Santos Main

Triangulations of polytopes and point
configurations
Francisco Santos, Mónica Blanco and Federico Castillo
Abstract These lecture notes summarize the theory of triangulations of polytopes

and point configurations, including the notions of regular triangulations and bistellar
flips, using tools from oriented matroid theory, and leading to the definition of the
secondary polytope of a point configuration. They are self-contained except that
some, but not all, of the proofs are omitted. They cover more or less Chapters 2
and 4 (plus Section 5.1) of the book “Triangulations: Structures for Algorithms and
Applications” by J. A. De Loera, J. Rambau, and F. Santos (Springer, 2010).
The idea of these notes is to learn from the basics. Our motto is: pasito a pasito,
suave suavecito.
The content of this course is based on the following sections of the book on
triangulations coauthored by Francisco Santos [1]: Chapter 2, Sections 2.1 to 2.5;
Chapter 4, Sections 4.1, 4.3 and 4.4; Chapter 5, Section 5.1. All throughout we
give pointers to that book so that the interested reader can find more details and
explanations.
Additional references that we recommend are:
• For polytope theory, including an introduction to oriented matroids similar in spirit
to the one in Section 2, we recommend Ziegler’s book “Lectures on polytopes”
[5].
Francisco Santos
Universidad de Cantabria, e-mail: francisco.santos@unican.es. Work of F. Santos was sup-
ported by grants MTM2017-83750-P and PID2019-106188GB-I00 of the Spanish State Research
Agency (AEI/10.13039/501100011033).
Mónica Blanco
e-mail: m.blanco.math@gmail.com
Federico Castillo
Max Planck Institute for Natural Sciences, e-mail: castillo@mis.mpi.de
1
2 Francisco Santos, Mónica Blanco and Federico Castillo
• For a comprehensive overview of several topics on triangulations that complement

the contents of this course, including references to the original sources, see
the chapter “Subdivisions and triangulations of polytopes” in the Handbook of
Discrete and Computational Geometry [3].
• Another excellent source is the book “Lectures in combinatorial geometry” [4]
by Rekha Thomas. Chapters 1 to 6 are an introduction to polytope theory, and
Chapters 7 and 8 overlap heavily with the contents of this course.
0 Introduction to polytopes and affine geometry
We start with a brief summary of the basic concepts of affine and convex combina-
tions. For more details see [1, Section 2.1] or the first two chapters of [5].
Definition 1
Let A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 be a finite subset.
• An affine combination of the points of A is any point of the form 𝜆1 𝑝 1 + · · · +𝜆 𝑛 𝑝 𝑛
Í
with 𝜆𝑖 ∈ R and 𝜆𝑖 = 1.
• An affine dependence between the pointsÍ of A is a valid expression of the form
𝜆1 𝑝 1 + · · · + 𝜆 𝑛 𝑝 𝑛 = 0 with 𝜆𝑖 ∈ R and 𝜆𝑖 = 0. The points of A are affinely
independent if their only affine dependence is the trivial one 𝜆1 = · · · = 𝜆 𝑛 = 0,
and affinely dependent otherwise.
• A subset of R𝑑 is an affine (sub)-space, or a flat, if it is closed under affine
combinations.
• The affine span of A is the smallest affine space containing A. Equivalently, it
is the set of points that are affine combinations of points of A. It is denoted by
aff(A).
• The dimension of an affine space is the maximum number of affinely independent
points minus 1. The dimension of A is the dimension of aff (A).
Observe that every affine combination 𝑝 = 𝜆1 𝑝 1 + · · · + 𝜆 𝑛 𝑝 𝑛 of points gives rise

to an affine dependence between them and the new one: 𝜆1 𝑝 1 + · · · + 𝜆 𝑛 𝑝 𝑛 − 𝑝 = 0.
Conversely, each point with nonzero coefficient in an affine dependence can be
written as an affine combination of the rest.
Definition 2 Let A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 be a finite subset.

• A convex combination
Í of the points of A is any point of the form 𝜆1 𝑝 1 + · · · +𝜆 𝑛 𝑝 𝑛
with 𝜆𝑖 ∈ R, 𝜆𝑖 = 1 and 𝜆𝑖 ≥ 0. That is, an affine combination with non-negative
coefficients.
• A subset of R𝑑 is a convex set if it is closed under convex combinations.
• The convex hull of A is the smallest convex set containing A. Equivalently, it is
the set of points that are convex combinations of points of A. It is denoted by
conv(A).
Triangulations of polytopes and point configurations 3
• A polytope P ⊂ R𝑑 is the convex hull of a finite set of points A. Its dimension 𝑘 is

the dimension of its affine span aff(P) = aff(A), and we say that P is a 𝑘-polytope.
Polytopes of dimensions 1, 2, 3 are called segments, polygons and polyhedra (in
the classical sense).
Example: Affine and convex hulls

Let us see two small examples in R2 . For two distinct points 𝑝 1 and 𝑝 2 , the set of
affine combinations (affine span) is the unique straight line (dashed in Figure 1, left)
that passes through them, and the set of convex combinations (convex hull) is the
segment with 𝑝 𝑖 as endpoints. The two points span an affine space and a convex hull
of dimension 1.
For three non collinear points 𝑝 3 , 𝑝 4 , 𝑝 5 ∈ R2 , their affine span is the entire plane
2
R and their convex hull is the triangle with the 𝑝 𝑖 as vertices. The dimension in this
case is 2.
In both pictures, the point 𝑝 represents a point in the convex hull (hence in the
affine span), and 𝑞 is a point in the affine span, but not in the convex hull.
a↵(p1 , p2 ) = { 1 p1 + 2 p2 : 1 + 2 = 1}
1 <0
p4
1 =0
2 >1 q
i <1 2 =1
1 >1
1 =1
2 =0
i >0 q p
2 <0 p2 p3
p1 p
p5
Fig. 1 Affine and convex hulls in dimension 2.
Theorem 3 (Minkowski-Weyl [5, Theorem 1.1]) Every polytope is bounded and

equals the set of solutions of a finite set of affine inequalities:


 𝑎 11 𝑥1 + · · · + 𝑎 1𝑛 𝑥 𝑑 ≤ 𝑏 1 



 𝑑 .. 

P = (𝑥 1 , . . . , 𝑥 𝑑 ) ∈ R : .
 
𝑎 𝑚1 𝑥1 + · · · + 𝑎 𝑚𝑛 𝑥 𝑑 ≤ 𝑏 𝑚 

 
 
for some (𝑎 𝑖 𝑗 ) ∈ R𝑚×𝑛 and some (𝑏 𝑖 ) ∈ R𝑚 .
Conversely: the set of solutions of any finite system of affine inequalities, if
bounded, is a polytope.
Related to these inequalities are the faces of a polytope:
Definition 4 Let P ⊂ R𝑑 be a polytope. For each linear functional 𝑓 : R𝑑 → R we

call face of P induced by 𝑓 the set
P 𝑓 := {𝑥 ∈ P : 𝑓 (𝑥) ≥ 𝑓 (𝑦) ∀𝑦 ∈ P},
that maximizes 𝑓 in P. See Figure 2 for an example.

The empty set is considered a face of dimension −1. The faces of dimension 0
are called vertices, the 1-dimensional ones are edges, and the faces of dimension
𝑘 − 1 (for a 𝑘-polytope) are called facets. Notice that the entire polytope P is a face
of itself (by letting 𝑓 = 0) and that every face of a polytope is a polytope.
The boundary of P is the union of all its proper faces (those of dimension < 𝑘).
The faces of a polytope are a poset (partially ordered set) by containment. In fact,
this poset is a graded distributive lattice.
Notation
We will be constantly dealing with a set of points 𝑝 1 , . . . , 𝑝 𝑛 and its dif-
ferent subsets. In order to simplify notation, we will identify each point with
its label i ≡ 𝑝 𝑖 ≡ {𝑝 𝑖 }. Similarly, for 𝑖1 , . . . , 𝑖 𝑚 ∈ {1, . . . , 𝑛}, we will write
i1 . . . im ≡ {𝑝 𝑖1 , . . . , 𝑝 𝑖𝑚 } for the set of the points corresponding to those labels,
and conv(i1 . . . im ) ≡ conv{𝑝 𝑖1 , . . . , 𝑝 𝑖𝑚 } for their convex hull. This notation will
also appear in the figures.
Example: faces of a polytope
The faces of the polygon P = conv{𝑝 1 , 𝑝 2 , 𝑝 3 }, with 𝑝 1 = (0, 0), 𝑝 2 = (1, 0) and
𝑝 3 = (0, 1) (see Figure 2), are:
• ∅ as the only face of dimension −1;
• P−𝑥−𝑦 = conv(1), P 𝑥 = conv(2) and P 𝑦 = conv(3) as its vertices;
• P−𝑦 = conv(12), P−𝑥 = conv(13) and P 𝑥+𝑦 = conv(23) as its edges.
• P = conv(123) as its only face of dimension 2.
The polytope that is of most interest to us is the one with the minimum number
of vertices:
Definition 5 A 𝑘-dimensional simplex is a 𝑘-polytope with exactly 𝑘 + 1 vertices.

Equivalently, it is the convex hull of a set of 𝑘 affinely independent points. Simplices
of dimensions 0, 1, 2, 3 are points, segments, triangles and tetrahedra, respectively.
Every face of a simplex is a simplex, and each subset of vertices defines a face.
That is, a 𝑘-simplex has 𝑘+1𝑖+1 faces of each dimension 𝑖 = −1, 0, . . . , 𝑘. (The poset
of faces of a simplex is the Boolean poset).
−x − y x
3 3 3
y
1 2 1 2 1 2 3
x+y x
3 3 3
y 1 2
1 2 1 2 1 2
Fig. 2 Faces of a polygon and the normals of the corresponding functionals.
1 Triangulations and subdivisions of point configurations
From now on let us consider A := {𝑝 1 , . . . , 𝑝 𝑛 } a finite set of points in R𝑑 that is

𝑑-dimensional (that is, its affine span is the whole R𝑑 ). We will call a set like A, with
labeled and possibly repeated points, a point configuration, and we will be interested
in some combinatorial and geometrical aspects of it. Let us also denote from now
on P := conv(A).
Remark
There is no (much) loss of generality in assuming that A is full-dimensional because
if A ⊂ R𝐷 and dim(A) = 𝑑 < 𝐷 then we can forget some coordinates and get a
𝑑-dimensional configuration A0 in R𝑑 that is affinely isomorphic to A; all we are
going to do is invariant under affine isomorphism.
Nonetheless, sometimes we need to consider a subset B of A as a configuration
in itself and, of course, it may be lower dimensional.
For a point configuration, we need to consider the concept of face in a slightly

different manner.
Definition 6 ([1, Def. 2.1.7]) Let B be a subset of A. We say that B is a face of A if
there is a face F of the polytope P = conv(A) for which B = A ∩ F. In particular,
F = conv(B), but notice that B may include points that are not vertices of F.
Running example
The point configuration depicted in Figure 3, will serve as our running example. The
faces of this point configuration are: ∅ of dimension −1; 1, 2, 3 and 4 of dimension
0; 12, 13, 24 and 34 of dimension 1; and the unique two-dimensional face 12345,
which includes the non-vertex 5.
3 4
Fig. 3 Point configuration 5

