Homotopy Theory [RW]

Homotopy Theory
Oscar Randal-Williams
https://www.dpmms.cam.ac.uk/∼or257/teaching/notes/HomotopyTheory.pdf
1 Homotopy groups, CW complexes, and brations 1

1.1 Homotopy groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Change of basepoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Relative homotopy groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 The Hurewicz homomorphism . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Compression criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 The long exact sequence of a pair . . . . . . . . . . . . . . . . . . . . . . . 7
1.7 Maps are pairs (up to homotopy) . . . . . . . . . . . . . . . . . . . . . . . 9
1.8 CW complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.9 Weak homotopy equivalences and Whitehead's Theorem . . . . . . . . . . 12
1.10 Connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.11 Cellular approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.12 CW approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.13 Hurewicz's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.14 EilenbergMac Lane spaces and cohomology . . . . . . . . . . . . . . . . . 22
1.15 Fibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.16 Comparing bres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.17 Function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.18 The MoorePostnikov tower . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.19 A motivating strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2 The Serre spectral sequence and applications 38

2.1 Spectral sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2 The Serre spectral sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3 The Serre spectral sequence in cohomology . . . . . . . . . . . . . . . . . . 49
2.4 Multiplicative structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.5 mod C theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.6 The transgression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.7 Freudenthal's suspension theorem . . . . . . . . . . . . . . . . . . . . . . . 63
3 Cohomology operations 66
3.1 Steenrod squares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2 Vector elds on spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3 Wu and StiefelWhitney classes . . . . . . . . . . . . . . . . . . . . . . . . 71
3.4 Constructing the Steenrod squares . . . . . . . . . . . . . . . . . . . . . . 74
3.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Last updated March 30, 2021. Corrections to or257@cam.ac.uk.

Chapter 1
Homotopy groups, CW complexes, and brations

Conventions. The term map means continuous function. We write I = [0, 1], I n for
the n-fold product, and
∂I n = {(t1 , t2 , . . . , tn ) ∈ I n | some ti is 0 or 1}.
We write D d ⊂ Rd for the unit disc, and S d−1 ⊂ Dd ⊂ Rd for the unit sphere. In
particular
D0 = {0}, S 0 = {±1}, D−1 = ∅, S −1 = ∅.
A homotopy between maps f0 , f1 : X → Y is a map F : X × [0, 1] → Y such that
F (−, 0) = f0 (−) and F (−, 1) = f1 (−). If A ⊂ X is a subspace we say that the homotopy
F is relative to A if F (a, t) is independent of t for all a ∈ A.
We write [X, Y ] := {maps f : X → Y }/homotopy for the set of homotopy classes of
maps from X to Y .
1.1 Homotopy groups

Lecture 1
Let (X, x0 ) be a based space, i.e. a space with a distinguished point x0 ∈ X .

Denition 1.1.1. Let the nth homotopy group πn (X, x0 ) denote the quotient of the
set of maps
f : I n −→ X such that f (∂I n ) ⊂ {x0 },
by the equivalence relation ', where f0 ' f1 if there is a homotopy
F : I n × [0, 1] −→ X
such that
(i) F (∂I n × [0, 1]) ⊂ {x0 },
(ii) F (−, 0) = f0 (−), and F (−, 1) = f1 (−).

Let us condense this denition a bit. For a pair (X, A) of a space X and a subspace
A ⊂ X , and another pair (Y, B), a map of pairs f : (X, A) → (Y, B) is a map f : X → Y
such that f (A) ⊂ B . A homotopy between maps of pairs, F : (X, A)×[0, 1] → (Y, B)
is a homotopy F : X × [0, 1] → Y such that F (A × [0, 1]) ⊂ B . In these terms we can
just say that
πn (X, x0 ) = {homotopy classes of maps f : (I n , ∂I n ) → (X, {x0 })}.
Sometimes we will write elements of these sets as [f ], to emphasise that they are equiv-
alence classes and that f is merely a representative, but often we will not.
1
2 Chapter 1 Homotopy groups, CW complexes, and brations
Example 1.1.2. If n=1 then π1 (X, x0 ) is the usual fundamental group (but so far we
have only descried it as a set, not as a group).
If n=0 I 0 = {∗} and ∂I 0 = ∅, π0 (X, {x0 })
then, as is the set of path components of
X (and is independent of x0 ). 4
For n≥1 we dene a composition law · on πn (X, x0 ) by the formula
(
f (2t1 , t2 , . . . , tn ) 0 ≤ t1 ≤ 1/2
(f · g)(t1 , t2 , . . . , tn ) = . (1.1.1)
g(2t1 − 1, t2 , . . . , tn ) 1/2 ≤ t1 ≤ 1
Theorem 1.1.3. The operation · is well-dened on πn (X, x0 ), and endows this set with
the structure of a group with identity element given by the constant map:
constx0 : I n −→ X
(t1 , t2 , . . . , tn ) 7−→ x0 .
Furthermore, if n≥2 then this group is abelian.
Proof. The rst part is identical to the argument that the fundamental group is indeed
a group. For the second part we use the homotopy shown below
"
x.tt#iiEHiiiFttEH
Figure 1.1
to swap the order of operations.

r .
If ϕ : (X, x0 ) → (Y, y0 ) is a based map then
ϕ∗ : πn (X, x0 ) −→ πn (Y, y0 )
[f ] 7−→ [ϕ ◦ f ]
denes a group homomorphism. If ψ : (Y, y0 ) → (Z, z0 ) is another based map then
(ψ ◦ ϕ)∗ = ψ∗ ◦ ϕ∗ , and (IdX )∗ = Idπn (X,x0 ) .
1
If two maps ϕ0 , ϕ1 : (X, x0 ) → (Y, y0 ) are based homotopic then the evident homo-
topy shows that (ϕ0 )∗ = (ϕ1 )∗ . From this it is immediate that if ϕ has a based homotopy
inverse (i.e. a based map ψ : (Y, y0 ) → (X, x0 ) such that ψ ◦ ϕ and ϕ ◦ ψ are both based
homotopic to the identity) then ϕ∗ is an isomorphism. In fact, if ϕ is any homotopy
equivalence then ϕ∗ : πn (X, x0 ) → πn (Y, y0 ) is an isomorphism: this is proved in the
same way as for the fundamental group; it is Example Sheet 1 Q1.
"
"
¥
1
i.e. homotopic as maps of pairs
" 1.2 Change of basepoint 3
1.2
If u:I→X
Change of basepoint
we get a map of pairs x.tt#iiEHiiiFttEH

is a path from x0 to x1 , and f : (I n , ∂I n ) → (X, x1 )
u# (f ) : (I n , ∂I n ) −→ (X, x0 )
given graphically as shown in Figure 1.2 below: that is, apply f

is a map of pairs, then
to a smaller cube, and

interpolate between the smaller and larger cube using u on each radial ray.
r .
Figure 1.2
"
"
¥
This construction is well-dened on homotopy classes, and yields a function
u# : πn (X, x1 ) −→ πn (X, x0 )
satisfying
(i) u# = u0# if u is homotopic to u0 relative to their end-points,
(ii) (constx0 )# is the identity,
(iii) u# is a group homomorphism,
(iv) if v : I → X is a path from x1 to x2 and u·v is the concatenation of these paths,

then (u · v)# = u# ◦ v# .
" '
These properties all follow from fairly obvious homotopies.
÷÷:
It follows that if x0 and x1 lie in the same path component, then the groups πn (X, x0 )
and πn (X, x1 ) are isomorphic, but a choice of path from x0 to x1 is necessary to obtain
*
a specic isomorphism.
In particular, taking x1 = x0 gives a left action of the group π1 (X, x0 ) on each of
the groups πn (X, x0 ) for n ≥ 1, which for n = 1 is simply the action by conjugation:
"
u# (f ) = u · f · u−1 . Thus for n≥2 the abelian group πn (X, x0 ) has the structure of a
x.tt#iiEHiiiFttEH
Z[π1 (X, x0 )]-module2 .
1.3 Relative homotopy groups

Let (X, A) be a pair and x0 ∈ A be a basepoint. Let un−1 ⊂ ∂I n denote the closure of
the complement of I n−1 × {0} ⊂ ∂I n .
Denition 1.3.1. For n≥1 let πn (X, A, x0 ) denote the quotient of the set of maps
f : I n −→ X such that f (∂I n ) ⊂ A and f (un−1 ) ⊂ {x0 }
by the equivalence relation given by homotopies through such maps. Extending the
r .
notation for maps of pairs, we can write such a map as a map of triples
f : (I n , ∂I n , un−1 ) −→ (X, A, {x0 }).
n ≥ 2 the formula (1.1.1) denes a group structure on πn (X, A, x0 ), with identity

For
element given by the constant map constx0 , and for n ≥ 3 the homotopy of Figure 1.1
shows that this group is abelian. For n = 1 the sets π1 (X, A, x0 ) do not have a group
structure, but constx0 still provides a canonical element of these sets. Analogously to
Figure 1.2 above, if u : I → A is a path from x0 to x1 then the following gure describes
the change-of-basepoint isomorphism in relative homotopy groups, so that in particular
π1 (A, x0 ) acts on each πn (X, A, x0 ).
"
"
¥
Maps of triples ϕ : (X, A, x0 ) → (Y, B, y0 ) induce functions ϕ∗ : πn (X, A, x0 ) →
πn (Y, B, y0 ) with all the properties one expects:
(i) If
"ϕ '
' ϕ1 through such maps then (ϕ0 )∗ = (ϕ1 )∗ ,

0
÷÷:
(ii) ϕ∗ is a homomorphism for n ≥ 2,
*
2
A group G Z[G] whose elements are nite Z-linear sums of elements of G, and
has a group ring
whose multiplication is determined by Z-linearity and the group structure of G. A left Z[G]-module is
precisely the same as an abelian group with a left G-action by group homomorphisms.
"
"
(iii)
Remark 1.3.2
¥
(ψ ◦ ϕ)∗ = ψ∗ ◦ ϕ∗ .
(Cubes and discs) . As

1.4
I n /∂I n ∼
= Sn,
The Hurewicz homomorphism
with ∂I n /∂I n
∗ ∈ S n , we may equally well take πn (X, x0 ) to be given by the homotopy classes of based
n
maps f : (S , ∗) → (X, x0 ). The group operation is then given as follows:
5
giving a basepoint
" '
*
÷÷:
(I n /un−1 , ∂I n /un−1 ) ∼
Similarly, as = (Dn , S n−1 ), we may take πn (X, A, x0 ) to be
given by
n
homotopy classes of maps of triples f : (D , S
n−1 , ∗) → (X, A, x ). The group
0
operation is given as follows:
"
.
:÷÷÷ 8. HAH
We will use these points of view interchangeably.
° 1.4 The Hurewicz homomorphism
;i
'
We know that Hn (Dn , S n−1 ; Z) is isomorphic to Z, with generator un . As homotopic

.
maps induce equal maps on homology, we may therefore dene a function
* am
h : πn (X, A, x0 ) −→ Hn (X, A; Z)
[f ] 7−→ f∗ (un ).
U
By contemplating the previous gure, this is a homomorphism: it is called the Hurewicz
homomorphism. For a while this will be the only tool we have to show that elements
of πn (X, A, x0 ) are nonzero, but we will develop more tools.
The action ofπ1 (A, x0 ) on πn (X, A, x0 ) is shown in the following gure, which makes
7. clear that the
&
Hurewicz homomorphism is insensitive to this action, i.e. h(u# ([f ])) =
l
h([f ]).
;
.
:÷÷÷ 8. HAH
° "
;i 8.
'
.
:÷÷÷ * am
1.5 Compression criterion
7. The constant map constx0 : (I n , ∂I n , un−1 ) → (X, A, x0 )

& represents an element of
l
πn (X, A, x0 ).
°
When is [f ] = [constx0 ] in πn (X, A, x0 )? To answer this we adopt the
;i
'
disc perspective.
Lemma 1.5.1. A mapf : (Dn , S n−1 , ∗) → (X, A, x0 ) represents [constx0 ] ∈ πn (X, A, x0 )
;
if and only if f : D → X is.homotopic relative to S n−1 to a map with image in A.
n
Lecture 2
Proof. Suppose rst that f is homotopic relative to S n−1 to a map g which has image
in A. Such a homotopy shows that [f ] = [g] ∈ πn (X, A, x0 ). Now choose a deformation
retraction
3
U
r : Dn × [0, 1] → Dn
of Dn∗ ∈ S n−1 ⊂ Dn , and consider g ◦ r : Dn × [0, 1] → X . At time 0 this is the map
to
g , and at time 1 it is constx0 : furthermore, for all times it sends S n−1 into A (as the
:E÷:÷
"∗ throughout). Thus this homotopy
i it
map g has image in A) and sends ∗ to x0 (as r xes
' shows that [g] = [constx0 ].
7. &
: . : l
;
Now suppose that [f ] = [constx0 ], and let H : Dn × [0, 1] → X be a homotopy giving
this identity, so H(D
n × {1}) ⊂ {x0 } ⊂ A and H(S n−1 × [0, 1]) ⊂ A. As Dn × [0, 1]
3
In this course a deformation retraction of X to a subspace A is a homotopy from the identity map
of X to a map into A, and this homotopy should be constant on A.
"
'
:
i deformation retracts to U
Dn × {1} ∪ S n−1 × [0, 1],
* am
1.6 The long exact sequence of a pair
by radial projection from (0, −1)

7
as in
Figure 1.3, it follows that f (−) = H(−, 0) is homotopic relative to S n−1 to a map into
A.
&
l
Figure 1.3
:E÷:÷
"
1.6 The long exact sequence of a pair
:
The map of pairs
induce maps
i : (A, x0 ) → (X, x0 )
.
i∗ : πn (A, x0 ) → πn (X, x0 )
and map of triples
and
:
j : (X, x0 , x0 ) → (X, A, x0 )
j∗ : πn (X, x0 ) → πn (X, A, x0 ).
In addition, given a map f : (I n , ∂I n , un−1 ) → (X, A, x0 ) its restriction to I n−1 × {0} ⊂

∂I n gives a map ∂f : (I n−1 , ∂I n−1 ) → (A, x0 ), inducing a function
∂ : πn (X, A, x0 ) −→ πn−1 (A, x0 ).
If n≥2 then ∂(f · g) = ∂f · ∂g , so this function is a homomorphism in this case.

Theorem 1.6.1. The sequence
··· πn+1 (X, A, x0 )

∂
i∗ j∗
πn (A, x0 ) πn (X, x0 ) πn (X, A, x0 )
∂
i∗ j∗
πn−1 (A, x0 ) πn−1 (X, x0 ) ···
i∗ j∗
··· π1 (X, x0 ) π1 (X, A, x0 )
∂
i∗
π0 (A, x0 ) π0 (X, x0 )
is exact. Exactness at the last three positions must be interpreted carefully: it means that
Im(j ∗ ) = ∂ −1 ([constx0 ]) and Im(∂) = (i∗ )−1 ([constx0 ]).
This long exact sequence is natural with respect to maps ϕ : (X, A, x0 ) → (Y, B, y0 ).
Proof.
Exactness at πn (X, x0 ): f : (I n , ∂I n ) → (X, x0 ) be such that j∗ ([f ]) = [constx0 ].
Let
By the compression criterion this means that f is homotopic relative to ∂I to a map
n
with image in A, so [f ] ∈ Im(i∗ ). The compression criterion also shows that j∗ ◦ i∗ = 0.
Exactness at πn (X, A, x0 ): We have ∂j∗ ([f ]) = [f |I n ×{0} ] = [constx0 ] because f sends

∂I n to x0 . Now let f : (I n , ∂I n , un−1 ) → (X, A, x0 ) be such that ∂f : (I n−1 , ∂I n−1 ) →
(A, x0 ) is homotopic to constx0 , say via a homotopy H : I
n−1 × [0, 1] → A. Form the
map
g : (I n , ∂I n ) −→ (X, x0 )
(
H(t2 , . . . , tn ; 1 − 2t1 ) 0 ≤ t1 ≤ 1/2
(t1 , t2 , . . . , tn ) 7−→
f (2t1 − 1, t2 , . . . , tn ) 1/2 ≤ t1 ≤ 1.
These indeed glue as H(t2 , . . . , tn ; 0) = (∂f )(t2 , . . . , tn ) = f (0, t2 , . . . tn ), and g indeed

takes the value x0 if any ti is 0 or 1. See Figure 1.4 (a).
Thus [g] ∈ πn (X, x0 ), and we can form j∗ ([g]) ∈ πn (X, A, x0 ). This is equal to
[f ] ∈ πn (X, A, x0 ) via the homotopy of triples
G : (I n , ∂I n , un−1 ) × [0, 1] −→ (X, A, x0 )
(
H(t2 , . . . , tn ; 1 − 2t1 ) 0 ≤ t1 ≤ 1/2 · s
(t1 , t2 , . . . , tn , s) 7−→ 1 −s
f ( 2t2−s , t2 , . . . , tn ) 1/2 · s ≤ t1 ≤ 1,
as the homotopy H has image in A. See Figure 1.4 (b).
Exactness at πn−1 (A, x0 ): For f : (I n , ∂I n , un−1 ) → (X, A, x0 ), the element i∗ ∂([f ]) is

zero because
∂f i
(I n−1 × {0}, ∂I n−1 × {0}) −→ (A, x0 ) −→ (X, x0 )
l
;
1.7 Maps are pairs (up to homotopy) 9
:E÷:÷
"
i it
'
: . :
Figure 1.4
is homotopic to constx0 via the homotopy f ! Conversely, if g : (I n−1 , ∂I n−1 ) → (A, x0 )

has i ◦ g homotopic to constx0 via a homotopy H : (I
n−1 , ∂I n−1 ) × [0, 1] → (X, x ), then
0
n
the map H : I → X satises
H|I n−1 ×{0} = g, H(∂I n ) ⊂ A, H(un−1 ) ⊂ {x0 },
and so [H] ∈ πn (X, A, x0 ) satises ∂([H]) = [g].
For a path u : [0, 1] → A, the change-of-basepoint bijections
u# : πn (X, A, x0 ) −→ πn (X, A, x1 ),
u# : πn (X, x0 ) −→ πn (X, x1 ) and u# : πn (A, x0 ) −→ πn (A, x1 )

give a map of long exact sequences from that based at x0 to that based at x1 . In particular
π1 (A, x0 ) acts on the terms of the long exact sequence of Theorem 1.6.1, making it a
long exact sequence of Z[π1 (A, x0 )]-modules (and of sets-with-a-π1 (A, x0 )-action near the
bottom).
1.7 Maps are pairs (up to homotopy)

A pair of spaces (Y, B) in particular gives an (inclusion) map i:B →Y.
If f :X→Y is a map, let
Mf := (X × [0, 1] t Y )/(x, 1) ∈ X × [0, 1] ∼ f (x) ∈ Y,
called the mapping cylinder of f .

There is an inclusion
i : X −→ Mf
x 7−→ [(x, 0)]
"
Him .
and linear interpolation gives a deformation retraction
f#FtI
""
"
r : Mf × [0, 1] −→ Mf " "'
([y], t) 7−→ y
([x, s], t) 7−→ [(x, s(1 − t) + t)]
to the subspace Y ⊂ Mf . Thus the pair (Mf , X) has inclusion map satisfying
i
X Mf
' r(−,0)
" f
Y
In other words, Mf
.
is a replacement of Y, up to homotopy equivalence, for which the
map f :X→Y is represented by the inclusion of a subspace.
Using this device we may consider any map f : X → Y as being, morally, the inclusion
of a subspace, and so can pretend that (Y, X) is a pair: we really mean the pair (Mf , X).
For example, for x0 ∈ X we can dene
πn (Y, X, x0 ) := πn (Mf , X, x0 ),
" '
whereupon, using πn (Y ) ∼
= πn (Mf ), we get a long exact sequence
. . .
.
··· πn+1 (Y, X, x0 )
∂
f∗ j∗
πn (X, x0 ) πn (Y, f (x0 )) πn (Y, X, x0 )
∂
f∗ j∗
πn−1 (X, x0 ) πn−1 (Y, f (x0 )) ···
1.8 CW complexes
Denition 1.8.1. A CW complex is a space X obtained by the following process.
(i) Let X −1 := ∅.
1.8 CW complexes 11
(ii) Supposing the space X n−1 has been dened, let {ϕα : S n−1 → X n−1 }α∈In be a set
of maps, and let
X n := (X n−1 t (In × Dn ))/(α, x) ∈ In × S n−1 ∼ ϕα (x) ∈ X n−1 ,
with the quotient topology.
We let enα be the image of {α} × int(Dn ) under the quotient map, called an open cell.
n
Let X := ∪n≥0 X with the following topology: a set S ⊂ X is open (resp. closed) if
n 4
and only if each S ∩ X is open (resp. closed) . We call X the n-skeleton of X .
n
A pair of spaces (X, A) is a relative CW complex if X is obtained as above but

starting from X
−1 := A.
If X is a CW complex then a subcomplex is a closed subspace A ⊂ X which is a

union of open cells. In this case A is a CW complex in its own right, and (X, A) is called
a CW pair.
Denition 1.8.2. A pair (X, A) has the homotopy extension property if given a
map f :X→Y and a homotopy H : A × [0, 1] → Y such that H(−, 0) = f |A (−), there
exists a homotopy H : X × [0, 1] → Y such that H 0 |A×[0,1] = H and H 0 (−, 0) = f (−).
0
5
Equivalently , we start with the data of a map
f ∪ H : (X × {0}) ∪ (A × [0, 1]) −→ Y
and ask for an extension to a map H 0 : X × [0, 1] → Y .

Theorem 1.8.3. A relative CW complex (X, A) has the homotopy extension property.
Proof. As a preliminary step, observe that just as in Figure 1.3 the space Dn × [0, 1]
(deformation) retracts to
n n−1
(D × {0}) ∪ (S × [0, 1]) via radial projection from (0, 2). It
n
follows that the pair (D , S
n−1 ) has the homotopy extension property: the initial data
is a map
f ∪ H : (Dn × {0}) ∪ (S n−1 × [0, 1]) −→ Y
and precomposing with the retraction r : Dn × [0, 1] → (Dn × {0}) ∪ (S n−1 × [0, 1]) gives
the required extension. Lecture 3
Now, let (X, A) be a relative CW complex, f : X → Y and H : A × [0, 1] → Y be

given. We will construct compatible maps H n : X n × [0, 1] → Y by induction over n: we
start with H −1 := H .
Suppose H
n−1 is given. Now
(X n−1 × [0, 1]) t (In × Dn × [0, 1])

X n × [0, 1] = ,
(α, x, t) ∈ In × Dn × [0, 1] ∼ (ϕα (x), t) ∈ X n−1 × [0, 1]
so the retraction r : Dn × [0, 1] → (Dn × {0}) ∪ (S n−1 × [0, 1]) for each α ∈ In gives a
retraction
rn : X n × [0, 1] −→ (X n × {0}) ∪ (X n−1 × [0, 1]),
4
So a function f : X → Y is continuous if and only if all its restrictions f |X n : X n → Y are
continuous.
5
Though this equivalence is subtle if A⊂X is not closed, see Proposition A.18 of Hatcher's book.
and we can set H n := (f |X n ∪ H n−1 ) ◦ rn . By denition of the topology on X , the map

H0 := ∪n≥0 Hn : X × [0, 1] = ∪n≥0 X n × [0, 1] → Y is continuous, and gives the required
extension.
This argument shows the most useful point of working with CW complexes: we can
make arguments by induction over cells, or skeleta, and often reduce to proving things
for Dn . The following is another example of this strategy.
Lemma 1.8.4 (Compression Lemma) . Let(X, A) be a relative CW complex and (Y, B)

be a pair with B 6= ∅. Suppose that for each n such that X has an n-cell relative to A,
πn (Y, B, y0 ) = 0 for all y0 ∈ B . Then any map f : (X, A) → (Y, B) is homotopic relative
to A to a map into B .
Proof. We will construct compatible homotopies
H n : X n × [0, 1] −→ Y
relative to A, such that H n |X n ×{0} = f |X n and H n (X n × {1}) ⊂ B . We start by taking

H −1 (x, t) = f (x) to be the constant homotopy on X 0 .
If H n−1 has been dened let In index the n-cells of X relative to A. For each α ∈ In
we have a map
mα := f |{α}×Dn ∪ H n−1 |{α}×S n−1 ×[0,1] : {α} × (Dn × {0}) ∪ (S n−1 × [0, 1]) −→ Y.

