Algebraic Topology

Oscar Randal-Williams
https://www.dpmms.cam.ac.uk/∼or257/teaching/notes/at.pdf

1 Introduction 1
1.1 Some recollections and conventions . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Cell complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Homotopy and the fundamental group 4

2.1 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 The fundamental group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Covering spaces 15
3.1 Covering spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 The fundamental group of the circle and applications . . . . . . . . . . . . 21
3.3 The construction of universal covers . . . . . . . . . . . . . . . . . . . . . 22
3.4 The Galois correspondence . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Some group theory 28

4.1 Free groups and presentations . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 Another view of free groups . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Free products with amalgamation . . . . . . . . . . . . . . . . . . . . . . . 32

5 The Seifert–van Kampen theorem 33

5.1 The Seifert–van Kampen theorem . . . . . . . . . . . . . . . . . . . . . . . 33
5.2 The effect on the fundamental group of attaching cells . . . . . . . . . . . 39
5.3 A refinement of the Seifert–van Kampen theorem . . . . . . . . . . . . . . 41
5.4 The fundamental group of a surface . . . . . . . . . . . . . . . . . . . . . . 42

6 Simplicial complexes 45
6.1 Simplicial complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2 Simplicial approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

7 Homology 54
7.1 Simplicial homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.2 Some homological algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.3 Elementary calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.4 The Mayer–Vietoris sequence . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.5 Continuous maps and homotopy invariance . . . . . . . . . . . . . . . . . 64
7.6 Homology of spheres and applications . . . . . . . . . . . . . . . . . . . . 68
7.7 Homology of surfaces, and their classification . . . . . . . . . . . . . . . . 70
7.8 Rational homology, Euler and Lefschetz numbers . . . . . . . . . . . . . . 75
Last updated December 13, 2016. Corrections to [email protected].
 Chapter 1

Introduction

Algebraic Topology is the art of turning existence questions in topology into existence
questions in algebra, and then showing that the algebraic object cannot exist: this then
implies that the original topological object cannot exist. This procedure usually has a
loss of information, and if the algebraic object does exist it does not typically allow us
to show that the original topological object does.
Many interesting topological problems can be expressed in the following form.

The extension problem. Let X be a topological space, A ⊂ X a subspace, and

f : A → Y a continuous function into a third space Y . Does there exist a continuous
function F : X → Y so that F |A = f ?

The most basic instance of this problem arises when taking X = Dn to be the n-
dimensional disc, A = Y = S n−1 to be the (n − 1)-dimensional sphere, and f = IdS n−1 :
S n−1 → S n−1 to be the identity. The problem then asks whether there is a continuous
function from the disc to its boundary which fixes each point on the boundary. In the
course we shall prove the following theorem.
Theorem. There is no continuous function F : Dn → S n−1 such that the composition

S n−1
incl. / Dn F / S n−1

is the identity.
You may find it intuitively quite clear that there is no such F , but the difficulty in
proving this is apparent: while it is obvious that various naïve choices of F : Dn →
S n−1 do not work, we must show that no choice can work, and in principle there are a
great many potential choices with very little mathematical structure to work with. The
machinery of Algebraic Topology translates proving the theorem above to proving:
Theorem. There is no group homomorphism F : {0} → Z such that the composition

Z
0 / {0} F /Z

is the identity.
I hope you will agree that this algebraic theorem is considerably easier to prove. There
are several other classical results in mathematics which we shall be able to rephrase as
an extension problem—or something like it—and hence solve using the machinery of
Algebraic Topology. One is that the notion of dimension is well-defined:
Theorem. If there is a homeomorphism Rn ∼ = Rm , then n = m.
Another is the fundamental theorem of algebra:

1
2 Chapter 1 Introduction

Theorem. Any non-constant polynomial with coefficients in C has a root in C.

This introduction has already mentioned several times the machinery of Algebraic
Topology. It is a characteristic of this subject that much foundational work must be put
in before any profit can be taken, but the foundational work then renders interesting and
apparently difficult problems almost trivial. Take this as consolation during the drier
parts of the course.

1.1 Some recollections and conventions

In order to write fewer words, we make the following convention.

Definition 1.1.1. A continuous function will be called a map.

We will need to repeatedly justify why various formulas that we write down do indeed
defione maps, i.e. are continuous. Almost all of these situations will be dealt with using
the following convenient lemma.

Lemma 1.1.2 (Gluing lemma). Let f : X → Y be a function between topological spaces,

and C, K ⊂ X be closed subsets such that C ∪ K = X. Then f is continuous if and only
if f |C : C → Y and f |K : K → Y are continuous.

Proof. As a restriction of a continuous map is continuous, the condition is clearly neces-

sary. To show it is sufficient, we use the characterisation of continuity in terms of closed
sets: f is continuous if and only if for each closed subset D ⊂ Y , f −1 (D) is closed.
In the situation of the lemma we have f −1 (D)∩C = (f |C )−1 (D), which is closed in C
as f |C is continuous, and so is closed in X because C is closed in X. Similarly f −1 (D)∩K
is closed in X. But then f −1 (D) = f −1 (D) ∩ (C ∪ K) = (f −1 (D) ∩ C) ∪ (f −1 (D) ∩ K)
is a finite union of closed sets, and so closed.

Later in the course we will need the following standard lemma.

Lemma 1.1.3 (Lesbegue number lemma). Let (X, d) be a metric space which is compact.
For any open cover U = {Uα }α∈I of X there exists a δ > 0 such that each ball Bδ (x) ⊂ X
is contained entirely in some Uα .

Proof. Suppose no such δ exists. Then for each n ∈ N there exist an xn ∈ X such that
B1/n (xn ) is not contained entirely in some Uα . As X is compact the set {xn }n∈N has a
limit point, y. This lies in some Uα , so there exists an r > 0 such that Br (y) also lies in
Uα . But now let N > 0 be such that
1
(i) N < 2r , and

(ii) d(xN , y) < 2r .

Then B1/N (xN ) is contained inside Br (y), and hence inside Uα , a contradiction.
 1.2 Cell complexes 3

1.2 Cell complexes

A convenient way to construct topological spaces, which we shall see interacts well with
the tools of Algebraic Topology, is the notion of attaching cells.

Definition 1.2.1. For a space X and a map f : S n−1 → X, the space obtained by
attaching an n-cell to X along f is the quotient space

X ∪f Dn := (X t Dn )/ ∼

where ∼ is the equivalence relation generated by x ∼ f (x) for x ∈ S n−1 ⊂ Dn .

Definition 1.2.2. A (finite) cell complex is a space X obtained by

(i) starting with a finite set X 0 with the discrete topology, called the 0-skeleton,

(ii) having defined the (n − 1)-skeleton X n−1 , form the n-skeleton X n by attaching a
collection of n-cells along finitely many maps {fα : S n−1 → X n−1 }α∈I , i.e.
!
G
n n−1
X = X t Dα /x ∈ Sαn−1 ⊂ Dαn ∼ fα (x) ∈ X n−1 ,
n

α∈I

(iii) stop at some finite stage k, so X = X k ; this k is the dimension of X.

 Chapter 2

Homotopy and the fundamental group

2.1 Homotopy

As we shall use the unit interval [0, 1] ⊂ R repeatedly, we denote it by I.

Definition 2.1.1 (Homotopy). Let f, g : X → Y be maps1 . A homotopy from f to

g is a map H : X × I → Y such that

H(x, 0) = f (x) and H(x, 1) = g(x).

If such an H exists, we say that f is homotopic to g, and write f ' g. If we wish to

record which homotopy between these maps we have in mind, then we write f 'H g.
If A ⊂ X is a subset, then we say that the homotopy H is relative to A if in addition

H(a, t) = f (a) = g(a) for all a ∈ A and all t ∈ I.

In this case we write f ' g rel A.

Proposition 2.1.2. For spaces X and Y and a subspace A ⊂ X, the relation “homotopic
relative to A" on the set of maps from X to Y is an equivalence relation.

Proof. We must verify three properties.

(i) For a map f : X → Y , define a homotopy H : X × I → Y by H(x, t) = f (x). We

may write this as the composition
proj f
H : X × I −→ X −→ Y

of continuous functions, so it is continuous. It satisfies H(x, 0) = f (x) = H(x, 1) =

H(x, t) for all x and t, so is a homotopy from f to f relative to any subset A ⊂ X.
Thus f ' f rel A.

(ii) Let H : X × I → Y be a homotopy from f to g relative to A. Define H̄(x, t) =

H(x, 1 − t). This is continuous as the map t 7→ 1 − t : I → I is continuous (in fact,
a homeomorphism). This new homotopy satisfies

H̄(x, 0) = H(x, 1) = g(x) H̄(x, 1) = H(x, 0) = f (x)

H̄(a, t) = H(a, 1 − t) = f (a) = g(a) for a ∈ A

and so is a homotopy from g to f relative to A.
1
Recall that a map is a continuous function.

4
 2.1 Homotopy 5

(iii) Let H : X × I → Y be a homotopy from f to g relative to A, and H 0 : X × I → Y

be a homotopy from g to h relative to A. then define a function H 00 : X × I → Y
by the formula (
00 H(x, 2t) 0 ≤ t ≤ 1/2
H (x, t) =
H 0 (x, 2t − 1) 1/2 ≤ t ≤ 1.
This function is well defined, as when t = 1/2 we have

H(x, 2 · 1/2) = H(x, 1) = g(x) = H 0 (x, 0) = H 0 (x, (2 · 1/2) − 1).

Furthermore, by the Gluing Lemma it is continuous, as it restricts to the continuous

function (x, t) 7→ H(x, 2t) on X × [0, 1/2] and to the continuous function (x, t) 7→
H 0 (x, 2t − 1) on X × [1/2, 1].

Definition 2.1.3 (Homotopy equivalence). A map f : X → Y is a homotopy equiv-

alence if there exists a map g : Y → X such that g ◦ f ' IdX and f ◦ g ' IdY . We call
g a homotopy inverse to f .
We say X is homotopy equivalent to Y if a homotopy equivalence f exists, and
write X ' Y .

Example 2.1.4. Let X = S 1 , Y = R2 \ {0}, and i : S 1 → R2 \ {0} be the standard

inclusion. Define a map

r : R2 \ {0} −→ S 1
x
x 7−→ |x| .

Figure 2.1 The retraction of R2 \ {0} to S 1 .

Then r ◦ i = IdS 1 , so we may take the constant homotopy. On the other hand the
x
composition i ◦ r : R2 \ {0} → R2 \ {0} is x →
7 |x| . Define a homotopy

H : (R2 \ {0}) × I −→ R2 \ {0}

x
(x, t) 7−→ t+(1−t)·|x| ,
6 Chapter 2 Homotopy and the fundamental group

t
which is well-defined as if t+(1−t)·|x| = 0 then |x| = t−1 ≤ 0, which is impossible. This
is a homotopy from i ◦ r to IdR2 \{0} , and so i is a homotopy equivalence with homotopy
inverse r; S 1 ' R2 \ {0}.

Example 2.1.5. Let Y = Rn , X = {0}, and i : {0} → Rn be the inclusion. Let

r : Rn → {0} be the only map there is. Then r ◦ i = Id{0} , and i ◦ r : Rn → Rn is the
constant map to 0. Define H : Rn × I → Rn by H(x, t) = t · x, which is a homotopy from
i ◦ r to IdRn . Thus i is a homotopy equivalence.

This property is central in the subject, and gets its own name.

Definition 2.1.6 (Contractible). A space X is contractible if it homotopy equivalent

to the 1-point space: X ' {∗}.

We wish to say that the relation of homotopy equivalence between spaces is an equiv-
alence relation, but first need a tool.

Lemma 2.1.7. Let f0 , f1 : X → Y be homotopic maps, and g0 , g1 : Y → Z be homotopic

maps. Then the maps g0 ◦ f0 , g1 ◦ f1 : X → Z are homotopic.

Proof. Let H be a homotopy from f0 to f1 , and G be a homotopy from g0 to g1 . We will

show that both g0 ◦ f0 and g1 ◦ f1 are homotopic to g0 ◦ f1 , and then use transitivity of
the homotopy relation.

(i) g0 ◦ f0 ' g0 ◦ f1 : Let H 00 : X × I → Z be the map g0 ◦ H (a composition of

continuous functions, so continuous). This gives the desired homotopy.

(ii) g1 ◦ f1 ' g0 ◦ f1 : Let H 000 : X × I → Z be the map H 0 ◦ (f1 × IdI ) (a composition

of continuous functions, so continuous). This gives the desired homotopy.

In particular, this lemma says that given maps

f
i /X ) h /Z
T 5Y
g

such that f ' g, then h ◦ f ' h ◦ g and f ◦ i ' g ◦ i. This is the form in which we will
generally use this lemma.

Proposition 2.1.8. The relation of homotopy equivalence satisfies

(i) X ' X for any space X,

(ii) if X ' Y then Y ' X,

(iii) if X ' Y and Y ' Z then X ' Z.

Proof.

(i) Let f = g = IdX and all the homotopies be constant.

 2.2 Paths 7

(ii) If f : X → Y is a homotopy equivalence with homotopy inverse g, then g is clearly

also a homotopy equivalence (with homotopy inverse f ).

(iii) Suppose that we have maps

f f0
) )
Xi Y i Z
g g0

such that

f ◦ g ' IdY f 0 ◦ g 0 ' IdZ g ◦ f ' IdX g 0 ◦ f 0 ' IdY .

Then applying Lemma 2.1.7 several times we obtain

(g ◦ g 0 ) ◦ (f 0 ◦ f ) = g ◦ (g 0 ◦ f 0 ) ◦ f ' g ◦ IdY ◦ f = g ◦ f ' IdX

and
(f 0 ◦ f ) ◦ (g ◦ g 0 ) = f 0 ◦ (f ◦ g) ◦ g 0 ' f 0 ◦ IdY ◦ g 0 = f 0 ◦ g 0 ' IdZ .

Definition 2.1.9 (Deformation retraction). For a space X and a subspace A, a defor-

mation retraction of X to A is a map r : X → A so that r|A = IdA and i ◦ r ' IdX 2 .

In particular, if there is a deformation retraction of X to A then the inclusion map

i : A ,→ X is a homotopy equivalence with homotopy inverse r: the homotopy i◦r ' IdX
is part of the definition of deformation retraction, and r ◦ i = IdA so we can take the
constant homotopy.
There is a related notion to that of a deformation retraction, which we introduce
now.

Definition 2.1.10 (Retraction). For a space X and a subspace A, a retraction of X

to A is a map r : X → A so that r|A = IdA .

Warning 2.1.11. A retraction need not be a homotopy equivalence! For example, if X

is a non-empty space and x0 ∈ X a point, there is a retraction of X to the subspace {x0 }
given by sending every point of X to x0 . This will only be a homotopy equivalence if X
is contractible.

2.2 Paths

Definition 2.2.1 (Paths and loops). For a space X and points x0 , x1 ∈ X, a path from
x0 to x1 is a map γ : I → X such that γ(0) = x0 and γ(1) = x1 . We write γ : x0 x1 .
If x0 = x1 then we call γ a loop based at x0 .
2
If we insist that i ◦ r ' IdX rel A then this is called a strong deformation retraction. Some
authors call this stronger notion “deformation retraction", so beware.
8 Chapter 2 Homotopy and the fundamental group

If γ is a path from x0 to x1 and γ 0 is a path from x1 to x2 , then we define a new path

γ · γ 0 : I → X, the concatenation of γ and γ 0 , by the formula
(
γ(2t) 0 ≤ t ≤ 1/2
(γ · γ 0 )(t) = 0
γ (2t − 1) 1/2 ≤ t ≤ 1.

(This is continuous by the Gluing Lemma.) This is a path from x0 to x2 . We also define
a path γ −1 : T → X, the inverse of γ, by the formula

γ −1 (t) = γ(1 − t),

which gives a path from x1 to x0 . Finally, we define a path cx0 : I → X, the constant
path at the point x0 , by cx0 (t) = x0 .

If we define a relation ∼ on the space X by

x ∼ y ⇔ there is a path in X from x to y

then the above three constructions (concatenation, inverse, constant paths) show that
this is an equivalence relation.

Definition 2.2.2 (Path components and π0 ). The equivalence classes of ∼ are called
path components, and the set of equivalence classes is denoted π0 (X). If there is a
single equivalence class then we say X is path connected.

Proposition 2.2.3. To a map f : X → Y there is a well-defined associated function

π0 (f ) : π0 (X) −→ π0 (Y )

given by π0 (f )([x]) = [f (x)]. Furthermore,

(i) if f ' g then π0 (f ) = π0 (g),

h k
(ii) for maps A → B → C we have π0 (k ◦ h) = π0 (k) ◦ π0 (h),

(iii) π0 (IdX ) = Idπ0 (X) .

Proof. We must show that the proposed function is well-defined, so let [x] = [x0 ] ∈ π0 (X):
this means that there is a path γ : I → X from x to x0 . Then f ◦ γ : I → Y is a path
from f (x) to f (x0 ), and so [f (x)] = [f (x0 )] ∈ π0 (Y ), as required.
Properties (ii) and (iii) are clear from the definition. For (i), let H : X × I → Y be
a homotopy from f to g. For x ∈ X the map H(x, −) : I → Y is a path from f (x) to
g(x), and so [f (x)] = [g(x)] ∈ π0 (Y ). Thus π0 (f )([x]) = π0 (g)([x]), and this holds for
any x so π0 (f ) = π0 (g) as required.

Corollary 2.2.4. If f : X → Y is a homotopy equivalence then π0 (f ) : π0 (X) → π0 (Y )

is a bijection.
 2.2 Paths 9

Proof. If g : Y → X is a homotopy inverse to f then we have

π0 (f ) ◦ π0 (g) = π0 (f ◦ g) = π0 (IdY ) = Idπ0 (Y )

and similarly
π0 (g) ◦ π0 (f ) = π0 (g ◦ f ) = π0 (IdX ) = Idπ0 (X)
so π0 (g) is inverse to π0 (f ).
Example 2.2.5. The space {−1, 1} with the discrete topology is not contractible, i.e. is
not homotopy equivalent to the one-point space {∗}. This is because continuous paths in
{−1, 1} must be constant, so π0 ({−1, 1}) = {−1, 1} has cardinality 2. But π0 ({∗}) = {∗}
has cardinality 1, so these are not in bijection.
Example 2.2.6. The space R2 does not retract (cf. Definition 2.1.10) on to the subspace
{(−1, 0), (1, 0)}. If we write i : {(−1, 0), (1, 0)} → R2 for the inclusion, and suppose
r : R2 → {(−1, 0), (1, 0)} is a retraction, so r ◦ i = Id{(−1,0),(1,0)} , then we obtain a
factorisation
π0 (i) π0 (r)
π0 ({(−1, 0), (1, 0)}) −→ π0 (R2 ) −→ π0 ({(−1, 0), (1, 0)})

of the identity map. By the previous example π0 ({(−1, 0), (1, 0)}) has cardinality 2, but
R2 is path-connected so π0 (R2 ) has cardinality 1. In particular, π0 (i) cannot be injective,
so π0 (r) ◦ π0 (i) cannot be the identity map.
Definition 2.2.7. Two paths γ, γ 0 : I → X from x0 to x1 are homotopic as paths
if they are homotopic relative to {0, 1} ⊂ I in the sense of Definition 2.1.1. We write
γ ' γ 0 as paths from x0 to x1 .
Lemma 2.2.8. If γ0 ' γ1 as paths from x0 to x1 , and γ00 ' γ10 as paths from x1 to x2 ,
then γ0 · γ00 ' γ1 · γ10 as paths from x0 to x2 .
Proof. Let H be a relative homotopy from γ0 to γ1 , and H 0 be a relative homotopy from
γ00 to γ10 . As H(1, t) = x1 = H 0 (0, t) for all t, these homotopies fit together to a function
H 00 : I × I → X given by
(
H(2s, t) 0 ≤ s ≤ 1/2
H 00 (s, t) = 0
H (2s − 1, t) 1/2 ≤ s ≤ 1

which is continuous by the Gluing Lemma.

10 Chapter 2 Homotopy and the fundamental group

This satisfies

H 00 (−, 0) = (γ0 · γ00 )(−) H 00 (−, 1) = (γ1 · γ10 )(−) H 00 (0, t) = x0 H 00 (1, t) = x2

as required.

Proposition 2.2.9. Let γ0 : x0 x1 , γ1 : x1 x2 , and γ2 : x2 x3 be paths in X.

Then

(i) (γ0 · γ1 ) · γ2 ' γ0 · (γ1 · γ2 ) as paths from x0 to x3 ,

(ii) γ0 · cx1 ' γ0 ' cx0 · γ0 as paths from x0 to x1 ,

(iii) γ0 · γ0−1 ' cx0 as paths from x0 to x0 , and γ0−1 · γ0 ' cx1 as paths from x1 to x1 .

Proof. First contemplate the following figures.

From these, we arrive at the following formulæ.

(i) The homotopy 

4s
γ0 ( t+1 )
 0 ≤ s ≤ t+1
4
H(s, t) = γ1 (4s − 1 − t) t+1
4 ≤ s ≤ t+2
4
γ2 (1 − 4(1−s)
 t+2
) ≤ s ≤ 1

2−t 4

goes from (γ0 · γ1 ) · γ2 to γ0 · (γ1 · γ2 ) as paths from x0 to x3 .

(ii) The homotopy (

2s
γ0 ( t+1 ) 0 ≤ s ≤ t+1
2
H(s, t) = t+1
x1 2 ≤ s ≤ 1,
goes from γ0 · cx1 to γ0 as paths from x0 to x1 .

(iii) The homotopy 

γ0 (2s)
 0 ≤ s ≤ 1−t
2
H(s, t) = γ0 (1 − t) 1−t 1+t
2 ≤ s ≤ 2

γ0 (2 − 2s) 1+t ≤ s ≤ 1

2

goes from γ0 · γ0−1 to cx0 as paths from x0 to x0 .

 2.3 The fundamental group 11

2.3 The fundamental group

Theorem 2.3.1. Let X be a space and x0 ∈ X be a point. Let π1 (X, x0 ) denote the set
of homotopy classes of loops in X based at x0 . Then the rule [γ] · [γ 0 ] := [γ · γ 0 ] and the
element e := [cx0 ] define a group structure on π1 (X, x0 ).

We call the resulting group the fundamental group of X based at x0 .

