AN INTRODUCTION TO RELATIVELY

HYPERBOLIC GROUPS
A thesis

submitted by

Murphy A. Page

in partial fulfillment of the requirements

for the degree of

Master of Science

in

Mathematics

TUFTS UNIVERSITY

August 2017

© Copyright 2017 by Murphy A. Page

Adviser: Genevieve Walsh

 ii

Abstract

We look at a few of the various ways in which relatively hyperbolic groups are

defined. It has been shown that, if a little care taken with hypotheses, all definitions

are equivalent; here we examine the relationship between the definitions using the

coned-off Cayley graph and fine δ-hyperbolic graphs. We also go over the definition

using geometrically finite convergence groups. Finally, we use the free group F2 as

an example of a relatively hyperbolic group to explore the different definitions.

 iii

To my bicycle, Bruce, for bringing me everywhere I needed to be and never bringing

me down.
 iv

Acknowledgements

I am eternally grateful to the professors that helped get me through this graduate

school experience in (more or less) one piece. Thanks to the graduate director Kim

Ruane, for always making everything go; to Moon Duchin, for being a stellar mentor;

and to my advisor Genevieve Walsh, for convincing me to make good decisions for

myself and for keeping me going through the last stretch of this.
 v

Contents

List of Figures vi

1 Introduction 2

1.1 History and Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Overview of Hyperbolic Groups 4

2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 δ-hyperbolic Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.2.1 The Boundary of a Hyperbolic Space . . . . . . . . . . . . . . . 7

2.3 Hyperbolic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Relatively Hyperbolic Groups 11

3.1 The Coned Off Cayley Graph . . . . . . . . . . . . . . . . . . . . . . . . 11

3.2 Fine Hyperbolic Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.3 Geometrically Finite Convergence Groups . . . . . . . . . . . . . . . . . 17

4 Examples of Relatively Hyperbolic Groups 20

4.1 F2 as a Geometrically Finite Convergence Group . . . . . . . . . . . . . 20

4.2 The Coned-off Cayley Graph of (F2 , ⟨a⟩) . . . . . . . . . . . . . . . . . . 22

Bibliography 23
 vi

List of Figures

2.1 The Cayley graph for the free group on two generators . . . . . . . . 9

2.2 The two-holed torus arises as a quotient of H2 by “folding” the hy-

perbolic octagon. Image made with Curt McMullen’s LIM (http://

www.math.harvard.edu/\protect\unhbox\voidb@x\penalty\@M\{}ctm/

math101/www/programs/index.html . . . . . . . . . . . . . . . . . . . 10

3.1 The coned off Cayley graph of Z ⊕ Z relative to Z. . . . . . . . . . . . 13

3.2 Two paths in Γ.

̂ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4.1 The Cayley graph of F2 (see 2.1) overlayed on H2 , tiled with a

fundamental domain for the action of F2 on the plane. Image gener-

ated with Curt McMullen’s LIM (http://www.math.harvard.edu/

\protect\unhbox\voidb@x\penalty\@M\{}ctm/math101/www/programs/

index.html . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
 1

An Introduction to Relatively Hyperbolic Groups

 2

Chapter 1

Introduction

The study of relatively hyperbolic groups is relatively new in the field of geometric

group theory but has developed extensively over the past twenty years. There is an

abundance of definitions for the notion of “relatively hyperbolic,” which, although

more or less equivalent, can feel rather different from one another. As a result, the

literature on the topic is somewhat labythinthine. This thesis is intended to be an

approachable introduction to relatively hyperbolic groups, reviewing some of the

various definitions and looking at simple, specific examples.

1.1 History and Significance

Hyperbolic groups were introduced and explored in depth by Gromov in his seminal

1987 paper [7]. In it Gromov also describes the notion of relative hyperbolicity as a

generalization of hyperbolicity; this was specifically motivated by the fundamental

groups of finite volume, cusped manifolds of negative curvature. Unlike hyperbolic

groups, however, relatively hyperbolic groups were left largely untouched for about

a decade after this publication.

A precise timeline is difficult to establish, but 1997/98 saw a surge in papers

on the topic. Two of the main pioneers here are Bowditch, who developed a the-

ory grounded in Gromov’s original proposal in [2] (first draft in ‘97, not published

until ‘12), and Farb, who proposed a new definiton based on the construction of a

“coned-off Cayley graph” in [5]. There now exist around six definitions of relatively

hyperbolic groups which have been shown to be equivalent (see [8], in which Hruska

demonstrates the equivalence of definitions without requiring the groups to be finitely

generated). However, this thesis will focus on only a few of these formulations.

