Claremont Colleges

Scholarship @ Claremont
HMC Senior Theses HMC Student Scholarship

2013

A Discrete Approach to the Poincare-Miranda

Theorem
Connor Thomas Ahlbach
Harvey Mudd College

Recommended Citation
Ahlbach, Connor Thomas, "A Discrete Approach to the Poincare-Miranda Theorem" (2013). HMC Senior Theses. 47.
https://scholarship.claremont.edu/hmc_theses/47

This Open Access Senior Thesis is brought to you for free and open access by the HMC Student Scholarship at Scholarship @ Claremont. It has been
accepted for inclusion in HMC Senior Theses by an authorized administrator of Scholarship @ Claremont. For more information, please contact
[email protected].
A Discrete Approach to the Poincaré-Miranda
Theorem

Connor Ahlbach

Francis Su, Advisor

Michael Orrison, Reader

Department of Mathematics

May, 2013
Copyright c 2013 Connor Ahlbach.

The author grants Harvey Mudd College and the Claremont Colleges Library the
nonexclusive right to make this work available for noncommercial, educational
purposes, provided that this copyright statement appears on the reproduced ma-
terials and notice is given that the copying is by permission of the author. To dis-
seminate otherwise or to republish requires written permission from the author.
Abstract

The Poincaré-Miranda Theorem is a topological result about the existence

of a zero of a function under particular boundary conditions. In this the-
sis, we explore proofs of the Poincaré-Miranda Theorem that are discrete
in nature - that is, they prove a continuous result using an intermediate
lemma about discrete objects. We explain a proof by Tkacz and Turzański
that proves the Poincaré-Miranda Theorem via the Steinhaus Chessboard
Theorem, involving colorings of partitions of n-dimensional cubes. Then,
we develop another new proof that relies on a polytopal generalization of
Sperner’s Lemma of DeLoera - Peterson - Su. Finally, we extend these dis-
crete ideas to prove the existence of a zero with the boundary condition of
Morales, in dimension 2.
Contents

Abstract iii

1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Sperner’s Lemma and Brouwer Fixed Point Theorem 7

2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Sperner’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Proving the Brouwer Fixed Point Theorem with Sperner’s
Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 The Steinhaus Chessboard Theorem 17

3.1 Background and Setup . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Proving the Steinhaus Chessboard Theorem . . . . . . . . . . 22
3.3 Proving the Poincaré-Miranda Theorem with the Steinhaus
Chessboard Theorem . . . . . . . . . . . . . . . . . . . . . . . 27

4 Polytopal Sperner’s Lemma 31

4.1 Polytopal Sperner’s Lemma . . . . . . . . . . . . . . . . . . . 31
4.2 Proving the Poincaré-Miranda Theorem with Polytopal Sperner’s
Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

5 Another Zero Theorem 37

5.1 Morales’s Generalization of the Intermediate Value Theorem 37
5.2 Discrete Proof in Dimension 2 . . . . . . . . . . . . . . . . . . 38

6 Conclusion 45
vi Contents

Bibliography 47
Chapter 1

Introduction

1.1 Overview
In this thesis, we study the Poincaré-Miranda Theorem (Poincaré (1886),
Miranda (1940)). First, we motivate the significance of the Poincaré-Miranda
Theorem as an n-dimensional generalization of the Intermediate Value The-
orem. We describe and discuss the advantages of discrete approaches to
topological results, such as the Poincaré-Miranda Theorem, and exemplify
this approach by proving the Brouwer Fixed Point Theorem using Sperner’s
Lemma. Next, we describe Tkacz’s and Turzański’s discrete approach to
the Poincaré-Miranda Theorem, which involves proving the existence of a
particular connected chain of cubes. We then explore an original discrete
proof of the Poincaré-Miranda Theorem using a generalization of Sperner’s
Lemma to polytopes. Finally, we investigate another theorem guaranteeing
the existence of a zero of a function and prove a special case.

1.2 Background
Let I = [0, 1]. We can state the Bolzano Intermediate Value Theorem as
follows:

Theorem 1 (Intermediate Value). If f : I → R is continuous,

f (0) < 0, and f (1) > 0, (1.1)

then there exists x ∈ I such that f ( x ) = 0.

2 Introduction

We visualize the Intermediate Value Theorem in Figure 1.1:

0
− +

Figure 1.1 Any continuous function on a line segment that is negative on one
endpoint and positive on the other endpoint must have a zero.

This classical result follows easily from the fact that continuous func-
tions preserve connectedness. Can we generalize Theorem 1 to higher di-
mensions?
Suppose f : I n → Rn is continuous. What conditions, if any, for f on
the boundary of I n would guarantee the existence of a zero of f ? For n = 1,
Theorem 1 tells us that condition 1.1 will guarantee the existence of a zero.
In n dimensions, the general question is answered by the Poincaré-Miranda
Theorem.
Before we state the Poincaré-Miranda Theorem, we will need some no-
tation. Let xi denote the i-th coordinate of x ∈ Rn . Then, we denote the i-th
opposite faces of I n by

Ii− = { x ∈ I n | xi = 0}, Ii+ = { x ∈ I n | xi = 1}.

In this paper, n will always represent the dimension of the cube. We visu-
alize these opposite faces in Figure 1.2 for n = 1, 2, 3:

I2+

I3+
I1− I1+ I1− I1+
I2+
I1−
I2− I1+
I2− I3−

Figure 1.2 The opposite sides of the n-cube for n = 1, 2, 3.

We state the Poincaré-Miranda Theorem (Poincaré (1886), Miranda (1940)).

Theorem 2 (Poincaré-Miranda). Suppose f = ( f 1 , . . . f n ) : I n → Rn is con-

tinuous and
f i ( Ii− ) ⊂ (−∞, 0), f i ( Ii+ ) ⊂ (0, ∞). (1.2)
 Background 3

Then, there exists x ∈ I n such that f ( x ) = 0.

The main significance of the Poincaré-Miranda Theorem is its ability to
guarantee the existence of a zero of a function without having to solve for
one explicitly. It can be used in more than 1 dimension unlike Theorem 1.
If f satisfies condition 1.2, f takes on opposite signs in the i-th coordi-
nate on the i-th opposite sides. In particular,

f i < 0 on Ii− , f i > 0 on Ii+ .

We visualize condition 1.2 for n = 1, 2 in Figure 1.3:

(·, +)

(−) (+) (−, ·) (+, ·)

(·, −)
Figure 1.3 The boundary conditions of the Poincaré-Miranda Theorem for n =
1, 2.

Let us give some examples so that we can see why Theorem 2 is a rea-
sonable guess for a generalization of Theorem 1.

1. If n = 1, Theorem 2 reduces to Theorem 1. Thus, the Poincaré-

Miranda Theorem generalizes the Intermediate Value Theorem.

2. If f i depends only on xi , then we can write

f ( x1 , . . . xi , . . . xn ) = ( f 1 ( x1 ), . . . f i ( xi ), . . . f n ( xn ))

where each f i is a continuous function of 1 variable. From Equation

1.2, we have
f i (0) < 0, f i (1) > 0.
Then, there exists pi ∈ I such that f i ( pi ) = 0, meaning that f ( p1 , . . . pn ) =
0, proving the existence of a zero of f .

3. Consider f : I 2 → R2 given by

f ( x, y) = ( x + y − 1, y − x ).
4 Introduction

It is easy to see that f satisfies condition 1.2. The zeros of f occurs at

the solution(s) to the system

x + y = 1, y = x,
1 1
∈ In.


which can easily be solved to get the zero 2, 2

4. Of course, not all instances of Theorem 2 follow directly from The-

orem 1, or have a 0 that can be solved algebraically. For example,
consider f : I 2 → R2 given by
!
1 2 1
 
1
f ( x, y) = sin x + cos y + 2( x − 1), y − x− + .
2 2 2

Here, f satisfies condition 1.2, but the existence of a 0 does not follow
easily from Theorem 1, and one cannot solve for a 0 using algebraic
techniques.

