Covering Spaces and Graph Theory PDF

Covering Spaces and their Applications to Graph Theory
Ben Ashby 00638208,

Syafiq Johar 00608118,
Kevin Pan 00640443,
Claire Rebello 00639944,
Ryutaro Tanno 00640324,
Steven Yang 00643297
This is our own unaided work unless stated otherwise.
1
Contents
I Covering Spaces 4
1 Introductory Topological Concepts 4

1.1 Loops and Homotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 The Fundamental Group 5

2.1 Induced Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Covering Spaces 7
4 The Lifting Theorems 9

4.1 Homotopy Equivalence and the Homotopy Type of a Space . . . . . . . . 15
5 Equivalent Covering Spaces and Covering Transformations 16

5.1 Conjugate Subgroups of π1 . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2 Deck Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3 Criterion for Equivalence of Covers . . . . . . . . . . . . . . . . . . . . . . 18
6 The Universal Covering Space 19

6.1 Local Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2 Quotient Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.3 Existence of a Universal Cover . . . . . . . . . . . . . . . . . . . . . . . . 20
7 Summary of Classification of Covering Spaces 20
II Applications to Graph Theory 22
8 Some Basic Graph Theory 22

8.1 Essential Definitions and Concepts . . . . . . . . . . . . . . . . . . . . . . 22
8.2 Simplices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
9 Euler Characteristics 24
9.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
10 Results on Polygonal Representations 27

10.1 Connected Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
11 Orientable Surfaces 30
11.1 Examples of Orientable and Non-orientable Closed Surfaces . . . . . . . . 30
12 Classification of Surfaces 31
III A More Topological View of Graphs 34
13 Some Useful Theorems 34
14 The Fundamental Group of a Graph 35
2
15 Fundamental Groups and Free Groups 35
16 Covering Graphs 38
16.1 A Purely Algebraic Result . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
17 Folding: An Application 41
IV Summary 43
V Bibliography 44
3
Part I
Covering Spaces
Having studied some basic point-set topology, we now wish to branch out into algebraic
topology. As the name may well suggest, this area of mathematics aims to use algebra
as a tool for studying topology. In fact, algebraic structures turn out to be extremely
useful in pursuing this goal. We will begin by examining topological spaces and some of
the deep correspondences that exist between groups and topological spaces. The focus
then moves to applications of this theory: surfaces are studied and classified, graphs are
introduced, and finally the theory will be used to prove results in algebra that are difficult
to approach from a purely algebraic angle.
Before we can leap into the theory, a few basic definitions must be stated:
1 Introductory Topological Concepts

In this section we proceed largely as in Munkres’ book [18].
1.1 Loops and Homotopies

The first way that a mathematician learns to apply algebra to problems in topology
is to use paths and loops; we can learn a lot about a given topological space by looking
at the equivalence classes of loops that it is possible to have in that space. The algebra
comes in when we consider these equivalence classes as algebraic objects, in particular,
as elements of a group. It is probably intuitively very clear what we mean by a path, but
a formal definition is given below.
Definition 1.1 (Paths and loops). A path in a topological space X with endpoints x0 ,
x1 ∈ X is a continuous map γ : [0, 1] → X such that γ(0) = x0 and γ(1) = x1 . A loop in
X based at x0 is a path γ : [0, 1] → X such that γ(0) = γ(1) = x0 (i.e. a loop is a closed
path).
We were first introduced to the idea of a homotopy in the context of “equivalence” of

paths, but there is a more general definition:
Definition 1.2 (Homotopy). If f0 and f1 are continuous maps from space X to space
Y , we say that the two maps are homotopic if there is a continuous map F : X × I → Y
such that F (x, 0) = f0 (x) and F (x, 1) = f1 (x) for all x ∈ X, where I = [0, 1]. If f0 and
f1 are homotopic, we write f0 ' f1 .
Now we see that a path homotopy is just a special case of the more general idea above,
but is nonetheless a very useful notion.
Definition 1.3 (Path homotopy). If, this time, the two maps are paths in a space X we
call them path homotopic if there is a homotopy between them in the above sense, and
in addition, for any t ∈ I, F (0, t) = x0 and F (1, t) = x1 where x0 and x1 are respectively
the common beginning and endpoints of the two paths.
Both of these equivalences define equivalence relations. Note the slight difference
in the definitions; for a normal homotopy anything goes so long as the homotopy is a
4
continuous function. In the case of path homotopy, at every stage of the deformation of
the path (i.e. every intermediate path) the beginning and endpoints are the same, so of
course path homotopies preserve endpoints, meaning that any two homotopic paths must
have the same start and endpoints.
Path homotopies are also useful for the intuition they offer; we can imagine a path
as a broken elastic band that we regard as being able to stretch infinitely far. Our
definition of a path homotopy says that if we can deform our elastic band from one shape
to another, then the two paths are homotopic. These definitions make precise some of
the basic notions of what is sometimes known as “rubber sheet geometry”. We now have
a mathematically sound notion of continuous deformations of paths.
2 The Fundamental Group

An operation is defined on paths which we shall call a product. The product of two
paths f and g is denoted as f • g which we informally describe as “tagging one path onto
the end of the other”. More formally:
Definition 2.1 (Product of Paths). If f and g are both continuous functions from the
interval [0,1] to a space X (i.e., two paths in X) and we have f (1) = g(0), then we define
f • g(s) to be f (2s) for s ∈ [0, 12 ] and g(2s − 1) for s ∈ [ 21 ,1], which is continuous as long
as f (1) = g(0) .
This is the group operation of the fundamental group, which we define to be the set
of (homotopy) equivalence classes of loops (closed paths) starting and ending at a given
point x0 in a topological space. This is a group under the operation described above
(which, of course, is well defined, since loops at a given point all have the same beginning
and endpoints) [18, p.328]. We denote this π1 (X, x0 ), and a precise definition is given
below:
Definition 2.2 (Fundamental Group). The fundamental group of X with base point x0
is defined by:
π1 (X, x0 ) = {[f ] | f is a loop based at x0 }
with group operation •.
The fundamental group is a topological invariant, which means that if two topological
spaces are homeomorphic,then they have isomorphic fundamental groups. In fact, if f is
a homeomorphism between the two spaces, the induced homomorphism (defined below,
in Definition 2.4) f∗ is an isomorphism between the fundamental groups. This means that
the fundamental group is a useful tool for determining whether or not two spaces are not
homeomorphic. In particular, if the fundamental groups of two spaces are not isomorphic,
then the two spaces are not homeomorphic. Now we know about the fundamental group,
we can make the following definition:
Definition 2.3 (Simple Connectedness). A space is simply connected if it is path con-

nected, and the fundamental group is trivial for all base points in the space.
A worthy remark (see Munkres for a proof, although it should be intuitively clear)
is that for a path connected space, the fundamental groups based at any two points are
isomorphic, so in the above definition, we only really need to show that the fundamental
group at one base point is trivial, and the path connectedness assumption does the rest.
5
We must unfortunately note here that the fundamental group is not the whole story.
If we imagine that we have a topological space Y that is not path connected, then all
loops based at a point in a given path component will be able to tell us about that path
component only, since we can’t reach the others by moving along paths and loops. Thus,
the fundamental group can give us detailed information about the areas that our loops
can reach, but its usefulness will be limited to path connected spaces.
2.1 Induced Homomorphisms

Let X and Y be topological spaces. Suppose that h is a continuous map sending the
point x0 ∈ X to y0 ∈ Y . This will be denoted:
h : (X, x0 ) → (Y , y0 )
Now, continuous functions are the structure preserving maps in topology, and so we
might expect, since the fundamental group is a way of describing the structure of a
space, that there would be a relationship between the fundamental groups of two spaces
when one is mapped continuously onto the other. In fact, this continuous map induces
a homomorphism of the fundamental groups of the respective spaces, which we define as
follows:
Definition 2.4 (Induced Map/Homomorphism). Let h∗ ([f ]) = [h◦f ] for all elements [f ]
of π1 (X, x0 ) This is the homomorphism induced by h.
Remark 2.1. It is important to note the following:

(i) Every induced map is a homomorphism (which is clear from the definition). From here
on we shall stick to the term induced homomorphism.
(ii) Let f : Y → Z and g : X → Y be continuous maps, then (f ◦ g)∗ = f∗ ◦ g∗ . [20, p.127]
The induced homomorphism can be viewed as follows: we examine the effect of a

continuous map on loops in a space, and so the induced homomorphism sends a loop in
X̃ to a loop in X, that is, it maps one fundamental group into the other. The situation
is described in the following diagram:
X̃
X
γ̃ γ
p
Figure 1: Looking at the induced homomorphism.
So the term induced is appropriate; the deformation under a continuous function

causes the fundamental group, but since continuous functions preserve topological struc-
ture, the map between the groups is a homomorphism.
6
3 Covering Spaces
Our first big new idea, covering spaces are a class of objects associated with a topo-
logical space that can help us to study the space in more detail. As we will see, several
important properties of the space will be preserved in a covering space, such as classes of
paths and loops.
There are several different definitions of a covering space which arise due to the con-
ditions on the space X. Here we place no restrictions on the space X but require the
covering map p to be surjective.
Definition 3.1 (Covering Space). A covering space of a topological space X is a pair

(X̃, p) where X̃ is a space and p : X̃ → X is a surjective continuous map such that there
exists an open cover {Uα } of X such that, for each α, p−1 (Uα ) is a disjoint union of open
subsets in X̃, each of which is mapped homeomorphically onto Uα by p. We refer to p as
a covering map, although it is sometimes referred to as a projection.
So — covering maps are examples of local homeomorphisms. We can think of a

covering space as a space that divides up into disjoint copies of our base space X, where
by copies we mean spaces that are each homeomorphic to X. A particular example is
the universal cover (we shall justify saying ‘the’ universal cover later) which is a simply
connected covering space. The word universal is used since, as we will show, the universal
covering space of a space X is a covering space for all other connected covering spaces of
X.
Example 3.1 (The Covering Space of the Torus). It is a fact [18] that the torus is a
product of two circles. We may ask ourselves whether or not the covering space of the
R R
torus is × ? The answer to this question is yes; the covering map is fiddly to explicitly
R
define, but we shall describe it here. We can first divide the 2 into equal squares and
for each square, label the edges as in Figure 1, then we join edge b and edge d together
to get a cylinder (without its top and bottom) as in Figure 2. Next, we join a and c, as
in Figure 3, to get a torus. By doing this, we can show that each square covers the torus
and all squares agree at edges, because for each square, we join a with c and b with d.
Finally, the preimage of every open set with respect to the covering map to these squares
R
is a union of disjoint sets, and hence 2 is a covering space of the torus. (Note that by
‘join’ we mean ‘identify’ in the sense of quotient spaces, which are briefly discussed in the
section on the universal covering space).
d b
Figure 2: Subdividing the plane into squares.
7
a
b, d
b, d
a, c
Figure 3: Forming a cylinder. Figure 4: Creating a torus from a cylinder.
The following lemma is the first result in our general theory of covering spaces and
covering maps:
Lemma 3.1. Let (X̃, p) be a covering space. Then p is an open map, that is, p(V ) is
open in X whenever X̃ is open in X̃. Moreover, the covering map p is a homeomorphism
if and only if it is a bijection. [19, p.34]
Proof. For the first part, let V be an open subset of X̃. Pick an arbitrary point x ∈ p(V ).
It suffices to find a neighbourhood of x that is fully contained in p(V ). We have that
p(v) = x for some v ∈ V . Since p is a covering map there is a neighbourhood U of x
which is evenly covered by p, which means that p−1 (U ) is a disjoint union of open sets,
each of which is mapped homeomorphically on to U by p (this is just the definition of
a covering map). Clearly v must be in one of these sets, which we will call Ũ . Now let
N (x) = p(V ∩ Ũ ) so that x ∈ N (x) because v ∈ V and v ∈ Ũ . Now, since p is a covering
map, the restriction p |Ũ : Ũ → U is a homeomorphism, so that p−1 is well defined and
continuous on U . Now, we have the following equality:
N (x) = p(V ∩ Ũ ) = (p−1 )−1 (V ∩ Ũ )
and we note that V ∩ Ũ is open in Ũ . Hence, by the continuity of p−1 , N (x) is open, and
N (x) = p(V ∩ Ũ ) ⊆ p(V ) so we’ve found the desired neighbourhood and p(V ) is open,
as required.
The second part follows directly from the first since p(U ) = (p−1 )−1 (U ) is open from
the first part for any open set U ⊆ X. Hence, the inverse of p is continuous and hence p
is a homeomorphism.
The following definitions will be useful later, and also to prove the next theorem.
Definition 3.2 (Fiber). Let (X, p) be a covering space of a space X, and x0 ∈ X. The
set p−1 (x0 ) is called the fiber of x0 , and consists of all the points of the covering space
that map to x0 .
Definition 3.3 (Number of sheets). The cardinality of the set p−1 (x) for some x ∈ X is
sometimes called the number of sheets of a covering. The following theorem shows that
this value is well defined, that is, the same for any choice of x.
The aforementioned theorem concerns the uniformity of the cover; we will show that
the fiber of any two points of the base space have the same cardinality, and hence any
two points are “ covered” by a set of the same cardinality.
8
Theorem 3.1. If the space X is path connected, and p : X̃ → X is a covering map, then
for any x1 , x2 ∈ X, the sets p−1 (x1 ) and p−1 (x2 ) have the same cardinality. [5, p.125]
Proof. Choose arbitrarily any two points x0 , x1 ∈ X. We se up a bijection between the

sets p−1 (x0 ) and p−1 (x1 ). Making use of the path connectedness hypothesis, we choose
a path γ in X from X0 to x1 . For any point y ∈ p−1 (x0 ), by the path lifting theorem
we can lift γ to a path γ e in X̃. Moreover, this lift is unique for a given starting point
in p−1 (x0 ). If we let f (y) be the endpoint γ e(1), then f is an injective mapping from
−1 −1
p (x0 ) → p (x1 ) by the uniqueness described above. But, defining the analogous map
from p−1 (x1 ) → p−1 (x0 ) using instead the reverse path γ− we get the inverse map of the
map f defined above. We’ve set up a bijection between the sets p−1 (x0 ) and p−1 (x1 ), and
so by definition they have the same cardinality.
Additionally, since p∗ is a homomorphism, we have that the image p∗ (π1 (X, x0 )) is a

subgroup of π1 (Y, y0 ). The image is called the characteristic subgroup, or sometimes the
image subgroup, of the cover. This relation between a covering space of a topological space
and subgroups of the fundamental group of a space will be central to our classification of
the covers of a given space.
An important thing to note is that the elements in the image subgroup are homotopy
classes of loops in X that lift to loops in X̃.
4 The Lifting Theorems

