f −1 (Vi ) → g −1 (Vi ) to be the Vi -morphism corresponding to the
OS (Vi )-algebra homomorphism
φ
OY (g −1 (Vi )) = (g∗ OY )(Vi ) → (f∗ OX )(Vi ) = OX (f −1 (Vi )).
One can verify hi |f −1 (Vi )∩f −1 (Vj ) = hj |f −1 (Vi )∩f −1 (Vj ) . So we can glue hi to-
gether to get an S-morphism h : X → Y . Define G(φ) = h. One can verify that
F and G are inverse to each other.
Let f : X → S be an affine morphism. By Proposition 4.3 (ii), the sheaf of

OS -algebras f∗ OX is a quasi-coherent OS -module. Conversely, we have
Proposition 4.12. Let S be a scheme and let B be a sheaf of OS -algebras

which is quasi-coherent as an OS -module. (We call such B a quasi-coherent
OS -algebra.) Then there exists an affine morphism f : X → S unique up to
isomorphism such that f∗ OX ∼ = B. Moreover, for any sheaf F of B-modules

which is quasi-coherent as an OS -module, there exists a quasi-coherent OX -
module F 0 unique up to isomorphism such that f∗ F 0 ∼= F as sheaves of B-
modules.
Proof. Suppose f1 : X1 → S and f2 : X2 → S are two affine morphisms

such that fi∗ OX ∼
= B. By Proposition 4.11, there exists a unique morphism
g : X1 → X2 such that f2 g = f1 and the following diagram commutes:
f2∗ (g ] )
f2∗ OX2 → f1∗ OX1
∼
=↓ ↓∼
=
B = B.
Similarly, there exists a unique morphism h : X2 → X1 such that f1 h = f2 and
f1∗ (h] )
f1∗ OX1 → f2∗ OX2
∼
=↓ ↓∼
=
B = B.
Using Proposition 4.11 again, one can verify that hg = idX1 and gh = idX2 . This
proves the uniqueness of the affine morphism f : X → S satisfying f∗ OX ∼ = B.
Similarly, using Proposition 4.10, one can prove the uniqueness of the quasi-
coherent OX -module F 0 satisfying f∗ F 0 ∼ = F.
Cover S by affine open subschemes Vi = SpecAi (i ∈ I). Then B|Vi ∼ = Bi∼
for some Ai -modules Bi . By Proposition 4.2 (iv) the morphism B ⊗OS B →
B which gives the multiplication of the OS -algebra B defines homomorphisms
Bi ⊗Ai Bi → Bi . These homomorphisms define multiplications on Bi which
are commutative and associative. So Bi are Ai -algebras. Similarly we have
F|Vi = Mi∼ for some Bi -modules Mi . For each i, let Xi = SpecBi , let Fi0 be
the OXi -module defined by the Bi -module Mi , and let fi : Xi → Vi be the
morphism defined by the Ai -algebra structure on Bi . We have fi ∗ OXi ∼ = B|Vi
and fi ∗ Fi0 ∼ = F|Vi . So for any i, j, we have (fi ∗ OXi )|Vi ∩Vj ∼
= (fj ∗ OXj )|Vi ∩Vj and
(fi ∗ Fi0 )|Vi ∩Vj ∼
= (fj ∗ Fj0 )|Vi ∩Vj . By Proposition 4.11, we have
fi−1 (Vi ∩ Vj ) ∼
= fj−1 (Vi ∩ Vj )
as schemes over Vi ∩ Vj , and by Proposition 4.10 we have
Fi0 |f −1 (Vi ∩Vj ) ∼
= Fj0 |fj−1 (Vi ∩Vj ) .
i
We leave to the reader to show that these isomorphisms define glueing data for
Xi and Fi0 . Glueing them together, we get X and F 0 with the required property.
Let (X, OX ) be a ringed space. A sheaf of ideals of OX is an OX -module

which is subsheaf of OX . For any OX -module F, we define IF to be the image
of the morphism
I ⊗OX F → OX ⊗OX F ∼ = F.
Let (f, f ] ) : (X, OX ) → (Y, OY ) be a morphism of ringed spaces and let J be a

sheaf of ideals of OY . We define J OX to be the image of the morphism
f ∗ J → f ∗ OY ∼
= OX .
It is a sheaf of ideals of OX . More generally, for any OX -module F, we define

J F to be the image of the morphism
f ∗ J ⊗OX F → f ∗ OY ⊗OX F ∼
= F.
Obviously, we have
J F = (J OX )F.
Proposition 4.13. Let (f, f ] ) : (X, OX ) → (Y, OY ) be a morphism of ringed

spaces and let J be a sheaf of ideals of OY . Then for any OX -module F, we
have
F ⊗OX f ∗ (OY /J ) ∼= F/J F.
Proof. We have an exact sequence
0 → J → OY → OY /J → 0
and hence an exact sequence
0 → f −1 J → f −1 OY → f −1 (OY /J ) → 0.
Tensoring with F ⊗OX OX over the sheaf of rings f −1 OY , we get an exact

sequence
F ⊗OX (OX ⊗f −1 OY f −1 J ) → F ⊗OX (OX ⊗f −1 OY f −1 OY )

→ F ⊗OX (OX ⊗f −1 OY f −1 (OY /J )) → 0,
that is,
F ⊗OX f ∗ J → F → F ⊗OX f ∗ (OY /J ) → 0.
This implies that
F/J F ∼
= F ⊗OX f ∗ (OY /J ).
Let X be a ringed space and F an OX -module. We define the support of F

to be the subset
suppF = {P ∈ X|FP 6= 0}.
Proposition 4.14. Let X be a scheme and F a quasi-coherent sheaf. Sup-

pose F locally is of finite type. Then suppF is a closed subset X. Suppose
furthermore that X is noetherian. Then for any coherent sheaf of ideals I of
OX such that suppF ⊂ supp(OX /I), there exists a natural number n such that
I n F = 0.
Proof. Cover X by affine open subschemes Ui = SpecAi (i ∈ I) such that

F|Ui ∼= Mi∼ for some finitely generated Ai -modules Mi . By Lemma 3.28, we
have suppF ∩ Ui = V (AnnAi (Mi )). So suppF is closed. For each i, let Ii be the
ideal of Ai such that I|Ui ∼= Ii∼ . Then supp(OX /I)p∩ Ui = V (Ii ). If suppF ⊂
supp(OX /I), then V (AnnAi (Mi )) ⊂ V (Ii ). So Ii ⊂ AnnAi (Mi ). If we assume
X is noetherian, then Ii is finitely generated and hence Iini ⊂ AnnAi (Mi ) for
some natural number ni , that is, Iini Mi = 0. The open covering may be taken
to be finite. So there exists a natural number n such that I n F = 0.
Let (f, f ] ) : (Z, OZ ) → (X, OX ) be a closed immersion of schemes. The

kernel IZ of f ] : OX → f∗ OZ is called the ideal sheaf of the closed immersion.
Proposition 4.15. Let X be a scheme. The ideal sheaf of any closed immersion
into X is quasi-coherent. Conversely for any quasi-coherent sheaf of ideals I of
OX , the ringed space (supp(OX /I), (OX /I)|supp(OX /I) ) is a scheme, and the
pair (i, i] ) consisting of the inclusion i : supp(OX /I) → X and the canonical
morphism i] : OX → i∗ i−1 (OX /I) = i∗ ((OX /I)|supp(OX /I) ) is a closed immer-
sion with ideal sheaf I. Moreover, any two closed immersions into X with ideal
sheaf I are isomorphic.
Proof. Any closed immersion f : Z → X is quasi-compact and separated. By

Proposition 4.9 (iii), f∗ OZ is quasi-coherent. So the kernel of f ] : OX → f∗ OZ
is quasi-coherent, that is, the ideal sheaf of f is quasi-coherent.
Given a quasi-coherent sheaf of ideals I, denote by (Z, OZ ) the ringed space
(supp(OX /I), (OX /I)|supp(OX /I) ). To prove (Z, OZ ) is a scheme and the pair
(i, i] ) : (Z, OZ ) → (X, OX ) is a closed immersion with ideal sheaf I, we may
assume X = SpecA is affine. Then I = I ∼ for some ideal I of A. We have
OX /I = (A/I)∼ and supp(OX /I) = V (I). Consider the closed immersion
ϕ : SpecA/I → SpecA induces by the canonical homomorphism A → A/I. It
induces a homeomorphism from SpecA/I to V (I) and ϕ∗ OSpecA/I = (A/I)∼
by Proposition 4.3 (i). So (Z, OZ ) is isomorphic to SpecA/I as ringed spaces
and the morphism (i, i] ) : (Z, OZ ) → (X, OX ) can be identified with the closed
immersion ϕ. One can verify directly that the ideal sheaf of the closed immersion
i is I.
Let’s prove the uniqueness of the closed subscheme with ideal sheaf I. Let
f : Z → X be a closed immersion with ideal sheaf I. f ] induces an isomorphism
OX /I ∼ = f∗ OZ . Since f induces a homeomorphism of Z with a closed subset
of X, we have supp(f∗ OZ ) = f (Z). So supp(OX /I) = f (Z) and f ] induces an
isomorphism f −1 (OX /I) ∼ = OZ . Hence Z is isomorphic to the closed subscheme
(supp(OX /I), (OX /I)|supp(OX /I) ).
Note that Proposition 4.15 gives another proof of Proposition 3.16 (ii).
Proposition 4.16. Let f : X → Y be a quasi-compact and quasi-separated

morphism of schemes. Then we have a commutative diagram
f
X → Y
g↓ %i
Z
such that i : Z → Y is a closed immersion and it has the following universal
property: For any commutative diagram
f
X → Y
g0 ↓ % i0
Z0
such that i0 : Z 0 → Y is a closed immersion, there exists a unique morphism
j : Z → Z 0 such that i0 j = i and jg = g 0 . We have i(Z) = f (X). We call Z the
scheme theoretic image of f .
Proof. Let I be the kernel of f ] : OY → f∗ OX . Since f is quasi-compact and

quasi-separated, f∗ OX is quasi-coherent by Proposition 4.9 (iii). So I is a quasi-
coherent sheaf of ideals of OY . Let (Z, OZ ) = (supp(OX /I), (OX /I)|supp(OX /I) )
be the closed subscheme of Y with ideal sheaf I and let i : Z → Y be the closed
immersion. Note that f ] induces a monomorphism OY /I ,→ f∗ OX . So Z =
supp(OZ /I) = supp(f∗ OX ). Obviously we have f (X) ⊂ supp(f∗ OX ) ⊂ f (X).
Since Z is closed, we must have Z = f (X).
Define a morphism (g, g ] ) : (X, OX ) → (Z, OZ ) as follows: The map g :
X → Z on the underlying topological spaces is induced by f . The morphism
f ] : OY → f∗ OX induces a morphism OY /I → f∗ OX = i∗ g∗ OX and hence a
morphism OZ = i−1 (OY /I) → g∗ OX . We define g ] : OZ → g∗ OX to be this
last morphism. It is not hard to verify f = ig as morphisms of schemes.
Suppose i0 : Z 0 → Y is a closed immersion such that there exists a morphism
g 0 : X → Z 0 satisfying f = i0 g 0 . Then we have f (X) ⊂ i0 (Z 0 ) and hence
f (X) ⊂ i0 (Z 0 ). So there exists a continuous map j : Z → Z 0 such that i = i0 j
as maps between topological spaces. Let I 0 be the ideal sheaf of Z 0 . It is the
]
kernel of i0 : OY → i0∗ OZ 0 . Since we have f = i0 g 0 , I 0 is contained in the kernel
of f : OY → f∗ OX , that is, I 0 is a subsheaf of I. So we have a canonical
]
morphism OY /I 0 → OY /I. We have OY /I 0 ∼ = i0∗ OZ 0 and OY /I ∼ = i∗ OZ .

0 0
So we have a morphism i∗ OZ 0 → i∗ OZ = i∗ j∗ OZ , which induces a morphism
OZ 0 → j∗ OZ . We define j ] to be this last morphism. One then verifies that
i0 j = i and jg = g 0 as morphisms of schemes.
Let j 0 : Z → Z 0 be another morphism such that i0 j 0 = i and j 0 g = g 0 as
morphisms of schemes. Then we have i0 j = i0 j 0 . Since i0 is a closed immersion,
it is a monomorphism in the category of schemes. So we have j = j 0 . This
proves the uniqueness of j.
Corollary 4.17. Let f : Z → X be an immersion. If f is quasi-compact,

then f can be factorized as Z → F → X such that the first arrow is an open
immersion and the second arrow is a closed immersion.
Proof. Since f is an immersion, it is separated. If f is quasi-compact, then

g i
by Proposition 4.16, we may factorize f as Z → F → X such that F is the
scheme theoretic image of f . Then i : F → X is a closed immersion. Since
i(F ) = f (X) and f is an immersion, g : Z → F induces a homeomorphism of Z
with an open subset of F . By the definition of scheme theoretic image, for any
P ∈ Z, the homomorphism gP] : OF,g(P ) → OZ,P is a monomorphism. Since f
is an immersion, the homomorphism fP] : OX,f (P ) → OZ,P is surjective. This
implies gP] is surjective. So gP] is an isomorphism. By Proposition 3.15 (i), g
must be an open immersion.
Proposition 4.18. (Chow’s Lemma) Let S be a noetherian scheme and let

f : X → S be a proper morphism. Then there exists a projective morphism
g : X 0 → X such that f g is projective, and there exists an open dense subset U
of X such that g induces an isomorphism from g −1 (U ) to U .
Proof. Note that X is noetherian. Let Xi (i = 1, . . . , n) be the irreducible

components of X. For each i, let Vi = Xi − ∪j6=i Xj . Then Vi is open in
X and dense in Xi and the open immersion Vi → X is quasi-compact since
Vi is necessarily noetherian. Put a scheme structure on Xi so that it is the
scheme theoretic image of Vi in X. Then Vi , with its open subscheme structure
induced from X, is also an open subscheme of Xi . Note that the morphism
fi : Xi → S induced by f is proper. Suppose we can prove the proposition for
each fi : Xi → S and let gi : Xi0 → Xi be a projective morphism such that
fi gi is projective and let Ui be an open dense subset
`n of Xi such that gi induces
an isomorphism from gi−1 (Ui ) to Ui . Let X 0 = i=1 Xi0 and g : X 0 → X the
morphism induced by gi . Then g and f g are projective. Take U = ∪ni=1 (Ui ∩Vi ).
Then U is dense in X and g induces an isomorphism from g −1 (U ) to U . So to
prove our proposition, we may assume X is irreducible.
Cover S by finitely many affine open subschemes Vi = SpecAi (i ∈ I) and
cover each f −1 (Vi ) by finitely many affine open subschemes Uij = SpecBij
(j ∈ Ji ) such that each Bij is a finitely generated Ai -algebra. Let b1 , · · · , bm be
a finite family of generators of Bij . Then the Ai -algebra homomorphism
Ai [x1 , . . . , xm ] → Bij , xk 7→ bk
is surjective. It induces a closed immersion Uij → SpecAi [x1 , . . . , xm ]. On the

other hand, we have an isomorphism
xk
Ai [x1 , . . . , xm ] → (Ai [x0 , . . . , xm ])(x0 ) , xk 7→
x0
and an open immersion
Spec(Ai [x0 , . . . , xm ])(x0 ) → ProjAi [x0 , . . . , xm ].
So we have an immersion Uij → ProjAi [x0 , . . . , xm ] = Pm

Ai . The open immersion
SpecAi = Vi → S induces an open immersion Pm Ai → P m
S . So finally we get an
immersion Uij → Pm S . Since Uij is noetherian, this immersion is quasi-compact.

By Corollary 4.17, we may factorize this immersion into Uij → Pij → PnS such
that the first arrow is an open immersion and the second arrow is a closed
immersion. Changing notations, we may assume that X can be covered by
open subschemes Ui (i = 1, . . . , n) such that each morphism Ui → S can be
factorized as Ui → Pi → S with the first arrow being an open immersion and
the second arrow being projective.
Let U = ∩ni=1 Ui . Since X is irreducible, U is dense in X. Consider the
morphism h : U → X ×S P1 ×S · · · ×S Pn defined by the open immersions
U → X and U → Pi . Denote P = P1 ×S · · · ×S Pn for convenience. Let X 0 be
the scheme theoretic image of h : U → X ×S P , let g : X 0 → X and g 0 : X 0 → P
be the compositions of the closed immersion X 0 → X ×S P with the projections
of X ×S P to its factors.
Note that g : X 0 → X is projective since it is the composition of the closed
immersion X 0 → X ×S P with the projective morphism X ×S P → X. We have
g −1 (U ) = X 0 ∩ (U ×S P ). Note that X 0 ∩ (U ×S P ) → U ×S P is the scheme
theoretic image of U → U ×S P . But U → U ×S P is a closed immersion since it
is the graph of U → P . So U → U ×S P is the scheme theoretic image of itself.
Hence the morphism U → X 0 induces an isomorphism U ∼ = X 0 ∩ (U ×S P ). The
g
composition of this isomorphism with X ∩ (U ×S P ) = g −1 (U ) → U is idU . So
0
−1
g induces an isomorphism from g (U ) to U .
Obviously X 0 ∩ (Ui ×S P ) (i = 1, . . . , n) form an open covering of X 0 . Let
Wi be the inverse images of Ui under the projections P → Pi . We claim that
X 0 ∩(Ui ×S P ) ⊂ X 0 ∩(X ×S Wi ). This would imply X 0 ∩(X ×S Wi ) (i = 1, . . . , n)
form an open covering of X 0 . For each i, we have a commutative diagram
U → P
↓ ↓
Ui → Pi ,
where the top horizontal arrow is the morphism U → P whose composition with
each projection P → Pk is equal to the composition U → Uk → Pk . Since the
image of U in X 0 is dense, the diagram
X 0 ∩ (Ui ×S P ) → P
↓ ↓
Ui → Pi ,
commutes, at least as a diagram of maps on the underlying topological spaces.

This implies that the image of X 0 ∩ (Ui ×S P ) → P is contained in Wi . So
X 0 ∩ (Ui ×S P ) ⊂ X 0 ∩ (X ×S Wi ).
Finally let’s prove g 0 : X 0 → P is a closed immersion. This would imply
X 0 is projective over S and hence f g : X 0 → S is projective. Since X and P
are proper over S, X ×S P is proper over S. Since X 0 is a closed subscheme
of X ×S P , X 0 is also proper over S. By Proposition 3.26 (vi), g 0 : X 0 → P is
proper. To prove it is a closed immersion, it suffices to show it is an immersion.
−1
We have g 0 (Wi ) = X 0 ∩ (X ×S Wi ). Since X 0 ∩ (X ×S Wi ) (i = 1, . . . , n)
g0
form an open covering of X 0 , it suffices to show each X 0 ∩ (X ×S Wi ) → Wi is
a closed immersion.
Consider the morphism defined by the composition Wi → Ui → X, where
the first arrow is induced by the projection P → Pi . Let Wi → X ×S Wi be the
graph of this morphism. It is a closed immersion. Let U → P be the morphism
whose composition with each projection P → Pi is equal to U → Ui → Pi . The
image of U → P is contained in Wi and hence it induces a morphism U → Wi .
The image of U → X ×S P is contained in X ×S Wi and hence it induces a
morphism U → X ×S Wi . Moreover, we have a commutative diagram
Wi
% ↓
U → X ×S Wi .
But X 0 ∩ (X ×S Wi ) → X ×S Wi is the scheme theoretic image of U → X ×S Wi .
Therefore X 0 ∩ (X ×S Wi ) → X ×S Wi factorizes through the closed immersion
Wi → X ×S Wi :
Wi
θ% ↓
X 0 ∩ (X ×S Wi ) → X ×S Wi .
θ
Note that X 0 ∩ (X ×S Wi ) → Wi is necessarily a closed immersion. Consider
the commutative diagram
Wi
θ% ↓
X 0 ∩ (X ×S Wi ) → X ×S Wi
g0 & ↓
Wi .
Since the composition of the vertical arrows in this diagram is identity, the
g0
morphism X 0 ∩ (X ×S Wi ) → Wi coincides with θ and hence g 0 is a closed
immersion. This proves our assertion.
∞
L
Let S = Sd be a graded ring. A graded S-module is an S-module M
d=0
∞
L
together with a decomposition M = Mi of abelian groups such that
i=−∞
Sd Mi ⊂ Md+i . Elements in Mi are called homogeneous of degree i. For any
integer n, define M (n) to be the graded S-module whose underlying S-module
is M and whose grading is defined by (M (n))d = Md+n . For any homogeneous
prime ideal p of S, define M(p) to be the S(p) -module
m
M( p ) = { ∈ Mp |m ∈ M and t ∈ S−p are homogeneous of the same degree }.
t
For any homogeneous element f in S, define M(f ) to be the S(f ) -module
m
M(f ) = { ∈ Mf |k ∈ N and m ∈ M is homogeneous of degree kdegf }.
fk
Define a sheaf M ∼ of OProjS -modules as follows:

of ProjS, M ∼ (U ) consists of functions s : U → p∈ProjS M(p) satisfying the
(a) For any p ∈ U , we have s(p) ∈ M(p) .
homogeneous elements m ∈ M and f ∈ S of the same degree such that for any
q ∈ Up , we have f 6∈ q and s(q) = mf in M(q) .
For any inclusion of open subsets V ⊂ U , we define M ∼ (U ) → M ∼ (V ) to be
the restriction of functions.
Proposition 4.19. Let S be a graded ring and M a graded S-module.

(i) For any p ∈ ProjS, we have (M ∼ )p ∼ = M( p ) .
(ii) For any homogeneous element f in S with positive degree, M ∼ |D+ (f )
can be identified with (M(f ) )∼ through the isomorphism D+ (f ) ∼= SpecS(f ) .
(iii) M ∼ is quasi-coherent. If S is noetherian and M is a finitely generated
S-module, then ProjS is noetherian and M ∼ is coherent.
We leave to the reader to prove this proposition. We only explain the fact
that if S is noetherian and M is finitely generated, then M(f ) is noetherian
for any homogeneous element f in S. Let N1 ⊂ N2 ⊂ · · · be an ascending
chain of submodules of M(f ) . For each i, let Sf Ni be the submodule of Mf
generated by Ni . Then we have Sf N1 ⊂ Sf N2 ⊂ · · ·. Since Mf is noetherian,
we have Sf Nn = Sf Nn+1 = · · · for large n. We have Sf Ni ∩ M(f ) = Ni . So
Nn = Nn+1 = · · ·. Hence M(f ) is noetherian.
For any integer n, define OProjS (n) = S(n)∼ . We call OProjS (1) the twisting
sheaf. For any OProjS -module F, define F(n) = F ⊗OProjS OProjS (n).
Let M and N be graded S-modules. We define a graded S-module M ⊗S N
as follows: We have
M M M M
M ⊗Z N = Md ⊗Z Ne = Md ⊗Z Ne .
d e n d+e=n
It gives a decomposition of M ⊗Z N . M ⊗S N is the quotient of M ⊗Z N by the

subgroup generated by elements of the form sm ⊗ n − m ⊗ sn. Note that this
subgroup is a homogeneous subgroup of M ⊗Z N . Hence the decomposition of
M ⊗Z N induces a grading on M ⊗S N .
Proposition 4.20. Let S be a graded ring which is generated by S1 as an

S0 -algebra.
(i) OProjS (n) is an invertible OProjS -module for every n.
(ii) Let M and N be graded S-modules. Then
(M ⊗S N )∼ ∼
= M ∼ ⊗OProjS N ∼ .
In particular, we have
(M (n))∼ = (M ⊗S S(n))∼ = M ∼ ⊗OProjS S(n)∼ = M ∼ (n),
(S(m + n))∼ = (S(m) ⊗S S(n))∼ = S(m)∼ ⊗OProjS S(n)∼ .
Hence
OProjS (m + n) = OProjS (m) ⊗OProjS OProjS (n),

OProjS (n) = OProjS (1)⊗n ,
OProjS (−1) = HomOProj S (OProjS (1), OProjS ).
(iii) Let T be another graded ring which is generated by T1 as a T0 -algebra,

let φ : S → T be a homomorphism preserving the gradings, and let
U = {q ∈ ProjT |S+ 6⊂ φ−1 (q )}.
Then U is an open subset of ProjT and φ induces canonically a morphism of

schemes f : U → ProjS. For any graded S-module M , we have
f ∗ M ∼ = (T ⊗S M )∼ |U ,
and for any graded T -module N , we have
f∗ (N ∼ |U ) = (S N )∼ ,
where on the right-hand side, N is regarded as an S-module. In particular, we

have
f ∗ OProjS (n) ∼
= OProjT (n)|U .
Proof. (i) For any homogeneous element f ∈ S+ , we have OProjS (n)|D+ (f ) ∼

=
(S(n)(f ) )∼ on D+ (f ) ∼
= SpecS(f ) . When f ∈ S1 , we have an isomorphism of
S(f ) -modules
S(f ) → S(n)(f ) , x → f n x.
Hence OProjS |D+ (f ) ∼
= OProjS (n)|D+ (f ) . Since S is generated by S1 as an S0 -
algebra, D+ (f ) (f ∈ S1 ) form an open covering of ProjS. So OProjS (n) is
invertible.
(ii) Define a morphism
φ : M ∼ ⊗OProjS N ∼ → (M ⊗S N )∼
as follows: Since M ∼ ⊗OProjS N ∼ is the sheaf associated to the presheaf
U 7→ M ∼ (U ) ⊗OProjS (U ) N ∼ (U ),
∼
it suffices to defineà morphism from this presheaf to (M ⊗` S N ) . For any
∼
sections s : U → p∈ProjS M(p) of M (U ) and t : U → p∈ProjS N(p) of
N ∼ (U ), define a section φ(s ⊗ t) : U → p∈ProjS (M ⊗S N )(p) of (M ⊗S N )∼ (U )
`
so that for any p ∈ U , φ(s ⊗ t)(p) is the image of s(p) ⊗ t(p) under the canonical
homomorphism
M(p) ⊗S(p) N(p) → (M ⊗S N )(p) .
For any f ∈ S1 , through the isomorphism D+ (f ) ∼

= SpecS(f ) , we have
(M ∼ ⊗OProjS N ∼ )|D+ (f ) ∼
= M(f ) ∼ ⊗OSpecS(f ) N(f ) ∼ = (M(f ) ⊗S(f ) N(f ) )∼ ,
∼ ∼
(M ⊗S N )∼ |D+ (f ) = (M ⊗S N )(f ) .
Since D+ (f ) (f ∈ S1 ) form an open covering of ProjS, to prove φ is an isomor-
phism, it suffices to show the homomorphism
φ(f ) : M(f ) ⊗S(f ) N(f ) → (M ⊗S N )(f )
induced by φ is an isomorphism for any f ∈ S1 . Obviously it is onto. We have

a commutative diagram
φ(f )
M(f ) ⊗S(f ) N(f ) → (M ⊗S N )(f )
↓ ↓
Mf ⊗Sf Nf ∼
= (M ⊗S N )f .
To prove φ(f ) is injective, it suffices to show the first vertical arrow is injective.
Indeed, it has a left inverse
P P
i mi i ni
X mi X ni
Mf ⊗Sf Nf → M(f ) ⊗S(f ) N(f ) , ⊗ →
7 ( ) ⊗ ( ),
fk fl i
fi i
fi
where for each i, mi and ni are homogeneous elements of degree i in M and N ,

respectively. We leave to the reader to verify that this map is well-defined.
(iii) We have U = ProjT − V (φ(S+ )T ). So U is open. Define f : U → ProjS
by
f (q) = φ−1 (q)
for any q ∈ U . We have f −1 (V (a)) = V (φ(a)T )∩U for any homogeneous ideal a
of S. So f is continuous. Define f ] : OProjS → f∗ (O`ProjT |U ) as follows: For any
open subset V of ProjS and any section s : V → p∈ProjS S(p) in OProjS (V ),
define a section f ] (s) : f −1 (V ) → q∈ProjT T(q) in OProjT (f −1 (V )) by
`
(f ] (s))(q) = φq (s(f (q)))
for any q ∈ f −1 (V ), where φq : S(φ−1 (q)) → T(q) is the homomorphism induced

by φ. One can verify that (f, f ] ) is a morphism of schemes. Note that for any
homogeneous element a ∈ S+ , we have f −1 (D+ (a)) = D+ (φ(a)).
For any T -module N , define
α : (S N )∼ → f∗ (N ∼ |U )
as follows: For any section s : V → p∈ProjS (S N )(p) of (S N )∼ (V ), we define

`
a section α(s) : f −1 (V ) → q∈ProjT N(q) of f∗ (N ∼ |U )(V ) = N ∼ (f −1 (V )) so

`
that for any q ∈ f −1 (V ), (α(s))(q) is the image of s(φ−1 (q)) under the canonical
homomorphism
n n
(A N )(φ−1 (q)) → N(q) , → .
s φ(s)
To prove α is an isomorphism, it suffices to show that for any homogeneous

α
element a ∈ S+ the homomorphism (S N )∼ (D+ (a)) → f∗ (N ∼ |U )(D+ (a)) is an
isomorphism. We have
(S N )∼ (D+ (a)) = (S N )(a) ,

f∗ (N ∼ |U )(D+ (a)) = N ∼ (D+ (φ(a))) = N(φ(a)) .
Our assertion follows from the fact that the canonical homomorphism
n n
(S N )(a) → N(φ(a)) , k
7→
a φ(a)k
is an isomorphism.
For any S-module M , the S-module homomorphism
M →S (T ⊗S M ), m 7→ 1 ⊗ m
induces a morphism
M ∼ → (S (T ⊗S M ))∼ = f∗ ((T ⊗S M )∼ |U ).
Since f ∗ is left adjoint to f∗ , it induces a morphism
β : f ∗ M ∼ → (T ⊗S M )∼ |U .
For any homogeneous element a ∈ S+ , we have
(f ∗ M ∼ )|D+ (φ(a)) ∼
= (f |D+ (φ(a)) )∗ (M ∼ |D+ (a) ) ∼
= (f |SpecT(φ(a)) )∗ (M(a)
∼
)
∼
= (T(φ(a)) ⊗S(a) M(a) ) ,
(T ⊗S M )∼ |D+ (φ(a)) = (T ⊗S M )(φ(a)) .
One can verify that for any a ∈ S1 , β induces an isomorphism
T(φ(a)) ⊗S(a) M(a) ∼

= (T ⊗S M )(φ(a)) .
Since D+ (φ(a)) (a ∈ S1 ) form an open covering of U , β is an isomorphism.
We leave to the reader to prove the following lemma: (Confer the proof of
Proposition 3.9 (iii) and Lemma 4.4.)
Lemma 4.21. Let X be a scheme, L an invertible OX -module and f ∈ L(X).

Define Xf to be the subset of X consisting of those points x ∈ X such that the
germ of f at x does not lie in mx Lx , where mx is the maximal ideal of the local
ring OX,x . Then Xf is open. Let F be a quasi-coherent OX -module.
(i) Suppose X is quasi-compact. If s ∈ F(X) is a section whose restriction
to Xf vanishes, then there exists a natural number n such that the section
s ⊗ f ⊗n ∈ (F ⊗OX L⊗n )(X) vanishes.
(ii) Suppose X is quasi-compact and quasi-separated. Given a section t in

F(Xf , F), there exists a natural number n such that the section t ⊗ f ⊗n in
(F ⊗OX L⊗n )(Xf ) can be extended to a section in (F ⊗OX L⊗n )(X).
Let S be a graded ring which is generated by S1 as an S0 -algebra and let

X = ProjS. Denote OProjS (n) by O(n). For any OX -module F, define the
graded S-module associated to F to be
∞
M
Γ∗ (F) = F(n)(X),
n=−∞
where the scalar product of S on Γ∗ (F) is defined by the following composition

of the canonical homomorphisms:
Sd ⊗Z F(n)(X) → O(d)(X) ⊗Z F(n)(X) → F(d + n)(X).
Proposition 4.22. Let S be a graded ring which is finitely generated by S1 as

an S0 -algebra and let F be a quasi-coherent OProjS -module. Then we have a
canonical isomorphism Γ∗ (F)∼ ∼= F.
Proof. Denote ProjS by X and OProjS (n) by O(n). For any homogeneous
element f in S of degree d, define
φ(D+ (f )) : Γ∗ (F)∼ (D+ (f )) = Γ∗ (F)(f ) → F(D+ (f ))

s
as follows: For any fk
∈ Γ∗ (F)(f ) , where s ∈ F(dk)(X), we define φ(D+ (f ))( fsk )
1
to be the image of fk
⊗ s|D+ (f ) under the canonical homomorphism
O(−dk)(D+ (f )) ⊗OX (D+ (f )) F(dk)(D+ (f )) → F(D+ (f )).
Applying Lemma 4.21 to L = O(d), we see that φ(D+ (f )) is an isomorphism.

For any open subset U of X, we choose an open covering {D+ (fi )}i∈I of U and
define
φ(U ) : Γ∗ (F)∼ (U ) → F(U )
to be the isomorphism making the following diagram commute:
0 → Γ∗ (F)∼ (U ) → Γ∗ (F)∼ (D(fi )) → Γ∗ (F)∼ (D(fi ) ∩ D(fj ))
Q Q
i∈I i,j∈I
φ(U ) ↓ Q ↓ Q ↓
0 → F(U ) → F(D+ (fi )) → F(D+ (fi ) ∩ D+ (fj )),
i∈I i,j∈I
where in each row, the second horizontal arrow is defined by s 7→ (s|D(fi ) ) and
the third horizontal arrow is defined by (si ) 7→ (sj |D(fi )∩D(fj ) − si |D(fi )∩D(fj ) ).
One can verify φ(U ) is independent of the choice of the open covering {D(fi )}i∈I .
Lemma 4.23. Let S be the ring A[x0 , . . . , xm ] with the grading defined by the
degrees of polynomials. We have Γ∗ (OProjS ) ∼= S.
Proof. Denote OProjS (n) by O(n). Since D+ (xi ) (i = 0, . . . , m) form an open

covering of ProjS, we have an exact sequence
m
Y m
Y
0 → O(n)(ProjS) → O(n)(D+ (xi )) → O(n)(D+ (xi ) ∩ D+ (xj )),
i=0 i,j=0
where the second arrow is defined by s 7→ (s|D+ (xi ) ) and the third arrow is
defined by (si ) 7→ (sj |D+ (xi )∩D+ (xj ) − si |D+ (xi )∩D+ (xj ) ). Taking the direct sum
over all integers n, we get an exact sequence
m
Y m
Y
0 → Γ∗ (OProjS ) → A[x0 , . . . , xm ]xi → A[x0 , . . . , xm ]xi xj .
i=0 i,j=0
We have canonical monomorphisms
A[x0 , . . . , xm ]xi ,→ A[x0 , . . . , xm ]xi xj ,→ A[x0 , . . . , xm ]x0 ···xm .
Through this monomorphisms, we identify A[x0 , . . . , xm ]xi (i = 0, . . . , m) with

subrings of A[x0 , . . . , xm ]x0 ···xm . The above exact sequence then shows that
m
\
Γ∗ (OProjS ) = A[x0 , . . . , xm ]xi .
i=0
Any element in A[x0 , . . . , xm ]x0 ···xm can be written uniquely as
xk00 · · · xkmm f,
where k0 , . . . , km are integers and f ∈ A[x0 , . . . , xm ] is a polynomial with no

factors xi (i = 0, . . . , m). For each i, we have xk00 · · · xkmm f ∈ A[x0 , . . . , xm ]xi if
m
and only if kj ≥ 0 for any j 6= i. So xk00 · · · xkmm f ∈
T
A[x0 , . . . , xm ]xi if and
i=0
only if kj ≥ 0 for all j. Hence Γ∗ (OProjS ) = A[x0 , . . . , xm ].
Corollary 4.24.
(i) Let X be a closed subscheme of Pm A = ProjA[x0 , . . . , xm ]. Then X is iso-
morphic to ProjA[x0 , . . . , xm ]/a for some homogeneous ideal a of A[x0 , . . . , xm ].
(ii) A scheme X over SpecA is projective if and only it is isomorphic to
ProjS for some graded ring S such that S0 = A and S is finitely generated by
S1 as an S0 -algebra.
Proof. (i) Let I be the ideal sheaf of the closed immersion X → Pm

A . By
Proposition 4.22, we have I ∼
= Γ∗ (I)∼ . Since
Γ∗ (I) ⊂ Γ∗ (OPm
A
)∼
= A[x0 , . . . , xm ],
Γ∗ (I) corresponds to a homogeneous ideal a of A[x0 , . . . , xm ]. Then X is iso-

morphic to ProjA[x0 , . . . , xm ]/a.
(ii) We leave to the reader to prove the “if” part. Let’s prove the “only if”
part. Suppose X is projective over SpecA. The morphism X → SpecA can be
factorized as X → Pm A → SpecA such that the first arrow is a closed immersion.
By (i), X is isomorphic to ProjA[x0 , . . . , xm ]/a for some homogeneous ideal a.
Let a0 = a ∩ (x0 , . . . , xm ). One can show
ProjA[x0 , . . . , xm ]/a ∼
= ProjA[x0 , . . . , xm ]/a0 .
It is easy to see that S = A[x0 , . . . , xm ]/a0 has the required property.
For any scheme S, define the twisting sheaf OPm S

(1) on Pm m
S = PZ × S to
∗ m m
be p (OPm Z
(1)), where p : PZ × S → PZ is the projection. When S = SpecA,
this coincides with the twisting sheaf of Pm
A . In general, if we cover S by affine
open subschemes Ui = SpecAi (i ∈ I), then OPm S
(1) is obtained by glueing the
twisting sheaves OPm Ai
(1) (i ∈ I).
Let X be an S-scheme. An invertible OX -module L is called very ample
over S if there exists an immersion i : X → Pm S which is an S-morphism such
that L ∼= i∗ (OPmS
(1)).
We say an OX -module F is generated by global sections if there exists a
family of sections si ∈ F(X) (i ∈ I) such that for any P ∈ X, FP is generated
by the germs siP (i ∈ I) as an OX,P -module. This is equivalent to saying that
the morphism
⊕i∈I OX → F, ei 7→ si
is surjective, where for each i, ei is the section of ⊕i∈I OX whose i-th component
is 1 and whose other components are 0. For example, if X = SpecA, then for any
A-modules M , elements in M define global sections of M ∼ which generate M ∼ .
If S is a graded ring which is generated by S1 as an S0 -algebra and X = ProjS,
then elements in S1 define global sections of OX (1) which generate OX (1).
Proposition 4.25. (Serre) Let X be a scheme proper over SpecA for some
noetherian ring A, let OX (1) be an invertible OX -module very ample over
SpecA, and let F be a coherent OX -module. Then there exists a natural number
N such that for any n ≥ N , the OX -module F(n) = F ⊗OX OX (1)⊗n can be
generated by finitely many global sections, that is, there exists an epimorphism
m
OX → F(n)
for some natural number m.
∗
Proof. Let i : X → Pm A be an immersion such that i (OPA (1)) = OX (1).
m
Since X is proper over SpecA, i is proper by Proposition 3.26 (vi) and hence
is a closed immersion. We have i∗ (F(n)) ∼ = (i∗ F)(n) by Lemma 4.26 below.
Moreover, for any P ∈ X, we have F(n)P ∼ = (i∗ (F(n)))i(P ) . To prove our
assertion, it suffices to show (i∗ F)(n) can generated by finitely many global
sections for large n. So we may assume X = Pm A . For each i, let Mi be a finitely
generated A[x0 , . . . , xm ](xi ) -module such that F|D+ (xi ) ∼
= Mi∼ . Let {sij } be a
finite family of generators for Mi . Then by Lemma 4.21, there exists a natural
number N such that for any n ≥ N , each section sij ⊗ xni ∈ F(n)(D+ (xi )) can
be extended to a global section tij ∈ F(n)(PmA ). Note that {tij } generate F(n).
Lemma 4.26. Let f : X → Y be a morphism of ringed spaces, F an OX -module

and G a locally free OY -module. Then we have a canonical isomorphism
f∗ F ⊗OY G ∼
= f∗ (F ⊗OX f ∗ G).
Proof. Let F 0 be an OX -module. We have a canonical morphism
f∗ F ⊗OY f∗ F 0 → f∗ (F ⊗OX F 0 )
defined in the obvious way. We define
f∗ F ⊗OY G → f∗ (F ⊗OX f ∗ G)
to be the composition
f∗ F ⊗OY G → f∗ F ⊗OY f∗ f ∗ G → f∗ (F ⊗OX f ∗ G).
To prove it is an isomorphism, we may work locally. Hence we may assume G

is free. Our assertion is then trivial.
1.5 Formal Completions of Schemes and Sheaves
The main part of this section is used only in the last section of this book.
Let (Mn , φmn )n∈N be an inverse system of sets. We say it satisfies the
Mittag-Leffler condition if for any n, there exists a natural number n0 such that
for any m ≥ n + n0 , we have
φn+n0 ,n φmn
im(Mn+n0 → Mn ) = im(Mm → Mn ).
Suppose (Mn , φmn )n∈N satisfies the Mittag-Leffler condition and each Mn is
nonempty. Let Mn0 = ∩m≥n imφmn . Then (Mn0 )n∈N is an inverse system of
0 φmn
nonempty sets and inv. limn Mn0 ∼ = inv. limn Mn . Moreover each Mm → Mn0 is
0
surjective. This implies inv. limn Mn 6= ∅.
Note that if each φmn is surjective, then (Mn , φmn )n∈N satisfies the Mittag-
Leffler condition. If (Mn , φmn )n∈N is an inverse system of A-modules and each
Mn has finite length, then (Mn )n∈N satisfies the Mittag-Leffler condition. In-
deed, for each n, the descending chain
imφnn ⊃ imφn+1,n ⊃ imφn+2,n ⊃ · · ·

1.5. FORMAL COMPLETIONS OF SCHEMES AND SHEAVES 77
is stationary.
Proposition 5.1. Let (Mn0 , φ0mn )n∈N , (Mn0 , φmn )n∈N and (Mn00 , φ00mn )n∈N be
un vn
inverse systems of abelian groups and let (Mn0 → Mn )n∈N and (Mn → Mn00 )n∈N
be morphisms of inverse systems such that for each n, the sequence
u v
0 → Mn0 →
n n
Mn → Mn00 → 0
is exact. If (Mn0 )n∈N satisfies the Mittag-Leffler condition, then the sequence
0 → inv. lim Mn0 → inv. lim Mn → inv. lim Mn00 → 0

n n n
is exact.
Proof. Only the surjectivity of inv. limn Mn → inv. limn Mn00 is nontrivial.
Given (x00n ) ∈ inv. limn Mn00 , let Nn = vn−1 (x00n ). Then (Nn , φmn )n∈N is an
inverse system of sets. Since each vn is surjective, each Nn is nonempty. We
claim (Nn )n∈N satisfies the Mittag-Leffler condition. Then inv. limn Nn 6= ∅
by the discussion at the beginning of this section. Any element in inv. limn Nn
lies in inv. limn Mn and is mapped to (x00n ). This proves the surjectivity of
inv. limn Mn → inv. limn Mn00 . Let’s prove our claim. Given n, since (Mn0 )n∈N
satisfies the Mittag-Leffler condition, we can find a natural number n0 such that
for any m ≥ n + n0 , we have
φ0n+n ,n φ0
0
im(Mn+n 0
→0 Mn0 ) = im(Mm
0 mn
→ Mn0 ).
Fix an m ≥ n + n0 and an element xm ∈ Nm . For any n ≤ k ≤ m, let

xk = φmk (xm ). Then the family of maps
Mk0 → Nk , x0k 7→ uk (x0k ) + xk
form an isomorphism between the inverse systems of sets (Mk0 )n≤k≤m and
(Nk )n≤k≤m . So we have
φn+n0 ,n φmn
im(Nn+n0 → Nn ) = im(Nm → Nn ).
This is true for any m ≥ n + n0 . So (Nn )n∈N satisfies the Mittag-Leffler condi-
tion.
Let A be a ring, I an ideal of A, and M an A-module. A decreasing filtration
M = M0 ⊃ M 1 ⊃ M 2 ⊃ · · ·
of M is called an I-filtration if IMn ⊂ Mn+1 for every n. This implies that

I n M ⊂ Mn for every n. The filtration is called I-stable if it is an I-filtration
and there exists a natural number n0 such that for any n ≥ n0 , we have IMn =
Mn+1 . This implies that
I n M ⊂ Mn = I n−n0 Mn0 ⊂ I n−n0 M

for any n ≥ n0 , and hence
inv. lim M/I n M ∼

= inv. lim M/Mn .
n n
Proposition 5.2. Let A be a ring, I an ideal of A, M an A-module,
M = M0 ⊃ M1 ⊃ M2 ⊃ · · ·
∞
I n . Suppose each Mn is a finitely
L
an I-filtration of M , and S the graded ring
n=0
generated A-module. Then the filtration is I-stable if and only if the graded
∞
L
S-module Mn is finitely generated.
n=0
Proof. Since the filtration is an I-filtration, we have I n Mm ⊂ Mm+n . This

L∞
defines a graded S-module structure on Mn . Suppose the filtration is I-
n=0
stable. Let n0 be a natural number such that Mn = I n−n0 Mn0 for any n ≥ n0 .
∞
L n0
L
Then Mn can be generated by Mn as an S-module. Since each Mn is a
n=0 n=0
∞
L
finitely generated S0 -module, Mn is a finitely generated S-module.
n=0
∞
L
Conversely, suppose Mn is a finitely generated S-module. Let xi (i =
n=0
∞
L
1, . . . , k) be a finite family of homogeneous elements generating Mn , let
n=0
ni = degxi and let n0 = max(n1 , . . . , nk ). For any n ≥ n0 , any x ∈ Mn can
as r1 x1 + · · · + rk xk for some ri ∈ Sn−ni = I n−ni . We may write
be written P
each ri as j aij rij for some aij ∈ I n−n0 and rij ∈ I n0 −ni . Then we have
x = aij (rij xi ) ∈ I n−n0 Mn0 . So we have Mn = I n−n0 Mn0 for any n ≥ n0 and
P
i,j
hence the filtration is I-stable.
Corollary 5.3. (Artin-Rees) Let A be a noetherian ring, I an ideal of A, M a

finitely generated A-module, and N a submodule of M . Then the filtration
N ⊃ N ∩ IM ⊃ N ∩ I 2 M ⊃ · · ·
of N is I-stable and hence
inv. lim N/(N ∩ I n M ) ∼

