Algebraic Topology

Summer term 2016

Christoph Schweigert
Hamburg University
Department of Mathematics
Section Algebra and Number Theory
and
Center for Mathematical Physics
(as of: 16.7.2018)

Contents
1 Homology theory 1
1.1 Chain complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Singular homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The homology groups H0 and H1 . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Homotopy invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 The long exact sequence in homology . . . . . . . . . . . . . . . . . . . . . . . . 17
1.6 The long exact sequence of a pair of spaces . . . . . . . . . . . . . . . . . . . . . 22
1.7 Excision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.8 Mayer-Vietoris sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.9 Reduced homology and suspension . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.10 Mapping degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.11 CW complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.12 Cellular homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
1.13 Homology with coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.14 Tensor products and the universal coefficient theorem . . . . . . . . . . . . . . . 62
1.15 The topological Künneth formula . . . . . . . . . . . . . . . . . . . . . . . . . . 72

2 Singular cohomology 77
2.1 Definition of singular cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.2 Universal coefficient theorem for cohomology . . . . . . . . . . . . . . . . . . . . 79
2.3 Axiomatic description of a cohomology theory . . . . . . . . . . . . . . . . . . . 82
2.4 Cap product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.5 Cup product on cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.6 Orientability of manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.7 Cohomology with compact support . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.8 Poincaré duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.9 Alexander-Lefschetz duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

1
 2.10 Application of duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
2.11 Duality and cup products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
2.12 The Milnor sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
2.13 Lens spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.14 A first quick glance on homotopy theory . . . . . . . . . . . . . . . . . . . . . . 130

A English - German glossary 135

Literature:
Some of the literature I used to prepare the course:
A. Hatcher, Algebraic Topology
Cambridge University Press, 2002.
G. Bredon, Topology and Geometry.
Springer Graduate Text in Mathematics 139, Springer, New York, 2010
R. Stöcker, H. Zieschang, Algebraische Topologie.
Teubner, Stuttgart, 1994

The current version of these notes can be found under

http://www.math.uni-hamburg.de/home/schweigert/ss16/atskript.pdf
as a pdf file.
Please send comments and corrections to [email protected]!

These notes are based on lectures delivered at the University of Hamburg in the summer
term 2016. I would like to thank Birgit Richter for making her notes available to me. The
present notes are indeed just a minor extension of the notes for her course. I am grateful to F.
Bartelmann, V. Koppen, A. Lahtinen, M. Manjunatha, S. Bunk, L. Wienke and L. Woike for
helpful comments.

2
1 Homology theory
We recall a few facts about the fundamental group:
• It assigns to any path connected topological space X with a base point x ∈ X an algebraic
object, a group π1 (X, x). The assignment is functorial, i.e. for a continuous map f : X →
Y we get a group homomorphism f∗ : π(X, x) → π1 (Y, f (x)). Homotopic maps f ' g
induce the same maps on the fundamental group, f∗ = g∗ .

• The fundamental group is computable invariant, most notably due to the theorem of
Seifert-van Kampen.

• The invariant crucially enters in covering theory: if a topological space X is sufficiently

connected, the equivalence classes of path-connected coverings are classified by conjugacy
classes of subgroups of π1 (X).

• However, for CW complexes, it is insensitive to n-cells with n > 3. As a consequence, it

cannot distinguish spheres Sn for different n > 2.
A possible remedy is to consider continuous maps I n → M , with I = [0, 1], up to homo-
topy relative boundary. But the corresponding homotopy groups πn (M ) are difficult to
compute, even for spaces as fundamental as spheres. For example, for the 2-sphere πn (S2 )
is non-zero, although the 2-sphere does not have cells in dimensions greater than 2.
Homology is a computable algebraic invariant that is sensitive to higher cells as well; but it
takes some effort to define it. In particular, we will have rather huge objects in intermediate
steps to which we turn now:

1.1 Chain complexes

Homology is defined using algebraic objects called chain complexes.

Definition 1.1.1
A chain complex is a sequence of abelian groups, (Cn )n∈Z , together with homomorphisms
dn : Cn → Cn−1 for n ∈ Z, such that dn−1 ◦ dn = 0.
Let R be an (associative) ring with unit 1 R . A chain complex of R-modules can analogously
be defined as a sequence of R-modules (Cn )n∈Z with R-linear maps dn : Cn → Cn−1 such that
dn−1 ◦ dn = 0.

Definition 1.1.2
We fix the following terminology:
• The homomorphisms dn are called differentials or boundary operators.

• The elements x ∈ Cn are called n-chains.

• Any x ∈ Cn such that dn x = 0 is called an n-cycle. We denote the group of n-cycles by

Zn (C) := ker(dn ) = {x ∈ Cn | dn x = 0}.

• Any x ∈ Cn of the form x = dn+1 y for some y ∈ Cn+1 is called an n-boundary.

Bn (C) := Im(dn+1 ) = {dn+1 y, y ∈ Cn+1 }.

1
The cycles and boundaries form subgroups of the group of chains. The identity dn ◦ dn+1 = 0
implies that the image of dn+1 is a subgroup of the kernel of dn and thus

Bn (C) ⊂ Zn (C).

We often drop the subscript n from the boundary maps and just write C∗ for the chain
complex.

Definition 1.1.3
The abelian group Hn (C) := Zn (C)/Bn (C) is called the nth homology group of the complex
C∗ .
We denote by [c] ∈ Hn (C) the equivalence class of a cycle c ∈ Zn (C). If c, c0 ∈ Cn are such
that c − c0 is a boundary, then c is said to be homologous to c0 . This defines an equivalence
relation on chains. A complex is called acyclic, if its homology except in degree 0 vanishes.
Examples 1.1.4.
1. Consider the complex with (
Z n = 0, 1
Cn =
0 otherwise
Here, the only non-zero differential is d1 ; let it be the multiplication with N ∈ N, then
(
Z/N Z n = 0
Hn (C) =
0 otherwise.

2. Take Cn = Z for all n ∈ Z and consider differentials

(
idZ n odd
dn =
0 n even.

The homology of this chain complex vanishes in all degrees.

3. Consider Cn = Z for all n ∈ Z again, but let all boundary maps be trivial. The homology
of this chain complex equals Z in all degrees.
We need morphisms of chain complexes:

Definition 1.1.5
Let C∗ and D∗ be two chain complexes. A chain map f : C∗ → D∗ is a sequence of homomor-
phisms fn : Cn → Dn such that dD C
n ◦ fn = fn−1 ◦ dn for all n, i.e., the diagram

dC
n
Cn / Cn−1
fn fn−1
 dD 
n
Dn / Dn−1

commutes for all n.

A chain map f sends cycles to cycles, since

dD C
n fn (c) = fn−1 (dn c) = 0 for a cycle c ,

2
and boundaries to boundaries, since

fn (dC D
n+1 λ) = dn+1 fn+1 (λ) .

We therefore obtain an induced map of homology groups

Hn (f ) : Hn (C) → Hn (D)

via Hn (f )[c] = [fn c].

Examples 1.1.6.
1. There is a chain map from the chain complex mentioned in Example 1.1.4.1 to the chain
complex D∗ that is concentrated in degree zero and has D0 = Z/N Z.

/Z ∙N /Z /0
0

   
0 / 0 / ZN /0

Note that (f0 )∗ is an isomorphism on the zeroth homology group; all homology groups
are isomorphic.
2. Are there chain maps between the complexes from Examples 1.1.4.2. and 1.1.4.3?
Lemma 1.1.7.
If f : C∗ → D∗ and g : D∗ → E∗ are two chain maps, then Hn (g) ◦ Hn (f ) = Hn (g ◦ f ) for all n.
We next study situations in which two chain maps induce the same map on homology.

Definition 1.1.8
A chain homotopy H between two chain maps f, g : C∗ → D∗ is a sequence of homomorphisms
(Hn )n∈Z with Hn : Cn → Dn+1 such that for all n

dD C
n+1 ◦ Hn + Hn−1 ◦ dn = fn − gn .

dC dC dC dC
... n+2
/ Cn+1 n+1
/ Cn n / Cn−1 n−1
/ ...
o ooo o
Hn+1oooo H oooo Hn−1 oooo
oo f n+1 gn+1 on o f n gn ooo f n−1 gn−1
ooooodD 
w
o ooodD  w
o oooo dD 
dD
. . .wo n+2
/ Dn+1 n+1
/D
n
n / Dn−1 n−1
/ ...
If such an H exists, then the chain maps f and g are said to be (chain) homotopic. We
write f ' g.
We will see in Section 1.4 geometrically defined examples of chain homotopies.
Proposition 1.1.9.
1. Being chain homotopic is an equivalence relation on chain maps.
2. If f and g are homotopic, then Hn (f ) = Hn (g) for all n.

Proof.
1. If H is a homotopy from f to g, then −H is a homotopy from g to f . Each chain map f
is homotopic to itself with chain homotopy H = 0. If f is homotopic to g via H and g is
homotopic to h via K, then f is homotopic to h via H + K.

3
 2. We have for every cycle c ∈ Zn (C∗ ):

Hn (f )[c] − Hn (g)[c] = [fn c − gn c] = [dD C

n+1 ◦ Hn (c)] + [Hn−1 ◦ dn (c)] = 0.

Here, the class of the first term vanishes; in the second term dC
n c = 0, since c is a cycle.

Definition 1.1.10
Let f : C∗ → D∗ be a chain map. We call f a chain homotopy equivalence, if there exists a
chain map g : D∗ → C∗ such that g ◦ f ' idC∗ and f ◦ g ' idD∗ . The chain complexes C∗ and
D∗ are said to be chain homotopically equivalent.
Chain homotopically equivalent chain complexes have isomorphic homology. However, chain
complexes with isomorphic homology do not have to be chain homotopically equivalent, cf.
Example 1.1.6.1: there is no non-zero morphism of abelian groups ZN → Z.

Definition 1.1.11
If C∗ and C∗0 are chain complexes, then their direct sum, C∗ ⊕ C∗0 , is the chain complex with

(C∗ ⊕ C∗0 )n = Cn ⊕ Cn0 = Cn × Cn0

with differential d = d ⊕ d0 given by

d⊕ (c, c0 ) = (dc, d0 c0 ).

(j)
Similarly, if (C∗ , d(j) )j∈J is a family of chain complexes, then we can define their direct
sum as follows:
M M
( C∗(j) )n := Cn(j)
j∈J j∈J

as abelian groups and the differential d⊕ is defined via the property that its restriction to the
jth summand is d(j) .

1.2 Singular homology

In the definition of the fundamental group, we test a topological space X by (homotopy classes
of) maps S1 → X. In the definition of singular homology, we use maps from higher-dimensional
objects, simplices.
Let v0 , . . . , vn be n + 1 points in Rn+1 . Consider the convex hull
n
X n
X
K(v0 , . . . , vn ) := { ti vi | ti = 1, ti > 0} ⊂ Rn+1 .
i=0 i=0

Definition 1.2.1
If the vectors v1 − v0 , . . . , vn − v0 are linearly independent, then K(v0 , . . . , vn ) is the simplex
generated by v0 , . . . , vn . We denote such a simplex by simp(v0 , . . . , vn ).
Note that simplex really means “simplex with an ordering of its vertices”.

4
Example 1.2.2.
1. Denote by ei ∈ Rn+1 the vector that has an entry 1 in coordinate i + 1 and is zero in all
other coordinates. The standard topological n-simplex is Δn := simp(e0 , . . . , en ).
2. The first examples of standard topological simplices are:
• Δ0 is the point e0 = 1 ∈ R.
• Δ1 is the line segment in R2 between e0 = (1, 0) ∈ R2 and e1 = (0, 1) ∈ R2 .
• Δ2 is a triangle in R3 with vertices e0 , e1 and e2 and Δ3 is homeomorphic to a
tetrahedron.
3. The coordinate description of the standard n-simplex in Rn+1 is
n
X
Δ = {(t0 , . . . , tn ) ∈ R
n n+1
| ti = 1, ti > 0}.
i=1

We consider the standard simplex Δ n as a subset Δn ⊂ Rn+1 ⊂ Rn+2 ⊂ . . ..

The boundary of Δ1 consists of two copies of Δ 0 , the boundary of Δ 2 consists of three
copies of Δ1 . In general, the boundary of Δ n consists of n + 1 copies of Δn−1 .
We describe the boundary by the following (n + 1) face maps for 0 6 i 6 n

di = din−1 : Δn−1 ,→ Δn ; (t0 , . . . , tn−1 ) 7→ (t0 , . . . , ti−1 , 0, ti , . . . , tn−1 ).

The image of din−1 in Δn is the face that is opposite to the vertex ei . It is the (n−1)-simplex
generated by the n − 1-tuple e0 , . . . , ei−1 , ei+1 , . . . , en of vectors in Rn+1 .
Lemma 1.2.3.
Concerning the composition of face maps, the following rule holds:
din−1 ◦ djn−2 = djn−1 ◦ di−1
n−2
, for all 0 6 j < i 6 n.
Example: face maps for Δ 0 and composition into Δ 2 : d2 ◦ d0 = d0 ◦ d1 .

Proof.
Both expressions yield
din−1 ◦ djn−2 (t0 , . . . , tn−2 ) = (t0 , . . . , tj−1 , 0, tj . . . , ti−2 , 0, ti−1 , . . . , tn−2 )
= djn−1 di−1n−2
(t0 , . . . , tn−2 ).


Definition 1.2.4
Let X be an arbitrary topological space, X 6= ∅. A singular n-simplex in X is a continuous
map α : Δn → X.
Note, that α is just required to be continuous. (It does not make sense to require it to be
smooth. We do not require α to be injective either.) In comparison to the definition of the
fundamental group, note that we do not identify simplices and we do not fix a base point.
We want to be able to express the idea that the boundary of a 1-simplex, i.e. of an interval,
is the the difference of its endpoints. To this end, we have to be able to add and subtract
0-simplices.
We recall some algebraic notions:

5
Remark 1.2.5.
1. Any abelian group A can be seen as a Z-module with n.a := a . . + a} for n ∈ N and
| + .{z
n−times
a ∈ A and (−n).a := −n.a. Thus, abelian groups are in bijection with Z-modules. An
abelian group A is called free over a subset B ⊂ A, if B is a Z-basis, i.e. if any element
a ∈ Z can be uniquely written as a Z-linear combination of elements in B.
2. The group Zr is free abelian with basis {e1 , . . . , er } with ei = (0, . . . , 0, 1, 0, . . . , 0). The
group Z2 is not free, since it does not admit a basis: the vector 1 ∈ Z2 is not free since
2 ∙ 1 = 0.
3. A free abelian group F with basis B can be characterized by the following universal
property: any map f : B → A of sets into an arbitrary abelian group A can be extended
uniquely to a group homomorphism h : F → A, i.e. h(b) = f (b) for all b ∈ B,
HomSet (B, A) ∼
= Homgroup (F, A) .

4. Any subgroup of a free abelian group F is a free abelian group of smaller rank.

Definition 1.2.6
Let X be a topological space. Let Sn (X) be the free abelian group generated by all singular
n-simplices in X. We call Sn (X) the n-th singular chain module of X.

Remarks 1.2.7.
P
1. Elements of the singular chain group Sn (X) are thus sums i∈I λi αi with λi ∈ Z and
λi = 0 for almost all i ∈ I and αi : Δn → X a singular n-simplex. All sums are effectively
finite sums.
2. For all n > 0 there are non-trivial elements in Sn (X), because we assumed that X =
6 ∅:
n
we can always chose a point x0 ∈ X and consider the constant map κx0 : Δ → X as a
singular n-simplex α. By convention, we define Sn (∅) = 0 for all n > 0.
3. By the universal property 1.2.5.3, to define group homomorphisms from Sn (X) to some
abelian group, it suffices to define such a map on generators.
Example 1.2.8.
Let X be any topological space. As an example, we compute S0 (X): a continuous map α : Δ0 →
P is determined by its value α(e0 ) =: xα ∈ X, which is a point
X P in X. A singular 0-simplex
λ α
i∈I i i can thus be identified with the formal sum of points i∈I λi xαi with λi ∈ Z.
Such objects appear in complex analysis: counting the zeroes and poles of a meromorphic
function with multiplicities then this gives an element in S0 (X). In algebraic geometry, a divisor
is an element in S0 (X).

Definition 1.2.9
Using the face maps din−1 : Δn−1 → Δn from Example 1.2.3.3, we define a group homomorphism
∂i : Sn (X) → Sn−1 (X) on generators by precomposition with the face map
∂i (α) = α ◦ din−1
and call it the ith face of the singular simplex α.
P P
On Sn (X), we thus get by Z-linear extension ∂i ( j λj α j ) = j λj (αj ◦ din−1 ).

6
Lemma 1.2.10.
The face maps on Sn (X) satisfy the simplicial relations

∂j ◦ ∂i = ∂i−1 ◦ ∂j , 0 6 j < i 6 n.

Proof.
The relation follows immediately from the relation

din−1 ◦ djn−2 = djn−1 ◦ di−1

n−2
, for all 0 6 j < i 6 n.

in Lemma 1.2.3. 

Definition 1.2.11
the boundary operator on singular chains as ∂ : Sn (X) → Sn−1 (X) as the alternating
We defineP
sum ∂ = ni=0 (−1)i ∂i .

Lemma 1.2.12.
The map ∂ is a boundary operator, i.e. ∂ ◦ ∂ = 0.

Proof.
This is an immediate consequence of the simplicial relations in Lemma 1.2.10
n−1
X n
X XX
j
∂◦∂ =( (−1) ∂j ) ◦ ( (−1)i ∂i ) = (−1)i+j ∂j ◦ ∂i
j=0 i=0 i j
X X
= (−1)i+j ∂j ◦ ∂i + (−1)i+j ∂j ◦ ∂i
06j<i6n 06i6j6n−1
1.2.10
X X
= (−1)i+j ∂i−1 ◦ ∂j + (−1)i+j ∂j ◦ ∂i = 0.
06j<i6n 06i6j6n−1


We therefore obtain for a topological space X a complex of (free) abelian groups,
∂ ∂ ∂ ∂
. . . → Sn (X) −→ Sn−1 (X) −→ . . . −→ S1 (X) −→ S0 (X) → 0 ,

the singular chain complex, S∗ (X). We abbreviate the group Zn (S∗ (X)) of cycles by Zn (X),
the group Bn (S∗ (X)) of boundaries by Bn (X) and the n-th homology group Hn (S∗ (X)) by
Hn (X).

Definition 1.2.13
For a space X, the abelian group Hn (X) is called the nth singular homology group of X.

Example 1.2.14.
1. Note that all 0-chains are 0-cycles, Z0 (X) = S0 (X).
2. The boundary of a 1-chain α : Δ1 → X is

∂α = α ◦ d0 − α ◦ d1 = α(e1 ) − α(e0 )

which justifies the name “boundary”.

7
 3. To find an example of a 1-cycle, consider a 1-chain c = α + β + γ, where we take singular
1-simplices α, β, γ : Δ1 → X such that α(e1 ) = β(e0 ), β(e1 ) = γ(e0 ) and γ(e1 ) = α(e0 ).
Calculate ∂α = ∂0 α − ∂1 α = α(e1 ) − α(e0 ) and similarly for β and γ to find ∂c = 0. This
motivates the word “cycle”.

We need to understand how continuous maps of topological spaces interact with singular
chains and singular homology.

Definition 1.2.15
Let f : X → Y be a continuous map. The map fn = Sn (f ) : Sn (X) → Sn (Y ) is defined on
generators α : Δn → X by postcomposition
α f
fn (α) = f ◦ α : Δn −→ X −→ Y.

Lemma 1.2.16.
For any continuous map f : X → Y we have commuting diagrams
fn
Sn (X) / Sn (Y )
∂X ∂Y
 fn−1

Sn−1 (X) / Sn−1 (Y ),

i.e. (fn )n∈Z is a chain map and hence induces by the remarks following Definition 1.1.5, a map
Hn (f ) : Hn (X) → Hn (Y ) of the homology groups.

Proof.
By definition, we have for a singular n-simplex α : Δn → X by the associativity of the compo-
sition of maps
n
X n
X
Y i
∂ (fn (α)) = (−1) (f ◦ α) ◦ di = (−1)i f ◦ (α ◦ di ) = fn−1 (∂ X α).
i=0 i=0

Remarks 1.2.17.
1. The identity map on X induces the identity map on Hn (X) for all n > 0 and if we have
a composition of continuous maps
f g
X −→ Y −→ Z,

then Sn (g ◦ f ) = Sn (g) ◦ Sn (f ) and thus Hn (g ◦ f ) = Hn (g) ◦ Hn (f ).

2. In categorical language, this says precisely that Sn (−) and Hn (−) are functors from
the category of topological spaces and continuous maps into the category of abelian
groups. Taking all Sn (−) together turns S∗ (−) into a functor from topological spaces
and continuous maps into the category of chain complexes of abelian groups with chain
maps as morphisms.

8
 3. One implication of Lemma 1.2.16 is that homeomorphic spaces have isomorphic homology
groups:
X∼= Y ⇒ Hn (X) ∼ = Hn (Y ) for all n > 0.
In Theorem 1.4.7, we will see the stronger statement that homotopic maps induce the
same morphism in homology.

Our first (not too exciting) calculation is the following:

Proposition 1.2.18.
The homology groups of a one-point space pt are trivial but in degree zero,
(
0, if n > 0,
Hn (pt) ∼
=
Z, if n = 0.

Proof.
For every n > 0 there is precisely one continuous map α : Δn → pt, the constant map. We
denote this map by κn . Then the boundary of κn is
n n
(
X X κn−1 , n even,
∂κn = (−1)i κn ◦ di = (−1)i κn−1 =
i=0 i=0
0, n odd.

For all n we have Sn (pt) ∼

= Z generated by κn and therefore the singular chain complex looks
as follows:
∂=0 / ∂=idZ /
... Z Z ∂=0 / Z → 0 ,
cf. Example 1.1.4.2. 

1.3 The homology groups H0 and H1

We start with the following observation:

Proposition 1.3.1.
For any topological space X, there is a homomorphism ε : H0 (X) → Z with ε 6= 0 for X 6= ∅.

Proof.
• If X 6= ∅, we have a unique morphism X → pt of topological spaces which induces by
Lemma 1.2.16 a morphism of chain complexes S∗ (X) → S∗ (pt). It maps any 0-simplex
α : Δ0 → X to
α
Δ0 → X → pt ,
the generator of H0 (pt), the constant map κ0 : Δ0 → pt, cf. Proposition 1.2.18.

• It is instructive to show directly that the map

ε̃ :S0 (X) → Z
P P
with ε̃(α) = 1 for any generator α : Δ0 → X, thus ε̃( i∈I λi αi ) = i∈I λi on S0 (X) gives
a well-defined map on homology. (As only finitely many λi are non-trivial, this is in fact
a finite sum.)

9
 P
Let S0 (X) 3 c = ∂b be a boundary and write b = i∈I νi βi with βi : Δ1 → X and a finite
set I. Then we get
X X X X
∂b = ∂ νi βi = νi (βi ◦ d0 − βi ◦ d1 ) = ν i βi ◦ d0 − ν i βi ◦ d 1
i∈I i∈I i∈I i∈I

and hence X X
ε̃(c) = ε̃(∂b) = νi − νi = 0.
i∈I i∈I

We said that S0 (∅) is zero, so H0 (∅) = 0. In this case, we define ε to be the zero map.
If X 6= ∅, then any singular 0-simplex α : Δ0 → X can be identified with its image point,
so the map ε on S0 (X) counts points in X with multiplicities.
Proposition 1.3.2.
∼
=
If X is a path-connected, non-empty space, then ε : H0 (X) → Z.

Proof.
1. As X is non-empty, there is a point x ∈ X and the constant map κx with value x is an
element in S0 (X) with ε(κx ) = 1. Therefore, the group homomorphism ε is surjective.
2. For any other point y ∈ X there is a continuous path ω : [0, 1] → X with ω(0) = x and
ω(1) = y. We define a singular 1-simplex αω : Δ1 → X as

αω (t0 , t1 ) = ω(1 − t0 )

for t0 + t1 = 1, 0 6 t0 , t1 6 1. Then

∂(αω ) = ∂0 (αω ) − ∂1 (αω ) = αω (e1 ) − αω (e0 ) = αω (0, 1) − αω (1, 0) = κy − κx ,

and the two singular 0-simplices κx , κy in the path connected space X are homologous.
This shows that ε is injective.


Note that in the proof, we associated to a continuous path ω in X from x to y a 1-simplex

αω on X with ∂αω = κy − κx . In the sequel, we will identify them frequently.
Corollary 1.3.3. F
If X is a disjoint union, X = i∈I Xi , such that all Xi are non-empty and path-connected, then
M
H0 (X) ∼= Z.
i∈I

This gives an interpretation of the zeroth homology group of X: it is the free abelian group
generated by the path-components of X.

Proof.
The singular chain complex of X splits as the direct sum of chain complexes of the Xi :
M
Sn (X) ∼
= Sn (Xi )
i∈I

10
for all n. Boundary summands ∂i stay in a component, in particular,
M M
∂ : S1 (X) ∼
= S1 (Xi ) → S0 (Xi ) ∼
= S0 (X)
i∈I i∈I

is the direct sum of the boundary operators ∂ : S1 (Xi ) → S0 (Xi ) and the claim follows from
Proposition 1.3.2. 

Next, we relate the homology group H1 to the fundamental group π1 . To this end, we see
continuous paths ω in X as 1-simplices αω , as in the proof of Proposition 1.3.2.

Lemma 1.3.4.
Let ω1 , ω2 , ω be paths in a topological space X.

1. Constant paths are null-homologous.

2. If ω1 (1) = ω2 (0), we can define the concatenation ω1 ∗ ω2 of ω1 followed by ω2 . Then

αω1 ∗ω2 − αω1 − αω2 is a boundary.

3. If ω1 (0) = ω2 (0), ω1 (1) = ω2 (1) and if ω1 is homotopic to ω2 relative to {0, 1}, then αω1
and αω2 are homologous as singular 1-chains.

ˉ (t) := ω(1 − t).

4. Any 1-chain of the form αωˉ ∗ω is a boundary. Here, ω

Proof.
1. Denote by cx the constant path on x ∈ X. Consider the constant singular 2-simplex
α(t0 , t1 , t2 ) = x. Then ∂α = cx − cx + cx = cx .

2. We define a singular 2-simplex β : Δ2 → X on X as follows.

e2
KA


QA

Q
ω1 ∗ ω2
QA ω2

QQ A

Q Q A
Q Q

Q -A
Q
e0 ω1 e1

We define β on the boundary components of Δ 2 as indicated and prolong it constantly

along the sloped inner lines. Then

∂β = β ◦ d0 − β ◦ d1 + β ◦ d2 = ω2 − ω1 ∗ ω2 + ω1 .

3. Let H : [0, 1] × [0, 1] → X a homotopy from ω1 to ω2 . As we have that H(0, t) = ω1 (0) =

ω2 (0), we can factor H over the quotient [0, 1] × [0, 1]/{0} × [0, 1] ∼
= Δ2 with induced map
h : Δ2 → X. Then
∂h = h ◦ d0 − h ◦ d1 + h ◦ d2 .
The first summand is null-homologous by 1., because it is constant (with value ω1 (1) =
ω2 (1)), the second one is ω2 and the last is ω1 , thus ω1 − ω2 is null-homologous.

11
 4. Consider a singular 2-simplex γ : Δ2 → X as indicated below.
e2
KA



A
ω(1)


A ω



A



A





-A
e0 ω
ˉ e1

Definition 1.3.5
Let X be path-connected and x ∈ X. Let h : π1 (X, x) → H1 (X) be the map, that sends the
homotopy class [ω]π1 of a closed path ω to its homology class [ω] = [αω ]H1 . This map is called
the Hurewicz-homomorphism. 1
Lemma 1.3.4.3 ensures that h is well-defined and by Lemma 1.3.4.2
1.3.4.2
h([ω1 ][ω2 ]) = h([ω1 ∗ ω2 ]) = [ω1 ] + [ω2 ] = h([ω1 ]) + h([ω2 ])
thus h is a group homomorphism. For a closed path ω we have by Lemma 1.3.4.4 that [ˉω ] = −[ω]
in H1 (X).
Recall that the commutator subgroup [G, G] of G is the smallest subgroup of a group G
containing all commutators [g, h] := ghg −1 h−1 for all g, h ∈ G.
It is a normal subgroup of G: If c ∈ [G, G], then for any g ∈ G the element gcg −1 c−1 is a
commutator and also by the closure property of subgroups the element gcg −1 c−1 c = gcg −1 is
in the commutator subgroup [G, G].

Definition 1.3.6
Let G be an arbitrary group, then its abelianization, Gab , is the quotient group G/[G, G].

Remark 1.3.7.
The abelianization comes with a projection G → Gab . It can be characterized by the universal
property that any group homomorphism G → A with A abelian factorizes uniquely as
G /A
|=
|
|
 |
Gab
Proposition 1.3.8.
Let X be a path-connected non-empty space. Since H1 (X) is abelian, the Hurewicz homomor-
phism factors over the abelianization of π1 (X, x). It induces an isomorphism
π1 (X, x)ab ∼
= H1 (X) ,
i.e.
h
π1 (X, x) / H (X)
ffffffff3 1
∼
= ffffff
p
ffffffffhf
 fff ab

π1 (X, x)ab = π1 (X, x)/[π1 (X, x), π1 (X, x)]

1
Witold Hurewicz: 1904–1956.

12
Proof.
• We construct an inverse φ to hab . To this end, choose as an auxiliary datum for any point
y ∈ X a path uy from the base point x to y. For the base point x itself, we take ux to be
the constant path on x.
Let α be an arbitrary singular 1-simplex and let yi := α(ei ). Define

φ̃ : S1 (X) → π1 (X, x)ab

on the generator α of S1 (X) as the class of the closed path φ̃(α) = [uy0 ∗ ωα ∗ uˉy1 ] and
extend φ linearly to all of S1 (X), keeping in mind that the composition in π1 is written
multiplicatively.
• We have to show that φ̃ is trivial on boundaries, so let β : Δ2 → X a singular 2-simplex.
Then
φ̃(∂β) = φ̃(β ◦ d0 − β ◦ d1 + β ◦ d2 ) = φ̃(β ◦ d0 )φ̃(β ◦ d1 )−1 φ̃(β ◦ d2 ).
Abbreviating β ◦ di with αi , we get as a result

[uy1 ∗ α0 ∗ uˉy2 ][uy0 ∗ α1 ∗ uˉy2 ]−1 [uy0 ∗ α2 ∗ uˉy1 ] = [uy0 ∗ α2 ∗ uˉy1 ∗ uy1 ∗ α0 ∗ uˉy2 ∗ uy2 ∗ αˉ1 ∗ uˉy0 ].

Here, we have used that the image of φ is abelian. We can reduce the paths u ˉ y1 ∗ uy1 and
uˉy2 ∗ uy2 and are left with [uy0 ∗ α2 ∗ α0 ∗ αˉ1 ∗ uˉy0 ] but α2 ∗ α0 ∗ αˉ1 is the closed path tracing
the boundary of the singular 2-simplex β and therefore it is null-homotopic in X. Thus
φ̃(∂β) = 1 and φ̃ passes to a map

φ : H1 (X) → π1 (X, x)ab .

• The composition φ ◦ hab evaluated on the class of a closed path ω gives

φ ◦ hab [ω]π1 = φ[ω]H1 = [ux ∗ ω ∗ uˉx ]π1 .

But we chose ux to be constant, thus φ ◦ hab = id.

P
If c = λi αi is a 1-cycle, then hab ◦ φ(c) is of the form [c + D∂c ] where the D∂c -part comes
from the contributions of the uyi . The fact that ∂(c) = 0 implies that the summands in
D∂c cancel off and thus hab ◦ φ = idH1 (X) .


Note that abelianization of an abelian group is the group itself: G ∼

= Gab . Whenever the
∼
fundamental group is abelian, we thus have H1 (X) = π1 (X, x).
Corollary 1.3.9.
Standard results on the fundamental group π1 yield explicit results for the following first ho-
mology groups:
H1 (Sn ) = 0, for n > 1, H1 (S1 ) ∼ = Z,
1
. . × S}1 ) ∼
| × .{z
H1 (S = Zn ,
n

H1 (S1 ∨ S1 ) ∼
= (Z ∗ Z)ab ∼= Z ⊕ Z,
(
Z, if n = 1,
H1 (RP n ) ∼
=
Z/2Z, for n > 1.

13
1.4 Homotopy invariance
The main goal of this section is to show that two continuous maps that are homotopic induce
identical maps on homology groups.

Observation 1.4.1.
• Let α : Δn → X a singular n-simplex; consider two homotopic maps f, g : X → Y . The
homotopy
H : X × [0, 1] → Y
from f to g induces a homotopy
α×id H
Δn × [0, 1] → X × [0, 1] → Y

from f ◦ α to g ◦ α. This is a map with codomain Δ n × [0, 1], i.e. from a prism over Δ n .
From this geometric homotopy, we want to obtain a chain homotopy from the chain map
S(f ) to the chain map S(g) of singular chain complexes.

• To that end we have to cut the (n+1)-dimensional prism Δ n ×[0, 1] into (n+1)-simplices.
In low dimensions this is intuitive:

– The one-dimensional prism Δ 0 × [0, 1] is homeomorphic to the standard 1-simplex

Δ1 .
– The two-dimensional prism Δ 1 × [0, 1] ∼ = [0, 1]2 (which has the shape of a square)
can be cut in two triangles, i.e. into two copies of the standard 2-simplex Δ 2 .
– Δ2 × [0, 1] is a 3-dimensional prism and that can be glued together from three
tetrahedrons, e.g. like

 
  
  
@
@ @ @   
 
   
   
   
   
@  @

@
 
@ @ @ 

In general, we compose the (n + 1)-dimensional prism Δ n × [0, 1] from n + 1 copies of the

standard simplex Δn+1 :

Definition 1.4.2
For a given n ∈ N0 , define n + 1 injections

pi : Δn+1 → Δn × [0, 1]
(t0 , . . . , tn+1 ) 7→ ((t0 , . . . , ti−1 , ti + ti+1 , ti+2 , . . . , tn+1 ), ti+1 + . . . + tn+1 )

with i = 0, . . . , n. These are (n + 1)-simplices on the prism on the topological space Δn × [0, 1].

14
Remark 1.4.3.
• The image of the standard basis vectors ek with k = 0, 1, . . . n + 1 is
(
(ek , 0), for 0 6 k 6 i,
pi (ek ) =
(ek−1 , 1), for k > i.

For example, in the case n = 1, we have

p0 e0 →
7 e0 p1 e0 →
7 e0
e1 → 7 e0 + e 2 e1 → 7 e1
e2 → 7 e1 + e 2 e2 → 7 e1 + e2

• For all n > 0, we obtain n + 1 group homomorphisms

Pi : Sn (X) → Sn+1 (X × [0, 1])

for i = 0, 1, . . . n which is defined on a generator α : Δn → X of Sn (X) via precomposition:

pi α×id
Pi (α) = (α × id) ◦ pi : Δn+1 −→ Δn × [0, 1] −→ X × [0, 1].

• For k = 0, 1 let jk : X → X × [0, 1] be the inclusion x 7→ (x, k).

Lemma 1.4.4.
The group homomorphisms Pi satisfy the following relations:
1. ∂0 ◦ P0 = Sn (j1 ) as group homomorphisms Sn (X) → Sn (X × [0, 1]).
2. ∂n+1 ◦ Pn = Sn (j0 ),
3. ∂i ◦ Pi = ∂i ◦ Pi−1 for 1 6 i 6 n.
4. (
Pi ◦ ∂j−1 , for i 6 j − 2
∂j ◦ Pi =
Pi−1 ◦ ∂j , for i > j + 1.

Proof.
• For the first point, note that for α : Δn → X, ∂0 ◦ P0 (α) is the singular n-simplex
d p0 α×id
Δn →0 Δn+1 → Δn × [0, 1] → X × [0, 1] .

The composition of the first two maps on Δ n evaluates to

n
X
p0 ◦ d0 (t0 , . . . , tn ) = p0 (0, t0 , . . . , tn ) = ((t0 , . . . , tn ), ti ) = ((t0 , . . . , tn ), 1)
i=0

and thus the whole map equals

α j1
Sn (j1 )(α) : Δn → X → X × [0, 1]

• Similarly, we compute

pn ◦ dn+1 (t0 , . . . , tn ) = pn (t0 , . . . , tn , 0) = ((t0 , . . . , tn ), 0)

and deduce ∂n+1 ◦ Pn = Sn (j0 ).

15
 n
• For the third Pn identity, one checks that pi ◦ di = pi−1 ◦ di on Δ : both give
((t0 , . . . , tn ), j=i tj ) on (t0 , . . . , tn ).

• For d) in the case i > j + 1, consider the following diagram

pi
Δ n+1 / Δn × [0, 1]
dj qqq
q8
q
qqq
qqq
Δn MM
MMMp
MMi−1
MMM
& dj ×id
Δn−1 × [0, 1] / Δn × [0, 1]

Checking coordinates one sees Pn that this diagram commutes: both give
((t0 , . . . , tj−1 , 0, . . . ti−1 + ti , . . . tn ), j=i tj ) on (t0 , . . . , tn ).
The remaining case follows from a similar observation.


Definition 1.4.5
For each n > 0, we define a group homomorphism

P: Sn (X) → Sn+1 (X × [0, 1])

Pn i
as the alternating sum P = i=0 (−1) Pi .

Lemma 1.4.6.
The group homomorphisms P provide a chain homotopy between the chain maps S(j0 ), S(j1 ) :
S∗ (X) → S∗ (X × [0, 1]), i.e. we have for all n

∂ ◦ P + P ◦ ∂ = Sn (j1 ) − Sn (j0 ) .

Proof.
We evaluate the left hand side on a singular n simplex α : Δn → X and find from the definitions

X n+1
n X n−1 X
X n
i+j
∂P α + P ∂α = (−1) ∂ j Pi α + (−1)i+j Pi ∂j α.
i=0 j=0 i=0 j=0

If we single out the terms involving the pairs of indices (0, 0) and (n, n + 1) in the first sum,
we are left with by Lemma 1.4.4.1 and 2.

X X n
n−1 X
i+j
Sn (j1 )(α) − Sn (j0 )(α) + (−1) ∂j Pi α + (−1)i+j Pi ∂j α.
(i,j)6=(0,0),(n,n+1) i=0 j=0

Using Lemma 1.4.4 we see that only the first two summands survive: the terms in the first
sum with i = j and i = j − 1 cancel by Lemma 1.4.4.3. The remaining terms cancel by the
same mechanism as in the proof of Lemma 1.2.12. 

So, finally we can prove the main result of this section:

16
Theorem 1.4.7 (Homotopy invariance).
If f, g : X → Y are homotopic maps, then they induce the same map on homology.

Proof.
Let H : X × [0, 1] → Y be a homotopy from f to g, i.e. H ◦ j0 = f and H ◦ j1 = g. Set
P Sn+1 (H)
Kn := Sn+1 (H) ◦ P : Sn (X) → Sn+1 (X × [0, 1]) −→ Sn+1 (Y ) .

We claim that (Kn )n is a chain homotopy between the two chain maps (Sn (f ))n and (Sn (g))n .
Note that H : X × I → Y induces a chain map (Sn (H))n . Therefore we get

∂ ◦ Sn+1 (H) ◦ P + Sn (H) ◦ P ◦∂ = Sn (H) ◦ ∂ ◦ P + Sn (H) ◦ P ◦ ∂ [S• (H) is a chain map]

| {z } | {z }
Kn Kn−1

= Sn (H) ◦ (∂ ◦ P + P ◦ ∂)
1.4.6
= Sn (H) ◦ (Sn (j1 ) − Sn (j0 ))
= Sn (H ◦ j1 ) − Sn (H ◦ j0 )
= Sn (g) − Sn (f ).

Hence the two chain maps S(f ) and S(g) are chain homotopic; by Proposition 1.1.9.2, we have
Hn (g) = Hn (f ) for all n. 

Corollary 1.4.8.
1. If two spaces X, Y are homotopy equivalent, then H∗ (X) ∼
= H∗ (Y ).
2. In particular, if X is contractible, then
(
Z, for ∗ = 0
H∗ (X) ∼
=
0, otherwise.

Examples 1.4.9.
1. Since Rn is contractible for all n, the above corollary implies that its homology is trivial
but in degree zero where it consists of the integers.

2. As the Möbius strip is homotopy equivalent to S1 , we know that their homology groups
are isomorphic.

3. The zero section of a vector bundle induces a homotopy equivalence between the base
and the total space, hence these two have isomorphic homology groups.

1.5 The long exact sequence in homology

In a typical situation, we have a subspace A of a topological space X and might know some-
thing about A or X and want to calculate the homology of the other space using that partial
information.
Before we can move on to topological applications, we need some algebraic techniques for
chain complexes. We need to know that a short exact sequence of chain complexes gives rise to
a long exact sequence in homology.

Definition 1.5.1
17
 1. Let A, B, C be abelian groups and
f g
A /B /C

a sequence of homomorphisms. Then this sequence is exact, if the image of f equals the
kernel of g.

2. A sequence
fi+1 fi fi−1
... / Ai / Ai+1 / ...
of homomorphisms of abelian groups (indexed over the integers) is called (long) exact, if
it is exact at every Ai , i.e. the image of fi+1 equals the kernel of fi for all i.

3. An exact sequence of the form

f g
0 /A /B /C /0

is called a short exact sequence.

Examples 1.5.2.
1. The sequence
0 / Z 2∙ / Z π / Z/2Z /0

is a short exact sequence.