The lifting properties are one of the main reasons that covering spaces are so useful.
The next few results show that there are many important types of function that can
be lifted to a covering space. When writing this section we referred to the books in
the bibliography by Munkres, Massey and Hatcher, respectively [18], [5] and [7], and
we combined these sources to present the proofs in an accessible way. The term ‘lift’ is
defined precisely below.
Definition 4.1 (Lifting of Functions). Let X, Y and Z be spaces. Suppose that g : Y → Z

and f : X → Z are continuous maps. If fe : X → Y is a continuous map with fe ◦ g = f ,
then we say that fe is a lifting, or lift, of f to Y .
Before we prove the big theorems, we shall need the help of a few lemmas. As men-
tioned above, these are a hint at the usefulness of covering spaces, and the functions
alluded to are usually homotopies, loops and paths. There is quite a lot of theory here.
The aim is to prove that we can lift uniquely paths and homotopies to covering spaces,
and ultimately to determine precisely when maps can be lifted to covering spaces.
Theorem 4.1 (Unique Lifting Property). Let (X̃, p) be a covering space of a topological
space X and let Z be a connected topological space. Suppose that g : Z → X̃ and h :
Z → X̃ are both continuous maps. Then, if p ◦ g = p ◦ h and there is a point z ∈ Z at
which g and h agree, then g = h.
Proof. Let Z0 = {z ∈ Z : g(z) = h(z)} . If we could prove Z0 is clopen (both open and
closed), then connectedness of Z gives Z = Z0 and hence, g = h (because Z0 is non-empty
and any clopen subset of a connected topological space Z is either the empty set or Z
itself).
Preliminary to establishing that Z0 is clopen, there are some preparations to be made;
let z be an arbitrary point in Z. Since p ◦ g(z) ∈ X and p is a covering map, there exists
9
a neighbourhood u of p ◦ g(z) such that p−1 (u) is a disjoint union of open subsets in X, e
each of which is homeomorphically mapped onto u. One of these open sets contains g(z)
and we denote this open set by u e ⊂ p−1 (u)). By hypothesis, we also know
e (i.e. g(z) ∈ u
that p ◦ g(z) = p ◦ h(z), and so h(z) belongs to p−1 (u), which similarly implies that one of
these open sets also contains h(z); let this set be denoted by ve (i.e. h(z) ∈ ve ⊂ p−1 (u)).
Finally, we are ready to prove the “clopenness” Z0 . Here we employ this clever trick;
we define N (z) = g −1 (e
u) ∩ h−1 (ev ) and this miraculously turns out to be a neighbourhood
of z, which is entirely contained in either Z0 or Z\Z0 , depending on which one of the two
sets the point z belongs to. If z ∈ Z0 , then it immediately follows that u e = ve, and thus
both g(N (z)) and h(N (z)) are subsets of u e (because N (z) is equal to g −1 (e u) ∩ h−1 (e
u)).
By definition of a covering map, we know that p|ue : u e → u is a homeomorphism, and thus
p ◦ g = p ◦ h on Z. Let n be a point in N (z). Then, we have; p ◦ g(n) = p ◦ h(n) ⇒ g(n) =
h(n) as p|ue is a bijection. Therefore, g|N (z) = h|N (z) , and thus N (Z) ⊂ Z0 . As z is an
arbitrary point of Z, it follows that ∀z ∈ Z0 , ∃ an open set N (z) such that z ∈ N (z) and
N (z) ⊂ Z0 . Hence, Z0 is an open subset of Z. Now, suppose that z is a point in Z\Z0 .
Then, g(z) 6= h(z) and, by definition, we have either u e ∩ ve = ∅. If u
e = ve or u e = ve, then, as
shown in the case where z is a point in Z0 , we have N (z) ⊂ Z0 and so z ∈ Z0 , which is a
contradiction. Therefore, u e ∩ ve = ∅. It can be readily deduced from the definition of N (z)
that g(N (z)) ⊂ u e and h(N (z)) ⊂ ve. These results imply that g(z , ) 6= h(z , ) ∀z , ∈ N (z),
and thus N (z) ⊂ Z\Z0 . The arbitrary choice of z implies that Z\Z0 is open, and so Z0
is closed. We have shown that Z0 is a clopen subset of Z and hence the required result
follows.
Corollary 4.1. Let (X,

e p) be a covering space of X and let Y be a connected topological
space. Given any two continuous maps g, h : Y → X e such that p ◦ g = p ◦ h, the set
Z0 = {z ∈ Z : g(z) = h(z)} is either empty or all of Y . [5, p.152]
Proof. This corollary directly follows from Theorem 4.1.
Lemma 4.1. Let (X̃, p) be a covering space of a topological space X. Let Z be another
topological space, and let A be a connected subset of Z. Suppose also that f is continuous
map from z to X and g is a continuous map from A to X̃ such that p ◦ g = f |A , and
also f (Z) is contained in some open subset of X that is evenly covered by p, then there
is a continuous map f˜ : Z → X̃ such that f˜ |A = g and p ◦ f˜ = f . We can think of the
map f˜ as an extension of g to the whole of Z, since the maps f˜ agree on A. [19]
Proof. We know that p−1 (u) is a disjoint union of open sets in X. e Pick a point a ∈ A.
Then, we have p ◦ g(a) = f |A (a) ∈ u since f (Z) ⊂ u, and hence g(a) ∈ p−1 (u). Therefore,
one of the disjoint open sets contains g(a) and we denote this open set as u e. We define
a map f from Z to X by f = σ ◦ f where σ : u → u
e e e e is the inverse function of the
homeomorphism p|ue : u e → u, and this turns out be the required map. We will show that
p ◦ fe = f . Pick an arbitrary point z ∈ Z. Then, by definition, fe(z) = p ◦ f (z) . It
trivially follows that p ◦ fe(z) = p ◦ σ(f (z)), and as f (z) ∈ u and σ ◦ p = idu , we have
p ◦ σ(f (z)) = f (z). Hence, p ◦ fe(z) = f (z) and the arbitrary choice of z gives p ◦ fe = f .
We complete the proof by establishing the restriction of f to A is equal to p ◦ g. The two
equalities p ◦ fe = f and p ◦ g = f |A imply that p ◦ fe|A = p ◦ g. Hence, it follows that
fe(a) = σ ◦ f (a) = σ ◦ (p ◦ g)(a) = g(a). As A is connected and a is an arbitrary point of
A, Theorem 1 implies that fe|A = g.
10
Theorem 4.2 (Path Lifting Theorem). Let (X̃, p) be a covering space of a topological
space X and let γ : [0,1]→ X be a path in X starting at γ(0) = x0 . If x˜0 ∈ X̃ is such
that p(x˜0 ) = x0 then there exists a unique path γ̃ in X̃ beginning at x˜0 and such that
p ◦ γ̃ = γ. That is, γ lifts to a unique path in X̃ that begins at x˜0 . [19]
X̃
γ̃
p
[0, 1] X
γ
Figure 5: Path lifting
Proof. First of all, we shall prove the existence of a lift. As p is a covering map, there
exists an open cover A = {ui : i ∈ I} of X such that each ui is evenly covered by p.
Then, {r−1 (ui ) : i ∈ I} is an open cover of [0, 1]. By the Heine-Borel theorem, [0, 1] is
compact, it follows from the Lebesgue Lemma that there exists some δ > 0 such that every
subinterval of [0, 1] of length less than δ is mapped into an element of A. Partition [0, 1]
into n subintervals [ti−1 , ti ] of length less than δ where 0 = t0 < t1 < ... < tn−1 < tn = 1.
Then, each of [ti−1 , ti ] is mapped by r into some open set u∗i−1 in X, which is evenly
covered by p.
Let i be an element of {1, 2, ..., n}, and let ri−1 denote the restriction of r to [ti−1 , ti ],
and let xi−1 denote r(ti−1 ). As the point xi−1 is contained in X and the covering map
p is surjective, there exists a point xg −1
i−1 in p (xi−1 ). Furthermore, we define a map
ggi−1 : {ti−1 } → X by gg i−1 (ti−1 ) = xi−1 . Then, it clearly follows that p ◦ gg i−1 = r|{ti−1 } .
e
As we know {ti−1 } is a connected subset of [ti−1 , ti ], Lemma 4.1 implies that there exists
a map rg i−1 : [ti−1 , ti ] → X such that rg i−1 |{ti−1 } = ggi−1 and p ◦ rg i−1 = ri−1 . We define
e
a map re : [0, 1] → X by re = rg
e i−1 on [ti−1 , ti ] ∀i ∈ {1, ..., n}. Then, it is obvious from
our method of defining re that the required commutativity relation p ◦ re = r holds and
re(0) = xf0 , and hence, we have constructed a lift of r starting at x f0 .
Now, we prove the uniqueness of the lift. Let re1 be another lift of r starting at x f0 .
Then, p ◦ re1 = p ◦ re and we know that r and re agree at 0. Therefore, Theorem 4.1 implies
that r = re and thus, the lift is unique.
Remark 4.1. It would be tempting to think that loops (closed paths) in the base space
lift to loops in the cover space. This, unfortunately, is not true. If a loop is based at x
then it is certainly true that the endpoint of the lift will lie in p−1 (x), but it need not be
the same as the start point.
The following theorem is the extension of Theorem 4.2 (Path Lifting Theorem) to
the case where the domain of a map which we intend to lift is [0, 1] × [0, 1], as illustrated
by Figure 6.
11
X̃ a0 a1
X a
Figure 6: Lift of a loop
Theorem 4.3 (Monodromy Theorem). Let p : X e → X be a covering map, let H :