The domain is homeomorphic to Dn , and mα (S n−1 × {1}) ⊂ H n−1 (X n−1 × {1}) ⊂ B , so

mα represents an element of πn (Y, B, mα (α, ∗, 1)). As this homotopy group vanishes by
assumption, by the compression criterion (Section 1.5) the map mα is homotopic relative
to {α} × S
n−1 × {1} to a map into B : this homotopy gives an extension to a map
Mα : {α} × Dn × [0, 1] −→ Y
satisfying Mα ({α} × Dn × {1}) ⊂ B . As
(X n−1 × [0, 1]) t (In × Dn × [0, 1])

X n × [0, 1] =
(α, x, t) ∈ In × Dn × [0, 1] ∼ (ϕα (x), t) ∈ X n−1 × [0, 1]
H n = (H n−1 t α∈In Mα )/ ∼.
F
we can let
1.9 Weak homotopy equivalences and Whitehead's Theorem

Denition 1.9.1. A map f : X → Y is a weak homotopy equivalence if the functions
f∗ : πn (X, x0 ) → πn (Y, f (x0 )) are bijections for all n and for all basepoints x0 ∈ X .
More generally, we say that spaces X and Y are weakly homotopy equivalent if
there exists a zig-zag
X −→ Z1 ←− Z2 −→ Z3 ←− · · · −→ Zn ←− Y
of weak homotopy equivalences.

1.9 Weak homotopy equivalences and Whitehead's Theorem 13
We have seen (Example Sheet 1 Q1) that homotopy equivalences are weak homotopy
equivalences. Our goal is to show that the converse is also true, as long as the spaces
involved are CW complexes.
Theorem 1.9.2. Let (Z, A) be a relative CW complex, f :X→Y be a weak homotopy

equivalence, and a commutative square
h
A X
i f
g
Z Y
be given. There there is a map ĝ : Z → X such that
(i) ĝ ◦ i = h,
(ii) f ◦ ĝ is homotopic to g, relative to A.
Proof. Suppose rst that f :X →Y is the inclusion of a subspace, so that (Y, X) is a

pair. The long exact sequence
··· πn+1 (Y, X, x0 )

∂
f∗ j∗
πn (X, x0 ) πn (Y, f (x0 )) πn (Y, X, x0 )
∂
f∗ j∗
πn−1 (X, x0 ) πn−1 (Y, f (x0 )) ···
then shows that πn (Y, X, x0 ) = 0 for all x0 ∈ X . The claim then follows from the
Compression Lemma (Lemma 1.8.4).
For a general map f :X→Y we reduce to the case above following Section 1.7. Let I gave a fallacious ver-
sion of this argument
Mf denote the mapping cylinder of f, with inclusions inc : X → Mf and j : Y → Mf in lectures: here it is
now corrent.
and retraction r : Mf → Y . Consider the diagram
h
A X X X
i f inc f
g j r
Z Y Mf Y,
in which the left and right squares commute (using that r ◦ j = IdY ) and the middle
square commutes up to homotopy. Combining the two leftmost squares gives a new
square
h
A X
i inc
g 0 :=j◦g
Z Mf
which also commutes up to homotopy. Using the homotopy extension property, we may
change g0 by a homotopy to a map g 00 making the square literally commute. This puts
us in a situation to apply the version of the statement proved above, as inc : X → Mf
is the inclusion of a subspace. Thus there is a map ĝ 00 : Z → X such that ĝ 00 ◦ i = h and
such that inc ◦ ĝ 00 ' g 00 ' g 0 relative to A. Applying r to this homotopy, it follows that
f◦ ĝ 00 is homotopic to g relative to A, as required.
Corollary 1.9.3 (Whitehead's theorem) . If f :X→Y is a weak homotopy equivalence

between CW complexes, then it is a homotopy equivalence.
Proof. First apply the previous theorem to
∅ X
f
IdY
Y Y,
which provides a map g : Y → X such that f ◦ g is homotopic to IdY . We claim that g

is a homotopy inverse to f so we must also show that g ◦ f is homotopic to IdX .
As f ◦ g is homotopic to IdY , there is a homotopy H from f ◦ (g ◦ f ) = (f ◦ g) ◦ f to
f ◦ IdX = IdY ◦ f . Thus we may form the commutative square
f ◦gtIdX
X × {0} t X × {1} X
inc f
H
X × [0, 1] Y.
The left-hand map is the inclusion of a sub-CW complex, and the right-hand map f is
still a weak homotopy equivalence, so by the previous theorem we may nd is a map
Ĥ : X × [0, 1] → X such that Ĥ|X×{0} = f ◦ g and Ĥ|X×{1} = IdX , as required.
There is another consequence of Theorem 1.9.2 with further justies focussing on the
notion of weak homotopy equivalences.
Corollary 1.9.4. A weak homotopy equivalence induces an isomorphism on homology

Lecture 4 and cohomology with any coecients.
Proof sketch. It is enough to prove that a weak homotopy equivalence f :X→Y induces
an isomorphism on homology with Z coecients: it then follows for (co)homology with
any coecients by Universal Coecient Theorems. By replacing Y by the mapping
cylinder Mf , it is also enough to treat the case where f is the inclusion of a subspace.
In this case we need to show that Hn (Y, X; Z) = 0.
σ : ∆n → Y are singular
P
Let [ nσ · σ] ∈ Cn (Y, X) be a relative cycle, where
simplices. We must show that this cycle represents zero in Hn (Y, X; Z). By considering
nσ · σ as |nσ | copies of σ (or of −σ ), we may glue copies of ∆n together along faces as
follows. If τ ⊂ σ is a (n − 1)-dimensional face which is not sent into X , then because
∂( nσ · σ) ∈ Cn−1 (X) there is another simplex σ 0 which also has τ as a face. Let
P
1.10 Connectivity 15
K denote a simplicial complex obtained by gluing together in this manner all (n − 1)-
dimensional faces which are not mapped into X, and let L⊂K denote the union of the
unglued (n − 1)-dimensional faces. By construction there is a map of pairs
f : (|K|, |L|) −→ (Y, X).
Now
x := sumn-simplices of K ∈ Cn (|K|, |L|)
of all
P
is a cycle, and by construction f∗ (x) = [ nσ · σ] ∈ Hi (Y, X). Now as (|K|, |L|) is a
relative CW complex, by Theorem 1.9.2 the map f : (|K|, |L|) → (Y, X) is homotopic
relative to |L| to a map with image in X , and so f∗ (x) = 0.
1.10 Connectivity
Denition 1.10.1. For n ≥ 0 say that a pair (X, A) is n-connected if π0 (A) → π0 (X)
is onto
6 and πi (X, A, x0 ) = 0 for all 1 ≤ i ≤ n and all x0 ∈ A. Equivalently, for all
x0 ∈ A the maps
πi (A, x0 ) −→ πi (X, x0 )
are epimorphisms for i ≤ n and isomorphisms for i < n. More generally, say that a map
f :X→Y n-connected if the pair (Mf , X) is n-connected.
is
Say that a space X is n-connected if the pair (X, {x0 )}) is n-connected for all
x0 ∈ X , i.e. πi (X, x0 ) = 0 for all 0 ≤ i ≤ n and all x0 ∈ X .
Theorem 1.10.2. S n is (n − 1)-connected.
Proof. Let ∗ ∈ S n be a basepoint, i < n, and f : (S i , ∗) → (S n , ∗) be a continuous map,

n
representing [f ] ∈ πi (S , ∗). By the Simplicial Approximation Theorem we may change
f by a homotopy to a map f 0 that is simplicial with respect to some triangulations
= |K| and S n ∼
Si ∼ = |L|; as i < n the map f 0 : S i → S n is not surjective, so lands inside
n n n n
some (int(D ), ∗) ⊂ (S , ∗). As int(D ) deformation retracts to ∗ ∈ int(D ), it follows
0
that f is homotopic, relative to ∗, to a constant map, so [f ] = 0 ∈ πi (S ).
n
Corollary 1.10.3. The pair (Dn , S n−1 ) is (n − 1)-connected.
Proof. The long exact sequence on homotopy groups for this pair, and the fact that
πi (Dn , ∗) = 0, shows that the boundary map
∂ : πi (Dn , S n−1 , ∗) −→ πi−1 (S n−1 , ∗)
is an isomorphism. By the previous theorem the latter group vanishes for i − 1 < n − 1,
so the former group vanishes for i < n.
Corollary 1.10.4. If (X, A) is a relative CW complex only having relative cells of di-
mension < n, then any map f : (X, A) → (Dn , S n−1 ) is homotopic relative to A to a
map into S
n−1 .
6
You can interpret this condition as π0 (X, A, x0 ) = 0, but we did not dene such a relative homo-
topy group.
Proof. Apply the Compression Lemma (Lemma 1.8.4).
Corollary 1.10.5. If (X, A) is a relative CW complex only having relative cells of di-
mension ≥ n, then it is (n − 1)-connected.
Proof. Let i < n and f : (I i , ∂I i , ui−1 ) → (X, A, x0 ) represent an element [f ] ∈

" πi (X, A, x0 ), which we must show is zero, i.e. we must nd a homotopy compressing
Him
f into A.
i i 0
As I is compact so is f (I ), so it lies in a sub relative CW complex (X , A) having
nitely-many cells (see Example Sheet 1 Q5). .
Thus we may suppose without loss of
generality that (X, A) has nitely-many cells. Let A ⊂ X0 ⊂ X be a sub-CW complex
having one fewer relative cell than X. By induction it suces to show that we can
homotope f relative to ∂I n until it has image in X 0.
Thus we may suppose that we are in the following simplied situation: we have a
map g : (I i , ∂I i ) → (X, X 0 ), where X = X 0 ∪ Dk with k ≥ n and i < n, and we wish to
i
homotope g relative to ∂I to that it has image in X .
0
f#FtI
""
"
" "'
Cover X = X 0 ∪ Dk by the open sets U = X 0 ∪ (Dk \ {0}) and V = int(Dk ). Then

g −1 (U ) and g −1 (V ) form an open cover of I i , so by the Lesbegue number lemma we may
i
subdivide I into small cubes each of which maps into U or into V (or into both). Let
" B ⊂ I n be the bad cubes: those closed cubes which map into V . Then g(∂B) ⊂ V ∩ U ,
and so we have a map
g|B : (B, ∂B) −→ (V, V ∩ U ) ∼

= (int(Dk ), int(Dk ) \ {0}) ' (Dk , S k−1 ).
.
As (B, ∂B) is a relative CW complex only having cells of dimension ≤ i < n, and
(V, V ∩ U ) is homotopy equivalent to (Dk , S k−1 ) with k ≥ n, it follows from Corollary
1.10.4 that g|B is homotopic relative to ∂B to a map into U ∩ V , and hence that g is
homotopic to a map into U . Finally, U deformation retarcts to X .
0
1.11 Cellular approximation

Corollary 1.11.1 (Cellular approximation) . If f : (X, A) → (Y, B) is a map between
" '
relative CW complexes, then it is homotopic to a map f¯ satisfying f¯(X n ) ⊂ (Y n ) for all
n ≥ 0.
Proof. See Example Sheet 1 Q7.
. . .
.
1.12 CW approximation 17
Corollary 1.11.2. If X
is a CW complex and ϕ : S
n−1 → X is an attaching map for
an n-cell then ϕ is homotopic to a map into X
n−1 , so X ∪ϕ Dn is homotopy equivalent
to a CW complex.
1.12 CW approximation
From the point of view of weak homotopy equivalence, every space is equivalent to a CW
complex:
Theorem 1.12.1. For any space X there is a CW complex C and a weak homotopy
equivalence f : C → X.
Furthermore, if g : D → X is another weak homotopy equivalence from a CW com-
plex, there is a weak homotopy equivalence φ:C→D such that g ◦ φ ' f.
Lecture 5
Proof. Suppose that X is path connected (otherwise repeat the below for each path
component), and choose a basepoint x0 . We will construct CW complexes C0 ⊂ C1 ⊂
C2 ⊂ · · · and maps f n : Cn → X such that
(i) Cn is n-dimensional, and
(ii) fn is f∗n : πi (C n , ∗) → πi (X, x0 ) is an isomorphism for i < n and

n-connected, i.e.
an epimorphism for i = n.
n : C := n
S S
Then the map f := n≥0 f n≥0 C → X induces an isomorphism on all
homotopy groups and so is a weak homotopy equivalence.
We start with C 0 = {∗} and f 0 (∗) = x0 . Assuming that C n−1 has been constructed,
let {ψα : (S n−1 , ∗) → (C n−1 , ∗)}α∈In be a set of maps such that [ψα ] ∈ πn−1 (C n−1 , ∗)
generate
Ker(f∗n−1 : πn−1 (C n−1 , ∗) → πn−1 (X, x0 )),
and let [
n
Cprelim := C n−1 ∪ Dαn
α∈In
be the CW complex obtained by attaching n-cells to C n−1 along the maps ψα . As

f n−1 ◦ ψα : (S n−1 , ∗) → (X, x0 ) is nullhomotopic, a choice of nullhomotopy gives an
extension of this map over the cell Dαn , and so an extension of f n−1 to a map
n n
fprelim : Cprelim −→ X.
n
fprelim
i
Consider the factorisation f n−1 : C n−1 → Cprelim
n → X as giving
n
(fprelim )∗
i∗
f∗n−1 : πn−1 (C n−1 , ∗) n
πn−1 (Cprelim , ∗) πn−1 (X, x0 ).
The map f∗n−1 is surjective by assumption. The map i∗ is surjective, because we have
πn−1 (Cprelim , C n−1 , ∗) = 0 by an application of cellular approximation (Corollary 1.11.1).
n
On the other hand the ψα map to 0 under i∗ by construction, and as these generate the
kernel of f∗n−1 it follows that
n
(fprelim )∗ is a bijection. For i<n−1 we have
n
(fprelim )∗
i∗
f∗n−1 : πi (C n−1 , ∗) n
πi (Cprelim , ∗) πi (X, x0 ),
the map f∗n−1 is an isomorphism by assumption, and the map i∗ is an isomorphism

n
as both πi (Cprelim , C
n−1 , ∗) and π n n−1 , ∗) vanish by cellular approximation
i+1 (Cprelim , C
n
(Corollary 1.11.1): it follows that (fprelim )∗ is an isomorphism too.
n
Now let {ϕα : (S , ∗) → (X, x0 )}β∈Jn be a set of maps such that [ϕβ ] ∈ πn (X, x0 )
generate this group, let
_
C n := Cprelim
n
∨ Sβn
β∈Jn
n-cell trivially for each β ∈ Jn , and set f n := fprelim

n
W
be obtained by attaching an ∨ ϕβ :
Cn → X. By
n n
construction the map f∗ : πn (C , ∗) → πn (X, x0 ) is an epimorphism.
Consider the maps
n−1 n i∗ f∗n
(fprelim )∗ : πi (Cprelim , ∗) πi (C n , ∗) πi (X, x0 ).
The map i∗ is a split monomorphism as the inclusion → C n has a retraction

n
i : Cprelim
n
(by collapsing all Sβ 's to the basepoint).
n n
As πi (C , Cprelim , ∗)
= 0 for i ≤ n − 1 by
cellular approximation (Corollary 1.11.1), it follows that the map i∗ is an isomorphism
n
for i ≤ n − 1. Together with the fact that (fprelim )∗ is also an isomorphism for i ≤ n − 1,
n
it follows that f∗ is an isomorphism in this range too. This nishes the construction of
n n
the C 's and f 's.
For the last part, apply Theorem 1.9.2 to
∅ D
g
f
C X
to get φ:C→D such that g ◦ φ ' f; as f and g are weak homotopy equivalences, so is
φ.
Corollary 1.12.2. If X is n-connected, there is a weak homotopy equivalence f :C→X

where C has a single 0-cell, and all other cells of dimension > n.
Proof. As in the last proof, but start with {∗} = C 0 = C 1 = · · · = C n .
1.13 Hurewicz's Theorem

By Theorem 1.10.5, if (X, A) is a relative CW complex only having cells of dimension
≥ n then πi (X, A, x0 ) = 0 for i < n. What about πn (X, A, x0 )? We start by considering
the situation where (X, A) consists of a single relative n-cell.
"
Him 1.13
.
Hurewicz's Theorem 19
Theorem 1.13.1. If X = A ∪ Dn , then the map
Φ : (Dn , S n−1 , ∗) −→ (X, A, x0 )
f#FtI
""
" given by the
Proof. Let
n-cell generates πn (X, A, x0 )
'
as a
[f ] ∈ πn (X, A, x0 ) with n ≥ 3; our goal is to"write

"
Z[π1 (A, x0 )]-module.7
it as a Z-linear combination
of elements γ# ([Φ]) for γ ∈ π1 (A, x0 ).
We have a map f : (D , S
n n−1 , ∗) → (X, A, x ). There is an open set U = int(D n ) ⊂
0
X , and so f −1 (U ) ⊂ Dn is an open subset contained in the interior. Now f : f −1 (U ) → U
n
is a map between open subsets of R , so it may be homotoped near f
−1 (0) so that it is
smooth and transverse to 0 ∈ U near this set: continue to call the map f . Then there is
a small closed disc 0 ∈ D ⊂ U such that f
−1 (D) = tk D is a disjoint union of closed
j=1 j
discs, each of which is sent homeomorphically to D by f .
"
Let x1 be a point on the boundary of D, and u be a path in Dn ⊂ X from x0 to x1 .

Let yj ∈ Dj be the point that is sent by f to x1 .
n n
Choose paths vj in D from ∗ to yj , and let V = A ∪ (D \ {0}) ⊂ X , another open
" '
set. Then the disc D gives an element
[incD ] ∈ πn (X, V, x1 )
which under the isomorphisms . .

.
. u#
πn (X, A, x0 ) −→ πn (X, V, x0 ) ←− πn (X, V, x1 )
corresponds to [Φ].
Each f ◦ vj is a path from x0 to x1 lying in V , but need not be homotopic to the
path u. However, it is homotopic to γj · u for some [γj ] ∈ π1 (V, x0 ) = π1 (A, x0 ). Thus
from the geometric description of sum of relative homotopy classes (Remark 1.3.2) we
see that
k
X
[f ] = ±(γj )# ([Φ]) ∈ πn (X, V, x0 )
j=1
∼
=
as required, where the signs come from whether the homeomorphisms f |Dj : Dj → D
preserve or reverse orientation.
Lecture 6
Corollary 1.13.2. If X is obtained from A by attaching a single n-cell along a map ϕ :

(S n−1 , ∗) → (A, x0 ), then πi (A, x0 ) → πi (X, x0 ) is an isomorphism for all i < n − 1, and
πn−1 (A, x0 ) → πn−1 (X, x0 ) is an epimorphism with kernel the Z[π1 (A, x0 )]-submodule
generated by [ϕ] ∈ πn−1 (A, x0 ).
8
Proof. The rst two parts follows from the fact that πi (X, A, x0 ) = 0 for i < n, by
Corollary 1.10.5. For the last part we consider the portion of the long exact sequence of
(X, A) given by
∂
πn (X, A, x0 ) πn−1 (A, x0 ) πn−1 (X, x0 ) 0,
which has ∂([Φ]) = [ϕ], and apply the previous theorem.
Corollary 1.13.3. X = A ∪ Dn
If and A is simply connected, then πn (X, A, x0 ) is
isomorphic to Z, generated by [Φ].
Proof. By the theorem this group is generated as a Z-module by [Φ], so it remains to

show that [Φ] has innite order. For this consider the Hurewicz homomorphism
h : πn (X, A, x0 ) −→ Hn (X, A; Z) ∼
=excision Hn (Dn , S n−1 ; Z) = Z
and observe that [Φ] is tautologically sent to the generator [Dn ] ∈ Hn (Dn , S n−1 ; Z). This
does indeed have innite order.
Corollary 1.13.4 (Hopf 's theorem) . We have πn (S n , ∗) ∼

= Z, generated by the identity
map IdS n : Sn → S n . That is, maps f, g : S → S n are homotopic if and
n only if they
have the same degree.
Proof. Take A = {x0 } in the previous corollary.
Corollary 1.13.5. Sαn

W
If X= α∈J is a wedge of n-spheres with n ≥ 2, then
πn (X, ∗) ∼
M
= Z,
α∈J
generated by the inclusions iα : Sαn → X .
Sαn → Sαn is a
W Q
Proof. Suppose rst that J is nite. Then the inclusion α∈J α∈J
sub-CW complex, and the relative cells have dimension ≥ 2n, so this pair is (2n − 1)-
connected. Thus the inclusion induces an isomorphism on πn (−). On the other hand a
map into a product is a product of maps, so
!
∼
Y Y M M
πn Sαn = πn (Sαn ) = πn (Sαn ) = Z.
α∈J α∈J α∈J α∈J
7
For n ≥ 3; for n=2 the π1 (A, x0 )-orbit of [Φ] generates the non-abelian group π2 (X, A, x0 ); for
n=1 there is no sensible statement.
8
For n − 1 ≥ 2; for n−1=1 the kernel is the normal subgroup of π1 (A, x0 ) generated by [ϕ].
1.13 Hurewicz's Theorem 21
X = J 0 ⊂J nite ( J 0 S n ), and as any map from a compact

S W
Now in general we have
space (such as Sn or S n × [0, 1]) has image in a nite subcomplex, and taking homotopy
groups preserves direct limits, the claim follows.
Theorem 1.13.6 (Hurewicz) . Let X be a path-connected space with basepoint x0 , which

is (n − 1)-connected for some n ≥ 2. Then Hi (X; Z) = 0 for 0 < i < n, and the Hurewicz
homomorphism
h : πn (X, x0 ) −→ Hn (X; Z)
is an isomorphism.
Proof. By Corollary 1.12.2 there is a weak equivalence f : C → X from a CW complex

having a single 0-cell, and no other cells of dimension < n. By Corollary 1.9.4 the map
f∗ : Hi (C; Z) → Hi (X; Z) is an isomorphism, but computing with cellular homology
gives Hi (C; Z) = Hi
cell (C; Z) = 0 for 0 < i < n as claimed.
We now consider the Hurewicz homomorphism in degree n. The inclusion i : C

n+1 →
C is (n + 1)-connected, so i∗ : πn (C n+1 , ∗) → πn (C, ∗) is an isomorphism, and i∗ :