A0 : 𝑝1 = (0, 0), 𝑝2 = (3, 0),
𝑝3 = (0, 3), 𝑝4 = (3, 3),
𝑝5 = (1, 1). 1 2
Definition 7 In terms of point configurations, we say that a set A is a simplex if its

elements are affinely independent. In particular, if A is a 𝑘-simplex, it has cardinality
𝑘 + 1.
Notice that we call 𝑘-simplex both a set of 𝑘 + 1 affinely independent points and the
polytope that is their convex hull.
Definition 8 ([1, Def. 2.3.1]) A (polyhedral) subdivision of A is a finite collection
S = {S1 , . . . , S𝑚 } of subsets of A, called cells, such that:
(DP) For each 𝑖 ∈ {1, . . . , 𝑚}, conv(S𝑖 ) is a 𝑑-dimensional polytope;
Ð
(UP) conv(A) = 𝑖 conv(S𝑖 )
(IP) If 𝑖 ≠ 𝑗 then
• S𝑖 ∩ S 𝑗 is a common (possibly empty) proper face of both S𝑖 and S 𝑗 ,
• and conv(S𝑖 ) ∩ conv(S 𝑗 ) = conv(S𝑖 ∩ S 𝑗 ).
The trivial subdivision is the subdivision {A} with a single cell, the set A. A
subdivision of A is a triangulation if every cell is a simplex. We call faces of a
subdivision the faces of its cells.
Note that we are calling cells only the pieces of full-dimension. This deviates
from the standard use in much of the literature on cell complexes, where the word
“cells” refers also to the lower dimensional pieces.
Since every subset of a simplex is a face of it, the property (IP) (as in Definition 8)
applied to triangulations reduces to
(IP) If 𝑖 ≠ 𝑗 then
conv(S𝑖 ) ∩ conv(S 𝑗 ) = conv(S𝑖 ∩ S 𝑗 ) (1)
Running example: subdivisions and triangulations

In the A0 of Figure 3, {1245, 1345} and {124, 134} are subdivisions (see Figure 4),
and we consider them different subdivisions. Observe that the second one does not
use point 5, but that is ok.
In A0 , the polytope conv(1245) is a 2-simplex, but the set 1245 is not. Therefore,
{124, 134} is a triangulation of A0 but {1245, 1345} is not.
{1245, 1345} {124, 134}

3 4 3 4
Fig. 4 Two different subdi- 5 5

visions of A0 with the same
polytopal cells. 1 2 1 2
However, {1245, 134} is not a subdivision: whereas the polytopal triangles

conv(1245) and conv(134) intersect in a common face (the segment conv(14)),
the sets 1245 and 134 do not: their intersection is 14, which is not a face of 1245.
Definition 9 ([1, Def. 2.3.8]) Let T = {T1 , . . . , T𝑙 } and S = {S1 , . . . , S𝑚 } be two

subdivisions of A. We say T is a refinement of S if for each 𝑖 ∈ {1, . . . , 𝑙}, there
exists 𝑗 ∈ {1, . . . , 𝑚}, such that T𝑖 ⊆ S 𝑗 . In that case we write T S.
Refinement of subdivisions of a point configuration A is a partial order (poset),

where the maximal element is the trivial subdivision {A} and the minimal elements
are the triangulations.
Running example: refinement of subdivisions

Notice that refinement of subdivisions is on the sets and not on the polytopes.
Figure 5 shows three subdivisions R, S and T of A0 . The only refinements are
R S and T S.
3 4 3 4 3 4
5 5
1 2 1 2 1 2
R = {125, 135, 234, 235} S = {1235, 234} T = {123, 234}

Fig. 5 Refinements among subdivisions of A0 .
Remark
We may sometimes refer to subdivisions or triangulations of a polytope, in which
case we are refering to subdivisions on the point configuration consisting of the
vertices of the polytope.
Problems
The easier problems are marked with an *.

1.1 *Draw all triangulations of one of the following three point configurations.
Clue: there are 18, 14, and 4, respectively.
1 6
2
3 4
5
2 5
5
4 6
1 3 1 2
3 4
The m. o. a. e.1 A hexagon 5 points in the plane
1.2 a. *Construct all the subdivisions of a triangular prism (the product of an equilateral triangle
and a segment perpendicular to it). Draw also the poset of subdivisions.
Clue: modulo symmetries there are only three subdivisions: triangulations (all equivalent
under symmetry), the trivial subdivision, and another class.
b. Describe all the triangulations of Δ 𝑘 × Δ1 , the product of a 𝑘-simplex and a segment. (See
[1, Sec. 6.2.1] for more details)
Clue: there is a natural bĳection between triangulations of Δ 𝑘 × Δ1 and permutations of
{1, . . . , 𝑘 + 1}, where the 𝑘 + 1 symbols {1, . . . , 𝑘 + 1} correspond to the vertices of Δ 𝑘 .
1.3 *Let A consist of 4 points in R2 , not all in a line (that is, they affinely span R2 ). Show that
there are three subdivisions of A, no matter what the points are: the trivial subdivision and two
triangulations.
1.4 *Do the same for 5 points in R3 , not all in a plane.
1.5 Do the same for 𝑑 + 2 points in R𝑑 , not all in a hyperplane. For this:
a. Observe that there is (modulo multiplication by a constant) a unique affine dependence
among the points. Let A+ and A− be the sets of points with positive and negative coefficient
in it.
b. Show that every full-dimensional simplex in A is of the form A \ { 𝑝𝑖 } where 𝑝𝑖 is in
A+ ∪ A− .
c. Show that two such simplices A \ { 𝑝𝑖 } and A \ { 𝑝 𝑗 } intersect properly (that is, they satisfy
(IP)) if and only if 𝑝𝑖 and 𝑝 𝑗 lie in the same subset A+ or A− .
d. Show that {A \ { 𝑝𝑖 } : 𝑝𝑖 ∈ A+ } and {A \ { 𝑝𝑖 } : 𝑝𝑖 ∈ A− } are two triangulations of A,
and conclude from the above that they are the only non-trivial subdivisions of it.
Clue: Lemma 20 below is essentially solving this exercise.
1.6 a. *Consider the vertex set of a 1 × 2 grid. That is, let A = { (𝑖, 𝑗) : 𝑖 ∈ {0, 1}, 𝑗 ∈ {0, 1, 2} }.
Construct all the triangulations of A. How many of them use all the points?
1 “Mother Of All Examples”. See [1]

b. Consider the vertex set of a 1 × 𝑘 grid. That is, let A = { (𝑖, 𝑗) : 𝑖 ∈ {0, 1}, 𝑗 ∈
{0, 1, . . . , 𝑘 } }. Show that A has 2𝑘
𝑘 triangulations that use all points.
1.7 *Show that the 3-dimensional cube [0, 1] 3 has six symmetry classes of triangulations, one with
five tetrahedra and five with six tetrahedra. (See [1, Sec. 6.3.4] for more details)
Clue: first show that every triangulation uses either one of the big, regular, tetrahedra inscribed
in the cube, or one of the four long diagonals between opposite vertices. In the first case this
completely determines the triangulation; in the second case look at the possible configurations
of tetrahedra around that diagonal.
1.8 For each 𝑛 ≥ 3, let 𝑇𝑛 be the set of all triangulations of a convex 𝑛-gon. The goal is to show
1 2𝑛−4
that |𝑇𝑛 | equals 𝑛−1 𝑛−2 . For this:
a. Show that every triangulation of the 𝑛-gon has 2𝑛 − 3 edges (the 𝑛 edges of the 𝑛-gon plus
𝑛 − 3 internal diagonals). Deduce that the average degree of vertices in a triangulation is
4 − 𝑛6 .
Note: The degree here is in terms of graphs: the number of edges a particular vertex belongs
to.
b. Look at a particular vertex, say vertex 1. Explain why the average degree of *that vertex*
among *all triangulations* is also 4 − 𝑛6 .
c. Consider the map 𝑓 : 𝑇𝑛+1 → 𝑇𝑛 that “contracts” the edge {1, 𝑛 + 1}, converting each
triangulation of the (𝑛 + 1)-gon into a triangulation of the 𝑛-gon.2 Show that for each
T ∈ 𝑇𝑛 , the number of triangulations on 𝑓 −1 ( T) equals degT (1) (the degree of vertex 1 in
T).
d. Conclude from (b) and (c) that the average size of 𝑓 −1 ( T) is exactly 4𝑛−6
𝑛 .
e. Deduce that
4𝑛 − 6
|𝑇𝑛+1 | = |𝑇𝑛 |,
𝑛
and from this prove the formula for |𝑇𝑛 | by induction.
1 2𝑛

Remark: the numbers 𝐶𝑛 = 𝑛+1 𝑛 are called Catalan numbers. The exercise shows that |𝑇𝑛 |
equals the (𝑛 − 2)-nd Catalan number 𝐶𝑛−2 . 3
1.9 *Show bĳections between the following two sets.

a. Triangulations of a convex (𝑛 + 2)-gon. For example, for 𝑛 = 3 there are 𝐶3 = 14 63 = 5 of

them.
b. Ways of putting 𝑛 pairs of parentheses in the (𝑛 + 1)-term multiplication 𝑎1 · · · 𝑎𝑛+1 , in
order to break it into 2-term multiplications. For example, for 𝑛 = 3 there are five ways:
( ( (𝑎1 𝑎2 ) 𝑎3 ) 𝑎4 ), ( (𝑎1 (𝑎2 𝑎3 )) 𝑎4 ), (𝑎1 ( (𝑎2 𝑎3 ) 𝑎4 )), (𝑎1 (𝑎2 (𝑎3 𝑎4 ))), ( (𝑎1 𝑎2 ) (𝑎3 𝑎4 )).
Clue: the bĳection becomes easier to visualize if you draw the 𝑛-gon with a long horizontal
edge 1n on top and a chain of 𝑛 − 1 small edges below it. Think of half an arepa; the edge 1n
corresponds to the cut you made:
1.10 Show a bĳection between the above two sets and Dyck paths of length 2𝑛: monotone paths
in the integer grid going from (0, 0) to (𝑛, 𝑛) and staying always above the diagonal 𝑥 =
𝑦. For example, for 𝑛 = 3 there are five paths: 𝑁 𝑁 𝑁 𝐸 𝐸 𝐸, 𝑁 𝑁 𝐸 𝐸 𝑁 𝐸, 𝑁 𝐸 𝑁 𝑁 𝐸 𝐸,
𝑁 𝐸 𝑁 𝐸 𝑁 𝐸, 𝑁 𝑁 𝐸 𝑁 𝐸 𝐸. Here 𝑁 and 𝐸 denote a “north” and “east” step, respectively.
2 We consider the vertices of the 𝑛-gon to be labeled from 1 to 𝑛 in cyclic order.