The subject of relatively hyperbolic groups continues to be fruitful, and much

 3
work has been done translating concepts from hyperbolic groups to their more gen-

eral, “relative” versions. For example, Dahmani constructs a relative Rips complex

in [3], and there are many papers looking at boundaries of relatively hyperbolic

groups. More recently, acylindrically hyperbolic groups have been a popular topic

in which relatively hyperbolic groups naturally arise as examples. Specifically, it is

known that non-virtually cyclic relatively hyperbolic groups are acylindrically hy-

perbolic [9].

1.2 Summary

In Chapter 2, we provide a brief overview of hyperbolic groups, going over relevant

definitions, properties, etc. Note that we assume the reader is familiar with funda-

mental ideas from geometric group theory and has a little experience with hyperbolic

spaces and groups.

Chapter 3 goes over various definitons of relatively hyperbolic groups. First we

look Farb’s construction using the coned-off Cayley graph (as an aside, this is the

author’s favorite definition). We then prove that when this graph is δ-hyperbolic,

the bounded coset penetration property is equivalent to fineness, which brings us to

a second definition. To close the chapter, we present geometrically finite convergence

groups as a dynamical formulation for relatively hyperbolic groups.

Finally, in Chapter 4 we look at F2 as a relatively hyperbolic group, both as the

fundamental group of a punctured torus with a cusp and using the Coned-off Cayley

graph.
 4

Chapter 2

Overview of Hyperbolic Groups

In this chapter, we quickly review some fundamental and relevant information about

δ-hyperbolic spaces and hyperbolic groups. Sources for this chapter are [1] and [6].

2.1 Preliminaries

Let X be a metric space with distance d ∶ X × X → [0, ∞). A path in X is a

continuous function from an interval to X, p ∶ [0, t] → X. A geodesic is a path whose

length realizes the distance between endpoints. Throughout this chapter we will

let X be a proper geodesic metric space, i.e. there exists a (possibly non-unique)

geodesic between any two points of X and every closed ball is compact.

Let λ ≥ 1, c ≥ 0 be constants. A (λ, c)-quasigeodesic between x0 , x1 ∈ X is a path

with p(0) = x0 , p(t) = x1 and

1
d(x0 , x1 ) − c ≤ d(x0 , x1 ) ≤ λd(x0 , x1 ) + c.
λ

Let (X, dX ), (Y, dY ) be proper geodesic metric spaces. An isometry from X to

Y is a map f ∶ X → Y that preserves distances, i.e. for all x, y ∈ X we have

dY (f (x), f (y)) = dX (x, y).

Given constants λ ≥ 1 and c ≥ 0, a (λ, c)-quasiisometry between X and Y is a map

g ∶→ Y such that

• for all x, y ∈ X we have 1

λ dX (x, y) − c ≤ dY (g(x), g(y)) ≤ λdX (x, y) + c, and

• for all y ∈ Y there exists x ∈ X such that dY (g(x), y) ≤ c.

The second condition here is known as coarse surjectivity.

 5
2.2 δ-hyperbolic Spaces

There are various ways of defining δ-hyperbolic spaces which make prescise the

heuristic notion that these spaces are “approximately trees.”

Definition 2.2.1 A geodesic metric space X is δ-hyperbolic (sometimes called “Gro-

mov hyperbolic”) if there is a constant δ > 0 such that, given any geodesic triangle

in X, one side of the triangle is contained in the union of the δ-neighborhoods of the

other two.

w y
x
z

x y z
(a) δ-thin triangle (b) 0-thin triangle

The second way we’ll look at defining δ-hyperbolic spaces uses the Gromov prod-

uct of two points.

Definition 2.2.2 Fix a basepoint z ∈ X. The Gromov product based at z of two

points x, y ∈ X is
1
(x∣y)z ∶= (d(x, z) + d(y, z) − d(x, y)) .
2

Keeping our heuristic in mind, given a triangle (or tripod) in a tree, the Gromov

product tells you the length of one of the legs; for example, in figure 2.1b we see that

(x∣y)z = d(w, z).

Definition 2.2.3 A geodesic metric space X is δ-hyperbolic if there is some δ > 0

such that for all x, y, z, w ∈ X we have

(x∣z)w ≥ min((x∣y)w , (y, z)w ) − δ.

 6
Note that if for some fixed basepoint w0 ∈ X, the above inequality holds for all

x, y, z ∈ X, then the space X is δ-hyperbolic.

The following properties of δ-hyperbolic spaces will be useful for us in the next

chapter:

Lemma 2.2.1 Let X be δ-hyperbolic and p, p′ be distinct geodesics in X beginning

and ending at the same points. Then p and p′ lie within a δ-neighborhood of each

other.

