0.

1 Homotopic maps induce the same map in cohomology

Let f , g : M → N be smooth maps between smooth manifolds. A (smooth) homotopy between f

and g is a smooth map F : M × [0, 1] → N such that
½
F (p, 0) = f (p)
F (p, 1) = g(p)

for p ∈ M . If there exists a homotopy between f and g, we say that they f and g are homotopic.

0.1.1 Proposition Let f , g be homotopic maps. Then the induced maps in cohomology
k k
f ∗ , g ∗ : HdR (N ) → HdR (M )

are equal.
The proof of this propositon is given below. First, we need to make some remarks. For t ∈ [0, 1],
consider the inclusions it given by
it (p) = (p, t)
for p ∈ M , and consider the natural projection π : M × [0, 1] → M given by π(p, t) = p. Then,
obviously,
π ◦ it = idM
implying that
i∗t π ∗ = id in Ωk (M ) and HdR
k (M ).

∂
We consider the projection t : M × [0, 1] → [0, 1]. Then there exists a “vertical” vector field ∂t
∂
and a 1-form dt. Note that ker dπ is spanned by ∂t .

0.1.2 Lemma Let ω ∈ Ωk (M × [0, 1]). Then we can write

(0.1.3) ω = ζ + dt ∧ η

where ζ ∈ Ωk (M ×[0, 1]) has the property that it vanishes if some of its arguments belongs to ker dπ,
and η ∈ Ωk−1 (M × [0, 1]) has the same property.
Proof. Set η = i ∂ ω and ζ = ω − dt ∧ η. Since
∂t

i ∂ η = i ∂ i ∂ ω = 0,
∂t ∂t ∂t

it is clear that η has the claimed property. Similarly,

i ∂ ζ = i ∂ ω − i ∂ (dt ∧ η)
∂t ∂t ∂t
= η − i ∂ dt ∧ η + dt ∧ i ∂ η
∂t ∂t
= η−η+0
= 0,

as desired. ¤
We define the homotopy operator

Hk : Ωk (M × [0, 1]) → Ωk−1 (M )

1
by the formula Z 1
(Hk ω)p (v1 , . . . , vk−1 ) = η(p,t) (dit (v1 ), . . . , dii (vk−1 )) dt,
0
where ω is decomposed as in (0.1.3) and p ∈ M , v1 , . . . , vk−1 ∈ Tp M . Note that Hk is “integration
along the fiber of π”. For simplicity, we henceforth drop the subscript and just write H for the
homotopy operator.
k (M × [0, 1]). We first claim that
Proof of Propostion 0.1.1. Let ω ∈ HdR

(0.1.4) dHω + Hdω = i∗1 ω − i∗0 ω.

The proof is by direct computation: since this is a pointwise identity, we can work in a coordinate
system. Let (U, x1 , . . . , xn ) be a coordinate system in M . Then (U × [0, 1], x1 ◦ π, . . . , xn ◦ π, t) is
a coordinate system in M × [0, 1] and we can write
X X
ω|U ×[0,1] = aI dxI + dt ∧ bJ dxJ
I J

where ai , bJ are smooth functions on U × [0, 1] and I, J are increasing multi-indices. In U × [0, 1],
we have:
X µZ 1 ¶
Hω = bJ dt dxJ ,
J 0

X µZ 1
∂bJ
¶
dHω = dt dxi ∧ dxJ ,
0 ∂xi
J,i
X ∂aI X ∂aI X ∂bJ
dω = dxi ∧ dxI + dt ∧ dxI − dt ∧ dxi ∧ dxJ ,
∂xi ∂t ∂xi
I,i I J,i

X µZ 1
∂aI
¶ X µZ 1 ∂bJ ¶
Hdω = dt dxI − dt dxi ∧ dxJ .
0 ∂t 0 ∂xi
I J,i

It follows that
X µZ 1
∂aI
¶
dHω + Hdω|p = (p, t) dt dxI
0 ∂t
I
X
= (aI (p, 1) − aI (p, 0))dxI
I
∗
= i1 ω − i∗0 ω|p ,

as claimed.
Suppose now that F : M × [0, 1] → N is a homotopy between f and g. Let α be a closed k-form
K (N ). Applying identity (0.1.4) to ω = F ∗ α yields
in N representing the cohomology class [α] ∈ HdR

dHF ∗ α + HF ∗ dα = i∗1 F ∗ α − i∗0 F ∗ α.