= inv. lim N/I n N.
n n
∞
I n is noetherian. Indeed, if r1 , . . . , rk is a
L
Proof. The graded ring S =
n=0
family of generators for the ideal I, then the homomorphism of A-algebras
A[x1 , . . . , xk ] → S, xi 7→ ri ∈ S1
is surjective. Our assertion then follows from the fact that A[x1 , . . . , xk ] is
noetherian. Obviously
N ⊃ N ∩ IM ⊃ N ∩ I 2 M ⊃ · · ·
is an I-filtration of N . Since
M ⊃ IM ⊃ I 2 M ⊃ · · ·
∞
I n M is finitely generated.
L
is an I-stable filtration of M , the graded S-module
n=0
∞
(N ∩I n M ) is also finitely generated. Hence by Proposition
L
So its submodule
n=0
5.2, the filtration
N ⊃ N ∩ IM ⊃ N ∩ I 2 M ⊃ · · ·
of N is I-stable.
Let A be a ring and I an ideal. The ring

b = inv. lim A/I n
A
n
is called the I-adic completion of A. We have a canonical homomorphism
A → A.
b
A is called I-adically complete if this homomorphism is an isomorphism. For

any A-module M , the A-module
b
c = inv. lim M/I n M

M
n
is called the I-adic completion of M . We have a canonical homomorphism
M →M
c
which is compatible with the module structures and the ring homomorphism
A → A.
b M is called I-adically complete if M → M
c is an isomorphism.
Proposition 5.4. Let A be a noetherian ring, I an ideal of A, and
0 → M 0 → M → M 00 → 0
an exact sequence of finitely generated A-modules. Then the sequence

c0 → M
0→M c00 → 0
c→M
is exact.
Proof. For each n, we have an exact sequence
0 → M 0 /M 0 ∩ I n M → M/I n M → M 00 /I n M 00 → 0.
Passing to inverse limit and applying Proposition 5.1, we see that
0 → inv. lim M 0 /M 0 ∩ I n M → inv. lim M/I n M → inv. lim M 00 /I n M 00 → 0

n n n
is exact. By Corollary 5.3, this sequence is exactly

c0 → M
0→M c00 → 0.
c→M
Proposition 5.5. Let A be a noetherian ring, I an ideal of A, M a finitely

generated A-module, A
b and Mc the I-adic completions of A and M , respectively.
Then we have a canonical isomorphism
b ⊗A M ∼
A =Mc.
In particular, if A is I-adically complete, then any finitely generated A-module

M is also I-adically complete.
Proof. The homomorphism M → M

c induces a homomorphism
b ⊗A M → M
A c.
Since M is a finitely generated A-module and A is noetherian, we may find an

exact sequence
Am → An → M → 0.
Consider the commutative diagram
b ⊗A Am
A → b ⊗A An
A → b ⊗A M
A → 0
↓ ↓ ↓
(Am )
[ → (An )
[ → Mc → 0.
The first row is exact since the functor A b ⊗A − is right exact. The second
row is exact by Proposition 5.4. The first two vertical arrows are obviously
isomorphisms. So the last vertical arrow is also an isomorphism.
Corollary 5.6. Let A be a noetherian ring, I an ideal of A, and A b the I-adic

completion of A.
(i) For any ideal J of A, the I-adic completion Jb of J is an ideal of A.
b
n ∼ b c n n
(ii) For any n, we have A/I = A/I . Hence I is the kernel of the canonical
c
homomorphism A b → A/I n .
(iii) For any 0 ≤ m ≤ n, we have Ic m and I m /I n ∼
n ⊂ Ic = Icm /I
c n.
(iv) For any ideal J of A and any n, we
Jcn ∼
=Ab ⊗A J n ∼
= J nA bn∼
b = (J A) = Jbn .
Proof. By Proposition 5.4, the sequence
0 → Jb → A
b
b For any 0 ≤ m ≤ n, we have an exact sequence

is exact. So Jb is an ideal of A.
0 → I n → I m → I m /I n → 0.
By Proposition 5.4, the sequence
n → Ic
0 → Ic m → I\
m /I n → 0
is exact. Using the definition of the I-adic completion, one can show
m /I n ∼
I\ = I m /I n .
n ⊂ Ic
So Ic m and Ic
m /I
c n ∼
= I m /I n . Taking m = 0, we get
A/ n ∼
b Ic = A/I n .
By Proposition 5.4 and 5.5, the sequence

b ⊗A J n → A
0→A b ⊗A A/J n → 0
b ⊗A A → A
b ⊗A J n → A
is exact. In particular, the canonical homomorphism A b is injective.
n b
Its image is J A. So we have
b ⊗A J n ∼
A = J n A.
b
The other assertions in (iv) are obvious.
Proposition 5.7. Let A be a noetherian ring, I an ideal of A, and Ab the I-adic

completion of A. Then A is I-adically complete and I ⊂ rad(A). (Recall that
b b b b
the Jacobson radical rad(A) of a ring A is defined to be the intersection of all
maximal ideals of A.)
Proof. We have
b = inv. lim A/I n ∼
A b Ibn .
= inv. lim A/
n n
So Ab is I-adically
b complete. For any x ∈ Ib and any a ∈ A,
b consider the formal
series
1 + rx + (rx)2 + · · · .
b Ibn makes sense since it is the finite sum
Note that its image in each A/
1 + rx + (rx)2 + · · · + (rx)n−1 .
So 1 + rx + (rx)2 + · · · defines an element in A b Ibn . The image of

b = inv. limn A/
2 n
the product (1 − rx)(1 + rx + (rx) + · · ·) in each A/I is 1. So
b b
(1 − rx)(1 + rx + (rx)2 + · · ·) = 1
and hence 1 − rx is a unit in A.

b Let m be a maximal ideal of A. b If x 6∈ m,
then (x) + m = A and hence rx + y = 1 for some r ∈ A and y ∈ m. But then
b b
1 − rx = y can not be a unit. Contradiction. So x lies in every maximal ideal
of A.
b
Proposition 5.8. Let A be a noetherian ring, I an ideal of A, M a finitely

generated A-module, and M
c the I-adic completion of M . Then we have
∞
\
ker(M → M
c) = I n M = {x ∈ M |(1 + r)x = 0 for some r ∈ I}.
n=1
∞
I n M = 0.
T
If I ⊂ rad(A), then
n=1
Proof. If x ∈ M and (1 + r)x = 0 for some r ∈ I, then

∞
\
x = (−r)x = (−r)2 x = · · · ∈ I n M.
n=1
∞
c) = T I n M . We have N ∩ I n M = N for any n. By
Let N = ker(M → M
n=1
Corollary 5.3, we must have IN = N . Let x1 , . . . , xk be a finite family of
generators of N . Then we can write
X
xi = aij xj
j
for some aij ∈ I. Let P be the matrix (aij ) and E the identity matrix. Then
we have  
x1
(E − P )  ...  = 0.
 
xk
Let (E − P )∗ be the adjoint matrix of E − P . We have
 
x1
(E − P )∗ (E − P )  ...  = 0,
 
xk
that is,  
x1
det(I − P )  ...  = 0.
 
xk
Since the entries of P lie in I, we have det(I − P ) = 1 + r for some r ∈ I. Then
∞
I nM .
T
have (1 + r)xi = 0 for each i and hence (1 + r)x = 0 for any x ∈ N =
n=1
Suppose furthermore that I ⊂ radA. Then 1 + r is a unit in A. Hence N = 0.
Proposition 5.9. Let A be a noetherian ring, I an ideal of A, and A

b the I-adic
completion of A. Then A is noetherian.
b
Proof. Consider the graded ring

∞ ∞
Ibn /Ibn+1 ∼
M M
G(A)
b = = I n /I n+1 .
n=0 n=0
It is noetherian. Indeed, let r1 , . . . , rm be a finite family of generators of the

ideal I and let r̄1 , . . . , r̄m be their images in I/I 2 . Then the homomorphism of
A/I-algebras
A/I[x1 , . . . , xm ] → G(A),
b xi → r̄i
is surjective. Our assertion then follows from the fact that A/I[x1 , . . . , xm ] is
noetherian.
Let J be an ideal of A.
b We need to show it is finitely generated. Consider
the graded G(A)-module
b
∞
M
G(J) = J ∩ Ibn /J ∩ Ibn+1 .
n=0
It is a graded submodule of G(A). b Since G(A) b is noetherian, G(J) is finitely

ni
generated. Let ai ∈ J ∩ I (i = 1, . . . , k) such that their images in G(J) form
b
a family of generators. Let A bi be Ab itself provided with the I-stable
b filtration
(
A
b if 0 ≤ n ≤ ni ,
(Ai )n =
b
n−ni
I
b if n ≥ ni .
Consider the homomorphism
φ : ⊕ki=1 A
bi → J, ei 7→ ai ,
where for each i, ei is the element in ⊕ki=1 A

bi whose i-th component is 1 and
whose other components are 0. Then we have
φ((⊕ki=1 A
bi )n ) ⊂ J ∩ Ibn .
So φ induces a homomorphism of graded G(A)-modules

b
∞
M ∞
M
G(φ) : G(⊕ki=1 A
bi ) = (⊕ki=1 A
bi )n /(⊕ki=1 A
bi )n+1 → G(J) = J∩Ibn /J∩Ibn+1 .
n=0 n=0
By our definition of φ, G(φ) is surjective. For any y ∈ J, the image of y in G(J)

lies in the image of G(φ). So there exists an x0 ∈ (⊕ki=1 Abi )0 such that
φ(x0 ) = y mod(J ∩ I).

b
The image of y − φ(x0 ) ∈ J ∩ Ib in G(J) lies in the image of G(φ). So there

exists an x1 ∈ (⊕ki=1 A
bi )1 such that
φ(x1 ) = y − φ(x0 ) mod(J ∩ Ib2 ).
In this way, for each natural number n, we can find an xn ∈ (⊕ki=1 A

bi )n such
that
φ(x1 + · · · + xn ) = y mod(J ∩ Ibn+1 ).
The formal series x = x0 + x1 + · · · defines an element in
⊕ki=1 A
bi = inv. lim(⊕k A k
i=1 i )/(⊕i=1 Ai )n
b b
n
since its image in each (⊕ki=1 A

bi )/(⊕k A
i=1 i )n is a finite sum. Since
b
φ(x1 + · · · + xn ) − y, φ(xn+1 + xn+2 + · · ·) ∈ Ibn+1 ,

we have
φ(x) − y = (φ(x1 + · · · + xn ) − y) + φ(xn+1 + xn+2 + · · ·) ∈ ∩∞ bn+1 .
n=1 I
But ∩∞ bn+1 = 0 since A b ∼ b Ibn . So φ(x) = y and hence φ is

n=1 I = inv. limn A/
surjective. This proves J is finite generated. Hence A b is noetherian.
Proposition 5.10. Let (An , φmn )n∈N be an inverse system of rings. For every
n, let In be the kernel of the homomorphism φn1 : An → A1 . Suppose the
following conditions hold:
n
(a) For any m ≥ n, φmn : Am → An is surjective with kernel Im . (In
m
particular, the kernel of φmm = idAm is Im = 0.)
(b) I2 /I22 = I2 is finitely generated as an A2 /I2 ∼= A1 -module.
Let A = inv. limn An . For each n, let φn : A → An be the canonical
homomorphism and let I (n) be its kernel. We have
(i) For any n, I (n) is finitely generated and I (n) = (I (1) )n .
(ii) A ∼ = inv. limn A/(I (1) )n .
(iii) I (1) /(I (1) )2 ∼
= I2 /I22 .
Proof. Note that (i) implies (ii) and (iii). Indeed, since each φmn is surjective,
each φn : A → An is also surjective. It induces an isomorphism
A/I (n) ∼
= An .
We have
A = inv. lim An ∼
= inv. lim A/I (n) ∼
= inv. lim A/(I (1) )n .
n n n
This proves (ii). We have a commutative diagram
A/(I (1) )2 ∼
= A2
↓ ↓
A/I (1) ∼
= A1 .
The kernel of the first vertical arrow is I (1) /(I (1) )2 and the kernel of the second
vertical arrow is I2 = I2 /I22 . So I (1) /(I (1) )2 ∼
= I2 /I22 . This proves (iii).
Let’s prove (i). Let ai = (. . . , ai3 , ai2 , 0) (i = 1, . . . , k) be a finite family of
elements in A = inv. limn An such that ai2 generate I2 . Note that ai ∈ I (1) and
ain ∈ In . For each n, let Jn be the ideal of An generated by ain (i = 1, . . . , k).
Since φn2 : An → A2 (n ≥ 2) is surjective, we have
φn2 (In ) = I2 = J2 = φn2 (Jn ).
Since kerφn2 = In2 , this implies
In = Jn + In2 .
Combining with the fact that Inn = 0, we get
Inn−1 = Jnn−1 .
Let J (1) be the ideal of A generated by ai (i = 1, . . . , k). We claim that
I (n) = (J (1) )n .
Then (i) follows. Obviously, we have J (1) ⊂ I (1) and hence (J (1) )n ⊂ (I (1) )n .
Using the fact that Inn = 0, one can show (I (1) )n ⊂ I (n) . So we have (J (1) )n ⊂
I (n) . Given x = (. . . , xn+2 , xn+1 , 0, . . . , 0) ∈ I (n) , we have
n n
xn+1 ∈ kerφn+1,n = In+1 = Jn+1 .
So X
xn+1 = λi1 ...ik ,n+1 ai1,n+1
1
· · · aik,n+1
k
i1 +···+ik =n
for some λi1 ...ik ,n+1 ∈ An+1 . Choose λi1 ...ik ∈ A so that their images in An+1
are λi1 ...ik ,n+1 . Then
X
x− λi1 ...ik ai11 · · · aikk ∈ I (n+1) .
i1 +···+ik =n
By the same argument, we can find λi1 ...ik ∈ A for any i1 + · · · + ik = n + 1

such that
X X
x− λi1 ...ik ai11 · · · aikk − λi1 ...ik ai11 · · · aikk ∈ I (n+2) .
i1 +···+ik =n i1 +···+ik =n+1
Similarly, we can find λi1 ...ik ∈ A for any i1 + · · · + ik ≥ n such that for any
m ≥ 0, we have
X
x− λi1 ...ik ai11 · · · aikk ∈ I (n+m+1) .
n≤i1 +···ik ≤n+m
Consider the formal series

X
λi1 ...ik ai11 · · · aikk .
i1 +···+ik ≥n
It is a well-defined element in A = inv. limi Ai since its image in each Ai is a

finite sum. We have
X \
x− λi1 ...ik ai11 · · · aikk ∈ I (n+m+1) = 0.
i1 +···+ik ≥n m≥0
So X
x= λi1 ...ik ai11 · · · aikk .
i1 +···+ik ≥n
λi1 ...ik ai11 · · · aikk as

P
We can write
i1 +···+ik ≥n
X
fi1 ...ik (a1 , . . . , ak )ai11 · · · aikk ,
i1 +···+ik =n
where the coefficients fi1 ...ik (a1 , . . . , ak ) are formal power series in a1 , . . . , ak
which give well-defined elements in A. Then we have
X
x= fi1 ...ik (a1 , . . . , ak )ai11 · · · aikk ∈ (J (1) )n .
i1 +···+ik =n
So I (n) ⊂ (J (1) )n and hence I (n) = (J (1) )n .
Proposition 5.11. Keep the assumptions and notations in Proposition 5.10.

Let (Mn , ψmn )n∈N be an inverse system such that each Mn is an An -module
and for any m ≥ n, ψmn : Mm → Mn is a homomorphism compatible with the
module structures and the ring homomorphism φmn : Am → An . Suppose M1
n
is a finitely generated A1 -module and each ψmn is surjective with kernel Im Mm .
Let M = inv. limn Mn and let ψn : M → Mn be the canonical homomorphisms.
Then M is a finitely generated A-module, and each ψn induces an isomorphism
M/(I (1) )n M ∼ = Mn .
Proof. Let zi = (. . . , zi2 , zi1 ) (i = 1, . . . , k) be a finite family of elements

in M = inv. limn Mn such that zi1 generate M1 . For each n, let Nn be the
submodule of Mn generated by zin (i = 1, . . . , k). Since ψn1 : Mn → M1 is
surjective, we have
ψn1 (Mn ) = M1 = N1 = ψn1 (Nn ).
Since kerψn1 = In Mn , this implies
Mn = Nn + In Mn .
Combining with that fact that Inn = 0, we get
Inn−1 Mn = Inn−1 Nn .
Let N be the submodule of M generated by zi (i = 1, . . . , k) and let M (n) =

kerψn . We claim that
M (n) = I (n) N.
Taking n = 0, we get M = N and hence M is a finitely generated A-module.

Since each ψmn is surjective, each ψn is surjective and it induces an isomorphism
M/M (n) ∼ = Mn . But M/(I (1) )n M = M/I (n) M = M/M (n) . The proposition
follows. Let’s prove our claim. For any
z = (. . . , zn+2 , zn+1 , 0, . . . , 0) ∈ M (n) ,
we have
n n
zn+1 ∈ kerψn+1,n = In+1 Mn+1 = In+1 Nn+1 .
So we have
k
X
zn+1 = ri,n+1 zi,n+1
i=1
n (n+1)
for some ri,n+1 ∈ In+1 . Choose ri ∈ A so that their images in An+1 are
(n+1) (n)
ri,n+1 . Then we have ri ∈I and
k
(n+1)
X
z− ri zi ∈ M (n+1) .
i=1
(n+2)
By the same argument, we can find ri ∈ I (n+1) such that
k k
(n+1) (n+2)
X X
z− ri zi − ri zi ∈ M (n+2) .
i=1 i=1
(n+m)
Similarly, we can find ri ∈ I (n+m−1) (i = 1, . . . , k) for any m ≥ 1 such that
k
(n+1) (n+2) (n+m)
X
z− (ri + ri + · · · + ri )zi ∈ M (n+m) .
i=1
We then have
k X ∞
(n+m)
X
z= ( ri )zi ∈ I (n) N.
i=1 m=1
(n) (n)
So M ⊂I N . Obviously we have M (n) ⊃ I (n) N . Hence M (n) = I (n) N .
Let X be a noetherian scheme and I a coherent sheaf of ideals of OX .

b O b ), where X
Consider the ringed space (X, b is the closed subset
X
X
b = supp(OX /I)
of X and OXb is the sheaf of rings
OXb = inv. lim(OX /I n )|Xb .

n
b O b ) the formal completion of X with respect to I. We often denote

We call (X, X
b O b ) by X.
(X, b We have a canonical morphism of ringed spaces
X
(i, i] ) : (X,
b O b ) → (X, OX ),
X
b → X is the inclusion, and

where i : X
i] : OX → i∗ OXb = inv. lim i∗ i−1 (OX /I n )

n
is induced by the family of the canonical morphisms OX → i∗ i−1 (OX /I n ). For

each n ∈ N, let
(Xn , OXn ) = (supp(OX /I n ), (OX /I n )|supp(OX /I n ) )
be the closed subscheme of X with the ideal sheaf I n . Then Xn has the same
underlying topological space as X.b Let in : Xn → X and inm : Xn → Xm
(m ≥ n) be the canonical closed immersions. For each n, we have a canonical
morphism
în : (Xn , OXn ) → (X,
b O b ),
X
where în : Xn → X
b is identity, and
î]n : OXb = inv. lim(OX /I n )|Xb → OXn = (OX /I n )|Xn

n
is the canonical morphism. The following diagram commutes:

n i
(Xn , OXn ) → (X, OX )
inm ↓ & în ↑i
î
m
(Xm , OXm ) → b O b ).
(X, X
For any OX -module F, we define the formal completion Fb of F with respect

to I to be the OXb -module
Fb = inv. lim(F/I n F)|Xb .

n
We have
Fb = inv. lim în∗ i∗n F
n
and the family of canonical morphisms F → in∗ i∗n F (n ∈ N) induces a mor-

phism
F → i∗ Fb = inv. lim in∗ i∗n F.
n
Let I be coherent sheaf of OX -ideals such that supp(OX /I) = supp(OX /I 0 ).

0
k0
Then by Proposition 4.14, we have I k ⊂ I 0 and I 0 ⊂ I for some natural
numbers k and k 0 . So the formal completions of X and F with respect to I
are isomorphic to the formal completions of X and F with respect to I 0 , re-
spectively. Hence formal completions with respect to I depend only on the
underlying topological space of the closed subscheme of X with ideal sheaf I.
Proposition 5.12. Let X be a noetherian scheme, I a coherent sheaf of ideals

of OX , and (X, b O b ) the formal completion of X with respect to I. Then
X
b O b ) is a locally ringed space and the canonical morphism (i, i] ) : (X, OX ) →
(X, X
b O b ) is a morphism of locally ringed space.
(X, X
Proof. The problem is local. We may assume X = SpecA for some noetherian
ring A. Let I be the ideal of A corresponding to I. We have
b = supp(A/I)∼ = V (I).
X
A point in Xb corresponds to a prime ideal p of A satisfying p ⊃ I. Note that

D(f ) ∩ X (f ∈ A − p) form a fundamental system of neighborhoods of p in X.
b b
So
OX,
b p = dir. lim OXb (D(f ) ∩ X).
b
f 6∈p
We have
b = inv. lim(A/I n )f = inv. lim Af /(If )n = A

OXb (D(f ) ∩ X) cf ,
n n
where A
cf is the If -adic completion of Af . So
OX,
b p = dir. lim Af .
c
f 6∈p
We need to show this is a local ring. Let

n
f = inv. lim pf /(If ) .
pc
n
By Corollary 5.3, it is isomorphic to the If -adic completion of pf . It is an ideal

of A f is an ideal of OX,b p = dir. lim Af . To prove OX,
cf . So dir. lim pc c b p is a local
f 6∈p f 6∈p
ring, it suffices to show every element in dir. lim A cf − dir. lim pcf is a unit. It is
f 6∈p f 6∈p
enough to show that for any element z ∈ A cf − pc f , there exists a g ∈ A − p which
is divisible by f such that the image of z under the canonical homomorphism
cf → A
A cg is a unit. Let z0 be the image of z under the canonical homomorphism
Af → Af /If . Choose fxk ∈ Af so that its image in Af /If is z0 . Since z 6∈ pc
c f , we
have x 6∈ p. Let g = f x. Then g 6∈ p and the image z00 of z0 under the canonical
k+1
homomorphism Af /If → Ag /Ig is a unit. (Its inverse is the image of f g ∈ Ag
in Ag /Ig .) Let z 0 be the image of z under the homomorphism A cf → A cg . Then
the image of z under Ag → Ag /Ig is z0 , which is a unit. Since Ag /Ig ∼
0 c 0 c b = Ag /Ig ,
the image of z 0 in A cg /Ibg is a unit and hence does not lie in any maximal ideal of
cg /Ibg . But Ibg ⊂ rad(A
A cg ) by Proposition 5.7, so z 0 does not lie in any maximal
ideal of A cg and hence is a unit. This proves O b = dir. limf 6∈p A cf is a local
X,p
ring with maximal ideal dir. limf 6∈p pc f . Note that OX,p = dir. limf 6∈p Af and its
maximal ideal is dir. limf 6∈p pf . It is easy to see the canonical homomorphism
dir. lim Af → dir. lim A cf

f 6∈p f 6∈p
is local. So the canonical morphism (i, i] ) : (X, OX ) → (Y, OY ) is a morphism

of locally ringed space.

of OX , and Xb the formal completion of X with respect to I.
(i) For any exact sequence of coherent OX -modules
0 → F 0 → F → F 00 → 0,
the sequence
0 → Fb0 → Fb → Fb00 → 0
is exact.
b → X be the canonical morphism. For any coherent OX -module
(ii) Let i : X
F, we have a canonical isomorphism
i∗ F ∼
= F.
b
(iii) For any coherent OX -modules F and G, we have
Fb ⊗OXc Gb ∼
= (F ⊗OX G)∧ ,
HomOXc (F, b ∼
b G) = (HomOX (F, G))∧ .
Proof. (i) Let

0 → F 0 → F → F 00 → 0
be an exact sequence of coherent OX -modules. Let U = SpecA be an affine
∼ ∼
open subscheme of X. Suppose F 0 |U ∼ = M 0 , F|U ∼= M ∼ , and F 00 |U ∼
= M 00
0 00
for some finitely generated A-modules M , M , and M . Then we have an exact
sequence
0 → M 0 → M → M 00 → 0.
Let I be the ideal of A corresponding to I. We have
F(U
b ∩ X) b ∼
b = inv. lim(F/I n F)| b (U ∩ X) = inv. lim M/I n M = M
c,
n X n
where M b ∼
c is the I-adic completion of M . Similarly, we have Fb0 (U ∩ X) c0
=M
00 ∼ c00
and F = M . By Proposition 5.4, the sequence
b
0 → Fb0 (U ∩ X)
b → F(U b → Fb00 (U ∩ X)
b ∩ X) b →0
is exact. When U goes over the family of affine open subschemes of X, U ∩ X

b
forms a basis for the topology of X. So the sequence
b
0 → Fb0 → Fb → Fb00 → 0
is exact.
(ii) The canonical morphism F → i∗ Fb induces a morphism i∗ F → F. Let’s
prove it is an isomorphism. This is a local question, so we may assume there
exists an exact sequence
m n
OX → OX → F → 0.
i∗ OXm
→ i∗ OXn
→ i∗ F → 0
↓ ↓ ↓
Obm → Obn → Fb → 0.
X X
The two rows are exact. Obviously the first two vertical arrows are isomor-
phisms, so the third vertical arrow is also an isomorphism.
(iii) The first equality follows from (ii) and the fact that
i∗ F ⊗i∗ OX i∗ G ∼
= i∗ (F ⊗OX G).
We have a canonical morphism
HomOX (F, G) → i∗ HomOXc (i∗ F, i∗ G).
It induces a morphism
i∗ HomOX (F, G) → HomOXc (i∗ F, i∗ G).
Let’s prove it is an isomorphism. We may work locally and assume there exists
an exact sequence
m n
OX → OX → F → 0.
0 → i∗ HomOX (F, G) → i∗ HomOX (OX n

, G) → i∗ HomOX (OXm
, G)
↓ ↓ ↓
0 → HomOXc (i∗ F, i∗ G) → HomOXc (i∗ OX
n ∗
, i G) → HomOXc (i∗ OX
m ∗
, i G).
The two rows are exact and the last two vertical arrows are isomorphisms. So
the first vertical arrow is an isomorphism. Our assertion then follows from (ii).

of OX , and Xb the formal completion of X with respect to I.
(i) For any coherent sheaf of ideals J of OX , the formal completion Jb of J
with respect to I is a sheaf of ideals of OXb .
(ii) For any n, we have OXb /Icn ∼

= (OX /I n )|Xb .
(iii) For any 0 ≤ m ≤ n, we have Icn ⊂ Icm and I cn ∼
m /I
c = (I m /I n )|Xb .
(iv) For any coherent sheaf of ideals J of OX and any n, we have
Jcn ∼
= i∗ (J n ) ∼
= J n OXb ∼
= Jbn .
Proof. We have an exact sequence
0 → J → OX .
By Proposition 5.13 (i), the sequence
0 → Jb → O
d X
is exact, that is,

0 → Jb → OXb
is exact. So Jb is a sheaf of ideals of OXb .
For any 0 ≤ m ≤ n, we have an exact sequence of coherent OX -modules
0 → I n → I m → I n /I m → 0.
0 → Icn → Ic
m → I\
n /I m → 0
is exact. Using the definition of the formal completion, one can show
m /I n ∼
I\ = (I m /I n )|Xb .
This shows Icn ⊂ Ic

m and
cn ∼
m /I
Ic = (I m /I n )|Xb .
Taking m = 0, we get
OXb /Icn ∼
= (OX /I n )|Xb .
By Proposition 5.13 (i) and (ii), the canonical morphism
i∗ J n → i∗ OX ∼
= OXb
is injective. Its image is J n OXb . So
i∗ J n ∼
= J n OXb .
The other assertions in (iv) are obvious.

of OX , F a coherent OX -module, X
b and Fb the formal completions of X and
F with respect to I, respectively. Then the kernel of the canonical homomor-

phism F(X) → F( b consists of those sections s ∈ F(X) which vanish in a
b X)
neighborhood of Xb in X.
Proof. If s ∈ F(X) vanishes in a neighborhood of X b in X, then its image in

each (F/I F)|Xb (X) is 0. Hence its image in F(X) = inv. limn (F/I n F)|Xb (X)
n b b b b
is 0. Conversely, suppose s ∈ F(X) lies in the kernel of F(X) → F(X). b b
Then for any affine open subscheme U = SpecA of X, s|U lies in the kernel
of F(U ) → F(U
b ∩ X).
b Suppose F|U ∼ = M ∼ for some finitely generated A-
module M and let I be the ideal of A corresponding to I. We have F(U ) ∼ =M
and F(U ∩ X) = M , where M is the I-adic completion of M . Let x ∈ M
b b c c
be the element corresponding to s|U ∈ F(U ). Then x lies in the kernel of the
canonical homomorphism M → M c. By Proposition 5.8, there exists an r ∈ I
such that (1 + r)x = 0. Note that the image of 1 + r in Ap is a unit for any
p ∈ V (I) ∼
= U ∩ X.
b So there exists a neighborhood V of V (I) in U = SpecA
such that the image of x in M ∼ (V ) vanishes. Since U is arbitrary, there exists
a neighborhood of Xb in X over which s vanishes.
Corollary 5.16. Let X be a noetherian scheme, I a coherent sheaf of ideals of

OX , X b the formal completions of X with respect to I, u : F → G a morphism
of coherent OX -modules and û : Fb → Gb the morphism induced by u on formal
completions with respect to I.
(i) û vanishes if and only if u vanishes in a neighborhood of X
b in X.
(ii) û is a monomorphism (resp. epimorphism, resp. isomorphism) if and
only if u is a monomorphism (resp. epimorphism, resp. isomorphism) in a
neighborhood of X b in X.
Proof. (i) By Proposition 4.6 (i), HomOX (F, G) is coherent. Regarding u as

a section of HomOX (F, G), our assertion follows from Proposition 5.15 applied
to this coherent sheaf.
(ii) We have an exact sequence
u
0 → keru → F → G → cokeru → 0.

û
0 → keru
d → Fb → Gb → cokeru
[ →0
d → Fb vanishes. By (i), this

is exact. So û is a monomorphism if and only if keru
is equivalent to saying keru → F vanishes in a neighborhood of X b in X, that
is, u is a monomorphism in a neighborhood of X b in X. Similarly, we can prove
the other assertions in (ii).
Let X be a noetherian scheme, I a coherent sheaf of ideals of OX , X

b the
formal completion of X with resect to I, and Xn the closed subschemes of X
with ideal sheaves I n (n ∈ N). Fix notations by the following commutative

diagram (m ≥ n).
n i
(Xn , OXn ) → (X, OX )
inm ↓ % im ↑i
î
m
(Xm , OXm ) → b O b ),
(X, X
where all the arrows are the canonical morphisms. A coherent inverse system
(Fn , φmn )n∈N consists of an OXn -module Fn on Xn for each n and a morphism
φmn : Fm → inm∗ Fn for each pair m ≥ n such that the following conditions
hold:
(a) Each Fn is a coherent OXn -module.
(b) φnn = idFn for every n and imk∗ (φmn ) ◦ φkm = φkn for every triple
k ≥ m ≥ n.
(c) For every pair m ≥ n, φmn induces an isomorphism i∗nm Fm ∼ = Fn , that
is, Fm ⊗OXm OXn ∼ = Fn .
An OXb -module F is called coherent if
F∼
= inv. lim în∗ Fn
n
for some coherent inverse system (Fn , φmn )n∈N .

Suppose F is a coherent OX -module. Then (i∗n F, φmn )n∈N form a coherent
inverse system, where φmn is the canonical morphism
i∗m F → inm∗ i∗nm i∗m F ∼

= inm∗ i∗n F
for any m ≥ n. So
Fb = inv. lim în∗ i∗n F
n
is a coherent OXb -module. A coherent OXb -module is called algebraizable if it is

of the form Fb for some coherent OX -module F. Not every coherent OXb -module
is algebraizable. But see Proposition 5.20 and Theorem 2.7.10 for cases where
the answer is positive.
Proposition 5.17. Notation as above. Suppose (Fn , φmn )n∈N is a coherent

inverse system and F = inv. limn în∗ Fn . Then for each n, we have canonical
isomorphisms
î∗n F = F ⊗OXc OXn ∼= F/Ibn F ∼
= Fn ,
and for each pair m ≥ n, φmn is identified with the canonical morphism
î∗m F → inm∗ i∗nm î∗m F ∼

= inm∗ î∗n F.
Proof. Each canonical morphism F = inv. limn în∗ Fn → în∗ Fn induces a mor-
phism
î∗n F → Fn .
Let’s show it is an isomorphism. It suffices to show that for any affine open
subscheme U = SpecA of X, the canonical homomorphism
F(U ∩ X)
b ⊗ b OXn (U ∩ Xn ) → Fn (U ∩ Xn )
O c(U ∩X)
X
is an isomorphism. For each n, let Mn be a finitely generated A/I n -module

such that Fn |U ∩Xn ∼
= Mn∼ . For each pair m ≥ n, the morphism φmn : Fm →
inm∗ Fn induces a homomorphism ψmn : Mm → Mn which is compatible with
the modules structures and the ring homomorphism A/I m → A/I n . Since φmn
induces an isomorphism i∗nm Fm ∼ = Fn , ψmn is surjective with kernel I n Mm .
Let M = inv. limn Mn . By Proposition 5.11, the canonical homomorphism
M → Mn induces an isomorphism
M ⊗Ab A/I n ∼
= Mn .
But we have
F(U ∩ X)
b = inv. lim Fn (U ∩ Xn ) ∼
= inv. lim Mn ,
n n
OXb (U ∩ X)
b ∼
= A,
b
OXn (U ∩ Xn ) ∼
= A/I n ,
Fn (U ∩ Xn ) ∼
= Mn ,
so
F(U ∩ X)
b ⊗ ∼
b OXn (U ∩ Xn ) = Fn (U ∩ Xn ).
O c(U ∩X)
X
By Propositions 5.14 and 4.13, we have
î∗n F = F ⊗OXc OXn ∼

= F ⊗OXc OXb /Ibn OXb ∼
= F/Ibn F.
For each pair m ≥ n, the morphism inm : Xn → Xm is identity on the underlying

topological spaces. So φmn : Fm → inm∗ Fn can be considered as a morphism
Fm → Fn . Similarly the canonical morphism î∗m F → inm∗ i∗nm î∗m F ∼
= inm∗ î∗n F
m n
can be considered as a morphism F/Ib F → F/Ib F. Obviously we have a
commutative diagram
F/Ibm F → F/Ibn F
↓ ↓
φmn
Fm → Fn .
So φmn can be identified with the canonical morphism î∗m F → inm∗ î∗n F.
Corollary 5.18. Keep the above notations. Let F be an OXb -module and
let (Fn , φmn )n∈N be the inverse system defined by Fn = î∗n F for each n and
the canonical morphism φmn : î∗m F → inm∗ i∗nm î∗m F ∼
= inm∗ î∗n F for each pair
m ≥ n. Then F is a coherent OX -module if and only if each Fn is a coherent
OXn -module and F ∼ = inv. limn Fn .
Proof. Obviously we have i∗nm Fm ∼ = Fn . The “if” part follows from the
definition of coherent OXb -modules. The “only if” part follows from Proposition
5.17.
Corollary 5.19. Notation as above. Let (Fn , φmn )n∈N and (Gn , ψmn )n∈N be
two coherent inverse systems and let F = inv. limn în∗ Fn and G = inv. limn în∗ Gn .
Then we have a bijection
HomOXc (F, G) ∼
= inv. lim HomOXn (Fn , Gn ).
n
Proof. Every element (un ) ∈ inv. limn HomOXn (Fn , Gn ) defines a family of
morphisms
în∗ (un ) : în∗ Fn → în∗ Gn .
Passing to inverse limit, we get a morphism
inv. lim în∗ (u) : inv. lim în∗ Fn → inv. lim în∗ Gn .
n n n
We define
S : inv. lim HomOXn (Fn , Gn ) → HomOXc (F, G), (un ) 7→ inv. lim în∗ (u).
n n
Any u ∈ HomOXc (F, G) induces a family of morphisms î∗n (u) : î∗n F → î∗n G.
By Proposition 5.17, this family can be identified with a family of morphisms
un : Fn → Gn . One can verify (un ) ∈ inv. lim HomOXn (Fn , Gn ). We define
T : HomOXc (F, G) → inv. lim HomOXn (Fn , Gn ), u 7→ (un ).

n
One can verify that S and T are inverse to each other.
Proposition 5.20. Let A be a noetherian ring, I an ideal of A, X = SpecA, and

Xb the formal completion of X with respect to I ∼ . If A is I-adically complete,
then any coherent OXb -module is of the form (M ∼ )∧ , where (M ∼ )∧ is the formal
completion of M ∼ with respect to I ∼ .
Proof. Let F be a coherent OXb -module. Then F ∼ = inv. limn în∗ Fn for some
coherent inverse system (Fn , φmn )n∈N . On each Xn = SpecA/I n , we have
Fn ∼= Mn∼ for some finitely generated A/I n -module Mn . For every pair m ≥ n,
the morphism φmn : Fm → inm∗ Fn induces a homomorphism ψmn : Mm → Mn
which is compatible with the module structures and the ring homomorphism
A/I m → A/I n . Moreover, since φmn induces an isomorphism i∗nm Fm ∼ = Fn ,
ψmn is surjective with kernel I n Mn . Let M = inv. limn Mn . By Proposition
5.11, M is a finitely generated Ab = inv. limn A/I n -module and the canonical
homomorphism M → Mn induces an isomorphism M ⊗Ab A/I n ∼ = Mn . Since
A is I-adically complete, M is a finite generated A module and the canonical

homomorphism M → Mn induces an isomorphism M/I n M ∼ = Mn . We have
(M ∼ )∧ = inv. lim(M ∼ /(I ∼ )n M ∼ )|Xb

n
∼
= inv. lim(M/I n M )∼ |Xb
n
∼
= inv. lim în∗ Mn∼
n
∼
= inv. lim în∗ Fn
n
∼
= F,
that is, F ∼
= (M ∼ )∧ .

of OX , X b the formal completion of X with respect to I, and F and G coherent
OXb -modules.
(i) F ⊗OXc G and HomOXc (F, G) are coherent OXb -modules.
(ii) For any morphism u : F → G, the OXb -modules keru, cokeru and imu
are coherent.
Proof. We only prove (ii). The proof of (i) is similar. Let K = kerφ and let
Kn = î∗n K for each n. By Corollary 5.18, we need to show each Kn is a coherent
OXn -module and K ∼ = inv. limn în∗ Kn . This is a local problem. We may assume
X = SpecA for some noetherian ring A. Let I be the ideal of A corresponding
to I. Note that replacing A and I by their I-adic completions A b and Ib doesn’t
change X and Xn and thus has no effect on K and Kn . Making this replacement,
b
we are reduced to the case where A is I-adically complete. By Proposition 5.20,
we have F ∼ = (M ∼ )∧ and G ∼ = (N ∼ )∧ for some finitely generated A-modules M
and N . By Corollary 5.19, the morphism u : F → G is induced by a compatible
family of homomorphisms un : M/I n M → N/I n N . Passing to inverse limit and
using the fact that M and N are necessarily I-adically complete (Proposition
5.5), we get a homomorphism û = inv. limn un : M → N . By Propositions 4.2
(iii) and 5.13 (i), the sequence
0 → ((kerû)∼ )∧ → (M ∼ )∧ → (N ∼ )∧
is exact. So we have K ∼ = ((kerû)∼ )∧ and hence K is coherent. Similarly, we

can prove the other assertions in (ii).
Proposition 5.22. Let f : X → Y be a morphism of noetherian schemes, I

a coherent sheaf of ideals of OY , X b and Yb the formal completions of X and
Y with respect to IOX and I, respectively, and fˆ : X b → Yb the morphism of
locally ringed spaces induced by f .
(i) For any coherent OY -module G, we have a canonical isomorphism fd ∗G ∼
=
fˆ∗ G,
b where fd b are the formal completions of f ∗ G and G with respect to
∗ G and G
IOX and I, respectively.

(ii) For any coherent OYb -module G, fˆ∗ G is a coherent OXb -module.
Proof. (i) We have a commutative diagram

iX
b →
X X
fˆ ↓ ↓f
i Y
Yb → Y,
where the horizontal arrows are the canonical morphisms. Both G and f ∗ G are
coherent. By Proposition 5.13 (ii), we have Gb ∼ ∗G ∼
= i∗Y G and fd = i∗X f ∗ G. Our
∗ ∗ ∼ ˆ∗ ∗
assertion then follows from the fact that iX f G = f iY G.
(ii) By Corollary 5.18, proving fˆ∗ G is coherent is local problem. So we may
assume X = SpecB and Y = SpecA for some noetherian ring A and some
noetherian A-algebra B. Let I be the ideal of A corresponding to I. Replacing
A, I and B by their I-adic completions A, b Ib and B b doesn’t change X, b Yb
and fˆ. So we may assume A is I-adically complete. By Proposition 5.20, we
have G ∼ = (M ∼ )∧ for some finitely generated A-module M . By (i), we have
fˆ∗ ((M ∼ )∧ ) ∼
= (f ∗ M ∼ )∧ . So G ∼
= (f ∗ M ∼ )∧ is coherent.
Chapter 2
Cohomology
2.1 Derived Functors
Let C be an abelian category. A complex in C is a sequence

di di+1
· · · → Ai → Ai+1 → Ai+2 → · · ·
of morphisms in C such that di+1 di = 0 for each i. The morphisms di are called
the derivation operators of the complex. We often denote the above complex by
A· . If Ai is only specified for a certain range of i, say i ≥ 0, then we set Ai = 0
for all other i. Let A· and B · be two complexes. A morphism of complexes
f · : A· → B · consists of morphisms f i : Ai → B i (i ∈ Z) such that the diagram
di
Ai → Ai+1
i
f ↓ ↓ f i+1
di+1
Bi → B i+1
commutes for every i. We define the compositions of morphisms in the obvious

way. We thus get the category of complexes in C, which is an abelian category.
Define respectively the i-th cocycle object Z i (A· ), coboundary object B i (A· ), and
cohomology object H i (A· ) of a complex A· by
Z i (A· ) = kerdi ,
B i (A· ) = imdi−1 ,
H i (A· ) = Z i (A· )/B i (A· ).
Any morphism f · : A· → B · of complexes induces morphisms on cohomology

objects H i (A· ) → H i (B · ). Two morphisms f · , g · : A· → B · are homotopic if
there exist morphisms hi : Ai → B i−1 (i ∈ Z) such that
f i − g i = hi+1 di + di−1 hi .
99
100 CHAPTER 2. COHOMOLOGY
Then f · and g · induce the same morphisms on cohomology objects. We write

f · ∼ g · if f · and g · are homotopic. Two complexes A· and B · are called ho-
motopically equivalent if there exist morphisms f · : A· → B · and g · : B · → A·
such that g · f · ∼ idA· and f · g · ∼ idB · . Then f · and g · induce isomorphisms on
cohomology objects. A sequence of morphisms of complexes
A· → B · → C ·
is called exact if
Ai → B i → C i
is exact for each i. Given a short exact sequence
0 → A· → B · → C · → 0,
we can define canonically a morphism δ i : H i (C · ) → H i+1 (A· ) for each i such

that the following sequence is exact:
δ i−1 δi
· · · → H i (A· ) → H i (B · ) → H i (C · ) → H i+1 (A) → · · · .
It is called the long exact sequence of cohomology objects associated to the above
short exact sequence of complexes.
An additive covariant functor F : C → D between abelian categories is called
left exact if for any exact sequence
0→A→B→C→0
in C, the sequence
0 → F (A) → F (B) → F (C)
is exact. F is called right exact if
F (A) → F (B) → F (C) → 0
is exact. F is called exact if it is both left exact and right exact. This is
equivalent to saying that F transforms any exact sequence in C into an exact
sequence in D. An additive contravariant G : C → D is called left exact if the
sequence
0 → G(C) → G(B) → G(A)
is exact. It is called right exact if
G(C) → G(B) → G(A) → 0
is exact. It is called exact if it is both left exact and right exact.
Examples.
1. For any abelian category C and any object A in C, the functor Hom(A, −)
from C to the category of abelian groups is left exact and covariant, and the
functor Hom(−, A) is left exact and contravariant.
2.1. DERIVED FUNCTORS 101
2. The functor F 7→ F(X) from the category of sheaves of abelian groups

on a topological space X to the category of abelian groups is left exact.
3. Let f : X → Y be a continuous map between topological spaces. Then
F 7→ f∗ F is a left exact functor from the category of sheaves of abelian groups
on X to that on Y .
4. Let (X, OX ) be a ringed space and F an OX -module. Then the functor
F ⊗OX − on the category of OX -modules is right exact.
An object I in an abelian category C is called injective if Hom(−, I) is an

exact functor. This is equivalent to saying that for any object A in C and
any sub-object B of A, any morphism B → I can be extended to a morphism
A → I. An object P in C is called projective if Hom(P, −) is an exact functor.
This is equivalent to saying that for any object A and any quotient B of A, any
morphism P → B can be lifted to a morphism P → A. Note that an object is
injective if and only if it is projective in the dual category. So if an assertion
holds for injective objects, then its dual assertion holds for projective objects.
In the following, we mainly state results for injective objects.
Let
i
0→A→B→C→0
be a short exact sequence. If A is injective, then the sequence splits. Indeed, idA
can be extended to a morphism π : B → A. We have πi = idA . Our assertion
then follows from Proposition 1.1.3.
An injective resolution of an object A is complex of the form
0 → I0 → I1 → · · ·
together with a monomorphism A → I 0 such that each I i is injective and the

sequence
0 → A → I0 → I1 → · · ·
is exact. A projective resolution of A is a complex of the form
· · · → P −1 → P 0 → 0
together with an epimorphism P 0 → A such that each P i is projective and the

sequence
· · · → P −1 → P 0 → A → 0
is exact. We say a category C has enough injective objects if for any object A in
C, there exists a monomorphism from A to an injective object in C. We say C has
enough projective objects if for any object A in C, there exists an epimorphism
from a projective object to A.
Proposition 1.1. Let C be an abelian category.