However, the simple connectivity argument used to prove Theorem 1

fails for general n. Of course, Theorem 2 has been established, and mutliple
proofs are known. The original proof by Poincaré in 1886 used homotopy
invariance (Poincaré (1886)). In 1940, Miranda showed that the Poincaré-
Miranda Theorem is equivalent to the Brouwer Fixed Point Theorem (Mi-
randa (1940)). The Brouwer Fixed Point Theorem which was established in
general by Hadamard in 1910 (Hadamard (1910)).
We next introduce the Brouwer Fixed Point Theorem, which will be rel-
evant to our discussion, especially in Chapter 2. Let B n = { x ∈ Rn |
∑in=1 xi2 ≤ 1} be the n-dimensional ball. Recall that a fixed point of a func-
tion f is a point x such that f ( x ) = x.
Theorem 3 (Brouwer Fixed Point). Every continuous map f : B n → B n has a
fixed point.
The proof of Theorem 3 using Sperner’s Lemma is well-known. Sperner’s
Lemma was established by Sperner in 1928 (Sperner (1928)), but it was
Knaster, K. Kuratowski, and S. Mazurkiewicz who used it to prove The-
orem 3 (Knaster et al. (1929)). This proof exemplifies a discrete approach
to a topological result. First, one proves an intermediary discrete result
or analog (in this case Sperner’s Lemma), and then uses it to prove the
theorem (in this case the Brouwer Fixed Point Theorem). This discrete ap-
proach has some advantages. First, the intermediate discrete analog is often
easier to prove, or at least requires less sophisticated techniques, than the
 Layout 5

topological result. This approach tends to avoid higher-level concepts, like

in this case degree or homology, that more advanced proofs use, making
them more accessible to less experienced mathematicians. Then, if our dis-
crete analog was chosen wisely, the topological result follows easily. We are
breaking the problem into two parts, both simpler than the whole. Further-
more, the discrete analog may be remarkable in and of itself. Finally, dis-
crete proofs can be, to some extent, constructive. So, for example, Sperner’s
Lemma, which proves the Brouwer Fixed Point Theorem, does give us an
idea where the fixed point is.
Since Theorem 2 and Theorem 3 are equivalent (Miranda (1940)), and
there is a discrete proof for Theorem 3, analogously, shouldn’t there be an
discrete proof for Theorem 2?
Yes! In fact, I have found two papers with such approaches, and found
an original proof as well (Kulpa (1997), Tkacz and Turzański (2008)) .
In his paper on the Poincaré-Miranda Theorem, Kulpa proves Theorem
2 by constructing a labelling on the cube, proving a variation, but not a
generalization, of Sperner’s Lemma, and finally using it to prove Theorem
2 (Kulpa (1997)). In fact, we will prove Theorem 2 using a generalization
of Sperner’s Lemma to polytopes. It makes sense that such similarity is
found to Sperner’s Lemma because Theorem 2 and Theorem 3 are equiva-
lent. Also, in their paper on the Steinhaus Chessboard Theorem, Tkacz and
Turzański also prove a discrete result about colorings on partitions of cubes
first and then use it to prove Theorem 2 (Tkacz and Turzański (2008)).

1.3 Layout
In Chapter 2, we give an example of how discrete approaches to topo-
logical results work by proving Sperner’s Lemma and using it to prove
Theorem 3. Furthermore, the proof of the Brouwer Fixed Point Theorem
using Sperner’s Lemma is a classical proof mathematicians should know,
and it is not a diversion because Sperner’s Lemma is quite relevant to our
work here. Many of the same ideas that show up in Sperner’s Lemma
will show up later. Then, in Chapter 3, we carefully clarify and proceed
through Tkacz’s and Turzański’s proof of Theorem 2 using the Steinhaus
Chessboard Theorem. In Chapter 4, we explain an original proof of the
Poincaré-Miranda Theorem using a polytopal generalization of Sperner’s
Lemma. This proof is similar in spirit to Kulpa’s (Kulpa (1997)), but its use
of a more general labelling and stronger result make it simpler than Kulpa’s
proof. In Chapter 5, we explore another zero theorem of Morales and give
a discrete proof in dimension 2.
Chapter 2

Sperner’s Lemma and Brouwer

Fixed Point Theorem

2.1 Background
How would one prove a topological result by first proving a discrete ana-
log? In this section, we demonstrate this technique with the proof of the
Brouwer Fixed Point Theorem via Sperner’s Lemma (Knaster et al. (1929)
Sperner (1928)). Note that similar techniques will show up when we present
new discrete proofs, so it is well worth discussing this example. To state
Sperner’s Lemma, we need to define simplices and Sperner labellings. First,
let us first go over some background.
Recall that the n-dimensional simplex, or n-simplex, which we denote
∆n , is the convex hull of n + 1 affinely independent points v1 , . . . vn+1 . For
n = 0, 1, 2, and 3,

• ∆0 is a point.

• ∆1 is a line segment.

• ∆2 is a triangle.

• ∆3 is a tetrahedron.

• See Figure 2.1 for a geometric visualization.

Here, in general, we denote the convex hull of x1 , . . . xk by conv( x1 , . . . xk ).

For each {i1 , . . . ik } ⊂ {1, . . . (n + 1)}, we say conv(vi1 , . . . vik ) is a face, in
particular, a (k − 1)-face, of ∆n . Note that each face of a simplex is also a
8 Sperner’s Lemma and Brouwer Fixed Point Theorem

v1
v1 v2
v3

v1 v2

v4

v1 v2

v3

Figure 2.1 From top to bottom: 0-simplex, 1-simplex, 2-simplex, and 3-simplex

simplex, of lesser or equal dimension. For notation, let [k ] = {1, 2 . . . k }

for all k ∈ N. A triangulation is simply a partition into a finite number of
smaller n-simplices where the intersection of 2 simplices is always a face of
both simplices. We give an example of a triangulation in Figure 2.2:

v3

v1 v2

Figure 2.2 An example of a trinagulation in dimension 2.

 Background 9

Note that triangulations do not allow two simplices to intersect so their

intersection is not a face of some simplex. We give an example of intersec-
tions that are not allowed in triangulations in Figure 2.3.

Figure 2.3 Intersections that are not allowed in triangulations

Of course, the vertices of triangulation T is simply union of the sets of

vertices of each simplex in T. In general, let T be a triangulation of ∆n into
n-simplices, and let V ( T ) be the set of vertices of T. Now, we have enough
background to state Sperner’s Lemma.

Definition 1 (Labelling). A labelling of T is simply a map L : V ( T ) → [n + 1],

where we say vertex v has label L(v).

Definition 2 (Sperner Labelling). We say L is a Sperner Labelling if the vertices

of ∆n have all the labels 1 through (n + 1), and every vertex belonging to a face F
of ∆n has the same label as one of the vertices of F.

Without loss of generality, we label vertex vi by i for all i ∈ [n + 1]. In

general, to be a Sperner labelling, each vertex in V ( T ) ∩ conv(vi1 , . . . vik )
must be labelled one of i1 , i2 , . . . ik . For example, with n = 2, any Sperner
labelling is of the form given in Figure 2.4.

Labelled 1 or 3 Labelled 2 or 3

1 2
Labelled 1 or 2

Figure 2.4 Form of a Sperner labelling in dimension 2.

10 Sperner’s Lemma and Brouwer Fixed Point Theorem

Here, the vertices on face conv(v1 , v2 ) must be labelled 1 or 2, vertices

on conv(v2 , v3 ) must be labelled 2 or 3, vertices on conv(v1 , v3 ) must be
labelled 1 or 3, and interior vertices can have any label from {1, 2, 3}. Next,
we state and prove Sperner’s Lemma, which naturally deals with Sperner
labellings (Sperner (1928)). We say that a simplex in T is fully labelled if its
vertices have every label in [n + 1].

2.2 Sperner’s Lemma

Theorem 4 (Sperner’s Lemma). Let T be a triangulation of ∆n . Then, in every
Sperner labelling of T, there exists an odd number of fully labelled simplices in T.
In particular, there is at least one fully labelled simplex in T.

Proof. We prove Theorem 4 by induction on n.

Base Case: Suppose n = 0. Then, ∆0 is a point p. Any Sperner labelling of T
must be p labelled with a 1, which yields a fully labelled simplex.

Induction Hypothesis: Suppose Theorem 4 holds for dimension n − 1.

Now, consider ∆n . Let T be a partition of ∆n and L : V ( T ) → [n +
1] be any Sperner labelling of T. We refer to the vertices of ∆n by their
label. By restricting L to conv(1, 2, . . . n), we get a Sperner labelling of the
(n − 1)-face conv(1, 2, · · · n). By the induction hypothesis, there exists an
odd number of faces on the face conv(1, 2 . . . n) with the labels 1 through n.
Consider any such face F0 .
Let a "door" refer to a (n − 1)-face in the triangulation with all the labels
1, 2, · · · n. Thus, each such F0 is a door. Now, construct a path Q of "pebbles"
as follows. Start at the simplex S0 that has F0 as a face. Put a pebble by face
F0 . Then, if we are in simplex S, and there is a door D in S without a pebble
by it, put one pebble on both sides of D, and move through that face into
the next simplex. Continue moving through doors and placing pebbles on
both sides of them until there are no unused doors in the simplex we are in.
Note that this path must terminate since we can only have a finite number
of doors, and we never go through the same door twice.
We have two types of such paths based on where they end. We give an
example of each below in dimension 2, where the doors are the 1-faces with
vertices labelled 1 and 2.
 Sperner’s Lemma 11

• Type 1 Paths: The path Q ends in another door on the boundary. If we

ever reach another door on the boundary, which must be on the face
conv(1, 2 . . . n) by the fact that we have a Sperner labelling, Q ends.