[0, 1]×[0, 1] → X be a continuous map and let x
e0 be a point in X
e such that p(e
x0 ) = H(0, 0).
Then, there exists a unique continuous map H : [0, 1] × [0, 1] → X such that H(0,
e e e 0) = x e0
and p ◦ He = H. [19]
Proof. First of all, we shall prove the existence of a lift. The essence of the proof is exactly
the same as that of the unique lifting property. Let A = {ui : i ∈ I} be an open cover
of X such that each ui is evenly covered by p. Then, {H −1 (ui ) : i ∈ I} is an open cover
of [0, 1] × [0, 1]. The unit square [0, 1] × [0, 1] is compact and thus, the Lebesgue Lemma
implies that there exists some δ > 0 such that every square whose sides have length less
than δ is mapped by H into some element of A. Subdivide [0, 1] × [0, 1] into n × n squares
Sj,k where Sj,k = {(s, t) ∈ [0, 1] × [0, 1] : j−1 j k−1 k
n ≤ s ≤ n , n ≤ t ≤ n } such that n ≤ δ.
1
We will construct a lift H e over the S1,1 . Let U ∗ denote the element of A which contains
H(S1,1 ) and let H1,1 denote the restriction of H to S1,1 . We know p(e x0 ) = H(0, 0) and
{(0, 0)} is a connected subset of S1,1 . Define a map gg 1,1 : {(0, 0)} → X e by gg1,1 (0, 0) = x
e0
i.e. p ◦ gg ∗
1,1 = H1,1 |{(0,0)} . Since we know that U is an open subset of X, which is
evenly covered by p, and H(S1,1 ) is a subset of U ∗ , Lemma 4.1 implies that there exists a
continuous map H 1,1 : S1,1 → X such that p ◦ H1,1 = H1,1 . Hence, we have constructed a
g e g
lift of H over S1,1 . In a similar manner, we can continuously extend H e successively over
every other square and eventually cover the whole of [0, 1] × [0, 1]. Hence, the existence
of a lift follows. The uniqueness of the lift follows from Theorem 4.1 in the exactly same
manner as in the proof of the unique lifting property.
Now we are ready to prove an important result about the homomorphisms induced
by covering maps:
Theorem 4.4. Suppose that (X, p) is a covering space of Y . Then the group homomor-
phism p∗ is injective.
Proof. Suppose that [f ] is an element of the kernel of p∗ . Then any loop in [p ◦ f ] is
homotopic to a constant loop in X. But, we can lift this homotopy to a homotopy in the
covering space, and so any γ ∈ [f ] is homotopic to the identity element in the fundamental
group of the cover. Thus, since the kernel has only the identity as an element, by results
from algebra the homomorphism is injective.
Remark 4.2. The monodoromy theorem implies that given any homotopic paths in X,
you can find the unique lift of the homotopy between these paths in X.
e Additionally, this
12
theorem give the following corollaries, which will be frequently used in the subsequent
section of this project.
Corollary 4.2. Let p : X e → X be a covering map. Given two paths g0 , g1 in X,

e if
p ◦ g0 ' p ◦ g1 , then g0 ' g1 .
Proof. It would suffice to define a homotopy between g0 and g1 . As p ◦ g0 and p ◦ g1 are

homotopic paths in X, there exists a homotopy H : [0, 1] × [0, 1] → X between them.
By applying Theorem 4, we obtain a map H e : [0, 1] × [0, 1] → X
e such that p ◦ H
e = H.
It turns out that H is a homotopy between g0 and g1 (for the details of the proof, see
e
Massey’s book).
Corollary 4.3. Let p : X e → X be a covering map. Suppose that f0 and f1 are two
homotopic paths in X with the same beginning point, with respective lifts fe0 and fe1 to
X.
e Then the lifts fe0 and fe1 both end at the same point, and are homotopic in X.
e
Proof. Since we can lift f0 and f1 to fe0 and fe1 respectively, and so by definition of a lift
we must have fe0 ◦ p = f0 and fe1 ◦ p = f1 , and thus we directly apply Corollary 4.2, s to
see that fe0 and fe1 are homotopic.
So far, we have considered the cases where the domain of the continuous functions
we intend to lift is either [0, 1] or [0, 1] × [0, 1]. However, there is a criterion which can
be employed to determine whether continuous maps to X from any general topological
space Y may be lifted to X e (which is known as the “Lifting Criterion”). Preliminary to
proving this big result, we need the following lemma:
Lemma 4.2. A locally path connected and connected topological space is path connected.
Proof. Let X be a locally path connected and connected topological space and fix x ∈ X.
Let A be the set of all points in X which can be joined by a path to x. As A is non-empty
(as x ∈ A), it suffices to show that A is clopen (since the connectedness of X implies that
A is all of X). First, we will show that A is open. Fix a ∈ A and take a path connected
neighbourhood U of a. Then, any given u ∈ U and a can be joined by a path, and we
can also find a path between a and x. Hence, by addition of paths, u can be joined to x
by a path and thus, u is a point in A. Hence, U is a subset of A and so, A is an open
set. Similarly, fix a point b ∈ X/A and choose an path connected neighbourhood V of b.
To show that A is closed, we first establish that A ∩ V = ∅. Suppose the contrary that
A ∩ V 6= ∅. Take q ∈ A ∩ V . Then, we can find paths joining q to b and q to x respectively,
which implies b is a point in A. This is clearly a contradiction as b ∈ X/A, and thus we
obtain that A ∩ V = ∅. This equality implies that V ⊂ X/A and so, X/A is open i.e. A
is closed.
Finally, we are ready to prove the lifting criterion. We should note that this theorem
deals with existence only, but will be sufficient for our purposes. When presenting this
theorem we took ideas from Massey and Hatcher.
Theorem 4.5 (Lifting Criterion). Let (X, e p) be a covering space of (X, x0 ), let (Y, y0 ) be
a locally path connected and connected space. Given a map ϕ : (Y, y0 ) → (X, x0 ), there
e : (Y, y0 ) → (X,
exists a lifting ϕ e xf0 ) if and only if ϕ∗ (π1 (Y, y0 )) ⊂ p∗ (π1 (X,
e xe0 )).
13
(X̃, x̃0 ) π1 (X̃, x̃0 )
ϕ̃ ϕ̃∗
(Y, y0 ) p π1 (Y, y0 ) p∗
ϕ ϕ∗
(X, x0 ) π1 (X, x0 )
(a) (b)
Figure 7: Lifting criterion
Proof. We shall prove first that ϕ∗ (π1 (Y, y0 )) ⊂ p∗ (π1 (X,e x e0 )) is a necessary condition for
the existence of such a lifting ϕ e by considering the fundamental groups of the topological
spaces involved. For, if we suppose such a lift ϕ e exists, then we obtain ϕ∗ = p∗ ◦ ϕ e∗ . The
proof is easy: ϕ([r])
e = [ϕ◦r] = [p◦ ϕ◦r]
e e ∗ ([r]) = p∗ ◦ ϕ
= (p◦ ϕ) e∗ ([r]) for any [r] ∈ π1 (Y, y0 ).
Clearly, ϕ e∗ (π1 (Y, y0 )) ⊂ π1 (X,
e xe0 ) and hence, the injectivity of p∗ (which follows from
Theorem 4.4) implies that ϕ∗ (π1 (Y, y0 )) = p∗ ◦ ϕ e∗ (π1 (Y, y0 )) = p∗ (ϕe∗ (π1 (Y, y0 )) is a sub-
set of p∗ (π1 (X,
e xe0 )). See Figure 7b for clarity.
Now, we prove that ϕ∗ (π1 (Y, y0 )) ⊂ p∗ (π1 (X,

e xe0 )) is a sufficient condition for the ex-
istence of the lift ϕ.
e To do so, we shall actually construct the map ϕ e as follows. As Y
is locally path connected and connected, Lemma 4.2 implies that Y is path connected.
Hence, given any y ∈ Y , we may take a path γ : [0, 1] → Y with the initial point y0 and
the endpoint y. Then, ϕ ◦ γ is a path in X starting at x0 and apply the unique lifting
property to obtain the lift ϕ
] ◦ γ with the initial point xe0 . Define:
ϕ(y)
e =ϕ
] ◦ γ(1) = the endpoint of ϕ
] ◦ γ.
Then, it is obvious from our way of defining ϕ e that the required commutativity relation
p◦ϕ e = ϕ holds (the proof is easy: for any given y ∈ Y , we have p ◦ ϕ(y) e = p◦ϕ ] ◦ f (1) =
ϕ ◦ γ(1) = ϕ(y)). To show that ϕ e is the required lift, it would suffice to show that ϕ e is
well-defined and continuous. We will first establish that ϕ e is well-defined: we will show
that ϕ(y)
e is independent of the choice of the path γ.
Choose another path γ1 from y0 to y which is homotopic to γ. As γ ' γ1 , it easily
follows that ϕ ◦ γ ' ϕ ◦ γ1 . Then we can lift to paths ϕ ◦ γ and ϕ ◦ γ1 in X, e and denote
them by g and g1 (i.e. p ◦ g = ϕ ◦ γ and p ◦ g1 = ϕ ◦ γ1 ). Hence, p ◦ g ' p ◦ g1 and
thus, by using Corollary 4.2, we see that g ' g1 . Hence, ϕ(y) e = g(1) = g1 (1) i.e. ϕ(y)
e
only depends on the homotopy class of the path γ. We now prove that replacing γ with
another path from y0 to y in a different homotopy class does not alter ϕ(y). e Suppose
α and β are two different homotopy classes of paths in Y from y to y0 . Then, αβ −1 is
a set of loops based at y0 i.e. αβ −1 ∈ π1 (Y, y0 ) and so, by hypothesis, ϕ∗ (π1 (Y, y0 )) ⊂
p∗ (π1 (X,
e xe0 )) ⇒ ϕ∗ (αβ −1 ) ∈ p∗ (π1 (X,
e xe0 )). By using Remark 1 (i), we rewrite ϕ∗ (αβ −1 )
as follows: ϕ∗ (αβ −1 ) = ϕ∗ (α)ϕ∗ (β −1 ) = ϕ∗ (α)ϕ∗ (β)−1 . Hence, there exists ec ∈ π1 (X,
e xe0 )
such that p∗ (e −1
c) = ϕ∗ (α)ϕ∗ (β) (i.e. there exists a homotopy class of loops based at x e0
in Xe which projects onto ϕ∗ (α)ϕ∗ (β)−1 ). Therefore, the lift of any path in ϕ∗ (α)ϕ∗ (β)−1
is a loop in X e based at xe0 and this implies that if ϕ∗ (α) and ϕ∗ (β) are lifted to paths in
e then they have the same endpoints. Therefore, if you choose f γ ∈ α and γ1 ∈ β, then
X,
14
ϕ
] ◦ γ(1) = ϕ^ ◦ γ1 (1), and thus ϕ(y)
e is independent of the choice of the path γ. We note
here that the lift ϕe is unique in the view of Corollary 1.
Now, we prove that ϕ e is continuous. Here we shall use the definition in terms of
open sets of continuity of a map at a point. Suppose that y ∈ Y and let U be an open
neighbourhood of ϕ(y) such that p−1 (U ) is a disjoint union of open sets in the covering
space, each mapped homeomorphically onto U by the restriction of the covering map p to
each particular sets. Now, ϕ(y)
e is in one of these sets, say Ũ , and then the restriction of
p to Ũ is a homeomorphism. From now on, for ease of exposition, we shall assume that
when we see p we really mean p |Ũ , which is a homeomorphism.
Since Y is locally path connected, we can choose a path connected open neighbour-
hood V of y such that ϕ(V ) ⊂ U . Now, consider an arbitrary point y1 ∈ V . We know
that ϕe is well defined for any path we might choose to take from y0 to y1 , so we shall
divide the path into two parts as follows: fix a path γ from y0 to y, then let η be a path
from y to y1 that is wholly contained in V , which we can do by the path connectedness
of Y . Now we map these paths via ϕ to X to get ϕ ◦ γ and ϕ ◦ η which are both paths in
X, and which we can lift to paths in X̃, respectively ϕ ] ◦ γ and ϕ ]◦ η . By the definition
of ϕ
e given above, we have that
ϕ(y ◦ γ • η (1) = ϕ
e 1 ) = ϕ^ ] ◦ η (1)
because we are interested in the endpoint of the path only. But this is what we want;
note that the path η is contained in V so that ϕ ◦ η (1) ∈ ϕ(V ) ⊂ U and since p is a
homeomorphism between U and U e , p−1 ◦ ϕ ◦ η (1) = ϕ ] ◦ η (1) ∈ Ue . The choice of y1
was arbitrary, and so for any point v ∈ V we have ϕ(v) e ∈ U , which says exactly that
e
e )⊂U
ϕ(V e , that is, ϕ
e is continuous at y. Sine y is an arbitrary point in Y , ϕ
e is a continuous
map.
Remark 4.3. This theorem beautifully illustrates how a purely topological question (the
existence of a lift of a general continuous map) can be reduced to a purely algebraic
question and this is a very good example of the general strategy of algebraic topology.
4.1 Homotopy Equivalence and the Homotopy Type of a Space

Here we describe another way to classify spaces, this time in terms of their homotopy
type, which is defined below.
Definition 4.2 (Homotopy Equivalence). Suppose that X and Y are topological spaces.
X and Y are said to have the same homotopy type if we can find continuous maps f :
X → Y and g : Y → X such that the compositions g ◦ f and f ◦ g are homotopic to the
identity function on X and Y respectively. If this is the case, we call the maps f and g
homotopy equivalences.
If two spaces are homotopy equivalent, then they have isomorphic fundamental groups
in the following sense: if f is such an equivalence between spaces X and Y then π1 (X, x0 ) ∼
=
π1 (Y, f (x0 )) via the isomorphism f∗ (not proved here, see Munkres for proof). It’s also a
fact (not proved here) that path-connectedness is an invariant of homotopy equivalence,
so that if X is path connected, so is Y , so if X is path connected, we need not con-
sider the base point at all, as the fundamental group of a path connected space is, up to
isomorphism, the same at all points.
Theorem 4.6. If X and Y are path connected and have the same homotopy type,
then π1 (X, xo ) ∼
= π1 (Y, f (x0 )) for all x0 ∈ X, where f is a homotopy equivalence. The
isomorphism is given by f∗
15
5 Equivalent Covering Spaces and Covering Transformations
One of the aims of our project is to find a link between the subgroups of the fundamen-
tal group and the covering spaces of a space. Notice that for a finite group the number of
all possible subgroups is also finite, but as far as we know, the number of covering spaces
might be infinite. In order to make a connection we need to define an equivalence relation
on covering spaces. Assuming for now that such covering spaces do exist (their existence
will be dealt with later) this section aims to classify them in terms of subgroups of the
fundamental group of the base space. The material in this section is largely inspired by
Massey’s book, listed as [5] in the bibliography.
5.1 Conjugate Subgroups of π1