Hncell (C n+1 ; Z) → Hncell (C; Z) is clearly an isomorphism too: thus we may suppose that
C = C n+1 is (n + 1)-dimensional and apart from a 0-cell has no cells of dimension < n.
n
As any map from a compact set (such as S or ∆ ) into C
n n+1 intersects the interiors
of nitely-many (n + 1)-cells, we may suppose without loss of generality that C

n+1 has
nitely-many (n + 1)-cells: we proceed by induction on the number of (n + 1)-cells. By

n n
the previous corollary the map h : πn (C , ∗) → Hn (C ; Z) is an isomorphism, providing
the base of the induction. Suppose then that C
n+1 = C 0 ∪ D n and that the Hurewicz
ϕ
0
homomorphism for C is an isomorphism. The Hurewicz homomorphism gives a map of
long exact sequences
Z∼
= πn+1 (C n+1 , C 0 , ∗) πn (C 0 , ∗) πn (C n+1 , ∗) 0 0
∼
= ∼
= ∼
= ∼
=
Z∼
= Hn+1 (C n+1 , C 0 ; Z) Hn (C 0 ; Z) Hn (C n+1 ; Z) 0 0
where the rst vertical map is an isomorphism by Corollary 1.13.3, the next is an iso-
morphism by assumption, and the other two are isomorphisms by observation: it follows
from the 5-lemma that the middle vertical map is an isomorphism too.
Theorem 1.13.7. If X is a path-connected space with basepoint x0 , then the Hurewicz

homomorphism
h : π1 (X, x0 ) −→ H1 (X; Z)
is an epimorphism, with kernel the commutator subgroup of π1 (X, x0 ). That is, this
homomorphism is the abelianisation of π1 (X, x0 ).
S1
W
Proof sketch. It holds for X = J by direct calculation of both sides: the left-hand
side is the free group on J and the right-hand side is the free abelian group on J. Then
proceed as in the proof of the last theorem, looking at the eect on both sides of adding
a cell.
1.14 EilenbergMac Lane spaces and cohomology

Denition 1.14.1. A path-connected based space (X, x0 ) is an EilenbergMac Lane
space of type (G, n) if (
G i=n
πi (X, x0 ) ∼
=
6 n.
0 i=
(Note that if n≥2 then the group G must necessarily be abelian, but if n=1 it need
not be.)
Example 1.14.2. Recall from Q9 on Example Sheet 1 that if π : X → X is a covering

space then π∗ : πi (X, x̄0 ) → πi (X, π(x̄0 )) is an isomorphism for i ≥ 2. Thus if a path-
connected based space (X, x0 ) has contractible universal cover then πi (X, x0 ) = 0 for all
i ≥ 2, and so X is an EilenbergMac Lane space of type (G, 1) for G := π1 (X, x0 ).
In particular,
(i) S1 is EilenbergMac Lane space of type (Z, 1), and
(ii) RP∞ is an EilenbergMac Lane space of type (Z/2, 1). 4

Lecture 7
Lemma 1.14.3. If either n = 1 and G is any group, or n ≥ 2 and G is an abelian group,

then an EilenbergMac Lane space of type (G, n) exists, and can be taken to be a CW
complex.
Proof. Suppose rst that n≥2 and G is abelian. Let
M f M
Z −→ Z −→ G −→ 0
β∈J α∈I
be a partial free resolution of the abelian group G. 9 Form the CW complex X 00 :=

n X 0 by attaching an 1)-cell to X 00 for each
W
α∈I S , then form (n + β ∈ J, along a map
ϕβ with
M
[ϕβ ] = f (1β ) ∈ Z = πn (X 00 , ∗).
α∈I
Now X0 has no cells of dimension< n apart from a 0-cell, so is (n − 1)-connected, and by

= ( α∈I Z)/im(f ) ∼
(X 0 ; Z) = G, so by Hurewicz's theorem πn (X 0 ; ∗) ∼
L
construction Hn =
G. Now form X from X by attaching cells of dimension ≥ n + 2 to kill πi (X 0 , ∗) for all
0
i > n: it is then the required EilenbergMac Lane space.

00 1
W
If n = 1 we choose a presentation G = hI|Ji of the group G, let X = α∈I S ,
0 00
and form X by attaching 2-cells along ϕβ with [ϕβ ] = β ∈ Free(I) = π1 (X , ∗). Then
proceed as above.
9
For example, the free abelian group on G surjects onto G, and its kernel is again free abelian as
any subgroup of a free abelian group is free abelian.
1.14 EilenbergMac Lane spaces and cohomology 23
Let (X, x0 ) be an EilenbergMac Lane space of type (G, n). If n ≥ 2, or more

generally if n ≥ 1 and G is abelian, then by the Hurewicz theorem we have Hn−1 (X; Z) =
0 (or it is the free abelian group Z if n = 1) and an isomorphism
∼
φX : Hn (X; Z) ←− πn (X, x0 ) = G,
so by the Universal Coecient Theorem
∼
(
(((
1 ( Z), G) −→ H n (X; G) −→ Hom (H (X; Z), G) −→ 0
0 −→ (
Ext
(Z( (H ((
(n−1 (X, Z n
the isomorphism φX corresponds to a canonical cohomology class
ιX ∈ H n (X; G).
For any space Y there is therefore a function
[Y, X] −→ H n (Y ; G)
(1.14.1)
[f : Y → X] 7−→ f ∗ (ιX ).
Theorem 1.14.4. If Y is a CW complex, then the function (1.14.1) is a bijection.
Proof. To show surjectivity, let [φ] ∈ H n (Y ; G) ∼ n (Y ; G), so that φ : C cell (Y ) → G

= Hcell n
is a cocycle, i.e. a homomorphism such that φ ◦ d = 0. For each n-cell of Y , φ gives an
element of G = πn (X, x0 ). This determines a map
_
f¯ : Y n /Y n−1 = S n −→ X,
α∈In
and therefore a map

f ¯
f¯n : Y n −→ Y n /Y n−1 −→ X.
satises (f¯ ) (ιX ) = [φ]|Y n ∈ H (Y ; G).
By construction this
n ∗ n n As the restriction map
H n (Y ; G) → H n (Y n ; G) is injective (by considering cellular cohomology, say), it suces
to show that the map f¯ : Y
n n → X extends to a map Y → X .
We rst extend it to Y
n+1 . To do so, note that for each (n + 1)-cell Φ attached along
a map ϕ : S → Y , the composition [f¯ ◦ ϕ] ∈ πn (X, x0 ) = G is precisely φ ◦ d(Φ),
n n n
which vanishes as φ is a cocycle. Thus f extends to a map f¯

¯n n+1 : Y n+1 → X . Now Y is
obtained from Y
n+1 by attaching cells of dimension i > n + 1, but then πi−1 (X, x0 ) = 0
ϕ f¯i−1
so for an i-cell attached along ϕ the composition S i−1 → Y i−1 → X is nullhomotopic
map f¯
so the
n+1 extends to Y .
Injectivity is proved similarly, constructing a nullhomotopy out of a coboundary.
Corollary 1.14.5. If (X, x0 ) is an EilenbergMac Lane space of type (G, n), and (Z, z0 )
is another which is a CW complex, then there is a weak homotopy equivalence f : Z → X.
Proof. The class ιZ ∈ H n (Z; G) corresponds by the previous theorem to a map f :Z→
X ∗
(which we may homotope so that it sends z0 to x0 ) such that f (ιX ) = ιZ . That is,
the map
f∗
G∼
= πn (Z, z0 ) −→ πn (X, x0 ) ∼
=G
is the identity map of G: but as X and Z only have this homotopy group non-trivial, f
is a weak homotopy equivalence.
We will therefore generally write K(G, n) for any space which is an Eilenberg
Mac Lane space of type (G, n): by Lemma 1.14.3 such spaces exist for all (G, n), and by
Corollary 1.14.5 they are unique up to weak homotopy equivalence.
1.15 Fibrations
Denition 1.15.1. Let C be a class of spaces. A map p : E → B has the homotopy
lifting property with respect to C if for each X ∈ C and for each commutative square
f
X × {0} E
p (1.15.1)
H
X × [0, 1] B
there is a homotopy H̃ : X × [0, 1] → E such that
p ◦ H̃ = H and H̃(−, 0) = f (−).
If p has the homotopy lifting property with respect to C = {all spaces} then it is
called a Hurewicz bration. If it has the homotopy lifting property with respect to
C = {D0 , D1 , D2 , D3 , . . .} then it is called a Serre bration.
We call B the base, E the total space, and p
−1 (b) the bre over b ∈ B.
Example 1.15.2. Let p = proj1 : E := B × F → B denote projection to the rst

coordinate. Then given a lifting diagram as above we can take
H̃(x, t) := (H(x, t), proj2 ◦ f (x)).
This satises the required properties, so p is a Hurewicz bration. 4
Example 1.15.3. Any covering space p:B→B is a Hurewicz bration (by the homo-
topy lifting lemma). 4
Example 1.15.4. A Hurewicz bration is a Serre bration. 4
Example 1.15.5. A composition of Hurewicz (or Serre) brations is a Hurewicz (or

Serre) bration. 4
Example 1.15.6. Let p:E→B be a map, and φ : B0 → B be a map. Let
E 0 = B 0 ×B E := {(b0 , e) ∈ B 0 × E | φ(b0 ) = p(e)}
and p0 : E 0 → B 0 be p0 (b0 , e) = b0 . Then p0 is called the pullback of p along φ; it ts

into a diagram
(b0 ,e)7→e
E0 E
p0 p
φ
B0 B
1.15 Fibrations 25
and has the following universal property: Given maps X → E and X → B 0 which become
Lecture 8 equal in B, there is a unique map X→ E 0 making the evident diagram commute.
Suppose now that p has the homotopy lifting property with respect to C , and consider
a lifting problem
f
X × {0} E0
p0
H
X × [0, 1] B0
with X ∈ C. We can incorporate this into the larger commutative diagram
f (b0 ,e)7→e
X × {0} E0 E
p0 p
H φ
X × [0, 1] B0 B,
and the outer square is a homotopy lifting problem for p, X∈C

which can be solved as
and p has the homotopy lifting property with respect to C : let φ ◦ H : X × [0, 1] → E
^
0
be the lift. But by the universal property this produces a lift H̃ : X × [0, 1] → E of the
original lifting problem. Concretely, it is given by
^
H̃(x, t) := (H(x, t), φ ◦ H(x, t)) ∈ E 0 .
This argument shows that the class of maps having the homotopy lifting property
with respect to C is closed under the formation of pullbacks. In particular, the pullback
of a Hurewicz (or Serre) bration is again a Hurewicz (or Serre) bration. We often write
φ∗ E := E 0 and φ∗ p := p0 : φ∗ E → B 0 . 4
Lemma 1.15.7. Let p:E→B be a Serre bration, (X, A) be a relative CW complex,

and a commutative diagram
f
X × {0} ∪ A × [0, 1] E
p
H
X × [0, 1] B
be given. Then there is a map H̃ : X × [0, 1] → E extending f and lifting H.
Proof. Consider rst the case (X, A) ∼

= (Dn , S n−1 ). There is a homeomorphism of pairs
(Dn × [0, 1], Dn × {0} ∪ S n−1 × [0, 1]) ∼

= (Dn × [0, 1], Dn × {0})
which translates this lifting problem to that of (1.15.1) for X = Dn , which can be solved
as p is a Serre bration.
For a general relative CW complex (X, A), we can construct the lift one cell at a time
using the above.
Theorem 1.15.8 (Local-to-global principle for Serre brations) . If p : E → B is a map

such that there is an open cover {Uα }α∈I with each p|p−1 (Uα ) : p−1 (Uα ) → Uα is a Serre
bration, then p is a Serre bration.
Proof. Using a homeomorphism Dn ∼

= I n, let a lifting problem
f
I n × {0} E
p
H
I n × [0, 1] B
be given. Now {H −1 (Uα )}α∈I is an open cover of I n × [0, 1], so by the Lesbegue number
lemma we may choose a grid ne enough that each cube lies in some H −1 (Uα ). Order
the cubes as shown in the gure below, so that each cube except the rightmost ones has
a face to its right labelled higher.
i÷÷
"
:
Now we can lift the map H on the 1st cube, extending the lift given on its red face,
using that this cube has image in some Uα , and that p is a Serre bration over Uα .
Similarly, supposing that the rst (i − 1) cubes have been compatibly lifted, we may nd
a lift on the ith cube, extending the lift already given over part of its boundary, as no
lift has yet been given on its rightmost face. Continuing in this way gives the desired
lift.
Corollary 1.15.9. A bre bundle is a Serre bration.
Proof. If p:E→B is a bre bundle then by denition there is an open cover {Uα }α∈I of
B and homeomorphisms p
−1 (U )
α
∼
= Uα ×F over Uα , identifying p|p−1 (Uα ) : −1
p (Uα ) → Uα
with proj1 : Uα × F → Uα . By Example 1.15.2 such maps are Serre brations, so by the
previous theorem p is too.
Theorem 1.15.10. (B, b0 ) be a path-connected

Let based space, p:E →B be a Serre
bration, and x0 ∈ F := p−1 (b0 ). Then the map
p∗ : πn (E, F, x0 ) −→ πn (B, b0 )
1.15 Fibrations 27
is an isomorphism for all n ≥ 1, so there is a long exact sequence
··· πn+1 (B, b0 )

∂
i∗ p∗
πn (F, x0 ) πn (E, x0 ) πn (B, b0 )
∂
i∗ j∗
πn−1 (F, x0 ) πn−1 (E, x0 ) ···
Proof. Let f : (I n , ∂I n ) → (B, b0 ) be a map. Construct the relative lifting problem
constx0
un−1 I n−1 × {0} ∪ ∂I n−1 × [0, 1] E
p
f
I n−1 × [0, 1] B,
which has a solution fˆ : I n−1 × [0, 1] → E . By construction this sends ∂I n to F and

un−1 to x0 , so we have [fˆ] ∈ πn (E, F, x0 ), and by construction this satises p∗ ([fˆ]) =
[p ◦ fˆ] = [f ]. Thus p∗ is surjective.
Suppose that [g] ∈ πn (E, F, x0 ) satises p∗ ([g]) = 0, so that p ◦ g is homotopic to
constx0 : let H : I n × [0, 1] → B be such a homotopy, which sends ∂I n × [0, 1] to b0 .
Now constx0 : u
n−1 × [0, 1] → E is a lift of H|
un−1 ×[0,1] , so we may construct the relative
lifting problem
g∪constx0
I n × {0} ∪ un−1 × [0, 1] E
p
H
I n × [0, 1] B.
This has a solution H̃ : I n × [0, 1] → E , which is a homotopy relative to un−1 from g to

a map into F = p−1 (b0 ), which by the Compression Criterion (Section 1.5) shows that
[g] = 0 ∈ πn (E, F, x0 ).
Example 1.15.11. Let S 2n+1 ⊂ Cn+1 be the unit sphere. The group S 1 = U (1) of unit
complex numbers acts freely on S
2n+1 with quotient the complex projective space CPn .
The map
p : S 2n+1 −→ CPn
is a bre bundle, so a Serre bration, with bre over each point homeomorphic to S1.
Thus there is a long exact sequence
i p∗ ∂
· · · −→ πi (S 1 , x0 ) −→
∗
πi (S 2n+1 , x0 ) −→ πi (CPn , ∗) −→ πi−1 (S 1 , x0 ) −→ · · ·
Combining this with (

Z i=1
πi (S 1 , x0 ) =
0 else
and πi (S 2n+1 , x0 ) = 0 for i ≤ 2n, we nd that

0 i = 1

n
πi (CP , ∗) = Z i = 2

0 2 < i ≤ 2n.

In particular, for CP∞ = ∪n≥0 CPn we nd
(
0 i 6= 2
πi (CP∞ , ∗) =
Z i = 2,
so that CP∞ is an EilenbergMac Lane space of type (Z, 2), i.e. CP∞ ' K(Z, 2).
The map H := p : S3 → S2 = CP1 , with bre S 1 , is known as the Hopf map. The
sequence
i H ∂
0 = π3 (S 1 , x0 ) −→
∗
π3 (S 3 , x0 ) −→
∗
π3 (S 2 , ∗) −→ π2 (S 1 , x0 ) = 0
shows that π3 (S 2 , ∗) ∼
=Z generated by [H]. 4
Lecture 9
1.16 Comparing bres

Theorem 1.16.1. Let p : E → B be a Serre bration, b0 , , b1 ∈ B be points with bres
−1
Fbi := p (bi ), and u : [0, 1] → B be a path from b0 to b1 . Then there is a space Fu and
weak homotopy equivalences
iu
b0 iu
b1
Fb0 −→ Fu ←− Fb1 .
Furthermore, if v is a path from b1 to b2 then there is a commutative diagram of weak
homotopy equivalences
iu·v
b Fu·v iu·v
b
0 2
Fu Fv
iu
b iu
b ivb ivb
0 1 1 2
Fb0 Fb1 Fb2 .
Proof. Dene Fu := u∗ E . There are maps of Serre brations
Fb0 Fu Fb1
u∗ p
{0} [0, 1] {1}.

1.16 Comparing bres 29
The long exact sequence on homotopy groups for the bration u∗ p based at a point
x0 ∈ Fu with (u∗ p)(x0 ) =0 is
· · · −→ πi+1 ([0, 1], {0}) −→ πi (Fb0 , x0 ) −→ πi (Fu , x0 ) −→ πi ([0, 1], {0}) −→ · · ·
and as [0, 1] is contractible the middle map is an isomorphism: thus Fb0 → Fu is a weak
homotopy equivalence. The map Fb1 → Fu is also a weak homotopy equivalence by the
same argument.
For the second part, note that the composition
∼
= u·v
[0, 1] −→ [0, 1/2] ⊂ [0, 1] −→ B
is the path u, so pulling p back along these maps gives a map of Serre brations
Fu Fu·v
[0, 1] rescale [0, 1/2] inc

[0, 1];
the map of total spaces is the required map Fu → Fu·v . These brations have the same
bres, and their bases are both contractible, so it follows from the map of long exact
sequences for these brations that the map Fu → Fu·v is a weak equivalence. The map
Fv → Fu·v is constructed analogously, and it is then easy to verify that the diagram so
obtained commutes.
By the rst part, if B is path-connected then the bres over dierent points are all
weakly homotopy equivalent to each other. Because of this, we will often say that
i p
F −→ E −→ B is a bration sequence,
meaning that p is a bration, B is path-connected, and that F is the bre of p over some
point: the bre over any other point is then weakly homotopy equivalent to F .
Corollary 1.16.2. Let p:E→B be a Serre bration, b0 , b1 ∈ B be points with bres

Fbi := p−1 (bi ), and u : [0, 1] → B be a path from b0 to b1 . Then there is an induced
isomorphism
∼
u# : H∗ (Fb0 ) −→ H∗ (Fb1 ).
Ifv is a path from b1 to b2 , then v# ◦ u# = (u · v)# . In particular π1 (B, b0 ) acts on
H∗ (Fb0 ). Similarly, it acts on the cohomology of Fb0 .
Proof. As weak homotopy equivalences induce isomorphisms on homology, we dene u#

to be the composition of the isomorphisms
∼ ∼
H∗ (Fb0 ) −→ H∗ (Fu ) ←− H∗ (Fb1 ).
The formula v# ◦ u# = (u · v)# then follows from the commutative diagram in the last
theorem.
1.17 Function spaces

Denition 1.17.1. For spaces X and Y, we let map(X, Y ) denote the set of continuous
maps f :X →Y. We endow this set with the compact-open topology: the topology
generated by the subbasis of sets
W (K, U ) := {f : X → Y | f (K) ⊂ U },
ranging over all compact K⊂X and all open U ⊂Y.
Lemma 1.17.2. If X is locally compact

10 then the evaluation map
eX,Y : map(X, Y ) × X −→ Y
(f, x) 7−→ f (x)
is continuous.
Proof. Let U 3 f (x) be an open neighbourhood. As X is locally compact there is a

compact neighbourhood x ∈ K ⊂ f −1 (U ). Then eX,Y sends W (K, U ) × K into U, so
the open set W (K, U ) × int(K) 3 (f, x) lies in U.
Lemma 1.17.3. Let f :Z ×X →Y be continuous. Then its adjoint
f ad : Z −→ map(X, Y )
z 7−→ (x 7→ f (z, x))
is continuous.
Proof. We must show that (f ad )−1 (W (K, U )) is open, for K ⊂ X compact and U ⊂ Y
ad
open. If f (z) ∈ W (K, U ) then {z} × K ⊂ f
−1 (U ). As K is compact there is an open
V 3 z such that V × K ⊂ f (U ), but then z ∈ V ⊂ (f ad )−1 (W (K, U )).

−1
Corollary 1.17.4. If X is locally compact and f ad : Z → map(X, Y ) is continuous,

then f :Z ×X →Y is continuous.
Proof. The map f is
f ad ×IdX eX,Y
Z ×X −→ map(X, Y ) × X −→ Y.
Corollary 1.17.5. If X and Z are locally compact, then the map
α : map(Z × X, Y ) −→ map(Z, map(X, Y ))

f 7−→ f ad
is a homeomorphism.
10
i.e. for every point x∈X and every open neighbourhood U 3 x, there is a compact set K with
x ∈ int(K) ⊂ K ⊂ U .
1.17 Function spaces 31
Proof. By the last two results this function is well-dened and a bijection. Also, by the
last results,
α is continuous ⇐⇒ map(Z × X, Y ) × Z −→ map(X, Y ) is continuous
⇐⇒ map(Z × X, Y ) × Z × X −→ Y is continuous
Z × X is locally compact. Similarly α−1 is continuous

which it is as if and only if
map(Z, map(X, Y )) × Z × X → Y is continuous, but this factors as
eZ,map(X,Y ) ×IdX eX,Y
map(Z, map(X, Y )) × Z × X −→ map(X, Y ) × X −→ Y
so is indeed continuous.
Theorem 1.17.6. If (X, A) is a relative CW complex and X is locally compact, then

the restriction map
res : map(X, Y ) −→ map(A, Y )

f 7−→ f |A
is a Serre bration.
Proof. Suppose given a homotopy lifting problem
f ad
Dn × {0} map(X, Y )
res
H ad
Dn × [0, 1] map(A, Y ),
where f ad and H ad are adjoint to maps f : X ×Dn ×{0} → Y and H : A×Dn ×[0, 1] → Y .
n n
As (X × D , A × D ) is again a relative CW complex, it has the homotopy extension
n
property, so H extends to a H̄ : X × D × [0, 1] → Y starting at f . By adjunction this
gives a map
H̄ ad : Dn × [0, 1] −→ map(X, Y ),
lifting H ad and extending f ad .
Example 1.17.7. Consider the relative CW complex ([0, 1], {0}). The theorem shows
that the evaluation map
ev0 : P X := map([0, 1], X) −→ X

γ 7−→ γ(0)
is a Serre bration, called the path bration. The bre ev0−1 (x0 ) =: Px0 X is the space
of paths in X starting at x0 , and this is contractible: the homotopy
Px0 X × [0, 1] −→ Px0 X

( !
γ(s) 0 ≤ s ≤ t
(γ, t) 7−→ s 7→
γ(t) t ≤ s ≤ 1
is easily checked to be a deformation retraction. 4
Lemma 1.17.8. In fact ev0 : P X → X is a Hurewicz bration.
Proof. Supposing given a homotopy lifting problem
f
Z × {0} PX
ev0
H
Z × [0, 1] X,
with Z an arbitrary space. Dene a map H̃ ad : Z × [0, 1] × [0, 1] → X by the formula
(
ad H(z, t − 2s) 0 ≤ s ≤ t/2
H̃ (z, t, s) := s−t/2 .
f (z, 0)( 1−t/2 ) t/2 ≤ s ≤ 1
This is continuous by the gluing lemma, so its adjoint H̃ : Z × [0, 1] → P X is continuous

by Lemma 1.17.3. We easily check that H̃ lifts H and extends f.
The same argument, with more notation, shows that
ev0 × ev1 : P X −→ X × X
γ 7−→ (γ(0), γ(1))
is a Hurewicz bration.
Let f :Y →X be a map, and form the pullback
Ef (f × IdX )∗ P X PX
ev0 ×ev1
f ×IdX
Y ×X X × X.
Unravelling the denition, we have
Ef ∼
= {(y, γ) ∈ Y × P X | γ(0) = f (y)},
and the map (y, γ) 7→ (y, γ(1)) : Ef → Y × X , being the pullback of ev0 × ev1 , is a
Hurewicz bration. As the projection Y × X → X is also a Hurewicz bration, the map
pf : Ef −→ X
(y, γ) 7−→ γ(1)
Lecture 10 is a Hurewicz bration.