1 3
1
3 2

3 2
2 2

1 2 1 2 2
Figure 2.5 Example of a Type 1 Path

• Type 2 Paths: The path Q ends in a fully labelled simplex. Otherwise,

if Q does not end in a door on the boundary, it must end somewhere
in the interior, as it never goes through a simplex twice and there are
a finite number of simplices in T. But, Q can only end in the interior
if the simplex it ends in has precisely 1 door, which occurs if and only
if Q ends in a fully labelled simplex.

Thus, type 1 paths use exactly 2 doors on the boundary, and type 2
paths use exactly 1 door on the boundary. But by the Inductive Hypthesis,
we have an odd number of doors on the boundary, which all must occur
on face conv(1, 2 . . . n). It follows that we must have an odd number of
type 2 paths, each giving us one fully labelled simplex. So, the type 2 paths
correspond bijectively with an odd number of fully labelled simplices.
However, there may exist other fully labelled simplices in ∆n which
are NOT the end of such a type 2 path of pebbles. Yet, for each of these
simplices S, we may again construct paths P usign the same technique by
starting at S and moving through the doors. We call these type 3 paths.
Then, P cannot terminate at a door on the boundary or else S would have
been the end of some type 2 path above. By the same argument above, this
means that P must terminate in another fully labelled simplex U not hit by
a type 1 or type 2 path, because each door belongs to a unique path. And
12 Sperner’s Lemma and Brouwer Fixed Point Theorem

1 3
1
3 2

3 2
2 1

1 1 1 2 2
Figure 2.6 Example of a Type 2 Path

the type 3 path through the doors starting from U reverses P. We illustrate
this in Figure 2.7:
Any type 3 path pairs two fully labelled simplices in T that are the NOT
the end of type 2 paths. Thus, all fully labelled simplices that are not at the
end of type 2 paths come in pairs, so there are an even number of them. But
an odd number of fully labelled simplices are at the end of type 2 paths, so
we conclude that T contains an odd number of fully labelled simplices.

This completes the discrete analog part of the proof. We proceed to the
topological part.

2.3 Proving the Brouwer Fixed Point Theorem with

Sperner’s Lemma
We will use Theorem 4 to prove the No Retraction Theorem, another topo-
logical result, and then prove the Brouwer Fixed Point Theorem using the
No Retraction Theorem. Next, we define a retract and state the No Retrac-
tion Theorem.

Definition 3. For topological spaces A ⊆ B, f : B → A is a retract if f is

continuous and
f ( a) = a for all a ∈ A. (2.1)
 Proving the Brouwer Fixed Point Theorem with Sperner’s Lemma 13

1 3
1
3 2
2
3 2
1

1 1 1 2 2
Figure 2.7 Example of a Type 3 Path

Recall that B n = { x ∈ Rn | ∑in=1 xi2 ≤ 1} is the standard n-dimenional

ball, and Sn−1 = { x ∈ Rn | ∑in=1 xi2 = 1} is the standard (n − 1)-dimensional
sphere. Here, Sn−1 is the boundary of B n .

Theorem 5 (No Retraction). There does not exist a retract f : B n → Sn−1 .

Theorem 5 makes sense because it seems like however we retract the

ball to its boundary, the sphere, we would have to rip apart the ball and
violate continuity. We now use Theorem 4 to prove the Theorem 5.

Proof. Suppose to the contrary that there exists a retract f : B n → Sn−1 .

Then, we can deform this to a retract of ∆n to its boundary g : ∆n → ∂∆n ,
because B n and ∆n are homeomorphic topological spaces. Label the ver-
tices of ∆n by 1 through (n + 1). We refer to these vertices of ∆n by their
label. Let e be half of the distance from the center of each (n − 1)-face
F 6= conv(1, 2, · · · n) to the face conv(1, 2, · · · n) in ∆n . By symmetry, this
minimum distance is the same for each (n − 1)-face F 6= conv(1, 2, · · · n).
Let d : Rn × Rn → R≥0 be the Euclidean metric. Then, by the uniform
continuity of f , since ∆n is compact, there exists δ > 0 such that for all
x, y ∈ ∆n ,
d( x, y) < δ =⇒ d( g( x ), g(y)) < e.
Let T be a triangulation of ∆n such that for each simplex Q in T,

sup d( x, y) < δ.
x, y∈ Q
14 Sperner’s Lemma and Brouwer Fixed Point Theorem

Face G’s center

2e
g(vk )
<e
Face F (contains g(v1 ))

Figure 2.8 2-dimensional projection of where g(vk ), g(v1 ) appear on ∆n

It follows immediately that for all x, y ∈ Q for any simplex Q in T,

d( g( x ), g(y)) < e. (2.2)

Now, let us label each vertex v of T with the vertex of ∆n that g(v) is closest
to under the Euclidean metric in ∂∆n . Since a vertex on face F is closer
to some vertex on face F than any other vertex in ∆n , this labelling L is
Sperner. Then, by Theorem 4, there exists a fully labelled simplex S =
conv(v1 , . . . vn+1 ) in T, where L(vi ) = i.. So, by Equation 2.2,

d( g(v1 ), g(vi )) < e for all i ∈ [n + 1].

Now, f (v1 ) belongs to some (n − 1)-face F. Let k be the vertex of ∆n oppo-

site to face F and let G denote a face containing g(vk ). If F = G, then g(vk )
would be closer to some vertex of F than it is to k, contradiction the fact that
L(vk ) = k. Thus, G 6= F. By construction of e, this must mean that g(vk )
lies below the center of G, where k is on top. So, g(vk ) must be closer to
another vertex of G than it is to k. We visualize a 2-dimensional projection
of this phenomenon in Figure 2.8.
This picture illustrates why g(vk ) must be closer to another vertex of
G than it is to k, which we stated above. But this contradicts the fact that
vk was labelled k. From this contradiction, we conclude that no retract f :
B n → Sn−1 exists.

Finally, we prove the Brouwer Fixed Point Theorem. Recall the Brouwer
Fixed Point Theorem states that every continuous map f : B n → B n has a
fixed point.
 Proving the Brouwer Fixed Point Theorem with Sperner’s Lemma 15

Proof of Brouwer Fixed Point Theorem. Suppose to the contrary that there ex-
ists a continuous map f : B n → Bn without a fixed point. Then, define the
map g : B n → Sn−1 by g( x ) being the intersection of the ray from f ( x ) to x
with the boundary Sn−1 . We illustrate this map below:

g( x )

f (x)

Figure 2.9 Creating a retraction g from f , which has no fixed points

As f is continuous, and f ( x ) 6= x for all x ∈ B n , g is continuous. Fur-

thermore, for any x ∈ Sn−1 , g( x ) = x. Thus, g : B n → Sn−1 is a retract,
contradicting the Theorem 5. Hence, every continuous map f : B n → B n
has a fixed point.

Thus, we have proven Sperner’s Lemma, a discrete result, and used it

to establish the Brouwer Fixed Point Theorem. Our discrete approach is
complete. This gives us some idea of how discrete proofs of topological
results work. The labelling and path ideas in Sperner’s Lemma will be
similar to ideas we explore later in the paper. Understanding the proof of
Sperner’s Lemma will help the reader understand the rest of the paper.
Chapter 3

The Steinhaus Chessboard

Theorem

3.1 Background and Setup

In this chapter, we present Tkacz’s and Turzański’s proof of the Poincaré-
Miranda Theorem, which is also a discrete proof (Tkacz and Turzański
(2008)). The intermediary discrete analog is an n-dimensional generaliza-
tion of the Steinhaus Chessboard Theorem. In this chapter, we first de-
scribe the Steinhaus Chessboard Theorem and its generalization by Tkacz
and Turzański. Then, we present their proof of this generalization, which is
done by finding a chain of simplices with certain properties first and then
using that to establish the generalized Steinhaus Chessboard Theorem. Fi-
nally, we see how the generalized Steinhaus Chessboard Theorem can be
used to prove the Poincaré-Miranda Theorem.
We first state the original Steinhaus Chessboard Theorem:
Theorem 6. Consider an n by n chessboard. Place mines on any set of squares.
Then, one of the following holds:

1. A king can move from the left to the right of the board without passing
through any mined squares
2. A rook can move from the bottom to the top of the board using only mined
squares.