(i) If C has enough injective objects, then any object in C has an injective
resolution.
(ii) Suppose
0 → A → A0 → A1 → · · · ,
0 → B → I0 → I1 → · · ·
are two complexes such that the first sequence is exact and each I i is injective.
Then for any morphism f : A → B, there exists a morphism of complexes
f · : A· → I · such that
0 → A → A0 → A1 → · · ·
0 1
f↓ f ↓ f ↓
0 → B → I0 → I1 → · · ·
commutes. We call such f · a morphism extending f . If g · : A· → I · is another

morphism extending f , then f · and g · are homotopic.
(iii) Any two injective resolutions of an object A are homotopically equiva-
lent.
(iv) Let
0→A→B→C→0
be an exact sequence in C and let I · and J · be injective resolutions of A and C,
respectively. Then there exists an exact sequence
0 → B → I0 ⊕ J0 → I1 ⊕ J1 → · · ·
such that the following diagram commutes:
0 0 0
↓ ↓ ↓
0 → →
A B → C → 0
↓ ↓ ↓
0 → I0
→ I0 ⊕ J0 → J0 → 0
↓ ↓ ↓
0 → I1
→ I1 ⊕ J1 → J1 → 0
↓ ↓ ↓
.. ..
··· . .
where the morphisms I · → I · ⊕ J · and I · ⊕ J · → J · are the canonical ones.
Proof. (i) Let A be an object in C. Choose a monomorphism A → I 0 such that

I 0 is an injective object. Then we have an exact sequence
0 → A → I 0.
Suppose we have found an exact sequence
0 → A → I 0 → · · · → I n−1 .
Choose a monomorphism
coker(I n−2 → I n−1 ) → I n
such that I n is an injective object. (When n = 1, we make the convention that
A = I n−2 .) Then we have an exact sequence
0 → A → I 0 → · · · → I n.
In this way, we get an injective resolution of A.
(ii) Since I 0 is injective and A is a sub-object of A0 , the composition
f
A → B → I0
can be extended to a morphism f 0 : A0 → I 0 . The diagram
0 → A → A0
0
f↓ f ↓
0 → B → I0
commutes. Suppose we have defined f i for any i < n such that the diagram
0 → A → A0 → A1 → · · · → An−1
0 1 n−1
f↓ f ↓ f ↓ f ↓
0 → B → I0 → I1 → · · · → I n−1
commutes. Note that the composition
f n−1
An−1 → I n−1 → I n
vanishes on im(An−2 → An−1 ). Hence it induces a morphism
coker(An−2 → An−1 ) → I n .
Since I n is injective and coker(An−2 → An−1 ) is a sub-object of An , this mor-
phism can be extended to a morphism f n : An → I n . The following diagram
commutes:
0 → A → A0 → A1 → ··· → An
0 1 n
f↓ f ↓ f ↓ f ↓
0 → B → I0 → I1 → ··· → I n.
In this way, we get a morphism f · : A· → I · extending f : A → B.
Suppose g · : A· → I · is another morphism extending f . Let I −1 = 0 and
define h0 : A0 → I −1 to be the zero morphism. Suppose we have defined
hi : Ai → I i−1 for any i ≤ n such that hi+1 di + di−1 hi = f i − g i for any
i ≤ n − 1. We have
(f n − g n − dn−1 hn )dn−1
= dn−1 (f n−1 − g n−1 ) − dn−1 hn dn−1
= dn−1 (hn dn−1 + dn−2 hn−1 ) − dn−1 hn dn−1
= 0.
So f n −g n −dn−1 hn vanishes on im(An−1 → An ) and hence induces a morphism

from coker(An−1 → An ) to I n . But coker(An−1 → An ) is a sub-object of
An+1 and I n is injective, so this morphism can be extended to a morphism
hn+1 : An+1 → I n . We then have hn+1 dn + dn−1 hn = f n − g n .
(iii) Let I · and J · be two injective resolutions of A. By (ii), there exists a
morphism f · : I · → J · extending idA and a morphism g · : J · → I · extending
idA . Note that both g · f · : I · → I · and idI · are morphisms extending idA . So by
(ii), we have g · f · ∼ idI · . Similarly f · g · ∼ idJ · . So I · and J · are homotopically
equivalent.
(iv) Since I 0 is injective and A is a sub-object of B, the morphism A → I 0
can be extended to a morphism B → I 0 . Let
B → I0 ⊕ J0
be the morphism defined by B → I 0 and B → C → J 0 . We have a commutative
diagram
0 → A → B → C → 0
↓ ↓ ↓
0 → I 0 → I 0 ⊕ J 0 → J 0 → 0.
The first and the third vertical arrows are monomorphisms. So B → I 0 ⊕ J 0 is
also a monomorphism. Suppose we have defined the morphism I i−1 ⊕ J i−1 →
I i ⊕ J i for any i ≤ n such that the diagram
0 0 0
↓ ↓ ↓
0 → A → B → C → 0
↓ ↓ ↓
0 → I0 → I0 ⊕ J0 → J0 → 0
↓ ↓ ↓
.. .. ..
. . .
↓ ↓ ↓
0 → In → In ⊕ Jn → Jn → 0
commutes and all the columns are exact. Then the sequence
0 → coker(I n−1 → I n ) → coker(I n−1 ⊕ J n−1 → I n ⊕ J n ) → coker(J n−1 → J n ) → 0
is exact. Actually this is a part of the long exact sequence of the cohomology
objects associated to the short exact sequence of complexes
0 → I · → I · ⊕ J · → J · → 0,
where I · is the complex
0 → A → I0 → · · · → In → 0
and similarly for J · and I · ⊕ J · . Since I n+1 is injective, we can extend the
morphism coker(I n−1 → I n ) → I n+1 to a morphism
coker(I n−1 ⊕ J n−1 → I n ⊕ J n ) → I n+1 .
This last morphism together with the composition
coker(I n−1 ⊕ J n−1 → I n ⊕ J n ) → coker(J n−1 → J n ) → J n+1
defines a morphism
coker(I n−1 ⊕ J n−1 → I n ⊕ J n ) → I n+1 ⊕ J n+1
which is necessarily a monomorphism. Then the diagram
0 0 0
↓ ↓ ↓
0 → A → B → C → 0
↓ ↓ ↓
0 → I0 → I0 ⊕ J0 → J0 → 0
↓ ↓ ↓
.. .. ..
. . .
↓ ↓ ↓
0 → I n+1 → I n+1 ⊕ J n+1 → J n+1 → 0
commutes and all the columns are exact.
Let F : C → D be a left exact covariant functor between abelian categories.

Assume C has enough injective objects. We define the right derived functors
·
Ri F (i = 0, 1, . . .) of F as follows: Choose an injective resolution IA for each
i
object A in C once and for all. For each i, we define R F (A) to be the i-th
·
cohomolgy object of the complex F (IA ). Given a morphism A → B in C, there
· ·
exists a morphism of complexes IA → IB extending A → B which is unique up
to homotopy by Proposition 1.1 (ii). It induces a well defined morphism
· ·
Ri F (A) = H i (F (IA )) → Ri F (B) = H i (F (IB )).
The isomorphic class of the functor Ri F is independent of the choice of injective

resolutions by Proposition 1.1 (iii). Using the assumption that F is left exact,
one can verify R0 F = F . For any injective object I in C, we have Ri F (I) = 0
for any i ≥ 1. Given a short exact sequence
0 → A → B → C → 0,
by Proposition 1.1 (iv), we may find a short exact sequence of complexes

· ·
0 → IA → IB → IC· → 0
· ·
such that IA , IB and IC· are injective resolutions of A, B and C, respectively,
and for each i, we have
i ∼ i
IB = IA ⊕ ICi .
Then
· ·
0 → F (IA ) → F (IB ) → F (IC· ) → 0
is a (split) short exact sequence of complexes. So there exists a morphism
δ i : Ri F (C) → Ri+1 F (A)
for each i such that the sequence
δ i−1 δi
· · · → Ri F (A) → F i F (B) → Ri F (C) → Ri+1 F (A) → · · ·
is exact. We call this sequence the long exact sequence of Ri F associated to the
short exact sequence
0 → A → B → C → 0.
Given a morphism of short exact sequences
0 → A → B → C → 0
↓ ↓ ↓
0 → A0 → B0 → C0 → 0,
the following commutative diagram commutes
δi
Ri F (C) → Ri+1 F (A)
↓ ↓
δi
Ri F (C 0 ) → Ri+1 F (A0 )
for each i.
An object J is called F -acyclic if Ri F (J) = 0 for any i ≥ 1. An F -acyclic
resolution of an object A is a complex of the form
0 → J0 → J1 → · · ·
together with a monomorphism A → J 0 such that each J i is F -acyclic and the

sequence
0 → A → J0 → J1 → · · ·
is exact.
Proposition 1.2. Let F : C → D be a left exact covariant functor between

abelian categories. Suppose C has enough injective objects. Let J · be an F -
acyclic resolution of A. Then for any i, we have Ri F (A) ∼
= H i (F (J · )).
Proof. By the long exact sequence for Ri F associated to the short exact se-
quence
0 → A → J 0 → Z 1 (J · ) → 0
and the fact that Ri F (J 0 ) = 0 for any i ≥ 1, we have
Ri−1 F (Z 1 (J · )) ∼
= Ri F (A)
for any i ≥ 2. By the long exact sequence for Ri F associated to the short exact
sequence
0 → Z 1 (J · ) → J 1 → Z 2 (J · ) → 0
and the fact that Ri F (J 1 ) = 0 for any i ≥ 1, we have
Ri−2 F (Z 2 (J · )) = Ri−1 F (Z 1 (J · )) = Ri F (A)
for any i ≥ 3. In this way, we get
Ri F (A) = Ri−1 F (Z 1 (J · )) = · · · = R1 F (Z i−1 (J · ))
for any i ≥ 1. By the long exact sequence for Ri F associated to the short exact
sequence
0 → Z i−1 (J · ) → J i−1 → Z i (J · ) → 0
and the fact that R1 F (J i−1 ) = 0, we have an exact sequence
0 → F (Z i−1 (J · )) → F (J i−1 ) → F (Z i (J · )) → R1 F (Z i−1 (J · )) → 0
and hence
R1 F (Z i−1 (J · )) = F (Z i (J · ))/im(F (J i−1 ) → F (Z i (J · ))).
We have an exact sequence
0 → Z i (J · ) → J i → J i+1 .
Since F is left exact, we have an exact sequence
0 → F (Z i (J · )) → F (J i ) → F (J i+1 )
and hence
F (Z i (J · )) = ker(F (J i ) → F (J i+1 )).
So
R1 F (Z i−1 (J · )) = ker(F (J i ) → F (J i+1 ))/im(F (J i−1 ) → F (Z i (J · )))

= H i (F (J · )).
Therefore Ri F (A) = H i (F (J · )) for any i ≥ 1. Using the fact that F is left

exact, one can show R0 F (A) = H 0 (F (J · )).
Proposition 1.3. Let F : C → D be a left exact covariant functor between

abelian categories. Suppose C has enough injective objects. Let S be a family
of objects satisfying the following conditions:
(a) For any object A in C, there exists a monomorphism from A to an object
in S.
(b) If an object A in C is isomorphic to a direct factor of an object in S,
then A is an object in S.
(c) Let
0→A→B→C→0
be an exact sequence in C. If A and B are objects in S, then C is also an object
in S and the sequence
0 → F (A) → F (B) → F (C) → 0
is exact.
Then any injective object in C is an object in S, and any object in S is
F -acyclic.
Proof. Let I be an injective object in C and let I → S be a monomorphism

from I to an object S in S. Denote the cokernel of this monomorphism by S/I.
Then we have an exact sequence
0 → I → S → S/I → 0.
This sequence necessarily splits. So I is isomorphic to a direct factor of S. By

(b), I is an object in S.
Let S be an object in S and let I · be an injective resolution of S. We then
have exact sequences
0 → S → I0 → Z 1 (I · ) → 0,
1 ·
0 → Z (I ) → I 1 → Z 2 (I · ) → 0,
0 → Z 2 (I · ) → I 2 → Z 3 (I · ) → 0,
..
.
By (c), these short exact sequences imply successively that Z 1 (I · ), Z 2 (I · ), . . .

are objects in S and the following sequences are exact:
0 → F (S) → F (I 0 ) → F (Z 1 (I · )) → 0,
0 → F (Z (I )) → F (I 1 ) → F (Z 2 (I · )) → 0,
1 ·
0 → F (Z 2 (I · )) → F (I 2 ) → F (Z 3 (I · )) → 0,
..
.
This implies that
0 → F (S) → F (I 0 ) → F (I 1 ) → F (I 2 ) → · · ·
is exact. So Ri F (S) = H i (F (I · )) = 0 for any i ≥ 1. Hence objects in S are

F -acyclic.
Let C and D be abelian categories. A covariant cohomological functor from

C to D is a family of covariant functors T = (T i ) (i = 0, 1, . . .) together with
morphisms δ i : T i (C) → T i+1 (A) for each short exact sequence
0→A→B→C→0
satisfying the following conditions:

(i) Associated to the above short exact sequence, we have a long exact se-
quence
δ0
0 → T 0 (A) → T 0 (B) → T 0 (C) → T 1 (A) → · · · .
(ii) For any morphism of short exact sequences
0 → A → B → C → 0
↓ ↓ ↓
0 → A0 → B0 → C0 → 0,
the following diagram commutes for each i:

δi
T i (C) → T i+1 (A)
↓ ↓
δi
T i (C 0 ) → T i+1 (A0 ).
We say T = (T i ) is universal if for any cohomological functor T 0 = (T 0i ) and any

natural transformation f 0 : T 0 → T 00 , there exists a unique family of natural
transformations f i : T i → T 0i (i ≥ 1) such that for any short exact sequence
0 → A → B → C → 0,
the following diagram commutes for each i:

δi
T i (C) → T i+1 (A)
fCi ↓ ↓ fAi+1
δi
T 0i (C) → T 0i+1 (A).
A functor F : C → D is called effaceable if for any object A in C, there exists

a monomorphism u : A → M such that F (u) = 0. It is called coeffaceable if for
any object A in C, there exists an epimorphism u : M → A such that F (u) = 0.
Proposition 1.4. Let T = (T i ) : C → D be a covariant cohomological functor

between abelian categories. If T i is effaceable for any i ≥ 1, then T is universal.
Proof. Let T 0 = (T 0i ) be a cohomological functor and f 0 : T 0 → T 00 a natural

transformation. Suppose we have shown there exists a unique family of natural
transformations f i : T i → T 0i (0 ≤ i ≤ n − 1) such that for any short exact
sequence
0 → A → B → C → 0,
the following diagram commutes whenever 0 ≤ i ≤ n − 2:
δi
T i (C) → T i+1 (A)
fCi ↓ ↓ fAi+1
δi
T 0i (C) → T 0i+1 (A).
For any object A in C, choose a monomorphism u : A → M such that T n (u) = 0

and let M/A be its cokernel. We have a short exact sequence
0 → A → M → M/A → 0.
Consider the following diagram:
T n−1 (M ) → T n−1 (M/A) → T n (A) → 0
n−1 n−1
fM ↓ fM/A ↓ fAn ↓
0n−1 0n−1 0n
T (M ) → T (M/A) → T (A).
Note that the two rows are exact. So there exists a unique morphism
fAn : T n (A) → T 0n (A)
making the diagram commute. We claim the morphism fAn is independent of
the choice of the monomorphism u : A → M satisfying T n (u) = 0. Indeed, let
v : A → N be another monomorphism satisfying T n (v) = 0 and let L be the
cokernel of the morphism A → M ⊕ N defined by −u : A → M and v : A → N .
Then we have a commutative diagram
u
A → M
v↓ &w ↓
N → L,
and the morphism w : A → L is a monomorphism satisfying T n (w) = 0. The
morphism M → L induces a morphism M/A → L/A. The commutativity of
the diagram
T n−1 (M/A) → T n−1 (L/A)
n−1 n−1
fM/A ↓ ↓ fL/A
0n−1 0n−1
T (M/A) → T (L/A).
implies that the morphism fAn : T n (A) → T 0n (A) defined by the monomorphism
u coincides with the one defined by w. Similarly the morphism fAn defined by
the monomorphism v also coincides with the one defined by w.
Let α : A → B be a morphism in C. Choose monomorphisms u : A → M
and v0 : B → N0 such that T n (u) = 0 and T n (v0 ) = 0. Let N be the cokernel
of the morphism A → M ⊕ N0 defined by −u : A → M and v0 α : A → N0 .
Then we have a commutative diagram
u
A → M
α↓ ↓
v
B → N
such that u and v are monomorphisms satisfying T n (u) = 0 and T n (v) = 0. We
T n−1 (M/A) → T n−1 (N/B)
n−1 n−1
fM/A ↓ ↓ fN/B
T 0n−1 (M/A) → T 0n−1 (N/B).
This implies that the following diagram commutes:
T n (A) → T n (B)
fAn ↓ ↓ fBn
T (A) → T 0n (B).
0n
So f n is a natural transformation.
Given a short exact sequence
0 → A → B → C → 0,
choose a monomorphism v : B → M such that T n (v) = 0. Let u be the

v
composition A → B → M . Then T n (u) = 0. We have a commutative diagram
0 → A → B → C → 0
k v↓ ↓
u
0 → A → M → M/A → 0.
It gives rise to the following commutative diagram:

δ n−1
T n−1 (C) → T n (A)
↓ k
δ n−1
T n−1 (M/A) → T n (A)
n−1
fM/A ↓ ↓ fAn
δ n−1
T 0n−1 (M/A) → T 0n (A)
↑ k
δ n−1
T 0n−1 (C) → T 0n (A).
Combining with the commutativity of the diagram
T n−1 (C) → T n−1 (M/A)

fCn−1 ↓ n−1
↓ fM/A
T 0n−1 (C) → T 0n−1 (M/A),
this implies the following diagram commutes:

δ n−1
T n−1 (C) → T n (A)
fCn−1 ↓ ↓ fAn
δ n−1
T 0n−1 (C) → T 0n (A).
Corollary 1.5. Let F : C → D be a left exact covariant functor between

abelian categories. Assume C has enough injective objects. Then (Ri F ) is a
universal cohomological functor. Conversely, if T = (T i ) is a covariant universal
cohomological functor, then T 0 is left exact and T i is isomorphic to Ri T 0 for
any i ≥ 0.
Proof. Obviously (Ri F ) is a cohomological functor. For any object A in C,

choose a monomorphism u : A → I such that I is an injective object. Since
Ri F (I) = 0 for any i ≥ 1, we have Ri F (u) = 0 for any i ≥ 1. This shows that
Ri F is effaceable. So (Ri F ) is universal by Proposition 1.4.
Conversely, suppose T = (T i ) is a covariant universal cohomological functor.
Then T 0 is left exact by the definition of the cohomological functor. Both (T i )
and (Ri T 0 ) are universal cohomological functors and they have the same 0-th
component. So they are isomorphic.
We have seen that when C has enough injective objects, we can define the
right derived functors Ri F of any left exact covariant functor F using injective
resolutions and they form a universal cohomological functor. Similarly if C has
enough projective objects, we can define the left derived functors Li F of any
right exact covariant functor F using projective resolutions. We can also define
the left derived functors (resp. right derived functors) of any contravariant right
exact (resp. left exact) functor using injective (resp. projective) resolutions. We
leave to the reader to formulate their universal properties.
Proposition 1.6. Let (X, OX ) be a ringed space. Then the abelian category
of sheaves of OX -modules has enough injective objects.
Proof. For any open subset U of X, define a presheaf PU by

OX (V ) if V ⊂ U,
PU (V ) =
{0} if V ⊂
6 U.
Let GU be the sheaf associated to the presheaf PU . For any OX -module F

and any section s ∈ F(U ), define a morphism PU → F as follows: If V ⊂ U ,
we define PU (V ) = OX (V ) → F(V ) by a 7→ as|V . If V 6⊂ U , we define
PU (V ) = {0} → F(V ) to be 0. The morphism PU → F induces a morphism
GU → F, which we call the morphism defined by the section s ∈ F(U ). Let
G = ⊕U GU , where the direct sum is taken over all open subsets of X. Then G
is an OX -module.
First we prove an OX -module I is injective if for any OX -submodule G 0 of G,
any morphism from G 0 to I can be extended to a morphism from G to I. Given
an OX -module F, an OX -submodule F 0 of F and a morphism φ0 : F 0 → I,
we need to show there exists a morphism φ : F → I such that φ|F 0 = φ0 .
Let S be the family of pairs (F 00 , φ00 ) such that F 00 is an OX -submodule of F
containing F 0 and φ00 : F 00 → I is a morphism satisfying φ00 |F 0 = φ0 . By Zorn’s
lemma, S has a maximal element. To prove our assertion, it suffices to show for
any maximal element (F 00 , φ00 ) in S, we have F 00 = F. To simplify notations,
suppose (F 0 , φ0 ) is maximal and let’s prove F 0 = F. If F 0 6= F, then there
exists a section s ∈ F(W )−F 0 (W ) for some open subset W . Define a morphism
ψ : G = ⊕U GU → F so that ψ|GU = 0 if U 6= W and ψ|GW : GW → F is the
morphism defined by the section s. Let G 0 be the inverse image of F 0 under
ψ and let ψ 0 : G 0 → F 0 be the morphism induced by ψ. By assumption, there
exists a morphism θ : G → I such that θ|G 0 = φ0 ψ 0 . We have an exact sequence

α β
0 → G 0 → F 0 ⊕ G → F,
where α is defined by g 0 7→ (ψ 0 (g 0 ), −g 0 ) and β is defined by (f 0 , g) 7→ f 0 + ψ(g).

Consider the morphism
δ 0 : F 0 ⊕ G 7→ I, (f 0 , g) 7→ φ0 (f 0 ) + θ(g).
We have δ 0 α = 0. So there exists a morphism δ : imβ → I such that δβ = δ 0 .

Note that imβ is an OX -submodule of F containing F 0 and δ|F 0 = φ0 . Moreover
imβ strictly contains F 0 since s is a section of imβ over W but not a section of
F 0 over W . This contradicts to the maximality of (F 0 , φ0 ). So we have F 0 = F.
This proves our assertion.
Let F be an OX -module and let I(F) be the set of all morphisms φi : Gi → F
from OX -submodules Gi of G to F. Consider the morphism
M
φ: Gi → F ⊕ G |I(F )| ,
i∈I(F )
where G |I(F )| is the direct sum of copies G and each element in I(F) corre-
sponds to a copy of G, and φ|Gi : Gi → F ⊕ G |I(F )| is the morphism defined by
−φi : Gi → F and the canonical inclusion of Gi into the i-th copy of G in G |I(F )| .
Let I1 (F) be the cokernel of φ and let φ(F) : F → I1 (F) be the morphism de-
fined by composing the canonical inclusion F → F ⊕ G |I(F )| with the canonical
epimorphism F ⊕ G |I(F )| → cokerφ. Note that φ(F) is a monomorphism. More-
over, each morphism φi : Gi → F can be extended to the morphism G → I1 (F)
induced by the morphism G → F ⊕ G |I(F )| which embeds G into the i-th copy
of G in F ⊕ G |I(F )| .
φi
Gi → F
↓ ↓
G → I1 (F).
Next we use transfinite induction to define Ii (F) for any ordinal i and a
monomorphism Ij (F) → Ii (F) for any pair of ordinals j < i such that for any
triple of ordinals k < j < i, the following diagram commutes:
Ik (F) → Ij (F)
& ↓
Ii (F).
We define I0 (F) = F. Let i be an ordinal and suppose we have defined Ij (F) for
any j < i and monomorphisms Ik (F) → Ij (F) for any k < j < i. If i = i0 + 1
for some ordinal i0 , we define Ii (F) = I1 (Ii0 (F)), and for any j < i, define
Ij (F) → Ii (F) to be the composition of the monomorphism Ij (F) → Ii0 (F)
with the monomorphism Ii0 (F) → I1 (Ii0 (F)). If i is a limit ordinal, we define
Ii (F) = dir. limj<i Ij (F), and for any j < i, define Ij (F) → dir. limj<i Ij (F)
to be the canonical morphism which is necessarily a monomorphism.
Let n be the smallest ordinal whose cardinal |n| is strictly larger than the
cardinal |S| of the set S consisting of all OX -submodules of G. Note that n is
a limit ordinal. We have a monomorphism F = I0 (F) → In (F). Let’s show
In (F) is an injective sheaf. This would prove the category of OX -modules has
enough injective objects. It suffices to show any morphism φ0 : G 0 → In (F)
from an OX -submodule G 0 of G to In (F) has image contained in In0 (F) for
some ordinal n0 < n. Indeed, if this is true, then we can extend the morphism
G 0 → In0 (F) induced by φ0 to a morphism G → In0 +1 (F). Composing this
last morphism with the monomorphism In0 +1 (F) → In (F), we get a morphism
G → In (F) extending the morphism φ : G 0 → In (F). By our discussion at the
beginning, this implies In (F) is an injective sheaf.
It suffices to show there exists an ordinal n0 < n such that φ0−1 (Ii (F)) =
0−1
φ (In0 (F)) for any n0 ≤ i < n. If this is not true, then we can find a set
I of ordinals strictly less than n with limit n such that φ0−1 (Ii (F)) is strictly
contained in φ0−1 (Ii+1 (F)) for any i ∈ I. Each φ0−1 (Ii (F)) is an OX -submodule
of G. So we have |I| ≤ |S|, where |I| denote the cardinal of the set I. By the
minimality of n, we have |i| S ≤ |S| for any i < n. Since n is the limit of the
ordinals in I, we have n = i. Hence
i∈I
X X
|n| ≤ |i| ≤ |S| ≤ |I||S| ≤ |S|.
i∈I i∈I
This contradicts to |n| > |S|.
Let (X, OX ) be a ringed space. By Proposition 1.6, we may talk about

the right derived functors of left exact covariant functors on the category of
OX -modules. Here are some examples:
Examples.
1. On the category of sheaves of abelian groups on a topological space X,
the i-th right derived functor of the left exact functor F 7→ F(X) is denoted
by H i (X, F) and is called the i-th cohomology group of F. For any open subset
U of X, we define H i (U, F) to be the i-th right derived functor of the left
exact functor F 7→ F(U ). We often denote H 0 (U, F) by Γ(U, F). We have
Γ(U, F) = F(U ).
Even when (X, OX ) is a ringed space and F is an OX -module, we calculate
H i (X, F) by regarding F as a sheaf of abelian groups. In this case, H i (X, F)
has an OX (X)-module structure. We will see later that the functor H i (X, −)
defined in this way is also the i-th derived functor of Γ(X, −) on the category
of OX -modules.
2. Let (X, OX ) be a ringed space and F an OX -module. On the cate-
gory of OX -modules, the i-th right derived functor of the left exact functor
G 7→ HomOX (F, G) is denoted by ExtiOX (F, G) and is called a higher exten-
sion group of F by G. The i-th right derived functor of the left exact functor
G → HomOX (F, G) is denoted by ExtiOX (F, G) and is called a higher extension
sheaf of F by G.
3. Let f : X → Y be a continuous map of topological spaces. On the

category of sheaves of abelian groups, the i-th derived functor of the left exact
functor F 7→ f∗ F is denoted by Ri f∗ F and is called a higher direct image of F.
Even when (f, f ] ) : (X, OX ) → (Y, OY ) is a morphism of ringed spaces and
F is an OX -module, we calculate Ri f∗ F by regarding F as a sheaf of abelian
groups. In this case, Ri f∗ F has an OY -module structure. We will see later that
the functor Ri f∗ defined in this way is also the i-th right derived functor of f∗
on the category of OX -modules.
4. Let A be a ring. For any A-module M , we may find an epimorphism
F → M such that F is a free A-module. In particular, the category of A-
modules has enough projective objects. So we may talk about the left derived
functors of right exact covariant functors on the category of A-modules. For any
A-module M , the i-th left derived functor of the right exact functor N → M ⊗N
is denoted by TorA i (M, N ). Given any projective resolution
· · · → P −1 → P 0 → N → 0
of N , we have
TorA ∼ −i ·
i (M, N ) = H (M ⊗A P ).
Lemma 1.7. Let (X, OX ) be a ringed space and I an injective OX -module.

Then for any open subset U of X, I|U is an injective OU -module.
Proof. Let j : U → X be the inclusion. For any OU -modules G, define j! G to

be the sheaf on X associated to the presheaf

G(V ) if V ⊂ U,
V 7→
0 otherwise.
Obviously G 7→ j! G is an exact functor. We leave to the reader show that j! is
left adjoint to j ∗ . For any OU -module G, we have
HomOU (G, I|U ) ∼

= HomOX (j! G, I).
Since j! is an exact functor and I is injective, HomOX (j! −, I) is an exact functor.
So HomOU (−, I|U ) is an exact functor and hence I|U is an injective OU -module.
Corollary 1.8. Let X be a topological space and F a sheaf of abelian groups on

X. Then for any open subset U of X and any i, we have H i (U, F) ∼
= H i (U, F|U ).
Proof. Let I · be an injective resolution of F. Then I · |U is an injective resolu-

tion of F|U . So we have
H i (U, F|U ) ∼
= H i (I · |U (U )) = H i (I · (U )) = H i (U, F).
Proposition 1.9. Let f : X → Y be a continuous map. For any sheaf of

abelian groups F on X and any i, Ri f∗ F is the sheaf associated to the presheaf
V 7→ H i (f −1 (V ), F).
Proof. Let I · be an injective resolution of F. Then
Ri f∗ F = ker(f∗ I i → f∗ I i+1 )/im(f∗ I i−1 → f∗ I i ).
Hence Ri f∗ F is the sheaf associated to the presheaf
V 7→ ker((f∗ I i )(V ) → f∗ (I i+1 )(V ))/im((f∗ I i−1 )(V ) → (f∗ I i )(V ))

= ker(I i (f −1 (V )) → I i+1 (f −1 (V )))/im(I i−1 (f −1 (V )) → I i (f −1 (V )))
∼
= H i (f −1 (V ), F).
Our assertion follows.
A sheaf of abelian groups F on a topological space X is called flasque if for

any nonempty open subset U of X, the restriction F(X) → F(U ) is surjective.
This is equivalent to saying that for any inclusion V ⊂ U of nonempty open
subsets of X, the restriction F(U ) → F(V ) is surjective. For any sheaf F on
X, define a sheaf C 0 (F) so that for any open subset U of X, we have
a
C 0 (F)(U ) = {s : U → FP |s(P ) ∈ FP }.
P ∈X
Obviously C 0 (F) is flasque and we have a canonical monomorphism
F → C 0 (F)
defined by mapping each section s ∈ F(U ) to the section

a
U→ FP , P 7→ sP
P ∈X
in C 0 (F)(U ). Define
C 1 (F) = C 0 (coker(F → C 0 (F)),

C n+1 (F) = C 0 (coker(C n−1 (F) → C n (F))).
0 → F → C 0 (F) → C 1 (F) → · · · .
So C · (F) is a flasque resolution of F. We call it the Godement resolution of F.
Proposition 1.10. For any i, the functor F 7→ C i (F) on the category of sheaves
of abelian groups on X is exact.
Proof. For any sheaf F, denote the morphism F → C 0 (F) by d−1 F and the
morphism C i (F) → C i+1 (F) by diF . We will show simultaneously that the
functors F 7→ C i (F) and F 7→ cokerdi−1
F are exact by induction on i. Suppose
0→F →G→H→0
is an exact sequence of sheaves. Then for any P ∈ X,
0 → FP → GP → HP → 0
is exact. This implies that
0 → C 0 (F)(U ) → C 0 (G)(U ) → C 0 (H)(U ) → 0
is exact for any open subset U of X. Hence
0 → C 0 (F) → C 0 (G) → C 0 (H) → 0
is exact. This proves F 7→ C 0 (F) is exact. We have a commutative diagram
0 → F → G → H → 0
d−1
F ↓ d−1
G ↓ d−1
H ↓
0 → C 0 (F) → C 0 (G) → C 0 (H) → 0.
Note that all the vertical arrows are monomorphisms. So by the Snake Lemma,
we have an exact sequence
0 → cokerd−1 −1 −1
F → cokerdG → cokerdH → 0.
Hence F 7→ cokerd−1 i
F is an exact functor. Suppose we have proved F 7→ C (F)
i−1 i+1
and F 7→ cokerdF are exact functors. Then F 7→ C (F) is an exact functor
since
C i+1 (F) = C 0 (cokerdi−1
F ).
For any short exact sequence
0 → F → G → H → 0,
0 → cokerdi−1
F → cokerdGi−1 → cokerdi−1
H → 0
↓ ↓ ↓
0 → C i+1 (F) → C i+1 (G) → C i+1 (H) → 0.
All the vertical arrows are monomorphisms. So by the Snake Lemma,
0 → cokerdiF → cokerdiG → cokerdiH → 0
is exact. Hence F 7→ cokerdiF is an exact functor.
Lemma 1.11. Let

0 → F 0 → F → F 00 → 0
be an exact sequence of sheaves of abelian groups on a topological space X. If
F 0 is flasque, then
0 → F 0 (X) → F(X) → F 00 (X) → 0

is exact, and for any continuous map f : X → Y ,
0 → f∗ F 0 → f∗ F → f∗ F 00 → 0
is exact. If both F 0 and F are flasque, then F 00 is also flasque.
Proof. Suppose F 0 is flasque. Let s00 ∈ F 00 (X) be a section. Since F → F 00

is surjective, there exists an open covering {Ui }i∈I of X such that each section
s00 |Ui in F 00 (Ui ) can be lifted to a section si in F(Ui ). Note that
sij = si |Ui ∩Uj − sj |Ui ∩Uj
are mapped to 0 under the morphism F → F 00 . But F 0 is the kernel of this

morphism. So we may regard sij as sections in F 0 (Ui ∩ Uj ). We claim that there
exist sections t0i ∈ F 0 (Ui ) such that
sij = t0i |Ui ∩Uj − t0j |Ui ∩Uj .
We then have
(si − t0i )|Ui ∩Uj = (sj − t0j )|Ui ∩Uj .
So we can glue si − t0i together to get a section in F(X). It is mapped to the
section s00 ∈ F 00 (X). So F(X) → F 00 (X) is surjective. Now it is easy to see
0 → F 0 (X) → F(X) → F 00 (X) → 0
is exact. Since F 0 |U is necessarily flasque for any open subset U of X,
0 → F 0 (U ) → F(U ) → F 00 (U ) → 0
is also exact. In particular, for any continuous map f : X → Y and any open
subset V of Y , the sequence
0 → (f∗ F 0 )(V ) → (f∗ F)(V ) → (f∗ F 00 )(V ) → 0
is exact. So
0 → f∗ F 0 → f∗ F → f∗ F 00 → 0
is exact.
Let’s prove our claim. Let S be the set of families of sections (t0i )i∈J such
that J ⊂ I, t0i ∈ F 0 (Ui ) and sij = t0i |Ui ∩Uj − t0j |Ui ∩Uj for any i, j ∈ J. By Zorn’s
lemma, S has a maximal element. It suffices to show that for any maximal
element (t0i )i∈J in S, we have J = I. Suppose J 6= I and let i0 ∈ I − J. For any
i, j ∈ J, we have
t0i − t0j = sij = si0 j − si0 i
on Ui ∩ Uj ∩ Ui0 . Hence
t0i + si0 i = t0j + si0 j
on Ui ∩ Uj ∩ Ui0 . Since Ui ∩ Ui0 (i ∈ J) form an open covering (∪i∈J Ui ) ∩ Ui0 , we
may glue (t0i + si0 i )|Ui ∩Ui0 together to get a section in F 0 ((∪i∈J Ui ) ∩ Ui0 ). Since
F 0 is flasque, we may then extend this section to a section t0i0 ∈ F 0 (Ui0 ). But
then (t0i )i∈J∪{i0 } defines an element in S. This contradicts to the maximality of
(t0i )i∈J . So we must have I = J. This proves our claim.
Finally assume both F 0 and F are flasque. For any open subset U of X, we
have the following commutative diagram:
0 → F 0 (X) → F(X) → F 00 (X) → 0
↓ ↓ ↓
0 → F 0 (U ) → F(U ) → F 00 (U ) → 0.
We have proved each row is exact. The first two vertical arrows are surjective.
So the last vertical arrow is also surjective and hence F 00 is flasque.
By Proposition 1.3 and Lemma 1.11, flasque sheaves are acyclic with respect
to the functor F 7→ F(X) on the category of sheaves of abelian groups. So by
Proposition 1.2, we may use flasque resolutions and in particular the Gode-
ment resolution of F to calculate H i (X, F). If X is a ringed space, then by
Proposition 1.3 and Lemma 1.11, any injective OX -module is flasque. Hence
any injective resolution in the category of OX -modules is also a flasque resolu-
tion. So the i-th derived functor of F 7→ F(X) on the category of OX -modules
coincides with F → H i (X, F).
Similarly, for any continuous map f : X → Y and any sheaf of abelian
groups F, we can calculate Ri f∗ F using flasque resolutions and in particular
the Godement resolution of F. If f : X → Y is a morphism of ringed space,
the functor F 7→ Ri f∗ F is the i-th derived functor of f∗ on the category of
OX -modules.
Proposition 1.12. Let X be a noetherian topological space.

(i) Let Fi (i ∈ I) be a family of sheaves of abelian groups on X. We have
H n (X, ⊕i∈I Fi ) ∼
= ⊕i∈I H n (X, Fi ) for any n.
(ii) Let I be a direct set and (Fi )i∈I a direct system of sheaves of abelian
groups on X. We have H n (X, dir. limi Fi ) ∼ = dir. limi H n (X, Fi ) for any n.
Proof. Note that (i) follows from (ii). Indeed any direct sum of sheaves is the
direct limit of direct sums with finitely many factors, and (i) holds for any finite
direct sum since each H n (X, −) is an additive functor and hence commutes with
finite direct sums. Let’s prove (ii). Since X is a noetherian topological space,
the presheaf U 7→ dir. limi Fi (U ) is a sheaf and coincides with dir. limi Fi . Let
C · (Fi ) be the Godement resolution of Fi . Then (C · (Fi ))i∈I is a direct system
of complexes of sheaves. One can show dir. limi C · (Fi ) is a flasque resolution of
dir. limi Fi . So we have
H n (X, dir. lim Fi ) = H n (Γ(X, dir. lim C · (Fi )))
i i
= H (dir. lim Γ(X, C · (Fi )))
n
i
= dir. lim H n (Γ(X, C · (Fi )))
i
= dir. lim H n (X, Fi ).
i
Proposition 1.13. Let (X, OX ) be a ringed space and let F and G be two OX -
modules. Then for each i, ExtiOX (F, G) is the sheaf associated to the presheaf
U 7→ ExtiOU (F|U , G|U ).
Proposition 1.14. Let (X, OX ) be a ringed space, let F be an OX -module

and let
0 → G 0 → G → G 00 → 0
be an exact sequence of OX -modules. Then we have long exact sequences
· · · → ExtiOX (F, G 0 ) → ExtiOX (F, G) → ExtiOX (F, G 00 ) → Exti+1 0

OX (F, G ) → · · ·
and
· · · → ExtiOX (G 00 , F) → ExtiOX (G, F) → ExtiOX (G 0 , F, ) → Exti+1 00

OX (G , F) → · · · .
We have similar long exact sequences for Ext·OX (−, −).
Proof. The first long exact sequence follows from the fact that (Exti (F, −))
is a cohomological functor. Let I · be an injective resolution of F. Since
HomOX (−, I) is an exact functor for any injective OX -module I, we have a
short exact sequence of complexes
0 → HomOX (G 00 , I · ) → HomOX (G, I · ) → HomOX (G 0 , I · ) → 0.
The second long exact sequence in the proposition is the long exact sequence of
cohomology groups associated to this short exact sequence of complexes.
Let F and G be OX -modules on a ringed space (X, OX ). An extension of F

by G is a short exact sequence of OX -modules of the form
0 → G → E → F → 0.
Two extensions
0 → G → E1 → F → 0
and
0 → G → E2 → F → 0
are called isomorphic if there exists a commutative diagram
0 → G → E1 → F → 0
k ↓ k
0 → G → E2 → F → 0.
Note that the middle vertical arrow is necessarily an isomorphism.

OX -modules. Then there is a one-to-one correspondence between Ext1OX (F, G)
and the set of isomorphic classes of extensions of F by G.
Proof. Fix an injective resolution

d0 d1
0 → G → I0 → I1 → · · ·
of G. We have
Ext1OX (F, G)
= ker(HomOX(F, I 1) → HomOX(F, I 2))/im(HomOX(F, I 0) → HomOX(F, I 1))
= HomOX (F, Z 1 (I · ))/im(HomOX (F, I 0 ) → HomOX (F, I 1 )).
Let S be the set of isomorphic classes of extensions of F by G. Define a map
Φ : Ext1OX (F, G) → S
as follows: Given any element
e ∈ Ext1OX (F, G) = HomOX (F, Z 1 (I · ))/im(HomOX (F, I 0 ) → HomOX (F, I 1 )),
let e0 : F → Z 1 (I · ) be an element in HomOX (F, Z 1 (I · )) whose image in

Ext1OX (F, G) is e. We define Φ(e) to be the isomorphic class of the pulling-
back of the short exact sequence
0 → G → I 0 → Z 1 (I · ) → 0
by the morphism e0 : F → Z 1 (I · ), that is, Φ(e) is the isomorphic class of the

extension
0 → G → I 0 ⊕e0 F → F → 0,
where I 0 ⊕e0 F is the subsheaf of I 0 ⊕ F consisting of sections (s, t) such that
d0 (s) = e0 (t), the arrow G → I 0 ⊕e0 F is the morphism defined by G → I 0 and
0
the zero morphism G → F, and the arrow I 0 ⊕e0 F → F is the projection. We
claim Φ(e) is independent of the choice of e0 : F → Z 1 (I · ) and hence Φ(e) is
well-defined. Indeed, if e00 : F → Z 1 (I · ) is another morphism whose image in
Ext1OX (F, G) is e, then e0 − e00 = d0 f for some morphism f ∈ HomOX (F, I 0 ).
We have a commutative diagram
0 → G → I 0 ⊕e0 F → F → 0
k ↓ k
0 → G → I 0 ⊕e00 F → F → 0,
where the middle vertical arrow I 0 ⊕e0 F → I 0 ⊕e00 F maps each section (s, t)
of I 0 ⊕e0 F to the section (s − f (t), t) of I 0 ⊕e00 F. This proves the pulling back
of
0 → G → I 0 → Z 1 (I · ) → 0
by e0 and by e00 are isomorphic. Hence Φ(e) is independent of the choice of e0 .

Define Ψ : S → Ext1OX (F, G) as follows: Let E ∈ S be the isomorphic class
of the extension
0 → G → E → F → 0.
We can extend the morphism G → I 0 to a morphism ψ : E → I 0 since I 0
is injective. Let ψ̄ : F → Z 1 (I · ) be the morphism which makes the following
diagram commute:
0 → G → E → F → 0
k ψ↓ ψ̄ ↓
0 → G → I 0 → Z 1 (I · ) → 0.
We define Ψ(E) to be the image of ψ̄ : F → Z 1 (I · ) in
Ext1OX (F, G) = HomOX (F, Z 1 (I · ))/im(HomOX (F, I 0 ) → HomOX (F, I 1 )).
We claim that Ψ(E) is independent of the choice of ψ : E → I 0 . Indeed, if

ψ 0 : E → I 0 is another morphism extending G → I 0 , then ψ − ψ 0 vanishes on
im(G → E) and hence induces a morphism δ : F → I 0 . We have ψ̄−ψ̄ 0 = d0 δ. So
the images of ψ̄ and ψ̄ 0 in Ext1OX (F, G) are the same. Hence Ψ(E) is independent
of the choice of ψ. Now it is easy to see Ψ is well-defined.
One can verify that the two maps Φ and Ψ are inverse to each other. This
proves our proposition.
Proposition 1.16. Let A be a ring, let M be an A-module, and let
0 → N 0 → N → N 00 → 0
be an exact sequence of A-modules. Then we have long exact sequences

00 0 00
· · · → TorA A A A
i+1 (M, N ) → Tori (M, N ) → Tori (M, N ) → Tori (M, N ) → · · ·
and
00 0 00
· · · → TorA A A A
i+1 (N , M ) → Tori (N , M ) → Tori (N, M ) → Tori (N , M ) → · · · .
An A-module M is called flat if the functor M ⊗A − is exact. Note that any

free A-module is flat. Since any projective A-module is a direct factor of a free
A-module, any projective A-module is also flat.
Proposition 1.17. Let A be a ring and M an A-module. The following

conditions are equivalent:
(i) M is flat.
(ii) For any A-module N and any i ≥ 1, we have TorA i (M, N ) = 0.
(iii) For any A-module N , we have TorA
1 (M, N ) = 0.
(iv) For any exact sequence of finitely generated A-modules
0 → N 0 → N,
the sequence
0 → M ⊗A N 0 → M ⊗A N
is exact.
Proof. (i)⇒(ii) Let

· · · → P −1 → P 0 → N → 0
be a projective resolution of N . Since M is flat, the sequence
· · · → M ⊗A P −1 → M ⊗A P 0 → M ⊗A N → 0
is exact. So for any i ≥ 1, we have TorA −i ·

i (M, N ) = H (M ⊗A P ) = 0.
(ii)⇒(iii) is trivial.
(iii)⇒(iv) Let N/N 0 be the cokernel of N 0 → N . We have a short exact
sequence
0 → N 0 → N → N/N 0 → 0
and hence a long exact sequence
0 0 0
· · · → TorA
1 (M, N/N ) → M ⊗A N → M ⊗A N → M ⊗A N/N → 0.
By (iii), we have TorA 0

1 (M, N/N ) = 0. So a part of the above long exact sequence
is
0 → M ⊗A N 0 → M ⊗A N.
(iv)⇒(i) Since M ⊗A − is right exact, it suffices to show for any exact
sequence
φ
0 → N0 → N
of A-modules, the sequence
idM ⊗φ
0 → M ⊗A N 0 → M ⊗A N
Pk
is exact. Suppose i=1 mi ⊗ ni ∈ ker(idM ⊗ φ). Then in M × N , we can write
k
X k1
X
(mi , φ(ni )) = ((x1i + x2i , yi ) − (x1i , yi ) − (x2i , yi ))
i=1 i=1
k2
X
+ ((xi , y1i + y2i ) − (xi , y1i ) − (xi , y2i ))
i=1
k3
X
+ ((ai ui , vi ) − (ui , ai vi )).
i=1
Let N1 be the submodule of N generated by φ(ni ), yi , y1i , y2i and vi , let

N10 be the submodule of N 0 generated by ni , and let φ1 : N10 → N1 be the
homomorphism induced by φ. Then
φ1
0 → N10 → N1
Pk
is an exact sequence of finitely generated A-modules and i=1 mi ⊗ ni lies in
Pk
the kernel of idM ⊗ φ1 . By (iv), i=1 mi ⊗ ni vanishes in M ⊗ N10 and hence
vanishes in M ⊗ N 0 .
By Proposition 1.2 and 1.17, we can use flat resolutions of N to calculate

TorAi (M, N ).
Note that for any multiplicative set S in a ring A, the ring S −1 A is a flat A-
algebra since the functor M 7→ S −1 A ⊗A M ∼ = S −1 M is an exact functor on the
category of A-modules. If A is noetherian, then for any ideal I of A, the I-adic
completion A b of A is a flat A-algebra. Indeed, by Proposition 1.5.5, we know
b ∼
A ⊗A M = M for any finitely generated A-module M , and by Proposition 1.5.4,
c
the functor M 7→ M c is exact on the category of finitely generated A-modules.
Suppose A → B is a homomorphism of rings. If M is a flat A-module, then
B ⊗A M is a flat B-module.
Proposition 1.18. Let A be a ring and M an A-module with finite presenta-

tion. Then M is flat if and only if M is projective.
Proof. We have seen any projective A-module is flat. Suppose M is a flat A-

module with finite presentation. To show M is projective, it suffices to show M ∼
is a locally free sheaf on the scheme SpecA by Proposition 1.4.7. By Proposition
1.4.1 (ii), it suffices to show for any p ∈ SpecA, Mp is a free Ap -module. So we
are reduced to proving that for any local ring A, any flat A-module with finite
presentation is free. Let m be the maximal ideal of A. Choose x1 , . . . , xn ∈ M
such that their images in M/mM form a basis for the A/m-vector space M/mM .
Consider the homomorphism
An → M, ei 7→ xi (i = 1, . . . , n),
where for each i, ei is the element in An whose i-th component is 1 and whose
other components are 0. By Nakayama’s lemma, An → M is surjective. Let R
be its kernel. We have an exact sequence
0 → R → An → M → 0.
Since M is flat, we have TorA 1 (M, A/m) = 0. The second long exact sequence in
Proposition 1.16 of TorA i (−, A/m) associated to the above short exact sequence
gives rise to the exact sequence
0 → R ⊗A A/m → An ⊗A A/m → M ⊗A A/m → 0,

that is,
0 → R/mR → (A/m)n → M/mM → 0.
By our definition of An → M , the homomorphism (A/m)n → M/mM is an
isomorphism. So R/mR = 0. Since M has finite presentation, R is finitely gen-
erated. By Nakayama’s lemma, we have R = 0. So An → M is an isomorphism
and hence M is free.
Proposition 1.19. Let φ : A → B be a homomorphism of rings and let M be

a B-module. The following conditions are equivalent:
(i) M is flat as an A-module.
(ii) For any q ∈ SpecB, Mq is flat as an A-module.
(iii) For any q ∈ SpecB, Mq is flat as an Aφ−1 (q) -module.
Proof. (i)⇒(ii) For any A-module N , we have Mq ⊗A N = Bq ⊗B (M ⊗A N ).