On the other hand, the maps y 7→ (y, constf (y) ) : Y → Ef and (y, γ) 7→ y : Ef → Y
are easily seen to be homotopy inverses. In total, the diagram
Ef
y7→(y,constf (y) )
pf
f
Y X
1.17 Function spaces 33
shows that the map f : Y → X may be replaced up to homotopy equivalence with a

Hurewicz bration pf : Ef → X . We call the bre
p−1
f (x0 ) = {(y, γ) ∈ Y × P X | γ(0) = f (y), γ(1) = x0 }
the homotopy bre of f at the point x0 ∈ X . The long exact sequence on homotopy
groups for the bration pf , along with the homotopy equivalence Y ' Ef , gives a long
exact sequence
··· πn+1 (X, x0 )

∂
f∗
πn (p−1
f (x0 ), (y0 , constx0 )) πn (Y, y0 ) πn (X, x0 )
∂
f∗
πn−1 (p−1
f (x0 ), (y0 , constx0 )) πn−1 (Y, y0 ) ···
Applied to the inclusion map f : {x0 } → X , we get
Ef = {γ ∈ P X | γ(0) = x0 } = Px0 X
and
p−1
f (x0 ) = {γ ∈ P X | γ(0) = x0 , γ(1) = x0 } =: Ωx0 X
the loop space of X based at x0 . The long exact sequence on homotopy groups for the
bration pf shows that
∼
∂ : πn (X, x0 ) −→ πn−1 (Ωx0 X, constx0 )
is an isomorphism for n ≥ 1. In this sense ΩX has the analogous eect on homotopy

groups as the suspension ΣX does on homology groups.
As the space Ωx0 X is again based, at constx0 , we can iterate this construction to
form Ωkx0 X . Unravelling denitions shows that
Ωkx0 X ∼
= {f ∈ map(I k , X) | f (∂I k ) = x0 }.
If x1 lies in the same path-component as x0 then Theorem 1.16.1 applied to the bres
of the bration
ev0 × ev1 : P X −→ X × X
over (b0 , b0 ) and (b1 , b1 ) shows that Ωx1 X and Ωx0 X are weakly equivalent
11 , so if X
is path-connected then we often just write ΩX to mean the loop space taken at some
basepoint.
11
In fact we easily see that they are homotopy equivalent: conjugating by a path u from x0 to x1
gives a map Ωx0 X → Ωx1 X , conjugating by the reverse path gives a map Ωx1 X → Ωx0 X , and the two
compositions are easily seen to be homotopic to the identity.
1.18 The MoorePostnikov tower

Lemma 1.18.1. Let f : (A, a0 ) → (X, x0 ) be a map of path-connected and based spaces,
and n ≥ 0. There is a relative CW complex (Zn , A) and an extension gn : Zn → X of f
such that
(i) (Zn , A) only has relative cells of dimension > n, so in particular is n-connected,
(ii) the map (gn )∗ : πn (Zn , a0 ) → πn (X, x0 ) is injective, and for all i > n the maps
(gn )∗ : πi (Zn , a0 ) → πi (X, x0 ) are isomorphisms.12
To emphasise: the factorisation
gn
f : A −→ Zn −→ X
has the rst map an isomorphism on homotopy groups in degrees < n, the second map
an isomorphism in degrees > n, and in degree n
f∗ : πn (A, a0 ) −→ πn (Zn , a0 ) −→ πn (X, x0 )
is the (unique up to isomorphism) factorisation of f∗ as an epimorphism followed by a

monomorphism.
Proof. First construct (T n , A) by attaching (n + 1)-cells to A to kill
ker(f∗ : πn (A, a0 ) → πn (X, x0 )),
and using the nullhomotopies in X to extend the map f to a gnn : T n → X . Now construct
relative CW complexes T
n ⊂ T n+1 ⊂ T n+2 ⊂ ··· i
with gn : T i → X such that
(i) (T i , A) only has relative cells of dimensions n+1≤∗≤i+1
(ii) the map πj (gni ) is injective for n≤j <i+1 and surjective for n < j ≤ i + 1.
This can be done exactly as in the proof of the CW approximation theorem: if T i−1 has
been constructed, rst attach (i + 1)-cells to kill ker(f∗ : πi (T i−1 ) → πi (X)), and then
wedge on (i + 1)-spheres to generate πi+1 (X).
i
S
Finally, let Zn := i≥n T .
If gn : Zn → X and gn+1 : Zn+1 → X are maps given by the lemma, consider the
commutative square
A Zn
gn
gn+1
Zn+1 X.
The relative CW complex (Zn+1 , A) only has cells of dimension ≥ n + 2. On the other
hand we have a long exact sequence
· · · πi (Zn ) −→ πi (X) −→ πi (X, Zn ) −→ πi−1 (Zn ) −→ πi−1 (X) · · · ,

12
This property is often phrased as the map gn : Zn → X is n-co-connected.
1.18 The MoorePostnikov tower 35
the left-hand map is surjective for i > n, so for i ≥ n + 1, and the right-hand map is
injective for i − 1 ≥ n, so for i ≥ n + 1. Thus πi (X, Zn ) = 0 for i ≥ n + 1. It then follows
from the compression lemma that there is a map
pn+1 : Zn+1 −→ Zn
which is the identity on A, such that gn ◦ pn+1 is homotopic to gn+1 relative to A. In

total we obtain a diagram
.
.
.
Z2
p2
g2
Z1
g1
p1
g0
f :A Z0 X
where the left-hand triangles commute, and the right-hand triangles commute up to
homotopy. By redening gn0 = g0 ◦ p1 ◦ p2 ◦ · · · ◦ pn we may assume that the right-hand
triangles commute too. This is the MoorePostnikov tower of the map f : A → X .
Example 1.18.2. For a space X consider the map f : X → {∗}. Its MoorePostnikov
tower gives a diagram
.
.
.
X2
p2
f2
X1
f1
p1
f0
X X0
where πi (Xn ) = 0 for i≥n and
(fn )∗ : πi (X, x0 ) −→ πi (Xn , fn (x0 )) is an isomorphism for i < n.
This is the Postnikov tower of X .

It is perhaps suggestive to use interval" notation and write Xn = X[0, n − 1]: this
space has the same homotopy groups as X in the range of degrees 0 ≤ ∗ ≤ n − 1 and
trivial homotopy groups outside this range. 4
Example 1.18.3. Choose a basepoint x0 ∈ X , and consider f : {x0 } → X . Its Moore

Postnikov tower gives a diagram
.
.
.
Z2
p2
g2
Z1
g1
p1
g0
{x0 } Z0 X
where Zn n-connected, and (gn )∗ : πi (Zn , x0 ) → πi (X, x0 ) is an isomorphism for i > n.

is
This is the Whitehead tower of X (at the basepoint x0 ).
It is again suggestive to write Zn = X[n + 1, ∞]: this space has the same homotopy
groups as X in the range of degrees n + 1 ≤ ∗ < ∞ and trivial homotopy groups outside
this range. 4
In the Whitehead tower consider the long exact sequence
(pn+1 )∗ ∂ (pn+1 )∗
πi (Zn+1 ) πi (Zn ) πi (Zn , Zn+1 ) πi−1 (Zn+1 ) πi−1 (Zn )
(gn+1 )∗ (gn )∗ (gn+1 )∗ (gn )∗
πi (X) πi (X) πi−1 (X) πi−1 (X).
If i ≤ n πi (Zn ) = 0 = πi−1 (Zn+1 ), so πi (Zn , Zn+1 ) = 0. If i ≥ n + 2 then the

then
vertical maps are all isomorphisms, so πi (Zn , Zn+1 ) = 0 too. In degree n + 1 we have
πn+1 (Zn+1 ) = 0, πn+1 (Zn ) = πn+1 (X) and πn (Zn+1 ) = 0, and so
(
πn+1 (X, x0 ) i = n + 1
πi (Zn , Zn+1 , x0 ) ∼
=
0 else.
In other words, the homotopy bre of pn : Zn+1 → Zn is a K(πn+1 (X, x0 ), n).

Similarly, for the Postnikov tower of X we nd
(
πn (X, x0 ) i = n + 1
πi (Xn , Xn+1 , x0 ) ∼
=
0 else,
so the homotopy bre of pn : Xn+1 → Xn is a K(πn (X, x0 ), n).

Both of these become easier to remember using the interval" notation: the rst says
that the homotopy bre of X[n, ∞] → X[n − 1, ∞] is a K(πn−1 (X, x0 ), n − 2), and the
second says that the homotopy bre of X[0, n] → X[0, n − 1] is a K(πn (X, x0 ), n).
1.19 A motivating strategy 37
1.19 A motivating strategy

At this point we have developed various techniques for manipulating homotopy groups
of spaces, but we do not yet know how to compute many of them. We know Hopf 's
theorem, that πn ∼
(S n )
= Z generated by the identity map, and in Example 1.15.11 we
2 ∼ Z generated by the Hopf map H : S 3 → S 2 .
calculated that π3 (S ) =
If X is a simply-connected space and Hi (X; Z) = 0 for 0 < i < n then the Hurewicz
theorem shows that
∼
h : πn (X, x0 ) −→ Hn (X; Z).
As we are quite good at calculating homology, this means that we can often calculate
the lowest non-trivial homotopy group.
If X is a path-connected space and
· · · −→ X[n, ∞] −→ X[n − 1, ∞] −→ · · · −→ X[3, ∞] −→ X[2, ∞] −→ X[1, ∞] ' X
is the Whitehead tower of X , then as X[n, ∞] is (n − 1)-connected we have isomorphisms

∼ ∼
πn (X) ←− πn (X[n, ∞]) −→ Hn (X[n, ∞]; Z),
so we can calculate πn (X) as the lowest non-trivial homology group of the space X[n, ∞].
The problem is that X[n, ∞] is a space that we have shown exists by highly inexplicit
means, and we have no idea how to express its homology in terms of e.g. the homology
of X.
However, supposing that we know the homology of X we could try to inductively cal-
culate the homology of the spaces X[n, ∞] (and hence by the above the homotopy groups
of X ) as follows. As the homotopy bre of X[n, ∞] → X[n−1, ∞] is a K(πn−1 (X), n−2),
and we may suppose that we know πn−1 (X) by induction, the strategy would have some
hope if
(i) given a bration p : E → B with bre F, we had a mechanism to calculate the

homology of E given the homology of B and of F, and
(ii) given an abelian group G we knew how to calculate the homology of K(G, n).
Developing tools to do this will be the rest of the course: the rst is the Serre spectral
sequence, and the second is the subject of cohomology operations. These tools will
have broad applications, well beyond calculating homotopy groups.
Chapter 2
The Serre spectral sequence and applications

Lecture 11
2.1 Spectral sequences

Denition 2.1.1. ALbigraded abelian group A•,• is an abelian group A with a
A =
decomposition p,q∈Z Ap,q . A degree (a, b) map f : A•,• → B•,• of bigraded
abelian groups is a homomorphism f :A→B such that f (Ap,q ) ⊂ Bp+a,q+b .
Denition 2.1.2. spectral sequence is a sequence E•,•

A (homological)
1 , E2 , E3 , . . .
•,• •,•
of bigraded abelian groups, called pages, equipped with maps
dr : E•,•
r r
−→ E•,• of degree (−r, r − 1)
such that dr ◦ dr = 0, so that dr is a dierential, and

r+1 = H(E r , dr ).
E•,• •,• That is,
r+1
r → Er
ker(dr : Ep,q p−r,q+r−1 )
Ep,q = r .
im(dr : Ep+r,q−r+1 → Ep,q
r )
Remark 2.1.3. All the gradings I have given are optional and variable.
Denition 2.1.4. An exact couple of type r consists of bigraded abelian groups E•,•
and A•,• and maps
i : A•,• −→ A•,• of degree (1, −1)

j : A•,• −→ E•,• of degree (−r, r)
k : E•,• −→ A•,• of degree (−1, 0)
so that the triangle

i
A•,• A•,•
k j
E•,•
is exact at each vertex (i.e. im(i) = ker(j), and so on).

In this case d := j ◦ k , which has degree (−r − 1, r), is a dierential on E•,• , as
d ◦ d = j ◦ (k ◦ j) ◦ k and k ◦ j = 0.
The derived couple of such an exact couple is given by
ker(d)
A0•,• := im(i) 0
E•,• :=
im(d)
i0 := i|A0•,• j 0 (i(a)) := [j(a)] k 0 ([e]) := k(e).
38
2.1 Spectral sequences 39
Theorem 2.1.5. The derived couple of an exact couple of type r is well-dened and is
an exact couple of type (r + 1).
Proof. First show that j 0 : A0•,• → E•,•

0 is well-dened. If i(a) = i(b), then i(a − b) = 0
so a − b = k(c). Then
j(a) = j(b) + j(k(c)) = j(b) + d(c)
so
0 = ker(d) .
[j(a)] = [j(b)] ∈ E•,• im(d)
0 0 0
Now show that k : E•,• → A•,• is well-dened. If
0 ,
[e] = [f ] ∈ E•,• then d(e) =
d(f ) = 0, and e = f + d(g) = f + j(k(g)). Then
k(e) = k(f ) + k(j(k(g))) = k(f )
as k ◦ j = 0.
That the resulting triangle is exact at each vertex I leave as an exercise.
i(a) ∈ Ap,q then a ∈ Ap−1,q+1 , so j(a) ∈ Ep−1−r,q+1+r , so j 0 has degree
Finally, if
(−r − 1, r + 1). The map i0 has degree (1, −1), and k 0 has degree (−1, 0). Thus the
derived couple is indeed an exact couple of type (r + 1).
In particular, if
r be
(A•,• , E•,• , i, j, k) is an exact couple of type 0, then letting E•,•
the (r − 1)st r
derived E•,• , and d be the (r − 1)st derived d, gives a spectral sequence.
Example 2.1.6 (The spectral sequence of a ltered space) . Let
∅ ⊂ X0 ⊂ X1 ⊂ X2 ⊂ · · · ⊂ X
be a sequence of subspaces. Using the long exact sequences for the pairs (Xn , Xn−1 ) for
homology with arbitrary coecients (which we omit from the notation), let
Ap,q := Hp+q (Xp )

Ep,q := Hp+q (Xp , Xp−1 )
i : Hp+q (Xp ) −→ H(p+1)+(q−1) (Xp+1 ) of degree (1, −1)
j : Hp+q (Xp ) −→ Hp+q (Xp , Xp−1 ) of degree (0, 0)
k : Hp+q (Xp , Xp−1 ) −→ H(p−1)+q (Xp−1 ) of degree (−1, 0).
This is an exact couple of type 0 (it is exact by the exactness of the long exact sequence
of a pair), so gives a spectral sequence with
1
Ep,q = Hp+q (Xp , Xp−1 ).
The form of this spectral sequence is shown in Figure 2.1.

The dierential d1 : Ep,q
1 → E1
p−1,q is given by the composition
∂
Hp+q (Xp , Xp−1 ) −→ H(p−1)+q (Xp−1 ) −→ H(p−1)+q (Xp−1 , Xp−2 ). 4
40 Chapter 2 The Serre spectral sequence and applications
F÷if÷ f:÷÷
"
Figure 2.1 The homological spectral sequence for a ltered space.

not e
18
if
Z
÷÷
Example 2.1.7.
.
Z
X beD a CW complex, ∅ ⊂ X 0 ⊂ X 1 ⊂ · · ·
3h -
T
Let be its skeletal ltration.
Then we have (
1 p p−1 Cpcell (X) ifq=0
Ep,q = Hp+q (X , X )=
0 else,
.
and "
.
∂
Cpcell (X) = Hp (X p , X p−1 ) −→ H(p−1) (X p−1 ) −→ Hp−1 (X p−1 , X p−2 ) = Cp−1
cell
(X)
is by denition the cellular boundary map. Thus
n - I 72 ( 7L
Hpcell (X) q=0
#
if
h 2
¥¥p
Ep,q =
0 else.
0 I
The dierential dr for r ≥ 2 changes the q -degree, so must necessarily be zero: thus the
7#
←
.DZ
-
above describes
r for all r ≥ 2.
Ep,q 4
.
p
. o
.
Theorem 2.1.8 (Convergence) . Let ∅ ⊂ X0 ⊂ X1 ⊂ X2 ⊂ · · · ⊂ X be a sequence of

n
X i lies in some Xn , and let {(E•,•
-
o
- -
subspaces such that any simplex in

r , dr )}
r≥1 be the
associated spectral sequence.
r r
If r ≥ p + 1 then d : Ep,q
r
→ Ep−r,q+r−1 is zero, so
r+1
Ep,q is a quotient of
r .
Ep,q Write
1 2
l÷E "f÷÷÷
∞ p+1 p+2 p+3
Ep,q := lim(Ep,q Ep,q Ep,q · · · ).
−→
o o o O
Then there is a ltration Z
0 ≤ F 0 Hd (X) ≤ F 1 Hd (X) ≤ F 2 Hd (X) ≤ · · ·
. . . . .
of Hd (X) such that
.
n
S
(i) n≥0 F Hd (X) = Hd (X), and
l 2 3 4 5
2.1 Spectral sequences 41
F p Hp+q (X) ∼ ∞.
(ii)
F p−1 Hp+q (X) = Ep,q
We say the spectral sequence converges to H∗ (X), 1

and write Ep,q ⇒ Hp+q (X).
Lecture 12
r 1
Proof. Firstly, Ep−r,q+r−1 is a subquotient of Ep−r,q+r−1 = Hp+q−1 (Xp−r , Xp−r−1 ), and
r r r
this vanishes if p − r < 0. Thus d : Ep,q → Ep−r,q+r−1 is indeed zero if r ≥ p + 1.
Now, dene
F p Hp+q (X) := im(Hp+q (Xp ) → Hp+q (X)).
nH
S
By our assumption that any simplex in X lies in some Xn , we have n≥0 F p+q (X) =
Hp+q (X) as required.
To prove property (ii) we rst establish some intermediate results.
Claim: If x ∈ Ap,q then
1
j(x) ∈ Ep,q = Ep,q is a cycle with respect to every dr .
Proof of claim. We have d(j(x)) = j(k(j(x))) = 0 as k ◦ j = 0. More generally, if j(x)

survives until
r then
Ep,q
dr ([j(x)]) = j r ◦ k r ([j(x)]) = j r ([k(j(x))])
which also vanishes as k ◦ j = 0.
This denes a homomorphism

r
j : Ap,q → Ep,q for all r ≥ 1, so a homomorphism
∞
j∞ : Ap,q −→ Ep,q .
Claim: The homomorphism j∞ is onto.
∞ is represented by an x ∈ E
Ep,q 1
Proof of claim. An element of p,q = Ep,q such that
d ([x]) = 0 for all r ≥ 1. Thus j r (k r ([x])) = 0, so k r ([x]) = i◦r (ar ) for some ar ∈ A•,•
r
◦r−1 : A
(as it is the image under i of an element which lies in the image of i •,• → A•,• ).
r
But k ([x]) = k(x), so we nd that
k(x) ∈ im(i◦r : Ap−r−1,q+r → Ap−1,q )
r. But Ap−r−1,q+r = 0
for all for p − r − 1 < 0, so for all r 0, and hence k(x) = 0, so
x = j(y), so [x] = j∞ (y).
Claim: ker(j∞ ) = i(Ap−1,q+1 ) + ◦s ).