First, we say that the sets A1 , . . . Am form a (connected) chain if

Ai ∩ Ai+1 6= ∅ for all i ∈ [m − 1].

18 The Steinhaus Chessboard Theorem

This is very intutive definition of a chain. We just have that consecutive

sets must have nonempty intersection. An implication of this theorem is
that there is either a connected chain of non-mined squares from the left to
the right, or a chain of mined sqares form the top to the bottom. (The chain
from the top to the bottom need not use diagonals, but that is irrelevant for
or purposes.) Tkacz and Turzański generalize this theorem to n-dimensions
as below. We present an informal version now and make it more formal
later.

Theorem 7 (Generalized Steinhaus Chessboard Theorem - Informal). Create

a grid of n-dimensional cubes. Color each cube with one of n colors from 1, 2 . . . n.
Then there exists a set of cubes all colored i which connect the corresponding oppo-
site sides of the grid.

For example, consider the 2-dimensional 4 by 4 grids colored white and

black in Figure 3.1:

Figure 3.1 Example of the Steinhaus Chessboard Theorem in dimension 2

The Steinhaus Chessboard Theorem tells us that there either be a white

chain from left to right, or a black chain from top to bottom. In the left
grid, we have a white chain from the left to the right side, but no black
chain from top to bottom, and vice versa for the right grid. Yet, in the left
grid, note that we also have a black chain from left to right. This occurs
because we can exchange which colors are associated to which sides in the
Steinhaus Chessboard Theorem: we must either have a black chain from
left to right or a white chain from top to bottom. As we don’t have a white
chain from top to bottom, we must have a black chain from to left to right.
In fact, we can associate the colors with the sides in any 1-1 fashion. So, in
fact, for any permutation σ of [n], there exists an i-colored connected chain
between the σ(i )-th opposite sides of I n for some i.
 Background and Setup 19

From now on, the Steinhaus Chessboard Theorem refers to this gener-
alized result, not the 2-dimensional original.
Why does this make sense as a discrete analog for the Poincaré-Miranda
Theorem? Well, consider f = ( f 1 , . . . f n ). One observes that the zero sets

Z ( f i ) = { x ∈ I n | f i ( x ) = 0}

are hypersurfaces that must block all paths from Ii− , where f i < 0 to Ii+ ,
where f i > 0, because we can’t get from negative to positive continuously
without passing through 0. Hence, we expect that Z ( f i ) to be a hypersur-
face connecting all but the i-th opposite faces. Then, one would also expect
that these n zero sets, which we expect to be (n − 1)-dimensional, intersect
at at least a point. For example, in 2 dimensions, we expect these zero sets
to be curves between the first and second pairs of opposite faces, like in
Figure 3.2:

Figure 3.2 Any two curves between the pairs of opposite sides of a square
must intersect.

where the curves represent zero sets. In 2 dimensions, it is very intuitive

that 2 curves connecting opposite sides must intersect. Geometric intution
also suggests that this should generalize to n dimensions - the n zero set
hypersurfaces have at least one common point of intersection. This inter-
section is of course the set of zeros of f , and showing this is nonempty will
prove the Poincaré-Miranda Theorem.
Conversely, is there some way that f 6= 0 anyhere within I n make sense?
Now, if we color the our cubes such that f i 6= 0 on all i colored cubes,
then an i-colored connected chain between the i-th opposite side would
represent the fact that the zero set Z ( f i ) does not block all chains between
the i-th opposite sides. How would this help us? The existence of colored
chain between the i-th opposite sides, on which f i 6= 0, makes no sense. So,
20 The Steinhaus Chessboard Theorem

no such chain can exist, and then we see that each zero set Z ( f i ) does block
all chains between the i-th opposite sides. Then, using this property, we see
intuitively that this means the n zero sets must interesect in at least a point.
We will see how this idea plays out in the proof. Of course, this is not
a rigorous argument, but we can make it rigorous by discretizing the prob-
lem first, and then bringing it back to a topological problem. Discrete ap-
proaches have the potential to turn ideas like this into rigorous proofs.
We will present a more formal statement of the generalized Steinhaus
Chessboard Theorem later as we go through their argument, which we re-
fer to as the connected chain argument since they argue for the existence
of a conneected chain. First, we prove the existence of such a chain. Then,
we prove the Poincaré-Miranda Theorem using the generalized Steinhaus
Chessboard Theorem.
We now present my clarified interpretation of Tkacz’s and Turzański’s
proof. First, we will discuss relevant notation and preliminaries. Consider
a decomposition of I n with cubes of size 1k :
    
i1 i1 + 1 in in + 1
T (k) = , ×...× , | i j ∈ {0, 1, . . . k − 1} .
k k k k

Here, T (k ) is just an n-dimensional grid of cubes. In figure 3.3, we show

T (4) in dimension 2:

Figure 3.3 T(4) is a 4 by 4 grid of squares.

Next, define
 n
1 1 1 1
C (k) = − , ,...,1− ,1+ .
2k 2k 2k 2k

which is the set of the centers of all the cubes in T (k ) plus one layer of
points surrounding the entire cube. Adding the vertices of C (4) to T (4),
we get Figure 3.4:
 Background and Setup 21

Figure 3.4 C (k ) is the set of centers of the cubes in T (k ), plus one exterior
layer.

Note that for each cube t ∈ T (k ), there is a unique z ∈ C such that z ∈ t.

Furthermore, we have the points in C on the sides given by
   
1 1
C (k )i− = x ∈ C (k ) | x (i ) = − , C (k )i+ = x ∈ C (k) | x (i ) = 1 +
2k 2k

and set of exterior points in C (k ) given by

n
(C (k)i− ∪ C (k)i+ ).
[
∂C (k ) =
i =1

Let
1
ei = (0, . . . 0, , 0 . . . 0).
k
with ei (i ) = 1k be the i-th basis vector for i ∈ [n]. Now, we can triangulate
n
the cube 0, 1k using the simplices


(" # )
r 
1 1
( Tr )n,k = 0, eα(1) , . . . , ∑ eα(i) , . . . ,... | α ∈ Sn . (3.1)
i =1
k k

where Sn is the set of all permutations of [n] = {1, . . . n}. We visualize this
in two dimensions in Figure 3.5:
Next, let
2i1 − 1 2i1 + 1 2in − 1 2in + 1
    
0
T (k) = , ×...× , | i j ∈ {0, 1, . . . k} .
2k 2k 2k 2k

represent the set of n-dimensional cubes with side length 1k using vertices
in C (k ). This is basically a shifted version of T with another layer of cubes
22 The Steinhaus Chessboard Theorem

h i2
Figure 3.5 The triangulation ( Tr )2,k of 0, 1k .

Figure 3.6 T 0 (k ) is the set of cubes of side length 1

k using the vertices in C ( k ).

on the outside. Note that the vertices of T 0 (k ) lie in C (k ). We add in T 0 (4)

to T (4) and C (4) in Figure 3.6:
1 n
Let T be the triangulation of − 2k
 1 
, 1 + 2k into simplices where each
0
n
t ∈ T (k ) is triangulated using a shifted triangulation ( Tr )n,k of 0, 1k in


Equation 3.1. By the symmetry of ( Tr )n,k in Equation 3.1, T is a triangu-

1 n
 1 
lation of − 2k , 1 + 2k . Finally, for any simplex S, let V (S) refer to the
vertices of S.

3.2 Proving the Steinhaus Chessboard Theorem

Next, we extend an arbitrary coloring of T (k ) to a labelling on C (k ), and
then prove that we have a chain of simplices in T with certain properties,
to be explained below.

Lemma 8. For a coloring function F : T (k ) → [n], define a labelling φ : C (k ) →

[n] by
φ(z) = F (t) for z ∈ t ∈ T (k )
for z ∈
/ ∂C (k ). And, let

φ(z) = min{i | z ∈ C (k )i− ∪ C (k )+

(i +1) mod n
} (3.2)
 Proving the Steinhaus Chessboard Theorem 23

for z ∈ ∂C. Then, there exists a chain of simplices S1 , . . . Sm ∈ T such that

φ(V (S p ∩ S p+1 )) = [n]

for all p ∈ [m − 1]. Also, for the end simplices S1 and Sm , we have

S1 ∩ C (k )i− 6= ∅, Sm ∩ C (k )i+ 6= ∅.

for all i ∈ [n].

