Algebraic Topology by Tim Perutz

ALGEBRAIC TOPOLOGY I: FALL 2008
TIM PERUTZ
Contents
I. The fundamental group
1. Introduction
1.1. Homotopy equivalence
1.2. The fundamental groupoid
2. The fundamental group of the circle
2.1. Trivial loops
2.2. Computing 1 (S 1 )
2.3. Applications
3. Van Kampen in theory
3.1. Group presentations
3.2. Push-outs
3.3. Van Kampens theorem
4. Van Kampen in practice
4.1. Fundamental groups of spheres
4.2. A useful lemma
4.3. Fundamental groups of compact surfaces
4.4. The complement of a trefoil knot
5. Covering spaces
5.1. Deck transformations
5.2. Examples
5.3. Unique path lifting
6. Classifying covering spaces
6.1. An equivalence of categories
6.2. Existence of a simply connected covering space
II. Singular homology theory
7. Singular homology
7.1. The definition
7.2. The zeroth homology group
7.3. The first homology group
8. Simplicial complexes and singular homology
8.1. -complexes
8.2. The Hurewicz map revisited
9. Homological algebra
9.1. Exact sequences
9.2. Chain complexes
10. Homotopy invariance of singular homology
11. The locality property of singular chains
12. MayerVietoris and the homology of spheres
1
3
3
3
4
6
6
6
7
9
9
9
11
13
13
13
13
15
17
18
18
19
21
21
23
25
25
25
26
26
28
28
29
30
30
31
34
37
39
TIM PERUTZ
12.1. The MayerVietoris sequence

12.2. Degree
13. Relative homology and excision
13.1. Relative homology
13.2. Suspension
13.3. Summary of the properties of relative homology
14. Vanishing theorems for homology of manifolds
14.1. Local homology
14.2. Homology in dimension n
15. Orientations and fundamental classes
15.1. Homology with coefficients
15.2. What its good for
15.3. The local homology cover
15.4. Orientations
15.5. Fundamental classes
16. Universal coefficients
16.1. Homology with coefficients
16.2. Tor
16.3. Universal coefficients
III. Cellular homology
17. CW complexes
17.1. Compact generation
17.2. Degree matrices
17.3. Cellular approximation
18. Cellular homology
19. Cellular homology calculations
19.1. Calculations
20. The EilenbergSteenrod axioms
IV. Product structures
21. Cohomology
21.1. Ext
22. Product structures, formally
22.1. The evaluation pairing
22.2. The cup product
22.3. The cap product
23. Formal computations in cohomology
23.1. The K
unneth formula
23.2. An algebraic application of cup product
24. Cup products defined
24.1. The basic mechanism
24.2. Cup products in cellular cohomology
24.3. Cup products in singular cohomology
25. Non-commutativity
26. Poincare duality
39
41
42
42
43
44
45
46
46
49
49
49
49
50
50
53
53
53
55
57
57
59
59
59
61
64
64
67
70
70
71
74
74
74
75
77
78
78
80
80
80
81
83
84
I. The fundamental group

1. Introduction
We explain that algebraic topology aims to distinguish homotopy types. We introduce the fundamental groupoid and the fundamental group.
1.1. Homotopy equivalence.
1.1.1. A topological space is a set X equipped with a distinguished collection of
subsets, called open. The collection must be closed under finite intersections and
arbitrary unions. In particular, it includes the empty union , and the empty
intersection X.
A map X Y between topological spaces is continuous if the preimage of
every open set in Y is open in X. A homeomorphism is a continuous map with a
continuous two-sided inverse.
Convention: In this course, map will mean continuous map.
1.1.2. Elementary properties of spaces that are preserved by homeomorphism (e.g.
the Hausdorff property, compactness, connectedness, path-connectedness) allow us
to distinguish some spaces. For instance, the interval [0, 1] is not homeomorphic
to the circle S 1 = R/Z because [0, 1] \ {1/2}Wis disconnected, whilst S 1 \ {x} is
n
connected for any x S 1 . The spaces Xn = i=1 S 1 (the wedge product, or one1
point union, of n copies of S ) are all distinct, because it is possible to delete n,
but not n + 1, distinct points of Xn without disconnecting it.
However, if we thickened the circles in Xn to ribbons, making a space Yn , the
argument would fail. In algebraic topology, one looks for invariants of spaces which
are insensitive to such thickenings, so that if they distinguish the Xn they also
distinguish the Yn .
Definition 1.1. If f0 , f1 : X Y are maps, a homotopy from X Y is a map
F : [0, 1]X Y such that F it = ft for t {0, 1}, where it (x) = (t, x) [0, 1]X.
We often think of F as a path {ft }t[0,1] of maps ft : X Y .
Homotopy defines an equivalence relation on the set of maps f : X Y , which
we denote by the symbol '.
Definition 1.2. A homotopy equivalence is a map f : X Y such that there exists
g : Y X which is an inverse up to homotopy. That is, f g ' idY and g f ' idX .
Exercise 1.1: Homotopy equivalence defines an equivalence relation on spaces.
The equivalence classes are called homotopy types. Algebraic topology provides
a collection of invariants of homotopy types. The principal invariants are the fundamental group and the homology groups, and the homomorphisms between these
groups associated with maps between spaces.
Exercise 1.2: The following equivalent conditions define what is means for a non-empty
space X to be contractible. Check their equivalence.
X is homotopy equivalent to a one-point space.
For every x X, the inclusion {x} X is a homotopy equivalence.
For some x X, the inclusion {x} X is a homotopy equivalence.
For some x X, the constant map cx : X X at x is homotopic to idX .
Exercise 1.3: Any convex subset of Rn is contractible.
TIM PERUTZ
Convex subsets of Rn are contractible for a particular reason: their points are
deformation retracts. In general, if X is a space and i : A X the inclusion of a
subspace, we say that A is a deformation retract of X if there is a map r : X A
such that r i = idA and i r ' idX by a homotopy {ht } so that (in addition to
h0 = i r and h1 = idX ) one has ht (a) = a for all t and a A. Such a map r,
called a deformation retraction, is obviously a homotopy equivalence.
Exercise 1.4: Show carefully that the letter A, considered as a union of closed line
segments in R2 , is homotopy equivalent but not homeomorphic to the letter O. Show
briefly that all but one of the capital letters of the alphabet is either contractible or
deformation-retracts to a subspace homeomorphic to O. Show that the letters fall into
exactly three homotopy types. How many homeomorphism types are there? (View a
letter as a finite union of the images of paths [0, 1] R2 . Choose a typeface!)
Exercise 1.5: Let {X }A be a collection of spaces indexed
by a set A. Let x X
W
be basepoints. Define `
the wedge sum (or 1-point union) A X as the quotient space
of the disjoint union X by the equivalence relation x x for all , A.
Show carefully that, for n 1, the complement of p distinct points in Rn is homotopyequivalent to the wedge sum of p copies of the sphere S n1 = {x Rn : |x| = 1}.
Remark. Lets look ahead. Theorems of Hurewicz and J. H. C. Whitehead imply
that, among all spaces which are cell complexes, the sphere S n = {x Rn+1 :
|x| = 1}, with n > 1, is characterized up to homotopy equivalence by its homology
/ {0, 1} and its trivial fundamental
groups H0 (S n )
= Z, Hi (S n ) = 0 for i
= Hn (S n )
group. In general, distinct homotopy types can have trivial fundamental groups and
isomorphic homology groups (e.g. S 2 S 2 , CP 2 #CP 2 ). Another invariant, the
cohomology ring, distinguishes these two examples. When it fails to distinguish
spaces, one localizes the problem and works over Q and mod primes p. Over Q, a
certain commutative differential graded algebra gives a new invariant [D. Sullivan,
Infinitesimal computations in topology, Publ. Math. I.H.E.S. (1977)]. Mod p, one
considers the Steenrod operations on cohomology. There is an algebraic structure
which captures all this at once, and gives a complete invariant for the homotopy
type of cell complexes with trivial fundamental group [M. Mandell, Cochains and
homotopy type, Publ. Math. I.H.E.S. (2006)].
1.2. The fundamental groupoid.
1.2.1. Our first invariants of homotopy type are the fundamental groupoid and the
isomorphism class of the fundamental group.
A path in a space X is a map f : I X, where I = [0, 1]. Two paths f0 and f1
are homotopic rel endpoints if there is a homotopy {ft }t[0,1] between them such
that ft (0) and ft (1) are both independent of t. Write for the equivalence relation
of homotopy rel endpoints.
Two paths f and g are composable if f (1) = g(0). In this case, their composite
f g is the result of traversing first f , then g, both at double speed: (f g)(t) = f (2t)
for t [0, 1/2] and (f g)(t) = g(2t 1) for t [1/2, 1].
The composition operation is not associative: (f g) h 6= f (g h). What is
true, however, is that (f g) h f (g h). (Proof by picture.)
If f is a path, let f 1 denote the reversed path: f 1 (t) = f (1 t). One has
f f 1 cf (0) and f 1 f cf (1) , where cx denotes the constant path at x.
(Picture.) Moreover, cf (0) f ' f and f cf (1) ' f .
We now define a category 1 (X), the fundamental groupoid of X. A category

consists of a collection (for instance, a set) of objects, and for any pair of objects
(x, y), a set Mor(x, y) of morphisms (or maps) from x to y. Also given is an
associative composition rule
Mor(x, y) Mor(y, z) Mor(x, z).
Each set Mor(x, x) must contain an identity ex (meaning that composition with ex
on the left or right does nothing).
The objects of the category 1 (X) are the points of X. Define 1 (x, y) as the
set of equivalence classes of paths from x to y under the relation of homotopy rel
endpoints. 1 (x, y) will be the morphism set Mor(x, y) in the category. One has
well-defined composition maps 1 (x, y)1 (y, z) 1 (x, z), which are associative
by our discussion. The class [cx ] of the constant path at x defines an identity element
ex for 1 (x, x). This shows that 1 (X) is a category.
A category in which every morphism has a 2-sided inverse is called a groupoid.
Every morphism [f ] 1 (x, y) has a 2-sided inverse [f 1 ] 1 (y, x).
1.2.2. Groupoids are too complicated to be really useful as invariants. However,
as with any groupoid, the sets 1 (x, x) form groups under composition, and we
can use this to extract a practical invariant. When a basepoint x X is fixed,
1 (X, x) := 1 (x, x) is called the fundamental group. It is the group of based
homotopy classes of loops based at x.
If X is path connected, the fundamental groups for different basepoints are
all isomorphic. Indeed, if f is a path from x to y then the map
1 (X, x) 1 (Y, y),
[] 7 [f ] [] [f 1 ]
is an isomorphism.
If F : X Y is a map, there is an induced homomorphism
F : 1 (X, x) 1 (Y, F (x)),
[f ] 7 [F f ].
If G : Y Z is another map, one clearly has G F = (G F ) .

Maps F0 and F1 which are based homotopic (i.e. homotopic through maps
Ft with Ft (x) constant for all t) give the same homomorphism 1 (X, x)
1 (Y, F0 (x)).
Exercise 1.6:
(a) If f0 and f1 are loops (I, I) (X, x), we say they are homotopic through loops if they are joined by a homotopy ft with ft (0) equal to
ft (1) but not necessarily to x. Show that f0 is homotopic to f1 through loops
iff [f0 ] is conjugate to [f1 ] in 1 (X, x).
(b) Show that a homotopy equivalence between path connected space induces an
isomorphism on 1 , regardless of the choices of basepoints.
A point clearly has trivial 1 (theres only one map I ). By (b) from the
exercise, 1 (X, x) = {1} for any contractible space X and any x X.
A space is called simply connected if it is path-connected and has trivial 1 . We
have just seen that contractible spaces are simply connected.
Exercise 1.7: (*) Prove directly that the 2-sphere S 2 = {x R3 : |x| = 1} is simply
connected.
TIM PERUTZ
2. The fundamental group of the circle

Our first calculation of a non-trivial fundamental group has already has remarkable consequences.
2.1. Trivial loops. We begin by interpreting what it means for a loop to be trivial
in the fundamental group. It is convenient to regard a loop not as a map f : I X
with f (1) = f (0) but as a map from the unit circle S 1 = D2 C into X.
Proposition 2.1. A loop f : S 1 X represents the identity element e 1 (X, f (1))
if and only if it extends to a map from the closed unit disc D2 into X.
Thus a simply connected space is a path-connected space in which every loop
bounds a disc.
Proof. If [f ] = 1 1 (X, f (1)), let {ft }t[0,1] be a homotopy rel endpoints from
the constant map cf (1) to f = f1 . Define a continuous extension F : D2 X of f
by setting F (z) = f|z| (z/|z|) if z 6= 0 and F (0) = f (1).
Conversely, if f extends to F : D2 X, define a map I S 1 X, (t, z) 7 F (tz).
Then F is a homotopy from the constant map cF (0) to f . The latter is in turn is
homotopic through constant maps to cf (1) . Hence f is homotopic through loops to
cf (1) . By (a) from Exercise 1.6, [f ] is conjugate to [cf (1) ]. But [cf (1) ] = e, hence
[f ] = e.

2.2. Computing 1 (S 1 ).
Theorem 2.2. The fundamental group of S 1 is infinite cyclic: there is a (unique)
homomorphism deg : 1 (S 1 )
= Z such that deg(idS 1 ) = 1.
We think of S 1 as R/Z, and take [0] as basepoint. Note that two maps S 1 S 1
taking [0] to [0] are homotopic through loops iff they represent conjugate elements
in 1 (S 1 , [0]). By the theorem, 1 is abelian, so conjugate elements are actually
equal. Hence deg is actually an invariant of homotopy through loops, indeed a
complete invariant.
The key idea of the proof is to look at the quotient map p : R R/Z = S 1 . This
map is the prototypical example of a covering map.
Lemma 2.3. Every map f : (I, I) (S 1 , [0]) lifts uniquely to a map f: I R
such that (i) f(0) = 0, and (ii) p f = f .
Proof. Let T be the set of t I such that f exists and is unique on [0, t]. For any
[x] = p(x) S 1 , the open set U[x] = p(x 1/4, x + 1/4) S 1 contains [x] and
has the following property: the preimage p1 (U ) is the disjoint union of open sets
Vxn := (n+x1/4, n+x1/4), n Z. Moreover, p maps each Vxn homeomorphically
onto U .
If f has been defined on [0, t], with t < 1, there exists > 0 so that f (t, t+)
Uf (t) . Since f(t) Vf0(t) , we are forced to define f on [t, t + ) as the composite
f
p1
[t + ) Uf (t) Vf0(t) .
This does indeed define an extension of f to [0, t + ). So T is an open set.
Now suppose f exists and is unique on [0, t). Since f (s) f (t) as s t,
when 0 < t s 1 the lifts f(s) must lie in one of the open sets V projecting
homeomorphically to Uf (t) , independent of s. Thus we can define f(t) to be the

preimage of f (t) that lies in V , and this defines the unique continuous lift of f on
[0, t]. Hence T is closed. Since I is connected and T non-empty, we have T = I.

Proof of the theorem. Given f : (I, I) (S 1 , [0]), construct f as in the lemma.
Since p f(1) = [0], f(1) is an integer. Define the degree deg(f ) to be this integer.
Since f was uniquely determined by f , deg(f ) is well-defined. We now observe that
if {ft }t[0,1] is a based homotopy then deg(f0 ) = deg(f1 ). Indeed, we can lift each
ft to a unique map ft : I R, p(ft (0)) = [0], and p ft = ft . It is easy to check
that the ft vary continuously in t, hence define a homotopy {ft } from f0 to f1 .
Thus deg(ft ) = ft (1) is a continuous Z-valued function, hence constant.
Thus deg defines a map 1 (S 1 ) Z. It is a homomorphism because fg
g is
given on [0, 1/2] by the unique lift of t 7 f (2t) which begins at 0 (this ends at
deg(f )), and on [1/2, 1] by the unique lift of t 7 f (2t 1) which begins at deg(f )
(this ends at deg(g) + deg(f )).
The degree homomorphism is surjective because deg(idS 1 ) = 1. To see that it
is injective, suppose deg f = 0. Then f is a loop in R, based at 0. Since R is
simply connected, f is based-homotopic to the constant map, and applying p to
this homotopy we see that the same is true of f .

2.3. Applications.
Corollary 2.4 (The fundamental theorem of algebra). Every non-constant polynomial p(z) C[z] has a complex root.
Proof. We may assume p is monic. If p(z) = z n + cn1 z n1 + + c0 has no root,
p(z)/|p(z)| is a well-defined function C S 1 C. Let f denote its restriction to
the circle {|z| = 1}. Now, f extends to a map from the unit disc to S 1 , whence f
is null-homotopic (cf. the last lecture) so deg(f ) = 0 by the homotopy-invariance
of degree.
Now define ft : S 1 S 1 for t > 1 by ft (z) = p(tz)/|p(tz)|. The ft are all
homotopic, and f1 = f , so deg(ft ) = 0 for all t. But for |z| 0, |cn1 z n1 + +
c0 | < |z n |, and hence ps (z) := z n + s(cn1 z n + + c0 ) has no root for 0 s 1.
Thus, for some fixed t 0, we can define gs : S 1 S 1 by gs (z) = ps (tz)/|ps (tz)|,
and this defines a homotopy from ft = g1 to g0 . But g0 (z) = z n /|z n |, and so
deg g0 = n (check this!). Hence n = 0.

Remark. Some proofs of FTA invoke Cauchys theorem from complex analysis. To
make the link with our approach, note that if : S 1 C is a loop then, by the
residue theorem (a consequence of Cauchys theorem) the complex number
Z
1
d() =
z 1 dz
2i
is actually an integer depending on only through its homotopy class in C . When
is a based loop S 1 S 1 C , d() = deg() (this follows from our theorem,
bearing in mind that d defines a homomorphism d : 1 (S 1 ) Z and that d(idS 1 ) =
1).
Another corollary is the Brouwer fixed point theorem.
Corollary 2.5. Every continuous map g : D2 D2 has a fixed point.
TIM PERUTZ
(Here D2 denotes the closed unit disc.)

Proof. Suppose g has no fixed point. Then, for any x D2 , there is a unique
line passing through x and g(x). Define r(x) S 1 to be the point where this line
hits S 1 = D2 when one starts at g(x) and moves along the line towards x. Thus
r(x) = x when x D2 . Writing r(x) = x + t(g(x) x), one calculates from the
requirements that |r(x)| = 1 and t 0 that
p
hx, x g(x)i hx, x g(x)i2 (|x|2 1)|x g(x)|2
.
t=
|x g(x)|2
Thus r is continuous.
On the other hand, there can be no continuous r : D2 D2 with r|D2 = id,
for if such an r existed, the degree of its restriction r0 to the boundary would be 1
(because r0 = id) but also 0 (because r0 extends over D2 ). Hence there must be a
fixed point.

Remark. The Brouwer fixed point theorem holds in higher dimensions too: every
continuous map g : Dn Dn has a fixed point. One can attempt to prove it using
the same argument. For this to work, what one needs is a homotopy-invariant,
integer-valued degree for maps S n1 S n1 . The identity map should have degree
1 and the constant map degree 0. With such a function in place, the same argument
will run.
There are many ways of defining a degree function (actually, the same degree
function): one can use homology theory, homotopy theory, differential topology or
complex analysis.
Exercise 2.1: Show that every matrix A SL2 (R) can be written uniquely as a product
KL with K SO(2) and L lower-triangular with positive diagonal entries. Use this
to write down (i) a deformation-retraction of SL2 (R) (topologized as a subspace of
R4 ) onto its subspace SO(2); and (ii) a homeomorphism S 1 (0, ) R SL2 (R).
Deduce that SL2 (R) is path-connected and that 1 (SL2 (R))
= Z.
Exercise 2.2: The polar decomposition. It is known that every matrix A SL2 (C)
can be written uniquely as a product U P with U SU (2) and P positive-definite
hermitian. Assuming this, deduce a homeomorphism S 3 (0, ) C SL2 (C). (We
will soon see that this implies 1 SL2 (C) = {1}.)
3. Van Kampen in theory

There are two basic methods for computing fundamental groups. One, the method
of covering spaces, generalises our proof that 1 (S 1 ) = Z. The other, which we shall
discuss today, is to cut the space into simpler pieces and use a locality property of
1 called van Kampens theorem (a.k.a. the Seifertvan Kampen theorem).
3.1. Group presentations.
Definition 3.1. A free group on a set S is a group FS equipped with a map
i : S FS enjoying a universal property: for any map f from S to a group G
there is a unique homomorphism f: FS G with f i = f .
If FS and FS0 are both free groups on S, and i : S FS and i0 : S FS0
the defining maps, then there are unique homomorphisms h : FS FS0 such that
h i = i0 and h0 : FS0 FS such that h i0 = i. Thus h0 h i = i. It follows that
h0 h = id, since both sides are homomorphisms FS FS extending i. Hence h
and h0 are inverse isomorphisms.
The free group Fn := F{1,...,n} can be realised as the group of all words made
1 2 7
1
up of letters a1 , . . . , an and their formal inverses a1
1 , . . . , an , e.g. a4 a3 a4 a1 .
1
1
Expressions ai ai and ai ai can be deleted or inserted. The group operation is
1
3
3
3
concatenation of words, e.g. (a4 a1
3 ) (a3 a2 ) = a4 a3 a3 a2 = a4 a2 . The identity
element is the empty word. The map i sends m to am , and given an f : {1, . . . , n}
2
1
G we extend it to f by sending, for example, a2 a1
f (a3 )2 .
1 a3 to f (a2 )f (a1 )
We often write this group as ha1 , . . . , an i. For example, F1 = hai = Z.

Lemma 3.2. The groups Fn , for different n, are all distinct.
Proof. The abelianization (Fn )ab := Fn /[Fn , Fn ] is isomorphic to Zn , and Zn /2Zn
has 2n elements.

Now suppose that r1 , . . . , rm are elements of ha1 , . . . , an i. Let R be the smallest
normal subgroup containing the ri (R is thought of as a group of relations). Define
ha1 , . . . , an | r1 , . . . , rm i = ha1 , . . . , ar i/R.
If G is a group, and g1 , . . . , gn G group elements, theres a unique homomorphism
f : ha1 , . . . , an i G sending each ai to gi . It is surjective iff g1 , . . . , gn generate G.
In this case, G
= ha1 , . . . , an i/ ker f . Thus, if g1 , . . . gn generate G, and r1 , . . . , rm
are elements of ha1 , . . . , an i which generate ker f as a normal subgroup, then f
induces an isomorphism
ha1 , . . . , an | r1 , . . . , rm i G.
Such an isomorphism is called a (finite) presentation for G. As examples, we have(!)
Z/(n)
= ha | an i,
D2n
= ha, b | an , b2 , (ba)2 i,
Z2
= ha, b | aba1 b1 i.
3.2. Push-outs.
Definition 3.3. Consider three groups, G1 , G2 and H, and a pair of homomorphisms
f1
f2
G1 H G2 .
10
TIM PERUTZ
A push-out for (f1 , f2 ) is another group P and a pair of homomorphisms p1 : G1

P and g2 : G2 P forming a commutative square
f1
f2 y
G1
p1
y
p2
G2 P
and satisfying a universal property: given any other such square (a group K and
homomorphisms k1 : G1 K and k2 : G2 K such that k1 f1 = k2 f2 ), there
is a unique homomorphism h : P K such that k1 = h p1 and k2 = h p2 .
Exercise 3.1: Prove that the universal property determines P up to isomorphism. In
what sense is the isomorphism unique?
We can understand push-outs concretely using group presentations. Suppose
G1 = ha1 , . . . , an | r1 , . . . rm i and that G2 = hb1 , . . . , bp | s1 , . . . , sq i. Also suppose
that H has generators h1 , . . . , ho . In the push-out square above, the group P is then
a group called the free product of G1 and G2 amalgamated along H and notated
G1 H G2 . It has the presentation
G1 H G2 = ha1 , . . . , an , b1 , . . . , bp | r1 , . . . rm , s1 , . . . , sq , c1 , . . . , co i,
where ci = f1 (hi )f2 (hi )1 .
Exercise 3.2: Check that K = G1 H G2 fits into a push-out square for f1 and f2 .
Note that there was no need for our group presentations to be finite, except for
notational convenience: we can allow infinite sets of generators and relations.
Exercise 3.3: The free product G1 G2 of groups G1 and G2 is the push-out of the
diagram G1 {1} G2 . Define D as the subgroup of the group of affine transformations R R generated by x 7 x and x 7 x+1. Prove that D
= (Z/2)(Z/2).
Exercise 3.4: In this exercise we show that the modular group, P SL2 (Z) = SL2 (Z)/{I},
is the free product (Z/2) (Z/3). Define three elements of SL2 (Z),

0 1
1 1
0 1
S=
, T =
, U = ST =
.
1 0
0
1
1 1
(a) Verify that S 2 = U 3 = I.
(b)
Show that,
for any A SL2 (Z), there is an n Z such that the matrix
a b
= AT n has c = 0 or |d| |c|/2.
c d
(c) Explain how to find an integer l 0 and a sequence of integers n1 , . . . , nl
such that either AT n1 ST n2 S . . . ST l or AT n1 ST n2 S . . . ST l S has 0 as its
lower-left entry.
(d) Show that S and T generate SL2 (Z).
(e)* Define : ha, b | a2 , b3 i = (Z/2) (Z/3) P SL2 (Z) to be the unique
homomorphism such that (a) = S and (b) = U . Remind yourself
how P SL2 (R) acts on the upper half-plane H C by Mobius maps. Take
1 6= w (Z/2) (Z/3). Prove that the Mobius map w corresponding to
(w) P SL2 (R) has the property that w (D) D = , where
D = {z H : 0 < Re z < 1/2, |z 1| > 1}.
[Hint: consider A := {z H : Re z > 0} and B := {z H : |z 1| >
max(1, |z|)}.] Deduce that is an isomorphism.
11
3.3. Van Kampens theorem.