3 See [1, Thms. 1.1.2 and 1.1.3] for more details on this and the next two exercises
2 Oriented matroids
2.1 Oriented matroids of point configurations
In this section we look at the oriented matroid of a point configuration, which is a

combinatorial object that stores information about the relative positions of the points.
See [1, Sect. 4.1] for more details.
Let A := {𝑝 1 , . . . , 𝑝 𝑛 } be a full-dimensional point configuration in R𝑑 .
We are goint to consider two types of vectors of R𝑛 related to A:
• Affine dependences (see Definition 1) of the points of A, considered as vectors
𝜆 = (𝜆1 , . . . , 𝜆 𝑛 ) ∈ R𝑛 ,
Í
for 𝜆𝑖 ∈ R such that 𝜆𝑖 = 0 and 𝜆1 𝑝 1 + · · · + 𝜆 𝑛 𝑝 𝑛 = 0.
• For any affine functional 𝑓 : R𝑑 → R, the vector

𝑓 ( 𝑝 1 ), . . . , 𝑓 ( 𝑝 𝑛 ) ∈ R𝑛
which we call affine evaluation of A by 𝑓 .

Lemma 10 ([1, Lemma 4.1.34]) The sets of affine dependences and affine evalua-
tions of A are complementary linear subspaces of R𝑛 of dimensions 𝑛 − 𝑑 − 1 and
𝑑 + 1, respectively, where 𝑛 = |A| and 𝑑 = dim(A).
Proof Every affine dependence is orthogonal to every affine evaluation since
Í Í
𝜆 𝑖 𝑓 ( 𝑝 𝑖 ) = 𝑓 (𝜆𝑖 𝑝 𝑖 ) = 0.
Since A is full-dimensional, there exist 𝑑 + 1 points of A that are affinely inde-
pendent. Each of the remaining 𝑛 − 𝑑 − 1 points of A can be written as an affine
combination of them. Those 𝑛 − 𝑑 − 1 affine dependences are linearly independent
vectors of R𝑛 . Thus, affine dependences have dimension at least 𝑛 − 𝑑 − 1.
Finally, every affine functional is of the form 𝑓 (𝑥1 , . . . , 𝑥 𝑑 ) = 𝑎 0 + 𝑎 1 𝑥1 + · · · +
𝑎 𝑑 𝑥 𝑑 , hence affine evaluations are generated by the evaluations corresponding to
a constant functional and to the coordinate functionals. The fact that A is full-
dimensional implies these 𝑑 + 1 evaluations are linearly independent vectors in R𝑛 .
Thus, evaluations have dimension at least 𝑑 + 1.
Since the spaces are orthogonal, those are their exact dimensions.
Of these dependences and evaluations of a point configuration, we are sometimes
only interested in the signs of their entries:
Definition 11 ([1, Def. 4.1.1]) The signature of a vector 𝑥 = (𝑥 1 , . . . , 𝑥 𝑛 ) ∈ R𝑛 is
the vector of signs of 𝑥. That is, the vector
𝑥𝑖
(𝜖 1 , . . . , 𝜖 𝑛 ), 𝜖𝑖 = ∈ {−1, 0, +1},
|𝑥 𝑖 |
with 0/0 taken as 0.

A signature (𝜖1 , . . . , 𝜖 𝑛 ) can be uniquely represented by the pair (𝐸 + , 𝐸 − ) where
𝐸 + := {𝑖 : 𝑥𝑖 > 0} and 𝐸 − := {𝑖 : 𝑥𝑖 < 0}.
That is, there is a bĳection between the 3𝑛 possible signatures of vectors in R𝑛 and
pairs (𝐸 + , 𝐸 − ) of disjoint subsets of {1, . . . , 𝑛}.
Signatures of affine dependences [1, Sec. 4.1.1]

Í
For a non-zero affine dependence 𝜆1 𝑝 1 + · · · + 𝜆 𝑛 𝑝 𝑛 = 0, since 𝜆𝑖 = 0 we have
both positive and negative coefficients, and we can write:
1 ∑︁ 1 ∑︁ ∑︁ ∑︁
𝑝 := · 𝜆𝑖 𝑝 𝑖 = · (−𝜆𝑖 𝑝 𝑖 ); where 𝛼 := 𝜆𝑖 = (−𝜆𝑖 )
𝛼 𝑖: 𝜆 >0 𝛼 𝑖: 𝜆 <0 𝑖: 𝜆𝑖 >0 𝑖: 𝜆𝑖 <0
𝑖 𝑖
Thus, 𝑝 is both a convex combination of the sets {𝑝 𝑖 : 𝜆𝑖 > 0} and {𝑝 𝑖 : 𝜆𝑖 < 0}.
In particular, it is a point in the intersection of their respective convex hulls.
That is, for each non-zero dependence 𝜆 = (𝜆1 , . . . , 𝜆 𝑛 ) ∈ R𝑛 of a point config-
uration A, its signature defines a pair (A+𝜆 , A−𝜆 ) where A+𝜆 (resp. A−𝜆 ) is the set of
points of A with positive (resp. negative) coefficients in 𝜆. These two sets have the
property that conv(A+𝜆 ) ∩ conv(A−𝜆 ) ≠ ∅.
In particular, we are interested in the affine dependences among the smallest
possible subsets of points, which are called circuits.
Definition 12 ( [1, Def. 2.4.1] ) A circuit is a set C of affinely dependent points such
that every proper subset is affinely independent.
In particular, for a circuit C = {𝑐 1 , . . . , 𝑐 𝑘 }, we have 𝑘 = dim(C) + 2 and there is a
unique (up to rescaling) affine dependence 𝜆 = (𝜆1 , . . . , 𝜆 𝑘 ) between the points of
C. Since every proper subset of C is affinely independent, 𝜆 has no zero entries. Its
signature determines a partition (C+ , C− ) of the point configuration C that is called
the Radon partition and is the unique partition of C into two disjoint subsets such
that conv(C+ ) ∩ conv(C− ) ≠ ∅.
Definition 13 An oriented circuit is the signature of the unique dependence of a

circuit.
Example: signatures of circuits

In dimension 2, a circuit can have up to 4 points (two points more than the dimension).
A circuit {𝑐 1 , 𝑐 2 } with two elements can only have signature (1, 2): that is, 1 = 2.
A circuit {𝑐 1 , 𝑐 2 , 𝑐 3 } with three elements can only have signature (1, 23): that is,
1 ∈ conv(23).
A circuit {𝑐 1 , 𝑐 2 , 𝑐 3 , 𝑐 4 } with four elements has two options for its signature:
(12, 34), in which case conv(12) ∩ conv(34) ≠ ∅ (two segments that intersect); and
(1, 234), which implies 1 ∈ conv(234).
See Figure 6 for an example of each type of 2-dimensional circuit.
(1, 2) (1, 23) (12, 34) (1, 234)

3 3
1=2 2 1
1 2
3 1
2
4 4
Fig. 6 Possible signatures of circuits in dim ≤ 2.
Signatures of affine evaluations [1, Sec. 4.1.2]
A non-constant affine functional 𝑓 : R𝑑 → R is 0 in the points 𝑥 ∈ R𝑑 of an

affine hyperplane (of dimension 𝑑 − 1) 𝐻0 , is positive in one of the open halfspaces
defined by that hyperplane, say 𝐻+ , and is negative in the remaining open halfspace
𝐻− .
Taking A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 a point configuration and the affine evaluation
( 𝑓 ( 𝑝 1 ), . . . , 𝑓 ( 𝑝 𝑛 )) ∈ R𝑛 , its signature defines a pair (A+𝑓 , A−𝑓 ) where A+𝑓 (resp.
A−𝑓 ) is the set of points of A where 𝑓 takes a positive (resp. negative) value. That is,
the signature vector tells us which points lie in a hyperplane, and which points lie in
each of the open halfspaces.
H− H0
H− H0
H+
3 4 3 4 H+
5 5
Fig. 7 Affine evaluation
signatures on A0 . The one 1 2 1 2
on the right has minimal
support, giving cocircuit 23
and oriented cocircuit (2, 3). (+, +, −, +, +) (0, +, −, 0, 0)
Clearly, the information given by an affine evaluation on A is more significant as

more points of A lie in 𝐻0 (see Figure 7). That is, we will be interested in the affine
evaluations of minimal support in A, which correspond to affine functionals that are
0 in a hyperplane spanned by points of A (at least 𝑑 affinely independent points of

A are in such hyperplane). The points of A not lying in the hyperplane will form a
cocircuit, and the signature of the evaluation will be an oriented cocircuit.
Definition 14 ( [1, Defs. 4.1.6 and 4.1.18] ) For A a point configuration:

• The circuits of A are its minimal affinely dependent subsets. The oriented circuits
of A are the signatures of those dependences in A.
• The cocircuits of A are the minimal supports of affine evaluations on A. That
is, the minimal (under inclusion) subsets of A of the form A \ 𝐻0 , where 𝐻0
is an affine hyperplane. The oriented cocircuits are the signatures of the affine
evaluation vectors of A under those functionals.
Notice that both the circuits and the cocircuits are merely subsets of A (those
with non-zero coefficient in a particular dependence or evaluation, respectively), and
that oriented circuits and cocircuits, as signatures, are completely determined by the
pairs (A+𝜆 , A−𝜆 ) or (A+𝑓 , A−𝑓 ), respectively. Also note that each circuit (resp. cocircuit)
produces two opposite oriented circuits (resp. cocircuits).
Definition 15 ( [1, Def. 4.1.43] ) The oriented matroid of a point configuration A is

the set of its oriented circuits, or its set of oriented cocircuits (both carry the same
information about A).
Notice that, for A of dimension 𝑑, any 𝑑 +2 points of A that span aff(A) determine
a 𝑑-dimensional circuit, and that any 𝑑 affinely independent points determine a
cocircuit.
Running example: oriented matroid

The following is the full list of (oriented) circuits of configuration A0 of Figure 3.
(14, 5), (123, 5), (14, 23), (45, 23),

(5, 14), (5, 123), (23, 14), (23, 45).
The list of (oriented) cocircuits of A0 is:
(∅, 125), (∅, 135), (∅, 245), (∅, 345), (1, 24), (1, 34), (2, 3),
(125, ∅), (135, ∅), (245, ∅), (345, ∅), (24, 1), (34, 1), (3, 2).
Remark
Observe that from the oriented matroid of A we can recover several things:
• The facets of A (as a point configuration) are the complements of the positive
cocircuits. That is, F is a facet of A if and only if (A \ F, ∅) is a cocircuit.
• The faces of A are all possible intersections of facets.