2. The sequence
ι
0 /U /A

is exact, iff ι : U → A is a monomorphism. The sequence

%
B /Q /0

is exact, iff % : B → Q is an epimorphism. Finally, Φ : A → A0 is an isomorphism, iff the

φ
sequence 0 /A / A0 /0 is exact.

3. A sequence
f g
0 /A /B /C /0

is exact, iff f is injective, the image of f equals the kernel of g and g is an epimorphism.

4. Another equivalent description is to view a long exact sequence as a chain complex with
vanishing homology groups. Homology measures the deviation from exactness.

Definition 1.5.3
If A∗ , B∗ , C∗ are chain complexes and f∗ : A∗ → B∗ , g∗ : B∗ → C∗ are chain maps, then we call
the sequence of chain complexes
f∗
A∗ / B ∗ g∗ / C∗

exact, if the image of fn is the kernel of gn for all n ∈ Z.

18
 Thus such an exact sequence of chain complexes is a commuting double ladder
.. .. ..
. . .
d d d
 fn+1  gn+1 
An+1 / Bn+1 / Cn+1
d d d
 fn  gn 
An / Bn / Cn
d d d
 fn−1  gn−1 
An−1 / Bn−1 / Cn+1
d d d
  
.. .. ..
. . .
in which every row is exact and where in the columns we have differentials, i.e. d ◦ d = 0.
Example 1.5.4.
Let p be a prime, then the diagram
0 0 0

 id
 0 
Z / Z / 0
∙p ∙p2
 ∙p  
π / Z/pZ
Z / Z
π π id
 ∙p
 
π
Z/pZ / Z/p2 Z / Z/pZ

  
0 0 0
has exact rows and columns. It is an exact sequence of chain complexes. Here, π denotes the
appropriate canonical projection map.
Proposition 1.5.5.
f g
If 0 / A∗ / B∗ / C∗ / 0 is a short exact sequence of chain complexes, then there exists
a homomorphism δn : Hn (C∗ ) → Hn−1 (A∗ ) for all n ∈ Z which is natural, i.e. if
f g
0 / A∗ / B∗ / C∗ / 0
α β γ
 f0
 g0

0 / A0∗ / B∗0 / C∗0 / 0
is a commutative diagram of chain complexes in which the rows are exact, then Hn−1 (α) ◦ δn =
δn0 ◦ Hn (γ),
δn
Hn (C∗ ) / Hn−1 (A∗ )
Hn (γ) Hn−1 (α)
 0
δn 
Hn (C∗0 ) / Hn−1 (A0∗ )

19
 The method of proof is an instance of a diagram chase. The homomorphism δn is called
connecting homomorphism.

Proof.
We show the existence of a δ first and then prove that the constructed map satisfies the
naturality condition.

a) Definition of δ:
We work with the following maps:
gn
Bn 3 b  / c ∈ Cn
d
fn−1  gn−1
An−1 3 a  / db ∈ Bn−1  / 0

For c ∈ Cn with d(c) = 0, we choose a preimage b ∈ Bn with gn b = c. This is possible

because gn is surjective. We know that dgn b = dc = 0 = gn−1 db thus db is in the kernel
of gn−1 , hence it is in the image of fn−1 . Thus there is an a ∈ An−1 with fn−1 a = db. We
have that fn−2 da = dfn−1 a = ddb = 0 and as fn−2 is injective, this shows that a is a cycle.
We define δ[c] := [a].
In order to check that δ is well-defined, we assume that there are two different preimages
b and b0 with gn b = gn b0 = c. Then gn (b − b0 ) = 0 and thus there is an ã ∈ An with
fn ã = b − b0 . Define a0 := a − dã ∈ An−1 . Then

fn−1 a0 = fn−1 a − fn−1 dã = db − db + db0 = db0

because fn−1 dã = dfn ã = db − db0 . As fn−1 is injective, we get that a0 is uniquely deter-
mined with this property. As a is homologous to a0 we get that [a] = [a0 ] = δ[c], thus the
latter is independent of the choice of the preimage b.
In addition, we have to make sure that the value stays the same if we add a boundary
term to c, i.e. take c0 = c + dc̃ for some c̃ ∈ Cn+1 . Choose preimages of c, c̃ under the
surjective maps gn and gn+1 , i.e., b and b̃ with gn b = c and gn+1 b̃ = c̃. Then the element
b0 = b + db̃ has boundary db0 = db and thus both choices will result in the same a.
Therefore the connecting morphism δn : Hn (C∗ ) → Hn−1 (A∗ ) is well-defined.

b) We have to show that δ is natural with respect to maps of short exact sequences.
Let c ∈ Zn (C∗ ), then δ[c] = [a] for some b ∈ Bn with gn b = c and a ∈ An−1 with
fn−1 a = db. Therefore, Hn−1 (α)(δ[c]) = [αn−1 (a)].
On the other hand, we have
0
fn−1 (αn−1 a) = βn−1 (fn−1 a) = βn−1 (db) = dβn b

and
gn0 (βn b) = γn gn b = γn c
and we can conclude that by the construction of the connecting homomorphism δ 0 in the
second long exact sequence
δ 0 [γn (c)] = [αn−1 (a)]
and this shows δ 0 ◦ Hn (γ) = Hn−1 (α) ◦ δ.

20
 

With this auxiliary result at hand we can now prove the main result in this section:
Proposition 1.5.6 (Long exact sequence in homology).
For any short exact sequence
f g
0 / A∗ / B∗ / C∗ /0

of chain complexes we obtain a long exact sequence of homology groups

δ Hn (f ) Hn (g) δ Hn−1 (f )
... / Hn (A∗ ) / Hn (B∗ ) / Hn (C∗ ) / Hn−1 (A∗ ) /. . .

Proof.
a) Exactness at Hn (B∗ ):
We have Hn (g) ◦ Hn (f )[a] = [gn (fn (a))] = 0, because the composition of gn and fn is zero.
This proves that the image of Hn (f ) is contained in the kernel of Hn (g).
For the converse, let [b] ∈ Hn (B∗ ) with [gn b] = 0. Since gn b is a boundary, there exists
c ∈ Cn+1 with dc = gn b. As gn+1 is surjective, we find a b0 ∈ Bn+1 with gn+1 b0 = c. Hence

gn (b − db0 ) = gn b − dgn+1 b0 = dc − dc = 0.

Exactness at Bn allows to find a ∈ An with fn a = b − db0 . Now

fn−1 (da) = dfn (a) = d(b − db0 ) = db = 0

since b is a cycle. The map fn−1 is injective, thus da = 0. Therefore fn a is homologous to

b and Hn (f )[a] = [b]. Thus the kernel of Hn (g) is contained in the image of Hn (f ).

b) Exactness at Hn (C∗ ):
Let [b] ∈ Hn (B∗ ), then δ[gn b] = 0 because b is a cycle, so 0 is the only preimage under
the injective map fn−1 of db = 0. Therefore the image of Hn (g) is contained in the kernel
of the connecting morphism δ.
Now assume that δ[c] = 0, thus in the construction of δ, the a is a boundary, a = da0 .
Then for a preimage b of c under gn , we have by the definition of a

d(b − fn a0 ) = db − dfn a0 = db − fn−1 a = 0.

Thus b − fn a0 is a cycle and gn (b − fn a0 ) = gn b − gn fn a0 = gn b − 0 = gn b = c, so we found

a preimage for [c] under Hn (g) and the kernel of δ is contained in the image of Hn (g).

c) Exactness at Hn−1 (A∗ ):

Let c be a cycle in Zn (C∗ ). Again, we choose a preimage b of c under gn and an a with
fn−1 (a) = db. Then Hn−1 (f )δ[c] = [fn−1 (a)] = [db] = 0. Thus the image of δ is contained
in the kernel of Hn−1 (f ).
If a ∈ Zn−1 (A∗ ) with Hn−1 (f )[a] = 0. Then fn−1 a = db for some b ∈ Bn . Take c = gn b.
Then by definition δ[c] = [a].


21
1.6 The long exact sequence of a pair of spaces
Let X be a topological space and A ⊂ X a subspace of X.
Remarks 1.6.1.
1. Consider the inclusion map i : A → X, i(a) = a. We obtain an induced map of chain
complexes Sn (i) : Sn (A) → Sn (X). The inclusion of spaces does not have to yield a
monomorphism on homology groups. For instance, we can include A = S1 into X = D2 .
By Corollary 1.4.8.2, since D is contractible, we know that Hn (D) = 0 for n > 1 and by
Corollary 1.3.9 that H1 (S1 ) = Z.
2. Consider the quotient groups Sn (X, A) := Sn (X)/Sn (A). Since dn (Sn (A)) ⊂ Sn−1 (A), the
differential induces a well-defined map
dn Sn (X)/Sn (A) → Sn−1 (X)/Sn−1 (A)
cn + Sn (A) 7→ dn (cn ) + Sn−1 (A)
that squares to zero.
3. Alternatively, Sn (X, A) is isomorphic to the free abelian group generated by all n-simplices
β : Δn → X whose image is not completely contained in A, i.e. β(Δn ) ∩ (X \ A) 6= ∅.
We consider pairs of spaces (X, A).

Definition 1.6.2
The relative chain complex of the pair (X, A) is
S∗ (X, A) := S∗ (X)/S∗ (A)
with the differentials described in Remark 1.6.1.2.

Definition 1.6.3
• Elements in Sn (X, A) are called relative chains in (X, A).
• Cycles in Sn (X, A) are chains c with ∂ X (c) a linear combination of generators with image
in A. These are called relative cycles.

• Boundaries in Sn (X, A) are chains c in X such that c = ∂ X b + a where a is a chain in A.

These are called relative boundaries.

The following facts are immediate from the definition:

1. Sn (X, ∅) ∼
= Sn (X).
2. Sn (X, X) = 0.
3. Sn (X t X 0 , X 0 ) ∼
= Sn (X).

Definition 1.6.4
The relative homology groups of the pair of spaces (X, A) are the homology groups of the
relative chain complex S∗ (X, A) from Definition 1.6.2:
Hn (X, A) := Hn (S∗ (X, A)).

22
Theorem 1.6.5 (Long exact sequence for relative homology).
1. For any pair of topological spaces A ⊂ X we obtain a long exact sequence

δ Hn (i) Hn−1 (i)

... / Hn (A) / Hn (X) / Hn (X, A) δ / Hn−1 (A) /. . .

2. For a map of spaces f : X → Y with f (A) ⊂ B ⊂ Y , we get an induced map of long

exact sequences

δ Hn (i) δ Hn−1 (i)

... / Hn (A) / Hn (X) / Hn (X, A) / Hn−1 (A) / ...
Hn (f |A ) Hn (f ) Hn (f ) Hn−1 (f |A )
   
... / Hn (B) / Hn (Y ) / Hn (Y, B) / Hn−1 (B) H / ...
δ Hn (i) δ n−1 (i)

A map f : X → Y with f (A) ⊂ B is denoted by f : (X, A) → (Y, B).

Proof.
1. By definition of the relative chain complex S∗ (X, A) the sequence

0 / S∗ (A) S∗ (i) / S∗ (X) π / S∗ (X, A) /0

is an exact sequence of chain complexes and by Proposition 1.5.6 we obtain the long exact
sequence in the first claim.

2. For a map f : (X, A) → (Y, B) the diagram

Sn (i) π
0 / Sn (A) / Sn (X) / Sn (X, A) / 0
Sn (f |A ) Sn (f ) Sn (f )/Sn (f |A )
 Sn (i)
 
π
0 / Sn (B) / Sn (Y ) / Sn (Y, B) / 0

commutes. We now use Proposition 1.5.5.

Example 1.6.6.
Consider the embedding
ι: Sn−1 ,→ Dn .
We obtain a long exact sequence
δ
. . . → Hj (Sn−1 ) → Hj (Dn ) → Hj (Dn , Sn−1 ) → Hj−1 (Sn−1 ) → Hj−1 (Dn ) → . . .

The disc Dn is contractible and by Corollary 1.4.8, we have Hj (Dn ) = 0 for j > 0. From the
long exact sequence we get that δ : Hj (Dn , Sn−1 ) ∼
= Hj−1 (Sn−1 ) for j > 1 and n > 1.
Recall the following definitions:

Definition 1.6.7
1. A subspace ι : A ,→ X is a weak retract, if there is a map r : X → A with r ◦ ι ' idA .

23
 2. A subspace ι : A ,→ X is a deformation retract, if there is a homotopy R : X × [0, 1] → X
such that

(a) R(x, 0) = x for all x ∈ X,

(b) R(x, 1) ∈ A for all x ∈ X, and
(c) R(a, 1) = a for all a ∈ A.

Any deformation retract is a weak retract: take r := R(−, 1) : X → A. Condition (c) then
amounts to r ◦ ι = idA .

Proposition 1.6.8.
If i : A ,→ X is a weak retract, then

Hn (X) ∼
= Hn (A) ⊕ Hn (X, A), 0 6 n.

Proof.
From the defining identity of a weak retract r ◦ ι ' idA , we get by Theorem 1.4.7 that Hn (r) ◦
Hn (i) = Hn (idA ) = idHn (A) for all n. Hence Hn (i) is injective for all n. This implies that
Hn (i)
0 → Hn (A) −→ Hn (X) is exact. Injectivity of Hn−1 (i) yields that the image of δ : Hn (X, A) →
Hn−1 (A) is trivial. Therefore, the long exact sequence of Theorem 1.6.4 decomposes into short
exact sequences
Hn (i) ∗ π
0 → Hn (A) −→ Hn (X) −→ Hn (X, A) → 0
for all n. As Hn (r) is a left-inverse for Hn (i) we obtain a splitting

Hn (X) ∼
= Hn (A) ⊕ Hn (X, A) .

Indeed, we have a map

Hn (X) → Hn (A) ⊕ Hn (X, A)
[c] 7→ ([rc], π∗ [c])
with inverse
Hn (A) ⊕ Hn (X, A) → Hn (X)
([a], [b]) 7→ Hn (i)[a] + [a0 ] − Hn (i ◦ r)[a0 ]
for any [a0 ] ∈ Hn (X) with π∗ [a0 ] = [b]. The second map is well-defined: if [a00 ] is an-
other element with π∗ [a00 ] = [b], then [a0 − a00 ] is of the form Hn (i)[ã] because this element
is in the kernel of π∗ and hence [a0 −a00 ]−Hn (i◦r)[a0 −a00 ] = Hn (i)[ã]−Hn (i◦r ◦i)[ã] is trivial. 

Proposition 1.6.9.
For any ∅ 6= A ⊂ X such that A ⊂ X is a deformation retract, then

Hn (i) : Hn (A) ∼
= Hn (X), Hn (X, A) ∼
= 0, 0 6 n.

Proof.
Consider the map r := R(−, 1) : X → A. Then R is a homotopy from id X to i ◦ r. The third
condition defining a deformation retract can be rewritten as r ◦ i = idA , i.e. r is a retraction.
Together, this implies that A and X are homotopically equivalent and by Corollary 1.4.8 Hn (i)

24
is an isomorphism for all n > 0. 

Definition 1.6.10
If X has two subspaces A, B ⊂ X, then (X, A, B) is called a triple, if B ⊂ A ⊂ X.
Any triple gives rise to three pairs of spaces (X, A), (X, B) and (A, B) and accordingly we
have three long exact sequences in homology. But there is another long exact sequence:
Proposition 1.6.11.
For any triple (X, A, B), there is a natural long exact sequence
δ
... / Hn (A, B) / Hn (X, B) / Hn (X, A) / Hn−1 (A, B) /. . .

This sequence is part of the following braided commutative diagram displaying four long exact
sequences
. . .MM .9 . .
MMM
MMM rrrrr
MMM rr
&  rrr
Hn+1 (X, A) Hn (A, B) Hn−1 (B)
s9 MMM r9 NNN pp8 II
ssss MMM
MMM rrrrr NNN
NNN pppp II
II
sss r pp II
sss
MM& rrr NN& ppp II
I$
s
. . .KK .: . .
Hn (A) Hn (X, B)
KKK
qq8 LLL pp8 NNN uuu
KKK q LLL pp NNN u
qqq
ppp NNN uu
KKK
qqq LLL
pp N uuu
K% qq L& pp N' uu
Hn (B) H? n (X) Hn>(X, A)
qqq8 LLL
LLL
qq
qqqqq LLL
L%
... q ...
In particular, the connecting homomorphism δ : Hn (X, A) → Hn−1 (A, B) is the composite
(A,B)
δ = π∗ ◦ δ (X,A) .

Proof.
Note that Sn (B) ⊂ Sn (A) ⊂ Sn (X); by the homomorphism theorem, the sequence

0 / Sn (A)/Sn (B) / Sn (X)/Sn (B) / Sn (X)/Sn (A) / 0.

is exact. Now apply Proposition 1.5.6 to obtain the long exact sequence. 

1.7 Excision
The aim is to simplify relative homology groups. Let A ⊂ X be a subspace. Then it is easy to
see that H∗ (X, A) is not isomorphic to H∗ (X \ A): Consider the figure eight as X and A as the
point connecting the two copies of S1 .

'$
'$

•
&%
&%

25
 • X \ A has two connected components. By Corollary 1.3.3 we have H0 (X \ A) ∼
= Z ⊕ Z.
• Any x ∈ X \A is a generator for the group of 0-cycles. Since the space X is path connected,
it is homologous to the point a ∈ A and thus vanishes in relative homology. The group
H0 (X, A) is trivial.

So if we want to simplify the relative homology group H∗ (X, A) by excising something, then
we have to be more careful. The first step towards that is to make singular simplices ’smaller’.
The technique is called barycentric subdivision; it is a tool that is frequently used.
First, we construct cones. Let v ∈ Δp and let α : Δn → Δp be a singular n-simplex on Δp .

Definition 1.7.1
The cone of α : Δn → Δp with respect to v ∈ Δp is the singular (n+1)-simplex Kv (α) : Δn+1 →
Δp , (
(1 − tn+1 )α( 1−tt0n+1 , . . . , 1−ttnn+1 ) + tn+1 v, tn+1 < 1,
(t0 , . . . , tn+1 ) 7→
v, tn+1 = 1.
This map is well-defined and continuous. On the standard basis vectors Kv gives Kv (ei ) = α(ei )
for 0 6 i 6 n but Kv (en+1 ) = v. Extending Kv linearly gives a map on chain groups

Kv : Sn (Δp ) → Sn+1 (Δp ).

Lemma 1.7.2.
The map Kv satisfies:
1. For c ∈ S0 (Δp ), the boundary of the cone Kv (c) is the 0-chain

∂Kv (c) = ε(c).κv − c

with κv (e0 ) = v and ε the augmentation as introduced in Proposition 1.3.1.

2. For n > 0 we have that ∂ ◦ Kv − Kv ◦ ∂ = (−1)n+1 id.

Proof.
1. For a singular 0-simplex α : Δ0 → Δp we know that ε(α) = 1 and we calculate

∂Kv (α)(e0 ) = Kv (α) ◦ d0 (e0 ) − Kv (α) ◦ d1 (e0 ) = Kv (α)(e1 ) − Kv (α)(e0 ) = v − α(e0 ).

Extending linearly shows the claim.

2. For n > 0 we have to calculate ∂i Kv (α) and it is straightforward to see that ∂n+1 Kv (α) = α
and ∂i (Kv (α)) = Kv (∂i α) for all 0 6 i < n + 1.


Definition 1.7.3
For an n-simple
P α : Δn → Δp on Δp , choose as the additional vertex the barycenter v(α) =
1 n p p
v := n+1 i=0 α(ei ) of the vertices. The barycentric subdivision B : Sn (Δ ) → Sn (Δ ) is defined
inductively as B(α) = α for α ∈ S0 (Δ ) and B(α) = (−1) Kv (B(∂α)) for n > 0. For n > 1
p n

P
this equals B(α) = ni=0 (−1)n+i Kv (B(∂i α)).

26
 If we take n = p and α = idΔn , then for small n this looks as follows: You cannot subdivide
a point any further. For n = 1 we get
•
@
@
R
@
•
@
I
@
@•

And for n = 2 we get (up to tilting) •

@
@
@
@@
•PPP  •
@
P
• P
  PP@
 P@
• • P•

Lemma 1.7.4.
The barycentric subdivision is a chain map
B: S∗ (Δp ) → S∗ (Δp ) .

Proof.
We have to show that ∂B = B∂.
• If α is a singular zero chain, then the fact Bα = α from the definition implies ∂Bα =
∂α = 0 and B∂α = B(0) = 0.
• Let n = 1. Then by Definition 1.7.3
∂Bα = −∂Kv B(∂0 α) + ∂Kv B(∂1 α).
But the boundary terms are zero chains on which B is the identity, so we get with Lemma
1.7.2.1
1.7.2.1
−∂Kv (∂0 α) + ∂Kv (∂1 α) = −κv + ∂0 α + κv − ∂1 α = ∂α = B∂α.
In the last step, we used that B is the identity on the 0-chain ∂α. Note, that the v is
v(α), not a v(∂i α).
• We prove the claim inductively, so let α ∈ Sn (Δp ). Then
def
∂Bα =(−1)n ∂Kv (B∂α)
1.7.2.2
= (−1)n ((−1)n B∂α + Kv ∂B∂α)
ind.
= B∂α + (−1)n Kv B∂∂α = B∂α.
Here, the first equality is by definition, the second one follows by Lemma 1.7.2.2 and then
we use the induction hypothesis and the fact that ∂∂ = 0.

Our aim is to show that barycentric subdivision B does not change anything on the level of
homology groups and to that end we prove that the chain map B : S∗ (Δp ) → S∗ (Δp ) is chain
homotopic to the identity.
To this end, we construct ψn : Sn (Δp ) → Sn+1 (Δp ) again inductively on generators as
ψ0 (α) := 0, ψn (α) := (−1)n+1 Kv (Bα − α − ψn−1 ∂α)
1
Pn
with v := n+1 i=0 α(ei ) the barycenter.

27
Lemma 1.7.5.
The sequence (ψn )n is a chain homotopy from B to the identity on S∗ (Δp ).

Proof.
• For n = 0 we have ∂ψ0 = 0 and this agrees with B − id in that degree.
• For n = 1, we get
def
∂ψ1 + ψ0 ∂ = ∂ψ1 = ∂(Kv B − Kv − Kv ψ0 ∂) = ∂Kv B − ∂Kv .

With Lemma 1.7.2.2 we can transform the latter to B + Kv ∂B − ∂Kv and as B is a

chain map, this equals B + Kv B∂ − ∂Kv . In chain degree one B∂ agrees with ∂, thus this
reduces to
1.7.2.2
B + Kv ∂ − ∂Kv = B − (∂Kv − Kv ∂) = B − id.

• So, finally we can do the inductive step:

∂ψn =(−1)n+1 ∂Kv (B − id − ψn−1 ∂) [defn.]

=(−1) ∂Kv B − (−1) ∂Kv − (−1)n+1 ∂Kv ψn−1 ∂
n+1 n+1

=(−1)n+1 ((−1)n+1 B + Kv ∂B) [Lemma 1.7.2.2]

− (−1) ((−1)n+1 id + Kv ∂)
n+1
[Lemma 1.7.2.2]
− (−1)n+1 ((−1)n+1 ψn−1 ∂ + Kv ∂ψn−1 ∂) [1.7.2.2]
=B − id − ψn−1 ∂ + remaining terms

The equation
ind. ass.
Kv ∂ψn−1 ∂ + Kv ψn−2 ∂ 2 = Kv (∂ψn−1 + ψn−2 ∂) ∂ = Kv B∂ − Kv ∂

from the inductive assumption ensures that these terms give zero.


Definition 1.7.6
A singular n-simplex α : Δn → Δp is called affine, if
n
X n
X
α( ti e i ) = ti α(ei ).
i=0 i=0
Pn Pn
We abbreviate vi := α(ei ), so α( i=0 ti ei ) = i=0 ti vi and we call the elements vi ∈ Δp the
vertices of α.

Definition 1.7.7
Let A be a subset of a metric space (X, d). The diameter of A is

sup{d(x, y)|x, y ∈ A}

and we denote it by diam(A).

Accordingly, the diameter of an affine n-simplex α in Δp is the diameter of its image, and
we abbreviate that with diam(α).

28
Lemma 1.7.8.
For any affine singular n-simplex α every simplex in the chain Bα has diameter 6 n
n+1
diam(α).

Thus barycentric subdivision of affine simplices decreases the diameter. Either you believe
this lemma, or you prove it, or you check Bredon, Proof of Lemma 13.7 (p. 226).
Each simplex in the chain Bα is again affine; this allows us to iterate the application of B
and get smaller and smaller diameter of individual simplices. Thus, the k-fold iteration, B k (α),
n

k
has diameter at most n+1 diam(α).
In the following we use the easy, but powerful trick to express the singular n-simplex α :
Δn → X as
α = α ◦ idΔn = Sn (α)(idΔn ) ,
i.e. as the image of an n-simplex on Δn . This allows us to use the barycentric subdivision for
general spaces: note that id Δn : Δn → Δn can be seen as an n-simplex on Δn . To this simplex, we
can apply barycentric subdivision to get a chain B(idΔn ) ∈ Sn (Δn ). Now a singular n-simplex
on X is a map α : Δn → X and thus gives rise to a morphism of abelian groups

Sn (α) : Sn (Δn ) → Sn (X) .

Therefore, Sn (α)B(Δn ) ∈ Sn (X).

Definition 1.7.9
1. We define BnX : Sn (X) → Sn (X) as

BnX (α) := Sn (α) ◦ B(idΔn ).

2. Similarly, ψnX : Sn (X) → Sn+1 (X) is defined as

ψnX (α) := Sn+1 (α) ◦ ψn (idΔn ).

Lemma 1.7.10.
f
1. The maps B X are natural in X , i.e. for any map X → Y of topological spaces the
diagram
BX
S∗ (X) / S∗ (X)

S∗ (f ) S∗ (f )
 
BY
S∗ (Y ) / S∗ (Y )
commutes.

2. The maps
B X : S∗ (X) → S∗ (X)
are homotopic to the identity on Sn (X).

Proof.

29
 • Let f : X → Y be a continuous map. We have

Sn (f )BnX (α) =Sn (f ) ◦ Sn (α) ◦ B(idΔn )

=Sn (f ◦ α) ◦ B(idΔn )
=BnY (f ◦ α) = BnY Sn (f )(α).

In the first step, we used the definition of BnX ; in the second step the functorality of
Sn (−). In the last step, we used the definition of Sn (f ). Thus the maps B X are natural
in X.

• The calculation for ∂ψnX + ψn−1

X
∂ = BnX − idSn (X) uses that α induces a chain map and
thus we get
defn ∂ chain map
∂ψnX (α) = ∂ ◦ Sn+1 (α) ◦ ψn (idΔn ) = Sn (α) ◦ ∂ ◦ ψn (idΔn ).

Hence
(∂ψnX + ψn−1
X
∂)(α) = Sn (α) ◦ (∂ ◦ ψn (idΔn ) + ψn−1 ◦ ∂(idΔn ))
[1.7.5]
= Sn (α) ◦ (B − id)(idΔn ) = BnX (α) − α.

Now we consider singular n-chains that are spanned by ’small’ singular n-simplices. Here,
“smallness’ is defined in terms of an open covering.

Definition 1.7.11
Let U = {Ui , i ∈ I} be an open covering of X. Then SnU (X) is the free abelian group generated
by all singular n-simplices α : Δn → X such that the image of Δn under α is contained in one
of the open sets Ui ∈ U.
Note that SnU (X) is an abelian subgroup of the singular chain group Sn (X). The restriction
of the differential of Sn (X) gives a chain complex

. . . → SnU (X) → Sn−1

U
(X) → . . . .

We denote its homology by HnU (X). As we will see now, these chains suffice to detect everything
in singular homology.

Lemma 1.7.12.
1. For any subspace A ⊂ X, the barycentric subdivision of c ∈ Sn (A) is again in Sn (A), i.e.
B k (c) ∈ Sn (A).

2. If c ∈ Sn (X) is a cycle relative A ⊂ X, then B(c) is a cycle relative A as well that is

homologous to c relative A.

3. Let U be an open covering of X. Then every cycle in Sn (X) is homologous to a cycle in

SnU (X).

Proof.
1. This follows at once from the definition of barycentric subdivision.

30
 2. We note that the map ψn : Sn (X) → Sn+1 (X) maps α ∈ Sn (A) to ψn (α) ∈ Sn+1 (A).
Now consider for a relative cycle c the equation. cf. 1.7.10.2

Bc = c + ψn−1 ∂c + ∂ψn c .

Since c is a relative cycle, ∂c ∈ Sn−1 (A) and by part 1, ψn−1 ∂c ∈ Sn (A). Thus Bc is
homologous to c relative A. Its boundary is

∂Bc = ∂c + ∂ψn−1 ∂c .

Thus Bc is a relative cycle as well.

P
3. Consider a singular n-chain α = m j=1 λj αj ∈ Sn (X) on X and let Lj for 1 6 j 6 m be
the Lebesgue numbers for the m coverings {αj−1 (Ui ), i ∈ I} of the simplex Δn . Choose k,

k
n
such that n+1 6 L1 , . . . , Lm . Then B k α1 up to B k αm are all chains in SnU (X). Therefore
m
X
B k (α) = λj B k (αj ) =: α0 ∈ SnU (X).
j=1

From part 2, we know that B k α is a cycle as well that is homologous to α.



We conclude:
Corollary 1.7.13.
For any open covering U, the injective chain map

S∗U (X) ,→ S∗ (X)

induces an isomorphism in homology, HnU (X) ∼

= Hn (X).

Proof.
The map on homology is surjective, since for any cycle c ∈ Sn (X), we find by Lemma 1.7.12.3
a homologous cycle c0 ∈ SnU (X).
The map is injective as well: suppose c ∈ SnU (X) is a boundary in Sn (X), i.e. c = ∂e with
e ∈ Sn+1 (X). Find k ∈ N such that B k (e) ∈ Sn+1
U
(X) and

B k (e) − e = ψ̃n−1 (∂e) + ∂ ψ̃n (e) = ψ̃n−1 (c) + ∂ ψ̃n (e) .

Thus
∂B k (e) − ∂e = ∂ ψ̃n−1 (c)
is a boundary in SnU (X). Thus,

c = ∂e = ∂(B k (e) − ψ̃n−1 (e))

is a boundary in SnU (X) as well. 

We remark that this isomorphism actually comes from a homotopy equivalence of chain
complexes.
With this we get the main result of this section:

31
Theorem 1.7.14 (Excision).
Let W ⊂ A ⊂ X such that W ˉ ⊂ Å. Then the inclusion i : (X \ W, A \ W ) ,→ (X, A) induces
an isomorphism of relative homology groups
Hn (i) : Hn (X \ W, A \ W ) ∼
= Hn (X, A)
for all n > 0.

Proof.
• We first prove that Hn (i) is surjective.
Let c ∈ Sn (X, A) be a relative cycle, i.e. ∂c ∈ Sn−1 (A). Consider the open covering
U = {Å, X \ W ˉ } =: {U, V } of X. Now subdivide and find k such that c0 := B k c is a
chain in SnU (X). It follows from Lemma 1.7.12.2 that c0 is homologous to c relative A.
Decompose c0 = cU + cV with cU and cV being chains on the corresponding open sets.
(This decomposition is not unique.)
The boundary of c0 is ∂c0 = ∂B k c = B k ∂c; by assumption this is a chain in Sn−1 (A).
Moreover, we find from the decomposition c0 = cU + cV
∂c0 = ∂cU + ∂cV
with ∂cU ∈ Sn−1 (U ) ⊂ Sn−1 (A). Thus, ∂cV = ∂c0 − ∂cU ∈ Sn−1 (A). Since ∂cV ∈ V is
supported in X \ W , we have ∂cV ∈ Sn−1 (A \ W ). Therefore, cV is a relative cycle in
Sn (X \ W, A \ W ).
In Hn (X, A), we find [c] = [c0 ] = [cU + cV ] = [cV ], where we used in the first step Lemma
1.7.12.2. This shows that Hn (i)[cV ] = [c] ∈ Hn (X, A). Thus [cV ] is a preimage of [c] in
Hn (X \ W, A \ W ).
• The injectivity of Hn (i) is shown as follows.
Assume that there exists c ∈ Sn (X \ W ) with ∂c ∈ Sn−1 (A \ W ) such that Hn (i)[c] = 0.
The last statement means that c is of the form c = ∂b + a0 for some b ∈ Sn+1 (X) and
a0 ∈ Sn (A). We can choose all summands such that they avoid W .
We write b as bU + bV with bU ∈ Sn+1 (U ) ⊂ Sn+1 (A) and bV ∈ Sn+1 (V ) ⊂ Sn+1 (X \ W ).
Then
c = ∂bU + ∂bV + a0 .
Note that a0 and ∂bU are chains in Sn (A\W ). So we have written c as a boundary of a chain
bV in Sn+1 (X \ W ) plus a chain a0 + ∂bU in Sn (A \ W ). Thus [c] = 0 ∈ Hn (X \ W, A \ W ).


1.8 Mayer-Vietoris sequence

We consider the following situation: there are subspaces X1 , X2 ⊂ X such that X1 and X2 are
open in X and such that X = X1 ∪ X2 . We consider the open covering U = {X1 , X2 }. We need
the following maps:
: X1 B
i1uu BB κ
uuu BB 1
uu BB
uu B
X1 ∩ XI2 > X.
II
II |||
I |
i2 III || κ
$ || 2
X2

32
Note that by definition, the sequence of complexes
(i1 ,i2 )
0 / S∗ (X1 ∩ X2 ) / S∗ (X1 ) ⊕ S∗ (X2 ) / S U (X) /0 (1)
∗

is exact. Here, the second map is

(α1 , α2 ) 7→ κ1 (α1 ) − κ2 (α2 ).
Note that here the open sets are ordered to define the difference.
Theorem 1.8.1 (Mayer-Vietoris sequence).
There is a long exact sequence
δ δ
... / Hn (X1 ∩ X2 ) / Hn (X1 ) ⊕ Hn (X2 ) / Hn (X) / Hn−1 (X1 ∩ X2 ) /. . .

Proof.
The proof follows from the exact sequence (1) of chain complexes by Lemma 1.7.12, because
HnU (X) ∼
= Hn (X), by Corollary 1.7.13 

Observation 1.8.2.
1. As an application, we calculate the homology groups of spheres. Let X = Sm and let
X ± := Sm \{∓em+1 }. The subspaces X + and X − are contractible and therefore H∗ (X ± ) =
0 for all positive ∗.
The Mayer-Vietoris sequence is as follows
δ/ δ
... Hn (X + ∩ X − ) / Hn (X + ) ⊕ Hn (X − ) / Hn (Sm ) / Hn−1 (X + ∩ X −) /. . .

For n > 1 we can deduce from Hn (X ± ) = 0

Hn (Sm ) ∼
= Hn−1 (X + ∩ X − ) ∼
= Hn−1 (Sm−1 ).
The first map is the connecting homomorphism δ and the second map is
Hn−1 (i) : Hn−1 (Sm−1 ) → Hn−1 (X + ∩ X − ) where i is the inclusion of Sm−1 into X + ∩ X −
and this inclusion is a homotopy equivalence. Thus define
D := Hn−1 (i)−1 ◦ δ : Hn (Sm ) → Hn−1 (Sm−1 ) .
This D is an isomorphism for all n > 2.
We have to control what is going on in small degrees and dimensions.
2. We know from the Hurewicz isomorphism that H1 (Sm ) is trivial for m > 1, cf. Corollary
1.3.9. Here, we show this directly via the Mayer-Vietoris sequence:
... → 0 ∼
= H1 (X + ) ⊕ H1 (X − ) → H1 (Sm )
δ
→Z∼ = H0 (X + ∩ X − ) → H0 (X + ) ⊕ H0 (X − ) ∼
= Z ⊕ Z.
We have to understand the map in the second line. Let 1 be a base point of X + ∩ X − .
Then the map on H0 is
[1] 7→ ([1], [1]).
This map is injective and therefore the connecting homomorphism δ : H1 (Sm ) → H0 (X + ∩
X − ) is zero. We find
H1 (Sm ) ∼
= 0, m > 1.

33
 3. We also compute H1 (S1 ) using a Mayer-Vietoris argument and consider the case of n =
1 = m. In this case, the intersection X + ∩ X − splits into two components. We choose
base points P+ ∈ X + and P− ∈ X − . Consider the exact sequence
δ (H0 (i1 ),H0 (i2 ))
0 /H
1 (S
1
) / H0 (X + ∩ X −) / H0 (X + ) ⊕ H0 (X − ) /H
0 (S
1
)
which gives
1 δ /Z ⊕ Z /Z ⊕ Z / Z.
0 /H
1 (S )
The kernel of the last map, the difference of H0 (κ1 ) and H0 (κ2 ),
H0 (X + ) ⊕ H0 (X − ) → H0 (S1 )
is spanned by ([P+ ], [P− ]) and thus isomorphic to Z. This is the image of (H0 (i1 ), H0 (i2 )).
Therefore, the sequence
1 δ /Z ⊕ Z /Z
0 /H
1 (S ) /0

is short exact; thus H1 (S1 ) ∼

= Z is a free abelian group of rank 1. We already knew this
from the Hurewicz isomorphism.
4. We now combine the arguments.
• For 0 < n < m we get by applying D repeatedly,
∼ ∼ ∼
)∼
= = =
Hn (Sm ) /H
n−1 (S
m−1
) /. . . /H
1 (S
m−n+1
= π1 (Sm−n+1 ).
and the latter is trivial by 2.
• Similarly, for 0 < m < n we have similarly
∼ ∼ ∼
)∼
= = =
Hn (Sm ) /H
n−1 (S
m−1
) /. . . /H
n−m+1 (S
1
= 0.
The last claim follows directly by another simple Mayer-Vietoris argument.
• The remaining case 0 < m = n gives a non-vanishing result:
∼ ∼ ∼
)∼
= = =
Hn (Sn ) /H
n−1 (S
n−1
) /. . . /H
1 (S
1
= Z.
We can summarize the result as follows.
Proposition 1.8.3.
The homology groups of spheres are:


 Z ⊕ Z, n = m = 0,

Z, n = 0, m > 0,
Hn (Sm ) ∼
=


 Z, n = m > 0,

0, otherwise.
We specify a generator of Hn (Sn ).

Definition 1.8.4
Let μ0 := [P+ ] − [P− ] ∈ H0 (X + ∩ X − ) ∼
= H0 (S0 ) and let μ1 ∈ H1 (S1 ) ∼
= π1 (S1 ) be given by the
degree one map (i.e. the class of the identity on S , i.e. the class of the loop t 7→ e2πit ).
1

Define the higher μn via the map D from 1.8.2.1 as Dμn = μn−1 . Then μn is called the
fundamental class in Hn (Sn ).
In order to obtain a relative version of the Mayer-Vietoris sequence, we need a tool from
homological algebra.

34
Lemma 1.8.5 (Five-Lemma).
Let
α1 α2 α3 α4
A1 / A2 / A3 / A4 / A5
f1 f2 f3 f4 f5
 β1  β2  β3  β4 
B1 / B2 / B3 / B4 / B5
be a commutative diagram of exact sequences. If the four maps f1 , f2 , f4 , f5 are isomorphisms,
then so is f3 .

Proof.
Again, we are chasing diagrams.

• We show that f3 is injective.

Assume that there is an a ∈ A3 with f3 a = 0. Then β3 f3 a = f4 α3 a = 0, as well. But f4
is injective, thus α3 a = 0. Exactness of the top row gives, that there is an a0 ∈ A2 with
α2 a0 = a. This implies
f 3 α 2 a0 = f 3 a = 0 = β 2 f2 a0 .
Exactness of the bottom row gives us a b ∈ B1 with β1 b = f2 a0 , but f1 is an isomorphism
so we can lift b to a1 ∈ A1 with f1 a1 = b.
Thus f2 α1 a1 = β1 b = f2 a0 and as f2 is injective, this implies that α1 a1 = a0 . So finally we
get that a = α2 a0 = α2 α1 a1 , but the latter is zero, thus a = 0.

• For the surjectivity of f3 , assume b ∈ B3 is given. Move b over to B4 via β3 and set
a := f4−1 β3 b. (Note here, that if β3 b = 0 we actually get a shortcut: Then there is a
b2 ∈ B2 with β2 b2 = b and thus an a2 ∈ A2 with f2 a2 = b2 . Then f3 α2 a2 = β2 b2 = b.)
Consider f5 α4 a. This is equal to β4 β3 b and hence trivial. Therefore α4 a = 0 and thus
there is an a0 ∈ A3 with α3 a0 = a. Then b − f3 a0 is in the kernel of β3 , because

β3 (b − f3 a0 ) = β3 b − f4 α3 a0 = β3 b − f4 a = 0.

Hence we get a b2 ∈ B2 with β2 b2 = b − f3 a0 . Define a2 as f2−1 (b2 ), so a0 + α2 a2 is in A3

and
f3 (a0 + α2 a2 ) = f3 a0 + β2 f2 a2 = f3 a0 + β2 b2 = f3 a0 + b − f3 a0 = b.