Since M is flat as an A-module, M ⊗A − is an exact functor on the category
of A-modules. But Bq ⊗B − is an exact functor on the category of B-modules.
So the functor Mq ⊗A − is exact on the category of A-modules and hence Mq
is flat as an A-module.
(ii)⇒(iii) Since Mq is flat as an A-module, (Mq )φ−1 (q) is flat as an Aφ−1 (q) -
module. But (Mq )φ−1 (q) ∼ = Mq .
(iii)⇒(i) We need to show M ⊗A − is an exact functor. It suffices to show
that Bq ⊗B M ⊗A − is an exact functor for any q ∈ SpecB. For any A-module
N , we have
Bq ⊗B M ⊗A N = Mq ⊗A N = Mq ⊗Aφ−1 (q) Aφ−1 (q) ⊗A N = Mq ⊗Aφ−1 (q) Nφ−1 (q) .
Our assertion follows from the assumption that Mq is flat as an Aφ−1 (q) -module.
Proposition 1.20. Let A be a ring and M an A-module. The following

(i) A sequence
φ ψ
N 0 → N → N 00
of A-modules is exact if and only if
idM ⊗φ idM ⊗ψ
M ⊗A N 0 → M ⊗A N → M ⊗A N 00
is exact.
(ii) M is a flat A-module, and for any A-module N , the condition M ⊗A N =
0 implies that N = 0.
(iii) M is a flat A-module, and for any homomorphism φ : N → N 0 of
A-modules, the condition idM ⊗ φ = 0 implies that φ = 0.
(iv) M is a flat A-module and for any maximal ideal m of A, we have mM 6=
M.
When M satisfies one of the above conditions, we say M is a faithfully flat
A-module.
Proof. (i)⇒(ii) Suppose M ⊗A N = 0. Then the sequence 0 → M ⊗A N → 0

is exact. So 0 → N → 0 is exact. Hence N = 0.
(ii)⇒(iii) Since M is a flat A-module, we have im(idM ⊗ φ) ∼
= M ⊗A imφ. If
idM ⊗ φ = 0, then M ⊗A imφ = 0. Hence imφ = 0.
(iii)⇒(iv) Suppose there exists a maximal ideal m of A such that mM = M .
Then M ⊗A A/m ∼ = M/mM = 0. So idM ⊗ idA/m = 0. This implies idA/m = 0
and hence A/m = 0. This is impossible.
(iv)⇒(ii). Suppose N 6= 0. Let x be a nonzero element in N . Consider the
homomorphism A → N, a 7→ ax. Let a be its kernel. Note that a is contained in
some maximal ideal m of A since a 6= A. Since mM 6= M , we have aM 6= M and
hence M ⊗A A/a 6= 0. Since M is flat, the monomorphism A/a → N induces a
monomorphism M ⊗A A/a → M ⊗A N . So we have M ⊗A N 6= 0.
(iii)⇒(i) Suppose
idM ⊗φ idM ⊗ψ
M ⊗A N 0 → M ⊗A N → M ⊗A N 00
is exact. Then idM ⊗ ψφ = 0. So ψφ = 0. Since M is flat, we have
M ⊗A (kerψ/imφ) ∼
= ker(idM ⊗ ψ)/im(idM ⊗ φ) = 0.
So kerψ/imφ = 0.
Proposition 1.21. Let φ : A → B be a homomorphism of rings. The following

(i) B is a faithfully flat A-algebra.
(ii) B is a flat A-algebra and for any A-module M , the homomorphism
M → B ⊗A M, x 7→ 1 ⊗ x
is injective. (In particular, A → B is injective.)

(iii) B is a flat A-algebra and for any ideal a of A, we have φ−1 (aB) = a.
(iv) B is a flat A-algebra and for any maximal ideal m of A, there exists a
maximal ideal n of B such that φ−1 (n) = m.
(v) B is a flat A-algebra and for any prime ideal p of A, there exists a prime
ideal q of B such that φ−1 (q) = p.
Proof. (i)⇒(ii) It suffices to show B ⊗A M → B ⊗A (B ⊗A M ) is injective.

Indeed it has a right inverse
B ⊗A (B ⊗A M ) → B ⊗A M, b1 ⊗ (b2 ⊗ x) 7→ b1 b2 ⊗ x.
(ii)⇒(iii) Applying (ii) to M = A/a, we see that φ induces a monomorphism

A/a → B/aB. This implies that a = φ−1 (aB).
(iii)⇒(iv) Note that mB 6= B. Otherwise φ−1 (mB) = A 6= m. So there
exists a maximal ideal n of B such that mB ⊂ n. We then have m ⊂ φ−1 (n).
Since m is a maximal ideal, we must have m = φ−1 (n).
(iv)⇒(i) Follows from Proposition 1.20 (iv).
2.2. SPECTRAL SEQUENCES 127
(i)⇒(v) It is easy to show that Bp is a faithfully flat Ap -algebra. By the

equivalence of (i) and (iv) that we have already proved, there exists a maximal
ideal n of Bp such that φ−1 p (n) = pAp , where φp : Ap → Bp is the homo-
morphism induced by φ. Let q be the inverse image of n under the canonical
homomorphism B → Bp . Then q is a prime ideal of B and φ−1 (q) = p.
Corollary 1.22. Let A → B be a local homomorphism of local rings. If B is

a flat A-algebra, then it is faithfully flat.
Proof. This follows from Proposition 1.21 (iv).
Corollary 1.23. Let A be a noetherian ring, I an ideal of A, and A b the I-adic

b is a faithfully flat A-algebra if and only if I ⊂ rad(A).
completion of A. Then A
Proof. We have seen A b is a flat A-algebra. Let φ : A → A b be the canonical

homomorphism. Suppose A b is faithfully flat. Then for any maximal ideal m of
A, there exists a maximal ideal n of A b such that m = φ−1 (n) by Proposition
1.21 (iv). By Proposition 1.5.7, we have Ib ⊂ n. So
I ⊂ φ−1 (I)
b ⊂ φ−1 (n) = m.
Hence I is contained in every maximal ideal of A. Therefore I ⊂ rad(A).

Conversely, suppose I ⊂ rad(A). Let m be a maximal ideal of A. Then
m/I is a maximal ideal of A/I. There exists a maximal ideal n of Ab such that
m/I is identified with n/Ib through the isomorphism A/I ∼
= A/
b I.
b We then have
−1
m = φ (n). Hence A is faithfully flat by Proposition 1.21 (iv).
b
2.2 Spectral Sequences
A spectral sequence E in an abelian category C consists of the following data:

(a) A family of objects Erpq in C for any p, q, r ∈ Z and r ≥ 2.
(b) A family of morphisms
dpq pq p+r,q−r+1
r : Er → E r
such that dp+r,q−r+1

r dpq
r = 0. Let Zr+1 (Erpq ) = kerdpq pq
r and Br+1 (Er ) =
p−r,q+r−1 pq pq pq
imdr . We have Br+1 (Er ) ⊂ Zr+1 (Er ) ⊂ Er .
(c) A family of isomorphisms
αrpq : Zr+1 (Erpq )/Br+1 (Erpq ) ∼ pq

= Er+1 .
For any k ≥ r+1, define by induction on k−(r+1) some sub-objects Bk (Erpq )

and Zk (Erpq ) of Erpq as follows: When k − (r + 1) = 0, Bk (Erpq ) and Zk (Erpq )
have been defined in (b). Suppose we have defined Bi (Erpq ) and Zi (Erpq ) for
pq
any 0 ≤ i − (r + 1) < k − (r + 1). In particular, the sub-objects Bk (Er+1 ) and
pq pq
Zk (Er+1 ) of Er+1 have been defined. Through the isomorphisms
αrpq : Zr+1 (Erpq )/Br+1 (Erpq ) ∼ pq

= Er+1 ,
pq pq pq
the sub-objects Bk (Er+1 ) and Zk (Er+1 ) of Er+1 are identified with some sub-
objects of Zr+1 (Er )/Br+1 (Er ). We define Bk (Erpq ) and Zk (Erpq ) as the inverse
pq pq
images of these sub-objects under the canonical epimorphism Zr+1 (Erpq ) →

Zr+1 (Erpq )/Br+1 (Erpq ). Obviously we have Bk (Erpq ) ⊂ Zk (Erpq ) and
Zk (Erpq )/Bk (Erpq ) ∼

= Zk (Er+1 )/Bk (Er+1 ) ∼
pq pq
= ··· ∼
= Zk (Ek−1 )/Bk (Ek−1 ) ∼
pq pq pq
= Ek .
If we set Br (Erpq ) = 0 and Zr (Erpq ) = Erpq , then we have
0 = Br (Erpq ) ⊂ Br+1 (Erpq ) ⊂ Br+2 (Erpq ) ⊂ · · · ,

· · · ⊂ Zr+2 (Erpq ) ⊂ Zr+1 (Erpq ) ⊂ Zr (Erpq ) = Erpq ,
and any object on the first line is a sub-object of any object on the second line.
The other data of the spectral sequence E are the following:
(d) Two sub-objects B∞ (E2pq ) and Z∞ (E2pq ) of E2pq such that
Bk (E2pq ) ⊂ B∞ (E2pq ) ⊂ Z∞ (E2pq ) ⊂ Zk (E2pq )
for any k ≥ 2. Let E∞ pq

= Z∞ (E2pq )/B∞ (E2pq ).
(e) A family of objects H n (n ∈ Z) in C and each H n is provided with a
decreasing filtration
· · · ⊃ F p H n ⊃ F p+1 H n ⊃ · · · .
(f) A family of isomorphisms

pq ∼
β pq : E∞ = F p H p+q /F p+1 H p+q .
This finishes the definition of the spectral sequence E. We often denote it

by
E2pq ⇒ H p+q .
Let E and E 0 be two spectral sequences. A morphism of spectral sequences

u : E → E 0 consists of morphisms upq pq 0pq
r : E r → Er and un : H n → H 0n
satisfying the following conditions:
(i) un (F p H n ) ⊂ F p H 0n and the diagrams
dpq
Erpq →r
Erp+r,q−r+1
upq
r ↓ ↓ urp+r,q−r+1
d0pq
Er0pq →r
Er0p+r,q−r+1
commute. So un induce morphisms
ūn : F p H n /F p+1 H n → F p H 0n /F p+1 H 0n

and upq
r induce morphisms
0pq 0pq
ūpq pq pq
r : Zr+1 (Er )/Br+1 (Er ) → Zr+1 (Er )/Br+1 (Er ).
(ii) The diagrams

αpq
r
Zr+1 (Erpq )/Br+1 (Erpq ) ∼
=
pq
Er+1
ūpq
r ↓ ↓ upq
r+1
α0pq
r
Zr+1 (Er0pq )/Br+1 (Er0pq ) ∼
=
0pq
Er+1
0pq 0pq
commute. So upq pq pq pq
r (Bk (Er )) ⊂ Bk (Er ) and ur (Zk (Er )) ⊂ Zk (Er ) for any
k, r ≥ 2.
0pq 0pq
(iii) upq pq pq pq
2 (B∞ (E2 )) ⊂ B∞ (E2 ) and u2 (Z∞ (E2 )) ⊂ Z∞ (E2 ). So u2
pq
pq pq 0pq
induce morphisms u∞ : E∞ → E∞ .
(iv) The diagrams
pq β pq
E∞ → F p H p+q /F p+1 H p+q
pq
u∞ ↓ ūp+q ↓
0pq β 0pq
E∞ → F p H 0p+q /F p+1 H 0p+q
commute.
We define the composition of morphisms of spectral sequences in the obvious
way. An isomorphism of spectral sequences is a morphism with a two-sided
inverse.
A spectral sequence E is called biregular if it has the following properties:

(i) For any pair (p, q), we have B∞ (E2pq ) = Bk (E2pq ) and Z∞ (E2pq ) = Zk (E2pq )
for sufficiently large k.
0
(ii) For each n, we have F p H n = 0 for some p and F p H n = H n for some
p0 .
Suppose in the abelian category C, the direct sum of any family of objects
exists. Then for any object A in C and any family Ai (i ∈ I) of sub-objects of
A, we define ∪i∈I Ai to be the image of the canonical morphism ⊕i∈I Ai → A.
A spectral sequence E in C is called regular if it has the following properties:
(i) For any pair (p, q), we have B∞ (E2pq ) = ∪2≤k<∞ Bk (E2pq ) and Z∞ (E2pq ) =
Zk (E2pq ) for sufficiently large k.
(ii) For each n, we have F p H n = 0 for some p and H n = ∪p F p H n .
In the following, we only state results for biregular spectral sequences. With
slight modifications, they also hold for regular spectral sequences.
Proposition 2.1. Let u : E → E 0 be a morphism of biregular spectral se-

quences. If upq
2 is an isomorphism for any p, q ∈ Z, then u is an isomorphism.
Proof. Since upq pq

2 is an isomorphism for any p, q ∈ Z, ur is an isomor-
0pq
phism for any p, q, r ∈ Z, r ≥ 2. So we have upq
2 (B k (E pq
2 )) = Bk (E2 )
0pq
and upq pq 0
2 (Zk (E2 )) = Zk (E2 ) for any k ≥ 2. Since E and E are biregu-
lar, this implies u2 (B∞ (E2 )) = B∞ (E2 ) and u2 (Z∞ (E2 )) = Z∞ (E20pq ).
pq pq 0pq pq pq
So upq
∞ is an isomorphism for any p, q ∈ Z. Hence ū
p,n−p
: F p H n /F p+1 H n →
p 0n p+1 0n
F H /F H is an isomorphism for any p, n ∈ Z. Let’s prove un : H n → H 0n
is an isomorphism for any n ∈ Z. Since E and E 0 are biregular, it suffices to
show un induces an isomorphism F p H n → F p H 0n for every p. We prove this by
descending induction on p. For p sufficiently large, we have F p H n = F p H 0p = 0
and F p H n → F p H 0n is trivially an isomorphism. Suppose we have shown that u
induces an isomorphism F p H n → F p H 0n . Consider the following commutative
diagram:
0 → F pH n → F p−1 H n → F p−1 H n /F p H n → 0
↓ ↓ ↓
0 → F p H 0n → F p−1 H 0n → F p−1 H 0n /F p H 0n → 0.
By the induction hypothesis, the first vertical arrow is an isomorphism. We

have seen the third vertical arrow is an isomorphism. So the second vertical
arrow is also an isomorphism. This proves our assertion.
Proposition 2.2. Let E be a biregular spectral sequence. Suppose there exist

an r ≥ 2 and a q0 ∈ Z (resp. p0 ∈ Z) such that Erpq = 0 whenever q 6= q0 (resp.
p 6= p0 ). (Such spectral sequence is called degenerate). Then Ern−q0 ,q0 ∼
= H n for
p0 ,n−p0 ∼ n
any n ∈ Z (resp. Er = H ).
Proof. Suppose Erpq = 0 whenever q 6= q0 . Then Ekpq = 0 whenever q 6= q0 and

k ≥ r. So dpq
k = 0 for any p, q ∈ Z and k ≥ r. This implies
Br (E2pq ) = Br+1 (E2pq ) = · · · ,

· · · = Zr+1 (E2pq ) = Zr (E2pq ).
Since E is biregular, we have
Br (E2pq ) = Br+1 (E2pq ) = · · · = B∞ (E2pq ),

Z∞ (E2pq ) = · · · = Zr+1 (E2pq ) = Zr (E2pq ).
So
pq
E∞ = Z∞ (E2pq )/B∞ (E2pq ) ∼
= Zr (E2 )/Br (E2 ) ∼
pq pq
= Erpq .
We have
F p H n /F p+1 H n ∼ p,n−p
= E∞ .
So F p H n /F p+1 H n = 0 whenever p 6= n − q0 and hence
· · · = F n−q0 −1 H n = F n−q0 H n ⊃ F n−q0 +1 H n = F n−q0 +2 H n = · · · .
Since E is regular, this implies that F n−q0 H n = H n and F n−q0 +1 H n = 0. So
Hn ∼
= F n−q0 H n /F n−q0 +1 H n ∼ n−q0 ,q0 ∼
= E∞ = Ern−q0 ,q0 .
Proposition 2.3. Let E be a biregular spectral sequence. Suppose E2pq = 0

when p < 0, or q < 0, or 0 < q < n. Then E2i,0 ∼
= H i for any i < n and we have
an exact sequence
0 → E2n,0 → H n → E20,n → E2n+1,0 → H n+1 .
In particular, if E2pq = 0 when p < 0 or q < 0, then E20,0 ∼

= H 0 and we have an
exact sequence
0 → E21,0 → H 1 → E20,1 → E22,0 → H 2 .
Proof. When p < 0, or q < 0, or 0 < q < n, we have Erpq = 0 for any
2 ≤ r ≤ ∞. Suppose i < n. Then we have F p H i /F p+1 H i = E∞
p,i−p
= 0 when
p 6= i and hence
· · · = F i−1 H i = F i H i ⊃ F i+1 H i = F i+1 H i = · · · .
Since E is biregular, this implies F i H i = H i and F i+1 H i = 0 and hence

Hi ∼
= E∞i,0
. Consider the morphisms
dri−r,r−1 di,0
Eri−r,r−1 → Eri,0 →
r
Eri+r,−r+1 .
We have Eri+r,−r+1 = 0 since −r + 1 < 0. We have Eri−r,r−1 = 0 since either

i − r < 0 or 0 < r − 1 < n. So dri−r,r−1 and di,0
r vanish. Hence
B2 (E2i,0 ) = B3 (E2i,0 ) = · · · ,
· · · = Z3 (E2i,0 ) = Z2 (E2i,0 ).
Since E is biregular, this implies that
B2 (E2i,0 ) = B3 (E2i,0 ) = · · · = B∞ (E2i,0 ),

Z∞ (E2i,0 ) = · · · = Z3 (E2i,0 ) = Z2 (E2i,0 ).
i,0 ∼
= E2 and hence H i ∼
i,0 i,0
So E∞ = E2 for any i < n.
p n p+1 n p,n−p
We have F H /F H = E∞ = 0 when p 6= 0, n. So we have
· · · = F −1 H n = F 0 H n ⊃ F −1 H n = · · · = F n H n ⊃ F n+1 H n = · · · .
Since E is biregular, this implies that H n = F 0 H n and F n+1 H n = 0. So the

exact sequence
0 → F n H n → H n → H n /F n H n → 0
is identified with
n,0
0 → E∞ → H n → E∞
0,n
→ 0. (1)
We have F p H n+1 /F p+1 H n+1 = E∞
p,n+1−p
= 0 when p ≥ n + 2. So
F n+2 H n+1 = F n+3 H n+1 = · · · .

Since E is biregular, we must have F n+2 H n+1 = 0. So the exact sequence
0 → F n+1 H n+1 → H n+1
can be identified with

n+1,0
0 → E∞ → H n+1 . (2)
Consider the morphisms
dn−r,r−1 dn,0
Ern−r,r−1 r
→ Ern,0 →
r
Ern+r,−r+1 .
We have Ern−r,r−1 = 0 since either n − r < 0 or 0 < r − 1 < n. We have

Ern+r,−r+1 = 0 since −r + 1 < 0. So drn−r,r−1 and dn,0
r vanish. Combining with
the fact that E is biregular, this implies that
B2 (E2n,0 ) = B3 (E2n,0 ) = · · · = B∞ (E2n,0 ),

Z∞ (E2n,0 ) = · · · = Z3 (E2n,0 ) = Z2 (E2n,0 ).
So we have
n,0
E∞ = E2n,0 . (3)
d−r,n+r−1 d0,n
Er−r,n+r−1 r
→ Er0,n →
r
Err,n−r+1 .
We have Er−r,n+r−1 = 0 since −r < 0. When r 6= n + 1, we have Err,n−r+1 = 0

since either n − r + 1 < 0 or 0 < n − r + 1 < n. So we have dr−r,n+r−1 = 0, and
when r 6= n + 1, we have d0,n
r = 0. Combining with the fact that E is biregular,
we get
B2 (E20,n ) = · · · = B∞ (E20,n ),
Z∞ (E20,n ) = · · · = Zn+2 (E20,n ) ⊂ Zn+1 (E20,n ) = · · · = Z2 (E20,n ).
So we have
0,n
E∞ = Zn+2 (E20,n )/Bn+1 (E20,n ), (4)
E20,n = 0,n
En+1 . (5)
Consider the morphism
drn+1−r,r−1 dn+1,0
Ern+1−r,r−1 → Ern+1,0 r→ Ern+1+r,−r+1 .
We have Ern+1+r,−r+1 = 0 since −r + 1 < 0. When r 6= n + 1, we have

Ern+1−r,r−1 = 0 since either n + 1 − r < 0 or 0 < r − 1 < n. So we have
dn+1,0
r = 0, and when r 6= n + 1, we have drn+1−r,r−1 = 0. Combining with the
fact that E is biregular, we get
B2 (E2n+1,0 ) = · · · = Bn+1 (E2n+1,0 ) ⊂ Bn+2 (E2n+1,0 ) = · · · = B∞ (E2n+1,0 ),

Z∞ (E2n+1,0 ) = · · · = Z2 (E2n+1,0 ).
So we have
n+1,0
E∞ = Zn+1 (E2n+1,0 )/Bn+2 (E2n+1,0 ), (6)
E2n+1,0 = n+1,0
En+1 . (7)
Through the identifications
0,n
En+1 = Zn+1 (E20,n )/Bn+1 (E20,n ),
n+1,0
En+1 = Zn+1 (E2n+1,0 )/Bn+1 (E2n+1,0 ),
the kernel of the morphism
d0,n 0,n n+1,0

n+1 : En+1 → En+1
is identified with Zn+2 (E20,n )/Bn+1 (E20,n ) and the cokernel is identified with
Zn+1 (E2n+1,0 )/Bn+2 (E2n+1,0 ). Combining with the equations (4), (5), (6), (7),
we get a morphism
E20,n → E2n+1,0
induced by d0,n 0,n
n+1 whose kernel is isomorphic to E∞ and whose cokernel is iso-
n+1,0
morphic to E∞ . Combining with the exact sequences (1), (2) and the equa-
tion (3), we get an exact sequence
0 → E2n,0 → H n → E20,n → E2n+1,0 → H n+1 .

for any p 6= p1 , p2 (p1 < p2 ). If p2 − p1 = 1, then we have short exact sequences
0 → E2p2 ,n−p2 → H n → E2p1 ,n−p1 → 0.
If p2 − p1 ≥ 2, then we have a long exact sequence
· · · → E2p2 ,n−p2 → H n → E2p1 ,n−p1 → E2p2 ,n+1−p2 → H n+1 → · · · .
Proof. We have Erpq = 0 for any 2 ≤ r ≤ ∞ and any p 6= p1 , p2 . So

F p H n /F p H n = 0 for any p 6= p1 , p2 and hence
· · · = F p1 H n ⊃ F p1 +1 H n = · · · = F p2 H n ⊃ F p2 +1 H n = · · · .
Since E is biregular, this implies F p1 H n = H n and F p2 +1 H n = 0. The exact

sequence
0 → F p2 H n → H n → H n /F p2 H n → 0
is identified with
p2 ,n−p2
0 → E∞ → H n → E∞
p1 ,n−p1
→ 0. (1)
Suppose p2 − p1 = 1. Consider the morphisms

dp 1 −r,n−p1 +r−1 dp 1 ,n−p1
Erp1 −r,n−p1 +r−1 r
→ Erp1 ,n−p1 r
→ Erp1 +r,n−p1 −r+1
and the morphisms

drp2 −r,n−p2 +r−1 dp 2 ,n−p2
Erp2 −r,n−p2 +r−1 → Erp2 ,n−p2 r
→ Erp2 +r,n−p2 −r+1 .
Note that Erp1 −r,n−p1 +r−1 , Erp1 +r,n−p1 −r+1 , Erp2 −r,n−p2 +r−1 and Erp2 +r,n−p2 −r+1
all vanish since p1 − r, p1 + r, p2 − r and p2 + r are different from p1 and p2 .
So dpr 1 −r,n−p1 +r−1 , drp1 ,n−p1 , drp2 −r,n−p2 +r−1 and drp2 ,n−p2 all vanish. Combin-
ing with the fact that E is biregular, this implies E∞ p2 ,n−p2
= E2p2 ,n−p2 and
p1 ,n−p1 p1 ,n−p1
E∞ = E2 . So the exact sequence (1) is identified with
0 → E2p2 ,n−p2 → H n → E2p1 ,n−p1 → 0.
Suppose p2 − p1 ≥ 2. Consider the morphisms
dp 1 −r,n−p1 +r−1 dp 1 ,n−p1
Erp1 −r,n−p1 +r−1 r
→ Erp1 ,n−p1 r
→ Erp1 +r,n−p1 −r+1 .
We have Erp1 −r,n−p1 +r−1 = 0, and when r 6= p2 − p1 , we have Erp1 +r,n−p1 −r+1 =
0. Combining with the fact that E is biregular, we get
B2 (E2p1 ,n−p1 ) = · · · = B∞ (E2p1 ,n−p1 ),
Z∞ (E2p1 ,n−p1 ) = · · · = Zp2 −p1 +1 (E2p1 ,n−p1 ) ⊂
⊂ Zp2 −p1 (E2p1 ,n−p1 ) = · · · = Z2 (E2p1 ,n−p1 ).
So we have
p1 ,n−p1
E∞ = Zp2 −p1 +1 (E2p1 ,n−p1 )/Bp2 −p1 (E2p1 ,n−p1 ), (2)
E2p1 ,n−p1 = Epp21−p
,n−p1
1
. (3)
drp2 −r,n−p2 +r dp 2 ,n+1−p2
Erp2 −r,n−p2 +r → Erp2 ,n+1−p2 r
→ Erp2 +r,n+2−p2 −r .
We have Erp2 +r,n+2−p2 −r = 0, and when r 6= p2 − p1 , we have Erp2 −r,n−p2 +r = 0.
Combining with the fact that E is biregular, we get
B2 (E2p2 ,n+1−p2 ) = · · · = Bp2 −p1 (E2p2 ,n+1−p2 ) ⊂
⊂ Bp2 −p1 +1 (E2p2 ,n+1−p2 ) = · · · = B∞ (E2p2 ,n+1−p2 ),
Z∞ (E2p2 ,n+1−p2 ) = · · · = Z2 (E2p2 ,n+1−p2 ).
So we have
p2 ,n+1−p2
E∞ = Zp2 −p1 (E2p2 ,n+1−p2 )/Bp2 −p1 +1 (E2p2 ,n+1−p2 ), (4)
E2p2 ,n+1−p2 = Epp22−p

,n+1−p2
1
. (5)
Through the identifications
Epp21−p
,n−p1
1
= Zp2 −p1 (E2p1 ,n−p1 )/Bp2 −p1 (E2p1 ,n−p1 ),
Epp22−p
,n+1−p2
1
= Zp2 −p1 (E2p2 ,n+1−p2 )/Bp2 −p1 (E2p2 ,n+1−p2 ),
the kernel of the morphism

dpp12 −p
,n−p1
1
: Epp21−p
,n−p1
1
→ Epp22−p
,n+1−p2
1
is identified with Zp2 −p1 +1 (E2p1 ,n−p1 )/Bp2 −p1 (E2p1 ,n−p1 ) and the cokernel is iden-
tified with Zp2 −p1 (E2p2 ,n+1−p2 )/Bp2 −p1 +1 (E2p2 ,n+1−p2 ). Combining with the equa-
tions (2), (3), (4), (5), we get a morphism
E2p1 ,n−p1 → E2p2 ,n+1−p2
induced by dpp12 ,n−p
−p1
1 p1 ,n−p1
whose kernel is identified with E∞ and whose cokernel
p2 ,n+1−p2
is identified with E∞ . Combining with the exact sequence (1), we get a
long exact sequence
· · · → E2p2 ,n−p2 → H n → E2p1 ,n−p1 → E2p2 ,n+1−p2 → H n+1 → · · · .

for any q 6= q1 , q2 (q1 < q2 ). Then we have a long exact sequence
· · · → E2n−q1 ,q1 → H n → E2n−q2 ,q2 → E2n+1−q1 ,q1 → H n+1 → · · · .
The following lemma is obvious:
Lemma 2.6. Suppose we have a commutative diagram

D
% ↓γ
α β
A → B → C.
such that the bottom row is exact. Then β induces an isomorphism
imγ/imα → imβγ.
Let K · be a complex in an abelian category C endowed with a decreasing

filtration
· · · ⊃ F p K · ⊃ F p+1 K · ⊃ · · · .
We are going to construct a spectral sequence from this filtered complex.
For any p, q, r ∈ Z and r ≥ 2, let
Zrpq = im(H p+q (F p K · /F p+r K · ) → H p+q (F p K · /F p+1 K · )),
where the morphism H p+q (F p K · /F p+r K · ) → H p+q (F p K · /F p+1 K · ) is the mor-
phism in the long exact sequence of cohomology objects associated to the short
exact sequence
0 → F p+1 K · /F p+r K · → F p K · /F p+r K · → F p K · /F p+1 K · → 0.
Let
Brpq = im(H p+q−1 (F p−r+1 K · /F p K · ) → H p+q (F p K · /F p+1 K · )),
where the morphism H p+q−1 (F p−r+1 K · /F p K · ) → H p+q (F p K · /F p+1 K · ) is the

morphism in the long exact sequence of cohomology objects associated to the
0 → F p K · /F p+1 K · → F p−r+1 K · /F p+1 K · → F p−r+1 K · /F p K · → 0.

pq pq
We define Z∞ and B∞ using the same formulas by making the convention that
∞ · −∞ ·
F K = 0 and F K = K · . So we have
pq
Z∞ = im(H p+q (F p K · ) → H p+q (F p K · /F p+1 K · )),
pq
B∞ = im(H p+q−1 (K · /F p K · ) → H p+q (F p K · /F p+1 K · )).
Obviously, we have
B2pq ⊂ B3pq ⊂ · · · ⊂ B∞
pq
,
pq pq pq
Z∞ ⊂ · · · ⊂ Z3 ⊂ Z2 .
Since we have a commutative diagram
0 → F pK · → K· → K · /F p K · → 0
↓ ↓ k
0 → F p K · /F p+1 K · → K · /F p+1 K · → K · /F p K · → 0,
H p+q−1 (K · /F p K · ) → H p+q (F p K · )
k ↓
H p+q−1 (K · /F p K · ) → H p+q (F p K · /F p+1 K · ).
So we have
pq pq
B∞ ⊂ Z∞ .
Define
Erpq = Zrpq /Brpq (r ≥ 2), E∞
pq pq
= Z∞ pq
/B∞ .
Applying Lemma 2.6 to the commutative diagram
Hp+q(F p K ·/F p+r K ·)

% ↓
· ·
H p+q p
(F K /F p+r+1
K )→ Hp+q(F p K ·/F p+1 K ·) → Hp+q+1(F p+1 K ·/F p+r+1 K ·),
pq
we see that Zrpq /Zr+1 is isomorphic to the image of the composition
H p+q(F p K · /F p+r K · ) → H p+q(F p K · /F p+1 K · ) → H p+q+1(F p+1 K · /F p+r+1 K · ).

Applying Lemma 2.6 to the commutative diagram
H p+q (F p K · /F p+r K · )
% ↓
· ·
Hp+q
(F p+1
K /F p+r
K )→ Hp+q+1(Fp+rK ·/Fp+r+1K ·) → Hp+q+1(Fp+1K ·/Fp+r+1K ·),
p+r,q−r+1
we see that Br+1 /Brp+r,q−r+1 is isomorphic to the image of the composi-
tion
H p+q(F p K ·/F p+r K ·)→H p+q+1(F p+r K ·/F p+r+1 K ·)→H p+q+1(F p+1 K ·/F p+r+1 K ·).
H p+q (F p K · /F p+r K · ) → H p+q+1 (F p+r K · /F p+r+1 K · )

↓ ↓
H p+q (F p K · /F p+1 K · ) → H p+q+1 (F p+1 K · /F p+r+1 K · ),
we see that
pq ∼ p+r,q−r+1
Zrpq /Zr+1 = Br+1 /Brp+r,q−r+1 .
Define
dpq pq p+r,q−r+1
r : Er → Er
pq ∼ p+r,q−r+1
Zrpq /Brpq → Zrpq /Zr+1 = Br+1 /Brp+r,q−r+1 → Zrp+r,q−r+1 /Brp+r,q−r+1 .
Obviously, we have
pq
kerdpq
r = Zr+1 /Brpq ,
p+r,q−r+1
imdpq
r = Br+1 /Brp+r,q−r+1 .
This implies that

drp+r,q−r+1 dpq
r =0
and
p−r,q+r−1 ∼ pq pq pq
kerdpq
r /imdr = Zr+1 /Br+1 = Er+1 .
For any n ∈ Z, define H n = H n (K · ) and define a decreasing filtration on H n
by
F p H n = im(H n (F p K · ) → H n (K · )).
Applying Lemma 2.6 to the diagram
H p+q (F p K · )
% ↓
H p+q (F p+1 K · ) → H p+q (K · ) → H p+q (K · /F p+1 K · ),
we see that F p H p+q /F p+1 H p+q is isomorphism to the image of the composition
H p+q (F p K · ) → H p+q (K · ) → H p+q (K · /F p+1 K · ).

Applying Lemma 2.6 to the diagram
H p+q (F p K · )
% ↓
H p+q−1 (K · /F p K · ) → H p+q (F p K · /F p+1 K · ) → H p+q (K · /F p+1 K · ),
pq pq
we see that Z∞ /B∞ is isomorphic to the image of the composition
H p+q (F p K · ) → H p+q (F p K · /F p+1 K · ) → H p+q (K · /F p+1 K · ).
H p+q (F p K · ) → H p+q (K · )
↓ ↓
H p+q (F p K · /F p+1 K · ) → H p+q (K · /F p+1 K · ),
we see that
F p H p+q /F p+1 H p+q ∼ pq
= Z∞ pq
/B∞ .
Finally we define B∞ (E2pq ) and Z∞ (E2pq ) to be the sub-objects B∞
pq
/B2pq and
pq pq pq
Z∞ /B2 of E2 . This finishes the construction of the spectral sequence.
Note that if we define
E1pq = H p+q (F p K · /F p+1 K · )
and define dpq pq p+1,q

1 : E1 → E1 to be the morphism
H p+q (F p K · /F p+1 K · ) → H p+q+1 (F p+1 K · /F p+2 K · )
in the long exact sequence of cohomology groups associated to the short exact
sequence
0 → F p+1 K · /F p+2 K · → F p K · /F p+2 K · → F p K · /F p+1 K · → 0,
then we have
E2pq = kerdpq p−1,q
1 /imd1 .
We often regard (E1pq , dpq
1 ) as part of the above spectral sequence and denote
the spectral sequence by
E1pq = H p+q (F p K · /F p+1 K · ) ⇒ H p+q (K · ).
Proposition 2.7. Let K · be a complex provided with a decreasing filtration
· · · ⊃ F p K · ⊃ F p+1 K · ⊃ · · · .
Suppose for each n, there exist integers f (n) and g(n) such that F p K n = K n
for any p ≤ f (n) and F p K n = 0 for any p ≥ g(n). Then the spectral sequence
constructed above is biregular.
Proof. For each fixed p and q, when r > max(g(p+q−1), g(p+q), g(p+q+1))−p,
we have F p+r K n = 0 for n = p + q − 1, p + q, p + q + 1 and hence
H p+q (F p K · /F p+r K · ) = H p+q (F p K · ).
So when r > max(g(p+q−1), g(p+q), g(p+q+1))−p, we have Zrpq = Z∞ pq

. When
p−r+1 n
r > p + 1 − min(f (p + q − 2), f (p + q − 1), f (p + q)), we have F K = Kn
for n = p + q − 2, p + q − 1, p + q and hence
H p+q−1 (F p−r+1 K · /F p K · ) = H p+q−1 (K · /F p K · ).
So when r > p + 1 − min(f (p + q − 2), f (p + q − 1), f (p + q)), we have Brpq = B∞ pq

.
p n n p ·
For each fixed n, we have F K = 0 for p > g(n) and hence H (F K ) = 0. So
when p > g(n), we have F p H n = 0. When p < min(f (n − 1), f (n), f (n + 1)),
we have F p K i = K i for i = n − 1, n, n + 1 and hence H n (F p K · ) = H n (K · ). So
when p < min(f (n − 1), f (n), f (n + 1)), we have F p H n = H n .
A bicomplex K ·· in an abelian category C is a family of objects K ij (i, j ∈ Z)

together with morphisms dij1 :K
ij
→ K i+1,j and dij
2 :K
ij
→ K i,j+1 such that
di+1,j
1 dij i,j+1 ij
1 = 0, d2 d2 = 0, di,j+1
1 dij i+1,j ij
2 = d2 d1 .
Suppose either the direct sum of any family of objects in C exists, or for any n,
there are only finitely many pairs (i, j) satisfying i + j = n and K ij 6= 0. Then
the (simple) complex sK ·· associated to K ·· is defined by
M
(sK ·· )n = K ij
i+j=n
and d : (sK ·· )n → (sK ·· )n+1 is given by
dx = d1 x + (−1)i d2 x
for any x ∈ K ij . One can verify that dd = 0. We define the n-th cohomology
object H n (K ·· ) of K ·· to be the n-th cohomology object of sK ·· . For each i,
d
· · · → K ij →2 K i,j+1 → · · ·
is a complex. We denote its q-th cocycle, coboundary, and cohomology objects

q q q
by ZII (K i,· ), BII (K i,· ), and HII (K i,· ), respectively. Note that
q d q
··· → ZII (K i,· ) →1 ZII (K i+1,· ) → · · ·
q d q
··· → BII (K i,· ) →1 BII (K i+1,· ) → · · ·
q d q
··· → HII (K i,· ) →1 HII (K i+1,· ) → · · ·
q
are complexes. We denote the p-th cohomology object of the complex HII (K ·· )
p q q q q
by HI HII (K ·· ). Similarly, we have objects ZI (K ·,j ), BI (K ·,j ), HI (K ·,j ), and
p
HII HIq (K ·· ).
Define two decreasing filtrations on sK ·· as follows:

M M
FIp (sK ·· )n = K ij , FIIp (sK ·· )n = K ij .
i+j=n,i≥p i+j=n,j≥p
These two filtrations define two spectral sequences EI and EII . For the first
filtration, we have EI pq
1 =H
p+q
(FIp (sK ·· )/FIp+1 (sK ·· )). Obviously we have
(FIp (sK ·· )/FIp+1 (sK ·· ))n = K p,n−p ,
H p+q (FIp (sK ·· )/FIp+1 (sK ·· )) q
= HII (K p,· ).
One can then verify
EI pq p q ··
2 = HI HII (K ).
So the spectral sequence defined by the first filtration is
EI pq p q ··
2 = HI HII (K ) ⇒ H
p+q
(K ·· ).
Similarly the spectral sequence defined by the second filtration is
EII pq p q ··
2 = HII HI (K ) ⇒ H
p+q
(K ·· ).
The following proposition follows directly from Proposition 2.7:
Proposition 2.8. Let K ·· be a double complex. Suppose one of the following

conditions holds:
(i) There exist p0 , q0 such that K pq 6= 0 only when p ≥ p0 and q ≥ q0 .
(ii) There exist p0 , q0 such that K pq 6= 0 only when p ≤ p0 and q ≤ q0 .
(iii) There exist p1 ≤ p2 such that K pq 6= 0 only when p1 ≤ p ≤ p2 .
(iv) There exist q1 ≤ q2 such that K pq 6= 0 only when q1 ≤ q ≤ q2 .
Then the two spectral sequences of the double complex K ·· are biregular.
Proposition 2.9. Let K · be a complex in an abelian category with enough

injective objects. There exists a double complex I ·· such that for each q, I ·,q ,
q q q
ZII (I ·· ), BII (I ·· ) and HII (I ·· ) are injective resolutions of K q , Z q (K · ), B q (K · )
q ·
and H (K ), respectively, and for each i, the following short exact sequences
are split:
q q q
0 → BII (I i,· ) → ZII (I i,· ) → HII (I i,· ) → 0,
q q+1
0 → ZII (I i,· ) → I i,q → BII (I i,· ) → 0.
We call I ·· a Cartan-Eilenberg resolution of K · .
.. ..
. .
↑ ↑
··· → I 1,q → I 1,q+1 → ···
↑ ↑
· · · → I 0,q → I 0,q+1 → ···
↑ ↑
· · · → Kq → K q+1 → ···
↑ ↑
0 0
· q ·
Proof. For each q, choose an injective resolution IB q (K · ) of B (K ) and an
· q ·
injective resolution IH q (K · ) of H (K ). We have an exact sequence
0 → B q (K · ) → Z q (K · ) → H q (K · ) → 0.
By Proposition 1.1 (iv), there exists an injective resolution IZ· q (K · ) together with
a short exact sequence of complexes
· · ·
0 → IB q (K · ) → IZ q (K · ) → IH q (K · ) → 0
such that IZi q (K · ) = IB

i i
q (K · ) ⊕ IH q (K · ) for each i. We have an exact sequence
0 → Z q (K · ) → K q → B q+1 (K · ) → 0.
By Proposition 1.1 (iv) again, there exists an injective resolution I ·,q of K q

together with a short exact sequence of complexes
0 → IZ· q (K · ) → I ·,q → IB
·
q+1 (K · ) → 0
··
such that I i,q = IZi q (K · ) ⊕ IB
i
q+1 (K · ) for each i. Then the double complex I has
the required properties.
Proposition 2.10. Let F : C → C 0 and G : C 0 → C 00 be two left exact covariant

functors between abelian categories. Suppose C and C 0 have enough injective
objects, and for any injective object I in C, the object F (I) is G-acyclic. Then
for any object A in C, we have a biregular spectral sequence
E2pq = Rp G(Rq F (A)) ⇒ Rp+q (GF )(A).
Proof. Let I · be an injective resolution of A and let J ·· be a Cartan-Eilenberg

resolution of F (I · ). Consider the double complex G(J ·· ). Since for any i and q,
the short exact sequences
q q q
0 → BII (J i,· ) → ZII (J i,· ) → HII (J i,· ) → 0,
q q+1
0 → ZII (J i,· ) → J i,q → BII (J i,· ) → 0
are split, the sequences

q q q
0 → G(BII (J i,· )) → G(ZII (J i,· )) → G(HII (J i,· )) → 0,
q i,· i,q q+1
0 → G(ZII (J )) → G(J ) → G(BII (J i,· )) → 0
are exact (split). This implies that

q q
HII (G(J ·· )) = G(HII (J ·· )).
q
Since HII (J ·· ) is an injective resolution of H q (F (I · )) = Rq F (A), we have
HIp HII
q
(G(J ·· )) = HIp (G(HII
q
(J ·· ))) = Rp G(Rq F (A)).
So the first spectral sequence of the double complex G(J ·· ) is
EI pq p q
2 = R G(R F (A)) ⇒ H
p+q
(G(J ·· )).
Since J ·,p is an injective resolution of F (I p ), we have
HIq (G(J ·,p )) = Rq G(F (I p )).
Since F (I p ) is G-acyclic, we have

0 if q ≥ 1,
HIq (G(J ·,p ) q p
= R G(F (I )) =
GF (I p ) if q = 0,
and hence

p 0 if q ≥ 1,
HII HIq (G(J ·· )) =
H p (GF (I · )) = Rp (GF )(A) if q = 0.
So the second spectral sequence
EII pq p q ··
2 = HII HI (G(J )) ⇒ H
p+q
(G(J ·· ))
of the double complex G(J ·· ) degenerates. These two spectral sequences are
biregular by Proposition 2.8 (i). So by Proposition 2.2, we have
Rn (GF )(A) = H n (G(J ·· ))
for any n. Hence the first spectral sequence of the double complex G(J ·· ) can
be written as
E2pq = Rp G(Rq F (A)) ⇒ Rp+q (GF )(A).
Lemma 2.11. Let f : X → Y be a continuous map. For any injective sheaf I

of abelian groups, f∗ I is injective.
Proof. For any sheaf of abelian groups G on Y , we have
Hom(G, f∗ I) ∼
= Hom(f −1 G, I).
Since f −1 is an exact functor and I is injective, Hom(f −1 −, I) is an exact

functor. So Hom(−, f∗ I) is an exact functor and hence f∗ I is injective.
Corollary 2.12. Let f : X → Y be a continuous map and F a sheaf of abelian

groups on X. Then we have a biregular spectral sequence
E2pq = H p (Y, Rq f∗ F) ⇒ H p+q (X, F).
Proof. Apply Proposition 2.10 to the functors F = f∗ and G = Γ(Y, −). Note
that GF = Γ(X, −) and F (I) is G-acyclic for any injective sheaf I on X by
Lemma 2.11.
Similarly we have
Corollary 2.13. Let f : X → Y and g : Y → Z be continuous maps and let F

be a sheaf of abelian groups on X. Then we have a biregular spectral sequence
E2pq = Rp g∗ Rq f∗ F ⇒ Rp+q (gf )∗ F.
Lemma 2.14. Let (X, OX ) be a ringed space and let I be an injective OX -

module. Then for any OX -module F, HomOX (F, I) is flasque.
Proof. Let U be an open subset of X and let j : U → X be the inclusion.