Proof. Lemma 2.2.8 shows that the proposed composition law is well-defined, and Propo-
sition 2.2.9 shows that the group axioms are satisfied.

Definition 2.3.2. A based space is a pair (X, x0 ) of a space X and a point x0 ∈ X,

called the basepoint. A map of based spaces f : (X, x0 ) → (Y, y0 ) is a continuous
map f : X → Y such that f (x0 ) = y0 . A based homotopy is a homotopy relative to
{x0 } ⊂ X in the sense of Definition 2.1.1.

Proposition 2.3.3. To a based map f : (X, x0 ) → (Y, y0 ) there is a well-defined associ-

ated function π1 (f ) : π1 (X, x0 ) → π1 (Y, y0 ) given by π1 (f )([γ]) = [f ◦ γ]. It satisfies

(i) π1 (f ) is a group homomorphism,

(ii) if f is based homotopic to f 0 then π1 (f ) = π1 (f 0 ),

h k
(iii) for based maps (A, a) → (B, b) → (C, c) we have π1 (k ◦ h) = π1 (k) ◦ π1 (h),

(iv) π1 (IdX ) = Idπ1 (X,x0 ) .

Proof. The proposed function π1 (f ) is well-defined, as if γ ' γ 0 as paths then f ◦γ ' f ◦γ 0

as paths.

(i) Note that f ◦cx0 = cy0 , so π1 (f ) preserves the unit. Also f ◦(γ ·γ 0 ) = (f ◦γ)·(f ◦γ 0 ),
so π1 (f ) preserves the composition law.

(ii) if f ' f 0 rel {x0 } then f ◦ γ ' f 0 ◦ γ rel {0, 1}, as γ({0, 1}) = {x0 }.

(iii) π1 (k ◦h)([γ]) = [k ◦h◦γ] = π1 (k)([h◦γ]) = π1 (k)(π1 (h)([γ])), for any [γ] ∈ π1 (A, a).

(iv) π1 (IdX )([γ]) = [IdX ◦ γ] = [γ] for any [γ] ∈ π1 (X, x0 ).

After having introduced the notation π1 (f ) for the group homomorphism induced by
a based map f , will will now discard it in favour of the shorter notation f∗ .

Proposition 2.3.4 (Change of basepoint). Let u : x0 x1 be a path in X. It induces

a group isomorphism

u# : π1 (X, x0 ) −→ π1 (X, x1 )
[γ] 7−→ [u−1 · γ · u]

satisfying

(i) if u ' u0 as paths from x0 to x1 then u# = u0# ,

12 Chapter 2 Homotopy and the fundamental group

(ii) (cx0 )# = Idπ1 (X,x0 ) ,

(iii) if v : x1 x2 is a path then (u · v)# = v# ◦ u# ,
(iv) if f : X → Y sends x0 to y0 and x1 to y1 then
(f ◦ u)# ◦ f∗ = f∗ ◦ u# : π1 (X, x0 ) −→ π1 (Y, y1 ),
or in other words the square
f∗
π1 (X, x0 ) / π1 (Y, y0 )

u# (f ◦u)#
 f∗ 
π1 (X, x1 ) / π1 (Y, y1 )

commutes: going either way around it from the top left corner to the bottom right
corner yields the same homomorphism.
(v) if x1 = x0 then u# is the automorphism of π1 (X, x0 ) given by conjugation by
[u] ∈ π1 (X, x0 ).
Proof. That u# is a group homomorphism is a consequence of invoking Proposition 2.2.9
several times. Properties (i) – (iii) and (v) are by now standard arguments. For (iv) we
calculate
((f ◦ u)# ◦ f∗ )([γ]) = (f ◦ u)# ([f ◦ γ]) = [(f ◦ u)−1 · (f ◦ γ) · (f ◦ u)]
= [f ◦ (u−1 · γ · u)] = f∗ ([u−1 · γ · u]) = f∗ (u# ([γ])).
Warning 2.3.5. By this proposition the fundamental groups based at different points
in a path connected space X are isomorphic, but they are not canonically isomorphic!
One has to choose a path between their basepoints to obtain an isomorphism.
Thus it makes sense to say “the fundamental group is trivial / abelian / has group-
theoretic property P" without referring to a basepoint. But it does not make sense
to have an element a0 ∈ π1 (X, x0 ) and say “let a1 ∈ π1 (X, x1 ) be the corresponding
element".
If H : X × I → Y is a homotopy from f to g, and x0 ∈ X is a basepoint, then
u := H(x0 , −) : I → Y is a path from f (x0 ) to g(x0 ) in Y . We can ask about the
relationship between the three homomorphisms
f∗ : π1 (X, x0 ) −→ π1 (Y, f (x0 ))
g∗ : π1 (X, x0 ) −→ π1 (Y, g(x0 ))
u# : π1 (Y, f (x0 )) −→ π1 (Y, g(x0 )),
which fit into the diagram
π1 (Y, f (x0 ))
7
f∗
∼
= u# (2.3.1)
g∗ 
π1 (X, x0 ) / π1 (Y, g(x0 )).
 2.3 The fundamental group 13

Lemma 2.3.6. u# ◦ f∗ = g∗ , or in other words the triangle (2.3.1) commutes.

Proof. For a loop γ : I → X based at x0 consider the composition

γ×Id H
F : I × I −→I X × I −→ Y.

Consider the path `+ : I → I × I from (0, 1) to (1, 1) given by s 7→ (s, 1), and the path
`− : I → I × I from (0, 1) to (1, 1) given by concatenating the paths

s 7→ (0, 1 − s) s 7→ (s, 0) s 7→ (1, s).

These are paths between the same points, and we can define a homotopy

L(s, t) = t · `− (s) + (1 − t) · `+ (s)

between them, using the vector space structure on R2 ⊃ I × I and the fact that I × I is
convex. This shows that `+ ' `− as paths from (0, 1) to (1, 1).
Hence F ◦ `+ ' F ◦ `− as paths from F (0, 1) = H(x0 , 1) = g(x0 ) to F (1, 1) =
H(x0 , 1) = g(x0 ). But [F ◦ `+ ] = [g ◦ γ] and [F ◦ `− ] = [u−1 · (f ◦ γ) · u], as required.

Theorem 2.3.7. Let f : X → Y be a homotopy equivalence, and x0 ∈ X. Then

f∗ : π1 (X, x0 ) −→ π1 (Y, f (x0 ))

is an isomorphism.

Proof. Let g : Y → X be a homotopy inverse to f , so f ◦ g 'H IdY and g ◦ f 'H 0 IdX .

Let u0 : I → X be the path u0 (t) = H 0 (x0 , 1 − t), going from x0 to g ◦ f (x0 ). Then
applying Lemma 2.3.6 gives
f∗ g∗
u0# = (g ◦ f )∗ = g∗ ◦ f∗ : π1 (X, x0 ) −→ π1 (Y, f (x0 )) −→ π1 (X, g ◦ f (x0 )),

which is an isomorphism: thus the map f∗ we are studying is injective, and the map

g∗ : π1 (Y, f (x0 )) −→ π1 (X, g ◦ f (x0 )) (2.3.2)

is surjective. If can show that g∗ is also injective then we are done, as then it is an
isomorphism and so f∗ = (g∗ )−1 ◦ u0# is also an isomorphism.
To show that g∗ is injective we consider the path u : I → Y given by u(t) =
H(f (x0 ), 1 − t), going from f (x0 ) to f ◦ g ◦ f (x0 ). Applying Lemma 2.3.6 again gives
g∗
u# = (f ◦ g)∗ = f∗ ◦ g∗ : π1 (Y, f (x0 )) −→ π1 (X, g ◦ f (x0 )) −→ π1 (Y, f ◦ g ◦ f (x0 )),

which is an isomorphism. (Note the second map here is induced by f but is not the map
f∗ we are studying, as the basepoints involved are different.) Thus the map (2.3.2) is
also injective, as required.

Definition 2.3.8. A space X is simply connected if it is path connected and its

fundamental group is trivial an some (and hence every) basepoint.
14 Chapter 2 Homotopy and the fundamental group

Example 2.3.9. Any contractible space is simply connected.

Lemma 2.3.10. A space X is simply connected if and only if for each pair of points
x0 , x1 ∈ X there is a unique homotopy class of path between them.

Proof. Suppose X is simply connected, and let x0 , x1 ∈ X. As X is path connected

there exists a path from x0 to x1 . If γ and γ 0 are two such paths, then γ −1 · γ 0 is a loop
based at x0 and so [γ −1 · γ 0 ] ∈ π1 (X, x0 ) = {[cx0 ]}. Thus there is a homotopy of paths
γ −1 · γ 0 ' cx0 . Preconcatenating with γ, there is a homotopy of paths from γ 0 to γ, as
required.
Now suppose X has the property described in the statement of the lemma. In particu-
lar there exists a path between any pair of points, so X is path connected. Furthermore,
if γ is a loop based at x0 then so is cx0 , and so these are homotopic as paths. Thus
[γ] = [cx0 ] ∈ π1 (X, x0 ), and so π1 (X, x0 ) is trivial.
 Chapter 3

Covering spaces

3.1 Covering spaces

Definition 3.1.1. A covering space of a space X is a pair (X, e p) of a space X

e and a
map p : Xe → X such that for any x ∈ X there is an open neighbourhood U of x such that
−1
p (U ) is a disjoint union of open sets of Y each of which is mapped homeomorphically
onto U . In this situation we call p a covering map.

For such an open neighbourhood U , we shall usually write p−1 (U ) = α∈I Vα where
`
each p|Vα : Vα → U is a homeomorphism. We call these open sets U evenly covered.

Figure 3.1 The generic picture of an evenly covered set U .

Example 3.1.2. A homeomorphism is a covering map.

e → X and q : Ye → Y are covering maps, so is the product
Example 3.1.3. If p : X
p×q :X ×Y →X ×Y.
e e

Example 3.1.4. Let S 1 ⊂ C be the set of complex numbers of modulus 1, and consider
the map p : R → S 1 given by p(t) = e2πit . For the set Uy>0 := {x + iy ∈ S 1 | y > 0} we
have a
p−1 (Uy>0 ) = (j, j + 12 ),
j∈Z

the disjoint union of Z-many open intervals. Furthermore

p| 1 : (j, j + 12 ) −→ Uy>0
(j,j+ 2 )

is a continuous bijection, with inverse

Uy>0 −→ (j, j + 12 )
x + iy 7−→ j + arccos(x)/2π.

15
16 Chapter 3 Covering spaces

Similarly with Uy<0 , Ux>0 and Ux<0 , and these four open sets cover S 1 . Thus p is a
covering map.

Example 3.1.5. Let S 1 ⊂ C, and p : S 1 → S 1 be given by p(z) = z n . Let η := e2πi/n ,

and for y ∈ S 1 let ξ be some nth root of y. Then

p−1 (y) = {ξ, ξ · η, ξ · η 2 , . . . , ξ · η n−1 }

and p−1 (S 1 \ {y}) is the complement of this set. Let

V0 := {z ∈ S 1 | 0 < arg(z/ξ) < 2π/n}

`n−1
and Vi := V0 · η i for i = 0, 1, 2, . . . , n − 1, so p−1 (s1 \ {y}) = i=0 Vi . Then p|Vi : Vi →
S 1 \ {y} is a homeomorphism.

Figure 3.2 The set V0 for n = 4.

Example 3.1.6. Let S 2 ⊂ R3 be the unit sphere, and ∼ be the equivalence relation on
S 2 generated by x ∼ −x, with quotient space RP2 := S 2 / ∼ (the real projective plane)
and quotient map p : S 2 → RP2 . Let V := {(x, y, z) ∈ s2 | z 6= 0}, an open set, and
U = p(V ). Then p−1 (U ) = V`, and so U ⊂ RP2 is open.
Furthermore, V = Vz<0 Vz>0 , and we claim that p|Vz>0 : Vz>0 → U is a homeo-
morphism (similarly for Vz<0 of course). To see this, we shall construct an inverse. By
the definition of the quotient topology, a continuous map U → Y for some space Y is
the same as a continuous map V → Y which in constant on ∼-equivalence classes. Thus
define the map

t : V −→ Vz>0
(
(x, y, z) if z > 0,
(x, y, z) 7−→
(−x, −y, −z) if z < 0.

This is clearly continuous and constant on ∼-equivalence classes, and so defines a con-
tinuous function t̄ : U → Vz>0 . It is inverse to p|Vz>0 , and so p is a covering map.

Our interest in covering spaces comes from the following important property they
have with respect to paths and homotopies.
 3.1 Covering spaces 17

Definition 3.1.7 (Lifting). Let p : X e → X be a covering space, and f : Y → X be a

map. A lift of f along p is a map f˜ : Y → X
e such that p ◦ f˜ = f .

Lemma 3.1.8 (Uniqueness of lifts lemma). Let p : X e → X be a covering space, f : Y →

˜ ˜
X be a map, and f0 , f1 : Y → X be two lifts of f . Then the set
e

S := {y ∈ Y | f˜0 (y) = f˜1 (y)}

is both open and closed in Y . Thus if Y is connected then either S = ∅ or S = Y .
Proof. Let us first show that S is open. Let y ∈ S`and U be an open neighbourhood
of f (s) ∈ X which is evenly covered, so p−1 (U ) = α∈I Vα . Then f˜0 (y) and f˜1 (y) are
equal so lie in the same Vβ . Thus, on N := f˜0−1 (Vβ ) ∩ f˜1−1 (Vβ ) we have
p|Vβ ◦ f˜0 |N = f |N = p|Vβ ◦ f˜1 |N

but p|Vβ is a homeomorphism, so f˜0 |N = f˜1 |N and y ∈ N ⊂ S. Thus S is open.

Let us now show that S is closed. Let y ∈ S and suppose f˜0 (y) 6= f˜` 1 (y). Let U be
an open neighbourhood of f (y) ∈ X which is evenly covered; p−1 (U ) = α∈I Vα . Thus
f˜0 (y) ∈ Vβ and f˜1 (y) ∈ Vγ for some β, γ ∈ I. As each Vα is mapped homeomorphically
to U , and f˜0 (y) 6= f˜1 (y), we must have β 6= γ. Now f˜0−1 (Vβ ) ∩ f˜1−1 (Vγ ) is an open
neighbourhood of y, so muct intersect S by the definition of closure. But then Vβ and
Vγ must intersect, a contradiction.

Lemma 3.1.9 (Homotopy lifting lemma). Let p : X e → X be a covering space, H :

Y × I → X be a homotopy from f0 to f1 , and f˜0 be a lift of f0 . Then there exists a
e :Y ×I →X
unique homotopy H e such that

(i) H(−,
e 0) = f˜0 (−), and
(ii) p ◦ H
e = H.

{Uα }α∈I be an open cover of X by sets which are evenly covered, and write
Proof. Let `
∼
p−1 (Uα ) = β∈Iα Vβ with p|Vβ → Uα .
Then {H −1 (Uα )}α∈I is an open cover of Y × I. For each y0 ∈ Y this restricts to an
open cover of the compact space {y0 } × I, and by the Lesbegue number lemma there is
a N = N (y0 ) such that each path
H|{y0 }×[i/N,(i+1)/N ] : {y0 } × [ Ni , i+1
N ] −→ X

lies entirely inside some Uα . In fact, as {y0 }×I is compact there is an open neighbourhood
Wy0 of y0 such that H(Wy0 × [ Ni , i+1 N ]) lies entirely inside some Uα for each i. We are
now in the situation depicted in Figure 3.3.
We obtain a lift H|
e Wy ×I of H|Wy ×I as follows:
0 0

(i) We have H| 1 : Wy0 × [0, N1 ] −→ Uα and a lift f˜0 |Wy0 : Wy0 → X

e with image
Wy0 ×[0, N ]
in some Vβ which is mapped homeomorphically to Uα via p|Vβ , so we can let

H|
e 1 := (p|Vβ )−1 ◦ H| 1 : Wy0 × [0, N1 ] −→ Vβ ⊂ X.
e
Wy0 ×[0, N ] Wy0 ×[0, N ]
18 Chapter 3 Covering spaces

Figure 3.3

(ii) Proceed in the same way, lifting the (short) homotopy H| 1 2 starting at
Wy0 ×[ N , N ]
H|
e 1 , and so on.
Wy0 ×{ N }

At the end of this process, we obtain a map H| e Wy ×I lifting H|Wy ×I and extending
0 0
f˜0 |Wy0 .
We may do this for every point y0 ∈ Y , so it is now enough to check that on (Wy0 ×
I) ∩ (Wy1 × I) = (Wy0 ∩ Wy1 ) × I the two lifts we have constructed agree. The two
lifts do agree on (Wy0 ∩ Wy1 ) × {0}, as here they both restrict to f0 |Wy0 ∩Wy1 , and so
by Lemma 3.1.8 they agree on an open and closed subset of (Wy0 ∩ Wy1 ) × I containing
(Wy0 ∩ Wy1 ) × {0}. Restricting to each {y} × I we must obtain an open and closed subset
containing {y} × {0}, and so the whole of {y} × I (as I is connected): thus the two maps
agree on the whole of (Wy0 ∩ Wy1 ) × I.

The following corollary is obtained by taking Y = {∗} in the homotopy lifting lemma.

Corollary 3.1.10 (Path lifting). Let p : X e → X be a covering space, γ : I → X be a

path, and x̃0 ∈ X
e be such that p(x̃0 ) = γ(0). Then there exists a unique path γ̃ : I → X
e
such that

(i) γ̃(0) = x̃0 , and

(ii) p ◦ γ̃ = γ.

Corollary 3.1.11. Let p : X e → X be a covering space, γ, γ 0 : I → X be paths from x0

to x1 , and γ̃, γ̃ 0 : I → X
e be lifts starting at x̃0 ∈ p−1 (x0 ). If γ ' γ 0 as paths, then γ̃ ' γ̃ 0
as paths; in particular γ̃(1) = γ̃ 0 (1).

Proof. Let H : I × I → X be a homotopy of paths from γ to γ 0 , and lift it starting at γ̃

to obtain a homotopy He : I × I → X.
e Then

(i) H(−,
e 1) is a lift of γ 0 starting at x̃0 , so is equal to γ̃ 0 ,

(ii) H(0, −) and H(1, −) are constant paths, so their lifts H(0,
e −) and H(1,
e 0) are too,

e is a homotopy of paths from γ̃ to γ̃ 0 .

and hence H
e → X be a covering space, and X be path connected. Then
Corollary 3.1.12. Let p : X
−1
the sets p (x) for x ∈ X are all in bijection.
 3.1 Covering spaces 19

Proof. Let γ : I → X be a path from x0 to x1 , and for each y0 ∈ p−1 (x0 ) let γ̃y0 be the
unique lift of γ starting at y0 . Define the function

γ∗ : p−1 (x0 ) −→ p−1 (x1 )

y0 7−→ γ̃y0 (1).

Similarly, the inverse path γ −1 defines a function (γ −1 )∗ : p−1 (x1 ) → p−1 (x0 ). Now for
any y0 ∈ p−1 (x0 ) we calculate

(γ −1 )∗ ◦ γ∗ (y0 ) = endpoint of lift of γ −1 starting at γ̃y0 (1)

= endpoint of lift of γ −1 · γ starting at y0
= endpoint of lift of cx0 starting at y0
= y0 ,

so (γ −1 )∗ is a left inverse for γ. The analogous calculation with the other composition
shows it is also a right inverse.
e → X is an n-sheeted cover (for n ∈
Definition 3.1.13. Say a covering map p : X
−1
N ∪ {∞}) if each p (x) has cardinality n.
We now come to the first connection between covering spaces and the fundamental
group.
Lemma 3.1.14. Let p : X e → X be a covering map, x0 ∈ X, and x̃0 ∈ p−1 (x0 ). Then
e x̃0 ) → π1 (X, x0 ) is injective.
the map p∗ : π1 (X,
Proof. Let γ : I → X e be a loop based at x̃0 such that p∗ ([γ]) = [cx ], so p ◦ γ ' cx as
0 0
paths. Let H be a homotopy of paths from p ◦ γ to cx0 , and lift it starting at the lift γ
of p ◦ γ. This gives a homotopy of paths H e from γ to a lift of cx , which must be cx̃ ,
0 0
and hence [γ] = [cx̃0 ].
In the proof of Corollary 3.1.12 we constructed, for a covering map p : X e → X and
−1 −1
path γ : I → X from x0 to x1 , a function γ∗ : p (x0 ) → p (x1 ). By Corollary 3.1.11
this function only depends on the homotopy class of γ as a path. In particular, this
construction defines a (right) action • of the group π1 (X, x0 ) on the set p−1 (x0 ).
e → X be a covering map, and X be path connected.
Lemma 3.1.15. Let p : X
(i) π1 (X, x0 ) acts transitively on p−1 (x0 ) if and only if X
e is path connected.

(ii) The stabiliser of y0 ∈ p−1 (x0 ) is the subgroup

e y0 ) → π1 (X, x0 )) ≤ π1 (X, x0 ).
Im(p∗ : π1 (X,

(iii) If X
e is path connected then there is a bijection

π1 (X, x0 )
−→ p−1 (x0 )
p∗ π1 (X,
e y0 )

induced by acting on the point y0 .

20 Chapter 3 Covering spaces

Proof. For (i), let us first show that if X

e is path connected then π1 (X, x0 ) acts transitively
−1 0 −1
on p (x0 ). Let y0 , y0 ∈ p (x0 ), and let γ : I → X e be a path between them, which
exists by hypothesis. Then p ◦ γ is a loop in X based at x0 , and γ is the lift of it starting
at y0 . Thus
y0 • [p ◦ γ] = y00 ,
as required.
Now let us show that if π1 (X, x0 ) acts transitively on p−1 (x0 ) then X e is path con-
nected. Suppose, for a contradiction, that y0 , z0 ∈ X e lie in different path components.
Choose a path γ : I → X from p(y0 ) to p(z0 ), and lift it starting at y0 . This gives a
path in X e from y0 to a z 0 , and p(z 0 ) = p(z0 ). As z 0 is in the path component of y0 , it is
0 0 0
not in the path component of z0 . But as π1 (X, x0 ) acts transitively on p−1 (x0 ) there is
a [γ 0 ] ∈ π1 (X, x0 ) such that z0 • [γ 0 ] = z00 , and so the lift of γ 0 starting at z0 ends at z00 ,
which is a contradiction.
For (ii), suppose that y0 • [γ] = y0 , so the lift γ̃ of γ starting at y0 also ends at y0 .
Thus γ̃ is a loop based at y0 , and γ = p ◦ γ̃, so

[γ] = p∗ ([γ̃]) ∈ Im(p∗ : π1 (X,

e y0 ) → π1 (X, x0 )).

