INTRODUCTION TO GEOMETRIC

GROUP THEORY
A REPORT
submitted in partial fulfillment of
the requirements for the award of
the degree of

Bachelor of Science
in

MATHEMATICS
by

RAJ KISHAN BARI

DEPARTMENT OF MATHEMATICS
RAADHANATH RATH VIGYAN DEGREE MAHAVIDYALAYA KHUNTUNII,
CUTTACK

Regd no-2102010620420080
Roll no- 2102010620420080
 ACKNOWLEDGMENT

First of all, I would like to thank my project advisor Mr. T.P.Gadnayak for his expert
advice and encouragement throughout the project. It would have been almost
impossible to carry out the project without his help and valuable comments. I would also
like to thank my teacher Mrs.Gayatree Priyadarshini for their positive comments and for
showing a keen interest in my project through careful evaluation of my presentations. I
would like to thank my batch mates and seniors at the department for constantly
motivating me and carrying out wonderful discussions on various topics of interest
during the project. Thanks to Swayam,my best friend at RRVDM, for always supporting
me and making life at RRVDM joyous and wonderful. I would also like to express my
gratitude to my numerous friends, for always being there by my side throughout this
journey. Finally, I would like to thank my family for their love, support and faith in me.

Raj kishan Bari

 ABSTRACT

In this project, we will try to perceive groups as geometric objects to study their
properties relatively easily. We begin with introducing the notions of Cayley graphs, and
the action of groups on trees endowed with a path metric. By studying the action of
SL(2,Z) on the Farey tree, we show that for m ≥ 3, a level m congruence subgroup of
SL(2,Z) is free. Further, we show that if a group acts freely and transitively on the edges
of a tree, then it is isomorphic to the free product of the stabilizers of the vertices under
the action. Finally, applying this result to the action of PSL(2,Z) on
∼
the Farey tree, we prove that PSL(2, Z) = Z2 ∗ Z3.

We will go on to define the geometric realizations of Cayley graphs, and look at how
Cayley graphs for the same group with two different generating sets are equivalent to
each other through the notion of quasi-isometry. Further, we discuss the word problem
for a group and how Dehn functions can be used to measure the complexity of its
solvability. We will then try to validate the importance of Dehn functions, as a measure
for the complexity of the word problem.
Finally, we will discuss hyperbolic groups and the word problem for these groups.
 LIST OF SYMBOLS OR
ABBREVIATIONS

τ (G, S) the Cayley graph of a group G w.r.t. a generating set S.

Fn the free group of n letters.
SL(2, Z) the special linear group of 2 × 2 matrices.
SL(2, Z[m]) the level m congruence subgroup of SL(2, Z).
PSL(2, Z) the quotient group of SL(2, Z) w.r.t subgroup generated by −I.
M∗ N the free product of two groups M and N .
⟨ S | R⟩ the presentation of a group with generating set S
and defining relations R.
π1(G) the fundamental group of G.
e(τ ) the number of ends of a graph τ .
e(G) the number of ends of a group G.
dim X the dimension of a metric space X.
asdim(X) the asymptotic dimension of a metric space X.
asdim(G) the asymptotic dimension of a group G.
diam(X) the max distance between two points of a metric space X.
Contents

Academic Integrity and Copyright Disclaimer ii

Acknowledgment iii

Abstract iv

List of Symbols or Abbreviations v

1 Introduction 1

2 Group actions on trees 3

2.1 Group action on sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Cayley graphs and group actions on graphs . . . . . . . . . . . . . . . 3
2.3 Free groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Quasi-isometries 10
3.1 The need for defining Quasi-isometries .. . . . . . . . . . . . . . . . 10
3.2 Quasi-isometric embeddings and euivalences . . . . . . . . . . . . . . 12
3.2.1 Notion of quasi-isometries . . . . . . . . . . . . . . . . . . . . 12
3.2.2 Geometric realizations of Cayley graphs of a group .. . . . . 12

4 Word problem and its solvability 14

4.1 Dehn Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.1.1 The importance of the Dehn Function .. . . . . . . . . . . . 16
4.2 A semigroup with an unsolvable word problem . . . . . . . . . . . . . 17

5 Hyperbolic groups 19
5.1 δ–hyperbolicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.2 Hyperbolic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3 Surface groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.4 Word problem for hyperbolic groups .. . . . . . . . . . . . . . . . . 22
 vii

A Ends of groups 24
A.1 Number of ends of a group ........................................................................... 24
A.2 Ends of groups and quasi-isometries ............................................................. 25

B Asymptotic dimension 27
B.1 Topology and dimension ............................................................................... 28
B.2 Large-scale dimension ................................................................................... 28
B.2.1 Changing the scale ............................................................................ 28
B.2.2 Asymptotic dimension of metric spaces and groups .......................... 28
Chapter1

Introduction

The emergence of geometric group theory as a distinct area of mathematics is usually

traced to the late 1980s and early 1990s. The 1987 monograph of Mikhail Gromov titled
“Hyperbolic groups” [1] introduced the notion of a hyperbolic group, which captures the
idea of a finitely generated group having large-scale negative curva- ture, and by his
subsequent monograph “Asymptotic Invariants of Infinite Groups” [2], that outlined
Gromov’s program of understanding discrete groups up to quasi- isometry. The work of
Gromov had a transformative effect on the study of discrete groups and the phrase
“geometric group theory” started appearing soon afterward [3].
An important aspect of mathematics consists of the study of symmetries of an
object, whether the object is a simple 3-dimensional object seen in daily life such as a
cube or a complicated abstract object such as a group. Geometric group theory tries to
study every group as a group of symmetries of some object so that we can infer some of
its properties which are relatively harder to study with respect to the abstract group
structure [4]. This idea is illustrated in Figure 1.1.

(a) The group Z/nZ as rotations of a regular n-gon.

(b) The group Z2 as translations in R2.