At this point, we give a reminder of a definition from group theory:
Definition 5.1. Two subgroups H1 , H2 , of a group G are called conjugate if there exists
an element g ∈ G so that H1 = gH2 g −1 ; that is each element of H1 is of the form ghg −1
where h ∈ H2 .
Lemma 5.1. Conjugacy is an equivalence relation on the set of subgroups of G.
In light of this, the relation defines equivalence classes, which we shall call conjugacy
classes.
Now Theorem 4.4 leads to the following question: suppose x˜0 and x˜1 are points of X̃
such that p(x˜0 ) = p(x˜1 ) = x0 . Is there a link between the projections of the fundamental
groups based at x˜0 and x˜1 ? The fact that both points are in the fiber of x suggests a
connection, and in fact there is one, which is described in the next theorem.
Theorem 5.1. Suppose that x˜0 , x˜1 satisfy p(x˜0 ) = p(x˜1 ) = x let [γ̃] be a class of
paths from x˜0 to x˜1 . Let g = [γ] = p([γ̃]) = [p(γ̃)] and let G0 = p∗ (π1 (X̃, x˜0 )) and
G1 = p∗ (π1 (X̃, x˜1 )). Then, the following are true:
1. g is in π1 (X, x).
2. The relationship between the subgroups G0 and G1 is G0 = gG1 g −1 , that is, the
subgroups are conjugate — they belong to the same conjugacy class in π1 (X, x0 ).
3. If another subgroup G2 is conjugate to G0 , then G2 = p∗ (π1 (X̃, x˜2 )) for some

x˜2 ∈ p−1 (x).
[3, p.154][24, p.248]
Proof.
1. Clear that g is a class of loops based at x, since loops project to loops.
2. Already given g so direct check.
3. Note that any subgroup in the conjugacy class is of the form α−1 [G0 ]α for some
choice of the element α ∈ π0 (X, x). Choose a closed path f : [0.1] → X representing
α. Apply the path lifting theorem to obtain a path g : I → X̃ covering α with
initial point x˜0 and then let x˜2 be the terminal point of this lifted path. Then
p∗ π1 (X̃, x˜2 ) = α−1 p∗ π1 (X̃, x˜0 )α = G2 .
16
5.2 Deck Transformations
Definition 5.2 (Homomorphism). Let (X̃1 , p1 ) and (X̃2 , p2 ) be two covering spaces of a
topological space X. If ϕ : X̃1 → X̃2 is a continuous map such that the diagram in Figure
8 commutes, then we say it is a homomorphism.
ϕ
X̃1 X̃2
p1 p2
Figure 8: Commutative diagram of a homomorphism.
It is clear that the composition of two homomorphisms is a homomorphism, and, for

example, that the identity map on X̃ is a homomorphism.
Definition 5.3 (Isomorphism). A homomorphism ϕ of (X̃1 , p1 ) into (X̃2 , p2 ) is called an
isomorphism if there exists a homomorphism ψ of (X̃2 , p2 ) into (X̃1 , p1 ) such that both
composition ϕ ◦ ψ and ψ ◦ ϕ are identity maps on X̃1 and X̃2 respectively. Two covering
spaces are said to be isomorphic, or equivalent, if there exists an isomorphism between
them. An automorphism is an isomorphism of a covering space on to itself; it need not
be the identity map though.
Automorphisms are usually called deck transformations or covering transformations.
A homomorphism of covering space is an isomorphism if and only if it is a homeomorphism,
so we can condense to the following definition:
Definition 5.4 (Equivalence of Covering Spaces). We say that two covering spaces
(X̃0 , p0 ) and (X̃1 , p1 ) are equivalent if there exists a homeomorphism h from X̃0 to X̃1
such that p0 = p1 ◦ h.
So, the spaces must be topologically equivalent and the maps must agree. This ensures
that the structure of the covering space is retained, of which the meaning becomes clear
if we note that, since p0 = p1 ◦ h, the image of the set p−10 (x) under an equivalence of
covering spaces is p−1
1 (x), so the important covering properties are retained.
Lemma 5.2. The set of all deck transformations of a covering space (X̃, p) of X is a
group under the operation of composition of functions, and we use the notation A(X̃, p)
for this this group. [3, 159]
Proof.
1. Closure: If ϕ and ψ are two deck transformations, then ϕ ◦ ψ is also a deck trans-
formation as composition of two homeomorphisms is a homeomorphism.
2. Identity: The identity map i: X̃ → X̃ is a homeomorphism.
3. Inverse: If ϕ is a deck transformation then ϕ is a homeomorphism so there exists

ϕ−1 such that ϕ ◦ ϕ−1 = i.
17
Lemma 5.3. Suppose that ϕ0 and ϕ1 are two deck transformations of (X̃1 , p1 ) into
(X̃2 , p2 ). If there exists a point x ∈ X̃1 such that ϕ0 (x) = ϕ1 (x) then these two deck
transformations are the same, i.e ϕ0 = ϕ1 . [3, 159]
Proof. This is a special case of Theorem 4.1.
Lemma 5.4. If ϕ ∈ A(X̃, p) and ϕ 6= i, then ϕ has no fixed point. [3, 159]
Proof. Follows directly from the previous lemma.
Lemma 5.5. Let (X̃1 , p1 ) and (X̃2 , p2 ) be two covering spaces of X with x˜1 ∈ X̃1 and
x˜2 ∈ X̃2 points such that p1 (x˜1 )=p2 (x˜2 ). Then there exists a homomorphism ϕ of (X̃1 , p1 )
into (X̃2 , p2 ) such that ϕ(x˜1 ) = x˜2 if and only if p1∗ (π1 (X̃, x˜1 )) ⊂ p2∗ (π1 (X̃, x˜2 )). [3, 159]
Proof. We apply the lifting criterion (proved above) to (X̃1 , p1 ) and (X̃2 , p2 ) instead of
(Y, y0 ) and (X, x0 ). So we get p1∗ (π1 (X̃, x˜1 )) ⊂ p2∗ (π1 (X̃, x˜2 )).
Lemma 5.6. Let (X̃1 , p1 ) and (X̃2 , p2 ) be two covering spaces of X with x˜1 ∈ X̃1 and
x˜2 ∈ X̃2 points such that p1 (x˜1 )=p2 (x˜2 ). Then there exists an isomorphism ϕ of (X̃1 , p1 )
into (X̃2 , p2 ) such that ϕ(x˜1 ) = x˜2 if and only if p1∗ (π1 (X̃, x˜1 )) = p2∗ (π1 (X̃, x˜2 )). [3, 159]
Proof. Obvious consequence of previous lemma.
Theorem 5.2. Let(X̃, p) be a covering space of X and x˜1 , x˜2 ∈ p−1 (x), x ∈ X.
There exists a deck transformation ϕ ∈ A(X̃, p) such that ϕ(x˜1 ) = x˜2 if and only if
p∗ (π1 (X̃, x˜1 )) = p∗ (π1 (X̃, x˜2 )). [3, 159]
Proof. We apply previous lemma to (X̃, p) and (X̃, p) with x̃i ∈ X̃,i = 1, 2 .
This theorem links the conjugacy classes of π1 (X, x0 ) and the covering spaces of X
up to equivalence.
5.3 Criterion for Equivalence of Covers

Here we have a fascinating correspondence. The next theorem tells us that we can
examine the conjugacy classes of subgroups of π1 to find out whether or not two covering
spaces are equivalent - a perfect example of using algebra to solve problems in topology:
Theorem 5.3 (Cover Equivalence). [3, 159] Two covering spaces (X̃1 , p1 ) and (X̃2 , p2 ) of
X are equivalent (or sometimes isomophic) if and only if, for any two points x˜1 ∈ X̃1 and
x˜2 ∈ X̃2 such that p1 (x˜1 ) = p2 (x˜2 ) = x0 , the subgroups p1∗ (π1 (X̃, x˜1 )) and p2∗ (π1 (X̃, x˜2 ))
belong to the same conjugacy class in π1 (X, x0 ).
Proof. Follows from Theorem 5.1 and Lemma 5.2.
The following Lemma is important in the construction of the universal cover, which
is in the next charpter.
Lemma 5.7. Let (X̃1 , p1 ) and (X̃2 , p2 ) be two covering spaces of X, and let ϕ be a
homomorphism of the first covering space into the second. Then (X̃1 , p1 ) is a covering
space of X̃2 . [3, 160]
18
Proof. First let y be any point in X̃2 , we want to show that there exist a point x ∈ X̃1
such that ϕ(x) = y. Choose a base point x1 ∈ X̃1 and ϕ(x1 ) = x2 , by Definition 5.2 we
have p1 (x1 ) = p2 (x2 ) = x0 . Now we choose a path α in X̃2 from x2 to y. Then β = p2 (α)
is a path in X with initial point x. By the lifting theorem, there exists a unique path γ
in X̃1 with initial ponit x1 such that p1 (γ) = β. We now define the endpoint of γ to be
x. Then ϕ(γ) and α are two paths in X̃2 with same initial point and the projection are
the same i.e p2 (ϕ(γ)) = β = p2 (α), hence by the lifting theorem ϕ(γ) = α and hence we
find x such that ϕ(x) = y.
6 The Universal Covering Space

Here we digress for a time to explain a couple of ideas that will be of use later on.
6.1 Local Properties

A useful notion for later will be that of semi-local simple connectedness:
Definition 6.1 (Semi-local Simple Connectedness). Say a topological space X is semi −

locally simply connected if, for all points x ∈ X there is an open neighbourhood Ux of x
such that the image of π1 (Ux , x) under the homomorphism induced by the inclusion map
i : Ux → X is the trivial group.
So, any loop in Ux must be homotopic to a constant loop when viewed as a loop in X,
we note that Ux need not be itself simply connected. The homotopy may take the loop
outside of Ux , so this is not the same as local simple connectedness.
A worthy remark made by Hatcher on local properties is the following. When we
speak of a space satisfying properties locally, we mean that the space has the property on
arbitrarily small neighbourhoods of points in the space.
6.2 Quotient Spaces

Given a topological space (X, τ ) and an equivalence relation ∼ on X, we introduce the
idea of “identification” of points. We look at the equivalence classes defined by ∼ and say
that in our new space, we will regard all points in a given equivalence class as one point.
Thus, we make the following definition, as given in [21]:
Definition 6.2 (Quotient Space). A quotient space of X, (X/ ∼, T̃ ) is the set of equiva-
lence classes, X/ ∼, with the quotient topology T̃ consisting of all subsets of X/ ∼ whose
pre-images under the natural map (that is, the map from X → X/ ∼ sending each point
to the equivalence class that it is in) are open in X .
Remark 6.1. The natural map between the two topological spaces is continuous, which
is clear from the definition.
In this section we state the conditions under which a space has a universal cover, then
we describe the construction of a universal cover assuming this condition is satisfied, and
use this to link all the covers of a space with subgroups of the fundamental group.
19
6.3 Existence of a Universal Cover
The big result of this section: if a topological space is sufficiently “nice” (i.e. satisfies
the semi-local simple connectedness condition) then it has a universal cover, of which we
will sketch the construction below [7]. Also, we will see that any two simply connected
covers are equivalent in the sense that was described above. Also, any other covering
space is covered by the universal cover, so it is, in a sense, a more general idea of a
covering space.
The proof that a semi-locally simply connected topological space has a universal cover
is entirely constructive. Given a topological space X, we describe another topological
space X̃ and a map p. We then put a topology on X so that these two objects form a
covering space for X, and finally that X̃ is simply connected.
The idea is to define a space X̃ in terms of X so that our construction works for
general spaces. In a simply connected space Y , any two paths that that have the same
start and endpoint are homotopic, and so for any pair of points there is a unique homotopy
class of paths joining them. This means that if y0 ∈ Y , we can in fact imagine points
in the space as homotopy classes of paths that begin at our base point y0 . The clever
bit is that if Y is a covering space for some space X then by our results on path and
homotopy lifting, these homotopy classes correspond exactly to the homotopy classes in
X. This will give us our method of constructing a universal cover for a given space, which
is actually defined to be the set of all homotopy classes of paths based at a point in X.
The covering map is the map that sends a homotopy class of paths, i.e. an element of our
constructed universal cover X̃, to the common endpoint of all the paths in this class. The
description of the topology on this space is rather technical, so we will not go into the
details here. For a complete proof of the existence theorem, see Hatcher’s book Algebraic
Topology. Also laid out in Hatcher’s book are the details of how to get a covering space
X˜H , for each subgroup H of π1 (X, x), such that the projection subgroup associated with
X̃ is this subgroup. These covers are quotient spaces of the universal cover, and so the
name universal cover is most definitely justified, since once we have it, we can construct
all other covers. The equivalence relation that defines this quotient space relates classes
of paths that have the same endpoints, and satisfy an additional condition. Precisely, we
have that [γ] ∼ [γ 0 ] if γ(1) = γ 0 (1) and additionally [γ •γ−
0 ], where the subscript - denotes
that we take the reversed orientation of γ 0 . This means that the paths γ and γ 0 can be
thought of as respective halves of a loop in H. If we project the fundamental group based
at x̃ where x̃ is the homotopy class of the constant loop at x, then we get exactly the
subgroup H.
7 Summary of Classification of Covering Spaces