We now consider a relative situation, so let X be a topological space with A, B ⊂ X open

in A ∪ B and set U := {A, B}. This is an open covering of A ∪ B ⊂ X. The following diagram

35
of exact sequences combines absolute chains with relative ones:


0 0 0 S (A ∪ B)
4 n
ϕ hhhhhh
hhh
hhhh
hh
   hhhh 
/ Sn (A ∩ B) / Sn (A) ⊕ Sn (B) U
/ S (A ∪ B) /0 S (X)
0 n hhh n
hhhhhhh
h
hhhh
   hh hhhhh 
Δ diff h 
0 / Sn (X) / Sn (X) ⊕ Sn (X) / Sn (X) /0 Sn (X, A ∪ B)
ψ hhhhhh
hh4
hh
hhhh
   hhhh  
0 / Sn (X, A ∩ B) / Sn (X, A) ⊕ Sn (X, B) / Sn (X)/S U (A ∪ B)
n
/0 0

  
0 0 0

Here, ψ is induced by the inclusion ϕ : SnU (A ∪ B) → Sn (A ∪ B), Δ denotes the diagonal

map and diff is the difference map. It is clear that the first two rows are exact. That the third
row is exact follows by a version of the nine-lemma or a direct diagram chase.
Consider the two right-most non-trivial columns in this diagram. Each gives a long exact
sequence in homology and we focus on five terms:
δ
Hn (S∗U (A ∪ B)) / Hn (X) /H U
n (S∗ (X)/S∗ (A ∪ B)) / Hn−1 (S∗U (A ∪ B)) / Hn−1 (X)
Hn (ϕ) Hn (ψ) Hn−1 (ϕ)
  
δ
Hn (A ∪ B) / Hn (X) / Hn (X, A ∪ B) / Hn−1 (A ∪ B) / Hn−1 (X)

Then by the five-lemma 1.8.5, as Hn (ϕ) and Hn−1 (ϕ) are isomorphisms by Corollary 1.7.13, so
is Hn (ψ). This observation, together with the bottom non-trivial exact row of the first diagram,
proves the following
Theorem 1.8.6 (Relative Mayer-Vietoris sequence).
If A, B ⊂ X are open in A ∪ B, then the following sequence is exact:
δ / Hn (X, A ∩ B) / Hn (X, A) ⊕ Hn (X, B) / Hn (X, A ∪ B) δ
... / ...

1.9 Reduced homology and suspension

For any path-connected space, the zeroth homology is isomorphic to the integers, so this copy of
Z is superfluous information and we want to get rid of it. Let pt denote the one-point topological
space. Then for any space X there is a unique continuous map ε : X → pt.

Definition 1.9.1
We define He n (X) := ker(Hn (ε) : Hn (X) → Hn (pt)) and call it the reduced nth homology group
of the space X.

Remarks 1.9.2.
e n (X) ∼
1. Note that H = Hn (X) for all positive n.

36
 e 0 (X) = 0, cf. Proposition 1.3.1.
2. If X is path-connected, then H

3. Choose a base point x ∈ X. Then the composition

{x} ,→ X → {x}

is the identity. Because of Hn (pt) ∼= Hn ({x}), we get from proposition 1.6.8 about weak
retracts
He n (X) ⊕ Hn ({x}) ∼
= Hn (X) .
The retraction r : X → {x} splits the long exact sequence of relative homology for {x} →
X
. . . Hn ({x}) → Hn (X) → Hn (X, {x}) → . . .
e n (X) ∼
and thus we identify reduced homology as relative homology, H = Hn (X, {x}).
4. We can prolong the singular chain complex S∗ (X) and consider the chain complex of free
abelian groups Se∗ (X):
ε
. . . → S1 (X) → S0 (X) −→ Z → 0.
where ε(α) = 1 for every singular 0-simplex α. This is precisely the augmentation we
considered in Proposition 1.3.1. Then for all n > 0,
e ∗ (X) ∼
H = H∗ (Se∗ (X)).

For every continuous map f : X → Y induces a chain map S∗ (f ) : S∗ (X) → S∗ (Y ); for the
evaluation, we have εY ◦ S0 (f ) = εX . We thus obtain the following result:

Lemma 1.9.3.
The assignment X 7→ H∗ (Se∗ (X)) is a functor, i.e. for a continuous map f : X → Y we get
an induced map H∗ (Se∗ (f )) : H∗ (Se∗ (X)) → H∗ (Se∗ (Y )) such that the identity on X induces the
identity and composition of maps is respected.
e ∗ (−) is a functor.
As a consequence, H

Definition 1.9.4
For ∅ 6= A ⊂ X we define
e n (X, A) := Hn (X, A).
H

Since we identified in Remark 1.9.2.3 reduced homology groups with relative homology
groups Hn (X, {x}), we obtain a reduced version of the Mayer-Vietoris sequence. A similar
remark applies to the long exact sequence for a pair of spaces.

Proposition 1.9.5.
For each pair (X, A) of spaces, there is a long exact sequence

... / e n (A)
H / e n (X)
H / e n (X, A)
H / e n−1 (A)
H / ...

and a reduced Mayer-Vietoris sequence, if X1 ∩ X2 6= ∅, which is identical in positive degrees

and ends as
. . . H̃0 (X1 ∩ X2 ) → H̃0 (X1 ) ⊕ H̃0 (X2 ) → H̃0 (X) → 0

Examples 1.9.6.
37
1. Recall that we can express the real projective plane RP 2 as the quotient space of S2
modulo antipodal points or as a quotient of D2 :

RP 2 ∼
= S2 / ± id ∼
= D2 /z ∼ −z for z ∈ S1 .

We use the latter definition and set X = RP 2 , A = X \ {[0, 0]} (which is an open Möbius
strip and hence homotopically equivalent to S1 ) and B = D̊2 . Then

A ∩ B = D̊2 \ {[0, 0]} ' S1 .

Thus we know that H1 (A) ∼ = Z, H1 (B) ∼ = 0 and H2 (A) = H2 (B) = 0. We choose

generators for H1 (A) and H1 (A ∩ B) as follows:

'$
a


− • γ−
&%

Let a be the path that runs along the outer circle in mathematical positive direction half
around starting from the point (1, 0). This is the generator for H1 (A). Let γ be the loop
that runs along the inner circle in mathematical positive direction. This is the generator
for H1 (A ∩ B); note that A ∩ B ' D \ {0}. Then the inclusion iA∩B : A ∩ B → A induces

H1 (iA∩B )[γ] = 2[a].

This suffices to compute H∗ (RP 2 ) up to degree two because the long exact sequence is

e 1 (A∩B) ∼
e 2 (X) → H ∙2
e 1 (A) ∼ e 1 (X) → H
e 0 (A∩B) = 0.
H̃2 (A)⊕H̃2 (B) = 0 → H =Z→H =Z→H

On the two copies of the integers, the map is given by multiplication by two and thus we
obtain:

H2 (RP 2 ) ∼
= ker(2∙ : Z → Z) = 0,
H1 (RP ) = coker(2∙ : Z → Z) ∼
2 ∼
= Z/2Z,
2 ∼
H0 (RP ) = Z.

The higher homology groups are trivial, because there Hn (RP 2 ) is located in a long exact
sequence between trivial groups.

2. We can now calculate the homology groups of bouquets of spaces in terms of the homology
groups of the single spaces, at least in good cases. Let (Xi )i∈I be a family of topological
spaces with chosen basepoints xi ∈ Xi . Consider the bouquet
_
X= Xi .
i∈I

If the inclusion of xi into Xi is pathological, we cannot apply the Mayer-Vietoris sequence

However, we get the following:

38
Proposition 1.9.7.
If there are neighbourhoods Ui of xi ∈ Xi together with a deformation of Ui to {xi }, then we
have for any finite E ⊂ I _ M
e n ( Xi ) ∼
H = e n (Xi ).
H
i∈E i∈E

In the situation above, we say that the space Xi is well-pointed with respect to the point
xi ∈ Xi .

Proof.
First we consider the case of two bouquet summands. We have X1 ∨ U2 ∪ U1 ∨ X2 as an open
covering of X1 ∨ X2 . Since (X1 ∨ U2 ) ∩ (U1 ∨ X2 ) = U1 ∩ U2 is contractible, the Mayer-Vietoris
sequence then gives that Hn (X) ∼ = Hn (X1 ∨ U2 ) ⊕ Hn (U1 ∨ X2 ) for n > 0. For H0 we get the
exact sequence
0→H e 0 (X1 ∨ U2 ) ⊕ H
e 0 (U1 ∨ X2 ) → H̃0 (X) → 0.

By induction we obtain the case of finitely many bouquet summands. 

We also get _ M
He n ( Xi ) ∼
= He n (Xi )
i∈I i∈I

but for this one needs a colimit argument. We postpone that for a while.
We can extend such results to the full relative case. Let A ⊂ X be a closed subspace and
assume that A is a deformation retract of an open neighbourhood A ⊂ U . Let π : X → X/A be
the canonical projection and b = {A} ∈ X/A the image of A. Then X/A is well-pointed with
respect to the point b ∈ X/A by the neighborhood π(U ).

Proposition 1.9.8.
In the situation above
Hn (X, A) ∼
=He n (X/A), 0 6 n.

Proof.
The canonical projection π : X → X/A induces a homeomorphism of pairs (X \ A, U \ A) ∼
=
(X/A \ {b}, π(U ) \ {b}). Consider the following diagram:
∼
= ∼
=
Hn (X, A) / Hn (X, U ) o Hn (X \ A, U \ A)
Hn (π) ∼
= Hn (π)
 ∼ ∼ 
= =
Hn (X/A, b) / Hn (X/A, π(U )) o Hn (X/A \ {b}, π(U ) \ {b})

The upper and lower left arrows are isomorphisms because A is a deformation retract of U ,
the isomorphism in the upper right is a consequence of excision, because A = Aˉ ⊂ U , cf.
Theorem 1.7.14. The lower right one follows from excision as well. The right vertical arrow is
an isomorphism, because we have a homeomorphism of pairs. 

Definition 1.9.9
1. The cone of a topological space X is the topological space

CX := X × [0, 1]/X × {0} .

39
 2. The suspension of a topological space X is the topological space

ΣX := X × [0, 1]/(x1 , 0) ∼ (x2 , 0) and (x1 , 1) ∼ (x2 , 1) for all x1 , x2 ∈ X .

Remarks 1.9.10.
1. The cone over a point p is an interval. The cone over an interval is a triangle, a 2-simplex.
The cone over an n-simplex is an (n+1)-simplex. The cone over Sn is a closed (n+1)-ball.

2. Note that for any topological space X, the cone CX is contractible to its apex. Thus
H̃n (CX) = 0 for all n > 0. Similarly, for A ⊂ X, we have CA ⊂ CX and H̃n (CX, CA) = 0
for all n > 0.

3. The suspension of Sn is ΣSn ∼

= Sn+1 .
4. We have natural embeddings X → CX with x 7→ [x, 1] and CX → ΣX with x 7→ [x, 12 ].
We can see the suspension as two cones, glued together at their bases.

Theorem 1.9.11 (Suspension isomorphism).

Let A ⊂ X be a closed subspace and assume that A is a deformation retract of an open
neighbourhood A ⊂ U . Then

Hn (ΣX, ΣA) ∼
= H̃n−1 (X, A), for all n > 0.

Proof.
1. We first note two equivalences:

X ∪ CA/CA ' X/A, where the cone CA is attached to X by identifying A ⊂ X and the
base A ⊂ CA:

CA

 CA

 CA CA

 C A

 C A

 C A
A X
and CX/(CA ∪ X) ' ΣX/ΣA:
r r
r r
r
A@r r
A@r
r
r A@ r r
rA@
r r r r r r
r
r r Ar @ r
r Ar @
r r r r r r r r r
r
r r r Ar @
CX
r
r r Ar @
ΣX
r r r r r r r r r r r
r
r r r r rA r @ r
r r r Ar @

r r r r r r rA
r r r r r Ar
A rr rr rr rr r
r
@ @
@ A rr rr rr r
r
CA @

@ A rr rr r
r
@ A r
r
@Ar r
r
@ A
r
ΣA

40
 2. Consider the triple (CX, X ∪ CA, CA). We obtain from Proposition 1.6.11 the long exact
sequence on homology groups
/ Hn (CX, CA ∪ X) δ
... / Hn (CX, CA) / H̃
n−1 (X ∪ CA, CA) /. . .

Since cones are contractible, the connecting morphism δ gives us isomorphisms

Hn (CX, CA ∪ X) ∼
= H̃n−1 (X ∪ CA, CA)

3. Using Proposition 1.9.8 and the equivalences from part 1. of the proof, we compute the
right hand side:
1.9.8 1. 1.9.8
H̃n−1 (X ∪ CA, CA) ∼
= H̃n−1 (X ∪ CA/CA) ∼
= H̃n−1 (X/A) ∼
= H̃n−1 (X, A) .

Similarly, we get for the left hand side

1.9.8 1. 1.9.8
H̃n (CX, CA ∪ X) ∼
= H̃n (CX/CA ∪ X) ∼
= H̃n (ΣX/ΣA) ∼
= Hn (ΣX, ΣA).

Note that the corresponding statement is wrong for homotopy groups. We have Σ S2 ∼ = S3 ,
2 ∼ 3 ∼
but π3 (S ) = Z, whereas π4 (S ) = Z/2Z, so homotopy groups (unlike homology groups) do
not satisfy such an easy form of a suspension isomorphism. There is a Freudenthal suspension
theorem for homotopy groups, but that is more complicated. For the above case it yields:
π1+3 (S3 ) ∼
= π1+4 (S4 ) ∼
= . . . =: π1s
where π1s denotes the first stable homotopy group of the sphere.

1.10 Mapping degree

Recall that we defined in Definition 1.8.4 fundamental classes μn ∈ H̃n (Sn ) ∼
= Z for all n > 0.

Definition 1.10.1
A continuous map f : Sn → Sn induces a homomorphism
H̃n (f ) : H̃n (Sn ) → H̃n (Sn )
and therefore we get
H̃n (f )μn = deg(f )μn
with deg(f ) ∈ Z. We call this integer the degree of f .
In the case n = 1 we can relate this notion of a mapping degree to the one defined via the
fundamental group of the 1-sphere: if we represent the generator of π1 (S1 , 1) as the class given
by the loop
ω : [0, 1] → S1 , t 7→ e2πit ,
then the abelianized Hurewicz map, hab : π1 (S1 , 1) → H1 (S1 ), sends by definition 1.8.5 the class
of ω precisely to μ1 ∈ H1 (S1 ) and therefore the naturality of hab
π1 (f )
π1 (S1 , 1) /π
1 (S
1
, 1)
hab hab
 H1 (f )

H1 (S1 ) / H1 (S1 )

41
shows that
def nat def
deg(f )μ1 = H1 (f )μ1 = hab (π1 (f )[w]) = hab (k[w]) = kμ1 .
where k is the degree of f defined via the fundamental group. Thus both notions coincide for
n = 1.
As we know that the connecting homomorphism in the long exact sequence in relative
homology induces an isomorphism between Hn (Dn , Sn−1 ) and H̃n−1 (Sn−1 ), we can consider
degrees of maps f : (Dn , Sn−1 ) → (Dn , Sn−1 ) by defining a fundamental class μ ˉ n := δ −1 μn ∈
Hn (D , S ). Then Hn (f )(ˉ
n n
μn gives a well-defined integer deg(f ) ∈ Z.
μn ) := deg(f )ˉ
The degree of self-maps of S satisfies the following properties:
n

Proposition 1.10.2.
1. If f is homotopic to g, then deg(f ) = deg(g).

2. The degree of the identity on Sn is one.

3. The degree is multiplicative, i.e., deg(g ◦ f ) = deg(g)deg(f ).

4. If f is not surjective, then deg(f ) = 0.

Proof.
The first three properties follow directly from the definition of the degree. If f is not surjective,
then it is homotopic to a constant map and this has degree zero. 

It is true that the group of (pointed) homotopy classes of self-maps of Sn is isomorphic to

Z and thus the first statement in Proposition 1.10.2 can be upgraded to an ’if and only if’, but
we will not prove that here.
Recall that ΣSn ∼= Sn+1 . If f : Sn → Sn is continuous, then the suspension Σ(f ) : ΣSn → ΣSn
n
is given as ΣS 3 [x, t] 7→ [f (x), t].

Lemma 1.10.3.
Suspensions leave the degree invariant, i.e. for f : Sn → Sn we have

deg(Σ(f )) = deg(f ).

In particular, for every integer k ∈ Z there is a continuous map f : Sn → Sn with deg(f ) = k.

Proof.
The suspension isomorphism of Theorem 1.9.11 is induced by a connecting homomorphism
which is functorial by Proposition 1.5.5. Using the isomorphism Hn+1 (Sn+1 ) ∼
= Hn+1 (ΣSn ),
the connecting homomorphism sends μn+1 ∈ Hn+1 (Sn+1 ) to ±μn ∈ H̃n (Sn ). But then the
commutativity of
∼
= Hn+1 (Σf ) ∼
Hn+1 (Sn+1 ) / Hn+1 (ΣSn ) / Hn+1 (ΣSn ) o = Hn+1 (Sn+1 )
δ δ
 
Hn (f )
H̃n (Sn ) / H̃n (Sn )

ensures that ±deg(f )μn = ±deg(Σf )μn with the same sign. 

42
 For the degree of a based self-map of S1 one has an additivity relation deg(ω 00 ?ω 00 ) = degω 00 +
degω 0 with respect to concatenation of paths. We can generalize this to higher dimensions.
Consider the pinch map T : Sn → Sn /Sn−1 ' Sn ∨ Sn and the fold map F : Sn ∨ Sn → Sn . Here,
F is induced by the identity of Sn .

'$

'$ '$
T
'$
&% F
• • −→ • −→
&% &%
&%

Proposition 1.10.4.
For f, g : Sn → Sn based, we have

deg(F ◦ (f ∨ g) ◦ T ) = deg(f ) + deg(g).

Proof.
The map Hn (T ) sends μn to (μn , μn ) ∈ H̃n Sn ⊕ H̃n Sn ∼ = H̃n (Sn ∨ Sn ). Under this isomor-
phism, the map Hn (f ∨ g) corresponds to (μn , μn ) 7→ (H̃n (f )μn , H̃n (g)μn ) and this yields
(deg(f )μn , deg(g)μn ) which under the fold map is sent to the sum. 

We use the mapping degree to show some geometric properties of self-maps of spheres.
Proposition 1.10.5.
Let f (n) : Sn → Sn be the map

(x0 , x1 , . . . , xn ) 7→ (−x0 , x1 , . . . , xn ).

Then f (n) has degree −1.

Proof.
We prove the claim by induction. μ0 was by definition 1.8.4 the difference class [+1] − [−1], and

f (0) ([+1] − [−1]) = [−1] − [+1] = −μ0 .

We defined μn in such a way that Dμn = μn−1 . Therefore, as D is obtained from a connecting
homomorphism and thus by proposition 1.5.5 natural,

Hn (f (n) )μn = Hn (f (n) )D−1 μn−1 = D−1 Hn−1 (f (n−1) )μn−1 = D−1 (−μn−1 ) = −μn .

Corollary 1.10.6.
The antipodal map
A: S n → Sn
x 7→ −x
has degree (−1)n+1 .

43
Proof.
(n)
Let fi : Sn → Sn be the map (x0 , . . . , xn ) 7→ (x0 , . . . , xi−1 , −xi , xi+1 , . . . , xn ). As in Proposition
(n) (n) (n)
1.10.5, one shows that the degree of fi is −1. As A = fn ◦ . . . ◦ f0 , the claim follows from
Proposition 1.10.2.3. 
In particular, for even n, the antipodal map cannot be homotopic to the identity.
Proposition 1.10.7.
Let f, g : Sn → Sn with f (x) 6= g(x) for all x ∈ Sn , then f is homotopic to A ◦ g, with A the
antipodal map. In particular,
deg(f ) = deg(A ◦ g) = deg(A) ∙ deg(g) = (−1)n+1 deg(g).

Proof.
By assumption, for all x ∈ Sn the segment t 7→ (1 − t)f (x) − tg(x) does not pass through the
origin for 0 6 t 6 1. Thus the homotopy
(1 − t)f (x) − tg(x)
H(x, t) =
||(1 − t)f (x) − tg(x)||
with values in Sn connects f to −g = A ◦ g. 

Corollary 1.10.8.
For any f : Sn → Sn with deg(f ) = 0 there exists a point x+ ∈ Sn with f (x+ ) = x+ and a point
x− with f (x− ) = −x− .

Proof.
If f (x) 6= x = id(x) for all x, then by Proposition 1.10.7, f is homotopic to A ◦ id = A. Thus
deg(f ) = deg(A) 6= 0. If f (x) 6= −x for all x, then f is homotopic to A ◦ (−id) and thus
deg(f ) = (−1)n+1 deg(A) 6= 0. 

Corollary 1.10.9.
If n is even, then for any continuous map f : Sn → Sn , there is an x ∈ Sn with f (x) = x or
f (x) = −x.

Proof.
Because n is even, deg(A) = −1 If f (x) 6= x for all x ∈ Sn , by the argument given in the
proof of Corollary 1.10.8, we have deg(f ) = deg(A) = −1. If f (x) 6= −x for all x ∈ Sn , then
deg(f ) = deg(A) deg(−id|Sn ) = 1. Both at the same time is impossible. 

Finally, we can say the following about hairstyles of hedgehogs of arbitrary even dimension:
Proposition 1.10.10 (Hairy Ball theorem).
Any tangential vector field on an even-dimensional sphere S2k vanishes in at least one point.

Proof.
Recall that we can describe the tangent space at a point x ∈ S2k ⊂ R2k+1 as
Tx (S2k ) = {y ∈ R2k+1 |hx, yi = 0}.

44
Assume that V is a tangential vector field which does nowhere vanish, i.e. V (x) 6= 0 for all
x ∈ S2k and V (x) ∈ Tx (S2k ) for all x. Consider the continuous map

f: S2k → S2k
x 7→ ||VV (x)||
(x)

Assume f (x) = x, which amounts to V (x) = ||V (x)||x. But this means that V (x) points into
the direction of x and thus it cannot be tangential. Thus f (x) 6= x for all x ∈ S2k . Similarly,
f (x) = −x yields the same contradiction. Thus the existence of such a V is in contradiction to
Corollary 1.10.9. 

1.11 CW complexes

Definition 1.11.1
A topological space X is called an n-cell, if X is homeomorphic to Rn . The number n is called
the dimension of the cell.

Examples 1.11.2.
1. Every point is a zero cell. The spaces D̊n ∼
= Rn ∼
= Sn \ N are n-cells.
2. Note that an n-cell cannot be an m-cell for n 6= m, because Rn  Rm for n 6= m. This
follows, since Rn ∼
= Rm would imply
Sn−1 ' Rn \ {0} ∼
= Rm \ {0} ' Sm−1 ,
but H̃n−1 (Sn−1 ) ∼= Z for all n and H̃n−1 (Sm−1 ) = 0 for n 6= m. Hence the dimension of a
cell is well-defined.

Definition 1.11.3
A cell decomposition of a space X is a decomposition of X into subspaces, each of which is a
cell of some dimension, i.e., G
X= Xi , X i ∼
= Rni .
i∈I

Here, this decomposition is meant as a set, not as a topological space.

Examples 1.11.4.
1. The boundary of a 3-dimensional cube has a cell decomposition into 8 points, 12 open
edges, and 6 open faces.
2. The standard 3-simplex can be decomposed into 4 zero-cells, six 1-cells, four 2-cells, and
a 3-cell.
3. The n-dimensional sphere (for n > 0) has a cell decomposition into the north pole and
its complement, thus into a single zero-cell and n-cell.

Definition 1.11.5
A topological Hausdorff space X together with a cell decomposition is called a CW complex,
if it satisfies the following conditions:

45
 (a) [Characteristic maps] For every n-cell σ ⊂ X, there is a continuous map Φσ : Dn → X
such that the restriction of Φσ to the interior D̊n is a homeomorphism
∼
=
Φσ |D̊n : D̊n −→ σ

and such that Φσ maps Sn−1 ∼

= ∂Dn to the union of cells of dimension at most n − 1.

ˉ ⊂ X has a non-trivial intersection

(b) [Closure finiteness] For every n-cell σ, the closure σ
with only finitely many cells of X.

ˉ ⊂ σ is closed for all cells

(c) [Weak topology] A subset A ⊂ X is closed, if and only if A ∩ σ
σ in X.

Remarks 1.11.6.
1. The map Φσ as in (a) is called a characteristic map of the cell σ. Its restriction Φσ |Sn−1
to the boundary ∂Dn ∼
= Sn−1 is called an attaching map.

2. Property (b) is the closure finite condition: the closure of every cell is contained in finitely
many cells. This is the ’C’ in CW.

ˉ is closed in X, condition (c) is equivalent to requiring that A ∩ σ

3. Since σ ˉ is closed in X.
Condition (c) can be replaced by the equivalent axiom that a subset A ⊂ X is open, if
and only if A ∩ σˉ is open in σ
ˉ for all cells σ in X.

4. If X is a CW complex with only finitely many cells, then we call X finite. Conditions (b)
and (c) are then automatically fulfilled.

5. Every non-empty CW complex must contain at least one zero cell. Indeed, if n > 0 would
be the lowest dimension of a cell, its boundary Sn−1 could not be taken into cells of
dimension at most n − 1.

6. It follows from axiom (a) that for every n-cell σ, we have σ = Φσ (Dn ).
Proof: From the general inclusion f (B) ⊂ f (B) for continuous maps, we conclude

σ = Φσ (D̊n ) ⊃ Φσ (Dn ) ⊃ σ .

As a compact subspace of a Hausdorff space, Φ σ (Dn ) is closed; since it lies between σ and
σ, we conclude Φσ (Dn ) = σ. In particular the closure σ is compact in X as the continuous
image of the compact set Dn .

ˉ \ σ for an n-cell σ is contained in the union of cells of dimension at most

7. It follows that σ
n − 1.

8. Every finite CW complex is compact, since it is the union of finitely many compact
subspaces Φσ (Dn ).

Examples 1.11.7.
1. The CW structures on a fixed topological space are not unique. For example, S2 with the
CW structure from the cell-decomposition S2 \ {N } t {N } has a single 0-cell consisting
of the north pole and one 2-cell. Projections of a tetrahedron, cube, octahedron or even
less regular bodies to the sphere provide other CW structures.

46
 2. Consider the following spaces with cell decomposition:

Figure 1 has two 0-cells and two 1-cells. The cell boundary of one of the 1-cells is not
contained in 0-cells, cf. Remark 1.11.6.7. Hence axiom (a) is violated. It is satisfied in
figure 2, where we have four 1-cells and four 0-cells. Figure 3 with three 1-cells and three
0-cells is not a CW complex, since the cell closure of one of the 1-cells is not compact.
Figure 4 is again a CW complex.

3. Consider the topological space X = X1 ∪ X2 ⊂ R2 with

1
X1 := {(x, sin ) | 0 < x < 1} ⊂ R2 X2 := {(0, y) | − 1 6 y 6 1}
x
with the topology induced from R2 . We consider a cell decomposition with (0, ±1) as
0-cells and X̊2 and X1 as 1-cells. Here the axiom (a) is violated, since the boundary of
X̊1 is not in the 0-skeleton. (This space is indeed not CW decomposable.)

4. Consider the disc with the following two different cell decompositions:

• The center 0 ∈ D2 and any point on the boundary are declared to be a 0-cell. Every
radius is a 1-cell. Axioms (a) and (b) are satisfied, but axiom (c) is not: take an open
interval on the boundary. Then all intersections with all closures of cells are closed,
but it is not a closed subspace of D2 .
• Any point in the boundary is a 0-cell, the only 2-cell is D̊2 . Axiom (b) is not satisfied,
since the closure of the two-cell has a non-trivial intersection with infinitely many
0-cells. But axioms (a) and (c) hold.

5. The unit interval [0, 1] has a CW structure with two zero cells and one 1-cell. But for
instance the decomposition σ00 = {0}, σk0 = { k1 }, k > 0 and σk1 = ( k+1
1
, k1 ) does not give a
CW structure on [0, 1]. Consider the following countable subset A ⊂ [0, 1]
   
1 1 1
A := + |k ∈ N .
2 k k+1

Then A ∩ σ ˉk1 is precisely the point 12 ( k1 + k+1

1
). This is closed, but the subset A is not
closed in [0, 1], since it does not contain the limit point 0 of A.
Remark 1.11.8.
• Historically, the notion of a simplicial complex plays an important role: a set K of simplices
in Rn is called a simplicial complex or polyhedron, if the following conditions are satisfied:

(a) If K contains a simplex, it contains all faces of this simplex.

(b) The intersection of two simplices of K is either empty or a common face.
(c) K is locally finite, i.e. every point of Rn has a neighborhood that intersects only
finitely many simplices of K.

47
 A simplicial map is a map that takes any k-simplex affinely into a k 0 -simplex with k 0 6 k.
• The subspace |K| := ∪s∈K s ⊂ Rn is called the topological space underlying the complex
K. Simplicial homology can be defined for a simplicial complex; it depends only on |K|.
Simplicial homology has various disadvantages: 2 for example, S2 can be written as a
simplicial complex with 14 simplices only (obtained from the projection of the tetrahedron
to the sphere), but as a CW complex with a 0-cell and a 2-cell only. A 2-torus S1 × S1
can be written as a CW complex with 4 cells, but the smallest simplicial complex has 42
cells.

Definition 1.11.9
S
1. The union X n := σ⊂X,dim(σ)6n σ of cells of dimension at most n is called the n-skeleton
of X.
2. If we have X = X n , but X n−1 ( X, then we say that X is n-dimensional, i.e., dim(X) = n.
3. A subset Y ⊂ X of a CW complex X is called a subcomplex (sub-CW complex), if it has
a cell decomposition by cells of X and if for any cell σ ⊂ Y , also its closure σ in X is
ˉ ⊂Y.
contained in Y , i.e. σ
4. For any subcomplex Y ⊂ X, (X, Y ) is called a CW pair.

We characterize subcomplexes:
Lemma 1.11.10.
Let X be a CW complex and Y ⊂ X be a subspace, together with a cell decomposition by a
subset of cells of X. Then the following conditions are equivalent:
1. Y is a subcomplex, i.e. for any cell σ ⊂ Y , the closure σ in X is contained in Y .
2. Y is closed in X.
3. The cell decomposition (with the cells of X) endows Y with the structure of a CW
complex.

Proof.
2⇒ 1 is trivial: σ ⊂ Y = Y . (Here the bar denotes closure in X, of course.)
1⇒ 2 The topology of X is such by axiom ‘W‘ that Y is closed, if and only if Y ∩ σ is closed
in X for all cells σ in X. Since X is closure finite, σ hits only finitely many cells of X.
Since Y is the union of cells of X, only finitely many of these cells appear in

σ ∩ Y = σ ∩ (σ1 ∪ . . . ∪ σr )

with σi cells of Y . By 1., σ i ⊂ Y , thus

σ ∩ Y = σ ∩ (σ 1 ∪ . . . ∪ σ r )

The intersection of finite unions of closed subsets of X is closed in X, thus this is closed
in X.
Rb
“Computing homology with simplicial chains is like computing integrals
2
a
f (x)dx with approximating
Riemann sums.” (Dold, Lectures in algebraic topology, 1972)

48
 3⇒ 1 For any cell σ ⊂ Y , a characteristic map Φ 0σ for Y exists by 3. It is also characteristic for
X. Remark 1.11.6.6 that σ = Φ0σ (D) now implies that the closure of σ in Y agrees with
the closure of σ in X.

1,2⇒3 1 implies that a characteristic map for a cell σ ⊂ X relative to the complex X is also char-
acteristic relative to Y . This implies the existence of characteristic maps for Y . Closure
finiteness for Y is immediate from the one for X. Thus axioms (a) and (b) hold.
We still have to show that for A ⊂ Y the condition that A ∩ σ Y is closed in Y for all cells
σ ⊂ Y implies that A is closed in Y . (Here σ Y is the closure of σ in Y .) By 2), a set is
closed in Y , if and only if it is closed in X.
It follows that the closure of each cell σ ⊂ Y in Y agrees with the closure of σ in X. It
also follows that it is enough to show that A ∩ σ X is closed in X for all cells σ ⊂ X.
Closure finiteness implies

σ ∩ A = σ ∩ (σ1 ∪ . . . ∪ σr ) ∩ A

where we can assume that σi are all cells in Y . Now

σ ∩ A = σ ∩ (σ 1 ∪ . . . ∪ σ r ) ∩ A

and by assumption A ∩ σ i is closed in X for all i. Thus A ∩ σ is closed in X.

Corollary 1.11.11.
1. Arbitrary intersections and arbitrary unions of subcomplexes are again subcomplexes.

2. The skeleton X n is a subcomplex.

3. Every union of n-cells in X with X n−1 forms a subcomplex.

4. Every cell lies in a finite subcomplex.

Proof.
1. The subcomplexes are closed in X by Lemma 1.11.10, hence their intersection is closed
and by Lemma 1.11.10 a subcomplex. The statement about the union follows directly
from the definition of a subcomplex.

2. and 3. follow from the observation that for an n-cell σ we have that σ = (σ \ σ) ∪ σ is contained
in X n−1 ∪ σ.

4. Induction on the dimension of the cell; then use closure finiteness and σ = Φσ (Dn ).

We want to understand the topology of CW complexes.

Remarks 1.11.12.
1. Cells do not have to be open in X. For example, in the CW structure on [0, 1] with two
zero cells 0 and 1, the 0-cells are not open in [0, 1].

49
 2. If X is a CW complex and σ is an n-cell, then σ is open in the n-skeleton X n . Indeed,
for x ∈ σ, choose a neighborhood U that is open in σ. The intersection U ∩ σ 0 for any
other cell σ 0 is empty, unless σ 0 = σ. (Since there are no cells of higher dimension, only
the boundary of σ intersects other cells.) By the weak topology, then U is also open in
X.
The n-skeleton X n is by corollary 1.11.11.2 a subcomplex and thus by lemma 1.11.10.2
closed in X.

3. We can replace condition (c), that A is closed (resp. open) in X, if and only if the
intersection of A with σˉ is closed (resp. open) in σ for any cell σ, by the equivalent
condition that A is closed (resp. open) in X if and only if A ∩ X n is closed (resp. open)
in X n for all n > 0.

4. A CW-complex X is the direct limit of its skeleta, lim −→ X n . Recall that a direct limit
n
of a of a directed system of topological spaces (X
S )n∈N that is an ascending system of
subspaces X ⊂ X ⊂ . . . is the union X = n>0 X n = tX n / ∼ with the quotient
0 1

topology. Thus a CW complex has the final (“weak”) topology. This is the ’W’
Such a direct limit has the following universal property: for any system of maps ( fn : X n →
Z)n>0 such that fn+1 |X n = fn there is a uniquely determined continuous map f : X → Z
such that f |X n = fn . This is important for constructing maps out of CW complexes by
extending maps recursively on n-cells.
Using the universal property of the sum of topological spaces, the characteristic maps of
all cells combine into a single map

Φ: t σ D nσ → X

which endows X with the quotient topology.

Definition 1.11.13
Let X and Y be CW complexes. A continuous map f : X → Y is called cellular, if f (X n ) ⊂ Y n
for all n > 0.
The category of CW complexes together with cellular maps is rather flexible. Most of the
classical constructions do not lead out of it: for example, CW complexes nicely behave with
respect to collapsing subspaces to points. If X is a CW complex and A ⊂ X a subcomplex,
the cell decomposition of X/A consisting the zero-cell A and the cells of X \ A is again a CW
decomposition. Thus X/A is a CW complex in a canonical way.
However, one has to be careful with respect to products:

Proposition 1.11.14.
If X and Y are CW complexes, then X × Y is a CW complex, if one of the factors is locally
compact.

Proof.
As products of cells are cells, X ×Y inherits a cell decomposition from its factors. Characteristic
maps are products of the characteristic maps for the factors. Closure finitenss follows from
σ × τ = σ × τ . We need to ensure that X × Y carries the weak topology.
To this end, we need a few auxiliary facts:

50
• For two spaces U, V , let C(U, V ) be the set of all continuous maps from U to V . The
topology of C(U, V ) is generated (under finite intersections and arbitrary unions) by the
sets V (K, O) := {f ∈ C(U, V )|f (K) ⊂ O} for compact K ⊂ U and open O ⊂ V . This
is called the compact-open topology. (If U is compact, the compact-open topology is the
one of the metric of uniform convergence,
d(f, g) := sup(f (u), g(u)) ,
u∈U

see [Laures-Szymik, p. 72].)

• If Z is locally compact and all spaces are Hausdorff, there is a homeomorphism
C(X × Z, W ) ∼
= C(X, C(Z, W )) (∗)
of topological spaces. Here, for f : X × Z → W we consider for any given x ∈ X the map
f # (x) : Z → W
z 7→ f (x, z)
This yieds a continous map f # : x 7→ f # (x). The homeomorphism (∗) sends f to f # ∈
C(X, C(Z, W )).
• Using these facts, we show the following Lemma:
Let X, Y and Z be topological spaces satisfying the Hausdorff condition and suppose
that π : X → Y gives Y the quotient topology and that Z is locally compact. Then
π × id : X × Z → Y × Z
gives Y × Z the quotient topology.
We have to show that Y ×Z has the universal property of a quotient space. Hence suppose
that g : Y ×Z → W is a map of sets and assume that the composition g◦(π×id) : X ×Z →
Y × Z → W is continuous.
Under the adjunction (∗), the map g ◦ (π × id) corresponds to the composite
π g#
g̃ : X /Y / C(Z, W ).

which is continuous as the image under the adjunction (∗). Since Y carries the quotient
topology, the map g # is continuous and hence, again by (∗), the map g : Y × Z → W is
continuous, too.
• With the help of this result we consider the characteristic maps of X and Y ,
Φσ : Dnσ → X, for σ an nσ -cell in X
Ψτ : Dmτ → Y, for τ an mτ -cell in Y.
We use the product of topological spaces to combine these maps to a single map and
write X × Y as a target of a map
G G
Φ × Ψ : ( Dnσ ) × ( Dmτ ) → X × Y.
σ τ

To establish that X × Y has the weak topology, we can show that X × Y carries the
quotient topology with respect to this map. We know that each Dnσ is locally
F nσ compact,
thus so is the disjoint union of closed discs. The map id D σ × Ψ gives ( D ) × Y the
F n

quotient topology and by assumption Y is locally compact and therefore, by the result of
the previous point, Φ × idY induces the quotient topology on X × Y .

51
 

Lemma 1.11.15.
If D is a subset of a CW complex X and D intersects each cell in at most one point, then D
is discrete.

Proof.
Let S be an arbitrary subset of D. It suffices to show that S is closed. The closure σ of any
cell σ of X is covered by finitely many cells. Hence S ∩ σ ˉ is finite. Since X is by definition
Hausdorff (thus T1 ), S ∩ σ
ˉ is closed in σ
ˉ . Since this holds for all cells σ, the weak topology
guarantees that S is closed in X. 

Corollary 1.11.16.
Let X be a CW complex.
1. Every compact subset K ⊂ X is contained in a finite union of cells.

2. The space X is compact, if and only if it is a finite CW complex.

3. The space X is locally compact, if and only if it is locally finite, i.e. every point has a
neighborhood that is contained in finitely many cells.

Proof.
• We show that 1. implies one implication in 2. If X is compact, then by 1. it is contained
in a finite union of cells. The converse was shown in Remark 1.11.6.8.

• It is clear that 2. implies 3.

• Thus we only prove 1: consider the intersections K ∩ σ with all cells σ and choose a point
pσ in every non-empty intersection. Then D := {pσ |σ a cell in X} is discrete by Lemma
1.11.15. It is also compact and therefore finite.


Corollary 1.11.17.
If f : K → X is a continuous map from a compact space K to a CW complex X, then the
image of K under f is contained in a finite skeleton.
For the proof just note that the image f (K) is compact in X and apply 1.11.16.1.
Proposition 1.11.18.
Let A be a subcomplex of a CW complex X. Then X × {0} ∪ A × [0, 1] is a strong deformation
retract of X × [0, 1].

Proof.
• Consider first the case when X = Dn and A = ∂Dn = Sn−1 . For r : Dn × [0, 1] →
Dn × {0} ∪ Sn−1 × [0, 1] we can choose the standard retraction of a cylinder onto its
bottom and sides, cf. Figure VII-6 in Bredon, p. 451.

52
 • We inductively construct retractions
ρr : X × {0} ∪ (A × I ∪ X r × I) → X × {0} ∪ A × [0, 1] ,
where ρr+1 extends ρr . Suppose that ρr−1 is given. Then extending to an r-cell of X
amounts to extending on Dr × [0, 1] along Dr × {0} ∪ Sr−1 × [0, 1]. As we have seen, this
can be done.
These maps for all r-cells fit together to a map on the r-skeleton (X × [0, 1])r which, by
the weak topology, fit together to a retract X × [0, 1] → X × {0} ∪ A × [0, 1].


Definition 1.11.19
1. A map p : E → B has the homotopy lifting property, if for any space Y and any homotopy
h : Y × [0, 1] → B
and any map g : Y → E such that p ◦ g = h0 there exists a map H : Y × I → E with
p ◦ H = h such that H(y, 0) = g(y) for all y ∈ Y . As a diagram:
g
Y /
x; E
Hx x p
ι0
x
 x h 
Y ×I / B
Then the continuous map p : E → B is called a fibration.
2. A map ι : A → X has the homotopy extension property, if for any space Y and any map
g : X → Y and h : A × [0, 1] → Y a homotopy such that h|A×{0} = g ◦ ι, then there is an
extension of h to H : X × [0, 1] → Y , compatible with g and h.
As a diagram:
g
YO o v XO
p0 Hv v
v ι
{v
C(I, Y ) o h
A
(Note that by adjunction (∗) a map A → C(I, Y ) amounts to a homotopy A × I → Y .)
Then ι : A → X is called a cofibration.

The property in Proposition 1.11.18 implies that any subcomplex of a CW complex has the
homotopy extension property. Indeed, two maps
g: X→Y and h : A × [0, 1] → Y
such that h|A×{0} = g can be combined to a single map
g̃ : X × {0} ∪ A × [0, 1] → Y ;
a retraction r : X × [0, 1] → X × {0} ∪ A × [0, 1] provides the homotopy extension g̃ ◦ r :
X ×I →Y.