Conversely, if [γ 0 ] ∈ π1 (X,
e y0 ) then γ 0 is a (and so the) lift of p ◦ γ 0 starting at y0 . It ends
at y0 too, and so y0 • [p ◦ γ 0 ] = y0 .
For (iii), we simply apply the Orbit-Stabiliser theorem from the theory of group
actions.
e → X is a universal cover if X
Definition 3.1.16. We say that a covering map p : X e
is simply connected.

The following connection between universal covers and the fundamental group is
immediate from this definition and Lemma 3.1.15.
e → X is a universal cover, then each x̃0 ∈
Corollary 3.1.17. If a covering map p : X
−1
p (x0 ) determines a bijection
∼
` : π1 (X, x0 ) −→ p−1 (x0 )
[γ] 7−→ x̃0 • [γ].

We wish to use this bijection to understand the fundamental group, but it must be
used carefully, as π1 (X, x0 ) is a group but p−1 (x0 ) does not have a group structure.
Unravelling the definitions shows that the multiplication law on p−1 (x0 ) induced by this
bijection can be described as follows: for y0 , z0 ∈ p−1 (x0 ), the product `(`−1 (y0 )·`−1 (z0 ))
is obtained by

(i) choose a path γ̃ : I → X

e from x̃0 to z0 (this is unique up to homotopy, by Lemma
2.3.10),

(ii) let γ be the lift of the loop p ◦ γ̃ starting at y0 ,

(iii) then `(`−1 (y0 ) · `−1 (z0 )) = γ(1).

 3.2 The fundamental group of the circle and applications 21

3.2 The fundamental group of the circle and applications

We will now use this description to compute the group π1 (S 1 , 1).

Theorem 3.2.1. Let u : I → S 1 be the loop given by t 7→ e2πit , which is based at

1 ∈ S 1 ⊂ C. Then there is a group isomorphism π1 (S 1 , 1) ∼
= (Z, +) sending [u] to 1.

Proof. Recall from Example 3.1.4 that we constructed a covering space p : R → S 1 via
p(t) = e2πit . The space R is contractible, and so simply connected, so this is a universal
cover. As a set we have p−1 (1) = Z ⊂ R, and we may choose x̃0 = 0 ∈ Z = p−1 (1) as a
basepoint. Acting on this basepoint then gives a bijection

` : π1 (S 1 , 1) −→ Z.

To compute `−1 (m) we can choose the path ũm : I → R given by ũn (t) = mt, which
goes from 0 to m, so that `−1 (m) = [p ◦ ũm ]. Then we find that

`(`−1 (n) · `−1 (m)) = endpoint of lift of [p ◦ ũm ] starting at n

= endpoint of t 7→ n + mt : I → R
=n+m

Thus the multiplication law on Z = p−1 (1) induced by ` is the usual addition of integers.
Furthermore, `([u]) = 1.

In the following theorem, we consider D2 ⊂ C to be the set of complex numbers of

modulus at most 1.

Theorem 3.2.2. The disc D2 does not retract to S 1 ⊂ D2 .

Proof. Suppose that r : D2 → S 1 is a retraction, and let i : S 1 → D2 be the inclusion.

By definition of retraction, r ◦ i = IdS 1 , and so taking fundamental groups based at
1 ∈ S 1 ⊂ D2 we obtain a factorisation
i r
Z∼
= π1 (S 1 , 1) −→ π1 (D2 , 1) −→ π1 (S 1 , 1) ∼
∗ ∗
=Z

of the identity map. But D2 is contractible, and so π1 (D2 , 1) = {[c1 ]} is the trivial
group. This is a contradiction, as the identity map of Z cannot factor through a trivial
group.

Corollary 3.2.3 (Brouwer’s fixed point theorem, 1909). Every continuous map f : D2 →
D2 has a fixed point.

Proof. Suppose f is a map without fixed points. Then define a function r : D2 → S 1 by

sending x to the first point on S 1 hit by the ray from f (x) through x. (This definition
makes sense only because we know that f (x) 6= x.) One may convince oneself that r
is continuous if f is. If x ∈ S 1 ⊂ D2 then such a ray will first hit S 1 at x, and so
r(x) = x. Thus r is a continuous retraction of D2 to S 1 , but this is impossible by
Theorem 3.2.2.
22 Chapter 3 Covering spaces

Theorem 3.2.4 (Fundamental theorem of algebra). Every nonconstant polynomial over

C has a root in C.

Proof. Suppose not, and let p(z) = z n +a1 z n−1 +· · ·+an be a (monic) polynomial having
no root. Fix (
|a1 | + |a2 | + · · · + |an |
r≥
1
and note that on the circle |z| = r we have the estimate

|z|n = rn−1 · r > |z|n−1 · (|a1 | + |a2 | + · · · + |an |) ≥ |a1 z n−1 + · · · + an |

Thus for t ∈ [0, 1] the polynomial

pt (z) := z n + t · (a1 z n−1 + · · · + an )

has no roots on the circle |z| = r.

Consider the homotopy of loops in S 1 ⊂ C

pt (r · e2πis )/pt (r)

F (s, t) = ,
|pt (r · e2πis )/pt (r)|

which is well-defined as the loop s 7→ pt (r · e2πis ) is never 0. When t = 0 this gives the
loop s 7→ e2πins , which represents n ∈ π1 (S 1 , 1). When t = 1 this gives the loop

p(r · e2πis )/p(r)

fr (s) :=
|p(r · e2πis )/p(r)|

in S 1 ⊂ C, so we must have [fr ] = n ∈ π1 (S 1 , 1).

As p is assumed to have no roots, by varying r we see that fr is homotopic to f0 ,
which is constant: thus [fr ] = 0 ∈ π1 (S 1 , 1). Hence n = 0, and so the polynomial p is
constant.

3.3 The construction of universal covers

It practice we can often directly construct a universal cover of a given space X, but for
theoretical work we need a general construction.

Observation 1. Suppose that p : X e → X is a universal cover, x0 ∈ X is a basepoint,

and U 3 x0 is a neighbourhood which is evenly covered, so p−1 (U ) = α∈I Vα and
`
each p|Vα : Vα → U is a homeomorphism. Fix some α. Then for any loop γ : I → U
based at x0 there is a lift to a loop γ̃ : I → Vα based at x̃0 = Vα ∩ p−1 (x0 ). Thus the
homomorphism
e x̃0 ) −→ π1 (X, x0 )
p∗ : π1 (X,
sends [γ̃] to [γ]. But X
e is simply connected, being a universal cover, so [γ̃] = [cx̃ ] and
0
so [γ] = [cx0 ].
Thus if X has a universal cover, then any point x ∈ X has a neighbourhood Ux
so that any loop in Ux based at x is contractible in X. This property is called being
 3.3 The construction of universal covers 23

semilocally simply connected, and we have just argued that it is necessary in order
for a space to have a universal cover.

Observation 2. Suppose that p : X e → X is a universal cover and x0 = p(x̃0 ) is a

basepoint. As X is simply connected, for any point y ∈ X
e e there is a unique homotopy
class of paths [α] from x̃0 to y, and [p ◦ α] gives a preferred homotopy class of path from
x0 to p(y). Thus we can recover y as the endpoint of the lift of [p ◦ α] starting at x̃0 ,
which defines a bijection
 
e −→ homotopy classes of paths in X
X .
starting at x0 and ending anywhere

We may use this bijection in order to construct a universal cover, when we don’t yet have
one.

Theorem 3.3.1. Let X be path connected, locally path connected1 , and semilocally simply
e → X with X
connected. Then there exists a covering map p : X e simply connected.

Proof. (Non examinable.) As a set, define

 
homotopy classes of paths in X
X
e :=
starting at x0 and ending anywhere

with the function p : Xe → X given by p([γ]) = γ(1). We must construct a topology

on X, show that p is continuous in this topology, show that p is a covering map in this
e
topology, and show that X e is simply connected.
Let us first construct the topology on X.
e Consider the set
 
U is open in X, is path connected, and
U := U ⊂X .
π1 (U, x) → π1 (X, x) is trivial for all x ∈ U

Claim: U is a basis for the topology on X.

Proof of claim. Let V be an open neighbourhood of a point x ∈ X.

(i) As X is semilocally simply connected there is a neighbourhood U 0 3 x such that

π1 (U 0 , x) → π1 (X, x) is trivial.

(ii) As X is locally path connected there is an open neighbourhood x ∈ U ⊂ V ∩ U 0

which is path connected.

(iii) The map π1 (U, x) → π1 (X, x) factors through the trivial map π1 (U 0 , x) → π1 (X, x),
so is trivial.
1
Recall from Example Sheet 1 that X is locally path connected if for every point x ∈ X and every
neighbourhood U 3 x, there exists a smaller neighbourhood x ∈ V ⊂ U such that V is path connected.
24 Chapter 3 Covering spaces

(iv) Let y ∈ U be another point, and u : I → U a path from x to y. Then we have a

commutative square
π1 (U, y) / π1 (X, y)

u# u#
 
trivial /
π1 (U, x) π1 (X, x)
where the vertical maps are isomorphisms, so the top map is trivial too.

Thus U is a neighbourhood of x contained in V which is in the collection U, and hence

U is indeed a basis for the topology on X.

Note that if x, y ∈ U ∈ U then there exists a path from x to y in U , and all such
paths are homotopic in X. For a [α] ∈ Xe and a U ∈ U such that α(1) ∈ U , define

[β] = [α · α0 ] for some path

 
(α, U ) := [β] ∈ X
e .
α0 in U starting at α(1)

Claim: These sets form a basis for a topology on X.

e

Proof of claim. We must check that [β] ∈ (α0 , U0 ) ∩ (α1 , U1 ) has a neighbourhood of this
form.
Let W ⊂ U0 ∩ U1 be a neighbourhood of β(1) ∈ U0 ∩ U1 in the collection U, which
exists as U is a basis for the topology on X. Then (β, W ) is a neighbourhood of [β],
and it is enough to show that (β, W ) ⊂ (α0 , U0 ) ∩ (α1 , U1 ). If [γ] ∈ (β, W ) then it is a
concatenation of β and a path in W , but then it is a concatenation of α0 and a path in
U0 too; similarly, it is a concatenation of α1 and a path in U1 .

We give X e the topology generated by this basis.

Let us now show that p is continuous in this topology; as U is a basis for the topology
on X, it is enough to show that p−1 (U ) ⊂ X
e is open for each U ∈ U. But if [α] ∈ p−1 (U ),
so α is a path starting at x0 and ending in U , then [α] ∈ (α, U ) ⊂ p−1 (U ).
Let us now show that p is a covering map. We first claim that each map

p|(α,U ) : (α, U ) → U

is a homeomorphism. It is surjective as U is path connected, and hence every point in

U may be reached by a path from α(1). If [β], [β 0 ] ∈ (α, U ) are homotopy classes of
path which end at the same point, and each may be obtained from α by concatenation
of a path in U , then they differ by concatenation of a loop in U . As every loop in
U is contractible in X, it follows that [β] = [β 0 ] and so the map is injective. Finally
p((γ, V )) = V , so p is an open map, and hence p|(α,U ) is too.
We now claim that p−1 (U ) is partitioned into sets of the form (α, U ). We have already
show that it is covered by sets of this form, so it remains to show that any two such sets
are either disjoint or equal. Thus suppose [γ] ∈ (α, U )∩(β, U ). Then [γ] = [α·α0 ] = [β·β 0 ]
for paths α0 and β 0 in U , and so [α] = [β · β 0 · (α0 )−1 ] and β 0 · (α0 )−1 is a path in U , so
[α] ∈ (β, U ). But then if [δ] ∈ (α, U ), so [δ] = [α · α00 ], then [δ] = [β · β 0 · (α0 )−1 · α00 ]
 3.4 The Galois correspondence 25

and so [δ] ∈ (β, U ). Thus (α, U ) ⊂ (β, U ), and the reverse inclusion follows by the same
argument.
Finally, we must show that X e is simply connected. The fundamental observation is
e starting at [cx ] ∈ X
that if γ : I → X is a path starting at x0 , then its lift to X 0
e is the
path
s 7−→ [t 7→ γ(st)] : I −→ X,e

which ends at the point [γ] ∈ X. e Thus if a loop γ in X based at x0 lifts to a loop in X e
e x̃0 )) = {e} ≤ π1 (X, x0 ). But
based at [cx0 ] then [γ] = [cx0 ], which shows that p∗ (π1 (X,
then by Lemma 3.1.14 the group π1 (X, x̃0 ) must be trivial.
e

3.4 The Galois correspondence

We have seen in Lemma 3.1.14 that if p : X e → X is a path connected covering space,

−1
x0 ∈ X and x̃0 ∈ p (x0 ), then p∗ : π1 (X, e x̃0 ) → π1 (X, x0 ) is injective, and we obtain a
subgroup p∗ (π1 (X, x̃0 )) ≤ π1 (X, x0 ).
e
If x̃00 ∈ p−1 (x0 ) is another point, then choose a path γ from x̃0 to x̃00 . Then p ◦ γ is a
loop in X based at x0 , so we have an element [p ◦ γ] ∈ π1 (X, x0 ), and

[p ◦ γ]−1 · p∗ (π1 (X, e x̃0 ))

e x̃0 )) · [p ◦ γ] = p∗ (π1 (X,
0

i.e. the subgroups obtained from these two basepoints in X

e are conjugate. Thus fixing a
based space (X, x0 ) we obtain functions
 
based covering maps
e x̃0 ) → (X, x0 ) −→ {subgroups of π1 (X, x0 )}
p : (X,

and  
e → X −→ conjugacy classes of
n o
covering maps p : X .
subgroups of π1 (X, x0 )
We wish to show that dividing out by an appropriate equivalence relation on the left
hand side turn both of these functions into bijections.

Proposition 3.4.1 (Surjectivity). Suppose that X is path connected, locally path con-
nected, and semilocally simply connected. Then for any subgroup H ≤ π1 (X, x0 ) there is
e x̃0 ) → (X, x0 ) such that p∗ (π1 (X,
a based covering map p : (X, e x̃0 )) = H.

Note that taking H to be the trivial group, this proposition implies the existence of
a universal cover. Thus it requires all the technical hypotheses of Theorem 3.3.1.

Proof. Let q : X → X be the universal cover constructed in Theorem 3.3.1, whose

underlying set consists of the homotopy classes of paths in X starting at x0 . Define a
relation ∼H on X by

[γ] ∼H [γ 0 ] if γ(1) = γ 0 (1) and [γ · (γ 0 )−1 ] ∈ H ⊂ π1 (X, x0 ).

(i) [γ] ∼H [γ] as [cx0 ] ∈ H,

26 Chapter 3 Covering spaces

(ii) if [γ] ∼H [γ 0 ] then [γ · (γ 0 )−1 ] ∈ H, and so [γ 0 · γ −1 ] ∈ H, and hence [γ 0 ] ∼H [γ],

(iii) if [γ] ∼H [γ 0 ] ∼H [γ 00 ], then [γ · (γ 0 )−1 ], [γ 0 · (γ 00 )−1 ] ∈ H and so the product

[γ · (γ 00 )−1 ] ∈ H too.

Thus ∼H is an equivalence relation. (We could have defined ∼H for any subset H of
π1 (X, x0 ), but this argument shows that such a ∼H is an equivalence relation if and only
if H is a subgroup.)
Let us define X eH := X/ ∼H to be the quotient space, and pH : X eH → X to be the
induced map. Note that if [γ] ∈ (α, U ) and [γ ] ∈ (β, U ) have [γ] ∼H [γ 0 ] then (α, U ) and
0

(β, U ) are identified by ∼H , as [γ · η] ∼H [γ 0 · η] for η a path in U . Thus pH is a covering

map.
It remains to show that (pH )∗ (π1 (X eH , [[cx ]])) = H. If [γ] ∈ H then the lift if γ̄ of
0
γ to X starting at [cx0 ] ends at γ], so the lift to X eH ends at [[γ]] = [[cx ]] so is a loop.
0
This shows the containment ⊇. On the other hand, if [γ] = (pH )∗ ([γ̃]) for γ̃ a loop in
XeH , then the lift of γ̃ to X starting at [cx ] ends at [γ]. As γ̃ is a loop in X
0
eH , [γ] must
be identified with [cx0 ], so lies in H; this shows the containment ⊆.

Proposition 3.4.2 (Based uniqueness). If

e1 , x̃1 ) −→ (X, x0 )
p1 : (X and e2 , x̃2 ) −→ (X, x0 )
p2 : (X

are path connected covering spaces, then there is a based homeomorphism

e1 , x̃1 ) −→ (X
h : (X e2 , x̃2 )

such that p2 ◦ h = p1 if and only if (p1 )∗ (π1 (X

e1 , x̃1 )) = (p2 )∗ (π1 (X
e2 , x̃2 )).

Proof. If h exists, we have (p1 )∗ = (p2 )∗ ◦ h∗ and h∗ is an isomorphism, so (p1 )∗ and

(p2 )∗ have the same image.
For the converse direction, let H := (p1 )∗ (π1 (X
e1 , x̃1 )); it is then enough to give a
based homeomorphism h : (XH , [[cx0 ]]) → (X1 , x̃1 ). We begin with the map
e e

r : (X, [cx0 ]) −→ (X
e1 , x̃1 )

given by sending the point [γ] to the end point of the lift γ̃ of γ to X
e1 starting at x̃1 .
Now

r([γ]) = r([γ 0 ]) ⇐⇒ γ̃ and γ̃ 0 end at the ssame point in X

e1
⇐⇒ [γ 0 · γ −1 ] ∈ π1 (X
e1 , x̃1 ) = H
⇐⇒ [γ] ∼H [γ 0 ]

and so the map r descends to a continuous bijection

eH , [[cx ]]) −→ (X
q : (X e1 , x̃1 )
0

of covering spaces over (X, x0 ). This map is also open, as both covering maps to X are
local homeomorphisms, and so q is a homeomorphism.
 3.4 The Galois correspondence 27

Proposition 3.4.3 (Unbased uniqueness). If

e1 −→ X
p1 : X and e2 −→ X
p2 : X

are path connected covering spaces, then there is a homeomorphism

e1 −→ X
h:X e2

such that p2 ◦ h = p1 if and only if (p1 )∗ (π1 (X

e1 , x̃1 )) is conjugate to (p2 )∗ (π1 (X
e2 , x̃2 ))
−1
for some x̃i ∈ pi (x0 ).

Proof. If h exists, choose a x̃1 ∈ p−1

1 (x0 ) and let x̃2 = h(x̃1 ). With these basepoints
everything in sight is based, and so (p1 )∗ (π1 (X
e1 , x̃1 )) = (p2 )∗ (π1 (X
e2 , x̃2 )).
Conversely, suppose we have chosen x̃1 and x̃2 such that

[γ]−1 · (p1 )∗ (π1 (X

e1 , x̃1 )) · [γ] = (p2 )∗ (π1 (X
e2 , x̃2 ))

for some [γ] ∈ π − 1(X, x0 ). If we lift γ starting at x̃1 then it ends at some x̃01 ∈ X
e1 , and

e1 , x̃01 )) = (p2 )∗ (π1 (X

(p1 )∗ (π1 (X e2 , x̃2 )).

e1 , x̃0 ) → (X
Then the previous proposition gives a based homeomorphism h : (X e2 , x̃2 ).
1
 Chapter 4

Some group theory

4.1 Free groups and presentations

Let S = {sα }α∈I be a set, called the alphabet, and let S −1 = {s−1
α }α∈I (we suppose
that S ∩ S −1 = ∅). A word in the alphabet S is a (possibly empty) finite sequence

(x1 , x2 , . . . , xn )

of elements of S ∪ S −1 . A word is reduced if it contains no subwords

(sα , s−1
α ) or (s−1
α , sα ).

An elementary reduction of a word (x1 , . . . , xi , sα , s−1 α , xi+3 , . . . , xn ) means replacing

it by (x1 , . . . , xi , xi+3 , . . . , xn ), and similarly with words containing (s−1 α , sα ).

Definition 4.1.1 (Free group). The free group on the alphabet S, F (S), is the set of
reduced (possibly empty) words in S. The group operation is given by concatenating
words and then applying elementary reductions until the word is reduced.

It is not really clear that this group operation is well-defined, or that it is associative.
It is clear though that it is unital

() · (x1 , x2 , . . . , xn ) = (x1 , x2 , . . . , xn ) = (x1 , x2 , . . . , xn ) · ()

and that it has inverses

−1 −1
(x1 , x2 , . . . , xn ) · (x−1
n , xn−1 , . . . , x1 ) = ().

In Section 4.2 we give an alternative description of F (S) which is obviously a group, and
show it agrees with that of Definition 4.1.1.
By construction there is a function ι : S → F (S) given by sending the element sα to
the reduced word (sα ).

Lemma 4.1.2 (Universal property of free groups). For any group H, the function
   
group homomorphisms functions
−→ ,
ϕ : F (S) → H φ:S→H

given by precomposing with ι, is a bijection.

Proof. Given a function φ : S → H we want a homomorphism ϕ such that ϕ((sα )) =

φ(sα ). But there is a unique way to do this, by defining, on not necessarily reduced
words,
ϕ((sα11 , . . . , sαnn )) = φ(sα1 )1 · · · φ(sαn )n .

28
 4.1 Free groups and presentations 29

Note that if (sα11 , . . . , sαnn ) is not reduced, so contains for example (sα , s−1
α ), then the
product φ(sα1 )1 · · · φ(sαn )n contains φ(sα )·φ(sα )−1 = 1 and so we may reduce the word
(sα11 , . . . , sαnn ) without changing the value of ϕ on it. As the group operation in F (S) is
given by concatenation and reduction of words, this shows that ϕ is a homomorphism.

Definition 4.1.3 (Presentation). Let S be a set and R ⊂ F (S) be a subset. The group
hS | Ri is defined to be the quotient F (S)/hhRii of (S) by the smallest normal subgroup
containing R. More concretely,

hhRii = {(r11 )g1 · (r22 )g2 · · · (rnn )gn | ri ∈ R, i ∈ {±1}, gi ∈ F (S)}.

We call hS | Ri the group with generators S and relations R, and call the data (S, R) a
presentation of this group. If S and R are both finite, we call it a finite presentation.

