Académique Documents
Professionnel Documents
Culture Documents
2012–13
Part B: Semester 2
Dr E Cheng
School of Mathematics and Statistics, University of Sheffield
E-mail: e.cheng@sheffield.ac.uk
http://cheng.staff.shef.ac.uk/mas435/
Contents
1 Chain complexes and homology 3
1.1 Chain complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Products of abelian groups . . . . . . . . . . . . . . . . . . . . . 4
1.4 Free abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Maps between chain complexes . . . . . . . . . . . . . . . . . . . 6
2 Low-dimensional cell-complexes 7
2.1 Low-dimensional examples . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Simplices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
8 Further remarks 25
8.1 Moore spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
8.2 Wedge sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
8.3 Suspension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Introduction
We have seen that Algebraic Topology is about studying topological spaces using
algebra. In the first part of the course the algebra we used was group theory.
We saw how to study a topological space X via its “fundamental group” π1 X
which helped us some features of spaces but not others. The main limitation
was that the fundamental group is constructed from loops in a space, and cannot
detect higher-dimensional features.
S 1 −→ X
S n −→ X
for higher values of n; these loops form a group called the nth homotopy group
πn X. The trouble with this approach is that it is very difficult to compute.
For the first step, it helps if we know how our how space is built up dimension
by dimension from “cells”, that is, some form of disk/ball. Homology will then
Formally, this is done by taking a quotient group. This quotient group measures
how “holey” the space is, because it measures the “holes” that were not filled
in by disks.
We need to start by looking at the algebraic objects we’ll be using: chain com-
plexes of abelian groups. All the groups in this part of the course are abelian,
which makes a lot of things easier.
δn+2 δn+1 δn δ2 δ1 δ0
··· Cn+1 Cn Cn−1 · · · C2 C1 C0 0
1.2 Homology
0 ×2 0 0
0 Z Z Z 0
0 0 0 0
0 Z Z Z 0
0 1 1 0
0 Z Z Z 0
0 δn = 0 0 0
0 Z 0 0 ··· 0 0
Recall that given two abelian groups A and B we can form their direct product
A × B, elements of which are pairs (a, b) where a ∈ A and b ∈ B. This is again
an abelian group, inheriting its group operation from A and B: (a, b)+ (a′ , b′ ) =
(a + a′ , b + b′ ). As A × B is abelian here, the group operation will be written
additively.
Similarly, recall the direct sum of A and B, the abelian group A ⊕ B. This has
elements a ⊕ b where again a ∈ A and b ∈ B. The group operation is given by
(a ⊕ b) + (a′ ⊕ b′ ) = (a + a′ ) ⊕ (b + b′ ), again using the group operations of A and
B. There is then an obvious isomorphism of abelian groups A × B ∼ = A ⊕ B.
Exercise 1.8. Construct this isomorphism. You should construct the map,
show it’s a group homomorphism, and show it’s an isomorphism.
The free abelian group on one generator is isomorphic to Z. For two generators
it is isomorphic to Z ⊕ Z. Carrying on in this way, for k generators, it is
isomorphic to Z⊕k . If we write the generators/basis elements as e1 , . . . , ek then
a general element of Z⊕k is Σki=1 λi ei , where λi ∈ Z. All of our chain complexes
will involve free abelian groups generated by cells of the space.
δ2 δ1 δ0
0 Z⊕Z Z 0
α a
β a
where δ2 and δ0 are the usual zero maps and δ1 maps the generators of Z ⊕ Z as
shown. Notice that ker δ0 = Z and Im δ1 = Z, so H0 = Z/Z = 0. Also notice
that the kernel of δ1 is generated by one element, α − β, and that Im δ2 = 0,
thus H1 = Z/0 = Z. The higher homology groups, for n ≥ 2, are then all the
trivial group 0.
1.5 Maps between chain complexes 6
Just as there are maps between abelian groups, we can define maps between
chain complexes.
δn
An An−1
fn fn−1
Bn Bn−1
ǫn
Proof. Given the chain map f , as above, we need to define a group homo-
morphism Hn (f ) : ker δn / Im δn+1 → ker ǫn / Im ǫn+1 . Given a ∈ ker δn , is
fn (a) ∈ ker ǫn ? Yes, by commutativity of the following diagram.