In the following we collect some facts about the topology of CW complexes that we do not
prove:

53
Lemma 1.11.20.
1. For any subcomplex A ⊂ X, there is an open neighborhood U of A in X together with
a strong deformation retract to A. In particular, for each skeleton X n there is an open
neighborhood U in X (and as well in X n+1 ) of X n such that X n is a strong deformation
retract of U .

2. Every CW complex is paracompact, locally path-connected and locally contractible. (A

topological space X is paracompact, if every open cover has a locally finite open refine-
ment.)

3. Every CW complex is semi-locally 1-connected, hence possesses a universal covering space

which has a natural structure of a CW complex.
Lemma 1.11.21.
For the skeleta of a CW complex X, the following homeomorphisms hold:
1. G G
X n \ X n−1 = σ∼
= D̊n .
σ an n-cell σ an n-cell

2. _
X n /X n−1 ∼
= Sn .
σ an n-cell

Proof.
The first claim follows directly from the definition of a CW complex. For the second claim
note that the characteristic maps send the boundary ∂Dn to the (n − 1)-skeleton and hence
for every n-cell in X we get a copy of Sn in the quotient X n /X n−1 . 

Example 1.11.22.
Consider
W6 the two-dimensional CW complex given by the hollow cube W 2 . Then W 2 /W 1 ∼
=
i=1 S 2
, a bouquet of 6 two-dimensional spheres.

1.12 Cellular homology

In the following, X will always be a CW complex.
Lemma 1.12.1.
For the relative homology of the skeleta, we have Hq (X n , X n−1 ) = 0 for all q 6= n > 1.

Proof.
Using the identification of relative homology and reduced homology of the quotient gives
M
Hq (X n , X n−1 ) ∼
= H̃q (X n /X n−1 ) ∼
= H̃q (Sn ).
σ an n-cell

The first isomorphism uses Lemma 1.11.20.1 and Proposition 1.9.8. The last isomorphism uses
Lemma 1.11.21.2 and Proposition 1.9.7. 

Lemma 1.12.2.
Consider the inclusion in : X n → X of the n-skeleton X n into X.

54
 1. The induced map Hn (in ) : Hn (X n ) → Hn (X) is surjective.

2. On the (n + 1)-skeleton we get an isomorphism

Hn (in+1 ) : Hn (X n+1 ) ∼
= Hn (X).

Proof.
• Using the inclusion of skeleta, we can factor in : X n → X as
in
Xn g
g3/5
gkgkgkgkgkxkx; X
ggg
in+1 gggg kkk x
α1 gg k xx
g gggggggg kkkkkk xxxin+3
 gggg / n+2 kk i/ n+2n+3x / ...
X n+1 α2 X X α3 α4

The map Hn (α1 ) : Hn (X n ) → Hn (X n+1 ) is surjective, because Lemma 1.12.1 asserts that
Hn (X n+1 , X n ) = 0. For i > 1 we have the following piece of the long exact sequence of
the pair (X n+i , X n+i−1 )

n+i−1 Hn (αi ) /
0∼
= Hn+1 (X n+i , X n+i−1 ) /H
n (X ) Hn (X n+i ) /H
n (X
n+i
, X n+i−1 ) ∼
= 0.

Therefore Hn (αi ) is an isomorphism in this range. If the complex X is finite-dimensional,

this already proves both claims.

• To deal with the general case, observe that every singular simplex in X, as the continuous
image of the compact standard simplex, has compact image which, by Corollary
Pm 1.11.17
n
is contained in one of the skeleta X . Let a ∈ Sn (X) be a chain, a = i=1 λi βi . Then
we can find an M such that the images of all the βi ’s are contained in the skeleton X M ,
say for M = n + q. Therefore [a] ∈ Hn (X) can be written as Hn (iM )[b], for some class
[b] ∈ Hn (X M ).
Now αq ◦ . . . ◦ α1 induces a surjective map in homology, hence [b] can be written as
Hn (αq ) ◦ . . . ◦ Hn (α1 )[c] for some [c] ∈ Hn (X n ). This implies

[a] = Hn (iM ) ◦ Hn (αq ) ◦ . . . ◦ Hn (α1 )[c] = Hn (in )[c]

thus Hn (in ) is surjective, showing the first assertion.

• Since Hn (in ) = Hn (in+1 ) ◦ Hn (α1 ) and Hn (in ) is surjective by the preceding assertion, it
is clear that Hn (in+1 ) is surjective as well.
Suppose that Hn (in+1 )[a] = 0 for some n-chain a. Then there exists an n + 1-chain β such
that Sn (in+1 a) = ∂b. Using the same argument, there exists M = n + q such that β can
be defined in terms of the M -skeleton. Thus Sn (αn+q−1 ) ◦ . . . ◦ Sn (αn+1 )(a) = ∂b and thus
Hn (αn+q−1 ) ◦ . . . ◦ Hn (αn+1 )([a]) = 0. But all maps are isomorphisms, thus [a] = 0.

Corollary 1.12.3.
For CW complexes X, Y we have

1. If the n-skeleta X n and Y n are homeomorphic, then Hq (X) ∼

= Hq (Y ), for all q < n.

55
 2. If X has no q-cells, then Hq (X) ∼
= 0.

3. In particular, if q exceeds the dimension of X, then Hq (X) ∼

= 0.

Proof.
1. The first claim is a direct consequence of Lemma 1.12.2 which asserts that Hn (X n+1 ) ∼
=
Hn (X).

2. By assumption in 2. X q−1 = X q , therefore we have Hq (X q−1 ) ∼

= Hq (X q ) and the latter
surjects by Lemma 1.12.2.2 onto Hq (X). Hence 2. is reduced to the statement in 3, applied
to X q−1 .

3. We show that Hn (X r ) ∼
= 0 for n > r. Consider the long exact sequence o relative homology

→ Hn+1 (X i , X i−1 ) → Hn (X i−1 ) → Hn (X i ) → Hn (X i , X i−1 ) → . . .

For i < n, the the adjacent relative groups Hn (X i , X i−1 ) are trivial by Lemma 1.12.1. In
this way, we get a chain of isomorphisms

Hn (X r ) ∼ = ... ∼
= Hn (X r−1 ) ∼ = Hn (X 0 ) .

Observation 1.12.4.
1. Let X be a CW complex. Note that by the proof of Lemma 1.12.1
M M
Cn (X) := Hn (X n , X n−1 ) ∼
= H̃n (Sn ) ∼
= Z
σ an n-cell σ an n-cell

is a free abelian group. For n < 0, we let Cn (X) be trivial.

2. If X has only finitely many n-cells, then the abelian group Cn (X) is finitely generated.
If X is a finite CW complex, then C∗ (X) is finitely generated as a chain complex, i.e.
Cn (X) is only non-trivial in finitely many degrees n, and in these degrees, Cn (X) is finitely
generated.

3. Consider the map

δ %
d : Hn (X n , X n−1 ) /H
n−1 (X
n−1
) /H
n−1 (X
n−1
, X n−2 )

where δ is the connecting morphism in the long exact sequence in relative homology from
Theorem 1.6.5 for the pair X n−1 ⊂ X n and % is the map induced by the projection map
Sn−1 (X n−1 ) → Sn−1 (X n−1 , X n−2 ).
The map d is a boundary operator: the composition d2 is % ◦ δ ◦ % ◦ δ, but
ρ δ
δ ◦ % : Hn−1 (X n−1 ) → Hn−1 (X n−1 , X n−2 ) → Hn−2 (X n−2 )

is a composition in a long exact sequence and thus vanishes.

56
Definition 1.12.5
Let X be a CW complex. The cellular chain complex C∗ (X) consists of the free abelian groups
Cn (X) := Hn (X n , X n−1 ) with boundary operator
δ %
d : Hn (X n , X n−1 ) /H
n−1 (X
n−1
) /H
n−1 (X
n−1
, X n−2 )
where % is the map induced by the projection map Sn−1 (X n−1 ) → Sn−1 (X n−1 , X n−2 ).

Theorem 1.12.6 (Comparison of cellular and singular homology).

For every CW complex X, there is an isomorphism Υ : H∗ (C∗ (X), d) ∼
= H∗ (X) relating cellular
and singular homology.

Proof.
Consider the diagram
Cn+1 (X) Hn+1 (X n+1 , X n )
SSS
SSS δ
λ
SSS
SSS
 SS)
n δ 0 Hn (in )
d Hn+1 (X, X ) / Hn (X n ) / Hn (X)
kkk
kkk
kkkk%k
 ukkk
Cn (X) Hn (X n , X n−1 )
SSS
SSS δ
λ
SSS
SSS
 SS)
n−1 Hn−1 (in−1 ) /
δ 0
n−1
d Hn (X, X ) /H
n−1 (X ) Hn−1 (X)
l ll
ll
lll
lllll %
 ul
Cn−1 (X) Hn−1 (X n−1 , X n−2 )
SSS
SSS δ
λ
SSS
SSS
 SS)
n−1 δ 0 Hn−2 (in−2 )
d Hn−1 (X, X ) / Hn−2 (X n−2 ) / Hn−2 (X)
jj j
jjjj
jjjjjjj %
. . . ...tjj

1. The fact that Hk (X k−1 ) ∼

= 0 for all k by Corollary 1.12.3, combined with the long exact
sequence
ρ
0 = Hn−1 (X n−2 ) → Hn−1 (X n−1 ) → Hn−1 (X n−1 , X n−2 )
implies that all occurring %-maps are injective.
2. For every a ∈ Hn (X n ) the element %(a) ∈ Hn (X n , X n−1 ) = Cn (X) is a cycle for d:
d%(a) = %δ%(a) = 0 ,
since δ ◦ % = 0, cf. 1.12.4.3.
3. Conversely, let c ∈ Cn (X) be a d-cycle, thus 0 = dc = %δc. As % is injective by 1.,
we obtain δc = 0. Exactness of the long exact sequence yields that c = %(a) for some
a ∈ Hn (X n ). Hence, % induces an isomorphism
Hn (X n ) ∼
= ker(d : Cn (X) → Cn−1 (X)) .
We have thus expressed the cycles of the cellular complex in terms of the simplical ho-
mology group Hn (X n ).

57
 4. We define Υ̃ : ker(d) → Hn (X) as Υ̃[c] = Hn (in )(a) for c = %(a) with a ∈ Hn (X n ) and
Hn (in ) : Hn (X n ) → Hn (X).

5. The map Υ̃ is surjective, because Hn (in ) is surjective by Lemma 1.12.2.1.

6. In the diagram, the triangles commute, i.e. δ = δ 0 ◦ λ, where λ : Hn+1 (X n+1 , X n ) →

H(X, X n ) comes from the triple (X, X n+1 , X n ).

7. Consider the long exact sequence for the pair (X, X n+1 ):
∼
=
Hn+1 (X n+1 ) / / Hn+1 (X) /H
n+1 (X, X
n+1
) /H
n (X
n+1
) / Hn (X) .

The surjectivity of the first morphism is Lemma 1.12.2.1. The last morphism is an
isomorphism by Lemma 1.12.2.2. The second morphism is zero, hence the morphism
Hn+1 (X, X n+1 ) → Hn (X n+1 ) is injective. Its image is the kernel of an isomorphism and
thus zero. This tells us that Hn+1 (X, X n+1 ) = 0.
Now consider the triple (X, X n+1 , X n ) which by Proposition 1.6.11 yields the exact se-
quence
Hn+1 (X n+1 , X n ) → Hn+1 (X, X n ) → Hn+1 (X, X n+1 ) = 0
which implies that λ : Hn+1 (X n+1 , X n ) → Hn+1 (X, X n ) is surjective.

8. Using this we obtain

7. les
im(δ) = im(δ 0 ) = ker(Hn (in )) ,
where ‘les’ indicates that we used the long exact sequence. As d = % ◦ δ, the injective map
% induces an isomorphism between the image of d and the image of δ. Thus

imd ∼
= imδ = ker Hn (in ) .

9. Taking all facts into account we get that % induces an isomorphism

ker(d : Cn (X) → Cn−1 (X)) ∼ Hn (X n )

=
im(d : Cn+1 (X) → Cn (X)) ker(Hn (in ))

The numerator is 3., the denominator is 8. and the injectivity of %, cf. 1. But for any n,
the sequence
0 / kerHn (in ) / Hn (X n ) / im(Hn (in )) /0

is exact and therefore

Hn (X n )/ker(Hn (in )) ∼
= imHn (in ) ∼
= Hn (X) ,

where the last isomorphism comes from the surjectivity in Lemma 1.12.2.1.

The differential of the cellular complex is very explicitly computable as well:

Proposition 1.12.7 (Cellular Boundary Formula).

Let X be a CW complex. Identify by Observation 1.12.4.1 cells σαn with the generators of the
cellular chain group Cn (X). Denote by dαβ ∈ Z the degree of the map

Δαβ : Sαn−1 → X n−1 → Sβn−1

58
that is the composition of the attaching map of the n-cell σnα with the quotient map collapsing
the complement X n−1 \ σβn−1 of a given (n − 1)-cell σβn−1 to a point. Then the differential of the
cellular chain complex is X
d(σαn ) = dαβ σβn−1 .
β

(This is a finite sum, since the attaching map of σαn has compact image and thus only meets
finitely many cells σβn−1 .)

Proof.
Consider the commuting diagram:

δ∼
= Δαβ∗
Hn (Dnα , ∂Dnα ) / H̃n−1 (∂Dnα ) / H̃n−1 (Sβn−1 )
O
Φα∗ φα∗ qβ∗
 
δ q∗
Hn (X n , X n−1 ) / H̃
n−1 (X
n−1
) / H̃
n−1 (X
n−1
/X n−2 )
SSS j j
SSS jjj
SSS
SSS ρ
jjjjj∼jj
d SS)  ujjj =

Hn−1 (X n−1 , X n−2 )

where
• Φα : Dn → X n is the characteristic map for the cell σαn and φα : ∂Dnα → X n−1 the
attaching map.

• q : X n−1 → X n−1 /X n−2 is the quotient map.

• qβ : X n−1 /X n−2 → Sβn−1 collapses the complement X n−1 \ σβn−1 of the cell σβn−1 to a point.
The resulting quotient sphere is identified with Dβn−1 /∂Dβn−1 via the characteristic map
Φβ .

• Finally, Δαβ := qβ ◦ q ◦ φα is defined as the composition of the attaching map φα of the

cell σαn , following by collapsing the complement of σβn−1 in X n−1 .
The characteristic map Φ α∗ takes a fundamental class μn ∈ Hn (Dnα , ∂Dnα ), cf. comments
before Proposition 1.10.2, to a generator enα of the summand in Hn (X n , X n−1 ) corresponding to
the cell σαn . The commutativity of the left part of the diagram implies that d(enα ) = ρφα∗ δμn =
ρ ◦ φα∗ (μn−1 ).
In terms of the canonical basis of Hn−1 (X n−1 , X n−2 ), the map qβ∗ is the projection on the
Z-summand corresponding to σβn−1 . The commutativity of the diagram now shows the claim.


Examples 1.12.8 (Projective Spaces).

Let K be R, C or H; set m := dimR K. The multiplicative group K ∗ := K \ {0} acts on the
vector space K n+1 via scalar multiplication,

K ∗ × K n+1 \ {0} → K n+1 \ {0}, (λ, v) 7→ λv.

We define KP n = (K n+1 \ {0})/K ∗ and we denote the equivalence class of (x0 , . . . , xn ) in KP n

by [x0 : . . . : xn ]. The n + 1-tuple (x0 , . . . , xn ) is called the homogeneous coordinates of the
point [x0 : . . . : xn ] ∈ KP n .

59
 We define subsets for 0 6 i 6 n

Xi := {[x0 : . . . : xn ]|xi 6= 0, xi+1 = . . . = xn = 0} ⊂ KP n

and consider the map

x0 xi−1
ξ i : Xi → K i , ξi [x0 : . . . : xn ] = ( ,..., ).
xi xi
The map ξi is a homeomorphism; thus Xi is a cell of KP n of real dimension idimR (K) = im.
We can write KP n as X0 t . . . t Xn and we have characteristic maps Φ i : Dmi → KP n as

Φi (y) = Φi (y0 , . . . , yi−1 ) = [y0 : . . . : yi−1 : 1 − ||y|| : 0 : . . . : 0]

with Xi = Φi (D̊mi ). This defines a structure of a CW complex on KP n .

1. First, consider the case K = C. Here, we have a cell in each even dimension 0, 2, 4, . . . , 2n
for CP n . Therefore the cellular chain complex is
(
Z k = 2i, 0 6 i 6 n,
Ck (CP n ) =
0 k = 2i − 1 or k > 2n.

The boundary operator is zero in each degree and thus

(
Z, ∗ = 2i, 0 6 ∗ 6 2n,
H∗ (CP n ) =
0, otherwise.

2. The case of the quaternions, K = H, is similar. Here the cells are in degrees congruent to
zero modulo four, thus
(
Z, ∗ = 4i, 0 6 ∗ 6 4n,
H∗ (HP n ) =
0, otherwise.

3. Non-trivial boundary operators occur in the case of real projective space, RP n . Here, we
have a cell in each dimension up to n and thus the homology of RP n is the homology of
the chain complex
d d d
0 → Cn ∼
= Z → Cn−1 ∼
= Z → . . . → C0 ∼
= Z.

We first consider the case of RP 2 , which we write as a CW-complex with one 0-,1- and
2-cell. The 1-cell is attached to the 0-cell to form a circle. Thus d[a] = 0. The 2-cell, a
disc, is attached to the circle using a map of degree ±2, where the sign is undetermined,
since we did not fix orientations.
Thus the complex becomes
±2 0
0→Z→Z→Z→0
and we read off the homology groups we computed in Example 1.9.6.1:

H0 (RP 2 ) = Z, H1 (RP 2 ) = Z/2Z and H2 (RP 2 ) = 0 .

60
 4. In general, X = RP n has, for 0 6 k 6 n the k-skeleton X k = RP k . The attaching map
of the single k-cell is
φ : ∂Dk ∼= Sk−1 → RP k−1 = Sk−1 /A ,
where A is the antipodal map. We have to compute the degree of the composition
φ
Sk−1 QQQ / Sk−1 / ± id = RP k−1
QQQ
QQQ
QQQ π
φˉk QQ( 
RP k−1 /RP k−2 ∼
= Sk−1

By construction φˉk ◦ A = φˉk , with A the antipodal map, and thus

deg(φˉk ) = deg(φˉk ◦ A) = (−1)k deg(φˉk )

and hence the degree of φˉk is trivial for odd k. The complement Sk−1 \ Sk−2 has two
components X+ , X− and A exchanges these two components. The map φˉk sends X+ and
X− to [X+ ]. Therefore the degree of φˉk is
1.10.4
deg(φˉk ) = deg(F ◦ (id ∨ A) ◦ T ) = deg(id) + deg(A) = 1 + (−1)k .

and d is either zero or two. For the cellular complex, we find

0 ∙2 0
... → Z → Z → Z → 0 .

Thus, depending on n we get


Z
 k=0
n
Hk (RP ) = Z/2Z k < n, k odd


0 otherwise.

for n even.
For odd dimensions n we get

Z
 k = 0, n
n
Hk (RP ) = Z/2Z 0 < k < n, k odd


0 otherwise.

Note that RP 1 ∼
= S1 and RP 3 ∼
= SO(3).

1.13 Homology with coefficients

Let G be an arbitrary abelian group.

Definition 1.13.1
The singular chain complex of a topological space X with coefficients in G, S∗ (X; G), has as
P
elements in Sn (X; G) finite sums of the form N n
i=1 gi αi with gi in G and αi : Δ → X a singular
n-simplex. Addition in Sn (X; G) is given by
N
X N
X N
X
gi αi + hi αi = (gi + hi )αi .
i=1 i=1 i=1

61
The nth (singular) homology group of X with coefficients in G is
Hn (X; G) := Hn (S∗ (X; G))
where the boundary operator ∂ : Sn (X; G) → Sn−1 (X; G) is given by
N
X n
X N
X
j
∂( gi αi ) = (−1) ( gi (αi ◦ dj )) .
i=1 j=0 i=1

We use a similar definition for cellular homology of a CW complex X with coefficients

n n−1 ∼
in
L G. Recall from Observation 1.12.4 that the chain groups are Cn (X) = Hn (X , X ) =
σ an n-cell Z.

Definition 1.13.2 P L
We write c ∈ Cn (X; G) as c = N i=1 gi σi ∈ σ an n-cell G and let the boundary operator d˜ be
˜ = PN gi d(σi ) where d : Cn (X) → Cn−1 (X) is the boundary in the cellular chain
defined by dc i=1
complex of X defined in Observation 1.12.4.
We can transfer Theorem 1.12.6 to the case of homology with coefficients:
Hn (X; G) ∼ ˜
= Hn (C∗ (X; G), d)
for every CW complex X and therefore we denote the latter by Hn (X; G) as well.
Note that Hn (X; Z) = Hn (X) for every space X.
Example 1.13.3.
In the case X = RP 2 , we see that coefficients really make a difference.

• Recall from Example 1.9.6 that for coefficients G = Z we had that H0 (RP 2 ) ∼
= Z,
2 ∼
H1 (RP ) = Z/2Z and H2 (RP ) = 0.
2

• However, for coefficients G = Z/2Z the cellular chain complex looks rather different:

0 / Z/2Z 2=0 / Z/2Z 0 / Z/2Z /0

and therefore Hi (RP 2 ; Z/2Z) ∼

= Z/2Z for 0 6 i 6 2.
• For rational coefficients, we we consider H∗ (RP 2 ; Q). We obtain the cellular complex

/Q 2 /Q 0 /Q
0 /0

But here, multiplication by 2 is an isomorphism and we get H0 (RP 2 ; Q) = Q,

H1 (RP 2 ; Q) = Q/2Q = 0 and H2 (RP 2 ; Q) = 0.

1.14 Tensor products and the universal coefficient theorem

We need to clarify whether homology H∗ (X, G) with coefficients in an abelian group G is
computable from singular homology H∗ (X) and the group G. To see that this can indeed be
done, we need a few more algebraic facts.

Definition 1.14.1
Let A and B be abelian groups. The tensor product A ⊗ B of A and B is the quotient of the
free abelian group generated by the set A × B by the subgroup generated by

62
 (a) (a1 + a2 , b) − (a1 , b) − (a2 , b),

(b) (a, b1 + b2 ) − (a, b1 ) − (a, b2 )

for a1 , a1 , a ∈ A and b1 , b2 , b ∈ B.
We denote the equivalence class of (a, b) in A ⊗ B by a ⊗ b.

Remarks 1.14.2.
1. The relations (a) and (b) imply that λ(a ⊗ b) = (λa) ⊗ b = a ⊗ (λb) for any integer λ ∈ Z
and a ∈ A, b ∈ B.
P
2. Elements of the abelian group A ⊗ B are finite sums of equivalence classes ni=1 λi ai ⊗ bi .

3. The group A ⊗ B is generated by elements a ⊗ b with a ∈ A and b ∈ B.

4. The tensor product is symmetric up to isomorphism and the isomorphism A ⊗ B ∼

= B ⊗A
is given by
X n n
X
λi ai ⊗ bi 7→ λ i b i ⊗ ai .
i=1 i=1

5. The tensor product is associative up to isomorphism:

A ⊗ (B ⊗ C) ∼
= (A ⊗ B) ⊗ C

for all abelian groups A, B, C.

6. For homomorphisms f : A → A0 and g : B → B 0 we get an induced homomorphism

f ⊗ g : A ⊗ B → A0 ⊗ B 0

which is given by (f ⊗ g)(a ⊗ b) = f (a) ⊗ g(b) on generators.

7. The tensor product has the following universal property. For abelian groups A, B, C, the
bilinear maps from A × B to any abelian group C are in bijection to linear maps from
A ⊗ B to C,
A × BK /
x; C
KKK x
KKK x
⊗ KKK x ∃!
% x
A⊗B

8. We have already encountered tensor products in Section 1.13: we have group isomorphisms
for cellular and singular homology with values in an abelian group G:

Sn (X) ⊗ G ∼
= Sn (X; G) and Cn (X) ⊗ G ∼
= Cn (X; G) .

We collect the following properties of tensor products:

Remarks 1.14.3.
1. For every abelian group A, we have isomorphisms

A⊗Z∼
=A∼
=Z⊗A

with a ⊗ n 7→ n.a and inverse a 7→ a ⊗ 1.

63
 2. For every abelian group A, we have

A ⊗ Z/nZ ∼
= A/nA.

Here, nA = {na|a ∈ A} is a subgroup of A for any abelian group A. The isomorphism is

given by
a ⊗ ˉi 7→ ia
where ˉi denotes an equivalence class of i ∈ Z in Z/nZ and ia
ˉ the class of ia ∈ A in A/nA.
α β
3. If 0 → A −→ B −→ C → 0 is a short exact sequence of abelian groups, then for an
arbitrary abelian group D, the sequence
β⊗id
0 / A ⊗ D α⊗id / B ⊗D /C ⊗D /0

is not necessarily exact. For example, the sequence

0 → Z −→ Q −→ Q/Z → 0

is exact, but tensoring with Z/2Z yields

0 → Z ⊗ Z/2Z −→ Q ⊗ Z/2Z −→ Q/Z ⊗ Z/2Z → 0

= Q/2Q ∼
is not, because Q ⊗ Z/2Z ∼ = 0 and tensoring yields

0 → Z/2Z → 0 → 0 → 0

which is obviously not exact.

Lemma 1.14.4.
α β
1. For every abelian group D, (−) ⊗ D is right exact, i.e. if 0 → A −→ B −→ C → 0 is a
short exact sequence, then
α⊗id / β⊗id
A⊗D B⊗D /C ⊗D /0

is exact.
α β
2. If the exact sequence 0 → A −→ B −→ C → 0 is a split short exact sequence, then
β⊗id
0 / A ⊗ D α⊗id / B ⊗D /C ⊗D /0

is exact for any abelian group D.

Proof.
Exercise. 

Suppose, tensoring with the abelian group D would be exact. Then, we could tensor the
exact sequence of abelian groups

0 → Bn (X) → Zn (X) → Hn (X) → 0

with the abelian group G and get the isomorphism between

def def
Hn (X; G) = Hn (S∗ (X) ⊗ G) = Zn (X) ⊗ G/Bn (X) ⊗ G

64
and
Hn (X) ⊗ G = Hn (S∗ (X)) ⊗ G .
A consequence of the failure of the functor (−) ⊗ D to be exact on the left hand side is that
this isomorphism is, in general, wrong.

Definition 1.14.5
Let A be an abelian group. A short exact sequence 0 → R −→ F −→ A → 0 with F a free
abelian group is called a free resolution of A.
Note that in the situation above R is also free abelian, because it can be identified with a
subgroup of the free abelian group F .
Example 1.14.6.
n
For every n > 1, the sequence 0 → Z −→ Z −→ Z/nZ → 0 is a free resolution of the cyclic
group Z/nZ.
Proposition 1.14.7.
Every abelian group possesses a free resolution.
The resolution that we will construct in the proof is called the standard resolution of A.

Proof.
Let F be the free abelian group generated by the elements of the underlying set of A. We denote
by ya the basis element in F corresponding to a ∈ A. Define a homomorphism
p: F → A
P P
a∈A λa ya 7→ a∈A λa a.

Here, λa ∈ Z and this integer is non-zero for only finitely many a ∈ A. By construction, p is
an epimorphism. We set R to be the kernel of p. Since R is a subgroup of a free abelian group
and thus a free abelian group as well, we obtain the desired free resolution of A. 

Definition 1.14.8
i
For two abelian groups A and B and for 0 → R −→ F −→ A → 0 the standard resolution of
A we define
Tor(A, B) := ker(i ⊗ id : R ⊗ B → F ⊗ B).

In general, i ⊗ id is not injective, thus Tor(A, B) is in general not trivial. Unfortunately,

the standard resolution constructed in Proposition 1.14.7 is typically very large. We show that
we can calculate Tor(A, B) via an arbitrary free resolution of A. To that end, we prove the
following result.
Proposition 1.14.9.
i
For every homomorphism f : A → B of abelian groups and for free resolutions 0 → R −→
i 0
F −→ A → 0 and 0 → R0 −→ F 0 −→ B → 0 we have:
1. There exist homomorphisms g : F → F 0 and h : R → R0 , such that the diagram
i p
0 / R / F / A / 0
h g f
 i0
 p0 
0 / R0 / F0 / B / 0

65
 commutes.
If g 0 , h0 are other homomorphisms with this property, then there is a group homomorphism
α : F → R0 with i0 ◦ α = g − g 0 and α ◦ i = h − h0 .
2. For every abelian group D, the map h ⊗ id : R ⊗ D → R0 ⊗ D maps the kernel of i ⊗ id to
the kernel of i0 ⊗ id. The restriction h ⊗ id|ker(i⊗id) is independent of the choice of g and
h. We denote this map by ϕ(f, R → F, R0 → F 0 ).
3. For a homomorphism f 0 : B → C the map ϕ(f 0 ◦ f, R → F, R00 → F 00 ) is equal to the
composition ϕ(f 0 , R0 → F 0 , R00 → F 00 ) ◦ ϕ(f, R → F, R0 → F 0 ).
Note that we can view the morphism α in 1.14.9.1 as a chain homotopy between the chain
maps g, h and g 0 , h0 of free chain complexes
/ i /F /
0 R 0
||
α |
h0 h ||| g 0 g
|

~| i0 

0 / R0 / F0 / 0

Proof.
• To show 1., let {xi } be a basis of F and choose yi ∈ F 0 such that p0 (yi ) = f p(xi ). This
is posisble, since p0 is surjective. We define g : F → F 0 on this basis by g(xi ) = yi . Thus
p0 ◦g(xi ) = p0 (yi ) = f p(xi ). For every r ∈ R we find p0 ◦g(i(r)) = f ◦p◦i(r) = 0. Therefore
g(i(r)) is contained in the kernel of p0 which is equal to the image of i0 . In order to define
h we use the injectivity of i0 , thus h(r) is the unique preimage of g(i(r)) under i0 . This
shows the first claim in 1.
• Given h, h0 and g, g 0 as in 1., we get for x ∈ F that g(x) − g 0 (x) is in the kernel of p0 which
is the image of i0 . Define α as (i0 )−1 (g − g 0 ). Then by construction i0 α = g − g 0 and
i0 (h − h0 ) = (g − g 0 )i = i0 αi .
Here, we first used that the square commutes and then the equation i0 α = g − g 0 . As i0 is
injective, this yields the second relation h − h0 = αi.
• For 2., we consider an element z ∈ ker(i ⊗ id) ⊂ R ⊗ D. Then
(i0 ⊗ id) ◦ (h ⊗ id)(z) = (g ⊗ id) ◦ (i ⊗ id)(z) = 0
and thus (h ⊗ id)(z) is in the kernel of (i0 ⊗ id). If h0 is any other map satisfying the
properties, then we find α as in 1. and compute
(h0 ⊗ id)(z) − (h ⊗ id)(z) = ((h0 − h) ⊗ id)(z) = ((α ◦ i) ⊗ id)(z) = (α ⊗ id)(i ⊗ id)(z) = 0.

• The uniqueness in 2. implies 3.



Corollary 1.14.10.
i0
1. For every free resolution 0 → R0 −→ F 0 −→ A → 0 and any abelian group D, we get a
unique isomorphism
ϕ(idA , R0 → F 0 , R → F ) : ker(i0 ⊗ id) → Tor(A, D).
Thus we can calculate Tor(A, D) with every free resolution of A.

66
 f g
2. Tor is functorial: if A → A0 and B → B 0 are morphisms of abelian groups, we have
morphisms
Tor(f, g) : Tor(A, B) → Tor(A0 , B 0 ) .

Proof.
ι
For the second statement, note that given a free resolution 0 → R → F → A of A, the morphism
0 / R⊗B / F ⊗B / A⊗B / 0
id⊗g id⊗g id⊗g
  
0 / R ⊗ B0 / F ⊗ B0 / A ⊗ B0 / 0
induces a morphism ker(ι ⊗ idB ) → ker(ι ⊗ idB 0 ). 

Examples 1.14.11.
n
1. We compute Tor(Z/nZ, D) for any abelian group D using the free resolution 0 → Z −→
Z −→ Z/nZ → 0. By Definition 1.14.8 and by Corollary 1.14.10, we have
Tor(Z/nZ, D) ∼
= ker(n ⊗ id : Z ⊗ D → Z ⊗ D).
As Z ⊗ D ∼ = D and as n ⊗ id induces the multiplication by n, we get
Tor(Z/nZ, D) ∼ = {d ∈ D|nd = 0} for all n > 1. We thus get the elements in D that are
n-torsion. For this reason, Tor is sometimes called torsion product.
2. From the first example we obtain Tor(Z/nZ, Z/mZ) ∼
= Z/gcd(m, n)Z, because the n-
torsion subgroup in Z/mZ is Z/gcd(m, n)Z.
id
3. For A free abelian, Tor(A, D) ∼= 0 for arbitrary D. To see this, note that 0 → 0 → A −→
A → 0 is a free resolution of A. The kernel is a subgroup of 0 ⊗ D = 0 and hence trivial.
4. For two abelian groups A1 , A2 , D there is an isomorphism
Tor(A1 ⊕ A2 , D) ∼
= Tor(A1 , D) ⊕ Tor(A2 , D).
Consider two free resolutions
0 → Ri → Fi → Ai → 0, i = 1, 2.
Their direct sum
0 → R 1 ⊕ R 2 → F 1 ⊕ F 2 → A 1 ⊕ A2 → 0
is a free resolution of A1 ⊕ A2 with
ker((i1 ⊕ i2 ) ⊗ id) = ker(i1 ⊗ id) ⊕ ker(i2 ⊗ id).

We extend the definition of tensor products to chain complexes of abelian groups:

Definition 1.14.12
Are (C∗ , d) and (C∗0 , d0 ) two chain complexes, then (C∗ ⊗ C∗0 , d⊗ ) is the chain complex with
M
(C∗ ⊗ C∗0 )n = Cp ⊗ Cq0
p+q=n

and with d⊗ (cp ⊗ c0q ) = (dcp ) ⊗ c0q + (−1)p cp ⊗ d0 c0q .

67
Lemma 1.14.13.
The map d⊗ is a differential.

Proof.
The composition is
d⊗ ((dcp ) ⊗ c0q + (−1)p cp ⊗ d0 c0q ) = 0 + (−1)p−1 (dcp ) ⊗ (d0 c0q ) + (−1)p (dcp ) ⊗ (d0 c0q ) + 0 = 0.


Remarks 1.14.14.
1. Let G be an abelian group, then let CG be the chain complex that is concentrated in
degree 0, i.e. with (
G, n = 0,
(CG )n =
0, n 6= 0.
Then for every chain complex (C∗ , d), the tensor product is
(C∗ ⊗ CG )n = Cn ⊗ G, d⊗ = d ⊗ id.
In particular, for every topological space X,
S∗ (X) ⊗ CG ∼
= S∗ (X) ⊗ G = S∗ (X; G) .
This allows us to identify the singular chain complex with values in the abelian group
G with a tensor product of chain complexes. Similarly, for a CW complex X, we get
C∗ (X; G) = C∗ (X) ⊗ CG for cellular homology with values in G.
2. For every pair of spaces (X, A), we therefore introduce the chain complex
S∗ (X, A; G) := S∗ (X, A) ⊗ CG .

3. A map f : (C∗ , d) → (D∗ , dD ) induces a map of chain complexes

f ⊗ id : C∗ ⊗ C∗0 → D∗ ⊗ C∗0 .
In particular, for every continuous (cellular) map we get induced maps on singular (cel-
lular) homology with coefficients.
(
G, ∗ = 0
4. Note that, in generalization of proposition 1.2.18 we have H∗ (pt; G) ∼
=
0, ∗ 6= 0.

Definition 1.14.15
A chain complex C∗ is called free, if the chain group Cn is a free abelian group for all n ∈ Z.
For example, the chain complexes S∗ (X, A) and C∗ (X) are free.
Theorem 1.14.16 (Universal coefficient theorem (algebraic version)).
Let C∗ be a free chain complex and G an abelian group, then for all n ∈ Z we have a split
short exact sequence
0 → Hn (C∗ ) ⊗ G → Hn (C∗ ⊗ G) → Tor(Hn−1 (C∗ ), G) → 0 .
In particular
Hn (C∗ ⊗ G) ∼
= Hn (C∗ ) ⊗ G ⊕ Tor(Hn−1 (C∗ ), G).

68
 This theorem will be a consequence of the more general Theorem 1.14.19. Applying Theorem
1.14.16 to the singular chain complex C∗ := S∗ (X) of a topological space X, cf. Remark
1.14.14.1, we obtain the following:
Theorem 1.14.17 (Universal coefficient theorem (topological version)).
For every space X, there is a split short exact sequence

0 → Hn (X) ⊗ G → Hn (X; G) → Tor(Hn−1 (X), G) → 0 .

Therefore, we get an isomorphism

Hn (X; G) ∼
= Hn (X) ⊗ G ⊕ Tor(Hn−1 (X), G) .

Example 1.14.18.
For the real projective space X = RP 2 , we obtain

Hn (RP 2 ; G) ∼
= Hn (RP 2 ) ⊗ G ⊕ Tor(Hn−1 (RP 2 ), G) .

Recalling from Example 1.9.6.1 and Example 1.12.8

H0 (RP 2 ) = Z , H1 (RP 2 ) = Z2 and H2 (RP 2 ) = 0

we find
H0 (RP 2 ; G) ∼ =Z⊗G∼
= H0 (RP 2 ) ⊗ G ⊕ Tor(H−1 (RP 2 ), G) ∼ = G,
H1 (RP 2 ; G) ∼
= H1 (RP 2 ) ⊗ G ⊕ Tor(H0 (RP 2 ), G) ∼
= G/2G ⊕ 0 ∼
= G/2G,
and
H2 (RP 2 ; G) ∼
= H2 (RP 2 ) ⊗ G ⊕ Tor(H1 (RP 2 ), G) ∼
= Tor(Z/2Z, G).
This reproduces the findings in example 1.13.3.
The universal coefficient theorems 1.14.16 and 1.14.17 are both corollaries of the following
more general statement.
Theorem 1.14.19 (Künneth formula).
For a free chain complex C∗ and a chain complex C∗0 we have the following split exact sequence
for every integer n
L 0 λ
L 0
/ Hn (C∗ ⊗ C 0 )
0 p+q=n Hp (C∗ ) ⊗ Hq (C∗ ) p+q=n−1 Tor(Hp (C∗ ), Hq (C∗ ))
/ / / 0,
∗

i.e. M M
Hn (C∗ ⊗ C∗0 ) ∼
= Hp (C∗ ) ⊗ Hq (C∗0 ) ⊕ Tor(Hp (C∗ ), Hq (C∗0 )) .
p+q=n p+q=n−1
L
The map λ : p+q=n Hp (C∗ ) ⊗ Hq (C∗0 ) → Hn (C∗ ⊗ C∗0 ) in the theorem is given on the
(p, q)-summand by
λ([cp ] ⊗ [c0q ]) := [cp ⊗ c0q ]
for cp ∈ Cp and c0q ∈ Cq0 . By the definition of the tensor product of complexes, this map is
well-defined.
Lemma 1.14.20.
For any free chain complex C∗ with trivial differential and an arbitrary chain complex, C∗0 , λ
is an isomorphism M
λ: Hp (C∗ ) ⊗ Hq (C∗0 ) ∼
= Hn (C∗ ⊗ C∗0 ).
p+q=n

69
Proof.
• We abbreviate the subgroup of cycles in Cq0 with Zq0 and the subgroup of boundaries in Cq0
with Bq0 and use analogous abbreviations for the complex C∗ . By definition 0 → Zq0 −→
d
Cq0 −→ Bq−1
0
→ 0 is a short exact sequence. Since Zp = Cp , the group Zp is free so that
tensoring Zp ⊗ (−) is exact by Remark 1.14.11.3. Thus

0 → Zp ⊗ Zq0 −→ Zp ⊗ Cq0 −→ Zp ⊗ Bq−1

0
→0

is a short exact sequence. This implies that Zp ⊗ Zq0 is the subgroup of cycles in Zp ⊗ Cq0 =
Cp ⊗ Cq0 . Summation over p + q = n yields that the n-cycles in the complex C∗ ⊗ C∗0 are
M
Zn (C∗ ⊗ C∗0 ) = Zp ⊗ Zq0
p+q=n

and the n-boundaries are given by

M
Bn (C∗ ⊗ C∗0 ) = Zp ⊗ Bq0 .
p+q=n

• The sequence
0 → Bq0 −→ Zq0 −→ Hq (C∗0 ) → 0
is exact by definition. Tensoring with Zp is exact, since Zp is free. Tensoring and then
summing over p + q = n yields the exact sequence
M M M
0→ Zp ⊗ Bq0 −→ Zp ⊗ Zq0 −→ Zp ⊗ Hq (C∗0 ) → 0
p+q=n p+q=n p+q=n

The identification of Zn (C∗ ⊗ C∗0 ) and Bn (C∗ ⊗ C∗0 ) in the previous part of the proof
implies that the right-most term is isomorphic to the nth homology group of the complex
C∗ ⊗ C∗0 and therefore
M M
Hn (C∗ ⊗ C∗0 ) ∼
= Zp ⊗ Hq (C∗0 ) = Hp (C∗ ) ⊗ Hq (C∗0 ).
p+q=n p+q=n

Lemma 1.14.21.
Let C∗ be a chain complex of abelian groups. Then there exists a free chain complex F∗ and a
∼
chain map ϕ : F∗ → C∗ which induces an isomorphism in homology, ϕ∗ : H(F∗ ) → H(C∗ ).