Figure 1.1: Representation of two groups as groups of symmetries of mathematical

objects.
 2

The object that captures the symmetries of a group is called a Cayley graph asso- ciated
with the group.
Using the notion of group actions on spaces, geometric group theory tries to
understand the properties of that group by studying the geometric properties of its
associated Cayley graphs. That is why it is important for us to study how a group acts on
graphs and trees. We will study this notion in the next Chapter. Later, we will study the
notions of quasi-isometries, word problems, ends of groups and asymptotic dimensions
to work with such groups and graphs. These chapters in the thesis are mostly based on
[4] “Office hours with a geometric group theorist.”
Chapter2

Group actions on trees

2.1 Group action on sets

We start by introducing the action of a group on a set.

Definition 2.1. An action of a group G on a set X is a function G × X → X

where the image of (g, x) is written g · x and where

(i) 1 · x = x for all x ∈ X and

(ii) g · (h · x) = (gh) · x for all g, h ∈ G and x ∈ X.

If the group SX is the group of symmetries of X, thought of as a set. An action of G on

X is the same thing as a homomorphism G → SX . So an action of G on X is the formal
way to realize G as a group of symmetries of the set X.
We need the notions of graphs which can be naturally perceived as metric spaces that
helps us understand the properties of groups via their actions on them.

Definition 2.2. A graph G is a pair (V, E) where:

(i) V /= φ is a set of vertices and

(ii) every e ∈ E joins a pair of (not necessarily distinct) v1, v2 ∈ V .

A type of graph which encodes the information about a group is called a Cayley graph.
Before diving into the notion of Cayley graphs, we first understand the action of a group
on a graph.

2.2 Cayley graphs and group actions on graphs

Definition 2.3. Let G be an arbitrary group and let S be a generating set for
G. The Cayley graph for G with respect to S is a directed, labeled graph
τ (G, S) := (V, E) where, V = G, and E = {(g, gs) : g ∈ G and s ∈ S}.
 4

Theorem 2.4. Let S be a generating set for G. Then the map f : G → Aut(τ (G, S))
such that f (g) = fg maps s ∈ G to gs ∈ G is an isomorphism.

Proof. The map f is a homomorphism. It is injective as ∀ g ∈ G and 1 ∈ V of

τ , fg(1) = g. Therefore g1(1) g2(1) ∀ g1 g2. Let φ ∈ Aut(τ ), that is, φ is an
isomorphism τ → τ . Suppose that φ(1) = g. We will show that φ = φg.
We use induction on word length in G with respect to S. The base case is word length
0, which is just the statement φ(1) = g = φg(1). Assume φ agrees with φg on all elements
of G of word length n with respect to S. Suppose v ∈ G has word length n + 1. This means
that v = ws, where the word length of w is n and s ∈ S ∪ S− 1 . For simplicity, we assume
that s ∈ S, as the other case is similar. By assumption, φ(w) = φg(w). There is a unique
edge labeled s from w, namely, the edge from w to ws = v. Similarly, there is a unique
edge labeled s from φ(w), with ending point φ(w)s. Since φ and φg both respect edge
labels, it must be that φ(v)

= φg(v) = φ(w)s. In particular, φ(v) = φg(v).

We define a group action more specifically on graphs in the following manner.

Definition 2.5. An action of a group G on a graph (V, E) is a homomorphism

G → Aut((V, E)) with the following properties.

(i) Any g ∈ G acting on v ∈ V takes it to some g · v ∈ V ;

(ii) Any g ∈ G acting on e ∈ E takes it to some g · e ∈ E;

(iii) For any x ∈ V or x ∈ E, we have 1 · x = x;

(iv) For g, h ∈ G and x ∈ V or x ∈ E, g.(h.x) = (g.h).x;

(v) If e ∈ E connects v, w ∈ V then g · e connects g · v and g · w.

Example 2.5.1. Z/3Z acts on a regular 3–valent tree T3 such that image of 1 ∈ Z/3Z is
the identity map, the image of 2, 3 ∈ Z/3Z maps each vertex to its two descendants.

2.3 Free groups

Definition 2.6. A free group Fn of ‘nJ letters is ⟨ a1, . . . , an⟩ with no defining relations.
The set {a1, . . . , an} is said to be the generating set for the free group.

Definition 2.7. An action of G on a set X is said to be free if ∀ g ∈ G and ∀ x ∈ X, if g · x

= x then g = 1.
 5

Figure 2.1: Cayley graph for a free group of two letters.

Theorem 2.8. If a group G acts freely on a tree, then G is a free group.

Before trying to prove this theorem, we need to define the tiling of a tree.
Definition 2.9. A tile is a subtree T0 of the barycentric subdivision T J of T (the barycentric
subdivision of a graph is the graph obtained by subdividing each edge; that is, we place
a new vertex at the center of each edge of the original graph). A tiling of T is a collection
of tiles with the following properties:

(i) No two tiles share an edge, so two tiles can only intersect at one vertex.

(ii) The union of the tiles is the entire tree T J.

Proof. The key to the proof is to obtain a tiling of our tree T . For each g ∈ G, let Tg be
the subtree of the barycentric subdivision whose vertex set is the set of vertices w of T J
so that d(w, gv) ≤ d(w, gJv) for all gJ ∈ G and whose edge set is the set of edges e of T J
so that both vertices of e lie in Tg.

We claim that the following set is a generating set for our group G

S = {g ∈ G|(gT0) ∩ T0 = φ}.

We need to show that our set S is a symmetric generating set for G. Let s ∈ S. This
means that (sT0) ∩ T0 = w for some vertex w of T J . Applying s− 1 we can
 6

conclude that T0 ∩ s− 1T0 = {s− 1(w)}. This means that s− 1 ∈ S, as desired.

Now, we need to show that S generates G. Let g ∈ G. We want to write g as a product
of elements of S. We look at the vertex gv. We can draw the unique path from gv back
to v and keep track of the tiles encountered along this path:
Tgn, Tgn− 1, . . . , Tg1, Tg0, where gn = g and g0 = e. If a path travels through tiles
Tgi+1 and Tgi without traveling through any tiles in between, then Tgi+1 ∩ Tgi must be
nonempty. Applying g− 1 , we see that (g− 1Tgi+1) ∩ (g− 1Tgi) = Tgi− 1gi+1 ∩ T0 =/
i i i
φ
.
But this means exactly that every g− 1gi i+1 is in S, and g can be written as a product
of elements in S. Thus, S is a symmetric generating set of G.
Since there is a unique (non-backtracking) path from gv to v in T ∀ g = s1s2 . . . sk we
can argue that that the unique path from gv to v completely determines the word g, ∀ g
∈ G.