• If B is a subset of A, the circuits of B are the circuits of A contained in B and the
cocircuits of B are a subset of the restrictions to B of the cocircuits of A.
Note: every evaluation on A, restricted to B, is an evaluation on B, but some of
the evaluations that have minimal support in A may not have minimal support in
B; those have to be discarded from the list of cocircuits.
• In particular, we can compute faces of all subsets of A.
• Affinely independent subsets of A are those that do not contain any circuit. In
particular, from the oriented matroid we can compute the dimension of A and
also of every subset of A.
With some extra effort one can prove the following:
Theorem 16 ( [1, Thms. 4.1.14, 4.1.29 and 4.1.31] ) Let S be a collection of

subsets of a point configuration A. Then, knowing the oriented matroid of A (and
forgetting the coordinates of the points) is enough information to check whether S
is a subdivision of A, and whether it is a triangulation.
Proof (sketch) Let S = {S1 , . . . , S𝑚 } where the S𝑖 are subsets of A. Let us check
the properties of subdivisions as in Definition 8:
(DP) S𝑖 is 𝑑-dimensional if, and only if, it contains 𝑑 + 1 points that are affinely
independent. A subset of A is affinely independent if it does not contain a circuit.
(UP) It can be checked that, in the presence of (IP), property (UP) is equivalent to
the following: for every cell S𝑖 in S and every facet F of S𝑖 such that conv(F) is
not contained in the boundary of conv(A), there is another cell S 𝑗 containing the
same facet F and lying in the opposite side of the hyperplane containing conv(F).
Facets of every S𝑖 ∈ S and the existence of such an S 𝑗 for each of them can be
checked knowing the oriented matroid alone.
(IP) Claim: If 𝑖 ≠ 𝑗 then conv(S𝑖 ) ∩ conv(S 𝑗 ) = conv(S𝑖 ∩ S 𝑗 ) if, and only if,
there exists no (oriented) circuit (C+ , C− ) with C+ ⊆ S𝑖 and C− ⊆ S 𝑗 .
See Figure 8 for an example when the (IP) property does not hold: the circuit
(C+ , C− ), with C+ := i2 i3 ⊂ 𝑆𝑖 and C− := j1 j2 ⊂ 𝑆 𝑗 is such that the point 𝑝 :=
conv(i2 i3 ) ∩ conv(j1 j2 ) lies in conv(S𝑖 ) ∩ conv(S 𝑗 ) but not in conv(S𝑖 ∩ S 𝑗 ) = ∅:
j2
i2
Si = i1 i2 i3 p Sj = j1 j2 j3
j1 j3
Fig. 8 Setting of proof of
Theorem 16. i1 i3
2.2 Oriented matroids of vector configurations and matrix

representation
Oriented matroids can be defined (and are usually defined) for vector configurations
rather than point configurations. If V = {𝑣 1 , . . . , 𝑣 𝑛 } ⊂ R𝐷 is a vector configuration
(a finite subset of vectors of R𝐷 ) then we define its dependences, evaluations, circuits
and cocircuits as above, simply changing the word affine to linear everywhere. Of
course, one can say that a point configuration and a vector configuration are the same
thing, a finite subset of R𝐷 . What makes a difference is whether we are doing affine
geometry or linear algebra in R𝐷 . In particular, 𝑘 linearly independent vectors span
a space of dimension 𝑘, whereas 𝑘 + 1 affinely independent points are needed for an
affine space of dimension 𝑘.
We typically represent a point configuration A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 as the
(𝑑 + 1) × 𝑛 matrix
𝑝 . . . 𝑝𝑛
𝑀A = 1 .
1 ... 1
and a vector configuration V = {𝑣 1 , . . . , 𝑣 𝑛 } ⊂ R𝐷 as the 𝐷 × 𝑛 matrix

𝑀V = 𝑣 1 . . . 𝑣 𝑛 .
With this matrix representation, we have the following, which explains and re-
proves Lemma 10:
Lemma 17 ( [1, Lem. 4.1.31] ) Affine dependences of A are the kernel of 𝑀A and
affine evaluations of A are the rowspace of 𝑀A (the linear subspace generated by
rows, which is the same as the image of 𝑀A𝑡 ).
Linear dependences of V are the kernel of 𝑀V and linear evaluations are the
rowspace of 𝑀V .
Remark
Let A ⊂ R𝑑+1 be a 𝑑-dimensional point configuration contained in an affine

hyperplane not passing through the origin. Observe that the oriented matroid of A
as a point configuration is the same as its oriented matroid as a (𝑑 + 1)-dimensional
vector configuration.
In particular, the (affine) oriented matroid of the point configuration A =
{𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 is
the same as the (linear) oriented matroid of the vector config-
uration 𝑝11 , . . . , 𝑝1𝑛 ⊂ R𝑑+1 (the embedding of A in the hyperplane {𝑥 𝑑+1 = 1}
of R𝑑+1 ), which corresponds to the matrix representation 𝑀A .
Notice that if we have a 𝑑-dimensional point configuration A ⊂ R𝑑+1 in ho-
mogeneous coordinates (embedded in an affine hyperplane not passing through the
origin), we can automatically treat it as a vector configuration and hence use the
matrix representation 𝑀V . This is so because in this case the row (1, . . . , 1) that we
normally add in the matrix representation is a linear combination of the other rows.
2.3 Duality and the Gale transform
Let V ⊂ R 𝑘 and W ⊂ Rℓ be two vector configurations with 𝑛 elements, and represent

them as a 𝑘 × 𝑛 matrix 𝑀V and an ℓ × 𝑛 matrix 𝑀W , respectively.
Definition 18 ( [1, Def. 4.1.35] ) We say that V and W are oriented matroid duals if
the oriented circuits of V are the oriented cocircuits of W, and viceversa.
We say that W is the Gale transform of V if the row spaces of 𝑀V and 𝑀W are
orthogonal complements to one another.
Following the linear part of Lemma 10 we get the following:

Lemma 19 If W is a Gale transform of V, then V and W are oriented matroid duals.
For every V there is some Gale transform W: simply write V as a matrix 𝑀V ,
compute a basis of the orthogonal complement of the rowspace of 𝑀V , and use that
as the rows for a matrix 𝑀W , whose columns are the elements of W.
Running example: Gale transform
Let us compute, as an example, a Gale transform of the point configuration

A0 = {𝑝 1 , . . . , 𝑝 5 } of Figure 3. We take its embedding V0 := {𝑣 1 = ( 𝑝 1 , 1), . . . , 𝑣 5 =
( 𝑝 5 , 1)} in R3 , so that the matrix of this vector configuration is
0303 1
𝑀V0 = 0 0 3 3 1® .
© ª
«1 1 1 1 1¬
The following matrix 𝑁 has row space orthogonal to that of 𝑀V0

−1 −1 −1 0 3
𝑁= .
1 −1 −1 1 0
Hence its columns are the elements 𝑤 𝑖 of a vector configuration W which is the Gale
transform of V0 . This vector configuration is drawn in Figure 9.
Observe that vectors 𝑤 2 and 𝑤 3 coincide. This is in agreement with the fact
that (2, 3) was an oriented cocircuit in V0 (corresponding to the linear evaluation
(0, 3, −3, 0, 0) of the functional 𝑥 − 𝑦), so (2, 3) must now be an oriented circuit of
W and (0, 3, −3, 0, 0) its linear dependence in W. These two “equal vectors” must be
considered different elements in the oriented matroid and in the vector configuration,
distinguished by their labels.
w1 w4
w5
Fig. 9 A Gale transform of

V0 (or A0 ). w2 = w3
Problems
2.1 *Show that there are only three possible oriented matroids for five points in
general position (=no 3 collinear) in the plane.
2.2 *Write down the whole list of circuits and cocircuits of the first two point config-
urations of Problem 1.1.
6
Clue: since the configurations are in general position there should be exactly 4 pairs of opposite
circuits and 62 pairs of opposite cocircuits in each of them.

2.3 Write down the circuits and cocircuits of the following vector configuration of
rank 3:
{(0, 2, 1), (0, −1, −2), (1, 0, 2), (−2, 0, −1), (2, 1, 0), (−1, −2, 0)}.
Conclude that it is dual to the vertices of the hexagon.
2.4 Show that the above list of vectors is actually a Gale transform of the set of vertices
of the hexagon (the second configuration in Problem 1.1) with coordinates given
as follows:
𝑝 1 = (0, 2, 1), 𝑝 2 = (0, 1, 2), 𝑝 3 = (1, 0, 2),
𝑝 4 = (2, 0, 1), 𝑝 5 = (2, 1, 0), 𝑝 6 = (1, 2, 0)
Remark: the extra coordinate that you have (the configuration is two dimensional but it is
embedded in an affine plane in R3 ) implies that you can already think of the vertices of the
hexagon as vectors, without the need of an additional coordinate constantly equal to 1 (see
Remark 2.2 in the notes).
2.5 *Construct a Gale transform of the m.o.a.e. (the first configuration in Problem
1.1), with coordinates given as follows:
𝑝 1 = (4, 0, 0), 𝑝 2 = (0, 0, 4), 𝑝 3 = (0, 4, 0),

𝑝 4 = (2, 1, 1), 𝑝 5 = (1, 1, 2), 𝑝 6 = (1, 2, 1)
Same remark as above about the third coordinate.
2.6 *Consider versions of the m.o.a.e. where the inner triangle is rotated with respect
to the outer one. (That is, you have the six vertices of two equilateral triangles
with the same barycenter and one inside the other, but the two triangles are no
longer parallel to one another). Convince yourself of the following behavior: if the
rotation is small, neither the oriented matroid nor the set of triangulations of your
point set changes. As you keep rotating, there is a precise point when both the
oriented matroid and the set of triangulations change. Try to describe the change.
(How many and which circuits, cocircuits, and triangulations are affected?).
2.7 Oriented matroid of a directed graph. Let 𝐺 be a directed graph: a pair (𝑉, 𝐸)
where 𝑉 = {1, . . . , 𝑘 } is a set of 𝑘 vertices and 𝐸 a list of 𝑛 directed edges (each
→−
directed edge is an ordered pair 𝑖 𝑗 := (𝑖, 𝑗) with 𝑖, 𝑗 ∈ 𝑉).
Consider the directed incidence matrix 𝐴 of 𝐺: the 𝑘 × 𝑛 matrix whose 𝑙-th
column is the vector 𝑒 𝑖 − 𝑒 𝑗 , where 𝑒 𝑠 ∈ {0, 1} 𝑘 is the vector with all zeroes and
→
−
coordinate 𝑠 equal to 1, and the 𝑙-th edge of the graph is 𝑖 𝑗 .
Let X := {𝑎 1 , . . . , 𝑎 𝑛 } ⊂ R 𝑘 be the set of columns of 𝐴 (so that the elements of
X are in bĳection with edges of 𝐺, and the number of coordinates is the number
of vertices of 𝐺). Show that:
a. A subset of X is linearly dependent if, and only if, the corresponding set of
edges of 𝐺 contains a (not necessarily well-directed) cycle.
Clue: for the if part, show how to get a dependence from the cycle; for the only if part, argue
that if the set of edges contains no cycle then some vertex is used in a single edge, and use
that to induct on the number of edges: removing that edge from the list reduces the rank of
your set of vectors by one.
b. Circuits of X are in bĳection to (not necessarily well-directed) cycles of 𝐺.
c. For each circuit, the two sides of the corresponding oriented circuit are the
edges pointing to one or the other direction along the cycle.
2.8 a. *Write down all the oriented circuits and cocircuits of the point configuration
A = {(6, 0), (3, 0), (2, 2), (0, 0), (0, 3), (0, 6)}.
Clue: there are seven circuits and seven cocircuits.