S
s ker(i
Proof of claim. If j∞ (x) = 0 then

r
[j(x)] = 0 ∈ Ep,q for some r. Thus
[j(x)] = j r−1 ◦ k r−1 (yr ) ∈ Ep,q

r−1
for some yr .
k r−1 (yr ) ∈ Ar−1 r−1 (y ) = i◦r−2 (a ar−1 ∈ A•,• ,

Now •,• , so k r r−1 ) for some and thus
i◦i−1 (ar−1 ) = 0, so ar−1 ∈ ker(i◦r−1 ). Now
[j(x)] = j r−1 (i◦r−2 (ar−1 )) = [j(ar−1 )] ∈ Ep,q

r−1
,
so
r−1
[j(x − ar−1 )] = 0 ∈ Ep,q .
Continuing in this way, get
1
j(x − ar−1 − ar−2 − · · · − a1 ) = 0 ∈ Ep,q
with as ∈ ker(i◦s ), so
x = i(y) + ar−1 + ar−2 + · · · + a1
as required.
F p Hp+q (X) ∼ ∞.
We now prove that
F p−1 Hp+q (X) = Ep,q By the above we have
Ap,q / s ker(i◦s )
S
∞ ∼ Ap,q ∼
Ep,q = = ,
i(Ap−1,q+1 ) + s ker(i◦s ) i(Ap−1,q+1 / s ker(i◦s ))
S S
÷F÷if÷
but
"
f:÷÷
"
[
ker(i◦s ) = ker(Hp+q (Xp ) → Hp+q (X)),
s
because a homology class on Xp vanishes in X if and only it vanishes in some Xp+s ,
again using the assumption that every simplex in X lies in some Xn . The fact that
Hp+q (Xp )
F p Hp+q (X) = im(Hp+q (Xp ) → Hp+q (X)) ∼
=
ker(Hp+q (Xp ) → Hp+q (X))
nishes the proof.
In practice the spectral sequences we will look at have even better vanishing proper-
ties: they satisfy
1
Ep,q =0 for p<0 or q < 0.
not
18
i:÷÷
if
" - .
3h D Z
-
T
. "
In this case there are only nitely-many non-zero groups

∞ along each diagonal
Ep,q
p+q = d, so the ltration F • Hd (X) has nite length, and ∞ n I
{Ep,q }p+q=d gives a composition-
72
#
Hd (X). "
oh
¥¥p
series for
17 .
Example 2.1.9. Returning to Example 2.1.7, on the skeletal ltration of a CW complex

X, we have / I ←(
0 I
∞
Ep,q ∼
=
Hpcell (X)
-
.
. .DZ q = 0
if o 7#
0 else,
i
-
o
- -
D 1 2
19 K o o o O
Z
.
2.2 The Serre spectral sequence 43
so by the convergence theorem Hp (X) has a ltration with a single nontrivial ltration
quotient, Hpcell (X), and hence
Hp (X) ∼
= Hpcell (X).
This gives a new proof that cellular homology calculates singular homology. 4
2.2 The Serre spectral sequence

Theorem 2.2.1. p : E → B be a Hurewicz bration over a CW complex B with
Let
−1
a single 0-cell b0 ∈ B and bre F := p (b0 ), such that π1 (B, b0 ) acts trivially (for the
action of Corollary 1.16.2) on H∗ (F ; G).
r r
Then there is a spectral sequence {(E•,• , d )} with
2 ∼
Ep,q = Hp (B; Hq (F ; G))
F p Hp+q (E;G) ∼ ∞.
and a ltration of H∗ (E; G) such that
F p−1 Hp+q (E;G) = Ep,q
Corollary 2.2.2. Let p : E → B be a Serre bration over a path-connected space B ,

such that π1 (B, b0 ) acts trivially on the homology of F := p−1 (b0 ) with G-coecients.
Then there is a spectral sequence with precisely the same properties.
Proof. Let α : C → B be a CW approximation, where C has a single 0-cell c0 and

a(c0 ) = b0 . Construct the following commutative diagram
'weak
p−1
α∗ p (c0 ) F F
' 'weak
Eα∗ p α∗ E E
pα∗ p α∗ p p
α
C C B
by rst forming the pullbackα∗ p : α∗ E → C , which is again a Serre bration, and whose
∗
bre over c0 is again F , and then replacing the map α p by a Hurewicz bration as
∗
in Section 1.17. The replacement map α E → Eα∗ p is a homotopy equivalence, so the
−1
induced map F → pα∗ p (c0 ) on bres is a weak homotopy equivalence (by the 5-lemma
applied to the map of long exact sequences of homotopy groups for the Serre brations
α∗ p and pα∗ p ). Similarly, as α is a weak equivalence the map α∗ E → E is too (by the
5-lemma applied to the map of long exact sequence for the
∗
brations p and α p). In
particular all the horizontal maps induce isomorphisms on homology, and the left-hand
column is a Hurewicz bre sequence to the theorem above applies to it: this gives the
analogous spectral sequence for the right-hand column.
Example 2.2.3. Consider the Hopf bration H : S3 → S2, whose bres are all homeo-
1
morphic to S . This is a bre bundle, and so a Serre bration. The group π1 (S 2 , ∗) is
not e
18
i:÷÷
if
" - .
3h D Z -
1
T
( .
trivial so necessarily acts trivially on H∗ (S ; Z). We have
"
2 2 1 Z (p, q) = (0, 0), (2, 0), (0, 1), (2, 1)
Ep,q = Hp (S ; Hq (S ; Z)) =
0 otherwise
as shown in Figure 2.2.
n - I 72
#
"
oh
¥¥p
17 .
/ I
0 I
7#
←
.DZ
-
.
. o
.
i
-
o
- -
D 1 2
l÷E "f÷÷
Figure 2.2 The homological Serre spectral sequence for the Hopf bration.
19 K o o o O
Z
.
This spectral sequence converges to
• . . . .
Z ∗.
(
= 0, 3
H∗ (S 3 ; Z) =
0 otherwise,
so the copies of Z at
2
E2,0 and
2
E0,1 cannot survive to
∞.
E•,• Thus the dierential
2 ∼ 2 ∼
d2 : E2,0 = Z −→ E0,1 =Z
must be an isomorphism, as this is the only way these groups can die. 4
Lecture 13
Example 2.2.4. l
n > 1Dand consider the path2
Let
3 4 n
bration γ 7→ γ(1) : P∗ S → 5S n , with
n n n n
bre over ∗ ∈ S given by Ω∗ S = ΩS . As n > 1 the space S is simply-connected, so
n
its fundamental group acts trivially on H∗ (Ω∗ S ; Z). Thus we have a spectral sequence
2
Ep,q = Hp (S n ; Hq (ΩS n ; Z)) ⇒ Hp+q (P∗ S n ; Z).
P∗ S n ∞ =0
Ep,q p+q >0 ∞ ∼ Z).
E0,0
Now is contractible, so we must have for (and =
We will reverse-engineer this spectral sequence,
n
using as input just H∗ (S ; Z), that
the spectral sequence converges to zero in degree p + q > 0, and that ΩS n is path
connected, which follows from the long exact sequence on homotopy groups for the path
bration shows that ΩS n is path-connected. It will be useful to refer to the chart at the
end of the example throughout.
As ΩS n is path-connected we have H0 (ΩS n ; Z) ∼
= Z, and we can determine the bottom
row
2
of E•,• to be
(
2 Z ∗ = 0, n
Ep,0 = Hp (S n ; Z) =
0 otherwise.
Hi (ΩS n ; Z) 6= 0 for 0 < i < n − 1,

Suppose that and let i be minimal with this
2 n n
property. Then E0,i = H0 (S ; Hi (ΩS ; Z)) 6= 0. The dierential entering
r
E0,i is dr :
r
Er,i−r+1 r , and E r
→ E0,i r,i−r+1 is a subquotient of
2
Er,i−r+1 = Hr (S n ; Hi−r+1 (ΩS n ; Z))
but this vanishes as i−r+1 < i for r ≥ 2, and we supposed that i was minimal.
∞
Thus E0,i = 2
E0,i 6= 0, which is impossible as the spectral sequence converges to zero for
p + q > 0. Thus
Hi (ΩS n ; Z) = 0 for 0 < i < n − 1.
Now in degree n there is a unique possible non-zero dierential leaving position (n, 0),
which is
dn : En,0
n 2
= En,0 n
= Z −→ E0,n−1 = Hn−1 (ΩS n ; Z).
This must be injective, but is must also be surjective as no other dierential can kill the
group at position (0, n − 1). Thus
Hn−1 (ΩS n ; Z) ∼
= Z.
But now we can ll in the entire row q = n − 1, and we nd that
2
En,n−1 ∼
=Z too. The
pattern of rows now repeats exactly, giving
(
Z q = 0, n − 1, 2(n − 1), 3(n − 1), . . .
Hq (ΩS ; Z) ∼
n
=
0 otherwise.
The spectral sequence is therefore as shown in Figure 2.3. 4
0 1
Proof of Theorem 2.2.1. Let {b0 } = B ⊂ B ⊂ B 2 ⊂ · · · be the skeleta of B , and lter
−1 n
E by En := p (B ). Note that any simplex in E lies in some En , as any simplex in B
lies in some
n B . Following Example 2.1.6 there is an associated spectral sequence, with
1
Ep,q = Hp+q (Ep , Ep+1 ; G),
and by Theorem 2.1.8 this spectral sequence converges to Hp+q (E; G). To prove the
theorem we must therefore show that
2 ∼
Ep,q = Hp (B; Hq (F ; G))
under the assumption that π1 (B, b0 ) acts trivially on Hq (F ; G). To do that, we will
compute the homology of the chain complex
1 , d1 ). Let us omit the coecients G
(E•,•
from the notation: it plays no role.
If {iα : Dp → B p }α∈Ip are the characteristic maps of the p-cells of B, then we have
Hurewicz brations
pα := i∗α p : Eα := i∗α E −→ Dp .
Let ∂Eα = p−1
α (S
p−1 ), so that ∂Eα → S p−1 is also a bration.
Claim: The natural map

M
H∗ (Eα , ∂Eα ) −→ H∗ (Ep , Ep−1 )
α∈Ip
is an isomorphism.
F÷if÷ f:÷÷ .
not e
18
if
Z
÷÷
.
3h D Z -
T
. "
.
n - I 72 7L
#
¥¥p
0 I
-
.
.DZ 7#
p
. o
.
i
-
n
o
- -
1 Figure 2.3 The homological Serre spectral sequence for the path bration of S n .
2
l÷E "f÷÷÷
1
o oProof of claim.
o
tion retracts to
There
p−1 E
Z
O is an open neighbourhood of
p−1 p
(this is true for B ⊂ B
E p−1 in Ep which weakly
, by taking the complements of the
deforma-
centres of the discs: then use the homotopy lifting property). Thus by excision
. . . . .
∼
H∗ (Ep , Ep−1 ) −→ H∗ (Ep /Ep−1 , ∗),
.
but we also have
Ep /Ep−1 ∼
_
= Eα /∂Eα .
α∈Ip
pα : Eα → Dp , let
Now consider a single bration 0 ∈ Dp be a basepoint, and set
Fα := −1
pα3(0). Consider
l 2 4 the5homotopy lifting problem
proj inc
S p−1 × Fα × {0} Fα Eα
proj (v,t)7→t·v
S p−1 × Fα × [0, 1] S p−1 × [0, 1] Dp .
1
A weak deformation retraction of a space X to a subspace A is a homotopy from the identity map
of X to a map into A, which restricts to a homotopy of maps from A into A. It is weak" in that it need
not x A pointwise.
The lifted homotopy H̃ : S p−1 × Fα × [0, 1] → Dp satises H̃(v, x, 0) = x for all v ∈ S p−1 ,
so it descends to a map φ of brations
φ
Dp × Fα Eα
Dp Dp .
As the bres over 0 ∈ Dp Fα , it follows from the 5-lemma that φ is

of both brations are
a weak homotopy equivalence. Restricting to S
p−1 ⊂ Dp gives a map ∂φ : S p−1 × Fα →
∂Eα which is a weak homotopy equivalence for the same reason. Thus we have
H∗ (Eα , ∂Eα ) ∼
= H∗ (Dp × Fα , S p−1 × Fα )
∼
= H∗−p (Fα )
by the Künneth theorem for pairs. We therefore have
1 ∼
M
Ep,q = Hq (Fα ).
α∈Ip
Choosing a path from 0 ∈ int(Dp ) ⊂ B to b0 , Corollary 1.16.2 gives an isomorphism

∼
H∗ (Fα ) → H∗ (F ). Dierent choices of paths in principle give dierent isomorphisms
but by our assumption that π1 (B, b0 ) acts trivially on the homology of F in the current
situation they do not: thus we may canonically identify H∗ (Fα ) with H∗ (F ).
Thus we may write
1 ∼
M
Ep,q = Hq (F ) = Cpcell (B) ⊗ Hq (F ). (2.2.1)
α∈Ip
Claim: Under this isomorphism d1 = dcell ⊗ 1.
Proof of claim. Before starting the proof proper, we make two observations about natu-
rality:
(i) If f : C → B is a cellular map, C also having a single 0-cell c0 , then the ltration
∗ ∗
of f E given by (f E)p = (f p)
∗ −1 (C p ) is compatible with the ltration of E by
Ep , giving a map of exact couples and hence a map of spectral sequences. Under
1
the isomorphism (2.2.1) the induced map on E•,• is f# ⊗ 1.
(ii) If g : E 0 → E is a map of brations over B , with h : F 0 → F the induced map

0
on bres over b0 , then g(Ep ) ⊂ Ep , so again get a map of exact couples and hence
of spectral sequences. Under the isomorphism (2.2.1) the induced map on E•,• is
1
1 ⊗ h∗ .
Now, as it is enough to check the claim on basis elements of Cpcell (B), by (i) we may
suppose that B= Dp , having a single 0-cell, a (p − 1)-cell, and a p-cell. Then, using the
48 Chapter 2
÷F÷if÷ The Serre spectral sequence and applications
f:÷÷ .
equivalence φ : Dp × Fα → Eα over Dp , by (ii) it is enough to check the claim for the

trivial
p p
bration D × Fα → D , where we can write F = Fα . In this case we have
not e
18
if
Z
i:÷÷
" 1
Ep,q - .
3h D Z
-
1
Ep−1,q
Hp+q (Dp × F, S p−1 × F )

∂
dr
T
Hp+q−1 (S p−1 × F, {b0 } × F )
. "
.
Hp+q−1 (S p−1 × F)
which by naturality of the Künneth theorem for pairs is
∼
7L
∂⊗1
Hp (Dp , S p−1 ) ⊗ Hq (F ) −→ Hp−1 (S p−1 ) ⊗ Hq (F ) −→ Hp−1n(S Ip−172
, ∗) ⊗ Hq (F ) -
#
"
oh
¥¥p
17
as required.
.
1 , d1 ) with the cellular chain complex C•cell (B; Hq (F ; G)),

Having identied
2
(E•,q
I
0 I we
nd that Ep,q /is isomorphic to Hp (B; Hq (F ; G)) as required.
7#
←
.DZ
-
.
. o
Example 2.2.5. .
Acting on the unit vector (1, 0, 0) ∈ C3 gives a mapq : SU (3) → S 5,
which is easily checked to be a bre bundle and hence a Serre bration. The bre
i over -
n
o
- -
(1, 0, 0) ∈ S 5 is
• the stabiliser of this vector, so SU (2) ∼
= S3. Thus we have a spectral
sequence
2
D Ep,q = H1 p (S 5 ; Hq (S2
3
)) ⇒ Hp+q (SU (3)).
l÷E "f÷÷÷
19 K o o o O
Z
.
• . . . . .
D l 2 3 4 5
By considering the chart, there is no space for non-trivial dierentials, and in each
total degree p+q =d there is at most one non-trivial group. Thus
(
Z n = 0, 3, 5, 8
Hn (SU (3); Z) ∼
= 4
0 otherwise.
Lecture 14
f÷ f:÷÷ .
2.3 The Serre spectral sequence in cohomology 49
2.3 notspectral sequence

The Serre e
in cohomology
18
if
Z
÷
.
3h D Z -
T
There is similarly a spectral sequence for the cohomology of a ltered space, with two
changes: the indexing is slightly dierent, and the matter of convergence is, in general,
more complicated.
If ∅ ⊂ X0 ⊂ X1 ⊂ X2 ⊂ · · · ⊂ X is a sequence of subspaces, let
. " Ap,q = H p+q (X, Xp ; G)

.
E p,q = H p+q (Xp , Xp−1 ; G)
2 (X, X , X
so the long exact sequence on cohomology for the triple p p−1 ) gives maps
i : Ap,q −→ Ap−1,q+1 of degree (−1, 1)
n - I 72 j:A p,q
−→ E p+1,q−1
7Lof degree (1, −1)
#
p,q p,q+1
k:E −→ A of degree (0, 1)
¥p
Based on this one can invent the notion of a cohomological exact couple, which is
I precisely the same as an exact couple but with a dierent convention on degrees: the
Z 7#
above gives a cohomological exact couple of type 0 (by denition), and its rth derived
p
o
couple will have type r . The corresponding spectral sequence has dierentials
np+r,q−(r−1)
i dr : Erp,q −→ Er ,
-
o
- -
so have minus the degree of the homological dierentials. Its form is shown in Figure
2 2.4.
÷E "f÷÷÷
o o O
Z
. . . .
2 3 4 5
Figure 2.4 The cohomological spectral sequence for a ltered space.
In this situation each position (p, q) can only be the target of nitely-many dier-
entials, but can be the source of innitely-many. Thus for r 0 we have inclusions
Erp,q ⊃ Er+1
p,q p,q
⊃ Er+2 ⊃ · · · , and we dene
p,q p,q
\
p,q
E∞ := Erp,q = lim(· · · ,→ Er+2 ,→ Er+1 ,→ Erp,q ).
←−
r0
2
This is the long exact sequence on cohomology associated to the short exact sequence of cochain
∗ ∗ ∗
complexes 0 → C (X, Xp ) → C (X, Xp−1 ) → C (Xp , Xp−1 ) → 0 given by the evident maps.
Inverse limits are more subtle than direct limits, and in general one has to analyse this
quite carefully to understand what it means to converge to H ∗ (X). In the case of the
cohomological Serre spectral sequence the following will suce.
Theorem 2.3.1. ∅ ⊂ X0 ⊂ X1 ⊂ X2 ⊂ · · · ⊂ X
If is a sequence of subspaces such that
any simplex in X lies in some Xn and
E1p,q = H p+q (Xp , Xp−1 ; G) = 0 unless p≥0 and q ≥ 0,
then dening a ltration by
F p H n (X; G) := im(H n (X, Xp−1 ; G) → H n (X; G))
we have
p H n (X; G) = H n (X; G), p H n (X; G)

S T
(i) p≥0 F p≥0 F = 0, and
p,q ∼ F p H p+q (X;G)

(ii) E∞ = F p+1 H p+q (X;G)
.
Proof. Let us omit the coecients G from the notation. We have F 0 H n (X) = H n (X),
which gives the rst part of (i). For the second part consider
3 2 1
··· H n (X, Xp+1 ) H n (X, Xp ) H n (X, Xp−1 ) ···
H n (Xp+1 , Xp ) H n (Xp , Xp−1 ).
From the long exact sequence of the triple (X, Xp , Xp−1 ) and our assumption that
H n (Xp , Xp−1 ) = 0 if n − p < 0, we see that the map 1 is an epimorphism for p ≥ n + 1,
and by the analogous argument the maps 2 and 3 and so on are then all epimorphisms
too.
Suppose then that p ≥ n + 1, let ϕp−1 ∈ C n (X, Xp−1 ) be a relative cocycle, which
we consider as a homomorphism ϕp−1 : Cn (X; Z) → G which vanishes on Cn (Xp−1 ; Z).
Choose a sequence
ϕp ∈ C n (X, Xp ), ϕp+1 ∈ C n (X, Xp+1 ), ...
of relative cocycles whose cohomology classes correspond under the maps 1, 2, 3,
and so on. Now we must have
ϕp−1 = ϕp + ρp−1 ◦ d for some ρp−1 ∈ C n−1 (X, Xp−1 ),
as [ϕp−1 ] = [ϕp ] ∈ H n (X, Xp−1 ), and similarly there must be cochains
ρp ∈ C n−1 (X, Xp ), ρp+1 ∈ C n−1 (X, Xp+1 ), ρp+2 ∈ C n−1 (X, Xp+2 ), ...
such that ϕi = ϕi+1 + ρi ◦ d. Consider
φ : ρp−1 + ρp + ρp+1 + · · · : Cn−1 (X; Z) −→ G,

2.4 Multiplicative structure 51
which is well-dened as any x ∈ Cn−1 (X; Z) lies in Cn−1 (Xp−1+r ; Z) for some r 0, so
all but nitely-many ρi 's vanish on it.
Now we observe that if x ∈ Cn (Xq ; Z) then
ϕp−1 (x) = ρp−1 (dx) + ρp (dx) + · · · +

ρq
(dx) q+1 (dx) + · · ·
+ ρ

= φ(dx),
and every element of Cn (X; Z) lies in some Cn (Xq ; Z) so ϕp−1 = φ ◦ d, so [ϕp−1 ] = 0 ∈

H n (X, Xp−1 ). As ϕp−1 was arbitrary, H n (X, Xp−1 ) = 0, and hence F p H n (X; G) = 0,
which nishes the proof of (i).
Part (ii) is just as in Theorem 2.1.8.
Theorem 2.3.2. Let p : E → B be a Serre bration over a path-connected space with

bre F := p−1 (b0 ) so that π1 (B, b0 ) acts trivially on H ∗ (F ; G). Then there is a spectral
sequence with
E2p,q = H p (B; H q (F ; G))
p,q ∼ F p H p+q (E;G)
and with E∞ = F p+1 H p+q (E;G) , for a certain descending ltration F • H n (E; G) having
F 0 H n (E; G) = H n (E; G) and F m H n (E; G) = 0 for m > n.
Proof. Using the above convergence result this is just as for the homology Serre spectral
sequence, with a little care regarding direct sums vs. direct products.
2.4 Multiplicative structure

The great advantage of the cohomological Serre spectral sequence of a bration sequence
F →E→B is that it relates the cup-product structure on H ∗ (E) to that on H ∗ (B) and
H ∗ (F ). The following encapsulates all the necessary properties. We do not include the
proof: it is no more dicult than what we have do so far, but is not very enlightening.
Theorem 2.4.1. be a commutative ring, p : E → B be a Serre bration with path-

Let R
−1 ∗
connected base, and bre F = p (b0 ) such that π1 (B, b0 ) acts trivially on H (F ; R).
•,•
Then the Serre spectral sequence {(Er , dr )}r≥2 admits product maps
0 0 0 0
− · − : Erp,q ⊗ Erp ,q −→ Erp+p ,q+q for r≥2
such that
(i) Er•,• is a bigraded ring for each r ≥ 2,
(ii) dr : Er•,• → Er•,• is a derivation, i.e.
dr (x · y) = dr (x) · y + (−1)deg(x) x · dr (y),
where deg(x) = p + q when x ∈ Erp,q ,

•,•
(iii) Er+1 = H(Er•,• , dr ) as bigraded rings,
(iv) the isomorphisms E2p,q ∼

= H p (B; H q (F ; R)) assemble to an isomorphism of bigraded
rings, where the latter is given the product
0 0 ^ 0 0
H p (B; H q (F ; R)) ⊗ H p (B; H q (F ; R)) −→ H p+p (B; H q (F ; R) ⊗ H q (F ; R))
^ 0 0
−→ H p+p (B; H q+q (F ; R)),
(v) the ltration F • H ∗ (E; R) satises

0 0 0 0
F m H n (E; R) ^ F m H n (E; R) ⊂ F m+m H n+n (E; R),
so the cup product induces well-dened maps

0 0 0 0
F m H n (E; R) F m H n (E; R) F m+m H n+n (E; R)
⊗ −→ ,
F m+1 H n (E; R) F m0 +1 H n0 (E; R) F m+m0 +1 H n+n0 (E; R)
p,q ∼ F p H p+q (E;R)

(vi) the isomorphisms E∞ = F p+1 H p+q (E;R)
assemble to an isomorphism of bigraded
rings.
Remark 2.4.2. When the Universal Coecient Theorem applies to let us write
H p (B; H q (F ; R)) ∼
= H p (B; R) ⊗R H q (F ; R)
for all p and q, the multiplication in (iv) is given by
0
(x ⊗ y) · (x0 ⊗ y 0 ) = (−1)deg(y)deg(x ) (x ^ x0 ) ⊗ (y ^ y 0 ).
This sign may seem strange at rst, but it is necessary in order to make the Künneth
∼
isomorphism H ∗ (B; R) ⊗R H ∗ (F ; R) → H ∗ (B × F ; R) into a ring isomorphism.
Example 2.4.3. Let us revisit the bration sequence
ΩS n −→ P∗ S n −→ S n
from Example 2.2.4, with n > 1. We showed there that H∗ (ΩS n ; Z) is Z in degrees
divisible by (n − 1) and 0 otherwise; by the Universal Coecient Theorem H ∗ (ΩS n ; Z)
has the same description. The cohomological Serre spectral sequence must therefore be
as shown below.
Let us be concrete about generators. Use the Universal Coecient Theorem to write
E2p,q = H p (S n ; Z) ⊗ H q (ΩS n ; Z).
Let u ∈ H n (S n ; Z) be the standard generator, giving an element u ⊗ 1 ∈ E2n,0 and let

x1 ∈ H n−1 (ΩS n ; Z) be such that
dn (1 ⊗ x1 ) = u ⊗ 1.
More generally, assuming that xi−1 ∈ H (i−1)(n−1) (ΩS n ; Z) has been chosen, dene xi ∈
H i(n−1) (ΩS n ; Z) to be such that dn (1⊗xi ) = u⊗xi−1 . We have therefore chosen preferred
generators for all the
n
cohomology groups of ΩS , and so have
xi ^ xj = A(i, j)xi+j (2.4.1)

it s,
21 Zu④X,
.
3W ZX ,
×.
"
n - l IM U⑧X,
→
of#
n
p
l
-
D
- -
for some integers A(i, j). What are they?