Theorem 3.4. Suppose that X is the union of two path-connected open subsets
U and V with path-connected intersection U V . Take x U V . Then the
commutative diagram
1 (U V, x) 1 (U, x)
y
y
1 (V, x) 1 (X, x)
of maps induced by the inclusions is a push-out square.
Example 3.5. Let Cn be the complementWof n points in the plane. Observe that
n
Cn deformation-retracts
to the wedge sum i=1 S 1 . We have 1 (Cn )
= Fn . Indeed,
Wn
1
when n > 0, i=1 S is the union of a subspace U which deformation-retracts to
Wn1 1
1
i=1 S , and a subspace V which deformation-retracts to S , where the subspace
U V is contractible. By induction, 1 (U )
F
.
We
know
1 (V )
= n1
= Z = F1 .
The push-out of Fn1 and Z along the trivial group H is Fn1 F1
F
= n . Thus the
result follows from van Kampens theorem.
Lemma 3.6. For any loop : (I, I) (X, x), there exists a finite, strictly increasing sequence 0 = s0 < s1 < s2 < < sn = 1 such that maps each interval
[si , si+1 ] into U or into V .
Proof. Every x I has a connected open neighbourhood whose closure maps to U
or to V . Since I is compact, finitely many of these intervals cover I. The endpoints
of the intervals in the finite cover form a finite subset of I, which we may enumerate
in ascending order as (s0 , . . . , sn ).

Let us call the sequence (si ) a subdivison for .
Lemma 3.7. Suppose = {t }t[0,1] is a homotopy of paths (I, I) (X, x).
Then there are increasing sequences 0 = t0 < t1 < < tm = 1, and 0 = s0 <
< sn = 1, such that maps each rectangle [ti , ti+1 ] [sj , sj+1 ] into U or into
V . Moreover, we can take the sequence (si ) to refine given subdivisions of 0 and
1 .
Exercise 3.5: Prove the lemma.
Proof of Van Kampens theorem. Suppose we are given a group G and homomorphisms f : 1 (U ) G, g : 1 (V ) G which agree on the images of 1 (U V ). We
construct a map : 1 (X) G so that f = (iU ) and g = (iV ) , where
iU : U X and iV : V X are the inclusions.
Take : (I, I) (X, x), and choose a subdivision s0 < < sn . Label the
intervals [si , si+1 ] as red or blue, in such a way that maps red intervals to U
and blue intervals to V . For 0 < i < n, connect (si ) to x by a path i inside
U (if both adjacent intervals [si1 , si ] and [si , si+1 ] are red), inside V (if both
adjacent intervals are blue), or inside U V (if the adjacent intervals are different
colours). Then i := i1 |[si ,si+1 ] i is a loop in either U or V . Define [] =
1 [0 ] n1 [n1 ], where i is either f or g according to whether [si , si+1 ] is
red or blue.
We need to see that is well-defined, and does not depend on the choices of
path, subdivision and colouring. Observe that for a fixed and fixed subdivision,
12
TIM PERUTZ
changing the colouring does not affect , because f and g agree on the image of
1 (U V ). Moreover, refining a subdivision for given does not affect the definition
of . Nor does changing the choice of a path i (instead of trying to replace i by a
rival path i0 , insert an extra point into the subdivision, and use both paths i and
i0 ).
Hence we are left with considering homotopic paths 0 and 1 with a common
subdivision s0 < < sn .
Given a homotopy = {t }, we can subdivide [0, 1] [0, 1] into rectangles
Rij = [ti , ti+1 ] [sj , sj+1 ] and color the Rij as red or blue in such a way so that
maps the red rectangles to U and the blue ones to V . It will suffice to show that
0 and t1 give the same definition for .
This last part of the argument requires pictures, which I will draw in class.
(Consult Hatcher if you need to.) The idea is this: rather than going along the
bottom edge of R0j we can go around the other three sides. We have 0 ' 0 n ,
but by going round these three sides we can replace i by a new loop i0 , and this
will not affect . By eliminating backtracking we can get from 0 n to t1 ,
again without affecting .
Knowing it is well-defined, one can check that is a homomorphism making the
two triangles commute (do so!). Note also that it is the unique such homomorphism:
since is homotopic to the composite of the i |[si ,si+1 ] i1 , we have no choice
but to define this way. This concludes the proof.
13
4. Van Kampen in practice

We compute some fundamental groups using van Kampens theorem.
4.1. Fundamental groups of spheres. A first use of van Kampens theorem is
to show that spaces that should be simply connected are simply connected.
Proposition 4.1. Let S n = {x Rn+1 : |x| = 1} be the n-sphere. When n 2,
1 (S n ) is trivial.
Proof. Notice that the subspace U = {x = (x0 , . . . , xn ) S n : x0 6= 1} is homeomorphic to Rn . Similarly, V := {x = (x0 , . . . , xn ) S n : x0 6= 1} is homeomorphic to Rn . Thus U and V are contractible open sets, and their intersection is
path connected: it deformation-retracts to the equator {x0 = 0}
= S n1 , which is
n
path connected when n 1 > 0. By van Kampen, 1 (S ) is the push-out of two
homomorphisms to the trivial group; it is therefore trivial.

4.2. A useful lemma.
Lemma 4.2. Suppose
f
p
y
{1} P
is a pushout square. Then p is surjective, and its kernel is the normalizer of im f .
Proof. Put P 0 = G/N , where N is the normalizer of im f , and define p0 : G P 0
to be the quotient map. It is easy to check that P 0 and p0 fit into a push-out square
for the homomorphisms f : H G and H {1}. Thus P is isomorphic to P 0 so
that p is identified with p0 .

In conjunction with van Kampens theorem, this lemma has the following consequence.
Proposition 4.3. Suppose that X is the union of a path-connected open set U and
a simply connected open set V , with U V path-connected. Let x U V . Then
1 (X, x) is generated by loops in U . A based loop in U becomes trivial in 1 (X) iff
it lies in the normal subgroup of 1 (U, x) generated by loops in U V .
4.3. Fundamental groups of compact surfaces.
Proposition 4.4. Let T 2 be the 2-torus, RP 2 the real projective plane, K 2 the
Klein bottle. Then
1 (T 2 )
= Z2 ; 1 (RP 2 )
= Z/2; 1 (K 2 )
= ha, b | aba1 bi.
No two of these spaces are homotopy-equivalent.
Proof. These spaces X are all quotient spaces q : I 2 X of the square I 2 R2 ,
obtained by gluing together its sides in pairs. Take p in int(I 2 ).
Let U = q(I 2 \ {p}), and V = q(D) with D a small open disc containing p.
Thus U V deformation-retracts to a circle and V is simply connected. By the last
proposition, 1 (X) is generated by loops in the subspace U , which deformationretracts to q(I 2 ).
Going anticlockwise round I 2 , we label the sides as s1 , s2 , s3 , s4 (as directed
paths).
14
TIM PERUTZ
1
In T 2 , q(s1 ) = q(s1
3 ) and q(s2 ) = q(s4 ). Thus U deformation-retracts to
a wedge of two circles a = q(s1 ) and b = q(s2 ), and U ' s1 s2 s3 s4 '
a b a1 b1 . To apply van Kampen, note that 1 (U V ) = Z and 1 (U ) = ha, bi.
The homomorphism Z F2 induced by U V , U sends 1 to aba1 b1 . Thus,
by the last proposition,
1 (T 2 )
= ha, b | aba1 b1 i
= Z2 .
In K 2 , q(s1 ) = q(s3 ) and q(s2 ) = q(s1

4 ). The argument is just the same as for the
torus, except that now the homomorphism Z F2 sends 1 to aba1 b. Thus
1 (K 2 )
= ha, b | aba1 bi.
In RP 2 , q(s1 ) = q(s3 ) and q(s2 ) = q(s4 ). Thus q(I 2 ) is a single circle, and the
map q : I 2 1(I 2 ) has degree 2. So 1 (U ) = Z and 1 (U V ) = Z. The map
1 (U V ) 1 (U ) corresponds to x 7 2x as a map Z Z. Hence
1 (RP 2 )
= Z/2.
It follows easily that these three spaces are homotopically inequivalent: the abelianized fundamental groups (in which everything commutes) are 1 (T 2 )ab
= Z2 ,
2 ab
2 ab
1 (RP ) = Z/2 and 1 (K ) = Z/2 Z.

As part of the last proposition, we showed 1 (T 2 ) = Z2 . We now compute 1
for a torus with n punctures.
Lemma 4.5. Let p1 , . . . , pn be distinct points of T 2 . There are isomorphisms
n : 1 (T 2 \ {p1 , . . . , pn }) Gn := h1 , . . . , n , a, b | aba1 b1 (1 n )1 i
so that filling in pn induces the following commutative diagram:
1 (T 2 \ {p1 , . . . , pn }) 1 (T 2 \ {p1 , . . . , pn1 })
n y
y n1
Gn
gn
Gn1 ,
where gn (n ) = 1, gn (i ) = i for i < n, gn (a) = a and gn (b) = b.

Proof. Apply van Kampen to a decomposition of T 2 \ {p1 , . . . , pn } into a oncepunctured torus U and an (n + 1)-punctured 2-sphere.

Proposition 4.6. Let g be the closed orientable surface of genus g. Then
ha1 , b1 , . . . , ag , bg | [a1 , b1 ] [ag , bg ]i,
1 (g ) =
where [a, b] := aba1 b1 . If p1 , . . . , pn are distinct points in g then
1 (g \ {p1 , . . . , pn })
= ha1 , b1 , . . . , ag , bg , 1 , . . . , n | [a1 , b1 ] [ag , bg ] = 1 n i.
Proof. By induction on g. We have already proved it for g = 0 and for g = 1.
Decompose g \ {p1 , . . . , pn } as the union of U ' g \ {p1 , . . . , pn , q} and V '
T 2 \ {q 0 } along an annulus U V (wrapping round q in U and around q 0 in V ). By
induction on g, van Kampen, and the last lemma, we find that 1 (g \{p1 , . . . , pn })
has generators
a1 , b1 , . . . , ag1 , bg1 , 1 , . . . , n ,
coming from U ;
ag , bg , n+1 , 0
15
coming from V ; a relation = 0 from U V ; and relations

[a1 , b1 ] [ag1 , bg1 ] = 1 n ,
[ag , bg ] = 01 n+1
from U and V . It is easy to check that this system of generators and relations are
equivalent to those given.

Another standard way to prove this is to think of g as an identification-space
of the 4g-gon.
4.4. The complement of a trefoil knot. The left-handed trefoil knot K is the
image of the embedding f : S 1 S 3 = {(z, w) C2 : |z|2 + |w|2 = 1} given by
1
1
f (e2it ) = ( e4it , e6it ).
2
2
3
2 3
Proposition 4.7. 1 (S \ K) = ha, b | a b i.

Proof. We decompose S 3 as the union of two subspaces Y = {(z, w) : |z| |w|}
and Z = {(z, w) : |z| w}. Both are solid tori S 1 D2 , and Y Z is a torus
S 1 S 1 . There results a decomposition S 3 \ K = (Y \ K) (Z \ K). Though the
sets in this decomposition are not open, van Kampen is applicable because we can
thicken up K to a rope R, and then take thin open neighbourhoods of Y \ R and
Z \ R which deformation-retract onto them. Now, Y \ K deformation-retracts to
the core circle |w| = 0, and Z \ K to the core circle |z| = 0, while (Y \ K) (Z \ K)
deformation retracts to a circle K 0 parallel to K inside the torus Y Z. Now K 0
wraps twice around the core circle in Y \ K, three times around that in Z \ K. Van
Kampen shows that 1 (S 3 \ K) is a push-out of the diagram
2
Z Z Z,
and this gives the presentation claimed.
Exercise 4.1: An n-dimensional manifold is a Hausdorff space X covered by open sets

homeomorphic to Rn . Let X1 and X2 be connected n-dimensional manifolds. A
connected sum X1 #X2 is constructed by choosing embeddings i1 : Dn X1 and
i2 : Dn X2 of the closed n-disc Dn , and letting
X1 #X2 = (X1 \ i1 (int D0 )) q (X2 \ i2 (int D0 ))/ ,
D0 = 21 Dn Dn , where identifies i1 (x) with i2 (x) for all x S n1 = D0 .
(a) Prove that if n > 2 then 1 (X1 #X2 )
= 1 (X1 ) 1 (X2 ).
(b) Let X be an iterated connected sum of r copies of S 1 S n1 , where n 3.
Compute 1 (X).
(c)* Given a finitely presented group G = hg1 , . . . , gk | r1 , . . . , rl i, find a connected,
compact, 4-dimensional manifold M with 1 (M )
= G. [Hint: Start with the
case of no relations. Use the fact that (S 1 D3 ) = S 1 S 2 = (D2 S 2 ).]
Exercise 4.2: Let K be the trefoil knot. Weve seen that 1 (S 3 \ K) = ha, b | a2 = b3 i.
How do you find a word representing a given loop in S 3 \ K? Find words representing a
meridian for K (i.e., the boundary of a small normal disc) and a longitude (parallel to
the knot; not unique!). Let Z be the the kernel of the homomorphism 1 (S 3 \ K)
ha, b | a2 , b3 i which sends a to a and b to b. Show that Z
= Z, generated by a
longitude, and that Z is contained in the center of 1 (S 3 \ K). [Interestingly, by an
earlier exercise we have ha, b | a2 , b3 i
= P SL2 (Z).]
16
TIM PERUTZ
Exercise 4.3: The braid group on 3 strings. In this extended exercise (based on one in
Serres book Trees) well see that the following five groups are isomorphic:
1 (S 3 \ K), where K is a (left-handed) trefoil knot.
The group ha, b | a2 = b3 i.
The algebraic braid group on 3 strings, hs, t | sts = tsti.
The geometric braid group on 3 strings B3 , defined as the fundamental group
of the configuration space C3 of 3-element subsets of C.
1 (C2 \ C), where C C2 is the cuspidal cubic {(X, Y ) : X 2 = Y 3 }.
(a) We already know that 1 (S 3 \ K)
= ha, b | a2 = b3 i. Show that a 7 sts,
b 7 ts defines an isomorphism
ha, b | a2 = b3 i hs, t | sts = tsti.
(b) Take as basepoint {2, 0, 2} C3 . Define loops and in X3 , (t) =
{1 eit , 1 + eit , 2} and (t) = {2, 1 eit , 1 + eit } for t [0, 1].
Let s = [] and t = [ ] in B3 . Check that sts = tst, so that one has a
homomorphism hs, t | sts = tsti B3 .
(b) C3 is the subspace of Sym3 (C) (the quotient of C3 by the action of the
symmetric group S3 permuting coordinates) where the three points are distinct. Let Sym30 (C) = {{a, b, c} Sym3 (C) : a + b + c = 0}. Show that
Sym3 (C)
= C Sym30 (C). Define a homeomorphism h : Sym30 (C) C2 by
sending {a, b, c} to the point (x, y) such that
(t a)(t b)(t c) t3 + xt + y.
Verify that the points a, b and c are distinct iff 4x3 + 27y 2 6= 0. Deduce that
C3
= C (C2 \ C), hence that B3
= 1 (C2 \ C).
2
(d) Show that C \ C is homotopy-equivalent to S 3 \ K, whence 1 (C2 \ C)
=
1 (S 3 \ K).
(e)* Show that going round the full circle of homomorphisms, the resulting homomorphism 1 (S 3 \ K) 1 (S 3 \ K) is an isomorphism.
17
5. Covering spaces
Another basic method of computing fundamental groups is to identify the space
by a discrete group G acting

X as the quotient X/G
of a simply connected space X
freely on it by homeomorphisms. Under certain additional conditions, one then has
1 (X)
= G (just as 1 (S 1 ) = 1 (R/Z) = Z). In this lecture we will explore how
covering spaces arise in practice. We also see how a covering map gives rise to
two groups: (i) its group of deck transformations, and (ii) the image of 1 of the
covering space in 1 of the base.
X such that X has a
Definition 5.1. A covering map is a surjective map p : X
cover by open sets U with the property that p1 (U ) is the disjoint union of open
of a
sets, each of which is mapped by p homeomorphically onto U . The domain X
covering map is called a covering space of X.
The fibre F = p1 (x) is a discrete space. For an open set U as in the definition,
and x U , there is a homeomorphism t : p1 (U ) F U such that pr2 t = p as
maps p1 (U ) U (t is called a trivialisation for p over U ). Thus the fibres over
points of U are all homeomorphic, and hence, if X is path-connected, all the fibres
of p are homeomorphic. The covering map is trivial if there exists a trivialisation
over X.
Remark. In the theory of covering spaces its a useful safety precaution to assume
that all spaces are locally path connected (i.e., for any point x and any neighbourhood of x there is a smaller neighbourhood which is path connected).
Exercise 5.1: The following are covering maps:
(1) The quotient map R R/Z.
(2) The map S 1 S 1 , eit 7 eint .
(3) The product of covering maps (e.g. Rn (R/Z)n = Rn /Zn ).
(4) The quotient map S n RP n .
Example 5.2. Let (X, x) be a based space. A covering space Y for S 1 X can
be obtained by taking a family (Xn )nZ of identical copies of X, then letting Y be
the result of attaching Xn to R by identifying x Xn = X to n Z. The covering
map p : Y X is given on R by the quotient map R R/Z = S 1 S 1 X and
on Xn by the identification Xn = X.
Graphs. A graph is a topological space obtained by the following procedure.
One takes a discrete space V (the vertices), a set E (the edges) and for
` each e E
a map ae : {0, 1} V . One forms the identification space of V q eE [0, 1] in
which 0 [0, 1]e is identified with its image ae (0) V , and 1 [0, 1]e is identified
with ae V .
Example 5.3. A covering space of a graph is again a graph. For example S 1 S 1
is a graph with one vertex and 2 edges. The vertex has valency 4 (i.e., 4 intervals
emanate from it). Any covering space of S 1 S 1 is a graph in which each vertex
has valency 4. The edges of can be coloured red and blue so that each vertex
has two red and two blue intervals emanating from it. Moreover, can be oriented
(i.e., each edge given a direction) so that at each vertex, exactly one red interval
is outgoing and exactly one blue interval is outgoing. Conversely any oriented,
coloured graph with these properties defines a covering of S 1 S 1 .
18
TIM PERUTZ
5.1. Deck transformations.

Definition 5.4. Fix covering maps p1 : Y1 X and p2 : Y2 X. A map of
covering spaces from (Y1 , p1 ) to (Y2 , p2 ) is a map f : Y1 Y2 such that p1 f = p2 .
A deck transformation for a covering space p : Y X is a map of covering spaces
h from (Y, p) to itself which is also a homeomorphism.
The inverse of a deck transformation is another deck transformation. Hence the
deck transformations form a group Aut(Y /X).
Example 5.5. In Example 5.2, the covering space p : Y S 1 X has Z as its
group of deck transformations. The generator is the shift homeomorphism, acting
on R by t 7 t + 1 and sending Xn identically to Xn+1 .
Coverings arise in nature via group actions. Suppose given a continuous action
G Y Y of the discrete group G on the space Y .
Proposition 5.6. The quotient map q : Y Y /G is a covering map provided the
action is a covering action: Y is covered by open sets V such that gV V = for
all g G \ {e}. If Y is path connected, the group of deck tranformations is G.
Proof. Given x Y , take a neighbourhood V of x as in the statement. We may
assume V is connected. Let U = q(V ). Then q 1 (U ) is the disjoint union of the
open sets gV for g G. Each is mapped bijectively to U ; the map is open by
definition of the quotient topology, hence a homeomorphism. This shows that q is
a covering map.
Any g G determines a deck transformation x 7 g x, and these give a homomorphism G G0 , where G0 is the group of deck transformations. Since the
action is free, the kernel of this homomorphism is trivial. To see that it is surjective, suppose f is a deck transformation. Pick a point y Y , choose g G such
that f (y) = hg y, where hg is the action of g. Then hg1 f fixes y. By the last
theorem, G0 acts freely, hence hg1 f = id, i.e. f = hg .

5.2. Examples.
The action of Zn on Rn by translations is a covering action (take the cover
to be by balls of radius 1/3).
The action of the cyclic group Z/p on S 2n1 = {z Cn : |z| = 1}, where
the generator acts by scalar multiplication by e2i/p , is a covering action.
Indeed, a non-trivial group element moves every point by a distance d
(for the Euclidean metric in Cn ), where d = mink{1,...,p1} |1 e2ik/p |,
hence the open sets S 2n1 B(z; d/2) (with |z| = 1) provide a suitable
cover. The quotient L2n1 (p) = S 2n1 /(Z/p) is called a lens space.
The last example generalises: if Y is compact and simply connected, and a
finite group G acts freely on Y , then 1 (Y /G)
= G.
If G is a compact, simply connected topological group, and Z G a finite
subgroup, then the action of Z on G is a covering action. An interesting example is G = SU(2) and Z = {I} G. The quotient P U (2) := SU(2)/Z
is isomorphic to SO(3). Indeed, PU(2) is the group of conformal symmetries of C {}, while SO(3) the group of orientation-preserving isometries of S 2 . These symmetries coincide under the standard homeomorphism
19
C {} = S 2 . Moreover, there is a homeomorphism

3
SU(2) S = {(, ) C : || + || = 1},

(, ) 7

.
The involution A A on SU(2) corresponds to the antipodal map on S 3 ,

hence PU(2)
= RP 3 .
Exercise 5.2: Do this exercise if you know the basic facts about smooth manifolds.
Suppose Y and X are smooth n-manifolds, and p : Y X a smooth, proper map
whose derivative Dp : Tx Y Tp(x) X is an isomorphism for all x Y . Then p is a
(finite-sheeted) covering map.
Exercise 5.3: Show that T 2 \ {4 points} is a 2-sheeted covering of S 2 \ {4 points}.
Some possible approaches are (a) a direct topological argument; (b) the Weierstrass
-function from complex analysis; (c) a pencil of divisors of degree 2 on an elliptic
curve.
5.3. Unique path lifting.
X be a covering map. Fix basepoints x X and
Lemma 5.7. Let p : X
1
x
p (x).
(1) If : I X a path, and (0) = x, then there is a unique path : I X

such that (0) = x which lifts in the sense that p = .
: I2 X
once we specify
(2) A homotopy : I 2 X lifts uniquely to a map
(0, 0).
x
(3) The map p : 1 (X,
) 1 (X, x) is injective.
0
x
x
(4) If x
also lies in p1 (x) then p (1 (X,
0 )) and p (1 (X,
)) are conjugate
subgroups of 1 (X, x).
x
(5) All conjugates of p (1 (X,
)) arise in this way.
Proof. (1) The proof is exactly the same as the proof of unique path lifting for
R S 1 that we gave in our proof that 1 (S 1 ) = Z. Similarly (2).
(3) If p (0 ) and p (1 ) are homotopic rel endpoints then the unique lift of the
defines a homotopy rel endpoints between 0 and 1 .
homotopy to X
joining x
(4) Choose a path in X
0 to x
. Then we have
0
x
p (1 (X, x
)) = (p ) p (1 (X,
)) (p )1 .
(5) Follows from (1).
Exercise 5.4: Write out the missing details.

Exercise 5.5: A surjective map p : Y X which has unique path-lifting need not be
a covering map. (You may choose Y not to be locally path connected. For a harder
exercise, find an example where Y is locally path connected.)
X gives rise
Let us summarise where we have got to. A covering space p : X
(a) to a group of deck transformations Aut(X/X); and (b) to a conjugacy class of

x
subgroups of 1 (X, x), the images of 1 (X,
), for basepoints x
p1 (x).
= X = S 1 , and p is the covering eit 7 eint , then Aut(X/X)
Example 5.8. If X
=
2i/n
in
Z/n (the generator being multiplication by e
), while the image of 1 (X)
in 1 (X) is
1 (X) is nZ Z. We see in this example that the image of 1 (X)
a subgroup whose index is equal to the number of sheets of the covering. It is a
normal subgroup, and the quotient group is isomorphic to Aut(X/X).

If we take
20
TIM PERUTZ
= R, with p : R R/Z = S 1 the quotient map, then Aut(X/X)
X
= Z and
1 (R) = {1}, so again Aut(X/X)

= 1 (X)/p 1 (X).
In the next lecture we will see that these observations generalise (except that
is not always normal). In particular, if X
is simply connected
the image of 1 (X)
then Aut(X/X) = 1 (X).
21
6. Classifying covering spaces

In the previous lecture we introduced covering spaces. Today we classify the
covering spaces of a given space X.
The following theorem could be called the fundamental lemma of covering space
theory.
X be a covering map, with X
pathTheorem 6.1 (lifting criterion). Let p : X
connected, and f : B X a map from a path-connected and locally path-connected
such that p(
space B. Choose b B and x
X
x) = f (b). Then f lifts to a map
f : B X with p f = f and f (b) = x

if and only if
x
f (1 (B, b)) p (1 (X,
))
in 1 (X, f (b)). When it exists, the lift is unique.
Proof. If the lift exists then p f = f , hence im f im p . Uniqueness follows
from the uniqueness of lifts of paths. We now consider existence. Take y B and
is the
a path from b to y. We attempt to define f(y) = (1), where : B X
unique lift of f with (0) = x
. If this make sense and is continuous then it will
certainly fulfil the requirements. We need to prove that (1) is independent of the
choice of . If 0 is another such path then followed by ( 0 )1 is a loop l in B.
x
But if f (1 (Y, y)) p (1 (X,
)) then l is homotopic rel endpoints to the image
Lifting the homotopy gives a homotopy rel endpoints between
of a loop in X.
and the lift of ( 0 )1 , which shows that the lift 0 of f 0 ends at the same
point as does f . Continuity of f follows from local path-connectedness of B (cf.
Hatcher).