Proof. In this proof, one should notice many similarities to the proof of
Sperner’s Lemma. First, note that φ is well defined because, for all t ∈ T (k ),
there exists a unique z ∈ C such that z ∈ t.
We say a simplex S is n-colored if φ(V (S)) = [n]. First, let S1 = conv(z0 , z1 , · · · zn )
where
1
 
− 2k if j > i
zi ( j ) = 1 .
2k if j ≤ i
1 1 1
corner of T 0 (k ). Then, by Equation


This face lies by the − 2k , − 2k , · · · − 2k
3.2, φ(zi ) = i + 1 for all 0 ≤ i ≤ n − 1, so S1 is n-colored. Now, let us
construct a chain of simplices inductively. Suppose that S j has a single
unused ( that is, not shared by another simplex in the chain {Si } ) n- colored
face that does not intersect ∂C (k ). Now, any n-colored simplex, which has
n + 1 vertices, has exactly 2 n-colored faces. As S1 has one n-colored face
conv(z0 . . . zn−1 ) on the boundary, it has exactly one n-colored face, say F,
not used by another simplex or on the boundary.
Let us begin an algorithm staring at this S1 . Let F be the single unused
n-colored face of S j . Note that F is shared by another unique simplex in T
if and only if F does not intersect ∂C (k ). Then, let S j+1 ∈ T be the unique
simplex sharing face F with S j . Now both of S j ’s n-colored faces are used,
and one of S j+1 n-colored faces is left unused. So, we add S j+1 to the chain
and repeat inductively unless F’s vertices are all in ∂C (k). This process
must terminate because T has a finite number of simplices. Thus, we have
a chain of simplices S1 , . . . Sm that proceeds from S1 to intersect ∂C (k ) such
that
φ(V (S p ∩ S p+1 )) = [n].
However, an (n − 1)-face intersecting ∂C (k ) that is n-colored must have a
vertex belonging to Ci− ∪ C(+i+1) mod n) for all i ∈ [n] by Equation 3.2. So, it
must be either conv(z0 , . . . zn−1 ) or conv(w0 , . . . wn−1 ) where
1
 
1 + 2k if j > i
wi ( j ) = 1
1 − 2k if j ≤ i
24 The Steinhaus Chessboard Theorem

for 0 ≤ i ≤ n. This simplex is simply a reflection of conv(z0 , . . . zn−1 ) lies

1 1 1
corner of T 0 (k ).. We see that φ(wi ) = i


by the 1 + 2k , 1 + 2k , . . . 1 + 2k
for all i by definition, so φ(conv(w0 , . . . wn−1 )) = [n]. So, the last simplex
in our chain, Sm must be conv(w0 , . . . wn ), which is the n-simplex in our
triangulation T with conv(w0 , . . . wn−1 ) as a face. Furthermore, S1 and Sm
defined above satisfy

S1 ∩ Ci− 6= ∅, Sm ∩ (Ci+ ) 6= ∅.

for all i ∈ [n]. Thus, we have our chain of simplices S1 . . . Sm with the
desired properties.

This proof is remarkably similar to Sperner’s Lemma: mainly, we con-

struct the paths in basically the same way. We give a 2-dimensional exam-
ple of this theorem in action in Figure 3.7.
The next idea is to use this chain of simplices to construct a chain of
cubes in T (k ) with similar properties, and this will give us the Steinhaus
Chessboard Theorem we have alluded to earlier. Once we have the lemma
above, constructing this chain is relatively straightforward. We next state
and prove a formal version of the Steinhaus Chessboard Theorem using
Lemma 8.
Theorem 9 (Generalized Steinhaus Chessboard Theorem - Formal). For any
coloring function F : T (k ) → [n], there exists cubes P1 , . . . Pr , all colored i, in
T (k ) such that

Pj ∩ Pj+1 6= ∅, P1 ∩ Ii− 6= ∅, Pr ∩ Ii+ 6= ∅

for some i ∈ [n].

Proof. Define φ : C (k ) → [n] as in Lemma 8. By Lemma 8, there exists a
chain of simplices S1 , . . . Sm in T satisfying the properties of Lemma 8. As
Sm ∩ Ci+ 6= ∅ for all i, there exists a smallest index `1 such that

S`1 ∩ C (k )i+ 6= ∅

for some i ∈ [n]. Then, for this i, let `2 be the largest index such that

S`2 ∩ C (k )i− 6= ∅

which exists because S1 ∩ C (k )i− 6= ∅. Here, we assume that `2 ≤ `1 . A

similar approach will work if `1 ≤ `2 . For `2 ≤ `1 , the chain S1 , . . . Sm does
not hit C (k )i+ or C (k )i− strictly between `2 and `1 .
 Proving the Steinhaus Chessboard Theorem 25

C (k )2+

1 1 1 1 1 1

S13
1 2 1 1 1 2
S12
S11
1 1 1 1 1 2
S10
C (k )1− C (k )1+
S9
1 1 1 1 2 2
S4 S6 S8
S3 S5 S7
1 2 2 2 1 2
S2
S1
1 2 2 2 2 2

C (k )2−

Figure 3.7 We get a chain of simplices between opposite corners in I n , where

V (S p ∩ S p+1 ) is n-colored for all p.

1 −`2 −1
As φ(S j ∩ S j+1 ) = [n], we can find a sequence {z j }`j= 0 with z j ∈
S`2 + j ∩ S`2 + j+1 such that φ(z j ) = i for all j.
Let us show that for all j ∈ [`1 − `2 − 1], z j does not belong to ∂C (k ). By
definition, the only points in ∂C that are colored i must belong to C (k )i− ∪
C (k )+(i +1) mod n
. But by construction, z j ∈ / Ci− for j ≥ 1. Otherwise, we
would contradict the defintion of z0 and `2 . Furthermore, we must have
zj ∈/ C(+i+1) mod n for j ≤ `1 − `2 − 1. Otherwise, we would contradict our
definition of i, which was chosen to be the index such that the chain of
simplices S1 , . . . Sm touched Ci+ before Ck+ for any k 6= i. Therefore, for
1 ≤ j ≤ `1 − `2 − 1, z j does not belong to C (k )i− ∪ C (k)+ (i +1) mod n
. Because
the z j are all colored i, we conclude that z j ∈ / ∂C (k) for all j ∈ [`1 − `2 − 1].
26 The Steinhaus Chessboard Theorem

Hence, there exists a cube Pj ∈ T (k ) containing z j for all j ∈ [`1 − `2 − 1].

As each cube is centered at z j , we have F ( Pj ) = φ(z j ) = i for all j. Now,
S`2 ∩ C (k )i− is a single point z0 on the exterior ∂C (k ). So, the face opposite
to this point in S`2 , say F, must be the one we went through in our chain, or
S1 ∩ S2 , and is n-colored. Here, z1 ∈ S1 ∩ S2 is colored i. Then, z1 belongs
to the same simplex as z0 , so z1 is within 2k 1
of Ii− . Thus, we must have
P1 ∩ Ii− 6= ∅. Similarly, P`1 −`2 −1 ∩ Ii+ 6= ∅. Furthermore, we know that
z j , z j+1 ∈ S j+1 ∈ T . But by construction of T , this means that
1
|z j (s) − z j+1 (s)| ≤
k
for all s, or that z j and z j+1 differ by no more than 1k in every coordinate.
Now, each z j is the center of Pj , so that any point away from z j by no more
1
in each coordinate is in Pj . So, as 21 z j + z j+1 is away from z j by no


than 2k
1
more than 2k in each coordinate,
1

z j + z j+1 ∈ Pj .
2
Similarly,
1

z j + z j+1 ∈ Pj+1 .
2
Hence,
Pj ∩ Pj+1 6= ∅ for all j ∈ [`1 − `2 − 1].
As C (k )i− ⊂ Ii− , C (k )i+ ⊂ Ii+ , we have that
P1 ∩ Ii− 6= ∅, P`1 −`2 −1 ∩ Ii+ 6= ∅.
Thus, after ignoring repeats of the same cube, we have a chain P1 , . . . Pr
with the desired properties.