We will apply this theorem for the action of a congruence subgroup of SL(2, Z) on a
Farey tree to see how a free action on a tree implies that the group is free.
Definition 2.10. We say that (m, n) ∈ Z2 is primitive if gcd(|m|, |n|) = 1. We define an
equivalence relation ∼ on the primitive elements of Z2 by declaring (m, n) to be
equivalent to −(m, n) := (−m, −n). We represent their equivalence class by
±(m, n).
Definition 2.11. The Farey graph is the graph whose vertex set is the set
{±(m, n) ∈ Z2 | ± (m, n) is primitive}. We denote a vertex ±(m, n) by (m, n) in Figure 2.2.
Two vertices ±(p, q) and ±(r, s) are connected by an edge if ps−qr = ±1. The Farey graph
thus obtained can be seen in Figure 2.2.

Figure 2.2: The Farey graph.

 7

Now, we can obtain the Farey tree by the following steps.

(i) Mark the centre of each edge of the Farey graph as a vertex for the Farey tree.

(ii) Mark the centroid of each triangle formed by the edges of the Farey graph as a
vertex for the Farey tree.

(iii) An endpoint function for the Farey tree connects the triangle vertices to the
corresponding edge vertices if and only if that edge is a side of that triangle.

Figure 2.3: The Farey tree.

Definition 2.12. The level m congruence subgroup of SL(2,Z) denoted by SL(2, Z)[m] is
the kernel of the homomorphism φ : SL(2, Z) → SL(2, Z/mZ) which reduces the entries
of all the matrices in SL(2,Z) modulo m.

Corollary 2.13. For m ≥ 3, the group SL(2, Z)[m] is isomorphic to a free group.

Proof. Since the action of SL(2, Z) on the Farey complex cannot interchange an edge
and a triangle, it means the action cannot interchange the two endpoints of the same
 8

edge. In other words, SL(2, Z) acts on the Farey tree without inversions. It remains to
understand the stabilizer in SL(2, Z) of each vertex of the Farey tree. Let us first consider
stabilizers of vertices corresponding to edges of connecting the vertices
±(1, 0) and ±(0, 1). For an element of SL(2, Z) to stabilize v, it simply must preserve
the set S = {±(1, 0), ±(0, 1)}. As the columns of a matrix are just the images of the
standard basis vectors under the action of that matrix, the columns of our stabilizers
must lie in S. That gives exactly 6 matrices to think about. But we cannot choose a vector
and its negative, for then the determinant will be 0. Therefore, we have the following 4
matrices as candidates.
! ! ! !
1 0 , 0 1 , −1 0 , 0 −1 .
0 1 −1 0 0 −1 1 0

These are the elements of the cyclic group generated by the second matrix on the list.
So the stabilizer of v is a cyclic group of order 4. The only matrices that lie in SL(2, Z)[m]
for m ≥ 2 are identity and its negative. However, only the identity lies in SL(2, Z)[m] for
m ≥ 3. But, we have proven that the required condition holds true only for one vertex.
But, as the action of SL(2, Z)[m] is transitive on Farey tree, the stabilizer of any vertex
can be calculated as follows,

Stab(Mv) = M Stab(v)M − 1,

Therefore the stabilizer for all the vertices will be identity, and hence the action is free.
Using Theorem 2.8, SL(2, Z)[m] is a free group.

Theorem 2.14. Suppose that a group G acts without inversions on a tree T in such a way
that G acts freely and transitively on edges. Choose one edge e of T and say
that the stabilizers of its vertices are H1 and H2. Then G ∼= H1 ∗ H2.

Proof. Since G acts without inversions, we avoid the barycentric subdivision, so the
definition of the tiling is the same as before except that the tiles are subgraphs of T itself.
A path in T from e to ge will give us a unique alternating word in the elements of H1 and
K1 , and an alternating word in the elements of H1 and K1 gives a unique path in T from
e to gv. Since there is only one path from e to ge, it follows that the product is a free
product.
∼
Corollary 2.15. Z/2Z ∗ Z/3Z = PSL(2, Z)

Proof. The action of SL(2, Z) on the Farey tree can be restricted to the action of PSL(2,
Z) since (m, n) and −(m, n) represents the same vertex. This means that PSL(2, Z) also
acts without any inversions and transitively on the Farey tree.
We, name the vertex corresponding to {±(1, 0), ±(0, 1)} as v0 and the vertex
corresponding to {±(1, 0), ±(0, 1), ±(1, 1)} as w0. Now, applying Theorem 2.14, we
 9

know that PSL(2, Z) is the free product of the stabilizers of v0 and w0. SL(2, Z), the
stabilizers were isomorphic to Z/4Z and Z/6Z. The negative of the identity corresponds
to 2 and 3 in these two groups. Thus the images of these stabilizers in PSL(2, Z) are
isomorphic to Z/2Z and Z/3Z. Therefore,

∼
PSL(2, Z) = Z/2Z ∗ Z/3Z.

Corollary 2.16. Every finitely generated subgroup of a free group is free.

Proof. As a consequence of Theorem 2.8, if we have a subgroup H of a free group G, then

H acts freely on a tree as well. We can take any free action of G on a tree and restrict
the action to H. Again using Theorem 2.8, H is also a free group.
Chapter 3

Quasi-isometries

One of the problems we encounter when using Cayley graphs as a geometric repre-
sentation of a group is that there can be different Cayley graphs for the same group with
respect to different generating sets. These graphs are not isometric to each other. So,
we need to define another form of equivalence called quasi-isometries or “coarse
isometries”.

3.1 The need for defining Quasi-isometries

Definition 3.1. The geometric realization of a Cayley graph is defined as follows

(i) Each edge in the graph is associated to an inteval [0, 1]

(ii) All points on the edges are now a part of the metric space.