b. *Using the description in parts (b) and (c) of the Problem 2.7 (even if you
did not solve the problem), write down the oriented circuits of the oriented
matroid of the graph
→
− → − → − → − → − → −
𝐺 = ({1, 2, 3, 4}, {12, 13, 14, 23, 24, 34})
(The complete graph on 4 vertices, directed from smaller indices to bigger

indices).
c. *In both cases you get the same oriented circuits, so these oriented matroids
are the same. What do the cocircuits of A correspond to in 𝐺?
3 Flips. Graph of triangulations
Flips are an adjacency notion among different triangulations of a point configuration,

and each flip corresponds to a circuit. For more details on the contents of this section
see [1, Sections 2.4 and 4.4].
Lemma 20 ( [1, Lem. 2.4.2] ) Let C = (C+ , C− ) be an oriented circuit. Then C has
exactly two triangulations:
TC+ = {C \ {𝑐 𝑖 } | 𝑐 𝑖 ∈ C+ } and TC− = {C \ {𝑐 𝑖 } | 𝑐 𝑖 ∈ C− }.
which we will refer to as the positive and negative triangulations of the circuit.
Proof Let 𝑑 be the dimension of the circuit. In particular C = {𝑐 1 , 𝑐 2 , . . . , 𝑐 𝑑+2 } and

the 𝑑-simplices of C are the sets C \ {𝑐 𝑖 }, for each 𝑖 = 1, . . . , 𝑑 + 2.
Let S𝑖 = C \ {𝑐 𝑖 } and S 𝑗 = C \ {𝑐 𝑗 } be two such simplices. Then, S𝑖 and S 𝑗 will
belong to the same triangulation of C if, in particular, they verify the intersection
property (IP) (see (1)).
Remember, from the proof of Theorem 16, that S𝑖 and S 𝑗 verify conv(S𝑖 ∩ S 𝑗 ) =
conv(S𝑖 ) ∩ conv(S 𝑗 ) if, and only if, it does not happen that C+ ⊆ S𝑖 and C− ⊆ S 𝑗
(or viceversa).
This implies that, if S𝑖 and S 𝑗 belong to the same triangulation, both 𝑐 𝑖 and 𝑐 𝑗
must belong to either C+ or C− .
That is, for some triangulation T of C, we have that either
T ∈ C \ {𝑐 𝑖 } | 𝑐 𝑖 ∈ C+ , ∀T ∈ T

or
T ∈ C \ {𝑐 𝑖 } | 𝑐 𝑖 ∈ C− ,

∀T ∈ T
Since both those sets are triangulations of C, that completes the proof.
Example: triangulations of circuits

Let us see the triangulations of the two combinatorial types of circuits of dimension
2. That is, the oriented circuits C = (14, 23) (Figure 10, left), and D = (123, 4)
(Figure 10, right).
TC+ = {123, 234} TC− = {124, 134} TD+ = {124, 134, 234} TD− = {123}
3 4 3 4 3 3
1 2 1 2 1 2 1 2
Fig. 10 Triangulations of 2-dimensional circuits.
Let A be a point configuration, T a triangulation of A, and F ⊆ A a face of T (in

particular, F is a simplex).
Definition 21 ( [1, Def. 2.1.6] ) The link of F in T is

L T (F) = G ⊂ A | F ∪ G ∈ T , F ∩ G = ∅ .
Notice that, since F ∪ G is a simplex of the triangulation, then G is also a face

of F ∪ G, a face of T , and a simplex (of lower dimension) in itself. For example, in
Figure 10, the link of 23 in TC + is {1, 4}, and the link of 4 in TD + is {12, 13, 23}.
Observe as well that if F is full-dimensional (i.e., a cell of T ), the link is {∅}.
Definition 22 ( [1, Defs. 2.4.7 and 4.4.2, and Thm. 4.4.1] ) Let T be a triangulation
of A. Suppose there is an oriented circuit C = (C+ , C− ) in A such that T contains
the positive triangulation TC+ of C, and suppose further that the links in T of all the
cells of TC+ are identical. Let L be that link.
Then it is possible to construct a new triangulation T 0 of A as follows:
T 0 = T \ T ∪ L | T ∈ TC+ , L ∈ L ∪ T ∪ L | T ∈ TC− , L ∈ L .

That is, we remove TC+ (together with its link) from T and insert TC− (with the
same link). We say that T 0 is obtained from T by a (geometric bistellar) flip on the
circuit (C+ , C− ), and that T and T 0 are adjacent or flip-neighbors.
Observe that if T 0 is obtained from T by a flip on the circuit (C+ , C− ) then T is

obtained from T 0 by a flip on the opposite circuit (C− , C+ ). As an example, the two
triangulations of each circuit of Figure 10 are adjacent.
Remark
When the circuit C is full-dimensional (consists of exactly 𝑑 + 2 points), then the
cells of TC+ are full-dimensional and the condition about the links is vacuous. Hence
T 0 = T \ T ∈ TC+ ∪ T ∈ TC− .

However, when the circuit is not full-dimensional (the circuit involves less points)
the link condition has to be verified. Notice that in this case, when we say that “T
contains’ TC+ ”, we mean that every cell (or face) of TC+ is a face (but not necessarily
a cell) of T .
Definition 23 ( [1, Def. 2.4.8] ) The set of all triangulations of A, under adjacency
by flips, forms the graph of triangulations, or flip-graph, of A.
Running example: flips of circuits and flip-graph

In Figure 11, we have all the triangulations of our running example A0 of Figure 3,
which are:
T1 = {124, 134}, T3 = {125, 135, 234, 235},
T2 = {123, 234}, T4 = {125, 135, 245, 345}
T1 T2 T3 T4
3 4 3 4 3 4 3 4
5 5
1 2 1 2 1 2 1 2
Fig. 11 Triangulations of A0 .
Let us look at the circuits of A0 :
• The circuit (14, 23) is full-dimensional, and its two triangulations are contained,
respectively, in triangulations T1 and T2 of A0 .
• The circuit (14, 5) is lower dimensional, and it is a bit more complicated to
study. The two triangulations of 145 are {14} and {15, 45}. Notice that these are
contained in T1 and T4 , respectively, and that the link of 14, 15 and 45 in both
triangulations is the same, namely {2, 3}. Hence the conditions are met and these
two triangulations are adjacent via a flip on the circuit.
Figure 12 shows the flip-graph of A0 .
3 4 3 4
(14, 23)
1 2 1 2
(14, 5) (123, 5)
3 4 3 4
Fig. 12 Graph of triangula-

tions of A0 related by flips. 5 5
Observe that it is the graph of (23, 45)
a square. 1 2 1 2
Example: Flips in dimension 3

We now describe an example of flip in three dimensions. Our point configuration
is going to be {0, 1}3 , the vertices of a regular cube. Triangulation T1 corresponds
to a large simplex 1368 and four simplices that are pyramids over its four faces.
This triangulation contains the positive triangulation of the circuit (28, 136), namely
{1236, 1368}, so we can perform a flip to obtain a new triangulation T2 : remove those
two simplices and add the three simplices 1238, 1268 and 2368 (see Figure 13).

T1 = 1236, 1348, 1368, 1568, 3678

8 7
5 6
8 7
5 6
4 3
4 1 2
3
1 2

T2 = 1238, 1268, 1348, 1568, 2368, 3678}
Fig. 13 Flip of 3-dimensional circuit (28, 136) on a triangulation of the 3-cube.
Notice that both T1 and T2 contain the (2-dimensional) triangulation {123, 134}
of the circuit (24, 13). However, on T1 these two triangles have different links: the
link of 123 is {6} whereas that of 134 is {8}. Thus, in T1 we cannot perform a flip
on that circuit. On the other hand, in T2 the link of both triangles is {8} so we can
perform a flip to obtain a third triangulation T3 : remove the two tetrahedra 1238 and
1348 and add 1248, 2348 (see Figure 14).

T2 = 1238, 1268, 1348, 1568, 2368, 3678

8 7
5 6
8 7
5 6
4 3
4 1 2
3
1 2

T3 = 1248, 1268, 1568, 2348, 2368, 3678

Fig. 14 Flip of 2-dimensional circuit (24, 13) on a triangulation of the 3-cube.

Flip-graph of moae
The moae (the “mother of all examples”) is the point configuration of Figure 15.
Figure 16 shows its graph of triangulations.
4 6
Fig. 15 The mother of all
examples. 1 3
Fig. 16 The graph of flips of the moae. The two non-regular triangulations are in blue and red, and
the flips joining them to the regular ones are dashed.
It is a fact that the flips among triangulations are in bĳection with subdivisions
whose proper refinements are triangulations. See Section 2.4.2 of [1] for more on
this.
Definition 24 Let A be a point configuration, and let C = (C+ , C− ) be an oriented
circuit of A. A flip on (C+ , C− ) is said to be of type (𝑖, 𝑗), where |C+ | = 𝑖 and
|C− | = 𝑗.
In Figure 12, we have one lower dimensional circuit of type (1, 2), a full-
dimensional circuit of type (1, 3), and two full-dimensional circuits of type (2, 2).
How to identify flips in a triangulation

Suppose we are given a triangulation T of a point configuration A. Let us see how
to find all the flips that can be performed in T . We first look at the full-dimensional
circuits / flips:
• Flip of type (1, 𝑑 + 1): In this case the flip is on a circuit ({𝑝}, F), where 𝑝 is
a point and F is a 𝑑-simplex. The positive triangulation of this flip does not use
𝑝, but has F as a cell, and the negative one contains the following 𝑑 + 1 cells:
F𝑞 := F ∪ {𝑝} \ {𝑞}, for each 𝑞 ∈ F. That is, the flip can be performed if and only
if T does not use 𝑝 and 𝑝 lies in the relative interior of a cell of T . See [1, Exm.
4.4.7].
• Flip of type (𝑖, 𝑗), for 𝑖 ≥ 2: Our 𝑑-dimensional circuit C = (C+ , C− ), is such that
TC+ has 𝑖 ≥ 2 simplices. In particular, C is contained in two adjacent cells of T
(by adjacent, we mean that the cells intersect in a (𝑑 − 1)-dimensional face of the
triangulation). See [1, Lemma 4.4.9]. To detect these, we look at all adjacent pairs
of cells in T and check whether the unique circuit contained in them is flippable.
For the lower dimensional flips/circuits, we need to do the same, but taking
into account that our circuit now lives in a lower dimensional subset of the point
configuration and that whether we can flip or not will also depend on the link of the
(lower-dimensional) cells of the triangulation of the circuit.
Example: finding flips in triangulations