Proof: Two such geodesics form a bi-gon, which is geodesic triangle with one degen-

erate side. By the thin-triangles condition for δ-hyperbolic spaces, the bi-gon cannot

be fatter than δ. ◻

Extending this idea to quasi-geodesics requires an important theorem known as

the Morse lemma:

Theorem 2.2.1 Given λ ≥ 1, c ≥ 0 there is a constant r > 0 depending on δ, λ, c such

that any (λ, c)-quasigeodesic in X lies within an r neighborhood of some geodesic in

X.

By combining this theorem with lemma 2.2.1, we achieve the result that will be

used later.

Lemma 2.2.2 Let p, p′ be (λ, c)-quasigeodesics beginning and ending at the same

points. Then there is some constant R > 0 such that p and p′ lie within an R-

neighborhood of each other.

Another important result, particularly to the study of hyperbolic groups, is that

δ-hyperbolicity is a quasiisometry invariant:

Theorem 2.2.2 Let X be a δ-hyperbolic space, and suppose there is a quasiisometry

f ∶ X → Y where Y is some geodesic metric space. Then there exists δ ′ > 0 such that

Y is a δ ′ -hyperbolic space.
 7
2.2.1 The Boundary of a Hyperbolic Space

Fix a basepoint w ∈ X. We can define the Gromov boundary of a δ-hyperbolic

space using either geodesic rays or sequences that tend to infinity. Two rays r1 ∶

[0, ∞) → X, r2 ∶ [0, ∞) → X originating at w are equivalent if they remain within

some bounded distance of each other, and the boundary ∂ r X is the set of equivalence

classes of rays. A sequence {xi } in X tends to infinity if lim inf i,j→∞ (xi ∣xj )w =

∞. Two such sequences {xi }, {yj } are equivalent if lim inf i,j→∞ (xi , yj )w = ∞. The

boundary ∂ s X is the set of equivalence classes of sequences tending to infinity.

The following proposition combines a number of well-known facts about these

boundaries, the proofs of which can all be found in [1].

Proposition 2.2.1 The definitions of ∂r X and ∂s X are equivalent, allowing us to

simply refer to ∂X. Additionally,

• each definition is independent of the choice of basepoint,

• there are topologies for ∂r X and ∂s X which are homeomorphic, and

• ∂X is compact and metrizable.

A relevant example of the boundary of a hyperbolic space is the hyperbolic plane

H2 , whose boundary is the circle S 1 . This is easily seen by looking at the Poincare

disc model and thinking about ∂ r X: no two rays originating from the center of

the disc will ever remain within bounded distance from each other, so none will be

equivalent to another.

2.3 Hyperbolic Groups

Let G be a group acting on a space X. The action is by isometries if every g ∈ G

acts as an isometry of X. It is a cocompact action if the quotient space X/G is

compact. When there are no accumulation points of the action, it is called properly

discontinuous; that is, for all K ⊂ X compact, there are only finitely many g ∈ G

such that gK ∩ K =/ ∅.
 8
Definition 2.3.1 The group G acts on X geometrically if the action is properly

discontinuous, cocompact, and by isometries.

Definition 2.3.2 A group G is hyperbolic if there exists some δ-hyperbolic space

on which G acts geometrically.

A useful fact is that any group acts geometrically on its Cayley graph; thus, if a

group G has a presentation such that the corresponding Cayley graph is δ-hyperbolic,

G is hyperbolic. Indeed, it is known that the Cayley graphs for any presentations of

G are quasiisometric, so by theorem 2.2.2, if any Cayley graph is δ-hyperbolic, every

Cayey graph is.

The free group on two generators is a wonderful example of a hyperbolic group, as

we can see its hyperbolicity in various ways. Its Cayley graph is a tree and therefore

0-hyperbolic (see figure 2.1).

Now F2 is also the fundamental group of a punctured torus, so we can see its hy-

perbolicity through a geometric action on a closed, convex subset S of the hyperbolic

plane H2 (which is quasi-isometric to a tree). By choosing the appropriate Moebius

transformations as the two generators for the group, specifically ensuring that their

commutator is a hyperbolic transformation, we can obtain a compact punctured

torus as the quotient of the subset S by F2 . We can similiary see that the funda-

mental group of the two-holed torus is hyperbolic; figure 2.2 shows a fundamental

domain for the action on H2 .