22
:/
Using the multiplicative properties of the Serre spectral sequence we calculate
ds
.
7-
O⊗ xi ) · (1 ⊗ xj ))
d,
dn (1 ⊗ xi ^ xj ) = dn ((1
= (u ⊗ xi−1 ) · (1 ⊗ xj ) + (−1)i(n−1) (1 ⊗ xi ) · (u ⊗ xj−1 )
G N K 702 2B
:X
= u ⊗ (xi−1 ^ xj ) + (−1)i(n−1) (−1)i(n−1)(n) u ⊗ (xi ^ xj−1 )
= u ⊗ (xi−1 ^ xj ) + (−1)i(n−1) u ⊗ (xi ^ xj−1 ),
using that (n − 1)(n) is even. By the ansatz (2.4.1) this gives the recurrence relation
.
\g
x\
A(i, j) = A(i − 1, j) + (−1)i(n−1) A(i, j − 1).
3 D
Case 1: (n − 1) is even. Then we have A(i, j) = A(i − 1, j) + A(i, j − 1) which is solved
i+j ∗ n
by A(i, j) =
i . Thus zathe cup-product structure on H (ΩS ; Z) is given by
as
'T \ X

i+j
- o xi ^ xj = xi+j .
i
.
This is known as a freeo
#EW2
divided power algebra ΓZ [x1 ] on the class x1 .
Case 2: (n − 1) is odd. Then
O
we have
l A(i, j)
2
(−1)i A(i,
= A(i3− 1, j) + K 5 j − 1). 6
One
7 8
may verify that
i+j
A(2i + 1, 2j + 1) = 0, A(2i, 2j) =
i

i + j i + j
it A(2i, 2js
+ ,1) =
i
A(2i + 1, 2j) =
i
21 ZX , Zu④X,
.
3W
satises this recurrence.
This may be understood a little as follows. As x1 ∈ H n−1 (ΩS n ; Z) has odd degree,
2 2 2
by the graded-commutativity of the cup product we have x1 = −x1 and so 2x1 = 0, but
H 2(n−1) (ΩS n ; Z) = Z{x2 } is torsion-free, so x21 = 0. This ts with A(1, 1) = 0 above.
×
"
Now the above says that x1 ^ x2i = x.2i+1 . The above is then saying that we have a
ring isomorphism
H ∗ (ΩS n ; Z) = Z[x1 ]/(x21 ) ⊗ ΓZ [x2 ]. 4
Lecture 15
n l IM
Example
-
2.4.4. The long exact sequence
, U⑧X
on homotopy groups for the bration sequence
→ ΩK(Z, 3) −→ P∗ K(Z, 3) −→ K(Z, 3),
P K(Z, 3) is contractible, shows that ΩK(Z, 3) is a K(Z, 2).3

of#
and the fact that ∗ Thus it is
∞ ∗
weakly equivalent
Z[ι2 ]
to CP
for a class ι2 p
and so we know its cohomology ring: we have H (K(Z, 2); Z) =
of degree 2. We can take this to be the class we described in Section
n
l
-
D
- -
1.14, which also gives a ι3 ∈ H 3 (K(Z, 3); Z). We will explain how to produce the chart
shown in Figure 2.5 for the cohomological Serre spectral sequence of this bration
22
:/
ds
.
7-
O
d,
G N K 702 2B 782
:X . .
\g
x\
3 D
za as Nz
- o
'T \ X . ,
o
#EW2
O l 2 3 K 5 6 7 8 9
Figure 2.5 The cohomological Serre spectral sequence for the path bration of K(Z, 3).
We have H i (K(Z, 3); Z) = 0 for 0<i<3 and that H 3 (K(Z, 3); Z) = Z{ι3 }, by the
Universal Coecient Theorem and the Hurewicz Theorem. This lets us complete the
3
This is completely general: ΩK(G, n) ' K(G, n − 1).
E2 -page for p ≤ 3. As the cohomological Serre spectral sequence converges to zero in

positive degrees, we must have d3 (1 ⊗ ι2 ) = ±ι3 ⊗ 1; by rechoosing generators we can
suppose it is +ι3 ⊗ 1. By the derivation property we then have d3 (1 ⊗ ι22 ) = 2ι3 ⊗ ι2 , and
more generally
d3 (1 ⊗ ιn2 ) = n · ι3 ⊗ ι2n−1 ,
which completes the rst column of dierentials. We also deduce that H i (K(Z, 3); Z) = 0
for 3 < i < 6, as there is no way these groups could die in the spectral sequence.
2
As ι3 has odd degree, we have 2ι3 = 0. We also have
d3 (ι3 ⊗ ι2 ) = ι23
3,2
by the derivation property. In order to leave nothing in E∞ we must therefore have
H 6 (K(Z, 3); Z) = Z/2{ι23 }. This lets us complete the E2 -page for p ≤ 6. By the deriva-
tion property again
(
0 if n is even
d3 (ι3 ⊗ ιn2 ) = n · ι23 ⊗ ιn−1
2 = 2
ι3 ⊗ ι2n−1 if n is odd,
which completes the second column of dierentials. We also see H 7 (K(Z, 3); Z) = 0 as
there is no way this group could die in the spectral sequence.
We have found that E43,4 = Z/3{ι3 ⊗ ι22 }, and the only way this can die is if the
dierential
d5 : E53,4 −→ E58,0
is injective: similarly, the only way the group at (8, 0) can die is if this group is surjective.
Thus we have
H 8 (K(Z, 3); Z) ∼
= Z/3.
In the same way, d3 : E36,2 = Z/2{ι23 ⊗ ι2 } → E39,0 must be an isomorphism, so
H 9 (K(Z, 3); Z) ∼
= Z/2. In total, we have calculated
n 0 1 2 3 4 5 6 7 8 9 10
H n (K(Z, 3); Z) Z 0 0 Z 0 0 Z/2 0 Z/3 Z/2 ?
and so, using the Universal Coecient Theorem backwards,
n 0 1 2 3 4 5 6 7 8 9 10
Hn (K(Z, 3); Z) Z 0 0 Z 0 Z/2 0 Z/3 Z/2 ? ?
describes the homology of K(Z, 3) in degrees ≤ 8. 4
Example 2.4.5. Let f : S 3 → K(Z, 3) be a map that generates π3 (K(Z, 3)) ∼

= Z. Then
∼
there is a bration p : Ef → K(Z, 3) with bre F , and a weak equivalence w : S 3 → Ef
such that p ◦ w = f.
The long exact sequence on homotopy groups
··· π5 (K(Z, 3)) = 0

∂
∼
π4 (F ) π4 (S 3 ) π4 (K(Z, 3)) = 0
∂
0 ∼
π3 (F ) π3 (S 3 ) = Z π3 (K(Z, 3)) = Z ···
shows that F is 3-connected and that π4 (S 3 ) ∼

= π4 (F ): thus, by the Hurewicz theorem
we have π4 (S 3 ) ∼
= π4 (F ) ∼
= H4 (F ; Z).
Considering the homology Serre spectral sequence
1 ∼
Ep,q = Hp (K(Z, 3); Hq (F ; Z)) ⇒ Hp+q (Ef ; Z) ∼
= Hp+q (S 3 ; Z)
we see that the dierential

d5 : E0,4
4 4
−→ E5,0 = Z/2
must be an isomorphism, as its kernel or cokernel would contribute to H∗ (S 3 ; Z) in
degree ∗ ∈ {4, 5}, which is impossible. Thus H4 (F ; Z) ∼
= Z/2, so by the discussion above
π4 (S 3 ) ∼
= Z/2.
Together with Hopf 's theorem πn (S n ) ∼
= Z, and the Hopf bration sequence S1 →
S3 → 2
S , we now know the following homotopy groups of spheres
n 1 2 3 4
πn (S 3 ) 0 0 Z Z/2
πn (S 2 ) 0 Z Z Z/2
2.5 mod C theory

Denition 2.5.1. A Serre class C is a class of abelian groups closed under the opera-
tions of taking subgroups, quotients, and forming extensions.
Example 2.5.2.
(i) The class of nitely-generated abelian groups (using that Z is Noetherian).
(ii) The class of torsion abelian groups where each element is annihilated by products
of prime numbers in some set P.
(iii) The class of nite abelian groups in the above class. 4
Denition 2.5.3. For a Serre class C, a map f :A→B of abelian groups is
(i) injective mod C if Ker(f ) ∈ C ,

2.5 mod C theory 57
(ii) surjective mod C if B/Im(f ) ∈ C ,
(iii) an isomorphism mod C if both the above hold.
Lemma 2.5.4. A ⊗ B, Tor(A, B) ∈ C when A, B ∈ C . (The

Suppose in addition that
three examples above have this property.) Let p : E → B be a Serre bration with path-
−1
connected base, and bre F = p (b0 ) such that π1 (B, b0 ) acts trivially on H∗ (F ; Z), then
if any two of
H
e ∗ (F ; Z), H
e ∗ (E; Z), H
e ∗ (B; Z)
lie in C then so does the third. Lecture 16
Proof. Suppose rst that H e ∗ (B; Z) ∈ C ,

e ∗ (F ; Z), H and consider the Serre spectral se-
quence
2
Ep,q = Hp (B; Hq (F ; Z)) ⇒ Hp+q (E; Z).
As the class C is closed under extensions, to see that Hp+q (E; Z) ∈ C for p + q > 0 it
∞
suces to show that Ep,q ∈ C for p + q > 0. As the class C is closed under forming
subquotients, it suces to
2
show that Ep,q ∈ C for p + q > 0. The Universal Coecient
Theorem gives
0 −→ Hp (B; Z) ⊗ Hq (F ; Z) −→ Hp (B; Hq (F ; Z)) −→ Tor(Hp−1 (B; Z), Hq (F ; Z)) −→ 0
so as C is closed under ⊗ and Tor (and Z ⊗ A = A and Tor(Z, A) = 0) it follows that we

2
do indeed have Ep,q ∈ C for p + q > 0.
Suppose now that H e ∗ (E; Z) ∈ C . The lower corner of the Serre spectral
e ∗ (F ; Z), H
sequences gives an exact sequence
d2
H2 (B; Z) −→ H1 (F ; Z) −→ H1 (E; Z) −→ H1 (B; Z) −→ 0,
so H1 (B; Z) is a quotient of H1 (E; Z) and so lies in C. To generalise this, suppose then

that Hp (B; Z) ∈ C for 0 < p < k . We have an exact sequence
r+1 r r dr
0 −→ Ek,0 −→ Ek,0 −→ Ek−r,r−1
and
r
Ek−r,r−1 2
is a subquotient of Ek−r,r−1 = Hk−r (B; Hr−1 (F ; Z)). As we have supposed
that Hk−r (B; Z), Hk−r−1 (B; Z) ∈ C , by an application of the Universal Coecient The-
2 r r+1
orem as above we nd that Ek−r,r−1 ∈ C and hence Ek−r,r−1 ∈ C . Thus if Ek,0 ∈ C then
E r ∈ C . As He k (E; Z) ∈ C so are its ltration quotients, so E k+1 = E ∞ ∈ C . Thus by

k,0 k,0 k,0
downwards induction we nd that
2 = H (B; Z) ∈ C as required.
Ek,0 k
Under the assumption that H∗ (B; Z), H∗ (E; Z) ∈ C , similar reasoning
e e shows that
e ∗ (F ; Z) ∈ C .
H
Lemma 2.5.5. Let C be either
(i) the class of nitely-generated abelian groups, or
(ii) the class of nite P -torsion abelian groups.

If A∈C then Hi (K(A, n); Z) ∈ C for all i, n > 0.
Proof. By considering the bration sequences
K(A, n − 1) ' ΩK(A, n) −→ P∗ K(A, n) −→ K(A, n),
with P∗ K(A, n) ' ∗, the previous lemma reduces the claim to the case n = 1.
Finitely-generated: A = Then Zr
⊕ Z/n1 ⊕ · · · ⊕ Z/nk so we can take K(A, 1) =
(S 1 )r
× K(Z/n1 , 1) × · · · × K(Z/nk , 1), so by the Künneth theorem we are reduced to
showing that each He i (K(Z/n, 1); Z) is nitely-generated.
2
The class n · ι2 ∈ H (K(Z, 2); Z) corresponds to a map fn : K(Z, 2) → K(Z, 2),
which on π2 (−) induces multiplication by n. It follows from the long exact sequence on
homotopy groups that the homotopy bre of fn is K(Z/n, 1), so there is a homotopy
bre sequence
fn
K(Z/n, 1) −→ K(Z, 2) −→ K(Z, 2).
As K(Z, 2) ' CP∞ has nitely-generated homology groups, it follows from the previous
lemma that K(Z/n, 1) does too.
Finite P -torsion: Then A = Z/pn1 1 ⊕ · · · ⊕ Z/pnk k for pj ∈ P , so as above it suces

to show that each e i (K(Z/pn , 1); Z) is nite p-torsion. We have the homotopy bre
H
sequence
i fp n
K(Z/pn , 1) −→ K(Z, 2) −→ K(Z, 2)
and by the long exact sequence on homotopy groups the homotopy bre of i is a K(Z, 1) '
S 1 , so we have a homotopy bre sequence
i
S 1 −→ K(Z/pn , 1) −→ K(Z, 2). (2.5.1)
By the Hurewicz theorem we have H1 (K(Z/pn , 1); Z) ∼

= Z/pn , and so by the Universal
Coecient Theorem we have
H 1 (K(Z/pn , 1); Z) = 0 H 2 (K(Z/pn , 1); Z) ⊃ Z/pn .
Thus the cohomology Serre spectral sequence for (2.5.1) must take the following form
23
'
•
tune
ku
Epi
0 I
"
2 3
.
"
U 5
"
6
I
24 .
in •
""
"t '
2.5 mod C theory 59
so that 
Z
 i=0
i n ∼
H (K(Z/p , 1); Z) = Z/pn i > 0 even

0 i odd.

By the Universal Coecient Theorem (backwards!) each e i (K(Z/pn , 1); Z)

H is therefore
nite p-torsion, as required.
Theorem 2.5.6 (mod C Hurewicz) . Let C be either
(i) the class of nitely-generated abelian groups, or
(ii) the class of nite P -torsion abelian groups.
If X is 1-connected and πi (X, x0 ) ∈ C for 0 < i < n, then Hi (X; Z) ∈ C for 0<i<n
and
h : πn (X, x0 ) −→ Hn (X; Z)
is an isomorphism mod C.
Proof. Consider the Postnikov tower of X,
.
.
.
X2
p2
f2
X1
f1
p1
f0
X X0
and recall that the homotopy bre of pk : Xk → Xk−1 is a K(πk (X, x0 ), k).
So if πi (X, x0 ) ∈ C for 0 < i < n, then by induction using the homotopy bre
sequences
pi
K(πi (X, x0 ), i) −→ Xi −→ Xi−1
we nd that e ∗ (Xn−1 ; Z) ∈ C .
H As fn−1 : X → Xn−1 is n-connected, we have
∼
Hi (X; Z) −→ Hi (Xn−1 ; Z) for i<n
so Hi (X; Z) ∈ C for 0 < i < n. Lecture 17
The map fn : X → Xn is (n + 1)-connected, so an isomorphism on nth homology.

The homology Serre spectral sequence for the homotopy bre sequence
pn
K(πn (X, x0 ), n) −→ Xn −→ Xn−1
has the form

60
'
•
tune
ku
Chapter 2
Epi
0 I
"
2 3
The Serre spectral sequence and applications
.
"
U 5
"
6
I
24
§
.
in •
""
"t '
I 0
°
#•
n I n htt ntz
3
-
O l 2
so gives an exact sequence

25 ② e3 ⑥ 250×3
/
.
if
dn+1
Hn+1 (Xn−1 ; Z) Hn (K(πn (X, x0 ), n); Z) Hn (Xn ; Z) Hn (Xn−1 ; Z)
• h
πn (X, x0 )
④
'
K2
Q2 @ K2
where the two outer terms lie in C : thus the kernel and cokernel of map h lie in C , so h
is an isomorphism mod C.
:
Corollary 2.5.7. πi (S n ) is a nitely-generated abelian group for all i and n.
Proof. For n=1

we know these homotopy groups, so suppose n > 1. Suppose not, and
let i n .
be minimal such that πi (S ) is not .
on
nitely-generated.
n n
Then h : πi (S ) → Hi (S ; Z)
of
is an isomorphism mod nitely-generated abelian groups, so Hi (S n ; Z) is also not nitely-
generated, a contradiction.
More generally, this argument shows that if X is a 1-connected space with each
Hi (X; Z) nitely-generated, then
, each πi (X, x0 ) is also nitely-generated.
←¥
Corollary 2.5.8. The groups π (S 3 ) are nite for i > 3.
i
Proof. Let f : S 3 → K(Z, 3) represent a generator of H 3 (S 3 ; Z), and X denote its

homotopy bre. By Lemmas 2.5.4 and 2.5.5 the homology groups of X are nitely-
generated.
Let F denote the Serre class of nite abelian groups. We have
πi (S 3 ) nite for i > 3 ⇐⇒ πi (X) nite for all i (by LES of homotopy groups)
⇐⇒ H
e i (X; Z) nite for all i (by mod F Hurewicz Theorem)
⇐⇒ H e i (X; Z) ⊗ Q = 0
e i (X; Q) = H for all i
'
•
tune
ku
0 I
"
2
Epi
32.5
.
mod C theory
U
"
61 5
"
where the last implication is because the homology of X is nitely-generated.
24 X has trivial Q-homology, we consider the Q-cohomology Serre spectral
§
To show that
.
sequence for homotopy bre sequence

in •
K(Z, 2) ' ΩK(Z, 3) −→ P∗ K(Z, 3) −→ K(Z, 3)

""
"t '
which takes the form shown in Figure 2.6, showing that
(
i Q i = 0, 3
H (K(Z, 3); Q) =
0 else.
I 0
It follows that the map f : S 3 −→ K(Z, 3) is an isomorphism on H ∗ (−; Q), and therefore
3 f
°
from the Serre spectral
# •
X → S → K(Z, 3)
sequence for the homotopy bre sequence
that H∗ (X; Q) = Q, so H∗ (X; Q) = 0
e as required.
n I n htt ntz
3
-
O l 2
25 ② e3 ⑥ 250×3
/
.
④ K2
if '
Q2 @ K2
: on . .
of ,
←¥
Figure 2.6 The cohomological Serre spectral sequence for the path bration of K(Z, 3) with
Q-coecients.
2.6 The transgression

p
For F →E→B a Serre bration of path-connected spaces, consider
∂
Hi (E, F ) Hi−1 (F )
p∗
j∗
Hi (B) Hi (B, b0 ).
Say that (x, y) ∈ Hi (B) × Hi−1 (F ) is a transgressive pair if there is a z ∈ Hi (E, F )

such that
p∗ (z) = j∗ (x) and ∂(z) = y.
This gives a partially-dened and many-valued function
Hi (B) Hi−1 (F ),
called the transgression. The following lemma relates it to those dierentials in the
Serre spectral sequence going from the horizontal to the vertical edge.
Lemma 2.6.1. A pair (x, y) ∈ Hi (B) × Hi−1 (F ) is transgressive if and only if in the
Serre spectral sequence
2
the class x ∈ Hi (B) = Hi (B; H0 (F )) = Ei,0 survives until Ei,0
i
and
di (x) = [y] ∈ E0,i−1

i
, a quotient of
2
E0,i−1 = H0 (B; Hi−1 (F )) = Hi−1 (F ).
Proof. Let z ∈ Hi (E, F ) exhibit (x, y) as a transgressive pair, recall that En := p−1 (B n ),
and consider
Hi (Ei , F ) Hi (E, F ) Hi (B, b0 )
z̄ z j∗ (x),
where the rst map is surjective as Hi (E, Ei ) = 0, and so a z̄ can indeed be chosen. The
spectral sequence comes from the exact couple
L i L
p,q Hp+q−1 (Ep−1 ) p,q Hp+q (Ep )
k j
L
p,q Hp+q (Ep , Ep−1 )
and the class

2
x ∈ Ei,0 is represented by the image z̄¯ of z̄ under
1
Hi (Ei , F ) = Hi (Ei , E0 ) −→ Hi (Ei , Ei−1 ) = Ei,0 .
Thus dr (x) is given by j(i−1 )r k(z̄¯). But k(z̄¯) = ∂(z̄¯) ∈ Hi−1 (Ei−1 ) and we know this
r
lifts to ∂(z̄) ∈ Hi−1 (F ) = Hi−1 (E0 ). This shows that d (x) = 0 for r < i, and that
∼
d (x) = j∂(z̄) = j∂(z), but this is j : Hi−1 (E0 ) → Hi−1 (E0 , ∅), so di (x) is represented
i
by ∂(z) = y ∈ Hi−1 (F ).
The reverse direction is similar.
2.7 Freudenthal's suspension theorem 63
There is a similar discussion in cohomology. A pair (x, y) ∈ H i+1 (B) × H i (F ) is

called transgressive if there is a z∈ H i+1 (B, b0 ) such that in the diagram
δ
H i+1 (E, F ) H i (F )
p∗
j∗
H i+1 (B) H i+1 (B, b0 ).
we have j ∗ (z) = x and p∗ (z) = δ(y). As in the lemma, this is equivalent to y ∈ H i (F ) =

E20,i surviving until 0,i
Ei+1 and satisfying di+1 (y) = [x]. Lecture 18
2.7 Freudenthal's suspension theorem

Theorem 2.7.1. Let X be an (n − 1)-connected based space. Then the suspension map
Σ : πi (X) −→ πi+1 (ΣX)

[f : S i → X] 7−→ [Σf : ΣS i = S i+1 → ΣX]
is an epimorphism for i ≤ 2n − 1 and an isomorphism for i ≤ 2n − 2.
Proof. For n=1 this is just Theorem 1.13.7, so suppose that n − 1 ≥ 1. Consider the
path bration
ΩΣX −→ P∗ ΣX −→ ΣX,
and note that ΣX is n-connected (by the Hurewicz theorem and the suspension isomor-
phism e ∗ (ΣX) ∼
H =He ∗−1 (X)), so ΩΣX is (n − 1)-connected. The Serre spectral sequence
has the form
ZG '
nt , ⑧ • • •
n ⑧ • A O
n
tdndegree
i
i D
o
##-
O l n htt ntz nt }
27 .
'
e5 et Et +8
its
'
1 x x x
/ \ sq
'
• • • • • • • • → •
-
-
Thus the dierentials di : Hi (ΣX) −→ Hi−1 (ΩΣX) are isomorphisms for i − 1 < 2n.
These dierentials are transgressions so are given by
∼
Hi (P∗ ΣX, ΩΣX) Hi−1 (ΩΣX)
p∗
∼
Hi (ΣX) Hi (ΣX, ∗)
and hence
p∗ : Hi (P∗ ΣX, ΩΣX) −→ Hi (ΣX, ∗)
is an isomorphism for i ≤ 2n.
Consider the map
f : CX −→ P∗ ΣX
[t, x] 7−→ (s 7→ [st, x])
where we consider CX and ΣX as quotients of [0, 1] × X . Then there is a commutative

square
fˆ
X ΩΣX
x7→[x,1]
f
CX P∗ ΣX
where fˆ(x)(s) = [s, x]. The map on long exact sequences on homology gives
∼
Hi (CX, X) Hi−1 (X)
(f,fˆ)∗ fˆ∗
∼
∼ Hi (P∗ ΣX, ΩΣX) Hi−1 (ΩΣX)
p∗
Hi (ΣX, ∗)
where the curved arrow is an isomorphism by excision. As the map p∗ is an isomorphism

for i ≤ 2n, it follows that
fˆ∗ : H∗ (X) −→ H∗ (ΩΣX)
is an isomorphism for ∗ ≤ 2n − 1.
As n − 1 ≥ 1, ΩΣX is 1-connected. Letting F denote the homotopy bre of the map
ˆ e ∗ (F ) = 0 for ∗ ≤ 2n − 2. By the
f , it then follows from the Serre spectral sequence that H
Hurewicz theorem (as π1 (F ) must be a quotient of π2 (ΩΣX) and so abelian) it follows
that F is (2n − 2)-connected, so by the long exact sequence
fˆ∗ ∂
· · · −→ πi (F ) −→ πi (X) −→ πi (ΩΣX) −→ πi−1 (F ) −→ · · ·
the map fˆ∗ : πi (X) −→ πi (ΩΣX) ∼