We will show how the construction of the cubes plays out in Figure 3.8.
In the example for Lemma 8, S10 was the first simplex to touch the right side
C1+ . And, S3 was the last simplex to touch the left hand side C1− . Using this,
we construct the sequence {z j } described in Theorem 9. For convenience,
we ignore repeats of the same vertex for the sequence {z j }.
Each vertex z j in the sequence in Figure 3.8 is labelled 1. From the above
sequence, we get the desired chain P1 , P2 , P3 , P4 , which are colored 1, shown
in Figure 3.9.
Now we have established our discrete analog. Finally, we use it to prove
the Poincaré-Miranda Theorem.
 Proving the Poincaré-Miranda Theorem with the Steinhaus Chessboard Theorem 27

C (k )2+

1 1 1 1 1 1

S13
1 2 1 1 1 2
S12
S11
1 1 1 1 z4 2
S10
C (k )1− C (k )1+
S9
1 z1 z2 z3 2 2
S4 S6 S8
S3 S5 S7
1 2 2 2 1 2
S2
S1
1 2 2 2 2 2

C (k )2−

Figure 3.8 The chain of simplices in Lemma 8 gives rise to a sequence of i-

colored points that are no more than 1k apart in each coordinate going between
Ii− and Ii+ .

3.3 Proving the Poincaré-Miranda Theorem with the

Steinhaus Chessboard Theorem
Proof of Poincaré-Miranda Theorem. Suppose f = ( f 1 , . . . f n ) : I n → Rn is
continuous and

f i ( Ii− ) ⊂ (−∞, 0), f i ( Ii+ ) ⊂ (0, ∞).

Suppose to the contrary that f does not have a zero in I n . Then, let

Ui = { x ∈ I n | f i ( x ) 6= 0}.
28 The Steinhaus Chessboard Theorem

C (k )2+

P4 (1)
C (k )1− C (k )1+
P1 (1) P2 (1) P3 (1)

C (k )2−

Figure 3.9 The sequence of points corresponds a sequence of cubes connect-

ing Ii− and Ii+ .

Note that each Ui is open because R − {0} is open and each f i is continuous.
As f 6= 0 anywhere, {Ui }in=1 is an open cover for I n , which is compact. So,
by the Riemann Lebesgue Lemma, there exists e > 0 such that for each
p ∈ In ,
B( p, e) ⊂ Uj
√
for some j ∈ [n]. So, choosing k such that 1k < 2 ne, so that each cube of
side length 1k is contained ina ball of radius e, we have

t ∈ T (k ) =⇒ t ∈ Uj for some j ∈ [n].

So, each cube in T (k ) is contained within one of these open sets. Hence, the
coloring function F : T (k ) → [n] given by

F (t) = min{ j | t ∈ Uj }

is well-defined. So, by the Steinhaus Chessboard Theorem, there exists an

i-colored chain P1 , . . . Pr such that

Pj ∩ Pj+1 6= ∅, P1 ∩ Ii− 6= ∅, Pj ∩ Ii+ 6= ∅.

 Proving the Poincaré-Miranda Theorem with the Steinhaus Chessboard Theorem 29

Thus there exists x ∈ P1 ∩ Ii− , y ∈ Pr ∩ Ii+ . Then, f i ( x ) < 0, f i (y) > 0.

Furthermore, as we have a chain, K := rj=1 Pj is connected, so f (K ) must
S

be connected as f is continuous. Hence, 0 ∈ [ f i ( x ), f i (y)] ⊂ f (K ). So, there

exists p ∈ K such that
f i ( p) = 0.
However, p ∈ Pj for some j, so as Pj is i-colored, p ∈ Ui . But this means
that
f i ( p) 6= 0,
contradicting the above. We visualize this chain in Figure 3.10:

f i 6= 0 Ii+
fi (x) < 0
f i (y) > 0
Ii−

Figure 3.10 We have an i-colored chain, where f i 6= 0, from Ii− , where f i < 0,
to Ii+ , where f i > 0.

We conclude that f has a zero in I n . This completes or proof of the

Poincaré-Miranda Theorem via the Steinhaus Chessboard Theorem .
Chapter 4

Polytopal Sperner’s Lemma

4.1 Polytopal Sperner’s Lemma

In this chapter, we present an original proof of the Poincaré-Miranda Theo-
rem using a generalization of Sperner’s Lemma to polytopes. Although it is
similar in spirit to the proof of the Poincare Miranda Theorem presented in
Kulpa, it differs in the labelling it uses and the version of Sperner’s Lemma
used. In this proof, we have a simpler labelling and argument than that in
Kulpa because we employ a stronger result. We present our original proof
below.
Let P ⊂ Rn be a polytope, and let T be a triangulation of P into n-
dimensional simplices. Let {v1 , . . . vm } be the set of vertices of polytope P
and let V ( T ) be the set of vertices in triangulation T. Then, a labelling of T
is a map ` : V ( T ) → [m].

Definition 4 (Sperner Labelling). Let P be a polytope with vertices {v1 , . . . vm }.

Then, a labelling ` : V ( T ) → [m] is Sperner if for all faces F of polytope P,

v∈F implies `(v) ∈ {i | vi ∈ F }.

This is a straightforward extension of a Sperner labelling from simplices

to polytopes. In particular, this means that `(vi ) = i for all vertices vi of
polytope P. So, a labelling of a triangulation is Sperner if every vertex
within a face F is labelled the same as the index of some vertex of F. This
must hold for all faces F of the polytope. In fact, using this generalized
notion of a Sperner labelling, we get a generalized version of Sperner’s
Lemma.
32 Polytopal Sperner’s Lemma

Theorem 10 (Polytopal Sperner’s Lemma). Let ` : V ( T ) → [m] be a Sperner

Labelling of a triangulation T of polytope P. Then, there exists a simplex in T with
distinct labels j(0) , j(1) , . . . j(n) such that

v j(0) , v j(1) , . . . v j(n) do not all lie in the same proper face. (4.1)

By a proper face, we simply mean a face of dimension less than n. This

result follows from of a stronger result due to DeLoera-Peterson-Su (De Lo-
era et al. (2002)), but it is all we need for our purposes here. Let us ap-
ply it in particular to I n , and understand how 4.1 restricts the simplices
conv(v j(0) , v j(1) , . . . v j(n) ) on vertices. Let us label the vertices of I n in binary.
In particular, let
( x1 , · · · xn ) = v x1 ,···xn
when xi ∈ {0, 1} for all i. Furthermore, we construct our Sperner labelling
in binary as well, so for vertices,

`(v x1 ,···xn ) = ( x1 , · · · xn ).

We let ji denote the i-th coordinate of binary label j. We call a simplex fully
labelled as long as its vertices have distinct labels.

Theorem 11 (Polytopal Sperner’s Lemma on a Cube). Let ` : V ( T ) → [m]

be a Sperner Labelling of a triangulation T of I n . Then, there exists a simplex in T
whose vertices have distinct binary labels j(0) , j(1) , . . . j(n) such that for all i, there
exists ai , bi ∈ {0, . . . n} with
( ai ) ( bi )
ji = 0, ji = 1. (4.2)

Proof. By Polytopal Sperner’s Lemma, there exists a simplex in T with dis-

tinct binary labels j(0) , j(1) , . . . j(n) satisfying inequality 4.1. Now, suppose
(k)
to the contrary there exists some i such that ji is constant for all k ∈ [n].
−
Then, by defintion, we would have v jk ∈ Ii or v jk ∈ Ii+ for all k ∈ [n]. But,
either way,
v j(0) , v j(1) , . . . v j(n) all lie in Ii− or Ii+ ,
contradicting Condition 4.1. So we conclude that for all i, there exists
ai , bi ∈ {0, . . . n} with
(a ) (b )
ji i = 0, ji i = 1.
 Proving the Poincaré-Miranda Theorem with Polytopal Sperner’s Lemma 33

We refer to a simplex as fully labelled as long it has distinct labels. Con-

dition 4.2 says that the binary labels of the fully labelled simplex must, as
a whole, contain a 0 and a 1 in each coordinate. We give an example in 2
dimensions of a binary Sperner labelling on a square and we point out 2
fully labelled simplices whose corresponding vertices satisfy Condition 4.2
in Figure 4.1:

01 11 01 11

01 10
01 11
00 11
00 11

00 10 00 10
Figure 4.1 We have a Sperner labelling of a triangulation of I 2 , and we have
two fully labelled simplexes.

Then, the triangle labelled in red has distinct labels whose correspond-
ing vertices are not all contained in a proper face. This triangle also has
the labels 00, 01, 11, which, considering the labels as a whole have a 0 and
1 in each coordinate. So, they satisfy the condition in Polytopal Sperner’s
Lemma on a Cube.

Now, we are ready to prove the Poincaré-Miranda Theorem via Theo-

rem 11.

4.2 Proving the Poincaré-Miranda Theorem with Poly-

topal Sperner’s Lemma
Proof of Poincaré-Miranda Theorem. Consider any continuous function f =
( f 1 , . . . f n ) : I n → Rn satisfying

f i ( Ii− ) ⊂ (−∞, 0), f i ( Ii+ ) ⊂ (0, ∞)

34 Polytopal Sperner’s Lemma

for all i ∈ [n]. Let T be any triangulation of I n . We then define a binary

labelling ` : V ( T ) → {0, 1}n whose coordinates are given as follows:

1 if f i (v) ≥ 0
`i ( v ) = . (4.3)
0 if f i (v) < 0
Lemma 12. The labelling ` defined above is a Sperner labelling on I n .