 9

Figure 2.1: The Cayley graph for the free group on two generators
 10

Figure 2.2: The two-holed torus arises as a quotient of H2 by “folding” the hyperbolic
octagon. Image made with Curt McMullen’s LIM (http://www.math.harvard.edu/
~ctm/math101/www/programs/index.html
 11

Chapter 3

Relatively Hyperbolic Groups

We will now look at some of the ways to define relatively hyperbolic groups. A group

is hyperbolic when it acts geometrically on some δ-hyperbolic space, so there are

various ways in which a group can fail to be hyperbolic: it could act geometrically

only on spaces which are not δ-hyperbolic (such as Z ⊕ Z) or it could fail to act

geometrically on a δ-hyperbolic space in some way (such as the fundamental group

of a cusped space, where the action fails to be cocompact). The hope of relatively

hyperbolicity is to find the groups who miss the hyperbolic mark one way or another,

but not by too much. When we think of a cusped space, for example, there is a sense

that the cusp (which prevents hyperbolicity) can be easily controlled. This chapter

will make precise the notion of being “not too far off” from hyperbolic.

3.1 The Coned Off Cayley Graph

In this section we will let G = ⟨S∣R⟩ be a group with and Γ be its Cayley graph. For

simplicity, we will conflate words in G and their corresponding paths in Γ, and we

will use the word metric dS to measure distances in G and Γ.

Definition 3.1.1 For a H < G, we construct the coned off Cayley graph Γ(G,
̂ H)

from Γ by adding a cone vertex v(gH) for each distinct coset gH and an edge from

a vertex g0 ∈ Γ to v(gH) if g0 ∈ gH. To make Γ(G,

̂ H) a metric space, we declare

that each new edge to a cone vertex has length 1/2, and when there is no ambiguity

around the subgroup, we will simply use the notation Γ.

̂

(Note that we may replace H with a finite collection of subgroups {Hi } with no com-

plications, but for simplicity of notation in this section, we will only be considering

one subgroup.)

We associate a path w in Γ to a path ŵ in Γ

̂ as follows: For every maximal subpath
 12
z = z0 z1 ⋯zn−1 zn of w such that n > 0 and every vertex of z belongs to the same coset

gH, replace z1 ⋯zn−1 with v(gH) using the edges of length 1/2, (z0 , v(gH)) and

(v(gH), zn ). In the case that n = 1, replace the edge from the Cayley graph (z0 , z1 )

with the two edges of length 1/2 passing through v(gH).

A path w whose associated path ŵ contains such a subpath ẑ = z0 ⋅ v(gH) ⋅ zn is

said to penetrate the coset gH with entering vertex z0 and exiting vertex zn . A path

w is without backtracking if it never penetrates the same coset more than once. We

call a path w a relative (quasi)geodesic if ŵ is a (quasi)geodesic in Γ.

̂

Definition 3.1.2 The coned off Cayley graph Γ

̂ has the Bounded Coset Penetration

(BCP) property if for every λ ≥ 1 there is a constant c(λ) > 0 such that whenever

w and w′ are (λ, 0)-quasigeodesics without backtracking such that their initial end-

points are w− = w−′ and their terminal endpoints are w+ , w+′ with dS (w+ , w+′ ) ≤ 1,

then the following hold:

1. If w penetrates a coset gH but w′ does not, then the entering and exiting

vertices of w in gH are at most c far from each other.

2. If w and w′ both penetrate a coset gH, then their entering vertices are at most

c far from each other, and similarly their exiting vertices are at most c far.

We are now ready to give the definiton of relative hyperbolicity attributed to Farb.

Definition 3.1.3 A group G is hyperbolic relative to the subgroup H if the coned off

Cayley graph Γ(G,

̂ H) is δ-hyperbolic and has the BCP property. If Γ
̂ is δ-hyperbolic

without the BCP property, we say that G is weakly relatively hyperbolic.

In the original formulation, weakly relatively hyperbolic was called relatively

hyperbolic while the added condition of the BCP property made a group strongly

relatively hyperbolic. However, the coned off Cayley graph can be δ-hyperbolic even

in situations that feel very far from hyperbolic, and eventually the terminology was

reformulated. One example where we see this is Z ⊕ Z = ⟨a, b ∣ [a, b] ⟩, which is weakly

hyperbolic relative to Z =< b >. The coned off Cayley graph is pictured in Figure
 13
??, and it is simple to verify that it is indeed δ-hyperbolic. To see that it does not

satisfy the BCP property, consider the paths bn and abn for increasingly large n.

Figure 3.1: The coned off Cayley graph of Z ⊕ Z relative to Z.

3.2 Fine Hyperbolic Graphs

When Γ̂ is δ-hyperbolic, an equivalent property to the BCP property that we may

look for instead is fineness.