From now on, the base spaces of our covering maps will be assumed pathconnected and locally path-connected.
Corollary 6.2. If p1 : Y1 X and p2 : Y2 X are covering maps, and p(y1 ) =
x = p2 (y2 ), then there exists a homeomorphism h : Y1 Y2 with p2 h = p1 and
h(y1 ) = y2 if and only if p1 1 (Y1 , y1 ) = p2 1 (Y2 , y2 ) in 1 (X, x). Hence two
coverings of X are isomorphic iff they define conjugate subgroups of 1 (X, x).
Corollary 6.3. Any two simply connected covering spaces of X are isomorphic.
Because of this result, we shall refer to a simply connected covering space of X
as a universal cover of X.
X is a universal cover, Aut(X/X)
Corollary 6.4. If p : X
acts freely and tran1
sitively on any fibre p (x). We obtain an isomorphism Ix : 1 (X, x) Aut(X/X)

1
by fixing a base-point x
p (x), then mapping [] to the unique deck transformation which sends x
to (1), being the unique lift of with (0) = x
.
X
intertwining p are necProof. According to the lifting criterion, maps h : X
essarily homeomorphisms, and they are in natural bijection with the fibre p1 (x).

6.1. An equivalence of categories. We now formulate the classification theorem
for coverings of X. In a nutshell, this says that isomorphism classes of path connected covering spaces correspond to conjugacy classes of subgroups of 1 (X, x).
We give a sharper statement, which classifies not only the coverings, but also the
maps between them.
22
TIM PERUTZ
We shall define two categories and prove their equivalence. An equivalence of

categories F : C C 0 is a functor such that there exists a functor G : C 0 C so
that F G and G F are naturally isomorphic to the identity functors on C 0 and
C respectively. A standard result in category theory says that F is an equivalence
provided that (i) F : Hom(X, Y ) Hom(F(X), F(Y )) is bijective for all objects
X and Y , and (ii) every object of C 0 is isomorphic to some C(X).
Definition 6.5. Let G be a group. Its orbit category O(G) is the category whose
objects are the subgroups H G. For any H, the set G/H of left cosets of H
is a transitive G-set. We define the morphisms H K to be maps of G-sets
G/H G/K.
Definition 6.6. If X is a path-connected space, we define a category Cov(X)
whose objects are path-connected covering spaces p : Y X and whose morphisms
are maps of covering spaces.
Theorem 6.7. Suppose that (X, x) is a based space. Fixing a universal cover
X and a basepoint x
p: X
p1 (X) determines an equivalence of categories
G : O(1 (X, x)) Cov(X).
X be a
Proof. We define a functor G : O(1 (X, x)) Cov(X). Thus let X
simply-connected covering space, and fix a basepoint x
over x X. Path-lifting
starting at x
defines an isomorphism Ix : G Aut(X/X)
where G = 1 (X, x).
x (H). It comes with a projection map
Take H 1 (X, x), and define G(H) = X/I
X, and this is certainly a covering. Its fibre over x
G(H) X, induced by p : X
is canonically identified with G/H, and 1 (G(X), [
x]) maps to H under the covering
map.
Every path-connected covering Y X is isomorphic to G(H) for some H.
Indeed, we take H to be the image of 1 (Y, y) in 1 (X, x) for some y lying over x,
cf. Corollary 6.2.
If K is another subgroup, and f : G/H G/K a map of G-sets, let f (H) = K.
Then, for all g G, we have f (gH) = gK. Notice that if h H then f (hH) =
hK = K, hence 1 H K; conversely, an element such that 1 H K
defines a map of G-sets.
x (H)
We shall define G(f ) via the lifting criterion. We are looking for a map X/I
X/Ix (K) covering the identity on X. Such a map will be unique once we specify its
x (H) such that the image
effect on a point. For existence, take a basepoint z X/I
x (H), z) in G is 1 H (cf. Lemma 5.7, (5)). By the lifting criterion,
of 1 (X/I
x (H) X/I
x (K) which sends z to
there is a unique map of covering spaces X/I
[
x]. This is G(f ). Its straightforward to check this gives a functor.
It remains to see that G gives a bijection between morphism sets. This is another
application of the lifting criterion, but we omit the details.

Let us spell out some aspects of this correspondence.
At one extreme, we can consider the trivial subgroup {1} G, which
corresponds to the universal cover. At the other extreme, G G gives the
trivial cover X X.
In general, the fibre of the covering G(H) corresponding to H G is G/H.
Thus finite index subgroups correspond to coverings with finite fibres.
23
We can recover the conjugacy class of H G from G(H) as the image of

1 (G(H)) in 1 (X, x). (To recover H on the nose, we have to remember
the basepoint [
x] coming from x
X.)
The normal (or regular, or Galois) coverings of X are those coverings
q : Y X for which q 1 (Y ) is a normal subgroup of G. Equivalently,
Aut(Y /X) acts transitively on the fibre. A normal covering determines an
actual subgroup, not just a conjugacy class of subgroups.
The similarity of the classification theorem with the fundamental theorem of Galois
theory is not coincidental; the theory of etale maps in algebraic geometry unites
them. In particular, finite extensions of the function field K(X) of a variety X
correspond to finite (etale) coverings of X.
6.2. Existence of a simply connected covering space. Under very mild hypotheses, a simply connected covering exists. Assume X locally path connected.
X from a simply
Proposition 6.8. Suppose that X admits a covering map p : X
Then X is semi-locally simply connected, meaning that each
connected space X.
x X has a path-connected neighbourhood U such that im(1 (U ) 1 (X)) is
trivial.
Proof. Let U be a neighbourhood over which p is trivial. Then any loop in U
which is nullhomotopic (rel I). Projecting the nullhomotopy
lifts to a loop in X,
to X, we see that is nullhomotopic in X.

Exercise 6.1: Find a path connected, locally path connected space which is not semilocally simply connected.
as the set of homotopy classes [], where
Now fix a basepoint x X. Define X
X to
: I X with (0) = x and [] its homotopy class rel I. Define p : X
ought to be generated by
be the evaluation map [] 7 (1). The topology on X
the path components of the sets p1 (V ) with V X open. Path components do
not make sense a priori, but we can make sense of them, via path-lifting, when V
is path connected and im(1 (V ) 1 (X)) is trivial.
Proposition 6.9. If X is path-connected, locally path-connected and semi-locally
X is a covering map and X
is simply connected.
simply connected then p : X
Thus X admits a simply connected covering space.
more precise, then prove the proposition.
Exercise 6.2: Make the topology on X
Exercise 6.3: (From Mays book.) Identify all index 2 subgroups of the free group F2 .
Show that they are all free groups and identify generators for them.
Exercise 6.4: (a) The universal cover of the torus T 2 is R2 . Identify all the deck
transformations and hence determine (once again) the fundamental group. Which
surfaces can cover T 2 ? (b) Show that the Klein bottle is also covered by R2 ; identify
the deck transformations and hence the fundamental group.
Exercise 6.5: Let p : Y X be a covering (with Y path connected and X locally path
connected) such that p 1 (Y, y) = H G = 1 (X, p(y)). Show that Aut(X/X)

=
(NG H)/H, where NG H = {g G : gHg 1 = H}.
For the next exercise, you may use the following fact: the quotient SU(2)/{I}
is isomorphic, as a topological group, to SO(3).
24
TIM PERUTZ
Exercise 6.6: Define a regular tetrahedron as a set of four distinct, unordered, equidistant points on S 2 R3 . Let T be the space of regular tetrahedra. (a) Show that
1 (T ) has a central subgroup Z
= Z/2 such that 1 (T )/Z
= A4 . (b) Identify several
(at least 5) pairwise non-isomorphic, path connected covering spaces of T , describing
them geometrically. (c) Show that the fundamental group of the space P of regular
icosahedra (unordered collections of 20 distinct points on S 2 forming the vertices of a
regular icosahedron) has order 120, but that the abelianization 1 (P)ab has order at
most 2. (In fact it is trivial.) [Recall that the icosahedral group A5 is simple.]
Exercise 6.7: Rotation about a fixed axis, by angles increasing from 0 up to 2, determines a loop in SO(3). Show that is nullhomotopic.
25
II. Singular homology theory

7. Singular homology
We explain a fundamental construction of algebraic topologysingular homology.
We compute the 0th homology groups in terms of the path components of the space,
and show that 1 maps onto the first homology group.
Precursors of homology theory go back to the 18th Century and Eulers formula
v e + f = 2 for the numbers of vertices, edges and faces of a convex polyhedron.
Its systematic development began with Poincare in the 1890s. The definition of
singular homology we shall give is due to Eilenberg (1944), but it rests on fifty
years of exploration and refinement by many mathematicians. Every aspect of it is
the result of a gradual process of experiment and abstraction. It is perfectly simple
and, at first, perfectly mysterious.
7.1. The definition. The geometric n-simplex is
n = {(x0 , . . . , xn ) [0, 1]n+1 :
xi = 1}.
It is the convex hull [v0 , . . . , vn ] of the points vi = (0, . . . , 0, 1i , 0, . . . , 0).

Define the ith face map
i : n1 n ,
(x0 , . . . , xn1 ) 7 (x0 , . . . xi1 , 0, xi , . . . , xn1 ).
It is homeomorphism onto the face [v0 , . . . , vi , . . . , vn ].

An n-simplex in the space X is a continuous map : n X. Let n (X) be
the set of all n-simplices. Define the nth singular chain group Sn (X) as
Sn (X) = Zn (X) ,
the
P free abelian group generated by n (X). It is the group of finite formal sums
i ni i with ni Z and i n (X). For n > 0, define n : Sn Sn1 as the
Z-linear map such that
n
X
n =
(1)i ( i ), n (X).
i=0
In alternative notation, n =
Pn
i=0
(1)i (|[v0 ,...,vi ,...,vn ] ).
Lemma 7.1. n n+1 = 0.

Proof. This is a consequence of the following relations among the face maps:
i j = j i1 ,
j < i.
For any n + 1-simplex , we have

X
X
n n+1 =
(1)i+j i j
0jn 0in+1
0j<in+1
=
=
0kln
=0.
(1)i+j i j
0ijn
(1)
i+j
j i1 +
0j<in+1
(1)i+j i j +
(1)k+l+1 k l +
(1)i+j i j
0ijn
X
0ijn
(1)i+j i j
26
TIM PERUTZ
Since simplices generate Sn+1 (X), the result follows.
It is convenient to let 0 : S0 (X) 0 be the zero-map. The nth singular homology of X is the abelian group
Hn (X) := ker n / im n+1 .
Elements of ker n are called n-cycles; elements of im n+1 are n-boundaries. By
the lemma, an n-boundary is an n-cycle, and the nth homology is the group of
n-cycles modulo n-boundaries.
In future lectures we will develop these groups systematically. Today we will
look only at the zeroth and first homology groups.
7.2. The zeroth homology group.
P
P
Proposition 7.2. The map : S0 (X) Z, ( ni i ) = ni induces a surjection
H0 (X) Z provided only that X is non-empty. When X is path-connected, this
map is an isomorphism.
Proof. We have to show that descends to H0 (X) = S0 (X)/ im 1 . If is a 1simplex then = 0 1 . Thus ( ) = 1 1 = 0. Hence (im 1 ) = 0,
and descends to H0 (X). For
P any 0-simplex , (n) = n, so is surjective. If X
is path-connected, take s = ni i ker . We may assume ni = 1 for all i. The
number of + and signs is equal, so we may partition the 0-simplices into pairs
(i , j ) with ni = 1 and nj = 1. But i j is the boundary of a 1-simplex (i.e.,
of a path), since X is path-connected. Hence s im 1 .

L
Exercise 7.1: Show that, in general, Hn (X) = Y 0 (X) Hn (Y ), where 0 (X) is the
Z0 (X) .
set of path-components of X. Thus H0 (X) =
So, whilst S0 (X) is typically very large (often uncountably generated), H0 (X)
is finitely generated for all compact spaces.
7.3. The first homology group. Theres a homeomorphism I 1 given by
t 7 tv1 +(1t)v0 . Thus a path : I X defines a 1-simplex . When (0) = (1),
is a 1-cycle.
Lemma 7.3. Fix a basepoint x X. The map 7 induces a homomorphism
h : 1 (X, x) H1 (X).
Proof. A constant loop is the boundary of a constant 2-simplex. Loops which are
homotopic rel endpoints give homologous 1-simplices (by subdividing a square into
two triangles and using the fact that constant loops are boundaries). Thus h is
well-defined. If f and g are composable paths, the composition f g maps under
define a 2-simplex = (f g) p : 2 X, where p is the projection
h to f + h:
[v0 , v1 , v2 ] [v0 , v2 ], t0 v0 + t1 v1 + t2 v2 7 t1 v1 + t2 v2 . We have = g f[
g + f.

The map h is sometimes called the Hurewicz map.
Proposition 7.4. The kernel of the Hurewicz map h : = 1 (X, x) H1 (X)
contains the commutator subgroup [, ], and hence h induces a homomorphism
ab := /[, ] H1 (X).
When X is path-connected, h is surjective.
27
Proof. Since h is a homomorphism, h(f gf 1 g 1 ) = h(f )+h(g)h(f )h(g) = 0.

Thus [, ] ker h. To obtain surjectivity in the path-connected case,P
note that
the group of 1-cycles is generated by loops, where a loop is a 1-cycle iZ/N i
with 1 i + 0 i+1 = 0 for all i Z/N . Thus it suffices to show that any loop lies
in the image of h. But any loop is homologous to a loop based at x (insert a path
from x to 0 (0), and 1 from N 1 (1) to x). The composition of all the paths
making up the based loop is homologous to their sum, and it lies in im(h).

Example 7.5. Any simply connected space X has H1 (X) = 0.
Remark. By analogy, one can look at the group H2 (X)/s(X), where s(X) H2 (X)
is the subgroup generated by the spherical cycles: those represented by a map from
a tetrahedron (built from four 2-simplices) into X. This group is zero when X is
simply connected. A theorem of Hopf [Comment. Math. Helv. 14, (1942), 257309]
says that, for a general path-connected X, H2 (X)/s(X) depends only on 1 (X).
It is naturally isomorphic to a group which is now understood as H2 (1 (X)), the
second group homology of 1 (X). Indeed, group homology was developed partly in
response to Hopfs theorem. See, e.g., Brown, Cohomology of groups (GTM 87).
28
TIM PERUTZ
8. Simplicial complexes and singular homology

We show that a singular n-cycle can be represented by a map from an n-dimensional
-complex. We complete our calculation of H1 in terms of 1 .
8.1. -complexes.
Definition 8.1. A -complex (or semi-simplicial complex) is a space X equipped
with sets Sn , empty for n 0, and for each n and each Sn a continuous maps
n : n X. We require that (i) n is injective on int(n ), and X (as a set) is
the disjoint union over n and of the images n (int(n ); (ii) when n > 0, the
restriction to a face, n i , is equal to n1 for some Sn1 ; and (iii), a set
U X is open iff (n )1 (U ) is open for all n and all .
There are a number of connections between -complexes and singular homology.
If a space is given the structure of a -complex, there is a distinguished sub-space
Snsimp (X) Sn (X), spanned by the n-simplices n . One has (Snsimp )(X)
simp
Sn1
(X), so it makes sense to form the simplicial homology group
Hnsimp (X) =
simp
ker(n : Snsimp (X) Sn1
(X))
simp
im(n+1 : Sn+1
(X) Snsimp (X)
This comes with a natural homomorphism Hnsimp (X) Hn (X), induced by the
inclusion Snsimp (X) Sn (X).
Exercise 8.1: Think of S 2 as a tetrahedron, i.e., a -complex with four 2-simplices, six
1-simplices and four 0-simplices. Show that for this structure
H0simp (S 2 ) = Z,
H1simp (S 2 ) = 0,
H2simp (S 2 ) = Z,
simp
H>2
(S 2 ) = 0.
Exercise 8.2: Compute Hsimp for the spaces T 2 , RP 2 and K 2 , each thought of as a
-complex with two 2-simplices (and some 1- and 0-simplices).
Remark. You may like to keep in mind the following fact, even though its not
part of the logical development of this course: the map Hnsimp (X) Hn (X) an
isomorphism. So, for example, the homology of a -complex is finitely generated.
The following simple observation gives some geometric insight into singular homology.
Lemma 8.2. Let z be a singular n-cycle in X, so n z = 0. Write it as z =
PN
i=1 i i with i = 1. Then there is an -complex Z, with precisely N nsimplices (1 , P
. . . , N ) and no higher-dimensional simplices, and a map f : Z X,
such that (i)
i i represents a simplicial n-cycle for Z, and (ii) i = f i for
each i.
Proof. Since n z = 0, each face i j must cancel with another face i0 j 0 .
Thus, we can partition the set of faces of all i into pairs. We define a -complex
Z by gluing N n-simplices together along their faces, paired up in the way just
determined. This has the right properties.

More generally, if n z = y, we can build a -complex and a map from it into X
so that the summed boundary of the n-simplices in the complex maps to X as the
cycle y.
29
8.2. The Hurewicz map revisited. Last lecture, we introduced the Hurewicz
map h : 1 (X)ab H1 (X) and proved its surjectivity (assuming X path connected). We did not analyse its kernel. We now finish the job.
Theorem 8.3. When X is path-connected, the Hurewicz map h : 1 (X)ab H1 (X)
is an isomorphism.
Example 8.4.
Recall that 1 (S 1 )
= Z. Since this group is already abelian,
1
H1 (S ) = Z also.
Recall that 1 (T 2 ) = 1 (T 2 )ab
= Z2 , 1 (K 2 )ab
= Z Z/2 and 1 (RP 2 ) =
2 ab
2
2
1 (RP ) = Z/2, where T is the 2-torus, K the Klein bottle, and RP 2
the real projective plane.
Recall that the closed, oriented surface g of genus g has
1 (g )
= ha1 , . . . , ag ; b1 , . . . , bg | [a1 , b1 ] [ag , bg ]i.
Thus H1 (g )
= Z2g , generated by the classes of a1 , . . . , ag and b1 , . . . , bg .
Proof of the theorem. Take a loop ker h: say = 2 . We will show that [] is
a product of commutators in , hence lies in [, ]. We have already proved that
h is onto, so this will complete the proof. But we can build from the 2-chain a
2-dimensional -complex K, and a map f : K X, with the following properties:
if z is the sum of the 2-simplices, then 2 z = for a 1-simplex such that f = .
Moreover, the image of in K is a loop K. It suffices, then, to show that K
is in the commutator subgroup of 1 (K, b) (for the obvious basepoint b K), for
then the corresponding result will hold in X just by applying f . Thus we deduce
the theorem from the following lemma.

Lemma 8.5. Let K be a compact, connected, 2-dimensional -complex. Suppose
PN
z is the sum of the 2-simplices, and that 2 z = i=1 i for some 1-simplices i
such that i (1) = i+1 (0), i Z/N . Fix a basepoint b; take it to be a vertex lying
on K. Then 2 z represents an element of = 1 (K, b). This element lies in the
normal subgroup [, ] generated by commutators.
Proof. First observe (exercise!) that in general, if we have a free homotopy through
loops t : S 1 X, then we have two fundamental groups = 1 (X, 0 (1)) and
0 = 1 (X, 1 (1)); and [0 ] [, ] iff [1 ] [ 0 , 0 ] 0 .
We now proceed by induction on the number of 2-simplices. The lemma is
obvious when there is only one 2-simplex. When there is more than one, remove a
2-simplex adjacent to the boundary which has the basepoint as one of its vertices,
so as to create a new -complex K which again satisfies the hypotheses(!). Pick a
new basepoint b0 on K 0 which was one of the vertices of . By induction, K 0 is
a product of commutators in 1 (K 0 , b0 ), hence in 1 (K, b0 ). But K is homotopic
through loops to K 0 , so the result follows from our observation.

Remark. The lemma is connected with the geometric interpretation of the algebraic
notion of commutator length. In general, for a group , the commutator length
cl() of [, ] is the least integer g such that is the product of g commutators
in . If = 1 (X, x), then one can show that cl() is the minimal genus g of a
compact oriented surface K bounding . Here by a compact oriented surface I mean
a -complex K, equipped with a map f : K X, which satisfies the conditions of
the lemma and which is locally homeomorphic to R2 . The genus of K is half the
rank of H1simp (K).
30
TIM PERUTZ
9. Homological algebra
Having introduced singular homology, we now need an adequate algebraic language to describe it.
9.1. Exact sequences. We shall work with modules over a base ring R, which
we will assume to be commutative and unital. We write 0 for the zero-module. A
sequence of R-modules and linear maps
a
ABC
is exact if ker b = im a. A longer sequence of maps is called exact if it is exact at
each stage.
0 B C is exact iff b is injective.

a
A B 0 is exact iff a is surjective.
a
0 A B 0 is exact iff a is bijective, i.e., iff a is an isomorphism.
Exact sequences of the form
a
0 A B C 0,
are called short exact sequences. In such a sequence, coker a = B/ im a =
B/ ker b. But b induces an isomorphism B/ ker B im b = C. Thus a is
injective with cokernel C, while b is surjective with kernel A.
A short exact sequence is called split if it satisfies any of the following
equivalent conditions: (i) there is a homomorphism s : C B with bs =
idC ; (ii) there is a homomorphism t : B A with ta = idA ; or (iii) there is
an isomorphism f : B A C so that a(x) = f (x, 0) and b(f 1 (x, y)) = y.
If the six-term sequence
a
0ABCD0
ker b while c induces an isois exact then a induces an isomorphism A =
morphism D
= coker b.
Exact sequences are useful because if one has partial information about the groups
and maps in a sequence (in particular, the ranks of the groups) then exactness helps
fill the gaps.
Example 9.1. Suppose one has an exact sequence of Z-modules
i
0 Z A Z/2 0.
What can one say about A (and about the maps)? Choose x A with p(x) 6= 0.
Then 2x ker p = im i. There are two possibilities:
(i) 2x = i(2k) for some k. Let x0 = x i(k). Then 2x0 = 0. We can then
define a homomorphism s : Z/2 A with p s = id by sending 1 to x. Thus the
sequence splits, and so may be identified with the trivial short exact sequence
0 Z Z/2 Z Z/2 0.
(ii) 2x = i(2k + 1) for some k. Let x0 = x i(k). Then 2x0 = i(1) and
p(x0 ) = p(x) 6= 0. Given y A, either y = i(m) for some m, in which case
y = 2mx0 , or else y x0 = i(m) for some m, in which case y = (2m + 1)x0 . Thus
A = Zx0 . Moreover, x0 has infinite order (since Zx0 contains im i). So the sequence
may be identified with the sequence 0 Z Z Z/2 0, in which the map
Z Z is multiplication by 2 and Z Z/2 is the quotient map.
31
9.2. Chain complexes. A chain complex over R is a collection {Cp }pZ of Rmodules, together with linear maps dp : CpL
Cp1 , called differentials, satisfying
dp1 dp = 0. We write C for the sum
p Cp , which is a graded module, and
L
d=
dp : C C (an endomorphism map which lowers degree by 1, satisfying
d2 = 0). We define the homology
H(C ) = ker d/ im d.
L
L
Notice that ker d = p ker dp and im d =
im dp , so H(C ) = p Hp (C ), where
Hp (C ) = ker dp / im dp+1 .
Elements of Zp (C) := ker dp are called p-cycles; elements of Bp (C) := im dp+1 ,
p-boundaries. If H(C ) = 0, we say that C is acyclic.
A chain map from (C , dC ) to (D , dD ) is a linear map f : C D such
that f (Cp ) Dp and dD f = f dC . A chain map induces homomorphisms
Hp (f ) : Hp (C ) Hp (D ). A chain map which induces an isomorphism on homology is called a quasi-isomorphism.
L
9.2.1. Chain homotopies. We need a criterion for two chain maps f and g : C
D to induce the same map on homology. For this we introduce the notion of
chain homotopy. A chain homotopy from g to f is a collection of linear maps
hp : Cp Dp+1 such that
dD hp + hp1 dC = f g.
If x Zp (C) then f (x) = g(x) + dD (hp x), hence [f (x)] = [g(x)] H(D ). If, for
example, there is a chain homotopy from f : C C to the identity map idC , then
f is a quasi-isomorphism. If there is a null-homotopy, i.e., a chain homotopy from
f to the zero-map, then f induces the zero-map on homology. If both possibilities
occur then C must be acyclic, i.e., H (C) = 0.
9.2.2. Short and long exact sequences. We study the effect of passing to homology
on a short exact sequence
a
0 A B C 0
of chain complexes and chain maps.
Hp (a)
Hp (b)
Lemma 9.2.
(i) The sequence Hp (A) Hp (B) Hp (C) is exact.
(ii) Take x Ap with dA x = 0. Then [x] ker Hp (a) iff there exists y Bp+1
such that a(x) = dB y.
(iii) Take z Cp with dC z = 0. Then [z] im Hp (b) iff there exists y Bp1
such that b(y) = z and dB y = 0.
Proof. (i) Take y Bp with dB y = 0 and b(y) = dC z for some z Cp+1 . Then
z = b(y 0 ), say, and b(y dB y 0 ) = dC (z b(y 0 )) = 0, so y dB y 0 = a(x) for some
x Ap , i.e. y im a + im dB , as required.
(ii) is obvious, and (iii) almost so.

Points (ii) and (iii) can be pushed considerably further. Define the connecting
homomorphism
: Zp (C) Hp1 (A)
as follows:
(z) = [x] when there exists y Bp with b(y) = z and a(x) = dB y.
32
TIM PERUTZ
Lemma 9.3. is a well-defined map.

Proof. Note first that, since b is onto, there is some y with b(y) = z; and b(dB y) =
dC (by) = dC z = 0, hence dB y ker b = im a. Thus suitable y and x exist.
Moreover, x is determined by y, because of the injectivity of a. If b(y) = b(y 0 ) = z
then y y 0 ker b = im a, so y = y 0 +a(x0 ) for some x0 , and dB y = dB y 0 +dB a(x0 ) =
dB y 0 + a(dA x0 ). Thus replacing y by y 0 has the effect of replacing x by x + dA x0 ,
so the homology class [x] is well-defined.