Given a morphism φU : F|U → I|U , we need to show there exists a morphism
φ : F → I such that φ|U = φU . For any OU -module G on U , recall that j! G is
the sheaf on X associated to the presheaf

G(V ) if V ⊂ U,
V 7→
0 otherwise.
We leave to the reader to construct a canonical monomorphism j! (F|U ) → F.

j! (φU )
Let ψ : j! (F|U ) → I be the composition j! (F|U ) → j! (I|U ) → I. Since I is
injective, the morphism ψ can be extended to a morphism φ : F → I. We have
φ|U = φU .
OX -modules. Then we have a biregular spectral sequence
E2pq = H p (X, ExtqOX (F, G)) ⇒ ExtO

p+q
X
(F, G).
Proof. Apply Proposition 2.10 to the functors F = HomOX (F, −) and G =

Γ(X, −). Note that we have GF = HomOX (F, −) and F (I) is G-acyclic for any
injective OX -module I by Lemma 2.14.
Proposition 2.16. Let (X, OX ) be a ringed space, let F and G be OX -modules,

let I · be an injective resolution of G and let
· · · → E1 → E0 → F → 0
be an exact sequence of OX -modules such that each Ei is locally free of finite

rank. Then we have
ExtnOX (F, G) = H n (HomOX (F, I · )) ∼

= H n (HomOX (E· , G)) ∼
= H n (HomOX (E· , I · )).
Proof. Consider the double complex (K pq ) = (HomOX (Ep , I q )). Since each Ei
is locally free of finite rank, each functor HomOX (Ei , −) is exact and hence

q 0 if q ≥ 1,
HII (K ·· ) =
HomOX (E· , G) if q = 0.
So we have

0 if q ≥ 1,
HIp HII
q
(K ·· ) =
H p (HomOX (E· , G)) if q = 0.
Hence the first spectral sequence of the double complex K ·· degenerates. By
Proposition 2.2, this implies
H n (HomOX (E· , G)) = H n (HomOX (E· , I · )). (1)
Since each I i is injective, each functor HomOX (−, I i ) is exact and hence

q ·· 0 if q ≥ 1,
HI (K ) =
HomOX (F, I · ) if q = 0.
So we have
p 0 if q ≥ 1,
HII HIq (K ·· ) =
ExtpOX (F, G) if q = 0.
Hence the second spectral sequence of the double complex K ·· degenerates. By
Proposition 2.2, this implies
ExtnOX (F, G) = H n (HomOX (E· , I · )). (2)
Combining (1) and (2), we get
ExtnOX (F, G) ∼
= H n (HomOX (E· , I · )) ∼
= H n (HomOX (E· , G)).
Proposition 2.17. Let A be a ring, let M and N be two A-modules, and let
··· → E −1 → E0 → M → 0,
··· → F −1 → F0 → N → 0
be exact sequences such that E i and F i are flat A-modules. Then we have
TorA
n (M, N ) = H
−n
(M ⊗A F · ) ∼
= H −n (E · ⊗A N ) ∼
= H −n (E · ⊗A F · ).
In particular, we have TorA ∼ A
n (M, N ) = Torn (N, M ).
2.3 Čech Cohomology
Let X be a topological space and let U = {Ui }i∈I be an open covering of X. For
any i0 , . . . , in ∈ I, let Ui0 ...in = Ui0 ∩ · · · ∩ Uin . Given a presheaf P of abelian
groups on X, define a complex C · (U, P) as follows: For each n ≥ 0, let
Y
C n (U, P) = P(Ui0 ...in ),
i0 ,...,in ∈I
2.3. ČECH COHOMOLOGY 145
and define
d : C n (U, P) → C n+1 (U, P), s = (si0 ...in ) 7→ ds = ((ds)i0 ...in+1 )
by
n+1
X
(ds)i0 ...in+1 = (−1)k si0 ...îk ...in+1 |Ui0 ...in+1 ,
k=0
where si0 ...îk ...in+1 |Ui0 ...in+1 is the image of si0 ...îk ...in+1 ∈ P(Ui0 ...îk ...in+1 ) under
the restriction P(Ui0 ...îk ...in+1 ) → P(Ui0 ...in+1 ). One can verify that dd = 0. We
call C · (U, P) the Čech complex of P with respect to the open covering U. The
n-th cohomology group of C · (U, P) is denoted by Ȟ n (U, P) and is called the
n-th Čech cohomology group of P with respect to the open covering U.
Fix a total order on I. Define a complex C 0· (U, P) as follows: For each n,
let Y
C 0n (U, P) = P(Ui0 ...in ),
i0 <···<in
and define d : C 0n (U, P) → C 0n+1 (U, P) by the same formula as above. We call
C 0· (U, P) the alternating Čech complex of P with respect to the open covering
U.
Proposition 3.1. Let X be a topological space, U = {Ui }i∈I an open covering

of X and P a presheaf of abelian groups on X. Then there exists a morphism
of complexes f · : C 0· (U, P) → C · (U, P) inducing isomorphisms
H n (C 0· (U, P)) ∼
= H n (C · (U, P))
on cohomology groups.
Proof. Fix a total order on I. For each n ≥ 0, let Kn (I) be the free abelian
group with basis (i0 , . . . , in ) ∈ I n+1 and let Kn0 (I) be the free abelian subgroup
of Kn (I) generated by those (i0 , . . . , in ) such that i0 < · · · < in . Define
∂n : Kn (I) → Kn−1 (I)
by
n
X
∂n (i0 , . . . , in ) = (−1)k (i0 , . . . , îk , . . . , in ).
k=0
Then ∂n−1 ∂n = 0 and ∂n (Kn0 (I)) ⊂ Kn−1 0

(I) for any n. So K· (I) is a complex
0
and K· (I) is a sub-complex. (But the derivation operator ∂ has degree −1.)
Let i· : K·0 (I) → K· (I) be the inclusion. Define homomorphisms fn : Kn (I) →
Kn0 (I) (n ≥ 0) by

0 if ik = il for some k 6= l,
fn (i0 , . . . , in ) =
sgn(σ)(iσ(0) , . . . , iσ(n) ) otherwise,
where σ is a permutation of {0, . . . , n} such that iσ(0) < · · · < iσ(n) and sgn(σ)
is the sign of σ. Note that f· : K· (I) → K·0 (I) is a morphism of complexes and
we have
f· i· = idK·0 (I) .
We claim that there exist morphisms hn : Kn (I) → Kn+1 (I) (n ≥ 0) such that
∂n+1 hn + hn−1 ∂n = in fn − idKn (I) ,
and for any (i0 , . . . , in ) ∈ I n+1 , hn (i0 , . . . , in ) is a linear combination of those

(j0 , . . . , jn+1 ) ∈ I n+2 satisfying {j0 , . . . , jn+1 } ⊂ {i0 , . . . , in } as sets. Assume
the claim for a moment. Suppose
X
hn (i0 , . . . , in ) = Cji00...j
...in
n+1
(j0 , . . . , jn+1 ).
{j0 ,...,jn+1 }⊂{i0 ,...,in }
We define
h∗n : C n+1 (U, P) → C n (U, P), s = (sj0 ...jn+1 ) 7→ h∗n (s) = ((h∗n (s))i0 ...in )
by X
(h∗n (s))i0 ...in = Cji00...j
...in
s |
n+1 j0 ...jn+1 Ui0 ...in
.
{j0 ,...,jn+1 }⊂{i0 ,...,in }
We define : C (U, P) → C 0n (U, P) and fn∗ : C 0n (U, P) → C n (U, P) in a

i∗n n
similar way. One can show i∗· and f·∗ are morphisms of complexes and we have
i∗· f·∗ = idC 0· (U,P) ,

dn−1 h∗n−1 + h∗n dn = fn∗ i∗n − idC n (U,P) (n ≥ 0).
This implies that the homomorphisms on cohomology groups induced by f·∗ and
i∗· are inverse to each other.
Before proving our claim, we first show that for any set I, we have
Hn (K· (I)) = 0
for any n ≥ 1. Fix an element i in I. Consider the homomorphisms
gn : Kn (I) → Kn+1 (I), (i0 , . . . , in ) 7→ (i, i0 , . . . , in ).
One can verify

∂n+1 gn + gn−1 ∂n = idKn (I)
for any n ≥ 1. This implies Hn (K· (I)) = 0 for any n ≥ 1.
Let’s prove our claim. Note that i0 f0 : K0 (I) → K0 (I) coincides with
idK0 (I) . We define h0 : K0 (I) → K1 (I) to be 0. Suppose we have defined
hk : Kk (I) → Kk+1 (I) for any 0 ≤ k < n such that ∂k+1 hk + hk−1 ∂k = ik fk −
idKk (I) , and for any (i0 , . . . , ik ) ∈ I k+1 , hk (i0 , . . . , ik ) is a linear combination
of those (j0 , . . . , jk+1 ) ∈ I k+2 satisfying {j0 , . . . , jk+1 } ⊂ {i0 , . . . , ik }. Then
for any (i0 , . . . , in ) ∈ I n+1 , (in fn − idKn (I) − hn−1 ∂n )(i0 , . . . , in ) is a linear
combination of those (j0 , . . . , jn ) satisfying {j0 , . . . , jn } ⊂ {i0 , . . . , in }. We may
regard (in fn − idKn (I) − hn−1 ∂n )(i0 , . . . , in ) as an element in Kn ({i0 , . . . , in }).
Moreover, we have
∂n (in fn − idKn (I) − hn−1 ∂n )(i0 , . . . , in )

= (in−1 fn−1 − idKn−1 (I) )∂n (i0 , . . . , in ) − ∂n hn−1 ∂n (i0 , . . . , in )
= (∂n hn−1 + hn−2 ∂n−1 )∂n (i0 , . . . , in ) − ∂n hn−1 ∂n (i0 , . . . , in )
= 0.
Since Hn (K· ({i0 , . . . , in })) = 0, there exists a c ∈ Kn+1 ({i0 , . . . , in }) such that
∂n+1 (c) = (in fn − idKn (I) − hn−1 ∂n )(i0 , . . . , in ).
We define hn (i0 , . . . , in ) = c.
Corollary 3.2. Let X be a topological space, U = {U1 , . . . , Un } a finite open

covering of X, and P a presheaf on X. Then Ȟ q (U, P) = 0 for any q ≥ n.
Proof. We can use the alternating Čech complex to calculate Čech cohomology
groups. Our assertion follows from the fact that C 0q (U, P) = 0 for any q ≥ n
since there are n open subsets in U.
Proposition 3.3. Let X be a topological space and let U = {Ui }i∈I be an open
covering of X.
(i) The category of presheaves of abelian groups on X has enough injective
objects.
(ii) The functors (Ȟ i (U, −)) on the category of presheaves of abelian groups
form a cohomological functor.
(iii) For any injective presheaf I, we have Ȟ i (U, I) = 0 for any i ≥ 1.
(iv) (Ȟ i (U, −)) is a universal cohomological functor, Ȟ 0 (U, −) is left exact,
and Ȟ i (U, −) is the i-th right derived functor of Ȟ 0 (U, −) for any i ≥ 0.
Proof. We leave to the reader to prove (i) and (ii). We will prove (iii). This
implies H i (U, −) is effaceable for every i ≥ 1. So (iv) holds by Proposition 1.4
and Corollary 1.5.
Let’s prove (iii). We need to show the sequence
Y Y Y
I(Ui0 ) → I(Ui0 i1 ) → I(Ui0 i1 i2 ) → · · ·
i0 ∈I i0 ,i1 ∈I i0 ,i1 ,i2 ∈I
in the Čech complex is exact. For any open subset U of X, define a presheaf
ZU on X by
0 if V 6⊂ U,
ZU (V ) =
Z if V ⊂ U.
Then the above sequence is identified with

M M M
Hom( ZUi0 , I) → Hom( ZUi0 i1 , I) → Hom( ZUi0 i1 i2 , I) → · · ·
i0 ∈I i0 ,i1 ∈I i0 ,i1 ,i2 ∈I
Since I is an injective presheaf, it suffices to show the sequence

δ δ δ
M 0
M 1
M 2
ZUi0 ← ZUi0 i1 ← ZUi0 i1 i2 ← ···
i0 ∈I i0 ,i1 ∈I i0 ,i1 ,i2 ∈I
is an exact sequence of presheaves, that is, for any open subset V of X, the
sequence
δ δ δ
M 0
M 1
M 2
ZUi0 (V ) ← ZUi0 i1 (V ) ← ZUi0 i1 i2 (V ) ← ···
i0 ∈I i0 ,i1 ∈I i0 ,i1 ,i2 ∈I
is exact. Let’s describe

M M
δn : ZUi0 ,...,in+1 (V ) → ZUi0 ,...,in (V ).
i0 ,...,in+1 ∈I i0 ,...,in ∈I
Let S = {i ∈ I|V ⊂ Ui }. We have

M M M
ZUi0 ,...,in (V ) = Z= Z.
i0 ,...,in ∈I V ⊂Ui0 ,...,in (i0 ,...,in )∈S n+1
L
So ZUi0 ,...,in (V ) can be identified with Kn (S) defined in the proof of
i0 ,...,in ∈I
Proposition 3.1. Then
M M
δn : ZUi0 ,...,in+1 (V ) → ZUi0 ,...,in (V )
i0 ,...,in+1 ∈I i0 ,...,in ∈I
is identified with ∂n+1 : Kn+1 (S) → Kn (S). In the proof of Proposition 3.1, we
have shown H n (K· (S)) = 0 for any n ≥ 1. So the sequence
δ δ δ
M 0
M 1
M 2
ZUi0 (V ) ← ZUi0 i1 (V ) ← ZUi0 i1 i2 (V ) ← ···
i0 ∈I i0 ,i1 ∈I i0 ,i1 ,i2 ∈I
is exact.
Let U = {Ui }i∈I and V = {Vj }j∈J be two open coverings of X. We say U is
a refinement of V there exists a map τ : I → J such that Ui ⊂ Vτ (i) . Define
τ ∗ : C n (V, P) → C n (U, P), s = (sj0 ...jn ) 7→ τ ∗ (s) = ((τ ∗ (s))i0 ...in )
by
(τ ∗ (s))i0 ...in = sτ (i0 )...τ (in ) |Ui0 ...in .
Then τ ∗ : C · (V, P) → C · (U, P) is a morphism of complex and it induces mor-

phisms τ ∗ : Ȟ n (V, P) → Ȟ n (U, P) on cohomology groups.
Proposition 3.4. Notation as above. If τ 0 : I → J is another map such that

Ui ⊂ Vτ 0 (i) , then τ ∗ and τ ∗ induces the same homomorphisms Ȟ n (V, P) →
Ȟ n (U, P) on cohomology groups.
Proof. Define
h : C n (V, P) → C n−1 (U, P), s = (sj0 ...jn ) 7→ h(s) = ((h(s))i0 ...in−1 )
by
n−1
X
(h(s))i0 ...in−1 = (−1)k sτ (i0 )...τ (ik )τ 0 (ik )...τ 0 (in−1 ) |Ui0 ...in−1 .
k=0
One can check

hd + dh = τ 0∗ − τ ∗ .
So τ ∗ and τ 0∗ are homotopic and hence they induce the same homomorphisms
on Čech cohomology groups.
Given two open coverings U and V of X, we denote U ≥ V if U is a refinement

of V. For any two open coverings U = {Ui }i∈I and V = {Vj }j∈J , the open
covering W = {Ui ∩ Vj }i∈I,j∈J is a common refinement of U and V. So the
family of open coverings of X is a direct set. By Proposition 3.4, we have well
defined homomorphisms Ȟ n (V, P) → Ȟ n (U, P) whenever U ≥ V. We define
the n-th Čech cohomology group Ȟ n (X, P) of the presheaf P to be
Ȟ n (X, P) = dir. lim Ȟ n (U, P).

U
As a corollary of Proposition 3.3, we have
Corollary 3.5. Let X be a topological space.

(i) The functors (Ȟ i (X, −)) on the category of presheaves of abelian groups
form a cohomological functor.
(ii) For any injective presheaf I, we have Ȟ i (X, I) = 0 for any i ≥ 1.
(iii) (Ȟ i (X, −)) is a universal cohomological functor, Ȟ 0 (X, −) is left exact,
and Ȟ i (X, −) is the i-th right derived functor of Ȟ 0 (X, −) for any i ≥ 0.
Now we discuss the Čech cohomology of sheaves. For any open covering
U = {Ui }i∈I of X and any sheaf of abelian groups F on X, Ȟ 0 (U, F) is the
kernel of the homomorphism
Y Y
F(Ui ) → F(Ui ∩ Uj ), (si ) 7→ (sj |Ui ∩Uj − si |Ui ∩Uj ).
i∈I i,j∈I
So we have
Ȟ 0 (U, F) = F(X) = H 0 (X, F).
Proposition 3.6. Let F be a flasque sheaf on X. Then for any open covering
U = {Ui }i∈I of X, we have Ȟ n (U, F) = 0 for any n ≥ 1.
Proof. First we note that any injective sheaf is an injective object in the
category of presheaves. So we have Ȟ n (U, I) = 0 for any n ≥ 1 and any
injective sheaf I. Let I · be an injective resolution of the flasque sheaf F in the
category of sheaves. We have exact sequences
0 → F → I0 → Z 1 (I · ) → 0,
0 → Z (I) → I 1
1
→ Z 2 (I · ) → 0,
..
.
By Lemma 1.11, all the sheaves in the above exact sequences are flasque and
for any open subset U , the following sequences are exact:
0 → F(U ) → I 0 (U ) → Z 1 (I · )(U ) → 0,
0 → Z (I)(U ) → I 1 (U ) → Z 2 (I · )(U ) → 0,
1
..
.
So
0 → F(U ) → I 0 (U ) → I 1 (U ) → · · ·
is exact for any open subset U of X, that is,
0 → F → I0 → I1 → · · ·
is exact in the category of presheaves. Therefore I · is an injective resolution of

F in the category of presheaves. By Proposition 3.3 (iv), Ȟ n (U, F) is the n-th
cohomology group of the complex
0 → Ȟ 0 (U, I 0 ) → Ȟ 0 (U, I 1 ) → · · · ,
which can identified with
0 → I 0 (U ) → I 1 (U ) → · · · .
So we have Ȟ n (U, F) = 0 for any n ≥ 1.
Let U be an open covering of X, let F be a sheaf on X and let I · be an

injective resolution of F in the category of sheaves. Consider the double complex
(K pq ) = (C p (U, I q )). We have
HIp HII
q
(K ·· ) = Ȟ p (U, Hq (F)),
where Hq (F) is the presheaf defined by
Hq (F)(U ) = H q (U, F|U ).

On the other hand, we have
HIq (K ·,p ) = Ȟ q (U, I p ).
Since I p is an injective sheaf, we have Ȟ q (U, I p ) = 0 for any q ≥ 1 by Proposi-

tion 3.6. So

p 0 if q ≥ 1,
HII HIq (K ·· ) =
H p (Γ(X, I · )) = H p (X, F) if q = 0.
Hence the second spectral sequence of the double complex K ·· degenerates and
this gives H n (X, F) = H n (K ·· ). So the first spectral sequence of the double
complex can be written as
E2pq = Ȟ p (U, Hq (F)) ⇒ H p+q (X, F).
By Proposition 2.8 (i), this spectral sequence is biregular. Taking direct limit
with respect to open coverings U of X, we get a biregular spectral sequence
E2pq = Ȟ p (X, Hq (F)) ⇒ H p+q (X, F).
So we have the following proposition:
Proposition 3.7. Let F be a sheaf on X and let Hq (F) be the presheaf defined
by Hq (F)(U ) = H q (U, F|U ).
(i) For any open covering U of X, we have a biregular spectral sequence
E2pq = Ȟ p (U, Hq (F)) ⇒ H p+q (X, F).
(ii) We have a biregular spectral sequence
E2pq = Ȟ p (X, Hq (F)) ⇒ H p+q (X, F).
Corollary 3.8. Let F be a sheaf on X and let U = {Ui }i∈I be an open covering
of X. Suppose for any i0 , . . . , in ∈ I, we have H q (Ui0 ∩ · · · ∩ Uin , F) = 0 for any
q ≥ 1. Then for any n, we have Ȟ n (U, F) ∼ = H n (X, F).
Proof. Under our assumption, the biregular spectral sequence
E2pq = Ȟ p (U, Hq (F)) ⇒ H p+q (X, F)
degenerates. Our assertion then follows from Proposition 2.2.
Lemma 3.9. For any sheaf F on X, we have Ȟ 0 (X, Hq (F)) = 0 for any q ≥ 1.
Proof. Let U = {Ui }i∈I be an open covering of X. Note that Ȟ 0 (U, Hq (F)) is
the kernel of the homomorphism
Y Y
Hq (F)(Ui ) → Hq (F)(Ui ∩ Uj ), (si ) 7→ (sj |Ui ∩Uj − si |Ui ∩Uj ).
i∈I i,j∈I
Let (si ) be an element in Ȟ 0 (U, F), where si ∈ Hq (F)(Ui ) = H q (Ui , F|Ui ). Fix
an injective resolution I · of F. For any q ≥ 1, we have
H q (Ui , F) = ker(I q (Ui ) → I q+1 (Ui ))/im(I q−1 (Ui ) → I q (Ui )).
Let ti ∈ ker(I q (Ui ) → I q+1 (Ui )) such that its image in H q (Ui , F) is si . Since
I q−1 → I q → I q+1
is exact, we may find an open covering {Uij }j∈Ji of Ui such that each ti |Uij lies in
im(I q−1 (Uij ) → I q (Uij )). Then the image of each si under the homomorphism
Hq (F)(Ui ) → Hq (F)(Uij ) vanishes. Let U0 be the open covering defined by Uij
(i ∈ I, j ∈ Ji ). Then U0 is a refinement of U and the image of s = (si ) under
homomorphism Ȟ 0 (U, Hq (F)) → Ȟ 0 (U0 , Hq (F)) vanishes. So
Ȟ 0 (X, Hq (F)) = dir. lim Ȟ 0 (U, Hq (F)) = 0

U
for any q ≥ 1.
Corollary 3.10. For any sheaf F on X, we have Ȟ 0 (X, F) ∼

= H 0 (X, F),
1 ∼ 1 2 2
Ȟ (X, F) = H (X, F) and Ȟ (X, F) ,→ H (X, F).
Proof. For the spectral sequence in Proposition 3.7 (ii), we have
E2pq = Ȟ p (X, Hq (F)) = 0
when p < 0 or q < 0. By Proposition 2.3, we have E200 = Ȟ 0 (X, F) ∼

= H 0 (X, F)
and we have an exact sequence
0 → E21,0 → H 1 → E20,1 → E22,0 → H 2
By Lemma 3.9, we have E20,1 = Ȟ 0 (X, H1 (F)) = 0. So we have E21,0 ∼

= H 1 and
E2 ,→ H , that is, Ȟ (X, F) ∼
2,0 2 1 1 2 2
= H (X, F) and Ȟ (X, F) ,→ H (X, F).
Proposition 3.11. (Cartan) Let B be a family of open subsets of X which

form a basis for the topology of X and let F be a sheaf on X such that for
any U1 , . . . , Uk ∈B, we have Ȟ q (U1 ∩ · · · ∩ Uk , F) = 0 for any q ≥ 1. Then
H q (U1 ∩ · · · ∩ Uk , F) = 0 for any q ≥ 1. For any open subset V of X and any
open covering U of V by open subsets in B, we have Ȟ n (U, F) ∼ = H n (V, F) for
n ∼ n
any n. Moreover, we have Ȟ (V, F) = H (V, F) for any n.
Proof. Let U = U1 ∩ · · · ∩ Uk . We prove H q (U, F) = 0 by induction on q ≥ 1.

We have
H 1 (U, F) ∼
= Ȟ 1 (U, F) = 0.
Suppose we have shown H q (U, F) = 0 for 0 < q < n. Then for any open
covering U of U by open subsets in B, any 0 ≤ q ≤ n and any p, we have
C p (U, Hq (F)) = 0 and hence Ȟ p (U, Hq (F)) = 0. So we have Ȟ p (U, Hq (F)) = 0

when 0 < q < n. Applying Proposition 2.3 to the spectral sequence
E2pq = Ȟ p (U, Hq (F)) ⇒ H p+q (U, F),
we get an exact sequence
0 → E2n,0 → Hn → E20,n .
k k k
Ȟ n (U, F) H n (U, F) 0
where the third equality follows from Lemma 3.9. So we have
H n (U, F) ∼
= Ȟ n (U, F) = 0.
By Corollary 3.8, for any open subset V of X and any open covering U of V by
open subsets in B, we have Ȟ n (U, F) ∼
= H n (V, F) for any n. So
Ȟ n (V, F) = dir. lim H n (U, F) ∼

= H n (V, F).
U
Let (X, OX ) be a ringed space. Recall that isomorphic classes of invertible

OX -modules form a group, where the product of two elements is defined by
the tensor product of invertible sheaves. The isomorphic class of the trivial
invertible sheaf OX is the identity element, and for any invertible sheaf L, the
inverse of its isomorphic class is the isomorphic class of HomOX (L, OX ). This
group is called the Picard group of X and is denoted by Pic(X).
Proposition 3.12. Let (X, OX ) be a ringed space. Then we have a canonical

isomorphism
∗ ∼
H 1 (X, OX ) = Pic(X),
∗
where OX is the sheaf defined by
∗
OX (U ) = {s ∈ OX (U )|s is a unit in OX (U )}.
It is a sheaf of abelian groups whose group structure is induced by the multi-

plication of the sheaf of rings OX .
Proof. For any open covering U = {Ui }i∈I , let Pic(U) be the subgroup of
Pic(X) consisting of the isomorphic classes of those invertible sheaves L satis-
fying L|Ui ∼
= OUi . We will construct a canonical isomorphism
∗ ∼
Ȟ 1 (U, OX ) = Pic(U).
Obviously we have Pic(X) = dir. limU Pic(U). Passing to direct limit over cov-
erings of X, we get an isomorphism
∗ ∼
Ȟ 1 (X, OX ) = Pic(X).
By Corollary 3.10, this would prove our proposition.

Fix a total order on I. Let
∗
Z 1 (U, OX )
01 ∗
= ker(C (U, OX ) → C 02 (U, OX ∗
))
Y
∗
= {(si0 i1 ) ∈ OX (Ui0 i1 )|si1 i2 si0 i1 = si0 i2 on Ui0 i1 i2 for any i0 < i1 < i2 },
i0 <i1
1 ∗
B (U, OX )
= im(C 00 (U, OX
∗
) → C 01 (U, OX
∗
))
Y Y
= {(si0 i1 ) ∈ ∗
OX (Ui0 i1 )|si0 i1 = ti1 t−1
i0 on Ui0 i1 for some (ti0 ) ∈
∗
OX (Ui0 )}.
i0 <i1 i0
Then we have
∗ ∼ 1 ∗ ∗
Ȟ 1 (U, OX ) = Z (U, OX )/B 1 (U, OX ).
Define a map
∗
Φ : Z 1 (U, OX ) → Pic(U)
∗
as follows: For any (si0 i1 ) ∈ Z 1 (U, OX ) and any pair i0 < i1 , let
φi0 i1 : OUi0 |Ui0 i1 → OUi1 |Ui0 i1
be the isomorphism which maps any section s of OUi0 |Ui0 i1 to the section ssi0 i1
of OUi1 |Ui0 i1 . We define φii = idOUi for any i ∈ I and φi1 i0 = φ−1 i0 i1 for any
i0 < i1 . For any i0 , i1 , i2 ∈ I, we have si1 i2 si0 i1 = si0 i2 on Ui0 i1 i2 , so φi1 i2 φi0 i1 =
φi0 i2 on Ui0 i1 i2 . Hence we may glue {OUi } together through {φij } to get an
invertible OX -module. We define Φ((si0 i1 )) ∈ Pic(U) to be the isomorphic class
∗
of this invertible OX -module. One can verify that Φ : Z 1 (U, OX ) → Pic(U) is a
homomorphism.
Suppose (si0 i1 ) lies in the kernel of Φ and let L be the invertible OX -module
constructed above using the glueing data define by (si0 i1 ). Then we have an
isomorphism ψ : OX ∼ = L. By our construction, we have L|Ui ∼ = OUi for any
i ∈ I. Composition this isomorphism with ψ|Ui , we get an automorphism of
OUi . Any automorphism of OUi is induced by multiplication by an element
∗ ∗
in OX (Ui ). Let ti ∈ OX (Ui ) correspond to the above automorphism. One
can verify that si0 i1 = ti1 t−1 i0 on Ui0 i1 for any i0 < i1 and hence (si0 i1 ) lies in
∗ ∗
B 1 (U, OX ). Conversely, one can show every element in B 1 (U, OX ) lies in kerΦ.
1 ∗
So we have kerΦ = B (U, OX ).
Let’s show Φ is surjective. Suppose L is an invertible OX -module such that
we have an isomorphism φi : L|Ui ∼ = OUi for any i ∈ I. For any i1 < i2 , consider
the composition
φ−1
i φi
OUi0 |Ui0 i1 →0 L|Ui0 i1 →1 OUi1 |Ui0 i1 .
∗
It induces an automorphism of OUi0 i1 . Let si0 i1 ∈ OX (Ui0 i1 ) correspond to this
automorphism. Then we have si0 i1 si1 i2 = si0 i2 on Ui0 i1 i2 for any i0 < i1 < i2 .
∗
So (si0 i1 ) ∈ Z 1 (U, OX ). One can verify that Φ((si0 i1 )) is equal to the isomorphic
class of L. So Φ is surjective.
2.4. COHOMOLOGY OF AFFINE AND PROJECTIVE SCHEMES 155
∗ ∼ 1 ∗ ∗
Therefore Φ induces an isomorphism of Ȟ 1 (U, OX ) = Z (U, OX )/B 1 (U, OX )
with Pic(U).
2.4 Cohomology of Affine and Projective Schemes
Before calculating the cohomology of affine and projective schemes, we study

the cohomology of Koszul complex.
Let A be a ring and f1 , . . . , fr ∈ A. Define the Koszul complex
Vp K· (f1 , . . . , fr )
as follows: K1 is the free A-module Ar of rank r and Kp = K1 for any p ≥ 0.
For each i ∈ {1, . . . , r}, let ei be the element in K1 = Ar whose i-th component
is 1 and whose other components are 0. Then ei (i = 1, . . . , r) form
Vp a basis of
K1 and ei1 ∧ · · · ∧ eip (i1 < · · · < ip ) form a basis for Kp = K1 . Define
∂p : Kp → Kp−1 by
p
X
∂(ei1 ∧ · · · ∧ eip ) = (−1)k−1 fik ei1 ∧ · · · ∧ êik ∧ · · · ∧ eip .
k=1
Note that ∂p is nothing but the contraction by f1 e∗1 +· · ·+fr e∗r , where {e∗1 , . . . , e∗r }
is the basis of HomA (Ar , A) dual to {e1 , . . . , er }. One can verify that ∂∂ = 0.
For any A-module M , define
K· (f1 , . . . , fr ; M ) = K· (f1 , . . . , fr ) ⊗A M,
K · (f1 , . . . , fr ; M ) = HomA (K· (f1 , . . . , fr ), M ),
Hp (f1 , . . . , fp ; M ) = Hp (K· (f1 , . . . , fr ; M )),
H p (f1 , . . . , fp ; M ) = H p (K · (f1 , . . . , fr ; M )).
Any element z in Kp (f1 , . . . , fr ; M ) can be written as
X
z= ei1 ∧ · · · ∧ eip ⊗ zi1 ...ip
i1 <···<ip
for some uniquely determined zi1 ...ip ∈ M . For any distinct elements i1 , . . . , ip
in {1, . . . , r}, define
zi1 ...ip = sgn(σ)ziσ(1) ...iσ(p) ,
where σ is a permutation of {1, . . . , p} such that iσ(1) < · · · < iσ(p) . Then we
have
X p
X
∂z = (−1)k−1 fik ei1 ∧ · · · ∧ êik ∧ · · · ∧ eip ⊗ zi1 ...ip
i1 <···<ip k=1
X p
X
= fik ei1 ∧ · · · ∧ êik ∧ · · · ∧ eip ⊗ zik i1 ...îk ...ip
i1 <···<ip k=1
X X
= ei1 ∧ · · · ∧ eip−1 ⊗ ( fj zji1 ...ip−1 ).
i1 <···<ip−1 j∈{1,...,r}−{i1 ,...,ip−1 }
So we have X
(∂z)i1 ...ip−1 = fj zji1 ...ip−1 .
j∈{1,...,r}−{i1 ,...,ip−1 }
Any element g in K p (f1 , . . . , fr ; M ) = HomA (Kp , A) is completely determined

by gi1 ...ip = g(ei1 ∧ · · · ∧ eip ). The derivation operator is given by
(dg)i1 ...ip+1 = (dg)(ei1 ∧ · · · ∧ eip+1 ) = g(∂(ei1 ∧ · · · ∧ eip+1 ))

p+1
X
= (−1)k−1 fik gi1 ...îk ...ip+1 .
k=1
Proposition 4.1. We have Hp (f1 , . . . , fr ; M ) ∼

= H r−p (f1 , . . . , fr ; M ) for any p.
Proof. Define a homomorphism
∗ : Kp (f1 , . . . , fr ; M ) → K r−p (f1 , . . . , fr ; M )

P
as follows: For any z = i1 <···<ip ei1 ∧ · · · ∧ eip ⊗ zi1 ...ip in Kp (f1 , . . . , fr ; M ),
we define ∗z ∈ K r−p (f1 , . . . , fr ; M ) = Hom(Kr−p , M ) by
(∗z)j1 ...jr−p = (∗z)(ej1 ∧ · · · ∧ ejr−p ) = sgn(i1 . . . ip j1 . . . jr−p )zi1 ...ip ,
where {i1 , . . . , ip , j1 , . . . , jr−p } = {1, . . . , r}, and sgn(i1 . . . ip j1 . . . jr−p ) denotes

the sign of the permutation

1 ... p p + 1 ... r
.
i1 . . . ip j1 . . . jr−p
We claim that
∗∂ = (−1)p−1 d∗
on Kp (f1 , . . . , fr ; M ). Our assertion then follows. Suppose
{i1 , . . . , ip−1 , j1 , . . . , jr−p+1 } = {1, . . . , r}.
We have
(∗∂z)j1 ...jr−p+1
= sgn(i1 . . . ip−1 j1 . . . jr−p+1 )(∂z)i1 ...ip−1
X
= sgn(i1 . . . ip−1 j1 . . . jr−p+1 ) fj zji1 ...ip−1
j∈{1,...,r}−{i1 ,...,ip−1 }
r−p+1
X
= sgn(i1 . . . ip−1 j1 . . . jr−p+1 ) fjk zjk i1 ...ip−1 ,
k=1
and
(d ∗ z)j1 ...jr−p+1
r−p+1
X
= (−1)k−1 fjk (∗z)j1 ...ĵk ...jr−p+1
k=1
r−p+1
X
= (−1)k−1 fjk sgn(jk i1 . . . ip−1 j1 . . . ĵk . . . jr−p+1 )zjk i1 ...ip−1
k=1
r−p+1
X
= (−1)p−1 sgn(i1 . . . ip−1 j1 . . . jr−p+1 ) fjk zjk i1 ...ip−1 .
k=1
So we have (∗∂z)j1 ...jr−p+1 = (−1)p−1 (d ∗ z)j1 ...jr−p+1 .
Proposition 4.2. Suppose f1 , . . . , fr generate the unit ideal of A. Then for

any A-module M , we have Hp (f1 , . . . , fr ; M ) = 0 and H p (f1 , . . . , fr ; M ) = 0 for
any p.
Proof. By Proposition 4.1, it suffices to show H p (f1 , . . . , fr ; M ) = 0. Choose

a1 , . . . , ar ∈ A such that a1 f1 + · · · + ar fr = 1. Define
H : K p (f1 , . . . , fr ; M ) → K p−1 (f1 , . . . , fr ; M )
by
r
X
(Hg)i1 ...ip−1 = ai gii1 ...ip−1 .
i=1
We have
((dH + Hd)g)i1 ...ip
Xp r
X
= (−1)k−1 fik (Hg)i1 ...îk ...ip + ai (dg)ii1 ...ip
k=1 i=1
Xp Xr Xr p
X
k−1 k
= (−1) fik ai gii1 ...îk ...ip + ai fi gi1 ...ip +ai (−1) fik gii1 ...îk ...ip
k=1 i=1 i=1 k=1
Xr
= ai fi gi1 ...ip
i=1
= gi1 ...ip .
So dH + Hd = id, that is, the identity morphism of K · (f1 , . . . , fr ; M ) is homo-
topic to 0. This implies that H p (f1 , . . . , fr ; M ) = 0 for any p.
Proposition 4.3. Let f1 , . . . , fr ∈ A and let M be an A-module. Suppose for

any 1 ≤ k ≤ r, the homomorphism
k−1
X k−1
X
M/ fi M → M/ fi M, z 7→ fk z
i=1 i=1
is injective. Then Hp (f1 , . . . , fr ; M ) = 0 for any p 6= 0 and H p (f1 , . . . , fr ; M ) =

0 for any p 6= r.
Proof. By Proposition 4.1, it suffices to show Hp (f1 , . . . , fr ; M ) = 0 for any

p 6= 0. We prove this by induction on r.
When r = 1, K· (f1 ; M ) is identified with the complex
f1
0 → M → M → 0,
where the morphism f1 is multiplication by f1 . Since f1 : M → M is injective,
we have Hp (f1 ; M ) = 0 for any p 6= 0.
Suppose we have shown Hp (f1 , . . . , fr−1 ; M ) = 0 for any p 6= 0. Note
that K· (f1 , . . . , fr−1 ; M ) can be regarded as a sub-complex of K· (f1 , . . . , fr ; M ).
Moreover, for each p, we have an isomorphism
Kp−1 (f1 , . . . , fr−1 ; M ) → Kp (f1 , · · · , fr ; M )/Kp (f1 , . . . , fr−1 ; M ),
ei1 ∧ · · · ∧ eip−1 ⊗ z 7→ ei1 ∧ · · · ∧ eip−1 ∧ er ⊗ z
(1 ≤ i1 < . . . < ip−1 ≤ r − 1, z ∈ M ).
One can verify that these isomorphisms commute with ∂. So we have an exact
sequence of complexes
0 → K· (f1 , . . . , fr−1 ; M ) → K· (f1 , . . . , fr ; M ) → K· (f1 , · · · , fr−1 ; M )[−1] → 0,
where the complex K· (f1 , · · · , fr−1 ; M )[−1] is defined by
(K· (f1 , · · · , fr−1 ; M )[−1])p = Kp−1 (f1 , · · · , fr−1 ; M ).
Consider the long exact sequence of cohomology groups associated to this short
exact sequence
· · · → Hp (f1 , . . . , fr−1 ; M ) → Hp (f1 , . . . , fr ; M ) → Hp−1 (f1 , . . . , fr−1 ; M ) → · · · .
By the induction hypothesis, Hp (f1 , . . . , fr−1 ; M ) and Hp−1 (f1 , . . . , fr−1 ; M )
are 0 for any p ≥ 2. So we have Hp (f1 , . . . , fr ; M ) = 0 for any p ≥ 2. Moreover,
a part of the above long exact sequence is
0 → H1 (f1 , . . . , fr ; M ) → H0 (f1 , . . . , fr−1 ; M ) → H0 (f1 , . . . , fr−1 ; M ).
r−1
P fr r−1
P
One can verify the last arrow can be identified with M/ fr M → M/ fr M.
i=1 i=1
Since it is injective, we have H1 (f1 , . . . , fr ; M ) = 0.
Proposition 4.4. Let X = SpecA be an affine scheme and F a quasi-coherent

OX -module. Then we have Ȟ p (X, F) = 0 for any p ≥ 1.
Proof. For any A-module M and any f ∈ A, consider the direct system
(M (n) , fmn )n∈N defined by M (n) = M for any n ∈ N and
fmn : M (m) → M (n) , x 7→ f n−m x
for any m ≤ n. The family of homomorphisms fn : M (n) → Mf defined by

x 7→ fxn induces an isomorphism
dir. lim M (n) ∼

= Mf .
n
Suppose F ∼ = M ∼ . Let f1 , . . . , fr ∈ A such that U = {D(f1 ), . . . , D(fr )} is

an open covering of X. The alternating Čech complex is given by
Y
C 0p (U, F) = F(D(fi0 ) ∩ · · · ∩ D(fip ))
1≤i0 <···<ip ≤r
Y
= Mfi0 ···fip
1≤i0 <···<ip ≤r
(n)
Y
= dir. lim Mi0 ...ip ,
n
1≤i0 <···<ip ≤r
Q (n)
where the direct system ( Mi0 ...ip )n∈N is defined by
1≤i0 <···<ip ≤r
(n)
Y Y
Mi0 ...ip = M
1≤i0 <···<ip ≤r 1≤i0 <···<ip ≤r
for any n and

(m) (n)
Y Y
Mi0 ...ip → Mi0 ...ip , (xi0 ...ip ) 7→ ((fi0 · · · fip )n−m xi0 ...ip )
1≤i0 <···<ip ≤r 1≤i0 <···<ip ≤r
(n)
for any m ≤ n. Let Cnp (M ) =
Q
Mi0 ...ip and define
1≤i0 <···<ip ≤r
d : Cnp (M ) → Cnp+1 (M ), x = (xi0 ...ip ) 7→ dx = ((dx)i0 ...ip+1 )
by
p+1
X
(dx)i0 ...ip+1 = (−1)k fink xi0 ...îk ...ip+1 .
i=0
Then each Cn· (M ) is a complex and
C · (U, F) ∼
= dir. lim Cn· (M ).
n
So for any p, we have
Ȟ p (U, F) ∼
= dir. lim H p (Cn· (M )).
n
We will show H p (Cn· (M )) = 0 for any n and any p ≥ 1. This implies Ȟ p (U, F) =
0 for any p ≥ 1. Since any open covering of X has a refinement consisting of
open subsets of the form D(f ) (f ∈ A), we have Ȟ p (X, F) = 0 for any p ≥ 1.
Since {D(f1n ), . . . , D(frn )} covers X = SpecA, f1n , . . . , frn generate the unit
ideal of A. Note that for any p ≥ 1, the sequence
d d
Cnp−1 (M ) → Cnp (M ) → Cnp+1 (M )
can be identified with the sequence

d d
K p (f1n , . . . , frn ; M ) → K p+1 (f1n , . . . , frn ; M ) → K p+2 (f1n , . . . , frn ; M ).
So we have
H p (Cn· (M )) ∼
= H p+1 (f1n , . . . , frn ; M ).
Our assertion then follows from Proposition 4.2.
Corollary 4.5. Let X be a separated scheme and let F be a quasi-coherent

OX -module. Then for any covering U of X by affine open subschemes, we have
= H p (X, F) for any p. Moreover we have Ȟ p (X, F) ∼
Ȟ p (U, F) ∼ = H p (X, F) for
any p.
Proof. Let B be the family of affine open subschemes of X. Then for any
U1 , . . . , Uk in B, U1 ∩ · · · ∩ Uk is also affine by Proposition 1.3.25 (i) and hence
Ȟ q (U1 ∩ · · · ∩ Uk , F) = 0 for any q ≥ 1 by Proposition 4.4. The corollary then
follows from Proposition 3.11.
Theorem 4.6. Let X be an affine scheme and F a quasi-coherent OX -module.