(iii) Distance between two points on the same edge is the corresponding on the real
line w.r.t Euclidean metric.

(iv) Distance between two points not on the same edge can be calculated by sum-
ming up the smallest distance between the endpoints of the two concerned edges
and the distance of the points from those two endpoints.

Definition 3.2. If (X, dX) and (Y, dY ) are metric spaces, a function f : X → Y is called an
isometric embedding if f preserves distances, that is, for all
x1, x2 ∈ X,
dY (f (x1), f (x2)) = dX(x1, x2).

An isometric embedding f is called an isometry if it is also surjective.

Example 3.2.1. A map from f : Z → R2 such that f (x) = (x, 0) is an isometric embedding,
but not an isometry.
 11

Definition 3.3. If (X, dX) and (Y, dY ) are metric spaces, a function f : X → Y is called a bi-
Lipschitz embedding if f preserves distances by allowing distances to be stretched and
compressed by bounded amounts, that is, ∃ K > 0 such that ∀ x1, x2 ∈ X,

1
KdX(x1, x2) ≤ dY (f (x1), f (x2)) ≤ KdX(x1, x2).
A bi-Lipschitz embedding f is a bi-Lipschitz equivalence if it is also surjective.

Example 3.3.1. A map from f : Z → R2 defined by f (x) = (3x, 0) is a bi-Lipschitz

embedding, but not a bi-Lipschitz equivalence.

Theorem 3.4. Let G be a finitely generated group and let S and S J be two finite generating
sets for G. Then the identity map f : G → G is a bi-Lipschitz equivalence from the metric
space (G, dS) to the metric space (G, dS′ ).

Proof. Since S is finite, we can define a constant M ≥ 1 by

M = max{dS′ (1, s)|s ∈ S ∪ S− 1}.

Now consider any g ∈ G. Suppose g has word length n with respect to S, so we can write
g = s1s2 . . . sn,

where each si is in S ∪ S− 1 . Using the triangle inequality, we get:

dJS (1, g) = dS′ (1, s1s2 . . . sn)

≤ dS′ (1, s1) + dS′ (s1, s1s2 . . . sn)
≤ dS′ (1, s1) + dS′ (s1, s1s2) + dS′ (s1s2, s1s2 . . . sn)
≤ dS′ (1, s1) + dS′ (s1, s1s2) + dS′ (s1s2, s1s2s3) + . . . dS′ (s1s2 . . . sn− 1, s1s2 . . . sn).

Since the action of a group element preserves the distances between the elements of
the same group with respect to word metric we can rewrite this as

dS′ (1, g) ≤ dS′ (1, s1) + dS′ (1, s2) + · · · + dS′ (1, sn)
≤M+M+···+M
= Mn.

But n is the word length of g with respect to S, that is, dS(1, g) = n. Therefore, we have
shown that ∀ g ∈ G, dS′ (1, g) ≤ MdS(1, g). We can just reverse the roles of S and SJ to
find another bound N and the larger of two values can be assigned to K.
 12

We have shown that the Cayley graphs of a group with respect to two different
generating sets will be equivalent to each other under the bi-Lipschitz equivalence.

Example 3.4.1. Consider the Cayley graphs of R2 and Z2 with respect to the generating
set {(1, 0), (0, 1)} with the natural Taxicab metric for their geometric realizations. We
can take a function f : Z2 → R2 such that each point is marked onto itself, that is, f (x, y)
= (x, y). The distances are preserved by Taxicab metric and we get an isometric
embedding.But this map is not surjective. Hence this is not an isometry. Intuitively there
should be some equivalence relation between the two metric spaces which make Z2 ∈
R2 different from Z ∈ R2

We have a similar problem with regards to the Cayley graphs and their geometric realizations
not being equivalent to each other in any sense.

3.2 Quasi-isometric embeddings and euivalences

3.2.1 Notion of quasi-isometries

Definition 3.5. Let (X, dX) and (Y, dY ) be metric spaces. A function f : X → Y is called a
quasi-isometric embedding if there are constants K ≥ 1 and C ≥ 0 so that

1
dX(x1, x2) − C ≤ dY (f (x1), f (x2)) ≤ KdX(x1, x2) + C.
K
A quasi-isometric embedding f : X → Y is called a quasi-isometric equivalence, or just a
quasi-isometry, if there is a constant D > 0 so that for every point y ∈ Y , there is an x ∈
X so that
dY (f (x), y) ≤ D.
Example 3.5.1. A map from f : R2 → Z2 such that f (x, y) = ([x♩ , [y♩ ) is a quasi-isometry,
but it is neither an isometry or bi-Lipschitz equivalence.

3.2.2 Geometric realizations of Cayley graphs of a group

Theorem 3.6. Let G be a finitely generated group and let S and S J be two finite generating
sets for G. Then the geometric realization of the Cayley graph τ (G, S) is quasi-isometric
to the geometric realization of the Cayley graph τ (G, SJ)
Proof. First, there is a quasi-isometry from the geometric realization of any graph to its
set of vertices (with the path metric) obtained by sending every point on an edge to the
nearest vertex. In Theorem 3.4, we showed that there is bi-Lipschitz equivalence
between two Cayley graphs of the same group. Therefore, we can obtain a quasi-
isometry τ (G, S) → τ (G, SJ) as a composition of three quasi-isometries.
 13

Since we have proven that all Cayley graphs and their geometric realizations for a
given group are equivalent to each other, we can now talk about the geometric
representation of a group being equivalent to some other group without being con-
cerned about the choice of generating set influencing our geometric image of that
group. In the next chapter, we will try to solve the other problem we encounter while
perceiving groups as geometric objects.
Chapter 4

Word problem and its solvability

When we have a finitely presented group, we face the problem of being able to tell if
two words w1, and w2 represent the same group element. This can be equivalently
stated as the problem of being able to tell whether or not w1− 1w2 is the identity
element. This complexity is captured by the notion of a word problem of a group.

Definition 4.1. A group G = ⟨ X | R⟩ is a finitely presented if it is finitely generated, i.e., |X|

< ∞ and has a finite number of relations defined on them, i.e.
|R| < ∞.