In Figure 17 we can see four triangulations of a 2-dimensional point configuration
with 7 points. In T4 we find the circuit (145, 7) which, under a type (3, 1) flip,
gives us triangulation T3 . Both T1 and T2 contain the positive triangulation of the
1-dimensional circuit (14, 6), but only that of T1 is flippable: the links of 16 and 46
are equal in T1 but different in T2 .
Now, if we look at flips of type (𝑖, 𝑗) with 𝑖, 𝑗 ≥ 2, notice that in dimension 2,
we can only find type (2, 2) circuits, which in particular are full-dimensional. These
can be found by looking at pairs of adjacent cells, as explained above. The ones that
are flippable in the triangulations of Figure 17 are shadowed in Figure 18:
2 3
T1
6 4
1 T3 T4
type (2, 1)
2 3 2 3
5
type (3, 1)
4 4
2 3 1 1
7
T2
6 4 5 5
1
Fig. 17 Flippable circuits of type (𝑘, 1).
2 3 2 3 3
6 6 6 4
1 1
3 2 3 3
4 4 4
1 1 1
7
Fig. 18 Flippable circuits of
type (2, 2). 5
Problems
3.1 *Show that all the triangulations of a convex 𝑛-gon have exactly 𝑛 − 𝑑 − 1 = 𝑛 − 3
flips, where 𝑛 is the number of vertices and 𝑑 = 2 is the dimension.
3.2 *Show that for the triangular prism Δ2 × Δ1 all triangulations have exactly 𝑛 −
𝑑 − 1 = 2 flips, where 𝑛 = 6 is the number of vertices and 𝑑 = 3 is the dimension.
3.3 If you did Problem 1.2b, show that flips among triangulations of Δ 𝑘 × Δ1 corre-
spond to permutations that differ in a transposition of consecutive elements. In
particular, all triangulations have exactly 𝑛 − 𝑑 − 1 = 𝑘 flips, where 𝑛 = 2𝑘 + 2 is
the number of vertices and 𝑑 = 𝑘 + 1 is the dimension.
3.4 *Show that m.o.a.e. has triangulations with three and with four flips. (Here 𝑛−𝑑−1
is three).
3.5 *Show that the 3-cube [0, 1] 3 has triangulations with four, and with six flips.
(Here 𝑛 − 𝑑 − 1 is four).
3.6 *Draw the graphs of flips among triangulations of the configuration you chose in
Problem 1.1.
3.7 *Draw the graph of flips among triangulations of the following six points: the five
vertices of a regular pentagon, together with its center.
3.8 Let T1 and T2 be two triangulations of the convex 𝑛-gon. Show that:
a. *If one of them is the triangulation that joins one vertex to all the others, then
we can go from T1 to T2 in 𝑛 − 3 or less flips.
b. *No matter what they are, we can go from T1 to T2 in 2𝑛 − 6 or less flips.
c. In fact, we can go from T1 to T2 in 2𝑛 − 10 or less flips. (Clue: [1, Prop. 1.1.5]).
d. Let 𝑛 be even and suppose that T1 uses all the even ears4 and T2 uses all the
odd ears5. Show that it is impossible to go from T1 to T2 in less than 32 𝑛 − 5
flips.
Clue: classify the possible diagonals of an 𝑛-gon in three types: even-even, even-odd, and
odd-odd. This gives you, for each triangulation, a vector of length three that counts the
number of diagonals of each type. Study how that vector changes by a flip (there are several
cases), and what change you need in order to go from T1 to T2 .
3.9 Let A = {±𝑒 𝑖 ± 𝑒 𝑗 : 𝑖, 𝑗 ∈ {1, 2, 3}} ∪ {(0, 0, 0)} (the vertices and barycenter
of a regular cube-octahedron; see [1, Exm. 3.6.16 and Prop. 3.6.17]). Let T be
a triangulation in which all tetrahedra use (0, 0, 0) as a vertex (put differently, T
is one of the 64 triangulations obtained triangulating the 6 quadrilateral facets of
conv(A) and coning the boundary to (0, 0, 0)). Show that:
a. The number of flips in T equals 6 plus twice the number of vertices of degree
four that you see in the triangulation of the boundary.
b. There are two triangulations with only 6 flips (observe that 𝑛 − 𝑑 − 1 = 9 in
this example).
c. Are there any triangulations with 8 flips?
(−1,0,1)
(0,−1,1)
(0,1,1)
(1,0,1) (−1,−1,0)
(0,0,0) (−1,1,0)
(1,−1,0)
(1,1,0) (−1,0,−1)
(0,−1,−1)
(0,1,−1)
(1,0,−1)
4 that is, suppose that for every even 𝑖, the triangle 𝑖 − 1, 𝑖, 𝑖 + 1 is in T1

5 the same, for all odd 𝑖
4 Regular subdivisions and triangulations
Let A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 be a point configuration and let 𝛼 = (𝛼1 , . . . , 𝛼𝑛 ) ∈ R𝑛

be a vector.
Definition 25 ([1, Def. 2.2.10]) The regular subdivision of A obtained by the lifting
vector 𝛼 is defined as follows:
(i) Let 𝑝˜𝑖 = ( 𝑝 𝑖 , 𝛼𝑖 ) ∈ R𝑑+1 for each 𝑖 and compute the facets of Ã := { 𝑝˜1 , . . . , 𝑝˜ 𝑛 }.
(ii) Project the lower facets of Ã onto R𝑑 .
Here, a lower facet of Ã is a facet that is visible from below. That is, a facet whose
outer normal vector has its last coordinate negative.
Observe that the “projection” step is combinatorially trivial. For each lower facet
{ 𝑝˜𝑖1 , . . . , 𝑝˜𝑖𝑘 } of Ã we simply make {𝑝 𝑖1 , . . . , 𝑝 𝑖𝑘 } a cell of the subdivision.
Example: regular subdivisions
Figure 19 shows two liftings of the point configuration 12345 which have same
lifted convex hull but yield two different regular subdivisions. Observe that the
subdivision on the right is a refinement of the subdivision on the left.
1̃ 5̃ 1̃ 3̃ 5̃
2̃ 3̃ 4̃ 2̃ 4̃
1 2 3 4 5 1 2 3 4 5
{12, 234, 45} {12, 24, 45}

Fig. 19 Regular subdivisions in dimension 1.
In Figure 20 we can see a regular subdivision of A0 given by the lifting vector

(4, 3, 3, 14, 2).
Example of non-regular subdivision: the moae
The trianguation of moae depicted in Figure 21 is not regular. To see this we

argue by contradiction. If 𝛼 is a lifting vector inducing this subdivision, by Problem
4.2 below we can assume without loss of generality that 𝛼4 = 𝛼5 = 𝛼6 = 0, so that
𝛼 = (𝛼1 , 𝛼2 , 𝛼3 , 0, 0, 0). In order for the edge 16 to be a face of the subdivision
3 4
Fig. 20 A regular triangula-
tion of A0 , with lifting vector 5
(4, 3, 3, 14, 2). 1 2
we need 𝛼1 < 𝛼3 . The same argument for the edges 35 and 24 leads us to the
contradiction 𝛼1 < 𝛼3 < 𝛼2 < 𝛼1 . Hence, no lifting vector yields the subdivision
of Figure 21. Section 7.1.2 of [1] contains twelve proofs of non-regularity of this
triangulation.
4
6
Fig. 21 A non-regular subdi-
vision. 1 3
For example, if 𝛼 = 0 then we get the trivial subdivision {A}. If 𝛼 has a single
non-zero coordinate 𝛼𝑖 , corresponding to a certain point 𝑝 𝑖 ∈ A, then we get the
following two subdivisions, depending on the sign of 𝛼𝑖 .
Pulling a point 𝑝 𝑖 [1, Sec. 4.3.2]

Taking 𝛼𝑖 < 0, the subdivision obtained is

F ∪ {𝑝 𝑖 } : F is a facet of A not containing 𝑝 𝑖 .
In particular, if 𝑝 𝑖 is in the interior of conv(A) then the subdivision is the cone of

the boundary of A with apex at 𝑝 𝑖 .
Pushing a point 𝑝 𝑖 [1, Sec. 4.3.1]

Taking 𝛼𝑖 > 0, the subdivision obtained is

A \ {𝑝 𝑖 } ∪ F ∪ 𝑝 𝑖 : 𝐹 is a facet of A \ {𝑝 𝑖 } visible from 𝑝 𝑖 .
Here, we say that a facet F of A \ {𝑝 𝑖 } is visible from 𝑝 𝑖 if 𝑝 𝑖 lies in the opposite

side of the hyperplane containing F than the rest of points of A \ F.
In particular, if 𝑝 𝑖 is not a vertex of A then the subdivision is {A \ {𝑝 𝑖 }}, the
trivial subdivision of A \ {𝑝 𝑖 }.
Figure 22 shows an example of pulling or pushing of a single point in a point

configuration, starting at the trivial subdivision.
6 6
pulling 1 pushing 1
5 5
7
4 1 4 1
8
trivial subdivision
3 6 3
2 5 2
{123, 1345, 156} 7 {128, 167, 178, 2345678}
4 1
8
6 3 6
pulling 4 2 pushing 4
5 5
{12345678} 7
4 1 1
8
3 3
2 2
{124, 146, 234, 456} {1235678}
Fig. 22 Pulling or pushing a point from the trivial subdivision.
Running example: pulling or pushing a point

The following table collects the subdivisions of A0 obtained by pulling or pushing
a single point of the configuration:
𝛼 𝑠𝑢𝑏𝑑𝑖𝑣𝑖𝑠𝑖𝑜𝑛
(−1, 0, 0, 0, 0) {124, 134}
(0, −1, 0, 0, 0) {123, 234}
(0, 0, −1, 0, 0) {123, 234}
(0, 0, 0, −1, 0) {124, 134}
(0, 0, 0, 0, −1) {125, 135, 245, 345}
(1, 0, 0, 0, 0) {125, 135, 2345}
(0, 1, 0, 0, 0) {1245, 1345}
(0, 0, 1, 0, 0) {1245, 1345}
(0, 0, 0, 1, 0) {1235, 234}
(0, 0, 0, 0, 1) {1234}
Lexicographic subdivision [1, Def. 4.3.8]