Proof. Suppose that w is in some proper face F on I n . (The Sperner condi-

tion is trivially true for the only non-proper face - the whole of I n .) Then,
we can find a nonempty set of coordinates {i1 , i2 . . . ik } ⊂ [n] such that

F = { x ∈ I n | xir = g(ir ) for all r ∈ [k ]}

for some map g : {i1 , i2 . . . ik } → {0, 1}. That is, face F is defined by each
element of this set of coordinates as equal to either 0 or 1. Then, we can
write F as an intersection of (n − 1)-dimensional faces:
   

Ii−r  Ii+r  .
\ \ \
F= 
{r | g(ir )=0} {r | g(ir )=1}

Hence, by condtion 1.2, we know that for each coordinate ir ,

f ir (w) < 0 if g(ir ) = 0, f ir (w) > 0 if g(ir ) = 1.

So, by Definition 4.3,

` ir ( w ) = g ( i r )
for all r ∈ [k]. By the definition of face F, this means that the vertex v`(w)
belongs to face F. Thus, w shares a label with some vertex of F, meaning
that ` is Sperner.

Using this fact, and the compactness of I n , we can prove the existence
of a zero. Consider e > 0, and let d be the Euclidean metric in Rn . Since f is
continuous on a compact set and hence uniformly continuous, there exist
δ > 0 such that
e
d( x, y) < δ implies d( f ( x ), f (y)) < √ .
n
Now, let Rk refer to a sequence of triangulations of I n into simplices such
that
1
max d( x, y) < for all Q ∈ Rk .
x,y∈ Q k
 Proving the Poincaré-Miranda Theorem with Polytopal Sperner’s Lemma 35

There exists simplices Qk ∈ Rk such that Qk has distinct binary labels satis-
fying the condition of Theorem 11.
Consider any sequence { pk } with pk ∈ Qk . Now, consider K0 large
enough such that K10 ≤ δ. Fix any k ≥ K0 . Then, for all x, y ∈ Qk ,
1 1 e
d( x, y) < < < δ =⇒ d( f ( x ), f (y)) < √ .
k K0 n
As Qk satisfies the condition 4.2 in Theorem 11, for all i, there exists vertices
xi , yi of Qk such that
`i ( xi ) = 0, `i (yi ) = 1.
By Definition 4.3,
f i ( xi ) < 0, f i (yi ) ≥ 0.
As xi , yi ∈ Qk ,
e e
d( f ( xi ), f ( pk )) < √ , d( f (yi ), f ( pk )) < √ .
n n
So, in particular, we must have
e e
f i ( yi ) − √ < f i ( p k ) < f i ( xi ) + √ .
n n
As f i ( xi ) < 0, f i (yi ) ≥ 0, the previous inequality implies
e
| f i ( pk ) | < √ .
n
So, we have v
e 2
u n  
| f ( pk )| < ∑ √
u
t = e.
i =1 n
But this tells us that in Rn ,
f ( pk ) → 0.
Now, we know In is compact, so there exists a convergent subsequence
{ pkr } of { pk } such that pkr → p for some p ∈ I n - see Figure 4.2 for a
visualization of this. Then, { f ( pkr )} is also a subsequence of { f ( pk )}, so
f ( pkr ) → 0 as well. By the continuity of f , we conclude that
f ( p) = lim f ( pkr ) = 0.
r →∞

So, f has a zero in In.

36 Polytopal Sperner’s Lemma

Figure 4.2 We can find a convergent subsequence of points within the sim-
plices from Polytopal Sperner’s Lemma. The limit point of this convergent sub-
sequence is a zero of f .
Chapter 5

Another Zero Theorem

5.1 Morales’s Generalization of the Intermediate Value

Theorem
There is a another way we can generalize the Intermediate Value Theorem.
In this section, we let I = [−1, 1], not [0, 1], which we used in all previous
sections. We can restate the Intermediate Value Theorem as follows: if f :
I → R is continuous and

h f ( x ), x i > 0 for x = ±1,

then there exists x ∈ I n such that f ( x ) = 0 (Morales (2002)). Here, h·, ·i is
the Euclidean inner product on Rn given by
n
h x, yi = ∑ xi yi .
i =1

We can generalize the above by simply keeping the boundary condition as

the above inner product, and this turns out to hold for I n as well (Morales
(2002)). Let
n
∂I n = ( Ii+ ∪ Ii− )
[

i =1
denote the boundary of the cube n
I .
Theorem 13 (Morales). If f : I n → Rn is continuous and satisfies

h f ( x ), x i > 0 for all x ∈ ∂I n ,

then there exists x ∈ I n such that f ( x ) = 0.
38 Another Zero Theorem

5.2 Discrete Proof in Dimension 2

We are interested in proving this geometric result using combinatorial meth-
ods. In this case, we have found an original discrete proof of this result in
dimension 2. The proof again finds a fully labelled simplex of a particular
labelling. We shall prove the following:
Theorem 14. Suppose that f : I 2 → R2 is continuous and satisfies

h f ( x ), x i > 0 for all x ∈ ∂I 2 . (5.1)

Then there exists x ∈ I n such that f ( x ) = 0.

Proof. First, let f = ( f 1 , f 2 ). Let T be a triangulation of I 2 . Let us label the
vertices of the triangulation T as follows:


 1 if f 1 (v) ≥ 0, f 2 (v) ≥ 0
2 if f 1 (v) < 0, f 2 (v) ≥ 0

`(v) = .

 3 if f 1 (v) < 0, f 2 (v) < 0
4 if f 1 (v) ≥ 0, f 2 (v) < 0


This creates a labelling ` : V ( T ) → {1, 2, 3, 4}. This is actually the same

labelling we used for the square in our proof of the Poincare Miranda Theo-
rem using Polytopal Sperner Lemma’s except that we use the labels {1, 2, 3, 4}
instead of the binary labels {00, 01, 10, 11}. Equivalently, v is labelled by the
quadrant f (v) lies in, with the quadrants given by

Q1 = {( x, y) ∈ R2 | x ≥ 0, y ≥ 0}, Q2 = {( x, y) ∈ R2 | x ≤ 0, y ≥ 0}
Q3 = {( x, y) ∈ R2 | x ≤ 0, y ≤ 0}, Q4 = {( x, y) ∈ R2 | x ≥ 0, y ≤ 0},

with some negligible choices made at the axes. However, because of the
different boundary condition, ` need not be Sperner.
Now, since ∂I 2 is compact, the image f (∂I 2 ) is compact as well, and it
can’t include 0 by Equation 5.1. This means that there exists a ball B(0, e)
in R2 for some e > 0 such that B(0, e) f (∂I 2 ) = ∅. Next, by continuity,
T

there exists δ > 0 such that

d( x, y) < δ =⇒ d( f ( x ), f (y)) < e. (5.2)

We call a triangulation "fine-enough" if for any simplex S in T,

sup d( x, y) < δ. (5.3)

x,y∈S
 Discrete Proof in Dimension 2 39

Lemma 15. Let T be the set of all fine-enough triangulations that includes the
points (±1, 0), (0, ±1) and at least 1 other vertex in each quadrant as vertices.
Then, for each T ∈ T ,

1. No 13 or 24 edges appear in T.

2. If v ∈ Qi for i ∈ {1, 2, 3, 4}, then `(v) 6= the number of the quadrant

opposite Qi . Or,

v ∈ Q1 =⇒ `(v) 6= 3, v ∈ Q2 =⇒ `(v) 6= 4
v ∈ Q3 =⇒ `(v) 6= 1, v ∈ Q4 =⇒ `(v) 6= 2.

3. There exists a vertex v ∈ Qi labelled i for all i ∈ {1, 2, 3, 4}.

Proof. Let T ∈ T . Let x, y ∈ V ( T ) ∩ ∂I 2 be adjacent vertices on the bound-

ary. So, f ( x ), f (y) ∈ f (∂I 2 ), the image of the boundary. Then, by conditions
5.2 and 5.3,
d( f ( x ), f (y)) < e.
Now, because B(0, e) f (∂I 2 ) = ∅, it is impossible for f ( x ) and f (y) to be
T

in opposite quadrants, namely Q1 and Q3 or Q2 and Q4 . We illustrate this

in Figure 5.1:

y f B(0, e)
<δ f (x)
x

f (y)

Figure 5.1 In fine enough triangulations, two adjacent vertices on ∂I 2 cannot

be mapped to opposite qudrants by f .