Informally, we consider elements of hS | Ri as words in the alphabet S, up to the

relation that we may apply elementary reductions (or the converse) and cancel out any
subword lying in R (or the converse).

Lemma 4.1.4 (Universal property of group presentations). For any group H, the func-
tion    
group homomorphisms functions φ : S → H such
−→ ,
ψ : hS | Ri → H that ϕ(r) = 1 for each r ∈ R
ι quot. ψ
given by sending ψ to the composition φ : S → F (S) → hS | Ri → H, is a bijection

Proof. If ψ and ψ 0 give the same functions φ = φ0 : S → H, then by Lemma 4.1.2 the
two homomorphisms
φ
quot.
/ hS | Ri *
F (S) 4H
φ0

are equal, and as the quotient map is surjective it follows that ψ = ψ 0 .

On the other hand, let a function φ : S → H be given such that the induced map
ϕ : F (S) → H satisfied ϕ(r) = 1 for all r ∈ R. Then R ⊂ Ker(ϕ), and Ker(ϕ) is a normal
subgroup of F (S) so hhRii ⊂ Ker(ϕ) too (as hhRii was the minimal normal subgroup
containing R). Thus ϕ descends to a homomorphism ψ : F (S)/hhRii → H.

Example 4.1.5 (The canonical1 presentation). Let G be a group, and φ = IdG : G → G

be the identity homomorphism. By Lemma 4.1.2 there is an induced homomorphism
ϕ : F (G) → G. This is certainly surjective, as the word (g) maps to g ∈ G, and we can
let R := Ker(ϕ). By Lemma 4.1.4 there is an induced homomorphism

ψ : hG | Ri −→ G

which is an isomorphism (by the first isomorphism theorem).

By this example every group may be given a presentation, though of course this is
quite useless for practical work.
1
or “stupid"
30 Chapter 4 Some group theory

Example 4.1.6. Let G = ha, b | ai and H = ht | i, and let us show that these groups are
isomorphic. First consider the function

φ : {a, b} −→ H
a 7−→ e
b 7−→ [t]

which satisfies ϕ(a) = e and so defines a homomorphism ψ : G → H. Now consider the

function

φ0 : {t} −→ G
t 7−→ [b]

which defines a homomorphism ψ 0 : H → G.

Now ψ ◦ ψ 0 ([t]) = ψ([b]) = [t], and as [t] generates H it follows that ψ ◦ ψ 0 = IdH . On
the other hand ψ 0 ◦ ψ([b]) = ψ 0 ([t]) = [b] and

ψ 0 ◦ ψ([a]) = ψ 0 (1) = e = [a]

in the group G, as a is a relation. As [a] and [b] generate G, it follows that ψ 0 ◦ ψ = IdG .

Example 4.1.7. Let G = ha, b | ab−3 , ba−2 i. In this group we have [a] = [b]3 and
[b] = [a]2 , so putting these together [a] = [a]6 , and multiplying by [a]−1 we get [a]5 = e.
Thus using [b] = [a]2 we can write every word in a and b just using the letter a, and
using [a]5 = e we may ensure that the largest power of a appearing is 4. Thus the group
G has at most 5 elements, e, [a], [a]2 , [a]3 , [a]4 . Lets guess that it might be Z/5, and
prove this.
Consider the function

φ : {a, b} −→ Z/5
a 7−→ 1
b 7−→ 2.

Then ϕ(ab−3 ) = 1 − 2 · 3 = −5 = 0 ∈ Z/5 and ϕ(ba−2 ) = 2 − 2 = 0 ∈ Z/5 so by Lemma

4.1.4 this gives a homomorphism

ψ : ha, b | ab−3 , ba−2 i −→ Z/5.

This homomorphism sends [a] to 1 ∈ Z/5 so is surjective, but we already argued that G
had at most 5 elements, so it is also injective: thus ψ is an isomorphism.

4.2 Another view of free groups

Fix a set S and let W be the set of reduced words in the alphabet S, and P (W ) be the
group of permutations of the set W .
 4.2 Another view of free groups 31

Definition 4.2.1. For each α ∈ I, define a function Lα : W → W by the formula

(
(sα , x1 , x2 , . . . , xn ) if x1 6= s−1
α
Lα (x1 , x2 , . . . , xn ) = −1
(x2 , . . . , xn ) if x1 = sα .

Note that in the second case x2 6= sα , otherwise (x1 , x2 , . . . , xn ) would not be reduced.

Lemma 4.2.2. Lα is a bijection, so represents an element of P (W ).

Proof. Let (x1 , . . . , xn ) ∈ W . If x1 = sα then x2 6= s−1 α , so Lα (x2 , . . . , xn ) = (x1 , . . . , xn ).

If x1 6= sα then (s−1 α , x1 , . . . , x n ) is a reduced word, and Lα (s−1
α , x1 , . . . , xn ) = (x1 , . . . , xn ).
Thus Lα is surjective.
If Lα (x1 , x2 , . . . , xn ) = Lα (y1 , y2 , . . . , ym ) and this reduced word starts with sα then
x1 6= s−1 −1
α and y1 6= sα , and so xi = yi for each i. If this reduced word does not start
with sα then x1 = y1 = s−1 α , and

(x2 , . . . , xn ) = Lα (x1 , x2 , . . . , xn ) = Lα (y1 , y2 , . . . , ym ) = (y2 , . . . , ym ),

so (x1 , x2 , . . . , xn ) = (y1 , y2 , . . . , ym ).

We can now give an alternative definition of the free group, which is manifestly a
group.

Definition 4.2.3 (Free group). The free group F (S) is the subgroup of P (W ) generated
by the elements {Lα }α∈I .

Lemma 4.2.4. The function φ : F (S) → W given by σ 7→ σ · () is a bijection.

This identifies F (S) with the set of reduced words W in the alphabet S, and shows
that the group operation is given by concatenation of words followed by word reduction.
Thus the definition given in Definition 4.1.1 is indeed a group.

Proof. If (sα11 , . . . , sαnn ) is a reduced word, with i ∈ {±1}, then

(sα11 , . . . , sαnn ) = Lα11 · · · Lαnn · () = φ(Lα11 · · · Lαnn ),

and so φ is surjective.
As the {Lα }α∈I generate F (S), any element σ may be represented by a concatenation

σ = Lα11 · · · Lαnn ∈ P (W ).

As Lα · L−1 −1 1 n
α = IdW and Lα · Lα = IdW , if the word (sα1 , . . . , sαn ) is not reduced then we
can simplify Lα11 · · · Lαnn while giving the same element σ ∈ P (W ). Thus we may suppose
that any σ is represented by Lα11 · · · Lαnn such that the associated word (sα11 , . . . , sαnn ) is
reduced. But then
φ(σ) = σ · () = (sα11 , . . . , sαnn ),
from which we can recover σ = Lα11 · · · Lαnn , which shows that φ is injective.
32 Chapter 4 Some group theory

4.3 Free products with amalgamation

Let H, G1 , and G2 be groups, and let

i1 i2
G1 o H / G2

be homomorphisms. Suppose that we have presentations Gi = hSi | Ri i (which is no loss

of generality, by Example 4.1.5). Then the free product of G1 and G2 is

G1 ∗ G2 := hS1 ∪ S2 | R1 ∪ R2 i.

The functions Si → S1 ∪ S2 → F (S1 ∪ S2 ) → G1 ∗ G2 and Lemma 4.1.4 give canonical

homomorphisms
j1
G1 / G1 ∗ G2 o j2 G2
and we define the free product with amalgamation over H to be the quotient

G1 ∗H G2 := (G1 ∗ G2 )/hhj1 i1 (h)(j2 i2 (h))−1 | h ∈ Hii.

Note that the free product is the free product with amalgamation for H = {e}. By
construction, the square
i1
H / G1

i2 j1
 j2 
G2 / G1 ∗H G2

commutes, that is j1 ◦ i1 = j2 ◦ i2 : H → G1 ∗H G2 .

Lemma 4.3.1 (Universal property of amalgamated free products). For any group K the
function
   
group homomorphisms group homomorphisms φ1 : G1 → K
−→ ,
φ : G1 ∗H G2 → K and φ2 : G2 → K such that φ1 ◦ i1 = φ2 ◦ i2

given by sending φ to φ1 = φ ◦ j1 and φ2 = φ ◦ j2 , is a bijection.

Proof. Given φ1 and φ2 , define φ̂ : F (S1 ∪ S2 ) → K via the function

ˆ
φ̂ : S1 ∪ S2 −→ K
s ∈ Si 7−→ φi ((s)).

If r ∈ R1 then φ̂(r) = φ1 (r) = e as [r] = e ∈ G1 = hS1 | R1 i; similarly for r ∈ R2 .

Thus φ̂ descends to a homomorphism φ̃ : G1 ∗ G2 → K. Finally, for h ∈ H we have

φ̃(j1 i1 (h)(j2 i2 (h))−1 ) = φ̃(j1 i1 (h))φ̃(j2 i2 (h))−1 = φ1 i1 (h)(φ2 i2 (h))−1 = e

and so hhj1 i1 (h)(j2 i2 (h))−1 | h ∈ Hii lies in Ker(φ̃), so φ̃ descends to a homomorphism

φ : G1 ∗H G2 → K as required.
 Chapter 5

The Seifert–van Kampen theorem

5.1 The Seifert–van Kampen theorem

Let X be a space, A, B ⊂ X be subsets, and x0 ∈ A ∩ B be a point. There is then a

commutative diagram
π1 (A ∩ B, x0 ) / π1 (A, x0 )

 
π1 (B, x0 ) / π1 (X, x0 )

where all the homomorphisms are those induced on the fundamental group by the natural
inclusions. By the universal property of amalgamated free products (Lemma 4.3.1) there
is a corresponding homomorphism

φ : π1 (A, x0 ) ∗π1 (A∩B,x0 ) π1 (B, x0 ) −→ π1 (X, x0 ).

The Seifert–van Kampen theorem concerns conditions under which this homomorphism
is an isomorphism.
Theorem 5.1.1 (Seifert–van Kampen). Let X be a space, and A, B ⊂ X be open subsets
which cover X and such that A ∩ B is path connected. Then for any x0 ∈ A ∩ B the
homomorphism
φ : π1 (A, x0 ) ∗π1 (A∩B,x0 ) π1 (B, x0 ) −→ π1 (X, x0 )
is an isomorphism.
Before giving the proof of this theorem, we give some important examples.
Example 5.1.2 (Spheres). Consider the sphere S n for n ≥ 2. This may be covered by
the open sets
U := {(x1 , x2 , . . . , xn+1 ) ∈ S n | xn+1 > −1}
V := {(x1 , x2 , . . . , xn+1 ) ∈ S n | xn+1 < 1}
the complements of the north and south poles. Each of these is open, and stereographic
projection shows that they are homeomorphic to Rn and so contractible. Finally, cylin-
drical projection gives a homeomorphism

U ∩V ∼
= S n−1 × (−1, 1)
and as n ≥ 2 this is path connected. Thus by the Seifert–van Kampen theorem we have

π1 (S n , x0 ) ∼
= {e} ∗π1 (U ∩V,x0 ) {e}
and so π1 (S n , x0 ) ∼
= {e}. Thus the n-sphere is simply connected for n ≥ 2.

33
34 Chapter 5 The Seifert–van Kampen theorem

Example 5.1.3 (Real projective space). Recall that we have shown that the quotient
map p : S n → RPn is a covering map, with two sheets. If n ≥ 2 then S n is simply
connected, by the previous example, and so this is the universal cover of RPn . By Lemma
3.1.15 (iii), there is then a bijection between the group π1 (RPn , x0 ) and the set p−1 (x0 )
with two elements: there is only one group of cardinality two, so π1 (RPn , x0 ) = Z/2.

For based spaces (X, x0 ) and (Y, y0 ), the wedge product X ∨ Y is the quotient
space of X t Y by the relation generated by x0 ∼ y0 . This common point is taken to be
the basepoint of X ∨ Y .

Example 5.1.4 (Wedge of two circles). Let the circle S 1 ⊂ C have the basepoint 1 ∈ S 1 ,
and consider the space S 1 ∨ S 1 with basepoint x0 the wedge point. Let us take the open
cover
U := S 1 ∨ (S 1 \ {−1}) V := (S 1 \ {−1}) ∨ S 1 .
As S 1 \ {−1} deformation retracts to the basepoint 1 ∈ S 1 , notice that U deformation
retracts to S 1 ∨ {1} ∼
= S 1 , V deformation retracts to {1} ∨ S 1 , and U ∩ V = (S 1 \ {−1}) ∨
1
(S \ {−1}) deformation retracts to {x0 }. Thus by the Seifert–van Kampen theorem we
have
π1 (S 1 ∨ S 1 , x0 ) ∼
= π1 (S 1 ∨ {1}, x0 ) ∗ π1 ({1} ∨ S 1 , x0 ).

If we let a be the standard loop around the first circle and b be the standard loop
around the second circle, as shown above, we thus have

π1 (S 1 ∨ S 1 , x0 ) ∼
= ha, b | i,

a free group on two generators.

This is our first example of a space with a really complicated fundamental group. Let
us apply the theory of Section 3.4 to it, to investigate some of its covering spaces.

Example 5.1.5. The function

φ : {a, b} −→ Z/3
a 7−→ 1
b 7−→ 1

induces a surjective homomorphism ϕ : π1 (S 1 ∨ S 1 , ∗) = ha, b | i → Z/3. Thus K :=

Ker(ϕ) is an index 3 subgroup of π1 (S 1 ∨ S 1 , ∗). By the results of Section 3.4 there is a
corresponding unique based covering space
e x̃0 ) −→ (S 1 ∨ S 1 , ∗)
p : (X,
e x˜0 )) = K ≤ π1 (S 1 ∨ S 1 , ∗). Let us work out what it looks like.
with p∗ (π1 (X,
 5.1 The Seifert–van Kampen theorem 35

By Lemma 3.1.15 (iii), the group π1 (S 1 ∨ S 1 , ∗) acting on the point x̃0 ∈ p−1 (∗)
induces a bijection
π1 (S 1 ∨ S 1 , ∗)
−→ p−1 (∗)
p∗ (π1 (X, x˜0 ))
e

and so p is a 3-sheeted cover, because p∗ (π1 (X,

e x˜0 )) = K has index 3. The 3 elements of
−1
p (∗) are thus identified with the right cosets of K, lets say x̃0 = K, Ka, and Ka2 .
Each of the edges a and b in S 1 ∨ S 1 has 3 lifts to edges in X,
e and we wish to know
1 1
where these go from and to. This is easy: if x ∈ π1 (S ∨ S , ∗) = ha, b | i is a word in a
and b, then lifting it starting at the point Ky ∈ p−1 (∗) gives a path ending at Kyx.
Thus the lift of a starting at K ends at Ka, the lift starting at Ka ends at Ka2 , and
the lift starting at Ka2 ends at Ka3 = K. Similarly, the lift of b starting at K ends at
Kb = Ka, the lift starting at Ka ends at Kab = Ka2 , and the lift starting at Ka2 ends
at Ka2 b = K. Thus the covering space X e looks like

where the labels on the edges show to which edge in S 1 ∨ S 1 they map to.

Example 5.1.6. The function

φ : {a, b} −→ Z/3
a 7−→ 1
b 7−→ 0

induces a surjective homomorphism ϕ : π1 (S 1 ∨ S 1 , ∗) = ha, b | i → Z/3. Thus K :=

Ker(ϕ) is an index 3 subgroup of π1 (S 1 ∨ S 1 , ∗), and there is a corresponding covering
e x̃0 ) → (S 1 ∨ S 1 , ∗). The same analysis as in the last example gives the
space p : (X,
covering space

Example 5.1.7. The universal cover of S 1 ∨ S 1 may be described as the infinite 4-valent
tree (i.e. graph with no loops), where each vertex has a copy of the edge a coming in, a
copy of the edge a going out, a copy of the edge b coming in, and a copy of the edge b
36 Chapter 5 The Seifert–van Kampen theorem

going out. This certainly describes a covering space of S 1 ∨ S 1 ; it is easy to see that this
tree is simply connected, by showing that it is in fact contractible.

Let us now move on to the proof of the Seifert–van Kampen theorem.

Proof of Theorem 5.1.1. First observe that we may assume that A and B are also path
connected: if not, there is a path component of A or B which does not intersect A ∩ B,
and so this lies in a different path component of X to x0 . Hence loops based at x0 cannot
reach it, so removing it does not change any of π1 (A, x0 ), π1 (B, x0 ), π1 (A ∩ B, x0 ) or
π1 (A ∪ B, x0 ).
The map φ is surjective. Let γ : I → X be a loop based at x0 , so {γ −1 (A), γ −1 (B)} is
an open cover of I; let n1 be a Lesbegue number for this cover. Then each path γ| i i+1
[n, n ]
has image entirely in A or entirely in B (or both). If subsequent path pieces γ| i i+1
[n, n ]
and γ| i+1 i+2 both lie in A or both lie in B then we may concatenate them. Proceeding
[ n , n ]
in this way, we may express γ as a concatenation γ1 · γ2 · · · γk where each γi (0) and each
γi (1) lie in A ∩ B.
Now for each point γ1 (1), γ2 (1), . . . , γk−1 (1) ∈ A ∩ B choose a path ui from γi (1) to
x0 in A ∩ B. Then γ is homotopic to

(γ1 · u1 ) · (u−1 −1
1 · γ2 · u2 ) · · · (uk−1 · γk )

as a path, and so is a product of elements coming from π1 (A, x0 ) and π1 (B, x0 ), as

required.
The map φ is injective. By considering the description of the amalgamated free
product, the group π1 (A, x0 ) ∗π1 (A∩B,x0 ) π1 (B, x0 ) has the following description. It is
generated by

(i) for each loop γ : I → A, an element [γ]A ,

(ii) for each loop γ : I → B, an element [γ]B ,

subject to the relations

 5.1 The Seifert–van Kampen theorem 37

(i) if γ ' γ 0 as paths in A, then [γ]A = [γ 0 ]A , and similarly for B,

(ii) if γ, γ 0 : I → A then [γ]A · [γ 0 ]A = [γ · γ 0 ]A , and similarly for B,

(iii) if γ : I → A ∩ B is a loop then [γ]A = [γ 0 ]B .

Let S1 , . . . , Sn be a sequence of elements of {A, B}, and γi : I → Si be a sequence of

loops based at x0 , and suppose that

φ([γ1 ]S1 · [γ2 ]S2 · · · [γn ]Sn ) = [cx0 ] ∈ π1 (X, x0 ).

Temporary definition. For a sequence γ1 , γ2 . . . , γn of loops based at x0 , let γ1 ·

γ2 · · · γn , unbracketed, be the concatenation formed by cutting I into n pieces of length
1
n , and running each path γi n times as fast.
By assumption γ1 · γ2 · · · γn ' cx0 as paths in X; let H be a homotopy between them.
We can subdivide I × I into squares of side length N1 for n | N so that H sends each
square entirely into A or entirely into B. Choose for each square a label A or B saying
into which set it is mapped (some squares will admit either choice, and some not).

For each point ( Ni , Nj ) ∈ I ×I choose a path ui,j from H( Ni , Nj ) to x0 , such that if ( Ni , Nj )

is a corner of a square with label A then the path ui,j lies in A, and if it is a corner of a
square with label B then the path ui,j lies in B (so if it is a corner of squares with both
labels then the path lies in A ∩ B).
Step 1. For the path H| i−1 i j which ends at ( Ni , Nj ), there is a homotopy to the
[ N , N ]× N
path u−1
i−1,j · H| i−1 i j · ui,j which ends at x0 given by
[ N , N ]× N
38 Chapter 5 The Seifert–van Kampen theorem

j
This homotopy can be taken to lie in A if either the square above or below [ i−1 i
N , N]× N
is labelled A, and similarly for B, as the paths ui−1,j and ui,j then lie in the required
subspace. Call this homotopy hi,j , and write h̄i,j for the reverse homotopy.
If i − 1 = 0 or i = N the homotopy can be suitably modified, as then the path
H| i−1 i j already starts or ends at x0 .
[ N , N ]× N

Step 2. Form a homotopy by gluing together the homotopies h̄i,0 as shown.

This shows that each path γi in Si may be homotoped (in Si ) to be the composition of
N/n
N/n loops γi1 · γi2 · · · γi given by

(γi |[0, n ] · u(i−1)n+1,0 ) · (u−1

(i−1)n+1,0 · γi | n 2n · u(i−1)n+2,0 ) · · · (u−1
in−1,0 · γi | N −n )
N [N , N ] [ N ,1]

N/n
and so [γi ]Si = [γi1 ]Si · · · [γi ]Si in the amalgamated free product. Thus it is enough to
show that
N/n N/n
([γ11 ]S1 · · · [γ1 ]S1 ) · ([γ21 ]S2 · · · [γ2 ]S2 ) · · · ([γn1 ]Sn · · · [γnN/n ]Sn )

is trivial in the amalgamated free product.

Step 3. We now form a homotopy by gluing together the pieces H| i−1 i j−1 j and
[ N , N ]×[ N , N ]
the homotopies hi,j and h̄i,j together. For each row H| j−1 j of the homotopy H
[0,1]×[ N , N ]
we form a homotopy Gj by

Note that each column of the above maps entirely into A or entirely into B, and the
points i/N along the top or bottom all map to x0 .
Step 4. As each column of the homotopy Gj lies entirely in A or B, and has its four
vertices mapping to x0 , it follows that the loop given by the top of each column in
homotopic to the loop given by the bottom of each column by a homotopy which lies
entirely in A or entirely in B. Thus the sequence of composable loops along the bottom
and the sequence of composable loops along the top represent the same element of the
amalgamated free product. As the sequence
N/n N/n
([γ11 ]S1 · · · [γ1 ]S1 ) · ([γ21 ]S2 · · · [γ2 ]S2 ) · · · ([γn1 ]Sn · · · [γnN/n ]Sn )
 5.2 The effect on the fundamental group of attaching cells 39

arises as the bottom of G0 , this argument shows that it is equivalent in the amalgamated
free product to the sequence of composable loops arising from the top of GN −1 . But by
reversing the argument of Step 2, the sequence of composable loops arising from the top
of GN −1 is a refinement of the loop H(−, t), which was constant.

5.2 The effect on the fundamental group of attaching cells

Let f : (S n−1 , ∗) → (X, x0 ) be a based map, and

Y := X ∪f Dn = (X t Dn )/z ∈ S n−1 ⊂ Dn ∼ f (z) ∈ X.

We let y0 = [x0 ] be the equivalence class of x0 , so the inclusion i : X → Y is a based

map. Thus there is an induced homomorphism

i∗ : π1 (X, x0 ) −→ π1 (Y, y0 ).

The space Y has an open cover by the interior of the disc

U := int(Dn )

and the complement of the centre of the disc

V := X ∪f (Dn \ {0}).

These satisfy U ∩ V ∼
= S n−1 × (0, 1), which is path connected as long as n ≥ 2.

Theorem 5.2.1.

(i) If n ≥ 3 then the map i∗ is an isomorphism.

(ii) If n = 2 then the map i∗ is surjective with kernel the normal subgroup generated by
[f ] ∈ π1 (X, x0 ).

Proof. Choose a point y1 ∈ U ∩ V ⊂ int(Dn ), and a path u in Dn \ {0} from y1 to the

basepoint ∗ ∈ S n−1 ⊂ Dn (which is identified with x0 ∈ X).

If n ≥ 3, then U ∩ V ' S n−1 is simply connected. By Seifert–van Kampen the map

φ : π1 (U, y0 ) ∗π1 (U ∩V,y1 ) π1 (V, y1 ) −→ π1 (Y, y1 )

40 Chapter 5 The Seifert–van Kampen theorem

is an isomorphism, but both π1 (U ∩ V, y1 ) and π1 (U, y1 ) are trivial and so the map
π1 (V, y1 ) → π1 (Y, y1 ) is an isomorphism. Hence, the change-of-basepoint isomorphism
given by u shows that π1 (V, y0 ) → π1 (Y, y0 ) is an isomorphism. Finally, the inclusion
(X, x0 ) → (V, y0 ) is a based homotopy equivalence (by pulling Dn \ {0} on to its bound-
ary), and the theorem follows.
If n = 2 then U ∩ V ' S 1 has fundamental group Z. The map

Z∼
= π1 (U ∩ V, y1 ) −→ π1 (V, y1 )
is seen to send the generator to the loop u# ([f ]). Thus, as π1 (U, y1 ) = {e}, the Seifert–
van Kampen theorem shows that

φ : π1 (V, y1 )/hhu# [f ]ii −→ π1 (Y, y1 )

is an isomorphism. Under the change-of-basepoint isomorphism given by the path u, and

again using that the inclusion (X, x0 ) → (V, y0 ) is a based homotopy equivalence, the
theorem follows.
Example 5.2.2. The standard cell structure on the torus T has two 1-cells a and b
attached to a single vertex v, and a 2-cell attached along a loop given by a · b · a−1 · b−1 .
The 1-skeleton T 1 is thus a wedge of two circles, so π1 (T 1 , v) ∼ = ha, b | i, and attaching
the 2-cell thus gives
π1 (T, v) ∼
= ha, b | a · b · a−1 · b−1 i.
One may easily check that this is a presentation of the abelian group Z × Z.
Corollary 5.2.3. For any group presentation G := hS | Ri with S and R finite, there is
a 2-dimensional based cell complex (X, x0 ) with π1 (X, x0 ) ∼
= G.
Proof. Let Y be a wedge of |S|-many circles, each identified with an element of S. Let
Y have the wedge point as basepoint, and call it y0 . For s ∈ S let xs ∈ π1 (Y, y0 ) be a
loop which transverses the sth circle precisely once. Then sending s to xs gives a group
isomorphism
hS | i ∼
= π1 (Y, y0 ).
Each element r ∈ R ⊂ hS | i is represented by a word in the alphabet S, and so by an
element of π1 (Y, y0 ). Let γr0 : I → Y be a loop in Y such that [γr0 ] corresponds to r under
the isomorphism above, and γr : (S 1 , 1) → (Y, y0 ) be the map from the circle obtained
by gluing the endpoints of I together. By Theorem 5.2.1 attaching a 2-cell to Y along
γr has the effect of dividing out by the normal subgroup generated by [γr ]. Thus if we
attach a 2-cell along γr for each r ∈ R to form a cell complex X, with x0 = [y0 ], then we
have
π1 (X, x0 ) ∼
= π1 (Y, y0 )/hh[γr ], r ∈ Rii
But this is precisely the description of hS | Ri.
The assumption that S and R are finite may be removed, by developing the Seifert–
van Kampen theorem for covers by infinitely many open sets such that every possible
intersection of them is path connected. Hence every group occurs as the fundamental
group of a 2-dimensional cell complex (possibly having infinitely many cells).
 5.3 A refinement of the Seifert–van Kampen theorem 41

5.3 A refinement of the Seifert–van Kampen theorem

The requirement in Theorem 5.1.1 that A and B be a pair of open sets does not fit with
how one typically wants to use the theorem. In this section we shall describe weaker
hypotheses that are usually more convenient.

Definition 5.3.1. A subset A ⊂ X of a space X is called a neighbourhood deforma-

tion retract if there is an open neighbourhood A ⊂ U ⊂ X and a strong1 deformation
retraction of U onto A.

Theorem 5.3.2. Let X be a space, and A, B ⊂ X be closed subsets which cover X and
such that A ∩ B is both path connected and a neighbourhood deformation retract in both
A and B. Then for any x0 ∈ A ∩ B the homomorphism

φ : π1 (A, x0 ) ∗π1 (A∩B,x0 ) π1 (B, x0 ) −→ π1 (X, x0 )

is an isomorphism.

Proof. Choose open neighbourhoods A ∩ B ⊂ U ⊂ A and A ∩ B ⊂ V ⊂ B which strongly

deformation retract to A ∩ B. Now the complement of A ∪ V in X is B \ V , which is
closed as V was open in B; thus A ∪ V is open in X. Similarly U ∪ B is open in X. The
strong deformation retraction of V to A ∩ B shows that the inclusion A → A ∪ V is a
homotopy equivalence, and similarly for B → U ∪ B.
Finally, (A ∪ V ) ∩ (U ∪ B) = U ∪ V and the strong deformation retractions of U to
A ∩ B and V to A ∩ B glue together to give a (strong) deformation retraction of U ∪ V
to A∩V.
We may now apply Theorem 5.1.1 to the open cover given by A ∪ V and U ∪ B, which
by the above discussion has (A ∪ V ) ∩ (U ∪ B) = U ∪ V ' A ∩ B path connected. The
theorem then follows from noting that the maps

π1 (A, x0 ) −→ π1 (A ∪ V, x0 )
π1 (B, x0 ) −→ π1 (U ∪ B, x0 )
π1 (A ∩ B, x0 ) −→ π1 (U ∪ V, x0 )

are all isomorphisms.

Recall that for based spaces (X, x0 ) and (Y, y0 ) the wedge product X ∨ Y is the
quotient space of X t Y by the relation generated by x0 ∼ y0 . If the subspaces {x0 } ⊂ X
and {y0 } ⊂ Y are neighbourhood deformation retracts2 then it follows from Theorem
5.3.2 that
π1 (X ∨ Y, [x0 ]) ∼
= π1 (X, x0 ) ∗ π1 (Y, y0 ).
1
That is, a homotopy from IdU to a map r : U → A, which is constant when restricted to A.
2
Some authors say that a based space (X, x0 ) is well-based if {x0 } ⊂ X is a neighbourhood
deformation retract.
42 Chapter 5 The Seifert–van Kampen theorem

Example 5.3.3. The based space (S 1 , 1) has {1} ⊂ S 1 a neighbourhood deformation

retract, and so

π1 (S 1 ∨ S 1 , [1]) ∼
= π1 (S 1 , 1) ∗ π1 (S 1 , 1) ∼
=Z∗Z∼
= ha, b | i

is a free group on two generators. By induction on n, for any finite n we thus have
n
S 1 , [1]) ∼
_
π1 ( = hx1 , x2 , . . . , xn | i,

a free group on n generators.

5.4 The fundamental group of a surface

Example 5.4.1 (Orientable surfaces). Consider the based space (X, x0 ) obtained from
the torus by removing an open triangle, as shown

By stretching the hole, we see that the inclusion

ι : (S 1 ∨ S 1 , [1]) −→ (X, x0 )

of the loops a and b is a based homotopy equivalence. Thus we have π1 (X, x0 ) ∼ = ha, b | i.
The loop r given by the boundary of the open disc we removed represents the class
[a, b] := aba−1 b−1 ∈ π1 (X, x0 ), if we orient it as shown in the figure above.

Gluing g copies of X together as shown above gives a based space (Fg , f0 ). As the
gluings are performed along intervals, which are contractible and are neighbourhood
deformation retracts in X, Theorem 5.3.2 shows that there is an isomorphism

π1 (Fg , f0 ) ∼
= ha1 , b1 , a2 , b2 , . . . , ag , bg | i
 5.4 The fundamental group of a surface 43

where ai and bi are the loops a and b in the ith copy of X. The boundary of Fg is a
circle, and the loop it represents is given by r1 · r2 · · · rg ∈ π1 (Fg , f0 ), the product of the
boundary loops on theQ individual copies of X. In terms of the generators ai and bi , the
boundary loop is thus gi=1 [ai , bi ]. Attaching D2 to Fg gives a closed surface Σg , and by
Theorem 5.2.1 we have
g
π1 (Σg ) ∼
Y
= ha1 , b1 , a2 , b2 , . . . , ag , bg | [ai , bi ]i.
i=1

Example 5.4.2 (The projective plane again). First recall from Example 5.1.3 that we
have shown that π1 (RP2 , x0 ) is Z/2. Let us try to understand the generator of this group.

The description of RP2 as a quotient space of S 2 also shows that it may be obtained
from D2 (thought of as the upper hemisphere of S 2 ) by dividing out by the equivalence
relation
x ∈ S 1 ⊂ D2 ∼ −x ∈ S 1 ⊂ D2
given by identifying opposite points on the boundary of D2 . If we give D2 the cell
structure shown in the figure above then we see that there is an induced cell structure
on RP2 having a single 0-cell, a single 1-cell, and a single 2-cell. If the loop around the
1-cell is called a, we see that the fundamental group of RP2 has the presentation

π1 (RP2 , x0 ) ∼
= ha | a2 i.

Example 5.4.3 (Nonorientable surfaces). Consider the based space (Y, y0 ) obtained
from RP2 by removing an open triangle, as shown

By stretching the hole, we see that Y is homotopy equivalent to S 1 , and so π1 (Y, y0 ) ∼ =

ha | i. The boundary loop r is now given by a2 .
Gluing n copies of Y together as shown gives a based space (En , e0 ) with π1 (En , e0 ) ∼ =
ha1 , a2 , . . . , an | i, and with boundary loop given by r1 · r2 · · · rn = a21 a22 · · · a2n . Hence, if
44 Chapter 5 The Seifert–van Kampen theorem

we form the closed surface Sn by attaching a copy of D2 to En along its boundary, we

have
π1 (Sn ) ∼
= ha1 , a2 , . . . , an | a21 a22 · · · a2n i.
 Chapter 6

Simplicial complexes

6.1 Simplicial complexes

Definition 6.1.1. A finite set of points a0 , a1 , . . . , an ∈ Rm are affinely independent

if ( n n
)
X X
ti ai = 0 and ti = 0 ⇔ (t0 , t1 , . . . , tn ) = 0.
i=0 i=0

Lemma 6.1.2. The set a0 , a1 , . . . , an ∈ Rm are affinely independent if and only if the
vectors a1 − a0 , a2 − a0 , . . . , an − a0 are linearly independent.

Proof. Suppose that the a0 , a1 , . . . , an ∈ Rm are affinely independent, and that

n
X
si (ai − a0 ) = 0.
i=1

Then (− ni=1 si )a0 + s1 a1 + · · · + sn an = 0 and the sum of the coefficients is 0, so by

P
definition of affinely independent we must have si = 0 for all i, so the vectors a1 −a0 , a2 −
a0 , . . . , an − a0 are linearly independent.
Suppose now that the vectors Pn a1 − a0 , a2 − a0P
, . . . , an − a0 are linearly independent,
n
and that P there are ti with i=0 ti ai = 0 and i=0 ti =P0. The second sum gives
t0 a0 = − ni=1 ti ai , and so we may rewrite the first sum as ni=1 ti (ai − a0 ) = 0. As the
vectorsP ai − a0 are linearly independent, it follows that ti = 0 for all i ≥ 1, but then
t0 = − ni=1 ti is 0 too.

Definition 6.1.3. If a0 , a1 , . . . , an ∈ Rm are affinely independent, they define an n-

simplex ( n n
)
X X
σ = ha0 , a1 , . . . , an i := ti ai | ti = 1, ti ≥ 0
i=0 i=0

given by the convex hull of the points ai . The ai are called the vertices of σ, and are
said to span σ.
If x ∈ ha0 , a1 , . . . , an i then x can be written as x = ni=0 ti ai for unique real num-
P
bers ti (the uniqueness follows from the affine independence). The ti are called the
barycentric coordinates of x.
A face of a simplex σ = ha0 , a1 , . . . , an i is a simplex τ spanned by a subset of
{a0 , a1 , . . . , an }; we write τ ≤ σ, and τ < σ if τ is a proper face of σ.
The boundary of a simplex σ, written ∂σ, is the union of all of its proper faces. Its
interior, written σ̊, is σ \ ∂σ.

45
46 Chapter 6 Simplicial complexes

Lemma 6.1.4. Let σ be a p-simplex in Rm and τ be a p-simplex in Rn . Then σ and τ

are homeomorphic.

Proof. Let σ = ha0 , a1 , . . . , ap i and τ = hb0 , b1 , . . . , bp i. Define a function

h : σ −→ τ
p
X p
X
ti ai 7−→ ti bi
i=0 i=0

which is well-defined and a bijection, by the uniqueness of the barycentric coordinates

ti . As the vectors ai − a0 are linearly independent, we may extend h to an affine map
ĥ : Rm → Rn , so it is continuous.

Definition 6.1.5. A geometric (or euclidean) simplicial complex in Rm is a finite

set K of simplices in Rm such that

(i) if σ ∈ K and τ ≤ σ, then τ ∈ K,

(ii) if σ, τ ∈ K then σ ∩ τ is either empty or is a face of both σ and τ .

Figure 6.1 a) a geometric simplicial complex; b) not a geometric simplicial complex.

The dimension of K is the largest n for which K contains an n-simplex. The

polyhedron |K| associated to K is the subspace of Rm given by the union of all the
simplices in K. The d-skeleton K(d) of a simplicial complex K is the sub simplicial
complex given by the simplices of K of dimension at most d.

Note that a simplex is closed and bounded, so it is compact; as a polyhedron is a finite

union of simplices, it follows that it is compact. As Rm is Hausdorff, any polyhedron is
also Hausdorff.

Definition 6.1.6. A triangulation of a space X is a pair of a simplicial complex K

and a homeomorphism h : |K| → X.

Thus if X has a triangulation it must in particular be compact and Hausdorff; this

is however not sufficient.
 6.1 Simplicial complexes 47

Example 6.1.7. The standard n-simplex ∆n ⊂ Rn+1 is the simplex spanned by the
basis vectors e1 , e2 , . . . , en+1 . It, along with all its faces, defines a simplicial complex.

Example 6.1.8. The simplicial (n − 1)-sphere is the simplicial complex given by the
proper faces of ∆n , and all their faces. Its polyhedron is ∂∆n ⊂ Rn+1 .

Example 6.1.9. In Rn+1 consider the 2n+1 n-simplices given by h±e1 , ±e2 , . . . , ±en+1 i,
and let K be the simplicial complex given by these simplices and their faces. Define a
function

h : |K| ⊂ Rn+1 −→ S n
x
x 7−→
|x|

which is a bijection and continuous (it is the restriction of the map Rn+1 \ {0} → S n
given by the same formula) so is a homeomorphism (as the source is compact and the
target is Hausdorff). This defines a triangulation of S n .

Figure 6.2 a) The polyhedron |K|; b) radial projection to the sphere.

Lemma 6.1.10. Any point of the polyhedron |K| lies in the interior of a unique simplex.

Proof. If x ∈ |K| = ∪σ∈KP σ then certainly x lies in some simplex σ, say with σ =
, ap i. Thus x = pi=0 ti ai . If τ ≤ σ is the face given by {ai | ti > 0} then we
ha0 , a1 , . . .P
have x = ni=0 si bi with τ = hb0 , b1 , . . . , bn i and si > 0 for all i, and so x is in τ̊ .
If x ∈ τ̊ 0 too, then τ̊ ∩ τ̊ 0 6= ∅. Thus τ ∩ τ 0 is not empty, so is a face of both τ and τ 0 .
But it contains an interior point of either simplex, so is not a proper face of either: thus
τ = τ ∩ τ 0 = τ 0.

Definition 6.1.11. For a simplicial complex K, let VK denote the set of 0-simplices of
K, which we call the vertices of K.

Definition 6.1.12. A simplicial map f from K to L is a function f : VK → VL such

that if σ = ha0 , a1 , . . . , ap i is a simplex of K, then the set {f (a0 ), f (a1 ), . . . , f (ap )} spans
a simplex of of L, which we call f (σ). We will write f : K → L for a simplicial map.
48 Chapter 6 Simplicial complexes

The reason for this perhaps surprising formulation is that the list f (a0 ), f (a1 ), . . . , f (ap )
is allowed to have repeats. Thus these elements to not need to be affinely independent,
they just need to become so when we throw away repeated elements.

Example 6.1.13. The function

f : ∆1 ⊂ R2 −→ ∆2 ⊂ R3
(1, 0) 7−→ (1, 0, 0)
(0, 1) 7−→ (0, 0, 1)

defines a simplicial map, as does the function

g : ∆2 ⊂ R3 −→ ∆1 ⊂ R2
(1, 0, 0) 7−→ (1, 0)
(0, 1, 0) 7−→ (1, 0)
(0, 0, 1) 7−→ (0, 1).

Lemma 6.1.14. A simplicial map f : K → L induces a continuous map |f | : |K| → |L|,

and |g ◦ f | = |g| ◦ |f |.

Proof. For a simplex σ ∈ K, σ = ha0 , a1 , . . . , ap i, define

fσ : σ −→ |L|
p
X p
X
ti ai 7−→ ti f (ai ),
i=0 i=0

which is linear in the barycentric coordinates ti , so continuous.

If τ ≤ σ then fτ = fσ |τ , so for simplices σ and σ 0 we have fσ |σ∩σ0 = fσ∩σ0 = fσ0 |σ∩σ0
and so the functions fσ glue to a well-defined function
[
|f | : |K| = σ −→ |L|,
σ∈K

which is continuous by the gluing lemma. The formula for fσ shows that |f | behaves as
claimed under composition.
 6.2 Simplicial approximation 49

Note that we can recover the simplicial map f from the continuous map |f | and the
subsets VK ⊂ |K| and VL ⊂ |L|. Thus a simplicial map from K to L could also be
defined as a continuous map |K| → |L| which sends vertices to vertices and is linear on
each simplex.
Definition 6.1.15. For a point x ∈ |K|,
(i) The (open) star of x is the union of the interiors of those simplices in K which
contain x, [
StK (x) = σ̊ ⊂ Rm .
x∈σ
The complement of StK (x) is the union of those simplices of K which do not contain
x: this is a polyhedron, so closed in |K|, so StK (x) is indeed open.
(ii) The link of x, LkK (x), is the union of those simplices of K which do not contain x,
but are a face of a simplex which does contain x. This is a subpolyhedron of |K|,
so is closed.

Figure 6.3 Three examples of points x with their open stars (in grey) and links (dotted).

6.2 Simplicial approximation

Definition 6.2.1. Let f : |K| → |L| be a continuous map. A simplicial approxima-

tion to f is a function g : VK → VL such that

f (StK (v)) ⊂ StL (g(v))

for every v ∈ VK .
Lemma 6.2.2. If g is a simplicial approximation to a map f , then g defines a simplicial
map and f is homotopic to |g|. Furthermore, this homotopy may be supposed to be relative
to the subspace {x ∈ |K| |f (x) = |g|(x)} where the two maps already agree.
Proof. To show that g defines a simplicial map, for each σ ∈ K we must show that
the images under g of the vertices of σ span a simplex in L. For x ∈ σ̊ we have x ∈
∩v∈Vσ StK (v), and so
\ \
f (x) ∈ f (StK (v)) ⊂ StL (g(v)).
v∈Vσ v∈Vσ
50 Chapter 6 Simplicial complexes

Thus if τ ∈ L is the unique simplex such that f (x) ∈ τ̊ , then every g(v) is a vertex of τ ,
and so the collection {g(v) | v ∈ Vσ } span a face of τ , so span a simplex of L.
We now show that f is homotopic to |g|. If |L| ⊂ Rm , we try to define a homotopy
by the formula

H : |K| × I −→ |L|
(x, t) 7−→ t · f (x) + (1 − t) · |g|(x),

where the linear interpolation takes place in Rm . This certainly defines a continuous
map to Rm , and we must check that it lands in |L| ⊂ Rm .
Let x ∈ σ̊ ⊂ |K| and suppose that f (x) ∈ τ̊ ⊂ |L|. If σ = ha0 , a1 , . . . , ap i then by the
argument above each g(ai ) is a vetex of τ . Hence
p p
!
X X
|g|(x) = |g| ti ai = ti g(ai )
i=0 i=0

is a linear combination of vertices of τ , so lies in τ too. But then f (x) and |g|(x) both
lie in τ , so the straight line between them does too.

Definition 6.2.3. The barycentre of the simplex σ = ha0 , a1 , . . . , ap i is the point

a +a1 +···+ap
σ̂ = 0 p+1 . The (first) barycentric subdivision of a simplicial complex K is the
simplicial complex

K 0 := {hσ̂0 , σ̂1 , . . . , σ̂p i | σi ∈ K and σ0 < σ1 < · · · < σp }.

We define the rth barycentric subdivision K (r) inductively as (K (r−1) )0 .

Figure 6.4 The barycentric subdivision of the 2-simplex ha0 , a1 , a2 i, with the simplex
hha
d \ \
2 i, ha0 , a2 i, ha0 , a1 , a2 ii indicated in grey.

It is not obvious that the collection of simplices K 0 described in this definition actually
forms a simplicial complex. To see that it does is rather involved, and we collect the
necessary arguments in the following proposition.

Proposition 6.2.4. The barycentric subdivision K 0 of a simplicial complex K is indeed

a simplicial complex, and |K 0 | = |K|.
 6.2 Simplicial approximation 51

Proof.
If σ0 < σ1 < · · · < σp then the σ̂i are affinely independent. Suppose that
p
X p
X
ti σ̂i = 0 and ti = 0.
i=0 i=0

Let j = max{i | ti 6= 0}; then

j−1
X ti
σ̂j = − σ̂i ∈ σj−1 ,
tj
i=0

so σ̂j lies in a proper face of σj , which is impossible.

K0 is a simplicial complex. Let hσ̂0 , σ̂1 , . . . , σ̂p i ∈ K 0 . A face is given by omitting
some σ̂j ’s from σ0 < σ1 < · · · < σp , but omitting some σ̂j ’s from such a sequence still
gives a sequence of proper faces.
Let σ 0 = hσ̂0 , σ̂1 , . . . , σ̂p i and τ 0 = hτ̂0 , τ̂1 , . . . , τ̂q i, and consider σ 0 ∩ τ 0 . This lies inside
σp ∩ τq , which is a simplex δ of K, and there are simplices

σ 00 = hσ̂0 ∩ δ, σ̂1 ∩ δ, . . . , σ̂p ∩ δi and τ 00 = hτ̂0 ∩ δ, τ̂1 ∩ δ, . . . , τ̂q ∩ δi

of K 0 (where if σ̂i ∩ δ = ∅ we ignore it). Now σ 0 ∩ τ 0 = σ 00 ∩ τ 00 ⊂ δ, which reduces to the

case that everything lies within a single simplex of K, namely δ.
We now proceed by induction on the dimension of δ, in cases:

(i) σ 00 and τ 00 contain δ̂: let σ̄ 00 be the face of σ 00 obtained by removing δ̂, and similarly
τ̄ 00 . Then σ 00 ∩ τ 00 is the simplex spanned by σ̄ 00 ∩ τ̄ 00 and δ̂, and σ̄ 00 ∩ τ̄ 00 lies in the
boundary ∂δ. This is smaller dimensional than δ, so by induction we may assume
that σ̄ 00 ∩ τ̄ 00 is a simplex of K 0 , but then σ 00 ∩ τ 00 is too.

(ii) σ 00 or τ 00 does not contain δ̂: then σ 00 ∩ τ 00 ⊂ ∂δ lies in a smaller dimensional

simplex, so by induction we may assume it is a simplex of K 0 .

|K0 | = |K|. Note that hσ̂0 , σ̂1 , . . . , σ̂p i ⊆Pσp , so |K 0 | ⊆ |K|. On the other hand, for
x ∈ σ = ha0 , a1 , . . . , ap i written as x = pi=0 ti ai , we may reorder the ai so that t0 ≥
t1 ≥ · · · ≥ tp . Then

a0 + a1 a0 + a1 + a2
x = (t0 − t1 )a0 + 2(t1 − t2 ) + 3(t2 − t3 ) + ···
2 3
= (t0 − t1 )ha
d \ \
0 i + 2(t1 − t2 )ha0 , a1 i + 3(t2 − t3 )ha0 , a1 , a2 i + · · ·
D E
∈ had \ \
0 i, ha0 , a1 i, . . . , ha0 , a1 , . . . , ap i

which is a simplex of K 0 .

We now consider the following construction, which compares K and K 0 . First note
that the vertices of K 0 are precisely the barycentres of simplices of K, so there is a
52 Chapter 6 Simplicial complexes

bijection VK 0 ∼
= K. Thus if we choose a function K → VK which sends each simplex σ
to a vertex vσ of σ, then we obtain a function

g : VK 0 −→ VK
σ̂ 7−→ vσ .

If hσ̂0 , σ̂1 , . . . , σ̂p i is a simplex of K 0 then σ0 < σ1 < · · · < σp , and so all vσi are vertices
of σp , and so hvσ0 , vσ1 , . . . , vσp i is a simplex of K. Thus g defines a simplicial map.
Furthermore, if the point σ̂ lies in a simplex τ 0 = hτ̂0 , τ̂1 , . . . , τ̂p i ∈ K 0 , then σ̂ ∈ τp
and so σ is a face of τp . In particular, vσ ∈ σ ⊂ τp . Thus τ̊ 0 ⊂ τ̊p ⊂ StK (vσ ), and hence

StK 0 (σ̂) ⊂ StK (vσ ),

so the simplicial map g is a simplicial approximation to Id : |K 0 | → |K|.

Definition 6.2.5. The mesh of K is

µ(K) := max{kv0 − v1 k | hv0 , v1 i ∈ K}.

We will use the mesh as a measure of how fine a triangulation is, and will need to
know that sufficiently many barycentric subdivisions will make the mesh as small as we
like.
n r
Lemma 6.2.6. Suppose that K has dimension at most n. Then µ(K (r) ) ≤ ( n+1 ) ·µ(K),
(r)
and so µ(K ) → 0 as r → ∞.

Proof. This will follow by indutcion from the case r = 1. Let hτ̂ , σ̂i ∈ K 0 , so τ < σ as
simplices of K. Firstly we estimate

kτ̂ − σ̂k ≤ max{kv − σ̂k | v a vertex of σ},

as the euclidean distance to a point on a simplex is maximised at the vertices. Now

write σ = hv0 , v1 , . . . , vm i, with m ≤ n, and suppose that v0 is chosen so that kv0 − σ̂k
is maximal. Then we calculate
m
X
1
kv0 − σ̂k = kv0 − m+1 vi k
i=0
m
X
= k m+1
m+1 v0 −
1
m+1 vi k
i=0
m
X
1
= m+1 k (vi − v0 )k
i=1
m
X
1
≤ m+1 kvi − v0 k
i=1
m
≤ ·
m+1 µ(σ)
n
≤ n+1 · µ(K).
 6.2 Simplicial approximation 53

Theorem 6.2.7 (Simplicial approximation theorem). Let f : |K| → |L| be a continuous

map. Then there is a simplicial map g : K (r) → L for some r, such that g is a simplicial
approximation to f .
Furthermore, if N is a subcomplex of K and f ||N | : |N | → |L| is already simplicial,
then we may choose g to agree with f ||N | on VN ⊂ VK ⊂ VK (r) .

Proof. Consider the open cover {f −1 (StL (w))}w∈VL of |K|. This has a Lesbegue number
δ (with respect to the euclidean metric on |K|), and by Lemma 6.2.6 we may choose an
r such that µ(K (r) ) < δ. Then for each v ∈ VK (r) we have

StK (r) (v) ⊂ Bµ(K (r) ) (v) ⊂ f −1 (StL (w))

for some w ∈ VL , and setting g(v) = w we have

f (StK (r) (v)) ⊂ StL (g(v)).

Thus g is a simplicial approximation to f .

For the addendum, note that if v ∈ VN then f (v) is a vertex of L, so by increasing r
we may in addition suppose that

StK (r) (v) ⊂ Bµ(K (r) ) (v) ⊂ f −1 (StL (f (v)))

for each v ∈ VN , and hence choose g(v) := f (v) for these v.

 Chapter 7

Homology

7.1 Simplicial homology

Definition 7.1.1. Let K be a simplicial complex, and On (K) be the free abelian group
with basis given by the symbols

{[v0 , v1 , . . . , vn ] | v0 , v1 , . . . vn span a simplex of K}.

Note that the vi are considered to be ordered, and they could span a simplex of dimension
less than n (i.e. the list could have repeats).
Let Tn (K) ≤ On (K) be the subgroup generated by the elements
(i) [v0 , v1 , . . . , vn ] if the sequence contains repeated vertices,

(ii) [v0 , v1 , . . . , vn ] − sign(σ) · [vσ(0) , vσ(1) , . . . , vσ(n) ] for a permutation σ of {0, 1, . . . , n}.
Then we define Cn (K) := On (K)/Tn (K) to be the quotient group.
Lemma 7.1.2. There is a (non-canonical) isomorphism Cn (K) ∼
= Z# n-simplices of K .
Proof. Choose a total order ≺ on VK . Then each n-simplex σ ∈ K determines a canonical
ordered simplex [σ] given by ordering the vertices ai of σ so that a0 ≺ a1 ≺ · · · ≺ an .
This defines a homomorphism

φ : Zn-simplices of K −→ On (K)
σ 7−→ [σ]

and so a homomorphism

φ0 : Zn-simplices of K −→ Cn (K)
σ 7−→ [[σ]] .

For each symbol [a0 , a1 , . . . , an ] there is a unique permutation τ of {0, 1, . . . , n} such

that aτ (0) ≺ aτ (1) ≺ · · · ≺ aτ (n) , and we write sign([a0 , a1 , . . . , an ]) := sign(τ ). Using this
we define a homomorphism

ρ : On (K) −→ Zn-simplices of K
(
sign([a0 , a1 , . . . , an ]) · ha0 , a1 , . . . , an i if there are no repeats
[a0 , a1 , . . . , an ] 7−→
0 if there are repeats.

By construction Tn (K) ≤ Ker(ρ), and so ρ induces a homomorphism

ρ0 : Cn (K) −→ Zn-simplices of K .

54
 7.1 Simplicial homology 55

Now it is easy to see that ρ0 ◦ φ0 (σ) = σ, and so ρ0 ◦ φ0 = Id. Furthermore if

[a0 , a1 , . . . , an ] has no repeats then

φ0 ◦ ρ0 ([a0 , a1 , . . . , an ]) = sign([a0 , a1 , . . . , an ]) · [ha0 , a1 , . . . , an i]

= sign(τ ) · [aτ (0) , aτ (1) , . . . , aτ (n) ]

where τ is such that aτ (0) ≺ aτ (1) ≺ · · · ≺ aτ (n) , but in the quotient Cn (K) this is
equivalent to [a0 , a1 , . . . , an ]. As symbols [a0 , a1 , . . . , an ] with repeats are trivial in Cn (K),
it follows that φ0 ◦ ρ0 = Id.
We now define a homomorphism

dn : On (K) −→ On−1 (K)

n
X
[v0 , v1 , . . . , vn ] 7−→ (−1)i · [v0 , v1 , . . . , v̂i , . . . , vn ],
i=0

where [v0 , v1 , . . . , v̂i , . . . , vn ] denotes the sequence obtained by removing vi .

Lemma 7.1.3. The homomorphism dn sends Tn (K) into Tn−1 (K).
Proof. First note that dn ([v0 , v1 , . . . , vn ] − sign(σ)[vσ(0) , vσ(1) , . . . , vσ(n) ]) is
n
X n
X
i
(−1) [v0 , v1 , . . . , v̂i , . . . , vn ] − (−1)i sign(σ)[vσ(0) , vσ(1) , . . . , v̂σ(i) , . . . , vσ(n) ].
i=0 i=0
(7.1.1)
We must show that this is trivial in On−1 (K)/Tn−1 (K).
If σ = (j, j + 1), so sign(σ) = −1, then we compute
n
X j−1
X
i
(−1) sign(σ)[vσ(0) , . . . , v̂σ(i) , . . . , vσ(n) ] = (−1)i+1 [v0 , . . . , v̂i , . . . , vj−1 , vj+1 , vj , . . . , vn ]
i=0 i=0
+ (−1)j+1 [v0 , . . . , vj−1 , vj , vj+2 , . . . , vn ]
+ (−1)j+2 [v0 , . . . , vj−1 , vj+1 , vj+2 , . . . , vn ]
Xn
+ (−1)i+1 [v0 , . . . , vj−1 , vj+1 , vj , . . . , v̂i , . . . vn ].
i=j+2

Now the terms in the first sum satisfy

[v0 , . . . , v̂i , . . . , vj−1 , vj+1 , vj , . . . , vn ] ≡ −[v0 , . . . , v̂i , . . . , vj−1 , vj , vj+1 , . . . , vn ] mod Tn−1 (K)

and those in the second sum satisfy

[v0 , . . . , vj−1 , vj+1 , vj , . . . , v̂i , . . . vn ] ≡ −[v0 , . . . , vj−1 , vj , vj+1 , . . . , v̂i , . . . vn ] mod Tn−1 (K)

so in On−1 (K)/Tn−1 (K) we get

n
X n
X
i
(−1) sign(σ)[vσ(0) , vσ(1) , . . . , v̂σ(i) , . . . , vσ(n) ] ≡ (−1)i [v0 , v1 , . . . , v̂i , . . . , vn ]
i=0 i=0
56 Chapter 7 Homology

as required. This establishes that the element (7.1.1) is trivial in On−1 (K)/Tn−1 (K)
whenever σ is a transposition: as any permutation σ is a product of transpositions of
the form (j, j + 1), it follows for arbitrary σ.
Now suppose that we have [v0 , v1 , . . . , vn ] with vj = vj+1 . Then

j−1
X
dn ([v0 , v1 , . . . , vn ]) = (−1)i [v0 , . . . , v̂i , . . . , vj , vj+1 , . . . , vn ]
i=0
+ (−1)j [v0 , . . . , . . . , vj−1 , vj+1 , . . . , vn ]
+ (−1)j+1 [v0 , . . . , . . . , vj , vj+2 , . . . , vn ]
Xn
+ (−1)i [v0 , . . . , vj , vj+1 , . . . , v̂i , . . . , vn ]
i=j+2

but the middle two terms cancel out, and the outer two terms contain repeated entries
so lie in Tn−1 (K). In general, if [v0 , v1 , . . . , vn ] with vj = vk then applying σ = (j, k + 1)
reduces to the case k = j + 1.

The consequence of this lemma is that dn : On (K) → On−1 (K) induces a homomor-
phism
dn : Cn (K) −→ Cn−1 (K).

Lemma 7.1.4. The homomorphism dn−1 ◦ dn : Cn (K) → Cn−2 (K) is trivial.

Proof. We compute, at the level of On (K),

n
!
X
dn−1 ◦ dn ([v0 , . . . , vn ]) = dn−1 (−1)i · [v0 , . . . , v̂i , . . . , vn ]
i=0
n
X i−1
X
= (−1)i (−1)k [v0 , . . . , v̂k , . . . v̂i , . . . , vn ]
i=0 k=0
n−1
!
X
+ (−1)k [v0 , . . . , v̂i , . . . v̂k+1 , . . . , vn ]
k=i

The coefficient of [v0 , . . . , v̂a , . . . v̂b , . . . , vn ] is (−1)a (−1)b from k = a and i = b plus
(−1)a (−1)b−1 from i = a and k + 1 = b. These cancel out, so dn−1 ◦ dn ([v0 , . . . , vn ]) = 0.
the symbols [v0 , . . . , vn ] generate Cn (K), so dn−1 ◦ dn = 0.

The consequence of this lemma is that we have the containment Im(dn ) ⊂ Ker(dn−1 ).
Homology is the measurement of how different these two groups are.

Definition 7.1.5. The ith simplicial homology group of a simplicial complex K is

Ker(dn : Cn (K) → Cn−1 (K))

Hi (K) := .
Im(dn+1 : Cn+1 (K) → Cn (K))
 7.2 Some homological algebra 57

Example 7.1.6 (Homology of the simplicial circle). Consider the simplicial complex
K inside R3 given by all proper faces of the standard 2-simplex ∆2 ⊂ R3 . If we write
{e1 , e2 , e3 } for the standard basis of R3 , then K has simplices
he1 i, he2 i, he3 i, he1 , e2 i, he1 , e3 i, he2 , e3 i.
Choosing the order e1 ≺ e2 ≺ e3 on the vertices, we obtain isomorphisms
C0 (K) ∼
= Z{[e1 ], [e2 ], [e3 ]}
∼
C1 (K) = Z{[e1 , e2 ], [e1 , e3 ], [e2 , e3 ]},
so both groups are free abelian of rank three, and have the bases shown. With respect
to these isomorphisms, we have
d1 ([ei , ej ]) = [ej ] − [ei ],
so in the given bases d1 is given by the matrix
 
−1 −1 0
 1 0 −1  .
0 1 1
Hence
H0 (K) = Z{[e1 ], [e2 ], [e3 ]}/h[ei ] − [ej ] | i, j ∈ {1, 2, 3}i ∼
=Z
and
H1 (K) = Ker(d1 ) = Z{[e1 , e2 ] − [e1 , e3 ] + [e2 , e3 ]} ∼
= Z.
Also Hi (K) = 0 for i ≥ 2, as K has no simplices of dimension 2 or higher.
Example 7.1.7 (Homology of the 2-simplex). Now consider the simplicial complex L
in R3 given by the standard 2-simplex ∆2 in R3 along with all its faces. The simplicial
complex L is similar to K from the previous example, but has a 2-simplex as well. Thus
we have
d2 d1
C2 (L) / C1 (L) / C0 (L)

[e1 , e2 , e3 ]  / [e1 , e2 ] − [e1 , e3 ] + [e2 , e3 ]

where C1 (L) = C1 (K) and C0 (L) = C0 (K), and the formula for d1 is the same as for K.
Now d2 is injective, so H2 (L) = 0. Also Im(d2 ) = Z{[e1 , e2 ] − [e1 , e3 ] + [e2 , e3 ]} =
Ker(d1 ) and so H1 (L) = 0. Finally, we obtain H0 (L) ∼
= Z just as in the previous example.

7.2 Some homological algebra

Definition 7.2.1. A chain complex is a sequence C0 , C1 , C2 . . . of abelian groups

equipped with homomorphisms dn : Cn → Cn−1 such that dn−1 ◦ dn = 0. We write it as
C• , and call the dn the differentials of C• . The nth homology of C• is
Ker(dn : Cn → Cn−1 )
Hn (C• ) := .
Im(dn+1 : Cn+1 → Cn )
58 Chapter 7 Homology

We write Zn (C• ) := Ker(dn ), and call this the abelian group of n-cycles in C• , and
Bn (C• ) := Im(dn+1 ), and call this the abelian group of n-boundaries in C• .
A chain map f• : C• → D• is a sequence of homomorphisms fn : Cn → Dn such
that fn ◦ dn+1 = dn+1 ◦ fn+1 ; in other words, such that the square

fn+1
Cn+1 / Dn+1

dn+1 dn+1
 fn 
Cn / Dn

commutes.
A chain homotopy from f• to g• (two chain maps from C• to D• ) is a sequence of
homomorphisms hn : Cn → Dn+1 such that

gn − fn = dn+1 ◦ hn + hn−1 ◦ dn : Cn → Dn .

Note that in the above definitions we use the same symbol, dn , for the differential in
any chain complex. Which one we mean in any given situation will be clear from context.

Lemma 7.2.2. A chain map f• : C• → D• induces a homomorphism

f∗ : Hn (C• ) −→ Hn (D• )
[x] 7−→ [fn (x)].

Furthermore, if f• and g• are chain homotopic then f∗ = g∗ .

Proof. We need to check that f∗ is well-defined.

Zn (C• )
(i) Let [x] ∈ Hn (C• ) = B n (C• )
be represented by x ∈ Zn (C• ). The element fn (x) ∈ Dn
is a cycle, as dn (fn (x)) = fn−1 (dn (x)) = fn−1 (0) = 0, because x is a cycle in Cn .
Thus fn (x) does represent a homology class.

(ii) If [x] = [y], then x − y ∈ Bn (C• ), and so x − y = dn+1 (z) for some z ∈ Cn+1 . Then
fn (x) − fn (y) = fn (dn+1 (z)) = dn+1 (fn+1 (z)) is a boundary, so [fn (x)] = [fn (y)] ∈
Hn (D• ).

Now suppose that f• and g• are chain homotopic via a chain homotopy h• , and let
x ∈ Zn (C• ). We have

gn (x) − fn (x) = dn+1 (hn (x)) + hn−1 (dn (x))

but x is a cycle so dn (x) = 0. Thus gn (x) − fn (x) is a boundary, but then [gn (x)] =
[fn (x)].

Just as we did with the notion of homotopy of maps between spaces, we check that

(i) being chain homotopic defines an equivalence relation on the set of chain maps from
C• to D• , written f• ' g• ,
 7.3 Elementary calculations 59

(ii) if a• : A• → C• is a chain map, and f• ' g• : C• → D• , then f• ◦ a• ' g• ◦ a• , and

similarly with precomposition.

Definition 7.2.3. A chain map f• : C• → D• is a chain homotopy equivalence if

there is a chain map g• : D• → C• and chain homotopies f• ◦g• ' IdD• and g• ◦f• ' IdC• .

Lemma 7.2.4. If f• : C• → D• is a chain homotopy equivalence, then f∗ : Hn (C• ) →

Hn (D• ) is an isomorphims for all n.

Proof. We have f∗ ◦ g∗ = (f• ◦ g• )∗ = (IdD• )∗ = IdHn (D• ) , and vice versa.

7.3 Elementary calculations

Returning to simplicial complexes, note that for a simplicial complex K we have defined
in Section 7.1 a chain complex C• (K), which we call the simplicial chain complex
of K. We want to show that a simplicial map f : K → L induces a chain map f• :
C• (K) → C• (L).

Lemma 7.3.1. Let f : K → L be a simplicial map. Then the formula

fn : Cn (K) −→ Cn (L)
[a0 , a1 , . . . , an ] 7−→ [f (a0 ), f (a1 ), . . . , f (an )]

defines a chain map f• : C• (K) → C• (L), and hence a homomorphism f∗ : Hn (K) →

Hn (L).

Proof. We first need that fn (Tn (K)) ⊂ Tn (L), but this is clear. Then we compute

n
!
X
i
fn−1 (dn ([a0 , a1 , . . . , an ])) = fn−1 (−1) · [a0 , . . . , âi , . . . , an ]
i=0
n
X
= (−1)i · [f (a0 ), . . . , f[
(ai ), . . . , f (an )]
i=0
= dn (fn ([a0 , a1 , . . . , an ])).

Definition 7.3.2. Say a simplicial complex K is cone with cone point v0 ∈ VK if

|K| = StK (v0 ) ∪ LkK (v0 ). In other words, every simplex of K is a face of a simplex
which contains v0 .