This section aims really to just collate the ideas we have regarding the intimate con-
nection between connected covering spaces of a topological space X and subgroups of the
fundamental group π1 (X, x0 ). We wll restrict our attention to connected covers, and the
reason for this is that the disconnected covers are just disjoint unions of the connected
covers, and so it suffices to classify the connected ones for us to get the full picture.
We note that for every covering space (X̃, p) we get a subgroup of π1 (X, x0 ), namely
p∗ (π1 (X̃, x˜0 )), where p∗ is the induced homomorphism defined above. This describes a
correspondence (sometimes called the Galois correspondence) between covering spaces of
X and subgroups of π1 (X, x0 ) which by the previous section is surjective - that is - for
any subgroup H of the fundamental group, there is a covering space (X˜H , p) such that
20
p∗ (π1 (X˜H , x˜0 )) = H for some x˜0 ∈ X̃.
We now have a way to think about covering spaces. The relationship described above
shows that they are a sort of topological analogue of the algebraic concept of a subgroup
of another group. Developing this, and to round off our theory of covering spaces, we
see that just as the trivial group is a subgroup of all subgroups of a given group G, the
universal cover is a cover of all covering spaces of a topological space X, as shown by the
following theorem:
Theorem 7.1 (Partial Ordering of Covers). Suppose that (X̃, p) is a universal cover of
a topological space X and suppose that (X̃0 , p0 ) is any other covering space of X. Then
(X̃, pe) is a covering of (X̃0 , p0 ) where pe is the lift of p to X̃.
Proof. Here we make use of the lifting criterion; since the universal cover is simply con-
nected, it has trivial fundamental group, so clearly the hypotheses for the lifting criterion
are satisfied, and so we have that there exists a lift of the map p to X̃0 , which we shall
call pe. Now we refer to Lemma 5.7. This theorem states that if (X1 , p1 ) and (X2 , p2 )
cover X and there exists a lift of p1 to X2 then X1 covers X2 . Our situation is clearly a
special case of this result, and hence our simply connected cover covers any other covering
space.
21
Part II
Applications to Graph Theory
As we have seen before, we can classify the covering spaces of surfaces according to
subgroups of the fundamental groups. If we approach surfaces from the view of graph
theory, we can categorise surfaces more explicitly. More precisely, we can prove that
every closed surface can be shown to be topologically equivalent to one and only one of
the sphere or a sum of tori or projective planes[14, p.145-149]. Graph theory also provides
us with an easy way to classify a given surface by way of the Euler characteristic. Using
the Euler characteristic, there is no need to provide an explicit deformation to one of the
standard forms in order to categorise a surface, saving us both time and effort.
8 Some Basic Graph Theory

8.1 Essential Definitions and Concepts
Definition 8.1 (Graph). Formally, a graph consists of two sets of objects: the vertices
V = {vi } and the edges E = {ei }, subject to incidence relations. Each edge ej consists of
−1
a pair {e+1
j , ej } called the positive and negative orientation of ej , and ej is incident on
two vertices, which are called the initial and final points of e+1
j or equivalently the final
−1
and initial points of ej . For simplicity we denote a graph, G , as the pair G = (V, E).
[2, p.91]
Definition 8.2 (Path). A path on a graph G from a vertex v1 to vn+1 is a finite sequence
of oriented edges e1 e2 . . . en−1 en , where e1 joins v1 to v2 , e2 joins v2 to v3 , . . . , and en
joins vn to vn+1 . A closed path (or loop) is the special case where v1 = vn+1 . [6, p.280]
Definition 8.3 (Simple Path). A path on a graph G is simple if the edges e1 , e2 , . . . , en−1 , en
are all distinct, and the vertices v1 , v2 , . . . , vn , vn+1 are all distinct (except in the case of
a loop). [6, p.280]
Definition 8.4 (Connected Graph). A graph G is connected if given any two vertices v
and w of G , there is a path from v to w. [6, p.280]
8.2 Simplices
When first considering graphs, we discussed the properties of the vertices and edges.
When we extend our idea of graphs to surfaces in three dimensions (and higher), we
introduce triangular faces to our analysis. We can summarise the general concepts of
these building blocks with the notion of a simplex.
R
Definition 8.5 (Simplex). The standard n-simplex is a subset of n +1, {(x0 , x1 , . . . , xn )| xi =
P
1, and 0 < xi < 1}. Any P arbitrary n-simplex can be obtained from the above by map-
ping (x0 , x1 , . . . , xn ) 7→ vi xi , with {vi } being the set of vertices of the new, arbitrary
simplex.
Example 8.1. Therefore, given the above definition, we can see that a 0-simplex is a
point (or vertex), a 1-simplex is a line (or edge), a 2-simplex is a triangle, a 3-simplex is
a tetrahedron and so on to higher dimensions.
22
The main aim of this chapter is to “translate” surfaces into graphs – we will call this
graph a triangulation of the surface. It has actually been proven that any two-dimensional
surface can be triangulated [22] (and methods to explicitly construct these triangulations
exist [23, p.124-135]). In order to be able to convert surfaces to graphs by means of
triangulation, we need to put some constraints on what a “valid” representation of a
surface is.
Definition 8.6 (Intersection Condition). We can impose an intersection condition upon
a graph G, which requires that any two triangles formed by the graph satisfy one of the
following:
1. they are disjoint,
2. they share one (and only one) vertex,
3. they share two vertices (that is, they have one common edge).
[6, p.51]
So what does it mean to triangulate a surface? We can now use triangles to construct
two-dimensional graphs which represent a surface (we may need to“fold” this graph in
order to realise the surface in three dimensions). This is akin to the idea of being able
to create a two-dimensional net for a three-dimensional cube, and also enables us to re-
examine our concept of closed surfaces within the context of graphs. Given that we may
want to study surfaces which are impossible to realise in three dimensions (for example,
the Klein Bottle), translating them to graphs can be an invaluable tool. In order to do
this, we first introduce the link of a vertex:
Definition 8.7 (Link). The link of a vertex, v is the path formed by considering the
edges opposite v in each triangle containing v.
For example:
Figure 9: The bold path is the link of the vertex v in a surface.
Definition 8.8 (Closed Surfaces). A closed surface is a collection of triangles such that
1. the above intersection condition is satisfied,
2. the collection is connected,
3. the link of every vertex v is a simple closed polygon.

[6, p.52]
If we look at these definitions more closely, we can deduce a couple of things.
Lemma 8.1. The word representing a closed surface always contains an even number of
letters, with each letter being included in a pair.
23
Remark 8.1. If we label the outer edges on our two-dimensional collection of triangles
with letters, we must have an even number of letters in the “word” representing the
surface. Why is this? In order for our surface to be closed, each of these outer edges must
coincide with one and only one other edge. The only possible candidates for this are the
other outer edges (as the inner edges already coincide with another edge), and therefore
the outer edges must identify with each other in pairs.
Definition 8.9 (Barycentric Subdivision). A barycentric subdivision of a surface G is
obtained by subdividing each triangle into six by constructing the medians.
Figure 10: An example of barycentric subdivision of a triangle (2-simplex)
It can be shown that any diagram where the endpoints of any one edge are distinct,
every pair of edges is distinct, and every pair of triangles is distinct, can be converted to
a closed surface by means of two repetitions of barycentric subdivision. [6, p.64]
9 Euler Characteristics
We now move on to examining the Euler characteristic of a graph or surface;
Definition 9.1 (Euler Characteristic). The Euler characteristic, χ(K), of a collection of
triangles K is defined as
χ(K) = a0 (K) − a1 (K) + a2 (K),
where a0 (K) is the number of vertices, a1 (K) the number of edges, and a2 (K) the number
of triangles in K.
(Note that this definition can be extended to the nigher n-dimensional case by con-
sidering the alternating sum of k-simplices for k = 1, . . . n)
Theorem 9.1 (The Euler Characteristic of a Graph is Unaffected by Barycentric Subdi-
vision.). Given a graph K, and an accompanying graph K0 obtained from K by means of
barycentric subdivision, the Euler characteristic is preserved.
Proof. Let us consider a graph K, with barycentric subdivision K0 . Then;
a0 (K0 ) = a0 (K) + a1 (K) + a2 (K),
as each triangle and edge produces a new vertex.
a1 (K0 ) = 2a1 (K) + 6a2 (K),
as each six new edges are formed in each triangle, and each existing edge is divided into
two, and
a2 (K0 ) = 6a2 (K),
24
as each triangle is divided into six smaller triangles. Combining these results, we see that
the Euler characteristic of K0 is
χ(K0 ) = a0 (K) − a1 (K) + a2 (K)
This proof shows us intuitively that our choice of triangulation of a surface does not
affect the result we will obtain when calculating the Euler characteristic. A rigorous proof
of this, however is much more complicated.
9.1 Examples
Example 9.1 (Sphere). There are multiple ways of triangulating the surface of a sphere,
but the easiest way is possibly to imagine the edges of an octahedron being projected
onto the sphere. As we do not alter the structure of the actual graph when we perform
this deformation, the Euler characteristic is preserved.
Figure 11: One possible triangulation of the sphere.
Looking at the octahedron:


a0 = 6 

a1 = 12 ⇒ χ(K) = 2

a2 = 8

Example 9.2 (Torus). To calculate the Euler characteristic of the torus, we again need
to triangulate the surface. We can represent such a triangulation in the plane as follows,
where the top edge is identified with the bottom edge, and the left and right edges are
also the same edge (that is, the torus can be formed by “sticking” the top and bottom
edges of this sheet together, and then the left and right edges together.).
25
Figure 12: One possible triangulation of the torus.
Now, we need to consider the number of vertices, edges and triangles (or equivalently,
0-simplices, 1-simplices and 2-simplices) in order to compute the Euler characteristic. In
the diagram above, vertices which are shown twice are labelled similarly to make this
computation clearer.

a0 = 9 
a1 = 27 ⇒ χ(K) = 0

a2 = 18

Example 9.3 (Klein Bottle). The Klein bottle is a surface which can only be realised in
four dimensions, and is similar to a higher dimensional Moebius strip. This is where the
power of being able to convert surfaces to two-dimensional representations is apparent;
we can now analyse a surface of which we cannot even visualise properly (the three-
dimensional projection requires self-intersection, which is rectified when working in four
dimensions). The triangulation of the Klein bottle is similar to that of the torus, but
when folding one pair of edges, we include a “twist”. This twist is what differentiates the
two surfaces, and as we shall see later, is of great importance when categorising closed
surfaces.
Figure 13: One possible triangulation of the Klein Bottle.
As before, we consider the sum of the simplices:


a0 = 9 

a1 = 27 ⇒ χ(K) = 0

a2 = 18

26
10 Results on Polygonal Representations
A pivotal part of our journey towards proving the classification theorem is showing that
we can reduce surfaces to a limited number of standard forms, and connected sums of
these. These standard forms are the sphere, the torus and the projective plane. To do
this, we employ “scissors and glue” transformations, [6, p.62]. First, it is necessary to
introduce some notational conventions.
Definition 10.1 (Letters and Words). We can label the boundary edges of a polygonal
representation with a succession of letters. We refer to this sequence as a word. If two
edges are identified, we endow them with the same letter. If a sequence of edges always
occurs together (for example, abc in the torus triangulation above), we can replace this
with another letter, say x. We must also take into account the orientation of each edge.
For example, the representation of the torus above can be written as xy −1 x−1 y, where
x = abc and y = f ed. Note that, similarly to group theory, (abc)−1 = c−1 b−1 a−1 . Another
important point is that the representation of a polygon is cyclic, that is abc = bca = cba.
Definition 10.2 (Pairs of the First Kind). Let a polygonal representation be described
by a word . . . a . . . a . . . . We refer to this pairing of a as a pair of the first kind.
Definition 10.3 (Pairs of the Second Kind). Let a polygonal representation be described
by a word . . . a . . . a−1 . . . . We refer to this pairing of a as a pair of the second kind.
10.1 Connected Sums