Linearity of is clear. Note that maps Bp (C) to 0, and hence descends to a
map on homology,
: Hp (C) Hp1 (A).
Theorem 9.4. The short exact sequence of chain complexes
a
0 A B C 0
induces an exact sequence
Ha
Hb
Ha
Hb
Hp (A) Hp (B) Hp (C) Hp1 (A) Hp1 (B) Hp1 (C) .

Proof. We have already established exactness at Hp (C) and well-definedness of the
connecting map .
Exactness at Hp (C): Say [z] ker . This means that z = b(y) and a(dA x0 ) =
dB y for some x0 Ap . Then dB (y ax0 ) = 0, and b(y ax0 ) = z, so z im b.
Exactness at Hp1 (A): Let x Zp1 (A), and suppose that ax = dB y. Then
[x] = (b(y)).

Exercise 9.1: We consider chain complexes (C , ) over a field k such that dimk H (C) <
. The Euler characteristic of C is then defined as the alternating sum
X
(C ) =
(1)p dimk Hp (C).
p
(i) Show that when

(ii) Show that if
dimk Cp < , one has (C ) =
(1)p dimk Cp .
Cp Cp1 Cp2 . . .
P
P
is an exact sequence, and p dim Cp < , then
(1)p dim Cp = 0.
Exercise 9.2: A collection C of Z-modules is called a Serre class if for every short exact
sequence 0 A B C 0 such that two out of the three Z-modules A, B,
C are in C, the third is in C also. Fix a prime p Z. Identify which of the following
properties of Z-modules M define Serre classes: (a) M is torsion; (b) M is torsion-free;
(c) M is torsion but has no p-torsion; (d) every element of M has p-power order; (e);
every element of M is divisible by p; (f) every M is finitely generated. (*) What if we
replace M by an arbitrary commutative ring R (and p by a prime of R?).
Exercise 9.3: Prove or give a counterexample: given two path connected open sets U
and V whose union is X and whose intersection U V is path connected, there exists
a short exact sequence of abelian groups
0 H1 (U V ) H1 (U ) H1 (V ) H1 (X) 0.
Exercise 9.4: If a : A B is a chain map its mapping cone, denoted cone(a) , is
the complex
cone(a)n = An1 Bn
33
with differential dcone(a) (x, y) = dA x + a(x) + dB y (check that this squares to zero).
The point of this construction is to convert questions about chain maps to questions
about chain complexes.
(i) Show that the induced map on homology, a = H(a) : H (A) H (B), is an
isomorphism iff H(cone(a) ) = 0.
(ii) Construct a short exact sequence
0 B cone(a) A1 0,
and identify the connecting map in the resulting long exact sequence.
(iii) Show that to give a map of complexes f = (h, b) : cone(a) C is to give a
chain map b : B C and a chain-homotopy h from b a to the zero map.
(iv) Show that if the map f from (iii) induces an isomorphism on homology then
there is a long exact sequence
a
Hn (A) Hn (B)
Hn (C) Hn1 (A) Hn1 (B)
Hn1 . . .
Exercise 9.5: (*) If a : A B and b : B C are chain maps, what can we

say about cone(b a)? Show how to arrange the six groups H (A), H (B), H (C),
H (cone(a)), H (cone(b)) and H (cone(ba)) as the vertices of an octahedral diagram
of maps. Four of the faces should be commuting triangles, the other four exact triangles
(i.e., long exact sequences visualised as triangles). The last of these triangles is a long
exact sequence
Hp (cone(a)) Hp (cone(ba)) Hp1 (cone(b)) . . . .
The hard part of the exercise is constructing this triangle and proving its exactness.
[Hint: define f : cone(a) cone(ba) by f (x, y) = (x, by). There is a natural inclusion
i : cone(b) cone(f ). Show that i is a chain-homotopy equivalence.]
This exercise shows shows that the derived category of the abelian category of chain
complexes satisfies Verdiers octahedral axiom for triangulated categories.
34
TIM PERUTZ
10. Homotopy invariance of singular homology

We prove that singular homology is an invariant of homotopy type. We thereby
we compute the homology of contractible spaces.
Proposition 10.1. Let denote a one-point space. Then Hi () = 0 for all i > 0.
Proof. There is exactly one simplex i in each dimension. Thus the singular complex is
Z2 Z1 Z0 0.
Since n i = n1 , the boundary operator is given by
n
X
1
n n =
(1)i n1 = [1 + (1)n ]n1 .
2
i=0
Thus the complex is
0
Z3 Z2 Z1 Z0 0.
So ker i = 0 when i is even and positive; and when i is odd, i+1 is onto. Thus
Hi () = 0 when i > 0. As expected, we find H0 ()

= Z.
Maps between spaces introduce homomorphisms between homology groups. Given
f : X Y , define f# : Sn (X) Sn (Y ) by
f# () = f .
It is clear that this is a chain map: n f# = f# n . Thus there is an induced map
f = Hn (f ) : Hn (X) Hn (Y ).
Notice that if g : Y Z is another map then (g f )# = g# f# , and hence
(g f ) = g f .
Remark. In categorical language, we can express this by saying that Hn defines a
functor from the category Top of topological space and continuous maps to the
category Ab of abelian groups and homomorphisms. That is, Hn associates with
each space X an abelian group Hn (X); with each map f : X Y a homomorphism
Hn (f ) : Hn (X) Hn (Y ); and the homomorphism Hn (g f ) associated with a
composite is the composite Hn (g) Hn (f ). Moreover, identity maps go to identity
maps.
Theorem 10.2. Suppose that F is a homotopy from f0 : X Y to f1 : X Y .
The homotopy then gives rise to a chain homotopy P F : S (X) S+1 (Y ) from
(f0 )# to (f1 )# , that is, a sequence of maps PnF : Sn (X) Sn+1 (Y ) such that
F
n+1 PnF + Pn1
n = (f1 )# (f0 )# .
Hence Hn (f0 ) = Hn (f1 ).

Corollary 10.3. If f : X Y is a homotopy equivalence then Hn (f ) is an isomorphism for all n.
Corollary 10.4. A contractible space X has Hi (X) = 0 for all i > 0.
Remark. We can express the theorem in categorical language. Define a category
hTop whose objects are topological spaces, and whose morphisms are homotopy
classes of continuous maps. Then singular homology defines a sequence of functors
Hn : hTop Ab.
35
To prove the theorem, we begin with low-dimensional cases, because there the
geometry is transparent.
Proof of the theorem when n is 0 or 1. First, given a 0-simplex : 0 X, note
that we have a 1-simplex P0F () := F : I = 1 Y , and 1 (P0F ()) = f1
f0 .
Next, consider some 1-simplex : 1 X. We want to examine f1 f0 .
Since 1 = I, our homotopy F defines a map on from the square 1 I to Y ,
F ( idI ) : 1 I.
But the square is a union of two 2-simplices along a common diagonal. To notate
this, let the bottom edge be 1 {0} = [v0 , v1 ], and the top edge 1 {1} = [w0 , w1 ].
Thus the square is the convex hull [v0 , v1 , w0 , w1 ] of its four vertices. It is the union
of the two triangles [v0 , v1 , w1 ] and [v0 , w0 , w1 ] along the common edge [v0 , w1 ]. Note
that by expressing these triangles as convex hulls, we implicitly identify them with
the geometric 2-simplex 2 : for the first of them, say, the point t0 v0 + t1 v1 + t2 w1
corresponds to the (t0 , t1 , t2 ) 2 .
Now define a 2-chain P1F () by applying F to each of these two simplices:
P1F () = F ( idI )|[v0 ,w0 ,w1 ] F ( idI )|[v0 ,v1 ,w1 ]
We have
2 P1F () = f1 f0 + P0F (1 ) :
the first terms come fom the top of the square, the second term from the bottom,
and the third from the two other sides. Thus, defining P1F : S1 (X) S0 (Y ) and
P2F : S2 (X) S1 (Y ) by linearly extending the definitions from simplices to singular
chains, we have that
2 P2F + P1F 1 = (f1 )# (f0 )# .

Proof of the theorem in arbitrary dimensions. We proceed in the same way. For an
n-simplex : n X, we have a map
F ( idI ) : n I Y
defined on the prism n I, and we want to express this as a sum of n+1-simplices.
To do this, we think of the prism as the convex hull [v0 , . . . , vn , w0 , . . . , wn ], where
vi is the ith vertex of {0} = , and wi the ith vertex of {1} = . Then
one can check (as Hatcher does) that
n
[
I =
[v0 , . . . , vi , wi , wi+1 , . . . , wn ],
i=0
that each [v0 , . . . , vi , wi , wi+1 , . . . , wn ] is an n + 1-simplex, and that these simplices

intersect along common faces. This gives I the structure of a -complex.
We now define
n
X
PnF () =
(1)i F ( idI )|[v0 ,...,vi ,wi ,wi+1 ,...,wn ] ,
i=0
extending by linearity to get a map PnF : Sn (X) Sn+1 (Y ). The boundary of

Pn () should then consist (geometrically and hence algebraically) of f1 , f0
and Pn1 (). Since we did not actually verify that we had a -complex, let us
instead verify algebraically that P F defines a chain homotopy.
36
TIM PERUTZ
We have
n+1 PnF () =
(1)i+j F ( idI )|[v0 ,...vj ...,vi ,wi ,...,wn ]
ji
(1)k+l F ( idI )|[v0 ,...,vk ,wk ,...wl1 ...,wn ] .
l>k
The term with j = i = 0 in the first sum is

F ( idI )|[w0 ,...,wn ] = f1 .
The term with l = k + 1 = n in the second sum is
F ( idI )|[v0 ,...,vn ] = f0 .
Next we look for the cancelling pairs of faces which we expect geometrically. These
appear as the equality of [v0 , . . . , vi , wi , . . . , wn ] = [v0 , . . . , vi1 , w
i1 , . . . , wn ]. Apart
from the exceptional cases i = j = 0 and l = k + 1 = n, the j = i term in the first
sum cancels with the l = k + 1 term in the second sum where k = i 1. So, at this
point we have
n+1 PnF () = f1 f0
X
+
(1)i+j F ( idI )|[v0 ,...vj ...,vi ,wi ,...,wn ]
j<i
(1)k+l+1 F ( idI )|[v0 ,...,vk ,wk ,...wl ...,wn ] .
l>k
We want the two sums here to total PnF (n ). But if j < i, then
X
PnF ( i ) =
(1)j+i1 ( idI )|[v0 ,...vj ...vi ,wi ,...,wn ] .
j
If j i then instead
PnF ( i ) =
(1)j+i ( idI )|[v0 ,......,vi ,wi ,...wj1 ...,wn ] .
So the desired equality does indeed hold.
Exercise 10.1: Suppose that X is a subspace of Rn such that there is a map r : Rn X

with r|X = idX . Show that X has the homology of a point.
Exercise 10.2: Compute the first homology group H1 of the n-torus T n = (S 1 )n . Use
this to construct a surjective homomorphism G GLn (Z), where G is the group of
homotopy equivalences T n T n . Show that when n = 1 its kernel consists of maps
homotopic to the identity.
Exercise 10.3: (*) The model Dehn twist on the annulus A = [1, 1] (R/Z) is the
homeomorphism t : A A of form (s, t) 7 (s, t + (s + 1)/2). A Dehn twist along an
embedded circle C in a surface S is a homeomorphism S S obtained by identifying
a neighbourhood of C with A, and transplanting a model Dehn twist into S. (I am
being careless about right/left-handed twists.)
Let be a genus 2 surface, and C a circle dividing it into two 1-holed tori.
Show that if f is a Dehn twist along C then f acts on H1 () as the identity, but f
is not homotopic to the identity map.
37
11. The locality property of singular chains

There is a locality theorem for singular chains, reminiscent of the proof of van
Kampens theorem on 1 . This may be regarded as the technical core of singular
homology theory. We do not give a complete proof, but we reduce it to a lemma
concerning the geometric p-simplex p .
Definition 11.1. An excisive triad is a triple (X; A, B) with X a space and A, B
subspaces of X such that X = int(A) int(B).
Theorem 11.2 (locality for singular chains). Suppose (X; A, B) is an excisive
triad. Let Sn (A + B) denote the subgroup of Sn (X) generated by the images of
n (A) and n (B). Notice that it is a subcomplex. Then the inclusion map
i : S (A + B) S (X)
is a quasi-isomorphism.
(Hatcher proves that i is a chain-homotopy equivalence, but this is more than
we need.)
Setting up the proof. Form the quotient complex Q = S (X)/ im(i). Then we
have a short exact sequence
i
0 S (A + B) S (X) Q 0
and hence a long exact sequence of homology groups
i
Hp+1 (Q ) Hp (S (A + B))
Hp (X) Hp (Q ) . . .
The theorem asserts that i is an isomorphism. From the long exact sequence, we
see that this is equivalent to the assertion that Q is acyclic, i.e., that
Hp (Q ) = 0
for all p Z.
Thus we have to show that if q is a singular p-chain representing a cycle in q, so

q = r + s
with r Sp1 (A) and s Sp1 (B), then there are p-chains a Sp (A) and b
Sp (B), and a (p + 1)-chain c Sp+1 (X), such that
q = a + b + c.
Clearly it suffices to do this when q is a simplex. So, in words: we must show that
any : p X is homologous to the sum of a p-chain in A and a p-chain in B.

We could hope to find such a homology by breaking up the simplex as a
union of sub-simplices making p into a -complex. If the p-simplices in this
decomposition are sufficiently small then they will map either to A or to B under
. When is a 1-simplex, i.e., a path I X, it is clear how we could do this: we
write I = [0, 1/2] [1/2, 1]. Then is homologous to |[0,1/2] + |[1/2,1] . Iterating
this subdivision k times, we break up the unit interval into sub-intervals of length
2k ; when k 0, each sub-interval will map into int(A) or int(B).
The proof in higher dimensions uses a generalization of this subdivision of 1 .
38
TIM PERUTZ
Lemma 11.3 (Subdivision lemma). The geometric p-simplex p can be decomposed as a p-dimensional -complex in such a way that all the p-simplices 1 , . . . , N
in this decomposition have diameter < 1, and such that in the singular chain complex S (p ), one has
X
idp
i im p+1 .
i
Here we regard idp as a p-simplex in p . (In fact, one can take N = p! and the
p
.)
diameters to be p+1
The particular subdivision we have in mind here is called barycentric subdivision.
For the proof of the lemma we refer to Hatcher (it can be extracted from Steps 1 and
2 of the proof of the Excision Theorem). We will at least say what the barycentric
subdivision is. The barycenter b of a p-simplex [v0 , . . . , vp ] is the point
1
(v0 + + vp ).
b=
p+1
We now define the barycentric subdivision by induction on p. When p = 0, the
subdivision of 0 = [v0 ] has just one simplex: [v0 ] itself. When p > 0, the psimplices of the barycentric subdivision of [v0 , . . . , vp ] are of form [b, w0 , . . . , wp1 ],
where [w0 , . . . , wp1 ] is a (p 1)-simplex in the barycentric subdivision of some face
[v0 , . . . , vi , . . . , vp ] of [v0 , . . . , vp ].
The subdivision lemma is a little fiddly to prove. Since we are omitting the proof,
let us emphasize that this is an entirely combinatorial lemma concerning convex
geometry in Euclidean spaces; the target space X does not appear at all.
P
Proof of locality, granting the subdivision lemma. Write = i i Sp (p ). Now,
each i is a map p p (actually an embedding), so we can iterate the subdivisionPprocess, considering the composed maps j i : p p . Lets write
2 = i,j j i , and more generally
X
n =
in i1 .
i1 ,...,in
By induction on n, we have that idp n im p+1 .

Let 1 be the maximum diameter of one of the i . Any x p has an open
neighbourhood Nx such that (N ) is contained in int(A) or in int(B), since these
are open sets that cover X. But the image of in i1 has diameter (1 )n ,
so for large enough n, Nx is contained is the image of such a simplex. Thus p
is covered by subdivided simplices in i1 which map either to int(A) or to
int(B). A priori, the number n depends on x, but because p is compact we can
use the same n = n0 for all these subdivided simplices.
We know that idp n0 Bp (p ) (recall that Bp denotes im p+1 ), and applying we find that
# n0 Bp (X).
But # n0 is the sum of simplices in0 i1 that map either to int(A) or
to int(B). This proves the theorem.
39
12. MayerVietoris and the homology of spheres

The locality theorem from the previous lecture has an important consequence:
the exact MayerVietoris sequence. Using this sequence, we can at last carry out
interesting calculations in singular homology. We show that the homotopy type of
S n , and hence the homeomorphism type of Rn , detects the dimension n.
12.1. The MayerVietoris sequence. We extract from the locality theorem an
extremely useful computational tool in singular homology.
Theorem 12.1. Suppose (X; A, B) is an excisive triad. Let a : A X, b : B X,
: A B A and : A B B be the inclusion maps. Then there is a canonical
long exact sequence

(a ,b )
Hp (A B) Hp (A) Hp (B) Hp (X) Hp1 (A B) . . . .

Remark. Since H1 (A B) = 0, the sequence ends with H0 (A) H0 (B)
H0 (X) 0.
Proof. Theres a short exact sequence of chain complexes
a+b
0 S (A B) S (A) S (B) S (A + B) 0,
simply because S (A B) = S (A) S (B). This results in a long exact sequence
of homology groups. But Hn (S (A + B)) = Hn (X) by the locality theorem, and
hence the long exact sequence has the form claimed.

Exercise 12.1: Show that the connecting map can be understood as follows. Take
a p-cycle z Sp (X). By locality, there is a homologous p-cycle z 0 = x + y with x a
chain in A and y a chain in B. Then x = y, hence x is a cycle in A B. We
have [z] = [x].
Exercise 12.2: Show that the MayerVietoris sequence is not merely canonical, but also
natural in the following sense. Given an another excisive triad (X 0 ; A0 , B 0 ) and a map
f : X X 0 such that f (A) A0 and f (B) B 0 , the two long exact sequences and
the maps between them induced by f form a commutative diagram.
Example 12.2. As a first example of the MayerVietoris sequence, let us prove
that
H (S 1 ) = Z Z
where the first Z is in degree 0, the second Z in degree 1. We have S 1 = A B
where A = S 1 \ {(1, 0)} and B = S 1 \ {(1, 0)}. Then A B ' S 0 . Since A and B
are contractible, and A B the disjoint union of two contractible components, the
exactness of the MayerVietoris sequence
Hp (A) Hp (B) Hp (S 1 ) Hp1 (A B),
tells us that Hp (S 1 ) = 0 for all p > 1. We already know H1 (S 1 ) = Z = H0 (S 1 ) (via
1 and path-connectedness), but lets see that we can recover this by the present
method. The sequence ends with the 6-term sequence
0 H1 (S 1 ) Z2 Z2 H0 (S 1 ) 0,
where the map Z2 = H0 (A B) H0 (A) H0 (B) = Z2 is given by (m, n) 7
(m n, m n). Thus H1 (S 1 ) is isomorphic to the kernel of this map, which is
Z(1, 1), and H0 (S 1 ) to its cokernel, which is also Z.
40
TIM PERUTZ
Proposition 12.3. We have

H (S n )
= Z Z,
n 0,
where the first Z is in degree 0, the second Z in degree n.

Proof. By induction on n. Since S 0 is a 2-point space, its true for that case. Weve
just proved it for n = 1, so well start the induction there.
So now assume n > 1. We have S n = A B where A = S n \ {N } and B =
n
S \ {S}, N and S being the north and south poles. Then A B ' S n1 . Since A
and B are contractible, MayerVietoris tells us that Hp (A) Hp (B) = 0 for p > 0.
Thus, from the exactness of
Hp (A) Hp (B) Hp (S n ) Hp1 (A B) Hp1 (A) Hp1 (B),
we see that Hp (S n )
= Hp1 (S n1 ) for all p > 1. We also have an exact sequence
0 H1 (S 1 ) Z Z2 ,
where the map Z = H0 (A B) H0 (A) H0 (B) = Z2 is n 7 (n, n), and so is
injective. Hence H1 (S n ) = 0 (which we knew anyway, S n being simply connected.)

Remark. The argument can be made slicker using reduced homology.
Note that, in all dimensions (even n = 1) the connecting map n : Hn (S n )
Hn1 (S n ) is an isomorphism.
Exercise 12.3: Describe an n-cycle cn Sn (S n ) that generates Hn (S n ). (Prove that
your cycle generates by induction, looking at the explicit form of the connecting map.)
We now take a break from the formal development of the theory and harvest
some applications.
Theorem 12.4. If Rn is homeomorphic to Rm then m = n.
Rn then Rm \ {0}
Proof. We may assume that m and n are positive. If Rm =
=
n
n
n1
R \ {pt.}
is homotopy equivalent to S m1 . But S n1 and
= R \ {0}, hence S
S m1 have different homology in degree n 1, unless m = n.

Theorem 12.5 (Brouwer fixed point theorem). Let Dn denote the closed n-ball in
Rn , n 1. Then every continuous map Dn Dn fixes a point.
Proof. Recall that we already proved this using 1 , when n = 2. By the same
argument that we gave there, it suffices to show there is no retraction r : Dn
Dn = S n1 , i.e. no r such that r i = idS n1 , where i : S n1 Dn denotes the
inclusion map. If such an r existed, we would have
idHn1 (X) = Hn1 (r) Hn1 (i).
But Hn1 (S n1 ) 6= 0, whereas Hn1 (Dn ) (the target of Hn1 (i)) is zero, assuming
n > 1. This proves the theorem (except when n = 1; but in that case it follows
from the intermediate value theorem).
41
12.2. Degree. Define the degree deg(f ) of a map f : S n S n by the equation

Hn (f )(m) = deg(f )m, m Hn (S n )
= Z.
The following properties of degree follow from the definition and the basic properties
of homology:
(1) Homotopic maps have the same degree.
(2) deg(idS n ) = 1.
(3) deg f = 0 when f extends to a map Dn+1 S n .
(4) deg(f g) = deg(f ) deg(g).
You will find solutions to the following exercise in Hatcher, but I recommend trying
to work it out for yourself.
Exercise 12.4: Degree has the following further properties:
(i) When n = 1, the homological degree defined here coincides with the degree as
we defined it earlier via 1 (S 1 ).
(ii) deg s = 1 when s is the restriction to S n of a reflection in Rn+1 .
(iii) deg a = (1)n+1 when a is the antipodal map x 7 x.
(iv) deg f = (1)n+1 when f has no fixed points.
Use the last point to show (a) that no group of order > 2 can act freely on S 2m , and
(b) that S 2m possesses no nowhere-vanishing vector field.
Exercise 12.5: Complex projective n-space CP n is defined as (Cn+1 \ {0})/C , where
C acts by scalar multiplication. Use MayerVietoris to show that
(
Z, if p is even and 0 p 2n,
Hp (CP n )
=
0, else.
Show, moreover, that the inclusion of CP n in CP n+1 , induced by the inclusion of Cn+1
as a linear subspace of Cn+2 , induces an isomorphism on homology up to degree 2n.
Exercise 12.6: Use MayerVietoris to compute H (T 2 ), H (RP 2 ) and H (K 2 ). In
each case, find an explicit (minimal) collection of cycles that span the homology.
Exercise 12.7: Use MayerVietoris to compute H (g ) where g is a 2-sphere with g
handles attached.
42
TIM PERUTZ
13. Relative homology and excision

13.1. Relative homology. Suppose i : A X is the inclusion of a subspace.
There is a short exact sequence of chain complexes
i#
0 S (A) S (X) S (X, A) 0,

where S (X, A) = S (X)/i# S (A). We denote by H (X, A) the homology of
S (X, A). Thus a chain in X defines a cycle in S (X, A) if its boundary is a
sum of simplices in A. A map of pairs f : (X, A) (Y, B) induces f : H (X, A)
H (Y, B).
The short exact sequence for the pair (X, A) induces the long exact sequence of
the pair
i
Hp (X) Hp (X, A) Hp1 (A) . . . .

Hp (A)
We can understand the connecting map as follows. If c Sp (X, A) is a cycle, [c]

is defined by lifting c to a chain c in Sp (X), and putting [c] = [i (p c)]. In other
words, [c] is the boundary of a chain representing c, considered as a cycle in A.
Exercise 13.1: Show that if B A X, one has a long exact sequence of the triple
Hi (A, B) Hi (X, B) Hi (X, A) Hi1 (A, B) . . . ,
and describe the maps in this sequence.
We now state the excision theorem. It has essentially the same content as the
MayerVietoris sequence.
Theorem 13.1 (Excision). Suppose (X; A, B) is an excisive triad. Then the map
H (A, A B) H (X, B)
induced by the inclusion (A, A B) (X, B) is an isomorphism.
Proof. We invoke the locality theorem for chains. Note that H (A, A B) is the
homology of the complex S (A)/S (A B) = S (A)/(S (A) S (B)) (where we
think of S (A) and S (B) as subcomplexes of S (X)). By a general property of
abelian groups, the inclusion of S (A) in S (A) + S (B) induces an isomorphism
S (A)/(S (A) S (B))
= (S (A) + S (B))/S (B).
But S (A)+S (B) = S (A+B) by definition of the latter, and the map S (A+B)
S (X) is a quasi-isomorphism. Let In = Hn (S (A + B)/S (B)). Then we have a
commutative diagram with exact rows
Hn (B) Hn (S (A + B))
=y
=y
Hn (B)
Hn (X)
In
Hn1 (B) Hn1 (S (A + B))
=y
=y
Hn (X, B) Hn1 (B)
Hn1 (X)
All the vertical arrows but the middle one are isomorphisms, and hence so is
the middle one (this is the 5-lemma). Putting things together, we find that
S (A)/S (A B) S (X)/S (B) is a quasi-isomorphism, which is the result we
want.