Proof.
We already know from the exercises that there exists a free chain complex F∗ whose homology
is isomorphic to the homology of C∗ . Fix an isomorphism ψ∗ : H∗ (F∗ ) → H∗ (C∗ ).
Consider the diagram of short exact sequences

0 / Bn (F∗ ) / Zn (F∗ ) / Hn (F∗ ) / 0

 
θ  ϕ1 ψn
n
  n 
0 / Bn (C∗ ) / Zn (C∗ ) / Hn (C∗ ) / 0

70
where we use the fact that Zn (F∗ ) is free to lift ψn to a map ϕ1n which induces by restriction a
d
map θn . Since Bn−1 (F∗ ) is free, the surjection Cn (F∗ ) → Bn−1 (F∗ ) implies that there is a direct
∼
sum decomposition Cn (F∗ ) = Zn ⊕ Yn such that d|Yn : Yn → Bn−1 (F∗ ). We use the fact that Yn
is free to lift
Y n / Bn−1 (F∗ )

ϕ2n
 θn−1
 
dC
Cn / Bn−1 (C∗ )
Then
ϕ1n ⊕ ϕ2n : Cn (F∗ ) = Zn ⊕ Yn → Cn
is the chain map inducing ψ in homology.


Proof. of Theorem 1.14.19

d
• We consider again the short exact sequence 0 → Zp −→ Cp −→ Bp−1 → 0. Since Bp−1
is free, this sequence is split. Tensoring it with Cq0 gives, by Lemma 1.14.4.2, an exact
sequence. Summing over p + q = n gives the short exact sequence
M M M
0→ Zp ⊗ Cq0 −→ Cp ⊗ Cq0 −→ Bp−1 ⊗ Cq0 → 0 (∗)
p+q=n p+q=n p+q=n

• We define two free chain complexes Z∗ and D∗ with trivial differential and chain groups

(Z∗ )p = Zp and (D∗ )p = Bp−1 .

Then the exact sequence (∗) can be interpreted as a short exact sequence of complexes.
This gives a long exact sequence
δn+1 δ
. . . → Hn+1 (D∗ ⊗C∗0 ) −→ Hn (Z∗ ⊗C∗0 ) → Hn (C∗ ⊗C∗0 ) → Hn (D∗ ⊗C∗0 ) −→
n
Hn−1 (Z∗ ⊗C∗0 ) → . . .

Lemma 1.14.20 gives us a description of H∗ (D∗ ⊗ C∗0 ) and H∗ (Z∗ ⊗ C∗0 ) and therefore we
can consider δn+1 as a map
L L j⊗id
δn+1 : Hp (D∗ ) ⊗ Hq (C∗0 ) = p+q=n+1 Bp−1 ⊗ Hq (C∗0 ) −→
p+q=n+1
L 0
L 0
p+q=n Zp ⊗ Hq (C∗ ) = p+q=n Hp (Z∗ ) ⊗ Hq (C∗ )

with j : Bp ,→ Zp .

• We can cut the long exact sequence in homology in short exact pieces and obtain that all
sequences
0 → coker(δn+1 ) −→ Hn (C∗ ⊗ C∗0 ) −→ ker(δn ) → 0
L 0
are exact. The cokernel of δn+1 is isomorphic to p+q=n (Zp /Bp ) ⊗ Hq (C∗ ) because the
tensor functor is right exact, thus
M
coker(δn+1 ) ∼
= Hp (C∗ ) ⊗ Hq (C∗0 ).
p+q=n

71
 As 0 → Bp −→ Zp −→ Hp (C∗ ) → 0 is a free resolution of the homology group Hp (C∗ ),
we obtain that

Tor(Hp (C∗ ), Hq (C∗0 )) ∼

= ker(j ⊗ id : Bp ⊗ Hq (C∗0 ) → Zp ⊗ Hq (C∗0 ))

and therefore M
Tor(Hp (C∗ ), Hq (C∗0 )) ∼
= ker(δn )
p+q=n−1

which proves the exactness of the Künneth sequence.

• We will first prove that the Künneth sequence is split in the case where both chain
complexes, C∗ and C∗0 , are free. In that case the sequences

0 → Zp → Cp → Bp−1 → 0, 0 → Zq0 → Cq0 → Bq−1

0
→0

are split and we chose retractions r : Cp → Zp and r0 : Cq0 → Zq0 . Consider the two com-
positions
r̃ : Cp −→ Zp  Hp (C∗ ), r̃0 : Cq0 −→Zq0  Hq (C∗0 )
and view H∗ (C∗ ) and H∗ (C∗0 ) as chain complexes with trivial differential. Then these
compositions yield a chain map
r̃⊗r̃ 0
C∗ ⊗ C∗0 −→ H∗ (C∗ ) ⊗ H∗ (C∗0 )

which on homology is
M
Hn (C∗ ⊗ C∗0 ) −→ Hn (H∗ (C∗ ) ⊗ H∗ (C∗0 )) = Hp (C∗ ) ⊗ Hq (C∗0 ).
p+q=n

This map gives the desired splitting.

If the complex C∗0 is not free, chose by the preceding lemma a free chain complex F∗0 ,
together with a chain maps
ψ0 : F 0 → C 0
inducing isomorphism in homology. The naturality of the Künneth exact sequence gives
a commutative diagram
L λ
L
0 /
p+q=n Hp (C∗ ) ⊗ Hq (F∗0 ) / Hn (C∗ ⊗ F∗0 ) /
p+q=n−1 Tor(Hp (C∗ ), Hq (F∗0 )) / 0,

id∗ ⊗ψ∗0 (id⊗ψ 0 )∗ Tor(id∗ ,ψ∗0 )

L 
λ  L 
0 /
p+q=n Hp (C∗ ) ⊗ Hq (C∗0 ) / Hn (C∗ ⊗ C∗0 ) /
p+q=n−1 Tor(Hp (C∗ ), Hq (C∗0 )) / 0,

Since id∗ and ψ∗0 are isomorphisms, so are id ∗ ⊗ ψ∗0 and Tor(id∗ , ψ∗0 ). Thus (id ⊗ ψ 0 )∗ is
an isomorphism and the two exact sequences are isomorphic. Hence both are split.

In the cases we are interested in (singular or cellular chains), the complexes will be free.
The splitting of the Künneth sequence is not natural. We have chosen a splitting of the short
exact sequences in the proof and usually, there is no canonical choice.

72
1.15 The topological Künneth formula
Let X and Y be topological spaces. The Künneth sequence for the singular chain complexes
C∗ = S∗ (X) and C∗0 = S∗ (Y ) of two topological spaces states that
M M
0→ Hp (X) ⊗ Hq (Y ) −→ Hn (S∗ (X) ⊗ S∗ (Y )) −→ Tor(Hp (X), Hq (Y )) → 0
p+q=n p+q=n−1

is exact. We will give a geometric meaning to the group Hn (S∗ (X) ⊗ S∗ (Y )) by showing that
it is actually isomorphic to Hn (X × Y ).
Lemma 1.15.1.
There is a homomorphism of chain complexes × : Sp (X) ⊗ Sq (Y ) −→ Sp+q (X × Y ) for all
p, q > 0 with the following properties.
1. For all points x0 ∈ X, viewed as zero chains, and for any singular q-simplex β : Δq → Y
on Y , the product is the following q-simplex on X × Y :
(x0 × β)(t0 , . . . , tq ) = (x0 , β(t0 , . . . , tq ))

Analogously, for all y0 ∈ Y and any singular p-simplex α : Δp → X on X, we require

(α × y0 )(t0 , . . . , tp ) = (α(t0 , . . . , tp ), y0 ) ∈ X × Y .

2. The map × is natural in X and Y : for f : X → X 0 and g : Y → Y 0

Sp+q (f, g) ◦ (α × β) = (Sp (f ) ◦ α) × (Sq (g) ◦ β).

3. The Leibniz rule holds:

∂(α × β) = ∂(α) × β + (−1)p α × ∂(β).

The map × is called the homology cross product.

Proof.
For p or q equal to zero, we define × as dictated by property (1). Therefore we can assume that
p, q > 1. The method of proof that we will apply here is called method of acyclic models.

• Consider first the specific topological spaces X = Δp , Y = Δq with the specific simplices
α = idΔp , and β = idΔq . If the homology cross product id Δp × idΔq of these simplices were
already defined, then property (3), the Leibniz rule, would force
∂(idΔp × idΔq ) = ∂(idΔp ) × idΔq + (−1)p idΔp × ∂(idΔq ) =: R ∈ Sp+q−1 (Δp × Δq ).
For the boundary of this element R, we get
∂R = ∂ 2 (idΔp )×idΔq +(−1)p−1 ∂(idΔp )×∂(idΔq )+(−1)p ∂(idΔp )×∂(idΔq )+(−1)2p−1 idΔp ×∂ 2 (idΔq ) = 0
so R is a cycle. But Hp+q−1 (Δp × Δq ) = 0 because p + q − 1 > 1 and the space Δp × Δq is
contractible and therefore the complex S∗ (Δp × Δq ) is acyclic, i.e. all its homology except
in degree zero vanishes. Thus R has to be a boundary, so there exists c ∈ Sp+q (Δp × Δq )
with ∂c = R.
We pick one such c and define the homology cross product as
idΔp × idΔq := c .

73
 • Now let X and Y be arbitrary spaces and α : Δp → X, β : Δq → Y arbitrary simplices.
Then Sp (α)(idΔp ) = α and Sq (β)(idΔq ) = β and therefore binaturality (2) dictates

α × β = Sp (α)(idΔp ) × Sq (β)(idΔq ) = Sp+q (α, β)(idΔp × idΔq ).

By construction, this definition satisfies all desired properties.

Note that for spaces X, Y with trivial homology in positive degrees, the Künneth Theorem
1.14.19 yields that Hn (S∗ (X) ⊗ S∗ (Y )) = 0 for positive n.
Lemma 1.15.2.
Suppose that C∗ and C∗0 are two chain complexes which are trivial in negative degrees and
such that Cn is free abelian for all n and Hn (C∗0 ) = 0 for all positive n. Then we have
1. Any two chain maps f∗ , g∗ : C∗ → C∗0 which agree in degree zero, f0 = g0 , are chain
homotopic.
2. If f0 : C0 → C00 is a homomorphism with f0 (∂C1 ) ⊂ ∂C10 then there is a chain map
f∗ : C∗ → C∗0 extending f0 .

Proof.
1. We will define a map Hn : Cn → Cn+1 0
for all n > 0 with ∂Hn + Hn−1 ∂ = fn − gn
inductively. For n = 0 we can take H0 = 0, because f0 = g0 by assumption. Assume that
we have found Hk for k 6 n − 1. Let {xi } be a basis of the free abelian group Cn and
define
yi := fn (xi ) − gn (xi ) − Hn−1 ∂(xi ) ∈ Cn0 .
Then

∂yi =∂fn (xi ) − ∂gn (xi ) − ∂Hn−1 ∂(xi )

=∂fn (xi ) − ∂gn (xi ) − Hn−2 ∂ 2 (xi ) − fn−1 ∂(xi ) + gn−1 ∂(xi )
=0.

But the complex C∗0 is acyclic by assumption. Therefore, yi has to be a boundary and we
define Hn (xi ) = zi by choosing some zi such that ∂zi = yi . Using this definition of Hn (xi )
and then the definition of yi , we find

(∂Hn + Hn−1 ∂)(xi ) = yi + Hn−1 ∂(xi ) = fn (xi ) − gn (xi ).

2. To show the second assertion, we define fn : Cn → Cn0 inductively such that ∂fn = fn−1 ∂
holds. Assume that {xi } is a basis of Cn . Then fn−1 ∂(xi ) is a cycle and thus there exists
yi with ∂yi = fn−1 ∂(xi ), due to the acyclicity of C∗0 . We define fn (xi ) := yi . Then

∂fn (xi ) = ∂yi = fn−1 ∂(xi )

so that (fn ) is a chain map.



We next have to show a uniqueness statement for the chain map constructed in Lemma
1.15.1.

74
Proposition 1.15.3.
Any two binatural families of chain maps fX,Y , gX,Y from S∗ (X) ⊗ S∗ (Y ) to S∗ (X × Y ) which
agree in degree zero and send the zero chain x0 ⊗ y0 ∈ (S∗ (X) ⊗ S∗ (Y ))0 = S0 (X) ⊗ S0 (Y ) to
(x0 , y0 ) ∈ S0 (X × Y ) are chain homotopic.

Proof.
• First we deal with the case X = Δp and Y = Δq for p, q > 0. If f, g : S∗ (Δp ) ⊗ S∗ (Δq ) −→
S∗ (Δp × Δq ) are two chain maps then the complex S∗ (Δp ) ⊗ S∗ (Δq ) is free abelian and the
complex S∗ (Δp × Δq ) is acyclic, so we can apply Lemma 1.15.2 and get a chain homotopy
(Hn )n ,
Hn : (S∗ (Δp ) ⊗ S∗ (Δq ))n −→ Sn+1 (Δp × Δq )
with ∂Hn + Hn−1 ∂ = fn − gn .

• Note that for arbitrary topological spaces X and Y binaturality of f and g implies

fX,Y ◦(S∗ (α)⊗S∗ (β)) = S∗ (α, β)◦fΔp ,Δq and gX,Y ◦(S∗ (α)⊗S∗ (β)) = S∗ (α, β)◦gΔp ,Δq

for all singular simplices α : Δp → X, β : Δq → Y .

We define
Hn : (S∗ (X) ⊗ S∗ (Y ))n → Sn+1 (X × Y )
α ⊗ β 7→ Sn+1 (α, β) ◦ Hn (idΔp ⊗ idΔq ).

This is well-defined and by construction:

∂Hn (α ⊗ β) = ∂Sn+1 (α, β) ◦ Hn (idΔp ⊗ idΔq ) [Definition]

= Sn (α, β)∂Hn (idΔp ⊗ idΔq ) [S∗ (α, β) is a chain map]
= Sn (α, β) ◦ (−Hn−1 ∂(idΔp ⊗ idΔq ) + fn (idΔp ⊗ idΔq ) − gn (idΔp ⊗ idΔq ))
= fn (α ⊗ β) − gn (α ⊗ β) − Hn−1 ∂(α ⊗ β).

For the last step, observe that

∂i (α) = α ◦ di = Sp (α)(idΔp ◦ di )

implies ∂α = Sp (α)(idΔp ◦ ∂) and thus

Sn (α, β) ◦ Hn−1 ∂(idΔp ⊗ idΔq ) = Sn (α, β) ◦ Hn−1 (idΔp ◦ ∂ ⊗ idΔq + (−1)p idΔp ⊗ idΔq ◦ ∂)
= Hn−1 (α ⊗ β) ◦ (∂ ⊗ id + (−1)p id ⊗ ∂)
= Hn−1 ∂(α ⊗ β)

where we used the definition of the differential and the definition of Hn−1 .

We finally need to ensure the existence of a homotopy inverse.

Proposition 1.15.4.
1. There is a chain map S∗ (X × Y ) −→ S∗ (X) ⊗ S∗ (Y ) for all spaces X and Y such that
this map is natural in X and Y and such that in degree zero this map sends (x0 , y0 ) to
x0 ⊗ y0 for all x0 ∈ X and y0 ∈ Y .

75
 2. Any two such maps are chain homotopic.

Proof.
• For the first assertion, let X = Δn = Y for n > 0 and set C∗ = S∗ (Δn × Δn ) and
C∗0 = S∗ (Δn ) ⊗ S∗ (Δn ). Set f0 : C0 → C00 as dictated by the condition. Then by Lemma
1.15.2 there is a chain map (fm )m , fm : Sm (Δn × Δn ) → (S∗ (Δn ) ⊗ S∗ (Δn ))m . For a
singular simplex α : Δn → X × Y , we define

f˜n (α) := (S∗ (p1 ◦ α)) ⊗ S∗ ((p2 ◦ α)) ◦ fn (ΔΔn ).

Here, ΔΔn : Δn −→ Δn × Δn is the diagonal map viewed as a singular n-simplex ΔΔn ∈

p1 p2
Sn (Δn × Δn ) and the pi are the projection maps X ←− X × Y −→ Y :

fn
Sn (Δn × Δn ) / (S∗ (Δn ) ⊗ S∗ (Δn ))n
S∗ (α)⊗S∗ (α)

(S∗ (X × Y ) ⊗ S∗ (X × Y ))n
S∗ (p1 )⊗S∗ (p2 )

(S∗ (X) ⊗ S∗ (Y ))n .

• The second assertion follows as in the proof of Proposition 1.15.3.

Theorem 1.15.5 (Eilenberg-Zilber).

The homology cross product × : S∗ (X) ⊗ S∗ (Y ) −→ S∗ (X × Y ) is a homotopy equivalence of
chain complexes.

Proof.
Let f be any natural chain map S∗ (X × Y ) → S∗ (X) ⊗ S∗ (Y ) from Proposition 1.15.4 with
f0 (x0 , y0 ) = x0 ⊗ y0 for any pair of points. Then the composition

f ◦ (− × −) : S∗ (X) ⊗ S∗ (Y ) → S∗ (X) ⊗ S∗ (Y )

is a chain map that sends x0 ⊗ y0 to itself. Using Lemma 1.15.2 for X = Δp and Y = Δq and
then extending by binaturality again, we get that the identity and f ◦ (− × −) are homotopic.
Similarly we get that the composition (− × −) ◦ f is homotopic to the identity. 

Corollary 1.15.6 (Topological Künneth formula).

For any pair of spaces X and Y the following sequence is split short exact
M M
0→ Hp (X) ⊗ Hq (Y ) −→ Hn (X × Y ) −→ Tor(Hp (X), Hq (Y )) → 0.
p+q=n p+q=n−1

The sequence is natural in X and Y , but the splitting is not.

Examples 1.15.7.

76
 1. For the n-torus T n = (S1 )n we get inductively
n
= Z( i ) ;
Hi (T n ) ∼

all Tor-groups vanish, since all homology groups are free.

2. For a space of the form X × Sn we obtain

Hq (X × Sn ) ∼
= Hq (X) ⊕ Hq−n (X).

Remarks 1.15.8.
1. There is also a relative version of the Künneth formula. The homology cross product in
its relative form is a map

× : Sp (X, A) ⊗ Sq (Y, B) −→ Sp+q (X × Y, A × Y ∪ X × B)

and the corresponding relative homology appears in a topological Künneth formula.

2. In particular for A and B a point we get a reduced Künneth formula which is based on a
homology cross product

S̃p (X) ⊗ S̃q (Y ) −→ Sp+q (X × Y, X ∨ Y )

and in good cases (see Proposition 1.9.8) the relevant relative homology H̃n (X ×Y, X ∨Y )
is isomorphic to the homology H̃p+q (X ∧ Y ) of the smash product X ∧ Y = X × Y /X ∨ Y .

2 Singular cohomology
2.1 Definition of singular cohomology

Definition 2.1.1
A cochain complex of abelian groups is a sequence (C n )n∈Z of abelian groups C n together with
homomorphisms δ : C n → C n+1 increasing the degree such that δ 2 = 0. The map δ is called
coboundary operator. The group

ker(δ : C n → C n+1 )
H n (C ∗ ) =
im(δ : C n−1 → C n )

is the nth cohomology group of C ∗ .

If (C∗ , d) is a chain complex, we can define Dn := C−n and this is a cochain complex because
the fact that d lowers degree by one gives a map d : C−n = Dn → C−n−1 = Dn+1 increasing the
degree. We therefore do not need a theory of cochain complexes; it is just convenient to switch
to cochain notation.

Definition 2.1.2

77
For two cochain complexes (C ∗ , δ) and (C̃ ∗ , δ̃) a map of cochain complexes from C ∗ to C̃ ∗ is a
sequence of homomorphisms f n : C n → C̃ n such that f n+1 ◦ δ = δ̃ ◦ f n for all n.
f n+1
C n+1
O
/
C̃ n+1
O
δ δ̃
fn
Cn /
C̃ n .

Maps of cochain complexes induce maps on cohomology. In particular, we get for a short
exact sequence of cochain complexes a long exact sequence with functorial connecting homo-
morphisms, as a consequence of Proposition 1.5.5.

Definition 2.1.3
1. Let G be any abelian group and X a topological space. Then the abelian group
S n (X; G) := Hom(Sn (X), G)
is called the nth cochain group of X with coefficients in G. In the special case G = Z, we
call S n (X) := Hom(Sn (X), Z) the nth singular cochain group of X.
2. The dual δ = ∂ ∗ = Hom(∂, idG ) of the boundary operator ∂ endows these groups with
the structure of a cochain complex.
3. The quotient group
ker(δ : S n (X; G) → S n+1 (X; G))
H n (X; G) =
im(δ : S n−1 (X; G) → S n (X; G))
is the nth cohomology group of X with coefficients in G.

Remarks 2.1.4.
1. Explicitly, the differential on a G-valued singular n-cochain ϕ : Sn (X) → G is given by
precomposition:
δ(ϕ) : Sn+1 (X) → G
α 7→ ϕ(∂α)
2. We evaluate δ 2 (ϕ) on a singular (n + 2)-simplex β : Δn+2 → X:
δ 2 (ϕ)(β) = (δϕ)(∂β) = ϕ(∂ 2 β) = 0 .
Thus δ 2 = 0 and we indeed have a cochain complex.
3. For a continuous map f : X → Y , denote the induced map S∗ (f ) of singular chains by
f∗ . Then the dual map
S ∗ (f ) = f ∗ : S ∗ (Y ; G) → S ∗ (X; G)
is defined as usual by precomposition: for ϕ ∈ S ∗ (Y ; G) and α ∈ S∗ (X),
f ∗ (ϕ)(α) = ϕ(f∗ α) ∈ G.

This is indeed a map S ∗ (Y ; G) → S ∗ (X; G) of cochain complexes:

hδ ◦ f ∗ ϕ, αi = hf ∗ ϕ, ∂αi = hϕ, f∗ ∂αi = hϕ, ∂f∗ αi = hδϕ, f∗ αi = hf ∗ δϕ, αi
for all chains α ∈ S∗ (X) and cochains ϕ ∈ S ∗ (Y ; G) and where we write hψ, αi := ψ(α).

78
 4. Note that
h(f ∗ ◦ g ∗ )ϕ, αi = hg ∗ ϕ, f αi = hϕ, gf αi = h(g ◦ f )∗ ϕ, αi .
Thus S n (−; G) and H n (−; G) are contravariant functors from the category of topological
spaces and continuous maps to the category of abelian groups.

Definition 2.1.5
1. For two abelian groups A and G, and ϕ ∈ Hom(A, G), a ∈ A the Kronecker pairing is
the G-valued evaluation
h−, −i : Hom(A, G) ⊗ A −→ G, hϕ, ai = ϕ(a) ∈ G.

2. For a homomorphism f : B → A and ϕ ∈ Hom(A, G), we have f ∗ (ϕ) ∈ Hom(B, G). On

b ∈ B, this takes the value
hf ∗ ϕ, bi = hϕ, f bi = ϕ(f (b)) ∈ G .

3. For a chain complex C∗ of abelian groups and the cochain complex C n := Hom(Cn , G),
we define a pairing with values in G:
h−, −i : C n ⊗ Cn → G, ϕ ⊗ a 7→ hϕ, ai = ϕ(a).

4. In particular, for A = Sn (X) a singular chain group and S n (X, G) = Hom(Sn (X), G), we
get a Kronecker pairing with values in G
h−, −i : S n (X; G) ⊗ Sn (X) → G.

5. For ∂ : Sn+1 (X) → Sn (X) and a ∈ Sn+1 (X) we get

hδϕ, ai = hϕ, ∂ai = ϕ(∂(a)).

Lemma 2.1.6.
Let C∗ be a complex of abelian groups and C n := Hom(Cn , G) for some abelian group G.
The Kronecker pairing h−, −i : C n ⊗ Cn → G induces a well-defined pairing on the level of
cohomology and homology, i.e. we obtain an induced map
h−, −i : H n (C ∗ ) ⊗ Hn (C∗ ) → G .

Proof.
Let ϕ be a cocycle, δϕ = 0. Then
hϕ, a + ∂bi = hϕ, ai + hϕ, ∂bi = hϕ, ai + hδϕ, bi = hϕ, ai.
Thus hϕ, −i descends to homology. Assume that ϕ is a coboundary, ϕ = δψ and a is a cycle,
∂a = 0. Then we get
hϕ, ai = hδψ, ai = hψ, ∂ai = 0.
Therefore h−, −i induces a well-defined G-valued pairing on Hn (C∗ ) and H n (C ∗ ). 

Changing perspective, this pairing induces a map

κ : H n (C ∗ ) −→ Hom(Hn (C∗ ), G)
via κ[ϕ][a] := hϕ, ai. How much of the cohomology H n (C ∗ ) does the map κ see, i.e. is it
surjective, does it have a kernel?

79
2.2 Universal coefficient theorem for cohomology
Dual to Tor which was defined using the tensor product (−) ⊗ (−), we consider a corresponding
construction for the functor Hom(−, −). Let R be a ring; for a short exact sequence of R-
modules
α β
0 → A −→ B −→ C → 0
given an R-module G, the sequence of abelian groups
β∗ α∗
0 → Hom(C, G) −→ Hom(B, G) −→ Hom(A, G) → 0
does not have to be exact. A problem can arise with respect to the surjectivity at the end.
n∙
As an example, consider the short exact sequence 0 → Z −→ Z −→ Z/nZ → 0 for a natural
number n > 1. Then the sequence
n∙
0 → Hom(Z/nZ, Z) = 0 −→ Hom(Z, Z) ∼ = Z −→ Hom(Z, Z) ∼ =Z
is exact, but multiplication by n is not surjective, so we cannot prolong this sequence to the
right with a zero.

Definition 2.2.1
i
Let A and G be abelian groups. For a free resolution 0 → R −→ F −→ A → 0 of A, we call
ι∗
Ext(A, G) the cokernel of Hom(i, G) : Hom(F, G) → Hom(R, G).

Remarks 2.2.2.
1. Ext comes from ’extension’, because one can describe Ext(A, G) in terms of extensions of
abelian groups.
2. As for Tor it is true that Ext(A, G) is independent of the free resolution of A.
3. Note that Ext(A, G) is covariant in G and contravariant in A: for homomorphisms f : A →
B and g : G → H we get morphisms of abelian groups
f ∗ : Ext(B, G) → Ext(A, G), g∗ : Ext(A, G) → Ext(A, H).

4. For any family of abelian groups (Gi , i ∈ I)

Y Y
Ext(A, Gi ) ∼
= Ext(A, Gi )
i∈I i∈I

and M Y
Ext( Gi , B) ∼
= Ext(Gi , B).
i∈I i∈I

5. The group Ext(A, G) is trivial, if A is free abelian. In this case, chose R = 0 and
∼
= 0
F = A. The free resolution 0 → A → A → 0 gives Ext(A, G) = coker(Hom(A, G) →
Hom(0, G)) = 0.
6. We compute
Ext(Z/nZ, G) ∼
= G/nG.
∙n
To this end, we use the free resolution 0 → Z → Z → Z/nZ → 0 and have to compute
the cokernel of
Hom(Z, G) → Hom(Z, G)
ϕ 7→ ϕ(n.−)
Identifying Hom(Z, G) ∼
= G via ϕ 7→ ϕ(1), the right hand side is identified with ϕ(n.1) =
n.ϕ(1).

80
 7. In particular, Ext(A, G) is trivial, if G is divisible, i.e. for all g ∈ G and n ∈ Z \ {0} there
exists t ∈ G with g = nt. For example this holds if G is isomorphic to one of the groups
Q, R, Q/Z, or C.

8. For natural numbers n and m

Ext(Z/nZ, Z/mZ) ∼
= Z/gcd(n, m)Z.

To state the two main results of this section, we need two simple observations:

Lemma 2.2.3.
ι π
Let 0 → A → B → C → 0 be a split short exact sequence of abelian groups. For any abelian
group G, the sequence
π∗ ι∗
0 → Hom(C, G) → Hom(B, G) → Hom(A, G) → 0

is split exact.

Proof.
The sequence is exact at Hom(C, G) and Hom(B, G) in any case, cf. one of the next exercises.
Chose a retract r : B → A for ι, i.e. r ◦ ι = idA . Then

Hom(A, G) → Hom(B, G)
ϕ 7→ ϕ ◦ r

is a section of ι∗ . 

Theorem 2.2.4 (Universal coefficient theorem for cochain complexes).

Let G be an abelian group. For every free chain complex C∗ and C ∗ := Hom(C∗ , G) the following
sequence is exact and splits
κ
0 → Ext(Hn−1 (C∗ ), G) −→ H n (C ∗ ) −→ Hom(Hn (C∗ ), G) → 0.

We specify to the singular chain complex, Cn = Sn (X) for a topological space X, which has
free chain groups.

Theorem 2.2.5 (Universal coefficient theorem for singular cohomology).

Let X be an arbitrary space. Then the sequence
κ
0 → Ext(Hn−1 (X), G) −→ H n (X; G) −→ Hom(Hn (X), G) → 0

is split exact, with κ as defined after Lemma 2.1.6.

Proof. of Theorem 2.2.4

• Let C∗ be a free chain complex and C ∗ := Hom(C∗ , G). Then the sequence 0 → Zn −→
∂
Cn −→ Bn−1 → 0 is split exact, since Bn is free as a subgroup of the free group Cn and
thus a section of ∂ can be constructed. By lemma 2.2.3, the G-dual sequence

0 → B n−1 −→ C n −→ Z n → 0

81
 is short exact. It gives a short exact sequence of cochain complexes, where we view B ∗
and Z ∗ as cochain complexes with trivial differential. This yields a long exact sequence
on the level of cohomology groups
∂ ∂
. . . −→ Z n−1 −→ B n−1 −→ H n (C ∗ ) −→ Z n −→ B n −→ . . . (∗)
Here, ∂ denotes the connecting homomorphism in the cohomological case. By the very
definition of the connecting homomorphism we get that ∂ is the dual of the inclusion
in : Bn ⊂ Zn , ∂ = i∗n :
Cn / Zn 3 ϕ

δ

Bn δ
/ C n+1
A preimage ψ ∈ C n of ϕ ∈ Z n is any morphism ψ : Cn → G that restricts to ϕ on
the subgroup Zn of cycles. It is mapped to ψ ◦ d ∈ C n+1 . Here, only the value of ψ on
boundaries matters, we can thus replace ψ ◦ d = ϕ ◦ d. We are looking for ϕ̃ : Bn → G
such that ϕ̃ ◦ d = ϕ ◦ d. This is achieved by the restriction of ϕ to the boundaries Bn .
• We cut the long exact sequence (∗) into the short one
0 → coker(i∗n−1 ) −→ H n (C ∗ ) −→ ker(i∗n ) → 0
and hence we have to compute the kernel and the cokernel of i∗n : Hom(Zn , G) →
Hom(Bn , G).
• The exact sequence obtained from applying the left exact Hom-functor to the short exact
ιn π
sequence 0 → Bn → Zn → Hn (C∗ ) → 0
π∗ n i∗
0 → Hom(Hn (C∗ ), G) −→ Hom(Zn , G) −→ Hom(Bn , G)
tells us that the kernel of i∗n is the image of π ∗ and due to the injectivity of π ∗ this is
isomorphic to Hom(Hn (C∗ ), G).
• The sequence
in−1
0 → Bn−1 −→ Zn−1 −→ Hn−1 (C∗ ) → 0
is a free resolution of the homology group Hn−1 (C∗ ) and therefore the cokernel of i∗n−1 by
Definition 2.2.1 equals Ext(Hn−1 (C∗ ), G).


Examples 2.2.6.
1. We know from Example 1.12.8.2 that the homology of complex porjective space CP n is
free with (
Z, 0 6 k 6 2n, k even,
Hk (CP n ) ∼
=
0, otherwise.
For free groups, the Ext-groups vanish by 2.2.2.5 and thus H k (CP n ) ∼
= Hom(Hk (CP n ), Z).
The cohomology is in this example given by the Z-dual of the homology.
2. Recall from Proposition 1.8.3 that
(
Z m = 0, n,
H m (Sn ) ∼
=
0 otherwise.
for n > 1. For later use we fix a class νn ∈ H n (Sn ) with hνn , μn i = 1.

82
2.3 Axiomatic description of a cohomology theory
Before we give an axiomatic description of singular cohomology, we establish some consequences
of some of the results we proved for singular homology.

Remarks 2.3.1.
1. For a chain map f : C∗ → C∗0 (such as the barycentric subdivision) the G-dual map

f ∗ = Hom(f, G) : Hom(C∗0 , G) −→ Hom(C∗ , G)

is a map of cochain complexes.

0
2. If (Hn : Cn → Cn+1 )n is a chain homotopy, then the G-dual
0
(H n := Hom(Hn , G) : Hom(Cn+1 , G) → Hom(Cn , G))n

is a cochain homotopy. Thus if ∂Hn + Hn−1 ∂ = fn − gn , then H n δ + δH n−1 = f n − g n .

3. We have seen in Lemma 2.2.3 that for a split exact sequence 0 → B1 −→ B2 −→ B3 → 0

the dual sequence 0 → Hom(B3 , G) −→ Hom(B2 , G) −→ Hom(B1 , G) → 0 is exact. For
instance, if A is a subspace of X, then the short exact sequence of chain complexes

0 → S∗ (A) −→ S∗ (X) −→ S∗ (X, A) → 0

is split. To see this, we define a retraction rn : Sn (X) → Sn (A) on a generator α : Δn → X

via (
α, if α(Δn ) ⊂ A,
rn (α) =
0, otherwise.
Therefore 0 → S ∗ (X, A) −→ S ∗ (X) −→ S ∗ (A) → 0 is a short exact sequence of cochain
complexes and gives rise to a long exact sequence in cohomology.

With the help of these facts we can show that singular cohomology satisfies the (Eilenberg-
Steenrod) axioms of a cohomology theory:

1. The assignment (X, A) 7→ H n (X, A) is a contravariant functor from the category of pairs
of topological spaces to the category of abelian groups.

2. If f, g : (X, A) → (Y, B) are two homotopic maps of pairs of topological spaces, then
H n (f ) = H n (g) : H n (Y, B) → H n (X, A).

3. For any subspace A ⊂ X there is a natural connecting homomorphism ∂ : H n (A) →

H n+1 (X, A) increasing the degree.

4. For any subspace A ⊂ X we get a long exact sequence

∂ H n (i) ∂
. . . −→ H n (X, A) −→ H n (X) −→ H n (A) −→ . . .

ˉ ⊂ Å ⊂ A ⊂ X
5. Excision holds, i.e. for W ⊂ W

H n (i) : H n (X, A) ∼
= H n (X \ W, A \ W ), for all n > 0.

83
 6. For the one-point space pt, we have
(
G, n = 0,
H n (pt) ∼
=
6 0.
0, n =

This is called the axiom about the coefficients or the dimension axiom.

7. Singular cohomology is additive under disjoint union:

G Y
H n ( Xi ) ∼
= H n (Xi ).
i∈I i∈I

For singular cohomology with coefficients in G we have an analogous set of axioms. There
are important so-called generalized cohomology theories like topological K-theory or cobordism
theories that satisfy all axioms but the dimension axiom. (For K-theory on a point, we get the
integers in every even degree.)

2.4 Cap product

The rough idea of the cap product is to evaluate a (relative) cochain of smaller or equal degree
on a piece of a (relative) chain to get a relative chain of smaller degree. (This is a partial
evaluation of cochains on chains.)

Definition 2.4.1
Let a : Δn → X be a singular n-simplex on X and let 0 6 q 6 n.

• The (n−q)-dimensional front face of the singular simplex a on X is the (n−q)-dimensional

singular simplex on X
a
F (a) = F n−q (a) = a ◦ i : Δn−q ,→ Δn −→ X

where i is the inclusion i : Δn−q ,→ Δn with i(ej ) = ej for 0 6 j 6 n − q.

• The q-dimensional back or rear face of of the singular simplex a is the q-simplex
a
R(a) = Rq (a) = a ◦ r : Δq ,→ Δn −→ X

where r : Δq ,→ Δn is the inclusion with r(e0 ) = en−q , . . . , r(eq ) = en , i.e., r(ei ) = en−(q−i) .

Definition 2.4.2
Let 0 6 q 6 n. Let R be an associative ring with unit. We define

∩ : S q (X, A; R)⊗Sn (X, A; R) = Hom(Sq (X, A), R)⊗Sn (X, A)⊗R −→ Sn−q (X)⊗R = Sn−q (X; R)

as
α ∩ (a ⊗ r) := F n−q (a) ⊗ hα, Rq (a)i ∙ r.

Remarks 2.4.3.

84
 1. The map ∩ is well-defined for relative (co-)homology for the pair (X, A): for a = a0 ∈
Sn (X, A), i.e. a = a0 + b with im(b) ⊂ A we get

α ∩ (a ⊗ r) = α ∩ ((a0 + b) ⊗ r) = α ∩ (a0 ⊗ r) + F (b) ⊗ hα, R(b)ir.

The image of R(b) is contained in A ⊂ X, but α ∈ Hom(Sq (X, A), R), thus α : Sq (X) → R
with α|Sq (A) = 0 and hα, R(b)i = 0.

2. We can express the (n − q)-dimensional front face of a in terms of the face maps from
Definition 1.2.9 as
F n−q (a) = ∂n−q+1 ◦ . . . ◦ ∂n (a).
Similarly,
Rq (a) = ∂0 ◦ . . . ◦ ∂0 (a) = ∂0n−q a .

3. There is a more general version of the cap product. Suppose that there is a pairing of
abelian groups
G ⊗ G0 → G00 ;
then we can define

∩ : S q (X, A; G) ⊗ Sn (X, A; G0 ) → Sn−q (X; G00 ).

Proposition 2.4.4.
1. The Leibniz formula holds for the cap product: for α ∈ S q (X, A), we have

∂(α ∩ (a ⊗ r)) = (δα) ∩ (a ⊗ r) + (−1)q α ∩ (∂a ⊗ r).

2. Naturality: for a map of pairs of spaces f : (X, A) → (Y, B), we have a map
f∗ : S∗ (X, A) → S∗ (Y, B) of chain complexes and a map f ∗ : S ∗ (Y, B) → S ∗ (X, A) cochain
complexes. Given a ⊗ r ∈ Sn (X, A) ⊗ R = Sn (X, A; R) and β ∈ Hom(Sq (Y, B), R), we
have
f∗ (f ∗ (β) ∩ (a ⊗ r)) = β ∩ (f∗ (a) ⊗ r) .

For the proof, we suppress the tensor product with R. It just adds to notational complexity.

Proof.
1. For the first claim we calculate the left hand side:

∂(α ∩ a) = ∂(F n−q (a) ⊗ hα, Rq (a)i)

= ∂(F n−q (a)) ⊗ hα, Rq (a)i
n−q
X
= (−1)i ∂i (∂n−q+1 ◦ . . . ◦ ∂n (a)) ⊗ hα, ∂0n−q (a)i
i=0

This has to be compared to the two terms on the right hand side:

(δα) ∩ a = F n−q−1 (a) ⊗ hδα, Rq+1 (a)i

= F n−q−1 (a) ⊗ hα, ∂Rq+1 (a)i
q
2.4.3.2
X n−(q+1)
= (−1)i ∂n−q ◦ . . . ◦ ∂n (a) ⊗ hα, ∂i ∂0 (a)i.
i=0

85
 and, noting that ∂j a is a (n − 1)-chain
n
X
α ∩ ∂a = (−1)j α ∩ ∂j a
j=0
n
X
= (−1)j F n−1−q (∂j a) ⊗ hα, Rq (∂j a)i
j=0
n
X
= (−1)j F n−1−q (∂j a) ⊗ hα, R(∂j (a))i
j=0
n
X
2.4.3.2 (n−1)−q
= (−1)j ∂n−q ◦ . . . ◦ ∂n−1 ◦ ∂j a ⊗ hα, ∂0 ∂j ai.
j=0

In order to get the result, use the simplicial relations ∂j ∂i = ∂i−1 ∂j for 0 6 j < i 6 n.

2. For the claim about naturality, we note that f∗ R = Rf∗ and f∗ F = F f∗ and plug in the
definitions to obtain

f∗ (f ∗ (β) ∩ a) = f∗ (F (a) ⊗ hf ∗ β, R(a)i)

= f∗ (F (a) ⊗ hβ, f∗ R(a)i)
= F (f∗ (a)) ⊗ hβ, R(f∗ (a))i)
= β ∩ f∗ (a).

Remark 2.4.5.
From the Leibniz formula, we conclude the following properties of the cap product:
• A cocycle cap a cycle is a cycle.
Indeed, for a cycle c with ∂c = 0 and for a q-cocycle e with δe = 0, we find

∂(e ∩ c) = (δe) ∩ c + (−1)q e ∩ (∂c) = 0 .

• A cocycle cap a boundary is a boundary.

Indeed, for a q-cocycle e with δe = 0 and a boundary b = ∂c, we find from the Leibniz
rule
∂(e ∩ (−1)q c) = (δe) ∩ (−1)q c + e ∩ b = e ∩ b
so that the cup poduct e ∩ b is a boundary.

• A coboundary cap a cycle is a boundary.

Therefore we obtain the following result:
Proposition 2.4.6.
The cap product induces a map

∩ : H q (X, A; R) ⊗ Hn (X, A; R) −→ Hn−q (X; R)

via
[α] ∩ [a] := [F (a) ⊗ hα, R(a)i]

86
Examples 2.4.7.
1. Let R be a ring and consider 1 ∈ S 0 (X; R). This is the cochain with value 1(a) = 1 ∈ R
for all 0-simplices a : Δ0 → X. We claim that the cap product obeys 1 ∩ a = a for any
singular simplex a : Δn → X. Indeed, we have q = 0 and thus F (a) = a. For the rear
face, we have R(a)(e0 ) = a(en ). Therefore, 1 ∩ a = a ⊗ h1, a(en )i = a ⊗ 1 and we identify
the latter with the n-simplex a.