We have already been introduced to some (fairly) simple surfaces, and have encountered
the concept of quotient spaces (see Definition 6.2 in Part I). We can create new surfaces
from our previous examples by performing an operation analogous to addition. The
connected sum of two surfaces A&B (denoted A # B) is the surface formed by removing
a ball from each surface, and then gluing the resulting the two new boundaries together.
Visually, this is an very intuitive way to “sum” spaces:
Figure 14: A visual representation of connected sums. [17]
Definition 10.4 (Connected Sums). Formally, let A&B be contained in Rn, and Dn ⊂
R n be a closed n-disk. Let α : Dn → A be a continuous injection, and β : Dn → B be a
similar function. Define the set:
S = (A \ α(( Dn)◦)) ∪ (B \ β((Dn)◦))

where: ( Dn)◦ denotes the interior of B.
27
Define an equivalence relation ∼ on S as:
x ∼ y ⇔ (x = y) or (α−1 (x) = β −1 (y))
Since the interiors of the disks were removed from the manifolds, it necessarily follows
D
that α−1 (x), β −1 (y) ∈ δ n
Then the connected sum A # B is defined as the quotient space of S under ∼. [15]
We now wish to show that the word representing any closed surface can be reduced
to one of three standard forms.
Lemma 10.1 (Transformation to Standard Forms). Any polygonal representation of a
closed surface can be reduced to one of the following:
• aa−1 (equivalently, the sphere)
• a1 a1 a2 a2 . . . ak ak , k ≥ 2 (equivalently, sphere with k cross caps, or the connected
sum of k projective planes)
• a1 b1 a−1 −1 −1 −1
1 b1 . . . ah bh ah bh (equivalently, sphere with h handles, or the connected
sum of h tori)
Proof. We introduce three different “scissors and glue” operations from Giblin’s Graphs,
Surfaces and Homology (which is highly recommended for more details on the following)
in order to transform our word representation to one of the above standard forms.
• Cancellation of aa−1
Similar to group theory, we can eliminate a sequence aa−1 from a representational
word, but only if this does not result in us reducing the sequence to an empty word.
This “folding” technique is encountered later in greater detail (see Part III, Section
17).
Figure 15: A visual representation of the process of eliminating aa−1 . [6]
• Rearrangement of pairs of the first kind

We can most simply convey this concept with a diagram. We simply “join” the two
matching edges, and “split” the remaining edge into two (which becomes bb). Note
that we need there to be at least one other pair of edges for this to work.
Figure 16: A visual representation of the process bringing together a pair of the first kind.
[6]
28
By repeated use of this technique, we can reduce any polygonal representation to
the form a1 a1 a2 a2 . . . an an X, where X is made up of solely pairs of the second kind.
• Rearrangement of pairs of the second kind

We have already reduced every word to the form a1 a1 a2 a2 . . . an an X above. If
there are at least two pairs of the second kind, and they interlock (that is we have
. . . a . . . b . . . a−1 . . . b−1 . . . somewhere in X), we can rearrange these to cdc−1 d−1 in
the following fashion:
Figure 17: A visual representation of the process rearranging a pair of the second kind.
[6]
Using these three operations, we have shown that any word which originally contained
a pair of the first kind can be reduced to the form a1 a1 a2 a2 . . . an an c1 d1 c−1 −1 −1 −1
1 d1 . . . cm dm cm dm .
Now, we have several possible scenarios to consider:
1. aa−1 (if there were no pairs of the first kind originally.)

This is homeomorphic to the sphere.
2. n = 0, m = h ∈ N
This is equivalent to c1 d1 c−1 −1 −1 −1
1 d1 . . . ch dh ch dh , and is homeomorphic to the con-
nected sum of h tori.
N
3. n = k ∈ , m = 0
This is equivalent to a1 a1 a2 a2 . . . ak ak , or the connected sum of k projective planes.
4. both n and m are non-zero.

This could potentially cause us a problem, as this does not appear to be the word
representation of any standard form. However, we can show that c1 d1 c−1 −1
1 d1 can
be transformed into pairs of the first kind, so this is actually equivalent to case 3.
Figure 18: Converting cdc−1 d−1 (the “handle” of a torus) to two pairs of the first kind.
[6]
29
Hence, we obtain the result required, and are well on our way to proving the Classification
Theorem of Surfaces.
11 Orientable Surfaces
We can also divide closed surfaces into two categories; orientable and non-orientable
surfaces.
Definition 11.1 (Orientable Surfaces). A closed surface M is orientable if every triangle

can be given an orientation such that two triangles, a & b, with a shared edge e are
coherently oriented, that is the direction of e in a is the opposite to the direction of e in
b.
Figure 19: Left: A coherent orientation. Right: A non-coherent orientation.
11.1 Examples of Orientable and Non-orientable Closed Surfaces

Example 11.1 (Sphere). If we consider the triangulation of the sphere above, we can
easily construct an orientation to show that the surface is orientable. If we arbitrarily
choose an orientation for one of the triangles, it is easy to then choose the orientations
for the remaining triangles such that the surface is coherently oriented.
Example 11.2 (Torus). Again, using the previous triangulation, we can show that the
torus is orientable.
Figure 20: Orientation of a torus.
30
Example 11.3 (Projective Plane). The projective plane is not orientable. This is a vital
point - while the torus and projective plane share the same Euler characteristic, they
do not share orientability. This can be seen by considering the ”twist” involved when
constructing the triangulation of the Klein Bottle (which is the sum of two projective
planes). The twist makes it impossible to coherently orient the triangles which make up
the surface, so the red arrows are not coherently oriented with the adjacent triangles when
we “stick” the left and right edges together. This is a key idea in being able to categorise
closed surfaces easily.
Figure 21: Attempting to orient a Klein bottle.
12 Classification of Surfaces
We now move onto the proof of our main statement, by first considering a more general
result about triangulations of surfaces. A preliminary lemma is required:
Lemma 12.1. The Euler characteristic of any simple, planar, closed polygon is equal to
1.
Proof. First, let us construct an arbitrary triangulation of our closed polygon, G, satisfying
the criteria set out in 8.6 and consider the effect of removing one of the boundary triangles
on the Euler characteristic:
χ(G) = a0 (G) − a1 (G) + a2 (G)
When we remove one triangle, we actually remove one vertex, two edges and one face,
therefore:
χ(G 0 ) = a0 (G 0 ) − a1 (G 0 ) + a2 (G 0 )
= (a0 (G) − 1) − (a1 (G) − 2) + (a2 (G) − 1)
= χ(G)
31
So if we continue to remove triangles in this manner, we will not affect the Euler
characteristic. Therefore, after a finite number of repetitions, we will obtain a single
triangle, and it can be easily verified that this has an Euler characteristic of 1.
We now look at a theorem which creates links between the Euler characteristic of a
closed surface and properties of its planar polygonal “net”, making it easier to evaluate
χ(K).
Theorem 12.1. If a closed surface K is represented by a word containing n letters and

a polygonal graph with n + r outer edges (recall that a letter can denote multiple edges).
If K has m distinct vertices, then:
n
χ(K) = m − +1
2
Proof. We consider the triangulation of the surface K which produces a net D when
“unfolded”. While the surface K is three dimensional, the net D is two dimensional. We
now evaluate the Euler characteristics of K and D separately.
We first look at D; as a planar surface, it is simple to compute the Euler characteris-
tic. By Lemma 12.1, we can collapse D to a single triangle without changing the Euler
characteristic — therefore
χ(D) = a0 − a1 + a2 = 3 − 3 + 1 = 1
Now we look at the boundary of the polygonal net. The number of vertices of D on
this boundary is n + r, while the number of vertices of K on the boundary is m + 2r .
Now, any vertices, edges and faces not on the boundary count equally in both surfaces
(D) and (K), so;
r r
a0 (K) − a0 (D) = (m + ) − (n + r) = m − n −
2 2
n r
a1 (K) − a1 (D) = −( + )
2 2
a2 (K) − a2 (D) = 0.
(We are justified in halving n as n must be even by Lemma 8.1.)
This implies that
χ(K) = χ(K) − χ(D) + χ(D)
r n r
= (m − n − ) + ( + ) + (0) + 1
2 2 2
n
=m− +1
2
Hence result.
Corollary 12.1 (The Classification Theorem). The Euler Characteristic of any closed
surface which can be deformed homeomorphically to (i.e. with standard form)
• the sphere is 2,
• the connected sum of k projective planes is 2 − k,
• the h-torus is 2 − 2h.
32
In addition, the standard form of a closed surface K is unique, and is determined by χ(K)
and whether or not K is orientable.
Note that the consideration of whether or not K is orientable is crucial to identifying

the standard form. For example, the Klein bottle has Euler characteristic 0, but both the
torus and double projective plane have Euler characteristic 0. The only way to determine
which standard surface the Klein bottle is equivalent to is to look at the orientability.
Proof. In order to prove that the Euler characteristic of closed surfaces with a specified
standard form satisfies the above, we simply need to verify that this statement holds
for the standard forms (following from Lemma 10.1). These are computed above, in the
examples section. The reason we only need to check these cases is that when we deform
a closed surface to a standard form by means of folding, we do not change the Euler
characteristic [6, p.62-65].
The statement about orientability can be deduced from the fact that folding again
has no effect on orientability. Now we note that none of the standard forms can have the
same Euler characteristic and orientability, and therefore, it is possible to categorise a
surface using only these two properties.
33
Part III
A More Topological View of Graphs
In the previous section, we started with a topological space, and used graphs as a tool
for the analysis and classification of surfaces. In this section, we take a slightly different
approach and treat the graphs themselves as the topological space. This means we can
apply the theory developed for generic topological spaces more effectively and therefore
arrive at some interesting results. However before we can do this, we need to enhance our
understanding of graphs with a few more definitions and concepts.
−1
Definition 12.1 (Reduced Path). The sequence e+1 j ej or e−1
j ej
+1
(an orientated edge
followed immediately by the edge in the reverse direction) is called a spur. A path on the
graph G is reduced if it contains no spurs. [2, p.91]
Definition 12.2 (Tree). A graph G is called a tree if it is connected and there are no
loops. For example, see Figure 22. [6, p.280] It is a spanning (or maximal) tree if it
contains every vertex of G . [2, p.91]
Figure 22: A tree and not a tree [6, p.21]
Definition 12.3 (Map of Graphs). Let G and H be graphs. A graph map f : G → H

consists of a vertex funcion fV : VG → VH and an edge function fE : EG → EH such
that the function preserves incidence i.e. for all edges e ∈ EG , the function f maps the
endpoints of the edge e to the endpoints of the edge fE (e). [1, p.5]
Definition 12.3 is required so that we can consider coverings in the future. Recall
that the fundamental group is defined by the equivalence classes of loops in the space.
Therefore we must define what we mean by equivalence of paths on a graph.
Definition 12.4 (Equivalence of Paths). Two paths p and p0 are equivalent if p0 can
be obtained from p by a finite number of removals or insertions of spurs from successive
oriented edges or endpoints. [2, p.92]
13 Some Useful Theorems

The following results will be required later, so are listed here for convenience.
Theorem 13.1. If an edge is removed from a tree, then 2 disjoint trees are formed.
34
Proof. Pick any edge of the tree. Let v0 and v1 be it’s endpoints. Since there are no
loops in a tree, there is only one path from v0 and v1 , i.e. the edge itself. So when this is
removed the resulting graphs are disjoint. Also, there are no loops in either subgraph.
Theorem 13.2. If T is a tree with |V | vertices, then the number of edges |E| is |V | − 1.
[1, p.5]
Proof. We will prove this by induction. For a tree with 1 vertex, it is obvious that the
number of edges is 0, otherwise any edge would form a loop: contradicting the fact that
T is a tree. Assume for induction that a tree with k vertices has k − 1 edges. Now we
show that this is true for a tree with k + 1 vertices. We remove one edge from the tree
and this results in two disjoint trees (by Theorem 13.1) such that each tree has less than
or equal to k vertices, say m and k + 1 − m. Then, we can apply the inductive hypothesis.
The total number of edges in this disjoint union of trees is (m−1)+(k +1−m−1) = k −1.
Hence, the total number of edges of the tree with k + 1 vertices is k − 1 + 1 = k.
14 The Fundamental Group of a Graph

The fundamental group π1 (G ) of a graph G is defined in almost exactly the same way to
the fundamental group of a topological space.
Definition 14.1 (The Fundamental Group). We begin by choosing a vertex P of G and

considering the equivalence classes of closed paths which begin and end at P , with path
equivalence as defined earlier.
If [p] denotes the equivalence class of such a path p, we define the product of equiva-
lence classes by
[p1 ] · [p2 ] = [p1 p2 ]
This product is well-defined on equivalence classes, since changing either p1 or p2 by
insertion or removal of spurs merely changes the product by insertion or removal of spurs,
and so the equivalence class remains unaltered. It is also clear that [p]−1 = [p−1 ] and
that the identity element e is the class of paths whose reduced form is the vertex P itself.
Thus π1 (G ) is indeed a group.
Again, in a similar way to topological spaces, the choice of a different base point
(vertex) leads to a isomorphic group. Therefore we can omit any mention of P in the
notation. [2, p.96]
15 Fundamental Groups and Free Groups