Exercise 13.2: Re-derive H (S n )
= Z(0) Z(n) using excision.
43
Example 13.2. When working with spaces X equipped with basepoints x, its
. We put
often useful to work with reduced homology H
n (X) = Hn (X, {x}).
H
Then, by the exact sequence of the pair, the natural map
n (X)
Hn (X) H
is an isomorphism for all n > 0. The end of the sequence looks like this:
H1 (X, {x}) H0 ({x}) H0 (X) H0 (X, {x}) 0.
1 (X) = H1 (X)), so
The map H0 ({x}) H0 (X) is injective (which explains why H
H0 (X) = H0 (X)/H0 ({x}). Thus, when X is path connected, we have

(
n=0
n (X) = 0,
H
Hn (X), n > 0.
This justifies the omission of x from the notation (when X is not path connected,
x is important).
Exercise 13.3: Let A be a non-empty closed subspace of X. There is a natural homo (X/A). Use excision to show that, when
morphism H (X, A) H (X/A, A/A) = H
A has a neighbourhood N that deformation-retracts to A, this map is an isomorphism.
Exercise 13.4: Show that if (X; A, B) is an excisive triad, and b A B a basepoint,
there is an exact MayerVietoris sequence in reduced homology (formally the same as
instead of H ).
the usual one, but with H
13.2. Suspension. We now set up the suspension isomorphisms. The (unreduced)
suspension SX of X is the space (X [1, 1])/ where (x, t) (x0 , t0 ) if t = t0 = 1
or t = t0 = 1. For example, S(S n )
= S n+1 .
Proposition 13.3. Fix a basepoint x X. There are natural isomorphisms
n+1 (SX) H
n (X), n 0.
sn : H
Proof. We have SX = C+ C , where
1
1
C+ = (X [ , 1])/(X {1}), C = (X [1, ])/(X {1}).
2
2
Both C+ and C are cones on X; they deformation retract to their respective cone
points c = (X {1})/(X {1}). Thus
n (SX).
Hn (SX, C+ ) = Hn (SX, {c+ }) = H
On the other hand (X; C+ , C ) is an excisive triad, so
Hn (SX, C+ )
= Hn (C , C+ C ).
Since C is contractible, the long exact sequence of the pair tells us that
Hn (C , C+ C ) Hn1 (C+ C ) = Hn1 (X)
0 (X). Putting this
is an isomorphism when n > 1, and that H1 (C , C+ C ) = H
together, we get isomorphisms
n (SX)
n1 (X).
H
= Hn (SX, C+ )
= Hn (C , C+ C )
=H
44
TIM PERUTZ
Exercise 13.5: Show that the two spaces S 2 S 4 and CP 2 have isomorphic homology
groups. Likewise the two spaces S 3 S 5 and S(CP 2 ).
Remark. We will see later that the homotopy types of S 2 S 4 and CP 2 can be distinguished by their cohomology rings, but that S 3 S 5 and S(CP 2 ) have isomorphic
cohomology rings.
13.3. Summary of the properties of relative homology.
To each pair of spaces (X, A), and each integer n, it assigns an abelian
group Hn (X, A) (and we write Hn (X) for Hn (X, )).
To each map f : (X, A) (X 0 , A0 ) and each n Z it assigns a homomorphism Hn (f ) : Hn (X, A) Hn (X 0 , A0 ). One has Hn (f g) = Hn (f )Hn (g)
and Hn (id(X,A) ) = idHn (X,A) . If f0 is homotopic to f1 via a homotopy {ft }
such that ft |A = f0 |A , then Hn (f1 ) = Hn (f0 ).
To each pair of spaces (X, A), and each integer n, it assigns a homomorphism
n : Hn (X, A) Hn1 (A).
These maps are natural transformations. That is, given f : (X, A)
(X 0 , A0 ), one has
n Hn (f ) = Hn1 (f ) n
as homomorphisms Hn (X, A) Hn1 (A0 ).
Besides these basic properties, the following also hold:
DIMENSION: If denotes a 1-point space then Hn () = 0 for n 6= 0, while
H0 () = Z.
EXACTNESS: The sequence
n
Hn (A) Hn (X) Hn (X, A)
Hn1 (A) Hn1 (A) . . .
is exact, where the unlabelled maps are induced by the inclusions (A, )
(X, ) and (X, ) (X, A).
EXCISION: if (X; A, B) is an excisive triad then the map
Hn (A, A B) Hn (X, B)
ADDITIVITY: If (X , A ) is a family of pairs, then one has an isomorphism
M
a
a
Hn (X , A ) Hn ( X ,
A )
given by the sum of the maps induced by the inclusions into the disjoint
union.
(Additivity is easy to check; for finite families, it follows from excision and exactness.)
Remark. As stated, these axioms do not uniquely characterise relative homology.
However, they do characterise it if one restricts the pairs (X, A) to be CW pairs.
Alternatively, if we include one more axiom, that weak equivalences induce isomorphisms on homology, then the axioms uniquely characterise the theory for arbitrary
pairs, because every pair is weakly equivalent to a CW pair. If one omits the dimension axiom, there are many different homology theories on CW pairs, including
stable homotopy, real or complex K-theory, and oriented, unoriented or complex
bordism.
45
14. Vanishing theorems for homology of manifolds

Theorem 14.1. Let M be an n-manifold, i.e., a space locally homeomorphic to
Rn . Then Hp (M ) = 0 for p > n.
The proof will use MayerVietoris sequences to increase the generality incrementally, starting from a very banal statement. So as to avoid low-dimensional
exceptions, we use MayerVietoris for reduced homology. However, we can and
shall assume throughout that n > 0.
Proof. Step 1: If n 1 and X is the union of strictly less than 2n of the faces of
p (X) = 0 for p n 1.
the n-cube I n , then H
This we prove by induction on n; the n = 1 case is clear. The induction step
is done by a further induction, on the number of faces of X. When X is empty
it it trivial. Otherwise, choose a face F of I n which is contained in X, and let
X 0 = X \ int(F ). Choose F so that not all its neighbours are in X. MayerVietoris
(and a few deformation retractions of open neighbourhoods) then gives an exact
seqence
p (F ) H
p (X 0 ) H
p (X) H
p1 (F X 0 ).
H
But F X 0 is a union of strictly less than 2(n 1) of the faces of the (n 1)-cube
p1 (F X 0 ) = 0. Trivially H
p (F ) = 0, and by induction on the
F , so inductively H
0
number of faces, Hp (X ) = 0. This completes the induction.

S
Step 2: If C1 , . . . , Cq are integer cubes in Rn then Hp ( i Ci ) = 0 for p n.
Here an integer cube C in Rn means a translate by some (m1 , . . . , mn ) Zn
of I n . We prove this by induction on the number of cubes q. A single cube is
contractible. For the induction step, observe that one of the cubes, which we may
take to be Cq , has maximal x1 -coordinate. It then intersects less than 2n of its
neighbours. Write Dq = C1 Cq . We then have a MayerVietoris sequence
p1 (Dq1 Cq ).
Hp (Cq ) Hp (Dq1 ) Hp (Dq ) H
By induction, the left-hand term is zero for p n, so it suffices to show that
Hp1 (Dq1 Cq ) = 0 also. But this follows from Step 1.
n
Step 3: Let
P U R be an open set. Then Hp (U ) = 0 for pS n.
Let z =
ai i be a p-cycle, and let K denote the union im i . Note that K
is compact. Thus we can find > 0 and a finite collection of integer cubes Cj such
that the interiors of the small cubes Cj cover K, but the union L of the Cj is
contained in U . Then z represents a p-cycle in L. But Hp (L) = 0 by Step 2; so z
is a boundary in L, and hence in U .
Step 4: For
0 when p > n.
P an n-manifold M , we have Hp (M ) = S
Let z =
ai i be a p-cycle in M , and let K = im i . Again it is compact,
and hence covered by finitely many open sets homeomorphic to Rn . We prove by
induction on N that a p-cycle
z contained in N open sets V1 , . . . , VN homeomorphic
S
to Rn is a boundary in Vi (hence in M ). The case N = 1 was Step 3. For the
induction step, assume N > 1 and let U = V1 VN 1 . MayerVietoris gives
an exact sequence
Hp (U ) Hp (VN ) Hp (U VN ) Hp1 (U VN ).
By induction, the left term vanishes. The term on the right vanishes by Step 3.
Hence the middle term also vanishes.
46
TIM PERUTZ
14.1. Local homology. The local homology of a space M at x M is defined as

H (M, M \ {x}).
Lemma 14.2. When M is an n-manifold, one has H (M, M \ {x})
= Z(n) .
Proof. Choose a neighbourhood D of x homeomorphic to a closed n-ball. By excision, H (M, M \ {x})
= H (D, D \ {x}). One finds from the long exact sequence of
p1 (D {x}) for all p. But D \ {x} deformationthe pair that Hp (D, D \ {x})
=H

retracts to S n1 , and H (S n1 )
= Z(p1) , whence the result.
We can form a natural covering space p : HM M whose fiber over x is
Hn (M, M \ {x}). Thus a point in HM is a pair (x, hx ) where x M and hx
Hn (M, M \ {x}), and p(x, hx ) = x. The topology is defined as follows. Let U be
an open set in X homeomorphic to Rn , and let B U be a closed neighbourhood
homeomorphic to the closed disc Dn . Then one has isomorphisms
Hn (M, M \ {y}) Hn (M, M \ B) Hn (M, M \ {x}).
Define a component of p1 (U ) to be an equivalence class of pairs (x, hx ) p1 (U ),
where (x, hx ) is equivalent to (y, hy ) if there is a ball B containing both x and y and
an element z Hn (M, M \ B) mapped by the above isomorphisms to hx and hy .
The components of p1 (U ), as U varies over open neighbourhoods homeomorphic
to Rn , form a basis for a topology making p a covering map.
14.2. Homology in dimension n.
Theorem 14.3. Hn (M ) = 0 when M is a connected but non-compact n-manifold.
We have already established a special case of this, the case where M is an open
set in Rn . The principle of the proof is standard, but the details here are from
Mays book. The proof is more technical than one would ideally like, but there is
only really new idea:
An n-dimensional homology class z determines a section sz of p : HM M , and
z = 0 if sz (x) = 0 for some x.
Lemma 14.4. Let U Rn be open. Then the natural homomorphism
Y
Hn (Rn , U )
Hn (Rn , Rn \ {x})
xRn \U
is injective. Equivalently, the product of maps induced by inclusions,

Y
n1 (U )
n1 (Rn \ {x}),
H
H
xRn \U
is injective.
The equivalence of the two assertions follows from the long exact sequences of
the pairs (Rn , U ) and (Rn , Rn \ {x}). Informally, the second assertion says that any
non-trivial (n 1)-cycle in U must wrap around some point outside U . We prove
the second assertion.
n1 (U ) maps to zero in H
n (Rn \ {x}) for all x
Proof. Suppose s H
/ U . We will
show that s = 0. The notation will be rather heavy: we fix a chain of subspaces
K V V U
47
n1 (K).
where K is compact, V open, V compact, and s is the image of some r H
n
Let T be a cube (d, d) large enough that V T . We discard everything outside

n1 (T U ), hence also in H
n1 (U ). We
T : we shall show that r maps to zero in H
n1 (T \ {x}) for all x T \ U . Our method will be
know that r maps to zero in H
n1 (T \ S)
to eat away more and more of T \ U , showing that r maps to zero in H
for progressively larger subsets E T \ U , starting at E = {x} and ending at
E = T \ U itself. Since T \ (T \ U ) = T U , this will do the job.
Now, T \ U is covered by a grid of small compact cubes in Rn , whose diameter is
chosen small enough that none of them hits V . Let C1 , . . . , Cq be the intersections
n1 (T \Ep ) for p q,
of these small cubes with T . We claim that r maps to zero in H
where Ep = C1 Cp . Indeed, the case p = 0 is trivial, and for p > 0 we have
T \Ep = (T \Ep1 )(Rn \Ep ). On the other hand, (T \Ep1 )(Rn \Ep ) = Rn \Ep .
MayerVietoris gives
n1 (T \ Ep ) H
n1 (T \ Ep1 ) H
n1 (Rn \ Ep ).
Hn (Rn \ Ep ) = 0 H
Here the group on the left is zero by Step 3 in the proof of the earlier theorem.
n1 (T \ Ep1 ) H
n1 (Rn \ Ep ), it must then map to zero
Since r maps to zero in H
in Hn1 (T \ Ep ).

Proof of the theorem. Any z Hn (M ) defines a continuous section sz : M HM .
If M is connected (hence path-connected), it follows from unique path-lifting that
any continuous section is determined by its value at a point. But z is represented by
a cycle that maps to a compact subset Z M . If M is connected but non-compact,
we can choosing x M \ Z. Then sz (x) = 0, and hence sz = 0.
Thus we take a class z Hn (M ) that maps to zero in some Hn (M, M \ {x}).
We must show that z = 0.
There is some compact set Z M such that z is in the image of Hn (Z). Now,
Z is contained in a finite union U1 Uq of coordinate neighbourhoods Ui ,
and it suffices to show [z] = 0 in Hn (U1 Uq ). We have already proved that
Hn (U1 ) = 0. Now take q > 1, and inductively suppose that weve shown that
Hn (U1 Uq1 ) = 0. MayerVietoris gives an exact sequence
n1 (U V ) H
n1 (U ) H
n1 (V ),
Hn (U ) Hn (V ) Hn (U1 Uq ) H
where V = U1 Uq1 and U = Uq . This reduces to
n1 (U V ) H
n1 (V ),
0 Hn (U1 Uq ) H
n1 (U V ) H
n1 (V ) induced by
and so the task is to show that the map j : H
inclusion is injective. This step is a little tricky. Take r ker j . From the long
exact sequence of the pair (V, U V ),
n1 (U V ) H
n1 (V ),
0 Hn (V, U V ) H
we find a (unique) t Hn (V, U V ) such that t = r. From the long exact sequence
of the pair (U, U V ),
n1 (U V ) 0,
0 Hn (U, U V ) H
we find a (unique) s Hn (U, U V ) with s = r. Now, s and t have images
s0 and t0 in a common group Hn (U V, U V ), and s0 t0 lies in the kernel of
n1 (V ), hence in the image of Hn (U V ): say s0 t0 comes
: Hn (U V, U V ) H
from w Hn (U V ). Since U V is non-compact, the composite Hn (U V )
Hn (M ) Hn (M, M \ {x}) maps w to zero (by the argument at the beginning of
48
TIM PERUTZ
this proof). Take x outside U V . Then we have a map Hn (U V, U V )

Hn (M, M \ {x}) factoring Hn (U V ) Hn (M, M \ {x}) which therefore carries
t0 s0 to zero. Moreover, t0 maps to zero in Hn (M, M \ {x}), because it comes from
t Hn (V, U V ), and the map Hn (V, U V ) Hn (M, M \ {x}) factors through
Hn (M \ {x}, M \ {x}) = 0. Hence s0 maps to zero in Hn (M, M \ {x}). It follows
that s maps to zero in Hn (U, U \ {c})
= Hn (M, M \ {c}).
Since U
= Rn , we have s Hn (Rn , Rn \V ) and s maps to zero in Hn (Rn , Rn \{y})
for every y lying outside V . We wish to show that s is zero (so that r = s = 0).
But that is what we proved in the last lemma.

Exercise 14.1: (a) Let X be a path connected space. Show how 2-sheeted covering
spaces of X, up to isomorphism, correspond to homomorphisms 1 (X, x) Z/2.
HM be the subspace consisting of pairs (x, hx ) with hx a generator
(b) Let M
M is a 2-sheeted covering space. The
for Hn (M, M \ {x})
= Z. Thus p : M
corresponding homomorphism 1 (M ) Z/2 is called the orientation character or first
StiefelWhitney class w1 (M ) of M . Compute it for X the Klein bottle and for RP 2 .
49
15. Orientations and fundamental classes

15.1. Homology with coefficients. Homology with coefficients is a simple generalisation of singular homology.
Let R be a commutative unital ring. The case we have been considering up to
this point is R = Z, but we could also take, for example, a field such as Fp = Z/p
(where p is prime) or Q.
Define a chain complex of R-modules S (X; R) by defining Sn (X; R) as the free
R-module generated by the n-simplices in X:
Sn (X; R) = Rn (X)
= Sn (X) Z R.
Define the differential on simplices by the familiar formula, and in general by
R-linearity. Let Hn (X; R) = Hn (S (X; R)), as an R-module.
If A X is a subspace then one has an injective chain map S (A; R) S (X; R).
Put
S (X, A; R) = S (X; R)/S (A; R),
Hn (X, A; R) = Hn (S (X, A; R)).
Then one has a long exact sequence of the pair. Locality, MayerVietoris and
excision hold just as before, as does homotopy invariance. One has H0 (X; R) =
R0 (X) . The homology of a point, Hi (; R), vanishes for i > 0. For an n-manifold
M , one has Hi (M ; R) = 0 for all i > n and, in the non-compact case, for i = n. All
these assertions follow from routine generalisations of the proofs we have given.
Slightly trickier is the assertion that, for X path connected, one has
1 (X)ab Z R.
H1 (X; R) =
The proof we gave for the case R = Z uses the integer coefficients in a significant
way, but the general case follows from it and the universal coefficients theorem, to
be given in the next lecture. (If you have a simpler proof, tell me!)
15.2. What its good for. Homology with coefficients contains no more information than ordinary homology, but is useful for the following reasons.
When R is a field, algebraic properties of homology simplify. For instance,
over a field, passing to homology commutes with the tensor product and
dualisation operations on chain complexes.
For many spaces, the Z-homology is more complicated than the homology
over the prime fields Z/p or over Q. Moreover, considering these collectively,
one does not lose information. Good examples are RP n and various Lie
groups, notably SO(n).
When working with manifolds, in Z-homology one must distinguish the orientable and non-orientable cases, but in Z/2-homology this is unnecessary.
15.3. The local homology cover. Recall from the last lecture that the local
homology of an n-manifold (now with R-coefficients) is given by
H (M, M \ {x}; R)
= R(n) .
Collectively, these form a covering space HM,R M with fibres Hn (M, M \{x}; R).
(Here one gives Hn (M, M \ {x}; R) the discrete topology.) The sections M,R of
this covering space form an R-module. A class z Hn (M ; R) gives classes zx
Hn (M, M \ {x}; R) for all x, and this determines a homomorphism Hn (M ; R)
M,R .
50
TIM PERUTZ
Theorem 15.1. The the natural homomorphism Hn (M ; R) M,R is injective.

Proof. We may assume M is connected. Fix x M , and observe that by our
vanishing theorem for homology of non-compact manifolds, Hn (M \ {x}) = 0. The
long exact sequence of the pair therefore reads
0 Hn (M ) Hn (M, M \ {x}).
That is, the homomorphism Hn (M ) Hn (M, M \ {x}) is injective. Our vanishing
theorem was proved for Z coefficients, but the proof applies to an arbitrary coefficient ring. Since it injects into Hn (M, M \ {x}), a fortiori Hn (M ; R) injects into
M,R .

15.4. Orientations. Depending on the ground ring R, there may be more than one
isomorphism of R-modules Hn (M, M \{x}; R)
= R: for instance, when R = Z, there
are exactly two isomorphisms (the one you first thought of, and its composition with
n 7 n). However, when R = Z/2, there is a unique isomorphism.
Definition 15.2. An R-orientation for the n-manifold M at x is an isomorphism
of R-modules
: Hn (M, M \ {x}; R) R.
If R is not specified, we understand R = Z.
Define the R-orientation cover
R = {(x, x ) : x M and is an R -orientation for M at x},
M
M be the projection (x, x ) 7 x. The topology is constructed in
and let pR : M
R is itself
much the same way as that of HM (work this through!). Notice that M
an n-manifold. When R = Z, it is a 2-sheeted covering of M . Assuming M is
Z is either connected, hence a non-trivial covering, or disconnected,
connected, M
hence(!) trivial.
Definition 15.3. An R-orientation for M is a section of p, i.e., a map s : M M

such that pR s = idM . If an orientation exists, M is called R-orientable. Again,
if no ring is specified we understand R = Z.
Thus an R-orientation for M is a coherent choice of orientations for all points.
Observe that there is automatically a unique (Z/2)-orientation.
The orientation cover is closely related to the local homology cover HM,R M .
We illustrate this with the case R = Z. If a Z then the subset
(a) := {(x, hx ) HM,R : x (hx ) = a for some orientation x }
M
(a) = M
(a). If a 6= 0 then the covering M
(a) M is isomorphic to
then M
MZ M , while if a = 0 it is a trivial covering M M .

15.5. Fundamental classes.
Definition 15.4. (i) If M is an n-manifold, an R-fundamental class for M is a
class z Hn (M ; R) whose image in Hn (M, M \ {x}; R) is a generator, for every
x M.
(ii) More generally, if K M is a subspace, an R-fundamental class for M at K
is a class z Hn (M, M \ K; R) whose image in Hn (M, M \ {x}; R) is a generator,
for every x K.
51
Theorem 15.5. An R-orientation for M determines an R-fundamental class zK at

K, for every compact subspace K M . The orientation maps zK (x) Hn (M, M \
{x}) to 1 R.
Proof. We drop the coefficients from the notation. It is technically easier to prove,
alongside the theorem, that Hp (M, M \ K) = 0 for p > n.
Start by supposing K U , with U
= Rn . Notice that by excision and a
deformation retraction, Hn (M, M \ U ) = Hn (Rn , Rn \ {0}). Thus an orientation

determines a fundamental class for M at U , and hence at K. Also, in this case
Hp (M, M \ K) = 0 for p > n by the long exact sequence of the pair and our
vanishing theorem from last time.
A general compact subset K M , is the union of finitely many compact subsets
Ki , each contained in a neighbourhood Ui
= Rn . So, by induction, it suffices to
show that if the theorem (including the statement about p > n) holds for K, L and
K L then it holds for K L.
To prove this, we need a relative form of MayerVietoris. If Y X, and (Y ; A, B)
is an excisive triad, then one has a long exact sequence
Hp (X, X \ (A B)) Hp (X, X \ A) Hp (X, X \ B) Hp (X, A B) . . . .
In our case we get
Hn+1 (M, M \ (K L)) Hn (M, M \ (K L)) Hn (M, M \ K) Hn (M, M \ L).
By hypothesis, the module on the left is zero, but we have fundamental classes
zK Hn (M, M \ (K L)) and zL Hn (M, M \ (K L)) determined by the
orientation. Their difference, in Hn (M, M \ (K L), is zero, and hence zK + zL
comes from a class zKL Hn (M, M \ (K L)) which is clearly a fundamental
class.

Exercise 15.1: Use the locality theorem for chains to derive this relative MayerVietoris
sequence.
Corollary 15.6. If M is compact, connected and R-oriented then Hn (M ; R) = R.
Hence any compact, connected n-manifold M has Hn (M ; Z/2)
= Z/2. If it is also
orientable then Hn (M )
= Z.
Proof. We know by the last theorem that a fundamental class exists, so the homomorphism
Hn (M ; R) M,R
is onto. We observed earlier that it is injective. Hence it is an isomorphism. But
M,R

= R, and an orientation specifies such an isomorphism.
Corollary 15.7. If M is connected but not orientable then Hn (M ) = 0.
Proof. The natural map Hn (M ) M is injective. If it had non-trivial image, the
(a) M with a 6= 0. But
image would contain a section of one of the covers M
these covers are copies of M M , and so have no sections by hypothesis.

Exercise 15.2: Convince yourself that the definition of an orientation at x is not crazy.
More specifically, take R = Z and M to be an n-dimensional vector space V . An
orientation for V in a more familiar sense would be an isomorphism V Rn , where
two isomorphisms 1 and 2 are considered equivalent if det( 21 ) > 0. Show how
an orientation for the vector space V determines an orientation for V as a manifold, at
any chosen point x.
52
TIM PERUTZ
Exercise 15.3: If M is given a smooth structure then it is orientable iff there is a smooth
atlas whose transition functions ij have positive Jacobian determinants det Dx ij .
Exercise 15.4: A manifold M is orientable if it is simply connected, or more generally,
if 1 (M ) is finite and has odd order.
Exercise 15.5: If p is odd then a Z/p-orientation determines, and is determined by, a
Z-orientation.
Exercise 15.6: Show that if a group G acts freely on M = S n , then the orientation
character 1 (M/G) = G Z/2 of the quotient manifold is given by G 3 g 7
deg(g) {1} = Z/2. For which n is RP n orientable?
Exercise 15.7: (i) Show that there is no homeomorphism Hn
= Rn , where Hn =
n
{(x1 , . . . , xn ) R : x1 0}.
(ii) Let M be a space, and N the subspace of points x M which have a neighbourhood U such that there is a homeomorphism U Hn sending x to 0. Say M is
an n-manifold with boundary if M \ N is an n-manifold. In that case write M for
the boundary N . Check that that M is an (n 1)-manifold.
Exercise 15.8: Suppose M is a compact, connected n-manifold with non-empty boundary M . It is a fact (see Hatcher) that there is a neighbourhood U of M and a
homeomorphism U M [0, 1) sending any x M to (x, 0).
(a) Show that Hp (M ) = 0 for p n and Hp (M, M ) = 0 for p > n.
(b) Show that if M \ M is orientable then Hn (M, M )
= Z.
(c) Let X be a space. Suppose h Hn1 (X) is a homology class that is representable by a manifold, in that there is a compact oriented n 1-manifold N with
fundamental class zN and a map f : N X with f zN = h. Show that h = 0 if
N = M for a compact oriented n-manifold with M = N such that f extends to
F : M X.
Exercise 15.9: Let h : S 3 S 2 denote the Hopf map, given by
h(z1 , z2 ) 7 z1 /z2 .
3
Here we think of S as the unit sphere in C2 and of S 2 as the Riemann sphere C{}.
Let be a generator for H3 (S 3 ) = Z. Then h = 0, since H3 (S 2 ) = 0. Can you find
a direct explanation for why the particular class h should be zero? In other words,
can you find a 4-chain bounding a 3-cycle representing h ?
53
16. Universal coefficients

In this lecture, we address the following question. Let C be a chain complex over
R, and Q an R-module. How does one compute H(C R Q) in terms of H(C )?
We obtain a solution when R is a principal ideal domain (PID) and C is a free
R-module.
16.1. Homology with coefficients. Last time, we introduced homology H (X; k)
with coefficients in a ring k. This was defined as the homology of the complex of
k-modules
k (X) = S (X) Z k.
What is its relation to H (X)? The answer is supplied by the universal coefficients
theorem:
Theorem 16.1 (Universal coefficients: homology version). There are short exact
sequences of k-modules,
0 Hn (X) k Hn (X; k) TorZ (Hn1 (X), k) 0,
Hn (X) TorZ (Hn1 (X), k),
natural in X. These sequence split, so Hn (X; k) =
but the splitting cannot be chosen naturally in X.
A first task will be to explain what Tor is.
16.2. Tor. Fix a commutative ring R (for instance Z) and an R-module Q. We
investigate the effect on exact sequences of the functor Q from R-modules to
R-modules. (Later we may suppose that Q is actually an R-algebra, and think of
Q as a functor from R-modules to Q-modules.)
i
Lemma 16.2. Let 0 M1 M2 M3 0 be a short exact sequence of Rmodules. Let Q be another R-module. Then the induced sequence
M1 Q M 2 Q M3 Q 0
is also exact. If the short exact sequence splits (for instance, if M3 is a free module)
then M1 Q M2 Q is injective.
Proof. Take m3 q M3 Q. Say q = p(m2 ). Then (p id)(m2 q) = m3 q.
Hence p id is onto.
Now, (p id) (i id) = (p i) id = 0, and hence there is an induced
map [p id] : (M2 Q)/ im(i id) M3 Q. Exactness of the sequence at
M2 Q is equivalent to the assertion that [p id] is injective. But im(i id) =
im i Q = ker p Q, and hence we can define s : M3 Q (M2 Q)/ im(i id)
by s(m q) = [p1 m q]. We have s [p id] = id, so [p id] is indeed injective.
Now suppose the sequence splits. Then there is a homomorphism l : M2 M2
with l i = id. Thus l id : M2 Q M1 Q satisfies (l id) (i id) = id id,
hence i id is injective.