δn+1 δn
An+1 An An−1
fn+1 fn fn−1
Bn+1 Bn Bn−1
ǫn+1 ǫn
That is to say fn−1 δn (a) = fn−1 (0) = 0 = ǫn fn (a). So we have a map ker δn →
ker ǫn . Now consider an element a ∈ Im δn+1 , so a = δn+1 (x) say. We need that
fn (a) ∈ Im ǫn+1 . However fn (a) = fn δn+1 (x) = ǫn+1 fn+1 (x) by commutativity
of the diagram. Thus fn (a) ∈ Im ǫn+1 . Thus the map ker δn → ker ǫn restricts
to a map Im δn+1 → Im ǫn+1 and so we can construct the map Hn (A) →
Hn (B), as required.
Note that for each n, Hn is a functor from the category ChCpx of chain com-
plexes and chain maps into the category Ab of abelian groups and group ho-
momorphisms. The above proposition gives us the action on morphisms.
2 Low-dimensional cell-complexes
Now that we’ve seen how to produce homology groups from a chain complex,
we need to see how to produce a chain complex from a topological space. There
are various different ways of doing this, like different “recipes”. They do not
all produce the same answer, but the clever part is that the resulting homology
groups are the same. This is a profound and amazing result that is beyond the
scope of this course.
One way of producing a chain complex from a space is to start with a cell
complex structure on the space. We’ve already seen that you can build the
same space using different cell complex structures, but again, this doesn’t matter
because you’ll get the same homology groups at the end, no matter what cell
complex structure you choose.
Cn = the free abelian group generated by the n-cells of our cell complex,
δn = boundaries, considering orientation in a way that we’ll see.
• For 2-cells the boundary is the sum of all the boundary 1-cells, taking
orientation into account (a bit like when we did van Kampen’s theorem
before, and were reading off the boundary of a cell). So if a 1-cell is
pointing “backwards” it gets a minus sign.
Example 2.2. A possible cell complex for the Torus is the following.
x b x
a α a
x x
b
2.1 Low-dimensional examples 8
This has
Recall that the free abelian group on one generator is isomorphic to Z, and with
two generators it is isomorphic to Z ⊕ Z.
0 0 0 0
··· 0 Z Z⊕Z Z 0
α 0 a 0
b 0
Exercise 2.3. Below are possible cell complexes for the circle S 1 , the sphere
S 2 , the Klein bottle and the real projective plane RP 2 . Find the associated
chain complexes and work out the homology.
x x b x
a a
a x α y a α a x α x
a a
x x
x
b
S1 S2 Klein bottle RP 2
Exercise 2.4. Try some different cell complex structures for the same spaces
and make sure you get the same homology groups. This is a useful way of
checking your answers.
To prove that we get the same homology for different cell structures on the same
space, it is easier to use singular homology, which is more general.
2.2 Simplices 9
2.2 Simplices
Simplices are topologically the same as discs and balls but are geometrically
different and more combinatorial. They look more complicated than discs and
balls, but the extra structure makes it easier1 to define boundaries and orienta-
tion at higher dimensions. In practice we won’t use this much in this course as
we’ll stick to lower dimensional things.
The above examples show that simplices are based around sticking together
triangles. The 2-simplex is homemorphic D2 the disc whilst the 3-simplex is
homemorphic to D3 the ball. The standard examples of discs and balls are the
unit discs and unit balls. For example, D2 = {(x, y) ∈ R2 | x2 + y 2 ≤ 1}. There
are also standard simplices which have a similar description.
∆0 ⊆ R ∆1 ⊆ R2 ∆2 ⊆ R3
(0, 1) (0, 1, 0)
1
(1, 0) (1, 0, 0)
(0, 0, 1)
Given any n + 1 points in Rm that do not lie in a hyperplane of dim < 2, these
span a simplex. That is, they are the vertices of an n-simplex. To specify an
n-simplex we just give its vertices and, given an ordering on the vertices, there
is a canonical linear homomorphism from any n-simplex to any other. We can
refer to an n-simplex as [v0 , . . . , vn ], i.e. just its vertices. If we delete one of the
vi ’s then we get n vertices spanning an (n − 1)-simplex, called a face. Thus an
n-simplex has n faces.
Example 2.7. The simplices below represent the torus and the Klein bottle.
x b x x b x
α α
c c
a a a a
β γ
x x x x
b b
Torus Klein Bottle
δn+1 δn
··· ∆n+1 (X) ∆n (X) ∆n−1 (X) ···
The boundaries here are the sum of the faces, taking into account the orienta-
tion.
The boundaries use (−1)i and the face inherits its ordering from the larger
simplex. Using the other notation, if we let ∆n = [v0 , . . . , vn ], then the ith
face is ∆in−1 = [v0 , . . . , vi−1 , vi+1 , . . . , vn ]. Sometimes this may be written as
[v0 , . . . , vi−1 , vˆi , vi+1 , . . . , vn ].