Definition 4.2. The word problem for a finitely generated and presented group is the
problem of determining whether or not a given word in the group represents identity.
A word problem is called solvable in finite number of steps if we are able to
devise an algorithm to find whether or not a word of certain length represents identity.

The complexity of word problem for a group presentation is governed by the Dehn
function. The faster the Dehn function grows, the greater the number of times relations
must be used to reduce the problem.

4.1 Dehn Function

A word w on S ∪ S− 1 represents the identity in G when it can be converted to the empty
word via

• a finite sequence of free reductions (aa− 1 → 1)

• free expansions (1 → aa− 1 for some a ∈ S ∪ S− 1)
• applications of defining relators (elements of R) or their cyclic permutations
 15

Such a sequence is called a null-sequence for w. The number of applications of defining

relators or its permutations is the length of that null sequence.
Counting the number of such moves it can take for the word of a given length to be
reduced to identity gives us a measure of the difficulty in working with the particular
presentation for a group. The Dehn function estimates this measure thereby measuring
the complexity of the solution to the word problem.

Definition 4.3. The Dehn function f : N → N maps n to the minimal number N such that,
if wi is a word of length ≤ n that represents the identity, li is the length of minimal null-
sequence for wi, then N = max{li}.
Since there are only finitely many words of length at most n (S is finite) the Dehn
function is well–defined.

Example 4.3.1. Consider the group Z/mZ = ⟨ a | am⟩ . The Dehn function f (n) is given by
[m ♩
n

Example 4.3.2. The Dehn function f (n) of Z × Z = ⟨ a, b | a− 1b− 1ab⟩ satisfies

n2 f
(n) ≤ 16 .

This bound is realized by the words a− kb− kakbk.

Proof. To see this, first note that a− kb− kakbk is the hardest word to reduce of a given
length n since all pairs of aa− 1 and bb− 1 are as far apart as possible. The permutations of
defining relators tell us that the multiplication on a, b, a− 1, b− 1 is commutative. If we
apply b− 1a = ab− 1 in the middle of the word, we will get a− kb− k+1ab− 1ak− 1bk.
We can continue this operation on the first a to reduce the first pair of a− 1a to
identity. This will take k applications of the operation b− 1a = ab− 1. Since the number of
such a is k, the total number of moves required will be k2. As k = n,
4
where n is the word length, we have

n2 f
(n) ≤ 16 .

Lemma 4.4. Any finitely generated abelian group has a quadratically growing Dehn
function.

Proof. If there are h elements in the generating set, then a− kb− k · h− kakbk . . . hk will be
the hardest word to reduce. It will take us (h − 1)k operations for each h to reach
 16

h− 1 and so, total operations only for h will be (h − 1)k2. Repeating the process for each
element of the generating set, we have

f (n) ≤ (h − 1)k2 + (h − 2)k2 + · · · + k2.

Since, k = n , 2h

n 2 n 2 n2
f (n) ≤ (h − 1)( 2h ) + (h − 2)( 2h ) + · · · + (2h ) .

Example 4.4.1. Consider Z2 = ⟨ a, b | ab = ba⟩ which has a quadratically growing Dehn

function. To tell whether or not a word on {a±1, b±1} represents the identity, it is enough
just to add up the exponents of the a±1 and b±1 present and check whether both are zero.
Therefore, we can solve the word problem in linear time, which shows that Dehn
function is not a good measure of the difficulty of the word problem. It’s more of a worst
case scenario.

4.1.1 The importance of the Dehn Function

Since the Dehn function measures the complexity of the solution to the word problem
for a group, it gives us an upper bound of the time required to solve the word problem.

Definition 4.5. A recursive function f : M → N is a function defined on a discrete, well

ordered domain with a least element that calls itself, that is, f (n) is defined in terms of f
(m), where m < n and m, n ∈ M .

Theorem 4.6. For a finitely presented group ⟨ S | R⟩ of a group with Dehn function
f : N → N, the following are equivalent.

(i) There is an algorithm that, given the input of a word on S ±1 , will compute whether
or not that word represents the identity (i.e., the presentation has solvable word
problem).

(ii) There is a recursive function g : N → N such that f (n) ≤ g(n) , ∀ n.

(iii) f itself is a recursive function.

Proof. (ii =⇒ i) Given an upper bound g(n) on the Dehn function, it is always possible to
reduce a word of length w that represents the identity to the empty word using a null-
sequence with at most g(n) applications of defining relators. So, if g(n) is recursive and f
(n) ≤ g(n), we can reduce the problem for words of length n to
 17

words of length < n using the recursion on g(n). Thus, we have g(n) = k(g(n − 1)) and f (n
−1) ≤ g(n −1). We can repeat this process until the word length reduces to a very small
number and the word problem becomes trivial. Thus, we have devised an algorithm to
solve the word problem given a recursive upper bound for Dehn function.

(i =⇒ iii) If, on the other hand, we have an algorithm that solves the word problem,
then we can calculate f (n) by the following procedure.

• First list all words on S±1 of length at most n

• Discard from the list all that fail to represent the identity

• For each word w that remains, calculate the minimal number of application of
defining relator moves necessary to reduce w to identity.

Since we have devised using an algorithm to solve the problem for words of length
n, we can simply use some free reductions to write the Dehn function f (n) as a function
of some f (n − k). Thus the Dehn function is recursive if we have an algorithm and the
word problem is solvable.
(iii =⇒ i) If f (n) is a recursive function, we can simply define

g(n) = f (n) + 1,

and we will get a recursive upper bound for f (n). Thus, we have shown that

ii =⇒ i =⇒ iii =⇒ ii.

Hence, our proof for the theorem is complete.

4.2 A semigroup with an unsolvable word prob- lem

It is not common to encounter a group with an unsolvable word problem. Alan Turing
[5] studied the importance of semigroups with unsolvable word problems in
cryptography and cryptanalysis. We conclude this chapter with an example of one such
semigroup.