In general, i id is not injective, but its kernel can be measured by the introduction of the torsion products. For simplicity, we assume R is Z, a field, or more
generally a principal ideal domain (PID). A PID is a commutative ring R such that
(i) xy = 0 implies x = 0 or y = 0, and (ii) every ideal is of form (x) = {ax : a R}.
The simplification in the present context because if R is a PID then every submodule
of a free R-module is free, i.e., has a basis. For the proof see, e.g., Langs Algebra,
54
TIM PERUTZ
Appendix 2 (in the general case, not restricted to finite rank free modules, this uses
Zorns lemma).
This has the following consequence: any R-module M has a two-step free resolution, i.e. an exact sequence
f1
f0
0 F1 F0 M 0
with F0 and F1 free. Just take any generating set S for M , let F0 be free on S, and
let F1 be the kernel of the natural map F0 M .
Applying Q to this sequence, we obtain a complex
f1 id
f0 id
0 F1 Q F0 Q M Q 0
which is exact at M Q and at F0 Q. We prefer to truncate this to the complex
f1 id
0 F1 Q F0 Q 0.
By the lemma, its homology at F0 Q (i.e., the cokernel of f1 id) is naturally
isomorphic to M Q. We define Tor(M, Q) to be its homology at F1 Q:
Tor(M, Q) = ker f1 id.
In brief: to define Tor, we took a free resolution of M , tensored it with Q, and
measured its failure to be exact by taking homology.
The next point is that Tor(M, Q) is independent of the choice of free resolution.
It is convenient to think of 0 F1 F0 as a chain complex F , and of f0 as an
augmentation (F M ) of the complex.
f0
g0
Lemma 16.3. Take two free resolutions (F M ) and (G M ). Then there is

a chain map : F G such that g0 = f0 . Moreover, is unique up to chain
homotopy.
The proof is left as an exercise.
Proposition 16.4. Tor(M, Q) is independent of the choice of two-step free resolution F M 0.
Proof. In the notation of the last lemma, we have to compare the homologies of the
complexes F Q and G Q. But the lemma supplies us with an augmentationpreserving chain maps : F G and : G F . Moreover, the uniqueness
clause tells us that is chain homotopic to the identity; likewise . Thus id
and id give chain-homotopy equivalences between F Q and G Q, showing
that they have canonically isomorphic homologies.

Example 16.5.
When R is a field we may take F0 = M and F1 = 0. Thus
Tor(M, Q) = 0, consistent with our first lemma.
If M is torsion-free over the PID R, we may take F0 = M have Tor(M, Q) =
0 for any Q and F1 = 0. Hence Tor(M, Q) = 0 for all Q.
x
The R-module M = R/(x) has the free resolution 0 R R R/(x)
0. Thus
Tor(R/(x), Q) = ker(x id : R Q R Q) = ker(x : Q Q),
i.e., Tor(R/(x), Q) is the x-torsion in Q. For instance, if p is prime then
Tor(R/(pn ), R/(p)) = R/(p).
55
16.3. Universal coefficients.

Theorem 16.6 (Universal coefficients). Let C be a chain complex of free Rmodules over a PID R. Then, for any R-module Q, one has short exact sequences
j
0 Hp (C ) Q Hp (C Q) Tor(Hp1 (C ), Q) 0,
where the map j is induced by the inclusion ker dp Q ker(dp idQ ). These
sequences are functorial in C . They always split, but not in a natural way.
Proof of universal coefficients. We let ip : Bp Zp be the inclusion of the pboundaries into the p-cycles. From the free resolution
ip
0 Bp Zp Hp (C ) 0
for the homology Hp (C ), we see that
coker(ip id) = Hp (C ) Q,
ker(ip1 id) = Tor(Hp1 (C ), Q).
Now consider the short exact sequence

dp
0 Zp Cp Bp1 0.
Since the terms are free modules, the sequence splits, and so the sequence
jp id
dp id
0 Zp Q Cp Q Bp1 Q 0
is also exact. It is, moreover, a short exact sequence of chain complexes, so it induces
a long exact sequence of homology groups. With a little thought one identifies the
connecting maps; the long exact sequence reads
ip id
Hp+1 (C Q) Bp Q Zp Q Hp (C Q) . . . ,
with ip : Bp Zp the inclusion. Its exactness tells us that there are short exact
sequences
0 coker(ip id) Hp (C Q) ker(ip1 id) 0,
i.e.,
0 Hp (C ) Q Hp (C Q) Tor(Hp1 (C ), Q) 0.
It is left as an exercise to think through why the map on the left is j. A splitting
dp
arises from a chosen splitting of 0 Zp Cp Bp1 0; again, this is left as

an exercise.
It is also left as an exercise to see that the whole sequence is functorial in C .
The non-naturality of the splittings will follow from a topological example in one
of the exercises below.

Notice that if Q is a commutative R-algebra, then Hp (C )Q and Tor(Hp1 (C ), Q)
are naturally Q-modules. And, rather obviously, the universal coefficients theorem
is valid as a statement about Q-modules. In particular taking R = Z and Q = k,
we obtain the topological universal coefficients theorem stated at the outset.
Exercise 16.1: Compute H (K), where K is the Klein bottle. Now compute H (K; Z/2n )
(i) by a direct argument, and (ii) using universal coefficients.
56
TIM PERUTZ
Exercise 16.2: Prove that the splittings of the universal coefficient theorem cannot
be chosen naturally in X by means of the following example. Take an embedding
i : R2 RP 2 , let D = i({z R2 : |z| < 1}), and let q be the quotient map
RP 2 RP 2 /(RP 2 \D)
= S 2 . Show that the k = Z/2 universal coefficients sequences
2
2
for RP and S cannot be split compatibly with q.
Exercise 16.3: This exercise develops an alternative approach to a special case of universal coefficients. Let C be a chain complex of free abelian groups. Show that
tensoring by the short exact sequence 0 Z Z Z/n 0 gives rise to a short
exact sequence of chain complexes 0 C C C Z/n 0. By analyzing the
resulting long exact sequence of homology groups, deduce a short exact sequence
0 Hp (C ) Z Z/n Hp (C Z/n) {x Hp1 (C ) : nx = 0} 0.
57
III. Cellular homology

17. CW complexes
We introduce a manageable category of spaces and maps, the category of CW
complexes and cellular maps.
CW complexes are a particularly convenient and useful class of spaces to work
with, for a number of reasons.
There is an efficient and geometrically clear way of computing the homology
groups of a CW complex.
Maps between CW complexes behave far better than maps between general
spaces: a weak homotopy equivalence between CW complexes is a homotopy
equivalence. A weak equivalence f : X Y is a map which induces bijections hTop(S n , X) hTop(S n , Y ), h 7 f h for all n (here hTop(S n , X)
denotes the set of homotopy class of maps).
One can localise the category hTop, artificially inverting the weak equivalences to form the topological derived category. Simplistically, performing
this localisation is equivalent to working with CW complexes. The precise
statement is that the topological derived category is equivalent to the homotopy category of CW spectra, see C. Weibel, Introduction to homological
algebra, Chapter 10.
On a smooth compact manifold M , a Morse function gives rise to a cell
structure making M a CW complex.
Definition 17.1. The n-cell en is the closed unit disc Dn Rn . For n > 0, its
boundary is en = S n1 . We put e0 = . If A is a space, and f : S n1 A a
map, we can build a space
Cf = en f A = (A q en )/
where f (x) x for x en .
We say that A f en results from attaching an n-cell to A via the attaching map f .
The notation Cf refers to the general notion of the homotopy cofiber Cf =
CX f Y of a map f : X Y , where CX = X I/X {1} is the cone on X.
Example 17.2. Let c : S n1 e0 be the constant map. Then S n
= Cc = en c e0 .
Two crucial observations are:
that Cf contains A as a closed subspace; and
that the quotient Cf /A is homeomorphic to en /en , hence to S n .
`
`
More generally, given an indexing set I and a map f = iI fi : iI en A,
we can build a space
a
Cf = (A q
en )/ where fi (x) x for x en .
i
The images of the fi need not be disjoint. We say Cf is obtained by attaching

n-cells to A.
W
Note that Cf /A
= iI S n .
Definition 17.3. A cell complex of dimension d is a space X together with a
sequence of closed subspaces
= X 1 X 0 X 1 X d = X
58
TIM PERUTZ
such that each X k+1 is obtained from X k by attaching a (possibly empty) collection
of n-cells. We call X k the k-skeleton of X. A cell complex is a space X together
with a sequence of closed subspaces
= X 1 X 0 X 1 X 2 . . .
such that X = k X k . A cell complex is a CW complex if X has the weak topology,
i.e. U X is open iff U X k is open in X k for each k.
S
(Im not sure how standard my usage of the term cell complex is here. However,
the term CW complex is certainly standard.)
Mostly we shall work with cell complexes with finitely many cells, which are
automatically CW complexes.
Remark. It is perhaps a pity that J. H. C. Whiteheads term CW complex, which
refers to certain technical properties, has not been replaced by something more
descriptive. W is for weak topology; C for closure finiteness, the property
that the image of each closed cell intersects the interiors of only finitely many cells
of lower dimension. In this as in other aspects of algebraic topology, there were
many possible variants of the definitionthe ones that we use lead to a streamlined
theory, and most of the others do not, but this is far from obvious.
Example 17.4. A graph is just a CW complex of dimension 1.
S
Example 17.5. The Hawaiian earring, i.e. the union n1 Cn of circles Cn in
R2 of radius n1 centered at n1 , is naturally a cell S
complex, but the topology
inherited from R2 does not make it a CW complex ( C2n \ {0} is open in the
weak topology but not the subspace topology).
Example 17.6. The 2-torus T 2 has the structure of a CW complex with one 0cell, one 1-cell (so the 1-skeleton is S 1 ) and one 2-cell. The same goes for the Klein
bottle K 2 . The real projective plane RP 2 has the structure of a CW complex with
two 0-cells, one 1-cell and one 2-cell.
Example 17.7. Complex projective space CP n = (Cn+1 \{0})/C is a cell complex
e2n e2n2 e0 . To see this, define
X 2k = {[z0 , . . . , zn ] CP n : zj = 0 for all j > k}.
Thus X 0 X 2 X 2n = CP n , and X 2k
= CP k . We will exhibit X 2k
as the 2k-skeleton of a cell decomposition. If [z] X 2k \ X 2(k1) CP n then
zk+1 = = zn = 0 but zk 6= 0, so [z] = [w1 , . . . , wk1 , 1, 0 . . . , 0] for a unique
(w0 , . . . , wk1 ) Ck . Thus X 2k \ X 2(k1)
= Ck . Thus CP n is a disjoint union of
open cells, one of each even dimension up to 2n. To see that they are attached in
the proper way, think of e2k as {w = (w0 , . . . , wk1 ) Ck : |w| 1} and define
i2k : e2k X 2k2 as follows:
ik (w) = (w0 , . . . , wk1 , (1 |w|2 )1/2 , 0, . . . , 0).
This map extends to a homeomorphism e2k fk X 2k2 X k which restricts to
the inclusion on X 2k2 , where fk : S 2k1 X 2k2 is given by fk (0 , . . . , k ) =
[0 , . . . , k , 0, . . . , 0].
Example 17.8. Let C[t] be the C-vector space of polynomials in t. Denote by CP
its projective space (C[t] \ {0})/C . Then CP is a CW complex with 2k-skeleton
(and 2k + 1-skeleton) CP k .
59
17.1. Compact generation. CW complexes are examples of compactly generated

spaces. A space X is compactly generated if (i) it is weak Hausdorff, meaning
that any compact Hausdorff subspace is closed, and (ii) every compactly closed
subspace is closed. Here a subspace A X is compactly closed if g 1 (A) is closed
in K for every compact Hausdorff space K and every map g : K X.
Theres a functor k from weak Hausdorff spaces to compactly generated spaces
(the closed sets of kX are the compactly closed sets of X). Algebraic topologists like
to work in the category of compactly generated spaces because compact generation
is a more intrinsic notion than that of a (space homotopy equivalent to a) CW
complex, and because this category has excellent properties. The quotient of a
compactly generated space by a closed equivalence relation is compactly generated.
The direct limit of compactly generated spaces is compactly generated. Define a
product X Y in this category to be k(X Y ) (k applied to the usual product);
and define Y X as kMap(X, Y ) where Map(X, Y ) has its compactopen topology.
Then the canonical bijection Z XY
= (Z Y )X is a homeomorphism. For instance,
a homotopy I X Y is the same thing as a map X Y I .
17.2. S
Degree matrices. A good deal of information about how a CW complex
X = k X k is built is encoded in its degree matrices Dn .
Suppose that for each n, one has a labelling of the n-cells as e1 , . . . , eN (n) . Define
an N (n 1) N (n) matrix Dn with integer entries as follows: the matrix entry
Dnij is equal to the degree of the map
j qn1 fi : S n1 S n1 ,
where fi : S n1 X n1 is the attaching map
ei ; qn1 : X n1 X n1 /X n2
W for
n1
n2
n1
the quotient map; and j : X
/X
= jS
S n1 the map which collapses all the spheres in the wedge except the jth.
Exercise 17.1: Compute degree matrices for the above-mentioned CW structures on
T 2 , RP 2 and K 2 .
Remark. The degree matrix has a geometric interpretation in Morse theory. Fix a
smooth Riemannian manifold (M, g). If f is a Morse function on M then the union
of the unstable manifolds (with respect to g) of the critical points is a subspace M 0
such that M 0 M is a homotopy equivalence. M 0 has a canonical cell decomposition in which the p-cells are the unstable manifolds of the index p critical points
p
for the
cip . If now cjp1 is an index p 1 critical point then the matrix entry Dij
cell decomposition is the signed count of downward gradient flow lines from cip to
cjp1 . (To make this count finite and meaningful, we perturb the pair (f, g) so that
it satisfies the MorseSmale transversality condition.)
17.3. Cellular approximation. The following theorem is not especially difficult
to prove, but we shall nonetheless omit the proof.
Theorem 17.9 (Cellular approximation of maps). Let f : X Y be a map between
CW complexes. Then f is homotopic to a cellular map, i.e. a map g such that
g(X k ) Y k for all k.
The next theorem is best proved using a little homotopy theory. Again we shall
omit the proof.
Theorem 17.10 (CW approximation of spaces). There is a procedure which assigns
to any space X a CW complex (X) and a weak homotopy equivalence X X.
60
TIM PERUTZ
Moreover, the procedure can be made functorial, in the sense that if f : X Y is

a map then there is a map f : X Y so that the obvious square commutes up
to homotopy.
This theorem has an important consequence. If one wants to prove a theorem
about the homology of general spaces, it suffices to prove it for CW complexes; by
cellular approximation of maps, one can also take the maps to be cellular.
61
18. Cellular homology

We compute the singular homology of cell complexes in terms of the cells and
the degree matrix obtained from their attaching maps.
Let X be a CW complex, and X k its k-skeleton; thus
[
= X 1 X0 X1 X2
X k = X.
k0
We wish to compute H (X) in terms of the cells and their attaching maps.
One of the convenient properties of CW complexes is that the subspace X n1
of X n has an open neighbourhood U which deformation-retracts to X n1 . Also
X n1 is closed in X n . Hence (this was an exercise using the excision theorem), one
(X n /X n1 ) when X n1 6= .
has H (X n , X n1 ) = H
Now, the idea of our calculation is to exploit the fact that X n /X n1 is a very
simple space:
_
_
Dn+1 /Dn+1 =
Sn.
X n /X n1
=
n-cells
n-cells
Thus Hn (X n , X n1 ) = Z{n-cells} .
Define n+1 as the connecting homomorphism
n+1
Hn+1 (X n+1 , X n ) Hn (X n )
from the exact sequence of the pair (X n+1 , X n ). Define qn as the natural map
qn
Hn (X n ) Hn (X n , X n1 ).
Theorem 18.1. Let Cn = Hn (X n , X n1 ) = Z{n-cells} . Define dn = qn1
n : Cn Cn1 . Then one has dn dn+1 = 0. The homology Hn (C ) of the
chain complex
dn+1
n
Cn+1 Cn
Cn1 . . .
is canonically isomorphic to the singular homology Hn (X).
This theorem is very useful as a computational tool, as we will see next time.
For now, we record two simple corollaries.
Corollary 18.2. If X is a d-dimensional CW complex then Hn (X) = 0 for all
n > d. If X is a CW complex with a finite number l of n-cells then the rank of
Hn (X) is finite and l.
Corollary 18.3. If X is a CW with only even-dimensional cells then Hn (X) =
Z{n-cells} for all n. In particular, H (CP n ) = Z(0) Z(2) Z(2n) , where the
subscripts denote degrees of the Z-summands.
Proof of the theorem. We will simplify by assuming that X is finite-dimensional,
i.e., that X = X d for some d 0. We also assume that X 6= , which forces
X 0 6= .
Step 1. We have already mentioned this step. By excision (which applies because
X n1 is a deformation retract of an open neighbourhood in X n ) one has, for n > 0,
_
_
M
k(
k (S n ).
Hk (X n , X n1 )
S n , ) = H
Sn)
H
= Hk (
=
n-cells
n-cells
Thus one has

Hn (X n , X n1 )
= Z{n-cells} ,
n-cells
62
TIM PERUTZ
and this is valid even when n = 0.

Step 2. From the long exact sequence of the pair (X n , X n1 ), one sees that, for
k > n 0, Hk (X n ) = Hk (X n1 ), and thus iteratively Hk (X n ) = Hk (X 0 ) = 0. So,
when X = X d , X has homology only up to dimension d.
Step 3. From the long exact sequence of the pair (X n+1 , X n ), one sees that, for
k < n+1, Hk (X n ) = Hk (X n+1 ). Thus, iteratively, Hk (X n ) = Hk (X n+p ) = Hk (X)
since X n+p = X for all n + p d. So to compute nth homology we can get away
with considering the n + 1-skeleton of X:
Hn (X) = Hn (X n+1 )
Step 4. Part of the long exact sequence of the pair (X n+1 , X n ) reads
n+1
Hn+1 (X n+1 , X n ) Hn (X n ) Hn (X n+1 ) 0.

Thus Hn (X) = Hn (X n+1 ) is isomorphic to coker n+1 . This represents progress,
since the domain of n+1 is a known group Z{(n+1)-cells} .
Step 5. The target of n+1 , Hn (X n ), sits in another exact sequence
qn
0 Hn (X n ) Hn (X n , X n1 ) Hn1 (X n1 )
(the zero on the left comes from Step 3). So, crucially, the map qn is injective. Let
dn+1 = qn n+1 . Heres the tricky step: one has
Hn (X)
= Hn (X n )/ im n+1
= im qn / im dn+1 ,
since qn carries Hn (X n ) isomorphically to im qn and im n+1 isomorphically to
im dn+1 .
Step 6. One has

im qn = ker n : Hn (X n , X n1 ) Hn1 (X n1 )

= ker dn1 : Hn (X n , X n1 ) Hn (X n1 , X n2 ) .
Thus
Hn (X)
= Hn (X n )/ im
= ker dn / im dn+1 .
Note that dn dn+1 = 0 because n qn = 0.
Remark. Let us make a note of what we have used in the proof. We needed excision
and the long exact sequence of the pair. We also needed the fact that the reduced
homology of a wedge is the direct sum of the reduced homologies. We needed
(S n ) = Z(n) . Recall that this was proved using excision (and
to know that H
homotopy invariance of singular homology) by an inductive argument beginning
k (S 0 ) = Z.
with H
We have Cn = Z{n-cells} and Cn1 = Z{(n1)-cells} , so dn is represented by an
integer-valued matrix (Dnij ):
X
Dn hii =
Dnij hji,
j
where hii represents an n-cell and the sum is over (n 1)-cells j.

Theorem 18.4. For any n > 1, the matrix Dn representing the differential dn
is equal to the degree matrix for the cellular attachments. The differential d1 is
characterized by dhii = hfi (1)i hfi (0)i.
63
Proof. We can think of the closure of the cell ek as an k-simplex (since it is homeomorphic to k . Thus the ith Z-summand in Hn (X n , X n1 ) is represented by the
cell ein , considered as a singular chain (notice that it is a relative n-cycle, since its
boundary lies in X n ).
Under n , the singular chain ein is mapped to its boundary, which is the image
of the fundamental class [S n ] Hn (S n ) under the cellular attaching map fi : that
is, n ein = (fi ) [S n1 ]. Next we have to apply the quotient map qn : X n1
X n1 /X n2 , since dn (ein ) = (qn1 ) (fi ) [S n1 ] Hn (X n1 /X n2 ).
The projection Hn (X n1 /X n2 ) Z to the jth Z-summand is induced by the
collapsing map
_
j : X n1 /X n2 =
S n1 S n1
(n1)cells
n1
which acts as the identity on jth copy of S

and sends everything else to the
wedge point. Thus the jth component of n ein is
(j ) (qn1 ) (fi ) [S n1 ] = (j qn1 fi ) [S n1 ] Hn1 (S n1 ).
But this is exactly the definition of the degree matrix Dn .
64
TIM PERUTZ
19. Cellular homology calculations

We compute some examples of cellular homology, and observe the uniqueness of
homology theories.
We begin by repeating the theorem from last time:
Theorem 19.1. If X is a CW complex then the homology of the cellular complex
(C , d) is canonically isomorphic to H (X). Here Cn = Z{n-cells} and Cn1 =
Z{(n1)-cells} , so the cellular differential dn is represented by an integer-valued matrix (Dnij ):
X
Dn hii =
Dnij hji,
j
where hii represents an n-cell and the sum is over (n 1)-cells j.