Then we have H p (X, F) = 0 for any p ≥ 1.
Proof. By Corollary 4.5, we have H p (X, F) = Ȟ p (X, F) for any p and by

Proposition 4.4, we have Ȟ p (X, F) = 0 for any p ≥ 1. Our assertion follows.
Corollary 4.7. Let

0→F →G→H→0
be an exact sequence of OX -modules on a scheme X. If F and H are quasi-
coherent, then G is also quasi-coherent.
Proof. The problem is local, so we may assume X is affine. Since F is quasi-

coherent, we have H 1 (X, F) = 0 by Theorem 4.6. So the long exact sequence
of cohomology groups associated to the above short exact sequence gives rise to
an exact sequence
0 → Γ(X, F) → Γ(X, G) → Γ(X, H) → 0.
0 → Γ(X, F)∼ → Γ(X, G)∼ → Γ(X, H)∼ → 0

↓ ↓ ↓
0 → F → G → H → 0.
The first and third vertical arrows are isomorphisms since F and H are quasi-
coherent. So the middle vertical arrow is also an isomorphism and hence G is
quasi-coherent.
Corollary 4.8. Let f : X → Y be an affine morphism and F a quasi-coherent

OX -module. Then Rp f∗ F = 0 for any p ≥ 1.
Proof. For any affine open subscheme V of Y , we have H p (f −1 (V ), F) = 0 for

any p ≥ 1 since f −1 (V ) is affine. Our assertion then follows from Proposition
1.9 and the fact that affine open subschemes form a basis for the topology of Y .
Let U = {Ui }i∈I be an open covering of X. For any open subsets Ui0 , . . . , Uin
in U, let ji0 ...in : Ui0 ...in ,→ X be the inclusion. Define the sheafified Čech
complex C · (U, F) as follows: Fix a total order on I. For any n ≥ 0, let
Y
C n (U, F) = ji0 ...in ∗ F.
i0 <···<in
Define
d : C n (U, F) → C n+1 (U, F)
as follows: For any Q open subset V of X and any section s = (si0 ...in ) in
C n (U, F)(V ) = F(Ui0 ...in ∩ V ), define a section ds = ((ds)i0 ...in+1 ) in
i0 <···<in
C n+1 (U, F)(V ) =
Q
F(Ui0 ...in+1 ∩ V ) by
i0 <···<in+1
n+1
X
(ds)i0 ...in+1 = (−1)k si0 ...îk ...in+1 |Ui0 ...in+1 ∩V .
k=0
Note that the n-th cohomology sheaf of C · (U, F) is the sheaf associated to the
presheaf V 7→ Ȟ n (U ∩ V, F), where U ∩ V is the open covering {Ui ∩ V }i∈I of
V.
Proposition 4.9. Let f : X → Y be a quasi-compact and separated morphism

of schemes. Then for any quasi-coherent OX -module F and any n, the OY -
module Rn f∗ F is quasi-coherent. Moreover, for any affine open subscheme V
of Y , we have a canonical isomorphism
H n (f −1 (V ), F) ∼
= Γ(V, Rn f∗ F).
In particular, if Y = SpecA is affine, then
Rn f∗ F = H n (X, F)∼ .
Proof. We may assume Y is affine. Let U be a covering of X by finitely

many affine open subschemes and let C · (U, F) be the sheafified Čech complex.
Consider the complex f∗ C · (U, F) of sheaves on Y . Its n-th cohomology sheaf

H n (f∗ C · (U, F)) is the sheaf associated to the presheaf
V 7→ Ȟ n (U ∩ f −1 (V ), F).
When V is an affine open subscheme of Y , U ∩ f −1 (V ) is a covering of f −1 (V )

by affine open subschemes. So by Corollary 4.5, we have
Ȟ n (U ∩ f −1 (V ), F) = H n (f −1 (V ), F)
for any affine open subscheme V of Y . Hence H n (f∗ C · (U, F)) is also the sheaf
associated to the presheaf
V 7→ H n (U ∩ f −1 (V ), F).
Therefore by Proposition 1.9, we have
Rn f∗ F ∼
= H n (f∗ C · (U, F)).
By Proposition 1.4.9 (iii), each f∗ C n (U, F) is quasi-coherent. So Rn f∗ F is also

quasi-coherent. Moreover, since Y is affine, we have
f∗ C · (U, F) ∼
= Γ(Y, f∗ C · (U, F))∼ .
Obviously we have
Γ(Y, f∗ C · (U, F)) = C · (U, F).
So we have
Rn f∗ F = H n (C · (U, F))∼
= Ȟ n (U, F)∼
= H n (X, F)∼ .
In particular, we have
Γ(Y, Rn f∗ F) = H n (X, F).
A morphism g : Y 0 → Y of schemes is called flat if for any y 0 ∈ Y 0 , OY 0 ,y0

is a flat OY,g(y0 ) -algebra. By Proposition 1.19, this is equivalent to saying that
for any affine open subscheme U 0 of Y 0 and any affine open subscheme U of Y
such that g(U 0 ) ⊂ U , OY 0 (U 0 ) is a flat OY (U )-algebra.
Proposition 4.10. Let f : X → Y be a quasi-compact and separated morphism

and let g : Y 0 → Y be a flat morphism. Fix notations by the following cartesian
diagram:
g0
X ×Y Y 0 → X
f0 ↓ ↓f
0 g
Y → Y.
For any quasi-coherent OX -module F on X and any p, we have a canonical

isomorphism
∼
g ∗ Rp f∗ F → Rp f∗0 g 0∗ F.
Proof. Let I · be an injective resolution of F and let J · be an injective res-

olution of g 0∗ F. Since g is flat, g 0 is also flat. So the functor g 0∗ is exact and
hence we have an exact sequence 0 → g 0∗ F → g 0∗ I · . By Proposition 1.1 (ii), we
have a morphism g 0∗ I · → J · unique up to homotopy extending idg0∗ F . It in-
duces a morphism f∗0 g 0∗ I · → f∗0 J · . On the other hand, the canonical morphism
I · → g∗0 g 0∗ I · induces a morphism f∗ I · → f∗ g∗0 g 0∗ I · = g∗ f∗0 g 0∗ I · . Since g ∗ is left
adjoint to g∗ , this morphism induces a morphism g ∗ f∗ I · → f∗0 g 0∗ I · . Composing
with the morphism f∗0 g 0∗ I 0 → f∗0 J · , we get a morphism g ∗ f∗ I · → f∗0 J · . Since
g is flat, g ∗ is an exact functor and hence the p-th cohomology sheaf of g ∗ f∗ I ·
is g ∗ Rp f∗ F. So the homomorphism on the p-th cohomology sheaves induced by
the morphism g ∗ f∗ I · → f∗0 J · is
g ∗ Rp f∗ F → Rp f∗0 g 0∗ F.
Let’s show it is an isomorphism. The problem is local with respect to Y and

Y 0 . So we may assume Y = SpecA and Y 0 = SpecA0 are affine. By Proposition
4.9, we have
Rp f∗ F = H p (X, F)∼ ,
p 0 0∗
R f∗ g F = H p (X ×Y Y 0 , g 0∗ F)∼ .
By Proposition 1.4.3 (ii), we then have
g ∗ Rp f∗ F = (A0 ⊗A H p (X, F))∼ .
To prove the proposition, it suffices to show the canonical homomorphism
A0 ⊗A H p (X, F) → H p (X ×Y Y 0 , g 0∗ F)
is an isomorphism. Let U = {Ui }i∈I be a covering of X by finitely many open

affine subschemes. Then U0 = {Ui ×Y Y 0 } is a covering of X ×Y Y 0 by affine
open subschemes. By Corollary 4.5, we have
H p (X, F) = Ȟ p (U, F),

H p (X ×Y Y 0 , g 0∗ F) = Ȟ p (U0 , g 0∗ F).
It is easy to see that
A0 ⊗A C · (U, F) ∼
= C · (U0 , g 0∗ F).
Taking p-th cohomology groups on both sides and using the fact that A0 is flat
over A, we get
A0 ⊗A H p (X, F) ∼
= H p (X ×Y Y 0 , g 0∗ F).
Proposition 4.11. Let A be a ring, let S be the ring A[x0 , . . . , xr ] with the
grading defined by the degrees of polynomials, and let X = ProjS.
(i) The canonical homomorphism
∞
M
S → Γ∗ (OX ) = H 0 (X, OX (n))
n=−∞
is an isomorphism.
∞
1
H r (X, OX (n)) is a free A-module with basis
L
(ii) k (ki > 0),
n=−∞ x0 0 ···xk
r
r
1
where k lies in H r (X, OX (−(k0 + · · · + kr ))).
x0 0 ···xk
r
r
∞
H p (X, OX (n)) = 0 for any p 6= 0, r.
L
(iii)
n=−∞
Proof. (i) is proved in Lemma 1.4.23. Let’s prove (ii) and (iii). Consider the
open covering U = {D+ (xi )|i = 0, . . . , r} of X. By Corollary 4.5, for any p, we
have
∞ ∞
∼
M M
p
H (X, OX (n)) = Ȟ p (U, OX (n)).
n=−∞ n=−∞
Consider the alternating Čech complex

∞
M ∞
M
C 0· (U, OX (n)) = C 0· (U, OX (n)).
n=−∞ n=−∞
We have
∞
M Y ∞
M
C 0p (U, OX (n)) = ( OX (n))(D+ (xi0 · · · xip ))
n=−∞ 0≤i0 <···<ip ≤r n=−∞
Y
= Sxi0 ···xip .
0≤i0 <···<ip ≤r
Note that
∞
M ∞
M ∞
M
C r−1 (U, OX (n)) → C r (U, OX (n)) → C r+1 (U, OX (n))
n=−∞ n=−∞ n=−∞

Y
Sxi0 ...xir−1 → Sx0 ...xr → 0.
0≤i0 <···<ir−1 ≤r
∞
H r (X, OX (n)) is the cokernel of the homomorphism
L
So
n=−∞
r
Y r
X
Sx0 ...x̂k ...xr → Sx0 ...xr , (gk ) 7→ (−1)k gk .
k=0 k=0
Note that Sx0 ...xr is a free A-module with basis xk00 · · · xkr r (ki ∈ Z) and the image
of the above homomorphism is the submodule generated by those xk00 · · · xkr r
∞
H r (X, OX (n)) is a free A-module with basis
L
with at least one ki ≥ 0. So
n=−∞
1
k (ki > 0). This proves (ii). On the other hand, we have
x0 0 ···xk
r
r
∞
(n)
M Y Y
C 0p (U, OX (n)) = Sxi0 ···ip = dir. lim Si0 ...ip ,
n
n=−∞ 0≤i0 <···<ip ≤r 0≤i0 <···<ip ≤r
Q (n)
where the direct system ( Si0 ...ip )n∈N is defined by
0≤i0 <···<ip ≤r
(n)
Y Y
Si0 ...ip = S
0≤i0 <···<ip ≤r 0≤i0 <···<ip ≤r
for any n and

(m) (n)
Y Y
Si0 ...ip → Si0 ...ip , (si0 ...ip ) 7→ ((xi0 · · · xip )n−m si0 ...ip )
0≤i0 <···<ip ≤r 0≤i0 <···<ip ≤r
(n)
for any m ≤ n. Let Cnp (S) =
Q
Si0 ...ip and define
0≤i0 <···<ip ≤r
d : Cnp (S) → Cnp+1 (S), s = (si0 ...ip ) 7→ ds = ((ds)i0 ...ip+1 )
by
p+1
X
(ds)i0 ...ip+1 = (−1)k xnik si0 ...îk ...ip+1 .
k=0
Each Cn· (S) is a complex and

∞
M
C 0· (U, OX (n)) = dir. lim Cn· (S).
n
n=−∞
So we have
∞
H p (X, OX (n)) ∼
M
= dir. lim H p (Cn· (S)).
n
n=−∞
Note that for any p ≥ 1, the sequence

d d
Cnp−1 (S) → Cnp (S) → Cnp+1 (S)
can be identified with the sequence

d d
K p (xn0 , . . . , xnr ; S) → K p+1 (xn0 , . . . , xnr ; S) → K p+2 (xn0 , . . . , xnr ; S).
So by Proposition 4.3, we have
H p (Cn· (S)) = H p+1 (xn0 , . . . , xnr ; S) = 0

∞
H p (X, OX (n)) = 0 for any p 6= 0, r.
L
when p 6= 0, r. Hence
n=−∞
Theorem 4.12 (Serre) Let A be a noetherian ring, X a scheme projective

over SpecA, OX (1) an invertible OX -module very ample over SpecA, and F a
coherent OX -module.
(i) (Finiteness Theorem) For any p, H p (X, F) is a finitely generated A-
module.
(ii) (Vanishing Theorem) There exists an integer n(F) depending on F such
that for any p ≥ 1 and any n ≥ n(F), we have H p (X, F(n)) = 0, where
F(n) = F ⊗OX OX (1)⊗n .
Proof. We can find a closed immersion i : X → PrA = ProjA[x0 , . . . , xr ] such

that i∗ OPrA (1) ∼
= OX (1). One can verify that i∗ is an exact functor and maps
injective sheaves on X to injective sheaves on PrA . So for any p, we have
H p (X, F) = H p (PrA , i∗ F).
Moreover we have i∗ (F(n)) ∼

= (i∗ F)(n) by Lemma 1.4.26. Hence we may assume
X = PrA . We prove the theorem by descending induction on p. By Proposition
1.4.25, we may find an epimorphism E → F such that E is a direct sum of finitely
many copies of OX (n) for some n. Let R be the kernel of this epimorphism.
0 → R → E → F → 0.
Since X can be covered by r + 1 affine open subschemes D+ (xi ) (i = 0, . . . , r),

we have H p (X, F(n)) = 0 for any p ≥ r + 1 and any n by Corollary 3.2 and
Corollary 4.5. In particular H p (X, F) is a finitely generated A-module and
H p (X, F(n)) = 0 for any n when p ≥ r + 1. Suppose p ≥ 0 and suppose for
any coherent OX -module F, we have shown H p+1 (X, F) is a finitely generated
A-module and there exists an integer n(F, p + 1) depending on F and p + 1 such
that H p+1 (X, F(n)) = 0 for any n ≥ n(F, p + 1). In particular, H p+1 (X, R) is
a finitely generated A-module and H p+1 (X, R(n)) = 0 for any n ≥ n(R, p + 1).
By Proposition 4.11, H p (X, E) is a finitely generated A-module. We have an
exact sequence
· · · → H p (X, E) → H p (X, F) → H p+1 (X, R) → · · · .
This implies H p (X, F) is a finitely generated A-module. By Proposition 4.11,

for any p ≥ 1, there exists an integer n(E, p) such that H p (X, E(n)) = 0 for
any n ≥ n(E, p). The long exact sequence of cohomology groups then shows
that H p (X, F(n)) = 0 for any n ≥ max(n(E, p), n(R, p + 1)). This shows that
for each p ≥ 1, there exists an integer n(F, p) such that H p (X, F(n)) = 0 for
any n ≥ n(F, p). Since H p (X, F(n)) 6= 0 only when 0 ≤ p ≤ r, if we take
n(F) = max n(F, p), then we have H p (X, F(n)) = 0 for any n ≥ n(F) and
1≤p≤r
any p ≥ 1.
2.5. COHOMOLOGICAL STUDY OF PROPER MORPHISMS 167
Corollary 4.13. Let f : X → Y be a projective morphism of noetherian

schemes, OX (1) an invertible OX -module very ample over Y , and F a coherent
OX -module.
(i) Rp f∗ F is a coherent OY -module for any p.
(ii) There exists an integer N such that for any n ≥ N , the canonical mor-
phism f ∗ f∗ F(n) → F(n) is surjective.
(iii) There exists an integer N such that Rp f∗ (F(n)) = 0 for any n ≥ N and
any p ≥ 1.
Proof. (i) By Proposition 4.9, Rp f∗ F is quasi-coherent and for any affine

open subscheme V = SpecA of Y , we have (Rp f∗ F)|V ∼ = H p (f −1 (V ), F)∼ . By
p −1
Theorem 4.12 (i), H (f (V ), F) is a finitely generated A-module. So Rp f∗ F
is coherent.
(ii) For any affine open subscheme V = SpecA of Y , by Proposition 1.4.25,
there exists an integer N , such that F(n)|f −1 (V ) is generated by sections in
F(n)(f −1 (V )) for any n ≥ N . Any section s in F(n)(f −1 (V )) can be regarded
as a section in (f∗ F(n))(V ). Denote by s0 its image under the canonical homo-
morphism
(f∗ F(n))(V ) → (f∗ f ∗ (f∗ F(n)))(V ) = f ∗ (f∗ F(n))(f −1 (V )).
One can verify that the image of s0 ∈ f ∗ f∗ F(n)(f −1 (V )) under the canonical
morphism f ∗ f∗ F(n) → F(n) is the original section s ∈ F(n)(f −1 (V )). So
(f ∗ f∗ F(n))|f −1 (V ) → F(n)|f −1 (V )
is surjective for any n ≥ N . Cover Y by finitely many affine open subschemes.

We see that there exists an integer N such that f ∗ f∗ F(n) → F(n) is surjective
for any n ≥ N .
(iii) For any affine open subscheme V = SpecA of Y , by Theorem 4.12 (ii),
there exists an integer N such that (Rp f∗ F(n))|f −1 (V ) = 0 for any n ≥ N and
any p ≥ 1. Cover Y by finitely many affine open subscheme. We see that there
exists an integer N such that Rp f∗ (F(n)) = 0 for any n ≥ N and any p ≥ 1.
2.5 Cohomological Study of Proper Morphisms
Let A be a noetherian ring, X a scheme proper over SpecA, and F a coherent

OX -module. In this section, we prove that H n (X, F) is a finitely generated
A-module for each n. Moreover, for any ideal I of A, we prove that
(H n (X, F))∧ ∼
= inv. lim H n (X, F/I k F).
k
where (H n (X, F))∧ = inv. limk H n (X, F)/I k H n (X, F) is the I-adic completion
of H p (X, F). We then use these results to obtain the Stein factorization of a
proper morphism, the Theorem of Connectedness and the Main Theorem of

Zariski.
Theorem 5.1. Let f : X → Y be a proper morphism of noetherian schemes.

Then for any coherent OX -module F and any n, Rn f∗ F is a coherent OY -
module.
Proof. We have seen Rn f∗ F is quasi-coherent in Proposition 4.9. We only need

to show it is of finite type. Let S be the family of those closed subsets Z of X
such that there exists a coherent OX -module F with suppF ⊂ Z but Rn f∗ F
is not coherent for some n. If S = ∅, we are done. Suppose S 6= ∅. Since X is
noetherian, S has a minimal element. Let Z be a minimal element in S, and let
F be a coherent OX -module with suppF ⊂ Z such that Rn f∗ F is not coherent
for some n.
Put the reduced induced closed subscheme structure on Z and let i : Z → X
be the closed immersion. We first show that we may choose F so that it is
of the form F = i∗ G for some coherent OZ -module G. Indeed, let I be the
ideal sheaf of the closed subscheme Z. We claim that there exists an m ≥ 0
such that Rn f∗ (I m F/I m+1 F) is not coherent for some n. We then take G =
i∗ (I m F/I m+1 F). It is a coherent OZ -module and i∗ G ∼ = I m F/I m+1 F. So
n ∼ n m
R f∗ (i∗ G) = R f∗ (I F/I m+1
F) is not coherent for some n. Let’s prove our
claim by contradiction. Suppose Rn f∗ (I m F/I m+1 F) is coherent for any m ≥ 0
and any n. Taking m = 0, we see Rn f∗ (F/IF) is coherent for any n. Suppose
we have proved Rn f∗ (F/I r F) is coherent for any n. We have an exact sequence
Rn f∗ (I r F/I r+1 F) → Rn f∗ (F/I r+1 F) → Rn f∗ (F/I r F).
This implies that Rn f∗ (F/I r+1 F) is coherent for any n. So Rn f∗ (F/I m F) is
coherent for any n and any m ≥ 0. Since suppF ⊂ Z, we have I k F = 0 for
some natural number k by Proposition 1.4.14. So Rn f∗ F = Rn f∗ (F/I k F) is
coherent for any n. This contradicts to our choice of F. So our claim holds.
If Z is not irreducible, then Z = Z1 ∪ Z2 for some proper closed subset Z1
and Z2 of Z. Put the reduced induced closed subscheme structures on Z1 and
on Z2 and let i1 : Z1 → Z and i2 : Z2 → Z be the closed immersions. Let
φ : i∗ G → i∗ i1∗ i1 ∗ G ⊕ i∗ i2∗ i2 ∗ G
be the morphism defined by the canonical morphisms G → i1∗ i1 ∗ G and G →
i2∗ i2 ∗ G. Note that φ is an isomorphism outside Z1 ∩ Z2 . So kerφ and cokerφ
are supported in Z1 ∩ Z2 . Since Z is a minimal element in S, the closed subsets
Z1 ∩ Z2 Z1 and Z2 are not in S. So the OY -modules Rn f∗ (kerφ), Rn f∗ (cokerφ),
Rn f∗ (i∗ i1∗ i1 ∗ G) and Rn f∗ (i∗ i2∗ i2 ∗ G) are coherent for any n. We have an exact
sequence
Rn−1 f∗ (cokerφ) → Rn f∗ (imφ) → Rn f∗ (i∗ i1∗ i1 ∗ G ⊕ i∗ i2∗ i2 ∗ G).
This implies that Rn f∗ (imφ) is coherent for any n. We have an exact sequence
Rn f∗ (kerφ) → Rn f∗ (i∗ G) → Rn f∗ (imφ).
This implies that Rn f∗ (i∗ G) = Rn f∗ F is coherent for any n. This contradicts

to our choice of F.
Suppose Z is irreducible. Let z be the generic point of Z. By Chow’s lemma
1.4.18, there exists a projective morphism g : Z 0 → Z such that f ig is also
projective, and there exists a dense open subset U of Z such that g induces
an isomorphism of g −1 (U ) with U . Note that z lies in U . Let OZ 0 (1) be
an invertible OZ 0 -module very ample over Z. By Corollary 4.13 (iii), we may
choose a sufficiently large k such that Rq g∗ OZ 0 (k) = 0 for any q ≥ 1. Then the
biregular spectral sequence
Rp (f i)∗ Rq g∗ OZ 0 (k) ⇒ Rp+q (f ig)∗ OZ 0 (k)
degenerates and hence
Rn (f i)∗ (g∗ OZ 0 (k)) ∼

= Rn (f ig)∗ OZ 0 (k)
for any n. Using the fact that i∗ is an exact functor and maps injective sheaves
on Z to injective sheaves on X, one can verify that
Rn f∗ (i∗ g∗ OZ 0 (k)) ∼
= Rn (f i)∗ (g∗ OZ 0 (k)).
By Corollary 4.13 (i), g∗ OZ 0 (k) and Rn (f ig)∗ OZ 0 (k) are coherent for any n.
So Rn f∗ (i∗ g∗ OZ 0 (k)) is coherent for any n. Since g induces an isomorphism
of g −1 (U ) with U , we have (g∗ OZ 0 (k))z 6= 0. Note that (g∗ OZ 0 (k))z and Gz
are finite dimensional vector spaces over the field OZ,z . So there exist natural
numbers l and m such that there exists an isomorphism of OZ,z -vector spaces
φz : (g∗ OZ 0 (k))lz ∼
= (Gz )m .
By Proposition 1.4.1 (ii), there exist an open neighborhood V of z in X and an

isomorphism
φV : (i∗ g∗ OZ 0 (k))l |V ∼
= (i∗ G)m |V
which induces the isomorphism φz on stalks. Let HV be the graph of φV , that
is, the image of the morphism

l l m
(i∗ g∗ OZ 0 (n)) |V → (i∗ g∗ OZ 0 (k)) ⊕ (i∗ G) |V , s 7→ (s, φV (s)).
Note that HV is a coherent OV -module. By Lemma 5.2 below, there exists a

coherent OX -submodule of (i∗ g∗ OZ 0 (k))l ⊕ (i∗ G)m such that H|V ∼ = HV and
H|X−Z = 0. Consider the projections p1 : H → (i∗ g∗ OZ 0 (k))l and p2 : H →
(i∗ G)m . Their kernels and cokernels are supported in Z − V . Since Z is a mini-
mal element in S, Rn f∗ (kerp1 ), Rn f∗ (cokerp1 ), Rn f∗ (kerp2 ) and Rn f∗ (cokerp2 )
are coherent for any n. Combining with the fact that Rn f∗ ((i∗ g∗ OZ 0 (k))l )
is coherent for any n, this implies that Rn f∗ ((i∗ G)m ) is coherent for any n.
So (Rn f∗ F)m = Rn f∗ ((i∗ G)m ) is coherent for any n. Since Rn f∗ F is quasi-
coherent, Rn f∗ F must be coherent for any n. This contradicts to our choice of
F.
In any case, the assumption S 6= ∅ leads to contradiction. So S = ∅.
Lemma 5.2. Let X be a noetherian scheme, U an open subscheme of X, F

a quasi-coherent OX -module and GU a coherent OU -submodule of F|U . Then
there exists a coherent OX -submodule G of F such that G|U = GU .
Proof. First consider the case where X = SpecA is affine. Let j : U → X be the
open immersion. Since GU is a submodule of j ∗ F = F|U , j∗ GU is a submodule
of j∗ j ∗ F. Let G 0 be the inverse image of j∗ GU under the canonical morphism
F → j∗ j ∗ F. Since F, j∗ j ∗ F and j∗ GU are quasi-coherent by Proposition 1.4.9,
G 0 is quasi-coherent. Note that j ∗ G 0 is the inverse image of j ∗ j∗ GU = GU under
the identity morphism j ∗ F → j ∗ (j∗ j ∗ F) = j ∗ F. So j ∗ G 0 = G|U . Assume
G 0 = M ∼ for some A-module M . We have M = dir. limn Mn , where the direct
limit is taken over the direct system of finitely generated A-submodules Mn of
M . Then G 0 = dir. limn Gn , where each Gn = Mn∼ is a coherent OX -submodule
of F. Note that GU = j ∗ G 0 = dir. limn j ∗ Gn . We claim that GU = j ∗ Gn for
some n. This would prove the lemma in the case where X is affine. Cover
U by finitely many affine open subschemes Ui = SpecAi (i ∈ I). Since GU is
a coherent OU -module, we have GU |Ui = Ni∼ for some finitely generated Ai -
modules Ni . We have GU |Ui = dir. limn Gn |Ui = dir. limn (Ai ⊗A Mn )∼ . So
Ni ∼ = dir. limn Ai ⊗A Mn . Since each Ni is a finitely generated Ai -module, there
exists an Mn such that Ni ∼ = Ai ⊗A Mn for all i. We then have GU = j ∗ Gn .
In general, we cover X by finitely many affine open subschemes U1 , . . . , Un .
Let Vi = U ∪ U1 ∪ · · · ∪ Ui (1 ≤ i ≤ n). We will construct coherent OVi -
submodules GVi of F|Vi by induction on i such that GVi |U = GU . Then G = GVn
would have the required property. By the affine case treated above, there exists
a coherent OU1 -submodule HU1 of F|U1 such that HU1 |U ∩U1 = GU |U ∩U1 . We
may glue HU1 and GU together to get a sheaf GV1 on V1 = U ∪ U1 . Suppose we
have constructed GVi . By the affine case treated above, we may find a coherent
OUi+1 -submodule HUi+1 of F|Ui+1 such that HUi+1 |Vi ∩Ui+1 = GVi |Vi ∩Ui+1 . We
can glue HUi+1 and GVi together to get a sheaf GVi+1 on Vi ∪ Ui+1 = Vi+1 .
Corollary 5.3. (Finiteness Theorem) Let A be a noetherian ring, X a scheme

proper over SpecA, and F a coherent OX -module. Then H n (X, F) is a finitely
generated A-module for any n.
Proof. Follows from Theorem 5.1 and Proposition 4.9.
Corollary 5.4.
(i) Let X be a noetherian scheme and let F and G be coherent OX -modules.
Then ExtnOX (F, G) is a coherent OX -module for any n.
(ii) Let A be a noetherian ring, X a scheme proper over SpecA, and F and
G coherent OX -modules. Then ExtnOX (F, G) is a finitely generated A-module
for any n.
Proof. (i) The problem is local. We may assume X = SpecA for some noethe-
rian ring A. Then F ∼
= M ∼ for some finitely generated A-module. Since M has
a resolution by free A-modules of finite ranks, we can find a resolution
· · · → E1 → E0 → F → 0
of F by free OX -modules of finite ranks. By Proposition 2.16, we have
ExtnOX (F, G) = H n (HomOX (E· , G)).
Note that each HomOX (Ei , G) is coherent. So ExtnOX (F, G) is also coherent.
H p (X, ExtqOX (F, G)) ⇒ ExtnOX (F, G).
By Corollary 5.3, each H p (X, ExtqOX (F, G)) is a finitely generated A-module.
∞
OX 0 -module F 0 such that g∗0 F 0 ∼ I k F. Note that F 0 is actually coherent.
L
=
k=0
Since g 0 is affine, the biregular spectral sequence
H p (X, Rq g∗0 F 0 ) ⇒ H p+q (X 0 , F 0 )
degenerates by Corollary 4.8. So we have

∞
M ∞
M
H n (X 0 , F 0 ) = H n (X, g∗0 F 0 ) = H n (X, I k F) = H n (X, I k F).
k=0 k=0
∞
By Corollary 5.3, H n (X 0 , F 0 ) is a finitely generated S-module. So H n (X, I k F)
L
k=0
is a finitely generated graded S-module.
Theorem 5.6. Let A be a noetherian ring, I an ideal of A, X a scheme proper

over SpecA, and F a coherent OX -module. Then for every n, we have
(H n (X, F))∧ ∼
= inv. lim H n (X, F/I k F),
k
where (H n (X, F))∧ = inv. limk H n (X, F)/I k H n (X, F) is the I-adic completion
of the A-module H n (X, F).
Proof. Obviously I k H n (X, F/I k F) = 0 for any n and any k. So the canonical
homomorphism H n (X, F) → H n (X, F/I k F) induces a homomorphism
H n (X, F)/I k H n (X, F) → H n (X, F/I k F).
Passing to inverse limit, we get a homomorphism
(H n (X, F))∧ → inv. lim H n (X, F/I k F)

k
for each n. Let’s show it is an isomorphism.

We have a long exact sequence
· · · → H n (X, I k F) → H n (X, F) → H n (X, F/I k F)

→ H n+1 (X, I k F) → H n+1 (X, F) → · · · .
Let
Rk = im(H n (X, I k F) → H n (X, F)),

Qk = ker(H n+1 (X, I k F) → H n+1 (X, F))
= im(H n (X, F/I k F) → H n+1 (X, I k F)).
Then we an exact sequence
0 → H n (X, F)/Rk → H n (X, F/I k F) → Qk → 0.

Passing to inverse limit and applying Proposition 1.5.1, we get an exact sequence
0 → inv. lim H n (X, F)/Rk → inv. lim H n (X, F/I k F) → inv. lim Qk → 0.
k k k
By Lemma 5.7 below, we have inv. limk H n (X, F)/Rk ∼ = (H n (X, F))∧ . By
Lemma 5.8 below, we have inv. limk Qk = 0. So we have
(H n (X, F))∧ ∼
k
Lemma 5.7. For any n, the decreasing filtration on H n (X, F) given by Rk

is I-stable, that is, I l Rk ⊂ Rk+l for any k, l ≥ 0, and there exists a natural
number k0 such that I l Rk = Rk+l for any k ≥ k0 and l ≥ 0. Moreover, we have
(H n (X, F))∧ ∼
= inv. lim H n (X, F)/Rk .
k
Proof. For any a ∈ I l , we have a commutative diagram

a
IkF → I k+l F
↓ ↓
a
F → F,
where the horizontal arrows are induced by multiplications by a and the vertical
arrows are inclusions. This diagram induces the following:
a
H n (X, I k F) → H n (X, I k+l F)
↓ ↓
n a n
H (X, F) → H (X, F).
So the multiplication by a maps the image of the first vertical arrow to the image
∞
of the second vertical arrow, that is, I l Rk ⊂ Rk+l . This makes
L
Rk a graded
k=0
∞ ∞
k
H n (X, I k F). By
L L
S= I -module. It is a quotient of the graded S-module
k=0 k=0
∞ ∞
H n (X, I k F) is finitely graded. So
L L
Lemma 5.5, S is noetherian and Rk is
k=0 k=0
a finitely generated graded S-module. Let xi (i = 1, . . . , m) be a finite family of
∞
L
homogeneous elements generating Rk and let k0 = max(degx1 , . . . , degxm ).
k=0
One can verify Rk0 +l = I l Rk0 for any l ≥ 0. When k ≥ k0 , we have
Rk = I k−k0 Rk0 ⊂ I k−k0 H n (X, F).
Moreover, for any k ≥ 0, we have
I k H n (X, F) = I k R0 ⊂ Rk .
So
(H n (X, F))∧ = inv. lim H n (X, F)/I k H n (X, F) ∼
= inv. lim H n (X, F)/Rk .
k k
Lemma 5.8. There exists an integer k0 such that the canonical homomorphism
Qk+k0 → Qk vanishes for any k ≥ k0 .
Proof. For any a ∈ I l , we have a commutative diagram

a
H n+1 (X, I k F) → H n+1 (X, I k+l F)
↓ ↓
n+1 a n+1
H (X, F) → H (X, F),
where the horizontal arrow are induced by multiplications by a. So the multi-
plication by a maps the kernel of the first vertical arrow to the kernel of the
∞
second vertical arrow, that is, I l Qk ⊂ Qk+l . This makes
L
Qk a graded S =
k=0
∞ ∞
k n+1 k
L L
I -submodule of H (X, I F). By Lemma 5.5, S is noetherian and
k=0 k=0
∞ ∞
H n+1 (X, I k F) is finitely generated. So
L L
Qk is a finitely generated graded
k=0 k=0
S-module. Let x1 , . . . , xm be a finite family of homogeneous elements generating
∞
L
Qk and let k0 = max(degx1 , . . . , degxm ). One can verify Qk+l = Sl Qk for
k=0
any k ≥ k0 and l ≥ 0. On the other hand, since the A-module H n (X, F/I k F)
is annihilated by I k and Qk = im(H n (X, F/I k F) → H n+1 (X, I k F)), the A-
module Qk is also annihilated by I k , that is, we have
αk (Sk )Qk = 0
for any k ≥ 0, where α : Sk → S0 is the inclusion I k ,→ A. This implies
αk0 (Sk0 )xi = 0 (i = 1, . . . , m). So
αk0 (Sk0 )Q = 0.
For any k ≥ k0 , since Qk+k0 = Sk0 Qk , the image of Qk+k0 → Qk is αk0 (Sk0 )Qk .
So Qk+k0 → Qk vanishes for any k ≥ k0 .
Corollary 5.9. Let f : X → Y be a proper morphism of noetherian schemes,

F a coherent OX -module, and I a coherent sheaf of ideals of OY . Then for
each n, we have a canonical isomorphism
inv. lim Rn f∗ F/I k Rn f∗ F ∼
= inv. lim Rn f∗ (F/I k F).
k k
Proof. Note that the canonical morphism Rn f∗ F → Rn f∗ (F/I k F) induces

a morphism Rn f∗ F/I k Rn f∗ F → Rn f∗ (F/I k F) for any n and k. Passing to
inverse limit, we get a morphism
inv. lim Rn f∗ F/I k Rn f∗ F → inv. lim Rn f∗ (F/I k F)
k k
for each n. Let’s show it is an isomorphism. It suffices to show
inv. lim(Rn f∗ F/I k Rn f∗ F)(V ) → inv. lim Rn f∗ (F/I k F)(V )

k k
is an isomorphism for any affine open subscheme V = SpecA of Y . Let I be the

ideal of A corresponding to I. Then by Proposition 4.9, we have
(Rn f∗ F/I k Rn f∗ F)(V ) ∼

= H n (f −1 (V ), F)/I k H n (f −1 (V ), F),
Rn f∗ (F/I k F)(V ) ∼
= H n (f −1 (V ), F/I k F).
Our assertion then follows from Theorem 5.6.
Theorem 5.10. (Theorem of Connectedness) Let f : X → Y be a proper

morphism of noetherian schemes. Suppose f ] : OY → f∗ OX is an isomorphism.
Then for any y ∈ Y , f −1 (y) is nonempty and connected.
Proof. First note that f is surjective and hence f −1 (y) is nonempty. Indeed,
since f is proper, f (X) is closed. Suppose f (X) 6= Y and let y ∈ Y − f (X). We
have (f∗ OX )y = 0. But (f∗ OX )y ∼ = OY,y 6= 0. Contradiction. So we must have
f (X) = Y .
We have a canonical flat morphism SpecOY,y → Y . Indeed, let V = SpecA
be an affine open neighborhood of y in Y and let p be the prime ideal of A
corresponding to y. Then it is the composition
SpecOY,y = SpecAp → SpecA = V ,→ Y.
The composition is independent of the choice of the affine open neighborhood

V of y. Fix notations by the following cartesian diagrams:
i i
Xk = X ×Y SpecOY,y /mky →
k
X 0 = X ×Y SpecOY,y → X
fk ↓ f0 ↓ f↓
j
SpecOY,y /mky → SpecOY,y → Y,
where my is the maximal ideal of OY,y . By Proposition 4.10, we have
f∗0 OX 0 ∼
= f∗0 i∗ OX ∼
= j ∗ f∗ OX ∼
= j ∗ OY ∼
= OSpecOY,y .
So we have
H 0 (X 0 , OX 0 ) ∼
= Γ(SpecOY,y , f∗0 OX 0 ) ∼
= Γ(SpecOY,y , OSpecOY,y ) ∼
= OY,y .
Taking my -adic completions, we get
(H 0 (X 0 , OX 0 ))∧ ∼
=ObY,y .
Combining with Theorem 5.6, we get
inv. lim H 0 (X 0 , OX 0 /mky OX 0 ) ∼

=ObY,y .
k
It is easy to see that OX 0 /mky OX 0 = ik∗ OXk . So we have
H 0 (X 0 , OX 0 /mky OX 0 ) ∼
= H 0 (Xk , OXk ).
Therefore
inv. lim H 0 (Xk , OXk ) ∼
=ObY,y .
k
0
In particular, inv. limk H (Xk , OXk ) is a local ring. Note that Xk has the same
underlying topological space as f −1 (y) for any k. Suppose f −1 (y) is not con-
nected. Then f −1 (y) is a union of two disjoint open subsets U1 and U2 . We
have
inv. lim H 0 (Xk , OXk ) = inv. lim H 0 (U1 , OXk ) ⊕ inv. lim H 0 (U2 , OXk ).
k k k
The elements (1, 0) and (0, 1) in inv. limk H 0 (U1 , OXk ) ⊕ inv. limk H 0 (U2 , OXk )
are not unit and their sum is a unit. So inv. limk H 0 (Xk , OXk ) can not be a
local ring. Contradiction. So f −1 (y) is connected.
Let f : X → Y be a proper morphism between noetherian schemes. By

Theorem 5.1, f∗ OX is a coherent OY -module. It is a sheaf of OY -algebras.
By Proposition 1.4.12, there exists an affine morphism g : Y 0 → Y such that
g∗ OY 0 ∼
= f∗ OX . By Proposition 1.4.11, the isomorphism g∗ OY 0 ∼
= f∗ OX induces
a morphism f 0 : X → Y 0 such that f can be factorized as
f0 g
X → Y 0 → Y.
We call it the Stein factorization of f . Note that g is a finite morphism. By

]
Proposition 1.4.10 (ii), f 0 : OY 0 → f∗0 OX is an isomorphism. So all the fibers
0
of f are connected by Theorem 5.10.
A morphism f : X → Y is called quasi-finite if it is of finite type and for any

y ∈ Y , f −1 (y) is finite as a set. Applying Lemma 5.11 below to the projection
f −1 (y) = X ×Y Speck(y) → Speck(y), where k(y) is the residue field of the
local ring OY,y , we see that a morphism f : X → Y of finite type is quasi-finite
if and only if f −1 (y) is discrete for any y ∈ Y .
Lemma 5.11. Let k be a field and let X → Speck be a morphism of finite

type. The following conditions are equivalent:
(i) The underlying topological space of X is finite as a set.
(ii) The underlying topological space of X is discrete.
(iii) There exists an artinian k-algebra A such that X ∼= SpecA.
Proof. (i)⇒(ii) For any x ∈ X, let U = SpecA be an affine open neighborhood

of x such that A is a finitely generated k-algebra. One can show A is a Jacobson
ring, that is, any prime ideal of A is an intersection of some maximal ideals.
(Confer Exercise 5.24 of [Atiyah-Macdonald].) Since the underlying space of X
T
is a finite set, A has only finitely maximal ideals, say m1 , . . . , mn . If p = mij
Q j
is a prime ideal, then we have p ⊃ mij and hence p ⊃ mij for some ij . On
T j
the other hand, we have p = mij ⊂ mij . So we have p = mij . So all the prime
j
ideals of A are maximal. Therefore all the points in the finite set U are closed.
This shows that U is discrete and hence X is discrete.
(ii)⇒(iii) Since X is of finite type over Speck, it is quasi-compact. Since the
underlying topological space of X is discrete, X has only finitely many points,
say x1 , . . . , xn , and each {xi } is an open subset of X. One can verify that
Qn n
Q
X = Spec( OX,xi ) and OX,xi is an artinian ring.
i=1 i=1
(iii)⇒(i) Follows from Propositions 8.1 and 8.3 in [Atiyah-Macdonald].
Proposition 5.12. Let f : X → Y be a morphism of noetherian schemes. The

following conditions are equivalent:
(i) f is finite.
(ii) f is proper and affine.
(iii) f is proper and quasi-finite.
Proof. Suppose f is finite. Then f is affine by definition. This implies that f is

separated. For any y ∈ Y , the projection f −1 (y) = X ×Y Speck(y) → Speck(y)
is a finite morphism and hence X ×Y Speck(y) = SpecA for some k(y)-algebra
A which is finitely generated as a k(y)-module. But then A is artinian and
hence f is quasi-finite by Lemma 5.11. Let’s prove f is a closed map on the
underlying topological spaces. This is a local problem with respect to Y . So
we may assume Y = SpecB is affine. Then X = SpecC for some B-algebra C
which is finitely generated as a B-module. Let φ : B → C be the homomorphism
corresponding to f . For any ideal c of C, using [Atiyah-Macdonald] Theorem
5.10, one can show f (V (c)) = V (φ−1 (c)). Hence f is a closed morphism. Since
any base change of a finite morphism is finite, f is universally closed. So f is
proper. This proves (i)⇒(ii) and (i)⇒(iii).
f0 g
Suppose f is proper. Let X → Y 0 → Y be the Stein factorization of f . If f is
affine, then f 0 is necessarily an isomorphism. So f can be identified with g and
hence is a finite morphism. This proves (ii)⇒(i). If f is quasi-finite, then for
any y 0 ∈ Y 0 , f −1 (g(y 0 )) is discrete and hence f 0−1 (y 0 ) ⊂ f −1 (g(y 0 )) is discrete.
But f 0−1 (y 0 ) is nonempty and connected. So f −1 (y 0 ) consists of only one point.
Hence f 0 is one-to-one and onto. Moreover, f 0 is proper by Proposition 1.3.26
(vi), so it is a homeomorphism on the underlying topological spaces. Combining
]
with the fact that f 0 : OY 0 → f∗0 OX is an isomorphism, we see that f 0 is an
isomorphism of schemes. So f can be identified with g and hence is finite. This
proves (iii)⇒(i).
Proposition 5.13. Let f : X → Y be a proper morphism of noetherian

schemes and let X 0 be the subset of X consisting of those points x which are
f0 g
isolated in f −1 (f (x)). Then X 0 is open in X. Let X → Y 0 → Y be the Stein
factorization of f . Then f 0 |X 0 : X 0 → Y 0 is an open immersion.
Proof. Let x be a point in X 0 . Since f 0−1 (f 0 (x)) ⊂ f −1 (f (x)), x is isolated

in f 0−1 (f 0 (x)). But f 0−1 (f 0 (x)) is connected, so we have f 0−1 (f 0 (x)) = {x}.
Note that f 0 is proper by Proposition 1.3.26 (vi). So for any open neighborhood
W of x in X, f 0 (X − W ) is closed. Since f 0−1 (f 0 (x)) = {x}, f 0 (x) lies in
Y 0 − f 0 (X − W ). Obviously f 0−1 (Y 0 − f 0 (X − W )) ⊂ W . This shows that
for any open neighborhood W of x in X, there exists an open neighborhood
W 0 = Y 0 − f 0 (X − W ) of f 0 (x) in Y 0 such that f 0−1 (W 0 ) ⊂ W . This implies
that
(f∗0 OX )f 0 (x) ∼
= OX,x .
Since OY 0 = ∼ f OX , we have
0
∗
OY 0 ,f 0 (x) ∼
= OX,x .
By Proposition 1.3.13 (iii), there exists a neighborhood Wx of x in X such that
f 0 |Wx : Wx → Y 0 is an open immersion. For any P ∈ Wx , since g is a finite
morphism, f 0 (P ) is isolated in g −1 (f (P )). So we may find an open neighborhood
Wf0 0 (P ) of f 0 (P ) in Y 0 such that Wf0 0 (P ) ∩ g −1 (f (P )) = {f 0 (P )}. We then have
Wx ∩ f 0−1 (Wf0 0 (P ) ) ∩ f −1 (f (P )) = {P }.
But Wx ∩ f 0−1 (Wf0 0 (P ) ) is a neighborhood of P . So P is isolated in f −1 (f (P )).