2. For a space X and a cochain α ∈ S n (X; R) and a chain a ∈ Sn (X) of same degree, we
have q = n and thus F (a)(e0 ) = a(e0 ) ∈ X and R(a) = a. We find

α ∩ a = a(e0 ) ⊗ hα, ai .

If X is path-connected, then [a(e0 )] ∈ H0 (X) ∼

= Z is a generator which we identify with
1 ∈ Z. In this sense, the cap product α ∩ a generalizes the Kronecker pairing of S n (X; G)
and Sn (X) with values in G, cf. Definition 2.1.5.4.

3. There is also a version of the cap product of the form

∩ : H q (X; R) ⊗ Hn (X, A; R) −→ Hn−q (X, A; R).

Remark 2.4.8.
We explain the notation ∩. We take a 2-torus T 2 . Its first homology is H1 (T 2 ) ∼ = Z2 and
2 2
generated by the class of a meridian b ⊂ T and of a longitude a ⊂ T . The second homology
H2 (T 2 ) ∼
= Z is generated by the class of the singular 2-simplex σ : Δ2 → T 2 that maps the
boundary ∂Δ2 to ab(a−1 b−1 ). We find F 1 (σ) = a and R1 (σ) = b.
We consider the class β ∈ H 1 (T 2 ) ∼
= Hom(H1 (T 2 ), Z) ∼
= Z ⊕ Z dual to [b] ∈ H1 (T 2 ). Then
β ∩ σ = F 1 (σ) ⊗ hβ, R1 (σ)i = a can be represented by the longitude a which is transversal to
the meridian b.

2.5 Cup product on cohomology

In the following, let R be a commutative ring with unit. We will consider homology and co-
homology with coefficients in R, but we will suppress the ring R in our notation, so Hn (X, A)
will stand for Hn (X, A; R) and similarly Sn (X, A) is Sn (X, A; R). We will use analogous ab-
breviations for cochains and cohomology. Sometimes, if we have to be explicit, we denote the
multiplication in R by μ.
We recall from Proposition 1.15.4 that an Eilenberg-Zilber map is a homotopy equivalence
of chain complexes

EZ : S∗ (X × Y ; X × B ∪ A × Y ) −→ S∗ (X, A) ⊗ S∗ (Y, B)

We use this structure in a first step to associate to a pair, consisting of a cochain on X and of
a cochain on Y a cochain on the product X × Y :

Definition 2.5.1
For a (relative) cochain α ∈ S p (X, A) on X and a (relative) cochain β ∈ S q (Y, B) on Y the
cohomology cross product α × β or external cup product is the (relative) (p + q)-cochain on
X ×Y
α × β := μ ◦ (α ⊗ β) ◦ EZ ∈ S p+q (X × Y, X × B ∪ A × Y )

87
Thus
Sn (X × Y ; X × B ∪ A × Y )
O O
O O
EZ O O
L  O Oα×β
p0 +q 0 =n Sp0 (X, A) ⊗ Sq 0 (Y, B) O O
O O
O O
O O
 O'
Sp (X, A) ⊗ Sq (Y, B) α⊗β
/R⊗R
μ
/R

Remarks 2.5.2.
1. Since the Eilenberg-Zilber map is natural, the cohomology cross product is natural, i.e.
for maps of pairs of spaces f : (X, A) → (X 0 , A0 ), g : (Y, B) → (Y 0 , B 0 ) we have

(f, g)∗ (α × β) = (f ∗ α) × (g ∗ β).

2. For the Kronecker pairing and for cohomology classes α ∈ H p (X, A) and β ∈ H q (Y, B)
and homology classes a ∈ Hp (X, A) and b ∈ Hq (Y, B), we have by definition of α × β

hα × β, a × bi = hα, aihβ, bi.

3. For 1 ∈ R and thus 1X ∈ S 0 (X, A) and 1Y ∈ S 0 (Y, B)

1X × β = p∗2 (β), α × 1Y = p∗1 (α)

where pi (i = 1, 2) denotes the projection onto the ith factor in X × Y . Indeed,

h1 × β, a × bi = h1, ai ∙ hβ, bi = hβ, bi

and
p∗2 β(a, b) = β ◦ p2 (a, b) = β(b) .

We next use the cohomology cross product in order to obtain a multiplication on the graded
abelian group H ∗ (X, G). Consider the diagonal map
Δ: X → X × X
x 7→ (x, x)
as a map of pairs
Δ: (X, A ∪ B) → (X × X, X × B ∪ A × X) .

Definition 2.5.3
For α ∈ H p (X, A) and β ∈ H q (X, B) we define the cup product of α and β as

α ∪ β = Δ∗ (α × β).
× /
H p (X, A) ⊗ H q (X, B) H p+q (X × X, X × B ∪ A × X)
W W
W W
W W Δ∗
∪
W W
W W+ 
H p+q (X, A ∪ B)

Conversely, we can express the cohomology cross product via the cup product:

88
Proposition 2.5.4.
Let X and Y be topological spaces. Consider the projections

p1 : X × Y → X and p2 : X × Y → Y .

Let α ∈ H p (X) and β ∈ H q (Y ) be cohomology classes. Then the external cup product satisfies

α × β = p∗1 (α) ∪ p∗2 (β) ,

where the cup product is on the product space X × Y

Proof.
Since by Remark 2.5.2.3, we have p∗1 (α) = α × 1Y and p∗2 (β) = 1X × β, we find

p∗1 (α) ∪ p∗2 (β) = (α × 1Y ) ∪ (1X × β).

Here, α × 1 and 1 × β live in the cohomology of X × Y and the cup product is to be taken
on the product space X × Y . By Definition 2.5.3, the cup product is the pull-back of the cross
product by the diagonal. Here, Δ X×Y : X × X → X × Y × X × Y . Therefore, the above is equal
to
Δ∗X×Y ((α × 1Y ) × (1X × β)) = α × β.


In the definition of the cup product, the map

Δ EZ
S∗ (X) →∗ S∗ (X × X) → S∗ (X) ⊗ S∗ (X)

enters. The Eilenberg-Zilber map was unique up to homotopy. We will get a simple explicit
formula of the cup product by choosing a simple morphism of complexes that is still homotopy
equivalent.

Definition 2.5.5
A diagonal approximation is a natural chain map D : S∗ (X) −→ S∗ (X) ⊗ S∗ (X) such that
D(x) = x ⊗ x for all 0-chains x ∈ S0 (X).
With the method of acyclic models, cf. Section 1.15, one can prove:

Proposition 2.5.6.
Any two natural diagonal approximations are chain homotopic.

Definition 2.5.7
The Alexander-Whitney map is the diagonal approximation
X
AW(a) = F p (a) ⊗ Rq (a) ∈ S∗ (X) ⊗ S∗ (X)
p+q=n

for a ∈ Sn (X).

Remarks 2.5.8.

89
 1. It is obvious that AW is a natural chain map and this map yields a cup product for
α ∈ H p (X, A) and β ∈ H q (X, B)
P 0 0
(α ∪ β)(a) = μ ◦ (α ⊗ β)AW(a) = μ ◦ (α ⊗ β) p0 +q0 =n (F p (a) ⊗ Rq (a))
= (−1)pq α(F p (a))β(Rq (a)) .

2. From the formula, we see that ∪ is associative and distributive on cochain level and not
just on the level of cohomology groups. Also a graded Leibniz rule immediately follows:
for α ∈ H p (X) and β ∈ H ∗ (X), we have

δ(α ∪ β) = (δα) ∪ β + (−1)p α ∪ δβ .

It implies that the cup product is well-defined in cohomology.

3. But note that it does not give a (graded) commutative product on singular cochains.
(The cup product is homotopy commutative and in fact it is homotopy commutative up
to coherent homotopies: it is an E∞ -algebra.)
Proposition 2.5.9.
Let X be a topological space and α, β, γ be cohomology classes on X. The cup product satisfies
1. Associativity:
α ∪ (β ∪ γ) = (α ∪ β) ∪ γ.

2. (Graded) commutativity:
α ∪ β = (−1)|α||β| β ∪ α.

3. Compatibility with the connecting homomorphism ∂ : H ∗ (A) → H ∗+1 (X, A) in relative

cohomology and ι : A → X, we find for α ∈ H ∗ (A) and β ∈ H ∗ (X):

∂(α ∪ i∗ β) = (∂α) ∪ β ∈ H ∗ (X, A).

4. Naturality: For f : X → Y and α, β ∈ H ∗ (Y ):

f ∗ (α ∪ β) = f ∗ α ∪ f ∗ β.

Proof.
Associativity and distributivity have already been discussed. Naturality follows from the
naturality of the external cup product. Graded commutativity follows from an explicit chain
homotopy that is constructed in [Hatcher, Theorem 3.14]. 

Using the relation

α × β = p∗1 α ∪ p∗2 β
from Proposition 2.5.4 that expreses the external cup product in terms of the cup product on
the product space X × Y , we conclude:
Corollary 2.5.10.
1. The cohomology cross product is associative

α × (β × γ) = (α × β) × γ

on the level of cohomology groups.

90
 2. It satisfies a graded version of commutativity. The twist map τ : X × Y → Y × X yields
on cohomology
α × β = (−1)|α||β| τ ∗ (β × α).

3. The Leibniz formula holds,

δ(α × β) = (δα) × β + (−1)|α| α × (δβ).
Here, |α| denotes the degree of α.
Proposition 2.5.11.
1. For all pairs of spaces (X, A) the cohomology groups H ∗ (X, A; R) have a structure of a
graded commutative ring with unit 1 ∈ H 0 (X, A; R).
2. The graded ring H ∗ (X, A; R) acts on the graded group H∗ (X, A; R) via the cap product
H ∗ (X, A; R) ⊗ H∗ (X, A; R) 3 α ⊗ a 7→ α ∩ a,
i.e. 1 ∩ a = a, (α ∪ β) ∩ a = α ∩ (β ∩ a). Thus H∗ (X, A; R) is a graded module over the
graded ring H ∗ (X, A; R).
Examples 2.5.12.
Many cup products are trivial for degree reasons.
1. Let Sn be a sphere of dimension n > 1. We know from Example 2.2.6.2 that H 0 (Sn ) ∼=
∼
Z = H (S ) and the cohomology is trivial in all other degrees. We have 1 ∈ H (S ) and
n n 0 n

νn ∈ H n (Sn ). We know that

1 ∪ ν n = νn = ν n ∪ 1 and 1∪1=1
but νn ∪νn ∈ H 2n (Sn ) = 0 and thus vanishes. Thus, H ∗ (Sn ) has the structure of a so-called
graded exterior algebra with one generator νn in degree n, H ∗ (Sn ) ∼ = ΛZ (νn ).
2. More generally, if X is a CW complex of finite dimension, then α ∪ β = 0 for all α, β for
|α| + |β| big enough.
3. In particular, H ∗ (X) often has nilpotent elements: if
αr := |α ∪ .{z
. . ∪ α} = 0 ,
r

then commutativity implies (α ∪ β)r = ±αr ∪ β r = 0 for any β ∈ H ∗ (X).

4. Assume that α ∈ H p (X; R) with p odd. Then
2
α2 = (−1)p α2 = −α2 ,
where we first used the graded commutativity 2.5.9.2. Therefore 2α2 = 0 and if R is a
field of characteristic not equal to 2 or if R is a torsionfree commutative ring, then α2 = 0.
5. Consider X = X1 ∨ X2 and assume that X1 , X2 are well-pointed. Then by Proposition
1.9.7
H ∗ (X) ∼
= H ∗ (X1 ) × H ∗ (X2 )
as rings. For α = α1 + α2 and β = β1 + β2 with αi , βi ∈ H ∗ (Xi ) in positive degrees, the
cup product can be shown to be
α ∪ β = (α1 + α2 ) ∪ (β1 + β2 ) = α1 ∪ β1 + α2 ∪ β2 .

91
 6. If X can be covered like X = X1 ∪ . . . ∪ Xr with H ∗ (Xi ) = 0 for ∗ > 1 and Xi path-
connected, then in H ∗ (X) all r-fold cup products of elements of positive degree vanish.
We prove the case r = 2; the general claim then follows by induction. So assume X =
X1 ∪ X2 such that the Xi have vanishing cohomology groups in all positive degrees and let
ij : Xj ,→ X be the inclusion of Xj into X (j = 1, 2). Then for all α ∈ H ∗ (X), i∗j (α) = 0.
Consider the exact sequence
i∗j
H ∗ (X, Xj ) −→ H ∗ (X) −→ H ∗ (Xj ) .

Therefore, for any α ∈ H ∗ (X), there exists α0 ∈ H ∗ (X, X1 ) that is mapped isomorphically
to α. Similarly, for β ∈ H ∗ (X) there is an β 0 ∈ H ∗ (X, X2 ) that corresponds to β. The
cup product α ∪ β then corresponds to α0 ∪ β 0 but this is an element of H ∗ (X, X1 ∪ X2 ) =
H ∗ (X, X) = 0.

7. Consider a product of spheres, X = Sn × Sm with n, m > 1. The Künneth formula and

the universal coefficient theorem imply that as an abelian group

H ∗ (Sn × Sm ) ∼
= H ∗ (Sn ) ⊗ H ∗ (Sm ).

We have four additive generators

1 × 1, αn := νn × 1, βm := 1 × νm and γn+m := νn × νm .

The square αn2 is trivial for degree reasons:

αn2 = (νn × 1) ∪ (νn × 1) = (νn ∪ νn ) × (1 ∪ 1) = 0.

2 2
Similarly, βm = 0 = γn+m . But the products

αn ∪ βm = νn × νm = γn+m and βm ∪ αn = (−1)mn γn+m

are non-trivial. This determines the ring structure of H ∗ (Sn × Sm ).

8. Additively, as a graded abelian group, this is isomorphic to the cohomology ring H ∗ (Sn ∨
Sm ∨ Sn+m ), which has generators α̃n , β̃m and γ̃n+1 in degrees n, m and n + m. However,
by 5.
α̃n ∪ β̃m = (α̃n + 0) ∪ (0 + β̃m ) = 0 + 0 = 0
so that the cohomology ring H ∗ (Sn × Sm ) is not isomorphic to the cohomology ring
H ∗ (Sn ∨ Sm ∨ Sn+m ) as a ring. Thus the graded cohomology ring is a finer invariant than
the cohomology groups.
Note that the cohomology rings of the suspensions Σ(Sn × Sm ) and Σ(Sn ∨ Sm ∨ Sn+m ) are
isomorphic (cf. exercise). But here, we actually have

Σ(Sn × Sm ) ' Σ(Sn ∨ Sm ∨ Sn+m ).

2.6 Orientability of manifolds

We now consider topological spaces with more properties.

Definition 2.6.1
A topological space X is called locally euclidean of dimension m, if every point x ∈ X has an
open neighborhood U which is homeomorphic to an open subset V ⊂ Rm .

92
 • A homeomorphism ϕ : M ⊃ U → V ⊂ Rm is called a chart.
• A set of charts is called atlas, if the corresponding U ⊂ X cover X.

Example 2.6.2.
Consider the line with two origins, i.e.
X = {(x, 1)|x ∈ R} ∪ {(x, −1)|x ∈ R}/ ∼, (x, 1) ∼ (x, −1) for x 6= 0.
Then X is locally euclidean, but X is not a particularly nice space. For instance, it is not
Hausdorff: one cannot separate the two origins.

Definition 2.6.3
A topological space X is an m-dimensional (topological) manifold (or m-manifold for short) if
X is a locally euclidean space of dimension m that is Hausdorff and has a countable basis for
its topology.
With this definition, topological manifolds are paracompact, i.e. every open cover has a
locally finite open refinement.
Examples 2.6.4.
1. Let U ⊂ Rm an open subset, then U is a topological manifold of dimension m.
2. The n-sphere Sn ⊂ Rn+1 is an n-manifold and Sn = (Sn \ N ) ∪ (Sn \ S) is an atlas of Sn .
3. The 2-dimensional torus T ∼= S1 × S1 is a 2-manifold and more generally, the surfaces Fg
are 2-manifolds. Charts can be easily given via the 4g-gon whose quotient Fg is.
4. The open Möbius strip [−1, 1] × (−1, 1)/ ∼ with (−1, t) ∼ (1, −t) for −1 < t < 1 is a
2-manifold.
Let M be a connected manifold of dimension m > 2. We denote the open charts by Uα ⊂ M .
Without loss of generality we can assume that the coordinate patches are homeomorphic to
open balls in Rm :
∼
ϕ : Uα → D̊m ⊂ Rm .
∼
For any x ∈ M , we can find a chart ϕ : Ux → D̊m with ϕ(x) = 0. Excision for (M \ Ux ) ⊂
(M \ {x}) ⊂ M tells us that for all x ∈ M
Hm (M, M \ {x}) ∼
= Hm (Ux , Ux \ {x}) ∼
= Hm (D̊m , D̊m \ {0}) ∼
= Hm−1 (D̊m \ {0}) ∼
=Z
for m > 2. Here the chart ϕ was used for the second isomorphism. Since we have fixed in
Definition 1.8.4 a generator in Hm−1 (D̊m \ {0}) = Hm−1 (Sm−1 ), any chart provides us with a
generator in Hm (M, M \ {x}).
For a triple B ⊂ A ⊂ M , there are maps of pairs
%B,A : (M, M \ A) −→ (M, M \ B) .
In particular, for x ∈ U ⊂ M , we get a map of pairs
%x,U : (M, M \ U ) → (M, M \ {x}) .

Definition 2.6.5
Let M be an m-manifold.

93
 1. A choice of generators ox ∈ Hm (M, M \ {x}) for all x ∈ M is called coherent, if for all
x ∈ M there is an open neighbourhood U of x and a class oU ∈ Hm (M, M \ U ) such that
for all y ∈ U we have that (%y,U )∗ oU = oy .

2. An m-manifold M is called orientable (with respect to homology with values in Z), if

there exists a coherent choice of generators ox ∈ Hm (M, M \ {x}).

3. If such a choice is possible, then the family (ox |x ∈ M ) is called an orientation of M .

In the sequel, we will write %x,U also for the map on homology, i.e. we drop the lower star
in (%x2 ,U )∗ .

Remarks 2.6.6.
1. Assume that U is a small ball in M so that (%x,U )∗ : Hm (M, M \ U ) → Hm (M, M \ {x})
is an isomorphism for each x ∈ U . For a coherent choice of generators, we have for all
x1 , x2 ∈ U the compatibility condition

ox2 = %x2 ,U ◦ (%x1 ,U )−1 (ox1 ).

Hm (M, M \ U )
iii UUUU
%x1 ,U UUU%Ux2 ,U
iiiiiii UUUU
ii UUUU
it iii *
ox1 ∈ Hm (M, M \ {x1 }) Hm (M, M \ {x2 }) 3 ox2

2. Given an orientation (ox |x ∈ M ), the family (−ox |x ∈ M ) is an orientation of M as well.

Example 2.6.7.
1. If M is the open Möbius strip and you pick a generator ox ∈ H2 (M, M \ {x}) and you
walk once around the Möbius strip, you end up at −ox .

2. If we choose other coefficients, these problems can disappear. For instance for G = Z/2Z
there is no problem to choose coherent generators for H2 (M, M \ {x}; Z/2Z) ∼ = Z/2Z, so
the Möbius strip is Z/2Z-orientable. In general, Z-orientability implies Zp orientability.
The converse holds for p an odd prime.
Orientability can also be considered for more general homology theories.

3. Now, we consider integral coefficients again. Suppose that the family ( ox |x ∈ M ) is an

orientation of M . In this case, we want to obtain a global class oM ∈ Hm (M ; Z) = Hm (M ),
an orientation class, that determines the orientation in the sense that

%x,M =: %x : Hm (M ) → Hm (M, M \ {x}), %x (oM ) = ox .

For example, for RP 2 , we have H2 (RP 2 ) = 0 by Example 1.9.6 and cannot have such a
class.

For questions of orientability, compact subsets play a particularly important role. We will
derive a global characterization of orientability for M compact in Theorem 2.6.11.

Definition 2.6.8
Let K ⊂ M be a compact subset of M . We call a class oK ∈ Hm (M, M \ K) an orientation of

94
M along K, if the collection of classes ox := (%x,K )∗ (oK ) for all x ∈ K constitutes a coherent
choice of generators for all x ∈ K.
Clearly, if we have a global class oM ∈ Hm (M ), then we get coherent generators ox for all
x ∈ M and also a class oK = (%K,M )∗ (oM ) as above for all compact K ⊂ M .
Lemma 2.6.9.
Let M be a connected topological manifold of dimension m and assume that M is orientable.
Let K ⊂ M be compact.
1. Then Hq (M, M \ K) = 0 for all q > m

2. Let a ∈ Hm (M, M \ K). Then a is trivial, if and only if (%x,K )∗ (a) = 0 for all x ∈ K.
In particular, if M is compact, then Hq (M, M \ M ) = Hq (M ) = 0 for q > m.
The following method of proof is a standard method in the theory of manifolds.

Proof.
1. We first show that, if the two claims hold for compact subsets A, B ⊂ M and for A ∩ B,
then they hold for the union A ∪ B.
Consider the following part of a relative Mayer-Vietoris sequence, cf. Theorem 1.8.6:
Φ
0 = Hn+1 (M, M \ (A ∩ B)) → Hn (M, M \ (A ∪ B)) →
Ψ
Hn (M, M \ A) ⊕ Hn (M, M \ B) → Hn (M, M \ A ∩ B)

For n > m, the leftmost zero comes from our assumption 1. on A∩B. All terms Hi (M \(A∪
B)) with i > m appear between terms equalling zero, hence are zero. This shows the first
claim for A ∪ B. If a class α ∈ Hm (M \ A ∪ B) has zero image in all Hm (M, M \ {x}), then
its images under Φ, as restrictions, have the same property, hence are zero by assumption.

2. First, consider the special case when M = Rm and K is convex and compact in M . In
this case we can assume without loss of generality that K ⊂ D̊m . We calculate

Hq (M, M \ K) = Hq (Rm , Rm \ K) ∼
= Hq (D̊m , D̊m \ x) = 0, for q > m.

All identifications are isomorphisms also for q = m and this gives the second claim as
well.

3. Using the statement in 1. and induction shows that claim for the case when M = Rm and
K = K1 ∪ . . . ∪ Kr is a union with Ki convex and compact, as in 2.

4. Let M = Rm and let K be an arbitrary compact subset and let a ∈ Hq (M, M \ K) with
q > m. Choose a chain ψ ∈ Sq (Rm ) representing the class a. The boundary of ψ, ∂(ψ),
has to be of the form
X̀
∂(ψ) = λ j τj
j=1

with finitely many (q − 1)-simplices τj : Δq−1 → Rm \ K with values in Rm \ K. As the

standard simplex Δq−1 is compact, the union of the images
`
[
τj (Δq−1 ) ⊂ Rm \ K
j=1

95
 is compact as well.
Hence, there exists an open neighborhood U of the compact subset K in Rm that does
not meet the simplices:
`
[
τj (Δq−1 ) ∩ U = ∅.
j=1

Therefore, the specific q-chain ψ on Rm also defines a cycle in S∗ (Rm , Rm \ U ); let a0 ∈

Hq (Rm , Rm \ U ) be the corresponding class. Since the classes a and a0 are defined by the
same cycle, we have
(%K,U )∗ (a0 ) = a.

Srclosed balls B1 , . . . , Br ⊂ R
m
To get compact convex subsets as in 3., choose finitely many
with Bi ⊂ U for all i and K ∩ Bi 6= ∅ such that K ⊂ i=1 Bi . Consider the chain of
restriction maps
%S Bi ,U Sr %K,S Bi
(R , R \ U )
m m / (Rm , Rm \ Bi ) / (Rm , Rm \ K).
i=1

Define a00 as a00 := (%S Bi ,U )∗ (a0 ). Note that (%K,S Bi )∗ (a00 ) = a.

The closed balls Bi are convex and compact subsets of Rm and therefore by 3.

(%S Bi ,U )∗ (a0 ) = 0 = a00 , for all q > m

and hence also a = 0. This shows the first claim for all compact subsets of Rm .
For the second claim, let q = m and assume that (%x,K )∗ (a) = 0 for all x ∈ K. We have
to show that a is trivial. We express (%x,K )∗ (a) as above as

(%x,K )∗ (a) = (%x,K )∗ ◦ (%K,S Bi )∗ (a00 ) = (%x,S Bi )∗ (a00 ) = 0

for all x ∈ K. For every x ∈ Bj ∩ K the above composition is equal to (%x,Bj )∗ ◦

(%Bj ,S Bi )∗ (a00 ), but (%x,Bj )∗ is an isomorphism and hence (%Bj ,S Bi )∗ (a00 ) = 0. This im-
pliesS(%y,Bj )∗ ◦ (%Bj ,S Bi )∗ (a00 ) = 0 for all y ∈ Bj and in addition (%y,S Bi )∗ (a00 ) = 0 for all
y ∈ Bi . According to case 3., this implies that a00 = 0 and therefore a = (%K,S Bi )∗ (a00 )
is trivial as well.
5. Now let M be an arbitrary manifold. Suppose that the compact subset K is contained in a
domain Uα of a chart, i.e. K ⊂ Uα ∼
= Rm . Therefore, by excision for M \Uα ⊂ M \K ⊂ M
Hq (M, M \ K) ∼
= Hq (Uα , Uα \ K) ∼
= Hq (Rm , Rm \ im(K)).
As the image of K is compact in Rm , the claim follows from 4.
6. If both the manifold M and the compact subset K are arbitrary, then write K = Kα1 ∪
. . . ∪ Kαr and each Kαi contained in the domain Uαi of a chart. Then 5. and 1. imply the
claim.


Proposition 2.6.10.
Assume that M is oriented with (ox ∈ Hm (M, M \ {x} | x ∈ M ). Let K ⊂ M be any compact
subset. Then there is a unique orientation of M along K, which is compatible with the orien-
tation of M , i.e. there is a unique class oK ∈ Hm (M, M \ K) such that (%x,K )∗ (oK ) = ox for all
x ∈ K.

96
Proof.
• First we show uniqueness. Let oK and õK be two orientations of M along K. By assump-
tion we have that

(%x,K )∗ (oK ) − (%x,K )∗ (õK ) = (%x,K )∗ (oK − õK ) ;

on the other hand, this equals ox − ox = 0. According to Lemma 2.6.9.2 this is only the
case if oK − õK = 0.

• In order to prove the existence of oK we first consider the case where K ⊂ Uα ∼

= D̊m
and hence M \ Uα ⊂ M \ K. Let x ∈ K. We denote the isomorphism Hm (M, M \ Uα ) ∼ =
Hm (M, M \ {x}) by φx .
We define oK as
oK := (%K,Uα )∗ ((φ−1
x )(ox )).

• For K = K1 ∪ K2 , with Ki contained in the codomain of a chart, the previous argument

ensures the existence of classes oK1 ∈ Hm (M, M \ K1 ) and oK2 ∈ Hm (M, M \ K2 ). Let
K0 = K1 ∩ K2 and consider the Mayer-Vietoris sequence
i κ
0 → Hm (M, M \ K) −→ Hm (M, M \ K1 ) ⊕ Hm (M, M \ K2 ) −→ Hm (M, M \ K0 ) → . . .

The uniqueness of the orientation along the intersection K0 implies that

κ(oK1 , oK2 ) = (%K0 ,K1 )∗ (oK1 ) − (%K0 ,K2 )∗ (oK2 ) = 0.

By exactness, there is a unique class oK ∈ Hm (M, M \ K) with i(oK ) = (oK1 , oK2 ).

• For the general case we consider a compact subset K and we know that K = K1 ∪ . . . ∪ Kr
with Ki ⊂ Uαi . An induction then finishes the proof.

Theorem 2.6.11.
Let M be a connected and compact manifold of dimension m. The following statements are
equivalent:

1. M is orientable,

2. There is an orientation class oM ∈ Hm (M ; Z),

3. Hm (M ; Z) ∼
= Z.

Proof.
• Proposition 2.6.10 yields that (1) implies (2).

• Now assume that (2) holds, thus there is a class oM ∈ Hm (M ) restricting to the lo-
cal orientation classes ox . Then the class oM cannot be trivial, because its restriction
(%x,M )∗ oM = ox is a generator and hence non-trivial. Furthermore, oM cannot be of finite
order: the relation koM = 0 implies kox = 0 for all x ∈ M , contradicting the property
that ox generates the free abelian group Hm (M, M \ {x}).

97
 Let a ∈ Hm (M ) be an arbitrary element. Thus (%x,M )∗ (a) = kox for some integer k. As
the ox are coherent in x, this k has to be constant on the connected manifold M . We let
b := koM − a and find (%x,M )∗ b = 0 for all x. Since M is compact, Lemma 2.6.9 implies
that b = 0. Therefore a = koM , thus every element in Hm (M ) is a multiple of oM and
Hm (M ) ∼
= Z.
• Assuming (3), there are two possible generators in Hm (M ). Choose one of them and call
it oM . Then ((%x,M )∗ oM |x ∈ M ) is an orientation of M .


Definition 2.6.12
Let M be a compact, connected, orientable manifold. Given an orientation on M , the class oM
as in Theorem 2.6.11 is also called fundamental class of M and is often denoted by [M ] = oM .

Example 2.6.13.
For the m-sphere, M = Sm we can choose μm ∈ Hm (Sm ) as in Definition 1.8.4 as a generator.
Thus
[Sm ] = oSm = μm .
In particular, spheres are orientable. It follows from Theorem 2.6.11 and Example 1.12.8 that
RP 2n is orientable, iff n is odd.
Remarks 2.6.14.
All results about orientations can be transferred to a setting with coefficients in a commutative
ring R with unit 1R .

1. Then M is called R-orientable if and only if there is a coherent choice of generators of

the group Hm (M, M \ x; R) ∼
= R for all x ∈ M .
2. Suppose M is a compact manifold. If M is not R-orientable, the map Hn (M ; R) →
Hn (M, M \ {x}; R) ∼
= R is injective for all x ∈ M with image {r ∈ R | 2r = 0}, cf.
Hatcher, Theorem 3.26 p. 236. In particular, for R = Z, M is not orientable, if and only
if Hn (M ; Z) = 0.
3. The results we obtained have formulations relative R: Lemma 2.6.9 goes through, and if M
has an R-orientation (oR x |x ∈ M ), then for all compact K ⊂ M there is an R-orientation
of M along K, i.e. a class oR K ∈ Hm (M, M \ K; R) that restricts to the local classes. The
R-version of Theorem 2.6.11 yields for a compact manifold M a class oR M ∈ Hm (M ; R)
R R
restricting to the ox . The class oM is then called the fundamental class of M with respect
to R and is denoted by [M ; R].

Returning to integral coefficients, we know from Theorem 2.6.11 that for compact connected
orientable manifolds of the same dimension we get a copy of the integers in the homology of
the highest degree, with the fundamental class as a generator. This motivates the following
definition:

Definition 2.6.15
Let M and N be two oriented compact connected manifolds of the same dimension m > 1 and
let f : M → N be continuous. Then the degree of f is the integer deg(f ) that is given by

Hm (f )[M ] = deg(f )[N ].

98
 Of course, this definition extends the notion of the degree of a map we introduced in Defi-
nition 1.10.1 for self-maps of spheres.
Proposition 2.6.16.
Let M, N1 , N2 be oriented compact connected manifolds and let f : M → N1 and g : N1 → N2
be continuous maps.
1. The degree is multiplicative,

deg(g ◦ f ) = deg(g)deg(f ).

ˉ is the same manifold as M but with opposite orientation, then

2. If M
ˉ →N
deg(f ) = deg(f : M ˉ → N1 ) = −deg(f : M → N
ˉ1 ) = −deg(f : M ˉ1 ).

3. If the degree of f is not trivial, then f is surjective.

Proof.
The first claim follows directly from the definition of the degree. For the second claim, note
ˉ ] = −[M ], because we need to have
that [M
ˉ ] = −ox
(%x,M )∗ [M

if (ox |x ∈ M ) is the given orientation of M .

For (3) assume that f is not surjective, thus there is a point y ∈ N , that is not contained
in the image of M under f . Consider the composition
Hm (f ) (%y,N )∗
Hm (M ) / Hm (N ) / Hm (N, N \ y).

This composition is trivial, since y 6∈ im(f ). On the other hand (%y,N )∗ is an isomorphism.
Hence Hm (f ) = 0 and f has trivial degree. 

2.7 Cohomology with compact support

So far, orientation theory works fine if we restrict our attention to compact manifolds. We are
aiming at Poincaré duality: if M is a compact connected oriented manifold of dimension m,
then taking the cap product with the orientation class [M ] = oM gives a map

(−) ∩ oM : H q (M ) → Hm−q (M ).

Our aim is to show that this gives an isomorphism, but we also want to extend the result to
non-compact manifolds M . To this end we start with the following:

Definition 2.7.1
Let X be an arbitrary topological space and let R be a commutative ring with unit 1R .

1. Then the singular n-cochains with compact support singular cochains with compact support
are
Scn (X; R) = {ϕ : Sn (X) → R|∃Kϕ ⊂ X compact, ϕ(σ) = 0
for all singular simplices σ : Δn → X with σ(Δn ) ∩ Kϕ = ∅.}

99
 2. The nth cohomology group with compact support of X with coefficients in R is

Hcn (X; R) := H n (Sc∗ (X; R)) .

Remarks 2.7.2.
1. The condition of compact support is formulated in a weak sense. One could have imagined
to restrict to cochains ϕ that are non-zero only on simplices contained in a given compact
subset Kϕ depending on ϕ. Then, however, a differential cannot be defined: if ϕ is a 0-
cochain on R assigning non-zero value only to the 0-simplex contained in x = 0, then its
differential assigns non-zero values to arbitrarily large 1-simplices, i.e. all those starting
or ending in x = 0.

2. Note that Sc∗ (X; R) ⊂ S ∗ (X; R) is a sub-complex. This inclusion of complexes induces a
map on cohomology
Hcn (X; R) −→ H n (X; R).

If X is compact, then obviously Hcn (X; R) ∼

= H n (X; R) for all n.
A map from singular cohomology to singular cohomology with compact support is much
more subtle; indeed, we only get in Proposition 2.7.7 a map from a collection of relative
singular cohomologies, involving all the compact subsets of a space.

Observation 2.7.3.
1. Let K ⊂ X be compact. The map of pairs

%K,X : (X, X \ X) = (X, ∅) −→ (X, X \ K)

induces a map of cochain complexes

%nK,X : S n (X, X \ K; R) −→ S n (X; R) .

We claim that the image of %nK,X is contained in Scn (X; R). Indeed, for an n-cochain
ϕ ∈ S n (X; R) in the image, there exists ψ ∈ S n (X, X \ K; R) with %nK,X (ψ) = ϕ. The
functional ψ is trivial on all simplices σ : Δn → X with σ(Δn ) ∩ K = ∅. Therefore, for
such a simplex σ
ϕ(σ) = %nK,X (ψ)(σ) = 0 .

2. For compact subsets K ⊂ L ⊂ X we have maps of pairs

%L,X %K,L
(X, X \ X) → (X, X \ L) → (X, X \ K)

such that %K,L ◦ %L,X = %K,X .

We summarize:

Lemma 2.7.4.
Let X be a topological space.

1. For any compact subset K ⊂ X, the map %∗K,X gives a cochain map S ∗ (X, X \ K; R) −→
Sc∗ (X; R). In particular we get an induced map

H ∗ (%K,X ) : H ∗ (X, X \ K; R) −→ Hc∗ (X; R).

100
 2. For compact subsets K ⊂ L ⊂ X we have

%K,L ◦ %L,X = %K,X

and therefore the diagram of cochain complexes:

S ∗ (X, X \ K; R)
QQQ ∗
QQQ%K,X
QQQ
QQQ
(
%∗K,L ∗
Sc (X; R)
mmm6
mmm
mmmmm%∗
 m L,X

S ∗ (X, X \ L; R)

commutes.
Remarks 2.7.5.
1. Recall that a poset I is called directed, if for all i, j ∈ I there is a k ∈ I with i, j 6 k
2. The compact subsets of a space X form a directed system: if K ⊂ X and L ⊂ X are
compact, both are subsets of the compact subset K ∪ L ⊂ X.
3. Given a poset I, we can consider diagrams (of modules, of abelian groups, of chain com-
plexes) of the shape I: for each i ∈ I, there is an object Mi and for all i 6 j there is a
map fji : Mi → Mj ; with fkj ◦ fji = fki for i 6 j 6 k and fii = idMi for all i. Then we call
(Mi )i∈I a direct system. If I is directed, then we call the system (Mi )i∈I a directed system.
4. Lemma 2.7.4 says that the system K 7→ S ∗ (X, X \ K; R) is a direct system of cochain
complexes: For K ⊂ L ⊂ L0 we have

%∗K,L0 = %∗L,L0 ◦ %∗K,L .

5. We recall some facts about the direct limit of a direct system (Mi )i∈I of R-modules and
of (co)chain complexes of R-modules.
The direct limit lim Mi of a direct system (Mi ) is an R-module lim Mi , together with
−→ −→
a family of maps (hi : Mi → lim Mi )i∈I with the following universal property: for every
−→
family of R-module maps gi : Mi → M that satisfy gj ◦ fji = gi for all i 6 j, there is a
unique morphism of R-modules g : lim Mi → M such that g ◦ hi = gi for all i ∈ I.
−→
As a commuting diagram:
Mi G gi
GG
GGhi
GG
G#
∃!g 
fji lim Mi _ _ _/ M
−
;x→ B
xx
xxxh
 xx j
Mj gj

This universal property determines the R-module lim Mi up to unique isomorphism.

−→
6. For a direct system (Mi , i ∈ I) of R-modules we can construct the direct limit lim Mi as
−→
!
M
lim Mi = Mi /U
−→
i∈I

101
 L
where U is the submodule of i∈I Mi generated by all differences of the form mi −
fji (mi ), i 6 j. The map gj : Mj → lim Mi is the composition of the injection for the
−→
direct sum, followed by the canonical projection to the quotient.

7. We need an explicit construction of the direct limit of a direct system (( Ci )∗ )i∈I of

(co)chain complexes: we write L := lim(Ci ). In degree n, we set
−→
Ln := lim((Ci )n ) .
−→
All diagrams constructed from the boundary operators
di
(Ci )n / (Ci )n−1
II
II
II
II
I$
fji fji Ln−1
uu:
u
uu
uuu
  u
(Cj )n dj
/ (Cj )n−1

commute and thus, by the universal property of Ln , induce a map

(Ci )n
EE
EE
EE
EE
" 
_ _d _/ Ln−1
< Ln
fji
B
yy
yyy
y
 yy
(Cj )n

This gives a boundary map

d : Ln = (lim(Ci ))n −→ (lim(Ci ))n−1 = Ln−1 .

−→ −→
More generally, any morphism of a directed system induces a morphism between the
direct limits.

8. Let (Ai )i∈I , (Bi )i∈I and (Ci )i∈I be three direct systems of R-modules. If
φi ψi
0 → Ai −→ Bi −→ Ci → 0

is a short exact sequence for all i ∈ I and if fjiB ◦ φi = φj ◦ fjiA , fjiC ◦ ψi = ψj ◦ fjiB for all
i 6 j, then we call
(φi ) (ψi )
0 → (Ai ) −→ (Bi ) −→ (Ci ) → 0
a short exact sequence of direct systems.
The composition of the map φi : Ai → Bi with hi : Bi → lim Bi gives maps Ai → lim Bi .
−→ −→
These maps yield, by the universal property of lim Ai , a unique map
−→
φ : lim Ai −→ lim Bi .
−→ −→
Similarly, we get a map ψ : lim Bi → lim Ci .
−→ −→
Lemma 2.7.6.
102
 1. If
(φi ) (ψi )
0 → (Ai ) −→ (Bi ) −→ (Ci ) → 0
is a short exact sequence of directed systems of R-modules, then the sequence of R-
modules
φ ψ
0 → lim Ai −→ lim Bi −→ lim Ci → 0
−→ −→ −→
is short exact.

2. If (Ai )i∈I is a directed system of chain complexes, then

lim Hm (Ai ) ∼
= Hm (lim A ).
−→ −→ i

Proof.
One has to show that i) φ is injective, ii) the kernel of ψ is the image of φ and iii) ψ is surjective.
We show i) and leave ii) and iii) and the second assertion asPan exercise.
Let a ∈ lim Ai with φ(a) = 0 ∈ lim Bi . Write a = [ nj=1 λj aj ] with aj ∈ Aij . Choose
−→ −→
k > i1 , . . . , in , then a = [ak ] for some ak ∈ Ak . By assumption φ(a) = [φk (ak )] = 0. Thus there
is an N > k with fNBk φk (ak ) = 0 and by the fact that the families φk are maps of directed
systems, we have 0 = fNBk ◦ φk (ak ) = φN ◦ fNAk (ak ). But φN is a monomorphism and therefore
fNAk (ak ) = 0 ∈ lim Ai , hence a = [ak ] = [fNAk (ak )] = 0. 
−→
We can use this algebraic result to approximate singular cohomology with compact support
via relative singular cohomology groups.

Proposition 2.7.7.
For all spaces X we have isomorphisms
∼
=
lim S ∗ (X, X \ K; R) −→ Sc∗ (X; R)
−→
and hence ∼
=
lim H ∗ (X, X \ K; R) −→ Hc∗ (X; R).
−→
Here the directed system runs over the poset of compact subsets K ⊂ X.