Having defined the fundamental group of graphs in the previous section, in this section
our main aim is to relate graphs and free groups, using the fundamental group as a tool.
By the end, we will show that the fundamental group of a connected graph is free. But
first, what exactly is a free group?
Let G be a group generated by S, i.e. any g ∈ G be written as a word g =
sλ1 1 sλ2 2 sλ3 3 . . . sλnn , where the si are not necessarily distinct and λi = ±1. We say that
g has a reduced product if si = si+1 ⇒ λi = λi+1 , i.e. any si is not immediately followed
by its inverse. [8, p.199-200]
Definition 15.1 (Free Groups). A group G is a free group if is is generated by a set S,

and the reduced product for every element of G is unique.
35
To proceed further, we require a couple of definitions. These are listed below.
Definition 15.2 (Bouquet of Circles). A bouquet of n circles is a graph with one vertex
and n loops based at the single vertex. The bouquet of n circles is usually denoted Bn .
Figure 23: A representation of B3
Definition 15.3 (Deformation Retract). A subspace A of a space X is a deformation

retract if there is a homotopy h : X × I → X such that h(x, 0) = x, h(a, t) = a and
h(x, 1) ∈ A for all x ∈ X, a ∈ A and t ∈ I. Such a homotopy is called a deformation of
X onto A.
Remark 15.1. Intuitively, a retract “continuously shrinks” a space onto a subspace. [1,
p.15]
Notice that the definition of deformation retract holds for any topological space. In
order to apply such topology ideas to graphs, we require a slightly more sophisticated
(and unfortunately less intuitive) definition of graph. This definition deals with ideas
such as open and closed sets, which will allow us to make sense of notions like continuity,
homotopy, etc in graphs.
Definition 15.4. A graph can be defined in topological terms as a pair consisting of a

Hausdorff space X and a subspace X 0 ⊂ X such that:
1. X 0 is a discrete, closed subspace of X. (Points of X 0 correspond to vertices of the

graph),
2. X \ X 0 is a disjoint union of open subsets ei , where each of the ei is homeomorphic

to an open interval of the real line. (The ei correspond to edges),
3. For each ei , its boundary ∂ei is a subset of X 0 consisiting of one or two points. If
∂ei consists of two points, the pair (e¯i , ei ) is homeomorphic to ([0, 1], (0, 1)) and if
∂ei consists of only one point, then (e¯i , ei ) is homeomorphic to (S 1 , S 1 \{1}) (This is
saying that an edge either connects two vertices or forms a loop when the endpoints
are the same vertex),
4. The topology on X: A subset A ⊂ X is closed/open iff A ∩ e¯i is closed/open ∀ei .
[3, p.198]
Equipped with this defintion, we are able to translate the idea of deformation retract
from topology to graph theory.
36
Theorem 15.1. Every edge of a graph is homotopic to a vertex. [11, p.42]
Proof. From our definition of graph, we know that any edge of a graph e is homeomorphic
to an open interval, say (0,1). Then, the closure of the edge, which is the edge and the
two vertices connected to it, is homeomorphic to [0, 1]. Let the vertices connected to this
edge be v0 and v1 . Then, there exists such a homeomorphism h : [0, 1] → ē with h(0) = v0
and h(1) = v1 . Now, we define a homotopy F : ē × [0, 1] → ē from ē to one of the vertex,
say v0 , by F (x, s) = h((1 − s)h−1 (x)). Note that this map is continuous (because, h is a
homeomorphism implies h−1 is continuous and therefore F is a composition of continuous
maps) and F (x, 0) = x and F (x, 1) = v0 for all x ∈ ē. So, the edge ē is homotopic to one
of its vertices. (Concept vaguely based upon [11, p.42])
Notice here how this theorem makes our previous definition of path equivalence in a
graph completely identical to path equivalence in a generic topological space (i.e. the
finite removal or insertion of spurs is actually a homotopy in the traditional sense). We
now use this to derive a crucial result:
Theorem 15.2. Any vertex of a tree T is a deformation retract of T . [9, p.35]
Proof. We prove this by induction on the number of edges. Let Tk be trees of any shape
with k edges. The case T1 is clear because we can find a homotopy to one of its vertices
by Theorem 15.1. Assume for induction that Tk is homotopic to any arbitrary vertex. If
we retract an outermost edge of Tk+1 into the vertex connecting to the rest of the tree,
then Tk+1 is reduced to Tk , which is homotopic to any by assumption.
Hence, a vertex is a deformation retract of Tk+1 . By induction any vertex is a defor-
mation retract of trees for all sizes.
For the purposes of rigour, we now need the following definitions and theorem. The
way in which we will utilise these will be explained in the final paragraph of the section.
Definition 15.5. The homotopy extension property is a property of a pair (X, A) where
A ⊂ X, X a topological space.
We say the pair has the property if we can find a homotopy ft : X → Y , when given
a map f0 : X → Y and a homotopy ft : A → Y on the restriction of the map f0 to A. [7,
p.14]
Definition 15.6. Given two spaces X and Y, we say they are homotopy equivalent if
there exist continuous maps f : X → Y and g : Y → X such that g ◦ f is homotopic to
idX and f ◦ g is homotopic to idY . [11, p.35]
Theorem 15.3. If the pair (X, A) satisfies the homotopy extension property and A is
contractible (i.e. A is homotopy equivalent to a point), then the quotient map q : X →
X/A is a homotopy equivalence. [7, p.16]
Proof. Since A is contractible there exists a homotopy ht : A → A such that h1 (A) = {a}
for some a ∈ A. Let ft : X → X be a homotopy extending ht with f0 = idX . This
can be done because (X, A) has the homotopy extension property. Since ft (A) ⊂ A for
all t, the composition q ◦ ft : X → X/A sends A to a point. Then, this is equivalent to
f¯t ◦ q : X → X/A where f¯t : X/A → X/A is a function such that Figure 24 commutes.
When t = 1 we have f1 (A) = {a}. So, f1 induces a map g : X/A → X such that
q ◦ g = f1 . By commutativity of Figure 25, we also have q ◦ g = f¯1 . Note that since
ft is a homotopy extension of the contraction of A, we have g ◦ q = f1 ' f0 = idX and
q ◦ g = f¯1 ' f¯0 = idX/A . Hence q and g are inverse homotopy equivalences.
37
ft f1
X X X X
g
q q q q
X/A X/A X/A X/A

f¯t f¯1
Figure 24 Figure 25
We will also need a few results from Hatcher’s Algebraic Topology.
(a) A graph is a 1 dimensional cell complex [7, Example 0.1 p.6]
(b) A tree is closed in G . [7, p.84]
(c) A pair (X, A) consisting of a cell complex X and a subcomplex A (a closed subset of
cell complex X) will be called a CW pair [7, p.7]
(d) Proposition 0.16. If (X, A) is a CW pair, (X, A) has the homotopy extension property.
[7, p.15]
Given our aims in this section, the precise definitions of the terms cell complex, CW
pair, etc. are not particularly relevant, however these properties are required to apply
Proposition 0.16. Let us consider G , a connected graph, with spanning tree T . By (a) G
is a cell complex. Since T is a closed subcomplex of G by (b), (G , T ) is a CW pair, by
(c). Therefore we may apply (d), giving the result (G , T ) has the homotopy extension
property.
Now by Theorem 15.2, the space T is contractible. Hence all conditions for Theorem
15.3 are satisfied, so the quotient space G /T is homotopy equivalent to G .
By collapsing T to a point v0 , say, we find that the quotient space G /T is Bn based at
v0 ; where n is the number of edges of G not in T . This is because every such edge of G
connects to two vertices in T and hence forms a loop based at v0 .
From earlier we know that Bn has a fundamental group isomorphic to the free group on
n generators. Using the Theorem 4.6 of Part I leads to our goal;
If G is a connected graph with spanning tree T , then π1 (G ) is a free group

with one generator for each edge of G not in T .
Furthermore, if G is finite, then π1 (G ) is free on 1 + |E| − |V | generators. This is

because the spanning tree T of G has the same number of vertices as G and |V | − 1
edges. Hence the the number of edges in G /T is |E| − (|V | − 1) = 1 + |E| − |V |.
16 Covering Graphs
The concept of a covering of a graph is analogous to the concept of a covering of a
topological space; it consists of a covering graph and a graph map with certain properties.
Definition 16.1 (Covering map). Let G = (V, E) and C = (V0 , E0 ) be graphs and
f : V0 → V is a surjection. Then f is a covering map from C to G if f is a locally
38
bijective map of graphs i.e. f maps edges incident to v0 ∈ V0 injectively to edges incident
to f (v0 ). [13, p.1]
(a) (b)
Figure 26: Covering map
For example (a) covers (b). This is because each black vertex in (a) is connected to
one white vertex and one black vertex, just as the black vertex in (b). Similarly for the
white vertices. Then, the obvious covering map maps black vertices in (a) to the black
vertex in (b) and the white vertices in (a) to the white vertex in (b).
Figure 27: Wedge of 2 circles
Example 16.1 (Wedge of two circles). Let us denote the wedge of two circles by G . G
is homotopy equivalent to the B2 , so it has the a fundamental group isomorphic to F2 ,
the free group on 2 generators.
b a
Figure 28: Free group with two generators a and b
In order to construct a cover, C , for G , we start with a vertex v̄0 ∈ C . Since there is
only one vertex v0 ∈ G , v̄0 must be mapped onto v0 . For it to be a cover, the map must
be locally bijective, so v̄0 has four oriented edges going out - each one corresponding to
a, b, a−1 , b−1 . If the other endpoint of these edges are all at another vertex v̄1 , we get the
covering graph shown in the Figure 29. This is because it can be easily checked that v̄1
is also mapped bijectively to v0 .
39
b
a
v̄1 v̄0
a−1
b−1
Figure 29: Example cover
However, this is certainly not the only cover of G . Another possible cover is con-
structed by insisting that the four outgoing edges from v̄0 terminate on four distinct
vertices. By the definition of covering map, these four vertices must also be locally
isomorphic to v0 . This recursive step must be repeated infinitely, leading to a fractal
structure.
We note that there are no loops formed at any point in the process, so C is an infinite
tree. From earlier we know that a tree is contractible, hence π1 (C ) is trivial. Therefore we
have found the universal cover. To summarise, we started with one of the simplest pos-
sible finite graphs, and arrived at the conclusion that its universal cover is a complicated
infinite graph with fractal structure.
b b
a−1 v̄0 a−1 v̄0
a a
b−1 b−1
(a) First iterative step (b) Second iterative step
Figure 30: Constructing the universal cover of F2
16.1 A Purely Algebraic Result

Any subgroup of a free group is also free
With the help of the theory developed in earlier chapters, we are now able to show this
group theoretic result without needing to indulge in group theory.
Let G be a free group with n generators. Then consider G = Bn . Hence, the funda-
mental group of G is isomorphic to the free group G i.e. π1 (G ) ∼
= G. For any subgroup H
of G, there is a covering map p : X̃ → G with X̃ a connected space such that the induced
map p∗ (π1 (X̃)) = H [See Section 6.3]. Since G is a graph, it follows that the covering
space X̃ is also a graph. Since p∗ is an injective homomorphism, we have π1 (X̃) ∼ = H.
From the result of the last section, the fundamental group of any connected graph is
isomorphic to a free group, hence H is free too.
Furthermore:
40
Theorem 16.1. If G is a free group with n generators and H is a subgroup of G with
finite index m, |G : H| = m. Then H is free with 1 + mn − m generators.
Proof. Given that H is a subgroup of G with index m, then H corresponds to a m sheeted