Since dα = 0 and F ◦ i0 = f , F ◦ i1 = g, we get

d (HF ∗ α) = g ∗ α − f ∗ α.

Hence g ∗ α and f ∗ α are cohomologous. ¤

2
0.2 Hairy ball theorem

Consider Euclidean space Rn+1 with coordinates (x0 , x1 , . . . , xn ) and the unit sphere ι : S n → Rn+1 .
Consider the n-form in Rn+1
n
X
ω= ci ∧ · · · dxn .
(−1)i xi dx0 ∧ · · · ∧ dx
i=0

Note that ω vanishes only at the origin. In particular,

n
X
(0.2.1) α = ι∗ ω = \
(−1)i (xi ◦ ι) d(x0 ◦ ι) ∧ · · · d(xi ◦ ι) ∧ · · · d(xn ◦ ι)
i=0

is a nowhere vanishing n-form on S n , hence it defines an orientation there. Of course,

dω = (n + 1)dx0 ∧ dx1 ∧ · · · ∧ dxn

and Z Z Z
α= ω= dω = (n + 1)vol(B̄ n ) > 0
Sn Sn B̄ n

(where the orientation of S n is induced from B̄ n ) so α is not exact by Stokes theorem. Thus [α] 6= 0
n (S n ).
in HdR
In the sequel, we consider n = 2m.

0.2.2 Theorem Let X be a smooth vector field on S 2m . Then there exists p ∈ S 2m such that
Xp = 0. In other words, every smooth vector field on an even-dimensional sphere has a zero.

Proof. Suppose, on the contrary, that X never vanishes. By rescaling, we may assume that X
is a unit vector field with respect to the metric induced from Euclidean space. Set

Ft : S 2m → S 2m , Ft (p) = cos t x + sin t X(p).

It is clear that Ft defines a homotopy between the identity map and the antipodal map of S 2m :

F0 = idS 2m and Fπ = −idS 2m .

Note that
Fπ∗ (xi ◦ ι) = −xi ◦ ι.
It follows that
Fπ∗ α = (−1)2m+1 α = −α.
On the other hand,
F0∗ α = α,
and by Proposition 0.1.1, [F0∗ α] = [Fπ∗ α], which contradicts the fact that [α] 6= 0. ¤

0.2.3 Remark The theorem can be extended to the case of continuous vector fields by using an
approximation result.

3
0.3 The smooth Brouwer fixed point theorem

Let B̄ n be the closed ball in Rn , and denote its boundary by ∂ B̄ n ; of course, ∂ B̄ n is diffeomorphic
to S n−1 . We first prove

0.3.1 Lemma There exists no smooth retraction r : B̄ n → ∂ B̄ n (that is, there exists no smooth
map r : B̄ n → ∂ B̄ n whose restriction to ∂ B̄ n is the identity).
Proof. The case n = 1 is easy as the closed interval B̄ 1 is connected and its boundary is
disconnected. Assume n ≥ 2 and suppose, to the contrary, that such a retraction r exists. Recall
that n-form α defined in (0.2.1). Since r is the identity along ∂ B̄ n ,
Z Z
∗
r α= α 6= 0.
∂ B̄ n ∂ B̄ n

On the other hand, by Stokes theorem,

Z Z Z
∗ ∗
r α= dr α = r∗ dα = 0,
∂ B̄ n B̄ n B̄ n

since dα = 0, which is a contradiction. ¤

0.3.2 Theorem Let f : B̄ n → B̄ n be a smooth map. Then there exists p ∈ B̄ n such that f (p) = p.
In other words, every smooth self-map of the closed n-ball admits a fixed point.
Proof. Suppose, on the contrary, that f (x) 6= x for all x ∈ B̄ n . The half-line originating at x
and going through f (x) meets ∂ B̄ n at a unique point; call it r(x). It is easy to see that this defines
a smooth retraction r : B̄ n → ∂ B̄ n which is prohibited by Lemma 0.3.1. ¤

0.3.3 Remark The theorem is not true in the case of the open ball B n , as is easily seen.