Hence Wx ⊂ X 0 . This shows X 0 is open in X. Moreover, there exists an open
covering {Wx }x∈X 0 of X 0 such that each f 0 |Wx : Wx → Y 0 is an open immersion.
Since for any x ∈ X 0 , we have f 0−1 (f 0 (x)) = {x}, f |X 0 is one-to-one. So f 0 |X 0
is an open immersion.
Theorem 5.14. (Main Theorem) Let f : X → Y be a quasi-projective quasi-

finite morphism of noetherian schemes. Then we can factorize f as
j f¯
X→X→Y
such that j is an open immersion and f¯ is finite.
Proof. By the definition of quasi-projective morphism and Corollary 1.4.17, we

can factorize f as
j0 0 f0
X ,→ X → Y
such that j 0 is an open immersion and f 0 is projective. Consider the Stein
factorization
0 f 00
X →X→Y
of f 0 . Since f is quasi-finite, any x ∈ X is isolated in f 0−1 (f 0 (x)). By Proposition
5.13, f 00 |X : X → X is an open immersion. The factorization
f 00 |X
X → X→Y
of f has the required property.
Corollary 5.15. Let Y be a noetherian integral scheme such that for any y ∈ Y ,
the integral domain OY,y is integrally closed in its fraction field, (schemes with
this property are called normal), and let X be a noetherian integral scheme.
Then any separated quasi-finite birational morphism f : X → Y must be an
open immersion.
Proof. We first show that locally f is an open immersion. This is a local

question. We may assume X = SpecB and Y = SpecA are affine, where B is a
finitely generated A-algebra. Let b1 , . . . , bm be a finite family of generators for
B. Consider the A-algebra homomorphism
A[x1 , . . . , xm ] → B, xi 7→ bi .
It is onto and hence induces a closed immersion X = SpecB → SpecA[x1 , . . . , xm ].
Since we have an isomorphism
xi
A[x0 , . . . , xm ](x0 ) → A[x1 , . . . , xm ], 7→ xi ,
x0
we have an open immersion SpecA[x1 , . . . , xm ] = ∼ D+ (x0 ) → Pm . So we have
A
m
an immersion X → PA . This shows that X → Y is quasi-projective. By the
Main Theorem 5.14, f can be factorized as
j f¯
X→X→Y
such that j is an open immersion and f¯ is finite. Replacing X by the scheme
theoretic image of j, we may assume X is integral and X is dense in X. Since
f¯ is finite and birational and Y is normal, f¯ must be an isomorphism. (Confer
[Atiyah-Macdonald] Chapter 5.) So f can be identified with j and hence is an
open immersion.
To prove f is an open immersion, we only need to show f is injective as a
map on the underlying topological spaces. Suppose f (x1 ) = f (x2 ) = y. Let ξ be
the generic point of X and let η be the generic point of Y . Since f is birational,
we have f (ξ) = η and fξ] : OY,η → OX,ξ is an isomorphism. Note that since
xi ∈ {ξ} for each i ∈ {1, 2}, any neighborhood U of xi is also a neighborhood
of ξ and hence we have a homomorphism OX (U ) → OX,ξ . Passing to direct
limit, we get a homomorphism OX,xi = dir. limxi ∈U OX (U ) → OX,ξ . Since X is
integral, this homomorphism identifies OX,ξ with the fraction field of OX,xi and
hence is a monomorphism. Similarly, we have a monomorphism OY,y → OY,η .
The following diagram commutes:
OY,y → OY,η
fx]i ↓ ↓ fξ]
OX,xi → OX,ξ .
Since locally f is an open immersion, the vertical arrows are isomorphisms.
Since the horizontal arrows are monomorphisms, we must have OX,x1 = OX,x2 .
Here we regard OX,xi (i = 1, 2) as subrings of OX,ξ through the monomorphism

OX,xi → OX,ξ . Let Ui (i = 1, 2) be affine open neighborhoods of xi so that f (Ui )
are contained in the same affine open neighborhood of y in Y . By Proposition
1.3.25 (i), U1 ∩ U2 is affine and the ring OX (U1 ∩ U2 ) is generated by the images
of the restrictions OX (U1 ) → OX (U1 ∩ U2 ) and OX (U2 ) → OX (U1 ∩ U2 ). We
OX (U1 )
↓ &
OX (U1 ∩ U2 ) → OX,ξ ,
↑ %
OX (U2 )
where all the arrows are monomorphisms. So we may regard OX (U1 ), OX (U2 )
and OX (U1 ∩ U2 ) as subrings of OX,ξ . Then OX (U1 ∩ U2 ) is the subring of OX,ξ
generated by OX (U1 ) and OX (U2 ). We have a commutative diagram
OX (U1 )
↓ &
OX,x1 = OX,x2 → OX,ξ ,
↑ %
OX (U2 )
where all the arrows are monomorphisms. So if we regard OX,x1 = OX,x2 as

a subring of OX,ξ , it contains the subring of OX,ξ generated by OX (U1 ) and
OX (U2 ), that is,
OX (U1 ∩ U2 ) ⊂ OX,x1 = OX,x2 .
Let mx be the maximal ideal of OX,x1 = OX,x2 . Then mx ∩ OX (U1 ∩ U2 ) is a
prime ideal of OX (U1 ∩ U2 ). Since U1 ∩ U2 is affine, it corresponds to a point
x ∈ U1 ∩ U2 . For each i, the prime ideal mx ∩ OX (Ui ) of OX (Ui ) corresponds
to the point xi in the affine scheme Ui . Under the inclusion U1 ∩ U2 ,→ Ui , x is
mapped to xi . So we have x1 = x2 = x and hence f is injective.
2.6 Local Freeness of Higher Direct Images
Let f : X → Y be a morphism and F a quasi-coherent OX -module. We

say F is flat over Y if for any x ∈ X, Fx is flat as an OY,f (x) -module. By
Proposition 1.19, this is equivalent to saying that for any affine open subscheme
U of X and any affine open subscheme V of Y such that f (U ) ⊂ V , F(U )
is flat as an OY (V )-module. For any y ∈ Y , let k(y) be the residue field of
the local ring OY,y , let Xy = f −1 (y) = X ×Y Speck(y), and let Fy = p∗ F,
where p : X ×Y Speck(y) → X is the projection. In this section, we study
how H n (Xy , Fy ) changes when y goes over Y under the assumption that f is
2.6. LOCAL FREENESS OF HIGHER DIRECT IMAGES 181
a proper morphism of noetherian schemes and F is a coherent OX -module flat

over Y . Moreover, we obtain some criteria for Ri f∗ F to be locally free.
Lemma 6.1. Let A be a ring, K · and C · two complexes of flat A-modules

such that K i = C i = 0 for large i, and φ : K · → C · a morphism inducing
isomorphisms on cohomology groups. Then for any A-module M , the morphism
φ ⊗ idM : K · ⊗A M → C · ⊗A M induces isomorphisms on cohomology groups.
Proof. For any morphism φ : K · → C · of complexes, define a complex L· as

follows: For each i, let
Li = C i ⊕ K i+1
and let di : Li → Li+1 be the homomorphism
C i ⊕ K i+1 → C i+1 ⊕ K i+2 , (ci , ki+1 ) 7→ (d(ci ) + φ(ki+1 ), −d(ki+1 )).
One can verify that dd = 0 and there exists a long exact sequence
· · · → H i (K · ) → H i (C · ) → H i (L· ) → H i+1 (K · ) → · · ·
We call L· the mapping cylinder of the morphism φ : K · → C · . Note that

L· ⊗A M is the mapping cylinder of φ⊗idM . Suppose φ induces isomorphisms on
cohomology groups. The above long exact sequence then shows that H i (L· ) = 0
for any i. Suppose furthermore that K i = C i = 0 for any i > n. Then Li = 0
for any i > n. We have short exact sequences
0 → Z n−1 (L· ) → Ln−1 → Ln → 0,

0 → Z n−2 (L· ) → Ln−2 → Z n−1
(L· ) → 0,
0 → Z n−3 (L· ) → Ln−3 → Z n−2 (L· ) → 0,
..
.
Suppose furthermore that K · and C · are complexes of flat A-modules. Then

L· is also a complex of flat A-modules. For any A-module M , the long exact
sequences of Tori (−, M ) associated to the above short exact sequences show
successively that Tor1 (Z n−1 (L· ), M ) = 0, Tor1 (M, Z n−2 (L· )) = 0, . . . and that
the following sequences are exact:
0 → Z n−1 (L· ) ⊗A M → Ln−1 ⊗A M → Ln ⊗A M → 0,

0 → Z n−2 (L· ) ⊗A M → Ln−2 ⊗A M → Z n−1
(L· ) ⊗A M → 0,
0 → Z n−3 (L· ) ⊗A M → Ln−3 ⊗A M → Z n−2
(L· ) ⊗A M → 0,
..
.
So
· · · → Li ⊗A M → Li+1 ⊗A M → · · ·
is exact. Hence H i (L· ⊗A M ) = 0 for any i. The long exact sequence
· · · → H i (K · ⊗A M ) → H i (C · ⊗A M ) → H i (L· ⊗A M ) → H i+1 (K · ⊗A M ) → · · ·
then shows that φ ⊗ idM : K · ⊗A M → C · ⊗A M induces isomorphisms on

cohomology groups.
Lemma 6.2. Let A be a ring, K · and C · two complexes of flat A-modules

such that K i = C i = 0 for large i, and φ : K · → C · a morphism inducing
isomorphisms on cohomology groups. Suppose C i = 0 for any i < 0. Then
K 0 /B 0 (K · ) is a flat A-module and the morphism of complexes
0 → K 0 /B 0 (K · ) → K 1 → K2 → ···
↓ ↓ ↓
0 → C0 → C1 → C2 → ···
induced by φ induces isomorphisms on cohomology groups.
Proof. Since C i = 0 for any i < 0 and H i (K · ) ∼

= H i (C · ), we have H i (K · ) = 0
for any i < 0 and hence
· · · → K −1 → K 0 → K 0 /B 0 (K · ) → 0
is a resolution of K 0 /B 0 (K · ) by flat A-modules. So for any an A-module M ,

we have
TorA 0 0 · ∼ −i ·
i (K /B (K ), M ) = H (K ⊗A M )
for any i ≥ 0. By Lemma 6.1, we have
H −i (K · ⊗A M ) ∼
= H −i (C · ⊗A M ).
Combining with the assumption that C i = 0 for any i < 0, we get
H −i (K · ⊗A M ) = 0
for any i ≥ 1. So
·
TorA 0 0
i (K /B (K ), M ) = 0
for any i ≥ 1 and any A-module M . Hence K 0 /B 0 (K · ) is flat. The other

assertion in the lemma is obvious.
Lemma 6.3. Let A be a noetherian ring and let C · be a complex of flat A-

modules such that each H i (C · ) is a finitely generated A-module and C i 6= 0
only when 0 ≤ i ≤ n. Then there exist a complex K · of finitely generated
projective A-modules and a morphism φ : K · → C · such that K i 6= 0 only when
0 ≤ i ≤ n, and for any A-module M , φ ⊗ idM : K · ⊗A M → C · ⊗A M induces
isomorphisms on cohomology groups.
Proof. For any i > n, we set K i = 0. Suppose we have defined K i and

K i → C i for any i > p such that each K i is a free A-module of finite rank,
H i (K · ) ∼
= H i (C · ) for any i > p + 1, and Z p+1 (K · ) → H p+1 (C · ) is surjective.
Consider the following diagram:
α β
A → B → Z p+1 (K · )
↓ ↓ ↓
Cp → C p /B p (C · ) → Z p+1 (C · ),
where the first row is the pulling back of the second row by Z p+1 (K · ) →
Z p+1 (C · ), that is,
B = {(c, k) ∈ C p/B p(C · )⊕Z p+1(K · )|c and k have the same image in Z p+1 (C · )},
A = {(c, k) ∈ C p ⊕ Z p+1 (K · )|c and k have the same image in Z p+1 (C · )}.
One can verify that the third vertical arrow in the above diagram induces a
monomorphism cokerβ → H p+1 (C · ). By our induction hypothesis, it is sur-
jective and hence is an isomorphism. The second vertical arrow induces an
isomorphism kerβ ∼= H p (C · ). So we have an exact sequence
β
0 → H p (C · ) → B → Z p+1 (K · ) → H p+1 (C · ) → 0.
Since H p (C · ), H p+1 (C · ) and K p+1 are finitely generated A-modules, B is also
finitely generated. One can verify α is surjective. Let a1 , . . . , ar be elements
in A such that α(a1 ), . . . , α(ar ) generate B, let K p be a free A-module with
rank r, and let γ : K p → A be a homomorphism which maps a basis of K p to
{a1 , . . . , ar }. Then αγ is surjective. This implies that
H p+1 (K · ) = cokerβ ∼
= H p+1 (C · )
and
αγ
Z p (K · ) = ker(βαγ) → kerβ ∼
= H p (C · )
is surjective. In this way, we get a complex K · of free A-modules of finite
ranks and a morphism K · → C · inducing isomorphisms on cohomology groups.
Moreover, we have K i 6= 0 only when i ≤ n. To get the complex with the
required property, we replace K · by
0 → K 0 /B 0 (K · ) → K 1 → · · · → K n → 0.
By Lemma 6.2, K 0 /B 0 (K · ) is a flat A-module. It has finite presentation since
K 0 and K −1 are free A-modules with finite ranks. So by Proposition 1.18,
K 0 /B 0 (K · ) is a projective A-modules. By Lemma 6.1 and 6.2, the complex
0 → K 0 /B 0 (K · ) → K 1 → · · · → K n → 0
has the required property.
Proposition 6.4. Let A be a noetherian ring, Y = SpecA, f : X → Y a

proper morphism, and F a coherent OX -module flat over Y . Then there exists
a complex K · of the form
0 → K0 → K1 → · · · → Kn → 0
such that for each i, K i is a finitely generated projective A-module, and for any
A-module M , there exists an isomorphism
H i (X, F ⊗A M ) ∼
= H i (K · ⊗A M )
functorial in M , where F ⊗A M = F ⊗OX f ∗ (M ∼ ). Moreover, for any A-algebra

B and any i, there exists an isomorphism
H i (X ×Y SpecB, F ⊗A B) ∼
= H i (K ⊗A B)
functorial in B, where F ⊗A B is the inverse image p∗ F of F under the projection

p : X ×Y SpecB → X. In particular, for any i and any y ∈ Y , we have
H i (Xy , Fy ) ∼
= H i (K · ⊗A k(y)) ∼
= H i (X, F ⊗A k(y)).
Proof. Let U = {U0 , . . . , Un } be a covering of X by finitely many affine open

subschemes. Consider the alternating Čech complex C · = C 0· (U, F). We have
Y
Ci = F(Uk0 ...ki ),
k0 <···<ki
Y
C 0i (U, F ⊗A M ) = F(Uk0 ...ki ) ⊗A M = C i ⊗A M.
k0 <···<ki
So we have C (U, F ⊗A M ) = C · ⊗A M . By Corollary 4.5, for every i, we have

0·
H i (X, F ⊗A M ) ∼
= H i (C · ⊗A M ).
For any A-algebra B, U0 = {p−1 (U0 ), . . . , p−1 (Un )} is a covering of X ×Y SpecB

by affine open subschemes. One can verify
C 0· (U0 , F ⊗A B) = C · ⊗A B.
So by Corollary 4.5 again, we have
H i (X ×Y SpecB, F ⊗A B) ∼
= H i (C · ⊗A M )
for every i. Since F is flat over Y , C · is a complex of flat A-modules. By

Corollary 5.3, H i (C · ) is a finitely generated A-module for each i. Moreover
C i 6= 0 only when 0 ≤ i ≤ n. Our proposition then follows from the Lemma
6.3.
Theorem 6.5. (Semicontinuity Theorem) Let f : X → Y be a proper mor-

phism of noetherian schemes and let F be a coherent OX -module flat over Y .
(i) For each i, the function
y 7→ dimk(y) H i (Xy , Fy )
on Y is upper semicontinuous. (Recall that a function g : Y → R is called

upper semicontinuous if g −1 ((−∞, r)) is open for any r ∈ R.)
(ii) The function

X
y 7→ χ(Xy , Fy ) = (−1)i dimk(y) H i (Xy , Fy )
i
on Y is locally constant.
Proof. The problem is local with respect Y . So we may assume Y = SpecA for
some noetherian ring A. Let K · be the complex in Proposition 6.4. We have
dimk(y) H i (Xy , Fy )
= dimk(y) H i (K · ⊗A k(y))
= dimk(y) ker(di ⊗ idk(y) ) − dimk(y) im(di−1 ⊗ idk(y) )
= dimk(y) (K i ⊗A k(y)) − dimk(y) im(di ⊗ idk(y) ) − dimk(y) im(di−1 ⊗ idk(y) ).
Taking the alternating sum of the above dimensions, we get

X
χ(Xy , Fy ) = (−1)i dimk(y) (K i ⊗A k(y)).
i
Since K i are finitely generated projective A-modules, the functions
y 7→ dimk(y) (K i ⊗A k(y))
on Y are locally constant by Proposition 1.4.7. So the function

X
y 7→ χ(Xy , Fy ) = (−1)i dimk(y) H i (Xy , Fy )
i
on Y is locally constant. This proves (ii). To prove (i), it suffices to show that
for every i, the function
y 7→ dimk(y) im(di ⊗ idk(y) )
on Y is lower semicontinuous. The problem is local. By Proposition 1.4.7 again,

we may assume K i and K i+1 are free A-modules of finite ranks. Fix bases for
K i and K i+1 and let P be the matrix for the homomorphism di : K i → K i+1
with respect these bases. We have
{y ∈ Y |dimk(y) im(di ⊗ idk(y) ) ≤ r} = V (a),
where a is the ideal of A generated by the determinants of (r+1)×(r+1)-minors

of the matrix P . So {y ∈ Y |dimk(y) im(di ⊗ idk(y) ) ≤ r} is closed and hence the
function y 7→ dimk(y) im(di ⊗ idk(y) ) on Y is lower semicontinuous.
Theorem 6.6. Let f : X → Y be a proper morphism of noetherian schemes

and let F be a coherent OX -module flat over Y . If Y is reduced and the function
is locally constant on Y for some i, then Ri f∗ F is locally free and the canonical
homomorphism
(Ri f∗ F)y ⊗OY,y k(y) → H i (Xy , Fy )
is an isomorphism for any y ∈ Y .
Proof. The problem is local with respect to Y . We may assume Y = SpecA

for some reduced noetherian ring A. Let K · be the complex in Proposition 6.4.
By assumption,
is locally constant on Y . Since K i is projective,
y 7→ dimk(y) (K i ⊗A k(y))
is also locally constant by Proposition 1.4.7. We have seen in the proof of

Theorem 6.5 that
dimk(y) H i (Xy , Fy ) + dimk(y) im(di−1 ⊗ idk(y) )

= dimk(y) (K i ⊗A k(y)) − dimk(y) im(di ⊗ idk(y) ).
The function on Y defined by the left-hand side is lower semicontinuous and

that defined by the right-hand side is upper semicontinuous. So they define a
continuous function on Y . But this function has integer values, so it must be
locally constant. Hence the functions
y 7→ dimk(y) im(di ⊗ idk(y) ),

y 7→ dimk(y) im(di−1 ⊗ idk(y) )
are locally constant on Y . Since the function
y 7→ dimk(y) (K i+1 ⊗A k(y))
is locally constant by Proposition 1.4.7, the function
y 7→ dimk(y) coker(di ⊗ idk(y) ) = dimk(y) (K i+1 ⊗A k(y)) − dimk(y) im(di ⊗ idk(y) )
is also locally constant. But the functor − ⊗A k(y) is right exact, so we have
coker(di ⊗ idk(y) ) ∼
= (cokerdi ) ⊗A k(y).
Hence the function

y 7→ dimk(y) ((cokerdi ) ⊗A k(y))
is locally constant on Y . By Lemma 6.7 below, cokerdi is a projective A-module.
Hence the short exact sequence
0 → imdi → K i+1 → cokerdi → 0

splits and we have

K i+1 ∼
= imdi ⊕ cokerdi .
Combining with the fact that K i+1 is projective, we see that imdi is projective.
So the short exact sequence
0 → kerdi → K i → imdi → 0
splits and we have

Ki ∼
= kerdi ⊕ imdi .
Combining with the fact that K i is projective, this implies that kerdi is projec-
tive. The morphism di : K i → K i+1 can be identified with
kerdi ⊕ imdi → imdi ⊕ cokerdi , (x, y) → (y, 0).
Under the decomposition
K i ⊗A k(y) ∼
= (kerdi ) ⊗A k(y) ⊕ (imdi ) ⊗A k(y),
im(di−1 ⊗ idk(y) ) is contained in the factor (kerdi ) ⊗A k(y). Since kerdi is pro-
jective, the function
y 7→ dimk(y) ((kerdi ) ⊗A k(y))
is locally constant. So the function
y 7→ dimk(y) ((kerdi ) ⊗A k(y)) − dimk(y) im(di−1 ⊗ idk(y) )

= dimk(y) ((kerdi ) ⊗A k(y)/im(di−1 ⊗ idk(y) ))
is locally constant. Note that (kerdi ) ⊗A k(y)/im(di−1 ⊗ idk(y) ) is the cokernel

of the homomorphism
K i−1 ⊗A k(y) → (kerdi ) ⊗A k(y).
By the right exactness of the functor − ⊗A k(y), it can be identified with
coker(K i−1 → kerdi ) ⊗A k(y) = H i (K · ) ⊗A k(y).
So the function
y 7→ dimk(y) (H i (K · ) ⊗A k(y))
is locally constant. By Lemma 6.7 below, H i (K · ) is a projective A-module. By
Proposition 6.4, we have H i (X, F) ∼
= H i (K · ), and by Proposition 4.9, we have
Ri f∗ F ∼
= (H i
(X, F))∼
. So
Ri f∗ F ∼
= (H i (K · ))∼ .
By Proposition 1.4.7, (H i (K · ))∼ is locally free. So Ri f∗ F is locally free. The

0 → imdi−1 → kerdi → H i (K · ) → 0
splits and hence

kerdi ∼
= imdi−1 ⊕ H i (K · ).
Since kerdi is projective, this implies imdi−1 is projective. So the short exact
sequence
0 → kerdi−1 → K i−1 → imdi−1 → 0
splits and hence
K i−1 ∼
= kerdi−1 ⊕ imdi−1 .
The sequence
di−1 di
K i−1 → K i → K i+1
kerdi−1 ⊕ imdi−1 → imdi−1 ⊕ H i (K · ) ⊕ imdi → imdi ⊕ cokerdi ,
where the first arrow is

(x, y) 7→ (y, 0, 0)
and the second arrow is
(x, y, z) 7→ (z, 0).
From this description, it is easy to see that
H i (K · ⊗A k(y)) ∼
= H i (K · ) ⊗A k(y),
that is,
H i (Xy , Fy ) ∼
= (Ri f∗ F)y ⊗OY,y k(y).
Lemma 6.7. Let Y be a reduced noetherian scheme and let F be a coherent

OY -module. If the function
y 7→ dimk(y) (Fy ⊗OY,y k(y))
is locally constant on Y , then F is locally free.
Proof. Let P be a point in Y . Choose an affine open connected neighborhood

V of P and sections s1 , . . . , sn ∈ F(V ) such that their images in FP ⊗OY,P k(P )
form a basis for the k(P )-vector space FP ⊗OY,P k(P ). Consider the morphism
OVn → F|V , ei 7→ si ,
where for each i, ei is the section of OVn whose i-th component is 1 and whose
other components are 0. Let C be cokernel of this morphism. By Nakayama’s
lemma, (OVn )P → FP is surjective and hence CP = 0. By Proposition 1.4.1,
after shrinking the neighborhood V of P , we may assume C|V = 0 and hence
OVn → F|V is surjective. Since V is connected, the locally constant function
y 7→ dimk(y) (Fy ⊗OY,y k(y)) is constant on V . So dimk(y) (Fy ⊗OY,y k(y)) = n

for any y ∈ V . Combining with the surjectivity of OVn → F|V , we see that
(OVn )y ⊗OY,y k(y) → Fy ⊗OY,y k(y)
is an isomorphism for any y ∈ V . So for any section (a1 , . . . , an ) of ker(OVn → F)

over some affine open subscheme U of V , the germ of each ai at y lies in the
maximal ideal of OY,y for any y ∈ U . Since U is reduced, we must have ai = 0.
So the kernel of the epimorphism OVn → F is 0. Hence OVn ∼ = F and F is locally
free.
Proposition 6.8. Let A be a noetherian local ring with the maximal ideal m,
X a scheme proper over SpecA, and F a coherent OX -module flat over SpecA.
For each i, the following conditions are equivalent:
(i) The canonical homomorphism
H i (X, F) ⊗A A/m → H i (X, F ⊗A A/m)
is surjective.
(ii) The canonical homomorphism
H i (X, F) ⊗A M → H i (X, F ⊗A M )
is surjective for any A-module M .

(iii) The functor
M 7→ H i (X, F ⊗A M )
on the category of A-modules is right exact.
(iv) The canonical homomorphism
H i (X, F) ⊗A M → H i (X, F ⊗A M )
is an isomorphism for any A-module M .
Proof. Any x ∈ M defines a morphism
F → F ⊗A M, s 7→ s ⊗ x
and hence a homomorphism
φx : H i (X, F) → H i (X, F ⊗A M ).
The canonical homomorphism
H i (X, F) ⊗A M → H i (X, F ⊗A M )
is defined by s ⊗ x 7→ φx (s).
(i)⇒(ii) We first prove
H i (X, F) ⊗A M → H i (X, F ⊗A M )
is surjective for any A-module M with finite length. This is trivial when
lengthA M = −1, that is, when M = 0. Suppose lengthA M = 0. Then M = Ax
for any nonzero element x in M . So M ∼
= A/Ann(x). Since
0 ⊂ m/Ann(x) ⊂ A/Ann(x)
is a chain in the A-module A/Ann(x) and lengthA (A/Ann(x)) = 0, we must

have m = Ann(x) and hence M ∼ = A/m. Our assertion then follows from the
assumption (i). Suppose we have proved (ii) for A-modules with lengths less
than n and let M be an A-module of length n. Then we can find a short exact
sequence of A-modules
0 → M 0 → M → M 00 → 0
such that lengthA (M 0 ), lengthA (M 00 ) < n. Consider the commutative diagram
H i (X, F) ⊗A M 0 → H i (X, F) ⊗A M → H i (X, F) ⊗A M 00 → 0

↓ ↓ ↓
H i (X, F ⊗A M 0 ) → H i (X, F ⊗A M ) → H i (X, F ⊗A M 00 ).
We will explain in a moment that the two rows in the diagram are exact. By
the induction hypothesis, the first and the third vertical arrows are surjective.
This implies that the second vertical arrow is surjective. This proves (ii) for
A-modules of finite lengths. The exactness of the first row in the above diagram
follows from the right exactness of the functor H i (X, F) ⊗A −. Since F is flat
over SpecA, the sequence
0 → F ⊗A M 0 → F ⊗A M → F ⊗A M 00 → 0
is exact. The second row in the above diagram is a part of the long exact
sequence of cohomology groups associated to this short exact sequence.
Next we show (ii) holds for any finitely generated A-module M . Note that
M/mn M is an A-module with finite length for each n. So
H i (X, F) ⊗A M/mn M → H i (X, F ⊗A M/mn M )
is surjective. Let Kn be its kernel. Each Kn has finite length. So the inverse
system (Kn )n∈N satisfies the Mittag-Leffler condition. Hence by Proposition
1.5.1, the homomorphism
inv. lim(H i (X, F) ⊗A M/mn M ) → inv. lim H i (X, F ⊗A M/mn M )

n n
is surjective. Let A
b be the m-adic completion of A. By Proposition 1.5.5, we
have
inv. lim(H i (X, F)⊗A M/Mn M ) = (H i (X, F)⊗A M )∧ ∼

= (H i (X, F)⊗A M )⊗A A,
b
n
and by Theorem 5.6, we have
inv. lim H i (X, F ⊗A M/mn M ) ∼

= (H i (X, F ⊗A M ))∧ ∼
= H i (X, F ⊗A M ) ⊗A A.
b
n
So the homomorphism
(H i (X, F) ⊗A M ) ⊗A A
b → H i (X, F ⊗A M ) ⊗A A
b
is surjective. By Corollary 1.23, A

b is a faithfully flat A-algebra. So
H i (X, F) ⊗A M → H i (X, F ⊗A M )
is surjective. This proves (ii) for finitely generated A-modules.

For any A-module M , we have M = dir. limn Mn , where (Mn ) is the direct
system of finitely generated A-submodules of M . We have
H i (X, F) ⊗A M = dir. lim(H i (X, F) ⊗A Mn ),

n
H i (X, F ⊗A M ) = dir. lim H i (X, F ⊗A Mn ),
n
where the second equality follows from Proposition 1.12 (ii). Since each
H i (X, F) ⊗A Mn → H i (X, F ⊗A Mn )
is surjective,
H i (X, F) ⊗A M → H i (X, F ⊗A M )
is also surjective.
(ii)⇒(iii) Given a short exact sequence of A-modules
0 → M 0 → M → M 00 → 0,
H i (X, F) ⊗A M 0 → H i (X, F) ⊗A M → H i (X, F) ⊗A M 00 → 0

↓ ↓ ↓
H i (X, F ⊗A M 0 ) → H i (X, F ⊗A M ) → H i (X, F ⊗A M 00 ),
where the two rows are exact and all the vertical arrows are surjective. This
implies that H i (X, F ⊗ M ) → H i (X, F ⊗A M 0 ) is surjective. So the sequence
H i (X, F ⊗A M 0 ) → H i (X, F ⊗A M ) → H i (X, F ⊗A M 00 ) → 0
is exact and hence the functor H i (X, F ⊗A −) is right exact.

(iii)⇒(iv) First assume M is a finitely generated A-module. Then we can
find an exact sequence
Am → An → M → 0
for some natural numbers m and n. Consider the following commutative dia-
gram:
H i (X, F) ⊗A Am → H i (X, F) ⊗A An → H i (X, F) ⊗A M → 0

↓ ↓ ↓
H i (X, F ⊗A Am ) → H i (X, F ⊗A An ) → H i (X, F ⊗A M ) → 0.
The two rows are exact since the functors H i (X, F) ⊗A − and H i (X, F ⊗A −)
are right exact. The first two vertical arrows are obviously isomorphisms. So the
last vertical arrow is also an isomorphism. This proves (iv) for finitely generated
A-modules. The general case follows by writing a module as the direct limit of
its finitely generated submodules.
(iv)⇒(i) Trivial.
Theorem 6.9. Let f : X → Y be a proper morphism of noetherian schemes

and let F be a coherent OX -module flat over Y . Assume for some i the canonical
homomorphism
is surjective for any y ∈ Y . Then it is an isomorphism. Moreover, under this
assumption, the following conditions are equivalent:
(i) The OY -module Ri f∗ F is locally free.
(ii) The canonical homomorphism
(Ri−1 f∗ F)y ⊗OY,y k(y) → H i−1 (Xy , Fy )
is surjective for any y ∈ Y .
Proof. Using Proposition 4.9 and 4.10, one can verify
(Ri f∗ F)y ∼
= H i (X ×Y SpecOY,y , p∗ F),
where p : X ×Y SpecOY,y → X is the projection. Note that the direct image of

the sheaf Fy on Xy under the closed immersion Xy → X ×Y SpecOY,y is the
sheaf p∗ F ⊗OY,y k(y) on X ×Y SpecOY,y . So we have
H i (Xy , Fy ) ∼
= H i (X ×Y SpecOY,y , p∗ F ⊗OY,y k(y)).
By our assumption,
is surjective. So the canonical homomorphism
H i (X ×Y SpecOY,y , p∗ F) ⊗OY,y k(y) → H i (X ×Y SpecOY,y , p∗ F ⊗OY,y k(y))
is surjective. Applying Proposition 6.8 to the scheme X ×Y SpecOY,y which is

proper over SpecOY,y and to the coherent sheaf p∗ F which is flat over SpecOY,y ,
we see that this homomorphism is an isomorphism and hence
is an isomorphism. Moreover, by Proposition 6.8, we have
H i (X ×Y SpecOY,y , p∗ F) ⊗OY,y M ∼
= H p (X ×Y SpecOY,y , p∗ F ⊗OY,y M )
for any OY,y -module M . By Proposition 1.4.1 (ii), the OY -module Ri f∗ F is

locally free if and only if (Ri f∗ F)y ∼
= H i (X ×Y SpecOY,y , p∗ F) is a free OY,y -
module for any y ∈ Y . By Proposition 1.18 and Lemma 1.4.8, this is equivalent
to saying that the functor on the category of OY,y -modules
M 7→ H i (X ×Y SpecOY,y , p∗ F) ⊗OY,y M ∼
= H i (X ×Y SpecOY,y , p∗ F ⊗OY,y M )
is exact for any y ∈ Y . Using the assumption that F is flat over Y and the long
exact sequence of cohomology groups, we see that the functor
M 7→ H i (X ×Y SpecOY,y , p∗ F ⊗OY,y M )
is exact if and only if the following two functors are right exact:
M 7→ H i (X ×Y SpecOY,y , p∗ F ⊗OY,y M ),
M 7→ H i−1 (X ×Y SpecOY,y , p∗ F ⊗OY,y M ).
By Proposition 6.8, this is equivalent to saying the following canonical homo-

morphisms are surjective:
H i (X ×Y SpecOY,y , p∗ F) ⊗OY,y k(y) → H i (X ×Y SpecOY,y , p∗ F ⊗OY,y k(y)),
H i−1 (X ×Y SpecOY,y , p∗ F)⊗OY,y k(y) → H i−1 (X ×Y SpecOY,y , p∗ F ⊗OY,y k(y)),

that is,
(Ri f∗ F)y ⊗OY,y k(y) → H i (Xy , Fy ),

are surjective. So under the assumption that
is surjective for any y ∈ Y , the OY -module Ri f∗ F is locally free if and only if
is surjective for any y ∈ Y .
Corollary 6.10. Let f : X → Y be a proper morphism of noetherian scheme

and let F be an OX -module flat over Y .
(i) If H i (Xy , Fy ) = 0 for some i and any y ∈ Y , then Ri f∗ F = 0 and
(Ri−1 f∗ F)y ⊗OY,y k(y) ∼

= H i−1 (Xy , Fy ).
(ii) If H 1 (Xy , Fy ) = 0 for any y ∈ Y , then the OY -module f∗ F is locally

free.
(iii) Let i0 be a natural number such that Ri f∗ F = 0 for any i ≥ i0 . Then
H i (Xy , Fy ) = 0 for any i ≥ i0 and any y ∈ Y .
Proof. (i) The canonical homomorphism
is trivially surjective for any y ∈ Y . By Theorem 6.9, it is an isomorphism. So

(Ri f∗ F)y ⊗OY,y k(y) = 0 for any y ∈ Y . By Nakayama’s lemma, we must have
(Ri f∗ F)y = 0 for any y ∈ Y and hence Ri f∗ F = 0. In particular, Ri f∗ F = 0 is
locally free. So by Theorem 6.9, the canonical homomorphism
is an isomorphism for any y ∈ Y .

(ii) By (i), the canonical homomorphism
(R0 f∗ F)y ⊗OY,y k(y) → H 0 (Xy , Fy )
is surjective and the canonical homomorphism
(R−1 f∗ F)y ⊗OY,y k(y) → H −1 (Xy , Fy )
is trivially surjective for any y ∈ Y . So f∗ F = R0 f∗ F is locally free by Theorem

6.9.
(iii) We know H i (Xy , Fy ) = 0 for any y ∈ Y when i is sufficiently large.
Suppose i > i0 and H i (Xy , Fy ) = 0 for any y ∈ Y . Then by (i), we have
H i−1 (Xy , Fy ) ∼
= (Ri−1 f∗ F)y ⊗OY,y k(y) = 0.
2.7 Grothendieck’s Existence Theorem
Let A be a noetherian ring, I an ideal of A, X a scheme proper over SpecA,

F a coherent OX -module, X b and Fb the formal completions of X and F with
respect to I, respectively. In this section, we prove that
H n (X, b ∼
b F) = inv. lim H n (X, F/I k F).
k
Moreover, we prove Grothendieck’s existence theorem, which says that any co-
herent OXb -module is the formal completion of some coherent OX -module. We
start with a series of lemmas:
Lemma 7.1. Let (Ck· )k∈N be an inverse system of complexes of abelian groups.
Suppose for any n, the inverse system (H n (Ck· ))k∈N satisfies the Mittag-Leffler
condition, and for any n and any pair l ≥ k, the homomorphism Cln → Ckn is
surjective. Then for any n, we have H n (inv. limk Ck· ) ∼
= inv. limk H n (Ck· ).
2.7. GROTHENDIECK’S EXISTENCE THEOREM 195
Proof. For any pair l ≥ k and any n, since Cln−1 → Ckn−1 is surjective,
B n (Cl· ) → B n (Ck· ) is also surjective. Since (H n (Ck· ))k∈N satisfies the Mittag-
Leffler condition, there exists a k0 such that for any l ≥ k + k0 , we have
im(H n (Cl· ) → H n (Ck· )) = im(H n (Ck+k

·
0
) → H n (Ck· )).
We have a commutative diagram

0 → B n (Cl· ) → Z n (Cl· ) → H n (Cl· ) → 0
↓ ↓ ↓
0 → B n (Ck· ) → Z n (Ck· ) → H n (Ck· ) → 0.
By diagram chasing, one can show that for any l ≥ k + k0 , we have
im(Z n (Cl· ) → Z n (Ck· )) = im(Z n (Ck+k

·
0
) → Z n (Ck· )).
So the inverse systems (Z n (Ck· ))k∈N , (B n (Ck· ))k∈N and (H n (Ck· ))k∈N satisfy
the Mittag-Leffler condition for any n. By Proposition 1.5.1, the following
sequences are exact:
0 → inv. lim Z n (Ck· ) → inv. limk Ckn → inv. lim B n+1 (Ck· ) → 0,
k k
0 → inv. lim B n (Ck· ) → inv. limk Z n (Ck· ) → inv. lim H n (Ck· ) → 0.
k k
This implies that H n

(inv. limk Ck· ) ∼
= inv. limk H n
(Ck· ).
Lemma 7.2. Let (Fk )k∈N be an inverse system of sheaves of abelian groups
on a topological space X. Suppose each Fk is flasque and for any pair l ≥ k,
the morphism Fl → Fk is surjective and its kernel is also flasque. Then for any
open subset U of X, the homomorphism
Γ(U, Fl ) → Γ(U, Fk )
is surjective for any pair l ≥ k and
H n (X, inv. lim Fk ) = 0

k
for any n ≥ 1.
Proof. For any pair l ≥ k, let Klk be the kernel of the epimorphism Fl → Fk .
The sequence
0 → Klk → Fl → Fk → 0
is an exact sequence of flasque sheaves. By lemma 1.11,
Γ(U, Fl ) → Γ(U, Fk )
is surjective for any open subset U of X. Moreover, for any open covering U of
U and any n, the homomorphism
C n (U, Fl ) → C n (U, Fk )
is surjective. Note that
H n (C · (U, Fl )) → H n (C · (U, Fk ))
is surjective for each n. Indeed, when n = 0, this follows from the surjectivity
of
Γ(U, Fl ) → Γ(U, Fk ).
When n ≥ 1, by Proposition 3.6, we have
H n (C · (U, Fk )) = 0.
So we can apply Lemma 7.1 to the inverse system (C · (U, Fk ))k∈N and we get
H n (inv. lim C · (U, Fk )) ∼

= inv. lim H n (C · (U, Fk )).
k k
It is easy to see that
inv. lim C · (U, Fk ) = C · (U, inv. lim Fk ).

k k
So we have
Ȟ n (U, inv. lim Fk ) ∼
= inv. lim Ȟ n (U, Fk ).
k k
n
When n ≥ 1, we have H (U, Fk ) = 0 by Proposition 3.6 and hence
Ȟ n (U, inv. lim Fk ) = 0.

k
This is true for any open covering U of any open set U . By Proposition 3.11,
this implies that
H n (X, inv. lim Fk ) = 0
k
for any n ≥ 1.
Lemma 7.3. Let X be a topological space, let B a family of open subsets of X

which form a basis for the topology of X, and let (Fk )k∈N be an inverse system
of sheaves of abelian groups on X. Suppose the following conditions hold:
(a) For any open subset U in B and any pair l ≥ k, the homomorphism
Γ(U, Fl ) → Γ(U, Fk )
is surjective.
(b) For any U in B, we have
H n (U, Fk ) = 0
for any k and any n ≥ 1,

(c) For each n, the inverse system (H n (X, Fk ))k∈N satisfies the Mittag-
Leffler condition.
Then we have
H n (X, inv. lim Fk ) ∼
= inv. lim H n (X, Fk ).
k k
Proof. By (a), the morphism Fl → Fk is surjective for any pair l ≥ k. Let Klk
be its kernel. By Proposition 1.10, for any i ≥ 0, we have an exact sequence
0 → C i (Klk ) → C i (Fl ) → C i (Fk ) → 0,
where C · (−) is the Godement resolution functor. So for each i, the inverse
system (C i (Fk ))k∈N satisfies the condition of Lemma 7.2. Hence
H n (X, inv. lim C i (Fk )) = 0

k
for any n ≥ 1 and any i. Moreover, for any open subset U in X and any pair
l ≥ k, the homomorphism
Γ(U, C i (Fl )) → Γ(U, C i (Fk ))
is surjective.
Let U be a member in B. By (a) and (b) and the fact that
H n (Γ(U, C · (Fk ))) ∼

= H n (U, Fk ),
the homomorphism
H n (Γ(U, C · (Fl ))) → H n (Γ(U, C · (Fk )))
is surjective for any n and any pair l ≥ k. We have seen
Γ(U, C n (Fl )) → Γ(U, C n (Fk ))
is surjective. So by Lemma 7.1, we have
H n (inv. lim Γ(U, C · (Fk ))) ∼

= inv. lim H n (Γ(U, C · (Fk ))) = inv. lim H n (U, Fk ).
k k k
When n ≥ 1, by (b), we have
H n (inv. lim Γ(U, C · (Fk ))) ∼

= inv. lim H n (U, Fk ) = 0.
k k
This is true for any U in the family B. So the sequence
0 → inv. lim Fk → inv. lim C 0 (Fk ) → inv. lim C 0 (Fk ) → · · ·

k k k
is exact. We have seen that
H n (X, inv. lim C i (Fk )) = 0

k
for any n ≥ 1 and any i. By Proposition 1.2, we have

= H n (Γ(X, inv. lim C · (Fk ))) = H n (inv. lim Γ(X, C · (Fk ))).
k k k
We have seen
Γ(X, C n (Fl )) → Γ(X, C n (Fk ))
is surjective. Moreover, for any n, the inverse system
(H n (Γ(X, C · (Fk ))))k∈N = (H n (X, Fk ))k∈N
satisfies the Mittag-Leffler condition by (c). So by Lemma 7.1 again, we have
H n (inv. lim Γ(X, C · (Fk ))) ∼

= inv. lim H n (Γ(X, C · (Fk ))) ∼
k k k
Therefore
k k
Let X be a noetherian scheme and let I be a coherent sheaf of ideals of OX .

Recall that the formal completion of X with respect to I is the ringed space
b O b ) defined by
(X, X
Xb = supp(OX /I),
OXb = inv. lim(OX /I k )|Xb .
k
For any OX -module F, its formal completion is the OXb -module
Fb = inv. lim(F/I k F)|Xb .

k
Theorem 7.4. Let A be a noetherian ring, I an ideal of A, X a scheme proper

over SpecA, F a coherent OX -module, X b and Fb the formal completions of X
and F with respect to the sheaf of ideals I ∼ OX , respectively. Then for each n,
we have an isomorphism
H n (X, b ∼
b F) = inv. lim H n (X, F/I k F),
k
where I k F = (I k )∼ F.
Proof. Let i : Xb → X be the canonical morphism of locally ringed spaces.

k
Since each F/I F is supported in the closed subset X,
b we have
F/I k F = i∗ ((F/I k F)|Xb )
and hence
i∗ Fb = i∗ (inv. lim(F/I k F)|Xb ) = inv. lim F/I k F.
k k
Using the fact that i∗ is an exact functor and transforms injective sheaves on
X
b to injective sheaves on X, one can verify that
H n (X, b ∼
b F) = H n (X, i∗ F).
b
So
H n (X, b ∼
b F) = H n (X, inv. lim F/I k F).
k
We have a long exact sequence
· · · → H n (X, F) → H n (X, F/I k F) → H n+1 (X, I k F) → · · ·
Let Qk = im(H n (X, F/I k F) → H n+1 (X, I k F)). Then we have an exact se-
quence
H n (X, F) → H n (X, F/I k F) → Qk → 0.
By Lemma 5.8, there exists a natural number k0 such that the canonical ho-
momorphism Qk+k0 → Qk vanishes for any k ≥ k0 . Then Qk+2k0 → Qk+k0
vanishes for any k. For any k and any l ≥ k + 2k0 , the canonical homomor-
phism Ql → Qk vanishes since it can be factorized as
Ql → Qk+2k0 → Qk+k0 → Qk .
Consider the following commutative diagram (l ≥ k + 2k0 ):
H n (X, F) → H n (X, F/I l F) → Ql → 0

k ↓ ↓
H n (X, F) → H n (X, F/I k F) → Qk → 0.
Since the third vertical arrow vanishes, we have
im(H n (X, F/I l F) → H n (X, F/I k F)) ⊂ im(H n (X, F) → H n (X, F/I k F)).
By the commutativity of the square on the left-hand side of the diagram, we

have
im(H n (X, F) → H n (X, F/I k F)) ⊂ im(H n (X, F/I l F) → H n (X, F/I k F)).
So
im(H n (X, F/I l F) → H n (X, F/I k F)) = im(H n (X, F) → H n (X, F/I k F))
for any l ≥ k + 2k0 . So the inverse system (H n (X, F/I k F))k∈N satisfies the
Mittag-Leffler condition. On the other hand, for any affine open subscheme U
of X, the homomorphism
Γ(U, F/I l F) → Γ(U, F/I k F)
is surjective for any l ≥ k, and
H n (U, F/I k F) = 0
for any k and any n ≥ 1. By Lemma 7.3, we have

H n (X, inv. lim F/I k F) ∼
k k
So
H n (X, b ∼
b F) = inv. lim H n (X, F/I k F).
k
Corollary 7.5. Let A be a noetherian ring, I an ideal of A, X a scheme proper

over SpecA, F a coherent OX -module, X b and Fb the formal completions of X
and F with respect to the sheaf of ideals I ∼ OX , respectively. Then for each n,
we have an isomorphism
H n (X, b ∼
b F) = (H n (X, F))∧ ,
where (H n (X, F))∧ is the I-adic completion of H n (X, F).
Proof. Follows from Theorems 5.6 and 7.4.
Corollary 7.6. Let f : X → Y be a proper morphism of noetherian schemes,

I a coherent sheaf of ideals of OY , F a coherent OX -module, Xk and Yk the
closed subschemes of X and Y with ideal sheaves I k OX and I k , respectively,
Fk = (F/I k F)|Xk , X,
b Fb and Yb the formal completions of X, F and Y with
respect to IOX , IOX and I, respectively, fk : Xk → Yk and fˆ : X b → Yb the
morphisms induced by f . Then for each n, we have isomorphisms
Rn fˆ∗ Fb ∼
= inv. lim Rn fk∗ Fk ∼ n f F,
= R\ ∗
k
n f F is the formal completion of Rn f F with respect to I.

where R\ ∗ ∗
Proof. Note that X b and Xk (resp. Yb and Yk ) have the same underlying
topological space, and fb and fk are the same as maps on these topological spaces.
The family of canonical morphisms Fb → Fk induces a family of morphisms
Rn fˆ∗ Fb → Rn fk∗ Fk
and hence a morphism
Rn fˆ∗ Fb → inv. lim Rn fk∗ Fk .
k
Let’s show it is an isomorphism. Let U = SpecA be an affine open subscheme

of Y . By Theorem 7.4, we have
H n (X b ∼
b ∩ f −1 (U ), F) = inv. lim H n (f −1 (U ), F/I k F).
k
Let iXk : Xk → X be the closed immersion. We have iXk ∗ Fk = F/I k F. Since

iXk ∗ is an exact functor and transforms injective sheaf into injective sheaf, we
have
H n (f −1 (U ), F/I k F) ∼
= H n (Xk ∩ f −1 (U ), Fk ).
So
H n (X b ∼
b ∩ f −1 (U ), F) = inv. lim H n (Xk ∩ f −1 (U ), Fk ),
k
that is,
b ∼
H n (fˆ−1 (Yb ∩ U ), F) = inv. lim H n (fk−1 (Yk ∩ U ), Fk ),
k
But R fˆ∗ Fb is the sheaf associated to the presheaf

n
V 7→ H n (fˆ−1 (V ), F)
b
and Rn fk∗ Fk is the sheaf associated to the presheaf
V 7→ H n (fk−1 (V ), Fk ),
and when U goes over the family of affine open subschemes of Y , Yb ∩U = Yk ∩U

forms a basis for the topology of Yb = Yk , so
Rn fˆ∗ Fb ∼
= inv. lim Rn fk∗ Fk .
k
Let iYk : Yk → Y be the closed immersions. We have f iXk = iYk fk and hence
Rn (f iXk )∗ Fk = Rn (iYk fk )∗ Fk .
Using the fact that iXk ∗ and iYk ∗ are exact functors and transform injective
sheaves to injective sheaves, one can verify
Rn (f iXk )∗ Fk ∼
= Rn f∗ (iXk ∗ Fk ) ∼
= Rn f∗ (F/I k F)
Rn (iYk fk )∗ Fk ∼ n
= iYk ∗ R fk∗ Fk .
So
Rn f∗ (F/I k F) ∼
= iYk ∗ Rn fk∗ Fk .
By Corollary 5.9, we have

= inv. lim Rn f∗ (F/I k F).
k k
So
= inv. lim iYk ∗ Rn fk∗ Fk .
k k
This implies that
inv. lim(Rn f∗ F/I k Rn f∗ F)|Yb ∼