Definition 3.2.1 A graph is fine if each of its edges is contained in only finitely

many circuits of length n for every n > 1.

This property in turn is equivalent to the graph being angularly locally finite:

Definition 3.2.2 Given two edges e1 = (v, v1 ), e2 = (v, v2 ) sharing one vertex, the

angle θ between e1 and e2 is the length of the shortest path between v1 and v2 that

does not include v. We define θ as being infinite if no such path exists.

A graph is angularly locally finite if for all angles θ, given an edge e, the set of

edges which make an angle of θ with e is finite.

The following lemmas, along with the second half of the proof of the main propo-

sition, come from the appendix of [4].

 14
Lemma 3.2.1 A graph is fine if and only if it is angularly locally finite.

Proof: Suppose that a graph G is not angularly locally finite. Then there is some

θ > 0 and some edge e in G such that there are infinitely many edges making angles

of θ with e. Then e is contained in infinitely many circuits of length θ + 2. Also note

that for two edges e1 , e2 making angles of θ with e, the resulting circuit for e1 does

not contain e2 and vice versa. This is by the definiton of a circuit as a simple loop

and because e, e1 , and e2 all have exactly one vertex in common. Thus there are

infinitely many distinct such circuits, so G is not fine.

On the other hand, suppose that G is angularly locally finite. Let n > 0 and e be

an edge of G. Any circuit of length n that contains e gives an edge e′ adjacent to e

making an angle of n − 2 with it. Thus there are only finitely many possibilities, so

G is fine. ◻

Note that in the definitions of fine and angularly locally finite, we may replace “loops

of length n” (or “angles of θ”) with “loops of lenth < n” (angles < θ).

Lemma 3.2.2 If the coned-off Cayley graph Γ

̂ satisfies the first criterion of BCP

then, up to a change in constant, it satisfies the second criterion as well.

Proof: Suppose the first criterion holds for λ ≥ 1 with constant r. Let w1 , w2 be two

(λ, 0)-quasigeodesic paths without backtracking in Γ

̂ hat penetrate the coset gH.

Take w1′ to be the path obtained by cutting off w1 after its entering vertex of gH.

This is also a (λ, 0)-quasigeodesic.

Now take w2′ to be the path obtained first by cutting off w2 after the vertex

v(gH) and then adding the edge from v(gH) to the last vertex of w1′ . This is a

(λ′ , 0)-quasigeodesic for some λ′ ≥ λ . So w1′ must also be a (λ′ , 0)-quasigeodesic.

Thus, w1′ and w2′ meet the conditions of the first criterion, with w2′ penetrating gH

and w1′ not.

So there is some constant c(λ′ ) such that the entering and exiting vertices of

w2′ are at most c far from each other. But the entering vertex of w2′ is the entering

vertex of w2 , and the exiting vertex of w2′ is the entering vertex of w1 , so we have
 15
shown that the entering vertices of w1 and w2 are at most c far from each other.

Since λ′ depends only on λ and is independent of choice of paths (we would use the

exact same construction) then the constant c holds across all paths that meet the

conditions of the second criterion. ◻

Proposition 3.2.1 A δ-hyperbolic coned off Cayley graph has the BCP property if

and only if it is fine.

Proof: (⇒ by contrapositive) Suppose that Γ

̂ is not fine. Then there is some edge e

contained in infinitely many circuits of length n for some n. Since the only infinite-

valence vertices are the cone vertices, then infinitely many of these circuits must

penetrate some coset gH. From these circuits we can choose finite subpaths to

obtain infinitely many paths wi each of which have e as their first edge (thus all

beginning at the same point) and penetrate the coset gH. Because they come from

circuits, each path will be without backtracking.

We must show that for every c > 0 we can find two paths wi , wj such that (WLOG)

their entering vertices are more than c far away from each other. Suppose not, i.e.

the entering vertices of the wi are all within some bounded distance from one another

in the Cayley graph Γ. Since there must be infinitely many distinct entering vertices

to give infinitely many paths penetrating gH, this leads to a bounded neighborhood

in Γ containing infinitely many vertices, contradicting finite generation. Thus, Γ

̂

cannot of the BCP property.

(⇐ By contradiction) Suppose that Γ

̂ is fine but does not have the BCP property

and let λ ≥ 1. Then for all c > 0 there are (λ, 0)-quasigeodesic paths w1 and w2 in Γ
̂

such that

1. the paths start at the same point,

2. the paths end at most distance one from each other,

3. w1 penetrates a coset gH with entering and exiting vertices further than c

apart in Γ, and

4. w2 does not penetrate the coset.