Proof.
By the universal property of lim S ∗ (X, X \ K; R), the chain maps
−→
%∗K,X : S ∗ (X, X \ K; R) −→ Sc∗ (X; R)

from Lemma 2.7.4 combine into a single chain map

lim S ∗ (X, X \ K; R) −→ Sc∗ (X; R)

−→
A cochain ϕ ∈ S n (X; R) is an element of Scn (X; R), if and only if there is a compact
K = Kϕ such that ϕ(σ) = 0 for all σ with σ(Δn ) ∩ K = ∅ and this is the case if and only if
ϕ ∈ S n (X, X \ K; R). This shows that the map is surjective. Injectivity is direct. Then apply
Lemma 2.7.6.2. 

To the eyes of compact cohomology, Rm looks like a sphere:

103
Proposition 2.7.8.
(
R, ∗ = m,
Hc∗ (Rm ; R) ∼
= H ∗ (Rm , Rm \ {0}; R) ∼
=
0, ∗ 6= m.

Proof.
If K ⊂ Rm is compact, then there is a closed ball BrK (0) of radius rK around the origin, with
K ⊂ BrK (0). Without loss of generality we can assume that rK is a natural number. Thus we
can take the direct limit over the subsystem of such balls:

lim H ∗ (Rm , Rm \ K; R) ∼
= lim H ∗ (Rm , Rm \ Br (0); R)
−→ −→
where the direct system on the right runs over all natural numbers r ∈ N. But

H ∗ (Rm , Rm \ Br (0); R) ∼
= H ∗ (Rm , Rm \ {0}; R)

for all r and the diagrams

H ∗ (Rm , Rm \ Br (0); R) / H ∗ (Rm , Rm \ Br+1 (0); R)

 
id
H ∗ (Rm , Rm \ {0}; R) / H ∗ (Rm , Rm \ {0}; R)

commute. Therefore

lim H ∗ (Rm , Rm \ Br (0); R) ∼

= lim H ∗ (Rm , Rm \ {0}; R)
−→ −→
is an isomorphism, but the system on the right is constant and therefore

Hc∗ (Rm ; R) ∼
= lim H ∗ (Rm , Rm \ Br (0); R) ∼
= H ∗ (Rm , Rm \ {0}; R).
−→


Note that Rm is homotopy equivalent to a one-point space which is compact and for which
compactly supported cohomology and ordinary cohomology coincide, cf. Remark 2.7.3.2. Thus
cohomology with compact support is not homotopy invariant; it cannot be characterized by
axioms of Eilenberg-Steenrod type.

2.8 Poincaré duality

Observation 2.8.1.
Let R be a commutative unital ring.

• Let M be a connected m-dimensional manifold with an R-orientation (ox |x ∈ M ). For

a compact subset L ⊂ M , following Proposition 2.6.10, let oL ∈ Hm (M, M \ L) be the
orientation of M along L. For an inclusion K ⊂ L of compact subsets, we have that

(%K,L )∗ (oL ) = oK

because (%x,K )∗ (oK ) = ox = (%x,L )∗ (oL ) = (%x,K )∗ ◦ (%K,L )∗ (oL ) for all x ∈ K and, by
Lemma 2.6.9.2, the class oK is uniquely characterized by this property.

104
 • Consider for any compact subset K ⊂ M the cap-product

(−) ∩ oK : H m−p (M, M \ K; R) −→ Hp (M ; R)

α 7→ α ∩ oK = F (oK ) ⊗ hα, R(oK )i .

For an inclusion K ⊂ L of compact subsets, we have for α ∈ H m−p (M, M \ K; R) that

(%K,L )∗ (α) ∈ H m−p (M, M \ L; R) and

(%K,L )∗ (α) ∩ oL = α ∩ (%K,L )∗ (oL ) = α ∩ oK ∈ Hp (M ; R) .

because by Proposition 2.4.4.2 the cap product is natural.

By the universal property, the maps produced by the cap products combine into a map
2.7.7
lim(− ∩ oK ) : lim H m−p (M, M \ K; R) = Hcm−p (M ; R) −→ Hp (M ; R).
−→ −→

Definition 2.8.2
Let M be a connected m-manifold with R-orientation (ox |x ∈ M ). The map

lim(− ∩ oK ) : Hcm−p (M ; R) → Hp (M ; R)
−→
is called Poincaré duality map and is denoted by PD or PDM .
We can now state the main result of this section:

Theorem 2.8.3 (Poincaré Duality).

Let M be a connected m-manifold with R-orientation (ox |x ∈ M ). Then the Poincaré duality
map
PD : Hcm−p (M ; R) −→ Hp (M ; R)
is an isomorphism for all p ∈ Z.

Corollary 2.8.4 (Poincaré duality for compact manifolds).

Let M be a connected compact manifold of dimension m with an R-orientation (ox |x ∈ M )
and let [M ] = oM be the fundamental class of M , then

PD = (−) ∩ [M ] : H m−p (M ; R) −→ Hp (M ; R)

is an isomorphism for all p ∈ Z. In particular, we have for a compact connected R-oriented

m-manifold H m (M, R) = H0 (M ; R) = R.

Example 2.8.5.
Any connected compact manifold of dimension m possesses a Z/2Z-orientation and thus a
Z/2Z
fundamental class oM ∈ Hm (M ; Z/2Z) ∼
= Z/2Z and thus for all p
Z/2Z
(−) ∩ oM : H m−p (M ; Z/2Z) ∼
= Hp (M ; Z/2Z).

For instance the cohomology of RP n and its homology satisfy Poincaré duality with Z/2Z-
coefficients, regardless of the parity of n.

Proof. of Theorem 2.8.3

105
1. First we consider the case of M = Rm . We know from Proposition 2.7.8 that
(
R, p = 0,
Hcm−p (Rm ) ∼
=
0, p 6= 0

and this is isomorphic to Hp (Rm ; R). Therefore, abstractly, both graded R-modules are
isomorphic. Let Br be the closed r-ball centered at the origin. We have to understand the
map
(−) ∩ oBr : Hcm (Rm ) → H0 (Rm ; R).
We know from Example 2.4.7.2 that h1, α∩oBr i = hα, oBr i for all α ∈ H m (Rm , Rm \Br ; R).
But
h−, oBr i : H m (Rm , Rm \ Br ; R) −→ R, u 7→ hu, oBr i
is bijective because of the universal coefficient theorem 2.2.5:

H m (Rm , Rm \ Br ; R) ∼
= Hom(Hm (Rm , Rm \ Br ), R) ⊕ Ext(Hm−1 (Rm , Rm \ Br ), R)

The last summand is trivial because Hm−1 (Rm , Rm \ Br ) = 0. Thus we obtain that for all
r the map (−) ∩ oBr is the map

HomZ (ZhoBr i, R) → R with ϕ 7→ ϕ(oBr )

and thus bijective and therefore its direct limit

lim(−) ∩ oBr : lim H m (Rm , Rm \ Br ; R) −→ H0 (Rm ; R)

−→ −→
is an isomorphism as well.

2. Now assume that M = U ∪V such that the claim holds for the open subsets U, V and U ∩V
which are m-dimensional manifolds themselves, i.e. the maps PD U , PDV and PDU ∩V are
isomorphisms and each of them uses the orientation that is induced from the orientation
of M . Assume that K ⊂ U and L ⊂ V are compact and consider the relative version of
the Mayer-Vietoris sequences in cohomology

... / H p (M, M \ (K ∩ L); R) / H p (M, M \ K; R) ⊕ H p (M, M \ L; R) / H p (M, M \ (K ∪ L); R)

BC
ED

GF
@A
/ H p+1 (M, M \ (K ∩ L); R) / ...

Excision for M \ U ⊂ M \ K tells us

H p (M, M \ K; R) ∼
= H p (U, U \ K; R) .

Similarly, we find for M \ V ⊂ M \ L and M \ (U ∩ V ) ⊂ M \ (K ∩ L)

H p (M, M \ (K ∩ L); R) ∼
= H p ((U ∩ V ), (U ∩ V ) \ (K ∩ L); R)
H p (M, M \ L; R) ∼
= H p (V, V \ L; R).

106
 We obtain a map of exact sequences
∩oU ∩V
Hcm−p (U ∩ V ; R) / Hp (U ∩ V ; R)

 ∩oU ⊕∩oV

Hcm−p (U ; R) ⊕ Hcm−p (V ; R) / Hp (U ; R) ⊕ Hp (V ; R)

 ∩oM

Hcm−p (M ; R) / Hp (M ; R)

 ∩oU ∩V

Hcm−p+1 (U ∩ V ; R) / Hp−1 (U ∩ V ; R)

 ∩oU ⊕∩oV

Hcm−p+1 (U ; R) ⊕ Hcm−p+1 (V ; R) / Hp−1 (U ; R) ⊕ Hp−1 (V ; R)
The right column is exact by the Mayer-Vietoris sequence 1.8.1 in homology; the exact-
ness of the left column follows from the Mayer-Vietoris sequence in cohomology we just
considered and the isomorphisms obtained by excision by taking the limit. By assump-
tion, the top two and the two bottom horizontal arrows are isomorphisms. The five lemma
1.8.5 thus proves the case M = U ∪ V .
S
3. Now assume M = ∞ i=1 Ui with open subsets Ui that exhaust M , i.e. such that U1 ⊂
U2 ⊂ . . .. We will show that if the claim holds for all open subsets Ui with the orientation
induced by the one of M , then the claim holds for M .
To that end, let U ⊂ M be an arbitrary open subset and let K ⊂ U be compact. Excision
for (M \ U ) ⊂ (M \ K) ⊂ M gives us an isomorphism
H p (M, M \ K; R) ∼
= H p (U, U \ K; R)
and we denote by ϕK the inverse of this map. The direct limit of these ϕK over all K ⊂ U
for fixed U induces a map
ϕM
U := lim ϕ : Hcp (U ; R) −→ Hcp (M ; R).
−→ K
In general, this map is not an isomorphism (U is ‘too small to see enough of M ’), but
now we vary the open set U . For U ⊂ V ⊂ W we get
ϕW W V
U = ϕV ◦ ϕU , ϕUU = id.
As the excision isomorphism is induced by the inclusion (U, U \ K) ,→ (M, M \ K), we
get that the following diagram commutes:
ϕM
U
Hcm−p (U ; R) / Hcm−p (M ; R)
PDU PDM
 (iM 
U )∗
Hp (U ; R) / Hp (M ; R)
and hence the corresponding diagram
lim ϕM
m−p →
−
lim Hc (Ui ; R)
Ui
/ Hcm−p (M ; R)
−→
lim PD PDM

→
− Ui M
lim(iU )∗ 
lim Hp (Ui ; R) →
− i / H (M ; R)
p
−→

107
 commutes as well. The map lim ϕM is an isomorphism because every compact subset K ⊂
−→ Ui
M ends up in some open set Ui eventually. By assumption, each PD Ui is an isomorphism
and so is their limit. Similarly the limit of the (iM
Ui )∗ is an isomorphism and therefore
PDM is an isomorphism.
4. We show that the claim Sis valid for arbitrary open subsets M ⊂ Rm . We express M as
a countable union M = ∞ r=1 B̊r , where the Br are m-balls. This is possible because the
topology of Rm has a countable basis.
Each open ball B̊r in Rm is homeomorphic to Rm , thus by (1) the claim holds for these
balls B̊r .
S
Set Ui := ir=1 B̊r , then of course
U 1 ⊂ U2 ⊂ . . .
The claim then holds by (2) for the Ui as finite unions, and because of (3) it then holds
for M .
5. Finally, we assume that M is as in the theorem: a connected m-manifold with some fixed
R-orientation. Every point in M has a neighborhood which is homeomorphic to some
open subset of Rm and we can choose the homeomorphism in such a way that it preserves
the orientation. We know that M has a countable basis for its topology and thus there
are open . ⊂ M such that Vi ∼
Si subsets V1 , V2 , . .S = Wi ⊂ Rm and the Vi cover M . Define
Ui := j=1 Vj , thus M = i Ui . The claim holds for the Vj by (4) and therefore it holds
for the finite unions Ui by (2) and thus by (3) for M .


2.9 Alexander-Lefschetz duality

We will derive a relative version of Poincaré duality H m−q (M ; R) ∼
= Hq (M ; R) and some geo-
metric applications. First, we consider Čech cohomology.
Observation 2.9.1.
Let X be an arbitrary topological space and let A ⊂ B ⊂ X a pair in X. We want to associate to
the pair a cohomology group. The rough idea of Čech cohomology is to approximate H q (B, A)
by H q (V, U ) where the open neighborhoods come closer and closer to (B, A).

• We consider open neighborhoods (V, U ) of (B, A), i.e. open subsets U ⊂ V ⊂ X with
A ⊂ U and B ⊂ V .
• From the inclusion (V, U ) ⊂ (V 0 , U 0 ) we get induced maps in relative cohomology
H q (V 0 , U 0 ) −→ H q (V, U ).

• We use this property to construct for a fixed pair A ⊂ B in X a directed system, so we

set (V 0 , U 0 ) 6 (V, U ) if and only if V ⊂ V 0 and U ⊂ U 0 .

Definition 2.9.2
Čech cohomology of the pair (B, A) with A ⊂ B ⊂ X is defined as the limit

Ȟ p (B, A) = lim H p (V, U ) .

−→
108
 In this generality, Čech cohomology has very bad properties.
Remarks 2.9.3.
1. A space Y is called a euclidean neighborhood retract, if Y is homeomorphic to a subset
X ⊂ Rn for some n such that X is a retract of a neighborhood X ⊂ U ⊂ Rn .
2. If the space X is a euclidean neighborhood retract and A ⊂ B ⊂ X are locally compact,
then Ȟ p (B, A) only depends on B and A and not on X.
3. If in addition A and B are euclidean neighborhood retracts themselves, then Ȟ p (B, A) is
actually isomorphic to H p (B, A). For more background on Čech cohomology see Dold’s
book Lectures on Algebraic Topology, reprint in: Classics in Mathematics. Springer-Verlag,
Berlin, 1995, VIII §6.
Observation 2.9.4.
• Now let M be a connected m-dimensional manifold and let K ⊂ L ⊂ M be compact
subsets in M . We assume that there is an orientation class oL ∈ Hm (M, M \L) of M along
L (possibly with coefficients in R, but we suppress coefficients from the notation). We
aim at a cap-pairing of Čech-cohomology Ȟ ∗ (L, K) with relative homology H∗ (M, M \ L)
in which the class oL is.
• For (L, K) ⊂ (V, U ) we set up a map on the level of chains and cochains
 
p Sk (U ) + Sk (V \ K)
S (V, U ) ⊗ −→ Sk−p (V \ K, V \ L) (∗).
Sk (V \ L)
For this note the trivial inclusion V \ L ⊂ (U ∪ (V \ K)) = V . For α ∈ S p (V, U ) and
k (V \K)
a + b ∈ Sk (US)+S
k (V \L)
we have

α ∩ (a + b) = α ∩ a + α ∩ b = 0 + α ∩ b
and this ends up in the correct chain group.
∗ (V \K)
• The homology of S∗ (US)+S
∗ (V \L)
is isomorphic to H∗ (V, V \L) and this in turn is isomorphic
to H∗ (M, M \ L) via excision for (M \ V ) ⊂ (M \ L) ⊂ M . This allows us to rewrite the
second tensorand on the left hand side of (∗), as desired.
• Excision for (M \ V ) ⊂ (M \ L) ⊂ (M \ K) tells us as well that
H∗ (V \ K, V \ L; R) ∼
= H∗ (M \ K, M \ L; R) .
This allows us to rewrite the right hand side.
• As Čech cohomology is the direct limit lim H ∗ (V, U ) and as everything is compatible
−→
(which we did not really show), the above gives a well-defined map
PD : Ȟ q (L, K) ⊗ Hm (M, M \ L) −→ Hm−q (M \ K, M \ L), α ⊗ oL 7→ α ∩ oL .

Proposition 2.9.5 (Alexander-Lefschetz duality).

Let M be a connected m-dimensional manifold and let K ⊂ L ⊂ M with K, L compact. Let
M be oriented along L with respect to R. Then the map
PD = (−) ∩ oL : Ȟ q (L, K; R) −→ Hm−q (M \ K, M \ L; R)
is an isomorphism for all integers q.

109
 Before we prove this result, we collect some properties of this form of the Poincaré duality
map.
Remarks 2.9.6.
1. This PD map still satisfies that PD(1) = oL for K = ∅ and 1 ∈ H 0 (L; R).
2. The PD-map is natural in the following sense: for any map of pairs i : (L, K) ,→ (L0 , K 0 )
in M , we have also a map ĩ : (M \ K 0 , M \ L0 ) ,→ (M \ K, M \ L), and the diagram
(−)∩oL0
Ȟ q (L0 , K 0 ) / Hm−q (M \ K 0 , M \ L0 )
Ȟ q (i) Hm−q (ĩ)
 (−)∩oL

Ȟ q (L, K) / Hm−q (M \ K, M \ L)
commutes.
3. We will not prove the following fact (cf. Bredon Lemma VI.8.1). The diagram
... / Ȟ q (L, K) / Ȟ q (L) / Ȟ q (K) / Ȟ q+1 (L, K) / ...

∩oL ∩oL ∩oK ∩oL

   
... / Hm−q (M \ K, M \ L) / Hm−q (M, M \ L) / Hm−q (M, M \ K) / Hm−q−1 (M \ K, M \ L) / ...

commutes, and therefore (using the five lemma) it suffices to show the absolute version
of Alexander-Lefschetz duality,
∩o
Ȟ q (L) −→
L
Hm−q (M, M \ L) .

Lemma 2.9.7.
Let K and L are compact subsets of M with an orientation class oK∪L along K ∪ L and induced
orientation classes oK and oL . Then the diagram
... ∂ / Ȟ q (K∪L) / Ȟ q (K)⊕Ȟ q (L) / Ȟ q (K∩L) ∂ / ...
∩oK∪L ∩oK ⊕∩oL oK∩L
δ
   δ
... / Hm−q (M,M \(K∪L)) / Hm−q (M,M \K)⊕Hm−q (M,M \L) / Hm−q (M,M \(K∩L)) / ...
commutes and has exact rows.

Proof.
• The only critical squares are the ones that are slightly out of the focus of the above
diagram, the ones with the connecting homomorphisms. The Ȟ ∗ -sequence in the upper
line comes from taking the direct limit of
0 → Hom(S∗ (U ) + S∗ (V ), R) −→ Hom(S∗ (U ), R) ⊕ Hom(S∗ (V ), R) −→ Hom(S∗ (U ∩ V ), R) → 0

over all open subsets U, V with K ⊂ U and L ⊂ V . (Note that by Lemma 2.7.6 taking
the direct limit is exact.)
• Let α ∈ Ȟ q (K ∩ L; R). Choose a representing cocycle f with α = [f ], i.e. δf = 0 on U ∩ V
and let ∂ be the connecting homomorphism for ordinary singular cohomology. What is
∂(α)? A preimage for f in the direct sum is a pair (f, 0) and its coboundary is (δf, 0),
so if we define h ∈ Hom(S∗ U + S∗ V, R) by the property h(u + v) = δf (u) for u ∈ S∗ (U ),
v ∈ S∗ (V ), then
∂(α) = [h].
We can extend h to a cochain on M (for instance by defining it to be trivial on the chains
that are supported on the complement).

110
 • We want to compare ∂(α) ∩ oK∪L and δ(α ∩ oK∩L ). For the first term we express the
orientation class oK∪L = [a] as a sum
a = b + c + d + e ∈ S∗ (U ∩ V ) + S∗ (U \ L) + S∗ (V \ K) + S∗ (M \ (K ∪ L)).
This is possible, since the subsets U ∩ V, U \ L, V \ K and M \ (K ∪ L) are open and
therefore we can work with small chains for this open cover. With the notation as above
we get
∂(α) ∩ oK∪L = [h ∩ (b + c + d + e)] = [h ∩ c] .
As h is only non-trivial on chains in U , only the terms involving b and c can contribute.
Since δ(f ) is trivial on U ∩ V , h is only non-trivial on the complement of V in U .
• For α ∩ oK∩L we write [f ∩ a] and as the lower exact row comes from the short exact
sequence of complexes
S∗ (M ) S∗ (M ) S∗ (M ) S∗ (M )
0→ −→ ⊕ −→ →0
S∗ (M \ K ∪ L) S∗ (M \ K) S∗ (M \ L) S∗ (M \ K) + S∗ (M \ L)
we view f ∩a as an element modulo S∗ (M \K)+S∗ (M \L). The connecting homomorphism
picks (f ∩ a, 0) as a pre-image of f ∩ a, then takes its boundary (∂(f ∩ a), 0). But the
latter is up to sign by the Leibniz rule
(∂(f ∩ a), 0) = (δ(f ) ∩ a), 0) ± (f ∩ ∂a, 0).
Writing a as a = b + c + d + e as above and using that f ignores b and e we obtain that
the above is (δf ∩ c + δf ∩ d ± f ∩ ∂a, 0). But δf ∩ d and f ∩ ∂a are elements in S∗ (M \ K)
and hence all that remains when we pick a preimage is (δf ∩ c, 0), thus
δ(α ∩ oK∩L ) = [δf ∩ c] = [h ∩ c].

Now we can prove Alexander-Lefschetz duality.

Proof. of Proposition 2.9.5
Remark 2.9.6.3 implies that it suffices to prove the absolute case, i.e. to show that for any
compact subset K ⊂ M
(−) ∩ oK : Ȟ q (K) −→ Hm−q (M, M \ K)
is an isomorphism for all q.
1. If K is empty, then we get the true statement that Ȟ q (∅) = 0 = Hm−q (M, M ). For K a
point we only get something non-trivial for degree q = 0 and here 1 ∈ R = Ȟ 0 (K) is sent
to oK = ox via Poincaré duality. Similarly, if M = Rm and K is convex and compact we
can proceed as in the case of a point.
2. If K = K1 ∪ . . . ∪ Kr with Ki compact and convex and M is still Rm an induction over r
using Lemma 2.9.7 and (1) proves the claim.
S
3. For M = Rm and K arbitrary we can find a neighborhood U of K of the form U = N i=1 Ui
q
with the Ui being convex. Such U suffice to calculate the direct limit lim H (U ) for the
−→
Čech cohomology of K. For such U we have
Hm−q (Rm , Rm \ K) ∼
= lim H (Rm , Rm \ U )
−→ m−q
S
because Rm \ K = U Rm \ U . The U satisfy Alexander-Lefschetz duality by (2) and
hence K does.

111
 4. Finally let M and K be arbitrary, but satisfying the conditions of Proposition 2.9.5.
Express K = K1 ∪. . .∪Kr such that the Ki are contained in a chart that is homeomorphic
to Rm and proceed as in the case before.

2.10 Application of duality

We specialize to the case when the manifold M is Rm with the standard orientation.

Proposition 2.10.1 (Classical Alexander duality).

Let K ⊂ Rm be compact. Then

Ȟ q (K) ∼
= Hm−q (Rm , Rm \ K) ∼
= H̃m−q−1 (Rm \ K).

Proof.
Here the first isomorphism is the absolute version of Alexander-Lefschetz duality 2.9.2 for
M = Rm . The second one is a result of the long exact sequence of pairs in homology. 

Remark 2.10.2.
• This is bad news for knot complements. A knot K is the homeomorphic image of S1 in
R3 . Proposition 2.10.1 implies that

H1 (R3 \ K) ∼
= Ȟ 1 (K)

but the circle is a euclidean neighborhood retract and therefore Čech cohomology concides
with ordinary singular cohomology. Since H 1 (K) ∼ = Z, the first homology group of any
knot complement is isomorphic to the integers, thus it does not help to distinguish knots.

• The fundamental group of the knot complement does a better job. Here the un-knot gives
the integers, but for instance the complement of the trefoil knot

has a fundamental group that is not isomorphic to the integers, but is isomorphic to the
group ha, b|a2 = b3 i. This group is actually isomorphic to the braid group on three strands.
(This can be computed using the Wirtinger presentation derived from the link diagram of
a knot, see Section 4.2.3 and 4.2.4 of J. Stillwell. Classical Topology and Combinatorial
Group Theory. Springer Graduate Text in Mathematics 72, 1993.

Proposition 2.10.3.
Let M be a compact oriented connected m-manifold and let ∅ 6= K ⊂ M be compact. If the
first homology H1 (M ) of the ambient manifold is trivial, then Ȟ m−1 (K) is a free abelian group.
The complement M \ K then has rankȞ m−1 (K) + 1 connected components.

112
Proof.
Let k = |π0 (M \ K)| be the number of components of the complement of K in M . By Corollary
1.3.3,
k = rankH0 (M \ K) = 1 + rankH̃0 (M \ K).
By assumption H1 (M ) = 0 = H̃0 (M ) and therefore we know from the long exact sequence and
duality that
H̃0 (M \ K) ∼= H1 (M, M \ K) ∼ = Ȟ m−1 (K).
Since the group Ȟ m−1 (K) is isomorphic to a zeroth homology group, it is free abelian. The
statement about k is now the combination of the two equations. 

Proposition 2.10.4.
If M is a compact connected orientable m-manifold and if the first homology group of M
with integral coefficients vanishes, then all compact submanifolds of M without boundary of
dimension (m − 1) are orientable.

Compact manifolds without boundary are often called closed.

Proof.
A submanifold N ⊂ M is a euclidean neighborhood retract and therefore

H m−1 (N ) ∼
= Ȟ m−1 (N ) ∼
= H1 (M, M \ N ) ∼
= H̃0 (M \ N ) .

Thus H m−1 (N ) is free abelian. Theorem 2.6.11 implies that the components of N are orientable.


Corollary 2.10.5.
It is not possible to embed real projective space RP 2 into R3 .

Proof.
If one could, then one could embed RP 2 into S3 as the one-point compactification of R3 . Due
to H1 (S3 ) = 0, the 2-manifold RP 2 would be orientable, but we know from Example 2.6.13
that this is not true. 

At Oberwolfach Research Institute for Mathematics there is a model of the Boy surface.
This is a model of an immersion of RP 2 into three-space. http://www.mfo.de/general/boy/

Proposition 2.10.6.
Let M be a compact connected and orientable m-manifold and let

βi := dimQ Hi (M ; Q)

be the ith Betti number of M . Then βi = βm−i .

Proof.
Note that in this case Čech cohomology Ȟ ∗ (M ) = Ȟ ∗ (M, ∅) is isomorphic to H ∗ (M ) because
a limit is to be taken over the relative cohomology groups H ∗ (M, U ) for the directed system of

113
pairs (M, U ) with U any open set which has (M, ∅) as a maximal element. Duality 2.9.5 then
implies that
2.9.5
βm−i = dimQ Hm−i (M ; Q) = dimQ H i (M ; Q)
As the group Q is divisible, Remark 2.2.2.7 implies that there is no Ext-term arising in the
universal coefficient theorem 2.2.4 and thus

dimQ H i (M ; Q) = dimQ (Hom(Hi (M ), Q)) .

The right hand side is equal to the dimension of the vector space of the homomorphisms from
the free part of the homology group Hi (M ) to Q which is equal to the rank of Hi (M ). Since
tensoring with Q is exact, there is no Tor-term and thus Hi (M ; Q) = Hi (M ) ⊗Z Q; thus the
rank of Hi (M ) is equal to βi = dimQ Hi (M ; Q). 

Corollary 2.10.7.
Let M be a compact connected
Pm and orientable m-manifold of odd dimension. Then the Euler
characteristic χ(M ) = i=0 (−1)i βi vanishes.

Proof.
We compute
m
X m
X
i 2.10.6
χ(M ) = (−1) βi = (−1)i βm−i = (−1)m χ(M ) .
i=0 i=0

Proposition 2.10.8.
For M a compact connected orientable m-manifold with boundary the duality holds

Ȟ q (M, ∂M ) ∼
= Hm−q (M ) .

Proof.
Glue a collar to M , i.e., consider the auxiliary manifold

W := M ∪ (∂M × [0, 1)) =: M ∪ W 0 .

Then W is an m-manifold without boundary; thus duality 2.9.5 applies to the pair of compact
subsets ∂M ⊂ M :
Ȟ q (M, ∂M ) ∼
= Hm−q (W \ ∂M, W \ M ) .
Now note that

W \ ∂M ' M \ ∂M t W 0 \ ∂M and W \ M = W 0 \ ∂M

so that the right hand side becomes

Hm−q (M \ ∂M ) ∼
= Hm−q (M ) .

For the last isomorphism, we used that taking the complement of the boundary ∂M in M
gives a space that is homotopy equivalent to M . 

114
Corollary 2.10.9.
If M is a compact connected orientable m-manifold, then the Euler characteristic of the bound-
ary ∂M is always even.

Proof.
With W as above, the homotopy equivalence W ' M implies χ(M ) = χ(W ). The long exact
sequence of the pair W \ M ⊂ W gives
χ(W ) = χ(W \ M ) + χ(W, W \ M ) .
The homotopy equivalence W \M ' ∂M yields χ(W \M ) = χ(∂M ) and duality 2.9.5 guarantees
that χ(W, W \ M ) = (−1)m χ(M ). Therefore
χ(∂M ) = (1 + (−1)m−1 )χ(M )
and this is always an even number. 

Remark 2.10.10.
1. Recall from from Example 1.12.8 that the real projective space has the structure of a CW
complex with one cell in each dimension. Thus RP 2m has Euler characteristic 1 and by
Corollary 2.10.9 cannot be a boundary.
2. For the calculations of of the Euler characteristic of complex and quaternionic projective
spaces, recall from Example 1.12.8 that for complex projective space of dimension 2m we
have cells in dimension up to 4m, but only in even dimensions. Similarly, for quaternion
projective space of dimension 2m cells occur up to dimension 8m, but only in degrees
divisible by 4.
Thus
2m
X
2m
χ(CP )= (−1)2i = 2m + 1
i=0
and
2m
X
2m
χ(HP )= (−1)4i = 2m + 1 .
i=0
By Corollary 2.10.9, all these projective spaces do not occur as boundaries of connected
compact orientable manifolds.
3. These facts are important in bordism theory: one can introduce an equivalence relation on
manifolds by saying that two m-manifolds M and N are bordant, if there is an (m + 1)-
manifold W whose boundary is the disjoint union of M and N , ∂W = M t N . Thus the
projective spaces give non-trivial equivalence classes under the bordism relation.

2.11 Duality and cup products

Let M be a connected closed m-manifold with an R-orientation oR
M for some commutative ring
R. We consider the composition
∪
H k (M ; R) ⊗R H m−k (M ; R) / H m (M ; R)
(−)∩oR
M

H0 (M ; R) ∼
=R

115
Definition 2.11.1
Let M be a connected closed m-manifold with an R-orientation for some commutative ring R.
For α ∈ H k (M ; R), β ∈ H m−k (M ; R) the map
(α, β) 7→ hα ∪ β, oR
Mi

with values in R is called the cup product pairing of M .

Proposition 2.11.2.
If R is a field or if R = Z and all homology groups of M are torsion-free, the cup product
pairing is non-singular in the sense that the two induced maps
H k (M ; R) → HomR (H m−k (M ; R), R)

and H m−k (M ; R) → HomR (H k (M ; R), R)

α 7→ β 7→ hα ∪ β, oR
Mi ∈ R β 7→ α 7→ hα ∪ β, oR
Mi ∈ R

are both isomorphisms.

Proposition 2.11.2 holds as long as one restricts attention to the free part of the cohomology
groups: let F H k (M ; R) denote the free part of H k (M ; R) then there is a non-singular pairing
F H k (M ; R) ⊗R F H m−k (M ; R) → R.
In geometric applications the ground ring is often R = R.
Proof.
The Kronecker pairing, cf. Lemma 2.1.6, yields a map
κ : H k (M ; R) → HomR (Hk (M ; R), R)
and Poincaré duality 2.8.3 tells us that capping with oR
M is an isomorphism between Hk (M ; R)
m−k
and H (M ; R). The composite is
κ
H k (M ; R) → HomR (Hk (M ; R), R) ∼
= HomR (H m−k (M ; R), R),
α 7→ hα, (−) ∩ oR
Mi .

Over a field, κ and hence the composite is an isomorphism. We finally use the duality relation
2.4.7.2 2.5.11.2 2.4.7.2
hα ∪ β, oR R
M i = h1, (α ∪ β) ∩ oM i = h1, α ∩ (β ∩ oR R
M )i = hα, β ∩ oM i

In the torsion-free setting, we obtain an isomorphism as well. 

Definition 2.11.3
Let M be a connected closed m-manifold with an R-orientation for some commutative ring R.
Dual to the cup product pairing, we define the intersection form:
Hp (M ; R) ⊗ Hm−p (M ; R) → R
with a ⊗ b 7→ hPD−1 (a) ∪ PD−1 (b), oR
M i.

For even-dimensional manifolds, the signature of this form is a particularly important in-
variant in differential topology. For instance one can show that for a compact oriented manifold
W such that ∂W = M with a 4n-dimensional manifold M , the signature of the intersection
form on M is trivial.
For explicit computations of cohomology rings, the following Lemma is useful:

116
Lemma 2.11.4.
Let M be a connected closed m-manifold with a Z-orientation and with torsion-free homology
groups. If H p (M ) ∼
= Z∼ = H m−p (M ) and if α ∈ H p (M ), β ∈ H m−p (M ) are generators, then
α ∪ β is a generator of the group H m (M ) ∼
= Z.

Proof.
Since by Proposition 2.11.2 the cup product pairing is non-degenerate for torsion free cohomol-
ogy, there exists for any generator α ∈ H p (M ) an element β 0 ∈ H m−p (M ) with

hα ∪ β 0 , oM i = 1 .

Note that this implies that α ∪ β 0 is a generator of H m (M ), as a dual of the generator oM .

As β is a generator of H m−p (M ), we know that β 0 = kβ for some integer k and hence

1 = hα ∪ β 0 , oM i = hα ∪ kβ, oM i = khα ∪ β, oM i.

But hα ∪ β, oM i is an integer as well, so k has to be ±1 and therefore α ∪ β generates the group

H m (M ) as well. 

We will use this result to calculate the cohomology rings of projective spaces.
Lemma 2.11.5.
If α ∈ H 2 (CP m ) is an additive generator, then αq = α∪q ∈ H 2q (CP m ) is an additive generator
as well for all q 6 m.

Proof.
We have to show by induction on the complex dimension m that αq is an additive generator of
H 2q (CP m ).
• For m = 1 there is nothing to prove because CP 1 ∼
= S2 and there α2 = 0.
• Consider the inclusion i : CP m−1 ,→ CP m . The CW structure of CP m explained in Ex-
ample 2.12.9 implies CP m ∼ = CP m−1 ∪f D2m for attaching the 2m-cell. For m > 1
i∗ : H 2i (CP m ) → H 2i (CP m−1 )

is an isomorphism for 1 6 i 6 m−1. In particular, i∗ (α) additively generates H 2 (CP m−1 ).

Induction over m then shows that (i∗ (α))q generates H 2q (CP m−1 ) for all 1 6 q 6 m − 1.
But (i∗ (α))q = i∗ (αq ), by Proposition 2.5.9.4, and i∗ is an isomorphism, so αq additively
generates H 2q (CP m ) for 1 6 q 6 m − 1. Lemma 2.11.4 then shows that α ∪ αm−1 = αm
generates H 2m (CP m ).


Corollary 2.11.6.
As a graded ring, we have

H ∗ (CP m ) ∼
= Z[α]/αm+1 with |α| = 2.
Similarly,
H ∗ (RP m ; Z/2Z) ∼
= Z/2Z[α]/αm+1 with |α| = 1.
(Taking coefficients in Z/2Z leads to a complex with vanishing differentials, cf. Example 1.13.3.)

117
 There are two geometric consequences that follow from this calculation.

Proposition 2.11.7.
For 0 < m < n the inclusion j : CP m ,→ CP n is not a weak retract.

Proof.
Let us assume that there exists r : CP n → CP m with r ◦ j ' id. On the second cohomology
groups, the map j induces an isomorphism

j ∗ : H 2 (CP n ) → H 2 (CP m ) .

Let α ∈ H 2 (CP m ) be an additive generator. Because of j ∗ ◦ r∗ = id, the element β := r∗ (α) ∈

H 2 (CP n ) is an additive generator as well. As αm+1 = 0 we get

β m+1 = r∗ (α)m+1 = r∗ (αm+1 ) = r∗ (0) = 0.

But by Corollary 2.11.6 H ∗ (CP n ) ∼

= Z[β]/β n+1 and hence β m+1 6= 0. 

Proposition 2.11.8.
The attaching map of the 2n-cell in CP n is not null-homotopic.

Proof.
Let ϕ : S2n−1 → CP n−1 be the attaching map, thus

CP n = CP n−1 ∪ϕ D2n .

If ϕ were null-homotopic, then there is a homotopy H : S2n−1 × [0, 1] → CP n−1 with H1 = ϕ

and H0 constant. Since H0 is constant, H factorizes to a map H̃
H
S2n−1 × [0, 1] CP
/ n−1
o o7
o o o
 o H̃
D2n

with the vertical map being (x, t) 7→ tx. Then

H̃ t idCP n−1 : D2n t CP n−1 → CP n−1

factorizes to a retract r : CP n → CP n−1 ,

H̃tid
D2n t CP n−1 i i4 CP
/ n−1
i
i i
i i ir
 i
CP n = D2n ∪ϕ CP n−1

in contradiction to Proposition 2.11.7. 

118
Remark 2.11.9.
A famous example of this phenomenon is the Hopf fibration

h : S3 → CP 1 = S2 = C ∪ ∞ .

Consider S3 ⊂ C2 and send S3 3 (u, v) to

(
u
v
,v=6 0,
h(u, v) :=
∞, v = 0.

Up to a homeomorphism of S2 , this is the attaching map for the 4-dimensional cell of CP 2

and thus by Proposition 2.11.8 not null-homotopic. In fact, the map h generates the homotopy
group π3 (S2 ) ∼
= Z.

2.12 The Milnor sequence

The aim is to calculate the cohomology rings of infinite dimensional projective spaces and more
generally to understand cohomology groups for infinite dimensional CW complexes.
We start with some algebraic structures: let (Mi )i∈N0 be a family of R-modules together
with a sequence of maps
f1 f2 f3
M0 ←− M1 ←− M2 ←− . . .
We call such a family (Mi , fi )i∈N0 an inverse system (over the poset (N0 , 6)).

Definition 2.12.1
The inverse limit of the inverse system (Mi )i∈N0 is the R-module
Y
lim Mi = {(x0 , x1 , . . .) ∈ Mi |fi+1 (xi+1 ) = xi , i > 0}.
←−
i∈N0

Remarks 2.12.2.
1. The restrictions of the projections of the product endow the the inverse limit with a
system of maps such that the diagrams
p
lim Mi j+1 / Mj+1
←− HH
HH
HH fj+1
pj HH
H$ 
Mj

commute for all j ∈ N0 . With their use, we can characterize the inverse limit by the
following universal property:

hj+1
0 Mj+1
v :
pj+1 vvv
v
vv
vv
∃!
W _ _ _/ lim M
←− iH
fj+1
HH
HH
H
pj HH
H$ 
hj . Mj

119
 Q
2. If ξ denotes the map that sends the element (x0 , x1 , . . .) ∈ i∈N0 Mi to (x0 − f1 (x1 ), x1 −
f2 (x2 ), . . .) then we can express the inverse limit as the kernel of ξ,
Y ξ
Y
0 → lim Mi −→ Mi −→ Mi .
←−
i∈N0 i∈N0

Definition 2.12.3
Let lim 1 Mi be the R-module coker(ξ).
←−
By definition, we have an exact sequence
Y ξ
Y
0 → lim Mi −→ Mi −→ Mi −→ lim 1 Mi → 0 .
←− ←−
i∈N0 i∈N0

Lemma 2.12.4.
If
0 → (Mi , fi ) −→ (Ni , gi ) −→ (Qi , hi ) → 0
is a short exact sequence of inverse systems (cf. Remark 2.7.6.8 for exact sequences of direct
systems), then the sequence

0 → lim Mi −→ lim Ni −→ lim Qi −→ lim 1 Mi −→ lim 1 Ni −→ lim 1 Qi → 0

←− ←− ←− ←− ←− ←−
is exact.

Proof. Q Q
Consider the map ξ : i Mi → i Mi as a chain complex C∗ that is non-trivial only in two
degrees 0 and 1. Then the first homology group is the inverse limit and the zeroth homology
group is the lim-one term

H1 C∗ = ker ξ ∼
= lim M and H0 C∗ = cokerξ ∼
= lim 1
Mi .
←− i ←−
We can translate the short exact sequence of inverse systems into a short exact sequence of
chain complexes
Q Q Q
i Mi i Ni i Qi
0 / / / /0

ξ ξ ξ
Q  Q  Q 
0 /
i Mi /
i Ni /
i Qi / 0
and the associated long exact sequence (cf. Proposition 1.5.6) gives precisely our claim. 

Therefore the lim-one terms measure how non-exact inverse limits are. We present a criterion
which ensures that we have exactness.

Lemma 2.12.5 (Mittag-Leffler condition).

Let (Mi , fi ) be an inverse system. Assume that for every n > 0 there exists N = N (n) such
that for all m > N that the image of fn+1 ◦ . . . ◦ fm : Mm → Mn is equal to the image of
fn+1 ◦ . . . ◦ fN : MN → Mn . Then
lim 1 Mi = 0.
←−

120
Proof.
Without loss of generality, we can assume that the sequence N (n) is monotonously increasing
in n. We have to showQthat the cokernel of ξ is trivial. This means that we have to show that
any sequence (ai )i ∈ i Mi is in the image of ξ, if the Mittag-Leffler condition holds.