= inv. lim Rn fk∗ Fk ,
k k
that is,
nf F ∼
R\ = inv. lim Rn fk∗ Fk .
∗
k
Lemma 7.7. Let X be a noetherian scheme, I a coherent sheaf of ideals of OX ,

b the formal completion of X with respect to I, and i : X → X
X b the canonical
morphism. Then the functor F 7→ i∗ F is exact on the category of OX -modules.
Proof. The functor F 7→ i∗ F is exact on the category of coherent OX -modules

by Proposition 1.5.13 (i) and (ii). In general, for any OX -module F and any
P ∈ X,
b we have
(i∗ F)P ∼
= OX,P
b ⊗OX,P FP .
To prove i∗ is an exact functor on the category of OX -modules, it suffices to
show OX,P
b is a flat OX,P -algebra for any P ∈ X.
b For any finitely generated
OX,P -module M , we can find an exact sequence
m n
OX,P → OX,P →M →0
for some natural numbers m and n. By Proposition 1.4.1 (i), there exists an
m n
open neighborhood U of P in X and a morphism OU → OU which induces the
m n m n
homomorphism OX,P → OX,P on stalks. Let F be the cokernel of OU → OU .
∼
Then F is a coherent OU -module and FP = M . Let
0 → M 0 → M → M 00 → 0
be an exact sequence of finitely generated OX,P -modules. We can find a neigh-

borhood U of P in X and coherent OU -modules F 0 , F and F 00 such that their
stalks at P are isomorphic to M 0 , M , and M 00 , respectively. Shrinking U if
necessary, we may assume there exist morphisms F 0 → F and F → F 00 which
induce M 0 → M and M → M 00 on stalks, respectively. Since M 0 → M is in-
jective, the stalk at P of ker(F 0 → F) is 0. By Proposition 1.4.1 (ii), we have
ker(F 0 → F) = 0 after shrinking U . By similar argument, we see that after
shrinking U , we have an exact sequence
0 → F 0 → F → F 00 → 0
of coherent OU -modules which induces the exact sequence
0 → M 0 → M → M 00 → 0
on stalks. By Proposition 1.5.13 (i) and (ii), the sequence
0 → (i∗ F 0 )P → (i∗ F)P → (i∗ F 00 )P → 0
is exact, where i is the canonical morphism from U to its formal completion U

b.
So the sequence
0 → OX,P
b ⊗OX,P M 0 → OX,P
b ⊗OX,P M → OX,P
b ⊗OX,P M 00 → 0
is exact. Hence OX,P

b is a flat OX,P -algebra. This proves our assertion.
Proposition 7.8. Let A be a noetherian ring, I an ideal of A, X a scheme

proper over SpecA, F and G coherent OX -module, X, b Fb and Gb the formal
∼
completions of X, F and G with respect to I OX , respectively.
(i) For each n, ExtnOX (F, G) is a coherent OX -module and
(ExtnOX (F, G))∧ ∼

= ExtnOXc (F,
b G),
b
where (ExtnOX (F, G))∧ is the formal completion of ExtnOX (F, G) with respect to
I ∼ OX .
(ii) For each n, ExtnOX (F, G) is a finitely generated A-module and

= ExtnOXc (F,
b G),
b
where (ExtnOX (F, G))∧ is the I-adic completion of ExtnOX (F, G).
Proof. (i) Let

0 → G → I·
be an injective resolution of the OX -module G and let
0 → Gb → I·
be an injective resolution of the OXb -module G.

b We have
ExtnOX (F, G) ∼
= H n (HomOX (F, I · )).
So by lemma 7.7, we have
i∗ ExtnOX (F, G) ∼
= H n (i∗ HomOX (F, I · )).
We have a canonical homomorphism
i∗ HomOX (F, I · ) → HomOXc (i∗ F, i∗ I · ).
Again by Lemma 7.7, the sequence
0 → i∗ G → i∗ I ·
is exact. By Proposition 1.5.13 (ii), we have i∗ G ∼

= G.
b By Proposition 1.1 (ii),
this isomorphism can be extended to a morphism of complexes
i∗ I · → I· .
By Corollary 5.4 (i), ExtnOX (F, G) is coherent. So by Proposition 1.5.13 (i), we

have
(ExtnOX (F, G))∧ = i∗ ExtnOX (F, G).
We define a morphism
(ExtnOX (F, G))∧ → ExtnO c (F,

b G)
b
X

=i∗ ExtnOX (F, G)
∼
=H n (i∗ HomOX (F, I · ))
→ H n (HomOXc (i∗ F, i∗ I · ))
→ H n (HomOXc (i∗ F, I· ))
∼
= ExtnO c (F,
b G).
b
X
Let’s prove it is an isomorphism. This is a local question. So we may assume

there exists an exact sequence of OX -modules
· · · → E1 → E0 → F → 0
such that each En is free of finite rank. By Lemma 7.7, the sequence
· · · → i∗ E1 → i∗ E0 → i∗ F → 0
is exact and each i∗ En is free of finite rank. By Proposition 2.16, we have
ExtnOX (F, G) ∼
= H n (HomOX (E· , G)),
ExtnO (F,
b G)
b ∼
= H n (HomOXc (i∗ E· , i∗ G)).
X
c
By Lemma 7.7, we have
= i∗ ExtnOX (F, G) ∼
(ExtnOX (F, G))∧ ∼ = H n (i∗ HomOX (E· , G)).
Since E· is a complex of free OX -modules, we have
i∗ HomOX (E· , G) ∼
= HomOXc (i∗ E· , i∗ G).
Putting all these isomorphisms together, we get

= ExtnOXc (F,
b G).
b
H p (X, ExtqOX (F, G)) ⇒ Extp+q

OX (F, G).
This spectral sequence involves only finitely generated A-modules by Corollary

5.4 (ii). Passing to I-adic completions and applying Proposition 1.5.4, we get a
biregular spectral sequence
(H p (X, ExtqOX (F, G)))∧ ⇒ (ExtO

p+q
X
(F, G))∧ (1)
On the other hand, we have a biregular spectral sequence

b Extq (F,
H p (X, b ⇒ Extp+q (F,
b G)) b G)
b (2)
Oc X
Oc X
We leave to the reader to construct a morphism from spectral (1) to spectral

sequence (2). By Corollary 7.5, we have
(H p (X, ExtqOX (F, G)))∧ ∼ b (Extq (F, G))∧ ) ∼

= H p (X, OX
b Extq (F,
= H p (X, Oc
b G)).
b
X
So by Proposition 2.1, we have

= ExtnOXc (F,
b G).
b
Corollary 7.9. Keep the assumptions and notations in Proposition 7.8. If A

is I-adically complete, then we have
HomOX (F, G) ∼
= HomOXc (F,
b G),
b
where the isomorphism maps each morphism u : F → G to the morphism

û : Fb → Gb induced by u on formal completions. Moreover, u is a monomorphism
(resp. epimorphism, resp. isomorphism) if and only û is a monomorphism (resp.
epimorphism, resp. isomorphism).
Proof. Taking n = 0 in Proposition 7.8 (ii), we get
(HomOX (F, G))∧ ∼

= HomOXc (F,
b G).
b
By Corollary 5.4 (ii), HomOX (F, G) is a finitely generated A-module. Since A

is I-adically complete, by Proposition 1.5.5, we have
HomOX (F, G) ∼
= (HomOX (F, G))∧ .
Hence
HomOX (F, G) ∼
= HomOXc (F,
b G).
b
By Corollary 1.5.16 (ii), the morphism û : Fb → Gb is a monomorphism (resp.

epimorphism, resp. isomorphism) if and only if u : F → G is a monomorphism
(resp. epimorphism, resp. isomorphism) in a neighborhood of X b in X. We claim
X is the only neighborhood of X in X. This would prove our assertion. Let U
b
be an open neighborhood of X b in X. Since f is proper, f (X − U ) is a closed
subset of SpecA. So there exists an ideal a of A such that f (X − U ) = V (a).
On the other hand, we have X b = f −1 (V (I)). So f (X − X) b ∩ V (I) = ∅. In
particular, we have f (X − U ) ∩ V (I) = ∅, that is, V (a) ∩ V (I) = ∅. Since A is
I-adically complete, we have I ⊂ rad(A) by Proposition 1.5.7 and hence all the
maximal ideals of A lie in V (I). So no maximal ideal of A lies in V (a). This
can happen only when a = A, that is, V (a) = ∅. So f (X − U ) = ∅ and hence
X = U.
Theorem 7.10. (Grothendieck Existence Theorem) Let A be a noetherian ring,

I an ideal of A, X a scheme proper over SpecA, and X b the formal completion
∼
of X with respect to the sheaf of ideals I OX . If A is I-adically complete, then
for any coherent OXb -module F, there exists a coherent OX -module F unique
up to isomorphism such that F ∼ = F,b where Fb is the formal completion of F
with respect to the sheaf of ideals I ∼ OX .
Taking into account of the definition of coherent OXb -modules and Proposi-
tions 1.5.13 (ii) and 1.5.17, the theorem can be restated in the following form
which avoids mentioning formal completions:
Theorem 7.11. (Grothendieck Existence Theorem) Let A be a noetherian

ring, I an ideal of A and X a scheme proper over SpecA. For each k, let Xk
be the closed subscheme of X with ideal sheaf (I k )∼ OX . Fix notations by the

following commutative diagram (l ≥ k).
k i
(Xk , OXk ) → (X, OX )
ikl ↓ % il ↑i
lî
(Xl , OXl ) → b O b ),
(X, X
where all the arrows are the canonical morphisms. Let (Fk , φlk )k∈N be a family
consisting of an OXk -module Fk for each k and a morphism φlk : Fl → ikl∗ Fk
for each pair l ≥ k such that the following condition holds:
(a) Each Fk is a coherent OXk -module.
(b) φkk = idFk for each k and ilm∗ (φlk )◦φml = φmk for each triple m ≥ l ≥ k.
(c) For each pair l ≥ k, φlk induces an isomorphism i∗kl Fl ∼ = Fk , that is,
Fl ⊗OXl OXk ∼ = Fk .
If A is I-adically complete, then there exists a coherent OX -module F unique
up to isomorphism such that the family (Fk , φlk )k∈N is isomorphic to the family
(i∗k F, ψlk )k∈N , where for each pair l ≥ k, ψlk is the canonical morphism
i∗l F → ikl∗ i∗kl i∗l F ∼

= ikl∗ i∗k F.
The uniqueness part of the theorem follows directly from Corollary 7.9. Be-
fore proving the existence part of the theorem, we need a series of lemmas.
Recall that a coherent OXb -module F is called algebraizable if F ∼
= Fb for some
coherent OX -module F. In the following, we work under the assumptions in
Theorem 7.10.
Lemma 7.12. Let F and G be two coherent OXb -module and let u : F → G
be a morphism. If F and G are algebraizable, then keru, cokeru and imu are
algebraizable.
Proof. Let F and G be coherent OX -modules such that F = ∼ Fb and G = ∼ G.

b
By Corollary 7.9, there exists a morphism u0 : F → G such that u = ub0 .
By Proposition 1.5.13 (i), we have keru ∼ [0 . Hence keru is algebraizable.
= keru
Similarly, cokeru and imu are algebraizable.
Lemma 7.13. Let

0→F→G→H→0
be an exact sequence of OXb -modules. Suppose F and H are algebraizable co-
herent OXb -modules, then G is also an algebraizable coherent OXb -module.
Proof. Let F and H be coherent OX -modules such that F ∼

= Fb and H ∼
= H.
b
By Proposition 7.8 (ii), we have
(Ext1OX (H, F))∧ ∼

= Ext1OXc (H, F).
By Corollary 5.4 (ii), Ext1OX (H, F) is a finitely generated A-module. Since A is

I-adically complete, by Proposition 1.5.5, we have
(Ext1OX (H, F))∧ ∼

= Ext1OX (H, F).
So
Ext1OX (H, F) ∼
= Ext1O c (H, F).
X
By Proposition 1.15, the exact sequence
0→F→G→H→0
corresponds to an element in Ext1O c (H, F) and hence an element in Ext1OX (H, F).
X
Let
0→F →G→H→0
be the short exact sequence corresponding to this element in Ext1OX (H, F). By
Corollary 4.7, G is a coherent OX -module. Moreover, we have G ∼ = G.
b So G is
an algebraizable coherent OXb -module.
Corollary 7.14. Let u : F → G be a morphism of OXb -modules. Suppose G,

keru and cokeru are algebraizable coherent OXb -modules. Then F is an alge-
braizable coherent OXb -module.
Proof. Applying Lemma 7.12 to the morphism G→ cokeru, we see that imu is
coherent and algebraizable. Applying Lemma 7.13 to the short exact sequence
0 → keru → F → imu → 0,
we see that F is coherent and algebraizable.
Lemma 7.15. Consider a commutative diagram of morphisms of noetherian

schemes.
g
Y → Z
& ↓
SpecA
Let Yb and Z b be the formal completion of Y and Z with respect to the sheaves
of ideals I ∼ OY and I ∼ OZ , respectively, and let ĝ : Yb → Z
b be the morphism
induced by g. Suppose g is a proper morphism. If F is an algebraizable coherent
OYb -module, then gb∗ F is an algebraizable coherent OZb -module.
Proof. Suppose F is a coherent OY -module such that F∼

= F.
b By Corollary 7.6,
we have
ĝ∗ F ∼
= ĝ∗ Fb ∼ ∗ F.
= gd
By Theorem 5.1, g∗ F is a coherent OZ -module. So ĝ∗ F is coherent and alge-
braizable.
Lemma 7.16. Let g : Y → Z be a proper morphism of noetherian schemes, I

a coherent sheaf of ideals of OZ , Yb and Zb the formal completions of Y and Z
with respect to the sheaves of ideals IOY and I, respectively, and ĝ : Yb → Z b
the morphism induced by g. Suppose U is an open subscheme of Z such that g
induces an isomorphism of g −1 (U ) with U . Let J be a coherent sheaf of ideals of
OZ such that Z − U = supp(OZ /J ). Then for any coherent OZb -module G, the
kernel and the cokernel of the canonical morphism G → ĝ∗ ĝ ∗ G are annihilated
by Jcn for some natural number n.
Proof. The problem is local with respect to Z. We may assume Z = SpecB for
some noetherian ring B. Let I and J be the ideals of B corresponding to I and
b be the I-adic completion of B, J 0 = J B,
J , respectively. Let B b Z 0 = SpecB,
b
0 0 0 0 0 0 0
U the inverse image of U in Z , Y = Y ×Z Z , and g : Y ×Z Z → Z the
projection.
ĝ
Yb → Z b
↓ ↓
g0
Y 0 = Y ×Z Z 0 → Z 0
↓ ↓
g
Y → Z.
Then g 0 induces an isomorphism of g 0−1 (U 0 ) with U 0 and Z 0−U 0 = supp(OZ 0 /J 0∼ ).
Note that Yb and Z b are also the formal completions of Y 0 and Z 0 with respect
b ∼ and (I)
to the sheaves of ideals (I) b ∼ OY 0 , respectively. Moreover, by Corollary
1.5.14 (iv), for any n, we have
\
(J 0n )∼ ∼
= (J 0n )∼ OZb = (J n )∼ OZb ∼ \
= (J n )∼ ,
that is,
\
(J n )∼ ∼ \
= (J 0n )∼ .
To prove the lemma, we may thus replace Z by Z 0 . So we may assume B is I-

adically complete. Then by Proposition 1.5.20, there exists a finitely generated
B-module M such that G ∼ = (M ∼ )∧ . Let iY : Yb → Y and iZ : Z b → Z be the
canonical morphism. We have
ĝ ∗ G ∼
= ĝ ∗ (M ∼ )∧ ∼
= ĝ ∗ i∗Z M ∼ ∼
= i∗Y g ∗ M ∼ ∼
= (g ∗ M ∼ )∧ .
So by Corollary 7.6, we have
ĝ∗ ĝ ∗ G ∼
= ĝ∗ (g ∗ M ∼ )∧ ∼
= (g∗ g ∗ M ∼ )∧ .
One can verify that the canonical morphism
G → ĝ∗ ĝ ∗ G
is induced by the canonical morphism
M ∼ → g∗ g ∗ M ∼
by passing to formal completions. By Theorem 5.1, g∗ g ∗ M ∼ is coherent. Let

K be a finitely generated B-module such that
ker(M ∼ → g∗ g ∗ M ∼ ) ∼
= K ∼.
Since g induces an isomorphism of g −1 (U ) with U , K ∼ is supported outside

U . By Lemma 1.4.14, there exists an integer n such that (J n )∼ K ∼ = 0. By
Proposition 1.5.13 (i), (K ∼ )∧ is isomorphic to the kernel of G → ĝ∗ ĝ ∗ G. So the
kernel of G → ĝ∗ ĝ ∗ G is annihilated by Jcn . Similarly, the cokernel of G → ĝ∗ ĝ ∗ G
is annihilated by Jcn for some natural number n.
Lemma 7.17. Let A be a noetherian ring, I an ideal of A, S = SpecA,

f : X → S a projective morphism, OX (1) an invertible OX -module very ample
b the formal completion of X with respect to the sheaf of ideals I ∼ OX ,
over S, X
and F a coherent OXb module. For each k, let Xk be the closed subscheme of X
with ideal sheaf (I k )∼ OX and let ik : Xk → X, îk : Xk → X b → X be
b and i : X
the canonical morphisms. Set
OX (n) = OX (1)⊗n ,
Fk = î∗k F,
Fk (n) = Fk ⊗OXk i∗k OX (n),
F(n) = F ⊗OXc i∗ OX (n).
Then there exists an integer N such that for any n ≥ N , the canonical homo-
morphism
H 0 (X,
b F(n)) → H 0 (Xk , Fk (n))
is surjective, and
H p (X, F(n)) = 0
for any p ≥ 1.
∞
I k /I k+1 . Fix notations by the following cartesian diagram:
L
Proof. Let B =
k=0
g0
X 0 = X ×S SpecB → X1
f0 ↓ ↓ f1
g
SpecB → SpecA/I,
where X1 is the closed subscheme of X with ideal sheaf I ∼ OX , f1 : X1 →

SpecA/I the morphism induced by f , and g the morphism induced by the
homomorphism
A/I → B = A/I ⊕ I/I 2 ⊕ · · · .
Note that g and g 0 are affine morphisms and
∞
g∗0 OX 0 ∼
M
= f1∗ B ∼ = f1∗ (I k /I k+1 )∼ .
k=0
∞
I k F/I k+1 F on X.
b Since each I k F/I k+1 F annihilated by
L
Consider the sheaf
k=0
I and OX1 ∼
= OXb /IOXb , we may regard I k F/I k+1 F as an OX1 -module. It is a
coherent OX1 -module. Indeed, we have an exact sequence of OXb -modules
0 → I k F/I k+1 F → F/I k+1 F → F/I k F → 0.
Since each sheaf in this sequence is annihilated by I k+1 and OXk+1 ∼ = OXb/I k+1 OXb ,
we may regard this sequence as an exact sequence of OXk+1 -modules. Then
F/I k+1 F is identified with Fk+1 and F/I k F is identified with (ik,k+1 )∗ Fk , where
ik,k+1 : Xk → Xk+1 is the canonical morphism. By Proposition 1.5.17, Fk and
Fk+1 are coherent. So I k F/I k+1 F is a coherent OXk+1 -module. This implies it
∞
I k F/I k+1 F is a quasi-coherent OX1 -module.
L
is a coherent OX1 -module. So
k=0
∞
f1∗ (I k /I k+1 )∼ -module.
L
It is also an So by Proposition 1.4.12, there exists a
k=0
quasi-coherent OX 0 -module F 0 such that
∞
g∗0 F 0 ∼
M
= I k F/I k+1 F.
k=0
Note that F 0 is actually coherent. For any n, by Lemma 1.4.26, we have

∞
g∗0 (F 0 (n)) ∼
M
= I k F(n)/I k+1 F(n),
k=0
where F 0 (n) is the twisting of F 0 by the inverse image of OX (n) in X 0 . Since

g 0 is affine, the biregular spectral sequence
H p (X1 , Rq g∗0 (F 0 (n))) ⇒ H p+q (X 0 , F 0 (n))
degenerates by Corollary 4.8 and hence
H p (X1 , g∗0 (F 0 (n))) ∼

= H p (X 0 , F 0 (n)),
that is,
∞
H p (X1 , I k F(n)/I k+1 F(n)) ∼
M
= H p (X 0 , F 0 (n)).
k=0
By Theorem 4.12 (ii), there exists an integer N such that for any n ≥ N , we
have H p (X 0 , F 0 (n)) = 0 for any p ≥ 1. So we have
H p (X1 , I k F(n)/I k+1 F(n)) = 0
for any k ≥ 0, p ≥ 1, and n ≥ N . By induction on k, one can then show for any
k ≥ 1, p ≥ 1 and n ≥ N , we have
H p (X,
b F(n)/I k F(n)) = 0,
that is,
H p (Xk , Fk (n)) = 0.
In particular, the inverse system (H p (Xk , Fk (n)))k∈N satisfies the Mittag-Leffler
condition for any p ≥ 1 and any n ≥ N . On the other hand, associated to the
short exact sequence of sheaves
0 → I k F(n)/I k+1 F(n) → F(n)/I k+1 F(n) → F(n)/I k F(n) → 0,
we have a long exact sequence of cohomology groups, a part of which is
H 0 (Xk+1 , Fk+1 (n)) → H 0 (Xk , Fk (n)) → H 1 (X1 , I k F(n)/I k+1 F(n)).
Since H 1 (X1 , I k F(n)/I k+1 F(n)) = 0 for any k ≥ 1 and any n ≥ N , the homo-
morphism
H 0 (Xk+1 , Fk+1 (n)) → H 0 (Xk , Fk (n))
is surjective and hence the inverse system (H 0 (Xk , Fk (n)))k∈N also satisfies
the Mittag-Leffler condition for any n ≥ N . Furthermore, for any affine open
subscheme U of X1 , the homomorphism
H 0 (U, Fl (n)) → H 0 (U, Fk (n))
is surjective for any l ≥ k, and
H p (U, Fk (n)) = 0
for any p ≥ 1. So by Lemma 7.3, we have

b F(n)) ∼
H p (X, = inv. lim H p (Xk , Fk (n))
k
for any p and any n ≥ N . From this formula and the properties of the inverse
system (H p (Xk , Fk (n)))k∈N , we see that for any n ≥ N , the homomorphism
H 0 (X, F(n)) → H 0 (Xk , Fk (n))
is surjective and
H p (X, F(n)) = 0
for any p ≥ 1.
Corollary 7.18. Keep the assumptions and notations in Lemma 7.17. There
exists an integer N such that F(n) is generated by finitely many global sections
m
for any n ≥ N , that is, there exists an epimorphism OX b → F(n) for some
natural number m.
Proof. By Proposition 1.4.25, there exists an integer N such that F1 (n) is

generated by finitely many global sections for any n ≥ N . Moreover, by Lemma
7.17, we may choose N sufficiently large such that
Γ(X, F(n)) → Γ(X, F1 (n))

is surjective. Let s1 , . . . , sm be elements in Γ(X, F(n)) so that their images in

Γ(X, F1 (n)) generate F1 (n). We claim s1 , . . . , sm generates F(n). Consider the
morphism
m
OXb → F(n), ei 7→ si ,
m
where for each i, ei is the section of OX
b whose i-th component is 1 and whose
m
other components are 0. Let’s show OX b → F(n) is surjective. For any affine
open subscheme U = SpecB of X, let B b the (IB)-adic completion of B. Re-
placing B by B has no effect on our problem. So we assume B is (IB)-adically
b
complete. By Proposition 1.5.20, we have F(n)|U ∩Xb = (M ∼ )∧ for some finitely
generated B-module M . The homomorphism
m
OX
b (U ∩ X) → F(n)(U ∩ X)
b b
can be identified with a homomorphism
B m → M.
m
b → F(n), the homomorphism
By the choice of the morphism OX
(B/IB)m → M/IM
is onto. We have IB ⊂ rad(B) by Proposition 1.5.7. So
Bm → M
is onto by Nakayama’s lemma. Hence
((B m )∼ )∧ → (M ∼ )∧
is surjective by Propositions 1.4.2 (iii) and 1.5.13 (i), that is, the restriction of
m
OX
b → F(n)
to U ∩ X
b is surjective. Since U is arbitrary,
m
OX
b → F(n)
is surjective.
Corollary 7.19. Theorem 7.10 holds if X is projective over SpecA.
Proof. Fix an invertible OX -module OX (1) very ample over SpecA. For any
coherent OXb -module F, we may find an epimorphism OXb (n)m → F by Corollary
7.18. Applying Corollary 7.18 to the kernel of this epimorphism, we see that
there exists an exact sequence of the form
0
OXb (n0 )m → OXb (n)m → F → 0
0
for some natural numbers m, n, m0 , n0 . Both OXb (n)m and OXb (n0 )m are alge-
braizable coherent OXb -module since they are the formal completions of OX (n)m
0
and OX (n0 )m respectively. By Lemma 7.12, F is also algebraizable.
Finally, we are ready to prove Theorem 7.10. Let S be the family of those
closed subset Z of X such that for any closed subscheme (Z, OX ) of X whose
underlying topological space is Z, there exists a coherent OZb -module which is
not algebraizable. If S = ∅, we are done. Otherwise, since X is noetherian,
S has a minimal element. Let Z be a closed subscheme of X such that there
exists a coherent OZb -module F which is not algebraizable and such that the
underlying topological space of Z is a minimal element of S. Applying Chow’s
Lemma 1.4.18 to Z → SpecA, we get a projective morphism g : Y → Z such
that Y is projective over SpecA and there exists a dense open subset U of Z
such that g induces an isomorphism of g −1 (U ) with U . By Proposition 1.5.22
(ii), ĝ ∗ F is coherent. Since Y is projective over SpecA, ĝ ∗ F is algebraizable
by Corollary 7.19. Hence ĝ∗ ĝ ∗ F is algebraizable by Lemma 7.15. Let J be a
coherent sheaf of ideals of OZ so that Z − U = supp(OZ /J ). Let K be the
kernel of the canonical morphism
F → ĝ∗ ĝ ∗ F.
By Lemma 7.16, K is annihilated by Jcn for some natural number n. Let
h : (Z − U, (OZ /J n )|Z−U ) → (Z, OZ )
be the closed immersion. We claim that K = ĥ∗ ĥ∗ K. Indeed, by Proposition

1.5.13 (i), the sequence
0 → Jcn → OZb → (h∗ OZ−U )∧ → 0
is exact. By Corollary 7.6 (or proving directly), we have
(h∗ OZ−U )∧ ∼
= ĥ∗ OZ−U
\.
So we have an exact sequence
0 → Jcn → OZb → ĥ∗ OZ−U

\ → 0.
Hence
OZ−U ∼ cn \ .
b /J )|Z−U
\ = (OZ
Moreover, since K is annihilated by Jcn , it is supported in Z\

− U . (Indeed, if V
is an open subset of Z disjoint from Z − U , then 1 is a section in Jbn (V ). But
b \
then for any section x of K(V ), we have x = 1 · x = 0.) So we have
K = ĥ∗ ĥ∗ K.
By Proposition 1.5.22 (ii), ĥ∗ K is coherent. By the minimality of the underlying

topological space of Z in S, ĥ∗ K is algebraizable. So by Lemma 7.15 (or proving
directly), K = ĥ∗ ĥ∗ K is algebraizable, that is, the kernel of F → ĝ∗ ĝ ∗ F is

algebraizable. Similarly, the cokernel is also algebraizable. Applying Corollary
7.14 to the morphism F → ĝ∗ ĝ ∗ F, we see that F is algebraizable. This contradicts
to the fact that F is chosen to be non-algebraizable. Hence S = ∅. This proves
our theorem.
Theorem 7.20. Let A be a noetherian ring, I an ideal of A, S = SpecA, X and

Y two schemes proper over S. For each k, let Sk = SpecA/I k , Xk = X ×S Sk
and Yk = Y ×S Sk . If A is I-adically complete, then we have a canonical bijection
HomS (X, Y ) ∼
= inv. lim HomSk (Xk , Yk ),
k
where HomS (X, Y ) is the set of S-morphisms from X to Y , and HomSk (Xk , Yk )
is the set of Sk -morphisms from Xk to Yk .
Proof. For any S-morphism f : X → Y and any k, let fk : Xk → Yk be the

Sk -morphism induced by f . Then (fk ) ∈ inv. limk HomSk (Xk , Yk ). Let’s prove
the map
HomS (X, Y ) → inv. lim HomSk (Xk , Yk ), f 7→ (fk )
k
is a bijection. Let f, g : X → Y be two S-morphisms such that they induce the

same morphism Xk → Yk for each k. Consider their graphs
Γf , Γg : X → X ×S Y.
They are closed immersions since Y is necessarily separated over S. Let Jf and
Jg be their ideal sheaves. Then the ideal sheaf of the graph of the morphism
Xk → Yk induced by f and g is
Jf OXk ×Sk Yk = Jg OXk ×Sk Yk .
We have
OXk ×Sk Yk /Jf OXk ×Sk Yk = OXk ×Sk Yk /Jg OXk ×Sk Yk .
Hence
i∗k (OX×S Y /Jf ) = i∗k (OX×S Y /Jg ),
where ik : Xk ×Sk Yk → X ×S Y is the canonical morphism. By the uniqueness
part of Theorem 7.11, we must have
OX×S Y /Jf = OX×S Y /Jg .
This implies that f and g have the same graph and hence f = g. So the
canonical map
HomS (X, Y ) → inv. lim HomSk (Xk , Yk )
k
is injective.
Suppose fk : Xk → Yk (k ∈ N) is a family of Sk -morphisms such that for

any l ≥ k, the morphism fk is induced by fl through the base change Sk → Sl .
Let
Γfk : Xk → Xk ×Sk Yk
be the graph of fk , which is necessarily a closed immersion, and let Jk be its
the ideal sheaf. Then the family (OXk ×Sk Yk /Jk )k∈N satisfies the condition of
Theorem 7.11. So there exists a coherent OX -module F such that the fam-
ily (OXk ×Sk Yk /Jk )k∈N is isomorphic to the family (i∗k F)k∈N . The compatible
family of epimorphisms
OXk ×Sk Yk → OXk ×Sk Yk /Jk
can be identified with a compatible family of epimorphism
i∗k OX×S Y → i∗k F.
By Corollary 1.5.19 and Corollary 7.9, this family of epimorphisms is induced

by an epimorphism
OX×S Y → F.
Let J be its kernel and let Z be the closed subscheme of X ×S Y with ideal
sheaf J . We have
= i∗k F ∼
i∗k (OX×S Y /J ) ∼ = OXk ×Sk Yk /Jk .
So Zk = Z ×S Sk is the closed subscheme corresponding to the graph Γfk of

fk : Xk → Yk . Let pk : Zk → Xk and p : Z → X be the following compositions:
Zk → Xk ×Sk Yk → Xk ,
Z → X ×S Y → X.
Since Zk is the graph of fk , each pk is an isomorphism. By Lemma 7.21 below,

p is also an isomorphism. Let f : X → Y be the composition
p−1
X → Z → X ×S Y → Y.
Then each fk : Xk → Yk is induced by f through base change Sk → S. This

proves the map
HomS (X, Y ) → inv. lim HomSk (Xk , Yk )

k
is onto.
Lemma 7.21. Let A be a noetherian ring, I an ideal of A, S = SpecA, Z and

X two schemes proper over S. For each k, let Sk = SpecA/I k , Xk = X ×S Sk
and Zk = Z ×S Sk . Let p : Z → X be an S-morphism such that the morphism
pk : Zk → Xk induced by p through the base change Sk → S is an isomorphism
for each k. If A is I-adically complete, then f is also an isomorphism.
Proof. By Proposition 1.3.26 (vi), p : Z → X is proper. Let
p0
Z → X0 → X
be its Stein factorization. Since pk : Zk → Xk is an isomorphism, any point

z of Z lying in Zk is isolated in p−1 (p(z)). By Proposition 5.13, there exists a
neighborhood Z 0 of Zk in Z such that p0 |Z 0 : Z 0 → X 0 is an open immersion.
As in the proof of Corollary 7.9, Z is the only neighborhood of Zk in Z. Hence
p0 : Z → X 0 is an open immersion. But p0 is proper. So p0 is an isomorphism.
Hence p : Z → X is a finite morphism. By Proposition 1.4.11, to prove p is an
isomorphism, it suffices to show
p ] : OX → p ∗ OZ
is an isomorphism. By Corollary 7.9, Proposition 1.5.17 and Corollary 1.5.19,

it suffices to show
i∗k OX → i∗k p∗ OZ
is an isomorphism for each k, where ik : Xk → X is the closed immersion. Using
the fact that p is an affine morphism, one can verify that
i∗k p∗ OZ ∼
= pk∗ OZk .
So the morphism
i∗k OX → i∗k p∗ OZ
p]k : OXk → pk∗ OZk .
Since pk is an isomorphism, p]k is an isomorphism. This proves our assertion.
Bibliography
[1] Atiyah, M. F. and Macdonald, I. G.
Introduction to Commutative Algebra, Addison-Wesley, Readings, Mass. (1969).
[2] Hartshorne, R.
Algebraic Geometry, Springer-Verlag, New York (1977).
[3] Matsumura, H.
Commutative Algebra, W. A. Benjamin Co., New York (1970).
[4] Grothendieck, A. and Dieudonné, J.
Éléments de Géometrie Algébrique.

EGA I Le Langage des Schémas, Publ. Math. IHES 4 (1960).
EGA II Étude Globale Élémentaire de Quelques Classes de Morphismes, Ibid.
8 (1961).
EGA III Étude Cohomologique des Faisceaux Cohérents, Ibid. 11 (1961), 17
(1963).
EGA IV Étude Locale des Schémas et des Morphismes de Schémas, Ibid. 20
(1964), 24 (1965), 28 (1966), 32 (1967).
[5] Grothendieck, A. et al.
Séminare de Géometrie Algébrique.

SGA 1 Revêtements Étales et Groupe Fondamental, Lecture Notes in Math.
224, Springer-Verlag, Heidelberg (1971).
SGA 4 (with Artin, M. and Verdier, J. L.) Théorie des Topos et Cohomologie
Étale des Schémas, Lecture Notes in Math. 269, 270, 305, Springer-Verlag,
Heidelberg (1972-1973).
SGA 4 21 (by Deligne, P. et al.) Cohomologie Étale, Lecture Notes in Math.
569, Springer-Verlag, Heidelberg (1977).
217
218 BIBLIOGRAPHY
SGA 5 Cohomologie l-adique et Fonctions L, Lecture Notes in Math. 589,

Springer-Verlag, Heidelberg (1977).
SGA 7 (with Raynaud, M., Rim, D. S., Deligne P., and Katz, N.) Groupes de
Monodromie en Géométrie Algébrique, Lecture Notes in Math. 288, 340,
Springer-Verlag, Heidelberg (1972-1973).
Index
Abelian category, 7 Closed subscheme, 35

with enough injective objects, of affine schemes, 35
101 of projective spaces, 74
with enough projective objects, reduced induced, 37
101 Coboundary object, 99
Acyclic object, 106 Cocycle object, 99
Acyclic resolution, 106 Coeffaceable functor, 109
Additive category, 5 Coherent inverse system, 94
Additive functor, 8 Coherent OX -module, 55
Affine morphism, 30, 31, 61-62 Coherent OXb -module, 94
Affine scheme, 18, Cohomological functor, 108-109
closed subscheme of, 35 universal, 109
cohomology of, 160 Cohomology group, H i (X, F), 114
criterion for, 29 commutation with dir. lim, 119
Algebraizable coherent sheaf, 94 commutation with direct sum,
Alternating Čech complex, 145 119
commutation with flat base change,
Base change, 40 162-163
Bicomplex, 139 commutation with inv. lim, 196-
Bijective morphism, 4 197,
Birational morphism, 34 Finiteness of, 166, 170
Biregular spectral sequence, 129 I-adic completion of, 172,200
of affine schemes, 160
Cartan’s theorem, 152 of formal completions, 198, 200
Cartan-Eilenberg resolution, 140 of PnA , 164
Cartesian diagram, 39 Cohomology object,
Category, of a bicomplex, 139
abelian, 7 of a complex, 99
additive, 5 Coimage of a morphism, 7
Čech cohomology, 145, 149 Cokernel of a morphism, 6-7
direct limit over coverings, 149 of presheaves, 9
Čech complex, 144-145 of sheaves, 10
alternating, 145 Complex, 99
sheafified, 161 associate to a bicomplex, 139
Chow’s Lemma, 66 Connected scheme, 27
Closed immersion, 34, 64
ideal sheaf of, 64 Degenerate spectral sequence, 130
219
220 INDEX
Derivation operator, 99 for proper morphism, 170

Derived functor, Five Lemma, 14
Cohomology group, 114, Flasque sheaf, 116
Ext, 114 Flat
Ext, 114 module, 122
higher direct images, 115, morphism, 162
Tor, 115 sheaf, 180
Diagonal morphism, 41 Formal completion
Direct image, f∗ F, 11, 51 of a scheme, 87-88
quasi-coherence of, 59 of a sheaf, 88
Direct limit, dir. lim, 5 Free OX -module, 49
of a direct system of sets, 3 Function field, 27
of a direct system of sheaves, 49 Functor,
Direct product, 4 additive, 8
of sheaves, 49 coeffaceable, 109
Direct set, 2 cohomological, 108-109
Direct sum, 4 effaceable, 109
of sheaves, 49 exact, 100
Direct system, 4-5 left exact, 100
Dominant morphism, 34 right exact, 100
Effaceable functor, 109 Generic point, 25

Epimorphism, 4 Germ, 3
Exact functor, 100 Glueing, 38-39
Exact sequence, 7 Glueing data, 38-39
long, 100, 106 Godement resolution, 116
of presheaves, 11 Graded module, 68
of sheaves, 11 associated to a sheaf, Γ∗ (F), 73
short, 7-8 Graded ring, 21
split, 8 Graph of a morphism, 42-43
Ext, 114 Grothendieck Existence Theorem, 205-
Finiteness of, 170 206
I-adic completion of, 203
Ext, 114 Higher direct images, Ri f∗ F, 115
Coherence of, 170 coherence of,
Formal completion of, 202-203 for projective morphisms, 167
Extension of OX -modules, 120 for proper morphisms, 168
formal completion of, 200
Faithfully flat local freeness of, 185-186, 192
algebra, 126 of affine morphisms, 161
module, 125 of formal completions, 200
Fiber, 40 quasi-coherence of, 161
Fibred product, 39 Higher extension group, Ext, 114
Finite morphism, 31 Higher extension sheaf, Ext, 114
Finiteness Theorem Hom, 49
for projective morphism, 166 Homogeneous
INDEX 221
element, 21, 68 Kernel of a morphism, 6

ideal, 21 of presheaves, 9
Homotopic morphisms of complexes, of sheaves, 9
99 Koszul complex, 155
Homotopically equivalent complexes, cohomology of, 156-158
100
Left adjoint functor, 13
I-adic completion Left derived functor, 112
of a module, 79 Left exact functor, 100
of a ring, 79 Locally closed subset, 35
I-adically complete Locally free OX -module, 49
module, 79 Locally integral scheme, 28
ring, 79 Locally noetherian scheme, 29-30
I-filtration, 77 Locally ringed space, 17
I-stable filtration, 77 Long exact sequence, 100, 106
Ideal sheaf of a closed immersion, 64
Image of a morphism, 7 Main Theorem, 178
of presheaves, 9 Mapping cylinder, 181
of sheaves, 10 Minimal prime ideal, 26
Immersion, 35 Mittag-Leffler condition, 76
closed, 34 Monomorphism, 4
open, 34 Morphism,
Injective morphism, 4 affine, 30, 31
Injective object, 101 bijective, 4
Injective resolution, 101 birational 34
Integral scheme, 27 diagonal, 41
Inverse image, dominant, 34
f −1 F, 12 finite, 31
f ∗ F, 51 flat, 162
Inverse limit, inv. lim, 5 injective, 4
of an inverse system of sets, 5 locally of finite type, 30-31
of an inverse system of sheaves, of complexes, 99
49 of direct systems, 5
Inverse system, 5 of finite type, 31
Invertible OX -module, 50 of inverse systems, 5
very ample, 75 of locally ringed spaces, 18
Irreducible of presheaves, 3
component, 26 of ringed spaces, 18
scheme, 27 of schemes, 18
topological space, 24 of sheaves, 3
Isomorphism, 4 of spectral sequences, 128-129
of locally ringed spaces, 18 projective, 45
of presheaves, 3 proper, 43
of sheaves, 3 quasi-compact, 30, 31
quasi-finite, 176
Jacobson radical of a ring, 81 quasi-projective, 45
222 INDEX
quasi-separated, 43 Restriction, 1
separated, 41 Right adjoint functor, 13
surjective, 4 Right derived functor, 105-106, 112,
universally closed, 43-44 Right exact functor, 100
√ Ringed space, 17
Nilpotent radical of an ideal, a, 15
Noetherian scheme, 30 Scheme, 18
Noetherian topological space, 25 affine, 18
Normal scheme, 179 connected, 27
integral, 27
Open immersion, 34 irreducible, 27
Open subscheme, 20 locally integral, 28
OX -module, 48 locally noetherian, 29,30
noetherian, 30
Picard group, 50, 153
normal, 179
Presheaf, 1
over another scheme, 39
ProjS, 22
quasi-separated, 43
sheaf M ∼ on, 69
reduced, 27
Projection, 4, 39
separated 41
Projective module, 57
Scheme theoretic image, 65
Projective morphism, 45
Section, 1
Projective object, 101
Segre embedding, 48
Projective resolution, 101
Semicontinuity theorem, 184-185
Projective space, PnY , 45
Separated morphism, 41
Proper morphism, 43
Separated scheme, 41
Quasi-coherent OX -module, 55 Serre’s theorem, 75, 166
on ProjS, 73 Sheaf, 2
on SpecA, 56 associated to a presheaf, 9
Quasi-compact morphism, 30, 31 coherent, 55
Quasi-finite morphism, 176 flasque, 116
Quasi-projective morphism, 45 flat, 180
Quasi-separated morphism, 43 free, 49
Quasi-separated scheme, 43 generated by global sections, 75
Quotient, 7 invertible, 50
locally free, 49
Rank, 49 locally of finite presentation, 50
Reduced induced closed subscheme locally of finite type, 50
structure, 37 M ∼,
Reduced scheme, 27 on ProjS, 69
Regular spectral sequence, 129 on SpecA, 51
Resolution, of finite presentation, 50
acyclic, 106 of finite type, 50
Cartan-Eilengberg, 140 of ideals, 62
Godement, 116 of modules, 48-49
injective, 101 quasi-coherent, 55
projective, 101 Sheafified Čech complex, 161
INDEX 223
Short exact sequence, 7-8

Snake Lemma, 14
Spectral sequence, 127-128
associated to a bicomplex, 140
associate to a complex with a
decreasing filtration, 135-
138
biregular, 129
degenerate, 130
regular, 129
Spectrum, SpecA, 16
closed subscheme of, 35
sheaf M ∼ on, 51
Split short exact sequence, 8
Stalk, 3
Stein factorization, 176
Structure sheaf, 18
Sub-object, 7
Subscheme,
closed, 35
open, 20
Support, 63
Surjective morphism, 4
Tensor product of sheaves, ⊗, 49

Theorem of connectedness, 175
Tor, 115
Twisting sheaf, OProjS (1), 69, 75
Underlying topological space, 18

Universal cohomological functor, 109
Universally closed morphism, 43-44
Upper semicontinuous function, 184
Vanishing theorem, 166

Very ample invertible OX -module,
75
Zariski
Main Theorem, 178
Theorem of Connectedness, 175
Zariski topology
on ProjS, 22
on SpecA, 15
Zero morphism, 5
Zero object, 7

Algebraic Geometry (Lei Fu)

Uploaded by

Document Information

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

Algebraic Geometry (Lei Fu)

Uploaded by

Copyright:

Available Formats

Algebraic Geometry

Tsinghua University Press

In this book we study the cohomology of coherent sheaves on schemes. An ex-

1 Schemes and Coherent Sheaves 1

Schemes and Coherent

1.1 Presheaves and Sheaves

Let X be a topological space. A presheaf F of sets on X consists of the following

Then a presheaf F of sets on X is just a contravariant functor from the category

1. Let X be a topological space and A an abelian group. For every nonempty

S(U ) = {s : U → π −1 (U )|πs = id, and s is continuous},

and for every inclusion V ⊂ U of nonempty open subsets, define ρU V : S(U ) →

We say a presheaf F of sets (resp. abelian groups, resp. rings) is a sheaf if

is exact, where the second arrow is

and the third arrow is

The last two examples of presheaves given above are sheaves.

FP = dir. lim F(U ),

where the direct limit is taken over the family of neighborhoods of P . So

Let F and G be presheaves of abelian groups on X. A morphism of presheaves

For any point P ∈ X, φ induces a homomorphism on stalks φP : FP → GP . If we

Proposition 1.1 Let φ : F → G be a morphism of sheaves on a topological

φ(s, UP )|UP ∩UQ = (t, U )|UP ∩UQ = φ(s, UQ )|UP ∩UQ .

By the injectivity of φ(UP ∩ UQ ) that we have proved above, we must have

It induces a morphism dir. limi ui : dir. limi Ai → dir. limi A0i .

A category C is called an additive category if for any objects A, B and C in

(ii) Suppose we have an object P and morphisms p1 : P → A, p2 : P → B,

Then P together with the morphisms p1 : P → A and p2 : P → B is the

Proof. (i) It is easy to verify that

By the universal property of the direct product, we have k1 p1 + k2 p2 = idA×B .

p1 k2 = p1 idP k2 = p1 (k1 p1 + k2 p2 )k2

This proves (P, k1 , k2 ) has the required universal property.

Let C be an additive category and f : A → B a morphism in C. We say a

commutes. We often denote K by kerf and call it the kernel of f . Similarly we

A → B → C 0 is 0, there exists a unique morphism C → C 0 such that the

Proposition 1.3. Let

Proof. (i)⇒(ii) and (i)⇒ (iii) are obvious.

(idA − i1 p1 )i1 = i1 − i1 (p1 i1 ) = 0.

Since A2 is the cokernel of i1 : A1 → A, there exists a morphism i2 : A2 → A

Let F : C → D be a covariant functor between abelian categories. We say F

Hom(A, B) → Hom(F (A), F (B))

is a homomorphism. We then have

F (p1 )F (k1 ) = idF (A) ,

0 → F (A) → F (B) → F (C) → 0

is also a split short exact sequence.

Note that the category of abelian groups is an abelian category. We leave

Proposition 1.4. Let X be a topological space. Then the category of presheaves

(kerφ)(U ) = ker(φ(U ) : F(U ) → G(U )),

for every open subset U of X. The stalks of these presheaves at a point P ∈ X

Proposition 1.5. Let F be a presheaf on a topological space X. There exists

Proof. ` For any open subset U of X, define F + (U ) to be the set of functions

Proposition 1.6. Let X be a topological space. Then the category of sheaves

(kerφ)(U ) = ker(φ(U )),

cokerφ is the sheaf associated to the presheaf

and imφ is the sheaf associated to the presheaf

Moreover, for any P ∈ X, we have

Proof. It is easy to show that the presheaf defined by U 7→ ker(φ(U )) is a

Lemma 1.7. Let f : A → B be a morphism in an abelian category. Then

vertical arrow is a monomorphism. So the upper horizontal arrow is a monomor-

Corollary 1.8. Let X be a topological space.

We leave to the reader to prove the following corollary:

Corollary 1.9. Let X be topological space.

Let f : X → Y be a continuous map of topological spaces. For any sheaf F

U 7→ dir. lim G(V )

as follows: Since f −1 f∗ F is the sheaf associated to the presheaf

U 7→ dir. lim (f∗ F)(V ) = dir. lim F(f −1 (V )),