 16
Since Γ
̂ is δ-hyperbolic, by lemma 2.2.2, there is some constant R(δ, λ) such that

any two paths satisfying 1 and 2 above stay at most R far from each other.

Now for any θ > 0 there is a constant r(θ) such that the following holds: when

g, g ′ ∈ H and dS (g, g ′ ) > r, the angle between the edges (v(H), g) and (v(H), g ′ ) in
̂ is greater than θ. Since for a fixed g ∈ H there are infinitely many g ′ ∈ H that
Γ

will be further than r from it, if this were not true it would lead to infinitely many

edges (v(H), g ′ ) adjacent to the edge (v(H), g) making an angle of less than θ with

it, contradicting lemma 3.2.1.

So now we fix θ to be 64 ⋅ λ ⋅ R(δ, λ) + 1 and choose w1 and w2 that satisfy the

four conditions above with constant c = r(θ). Since w1 penetrates gH for more that

r, the path passes through the vertex v(gH) (while w2 does not) and the angle of

the adjacent edges at this vertex is greater than θ. Our goal is to contradict that

this angle is > θ.

First we parametrize the two paths by arc length to have w1 ∶ [t−1 , t1 ] → Γ

̂ such

that w1 (0) = v(gH) and w2 ∶ [0, t2 ] → Γ.

̂ Let t = min(t1 , 10 ⋅ λ ⋅ R), so w1 (t) occurs

after the exiting vertex of w1 from gH, and let σ1 = w1 ∣[0,t] .

Now choose t′ so that d(w1 (t), w2 (t′ )) (the distance in Γ)

̂ is minimized, and thus

less than R. Let σ2 be a geodesic from w1 (t) to w2 (t′ ). Let t′′ = min(t2 , t′ + 20 ⋅ λ ⋅ R)

and let σ3 = w2 ∣[t′ ,t′′ ] . Next, choose t′′′ so that d(w2 (t′′ ), w1 (t′′′ )) is minimized

(again, less than R), and let σ4 be a geodesic from w2 (t′′ ) to w1 (t′′′ ). And lastly,

let σ5 = w1 ∣[t′′′ ,0] so that concatenating the σi in order gives us a loop. (See figure

3.2 for reference.)

w1 (0) = v(gH)

w1 (t′′′ ) w1 (t)
⋯ ⋯
w2 (t2 ) = w1 (t−1 ) <R <R w1 (t1 ) = w2 (0)
⋯
w2 (t′′ ) w2 (t′ )

Figure 3.2: Two paths in Γ.

̂

Since σ3 lies entirely in w2 it does not contain the vertex v(gH). We can see
 17
that σ2 also does not contain v(gH): first suppose t = 10 ⋅ λ ⋅ R. Then w1 (t) is at

a distance of at least 10 ⋅ R from v(gH), and since σ2 is shorter than R, it cannot

contain v(gH). Now suppose t = t1 . Then since the end points of w1 and w2 are

at most one away from each other, σ2 has length one and can be chosen as an edge

from the Cayley graph Γ. Thus is does not contain v(gH).

Now we want to see that σ4 does not contain v(gH) and that w1 (t′′′ ) in fact

occurs before (or at) the entering vertex of w1 in gH. First suppose that t′′ =

t′ + 20 ⋅ λ ⋅ R. Then the distance between w2 (t′ ) and w2 (t′′ ) is at least 20 ⋅ R. But if

w1 (t′′′ ) lies on the λ-quasigeodesic σ1 , we could construct a path starting at w2 (t′′ )

to w1 (t′′′ ) then w1 (t) and finally w2 (t′ ) which is shorter than 2 ⋅ 10R. So w2 (t′′′ )

must occur before v(gH). If t′′ = t2 , then it works out similarly to the second case

above.

Thus we have a loop that passes through v exactly once, so the angle at v, which

is the distance between the entering and exiting vertices, both of which are included

in the loop, is less than the length of the loop. This length in turn is less than

2(10 ⋅ λ ⋅ R + R + 20 ⋅ λ ⋅ R + R) ≤ 2(10 ⋅ λ ⋅ R + λ ⋅ R + 20 ⋅ λ ⋅ R + λ ⋅ R) ≤ 64 ⋅ λ ⋅ R = θ.

Thus we have arrived at a contradiction to the angle θ being large. ◻

This proposition brings us around to another definition of relative hyperbolicity:

Definition 3.2.3 Let G be a group acting on a fine δ-hyperbolic graph with finite

edge stabilizers and finitely many orbits of edges. Then if P is a set of representatives

of conjugacy classes of infinite-valence vertex stabilizers, G is hyperbolic relative to

P.