[1]
[0, 1] [1, 2]
[0, 1, 2]
The idea of singular homology is that we don’t have to start with a specific ∆-
complex structure. Recall that a path in a space X is a continuous map I → X.
A singular n-simplex in X is then a continuous map ∆n → X where ∆n is the
standard n-simplex. We can the form the singular chain complex for X where
Cn (X) is the free abelian group generated by all the singular n-simplices. The
boundary map is the same as before and the singular homology of X is then
the homology of the singular chain complex. NB. The 0-simplex is a point, so
a singular 0-simplex in X is any point of X. Thus C0 (X) is the free abelian
group generated by all of the points in X.
Example 3.1. We will compute the singular homology of a point. For all n ≥ 0
there is precisely one map ∆n → {∗}. The boundary map δn : Cn (∗) → Cn−1 (∗)
n−1
is given as before. So an n-simplex σ ∈ Cn (X) is sent to Σi=0 (i1)i σi . This is 0
when n is odd and σi when n is even. Thus we have a singular chain complex
1 0 1 0 0
··· Z Z Z Z 0
When we consider the homology of the point we still end up with the 0th ho-
mology group being Z which of course just doesn’t feel quite right. It would be
nice if H0 (∗) was the trivial group 0. We can achieve this by using an “aug-
mented chain complex” which replaces the boundary map, for each space X,
δ0 : C0 (X) → 0, with a group homomorphism ǫ : C0 (X) → Z as in the chain
complex below.
δ2 δ1 δ0 ǫ 0
··· C2 (X) C1 (X) C0 (X) Z 0
Σi (ni σi ) Σi ni
The homology of this chain complex is called the reduced homology of X, written
H̃n (X).
3.2 Homotopy invariance 12
Exercise 3.2. Check that this gives 0 for all the homology groups of the point.
ǭ
C0 (X)/ Im δ1 Z
[nσ] n
So far we have a way of taking a space X and creating a chain complex C(X).
We also have a way of taking a chain complex C(X) and getting homology
groups Hn (X). The idea now is then to take the idea of homotopy through
these processes.
Top ChCpx Ab
f f• Hn f
There is one thing in this picture that we haven’t come across yet.
δn+1 δn
··· An+1 An An−1 ···
1
fn+1 gn+1 Pn fn gn Pn
−
fn−1 gn−1
From the examples we have computed so far, it might be clear that when the
first homology group H1 (X) of a space is abelian then it is the same as the fun-
damental group π1 (X). In fact, there is a way to take any group and abelianise
it, i.e. make it abelian in the nicest possible way. We will see that H1 (X) is
always the abelianisation of π1 (X).
4.1 Abelianisation
Exercise 4.2. For any group G, the commutator subgroup [G, G] is normal.
Definition 4.3. Let G be a group, again not necessarily abelian. The abelian-
isation of G is defined to be Gab = G/[G, G].
G Gab
∃!f¯
f
The map along the top of the triangle is the quotient map G → G/[G, G] = Gab .
where each ni ∈ Z. The set {a1 , . . . , ak } is then said to be a generating set for
A.
Remark 4.5. Note that we are finitely generating A as a Z-module and this
expression is not necessarily unique. It is unique if A is a free group on the gen-
erators {a1 , . . . , ak }. In fact, every group can be expressed in terms of (maybe
infinitely many) generators and relations. There is a problem, called the word
problem for groups, which states that, given generators and relations for a group,
there is no systematic way to tell if two different representations of elements are
the same.
Zn ⊕ Zq1 ⊕ · · · ⊕ Zqk ,
Example 4.7.
Z6 = Z2 ⊕ Z3
If n has a prime factorisation as pk11 pk22 . . . pkmm , then
So abelian groups are much easier to deal with as we are able to classify the
finitely generated ones.
Remark 4.8. If you’re given a (non-abelian) group in terms of generators and
relations, you can often work out what its abelianisation is by just looking at
the relations and “collapsing” them using commutativity. Because the idea is,
informally, that the abelianisation makes everything commute.
Example 4.9 (Fundamental group of surfaces). Let X be an orientable surface
of genus g. Then
Thus
We used van Kampen’s theorem a lot when working out fundamental groups. It
primarily involved two things: disjoint unions and quotients. We already know
how to deal with disjoint unions, we just take direct sums:
Hn (X ∐ Y ) = Hn (X) ⊕ Hn (Y ).