Let B be a subset of the points of A, given in a specific order (this order may not be
their original order in A, but we denote B = {𝑝 1 , . . . , 𝑝 𝑘 } to simplify notation). For
each of these points choose a sign 𝜖𝑖 ∈ {+, −}.
Starting with the trivial subdivision, consider the points 𝑝 1 , . . . , 𝑝 𝑘 in order and
at each step:
• If 𝜖 𝑖 = +, push the point 𝑝 𝑖 in all the cells of the previously constructed subdivision
that contain 𝑝 𝑖 .
• If 𝜖𝑖 = −, pull the point 𝑝 𝑖 in all the cells of the previously constructed subdivision
that contain 𝑝 𝑖 .
In every step, the cells that do not contain 𝑝 𝑖 remain unchanged (recall that “𝑝 𝑖
is contained in a cell S” means 𝑝 𝑖 ∈ S). In particular, if a point 𝑝 𝑖 has “disappeared”
from the subdivision at some step, then it will never reappear, even if it is one of the
points to be pushed or pulled later (pushing or pulling a point that has disappeared
does not alter the subdivision).
Definition 26 A subdivision of A is called lexicographic if it can be obtained from

the trivial subdivision by a sequence of pullings and pushings.
Example: lexicographic triangulations

Let {0, 1, 2, 3, 4} × {0, 1, 2} be a point configuration with fifteen points. They are
labeled in the specific order depicted in Figure 23.
Figure 24 shows the lexicographic subdivisions that are the result of: pulling all
the points in their labeling order, pushing all the points in their labeling order, and
the mixed pushing-pulling given by the ordered list of signs:
10 2 14 4 7
Fig. 23 Point configuration 3 11 6 8 9

with a specified order of the
1 15 13 5 12
points.
Pulling Pushing Mixed

10 2 7 10 14 7 10 2 7
11
3 9 3 9
1 12 1 15 13 12 1 15 12
1+ , 2− , 3+ , 4− , 5+ , 6+ , 7+ , 8− , 9− , 10+ , 11− , 12− , 13− , 14+ , 15−

Fig. 24 Three examples of lexicographic subdivisions.
Lemma 27 ([1, Lemma 4.3.12]) The lexicographic subdivision coincides with the
regular subdivision obtained choosing positive numbers 𝜆1 · · · 𝜆 𝑘 0 and
taking 𝛼𝑖 = 𝜖𝑖 𝜆𝑖 for each point in the list and 𝛼𝑖 = 0 for each point not in the list.
The actual numbers 𝜆𝑖 in the lemma are not important, as long as each one is
sufficiently larger than the next one.
Problems
4.1 *Let 𝛼, 𝛽 ∈ R𝑛 be two lifting vectors such that 𝛼 − 𝛽 is an affine evaluation on

the set A. Show that 𝛼 and 𝛽 produce the same regular subdivision of a point
configuration A.
4.2 *Deduce from the previous problem that to construct all the regular subdivisions
of a set A, or to check whether a particular subdivision is regular or not, there is
no loss of generality in choosing a priori 𝑑 + 1 affinely independent points of A
(a 𝑑-simplex) and prescribing those 𝑑 + 1 coordinates in the lifting vector to be
zero.
4.3 Show that the following triangulation is not regular.

Clue: argue on the relative slopes of the lifted edges.
4.3 *Show that the following two triangulations of m.o.a.e. are not regular.
Clue: by the Problem 4.1, you can assume height zero to the three interior points.
2 2
5 5
4 6
6 4
32 Francisco
1 Santos, Mónica
3 Blanco
1 and Federico Castillo
3
4.4 *Show that, except for the non-regular triangulation in Fig. 21 and the mirror
image of it, all other triangulations of the sets in Problem 1.1 are lexicographic
(hence regular).
4.5 Consider configuration {𝑝 1 , 𝑝 2 , 𝑝 3 , 𝑝 4 , 𝑝 5 , 𝑝 6 }, where
𝑝 1 = (5, 0, 0), 𝑝 2 = (0, 5, 0), 𝑝 3 = (0, 0, 5),

𝑝 4 = (1, 2, 2), 𝑝 5 = (2, 1, 2), 𝑝 6 = (2, 2, 1)
4.4
Prove that the following regular triangulation is not lexicographic:
4.5 *Show that, except for the two non-regular triangulations in the previous exercise,
all other triangulations of the sets in Problem 1.1 are lexicographic (hence regular).
{456, 345, 246, 156, 126, 135, 234}
4.6 Consider configuration {p1, p2, p3, p4, p5, p6 }, where
4.6 *Let B be a subset of A. Show that there is a subdivision of A (in fact, a regular
p1 = (5, 0, 0), p2 = (0, 5, 0), p3 = (0, 0, 5),
one) having B as a face.
p4 = (1, 2, 2), p5 = (2, 1, 2), p6 = (2, 2, 1)
Prove that
4.7 Let B1 and B2 be two subsets ofthe
A following regular1triagulation
with conv(B ) ∩ conv(B is 2not
) =lexicographic:
∅. Show that
there is a subdivision of A (in fact, a regular one) in which both B1 and B2 are
faces.
4.8 Show that the same is not necessarily true for *three* subsets: taking B the vertex
set of a triangular prism, find three subsets B1 , B2 , B3 of it with conv(B𝑖 ) ∩
conv(B 𝑗 ) = ∅ for all 𝑖, 𝑗 ∈ {1, 2, 3}, but such that no subdivision of A has B1 , B2
and B3 as cells.
Remark: the m.o.a.e. is an example where no regular subdivision exists using B1 , B2 and B3 .
5 The Secondary polytope
The importance of regular triangulations is that the graph of flips among regular
triangulations of a 𝑑-dimensional point configuration A with 𝑛 points is the graph
of a polytope of dimension 𝑛 − 𝑑 − 1, where the graph of a polytope is the collection
of its vertices and edges. For example, in Figure 16 of the graph of triangulations
of the moae, if we remove the two non-regular subdivisions (i.e. the red and blue
colored ones), the remaining graph is planar. That is, it is the graph of a polytope of
dimension 3.
More specifically, it turns out that the whole poset of regular subdivisions of A
under refinement has the following very nice structure.
Theorem 28 (Gelfand-Kapranov-Zelevinsky, see [1, Thm. 5.1.9]) Let A be a point

configuration with 𝑛 points and dimension 𝑑. Then, there is a polytope 𝚺(A) of
dimension 𝑛 − 𝑑 − 1 whose poset of (non-empty) faces is isomorphic to the refinement
poset of regular subdivisions of A. In particular:
• Vertices of 𝚺(A) ←→ regular triangulations of A.

• Edges of 𝚺(A) ←→ (geometric bistellar) flips among regular triangulatons of A.
• Facets of 𝚺(A) ←→ “coarse” regular subdivisions of A.
• The full-dimensional face 𝚺(A) corresponds to the trivial subdivision {A}.
Hence, every regular triangulation has at least 𝑛 − 𝑑 − 1 flips, and the graph of flips
among regular triangulations is connected (and (𝑛 − 𝑑 − 1)-connected).
The two conclusions (≥ 𝑛 − 𝑑 − 1 flips and connected graph) hold also for non-
regular triangulations in dimension two, but not in higher dimensions: triangulations
with less than 𝑛 − 𝑑 − 1 flips exist starting in dimension 3 (there is one in Problem
3.9), and triangulations with no flips at all exist in dimension 6. Point sets with
disconnected graphs of triangulations exist in dimension 5 and higher. Whether they
exist in dimensions three and four is open.
Definition 29 The polytope 𝚺(A) is called the secondary polytope of the point
configuration A.
We finish by discussing briefly the main ingredients for the proof of Theorem 28.
Secondary fan [1, Sec 5.2]
Let A ⊂ R𝑑 be an 𝑛-point configuration. For any regular subdivision S we define

the secondary cone of S to be K(S) := {𝛼 ∈ R𝑑 : the lift vector 𝛼 induces S on A}.
By definition, any vector 𝛼 ∈ R𝑑 induces a regular subdivision, so the collection
of all secondary cones cover (without overlapping) the whole ambient space R𝑑 .
Furthermore, they form a fan, called the secondary fan of A.
Example: Running example
We keep studying our running example. The set of all lift vectors 𝛼 equals R5 but
we can pick any three non-collinear points in A and assume they have height zero
(see Problem 5.1). Then depending on 𝑥 and 𝑦 we get a different subdivision. See
Figure 25.
x y
Fig. 25 We choose points 0

1, 2, 5 to be at height zero
so we only get two free
parameters 𝑥, 𝑦. 0 0
In Table 1 we write down all possible subdivisions and the corresponding sec-
ondary cone (depending only on 𝑥, 𝑦). The first four are triangulations, the next four
are coarsenings, and the last one is the trivial subdivision.
Table 1 Triangulations and their secondary cones

Subdivision S The cone K( S)
{125, 135, 235, 234} 𝑥 > 0, 𝑦 > 2𝑥
{125, 135, 245, 345} 𝑦 > 0, 𝑥 > 𝑦/2
{134, 124} −𝑦 > 0, 𝑥 − 𝑦 > 0
{123, 234} −𝑥 > 0, 𝑦 − 𝑥 > 0
{125, 135, 2345} 𝑦 = 2𝑥, 𝑥 > 0
{1235, 234} 𝑦 > 0, 𝑥 = 0
{1234} 𝑥 = 𝑦, −𝑥 > 0
{1245, 1345} 𝑥 > 0, 𝑦 = 0
{12345} 𝑥=𝑦=0
In Figure 26 we show all possible lifts for each subdivision. We omit the trivial
subdivision at the origin.
To prove Theorem 28 one constructs the secondary polytope 𝚺(A) as a polytope

with normal fan equal to the secondary fan of A.
The GKZ vector

Suppose T is a triangulation of A = {𝑝 1 , . . . , 𝑝 𝑛 }. Define the vector
y
y = 2x > 0
Fig. 26 For example, when