Example 4.6.1. One of the simplest example of a semigroup with an unsolvable word
problem is G = ⟨ a, b, c, d, e|ac = ca, ad = da, bc = cb, bd = db, ce = eca, de = edb, cdca =
cdcae, caaa = aaa, daaa = aaa⟩ given by C´etjin [5]. Collins [5] came up with another
presentation for this semigroup.
 18

Recently, Wang [6] reduced the number of generators to two. But, there were 27
defining relators for that presentation the smallest of which had a word length of about
100.
Chapter 5 Hyperbolic

groups

Curvature is a fundamental way of understanding the intrinsic geometry of mani- folds.

There are three curvatures on 2-dimensional manifolds, namely zero, positive, and
negative. The most trivial examples for these three are the plane, the sphere, and the
saddle.

Figure 5.1: 2-dimensional manifolds with zero, positive and negative curvatures.

A nice space that exhibits saddle point behavior at each of its points is the hyperbolic
plane H2. This space can be interpreted by multiple models isometric to each other. To
understand one of them, we consider the M¨obius transformation

z → i−z
i+z
This transformation takes the upper half-plane U to the open unit disk D ∈ C, and it
takes the real line to the unit circle. Since we have already identified U with the
hyperbolic plane, we now have an identification of the hyperbolic plane with D. We refer
to D as the Poincar´e disk. If we denote the hyperbolic metric on D by du2 , it
turns out that du2 = 4 dE2 2 2 , where r denotes the distance from the origin
(1− r )
and dE2
is the Euclidean metric. It is known that surfaces with constant negative Gaussian
curvature admit a metric locally modeled on H2. But to study such groups, we want to
capture this behavior in a discrete model, which earns them the name “hyperbolic
groups.”
 20

5.1 δ–hyperbolicity
In the Euclidean metric, we define the incircle of a triangle to be the largest in- scribed
circle. The points of tangency are called inpoints, and they cut the three sides of the
triangle into six pieces that come in length-matched pairs. Now in a more general space,
we cannot necessarily find an inscribed circle in any nice way, but we can generalize the
other property. Let the inpoints of a geodesic triangle be the uniquely determined three
points that divide the sides into pairs of equal lengths, as shown in Figure 5.2.

Figure 5.2: Inpoints of a geodesic triangle.

They are uniquely determined because we are just solving the system a = r + s, b = s
+ t, c = r + t, and the triangle inequality guarantees a solution. If we consider our space
to be a tree, any ‘triangle’ has actually the same inpoints. So the ‘insize’ of our ‘triangle’
is zero.

Definition 5.1. We will call a metric space δ−hyperbolic if all geodesic triangles have
insize ≤ δ, where δ ∈ R, and δ ≥ 0.

Example 5.1.1. A tree is a 0−hyperbolic space because if we try to take any geodesic
triangle in a tree, the insize would be zero as the tree has no closed loops and thus no
triangles.

This definition works fine on geodesic spaces. For a general space, we say a space is
δ−hyperbolic if all four-tuples for any four points x, y, z, w in that space satisfy

(x∆y)w ≥ min{(y∆z)w, (x∆z)w} − δ,

where (x.yw) is the shortest distance between the line joining x, y and w.

A crucial property is that δ−hyperbolicity is stable under quasi-isometry but, it does

not preserve the constant δ. So a quasi isometry on one δ−hyperbolic space can take it
to some other δJ−hyperbolic space.
 21

5.2 Hyperbolic groups

Definition 5.2. A finitely generated group is called hyperbolic if any of its Cayley graphs
(for a finite generating set) is δ−hyperbolic.

Example 5.2.1. Since trees are always 0−hyperbolic, natural examples of hyper- bolic
groups are free groups Fn of any rank, as their Cayley graphs with respect to any finite
generating set S with |S| = n are just (2n)−regular trees.

Example 5.2.2. As proven in Chapter 2, PSL(2, Z) = ⟨ v, w | v2w3⟩ With this pre-

sentation, the Cayley graph looks just like a tree of triangles as in Figure 5.3, a graph that
clearly has the δ–hyperbolic property with δ = 1.

Figure 5.3: Cayley graph of PSL(2, Z).

Since PSL(2, Z) is a quotient group derived from SL(2, Z) and this induces a quasi-
isometry between the two, this implies that SL(2, Z) should also be a hyperbolic group.

5.3 Surface groups

Surface groups are fundamental groups of closed hyperbolic surfaces. Each manifold S
has an associated group π1(S), called the fundamental group, which can be said to be
group-theoretically encoding information about the topology of S. The idea started as
an attempt by Poincar´e [8] to classify manifolds by associating a group to them
which could be distinguished from other groups relatively easily.
 22

Definition 5.3. The fundamental group of a topological space X with some chosen base
point x0 , denoted by π1(X, x0), is the group of homotopy classes of loops, which are
closed paths on X starting and ending at x0 .

Example 5.3.1. A tree has no loops, so its fundamental group is trivial.

Example 5.3.2. All loops on a circle from a point to itself are just complete rota- tions
around the circle, so the fundamental group will be isomorphic to Z.

Example 5.3.3. Let Σg denote the closed orientable surface of genus g. π1(Σ2) can be
generated by the four loops in Figure 5.4.

Figure 5.4: Generating set for the fundamental group of Σ2 Its

fundamental group can be written as

π1(Σ2) ∼= ⟨ a, b, c, d | [a, b][c, d]⟩ .

In general, for a surface of genus g ≥ 1,

π1(Σg) ∼= ⟨ a1, b1, a2, b2 . . . ag, bg | [a1, b1][a2, b2] . . . [ag, bg]⟩ .

Since these surfaces are closed hyperbolic surfaces for g ≥ 2, their groups are surface
groups. Surface groups are important examples of hyperbolic groups them- selves and
their construction can lead us to an important result about word problem for hyperbolic
groups.

5.4 Word problem for hyperbolic groups

Definition 5.4. A presentation G = ⟨ a1, ..., an | r1, ..., rm⟩ for a group G is called a
Dehn presentation of G if the following conditions hold true:

(i) There is a set of strings u1, v1, ..., um, vm and each relator ri is of the form
ri = uiiv− 1. (Relator ri encodes the equivalence in the group of ui and vi.)
 23

(ii) For each i, the word length of vi is shorter than the word length of ui .