When n > 1, the matrix entry Dnij is the degree of j Qn1 fi : S n1
n1
n1
S
X n1 /X n2 =
i is the attaching map of the ith n-cell, Qn : X
W , where fn1
n1
n2
n1
the quotient map, and j : X
/X
S
the collapsing
(n1)cells S
map on the jth wedge-summand. The differential d1 is characterized by d1 hii =
hfi (1)i hfi (0)i.
19.1. Calculations. Let us use this theorem to compute the homology of T 2 , RP 2
and K 2 . The homology vanishes in degrees > 2, because these are 2-dimensional
cell complexes, but the interesting thing is to compute H1 and H2 .
X = T 2 has an obvious cell decomposition with one 0-cell, two 1-cells and
one 2-cell. These cells give bases v0 for C0 , (v1a , v1b ) for C1 , and v2 for C2 .
One has d1 v1a = v0 v0 = 0 = d1 v1b . The attaching map for the 2-cell is
a map S 1 Sa1 Sb1 . For T 2 it is aba1 b1 , so after collapsing Sb1 we
get a map homotopic to a a1 : S 1 Sa1 (which has degree 0), and after
collapsing Sb1 we get a map homotopic to b b1 : S 1 Sb1 (alslo degree 0).
Hence d2 v2 = 0. So H2 (T 2 ) = Z and H1 (T 2 ) = Z2 .
X = K 2 has a cell decomposition with X 0 = e0 , X 1 = ea1 eb1 X 0 , and
X = X 2 = e2 X 1 . These cells give bases v0 for C0 , (v1a , v1b ) for C1 , and
v2 for C2 . As for T 2 , one has d1 = 0. The attaching map for e2 is aba1 b,
so d2 v2 = 2v1b . Hence H2 (K 2 ) = 0 and H1 (K 2 ) = Z (Z/2).
X = RP 2 has a cell decomposition with two 0-cells, two 1-cells, and one
2-cell. These cells give bases (v0 , v00 ) for C0 , (v1a , v1b ) for C1 , and v2 for C2 .
Our conventions are such that d1 (v1a ) = v00 v0 = d1 (v1b ); thus ker d1 =
Z(v1a + v1b ). The attaching map for e2 is then abab, so d2 v2 = 2v1a + 2v1b .
Hence H2 (K 2 ) = 0 and H1 (K 2 ) = Z/2.
Happily, these calculations are consistent with our earlier results concerning 1 .
Example 19.2. Real projective space RP n = (Rn+1 \ {0})/R is a cell complex
with one k-cell of each dimension k {0, 1, . . . , n}. To see this, define
X k = {[x0 : : xn ] RP n : xj = 0 for all j > k}.
Thus X 0 X 1 X n = RP n , and X k
= RP k . We will exhibit X k as the kk
skeleton of a cell decomposition. If [x] X \X k1 RP n then xk+1 = = xn =
0 but xk 6= 0, so [x] = [y0 : . . . , yk1 : 1 : 0 : 0] for a unique (y0 , . . . , yk1 ) Rk .
Thus X k \ X k1
= Rk . Thus RP n is a disjoint union of open cells, one of each
dimension up to n. To see that they are attached in the proper way, think of ek as
65
{y = (y0 , : yk1 ) Rk : |y| 1} and define ik : ek X k as follows:

ik (y) = (y0 : : yk1 : (1 |y|2 )1/2 : 0 : : 0).
Then ik is injective on the interior of ek , but restricts to ek = S k1 as the map
fk (0 , . . . , k1 ) = [0 : . . . , k1 : 0 : : 0] which is the quotient map S k1
X k1 = RP k1 . Notice that ik extends to a homeomorphism ek fk X k1 X k
which restricts to the inclusion on X k1 .
To compute the cellular chain complex, we need to look at the composite Qk1
fk , where Qk1 : X k1 X k1 /X k2 = S k1 is the collapsing map. Well, when
|y| = 1, we have fk (y0 , . . . , yk1 ) = [y0 : : yk1 : 0 : . . . 0]. Thus fk (y) X k2
k1
precisely when yk1 = 0. Cut S k1 into two hemispheres D
= {(y0 , . . . , yk1 ) :
k1
|y| = 1, yk1 0}. Then, if y int(D+ ), we have
1
i1
k1 fk (y) = ik1 ik (y)
= i1
k1 (y0 : : yk1 : 0 : : 0)
= (y0 , . . . , yk2 )
k1
(check the last line for yourself!). If y int(D+
), then we have instead
i1
k1 fk (y) = (y0 , . . . , yk2 ).
We now see that the cellular differential dk is given by dk hek i hek1 i, where
is the degree of the composite map
p
ida
S k1 S k1 S k1 S k1 .
Here p is the pinching map that collapses the equator to the wedge point. The
k1
second map acts as the identity on the first wedge summand S+
and as the
k1
antipodal map a on the second wedge summand S . Now, on fundamental classes
we have p [S k1 ] = [S k1 ]+ + [S k1 ] , whilst (id a) maps [S k1 ]+ + [S k1 ] to
(1 + deg(a))[S k1 ] = (1 + (1))[S k1 ]. Thus = 1 + (1)k .
Hence the cellular complex reads
0
Z Z Z Z 0.
So we find the following: if n is even then
Z,
Z/2,
Hk (RP n ) =
0,
0,
k = 0,
k < n odd
k < n even
p n.
If n is odd then
Z,
k = 0,
Z/2,
k < n odd
n
Hk (RP ) = 0,
k < n even
Z,
k
= n,
0,
p > n.
66
TIM PERUTZ
Taking the last example further, lets now calculate H (RP n ; Z/2). We can do
this in two ways. One way is to use the cellular chain complex reduced mod 2,
which reads
0 Z/2 Z/2 Z/2 0
with a Z/2 in each degree between 0 and n. The maps are all zero. Hence
Hi (RP n ; Z/2) = Z/2 if 0 i n, and it is zero otherwise.
The other method is to use our calculation for Z-coefficients in conjunction with
universal coefficients. For n even, say, the Z/2-module Hk (RP n )Z/2 is Z/2 for
k = 0 or 0 < k < n odd, and 0 otherwise. However, Tor(Hk1 (RP n ), Z/2) is Z/2
if k 1 > 0 with k even, and is zero otherwise. Thus for any k between 0 and n,
the flanking terms in the universal coefficients exact sequence
0 Hk (RP n ; Z) Z Z/2 Hk (RP n ; Z/2) Tor(Hk1 (RP n ), Z/2) 0
always consist of Z/2 and 0 (in some order), hence Hk (RP n ; Z/2) = Z/2.
Exercise 19.1: Compute H (RP n ; Z/4). Do it in two ways: via the Z/4 cellular chain
complex; and via the homology over Z and universal coefficients.
Exercise 19.2: Use cellular homology to compute H (g ), where g is a standard
closed, orientable surface of genus g (defined, for instance, as a quotient of the 4ggon). Also, compute H (g #RP 2 ), the connected sum of g and RP 2 .
Exercise 19.3: Describe how the product X Y of finite CW complexes X and Y
inherits a structure of CW complex.
(1) Compute the Euler characteristic (X Y ) in terms of (X) and (Y ).
(2) Show that Ccell (X Y ) = Ccell (X) Ccell (Y ) as a graded abelian group.
(You are not asked to compute the differential.)
(3) Compute H (S n S n ).
(4) Let X = (S 3 )n = S 3 S 3 . Show that H3p (X) is isomorphic to the
exterior power p Zn , and that Hq (X) = 0 if q is not a multiple of 3.
67
20. The EilenbergSteenrod axioms

We show that the EilenbergSteenrod axioms uniquely characterize a homology
theory for CW pairs.
A CW pair (X, A) is a CW complex X and a subspace A which is a subcomplex,
i.e. it is a union of cells of X which form a CW complex.
Definition 20.1. An EilenbergSteenrod homology theory E on CW pairs assigns:
To each CW pair (X, A), and each integer n, an abelian group En (X, A)
(and we write En (X) for En (X, )).
To each map f : (X, A) (X 0 , A0 ) and each n Z it assigns a homomorphism En (f ) : En (X, A) Hn (X 0 , A0 ). One has En (f g) = En (f ) En (g)
and En (id(X,A) ) = idEn (X,A) . If f0 is homotopic to f1 via a homotopy {ft }
such that ft |A = f0 |A , then En (f1 ) = En (f0 ).
To each pair of spaces (X, A), and each integer n, it assigns a homomorphism
n : En (X, A) En1 (A).
These maps are natural transformations. That is, given f : (X, A)
(X 0 , A0 ), one has
n En (f ) = En1 (f ) n
as homomorphisms En (X, A) En1 (A0 ).
Besides these basic properties, the following axioms are required to hold:
DIMENSION: If denotes a 1-point space then En () = 0 for n 6= 0, while
E0 () = Z.
EXACTNESS: The sequence
n
En (A) En (X) En (X, A)
En1 (A) Hn1 (A) . . .
is exact, where the unlabelled maps are induced by the inclusions (A, )
(X, ) and (X, ) (X, A).
EXCISION: if (X; A, B) is an excisive triad then the map
En (A, A B) En (X, B)
ADDITIVITY: If (X , A ) is a family of pairs, then one has an isomorphism
M
a
a
En (X , A ) En ( X ,
A )
given by the sum of the maps induced by the inclusions into the disjoint
union.
Theorem 20.2 (EilenbergSteenrod). Take any homology theory E on CW pairs.
Then for any CW complex X, one has En (X, A)
= Hn (X, A). In fact, there is a
unique natural transformation E H extending a given isomorphism E ()
H (), and this natural transformation is a natural isomorphism.
Remark. The EilenbergSteenrod theorem is one of two general organising principles for (co)homology theories. The other is sheaf cohomology: the idea that one
can use different resolutions of the same sheaf to compute its cohomology. This
leads to a proof that singular cohomology is isomorphic to Cech

cohomology, and
68
TIM PERUTZ
that the real cohomology H (M ; R) of a smooth manifold is isomorphic to its de
Rham cohomology HdR

(M ).
We will not give a complete proof of the theorem: for a start, we will restrict our
attention, for simplicitys sake, to finite-dimensional CW complexes. We will prove
the existence of the natural isomorphism, but not its uniqueness. We will also not
prove that the natural transformations n are uniquely determined.
Proof. We have already proved the heart of this theorem in showing that H (X)
=
Hcell (X), for our proof only used properties of singular homology derivable from
the axioms.
To flesh this out, it is more convenient to work with reduced homology theories.
n (X) = En (X, {b}), where the basepoint b X is one of the 0-simplices.
Lets put E
Then En defines a reduced homology theory on based CW complexes, satisfying

analogues of the axioms above. To state them, we need the notion of the reduced suspension X of X: X = (X [1, 1])/ , where (x, 1) (y, 1), (x, 1) (y, 1),
and (unlike the unreduced suspension) (b, s) (b, t), where b is the basepoint. It
is made a CW complex by thinking of it as the smash product
S 1 X = (S 1 X)/(S 1 X)
(think this through). In general, the reduced suspension differs from the unreduced
suspension. However, one still has S n
= S n+1 . A map f : X Y preserving
basepoints induces f : X Y in a functorial manner.
The new axioms are as follows.
(S 0 ) = Z.
DIMENSION: E
EXACTNESS: if A is a CW subcomplex containing b then the sequence
(A) E
(X) E
(X/A)
E
is exact.
SUSPENSION: There is a natural (in X) isomorphism
(X) E
+1 (X).
: E
L
(W Xi ) is an isomor ADDITIVITY: The natural map
i E (Xi ) E
W i
phism. Here the Xi are based CW complexes, and i Xi their wedge sum
along the basepoints.
is a reduced homology
Exercise 20.1: Show that if E is a homology theory then E
theory.
:
Let CE (X) be the cellular chain complex defined via E
(X n /X n1 ).
CE (X) = E
n (X)
We can run our proof from two lectures ago to show that E
= Hn (CE (X)),
E
E
E
where C (X) is the reduced cellular chain complex C (X)/C (b). (This necessitates some slight adjustments in degree 0; otherwise the argument is identical.) To
be precise, the argument shows that we can obtain an isomorphism
n (X) Hn (CE (X))
: E
n (X) lifts to some x
n (X n ). We define (x) as the image of x
as follows: x E
E
n
n1
n (X /X
in E
).
(S n )
The dimension and suspension axioms tell us that E
= Z(n) , and hence that
n
n1
{ncells}
En (X /X
)=Z
. As before, the cellular differential may be identified
69
with a degree matrix, but at this point we hit a snag. The degree matrix is computed
rather than ordinary homology. Do these degrees agree?
using E
Let f : S n S n be a map preserving a basepoint. It has a homology degree
-degree, and we would like to prove that they are the same. This is easy
and an E
to check for S 0 , and quite easy also for S 1 , where we know that every map f is
homotopic rel basepoint to z 7 z d for some d Z. We can try to prove it in
general by induction on n, using naturality of the suspension isomorphism, but this
will only work for maps f homotopic to g for some g : S n1 S n1 . Does this
exhaust all possible maps S n S n ? Yes. This can be seen as a special case of the
Hurewicz theorem, one of the basic principles of homotopy theory. Alternatively,
it can be proved using a little differential topology (see Milnor, Topology from the
differentiable viewpoint).
and H
coincide.
The upshot is that the cellular chain complexes as defined via E
We deduce an isomorphism
(X) H
(X).
E
One now checks that this isomorphism is natural with respect to cellular maps
(those that send k-skeleta to k-skeleta) and so, by cellular approximation, with
(and
respect to arbitrary based maps. We have now succeeded in rebuilding E
hence E ) from the cellular homology of X.
. Indeed, it follows from excision
We now recover E from its reduced theory E
(X/A)
that the quotient map X X/A induces an isomorphism E (X, A)
=E
(X q ).
when A 6= ; and one has E (X) = E
One can also recover the natural transformation n , though this is trickier, and
I will omit it (see May chapters 14 and 8).
70
TIM PERUTZ
IV. Product structures

21. Cohomology
A cochain complex over R is a sequence of R-modules and maps
dp1
dp
C p1 C p C p+1 . . . ,
p Z,
such that dp dp1 = 0 for all p. Its just the same as a chain complex, except
that the indexing runs the other way. We use superscripts
L p for cochain complexes,
subscripts for chain complexes. We write C =
p C . The pth cohomology
module is then
H p (C ) = ker dp / im dp1 ,
L
and we put H (C ) =
H p (C ).
If (D , ) is a chain complex then one obtains a cochain complex by dualisation,
putting C p = HomR (Dp , R) and (dp f )(x) = f (p+1 x).
Remark. It is notQtrue that C = HomR (D , R), unlessLCp = 0 for |p| 0: one has
HomR (D , R) = p C p , which is usually bigger than p C p .
If X is a space, one defines the singular cochains S (X; R) by
S p (X; R) = HomZ (Sp (X), R),
with the differentials dp defined by dualising :
X
(dp c)() = c() =
(1)i c( i ).
i
p
One then puts H (X; R) = H (S (X; R)). Note that singular cochains, and hence
singular cohomology, are contravariantly functorial: a map f : X Y induces
f : H (Y ; R) H (X; R), and one has (f g) = g f .
Relative cohomology H p (X, A; R) for a subspace i : A X is defined as the
cohomology of ker(i : S (X; R) S (A; R)).
All the familiar properties of homology (long exact sequences of the pair, homotopy invariance, excision, MayerVietoris, etc.) have dual versions in cohomology,
with essentially identical proofs. Note that the connecting maps in long exact
sequences go up not down in degree.
Exercise 21.1: Formulate these properties of cohomology, and think through how you
would prove them.
Theorem 21.1 (Dual universal coefficients). If X is a space, and R any commutative unital ring, there are natural (in X), non-naturally split short exact sequences
j
0 Ext(Hp1 (X), R) H p (X; R) HomZ (Hp (X), R) 0.

In particular, there are non-canonical isomorphisms
H n (X)
= Hom(Hn (X), Z) Hn1 (X)tors .
where the Ators denotes the torsion subgroup of A.
This is a direct consequence of an algebraic theorem given below.
Exercise 21.2: Show that H 1 (X; R)
= HomZ (1 (X), R) for path connected X. Deduce that H 1 (X) = H 1 (X; Z) is a torsion-free abelian group for X locally path connected X.
71
21.1. Ext. Fix a commutative ring R and an R-module P . We investigate the

effect on exact sequences of the functor Hom(, P ) from R-modules to R-modules.
This problem is dual to the one we considered earlier involving tensor product,
because of the adjunction
Hom(M Q, P )
= Hom(Q, Hom(M, P )).
The arguments closely parallel those in the proof of homology universal coefficients,
and we shall mostly leave the verifications to the reader to work through.
p
Lemma 21.2. Let 0 M1 M2 M3 0 be a short exact sequence of Rmodules. Let P be another R-module. Then the induced sequence
p
0 Hom(M3 , P ) Hom(M2 , P ) Hom(M1 , P )

is also exact. If the short exact sequence splits (for instance, if M3 is a free module)
then Hom(M2 , P ) Hom(M1 , P ) is surjective.
In general, i is not surjective but its cokernel is measured by Ext. We assume
R is a PID, so (as discussed in the context of homology universal coefficients) any
R-module M has a two-step free resolution
f1
f0
0 F1 F0 M 0.
Applying Hom(, P ) to this sequence, we obtain a complex
f
0
1
0 Hom(M, P )
Hom(F0 , P )
Hom(F1 , P ) 0
which is exact at Hom(M, P ) and at Hom(F0 , P ). We prefer to truncate this to the

(non-exact) sequence
f
1
0 Hom(F0 , P )
Hom(F1 , P ) 0.
which we think of as a cochain complex. By the lemma, its cohomology at Hom(F0 , P )

(i.e., ker f1 ) is naturally isomorphic to Hom(M, P ). We define Ext(M, P ) to be its
cohomology at Hom(F1 , P ):
Ext(M, P ) = coker f1 .
f0
We showed in our treatment of Tor that given two free resolutions (F M )

g0
and (G M ), there is a chain map : F G , unique up to chain homotopy,
such that g0 = f0 . This shows that Ext(M, P ) is independent of the choice of
free resolution, up to canonical isomorphism.
Remark. Those who have studied Ext in other contexts should note that one can
compute Ext(M, P ) from a projective resolution of M or an injective resolution of
P . We have opted for a special case of the former.
Example 21.3.
When R is a field we may take F0 = M and F1 = 0. Thus
Ext(M, P ) = 0 for all P .
If M is a free module over the PID R, we may take F0 = M and F1 = 0.
Hence Ext(M, P ) = 0 for all P .
72
TIM PERUTZ
x
The R-module M = R/(x) has free resolution 0 R R R/(x) 0.

Thus
Ext(R/(x), P ) = coker(x : Hom(R, P ) Hom(R, P ))
= coker(x : P P )
= P/xP.
For instance, Ext(R/(x), R) = R/(x). Since Ext commutes with direct sum
(on either factor), one deduces that for a finitely-generated R-module M
one has Ext(M, R) = Mtors , the torsion submodule of M .
Theorem 21.4 (Dual universal coefficients). Let C be a chain complex of free Rmodules over a PID R. Then, for any R-module P , one has short exact sequences
j
0 Ext(Hp1 (C ), P ) H p (Hom(C , P )) Hom(Hp (C ), P ) 0.

These sequences split, but there is no preferred splitting. In particular, taking P =
R, we get short exact sequences
j
0 Hp1 (C )tors H p (C ) Hp (C ) 0,
where for any R-module M , M denotes Hom(M, R).
Proof. The proof is similar to that of the universal coefficient theorem for tensor
products. We take for our free resolution of the homology groups
ip
0 Bp Zp Hp (C ) 0
(Bp and Zp are free because Bp Zp Cp ). Thus we have
Ext(Hp1 (C ), R) = coker[ip1 : Hom(Zp1 , R) Hom(Bp1 , R)].
As in the tensor product argument, we have short exact sequences
p
0 Zp Cp Bp1 0
which split and therefore induce short exact sequences
dp
0 Hom(Bp1 , R) Hom(Cp , R) Hom(Zp , R) 0.

These constitute a short exact sequence of cochain complexes, and the resulting
long exact sequence on cohomology reads(!)
i
p1
dp
i
p
Hom(Zp1 , R) Hom(Bp1 , R) H p (Hom(C , R)) Hom(Zp , R) Hom(Bp , R).

We deduce from this short exact sequences
0 coker(ip1 ) H p (Hom(C , R)) ker(ip ) 0.
But ker ip = Hom(Hp (C ), R) and coker(ip1 ) = Ext(Hp1 (C ), R). These short
exact sequences split because Hom(Hp (C ), R) is a free R-module.

Remark. The universal coefficients theorems show that H (X) determines H (X; R)
and H (X; R) up to isomorphism, but not functorially. It is arguably better to
think of the singular chains as defining functors Sn : hTop hCh, where the category on the right is the category of chain complexes and chain homotopy classes
of chain maps. These functors can then be followed by algebraic functors R and
Hom(, R).
73
Exercise 21.3: Work over Z and fix distinct primes p and q. Compute (i) Ext(Z, Z/pn );
(ii) Ext(Z/pn , Z); (iii) Ext(Z/pn , Z/q m ); (iv) Ext(Z/pn , Z/pm ).
Exercise 21.4: Working over Z, compute Ext(A, Q/Z) for an arbitrary finitely generated
abelian group A.
Exercise 21.5: For any simply connected space X, one has H 2 (X)
= Hom(H2 (X), Z).
Exercise 21.6: Verify a case of the theorem by computing H (RP 2 ; Z) in two ways.
Do the same for H (RP 2 ; Z/2).
Exercise 21.7: What is the relation between H (X; R) and H (X)?
74
TIM PERUTZ
22. Product structures, formally

We explain the formal structure of the cup and cap products and state the
Poincare duality theorem.
Homology and cohomology are much more powerful invariants when one takes
account of their multiplicative structures. The cohomology H (X) is a graded ring
under the cup product. Homology H (X) is not a ring, but it is a graded module
over the graded ring H (X). This is expressed by the cap product.
22.1. The evaluation pairing. The most basic product structure is the evaluation pairing
H p (X) Hp (X), (c, h) 7 hc, hi.
In singular theory, this is induced by the evaluation
S p (X) Sp (X) = Hom(Sp (X), Z) Sp (X) Z;
similarly in cellular theory. One extends the evaluation pairing to a linear map
H (X) H (X) Z
which is zero on H p Hq when p 6= q.
Remark. It might be interesting to axiomatise this pairing in an arbitrary Eilenberg
Steenrod theory. Over a field k, H n (; k) is dual to Hn (; k) in a functorial manner.
Over Z this is not so, but there are specialisation maps H n () H n (; Z/p) and
Hn () Hn (; Z/p), so one could ask that an evaluation pairing should be naural,
and should specialise to the dual pairing between mod p homology and cohomology
for all primes p.
22.2. The cup product. The cup product is an associative bilinear product
H (X) H (X) H (X),
(a, b) 7 a ^ b.
Its fundamental properties are as follows:

One has H p (X) ^ H q (X) H p+q (X). Thus ^ makes H (X) into a
graded ring.
The ring has a unit element 1 H 0 (X): it is characterised by h1, hi = 1
when h H0 (X) is the class of a point.
The cup product is commutative in the graded sense:
a ^ b = (1)|a||b| b ^ a.
The cup product is functorial: if f : X Y is a map then
f (a ^ b) = f a ^ f b.
hence f is a homomorphism of unital graded rings. In particular, the
cohomology ring is an invariant
`of homotopy
Q type.
The natural isomorphism H ( iI Xi )
= iI H (Xi ) is an isomorphism
of rings.
`
W
So, for example, the pinching map iI Xi iI Xi induces a ring homomorphism
_
Y
H n ( Xi )
H n (Xi ),
iI
iI
which is an isomorphism in degrees n > 0. Since this isomorphism is induced by a

map between the spaces, it respects the cup product.
75
22.3. The cap product. The cap product is a bilinear product

H (X) H (X) H (X),
(c, h) 7 c _ h.
Its fundamental properties are as follows:

One has (a ^ b) _ h = a _ (b _ h).
One has 1 _ h = h.
One has H p (X) _ Hn (X) Hnp (X). Thus _ makes H (X) into a
graded unital module (with suitable grading conventions...) over the ring
H (X).
One has
ha ^ b, hi = ha, b _ hi.
If f : X Y is a map then
f (f c _ h) = c _ f h
for h H (X) and c H (Y ). This shows that homology, as a graded
module over cohomology, is an`invariant ofLhomotopy type.
The natural isomorphism H ( iI Xi )
= iI H (Xi ) is an isomorphism
of modules.
I dont know whether the axioms for the cap and cup products that I have
listed, when considered together, characterise both of them uniquely. However, the
relation between cup, cap and evaluation products does force them to be non-trivial
(e.g. one cant define the cup product of positive-degree cohomology classes to be
identically zero).
Most interesting computations of cohomology rings and homology modules invoke the Poincare duality theorem.
Theorem 22.1 (Poincare duality). Suppose M is a compact, connected, oriented
n-manifold with fundamental class [M ] Hn (M ). Then the duality map
D : H p (M ) Hnp (M ),
D(c) = c _ [M ],
is an isomorphism for all p Z.