However, we’re not sure what should be true about quotients. Given a space X
with a subspace A we might like to have that:
Hn (X/A) ∼
= H̃n (X)/H̃n (A).
5 What happened to van Kampen’s Theorem? 16
Fortunately this is not true. If it were, we would have a definite problem. Given
any space A we can make the cone on A:
X = A × I/A × {∗}.
Now this is contractible and so H̃n (X) = 0. So if H̃n (A) ⊂ H̃n (X), then we
have H̃n (A) = 0 for all n. Oops.
1. There is a relationship between H̃n (X), H̃n (A) and H̃n (X/A) via an exact
sequence involving all dimensions of homology at the same time.
The first point means that we can do calculations inductively, or implicitly. The
second means that we can sometimes do direct calculations in small examples.
The aim is that given CW-complexes i : A ֒→ X and j : X ֒→ X/A then we can
get an exact sequence:
i∗ j∗
··· H̃n (A) H̃n (X) H̃n (X/A)
β
i∗ j∗
H̃n−1 (A) H̃n−1 (X) H̃n−1 (X/A)
β
Note that if j∗ were surjective then the β would be zero maps. This long exact
sequence measures the failure of the j∗ to be surjective. The slightly complicated
part here is figuring out what the β should be—these maps are sometimes called
the Bockstein homomorphisms.
We first claim that we can make a chain complex of the following form:
δ̄n+1 δ̄n
··· Cn+1 (X)/Cn+1 (A) Cn (X)/Cn (A) Cn−1 (X)/Cn−1 (A) ···
5 What happened to van Kampen’s Theorem? 17
This works since δn (Cn (A)) ⊂ Cn−1 (A). This is called the relative chain com-
plex, the elements of Cn (X)/Cn (A) are called the relative chains and the ho-
mology is called the relative homology, Hn (X, A).
Example 5.1. Let X = I and A = {0, 1}. Then X/A = S 1 and, for all n ≥ 0,
Hn (X, A) ∼
= H̃n (S 1 ).
Now our third claim is that we have a short exact sequence of Cn ’s:
This unravels to a long exact sequence of homology which includes the relative
homology groups and the Bockstein homomorphisms.
Then in nice spaces this gives us the long exact sequence for quotients. In fact
we also have a short exact sequence of chain complexes which we will use after
proving the following lemma.
Lemma 5.2 (Snake Lemma). Consider the following diagram of abelian groups
and group homomorphisms, for which the central diagram is commutative and
the middle two rows are exact.
5 What happened to van Kampen’s Theorem? 18
f g
A B C 0
κ φ ψ
0 A′ B′ C′
f′ g′
A′ / Im κ B ′ / Im φ C ′ / Im ψ
β
ker κ ker φ ker ψ A′ / Im κ B ′ / Im φ C ′ / Im ψ.
Proof. We will construct the map and show that it is well-defined. Exactness
of the sequence is left as an exercise.
The short exact sequence of chain complexes is shown in the following diagram,
where Cn = Bn /An , the i’s are inclusions and the j’s are quotients.
5 What happened to van Kampen’s Theorem? 19
0 0 0
δ|A δ|A
··· An+1 An An−1 ···
i i i
δ δ
··· Bn+1 Bn Bn−1 ···
j j j
δ̄ δ̄
··· Cn+1 Cn Cn−1 ···
0 0 0
Now that we have the snake lemma and the short exact sequence of chain
complexes we can construct the connecting homomorphism
β : Hn (X, A) → Hn−1 A
and
Hn−1 A = ker(δn−1 |A )/ Im (δn |A ).
For clarity in the following explanation we will define, for any group homomor-
phism f : A → B, the cokernel of f as coker f = B/ Im f .
i j
0 Cn A Cn X Cn X/Cn A 0
δn |A δn δ̄n
In this section we will look more closely at the relationship between relative
homology and the homology of quotient spaces.
Theorem 6.1. Given CW-complexes A ⊂ X, then
2. In general, Hn (Y, ∗) ∼
= H̃n (Y ) if Y is a CW-complex.
x1 a x2 b x3 c x4
The chain complexes for X and A, with the obvious maps, as well as the relative
chain complex are then as follows.
The quotient space X/A is the wedge sum of three circles S 1 ∨ S 1 ∨ S 1 which
has the following augmented chain complex.
δ1 ǫ
0 Z⊕3 Z Z 0
The boundary map δ1 is a zero map, so the reduced homology of X/A is then
the same as the relative homology, for all n.