(iii) For any nonempty string w in S = {ai} that represents the identity element, if w
has been reduced by canceling all occurrences aia− 1 , then at least one of
i
ui or u− 1i must appear as a substring.

Theorem 5.5. Hyperbolic groups admit Dehn presentations.

Proof. Fix any K > 8δ and consider a Cayley graph for G with respect to a (finite)
generating set S = {ai | i ∈ N} for G. Now consider the list of all reduced words ti with
word length at most K. Now, we can check which of the ti represent the same word by
just following them in the graph. Let ui be the non-geodesic words from that list, and for
each ui , let vi be some geodesic word from that list reaching the same point in the graph,
so it is guaranteed to be strictly shorter. Then put R = {ri = uiv− 1 | i ∈ N}. We claim G =
⟨ S | R⟩ is a Dehn presentation. Conditions
i
(i) − (ii) of Definition 5.4 are satisfied by construction. For Condition (iii), we
can rule out 8δ−local geodesic loops of length at least 8δ. So any long loop has a non-
geodesic subsegment of length at most K, which is one of ui. This completes the proof.

Corollary 5.6. Hyperbolic groups have solvable word problems.

Proof. By construction of Dehn presentations, they have a solvable word problem, and
by Theorem 5.5 all hyperbolic groups have a solvable word problem.
Appendix A Ends of

groups

When we study the properties a real-valued function on R, we often ask ourselves, what
is happening to the function at infinity. It helps us predict properties such as if the
function is going to be negligibly small when our argument if large enough even if it may
never actually be zero. Here, we will study what happens to a finitely generated group
G at infinity, using its Cayley graph τ (G, S) with respect to a generating set S as its
geometric representation.

A.1 Number of ends of a group

Definition A.1. Let τ be a connected, locally finite (having a finite number of edges for
each vertex), infinite graph. Let C be a subgraph of τ . We define

||τ \C||

to be the number of disjoint, connected and unbounded components we get if we

remove C from τ .

Example A.1.1. If we remove any vertex from a free group of n letters, we will get 2n
components.

Definition A.2. For a locally finite graph τ , the number of ends of the graph
is
e(τ ) = sup{||τ \C|| | C is a finite subgraph of τ \C}

Definition A.3. The number of ends of a finitely generated group G is equal to the
number of ends of a Cayley graph τ (G, S) where S is a finite generating set for G

For this definition to be consistent, we need that the number of ends of G should not
depend on the choice of S.
 25

Theorem A.4. The number of ends of a finitely generated group is independent of the
choice of the finite generating set.

Before trying to prove this theorem, we look at how the number of ends of groups can
be related for groups that are quasi-isometric to each other.

A.2 Ends of groups and quasi-isometries

Theorem A.5. If two finitely generated groups are quasi-isometric, then they have the
same number of ends.

It can be clearly seen that this statement is stronger than the previous theorem. To prove
these statements; we will require the following result from Freudenthal and Hopf [4].

Theorem A.6. If G is a finitely generated group, then G has zero, one, two, or infinitely
many ends.

Proof. To prove this statement, all we need two show that if ∃ C ⊂ τ such that
||τ \C|| ≥ 3, then ∃ D ⊂ τ such that ||τ \D|| = ∞. (See Figure A.1)

Figure A.1: Construction of a group with more than 2 ends.

Let τ be a Cayley graph for a group G and let C0 be a finite subset of τ such that
||τ \C0|| = 3. Then there is then an element g in the group so that C1 = g · C0 is disjoint
from C0 . Then if D1 = C0 ∪ C1 , it follows that ||τ \D1|| > 3. But the same argument can be
applied inductively, and we can have a finite d∞ ⊂ τ such that
||τ \D∞|| = ∞.

Now, we can provide the proof of Theorem A.4 [7].

 26

Proof of Theorem A.4. Let S1 and S2 be two finite generating sets for a group G, and let
τ (G, S1) and τ (G, S2) be the corresponding Cayley graphs. Then the two Cayley graphs
must have a bi-Lipschitz eqivalence as proved earlier. Now, if we have a finite subgraph
C1 ⊂ S1, then the bi-Lipschitz function will map it to some finite subgraph C2 ⊂ S2, such
that ||τ (G, S1)\C1|| = ||τ (G, S2)\C2||. Hence, we are done.

This proves that the number of ends is the property of a group and not of its graphs.
The same logic can be applied while proving that two quasi-isometric groups have the
same number of ends, as all their Cayley graphs will have a quasi-isometric equivalence
relation between them. Now, we move onto some examples.

Example A.6.1. The number of ends of a Z × Z is 1. We can have two connected

components using a finite subgraph of its standard Cayley graph, but only one of them
will be connected.

Example A.6.2. Any free group of n letters(n ≥ 2) has infinitely many ends, as we can
remove just the identity vertex from the Cayley graph and get more than 4 ends. The
result then follows from Theorem A.6.

Corollary A.7. Suppose that G is a finitely generated group with a finite index subgroup
N. Then e(G) = e(N ).

Proof. The proof follows from Theorem A.5 and the fact that if G is quasi-isometric to
some group P , then it must have a finite index subgroup isometric to P and vice versa.