covering X̃ of G , i.e. the cardinality of the fiber of the vertex and each edge of G is m.
Since G is a bouquet of n circles, X̃ is a connected graph with mn edges and m vertices.
The spanning tree of X̃ has m vertices and m − 1 edges. This can be collapsed to a single
vertex. Hence, X̃ is equivalent to a bouquet of mn − (m − 1) = 1 + mn − m circles. Since
π1 (X̃) ∼
= H, H is a free group with 1 + mn − m generators. [10, p.7]
17 Folding: An Application
One application of covering maps of graphs is to find the rank and index of a subgroup
H of a free group G. Firstly we define an immersion as a graph map f : G → G 0 that
is locally injective. Based on this definition, we can see that the covering map is an
immersion since a covering map is locally bijective. Also, we can see that the inclusion
map is also an immersion.
Secondly, suppose that two orientated edges e1 and e2 in G have the same orientation
and label into or out of a common vertex v0 . Let the edge e1 connect vertices v0 and v1
while edge e2 connect vertices v0 and v2 . Then, a folding map f : G → G 0 identifies the
two edges e1 and e2 as well as the verticies v1 and v2 . Intuitively, this map “folds” the
graph by “sticking together” the specified vertices and edges.
These two definitions allow us to understand a theorem by John R. Stallings (1983).
Theorem 17.1 (Stallings Theorem). Every map G → G 0 of a finite graph can be factored
as
G = G0 → G1 → · · · → Gj → G 0
where the last map is an immersion and the other maps in the sequence are foldings. This
sequence is not unique but the final immersion is. [12, p.555]
We are not going to prove the theorem because the proof is out of the scope of this
project, but see [12, p.555] for the proof.
However, we will work though an algorithm on finding the rank (the lowest number
of generators required) and index (number of cosets) of a subgroup of a free group. [10,
p.8] Given a free group G with n generators {a1 , a2 , . . . , an } and a collection of k words
w1 , w2 , . . . , wk ∈ G. Let H be the subgroup of G generated by the words w1 , w2 , . . . , wk .
We wish to find the rank and the index of H.
This may seem trivial, but consider the following example: say we’re given a free
group, X, with two generators a and b. Let Y be the subgroup generated by the set
{a2 , b2 , ab, ba}. The rank is actually three (this will be shown later), meaning we only
need three of the four generators to construct the same subgroup, but this is not obvious.
We start by considering a graph G = Bn , which has the fundamental group isomorphic
to G. Then, consider a bouquet of k circles, each subdivided and oriented to spell out
the words wi for i = 1, 2, . . . , k. Call this graph H . The fundamental group of H is
isomorphic to the subgroup H of G [See Section 6.3]. We can do this because each element
of H can be written as a product of finitely many element of the generators wi and their
41
inverses. Note that H is a subgroup of G, so there exists a map H → G which is the
inclusion map.
Next, we then identify edges in H sharing a common vertex with the same label
and orientation and we “fold” these two edges together. We repeat until no more folding
is possible. Note that we can still spell out any word in H with the folded graph H 0 .
By carrying out the foldings from H to H 0 , we have factored the map H → G into
H → H 0 → G . Since no more folding can be done, it follows by Theorem 17.1 that there
must exist an immersion map from the folded graph H 0 to the bouquet of n circles G .
Now we count the number of edges incident on each vertex of H 0 . Firstly, we discount
the case where there are more than 2n incident. This is because if this is the case, there
must exist a pair of edges with the same orientation into or out of the vertex, and therefore
folding was not completed. This leaves the two possibilities: either the number of edges
incident on each vertex is exactly 2n or there exist vertices with less than 2n edges incident
on it.
If there are exactly 2n edges incident on each vertex of H 0 , then the immersion map
from H 0 to G is a covering map, hence H 0 is a covering graph for G . From here, we can
find the index of the subgroup H of G by counting the number of vertices in H 0 i.e. the
cardinality of the fiber of the base vertex in G .
If there are less than 2n edges incident on one of the vertices of H 0 , then the immersion
is not a covering map. However, we can extend the graph H 0 so that the number of edges
incident on each vertex is equal to 2n, turning it into a covering graph, by adding trees
onto each vertex with less than 2n edges incident on it. This is possible because adding
a tree onto a graph does not change its homotopy type. Then, there are infinitely many
number of vertices in the new graph, hence the index of subgroup H of G is infinite.
Then, from here, we can calculate the rank of H ⊂ G very easily. From Theorem
16.1, we have shown that if G is a group with n generators and the subgroup H of G has
index m, then H has 1 + mn − m generators. Going back to the example, with X the free
group and Y the specified subgroup, we construct Y ; a bouquet of 4 circles with each
edge subdivided to spell out each word of {a2 , b2 , ab, ba} (see Figure 31a). After folding
the graph, we will obtain the graph Y 0 (see Figure 31b).
However, we constructed this graph earlier in Section 16.1. Therefore we know that
this is a covering graph for B2 . Since B2 has only one vertex, both vertices in Y 0 are
mapped to the same vertex. Therefore the cardinality of the fiber of this vertex is 2, hence
the index of the subgroup Y is 2. We can calculate the number of generators (rank) of
the subgroup Y of X by using the formula: 1 + (2 × 2) − 2 = 3.
a b
bb a
a
a
bb a
a
b
(a) Y (b) Folded graph Y 0
Figure 31: Covering map
42
Part IV
Summary
Over the course of this project we explored the deep and meaningful links between topol-
ogy and algebra. In Part I, we explored the connections between covering spaces and
subgroups of a fundamental group, which enabled us to view a topological problem from
an algebraic perspective and vice versa. This is often more convenient to deal with, and
it is possible for us to solve problems in one area using the other, due to the one-to-one
correspondences which exist. For example, by endowing specially-structured spaces with
a topology, we can arrive at conclusions which are otherwise hard (or possibly impossible)
to prove algebraically. A prime example of this was explored in Part III when we proved
that the subgroup of a free group is also free by using graphs. Furthermore, in Part II
we again utilised graph theory to translate surfaces into graphs; thus, we were able to
simplify complex problems to a simpler form. By finding the relationships between these
concepts, we successfully linked two seemingly disparate areas of mathematics.
43
Part V
Bibliography
[1] Topological Graph Theory. J. L. Gross & T. W. Tucker, Wiley Interscience, New
York, 1st Edition, 1987. ISBN: 0-471-04926-3
[2] Classical Topology and Combinatorial Group Theory. J. Stillwell, Springer-Verlag,

New York Heidelberg Berlin, 1st Edition, 1980. ISBN: 0-387-90516-2
[3] Algebraic Topology: An Introduction. W. S. Massey, Harcourt, Brace & World, New
York, 1st Edition, 1967. ISBN: 0-387-90271-6
[4] Applications of Algebraic Topology. S. Lefschetz, Springer-Verlag, New York, 1st Edi-
tion, 1975. ISBN: 0-387-90137-X
[5] A Basic Course in Algebraic Topology. W. S. Massey, Springer, New York, 4th Edi-
tion, 1991. ISBN: 3-540-97430-X
[6] Graphs, Surfaces and Homology. P. J. Giblin, Chapman and Hall, London, 1st Edi-
tion, 1977. ISBN: 0-412-21440-7
[7] Algebraic Topology. A. Hatcher, Cambridge University Press, 1st Edition, 2002.
ISBN: 0-521-79540-0.
[8] Introduction to Group Theory. W. Ledermann and A. J. Weir, Pearson Education,

2nd Edition, 1996. ISBN: 0-582-25954-1.
[9] A Concise Course in Algebraic Topology. J. P. May, The University of Chicago, 1st
Edition, 1999. ISBN: 0-226-51183-9.
[10] Math 872 Algebraic Topology: Lecture Notes. M. Brittenham, University of Nebraska-
Lincoln, 2007. Retrieved from: (http://www.math.unl.edu/$\sim$mbrittenham2/
classwk/872s07/lecnotes/notes2.covering.spaces.pdf) on May 31st 2012.
[11] Algebraic Topology A Computational Approach. T. Kaczynski, Université de Sher-

brooke; K. Mischaikow, Georgia Institute of Technology; M. Mrozek, Jagellonian Uni-
versity, 2000. Retrieved from: (http://thiqaruni.org/mathpdf5/(26).pdf) on June
7th 2012 .
[12] Topology of Finite Graphs. J. R. Stallings, University of California, Berkeley 1983.

Retrieved from: (http://wwwmath.uni-muenster.de/u/linus.kramer/Stallings.pdf)
on June 5th 2012.
[13] Topological Cliques in Random Lifts of Graphs. M. Witkowski, Uniwersytet Adama

Mickiewicza w Poznaniu, 2010. Retrieved from: (http://ssdnm.mimuw.edu.pl/pliki/
prace-studentow/st/pliki/marcin-witkowski-3.pdf) on June 12th 2012.
[14] A Textbook of Topology. H. Seifert & W. Threlfall, Academic Press, New

York/London, 1980. Translation of Lehrbuch der Topologie. H. Seifert & W. Threlfall,
Teubner, Leipzig, 1934. ISBN: 0-126-34850-2
44
[15] Connected Sums. Retrieved from: (http://www.proofwiki.org/wiki/Definition:
Connected_Sum) on 14th June 2012.
[16] Retrieved from: (http://gfx.cs.princeton.edu/proj/sugcon/models/torus.png) on

5th June 2012. Courtesy of S. Rusinkiewicz, D. DeCarlo, A. Finkelstein, and A.
Santella.
[17] Connected Sums. Retrieved from: (http://en.wikipedia.org/wiki/File:Connected_

sum.svg) on 15th June 2012. Released under the GNU Free Documentation License.
Previous authors to be found at aforementioned address.
[18] Topology. J. R. Munkres. Prentice Hall, 2nd Edition, 1975. ISBN: 0-13-178449-8
[19] Topological Spaces and the Fundamental Group . D. R. Wilkins, 2002. Retrieved from:
(http://www.maths.tcd.ie/~dwilkins/Courses/421/421Mich_0203.pdf) on June 12th
2012.‘
[20] A First Course in Algebraic Topology. C. Kosniowski, Cambridge University press,

1980. ISBN: 0-521-29864-4
[21] An Introduction to Metric and Topological Spaces. W. A. Sutherland, Oxford Uni-

versity press, 2009. ISBN: 978-0-19-956308
[22] Compact Riemann Surfaces. J. Jost, Springer-Verlag, 1997. ISBN: 3-540-53334-6
[23] Geometric Integration Theory. H. Whitney, Princeton University Press, 1957. ISBN:
9780691079721
[24] Topology: A Geometric Approach. Terry Lawson, Oxford University press, 20030.
ISBN: 0-19-851597-9
45

Covering Spaces and Graph Theory PDF

Uploaded by

Copyright:

Available Formats

Covering Spaces and Graph Theory PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Covering Spaces and Graph Theory PDF

Uploaded by

Copyright:

Available Formats

Covering Spaces and their Applications to Graph Theory

Ben Ashby 00638208,

1 Introductory Topological Concepts 4

2 The Fundamental Group 5

4 The Lifting Theorems 9

5 Equivalent Covering Spaces and Covering Transformations 16

6 The Universal Covering Space 19

7 Summary of Classification of Covering Spaces 20

II Applications to Graph Theory 22

8 Some Basic Graph Theory 22

10 Results on Polygonal Representations 27

III A More Topological View of Graphs 34

13 Some Useful Theorems 34

14 The Fundamental Group of a Graph 35

1 Introductory Topological Concepts

1.1 Loops and Homotopies

We were first introduced to the idea of a homotopy in the context of “equivalence” of

2 The Fundamental Group

π1 (X, x0 ) = {[f ] | f is a loop based at x0 }

with group operation •.

Definition 2.3 (Simple Connectedness). A space is simply connected if it is path con-

2.1 Induced Homomorphisms

Remark 2.1. It is important to note the following:

The induced homomorphism can be viewed as follows: we examine the effect of a

Figure 1: Looking at the induced homomorphism.

So the term induced is appropriate; the deformation under a continuous function

Definition 3.1 (Covering Space). A covering space of a topological space X is a pair

So — covering maps are examples of local homeomorphisms. We can think of a

Figure 2: Subdividing the plane into squares.

Figure 3: Forming a cylinder. Figure 4: Creating a torus from a cylinder.

Proof. Choose arbitrarily any two points x0 , x1 ∈ X. We se up a bijection between the

Additionally, since p∗ is a homomorphism, we have that the image p∗ (π1 (X, x0 )) is a

4 The Lifting Theorems

Definition 4.1 (Lifting of Functions). Let X, Y and Z be spaces. Suppose that g : Y → Z

Corollary 4.1. Let (X,

Proof. This corollary directly follows from Theorem 4.1.

Figure 5: Path lifting

Figure 6: Lift of a loop

Theorem 4.3 (Monodromy Theorem). Let p : X e → X be a covering map, let H :

Corollary 4.2. Let p : X e → X be a covering map. Given two paths g0 , g1 in X,

Proof. It would suffice to define a homotopy between g0 and g1 . As p ◦ g0 and p ◦ g1 are

Figure 7: Lifting criterion

Now, we prove that ϕ∗ (π1 (Y, y0 )) ⊂ p∗ (π1 (X,

4.1 Homotopy Equivalence and the Homotopy Type of a Space

5.1 Conjugate Subgroups of π1

Lemma 5.1. Conjugacy is an equivalence relation on the set of subgroups of G.

3. If another subgroup G2 is conjugate to G0 , then G2 = p∗ (π1 (X̃, x˜2 )) for some

[3, p.154][24, p.248]

2. Already given g so direct check.

Figure 8: Commutative diagram of a homomorphism.

It is clear that the composition of two homomorphisms is a homomorphism, and, for

2. Identity: The identity map i: X̃ → X̃ is a homeomorphism.

3. Inverse: If ϕ is a deck transformation then ϕ is a homeomorphism so there exists

Proof. This is a special case of Theorem 4.1.

Proof. Follows directly from the previous lemma.

Proof. Obvious consequence of previous lemma.

5.3 Criterion for Equivalence of Covers

Proof. Follows from Theorem 5.1 and Lemma 5.2.

6 The Universal Covering Space