Example 6.3. Let X = D2 and let A be the boundary S 1 . The chain complexes
for X and A, with the obvious maps, as well as the relative chain complex are
as follows.
C(X) 0 Z Z Z 0
C(A) 0 0 Z Z 0
C(X, A) 0 Z 0 0 0
The quotient space X/A is the sphere S 2 , with the following augmented chain
complex.
δ2 δ1 ǫ
0 Z 0 Z Z 0
The reduced homology groups of X/A = S 2 are then the same as those of the
relative homology groups, for all n.
Example 6.4. It was mentioned above that we can use this relative homology
for working out quotients only when the spaces are somehow “nice enough”.
The Hawaiian earring (HE) immediately springs to mind as a “non-nice” space.
The Hawaiian earring is not a CW-complex. Compare it with the infinite wedge
of circles S 1 ∨ S 1 ∨ . . . ∨ S 1 ∨ . . . that all have the same radius and which is a
CW-complex.
A way to build the Hawaiian earring is to start with the interval, so let X = I,
and take the subspace of points A = { n1 | n ∈ N ∪ {0}}. The quotient space is
then just X/A. (Draw it!) The chain complexes for X and A, as well as the
relative chain complex are as follows.
C(X) 0 ZN ZN 0
C(A) 0 0 Z N 0
C(X, A) 0 Z N 0 0
We know that Dk is contractible and so H̃n (Dk ) = 0 for all n. Placing these
zeros into the long exact sequence we see that we have many instances of short
exact sequences which in turn means that the Bockstein homomorphisms, β, are
all isomorphism. I.e. H̃n (S k ) ∼
= H̃n (Dk , S k−1 ) ∼
= H̃n−1 (S k − 1) for all n > 0.
Also note that S 0 is a pair of points but otherwise S k is path connected for
k > 0. Thus the 0th homology group of the spheres is given by
Z if k = 0
H̃0 (S k )
0 if k 6= 0
The remaining piece to find is the homology groups of S 0 which are plainly
given by
Z if n = 0,
H̃n (S 0 ) =
0 if n 6= 0.
φ ψ
0 Cn (A ∩ B) Cn (A) ⊕ Cn (B) Cn (A + B) 0
x i(x) − j(x) a a
b b
Using the same techniques as before, this unravels to a long exact sequence of
homology.
Exercise 6.6. Let X = S n and take A and B to be the north and south hemi-
spheres, Dn , of X. Use the Mayer-Vietoris sequence to work out the homology
of X.
We will now look at the general idea of homology and what we expect it to do.
Formally, homology is a functor, for all n, from the category of pairs of spaces
(with appropriate maps) to Ab taking a pair (X, A) of spaces to a homology
group Hn (X, A). This should be homotopy invariant together with, for all n, a
map
δn : Hn (X, A) → Hn−1 (A, ∅)
such that for all maps f : (X, A) → (Y, B), the following square commutes.
Hn f
Hn (X, A) Hn (Y, B)
This in fact constitutes what is called a natural transformation. All of the above
is also required to satisfy the following axioms.
8 Further remarks 25
δ
··· Hn (A) Hn (X) Hn (X, A) Hn−1 (A) ···
Axiom 7.3 (Excision). If X is covered by the interiors of A and B then the in-
clusion (B, A∩B) ֒→ (X, A) induces an isomorphism Hn (B, A∩B)−→H ˜ n (X, A).
Axiom 7.4 (Additivity). If (X, A) = ∐i (Xi , Ai ) then there are inclusions
(Xi , Ai ) ֒→ (X, A) inducing an isomorphism
⊕i Hn (Xi , Ai )−→H
˜ n (X, A).
8 Further remarks
This is called the Moore space M (G, n). The version for homotopy groups is
denoted K(G, n), the Eilenberg-Mac Lane spaces.
as required.
8.3 Suspension
Given a space X we can form the suspension of X, denoted SX, given by the
quotient SX = (X × I)/ ∼ where the equivalence relation ∼ is the identification
(x1 , 0) ∼ (x2 , 0) and (x1 , 1) ∼ (x2 , 1). If we take X = S 1 we can picture this
attaching a cone above and below S 1 .
The homology of SX then moves up one dimension. I.e. for all n, H̃n (X) =
H̃n+1 (SX). Thus we can make a space Y with H̃1 (Y ) = Z/k, H̃2 (SY ) = Z/k,
H̃3 (SSY ) = Z/k, etc. Given groups G0 , G1 , G2 , . . ., we can make a CW-complex
X with H̃i (X) = Gi , for all i. Then we can make a space
X = ∨i M (Gi , i).