So by its construction, when the coned-off Cayley graph is δ-hyperbolic and has

the BCP property, then it meets the conditions of this definition.

3.3 Geometrically Finite Convergence Groups

This section presents the dynamical formulation of a relatively hyperbolic group

from [2], which nicely generalizes the action of a Kleinian group on its limit set.
 18
Definition 3.3.1 The action of a group G on a compact, metrizable space M with

more than two points is a (non-elementary) convergence group action if the action

of G on the space of distinct triples of points of M is properly discontinuous. A

convergence action is uniform if the action of the distinct space of triples is also

co-compact.

We may have elementary convergence group actions on spaces with two or fewer

points as well: if M is empty, G must be finite, if M has one point, G must be

countable, and if M has two points, G must be virtually cyclic.

A subgroup P < G is a parabolic subgroup if it is infinite and contains no lox-

odromic element (i.e. an infinite-order element which fixes exactly two poitns).

The unique fixed point of a parabolic subgroup P is called a parabolic point, and a

parabolic point is bounded if StabG (p) acts cocompactly on M − {p}. A point α ∈ M

is a conical limit point if there is a sequence of group elements {gi } and distinct

points β0 , β1 such that gi α → β0 , and for all γ ∈ M − {α}, gi γ → β1 .

Definition 3.3.2 If P is a set of representatives of the conjugacy classes of maximal

parabolic subgroups in G, then the action of the pair (G, P) on M is geometrically

finite if it is a convergence action and every point of M is either a conical limit point

or a bounded parabolic point.

It is a theorem of Bowditch that G acts as a uniform convergence action on M

if and only if every point is a conical limit point, i.e. there are no parabolic points.

This is relevant to another theorem of his havin to do with hyperbolic groups, which

leads directly into our next characterization of relative hyperbolicity.

Theorem 3.3.1 If G acts as a uniform convergence group on a compact, metrizable

space M , then G is hyperbolic and M is homeomorphic to ∂G.

From this, we have the following result of Yaman in [11], which is a natural

generalization of hyperbolicity to relative hyperbolicity.

Definition 3.3.3 If G acts as a geometrically finite convergence group on a compact,

metrizable space M , then G is hyperbolic relative to P, the parabolic stabilizers.

 19
Now it is also known that if a group acts properly discontinuously on a proper

δ-hyperbolic space X, the induced action on ∂X is a convergence group action [10].

Since ∂X is compact and metrizable (as stated in proposition 2.2.1), this brings us

to the final definition of a relatively hyperbolic group:

Definition 3.3.4 Let G be a group that acts properly discontinuously on a proper δ-

hyperbolic space X such that the induced convergence action on ∂X is geometrically

finite. Then if P is a set of representatives of conjugacy classes of the maximal

parabolic subgroups, (G, P) is relatively hyperbolic.

This second definition is essentially a more specific version of the first, requiring

M to be the boundary of a proper δ-hyperbolic space, but it is nice as it gives us

the convergence action requirement for free.

 20

Chapter 4

Examples of Relatively Hyperbolic

Groups

In this chapter, we will take a look at the free group as a relatively hyperbolic

group in two different ways. Rather than focusing on rigorously proving relative

hyperbolicity, we aim more to develop some intuition around the ideas presented in

the previous chapter.

4.1 F2 as a Geometrically Finite Convergence Group

First we will look at F2 as the fundamental group of a cusped, punctured torus.

This arises as the quotient of H2 by the properly discontinuous action of F2 = ⟨a, b⟩,

where a and b are hyperbolic Moebius transformations (i.e. isometries that each fix

exactly two boundary points) that have been chosen so their commutator [a, b] is

parabolic.

The large shaded square in figure 4.1 is a fundamental domain for the action;

the generator a takes the left side to the right, while b takes the bottom side to the

top. So when the square gets folded up, its four corners become the puncture, and

we see this is a cusp by noting that these corners lay on the boundary at infinity.

The commutator corresponds to the loop of the puncture, and so the commutator

subgroup ⟨[a, b]⟩ corresponds to the cusp. We will see that F2 is hyperbolic relative to

P, the set of conjugates of the commutator subgroup, a maximal parabolic subgroup,

by observing that the action on ∂H2 = S 1 is geometrically finite.