If f ( x ) and f (y) were in opposite quadrants, they would be at least e

apart, because they lie ouside B(0, e), but we know this cannot happen.
Thus, fine-enough triangulations avoid all instances of 13 or 24 edges on
the boundary ∂I 2 . This proves condition 1 in Lemma 15.
Furthermore, we can also interpret the boundary condition as stating
that f ( x ) and x can’t be apart by an angle of π2 or more. So, f can’t move
points from one quadrant to its opposite quadrant. Thus, we can’t have
40 Another Zero Theorem

labels of any quadrant in its opposite quadrant. This proves condition 2 in

Lemma 15.
Furthermore, as f can’t move points by an angle of π2 or more,

`(1, 0) ∈ {1, 4}, `(0, 1) ∈ {1, 2}, `(−1, 0) ∈ {2, 3}, `(0, −1) ∈ {3, 4}.

Suppose to the contrary that there is no label of 1 in Q1 . But then, Q1 must

look like Figure 5.2:

2 or 4
2

4
Figure 5.2 Without a 1 in Q1 , we would have only 2s and 4s, forcing a 24 edge.

where we may have many vertices between this 2 and 4. And, without any
1s in Q1 , we can only have 2s and 4s in Q1 , and then there is no way to
avoid a 24 edge, which we know we cannot have. Hence, there is at least
one 1 in Q1 . Similarly, there is at least one i in Qi for all i ∈ {1, 2, 3, 4},
proving condition 3 in Lemma 15.

Next, using these conditions, we show that there must be an odd num-
ber of 12 edges on ∂I 2 . Without loss of generality, suppose that label i in Qi
appears at the corner of the square in that quadrant, and that `(0, 1) = 1.
Then, we have at least the labels in Figure 5.3:
Consider the segment from (-1,1), labelled 2, to (0,1), labelled 1, in the
picture above. I claim there must exist an odd number of 12 edges of T on
this segment. We are in Q2 , so no 4 appears. Also, as we move from left
to right from the leftmost 2 to the rightmost 1 on this segment, there must
exist an odd number of switches from 1 to 2 or 2 to 1. During such a switch,
we can’t have any intervening 4’s because no 4’s appear in Q2 . Also, we
can’t have any intervening 3’s because we cannot have any 13 edges in T.
Hence, each 1 to 2 or 2 to 1 switch must take place along a 12 edge (in
either order). Moreover, each 12 edge causes a switch from 1 to 2 or 2 to 1.
It follows that there are an odd number of 12 edges in T along this segment.
This segment must look like Figure 5.4:
 Discrete Proof in Dimension 2 41

1
2 1

2 or 3 1 or 4

3 4
3 or 4
Figure 5.3 We know each corner must be labelled with its quadrant, and each
midpoint must be labelled with one its two adjacent quadrants.

2 3 2 1 2 3 2 ... 1
Figure 5.4 The (0,1) to (-1,1) edge must be of this form, where each number i
actually represents a sequence of i’s in a row, and the 3 sequences are allowed
to be empty, in which case the 2 sequence after it can be combined with the
previous 2 sequence.

We see that each 1 to 2 or 2 to 1 switch must take place on a 12 edge. And

we have an odd number of these because we start with 2 and end with 1.
In fact, similarly, each 1 to 2 or 2 to 1 switch in Q1 or Q2 has a one-to-
one correspondence with 12 edges in T. Next, consider the vertical segment
from (-1, 1), labelled 2, to (-1, 0), labelled 2 or 3 - above. Again, we are in Q2 ,
so we can have only 1s, 2s, or 3s. Suppose to the contrary that the last 1 or
2 before we end at (-1,1) is a 1. Then, this segment would have to end with
a 13 edge in T which we know we can’t have, followed by some number of
3s - see Figure 5.5:

2 ... 1 3
Figure 5.5 The (-1,1) to (-1,0) edge cannot end in a 13, where the last 3 rep-
resents a nonempty sequence of 3s.

Therefore, the last 1 or 2 must be a 2, so we have an even number of 1

to 2 or 2 to 1 switches. So, by the above, we have an even number of 12
edges in T on this segment. Furthermore, on the (0,1), labelled 1, to (1,1),
labelled 1, segment above, we must have an even number of 1 to 2 or 2 to
42 Another Zero Theorem

1 switches, and thus an even number of 12 edges. Also, similarly to the

previous argument, we must have an even number of 12 edges on the (1,1),
labelled 1, to (1,0), labelled 1 or 4 segment. Lastly, we can have no 12 edges
in Q3 or Q4 because we cannot have any 1s in Q3 or 2s in Q4 . We have
shown that the number of 12 edges in each section of ∂I 2 is as in Figure 5.6:

Odd 1 Even
2 1
Even Even
2 or 3 1 or 4

None
Figure 5.6 The parity of the number of 12 edges in each region of ∂I 2 .

This demonstrates that ∂I 2 contains an odd number of 12 edges in T.

Finally, we can use the same path chasing argument used to prove
Sperner’s Lemma on these 12 edges on ∂I 2 to find a simplex with 3 distinct
labels in T. Constructing paths using the same technizue as in Sperner’s
Lemma, we again see that type 1 paths, which start and end on the bound-
ary, use 2 doors, or 12 edges. As we have an odd number of 12 edges, there
is at least one type 2 path, which must end in a fully labelled simplex in T.
As we are in 2 dimensions, the vertices corresponding to these distinct
labels satisfy the nonzero volume condition 4.1 and condition 4.2 if treated
as binary labels. Therefore, we can find a sequence of finer and finer tri-
angulations { Rk }, all of which are "fine enough" to guarantee the labelling
boundary conditions above hold, such that
!
lim sup d( x, y) = 0,
k→∞ x,y∈S∈ Rk

where S ranges over all simplices in Rk . Then, by the same argument as

used in the Polytopal Sperner’s Lemma section to deduce that we have a
zero, we can deduce that f has a zero.
Discrete Proof in Dimension 2 43
Chapter 6

Conclusion

In this thesis, we demonstrated the utility of proving topological results

using intermediate lemmas about discrete objects. We presented two very,
but both elegant, discrete proofs of the Poincaré-Miranda Theorem. The
first involved the Steinhaus Chessboard Theorem, which states the exis-
tence of a particular colored chain in colorings of grids. The second in-
volved a generalization of Sperner’s Lemma to polytopes. Such discrete
proofs are simpler and more accessible than the higher-level topological
proofs of Poincaré-Miranda Theorem. The proof using Polytopal Sperner
can also be of use in actually finding a zero.
Possible future work includes finding a discrete proof for Morales’s
Theorem with the boundary condition h f ( x ), x i > 0 for general n. I do
not see how to generalize the discrete proof I gave for dimension 2.
Bibliography

De Loera, J., E. Peterson, and F. Su. 2002. A polytopal generalization of

sperner’s lemma". Journal of Combinatorial Theory Series A 100:1 – 26.
Hadamard, J. 1910. Sur quelques applications de l’indice de kronecker.
Introduction to: J Tannery, La The’orie des Fonctions d’une Variable, Hermann,
Paris 6.
Knaster, B., K. Kuratowski, and S. Mazurkiewicz. 1929. Ein beweis des
fixpunktsatzes fur n-dimensionale simplex. Fund Math 14:132–137.
Kulpa, Wladyslaw. 1997. The poincaré-miranda theorem. The American
Mathematical Monthly 104(6):pp. 545–550. URL http://www.jstor.org/stable/
2975081.

Miranda, C. 1940. Un osservazione su una teorema di brouwer. Boll Union

Mat Ital 527.
Morales, C. 2002. A bolzano’s theorem in the new millenium. Nonlinear
Analysis 51:679 – 691.
Poincaré, H. 1886. Sur les courbes definies par une equation differentielle.
J Math Pures Appl 85:151–217.
Sperner, E. 1928. Neuer beweis fur die invarianz der dimensionszahl und
des gebietes. Abhandlungen aus dem Mathematisches Seminar, Hamburg Uni-
versitat 6:265 – 272.
Tkacz, Przemyslaw, and Marian Turzański. 2008. An n-dimensional
version of steinhaus’ chessboard theorem. Topology and its Applica-
tions 155(4):354 – 361. doi:10.1016/j.topol.2007.07.005. URL http://www.
sciencedirect.com/science/article/pii/S0166864107003033. Proceedings of the
Tenth Prague Symposium on General Topology and its Relations to Mod-
ern Analysis and Algebra.