( 𝑥, 𝑦) = (3, 0) the in- x=y<0
duced regular subdivision
is {1245, 1345} (which is not
a triangulation).
∑︁
𝑧(T ) = (𝑧1 , . . . , 𝑧 𝑛 ) ∈ R𝑛 , 𝑧𝑖 = vol(F),
𝑝𝑖 ∈F
That is, 𝑧𝑖 is sum of the volumes of all 𝑑-simplices in T containing 𝑝 𝑖 (necessarily

as a vertex). 𝑧(T ) is called the GKZ-vector of T , to honor Gel’fand, Kapranov and
Zelevinsky. We can take
𝚺(A) = conv{𝑧(T ) : T a triangulation of A},

as the definition of secondary polytope (see [1, Def. 5.1.7].
Lemma 30 ([1, Thm. 5.2.16]) The secondary fan of A equals the normal fan of the
polytope 𝚺(A).
Some examples
In dimension 1, since all subdivisions are regular, it is not hard to see that for
the one dimensional point configuration A = {0, 1, 2, · · · , 𝑛 + 1} ⊂ R1 we have
that 𝚺( 𝐴) is combinatorially equivalent to an 𝑛-cube: there are 2𝑛 triangulations,
all regular, one for each possible subset of {1, . . . , 𝑛}. (The subset indicates which
of the intermediate points are used as vertices). Do explicitly the computation of
GKZ-vectors for 𝑛 = 2 and perhaps 𝑛 = 3.
Another especial example is given by the vertices of Δ 𝑘 ×Δ1 , with Δ 𝑘 a 𝑘-simplex.
In this case all triangulations are unimodular (all simplices have the same volume),
and you can compute that the secondary polytope is the permutohedron. (See [1, Sec
6.2.1] for details).
For a less trivial example, if we take A to be the vertices of an 𝑛-gon, the secondary
polytope is the so-called (𝑛 − 2)-associahedron.
The secondary polytope of the vertices of the standard cube [0, 1] 𝑑 has only been
computed for 𝑑 ≤ 4.
Remark
For any triangulation T , its GKZ vector will be a point in 𝚺(A) but not a
vertex unless T is regular. Moreover, the GKZ vector is unique for each regular
triangulation but may not be unique for non-regular ones (for example, the two non-
regular subdivisions of the moae have the same GKZ-vector; this cannot happen for
regular ones).
Problems
5.1 Check that the secondary fan of Figure 26 is correct.
5.2 Let S be a subdivision of a point configuration A with 𝑛 points. Let K(S) ⊂ R𝑛

be the set of all lifting vectors 𝛼 ∈ R𝑛 that produce S as a regular subdivision.
Show that:
a. K(S) is full-dimensional if, and only if, S is a regular triangulation.
b. K(S) is a convex, relatively open, polyhedral cone. That is, it is the set of
solutions to a set of linear equalities and strict linear inequalities.
c. 𝛽 is in the closure of K(S) if and only if the regular subdivision produced by
𝛽 is refined by S.
5.3 Draw the graph of flips among *regular* triangulations of the first two configura-
tions in Problem 1.1 and understand how they are the graphs of two 3-dimensional
polytopes (the corresponding secondary polytopes).
5.4 Let A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 be a configuration of 𝑛 points in dimension 𝑑. Prove

that even if the secondary polytope 𝚺(A) of A lives in R𝑛 , it has dimension
𝑛 − 𝑑 − 1.
a. To show that dim 𝚺(A) ≥ 𝑛 − 𝑑 − 1, show that pushing points one by one you
can get a chain of length 𝑛 − 𝑑 − 1 in the poset of regular subdivisions of A,
which coincides with the poset of faces of 𝚺(A).
b. To show that dim 𝚺( 𝐴) ≤ 𝑛−𝑑−1 show that the following 𝑑+1 affine equations
are independent and are satisfied by the GKZ-vector of every triangulation:
𝑛
∑︁
𝑧 𝑖 = (𝑑 + 1) vol(P),
𝑖=1
𝑛
∑︁
𝑗
𝑧𝑖 𝑝 𝑖 = = (𝑑 + 1) vol(P)𝑐 𝑗 , ∀ 𝑗 ∈ {1, . . . , 𝑑}.
𝑖=1
where 𝑐 = (𝑐 1 , . . . , 𝑐 𝑑 ) is the barycenter of P = conv(A) and 𝑧 = (𝑧 1 , . . . , 𝑧 𝑛 )

is the GKZ-vector of a triangulation.
Remark: the equations admit the following matrix form:

𝑝1 . . . 𝑝𝑛 𝑐
𝑧 = (𝑑 + 1) vol(P)
1 ... 1 1
5.5 In the first configuration of Problem 1.1, consider the GKZ-vectors of the eight
triangulations that use all points.
Remark: Since the secondary polytope is three dimensional, you can look at only three coordi-
nates; put differently, you are projecting R6 → R3 forgetting three coordinates. For the sake of
symmetry, use the coordinates of the three outer points).
With or without doing any computations show that:
a. These eight points lie in an affine plane.
b. The two non-regular triangulations produce the same point.
c. The mid-point of any two “opposite” triangulations is the same.
By symmetry, this implies the six regular triangulations produce the vertices of a
regular hexagon and the two non-regular ones both produce the barycenter of it.
5.6 Now consider the same configuration but with one of the two concentric triangles
in it slightly rotated with respect to the other one.
a. Are the eight GKZ-vectors still in an affine plane?
b. Do the two (formerly) non-regular triangulations still produce the same point?
c. Are the mid-points of any two “opposite” triangulations still the same?
By answering these (or other) questions argue that the eight points now are the
vertices of a 3-dimensional parallelepiped. What effect has this on the secondary
polytope, and on the set of regular triangulations?
Acknowledgements We thank Viviana Márquez for kindly sharing with us her board pictures of
Santos’ lectures, which have been of great help for producing these notes.
References
1. Jesús A. De Loera, Jörg Rambau and Francisco Santos. Triangulations: Structures for Al-
gorithms and Applications. Algorithms and Computation in Mathematics, vol. 25, Springer,
2010. DOI 10.1007/978-3-642-12971-1
2. Branko Grünbaum, Convex polytopes. Second edition revised and updated by V. Kaibel,
V. Klee and G. M. Ziegler. Graduate Text in Mathematics, vol. 221. Springer, 2003. DOI
10.1007/978-1-4613-0019-9
3. Carl Lee and Francisco Santos. Subdivisions and triangulations of polytopes.
In “Handbook of Discrete and Computational Geometry”, Third Edition, edited
by Csaba D. Toth, Joseph O’Rourke, Jacob E. Goodman, CRC Press, Novem-
ber 2017, pp 415–447. ISBN 9781498711395 A preprint version is available at
http://www.csun.edu/ ctoth/Handbook/HDCG3.html.
4. Rekha R. Thomas, Lectures in Geometric Combinatorics. Student Mathematical Library, vol.
33. IAS/Park City Mathematical Subseries. American Mathematical Society, Providence, RI;
Institute for Advanced Study (IAS), Princeton, NJ, 2006. viii+143 pp. ISBN: 0-8218-4140-8
5. Günter M. Ziegler, Lectures on polytopes. Graduate Texts in Mathematics, vol. 152. Springer,
1995. DOI 10.1007/978-1-4613-8431-1

Santos Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Santos Main

Uploaded by

Copyright:

Available Formats

Triangulations of polytopes and point

Francisco Santos, Mónica Blanco and Federico Castillo

Abstract These lecture notes summarize the theory of triangulations of polytopes

• For a comprehensive overview of several topics on triangulations that complement

0 Introduction to polytopes and affine geometry

Observe that every affine combination 𝑝 = 𝜆1 𝑝 1 + · · · + 𝜆 𝑛 𝑝 𝑛 of points gives rise

Definition 2 Let A = {𝑝 1 , . . . , 𝑝 𝑛 } ⊂ R𝑑 be a finite subset.

• A polytope P ⊂ R𝑑 is the convex hull of a finite set of points A. Its dimension 𝑘 is

Example: Affine and convex hulls

Fig. 1 Affine and convex hulls in dimension 2.

Theorem 3 (Minkowski-Weyl [5, Theorem 1.1]) Every polytope is bounded and

Related to these inequalities are the faces of a polytope:

Definition 4 Let P ⊂ R𝑑 be a polytope. For each linear functional 𝑓 : R𝑑 → R we

P 𝑓 := {𝑥 ∈ P : 𝑓 (𝑥) ≥ 𝑓 (𝑦) ∀𝑦 ∈ P},

that maximizes 𝑓 in P. See Figure 2 for an example.

Example: faces of a polytope

Definition 5 A 𝑘-dimensional simplex is a 𝑘-polytope with exactly 𝑘 + 1 vertices.

1 Triangulations and subdivisions of point configurations

From now on let us consider A := {𝑝 1 , . . . , 𝑝 𝑛 } a finite set of points in R𝑑 that is

For a point configuration, we need to consider the concept of face in a slightly

Fig. 3 Point configuration 5

Definition 7 In terms of point configurations, we say that a set A is a simplex if its

conv(S𝑖 ) ∩ conv(S 𝑗 ) = conv(S𝑖 ∩ S 𝑗 ) (1)

Running example: subdivisions and triangulations

{1245, 1345} {124, 134}

Fig. 4 Two different subdi- 5 5

However, {1245, 134} is not a subdivision: whereas the polytopal triangles

Definition 9 ([1, Def. 2.3.8]) Let T = {T1 , . . . , T𝑙 } and S = {S1 , . . . , S𝑚 } be two

Refinement of subdivisions of a point configuration A is a partial order (poset),

Running example: refinement of subdivisions

R = {125, 135, 234, 235} S = {1235, 234} T = {123, 234}

The easier problems are marked with an *.

1.4 *Do the same for 5 points in R3 , not all in a plane.

1 “Mother Of All Examples”. See [1]

1.9 *Show bĳections between the following two sets.

2 We consider the vertices of the 𝑛-gon to be labeled from 1 to 𝑛 in cyclic order.

2.1 Oriented matroids of point configurations

In this section we look at the oriented matroid of a point configuration, which is a

which we call affine evaluation of A by 𝑓 .

with 0/0 taken as 0.

A signature (𝜖1 , . . . , 𝜖 𝑛 ) can be uniquely represented by the pair (𝐸 + , 𝐸 − ) where

𝐸 + := {𝑖 : 𝑥𝑖 > 0} and 𝐸 − := {𝑖 : 𝑥𝑖 < 0}.

Signatures of affine dependences [1, Sec. 4.1.1]

Definition 13 An oriented circuit is the signature of the unique dependence of a

Example: signatures of circuits

(1, 2) (1, 23) (12, 34) (1, 234)

Signatures of affine evaluations [1, Sec. 4.1.2]

A non-constant affine functional 𝑓 : R𝑑 → R is 0 in the points 𝑥 ∈ R𝑑 of an

Clearly, the information given by an affine evaluation on A is more significant as

0 in a hyperplane spanned by points of A (at least 𝑑 affinely independent points of

Definition 14 ( [1, Defs. 4.1.6 and 4.1.18] ) For A a point configuration:

Definition 15 ( [1, Def. 4.1.43] ) The oriented matroid of a point configuration A is

Running example: oriented matroid

(14, 5), (123, 5), (14, 23), (45, 23),

The list of (oriented) cocircuits of A0 is:

• The faces of A are all possible intersections of facets.

With some extra effort one can prove the following:

Theorem 16 ( [1, Thms. 4.1.14, 4.1.29 and 4.1.31] ) Let S be a collection of

2.2 Oriented matroids of vector configurations and matrix

Let A ⊂ R𝑑+1 be a 𝑑-dimensional point configuration contained in an affine

2.3 Duality and the Gale transform

Let V ⊂ R 𝑘 and W ⊂ Rℓ be two vector configurations with 𝑛 elements, and represent

Following the linear part of Lemma 10 we get the following: