LAURENTIU MAXIM

UNIVERSITY OF WISCONSIN-MADISON

A L G E B R A I C TO P O L O G Y:
A COMPREHENSIVE
INTRODUCTION
 i

Contents

Preface ix

1 Foreword about invariants of spaces 1

2 Fundamental group 3
2.1 Definition 3
2.2 Basepoint (in)dependence 7
2.3 Functoriality 8
2.4 Homotopy invariance of fundamental group 9
2.5 Contractible spaces. Deformation Retracts 10
2.6 Fundamental group of a circle 12
2.7 Some Immediate Applications 15
Brower’s fixed point theorem 16
Fundamental Theorem of Algebra 17
Exercises 18
2.8 Seifert–Van Kampen’s Theorem 19
Free Groups 19
Free Products 21
Seifert-Van Kampen Theorem 25
Exercises 29

3 Classification of compact surfaces 31

 ii

3.1 Surfaces: definitions, examples 31

3.2 Fundamental group of a labeling scheme 36
3.3 Classification of surfaces 40
Exercises 45

4 Covering spaces 47
4.1 Definition. Properties 47
4.2 Covering transformations 52
4.3 Universal Covering Spaces 54
4.4 Group actions and covering maps 56
Exercises 60

5 Homology 63
5.1 Singular Homology 63
5.2 Homotopy Invariance 66
5.3 Homology of a pair 70
Exercises 79
5.4 π1 vs. H1 81
5.5 Cellular Homology 83
Degrees 84
Exercises 87
How to Compute Degrees? 88
CW Complexes 90
Exercises 92
Cellular Homology 92
Exercises 99
5.6 Euler Characteristic 101
Exercises 102
5.7 Lefschetz Fixed Point Theorem 103
Exercises 105
 iii

5.8 Homology with arbitrary coefficients 106

Tensor Products 106
Homology with Arbitrary Coefficients 108
Exercises 111
5.9 The Tor functor and the Universal Coefficient Theorem 111
Exercises 116

6 Basics of Cohomology 117

6.1 Cohomology of a chain complex: definition 117
6.2 Relation between cohomology and homology 118
Ext groups 118
Universal Coefficient Theorem 119
6.3 Cohomology of spaces 121
Definition and immediate consequences 121
Reduced cohomology groups 123
Relative cohomology groups 123
Induced homomorphisms 124
Homotopy invariance 125
Excision 126
Mayer-Vietoris sequence 126
Cellular cohomology 127
Exercises 131

7 Cup Product in Cohomology 133

7.1 Cup Products: definition, properties, examples 133
7.2 Application: Borsuk-Ulam Theorem 144
Exercises 146
7.3 Künneth Formula 147
Cross product 147
Künneth theorem in cohomology. Examples 149
Künneth exact sequence and applications 153
 iv

Künneth Formula for homology. 154

Universal Coefficient Theorem for homology 155
Künneth formula for cohomology 155
Exercises 156

8 Poincaré Duality 157

8.1 Introduction 157
8.2 Manifolds. Orientation of manifolds 158
Exercises 162
8.3 Cohomolgy with Compact Support 163
8.4 Cap Product and the Poincaré Duality Map 166
8.5 The Poincaré Duality Theorem 168
Exercises 172
8.6 Immediate applications of Poincaré Duality 174
8.7 Addendum to orientations of manifolds 175
8.8 Cup product and Poincaré Duality 180
Exercises 184
8.9 Manifolds with boundary: Poincaré duality and applications 184
Exercises 186
Signature 187
Connected Sums 188

9 Basics of Homotopy Theory 191

9.1 Homotopy Groups 191
9.2 Relative Homotopy Groups 197
9.3 Homotopy Extension Property 200
9.4 Cellular Approximation 201
9.5 Excision for homotopy groups. The Suspension Theorem 203
9.6 Homotopy Groups of Spheres 203
9.7 Whitehead’s Theorem 207
9.8 CW approximation 210
 v

9.9 Eilenberg-MacLane spaces 215

9.10 Hurewicz Theorem 218
9.11 Fibrations. Fiber bundles 219
9.12 More examples of fiber bundles 224
9.13 Turning maps into fibration 228
9.14 Exercises 229

10 Spectral Sequences. Applications 231

10.1 Homological spectral sequences. Definitions 231
10.2 Immediate Applications: Hurewicz Theorem Redux 234
10.3 Leray-Serre Spectral Sequence 236
10.4 Hurewicz Theorem, continued 240
10.5 Gysin and Wang sequences 242
10.6 Suspension Theorem for Homotopy Groups of Spheres 244
10.7 Cohomology Spectral Sequences 247
10.8 Elementary computations 249
10.9 Computation of πn+1 (Sn ) 253
10.10Whitehead tower approximation and π5 (S3 ) 256
Whitehead tower 256
Calculation of π4 (S3 ) and π5 (S3 ) 257
10.11Serre’s theorem on finiteness of homotopy groups of spheres 260
10.12Computing cohomology rings via spectral sequences 264
10.13Exercises 266

11 Fiber bundles. Classifying spaces. Applications 269

11.1 Fiber bundles 269
11.2 Principal Bundles 276
11.3 Classification of principal G-bundles 282
11.4 Exercises 287
 vi

12 Vector Bundles. Characteristic classes.

Cobordism. Applications. 289
12.1 Chern classes of complex vector bundles 289
12.2 Stiefel-Whitney classes of real vector bundles 292
12.3 Stiefel-Whitney classes of manifolds and applications 293
The embedding problem 293
Boundary Problem 297
12.4 Pontrjagin classes 299
Applications to the embedding problem 302
12.5 Oriented cobordism and Pontrjagin numbers 303
12.6 Signature as an oriented cobordism invariant 306
12.7 Exotic 7-spheres 307
12.8 Exercises 308

Bibliography 311
 vii

List of Figures

Figure 2.1: The map p : R → S1 . . . . . . . . . . . . . . . . 12

Figure 2.2: The map r : D2 → S1 . . . . . . . . . . . . . . . . 16
Figure 2.3: Here, m = 2, f 1 is the path in A1 := Aα from
x0 to f (s1 ) and f 2 is the path in A2 := A β from
f (s1 ) to x0 . . . . . . . . . . . . . . . . . . . . . . 27

Figure 3.1: Torus T 2 . . . . . . . . . . . . . . . . . . . . . . . 32

Figure 3.2: Klein bottle . . . . . . . . . . . . . . . . . . . . . 34
Figure 3.3: T 2 #T 2 . . . . . . . . . . . . . . . . . . . . . . . . 36
Figure 3.4: How to turn the Klein bottle into P2 . . . . . . 39
Figure 3.5: Removing a disc from RP2 yields a Möbius band. 40
Figure 3.6: Performing connected sum with a Klein bottle. 40
Figure 3.7: T 2 #RP2 . . . . . . . . . . . . . . . . . . . . . . . . 40
Figure 3.8: Every point has a neighborhood homeomor-
phic to a disc. . . . . . . . . . . . . . . . . . . . . 41
Figure 3.9: Step 1: Removing adjacent edges of the first kind. 42
Figure 3.10: Step 2: identifying all vertices. . . . . . . . . . . 42
Figure 3.11: Step 3: Making two Type II edges adjacent. . . 42
Figure 3.12: Step 4. . . . . . . . . . . . . . . . . . . . . . . . . 43
Figure 3.13: Step 5: Two pairs of the first kind being made
consecutive. . . . . . . . . . . . . . . . . . . . . . 43
Figure 3.14: Making all sides have the same orientation: cut
along d, glue along c. . . . . . . . . . . . . . . . 44
Figure 3.15: Completing Step 6. . . . . . . . . . . . . . . . . 44

Figure 4.1: Infinite earring . . . . . . . . . . . . . . . . . . . 55

Figure 5.1: Suspension of the circle S1 is homeomorphic to

S2 . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Figure 5.2: The map ∆ . . . . . . . . . . . . . . . . . . . . . 100

Figure 8.1: pinched torus . . . . . . . . . . . . . . . . . . . . 183

Figure 9.1: f + g . . . . . . . . . . . . . . . . . . . . . . . . . 192

viii

Figure 9.2: f +g ≃ g+ f . . . . . . . . . . . . . . . . . . . . 192

Figure 9.3: f + g, revisited . . . . . . . . . . . . . . . . . . . 193
Figure 9.4: βγ . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Figure 9.5: relative β γ . . . . . . . . . . . . . . . . . . . . . . 199
Figure 9.6: universal cover of S1 ∨ Sn . . . . . . . . . . . . . 205
Figure 9.7: The mapping cylinder M f . . . . . . . . . . . . 209

Figure 10.1: r-th page Er . . . . . . . . . . . . . . . . . . . . . 232

Figure 10.2: n-th diagonal of E∞ . . . . . . . . . . . . . . . . 232
Figure 10.3: p-axis and q-axis of E2 . . . . . . . . . . . . . . 234
 ix

Preface

Algebraic topology studies topological spaces via algebraic invariants

like fundamental group, homotopy groups, (co)homology groups, etc.
Topological (or homotopy) invariants are those properties of topological
spaces which remain unchanged under homeomorphisms (respectively,
homotopy equivalences). The ultimate goal of the theory is to classify
(at least special classes of) topological spaces up to homeomorphism or
homotopy equivalence. There are several success stories in this direction
(e.g., the classification of closed surfaces), but this is difficult to achieve
in general. Alternatively, one aims to develop enough invariants to be
able to distinguish between topological spaces.
While assuming minimal prerequisites (e.g., basic notions of alge-
bra and point set topology), these notes provide a comprehensive
introduction to algebraic topology. Topics covered here include: fun-
damental group, classification of compact surfaces, covering spaces,
homology, cohomology, Poincaré duality, higher homotopy groups,
spectral sequences, fiber bundles and classifying spaces, vector bun-
dles, characteristic classes. This material is intended as a two-semester
graduate course, but it may also serve as a quick reference for anyone
interested in geometry, topology and algebraic geometry.
The primary goal of these notes is to provide readers with a taste of
this beautiful subject by presenting concrete examples and applications
that motivate the abstract theory. Towards this goal, and in order to
keep the size of the material within a reasonable level, several important
results are stated without proof, while some of their main applications
are emphasized instead. For more complete details and further reading,
one one may also consult standard textbooks and references, such as
[Bott and Tu, 1982], [Bredon, 1993], [Hatcher, 2002], [Davis and Kirk,
2001], [Massey, 1991], [Milnor and Stasheff, 1974], [Munkres, 2000],
[Munkres, 1984], [Spanier, 1966], etc.
Acknowledgements. These notes grew out of lectures given by the
author at the City University of New York, University of Wisconsin-
Madison, and the University of Science and Technology of China (USTC)
in Hefei, China. I would like to thank my students, colleagues and
x

collaborators for valuable feedback.

 foreword about invariants of spaces 1

1
Foreword about invariants of spaces

As a warm-up example, let us consider the following simple example

of an invariant of a topological space with a finite number of path
components.

Example 1.0.1. If X is a topological space, let n( X ) be the number of

path components of X (by assumption, this is a positive integer). It is
easy to see that if f : X → Y is a continuous map, then n( f ( X )) ≤ n( X ).
Thus, if f is a homeomorphism, then n( X ) = n(Y ), so n(−) is a
topological invariant.

The invariant n( X ) can be used for proving the following one-

dimensional version of Brower’s fixed point theorem:

Theorem 1.0.2. Any continuous map f : [0, 1] → [0, 1] has a fixed point, i.e.,
there exists x ∈ [0, 1] so that f ( x ) = x.

Proof. Assume, by contradiction, that f ( x ) ̸= x, for any x ∈ [0, 1].

Define
f (x) − x
r(x) = .
| f (x) − x|
Then r is clearly a continuous map. Moreover, the image of r is the set
{±1}. Since f (0) ̸= 0 we must have f (0) > 0, so r (0) = 1. Similarly,
f (1) < 1, so r (1) = −1. Hence we have a surjective continuous function

r : [0, 1] → {−1, 1}.

By using the invariant n(−) on the map r, we get that

n({−1, 1}) ≤ n([0, 1]),

or 2 ≤ 1, which is clearly a contradiction.

In the following chapters, we will associate various algebraic invari-

ants to topological spaces, e.g., the fundamental group, (co)homology
groups, etc.
 fundamental group 3

2
Fundamental group

The first non-trivial algebraic invariant we associate to a topological

space is the fundamental group. This invariant is powerful enough
to provide us with a complete topological classification of compact
surfaces (see Section 3.3).

2.1 Definition

Let X be a connected topological space. For x, y ∈ X, consider the set

P ( X, x, y) = {γ : [0, 1] → X | γ(0) = x, γ(1) = y}

of all continuous paths in X from x to y. The loop space of X at x is then

defined by
Ω( X, x ) = P ( X, x, x ).
On P ( X, x, y), we define the following (equivalence) relation:

Definition 2.1.1. Two paths γ, δ ∈ P ( X, x, y) are called homotopic, denoted

as γ ∼ δ, if there exists a continuous map (called a homotopy between γ and
δ)
F : [0, 1] × [0, 1] → X.
so that

(t, 0) 7→ γ(t)
(t, 1) 7→ δ(t)
(0, s) 7→ x
(1, s) 7→ y

To emphasize the homotopy F between γ and δ, we usually use the

F
symbol γ ∼ δ. If we set F (t, s) = γs (t), then a homotopy F as above
satisfies the property that γ0 = γ and γ1 = δ, as well as γs (0) = x,
γs (1) = y. We can represent a homotopy schematically on the unit
square as follows:
4 algebraic topology

s
δ

γs

x y
F

t
γ

Lemma 2.1.2. The homotopy relation ∼ is an equivalence relation on the set

P ( X, x, y).

Proof. The homotopy relation is:

• reflexive, i.e., γ ∼ γ via F (t, s) = γ(t) for any s.

F F̄
• symmetric: if γ ∼ δ, then δ ∼ γ via F̄ (t, s) = F (t, 1 − s).

F G H
• transitive: let γ ∼ δ and δ ∼ φ, then γ ∼ φ via

1
 F (t, 2s) 0≤s≤ 2
H (t, s) =
 G (t, 2s − 1) 1
2 ≤s≤1

Note that

H (t, 0) = F (t, 0) = γ(t)

1
H (t, ) = F (t, 1) = G (t, 0) = δ(t)
2
H (t, 1) = G (t, 1) = φ(t)

In order to show that H is continuous, we use the following standard

fact from point set topology: if X = A ∪ B, with both A and B closed
subsets (or both open), and if f : X → Y is a map so that f | A and f | B
are continuous, then f is continuous.
 fundamental group 5

Definition 2.1.3. The fundamental group of X at the basepoint x ∈ X is

defined as the set of equivalence classes of loops at x under the homotopy
relation, i.e.,
π1 ( X, x ) := Ω( X, x )/ ∼ .

In order to justify the word “group” in the above definition, we

introduce the following concatenation operation on paths in X:

Definition 2.1.4. For x, y, z ∈ X, define

∗
P ( X, x, y) × P ( X, y, z) −→ P ( X, x, z)

γ(2t) 0 ≤ t ≤ 21
(γ ∗ δ)(t) =
δ(2t − 1) 1 ≤ t ≤ 1
2

0 γ 1 δ 1
2

Alternatively, one can define the path γ ∗s δ by


γ( t ) 0≤t≤s
s
(γ ∗s δ)(t) =
δ( t − s
) s≤t≤1
1− s

0 γ s δ 1

Lemma 2.1.5. The concatenation of paths is consistent with the homotopy

F G H
relation, i.e., if γ ∼ γ′ and δ ∼ δ′ , then γ ∗ δ ∼ γ′ ∗ δ′ .

Proof. The claimed homotopy H is defined by:


 F (2t, s) 0 ≤ t ≤ 21
H (t, s) =
 G (2t − 1, s) 1 ≤ t ≤ 1
2

1
γ′ 2 δ′

F G

γ 1 δ
2
6 algebraic topology

Corollary 2.1.6. The operation of concatenation of paths induces a binary

law on the set π1 ( X, x ) = Ω( X, x )/ ∼, by:

[γ] · [δ] := [γ ∗ δ]

Theorem 2.1.7. (π1 ( X, x ), ·) is a group.

Proof. In order to show the associativity of the binary law, we start by

noting that
γ ∗s δ ∼ γ ∗s′ δ,
for any s, s′ ∈ (0, 1). Indeed, this can be easily seen from the following
diagram:

0 γ s′ δ 1

γ δ
0 s 1

Then, for γ, δ, η ∈ Ω( X, x ), we have:

(γ ∗ δ) ∗ η ∼ (γ ∗ δ) ∗ 2 η ∼ γ ∗ 1 (δ ∗ η ) ∼ γ ∗ (δ ∗ η )
3 3

In order to find the identity element, consider the constant loop

ex (t) = x, for all t ∈ [0, 1]. We claim that if γ ∈ P ( X, x, y), then
F G
e x ∗ γ ∼ γ ∼ γ ∗ ey .

Indeed, we have,

2t s +1
γ(
s +1 ) 0≤t≤ 2
G (t, s) =
s +1
 y = γ (1)
2 ≤t≤1

x y
γ ey

γ ey

F
And similarly for ex ∗ γ ∼ γ. Therefore, ex is the identity element in
(π1 ( X, x ), ·).
 fundamental group 7

Finally, let
γ̄(t) = γ(1 − t)

and set,
[γ]−1 := [γ̄]
We claim that, γ ∗ γ̄ ∼ ex ∼ γ̄ ∗ γ, i.e., [γ̄] is the inverse of [γ] in
(π1 ( X, x ), ·). Indeed, γ ∗ γ̄ ∼ ex via:
ex

γ (1 − s )
s
γs γ̄s

γ x γ̄

Here, the homotopy H (t, s) = hs (t) between γ ∗ γ̄ and ex is given by

hs = γs ∗ γ̄s , where γs (t) is the path that equals γ on [0, 1 − s] and that
is stationary at γ(1 − s) on the interval [1 − s, 1], and γ̄s is the inverse
path of γs .
Similarly considerations apply for γ̄ ∗ γ ∼ ex .

Example 2.1.8. Here are some elementary examples, as well as some

which will be discussed later on:

a) If X = { x } is just a point space, then the only path (loop) in X is

the constant one, so π1 ( X, x ) = {[ex ]} is the trivial group.

b) If X is a convex subset of Rn , and x ∈ X, then π1 ( X, x ) = {[ex ]}.

Indeed, for any γ ∈ π1 ( X, x ), the map

H (t, s) = sex + (1 − s)γ(t)

is continuous, H (t, 0) = γ(t), H (t, 1) = ex , so H is a homotopy from

γ to ex .

c) For n ≥ 2, π1 (Sn , x ) = {[ex ]}. This will be explained later on.

d) As we will see later, one has: π1 (S1 , 1) ∼

= Z = ⟨γ(t)⟩, where
γ(t) = (cos 2πt, sin 2πt).

2.2 Basepoint (in)dependence

We can now ask the following:

8 algebraic topology

Question 2.2.1. How does π1 ( X, x ) change if we change the basepoint x, i.e.,

how are π1 ( X, x ) and π1 ( X, y) related, for y ̸= x?

In order to give an answer, let us assume that X is path-connected,

and let x ̸= y be two distinct points in X. Choose a path δ : I = [0, 1] →
X in X from x to y, δ(0) = x, δ(1) = y. Note that if γ ∈ Ω( X, x ), then
δ̄ ∗ γ ∗ δ ∈ Ω( X, y). It is easy to see that the assignment

γ 7→ δ̄ ∗ γ ∗ δ

is compatible with the homotopy relation (if γs is a homotopy starting

at γ, then δ̄ ∗ γs ∗ δ is a homotopy starting at δ̄ ∗ γ ∗ δ), hence it descends
to a map
δ# : π1 ( X, x ) → π1 ( X, y).

Proposition 2.2.2. δ# is an isomorphism.

Proof. It is easy to check that δ̄# : π1 ( X, y) → π1 ( X, x ), [η ] 7→ [δ ∗ η ∗ δ̄]

is the inverse of δ# . Moreover, δ# is a group homomorphism, since for
γ, η ∈ π1 ( X, x ) we have:

δ# ([γ] · [η ]) = δ# ([γ ∗ η ]) = [δ̄ ∗ (γ ∗ η ) ∗ δ] = [(δ̄ ∗ γ ∗ δ) ∗ (δ̄ ∗ η ∗ δ)]

= [(δ̄ ∗ γ ∗ δ)] · [(δ̄ ∗ η ∗ δ)]
= δ# ([γ]) · δ# ([η ])

2.3 Functoriality

The next question to ask is:

Question 2.3.1. How is the fundamental group affected by continuous maps

between topological spaces?

Let f : X → Y be a continuous map, with f ( x ) = y. Then the

γ f
composition I = [0, 1] → X → Y induces a map:

f ∗ : π1 ( X, x ) → π1 (Y, y)
[γ] 7→ [ f ◦ γ]

It is easy to see that f ∗ is well-defined: if γs is a homotopy for γ, then

f ◦ γs is a homotopy for f ◦ γ. Moreover, f ∗ is a homomorphism, since

 f (γ(2t)) 0 ≤ t ≤ 21
( f ◦ γ) ∗ ( f ◦ η ) = f ◦ (γ ∗ η ) : t 7→
 f (δ(2t − 1)) 1 ≤ t ≤ 1.
2

Using the above definition, one gets immediately the following:

Proposition 2.3.2. The following properties hold:

 fundamental group 9

f g
1. If ( X, x ) → (Y, y) → ( Z, z), then ( g ◦ f )∗ = g∗ ◦ f ∗

2. (id(X,x) )∗ = idπ1 (X,x)

As a consequence, we can now show the following:

Theorem 2.3.3. π1 ( X, x ) is a topological invariant, i.e., if

f : ( X, x ) → (Y, y)

is a homeomorphism, then

f ∗ : π1 ( X, x ) → π1 (Y, y)

is an isomorphism.

Proof. Let g = f −1 . Since f ◦ g = id(Y,y) , g ◦ f = id(X,x) , it follows by

the above two properties that ( f ∗ )−1 = g∗ , ( g∗ )−1 = f ∗ .

2.4 Homotopy invariance of fundamental group

In this section, we show that the fundamental group is a homotopy

invariant.

Definition 2.4.1. Let f , g : ( X, A) → (Y, B) be continuous maps of pairs,

so A ⊆ X, B ⊆ Y, with f ( A) ⊆ B, g( A) ⊆ B. We say that f and g
are “homotopic relative to A” (and write f ∼ A g) if there is a continuous
map F : X × [0, 1] → Y (called a homotopy) such that F ( A × [0, 1]) ⊆ B,
F ( x, 0) = f ( x ), and F ( x, 1) = g( x ) for all x ∈ X. If A = ∅, we say that f
is homotopic to g and write f ∼ g.

Lemma 2.4.2. If f , g : ( X, x0 ) → (Y, y0 ) are homotopic relative to x0 , then

f ∗ = g∗ : π1 ( X, x0 ) → (Y, y0 ).

Proof. If f ∼ x0 g via F, then for γ ∈ Ω( X, x0 ) it is easy to check that

H (t, s) := F (γ(t), s) is a homotopy between f ◦ γ and g ◦ γ. Hence
f ∗ ([γ]) = [ f ◦ γ] = [ g ◦ γ] = g∗ ([γ]) ∈ π1 (Y, y0 ).

Definition 2.4.3. We say that ( X, x0 ) and (Y, y0 ) are homotopy equivalent

(as pointed spaces) if there are continuous maps f : ( X, x0 ) → (Y, y0 ) and
g : (Y, y0 ) → ( X, x0 ) such that f ◦ g ∼y0 idY and g ◦ f ∼ x0 id X .

The following is an immediate consequence of the above lemma:

Theorem 2.4.4. If ( X, x0 ) and (Y, y0 ) are homotopy equivalent (as pointed

spaces), then π1 ( X, x0 ) ∼
= π1 (Y, y0 ).
Definition 2.4.5. We say that X and Y are homotopy equivalent (and write
X ≃ Y) if there are continuous maps f : X → Y and g : Y → X such that
f ◦ g ∼ idY and g ◦ f ∼ id X .
10 algebraic topology

It is easy to check that homotopy equivalence is an equivalence

relation. If X and Y are homotopy equivalent, we say that they have
the same homotopy type.
To prove that the fundamental group is preserved by a homotopy
equivalence, we need the following generalization of Lemma 2.4.2.

Lemma 2.4.6. Let h : X → Y and k : X → Y be continuous maps, x0 ∈

X, y0 = h( x0 ), y1 = k ( x0 ). If h ∼ k, there exists a path α in Y joining y0
to y1 , such that k ∗ = α# ◦ h∗ , i.e., the following diagram commutes:

h∗
π1 ( X, x0 ) π1 (Y, y0 )
α#
k∗
π1 (Y, y1 )

Proof. If H : X × [0, 1] → Y is a homotopy between h and k, we can take

α(t) = H ( x0 , t). Checking the commutativity of the above diagram is a
simple exercise.

Theorem 2.4.7. If f : X → Y is a homotopy equivalence, the induced homo-

morphism f ∗ : π1 ( X, x ) → π1 (Y, f ( x )) is an isomorphism, for any basepoint
x ∈ X.

Proof. Let g : Y → X be a homotopy inverse for f . Fix x0 ∈ X and

consider the maps

f g f
( X, x0 ) −
→ (Y, y0 ) −
→ ( X, x1 ) −
→ (Y, y1 )

where y0 = f ( x0 ), x1 = g(y0 ) and y1 = f ( x1 ). Since g ◦ f ∼ id X ,

by Lemma 2.4.6 and for a suitable choice of a path α between x0
and x1 in X we have that ( g ◦ f )∗ = α# is an isomorphism. Now,
( g ◦ f )∗ = g∗ ◦ ( f x0 )∗ is an isomorphism, which implies that g∗ is
surjective. Similarly, ( f ◦ g)∗ = ( f x1 )∗ ◦ g∗ is an isomorphism which
implies that g∗ is injective. (Here ( f x0 )∗ and ( f x1 )∗ are the maps induced
by f on the fundamental groups of the pointed spaces in the above
diagram.) Hence g∗ is an isomorphism. Using ( g ◦ f )∗ = α# we
conclude that,
( f x0 ) ∗ = ( g ∗ ) −1 ◦ α #

so that ( f x0 )∗ is also an isomorphism.

2.5 Contractible spaces. Deformation Retracts

Definition 2.5.1. A map f : X → Y is called nullhomotopic if f is homotopic

to a constant map. A space X is called contractible if the identity map
id X : X → X is nullhomotopic.
 fundamental group 11

Example 2.5.2. The euclidean space Rn , the n-dimensional disc D n ,

and the point space { x } are all contractible, while we will see later on
that S1 , S2 are not contractible.

It is a simple exercise to show the following:

Proposition 2.5.3. If X is contractible, then π1 ( X, x0 ) is trivial, for any

basepoint x0 ∈ X.

Definition 2.5.4. A space X is called simply-connected if π1 ( X, x0 ) is trivial

for any x0 ∈ X.

Remark 2.5.5. A contractible space is simply-connected. The converse

is not true, e.g., we will see that S2 is simply-connected, but it is not
contractible.

Proposition 2.5.6. A topological space X is simply-connected if, and only if,

there is a unique homotopy class of paths connecting any two points in X.

Proof. (=⇒) If x, y ∈ X and f , g : I = [0, 1] → X are paths from x to y

in X, then we have the following homotopies:

f ∼ f ∗ ey ∼ f ∗ ḡ ∗ g ∼ ex ∗ g ∼ g,

where we use the fact that ḡ ∗ g and f ∗ ḡ are loops in X at y and x,

resp., hence homotopic to the respective constant paths.
(⇐=) Take x = y. By hypothesis, any loop γ at x ∈ X is in the
homotopy class of ex .

Theorem 2.5.7. Let X be a topological space. The following are equivalent:

1. Every continuous map S1 → X is homotopic to the constant map.

2. Every continuous map S1 → X extends to a continuous map D2 → X,

where D2 is the 2-disc with boundary S1 .

3. π1 ( X, x0 ) is trivial, for all x0 ∈ X.

Proof. (3) =⇒ (1): Elements of π1 ( X, x0 ) can be regarded as homo-

topy classes of maps (S1 , s0 ) → ( X, x0 ), so the assertion follows.
(1) =⇒ (2): Let f : S1 → X be given. By (1), f is nullhomotopic, so
there is a map F : S1 × I → X with F (eiθ , 0) = f (eiθ ) and F (eiθ , 1) = c X ,
with c X a constant. Define fe : D2 → X by fe(reiθ ) = F (eiθ , 1 − r ). Then
fe is the required extension of f to D2 .
(2) =⇒ (3): Let f : S1 → X, f (1) = x0 , be a representative for
[ f ] ∈ π1 ( X, x0 ). By (2), f extends to some fe : D2 → X. If i : S1 ,→ D2
is the inclusion map, we have f = fe ◦ i, hence f ∗ = fe∗ ◦ i∗ . But D2 is
contractible, so fe∗ = 0 and f ∗ = 0. Hence [ f ] = [ f ◦ idS1 ] = f ∗ ([idS1 ]) =
0.
12 algebraic topology

Definition 2.5.8. A subset A ⊂ X is called a retract of X if there is a map

r : X → A, so that r| A = id A (i.e., if i : A ,→ X is the inclusion map, then
r ◦ i = id A ). A subset A ⊂ X is called a deformation retract if, in addition,
i ◦ r ∼ id X .

Remark 2.5.9. If A ⊂ X is a deformation retract of X, then A is

homotopy equivalent to X.

Lemma 2.5.10. The n-sphere Sn is a deformation retract of Rn+1 \{0}.

Proof. Let r : X = Rn+1 \ {0} → Sn be defined as r ( x ) = x/|| x ||. By

definition, we have r|Sn = idSn . Also, i ◦ r ∼ id X via H : X × [0, 1] → X
defined as
x
H ( x, t) =
(1 − t) + t|| x ||
x
Indeed, H is continuous, H ( x, 0) = x = id X ( x ) and H ( x, 1) = || x ||
=
i ◦ r ( x ).

2.6 Fundamental group of a circle

In this section, we sketch the proof of the following important result.

(More details will be given when we talk about covering spaces.)

Theorem 2.6.1. Let ϕ : Z → π1 (S1 ) be given by n 7→ [ωn ], where ωn : I =

[0, 1] → S1 ⊂ R2 is the loop ωn (t) = (cos(2πnt), sin(2πnt)). Then ϕ is a
group isomorphism.

Proof. Let p : R → S1 be defined by t 7→ (cos(2πt), sin(2πt)). Then

p−1 ((1, 0)) = Z.
Let us embed the real line into R3 as a helix via i : R ,→ R3 , t 7→
(cos(2πt), sin(2πt), t). Then p = pr12 ◦ i where pr12 ( x, y, z) = ( x, y).

Figure 2.1: The map p : R → S1 .

p
 fundamental group 13

Let ωe n : I → R be given by t 7→ nt. Note that ω e n (0) = 0 and

ωe n (1) = n. Also, ωn = p ◦ ω
e n , so ϕ(n) = [ p ◦ ωe n ]. In fact, ϕ(n) = [ p ◦ f˜]
for any path f˜ : I → R from 0 to n. Indeed ω e n and f˜ are homotopic in
R via the homotopy (1 − s)ω e n + s f˜. So p ◦ ω e n ∼ p ◦ f˜.
Define the translation τm : R → R by τm ( x ) = x + m, and notice that
ωe m is a path from 0 to m and τm (ω e n ) is a path from m to n + m; their
concatenation is thus a path in R from 0 to n + m. We have:

ϕ(m + n) = [ p ◦ ω
e n+m ] = [ p ◦ (ω
e m ∗ τm (ω
e n ))]
= [( p ◦ ω
e m ) ∗ ( p ◦ τm (ω
e n )]
= [ ωm ∗ ωn ] = [ ωm ] · [ ωn ]
= ϕ ( m ) · ϕ ( n ),
hence ϕ is a group homomorphism.
To prove that ϕ is bijective, we need two lemmas.

Lemma 2.6.2 (path lifting). For every f : I → S1 with f (0) = x0 ∈ S1 and

for any x̃0 ∈ p−1 ( x0 ), there is a unique f˜ : I → R such that p ◦ f˜ = f and
f˜(0) = x̃0 .

(R, x̃0 )
∃! f˜
p
f
( I, 0) ( S1 , x0 )

Lemma 2.6.3 (homotopy lifting). For every homotopy f s : I → S1 with

f s (0) = x0 ∈ S1 and for any x̃0 ∈ p−1 ( x0 ), there is a unique homotopy
f˜s : I → R such that p ◦ f˜s = f s and f˜s (0) = x̃0 .

Assuming the two lemmas for now, let x0 = (1, 0) and choose x̃0 = 0.
Let f : I → S1 be a loop at (1, 0) representing [ f ] ∈ π1 (S1 , x0 ). By the
path lifting lemma, there is a path f˜ : I → R such that p ◦ f˜ = f and
f˜(0) = 0 ∈ Z. Say f˜(1) = n ∈ Z, so f˜ is a path in R from 0 to n. Then
ϕ(n) = [ p ◦ f˜] = [ f ]. Since f was arbitrary, ϕ must be surjective.
Now suppose ϕ(m) = ϕ(n) for some m, n ∈ Z. So [ωm ] = [ωn ],
or ωm ∼ ωn . Let f s be a homotopy with f 0 = ωm and f 1 = ωn .
By homotopy lifting, there exists a homotopy f˜s : I → R such that
p ◦ f˜s = f s and f˜s (0) = 0. But f˜s (1) is independent of s, so f˜0 (1) =
f˜1 (1). Now f˜0 and ω e m are both lifts to R of f 0 = ωm which start at
0. By the uniqueness of path lifting, this gives f˜0 = ω̃m . In particular,
f˜0 (1) = ω
e m (1) = m. Similarly, f˜1 (1) = ωe n (1) = n. So m = n.

The path and homotopy lifting Lemmas 2.6.2 and 2.6.3 are conse-
quences of the following general lifting lemma which we prove here.

Lemma 2.6.4 (lifting). Let Y be a connected space. Given F : Y × I → S1

and Fe : Y × {0} → R which lifts F |Y ×{0} to R, there is a unique lift
Fe : Y × I → R of F which restricts to the given lift on Y × {0}.
14 algebraic topology

R
Fe p
∃! Fe
F
Y × {0} Y×I S1

Proof. First we define Fe locally, that is, on N × I for some neighborhood

N of a given point y0 ∈ Y. Then we show the uniqueness of Fe on sets
of the form {y0 } × I. This uniquely defines Fe on all of Y × I.

(Step 1) There is an open cover {Uα }α of S1 so that for each α, one has
p−1 (Uα ) = β U e β is an open interval in R that satisfies
F
e β , where each U
p (U e β ) = Uα and such that p restricts to a homeomorphism between U eβ
and Uα . For all pairs (y0 , s) ∈ Y × I, let α be such that F (y0 , s) ∈ Uα .
Since F is continuous, there is a neighborhood Ns × ( as , bs ) of (y0 , s)
so that F ( Ns × ( as , bs )) ⊆ Uα . Since {y0 } × I is compact, it can be
covered by finitely many such Ns × ( as , bs ). We can choose a single
neighborhood N of y0 and a partition of I given by 0 = s0 < s1 <
· · · < sm = 1 so that for each i there is an αi with F ( N × [si , si+1 ]) ⊂
Uαi . Assume (for induction) that Fe has been defined on N × [0, si ],
starting with the given lift on N × {0} for i = 0. We can extend it
to N × [si , si+1 ] as follows. Recall that since F ( N × [si , si+1 ]) ⊂ Uαi ,
we have Fe( N × {si }) ⊂ U e β , for a unique β i as above. Define Fe on
i
− 1
N × [si , si+1 ] by F = ( p |Uα : Uαi → U
e e β ) ◦ F.
i i

(Step 2) Now we show uniqueness for the case when Y is a single point.
Choose a partition of I by 0 = s0 < s1 < · · · < sm = 1 so that for all
i there is an open Uαi that completely contains F ([si , si+1 ]). Assume
we have Fe and Fe′ , two lifts of F : I → S1 . We have that Fe(0) = Fe′ (0),
since we are choosing a specific starting point x̃0 ∈ R. For induction,
suppose Fe and Fe′ coincide on [0, si ]. Since Fe is continuous and [si , si+1 ]
is connected, we have that Fe([si , si+1 ]) is connected. Thus there is a
unique U e β that completely contains Fe([si , si+1 ]). Similarly, there is a
i
unique Uβi′ ⊃ Fe′ ([si , si+1 ]).
e

⊂R
F
i Uβ i
e
Fe p

I ⊃ [ s i , s i +1 ] Uαi ⊂ S1
F

Since Fe(si ) = Fe′ (si ) by the induction hypothesis, and given that the sets
{Ue β } are either disjoint or equal, we must have that U
i
eβ = U
i
e β . Also,
i′
p| e is a homeomorphism, so p is injective on U e β and p ◦ Fe = p ◦ Fe′ .
Uβ i
i
Hence Fe = Fe′ on [si , si+1 ].
 fundamental group 15

(Step 3) The lifts Fe constructed on the sets N × I in (Step 1) are unique

by (Step 2) on each segment {y} × I, so two such lifts must agree on
their overlaps. This means, by gluing, that we get a well-defined lift
Fe : Y × I → R. Moreover, Fe is continuous since it is so on each set
N × I. Finally, Fe is unique by (Step 2).

Path lifting (Lemma 2.6.2) follows from Lemma 2.6.4 by letting Y be

a single point.
For homotopy lifting (Lemma 2.6.3), let Y = I in Lemma 2.6.4.
However, we are not being given a lift Fe : I × {0} → R of the homotopy
F : I × I → S1 . Let f s (t) = F (t, s). There is a unique lift Fe : I × {0} → R
obtained by applying the path-lifting Lemma 2.6.2 to f 0 : I → S1 . By
the general lifting Lemma 2.6.4, there is a then unique lift Fe : I × I → R.
So f˜s (t) = Fe(t, s) is a homotopy of paths lifting the homotopy f s , since
Fe |{0}× I and Fe |{1}× I are lifts of constant paths (indeed, p ◦ Fe(0, s) =
F (0, s) = F (0, 0), and similarly for Fe(1, s)), and by uniqueness, they are
also constant paths.

2.7 Some Immediate Applications

We start with the following:

Proposition 2.7.1. Sn is simply-connected if n ≥ 2.

Proof. Let γ : [0, 1] → Sn be a loop at x ∈ Sn . We claim that there

is a loop η in the homotopy class of γ which is not onto, i.e., there
exists y ̸= x with y ̸∈ Im(η ). Assuming this claim for now, η factors
as [0, 1] → Sn \{y} ∼
= Rn ,→ Sn , and since Rn is contractible, it follows
that η ∼ ex . Hence, by transitivity of the homotopy relation, we get
that γ ∼ ex .
To prove the claim, we can proceed in several different ways:
(a) A standard fact from differential topology is that any continuous
map between differentiable manifolds contains a smooth map in its
homotopy class. Using this fact, we have γ ∼ η : [0, 1] → Sn , where η
is smooth. Since dim([0, 1]) < dim (Sn ) = n ≥ 2, every value of η is
critical. But by Sard’s theorem, η has a regular value. If y ∈ Sn is such
a regular value of η, then y ̸∈ Im(η ).
(b) Point-set topology approach: Let y ̸= x. The goal is to homo-
top γ away from y. This can be done as follows. Let By be an
open ball in Sn around y. Note that γ−1 ( By ) is open in (0, 1), hence
γ−1 ( By ) = i∈ A ( ai , bi ), with A a possibly infinite index set. Since
F

γ−1 (y) is compact, only finitely many intervals ( ai , bi ) cover γ−1 (y).
Let ( a j , b j ), j ∈ A be so that ( a j , b j ) ∩ γ−1 (y) ̸= ∅. Let γ j := γ|[ a j ,bj ] ⊂ B̄y .
So γ( a j ), γ(b j ) ∈ ∂ B̄y = Syn−1 . As Syn−1 is path connected, there is a path
δj in Syn−1 from γ( a j ) to γ(b j ). Since B̄y is contractible, we then obtain
16 algebraic topology

that δj ∼ γ j in B̄y . Note that y ̸∈ Im δj . Homotop γ by deforming
γ j to δj , and keeping the rest of γ unchanged. Repeat the process
for all j’s such that ( a j , b j ) ∩ γ−1 (y) ̸= ∅. We get a loop η ∼ γ with
Im(η ) ∩ {y} = ∅.

Corollary 2.7.2. R2 is not homeomorphic to Rn if n ̸= 2.

Proof. If n = 1 and f : R2 → R is a homeomorphism then we have

R2 \{0} ∼= R\{ f (0)}. But R2 \{0} is path connected whereas R\{ f (0)}
is not path connected. Hence R2 ̸∼= R.
Now let n ≥ 2 and f : R2 → Rn be a homeomorphism. Then we have,

R2 \{0} ∼
= Rn \{ f (0)}

hence
π1 (R2 \{0}) ∼
= π1 (Rn \{ f (0)})
But we know that
(
Z, n=2
π1 (R \{ a}) ∼
n
= π 1 ( S n −1 ) =
0, n > 2.

Hence f cannot be a homeomorphism if n ̸= 2.

Brower’s fixed point theorem

Theorem 2.7.3. Any continuous map f : D2 → D2 has a fixed point.

Proof. Assume f ( x ) ̸= x for all x ∈ D2 . Let r : D2 → S1 be defined such

that r ( x ) is intersection of the line joining f ( x ) and x with S1 (with
/ S1 = ∂D2 ). We have r|S1 = idS1 , i.e.,
x between f ( x ) and r ( x ), if x ∈

S1 Figure 2.2: The map r : D2 → S1 .

r(x)

D2
f (x)

r ◦ i = idS1 for i : S1 ,→ D2 the inclusion map. We have the following

commutative diagram:
i
S1 D2
r
id
S1
 fundamental group 17

which, on the level of fundamental groups, yields the commutative

diagram:
i∗
Z 0
r∗
id
Z
This yields a contradiction since the identity map of Z cannot factor
through the zero map.

As an application of Brower’s fixed point theorem, we have the

following:

Proposition 2.7.4. Let A = ( aij ) ∈ M3 (R) be a 3 × 3 matrix with non-

negative real entries aij ≥ 0 for all i, j ∈ {1, 2, 3}. Assume det( A) ̸= 0. Then
A has a positive real eigenvalue.

Proof. Let T : R3 → R3 be the linear map corresponding to A. Let

B = S2 ∩ {( x1 , x2 , x3 ) ∈ R3 | x1 , x2 , x3 ≥ 0} ∼
= D2 .

If x ∈ B, then all coordinates of Tx = Ax are nonnegative, and not

all zero (since A is nonsingular and not all coordinates of x ∈ B can
be zero). So Tx/|| Tx || ∈ B. Let us now consider the continuous
map f : B → B defined as f ( x ) = Tx/|| Tx ||. By Brower’s fixed point
theorem, there exists x0 ∈ B so that f ( x0 ) = x0 , i.e., Tx0 = || Tx0 || x0 .
Setting λ = || Tx0 ||, we have that λ is an eigenvalue of A, with λ ∈ R
and λ > 0.

Fundamental Theorem of Algebra

Theorem 2.7.5. Let f (z) = zn + a1 zn−1 + · · · + an−1 z + an be a complex
polynomial. Then f has a complex root, i.e., f (z) = 0 has a solution in C.

Proof. If an = 0, then z = 0 is a solution. So we may assume an is

nonzero.
Define F (z, t) = zn + t( a1 zn−1 + · · · + an ), with z ∈ C and t ∈ [0, 1].
Clearly F is continuous, F (z, 0) = zn = pn (z) and F (z, 1) = f (z). So F
defines a homotopy between f : C → C and the n-th power function
pn .
Denote also by F its restriction to the circle Cr of radius r, i.e.,
Cr := {z ∈ C | |z| = r }. We see that for large enough r, F is nonzero.
Indeed, for large enough r,
 
| F (z, t)| ≥ |z|n − |t| | a1 ||z|n−1 + · · · + | an |
!
 |a | | a | 
1 n
= rn 1 − |t| +···+ n > 0.
r r
18 algebraic topology

So for large r, F is a homotopy Cr × I → C∗ = C\{0} from f to pn .

Assume, by contradiction, that f never vanishes. Define G (z, t) =
f (tz). Notice that G (z, 0) = f (0) = an and G (z, 1) = f (z). Restricting
to z ∈ Cr , G provides a homotopy Cr × I → C∗ from f to the constant
map ean . By transitivity, it follows that the power map pn (z) = zn and
the constant map are homotopic as maps Cr → C∗ . We thus obtain the
following commutative diagram

( pn )∗
Z∼
= π1 (Cr , r ) Z∼
= π 1 (C∗ , r n )
δ#
(e an )∗
Z∼
= π 1 (C∗ , a n )

with δ# the isomorphism induced from Lemma 2.4.6 by a certain path in

C∗ from r n to an . Let 1 denote the generator of π1 (Cr , r ), corresponding
to a loop at r going around the whole circle Cr . Then since (ean )∗ is
trivial, (ean )∗ (1) = 0, and since the diagram is commutative and δ♯ is
an isomorphism, δ♯ ◦ ( pn )∗ must be trivial as well, so ( pn )∗ (1) = 0. But
this contradicts the fact that ( pn )∗ (1) = n · 1. Hence our assumption
that f never vanishes is false, so f must have a complex root.

Exercises
1. Show that if h, h′ : X → Y are homotopic and k, k′ : Y → Z are
homotopic, then k ◦ h and k′ ◦ h′ are homotopic.

2. Let x0 and x1 be points of the path-connected space X. Show that

π1 ( X, x0 ) is abelian if and only if for every pair α and β of paths
from x0 to x1 , we have α# = β # : π1 ( X, x0 ) → π1 ( X, x1 ). (Recall
that α# : π1 ( X, x0 ) → π1 ( X, x1 ) is the group isomorphism defined by
α# ([γ]) := [α−1 ∗ γ ∗ α].)

3. Let A be a subspace of Rn ; let h : ( A, a0 ) → (Y, y0 ) be a continuous

map of pointed spaces. Show that if h is extendable to a continuous map
of Rn into Y, then h induces the trivial homomorphism on fundamental
groups (i.e., h∗ maps everything to the identity element).

4. Show that any two maps from an arbitrary space to a contractible

space are homotopic. As a consequence, prove that if X is a contractible
space, then any point in X is a deformation retract of X.

5. Show that if X and Y are path-connected spaces, and x ∈ X, y ∈ Y,

then π1 ( X × Y, ( x, y)) is isomorphic to π1 ( X, x ) × π1 (Y, y).

6. Using the fact that the fundamental group of the circle S1 is Z, show
that there are no retractions r : X → A in the following cases:
 fundamental group 19

(a) X = R3 , with A any subspace homeomorphic to S1 .

(b) X = S1 × D2 , with A its boundary torus S1 × S1 .

(c) X is the Möbius band and A its boundary circle.

7. Let V be a finite dimensional real vector space and W a subspace.

Compute π1 (V \ W ).

8. What is the fundamental group of RP2 minus a point?

9. Let A be a real 3 × 3 matrix, with all entries positive. Show that A has
a positive real eigenvalue. (Hint: Use Brower’s fixed point theorem.)

10. (Borsuk-Ulam theorem for S2 )

Given a continuous map f : S2 → R2 , there is a point x ∈ S2 such that
f ( x ) = f (− x ).
(Hint: show that there is no antipode-preserving map S2 → S1 .)

2.8 Seifert–Van Kampen’s Theorem

In this section, we explain how to compute the fundamental groups of

a space described by union of sets in terms of the fundamental groups
of these sets. We begin with a brief overview of free groups and free
products of groups, followed by the Seifert–Van Kampen theorem.

Free Groups
Definition 2.8.1. Let G be a group, and let { x j } j∈ J be a set of elements of
G. We say that the set { x j } j∈ J generates the group G if every element of G
can be written as a product of powers of the elements of { x j } j∈ J . If the family
{ x j } j∈ J is finite, we say that G is finitely generated.

Let X be a set. We want to construct a group F ( X ) generated by

the elements of X and which is “free”, in the sense that there are no
relations among its generators.

Definition 2.8.2. The set of words in X is the set

e
W ( X ) = {w = x11 . . . xnen | xi ∈ X, ei = ±1, n ∈ N}.

We also allow the empty word, denoted by 1 ∈ W ( X ).

We endow the set of words W ( X ) with the binary operation of

concatenation (or juxtaposition) of words.
We next define an equivalence relation on W ( X ). We need the
following:
20 algebraic topology

Definition 2.8.3. Let w and w′ be words in X. We say that w is equivalent

to w′ by an elementary reduction (and denote it by w ∼e w′ ) if one element of
the set {w, w′ } contains a subword of the form xx −1 or x −1 x, and the other
is obtained from it by deleting this subword.

Using this, we can define an equivalence relation on W ( X ) as follows.

Definition 2.8.4. Let w and w′ be words in X. We say that w is equivalent

to w′ (and write w ∼ w′ ) if there exist a sequence w1 , . . . , wk of words in X
such that
w = w1 ∼ e . . . ∼ e w k = w ′ .

Clearly, the relation ∼ defined above is reflexive, symmetric and

transitive, so it is an equivalence relation.

Remark 2.8.5. Each class of words contains a unique word of minimal

length (i.e., containing no subwords xx −1 or x −1 x), called a reduced
word.

Definition 2.8.6. We denote by F ( X ) := W ( X )/ ∼ the set of equivalence

classes of words.

It is easy to see that the relation ∼ on W ( X ) is consistent with

concatenation, that is, if w1 , w1′ , w2 , w2′ are words in X such that

w1 ∼ w1′
w2 ∼ w2′

then,
w1 w2 ∼ w1′ w2′ .
Thus, the binary law on W ( X ) given by concatenation descends to
F ( X ). Moreover, we have:

Theorem 2.8.7. The set F ( X ) of equivalence classes of words, with the induced
binary operation, is a group called the free group on the set X.

The group F ( X ) has the following universal mapping property (UMP):

Proposition 2.8.8 (UMP). Let

i : X −→ F ( X ), x 7→ [ x ]

be the map that sends every element of X to the equivalence class of the word
it defines, and let
j : X −→ G
be a set map from X to a group G. Then, there is a unique group homomorphism

f : F ( X ) −→ G

such that f ◦ i = j.
 fundamental group 21

Sketch of proof. We define the map f by

e
f ([ x11 . . . xnen ]) = j( x1 )e1 . . . j( xn )en ∈ G

This turns out to be well defined and it is a homomorphism of groups.

Example 2.8.9. Let X = { x }. Then,

F ( X ) = { x n | n ∈ Z} ∼
= Z.

Let G be the cyclic group of order n, that is,

G = ⟨ a | a n = 1⟩.

Then,
j : X → G, x 7→ a

gives a homomorphism

f : F ( X ) −→ G

which is surjective, with ker( f ) = ⟨ x n ⟩. Thus, we get that

G∼
= F ( X )/ ker( f ).

More generally, if G is a group generated by a set X, we can form

the free group F ( X ), and there is an epimorphism

f : F ( X ) −→ G.

Therefore,

G∼
= F ( X )/ ker( f ) = ⟨ x ∈ X | r ∈ ker( f )⟩

which gives us a presentation of G by generators (elements of X) and

relations (generators of ker( f )).

Free Products
Let H and K be groups. We form a new group H ∗ K from them, which
is called the free product of H and K, defined as follows. First we
consider the following set of words:

W ( H, K ) = { g1 g2 . . . gn | gi ∈ H or gi ∈ K }.

As before, we allow the empty word denoted by 1. The concatenation

of words defines a binary law on W ( H, K ).
We next define an equivalence relation on W ( H, K ) as follows:
22 algebraic topology

Definition 2.8.10. Let w and w′ be words in W ( H, K ). We say that w is

equivalent to w′ by an elementary reduction (and denote it by w ∼e w′ ) if
one of the elements of {w, w′ } contains a subword of the form ab, with both
a, b ∈ H or both a, b ∈ K, and the other is obtained from it by

• replacing the subword ab by the single element of H (or K) which is the

product a · b if a ̸= b−1 .
or

• removing the subword ab if a = b−1 .

Definition 2.8.11. Let w and w′ be words in W ( H, K ). We say that w and

w′ are equivalent, and write w ∼ w′ , if there exist a sequence w1 , . . . , wk of
words in W ( H, K ) such that

w = w1 ∼ e . . . ∼ e w k = w ′ .

Clearly, the relation ∼ is reflexive, symmetric and transitive, so it is

an equivalence relation.

Definition 2.8.12. Let H ∗ K := W ( H, K )/ ∼ be the set of equivalence

classes of words in W ( H, K ).

Remark 2.8.13. Any equivalence class contains a unique reduced word

h1 k 1 h2 k 2 . . . hr k r

with: 


 hi ∈ H for all i = 1, . . . , r

k ∈ K

for all j = 1, . . . , r
j
 i ̸= 1

 h for all i = 2, . . . , r


k ̸= 1 for all j = 1, . . . , r − 1.
j

The equivalence relation ∼ is consistent with concatenation, so the

binary law on W ( H, K ) descends to H ∗ K. We have the following result.

Theorem 2.8.14. The set H ∗ K, endowed with the operation induced from
concatenation, is a group, called the free product of H and K.

The group H ∗ K has the following universal mapping property (UMP):

Proposition 2.8.15 (UMP). Let

i : H −→ H ∗ K, h 7→ [h]

and
j : K −→ H ∗ K, k 7→ [k ]
be the maps that send every element of H (respectively, every element of K) to
the equivalence class of the word it defines, and let

p : H −→ G
 fundamental group 23

and
q : K −→ G
be any pair of group homomorphisms. Then, there is a unique group homomor-
phism
f : H ∗ K −→ G
such that f ◦ i = p and f ◦ j = q, or equivalently, such that the following
diagram is commutative.

p
i
∃! f
H∗K G

j
q

Sketch of proof. We define the map f on the unique reduced words as

f ( h1 k 1 h2 k 2 . . . hr k r ) = p ( h1 ) · q ( k 1 ) · p ( h2 ) · q ( k 2 ) · . . . · p ( hr ) · q ( k r ) ∈ G

This turns out to be well defined and it is a homomorphism of groups.

Corollary 2.8.16. The group H ∗ K is unique up to isomorphism.

Remark 2.8.17. Free products of any number of groups can be defined

similarly.

Example 2.8.18. Let X = { x1 , . . . , xn } be a set of n elements. We define

Fi := F ( xi ) ∼
=Z

for all i = 1, . . . , n. Then,

F(X ) ∼
= F1 ∗ . . . ∗ Fn ∼
= Z ∗ . . . ∗ Z = : Z∗ n

Example 2.8.19. If
H = ⟨h | rh ⟩
K = ⟨k | rk ⟩
are presentations of H and K by generators and relations, then

H ∗ K := ⟨h, k | rh , rk ⟩.

Example 2.8.20. The group Z2 ∗ Z2 is a free product, but it is not a free

group. Indeed,

Z2 ∗ Z2 = ⟨ a, b | a2 , b2 ⟩ = {1, a, b, ab, ba, aba, bab, abab, . . .}.

24 algebraic topology

Note that a−1 = a, b−1 = b, so ( ab)−1 = ba. Let

ω : Z2 ∗ Z2 −→ Z2 , x 7→ length of x mod 2.

Then ω is a homomorphism of groups, and

ker(ω ) = ⟨ ab⟩ ∼
=Z
We define the action ϕ of Z2 on Z = ⟨ ab⟩ by

ϕ : Z2 × Z −→ Z, ( a, ab) 7→ a( ab) a−1 = ba.

We have that ⟨ a⟩ ∩ ⟨ ab⟩ = {0}. Thus, Z2 ∗ Z2 = Z ⋊ Z2 , a semi-direct

product.
Remark 2.8.21. For a free product α∈∗A Hα , each group Hα is identified
with a subgroup of α∈∗A Hα , whose elements are the identity and the
one letter words h with h ∈ Hα . We have that
\
{1} = Hα ,
α∈ A

and, for all α, β ∈ A with α ̸= β,

( Hα \{1}) ∩ ( Hβ \{1}) = ∅.
For a free product of an arbitrary number of groups we also have
the Universal Mapping Property, namely:
Proposition 2.8.22 (UMP). Let { φα : Hα −→ G }α∈ A be a collection of
group homomorphisms, and let iα : Hα −→ α∈∗ A Hα be the inclusion for all
α ∈ A. Then, there exists a unique group homomorphism
∗
φ: α∈ A Hα −→ G
such that, for all α ∈ A,
φ ◦ iα = φα .
Sketch of proof. Let h1 h2 . . . hn be a word in α∈∗A Hα , with hi ∈ Hαi for all
i = 1, . . . , n. Define the map φ as:

φ ( h 1 h 2 . . . h n ) = φ α1 ( h 1 ) · φ α2 ( h 2 ) · . . . · φ α n ( h n ) ∈ G

This turns out to be well defined and it is a group homomorphism.

Example 2.8.23. Let

G = α∈×A Hα
be the cartesian product of the groups Hα , α ∈ A, and let

φ β : Hβ −→α∈×A Hα

be the inclusion for all β ∈ A. Then it follows from the UMP that there
exists a unique homomorphism
∗
φ: α∈ A Hα −→α∈×A Hα
that preserves every subgroup Hα .
 fundamental group 25

Seifert-Van Kampen Theorem

Let us now get back to calculating the fundamental group of a union of
sets.
Let X be a topological space, x0 ∈ X and let { Aα }α∈ J be open path
connected subsets of X such that
\
x0 ∈ Aα
α∈ J

and
[
X= Aα .
α∈ J
The inclusion
jα : Aα ,→ X
induces a homomorphism

( jα )∗ : π1 ( Aα , x0 ) −→ π1 ( X, x0 )
for all α ∈ J, so, by the UMP, there exists a unique homomorphism
∗
φ: α ∈ J π1 ( A α , x0 ) −→ π1 ( X, x0 ).
For every α, β ∈ J, with α ̸= β, we denote by iαβ the inclusion

iαβ : Aα ∩ A β ,→ Aα .

We then have a commutative diagram

iαβ Aα jα

Aα ∩ A β X

i βα jβ
Aβ

which induces the following commutative diagram on fundamental

groups

(iαβ )∗ π1 ( A α , x0 )
( jα )∗

π1 ( A α ∩ A β , x0 ) π1 ( X, x0 )

(i βα )∗ ( jβ )∗
π1 ( A β , x0 )

Thus, by the way φ is defined, we have that

(iαβ )∗ (ξ )((i βα )∗ (ξ ))−1 ∈ ker( φ)

for all ξ ∈ π1 ( Aα ∩ A β , x0 ), and for all α, β ∈ J.
With the above notations, we can state the Seifert-Van Kampen
Theorem.
 26 algebraic topology

S
Theorem 2.8.24 (Seifert-Van Kampen). If X = Aα is a topological
α∈ J
space, where Aα is a path-connected open set such that x0 ∈ Aα for all α ∈ J,
then

1) If Aα ∩ A β is path-connected for all α, β ∈ J, then

∗
φ: α ∈ J π1 ( A α , x0 ) −→ π1 ( X, x0 )

is surjective.

2) If Aα ∩ A β ∩ Aγ is path connected for all α, β, γ ∈ J, then

ker( φ) = N ⟨(iαβ )∗ (ξ )((i βα )∗ (ξ ))−1 | ξ ∈ π1 ( Aα ∩ A β , x0 ), α, β ∈ J ⟩

(2.8.1)
where N ⟨S⟩ denotes the normal subgroup generated by the set S.

Proof.
1) Let f : I −→ X be a loop at x0 ∈ X. By the continuity of f and the
compactness of I, there exists a partition 0 = s0 < s1 < . . . < sm = 1
such that
f ([si−1 , si ]) ⊂ Aαi
for some αi ∈ J. Denote by Ai the set Aαi , and let f i := f |[si−1 ,si ] . We
have that
f = f1 ∗ f2 ∗ . . . ∗ fm
with f i a path in Ai .
The set Ai ∩ Ai+1 is path-connected, and { x0 , f (si )} ⊂ Ai ∩ Ai+1 .
Thus, there exists a path gi in Ai ∩ Ai+1 from x0 to f (si ) for all i =
1, . . . , m − 1. Therefore,

f ∼ ( f 1 ∗ g 1 ) ∗ ( g1 ∗ f 2 ∗ g 2 ) ∗ . . . ∗ ( g m − 1 ∗ f m ) ,

and note that each of the paths in parentheses is a loop at x0 . Hence

[ f ] = [ f 1 ∗ g 1 ] · [ g1 ∗ f 2 ∗ g 2 ] · . . . · [ g m − 1 ∗ f m ] ,

where

• f 1 ∗ g1 is contained in A1 .

• gi ∗ f i+1 ∗ gi+1 is contained in Ai+1 for all i = 1, . . . , m − 1.

• gm−1 ∗ f m is contained in Am .
∗
Thus, if we see the classes of these loops as letters in α ∈ J π1 ( A α , x0 ), we
get that

[ f ] = φ([ f 1 ∗ g1 ] · [ g1 ∗ f 2 ∗ g2 ] · . . . · [ gm−1 ∗ f m ]),

and hence φ is surjective.

 fundamental group 27

Figure 2.3: Here, m = 2, f 1 is the path in

A1 := Aα from x0 to f (s1 ) and f 2 is the
path in A2 := A β from f (s1 ) to x0

g1 f ( s1 )
f1
Aα
x0 f2

Aβ

2) Clearly, (iαβ )∗ (ξ )((i βα )∗ (ξ ))−1 ∈ ker( φ) for all α, β ∈ J and for all
ξ ∈ π1 ( Aα ∩ A β , x0 ), so

N := N ⟨(iαβ )∗ (ξ )((i βα )∗ (ξ ))−1 | ξ ∈ π1 ( Aα ∩ A β , x0 ); α, β ∈ J ⟩ ⊂ ker( φ)

Hence, we get an induced homomorphism

∗
φ: α∈ J (π1 ( Aα , x0 )) /N −→ π1 ( X, x0 )

which is surjective since φ is.

To show that φ is injective, it suffices to show that it has a left
inverse, i.e., a homomorphism k such that k ◦ φ = id. Let H :=
φ : α∈∗ J (π1 ( Aα , x0 )) /N, and let φα : π1 ( Aα , x0 ) → H be given by the in-
clusion into the free product followed by the projection of the free prod-
uct onto its quotient by N. It is immediate to check that φα ◦ (iαβ )∗ =
φ β ◦ (i βα )∗ for any α ̸= β. With the help of the universal mapping
property, one can then construct a homomorphism k : π1 ( X, x0 ) → H
such that k ◦ ( jα )∗ = φα , for any α. To show that k is a left inverse for φ
it suffices to show that k ◦ φ acts as the identity on any generator of H,
i.e., on any coset of the form gN with g an element of some π1 ( Aα , x0 ).
But, for such a coset, w have

φ( gN ) = φ( g) = ( jα )∗ ( g),

so by applying k we get

k ( φ( gN )) = k(( jα )∗ ( g)) = φα ( g) = gN,

28 algebraic topology

as desired. (More details can be found Hatcher’s book, see also the
book of Munkres for the case | J | = 2.)

Most of the time, we will work with a covering consisting of two

open subsets. In this case, one gets the following:

Corollary 2.8.25. Let X = U ∪ V where U, V, and U ∩ V are open path

connected subsets of X. Fix x0 ∈ U ∩ V and consider the inclusion maps:

i U u
U∩V U∪V
j V v

Then
π1 (U, x0 ) ∗ π1 (V, x0 )
π1 ( X, x0 ) ∼
= .
N ⟨i∗ (ξ ) j∗ (ξ )−1 | ξ ∈ π1 (U ∩ V, x0 )⟩
Corollary 2.8.26. If U ∩ V is simply connected, then π1 (U ∪ V, x0 ) =
π1 (U, x0 ) ∗ π1 (V, x0 ).

Corollary 2.8.27. The union of two simply connected spaces is simply con-
nected, provided their intersection is nonempty and path connected.

Example 2.8.28. Let X = Sn for n ≥ 2. Let U to be the complement of

the north pole, and let V be the complement of the south pole. Then
U, V, and U ∩ V are all open and path connected, and U and V are
contractible. So, by Corollary 2.8.26, Sn is simply connected.

Definition 2.8.29. Given two spaces X and Y with distinguished points x0

and y0 respectively, the wedge of X and Y is defined by:

X ∨ Y := X ⊔ Y x0 ∼ y0 .

Example 2.8.30. Let Xn = in=1 S1 be a wedge of n circles at a single

W

point (called a bouquet of n circles). Then π1 ( Xn ) = Z∗n , that is, a free

product of n copies of Z. To see this, we proceed by induction on n.
For n = 1 we get a single circle, so the result is clear. For induction,
suppose we have shown that π1 ( Xn−1 ) = Z∗(n−1) . Let x0 be the wedge
point of n circles. For each i choose pi ̸= x0 to be a point on the i-th
circle. Let
−1
n_
U = Xn \{ pn } ≃ S1 = Xn−1 and V = Xn \{ p2 , · · · , pn−1 } ≃ S1 .
i =1

Then U ∩ V ≃ { x0 }, so by the Seifert-Van Kampen theorem (Corollary

2.8.25) we get that π1 ( Xn ) ∼
= π 1 (U ) ∗ π 1 ( V ) ∼
= π1 (U ) ∗ Z, and by the
induction hypothesis, this gives Z∗(n−1) ∗ Z ∼ = Z∗ n .
Example 2.8.31. Let X = R2 \{ x1 , · · · , xn }. Then X deformation retracts
to a bouquet of n circles, one going around each xi . So π1 ( X ) = Z∗n .
 fundamental group 29

Example 2.8.32. Let X = R3 \{coordinate axes}. Then X deformation

retracts (via x 7→ ||xx|| ) to S2 \{6 points} ∼
= R2 \{5 points}, where the
∼
last identification = is by stereographic projection. Then it is clear that
π 1 ( X ) = Z∗5 .

Example 2.8.33. Let X = S2 ∪ {equatorial disk ≈ D2 }, so S2 ∩ D2 =

{the equator S1 }. Take U = X \{north pole} and V = X \{south pole},
and note that both U and V are homotopic to S2 . Moreover, U ∩ V ≃ D2
which is contractible. Since U and V are simply connected, X is simply
connected.

Example 2.8.34. Let X = S2 ∪ {north-south diameter}. Let P be a

point on the diameter different from the poles. Let Q be a point on
the sphere different from the poles. Choose U = X \{ P} ≃ S2 and
V = X \{ Q} ≃ S1 . Notice that U ∩ V ≃ S2 \{ Q} ∼ = R2 . Since U ∩ V is
simply connected, we get that π1 ( X ) ∼
= π1 (U ) ∗ π1 (V ) = 0 ∗ Z = Z.

Exercises
1. Let X be the space obtained from D2 by identifying two distinct
points on its boundary. Is there a retract from X to its boundary?
Explain.

2. Calculate the fundamental group of the spaces below:

(i) R3 \ { x − axis and y − axis}.

(ii) The complement in R3 of a line and a point not on the line.

(iii) R3 minus two disjoint lines.

(iv) T 2 \ { x, y}, where x, y are two distinct points on the 2-torus.

(v) Möbius band. Are the cylinder and the Möbius band homeomor-
phic?

(vi) The complement in R3 of a line and a circle. Note: There are two
cases to consider, one where the line goes through the interior of the
circle and the other where it doesn’t. Are these two spaces homotopy
equivalent?

3. Show that RP3 and RP2 ∨ S3 have the same fundamental group.
Are they homeomorphic?

4. For a given a sequence of continuous maps

f1 f2 f3
X1 → X2 → X3 → · · ·
30 algebraic topology

define the quotient space

!
G
M := Xi × [0, 1] / (( xi , 1) ∼ ( f i ( xi ), 0))
i ≥1

obtained from the disjoint union of cylinders Xi × [0, 1] via the identifi-
cation of ( xi , 1) ∈ Xi × {1} with ( f i ( xi ), 0) ∈ Xi+1 × {0}. Compute the
fundamental group of M in the case when each Xi is a circle S1 and
f i : S1 → S1 is the map z 7→ zi (for each i ≥ 1).

5. For relatively prime positive integers m and n, the torus knot Km,n ⊂
R3 is the image of the embedding f : S1 → S1 × S1 ⊂ R3 , f (z) =
(zm , zn ), where the torus S1 × S1 is embedded in R3 in the standard
way. Compute π1 (R3 \ Km,n ).
 classification of compact surfaces 31

3
Classification of compact surfaces

The goal of this section is to show that the fundamental group is

powerful enough to classify real compact surfaces.

3.1 Surfaces: definitions, examples

Definition 3.1.1. An n-dimensional manifold with no boundary is a topologi-

cal space X such that every x ∈ X has a neighborhood Ux homeomorphic to
Rn .
Definition 3.1.2. A surface is a 2-dimensional manifold with no boundary.

In this section, we will work with (and classify) compact surfaces.

Let P be a polygonal region in the plane, with vertices p0 , p1 ,. . . ,
pm−1 and edges with oriented labels like in the picture below.
a1
p3 p2

a3 a2

p4 p1

a2 a1

p5 a3 p0

Going through the vertices starting at p0 in counter-clockwise order

gives us a labeling scheme. In the above example, the labeling scheme is

a1 a2 a1−1 a3−1 a2 a3 .

From P and the labeling scheme, we get an identification (quotient)

space X with a quotient map π : P → X as follows:

• The points in the interior of P are identified only to themselves.

32 algebraic topology

• Two edges carrying the same label are identified by an orientation

preserving linear homeomorphism.

Example 3.1.3 (The torus T 2 ). We start with the following polygonal

region,

P a P

b T2 b

P a P

with labeling scheme aba−1 b−1 . First, we glue the a labels together to
get a cylinder:

a
P P

b b

Next, we glue the b labels together to get the torus T 2 .

Figure 3.1: Torus T 2

a
P
b

Example 3.1.4 (The Sphere S2 ). From the polygonal region

 classification of compact surfaces 33

a a

with labeling scheme aa−1 , we get the sphere S2 by gluing the a labels
together.

Example 3.1.5 (The Projective Plane RP2 ). From the polygonal region

a a

with labeling scheme aa, we get the (real) projective plane RP2 by
gluing the a labels together.

Example 3.1.6 (The Klein Bottle K). We start with the following polygo-
nal region:

P a P

b K b

P a P

with labeling scheme aba−1 b. First, we glue the a labels together to

get a cylinder just like in Example 3.1.3, but this time, the b labels do
34 algebraic topology

not glue together so nicely, that is, the surface we get by gluing them
together cannot be embedded in R3 . The resulting surface is called the
Klein bottle (figure 3.2).

Figure 3.2: Klein bottle

Proposition 3.1.7. The identification space X obtained from a polygonal

region P as above is Hausdorff and compact.

Proof. Let π : P → X be the projection, where X has the quotient

topology. Note that π is continuous by the definition of the quotient
topology. Since P is compact, it follows that X = π ( P) is compact.
We next show that π is a closed map. If C is a closed set in P,
then π (C ) is closed if and only if X \π (C ) is open, or equivalently
π −1 ( X \π (C )) is an open set in P. We have that

π −1 ( X \π (C )) = P\π −1 (π (C ))

The only nontrivial identifications occur in the edges of P, which are

closed in P, and thus the intersection of C with any edge is again a
closed set. Therefore π −1 (π (C )) is just the union of C and a finite
number of other closed sets. Thus, P\π −1 (π (C )) is open, and π is
closed.
A quotient map f : Y → Z from a compact Hausdorff space Y is
closed if and only if Z is Hausdorff, so applying this result to π we get
that X is Hausdorff.

Definition 3.1.8. Let M, N be surfaces. We define the connected sum of M

and N, denoted by M#N, as follows:

M#N = ( M \ D1 ) ⊔ ( N \ D2 )/(∂D1 ∼ ∂D2 )

where D1 is a disk in M and D2 is a disk in N.

 classification of compact surfaces 35

Lemma 3.1.9. If L1 and L2 are labeling schemes for M and N, then their
concatenation L1 L2 is a labeling scheme for M#N.

Example 3.1.10. The connected sum T 2 #T 2 of two tori has a labeling

scheme a1 b1 a1−1 b1−1 a2 b2 a2−1 b2−1 . Indeed, let T12 be the following torus,
and D1 a disk inside of it
a1

b1 b1

D1

a1

and let T22 be the following torus, and D2 a disk inside of it:
a2

b2 b2

D2

a2

The following polygonal regions represent T12 \ D1 and T22 \ D2 respec-

tively.

a1 b1 a2 b2

b1 a1 b2 a2

∂D1 ∂D2

To get the connected sum of the two tori, we need to glue ∂D1 with
∂D2 , and we get the polygonal region which has the labeling scheme

a1 b1 a1−1 b1−1 a2 b2 a2−1 b2−1 ,

that is, the concatenation of the labeling schemes of T12 and T22 .
36 algebraic topology

a1 b1 Figure 3.3: T 2 #T 2

b1 a1

a2 b2

b2 a2

Definition 3.1.11. We introduce the following notation:

n times
z }| {
Tn := T 2 # . . . #T 2

and
n times
z }| {
Pn := RP2 # . . . #RP2 .

Our goal is to prove the following classification result.

Theorem 3.1.12. Any compact surface is homeomorphic to S2 , Tn or Pn for

some n ∈ N.

3.2 Fundamental group of a labeling scheme

Before giving a general result, we compute the fundamental group of

the torus T 2 . Consider the identification space P of the torus given
given by aba−1 b−1 . Let 0 be some point on the interior of the square.
Define U = P \ {0} and V = Bϵ (0), a ball of radius ϵ centered at 0. We
have that U ≃ S1 ∨ S1 , V is contractible, and U ∩ V = Bϵ (0) \ {0} ≃ S1 .
Then we have:

• π1 (U, A) = Z ∗ Z = ⟨ a, b⟩

• π1 (V, x0 ) is trivial

• π1 (U ∩ V, x0 ) ∼
=Z∼
= ⟨c⟩
 classification of compact surfaces 37

A a A

0
b b
x0
V

δ
A a A

Let i : U ∩ V ,→ U be the inclusion map. By the Seifert-Van Kampen

Theorem,

π1 ( T 2 , x0 ) ∼
= π1 (U, x0 )/N ⟨i∗ ξ | ξ ∈ π1 (U ∩ V, x0 )⟩.

Let δ be a path in T 2 from A to x0 . Then δ# : π1 (U, A) → π1 (U, x0 ) is

an isomorphism mapping [γ] 7→ [δ̄ ∗ γ ∗ δ]. Moreover, a, b ∈ π1 (U, A)
induce loops at x0 given by ã := δ̄ ∗ a ∗ δ and b̃ := δ̄ ∗ b ∗ δ, which freely
generate π1 (U, x0 ). In the above notations, we have

π1 ( T 2 , x0 ) ∼
= ⟨ ã, b̃⟩/N ⟨i∗ c⟩.

We next note that i∗ c is homotopic to δ̄ ∗ a ∗ b ∗ ā ∗ b̄ ∗ δ, and we have:

δ̄ ∗ a ∗ b ∗ ā ∗ b̄ ∗ δ ∼ (δ̄ ∗ a ∗ δ) ∗ (δ̄ ∗ b ∗ δ) ∗ (δ̄ ∗ ā ∗ δ) ∗ (δ̄ ∗ b̄ ∗ δ)

= ã ∗ b̃ ∗ 㯠∗ b̃¯ .

Hence π1 ( T 2 , x0 ) ∼
= ⟨ ã, b̃ | ãb̃ ã−1 b̃−1 ⟩ ∼
= Z × Z.
Similar calculations yield the following theorem:
Theorem 3.2.1. If X is the identification space of a labeling scheme
ϵ
a11 a2ϵ2 . . . aϵnn

with ϵi = ±1 whose vertices are all identified by the projection map π : P → X,

then:
ϵ
π1 ( X ) = ⟨ a1 , a2 , . . . , an | a11 a2ϵ2 . . . aϵnn = 1⟩.
Example 3.2.2. If K is the Klein bottle, with labelling scheme aba−1 b,
we get
π1 ( K ) ∼
= ⟨ a, b | aba−1 b = 1⟩.
So π1 (K ) is not abelian.
Example 3.2.3. Consider the labeling schemes for Tn and Pn .

. . #T}2 : a1 b1 a1−1 b1−1 . . . an bn a−

Tn = |T 2 # .{z 1 −1
n an
n−times
38 algebraic topology

2
Pn = RP
| . . . RP}2 : a1 a1 . . . an an
#{z
n−times

Notice that all vertices are identified in each labelling scheme. The
above theorem gives us:

π1 ( Tn ) = ⟨ a1 , b1 , . . . , an , bn | a1 b1 a1−1 b1−1 . . . an bn a− 1 −1
n bn = 1 ⟩
π1 ( Pn ) = ⟨ a1 , . . . , an | a21 . . . a2n = 1⟩

We deduce the following:

Proposition 3.2.4. The surfaces S2 , Pn , Tn (n ∈ N) have non isomorphic

fundamental groups, hence they are not homotopy equivalent nor homeomor-
phic.

Proof. First, π1 (S2 ) is trivial. Next, consider π1ab := π1 /[π1 , π1 ], the

abelianized fundamental group, where for a group G its commutator
subgroup is defined as [ G, G ] = {[ a, b] = aba−1 b−1 | a, b ∈ G }. We
have:
n
π1ab ( Tn ) = ⟨ a1 , b1 , . . . , an , bn | ∑ a i + bi − a i − bi = 0 ⟩
i =1
∼
=Z . . × Z = Z2n
× .{z
| }
2n−times

π1ab ( Pn ) = ⟨ a1 , . . . , a n | 2( a1 + . . . + a n ) = 0⟩
| {z }
An

= ⟨ a1 , . . . , an−1 , An | 2An ) = 0⟩ ∼
= Zn−1 × Z/2

The assertion follows now easily.

Recall that our goal is to show the following:

Theorem 3.2.5. Any compact surface is homeomorphic to one of S2 , Tn or

Pn , for some n ∈ N.

Corollary 3.2.6. If X is a simply connected compact surface, then it is

homeomorphic to S2 .

One dimension higher, things are much more complicated, but we

still have the following:

Theorem 3.2.7 (Poincaré Conjecture). If X is a simply connected closed

(i.e., compact, with no boundary) 3-manifold, then it is homeomorphic to S3 .

This is false in dimension 4, since S4 and S2 × S2 are simply con-

nected closed 4-manifolds, but they are not homeomorphic. (This fact
can be easily seen with homology or higher homotopy groups.) In
higher dimensions, one has the following important result:
 classification of compact surfaces 39

Theorem 3.2.8 (Smale, Freedman). If n ≥ 4 then any simply-connected

closed n-manifold which is homotopy equivalent to Sn is homeomorphic to Sn .

Before discussing the proof of the classification theorem for surfaces

(Theorem 3.2.5) it is an instructive exercise to see where the Klein bottle
and T 2 #RP2 fit on the list. This will be done by cutting and pasting on
the labeling scheme.

Example 3.2.9. In the notation of Figure 3.4, we cut along the diagonal
labeled c and glue along a to show that

K∼
= P2 .

(Note that cutting and pasting do not change the homeomorphism

type.)

a a Figure 3.4: How to turn the Klein bottle

into P2
I
c c
b b b
c
II b
a
a

b c
b II
a

c a
b I
b c
c

Example 3.2.10. We next claim that

K#RP2 ∼
= T 2 #RP2 ∼
= P3 .

We start by looking at RP2 \ disc ∼= S2 \(2 antipodal discs)/antipodal

identification; see Figure (3.5).
Attaching a torus is likened to attaching a handle, while attaching a
Klein bottle is likened to attaching an orientation-reversing (twisted)
handle, see Figure (3.6).
Therefore, T 2 #RP2 looks like a Möbius band with a handle attached to
it. Cutting the band away from the handle leads to the space pictured in
40 algebraic topology

Figure 3.5: Removing a disc from RP2

yields a Möbius band.

Figure 3.6: Performing connected sum

with a Klein bottle.

Figure (3.7). Similarly, K#RP2 looks like a Möbius band with a twisted
handle attached to it. Cutting the band between the legs of the handle
leads to the same space as in Figure (3.7). Hence the assertion follows.

Figure 3.7: T 2 #RP2 .

3.3 Classification of surfaces

We begin with two results whose proofs you can find in Munkres’ book.

Proposition 3.3.1. If P is a polygonal region with an even number of edges

which are identified in pairs (i.e., a regular labeling scheme), then the quotient
space X is a compact 2-dimensional manifold.
 classification of compact surfaces 41

Figure 3.8: Every point has a neighbor-

hood homeomorphic to a disc.

a a

Theorem 3.3.2. Every 2-dimensional compact surface is homeomorphic to the

identification space of a regular labeling scheme

The proof of the above theorem is based on the fact that each 2-
dimensional compact surface has a triangulation, and when we glue
half discs together along a common edge, we get a disc.
The arguments involved in the following classification of labelling
schemes provides the algorithm needed to identify any surface in the
list S2 , Tn , Pn , n ∈ N).

Theorem 3.3.3. A polygonal region of a regular labeling scheme is homeomor-

phic to a standard labelling scheme, i.e., one of the following:

• S2 : aa−1

• Tn : a1 b1 a1−1 b1−1 . . . an bn a− 1 −1
n bn

• Pn : a1 a1 a2 a2 . . . an an

Proof. Edges are of two kinds:

• first kind: a . . . a−1

• second kind: a . . . a

Here are the steps involved in the cut and paste algorithm.

Step 1: Adjacent edges of the first kind can be removed. (See Figure
(3.9), where the edge labeled a is removed.)

Step 2: All vertices get identified to one vertex.

In Figure (3.10), we cut along the edge labeled c and glue along a.
The effect is that the equivalence class of the vertex Q (consisting of
vertices identified to Q) is reduced by 1, while that of P is increased by
1.
Repeat until only one vertex labelled Q is left, and then we use Step
1 to remove it. Repeat this procedure until only one equivalence class
of vertices is left.
42 algebraic topology

Figure 3.9: Step 1: Removing adjacent

edges of the first kind.

a
a

Q c P Figure 3.10: Step 2: identifying all ver-

tices.
a b

c P
a
Q

c b
a Q P

Figure 3.11: Step 3: Making two Type II

edges adjacent.
a c a c a c
 classification of compact surfaces 43

Step 3: Make any pair of edges of second kind adjacent. (See Figure
(3.11).)
Here we cut along the edge c and, after flipping one of the two
pieces obtained, we glue along a. After removing the interior label a,
we created the subword cc, which corresponds to a pair of adjacent
edges of second kind.

Step 4: If a is an edge of the first kind, then there are two edges of the
−1
first kind which alternate: . . . a . . . a′ . . . a−1 . . . a′ . . ..
If this is not the case, the the edges of the region connecting the
vertices P in Figure (3.12) only get identified to edges from the same
region. The same applies for the region between the vertices labeled Q.
But then the endpoints of the edge a cannot be identified, contradicting
Step 2.

Q Q Figure 3.12: Step 4.

a
a a
P Q

P P

Step 5: Any two pairs of the first kind can be made consecutive. See
Figure (3.13).

b c Figure 3.13: Step 5: Two pairs of the first

kind being made consecutive.

a a
c
a a
b

b c

c
c d
a a
a
d

c c
44 algebraic topology

Here we first cut along c and glue along b, then cut along d and glue
along a.

At this point, the labeling scheme corresponds to a connected sum

of RP2 ’s and T 2 ’s. If there is no RP2 , then we get a Tn for some n ∈ N.
Otherwise, we proceed as in the following step.

Step 6: Transform . . . ccaba−1 b−1 . . . into . . . P3 . . ..

This was already explained geometrically in Example 3.2.10. We
sketch here the corresponding cut and paste procedure. It should be
clear at this point that we can just ignore the rest of the surface. See
Figure (3.14). The idea is to convert a . . . a−1 into a . . . a, so one gets 3
pairs of edges of the second kind, and apply Step 3 (see Figure (3.15)).

c d Figure 3.14: Making all sides have the

same orientation: cut along d, glue along
c.
b c b b

a a a a

b d

d e Figure 3.15: Completing Step 6.

e
b b e a

f
a a d d

d a

e e

e g e d
g

f g d f

f f

In Figure (3.14), we cut along d and glue along c. In Figure (3.15),

 classification of compact surfaces 45

we first cut along e and glue along b, then cut along f and glue along a,
and finally cut along g and glue along d.

Exercises
1. There are six ways to obtain a compact surface by identifying pairs
of sides in a square. In each case determine what surface one obtains.

2. The following labeling schemes describe two dimensional surfaces:

• abc−1 b−1 a−1 c

• abc−1 c−1 ba

• a 1 a 2 · · · a n a 1 −1 a 2 −1 · · · a n −1

In each case determine what standard surface it is homeomorphic

to.

3. Consider the space X obtained from a seven-sided polygonal region

by means of the labeling scheme abaaab−1 a−1 . Show that π1 ( X ) is the
free product of two cyclic groups.

4. Let X be the quotient space obtained from an eight-sided polygonal

region P by means of the labeling scheme abcdad−1 cb−1 . Let π : P → X
be the quotient map.

• Show that π does not map all the vertices of P to the same point of
X.

• Determine the space A = π (Bd P) (the boundary of P), and calculate

its fundamental group.

• Calculate the fundamental group of X. (Hint: first transform the

labeling scheme into a standard one by cutting and pasting opera-
tions.)

• What surface is X homeomorphic to?

5. Let X be a space obtained by pasting the edges of a polygonal region

together in pairs.

• Show that X is homeomorphic to exactly one of the spaces in the

following list: S2 , P2 , K, Tn , Tn #P2 , Tn #K, where K is the Klein bottle
and n ≥ 1.

• Show that X is homeomorphic to exactly one of the spaces in the fol-

lowing list: S2 , P2 , Km , Tn , P2 #Km , where Km is the m-fold connected
sum of K with itself and m ≥ 1.
46 algebraic topology

6. Let A be the annulus in the plane consisting of the set

A := {( x, y) ∈ R2 | 1 ≤ x2 + y2 ≤ 4}.

Let S denote the surface obtained from A by identifying antipodal

points of the inner circle and by identifying antipodal points of the
outer circle. Compute π1 (S) and write S as a connected sum of tori
and projective planes.

7. Let X be the topological space obtained by identifying by parallel

translation the opposite edges of a solid regular hexagon. Calculate the
fundamental group of X.
 covering spaces 47

4
Covering spaces

In this chapter we introduce covering spaces and show how they can be
used for computing fundamental groups. In addition, we will use the
fundamental group as a tool for studying covering spaces.

4.1 Definition. Properties

Definition 4.1.1. A map p : E → B is called a covering if

(a) p is continuous and onto.

(b) For all b ∈ B, there exists an open neighborhood U of b which is “evenly

covered”, i.e., p−1 (U ) = α Vα , where the Vα are disjoint and open, and
F

p|Vα : Vα → U is a homeomorphism for each α.

Example 4.1.2. It is easy to check that the following maps are coverings.

(i) p : R → S1 , t 7→ e2πit .

(ii) id X : X → X.

(iii) p : X × {1, . . . , n} → X, ( x, k) 7→ x.

(iv) p : S1 → S1 , z 7→ zn .

(v) p : Sn → RPn , x 7→ [ x ], the quotient map identifying antipodal

points of Sn .

(vi) p : C → C∗ , z 7→ ez .

(vii) Products of covering maps are covering maps, i.e., if pi : Ei → Bi , i =

1, 2, are coverings, then p1 × p2 : E1 × E2 → B1 × B2 is a covering.

Remark 4.1.3. 1. A covering map is open and locally a homeomor-

phism.

2. Not any local homeomorphism is a covering, e.g., p : R∗+ → S1 ,

t 7→ e2πit . Hence a restriction of a covering map does not have to be
a covering.
48 algebraic topology

3. If p : E → B is a covering, then each fiber p−1 (b), b ∈ B, is discrete.

Definition 4.1.4. Let p1 : E1 → B, p2 : E2 → B be two coverings. We say

that p1 and p2 are equivalent if there exists a homeomorphism f : E1 → E2
such that p2 ◦ f = p1 .

Remark 4.1.5. The equivalence of coverings is an equivalence relation.

In this chapter we elaborate on the following:

Problem 4.1.6. Classify all coverings of a space B (up to equivalence).

The proof of the following lemma is a simple exercise in point set

topology.

Lemma 4.1.7. If p : E → B is a covering, B0 ⊂ B, and E0 := p−1 ( B0 ), then

p| E0 : E0 → B0 is a covering.

Example 4.1.8. We know from Example 4.1.2 that p : R2 → T 2 is a

covering. Overlay the integer lattice on R2 , and identify each square
with a torus in the usual way. Let p0 = (1, 0) ∈ S1 , and let B0 = S1 ×
{ p0 } ∪ { p0 } × S1 . Then p−1 ( B0 ) = R × Z ∪ Z × R, and the restriction
of p to this space is a covering over B0 .

Theorems 4.1.9 and 4.1.10 are generalizations from the case of the
covering p : R → S1 , with similar proofs.

Theorem 4.1.9 (Path lifting property). Let p : E → B be a covering, b0 ∈ B,

and e0 ∈ p−1 (b0 ). If γ : I → B is a path in B starting at b0 , then there is a
ee0 : I → E such that γ
unique lift γ ee0 (0) = e0 .

Theorem 4.1.10 (Homotopy lifting property). Let p : E → B be a covering,

b0 ∈ B, and e0 ∈ p−1 (b0 ). Let F : I × I → B be a homotopy with b0 :=
F (0, s) for all s ∈ I. Then there is a unique lift Fe : I × I → E of F such that
Fe(0, s) = e0 for all s ∈ I.

Corollary 4.1.11. If γ1 , γ2 are paths in B starting at b0 which are homotopic

Fe
by some homotopy F, then (γ f1 )e0 ∼ (γ f2 )e0 . In particular, these lifts have the
same endpoints: (γ
f1 )e0 (1) = (γf2 )e0 (1).

Definition 4.1.12. Let b0 ∈ B. For e0 ∈ p−1 (b0 ), define

ϕe0 : π1 ( B, b0 ) −→ p−1 (b0 )

[γ] 7→ γ
ee0 (1)

Theorem 4.1.13. The map ϕe0 defined above is onto if E is path-connected,

and it is injective if E is simply connected.
 covering spaces 49

Proof. By Corollary 4.1.11, the map ϕe0 is well-defined.

Suppose that E is path-connected. Let e1 ∈ p−1 (b0 ), and let δ be a
path in E from e0 to e1 . Then γ := p ◦ δ : I → B is a loop in B at b0 .
Hence δ is a lift of γ starting at e0 , and we have ϕe0 ([γ]) = γ ee0 (1) =
δ(1) = e1 , so ϕe0 is surjective. Note that the equality γ ee0 (1) = δ(1)
comes from the uniqueness of lifts (Theorem 4.1.9).
Now suppose E is simply connected. Let γ1 , γ2 be loops in B at
b0 such that ϕe0 ([γ1 ]) = ϕe0 ([γ2 ]) = e1 . By definition, this means that
(γ
e1 )e0 (1) = (γ
e2 )e0 (1). To show that ϕe0 is injective, we must show that
γ1 ∼ γ2 . Since E is simply connected, there is a unique homotopy
class of paths from e0 to e1 , so (γ e1 )e0 ∼ (γe2 )e0 by some homotopy
F. This gives a homotopy p ◦ F : I × I → B from p ◦ (γ f1 )e0 = γ1 to
p ◦ (γ
f2 )e0 = γ2 , which shows that ϕe0 is injective.

Example 4.1.14. It is very easy to check that the antipodal identification

yields a covering map p : Sn → RPn . For n ≥ 2, Sn is path-connected
and simply connected. Then by Theorem 4.1.13,

Φe0 : π1 (RPn , b0 ) → p−1 (b0 )

is a bijection. Since the fiber p−1 (b0 ) has only two elements, we must
have that π1 (RPn , b0 ) ∼
= Z/2Z as a group isomorphism.
Example 4.1.15. Let p : R → S1 , t 7→ e2πit . Since R is both simply
connected and path-connected, Theorem 4.1.13 yields that

ϕe0 : π1 (S1 , b0 ) → Z

is a bijection. To show that the groups are isomorphic, we need to show

that ϕe0 is a homomorphism. Let γ, δ ∈ π1 (S1 , b0 ), and let γ e0 , δe0 be
their lifts in R starting at 0. Let γe0 (1) = n ∈ Z and δ0 (1) = m ∈ Z. By
e
definition, ϕe0 ([γ]) = n, ϕe0 ([δ]) = m. Hence we need to show that

ϕe0 ([γ] · [δ]) = n + m.

We have

ϕe0 ([γ] · [δ]) = ϕe0 ([γ ∗ δ]) = (^ e0 ∗ δe∗ )(1) = δe∗ (1)
γ ∗ δ )0 (1) = ( γ
= n + m,

where we set δe∗ (t) = n + δe0 (t) so that δe∗ (0) = n, δe∗ (1) = n + m. Thus,
ϕe0 is a homomorphism and therefore an isomorphism.
Proposition 4.1.16. If p : E → B is a covering and B is path-connected, then
for b0 , b1 ∈ B there is a bijection p−1 (b0 ) → p−1 (b1 ).

Proof. Let γ be a path in B from b0 to b1 (which exists since B is path-

connected). Define the bijection f γ : p−1 (b0 ) → p−1 (b1 ) by e0 7→ γ
ee0 (1).
− 1
It has the inverse ( f γ ) = f γ .
 50 algebraic topology

Proposition 4.1.17. Let E be path connected, p : E → B a covering, and

p(e0 ) = b0 . Then p∗ : π1 ( E, e0 ) → π1 ( B, b0 ) is injective. Further, if e0 is
changed to some other point e1 ∈ p−1 (b0 ), then the images under p∗ of the
groups π1 ( E, e0 ) and π1 ( E, e1 ) are conjugate in π1 ( B, b0 ).

Proof. Let γ1 , γ2 ∈ π1 ( E, e0 ) with p∗ ([γ1 ]) = p∗ ([γ2 ]). Then p ◦ γ1 ∼

p ◦ γ2 by some homotopy F. By homotopy lifting (Theorem 4.1.10), we
have that ( p^ ◦ γ1 )e0 ∼ ( p^
◦ γ2 )e0 , which implies that γ1 ∼ γ2 , by the
uniqueness of lifts. Indeed, for i = 1, 2, both γi and ( ^ p ◦ γi )e0 are lifts
of p ◦ γi starting at e0 , so they must coincide. Thus, p∗ is injective.
Now let e1 be a different point in the fiber of p over b0 . Let H0 =
p∗ π1 ( E, e0 ), H1 = p∗ π1 ( E, e1 ). We want to show these are conjugate
subgroups. First let δ be a path in E from e0 to e1 . Then the following
diagram commutes:
p∗
π1 ( E, e0 ) π1 ( B, b0 )
δ# ( p ◦ δ )#
p∗
π1 ( E, e1 ) π1 ( B, b0 )

Note that δ# is an isomorphism since E is path connected. So H0 and

( p ◦ δ)# H1 are conjugate subgroups via [ p ◦ δ].

Theorem 4.1.18. Let E be path-connected, p : E → B a covering map, b0 ∈ B

and e0 ∈ p−1 (b0 ). Let H := p∗ π1 ( E, e0 ) ≤ π1 ( B, b0 ). Then:
(a) A closed path γ in B based at b0 lifts to a loop in E at e0 if and only if
[γ] ∈ H.

(b) ϕe0 : H \π1 ( B, b0 ) → p−1 (b0 ), [γ] 7→ γ

ee0 (1) is a bijection. In particular,

#p−1 (b0 ) = [π1 ( B, b0 ) : p∗ π1 ( E, e0 )],

with # denoting cardinality.

Proof. Part ( a) is immediate. For (b), we first show that ϕe0 is well-
defined, i.e., if [δ] ∈ H, then ϕe0 ([δ] · [γ]) = ϕe0 ([γ]). We have

ϕe0 ([δ] · [γ]) = ϕe0 ([δ ∗ γ]) = (δ]

∗ γ)e0 (1) = (δee0 ∗ γ
eδ̃e (1) )(1)
0

=γ
eδ̃e (1) ( 1 ) .
0

By part (a), since [δ] ∈ H, we have that δee0 (1) = e0 . Thus, γ

eδ̃e = (1) ( 1 )
0
ee0 (1) = ϕe0 ([γ]), so ϕe0 is well defined. From Theorem 4.1.13 we know
γ
that ϕe0 is onto, so it remains to show that it is injective.
Suppose that ϕe0 ([γ1 ]) = ϕe0 ([γ2 ]). By definition, this means that
(γ
f1 )e0 (1) = (γf2 )e0 (1). Thus, (γ f1 )e0 ∗ (γ
e2 )e0 is a loop in E based at
e0 , which in turn is a lift of γ1 ∗ γ2 . By (a), [γ1 ∗ γ2 ] ∈ H. Finally,
[γ1 ] = [γ1 ∗ γ2 ∗ γ2 ] = [γ1 ∗ γ2 ] · [γ2 ]. Since [γ1 ∗ γ2 ] ∈ H, the cosets of
γ1 and γ2 coincide. Thus, ϕe0 is injective.
 covering spaces 51

Theorem 4.1.19 (Lifting Lemma). Let E, B, Y be path-connected and locally

path-connected spaces.1 Let p : E → B be a cover, b0 ∈ B, e0 ∈ p−1 (b0 ), 1
Recall that a topological space X is lo-
and f : Y → B a continuous map such that f (y0 ) = b0 . Then there exists cally path-connected if, for all x ∈ X and
for all neighborhoods Ux of x, there ex-
a lift fe: Y → E of f (i.e., p ◦ fe = f ) such that fe(y0 ) = e0 if and only if ists a neighborhood Vx which is path-
f ∗ π1 (Y, y0 ) ⊂ p∗ π1 ( E, e0 ). connected, contains x, and is contained
in Ux .
( E, e0 )
∃ fe
p
f
(Y, y0 ) ( B, b0 )

Proof. The “=⇒” direction is clear from p ◦ fe = f .

For the “⇐=” direction, let y ∈ Y, and we need to define fe(y). Let
α be a path in Y from y0 to y. Then f ◦ α is a path in B starting at b0 .
Define fe(y) := ( ] f ◦ α)e0 (1). We have ( p ◦ fe)(y) = p ◦ ( ] f ◦ α ) e0 ( 1 ) =
( f ◦ α)(1) = f (y). Thus, fe is a lift of f . It is also immediate that
fe(y0 ) = e0 .
Next we need to show fe is well defined (i.e., independent of α).
If β is another path in Y from y0 to y, then α ∗ β ∈ π1 (Y, y0 ), so
f ◦ (α ∗ β) ∈ f ∗ π1 (Y, y0 ) ⊂ p∗ π1 ( E, e0 ). It follows from Theorem 4.1.18
that ( f ◦^
(α ∗ β)) is a loop at e . Note that f ◦ (α ∗ β) = ( f ◦ α) ∗ ( f ◦ β).
e0 0
Then we have

( f ◦^ f ◦ α ) e0 ∗ ( ]
(α ∗ β))e0 = ( ] f ◦ β)(]
f ◦α)
f ◦ α)e0 ∗ (^
= (] f ◦ β)(]
f ◦α)
e0 ( 1 ) e0 ( 1 )

f ◦ α)e0 ∗ (^
= (^ f ◦ β)(]
f ◦α) e0 ( 1 )

This means that ( ] f ◦ α ) e0 ( 1 ) = ( ]

f ◦ β)e0 (1), hence the definition of fe
does not depend on the choice of α.
It remains to show that fe is continuous. Let y ∈ Y, and let U be
a path connected evenly covered neighborhood of f (y) ∈ B, which
exists by the locally path-connected assumption. Let V be the slice
in p−1 (U ) which contains fe(y). By the continuity of f , there is some
path-connected neighborhood of y, say W, in Y such that f (W ) ⊂ U.
Then fe(W ) ⊆ V (since fe(W ) is path-connected and contains fe(y)) and
fe|W = ( p|V )−1 ◦ f |W . Hence fe is continuous on W. Continuity on Y
follows from local continuity just proved.

Corollary 4.1.20. If Y is simply connected, then such a lift always exists.

Proposition 4.1.21 (Lift uniqueness). If Y is connected and fe1 , fe2 : Y → E
are two lifts as in the previous theorem (i.e., coinciding at y0 ∈ Y), then
fe1 = fe2 .

Proof. Let A = {y ∈ Y | fe1 (y) = fe2 (y)} ̸= ∅. We will show A = Y by

proving that A is both open and closed. Let y ∈ Y, and let U be an
52 algebraic topology

evenly covered neighborhood of f (y) in B. Then we have p−1 (U ) =

⊔α U e α such that p| e : U
Uα
e α → U is a homeomorphism. Let U e1, U
e 2 be
the slices containing fe1 (y) and fe2 (y), respectively. Since the fei are
continuous, there is a neighborhood N of y such that fe1 ( N ) ⊂ U e 1 and
fe2 ( N ) ⊂ Ue 2 . If y ∈/ A, we have fe1 (y) ̸= fe2 (y), hence Ue 1 ̸= U e 2 , so
U1 ∩ U2 = ∅. This means that f 1 ̸= f 2 on N, so A is closed. On the
e e e e
other hand, if fe1 (y) = fe2 (y), then U
e1 = Ue 2 , which implies that fe1 = fe2
on N (since p f 1 = p f 2 = f , and p is injective on U
e e e1 = U
e 2 ). Thus, A is
open.

4.2 Covering transformations

In this section, all spaces are assumed path-connected and locally path-
connected.

Definition 4.2.1. If p : E → B, p′ : E′ → B are coverings, a homomorphism

of coverings h : ( E, p) → ( E′ , p′ ) is a continuous map h : E → E′ such that
p′ ◦ h = p.

Definition 4.2.2. An isomorphism (or equivalence) of coverings is a homo-

morphism of coverings which is also a homeomorphism.

Theorem 4.2.3. Let p : E → B, p′ : E′ → B be coverings of B with p(e0 ) =

p′ (e0′ ) = b0 ∈ B. Then there is an equivalence of coverings h : E → E′ ,
h(e0 ) = e0′ if and only if H = p∗ (π1 ( E, e0 )) and H ′ = p′∗ (π1 ( E′ , e0′ )) are
equal as subgroups:
( E′ , e0′ )
∃h
p′
p
( E, e0 ) ( B, b0 )

Proof. “=⇒”: If h : E → E′ is an equivalence with h(e0 ) = e0′ , then

h∗ (π1 ( E, e0 )) = π1 ( E′ , e0′ ). Apply p′∗ and, using p′ ◦ h = p, we get
H = H′.
“⇐=”: Assume H = H ′ . Since H ⊂ H ′ , we get by the lifting lemma
(Theorem 4.1.19) that there exists h : ( E, e0 ) → ( E′ , e0 ) with h(e0 ) = e0′ ,
p′ ◦ h = p. Reversing the roles of p and p′ , we get that H ′ ⊂ H
implies the existence of a lift k : ( E′ , e0′ ) → ( E, e0 ) of p′ with p ◦ k = p′ ,
k (e0′ ) = e0 . Consider the diagram:

k◦h ( E, e0 )
id E p
p
( E, e0 ) ( B, b0 )
 covering spaces 53

Since p ◦ (k ◦ h) = ( p ◦ k) ◦ h = p′ ◦ h = p, we have that k ◦ h and id E

are lifts of p that agree at e0 . So, by the uniqueness of lifts, k ◦ h = id E .
By similar reasoning on p′ , we get that h ◦ k = id E′ .

Proposition 4.2.4. If h, k : ( E, p) → ( E′ , p′ ) are homomorphisms of cover-

ings p, p′ of B such that h(e) = k(e) for some e ∈ E, then h = k.

Proof. Consider the set A = {e ∈ E | h(e) = k(e)}. It is easy to see that

A is both open and closed, hence it is all of E.

Remark 4.2.5. If E = E′ and p = p′ , an equivalence of p interchanges

points in the fiber over each b ∈ B. Such a self-equivalence is called an
automorphism of ( E, p), or a deck transformation.

Definition 4.2.6. The deck transformations form a group under composition

of maps, called the deck group of ( E, p), and denoted D( E, p).

An immediate consequence of Theorem 4.2.3 is the following:

Corollary 4.2.7. If p : E → B is a covering and p(e1 ) = p(e2 ), then there is

h ∈ D( E, p) with h(e1 ) = e2 if and only if p∗ π1 ( E, e1 ) = p∗ π1 ( E, e2 ).

Moreover, Proposition 4.2.4 implies the following:

Corollary 4.2.8. If h ∈ D( E, p) so that h( x ) = x for some x ∈ E, then

h = idE .

We can now generalize Theorem 4.2.3 as follows:

Theorem 4.2.9 (Main Theorem). Let p : E → B and p′ : E′ → B be

covering maps. Let p(e0 ) = p′ (e0′ ) = b0 . The covering maps p and p′
are equivalent if and only if the subgroups H = p∗ π1 ( E, e0 ) and H ′ =
p′∗ π1 ( E′ , e0′ ) are conjugate in π1 ( B, b0 ).

Proof. “=⇒”: Assume we have an equivalence h : E → E′ , and let

h(e0 ) = e0′′ . By the previous theorem, H = p∗ π1 ( E, e0 ) equals H ′′ =
−1
p′∗ π1 ( E′ , e0′′ ). By changing e0′′ to any e0′ ∈ p′ (b0 ) we know that H ′′ is
conjugate to H ′ = p′∗ π1 ( E′ , e0′ ). So H and H ′ are conjugate.

“⇐=”: If H = p∗ π1 ( E, e0 ) and H ′ = p′∗ π1 ( E′ , e0′ ) are conjugate, we

need the following

Lemma 4.2.10. Let p : E → B be a covering, p(e0 ) = b0 , and H =

p∗ π1 ( E, e0 ). Given any subgroup K ⊂ π1 ( B, b0 ) conjugate to H, there
is an e1 ∈ p−1 (b0 ) such that K = H1 = p∗ π1 ( E, e1 ).

Proof. As K and H are conjugate in π1 ( B, b0 ), there is a loop α at b0

in B such that H = [α] · K · [α]−1 . Let eαe0 be a lift to E of α under p,
αe0 (1). Then H = [ p ◦ e
starting at e0 , let e1 = e αe0 ]−1 . So
αe0 ] · H1 · [ p ◦ e
K = H1 since p ◦ e αe0 = α.
54 algebraic topology

Using Lemma 4.2.10, there is e1 ∈ p−1 (b0 ) such that p′∗ π1 ( E′ , e0′ ) =
H ′ = p∗ π1 ( E, e1 ). By the lifting property (Theorem 4.1.19), there is an
equivalence h : E → E′ , thus finishing the proof of Theorem 4.2.9.

Definition 4.2.11. A covering p : E → B is called a universal covering map

if E is simply connected. In this case, E is called a universal cover of B.

In view of Theorem 4.2.9, we get the following consequence.

Corollary 4.2.12. If a universal cover of B exists, it is unique up to equivalence

of coverings, since the conjugacy class of the trivial subgroup in any group
has only one element.

Example 4.2.13. Let B be the Möbius band, with π1 ( B) ∼ = Z. Conjugacy

classes of subgroups of Z are given by nZ for n ∈ N. An even integer n
yields an n-fold covering of B by the cylinder S1 × I with (z, t) 7→ (zn , t).
An odd n yields an n-fold covering of B by the Möbius band under the
same map.

4.3 Universal Covering Spaces

In this section we investigate when a path connected, locally path

connected space B has a universal cover.

Definition 4.3.1. A topological space B is called semi-locally simply connected

if, for any b ∈ B, there is a neighborhood Ub of b such that the inclusion
i : Ub ,→ B induces a trivial homomorphism i∗ : π1 (Ub , b) → π1 ( B, b).

Example 4.3.2. If B is simply-connected, then B is semi-locally simply

connected.

In this section, we discuss the following:

Theorem 4.3.3. A topological space B has a universal cover if and only if B

is path connected, locally path connected and semi-locally simply connected.

The proof of the implication “=⇒” of Theorem 4.3.3 follows from

the following.

Proposition 4.3.4. Let p : E → B be a covering map, p(e0 ) = b0 . Assume

E is simply-connected. Then there exists a neighborhood U of b0 such that
the inclusion i : U ,→ B induces a trivial homomorphism i∗ : π1 (U, b0 ) →
π1 ( B, b0 ).

Proof. Let U be an evenly covered neighborhood of b0 and let U e be

− 1
the slice of p (U ) containing e0 . Let f be a loop in U at b0 . Since
p|Ue : Ũ → U is a homeomorphism, f lifts to a loop f˜ in Ue at e0 . Since E
˜
is simply-connected, there is a path homotopy F̃ from f to the constant
loop in E at e0 . Then p ◦ F̃ is a homotopy in B from p ◦ f˜ = f to the
constant loop in B at b0 .
 covering spaces 55

The proof of the converse implication “⇐=” of Theorem 4.3.3 follows

from the following.

Theorem 4.3.5. Let B be path connected, locally path connected and semi-
locally simply connected. Let b0 ∈ B and H ⊂ π1 ( B, b0 ) a subgroup.
Then there is a covering p : E → B and a point e0 ∈ p−1 (b0 ) such that
p∗ π1 ( E, e0 ) = H.

Sketch of proof. Let P be the set of all paths in B starting at b0 . Define

an equivalence relation on P by α ∼ β if α(1) = β(1) and [α ∗ β̄] ∈ H.
Let α# be the equivalence class of α ∈ P. Consider the set

E = { α# | α ∈ P }.

Define p : E → B by p(α# ) = α(1). Then p is surjective since B is

path-connected. Furthermore, one can define a topology on E so that p
becomes a covering map. (Details are left as an exercise.)

Example 4.3.6. The infinite earring has no universal cover, since it is not
semi-locally simply connected. The infinite earring is the space
[
X= Cn ,
n ≥1

1
where Cn is the circle of center (1/n, 0) and radius n. We claim that

Figure 4.1: Infinite earring

if U is any neighborhood of 0 ∈ X, then i∗ : π1 (U, 0) → π1 ( X, 0) is

nontrivial. Indeed, given n, there is a retraction r : X → Cn , defined by
mapping each circle Ci (i ̸= n) to 0, and as the identity on Cn . Choose
n large enough so that Cn ⊂ U, and consider the following diagram
with the induced homomorphisms on fundamental groups.

j j∗
Cn X π1 (Cn , 0) π1 ( X, 0)
k k∗
i i∗

U π1 (U, 0)

Since r∗ ◦ j∗ = idZ , we get that j∗ is injective. From j∗ = i∗ ◦ k ∗ , we

deduce that i∗ cannot be trivial.
56 algebraic topology

4.4 Group actions and covering maps

In this section we study in more depth the relation between the fun-
damental group of the base of a covering on the one hand, and deck
transformations of the covering on the other hand. All spaces are again
path connected and locally path connected.

Theorem 4.4.1. If p : E → B is a covering with p(e0 ) = b0 and

H = p∗ π1 ( E, e0 ) ⊂ π1 ( B, b0 ),

then
D( E, p) ∼
= N ( H )/H,
where N ( H ) = { g ∈ π1 ( B, b0 ) | gHg−1 = H } is the normalizer of H.
(Recall that N ( H ) is the largest subgroup of G which contains H as a normal
subgroup.)

Proof. Recall that

ϕ : H \π1 ( B, b0 ) → F := p−1 (b0 )

is a bijection. Define a map

ψ : D( E, p) → F, ψ(h) = h(e0 ).

Since each h ∈ D( E, p) is uniquely determined by its value on e0 , it

follows that ψ is injective. The assertion follows from the following two
facts:

(i) Im(ψ) = ϕ ( N ( H )/H ).

(ii) ϕ−1 ◦ ψ : D( E, p) → N ( H )/H is a group isomorphism.

For (i), recall that ϕ is defined as follows: given a loop α in B

at b0 , we set ϕ([α]) = e1 , where e1 = e αe0 (1) and e
αe0 is the lift of
α to E starting at e0 . The assertion in (i) is then equivalent to the
following statement: there is h ∈ D( E, p) with h(e0 ) = e1 if and only if
[α] ∈ N ( H ). By the Lifting Lemma (Theorem 4.1.19), such h exists if and
only if H = H ′ := p∗ (π1 ( E, e1 )). Moreover, we have [α] · H ′ · [α]−1 = H.
Hence h exists if and only if [α] · H · [α]−1 = H, which is the same as
[ α ] ∈ N ( H ).
For (ii), we only need to show that ϕ−1 ◦ ψ is a homomorphism, as
it is already bijective. So let h, k : E → E be covering transformations,
with h(e0 ) = e1 and k(e0 ) = e2 . Then ψ(h) = e1 and ψ(k) = e2 . Let γ, δ
be paths in E from e0 to e1 and e2 , respectively. Then, if α = p ◦ γ and
β = p ◦ δ, we get

ψ([α] H ) = e1 , ψ([ β] H ) = e2 .
 covering spaces 57

Now let e3 = h(k (e0 )), so that ψ(h ◦ k) = e3 . We need to show that

ψ([α ∗ β] H ) = e3 .

Since δ is a path from e0 to e2 , then h ◦ δ is a path from h(e0 ) = e1 to

h(e2 ) = h(k(e0 )) = e3 . So γ ∗ (h ◦ δ) is a path from e0 to e3 . Also note
that
p ◦ (γ ∗ (h ◦ δ)) = ( p ◦ γ) ∗ ( p ◦ h ◦ δ) = α ∗ β,

so γ ∗ (h ◦ δ) is a lift of α ∗ β. Thus ψ([α ∗ β] H ) = e3 , as desired.

Corollary 4.4.2. If π1 ( E, e) = 0, then D( E, p) = π1 ( B, p(e)).

Definition 4.4.3. A covering p : E → B is called regular if p∗ π1 ( E, e) is a

normal subgroup of π1 ( B, p(e)), for any e ∈ E.

Example 4.4.4. If π1 ( B) is abelian, any covering of B is regular.

We leave the following as an exercise.

Proposition 4.4.5. A covering p : E → B is regular if and only if the deck

group D( E, p) acts transitively on the fibers of p, that is, for all e1 , e2 ∈ E
with p(e1 ) = p(e2 ) = b ∈ B, there exists h ∈ D( E, p) such that h(e1 ) = e2 .

Corollary 4.4.6. If p : E → B is regular, then

D( E, p) ∼
= π1 ( B, p(e))/p∗ π1 ( E, e).

Remark 4.4.7. The universal cover p : E → B of B is regular (since the

trivial subgroup is normal) and D( E, p) ∼
= π1 ( B) acts transitively on
each fiber of p. Hence E/D( E, p) = E/π1 ( B) ∼
= B.

Example 4.4.8. Let p : R → S1 be the covering t 7→ exp(2πit). We have

that
D(R, p) = {t 7→ t + n | n ∈ Z} ∼ = Z.
Example 4.4.9. Consider the covering R2 → T 2 defined as p × p and p
as in the previous example. The deck group is in this case

{(t, s) 7→ (t + n, s + m) | n, m ∈ Z} ∼
= Z2 .

Example 4.4.10. Let p : S2 → RP2 be the covering defined by the

antipodal identification. Then D(R, p) = {±id} since π1 (RP2 ) ∼
= Z/2
is abelian.

Let X be a topological space, and G a subgroup of Homeo( X ),

the group of homeomorphisms of X. Then G acts on X, i.e., there
is a continuous map G × X → X given by ( g, x ) 7→ g · x := g( x ).
Let [ x ] = { gx | g ∈ G } be the orbit of x. Consider the orbit space
X/G := {[ x ] | x ∈ X }.
58 algebraic topology

Example 4.4.11. Consider the cylinder X = S1 × [0, 1]. Let h, k : X → X

be homeomorphisms defined by h( x, t) = (− x, t), k( x, t) = (− x, 1 − t).
Obviously, h, k are elements of order two in the group Homeo( X ). Let
G1 = ⟨h⟩ and G2 = ⟨k⟩. It is easy to see that X/G1 = X, while X/G2 is
a Möbius band.

Definition 4.4.12. Say that G acts freely on X if whenever g · x = x for some

x ∈ X, we have g = eG , the identity element of G.

Definition 4.4.13. The group G acts properly discontinuous on X if for any

x ∈ X, there is an open neighborhood Ux of x such that gUx ∩ Ux = ∅ for
all g ̸= eG . (Hence, gUx ∩ hUx = ∅ if h ̸= g ∈ G.)

The following result is left as an exercise.

Proposition 4.4.14. If X is Hausdorff and G is a finite group of homeomor-

phisms of X acting freely on X, the action of G is properly discontinuous.

The main result of this section is the following.

Theorem 4.4.15. Let X be a path-connected, locally path-connected topological

space, and G ≤ Homeo( X ). Then π : X → X/G is a covering if and only
if G acts properly discontinuous on X. Moreover, if this is the case, the deck
group D( X, π ) of the covering is isomorphic to G and the covering is regular.

Proof. We first show that π is an open map. Let U ⊂ X be open and

show that π (U ) is open in X/G. Since X/G has the quotient topology,
π (U ) is open in X/G if and only if π −1 (π (U )) is open in X. By the
definition of π, we have

π −1 (π (U )) =
[
gU
g∈ G

Since each g ∈ G is a homeomorphism of X, gU ⊂ X is open for every

g, so π −1 (π (U )) is open in X.
We now prove the “⇐=” direction. Assume G acts properly discon-
tinuous (p.d.) on X, and show that π is a covering map.
For x ∈ X, let U be a neighborhood of x such that gU ∩ U = ∅ for
all g ̸= eG . We claim that π (U ) is an evenly covered neighborhood of
[ x ] ∈ X/G. Indeed,

• π −1 (π (U )) = gU, and all { gU } g∈G are disjoint open sets in X.

S
g∈ G

• π | gU : gU → π (U ) is a homeomorphism. Indeed, π | gU is continu-

ous, open, and it is clearly onto. Moreover, if π ( gx1 ) = π ( gx2 ) for
x1 , x2 ∈ U, then there is g′ ∈ G with g′ gx1 = gx2 , or g−1 g′ gx1 = x2 .
But since hU ∩ U = ∅ for all h ̸= eG , one must have that g−1 g′ g = eG ,
or g′ = eG . In particular, gx1 = gx2 , thus proving the injectivity of
π | gU .
 covering spaces 59

To prove the “=⇒” direction, assume that π is a covering map, and

show that the action of G on X is p.d.
Let x ∈ X be arbitrary, and let Vx be a neighborhood of [ x ] = π ( x )
which is evenly covered by π. In particular, π −1 (Vx ) = α Uα , with
F

π |Uα : Uα → Vx a homeomorphism, for any α. Let Uα be the “slice”

containing x. We claim that for any g ̸= eG , we have gUα ∩ Uα = ∅. If
not, there is y ∈ gUα ∩ Uα , hence y, g−1 y ∈ Uα are distinct (note that g
and its inverse are covering transformations). But since [y] = [ g−1 y],
this contradicts the injectivity of π |Uα . Hence G acts p.d. on X.
Finally, we show that if π is a covering map, then G is its deck group
and π is regular.
First, any g ∈ G is a homeomorphism of X and π ◦ g = π, so
G ⊆ D( X, π ). Conversely, if h ∈ D( X, π ) with h( x1 ) = x2 , then since
π ◦ h = π we get that π ( x1 ) = π ( x2 ). In particular, there is g ∈ G such
that gx1 = x2 . Since g is also a covering transformation and h and g
agree on x1 , we have by uniqueness that h = g ∈ G.
The covering π is regular since G acts transitively on the fibers
of π. Indeed, if x1 , x2 ∈ π −1 ([ x ]), then [ x1 ] = [ x2 ], hence there is
g ∈ G = D( X, π ) with gx1 = x2 .

Using the above theorem and Corollary 4.4.2, we get the following.

Corollary 4.4.16. If X is simply connected and G acts properly discontinu-

ously on X, then π1 ( X/G ) ∼
= G.

Example 4.4.17. If G is finite and acts freely on a Hausdorff space X,

then we know by Proposition 4.4.14 that G acts properly discontinuous,
so π : X → X/G is a covering with D( X, π ) ∼ = G.

The following two results are left as exercises.

Proposition 4.4.18. If p : E → B is a cover (not necessarily regular), then

D( E, p) acts properly discontinuous on E.

Proposition 4.4.19. Any regular cover of B is of the form E/G, where E is

the universal cover of B and G acts properly discontinuous on E.

We conclude this chapter with some computations that follow easily

from the above results.

Example 4.4.20. The action of Z2 on R2 by translation (Example 4.4.9)

is properly discontinuous. So, since R2 /Z2 ∼
= T 2 , we have that R2 is a
2 2 ∼
universal cover of T and π1 ( T ) = Z .
2

Example 4.4.21. The action of Z on R2 by n ◦ ( x, y) = (n + x, y) is

also properly discontinuous. The quotient space, R2 /Z is an infinite
cylinder, S1 × R. Thus we have that R2 is the universal cover of S1 × R,
and the fundamental groups of the cylinder is Z.
 60 algebraic topology

Example 4.4.22. The action of Z on R2 by n ◦ ( x, y) = (n + x, (−1)n y) is

again properly discontinuous. The quotient space R2 /Z is the Möbius
band, which makes R2 the universal cover of the Möbius band and the
fundamental group of the Möbius band is Z.

Example 4.4.23. This example focuses on spaces called lens spaces.

Regard S2n+1 as a subspace of Cn+1 in the usual way. Let Z/q ⊂
C∗ be the q-th roots of unity. Define an action of Z/q on S2n+1 by
ξ ◦ (z1 , . . . , zn+1 ) = (ξz1 , ξ r2 z2 , . . . , ξ rn+1 zn+1 ). This action is free if and
only if gcd(ri , q) = 1, for all i. Assume this is the case, and define

L( p; r2 , r3 , . . . , rn+1 ) = S2n+1 /Z/q.

Since the action of Z/q is free, we have that

π : S2n+1 → L( p; r2 , . . . , rn+1 )

is a covering map with D(S2n+1 , π ) ∼ = Z/q. Since, for n ≥ 1, S2n+1 is

simply-connected, it is a universal cover of L( p; r2 , r3 , . . . , rn+1 ), so in
particular π1 ( L( p; r2 , . . . , rn+1 )) ∼
= Z/q.

Exercises
1. Show that the map p : S1 → S1 , p(z) = zn is a covering. (Here we
represent S1 as the set of complex numbers z of absolute value 1.)

2. Let p : E → B be a covering map, with E path connected. Show that

if B is simply-connected, then p is a homeomorphism.

3.

(i) Show that if n > 1 then any continuous map f : Sn → S1 is nullho-

motopic.

(ii) Show that any continuous map f : RP2 → S1 is nullhomotopic.

4.

(i) Classify all coverings of the Möbius strip up to equivalence.

(ii) Show that every covering of the Möbius strip is homeomorphic to

either R2 , S1 × R or the Möbius strip itself.

5.

(i) Show that the torus T 2 is a two-fold cover of the Klein bottle.

(ii) Is it possible to realize the Klein bottle as a two-fold cover of itself?

 covering spaces 61

(iii) Find the universal cover of the Klein bottle.

6. Let p : E → B be a covering map with E simply-connected. Show that

given any covering map r : Y → B, there is a covering map q : E → Y
such that r ◦ q = p.

7. Show that if G is a finite group with a fixed-point free action on a

Hausdorff space X, the quotient map p : X → X/G is a covering.

8. Let Z6 act on S3 = {(z, w) ∈ C2 , |z|2 + |w|2 = 1} via (z, w) 7→

(ϵz, ϵw) , where ϵ is a primitive sixth root of unity. Denote by L the
quotient space S3 /Z6 .

(i) What is the fundamental group of L?

(ii) Describe all coverings of L.

(iii) Show that any continuous map L → S1 is nullhomotopic.

 homology 63

5
Homology

Homology of a topological space X yields a collection of topological

invariants of X called homology groups which, roughly speaking, define
and categorize “holes” in a manifold. We first define singular homology
and study its properties. We then introduce CW complexes and define
their cellular homology, and show that in this context singular homol-
ogy and cellular homology coincide. Basic knowledge of homological
algebra will be assumed throughout this section.

5.1 Singular Homology

Definition 5.1.1. The standard n-simplex is the set

 n 
∆n := (t0 , . . . , tn ) ∈ Rn+1 | ∑ ti = 1; ti ≥ 0, ∀i ,
i =0

i.e., the convex span of the standard basis of Rn+1 .

Definition 5.1.2. An n-simplex is the convex span in Rm of n + 1 points,
v0 , . . . , vn that do not lie in a hyperplane of dimension less than n (i.e.,
v1 − v0 , . . . , vn − v0 are linearly independent).
Given n + 1 vectors v0 , . . . , vn as in the definition of the n-simplex,
we write [v0 , . . . , vn ] for the n-simplex that they generate, and we call
the vi ’s the vertices.
Note that there is a canonical linear homeomorphism from ∆n to any
n-simplex [v0 , . . . , vn ] defined by:
n
∆n −→ [v0 , . . . , vn ], (t0 , . . . , tn ) 7→ ∑ ti vi .
i =0

If we delete one vertex from the n-simplex [v0 , . . . , vn ], the remaining

n vertices span a (n − 1)-simplex, called a face of [v0 , . . . , vn ]. The union
of all faces is called the boundary of [v0 , . . . , vn ]. We denote faces by
[v0 , . . . , vbi , . . . , vn ], i = 0, . . . , n, where vbi indicates that vi is a deleted
vertex.
64 algebraic topology

Definition 5.1.3. A singular n-simplex in a topological space X is a continu-

ous map σ : ∆n −→ X.

We use the word “singular” because the image of such a map can
have “singularities”.
Let Cn ( X ) be the free abelian group with basis the singular n-
simplices in X, i.e.,
 
Cn ( X ) = ∑ i i i
n σ | n ∈ Z, σi : ∆ n
→ X continuous ,
i

where each formal sum ∑i ni σi is finite, i.e., all but finitely many ni are
zero. We call an element of Cn ( X ) an n-chain in X.
We define boundary maps

∂n : Cn ( X ) → Cn−1 ( X )

as follows. Since Cn ( X ) is the free abelian group on the singular

n-simplices of X, it suffices to define the map ∂n on the singular n-
simplices, and then extend it by linearity to all of Cn ( X ). If σ : ∆n → X
is such an n-simplex, we set
n
∂n (σ ) := ∑ (−1)i σ|[v0 ,...,vbi ,...vn ] .
i =0

A crucial lemma, whose proof is by a direct calculation using the

definition, is the following.

Lemma 5.1.4. For every n, we have that ∂n ◦ ∂n+1 = 0.

We often abbreviate the above fact as ∂2 = 0.

Definition 5.1.5. We call C• ( X ) = {Cn ( X ), ∂n }n∈N the singular chain

complex of X.

Note that both Im(∂n+1 ) and ker(∂n ) are subgroups of the abelian
group Cn ( X ). The above lemma yields that Im(∂n+1 ) is a subgroup of
ker(∂n ). Hence we can make the following.

Definition 5.1.6. The n-th singular homology group of X is defined by:

Hn ( X ) := ker(∂n )/ Im(∂n+1 ).

It is clear by definition that Hn ( X ) is a homeomorphism invariant.

Moreover, as we will see later, homology is in fact a homotopy invariant.

Definition 5.1.7. We introduce the following notations:

(i) Zn := ker(∂n ) is the group of n-cycles.

(ii) Bn := Im(∂n+1 ) is the group of n-boundaries.

 homology 65

We next prove some immediate consequences of the definition of

homology.

Proposition 5.1.8. Let x0 be a point. Then,


Z, n = 0
Hn ( x0 ) =
0, n > 0.

Proof. For every n, there is a unique map σn : ∆n → x0 . So Cn ( x0 ) is

the free abelian group generated by σn , hence it is isomorphic to Z.
Now,

n 0, n is odd
∂n (σn ) = ∑ (−1)i σn−1 =
i =0
 σn−1 , n is even, n ̸= 0.

So we get the chain complex:

∼
= 0 ∼
= 0
→Z−
··· − →Z−
→Z−
→Z→0

Taking homology of this complex yields the desired result.

Proposition 5.1.9. Suppose X is a space and { Xα }α∈ A are the path connected
components of X. Then, Hn ( X ) ∼
M
= Hn ( Xα ).
α∈ A

Proof. Since ∆n is path connected and an n-simplex σ : ∆n → X is a

continuous map, we have that Im(σ) ⊆ Xα for some α. Therefore, we
get a decomposition

Cn ( X ) ∼
M
= Cn ( Xα ).
α

The boundary maps preserve this decomposition, i.e., ∂(Cn ( Xα )) ⊆

Cn−1 ( Xα ). Hence ker(∂n ) and Im(∂n+1 ) split similarly as direct sums
and the result follows.

Proposition 5.1.10. If X ̸= ∅ is path connected, then H0 ( X ) ∼

= Z. More
∼
generally, H0 ( X ) = α Z, where X = α Xα is the union of X into its path
L S

connected components.

Proof. From
1∂ ∂0
C1 ( X ) −
→ C0 ( X ) −
→0
and ∂0 = 0, we get that H0 ( X ) ∼
= C0 ( X )/ Im(∂1 ). Define the augmenta-
tion map

ϵ : C0 ( X ) −→ Z
∑ ni σi 7→ ∑ ni
i i
66 algebraic topology

The map ϵ is clearly onto. We claim that if X is path connected then

ker(ϵ) = Im(∂1 ). This will then imply that H0 ( X ) ∼
= Z.
Let σ : ∆ → X be a singular 1-simplex. Then,
1

ϵ(∂1 (σ )) = ϵ(σ[v1 ] − σ[v0 ] ) = 1 − 1 = 0.

Therefore, Im(∂1 ) ⊆ ker(ϵ). Next, suppose that ϵ(∑i ni σi ) = 0, i.e.,

∑i ni = 0. Here, the σi ’s are singular 0-simplices, i.e., points of X. Let
x0 be a basepoint in X and let σ0 be the corresponding 0-simplex with
image x0 = σ0 (v0 ). Since X is path connected, for every i, there exists a
continuous path τi : I → X from x0 to σi (v0 ). The unit interval I is ∆1 .
So, we can regard τi ∈ C1 ( X ) and ∂1 (τi ) = σi − σ0 . Hence,
  !
∂1 ∑ ni τi = ∑ ni σi − ∑ ni σ0 = ∑ ni σi − ∑ ni σ0 = ∑ ni σi ,
i i i i i i

which shows that ker(ϵ) ⊆ Im(∂1 ).

Definition 5.1.11. The reduced homology groups of X, H e n ( X ), are the

homology groups of the augmented chain complex of X defined as:
∂ ∂
→ Z → 0,
2 1 ϵ
· · · → C2 ( X ) −
→ C1 ( X ) −
→ C0 ( X ) −

where ϵ is the augmentation map defined in Proposition 5.1.10 as ϵ(∑i ni σi ) =

∑i ni .
The above complex is a chain complex since, as shown above, we
have ϵ ◦ ∂1 = 0. Moreover, this formula also shows that ϵ induces an
onto map C0 ( X )/ Im(∂1 ) = H0 ( X ) ↠ Z with kernel H
e0 ( X ). Therefore,

H0 ( X ) ∼
=He0 ( X ) ⊕ Z

and it is clear that for n ≥ 1, we have that Hn ( X ) ∼= He n ( X ). So one

does not get any new information from the reduced homology groups,
but they allow us to state results in a cleaner way. For example, if x0 is
a point, then the previous proposition can be restated as H e n ( x0 ) = 0
for all n.

5.2 Homotopy Invariance

In this section we show that the homology groups are homotopy invari-
ants.
Let f : X → Y be a continuous map. Then, we have an induced
homomorphism
f # : Cn ( X ) → Cn (Y )
defined by f # (∑ ni σi ) = ∑ ni ( f ◦ σi ).

Lemma 5.2.1. f # is a chain map, i.e., f # ∂n = ∂n f # .

 homology 67

Proof. It suffices to show that this equality holds for a singular n-

simplex σ.
 
f # (∂n (σ )) = f # ∑(−1) σ|[v0 ,...,vbi ,...,vn ]
i
i
= ∑(−1)i f ◦ σ|[v0 ,...,vbi ,...,vn ]
i
= ∂n ( f ◦ σ )
= ∂n f # (σ )

We therefore get the following diagram with commutative squares:

∂ n +1 n ∂ ∂ n −1
. . . −−−−→ Cn+1 ( X ) −−−−→ Cn ( X ) −−−− → Cn−1 ( X ) −−−−→ . . .
  
f f f
y# y# y#
∂ n +1 n ∂ ∂ n −1
. . . −−−−→ Cn+1 (Y ) −−−−→ Cn (Y ) −−−− → Cn−1 (Y ) −−−−→ . . .
Corollary 5.2.2. f # takes n-cycles to n-cycles.

Proof. If ∂n (σ ) = 0, then ∂n ( f # (σ )) = f # (∂n (σ )) = f # (0) = 0.

Corollary 5.2.3. f # takes boundaries to boundaries.

Proof. Suppose σ = ∂n+1 (η ). Then

f # (σ ) = f # (∂n+1 (η )) = ∂n+1 ( f # (η )).

Therefore, we get the following corollary.

Corollary 5.2.4. The map f : X → Y induces a homomorphism f ∗ : Hn ( X ) →

Hn (Y ) for every n.

More generally, a chain map between chain complexes induces ho-

momorphisms between the homology groups of the two complexes.
From the properties of the map f # , we get the following proposition.
g f
Proposition 5.2.5.(a) If X −
→Y−
→ Z are maps, then ( f ◦ g)∗ = f ∗ ◦ g∗ .

(b) (id X )∗ = id H∗ (X )

We are ready to state our main theorem.

Theorem 5.2.6. If f , g : X → Y are homotopic maps, then they induce the

same homology homomorphisms f ∗ = g∗ : Hn ( X ) → Hn (Y ) for every n.

Before proving the theorem, let us state some important conse-

quences (deduced using Proposition 5.2.5):
68 algebraic topology

Corollary 5.2.7. If f : X → Y is a homotopy equivalence, then f ∗ : Hn ( X ) →

Hn (Y ) are isomorphisms for every n.

e n ( X ) = 0 for every n.
Corollary 5.2.8. If X is contractible, then H

Proof of Theorem 5.2.6. Let F : X × I → Y be the homotopy between f

and g. We will define an operator P : Cn ( X ) → Cn+1 (Y ), called a prism
operator, such that
∂P + P∂ = g# − f # (⋆)

Once defined, this will then show that g# and f # have the same effect on
homology. For, if α ∈ Cn ( X ) is a cycle, then by (⋆), we get that g# (α) −
f # (α) = ∂P(α) + P∂(α) = ∂P(α). Since f # and g# differ by a boundary,
they are homologous, so when quotient out by the boundaries, we get
that f ∗ ([α]) = g∗ ([α]) in homology.
It suffices to define P(σ ), for σ : ∆n → X a singular n-simplex, and
then we can extend P by linearity. We have the following maps:

(σ,id) F
∆n × I −−−→ X × I −
→Y

In order to define P(σ ), the idea is to divide ∆n × I into a linear

combination of (n + 1)-simplices.
For example, the following picture shows how to divide ∆1 × I into
two 2-simplices. If we let [v0 , v1 ] be the simplex ∆1 × {0}, and we let
[w0 , w1 ] be the simplex at ∆1 × {1}, then we can write ∆1 × I as the
union [v0 , w0 , w1 ] ∪ [v0 , v1 , w1 ].

w0 w1

v0 v1

The following picture shows shows how to divide ∆2 × I into three

3-simplices. If we let [v0 , v1 , v2 ] be the simplex ∆2 × {0}, and we let
[w0 , w1 , w2 ] be the simplex at ∆2 × {1}, then ∆2 × I can be written as
the union [v0 , w0 , w1 , w2 ] ∪ [v0 , v1 , w1 , w2 ] ∪ [v0 , v1 , v2 , w2 ].
 homology 69

w2

w0 w1

v2

v0 v1

It is an instructive exercise for the reader to show that

n
∆n × I =
[
[ v0 , . . . , v i , wi , . . . , w n ],
i =0

where we let ∆n × {0} = [v0 , . . . , vn ], ∆n × {1} = [w0 , . . . , wn ], with vi

and wi having the same image under the projection ∆n × I → ∆n .
We define
n
P(σ) := ∑ (−1)i F ◦ (σ, id)|[v0 ,...,vi ,wi ,...,wn ]
i =0

As we discussed earlier, if we can just show that (⋆) holds, we’re

done. We will sketch the proof of this fact below. We will see that
∂P corresponds to the boundary of the prism, g# corresponds to the
top of the prism, f # corresponds to the bottom of the prism, and P∂
corresponds to the sides of the prism.

∂P(σ ) = ∑ (−1) j (−1)i F ◦ (σ, id)|[v0 ,...,vbj ,...,vi ,wi ,...,wn ]

0≤ j ≤ i ≤ n

+ ∑ (−1) j+1 (−1)i F ◦ (σ, id)|[v0 ,...,vi ,wi ,...,c

w j ,...,wn ]
0≤ i ≤ j ≤ n

The terms with i = j in the two sums cancel, except for

F ◦ (σ, id)|[vb0 ,w0 ,...,wn ] = g ◦ σ = g# (σ )

and
− F ◦ (σ, id)|[v0 ,...,vn ,wcn ] = − f ◦ σ = − f # (σ).

The terms with i ̸= j in the sum for ∂P(σ ) are exactly − P∂(σ ).

Definition 5.2.9. A map P which satisfies property (⋆) is called a chain

homotopy between g# and f # .
70 algebraic topology

More generally, if (C• , ∂• ) and ( D• , ∂• ) are two chain complexes

with two chain maps h, k : C• → D• such that there exists a map (i.e.,
chain homotopy) P : Cn → Dn+1 satisfying P∂ + ∂P = h − k, then it
follows as in the above proof that h and k induce the same map on
homology. The chain homotopy condition says that the two ways of
going around the parallelogram from Cn to Dn add up to h − k.

··· Cn+1 Cn Cn−1 ···

h k
P P P
··· Dn + 1 Dn Dn − 1 ···

5.3 Homology of a pair

Given a space X and a subspace A ⊆ X, define

Cn ( X, A) := Cn ( X )/Cn ( A),
called the set of relative n-chains. Since ∂ : Cn ( X ) → Cn−1 ( X ) takes
Cn ( A) to Cn−1 ( A), we get induced boundary maps ∂ : Cn ( X, A) →
Cn−1 ( X, A). Since ∂2 = 0 on Cn ( X ), we have that ∂2 = 0 on Cn ( X, A).
Therefore, we get a chain complex {C• ( X, A), ∂• }, whose homology is
called the relative homology of the pair ( X, A), and is denoted Hn ( X, A).
Then, the natural question to ask is, how does the homology of the pair
( X, A) relate to, or can be computed from, the homologies of X and
and A.
This question is addressed by the following general construction. Let
i j
0 → A• −
→ B• −
→ C• → 0
be a short exact sequence of chain complexes. This means that we have
the following diagram, where every square commutes.

.. .. ..
. . .

i j
0 A n +1 Bn+1 Cn+1 0
∂ ∂ ∂
i j
0 An Bn Cn 0
∂ ∂ ∂
i j
0 A n −1 Bn−1 Cn−1 0

.. .. ..
. . .
 homology 71

For every n, we have homomorphisms

i∗ j∗
Hn ( A• ) −
→ Hn ( B• ) −
→ Hn (C• ).
We are going to define a map ∂ : Hn (C• ) → Hn−1 ( A• ), called a connect-
ing homomorphism.
Let c ∈ Cn be a cycle representative for α ∈ Hn (C• ). Then, since
j is surjective, there exists b ∈ Bn such that c = j(b). Therefore, we
have that ∂(b) ∈ Bn−1 . By the commutativity of the diagram, we know
that j(∂(b)) = ∂( j(b)) = ∂(c) = 0, since c is a cycle. Therefore, ∂(b) ∈
ker j = Im i. So, there exists a (unique, since i is injective) a ∈ An−1
with ∂(b) = i ( a). We show that a is a cycle. Since i (∂( a)) = ∂(i ( a)) =
∂(∂(b)) = 0, and since i is injective, this implies that ∂( a) = 0. Finally,
we define ∂(α) = [ a] ∈ Hn−1 ( A• ), which is clearly a homomorphism.
The next step is to show that this assignment is independent of all
choices.
(i) First, a is uniquely determined by ∂(b), since i is injective.

(ii) Next, suppose we choose b′ ∈ Bn such that j(b′ ) = c. Then, b′ − b ∈

ker j = Im i. So, there exists a′ ∈ An such that b′ − b = i ( a′ ).
Therefore,

∂(b′ ) = ∂(b) + ∂(i ( a′ ))

= ∂(b) + i (∂( a′ ))
= i ( a) + i (∂( a′ ))
= i ( a + ∂( a′ ))
So we see that changing b to b′ amounts to changing a by a homolo-
gous cycle a + ∂( a′ ). In particular, [ a] = [ a + ∂( a′ )] ∈ Hn−1 ( A• ).

(iii) Finally, suppose we choose a different representative for the class

[α]. So, if instead of c we use c + ∂(c′ ) for some c′ ∈ Cn+1 . But then,
c′ = j(b′ ) for some b′ ∈ Bn+1 . So,

c + ∂(c′ ) = c + ∂( j(b′ ))
= c + j(∂(b′ ))
= j(b + ∂(b′ ))
So then, b will be replaced by b + ∂(b′ ), which leaves ∂(b) unchanged,
hence a unchanged.
One can use the connecting homomorphism ∂ just defined to prove
the following statement.

Theorem 5.3.1. The sequence

i ∗ j∗ ∂
· · · → Hn ( A• ) −
→ Hn ( B• ) −
→ Hn (C• ) −
→ Hn−1 ( A• ) → · · ·
is exact.
72 algebraic topology

Proof. This is a routine check. We will show exactness of this diagram

∂ ∗ i
at the step Hn (C• ) −
→ Hn−1 ( A• ) −
→ Hn−1 ( B• ). The other two steps are
exercises for the reader.

1. Im ∂ ⊆ ker i∗ : for a as in the definition of ∂, we have: i∗ (∂(α)) =

i∗ ([ a]) = [∂(b)] = 0.

2. ker i∗ ⊆ Im ∂: let a ∈ An−1 with ∂( a) = 0 and i ([ a]) = 0 ∈ Hn−1 ( B• ).

Then, i ( a) = ∂(b) for some b ∈ Bn . But then, ∂( j(b)) = j(∂(b)) =
j(i ( a)) = 0, since j ◦ i = 0. Thus, j(b) is a cycle. From the construction
of the connecting homomorphism, we have that [ a] = ∂([ j(b)]). Thus,
[ a] ∈ Im ∂.

Let f : ( X, A) → (Y, B) be a continuous map f : X → Y such that

f ( A) ⊆ B. Then f induces f # : Cn ( X ) → Cn (Y ) so that f # (Cn ( A)) ⊆
Cn ( B) for all n. So we get an induced homomorphism

f # : Cn ( X, A) → Cn (Y, B).

Because f # ∂ = ∂ f # on Cn ( X ), this identity also holds for the induced

maps on the quotients. Therefore we get induced homomorphisms on
homology f ∗ : Hn ( X, A) → Hn (Y, B) for all n.
From the definition of relative chains, one also has that Cn ( X, ∅) =
Cn ( X ). So, let ( X, A) be a pair of spaces with A ⊆ X. Therefore, we
have a short exact sequence of chain complexes coming from natural
maps on the level of topological spaces:

0 → C• ( A) → C• ( X ) → C• ( X, A) → 0

Theorem 5.3.1 then yields the following result.

Theorem 5.3.2. Let X be a topological space and let A be a subspace of X.
Then, there is a long exact sequence:

· · · → Hn ( A) → Hn ( X ) → Hn ( X, A) → Hn−1 ( A) → · · ·

We list below a few more consequences of Theorem 5.3.1.

There is a long exact sequence for the reduced homology of a pair
( X, A). This is associated to the “augmented” short exact sequence for
( X, A):

0 / C• ( A) / C• ( X ) / C• ( X, A) /0

ε ε 0
  
0 /Z /Z /0 /0

  
0 0 0
 homology 73

Corollary 5.3.3. There is a long exact sequence for reduced homology of a

pair ( X, A):

· · · −→ H
e n ( A) −→ H
e n ( X ) −→ Hn ( X, A) −→ H
e n−1 ( A) −→ · · ·

Remark 5.3.4. In particular, if x0 ∈ X, the long exact sequence for

reduced homology of the pair ( X, x0 ) yields:

en (X) ∼
H = Hn ( X, x0 )

for all n.

Corollary 5.3.5. There is a long exact sequence for the homology of a triple
( X, A, B), where B ⊆ A ⊆ X:

· · · −→ Hn ( A, B) −→ Hn ( X, B) −→ Hn ( X, A) −→ Hn−1 ( A, B) −→ · · ·

Proof. Start with the short short exact sequence of chain complexes

0 / C• ( A, B) / C• ( X, B) / C• ( X, A) /0

where maps are induced by inclusions of pairs, then take the associated
long exact sequence for homology as in Theorem 5.3.1.

We next discuss properties of homology of pairs of spaces.

Proposition 5.3.6. If f , g : ( X, A) → (Y, B) are homotopic through maps of

pairs ( X, A) → (Y, B), then f ∗ = g∗ : Hn ( X, A) → Hn (Y, B) for all n.

Proof. The prism operator P : Cn ( X ) → Cn+1 (Y ) defined in Theorem

5.2.6, which satisfies ∂P + P∂ = g# − f # , takes Cn ( A) into Cn+1 ( B) by
construction. So we get a prism operator on quotients P : Cn ( X, A) →
Cn+1 (Y, B) which satisfies ∂P + P∂ = g# − f # on Cn ( X, A). Hence f #
and g# have the same effect on Hn ( X, A) for all n. That is, f ∗ = g∗ :
Hn ( X, A) → Hn (Y, B) for all n.

The next result is very important in homology calculations.

Theorem 5.3.7 (Excision Theorem). Given subspaces Z ⊂ A ⊂ X so that

Z ⊂ int( A), the inclusion ( X \ Z, A \ Z ) ,→ ( X, A) induces isomorphisms
∼
=
Hn ( X \ Z, A \ Z ) −
→ Hn ( X, A)

for all n. Equivalently, if A, B ⊆ X are such that X = int( A) ∪ int( B), the
inclusion ( B, A ∩ B) ,→ ( X, A) induces isomorphisms
∼
=
Hn ( B, A ∩ B) −
→ Hn ( X, A)

for all n.
74 algebraic topology

Remark 5.3.8. To see that the two statements of the Excision Theorem
are equivalent, just take B = X \ Z (or Z = X \ B). Then A ∩ B = A \ Z,
and the condition Z ⊂ int( A) is equivalent to X = int( A) ∪ int( B).

Proof of Excision Theorem 5.3.7 (Sketch). Given a topological space X, let

U = {Uj } j be a collection of subspaces of X whose interiors cover X.
Let

CnU ( X ) = {Σim=1 ni σi | m ∈ Z>0 , ni ∈ Z, σi ∈ Cn ( X ),

such that ∀i, ∃ j with σi (∆n ) ⊆ Uj }.

Then CnU ( X ) ≤ Cn ( X ). Furthermore, ∂n : Cn ( X ) → Cn−1 ( X ) induces

boundary maps ∂n on CnU ( X ) satisfying ∂2 = 0. So we get a chain
complex {C∗U ( X ), ∂∗ } whose nth homology group is denoted by HnU ( X ).
By subdividing simplices, it can be shown that the map

HnU ( X ) → Hn ( X )

induced by the inclusion is an isomorphism for all n. In fact, the

inclusion i : CnU ( X ) ,→ Cn ( X ) is a chain homotopy equivalence. That is,
there exists a chain map ρ : Cn ( X ) → CnU ( X ) such that iρ and ρi (the
latter of which is precisely the identity map) are both chain homotopic
to the identity map. So there exists P : Cn ( X ) → Cn+1 ( X ) such that
∂P + P∂ = id − iρ.
For proving the Excision Theorem, we take U = { A, B}, and we
let Cn ( A + B) denote CnU ( X ). Every operator appearing in ∂P + P∂ =
id − iρ takes chains in A to chains in A, so we can factor out the
chains in A to conclude that the inclusions Cn ( A + B)/Cn ( A) ,→
Cn ( X )/Cn ( A) = Cn ( X, A) also induce isomorphisms on homology.
But the map Cn ( B, A ∩ B) = Cn ( B)/Cn ( A ∩ B) ,→ Cn ( A + B)/Cn ( A)
induced by the inclusion is also an isomorphism since both quotient
groups are free with basis the singular n-simplices in B that do not lie
in A. Combining these statements, we obtain the desired isomorphisms
∼
=
Hn ( B, A ∩ B) −
→ Hn ( X, A)

induced by inclusion.

We will next discuss some applications of excision.

The first such application is the Suspension Theorem for homology.
For a space X, define its suspension ΣX to be the quotient of X × [−1, 1]
obtained by identifying X × {−1} to one point and X × {1} to another
point. For example, if X = Sn , then ΣX ∼= S n +1 .
Theorem 5.3.9 (Suspension Theorem). Let X be a topological space , with
suspension ΣX. There are isomorphisms
ei (X ) ∼
H =He i+1 (ΣX ), for all i ≥ 0.
 homology 75

Figure 5.1: Suspension of the circle S1 is

homeomorphic to S2

ΣS1 ∼
= S2

S1

Proof of Suspension Theorem.

 Let   ΣX be thequotient
π : X × [−1, 1] →
map. Let Σ+ X = π X × [− 4 , 1] , let Σ− X = π X × [−1, 14 ] , let S =
1

π ( X × {−1}), and let N = π ( X × {1}). Then we have the following:

e i (ΣX ) ∼
1. H = Hi (ΣX, S).

2. Hi (ΣX, S) ∼
= Hi (ΣX, Σ− X ). This can be seen in two ways:

(a) We observe that Σ− X deformation retracts to S and apply homo-

topy invariance for the homology of a pair.
(b) Hi (Σ− X, S) ∼= 0 by the long exact sequence for reduced homology
of the pair (Σ− X, S). Then the long exact sequence for homology
of the triple (ΣX, Σ− X, S) gives the desired isomorphism.

3. Hi (ΣX, Σ− X ) ∼
= Hi (Σ+ X, X ). This follows by excising int(Σ− X ) and
using homotopy invariance for the homology of a pair.

4. Hi (Σ+ X, X ) ∼ e i−1 ( X ) by applying the long exact sequence for

= H
reduced homology of the pair (Σ+ X, X ) and the fact that Σ+ X is
contractible.

(
Z, i=n
Corollary 5.3.10. H
ei (Sn ) =
0, i ̸= n.

Proof. We use induction on n ≥ 0. H e 0 ( S0 ) ∼

= Z because S0 is two
points. For i > 0, H e i (S0 ) = Hi (S0 ) = Hi ({−1}) ⊕ Hi ({1}) ∼ = 0. So
0 n
the statement holds for S . Assume it holds for S . If i = 0, we know
He i ( S n +1 ) ∼
= 0 because Sn+1 is connected. If i > 0, then H e i ( S n +1 ) =
n
Hi−1 (S ) by the Suspension Theorem. So if i = n + 1, then this group
e
is isomorphic Z, and if i ̸= n + 1 then this group is 0, by the induction
hypothesis.
76 algebraic topology

Theorem 5.3.11 (Brower). If U ⊆ Rm and V ⊆ Rn are nonempty homeo-

morphic open sets, then m = n.

Proof. For all x ∈ U and for all k ∈ Z, we have Hk (U, U \ { x }) = ∼

Hk (R , R \ { x }) by applying the second version of the Excision The-
m m

orem with X = Rm , B = U, and A = Rm \ { x }. Combining this with

the long exact sequence for the reduced homology of (Rm , Rm \ { x })
and the fact that Rm \ { x } is homotopy equivalent to Sm−1 , we obtain
for all x ∈ U and all k ∈ Z:

Hk (U, U \ { x }) ∼
= Hk (Rm , Rm \ { x }) ∼ =He k−1 (Rm \ { x })
(
∼ e k −1 ( S m −1 ) ∼ Z, k = m
=H =
0, k ̸= m.

Similarly, if y ∈ V, we have for all k ∈ Z:

(
Z, k=n
Hk (V, V \ {y}) ∼
=
0, k ̸= n.

But if f : U → V is a homeomorphism, then f : U \ { x } → V \ { f ( x )}

is a homeomorphism. Hence f induces isomorphisms
∼
=
Hk (U, U \ { x }) −
→ Hk (V, V \ { f ( x )})

for all k ∈ Z. Therefore, m = n.

Remark 5.3.12. If X is a topological space, x ∈ X, and U ⊆ X is an

open neighborhood of x, then for all n ∈ Z, the Excision Theorem
yields that
∼ Hn (U, U \ { x }) .
Hn ( X, X \ { x }) =

In particular, for all n ∈ Z, the group Hn ( X, X \ { x }) depends only on

the topology of a neighborhood of x. Therefore these homology groups
are called the local homology groups of X at x. They can be used to check
when a map f : X → Y is not a local homeomorphism.

We can now extend Brower’s fixed point theorem to arbitrary dimen-

sions.

Theorem 5.3.13 (Brower’s Fixed Point Theorem). (i) The boundary ∂D n

of the n-disc D n is not a retract of D n .

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

Proof. (i) Assume by contradiction that there exists a retraction r : D n →

∂D n = Sn−1 . Then, if i : Sn−1 ,→ D n is the inclusion, we have r ◦ i =
 homology 77

idSn−1 . By functoriality, for all k ∈ Z we have (r ◦ i )∗ = r∗ ◦ i∗ =

id He (Sn−1 ) . If k = n − 1 we obtain:
k

∼ i∗ r∗ ∼
Z
= /H
e n −1 ( S n −1 ) /H
e n −1 ( D n ) /H
e n −1 ( S n −1 ) =
4/ Z

idZ

But r∗ = 0 and i∗ = 0 because H e n−1 ( D n ) = 0. Therefore we have

arrived at a contradiction.
(ii) Let f : D n → D n be a continuous map. Assume by contradiction
that f ( x ) ̸= x for all x ∈ D n . Then we may define a function r : D n →
Sn−1 in the following way. Let x ∈ D n and let [ f ( x ), x ) denote the
(unique) ray based at f ( x ) passing through x. Define r ( x ) to be the
unique element in ([ f ( x ), x ) ∩ ∂D n ) \ { f ( x )}. Then r is continuous and
is a retraction D n → ∂D n , contradicting (i).

The following result is very useful in concrete calculations.

Theorem 5.3.14 (Mayer-Vietoris Sequence). Suppose X = A ∪ B =

int( A) ∪ int( B). Then there is a long exact sequence:

ϕ ψ ∂
· · · → Hn ( A ∩ B) → Hn ( A) ⊕ Hn ( B) → Hn ( X ) → Hn−1 ( A ∩ B) →
· · · → H0 ( X ) → 0.

Proof. Let Cn ( A + B) denote the subgroup of Cn ( X ) whose elements

are precisely sums of singular simplices in either A or B. The boundary
maps ∂ on Cn ( X ) restrict to boundary maps on Cn ( A + B), and we get
a chain complex {C∗ ( A + B), ∂∗ } whose homology is isomorphic to the
homology of X. Hence, we need only produce a long exact sequence
of the form specified in the theorem where each Hn ( X ) is replaced by
the n-th homology group of {C∗ ( A + B), ∂∗ }. To this end, for n ∈ Z≥0 ,
consider the following sequence:

/ Cn ( A ∩ B) / Cn ( A) ⊕ Cn ( B) / Cn ( A + B)
/0
ϕ ψ
0
(5.3.1)
where, ϕ( x ) = ( x, − x ) for all x ∈ Cn ( A ∩ B) and ψ( x, y) = x + y for all
( x, y) ∈ Cn ( A) ⊕ Cn ( B). We claim that this sequence is exact:

• ψ is surjective by the definition of Cn ( A + B).

• ϕ is injective, since a chain in A ∩ B which is zero as a chain in A (or

in B) must be the zero chain.

• For all x ∈ Cn ( A ∩ B), ψ ◦ ϕ( x ) = x − x = 0. Therefore Im(ϕ) ⊆

ker(ψ).
78 algebraic topology

• If ( x, y) ∈ ker(ψ), then x is a chain in A, y is a chain in B, and y = − x.

This implies that x is a chain in A ∩ B and ϕ( x ) = ( x, − x ) = ( x, y).
Therefore ker(ψ) ⊆ Im(ϕ).

It is easy to see that ϕ, ψ commute with the boundary operators, so

(5.3.1) yields a short exact sequence of chain complexes, and the Mayer-
Vietoris sequence is simply the associated long exact sequence in ho-
mology.

Remark 5.3.15. By using augmented chain complexes in (5.3.1), we

also obtain a corresponding Mayer-Vietoris sequence for the reduced
homology groups.

Example 5.3.16. Let X = Sn , A = Sn \ {S}, and B = Sn \ { N } where S

and N are the south pole and north pole, respectively. Then A ∼ = Rn ,
B∼= Rn , and A ∩ B ≃ Sn−1 . From the reduced Mayer-Vietoris sequence,
e i (Sn ) ∼
we get H =He i−1 (Sn−1 ) for all i. By induction, we find as before:
(
n ∼ Z, i = n
Hi (S ) =
e
0, i ̸= n.

Example 5.3.17 (Homology of the Klein Bottle). Let K be the Klein

bottle. It may be decomposed as K = M1 ∪ M2 where M1 and M2
are Möbius bands that are glued along their boundary circles (see the
figure below).

b K
M2
b
a M1 a M1

Each of M1 , M2 is homotopy equivalent to its core circle S1 , and

M1 ∩ M2 = S1 is the common boundary circle. By the reduced Mayer-
Vietoris sequence, Hn (K ) ∼
= 0 for all n > 2. Consider the segment of
the reduced Mayer-Vietoris sequence below:
ϕ ψ
0 → H2 (K ) → H1 ( M1 ∩ M2 ) → H1 ( M1 ) ⊕ H1 ( M2 ) → H1 (K ) → 0

Then ϕ : Z → Z ⊕ Z maps 1 to (2, −2). By exactness, H2 (K ) ∼ =

ker(ϕ) ∼= 0 and H1 (K ) ∼ = Coker(ϕ) ∼= (Z ⊕ Z) /⟨2(1, −1)⟩. If we con-
sider the basis {(1, 0), (1, −1)} of Z ⊕ Z, then (Z ⊕ Z) /⟨2(1, −1)⟩ ∼
=
Z ⊕ Z2 . We conclud the following:
(
∼
e i (K ) = Z ⊕ Z2 , i = 1
H
0, i ̸= 1.
 homology 79

Exercises
1. Show that if X is a path-connected topological space and f : X → X
is a continuous function, then the induced map f ∗ : H0 ( X ) → H0 ( X ) is
the identity map.

2. Show that H0 ( X, A) = 0 if and only if A meets each path-component

of X.

3. Show that H1 ( X, A) = 0 if and only if H1 ( A) → H1 ( X ) is surjective

and each path-component of X contains at most a path-component of
A.

4. A pair ( X, A) with X a space and A a nonempty closed subspace

that is a deformation retract of some neighborhood in X is called
a good pair. Show that for a good pair ( X, A), the quotient map
q : ( X, A) → ( X/A, A/A) obtained by collapsing A to a point, induces
isomorphisms q∗ : Hn ( X, A) → Hn ( X/A, A/A) ∼ =He n ( X/A), for all n.

W W
5. For a wedge sum α Xα , the inclusions iα : Xα ,→ α Xα induce an
isomorphism
M M _
iα ∗ : e n ( Xα ) → H
H en ( Xα ),
α α α
provided that the wedge sum is formed at basepoints xα ∈ Xα such
that the pairs ( Xα , xα ) are good.

6. Show that:

(i) Sn and Sm do not have the same homotopy type if n ̸= m.

(ii) Sn , for n > 1, is a simply-connected space which is not contractible.

7. Calculate the homology of the 2-torus T 2 .

8. Show that S1 × S1 and S1 ∨ S1 ∨ S2 have isomorphic homology

groups in all dimensions. Are these spaces homeomorphic?

9. Show that the quotient map S1 × S1 → S2 collapsing the subspace

S1 ∨ S1 to a point is not nullhomotopic by showing that it induces an
isomorphism on H2 . On the other hand, show that any map S2 →
S1 × S1 is nullhomotopic.

10. For ΣX the suspension of X, show by a Mayer-Vietoris argument

e n+1 (ΣX ) ∼
that there are isomorphisms H =He n ( X ) for all n.

11. For the case of the inclusion f : ( D n , Sn−1 ) ,→ ( D n , D n − {0}),

show that f is not a homotopy equivalence of pairs, i.e., there is no
80 algebraic topology

g : ( D n , D n − {0}) → ( D n , Sn−1 ) so that g ◦ f and f ◦ g are homotopic

to the identity through maps of pairs.

12. A graded abelian group is a sequence of abelian groups A• :=

( An )n≥0 . We say that A• is of finite type if

∑ rankAn < ∞.
n ≥0

The Euler characteristic of a finite type graded abelian group A• is the

integer
χ( A• ) := ∑ (−1)n · rankAn .
n ≥0

(i) Suppose

∂ ∂ ∂ ∂
· · · → Cn → Cn−1 → · · · → C1 → C0 → 0

is a chain complex such that the graded abelian group C• is of finite

type. Denote by Hn the n-th homology group of this complex and
form the corresponding graded group H• = ( Hn )n≥0 . Show that H•
is of finite type and
χ( H• ) = χ(C• ).

(ii) Suppose we are given three finite type graded abelian groups A• ,
B• , C• , which are part of a long exact sequence

i
k jk ∂
· · · → Ak → Bk → Ck →k Ak−1 → · · · → A0 → B0 → C0 → 0.

Show that
χ( B• ) = χ( A• ) + χ(C• ).
 homology 81

5.4 π1 vs. H1

Let X be a topological space. A continuous map f : I = [0, 1] → X can

be viewed as a path in X or as a singular 1-simplex. If f (0) = f (1),
then ∂ f = f (1) − f (0) = 0, so a loop in X can be viewed as a 1-cycle.
In this section, we discuss the following.
Theorem 5.4.1. By regarding loops as singular 1-cycles, one gets a homomor-
phism
h : π1 ( X, x0 ) → H1 ( X ).
If X is path-connected, then h is onto, with ker h = [π1 , π1 ], the commu-
tator subgroup of π1 := π1 ( X, x0 ). In this case, h induce an isomorphism
π1 ( X, x0 )ab ∼
= H1 ( X ), i.e., the first homology group can be seen as the
abelianization of the fundamental group.

Remark 5.4.2. An equivalent definition of h can be given as follows: if

f : S1 → X is an element of π1 ( X, x0 ), define

h([ f ]) := f ∗ (α),

for α ∈ H1 (S1 ) a generator represented by σ : I → S1 , s 7→ e2πis . Then

both [ f ] ∈ π1 ( X, x0 ) and f ∗ (α) are represented by the loop f σ : I →
X. A consequence of this formulation is that h([ f ]) = h([ g]) if f is
homotopic to g.

Proof. (i) If f = const x0 is the constant path, then f is a 1-cycle since

it is a loop, and f must be a boundary since H1 ( point) = 0. In fact,
f = ∂(σ), for σ the constant singular 2-simplex with the same image as
f , since

∂(σ ) = σ |[v1 ,v2 ] − σ |[v0 ,v2 ] + σ |[v0 ,v1 ] = f − f + f = f .

(ii) If f is homotopic to g through a path-homotopy preserving

basepoints, we show that f and g are homologous, hence correspond to
the same element in H1 ( X ). Indeed, let F : I × I → X be a homotopy
from f to g, so f (0) = g(0) = Fs (0) = x0 , f (1) = g(1) = Fs (1) = x0 ,
where F (t, s) = Fs (t).
g

σ2

x0 y0

σ1

f
82 algebraic topology

Let σ1 and σ2 be 2-simplices as in the above figure. Then:

∂(σ1 − σ2 ) = f − g − const x0 + const x0 .

Hence f − g is a boundary, whence f and g define the same element in

H1 ( X ).
(iii) We next show that multiplication (concatenation) of loops trans-
lates into cycle addition. i.e., if f , g : I → X are loops at x0 we show
that f · g is homologous to f + g, or equivalently, that f · g − f − g is a
boundary. Consider the singular 2-simplex σ depicted below. Then

∂(σ) = g − f · g + f .

v1

f g

v0 v2
f g

(iv) If f¯ is the inverse path of f , we show that f¯ is homologous

as a 1-cycle to − f . Indeed, f + f¯ − f · f¯ is a boundary by (iii) and
f · f¯ ∼ const x0 is (homologous to) a boundary by (i).

It then follows from (ii) and (iii) that h : π1 ( X, x0 ) → H1 ( X ) is a well

defined homomorphism. Hence, since H1 ( X ) is abelian, there is an
induced homomorphism π1 ( X, x0 )ab → H1 ( X ), also denoted by h. To
show that this is an isomorphism for X path connected, we construct
an inverse

j : H1 ( X ) → π1 ( X, x0 )ab .

For each x ∈ X, let ϕx be a fixed path in X from x0 to x, with ϕx0 =

const x0 the constant path at x0 . For σ a singular 1-simplex in X with
endpoints x1 and x2 , set

σ : = ϕx1 ∗ σ ∗ ϕx2 .
b

Then the map σ 7→ b σ defines a homomorphism C1 ( X ) → π1 ( X, x0 )ab

on its basis of singular 1-simplices.
Let us next note that if ρ is a singular 2-simplex, then ∂(ρ) maps to
the identity element. Indeed, if ∂(ρ) = σ0 − σ1 + σ2 ,
 homology 83

x3

σ2 σ1

x1 x2
σ0

then ∂(ρ) maps to the homotopy class of the path

ϕx1 ∗ σ0 ∗ ϕx2 ∗ ϕx2 ∗ σ1 ∗ ϕx3 ∗ ϕx3 ∗ σ2 ∗ ϕx1 ∼ ϕx1 ∗ (σ0 ∗ σ1 ∗ σ2 ) ∗ ϕx1 .

But σ0 ∗ σ1 ∗ σ2 = ρ∗ (γ), for γ a loop in ∆2 based at x1 . And since ∆2 is

simply connected, one has that σ0 ∗ σ1 ∗ σ2 ∼ const x1 . Therefore,

ϕx1 ∗ (σ0 ∗ σ1 ∗ σ2 ) ∗ ϕx1 ∼ ϕx1 ∗ const x1 ∗ ϕx1 ∼ const x0 .

Therefore, if we restrict C1 ( X ) → π1 ( X, x0 )ab to Z1 ( X ) and use the fact

that B1 ( X ) 7→ const x0 , we get an induced homomorphism

j : H1 ( X ) → π1 ( X, x0 )ab .

Finally, we show that h and j are inverse homomorphisms. First, if

σ is a loop at x0 ∈ X, then b σ = σ, hence j ◦ h = id. Suppose now that
c = ∑i ni σi is a singular 1-cycle. Then, under h ◦ j, σi maps to σbi which,
by (iii), is homologous to

(iv)
ϕ pi + σi + ϕqi = ϕ pi + σi − ϕqi ,

where pi , qi are the endpoints of σi . Hence c maps under h ◦ j to

∑ ni σi + ∑ ni (ϕpi − ϕqi ) = c + ∑ ni (ϕpi − ϕqi ).

i i i

At this end, note that that since 0 = ∂(c) = ∑i ni ( pi − qi ), it follows

readily that ∑i ni (ϕ pi − ϕqi ) = 0.

5.5 Cellular Homology

In this section, we introduce cellular homology, which is a new homology

theory for certain nice spaces called CW complexes, and show that for
such spaces cellular homology is isomorphic to singular homology. We
begin by introducing and studying the notion of degree of a self-map
of a sphere; the degree will play a fundamental role in computing the
boundary maps in the cellular chain complex, whose homology gives
the cellular homology.
84 algebraic topology

Degrees
Definition 5.5.1. The degree of continuous map f : Sn → Sn is defined as:

deg f := f ∗ (1) (5.5.1)

e n (Sn ) = Z → H
where f ∗ : H e n (Sn ) = Z is the homomorphism induced by f
in homology, and 1 ∈ Z denotes the generator.

The degree has the following properties:

1. deg idSn = 1.

Proof. This is because (idSn )∗ = id which is multiplication by the

integer 1.

2. If f is not surjective, then deg f = 0.

Proof. Indeed, if f is not surjective, there is some y ∈

/ Im f . Then we
can factor f in the following way:

g h
f : Sn −→ Sn \ {y} −→ Sn .

Since Sn \ {y} ∼ = Rn is contractible, H

e n (Sn \ {y}) = 0. Therefore
f ∗ = h∗ g∗ = 0, so deg f = 0.

3. If f ≃ g are homotopic maps, then deg f = deg g.1 1

By a theorem of Hopf, the converse of
this statement is also true.

Proof. This is because f ∗ = g∗ .

4. deg( g ◦ f ) = deg g · deg f .

Proof. Indeed, we have that ( g ◦ f )∗ = g∗ ◦ f ∗ .

5. If f is a homotopy equivalence, then deg f = ±1.

Proof. By definition, there exists a map g : Sn → Sn so that g ◦ f ≃

idSn and f ◦ g ≃ idSn . The claim follows directly from 1, 3, and
4 above, since f ◦ g ≃ idSn implies that deg f · deg g = deg idSn =
1.

6. If r : Sn → Sn is a reflection across some n-dimensional subspace of

Rn+1 , then deg r = −1.
 homology 85

Proof. Without loss of generality we can assume that the subspace is

Rn × {0} ⊂ Rn+1 , with

r ( x 0 , . . . x n ) = ( x 0 , . . . , x n −1 , − x n ).

The upper and lower hemispheres U and L of Sn can be regarded as

singular n-simplices, via their standard homeomorphisms with ∆n .
Then the generator of He n (Sn ) is [U − L]. The reflection map r maps
the cycle U − L to L − U = −(U − L). So

r∗ ([U − L]) = [ L − U ] = [−(U − L)] = (−1) · [U − L]

so deg r = −1.

7. If a : Sn → Sn is the antipodal map x 7→ − x, then deg a = (−1)n+1 .

Proof. Note that a is a composition of n + 1 reflections, since there

are n + 1 coordinates in x, each changing sign by an individual
reflection. From 4 above we know that composition of maps leads to
multiplication of degrees.

8. If f : Sn → Sn is a continuous map, and S f : Sn+1 → Sn+1 is the

suspension of f then deg S f = deg f .

Proof. Recall that if f : X → X is a continuous map and

ΣX = X × [−1, 1]/( X × {−1}, X × {1})

denotes the suspension of X, then S f := f × id[−1,1] / ∼, with the

same equivalence as in ΣX. Note that ΣSn = Sn+1 .
The Suspension Theorem states that

ei (X ) ∼
H =He i+1 (ΣX ), ∀i ≥ 0.

We already proved this fact by using the excision theorem 5.3.9. Here
we give another proof by using the Mayer-Vietoris sequence 5.3.14
for the decomposition

ΣX = C+ X ∪ X C− X,

where C+ X and C− X are the upper and lower cones of the suspen-
sion joined along their bases:

··· → H
e i+1 (C+ X ) ⊕ H
e i+1 (C− X ) → He i+1 (ΣX ) →
→H ei (X ) → H
e i (C+ X ) ⊕ H
e i (C− X ) → · · ·

Since C+ X and C− X are both contractible, the end groups in the

ei (X ) ∼
above sequence are both zero. Thus, by exactness, we get H =
86 algebraic topology

He i+1 (ΣX ), as desired.

Let C+ Sn denote the upper cone of ΣSn . Note that the base of C+ Sn
is Sn × {0} ⊂ ΣSn . The map f induces a map C+ f : (C+ Sn , Sn ) →
(C+ Sn , Sn ) whose quotient is S f . The long exact sequence of the pair
(C+ Sn , Sn ) in homology gives the following commutative diagram:

0 e i+1 (C+ Sn /Sn ) ∂- H

- Hi+1 (C+ Sn , Sn ) ≃ H e i (Sn ) - 0
∼

(S f )∗ f∗
? ?
e i +1 ( S n +1 ) ∂- e n
H Hi (S )
∼

Note that C+ Sn /Sn ∼= Sn+1 so the boundary map ∂ at the top and
bottom of the diagram are the same map. So by the commutativity
of the diagram, since f ∗ is defined by multiplication by some integer
m, then (S f )∗ must be given by multiplication by the same integer
m.

Example 5.5.2. Consider the reflection map: rn : Sn → Sn defined

by ( x0 , . . . , xn ) 7→ (− x0 , x1 , . . . , xn ). Since rn leaves x1 , x2 , . . . , xn
unchanged we can unsuspend one coordinate at a time to get

deg rn = deg rn−1 = · · · = deg r0 ,

where ri : Si → Si by ( x0 , x1 , . . . , xi ) 7→ (− x0 , x1 , . . . , xi ). So r0 : S0 →
S0 is given by x0 7→ − x0 . Note that S0 is two points but in reduced
homology we are only looking at one integer. Consider

e0 (S0 ) → H0 (S0 ) −
0→H →Z→0
ϵ

where H e0 (S0 ) = {( a, − a) | a ∈ Z}, H0 (S0 ) = Z ⊕ Z, and ϵ :

e 0 ( S0 ) → H
( a, b) 7→ a + b. Then (r0 )∗ : H e0 (S0 ) is given by ( a, − a) 7→
(− a, a) = (−1) · ( a, − a). So deg rn = −1.

9. If f : Sn → Sn has no fixed points then deg f = (−1)n+1 .

Sn
−x

f (x)
x
 homology 87

Proof. Consider the above figure. Since f ( x ) ̸= x, the segment

(1 − t) · f ( x ) + t · (− x ) from − x to f ( x ) does not pass through the
origin in Rn+1 . So we can normalize to obtain a homotopy:

((1 − t) · f ( x ) + t · (− x )
gt ( x ) : = : Sn → Sn .
||(1 − t) · f ( x ) + t · (− x )||

Note that this homotopy is well defined since (1 − t) · f ( x ) + t ·

(− x ) ̸= 0 for any x ∈ Sn and t ∈ [0, 1], because f ( x ) ̸= x for all x.
Then gt is a homotopy from f to a, the antipodal map, so they same
the same degree.

10. The n-sphere Sn has a continuous field of non-zero tangent vectors

if and only if n is odd.

Proof. Suppose x 7→ v( x ) is a tangent vector field on Sn , assigning

to a vector x ∈ Sn the vector v( x ) tangent to Sn at x. Regarding
v( x ) as a vector at the origin, tangency implies that x and v( x ) are
orthogonal in Rn+1 . If v( x ) ̸= 0 for all x, we may normalize so that
||v( x )|| = 1 for all x. Assuming this has been done, the vectors
(cos t) x + (sin t)v( x ) lie in the unit circle in the plane spanned by x
and v( x ). Letting t go from 0 to π, we obtain a homotopy:

f t ( x ) = (cos t) x + (sin t)v( x )

from the identity map of Sn to the antipodal map. In terms of degree,

this yields (−1)n+1 = 1, which implies that n is odd.
Conversely, if n = 2k − 1, the vector field defined by

v( x1 , x2 , · · · , x2k−1 , x2k ) = (− x2 , x1 , · · · , − x2k , x2k−1 )

is a nowhere vanishing tangent vector field, since v( x ) is orthogonal

to x, and ||v( x )|| = 1 for all x ∈ Sn .

Exercises
1. Let f : Sn → Sn be a map of degree zero. Show that there exist points
x, y ∈ Sn with f ( x ) = x and f (y) = −y.

2. Let f : S2n → S2n be a continuous map. Show that there is a point

x ∈ S2n so that either f ( x ) = x or f ( x ) = − x.

3. A map f : Sn → Sn satisfying f ( x ) = f (− x ) for all x is called an

even map. Show that an even map has even degree, and this degree is in
fact zero when n is even. When n is odd, show there exist even maps
of any given even degree.
88 algebraic topology

How to Compute Degrees?

Assume f : Sn → Sn is surjective, and that f has the property that there

exists some y ∈ Im(Sn ) so that f −1 (y) is a finite number of points, say
f −1 (y) = { x1 , x2 , . . . , xm }. Let Ui be a neighborhood of xi so that all
Ui ’s get mapped to some neighborhood V of y. So f (Ui \ xi ) ⊂ V \ y.
As f is continuous, we can choose the Ui ’s to be disjoint.

U1
x1

V
U2
x2
y

Um
xm

Let f |Ui : Ui → V be the restriction of f to Ui , with induced homomor-

phism
f ∗ : Hn (Ui , Ui \ xi ) −→ Hn (V, V \ y)

Note that by using excision and homology long exact sequences, one
has:
Hn (Ui , Ui \ xi ) ∼
= Hn (Sn , Sn \ xi ) ∼
=He n (Sn ) ∼
=Z

and
Hn (V, V \ y) ∼
= Hn (Sn , Sn \ y) ∼
=He n (Sn ) ∼
= Z.

Let us define the local degree of f at xi , denoted by deg f | xi , to be the

effect of f ∗ : Hn (Ui , Ui \ xi ) → Hn (V, V \ y). We then have the following
result:

Theorem 5.5.3. The degree of f equals the sum of local degrees at points in a
generic fiber, that is,
m
deg f = ∑ deg f |xi .
i =1

Proof. Consider the following commutative diagram, where the isomor-

phisms labelled by “exc” follow from excision, and “l.e.s” stands for a
 homology 89

long exact sequence.

f∗ -
Z∼
= Hn (Ui , Ui \ xi ) Hn (V, V \ y) ∼
=Z
· deg f | xi

∼ exc ∼
=, ki = exc


?
m ?
Pi M exc f∗
Z∼
= Hn (Sn , S
n
\ xi )  Hn (Ui , Ui \ xi ) ∼
= Hn (Sn , Sn \ f −1 (y)) - Hn (Sn , Sn \ y)
i =1

∼
6
6 = l.e.s.
∼
= , l.
e.s
l.e.s. j

f∗
Z∼
=He n (Sn ) - e n (Sn ) ∼
H =Z
· deg f

By examining the diagram above we have:

k i (1) = (0, . . . , 0, 1, 0, . . . , 0)

where the entry 1 is in the ith place. Also, Pi ◦ j(1) = 1, for all i, so
m
j(1) = (1, 1, . . . , 1) = ∑ k i (1).
i =1

The commutativity of the lower square gives:

 m 
deg f = f ∗ j(1) = f ∗ ∑ k i (1)
i =1
m
= ∑ f ∗ (0, . . . , 0, 1, 0, . . . , 0)
i =1
m
= ∑ deg f |xi ,
i =1

where the last equality follows from the commutativity of the upper
square.

Example 5.5.4. Let us consider the power map f : S1 → S1 , f ( x ) = x k ,

k ∈ Z. We claim that deg f = k. We distinguish the following cases:

• If k = 0 then f is the constant map which has degree 0.

• If k < 0 we can compose f with a reflection r : S1 → S1 by ( x, y) →

( x, −y). This reflection has degree −1. So since composition leads to
multiplication of degrees, we can assume that k > 0.

• If k > 0, then for all y ∈ S1 , f −1 (y) consists of k points (the k roots

of y), call them x1 , x2 , . . . , xk , and f has local degree 1 at each of
these points. Indeed, for the above y ∈ Sn we can find a small open
neighborhood centered at y, call this neighborhood V, so that he
pre-images of V are open neighborhoods Ui centered at each xi , with
90 algebraic topology

f |Ui : Ui → V a homeomorphism (which has possible degree ±1). In

our case, these homeomorphisms are restrictions of a rotation, which
is homotopic to the identity, and thus the degree of f |Ui equals 1, for
each i.
So the degree of f is indeed k. Note that this implies that we can
construct maps Sn → Sn of arbitrary degrees for any n, simply by
suspending the power map f .

CW Complexes
We next introduce cellular complexes (also referred to as cell-complexes
or CW complexes), and discuss a few important examples.
Start with a discrete set X0 , whose points are called 0-cells. In-
ductively, we form the n-skeleton Xn from Xn−1 by attaching n-cells
φn
eλn = Int( Dλn ) via maps ∂Dλn = Sλn−1 −→
λ
Xn−1 , i.e.,

Xn = Xn−1 ⨿λ Dλn ∼


with the identification x ∼ φnλ ( x ) for all x ∈ ∂Dλn . As a set, Xn =

Xn−1 ⨿λ eλn , where eλn is the homeomorphic image of Int( Dλn ) = Dλn \
∂Dλn under the quotient map. We can either stop this inductive process
at a finite stage, setting X = Xl for some l, or continue indefinitely, in
S
which case we set X = n Xn . Such a space X is called a CW (cell-)
complex.
Each cell eλn has a characteristic map Φnλ defined by the composition:

Dλn ,→ Xn−1 ⨿λ Dλn → Xn ,→ X.

Note that Φnλ | Int( Dn ) is a homeomorphism onto eλn , while the restriction
λ
of Φnλ to ∂Dλn is the attaching map φnλ .
A CW complex is endowed with the weak topology, i.e., A ⊂ X
is open ⇐⇒ A ∩ Xn is open for all n. An n-cell will be denoted by
eλn = Int( Dλn ). One can think of X as a disjoint union of cells of various
dimensions, or as ⨿n,λ Dλn ∼, where ∼ means that we are attaching


the cells via their respective attaching maps.

A CW complex X is finite if it has finitely many cells. A CW complex
is of finite type if it has finitely many cells in each dimension. Note
that a CW complex of finite type may have cells in infinitely many
dimensions. If X = ∪n Xn and Xm = Xn for all m > n for some n, then
X = Xn and we say that the skeleton stabilizes. The smallest n for
which X = Xn is called the dimension of X.
Remark 5.5.5. One space X may admit many CW structures, see the
case of Sn below.
Example 5.5.6. On the n-sphere Sn we have a CW structure with one
0-cell e0 and one n-cell en . The attaching map for the n-cell is the
 homology 91

constant map φ : Sn−1 = ∂D n → e0 = point, and there is only one such

map, the collapsing map. Think of taking the disk D n and collapsing
its entire boundary to a single point, giving Sn .
Example 5.5.7. A different CW structure on Sn can be constructed
so that there are two cells in each dimension from 0 to n. Start with
X0 = S0 = {e10 , e20 }. Then X1 = S1 where the two 1-cells D11 , D21
are attached to the 0-cells by homeomorphisms on their boundary.
Similarly, two 2-cells can be attached to X1 = S1 by homeomorphism
on their boundary giving X2 = S2 . Continuing in this manner, i.e.,
adding two cells in each new dimension, yields the above-mentioned
CW structure of Sn . Note that if we identify each pair of cells in the
same dimension by the antipodal map, we get a CW structure on the
real projective space RPn , with one cell in each dimension from 0 to n.
Example 5.5.8. The complex projective space CPn = Cn+1 \ {0} /C∗ is

identified with the collection of complex lines through the origin in

Cn+1 . It is also the orbit space of the C∗ -action on Cn+1 \ {0} given by

λ · (z0 , . . . , zn ) 7→ (λz0 , . . . , λzn ).

Let [z0 : . . . : zn ] ∈ CPn be the equivalence class of (z0 , · · · , zn ) ∈ Cn+1

under this action. Define

Φ : D2n → CPn

by v
 
n −1
u
: t1 − ∑ | z i |2  .
u
( z 0 , . . . , z n −1 ) 7 →  z 0 : . . . : z n −1
i =0

Then Φ takes ∂D2n into the set of points with zn = 0, i.e., into CPn−1 .
Let φ := Φ|∂D2n . It is easy to check that Φ factors through CPn−1 ∪ φ D2n
and, moreover, the resulting map

CPn−1 ∪ φ D2n → CPn

is a homeomorphism (it is a bijective map from a compact space to a

Hausdorff space, hence it is a homeomorphism onto its image). So it
follows inductively that CPn has a CW structure with one cell in each
even dimension 0, 2, . . . , 2n, where the attaching maps are the maps
labelled by φ. There are no cells of odd dimension.
Example 5.5.9. A covering space of a CW complex has a canonical
structure as a CW complex. Let f : X → Y be a covering map so that
Y is a CW complex with characteristic maps Φλ : Dλn → Y. As Dλn
is simply-connected, each Φλ lifts to a map Φ e n : D n → X, which are
λ λ
unique upon specification of the image of any point. The collection of
all such liftings of all Φnλ define a cell structure on X.
92 algebraic topology

Exercises
1. Let X and Y be finite CW complexes. Show that X × Y has the
structure of a finite CW complex with an (open) n + m dimensional cell
e × e′ for each n dimensional cell e in X and each m dimensional cell e′
in Y.

Cellular Homology
In this section, we show how to compute the homology of a CW
complex (assumed, for simplicity, to be of finite type). We first introduce
cellular homology and then we show that it can be identified with the
singular homology.
We start with the following preliminary result:

Lemma 5.5.10. If X is a CW complex of finite type, then:


0, k ̸= n
(a) Hk ( Xn , Xn−1 ) =
Z # n-cells , k = n.

(b) Hk ( Xn ) = 0 if k > n. In particular, if X is finite dimensional, then

Hk ( X ) = 0 if k > dim( X ).
∼
=
(c) The inclusion i : Xn ,→ X induces an isomorphism Hk ( Xn ) → Hk ( X ) if
k < n.

Proof. (a) We know that Xn is obtained from Xn−1 by attaching the

n-cells (eλn )λ . Pick a point xλ at the center of each n-cell eλn , and let
A := Xn − { xλ }λ . Then A deformation retracts to Xn−1 , so we have
that
Hk ( Xn , Xn−1 ) ∼
= Hk ( Xn , A).

Since the closure of Xn−1 is contained in the interior of A, by excising

Xn−1 the latter group is isomorphic to λ Hk ( Dλn , Dλn − { xλ }). More-
L

over, the homology long exact sequence of the pair ( Dλn , Dλn − { xλ })
yields that

Z, k=n
Hk ( Dλn , Dλn − { xλ }) ∼
=He k −1 ( S n −1 ) ∼
λ =
0, k ̸= n.

So the assertion follows.

(b) Consider the following portion of the long exact sequence of the
pair for ( Xn , Xn−1 ):

Hk+1 ( Xn , Xn−1 ) → Hk ( Xn−1 ) → Hk ( Xn ) → Hk ( Xn , Xn−1 )

 homology 93

If k + 1 ̸= n and k ̸= n, we have from part (a) that Hk+1 ( Xn , Xn−1 ) = 0

and Hk ( Xn , Xn−1 ) = 0. Thus Hk ( Xn−1 ) ∼
= Hk ( Xn ). Hence if k > n (so
in particular, n ̸= k + 1 and n ̸= k), we get by iteration that

Hk ( Xn ) ∼
= Hk ( Xn−1 ) ∼
= ··· ∼
= Hk ( X0 ).

Note that X0 is just a collection of points, so Hk ( X0 ) = 0. Thus when

k > n we have Hk ( Xn ) = 0 as desired.

(c) For simplicity, we only prove here the statement for finite dimen-
sional CW complexes. Let k < n and consider the following portion of
the long exact sequence for the pair ( Xn+1 , Xn ):

Hk+1 ( Xn+1 , Xn ) → Hk ( Xn ) → Hk ( Xn+1 ) → Hk ( Xn+1 , Xn )

Since k < n we have k + 1 ̸= n + 1 and k ̸= n + 1, so by part (a) we get

that Hk+1 ( Xn+1 , Xn ) = 0 and Hk ( Xn+1 , Xn ) = 0. Thus

Hk ( Xn ) ∼
= Hk ( Xn+1 ).

By repeated iterations, we obtain:

Hk ( Xn ) ∼ = Hk ( Xn+2 ) ∼
= Hk ( Xn+1 ) ∼ = ··· ∼
= Hk ( Xn+l ) = Hk ( X ),

where l is so that Xn+l = X (since we assumed X is finite dimensional).

This proves the claim.

In what follows we define the cellular homology of a CW complex X

in terms of a given cell structure, then we show that it coincides with
the singular homology, so it is in fact independent on the cell structure.
Cellular homology is a very useful tool for computations.

Definition 5.5.11. The cellular homology H∗CW ( X ) of a CW complex X is

the homology of the cellular chain complex (C∗ ( X ), d∗ ) indexed by the cells of
X, i.e.,
Cn ( X ) := Hn ( Xn , Xn−1 ), (5.5.2)

and with differentials

dn : Cn ( X ) −→ Cn−1 ( X )

defined by the following diagram, with diagonal arrows induced from long
exact sequences of pairs:
94 algebraic topology

Hn ( Xn+1 , Xn ) = 0
-

Hn ( Xn−1 ) = 0 Hn ( Xn+1 ) ∼
= Hn ( X )
-
in
-
Hn ( Xn )
-
1 jn
∂ n+
-
d n +1 dn
Hn+1 ( Xn+1 , Xn ) - Hn ( Xn , Xn−1 ) - Hn−1 ( Xn−1 , Xn−2 )
-
∂n
- jn −1

Hn−1 ( Xn−1 )
-

Hn−1 ( Xn−2 ) = 0

Here we use Lemma 5.5.10 for the identifications

Hn ( Xn−1 ) = 0, Hn−1 ( Xn−2 ) = 0, Hn ( Xn+1 ) ∼

= Hn ( X )
in the diagram. In the notations of the diagram, we set:

dn = jn−1 ◦ ∂n : Cn ( X ) → Cn−1 ( X ), (5.5.3)

and note that we have

dn ◦ dn+1 = 0. (5.5.4)
Indeed,
dn ◦ dn+1 = jn−1 ◦ ∂n ◦ jn ◦ ∂n+1 = 0,
since ∂n ◦ jn = 0 as the composition of two consecutive maps in a long
exact sequence. So {C∗ ( X ), d∗ } is a chain complex.
The following result asserts that cellular homology is independent
on the cell structure used for its definition:

Theorem 5.5.12. There are isomorphisms

HnCW ( X ) ∼
= Hn ( X )

for all n, where Hn ( X ) is the singular homology of X.

Proof. Since Hn ( Xn+1 , Xn ) = 0 and Hn ( X ) ∼

= Hn ( Xn+1 ), we get from
the diagram above that

Hn ( X ) ∼
= Hn ( Xn )/ ker in ∼
= Hn ( Xn )/Im ∂n+1 .

Now, Hn ( Xn ) ∼
= Im jn ∼= ker ∂n ∼= ker dn . The first isomorphism
comes from jn being injective, the second follows by exactness, and
 homology 95

ker ∂n = ker dn since dn = jn−1 ◦ ∂n and jn−1 is injective. Also, we have

Im ∂n+1 = Im dn+1 , since dn+1 = jn ◦ ∂n+1 and jn is injective.
Altogether, we have

Hn ( X ) ∼
= Hn ( Xn )/Im ∂n+1 = ker dn /Im dn+1 = HnCW ( X ),

thus proving the theorem.

Let us now discuss some immediate consequences of the above

theorem.

(a) If X has no n-cells, then Hn ( X ) = 0.

Indeed, in this case we have Cn = Hn ( Xn , Xn−1 ) = 0. Therefore,
HnCW ( X ) = 0.

(b) If X is connected and has a single 0-cell then d1 : C1 → C0 is the zero

map.
Indeed, since X contains only a single 0-cell, C0 = Z. Also, since X
is connected, H0 ( X ) = Z. So by the above theorem, Z = H0 ( X ) =
ker d0 /Im d1 = Z/Im d1 . This implies that Im d1 = 0, so d1 is the
zero map as desired.

(c) If X has no cells in adjacent dimensions then dn = 0 for all n and

Hn ( X ) ∼
= Z # n-cells for all n.
Indeed, in this case all maps dn vanish. So for any n, HnCW ( X ) ∼
=
∼
Cn = Z # n-cells .

Example 5.5.13. Recall that CPn has one cell in each even dimension
0, 2, 4, . . . , 2n. So CPn has no two cells in adjacent dimensions, meaning
we can apply Consequence (c) above to obtain:

Z, i = 0, 2, 4, . . . , 2n
Hi (CPn ) =
0, otherwise.

Example 5.5.14. When n > 1, Sn × Sn has one 0-cell, two n-cells, and
one 2n-cell. Since n > 1, these cells are not in adjacent dimensions so
again Consequence (c) above applies to give:

Z i = 0, 2n


n n
Hi (S × S ) = Z2 i=n


0 otherwise.


We next discuss how to compute in general the maps

dn : Cn ( X ) = Z # n-cells → Cn−1 ( X ) = Z # (n-1)-cells

96 algebraic topology

of the cellular chain complex. Let us consider the n-cells {eαn }α as the
basis for Cn ( X ) and the (n − 1)-cells {enβ−1 } β as the basis for Cn−1 ( X ).
In particular, we can write:

dn (eαn ) = ∑ dαβ · enβ−1 ,

β

with dαβ ∈ Z. The following result provides a way of computing the

coefficients dαβ :
Theorem 5.5.15. The coefficient dαβ is equal to the degree of the map
∆αβ : Sαn−1 → Snβ−1 defined by the composition:

φnα
Sαn−1 = ∂Dαn −→ Xn−1 = Xn−2 ⊔γ eγn−1
collapse
−−−−→ Xn−1 /( Xn−2 ⊔γ̸= β eγn−1 ) = Snβ−1 ,

where φnα is the attaching map of eαn , and the collapsing map sends Xn−2 ⊔γ̸= β
eγn−1 to a point.

enβ−1
X n −1

X n −2

Proof. We will proceed with the proof by chasing the following diagram:
(∆αβ )∗
Hn ( Dαn , Sαn−1 )
∂ /H
e n−1 (Sαn−1 ) /H
e n −1 ( S n −1 )
≃
O β
(Φnα )∗ ( φnα )∗ q β∗
  q∗
Hn ( Xn , Xn−1 )
∂n
/H
e n −1 ( X n −1 ) /H
e n−1 ( Xn−1 /Xn−2 )

dn
jn−1 ≃

' 
Hn−1 ( Xn−1 , Xn−2 )
≃ / Hn−1 ( Xn−1 , Xn−2 )
X n −2 X n −2

where:

• Φnα is the characteristic map of the cell eαn and φnα is its attaching map.

• q∗ : He n −1 ( X n −1 ) → H e n−1 ( D n−1 /∂D n−1 )

e n−1 ( Xn−1 /Xn−2 ) = L β H
β β
is induced by the quotient map q : Xn−1 → Xn−1 /Xn−2 .

• q β : Xn−1 /Xn−2 → Snβ−1 collapses the complement of the cell enβ−1

to a point, the resulting quoting sphere being identified with Snβ−1 =
Dβn−1 /∂Dβn−1 via the characteristic map Φnβ−1 .
 homology 97

• ∆αβ : Sαn−1 = ∂Dαn → Snβ−1 is the composition q β ◦ q ◦ φnα , i.e., the

attaching map of eαn followed by the quotient map Xn−1 → Snβ−1
collapsing the complement of enβ−1 in Xn−1 to a point.

Note that (∆αβ )∗ is defined so that the top right square commutes.
Recall that our goal is to compute dn (eαn ). The upper left square is
natural and therefore commutes (it is induced by the characteristic map
Φ : ( D ∗ , S∗−1 ) → ( X∗ , X∗−1 ) of a cell), while the lower left triangle
is part of the exact diagram defining the chain complex C∗ ( X ) and
is defined to commute as well. The map (Φnα )∗ takes the generator
[ Dαn ] ∈ Hn ( Dαn , Sαn−1 ) to a generator of the Z-summand of Hn ( Xn , Xn−1 )
corresponding to eαn , i.e.,

(Φnα )∗ ([ Dαn ]) = eαn .

Since the top left square and the bottom left triangle both commute,
this gives that

dn (eαn ) = dn ◦ (Φnα )∗ ([ Dαn ]) = jn−1 ◦ ( φnα )∗ ◦ ∂([ Dαn ]).

Looking to the bottom right square, recall that since X is a CW complex,

( Xn , Xn−1 ) is a good pair. This gives the isomorphism

Hn−1 ( Xn−1 , Xn−2 ) ≃ H

e n−1 ( Xn−1 /Xn−2 ).

Moreover, we also have that

e n−1 ( Xn−1 /Xn−2 ) ≃ Hn−1 ( Xn−1 /Xn−2 , Xn−2 /Xn−2 ).
H

The bottom right square commutes by the definition of jn−1 and q∗ ,

which combined with the commutativity of the top left square yields
that
dn (eαn ) = q∗ ◦ ∂n ◦ (Φnα )∗ ([ Dαn ]) = q∗ ◦ ( φnα )∗ ◦ ∂([ Dαn ]),
where formally we should precompose on the left hand side with the
isomorphism between Hn−1 ( Xn−1 , Xn−2 ) and H e n−1 ( Xn−1 /Xn−2 ) so
that everything is in the same space. This last map takes the generator
e n−1 ( D n−1 /∂D n−1 ).
[ Dαn ] to a linear combination of generators in ⊕ β H β β
To see which generators it maps to, we project down to the respective β
summands to obtain

dn (eαn ) = ∑ qβ∗ ◦ q∗ ◦ ( φnα )∗ ◦ ∂([ Dαn ]).

β

As noted before, we have defined (∆αβ )∗ = q β∗ ◦ q∗ ◦ ( φnα )∗ . So writing

dn (eαn ) = ∑(∆αβ )∗ ∂([ Dαn ]),

β

we see from the definition of the above maps and the fact that ∂([ Dαn ])
e n−1 (Sαn−1 ), that (∆αβ )∗ is multiplication by dαβ .
is a generator of H
98 algebraic topology

Example 5.5.16. Let Mg be the closed oriented surface of genus g, with

its usual CW structure: one 0-cell, 2g 1-cells { a1 , b1 , · · · a g , bg }, and
one 2-cell attached by product of commutators [ a1 , b1 ] · · · [ a g , bg ]. The
associated cellular chain complex of Mg is:

d3 d2 d1 d0
0 /Z / Z2g /Z /0

Since Mg is connected and has only one 0-cell, we get that d1 = 0.

We claim that d2 is also the zero map. This amounts to showing that
d2 (e) = 0, where e denotes the 2-cell. Indeed, let us compute the
coefficients deai and debi in our degree formula. As the attaching map
sends the generator to a1 b1 a1−1 b1−1 ...a g bg a− 1 −1
g b g , when we collapse all
1-cells (except ai , resp., bi ) to a point, the word defining the attaching
map a1 b1 a1−1 b1−1 ...a g bg a− 1 −1 −1
g b g reduces to ai ai and, resp., bi bi−1 . Hence
deai = 1 − 1 = 0. Similarly, debi = 1 − 1 = 0, for each i. Altogether,

d2 (e) = a1 + b1 − a1 − b1 + · · · a g + bg − a g − bg = 0.

So the homology groups of Mg are given by


Z i=0, 2


Hn ( Mg ) = Z2g i=1


0 otherwise.


Example 5.5.17. Let Ng be the closed nonorientable surface of genus g,

with its cell structure consisting of one 0-cell, g 1-cells { a1 , · · · , a g }, and
one 2-cell e attached by the word a21 · · · a2g . The cellular chain complex
of Ng is given by

d3 d2 d1 d0
0 /Z / Zg /Z /0

As before, d1 = 0 since Ng is connected and there is only one cell in

dimension zero. To compute d2 : Z → Zg we again apply the cellular
boundary formula, and obtain

d2 (1) = (2, 2, · · · , 2)

since each ai appears in the attaching word with total exponent 2,

which means that each map ∆αβ is homotopic to the map z 7→ z2 of
degree 2. In particular, d2 is injective, hence H2 ( Ng ) = 0. If we change
the standard basis for Zg by replacing the last standard basis element
en = (0, · · · , 0, 1) by en′ = (1, · · · , 1), then d2 (1) = 2 · en′ , so

H1 ( Ng ) ∼
= Zg /Im d2 ∼
= Zg /2Z ∼
= Zg−1 ⊕ Z/2.

Altogether,
 homology 99


Z i=0


Hn ( Ng ) = Zg−1 ⊕ Z2 i=1


0 otherwise.


Example 5.5.18. Recall that RPn has a CW structure with one cell ek in
each dimension 0 ≤ k ≤ n. Moreover, the attaching map of ek in RPn
is the two-fold cover projection φ : Sk−1 → RPk−1 . The cellular chain
complex for RPn looks like:

d n +1 dn d2 d1 d0
0 /Z / ··· /Z /Z /0

To compute the differential dk , we need to compute the degree of the

composite map
φ q
∆ : Sk−1 −→ RPk−1 −→ RPk−1 /RPk−2 = Sk−1 .

The map ∆ is a homeomorphism when restricted to each component of

Sk−1 \ Sk−2 , and these homeomorphisms are obtained from each other
by precomposing with the antipodal map a of Sk−1 , which has degree
(−1)k . Hence, by our local degree formula, we get that:

deg ∆ = deg id + deg a = 1 + (−1)k .

In particular, 
0 if k is odd
dk =
2 if k is even,

and therefore we obtain that


Z2 if k is odd , 0 < k < n


n
Hk (RP ) = Z k = 0, and k = n odd


0 otherwise.


Finally, note that an equivalent definition of the above map ∆ is obtained

by first collapsing the equatorial Sk−2 to a point to get Sk−1 ∨ Sk−1 , and
then mapping the two copies of Sk−1 onto Sk−1 , the first one by the
identity map, and the second by the antipodal map (see Figure 5.2).

Exercises
1. Describe a cell structure on Sn ∨ Sn ∨ · · · ∨ Sn and calculate H∗ (Sn ∨
S n ∨ · · · ∨ S n ).

2. Let f : Sn → Sn be a map of degree m. Let X = Sn ∪ f D n+1 be a

space obtained from Sn by attaching a (n + 1)-cell via f . Compute the
homology of X.
 100 algebraic topology

S k −1 Figure 5.2: The map ∆

S k −2

q
∆
S k −1

S k −1
id

S k −1

3. Let G be a finitely generated abelian group, and fix n ≥ 1. Construct a

CW-complex X such that Hn ( X ) ∼ = G and H̃i ( X ) = 0 for all i ̸= n. (Hint:
Use the calculation of the previous exercise, together with know facts
from Algebra about the structure of finitely generated abelian groups.)
More generally, given finitely generated abelian groups G1 , G2 , · · · , Gk ,
construct a CW-complex X whose homology groups are Hi ( X ) = Gi ,
i = 1, · · · , k, and H̃i ( X ) = 0 for all i ∈
/ {1, 2, · · · , k}.

4. Show that RP5 and RP4 ∨ S5 have the same homology and funda-
mental group. Are these spaces homotopy equivalent?

5. Let 0 ≤ m < n. Compute the homology of RPn /RPm .

6. The mapping torus T f of a map f : X → X is the quotient of X × I

X×I
Tf = .
( x, 0) ∼ ( f ( x ), 1)

Let A and B be copies of S1 , let X = A ∨ B, and let p be the wedge

point of X. Let f : X → X be a map that satisfies f ( p) = p, carries A
into A by a degree–3 map, and carries B into B by a degree–5 map.

(a) Equip T f with a CW structure by attaching cells to X ∨ S1 .

(b) Compute a presentation of π1 ( T f ).

(c) Compute H1 ( T f ; Z).

 homology 101

7. The closed oriented surface Mg of genus g, embedded in R3 in the

standard way, bounds a compact region R. Two copies of R, glued
together by the identity map between their boundary surfaces Mg ,
form a space X. Compute the homology groups of X and the relative
homology groups of ( R, Mg ).

8. Let X be the space obtained by attaching two 2-cells to S1 , one via

the map z 7→ z3 and the other via z 7→ z5 , where z denotes the complex
coordinate on S1 ⊂ C.

(a) Compute the homology of X with coefficients in Z.

(b) Is X homeomorphic to the 2-sphere S2 ? Justify your answer!

9. Homology of Lens Spaces.

Given m > 1 and integers l1 , · · · , ln so that (lk , m) = 1 for all k, define
the Lens space L = Lm (l1 , · · · , ln ) to be the orbit space S2n−1 /Zm of the
unit sphere S2n−1 with the Zm -action generated by the rotation:
 
ρ(z1 , · · · , zn ) = e2πil1 /m z1 , · · · , e2πiln /m zn ,

rotating the j-th C-factor of Cn by an angle 2πil j /m. (In particular,

when m = 2, ρ is the antipodal map, so L = RP2n−1 .)

(a) Show that one can construct a CW-structure on L with one cell ek in
each dimension k ≤ 2n − 1.

(b) Compute the differentials dk of the resulting cellular chain complex.

(c) Compute the homology of L.

5.6 Euler Characteristic

Definition 5.6.1. Let X be a finite CW complex of dimension n and denote

by ci the number of i-cells of X. The Euler characteristic of X is defined as:
n
χ( X ) = ∑ (−1)i · ci . (5.6.1)
i =0

It is natural to question whether or not the Euler characteristic

depends on the cell structure chosen for the space X. As we will see
below, this is not the case. For this, it suffices to show that the Euler
characteristic depends only on the cellular homology of the space X.
Indeed, cellular homology is isomorphic to singular homology, and the
latter is independent of the cell structure on X.
Recall that if G is a finitely generated abelian group, then G decom-
poses into a free part and a torsion part, i.e.,

G ≃ Zr × Z n 1 × · · · Z n k .
102 algebraic topology

The integer r := rk( G ) is the rank of G. The rank is additive in short

exact sequences of finitely generated abelian groups.

Theorem 5.6.2. The Euler characteristic can be computed as:

n
χ( X ) = ∑ (−1)i · bi (X ) (5.6.2)
i =0


with bi ( X ) := rk Hi ( X ) the i-th Betti number of X. In particular, χ( X ) is
independent of the chosen cell structure on X.

Proof. We use the following notation: Bi = Im(di+1 ), Zi = ker(di ), and

Hi = Zi /Bi . Consider a (finite) chain complex of finitely generated
abelian groups and the short exact sequences defining homology:

d n +1 dn d2 d1 d0
0 / Cn / ... / C1 / C0 /0


0 / Zi  ι / Ci di
/ / Bi−1 /0

d i +1 q
0 / Bi / Zi / Hi /0

The additivity of rank yields that

ci := rk(Ci ) = rk( Zi ) + rk( Bi−1 )

and
rk( Zi ) = rk( Bi ) + rk( Hi ).

Substitute the second equality into the first, multiply the resulting
equality by (−1)i , and sum over i to get that χ( X ) = ∑in=0 (−1)i · rk( Hi ).
Finally, we apply this result to the cellular chain complex Ci =
Hi ( Xi , Xi−1 ), and use the identification between cellular and singular
homology.

Example 5.6.3. If Mg and Ng denote the orientable and, resp., nonori-

entable closed surfaces of genus g, then χ( Mg ) = 1 − 2g + 1 = 2(1 − g)
and χ( Ng ) = 1 − g + 1 = 2 − g. So all the orientable and, resp., non-
orientable surfaces are distinguished from each other by their Euler
characteristic, and there are only the relations χ( Mg ) = χ( N2g ).

Exercises
1. A graded abelian group is a sequence of abelian groups A• :=
( An )n≥0 . We say that A• is of finite type if

∑ rankAn < ∞.
n ≥0
 homology 103

The Euler characteristic of a finite type graded abelian group A• is the

integer
χ( A• ) := ∑ (−1)n · rankAn .
n ≥0
A short exact sequence of graded groups A• , B• , C• , is a sequence of
short exact sequences

0 → An → Bn → Cn → 0, n ≥ 0.

Prove that if 0 → A• → B• → C• → 0 is a short exact sequence of

graded abelian groups of finite type, then

χ( B• ) = χ( A• ) + χ(C• ).

2. Suppose we are given three finite type graded abelian groups A• , B• ,

C• , which are part of a long exact sequence
ik jk ∂
· · · → Ak → Bk → Ck →k Ak−1 → · · · → A0 → B0 → C0 → 0.

Show that
χ( B• ) = χ( A• ) + χ(C• ).

3. For finite CW complexes X and Y, show that

χ ( X × Y ) = χ ( X ) · χ (Y ) .

4. If a finite CW complex X is a union of subcomplexes A and B, show

that
χ ( X ) = χ ( A ) + χ ( B ) − χ ( A ∩ B ).

5. For a finite CW complex and p : Y → X an n-sheeted covering space,

show that
χ (Y ) = n · χ ( X ) .

6. Show that if f : RP2n → Y is a covering map of a CW-complex Y,

then f is a homeomorphism.

5.7 Lefschetz Fixed Point Theorem

Let G be a finitely generated abelian group. Given an endomorphism

φ : G → G, define its trace by

Tr( φ) = Tr ( φ̄ : G/Torsion( G ) → G/Torsion( G )) (5.7.1)

where the latter trace is the linear algebraic trace of the map φ̄ : Zr →
Zr , with r = rk( G ). It is a fact that the trace is independent of the
choice of a basis for Zr .
104 algebraic topology

Definition 5.7.1. If X has the homotopy type of a finite CW complex and

f : X → X is a continuous map, then the Lefschetz number of f is defined as:

dim( X )
τ( f ) = ∑ (−1)i · Tr( f ∗ : Hi ( X ) → Hi ( X )). (5.7.2)
i =0

Remark 5.7.2. Notice that homotopic maps have the same Lefschetz
number since they induce the same maps on homology.

Example 5.7.3. If f ≃ id X , then τ ( f ) = χ( X ). This follows from the fact

that the map induced in homology by the identity map is the identity
homomorphism, and the trace of the latter is the corresponding Betti
number of X.
Theorem 5.7.4. (Lefschetz)
If X is a retract of a finite CW complex and if the continuous map f : X → X
satisfies τ ( f ) ̸= 0, then f has a fixed point.

Before sketching the proof of this theorem, let us consider a few

examples.

Example 5.7.5. Suppose that X has the homology of a point (up to

torsion). Then


τ ( f ) = Tr f ∗ : H0 ( X ) → H0 ( X ) = 1.

This follows from the fact that all the other homology groups are zero
and that the map induced on H0 is the identity.

This example leads immediately to two nontrivial results, the first of

which is the Brower fixed point theorem.
Example 5.7.6. (Brower) If f : D n → D n is continuous then f has a
fixed point.
Example 5.7.7. If X = RP2n , then modulo torsion X has the homology
of a point. Therefore any continuos map f : RP2n → RP2n has a fixed
point.
Finally we are led to an example which does not follow from the
computation for a point.
Example 5.7.8. If f : Sn → Sn is a continuous map and deg( f ) ̸=
(−1)n+1 , then f has a fixed point. To verify this, we compute

τ ( f ) = Tr( f ∗ : H0 (Sn ) → H0 (Sn )) + (−1)n · Tr( f ∗ : Hn (Sn ) → Hn (Sn ))

= 1 + (−1)n · deg( f )
̸= 0.

Corollary 5.7.9. If a : Sn → Sn is the antipodal map, then deg( a) =

(−1)n+1 .
 homology 105

Now we return to outlining the proof.

Definition 5.7.10. A map f : X → Y between CW complexes is called

cellular if f ( Xn ) ⊆ Yn for all n, with Xn denoting the n-skeleton of X and
similarly for Y.

We’ll need the following fundamental result from homotopy theory.

Theorem 5.7.11 (Cellular Approximation). Any continuous map f : X →

Y between CW complexes is homotopic to a cellular map.

The proof of this result is omitted for now. We proceed with sketch-
ing the proof of the Lefschetz theorem.

Proof. (sketch)
The general case reduces to the case when X is a finite CW complex.
Indeed, if r : K → X is a retraction of a finite CW complex K onto X,
the composition f ◦ r : K → X ⊂ K has exactly the same fixed points as
f and since r∗ : Hi (K ) → Hi ( X ) is projection onto a direct summand,
we have that Tr( f ∗ ◦ r∗ ) = Tr( f ∗ ), so τ ( f ◦ r ) = τ ( f ). We can therefore
assume that X is a finite CW complex.
Let us suppose that f has no fixed points.
By cellular approximation, the map f : X → X is homotopic to
a cellular map g : X → X. In particular, τ ( f ) = τ ( g). Moreover,
since f ( x ) ̸= x for all x ∈ X, it is possible to choose the cellular map
g : X → X so that g(eiλ ) ∩ eiλ = ∅, for all i and λ. Since the {eiλ }λ
generate Ci ( X ) := Hi ( Xi , Xi−1 ), we get that

∑(−1)i · Tr( g∗ : Ci (X ) → Ci (X )) = 0.
i

Furthermore, using the fact that the trace is additive for short exact
sequences, if follows as in the case of the Euler characteristic (Theorem
5.6.2) that
τ ( g) = ∑(−1)i · Tr( g∗ : Ci (X ) → Ci (X )).
i

Altogether, we get that τ ( f ) = τ ( g) = 0, which is a contradiction.

Exercises
1. Is there a continuous map f : RP2k−1 → RP2k−1 with no fixed
points? Explain.

1. Is there a continuous map f : CP2k−1 → CP2k−1 with no fixed points?

Explain. We will see later that any map f : CP2k → CP2k has a fixed
point.
 106 algebraic topology

5.8 Homology with arbitrary coefficients

Tensor Products
Let A, B be abelian groups. Define the abelian group

A ⊗ B = ⟨ a ⊗ b | a ∈ A, b ∈ B⟩/ ∼ (5.8.1)

where ∼ is generated by the relations ( a + a′ ) ⊗ b = a ⊗ b + a′ ⊗ b and

a ⊗ (b + b′ ) = a ⊗ b + a ⊗ b′ . The zero element of A ⊗ B is 0 ⊗ b =
a ⊗ 0 = 0 ⊗ 0 = 0 A⊗ B since, e.g., 0 ⊗ b = (0 + 0) ⊗ b = 0 ⊗ b + 0 ⊗ b so
0 ⊗ b = 0 A⊗ B . Similarly, the inverse of an element a ⊗ b is −( a ⊗ b) =
(− a) ⊗ b = a ⊗ (−b) since, e.g., 0 A⊗ B = 0 ⊗ b = ( a + (− a)) ⊗ b =
a ⊗ b + (− a) ⊗ b.
Lemma 5.8.1. The tensor product satisfies the following universal property
which asserts that if φ : A × B → C is any bilinear map, then there exists a
unique map φ : A ⊗ B → C such that φ = φ ◦ i, where i : A × B → A ⊗ B
is the natural map ( a, b) 7→ a ⊗ b.

A×B
i / A⊗B
φ
∃! φ
$ 
C
Proof. Indeed, φ : A ⊗ B → C can be defined by a ⊗ b 7→ φ( a, b).

Proposition 5.8.2. The tensor product satisfies the following properties:

(1) A ⊗ B ∼
= B ⊗ A via the isomorphism a ⊗ b 7→ b ⊗ a.

Ai ) ⊗ B ∼
= i ( Ai ⊗ B) via the isomorphism ( ai )i ⊗ b 7→ ( ai ⊗ b)i .
L L
(2) ( i

(3) A ⊗ ( B ⊗ C ) ∼
= ( A ⊗ B) ⊗ C via the isomorphism a ⊗ (b ⊗ c) 7→ ( a ⊗
b) ⊗ c.

(4) Z ⊗ A ∼
= A via the isomorphism n ⊗ a 7→ na.

(5) Z/nZ ⊗ A ∼
= A/nA via the isomorphism l ⊗ a 7→ la.

Proof. These are easy to prove by using the above universal property.
We sketch a few.
(1) The map φ : A × B → B ⊗ A defined by ( a, b) 7→ b ⊗ a is clearly
bilinear and therefore induces a homomorphism φ : A ⊗ B → B ⊗ A
with a ⊗ b 7→ b ⊗ a. Similarly, there is the reverse map ψ : B × A →
A ⊗ B defined by (b, a) 7→ a ⊗ b which induces a homomorphism
ψ : B ⊗ A → A ⊗ B with b ⊗ a 7→ a ⊗ b. Clearly, φ ◦ ψ = id B⊗ A and
ψ ◦ φ = id A⊗ B and A ⊗ B ∼
= B ⊗ A.
(4) The map φ : Z × A → A defined by (n, a) 7→ na is a bilinear
map and therefore induces a homomorphism φ : Z ⊗ A → A with
 homology 107

n ⊗ a 7→ na. Now suppose φ(n ⊗ a) = 0. Then na = 0 and n ⊗ a =

1 ⊗ (na) = 1 ⊗ 0 = 0Z⊗ A . Thus φ is injective. Moreover, if a ∈ A, then
φ(1 ⊗ a) = a and φ is surjective as well.
(5) The map φ : Z/nZ × A → A/nA defined by (l, a) 7→ la is a
bilinear map and therefore induces a homomorphism φ : Z/nZ ⊗ A →
A/nA with l ⊗ a 7→ la. Now suppose φ(l ⊗ a) = la = 0. Then
la = ∑ik=1 nai and l ⊗ a = 1 ⊗ (la) = 1 ⊗ (∑ik=1 nai ) = ∑ik=1 (n ⊗ ai ) =
0Z/nZ⊗ A , so φ is injective. Now let a ∈ A/nA. Then φ(1 ⊗ a) = a and
φ is surjective as well.

More generally, if R is a ring and A and B are R-modules, a tensor

product A ⊗ R B can be defined as follows:

(1) if R is commutative, define the R-module A ⊗ R B := A ⊗ B/ ∼,

where ∼ is the relation generated by ra ⊗ b = a ⊗ rb = r ( a ⊗ b).

(2) if R is not commutative, we need A a right R-module and B a left

R-module and the relation is ar ⊗ b = a ⊗ rb. In this case A ⊗ R B is
only an abelian group.

In both cases, A ⊗ R B is not necessarily isomorphic to A ⊗ B.

√ √
Example 5.8.3. Let R = Q[ 2] = { a + b 2 | a, b ∈ Q}. Now R ⊗ R R ∼
=
R which is a 2-dimensional Q-vector space. However, R ⊗ R as a
Z-module is a 4-dimensional Q-vector space.

Lemma 5.8.4. If G is an abelian group, then the functor − ⊗ G is right

i j i ⊗1 G j ⊗1 G
exact, that is, if A −
→B−
→ C → 0 is exact, then A ⊗ G −−→ B ⊗ G −−−→
C ⊗ G → 0 is exact.

Proof. Let c ⊗ g ∈ C ⊗ G. Since j is onto, there exists, b ∈ B such that

j(b) = c. Then ( j ⊗ 1G )(b ⊗ g) = c ⊗ g and j ⊗ 1G is onto.

Since j ◦ i = 0, we have ( j ⊗ 1G ) ◦ (i ⊗ 1G ) = ( j ◦ i ) ⊗ 1G = 0 and

thus, Im(i ⊗ 1G ) ⊆ ker( j ⊗ 1G ).

It remains to show that ker( j ⊗ 1G ) ⊆ Im(i ⊗ 1G ). It is enough to

show that
∼
=
ψ : B ⊗ G/Im(i ⊗ 1G ) −
→ C ⊗ G,
where ψ is the map induced by j ⊗ 1G . Construct an inverse of ψ,
induced from the homomorphism

φ : C × G → B ⊗ G/Im(i ⊗ 1G )

defined by (c, g) 7→ b ⊗ g, where j(b) = c. We must show that φ

is a well-defined bilinear map and that the induced map φ satisfies
108 algebraic topology

φ ◦ ψ = id and ψ ◦ φ = id.

If j(b) = j(b′ ) = c, then b − b′ ∈ ker j = Im i, so b − b′ = i ( a) for some

a ∈ A. Thus, b ⊗ g − b′ ⊗ g = (b − b′ ) ⊗ g = i ( a) ⊗ g ∈ Im(i ⊗ 1G ). So
φ is well defined.

Now φ((c + c′ , g)) = d ⊗ g where j(d) = c + c′ . Since j is surjective,

choose b, b′ ∈ B such that j(b) = c and j(b′ ) = c′ . Then d − (b + b′ ) ∈
ker j = Im i and so there exists a ∈ A such that i ( a) = d − (b + b′ ).
Thus, φ((c + c′ , g)) = d ⊗ g = (b + b′ ) ⊗ g = b ⊗ g + b′ ⊗ g = φ(c, g) +
φ(c′ , g) and φ is linear in the first component. For the second compo-
nent, φ(c, g + g′ ) = b ⊗ ( g + g′ ) = b ⊗ g + b ⊗ g′ = φ(c, g) + φ(c, g′ ).
Thus, φ is bilinear.

Now by the universal property of the tensor product, the bilinear

map φ induces a homomorphism

φ : C ⊗ G → B ⊗ G/Im(i ⊗ 1G )

defined by c ⊗ g 7→ φ(c, g) = b ⊗ g, where j(b) = c. For c ⊗ g ∈ C ⊗ G,

ψ ◦ φ(c ⊗ g) = ψ(b ⊗ g) = j(b) ⊗ g = c ⊗ g,

so ψ ◦ φ = idC⊗G . Similarly, for b ⊗ g ∈ B ⊗ G/Im(i ⊗ 1G ), φ ◦ ψ(b ⊗

g) = φ( j(b) ⊗ g) = φ( j(b), g) = b ⊗ g. Thus φ ◦ ψ = id.

Remark 5.8.5. Tensoring with a free abelian group is an exact functor.

Homology with Arbitrary Coefficients

Let G be an abelian group and X a topological space. We define the
homology of X with G-coefficients, denoted H∗ ( X; G ), as the homology
of the chain complex

Ci ( X; G ) = Ci ( X ) ⊗ G (5.8.2)

consisting of finite formal sums ∑i ηi · σi (with σi : ∆i → X and ηi ∈ G),

and with boundary maps given by

∂iG := ∂i ⊗ idG .

Since ∂i satisfies ∂i ◦ ∂i+1 = 0 it follows that ∂iG ◦ ∂iG+1 = 0, so

(C∗ ( X; G ), ∂∗G )

forms indeed a chain complex. We can construct versions of the usual

modified homology groups (relative, reduced, etc.) in the natural way.
Define relative chains with G-coefficients by

Ci ( X, A; G ) := Ci ( X; G )/Ci ( A; G ),
 homology 109

and reduced homology with G-coefficients via the augmented chain

complex

∂iG+1 ∂Gi ∂G
2 1 ∂G ϵ
· · · −→ Ci ( X; G ) −→ · · · −→ C1 ( X; G ) −→ C0 ( X; G ) −→ G → 0,

where ϵ(∑i ηi σi ) = ∑i ηi ∈ G. Notice that Hi ( X ) = Hi ( X; Z) by defini-

tion.
By studying the chain complex with G-coefficients, it follows that

G i = 0
Hi ( pt; G ) =
0 i ̸= 0.

Nothing (other than coefficients) needs to change in describing the

relationships between relative homology and reduced homology of
quotient spaces, so we can compute the homology of a sphere as before
by induction and using the long exact sequence of the pair ( D n , Sn ) to
be 
 G i = 0, n
Hi (Sn ; G ) =
0 otherwise.

Finally, we can build cellular homology with G-coefficients in the

same way, defining

CiG ( X ) = Hi ( Xi , Xi−1 ; G ) ∼
= G# i−cells .

The cellular boundary maps are given by:

diG (∑ ηα eiα ) = ∑ ηα dαβ eiβ−1 ,

α α,β

where dαβ is as before the degree of a map ∆αβ : Si−1 → Si−1 . This
follows from the easy fact that if f : Sk → Sk has degree m, then
f ∗ : Hk (Sk ; G ) ≃ G → Hk (Sk ; G ) ≃ G is the multiplication by m. As it
is the case for integers, we get an isomorphism

HiCW ( X; G ) ≃ Hi ( X; G )

for all i.

Example 5.8.6. We compute Hi (RPn ; Z2 ) using the cellular homology

with Z2 -coefficients. Notice that over Z the cellular boundary maps
are di = 0 or di = 2 depending on the parity of i, and therefore with
Z2 -coefficients all of boundary maps vanish. Therefore,

Z 0 ≤ i ≤ n
2
Hi (RPn ; Z2 ) =
0 otherwise.
110 algebraic topology

Example 5.8.7. Fix n > 0 and let g : Sn → Sn be a map of degree m.

Define the CW complex

X = S n ∪ g e n +1 ,

where the (n + 1)-cell en+1 is attached to Sn via the map g. Let f be the
quotient map f : X → X/Sn . Define Y = X/Sn = Sn+1 . The homology
of X can be easily computed by using the cellular chain complex:

d n +2 d n +1 dn d1 d1 d0
0 /Z /Z / ... /0 /Z /0
m

Therefore, 
Z i=0


Hi ( X; Z) = Zm i=n


0 otherwise.


Moreover, as Y = Sn+1 , we have


Z i = 0, n + 1
Hi (Y; Z) =
0 otherwise.

It follows that f induces the trivial homomorphisms in homology with

Z-coefficients (except in degree zero, where f ∗ is the identity). So it is
natural to ask if f is homotopic to the constant map. As we will see
below, by considering Zm -coefficients we can show that this is not the
case.
Let us now consider H∗ ( X; Zm ) where m is, as above, the degree of
the map g. We return to the cellular chain complex level and observe
that we have
d n +2 d n +1 dn d1 d1 d0
0 / Zm / Zm / ... /0 / Zm /0
m

Multiplication by m is now the zero map, so we get


Z
m i = 0, n, n + 1
Hi ( X; Zm ) =
0 otherwise.

Also, as already discussed,


Z i = 0, n + 1
m
Hi (Y; Zm ) =
0 otherwise.

We next consider the induced homomorphism f ∗ : Hn+1 ( X; Zm ) →

Hn+1 (Y; Zm ). The claim is that this map is injective, thus non-trivial,
so f cannot be homotopic to the constant map. As noted before, we
have an isomorphism H e n+1 (Y; Zm ) ≃ Hn+1 ( X, Sn ; Zm ). This leads us
 homology 111

to consider the long exact sequence of the pair ( X, Sn ) in dimension

n + 1. We have
f∗
· · · −→ Hn+1 (Sn ; Zm ) −→ Hn+1 ( X; Zm ) −→ Hn+1 ( X, Sn ; Zm ) −→ · · ·

But, Hn+1 (Sn ; Zm ) = 0 and so f ∗ is injective on Hn+1 ( X; Zm ). Since

Hn+1 ( X; Zm ) = Zm ̸= 0 and Hn+1 ( X, Sn ; Zm ) ≃ H e n+1 (Y; Zm ) it fol-
lows that f ∗ is not trivial on Hn+1 ( X; Zm ), which proves our claim.

Exercises
1. Calculate the homology of the 2-torus T 2 with coefficients in Z, Z2
and Z3 , respectively. Do the same calculations for the Klein bottle.

5.9 The Tor functor and the Universal Coefficient Theorem

In this section, we explain how to compute H∗ ( X; G ) in terms of G and

H∗ ( X; Z). More generally, given a chain complex

∂n ∂ n −1 1 ∂
C• : · · · → Cn −→ Cn−1 −−→ · · · −
→ C0 → 0

of free abelian groups and G an abelian group, we aim to compute

H∗ (C• ; G ) := H∗ (C• ⊗ G ) in terms of H∗ (C• ; Z) and G. The answer is
provided by the following result:

Theorem 5.9.1. (Universal Coefficient Theorem)

For each n, there are natural short exact sequences:

0 → Hn (C• ) ⊗ G → Hn (C• ; G ) → Tor( Hn−1 (C• ), G ) → 0. (5.9.1)

Naturality here means that if C• → C•′ is a chain map, then there is an

induced map of short exact sequences with commuting squares. Moreover,
these short exact sequences split, but not naturally.

In particular, if C• = C• ( X, A) is the relative singular chain complex

of a pair ( X, A), then there are natural short exact sequences

0 → Hn ( X, A) ⊗ G → Hn ( X, A; G ) → Tor( Hn−1 ( X, A), G ) → 0.

(5.9.2)
f
Naturality is with respect to maps of pairs ( X, A) −
→ (Y, B). The exact
sequence (5.9.2) splits, but not naturally. Indeed, if we assume that
A = B = ∅, then we have splittings


Hn ( X; G ) = Hn ( X ) ⊗ G ⊕ Tor( Hn−1 ( X ), G ),


Hn (Y; G ) = Hn (Y ) ⊗ G ⊕ Tor( Hn−1 (Y ), G ).
 112 algebraic topology

If these splittings were natural, and f induces the trivial map f ∗ = 0

on H∗ (−; Z) then f induces the trivial map on H∗ (−; G ), for any coef-
ficient group G. But this is in contradiction with Example 5.8.7.

Let us next explain the Tor functor appearing in the statement of the
universal coefficient theorem.
Definition 5.9.2. A free resolution of an abelian group H is an exact sequence:
f2 f1 f0
· · · → F2 −
→ F1 −
→ F0 −
→ H → 0,

with each Fn a free abelian group.

Given an abelian group G, from a free resolution F• of H, we obtain

a modified chain complex:

F• ⊗ G : · · · → F2 ⊗ G → F1 ⊗ G → F0 ⊗ G → 0.

We define
Torn ( H, G ) := Hn ( F• ⊗ G ). (5.9.3)
Moreover, the following holds:

Lemma 5.9.3. For any two free resolutions F• and F•′ of H there are canonical
isomorphisms Hn ( F• ⊗ G ) ∼ = Hn ( F•′ ⊗ G ) for all n. Thus, Torn ( H, G ) is
independent of the free resolution F• of H used for its definition.
Proposition 5.9.4. For any abelian group H, we have that

Torn ( H, G ) = 0 if n > 1, (5.9.4)

and
Tor0 ( H, G ) ∼
= H ⊗ G. (5.9.5)

Proof. Indeed, given an abelian group H, take F0 to be the free abelian

f0
group on a set of generators of H to get F0 ↠ H → 0. Let F1 := ker( f 0 ),
and note that F1 is a free (and abelian) group, as it is a subgroup of a
free abelian group F0 . Let F1 ,→ F0 be the inclusion map. Then

0 → F1 ,→ F0 ↠ H → 0

is a free resolution of H. Thus, Torn ( H, G ) = 0 if n > 1. Moreover, it

follows readily that Tor0 ( H, G ) ∼
= H ⊗ G.

Definition 5.9.5. In what follows, we adopt the notation:

Tor( H, G ) := Tor1 ( H, G ).

Proposition 5.9.6. The Tor functor satisfies the following properties:

(1) Tor( A, B) ∼
= Tor( B, A).
 homology 113

Ai , B ) ∼
L L
(2) Tor( i = i Tor( Ai , B).

(3) Tor( A, B) = 0 if either A or B is free or torsion-free.

(4) Tor( A, B) ∼
= Tor(Torsion( A), B), where Torsion( A) is the torsion sub-
group of A.
n
(5) Tor(Z/nZ, A) ∼
= ker( A −
→ A ).

(6) For a short exact sequence: 0 → B → C → D → 0 of abelian groups, there

is a natural exact sequence:

0 −→ Tor( A, B) −→ Tor( A, C ) −→ Tor( A, D )

−→ A ⊗ B −→ A ⊗ C −→ A ⊗ D −→ 0.
L
Proof. (2) Choose a free resolution for i Ai as the direct sum of free
resolutions for the Ai ’s.

n
(5) The exact sequence 0 → Z −
→ Z → Z/nZ → 0 is a free resolution
of Z/nZ. Tensoring with A and dropping the right-most term yields
n ⊗1
the complex Z ⊗ A −−−→ A
Z ⊗ A → 0, which by property (4) of the
n n
→ A → 0. Thus, Tor(Z/nZ, A) = ker( A −
tensor product is A − → A ).

(3) If A is free, we can choose the free resolution:

F1 = 0 → F0 = A → A → 0

which implies that Tor( A, B) = 0. On the other hand, if B is free,

tensoring the exact sequence 0 → F1 → F0 → A → 0 with B = Zs gives
a direct sum of copies of 0 → F1 → F0 → A → 0. Hence, it is an exact
sequence and so H1 of this complex is 0. For the torsion free case, see
below.

(6) Let 0 → F1 → F0 → A → 0 be a free resolution of A, and tensor

it with the short exact sequence 0 → B → C → D → 0 to get a
commutative diagram:

0 0 0

0 F1 ⊗ B F1 ⊗ C F1 ⊗ D 0

0 F0 ⊗ B F0 ⊗ C F0 ⊗ D 0

0 0 0

Rows are exact since tensoring with a free group preserves exactness.
Thus we get a short exact sequence of chain complexes. Recall now
114 algebraic topology

that for any short exact sequence of chain complexes 0 → B• → C• →

D• → 0, there is an associated long exact sequence of homology groups

· · · → Hn (B• ) → Hn (C• ) → Hn (D• ) → Hn−1 (B• ) → . . .

So in our situation, with B• = F• ⊗ B, C• = F• ⊗ C and D• = F• ⊗ D,

we obtain the homology long exact sequence:

0 → H1 ( F• ⊗ B) → H1 ( F• ⊗ C ) → H1 ( F• ⊗ D )
→ H0 ( F• ⊗ B) → H0 ( F• ⊗ C ) → H0 ( F• ⊗ D ) → 0

Since H1 ( F• ⊗ B) = Tor( A, B) and H0 ( F• ⊗ B) = A ⊗ B, the above long

exact sequence reduces to:

0 → Tor( A, B) → Tor( A, C ) → Tor( A, D )

→ A ⊗ B → A ⊗ C → A ⊗ D → 0.

(1) Apply (6) to a free resolution 0 → F1 → F0 → B → 0 of B, and get

a long exact sequence:

0 → Tor( A, F1 ) → Tor( A, F0 ) → Tor( A, B)

→ A ⊗ F1 → A ⊗ F0 → A ⊗ B → 0.

Because F1 , F0 are free, by (3) we have that Tor( A, F1 ) = Tor( A, F0 ) = 0,

so the long exact sequence becomes:

0 → Tor( A, B) → A ⊗ F1 → A ⊗ F0 → A ⊗ B → 0.

Also, by definition of Tor, we have a long exact sequence:

0 → Tor( B, A) → F1 ⊗ A → F0 ⊗ A → B ⊗ A → 0.

So we get a diagram:

0 Tor( A, B) A ⊗ F1 A ⊗ F0 A⊗B 0
ϕ ≃ ≃ ≃

0 Tor( B, A) F1 ⊗ A F0 ⊗ A B⊗A 0

with the arrow labeled ϕ defined as follows. The two squares on the
right commute since ⊗ is naturally commutative. Hence, there exists
ϕ : Tor( A, B) → Tor( B, A) which makes the left square commutative.
Moreover, by the 5-lemma, we get that ϕ is an isomorphism.

We can now prove the torsion free case of (3). Assume that B is torsion
f
free. Let 0 → F1 → F0 → A → 0 be a free resolution of A. The claim
about the vanishing of Tor( A, B) is equivalent to the injectivity of the
map f ⊗ id B : F1 ⊗ B → F0 ⊗ B. Assume ∑i xi ⊗ bi ∈ ker( f ⊗ id B ). So
 homology 115

∑i f ( xi ) ⊗ bi = 0 ∈ F1 ⊗ B. In other words, ∑i f ( xi ) ⊗ bi can be reduced

to zero by a finite number of applications of the defining relations for
tensor products. Only a finite number of elements of B, generating a
finitely generated subgroup B0 of B, are involved in this process, so in
fact ∑i xi ⊗ bi ∈ ker( f ⊗ id B0 ). But B0 is finitely generated and torsion
free, hence free, so Tor( A, B0 ) = 0. Thus ∑i xi ⊗ bi = 0, which proves
the claim. The case when A is torsion free follows now by using (1) to
reduce to the previous case.

(4) Apply (6) to the short exact sequence: 0 → Torsion( A) → A →

A/Torsion( A) → 0 to get:

0 → Tor( B, Torsion( A)) → Tor( B, A) → Tor( B, A/Torsion( A)) → · · ·

Because A/Torsion( A) is torsion free, Tor( B, A/Torsion( A)) = 0 by

(3), so:
Tor( B, Torsion( A)) ≃ Tor( G, A)
Now by (1), we get that Tor( A, B) ≃ Tor(Torsion( A), B).

Remark 5.9.7. It follows from (5) that

Z
Tor(Z/nZ, Z/mZ) = = Z/nZ ⊗ Z/mZ,
(n, m)Z
where (n, m) is the greatest common divisor of n and m. More generally,
if A and B are finitely generated abelian groups, then

Tor( A, B) = Torsion( A) ⊗ Torsion( B) (5.9.6)

where Torsion( A) and Torsion( B) are the torsion subgroups of A and

B respectively.

Let us conclude with some examples:

Example 5.9.8. Suppose G = Q, then Tor( Hn−1 ( X ), Q) = 0, so

Hn ( X; Q) ≃ Hn ( X ) ⊗ Q.

It follows that the n-th Betti number of X is given by

bn ( X ) := rkHn ( X ) = dimQ Hn ( X; Q).

Example 5.9.9. Suppose X = T 2 , and G = Z/4. Recall that H1 ( T 2 ) =

Z2 . So:
H0 ( T 2 ; Z/4) = H0 ( T 2 ) ⊗ Z/4 = Z/4

H1 ( T 2 ; Z/4) = H1 ( T 2 ) ⊗ Z/4 ⊕ Tor( H0 ( T 2 ), Z/4)

= Z2 ⊗ Z/4 = (Z/4)2

H2 ( T 2 ; Z/4) = H2 ( T 2 ) ⊗ Z/4 ⊕ Tor( H1 ( T 2 ), Z/4) = Z/4.

116 algebraic topology

Example 5.9.10. Suppose X = K is the Klein bottle, and G = Z/4.

Recall that H1 (K ) = Z ⊕ Z/2, and H2 (K ) = 0, so:

H2 (K; Z/4) = H2 (K ) ⊗ Z/4 ⊕ Tor( H1 (K ), Z/4)

= Tor(Z, Z/4) ⊕ Tor(Z/2, Z/4)

= 0 ⊕ Z/2
= Z/2.

Exercises
1. Prove Lemma 9.2.

2. Show that He n ( X; Z) = 0 for all n if, and only if, H

e n ( X; Q) = 0 and
Hn ( X; Z/p) = 0 for all n and for all primes p.
e
 basics of cohomology 117

6
Basics of Cohomology

Given a space X and an abelian group G, in this chapter we define

cohomology groups H i ( X; G ) by “dualizing” the definition of homol-
ogy, and study their properties and methods of computation. In the
next chapter we will show that, via the cup product operation, the
graded group i H i ( X; G ) becomes a ring. The ring structure will
L

help us distinguish spaces X and Y which have isomorphic homology

and cohomology groups but non-isomorphic cohomology rings, for
example X = CP2 and Y = S2 ∨ S4 .

6.1 Cohomology of a chain complex: definition

Let G be an abelian group, and let (C• , ∂• ) be a chain complex of free

abelian groups:
∂ n +1
n ∂ ∂ n −1
· · · −→ Cn+1 −→ Cn −→ Cn−1 −→ · · · (6.1.1)

By dualizing the chain complex (6.1.1), i.e., by applying Hom(−; G ) to

it, one gets the cochain complex:
δ n +1 δn δ n −1
· · · ←− C n+1 ←− C n ←− C n−1 ←− · · · (6.1.2)

with
C n := Hom(Cn , G ), (6.1.3)
and where the coboundary map

δ n : C n → C n +1 (6.1.4)

is defined by

(δn ψ)(α) = ψ(∂n+1 α), for ψ ∈ C n and α ∈ Cn+1 . (6.1.5)

It follows that

(δn+1 ◦ δn )(ψ) = ψ(∂n+1 ◦ ∂n+2 ) = 0, ∀ψ, (6.1.6)

since ∂n+1 ◦ ∂n+2 = 0 in the chain complex (6.1.1). We can therefore

make the following.
118 algebraic topology

Definition 6.1.1. The n-th cohomology group H n (C• ; G ) with G-coefficients

of the chain complex C• is defined by:

H n (C• ; G ) := H n (C • ; δ• ) := ker(δ : C n → C n+1 )/Im(δ : C n−1 → C n ).

(6.1.7)

6.2 Relation between cohomology and homology

In this section, we explain how each cohomology group H n (C• ; G )

can be computed only in terms of the coefficients G and the integral
homology groups H∗ (C• ) of (C• , ∂• ).

Ext groups
Let H and G be given abelian groups. Consider a free resolution of H,

f2 f1 f0
F• : · · · −→ F1 −→ F0 −→ H −→ 0.

Dualize it with respect to G, i.e., apply Hom(−, G ) to it, to get the

cochain complex

f∗ f∗ f∗
F1∗ ←− F0∗ ←− H ∗ ←− 0,
2 1 0
· · · ←−

where we set H ∗ = Hom( H, G ) and similarly for Fi∗ . After discard-

ing H ∗ , we get the cochain complex involving only the Fi∗ ’s, and we
consider its cohomology groups

H n ( F• ; G ) = ker f n∗+1 /Im f n∗

The Ext groups are defined as:

Extn ( H, G ) := H n ( F• ; G ). (6.2.1)

Then the following result, left here as an exercise, holds:

Lemma 6.2.1. The Ext groups are well-defined, i.e., they are independent of
the choice of the free resolution F• of H.

As in the case of the Tor functor, one can thus work with the free
resolution of H given by

0 −→ F1 −→ F0 −→ H −→ 0,

where F0 is the free abelian group on the generators of H, while F1 is

the free abelian group on the relations of H. In particular, it follows
that
Extn ( H, G ) = 0 , ∀n > 1,
 basics of cohomology 119

and we also get that

Ext0 ( H, G ) = Hom( H, G ).

For simplicity, we set:

Ext( H, G ) := Ext1 ( H, G ). (6.2.2)

Proposition 6.2.2. The Ext group Ext( H, G ) satisfies the following proper-
ties:

(a) Ext( H ⊕ H ′ , G ) = Ext( H, G ) ⊕ Ext( H ′ , G ).

(b) If H is free, then Ext( H, G ) = 0.

(c) Ext(Z/n, G ) = G/nG.

Proof. For ( a) use the fact that a free resolution of H ⊕ H ′ is a direct

sum of free resolutions of H and, resp., H ′ . For (b), if H is free, then
0 −→ H −→ H −→ 0 is a free resolution of H, so Ext( H, G ) = 0. For
part (c), start with the free resolution of Z/n given by
·n
0 −→ Z −→ Z −→ Z/n −→ 0,

dualize it and use the fact that Hom(Z, G ) = G to conclude that

Ext(Z/n, G ) = G/nG.

As an immediate consequence of these properties, we get the follow-

ing:
Corollary 6.2.3. If H is a finitely generated abelian group, then :

Ext( H, G ) = Ext(Torsion( H ), G ) = Torsion( H ) ⊗Z G. (6.2.3)

Proof. Indeed, H decomposes into a free part and a torsion part, and
the claim follows by Proposition 6.2.2.

Universal Coefficient Theorem

The following result shows that cohomology is entirely determined by
its coefficients and the integral homology:

Theorem 6.2.4. Given an abelian group G and a chain complex (C• , ∂• ) of

free abelian groups with homology H∗ (C• ), the cohomology group H n (C• ; G )
fits into a natural short exact sequence:
h
0 → Ext( Hn−1 (C• ), G ) −→ H n (C• ; G ) −→ Hom( Hn (C• ), G ) −→ 0
(6.2.4)
In addition, this sequence is split, that is,

H n (C• ; G ) ∼
= Ext( Hn−1 (C• ), G ) ⊕ Hom( Hn (C• ), G ). (6.2.5)
120 algebraic topology

Proof. (Sketch)
The homomorphism h : H n (C• ; G ) → Hom( Hn (C• ), G ) is defined as
follows. Let Zn = ker ∂n , Bn = Im ∂n+1 , in : Bn ,→ Zn the inclusion map,
and Hn (C• ) = Zn /Bn . Let [ϕ] ∈ H n (C• ; G ). Then ϕ is represented by a
homomorphism ϕ : Cn → G, so that δn ϕ := ϕ∂n+1 = 0, which implies
that ϕ| Bn = 0. Let ϕ0 := ϕ| Zn , then ϕ0 vanishes on Bn , so it induces a
quotient homomorphism ϕ¯0 : Zn /Bn → G, i.e., ϕ¯0 ∈ Hom( Hn (C• ), G ).
We define h by
h([ϕ]) = ϕ¯0 .
Notice that if ϕ ∈ Im δn−1 , i.e., ϕ = δn−1 ψ = ψ∂n , then ϕ| Zn = 0, so
ϕ¯0 = 0, which shows that h is well-defined. It is not hard to show that
h is an epimorphism, and

ker h = Coker(in∗ −1 : Zn∗−1 → Bn∗−1 ) = Ext( Hn−1 (C• ), G ), (6.2.6)

where the Ext group is defined with respect to the free resolution of
Hn−1 (C• ) given by
i n −1
0 −→ Bn−1 −→ Zn−1 −→ Hn−1 (C• ) −→ 0.

Remark 6.2.5. The splitting in the above universal coefficient theorem

is not natural; see Exercise 8 at the end of this chapter for an example.

The following special case of Theorem 6.2.4 is very useful in calcula-

tions:

Corollary 6.2.6. Let (C• , ∂• ) be a chain complex so that its (integral) homol-
ogy groups H∗ are finitely generated, and let Tn = Torsion( Hn ). Then we
have natural short exact sequences:

0 → Tn−1 −→ H n (C• ; Z) −→ Hn /Tn → 0 (6.2.7)

This sequence splits, so:

H n (C• ; Z) ∼
= Tn−1 ⊕ Hn /Tn . (6.2.8)

Finally, we have the following easy application of Theorem 6.2.4:

Proposition 6.2.7. If a chain map α : C• → C•′ between chain complexes C•
and C•′ induces isomorphisms α∗ on integral homology groups, then α induces
isomorphisms α∗ on the cohomology groups H ∗ (−; G ) for any abelian group
G.

Proof. By the naturality part of Theorem 6.2.4, we have a commutative

diagram:

0 −→ Ext( Hn−1 (C• ), G ) −→ H n (C• ; G ) −→ Hom( Hn (C• ), G ) −→ 0

↑ (α∗ )∗ ↑ α∗ ↑ (α∗ )∗
0 −→ Ext( Hn−1 (C• ), G ) −→ H (C• ; G ) −→ Hom( Hn (C•′ ), G ) −→ 0
′ n ′
 basics of cohomology 121

The claim follows by the five-lemma, since α∗ and its dual are isomor-
phisms.

6.3 Cohomology of spaces

We can now attach cohomology groups to topological spaces, by work-

ing, e.g., with the singular or cellular chain complex of such a space.

Definition and immediate consequences

Let X be a topological space with singular chain complex (C• ( X ), ∂• ).
The group of singular n-cochains of X with G-coefficients is defined as:

C n ( X; G ) := Hom(Cn ( X ), G ). (6.3.1)

So n-cochains are functions from singular n-simplices to G.

The coboundary map

δn : C n ( X; G ) → C n+1 ( X; G )

is defined as the dual of the corresponding boundary map ∂n+1 :

Cn+1 ( X ) → Cn ( X ), i.e., for ψ ∈ C n ( X; G ), we let
∂ n +1 ψ
δn ψ := ψ∂n+1 : Cn+1 ( X ) → Cn ( X ) → G. (6.3.2)

It follows that
δn+1 ◦ δn = 0, (6.3.3)
and for a singular (n + 1)-simplex σ : ∆ n +1 → X we have:
n +1
δn ψ(σ ) = ∑ (−1)i · ψ(σ|[v0 ,··· ,v̂i ,··· ,vn+1 ] ). (6.3.4)
i =0

Definition 6.3.1. The cohomology groups of X with G-coefficients are defined

as:
H n ( X; G ) := ker(δn )/Im(δn−1 ). (6.3.5)
Elements of ker δn are called n-cocycles, and elements of Im δn−1 are called
n-coboundaries.
Remark 6.3.2. Note that ψ is an n-cocycle if, by definition, it vanishes
on n-boundaries.
Since the groups Cn ( X ) of singular chains are free, we can employ
Theorem 6.2.4 to compute the cohomology groups H n ( X; G ) in terms
of the coefficients G and the integral homology of X. More precisely,
we have natural short exact sequences:

0 −→ Ext( Hn−1 ( X ), G ) −→ H n ( X; G ) −→ Hom( Hn ( X ), G ). −→ 0

(6.3.6)
Moreover, these sequences split, though not naturally.
Let us now derive some immediate consequences from (6.3.6):
122 algebraic topology

(a) If n = 0, (6.3.6) yields that

H 0 ( X; G ) = Hom( H0 ( X ), G ), (6.3.7)

or equivalently, H 0 ( X; G ) consists of all functions from the set of

path-connected components of X to the group G.

(b) If n = 1, the Ext-term in (6.3.6) vanishes since H0 ( X ) is free, so we

get:
H 1 ( X; G ) = Hom( H1 ( X ), G ). (6.3.8)

Remark 6.3.3. Theorem 6.2.4 also works for modules over a PID. In
particular, if G = F is a field, then

H n ( X; F ) ≃ Hom( Hn ( X ), F ) ≃ Hom F ( Hn ( X; F ), F ) = Hn ( X, F )∨

Thus, with field coefficients, cohomology is the dual of homology.

Example 6.3.4. Let X be a point space. From (6.3.6), we have:

H i ( X; G ) = Hom( Hi ( X ), G ) ⊕ Ext( Hi−1 ( X ), G ).

And since 
Z, i=0
Hi ( X ) =
0, otherwise,

we get

 G, i=0
Hom( Hi ( X ), G ) =
0, otherwise.

Furthermore, since Hi ( X ) is free for all i, we also have that

Ext( Hi−1 ( X ), G ) = 0, for all i.

Altogether,

 G, i=0
H i ( X; G ) =
0, otherwise.

Example 6.3.5. Let X = Sn . Then we have

(
Z, i = 0, n
Hi ( X ) =
0, otherwise.

Thus the Ext-term in the universal coefficient theorem vanishes and we

get:
(
i G, i = 0 or n
H ( X; G ) = Hom( Hi ( X ), G ) =
0, otherwise.
 basics of cohomology 123

Reduced cohomology groups

We start with the augmented singular chain complex for X:

· · · −→ C1 ( X ) −→ C0 ( X ) −→ Z −→ 0,
∂ ∂ ϵ

with ϵ(∑i ni xi ) = ∑i ni . After dualizing it (i.e., applying Hom(−; G )),

we get the augmented cochain complex

ϵ∗
· · · ←− C1 ( X; G ) ←− C0 ( X; G ) ←− G ←− 0.
δ δ

Note that since ϵ∂ = 0, we get by dualizing that δϵ∗ = 0. The homology

of this augmented cochain complex is the reduced cohomology of X with
G-coefficients, denoted by H e i ( X; G ).
It follows by definition that

e i ( X; G ) = H i ( X; G ), if i > 0,
H

and by the universal coefficient theorem (applied to the augmented

chain complex), we get

e 0 ( X; G ) = Hom( H
H e 0 ( X ), G ).

Relative cohomology groups

To define relative cohomology groups H n ( X, A; G ) for a pair ( X, A),
we dualize the relative chain complex by setting

C n ( X, A; G ) := Hom(Cn ( X, A), G ). (6.3.9)

The group C n ( X, A; G ) can be identified with functions from the set

of n-simplices in X to G that vanish on simplices in A, so we have a
natural inclusion
C n ( X, A; G ) ,→ C n ( X; G ). (6.3.10)

The relative coboundary maps

δ : C n ( X, A; G ) → C n+1 ( X, A; G ) (6.3.11)

are obtained by restricting the absolute ones, so they satisfy δ2 = 0. So

the relative cohomology groups H n ( X, A; G ) are defined.
We next dualize the short exact sequence

i j
0 −→ Cn ( A) −→ Cn ( X ) −→ Cn ( X, A) −→ 0

to get another short exact sequence

i∗ j∗
0 ←− C n ( A; G ) ←− C n ( X; G ) ←− C n ( X, A; G ) ←− 0, (6.3.12)
124 algebraic topology

where the exactness at C n ( A; G ) follows by extending a cochain in

A “by zero". More precisely, for ψ ∈ C n ( A; G ), we define a function
b : Cn ( X ) → G by
ψ
(
ψ(σ ), if σ ∈ Cn ( A)
ψ(σ) =
if Im(σ ) ∩ A = ∅.
b
0,

Then ψ b is a well-defined element of C n ( X; G ) since Cn ( X ) has a basis

made of simplices contained in A and those contained in X \ A. It is
clear that i∗ (ψ
b) = ψ.
Since i and j commute with ∂, it follows that i∗ and j∗ commute with
δ. So we obtain a short exact sequence of cochain complexes:

i∗ j∗
0 ←− C ∗ ( A; G ) ←− C ∗ ( X; G ) ←− C ∗ ( X, A; G ) ←− 0. (6.3.13)

By taking the associated long exact sequence of homology groups, we

get the long exact sequence for the cohomology groups of the pair
( X, A):

j∗ i∗
· · · → H n ( X, A; G ) → H n ( X; G ) → H n ( A; G ) → H n+1 ( X, A; G ) → · · ·
δ

(6.3.14)
We can also consider above the augmented chain complexes on X and
A, and get a long exact sequence for the reduced cohomology groups,
with H e n ( X, A; G ) = H n ( X, A; G ):

· · · → H n ( X, A; G ) → H e n ( A; G ) → H n+1 ( X, A; G ) → · · ·
e n ( X; G ) → H
(6.3.15)
In particular, if A = x0 is a point in X, we get by (6.3.15) that

e n ( X; G ) ∼
H = H n ( X, x0 ; G ). (6.3.16)

Induced homomorphisms
Recall that if f : X → Y is a continuous map, we have induced chain
maps
f# : Cn ( X ) / Cn (Y )

(σ : ∆n → X ) 
f
/ ( f ◦ σ : ∆n → σ
X → Y)

satisfying f # ∂ = ∂ f # . Dualizing f # with respect to G, we get maps

f # : C n (Y; G ) → C n ( X; G ),

with f # (ψ) = ψ( f # ) and δ f # = f # δ (which is obtained by dualizing

f # ∂ = ∂ f # ). Thus, we get induced homomorphisms on cohomology
groups:
f ∗ : H n (Y, G ) → H n ( X, G ).
 basics of cohomology 125

In fact, we can repeat the above construction for maps of pairs, say
f : ( X, A) → (Y, B). And note that the universal coefficient theorem
also works for pairs because Cn ( X, A) = Cn ( X )/Cn ( A) is free abelian.
So, by naturality, we get a commutative diagram for a map of pairs
f : ( X, A) → (Y, B):

0 / Ext( Hn−1 ( X, A), G ) / H n ( X, A; G ) / Hom( Hn ( X, A), G ) / 0

O O O
( f ∗ )∗ f∗ ( f ∗ )∗

0 / Ext( Hn−1 (Y, B), G ) / H n (Y, B; G ) / Hom( Hn (Y, B), G ) / 0

Homotopy invariance
In this subsection we show that cohomology groups are homotopy
invariants of spaces.

Theorem 6.3.6. If f ≃ g : ( X, A) → (Y, B) are homotopic maps of pairs and

G is an abelian group, then

f ∗ = g∗ : H n (Y, B; G ) → H n ( X, A; G ).

Proof. Recall from the proof of the similar statement for homology that
there is a prism operator

P : Cn ( X, A) → Cn+1 (Y, B) (6.3.17)

satisfying
f # − g# = P∂ + ∂P, (6.3.18)

with f # and g# the induced maps on singular chain complexes. In

fact, if F : X × I → Y denotes the homotopy, with F ( x, 0) = f ( x )
and F ( x, 1) = g( x ), then the prism operator is defined on generators
(σ : ∆n → X ) ∈ Cn ( X ) by pre-composing F ◦ (σ × id) : ∆n × I → Y
with an appropriate decomposition of ∆n × I into (n + 1)-dimensional
simplices. Then one notes that such a P takes Cn ( A) to Cn+1 ( B), hence
it induces the relative prism operator of (6.3.17).
So the difference of the middle maps in the following diagram equals
to the sum of the two side “paths”:

Cn ( X, A)
∂ / Cn−1 ( X, A)
P
f# g#
x   x P
Cn+1 (Y, B)
∂ / Cn (Y, B)

Then it follows from (6.3.18) that f ∗ = g∗ on relative homology groups.

The claim about cohomology follows by dualizing the prism operator
(6.3.17) to get
P∗ : C n+1 (Y, B; G ) → C n ( X, A; G ) (6.3.19)
126 algebraic topology

which satisfies an identity dual to (6.3.18), that is,

f # − g# = δP∗ + P∗ δ. (6.3.20)

This implies readily that f ∗ = g∗ on relative cohomology groups.

The following is an immediate consequence of Theorem 6.3.6:

Corollary 6.3.7. If f : X → Y is a homotopy equivalence, then f ∗ : H n (Y; G ) →
H n ( X; G ) is an isomorphism, for any coefficient group G.
Example 6.3.8. We have:
(
i n G, i = 0
H (R ; G ) =
0, otherwise.

This follows immediately by the homotopy invariance of cohomology

groups, since Rn is contractible.

Excision
Theorem 6.3.9. Given a topological space X, suppose that Z ⊂ A ⊂ X, with
cl ( Z ) ⊆ int( A). Then the inclusion of pairs i : ( X \ Z, A \ Z ) ,→ ( X, A)
induces isomorphisms

i∗ : H n ( X, A; G ) → H n ( X \ Z, A \ Z; G ) (6.3.21)

for all n. Equivalently, if A and B are subsets of X with X = int( A) ∪

int( B), then the inclusion map ( B, A ∩ B) ,→ ( X, A) induces isomorphisms
in cohomology.
Proof. By the naturality of universal coefficient theorem, we have the
commutative diagram:

0 / Ext( Hn−1 ( X, A), G ) / H n ( X, A; G ) / Hom( Hn ( X, A), G ) / 0

(i ∗ ) ∗ i∗ (i ∗ ) ∗

  
0 / Ext( Hn−1 ( X \ Z, A \ Z ), G ) / H n ( X \ Z, A \ Z; G ) / Hom( Hn ( X \ Z, A \ Z ), G ) / 0

By excision for homology, the maps i∗ , hence (i∗ )∗ , are isomorphisms.

So by the five-lemma, it follows that i∗ is also an isomorphism.

Mayer-Vietoris sequence
Theorem 6.3.10. Let X be a topological space, and A and B be subsets of X
so that
X = int( A) ∪ int( B).
Then there is a long exact sequence of cohomology groups:

ψ ϕ
· · · −→ H n ( X; G ) −→ H n ( A; G ) ⊕ H n ( B; G ) −→ H n ( A ∩ B; G )
−→ H n+1 ( X; G ) −→ · · · (6.3.22)
 basics of cohomology 127

Proof. There is a short exact sequence of cochain complexes, which at

level n is given by:

ψ ϕ
0 C n ( A + B; G ) C n ( A; G ) ⊕ C n ( B; G ) C n ( A ∩ B; G ) 0

Hom(Cn ( A + B), G )

where Cn ( A + B) is the set of simplices in X which are sums of simplices

in either A or B, and the maps are defined by

ψ(η ) = (η |Cn ( A) , η |Cn ( B) )

and
ϕ(α, β) = α|Cn ( A∩ B) − β|Cn ( A∩ B) .

Moreover, since C∗ ( A + B) ,→ C∗ ( X ) is a chain homotopy, it follows

by dualizing that C ∗ ( A + B; G ) and C ∗ ( X; G ) are chain homotopic,
and thus H ∗ ( A + B; G ) ∼
= H ∗ ( X; G ). The cohomology Mayer-Vietoris
sequence (6.3.22) is the long exact cohomology sequence of the above
short exact sequence of cochain complexes.

Remark 6.3.11. A similar Mayer-Vietoris sequence holds can be ob-

tained for the reduced cohomology groups.

Example 6.3.12. Let us compute the cohomology groups of Sn by

using the above Mayer-Vietoris sequence. Cover Sn by two open sets
A = Sn \ { N } and B = Sn \ {S}, where N and S are the North and,
resp., South pole of Sn . Then we have A ∩ B ≃ Sn−1 and A ≃ B ≃ Rn .
Thus by the Mayer-Vietoris sequence for reduced cohomology, together
with Example 6.3.8, homotopy invariance and induction, we get:

e i (Sn ; G ) ∼
H =He i −1 ( S n −1 ; G ) ∼
= ··· ∼
=He i − n ( S0 ; G )
(
∼ G, i = n
=
0, otherwise.

Cellular cohomology

Definition 6.3.13. Let X be a CW complex. The cellular cochain complex of

X, (C • ( X; G ), d• ), is defined by setting:

C n ( X; G ) := H n ( Xn , Xn−1 ; G ),

for Xn the n-skeleton of X, and with coboundary maps

dn = δn ◦ jn
128 algebraic topology

fitting in the following diagram (where the coefficient group for cohomology is
by default G):

H n −1 ( X n −1 )
:
j n −1 δ n −1

#
· · · / H n −1 ( X n −1 , X n −2 ) / H n ( X n , X n −1 ) / H n +1 ( X n +1 , X n ) / · · ·
d n −1 dn
<
jn δn
!
H n ( Xn )

Here, the diagonal arrows are part of cohomology long exact sequences for the
relevant pairs. For this reason, it follows that jn δn−1 = 0, and therefore

dn dn−1 = δn jn δn−1 jn−1 = 0.

So (C • ( X; G ), d• ) is indeed a cochain complex. The cellular cohomology of

X with G-coefficients is by definition the cohomology of the cellular cochain
complex (C • ( X; G ), d• ).

Just like in the case of cellular homology, we have the following

identification:

Theorem 6.3.14. The singular and cellular cohomology of X are isomorphic,

i.e.,
H n ( X; G ) ∼
= H n (C • ( X; G )) (6.3.23)

for all n and any coefficient group G. Moreover, the cellular cochain com-
plex (C • ( X; G ), d• ) is isomorphic to the dual of the cellular chain complex
(C• ( X ), d• ), obtained by applying Hom(−; G ).

Proof. Recall from Section 5.5 that for the cellular chain complex of X
we have that

Cn ( X ) := Hn ( Xn , Xn−1 ) ∼
= Z# of n-cells ,

and Hi ( Xn , Xn−1 ) = 0 whenever i ̸= n. So by the universal coefficient

theorem, we obtain:

C n ( X; G ) := H n ( Xn , Xn−1 ; G ) ∼
= Hom(Cn ( X ), G ) (6.3.24)

since the Ext term vanishes. The universal coefficient theorem also
yields that
H i ( Xn , Xn−1 ; G ) = 0 if i ̸= n, (6.3.25)

since the groups Hi ( Xn , Xn−1 ) are either free or trivial. From the long
exact sequence of the pair ( Xn , Xn−1 ), that is,

· · · −→ H k ( Xn , Xn−1 ; G ) −→ H k ( Xn ; G ) −→ H k ( Xn−1 ; G )
−→ H k+1 ( Xn , Xn−1 ; G ) −→ · · · ,
 basics of cohomology 129

we thus get for k ̸= n, n − 1 the isomorphisms

H k ( Xn ; G ) ∼
= H k ( X n −1 ; G ). (6.3.26)

Therefore, if k > n, we obtain by induction:

H k ( Xn ; G ) ∼
= H k ( X n −1 ; G ) ∼
= H k ( X n −2 ; G ) ∼
= ··· ∼
= H k ( X0 ; G ) = 0
(6.3.27)
since X0 is just a set of points.
We next claim that there is an isomorphism

H n ( X n +1 ; G ) ∼
= H n ( X; G ). (6.3.28)

First recall from Lemma 5.5.10(c) that the inclusion Xn+1 ,→ X induces
isomorphisms on homology groups Hk , for k < n + 1. So by the
naturality of the universal coefficient theorem, we get the following
diagram with commutative squares:

h
0 Ext( Hn−1 ( X ), G ) H n ( X; G ) Hom( Hn ( X ), G ) 0
(i∗ )∗ ∼
= i∗ (i∗ )∗ ∼
=
h
0 Ext( Hn−1 ( Xn+1 ), G ) Hn (X n +1 ; G ) Hom( Hn ( Xn+1 ), G ) 0

Then, by using the five-lemma, it follows that the middle map

i∗ : H n ( X; G ) → H n ( Xn+1 ; G )

is also an isomorphism.
Altogether, by using (6.3.27) and (6.3.28), we get the following dia-
gram (where the diagonal arrows are part of long exact sequences of
pairs):

H n −1 ( X n −2 ) ∼
=0
9

H n −1 ( X n −1 )
:
j n −1 δ n −1

%
··· / H n −1 ( X n −1 , X n −2 )
d n −1
/ H n ( X n , X n −1 )
dn
/ H n +1 ( X n +1 , X n ) / ···
<
jn δn
$
n
H ( Xn )
:

: "
α

H (X) ∼
n
= H ( X n +1 ) n
H n ( X n −1 ) ∼
=0

Thus, by using the definition dn = δn jn of the cellular coboundary

maps, and after noting that jn−1 and jn are onto and α is injective, we
130 algebraic topology

obtain the following sequence of isomorphisms:

H n ( X; G ) ∼
= H n ( X n +1 ; G )
∼
= Im(α)
∼
= ker(δn )
∼ ker(dn )/ ker( jn )
= (6.3.29)
∼
= ker(dn )/Im(δn−1 )
∼
= ker(dn )/Im(δn−1 jn−1 )
∼
= ker(dn )/Im(dn−1 ).

The only claim left to prove is that

d n = ( d n +1 ) ∗ . (6.3.30)

By definition, the cellular coboundary map dn is the composition:

jn δn
dn : H n ( Xn , Xn−1 ; G ) −→ H n ( Xn ; G ) −→ H n+1 ( Xn+1 , Xn ; G ),

and, similarly, the boundary map dn+1 of the cellular chain complex is
given by:
∂ n +1 jn
dn+1 : Hn+1 ( Xn+1 , Xn ) −→ Hn ( Xn ) −→ Hn ( Xn , Xn−1 ).
Let us now consider the following diagram:
jn
H n ( X n , X n −1 ; G ) / H n ( Xn ; G ) δn
/ H n +1 ( X n +1 , X n ; G )
∼
= h h ∼
= h
 ( jn )∗  )∗ 
Hom( Hn ( Xn , Xn−1 ), G ) / Hom( Hn ( Xn ), G) ( ∂ n +1
/ Hom( Hn+1 ( Xn+1 , Xn ), G)

The composition across the top is the cellular coboundary map dn , and
we want to conclude that it is the same as the composition (dn+1 )∗
across the bottom row. The extreme vertical arrows labelled h are iso-
morphisms by the universal coefficient theorem, since the relevant Ext
terms vanish (by using (6.3.25)). So it suffices to show that the diagram
commutes. The left square commutes by the naturality of universal
coefficient theorem for the inclusion map ( Xn , ∅) ,→ ( Xn , Xn−1 ), and
the right square commutes by a simple diagram chase.

Example 6.3.15. Let X = RP2 . Then X has one cell in each dimension
0, 1, and 2, and the cellular chain complex of X is:

0 /Z 2 /Z 0 /Z /0.

To compute the (cellular) cohomology H ∗ ( X; Z), we dualize (i.e., apply

Hom(−, Z)) the above cellular chain complex, and get:

0o Zo Zo Zo
2 0
0.
 basics of cohomology 131

Thus, we have

 Z,
 i=0
i 2
H (RP ; Z) = Z/2, i = 2

 0, otherwise.

Similarly, in order to calculate H ∗ ( X; Z/2), we dualize the cellular

chain complex of X with respect to Z/2 (i.e., by applying the functor
Hom(−, Z/2)) to get:

0o Z/2 o Z/2 o Z/2 o

0 0
0

We then have:
(
Z/2, i = 0, 1, or 2
H i (RP2 ; Z/2) ∼
=
0, otherwise.

Example 6.3.16. Let K be the Klein bottle. We compute H∗ (K; Z/3)

and H ∗ (K; Z/3). The cellular chain complex of K is given by:
(2,0)
0 /Z /Z ⊕ Z 0 /Z /0

So the cellular chain complex of K with Z/3-coefficients is given by:

(2,0)
0 / Z/3 / Z/3 ⊕ Z/3 0 / Z/3 /0

Note that the map (2, 0) : Z/3 → Z/3 ⊕ Z/3 is an isomorphism on

the first component, so we get:
(
Z/3, i = 0 or 1
Hi (K; Z/3) =
0, otherwise.

In order to compute the cohomology with Z/3-coefficients, we dualize

the cellular chain complex of K with respect to Z/3 to get:
(2,0)
0o Z/3 o Z/3 ⊕ Z/3 o Z/3 o
0
0

Therefore, we have
(
Z/3, i = 0 or 1
H i (K; Z/3) =
0, otherwise.

Exercises
1. Prove Lemma 6.2.1.

2. Show that the functor Ext(−, −) is contravariant in the first variable,

that is, if H, H ′ and G are abelian groups, a homomorphism α : H → H ′
induces a homomorphism α∗ : Ext( H ′ , G ) → Ext( H, G ).
 132 algebraic topology

3. For a topological space X, let

⟨ , ⟩ : C n ( X ) ⊗ Cn ( X ) → Z
be the Kronecker pairing given by ⟨ϕ, σ ⟩ := ϕ(σ ). In terms of this
pairing, the coboundary map δ : C n ( X ) → C n+1 ( X ) is defined by
⟨δ(ϕ), σ⟩ = ⟨ϕ, ∂σ⟩ for all σ ∈ Cn+1 ( X ). Show that this pairing induces
a pairing between cohomology and homology:
⟨ , ⟩ : H n ( X; Z) ⊗ Hn ( X; Z) → Z.

4. Compute H ∗ (Sn ; G ) by using the long exact sequence of a pair,

coupled with excision.

5. Compute the cohomology of the spaces S1 × S1 , RP2 and the Klein

bottle first with Z coefficients, then with Z/2 coefficients.

6. Show that if f : Sn → Sn has degree d, then f ∗ : H n (Sn ; G ) →

H n (Sn ; G ) is multiplication by d.

7. Show that if A is a closed subspace of X that is a deformation retract

of some neighborhood, then the quotient map X → X/A induces
isomorphisms
H n ( X, A; G ) ∼
=He n ( X/A; G )
for all n.

8. Let X be a space obtained from Sn by attaching a cell en+1 by a

degree m map.
• Show that the quotient map X → X/Sn = Sn+1 induces the trivial
map on H e i (−; Z) for all i, but not on H n+1 (−; Z). Conclude that the
splitting in the universal coefficient theorem for cohomology cannot
be natural.

• Show that the inclusion Sn ,→ X induces the trivial map on reduced

cohomology H e i (−; Z) for all i, but not on Hn (−; Z).

9. Let X and Y be path-connected and locally contractible spaces such

that H 1 ( X; Q) ̸= 0 and H 1 (Y; Q) ̸= 0. Show that X ∨ Y is not a retract
of X × Y.

10. Let X be the space obtained by attaching two 2-cells to S1 , one via
the map z 7→ z3 and the other via z 7→ z5 , where z denotes the complex
coordinate on S1 ⊂ C. Compute the cohomology groups H ∗ ( X; G ) of
X with coefficients:
(a) G = Z.

(b) G = Z/2.

(c) G = Z/3.
 cup product in cohomology 133

7
Cup Product in Cohomology

Let us motivate this chapter with the following simple, but hopefully
convincing example. Consider the spaces X = CP2 and Y = S2 ∨ S4 .
As CW complexes, both X and Y have one 0-cell, one 2-cell and one
4-cell. Hence the cellular chain complex for both X and Y is:
0 0 0 0
0 −→ Z −→ 0 −→ Z −→ 0 −→ Z −→ 0

So X and Y have the same homology and cohomology groups. Note

that X and Y also have the same fundamental groups, namely

π1 ( X ) = π1 (Y ) = 0.

A natural question is then whether X and Y are homotopy equivalent.

Similarly, one can ask if there is a map f : X → Y inducing isomor-
phisms on (co)homology groups. We will see below that by using cup
products in cohomology, we can show that the answer to both questions
is negative.

7.1 Cup Products: definition, properties, examples

Definition 7.1.1. Let X be a topological space, and fix a coefficient ring R

(e.g., Z, Z/nZ, Q). Let ϕ ∈ C k ( X; R) and ψ ∈ C l ( X; R). The cup product
ϕ ∪ ψ ∈ C k+l ( X; R) is defined by:

(ϕ ∪ ψ)(σ : ∆k+l → X ) = ϕ(σ|[v0 ,··· ,vk ] ) · ψ(σ|[vk ,··· ,vk+l ] ), (7.1.1)

where “ · ” denotes the multiplication in ring R.

The aim is to show that this cup product of cochains induces a cup
product of cohomology classes. We need the following result which
relates the cup product to coboundary maps.
Lemma 7.1.2.
δ(ϕ ∪ ψ) = δϕ ∪ ψ + (−1)k ϕ ∪ δψ, (7.1.2)
for ϕ ∈ C k ( X; R) and ψ ∈ C l ( X; R).
134 algebraic topology

Proof. For σ : ∆k+l +1 → X we have

k +1
(δϕ ∪ ψ)(σ) = ∑ (−1)i ϕ(σ|[v0 ,··· ,bvi ,··· ,vk+1 ] ) · ψ(σ|[vk+1 ,··· ,vk+l+1 ] )
i =0

and
k + l +1
(−1)k (ϕ ∪ δψ)(σ) = ∑ (−1)i ϕ(σ|[v0 ,··· ,vk ] ) · ψ(σ|[vk ,··· ,bvi ,··· ,vk+l +1 ] ).
i =k

When we add these two expressions, the last term of the first sum
cancels with the first term of the second sum, and the remaining terms
are exactly δ(ϕ ∪ ψ)(σ ) = (ϕ ∪ ψ)(∂σ ) since

k + l +1
∂σ = ∑ (−1)i σ |[v0 ,··· ,bvi ,··· ,vk+l +1 ] .
i =0

As immediate consequences of the above Lemma, we have:

Corollary 7.1.3. The cup product of two cocycles is again a cocycle. That is,
if ϕ, ψ are cocycles, then δ(ϕ ∪ ψ) = 0.

Proof. This is true, since δϕ = 0 and δψ = 0 imply by (7.1.2) that

δ(ϕ ∪ ψ) = 0.

Moreover, we have the following

Corollary 7.1.4. If one of ϕ or ψ is a cocycle and the other a coboundary, then

ϕ ∪ ψ is a coboundary.

Proof. Say δϕ = 0 and ψ = δη. Then ϕ ∪ ψ = ϕ ∪ δη = ±δ(ϕ ∪ η ).

Similarly, if δψ = 0 and ϕ = δη then ϕ ∪ ψ = δη ∪ ψ = δ(η ∪ ψ).

It follows from Corollary 7.1.3 and Corollary 7.1.4 that we get an

induced cup product on cohomology:
∪
H k ( X; R) × H l ( X; R) −→ H k+l ( X; R). (7.1.3)

It is distributive and associative since it is so on the cochain level. If R

has an identity element, then there is an identity element for the cup
product, namely the class 1 ∈ H 0 ( X; R) defined by the 0-cocycle taking
the value 1 on each singular 0-simplex.
Considering the cup product as an operation on the the direct sum
of all cohomology groups, we get a (graded) ring structure on the
cohomology ⊕i H i ( X; R). We will elaborate on the ring structure on
cohomology groups induced by the cup product after looking at a few
examples and properties of the cup product.
 cup product in cohomology 135

Example 7.1.5. Let us consider the real projective plane RP2 . Its Z/2Z-
cohomology is computed by:

Z/2Z for i = 0, 1, 2
H i (RP2 ; Z/2Z) =
0 otherwise.

Let α ∈ H 1 (RP2 ; Z/2Z) = Z/2Z be the generator, and consider

α2 := α ∪ α ∈ H 2 (RP2 ; Z/2Z).

We claim that α2 ̸= 0, so α2 is in fact the generator of H 2 (RP2 ; Z/2Z).

Consider the cell structure on RP2 with two 0-cells v and w, three
1-cells e, e1 and e2 , and two 2-cells T1 and T2 . The 2-cell T1 is attached
by the word e1 ee2−1 , and the 2-cell T2 is attached by the word e2 ee1−1
(see the figure below). We can of course regard these cells as singular
simplices as well.
w

e1

e T1 v T2 e

e2

Since α is a generator of H 1 (RP2 ; Z/2Z) ∼

= Hom( H1 (RP2 ), Z/2Z), it
is represented by a cocycle

ϕ : C1 (RP2 ) → Z/2Z

with ϕ(e) = 1, where we use the fact that e represents the generator of
H1 (RP2 ). The cocycle condition for ϕ translates into the identities:

0 = (δϕ)( T1 ) = ϕ(∂T1 ) = ϕ(e1 ) + ϕ(e) − ϕ(e2 ).

0 = (δϕ)( T2 ) = ϕ(∂T2 ) = ϕ(e2 ) + ϕ(e) − ϕ(e1 ).

As ϕ(e) = 1, without loss of generality we may take ϕ(e1 ) = 1 and
ϕ(e2 ) = 0.

Next, note that α2 = α ∪ α is represented by ϕ ∪ ϕ, and we have:

(ϕ ∪ ϕ)( T1 ) = ϕ(e1 ) · ϕ(e) = 1

136 algebraic topology

since T1 : [vww] → RP2 . Similarly,

(ϕ ∪ ϕ)( T2 ) = ϕ(e2 ) · ϕ(e) = 0.

Since the generator of H2 (RP2 ; Z/2Z) is T1 + T2 , and we have

(ϕ ∪ ϕ)( T1 + T2 ) = (ϕ ∪ ϕ)( T1 ) + (ϕ ∪ ϕ)( T2 ) = 1 + 0 = 1,

it follows that α2 (which is represented by ϕ ∪ ϕ) is the generator of
H 2 (RP2 ; Z/2Z). □

The cup product on cochains

C k ( X; R) × C l ( X; R) −→ C k+l ( X; R)

restricts to cup products:

C k ( X, A; R) × C l ( X; R) −→ C k+l ( X, A; R),

C k ( X, A; R) × C l ( X, A; R) −→ C k+l ( X, A; R),
and
C k ( X; R) × C l ( X, A; R) −→ C k+l ( X, A; R)
since Ci ( X, A; R) can be regarded as the set of cochains vanishing on
chains in A, and if ϕ or ψ vanishes on chains in A, then so does ϕ ∪ ψ.
So there exist relative cup products:
∪
H k ( X, A; R) × H l ( X; R) −→ H k+l ( X, A; R),
∪
H k ( X, A; R) × H l ( X, A; R) −→ H k+l ( X, A; R),
and
∪
H k ( X; R) × H l ( X, A; R) −→ H k+l ( X, A; R).
In particular, if A is a point, we get a cup product on the reduced
cohomology H e ∗ ( X; R).
More generally, there is a cup product
∪
H k ( X, A; R) × H l ( X, B; R) −→ H k+l ( X, A ∪ B; R)

when A and B are open subsets of X or subcomplexes of the CW

complex X. Indeed, the absolute cup product restricts first to a cup
product

C k ( X, A; R) × C l ( X, B; R) −→ C k+l ( X, A + B; R),

where C k+l ( X, A + B; R) is the subgroup of C k+l ( X; R) consisting of

cochains vanishing on sums of chains in A and chains in B. If A and
B are opens in X, then C k+l ( X, A ∪ B; R) ,→ C k+l ( X, A + B; R) induces
an isomorphism in cohomology, via the five-lemma and the fact that
the restriction maps Ci ( A ∪ B; R) → Ci ( A + B; R) induce cohomology
isomorphisms.

Let us now prove the following simple but important fact:

 cup product in cohomology 137

Lemma 7.1.6. Let f : X → Y be a continuous map with the induced maps on

cohomology f i : H i (Y; R) → H i ( X; R). If α ∈ H k (Y; R) and β ∈ H l (Y; R),
then
f ∗ ( α ∪ β ) = f ∗ ( α ) ∪ f ∗ ( β ), (7.1.4)
and similarly in the relative case.

Proof. It suffices to show the following cochain formula

f # ( ϕ ∪ ψ ) = f # ( ϕ ) ∪ f # ( ψ ),

with ϕ, ψ cochain representatives of α and β, respectively. For ϕ ∈

C k (Y; R) and ψ ∈ C l (Y; R) we have:

f # (ϕ) ∪ f # (ψ)(σ : ∆k+l → X ) = ( f # ϕ)(σ |[v0 ,··· ,vk ] ) · ( f # ψ)(σ |[vk ,··· ,vk+l ] )
= ϕ(( f # σ)|[v0 ,··· ,vk ] ) · ψ(( f # σ)|[vk ,··· ,vk+l ] )
= (ϕ ∪ ψ)( f # σ)
= ( f # (ϕ ∪ ψ))(σ).

Definition 7.1.7. A graded ring is a ring A with a sum decomposition

A = ⊕k Ak where the Ak are additive subgroups so that the multiplication of
A takes Ak × Al to Ak+l . Elements of Ak are called elements of degree k.

Definition 7.1.8. The cohomology ring of a topological space X is the graded

ring !
H ∗ ( X; R) := H k ( X; R), ∪ ,
M

k ≥0

with respect to the cup product operation. If R has an identity, then so does
H ∗ ( X; R). Similarly, we define the cohomology ring of a pair H ∗ ( X, A; R) by
using the relative cup product.
Remark 7.1.9. By scalar multiplication with elements of R, we can
regard these cohomology rings as R-algebras.
The following is an immediate consequence of Lemma 7.1.6:
Corollary 7.1.10. If f : X → Y is a continuous map then we get an induced
ring homomorphism

f ∗ : H ∗ (Y; R) → H ∗ ( X; R).

Example 7.1.11. The isomorphisms

∼
∏ H ∗ ( Xα ; R )
=
H∗ (
G
Xα ; R) −→ (7.1.5)
α α
F
whose coordinates are induced by the inclusions iα : Xα ,→ α Xα is a
ring isomorphism with respect to the coordinate-wise multiplication
 138 algebraic topology

in a ring product, since each coordinate function iα∗ is a ring homomor-

phism. Similarly, the group isomorphism

e∗( Xα ; R ) ∼ ∏ He ∗ (Xα ; R)
_
H = (7.1.6)
α α

is a ring isomorphism. Here the reduced cohomology is identified to

cohomology relative to a basepoint, and we use relative cup products.
(We also assume the basepoints xα ∈ Xα are deformation retracts of
neighborhoods.)

Example 7.1.12. From our calculations in Example 7.1.5 we have that:

H ∗ (RP2 ; Z/2Z) = { a0 + a1 α + a2 α2 | ai ∈ Z/2Z}

= (Z/2Z)[α]/(α3 ),

where α is a generator of H 1 (RP2 ; Z/2Z).

Example 7.1.13.
H ∗ ( S n , Z) = Z[ α ] / ( α2 )
where α is a generator of H n (Sn ; Z). Indeed, we have

Z for i = 0, n
H i ( S n ; Z) =
0 otherwise.

So if α is a generator of H n (Sn ; Z), then the only possible cup products

are α ∪ 1 and α ∪ α. However, α ∪ α ∈ H 2n (Sn ; Z) = 0. Hence α2 = 0.
Let us now recall that the cell structure on

RP∞ = RPn
[

n ≥0

consists of one cell in each non-negative dimension. The following

result will be proved later on in this section:
Theorem 7.1.14. The cohomology rings of the real (resp. complex) projective
spaces are given by:

(a)
H ∗ (RPn ; Z/2Z) ∼
= Z/2[α]/(αn+1 )
where α is the generator of H 1 (RPn ; Z/2Z).

(b)
H ∗ (RP∞ ; Z/2Z) ∼
= Z/2[α]
where α is the generator of H 1 (RPn ; Z/2Z).

(c)
H ∗ (CPn ; Z) = Z[ β]/( βn+1 )
where β is the generator of H 2 (CPn ; Z).
 cup product in cohomology 139

(d)
H ∗ (CP∞ ; Z) = Z[ β]

where β is the generator of H 2 (CPn ; Z).

Before discussing the proof of the above theorem, let us get back to
the following motivating example:

Example 7.1.15. We saw at the beginning of this chapter that the spaces
X = CP2 and Y = S2 ∨ S4 have the same homology and cohomology
groups, and even the same CW structure. The cup products can be
used to decide whether these spaces are homotopy equivalent. Indeed,
let us consider the cohomology rings H ∗ ( X; Z) and H ∗ (Y; Z). From
the above theorem, we have that:

H ∗ (CP2 ; Z) = Z[ β]/( β3 ),

where β is the generator of H 2 (CP2 ; Z). We also have a ring isomor-

phism
e ∗ ( S2 ∨ S4 ; Z) ∼
H =He ∗ ( S2 ; Z) ⊕ H
e ∗ ( S4 ; Z),

where H ∗ (S2 ; Z) = Z[α]/(α2 ) and H ∗ (S4 ; Z) = Z[γ]/(γ2 ), with de-

gree of α equal to 2 and degree of γ equal to 4. Moreover, α2 = 0,
γ2 = 0 and α ∪ γ = 0. Next, we consider the cohomology generators in
degree 2 and square them. In the case of H ∗ (CP2 ; Z), β2 is a generator
of H 4 (CP2 ; Z), hence β2 ̸= 0. However, in the case of H ∗ (S2 ∨ S4 ; Z),
α2 ∈ H 4 (S2 ; Z) = 0. Hence the two cohomology rings of the two spaces
are not isomorphic, hence the two spaces are not homotopy equivalent.

Let us now get back to the proof of Theorem 7.1.14. We will discuss
below the proof in the case of RPn . The result in the case of RP∞ fol-
lows from the finite-dimensional case since the inclusion RPn ,→ RP∞
induces isomorphisms on H i (−; Z/2) for i ≤ n by cellular cohomol-
ogy. The complex projective spaces are handled in precisely the same
manner, using Z-coefficients and replacing H k by H 2k and R by C.
We next prove the following result:

Theorem 7.1.16.

H ∗ (RPn ; Z/2) = Z/2[α]/(αn+1 ), (7.1.7)

where α is the generator of H 1 (RPn ; Z/2).

Proof. For simplicity, we use the notation

Pn := RPn

and all coefficients for the cohomology groups are understood to be

Z/2-coefficients.
140 algebraic topology

We prove (7.1.7) by induction on n. Let αi be a generator for H i (Pn )

and α j be a generator for H j (Pn ), with i + j = n. Since for any k < n
the inclusion map u : Pk ,→ Pn induces isomorphisms on cohomology
groups H l , for l ≤ k, it suffices by induction on n to show that αi ∪ α j ̸=
0.
Recall now that Pn = Sn /(Z/2), with
n
Sn = {( x0 , · · · , xn ) ∈ Rn+1 | ∑ xl2 = 1}.
l =0

Let
i
Si = {( x0 , · · · , xi , 0, · · · , 0) | ∑ xl2 = 1}
l =0
and
n
S j = {(0, · · · , 0, xn− j , · · · xn ) | ∑ xl2 = 1}
l =n− j

be the i-th and j-th (sub)sphere respectively. Note that since i + j = n,

we have that xn− j = xi . Hence Si ∩ S j = {(0, · · · , 0, ±1, 0, · · · , 0)} with
±1 in the i-th position, i.e., the intersection consists of the two antipodal
points with i-th coordinate ±1 and all other coordinates zero.

Sn

Sj

Si

Hence, Pi = Si /(Z/2) and P j = S j /(Z/2) are subsets of Pn =

Sn /(Z/2) so that

Pi ∩ P j = { p } = ( 0 : · · · : 0 : 1 : 0 : · · · : 0 )

with 1 in the i-th place.

Let U ⊂ Pn be the open subset consisting of points ( x0 : · · · : xn )
with xi ̸= 0, i.e.,

U = {( x0 : · · · : xi−1 : 1 : xi+1 : · · · : xn )},

and notice that the map

ϕ ( x 0 : · · · : x i −1 : 1 : x i +1 : · · · : x n ) = ( x 0 , · · · , x i −1 , x i +1 , · · · , x n )

is a homeomorphism U ∼
= Rn which takes p to 0 ∈ Rn .
 cup product in cohomology 141

We clearly have that Pn = Pn−1 ∪ U, where Pn−1 is identified to the

set of points in Pn with the i-th coordinate equal to zero. Regarding
U as the interior of the n-cell of Pn (attached to Pn−1 ), it follows that
Pn − { p} deformation retracts to Pn−1 . Similarly, as { p} = Pi ∩ P j ,
we have that Pi − { p} ≃ Pi−1 and P j − { p} ≃ P j−1 . All of this is
represented schematically in the figure below, where Pn is represented
by a disc with its antipodal boundary points identified.

P j −1

Pj

Pi − 1 p Pi − 1
Pi

P j −1

Let us now write Rn = Ri × R j , with coordinates of factors denoted

by ( x0 , · · · , xi−1 ) and ( xi+1 , · · · , xn ), respectively. Consider the follow-
ing commutative diagram with horizontal arrows given by the (relative)
cup product:

H i (Pn ) × H j (Pn ) / H n (Pn )

O O

H i ( P n , P n − P j ) × H j ( P n , P n − Pi ) / H n (Pn , Pn − { p})

 
H ( R , R − R ) × H j ( R n , R n − Ri )
i n n j / H n (Rn , Rn − {0})

The diagram commutes by the naturality of the cup product. Let us

examine the bottom row in the above diagram. Let Di denote a small
closed i-disc in Ri with boundary Si−1 . Then by homotopy equivalence
and excision we have:
H i (Rn , Rn − R j ) ∼
= H i (Rn , Rn − int( Di ) × R j )
∼
= H i ( D i × R j , S i −1 × R j )
∼
= H i ( D i × D j , S i −1 × D j )
∼
= H i (( Di , Si−1 ) × D j )
∼
= H i ( D i , S i −1 ).
Similarly,
H j ( R n , R n − Ri ) ∼
= H j (( D j , S j−1 ) × Di )
142 algebraic topology

∼
= H j ( D j , S j −1 )

and

H n (Rn , Rn − {0}) ∼
= H n ( D n , S n −1 )
∼
= H n ( D i × D j , S i −1 × D j ∪ S j −1 × D i ).

Since D n is an n-cell, its class [ D n ] (in the Z/2-cellular cohomology)

generates H n ( D n , Sn−1 ), and similar considerations apply to [ Di ] ∈
H i ( Di , Si−1 ) and [ D j ] ∈ H j ( D j , S j−1 ). So the above isomorphisms and
cellular cohomology show that the cup product of the bottom arrow in
the above commutative diagram takes the product of generators to a
generator, i.e., it is given by

[ D i ] × [ D j ] 7 → [ D n ].

The same will be true for the top row, provided we show that the four
vertical maps in the above diagram are isomorphisms.
For the bottom right vertical arrow, we have by excision that

H n (Pn , Pn − { p}) ∼
= H n (U, U − { p}) ∼
= H n (Rn , Rn − {0}), (7.1.8)

where the last isomorphism follows by using the homeomorphism

ϕ : U → Rn .
For the top right vertical arrow, we already noted that Pn − { p} defor-
mation retracts to Pn−1 , so we have

H n (Pn , Pn − { p}) ∼
= H n (Pn , Pn −1 ) ∼
= Z/2, (7.1.9)

where the second isomorphism follows by cellular cohomology. More-

over, by using the long exact sequence for the cohomology of the
pair (Pn , Pn−1 ) and the fact that H n (Pn−1 ) = 0, we get that the map
Z/2 = H n (Pn , Pn−1 ) → H n (Pn ) ∼ = Z/2 is onto, hence an isomor-
phism. Thus we get:

H n (Pn , Pn − { p}) ∼
= H n (Pn ) (7.1.10)

To show that the two left vertical arrows are isomorphisms, consider
the following commutative diagram.

(5)
H i (Pn ) o H i ( P n , Pi − 1 ) o H i (Pn , Pn − P j ) / H i (Rn , Rn − R j )
(2) (4)
(1) (3) (6) (7)
   (10)

H i ( Pi ) o H i ( Pi , Pi − 1 ) o H i (Pi , Pi − { p}) / H i (Ri , Ri − {0})
(8) (9)

It suffices to show that all these maps are isomorphisms. (Then

to finish the proof of the theorem, just interchange i and j.) First
note that (Rn , Rn − R j ) = (Ri , Ri − {0}) × R j deformation retract to
 cup product in cohomology 143

(Ri , Ri − {0}), so arrow (7) is an isomorphism. As already pointed out,

(10) is an isomorphism by (7.1.8). Moreover, (9) is an isomorphism as
in (7.1.9), and (8) is an isomorphism as in (7.1.10). The arrow (1) is
an isomorphism by cellular homology, and the arrow (3) is an isomor-
phism by cellular homology and the naturality of the cohomology long
exact sequence. By commutativity of the left square, it then follows that
(2) is an isomorphism. In order to show that (4) is an isomorphism,
we note that Pn − P j deformation retracts onto Pi−1 . Indeed, a point
v = ( x0 : · · · : xn ) ∈ Pn − P j has at least one of the first i coordinates
non-zero, so the function

f t (v) := ( x0 : · · · : xi−1 : txi : · · · : txn )

gives, as t decreases from 1 to 0, a deformation retract from Pn − P j

onto Pi−1 .
Since (3), (4) and (9) are isomorphisms, the commutativity of the
middle square yields that (6) is an isomorphism. Finally, since (6),
(7) and (10) are isomorphisms, the commutativity of the right square
yields that (5) is an is0morphism, which completes the proof of the
theorem.

Example 7.1.17. Let us consider the spaces RP2n+1 and RP2n ∨ S2n+1 .
First note that these spaces have the same CW structure and the same
cellular chain complex, so they have the same homology and cohomol-
ogy groups. However, we claim that RP2n+1 and RP2n ∨ S2n+1 are not
homotopy equivalent. In order to justify the claim, we first compute
their Z/2Z-cohomology rings.
From the above theorem, the cohomology ring of RP2n+1 is:

H ∗ (RP2n+1 ; Z/2Z) = Z/2Z[α]/(α2n+2 ),

where α is a degree one element generating H 1 (RP2n+1 ; Z/2Z).

We also have a ring isomorphism

e ∗ (RP2n ∨ S2n+1 ; Z/2Z) ∼

H =He ∗ (RP2n ; Z/2Z) ⊕ H
e ∗ (S2n+1 ; Z/2Z)

with H ∗ (RP2n ; Z/2Z) ∼ = Z/2Z[ β]/( β2n+1 ) for β the degree 1 genera-
tor of H 1 (RP2n ; Z/2Z), and H ∗ (S2n+1 ; Z/2Z) ∼ = Z/2Z[γ]/(γ2 ) for γ
the generator of H 2n + 1 (S 2n + 1 ; Z/2Z) of degree 2n + 1.
If there was a homotopy equivalence f : RP2n+1 → RP2n ∨ S2n+1 , then
the generators of degree one would correspond isomorphically to each
other, i.e., we would get f ∗ ( β) = α. But as f ∗ is a ring isomorphism,
this would then imply that: f ∗ ( β2n+1 ) = ( f ∗ ( β))2n+1 = α2n+1 . How-
ever, this yields a contradiction, since β2n+1 = 0, thus f ∗ ( β2n+1 ) = 0,
while α2n+1 ̸= 0 since α2n+1 generates H 2n+1 (RP2n+1 ; Z/2Z).
144 algebraic topology

7.2 Application: Borsuk-Ulam Theorem

In this section we use cup products in order to prove the following

result:
Theorem 7.2.1 (Borsuk-Ulam). If n > m ≥ 1, there are no maps g : Sn →
Sm commuting with the antipodal maps, i.e., for which g(− x ) = − g( x ), for
all x ∈ Sn .

Proof. We prove the theorem by contradiction. Assume that there is a

map g : Sn → Sm commuting with the antipodal maps. Then g carries
pairs of antipodal points ( x, − x ) in Sn to pairs of antipodal points
g( x ), g(− x ) = − g( x ) in Sm . So, by passage to the quotient, g induces

a map
f : RPn → RPm
[ x ] 7→ [ g( x )]
which makes the following diagram commutative:
g
Sn / Sm

p′ ↷ p
 f 
RPn / RPm

Here p and p′ are the two-sheeted covering maps.

We claim that there exists a lift f ′ of f , i.e., f = p f ′ in the following
diagram:
m
;S
f′
p
f 
RPn / RPm
Let us for now assume the claim and complete the proof of the theorem.
Consider the following diagram:
m
5;S
g
p
f′ 
Sn / RPn / RPm
p′ f

We have pg = f p′ = p f ′ p′ , the second equality following from the

above claim. This implies that both g and f ′ p′ are lifts of f p′ . Under
the two-sheeted covering map p, antipodal points in Sm are mapped
to the same point in RPm . Therefore, pg = p f ′ p′ implies that at a
point x ∈ Sn , we have g( x ) = f ′ p′ ( x ) or ag( x ) = f ′ p′ ( x ), where
a : Sm → Sm is the antipodal map. But ag( x ) = − g( x ) = g(− x ) and
f ′ p′ ( x ) = f ′ p′ (− x ). Thus at x ∈ Sn , one of following equalities holds:
g( x ) = f ′ p′ ( x ) or g(− x ) = f ′ p′ (− x ). Since g and f ′ p′ are lifts of f p′
 cup product in cohomology 145

and they coincide at a point, it follows by the uniqueness of the lift

that g = f ′ p′ . But this is a contradiction since p′ ( x ) = p′ (− x ), hence
f ′ p′ ( x ) = f ′ p′ (− x ), while g( x ) ̸= g(− x ) = − g( x ).
It remains to prove the claim. A lift for f exists iff

f ∗ (π1 (RPn )) ⊆ p∗ (π1 (Sm )). (7.2.1)

If m = 1, the only homomorphism

f ∗ : π1 (RPn ) ∼
= Z/2 → π1 (RP1 ) ∼
=Z

is the trivial one, so (7.2.1) is satisfied.

If m > 1, both groups π1 (RPn ) and π1 (RPm ) are Z/2. We will use
cup products to show that the induced map f ∗ : Z/2 → Z/2 on
fundamental groups is the trivial map. Let αm ∈ H ∗ (RPm ; Z/2) and
αn ∈ H ∗ (RPn ; Z/2) be the generators of degree 1, and consider the
induced ring homomorphism

f ∗ : H ∗ (RPm ; Z/2) → H ∗ (RPn ; Z/2).

We have:
0 = f ∗ (αm +1 ∗
m ) = f (αm )
m +1
,
so f ∗ (αm ) ∈ H 1 (RPn ; Z/2) has order m + 1 < n + 1. Therefore,

f ∗ (αm ) ̸= αn .

Since H 1 (RPn ; Z/2) = Z/2 = ⟨αn ⟩, this implies that

f ∗ (αm ) = 0.

Let i : RP1 ,→ RPn and j : RP1 ,→ RPm be the inclusions obtained

by setting all but the first two homogeneous coordinates equal to zero.
By cellular cohomology, the map j∗ : H 1 (RPm ) → H 1 (RP1 ) is an
isomorphism, so j∗ (αm ) is the generator of H 1 (RP1 ), and in particular,

j∗ (αm ) ̸= 0.

On the other hand,

( f ◦ i )∗ (αm ) = i∗ ( f ∗ (αm )) = 0.

So ( f ◦ i )∗ ̸= j∗ , hence the maps f ◦ i and j are not homotopic.

But the homotopy classes of i and j generate the groups π1 (RPn )
and π1 (RPm ), respectively. So the homomorphisms

f ∗ : π1 (RPn ) ≃ Z/2 −→ π1 (RPm ) ≃ Z/2

[i ] 7→ [ f ◦ i] ̸= [ j]

maps the generator [i ] to an element of Z/2 other than the generator

[ j], i.e., f ∗ = 0. This proves the claim, and completes the theorem.
 146 algebraic topology

Exercises
1. Show that if X is the union of contractible open subsets A and B,
then all cup products of positive-dimensional classes in H ∗ ( X ) are
zero. In particular, this is the case if X is a suspension. Conclude that
spaces such as RP2 and T 2 cannot be written as unions of two open
contractible subsets.

2. Is the Hopf map

f : S3 ⊂ C2 → S2 = C ∪ {∞}, (z, w) 7→ z
w

nullhomotopic? Explain.

3. Is there a continuous map f : X → Y inducing isomorphisms on

∼
=
all of the cohomology groups (i.e., f ∗ : H i (Y; Z) → H i ( X; Z), for all
i) but X and Y do not have isomorphic cohomology rings (with Z
coefficients)? Explain your answer.

4. Show that RP3 and RP2 ∨ S3 have the same cohomology rings with
integer coefficients.

5.
(a) Show that H ∗ (CPn ; Z) ∼
= Z[ x ]/( x n+1 ), with x the generator of
H 2 (CPn ; Z).

(a) Show that the Lefschetz number τ f of a map f : CPn → CPn is given
by
τ f = 1 + d + d2 + · · · + d n ,
where f ∗ ( x ) = dx for some d ∈ Z, and with x as in part ( a).

(c) Show that for n even, any map f : CPn → CPn has a fixed point.

(d) When n is odd, show that there is a fixed point unless f ∗ ( x ) = − x,

where x denotes as before a generator of H 2 (CPn ; Z).

6. Use cup products to compute the map H ∗ (CPn ; Z) → H ∗ (CPn ; Z)

induced by the map CPn → CPn that is a quotient of the map Cn+1 →
Cn+1 raising each coordinate to the d-th power, i.e.,

(z0 , · · · , zn ) 7→ (z0d , · · · , zdn ),

for a fixed integer d > 0. (Hint: First do the case n = 1.)

7. Describe the cohomology ring H ∗ ( X ∨ Y ) of a join of two spaces.

8. Let H = R · 1 ⊕ R · i ⊕ R · j ⊕ R · k be the skew-field of quaternions,

where i2 = j2 = k2 = −1 and ij = k = − ji, jk = i = −kj, ki = j = −ik.
For a quaternion q = a + bi + cj + dk, a, b, c, d ∈ R, its conjugate is
√
defined by q̄ = a − bi − cj − dk. Let |q| := a2 + b2 + c2 + d2 .
 cup product in cohomology 147

(a) Verify the following formulae in H: q · q̄ = |q|2 , q1 q2 = q̄2 q̄1 , |q1 q2 | =

| q1 | · | q2 |.

(b) Let S7 ⊂ H ⊕ H be the unit sphere, and let f : S7 → S4 = HP1 =

H ∪ {∞} be given by f (q1 , q2 ) = q1 q2 −1 . Show that for any p ∈ S4 ,
the fiber f −1 ( p) is homeomorphic to S3 .

(c) Let HPn be the quaternionic projective space defined exactly as in

the complex case as the quotient of Hn+1 \ {0} by the equivalence
relation v ∼ λv, for λ ∈ H \ {0}. Show that the CW structure of
HPn consists of only one cell in each dimension 0, 4, 8, · · · , 4n, and
calculate the homology of HPn .

(d) Show that H ∗ (HPn ; Z) ∼

= Z[ x ]/( x n+1 ), with x the generator of
H (HP ; Z).
4 n

(e) Show that S4 ∨ S8 and HP2 are not homotopy equivalent.

9. For a map f : S2n−1 → Sn with n ≥ 2, let X f = Sn ∪ f D2n be the

CW complex obtained by attaching a 2n-cell to Sn by the map f . Let
a ∈ H n ( X f ; Z) and b ∈ H 2n ( X f ; Z) be the generators of respective
groups. The Hopf invariant H ( f ) ∈ Z of the map f is defined by the
identity a2 = H ( f )b.

(a) Let f : S3 → S2 = C ∪ {∞} be given by f (z1 , z2 ) = z1 /z2 , for

(z1 , z2 ) ∈ S3 ⊂ C2 . Show that X f = CP2 and H ( f ) = ±1.

(b) Let f : S7 → S4 = H ∪ {∞} be given by f (q1 , q2 ) = q1 q2 −1 in

terms of quaternions (q1 , q2 ) ∈ S7 , the unit sphere in H2 . Show that
X f = HP2 and H ( f ) = ±1.

7.3 Künneth Formula

Cross product
Let us motivate this section by consider the spaces S2 × S3 and S2 ∨ S3 ∨
S5 . Both spaces are CW complexes with cells {e0 , e2 , e3 , e5 } in degrees,
0, 2, 3 and 5, respectively. So the cellular chain complex for both spaces
is:
0
0 → Z → 0 → Z −→ Z → 0 → Z → 0

Hence both spaces have the same homology and cohomology groups.
It is then natural to ask the following:

Question 7.3.1. Are the spaces S2 × S3 and S2 ∨ S3 ∨ S5 homotopy equiva-

lent?
148 algebraic topology

The aim of this section is to convince the reader that the answer is
No. More precisely, we will show that the two spaces have different
cohomology rings.
The cohomology ring H ∗ (S2 ∨ S3 ∨ S5 ; Z) can be computed from the
ring isomorphism

e ∗ ( S2 ∨ S3 ∨ S5 ; Z) ∼
H =He ∗ ( S2 ; Z) ⊕ H
e ∗ ( S3 ; Z) ⊕ H
e ∗ ( S5 ; Z),

with H ∗ (S2 ; Z) ∼
= Z[ α ] / ( α2 ), H ∗ ( S3 ; Z) ∼
= Z[ β]/( β2 ) and H ∗ (S5 ; Z) ∼
=
Z[γ]/(γ ), where α is the generator of H (S ; Z), β is the generator of
2 2 2

H 3 (S3 ; Z) and γ is the generator of H 5 (S5 ; Z). Moreover, we have that

α ∪ β = 0. Indeed, let

p : S2 ∨ S3 ∨ S5 → S2 ∨ S3

be the natural retraction map. Then p∗ induces isomorphisms on H 2

and H 3 . So if ᾱ and β̄ are the generators of H 2 (S2 ∨ S3 ) and H 3 (S2 ∨ S3 ),
then α = p∗ ᾱ and β = p∗ β̄. So

α ∪ β = p∗ ᾱ ∪ p∗ β̄ = p∗ (ᾱ ∪ β̄) = 0

since ᾱ ∪ β̄ = 0.
By the end of this section, we will show that the product of the gen-
erators of degree 2 and degree 3 in the cohomology ring of S2 × S3 is
the generator in degree 5, so it is non-zero. This will then completely
answer the above question.

The following result is proved in [Hatcher, Theorem 3.14]:

Theorem 7.3.2. Let R be a commutative ring, and α ∈ H k ( X, A; R) and

β ∈ H l ( X, A; R). Then the following holds:

α ∪ β = (−1)kl · β ∪ α. (7.3.1)

Definition 7.3.3. A graded ring which satisfies a condition as in the pre-

vious theorem is called graded commutative. Hence the cohomology ring
H ∗ ( X, A; R) is a graded commutative ring.

Corollary 7.3.4. If α ∈ H ∗ ( X; R) is of odd degree and if H ∗ ( X; R) has no

elements of order two, then α ∪ α = 0.

Definition 7.3.5. Cross product or External cup product

Let X and Y be topological spaces, and denote by p and q the projections
p : X × Y −→ X and q : X × Y −→ Y. By using the cohomology maps
defined by these projections, we have an induced map denoted by ×:
×
H ∗ ( X; R) × H ∗ (Y; R) −→ H ∗ ( X × Y; R)
a b 7→ a × b := p∗ ( a) ∪ q∗ (b)
 cup product in cohomology 149

All cohomology groups H i ( X; R) and H i (Y; R) have an R-module structure,

hence so do the corresponding cohomology rings H ∗ ( X; R) and H ∗ (Y; R).
Since the map × is bilinear, the universal property for tensor products yields a
group homomorphism called the cross product, which we again denote by ×:
×
H ∗ ( X; R) ⊗ R H ∗ (Y; R) −→ H ∗ ( X × Y; R) (7.3.2)

So, by definition, we have that:

×( a ⊗ b) := a × b.

The cross-product becomes a ring homomorphism if we put a ring structure on

H ∗ ( X; R) ⊗ R H ∗ (Y; R) by the following multiplication operation:

( a ⊗ b) · (c ⊗ d) = (−1)deg(b)·deg(c) ( ac ⊗ bd) (7.3.3)

Indeed, we have:

×(( a ⊗ b) · (c ⊗ d)) = (−1)deg(b)·deg(c) × ( ac ⊗ bd)

= (−1)deg(b)·deg(c) ( ac × bd)
= (−1)(deg b)·deg(c) p∗ ( a ∪ c) ∪ q∗ (b ∪ d)
= (−1)deg(b)·deg(c) p∗ ( a) ∪ p∗ (c) ∪ q∗ (b) ∪ q∗ (d)
(7.3.1)
= p∗ ( a) ∪ q∗ (b) ∪ p∗ (c) ∪ q∗ (d)
= ×( a ⊗ b) ∪ ×(c ⊗ d).

Künneth theorem in cohomology. Examples

The following result is very helpful for finding the cohomology ring of
a product of CW complexes:
Theorem 7.3.6. Künneth Formula
If X and Y are CW complexes, and H k (Y; R) is a finitely generated free
R-module for all k, then the cross product
×
H ∗ ( X; R) ⊗ R H ∗ (Y; R) −→ H ∗ ( X × Y; R)

is a ring isomorphism. Moreover, we have the following isomorphism of groups:

H n ( X × Y; R) ∼ H i ( X; R) ⊗ R H j (Y; R)
M
= (7.3.4)
i + j=n

In the next section, we will explain the content of Theorem 7.3.6 in a

more general context. Let us now work out some examples.
Example 7.3.7. Let us find the cohomology ring of S2 × S3 , which
appeared at the beginning of this section. According to the Künneth
formula, we have the following ring isomorphism:

H ∗ ( S2 × S3 ; Z) ∼
= H ∗ ( S2 ; Z) ⊗Z H ∗ ( S3 ; Z)
150 algebraic topology

If we let a ∈ H ∗ (S2 ; Z) denote the degree 2 element which generates

H 2 (S2 ; Z) and b ∈ H ∗ (S3 ; Z) the degree 3 element which generates
H 3 (S3 ; Z), then ×( a ⊗ 1) and ×(1 ⊗ b) (where 1 denotes the identity
in the respective cohomology rings) will be the generators in H ∗ (S2 ×
S3 ; Z) of degree 2 and 3, respectively. Moreover, ×( a ⊗ 1) ∪ ×(1 ⊗ b) =
×( a ⊗ b) will be a generator of degree 5 in H ∗ (S2 × S3 ; Z).
In order to simplify the notations, we make the following definition.

Definition 7.3.8. Exterior Algebra

Let R be a commutative ring with identity. The exterior algebra over R,
denoted
Λ R [ α1 , α2 , . . . ],
is the free R-module generated by products of the form:

αi1 αi2 · · · αik , with i1 < i2 < · · · < ik ,

and with associative and distributive multiplication defined by the rules:

αi α j = −α j αi , if i ̸= j
α2i = 0.

The empty product of αi ’s is allowed and it gives the identity element 1 ∈

Λ R [ α1 , α2 , . . . ].

Example 7.3.9. Let us now show that

H ∗ ( S3 × S5 × S7 ; Z) ∼
= ΛZ [ a3 , a5 , a7 ], (7.3.5)

where ai is the generator of degree i in H ∗ (S3 × S5 × S7 ; Z), for i =

3, 5, 7.
By the Künneth formula applied to the product of CW complexes
S3 × S5 × S7 , we have the following ring isomorphism:

H ∗ ( S3 × S5 × S7 ; Z) ∼
= H ∗ ( S3 ; Z) ⊗Z H ∗ ( S5 ; Z) ⊗Z H ∗ ( S7 ; Z).

Let αi be the generator of degree i in H ∗ (Si ; Z) for i = 3, 5, 7. Then

the generators of degree 3, 5 and 7 in H ∗ (S3 × S5 × S7 ; Z) are given
respectively by:

• a3 = ×(α3 ⊗ 1 ⊗ 1)

• a5 = ×(1 ⊗ α5 ⊗ 1)

• a7 = ×(1 ⊗ 1 ⊗ α7 )

The product of these generators produce generators of higher degrees,

i.e., 8, 10, 12 and 15, in the cohomology ring H ∗ (S3 × S5 × S7 ; Z). Let
us compute some products of the elements:

a23 = ×(α3 ⊗ 1 ⊗ 1) ∪ ×(α3 ⊗ 1 ⊗ 1)

 cup product in cohomology 151

 
= × ( α3 ⊗ 1 ⊗ 1) · ( α3 ⊗ 1 ⊗ 1)
= ×(α23 ⊗ 1 ⊗ 1)
= 0
and a similar result for a25 and a27 .

a3 a5 = ×(α3 ⊗ 1 ⊗ 1) ∪ ×(1 ⊗ α5 ⊗ 1)
 
= × ( α3 ⊗ 1 ⊗ 1) · (1 ⊗ α5 ⊗ 1)
= (−1)0·0 × (α3 ⊗ α5 ⊗ 1)
= ×(α3 ⊗ α5 ⊗ 1)

a5 a3 = ×(1 ⊗ α5 ⊗ 1) ∪ ×(α3 ⊗ 1 ⊗ 1)
 
= × (1 ⊗ α5 ⊗ 1) · ( α3 ⊗ 1 ⊗ 1)
= (−1)3·5 × (α3 ⊗ α5 ⊗ 1)
= − a3 a5
We have similar results for the other products too. The above cal-
culations show that we have an isomorphism H ∗ (S3 × S5 × S7 ; Z) ∼
=
ΛZ [ a3 , a5 , a7 ].

Remark 7.3.10. It is easy to see that a similar result holds for the
cohomology ring of any (finite) product of odd dimensional spheres.

Example 7.3.11. By the Künneth formula we have the following ring

isomorphism:

H ∗ (RP∞ × RP∞ ; Z/2) = H ∗ (RP∞ ; Z/2) ⊗Z2 H ∗ (RP∞ ; Z/2)

= Z/2[α] ⊗Z/2 Z/2[ β]
= Z/2[α, β]
where α and β are generators of degree 1, and they commute since we
work with Z/2-coefficients.
Example 7.3.12. Let us now investigate if the spaces CP6 and S2 × S4 ×
S6 are homotopy equivalent. Fortunately, there is an easy answer to this
question. Consider the usual CW structure for CP6 and the product
CW structure for S2 × S4 × S6 . Both spaces have cells only in even
dimensions, but CP6 has one cell in dimension 6, whereas S2 × S4 × S6
has two cells in dimension 6. It follows that H6 (CP6 ) = Z, whereas
H6 (S2 × S4 × S6 ) = Z ⊕ Z. So CP6 and S2 × S4 × S6 are not homotopy
equivalent. A more difficult approach to answer the question would be
to show that the cohomology rings for these spaces are not isomorphic.
We will do this in the following example.
n ( n +1)
Example 7.3.13. Let us show that if n > 1, the spaces CP 2 and S2 ×
S4 × · · · × S2n are not homotopy equivalent. Consider the following
cases:
152 algebraic topology

• If n = 1, then CP1 is homeomorphic to S2 .

• If n = 2, then both the spaces CP3 and S2 × S4 have one cell in each
of the dimensions {0, 2, 4, 6}. Thus they also have the same cellular
chain/cochain complex and, in particular, their homology/cohomol-
ogy groups are isomorphic. We will, however, distinguish these
spaces by their cohomology rings.

• If n ≥ 3, then CPn has one cell in each of the even dimensions

{0, 2, 4, . . . , 2n}, but the cell structure of S2 × S4 × · · · × S2n is differ-
ent from that of CPn since, for example, S2 × S4 × · · · × S2n has two
6-cells. As both spaces have cells only in even dimensions, we can
already conclude that they have different homology and cohomology
groups since they have different cell structures.

We will now show that for n > 1 the two spaces have non-isomorphic
cohomology rings. First, the Künneth formula yields that:

H ∗ (S2 × S4 × · · · × S2n ; Z)
∼
= H ∗ (S2 ; Z) ⊗Z H ∗ (S4 ; Z) ⊗Z · · · ⊗Z H ∗ (S2n ; Z)

So a degree 2 element in this ring looks like ×( a ⊗ 1 ⊗ 1 ⊗ · · · ⊗ 1),

where a ∈ H 2 (S2 ). The square of this element is:

[×( a ⊗ 1 ⊗ 1 ⊗ · · · ⊗ 1)]2 = ×[( a ⊗ 1 ⊗ 1 ⊗ · · · ⊗ 1)2 ]

= ×( a2 ⊗ 1 ⊗ 1 ⊗ · · · ⊗ 1)
= 0
n ( n +1)
since a2 ∈ H 4 (S2 ) = 0. However, in the case of CP 2 , we know that
square of a non-zero degree 2 element is a non-zero degree 4 element.
Hence the cohomology rings of the two spaces are not isomorphic.

Example 7.3.14. Let us use cup products and the Künneth formula in
order to show that Sn ∨ Sm is not a retract of Sn × Sm , for n, m ≥ 1.
First, consider the product CW structure on Sn × Sm : it consists of cells
{e0 , em , en , em+n } with attaching maps ϕ : ∂em → e0 and ϕ′ : ∂en → e0
coming from the factors. Hence Sn ∨ Sm is a subset of Sn × Sm . (Note
that we also allow the case n = m.) Next, suppose by contradiction that
there is a retract
r : Sn × Sm → Sn ∨ Sm .

So, if i : Sn ∨ Sm ,→ Sn × Sm denotes the inclusion, then the composition

r ◦ i is the identity map on Sn ∨ Sm . It follows that the cohomology map
(r ◦ i )∗ = i∗ ◦ r ∗ is the identity, so

r ∗ : H ∗ (Sn ∨ Sm ) −→ H ∗ (Sn × Sm )
 cup product in cohomology 153

is a monomorphism. By the Künneth formula, we have a ring isomor-

phism
×
H ∗ (Sn ) ⊗ H ∗ (Sm ) ∼
= H ∗ ( S n × S m ).
Hence, a non-zero element in H n (Sn × Sm ) is of the form a × 1 :=
×( a ⊗ 1), with a ∈ H n (Sn ) a non-zero class. Similarly, a non-zero
element in H m (Sn × Sm ) is of the form 1 × b := ×(1 ⊗ b), for some non-
zero class b ∈ H m (Sm ). Let us now consider the product of non-zero
elements a × 1 ∈ H n (Sn × Sm ) and 1 × b ∈ H m (Sn × Sm ) in the ring
H ∗ (Sn × Sm ). We get:

( a × 1) ∪ (1 × b) = ×( a ⊗ 1) ∪ ×(1 ⊗ b)
= ×[( a ⊗ 1) · (1 ⊗ b)]
= ×( a ⊗ b) (7.3.6)
= a×b
̸= 0,
since a ⊗ b ̸= 0 in H ∗ (Sn ) ⊗ H ∗ (Sm ). We also have a ring isomorphism
e ∗ (Sn ∨ Sm ) ∼
H =He ∗ (Sn ) ⊕ H
e ∗ ( S m ).

Let α, β ∈ H ∗ (Sn ∨ Sm ) be the generators of degree n and m, respectively.

Then
α ∪ β ∈ H n+m (Sn ∨ Sm ) = 0.
On the other hand, since r ∗ is a monomorphism, the classes r ∗ (α) and
r ∗ ( β) are non-zero elements of degree n and, resp., m in the cohomology
ring H ∗ (Sn × Sm ), so by the above calculation, their product is non zero.
But
r ∗ (α) ∪ r ∗ ( β) = r ∗ (α ∪ β) = r ∗ (0) = 0,
which gives us a contradiction.

Künneth exact sequence and applications

In this section, we provide the necessary background for Künneth-type
theorems.

Let us fix coefficients in a PID ring R.

Given two chain complexes (C• , ∂• ) and (C•′ , ∂′• ) of R-modules, we
define (C ⊗ C ′ )• to be the complex with:
n
(C ⊗ C ′ )n = (C p ⊗ Cn′ − p )
M
(7.3.7)
p =0

and boundary map dn : (C ⊗ C ′ )n → (C ⊗ C ′ )n−1 which on C p ⊗ Cn′ − p

is given by:

dn ( a ⊗ b) = (∂ p a) ⊗ b + (−1) p ( a ⊗ ∂′n− p b). (7.3.8)

154 algebraic topology

Then we have:
(d ◦ d)( a ⊗ b) = d (∂a) ⊗ b + (−1) p ( a ⊗ ∂′ b)

= (∂2 a) ⊗ b + (−1) p−1 (∂a) ⊗ (∂′ b)

+ (−1) p (∂a) ⊗ (∂′ b) + (−1) p a ⊗ (∂′2 b)
= 0.
So (C ⊗ C ′ )• , d• is a chain complex. It is therefore natural to ask the

following question:
Question 7.3.15. How is the homology H∗ ((C ⊗ C ′ )• ) related to H∗ (C• )
and H∗ (C•′ )?
The answer is provided by the following result from homological
algebra:
Theorem 7.3.16 (Künneth exact sequence). Let R be a PID, and assume
that for each i, Ci is a free R-module. Then for all n, there is a split short exact
sequence:

H p (C• ) ⊗ R Hn− p (C•′ ) −→ Hn ((C ⊗ C )• )

M

0 −→
p

TorR H p (C• ), Hn− p−1 (C•′ ) −→ 0 (7.3.9)

M

−→
p

In what follows we discuss several applications of Theorem 7.3.16.

Künneth Formula for homology.

Let X and Y be two spaces, and let C• and C•′ denote the singular
chain complexes of X and Y, respectively. Then it is not hard to see
that the singular chain complex C• ( X × Y ) of X × Y is chain homotopy
equivalent to (C ⊗ C ′ )• , so they have the same homology groups. We
thus have the following important consequence of Theorem 7.3.16:
Corollary 7.3.17 (Künneth Formula for homology). If X and Y are
topological spaces, then the following holds:
n −1

nM
Hn ( X × Y ) ∼
M

= H p ( X ) ⊗ Hn− p (Y ) ⊕ Tor H p ( X ), Hn− p−1 (Y ) .
p =0 p =0
(7.3.10)
In particular, if all homology groups of X or Y are free R-modules, then:
n
Hn ( X × Y ) ∼
M
= H p ( X ) ⊗ Hn− p (Y ). (7.3.11)
p =0

As a consequence of Corollary 7.3.17, we have:

Corollary 7.3.18. If the Euler characteristics χ( X ) and χ(Y ) are defined,
then χ( X × Y ) is defined, and:
χ ( X × Y ) = χ ( X ) · χ (Y ) . (7.3.12)
 cup product in cohomology 155

Universal Coefficient Theorem for homology

The Universal Coefficient Theorem for homology can be seen as a conse-
quence of Theorem 7.3.16 as follows: take C• to be the singular chain
complex of X and let C•′ to be the chain complex defined by: Cn′ = 0 if
n ̸= 0, C0′ = R, and ∂′n = 0 for all n ≥ 0. We then get by Theorem 7.3.16
that:

Hn ( X; R) ∼


= Hn ( X ) ⊗ R ⊕ Tor( Hn−1 ( X ), R). (7.3.13)

Remark 7.3.19. Note that (7.3.13) can also be obtained from (7.3.10) by
taking Y to be a point.

Künneth formula for cohomology

Finally, we also have the following cohomology Künneth formula:

Corollary 7.3.20. Künneth formula for cohomology

If R is a PID, and all homology groups Hi ( X; R) are finitely generated, then
there is a split exact sequence (with R-coefficients):
n
H p ( X ) ⊗ H n− p (Y ) −→ H n ( X × Y )
M

0 −→
p =0
n +1
Tor H p ( X ), H n− p+1 (Y ) −→ 0.
M

−→
p =0

Moreover, if all cohomology groups H i ( X ) of X (or Y) are free over R, we get

the following isomorphism:
n
Hn (X × Y) ∼ H p ( X ) ⊗ H n − p (Y ) .
M
= (7.3.14)
p =0

Proof. (Sketch.) Let us indicate how this result is obtained from Theo-
rem 7.3.16. We would like to apply the Künneth exact sequence to the
chain complexes defined by:

C−n := C n ( X; R), ∂−n := δX

n

and
′ n ′ n
C− n : = C (Y; R ), ∂−n : = δY .

However, note that Ci and Ci′ are not necessarily R-free. Indeed,

C n ( X; R) = HomR (Cn ( X; R), R),

but Cn ( X; R) is not necessarily a finitely generated R-module. In or-

der to get around this problem, the idea is to replace the chain com-
plex C• ( X; R) by a chain homotopic one, which has finitely generated
components. Here is where the assumption that Hi ( X; R) are finitely
generated is used.
156 algebraic topology

Exercises
1. Are the spaces S2 × RP4 and S4 × RP2 homotopy equivalent? Justify
your answer!

2. Using cup products, show that every map Sk+l → Sk × Sl induces the
trivial homomorphism Hk+l (Sk+l ) → Hk+l (Sk × Sl ), assuming k > 0
and l > 0.

3. Describe H ∗ (CP∞ /CP1 ; Z) as a ring with finitely many multiplica-

tive generators. How does this ring compare with H ∗ (S6 × HP∞ ; Z)?

4. Show that if Hn ( X; Z) is finitely generated and free for each n, then

H ∗ ( X; Z p ) and H ∗ ( X; Z) ⊗ Z p are isomorphic as rings, so in particular
the ring structure with Z-coefficients determines the ring structure with
Z p -coefficients.

5. Show that the cross product map H ∗ ( X; Z) ⊗ H ∗ (Y; Z) → H ∗ ( X ×

Y; Z) is not an isomorphism if X and Y are infinite discrete sets.

6. Show that for n even Sn is not an H-space, i.e., there is no map

µ : Sn × Sn → Sn so that µ ◦ i1 = idSn and µ ◦ i2 = idSn , where i1 , i2 are
the inclusions on factors.

7. Let A be the union of two once linked circles in S3 , and B be the

union of two unlinked circles. Show that the cohomology groups of
S3 \ A and S3 \ B are isomorphic, but their cohomology rings are not.

8. Compute the ring structure of H ∗ ( T n ; Z), where T n is the torus of

dimension n (i.e., a product of n circles S1 ). Do the same for H ∗ ( T n \
{ x }; Z), where x ∈ T n is any point.
 poincaré duality 157

8
Poincaré Duality

8.1 Introduction

In this chapter, we show that oriented n-manifolds enjoy a very special

symmetry on their (co)homology groups:

Theorem 8.1.1. Let M be a closed (i.e., compact without boundary), ori-

ented and connected manifold of dimension n. Then for all i ≥ 0 we have
isomorphisms:
Hi ( M; Z) ∼
= H n−i ( M; Z). (8.1.1)

In particular, we get:

Corollary 8.1.2. For all i ≥ 0, the isomorphisms

(8.1.1) (UCT )
Hi ( M; Q) ∼
= H n−i ( M; Q) ∼
= Hom( Hn−i ( M; Q), Q) (8.1.2)

yield a non-degenerate bilinear pairing

Hi ( M; Q) × Hn−i ( M; Q) → Q.

In particular, the complementary Betti numbers of M are equal, i.e.,

β i ( M ) = β n −i ( M ).

In the next section we will explain in more detail the notion of

orientability of manifolds. Later on, we will describe explicitly the
nature of the isomorphism (8.1.1) by using the cap product operation ∩,
i.e., we will show that it is realized by

∩[ M] : H n−i ( M; Z) −→ Hi ( M; Z), (8.1.3)

where [ M] ∈ Hn ( M; Z) is the “fundamental (orientation) class" of the

manifold M.
158 algebraic topology

8.2 Manifolds. Orientation of manifolds

Definition 8.2.1. A Hausdorff space M is a (topological) manifold if any

point x ∈ M has a neighborhood Ux homeomorphic to Rn (where such a
homeomorphism takes x to 0).

Let us now compute the local homology groups of a manifold M at

some point x ∈ M:

(1)
Hi ( M, M \ { x }; Z) ∼
= Hi (Ux , Ux \ { x }; Z)
(2)
∼
= Hi (Rn , Rn \ {0}; Z)
(3)
∼ e i −1 (Rn \ { 0 } ; Z )
= H (8.2.1)
(4)
∼ e i −1 ( S n −1 ; Z )
= H

Z, if i = n
=
0 , otherwise,

where (1) follows by excision, (2) by using the homeomorphism Ux = ∼

R , (3) by the homology long exact sequence of a pair, and (4) by using
n

a deformation retract.

Definition 8.2.2. The dimension of a manifold M, denoted dim( M ), is the

only non-vanishing degree of the local homology groups of M.

Definition 8.2.3. A local orientation of an n-manifold M at x ∈ M is a

choice µ x of one of the two generators of the local homology group Hn ( M, M \
{ x }; Z) = Z.

Remark 8.2.4. A local orientation µ x at x ∈ M induces local orientations

at all nearby points y, i.e., if x and y are contained in a small ball B,
then we have induced isomorphisms:

∼
=
µ x ∈ Z = Hn ( M, M \ { x }) ←
− Hn ( M, M \ B)
∼
=
→ Hn ( M, M \ {y}) = Z ∈ µy ,
−

where the above isomorphisms are induced by deformation retracts.

Definition 8.2.5. A (global) orientation on an n-manifold M is a continuous

choice of local orientations, i.e., for every x ∈ M there exists a closed ball
B ⊂ Ux ∼ = Rn and a (generating) class µ B ∈ Hn ( M, M \ B) such that
ρy : Hn ( M, M \ B) → Hn ( M, M \ {y}) takes µ B to µy for all y ∈ B.

Definition 8.2.6. The pair consisting of manifold and orientation is called an

oriented manifold.
 poincaré duality 159

Notation: Let M be an n-manifold and K ⊂ L ⊂ M be compact subsets.

Consider the map induced by inclusion of pairs:

ρK : Hi ( M, M \ L) → Hi ( M, M \ K ).

Then for a ∈ Hi ( M, M \ L), ρK ( a) is called the restriction of a to K.

In the above notations, we have the following important result:

Theorem 8.2.7. For any oriented manifold M of dimension n and any compact
K ⊂ M, there is a unique µK ∈ Hn ( M, M \ K; Z) such that ρ x (µK ) = µ x
for all x ∈ K.

An immediate corollary of the above theorem is the existence of the

fundamental class of compact oriented manifolds. More precisely, by
taking K = M in Theorem 8.2.7, we get the following:

Corollary 8.2.8. If M is a compact oriented n-manifold, there exists a unique

µ M ∈ Hn ( M; Z) so that ρ x (µ M ) = µ x for all x ∈ M.

Definition 8.2.9. The homology class [ M] := µ M of Corollary 8.2.8 is called

the fundamental class of M.

The proof of Theorem 8.2.7 uses the following:

Lemma 8.2.10. If K is a compact subset of an n-manifold M, we have:

(i) Hi ( M, M \ K ) = 0 if i > n.

(ii) a ∈ Hn ( M, M \ K ) is equal to 0 if and only if ρ x ( a) = 0 for all x ∈ K.

Before proving the above lemma, let us finish the proof of Theorem
8.2.7.

Proof. (of Theorem 8.2.7)

For the uniqueness part, if µ1K and µ2K are as in the statement of the
theorem, then for all x ∈ K we have ρ x (µ1K − µ2K ) = µ x − µ x = 0. Then
by using Lemma 8.2.10(ii), we get that µ1K − µ2K = 0, or µ1K = µ2K .
We prove the existence part in several steps:
Step I: If K is contained in a sufficiently small euclidean closed ball (of
finite positive radius) B centered at a point y ∈ M, as in the definition
of orientability, then for all x ∈ K, the composition
ρK ρx
Hn ( M, M \ B) −→ Hn ( M, M \ K ) −
→ Hn ( M, M \ { x }) (8.2.2)

is an isomorphism. Then set µK := ρK (µ B ), with µ B ∈ Hn ( M, M \ B)

as in the definition of orientability.
160 algebraic topology

Step II: If the theorem holds for compact subsets K1 and K2 and for their
intersection K1 ∩ K2 , we show that it holds for their union K = K1 ∪ K2 .
Indeed, the Mayer-Vietoris sequence for the open cover

M \ ( K1 ∩ K2 ) = ( M \ K1 ) ∪ ( M \ K2 ) ,

with intersection

M \ K = ( M \ K1 ) ∩ ( M \ K2 )

gives the long exact sequence:

φ
0 → Hn ( M, M \ K ) −
→ Hn ( M, M \ K1 ) ⊕ Hn ( M, M \ K2 )
ψ
−
→ Hn ( M, M \ (K1 ∩ K2 )) → . . .

where φ( a) = ρK1 ( a) ⊕ ρK2 ( a) and ψ(b ⊕ c) = ρK1 ∩K2 (b) − ρK1 ∩K2 (c).
By our assumption, there exist unique µK1 ∈ Hn ( M, M \ K1 ) and µK2 ∈
Hn ( M, M \ K2 ) restricting to local orientations at points x ∈ K1 and,
resp., x ∈ K2 , hence

ρ x ◦ ρ K1 ∩ K2 ( µ K i ) = ρ x ( µ K i ) = µ x (8.2.3)

for all x ∈ K1 ∩ K2 and i = 1, 2. Then we have

ρ x (ρK1 ∩K2 (µK1 ) − ρK1 ∩K2 (µK2 )) = µ x − µ x = 0 (8.2.4)

for all x ∈ K1 ∩ K2 . So by Lemma 8.2.10 we get that

ψ(µK1 ⊕ µK2 ) = ρK1 ∩K2 (µK1 ) − ρK1 ∩K2 (µK2 ) = 0, (8.2.5)

i.e., µK1 ⊕ µK2 ∈ ker ψ = Im φ. Since φ is injective, there exists a unique

µK ∈ Hn ( M, M \ K )

such that φ(µK ) = µK1 ⊕ µK2 . By the uniqueness part, we also have that
µK restricts to local orientations at points x ∈ K.

Step III: For an arbitrary compact K, we write K as a finite union

K = K1 ∪ K2 ∪ . . . ∪ Kr with each Ki as in Step I. Then the claim follows
by induction on r by using Step II.

Let us now get back to proving Lemma 8.2.10:

Proof. (of Lemma 8.2.10)

The proof is done in several steps, as indicated below.
Step I: Assume that M = Rn and K is a convex compact subset. Let
B be a large ball in Rn with K ⊂ B, and let S = ∂B be the bounding
 poincaré duality 161

sphere. Then for all x ∈ K, both M \ K and M \ { x } deformation retract

to S. So we have:

Hi ( M, M \ K ) ∼
= Hi ( M, M \ { x })
∼
= Hi (Rn , Sn−1 )
∼
=He i −1 ( S n −1 ) (8.2.6)

Z for i = n
=
0 otherwise.

Step II: We next show that if the Lemma holds for compact sets K1 ,
K2 and for their intersection K1 ∩ K2 , then it holds for K := K1 ∪ K2 .
Indeed, we have the Mayer-Vietoris sequence

· · · → Hi+1 ( M, M \ (K1 ∩ K2 )) → Hi ( M, M \ K )
φ ψ
−
→ Hi ( M, M \ K1 ) ⊕ Hi ( M, M \ K2 ) −
→ Hi ( M, M \ (K1 ∩ K2 )) → · · ·

If i > n, we have by our assumption that Hi+1 ( M, M \ (K1 ∩ K2 )) = 0,

Hi ( M, M \ K1 ) = 0 and Hi ( M, M \ K2 ) = 0. Therefore, Hi ( M, M \ K ) =
0.

If i = n, the Mayer-Vietoris sequence takes the form

φ
0 → Hn ( M, M \ K ) −
→ Hn ( M, M \ K1 ) ⊕ Hn ( M, M \ K2 )
ψ
−
→ Hn ( M, M \ (K1 ∩ K2 )) → . . .

with φ injective. So for a ∈ Hn ( M, M \ K ), we have the following

sequence of equivalences:

a = 0 ⇐⇒ 0 = φ( a) = ρK1 ( a) ⊕ ρK2 ( a)
⇐⇒ ρK1 ( a) = 0 and ρK2 ( a) = 0
⇐⇒ ρ x ρK1 ( a) = 0 ∀ x ∈ K1 , and ρy ρK2 ( a) = 0 ∀y ∈ K2 (8.2.7)
(since, by assumption, the lemma holds for K1 and K2 )
⇐⇒ ρ x ( a) = 0, ∀ x ∈ K1 ∪ K2 .

Step III: If M = Rn and K = K1 ∪ K2 ∪ · · · ∪ Kr with each Ki convex and

compact (which also implies that each Ki ∩ K j is convex and compact),
then the lemma holds for K by Step I and Step II.

Step IV: Assume that M = Rn and K is an arbitrary compact subset

in Rn . Choose a compact neighborhood N of K in Rn . Then for any
a ∈ Hi ( M, M \ K ) there exists a′ ∈ Hi ( M, M \ N ) such that ρK ( a′ ) = a.
Indeed, if γ is a cycle representative of a, we have that γ ∈ Ci (Rn ) and
∂γ ∈ Ci−1 (Rn \ K ). So ∂γ ∩ K = ∅. Choose N small enough so that
162 algebraic topology

∂γ ∩ N = ∅. Next, we cover K by a union of closed balls Bi such that

Bi ⊂ N and Bi ∩ K ̸= ∅. Then ρK factors as

ρK
Hi (Rn , Rn \ N ) - Hi (Rn , Rn \ K )
-
ρ∪
iB
i - ρK

Hi (Rn , Rn \ ∪i Bi )

If i > n, then Hi (Rn , Rn \ ∪i Bi ) = 0 by Step III. So for any a ∈

Hi (Rn , Rn \ K ), we have that

a = ρK ( a′ ) = ρK (ρ∪i Bi ( a′ )) = 0.

If i = n, then ρ x ( a) = 0 for all x ∈ K implies by a deformation retract

argument that ρ x ( a) = 0 for all x ∈ ∪i Bi . By using Step III, we then get
that ρ∪i Bi ( a′ ) = 0. Hence we have a = ρK (ρ∪i Bi ( a′ )) = 0.

Step V: If K is contained in some euclidean neighborhood in (arbitrary)

M, we have by excision

Hi ( M, M \ K ) ∼
= Hi (Rn , Rn \ K ). (8.2.8)

So the Lemma holds for K by Step IV.

Step VI: Finally, note that any compact subset K of M can be written
as a union K = K1 ∪ K2 ∪ . . . ∪ Kr with each Ki as in Step V. Then the
Lemma follows by using Step V, Step II and induction.

Exercises
1. Show that every covering space of an orientable manifold is an
orientable manifold.

2. Given a covering space action of a group G on an orientable manifold

M by orientation-preserving homeomorphisms, show that M/G is also
orientable.

3. For a map f : M → N between connected closed orientable n-

manifolds with fundamental classes [ M] and [ N ], the degree of f is
defined to be the integer d such that f ∗ ([ M ]) = d[ N ], so the sign of the
degree depends on the choice of fundamental classes. Show that for
any connected closed orientable n-manifold M there is a degree 1 map
M → Sn .

4. Show that a p-sheeted covering space projection M → N has degree

p, when M and N are connected closed orientable manifolds.
 poincaré duality 163

5. Given two disjoint connected n-manifolds M1 and M2 , a connected

n-manifold M1 #M2 , their connected sum, can be constructed by deleting
the interiors of closed n-balls B1 ⊂ M1 and B2 ⊂ M2 and identifying
the resulting boundary spheres ∂B1 and ∂B2 via some homeomorphism
between them. (Assume that each Bi embeds nicely in a larger ball in
Mi .)

(a) Show that if M1 and M2 are closed then there are isomorphisms

Hi ( M1 #M2 ; Z) ≃ Hi ( M1 ; Z) ⊕ Hi ( M2 ; Z), for 0 < i < n,

with one exception: If both M1 and M2 are non-orientable, then

Hn−1 ( M1 #M2 ; Z) is obtained from Hn−1 ( M1 ; Z) ⊕ Hn−1 ( M2 ; Z) by
replacing one of the two Z2 -summands by a Z-summand.

(b) Show that χ( M1 #M2 ) = χ( M1 ) + χ( M2 ) − χ(Sn ) if M1 and M2 are

closed.

8.3 Cohomolgy with Compact Support

Let X be a topological space and define the compactly supported i-cochains

on X by:

Cci ( X ) := Ci ( X, X \ K ) ⊂ Ci ( X ).
[
(8.3.1)
K compact in X

Equivalently,

Cci ( X ) = { φ : Ci ( X ) → Z | ∃ compact K φ ⊂ X
s.t. φ = 0 on chains in X \ K φ }.

Define a coboundary operator by

δφ(σ) := φ(∂σ),

and note that if φ ∈ Cci ( X ) vanishes on chains in X \ K φ then δφ is also

zero on all chains in X \ K φ , and so δφ ∈ Cci+1 ( X ). Therefore we get a
cochain (sub)complex (Cc• ( X ), δ• ).
Definition 8.3.1. The i-th cohomology of X with compact support is defined
by
Hci ( X ) := H i (Cc• ( X )).
In what follows, we give an alternative characterization of the coho-
mology with compact support, which is more useful for calculations.
We begin by recalling the notion of direct limit of groups.
Definition 8.3.2. Let Gα be abelian groups indexed by some directed set I,
i.e., I has a partial order ≤ and for any α, β ∈ I, there exists γ ∈ I such
164 algebraic topology

that α ≤ γ and β ≤ γ. Suppose also that for each pair α ≤ β there is a

homomorphism f αβ : Gα → Gβ such that f αα = idGα and f αγ = f βγ ◦ f αβ .
Consider the set
⨿α Gα / ∼

where the equivalence relation ∼ is defined as: if x ∈ Gα , x ′ ∈ Gα′ , then

x ∼ x ′ if f αγ ( x ) = f α′ γ ( x ′ ) with α, α′ ≤ γ. Any two equivalence classes [ x ]
and [ x ′ ] have representatives lying in the same Gγ , with α, α′ ≤ γ, so we can
define
[ x ] + [ x ′ ] = [ f αγ ( x ) + f α′ γ ( x ′ )].

This is a well-defined binary operation, and it gives an abelian group structure

on the set ⨿α Gα / ∼. The direct limit of the groups Gα is then the group
defined as:
lim Gα := ⨿α Gα / ∼ . (8.3.2)
−→
α∈ I

Remark 8.3.3. If J ⊂ I so that ∀α ∈ I, ∃ β ∈ J with α ≤ β, then

lim Gα = lim Gβ . In particular, if J = { β} (i.e, I contains a maximal
−→ −→
α∈ I β∈ J
element), then lim Gα = Gβ .
−→
α∈ I

We can now prove the following result:

Proposition 8.3.4. There is an isomorphism

Hci ( X ) ∼
= lim H i ( X, X \ K ) (8.3.3)
−→
K∈ I

where I := {K ⊂ X | K compact}.

Proof. First note that I is a directed set since it is partially ordered by

inclusion, and the union of two compact sets is also compact. Moreover,
if K ⊆ L are compact subsets of X, then there is a homomorphism
f KL : H i ( X, X \ K ) → H i ( X, X \ L) induced by inclusion. Hence the
direct limit group limK ∈ I H i ( X, X \ K ) is well-defined.
−→
Each element of limK ∈ I H i ( X, X \ K ) is represented by some cocycle
−→
φ ∈ Ci ( X, X \ K ) for some compact subset K of X. Regarding φ as an
i-cochain with compact support, its cohomology class yields an element
[ φ] ∈ Hci ( X ). Moreover, such a cocycle φ ∈ Ci ( X, X \ K ) is the zero
element in lim H i ( X, X \ K ) iff φ = δψ for some ψ ∈ Ci ( X, X \ L) with
−→
L ⊃ K, and so [ φ] = 0 in Hci ( X ).

Remark 8.3.5. If X is compact, then Hci ( X ) = H i ( X ), for all i ≥ 0, since

in this case there is a unique maximal compact set K ⊂ X, namely X
itself.
 poincaré duality 165

Example 8.3.6. Let us compute the cohomology with compact support

of Rn . By the above proposition,

Hci (Rn ) = lim H i (Rn , Rn \ K ),

−→
K

where the direct limit is over the directed set of compact subsets of Rn .
Note that it suffices to let K range over closed balls Bk of integer radius
k centered at the origin since each compact K ⊂ Rn is contained in such
a ball. So we have that

lim H i (Rn , Rn \ K ) = lim H i (Rn , Rn \ Bk ).

−→ −→
K k ∈Z≥0

Moreover, we have isomorphisms

H n (Rn , Rn \ Bk ) ∼
= H n (Rn , Rn \ Bk+1 )

induced by inclusion, since for all k:


Z if i = n
H i (Rn , Rn \ Bk ) ∼
= H i (Rn , Rn \ {0}) ∼
=
0 otherwise.

Altogether,

Hci (Rn ) ∼
= lim H i (Rn , Rn \ K ) = lim H i (Rn , Rn \ Bk )
−→ −→
k ∈Z≥0

Z if i = n
=
0 otherwise.

Remark 8.3.7. It follows from the previous example that the cohomol-
ogy with compact support Hc∗ (−) is not a homotopy invariant.
b = X ∪ x̂ be the one point compactification of X.
Remark 8.3.8. Let X
Then
Hci ( X ) ∼ b x̂ ) ∼
= H i ( X, =He i (X
b ). (8.3.4)
For example, Hci (Rn ) ∼=H e i (Sn ). This follows from the following gen-
eral fact. If U is an open subset of a topological space V, with closed
complement Z := V \ U, then there exists a long exact sequence for the
cohomology with compact support

· · · → Hci (U ) → Hci (V ) → Hci ( Z ) → Hci+1 (U ) → · · ·

b = X ∪ x̂, we get a long exact sequence

If we apply this fact to the case X

· · · → Hci ( X ) → Hci ( X
b ) → Hci ( x̂ ) → · · ·

Since Xb and x̂ are compact, this yields that Hci ( X ) ∼ b x̂ ) ∼

= H i ( X, =He i (X
b ),
as claimed.
166 algebraic topology

8.4 Cap Product and the Poincaré Duality Map

Definition 8.4.1. We define the cap product operation

∩
Ci ( X ) ⊗ Cn ( X ) −
→ Cn−i ( X ) (8.4.1)

as follows: for b ∈ Ci ( X ) and ξ ∈ Cn ( X ), b ∩ ξ ∈ Cn−i ( X ) is defined by

a(b ∩ ξ ) := ( a ∪ b)(ξ ) (8.4.2)

where a ∈ C n−i ( X ).
Remark 8.4.2. In view of the definition of the cup product, one can
reformulate the above definition of the cap product as follows: if
σ : ∆n → X is an n-simplex and b ∈ Ci ( X ), then

b ∩ σ = b(σ |[υn−i ,··· ,υn ] ) · σ |[υ0 ,··· ,υn−i ] . (8.4.3)

| {z } | {z }
∈Z ∈Cn−i ( X )

Moreover, for a, b ∈ C ∗ ( X ) and ξ ∈ C∗ ( X ) one has the identity

a ∩ (b ∩ ξ ) = ( a ∪ b) ∩ ξ.

The following result is a direct consequence of the definition:

Lemma 8.4.3. For any b ∈ Ci ( X ) and ξ ∈ Cn ( X ), we have:

∂(b ∩ ξ ) = (−1)n−i δb ∩ ξ + b ∩ ∂ξ. (8.4.4)

Proof. For any a ∈ C n−i−1 ( X ), we have

a (∂(b ∩ ξ )) = δa(b ∩ ξ )
= (δa ∪ b)(ξ )
 
= δ( a ∪ b) − (−1)n−i−1 a ∪ δb (ξ )
= ( a ∪ b)(∂ξ ) − (−1)n−i−1 a(δb ∩ ξ )
= a(b ∩ ∂ξ ) + (−1)n−i a(δb ∩ ξ ).

As a consequence, the cap product descends to (co)homology:

Corollary 8.4.4. There is an induced cap product operation
∩
H i ( X ) ⊗ Hn ( X ) −
→ Hn−i ( X ). (8.4.5)

Moreover, for a, b ∈ H ∗ ( X ) and ξ ∈ H∗ ( X ) one has the identity

a ∩ (b ∩ ξ ) = ( a ∪ b) ∩ ξ.

Hence the cap product makes the homology H∗ ( X ) a module over the ring
H∗ ( X ).
 poincaré duality 167

Remark 8.4.5. A relative cap product

∩
H i ( X, A) ⊗ Hn ( X, A) −
→ Hn−i ( X ) (8.4.6)

can be defined as follows. First note that the restriction

∩
Ci ( X, A) ⊗ Cn ( X ) −
→ Cn−i ( X )

of absolute cap product (8.4.1) vanishes on Ci ( X, A) ⊗ Cn ( A), so it

induces:
∩
Ci ( X, A) ⊗ Cn ( X, A) −
→ Cn−i ( X ).
Since (8.4.4) still holds in this relative setting, we get a relative cap
product operation:
∩
H i ( X, A) ⊗ Hn ( X, A) −
→ Hn−i ( X ).

The following result states that the cap product ∩ is functorial. Its
proof is a direct consequence of the definition of cap products and is
left as an exercise:

Lemma 8.4.6. If f : X → Y is a continuous map, then

φ ∩ f ∗ ξ = f ∗ (( f ∗ φ) ∩ ξ ) (8.4.7)

for all φ ∈ H i (Y ) and ξ ∈ Hn ( X ). This fact is illustrated in the following

diagram:
∩
H i ( X ) ⊗ Hn ( X ) −−−−→ Hn−i ( X )
x  
f ∗ f∗ y f∗ y
  

∩
H i (Y ) ⊗ Hn (Y ) −−−−→ Hn−i (Y )

Let us next move towards the definition of the Poincaré duality

map. Let M be a n-dimensional orientable connected manifold (not
necessarily compact), and let K ⊂ L ⊂ M where K, L are compact
subsets. Consider the diagram:
∩
H i ( M, M \ L) ⊗ Hn ( M, M \ L) −−−−→ Hn−i ( M)
x 
i∗ 
 
i∗ y

∩
H i ( M, M \ K ) ⊗ Hn ( M, M \ K ) −−−−→ Hn−i ( M)

By the functoriality of the cap product, we have for any φ ∈ H i ( M, M \

K ) that:
( i ∗ φ ) ∩ µ L = φ ∩ i ∗ ( µ L ), (8.4.8)

where µK and µ L denote the orientation classes of Theorem 8.2.7. More-

over, the following identification holds:
168 algebraic topology

Lemma 8.4.7. For compact subsets K ⊂ L of M, we have:

i∗ (µ L ) = µK . (8.4.9)

Proof. The claim follows from the commutativity of the following di-
agram and the uniqueness of µK in Hn ( M, M \ K ) which restricts to
local orientations µ x , ∀ x ∈ K.

µK ∈ Hn ( M, M \ K ) Hn ( M, M \ x )

i∗

µ L ∈ Hn ( M, M \ L)

Therefore, we have from (8.4.8) and (8.4.9) that:

(i ∗ φ ) ∩ µ L = φ ∩ i ∗ ( µ L ) = φ ∩ µ K , (8.4.10)

for all φ ∈ H i ( M, M \ K ). Let us now recall from Proposition 8.3.4 that

we have an isomorphism:

Hci ( M) ∼
= lim H i ( M, M \ K ), (8.4.11)
−→
K

where the direct limit on the right-hand side is taken over all compact
subsets K of M. We can now define the Poincaré duality map
∩
Hci ( M) −→ Hn−i ( M) (8.4.12)

as follows: its value on φ ∈ Hci ( M ) is defined as φK ∩ µK , where

φK ∈ H i ( M, M \ K ) is a representative of φ and µK ∈ Hn ( M, M \ K ) is
the orientation class defined by K (cf. Theorem 8.2.7). Note that the
Poincaré duality map (8.4.12) is well-defined (i.e., independent of the
choice of the representative φK ) by the commutativity of the following
diagram (which follows from the identity (8.4.10)):

i∗
H i ( M, M \ K ) H i ( M, M \ L)

∩µK ∩µ L

Hn−i ( M)

8.5 The Poincaré Duality Theorem

We have now all the necessary ingredients to formulate and prove the
main theorem of this chapter:
 poincaré duality 169

Theorem 8.5.1 (Poincaré Duality). If M is an n-dimensional oriented

connected manifold, then the Poincaré duality map:
∩
Hci ( M) −→ Hn−i ( M)

is an isomorphism for all i.

Proof. Recall that on an element

φ ∈ Hci ( M ) ∼
= lim H i ( M, M \ K ),
−→
K⊂X
K −compact

the Poincaré duality map takes the value φK ∩ µK , with φK ∈ H i ( M, M \

K ) a representative of φ, and µK the orientation class of Hn ( M, M \ K ).
The proof of the theorem will be divided into several steps. We first
show that the statement holds locally, then we glue the local isomor-
phisms by a Mayer-Vietoris argument.

Step I: We first show that the theorem holds for M = Rn .

Let Bk denote the closed ball of integer radius k in Rn . Then

Z if i = n
Hci (Rn ) ∼
= lim
−
→
H i
( R n
, R n
\ Bk ) ∼
=
B 0 otherwise
k

and 
Z if i = n
Hn−i (Rn ) ≃
0 otherwise.

The Universal Coefficient Theorem yields that

H n (Rn , Rn \ Bk ) ∼
= Hom( Hn (Rn , Rn \ Bk ); Z).

So H n (Rn , Rn \ Bk ) is generated by some class ak so that ak (µ Bk ) = 1 ∈

Z. Let 1 ∈ H 0 (Rn ) = Z be the generator. Then:

1 = ak (µ Bk ) = (1 ∪ ak )(µ Bk ) = 1( ak ∩ µ Bk ).

Hence ak ∩ µ Bk is a generator of H0 ( Rn ). In particular, the map

∩µ Bk : H n (Rn , Rn \ Bk ) → H0 (Rn )

is an isomorphism. Taking the direct limit over the Bk ’s, we get an

isomorphism
∼
=
Hcn (Rn ) −→ H0 (Rn ),
which by the above considerations coincides with the Poincaré duality
map. Also, both groups are trivial for i ̸= n, so the claim follows.
170 algebraic topology

Step II: Assuming the theorem holds for opens U, V ⊂ M and for their
intersection U ∩ V, we show that it holds for the union U ∪ V.
For this purpose, we construct a commutative diagram

··· Hci (U ∩ V ) Hci (U ) ⊕ Hci (V ) Hci (U ∪ V ) Hci+1 (U ∩ V ) ···

··· Hn−i (U ∩ V ) Hn−i (U ) ⊕ Hn−i (V ) Hn−i (U ∪ V ) Hn−i−1 (U ∩ V ) ···

(8.5.1)
Once the diagram is constructed, the claim follows by the 5-lemma.
The bottom row in (8.5.1) is just the Mayer-Vietoris homology sequence.
The top row of the above diagram can be constructed as follows. For
compact subsets K ⊂ U and L ⊂ V, consider the cohomology Mayer-
Vietoris sequence for the pairs ( M, M \ K ) and ( M, M \ L):

· · · → H i ( M, M \ (K ∩ L)) → H i ( M, M \ K ) ⊕ H i ( M, M \ L)
→ H i ( M, M \ (K ∪ L)) → · · ·

By excision, we get a long exact sequence:

· · · → H i (U ∩ V, U ∩ V \ K ∩ L) → H i (U, U \ K ) ⊕ H i (V, V \ L)
→ H i (U ∪ V, U ∪ V \ K ∪ L) → · · ·

Taking direct limits over K ⊂ U and L ⊂ V, we get the top long exact
sequence in (8.5.1):

· · · → Hci (U ∩ V ) → Hci (U ) ⊕ Hci (V ) → Hci (U ∪ V ) → · · ·

The commutativity follows by using the definition of the Poincaré dual-

ity map.

Step III: Assume M is a union of nested open subsets Uα so that the

theorem holds for each Uα . We show that the theorem holds for M.
First note that any compact subset in M (in particular, the support of
a singular (co)chain) is contained in some Uα . Then we claim that the
following identifications hold:

Hi ( M ) = lim
−
→
Hi (Uα ) (8.5.2)
α

and
Hci ( M ) = lim
−
→
Hci (Uα ). (8.5.3)
α

This claim and Poincaré duality for each Uα imply the Poincaré dual-
ity isomorphism for M, since the direct limit of isomorphisms is an
isomorphism. In order to prove the claim, we note that the inclusions
iα : Uα ,→ M induce homomorphisms iα∗ : Hi (Uα ) → Hi ( M) so that
for Uα ,→ Uβ the following diagram commutes:
 poincaré duality 171

Hi (Uα ) Hi (Uβ )

Hi ( M )

We therefore get a well-defined map

f : lim
−
→
Hi (Uα ) → Hi ( M).
α

We next show that f is an isomorphism.

• f is onto: any [ξ ] ∈ Hi ( M ) is represented by a cycle whose sup-

port is contained in a compact subset of M, thus in some Uα . The
corresponding homology class in Hi (Uα ) maps onto [ξ ].

• f is one-to-one: if ξ = ∂η, for η ∈ Ci+1 ( M), then ξ is a cycle in some

Uα , but not necessarily a boundary in Uα . On the other hand, η is
contained in some larger Uβ , so ξ can be regarded as a boundary in
Uβ . Therefore, [ξ ] = 0 ∈ Hi (Uβ ), hence it represents the zero class
in lim
−
→
Hi (Uα ).
α

So (8.5.2) follows. The identification in (8.5.3) is obtained similarly.

Step IV: We next show that the theorem holds when M is an open
subset of Rn .
If M is convex, then M is homeomorphic to Rn , so the theorem holds
S
by Step I. If M is not convex, then M = k∈Z>0 Vk , with each Vk open
and convex in Rn . By induction and Step II, the theorem holds for the
sets Uk = V1 ∪ · · · ∪ Vk . Note that {Uk }k forms a nested cover of opens
for M, hence the theorem follows by Step III.

Step V: Finally, we show that the Poincaré duality isomorphism holds

for an arbitrary M.
We first cover M by open sets Vα , each of which is homeomorphic to
an open subset of Rn . We next choose a well ordering < of the index
set, which exists by Zorn’s lemma (if M has a countable basis, the we
can choose the positive integer as index set). Then the sets
[
Uα := Vβ .
β<α

form a nested open cover of M. So by Step III, it suffices to show that the
theorem holds for each Uα . But Uα = ∪ β<α Vβ , and the theorem holds
for each Vβ by Step IV. By Step II, Step III, and transfinite induction,
the theorem holds for each Uα , and the claim follows.
 172 algebraic topology

Remark 8.5.2. By taking coefficients in any commutative ring R, we can

prove the Poincaré duality isomorphism over R via the coefficient map
Z → R. Moreover, for R = Z/2, Poincaré duality holds even without
the orientability assumption.

As an immediate consequence of Theorem 8.5.1, we get the following:

Corollary 8.5.3. If M is an n-dimensional closed oriented connected manifold,

then the map
∩
H i ( M ) −→ Hn−i ( M )

defined by the cap product with the fundamental class of M, that is, φ 7→
φ ∩ [ M], is an isomorphism for all i.

Exercises
1. Show that if Mn is a connected, non-compact manifold, then

Hi ( M; Z) = 0 for i ≥ n.

2. Show that the Euler characteristic of a closed, oriented, (4n + 2)-

dimensional manifold is even.

3. Let M be a closed oriented manifold with fundamental class [ M ].

Consider the cup product pairing between cohomology groups of com-
plementary dimensions (after moding out by the corresponding torsion
subgroups):

( , ) : H i ( M; Z)/Torsion ⊗ H n−i ( M; Z)/Torsion → Z

given by (α, β) = ⟨α ∪ β, [ M ]⟩. Here ⟨ , ⟩ : H n ( X; Z) ⊗ Hn ( X; Z) → Z

is the Kronecker pairing defined in Homework #1.

(i) Show that the cup product pairing is nonsingular in the following
sense: for each choice of a Z-basis { β 1 , · · · , β r } of the free abelian
group H n−i ( M; Z)/Torsion, there exists a Z-basis {α1 , · · · , αr } of
H i ( M; Z)/Torsion such that (αi , β j ) = δij . (Hint: Use the Universal
Coefficient Theorem and Poincaré Duality.)

(ii) As an application, re-prove the following facts about the ring struc-
tures on the cohomology of projective spaces:

(a) H ∗ (RPn ; Z2 ) ∼
= Z2 [ x ]/( x n+1 ), | x | = 1,
(b) H ∗ (CPn ; Z) ∼
= Z [ y ] / ( y n +1 ), |y| = 2,
(c) H ∗ (HPn ; Z) ∼
= Z [ w ] / ( w n +1 ), |w| = 4.
 poincaré duality 173

4. Let M be a closed, oriented 4n-dimensional manifold, with funda-

mental class [ M ]. The middle intersection pairing

( , ) : H 2n ( M; Z)/Torsion ⊗ H 2n ( M; Z)/Torsion → Z

given by (α, β) = ⟨α ∪ β, [ M ]⟩ is symmetric and nondegenerate. Let

{α1 , · · · , αr } be a Z-basis of H 2n ( M; Z)/Torsion, and let A = aij for

aij := (αi , α j ) ∈ Z. Then A is a symmetric matrix with det( A) = ±1, so

it is diagonalizable over R. Define the signature of M to be

σ ( M ) := #(positive eigenvalues) − #(negative eigenvalues)

(a) Compute σ (CPn ), σ (S2 × S2 ).

(b) Show that the signature σ ( M ) is congruent mod 2 to the Euler

characteristic χ( M ).

5. Show that if a connected manifold M is the boundary of a compact

manifold, then the Euler characteristic of M is even. Conclude that
RP2n , CP2n , HP2n cannot be boundaries.

6. Show that if M4n is a connected manifold which is the boundary of a

compact oriented (4n + 1)-dimensional manifold V, then the signature
of M is zero.

7. Show that if M is a compact contractible n-manifold then ∂M is a

homology (n − 1)-sphere, that is, Hi (∂M; Z) ≃ Hi (Sn−1 ; Z) for all i.

8. Let M be a closed, connected, orientable 4-manifold with fundamen-

tal group π1 ( M ) ∼
= Z/3 ∗ Z/3 and Euler characteristic χ( M) = 5.

(a) Compute Hi ( M, Z) for all i.

(b) Prove that M is not homotopy equivalent to any CW complex with

no 3-cells.

9. Let M be a closed, connected, oriented n-manifold and let f : Sn → M

be a continuous map of non-zero degree, i.e., the morphism

f ∗ : Hn (Sn ; Z) → Hn ( M; Z)

is non-trivial. Show that M and Sn have the same Q-homology.

10. Show that there is no orientation-reversing self-homotopy equiva-

lence CP2n → CP2n .
174 algebraic topology

8.6 Immediate applications of Poincaré Duality

In this section we derive several applications of the Poincaré duality

isomorphism of Theorem 8.5.1. (In particular, we provide answers to
some of the exercises listed in the previous section.)

Proposition 8.6.1. If Mn is a closed odd dimensional manifold, then

χ( M) = 0.

Proof. Let n = 2k + 1.
If M is oriented, then (with Z-coefficients):

( P.D.) (UCT )
rkHi ( M ) = rkH n−i ( M) = rkHn−i ( M).

So:
2k +1 k  
χ( M) = ∑ (−1)i · rkHi ( M) = ∑ (−1)i + (−1)n−i · rkHi ( M) = 0.
i =0 i =0

If M is non orientable, the Poincaré duality isomorphism holds with

Z/2-coefficients, and we get:

2k +1
χ( M) := ∑ (−1)n · rkHi ( M; Z)
n =0

(∗) 2k+1
= ∑ (−1)i · dimZ/2 Hi ( M; Z/2)
n =0
= 0,

where the vanishing follows as before by Poincaré duality (over Z/2).

The equality (∗) follows from the Universal Coefficient Theorem as
follows:

H i ( M, Z/2) = Hom( Hi ( M), Z/2) ⊕ Ext( Hi−1 ( M), Z/2).

• a Z-summand of Hi ( M; Z) contributes

– Hom(Z, Z/2) = Z/2 to H i ( M; Z/2), and

– Ext(Z, Z/2) = 0 to H i+1 ( M; Z/2).

• a Z/m summand of Hi ( M; Z), with m odd, contributes:

– Hom(Z/m, Z/2) = 0 to H i ( M; Z/2), and

– Ext(Z/m, Z/2) = 0 to H i+1 ( M; Z/2).

• a Z/m summand of Hi ( M; Z), with m even, contributes:

– Hom(Z/m, Z/2) = Z/2 to H i ( M; Z/2), and

 poincaré duality 175

– Ext(Z/m, Z/2) = Z/2 to H i+1 ( M; Z/2), so these Z/2 contribu-

tions cancel out in ∑i (−1)i · dimZ/2 H i ( M; Z/2).

Finally, note that dimZ/2 Hi ( M; Z/2) = dimZ/2 H i ( M; Z/2), so the

claim follows.

Proposition 8.6.2. If Mn is a closed, oriented, connected manifold, then

Torsion( Hn−1 ( M )) = 0.

Proof. Indeed,
( P.D.)
Torsion( Hi ( M )) = Torsion( H n−i ( M ))
(UCT )
= Ext( Hn−1−i ( M), Z)
= Torsion( Hn−1−i ( M))

Since M is connected, H0 ( M ) is free, so the claim follows.

We will show later the following:

Proposition 8.6.3. If Mn is a closed, connected, non-orientable manifold,

then
Torsion( Hn−1 ( M)) = Z/2
and
H n ( M ) = Z/2.
The second part of Proposition 8.6.3 follows from the Universal Coeffi-
cient Theorem and the following consequence of Poincaré duality (to
be proved in the next section):

Lemma 8.6.4. If Mn is an n-dimensional closed, connected manifold, then


Z , if M-oriented
Hn ( M ) =
0 , if M-non-oriented.

8.7 Addendum to orientations of manifolds

Before we explain the proof of Proposition 8.6.3, we need to elaborate

on orientations of manifolds.
Recall that if Mn is a n-manifold, a local orientation at x ∈ M is a
generator µ x ∈ Hn ( M, M \ x ) ∼= Z. We say that M is oriented if there
exists a global orientation, i.e., a continuous choice x → µ x of local
orientations. This means that for all x ∈ M, there is a closed euclidean
ball Bx (of finite positive radius) around x so that
ρy
Z∼
= Hn ( M, M \ Bx ) → Hn ( M, M \ y)

sends the generator µ Bx to the local orientation class µy , for all y ∈ Bx .

 176 algebraic topology

Proposition 8.7.1. Any manifold M (oriented or not) has an oriented double

cover M.
e

Proof. (Sketch)
Define

e := {µ x | x ∈ M, µ x a local orientation of M at x }
M

and π : Me → M by µ x → x. Clearly, π is a 2 : 1 map.

We need to put a topology on M e so that it becomes a manifold and
π is a covering map. For an open ball B ⊂ Rn ⊂ M of finite radius,
with a generator µ B ∈ Hn ( M, M \ B), define

U (µ B ) = {µ x ∈ M
e | x ∈ B, µ x = ρ x (µ B )},

where ρ x denotes the natural map Hn ( M, M \ B) → Hn ( M, M \ x ).

Then
π −1 ( B) = U (µ B ) ⊔ U (−µ B )

and both U (µ B ) and U (−µ B ) are in bijection to B. Moreover, it can be

shown that the sets {U (µ B )} B form basis of opens for the topology of
Me so that π is continuous. So π is 2-fold covering and M e is manifold.
Moreover, M e is orientable. Indeed, we have,

Hn ( M,
e Me \ µx ) ∼
= Hn (U (µ B ), U (µ B ) \ µ x ) ∼
= Hn ( B, B \ x )
(8.7.1)
∼ Hn ( M, M \ x ).
=

So at the point µ x ∈ M
e there exists a canonical local orientation

ex ∈ Hn ( M,
µ e Me \ µx ) ∼
=Z

corresponding to µ x under the above isomorphism (8.7.1). The consis-

tency of such local orientations follows by construction.

Example 8.7.2.(a) The oriented double cover of M = RP2 is M

e = S2 .

(b) The oriented double cover of the Klein bottle K is the 2-torus T 2 .

Proposition 8.7.3. If M is a connected manifold, then M is orientable if, and

e has two components. In particular, if π1 ( M) = 0 or has no index
only if, M
2 subgroup, then M is orientable.

Proof. The oriented double cover M e can have one or two components.
If M has two components, each is oriented and homeomorphic to M,
e
so M is orientable. Conversely, if M is orientable, it can have exactly
two orientations at each point, each defining a sheet of M.
e

Example 8.7.4. CPn is orientable.

 poincaré duality 177

The oriented double cover M

e can be embedded in a larger covering
space MZ of M as follows. Let

MZ = {α x | x ∈ M, α x ∈ Hn ( M, M \ x ) = Z}.

We then have the Z-fold projection map

πZ : MZ → M

defined by α x → x. A basis of opens {U ( B)} for MZ can be defined by

the following recipe: for an open ball B ⊂ Rn ⊂ M, set

U ( B) = {α x | x ∈ B, α x = ρ x (α B ) for α B ∈ Hn ( M, M \ B) ∼
= Z)}
∼
=
with ρ x : Hn ( M, M \ B) → Hn ( M, M \ x ) induced by inclusion as before.
For any k ∈ Z, we then get a subcover Mk ⊂ MZ by selecting ±kµ x in
the fibre above x. So
[
MZ = Mk
k ≥0

with M0 ∼
= M, Mk ∼
= M−k , and Mk ∼
= M,
e for any integer k.

Definition 8.7.5. A section of πZ : MZ → M is a continuous map α : M →

MZ defined by x 7→ α x ∈ Hn ( M, M \ x ) = Z. An orientation of M is a
section of πZ assigning µ x to each x ∈ M.
One can generalize the definition of orientability by replacing Z
with any commutative ring R with unit. Note that by the universal
coefficient theorem for homology, we have:

Hn ( M, M \ x; R) ∼
= Hn ( M, M \ x ) ⊗ R ∼
= Z⊗R ∼
= R.

The covering MZ can be generalized to:

MR = {α x | x ∈ M, α x ∈ Hn ( M, M \ x; R) ∼
= R }.

The corresponding covering map π R : MR → M is defined by α x 7→ x

(so the fibre over x ∈ M is R). Each r ∈ R determines a subcovering
Mr by selecting the points ±µ x ⊗ r ∈ Hn ( M, M \ x; R) in each fibre.
If r is an element of order 2 in R, then Mr is a copy of M. (Indeed,
±µ x ⊗ r = µ x ⊗ ±r = µ x ⊗ r.) Otherwise, Mr is homeomorphic to the
oriented double cover M.e We have
[
MR = Mr ,
r∈R

with all Mr being disjoint except for Mr = M−r , and Mr = M if 2r = 0.

Definition 8.7.6. An R-orientation of an n-dimensional manifold M is a
section of MR assigning to each x ∈ M a generator u of Hn ( M, M \ x; R) ∼
=
R.
 178 algebraic topology

Remark 8.7.7. Note that a generator of R is an element u so that Ru = R.

Since R has a unit, this is equivalent to saying that u is invertible in R.

Remark 8.7.8. An orientable manifold is R-orientable, for all commuta-

tive rings R with unit. A non-orientable manifold is R-orientable iff R
contains a unit of order 2. Thus every manifold is Z/2-orientable.

We are now ready to prove the following result, which shows that
orientability of a closed manifold is reflected in the structure of its
homology:

Theorem 8.7.9. Let M be a closed connected n-manifold. Then:

(a) if M is (R-)orientable, then Hn ( M; R) → Hn ( M, M \ x; R) ∼

= R is an
isomorphism for any x ∈ M.

(b) if M is not orientable, then Hn ( M; R) → Hn ( M, M \ x; R) ∼ = R is one-

to-one, with image the group generated by the set of elements of order 2 in
R.

(c) Hi ( M; R) = 0, for all i > n.

The proof of Theorem 8.7.9 is based on the Theorem 8.2.7 and Lemma
8.2.10 (which we formulate here with R-coefficients in parts (a) and (b)
below), together with a slight generalization of Theorem 8.2.7 (see part
(c) below) which holds without the orientability assumption:

Lemma 8.7.10. Let M be a connected n-manifold and K a compact subset of

M. Then:

(a) if M is R-oriented, there exists a unique µK ∈ Hn ( M, M \ K; R) such that

ρ x (µK ) = µ x ∈ Hn ( M, M \ x; R), for all x ∈ K.

(b) Hi ( M, M \ K; R) = 0 for i > n, and a class αK ∈ Hn ( M, M \ K; R) is

zero iff ρ x (αK ) = 0 for any x ∈ K.

(c) if x 7→ α x is a section of the covering space MR → M, then there is a unique

class αK ∈ Hn ( M, M \ K; R) so that ρ x (αK ) = α x ∈ Hn ( M, M \ x; R),
for all x ∈ K.

Note that the proof of part (c) of the above lemma is almost identical
to that of Theorem 8.2.7 (with the uniqueness following from part (b)),
with the only easy modification appearing in Step I of loc.cit. (where the
orientation assumption used in the proof of Theorem 8.2.7 is replaced
by the continuity of the section). We leave the details to the reader.

To deduce parts (a) and (b) of Theorem 8.7.9, choose K = M in the

above lemma, and let Γ R ( M ) be the set of sections of the covering map
MR → M. With respect to the addition of functions and multiplication
 poincaré duality 179

by scalars in R, Γ R ( M) becomes an R-module. Moreover, there exists a

homomorphism
Hn ( M; R) −→ Γ R ( M )

defined by
α → ( x 7 → α x ),

where α x is the image of α under the map ρ x : Hn ( M; R) → Hn ( M, M \

{ x }; R). The above lemma asserts that this is in fact an isomorphism.
Let us now translate the statements about Hn ( M; R) in Theorem 8.7.9
into statements about the R-module Γ R ( M ):

1. For the oriented case: Hn ( M; R) ∼

= Γ R ( M) → Hn ( M, M \ x; R) is an
isomorphism, defined by α 7→ ( x 7→ α x ) 7→ α x for a given x.

2. For the non-oriented case: Hn ( M; R) ∼

= Γ R ( M) → Hn ( M, M \ x; R) is
a monomorphism, with image the group generated by the elements
of order 2 in R.

Note that since M is connected, each section in Γ R ( M) is determined by

its value at one point x ∈ M. The injectivity statements in part (a) and
(b) of Theorem 8.7.9 follow from Lemma 8.7.10(b). Also, the surjectivity
in part (a), as reformulated in part 1 above, follows from Lemma
8.7.10(a). The remaining statement in part 2 above can be seen as follows.
Since π R is a covering map, the section group Γ R ( M) can be identified
with the connected components of MR which map homeomorphically
via π R to M. Since M is non-orientable, the oriented double cover
π:M e → M is non-trivial (i.e., connected), thus the components of MR
are of the form r ( M e → MR the continuous map defined by
e ), with r : M
µ 7→ µ ⊗ r. The only points in r ( Me ) which under π R map to x ∈ M are
µ x ⊗ r and −µ x ⊗ r = µ x ⊗ (−r ). Thus, π R |r M
e is a homeomorphism iff
r = −r, or 2r = 0. □

Corollary 8.7.11. If M is orientable, then Hn ( M; Z) ∼ = Z. If M is non-

orientable, then Hn ( M; Z) = 0. In either case, Hn ( M; Z/2) = Z/2.

We can now prove the following:

Corollary 8.7.12. Let M be a closed and connected n-manifold. If M is

oriented, then
Torsion( Hn−1 ( M )) = 0.

Otherwise,
Torsion( Hn−1 ( M )) = Z/2.

Proof. By the universal coefficient theorem for homology, and using

the fact that the homology groups of a closed manifold are finitely
180 algebraic topology

generated (e.g., see Corollaries A.8 and A.9 in Hatcher’s book), we

have:

Hn ( M; Z/p) = Hn ( M; Z) ⊗ Z/p ⊕ Tor( Hn−1 ( M; Z), Z/p)

= Hn ( M; Z) ⊗ Z/p ⊕ Torsion( Hn−1 ( M; Z)) ⊗ Z/p.

In the orientable case, if Hn−1 ( M) contained torsion, then for some

prime p, the group Hn ( M; Z/p) = Z/p would be larger than the Z/p
coming from the first summand (here we use that Hn ( M) = Z), which
is impossible. This means Torsion( Hn−1 ( M)) = 0.
In the non-orientable case, we have by Theorem 8.7.9 that Hn ( M; Z/m)
is either Z/2 or 0, depending on whether m is even or odd. (Indeed,
in this case the map Hn ( M; Z/m) → Z/m is injective with image the
elements of order 2 in Z/m. So, if m is odd, there are no elements of
order 2 in Z/m, while if m = 2k is even, then k is the only element of
order 2 in Z/m.) Since in this case we have Hn ( M; Z) = 0, this forces
the torsion subgroup of Hn−1 ( M ) to be Z/2.

Remark 8.7.13. By using the universal coefficient theorem for the coho-
mology of a closed n-manifold, we have:

H n ( M) = Free( Hn ( M)) ⊕ Torsion( Hn−1 ( M)).

So by using the result of and the previous corollary, we get that if M is

oriented then H n ( M) = Z. Otherwise, H n ( M ) = Z/2.

8.8 Cup product and Poincaré Duality

Let R be a fixed commutative coefficient ring, and fix φ ∈ C l ( M; R),

ψ ∈ C k ( M; R) and σ ∈ Ck+l ( M; R). Recall that the cap product ψ ∩ σ ∈
Cl ( M; R) is defined by

φ(ψ ∩ σ ) = ( φ ∪ ψ)(σ ) ∈ R. (8.8.1)

Alternatively, if σ is a (k + l )-simplex, then

ψ ∩ σ = ψ(σ |[vl ,vl +1 ,...,vk+l ] ) · σ|[v0 ,v1 ,...,vl ] . (8.8.2)

Indeed,

φ(ψ ∩ σ ) = ψ(σ |[vl ,vl +1 ,...,vk+l ] ) · φ(σ |[v0 ,v1 ,...,vl ] ) = ( φ ∪ ψ)(σ). (8.8.3)

This means that − ∪ ψ : C l ( M; R) → C k+l ( M; R) is dual to ψ ∩ − :

Ck+l ( M; R) → Cl ( M; R). Passing to (co)homology, we get the following
commutative diagram:

h
H l ( M; R) −−−−→ HomR ( Hl ( M; R), R)
 
∪ψy (ψ∩)∗ y
 

h
H k+l ( M; R) −−−−→ HomR ( Hk+l ( M; R), R)
 poincaré duality 181

In particular, if h is an isomorphism (e.g., R is a field, or we work over

Z but H∗ is torsion-free), then − ∪ ψ and ψ ∩ − determine each other.

Definition 8.8.1. Let M be a closed connected R-oriented n-manifold. Then

the cup product pairing

∩[ M]
H k ( M; R) × H n−k ( M; R) −→ H n ( M; R) −→ H0 ( M; R) = R (8.8.4)

is defined by
( φ, ψ) 7→ ( φ ∪ ψ) 7→ ( φ ∪ ψ) ∩ [ M].

Definition 8.8.2. Let A and B be R-modules. A pairing α : A × B → R

is non-singular if f : A → HomR ( B, R) is an isomorphism, with f defined
by f ( a)(b) = α( a, b), and g : B → HomR ( A, R) is an isomorphism, with
g(b)( a) = α( a, b).

We then have the following:

Proposition 8.8.3. The cup product pairing is non-singular if R is a field, or

if R = Z and torsion is factored out.

Proof. Consider the composition

h ( P.D.)∗
f : H k ( M; R) −→ HomR ( Hk ( M; R), R) −→ HomR ( H n−k ( M; R), R),

where ( P.D.)∗ denotes the dual of the Poincaré duality isomorphism.

Under our assumptions on R, h is isomorphism. Moreover, by Poincaré
Duality, ( PD )∗ is also an isomorphism, hence f is an isomorphism. For
φ ∈ H k ( M; R) and ψ ∈ H n−k ( M; R), we have:

f ( φ)(ψ) = (( P.D.)∗ ◦ h( φ))(ψ)

= h( φ)( P.D.(ψ))
= h( φ)(ψ ∩ [ M])
= φ(ψ ∩ [ M ])
= ( φ ∪ ψ)[ M].

We obtain a similar isomorphism by interchanging k with n − k, so the

claim follows.

Corollary 8.8.4. Let M be a closed connected Z-oriented n-manifold. Then

for any α ∈ H k ( M ) a generator of a Z-summand, there exists β ∈ H n−k ( M )
such that α ∪ β generates H n ( M ) ∼= Z.

Proof. By hypothesis, there exists a homomorphism (i.e., the projection

to some Z-summand)
φ : H k ( M) → Z
182 algebraic topology

such that φ(α) = 1. By the non-singularity of the cup product pairing,

φ is realized by taking the cup product with some β ∈ H n−k ( M) and
evaluating on the fundamental class [ M]. We therefore get

1 = φ(α) = (α ∪ β)[ M ].

This means α ∪ β is the generator of H n ( M ).

Corollary 8.8.5. H ∗ (CPn ; Z) ∼

= Z[α]/(αn+1 ), with deg(α) = 2.

Proof. Let α be the generator of H 2 (CPn ) = Z. By induction, we

can assume that αn−1 generates H 2n−2 (CPn ) = Z. Using the previ-
ous corollary, there exists β ∈ H 2 (CPn ) so that αn−1 ∪ β generates
H 2n (CPn ) = Z. Note that since α is the generator of H 2 (CPn ) = Z, it
follows that β = mα, for some m ∈ Z. This means that αn−1 ∪ β = mαn
generates Z. Thus m = ±1, whence αn generates H 2n (CPn ).

We can now ask the following:

Question 8.8.6. Does there exist a 2n-dimensional closed manifold whose

cohomology is additively isomorphic to that of CPn , but with a different cup
product structure?

If n = 2, the answer is No. Indeed, H ∗ (CP2 ; Z) = Z[α]/(α3 ), with

deg(α) = 2. If there is such a manifold M, then α also generates
H 2 ( M ) = H 2 (CP2 ) = Z, so there exists β ∈ H 2 ( M) such that α ∪ β gen-
erates H 4 ( M ) = Z. So, β = mα, for some m ∈ Z. Hence α ∪ β = mα2
generates H 4 ( M ), which yields m = ±1. This means that M has the
same cup product structure as CP2 .
If n ≥ 3, the answer is Yes. Indeed, S2 × S4 and CP3 have isomorphic
cohomology groups, but different cup product structures on their coho-
mology rings.

Another application of Poincaré duality is the following:

Corollary 8.8.7. If M is a closed oriented manifold of dimension m = 4n + 2,

then χ( M) is even.

Proof. By the definition of the Euler characteristic we have

4n+2
χ( M) = ∑ (−1)i · rk( Hi ( M)).
i =0

By Poincaré duality, we obtain

rk( Hi ( M )) = rk( Hm−i ( M)).

Therefore,
χ( M ) ≡ rk( H2n+1 ( M )) (mod 2).
 poincaré duality 183

Let us now consider the following cup product pairing

∪ ∩[ M]
H 2n+1 ( M ) × H 2n+1 ( M ) → H 4n+2 ( M) −→ Z

defined by
(α, β) 7→ (α ∪ β) 7→ (α ∪ β) ∩ [ M].
By Poincaré Duality, after moding out by torsion, this pairing is non-
singular. As a result, the matrix A of the cup product pairing is
non-singular and anti-symmetric. By linear algebra, A is similar to a
matrix with diagonal blocks
!
0 −1
−1 0

Therefore,
rk( H 2n+1 ( M )) = rk( A),
which is clearly even.

Remark 8.8.8. Dualizing the cup product pairing of Proposition 9.11.16,

we get the non-singular intersection pairing

Hk ( M ) × Hn−k ( M) → Z

defined by
([σ], [η ]) → ♯(σ ∩ η ′ ),
where η ′ is chosen so that it is homologous to η but transversal to σ (so
σ ∩ η ′ is a finite number of points).
Example 8.8.9. Let T be the 2-dimensional torus and S be a meridian
of T. Let M be the pinched torus T/S.

Figure 8.1: pinched torus

Then Poincaré duality fails for M. If not, let α be the longitude of M

(and T) and β be the a meridian of M. Then Poincaré duality for
184 algebraic topology

M would yield ([α], [ β]) → ♯(α ∩ β) = 1. However, [ β] = 0. This is

impossible since the intersection pairing is non-singular. The reason for
the failure of Poincaré duality is that the pinched torus M := T/S is
not a manifold. Indeed, a neighborhood of the pinch point is a wedge
of two 2-disks, thus it is not homeomorphic to R2 .

Exercises
1. Let Mg be a closed orientable surface of genus g ≥ 1. Show that for
each non-zero α ∈ H 1 ( M; Z) there exists β ∈ H 1 ( M; Z) with α ∪ β ̸= 0.
Deduce that M is not homotopy equivalent to a wedge sum X ∨ Y of
CW-complexes with non-trivial reduced homology. Do the same for
closed nonorientable surfaces using cohomology with Z2 -coefficients.

8.9 Manifolds with boundary: Poincaré duality and applications

In this section, we discuss the Poincaré duality theorem for manifolds

with boundary. The proofs are routine adaptation of those for closed
manifolds.

Definition 8.9.1. A Hausdorff topological space M is an n-manifold with

boundary if any point x ∈ M has a neighborhood Ux homeomorphic to either
Rn or Rn+ := {( x1 , · · · , xn ) ∈ Rn | xn ≥ 0}. In particular,

(a) if Ux ∼
= Rn , then Hn ( M, M \ x ) ∼
= Hn (Ux , Ux \ x ) ∼
= Z.

(b) if Ux ∼
= Rn+ , then

Hn ( M, M \ x ) ∼
= Hn (Ux , Ux \ x ) ∼
= Hn (Rn+ , Rn+ − {0}) ∼
= 0.

The boundary of M is defined to be

∂M := { x ∈ M | Hn ( M, M \ x ) = 0}.

Example 8.9.2. ∂( D n ) = Sn−1 , ∂(Rn+ ) = Rn−1 .

Remark 8.9.3. If M is an n-manifold with boundary, then the boundary

set ∂M is a manifold of dimension n − 1.

Definition 8.9.4. We say that a manifold with boundary ( M, ∂M) is ori-

entable if M \ ∂M is orientable as a manifold with no boundary.

We have the following:

Proposition 8.9.5. If ( M, ∂M ) is a compact, orientable n-manifold with ori-

ented boundary, then there exists a unique class µ M ∈ Hn ( M, ∂M ) inducing
local orientations µ x ∈ Hn ( M, M \ x ) at all points x ∈ M \ ∂M.
 poincaré duality 185

Remark 8.9.6. If ( M, ∂M ) is a compact, orientable n-manifold with

boundary, then in the long exact sequence for the pair ( M, ∂M ) we
have:
∂
Hn ( M, ∂M) −→ Hn−1 (∂M )
[ M] = µ M 7−→ [∂M]

Theorem 8.9.7 (Poincaré Duality). If ( M, ∂M) is a connected, oriented

n-manifold with boundary, then there are isomorphisms
∩µ M
Hci ( M) −−
∼
→ Hn−i ( M, ∂M) (8.9.1)
=

and
∩µ M
Hci ( M, ∂M) −−
∼
→ Hn−i ( M) (8.9.2)
=
where Hci ( M, ∂M ) := lim Kcompact H i ( M, ( M \ K ) ∪ ∂M ) is the cohomology
−→
K ⊂ M \∂M
with compact support for the pair ( M, ∂M ).
Let us now describe some applications of Poincaré duality for mani-
folds with boundary.

Proposition 8.9.8. If Mn = ∂V n+1 is a connected manifold with V a compact

(n + 1)-dimensional manifold with boundary, then the Euler characteristic
χ( M ) is even.
An immediate consequence of Proposition 8.9.8 is the following:

Corollary 8.9.9. RP2n , CP2n , HP2n cannot be boundaries of compact mani-

folds.

In order to prove Proposition 8.9.8, we need the following result:

Proposition 8.9.10. Assume V 2n+1 is an oriented, (2n + 1)-dimensional
compact manifold with connected boundary ∂V = M2n . If R is a field (e.g.,
Z/2Z if M is non-orientable), then dimR H n ( M; R) = dimR Hn ( M; R) is
even.

Proof of Proposition 8.9.10. Start with the cohomology long exact se-
quence for the pair (V, M ):
i∗ δ
H n (V; R) - H n ( M; R) - H n+1 (V, M; R)

∼
= ∩[ M] ∼
= ∩[V ]
? i∗ - ?
Hn ( M; R) Hn (V; R)
where i∗ , i∗ are induced by the inclusion i : M = ∂V ,→ V. By exactness,
P.D.
we have that Im i∗ ∼
= ker δ ∼
= ker i∗ , so

dim(Im i∗ ) = dim(ker i∗ ) = dim Hn ( M; R) − dim(Im i∗ ).

186 algebraic topology

Since i∗ , i∗ are Hom-dual, we have that dim(Im i∗ ) = dim(Im i∗ ).

Altogether,

dim H n ( M; R) = dim Hn ( M; R) = 2 dim(Im i∗ )

is even.

Proof of Proposition 8.9.8. If n = dim M is odd, then Proposition 8.6.1

yields that χ( M) = 0, thus even. If n = 2m is even, then we work with
Z/2Z-coefficients and get:
2m
χ( M) = ∑ (−1)i dimZ/2 Hi ( M; Z/2)
i =0
m −1
(1)
=2 ∑ (−1)i dimZ/2 Hi ( M; Z/2) + (−1)m dimZ/2 Hm ( M; Z/2)
i =0
≡ dimZ/2 Hm ( M; Z/2) (mod 2)
(2)
≡ 0 (mod 2),

where equation (1) follows by Poincaré Duality, and congruence (2) is

by Proposition 8.9.10.

The proof of Proposition 8.9.10 also yields the following:

Corollary 8.9.11. Under the assumptions of Proposition 8.9.10, we have
the following:

(a) Im i∗ ⊂ H n ( M2n ; R) is self-annihilating with respect to cup product

∪, i.e., if α, β ∈ Im i∗ , then α ∪ β = 0.

(b) dim(Im i∗ ) = 1
2 dim H n ( M2n ; R).

Proof. For any α = i∗ (α), β = i∗ ( β) with α, β ∈ H n (V; R), we have

δ(α ∪ β) = δ(i∗ (α) ∪ i∗ ( β)) = δi∗ (α ∪ β) = 0

Hence, α ∪ β ∈ ker δ : H 2n ( M; R) → H 2n+1 (V, M; R) ∼

= 0, where the
last isomorphism follows by the following commutative diagram

δ
H 2n ( M; R) - H 2n+1 (V, M; R)
∼
= P.D. ∼
= P.D.
? ?
H0 ( M; R) - H0 (V; R)

with the bottom arrow an injection.

Exercises
1. Let X be the cone on CPn . Show that X is a manifold with boundary
if and only if n = 1.
 poincaré duality 187

Signature
Definition 8.9.12. Let M be a closed oriented manifold. If dim M = 4k,
the signature σ ( M ) of M is defined to be the signature of the symmetric
non-singular cup product pairing

H 2k ( M; R) × H 2k ( M; R) −→ R
(α, β) 7→ (α ∪ β)[ M]

Otherwise, if dim M is not divisible by 4, we let σ ( M ) = 0.

Remark 8.9.13. Recall that a symmetric non-singular bilinear pairing
has only real (non-zero) eigenvalues, and its signature is defined by
subtracting the number of negative eigenvalues from the number of
positive eigenvalues.
Example 8.9.14.
!
2 2 0 1
σ(S × S ) = σ = 0,
1 0

σ(CP2n ) = 1,
σ(CP2 #CP2 ) = 2.
The signature σ is a cobordism invariant, i.e., if ∂W = M ⊔ − N, then
σ( M) = σ( N ). Here − N denotes the manifold N but with the opposite
orientation.

Here we prove the following version of this fact:

Theorem 8.9.15. If, in the above notations, M4k = ∂V 4k+1 is connected with
V compact and orientable, then σ ( M ) = 0.

Proof. Let A = H 2k ( M; R). The cup product yields a non-singular and

symmetric pairing
φ : A × A → R.
Let A+ be the subspace on which the pairing is positive-definite, and
A− the subspace on which the pairing is negative-definite. Let r =
dim A+ , 2l = dim A (which is even by Proposition 8.9.10). Then,
dim A− = 2l − r since the pairing is non-singular, and

σ ( M) = r − (2l − r ) = 2r − 2l.
188 algebraic topology

In order to prove that σ( M) = 0, it suffices to show that r = l.

Let B ⊂ A be the self-annihilating l-dimensional subspace given by
Proposition 8.9.8. Then A+ ∩ B = {0} and A− ∩ B = {0}. Hence,

dim A+ + dim B ≤ dim A = 2l, i.e., r + l ≤ 2l i.e., r ≤ l

dim A− + dim B ≤ dim A = 2l, i.e., 2l − r + l ≤ 2l i.e., r ≥ l

In conclusion, r = l and σ( M) = 0.

Connected Sums
Definition 8.9.16. Let Mn , N n be closed, connected, oriented n-manifolds.
Their connected sum is defined to be

M#N := ( M \ D1n ) ∪ f ( N \ D2n )

where f : ∂D1n = Sn−1 → ∂D2n = Sn−1 is an orientation-reversing homeo-

morphism.

Remark 8.9.17. The connected sum M#N of closed, connected, oriented

n-manifolds is itself a closed, connected, oriented n-manifold. The co-
homology ring H ∗ ( M#N ) is isomorphic to the ring resulting from the
direct product of H ∗ ( M) and H ∗ ( N ), with the unity elements identified,
and the orientation classes identified. In particular, H 0 ( M#N ) = Z,
H n ( M#N ) = Z and H k ( M#N ) ∼ = H k ( M) ⊕ H k ( N ), 0 < k < n. More-
over, cup products of positive dimensional classes, one from each of
the two original manifolds, are zero, i.e., α ∪ β = 0 for any α ∈ H k ( M )
and β ∈ H l ( N ) with k, l > 0.

Example 8.9.18. By the above description of cup products of a connected

sum, we get:
σ (CP2 # − CP2 ) = 0.

In fact, it can be shown that CP2 # − CP2 is the boundary of a connected,

oriented 5-manifold.
 poincaré duality 189

Example 8.9.19. The spaces S2 × S2 and CP2 #CP2 have the same coho-
mology groups,

H 0 = Z, H 2 = Z ⊕ Z = Zα ⊕ Zβ, H 4 = Z,

but different cohomology rings, since α ∪ β ̸= 0 in H ∗ (S2 × S2 ), but

α ∪ β = 0 in H ∗ (CP2 #CP2 ).
Example 8.9.20. We have

σ(CP2 #CP2 ) = 2 ̸= 0,

so in view of Theorem 8.9.15, CP2 #CP2 cannot be the boundary of a

compact, oriented 5-manifold. However, CP2 #CP2 = ∂W 5 , where W 5
is a compact non-orientable 5-manifold. The compact manifold W can
be constructed as follows:

(a) Start with (CP2 × I )#(RP2 × S3 ).

(b) Run an orientation reversing path γ from one CP2 to the other, by
traveling along an orientation reversing path in RP2 .

(c) Enlarge the path to a tube and remove its interior. What is left is a
5-dimensional non-orientable manifold with ∂W = CP2 #CP2 .
 basics of homotopy theory 191

9
Basics of Homotopy Theory

9.1 Homotopy Groups

Definition 9.1.1. For each n ≥ 0 and X a topological space with x0 ∈ X, the

n-th homotopy group of X is defined as

f : ( I n , ∂I n ) → ( X, x0 ) / ∼

πn ( X, x0 ) =

where I = [0, 1] and ∼ is the usual homotopy of continuous maps.

Remark 9.1.2. Note that we have the following diagram of sets:

f
( I n , ∂I n ) / ( X, x0 )
7
g

(
( I n /∂I n , ∂I n /∂I n )
with ( I n /∂I n , ∂I n /∂I n ) ≃ (Sn , s0 ). So we can also define

πn ( X, x0 ) = g : (Sn , s0 ) → ( X, x0 ) / ∼ .


Remark 9.1.3. If n = 0, then π0 ( X ) is the set of connected components

of X. Indeed, we have I 0 = pt and ∂I 0 = ∅, so π0 ( X ) consists of
homotopy classes of maps from a point into the space X.

Now we will prove several results analogous to the case n = 1, which

corresponds to the fundamental group.

Proposition 9.1.4. If n ≥ 1, then πn ( X, x0 ) is a group with respect to the

operation + defined as:

 f (2s , s , . . . , s )
1 2 n 0 ≤ s1 ≤ 12
( f + g)(s1 , s2 , . . . , sn ) =
 g(2s1 − 1, s2 , . . . , sn ) 1 ≤ s1 ≤ 1.
2

(Note that if n = 1, this is the usual concatenation of paths/loops.)

Proof. First note that since only the first coordinate is involved in this
operation, the same argument used to prove that π1 is a group is valid
192 algebraic topology

I n −1 Figure 9.1: f + g

f g

0 1/2 1 s1

here as well. Then the identity element is the constant map taking all
of I n to x0 and the inverse element is given by

− f ( s1 , s2 , . . . , s n ) = f (1 − s1 , s2 , . . . , s n ).

Proposition 9.1.5. If n ≥ 2, then πn ( X, x0 ) is abelian.

Intuitively, since the + operation only involves the first coordinate, if

n ≥ 2, there is enough space to “slide f past g”.

Figure 9.2: f + g ≃ g + f
f
f g ≃ f g ≃ ≃ g f
g

≃ g f

Proof. Let n ≥ 2 and let f , g ∈ πn ( X, x0 ). We wish to show that

f + g ≃ g + f . We first shrink the domains of f and g to smaller cubes
inside I n and map the remaining region to the base point x0 . Note that
this is possible since both f and g map to x0 on the boundaries, so
the resulting map is continuous. Then there is enough room to slide
f past g inside I n . We then enlarge the domains of f and g back to
their original size and get g + f . So we have “constructed” a homotopy
between f + g and g + f , and hence πn ( X, x0 ) is abelian.

Remark 9.1.6. If we view πn ( X, x0 ) as homotopy classes of maps

(Sn , s0 ) → ( X, x0 ), then we have the following visual representation
of f + g (one can see this by collapsing boundaries in the above cube
interpretation).
 basics of homotopy theory 193

Figure 9.3: f + g, revisited

f
c
X
g

Next recall that if X is path-connected and x0 , x1 ∈ X, then there is

an isomorphism
β γ : π1 ( X, x1 ) → π1 ( X, x0 )
where γ is a path from x1 to x0 , i.e., γ : [0, 1] → X with γ(0) = x1 and
γ(1) = x0 . The isomorphism β γ is given by

β γ ([ f ]) = [γ̄ ∗ f ∗ γ]

for any [ f ] ∈ π1 ( X, x1 ), where γ̄ = γ−1 and ∗ denotes path concatana-

tion. We next show that a similar fact holds for all n ≥ 1.
Proposition 9.1.7. If n ≥ 1 and X is path-connected, then there is an
isomorphism β γ : πn ( X, x1 ) → πn ( X, x0 ) given by

β γ ([ f ]) = [γ · f ],

where γ is a path in X from x1 to x0 , and γ · f is constructed by first shrinking

the domain of f to a smaller cube inside I n , and then inserting the path γ
radially from x1 to x0 on the boundaries of these cubes.

x0 Figure 9.4: β γ

x1

x0 x1 f x1 x0

x1

x0

Proof. It is easy to check that the following properties hold:

1. γ · ( f + g) ≃ γ · f + γ · g

2. (γ · η ) · f ≃ γ · (η · f ), for η a path from x0 to x1

3. c x0 · f ≃ f , where c x0 denotes the constant path based at x0 .

4. β γ is well-defined with respect to homotopies of γ or f .

Note that (1) implies that β γ is a group homomorphism, while (2)

and (3) show that β γ is invertible. Indeed, if γ(t) = γ(1 − t), then
β− 1
γ = βγ .
194 algebraic topology

So, as in the case n = 1, if the space X is path-connected, then πn is

independent of the choice of base point. Further, if x0 = x1 , then (2)
and (3) also imply that π1 ( X, x0 ) acts on πn ( X, x0 ) as:

π1 × π n → π n

(γ, [ f ]) 7→ [γ · f ]

Definition 9.1.8. We say X is an abelian space if π1 acts trivially on πn for

all n ≥ 1.

In particular, this implies that π1 is abelian, since the action of π1 on

π1 is by inner automorphisms, which must all be trivial.

We next show that πn is a functor.

Proposition 9.1.9. A continuous map ϕ : X → Y induces group homomor-

phisms ϕ∗ : πn ( X, x0 ) → πn (Y, ϕ( x0 )) given by [ f ] 7→ [ϕ ◦ f ], for all
n ≥ 1.

Proof. First note that, if f ≃ g, then ϕ ◦ f ≃ ϕ ◦ g. Indeed, if ψt is

a homotopy between f and g, then ϕ ◦ ψt is a homotopy between
ϕ ◦ f and ϕ ◦ g. So ϕ∗ is well-defined. Moreover, from the definition
of the group operation on πn , it is clear that we have ϕ ◦ ( f + g) =
(ϕ ◦ f ) + (ϕ ◦ g). So ϕ∗ ([ f + g]) = ϕ∗ ([ f ]) + ϕ∗ ([ g]). Hence ϕ∗ is a
group homomorphism.

The following is a consequence of the definition of the above induced

homomorphisms:

Proposition 9.1.10. The homomorphisms induced by ϕ : X → Y on higher

homotopy groups satisfy the following two properties:

1. (ϕ ◦ ψ)∗ = ϕ∗ ◦ ψ∗ .

2. (id X )∗ = idπn (X,x0 ) .

We thus have the following important consequence:

Corollary 9.1.11. If ϕ : ( X, x0 ) → (Y, y0 ) is a homotopy equivalence, then

ϕ∗ : πn ( X, x0 ) → πn (Y, ϕ( x0 )) is an isomorphism, for all n ≥ 1.

Example 9.1.12. Consider Rn (or any contractible space). We have

πi (Rn ) = 0 for all i ≥ 1, since Rn is homotopy equivalent to a point.

The following result is very useful for computations:

Proposition 9.1.13. If p : X e → X is a covering map, then p∗ : πn ( X,

e xe) →
πn ( X, p( xe)) is an isomorphism for all n ≥ 2.
 basics of homotopy theory 195

Proof. First we show that p∗ is surjective. Let x = p( xe) and consider

f : (Sn , s0 ) → ( X, x ). Since n ≥ 2, we have that π1 (Sn ) = 0, so
f ∗ (π1 (Sn , s0 )) = 0 ⊂ p∗ (π1 ( X,
e xe)). So f admits a lift to X,
e i.e., there
exists fe : (Sn , s0 ) → ( X,
e xe) such that p ◦ fe = f . Then [ f ] = [ p ◦ fe] =
p∗ ([ f ]). So p∗ is surjective.
e
( X,
e xe)
:
fe
p
f 
( S n , s0 ) / ( X, x )
Next, we show that p∗ is injective. Suppose [ fe] ∈ ker p∗ . So p∗ ([ fe]) =
[ p ◦ fe] = 0. Let p ◦ fe = f . Then f ≃ c x via some homotopy ϕt :
(Sn , s0 ) → ( X, x ) with ϕ1 = f and ϕ0 = c x the constant map. Again,
by the lifting criterion, there is a unique ϕ et : (Sn , s0 ) → ( X,
e xe) with
p◦ϕ et = ϕt .

( X,
e xe)
:
ϕet
p

/ ( X, x )
ϕt
( S n , s0 )

Then we have p ◦ ϕ e1 = ϕ1 = f and p ◦ ϕ e0 = ϕ0 = c x , so by the

uniqueness of lifts, we must have ϕ e1 = f and ϕ
e e0 = c xe. Then ϕ
et is a
homotopy between f and c xe. So [ f ] = 0. Thus p∗ is injective.
e e

Example 9.1.14. Consider S1 with its universal covering map p : R → S1

given by p(t) = e2πit . We already know that π1 (S1 ) = Z. If n ≥ 2,
Proposition 9.1.13 yields that πn (S1 ) = πn (R) = 0.

Example 9.1.15. Consider T n = S1 × S1 × · · · × S1 , the n-torus. We

have π1 ( T n ) = Zn . By using the universal covering map p : Rn → T n ,
we have by Proposition 9.1.13 that πi ( T n ) = πi (Rn ) = 0 for i ≥ 2.

Definition 9.1.16. If πn ( X ) = 0 for all n ≥ 2, the space X is called

aspherical.

Remark 9.1.17. As a side remark, the celebrated Singer-Hopf conjecture

asserts that if X is a smooth closed aspherical manifold of dimension
2k, then (−1)k · χ( X ) ≥ 0, where χ denotes the Euler characteristic.

Proposition 9.1.18. Let { Xα }α be a collection of path-connected spaces. Then

!
πn ∏ Xα = ∏ π n ( Xα )
∼
α α

for all n.
196 algebraic topology

Proof. First note that a map f : Y → ∏α Xα is a collection of maps

f α : Y → Xα . For elements of πn , take Y = Sn (note that since all spaces
are path-connected, we may drop the reference to base points). For
homotopies, take Y = Sn × I.

Example 9.1.19. A natural question to ask is whether there exist spaces

X and Y such that πn ( X ) ∼ = πn (Y ) for all n, but with X and Y not
homotopy equivalent. Whitehead’s Theorem (to be discussed later on)
states that if a map f : X → Y of CW complexes induces isomorphisms
on all πn , then f is a homotopy equivalence. So for the above question
to have a positive answer, we must find X and Y so that there is
no continuous map f : X → Y inducing the isomorphisms on πn ’s.
Consider

X = S2 × RP3 and Y = RP2 × S3 .

Then πn ( X ) = πn (S2 × RP3 ) = πn (S2 ) × πn (RP3 ). Since S3 is a cov-
ering of RP3 , for all n ≥ 2 we have that πn ( X ) = πn (S2 ) × πn (S3 ).
We also have π1 ( X ) = π1 (S2 ) × π1 (RP3 ) = Z/2. Similarly, we have
πn (Y ) = πn (RP2 × S3 ) = πn (RP2 ) × πn (S3 ). And since S2 is a cover-
ing of RP2 , for n ≥ 2 we have that πn (Y ) = πn (S2 ) × πn (S3 ). Finally,
π1 (Y ) = π1 (RP2 ) × π1 (S3 ) = Z/2. So

πn ( X ) = πn (Y ) for all n.
By considering homology groups, however, we see that X and Y are
not homotopy equivalent. Indeed, by the Künneth formula, we get that
H5 ( X ) = Z while H5 (Y ) = 0 (since RP3 is oriented while RP2 is not).

Just like there is a homomorphism π1 ( X ) −→ H1 ( X ), we can also

construct Hurewicz homomorphisms

h X : πn ( X ) −→ Hn ( X )

defined by
[ f : S n → X ] 7 → f ∗ [ S n ],
where [Sn ] is the fundamental class of Sn . A very important result in
homotopy theory is the following:

Theorem 9.1.20 (Hurewicz). If n ≥ 2 and πi ( X ) = 0 for all i < n, then

Hi ( X ) = 0 for i < n and πn ( X ) ∼
= Hn ( X ).
Moreover, there is also a relative version of the Hurewicz theorem
(see the next section for a definition of the relative homotopy groups),
which can be used to prove the following:
Corollary 9.1.21. If X and Y are CW complexes with π1 ( X ) = π1 (Y ) = 0,
and if a map f : X → Y induces isomorphisms on all integral homology groups
Hn , then f is a homotopy equivalence.
 basics of homotopy theory 197

We’ll discuss all of these in the subsequent sections.

9.2 Relative Homotopy Groups

Given a triple ( X, A, x0 ) where x0 ∈ A ⊆ X, we define relative homo-

topy groups as follows:

Definition 9.2.1. Let X be a space and let A ⊆ X and x0 ∈ A. Let

I n−1 = {(s1 , . . . , sn ) ∈ I n | sn = 0}

and set
J n−1 = ∂I n \ I n−1 .

Then define the n-th homotopy group of the pair ( X, A) with basepoint x0 as:

f : ( I n , ∂I n , J n−1 ) → ( X, A, x0 ) / ∼

πn ( X, A, x0 ) =

where, as before, ∼ is the homotopy equivalence relation.

sn J n −1

I n −1

Alternatively, by collapsing J n−1 to a point, we obtain a commutative

diagram

f
( I n , ∂I n , J n−1 ) / ( X, A, x0 )
7
g

(
( D n , S n −1 , s 0 )

where g is obtained by collapsing J n−1 . So we can take

πn ( X, A, x0 ) = g : ( D n , Sn−1 , s0 ) → ( X, A, x0 ) / ∼ .


A sum operation is defined on πn ( X, A, x0 ) by the same formulas

as for πn ( X, x0 ), except that the coordinate sn now plays a special role
and is no longer available for the sum operation. Thus, we have:
198 algebraic topology

x0
x0 S n −1
A
x0 x0
Dn

Proposition 9.2.2. If n ≥ 2, then πn ( X, A, x0 ) forms a group under the

usual sum operation. Further, if n ≥ 3, then πn ( X, A, x0 ) is abelian.

Remark 9.2.3. Note that the proposition fails in the case n = 1. Indeed,
we have that

π1 ( X, A, x0 ) = f : ( I, {0, 1}, {1}) → ( X, A, x0 ) / ∼ .

Then π1 ( X, A, x0 ) consists of homotopy classes of paths starting any-

where A and ending at x0 , so we cannot always concatenate two paths.

X
A

x0

Just as in the absolute case, a map of pairs ϕ : ( X, A, x0 ) → (Y, B, y0 )

induces homomorphisms ϕ∗ : πn ( X, A, x0 ) → πn (Y, B, y0 ) for all n ≥ 2.

A very important feature of the relative homotopy groups is the

following:

Proposition 9.2.4. The relative homotopy groups of ( X, A, x0 ) fit into a long

exact sequence

i
∗ j∗ n ∂
· · · → πn ( A, x0 ) → πn ( X, x0 ) → πn ( X, A, x0 ) −→ πn−1 ( A, x0 ) → · · ·
· · · → π0 ( X, x0 ) → 0,

where the map ∂n is defined by ∂n [ f ] = [ f | I n−1 ] and all others are induced by
inclusions.
Remark 9.2.5. Near the end of the above sequence, where group struc-
tures are not defined, exactness still makes sense: the image of one map
is the kernel of the next, which consists of those elements mapping to
the homotopy class of the constant map.

Example 9.2.6. Let X be a path-connected space, and

CX := X × [0, 1]/X × {0}

 basics of homotopy theory 199

be the cone on X. We can regard X as a subspace of CX via X × {1} ⊂

CX. Since CX is contractible, the long exact sequence of homotopy
groups gives isomorphisms

πn (CX, X, x0 ) ∼
= πn−1 ( X, x0 ).

In what follows, it will be important to have a good description of

the zero element 0 ∈ πn ( X, A, x0 ).

Lemma 9.2.7. Let [ f ] ∈ πn ( X, A, x0 ). Then [ f ] = 0 if, and only if, f ≃ g

for some map g with image contained in A.

Proof. (⇐) Suppose f ≃ g for some g with Im g ⊂ A.

x0

g
x0 A x0 X

Then we can deform I n to J n−1 as indicated in the above picture, and

so g ≃ c x0 . Since homotopy is a transitive relation, we then get that
f ≃ c x0 .
(⇒) Suppose [ f ] = 0 in πn ( X, A, x0 ). So f ≃ c x0 . Take g = c x0 .

Recall that if X is path-connected, then πn ( X, x0 ) is independent of

our choice of base point, and π1 ( X ) acts on πn ( X ) for all n ≥ 1. In the
relative case, we have:

Lemma 9.2.8. If A is path-connected, then β γ : πn ( X, A, x1 ) → πn ( X, A, x0 )

is an isomorphism, where γ is a path in A from x1 to x0 .

Figure 9.5: relative β γ

γ A γ

Remark 9.2.9. In particular, if x0 = x1 , we get an action of π1 ( A) on

πn ( X, A).

It is easy to see that the following three conditions are equivalent:

200 algebraic topology

1. every map Si → X is homotopic to a constant map,

2. every map Si → X extends to a map Di+1 → X, with Si = ∂Di+1 ,

3. πi ( X, x0 ) = 0 for all x0 ∈ X.

In the relative setting, the following are equivalent for any i > 0:

1. every map ( Di , ∂Di ) → ( X, A) is homotopic rel. ∂Di to a map

Di → A,

2. every map ( Di , ∂Di ) → ( X, A) is homotopic through such maps to a

map Di → A,

3. every map ( Di , ∂Di ) → ( X, A) is homotopic through such maps to a

constant map Di → A,

4. πi ( X, A, x0 ) = 0 for all x0 ∈ A.

Remark 9.2.10. As seen above, if α : Sn = ∂en+1 → X represents an

element [α] ∈ πn ( X, x0 ), then [α] = 0 if and only if α extends to a map
en+1 → X. Thus if we enlarge X to a space X ′ = X ∪α en+1 by adjoining
an (n + 1)-cell en+1 with α as attaching map, then the inclusion j :
X ,→ X ′ induces a homomorphism j∗ : πn ( X, x0 ) → πn ( X ′ , x0 ) with
j∗ [α] = 0. We say that [α] “has been killed” by adding an (n + 1)-cell.

The following is left as an exercise:

Lemma 9.2.11. Let ( X, x0 ) be a space with a basepoint, and let X ′ = X ∪α

en+1 be obtained from X by adjoining an (n + 1)-cell. Then the inclusion
j : X ,→ X ′ induces a homomorphism j∗ : πi ( X, x0 ) → πi ( X ′ , x0 ), which is
an isomorphism for i < n and surjective for i = n.

Definition 9.2.12. We say that the pair ( X, A) is n-connected if πi ( X, A) =

0 for i ≤ n. Say that X is n-connected if πi ( X ) = 0 for i ≤ n.

In particular, X is 0-connected if and only if X is connected. More-

over, X is 1-connected if and only if X is simply-connected.

9.3 Homotopy Extension Property

Definition 9.3.1 (Homotopy Extension Property). Given a pair ( X, A),

a map F0 : X → Y, and a homotopy f t : A → Y such that f 0 = F0 | A , we
say that ( X, A) satisfies the homotopy extension property (HEP) if there is
a homotopy Ft : X → Y extending f t and F0 . In other words, ( X, A) has
homotopy extension property if any map X × {0} ∪ A × I → Y extends to a
map X × I → Y.
 basics of homotopy theory 201

Proposition 9.3.2. Any CW pair has the homotopy extension property. In

fact, for every CW pair ( X, A), there is a deformation retract r : X × I →
X × {0} ∪ A × I, hence X × I → Y can be defined by the composition
r
X × I → X × {0} ∪ A × I → Y.
r
Proof. We have an obvious deformation retract D n × I −→ D n × {0} ∪
Sn−1 × I. For every n, consider the pair ( Xn , An ∪ Xn−1 ), where Xn
denotes the n-skeleton of X. Then

Xn × I = Xn × {0} ∪ ( An ∪ Xn−1 ) × I ∪ D n × I,
 

where the cylinders D n × I corresponding to n-cells D n in X \ A are

glued along D n × {0} ∪ Sn−1 × I to the pieces Xn × {0} ∪ ( An ∪ Xn−1 ) ×
I. By deforming these cylinders D n × I we get a deformation retraction

rn : Xn × I → Xn × {0} ∪ ( An ∪ Xn−1 ) × I.

Concatenating these deformation retractions by performing rn over 1 −
1 1
n − 1
, 1 − n , we get a deformation retraction of X × I onto X × {0} ∪
2 2
A × I. Continuity follows since CW complexes have the weak topology
with respect to their skeleta, so a map of CW complexes is continuous
if and only if its restriction to each skeleton is continuous.

9.4 Cellular Approximation

All maps are assumed to be continuous.

Definition 9.4.1. Let X and Y be CW-complexes. A map f : X → Y is called
cellular if f ( Xn ) ⊂ Yn for all n, where Xn denotes the n-skeleton of X and
similarly for Y.
Definition 9.4.2. Let f : X → Y be a map of CW complexes. A map
f ′ : X → Y is a cellular approximation of f if f ′ is cellular and f is homotopic
to f ′ .
Theorem 9.4.3 (Cellular Approximation Theorem). Any map f : X → Y
between CW-complexes has a cellular approximation f ′ : X → Y. Moreover, if
f is already cellular on a subcomplex A ⊆ X, we can take f ′ | A = f | A .
The proof of Theorem 9.4.3 uses the following key technical result.

Lemma 9.4.4. Let f : X ∪ en → Y ∪ ek be a map of CW complexes, with en ,

ek denoting an n-cell and, resp., k-cell attached to X and, resp., Y. Assume
h.e.
that f ( X ) ⊆ Y, f | X is cellular, and n < k. Then f ≃ f ′ (rel. X ) , with
Im( f ′ ) ⊆ Y.
Remark 9.4.5. If in the statement of Lemma 9.4.4 we assume that X
and Y are points, then we get that the inclusion Sn ,→ Sk (n < k) is
homotopic to the constant map Sn → {s0 } for some point s0 ∈ Sk .
202 algebraic topology

Lemma 9.4.4 is used along with induction on skeleta to prove the

cellular approximation theorem as follows.

Proof of Theorem 9.4.3. Suppose f | Xn−1 is cellular, and let en be an (open)

n-cell of X. Since en is compact, f (en ) (hence also f (en )) meets only
finitely many open cells of Y. Let ek be an open cell of maximal
dimension in Y which meets f (en ). If k ≤ n, f is already cellular on
en . If n < k, Lemma 9.4.4 can be used to homotop f | Xn−1 ∪en (rel. Xn−1 )
to a map whose image on en misses ek . By finitely many iterations
of this process, we eventually homotop f | Xn−1 ∪en (rel. Xn−1 ) to a map
f ′ : Xn−1 ∪ en → Yn , i.e., whose image on en misses all cells in Y of
dimension > n. Doing this for all n-cells of X, staying fixed on n-cells
in A where f is already cellular, we obtain a homotopy of f | Xn (rel.
Xn−1 ∪ An ) to a cellular map. By the homotopy extension property 9.3.2,
we can extend this homotopy (together with the constant homotopy on
A) to a homotopy defined on all of X. This completes the induction
step.
For varying n → ∞, we concatenate the above homotopies to define a
homotopy from f to a cellular map f ′ (rel. A) by performing the above
construction (i.e., the n-th homotopy) on the t-interval [1 − 1/2n , 1 −
1/2n+1 ].

We also have the following relative version of Theorem 9.4.3:

Theorem 9.4.6 (Relative cellular approximation). Any map f : ( X, A) →
(Y, B) of CW pairs has a cellular approximation by a homotopy through such
maps of pairs.

Proof. First we use the cellular approximation for f | A : A → B. Let

f ′ : A → B be a cellular map, homotopic to f | A via a homotopy H. By
the Homotopy Extension Property 9.3.2, we can regard H as a homotopy
on all of X, so we get a map f ′ : X → Y such that f ′ | A is a cellular
map. By the second part of the cellular approximation theorem 9.4.3,
h.e.
f ′ ≃ f ′′ , with f ′′ : X → Y a cellular map satisfying f ′ | A = f ′′ | A . The
map f ′′ provides the required cellular approximation of f .

Corollary 9.4.7. Let A ⊂ X be CW complexes and suppose that all cells of

X \ A have dimension > n. Then πi ( X, A) = 0 for i ≤ n.

Proof. Let [ f ] ∈ πi ( X, A). By the relative version of the cellular approxi-

mation, the map of pairs f : ( Di , Si−1 ) → ( X, A) is homotopic to a map
g with g( Di ) ⊂ Xi . But for i ≤ n, we have that Xi ⊂ A, so Im g ⊂ A.
Therefore, by Lemma 9.2.7, [ f ] = [ g] = 0.

Corollary 9.4.8. If X is a CW complex, then πi ( X, Xn ) = 0 for all i ≤ n.

Therefore, the long exact sequence for the homotopy groups of the
pair ( X, Xn ) yields the following:
 basics of homotopy theory 203

Corollary 9.4.9. Let X be a CW complex. For i < n, we have πi ( X ) ∼

=
π i ( Xn ).

9.5 Excision for homotopy groups. The Suspension Theorem

We state here the following useful result without proof:

Theorem 9.5.1 (Excision). Let X be a CW complex which is a union of
subcomplexes A and B, such that C = A ∩ B is path connected. Assume
that ( A, C ) is m-connected and ( B, C ) is n-connected, with m, n ≥ 1. Then
the map πi ( A, C ) −→ πi ( X, B) induced by inclusion is an isomorphism if
i < m + n and a surjection for i = m + n.
The following consequence is very useful for iterating homotopy
groups of spheres:
Theorem 9.5.2 (Freudenthal Suspension Theorem). Let X be an (n − 1)-
connected CW complex. For any map f : Si → X, consider its suspension,

Σ f : ΣSi = Si+1 → ΣX.

The assignment
[ f ] ∈ πi ( X ) 7→ [Σ f ] ∈ πi+1 (ΣX )
defines a homomorphism πi ( X ) → πi+1 (ΣX ), which is an isomorphism for
i < 2n − 1 and a surjection for i = 2n − 1.

Proof. Decompose the suspension ΣX as the union of two cones C+ X

and C− X intersecting in a copy of X. By using long exact sequences
of pairs and the fact that the cones C+ X and C− X are contractible, the
suspension map can be written as a composition:

πi ( X ) ∼
= πi+1 (C+ , X ) −→ πi+1 (ΣX, C− X ) ∼
= πi+1 (ΣX ),

with the middle map induced by inclusion.

Since X is (n − 1)-connected, from the long exact sequence of the pair
(C± X, X ), we see that the pairs (C± X, X ) are n-connected. Therefore,
the Excision Theorem 9.5.1 yields that πi+1 (C+ , X ) −→ πi+1 (ΣX, C− X )
is an isomorphism for i + 1 < 2n and it is surjective for i + 1 = 2n.

9.6 Homotopy Groups of Spheres

We now turn our attention to computing (some of) the homotopy

groups πi (Sn ). For i ≤ n, i = n + 1, n + 2, n + 3 and a few more cases,
these homotopy groups are known (and we will work them out later
on). In general, however, this is a very difficult problem. For i = n, we
would expect to have πn (Sn ) = Z by associating to each (homotopy
class of a) map f : Sn → Sn its degree. For i < n, we will show that
204 algebraic topology

πi (Sn ) = 0. Note that if f : Si → Sn is not surjective, i.e., there is

y ∈ Sn \ f (Si ), then f factors through Rn , which is contractible. By
composing f with the retraction Rn → x0 we get that f ≃ c x0 . However,
there are surjective maps Si → Sn for i < n, in which case the above
“proof” fails. To make things work, we “alter” f to make it cellular, so
the following holds.

Proposition 9.6.1. For i < n, we have πi (Sn ) = 0.

Proof. Choose the standard CW-structure on Si and Sn . For [ f ] ∈

πi (Sn ), we may assume by Theorem 9.4.3 that f : Si → Sn is cellular.
Then f (Si ) ⊂ (Sn )i . But (Sn )i is a point, so f is a constant map.

Recall now the following special case of the Suspension Theorem

9.5.2 for X = Sn :

Theorem 9.6.2. Let f : Si → Sn be a map, and consider its suspension,

Σ f : ΣSi = Si+1 → ΣSn = Sn+1 .

The assignment

[ f ] ∈ π i ( S n ) 7 → [ Σ f ] ∈ π i +1 ( S n +1 )

defines a homomorphism πi (Sn ) → πi+1 (Sn+1 ), which is an isomorphism

πi ( S n ) ∼
= πi+1 (Sn+1 ) for i < 2n − 1 and a surjection for i = 2n − 1.
Corollary 9.6.3. πn (Sn ) is either Z or a finite quotient of Z (for n ≥ 2),
generated by the degree map.

Proof. By the Suspension Theorem 9.6.2, we have the following:

Z∼
= π1 ( S1 ) ↠ π2 ( S2 ) ∼
= π3 ( S3 ) ∼
= π4 ( S4 ) ∼
= ···

To show that π1 (S1 ) ∼ = π2 (S2 ), we can use the long exact sequence for
the homotopy groups of a fibration, see Theorem 9.11.8 below. For any
fibration (e.g., a covering map)

F ,→ E −→ B

there is a long exact sequence

· · · −→ πi ( F ) −→ πi ( E) −→ πi ( B) −→ πi−1 ( F ) −→ · · · (9.6.1)

Applying the above long exact sequence to the Hopf fibration S1 ,→

S3 → S2 , we obtain:

· · · −→ π2 (S1 ) −→ π2 (S3 ) −→ π2 (S2 ) −→ π1 (S1 ) −→ π1 (S3 ) −→ · · ·

 basics of homotopy theory 205

Using the fact that π2 (S3 ) = 0 and π1 (S3 ) = 0, we therefore have an

isomorphism:
π2 ( S2 ) ∼
= π1 ( S1 ) ∼
= Z.
Note that by using the vanishing of the higher homotopy groups of S1 ,
the long exact sequence (9.11.8) also yields that

π3 ( S2 ) ∼
= π2 ( S2 ) ∼
= Z.

Remark 9.6.4. Unlike the homology and cohomology groups, the ho-
motopy groups of a finite CW-complex can be infinitely generated. This
fact is discussed in the next example.

Example 9.6.5. For n ≥ 2, consider the finite CW complex S1 ∨ Sn . We

then have that
πn (S1 ∨ Sn ) = πn (S^
1 ∨ S n ),

where S^1 ∨ Sn is the universal cover of S1 ∨ Sn , as depicted in the

attached figure. By contracting the segments between consecutive

Figure 9.6: universal cover of S1 ∨ Sn

2

−1

integers, we have that

Skn ,
_
1 ∨ Sn ≃
S^
k ∈Z

with Skn
denoting the n-sphere corresponding to the integer k. So for
any n ≥ 2, we have:

π n ( S1 ∨ S n ) = π n ( Skn ),
_

k ∈Z

which is the free abelian group generated by the inclusions Skn ,→

W n
k ∈Z Sk . Indeed, we have the following:

Sαn ) is free abelian and generated by the inclusions of

W
Lemma 9.6.6. πn ( α
factors.
206 algebraic topology

Proof. Suppose first that there are only finitely many Sαn ’s in the wedge
W n W n n
α Sα . Then we can regard α Sα as the n-skeleton of ∏α Sα . The cell
n 0
structure of a particular Sα consists of a single 0-cell eα and a single
n-cell, eαn . Thus, in the product ∏α Sαn there is one 0-cell e0 = ∏α e0α ,
which, together with the n-cells

( ∏ e0β ) × eαn ,
[

α β̸=α

form the n-skeleton α Sαn . Hence ∏α Sαn \ α Sαn has only cells of di-
W W

mension at least 2n, which by Corollary 9.4.8 yields that the pair
(∏α Sαn , α Sαn ) is (2n − 1)-connected. In particular, as n ≥ 2, we get:
W

 
πn ( Sαn ) ∼ = πn ∏ Sαn ∼ = ∏ πn (Sαn ) = πn (Sαn ) = Z.
_ M M

α α α α α

To reduce the case of infinitely many summands Sαn

to the finite case,
consider the homomorphism Φ : α πn (Sα ) −→ πn ( α Sαn ) induced
n
L W

by the inclusions Sαn ,→ α Sαn . Then Φ is onto since any map f : Sn →

W
W n
α Sα has compact image contained in the wedge sum of finitely many
Sαn ’s, so by the above finite case, [ f ] is in the image of Φ. Moreover, a
nullhomotopy of f has compact image contained in the wedge sum of
finitely many Sαn ’s, so by the above finite case we have that Φ is also
injective.

To conclude our example, we showed that πn (S1 ∨ Sn ) ∼ = πn ( k∈Z Skn ),

W

and πn ( k∈Z Skn ) is free abelian generated by the inclusion of each of

W

the infinite number of n-spheres. Therefore, πn (S1 ∨ Sn ) is infinitely

generated.
Remark 9.6.7. Under the action of π1 on πn , we can regard πn as a
Z[π1 ]-module. Here Z[π1 ] is the group ring of π1 with Z-coefficients,
whose elements are of the form ∑α nα γα , with nα ∈ Z and only finitely
many non-zero, and γα ∈ π1 . Since all the n-spheres Skn in the universal
cover k∈Z Skn are identified under the π1 -action, πn is a free Z[π1 ]-
W

module of rank 1, i.e.,

πn ∼
= Z[ π1 ] ∼
= Z[Z] ∼
= Z[t, t−1 ],
1 7→ t
− 1 7 → t −1
n 7→ tn ,

which is infinitely generated (by the powers of t) over Z (i.e., as an

abelian group).

Remark 9.6.8. If we consider the class of spaces for which π1 acts

trivially on all of πn ’s, a result of Serre asserts that the homotopy
groups of such spaces are finitely generated if and only if homology
groups are finitely generated.
 basics of homotopy theory 207

9.7 Whitehead’s Theorem

Definition 9.7.1. A map f : X → Y is a weak homotopy equivalence if it

induces isomorphisms on all homotopy groups πn .

Notice that a homotopy equivalence is a weak homotopy equivalence.

The following important result provides a converse to this fact in the
world of CW complexes.

Theorem 9.7.2 (Whitehead). If X and Y are CW complexes and f : X → Y

is a weak homotopy equivalence, then f is a homotopy equivalence. Moreover,
if X is a subcomplex of Y, and f is the inclusion map, then X is a deformation
retract of Y.

The following consequence is very useful in practice:

Corollary 9.7.3. If X and Y are CW complexes with π1 ( X ) = π1 (Y ) = 0,

and f : X → Y induces isomorphisms on homology groups Hn for all n, then
f is a homotopy equivalence.

The above corollary follows from Whitehead’s theorem and the fol-
lowing relative version of the Hurewicz Theorem 9.10.1 (to be discussed
later on):

Theorem 9.7.4 (Hurewicz). If n ≥ 2, and πi ( X, A) = 0 for i < n, with

A simply-connected and non-empty, then Hi ( X, A) = 0 for i < n and
πn ( X, A) ∼
= Hn ( X, A).
Before discussing the proof of Whitehead’s theorem, let us give
an example that shows that having induced isomorphisms on all ho-
mology groups is not sufficient for having a homotopy equivalence
(so the simply-connectedness assumption in Corollary 9.7.3 cannot be
dropped):

Example 9.7.5. Let

f : X = S1 ,→ (S1 ∨ Sn ) ∪ en+1 = Y ( n ≥ 2)

be the inclusion map, with the attaching map for the (n + 1)-cell of
Y described below. We know from Example 9.6.5 that πn (S1 ∨ Sn ) ∼ =
Z[t, t−1 ]. We define Y by attaching the (n + 1)-cell en+1 to S1 ∨ Sn by a
map g : Sn = ∂en+1 → S1 ∨ Sn so that [ g] ∈ πn (S1 ∨ Sn ) corresponds to
the element 2t − 1 ∈ Z[t, t−1 ]. We then see that

πn (Y ) = Z[t, t−1 ]/(2t − 1) ̸= 0 = πn ( X ),

since by setting t = 12 we get that Z[t, t−1 ]/(2t − 1) ∼

= Z[ 12 ] = { 2ak | k ∈
Z≥0 } ⊂ Q. In particular, f is not a homotopy equivalence. Moreover,
from the long exact sequence of homotopy groups for the (n − 1)-
connected pair (Y, X ), the inclusion X ,→ Y induces an isomorphism
208 algebraic topology

on homotopy groups πi for i < n. Finally, this inclusion map also

induces isomorphisms on all homology groups, Hn ( X ) ∼ = Hn (Y ) for all
n, as can be seen from cellular homology. Indeed, the cellular boundary
map
Hn+1 (Yn+1 , Yn ) → Hn (Yn , Yn−1 )

is an isomorphism since the degree of the composition of the attaching

map Sn → S1 ∨ Sn of en+1 with the collapse map S1 ∨ Sn → Sn is
2 − 1 = 1.

Let us now get back to the proof of Whitehead’s Theorem 9.7.2. To

prove Whitehead’s theorem, we will use the following:

Lemma 9.7.6 (Compression Lemma). Let ( X, A) be a CW pair, and (Y, B)

be a pair with Y path-connected and B ̸= ∅. Suppose that for each n > 0 for
which X \ A has cells of dimension n, πn (Y, B, b0 ) = 0 for all b0 ∈ B. Then
any map f : ( X, A) → (Y, B) is homotopic to some map f ′ : X → B fixing
A (i.e., with f ′ | A = f | A ).

Proof. Assume inductively that f ( Xk−1 ∪ A) ⊆ B. Let ek be a k-cell in

X \ A, with characteristic map α : ( D k , Sk−1 ) → X. Ignoring basepoints,
we regard α as an element [α] ∈ πk ( X, Xk−1 ∪ A). Then f ∗ [α] = [ f ◦
α] ∈ πk (Y, B) = 0 by our hypothesis, since ek is a k-cell in X \ A. By
Lemma 9.2.7, there is a homotopy H : ( D k , Sk−1 ) × I → (Y, B) such
that H0 = f ◦ α and Im( H1 ) ⊆ B.
Performing this process for all k-cells in X \ A simultaneously, we get
a homotopy from f to f ′ such that f ′ ( Xk ∪ A) ⊆ B. Using the homotopy
extension property 9.3.2, we can regard this as a homotopy on all of X,
i.e., f ≃ f ′ as maps X → Y, so the induction step is completed.
Finitely many applications of the induction step finish the proof if
the cells of X \ A are of bounded dimension. In general, we have

f ≃ f 1 , with f 1 ( X1 ∪ A) ⊆ B,
H1

f 1 ≃ f 2 , with f 2 ( X2 ∪ A) ⊆ B,
H2

···

f n−1 ≃ f n , with f n ( Xn ∪ A) ⊆ B,
Hn

and so on. Any finite skeleton is eventually fixed under these homo-
topies.
Define a homotopy H : X × I → Y as

H = Hi on 1 − 2i1−1 , 1 − 21i .
 

Note that H is continuous by CW topology, so it gives the required

homotopy.
 basics of homotopy theory 209

Proof of Whitehead’s theorem. We can split the proof of Theorem 9.7.2

into two cases:
Case 1: If f is an inclusion X ,→ Y, since πn ( X ) = πn (Y ) for all n, we
get by the long exact sequence for the homotopy groups of the pair
(Y, X ) that πn (Y, X ) = 0 for all n. Applying the above compression
lemma 9.7.6 to the identity map id : (Y, X ) → (Y, X ) yields a deforma-
tion retraction r : Y → X of Y onto X.
Case 2: The general case of a map f : X → Y can be reduced to the
above case of an inclusion by using the mapping cylinder of f , i.e.,

M f := ( X × I ) ⊔ Y/( x, 1) ∼ f ( x ).

Figure 9.7: The mapping cylinder M f

X × {0}
X×I
X × {1}
−→

f (X)

Note that M f contains both X = X × {0} and Y as subspaces, and M f

deformation retracts onto Y. Moreover, the map f can be written as the
composition of the inclusion i of X into M f , and the retraction r from
M f to Y:
i r
f : X = X × {0} ,→ M f → Y.

Since M f is homotopy equivalent to Y via r, it suffices to show that M f

deformation retracts onto X, so we can replace f with the inclusion
map i. If f is a cellular map, then M f is a CW complex having X as
a subcomplex. So we can apply Case 1. If f is not cellular, than f is
homotopic to some cellular map g, so we may work with g and the
mapping cylinder Mg and again reduce to Case 1.

We can now prove Corollary 9.7.3:

Proof. After replacing Y by the mapping cylinder M f , we may assume

that f is an inclusion X ,→ Y. As Hn ( X ) ∼
= Hn (Y ) for all n, we have
210 algebraic topology

by the long exact sequence for the homology groups of the pair (Y, X )
that Hn (Y, X ) = 0 for all n.
Since X and Y are simply-connected, we have π1 (Y, X ) = 0. So by
the relative Hurewicz Theorem 9.10.1, the first non-zero πn (Y, X ) is
isomorphic to the first non-zero Hn (Y, X ). So πn (Y, X ) = 0 for all n.
Then, by the homotopy long exact sequence for the pair (Y, X ), we get
that
πn ( X ) ∼
= π n (Y )
for all n, with isomorphisms induced by the inclusion map f . Finally,
Whitehead’s theorem 9.7.2 yields that f is a homotopy equivalence.

Example 9.7.7. Let X = RP2 and Y = S2 × RP∞ . First note that

π 1 ( X ) = π 1 (Y ) ∼
= Z/2. Also, since S2 is a covering of RP2 , we have
that
πi ( X ) ∼
= πi (S2 ), i ≥ 2.
= πi (S2 ) × πi (RP∞ ), and as RP∞ is covered by S∞ =
Moreover, πi (Y ) ∼
S n
n≥0 S , we get that

= πi (S2 ) × πi (S∞ ), i ≥ 2.
π i (Y ) ∼

To calculate πi (S∞ ), we use cellular approximation. More precisely, we

can approximate any f : Si → S∞ by a cellular map g so that Im g ⊂ Sn
for i ≪ n. Thus, [ f ] = [ g] ∈ πi (Sn ) = 0, and we see that

πi ( X ) ∼
= π i ( S2 ) ∼
= πi (Y ), i ≥ 2.

Altogether, we have shown that X and Y have the same homotopy

groups. However, as can be easily seen by considering homology
groups, X and Y are not homotopy equivalent. In particular, by White-
head’s theorem, there cannot exist a map f : RP2 → S2 × RP∞ inducing
isomorphisms on πn for all n. (If such a map existed, it would have to
be a homotopy equivalence.)

Example 9.7.8. As we will see later on, the CW complexes S2 and S3 ×

CP∞ have isomorphic homotopy groups, but they are not homotopy
equivalent.

9.8 CW approximation

Recall that map f : X → Y is a weak homotopy equivalence if it induces

isomorphisms on all homotopy groups πn . As seen in Theorem 9.10.3,
a weak homotopy equivalence induces isomorphisms on all homol-
ogy and cohomology groups. Furthermore, Whitehead’s Theorem
9.7.2 shows that a weak homotopy equivalence of CW complexes is a
homotopy equivalence.
 basics of homotopy theory 211

In this section we show that given any space X, there exists a (unique
up to homotopy) CW complex Z and a weak homotopy equivalence
f : Z → X. Such a map f : Z → X is called a CW approximation of X.

Definition 9.8.1. Given a pair ( X, A), with ∅ ̸= A a CW complex, an

n-connected CW model of ( X, A) is an n-connected CW pair ( Z, A), together
with a map f : Z → X with f | A = id A , so that f ∗ : πi ( Z ) → πi ( X ) is an
isomorphism for i > n and an injection for i = n (for any choice of basepoint).

Remark 9.8.2. If such models exist, by letting A consist of one point in

each path-component of X and n = 0, we get a CW approximation Z
of X.

Theorem 9.8.3. For any pair ( X, A) with A a nonempty CW complex such n-

connected models ( Z, A) exist. Moreover, Z can be built from A by attaching
cells of dimension greater than n. (Note that by cellular approximation this
implies that πi ( Z, A) = 0 for i ≤ n).

We will prove this theorem after discussing the following conse-

quences:

Corollary 9.8.4. Any pair of spaces ( X, X0 ) has a CW approximation ( Z, Z0 ).

Proof. Let f 0 : Z0 → X0 be a CW approximation of X0 , and consider

the map g : Z0 → X defined by the composition of f 0 and the inclusion
map X0 ,→ X. Let Mg be the mapping cylinder of g. Hence we get
the sequence of maps Z0 ,→ Mg → X, where the map r : Mg → X is a
deformation retract.
Now, let ( Z, Z0 ) be a 0-connected CW model of ( Mg , Z0 ). Consider
the composition:

(r, f 0 )
( f , f 0 ) : ( Z, Z0 ) −→ ( Mg , Z0 ) −→ ( X, X0 )

So the map f : Z → X is obtained by composing the weak homotopy

equivalence Z → Mg and the deformation retract (hence homotopy
equivalence) Mg → X. In other words, f is a weak homotopy equiva-
lence and f | Z0 = f 0 , thus proving the result.

Corollary 9.8.5. For each n-connected CW pair ( X, A) there is a CW pair

( Z, A) that is homotopy equivalent to ( X, A) relative to A, and such that Z
is built from A by attaching cells of dimension > n.

Proof. Let ( Z, A) be an n-connected CW model of ( X, A). By Theorem

9.8.3, Z is built from A by attaching cells of dimension > n. We
h.e.
claim that Z ≃ X (rel. A). First, by definition, the map f : Z → X
has the property that f ∗ is an isomorphism on πi for i > n and an
injection on πn . For i < n, by the n-connectedness of the given model,
212 algebraic topology

πi ( X ) ∼
= πi ( A ) ∼
= πi ( Z ) where the isomorphisms are induced by f
since the following diagram commutes,

f
ZO /X
O

A
id /A

(with A ,→ Z and A ,→ X the inclusion maps.) For i = n, by n-

connectedness of ( X, A) the composition

πn ( A) ↠ πn ( Z ) ↣ πn ( X )

is onto. So the induced map f ∗ : πn ( Z ) → πn ( X ) is surjective. Alto-

gether, f ∗ induces isomorphisms on all πi , so by Whitehead’s Theorem
we conclude that f : Z → X is a homotopy equivalence.
We make f stationary on A as follows. Define the quotient space

W f := M f /{{ a} × I ∼ pt, ∀ a ∈ A}

of the mapping cylinder M f obtained by collapsing each segment

{ a} × I to a point, for any a ∈ A. Assuming f has been made cellular,
W f is a CW complex containing X and Z as subcomplexes, and W f
deformation retracts onto X just as M f does.
Consider the map h : Z → X given by the composition Z ,→ W f →
X, where W f → X is the deformation retract. We claim that Z is
a deformation retract of W f , thus giving us that h is a homotopy
equivalence relative to A. Indeed, πi (W f ) ∼= πi ( X ) (since W f is a
∼
deformation retract of X) and πi ( X ) = πi ( Z ) since X is homotopy
equivalent to Z. Using Whitehead’s theorem, we conclude that Z is a
deformation retract of W f .

Proof of Theorem 9.8.3. We will construct Z as a union of subcomplexes

A = Zn ⊆ Zn+1 ⊆ · · ·

such that for each k ≥ n + 1, Zk is obtained from Zk−1 by attaching

k-cells.
We will show by induction that we can construct Zk together with
a map f k : Zk → X such that f k | A = id A and f k∗ is injective on πi for
n ≤ i < k and onto on πi for n < i ≤ k. We start the induction at k = n,
with Zn = A, in which case the conditions on πi are void.
For the induction step, k → k + 1, consider the set {ϕα }α of genera-
tors ϕα : Sk → Zk of ker ( f k∗ : πk ( Zk ) → πk ( X )). Define

Yk+1 := Zk ∪α ∪ϕα eαk+1 ,

where eαk+1 is a (k + 1)-cell attached to Zk along ϕα .

 basics of homotopy theory 213

Then f k : Zk → X extends to Yk+1 . Indeed, f k ◦ ϕα : Sk → Zk → X

is nullhomotopic, since [ f k ◦ ϕα ] = f k∗ [ϕα ] = 0. So we get a map
g : Yk+1 → X. It is easy to check that g∗ is injective on πi for n ≤ i ≤ k,
and onto on πk . In fact, since we extend f k on (k + 1)-cells, we only need
to check the effect on πk . The elements of ker( g∗ ) on πk are represented
by nullhomotopic maps (by construction) Sk → Zk ⊂ Yk+1 → X. So g∗
is one-to-one on πk . Moreover, g∗ is onto on πk since, by hypothesis,
the composition πk ( Zk ) → πk (Yk+1 ) → πk ( X ) is onto.
Let {ϕβ : Sk+1 → X } be a set of generators of πk+1 ( X, x0 ) and let
Zk+1 = Yk+1 ∨ Skβ+1 . We extend g to a map f k+1 : Zk+1 → X by defining
β
f k+1 |Sk+1 = ϕβ . This implies that f k+1 induces an epimorphism on πk+1 .
β
The remaining conditions on homotopy groups are easy to check.

Remark 9.8.6. If X is path-connected and A is a point, the construction

of a CW model for ( X, A) gives a CW approximation of X with a single
0-cell. In particular, by Whitehead’s Theorem 9.7.2, any connected CW
complex is homotopy equivalent to a CW complex with a single 0-cell.

Proposition 9.8.7. Let g : ( X, A) → ( X ′ , A′ ) be a map of pairs, where A, A′

are nonempty CW complexes. Let ( Z, A) be an n-connected CW model of
( X, A) with associated map f : ( Z, A) → ( X, A), and let ( Z ′ , A′ ) be an
n′ -connected model of ( X ′ , A′ ) with associated map f ′ : ( Z ′ , A′ ) → ( X ′ , A′ ).
Assume that n ≥ n′ . Then there exists a map h : Z → Z ′ , unique up to
homotopy, such that h| A = g| A , and the diagram

f
( Z, A) −−−−→ ( X, A)
 

hy g
y
f′
( Z ′ , A′ ) −−−−→ ( X ′ , A′ )

commutes up to homotopy.

Proof. The proof is a standard induction on skeleta. We begin with the

map g : A → A′ ⊆ Z ′ , and recall that Z is obtained from A by attaching
cells of dimension > n. Let k be the smallest dimension of such a cell,
thus ( A ∪ Zk , A) has a k-connected model, f k : ( Z k , A) → ( A ∪ Zk , A)
such that f k | A = id A . Composing this new map with g allows us to
consider g as having been extended to the k skeleton of Z. Iterating
this process produces our map.

Corollary 9.8.8. CW-approximations are unique up to homotopy equivalence.

More generally, n-connected models of a pair ( X, A) are unique up to homotopy
relative to A.

Proof. Assume that f : ( Z, A) → ( X, A) and f ′ : ( Z ′ , A) → ( X, A) are

two n-connected models of ( X, A). Then we may take ( X, A) = ( X ′ , A′ )
214 algebraic topology

and g = id in the above lemma twice, and conclude that there are two
maps h0 : Z → Z ′ and h1 : Z ′ → Z, such that f ◦ h1 ≃ f ′ (rel. A)
and f ′ ◦ h0 ≃ f (rel. A). In particular, f ◦ (h1 ◦ h0 ) ≃ f (rel. A) and
f ′ ◦ (h0 ◦ h1 ) ≃ f ′ (rel. A). The uniqueness in Proposition 9.8.7 then
implies that h1 ◦ h0 and h0 ◦ h1 are homotopic to the respective identity
maps (rel. A).

Remark 9.8.9. By taking n = n′ is Proposition 9.8.7, we get a functorial-

ity property for n-connected CW models. For example, a map X → X ′
of spaces induces a map of CW approximations Z → Z ′ .

Remark 9.8.10. By letting n vary, and by letting ( Z n , A) be an n-

connected CW model for ( X, A), then Proposition 9.8.7 gives a tower of
CW models


ZG 2


1
? Z

 
A / Z0 /X

with commutative triangle on the left, and homotopy-commutative

triangles on the right.

Example 9.8.11 (Whitehead towers). Assume X is an arbitrary CW

complex with A ⊂ X a point. Then the resulting tower of n-connected
CW modules of ( X, A) amounts to a sequence of maps

· · · −→ Z2 −→ Z1 −→ Z0 −→ X

with Z n being n-connected and the map Z n → X inducing isomor-

phisms on all homotopy groups πi with i > n. The space Z0 is path-
connected and homotopy equivalent to the component of X containing
A, so one may assume that Z0 equals this component. The space Z1
is simply-connected, and the map Z1 → X has the homotopy prop-
erties of the universal cover of the component Z0 of X. In general, if
X is connected, the map Z n → X has the homotopy properties of an
n-connected cover of X. An example of a 2-connected cover of S2 is the
Hopf map S3 → S2 .

Example 9.8.12 (Postnikov towers). If X is a connected CW complex,

the tower of n-connected models for the pair (CX, X ), with CX the
cone on X, is called the Postnikov tower of X. Relabeling Z n as X n−1 ,
 basics of homotopy theory 215

the Postnikov tower gives a commutative diagram


XG 3


2
X
?


X / X1

where the induced homomorphism πi ( X ) → πi ( X n ) is an isomorphism

for i ≤ n and πi ( X n ) = 0 if i > n. Indeed, by Definition 9.8.1 we get
π i ( X n ) = π i ( Z n +1 ) ∼
= πi (CX ) = 0 for i ≥ n + 1.

9.9 Eilenberg-MacLane spaces

Definition 9.9.1. A space X having just one nontrivial homotopy group

πn ( X ) = G is called an Eilenberg-MacLane space K ( G, n).

Example 9.9.2. We have already seen that S1 is a K (Z, 1) space, and

RP∞ is a K (Z/2Z, 1) space. The fact that CP∞ is a K (Z, 2) space will
be discussed in Example 9.11.16 by making use of fibrations and the
associated long exact sequence of homotopy groups.
Lemma 9.9.3. If a CW-pair ( X, A) is r-connected (r ≥ 1) and A is s-
connected (s ≥ 0), then the map πi ( X, A) → πi ( X/A) induced by the
quotient map X → X/A is an isomorphism if i ≤ r + s and onto if i =
r + s + 1.

Proof. Let CA be the cone on A and consider the complex

Y = X ∪ A CA

obtained from X by attaching the cone CA along A ⊆ X. Since CA is a

contractible subcomplex of Y, the quotient map

q : Y −→ Y/CA = X/A

is obtained by deforming CA to the cone point inside Y, so it is a

homotopy equivalence. So we have a sequence of homomorphisms
∼
= ∼
=
πi ( X, A) −→ πi (Y, CA) ←− πi (Y ) −→ πi ( X/A),

where the first and second maps are induced by the inclusion of pairs,
the second map is an isomorphism by the long exact sequence of the
pair (Y, CA)

0 = πi (CA) → πi (Y ) → πi (Y, CA) → πi−1 (CA) = 0,

216 algebraic topology

and the third map is the isomorphism q∗ . It therefore remains to

investigate the map πi ( X, A) −→ πi (Y, CA). We know that ( X, A) is
r-connected and (CA, A) is (s + 1)-connected. The second fact once
again follows from the long exact sequence of the pair and the fact that
A is s-connected. Using the Excision Theorem 9.5.1, the desired result
follows immediately.

n +1
Lemma 9.9.4. Assume n ≥ 2. If X = ( Sαn ) ∪
W S
α β eβ is obtained
n enβ+1
W
from by attaching (n + 1)-cells
α Sα via basepoint-preserving maps
ϕβ : Snβ → α Sαn , then
W

Sαn )/⟨ϕβ ⟩ = ( Z) / ⟨ ϕ β ⟩.
_ M
πn ( X ) = πn (
α α

Proof. Consider the following portion of the long exact sequence for
the homotopy groups of the n-connected pair ( X, α Sαn ):
W

Sαn ) −→ πn ( Sαn ) −→ πn ( X ) −→ πn ( X, Sαn ) = 0,

_ ∂ _ _
πn+1 ( X,
α α α

where the fact that πn ( X, α Sαn ) = 0 follows by Corollary 9.4.8 of the

W

Cellular Approximation theorem. So πn ( X ) ∼ = πn ( α Sαn )/Im(∂).

W
W n W n +1
We have the identification X/ α Sα ≃ β S β , so by Lemma 9.9.3
W n ∼
S ) = πn+1 ( Sn+1 ) is free
W
and Lemma 9.6.6 we get that πn+1 ( X, α α β β
with a basis consisting of the characteristic maps Φ β of the cells enβ+1 .
Since ∂([Φ β ]) = [ϕβ ], the claim follows.

Example 9.9.5. Any abelian group G can be realized as πn ( X ) with

n ≥ 2 for some space X. In fact, given a presentation G = ⟨ gα | r β ⟩, we
can can take  
Sαn ∪ enβ+1 ,
_ [
X=
α β

with the Sαn ’s corresponding to the generators of G, and with enβ+1

attached to α Sαn by a map f : Snβ → α Sαn satisfying [ f ] = r β . Note
W W

also that by cellular approximation, πi ( X ) = 0 for i < n, but nothing

can be said about πi ( X ) with i > n.

Theorem 9.9.6. For any n ≥ 1 and any group G (which is assumed abelian
if n ≥ 2) there exists an Eilenberg-MacLane space K ( G, n).

Proof. Let Xn+1 = ( α Sαn ) ∪ β enβ+1 be the (n − 1)-connected CW

W S

complex of dimension n + 1 with πn ( Xn+1 ) = G, as constructed in

Example 9.9.5. Enlarge Xn+1 to a CW complex Xn+2 obtained from
Xn+1 by attaching (n + 2)-cells eγn+2 via maps representing some set of
generators of πn+1 ( Xn+1 ). Since ( Xn+2 , Xn+1 ) is (n + 1)-connected (by
Corollary 9.4.8), the long exact sequence for the homotopy groups of
 basics of homotopy theory 217

the pair ( Xn+2 , Xn+1 ) yields isomorphisms πi ( Xn+2 ) = πi ( Xn+1 ) for

i ≤ n, together with the exact sequence

∂
· · · → πn+2 ( Xn+2 , Xn+1 ) → πn+1 ( Xn+1 ) → πn+1 ( Xn+2 ) → 0.

Next note that ∂ is an isomorphism by construction and Lemma 9.9.3.

Indeed, Lemma 9.9.3 yields that the quotient map Xn+2 → Xn+2 /Xn+1
induces an epimorphism

πn+2 ( Xn+2 , Xn+1 ) → πn+2 ( Xn+2 /Xn+1 ) ∼ Sγn+2 ),

_
= π n +2 (
γ

which is an isomorphism for n ≥ 2. Moreover, we also have an epimor-

phism πn+2 ( γ Sγn+2 ) → πn+1 ( Xn+1 ) by our construction of Xn+2 . As
W

∂ is onto, we then get that πn+1 ( Xn+2 ) = 0.

Repeat this construction inductively, at the k-th stage attaching (n +
k + 1)-cells to Xn+k to create a CW complex Xn+k+1 with vanishing
πn+k and without changing the lower homotopy groups. The union of
this increasing sequence of CW complexes is then a K ( G, n) space.

Corollary 9.9.7. For any sequence of groups { Gn }n∈N , with Gn abelian for
n ≥ 2, there exists a space X such that πn ( X ) ∼
= Gn for any n.

Proof. Call X n = K ( Gn , n). Then X = ∏n X n has the desired prescribed

homotopy groups.

Lemma 9.9.8. Let X be a CW complex of the form ( α Sαn ) ∪ β enβ+1 for

W S

some n ≥ 1. Then for every homomorphism ψ : πn ( X ) → πn (Y ) with Y a

path-connected space, there exists a map f : X → Y such that f ∗ = ψ on πn .

Proof. Recall from Lemma 9.9.4 that πn ( X ) is generated by the inclu-

sions iα : Sαn ,→ X. Let f send the wedge point of X to a basepoint
of Y, and extend f onto Sαn by choosing a fixed representative for
ψ([iα ]) ∈ πn (Y ). This then allows us to define f on the n-skeleton
Xn = α Sαn of X, and we notice that, by construction of f : Xn → Y,
W

we have that
f ∗ ([iα ]) = [ f ◦ iα ] = [ f |Sαn ] = ψ([iα ]).

Because the iα generate πn ( Xn ), we then get that f ∗ = ψ.

To extend f over a cell enβ+1 , we need to show that the composition
of the attaching map ϕβ : Sn → Xn for this cell with f is nullhomotopic
in Y. We have [ f ◦ ϕβ ] = f ∗ ([ϕβ ]) = ψ([ϕβ ]) = 0, as the ϕβ are precisely
the relators in πn ( X ) by Example 9.9.5. Thus we obtain an extension
f : X → Y. Moreover, f ∗ = ψ since the elements [iα ] generate πn ( Xn ) =
π n ( X ).

Proposition 9.9.9. The homotopy type of a CW complex K ( G, n) is uniquely

determined by G and n.
218 algebraic topology

Proof. Let K and K ′ be K ( G, n) CW complexes, and assume without loss

of generality (since homotopy equivalence is an equivalence relation)
that K is the particular K ( G, n) constructed in Theorem 9.9.6, i.e., built
from a space X as in Lemma 9.9.8 by attaching cells of dimension n + 2
and higher. Since X = Kn+1 , we have that πn ( X ) = πn (K ) = πn (K ′ ),
and call the composition of these isomorphisms ψ : πn ( X ) → πn (K ′ ).
By Lemma 9.9.8, there is a map f : X → K ′ inducing ψ on πn . To
extend this map over K, we proceed inductively, first extending it over
the (n + 2)-cells, than over the (n + 3)-cells, and so on.
Let eγn+2 be an (n + 2)-cell of K, with attaching map ϕγ : Sn+1 → X.
Then f ◦ ϕγ : Sn+1 → K ′ is nullhomotopic since πn+1 (K ′ ) = 0. There-
fore, f extends over eγn+2 . Proceed similarly for higher dimensional
cells of K to get a map f : K → K ′ which is a weak homotopy equiva-
lence. By Whitehead’s Theorem 9.7.2, we conclude that f is a homotopy
equivalence.

9.10 Hurewicz Theorem

Theorem 9.10.1 (Hurewicz). If a space X is (n − 1)-connected and n ≥ 2,

then He i ( X ) = 0 for i < n and πn ( X ) ∼
= Hn ( X ). Moreover, if a pair ( X, A)
is (n − 1)-connected with n ≥ 2, and π1 ( A) = 0, then Hi ( X, A) = 0 for all
i < n and πn ( X, A) ∼ = Hn ( X, A).

Proof. First, since all hypotheses and assertions in the statement deal
with homology and homotopy groups, if we prove the statement for
a CW approximation of X (or ( X, A)) then the results will also hold
for the original space (or pair). Hence, we assume without loss of
generality that X is a CW complex and ( X, A) is a CW-pair.
Secondly, the relative case can be reduced to the absolute case. In-
deed, since ( X, A) is (n − 1)-connected and that A is 1-connected,
Lemma 9.9.3 implies that πi ( X, A) = πi ( X/A) for i ≤ n, while
Hi ( X, A) = He i ( X/A) always holds for CW-pairs.
In order to prove the absolute case of the theorem, let x0 be a 0-cell
in X. Since X, hence also ( X, x0 ), is (n − 1)-connected, Corollary 9.8.5
tells us that we can replace X by a homotopy equivalent CW complex
with (n − 1)-skeleton a point, i.e., Xn−1 = x0 . In particular, H
ei (X ) = 0
for i < n. For showing that πn ( X ) ∼ = Hn ( X ), we may disregard any
cells of dimension greater than n + 1 since these have no effect on πn
or Hn . Thus we may assume that X has the form ( α Sαn ) ∪ β enβ+1 . By
W S

Lemma 9.9.4, we then have that πn ( X ) ∼ = ( α Z)/⟨ϕβ ⟩. On the other

L

hand, cellular homology yields the same calculation for Hn ( X ), so we

are done.

Remark 9.10.2. One cannot expect any sort of relationship between

πi ( X ) and Hi ( X ) beyond n. For example, Sn has trivial homology in
 basics of homotopy theory 219

degrees > n, but many nontrivial homotopy groups in this range, if

n ≥ 2. On the other hand, CP∞ has trivial higher homotopy groups
in the range > 2 (as a K (Z, 2) space), but many nontrivial homology
groups in this range.

Recall the Hurewicz Theorem has been already used for proving the
important Corollary 9.7.3. Here we give another important application
of Theorem 9.10.1:

Theorem 9.10.3. If f : X → Y induces isomorphisms on homotopy groups

πn for all n, then it induces isomorphisms on homology and cohomology
groups with G coefficients, for any group G.

Proof. By the universal coefficient theorems, it suffices to show that f

induces isomorphisms on integral homology groups H∗ (−; Z).
We only prove here the assertion under the extra condition that
X is simply connected (the general case follows easily from spectral
sequence theory, and it will be dealt with later on). As before, after re-
placing Y with the homotopy equivalent space defined by the mapping
cylinder M f of f , we can assume that f is an inclusion. Since by the
hypothesis, πn ( X ) ∼= πn (Y ) for all n, with isomorphisms induced by
the inclusion f , the homotopy long exact sequence of the pair (Y, X )
yields that πn (Y, X ) = 0 for all n. By the relative Hurewicz theorem (as
π1 ( X ) = 0), this gives that Hn (Y, X ) = 0 for all n. Hence, by the long
exact sequence for homology, Hn ( X ) ∼ = Hn (Y ) for all n, and the proof
is complete.

Example 9.10.4. Take X = RP2 × S3 and Y = S2 × RP3 . As seen in

Example 9.1.19, X and Y have isomorphic homotopy groups πn for
all n, but H5 ( X ) ̸∼
= H5 (Y ). So there cannot exist a map f : X → Y
inducing the isomorphisms on the πn .

9.11 Fibrations. Fiber bundles

Definition 9.11.1 (Homotopy Lifting Property). A map p : E → B has

the homotopy lifting property (HLP) with respect to a space X if, given a
homotopy gt : X → B, and a lift ge0 : X → E of g0 , there exists a homotopy
get : X → E lifting gt and extending ge0 .

ge0
X /E
?

p
get


X /B
gt
220 algebraic topology

Definition 9.11.2 (Lift Extension Property). A map p : E → B has the

lift extension property (LEP) with respect to a pair ( Z, A) if for all maps
f : Z → B and g : A → E, there exists a lift fe : Z → E of f extending g.

7? E

g
p
fe
 
A /Z /B
f

Remark 9.11.3. (HLP) is a special case of (LEP), with Z = X × [0, 1],

and A = X × {0}.

Definition 9.11.4. A fibration p : E → B is a map having the homotopy

lifting property with respect to all spaces X.

Definition 9.11.5 (Homotopy Lifting Property with respect to a pair).

A map p : E → B has the homotopy lifting property with respect to a pair
( X, A) if each homotopy gt : X → B lifts to a homotopy get : X → E starting
with a given lift ge0 and extending a given lift get : A → E.

Remark 9.11.6. The homotopy lifting property with respect to the pair
( X, A) is the lift extension property for ( X × I, X × {0} ∪ A × I ).
Remark 9.11.7. The homotopy lifting property with respect to a disk
D n is equivalent to the homotopy lifting property with respect to the
pair ( D n , ∂D n ), since the pairs ( D n × I, D n × {0}) and ( D n × I, D n ×
{0} ∪ ∂D n × I ) are homeomorphic. This implies that a fibration has
the homotopy lifting property with respect to all CW pairs ( X, A). Indeed,
the homotopy lifting property for disks is in fact equivalent to the
homotopy lifting property with respect to all CW pairs ( X, A). This
can be easily seen by induction over the skeleta of X, so it suffices to
construct a lifting get one cell of X \ A at a time. Composing with the
characteristic map D n → X of a cell then gives the reduction to the case
( X, A) = ( D n , ∂D n ).
Theorem 9.11.8 (Long exact sequence for homotopy groups of a fibra-
tion). Given a fibration p : E → B, points b ∈ B and e ∈ F := p−1 (b), there
∼
=
is an isomorphism p∗ : πn ( E, F, e) −→ πn ( B, b) for all n ≥ 1. Hence, if B is
path-connected, there is a long exact sequence of homotopy groups:
p∗
· · · −→ πn ( F, e) −→ πn ( E, e) −→ πn ( B, b) −→ πn−1 ( F, e) −→ · · ·
· · · −→ π0 ( E, e) −→ 0

Proof. To show that p∗ is onto, represent an element of πn ( B, b) by a

map f : ( I n , ∂I n ) → ( B, b), and note that the constant map to e is a
lift of f to E over J n−1 ⊂ I n . The homotopy lifting property for the
 basics of homotopy theory 221

pair ( I n−1 , ∂I n−1 ) extends this to a lift fe : I n → E. This lift satisfies

fe(∂I n ) ⊂ F since f (∂I n ) = b. So fe represents an element of πn ( E, F, e)
with p∗ ([ fe]) = [ f ] since p fe = f .
To show the injectivity of p∗ , let fe0 , fe1 : ( I n , ∂I n , J n−1 ) → ( E, F, e)
be so that p∗ ( fe0 ) = p∗ ( fe1 ). Let H : ( I n × I, ∂I n × I ) → ( B, b) be a
homotopy from p fe0 to p fe1 . We have a partial lift given by fe0 on I n × {0},
fe1 on I n × {1} and the constant map to e on J n−1 × I. The homotopy
lifting property for CW pairs extends this to a lift H e : I n × I → E giving
a homotopy fet : ( I n , ∂I n , J n−1 ) → ( E, F, e) from fe0 to fe1 .
Finally, the long exact sequence of the fibration follows by plugging
πn ( B, b) in for πn ( E, F, e) in the long exact sequence for the pair ( E, F ).
The map πn ( E, e) → πn ( E, F, e) in the latter sequence becomes the
p∗
composition πn ( E, e) → πn ( E, F, e) → πn ( B, b), which is exactly p∗ :
πn ( E, e) → πn ( B, b). The surjectivity of π0 ( F, e) → π0 ( E, e) follows
from the path-connectedness of B, since a path in E from an arbitrary
point x ∈ E to F can be obtained by lifting a path in B from p( x ) to
b.

Definition 9.11.9. Given two fibrations pi : Ei → B, i = 1, 2, a map

f : E1 → E2 is fiber-preserving if the diagram
f
E1 / E2

p1 p2
 
B
commutes. Such a map f is called a fiber homotopy equivalence if f is both
fiber-preserving and a homotopy equivalence, i.e., there is a map g : E2 → E1
such that f and g are fiber-preserving and f ◦ g and g ◦ f are homotopic to
the identity maps by fiber-preserving maps.
Definition 9.11.10 (Fiber Bundle). A map p : E → B is a fiber bundle
with fiber F if, for any point b ∈ B, there exists a neighborhood Ub of b with
a homeomorphism h : p−1 (Ub ) → Ub × F so that the following diagram
commutes:

p−1 (Ub )
h / Ub × F

p pr

" }
Ub

Remark 9.11.11. Fibers of fibrations are homotopy equivalent, while

fibers of fiber bundles are homeomorphic.
Theorem 9.11.12 (Hurewicz). Fiber bundles over paracompact spaces are
fibrations.
222 algebraic topology

Here are some easy examples of fiber bundles.

Example 9.11.13. If F is discrete, a fiber bundle with fiber F is a covering

map. Moreover, the long exact sequence for the homotopy groups
yields that p∗ : πi ( E) → πi ( B) is an isomorphism if i ≥ 2 and a
monomorphism for i = 1.

Example 9.11.14. The Möbius band I × [−1, 1]⧸(0, y) ∼ (1, −y) −→ S1

is a fiber bundle with fiber [−1, 1], induced from the projection map
I × [−1, 1] → I.

0 1

−1

Example 9.11.15. By glueing the unlabeled edges of a Möbius band, we

get K → S1 (where K is the Klein bottle), a fiber bundle with fiber S1 .

Example 9.11.16. The following is a fiber bundle with fiber S1 :

S1 ,→ S2n+1 (⊂ Cn+1 ) −→ CPn

( z0 , . . . , z n ) 7 → [ z0 : . . . : z n ] = [ z ]

For [z] ∈ CPn , there is an i such that zi ̸= 0. Then we have a neighbor-

hood
U[z] = {[z0 : . . . : 1 : . . . : zn ]} ∼
= Cn
(with the entry 1 in place of the ith coordinate) of [z], with a homeo-
morphism

p−1 (U[z] ) −→ U[z] × S1

(z0 , . . . , zn ) 7→ ([z0 : . . . : zn ], zi /|zi |).

By letting n go to infinity, we get a diagram of fibrations

S1  p = S1  q = ··· = S1  p

! " !
S2n+1 ⊂ S2n+3 ⊂ ... ⊂ S∞

  
CPn ⊂ CPn+1 ⊂ ... ⊂ CP∞
 basics of homotopy theory 223

In particular, from the long exact sequence of the fibration

S1 ,→ S∞ −→ CP∞

with S∞ contractible, we obtain that

(
∞ Z i=2
πi (CP ) ∼
= π i −1 ( S 1 ) =
0 i ̸= 2

i.e.,
CP∞ = K (Z, 2),
as already mentioned in our discussion about Eilenberg-MacLane
spaces.
Remark 9.11.17. As we will see later on, for any topological group
πG
G there exists a “universal fiber bundle” G ,→ EG −→ BG with EG
contractible, classifying the space of (principal) G-bundles. That is, any
G-bundle π : E → B over a space B is determined by (the homotopy
class of) a classifying map f : B → BG by pull-back: π ∼ = f ∗ πG :

E EG ≃ {pt}
π πG
 
B / BG
f

From this point of view, CP∞ can be identified with the classifying
space BS1 of (principal) S1 -bundles.
Example 9.11.18. By letting n = 1 in the fibration of Example 9.11.16,
the corresponding bundle

S1 ,→ S3 −→ CP1 ∼
= S2 (9.11.1)

is called the Hopf fibration. The long exact sequence of homotopy group
for the Hopf fibration gives: π2 (S2 ) ∼
= π1 (S1 ) and πn (S3 ) ∼
= πn (S2 ) for
∞
all n ≥ 3. Together with the fact that CP = K (Z, 2), this shows that S2
and S3 × CP∞ are simply-connected CW complexes with isomorphic
homotopy groups, though they are not homotopy equivalent as can be
easily seen from cellular homology.
Example 9.11.19. A fiber bundle similar to that of Example 9.11.16 can
be obtained by replacing C with the quaternions H, namely:

S3 ,→ S4n+3 −→ HPn .

(Note that S4n+3 can be identified with the unit sphere in Hn+1 .) In
particular, by letting n = 1 we get a second Hopf fiber bundle

S3 ,→ S7 −→ HP1 ∼
= S4 . (9.11.2)
224 algebraic topology

A third example of a Hopf bundle

S7 ,→ S15 −→ S8 (9.11.3)

can be constructed by using the nonassociative 8-dimensional algebra

O of Cayley octonions, whose elements are pair of quaternions ( a1 , a2 )
with multiplication defined by

( a1 , a2 ) · (b1 , b2 ) = ( a1 b1 − b̄2 a2 , a2 b̄1 + b2 a1 ).

Here we regard S15 as the unit sphere in the 16-dimensional vector space
O2 , and the projection map S15 −→ S8 = O ∪ {∞} is (z0 , z1 ) 7→ z0 z1−1
(just like for the other Hopf bundles). There are no fiber bundles
with fiber, total space and base spheres, other than those provided by
the Hopf bundles of (9.11.1), (9.11.2) and (9.11.3). Finally, note that
there is an “octonion projective plane” OP2 obtained by glueing a cell
e16 to S8 via the Hopf map S15 → S8 ; however, there is no octonion
analogue of RPn , CPn or HPn for higher n, since the associativity of
multiplication is needed for the relation (z0 , · · · , zn ) ∼ λ(z0 , · · · , zn ) to
be an equivalence relation.

Example 9.11.20. Other examples of fiber bundles are provided by the

orthogonal and unitary groups:

O(n − 1) ,→ O(n) → Sn−1

A 7→ Ax,

where x is a fixed unit vector in Rn . Similarly, there is a fibration

U (n − 1) ,→ U (n) → S2n−1
A 7→ Ax,

with x a fixed unit vector in Cn . These examples will be discussed in

some detail in the next section.

9.12 More examples of fiber bundles

Definition 9.12.1. For n ≤ k, the n-th Stiefel manifold associated to Rk is

defined as
Vn (Rk ) := {n-frames in Rk },
where an n-frame in Rk is an n-tuple {v1 , . . . , vn } of orthonormal vectors in
Rk , i.e., v1 , . . . , vn are pairwise orthonormal: ⟨vi , v j ⟩ = δij .
We assign Vn (Rk ) the subspace topology induced from

Vn (Rk ) ⊂ Sk−1 × · · · × Sk−1 ,

| {z }
n times

where Sk−1 × · · · × Sk−1 has the usual product topology.

 basics of homotopy theory 225

Example 9.12.2. V1 (Rk ) = Sk−1 .

Example 9.12.3. Vn (Rn ) ∼

= O ( n ).

Definition 9.12.4. The n-th Grassmann manifold associated to Rk is defined

as:
Gn (Rk ) := {n-dimensional vector subspaces in Rk }.

Example 9.12.5. G1 (Rk ) = RPk−1

There is a natural surjection

p : Vn (Rk ) −→ Gn (Rk )

given by
{v1 , . . . , vn } 7→ span{v1 , . . . , vn }.

The fact that p is onto follows by the Gram-Schmidt procedure. So

Gn (Rk ) is endowed with the quotient topology via p.

Lemma 9.12.6. The projection p is a fiber bundle with fiber Vn (Rn ) = O(n).

Proof. Let V ∈ Gn (Rk ) be fixed. The fiber p−1 (V ) consists on n-frames

in V ∼= Rn , so it is homeomorphic to Vn (Rn ). Let us now choose an
orthonormal frame on V. By projection and Gram-Schmidt, we get
orthonormal frames on all “nearby” (in some neighborhood U of V)
vector subspaces V ′ . Indeed, by projecting the frame of V orthogo-
nally onto V ′ we get a (non-orthonormal) basis for V ′ , then apply the
Gram-Schmidt process to this basis to make it orthonormal. This is a
continuous process. The existence of such frames on all n-planes in
U allows us to identify them with Rn , so p−1 (U ) is identified with
U × Vn (Rn ).

To conclude this discussion, we have shown that for k > n, there are
fiber bundles:

 / Vn (Rk ) / Gn (Rk )
O(n) (9.12.1)

A similar method gives the following fiber bundle for all triples
m < n ≤ k:


Vn−m (Rk−m ) 
p
/ Vn (Rk ) / Vm (Rk ) (9.12.2)

{ v1 , . . . , v n }  / { v1 , . . . , v m }

Here, the projection p sends an n-frame onto the m-frame formed

by its first m vectors, so the fiber consists of (n − m)-frames in the
(k − m)-plane orthogonal to the given frame.
226 algebraic topology

Example 9.12.7. If k = n in the bundle (9.12.2), we get the fiber bundle


O(n − m)  / O(n) / Vm (Rn ). (9.12.3)

Here, O(n − m) is regarded as the subgroup of O(n) fixing the first m

standard basis vectors. So Vm (Rn ) is identifiable with the coset space
O(n)⧸
O(n − m), or the orbit space of the free action of O(n − m) on
O(n) by right multiplication. Similarly,

= O(n)⧸O(m) × O(n − m),

Gm (Rn ) ∼

where O(m) × O(n − m) consists of the orthogonal transformations of

Rn taking the m-plane spanned by the first m standard basis vectors to
itself.
If, moreover, we take m = 1 in (9.12.3), we get the fiber bundle


O ( n − 1)  / O(n) / S n −1 (9.12.4)
!
 / A 0
A
0 1

B / Bu

with u ∈ Sn−1 some fixed unit vector. In particular, this identifies Sn−1
as an orbit (or homogeneous) space:

= O(n)⧸O(n − 1).
S n −1 ∼

Example 9.12.8. If m = 1 in the bundle (9.12.2), we get the fiber bundle


Vn−1 (Rk−1 )  / Vn (Rk ) / S k −1 . (9.12.5)

By using the long exact sequence for bundle (9.12.5) and induction on
n, it follows readily that Vn (Rk ) is (k − n − 1)-connected.

Remark 9.12.9. The long exact sequence of homotopy groups for the
bundle (9.12.4) shows that πi (O(n)) is independent of n for n large. We
call this the stable homotopy group πi (O). Bott Periodicity shows that
πi (O) is periodic in i with period 8. Its values are:

i 1 2 3 4 5 6 7 8
πi (O) Z/2 Z/2 0 Z 0 0 0 Z

Definition 9.12.10.
∞ ∞
Vn (R∞ ) := Vn (Rk ) Gn (R∞ ) := Gn (Rk )
[ [

k =1 k =1
 basics of homotopy theory 227

The infinite grassmanian Gn (R∞ ) carries a lot of topological informa-

tion. As we will see later on, the space Gn (R∞ ) is the classifying space
for rank-n real vector bundles. In fact, we get a “limit” fiber bundle:

O(n)  / Vn (R∞ ) / Gn (R∞ ). (9.12.6)

Moreover, we have the following:

Proposition 9.12.11. Vn (R∞ ) is contractible.

Proof. By using the bundle (9.12.5) for k → ∞, we see that πi (Vn (R∞ )) =
0 for all i. Using the CW structure and Whitehead’s Theorem 9.7.2
shows that Vn (R∞ ) is contractible.
Alternatively, we can define an explicit homotopy ht : R∞ → R∞ by

ht ( x1 , x2 , . . .) := (1 − t)( x1 , x2 , . . .) + t(0, x1 , x2 , . . .).

Then ht is linear for each t with ker ht = {0}. So ht preserves inde-

pendence of vectors. Applying ht to an n-frame we get an n-tuple of
independent vectors, which can be made orthonormal by the Gram-
Schmidt (G-S, for short) process. We then get a deformation retraction
of Vn (R∞ ) onto the subspace of n-frames with first coordinate zero.
Repeating this procedure n times, we get a deformation of Vn (R∞ ) to
the subspace of n-frames with first n coordinates zero.
Let {e1 , . . . , en } be the standard n-frame in R∞ . For an n-frame
{v1 , . . . , vn } of vectors with first n coordinates zero, define a homotopy
k t : Vn (R∞ ) → Vn (R∞ ) by
 
k t ({v1 , . . . , vn }) := (1 − t){v1 , . . . , vn } + t{e1 , . . . , en } ◦ (G − S).

Then k t preserves linear independence and orthonormality by Gram-

Schmidt.
Composing ht and k t , any n-frame is moved continuously to the
standard n-frame {e1 , . . . , en }. Thus k t ◦ ht is a contraction of Vn (R∞ ).

Similar considerations apply if we use C or H instead of R, so we

can define complex or quaternionic Stiefel and Grasmann manifolds,
by using the usual hermitian inner products in Ck and Hk , respectively.
In particular, O(n) gets replaced by U (n) if C is used, and Sp(n) is
the quaternionic analog of this. Then similar fiber bundles can be
constructed in the complex and quaternionic setting. For example, over
C we get fiber bundles

U (n) 
p
/ Vn (Ck ) / Gn (Ck ), (9.12.7)

with Vn (Ck ) a (2k − 2n)-connected space. As k → ∞, we get a fiber

bundle

U (n)  / Vn (C∞ ) / Gn (C∞ ), (9.12.8)
228 algebraic topology

with Vn (C∞ ) contractible. As we will see later on, this means that
Vn (C∞ ) is the classifying space for rank-n complex vector bundles. We
also have a fiber bundle similar to (9.12.4)

 / U (n) / S2n−1 ,
U ( n − 1) (9.12.9)

whose long exact sequence of homotopy groups then shows that

πi (U (n)) is stable for large n. Bott periodicity shows that this sta-
ble group πi (U ) repeats itself with period 2: the relevant groups are
0 for i even, and Z for i odd. Note that by (9.12.9), odd-dimensional
spheres can be realized as complex homogeneous spaces via

= U (n)⧸U (n − 1).
S2n−1 ∼

Many of these fiber bundles will become essential tools in the next
chapter for computing (co)homology of matrix groups, with a view
towards classifying spaces and characteristic classes of manifolds.

9.13 Turning maps into fibration

In this section, we show that any map is homotopic to a fibration.

Given a map f : A → B, define

E f := {( a, γ) | a ∈ A, γ : [0, 1] → B with γ(0) = f ( a)}.

E f is a topological space with respect to the compact-open topology.

Then A can be regarded as a subset of E f , by mapping a ∈ A to
( a, c f (a) ), where c f (a) is the constant path based at the image of a under
f . Define
p
E f −→ B
( a, γ) 7→ γ(1)

Then p| A = f , so f = p ◦ i where i is the inclusion of A in E f . Moreover,

i : A −→ E f is a homotopy equivalence, and p : E f −→ B is a fibration
with fiber A. So f can be factored as a composition of a homotopy
equivalence and a fibration:


A / Ef /9 B
h.e. fibration
i p
f

Example 9.13.1. If A = {b} ,→ B and f is the inclusion of b in B, then

E f =: PB is the contractible space of paths in B starting at b (called
the path-space of B). In this case, the above construction yields the path
fibration
ΩB = p−1 (b) ,→ PB −→ B,
 basics of homotopy theory 229

where ΩB is the space of all loops in B based at b, and PB −→ B is

given by γ 7→ γ(1). Since PB is contractible, the associated long exact
sequence of the fibration yields that

πi ( B ) ∼
= πi−1 (ΩB) (9.13.1)

for all i.
The isomorphism (9.13.1) suggests that the Hurewicz Theorem 9.10.1
can also be proved by induction on the degree of connectivity. Indeed,
if B is n-connected then ΩB is (n − 1)-connected. We’ll give the details
of such an approach by using spectral sequences.

The following result is useful for computations:

Proposition 9.13.2 (Puppé sequence). Given a fibration F ,→ E → B,
there is a sequence of maps

· · · −→ Ω2 B −→ ΩF −→ ΩE −→ ΩB −→ F −→ E −→ B

with any two consecutive maps forming a fibration.

9.14 Exercises

1. Let f : X → Y be a homotopy equivalence. Let Z be any other space.

Show that f induces bijections:

f ∗ : [ Z, X ] → [ Z, Y ] and f ∗ : [Y, Z ] → [ X, Z ] ,

where [ A, B] denotes the set of homotopy classes of maps from the

space A to B.

2. Find examples of spaces X and Y which have the same homology

groups, cohomology groups, and cohomology rings, but with different
homotopy groups.

3. Use homotopy groups in order to show that there is no retraction

RPn → RPk if n > k > 0.

4. Show that an n-connected, n-dimensional CW complex is con-

tractible.

5. (Extension Lemma)
Given a CW pair ( X, A) and a map f : A → Y with Y path-connected,
show that f can be extended to a map X → Y if πn−1 (Y ) = 0 for all n
such that X \ A has cells of dimension n.

6. Show that a CW complex retracts onto any contractible subcomplex.

(Hint: Use the above extension lemma.)
230 algebraic topology

7. If p : ( X̃, Ã, x̃0 ) → ( X, A, x0 ) is a covering space with à = p−1 ( A),

show that the map p∗ : πn ( X̃, Ã, x̃0 ) → πn ( X, A, x0 ) is an isomorphism
for all n > 1.

8. Show that a CW complex is contractible if it is the union of an

increasing sequence of subcomplexes X1 ⊂ X2 ⊂ · · · such that each
inclusion Xi ,→ Xi+1 is nullhomotopic. Conclude that S∞ is contractible,
and more generally, this is true for the infinite suspension Σ∞ ( X ) :=
n≥0 Σ ( X ) of any CW complex X.
S n

9. Use cellular approximation to show that the n-skeletons of homotopy

equivalent CW complexes without cells of dimension n + 1 are also
homotopy equivalent.

10. Show that a closed simply-connected 3-manifold is homotopy

equivalent to S3 . (Hint: Use Poincaré Duality, and also the fact that
closed manifolds are homotopy equivalent to CW complexes.)

11. Let X be a finite CW complex which is n-connected (i.e., πi ( X ) = 0

for all i ≤ n). Show that, for any 1 < k < n, the k-skeleton X k of X is
homotopy equivalent to a bouquet of k-spheres.

12. Show that a map f : X → Y of connected CW complexes is a

homotopy equivalence if it induces an isomorphism on π1 and if a
lift f˜ : X̃ → Ỹ to the universal covers induces an isomorphism on
homology.

13. Show that π7 (S4 ) is non-trivial. [Hint: It contains a Z-summand.]

14. Prove that the space SO(3) of orthogonal 3 × 3 matrices with

determinant 1 is homeomorphic to RP3 .

15. Show that if Sk → Sm → Sn is a fiber bundle, then k = n − 1 and

m = 2n − 1.

16. Show that if there were fiber bundles Sn−1 → S2n−1 → Sn for all n,
then the groups πi (Sn ) would be finitely generated free abelian groups
computable by induction, and non-zero if i ≥ n ≥ 2.

17. Let U (n) be the unitary group. Find πk (U (n)) for k = 1, 2, 3 and
n ≥ 2.

18. If p : E → B is a fibration over a contractible space B, then p is fiber

homotopy equivalent to the trivial fibration B × F → B.
 spectral sequences. applications 231

10
Spectral Sequences. Applications

Most of our considerations involving spectral sequences will be applied

to fibrations. If F ,→ E → B is such a fibration, then a spectral sequence
can be regarded as a machine which takes as input the (co)homology
of the base B and fiber F and outputs the (co)homology of the total
space E. Our emphasis here is on applications of the theory of spectral
sequences, and not so much on developing the theory itself.

10.1 Homological spectral sequences. Definitions

We begin with a discussion of homological spectral sequences.

Definition 10.1.1. A (homological) spectral sequence is a sequence

{ E∗r ,∗ , dr∗,∗ }r≥0

of chain complexes of abelian groups, such that

E∗r+ 1 r
,∗ = H∗ ( E∗,∗ ).

In more detail, we have abelian groups { Erp,q } and maps (called “differentials”)

drp,q : Erp,q → Erp−r,q+r−1

such that (dr )2 = 0 and

 
ker drp,q : Erp,q → Erp−r,q+r−1
+1
Erp,q :=  .
Im drp+r,q−r+1 : Erp+r,q−r+1 → Erp,q

We will focus on the first quadrant spectral sequences, i.e., with

Erp,q = 0 whenever p < 0 or q < 0. Hence, for any fixed ( p, q) in
the first quadrant and for sufficiently large r, the differentials drp,q and
drp+r,q−r+1 vanish, so that

+1
Erp,q = Erp,q = · · · = E∞
p,q .
232 algebraic topology

Figure 10.1: r-th page Er

dr

q+r−1 Erp−r,q+r−1
dr

q Erp,q
dr
..
.
0
0 ··· p−r p

In this case we say that the spectral sequence degenerates at page Er .

When it is clear from the context which differential we refer to, we
will simply write dr , instead of dr∗,∗ .

Definition 10.1.2. If { Hn }n are groups, we say the spectral sequence con-

verges (or abuts) to H∗ , and we write

( Er , dr ) ⇛ H∗ ,

if for each n there is a filtration

Hn = Dn,0 ⊇ Dn−1,1 ⊇ · · · ⊇ D1,n−1 ⊇ D0,n ⊇ D−1,n+1 = 0

such that, for all p, q,

E∞
p,q =
D p,q⧸
D p−1,q+1 .

Figure 10.2: n-th diagonal of E∞

D0,n

D1,n−1 /D0,n

Dn−1,1 /Dn−2,2

Hn /Dn−1,1
 spectral sequences. applications 233

To read off H∗ from E∞ , we need to solve several extension problems.

But if E∗r ,∗ and H∗ are vector spaces, then

Hn ∼ E∞
M
= p,q ,
p+q=n

since in this case all extension problems are trivial.

Remark 10.1.3. The following observation is very useful in practice:

• If E∞
p,q = 0, for all p + q = n, then Hn = 0.

• If Hn = 0, then E∞
p,q = 0 for all p + q = n.

Before explaining in more detail what is behind the theory of spectral

sequences, we present the special case of a spectral sequence associated
to fibrations, and discuss some immediate applications (including to
Hurewicz theorem).

Theorem 10.1.4 (Serre). If π : E → B is a fibration with fiber F, and with

π1 ( B) = 0 and π0 ( F ) = 0, then there is a first quadrant spectral sequence
with
E2p,q = H p ( B; Hq ( F )) ⇛ H∗ ( E) (10.1.1)

converging to H∗ ( E).

Remark 10.1.5. Fix some coefficient group K. Then, since B and F are
connected, we have:

• E2p,0 = H p ( B; H0 ( F; K)) = H p ( B; K),

2 = H ( B; H ( F; K)) = H ( F; K)
• E0,q 0 q q

The remaining entries on the E2 -page are computed by the universal

coefficient theorem.

Definition 10.1.6. The spectral sequence of the above theorem shall be referred
to as the Leray-Serre spectral sequence of a fibration, and any ring of coefficients
can be used.

Remark 10.1.7. If π1 ( B) ̸= 0, then the coefficients Hq ( F ) on B are acted

upon by π1 ( B), i.e., these coefficients are “twisted” by the monodromy
of the fibration if it is not trivial. As we will see later on, in this case
the E2 -page of the Leray-Serre spectral sequence is given by

E2p,q = H p ( B; Hq ( F )),

i.e., the homology of B with local coefficients Hq ( F ).

234 algebraic topology

Figure 10.3: p-axis and q-axis of E2

H∗ ( F )

H∗ ( B)

10.2 Immediate Applications: Hurewicz Theorem Redux

As a first application of the Leray-Serre spectral sequence, we can now

give a new proof of the Hurewicz Theorem in the absolute case:
Theorem 10.2.1 (Hurewicz Theorem). If X is (n − 1)-connected, n ≥ 2,
e i ( X ) = 0 for i ≤ n − 1 and πn ( X ) ∼
then H = Hn ( X ).

Proof. Consider the path fibration:


ΩX  / PX / X, (10.2.1)

and recall that the path space PX is contractible. Note that the loop
space ΩX is connected, since π0 (ΩX ) ∼= π1 ( X ) = 0. Moreover, since
π1 ( X ) = 0, the Leray-Serre spectral sequence (10.1.1) for the path
fibration has the E2 -page given by

E2p,q = H p ( X, Hq (ΩX )) ⇛ H∗ ( PX ).

We prove the statement of the theorem by induction on n. The

induction starts at n = 2, in which case we clearly have H1 ( X ) = 0
since X is simply-connected. Moreover,

π2 ( X ) ∼
= π1 (ΩX ) ∼
= H1 (ΩX ),

where the first isomorphism follows from the long exact sequence of
homotopy groups for the path fibration, and the second isomorphism
is the abelianization since π2 ( X ), hence also π1 (ΩX ), is abelian. So it
remains to show that we have an isomorphism

H1 (ΩX ) ∼
= H2 ( X ). (10.2.2)

Consider the E2 -page of the Leray-Serre spectral sequence for the path
fibration. We need to show that

d2 : E2,0
2 2
= H2 ( X ) → E0,1 = H1 (ΩX )
 spectral sequences. applications 235

is an isomorphism.

H∗ (ΩX )
E2

H1 (ΩX )
d2

H∗ ( X )
Z H1 ( X ) H2 ( X )

Since { E2p,q } ⇛ H∗ ( PX ) and PX is contactible, we have by Remark

10.1.3 that E∞ p,q = 0 for all p, q > 0. Hence, if d : H2 ( X ) → H1 ( ΩX )
2

is not an isomorphism, then E0,1 3 ̸ = 0 and E3 = ker d2 ̸ = 0. But the

2,0
3
differentials d and higher will not affect E0,1 3 and E3 . So these groups
2,0
remain unchanged (hence non-zero) also on E∞ , contradicting the fact
that E∞ = 0 except for ( p, q) = (0, 0). This proves (10.2.2).
Now assume the statement of the theorem holds for n − 1 and prove
it for n. Since X is (n − 1)-connected, we have by the homotopy long
exact sequence of the path fibration that ΩX is (n − 2)-connected. So
by the induction hypothesis applied to ΩX (assuming now that n ≥ 3,
as the case n = 2 has been dealt with earlier), we have that H e i (ΩX ) = 0
∼
for i < n − 1, and πn−1 (ΩX ) = Hn−1 (ΩX ).
Therefore, we have isomorphisms:

πn ( X ) ∼
= πn−1 (ΩX ) ∼
= Hn−1 (ΩX ),

where the first isomorphism follows from the long exact sequence
of homotopy groups for the path fibration, and the second is by the
induction hypothesis, as already mentioned. So it suffices to show that
we have an isomorphism

Hn−1 (ΩX ) ∼
= Hn ( X ). (10.2.3)

Consider the Leray-Serre spectral sequence for the path fibration. By

using the universal coefficient theorem for homology, the terms on the
E2 -page are given by

E2p,q = H p ( X, Hq (ΩX ))
∼
= H p ( X ) ⊗ Hq (ΩX ) ⊕ Tor( H p−1 ( X ), Hq (ΩX ))
=0

for 0 < q < n − 1, by the induction hypothesis for ΩX.

236 algebraic topology

H∗ (ΩX )

Hn−1 (ΩX )

.. dn
.

H∗ ( X )
...
Z H1 ( X ) H2 ( X ) Hn−1 ( X ) Hn ( X )

Hence, the differentials d2 , d3 · · · dn−1 acting on the entries on the p-axis

for p ≤ n, do not affect these entries. The entries Hn ( X ) and Hn−1 (ΩX )
are affected only by the differential dn . Also, higher differentials starting
with dn+1 do not affect these entries. But since the spectral sequence
converges to H∗ ( PX ) with PX contractible, all entries on the E∞ -page
(except at the origin) must vanish. In particular, this implies that
Hi ( X ) = 0 for 1 ≤ i ≤ n − 1, and dn : Hn ( X ) → Hn−1 (ΩX ) must be an
isomorphism, thus proving (10.2.3).

10.3 Leray-Serre Spectral Sequence

In this section, we give some more details about the Leray-Serre spectral
sequence. We begin with some general considerations about spectral
sequences.
Start off with a chain complex C∗ with a bounded increasing filtration
F C∗ , i.e., each F p C∗ is a subcomplex of C∗ , F p−1 C∗ ⊆ F p C∗ for any p,
•

F p C∗ = C∗ for p very large, and F p C∗ = 0 for p very small. We get an

induced filtration on the homology groups Hi (C∗ ) by

F p Hi (C∗ ) := Im( Hi ( F p C∗ ) → Hi (C∗ )).

The general theory of spectral sequences (e.g., see Hatcher or Griffiths-

Harris), asserts that there exists a homological spectral sequence with
E1 -page given by:

E1p,q = H p+q ( F p C∗ /F p−1 C∗ ) ⇛ H∗ (C∗ )

 spectral sequences. applications 237

and differential d1 given by the connecting homomorphism in the long

exact sequence of homology groups associated to the triple

( F p C∗ , F p−1 C∗ , F p−2 C∗ ).

Moreover, we have
Theorem 10.3.1.

E∞ p
p,q = F H p+q (C∗ ) /F
p −1
H p+q (C∗ )

So to reconstruct H∗ (C∗ ) one needs to solve a collection of extension

problems.
π
Back to the Leray-Serre spectral sequence, let F ,→ E → B be a
fibration with B a simply-connected finite CW-complex. Let C∗ ( E) be
the singular chain complex of E, filtered by

F p C∗ ( E) := C∗ (π −1 ( B p )),

where B p is the p-skeleton of B. Then,

F p C∗ ( E)/F p−1 C∗ ( E) = C∗ (π −1 ( B p ))/C∗ (π −1 ( B p−1 ))

= C∗ (π −1 ( B p ), π −1 ( B p−1 )).

By excision,

H∗ ( F p C∗ ( E)/F p−1 C∗ ( E)) = H∗ (π −1 (e p ), π −1 (∂e p ))

M

ep

where the direct sum is over the p-cells e p in B. Since e p is contractible,

the fibration above it is trivial, so homotopy equivalent to e p × F. Thus,

H∗ (π −1 (e p ), π −1 (∂e p )) ∼
= H∗ (e p × F, ∂e p × F )
∼
= H∗ ( D p × F, S p−1 × F )
∼
= H∗− p ( F )
∼
= H p ( D p , S p−1 ; H∗− p ( F )),

where the third isomorphism follows by the Künneth formula. Alto-

gether, there is a spectral sequence with E1 -page

E1p,q = H p+q ( F p C∗ ( E)/F p−1 C∗ ( E)) ∼ H p ( D p , S p−1 ; Hq ( F )).

M
=
ep

Here, d1 takes E1p,q to e p−1 H p−1 ( D p−1 , S p−2 ; Hq ( F )) by the boundary

L

map of the long exact sequence of the triple ( B p , B p−1 , B p−2 ). By cellular
homology, this is exactly a description of the boundary map of the CW-
chain complex of B with coefficients in Hq ( F ), hence

E2p,q = H p ( B, Hq ( F )).
238 algebraic topology

Remark 10.3.2. If the base B of the fibration is not simply-connected,

then the coefficients Hq ( F ) on B in E2 are acted upon by π1 ( B), i.e.,
these coefficients are “twisted” by the monodromy of the fibration if it
is not trivial, so taking the homology of the E1 -page yields

E2p,q = H p ( B; Hq ( F )),

regarded now as the homology of B with local coefficients Hq ( F ).

The above considerations yield Serre’s theorem:

i π
Theorem 10.3.3. Let F ,→ E → B be a fibration with π1 ( B) = 0 (or π1 ( B)
acts trivially on H∗ ( F )) and π0 ( E) = 0. Then, there is a first quadrant
spectral sequence with E2 -page

E2p,q = H p ( B, Hq ( F ))

which converges to H∗ ( E).

Therefore, there exists a filtration

Hn ( E) = Dn,0 ⊇ Dn−1,1 ⊇ . . . ⊇ D0,n ⊇ D−1,n+1 = 0

such that E∞
p,q = D p,q /D p−1,q+1 .

n-th diagonal of E∞

D0,n

D1,n−1 /D0,n

Dn−1,1 /Dn−2,2

Hn ( E)/Dn−1,1
 spectral sequences. applications 239

(a) We have the following diagram of groups and homomorphisms:

p p +1
H p ( B) = E2p,0 ⊇ ker d2p,0 = E3p,0 ⊇ ker d3p,0 = E4p,0 ⊇ . . . ⊇ ker d p,0 = E p,0
[ O
=

..
.O

E∞
p,0
O
π∗ =

H p ( E)/D p−1,1 H p ( E)
O
onto

H p ( E)

Moreover, the above diagram commutes, i.e., the composition

H p ( E) ↠ E∞ 2
p,0 ⊆ E p,0 = H p ( B ), (10.3.1)

which is also called the edge homomorphism, coincides with π∗ :

H p ( E ) → H p ( B ).

(b) We have the following diagram of groups and homomorphisms:

2
Hq ( F ) = E0,q / / E0,q
3
= Hq ( F )/Im(d2 ) // ... / / E q +2
0,q

=

..
.

=

∞
E0,q
i∗
=

D0,q H q ( E)
_

) 
Hq ( E)

Furthermore, this diagram commutes.

(c)

Theorem 10.3.4. The image of the Hurewicz map hnB : πn ( B) → Hn ( B)

n , which is called the group of transgression elements.
is contained in En,0
240 algebraic topology

Furthermore, the following diagram commutes:

hnB
πn ( B) / Hn ( B) = E2 ⊇ . . . ⊇ n
En,0
n,0

l.e.s. ∂ dn
 hnF−1 
π n −1 ( F ) / Hn−1 ( F ) = E2 // ... / / En
0,n−1 0,n−1

10.4 Hurewicz Theorem, continued

Under the assumptions of the Hurewicz theorem, consider the follow-

ing transgression diagram of Theorem 10.3.4:

hnX
πn ( X ) / Hn ( X ) = E2 = . . . = En
n,0 n,0

∼
= ∂ ∼
= dn
 ∼

πn−1 (ΩX )
= / Hn−1 (ΩX ) = E2 = . . . = E n
−1 0,n−1 0,n−1
hnΩX

−1
The Hurewicz homomorphism hnΩX is an isomorphism by the inductive
hypothesis, ∂ is an isomorphism by the homotopy long exact sequence
associated to the path fibration for X, and dn is an isomorphism by the
spectral sequence argument used in the proof of the Hurewicz theorem.
Therefore, hnX : πn ( X ) → Hn ( X ) is an isomorphism since the diagram
commutes.
Remark 10.4.1. It can also be shown inductively that under the assump-
tions of the Hurewicz theorem,

hnX+1 : πn+1 ( X ) −→ Hn+1 ( X )

is an epimorphism.

In what follows we give more general versions of the Hurewicz

theorem. Recall that even if X is a finite CW-complex the homotopy
groups πi ( X ) are not necessarily finitely generated. However, we have
the following result:
Theorem 10.4.2 (Serre). If X is a finite CW-complex with π1 ( X ) = 0 (or
more generally if X is abelian), then the homotopy groups πi ( X ) are finitely
generated abelian groups for i ≥ 2.

Definition 10.4.3. Let C be a category of abelian groups which is closed under

extension, i.e., whenever

0 /A /B /C /0

is a short exact sequence of abelian groups with two of A, B, C contained in C ,

then so is the third. A homomorphism φ : A → B is called a
 spectral sequences. applications 241

• monomorphism mod C if ker φ ∈ C ;

• epimorphism mod C if coker φ ∈ C ;

• isomorphism mod C if ker φ, coker φ ∈ C .

Example 10.4.4. Natural examples of categories C as above include
{finite abelian groups}, {finitely generated abelian groups}, as well as
{ p-groups}.
We then have the following:
Theorem 10.4.5 (Hurewicz mod C ). Given n ≥ 2, if πi ( X ) ∈ C for
e i ( X ) ∈ C for i ≤ n − 1, hn : πn ( X ) → Hn ( X ) is an
1 ≤ i ≤ n − 1, then H X
isomorphism mod C , and hnX+1 : πn+1 ( X ) → Hn+1 ( X ) is an epimorphism
mod C .

We need the following easy fact which guarantees that in the Leray-
Serre spectral sequence of the path fibration we have Enp,q ∈ C .

Lemma 10.4.6. If G ∈ C and X is a finite CW-complex, then Hi ( X; G ) ∈ C

for any i. More generally (even if X is not a CW complex), if A, B ∈ C , then
Tor( A, B) ∈ C .
Then the proof of Theorem 10.4.5 is the same as that of the classical
Hurewicz theorem, after replacing “∼ =” by “∼ = mod C ”, and “0” by
“C ”:
hnX
πn ( X ) / Hn ( X ) = E2 = . . . = En
n,0 n,0

∼
= ∂ ∼
= mod C dn
 ∼

πn−1 (ΩX ) n−1 / Hn−1 (ΩX ) = E0,n
= mod C 2 n
−1 = . . . = E0,n−1
hΩX

−1
Specifically, hnΩX is an isomorphism mod C by the inductive hypothesis,
∂ is an isomorphism by the long exact sequence associated to the path
fibration, and dn is an isomorphism mod C by a spectral sequence
argument similar to the one used in the proof of the Hurewicz theorem.
Therefore, hnX is an isomorphism mod C since the diagram commutes.

Proof of Serre’s Theorem 10.4.2. Let

C = {finitely generated abelian groups}.

e i ( X ) ∈ C since X is a finite CW-complex. By Theorem 10.4.5,
Then, H
we have πi ( X ) ∈ C for i ≥ 2.

As another application, we can now prove the following result:

Theorem 10.4.7. Let X and Y be any connected spaces and f : X → Y a
weak homotopy equivalence (i.e., f induces isomorphisms on homotopy groups).
Then f induces isomorphisms on (co)homology groups with any coefficients.
242 algebraic topology

Proof. By universal coefficient theorems, it suffices to show that f

induces isomorphisms on integral homology. As such, we can assume
that f is a fibration, and let F denote its fiber.
Since f is a weak homotopy equivalence, the long exact sequence
of the fibration yields that πi ( F ) = 0 for all i ≥ 0. Hence, by the
Hurewicz theorem, H e i ( F ) = 0, for all i ≥ 0. Also, H0 ( F ) = Z, since F
is connected.
Consider now the Leray-Serre spectral sequence associated to the
fibration f , with E2 -page given by (see Remark 10.1.7):

E2p,q = H p (Y, Hq ( F )) ⇛ H∗ ( X ),

where Hq ( F ) is a local coefficient system (i.e., locally constant sheaf)

on Y with stalk Hq ( F ). Since F has no homology, except in degree zero
(where H0 ( F ) = H0 ( F ) is always the trivial local system when F is
path-connected), we get:

E2p,q = 0 for q > 0,

and
E2p,0 = H p (Y ).

Therefore, all differentials in the spectral sequence vanish, so

E2 = · · · = E ∞ .

Recall now that

Hn ( X ) = Dn,0 ⊇ Dn−1,1 ⊇ · · · ⊇ 0

and E∞ p,q = D p,q /D p−1,q+1 . So if q > 0, then D p,q = D p−1,q+1 since

E∞
p,q = 0. In particular, Dn−1,1 = · · · = D0,n = D−1,n+1 = 0. Therefore,
∞ 2
Hn ( X ) = En,0 = En,0 = Hn (Y )

and, by our remarks on the Leray-Serre spectral sequence (and edge

homomorphism), the above composition of isomorphisms coincides
with f ∗ , thus proving the claim.

10.5 Gysin and Wang sequences

As another application of the Leray-Serre spectral sequence, we discuss

the Gysin and Wang sequences.
π
Theorem 10.5.1 (Gysin sequence). Let F ,→ E → B be a fibration, and
suppose that F is a homology n-sphere. Assume that π1 ( B) acts trivially on
Hn ( F ), e.g., π1 ( B) = 0. Then there exists an exact sequence
π π
· · · → Hi ( E) →∗ Hi ( B) → Hi−n−1 ( B) → Hi−1 ( E) →∗ Hi−1 ( B) → · · ·
 spectral sequences. applications 243

Proof. The Leray-Serre spectral sequence of the fibration has


 H ( B) , q = 0, n
p
E2p,q = H p ( B; Hq ( F )) =
0 , otherwise.

H∗ ( B)
n

0 E 2 = · · · = E n +1

0
H∗ ( B)

Thus the only possibly nonzero differentials are:

dn+1 : Enp,0
+1 +1
−→ Enp− n−1,n .

In particular,
+1
Enp,q = · · · = E2p,q
for any ( p, q), and

0

 , q ̸= 0, n
∞
E p,q = ker(dn+1 : Enp,0 +1
→ Enp−+1
n−1,n ) ,q = 0 (10.5.1)

+1
coker(dn+1 : Enp+ n +1

n+1,0 → E p,n ) , q = n.


The above calculations yield the exact sequences

d n +1
0 −→ E∞ n +1 n +1 ∞
p,0 −→ E p,0 −→ E p−n−1,n −→ E p−n−1,n −→ 0.

The filtration on Hi ( E) reduces to

0 ⊂ Ei∞−n,n = Di−n,n ⊂ Di,0 = Hi ( E)
and so the sequences
0 −→ Ei∞−n,n −→ Hi ( E) −→ Ei,0
∞
−→ 0 (10.5.2)
are exact for each i.
The desired exact sequence follows by combining (10.5.1), (10.5.2)
and the edge isomorphism (10.3.1).

Theorem 10.5.2 (Wang). If F ,→ E → Sn is a fibration, then there is an

exact sequence:
· · · −→ Hi ( F ) −→ Hi ( E) −→ Hi−n ( F ) −→ Hi−1 ( F ) −→ · · ·
Proof. Exercise.
244 algebraic topology

10.6 Suspension Theorem for Homotopy Groups of Spheres

We first need to compute the homology of the loop space ΩSn for n > 1.

Proposition 10.6.1. If n > 1, we have:


Z , ∗ = a ( n − 1), a ∈ N
H∗ (ΩSn ) =
0 , otherwise

Proof. Consider the Leray-Serre spectral sequence for the path fibration
(with π1 (Sn ) = π0 (ΩSn ) = 0)

ΩSn ,→ PSn ≃ ∗ → Sn ,

with E2 -page

 H (ΩSn ) , p = 0, n
q
E2p,q = H p (Sn ; Hq (ΩSn )) =
0 , otherwise

which converges to H∗ ( PSn ) = H∗ ( point). In particular, E∞

p,q = 0 for
all ( p, q) ̸= (0, 0).

H∗ (ΩSn )
E2 = · · · = E n

Hi (ΩSn ) Hi (ΩSn )

.. ..
. 0 .

H1 (ΩSn ) H1 (ΩSn )

... H∗ (Sn )
0 n

First note that we have H0 (ΩSn ) = Z since π0 (ΩSn ) = π1 (Sn ) = 0.

Moreover, Hi (ΩSn ) = E0,i2 = E3 = E∞ = 0 for 0 < i < n − 1, since
0,i 0,i
these entries are not affected by any differential. Furthermore, d2 =
d3 = . . . = dn−1 = 0 since these differential are too short to alter any of
the entries they act on. So

E2 = . . . = E n .

Similarly, we have dn+1 = dn+2 = . . . = 0, as these differentials are too

long, and so
E n +1 = E n +2 = . . . = E ∞ .
 spectral sequences. applications 245

Since E∞ n
p,q = 0 for all ( p, q ) ̸ = (0, 0), all nonzero entries in E (except at
the origin) have to be killed in En+1 . In particular,

dnn,q : En,q
n n
−→ E0,q + n −1

are isomorphisms.

H∗ (ΩSn )
E2 = · · · = E n

H2n−2 (ΩSn )

..
. dn

Hn (ΩSn ) 0
Hn−1 (ΩSn ) Hn−1 (ΩSn )
0 0
.. dn ..
. .
0 0

0Z ... n Z = H0 (ΩSn )

For instance, dn : Z = H0 (ΩSn ) = En,0

n −→ En
0,n−1 = Hn−1 ( ΩS )
n

is an isomorphism, hence Hn−1 (ΩS ) = Z. More generally, we get

n

isomorphisms
Hq (ΩSn ) ∼
= Hq+n−1 (ΩSn )

for any q ≥ 0. Since H0 (ΩSn ) ∼

= Z and Hi (ΩSn ) = 0 for 0 < i < n − 1,
this gives:

Z , ∗ = a ( n − 1), a ∈ N
H∗ (ΩSn ) =
0 , otherwise

as desired.

We can now give a new proof of the Suspension Theorem for homo-
topy groups.

Theorem 10.6.2. If n ≥ 3, there are isomorphisms πi (Sn−1 ) ∼

= π i +1 ( S n ),
for i ≤ 2n − 4, and we have an exact sequence:

Z → π2n−3 (Sn−1 ) → π2n−2 (Sn ) → 0.

Proof. We have Z ∼ = πn (Sn ) ∼

= πn−1 (ΩSn ). Let g : Sn−1 → ΩSn be a
generator of πn−1 (ΩS ). First, we claim that
n

g∗ is an isomorphism on Hi (−) for all i < 2n − 2.

246 algebraic topology

This is clear if i = 0, since ΩSn is connected. Given our calculation

for Hi (ΩSn ) in Proposition 10.6.1, it suffices to prove the claim for
i = n − 1. We have a commutative diagram:

g∗ : Hn−1 (Sn−1 ) - Hn−1 (ΩSn )

6 6
h ⟳ h

g∗
π n −1 ( S n −1 ) - πn−1 (ΩSn )

[id] 7→ [ g ◦ id] = [ g]

where h is the Hurewicz map. The bottom arrow g∗ is an isomorphism

on πn−1 by our choice of g. The two vertical arrows are isomorphisms
by the Hurewicz theorem (recall that n ≥ 3, so both Sn−1 and ΩSn are
simply-connected). By the commutativity of the diagram we get the
isomorphism on the top horizontal arrow, thus proving the claim.
Since we deal only with homotopy and homology groups, we can
moreover assume that g is an inclusion. Then the homology long exact
sequence for the pair (ΩSn , Sn−1 ) reads as:
g∗
· · · → Hi (Sn−1 ) −→ Hi (ΩSn ) → Hi (ΩSn , Sn−1 ) →
g∗
→ Hi−1 (Sn−1 ) −→ Hi−1 (ΩSn ) → · · ·

From the above claim, we obtain that Hi (ΩSn , Sn−1 ) = 0, for i < 2n − 2,
together with the exact sequence
∼
=
→ H2n−2 (ΩSn , Sn−1 ) → 0
0 → Z = H2n−2 (ΩSn ) −

Since Sn−1 is simply-connected (as n − 1 ≥ 2), by the relative Hurewicz

theorem, we get that πi (ΩSn , Sn−1 ) = 0 for i < 2n − 2, and

π2n−2 (ΩSn , Sn−1 ) ∼

= H2n−2 (ΩSn , Sn−1 ) ∼
= Z.

From the homotopy long exact sequence of the pair (ΩSn , Sn−1 ), we
then get πi (ΩSn ) ∼
= πi (Sn−1 ) for i < 2n − 3 and the exact sequence

· · · → Z → π2n−3 (Sn−1 ) → π2n−3 (ΩSn ) → 0

Finally, using the fact that πi (ΩSn ) ∼

= πi+1 (Sn ), we get the desired
result.

By taking i = 4 and n = 4, we get the first isomorphism in the

following:

Corollary 10.6.3. π4 (S3 ) ∼

= π5 ( S4 ) ∼
= ... ∼
= π n +1 ( S n )
 spectral sequences. applications 247

10.7 Cohomology Spectral Sequences

Let us now turn our attention to spectral sequences computing coho-

mology. In the case of a fibration, we have the following Leray-Serre
cohomology spectral sequence:
Theorem 10.7.1 (Serre). Let F ,→ E → B be a fibration, with π1 ( B) = 0
(or π1 ( B) acting trivially on fiber cohomology) and π0 ( F ) = 0. Then there
exists a cohomology spectral sequence with E2 -page
p,q
E2 = H p ( B, H q ( F ))

converging to H ∗ ( E). This means that, for each n, H n ( E) admits a filtration

H n ( E) = D0,n ⊇ D1,n−1 ⊇ . . . ⊇ D n,0 ⊇ D n+1,−1 = 0

so that
p,q
E∞ = D ⧸D p+1,q−1 .
p,q

p,q p,q p+r,q−r +1

Moreover, the differential dr : Er → Er satisfies (dr )2 = 0, and
Er+1 = H ∗ ( Er , dr ).

n-th diagonal of E∞
H n ( E)/D1,n−1

D1,n−1 /D2,n−2

D n−1,1 /D n,0

D n,0

The corresponding statements analogous to those of Remarks 10.1.3

and 10.1.5 also apply to the spectral sequence of Theorem 10.7.1.
The Leray-Serre cohomology spectral sequence comes endowed with
the structure of a product on each page Er , which is induced from a
product on E2 , i.e., there is a map
p,q p′ ,q′ p+ p′ ,q+q′
• : Er × Er −→ Er

satisfying the Leibnitz condition

dr ( x • y) = dr ( x ) • y + (−1)deg( x) x • dr (y)
248 algebraic topology

where deg( x ) = p + q. On the E2 -page this product is the cup product

induced from
′ ′ ′ ′
H p ( B, H q ( F )) × H p ( B, H q ( F )) −→ H p+ p ( B, H q+q ( F ))
m·γ×n·ν 7→ (m ∪ n) · (γ ∪ ν)
′ ′
with m ∈ H q ( F ), n ∈ H q ( F ), γ ∈ C p ( B) and ν ∈ C p ( B), so that
′ ′
m ∪ n ∈ H q+q ( F ) and γ ∪ ν ∈ C p+ p ( B).
As it is the case for homology, the cohomology Leray-Serre spectral
sequence satisfies the following property:
i π
Theorem 10.7.2. Given a fibration F ,→ E → B with F connected and
π1 ( B) = 0 (or π1 ( B) acts trivially on the fiber cohomology), the compositions
q,0 q,0 q,0 q,0 q,0
H q ( B) = E2 ↠ E3 ↠ · · · ↠ Eq ↠ Eq+1 = E∞ ⊂ H q ( E) (10.7.1)

and
0,q 0,q 0,q 0,q
H q ( E) ↠ E∞ = Eq+1 ⊂ Eq ⊂ · · · ⊂ E2 = H q ( F ) (10.7.2)

are the homomorphisms π ∗ : H q ( B) → H q ( E) and i∗ : H q ( E) → H q ( F ),

respectively.

Recall that for a space of finite type, the (co)homology groups are
finitely generated. By using the universal coefficient theorem in coho-
mology, we have the following useful result:
Proposition 10.7.3. Suppose that F ,→ E → B is a fibration with F connected
and assume that π1 ( B) = 0 (or π1 ( B) acts trivially on the fiber cohomology).
If B and F are spaces of finite type (e.g., finite CW complexes), then for a field
K of coefficients we have:
p,q
E2 = H p ( B; K) ⊗K H q ( F; K).

Sufficient conditions for the cohomology of the total space of a

fibration to be the tensor product of the cohomology of the fiber and
that of the base space are given by the following result.
i π
Theorem 10.7.4 (Leray-Hirsch). Suppose F ,→ E − → B is a fibration, with
B and F of finite type, π1 ( B) = 0 and π0 ( F ) = 0, and let K be a field of
coefficients. Assume that i∗ : H ∗ ( E; K) → H ∗ ( F; K) is onto. Then

H ∗ ( E; K) ∼
= H ∗ ( B; K) ⊗K H ∗ ( F; K).

Proof. Consider the Leray-Serre cohomology spectral sequence

p,q
E2 = H p ( B; H q ( F; K)) ⇛ H ∗ ( E; K)

of the fibration F ,→ E → B. By Proposition 10.7.3, we have:

p,q
E2 = H p ( B; K) ⊗K H q ( F; K).
 spectral sequences. applications 249

In order to prove the theorem, it suffices to show that

E2 = · · · = E∞ ,

i.e., that all differentials d2 , d3 , etc., vanish. Indeed, since we work with
field coefficients, all extension problems encountered in passing from
E∞ to H ∗ ( E; K) are trivial, i.e.,
p,q
H n ( E; K) ∼
M
= E∞ .
p+q=n

Recall from Theorem 10.7.2 that the composite

0,q 0,q 0,q 0,q
H q ( E; K) ↠ E∞ = Eq+1 ⊂ Eq ⊂ · · · ⊂ E2 = H q ( F; K)

is the homomorphism i∗ : H q ( E; K) → H q ( F; K). Since i∗ is assumed

onto, all these inclusions must be equalities. So all dr , when restricted
to the q-axis, must vanish. On the other hand, at E2 we have
p,q p,0 0,q
E2 = E2 ⊗ E2 (10.7.3)

p,0
since K is a field, and d2 is already zero on E2 since we work with a
first quadrant spectral sequence. Since d2 is a derivation with respect
to (10.7.3), we conclude that d2 = 0 and E3 = E2 . The same argument
applies to d3 and, continuing in this fashion, we see that the spectral
sequence collapses (degenerates) at E2 , as desired.

10.8 Elementary computations

Example 10.8.1. As a first example of the use of the Leray-Serre co-

homology spectral sequence, we compute here the cohomology ring
H ∗ (CP∞ ) of CP∞ .
Consider the fibration

S1 ,→ S∞ ≃ ∗ → CP∞ .

The E2 -page of the associated Leray-Serre cohomology spectral se-

quence starts with:

H ∗ ( S1 )
E2

Z
d2

H ∗ (CP∞ )
Z 0 Z
250 algebraic topology

Here, H 1 (CP∞ ) = E21,0 = 0 since it is not affected by any differential dr ,

and the E∞ -page has only zero entries except at the origin. Moreover,
since the cohomology of the fiber is torsion-free, we get by the universal
coefficient theorem in cohomology that

E2 = H p (CP∞ , H q (S1 )) = H p (CP∞ ) ⊗ H q (S1 ).

p,q

In particular, we have E21,1 = 0 and E20,1 = H 1 (S1 ) = Z.

Since S∞ has no positive cohomology, hence the E∞ -page has only
zero entries except at the origin, it is easy to see that d2 : E20,1 → E22,0 has
to be an isomorphism, since these entries are not affected by any other
differential. Hence we have H 2 (CP∞ ) = E22,0 ∼ = Z. Since all entries on
the E2 -page are concentrated at q = 0 and q = 1, the only differential
which can affect these entries is d2 . A similar argument then shows
p,1 p+2,0
that d2 : E2 → E2 is an isomorphism for any p ≥ 0. This yields
that H even (CP∞ ) = Z and H odd (CP∞ ) = 0.
Let Z = ⟨ x ⟩ = H 1 (S1 ). Let y = d2 ( x ) be a generator of H 2 (CP∞ ).

H ∗ ( S1 )
E2

x 0 xy
d2 d2

H ∗ (CP∞ )
1 0 y 0 y2

Then, after noting that xy = (1 ⊗ x )(y ⊗ 1) is a generator of Z = E22,1 ,

we have:

d2 ( xy) = d2 ( x )y + (−1)deg( x) xd2 (y) = y2 ,

Therefore, H 4 (CP∞ ) = Z = ⟨y2 ⟩, since the d2 that hits y2 is an iso-

morphism. By induction, we get that d2 ( xyn−1 ) = yn is a generator of
H 2n (CP∞ ). Altogether, H ∗ (CP∞ ) ∼
= Z[y], with deg(y) = 2.
Example 10.8.2 (Cohomology groups of lens spaces). In this example
we compute the cohomology groups of lens spaces. Let us first recall
the relevant definitions.
Assume n ≥ 1. Consider the scaling action of C∗ on Cn+1 \{0}, and
the induced S1 -action on S2n+1 . By identifying Z/r with the group
of r th roots of unity in C∗ , we get (by restriction) an action of Z/r on
S2n+1 . The quotient
2n+1
L(n, r ) := S ⧸Z/r

is called a lens space.

 spectral sequences. applications 251

The action of Z/r on S2n+1 is clearly free, so the quotient map

S2n+1 → L(n, r ) is a covering map with deck group Z/r. Since S2n+1
is simply-connected, it is the universal cover of L(n, r ). This yields that
π1 ( L(n, r )) = Z/r and all higher homotopy groups of L(n, r ) agree
with those of the sphere S2n+1 .
By a telescoping construction, which amounts to letting n → ∞, we
∞
get a covering map S∞ → L(∞, r ) := S ⧸Z/r with contractible total
space. In particular,
L(∞, r ) = K (Z/r, 1).
To compute the cohomology of L(n, r ), one may be tempted to
use the Leray-Serre spectral sequence for the covering map Z/r ,→
S2n+1 → L(n, r ). However, since L(n, r ) is not simply-connected, com-
putations may be tedious. Instead, we consider the fibration

S1 ,→ L(n, r ) → CPn (10.8.1)

whose base space is simply-connected. This fibration is obtained by

noting that the action of S1 on S2n+1 descends to an action of S1 =
S1 /(Z/r ) on L(n, r ), with orbit space CPn .
Consider now the Leray-Serre cohomology spectral sequence for the
fibration (10.8.1):
p,q
E2 = H p (CPn , H q (S1 ; Z)) ⇛ H p+q ( L(n, r ); Z)
p,q
and note that E2 = 0 for q ̸= 0, 1. This implies that all differentials d3
and higher vanish, so
E3 = · · · = E∞ .
On the E2 -page, we have by the universal coefficient theorem in coho-
mology that:
p,q
E2 = H p (CPn ; Z) ⊗ H q (S1 ; Z).
∼ H 1 (S1 ; Z), and let x be a generator
Let a be a generator of Z = E20,1 =
of Z = E22,0 ∼
= H 2 (CPn ; Z). We claim that

d2 ( a) = rx. (10.8.2)

H ∗ ( S1 )
E2

a 0 ax 0 ax2 ax n
d2 d2

... H ∗ (CPn )
1 0 x 0 x2 xn
252 algebraic topology

To find d2 , it suffices to compute H 2 ( L(n, r ); Z). Indeed, by looking

at the entries of the second diagonal of E∞ = · · · = E3 , we have:
0,2 1,1
0,2
H 2 ( L(n, r ); Z) = D0,2 , E∞ = D ⧸D1,1 = 0, E∞ 1,1
= D ⧸D2,0 = 0, and
2,0
E∞ = D2,0 = Z⧸Im(d ). In particular,
2

H 2 ( L(n, r ); Z) = D0,2 = D1,1 = D2,0 = Z⧸Im(d ). (10.8.3)

2

On the other hand, since H1 ( L(n, r ); Z) = π1 ( L(n, r )) = Z/r, we get

by the universal coefficient theorem that

H 2 ( L(n, r ); Z) = (free part) ⊕ Z/r. (10.8.4)

By comparing (10.8.3) and (10.8.4), we conclude that d2 ( a) = rx and

H 2 ( L(n, r ); Z) = Z/r.
By using the Künneth formula and the ring structure of H ∗ (CPn ; Z),
it follows from the Leibnitz formula and induction that d2 ( ax k−1 ) = rx k
for 1 ≤ k ≤ n, and we also have d2 ( ax n ) = 0. In particular, all the
nontrivial differentials labelled by d2 are given by multiplication by r.
Since multiplication by r is injective, the E3 = · · · = E∞ -page is given
by
E∞

0 0 0 0 0 Z

...
Z 0 Z/r 0 Z/r Z/r

The extension problems for going from E∞ to the cohomology of the

total space L(n, r ) are in this case trivial, since every diagonal of E∞
contains at most one nontrivial entry. We conclude that



 Z i=0

Z/r i = 2, 4, · · · , 2n

H i ( L(n, r ); Z) =


 Z i = 2n + 1


0 otherwise.

By letting n → ∞, we obtain similarly that


Z i=0


i
H (K (Z/r, 1); Z) = Z/r i = 2k, k ≥ 1


0 otherwise.


In particular, if r = 2, this computes the cohomology of RP∞ .

 spectral sequences. applications 253

10.9 Computation of πn+1 (Sn )

In this section we prove the following result:

Theorem 10.9.1. If n ≥ 3,

πn+1 (Sn ) = Z/2.

Theorem 10.9.1 follows from the Suspension Theorem (see Corollary

10.6.3), together with the following explicit calculation:
Theorem 10.9.2.
π4 (S3 ) = Z/2.
The proof of Theorem 10.9.2 given here uses the Postnikov tower
approximation of S3 , whose construction we recall here. (A different
proof of this fact will be given in the next section, by using Whitehead
towers.)
Lemma 10.9.3 (Postnikov approximation). Let X be a CW complex with
πk := πk ( X ). For any n, there is a sequence of fibrations

K (πk , k) ,→ Yk → Yk−1

and maps X → Yk with a commuting diagram

Y1 ko Y2 oi · · · Yn−1 o YO n
c

such that X → Yk induces isomorphisms πi ( X ) ∼

= πi (Yk ) for i ≤ k, and
πi (Yk ) = 0 for i > k.

Proof. To construct Yn we kill off the homotopy groups of X in degrees

≥ n + 1 by attaching cells of dimension ≥ n + 2. We then have πi (Yn ) =
πi ( X ) for i ≤ n and πi (Yn ) = 0 if i > n. Having constructed Yn , the
space Yn−1 is obtained from Yn by killing the homotopy groups of Yn
in degrees ≥ n, which is done by attaching cells of dimension ≥ n + 1.
Repeating this procedure, we get inclusions

X ⊂ Yn ⊂ Yn−1 ⊂ · · · ⊂ Y1 = K (π1 , 1),

which we convert to fibrations. From the homotopy long exact sequence

for each of these fibrations, we see that the fiber of Yk → Yk−1 is a
K (πk , k)-space.

Proof of Theorem 10.9.2. We consider the Postnikov tower construction

in the case n = 4, X = S3 , to obtain a fibration

K (π4 , 4) ,→ Y4 → Y3 = K (Z, 3), (10.9.1)

254 algebraic topology

where π4 = π4 (S3 ) = π4 (Y4 ). Here, Y3 = K (Z, 3) since to get Y3 we

kill off all higher homotopy groups of S3 starting at π4 . Since Y 4 is
obtained from S3 by attaching cells of dimension ≥ 6, it doesn’t have
cells of dimensions 4 and 5, thus

H4 (Y4 ) = H5 (Y4 ) = 0.

Let us now consider the homology spectral sequence for the fibration
(10.9.1). By the Hurewicz theorem,

0 p = 1, 2
H p (K (Z, 3); Z) =
Z p = 3

0 q = 1, 2, 3
Hq (K (π4 , 4); Z) =
 π4 ( S3 ) q = 4.

So the E2 -page looks like

H∗ (K (π4 , 4))

π4

d5
0

H∗ (K (Z, 3)
Z 0 0 Z H4 H5

Since H4 (Y4 ) = 0 = H5 (Y4 ), all entries on the fourth and fifth diagonals
of E∞ are zero. The only differential that can affect π4 (S3 ) = E0,4 2 =
5
· · · = E0,4 is
d5 : H5 (K (Z, 3), Z) −→ π4 (S3 ),
and by the previous remark, this map has to be an isomorphism (note
2 = H ( K (Z, 3), Z) can be affected only by d5 , and this
also that E5,0 5
element too has to be killed at E∞ ). Hence

π4 ( S3 ) ∼
= H5 (K (Z, 3), Z). (10.9.2)

In order to compute H5 (K (Z, 3), Z), we use the cohomology Leray-

Serre spectral sequence associated to the path fibration for K (Z, 3),
namely
ΩK (Z, 3) ,→ PK (Z, 3) → K (Z, 3),
 spectral sequences. applications 255

and note that, since PK (Z, 3) is contractible, we have πi (ΩK (Z, 3)) ∼ =
πi+1 (K (Z, 3)), i.e., ΩK (Z, 3) ≃ K (Z, 2) = CP∞ . Since each H j (CP∞ ) is
a finitely generated free abelian group, the universal coefficient theorem
yields that

E2 = H p (K (Z, 3); H q (CP∞ )) ∼

= H p (K (Z, 3)) ⊗ H q (CP∞ ),
p,q
(10.9.3)

and the product structure on E2 is that of the tensor product of

H ∗ (K (Z, 3)) and H ∗ (CP∞ ).
p,q
Since E2 = 0 for q odd, we have d2 = 0, so E2 = E3 . Similarly, all
the even differentials d2n are zero, so E2n = E2n+1 , for all n ≥ 1. Since
p,q
the total space of the fibration is contractible, we have that E∞ = 0 for
all ( p, q) ̸= (0, 0), so every non-zero entry on the E2 -page (except at the
origin) must be killed on subsequent pages.
Let a ∈ H 2 (CP∞ ) ∼ = Z be a generator. So ak is a generator of
H 2k (CP∞ ) = E2 , for any k ≥ 1. We create elements on E2∗,0 , which
0,2k

will sooner or later kill off all the non-zero elements in the spectral
sequence.

H ∗ (CP∞ )

5 0 E2 = E3

4 a2

d3
3 0
·2

2 a as

d3 d3
1 0 ∼
=

1 0 0 s 0 0 y = s2
0 H ∗ (K (Z, 3))
0 1 2 3 4 5 6

Note that E31,0 = E21,0 = H 1 (K (Z, 3)) is never touched by any differen-
tial, so
H 1 (K (Z, 3)) = E∞ 1,0
= 0.
Moreover, since d2 = 0, we also have that

H 2 (K (Z, 3)) = E22,0 = E32,0 = E∞

2,0
= 0.

The only differential that can affect ⟨ a⟩ = E20,2 = E30,2 is d0,2 0,2 3,0
3 : E3 → E3 ,
3,0
so there must be an element s ∈ E3 that kills off a, i.e., d3 ( a) = s. On
the other hand, since E33,0 is only affected by d3 and it must be killed
256 algebraic topology

off at infinity, we must have that d0,2 0,2 3,0

3 : E3 → E3 is an isomorphism,
so s generates
Z = E33,0 = E23,0 = H 3 (K (Z, 3)).

By (10.9.3), we also have that E33,2 = E23,2 = Z, generated by as. Note

that
d3 ( a2 ) = 2ad3 ( a) = 2as,

so d0,4
3 : E3
0,4
→ E33,2 is given by multiplication by 2. In particular,
E4 = 0. Next notice that H 4 (K (Z, 3)) = E34,0 and H 5 (K (Z, 3)) = E35,0
0,4

can only be touched by the differentials d3 , d4 , or d5 , but all of these are

trivial maps because their domains are zero. Thus, as H 4 (K (Z, 3)) and
H 5 (K (Z, 3)) can not killed by any differential, we have

H 4 (K (Z, 3)) = H 5 (K (Z, 3)) = 0.

Similarly, H 6 (K (Z, 3)) = E36,0 and ⟨ as⟩ = E33,2 are only affected by
d3 . Since d3 ( a2 ) = 2as, we have ker(d3 : ⟨ as⟩ = E33,2 → E36,0 ) =
Im(d3 : E30,4 → E33,2 = ⟨ as⟩) = ⟨2as⟩ ⊆ ⟨ as⟩, and hence H 6 (K (Z, 3)) =
Im(d3 : E33,2 → E36,0 ) ∼
= ⟨ as⟩ / ⟨2as⟩ = Z/2.
In view of the above calculations, we get by the universal coefficient
theorem that
H5 (K (Z, 3)) = Z/2. (10.9.4)

The assertion of the theorem then follows by combining (10.9.2) and

(10.9.4).

Corollary 10.9.4.
π4 (S2 ) = Z/2.

Proof. This follows from Theorem 10.9.2 and the long exact sequence
of homotopy groups for the Hopf fibration S1 ,→ S3 → S2 .

10.10 Whitehead tower approximation and π5 (S3 )

In order to compute π5 (S3 ) we make use of the Whitehead tower

approximation. We recall here the construction.

Whitehead tower
Let X be a connected CW complex, with πq = πq ( X ) for any q ≥ 0.

Definition 10.10.1. A Whitehead tower of X is a sequence of fibrations

· · · −→ Xn −→ Xn−1 −→ · · · → X0 = X

such that

(a) Xn is n-connected
 spectral sequences. applications 257

(b) πq ( Xn ) = πq ( X ) for q ≥ n + 1

(c) the fiber of Xn → Xn−1 is a K (πn , n − 1)-space.

Lemma 10.10.2. For X a CW complex, Whitehead towers exist.

Proof. We construct Xn inductively. Suppose that Xn−1 has already

been defined. Add cells to Xn−1 to kill off πq ( Xn−1 ) for q ≥ n + 1.
So we get a space Y which, by construction, is a K (πn , n)-space. Now
define the space
Xn := P∗ Xn−1 := { f : I → Y, f (0) = ∗, f (1) ∈ Xn−1 }
consisting of of paths in Y beginning at a basepoint ∗ ∈ Xn−1 and
ending somewhere in Xn−1 . Endow Xn with the compact-open topology.
As in the case of the path fibration, the map π : Xn → Xn−1 defined by
γ → γ(1) is a fibration with fiber ΩY = K (πn , n − 1).
From the long exact sequence of homotopy groups associated to the
fibration
K (πn , n − 1) ,→ Xn → Xn−1
we get that πq ( Xn ) = πq ( Xn−1 ) for q ≥ n + 1, and πq ( Xn ) = 0 for
q ≤ n − 2. Furthermore, the sequence
0 −→ πn ( Xn ) −→ πn ( Xn−1 ) −→ πn−1 (K (πn , n − 1)) −→ πn−1 ( Xn ) −→ 0

is exact. So we are done if we show that the boundary homomor-

phism ∂ : πn ( Xn−1 ) −→ πn−1 (K (πn , n − 1)) of the long exact sequence
is an isomorphism. For this, note that the inclusion Xn−1 ⊂ Y =
K (πn , n) = Xn−1 ∪ {cells of dimension ≥ n + 2} induces an isomor-
phism πn ( Xn−1 ) ∼= πn K (πn , n) ∼
= πn−1 (K (πn , n − 1)), which is pre-
cisely the above boundary map ∂.

Calculation of π4 (S3 ) and π5 (S3 )

In this section we use the Whitehead tower for X = S3 to compute
π5 ( S3 ).

Theorem 10.10.3.
π5 ( S3 ) ∼
= Z/2.
Proof. Consider the Whitehead tower for X = S3 . Since S3 is 2-
connected, we have in the notation of Definition 10.10.1 that X =
X1 = X2 . Let πi := πi (S3 ), for any i ≥ 0. We have fibrations
K ( π4 , 3) / X4


K ( π3 , 2) / X3


S3
258 algebraic topology

Since π3 = Z, we have K (π3 , 2) = CP∞ . Moreover, since X4 is 4-

connected, we get by definition and Hurewicz that

π5 ( S3 ) ∼
= π 5 ( X4 ) ∼
= H5 ( X4 ).

Similarly,
π4 ( S3 ) ∼
= π 4 ( X3 ) ∼
= H4 ( X3 ).
Once again we are reduced to computing homology groups. Using the
universal coefficient theorem, we will deduce the homology groups
from cohomology.
Consider now the cohomology spectral sequence for the fibration

CP∞ ,→ X3 → S3 .

The E2 -page is given by

E2 = H p (S3 , H q (CP∞ , Z)) = H p (S3 ) ⊗ H q (CP∞ ) ⇛ H ∗ ( X3 ).

p,q

p,q
In particular, E2 = 0 unless p = 0, 3 and q is even.

H ∗ (CP∞ )
E2 = E3

4 x2

d3
3 0
·2

2 x xu
d3
1 0 ∼
=
1 0 0 u
0 H ∗ ( S3 )
0 1 2 3
p,q
Since E2 = 0 for q odd, we have d2 = 0, so E2 = E3 . In addition, for
r ≥ 4, dr = 0. So E4 = E∞ .
Since X3 is 3-connected, we have by Hurewicz that H 2 ( X3 ) =
3
H ( X3 ) = 0, so all entries on the second and third diagonals of
E∞ = E4 are 0. This implies that d0,2 3 : E30,2 = Z → E33,0 = Z is
an isomorphism. Let H ∗ (CP∞ ) = Z[ x ] with x of degree 2, and let u be
a generator of H 3 (S3 ). Then we have d3 ( x ) = u. By the Leibnitz rule,
d3 x n = nx n−1 dx = nx n−1 u, and since x n generates E30,2n and x n−1 u
generates E33,2n−2 , the differential d0,2n
3 is given by multiplication by n.
This completely determines E4 = E∞ , hence the integral cohomology
and (by the universal coefficient theorem) homology of X3 is easily
computed as:
q 0 1 2 3 4 5 6 7 ··· 2k 2k + 1 ···
H q ( X3 ) Z 0 0 0 0 Z/2 0 Z/3 ··· 0 Z/k ···
Hq ( X3 ) Z 0 0 0 Z/2 0 Z/3 0 ··· Z/k 0 ···
 spectral sequences. applications 259

In particular, π4 = H4 ( X3 ) = Z/2, which reproves Theorem 10.9.1.

In order to compute π5 (S3 ) ∼

= H5 ( X4 ), we use the homology spectral
sequence for the fibration

K (π4 , 3) ,→ X4 → X3 ,

with E2 -page

E2p,q = H p ( X3 ; Hq (K (Z/2, 3))) ⇛ H∗ ( X4 ).

Note that, by the Hurewicz theorem, we have: Hi (K (π4 , 3)) = 0 for

i = 1, 2 and H3 (K (π4 , 3)) = π4 = Z/2. So E2p,q = 0 for q = 1, 2. Also,
E2p,0 = H p ( X3 ), whose values are computed in the above table.

H∗ (K (Z/2, 3))

5 Z/2

4 0

3 Z/2
d6

2 0 d4

1 0

Z 0 0 0 Z/2 0 Z/3
0 H∗ ( X3 )
0 1 2 3 4 5 6

Since X4 is 4-connected, we have by Hurewicz that H3 ( X4 ) = H4 ( X4 ) =

0, so all entries on the third and fourth diagonal of E∞ are zero. Since
the first and second row of E2 are zero, this forces d4 : E4,0
4 = E2 →
4,0
E0,3 = E0,3 to be an isomorphism (thus recovering the fact that π4 ∼
4 2
=
Z/2), and
2 ∞
H4 (K (Z/2, 3)) = E0,4 = E0,4 = 0.
Moreover, by a spectral sequence argument for the path fibration of
K (Z/2, 3), we obtain (see Exercise 6)
2
E0,5 = H5 (K (Z/2, 3)) = Z/2,

and this entry can only be affected by d6 : E6,0 6 ∼ Z/3 → E6 =

= 0,5
∼ ∞
E0,5 = Z/2, which is the zero map, so E0,5 = Z/2. Thus, on the fifth
2

diagonal of E∞ , all entries are zero except E0,5

∞ = Z/2, which yields

H5 ( X4 ) = Z/2, i.e., π5 (S ) = Z/2.

3
 260 algebraic topology

10.11 Serre’s theorem on finiteness of homotopy groups of spheres

In this section we prove the following result:

Theorem 10.11.1 (Serre).

(a) πi (S2k+1 ) is finite for i > 2k + 1.

(b) πi (S2k ) is finite for i > 2k, i ̸= 4k − 1, and

π4k−1 (S2k ) = Z ⊕ {finite abelian group}.

Proof of part (a). The case k = 0 is easy since πi (S1 ) is in fact trivial
for i > 1. For k > 0, recall Serre’s theorem 10.4.2, according to which
a simply-connected finite CW complex has finitely generated homo-
topy groups. In particular, the groups πi (S2k+1 ) are finitely generated
abelian for all i > 1. Therefore, πi (S2k+1 ) (i > 1) is finite if it is a torsion
group.
In what follows we show that

πi (S2k−1 ) ∼
= πi+2 (S2k+1 ) mod torsion, (10.11.1)

and part (a) of the theorem follows then by induction. The key to
proving the isomorphism (10.11.1) is the fact that

π2k−1 (Ω2 S2k+1 ) ∼

= π2k+1 (S2k+1 ) = Z.

Letting β : S2k−1 → Ω2 S2k+1 be a generator of π2k−1 (Ω2 S2k+1 ), we

will show that β induces an isomorphism mod torsion on H∗ (i.e., an
isomorphism on H∗ (−; Q)). Let us assume this fact for now. WLOG,
we assume that β is an inclusion, and then the homology long exact
sequence of the pair (Ω2 S2k+1 , S2k−1 ) yields that

H∗ (Ω2 S2k+1 , S2k−1 ) = 0 mod torsion.

The relative version of the Hurewicz mod torsion Theorem 10.4.5 then
tells us that
πi (Ω2 S2k+1 , S2k−1 ) = 0 mod torsion
for all i, so again by the homotopy long exact sequence of the pair
we get that πi (S2k−1 ) ∼
= πi (Ω2 S2k+1 ) ∼
= πi+2 (S2k+1 ) mod torsion, as
desired.
Thus, it remains to show that the generator β : S2k−1 → Ω2 S2k+1 of
π2k−1 (Ω2 S2k+1 ) induces an isomorphism on H∗ (−; Q). The bulk of the
argument amounts to showing that Hi (Ω2 S2k+1 ; Q) = 0 for i ̸= 2k − 1,
which we do by computing Hi (Ω2 S2k+1 ; Q)∨ = H i (Ω2 S2k+1 ; Q) with
the help of the cohomology spectral sequence for the path fibration
Ω2 S2k+1 ,→ ∗ → ΩS2k+1 . The E2 -page is given by
p,q
E2 = H p (ΩS2k+1 ; H q (Ω2 S2k+1 ; Q)) ⇛ H ∗ (∗; Q),
 spectral sequences. applications 261

p,q
and since the total space of the fibration is contractible, we have E∞ = 0
0,0 ∼
unless p = q = 0, in which case E∞ = Q.
It is a simple exercise (using the path fibration ΩS2k+1 ,→ ∗ → S2k+1 )
to show that

H ∗ (ΩS2k+1 ; Q) ∼
= Q[e], deg e = 2k.

Hence,
p,q
E2 = H p (ΩS2k+1 ; H q (Ω2 S2k+1 ; Q))
∼ H p (ΩS2k+1 ; Q) ⊗Q H q (Ω2 S2k+1 ; Q)
=

2kj,0 ∼
has possibly non-trivial columns only at multiples p of 2k, with E2 =
Q = ⟨e ⟩. This implies that d2 , d3 , . . . , d2k−1 are all zero, hence E2 = E2k .
j

Furthermore, since the first non-trivial homotopy group πq (Ω2 S2k+1 ) ∼ =

πq+2 (S2k+1 ) appears at q = 2k − 1, it follows by Hurewicz that

H q (Ω2 S2k+1 ; Q) = 0, for 0 < q < 2k − 1.

p,q
Therefore, E2 = 0 for 0 < q < 2k − 1.

H ∗ (Ω2 S2k+1 ; Q)
E2 = · · · = E2k

2k − 1 ω 0 eω 0
d2k d2k
0 1 0 e 0 e2
... H ∗ (ΩS2k+1 ; Q)
0 ... 2k ... 4k

2k,0 ∼ 0,2k −1 ∼
Since E2k = H 2k (ΩS2k+1 ) = ⟨e⟩ and E2k = H 2k−1 (Ω2 S2k+1 ) are
−1 0,2k −1 −1
only affected by d0,2k
2k : E2k 2k,0
→ E2k , we must have that d0,2k2k is an
2k,0 2k,0 0,2k −1 0,2k−1
isomorphism in order for E2k+1 = E∞ and E2k+1 = E∞ to be zero.
So H 2k−1 (Ω2 S2k+1 ) ∼
= Q = ⟨ω ⟩, with d2k (ω ) = e. As a consequence,
2jk,2k −1
E2k = H 2jk (ΩS2k+1 ; Q) ⊗Q H 2k−1 (Ω2 S2k+1 ) = ⟨e j ⟩ ⊗Q ⟨ω ⟩ = ⟨e j ω ⟩

2jk,2k −1 2jk,2k −1 2jk+2k,0

and d2k : E2k → E2k are isomorphisms since d2k (e j ω ) =
j j + 1
jd2k (e)ω + e d2k (ω ) = e . This implies that, except for q ∈ {0, 2k − 1},
p,q 0,i
E2k is always trivial, and in particular that H i (Ω2 S2k+1 ; Q) = E2k is
∗
trivial for i ̸= 0, 2k − 1. (If there was anything else in H (Ω S
2 2k + 1 ; Q),
it would have to also be present at infinity.)
Next note that S2k−1 and Ω2 S2k+1 are (2k − 2)-connected, so by
the Hurewicz theorem, their rational cohomology vanishes in degrees
i < 2k − 1. Hence, β : S2k−1 → Ω2 S2k+1 induces isomorphisms on
262 algebraic topology

H i (−; Q) if i ̸= 2k − 1. In order to show that β induces an isomorphism

on H2k−1 (−; Q), recall the commutative diagram:

/ H2k−1 (Ω2 S2k+1 )

β∗
H2k−1 (S2k−1 )
O O
h ∼
= h ∼
=

π2k−1 (S2k−1 ) / π2k−1 (Ω2 S2k+1 )

β∗

where the lower horizontal β ∗ is an isomorphism since β is the gen-

erator of π2k−1 (Ω2 S2k+1 ), and the vertical arrows are isomorphisms
by Hurewicz. Since the diagram commutes, the top horizontal map
labelled β ∗ is an isomorphism also, and the proof of part (a) is com-
plete.

Proof of part (b). We shall construct a fibration

S2k−1 ,→ E −
→ S2k
π

such that
πi ( E ) ∼
= πi (S4k−1 ) (mod torsion). (10.11.2)

Assuming for now that such a fibration exists, then since by part (a) we
have that 
finite i ̸= 4k − 1
πi (S4k−1 ) = ,
Z i = 4k − 1

we deduce that

finite i ̸= 4k − 1
πi ( E ) =
Z ⊕ finite i = 4k − 1.

The homotopy long exact sequence:

··· / πi (S2k−1 ) / πi ( E ) / πi (S2k ) / πi−1 (S2k−1 ) / ···

together with that fact proved in part (a) that


finite i ̸= 2k − 1
πi (S2k−1 ) = ,
Z i = 2k − 1

then yields that


finite i ̸= 2k, 4k − 1
πi (S2k ) =
Z ⊕ finite i = 4k − 1,

as desired.
 spectral sequences. applications 263

Note that in order to have (10.11.2), it is sufficient for E to satisfy

Hi ( E) ∼
= Hi (S4k−1 ) modulo torsion, i.e.,

finite i ̸= 0, 4k − 1
Hi ( E) =
Z ⊕ finite i = 4k − 1.

Indeed, by Hurewicz mod torsion, we then have that π4k−1 ( E) ∼ =

H4k−1 ( E) mod torsion, and let f : S 4k − 1 → E be a generator of the
Z-summand of π4k−1 ( E). WLOG, we can assume that f is an inclu-
sion. The homology long exact sequence of the pair ( E, S4k−1 ) then
implies that H∗ ( E, S4k−1 ) = 0 mod torsion. By Hurewicz mod torsion
this yields π∗ ( E, S4k−1 ) = 0 mod torsion. Finally, the homotopy long
exact sequence gives πi ( E) ∼ = πi (S4k−1 ) mod torsion.

Back to the construction of the space E, we start with the tangent

bundle TS2k → S2k , and let π : T0 S2k → S2k be its restriction to the
space of nonzero tangent vectors to S2k . Then π is a fibration, since it
is locally trivial, and its fiber is R2k \ {0} ≃ S2k−1 . We let

E = T0 S2k .

Let us now consider the Leray-Serre homology spectral sequence of

this fibration, with

E2p,q = H p (S2k ; Hq (S2k−1 )) = H p (S2k ) ⊗ Hq (S2k−1 ) ⇛ H∗ ( E).

Therefore, the page E2 has only four non-trivial entries at ( p, q) = (0, 0),
(2k, 0), (0, 2k − 1), (2k − 1, 2k), and all these entries are isomorphic to
Z.
H∗ (S2k−1 )
E2 = · · · = E2k

Z Z
d2k

H∗ (S2k )
Z Z

Clearly, the differentials d2 , d3 , . . . , d2k−1 are all zero, as are the dif-
ferentials d2k+1 , . . . . The only possibly non-zero differential in the
spectral sequence is d2k 2k 2k 2 2k
2k,0 : E2k,0 → E0,2k −1 . Thus, E = · · · = E and
E2k+1 = · · · = E∞ . Therefore, the space E has the desired homology if
and only if
d2k
2k,0 ̸ = 0.
264 algebraic topology

The map d2k

2k,0 fits into a commutative diagram

π2k (S2k )
∂ / π2k−1 (S2k−1 )

h ∼
= ∼
= h
 
d2k
H2k (S2k ) / H2k−1 (S2k−1 )

where ∂ is the connecting homomorphism in the homotopy long exact

sequence of the fibration, and h denotes the Hurewicz maps. Hence,
d2k ̸= 0 if and only if ∂ ̸= 0. If, by contradiction, ∂ = 0, then the
homotopy long exact sequence of the fibration π contains the exact
sequence
∗π
π2k (S2k ) −
∂
π2k ( E) −→ → 0.
In particular, there is [ϕ] ∈ π2k ( E) so that π∗ ([ϕ]) = [id], i.e., the
diagram
=E
ϕ
π

S2k S2k
id
commutes up to homotopy. By the homotopy lifting property of the
fibration, there is then a map ψ : S2k → E so that π ◦ ψ = id. In other
words, ψ is a section of the bundle π. This implies the existence of a
nowhere-vanishing vector field on S2k , which is a contradiction.

Remark 10.11.2. Serre’s original proof of Theorem 10.11.1 used the

Whitehead tower approximation of a sphere, together with the compu-
tation of the rational cohomology of K (Z, n) (see Exercise 13).

10.12 Computing cohomology rings via spectral sequences

The following computation will be useful when discussing about char-

acteristic classes:
Example 10.12.1. In this example, we show that the cohomology ring
H ∗ (U (n); Z) is a free Z-algebra on odd degree generators x1 , · · · , x2n−1 ,
with deg( xi ) = i, i.e.,

H ∗ (U (n); Z) = ΛZ [ x1 , · · · , x2n−1 ].

We will prove this fact by induction on n, by using the Leray-Serre

cohomology spectral sequence for the fibration

U (n − 1) ,→ U (n) → S2n−1 .

For the base case, note that U (1) = S1 , so H ∗ (U (1)) = ΛZ [ x1 ] with

deg( x1 ) = 1. For the induction step, we will show that

H ∗ (U (n)) = H ∗ (S2n−1 ) ⊗ H ∗ (U (n − 1)). (10.12.1)

 spectral sequences. applications 265

Since H ∗ (S2n−1 ) = ΛZ [ x2n−1 ] with deg( x2n−1 ) = 2n − 1, this will then

give recursively that H ∗ (U (n)) = ΛZ [ x1 , . . . , x2n−3 ] ⊗Z ΛZ [ x2n−1 ] =
ΛZ [ x1 , · · · , x2n−1 ], with odd-degree generators x1 , · · · , x2n−1 , with

deg( xi ) = i.

Assume by induction that H ∗ (U (n − 1)) = ΛZ [ x1 , · · · , x2n−3 ], with

deg( xi ) = i, and for n ≥ 2 consider the cohomology spectral sequence
p,q
E2 = H p (S2n−1 , H q (U (n − 1))) ⇛ H ∗ (U (n)).

By the universal coefficient theorem, we have that

p,q
E2 = H p (S2n−1 ) ⊗ H q (U (n − 1)) = 0 if p ̸= 0, 2n − 1.

So all the nonzero entries on the E2 -page are concentrated on the

columns p = 0 (i.e., q-axis) and p = 2n − 1. In particular,

d1 = · · · = d2n−2 = 0,

so
E2 = · · · = E2n−1 .

Furthermore, higher differentials starting with d2n are also zero (since
either their domain or target is zero), so

E2n = · · · = E∞ .

Recall now that x1 , · · · , x2n−3 generate the cohomology of the fiber

U (n − 1) and note that, due to their position on E2n−1 , we have that
d2n−1 ( x1 ) = · · · = d2n−1 ( x2n−3 ) = 0. Since d2n−1 ( x2n−1 ) = 0, we
conclude by the Leibnitz rule that

d2n−1 = 0.

(Here, x2n−1 denotes the generator of H ∗ (S2n−1 ).) Thus, E2n−1 = E2n ,
so in fact the spectral sequence degenerates at the E2 -page, i.e.,

E2 = · · · = E∞ .

Since the E∞ -term is a free, graded-commutative, bigraded algebra, it

is a standard fact (e.g., see Example 1.K in McCleary’s “A User’s guide
to spectral sequences”) that the abutement H ∗ (U (n)) of the spectral
sequence is also a free, graded commutative algebra isomorphic to the
∗,∗
total complex associated to E∞ , i.e.,
p,q
H i (U (n)) ∼
M
= E∞ ,
p + q =i

as desired.
266 algebraic topology

Example 10.12.2. We can similarly compute H ∗ (SU (n)) either directly

by induction from the fibration SU (n − 1) ,→ SU (n) → S2n−1 and
the base case SU (2) = S3 , or by using our computation of H ∗ (U (n))
together with the diffeomorphism

U (n) ∼= SU (n) × S1 (10.12.2)

 
1
given by A 7→ √
n
A, det A . In particular, (10.12.2) yields by the
det A
Künneth formula:

H ∗ (U (n)) = H ∗ (SU (n)) ⊗ H ∗ (S1 ),

hence
H ∗ (SU (n)) = ΛZ [ x3 , . . . , x2n−1 ]
with deg xi = i.

10.13 Exercises

1. Show that πi (ΣRP2 ) are finitely generated abelian groups for any
i ≥ 0. (Hint: Use Theorem 10.4.5, with C the category of finitely
generated 2-groups.

2. Compute the homology of ΩS1 . (Hint: Use the fibration ΩS1 ,→

Z → R obtained by “looping” the covering Z ,→ R → S1 , together
with the Leray-Serre spectral sequence.)

3. Prove Wang’s Theorem 10.5.2.

4. Let π : E → B be a fibration with fiber F, let K be a field, and

assume that π1 ( B) acts trivially on H∗ ( F; K). Assume that the Euler
characteristics χ( B), χ( F ) are defined (e.g., if B and F are finite CW
complexes). Then χ( E) is defined and

χ ( E ) = χ ( B ) · χ ( F ).

5. Use a spectral sequence argument to show that Sm ,→ Sn → Sl is a

fiber bundle, then n = m + l and l = m + 1.

6. Prove that H5 (K (π4 , 3)) = Z/2. (Hint: consider the two fibrations
K (Z/2, 2) = ΩK (Z/2, 3) ,→ ∗ → K (Z/2, 3), and RP∞ = K (Z/2, 1) ,→
∗ → K (Z/2, 2). Then compute H∗ (K (Z/2, 2)) via the spectral sequence
of the second fibration, and use it in the spectral sequence of the first
fibration to compute H∗ (K (Z/2, 3)).)

7. Compute the cohomology of the space of continuous maps f :

S1 → S3 . (Hint: Let X := { f : S1 → S3 , f is continuous} and define
 spectral sequences. applications 267

π : X → S3 by f 7→ f (1). Then π is a fibration with fiber ΩS3 . Apply

the cohomology spectral sequence for the fibration ΩS3 ,→ X → S3 to
conclude that H ∗ ( X ) ∼
= H ∗ (S3 ) ⊗ H ∗ (ΩS3 ).)

8. Compute the cohomology of the space of continuous maps f : S1 →

S2 .

9. Compute the cohomology of the space of continuous maps f : S1 →

CPn .

10. Compute the cohomology ring H ∗ (SO(n); Z/2).

11. Compute the cohomology ring H ∗ (Vk (Cn ); Z).

12. Show that H ∗ (SO(4)) ∼

= H ∗ (S3 ) ⊗ H ∗ (RP3 ).

13. Show that


Q[ z ] , if n is even
n
H ∗ (K (Z, n); Q) =
Λ(zn ) , if n is odd,

with deg(zn ) = n. Here, Λ(zn ) := Q[zn ]/(z2n ).

(Hint: Consider the spectral sequence for the path fibration

K (Z, n − 1) ,→ ∗ → K (Z, n)

and induction.)

14. Compute the ring structure on H ∗ (ΩSn ).

15. Show that the p-torsion in πi (S3 ) appears first for i = 2p, in which
case it is Z/p. (Hint: use the Whitehead tower of S3 , the homology
spectral sequence of the relevant fibration, together with Hurewicz mod
C p , where C p is the class of torsion abelian groups whose p-primary
subgroup is trivial.)

16. Where does the 7-torsion appear first in the homotopy groups of
Sn ?
 fiber bundles. classifying spaces. applications 269

11
Fiber bundles. Classifying spaces. Applications

11.1 Fiber bundles

Let G be a topological group (i.e., a topological space endowed with

a group structure so that the group multiplication and the inversion
map are continuous), acting continuously (on the left) on a topological
space F. Concretely, such a continuous action is given by a continuous
map ρ : G × F → F, ( g, m) 7→ g · m, which satisfies the conditions
( gh) · m = g · (h · m)) and eG · m = m, for eG the identity element of G.
Any continuous group action ρ induces a map
Adρ : G −→ Homeo( F )
given by g 7→ ( f 7→ g · f ), with g ∈ G, f ∈ F. Then Adρ is a group
homomorphism since
(Ad ρ)( gh)( f ) := ( gh) · f = g · (h · f ) = Adρ ( g)(Adρ (h)( f )).
Note that for nice spaces F (e.g., CW complexes), if we give Homeo( F )
the compact-open topology, then Adρ : G → Homeo( F ) is a continuous
group homomorphism, and any such continuous group homomor-
phism G → Homeo( F ) induces a continuous group action G × F → F.
We assume from now on that ρ is an effective action, i.e., Adρ is
injective.

Definition 11.1.1 (Atlas for a fiber bundle with group G and fiber F).
Given a continuous map π : E → B, an atlas for the structure of a fiber bundle
with group G and fiber F on π consists of the following data:
a) an open cover {Uα }α of B,

b) homeomorphisms hα : π −1 (Uα ) → Uα × F (called trivializing charts or

local trivializations) for each α so that the diagram
hα
π −1 (Uα ) / Uα × F

π pr1
# |
Uα
 270 algebraic topology

commutes,

c) continuous maps (called transition functions) gαβ : Uα ∩ Uβ → G so that

the horizontal map in the commutative diagram

π −1 (Uα ∩ Uβ )
hα hβ

v (
(Uα ∩ Uβ ) × F / (Uα ∩ Uβ ) × F
h β ◦ h−
α
1

is given by
( x, m) 7→ ( x, gβα ( x ) · m).

(By the effectivity of the action, if such maps gαβ exist, they are unique.)

Definition 11.1.2. Two atlases A and B on π are compatible if A ∪ B is an

atlas.

Definition 11.1.3 (Fiber bundle with group G and fiber F). A structure
of a fiber bundle with group G and fiber F on π : E → B is a maximal atlas
for π : E → B.

Example 11.1.4.

1. When G = {eG } is the trivial group, π : E → B has the structure

of a fiber bundle if and only if it is a trivial fiber bundle. Indeed,
the local trivializations hα of the atlas for the fiber bundle have
to satisfy h β ◦ h− 1
α : ( x, m ) 7 → ( x, e G · m ) = ( x, m ), which implies
h β ◦ h− 1
α = id, so h β = hα on Uα ∩ Uβ . This allows us to glue all
the local trivializations hα together to obtain a global trivialization
h : π −1 ( B ) = E ∼
= B × F.

2. When F is discrete, Homeo( F ) is also discrete, so G is discrete by

the effectiveness assumption. So for the atlas of π : E → B we have
π −1 (Uα ) ∼
= Uα × F = m∈ F Uα × {m}, so π is in this case a covering
S

map.

3. A locally trivial fiber bundle, as introduced in earlier chapters, is just

a fiber bundle with structure group Homeo( F ).

Lemma 11.1.5. The transition functions gαβ satisfy the following properties:

(a) gαβ ( x ) g βγ ( x ) = gαγ ( x ), for all x ∈ Uα ∩ Uβ ∩ Uγ .

−1
(b) g βα ( x ) = gαβ ( x ), for all x ∈ Uα ∩ Uβ .

(c) gαα ( x ) = eG .
 fiber bundles. classifying spaces. applications 271

Proof. On Uα ∩ Uβ ∩ Uγ , we have: (hα ◦ h− 1 −1 −1

β ) ◦ ( h β ◦ hγ ) = hα ◦ hγ .
Therefore, since Adρ is injective (i.e., ρ is effective), we get that

gαβ ( x ) g βγ ( x ) = gαγ ( x )

for all x ∈ Uα ∩ Uβ ∩ Uγ .
Note that (hα ◦ h− 1 −1
β ) ◦ ( h β ◦ hα ) = id, which translates into

( x, gαβ ( x ) gβα ( x ) · m) = ( x, m).

So, by effectiveness, gαβ ( x ) g βα ( x ) = eG for all x ∈ Uα ∩ Uβ , whence

−1
g βα ( x ) = gαβ ( x ).
Take γ = α in Property (a) to get gαβ ( x ) g βα ( x ) = gαα ( x ). So by
Property (b), we have gαα ( x ) = eG .

Transition functions determine a fiber bundle in a unique way, in the

sense of the following theorem.
Theorem 11.1.6. Given an open cover {Uα } of B and continuous functions
gαβ : Uα ∩ Uβ → G satisfying Properties (a)-(c), there is a unique structure
of a fiber bundle over B with group G, given fiber F, and transition functions
{ gαβ }.
e = Fα Uα × F × {α}, and define an equivalence rela-
Proof Sketch. Let E
tion ∼ on Ee by
( x, m, α) ∼ ( x, gαβ ( x ) · m, β),
for all x ∈ Uα ∩ Uβ , and m ∈ F. Properties (a)-(c) of { gαβ } are used
to show that ∼ is indeed an equivalence relation on E. e Specifically,
symmetry is implied by property (b), reflexivity follows from (c) and
transitivity is a consequence of the cycle property (a).
Let
e ∼
E = E/
be the set of equivalence classes in E, and define π : E → B locally by
[( x, m, α)] 7→ x for x ∈ Uα . Then it is clear that π is well-defined and
continuous (in the quotient topology), and the fiber of π is F.
It remains to show the local triviality of π. Let p : E e → E be the
quotient map, and let pα := p|Uα × F×{α} : Uα × F × {α} → π −1 (Uα ).
It is easy to see that pα is a homeomorphism. We define the local
trivializations of π by hα := p− 1
α .

Example 11.1.7.
1. Fiber bundles with fiber F = Rn and group G = GL(n, R) are called
rank n real vector bundles. For example, if M is a differentiable real
n-manifold, and TM is the set of all tangent vectors to M, then
π : TM → M is a real vector bundle on M of rank n. More precisely,
∼
=
if φα : Uα → Rn are trivializing charts on M, the transition functions
for TM are given by gαβ ( x ) = d( φα ◦ φ− 1
β ) φβ (x) .
272 algebraic topology

2. If F = Rn and G = O(n), we get real vector bundles with a Rieman-

nian structure.

3. Similarly, one can take F = Cn and G = GL(n, C) to get rank n

complex vector bundles. For example, if M is a complex manifold, the
tangent bundle TM is a complex vector bundle.

4. If F = Cn and G = U (n), we get real vector bundles with a hermitian

structure.

We also mention here the following fact:

Theorem 11.1.8. A fiber bundle has the homotopy lifting property with respect
to all CW complexes (i.e., it is a Serre fibration). Moreover, fiber bundles over
paracompact spaces are fibrations.

Definition 11.1.9 (Bundle homomorphism). Fix a topological group G

π′
acting effectively on a space F. A homomorphism between bundles E′ −→ B′
→ B with group G and fiber F is a pair ( f , fˆ) of continuous maps,
π
and E −
with f : B′ → B and fˆ : E′ → E, such that:

1. the diagram
fˆ
E′ /E

π′ π
 f 
B′ /B

commutes, i.e., π ◦ fˆ = f ◦ π ′ .
2. if {(Uα , hα )}α is a trivializing atlas of π and {(Vβ , Hβ )} β is a trivializing
atlas of π ′ , then the following diagram commutes:
Hβ fˆ
(Vβ ∩ f −1 (Uα )) × F o / π −1 (Uα ) / Uα × F
−1 hα
π′ (Vβ ∩ f −1 (Uα ))

π′ π
pr1 pr1
(   z
Vβ ∩ f −1 (Uα )
f
/ Uα

and there exist functions dαβ : Vβ ∩ f −1 (Uα ) → G such that for x ∈

Vβ ∩ f −1 (Uα ) and m ∈ F we have:

hα ◦ fˆ| ◦ Hβ−1 ( x, m) = ( f ( x ), dαβ ( x ) · m).

An isomorphism of fiber bundles is a bundle homomorphism ( f , fˆ) which

admits a map ( g, ĝ) in the reverse direction so that both composites are the
identity.

Remark 11.1.10. Gauge transformations of a bundle π : E → B are bundle

maps from π to itself over the identity of the base, i.e., corresponding
to continuous map g : E → E so that π ◦ g = π. By definition, such g
 fiber bundles. classifying spaces. applications 273

restricts to an isomorphism given by the action of an element of the

structure group on each fiber. The set of all gauge transformations
forms a group.
Proposition 11.1.11. Given functions dαβ : Vβ ∩ f −1 (Uα ) → G and dα′ β′ :
Vβ′ ∩ f −1 (Uα′ ) → G as in (2) above for different trivializing charts of π and
resp. π ′ , then for any x ∈ Vβ ∩ Vβ′ ∩ f −1 (Uα ∩ Uα′ ) ̸= ∅, we have

dα′ β′ ( x ) = gα′ α ( f ( x )) dαβ ( x ) g ββ′ ( x ) (11.1.1)

in G, where gα′ α are transition functions for π and g ββ′ are transition functions
for π ′ ,

Proof. Exercise.

The functions {dαβ } determine bundle maps in the following sense:

π′
Theorem 11.1.12. Given a map f : B′ → B and bundles E − → B, E′ −→ B′ ,
π

ˆ ′
a map of bundles ( f , f ) : π → π exists if and only if there exist continuous
maps {dαβ } as above, satisfying (11.1.1).

Proof. Exercise.

Theorem 11.1.13. Every bundle map fˆ over f = idB is an isomorphism. In

particular, gauge transformations are automorphisms.

Proof Sketch. Let dαβ : Vβ ∩ Uα → G be the maps given by the bundle

map fˆ : E′ → E. So, if dα′ β′ : Vβ′ ∩ Uα′ → G is given by a different choice
of trivializing charts, then (11.1.1) holds on Vβ ∩ Vβ′ ∩ Uα ∩ Uα′ ̸= ∅,
i.e.,
dα′ β′ ( x ) = gα′ α ( x ) dαβ ( x ) g ββ′ ( x ) (11.1.2)
in G, where gα′ α are transition functions for π and g ββ′ are transition
functions for π ′ . Let us now invert (11.1.2) in G, and set

d βα ( x ) = d− 1
αβ ( x )

to get:
d β′ α′ ( x ) = g β′ β ( x ) d βα ( x ) gαα′ ( x ).
So {d βα } are as in Definition 11.1.9 and satisfy (11.1.1). Theorem 11.1.12
implies that there exists a bundle map ĝ : E → E′ over idB .
We claim that ĝ is the inverse fˆ−1 of fˆ, and this can be checked
locally as follows:
fˆ ĝ
( x, m) 7→ ( x, dαβ ( x ) · m) 7→ ( x, d βα ( x ) · (dαβ ( x ) · m))
= ( x, d βα ( x )dαβ ( x ) ·m)
| {z }
eG

= ( x, m).

So ĝ ◦ fˆ = id E′ . Similarly, fˆ ◦ ĝ = id E
274 algebraic topology

One way in which fiber bundle homomorphisms arise is from the

pullback (or the induced bundle) construction.
π
Definition 11.1.14 (Induced Bundle). Given a bundle E − → B with group
G and fiber F, and a continuous map f : X → B, we define

f ∗ E := {( x, e) ∈ X × E | f ( x ) = π (e)},

with projections f ∗ π : f ∗ E → X, ( x, e) 7→ x, and fˆ : f ∗ E → E, ( x, e) 7→ e,

so that the following diagram commutes:

f ∗E E e

f ∗π π

X B
f

x f (x)

f ∗ π is called the induced bundle under f or the pullback of π by f , and as we

show below it comes equipped with a bundle map ( f , fˆ) : f ∗ π → π.
The above definition is justified by the following result:

Theorem 11.1.15.

(a) f ∗ π : f ∗ E → X is a fiber bundle with group G and fiber F.

(b) ( f , fˆ) : f ∗ π → π is a bundle map.

Proof Sketch. Let {(Uα , hα )}α be a trivializing atlas of π, and consider

the following commutative diagram:

hα
( f ∗ π )−1 ( f −1 (Uα )) π −1 (Uα ) Uα × F

f −1 (Uα ) Uα
f

We have

( f ∗ π )−1 ( f −1 (Uα )) = {( x, e) ∈ f −1 (Uα ) × π −1 (Uα ) | f ( x ) = π (e)}.

| {z }
∼
=Uα × F

Define
k α : ( f ∗ π )−1 ( f −1 (Uα )) −→ f −1 (Uα ) × F
by
( x, e) 7→ ( x, pr2 (hα (e))).
Then it is easy to check that k α is a homeomorphism (with inverse
k− 1 −1
α ( x, m ) = ( x, hα ( f ( x ), m )), and in fact the following assertions hold:
 fiber bundles. classifying spaces. applications 275

(i) {( f −1 (Uα ), k α )}α is a trivializing atlas of f ∗ π.

(ii) the transition functions of f ∗ π are f ∗ gαβ := gαβ ◦ f , i.e., f ∗ gαβ ( x ) =

gαβ ( f ( x )) for any x ∈ f −1 (Uα ∩ Uβ ).

Remark 11.1.16. It is easy to see that ( f ◦ g)∗ π = g∗ ( f ∗ π ) and (id B )∗ π =

π. Moreover, the pullback of a trivial bundle is a trivial bundle.

As we shall see later on, the following important result holds:

Theorem 11.1.17. Given a fibre bundle π : E → B with group G and

fiber F, and two homotopic maps f ≃ g : X → B, there is an isomorphism
f ∗π ∼
= g∗ π of bundles over X. (In short, induced bundles under homotopic
maps are isomorphic.)

As a consequence, we have:

Corollary 11.1.18. A fiber bundle over a contractible space B is trivial.

Proof. Since B is contractible, id B is homotopic to the constant map ct.

Let
i
b := Im(ct) ,→ B,

so i ◦ ct ≃ id B . We have a diagram of maps and induced bundles:

ct∗ i∗ E / i∗ E /E

ct∗ i∗ π i∗ π π
  
B
ct / {b} i /B
>
id B

Theorem 11.1.17 then yields:

π∼
= (id B )∗ π ∼
= ct∗ i∗ π.

Since any fiber bundle over a point is trivial, we have that i∗ π ∼

= {b} × F
∼ ∗ ∗ ∼
is trivial, hence π = ct i π = B × F is also trivial.

Proposition 11.1.19. If

f˜
E′ /E

π′ π
 f 
B′ /B

is a bundle map, then π ′ ∼

= f ∗ π as bundles over B′ .
276 algebraic topology

Proof. Define h : E′ → f ∗ E by e′ 7→ (π ′ (e′ ), f˜(e′ )) ∈ B′ × E. This is

well-defined, i.e., h(e′ ) ∈ f ∗ E, since f (π ′ (e′ )) = π ( f˜(e′ )).
It is easy to check that h provides the desired bundle isomorphism
over B′ .
E′
h f˜

fˆ
f ∗E /) E
π′

π′ π
  f 
B′ /B

Example 11.1.20. We can now show that the set of isomorphism classes
of bundles over Sn with group G and fiber F is isomorphic to πn−1 ( G ).
Indeed, let us cover Sn with two contractible sets U+ and U− obtained
by removing the south, resp., north pole of Sn . Let i± : U± ,→ Sn be
the inclusions. Then any bundle π over Sn is trivial when restricted
∗ π ∼ U × F. In particular, U provides a trivializing
to U± , that is, i± = ± ±
cover (atlas) for π, and any such bundle π is completely determined by
the transition function g± : U+ ∩ U− ≃ Sn−1 → G, i.e., by an element
in πn−1 ( G ).

More generally, we aim to “classify” fiber bundles on a given topo-

logical space. Let B ( X, G, F, ρ) denote the isomorphism classes (over
id X ) of fiber bundles on X with group G and fiber F, and G-action
on F given by ρ. If f : X ′ → X is a continuous map, the pullback
construction defines a map

f ∗ : B ( X, G, F, ρ) −→ B X ′ , G, F, ρ

so that (id X )∗ = id and ( f ◦ g)∗ = g∗ ◦ f ∗ .

11.2 Principal Bundles

As we will see later on, the fiber F doesn’t play any essential role in the
classification of fiber bundle, and in fact it is enough to understand the
set
P ( X, G ) := B ( X, G, G, mG )

of fiber bundles with group G and fiber G, where the action of G on

itself is given by the multiplication mG of G. Elements of P ( X, G ) are
called principal G-bundles. Of particular importance in the classification
theory of such bundles is the universal principal G-bundle G ,→ EG →
BG, with contractible total space EG.
 fiber bundles. classifying spaces. applications 277

Example 11.2.1. Any regular cover p : E → X is a principal G-bundle,

with group G = π1 ( X )⧸p π ( E). Here G is given the discrete topology.
∗ 1
e → X is a principal π1 ( X )-bundle.
In particular, the universal covering X

Example 11.2.2. Any free (right) action of a finite group G on a (Haus-

dorff) space E gives a regular cover and hence a principal G-bundle
E → E/G.

More generally, we have the following:

Theorem 11.2.3. Let π : E → X be a principal G-bundle. Then G acts freely

and transitively on the right of E so that E⧸G ∼
= X. In particular, π is the
quotient (orbit) map.

Proof. We will define the action locally over a trivializing chart for π.
Let Uα be a trivializing open in X with trivializing homeomorphism
∼
=
hα : π −1 (Uα ) → Uα × G. We define a right action on G on π −1 (Uα ) by

π −1 (Uα ) × G → π −1 (Uα ) ∼
= Uα × G
(e, g) 7→ e · g := h− 1
α ( π ( e ) , pr2 ( hα ( e )) · g )

Let us show that this action can be globalized, i.e., it is independent of

the choice of the trivializing open Uα . If (Uβ , h β ) is another trivializing
chart in X so that e ∈ π −1 (Uα ∩ Uβ ), we need to show that e · g =
h− 1



β π (e) , pr2 h β (e) · g , or equivalently,

h− 1 −1



α ( π ( e ) , pr2 ( hα ( e )) · g ) = h β π (e) , pr2 h β (e) · g . (11.2.1)

After applying hα and using the transition function gαβ for π (e) ∈
Uα ∩ Uβ , (11.2.1) becomes

(π (e) , pr2 (hα (e)) · g) = hα h− 1

β π (e) , pr2 h β (e) · g



= π (e) , gαβ (π (e)) · (pr2 h β (e) · g) ,

which is guaranteed by the definition of an atlas for π.

It is easy to check locally that the action is free and transitive. More-
over, E⧸G is locally given as Uα × G⧸G ∼ = Uα , and this local quotient
globalizes to X.

The converse of the above theorem holds in some important cases.

Theorem 11.2.4. Let E be a compact Hausdorff space and G a compact

Lie group acting freely on E. Then the orbit map E → E/G is a principal
G-bundle.

Corollary 11.2.5. Let G be a Lie group, and let H < G be a compact

subgroup. Then the projection onto the orbit space π : G → G/H is a
principal H-bundle.
278 algebraic topology

Let us now fix a G-space F. We define a map

P ( X, G ) → B ( X, G, F, ρ)

as follows. Start with a principal G bundle π : E → X, and recall from

the previous theorem that G acts freely on the right on E. Since G acts
on the left on F, we have a left G-action on E × F given by:

g · (e, f ) 7→ (e · g−1 , g · f ).

Let
E ×G F := E × F⧸G

be the corresponding orbit space, with projection map ω : E ×G F →

E⧸ ∼
G = X fitting into a commutative diagram

E×F (11.2.2)

pr1 $
} E × F⧸
E G
π
ω
! z
X

Definition 11.2.6. The projection ω := π ×G F : E ×G F → X is called the

associated bundle with fiber F.

The terminology in the above definition is justified by the following

result.

Theorem 11.2.7. ω : E ×G F → X is a fiber bundle with group G, fiber

F, and having the same transition functions as π. Moreover, the assign-
ment π 7→ ω := π ×G F defines a one-to-one correspondence P ( X, G ) →
B ( X, G, F, ρ).

Proof. Let hα : π −1 (Uα ) → Uα × G be a trivializing chart for π. Recall

that for e ∈ π −1 (Uα ), f ∈ F and g ∈ G, if we set hα (e) = (π (e), h) ∈
Uα × G, then G acts on the right on π −1 (Uα ) by acting on the right on
h = pr2 (hα (e)). Then we have by the diagram (11.2.2) that

ω −1 (Uα ) ∼ π −1 (Uα ) × F⧸
= (e, f ) ∼ (e · g−1 , g · f )
= Uα × G × F⧸(u, h, f ) ∼ (u, hg−1 , g · f ).
∼

Let us define
k α : ω −1 (Uα ) → Uα × F

by
[(u, h, f )] 7→ (u, h · f ).
 fiber bundles. classifying spaces. applications 279

This is a well-defined map since

[(u, hg−1 , g · f )] 7→ (u, hg−1 g · f ) = (u, h · f ).

It is easy to check that k α is a trivializing chart for ω with inverse

induced by Uα × F → Uα × G × F, (u, f ) 7→ (u, idG , f ). It is clear that
ω and π have the same transition functions as they have the same
trivializing opens.

The associated bundle construction is easily seen to be functorial in

the following sense.

Proposition 11.2.8. If

fb
E′ /E

π′ π
 f 
X′ /X

is a map of principal G-bundles (so fb is a G-equivariant map, i.e., fb(e · g) =

fb(e) · g), then there is an induced map of associated bundles with fiber F,

fb×G id F
E′ × G F / E ×G F

π′ π
 f 
X′ /X

Example 11.2.9. Let π : S1 → S1 , z 7→ z2 be regarded as a principal

Z/2-bundle, and let F = [−1, 1]. Let Z/2 = {1, −1} act on F by
multiplication. Then the bundle associated to π with fiber F = [−1, 1]
1
is the Möbius strip S1 ×Z/2 [−1, 1] = S × [−1, 1]⧸( x, t) ∼ ( a( x ), −t),
with a : S1 → S1 denoting the antipodal map. Similarly, the bundle
associated to π with fiber F = S1 is the Klein bottle.

Let us now get back to proving the following important result.

Theorem 11.2.10. Let π : E → Y be a fiber bundle with group G and fiber F,

and let f ≃ g : X → Y be two homotopic maps. Then f ∗ π ∼= g∗ π over id X .

It is of course enough to prove the theorem in the case of principal

G-bundles. The idea of proof is to construct a bundle map over id X
between f ∗ π and g∗ π:

f ∗E
? / g∗ E

~
X
280 algebraic topology

So we first need to understand maps of principal G-bundles, i.e., to

solve the following problem: given two principal G-bundles bundles
1 π 2 π
E1 −→ X and E2 −→ Y, describe the set maps(π1 , π2 ) of bundle maps

fˆ
E1 / E2

π1 π2
 f 
X /Y

Since G acts on the right of E1 and E2 , we also get an action on the left
of E2 by g · e2 := e2 · g−1 . Then we get an associated bundle of π1 with
fiber E2 , namely

ω := π1 ×G E2 : E1 ×G E2 −→ X.

We have the following result:

Theorem 11.2.11. Bundle maps from π1 to π2 are in one-to-one correspon-
dence to sections of ω.

Proof. We work locally, so it suffices to consider only trivial bundles.

Given a bundle map ( f , fb) : π1 7→ π2 , let U ⊂ Y open, and V ⊂
f −1 (U ) open, so that the following diagram commutes (this is the
bundle maps in trivializing charts)

fb
V×G U×G
π1 π2
f
V U

We define a section σ in

( V × G ) × G (U × G )

σ ω

as follows. For e1 ∈ V × G, with x = π1 (e1 ) ∈ V, we set

σ ( x ) = [e1 , fb(e1 )].

This map is well-defined, since for any g ∈ G we have:

[e1 · g, fb(e1 · g)] = [e1 · g, fb(e1 ) · g] = [e1 · g, g−1 · fb(e1 )] = [e1 , fb(e1 )].

Now, it is an exercise in point-set topology (using the local definition

of a bundle map) to show that σ is continuous.
ω
Conversely, given a section of E1 ×G E2 7→ X, we define a bundle by
( f , fb) by
fb(e1 ) = e2 ,
 fiber bundles. classifying spaces. applications 281

where σ (π1 (e1 )) = [(e1 , e2 )]. Note that this is an equivariant map
because
[e1 · g, e2 · g] = [e1 · g, g−1 · e2 ] = [e1 , e2 ],

hence fb(e1 · g) = e2 · g = fb(e1 ) · g. Thus fb descends to a map f : X → Y

on the orbit spaces. We leave it as an exercise to check that ( f , fb) is
indeed a bundle map, i.e., to show that locally fb(v, g) = ( f (v), d(v) g)
with d(v) ∈ G and d : V → G a continuous function.

The following result will be needed in the proof of Theorem 11.2.10.

Lemma 11.2.12. Let π : E → X × I be a bundle, and let π0 := i0∗ π :

E0 → X be the pullback of π under i0 : X → X × I, x 7→ ( x, 0). Then
π∼= ( pr1 )∗ π0 ∼
= π0 × id I , where pr1 : X × I → X is the projection map.

Proof. It suffices to find a bundle map ( pr1 , pr

b 1 ) so that the following
diagram commutes

bi0 pr
/E / E0
b1
E0
π0 π π0
   
X
i0 pr1
/ X×I /X

By Theorem 11.2.11, this is equivalent to the existence of a section

σ of ω : E ×G E0 → X × I. Note that there exists a section σ0 of
ω0 : E0 ×G E0 → X = X × {0}, corresponding to the bundle map
(id X , id E0 ) : π0 → π0 . Then composing σ0 with the top inclusion arrow,
we get the following diagram

X × { 0}
σ0
/ E ×G E0
_ 9
s
ω
 
X×I
id / X×I

Since ω is a fibration, by the homotopy lifting property one can extend

sσ0 to a section σ of ω.

We can now finish the proof of Theorem 11.2.10.

Proof of Theorem 11.2.10. Let H : X × I → Y be a homotopy between f

and g, with H ( x, 0) = f ( x ) and H ( x, 1) = g( x ). Consider the induced
bundle H ∗ π over X × I. Then we have the following diagram.
282 algebraic topology

f ∗E / H∗ E H
b
/E
:

f ∗π g∗ E H∗ π π

   
X × {0} 
i0
g∗ π / X×I H /Y
:
i1 pr1

 , !
X × {1} X

Since f = H (−, 0), we get f ∗ π = i0∗ H ∗ π. By Lemma 11.2.12, H ∗ π ∼ =

pr1∗ ( f ∗ π ) ∼
= pr1∗ ( g∗ π ), and thus f ∗ π = i0∗ H ∗ π = i0∗ pr1∗ g∗ π = g∗ π.

We conclude this section with the following important consequence

of Theorem 11.2.11

Corollary 11.2.13. A principle G-bundle π : E → X is trivial if and only if

π has a section.

Proof. The bundle π is trivial if and only if π = ct∗ π ′ , with ct : X →

point the constant map, and π ′ : G → point the trivial bundle over a
point space. This is equivalent to saying that there is a bundle map

E /G

π π′
 
X
ct / point

or, by Theorem 11.2.11, to the existence of a section of the bundle

ω : E ×G G → X. On the other hand, ω ∼ = π, since E ×G G → X looks
locally like

π −1 (Uα ) × G⧸ ∼ Uα × G × G⧸ ∼
∼= (u, g1 , g2 ) ∼ (u, g1 g−1 , gg2 ) = Uα × G,

with the last homeomorphism defined by [(u, g1 , g2 )] 7→ (u, g1 g2 ).

Altogether, π is trivial if and only if π : E 7→ X has a section.

11.3 Classification of principal G-bundles

Let us assume for now that there exists a principal G-bundle πG :

EG → BG, with contractible total space EG. As we will see below, such
a bundle plays an essential role in the classification theory of principal
G-bundles. Its base space BG turns out to be unique up to homotopy,
and it is called the classifying space for principal G-bundles due to the
following fundamental result:
 fiber bundles. classifying spaces. applications 283

Theorem 11.3.1. If X is a CW-complex, there exists a bijective correspondence

∼
=
Φ : P ( X, G ) −
→ [ X, BG ]
f ∗ π G ←[ f

Proof. By Theorem 11.2.10, Φ is well-defined.

Let us next show that Φ is onto. Let π ∈ P ( X, G ), π : E → X. We
need to show that π ∼= f ∗ πG for some map f : X → BG, or equivalently,
that there is a bundle map ( f , fb) : π → πG . By Theorem 11.2.11, this
is equivalent to the existence of a section of the bundle E ×G EG → X
with fiber EG. Since EG is contractible, such a section exists by the
following:
Lemma 11.3.2. Let X be a CW complex, and π : E → X ∈ B ( X, G, F, ρ)
with πi ( F ) = 0 for all i ≥ 0. If A ⊆ X is a subcomplex, then every section
of π over A extends to a section defined on all of X. In particular, π has a
section. Moreover, any two sections of π are homotopic.

Proof. Given a section σ0 : A → E of π over A, we extend it to a section

σ : X → E of π over X by using induction on the dimension of cells in
X − A. So it suffices to assume that X has the form

X = A ∪ϕ en ,

where en is an n-cell in X − A, with attaching map ϕ : ∂en → A. Since

en is contractible, π is trivial over en , so we have a commutative diagram
∼
π −1 ( e n )
= / en × F
; h F
σ0 pr1
π

 / en y
∂en
σ

with h : π −1 (en ) → en × F the trivializing chart for π over en , and σ

to be defined. After composing with h, we regard the restriction of σ0
over ∂en as given by

σ0 ( x ) = ( x, τ0 ( x )) ∈ en × F,

with τ0 : ∂en ∼
= Sn−1 → F. Since πn−1 ( F ) = 0, τ0 extends to a map
τ : e → F which can be used to extend σ0 over en by setting
n

σ ( x ) = ( x, τ ( x )).

After composing with h−1 , we get the desired extension of σ0 over en .

Let us now assume that σ and σ′ are two sections of π. To find
a homotopy between σ and σ′ , it suffices to construct a section Σ of
π × id I : E × I → X × I. Indeed, if such Σ exists, then Σ( x, t) =
(σt ( x ), t), and σt provides the desired homotopy. Now, by regarding
284 algebraic topology

σ as a section of π × id I over X × {0}, and σ′ as a section of π × id I

over X × {1}, the question reduces to constructing a section of π × id I ,
which extends the section over X × {0, 1} defined by (σ, σ′ ). This can
be done as in the first part of the proof.

In order to finish the proof of Theorem 11.3.1, it remains to show

that Φ is a one-to-one map. If π0 = f ∗ πG ∼
= g∗ πG = π1 , we will show
that f ≃ g. Note that we have the following commutative diagrams:

fb
E0 = f ∗ EG −−−−→ EG
 
π π
y 0 y G
f
X = X × {0} −−−−→ BG
gb
E0 ∼
= E1 = g∗ EG −−−−→ EG
 
π π
y 0 y G
g
X = X × {1} −−−−→ BG
where we regard gb as defined on E0 via the isomorphism π0 ∼
= π1 . By
putting together the above diagrams, we have a commutative diagram

←- α=( fb,0)∪( gb,1)

E0 × I ←−−−− E0 × {0, 1} −−−−−−−−→
b
EG
  
π × Id
yπ0 ×{0,1}
 π
y 0 y G
←- α=( f ,0)∪( g,1)
X × I ←−−−− X × {0, 1} −−−−−−−−→ BG

Therefore, it suffices to extend (α, b

α) to a bundle map ( H, Hb ) : π0 ×
Id → πG , and then H will provide the desired homotopy f ≃ g.
By Theorem 11.2.11, such a bundle map ( H, H b ) corresponds to a
section σ of the fiber bundle

ω : ( E0 × I ) ×G EG → X × I.

On the other hand, the bundle map (α, b

α) already gives a section σ0 of
the fiber bundle

ω0 : ( E0 × {0, 1}) ×G EG → X × {0, 1},

which under the obvious inclusion ( E0 × {0, 1}) ×G EG ⊆ ( E0 × I ) ×G

EG can be regarded as a section of ω over the subcomplex X × {0, 1}.
Since EG is contractible, Lemma 11.3.2 allows us to extend σ0 to a
section σ of ω defined on X × I, as desired.

Example 11.3.3. We give here a more conceptual reasoning for the

assertion of Example 11.1.20. By Theorem 11.3.1, we have

B(Sn , G, F, ρ) ∼
= P (Sn , G ) ∼
= [Sn , BG ] = πn ( BG ) ∼
= π n −1 ( G ),
 fiber bundles. classifying spaces. applications 285

where the last isomorphism follows from the homotopy long exact
sequence for πG , since EG is contractible.

Back to the universal principal G-bundle, we have the following

Theorem 11.3.4. Let G be a locally compact topological group. Then a uni-

versal principal G-bundle πG : EG → BG exists (i.e., satisfying πi ( EG ) = 0
for all i ≥ 0), and the construction is functorial in the sense that a continuous
group homomorphism µ : G → H induces a bundle map ( Bµ, Eµ) : πG →
π H . Moreover, the classifying space BG is unique up to homotopy.

Proof. To show that BG is unique up to homotopy, let us assume that

πG : EG → BG and πG ′ : E′ → B′ are universal principal G-bundles.
G G
′ , we get a
By regarding πG as the universal principal G-bundle for πG
′ ′ ∗
map f : BG → BG such that πG = f πG , i.e., a bundle map:

′ − fˆ
EG −−−→ EG
 
π ′ π
y G y G
′ − f
BG −−−→ BG

Similarly, regarding πG ′ as the universal principal G-bundle for π ,

G
there exists a map g : BG → BG ′ such that π = g∗ π ′ . Therefore,
G G

π G = g∗ π G
′
= g∗ f ∗ π G = ( f ◦ g)∗ π G .

On the other hand, we have πG = (id BG )∗ πG , so by Theorem 11.3.1

we get that f ◦ g ≃ id BG . Similarly, we get g ◦ f ≃ id B′ , and hence
G
′ → B is a homotopy equivalence.
f : BG G
We will not discuss the existence of the universal bundle here, instead
we will indicate the universal G-bundle, as needed, in specific examples.

Example 11.3.5. Recall from Section 9.12 that we have a fiber bundle

O(n)  / Vn (R∞ ) / Gn (R∞ ), (11.3.1)

with Vn (R∞ ) contractible. In particular, the uniqueness part of Theorem

11.3.4 tells us that BO(n) ≃ Gn (R∞ ) is the classifying space for rank n
real vector bundles. Similarly, there is a fiber bundle

U (n)  / Vn (C∞ ) / Gn (C∞ ), (11.3.2)

with Vn (C∞ ) contractible. Therefore, BU (n) ≃ Gn (C∞ ) is the classify-

ing space for rank n complex vector bundles.

Before moving to the next example, let us mention here without

proof the following useful result:
286 algebraic topology

Theorem 11.3.6. Let G be an abelian group, and let X be a CW complex.

There is a natural bijection

T : [ X, K ( G, n)] −→ H n ( X, G )

[ f ] 7→ f ∗ (α)
where α ∈ H n (K ( G, n), G ) ∼
= Hom( Hn (K ( G, n), Z), G ) is given by the
inverse of the Hurewicz isomorphism G = πn (K ( G, n)) → Hn (K ( G, n), Z).

Example 11.3.7 (Classification of real line bundles). Let G = Z/2

and consider the principal Z/2-bundle Z/2 ,→ S∞ → RP∞ . Since
S∞ is contractible, the uniqueness of the universal bundle yields that
BZ/2 ∼ = RP∞ . In particular, we see that RP∞ classifies the real line
(i.e., rank-one) bundles. Since we also have that RP∞ = K (Z/2, 1), we
get:

P ( X, Z/2) = [ X, BZ/2] = [ X, K (Z/2, 1)] ∼

= H 1 ( X, Z/2)

for any CW complex X, where the last identification follows from

Theorem 11.3.6. Let now π be a real line bundle on a CW complex X,
with classifying map f π : X → RP∞ . Since H ∗ (RP∞ , Z/2) ∼ = Z/2[w],
with w a generator of H 1 (RP∞ , Z/2), we get a well-defined degree one
cohomology class
w1 (π ) := f π∗ (w)
∼
=
called the first Stiefel-Whitney class of π. The bijection P ( X, Z/2) −→
H 1 ( X, Z/2) is then given by π 7→ w1 (π ), so real line bundles on X are
classified by their first Stiefel-Whitney classes.
Example 11.3.8 (Classification of complex line bundles). Let G = S1
and consider the principal S1 -bundle S1 ,→ S∞ → CP∞ . Since S∞ is
contractible, the uniqueness of the universal bundle yields that BS1 ∼
=
∞ ∞
CP . In particular, as S = GL(1, C), we see that CP classifies
1

the complex line (i.e., rank-one) bundles. Since we also have that
CP∞ = K (Z, 2), we get:

P ( X, S1 ) = [ X, BS1 ] = [ X, K (Z, 2)] ∼

= H 2 ( X, Z)

for any CW complex X, where the last identification follows from

Theorem 11.3.6. Let now π be a complex line bundle on a CW complex
X, with classifying map f π : X → CP∞ . Since H ∗ (CP∞ , Z) ∼
= Z[ c ],
∞
with c a generator of H (CP , Z), we get a well-defined degree two
2

cohomology class
c1 (π ) := f π∗ (c)
∼
=
called the first Chern class of π. The bijection P ( X, S1 ) −→ H 2 ( X, Z) is
then given by π 7→ c1 (π ), so complex line bundles on X are classified
by their first Chern classes.
 fiber bundles. classifying spaces. applications 287

Remark 11.3.9. If X is any closed oriented surface, then H 2 ( X, Z) ∼

= Z,
so Example 11.3.8 shows that isomorphism classes of complex line
bundles on X are in bijective correspondence with the set of inte-
gers. On the other hand, if X is a non-orientable closed surface, then
H 2 ( X, Z) ∼
= Z/2, so there are only two isomorphism classes of complex
line bundles on such a surface.

11.4 Exercises

1. Let p : S2 → RP2 be the (oriented) double cover of RP2 . Since RP2

is a non-orientable surface, we know by Remark 11.3.9 that there are
only two isomorphism classes of complex line bundles on RP2 : the
trivial one, and a non-trivial complex line bundle which we denote
by π : E → RP2 . On the other hand, since S2 is a closed orientable
surface, the isomorphism classes of complex line bundles on S2 are in
bijection with Z. Which integer corresponds to complex line bundle
p∗ π : p∗ E → S2 on S2 ?
π
2. Consider a locally trivial fiber bundle S2 ,→ E → S2 . Recall that
such π can be regarded as a fiber bundle with structure group G =
Homeo (S2 ) ∼
= SO(3). By the classification Theorem 11.3.1, SO(3)-
bundles over S2 correspond to elements in

[S2 , BSO(3)] = π2 ( BSO(3)) ∼

= π1 (SO(3)).

(a) Show that π1 (SO(3)) ∼

= Z/2. (Hint: Show that SO(3) is homeo-
morphic to RP3 .)

(b) What is the non-trivial SO(3)-bundle over S2 ?

3. Let π : E → X be a principal S1 -bundle over the simply-connected

space X. Let a ∈ H 1 (S1 , Z) be a generator. Show that

c1 ( π ) = d2 ( a ),

where d2 is the differential on the E2 -page of the Leray-Serre spectral

p,q
sequence associated to π, i.e., E2 = H p ( X, H q (S1 )) ⇛ H p+q ( E, Z).

4. By the classification Theorem 11.3.1, (isomorphism classes of) S1 -

bundles over S2 are given by

[S2 , BS1 ] = π2 ( BS1 ) ∼

= π1 ( S1 ) ∼
=Z

and this correspondence is realized by the first Chern class, i.e., π 7→

c1 ( π ).

(a) What is the first Chern class of the Hopf bundle S1 ,→ S3 → S2 ?

288 algebraic topology

(b) What is the first Chern class of the sphere (or unit) bundle of the
tangent bundle TS2 ?

(c) Construct explicitly the S1 -bundle over S2 corresponding to n ∈

Z. (Hint: Think of lens spaces, and use the above Exercise 3 and
Example 10.8.2.)
 vector bundles. characteristic classes. cobordism. applications. 289

12
Vector Bundles. Characteristic classes.
Cobordism. Applications.

12.1 Chern classes of complex vector bundles

We begin with the following

Proposition 12.1.1.

H ∗ ( BU (n); Z) ∼
= Z [ c1 , · · · , c n ] ,

with deg ci = 2i

Proof. Recall from Example 10.12.1 that H ∗ (U (n); Z) is a free Z-algebra

on odd degree generators x1 , · · · , x2n−1 , with deg( xi ) = i, i.e.,

H ∗ (U ( n ) ; Z ) ∼
= ΛZ [ x1 , · · · , x2n−1 ].

Then using the Leray-Serre spectral sequence for the universal U (n)-
bundle, and using the fact that EU (n) is contractible, yields the desired
result.
Alternatively, the functoriality of the universal bundle construction
yields that for any subgroup H < G of a topological group G, there
is a fibration G/H ,→ BH → BG. In our case, consider ! U (n − 1) as a
A 0
subgroup of U (n) via the identification A 7→ . Hence, there
0 1
exists fibration

U (n)/U (n − 1) ∼
= S2n−1 ,→ BU (n − 1) → BU (n).

Then the Leray-Serre spectral sequence and induction on n gives

the desired result, where we use the fact that BU (1) ≃ CP∞ and
H ∗ (CP∞ ; Z) ∼
= Z[c] with deg c = 2.

Definition 12.1.2. The generators c1 , · · · , cn of H ∗ ( BU (n); Z) are called

the universal Chern classes of U (n)-bundles.
290 algebraic topology

Recall from the classification theorem 11.3.1, that given π : E → X a

principal U (n)-bundle, there exists a “classifying map” f π : X → BU (n)
such that π ∼= f π∗ πU (n) .

Definition 12.1.3. The i-th Chern class of the U (n)-bundle π : E → X with

classifying map f π : X → BU (n) is defined as

ci (π ) := f π∗ (ci ) ∈ H 2i ( X; Z).

Remark 12.1.4. Note that if π is a U (n)-bundle, then by definition we

have that ci (π ) = 0, if i > n.

Let us now discuss important properties of Chern classes.

Proposition 12.1.5. If E denotes the trivial U (n)-bundle on a space X, then

ci (E ) = 0 for all i > 0.

Proof. Indeed, the trivial bundle is classified by the constant map ct :

X → BU (n), which induces trivial homomorphisms in positive degree
cohomology.

Proposition 12.1.6 (Functoriality of Chern classes). If f : Y → X is a

continuous map, and π : E → X is a U (n)-bundle, then ci ( f ∗ π ) = f ∗ ci (π ),
for any i.

Proof. We have a commutative diagram

fb
f ∗E /E / EU (n)

f ∗π π πU ( n )
 f  fπ 
Y /X / BU (n)

which shows that f π ◦ f classifies the U (n)-bundle f ∗ π on Y. Therefore,

ci ( f ∗ π ) = ( f π ◦ f )∗ ci
= f ∗ ( f π∗ ci )
= f ∗ ci ( π ) .

Definition 12.1.7. The total Chern class of a U (n)-bundle π : E → X is

defined by

c(π ) = c0 (π ) + c1 (π ) + · · · cn (π ) = 1 + c1 (π ) + · · · cn (π ) ∈ H ∗ ( X; Z),

as an element in the cohomology ring of the base space.

 vector bundles. characteristic classes. cobordism. applications. 291

Definition 12.1.8 (Whitney sum). Let π1 ∈ P ( X, U (n)), π2 ∈ P ( X, U (m)).

Consider the product bundle π1 × π2 ∈ P ( X × X, U (n) × U (m)), which
can be regarded as a U (n + m)-bundle via the!canonical inclusion U (n) ×
A 0
U (m) ,→ U (n + m), ( A, B) 7→ . The Whitney sum of the
0 B
bundles π1 and π2 is defined as:

π1 ⊕ π2 : = ∆ ∗ ( π1 × π2 ),

where ∆ : X → X × X is the diagonal map given by x 7→ ( x, x ).

Remark 12.1.9. The Whitney sum π1 ⊕ π2 of π1 and π2 is the U (n + m)-
bundle on X with transition functions (in a common refinement
! of the
1
gαβ 0
trivialization atlases for π1 and π2 ) given by i
where gαβ
0 2
gαβ
are the transition function of πi , i = 1, 2.

Proposition 12.1.10 (Whitney sum formula). If π1 ∈ P ( X, U (n)) and

π2 ∈ P ( X, U (m)), then

c ( π1 ⊕ π2 ) = c ( π1 ) ∪ c ( π2 ).

Equivalently, ck (π1 ⊕ π2 ) = ∑i+ j=k ci (π1 ) ∪ c j (π2 )

Proof. First note that

B(U (n) × U (m)) ≃ BU (n) × BU (m). (12.1.1)

Indeed, by taking the product of the universal bundles for U (n) and
U (m), we get a U (n) × U (m)-bundle over BU (n) × BU (m), with total
space EU (n) × EU (m):

U (n) × U (m) ,→ EU (n) × EU (m) → BU (n) × BU (m). (12.1.2)

Since πi ( EU (n) × EU (m)) ∼ = πi ( EU (n)) × πi ( EU (m)) ∼

= 0 for all i,
it follows that (12.1.2) is the universal bundle for U (n) × U (m), thus
proving (12.1.1).
Next, the inclusion U (n) × U (m) ,→ U (n + m) yields a map

ω : B(U (n) × U (m)) ≃ BU (n) × BU (m) −→ BU (n + m).

By using the Künneth formula, one can show (e.g., see Milnor’s book,
p.164) that:
ω ∗ c k = ∑ ci × c j .
i + j=k

Therefore,

ck (π1 ⊕ π2 ) = ck (∆∗ (π1 × π2 ))

= ∆ ∗ c k ( π1 × π2 )
292 algebraic topology

= ∆∗ ( f π∗1 ×π2 (ck ))

= ∆∗ ( f π∗1 × f π∗2 )(ω ∗ ck )
= ∑ ∆∗ ( f π∗1 (ci ) × f π∗2 (c j ))
i+ j=k

= ∑ ∆∗ (ci (π1 ) × c j (π2 ))

i+ j=k

= ∑ c i ( π1 ) ∪ c j ( π2 ).
i+ j=k

Here, we use the fact that the classifying map for π1 × π2 , regarded as
a U (n + m)-bundle is ω ◦ ( f π1 × f π2 ).

Since the trivial bundle has trivial Chern classes in positive degrees,
we get

Corollary 12.1.11 (Stability of Chern classes). Let E 1 be the trivial U (1)-

bundle. Then
c ( π ⊕ E 1 ) = c ( π ).

12.2 Stiefel-Whitney classes of real vector bundles

As in Proposition 12.1.1, one easily obtains the following

Proposition 12.2.1.

H ∗ ( BO(n); Z/2) ∼
= Z/2 [w1 , · · · , wn ] ,

with deg wi = i.

Proof. This can be easily deduced by induction on n from the Leray-

Serre spectral sequence of the fibration

O(n)/O(n − 1) ∼
= Sn−1 ,→ BO(n − 1) → BO(n),

by using the fact that BO(1) ≃ RP∞ and H ∗ (RP∞ ; Z/2) ∼

= Z/2[w1 ].

Definition 12.2.2. The generators w1 , · · · , wn of H ∗ ( BO(n); Z/2) are

called the universal Stiefel-Whitney classes of O(n)-bundles.

Recall from the classification theorem 11.3.1 that, given π : E → X a

principal O(n)-bundle, there exists a “classifying map” f π : X → BO(n)
such that π ∼= f π∗ πU (n) .

Definition 12.2.3. The i-th Stiefel-Whitney class of the O(n)-bundle π :

E → X with classifying map f π : X → BO(n) is defined as

wi (π ) := f π∗ (wi ) ∈ H i ( X; Z/2).
 vector bundles. characteristic classes. cobordism. applications. 293

The total Stiefel-Whitney class of π is defined by

w(π ) = 1 + w1 (π ) + · · · wn (π ) ∈ H ∗ ( X; Z/2),

as an element in the cohomology ring with Z/2-coefficients.

Remark 12.2.4. If π is a O(n)-bundle, then by definition we have that

wi (π ) = 0, if i > n. Also, since the trivial bundle is classified by the
constant map, it follows that the positive-degree Stiefel-Whitney classes
of a trivial O(n)-bundle are all zero.

Stiefel-Whitney classes of O(n)-bundles enjoy similar properties as

the Chern classes.

Proposition 12.2.5. The Stiefel-Whitney classes satisfy the functoriality

property and the Whitney sum formula.

12.3 Stiefel-Whitney classes of manifolds and applications

If M is a smooth manifold, its tangent bundle TM can be regarded as

an O(n)-bundle.

Definition 12.3.1. The Stiefel-Whitney classes of a smooth manifold M are

defined as
wi ( M ) := wi ( TM ).

Theorem 12.3.2 (Wu). Stiefel-Whitney classes are homotopy invariants, i.e.,

if h : M1 → M2 is a homotopy equivalence then h∗ wi ( M2 ) = wi ( M1 ), for
any i ≥ 0.

Characteristic classes are particularly useful for solving a wide range

of topological problems, including the following:

(a) Given an n-dimensional smooth manifold M, find the minimal inte-

ger k such that M can be embedded/immersed in Rn+k .

(b) Given an n-dimensional smooth manifold M, is there an (n + 1)-

dimensional smooth manifold W such that ∂W = M?

(c) Given a topological manifold M, classify/find exotic smooth struc-

tures on M.

The embedding problem

Let f : Mm → N m+k be an embedding of smooth manifolds. Then

f ∗ TN = TM ⊕ ν, (12.3.1)
294 algebraic topology

where ν is the normal bundle of M in N. In particular, ν is of rank

k, hence wi (ν) = 0 for all i > k. The Whitney product formula for
Stiefel-Whitney classes, together with (12.3.1), yields that

f ∗ w ( N ) = w ( M ) ∪ w ( ν ). (12.3.2)

Note that w( M ) = 1 + w1 ( M) + · · · is invertible in H ∗ ( M; Z/2), hence

w ( ν ) = w ( M ) −1 ∪ f ∗ w ( N ).

In particular, if N = Rm+k , one gets w(ν) = w( M )−1 .

The same considerations apply in the case when f : Mm → N m+k
is required to be only an immersion. In this case, the existence of the
normal bundle ν is guaranteed by the following simple result:

Lemma 12.3.3. Let

E1
i / E2

~
π1 π2
X
be a linear monomorphism of vector bundles, i.e., in local coordinates, i is
given by U × Rn → U × Rm (n ≤ m), (u, v) 7→ (u, ℓ(u)v), where ℓ(u)
is a linear map of rank n for all u ∈ U. Then there exists a vector bundle
π1⊥ : E1⊥ → X so that π2 ∼= π1 ⊕ π1⊥ .

To summarize, we showed that if f : Mm → N m+k is an embedding

or an immersion of smooth manifolds, than one can solve for w(ν) in
(12.3.2), where ν is the normal bundle of M in N. Moreover, since ν has
rank k, we must have that wi (ν) = 0 for all i > k.
The following result of Whitney states that one can always solve for
w(ν) if the codimension k is large enough. More precisely, we have:

Theorem 12.3.4 (Whitney). Any smooth map f : Mm → N m+k is homotopic

to an embedding for k ≥ m + 1.

Let us now consider the problem of embedding (or immersing) RPm

into Rm+k . If ν is the corresponding normal bundle of rank k, we have
that w(ν) = w(RPm )−1 .
We need the following calculation:

Theorem 12.3.5.
w(RPm ) = (1 + x )m+1 , (12.3.3)

where x ∈ H 1 (RPm ; Z/2) is a generator.

Before proving Theorem 12.3.5, let us discuss some examples.

 vector bundles. characteristic classes. cobordism. applications. 295

Example 12.3.6. Let us investigate constraints on the codimension k of

an embedding of RP9 into R9+k . By Theorem 12.3.5, we have:

w(RP9 ) = (1 + x )10 = (1 + x )8 (1 + x )2 = (1 + x8 )(1 + x2 ) = 1 + x2 + x8 ,

since x10 = 0 in H ∗ (RP9 ; Z/2). Therefore,

w(RP9 )−1 = 1 + x2 + x4 + x6 .

If an embedding (or immersion) f of RP9 into R9+k exists, then w(ν) =

w−1 (RP9 ), where ν is the corresponding rank k normal bundle. In
particular, w6 (ν) ̸= 0. Since we must have wi (ν) = 0 for i > k, we
conclude that k ≥ 6. For example, this shows that RP9 cannot be
embedded into R14 .
Example 12.3.7. Similarly, if m = 2r then
r r +1 r r
w(RP2 ) = (1 + x )2 = (1 + x )2 (1 + x ) = 1 + x + x 2 .
r r +k
If there exists an embedding or immersion RP2 ,→ R2 with normal
bundle ν, then
r r −1
w(ν) = w(RP2 )−1 = 1 + x + x2 + · · · + x2 ,

hence k ≥ 2r − 1 = m − 1. In particular, RP8 cannot be immersed in

r
R14 . In this case, one can actually construct an immersion of RP2 into
r
R2 +k for any k ≥ 2r − 1, due to the following result:

Theorem 12.3.8 (Whitney). An m-dimensional smooth manifold can be

embedded in R2m and immersed in R2m−1 .

Definition 12.3.9. A smooth manifold is parallelizable if its tangent bundle

TM is trivial.

Example 12.3.10. Lie groups, hence in particular S1 and S3 , are paral-

lelizable. Moreover, S7 is parallelizable (but not a Lie group).

Theorem 12.3.5 can be used to prove the following:

Theorem 12.3.11. w(RPm ) = 1 if and only if m + 1 = 2r for some r. In
particular, if RPm is parallelizable, then m + 1 = 2r for some r.

Proof. Note that if RPm is parallelizable, then w(RPm ) = 1 since TRPm

r r
is a trivial bundle. If m + 1 = 2r , then w(RPm ) = (1 + x )2 = 1 + x2 =
1 + x m+1 = 1. On the other hand, if m + 1 = 2r k, where k > 1 is an odd
integer, we have
r r r
w(RPm ) = [(1 + x )2 ]k = (1 + x2 )k = 1 + kx2 + · · · ̸= 1,
r
since x2 ̸= 0 (indeed, 2r < 2r k = m + 1).

In fact, the following result holds:

296 algebraic topology

Theorem 12.3.12 (Adams). RPm is parallelizable if and only if m ∈

{1, 3, 7}.
Let us now get back to the proof of Theorem 12.3.5

Proof of Theorem 12.3.5. The idea is to find a splitting of (a stabilization

of) TRPm into line bundles, then to apply the Whitney sum formula.
Recall that O(1)-bundles on RPm are classified by

[RPm , BO(1)] = [RPm , K (Z/2, 1)] ∼

= H 1 (RPm ; Z/2) ∼
= Z/2.

We’ll denote by E 1 the trivial O(1)-bundle, and let π be the non-trivial

O(1)-bundle on RPm . Since O(1) ∼ = Z/2, O(1)-bundles are regular
double coverings. It is then clear that π corresponds to the 2-fold cover
Sm → RPm .
We have w(E 1 ) = 1 ∈ H ∗ (RPn ; Z/2). To calculate w(π ), we notice
that the inclusion map i : RPn → RP∞ classifies the bundle π, as
the universal bundle S∞ → RP∞ pulls back under the inclusion to
Sm → RPm . In particular,

w1 (π ) = i∗ w1 = i∗ x = x,

where x is the generator of H 1 (RP∞ ; Z/2) = H 1 (RPm ; Z/2). There-

fore,
w(π ) = 1 + x.
We next show that

TRPm ⊕ E 1 ∼
=π ⊕ ·{z
· · ⊕ π}, (12.3.4)
|
m+1 times

from which the computation of w(RPm ) follows by an application of

the Whitney sum formula.
To prove (12.3.4), start with Sm ,→ Rm+1 with (rank one) normal
bundle denoted by Eν . Note that Eν is a trivial line bundle on Sm , as it
has a global non-zero section (mapping y ∈ Sm to the normal vector νy
at y). We then have

TSm ⊕ Eν ∼
= TRm+1 |Sm = E m+1 ∼
=E 1
· · ⊕ E }1 ,
⊕ ·{z
|
m+1 times

with E m+1 the trivial bundle of rank m + 1 on Sm , i.e., the Whitney sum
of m + 1 trivial line bundles E 1 on Sm , each of which is generated by
the global non-zero section y 7→ dxd |y , i = 1, · · · , m + 1.
i
Let a : Sm → Sm be the antipodal map, with differential da : TSm →
TSm . Let γ : (−ϵ, ϵ) → Sm , γ(0) = y, v = γ′ (0) ∈ Ty Sm . Then
da(v) = dt d
( a ◦ γ(t))|t=0 = −γ′ (0) = −v ∈ Ta(y) Sm . Therefore da is an
involution on TSm , commuting with a, and hence

TSm /da = TRPm .

 vector bundles. characteristic classes. cobordism. applications. 297

Next note that the normal bundle Eν on Sm is invariant under the

antipodal action (as da(νy ) = νa(y) ), so it descends to the trivial line
bundle on RPm , i.e.,
Eν /da ∼= E 1.
Finally,

d d ∼ n
Sm × R/da ∼
= Sm × R/(y, t ) ∼ (−y, −t ) = S ×Z/2 R,
dxi dxi
which is the associated bundle of π with fiber R. So,

E 1 /da ∼
= π.

This concludes the proof of (12.3.4) and of the theorem.

Remark 12.3.13. Note that RP3 = ∼ SO(3) is a Lie group, so its tangent
bundle is trivial. In this case, once can conclude directly that w(RP3 ) =
1, but this fact can also be seen from formula (12.3.3).

Boundary Problem
For a closed smooth manifold Mn , let µ M ∈ Hn ( M, Z/2) be the funda-
mental class. We will associate to M certain Z/2-invariants, called its
Stiefel-Whitney numbers.
Definition 12.3.14. Let α = (α1 , . . . , αn ) be a tuple of non-negative integers
such that ∑in=1 iαi = n. Set

w[α] ( M ) := w1 ( M )α1 ∪ · · · ∪ wn ( M )αn ∈ H n ( M; Z/2).

The Stiefel-Whitney number of M with index α is defined as

w(α) ( M ) := ⟨w[α] ( M ), µ M ⟩ ∈ Z/2,

where ⟨−, −⟩ : H n ( M; Z/2) × Hn ( M; Z/2) → Z/2 is the Kronecker

evaluation pairing.

We have the following result:

Theorem 12.3.15 (Pontrjagin-Thom). A closed n-dimensional smooth mani-
fold M is the boundary of a smooth compact (n + 1)-dimensional manifold W
if and only if all Stiefel-Whitney numbers of M vanish.

Proof. We only show here one implication (due to Pontrjagin), namely

that if M = ∂W then w(α) ( M ) = 0, for any α = (α1 , . . . , αn ) with
∑in=1 iαi = n.
If i : M ,→ W denotes the boundary embedding, then

i∗ TW ∼
= TM ⊕ ν1 ,

where ν1 is the rank-one normal bundle of M in W.

298 algebraic topology

Assume that TW has a Euclidean metric. Then the normal bundle ν1

is trivialized by picking the inward unit normal vector at every point in
M. Hence
i∗ TW ∼
= TM ⊕ E 1 ,
where E 1 is the trivial line bundle on M. In particular, the Whitney
sum formula yields that

w k ( M ) = i ∗ w k (W ) ,

for k = 1, · · · , n, so w[α] ( M) = i∗ w[α] (W ) for any α as above.

Let µW be the fundamental class of (W, M ) i.e., the generator of
Hn+1 (W, M; Z/2), and let µ M be the fundamental class of M as above.
From the long exact homology sequence for the pair (W, M) and
Poincaré duality, we have that

∂ ( µW ) = µ M .

Let δ : H n ( M; Z/2) → H n+1 (W, M; Z/2) be the map adjoint to ∂. The

naturality of the cap product yields the identity:

⟨y, µ M ⟩ = ⟨y, ∂µW ⟩ = ⟨δy, µW ⟩

for any y ∈ H n ( M; Z/2). Putting it all together we have:

w ( α ) ( M ) = ⟨ w [ α ] ( M ), µ M ⟩
= ⟨i∗ w[α] (W ), ∂µW ⟩
= ⟨δ(i∗ w[α] (W )), µW ⟩
= ⟨0, µW ⟩
= 0,

since δ ◦ i∗ = 0, as can be seen from the long exact cohomology sequence

for the pair (W, M).

Example 12.3.16. Suppose M = X ⊔ X, i.e., M is the disjoint union

of two copies of a closed n-dimensional manifold X. Then for any α,
w(α) ( M) = 2w(α) ( X ) = 0. This is consistent with the fact that X ⊔ X is
the boundary of the cylinder X × [0, 1].

Example 12.3.17. Every RP2k−1 is a boundary. Indeed, the total Stiefel-

Whitney class of RP2k−1 is (1 + x )2k = (1 + x2 )k , with x the generator
of H 1 (RP2k−1 ; Z/2). Thus, all the odd degree Stiefel-Whitney classes
of RP2k−1 are 0. Since every monomial in the Stiefel-Whitney classes of
RP2k−1 of total degree 2k − 1 must contain a factor w j with j odd, all
Stiefel-Whitney numbers of RP2k−1 vanish. This implies the claim by
the Pontrjagin-Thom Theorem 12.3.15.
 vector bundles. characteristic classes. cobordism. applications. 299

Example 12.3.18. The real projective space RP2k is not a boundary, for
any integer k ≥ 0. Indeed, the total Stiefel-Whitney class of RP2k is
   
2k + 1 2k + 1 2k
w(RP2k ) = (1 + x )2k+1 = 1 + x+···+ x
1 2k
= 1 + x + · · · + x2k

In particular, w2k (RP2k ) = x2k . It follow that for α = (0, 0, . . . , 1) we

have
w(α) (RP2k ) = 1 ̸= 0.

12.4 Pontrjagin classes

In this section, unless specified, we use the symbol π to denote real

vector bundles (or O(n)-bundles), and use ω for complex vector bundles
(or U (n)-bundles) on a topological space X.
Given a real vector bundle π, we can consider its complexification
π ⊗ C, i.e., the complex vector bundle with same transition functions
as π:
gαβ : Uα ∩ Uβ → O(n) ⊂ U (n),
and fiber Rn ⊗ C ∼= Cn .
Given a complex vector bundle ω, we can consider its realization
ωR , obtained by forgeting the complex structure, i.e., with transition
functions
gαβ : Uα ∩ Uβ → U (n) ,→ O(2n).
Given a complex vector bundle ω, its conjugation ω is defined by
transition functions
gαβ ¯·
gαβ : Uα ∩ Uβ → U (n) → U (n),

with the second homomorphism given by conjugation. ω has the same

underlying real vector bundle as ω, but the opposite complex structure
on its fibers.
Lemma 12.4.1. If ω is a complex vector bundle, then

ωR ⊗ C ∼
= ω ⊕ ω.

Proof. Let ȷ be the linear transformation on FR ⊗ C given by multiplica-

tion by i. Here F is the fiber of complex vector bundle ω, and FR is the
fiber of its realization ωR . Then ȷ2 = −id, so we have

FR ⊗ C ∼
= Eigen(i ) ⊕ Eigen(−i ),

where ȷ acts as multiplication by i on Eigen(i ), and it acts as multipli-

cation by −i on Eigen(−i ). Moreover, we have F ⊆ Eigen(i ) and F ⊆
Eigen(−i ). By a dimension count we then get that FR ⊗ C ∼ = F ⊕ F.
300 algebraic topology

Lemma 12.4.2. Let π be a real vector bundle. Then

π⊗C ∼
= π ⊗ C.

Proof. Indeed, since the transition functions of π ⊗ C are real-values

(same as those of π), they are also the transition functions for π ⊗ C.

Lemma 12.4.3. If ω is a rank n complex vector bundle, the Chern classes of

its conjugate ω are computed by

ck (ω ) = (−1)k · ck (ω ),

for any k = 1, · · · , n.

Proof. Recall that one way to define (universal) Chern classes is by

induction by using the fibration

S2k−1 ,→ BU (k − 1) → BU (k ).

In fact,
ck = d2k ( a),

where a is the generator of H 2k−1 (S2k−1 ; Z).

The complex conjugation on the fiber S2k−1 of the above fibration
is a map of degree (−1)k (it keeps k out of 2k real basis vectors in-
variant, and it changes the sign of the other k; each sign change is a
reflection and it has degree −1). In particular, the homomorphism
H 2k−1 (S2k−1 ; Z) → H 2k−1 (S2k−1 ; Z) induced by conjugation is defined
by a 7→ (−1)k · a. Altogether, this gives ck (ω ) = (−1)k · ck (ω ).

Combining the results from Lemma 12.4.2 and Lemma 12.4.3, we

have the following:

Corollary 12.4.4. For any real vector bundle π,

ck (π ⊗ C) = ck (π ⊗ C) = (−1)k ck (π ⊗ C).

In particular, for any odd integer k, ck (π ⊗ C) is an integral cohomology class

of order 2.

Definition 12.4.5 (Pontryagin classes of real vector bundles). Let π :

E → X be a real vector bundle of rank n. The i-th Pontrjagin class of π is
defined as:
pi (π ) := (−1)i c2i (π ⊗ C) ∈ H 4i ( X; Z).

If ω a complex vector bundle of rank n, we define its i-th Pontryagin class as

pi (ω ) := pi (ωR ) = (−1)i c2i (ω ⊕ ω ).

Remark 12.4.6. Note that pi (π ) = 0 for all i > n2 .

 vector bundles. characteristic classes. cobordism. applications. 301

Definition 12.4.7. If π is a real vector bundle on X, its total Pontrjagin class

is defined as
p(π ) = p0 + p1 + · · · ∈ H ∗ ( X; Z).

Theorem 12.4.8 (Product formula). If π1 and π2 are real vector bundles on

X, then

p(π1 ⊕ π2 ) = p(π1 ) ∪ p(π2 ) mod 2-torsion.

Proof. We have (π1 ⊕ π2 ) ⊗ C ∼

= (π1 ⊗ C) ⊕ (π2 ⊗ C). Therefore,

pi (π1 ⊕ π2 ) = (−1)i c2i ((π1 ⊕ π2 ) ⊗ C)

= (−1)i ∑ c k ( π1 ⊗ C) ∪ c l ( π2 ⊗ C)
k +l =2i

= (−1)i ∑ c2a (π1 ⊗ C) ∪ c2b (π2 ⊗ C) + {elements of order 2}

a + b =i

= ∑ p a (π1 ) ∪ pb (π2 ) + {elements of order 2},

a + b =i

thus proving the claim.

Definition 12.4.9. If M is a real smooth manifold, we define

p( M) := p( TM).

If M is a complex manifold, we define

p( M ) := p(( TM)R ).

Here TM is the tangent bundle of the manifold M.

In order to give applications of Pontrjagin classes, we need the

following computational result:

Theorem 12.4.10 (Chern and Pontrjagin classes of complex projective

space). The total Chern and Pontrjagin classes of the complex projective space
CPn are computed by:

c(CPn ) = (1 + c)n+1 , (12.4.1)

p(CPn ) = (1 + c2 )n+1 , (12.4.2)

where c ∈ H 2 (CPn ; Z) is a generator.

Proof. The arguments involved in the computation of c(CPn ) are very

similar to those of Theorem 12.3.5. Indeed, one first shows that there is
a splitting
TCPn ⊕ E 1 = γ ⊕ · · · ⊕ γ,
| {z }
n+1 times
302 algebraic topology

were E 1 is the trivial complex line bundle on CPn and γ is the complex
line bundle associated to the principle S1 -bundle S1 ,→ S2n+1 → CPn .
Then γ is classified by the inclusion

 / S∞
S2n+1

  
CPn  / CP∞ = BU (1)

and hence c1 (γ) = c, the generator of H 2 (CP∞ ; Z) = H 2 (CPn ; Z). The

Whitney sum formula for Chern classes then yields:

c(CPn ) = c( TCPn ) = c(γ)n+1 = (1 + c)n+1 .

By conjugation, one gets

c( TCPn ) = (1 − c)n+1 .

Therefore,

c(( TCPn )R ⊗ C) = c( TCPn ⊕ TCPn )

= c( TCPn ) ∪ c( TCPn )
= (1 − c 2 ) n +1 ,

from which one can readily deduce that p(CPn ) = (1 + c2 )n+1 .

Applications to the embedding problem

After forgetting the complex structure, CPn is a 2n-dimensional real
smooth manifold. Suppose that there is an embedding

CPn ,→ R2n+k ,

and we would like to find constraints on the embedding codimension k

by means of Pontrjagin classes.
Let ( TCPn )R be the realization of the tangent bundle for CPn . Then
the existence of an embedding as above implies that there exists a
normal (real) bundle νk of rank k such that

( TCPn )R ⊕ νk ∼
= TR2n+k |CPn ∼
= E 2n+k , (12.4.3)

with E 2n+k denoting the trivial real vector bundle of rank 2n + k.

By applying the Pontrjagin class p to (12.4.3) and using the product
formula of Theorem 12.4.8 together with the fact that there are no
elements of order 2 in H ∗ (CPn ; Z), we have

p(CPn ) · p(νk ) = 1.
 vector bundles. characteristic classes. cobordism. applications. 303

Therefore, we get
p(νk ) = p(CPn )−1 . (12.4.4)
And by the definition of the Pontryagin classes, we know that if
pi (νk ) ̸= 0, then i ≤ 2k .
Example 12.4.11. In this example, we use Pontrjagin classes to show that
CP2 does not embed in R5 . First,

p(CP2 ) = (1 + c2 )3 = 1 + 3c2 ,

with c ∈ H 2 (CP2 ; Z) a generator (hence c3 = 0). If there is an embedding

CP2 ,→ R4+k with normal bundle νk , then

p(νk ) = p(CP2 )−1 = 1 − 3c2 .

Hence p1 (νk ) ̸= 0, which implies that k ≥ 2.

12.5 Oriented cobordism and Pontrjagin numbers

If M is a smooth oriented manifold, we denote by − M the same mani-

fold but with the opposite orientation.

Definition 12.5.1. Let Mn and N n be smooth, closed, oriented real manifolds

of dimension n. We say M and N are oriented cobordant if there exists a
smooth, compact, oriented (n + 1)-dimensional manifold W n+1 , such that
∂W = M ⊔ (− N ).
Remark 12.5.2. Let us say a word of convention about orienting a
boundary. For any x ∈ ∂W, there exist an inward normal vector ν+ ( x )
and an outward normal vector ν− ( x ) to the boundary at x. By using
a partition of unity, one can construct an inward/outward normal
vector field ν± : ∂W → TW |∂W . By convention, a frame {e1 , · · · , en } on
Tx (∂W ) is positive if {e1 , · · · , en , ν− ( x )} is a positive frame for Tx W.

Lemma 12.5.3. Oriented cobordism is an equivalence relation.

Proof. M is clearly oriented cobordant to itself because M ⊔ (− M ) is

diffeomorphic to the boundary of M × [0, 1]. Hence oriented cobordism
is reflexive. The symmetry can be deduced from the fact that, if M ⊔
(− N ) ≃ ∂W, then N ⊔ (− M) ≃ ∂(−W ). Finally, if M1 ⊔ (− M2 ) ≃ ∂W,
and M2 ⊔ (− M3 ) ≃ ∂W ′ , then we can glue W and W ′ along the common
boundary and get a new manifold with boundary M1 ⊔ (− M3 ). Hence
oriented cobordism is also transitive.

Definition 12.5.4. Let Ωn be the set of cobordism classes of closed, oriented,

smooth n-manifolds.
Corollary 12.5.5. The set Ωn is an abelian group with the disjoint union
operation.
304 algebraic topology

Proof. This is an immediate consequence of Lemma 12.5.3. The zero

element in Ωn is the class of ∅, or equivalently, [ M] = 0 ∈ Ωn if and
only if M = ∂W, for some compact manifold W. The inverse of [ M ] is
[− M], since M ⊔ (− M) is a boundary.

A natural problem to investigate is to describe the group Ωn by

generators and relations. For example, both [CP4 ] and [CP2 × CP2 ]
are elements of Ω8 . Do they represent the same element, i.e., are CP4
and CP2 × CP2 oriented cobordant? A lot of insight is gained by using
Pontrjagin numbers.
Definition 12.5.6. Let Mn be a smooth, closed, oriented real n-manifold,
with fundamental class µ M ∈ Hn ( M; Z). Let k = [ n4 ] and choose a partition
α = (α1 , · · · , αk ) ∈ Zk such that ∑kj=1 4jα j = n. The Pontrjagin number of
M associated to the partition α is defined as:

p(α) ( M) = ⟨ p1 ( M)α1 ∪ · · · ∪ pk ( M)αk , µ M ⟩ ∈ Z.

Remark 12.5.7. If n is not divisible by 4, then p(α) ( M ) = 0 by definition.

Theorem 12.5.8. For n = 4k, each p(α) defines a homomorphism

Ωn −→ Z, [ M ] 7→ p(α) ( M ).

Hence oriented cobordant manifolds have the same Pontrjagin numbers. In

particular, if Mn = ∂W n+1 , then p(α) ( M ) = 0 for any partition α.

Proof. If M = M1 ⊔ M2 , then [ M] = [ M1 ] + [ M2 ] ∈ Ωn and µ M =

µ M1 + µ M2 ∈ Hn ( M; Z). It follows readily that p(α) ( M) = p(α) ( M1 ) +
p(α) ( M2 ).
If M = ∂N, then it can be shown as in the proof of Theorem 12.3.15
that p(α) ( M ) = 0 for any partition α.

Example 12.5.9. By Theorem 12.4.10, we have that p(CP2n ) = (1 +

c2 )2n+1 , where c is a generator of H 2 (CP2n ; Z). Hence pi (CP2n ) =
(2ni+1)c2i . For the partition α = (0, . . . , 0, 1), we find that p(α) (CP2n ) =
D E
(2nn+1)c2n , µCP2n = (2nn+1) ̸= 0. We conclude that CP2n is not an
oriented boundary.
Remark 12.5.10. If we reverse the orientation of a manifold M of real
dimension n = 4k, the Pontrjagin classes remain unchanged, but the
fundamental class µ M changes sign. Therefore, all Pontrjagin numbers
p(α) ( M) change sign. This shows that, if some Pontrjagin number
p(α) ( M) is nonzero, then M cannot have any orientation-reversing
diffeomorphism.

Example 12.5.11. The above remark and Example 12.5.9 show that
CP2n does not have any orientation-reversing diffeomorphism. How-
ever, CP2n+1 has an orientation-reversing diffeomorphism induced by
complex conjugation.
 vector bundles. characteristic classes. cobordism. applications. 305

Example 12.5.12. Let us consider Ω4 . As CP2 is not an oriented

boundary by Example 12.5.9, we have [CP2 ] ̸= 0 ∈ Ωn . Recall that
p(CP2 ) = 1 + 3c2 , so p1 (CP2 ) = 3c2 . For the partition α = (1), we then
get that p(1) (CP2 ) = 3. So
p (1)
Ω4 −→ 3Z −→ 0

is exact, thus rank(Ω4 ) ≥ 1.

Example 12.5.13. We next consider Ω8 . The partitions to work with in
this case are α1 = (2, 0) and α2 = (0, 1), and Theorem 12.5.8 yields a
homomorphism
( p(α ) ,p(α ) )
Ω8 −−−−−−−
→ Z ⊕ Z.
1 2

We aim to show that

rank(Ω8 ) = dimQ (Ω8 ⊗ Q) ≥ 2.

We start by noting that both CP4 and CP2 × CP2 are compact oriented
8-manifolds which are not boundaries. We calculate the Pontrjagin
numbers of these two spaces. First,

p(CP4 ) = (1 + c2 )5 = 1 + 5c2 + 10c4 ,

where c is a generator of H 2 (CP4 ; Z). Hence, p1 (CP4 ) = 5c2 and

p2 (CP4 ) = 10c4 . The Pontrjagin numbers of CP4 corresponding to the
partitions α1 = (2, 0) and α2 = (0, 1) are given as:

p(α1 ) (CP4 ) = ⟨ p1 (CP4 )2 , µCP4 ⟩ = 25,

p(α2 ) (CP4 ) = ⟨ p2 (CP4 ), µCP4 ⟩ = 10.

In order to compute the corresponding Pontrjagin numbers for CP2 ×
CP2 , let pri : CP2 × CP2 → CP2 , i = 1, 2, be the projections on factors.
Then
T (CP2 × CP2 ) ∼
= pr1∗ T (CP2 ) ⊕ pr2∗ T (CP2 ),
so Theorem 12.4.8 yields that

p(CP2 × CP2 ) = pr1∗ p(CP2 ) ∪ pr2∗ p(CP2 ) = p(CP2 ) × p(CP2 ),

where × denotes the external product. Let c1 and c2 denote the genera-
tors of the second integral cohomology of the two CP2 factors. Then:

p(CP2 × CP2 ) = (1 + c21 )3 · (1 + c22 )3 = (1 + 3c21 ) · (1 + 3c22 )

= 1 + 3c21 + 3c22 + 9c21 c22 .

Hence, p1 (CP2 × CP2 ) = 3(c21 + c22 ) and p2 (CP2 × CP2 ) = 9c21 c22 . There-
fore, the Pontrjagin numbers of CP2 × CP2 corresponding to the parti-
tions α1 and α2 are computed by (here we use the fact that c41 = 0 = c42 ):

p(α1 ) (CP2 × CP2 ) = 18, p(α2 ) (CP2 × CP2 ) = 9.

306 algebraic topology

The values of the homomorphism ( p(α1 ) , p(α2 ) ) : Ω8 −→ Z ⊕ Z on

" #
25 18
CP4 and CP2 × CP2 fit into the 2 × 2 matrix with nonzero
10 9
determinant. Hence rank(Ω8 ) ≥ 2.

More generally, we the following qualitative description of Ωn , which

we mention here without proof.

Theorem 12.5.14 (Thom). The oriented cobordism group Ωn is finitely

generated of rank | I |, where I is the collection of partitions α satisfying
∑ j 4jα j = n. In fact, modulo torsion, Ωn is generated by products of even
p(α) : Ωn → Z| I | is an injective
L
complex projective spaces. Moreover,
α∈ I
homomorphism onto a subgroup of the same rank.

Example 12.5.15. Our computations from Examples 12.5.12 and 12.5.13

together with Theorem 12.5.14 yield that in fact we have: rank(Ω4 ) = 1
and rank(Ω8 ) = 2.

12.6 Signature as an oriented cobordism invariant

Recall that if M4k is a closed, oriented manifold of real dimension

n = 4k, then we can define its signature σ ( M ) as the signature of the
bilinear symmetric pairing

H 2k ( M; Q) × H 2k ( M; Q) → Q,

which is non-degenerate by Poincaré duality. Recall also that if M

is an oriented boundary then σ( M) = 0. This suffices to deduce the
following result:

Theorem 12.6.1 (Thom). σ : Ω4k → Z is a homomorphism.

It follows from Theorems 12.5.14 and 12.6.1 that the signature is a

rational combination of Pontrjagin numbers, i.e.,

σ= ∑ aα p(α) (12.6.1)
α∈ I

for some coefficients aα ∈ Q. The Hirzebruch signature theorem provides

an explicit formula for these coefficients aα . In what follows we compute
by hand the coefficients aα in the cases of Ω4 and Ω8 .

Example 12.6.2. On closed oriented 4-manifolds, the signature is com-

puted by
σ = ap(1) , (12.6.2)

with a ∈ Q to be determined. Since a is the same for any [ M ] ∈ Ω4 ,

we will determine it by performing our calculations on M = CP2 .
Recall that σ (CP2 ) = 1, and if c ∈ H 2 (CP2 ; Z) is a generator then
 vector bundles. characteristic classes. cobordism. applications. 307

p1 (CP2 ) = 3c2 . Hence p(1) (CP2 ) = 3, and (12.6.2) implies that 1 = 3a,
or a = 31 . Therefore, for any closed oriented 4-manifold M4 we have
that
1 1
σ ( M ) = ⟨ p1 ( M ), µ M ⟩ = p(1) ( M ) ∈ Z.
3 3
Example 12.6.3. On closed oriented 8-manifolds, the signature is com-
puted by (12.6.1) as

σ = a(2,0) p(2,0) + a(0,1) p(0,1) , (12.6.3)

with a(2,0) , a(0,1) ∈ Q to be determined. Since Ω8 is generated rationally

by CP4 and CP2 × CP2 , we can calculate a(2,0) and a(0,1) by evaluating
(12.6.3) on CP4 and CP2 × CP2 . Using our computations from Example
12.5.13, we have:

1 = σ (CP4 ) = 25a(2,0) + 10a(0,1) , (12.6.4)

and
1 = σ (CP2 × CP2 ) = 18a(2,0) + 9a(0,1) . (12.6.5)
Solving for a(2,0) and a(0,1) in (12.6.4) and (12.6.5), we get:

1 7
a(2,0) = − , a(0,1) = .
45 45
Altogether, the signature of a closed oriented manifold M8 is computed
by the following formula:
1
σ ( M8 ) = ⟨7p2 ( M) − p1 ( M)2 , µ M ⟩. (12.6.6)
45

12.7 Exotic 7-spheres

Now we turn to the construction of exotic 7-spheres. Start with M

a smooth, 3-connected orientable 8-manifold. Up to homotopy, M ≃
(S4 ∨ · · · ∨ S4 ) ∪ f e8 . Assume further that β 4 ( M) = 1, i.e., M ≃ S4 ∪ f e8 ,
for some map f : S7 → S4 . By Whitney’s embedding theorem, there is
a smooth embedding S4 ,→ M. Let E be a tubular neighborhood of this
embedded S4 in M; in other words, E is a D4 -bundle on S4 inside M.
Such D4 -bundles on S4 are classified by

= π3 ( S3 × S3 ) ∼
π3 (SO(4)) ∼ = Z ⊕ Z.

(Here we use the fact that S3 × S3 is a 2-fold covering of SO(4).) That

means that E corresponds to a pair of integers (i, j).
Let X 7 be the boundary of E, so X is a S3 -bundle over S4 . If X is
diffeomorphic to a 7-sphere, one can recover M from E by attaching
an 8-cell to X = ∂E. So the question to investigate is: for which pairs of
integers (i, j) is X diffeomorphic to S7 ?
One can show the following:
308 algebraic topology

Lemma 12.7.1. X is homotopy equivalent to S7 if and only if i + j = ±1.

Suppose i + j = 1. Then for each choice of i, we get an S3 -bundle
over S4 , namely X = ∂E, which has the homotopy type of S7 . If X is in
fact diffeomorphic to S7 , then we can recover M by attaching an 8-cell
to X, and in this case the signature of M is computed by

1  
σ( M) = 7p(0,1) ( M ) − p(2,0) ( M ) .
45
Moreover, one can show that:
Lemma 12.7.2. p(2,0) ( M ) = 4(i − j)2 = 4(2i − 1)2 .
Note that σ ( M ) = ±1 since H 4 ( M; Z) = Z, and let us fix the
orientation according to which σ ( M ) = 1. Our assumption that X was
diffeomorphic to S7 leads now to a contradiction, since

4(2i − 1)2 + 45
p(0,1) ( M ) =
7
is by definition an integer for all i, which is contradicted e.g., for i = 2.
So far (for i = 2 and j = −1), we constructed a space X which is
homotopy equivalent to S7 , but which is not diffeomorphic to S7 . In
fact, one can further show the following:

Lemma 12.7.3. X is homeomorphic to S7 , so in fact X is an exotic 7-sphere.

This latest fact can be shown by constructing a Morse function
g : X → R with only two nondegenerate critical points (a maximum
and a minimum). An application of Reeb’s theorem then yields that X
is homeomorphic to S7 .

12.8 Exercises

1. Construct explicitly the bounding manifold for RP3 .

2. Let ω be a rank n complex vector bundle on a topological space X,

and let ci (ω ) ∈ H 2i ( X; Z) be its i-th Chern class. Via Z → Z/2, ci (ω )
determines a cohomology class c̄i (ω ) ∈ H 2i ( X; Z/2). By forgetting the
complex structure on the fibers of ω, we obtain the realization ωR of ω,
as a rank 2n real vector bundle on X.
Show that the Stiefel-Whitney classes of ωR are computed as follows:

(a) w2i (ωR ) = c̄i (ω ), for 0 ≤ i ≤ n.

(b) w2i+1 (ωR ) = 0 for any integer i.

3. Let M be a 2n-dimensional smooth manifold with tangent bundle

TM. Show that, if M admits a complex structure, then w2i ( M) is
 vector bundles. characteristic classes. cobordism. applications. 309

the mod 2 reduction of an integral class for any 0 ≤ i ≤ n, and

w2i+1 ( M ) = 0 for any integer i. In particular, Stiefel-Whitney classes
give obstructions to the existence of a complex structure on an even-
dimensional real smooth manifold.

4. Show that a real smooth manifold M is orientable if and only if

w1 ( M ) = 0.

5. Show that CP3 does not embed in R7 .

6. Show that CP4 does not embed in R11 .

7. Example 12.5.9 shows that CP2 is not the boundary on an oriented

compact 5-manifold. Can CP2 be the boundary on some non-orientable
compact 5-manifold?

8. Show that CP2n+1 is the boundary of a compact manifold.

 bibliography 311

Bibliography

Raoul Bott and Loring W. Tu. Differential forms in algebraic topology,

volume 82 of Graduate Texts in Mathematics. Springer-Verlag, New
York-Berlin, 1982.

Glen E. Bredon. Topology and geometry, volume 139 of Graduate Texts

in Mathematics. Springer-Verlag, New York, 1993.

James F. Davis and Paul Kirk. Lecture notes in algebraic topology, vol-
ume 35 of Graduate Studies in Mathematics. American Mathematical
Society, Providence, RI, 2001.

Allen Hatcher. Algebraic topology. Cambridge University Press, Cam-

bridge, 2002.

William S. Massey. A basic course in algebraic topology, volume 127

of Graduate Texts in Mathematics. Springer-Verlag, New York, 1991.

John W. Milnor and James D. Stasheff. Characteristic classes. Princeton

University Press, Princeton, N. J.; University of Tokyo Press, Tokyo,
1974. Annals of Mathematics Studies, No. 76.

James R. Munkres. Elements of algebraic topology. Addison-Wesley

Publishing Company, Menlo Park, CA, 1984.

James R. Munkres. Topology. Prentice Hall, Inc., Upper Saddle River,

NJ, 2000. Second edition of [ MR0464128].

Edwin H. Spanier. Algebraic topology. McGraw-Hill Book Co., New

York-Toronto, Ont.-London, 1966.
 INDEX 313

Index

K=∼ P2 , 39 characteristic map, 90

Pn , 36 Chern class, 286, 289, 290
Tn , 36 classification of surfaces, 40
π5 (S3 ), 257 classifying map, 290
πn+1 (Sn ), 253 classifying space, 282
n-cell, 90 cobordism, 187
cobordism group, 303
abelian space, 194 coboundary, 121
antipodal map, 85 coboundary map, 117, 121
aspherical, 195 cochain complex, 117
atlas, 269 cocycle, 121
attaching map, 90
cohomology group, 118, 121
augmentation map, 65
cohomology long exact
sequence, 124
Betti number, 102
cohomology ring, 137
Borsuk-Ulam theorem, 144
cohomology with compact
Bott periodicity, 226
support, 163
boundary, 63, 64
complex projective space, 91
boundary map, 64
complexification, 299
bouquet, 28
compression lemma, 208
Brower’s fixed point theorem,
concatenation, 5
1, 16, 76
connected sum, 34, 163, 188
connecting homomorphism, 71
cap product, 157, 166
cellular approximation, 105, contractible, 10
201 covering map, 47
cellular chain complex, 93 covering transformation, 52, 53
cellular cochain complex, 127 cross product, 149
cellular cohomology, 128 cup product, 133, 134
cellular homology, 93 cup product pairing, 172, 181
cellular map, 105, 201 CW approximation, 211
chain homotopy, 69 CW complex, 90
chain map, 66 cycle, 64
314 algebraic topology

deck group, 53 homotopy long exact sequence,

deck transformation, 53 198
deformation retract, 12 homotopy long exact sequence
degree, 84, 162 of a fibration, 220
dimension, 90, 158 homotopy type, 10
Hopf fibration, 223
edge homomorphism, 239 Hopf invariant, 147
Eilenberg-MacLane space, 215 Hurewicz homomorphism, 196
elementary reduction, 20, 22 Hurewicz theorem, 196, 218,
Euler characteristic, 80, 101 234
excision, 73, 126, 203
exotic sphere, 308 induced bundle, 274
Ext group, 118 infinite earring, 55
exterior algebra, 150 intersection pairing, 173, 183

face, 63 Künneth exact sequence, 154

fiber bundle, 221, 270 Künneth formula, 149, 154, 155
fibration, 220 Klein bottle, 33, 78
free action, 58 Kronecker pairing, 132
free group, 20
free product, 22 labeling scheme, 31
free resolution, 112 Lefschetz number, 104
fundamental class, 159 lens space, 60, 250
fundamental group, 5 Leray-Hirsch theorem, 248
fundamental theorem of Leray-Serre spectral sequence,
algebra, 17 233, 247
lift extension property, 220
good pair, 79 lifting lemma, 51
Grassmann manifold, 225 local homology group, 76
Gysin sequence, 242 local orientation, 158
loop space, 3, 229
homology long exact sequence
of a pair, 72 Möbius band, 39, 54
homotopic maps, 9 manifold, 31, 158
homotopic paths, 3 manifold with boundary, 184
homotopy, 3, 9 mapping cylinder, 209
homotopy equivalent, 9 Mayer-Vietoris sequence, 77,
homotopy extension property, 126
200
homotopy group, 191 nullhomotopic, 10
homotopy lifting, 13
homotopy lifting property, 48, orientation, 158, 177
219 oriented cobordism, 303
 INDEX 315

parallelizable, 295 Serre theorem, 260

path fibration, 228 set of words, 19
path lifting, 13 signature, 173, 187, 306
path lifting property, 48 simply-connected, 11
path-space, 228 singular chain complex, 64
pinched torus, 183 singular homology, 64
Poincaré Duality, 169, 185 singular simplex, 64
Pontrjagin number, 304 spectral sequence, 231
Pontrjagin-Thom theorem, 297 sphere, 32
Pontryagin class, 300 standard simplex, 63
Postnikov approximation, 253 Stiefel manifold, 224
Postnikov tower, 214 Stiefel-Whitney class, 286, 292
principal bundle, 276 Stiefel-Whitney number, 297
prism operator, 68, 125 surface, 31
projective plane, 33 suspension, 74, 85
properly discontinuous action, suspension theorem, 74, 203
58
Puppé sequence, 229 tensor product, 106
tor groups, 112
quaternionic projective space, torus, 32
147 transgression, 239
transition function, 270
reduced cohomology, 123
reduced homology, 66 universal coefficient theorem,
reduced word, 20, 22 111, 119, 155
reflection, 84 universal cover, 54
regular covering, 57 universal fiber bundle, 223
relative chains, 70 universal mapping property, 20,
relative cohomology, 123 22, 24
relative homology, 70
relative homotopy group, 197 Wang sequence, 243
retract, 12 weak homotopy equivalence,
207
Seifert-Van Kampen theorem, wedge, 28
26 Whitehead theorem, 207
semi-locally simply connected, Whitehead tower, 214, 256
54 Whitney sum, 291