= πi+1 (ΣX) is an epimorphism for i ≤ 2n − 1 and an
isomorphism for i ≤ 2n − 2.
2.7 Freudenthal's suspension theorem 65
Corollary 2.7.2. The map Σ : πi (S 3 ) → πi+1 (S 4 ) is an epimorphism for i≤5 and an

isomorphism for i ≤ 4, so in particular
∼
Z/2 = π4 (S 3 ) −→ π5 (S 4 ).
As we continue the range in Freudenthal's theorem only gets better, so πn+1 (S n ) = Z/2
for all n ≥ 3.
By Freudenthal's theorem πi+n (Σn X) is independent of n as long as n 0: we

call its stable value πi (X), the ith stable homotopy group of X . Tautologically it
s
satises πi (ΣX) ∼
s s (X), and in particular π s (S n ) = π s (S 0 ), so there is only a
= πi−1 i i−n
s
1-parameter family of stable homotopy groups of spheres, abbreviated πi := πi (S ).
s 0
We have calculated
π0s = Z π1s = Z/2
and the next few are
i 0 1 2 3 4 5 6 7 8
πis Z Z/2 Z/2 Z/24 0 0 Z/2 Z/240 Z/2 ⊕ Z/2
There is not an obvious pattern: these groups are a topic of active research.
Chapter 3
Cohomology operations
3.1 Steenrod squares

Theorem 3.1.1. There are natural (for maps of pairs) homomophisms
Sqi : H n (X, A; Z/2) −→ H n+i (X, A; Z/2), i≥0
satisfying
(i) Sq0 = Id,
(ii) Sqi (x) = x2 if i = |x|,
(iii) Sqi (x) = 0 i > |x|,

if
Sqk (x ^ y) = i+j=k Sqi (x) ^ Sqj (y),

P
(iv)
(v) δ ◦ Sqi = Sqi ◦ δ for the connecting map δ : H n (A; Z/2) → H n+1 (X, A; Z/2),
(vi) σ◦Sqi = Sqi ◦σ for the suspension isomorphism σ : H

e n (X; Z/2) → H
e n+1 (ΣX; Z/2),
(vii) Sq1 is the Bockstein operation associated to the exact sequence
0 −→ Z/2 −→ Z/4 −→ Z/2 −→ 0
of coecients.
P∞ i
Sometimes we write Sq = i=0 Sq . Then Sq(x) still makes sense, as the sum is
nite by (iii), and (iv) may then be expressed as Sq(x ^ y) = Sq(x) ^ Sq(y). We will
come back to construct the Sqi and prove this theorem in Section 3.4, but will rst give
Lecture 19 several applications which only use these properties.
Example 3.1.2. CP2 = S 2 ∪h D4

Recall that for h : S3 → S2 the Hopf map. As a ring
we have H ∗ (CP2 ; Z/2) = Z/2[x]/(x3 ) and so
Sq2 (x) = x2 6= 0.
Now Σn CP2 has cohomology
i 0 1 ··· n+1 n+2 n+3 n+4

H i (Σn CP2 ; Z/2) Z/2 0 ··· 0 Z/2 0 Z/2
generator 1 ··· σ n (x) σ n (x2 )
66
3.1 Steenrod squares 67
and we have
Sq2 (σ n (x)) = σ n (Sq2 (x)) = σ n (x2 ) 6= 0.
It follows that Σn CP2 = S n+2 ∪Σn h Dn+4 '
6 S n+2 ∨ S n+4 , and so Σn h : S n+3 → S n+2
is not homotopic to a constant map for any n ≥ 0.
In particular, in Example 2.4.5 we had calculated π4 (S ) ∼
3
= Z/2 as abstract groups,
but it now follows that this must be generated by the suspension Σh of the Hopf map.
n
Using Corollary 2.7.2 it then follows that πn+1 (S ) is generated by the appropriate
suspension of the Hopf map for any n ≥ 3. 4
Lemma 3.1.3. In H ∗ (RPn ; Z/2) = Z/2[x]/(xn+1 ) we have

i k k k+i
Sq (x ) = x ,
i
where the binomial coecient is taken modulo 2.
Proof. We have
Sq(x) = Sq0 (x) + Sq1 (x) + 2

Sq(x)

+ ···
= x + x2
using properties (i), (ii), and (iii) of the Steenrod squares, so by property (iv) we have
k
X k
Sq(xk ) = Sq(x)k = (x + x2 )k = xk+i
i
i=0
as required.
Lemma 3.1.4. In H ∗ (CPn ; Z/2) = Z/2[y]/(y n+1 ) we have

2i k k k+i
Sq (y ) = y ,
i
where the binomial coecient is taken modulo 2, and Sq2i+1 (y k ) = 0.
The following elementary lemma is convenient for making calculations with binomial
coecients modulo 2.
Lemma 3.1.5.
P`
k = ì=0 ki 2i are the binary
i
P
If n= i=0 ni 2 and expansions, then

n n0 n1 n`
= ··· mod 2.
k k0 k1 k`
The rst few Steenrod squares in H ∗ (RP∞ ; Z/2) can be visualised as in Figure 3.1.
n
In particular we see that every x may be linked to x by a zig-zag of Sq1 's and Sq2 's,
∞
so there can be no decomposition RP ' X ∨ Y with both X and Y having nontrivial
Z/2-cohomology. Similarly for Σn RP∞ with any n ≥ 0.
i
i D
o
##-
O l n htt ntz nt }
68 Chapter 3 Cohomology operations

27 .
'
e5 et Et +8
its
'
1 x x x
/ \ sq
'
• • • • • • • • → •
qi -y
-
sp
Figure 3.1 Steenrod operations in H ∗ (RP∞ ; Z/2).
28
!÷.
i
[ totgludegree
.
:
Example 3.1.6.
!
n k
k
Let n+1= 2r (2s + .1), and•consider RP /RP . If n − 2r > k there is
!
" '
• • . . . •
n−2r (RPn /RPk ; Z/2) which under the quotient map q : RPn → RPn /RPk
)
a class z ∈ H
∗
satises q (z) = x
n−2r . Thus
n − 2r n
r+1
∗ 2r 2r n−2r 2 s−1 n
q (Sq (z)) = Sq (x )= x = x = xn 6= 0
2r 2r
o l rk htt n -
I Zktz
using Lemma 3.1.5, and so Sq2 (z) 6= 0 ∈ H n (RPn /RPk ; Z/2) = Z/2.
If the map
f : RPn /RPk −→ RPn /RPn−1 = S n
had a right inverse up to homotopy, g : S n → RPn /RPk , then
g ∗ : H n (RPn /RPk ; Z/2) −→ H n (S n ; Z/2)

r r
would be an isomorphism, so we would have Sq2 (g ∗ (z)) = g ∗ (Sq2 (z)) 6= 0. But g ∗ (z)
lies in the group H
n−2r (S n ; Z/2) which vanishes: this is a contradiction.
n k n r
Thus the map f : RP /RP → S does not have a right inverse if n − 2 > k. 4
Theorem 3.1.7 . Let p : E → B be a Serre bration

(Kudo's trangesssion theorem)
−1
with path-connected base, and bre F = p (b0 ) such that π1 (B, b0 ) acts trivially on
2 survives until E i+1 and then di+1 (y) = [x], then
H ∗ (F ; Z/2). If y ∈ H i (F ; Z/2) = E0,i 0,i
i+n+1
Sqn (y) survives until E0,i+n and d
n+i+1 (Sqn (y)) = [Sqn (x)].
Proof. Under the assumptions (x, y) is a transgressive pair, so there is a z ∈ H i+1 (B, b0 )
so that in
δ
H i+1 (E, F ) H i (F )
p∗
j∗
H i+1 (B) H i+1 (B, b0 ).
we have j ∗ (z) = x and p∗ (z) = δ(y). But then Sqn (z) exhibits (Sqn (x), Sqn (y)) as being
a transgressive pair.
Lecture 20
3.2 Vector elds on spheres 69
3.2 Vector elds on spheres

For a 1-dimensional subspace ` ⊂ Rn , let r` denote reection in the line `: so r` |` = −Id
and r` |`⊥ = Id. Let r0 = r(1,0,...,0) . Then r0 r` is an orthogonal transformation of
determinant 1, so denes a map
Jn : RPn−1 −→ SO(n)
` 7−→ r0 r` .
` ∈ RPk−1 ⊂ RPn−1 , then Jn (`)

If xes the last (n − k) basis vectors, so lands in
SO(k) ⊂ SO(n). Ths we get map
qn,k : RPn−1 /RPk−1 −→ SO(n)/SO(k).
The two instances of / have dierent meanings : on the left RPn−1 /RPk−1 denotes
k−1
collapsing RP to a point, whereas on the right SO(n)/SO(k) denotes taking the
orbits of the SO(k)-action.
Lemma 3.2.1. Ifk = n − 1 then this is a map qn,n−1 : S n−1 → S n−1 and induces an
isomorphism on H∗ (−; Z) (i.e. it has degree ±1).
∼
=
Proof. The homeomorphism SO(n)/SO(n − 1) → S n−1 is given by sending a matrix in
SO(n) to its last column, considered as a unit vector in Rn . In other words, it is given
by acting on the last basis vector SO(n).
Thus the composition
qn,n−1
RPn−1 −→ RPn−1 /RPn−2 −→ SO(n)/SO(n − 1) ∼
= S n−1
sends ` to r0 r` (en ). The preimage of −en under this map consists of those ` such that
r0 r` (en ) = −en , so r` (en ) = −en . There is only one such `, namely ` = hen i, so the map
qn,n−1 has degree ±1.
Proposition 3.2.2. The map qn,k induces an isomorphism on H ∗ (−; Z) for ∗ ≤ n−1
as long as n ≤ 2k + 1.
Proof. When k = n − 1 we have proved this in the last lemma, so we proceed by down-
wards induction on k . Consider the commutative diagram
∼
=
RPk /RPk−1 SO(k + 1)/SO(k)
qn,k
RPn−1 /RPk−1 SO(n)/SO(k)
quotient π
qn,k+1
RPn−1 /RPk SO(n)/SO(k + 1)
where the right-hand column is a bration sequence but the left-hand column is not.
'
e5 et Et +8
its
'
1 x x x
/ \ sq
'
• • • • • • • • → •
qi -y
-
70 Chapter 3 Cohomology operations sp
28
!÷.
i
[ totgludegree
.
k
!
!
. • " '
• • . . . •
)
o l k htt n -
I Zktz
Figure 3.2
By inductive hypothesis the map qn,k+1 is an isomorphism in cohomology in degrees

∗ ≤ n − 1, so the Serre spectral sequence for the right-hand column takes the form shown
in Figure 3.2.
In degrees ∗ + 1 ≤ 2k + 1 we obtain a map of exact sequences
dk+1 SO(n) dk+1

H ∗ ( SO(k+1) ) H ∗ ( SO(n)
SO(k) ) H ∗ ( SO(k+1)
SO(k) )
SO(n)
H ∗+1 ( SO(k+1) )
∗
qn,k+1 ∗
qn,k ∼ ∗
qn,k+1
=
δ n−1 n−1 k δ n−1

H ∗ ( RP
RPk
) H ∗ ( RP
RPk−1
) H ∗ ( RP
RP
k−1 ) H ∗+1 ( RP
RPk
),
where the rightmost square commutes by the zig-zag description of the transgression
(Section 2.6) and the other squares clearly commute. By the 5-lemma we nd that
∗
qn,k
is an isomorphism for ∗ ≤ n − 1, as long as n − 1 ≤ 2k .
As RPn−1 /RPk−1 and SO(n)/SO(k) are easily seen to be 1-connected, we deduce
Corollary 3.2.3. If n ≤ 2k + 1 then the map qn,k is (n − 1)-connected.

Theorem 3.2.4. Let n+1 = 2r (2s+1). Then Sn does not admit 2r linearly-independent
vector elds.
Proof. Note if s = 0 then S n clearly does not admit 2r = n + 1 linearly independent

vector elds, and if r = 0 then n is even and so admits no nonvanishing vector elds.
r
Thus we may suppose that 2 ≤ n/2.
Consider the map
π : SO(n + 1)/SO(n − k) −→ S n
given by evaluating at the last basis vector. There is a bijection
∼
{orthonormal (k + 1)-tuples in Rn+1 } ←− SO(n + 1)/SO(n − k)
Aen−k+1 , Aen−k+2 , . . . Aen+1 ←−[ A
3.3 Wu and StiefelWhitney classes 71
under which the map π corresponds to recording the (k + 1)-st orthogonal vector.
If Sn admits k linearly independent vector elds then by applying the GramSchmidt
process it admits k orthonormal vector elds s1 , . . . sk , and so
s : S n −→ {orthonormal (k + 1)-tuples in Rn+1 } = SO(n + 1)/SO(n − k)

x 7−→ (s1 (x), s2 (x), . . . , sk (x), x)
π ◦ s = IdS n .
satises
n + 1 ≤ 2(n − k) + 1 (i.e. if 2k ≤ n) then by Corollary 3.2.3 the map qn+1,n−k :
If
RPn /RPn−k−1 → SO(n + 1)/SO(n − k) is n-connected, and so there exists a map
s0 : S n −→ RPn /RPn−k−1
such thatqn+1,n−k ◦ s0 ' s. This map s0 would be right inverse up to homotopy to

n n−k−1
the map RP /RP → RPn /RPn−1 = S n , but by Example 3.1.6 this means that
n − 2 ≤ n − k − 1, so k < 2r .
r
In fact if n + 1 = 2r (2s + 1) and r = c + 4d with 0 ≤ c < 4, then S n admits 2c + 8d − 1

linearly-independent vector elds and no more. Lecture 21
3.3 Wu and StiefelWhitney classes

Let M be a closed compact n-dimensional manifold, so the map
H i (M ; Z/2) −→ Hom(H n−i (M ; Z/2), Z/2)

x 7−→ hx ^ −, [M ]i
is an isomorphism by Poincaré duality. There is a linear map given by
H n−i (M ; Z/2) −→ Z/2

y 7−→ hSqi (y), [M ]i
which therefore corresponds under the Poincaré duality isomorphism to a class vi ∈

H i (M ; Z/2), uniquely determined by
hvi ^ y, [M ]i = hSqi (y), [M ]i ∀y ∈ H n−i (M ; Z/2).
This is the ith Wu class of M. We write
∞
M
v := 1 + v1 + v2 + · · · ∈ H i (M ; Z/2)
i=0
for the formal sum, the total Wu class. We then write

X
w := Sq(v) = Sqi (vj )
i,j
for the total StiefelWhitney class of M , with components w = 1 + w1 + w2 + · · · .

Example 3.3.1. RPn Sqi (xn−i ) = n−i

xn , n−i
xi .

On we have
i and so vi = i Thus
P∞ n−i
xi , and so

v= i=0 i
∞ ∞ X
i
X n−i i i
X n−i i i+j
w = Sq(v) = x (1 + x) = x .
i i j
i=0 i=0 j=0
It is an exercise with binomal coecients modulo 2 to see that this is (1 + x)n+1 , so

n+1
xi .

wi = i 4
Remark 3.3.2. Note that Sqi : H n−i (M ; Z/2) → H n (M ; Z/2) vanishes if i > n − i,
by property (iii) of Steenrod squares. Thus vi = 0 for 2i > n. Thus there are fewer
Wu classes than StiefelWhitney classes, which means that StiefelWhitney classes must
satisfy certain relations.
For example, if n=4 then v = 1 + v1 + v2 and so
w = 1 + (v1 ) + (v2 + v12 ) + (Sq1 (v2 )) + (v22 ).
Thus we have
w3 = Sq1 (v2 ) = Sq1 (v2 + v12 ) = Sq1 (w2 )

w4 = w2 + w12 .
So if w1 = w2 = 0 then w3 = w4 = 0 too.
Suppose that M is a smooth manifold and is embedded in Rn+k with normal bundle
ν, and let M ⊂ U ⊂ Rn+k be a tubular neighbourhood. There is a collapse map
S n+k
c : S n+k −→ = U+
S n+k \ U
and the space U+ may be identied with the Thom space Th(ν) = D(ν)
S(ν) . Recall that
the Thom class is a class e k (Th(ν); Z/2), and the (co)homology Thom isomorphisms
u∈H
are the maps
−û
H i (M ; Z/2) ∼
= H i (D(ν); Z/2) −→ H i+k (D(ν), S(ν); Z/2) ∼
=He i+k (Th(ν); Z/2)
e i+k (Th(ν); Z/2) ∼ u_−
H = Hi+k (D(ν), S(ν); Z/2) −→ Hi (D(ν); Z/2) ∼
= Hi (M ; Z/2).
Abusing notation slightly, we write
∼ e i+k
− ^ u : H i (M ; Z/2) −→ H (Th(ν); Z/2).
Proposition 3.3.3. We have
1 e ∗ (Th(ν); Z/2).
Sq(u) = û∈H
w
3.3 Wu and StiefelWhitney classes 73
Proof. As Sq0 = Id the operator Sq is formally invertible: call its inverse Sq−1 . Let us
write c∗ [S n+k ] =: [T h], so that u _ [T h] = [M ] under the homology Thom isomorphism.
The dening property of the Wu class gives
hSq(x) ^ u, [T h]i = hSq(x), u _ [T h]i

= hSq(x), [M ]i
= hv ^ x, [M ]i
= hv ^ x ^ u, [T h]i
for any x, and the left-hand side is hSq(x ^ Sq−1 (u)), [T h]i. Now [T h] = c∗ [S n+k ] so
this is
hc∗ (Sq(x ^ Sq−1 (u))), [S n+k ]i = hSq(c∗ (x ^ Sq−1 (u))), [S n+k ]i.
But in H ∗ (S n+k ; Z/2) we have Sq = Id (there is no space for any other operations) so
this is
hc∗ (x ^ Sq−1 (u)), [S n+k ]i = hx ^ Sq−1 (u), [T h]i.
In total we obtain the identity
hx ^ Sq−1 (u), [T h]i = hv ^ x ^ u, [T h]i

for all x ∈ H ∗ (M ; Z/2), so by the Thom isomorphism and Poincaré duality we have
−1
Sq (u) = v ^ u, and hence u = Sq(v) ^ Sq(u) giving Sq(u) = w1 ^ u as required.
Corollary 3.3.4. If Mn embeds into Rn+k then [ w1 ]i = 0 for all i ≥ k.
Proof. If i > k then Sqi (u) = 0 as |u| = k , so [ w1 ]i = 0.
i 1
2
If i = k then Sq (u) = u = [ ]k ^ u.
w If [ w1 ]k 6= 0 then by Poincaré duality there is
an x ∈ H n−k (M ; Z/2) such that
1 = hx ^ [ w1 ]k , [M ]i = hx ^ [ w1 ]k ^ u, [T h]i
= hx ^ u ^ u, [T h]i
= hc∗ (x ^ u) ^ c∗ (u), [S n+k ]i
which is a contradiction as all nontrivial cup products in H ∗ (S n+k ; Z/2) vanish.
Example 3.3.5.
k
On the manifold RP2 by Example 3.3.1 we have
k +1
w = (1 + x)2
k
= (1 + x)(1 + x)2
k
= (1 + x)(1 + x2 )
and so
1 1
=
w (1 + x)(1 + x2k )
k k
= (1 + x + x2 + x3 + · · · )(1 + x2 + x2·2 + · · · )
k −1 k k +1
= 1 + x + x2 + · · · + x2 ∈ H ∗ (RP2 ; Z/2) = Z/2[x]/(x2 ).
Thus RP2
k
does not embed into R2
k+1 −1 1
. 4
3.4 Constructing the Steenrod squares

Let us write Kn = K(Z/2, n) in this section, and take (co)homology with Z/2 coecients.
Recall that
n
there is a ιn ∈ H (Kn ) such that
∼
[X, Kn ] −→ H n (X)
f 7−→ f ∗ (ιn )
is a bijection for any CW-complex X. We are only interested in n > 0, so ιn may

be represented by a reduced cohomology class ιn ∈ H n (Kn , ∗). If (X, x0 ) is a based
CW-complex, the same argument shows that
∼
e n (X)
[X, Kn ]∗ −→ H
f 7−→ f ∗ (ιn ),
Lecture 22 where [−, −]∗ denotes the homotopy classes of based maps.
Suppose that we are given a class e n+i (Kn ).
θιn ∈ H Then for any based CW-complex
we have an operation
e n (X) −→ H
θ:H e n+i (X)
f ∗ (ιn ) 7−→ f ∗ (θιn )
which by construction is natural for maps of based CW-complexes. We can promote it

to an operation on the cohomology of all based spaces via CW approximation.
Question: Under what conditions on the class θιn is the function θ a homomorphism?
To answer this question, note that e n (Kn × Kn )

ιn ⊗ 1 + 1 ⊗ ιn ∈ H corresponds to a
based map
µn : Kn × Kn −→ Kn ,
well-dened up to homotopy.
Lemma 3.4.1. This map satises
(i) µn (∗, −) ' Id ' µn (−, ∗) : Kn → Kn ,
(ii) µn (µn (−, −), −) ' µn (−, µn (−, −)) : Kn × Kn × Kn → Kn .
Proof. We are claiming that certain maps to Kn are homotopic, which is the case if they
pull back ιn to the same class.
For (i) the composition
Id×inc µn
Kn = Kn × {∗} −→ Kn × Kn −→ Kn
1
It is a theorem of H. Whitney than any smooth n-manifold may be smoothly embedded in R2n .
3.4 Constructing the Steenrod squares 75
pulls back ιn to
(Id × inc)∗ (µn )∗ (ιn ) = (Id × inc)∗ (ιn ⊗ 1 + 1 ⊗ ιn ) = ιn
so this composition is homotopic to the identity; similarly for µn (∗, −).