• As a first case, we deal with sequences (ai )i such that every ai is in the image of fi+1 ◦
. . . ◦ fN (i) : MN (i) → Mi .
By induction on k, we construct elements b0 , . . . , bk with

bi ∈ im(fi+1 ◦ . . . ◦ fN (i) ) ⊂ Mi

such that ai = bi − fi+1 bi+1 for all i < k. Then we have (ai ) = ξ(bi ).
We start with a0 = b0 ∈ M0 . The condition is empty for k = 0.
Assume that elements b0 , b1 , . . . , bk have been found. Because both ak and bk are in
im(fk+1 ◦ . . . ◦ fN (k) ) and because by the assumption that the image of fk+1 ◦ . . . ◦ fN (k+1)
is equal to the image of fk+1 ◦ . . . ◦ fN (k) , we can find y ∈ MN (k+1) with

ak − bk = fk+1 ◦ . . . ◦ fN (k+1) (y).

Define
bk+1 := −fk+2 ◦ . . . ◦ fN (k+1) (y).
Then
bk − fk+1 bk+1 = bk + ak − bk = ak .
Thus (ak ) ∈ im ξ.

• If for some i the element ai is not in the image fi+1 ◦ . . . ◦ fN (i) : MN (i) → Mi , then we
consider the sum

a0i := ai + fi+1 ai+1 + . . . + fi+1 ◦ . . . ◦ fN (i) (aN (i) ).

We check that

ai − (a0i − fi+1 (a0i+1 )) = ai − ai − fi+1 (ai+1 ) − . . . − fi+1 ◦ . . . ◦ fN (i) (aN (i) )

+ fi+1 (ai+1 ) + fi+1 ◦ fi+2 (ai+2 ) + . . . + fi+1 ◦ . . . ◦ fN (i+1) (aN (i+1) )
= fi+1 ◦ . . . ◦ fN (i)+1 (aN (i)+1 ) + . . . + fi+1 ◦ . . . ◦ fN (i+1) (aN (i+1) )

and therefore ai −(a0i −fi+1 (a0i+1 )) is in the image of fi+1 ◦. . .◦fN (i+1) . As in the preceeding
case, we write ai − (a0i − fi+1 (a0i+1 )) as bi − fi+1 bi+1 . Thus

ai = ci − fi+1 (ci+1 )

with ci := bi + a0i .

Examples 2.12.6.

121
 1. If every map fi is surjective, then the inverse system (Mi , fi ) satisfies the Mittag-Leffler
criterion with N (n) = n + 1. For instance, the inverse system of rings
Z/pZ ←− Z/p2 Z ←− Z/p3 Z ←− . . .
satisfies this condition. The inverse limit of this system is a ring, called the p-adic integers.
These are denoted by Ẑp and they are the p-adic completion of the ring of integers.
2. We want to apply Lemma 2.12.5 to inverse systems of cochain complexes.
Assume that X is a CWS complex and that (Xn )n is a sequence of subcomplexes with
Xn ⊂ Xn+1 and X = n Xn . For instance, we could take Xn = X n , the n-skeleton of X.
Consider for each n the cochain complex
Sn∗ (X) := S ∗ (Xn ).
The inclusion maps Xn ⊂ Xn+1 induce maps of cochain complexes
∗
fn+1 : Sn+1 (X) −→ Sn∗ (X) .
We therefore have an inverse system
f1 f2
S0∗ (X) ←− S1∗ (X) ←− . . .
which are maps of cochain complexes, i.e. commute with the coboundary maps
i fn+1
Sn+1 (X) / Sni (X)
δ δ
 
i+1 fn+1
Sn+1 (X) / Sni+1 (X).

Lemma 2.12.7.
If (Cn∗ , fn ) is an inverse system of cochain complexes, such that for every cochain degree m the
inverse system (Cnm , fn ) satisfies the Mittag-Leffler condition, then the sequence
0 → lim 1 H m−1 (Cn∗ ) −→ H m (lim Cn∗ ) −→ lim H m (Cn∗ ) → 0
←− ←− ←−
is exact.

Proof.
We consider for fixed degree m the two obvious exact sequences
0 → Bnm −→ Znm −→ H m (Cn∗ ) → 0 (2)
and
δ
0 → Znm −→ Cnm −→
n
Bnm+1 → 0. (3)
1. As the Cnm are supposed to satisfy the Mittag-Leffler condition, Lemma 2.12.5 implies
that
lim 1 Cnm = 0, for all m. (4)
←−
Lemma 2.12.4 applied to the short exact sequence (3) implies that the sequence
lim 1 Cnm −→ lim 1 Bnm+1 → 0
←− ←−
is exact and thus lim 1 Bnm+1 = 0. Therefore the sequence (2) yields that
←−
lim 1 Znm ∼
= lim H (Cn∗ ) .
1 m
←− ←−

122
2. In addition we know, again from Lemma 2.12.4 applied to (3), that the sequence
lim δ
− n lim B m+1
0 → lim Znm −→ lim Cnm ←
−→
←− ←− ←− n
is exact and hence the inverse limit of the m-cocycles is equal to the module of m-cocycles
in the inverse limit complex, i.e.

lim Znm ∼
= Z m (lim C ∗) .
←− ←− n

3. As the lim-one term on the inverse system of coboundaries is trivial by 1., we obtain from
(2) that the sequence

0 → lim Bnm −→ lim Znm −→ lim H m (Cn∗ ) → 0

←− ←− ←−
is exact as well. Lemma 2.12.4 applied to (3) tells us that the kernel of the connecting
homomorphism
∂ : lim Bnm −→ lim 1 Znm−1 → 0 (∗)
←− ←−
is isomorphic to the image of the map
lim δ
− n lim B m
lim Cnm−1 ←
−→
←− ←− n
and thus to the coboundaries, i.e.

B m (lim Cn∗ ) ∼
= ker ∂ .
←−
Thus, we get an inclusion B m (lim Cn∗ ) ⊂ lim Bnm . Therefore we get the following sequence
←− ←−
of inclusions

B m (lim Cn∗ ) ⊂ lim Bnm ⊂ lim Znm = Z m (lim Cn∗ ) (∗∗) ,

←− ←− ←− ←−
where the last identity is 2.

4. Recall that for any inclusion A ⊂ B ⊂ C of submodules, the diagram

0 0 0

 id
 
0 / A / A /0 / 0

  
0 /B /C / C/B / 0
id
  
0 / B/A / C/A / C/B / 0

  
0 0 0

has exact rows and the first two lines are exact so that by the nine-lemma the last row is
exact.

123
 5. Applied to the inclusion (∗∗) of submodules this yields the short exact sequence
lim Bnm Z m (lim Cn∗ ) lim Znm
0→ ←− −→ ←− −→ ←− →0
B m (lim Cn∗ ) B m (lim Cn∗ ) lim Bnm
←− ←− ←−
m ∗
is exact. The middle term is the cohomology H (lim Cn ) of the inverse limit complex.
←−
The right term is isomorphic to lim H m (Cn∗ ) and the left term is isomorphic to the lim-one
←−
term lim 1 H m−1 (Cn∗ ) because the kernel of ∂ is B m (lim Cn∗ ) and thus by (∗) the quotient
←− ←−
is lim 1 Znm−1 ∼
= lim 1 m−1
H (Cn∗ ) by 2.
←− ←−


Theorem 2.12.8 (Milnor sequence).

If XSis a CW complex with a filtration X0 ⊂ . . . ⊂ Xn ⊂ Xn+1 ⊂ . . . of subcomplexes with
X = n Xn , then the sequence
0 → lim 1 H m−1 (Xn ; G) −→ H m (X; G) −→ lim H m (Xn ; G) → 0
←− ←−
is exact for all abelian groups G.

Proof.
• We define Cn∗ = Hom(S∗ (Xn ), G). This system of cochain complexes satisfies the Mittag-
Leffler condition because the inclusions of chains
Sm (Xn ) ,→ Sm (Xn+1 )
dualize to epimorphisms
Hom(Sm (Xn+1 ), G) −→ Hom(Sm (Xn ), G) .

• The only thing we have to show to apply Lemma 2.12.7 is that for the term in the middle
H m (X; G) ∼
= H m (lim Hom(S∗ (Xn ), G)) .
←−
By Remark 2.12.3.1, the inverse limit has a universal property dual to the one of the
direct limit and the maps induced from the inclusions Xn ,→ X
Hom(S∗ (X), G) −→ Hom(S∗ (Xn ), G)
induce a homomorphism
Hom(S∗ (X), G) → lim Hom(S∗ (Xn ), G) .
←−
• To see that this is an isomorphism, first note that if a space X is the union of a directed
system of subspaces Xα with the property that each compact subset of X is contained in
some Xα , then for homology the map
lim Si (Xα ; G) → Si (X; G)
−→
is an isomorphism for all abelian groups G. (Note that by Corollary 1.11.16.1, this applies
to a filtration by subcomplexes.)
Indeed, for surjectivity, represent a cycle on X by a sum of finitely many simplices. The
union of their images is compact in X and thus contained in some Xα , which ensures
surjectivity. For injectivity, if a cycle in some Xα is a boundary in X, by compactness, it
is a boundary in some Xβ ⊃ Xα , hence represents zero in lim Hi (Xα ; G).
−→
124
 • The dual of this argument then shows the claim.


Example 2.12.9.
We consider the infinite complex projective space CP ∞ . It is defined as a limit lim CP n . This
−→
space has a natural structure of a CW complex with a cell in every even dimension. To apply
Theorem 2.12.8, we consider the skeleton filtration, i.e.
X0 = pt ⊂ X1 = CP 1 ⊂ X2 = CP 2 ⊂ . . .
Thus Xn is the 2n-skeleton of CP ∞ . The Milnor sequence 2.12.8 in this case is for each m
0 → lim 1 H m−1 (CP n ) −→ H m (CP ∞ ) −→ lim H m (CP n ) → 0 (∗) .
←− ←−
m−1 n+1 m−1 n
However, the maps H (CP ) → H (CP ) are surjective. By Remark 2.12.6.1, this
inverse system satisfies the Mittag-Leffler condition and thus by Lemma 2.12.5
lim 1 H m−1 (CP n ) = 0
←−
and therefore (∗) gives isomorphisms
H m (CP ∞ ) ∼
= lim H m (CP n ).
←−
The inverse limit of truncated polynomial rings Z[α]/αn+1 is isomorphic to the ring of formal
power series. Recall that for a commutative ring R, the ring R[[z]] of formal power series is the
set RN of sequences with value in R withP addition (an ) + (bn ) = (an + bn ) and multiplication
given by a Cauchy product (an ) ∙ (bn ) = ( nk=1 ak bn−k ).
Corollary 2.12.10.

H ∗ (CP ∞ ) ∼
= Z[[α]], with |α| = 2 ,
where Z[[α]] denotes the ring of formal power series in α.
The arguments are analogous for the infinite real and quaternionic projective spaces, RP ∞
and HP ∞ .
Corollary 2.12.11.

H ∗ (RP ∞ ; Z/2Z) ∼
= Z/2Z[[α]], with |α| = 1
and
H ∗ (HP ∞ ) ∼
= Z[[α]], with |α| = 4.
Remark 2.12.12.
1. At times, the cohomology of a space is considered as a direct sum
M
H ∗ (X; G) = H n (X; G) .
n>0

From that point of view, we only have finite sums in H ∗ (X; G) so that this interpretation
yields the identification of H m (CP ∞ ) and H ∗ (RP ∞ ; Z/2Z) as a polynomial ring: the
formulae
H ∗ (CP ∞ ) ∼
= Z[α] with |α| = 2
and
H ∗ (RP ∞ ; Z/2Z) ∼
= Z/2Z[α] with |α| = 1.
can be found in the literature as well.

125
 L
2. However, viewing homology H∗ (X) as a direct sum n Hn (X) and Q
for free H∗ (X) the
∗
cohomology as a dual, then the description of H (X) as a product n H n (X) is more
natural.

2.13 Lens spaces

Observation 2.13.1.
1. Let m ∈ N and let `1 , . . . , `n be natural numbers with gcd(m, `i ) = 1 for all i and assume
2πi
n > 2. Choose a primitive n-root of unity ζ := e m and define an action of Z/mZ on
S2n−1 by
% : Z/mZ × S2n−1 → S2n−1 ,
(ζ; z1 , . . . , zn ) 7→ (ζ `1 z1 , . . . , ζ `n zn ) ,
where we view S2n−1 as a subspace of Cn .
2. This action is free: if %(ζ r ; z1 , . . . , zn ) = (z1 , . . . , zn ) for some (z1 , . . . , zn ), then we have
ζ r`i zi = zi for all i. Since there exists i such that zi 6= 0, we find ζ r`i = 1 and thus
r`i = 0 mod m. Since `i is invertible modulo m, we find r = 0 mod m.
Example 2.13.2.
If m = 2, then the all integers `i must be odd and therefore the action
% : Z/2Z × S2n−1 → S2n−1
is the antipodal action.
We consider the quotient spaces S2n−1 /(Z/mZ).

Definition 2.13.3
The space L = L(m; `1 , . . . , `n ) = S2n−1 /(Z/mZ) with the action as described in Observation
2.13.1 as above is called lens space with parameters (m; `1 , . . . , `n ).

Remarks 2.13.4.
1. For m = 2 we get the real projective spaces L(2; `1 , . . . , `n ) = RP 2n−1 as lens spaces.
2. The classical case is the three manifold case: For integers p, q with gcd(p, q) = 1 one
considers L(p, q) := L(p; 1, q) = S3 /Zp with (ζ; z1 , z2 ) 7→ (ζz1 , ζ q z2 )
3. Note that the projection map π : S2n−1 −→ L(m; `1 , . . . , `n ) is a covering map, because
the Z/mZ-action is free.
We now want to consider CW structures on lens spaces that generalize the CW structures
on projective spaces.
Observation 2.13.5. 2πij
1. We start with a CW structure on S1 that has m zero cells {e m , 1 6 j 6 m} and m one
cells.
2. Let Bj2n−2 be the subset of Cn
2πij
Bj2n−2 := {cos θ(0, . . . , 0, e m ) + sin θ(z1 , . . . , zn−1 , 0)|
0 6 θ 6 π/2, (z1 , . . . , zn−1 ) ∈ S2n−3 },
2πij
i.e., we connect the point (0, . . . , 0, e m ) with all the points (z1 , . . . , zn−1 ) ∈ S2n−3 via
quarters of a circle. Thus we obtain a space homeomorphic to a (2n − 2)-dimensional disc,
Bj2n−2 ∼
= D2n−2 . A calculation shows that Bj2n−2 ⊂ S2n−1 .

126
 2πij 2πi(j+1)
3. If we connect all points on the circular arc in S1 between e m and e m with S2n−3 ,
again via quarters of a circle, we get a (2n − 1)-dimensional ball Bj2n−1 contained in S2n−1
with boundary
∂Bj2n−1 = Bj2n−2 ∪ Bj+1
2n−2
. (∗)
The two boundary discs Bj2n−2 and Bj+1 2n−2
are attached to each other via their common
boundary S 2n−3 2n−1
. Thus Bj looks like a (2n−1)-dimensional lens. The union of all Bj2n−1
is S2n−1 .

4. We have to understand the Z/mZ-action % on these cells. It restricts to the subspace

2πij 2πi(j+1)
S2n−3 , i.e. %(S2n−3 ) ⊂ S2n−3 . The arcs between the points e m and e m on S1 are
2n−2
permuted by % and therefore % permutes the (2n − 2)-dimensional balls Bj and the
2n−1
(2n − 1)-dimensional balls Bj .
For any r ∈ N with r`n = 1 mod m, the map %r has order m as well and

%r |Bj2n−2 : Bj2n−2 −→ Bj+1

2n−2
,

because 2πij 2πij 2πi(j+1)

2πir`n
ζ r`n e m =e m e m =e m .

Thus, %r identifies the two faces of Bj2n−1 , cf. (∗). Each of the balls Bj2n−1 is a fundamental
domain of the %r -action. Thus
L∼= Bj2n−1 /%r
for any j = 1, . . . , m.

5. There is a natural inclusion

L(m; `1 , . . . , `n−1 ) ⊂ L(m; `1 , . . . , `n )

which is given by mapping the class [(z1 , . . . , zn−1 )] to [(z1 , . . . , zn−1 , 0)]. Representing
the (2n − 3)-dimensional lens space L(m; `1 , . . . , `n−1 ) as Bj2n−3 / ∼, we see that we can
build L(m; `1 , . . . , `n ) out of L(m; `1 , . . . , `n−1 ) by attaching the (2n − 1)-cell Bj2n−1 and a
(2n − 2)-cell Bj2n−2 . Note that we really just have to take one of the latter, because Bj2n−2
2n−2
is identified with its neighbour Bj−1 in the quotient.
Inductively we get a cell structure of L with one cell in each dimension up to 2n − 1.

Example 2.13.6.
2πi
For n = 2, the lens spaces are quotients of S3 . Let m = 5 and `1 = 1 and `2 = 2, so ζ = e 5 .
Then the Bj3 are 3-balls with boundary Bj2 and Bj+1 2
. The 2-balls Bj2 consist of elements
2πij
cos θ(0, e 5 ) + sin θ(z, 0) for z ∈ S1 and 0 6 θ 6 π2 ; these are the two-dimensional discs
2πij
(sin θz, cos θe 5 ) ∈ S3 ⊂ C2 .

Observation 2.13.7.
1. Let us consider the cellular chain complex of the lens spaces. In Observation 2.13.5.5, we
constructed a CW structure such that

C∗ (L) = Z, ∗ = 0, . . . , 2n − 1 .

Let σ k be the cell corresponding to the ball Bjk . We need to compute the boundary maps.

127
 2. The top cell has trivial boundary,

d(σ 2n−1 ) = σ 2n−2 − σ 2n−2 = 0

because the topological boundary of Bj2n−1 is the union of two balls one dimension lower
which are identified in the quotient.
3. The boundary of the cell σ 2n−2 is S2n−3 and the attaching map is the quotient map

S2n−3 −→ L(m; `1 , . . . , `n−1 ) .

The action % permutes the cells cyclically, and we get degree m:

d(σ 2n−2 ) = mσ 2n−3 .

By induction we see that the boundary maps are given by multiplication by zero respec-
tively m. Thus the homology of the lens space is the homology of the cellular complex
0 m 0 m 0
0 → Z −→ Z −→ Z −→ . . . −→ Z −→ Z → 0

and thus 
Z,
 ∗ = 0, 2n − 1,
H∗ (L(m; `1 , . . . , `n )) = Z/mZ, ∗ odd and < 2n − 1,


0, otherwise.
Note that we also get H1 (L) = π1 (L) = Z/mZ from covering theory because π1 (S2n−1 ) = 0
for n > 2 and thus S2n−1 is a universal cover of L.
4. As the top homology group is Z, Theorem 2.6.11 implies that lens spaces are compact
connected orientable manifolds of dimension 2n − 1.
The universal coefficient theorem 2.2.5 immediately gives for cohomology with coefficients
in Zm :
Lemma 2.13.8.
The additive cohomology groups are
(
Z/mZ, for all degrees 0 6 ∗ 6 2n − 1
H (L; Z/mZ) ∼
∗
=
0, ∗ > 2n − 1.

Note that the homology groups of L with coefficients in Z/mZ are isomorphic to the coho-
mology groups just by using the universal coefficient theorem 1.14.17

Hk (L; Z/mZ) ∼
= Hk (L; Z) ⊗ Z/mZ ⊕ Tor(Hk−1 (L), Z/mZ)
or by applying Poincaré duality 2.8.3, since L is compact and orientable.
We now focus on the case when m = p is a prime.
Proposition 2.13.9.
Let L = L(p; `1 , . . . , `n+1 ) be a lens space. Denote by α ∈ H 1 (L; Z/pZ) and β ∈ H 2 (L; Z/pZ)
(additive) generators. The cohomology group H j (L; Z/pZ) is then generated by
(
β i, for j = 2i
αβ i , for j = 2i + 1.

128
Proof.
We prove the claim by induction on n.

• For n = 1, we have a three-dimensional lens space L = L(p; `1 , `2 ). If α ∈ H 1 (L; Z/pZ)

and β ∈ H 2 (L; Z/pZ) are generators, then a cup product pairing argument from Lemma
2.11.4 shows that α ∪ β is a generator in degree three. We have to understand what α2 is:
if p is odd, then by the graded symmetry of the cup product, we have α2 = 0. For p = 2
we know that the lens space is RP 3 . In this case, by Corollary 2.11.6, α2 is a generator
of H 2 (L, Z/2Z). Thus, it is equal to β. In all other degrees, the cohomology groups are
trivial.

• Assume now that the claim is true up to degree n. We consider the inclusion

L(p; `1 , . . . , `n ) ,→ L(p; `1 , . . . , `n+1 ) =: L2n+1 .

Up to degree 2n − 1 this inclusion gives rise to an isomorphism on cohomology groups. We

know that β i generates the cohomology groups up in even degrees j = 2i < 2n−1 and αβ i
generates the cohomology groups in odd degrees j = 2i + 1 6 2n − 1. An argument as for
projective spaces, cf. Lemma 2.11.5, then shows that β ∪β n−1 generates H 2n (L2n+1 ; Z/pZ)
and β ∪ αβ n−1 = αβ n generates H 2n+1 (L2n+1 ; Z/pZ).

Corollary 2.13.10.
1. As graded rings
(
Λ(α) ⊗ Z/pZ[β]/β n+1 , p > 2,
H ∗ (L(p; `1 , . . . , `n+1 ); Z/pZ) ∼
=
Z/pZ[α]/α2n+2 , p = 2.

2. Fix a prime p and a sequence (`1 , `2 , . . .) of integers coprime to p. Let L denote the direct
limit of any system of the form

L(p; `1 , . . . , `n+1 ) ⊂ L(p; `1 , . . . , `n+2 ) ⊂ . . .

then (
Λ(α) ⊗ Z/pZ[[β]], p > 2,
H ∗ (L; Z/pZ) ∼
=
Z/pZ[[α]], p = 2.

Proof.
The second claim follows with the help of the Milnor sequence 2.12.8 as in Example 2.12.9. 

Remark 2.13.11.
Note that these cohomology groups do not dependent on the `i ’s.
Lens spaces of dimension three give rise to important examples of orientable connected and
compact 3-manifolds that have the same fundamental group and homology groups, but that are
not homotopy equivalent. For instance the lens spaces L(5; 1, 1) and L(5; 1, 2) are not homotopy
equivalent (cf. Hatcher, exercise 3.E.2), but have the same fundamental groups and the same
homology groups.

129
Observation 2.13.12.
We can interpret the generator β in terms of the so-called Bockstein-homomorphism.
The two short exact sequences of abelian groups
∙p ρ ∙p
0 → Z → Z → Z/pZ → 0 and 0 → Z/pZ → Z/p2 Z → Z/pZ → 0
give rise to short exact sequences of cochain complexes
0→ S ∗ (X; Z) → S ∗ (X; Z) → S ∗ (X; Z/pZ) → 0
0 → S ∗ (X; Z/pZ) → S ∗ (X; Z/p2 Z) → S ∗ (X; Z/pZ) → 0
and we get by Lemma 1.5.6 a corresponding long exact sequences of cohomology groups. Let
β̃ : H n (X; Z/pZ) → H n+1 (X; Z)
be the connecting homomorphism for the first sequence, let
β : H n (X; Z/pZ) → H n+1 (X; Z/pZ)
be the connecting homomorphism for the second sequence and let
ρ∗ : H n+1 (X; Z) → H n+1 (X; Z/pZ)
be induced by the reduction of the coefficients mod p. Then β is called the
Bockstein homomorphism.
Lemma 2.13.13.
For all n, the diagram
β̃
H n (X; Z/pZ) / H n+1 (X; Z)
RRR
RRR β
RRR ρ∗
RRR
R) 
H n+1 (X; Z/pZ)
commutes.

Proof.
For the proof just note that the diagram relating the two short exact sequences
∙p ρ
0 / Z / Z / Z/pZ / 0
ρ ρ2 id
 ∙p
 
0 / Z/pZ / Z/p2 Z / Z/pZ / 0
commutes and therefore we obtain the commutativity of the connecting homomorphisms, the
naturality statement of Proposition 1.5.5, implies
β̃
H n (X; Z/pZ) / H n+1 (X; Z)
id ρ∗
 β

H (X; Z/pZ)
n / H n+1
(X; Z/pZ).


With the help of this auxiliary result we will show that the class β ∈
H (L(p; `1 , . . . , `n+1 ); Z/pZ) in Proposition 2.13.9 is the image of the Bockstein homomorphism
2

applied to α, i.e. β = β(α). We discuss the example p = 2, i.e. the cases of real projective spaces
of odd dimension in detail; the cases for odd prime are similar.

130
Proposition 2.13.14.
The Bockstein homomorphism β : H n (RP ∞ ; Z/2Z) → H n+1 (RP ∞ ; Z/2Z) is an isomorphism
for odd n and is trivial for even n. In particular, β(α) = α2 .

Proof.
Consider the diagram

H n+1 (RP ∞ ; Z)
∙2
β̃

∙2
H n (RP ∞ ; Z/2Z) / H n+1 (RP ∞ ; Z) / H n+1 (RP ∞ ; Z)
TTTT
TTTTβ
TTTT ρ∗
TTT) 
H n+1 (RP ∞ ; Z/2Z)
β̃

H n+2 (RP ∞ ; Z)

• If n is odd, then n + 1 = 2k for some k and then H 2k (RP ∞ ; Z) ∼ = Z/2Z so that the
multiplication by 2 is trivial. The horizontal exact sequence then implies that β̃ is surjec-
tive. But both adjacent groups are Z/2Z, thus β̃ is an isomorphism, since any surjective
endomorphism of Z/2Z is an isomorphism.

• For even n, the groups H n+1 (RP ∞ ; Z) are trivial, hence in these degrees β̃ = 0, and also
the Bockstein homomorphism β = ρ∗ ◦ β̃ vanishes for even n.

• The same fact implies that for odd n, the lowest arrow in the exact column is zero. Thus
ρ∗ : H n+1 (RP ∞ ; Z) → H n+1 (RP ∞ ; Z/2Z) is surjective and, by the same arguments, an
isomorphism and therefore β is an isomorphism.

Remark 2.13.15.
Using that β is a connecting homomorphism and thus defined using a coboundary, one can use
the Leibniz rule 2.5.10.3 to show that it is a derivation with respect to the cup product:

β(α ∪ γ) = β(α) ∪ γ + (−1)|α| α ∪ β(γ).

The Bockstein homomorphism is one example of a cohomology operation.

2.14 A first quick glance at homotopy theory

The definition of a fundamental group has an obvious generalization:

Definition 2.14.1
Denote by I := [0, 1] the standard interval and by I n ⊂ Rn the n-dimensional unit cube.

1. For a space X with base point x0 ∈ X, we denote by πn (X, x0 ) set of homotopy classes
of maps
F : (I n , ∂I n ) → (X, x0 ) ,

131
 where homotopies are required to satisfy Ft (∂I n ) = x0 for all t ∈ [0, 1]. (For n = 0, take
I 0 to be a point and ∂I 0 = ∅ so that π0 (X) is the set of path-connected components of
X.)

2. For a subset A ⊂ X and a base point x0 ∈ A, take for n > 1 the subset I n−1 ⊂ ∂I n
the points with last coordinate sn = 0; they are homeomorphic to an (n − 1)-dimensional
cube. Finally let J n−1 be the closure of the complement ∂I n \ I n−1 . Then πn (X, A, x0 ) is
the set of homotopy classes of maps

(I n , ∂I n , J n−1 ) → (X, A, x0 ) .

Observation 2.14.2.
1. Generalizing the case of the fundamental group n = 1, we turn πn (X, x0 ) into a group
with composition defined by operations involving the first coordinate,

f (2s1 , . . . , sn ) for s1 ∈ [0, 12 ]
(f + g)(s1 , s2 , . . . , sn ) :=
g(2s1 − 1, . . . , sn ) for s1 ∈ [ 12 , 1]

This is well-defined on homotopy classes; since only the first coordinate is involved, the
same arguments as for the fundamental group show that we obtain a group structure.
The group πn (X, x0 ) is called the n-th homotopy group of X.

2. A sequence of homotopies shows that for n > 2, the group πn (X, x0 ) is abelian. Note that
for this process, we only need the two coordinates s1 , s2 .
Similar arguments as for the fundamental group show that the different base points yield
isomorphic homotopy groups. Indeed, the fundamental group π1 (X, x0 ) acts on all groups
πn (X, x0 ); for n = 1 this is the inner action of π1 (X, x0 ) on itself.

3. In the relative case, the last coordinate sn plays a special role. For this reason, πn (X, A, x0 )
has the structure of a group only for n > 2. It is abelian for n > 3. We call it the
relative homotopy group. (The set π1 (X, A, x0 ) is the set of homotopy classes of paths in
X from a varying point in A to the base point x0 .)
Familiar features of the fundamental group can be extended:
Observation 2.14.3.
1. A covering (X̃, x̃0 ) → (X, x0 ) induces an isomorphism p∗ : πn (X̃, x̃0 ) → πn (X, x0 ) for
n > 2.
Indeed, injectivity follows from [Topologie, 2.7.13] and surjectivity from the fact that the
maps I n → X factorize to maps Sn → X and that Sn is simply connected for n > 2.

2. As a consequence, πn (X, x0 ) with n > 2 vanishes for all spaces X with a contractible
cover, e.g. the sphere S1 with cover R or the torus T n with cover Rn . Such spaces are
called aspherical.

3. Base-point preserving maps of pairs ϕ : (X, A, x0 ) → (Y, B, y0 ) give rise to maps ϕ∗ :

πn (X, A, x0 ) → πn (Y, B, y0 ). They obey the familiar relations id ∗ = id and (f ◦g)∗ = f∗ ◦g∗ .
Homotopic maps ϕ, ϕ0 : (X, A, x0 ) → (Y, B, y0 ) give the same maps in relative homotopy,
ϕ∗ = ϕ0∗ .
Q Q
4. We have for arbitrary products πn ( α Xα ) = α πn (Xα ).

132
 5. Given an imbedding i : A → X and a base point x0 ∈ A, one gets a long exact sequence
ι ∗ j∗ ∂
. . . πn (A, x0 ) → πn (X, x0 ) → πn (X, A, x0 ) → πn−1 (A, x0 ) → . . . → π0 (X, x0 )

of homotopy groups. Here j : (X, x0 , x0 ) ,→ (X, A, x0 ) and and ∂ comes from restricting
maps (I n , ∂I n , J n−1 ) → X to I n−1 .

6. There is, however, no general analogue of the excision property for homology or the
Seifert-van Kampen theorem for the fundamental group, see however Theorem 2.14.8.
This makes homotopy groups difficult to compute, even for spheres.

However, homotopy is theoretically important, because it gives strong invariants, in partic-

ular for CW complexes:

Theorem 2.14.4 (Whitehead).

If a map f : X → Y between connected CW complexes induces isomorphisms f∗ : πn (X) →
πn (Y ) for all n > 0, then f is a homotopy equivalence.
In the case when f is the inclusion of a subcomplex with the same property, then X is even
a deformation retract of Y .

The proof is based on the following

Lemma 2.14.5 (Compression lemma).

Let (X, A) be a CW pair and let (Y, B) be any pair with B 6= ∅. Assume that for each n such
that X \ A has cells of degree n, we have πn (Y, B, y0 ) = 0 for all y0 ∈ B. Then every map
f : (X, A) → (Y, B) is homotopic relative A to a map X → B, i.e. we can homotop the map
such that the image is in B ⊂ Y .

Proof.
Assume inductively that f has been homotoped to take the skeleton X k−1 to the subspace B.
Let φ be the characteristic map of a k-cell ek of X \ A. The composition

f ◦φ: (Dk , ∂Dk ) → (Y, B)

can be homotoped relative ∂Dk into B, since we assumed πk (Y, B, y0 ) = 0 for all y0 ∈ B.
This homotopy induces on the quotient space

X k−1 t Dk → X k−1 ∪φ ek

a homotopy relative X k−1 . We do this for all k-cells of X \A at once, take the constant homotopy
on A and get a homotopy of the restriction f |X k ∪A to a map into B.
Inductively, we settle the case when the dimension of the cells of X \ A is bounded. In
general, deal with X k during the t-interval [1 − 2−k , 1 − 2−(k+1) ]. 

Proof.
of Whitehead’s Theorem 2.14.4.

• First suppose that f is the inclusion of a subcomplex. Consider the long exact sequence
2.14.3.5 for the pair (Y, X). Since f induces isomorphisms on the homotopy groups,
the relative homotopy groups πn (Y, X) vanish. Applying Lemma 2.14.5 to the identity
(Y, X) → (Y, X) yields a deformation retraction of Y onto X, as claimed.

133
 • Now consider the mapping cylinder Mf of f : X → Y : this is the quotient

X × I t Y → Mf

under the identification (x, 1) ∼ f (x). Thus Mf contains X = X ×{0} and Y as subspaces.
Mf deformation retracts to Y . Thus f is a composition

X ,→ Mf → Y

of an inclusion and a retraction. A retraction is a homotopy equivalence; thus it suffices

to show that Mf retracts onto X, if f induces isomorphisms on the homotopy groups (or,
equivalently, if all relative groups πn (Mf , X) vanish).
• If f happens to be cellular, then (Mf , X) is a CW pair and we are done by the first part of
the proof. In the general case, one can invoke a theorem that f is homotopic to a cellular
map.


Remarks 2.14.6.
1. We do not claim that any two CW complexes with isomorphic homotopy group are homo-
topy equivalent; rather the existence of a map f inducing the isomorphisms in homotopy
is required.
As a counterexample, consider X = RP 2 and Y = S2 × RP ∞ . Both have fundamental
group Z2 ; their universal covers are X̃ = S2 and Ỹ = S2 × S∞ which are homotopy
equivalent, since S∞ is contractible. Thus Observation 2.14.3.1 implies that the homotopy
groups are all isomorphic.
But the two spaces cannot be homotopy equivalent, since their homology differs: since
Y = S2 × RP ∞ retracts to RP ∞ , it has non-vanishing homology in infinitely many
components, in contrast to X.
2. There is a CW complex, unique up to homotopy, which has the property that it has a
single non-vanishing homotopy group G in degree n. Such a space K(G, n) is called an
Eilenberg-Mac Lane space. (The group G has to be abelian for n > 1.)
Cohomology classes for a CW complex X correspond bijectively to homotopy classes of
maps X → K(G, n): for any abelian group G and for all CW complexes X, there is for
n > 0 a natural bijection

T : [X, K(G, n)] → H n (X; G) .

More precisely, there is a distinguished class α ∈ H n (K(G, n); G) such that T (f ) = f ∗ (α).
This is a strong link between homotopy theory and cohomology theory.
3. For CW complexes, we can replace maps within the same homotopy class by cellular maps:
every map f : X → Y of CW complexes is homotopic to a cellular map. If f is already
cellular on a subcomplex A ⊂ X, then the homotopy may be taken to be stationary on
the subcomplex A.
As a consequence, πn (Sk ) = 0 for n < k. Indeed, with with usual CW structure on spheres
consisting of a 0-cell and a top-dimensional cell, cf. Example 1.11.4.3, and the 0-cells as
base points, any map Sn → Sk can be homotoped fixing the based point to a cellular map
which is thus constant.

134
 We finally state a generalization of Proposition 1.3.8 which relates homotopy groups to
homology groups:
Theorem 2.14.7 (Hurewicz).
If a space X is (n − 1)-connected with n > 2, i.e. if πi (X) = 0 for all i 6 n − 1, then H̃ i (X) = 0
for i 6 n − 1 and πn (X) = Hn (X).
Thus the first nonzero homotopy and homology group of a simply-connected space occur in
the same degree and are isomorphic.
We finally explain the best available analogue of excision:
Theorem 2.14.8.
Let X be a CW complex that is decomposed as the union of subcomplexes A and B with
non-empty connected intersection C := A ∩ B. Suppose that (A, C) is n-connected and (B, C)
is m-connected, with m, n > 0. Then the map

πi (A, C) → πi (X, B)

induced by the inclusion is an isomorphism for i < m + n and a surjection for i = m + n.

This yields
Corollary 2.14.9 (Freudenthal suspension theorem).
The suspension map
πi (Sn ) → πi+1 (Sn+1 )
is an isomorphism for i < 2n − 1 and a surjection for i = 2n − 1. More generally, this holds
for the suspension πi (X) → πi+1 (ΣX) of any (n − 1)-connected CW complex X. (Note that by
Remark 2.14.6.3 X = Sn is (n − 1)-connected.)

Proof.
Decompose the suspension ΣX as the union of two cones C± X intersecting in a copy of X.
Inclusion gives us by Theorem 2.14.8 a morphism

πi+1 (C+ X, X) → πi+1 (ΣX, C− X) (∗) .

The long exact sequence for (C+ X, X)

. . . → πi (X) → πi (C+ X) → πi (C+ X, X) → πi−1 (X) → πi−1 (C+ X) → . . .

together with the fact that the cone C+ X is contractible shows for the left hand side of (∗) the
isomorphism πi+1 (C+ X, X) ∼ = πi (X). The long exact sequence for (ΣX, C− X)
. . . → 0 = πi (C− X) → πi (ΣX) → πi (ΣX; C− X) → πi−1 (C− X) = 0 → . . .

shows that πi+1 (ΣX, C− X) ∼= πi+1 (ΣX).

Now suppose that X is (n − 1)-connected. Then the first long exact sequence implies that
the pairs (C± X, X) are n-connected. By Theorem 2.14.8 the inclusion map is an isomorphism
for i + 1 < 2n and surjective for i + 1 = 2n.


Corollary 2.14.10.
We have πn (Sn ) ∼
= Z for all n > 1, with the identity map as a generator. In particular, the
degree provides an isomorphism πn (Sn ) → Z.

135
Proof.
From Corollary 2.14.9, we know that in the suspension sequence

π1 (S1 ) → π2 (S2 ) → π3 (S3 ) → . . .

the first map is surjective and all other maps are isomorphisms. Since π1 (S1 ) is infinite cyclic,
generated by the identity map, it follows that all other groups πn (Sn ) are finite or infinite cyclic
groups generated by the identity map.
The group cannot be finite: there exist base-point preserving maps Sn → Sn of arbitrary
degree, cf. Lemma 1.10.3 for a weaker statement, and the degree is a homotopy invariant.
The degree map is an isomorphism, since the map z 7→ z of S1 has degree 1 and so do by
Lemma 1.10.3 its iterated suspensions. 

A English - German glossary

English German
boundary Rand
chain complex Kettenkomplex
chart Karte
cone Kegel
connecting homomorphism Verbindungshomomorphismus
cycle Zykel
excision Ausschneidung
hairy ball theorem Satz vom gekämmten Igel
lens space Linsenraum
manifold Mannigfaltigkeit
skeleton Gerüst
support Träger
suspension Einhängung

136
Index
p-adic integers, 132 degree, 41
Čech cohomology, 116 degree of a map, 104
diagonal approximation, 92
abelianization, 12 diagram chase, 20
acyclic complex, 2 diameter, 28
affine simplex, 28 differential, 1
Alexander-Whitney map, 92 dimension, 96
antipodal map, 43 dimension axiom, 85
aspherical space, 146 direct limit, 107
atlas, 96 direct sum, 4
attaching map, 46 direct system, 107
axioms of a cohomology theory, 85 directed poset, 107
barycentric subdivision, 26 directed system, 107
Betti number, 122 Eilenberg-Mac Lane space, 148
Bockstein homomorphism, 143 Eilenberg-Zilber theorem, 76
boundary, 1 euclidean neighborhood retract, 116
boundary operator, 1, 7 exact sequence, 18
cell, 45 excision, 32
cell decomposition, 45 external cup product, 90
cellular chain complex, 57 face map, 5
cellular map, 50 fibration, 53
chain, 1 finite CW complex, 46
chain complex, 1 five-lemma, 35
chain homotopy, 3 free abelian group, 6
chain homotopy equivalence, 4 free chain complex, 68
chain map, 2 free resolution, 65
characteristic map, 46 front face, 86
chart, 96 fundamental class, 34, 103
closed manifold, 122
closure finite condition, 46 Hairy Ball theorem, 44
coboundary operator, 77 homology cross product, 73
cochain complex, 77 homology group, 2
cofibration, 53 homotopy extension property, 53
cohomology cross product, 90 homotopy group, 145
cohomology group, 77, 78 homotopy lifting property, 53
cohomology group with compact support, 105 Hurewicz-homomorphism, 12
compact-open topology, 51
cone, 26, 39 intersection form, 126
connecting homomorphism, 20 inverse limit, 129
cup product pairing, 125 inverse system, 129
cup-product, 91 Kronecker pairing, 79
CW complex, 45
CW pair, 48 lens space, 138
cycle, 1 locally Euclidean space, 96
long exact sequence, 18
deformation retract, 24

137
manifold, 97
map of cochain complexes, 78
Milnor sequence, 135
Mittag-Leffler condition, 131

orientable manifold, 98
orientation, 98

paracompact, 54
Poincaré duality map, 112
polyhedron, 47

rear face, 86
reduced homology, 36
relative boundary, 22
relative chain, 22
relative chain complex, 22
relative cycle, 22
relative homology group, 22
relative homotopy group, 146
relative Mayer-Vietoris sequence, 36

short exact sequence, 18

simplex, 4
simplicial complex, 47
simplicial map, 48
singular n-cochains with compact support, 105
singular chain complex, 7
singular chain module, 6
singular cochain group, 78
singular simplex, 5
skeleton, 48
standard resolution, 65
subcomplex, 48
suspension, 40
suspension isomorphism, 40

tensor product, 62
topological Künneth formula, 76

universal coefficient theorem, 68

vertex, 28

weak retract, 23

138