Proposition 7.3.3. If K is a cone with cone point v0 , then the inclusion i : {v0 } → K
induces a chain homotopy equivalence i• : C• ({v0 }) → C• (K). Hence
(
Z{[v0 ]} n=0
Hn (K) =
0 else.
60 Chapter 7 Homology

Proof. The (only) map r : Vk → {v0 } defines a simplicial map r : K → {v0 }. We will
show that r• is a chain homotopy inverse to i• . One direction is easy: the composition
r• ◦ i• : C• ({v0 }) → C• ({v0 }) is equal to the identity map.
For the other direction, define a homomorphism
hn : On (K) −→ On+1 (K)
[a0 , . . . , an ] 7−→ [v0 , a0 , . . . , an ]
and observe that it takes Tn (K) into Tn+1 (K), and hence gives a homomorphism hn :
Cn (K) → Cn+1 (K). If n > 0 then
n
X
(hn−1 ◦ dn + dn+1 ◦ hn )([a0 , . . . , an ]) = (−1)i [v0 , a0 , . . . , âi , . . . , an ]
i=0
+ [a0 , . . . , an ]
Xn
+ (−1)i+1 [v0 , a0 , . . . , âi , . . . , an ]
i=0
= [a0 , . . . , an ] = (Id − in ◦ rn )([a0 , . . . , an ]).
If n = 0 then
(h−1 ◦ d0 + d1 ◦ h0 )([a0 ]) = d1 ([v0 , a0 ]) = [a0 ] − [v0 ]
= (Id − i0 ◦ r0 )([a0 ]).
Thus h• is a chain homotopy from i• ◦ r• to IdC• (K) .
Corollary 7.3.4. Let ∆n be the standard simplex in Rn+1 , and L be the simplicial
complex given by ∆n and all its faces. Then
(
Z i=0
Hi (L) =
0 else.
Proof. Any vertex of L is a cone point.
Corollary 7.3.5. Let K be the simplicial complex given by all the proper faces of ∆n ⊂
Rn+1 . Then for n ≥ 2 we have
(
Z i = 0 or n − 1
Hi (K) =
0 else.

Proof. Note that K is the (n − 1)-skeleton of the simplicial complex L from the previous
corollary, so Ci (K) = Ci (L) for 0 ≤ i ≤ n − 1, and Ci (K) vanishes in higher degrees.
Thus the two chain complexes are

dL dL dL dL dL
C0 (L) o C1 (L) o ··· o Cn−2 (L) o Cn−1 (L) o
1 2 n−2 n−1 n
Cn (L)

dK dK dK dK
C0 (K) o C1 (K) o ··· o Cn−2 (K) o Cn−1 (K) o
1 2 n−2 n−1
0,
 7.4 The Mayer–Vietoris sequence 61

where we have indicated which complex the differentials belong to.

Now for 0 ≤ i ≤ n − 1 the two chain complexes are equal, and we deduce that
Hi (K) = Hi (L) for 0 ≤ i ≤ n − 2. In particular in this range the homology if Z in degree
0 and 0 otherwise.
Near the top we have Cn (L) = Z{[e1 , . . . , en+1 ]}. Firstly Hn (L) = 0 = Ker(dLn ) so
dn is injective, and secondly Hn−1 (L) = 0 so Ker(dL
L
n−1 ) = Im(d L ). Thus we have
n
L ∼
Ker(dK L
n−1 ) = Ker(dn−1 ) = Im(dn ) = Z

and so Hn−1 (K) ∼

= Z.
Now that we have some example computations of the homology of certain simplicial
complexes, it is interesting to ask what it means.
Lemma 7.3.6. There is an isomorphism H0 (K) ∼
= Z# path components of |K| .
Proof. First note that every path component of a polyhedron contains a vertex (as every
point lies in an open simplex, so can be joined by a straight line to any vertex of that
simplex). By definition H0 (K) = ZVK /Im(d1 ). If vertices v and w lies in the same path
component, we can choose a path between them and then find a simplicial approximation
to this path. This gives a sequence v = v0 , v1 , . . . , vn = w of vertices such that each
adjacent pair span a 1-simplex. Then [vi ] − [vi−1 ] = d1 ([vi−1 , vi ]) so summing over all i
gives [w] − [v] = d1 ([v0 , v1 ] + [v1 , v2 ] + · · · + [vn−1 , vn ]), so [v] = [w] ∈ H0 (K).

7.4 The Mayer–Vietoris sequence

We will develop a tool akin to the Seifert–van Kampen Theorem, which will allow us to
compute the homology of a simplicial complex by decomposing it into pieces, computing
their homology, and then assembling them again using an algebraic machine.
Definition 7.4.1. We say that a pair of homomorphisms of abelian groups
f g
A −→ B −→ C
are exact at B if Im(f ) = Ker(g). More generally, we say that a collection of homo-
morphisms
· · · −→ Ai −→ Ai+1 −→ Ai+2 −→ Ai+3 −→ Ai+4 −→ · · ·
is exact if it is exact at Ai whenever there is both a map entering and a map exiting Ai .
A short exact sequence of abelian groups is an exact sequence
f g
0 −→ A −→ B −→ C −→ 0,
that is, f is injective (exactness at A), g is surjective (exactness at C), and Im(f ) =
Ker(g) (exactness at B).
Chain maps i• : A• → B• and j• : B• → C• form a short exact sequence of chain
complexes if for each n
n i jn
0 −→ An −→ Bn −→ Cn −→ 0
is a short exact sequence of abelian groups.
62 Chapter 7 Homology

• i j•
Theorem 7.4.2. If 0 → A• → B• → C• → 0 is a short exact sequence of chain
complexes, then there are natural homomorphisms ∂∗ : Hn (C• ) → Hn−1 (A• ) such that
∂∗ i∗ j∗
··· / Hn (A• ) / Hn (B• ) / Hn (C• )

∂∗
i∗ j∗
/ Hn−1 (A• ) / Hn−1 (B• ) / Hn−1 (C• ) / ···

is exact.
Let us defer the proof of this theorem.
Theorem 7.4.3 (Mayer–Vietoris). Let K be a simplicial complex, M and N be subcom-
plexes, and L = N ∩ M . Suppose that M and N cover K (i.e. every simplex of K lies
in M or N ). Let us write

L  _
j
/ N
_
i l
  k 
M /K
for the inclusion maps between these simplicial complexes. Then there are natural homo-
morphisms ∂∗ : Hn (K) → Hn−1 (L) such that
∂∗ i∗ +j∗ k∗ −l∗
··· / Hn (L) / Hn (M ) ⊕ Hn (N ) / Hn (K)

∂∗
/ Hn−1 (L) i∗ +j∗ / Hn−1 (M ) ⊕ Hn−1 (N ) k∗ −l∗ / Hn−1 (K) / ···

i∗ +j∗ k∗ −l∗
··· / H0 (M ) ⊕ H0 (N ) / H0 (K) /0

is exact.
Proof. Note that for each n the sequence of abelian groups
in +jn n n k −l
0 −→ Cn (L) −→ Cn (M ) ⊕ Cn (N ) −→ Cn (K) −→ 0

is exact:
(i) in and jn are both injective,

(ii) every simplex of K lies in M or N , so kn − ln is surjective,

(iii) if (x, y) ∈ Ker(kn − ln ) then kn (x) = ln (y), so x and y are both sums of simplices
in both M and N , so lie in L and are equal.
Thus the chain maps i• + j• : C• (L) → C• (M ) ⊕ C• (N ) and k• − l• : C• (M ) ⊕ C• (N ) →
C• (K) form a short exact sequence of chain complexes, so Theorem 7.4.2 applies, giving
the result.
 7.4 The Mayer–Vietoris sequence 63

Proof of Theorem 7.4.2. The proof consists of what is affectionately known as “diagram-
chasing", that is, following the logical consequences of an element being at a certain place
in a commutative diagram.
Constructing ∂∗ (The Snake Lemma). Consider the commutative diagram

in jn
0 / An / Bn / Cn /0

d d d
 in−1  jn−1 
0 / An−1 / Bn−1 / Cn−1 /0

and let [x] ∈ Hn (C• ) be a homology class, represented by a cycle x ∈ Cn . As jn is

surjective, there is a y ∈ Bn such that jn (y) = x. Then d(y) ∈ Bn−1 satisfies

jn−1 (d(y)) = d(jn (y)) = d(x) = 0

as x was a cycle. By exactness of the bottom row at Bn−1 , we thus have d(y) = in−1 (z)
for some z ∈ An−1 .
Now we have in−2 (d(z)) = d(in−1 (z)) = d(d(y)) = 0, and in−2 is injective, so d(z) =
0. Thus z representes a homology class ∂∗ ([x]) := [z] ∈ Hn−1 (A• ).
∂∗ is well-defined. If [x] = [x0 ] ∈ Hn (C• ) then x − x0 = d(a) for some a ∈ Cn+1 .
Proceeding as above for x0 , we get a y 0 ∈ Bn . As jn+1 is surjective, let b ∈ Bn+1 be such
that jn+1 (b) = a, then

jn (y − y 0 ) = x − x0 = d(a) = d(jn+1 (b)) = jn (d(b)),

so by exactness at Bn we have y − y 0 = d(b) + in (c) for some c ∈ An . Thus

d(y) − d(y 0 ) = d(in (c)) = in−1 (d(c)),

so by injectivity of in−1 we get z − z 0 = d(c). Thus [z] = [z 0 ] ∈ Hn−1 (A• ).

∂∗ is a homomorphism. Given [x], [x0 ] ∈ Hn (C• ), we can choose y + y 0 as the lift to Bn
of x+x0 ∈ Cn . It is then immediate that ∂n ([x+x0 ]) = [z +z 0 ], so ∂∗ is a homomorphism.
The sequence is exact at Hn (C• ). First let [x] ∈ Im(j∗ ). Then we can write x = jn (y)
for y ∈ Bn a cycle. The element y is then a lift of x, and it satisfies d(y) = 0. Thus
∂∗ ([x]) = 0.
Now suppose that ∂∗ ([x]) = 0. Thus the z obtained by the Snake Lemma construction
satisfies z = d(t) for some t ∈ An , and we have jn (y − in (t)) = x. But d(y − in (t)) =
d(y) − d(in (t)) = d(y) − in−1 (d(t)) = d(y) − in−1 (z) which is 0 by the way we chose z.
Thus y − in (t) is a cycle, and [x] = j∗ ([y − in (t)]), so [x] ∈ Im(j∗ ).
The sequence is exact at Hn (B• ). It is clear that Im(i∗ ) ⊂ Ker(j∗ ), as jn ◦ in = 0.
Let y be a cycle so that j∗ ([y]) = 0, so jn (y) = d(a) for some a ∈ Cn+1 . Let b ∈ Bn+1 be
such that jn+1 (b) = a. Then

jn (y − d(b)) = d(a) − d(jn+1 (b)) = 0

64 Chapter 7 Homology

and so y − d(b) = in (t) for some t ∈ An . As y is a cycle, so is in (t), and as in is injective

it follows that t is a cycle. But then i∗ ([t]) = [y − d(b)] = [y], so Im(i∗ ) ⊃ Ker(j∗ ).
The sequence is exact at Hn (A• ). If ∂∗ ([x]) = [z] as constructed via the Snake
Lemma, then in (x) = d(y), so i∗ ([x]) = 0. Thus Im(∂∗ ) ⊂ Ker(i∗ ). Let z be a cycle so
that i∗ ([z]) = 0, that is in (z) = d(y). Then d(jn+1 (y)) = jn (d(y)) = jn (in (z)) = 0, so
jn+1 (y) is a cycle. By construction ∂∗ ([jn+1 (y)]) = [z], so Im(∂∗ ) ⊃ Ker(i∗ ).

7.5 Continuous maps and homotopy invariance

In this section we want to show that to each map f : |K| → |L| we may associate a
homomorphism f∗ : Hi (K) → Hi (L), even when f does not come from a simplical map,
and further that if f ' f 0 then f∗ = f∗0 .
Definition 7.5.1. Simplicial maps f, g : K → L are contiguous if for each σ ∈ K the
simplices f (σ) and g(σ) are faces of some simplex τ ∈ L.
Lemma 7.5.2. If f, g : K → L are both simplicial approximations to the same map,
then they are contiguous.
Proof. Let F : |K| → |L| be a continuous map which f and g are simplicial approxima-
tions to. If x ∈ σ̊ ⊂ |K| and F (x) ∈ τ̊ ⊂ |L|, then as in the proof of Lemma 6.2.2 we
have that f (σ) and g(σ) are both faces of τ .
Lemma 7.5.3. If f, g : K → L are contiguous, then f• ' g• : C• (K) → C• (L) and so
f∗ = g∗ .
Proof. Choose an ordering ≺ on VK , so we can represent basis elements for Cn (K) by
ordered simplices [a0 , . . . , an ] such that a0 ≺ a1 ≺ · · · ≺ an . Define a homomorphism
hn : Cn (K) → Cn+1 (L) on basis elements by
n
X
hn ([a0 , . . . , an ]) = (−1)i [f (a0 ), . . . , f (ai ), g(ai ), . . . , g(an )].
i=0

Direct calculation shows that d ◦ hn + hn−1 ◦ d = gn − fn , so the hn provide a chain

homotopy from f• to g• .
Let K 0 be the barycentric subdivision of a simplicial complex K, and for each vertex
σ̂ ∈ VK 0 choose a vertex a(σ̂) of σ, to obtain a function a : VK 0 → VK .
Lemma 7.5.4. The function a : VK 0 → VK defines a simplicial approximation to the
identity map. Furthermore, every simplicial approximation g : K 0 → K to the identity
map is of this form.
Proof. The first claim is that StK 0 (σ̂) ⊆ StK (a(σ̂)). To see this, suppose that σ̂ is a
vertex of a simplex τ = hσ̂0 , σ̂1 , . . . , σ̂p i of K 0 , say σ̂ = σ̂i . Then a(σ̂) is a vertex of σi ,
and hence a vertex of σp . In particular τ̊ ⊂ σ̊p ⊂ StK (a(σ̂)).
If g : K 0 → K is a simplicial approximation to the identity map then we have
σ̊ ⊆ StK (g(σ̂)), but the open simplices in StK (g(σ̂)) are precisely those whose closure
contains g(σ̂), so g(σ̂) must be a vertex of σ.
 7.5 Continuous maps and homotopy invariance 65

Proposition 7.5.5. For a : K 0 → K a simplicial approximation to the identity map, the

induced map a∗ : Hn (K 0 ) → Hn (K) is an isomorphism for all n.

Proof. First consider the case where K consists of a single p-simplex, σ, and all its faces.
Note that K 0 is a cone with cone point σ̂, and we have seen that K is also a cone, with any
vertex as cone point. By Proposition 7.3.3 we can thus calculate the homology of either
of these simplicial complexes, and we find that it is 0 in strictly positive degrees, and
Z in degree zero, generated by any vertex. Thus by direct computation, any simplicial
approximation to the identity a : K 0 → K induces an isomorphism on homology in every
degree.
Now let us prove the proposition by double induction on (i) the dimension of K,
and (ii) the number of simplices of K of maximal dimension. If σ is a simplex of K of
maximal dimension, then it is not a proper face of any simplex and so L := K \ {σ} is
also a simplicial complex. Let S be the simplicial complex given by σ and all its faces,
and T = S ∩ L. Any simplicial approximation to the identity a : K 0 → K sends L0 into
L and S 0 into S (so also T 0 into T ). We thus obtain a map between Mayer–Vietoris
sequences

Hn (T 0 ) / Hn (S 0 ) ⊕ Hn (L0 ) / Hn (K 0 ) / Hn−1 (T 0 ) / Hn−1 (S 0 ) ⊕ Hn−1 (L0 )

(a|T 0 )∗ (a|S 0 )∗ ⊕(a|L0 )∗ a∗ (a|T 0 )∗ (a|S 0 )∗ ⊕(a|L0 )∗

    
Hn (T ) / Hn (S) ⊕ Hn (L) / Hn (K) / Hn−1 (T ) / Hn−1 (S) ⊕ Hn−1 (L)

in which every square commutes.

Now S is a simplex, so the proposition holds for it, T has strictly smaller dimension
than K, so the proposition holds for it, and L has either strictly smaller dimension than
K, or else has fewer simplices of maximal dimension, so the proposition also holds for it.
Thus all the vertical maps in the diagram above are known to be isomorphisms except
the middle one. An easy diagram-chase (known as the Five Lemma1 ) shows that the
middle map must then be an isomorphism too.

We will now use the isomorphisms of this proposition to identify Hn (K) with Hn (K 0 ),
and so by iteration to identify Hn (K) with Hn (K (r) ). Note that by Lemmas 7.5.2
and 7.5.3 all simplicial approximations to the identity a : K 0 → K induce the same
∼
isomorphism a∗ : H∗ (K 0 ) → H∗ (K). We call this isomorphism νK , and we write νK,r
for the composition νK (r−1) ◦ · · · ◦ νK 0 ◦ νK and, for r ≥ s, νK,r,s for the composition
νK (r−1) ◦ · · · ◦ νK (s+1) ◦ νK (s) .

Proposition 7.5.6. To each continuous map f : |K| → |L| there is an associated ho-
−1
momorphism f∗ : H∗ (K) → H∗ (L) given by f∗ = s∗ ◦ νK,r , where s : K (r) → L is a
simplicial approximation to f , which exists for some r by the Simplicial Approximation
Theorem.

(i) This homomorphism does not depend on r or s.

(ii) If g : |M | → |K| is another continuous map, then (f ◦ g)∗ = f∗ ◦ g∗ .

1
See Example Sheet 4, Q3
66 Chapter 7 Homology

Proof. For (i), let s : K (r) → L and t : K (q) → L be simplicial approximations to f ,

with r ≥ q. Choose a : K (r) → K (q) a simplicial approximation to the identity. Then
s, t ◦ a : K (r) → L are both simplicial approximations to f , so are contiguous by Lemma
7.5.2, and hence induce the same homomorphism on homology by Lemma 7.5.3. Thus
−1 −1 −1
s∗ = t∗ ◦ a∗ , but a∗ = νK,r,q , so s∗ ◦ νK,r = t∗ ◦ νK,r,q ◦ νK,r = t∗ ◦ νK,q .
For (ii), let s : K → L be a simplicial approximation to f , and t : M (q) → K (r) be
(r)

a simplicial approximation to g. Then s ◦ t is a simplicial approximation to f ◦ g, so

−1
(f ◦ g)∗ = (s ◦ t)∗ ◦ νM,q
−1 −1
= (s∗ ◦ νK,r ) ◦ (νK,r ◦ t∗ ◦ νM,q ) = f∗ ◦ g∗ .
Corollary 7.5.7. If f : |K| → |L| is a homeomorphism, then f∗ : Hn (K) → Hn (L) is
an isomorphism for every n.
For some applications it is convenient to have a chain-level description of the map
−1
νK : Hi (K) → Hi (K 0 ), that is, a chain map s• : C• (K) → C• (K 0 ) which induces
−1
νK on homology. Morally speaking it is clear what such a map should do: it should
send a simplex σ of K to the sum of all the simplices of the barycentric subdivision σ 0 .
However there are signs involved. We begin by letting s0 ([v0 ]) = [hv d 0 i], and supposing
that s0 , s1 , . . . , sn−1 have been defined, satisfying di ◦ si = si−1 ◦ di for i < n, then for an
n-simplex we let
sn (σ) = [σ̂, sn−1 dn (σ)].
The notation [σ̂, −] must be understood to be linear in −, and expanded out to obtain
a linear combination of ordered simplices. So for example
\
s1 ([v0 , v1 ]) = [hv \ d \ d
0 , v1 i, hv1 i − hv0 i] = [hv0 , v1 i, hv1 i] − [hv0 , v1 i, hv0 i].
d d
−1
Proposition 7.5.8. The maps sn define a chain map s• , and s∗ = νK .
Proof. We calculate

dn (sn ([v0 , . . . , vn ])) = dn ([hv0 \

, . . . , vn i, sn−1 dn ([v0 , . . . , vn ])])
= sn−1 dn ([v0 , . . . , vn ]) − [hv0 \
, . . . , vn i, dn−1 sn−1 dn ([v0 , . . . , vn ])]
and as dn−1 ◦ sn−1 = sn−2 ◦ dn−1 and dn−1 ◦ dn = 0 the second term vanishes, so the sn
define a chain map.
Let ≺ be a total order of VK , and let a : K ∼ = VK 0 → VK send each simplex to
its smallest vertex (with respect to ≺). This defines a simplicial approximation to the
identity a : K 0 → K. Certainly a0 ◦ s0 is the identity, so suppose that ai ◦ si is the
identity for i < n. If [v0 , . . . , vn ] is a simplex with v0 ≺ · · · ≺ vn then

an (sn ([v0 , . . . , vn ])) = an ([hv0 \

, . . . , vn i, sn−1 dn ([v0 , . . . , vn ])])
= [v0 , an−1 sn−1 dn ([v0 , . . . , vn ])]
and as an−1 ◦sn−1 is the identity this is [v0 , dn ([v0 , . . . , vn ])]. Expanding out dn ([v0 , . . . , vn ]),
the first term is [v1 , . . . , vn ] and the other all have a v0 in them, so when v0 is added they
become zero in Cn (K). Thus we obtain [v0 , . . . , vn ], as required.
 7.5 Continuous maps and homotopy invariance 67

To obtain the strongest applications of simplicial homology theory, we wish to know

that continuous maps between polyhedra which are homotopic induce equal homomor-
phisms on homology. We first show that maps of polyhedra which are sufficiently close
induce equal homomorphisms in homology.
Lemma 7.5.9. For a simplicial complex L in Rm , there is a  = (L) > 0 such that if
f, g : |K| → |L| satisfy |f (x) − g(x)| <  for all x ∈ |K|, then f∗ = g∗ : Hn (K) → Hn (L).
Proof. The sets {StL (w)}w∈VL give an open cover of |L|, which is compact, so by the
Lesbegue Number Lemma (Lemma 1.1.3) there is a  > 0 such that each ball of radius
2 in |L| lies in some StL (w).
Let f, g : |K| → |L| be as in the statement, using this . By the Lesbague Number
Lemma applied to the open cover {f −1 (B (y))}y∈|L| of |K|, there is a δ > 0 such that
each f (Bδ (x)) is contained in some B (y), and hence also each g(Bδ (x)) is contained in
B2 (y).
Choose r  0 so that µ(K (r) ) < 21 δ. Then for each v ∈ VK (r) the diameter of StK (r) (v)
is less than δ, so f (StK (r) (v)) and g(StK (r) (v)) both lie in some StL (w). Setting s(v) = w,
we obtain s : VK (r) → VL which is a simplicial approximation to both f and g. Thus
−1
f∗ := s∗ ◦ νK,r =: g∗ .
Theorem 7.5.10. If f ' g : |K| → |L|, then f∗ = g∗ .
Proof. Let H : |K| × I → |L| be a homotopy from f to g. As |K| × I is compact, H
is uniformly continuous. Thus for  = (L) provided by the previous lemma, there is a
δ > 0 such that
|s − t| < δ ⇒ |H(x, s) − H(x, t)| <  ∀x ∈ |K|.
Thus choose 0 = t0 < t1 < t2 < · · · < tk = 1 such that ti − ti−1 < δ for all i, and let
fi (x) = H(x, ti ). Then by the previous lemma (fi−1 )∗ = (fi )∗ for all i, but f0 = f and
fk = g.
This allows us to define and work with homology for a larger class of spaces than just
polyhedra.
Definition 7.5.11. A h-triangulation of a space X is a simplicial complex K and a
homotopy equivalence g : |K| → X. We define the homology of this h-triangulated
space to be Hn (X) := Hn (K).
Note that in particular a triangulation is an h-triangulation.
Lemma 7.5.12. The homology of an h-triangulated space does not depend on the choice
of h-triangulation.
Proof. Let ḡ : |K̄| → X be another h-triangulation, f : X → |K| be a homotopy inverse
to g, and f¯ : X → |K̄| be a homotopy inverse to ḡ. Then f ◦ ḡ : |K̄| → |K| is a homotopy
equivalence with homotopy inverse f¯ ◦ g. Thus
(f ◦ ḡ)∗ ◦ (f¯ ◦ g)∗ = IdHn (K)
and vice-versa, so (f ◦ ḡ)∗ : Hn (K̄) → Hn (K) is an isomorphism.
68 Chapter 7 Homology

If g : |K| → X and ḡ : |K̄| → X̄ are h-triangulated spaces, and h : X → X̄ is a

continuous map, then we obtain a continuous map
∼ h ∼
|K| −→ X −→ X̄ −→ |K̄|

and hence a homomorphism Hn (K) → Hn (K̄). Identifying these groups with Hn (X)
and Hn (X̄) respectively, we obtain a homomorphism

h∗ : Hn (X) → Hn (X̄).