The parabolic points are all the translates ot the four corners of the fundamental

domain, which appear to be accumulation points in the figure (and in fact, the set

of these points is dense in S 1 ). Heuristically, if we look at the checker-board-style

tiling of H′ by the fundamental domain, the parabolic points on the boundary are
2
 21

Figure 4.1: The Cayley graph of F2 (see 2.1) overlayed on H2 , tiled with a fundamen-
tal domain for the action of F2 on the plane. Image generated with Curt McMullen’s
LIM (http://www.math.harvard.edu/~ctm/math101/www/programs/index.html

those that come from rays that eventually “stay in the same color.” Note that for

any of these points p, the quotient S 1 − {p} by Stabp (G) is S 1 , and since this is

compact, each parabolic point is bounded (think about the bottom left corner of the

fundamental domain, which is stabilized by ⟨[a, b]⟩).

Now what’s left to see is that every other point on the boundary is a conical

limit point. Going back to our heuristic, these other points can be viewed as rays

that eventually pass through alternating colors in the checker board tiling. In our

figure, we can also see these points as the ones corresponding to the boundary of the

overlayed Cayley graph of F2 . We won’t examine this part in detail, but note that

the “north-south dynamics” of the hyperbolic isometries (e.g. the generators) show

us that the endpoints of the hyperbolic axes of these isometries are conical limit

points.
 22

4.2 The Coned-off Cayley Graph of (F2 , ⟨a⟩)

We’ll now look at ⟨a, b⟩ as being hyperbolic relative to the cyclic subgroup generated

by a (it would work the same if instead we chose b). We construct the coned-off

Cayley graph from figure 2.1 first by adding a vertex for every coset of ⟨a⟩ and then

connecting these to every element in the cosets. This is easily visualized by placing

a vertex “above” each of the horizonal axes in the graph, which correspond to group

elements eventually ending in repeating a’s and connecting it to each vertex of that

axis.

So we need to check to see that the newly constructed graph has the bounded

coset penetration property or that it is fine. In this case, the only circuits in the

graph are the ones we created with the cone vertices, and the symmetry of the

construction makes it very simple to check for fineness. In fact none of the original

edges from the Cayley graph are contained in more than one circuit of a given length.

Each of the new edges from the cone vertices is in only two circuits of a given length.

If we want to think about the bounded coset penetration property here, it is

useful to compare this coned-off Cayley graph with that of (Z ⊕ Z, Z), shown in

figure 3.1 which does not have the BCP property. With Z ⊕ Z, we’re “shortcutting”

across flat parts in one direction when we cone off the Cayley graph. But there

was flatness all over the graph, so while coning off created a δ-hyperbolic graph, the

absence of the BCP property tells us that the original graph was too flat to begin

with. In contrast, the flat copies of Z that we are coning off in the Cayley graph

of F2 are nicely separated from each other. This prevents us from having path that

begin and end together with one path penetrating a coset arbirarily deeply.
 23
Bibliography
[1] J. M. Alonso, T. Brady, D. Cooper, V. Ferlini, M. Lustig, M. Miha-
lik, M. Shapiro, and H. Short, Notes on word hyperbolic groups, in Group
theory from a geometrical viewpoint (Trieste, 1990), World Sci. Publ., River
Edge, NJ, 1991, pp. 3–63. Edited by Short.

[2] B. H. Bowditch, Relatively hyperbolic groups, Internat. J. Algebra Comput.,

22 (2012), pp. 1250016, 66.

[3] François Dahmani, Classifying spaces and boundaries for relatively hyperbolic
groups, vol. 86, 2003.

[4] , Les groupes relativement hyperboliques et leurs bords, 2003/13 (2003),

pp. viii+87. Thèse, l’Université Louis Pasteur (Strasbourg I), Strasbourg, 2003.

[5] B. Farb, Relatively hyperbolic groups, Geom. Funct. Anal., 8 (1998), pp. 810–
840.

[6] Étienne Ghys and Pierre de la Harpe, Espaces métriques hyperboliques,

in Sur les groupes hyperboliques d’après Mikhael Gromov (Bern, 1988), vol. 83
of Progr. Math., Birkhäuser Boston, Boston, MA, 1990, pp. 27–45.

[7] M. Gromov, Hyperbolic groups, in Essays in group theory, vol. 8 of Math. Sci.
Res. Inst. Publ., Springer, New York, 1987, pp. 75–263.

[8] G. Christopher Hruska, Relative hyperbolicity and relative quasiconvexity

for countable groups, Algebr. Geom. Topol., 10 (2010), pp. 1807–1856.

[9] D. Osin, Acylindrically hyperbolic groups, Trans. Amer. Math. Soc., 368 (2016),
pp. 851–888.

[10] Pekka Tukia, Convergence groups and Gromov’s metric hyperbolic spaces,
New Zealand J. Math., 23 (1994), pp. 157–187.

[11] Asli Yaman, A topological characterisation of relatively hyperbolic groups, J.

Reine Angew. Math., 566 (2004), pp. 41–89.