For (ii), µn (µn (−, −), −) is the composition
µn ×Id µn
Kn × Kn × Kn −→ Kn × Kn −→ Kn
so it pulls back ιn to
(µn × Id)∗ (µn )∗ (ιn ) = (µn × Id)∗ (ιn ⊗ 1 + 1 ⊗ ιn ) = (ιn ⊗ 1 + 1 ⊗ ιn ) ⊗ 1 + 1 ⊗ ιn .
The map µn (−, µn (−, −)) does too.
Corollary 3.4.2. The operation θ is a homomorphism if and only if
µ∗n (θιn ) = θιn ⊗ 1 + 1 ⊗ θιn ∈ H

e n+i (Kn × Kn ). (3.4.1)
Proof. The function θ is dened by naturality and θ(ιn ) = θιn , so if it is a homomorphism

then
µ∗n (θιn ) = µ∗n (θ(ιn )) = θ(µ∗n (ιn )) = θ(ιn ⊗ 1 + 1 ⊗ ιn )

= θ(π1∗ (ιn ) + π2∗ (ιn ))
= θ(π1∗ (ιn )) + θ(π2∗ (ιn ))
= π1∗ (θιn ) + π2∗ (θιn )
= θιn ⊗ 1 + 1 ⊗ θιn .
Conversely, if f ∗ (ιn ), g ∗ (ιn ) ∈ H

e n (X) then the composition
f ×g µn
X −→ Kn × Kn −→ Kn
pulls back ιn to
(f × g)∗ (µn )∗ (ιn ) = (f × g)∗ (ιn ⊗ 1 + 1 ⊗ ιn ) = f ∗ (ιn ) + g ∗ (ιn ).
Thus assuming (3.4.1) holds we have
θ(f ∗ (ιn ) + g ∗ (ιn )) := (f × g)∗ (µn )∗ (θιn )

= (f × g)∗ (θιn ⊗ 1 + 1 ⊗ θιn )
= f ∗ (θιn ) + g ∗ (θιn )
= θ(f ∗ (ιn )) + θ(g ∗ (ιn ))
as required.
Denition 3.4.3. Say that e ∗ (Kn )

x∈H is primitive if µ∗n (x) = x ⊗ 1 + 1 ⊗ x.
Lemma 3.4.4. Under the identication ΩKn+1 ' Kn , Ωµn+1 ' µn .
Proof. Let
π
Kn ' ΩKn+1 −→ P∗ Kn+1 −→ Kn+1
be the path bration. Then P∗ (Kn+1 × Kn+1 ) = P∗ (Kn+1 ) × P∗ (Kn+1 ) and there is a
map of brations
Ωµn+1
ΩKn+1 × ΩKn+1 ΩKn+1
P∗ µn+1
P∗ (Kn+1 × Kn+1 ) P∗ Kn+1
π×π π
µn+1
Kn+1 × Kn+1 Kn+1
In the Serre spectral sequence for the bration π, the class
ιn ∈ H (Kn ) ∼
n n 2
= H (ΩKn+1 ) = E0,n
transgresses to ιn+1 ∈ H n+1 (Kn+1 ) = En+1,0
2 . Thus there is a commutative square
dn+1
H n (Kn ) H n+1 (Kn+1 )
(Ωµn+1 )∗ µ∗n+1
d¯n+1
H n (Kn × Kn ) H n+1 (Kn+1 × Kn+1 )
where the horizontal maps are isomorphisms, and d¯n+1 is the dierential in the Serre
r
spectral sequence {Ē∗,∗ } for the bration π × π. We have
µ∗n+1 dn+1 (ιn ) = µ∗n+1 (ιn+1 ) = ιn+1 ⊗ 1 + 1 ⊗ ιn+1

and
ιn+1 ⊗ 1 + 1 ⊗ ιn+1 = d¯n+1 (ιn ⊗ 1 + 1 ⊗ ιn ),
so (Ωµn+1 )∗ (ιn ) = ιn ⊗ 1 + 1 ⊗ ιn , so Ωµn+1 ' µn as this is its dening property.
Theorem 3.4.5. If θιn ∈ H i+n (Kn ) is primitive, transgresses in the Serre spectral
sequence for
π
Kn ' ΩKn+1 −→ P∗ Kn+1 −→ Kn+1 ,
and i ≤ n, then it transgresses to a unique class, called θιn+1 ∈ H n+1+i (Kn+1 ), which
is also primitive.
Proof. From the Serre spectral sequence for π as shown in Figure 3.3 we see that the
lowest degree in which transgressions are not unique is 2n + 2. As n + i + 1 ≤ 2n + 1 by
assumption, the transgression of θιn is indeed unique.
To see that θιn+1 is primitive, we consider the map of Serre spectral sequences for
the map of brations in the previous lemma. We nd a commutative diagram
n+i+1 dn+i+1
H n+i (Kn ) ⊃ E0,n+i H n+i+1 (Kn+1 )
µ∗n µ∗n+1
n+i+1 d¯n+i+1
H n+i (Kn × Kn ) ⊃ Ēn+i+1,0 H n+i+1 (Kn+1 × Kn+1 ).
29
¥:
,
¥tdznfK
nti • ① in
in • . -
.
D htt Zntz
Figure 3.3 The Serre spectral sequence for the path bration over Kn+1 with Z/2-coecients.
30
it:±
,
We have
i
he
d¯n+i+1 µ∗n (θιn ) = d¯n+i+1 (θιn ⊗ 1 + 1 ⊗ θιn ) = θιn+1 ⊗ 1 + 1 ⊗ θιn+1
photog ,dgree
and dn+i+1 θιn = θιn+1 , so
µ∗n+1 (θιn+1 ) = µ∗n+1 (dn+i+1 θιn ) =, θιn+1 ⊗ 1 + 1 ⊗ θιn+1

as required. .
Lecture 23
Corollary 3.4.6. In the situation above, if θ: e n (−)

H → e n+i (−) is the function dened
H
by θιn , and e n+1 (−) → H
θ:H e n+i+1 (−) is the function dened by θιn+1 , then the square
e n (X) σ e n+1 (ΣX)

H H
θ θ .
e n+i (X) σ e n+i+1 (ΣX)

D H htt
H Zntz
commutes.
Proof. By naturality, it suces to show it commutes when X = Kn on the class ιn .

Choosing a nullhomotopy of the inclusion Kn = ΩKn+1 → P∗ Kn+1 gives maps of pairs
'
(CKn , Kn ) (P∗ Kn+1 , Kn )
π
f
(ΣKn , ∗) (Kn+1 , ∗)
and because π ∗ : H n+1 (Kn+1 , ∗) → H n+1 (P∗ Kn+1 , Kn ) is an isomorphism, as it is part

of the transgression of ιn to ιn+1 , it follows that
f ∗ : H n+1 (Kn+1 , ∗) −→ H n+1 (ΣKn , ∗)
is an isomorphism too, and so f represents the cohomology class e n+1 (ΣKn ).

σιn ∈ H The
denition of transgression gives the diagram
' δ
H n+i+1 (CKn , Kn ) H n+i+1 (P∗ Kn+1 , Kn ) e n+i (Kn )
H
' π∗
f∗
e n+i+1 (ΣKn )
H e n+i+1 (Kn+1 )
H
in which the square and outer boundary commutes. We have δ(θιn ) = π ∗ (θιn+1 ) by
denition of transgression, and so σ(θιn ) = f ∗ (θιn+1 ) = θ(σιn ) as required.
Corollary 3.4.7. In the situation above, if we dene
θ : H n (X, A) −→ H n+i (X, A)

∼
e ∗ (X/A) →
for a CW-pair (X, A) using H H ∗ (X, A), then the square
δ
H n (A) H n+1 (X, A)
θ θ
δ
H n+i (A) H n+i+1 (X, A)
commutes.
Proof. We can extend the inclusion A → CA ' ∗ to a map f : X → CA, and hence get
a map of pairs f : (X, A) → (CA, A), and an induced map fˆ : X/A → CA/A = ΣA. By
naturality the diagram
δ '
H n (A) H n+1 (X, A) e n+1 (X/A)
H
δ
f∗ fˆ∗
'
H n+1 (CA, A) e n+1 (ΣA)
H
σ
commutes, so as θ commutes with σ and with fˆ∗ the claim follows.

So far the discussion has been completely general; we now construct the Sqi .
Theorem 3.4.8. There are natural homomorphisms Sqi of degree i such that
(i) Sq0 = Id,

(ii) Sqi (x) = x2 if i = |x|,
(iii) Sqi (x) = 0 if i > |x|,
(iv) δ ◦ Sqi = Sqi ◦ δ for the connecting map δ : H n (A) → H n+1 (X, A),
(v) σ ◦ Sqi = Sqi ◦ σ for the suspension isomorphism e n (X) → H
σ:H e n+1 (ΣX).
Proof. Dene Sq0 = Id. For n > 0, dene Sqn ιn = ι2n ∈ H 2n (Kn ). Now
µ∗n (ι2n ) = (µn (ιn ))2 = (ιn ⊗ 1 + 1 ⊗ ιn )2 = ι2n ⊗ 1 + 1 ⊗ ι2n

so Sqn ιn is primitive. In the Serre spectral sequence for
Kn ' ΩKn+1 −→ P∗ Kn+1 −→ Kn+1

we have dn+1 (ι2n ) = 2ιn ⊗ ιn+1 = 0. Thus ι2n is transgressive: by Theorem 3.4.5 it
n 2n+1 (K
transgresses to a unique element Sq ιn+1 ∈ H n+1 ) which is again primitive.
Now consider the Serre spectral sequence for
Kn+1 ' ΩKn+2 −→ P∗ Kn+2 −→ Kn+2 .

The class not lying on an edge of lowest total degree is ιn+2 ⊗ ιn+1 of degree
2n + 3, so all
classes of degree ≤ 2n + 1 Sqn ιn+1 does, so by Theorem 3.4.5
transgress. In particular
n 2n+2 (K
it transgresses to a unique element Sq ιn+2 ∈ H n+2 ) which is again primitive.
n+k
Continuing in this way we dene Sq ιn for all k ≥ 0.
n n
Dening Sq ιj = 0 for j < n, we obtain an operation Sq dened on cohomology
classes of every degree, which satisfy all the required properties, except possibly
e n−1 (X) σ e n (ΣX)

H H
Sqn =0 Sqn =square
e 2n−1 (X) σ e 2n (ΣX)

H H
commuting: but this does in fact commute, as cup products vanish on any suspension
(the analogous naturality for δ follows from this as in Corollary 3.4.7).
To study how the Sqi interact with cup products, we must rst discuss products of
based spaces. If (X, x0 ) and (Y, y0 ) are based spaces, their smash product is
X ×Y
X ∧ Y := .
X × {y0 } ∪ {x0 } × Y
Note that ΣX = S 1 ∧ X . The usual Künneth theorem shows that
−^−
− ∧ − : H ∗ (X, x0 ) ⊗ H ∗ (Y, y0 ) −→ H ∗ (X × Y, X × {y0 } ∪ {x0 } × Y ) = H
e ∗ (X ∧ Y )
is an isomorphism (as we are working with coecients in the eld Z/2). Lecture 24
29
¥:
,
¥tdznfK
nti • ① in
Lemma 3.4.9. The map σιn : ΣKn → Kn+1 is injective on cohomology in degrees
∗ ≤ 2n + 1. in • .
. -
Proof. Return to the diagram
' δ
H n+i+1 (CKn , Kn ) H n+i+1 (P∗ Kn+1 , Kn ) e n+i (Kn )
H
' π∗
transgresson
(σιn )∗
e n+i+1 (ΣKn )
H e n+i+1 (Kn+1 )
H
D htt Zntz
which shows that (σιn )∗ is injective in the range of degrees in which the transgression is
single-valued. From the Serre spectral sequence
30
it:±
,
he
photog ,dgree
,
D htt Zntz
this is seen to be degrees ∗ ≤ 2n + 1.
Consider the maps
σιn−1 ∧Id
σL : ΣKn−1 ∧ Km −→ Kn ∧ Km
Id∧σιm−1
σR : Kn ∧ ΣKm−1 −→ Kn ∧ Km .
It follows from the previous lemma that the maps

M
σL∗ : e i (Kn ) ⊗ H
H e ∗ (ΣKn−1 ) ⊗ H
e j (Km ) −→ H e ∗ (Km )
i≤2n−1
j
M
∗ e i (Kn ) ⊗ H e ∗ (Kn ) ⊗ H
e j (Km ) −→ H e ∗ (ΣKm−1 )
σR : H
i
j≤2m−1
3.5 Outlook 81
are both injective, so in total degrees ∗ ≤ 2(n + m) − 1 the map
σL∗ ⊕ σR
∗ e ∗ (Kn ) ⊗ H
:H e ∗ (Km ) −→ (H
e ∗ (ΣKn−1 ) ⊗ H
e ∗ (Km )) ⊕ (H
e ∗ (Kn ) ⊗ H
e ∗ (ΣKm−1 ))
is injective.
Theorem 3.4.10. We have

X
Sqk (ιn ∧ ιm ) = e ∗ (Kn ∧ Km ).
Sqi (ιn ) ∧ Sqj (ιm ) ∈ H
i+j=k
Proof. If k =n+m this becomes the identity
(ιn ∧ ιm )2 = ι2n ∧ ι2m
which is certainly true. So we proceed by induction on the quantity n + m − k.

Let z∈H e n+m+k (Kn ∧Km ) be the dierence of the two terms. Under σ ∗ this becomes
L
X
σL∗ (z) = Sqk (σιn−1 ∧ ιm ) + Sqi (σιn ) ∧ Sqk (ιm ) ∈ H
e n+m+k (ΣKn−1 ∧ Km )
i+j=k
which under the suspension isomorphism is

X
Sqk (ιn−1 ∧ ιm ) + Sqi (ιn ) ∧ Sqk (ιm ) ∈ H
e (n−1)+m+k (Kn−1 ∧ Km ).
i+j=k
As (n − 1) + m − k < n + m − k this vanishes by induction. Similarly

∗ (z) = 0,
σR so
z = 0.
The nal property of Steenrod squares, that Sq1 agrees with Bockstein operation, is
Example Sheet 4 Q2.
3.5 Outlook
The Steenrod squares satisfy the Adem relations: if 0 < i < 2j then the identity
i/2
i
X
j j−k−1
Sq Sq = Sqi+j−k Sqk
i − 2k
k=0
a
holds. It is not hard to show using this that if a is not a power
of 2 then Sq is
j−1
decomposable (as we can write a = i+j with 0 < i < 2j and i ≡ 1 mod 2): the
rst few are
Sq3 = Sq1 Sq2

Sq5 = Sq1 Sq4
Sq6 = Sq2 Sq4 + Sq5 Sq1 = Sq2 Sq4 + Sq1 Sq4 Sq1
Sq7 = Sq1 Sq6 = Sq1 Sq2 Sq4 (using Sq1 Sq1 = 0)
Sq9 = Sq1 Sq8 .
Example 3.5.1. Suppose X is a space having H ∗ (X; Z/2) = Z/2[x]/(x3 ) with |x| = n
(so X is analogous to RP , CP2 , HP2 ). Then
2
0 6= x2 = Sqn (x).
If n is not a power of 2 then Sqn is decomposable, but Sqi (x) = 0 for all 0 < i < n
because the group it lies in is zero. Thus n must be a power of 2. 4
i
In fact, J. F. Adams has showed that Sq2 is decomposable in terms of higher-order
operations if i ≥ 4, which implies that in the above example we must have n = 1, 2, 4, 8;
the last example is provided by the octonionic projective plane OP2 . This is the famous
Hopf Invariant 1 Theorem.
SqI := Sqi1 Sqi2 · · · Sqir in Steenrod squares, if
In a dierent direction, given a word
ij < 2ij+1 then we can apply an Adem relation to write Sqij Sqij+1 as a linear combination
a b
of Sq Sq 's with a ≥ 2b. Iterating this, we can write any word in the Steenrod squares
as a linear combination of Sq 's with ij ≥ 2ij+1 : such an Sq is called admissible.
I I
Theorem 3.5.2. H ∗ (K(Z/2, n); Z/2) is a polynomial ring over Z/2 on the classes SqI ιn
such that
(i) SqI is admissible, and
excess e(I) :=
Pr
(ii) I = (i1 , i2 , . . . , ir ) has j=1 (ij − 2ij+1 ) < n.
Example 3.5.3. Only I = (0) has e(I) = 0. Only the admissible sequences
I = (1), (2, 1), (4, 2, 1), (8, 4, 2, 1), . . .
have e(I) = 1. Thus
H ∗ (K(Z/2, 2); Z/2) = Z/2[ι2 , Sq1 ι2 , Sq2 Sq1 ι2 , Sq4 Sq2 Sq1 ι2 , . . .].
The proof of this theorem does not go beyond the methods of this course: it holds
for n = 1 by observation, and can then be proved inductively using Kudo's transgression
theorem and some spectral sequence yoga: in particular it does not use the Adem rela-
tions. To prove the Adem relations one can proceed as follows. As the Steenrod squares
commute with the suspension isomorphism, it suces to check the Adem relation for
Sqi Sqj on cohomology classes of degree n ≥ i + j. From the theorem it is easy to check
that the map
ι1 × · · · × ι1 : K(Z/2, 1) × · · · × K(Z/2, 1) −→ K(Z/2, n)

| {z }
n times
is injective on cohomology in degrees ≤ 2n, and from this it is a matter of algebra to

verify that the Adem relation for Sqi Sqj ιn holds, as n + i + j ≤ 2n and we know how the
Steenrod squares act on
∗ H (K(Z/2, 1); Z/2) = Z/2[ι1 ]. (Of course one has to get quite
good at binomial coecients modulo 2.)

Homotopy Theory [RW]

Uploaded by

Copyright:

Available Formats

Homotopy Theory [RW]

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

Homotopy Theory [RW]

Uploaded by

Copyright:

Available Formats

Homotopy Theory

1 Homotopy groups, CW complexes, and brations 1

2 The Serre spectral sequence and applications 38

Last updated March 30, 2021. Corrections to or257@cam.ac.uk.

Homotopy groups, CW complexes, and brations

∂I n = {(t1 , t2 , . . . , tn ) ∈ I n | some ti is 0 or 1}.

1.1 Homotopy groups

Let (X, x0 ) be a based space, i.e. a space with a distinguished point x0 ∈ X .

(i) F (∂I n × [0, 1]) ⊂ {x0 },

(ii) F (−, 0) = f0 (−), and F (−, 1) = f1 (−).

πn (X, x0 ) = {homotopy classes of maps f : (I n , ∂I n ) → (X, {x0 })}.

to swap the order of operations.

If ϕ : (X, x0 ) → (Y, y0 ) is a based map then

we get a map of pairs x.tt#iiEHiiiFttEH

given graphically as shown in Figure 1.2 below: that is, apply f

to a smaller cube, and

(i) u# = u0# if u is homotopic to u0 relative to their end-points,

(ii) (constx0 )# is the identity,

(iii) u# is a group homomorphism,

(iv) if v : I → X is a path from x1 to x2 and u·v is the concatenation of these paths,

These properties all follow from fairly obvious homotopies.

1.3 Relative homotopy groups

f : I n −→ X such that f (∂I n ) ⊂ A and f (un−1 ) ⊂ {x0 }

f : (I n , ∂I n , un−1 ) −→ (X, A, {x0 }).

n ≥ 2 the formula (1.1.1) denes a group structure on πn (X, A, x0 ), with identity

' ϕ1 through such maps then (ϕ0 )∗ = (ϕ1 )∗ ,

(Cubes and discs) . As

We will use these points of view interchangeably.

° 1.4 The Hurewicz homomorphism

We know that Hn (Dn , S n−1 ; Z) is isomorphic to Z, with generator un . As homotopic

6 Chapter 1 Homotopy groups, CW complexes, and brations

1.5 Compression criterion

7. The constant map constx0 : (I n , ∂I n , un−1 ) → (X, A, x0 )

Lemma 1.5.1. A mapf : (Dn , S n−1 , ∗) → (X, A, x0 ) represents [constx0 ] ∈ πn (X, A, x0 )

by radial projection from (0, −1)

1.6 The long exact sequence of a pair

In addition, given a map f : (I n , ∂I n , un−1 ) → (X, A, x0 ) its restriction to I n−1 × {0} ⊂

∂ : πn (X, A, x0 ) −→ πn−1 (A, x0 ).

If n≥2 then ∂(f · g) = ∂f · ∂g , so this function is a homomorphism in this case.

Theorem 1.6.1. The sequence

··· πn+1 (X, A, x0 )

with image in A, so [f ] ∈ Im(i∗ ). The compression criterion also shows that j∗ ◦ i∗ = 0.

Exactness at πn (X, A, x0 ): We have ∂j∗ ([f ]) = [f |I n ×{0} ] = [constx0 ] because f sends

These indeed glue as H(t2 , . . . , tn ; 0) = (∂f )(t2 , . . . , tn ) = f (0, t2 , . . . tn ), and g indeed

as the homotopy H has image in A. See Figure 1.4 (b).

Exactness at πn−1 (A, x0 ): For f : (I n , ∂I n , un−1 ) → (X, A, x0 ), the element i∗ ∂([f ]) is

is homotopic to constx0 via the homotopy f ! Conversely, if g : (I n−1 , ∂I n−1 ) → (A, x0 )

H|I n−1 ×{0} = g, H(∂I n ) ⊂ A, H(un−1 ) ⊂ {x0 },

and so [H] ∈ πn (X, A, x0 ) satises ∂([H]) = [g].

For a path u : [0, 1] → A, the change-of-basepoint bijections

u# : πn (X, x0 ) −→ πn (X, x1 ) and u# : πn (A, x0 ) −→ πn (A, x1 )

1.7 Maps are pairs (up to homotopy)

Mf := (X × [0, 1] t Y )/(x, 1) ∈ X × [0, 1] ∼ f (x) ∈ Y,

called the mapping cylinder of f .

and linear interpolation gives a deformation retraction

X n := (X n−1 t (In × Dn ))/(α, x) ∈ In × S n−1 ∼ ϕα (x) ∈ X n−1 ,

with the quotient topology.

A pair of spaces (X, A) is a relative CW complex if X is obtained as above but

If X is a CW complex then a subcomplex is a closed subspace A ⊂ X which is a

f ∪ H : (X × {0}) ∪ (A × [0, 1]) −→ Y

1 Homotopy groups, CW complexes, and brations 1

Homotopy groups, CW complexes, and brations

n ≥ 2 the formula (1.1.1) denes a group structure on πn (X, A, x0 ), with identity

6 Chapter 1 Homotopy groups, CW complexes, and brations

and so [H] ∈ πn (X, A, x0 ) satises ∂([H]) = [g].

and we can set H n := (f |X n ∪ H n−1 ) ◦ rn . By denition of the topology on X , the map

Proof. Suppose rst that f :X →Y is the inclusion of a subspace, so that (Y, X) is a

X = J 0 ⊂J nite ( J 0 S n ), and as any map from a compact

of nitely-many (n + 1)-cells, we may suppose without loss of generality that C

nitely-many (n + 1)-cells: we proceed by induction on the number of (n + 1)-cells. By

1.14 EilenbergMac Lane spaces and cohomology

(i) S1 is EilenbergMac Lane space of type (Z, 1), and

(ii) RP∞ is an EilenbergMac Lane space of type (Z/2, 1). 4

Proof. Suppose rst that n≥2 and G is abelian. Let

i > n: it is then the required EilenbergMac Lane space.

Let (X, x0 ) be an EilenbergMac Lane space of type (G, n). If n ≥ 2, or more