The proof of this theorem mostly uses the naturality properties of cap and cup
products, though at one point it becomes necessary to look more closely at the
definition.
In practice, one works with a corollary. Next time well use this to compute the
cohomology ring of projective space.
Corollary 22.2 (Poincare duality: cup product version). Let M be a compact,
connected, oriented n-manifold, and [M ] Hn (M ) the fundamental class. Then
for any p Z the Poincare pairing
H p (M )/T p H np (M )/T np Z,
k
(a, b) 7 ha ^ b, [M ]i,
is non-degenerate. Here T H (M ) denotes the submodule of torsion classes, so

H k (M )/T k is torsion-free and hence free.
Remark. Note that the hence here uses a fact that we have not proved, that
the (co)homology of a compact manifold is finitely generated (a finitely generated
torsion-free abelian group is free). In fact, any compact manifold is homotopyequivalent to a finite CW complex. In the smooth case, this follows easily from
Morse theory; in general, it is a hard theorem.
76
TIM PERUTZ
Remark. Non-degeneracy means that the adjoint map

H p (M )/T p Hom(H np (M )/T np , Z),
a 7 (b 7 ha ^ b, [M ]i)
is an isomorphism. If one fixes bases for the free abelian groups H k (M )/T k , it is
the assertion that the matrix of the pairing is square and its determinant is 1.
Proof of the corollary. Recall a corollary of universal coefficients (which used finite
generation): H p (M ) = Hom(Hp (M ), Z) Hp1 (M )tors . Thus H p (M ; R)/T p
=
Hom(Hp (M ), Z). The adjoint map : H p (M ) Hom(H np (M ), Z) sends a to
the map b 7 ha ^ b, [M ]i = hb ^ a, [M ]i = hb, a _ [M ]i. If b is non-zero (mod
torsion) then it evaluates non-trivially on some homology class, and by the Poincare
duality theorem, we may take that class to be of form b _ [M ]. Hence ker = T p .
And is also surjective, because given any homomorphism f H np (M ) Z, we
can represent it as evaluation on some homology class h = D(c), and then f = (c).
77
23. Formal computations in cohomology

We show that Poincare duality leads to computations of cup product structures.
We explain an algebraic application.
Last time, we noted a corollary of Poincare duality, that on a compact oriented
manifold, the cup-product pairing on cohomology mod torsion is non-degenerate.
We now apply this to complex projective space. Take a polynomial ring Z[u], and
make it a graded ring by declaring u to have degree 2. Truncate it to the graded
ring Z[u]/(un+1 ). As a graded abelian group, we have
H (CP n )
= Z[u]/(un+1 ).
Theorem 23.1. There is an isomorphism of graded rings,
H (CP n )
= Z[u]/(un+1 ),
deg u = 2.
Proof. Induction on n. The space CP 0 is a point, and the result is trivially true
in this case. It is also trivially true when n = 1, and we will start the induction
there. If n > 1, observe that we have an inclusion i : CP n1 CP n , induced
by a linear inclusion Cn+1 Cn+2 . By induction, H (CP n1 )
= Z[t]/tn where
deg t = 2. Now, i is additively an isomorphism up to degree 2(n 1). Let

u H 2 (CP n ) be defined by i u = t. Since tn generates H 2(n1) (CP n1 ), un
generates H 2(n1) (CP n ). By Poincare duality, the pairing
H 2(n1) (CP n ) H 2 (CP n ) Z
is non-degenerate over Z. It follows that the generators of the two groups pair to
give 1, i.e.,
hun1 u, [CP n ]i = hun , [CP n ]i = 1.
Hence un generates H 2n (CP n ), and the result follows.
Exercise 23.1: Specify a suitable cohomology class u. Describe uk for all k. (Part of
the exercise is to work out what format the answer should sensibly take.)
Remark. A precisely similar argument, using the mod 2 version of Poincare duality,
shows that
H (RP n ; Z/2)
= (Z/2)[t]/tn+1 , deg t = 1.
The result for projective spaces shows that the cup product makes cohomology
a more powerful invariant.
Example 23.2. CP 2 is not homotopy equivalent to S 2 S 4 . Indeed, H (CP 2 )
contains a degree 2 class u such that u ^ u is non-tivial, whereas the cup-square
of a degree 2 class in H (S 2 S 4 ) is always zero.
Example 23.3. Any homeomorphism h : CP 2 CP 2 preserves orientation. For
it suffices to show that h [CP 2 ] = [CP 2 ] (where [CP 2 ] is the fundamental class),
i.e., that h has degree 1 rather than 1. But h u = u, since these are the two
generators for H 2 . So h (u ^ u) = h u ^ h u = u ^ u, hence h is the identity
on H 4 . Hence the dual map h on H4 is also the identity.
The argument also extends to CP 2n , but not to CP 2n+1 (indeed, it is false for
CP 1 ).
However, it still has limitations:
78
TIM PERUTZ
Example 23.4. S 3 S 5 and S(CP 2 ) have isomorphic cohomology rings: additively,

both are Z(0) Z(3) Z(5) where the subscripts denote degree. The product of two
classes of positive degree has to be zero, for reasons of degree.
Exercise 23.2: (a) Suppose that X and Y are compact, connected, orientable nmanifolds. Describe the cohomology ring of their connected sum X#Y . (b) Prove
that CP 2 #CP 2 and CP 2 #CP 2 are not homotopy-equivalent. Here CP 2 is the complex projective plane with one orientation, and CP 2 the same space with the opposite
orientation.
23.1. The K
unneth formula. There is one other general result which is very
useful in computing cup products. To state it I need to work with cohomology
with coefficients in a ring k. Cohomology with coefficients in k also carries a cup
product, which is linear over k and so makes H (X; k) a k-algebra.
The result is the K
unneth formula. In general, if A and B are graded k-algebras
then
their
tensor
product
A B is also a graded k-algebra, with ith graded part
L
A
B
,
and
product
given on homogeneous monomials by
ij
j j
0
(x y) (x0 y 0 ) = (1)|y||x | (x ^ x0 ) (y ^ y 0 ).
Theorem 23.5 (K
unneth formula). Let X and Y be spaces. Define the cross
product map
H (X; k) k H (Y ; k) H (X Y ; k),
(x, y) 7 x y := prX x ^ prY y,
where prX and prY the two projection maps from X Y . One checks that it is a map
of graded k-algebras. When X and Y are CW complexes, and H (Y ) is a finitely
generated free k-module (e.g. when k is a field), this map is an isomorphism.
Without the assumption that H (Y ) is a free module, there is still a K
unneth
theorem; it says that the cross product map is injective, and expresses its cokernel
using Tor.
Example 23.6.
H (CP n CP m )
= Z[t, u]/(tn+1 , um+1 )
with |t| = |u| = 2, and
H (RP n RP m ; Z/2)
= (Z/2)[x, y]/(xn+1 , y m+1 )
with |x| = |y| = 1 (the signs become irrelevant when k = Z/2).
23.2. An algebraic application of cup product. A division algebra over a field
F is an F -vector space A equipped with a bilinear map m : A A A, such that
for each x A \ {0} the linear maps m(x, ) : A A and m(, x) : A A are
isomorphisms.
Theorem 23.7 (Hopf). The dimension of a finite-dimensional division algebra
over R is a power of 2.
Proof. Let A be an n-dimensional real division algebra, with multiplication
m : A A A.
Bilinearity of m, and the fact that m(x, y) = 0 implies x = 0 or y = 0, tell us that
there is an induced map
M = Pm : PA PA PA.
79
Let k be the field Z/2. On mod 2 cohomology, M gives us a map

M : H (PA PA; k) H (PA; k).
We have H (PA; k) = k[x]/(xn ) where deg x = 1, and by the K
unneth formula,
H (PA PA; Z/2) = k[y, z]/(y n , z n ). Thus we have a map of graded rings
M : k[x]/(xn ) k[y, z]/(y n , z n ).
We claim that M (x) = y + z. Indeed, we know that for some (or indeed, any)
PA, the composite PA PA PA PA, 7 (, ) 7 M (, ) is a
homeomorphism, hence that M x = y mod z, and similarly that M x = z mod y.
This proves the claim.
Hence (y + z)n k[y, z] lies in the ideal generated by y n and z n . This implies
that the coefficient of y k z nj for 1 j < n, namely nj , must be zero in k, and
hence that the equation
(1 + t)n = 1 + tn
holds in k[t]. Now, (1 + t)2 = 1 + t2 in k[t]. Thus (1 + t)4 = (1 + t2 )2 = 1 + t4 , and
m
m
more generally, (1 + t)2 = (1 + t2 ). But if we write n in binary form, putting
n = 2m1 + 2m2 + + 2ma , m1 < m2 < < ma ,
Q
Q
mi
mi
m1
then (1 + t)n = (1 + t)2 = (1 + t2 ). But this expression is 1 + t2 plus
higher order terms. So the equation (1 + t)n = 1 + tn holds only if a = 1, i.e. n is
a power of 2.

Remark. The proof applies to a slightly more general kind of algebra: we can assume
that there exist some x and y in A so that m(x, ) and m(, y) are isomorphisms.
Exercise 23.3: Prove that the cohomology groups of a compact, simply connected,
oriented 4-manifold M are completely determined by the second Betti number
b2 (M ) = dimQ H 2 (M ; Q).
Prove that the cohomology ring is determined by the cup-product pairing H 2 H 2 Z.
Exercise 23.4: A special case of the Lefschetz fixed point theorem states that if X is a
finite CW complex, and g : X X a map without fixed points, then
X
(1)p tr H p (g) = 0.
p0
Here H (g) is the induced map on H (X; Q). What restrictions does this entail for
groups acting freely on CP n ?
p
80
TIM PERUTZ
24. Cup products defined

We define the cup product of cochains.
24.1. The basic mechanism. The definition of cup products combines two ideas:
(1) There is a chain map
D : C (X X) C (X)
induced by the diagonal map D : X X X, x 7 (x, x).
(2) There is a natural chain map (actually, a quasi-isomorphism)
: C (X) C (X) C (X X).
Here by C denotes cochains, but they could be singular or cellular. Given these
elements, one defines a chain map
^ : C p (X) C q (X) C p+q (X),
a ^ b = D# (a b).
24.2. Cup products in cellular cohomology. Suppose, for instance, we work

with cellular cohomology.
Proposition 24.1. Suppose X has a CW decomposition with p-cells e
p , and Y
a CW decomposition with q-cell fq . Then X Y has a cell decomposition with
(p + q)-cells e
p eq . The cellular chain complex of the product is given by
Ccell (X Y ) = Ccell (X) Ccell (Y ),
with the cellular boundary dXY (e f ) = dX e f + (1)deg e e dY f .
I wont give the proof (see Hatcher p. 268 or May p.99), but the basic point is
that the product of a p-cell and a q-cell is a p + q-cell.
Lets apply this to X X. It has a cell decomposition with cells e

p eq , and
one has an isomorphism of chain complexes
Ccell (X X) Ccell (X) Ccell (X),
where the differential on the right-hand side is given on the summand Cpcell (X)
Cqcell (X) by
dp id + (1)p id dq .
The cellular cochain complex Ccell

(X) is defined as the dual cochain complex to
(C (X), d). Thus by duality one has an isomorphism of cochain complexes
: Ccell
(X) Ccell
(X) Ccell
(X X).
Now, D is not a cellular map, but we have the cellular approximation theorem:
Theorem 24.2. Every map between CW complexes is homotopic to a cellular map,
i.e., one which maps the k-skeleton to the k-skeleton for all k.
So D is homotopic to a cellular map D0 : X X X.
Example 24.3. Make S 1 a CW complex with exactly two cells. Then one can see
from a picture how the diagonal S 1 S 1 S 1 is homotopic to a cellular map.
Exercise 24.1: Find a cellular approximation to the diagonal S n S n S n .
81
Define the cup product on the cellular cochain complex by

a ^ b = D0# (a b).
This is a perfectly reasonable (and correct) definition; for instance, it satisfies
d(a ^ b) = da ^ b + (1)|a| a ^ db.
Its only deficiency is that it is non-explicit, because of the cellular approximation
to D.
In some cases, one can write down an explicit cellular approximation and use it
compute the cup product:
Example 24.4. Use a cellular approximation to the diagonal S 1 S 1 S 1 to
construct another such approximation S 1 S 1 S 1 S 1 S 1 S 1 . Hence compute
the cup-product structure of H (S 1 S 1 ).
24.3. Cup products in singular cohomology. We have a canonically-defined
chain map D# : S (X X) S X. Explicitly, for c S p (X X) and p (X),
one has
hD# (c), i = hc, i.
We want to define
a ^ b = D# (a b).
We need to set up a suitable chain map , which should be natural in X. There is
an explicit formula for such a , called the AlexanderWhitney map:
Proposition 24.5 (AlexanderWhitney). Define a linear map
= X : S p (X) S q (X) S p+q (X X)
by
h(a b), i = ha, (pr1 )|[v0 ,...,vp ] )ihb, (pr2 )|[vp ,...,vp+q ] )i
p
for a S (X), b S q (X), and p+q (X X). Then is a chain map. It is

natural in X in that
(f f ) Y = X f
for f : X Y . One has
(1X 1X ) = 1XX ,
0
where 1X S (X) evaluates as 1 on any simplex (similarly 1XX ).

Exercise 24.2: Check the proposition.
We can use the AlexanderWhitney map to define the cup product as D# :
Definition 24.6. The cup product of cochains a and b is defined by
ha ^ b, i = ha, |[v0 ,...,vp ] )ihb, |[vp ,...,vp+q ] )i.
Remark. It is a theorem of EilenbergZilber that is a chain-homotopy equivalence.
Moreover, any other chain map 0 that is natural in X and satisfies 0 (1X 1X ) =
1XX is naturally chain-homotopic to . Thus one could define cup product via 0 ,
but on cohomology the definition would give the same product.
82
TIM PERUTZ
Lemma 24.7. One has

d(a ^ b) = da ^ b + (1)|a| a ^ db;
a ^ (b ^ c) = (a ^ b) ^ c;
a ^ 1X = a = 1X ^ a.
Thus S (X) is a differential graded algebra (DGA). For a map f , one has f (a ^
b) = f a ^ f b, so this DGA is natural in X.
Thus ^ descends to give a unital ring structure on cohomology H (X) bilinear
product on cohomology, natural in X.
Exercise 24.3: Use singular cohomology to compute H (S 1 S 1 ) as a ring. [You may
find it helpful to think of S 1 S 1 as a -complex.]
The cap product is defined by a similar procedure:
1D#
S p (X)Sn (X) S p (X)Sn (XX)) S p (X)[S (X)S (X)]n Snp (X).

Here the last map is
c y z 7 hc, yiz.
Exercise 24.4: Check that the cap product has the properties I claimed two lectures
back.
83
25. Non-commutativity
We sketch a definition of the simplest interesting Steenrod operation,
Sq n1 : H n (X; Z/2) H 2n1 (X; Z/2)
and use it to distinguish two homotopy types.
In the de Rham cohomology theory of smooth manifolds M , which is naturally
isomorphic to M 7 H (M ; R), the wedge product of differential forms is commutative (in the graded sense). However, the cup product of cochains is not commuative.
Lemma 25.1. The cochain-level cup product of singular cochains is not commutative. However, there exist natural maps : S p (X) S q (X) S p+q1 (X) satisfying
the chain homotopy identity
a ^ b (1)|a||b| b ^ a = d(a b) + (da b + (1)|a| a db).
Hence the cohomology cup product is commutative (in the graded sense).
I will not prove this lemma.
Suppose we work with cochains over Z/2. Construct the natural map as in
the lemma. For [c] H n (X; Z/2), define Sq n ([c]) = [c ^ c] H 2n (X; Z/2), and
Sq n1 ([c]) = [(c c)] H 2n1 (X; Z/2).
This makes sense because the identity satisfied by implies that, working mod 2,
we have that d(cc) = 0 when dc = 0, and also that ((c+db)(c+db))(cc)
is exact.
It is eminently plausibleand moreover truethat if (X; A, B) is an excisive
triad then the MayerVietoris connecting maps p : H p (A B) H p+1 (X) satisfy
Sq n1 (n c) = 2n1 (Sq n1 c),
c H n1 (A B).
(Here the Sq n1 on the right is the cup-square, that on the left the one defined
using .) It follows that one has (for n > 0)
Sq n1 (c) = (Sq n1 c) = (c ^ c),
c H n1 (X),
(X) H +1 (SX) is the suspension isomorphism.

where : H
Example 25.2. For instance, if 0 6= u H 2 (CP 2 ; Z/2) then in H 5 (SCP 2 ; Z/2)
one has
Sq 2 (u) = (u2 ) 6= 0.
2
So Sq is non-zero. On the other hand, in S 5 S 3 , which has the same cohomology
ring as S(CP 2 ), one has Sq 2 (H 3 ) = 0. So, once the operation Sq 2 : H 3 H 5 has
been put on a sound footing, we will have a proof that S 5 S 3 6' S(CP 2 ).
We have omitted several details in this discussion, but the basic points are these.
There is no commutative cochain-level cup product on mod 2 cochains
which is natural in X.
This non-commutativity has a manifestation in an operation on mod 2
cohomology which is sometimes non-trivial. This can be used to distinguish
homotopy types (such as S 5 S 3 versus S(CP 2 )).
84
TIM PERUTZ
duality
26. Poincare
We provide most but not all of the details of the proof of Poincare duality.
Poincare duality is primarily a statement about compact manifolds. Let us recall
the statement in that case.
Theorem 26.1 (Poincare duality: compact case). Suppose M is a compact, connected, R-oriented n-manifold with fundamental class [M ] Hn (M ; R). Then the
duality map
D : H p (M ; R) Hnp (M ; R),
D(c) = c _ [M ],
is an isomorphism for all p Z.

We shall prove the Poincare duality theorem using MayerVietoris sequences,
starting from simple cases and gradually expanding the generality. However, most
of the manifolds one encounter along the way are non-compact, and it is therefore
useful to have a formulation valid in the non-compact case. This involves the
compactly supported cohomology and relative cap products.
The compactly supported cohomology Hc (M ; R) is is the cohomology of a complex built from singular cochains c for which there is some compact subset K so
that c annihilates all chains in X \ K.1 More precisely, we define
Hcp (M ; R) = lim H p (X, X \ K; R)
where the direct limit is over compact subsets K X. To explain this, note that
the compact subsets of X form a direct system under inclusion. This means that
one has inclusion maps i : K1 K2 , and the composite of inclusion maps is again
an inclusion map. The identity map on X is then a map of pairs
(X, X \ K2 ) (X, X \ K1 ),
and hence induces a homomorphism
H p (X, X \ K2 ; R) H p (X, X \ K1 ; R).
In this way, the modules H p (X, X \ K) become a direct system as K ranges over
compact subsets. We defined Hcp (M ; R) as the direct limit of this system. Thus
one has a canonical map
H p (X, X \ K; R) Hcp (X; R)
for each compact K, and these commute with the maps in the inverse system.
Indeed, Hcp (X; R) is universal with respect to this property.
In practice, one computes Hcp (X) as the direct limit of H p (X, X \ K) as K
ranges over some compact exhaustion, i.e., a family of compact subspaces Ki such
that every compact subspace is contained in some Ki . This is valid by abstract
nonsense about cofinal families.
Example 26.2. Let us compute Hcp (Rn ). A compact exhaustion is given by the
discs Dn (m) centred at 0 and of radius m = 1, 2, . . . . One has H (Rn , Rn \
Dn (m))
= Z(n) . The inclusion of D(m) in D(m + 1) obviously induces an isomorphism on the relative cohomology groups. Hence Hc (Rn )
= Z(n) .
1If you know about de Rham theory, you should think of differential forms with compact
support.
85
Remark. Taking direct limits of (co)chain complexes is an exact functor, i.e. it

commutes with passing to (co)homology. This is not true of inverse limits; the
failure of exactness is measured by the derived functor lim1 of the inverse limit.
p
The cap product _ : H (X) Hq (X) Hqp (X) generalizes to a relative cap
product
H p (X, A) Hq (X, A) Hqp (X),
defined at (co)chain level by the same formula as the original cap product. Indeed,
one has ( _ a) = d _ a _ c. If represents a cocycle rel A, and a a
cycle rel A, then both terms vanish. When p = q, the cap product is essentially
just the evaluation pairing. More precisely,
h1, _ ai = h, ai.
Theorem 26.3 (Poincare duality: general case). If M is an R-oriented n-manifold
then the duality map
D : Hcp (M ; R) Hnp (M ; R),
is an isomorphism. Here D is defined as the direct limit of the maps
DK : H p (M, M \ K; R) Hnp (M ; R),
c 7 c _ [MK ],
where [MK ] Hn (M, M \ K) is the fundamental class of M relative to K, and _

the relative cap product.
In the proof we shall need the observation that the inclusion i : U M of
an open subspace U induces a covariant homomorphism i : Hcp (U ) Hcp (M ).
Indeed, if K U with K compact then excision gives an isomorphism
H p (U, U \ K)
= H p (M, M \ K),
and these isomorphisms commute with the maps in the direct system. We define
i as the composite
Hcp (U ) = lim H p (U, U \ K)
H p (M, M \ K) Hcp (M ).
= lim
KU
KU
Proof of the theorem. We drop the coefficients R from the notation. We shall prove
that D : Hcp (U ) Hnp (U ) is an isomorphism for each open subset D M
(including, eventually, M itself).
Step 1. The result holds when U = Rn .
Indeed, we saw above that Hc (Rn )
= R(n)
= Hn (Rn ). For any compact subset
K, the cap product of a generator for H n (Rn , Rn \ K) by the fundamental class
[RnK ] Hn (Rn , Rn \ K) is (up to a sign) the point class in H0 (Rn ), because of the
relation between relative cap product and the evaluation pairing. Passing to the
direct limit, we find that D is an isomorphism in this case.
Step 2: If the result holds for U , V and U V then it holds for U V .
I claim that there is a covariant MayerVietoris sequence in compactly supported
cohomology, and that one has a commutative diagram with exact rows of which
one portion reads
Hcp (U ) Hcp (V )
Dy
Hcp (U V )
Dy
Hcp+1 (U V )
Dy
Hnp (U ) Hnp (V ) Hnp (U V ) Hnp1 (U V )
86
TIM PERUTZ
Once this is verified, the 5-lemma assures us that D : Hcp (U V ) Hnp (U V ) is

an isomorphism. The proof of this claim involves locality for chains (unsurprising
since that underlies MayerVietoris) and a good deal of checking, for which I refer
to Hatcher.
Step 3: If the result holds for each S
Ui in a nested sequence U1 U2 U3 of
open subspaces, then it holds for V = Ui .
Any compact subset K V is contained in some Ui . The algebraic properties
of direct limits give
Hcp (V ) = lim lim H p (Ui , Ui \ K) = lim Hcp (Ui ).
KUi
But
[
Hnp (V ) = Hnp ( Ui ) = lim Hnp (Ui ),
because the singular chain complex of an expanding union is the direct limit of
the singular chain complexes, and taking homology commutes with direct limits.
Now D : Hcp (Ui ) Hnp (Ui ) is an isomorphism, and it commutes with the maps
induced by inclusion of Ui in Ui+1 ; hence D = limi (D : Hcp (V ) Hnp (M )) is an
isomorphism.
Step 4: The result holds for open subsets of Rn .
Every open set U Rn is the union of a countable set of open balls. Hence U
is the union of a nested family U1 U2 U3 . . . , where U1 is an open ball, and
Ui+1 is the union of Ui and a convex open set C homeomorphic to Rn . Note that
C Ui is convex, open, and has compact closure, and hence is homeomorphic to a
ball. Thus the result holds for Ui+1 by induction and steps 1 and 2.
Step 5: The result holds for M .
By Steps 1 and 4, and Zorns lemma, there is a non-empty, maximal open set O
for which the result holds. If this werent all of M , we could take the union of O
and a coordinate neighbourhood U
= Rn , disjoint from O; the result would then
hold for O U by steps 1 and 2. This contradicts maximality of O.

Remark. There is an almost trivial proof of duality for compact smooth manifolds
M in the context of Morse theory. The cellular chain complex can be understood
in terms of critical points and gradient flows for a Morse function f . Replacing f
by f does not change the homology (which is just H (M )) but it has the effect of
dualizing the chain complex and changing degree to degree n . Thus H (M )
is isomorphic to the cellular cohomology of M in degree n .
Exercise 26.1: Let M be a compact, connected, oriented 3-manifold. Determine the
graded ring H (M ) when (i) 1 (M ) is finite; (ii) H1 (M )
= Z; (iii) H1 (M )
= Z2 .
Exhibit two such 3-manifolds with non-isomorphic cohomology rings, both of which
have H1
= Z3 .
Exercise 26.2: Show that the Euler characteristic of a compact, orientable, odd-dimensional
manifold is zero.
Exercise 26.3: For which even dimensions 2n is it true that the Euler characteristic of
a compact, connected, orientable 2n-manifold is necessarily even?
Exercise 26.4: Show that for every map f : S 2n CP n , the induced map f on
reduced homology is zero.

Algebraic Topology by Tim Perutz

Uploaded by

Copyright:

Available Formats

Algebraic Topology by Tim Perutz

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Algebraic Topology by Tim Perutz

Uploaded by

Copyright:

Available Formats

ALGEBRAIC TOPOLOGY I: FALL 2008

12.1. The MayerVietoris sequence

ALGEBRAIC TOPOLOGY I: FALL 2008

I. The fundamental group

ALGEBRAIC TOPOLOGY I: FALL 2008

We now define a category 1 (X), the fundamental groupoid of X. A category

If G : Y Z is another map, one clearly has G F = (G F ) .

2. The fundamental group of the circle

ALGEBRAIC TOPOLOGY I: FALL 2008

homeomorphically to Uf (t) , independent of s. Thus we can define f(t) to be the

(Here D2 denotes the closed unit disc.)

ALGEBRAIC TOPOLOGY I: FALL 2008

3. Van Kampen in theory

We often write this group as ha1 , . . . , an i. For example, F1 = hai = Z.

A push-out for (f1 , f2 ) is another group P and a pair of homomorphisms p1 : G1

ALGEBRAIC TOPOLOGY I: FALL 2008

3.3. Van Kampens theorem.

ALGEBRAIC TOPOLOGY I: FALL 2008

4. Van Kampen in practice

In K 2 , q(s1 ) = q(s3 ) and q(s2 ) = q(s1

where gn (n ) = 1, gn (i ) = i for i < n, gn (a) = a and gn (b) = b.

ALGEBRAIC TOPOLOGY I: FALL 2008

coming from V ; a relation = 0 from U V ; and relations

Proposition 4.7. 1 (S \ K) = ha, b | a b i.

Exercise 4.1: An n-dimensional manifold is a Hausdorff space X covered by open sets

ALGEBRAIC TOPOLOGY I: FALL 2008

by a discrete group G acting

5.1. Deck transformations.

ALGEBRAIC TOPOLOGY I: FALL 2008

C {} = S 2 . Moreover, there is a homeomorphism

SU(2) S = {(, ) C : || + || = 1},

The involution A A on SU(2) corresponds to the antipodal map on S 3 ,

(1) If : I X a path, and (0) = x, then there is a unique path : I X

Exercise 5.4: Write out the missing details.

(a) to a group of deck transformations Aut(X/X); and (b) to a conjugacy class of

normal subgroup, and the quotient group is isomorphic to Aut(X/X).

= R, with p : R R/Z = S 1 the quotient map, then Aut(X/X)

1 (R) = {1}, so again Aut(X/X)

then Aut(X/X) = 1 (X).

ALGEBRAIC TOPOLOGY I: FALL 2008

6. Classifying covering spaces

f : B X with p f = f and f (b) = x

sitively on any fibre p (x). We obtain an isomorphism Ix : 1 (X, x) Aut(X/X)

We shall define two categories and prove their equivalence. An equivalence of

ALGEBRAIC TOPOLOGY I: FALL 2008

We can recover the conjugacy class of H G from G(H) as the image of

connected) such that p 1 (Y, y) = H G = 1 (X, p(y)). Show that Aut(X/X)

ALGEBRAIC TOPOLOGY I: FALL 2008

II. Singular homology theory

It is the convex hull [v0 , . . . , vn ] of the points vi = (0, . . . , 0, 1i , 0, . . . , 0).

(x0 , . . . , xn1 ) 7 (x0 , . . . xi1 , 0, xi , . . . , xn1 ).

It is homeomorphism onto the face [v0 , . . . , vi , . . . , vn ].

(1)i (|[v0 ,...,vi ,...,vn ] ).

Lemma 7.1. n n+1 = 0.

For any n + 1-simplex , we have

Since simplices generate Sn+1 (X), the result follows.

ALGEBRAIC TOPOLOGY I: FALL 2008

Proof. Since h is a homomorphism, h(f gf 1 g 1 ) = h(f )+h(g)h(f )h(g) = 0.

8. Simplicial complexes and singular homology

ALGEBRAIC TOPOLOGY I: FALL 2008