One can check that this is independent of choice of h-triangulations of either space, and
that if h ' h0 : X → X̄ then h∗ = h0∗ .

7.6 Homology of spheres and applications

As a first application, we can use homology to upgrade our proof of Brouwer’s fixed point
theorem (Corollary 3.2.3) to all dimensions.

Lemma 7.6.1. The sphere S n−1 is triangulable, and for n ≥ 2 we have

(
Z i = 0, n − 1
Hi (S n−1 ) ∼
=
0 else.

Proof. If ∆n ⊂ Rn+1 is the standard n-simplex, then its boundary ∂∆n is homeomorphic
to S n−1 . This boundary is the polyhedron of the simplicial complex of Corollary 7.3.5,
where its homology was computed.

Theorem 7.6.2 (Brouwer).

(i) There is no retraction of Dn to S n−1 ,

(ii) Any continuous map f : Dn → Dn has a fixed point.

Proof. The argument we gave in Corollary 3.2.3 shows that (i) implies (ii). To prove (i),
suppose that n ≥ 2, let r : Dn → S n−1 be a proposed retraction, and i : S n−1 → Dn be
the inclusion, and consider
i r
Z∼
= Hn−1 (S n−1 ) −→ Hn−1 (Dn ) −→ Hn−1 (S n−1 ) ∼
∗ ∗
= Z.

This is the identity map, as r ◦ i = IdS n−1 . But Dn is contractible, so is homotopy

equivalent to the polyhedron of the simplicial complex consisting of a single vertex. As
n − 1 > 0, it follows that Hn−1 (Dn ) = 0, which is a contradiction.

Recall that in Example 6.1.9 we gave a different triangulation of S n , using the sim-
plicial complex K having simplices

h±e1 , ±e2 , . . . , ±en+1 i

 7.6 Homology of spheres and applications 69

in Rn+1 , and all their faces. By Lemma 7.5.12 we must also have
(
∼ Z i = 0, n
Hi (K) =
0 else,

as different h-triangulations of S n have the same homology.

Lemma 7.6.3. The element

X
x := a1 a2 · · · an+1 [a1 e1 , . . . , an+1 en+1 ] ∈ Cn (K)
a∈{−1,1}n+1

is a cycle, and generates Hn (K) ∼

= Z.

Proof. When we apply the differential d to x, the simplex [a1 e1 , . . . , ad

i ei , . . . an+1 en+1 ]
appears twice, with ai = 1 and ai = −1, and these have opposite coefficients so cancel
out. Also, this element is not divisible by any integer, so generates Z ∼ = Hn (K).
The reflection ri : Rn+1 → Rn+1 in the ith coordinate defines a simplicial map from
K to itself, and this satisfies
X
ri (x) = a1 a2 · · · an+1 [a1 e1 , . . . , ai−1 ei−1 , −ai ei , ai+1 ei+1 , an+1 en+1 ]
a∈{−1,1}n+1
X
=− a1 a2 · · · an+1 [a1 e1 , . . . , an+1 en+1 ] = −x
a∈{−1,1}n+1

so (ri )∗ : Hn (S n ) → Hn (S n ) is multiplication by −1. The antipodal map a : S n → S n

may be written as r1 r2 · · · rn+1 , so a∗ : Hn (S n ) → Hn (S n ) is multiplication by (−1)n+1 .

Corollary 7.6.4. If n is even then the antipodal map a : S n → S n is not homotopic to

the identity2 .

2
See Example Sheet 1, Q1
70 Chapter 7 Homology

7.7 Homology of surfaces, and their classification

Definition 7.7.1. A surface is a Hausdorff topological space in which each point has an
open neighbourhood homeomorphic to an open subset of R2 . A triangulated surface
is a surface X equipped with a triangulation h : |K| → X.

Example 7.7.2 (Orientable surfaces). Let K be the simplicial complex shown in Figure
7.1, which is a triangulation of the torus with an open disc removed, X.

Figure 7.1 A triangulation of X, the torus with an open disc removed.

Its boundary is identified with the boundary of a 2-simplex. We can glue g copies
of K together along an edge to obtain a simplicial complex Kg which triangulates the
surface Fg from Example 5.4.1, with boundary given by a sub-simplicial complex L. We
can then glue on the cone CL on L to obtain a triangulation of the surface Σg from
Example 5.4.1.

Homology of K. The subcomplex W ⊂ K given by

 7.7 Homology of surfaces, and their classification 71

is such that the inclusion |W | ,→ |K| is a homotopy equivalence, and hence the inclusion
induces an isomorphism on homology. By an application of Mayer–Vietoris, decomposing
W into two circles glued together at 1, we obtain H0 (K) = Z{[1]} and H1 (K) = Z{a, b}
where
a = [1, 2] + [2, 3] + [3, 1] b = [1, 4] + [4, 7] + [7, 1],

and H2 (K) = 0. Furthermore, the boundary cycle r = [6, 5]+[9, 6]+[5, 9] of K represents
the same homology class as a + b − a − b, so the zero homology class.

Homology of Kg . We can decompose Kg = Kg−1 ∪ K where Kg−1 ∩ K is a 1-simplex,

∆1 . The Mayer–Vietoris sequence is then

0 / H2 (Kg−1 ) / H2 (Kg )

∂∗
i∗ +j∗ k∗ −l∗ /
/ H1 (∆1 ) / H1 (K) ⊕ H1 (Kg−1 ) H1 (Kg )

∂∗
/ H0 (∆1 ) i∗ +j∗ / H0 (K) ⊕ H0 (Kg−1 ) k∗ −l∗ / H0 (Kg ) /0

and ∆1 is contractible so H1 (∆1 ) = 0 and H0 (∆1 ) = Z generated by any vertex. Thus

we have
0 /0 / H2 (Kg )

∂∗
i∗ +j∗ k∗ −l∗ /
/0 / Z2 ⊕ H1 (Kg−1 ) H1 (Kg )

0
/ Z i∗ +j∗ / Z ⊕ H0 (Kg−1 ) k∗ −l∗ / H0 (Kg ) /0

and so by induction

H0 (Kg ) = Z generated by any vertex

H1 (Kg ) = Z2g generated by a1 , b1 , . . . , ag , bg
H2 (Kg ) = 0.

The boundary cycle r1 + · · · + rg represents the same homology class as a1 + b1 − a1 −

b1 + · · · + ag + bg − ag − bg , so is zero.
72 Chapter 7 Homology

Homology of Kg ∪ CL. The Mayer–Vietoris sequence for this decomposition is

0 / H2 (Kg ∪ CL)

∂∗
/ H1 (L) i∗ +j∗ / H1 (Kg ) ⊕ H1 (CL) k∗ −l∗ / H1 (Kg ∪ CL)

∂∗
/ H0 (L) i∗ +j∗ / H0 (Kg ) ⊕ H0 (CL) k∗ −l∗ / H0 (Kg ∪ CL) /0

The simplicial complex L is a triangulation of S 1 , and CL is contractible, so we obtain

0 / H2 (Kg ∪ CL)

∂∗
/Z i∗ / k∗ −l∗ /
Z2g ⊕0 H1 (Kg ∪ CL)

0
/ Z (1,1) / Z ⊕ Z k∗ −l∗ / H0 (Kg ∪ CL) /0

and the map i∗ : H1 (L) → H1 (Kg ) sends the generator to the boundary cycle r1 +· · ·+rg ,
which is the zero homology class, so this map is zero. We deduce that

H0 (Kg ∪ CL) = Z generated by any vertex

H1 (Kg ∪ CL) = Z2g generated by a1 , b1 , . . . , ag , bg
H2 (Kg ∪ CL) = Z.

The group H2 (Kg ∪ CL) is generated by any 2-cycle x such that ∂∗ (x) ∈ H1 (L) ∼
= Z is
a generator.

Example 7.7.3 (Nonorientable surfaces). Let K be the simplicial complex shown in

Figure 7.2, which is a triangulation of the projective plane with an open disc removed,
Y.

Figure 7.2 A triangulation of Y , the projective plane with an open disc removed.
 7.7 Homology of surfaces, and their classification 73

Its boundary is identified with the boundary of a 2-simplex. We can glue n copies
of K together along an edge to obtain a simplicial complex Kn which triangulates the
surface En from Example 5.4.3, with boundary given by a sub-simplicial complex L. We
can then glue on the cone CL on L to obtain a triangulation of the surface Sn from
Example 5.4.3.

Homology of K. The subcomplex W ⊂ K given by

is such that the inclusion |W | ,→ |K| is a homotopy equivalence, and hence the inclusion
induces an isomorphism on homology. We obtain H0 (K) = Z{[1]} and H1 (K) = Z{u}
where
u = a + b = [1, 2] + [2, 3] + [3, 1],
and H2 (K) = 0. Furthermore, the boundary cycle r = [4, 5]+[5, 6]+[6, 4] of K represents
the same homology class as 2u.

Homology of Kn . We can decompose Kn = Kn−1 ∪ K where Kn−1 ∩ K is a 1-simplex,

∆1 . The Mayer–Vietoris sequence is then

0 / 0 ⊕ H2 (Kn−1 ) / H2 (Kn )

∂∗
/ H1 (∆1 ) i∗ +j∗ / H1 (K) ⊕ H1 (Kn−1 ) k∗ −l∗ / H1 (Kn )

∂∗
/ H0 (∆1 ) i∗ +j∗ / H0 (K) ⊕ H0 (Kn−1 ) k∗ −l∗ / H0 (Kn ) /0

and ∆1 is contractible so H1 (∆1 ) = 0 and H0 (∆1 ) = Z generated by any vertex. Thus

by induction

H0 (Kn ) = Z generated by any vertex

H1 (Kn ) = Zn generated by u1 , . . . , un
H2 (Kn ) = 0.

The boundary cycle r1 +· · ·+rn represents the same homology class as 2u1 +2u2 +· · ·+2un .
74 Chapter 7 Homology

Homology of Kn ∪ CL. The Mayer–Vietoris sequence for this decomposition is

0 ⊕ H2 (CL) / H2 (Kn ∪ CL)

∂∗
/ H1 (L) i∗ +j∗ / H1 (Kn ) ⊕ H1 (CL) k∗ −l∗ / H1 (Kn ∪ CL)

∂∗
/ H0 (L) i∗ +j∗ / H0 (Kn ) ⊕ H0 (CL) k∗ −l∗ / H0 (Kn ∪ CL) /0

The simplicial complex L is a triangulation of S 1 , and CL is contractible, so we obtain

0 / H2 (Kn ∪ CL)

∂∗
/Z i∗ / Zn ⊕ 0 k∗ −l∗ / H1 (Kn ∪ CL)

0
/ Z (1,1) / Z ⊕ Z k∗ −l∗ / H0 (Kn ∪ CL) /0

and the map i∗ : H1 (L) → H1 (Kg ) sends the generator to the boundary cycle r1 +· · ·+rn ,
which is the 2u1 + 2u2 + · · · + 2un . Thus this map is 1 7→ (2, 2, . . . , 2), which is injective.
We deduce that
H0 (Kn ∪ CL) = Z generated by any vertex
H1 (Kn ∪ CL) = Zn /Z{(2, 2, . . . , 2)} generated by u1 , . . . , un subject to 2u1 + · · · + 2un = 0
H2 (Kn ∪ CL) = 0.
Note that the change of basis
e1 = u1 + u2 + · · · + un ei = ui 2 ≤ i ≤ n
gives an isomorphism of abelian groups Zn /Z{(2, 2, . . . , 2)} ∼
= Z/2 ⊕ Zn−1 .
To summarise these two examples, as well as Lemma 7.6.1 in dimension 2, the (tri-
angulable) surfaces S 2 , Σg for g ≥ 1, and Sn for n ≥ 1, all have non-isomorphic first
homology groups:
H1 (S 2 ) = 0 H1 (Σg ) ∼
= Z2g H1 (Sn ) ∼
= Z/2 ⊕ Zn−1 .
Thus no two of these surfaces are homeomorphic, or even homotopy equivalent. This
proves one half of the following theorem. The other half is much more elementary, but
quite long: it can be found for example in Maunder’s Algebraic Topology.
Theorem 7.7.4 (Classification of triangulable surfaces). Any triangulable surface is
homeomorphic to one of
S 2 , Σ1 , Σ2 , Σ3 , . . .
S1 , S2 , S3 , . . .
and no two of these are homeomorphic.
 7.8 Rational homology, Euler and Lefschetz numbers 75

We have been using the terms “orientable surface” for one obtained by gluing together
copies of the torus, and “nonorientable surface” for one obtained by gluing together copies
of the projective plane. These are in fact intrinsic properties of surfaces, and not just a
feature of how we choose to construct them: this may be seen homologically by

H2 (S 2 ) = H2 (Σg ) = Z but H2 (Sn ) = 0.

This difference accounts for our partition in Theorem 7.7.4 of surfaces into two series,
the orientable ones (S 2 , Σ1 , Σ2 , Σ3 , . . .) having second homology Z, and the nonorientable
ones (S1 , S2 , S3 , . . .) having second homology 0.

7.8 Rational homology, Euler and Lefschetz numbers

Definition 7.8.1. For a simplicial complex K, define the rational n-chains, Cn (K; Q),
by letting On (K; Q) be the Q-vector space with basis the oriented simplices, and Tn (K; Q)
be the sub vector space spanned by the usual collection of “trivial” simplices. We define
dn : Cn (K; Q) → Cn−1 (K; Q) by the usual formula, and write

Ker(dn : Cn (K; Q) → Cn−1 (K; Q))

Hn (K; Q) :=
Im(dn+1 : Cn+1 (K; Q) → Cn (K; Q))

for the nth rational homology of K. It is a Q-vector space.

Everything we did for ordinary homology remains true for rational homology (induced
maps, subdivision isomorphism, homotopy invariance, Mayer–Vietoris sequence). The
following lemma gives some idea of what rational homology measures. Its proof is Q8 on
Example Sheet 4.

Lemma 7.8.2. If Hn (K) ∼

= Zr ⊕ F for F a finite abelian group, then Hn (K; Q) ∼
= Qr .

We thus know the homology of several spaces. For spheres, with n > 0,
(
Q i = 0, n
Hi (S n ; Q) ∼
=
0 else.

For orientable surfaces 

Q
 i = 0, 2
∼
Hi (Σg ; Q) = Q2g i=1

0 else.


For nonorientable surfaces


Q
 i=0
∼
Hi (Sn ; Q) = Qn−1 i=1

0 else.

76 Chapter 7 Homology

Definition 7.8.3. Let X be a polyhedron and f : X → X be a continuous map. The

Lefschetz number of f is
X
L(f ) := (−1)i Tr (f∗ : Hi (X; Q) → Hi (X; Q)) .
i≥0

The Euler characteristic of X is

X
χ(X) = L(IdX ) = (−1)i dimQ Hi (X; Q).
i≥0

Example 7.8.4. Computing the Euler characteristic for the examples we have gives
(
n 2 n even
χ(S ) = χ(Σg ) = 2 − 2g χ(Sn ) = 2 − n.
0 n odd

Example 7.8.5. Recall that the antipodal map a : S n → S n has a∗ : Hn (S n ) → Hn (S n )

given by multiplication by (−1)n+1 . The same is true in rational homology, and so

L(a) = 1 + (−1)n · (−1)n+1 = 0.

Lemma 7.8.6. Let V be a finite-dimensional vector space and W ≤ V be a subspace,

Let A : V → V be a linear map such that A(W ) ⊂ W . Let B = A|W : W → W be the
restriction and C : V /W → V /W be the induced map. Then

Tr(A) = Tr(B) + Tr(C).

Proof. Let e1 , e2 , . . . , er be a basis for W , and extend it to a basis e1 , e2 , . . . , en of V .

The matrix of A with respect to this basis is of the form
 
X Y
0 Z

where X is the matrix for B with respect to the basis e1 , e2 , . . .
, er and Z is the matrix
for C with respect to the basis [er+1 ], . . . , [en ]. Now Tr X0 YZ = Tr(X) + Tr(Z), as
required.

Corollary 7.8.7. For a chain map f• : C• (K; Q) → C• (K; Q), we have

X X
(−1)i Tr(f∗ : Hi (K; Q) → Hi (K; Q)) = (−1)i Tr(fi : Ci (K; Q) → Ci (K; Q)).
i≥0 i≥0

Proof. There are exact sequence

0 −→ Bi (K; Q) −→ Zi (L; Q) −→ Hi (K; Q) −→ 0

and
0 −→ Zi (K; Q) −→ Ci (L; Q) −→ Bi−1 (K; Q) −→ 0.
 7.8 Rational homology, Euler and Lefschetz numbers 77

Let fiH , fiB , fiZ , and fiC be the maps induced by f• on homology, boundaries, cycles,
and chains respectively. Then
X
L(|f |) = (−1)i Tr(fiH )
i≥0
X
(−1)i Tr(fiZ ) − Tr(fiB )


=
i≥0
X
(−1)i Tr(fiC ) − Tr(fi−1
B
) − Tr(fiB )


=
i≥0
X
= (−1)i Tr(fiC ).
i≥0

Corollary 7.8.8. We have i≥0 (−1)i · #i-simplices of K = χ(|K|), and in particular

P
this number does not depend on the triangulation of the polyhedron |K|.
Theorem 7.8.9 (Lefschetz). Let f : X → X be a continuous map from a polyhedron to
itself. If L(f ) 6= 0 then f has a fixed point.
Proof. We will show the contrapositive to the theorem: if f has no fixed points then
L(f ) = 0. Supposing that f has no fixed points, let
δ := inf{|x − f (x)| : x ∈ X}.
As X is compact, δ > 0. Let K be a triangulation of X with µ(K) < δ/2, and choose g :
K (r) → K a simplicial approximation to f . For v ∈ K (r) we have f (v) ∈ f (StK (r) (v)) ⊂
StK (g(v)), and so |f (v) − g(v)| < δ/2. But |f (v) − v| ≥ δ, and so |g(v) − v| ≥ δ/2. Thus
if v ∈ σ ∈ K then g(v) 6∈ σ.
The map f∗ is defined to be g∗ ◦ (νK,r )−1
∗ . Using the subdivision map s• from Propo-
(r)
sition 7.5.8, iterated, the map f∗ is induced by the chain map g• ◦ s• : C• (K; Q) →
C• (K; Q). Thus by Lemma 7.8.7 we have
(r)
X
L(f ) = (−1)i Tr(gi ◦ si : Ci (K; Q) → Ci (K; Q)).
i≥0

(r) (r)
If σ ∈ K is an i-simplex then si (σ) is a sum of simplices inside σ, and so gi (si (σ)) is
(r)
a sum of simplices disjoint from σ. Thus the matrix for gi ◦ si with respect to the basis
of simplices has zeroes on the diagonal, so has trace zero.
Example 7.8.10. If X is a contractible polyhedron then L(f ) = 1 for any f : X → X,
so any f has a fixed point. This recovers Brouwer’s fixed point theorem, but is far more
general.
Example 7.8.11. let h : |K| → G be a triangulation of a topological group G which is
connected and nontrivial. The for g 6= 1 ∈ G multiplication by g is a homeomorphism
with no fixed points, so 0 = L(− · g). On the other hand, choosing a path from g to
1 ∈ G gives a homotopy from − · g to the identity, so
0 = L(− · g) = L(− · 1) = L(IdG ) = χ(G).
78 Chapter 7 Homology

Among the surfaces, only Σ1 and S2 have Euler characteristic zero, so no other
surface admits a topological group structure. The torus Σ1 = S 1 × S 1 does admit a
group structure, as R2 /Z2 . The surface S2 is the Klein bottle, and does not admit a
group structure. One way to see this is to use Q13 on Example Sheet 1, where the
fundamental group of the Klein bottle is computed and shown to be non-abelian: it is
an easy exercise to show that the fundamental group of a topological group (based at
the identity element) is always abelian.

Example 7.8.12. Let f : S3 → S3 be such that f ◦ f = IdS3 . We have


Q
 i=0
Hi (S3 ; Q) ∼
= Q2 i=1

0 else,


and f∗ : H0 (S3 ; Q) → H0 (S3 ; Q) is the identity map, as this group is generated by any
vertex of a triangulation of S3 . The linear map

f∗ : Q2 ∼
= H1 (S3 ; Q) −→ H1 (S3 ; Q) ∼
= Q2

satisfies (f∗ )2 = Id, so by elementary linear algebra its eigenvalues are all ±1. Thus its
trace is one of −2, 0, or 2, and so

L(f ) = 1 − Tr(f∗ : H1 (S3 ; Q) −→ H1 (S3 ; Q)) ∈ {−1, 1, 3}.

Hence f must have a fixed point.