This concludes our study for ends of groups. In the final Chapter, we look at the
topological notion of dimensions for finitely generated groups we have been working
with.
Appendix B Asymptotic

dimension

One of the most basic notions we study in topology is dimension. We think of the
building blocks of topology, such as points, lines and planes as being 0–dimensional 1–
dimensional and 2–dimensional respectively. Then our notion expands to the amount of
information required to represent a space. For instance, R2 is two di- mensional as it
requires two parameters to be presented, whether it is the standard coordinate system
or the polar coordinate system.
But, while defining the notion of dimension of groups, we cannot take the exact same
approach, as the groups we work with are often discrete objects having multiple defining
relations between their elements. So, once again we use the Cayley graphs as a metric
space representation of our groups and use the topological notion of dimensions to
construct the theory.
We can begin to think about the dimensions of groups with one of the simplest
examples, the group of integers Z. We can guess, that it should be 1–dimensional as it
sits in R. Similarly we can define the dimension for a free abelian group of two letters Z2
to be 2. (We also saw how Z ∈ R2 is different from Z2 ∈ R2 during our study of quasi-
isometries.) Similarly, the dimension of Zn can be defined as n. Now, if we talk about
the nonabelian free group of two letters F2, we can see that it does not follow a metric
similar to the euclidean metric in R2. Thus, it appears that it might have a dimension
different than 2, which seems non–intuitive. Also, if we look at nonstandard Cayley
graphs for Z, it appears that it no longer sits inside R.
So, if we are going to use Cayley graphs as a representation of our groups, we must make
sure that the defined notion of dimensions for our groups is invariant under quasi-
isometries.
 28

B.1 Topology and dimension

Definition B.1. We say that the dimension of a metric space X does not exceed n and
write dim X ≤ n if for every open cover U of X there is a refinement V with order at most
n + 1.

We will first characterize what it means to be zero dimensional and then use zero
dimensional sets to compute higher dimensions.

Lemma B.2. Suppose X is a separable metric space that is nonempty. Then dim X = 0 if
and only if for all p ∈ X and every open set U containing p, there is an open set V ⊂ U so
that X − V is also open.

Example B.2.1. By definition, Z, Q and Z2 are 0 dimensional.

This seems counter-intuitive until we recall that all these spaces are actually count- able
unions of discrete points, each having 0 dimension. This poses the challenge to
appreciate the fact that upon a change of perspective, if we look from far enough, Z
seems like R and not like R2. This notion rather covers the dimensions on a smaller scale.
We need a notion for large scale geometry which can look over the small nuances of
these spaces and differentiate between the dimensions of Z and Z2.

B.2 Large-scale dimension

Definition B.3. Let r > 0. A metric space X is said to have dimension 0 at scale
r if it can be expressed as a union X = ∪Xi where:

1. sup{diam(Xi)} < ∞

2. inf{d(x, xJ)|x ∈ Xi, xJ ∈ Xj} > r whenever i /= j.

B.2.1 Changing the scale

Considering Z as a metric space with the Euclidean metric, we see that Z has dimension
0 at every scale r < 1. However, if r ≥ 1, then Z does not have dimension 0 at scale r. This
agrees with our intuition: on small scales, Z looks like a discrete collection of points and
hence is 0−dimensional, but on large scales, Z no longer looks 0−dimensional.

B.2.2 Asymptotic dimension of metric spaces and groups

Definition B.4. We say that the asymptotic dimension of a metric space X
does not exceed n, and write asdim(X) ≤ n if for each r > 0, there exist subsets
 29

X0, X1, . . . , Xn with X = X0 ∪ X1 ∪ · · · ∪ Xn, and for each i, Xi has dimension 0 at scale r.

Definition B.5. We say that asymptotic dimension of a metric space X is n written

asdim(X) = n if and only if asdim(X) ≤ n and there is no integer q < n such that as asdim(X)
≤ q, i.e., it exceeds every integer less than n.

Since this definition is constructed in such a manner that the asymptotic dimension is
preserved under quasi-isometries, we can define:

Definition B.6. The asymptotic dimension of a group G is the asymptotic dimension of

one of its Cayley graphs constructed with respect to a generating set of G.

Example B.6.1. The asymptotic dimension of Zn which does not exceed n, but exceeds n
− 1 and hence asdim(Z) = n.

Example B.6.2. Consider the free group of n letters Fn. The Cayley graph is a tree T = (V,
E) in which each vertex is incident to four edges. We will show that any infinite tree has
asymptotic dimension 1. Since T contains an infinite geodesic asdim(T ) ≥ 1. Thus, it
remains to show that asdim(T ) ≤ 1. To prove this, for each r we must find two families
of uniformly bounded, r–disjoint sets whose union covers T . Let r > 0 be given. Fix some
vertex of the tree and call it x0 . We will use the notation |x| to denote the distance d(x,
x0) from x to the fixed vertex x0 . As a first step in the construction of the cover, for each
positive integer n, let

An = {v ∈ V | 2r(n − 1) ≤ |v| ≤ 2rn}.

Although the collections A0 = [ A2n+1 and Ae = [ A2n each consist of r–disjoint
subsets, they are not uniformly bounded. So, we subdivide them further. Clearly,
n∈N n∈N

the equivalence classes of Ao and Ae are uniformly bounded and r–disjoint. Hence
asdim(Fn) = 1.
We conclude this thesis with an illustration of the asymptotic dimension of F2
as shown in Figure B.1.
 30

Figure B.1: asdim(F2) = 1.

Bibliography

[1] Mikhail Gromov “Hyperbolic Groups” in “Essays in Group Theory” (Steve M.

Gersten, ed.), MSRI Publ. 8, 1987

[2] Mikhail Gromov “Asymptotic invariants of infinite groups” Vol. 2 (Sussex, 1991),
London Mathematical Society Lecture Note Series, 182, Cambridge University Press

[3] Graham A. Niblo and Martin A. Roller ”Geometric group theory. Vol. 1.” Math-
ematical Society Lecture Note Series, 181. Cambridge University Press, Cam- bridge,
1993

[4] Matt Clay and Dan Margalit “Office Hours with a Geometric Group Theorist”
Princeton University Press, 2017

[5] Donald J. Collins “A Simple Presentation of a Group with Unsolvable Word Problem”
Illinois Journal of Mathematics, 1986

[6] Xiaofeng Wang, Chen Xu, Guo Li, and Hanling Lin “Groups With Two Gen- erators
Having Unsolvable Word Problem And Presentations of Mihailova Sub- groups”
School of Mathematics, Shenzhen University, 2016

[7] John Meier “Groups, Graphs and Trees: An Introduction to the Geometry of Infinite
Groups” Cambridge University Press, 2008

[8] Henri Poincar´e “Papers on Topology Analysis Situs and Its Five Supplements
(Translations by John Stillwell)” Edinburgh University Press, 2009