M4P54 Lecture Notes

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 66

DIFFERENTIAL TOPOLOGY

Contents
1. Differential Forms on manifolds 1
1.1. Alternating p-forms on a vector space 1
1.2. Differential forms on a manifold 3
1.3. Local description of p-forms 4
1.4. Integrations on manifolds 6
1.5. Orientation 7
1.6. Partitions of unity 8
1.7. Manifolds with boundary 11
1.8. Stokes’ Theorem 12
2. de Rham Cohomology 15
2.1. Homotopy invariance 18
2.2. Some homological algebra 22
2.3. The Mayer-Vietoris sequence 24
2.4. Complactly supported de Rham cohomology 28
2.5. Poincaré duality 32
2.6. Degree of a morphism 38
3. Morse Theory 40
3.1. Introduction 40
3.2. CW complexes 43
3.3. CW-structure associated to a Morse function 44
3.4. Morse homology 55
4. Singular homology 61
4.1. de Rham homomorphism 65
References 66

1. Differential Forms on manifolds


1.1. Alternating p-forms on a vector space. We begin by recalling some linear alge-
bra. Let V be a vector space over R and, for any integer p ≥ 0, denote V p = V × · · · × V .
Definition 1.1. A multilinear map ω : V p → R is called an alternating p-form if for
all v1 , . . . , vp ∈ V and for all σ ∈ Sp , we have

ω vσ(1) , . . . , vσ(p) = (σ)ω (v1 , . . . , vp )
1
2 DIFFERENTIAL TOPOLOGY

where (σ) is the signature of σ, defined by


(σ) = (−1)m ,
where m is the number of transpositions in a decomposition of σ (note that m is not
uniquely defined, but (−1)m only depends on σ).
Notation 1.2. We denote
Λp V ∗ := {ω : V p → R|ω is an alternating p -form. } .
Note that Λp V ∗ is a vector space, which is called p-th exterior power of V .
Example 1.3.
• Λ0 V ∗ = R.
• Λ1 V ∗ = V ∗ = Hom(V, R) is the dual of V .
Definition 1.4. Given ω1 ∈ Λp V ∗ and ω2 ∈ Λq V ∗ , we can define the exterior product
ω1 ∧ ω2 ∈ Λp+q V ∗ by, for each v1 , . . . , vp+q ∈ V ,
X  
ω1 ∧ ω2 (v1 , . . . , vp+q ) := (σ)ω1 vσ(1) , . . . , vσ(p) ω2 vσ(p+1) , . . . , vσ(p+q)
σ∈Sp,q

where
Sp,q := {σ ∈ Sp+q |σ(1) < · · · < σ(p), σ(p + 1) < · · · < σ(p + q)} .
Example 1.5.
• Assume ω1 , ω2 ∈ Λ1 V ∗ . Then
ω1 ∧ ω2 (v, w) = ω1 (v)ω2 (w) − ω1 (w)ω2 (v).
• More in general, if ω1 , . . . , ωp ∈ Λ1 V ∗ and v1 , . . . , vp ∈ V , then
ω1 (v1 ) ω1 (v2 ) . . . ω1 (vp )
 
 ω2 (v1 ) ω2 (v2 ) . . . ω2 (vp ) 
ω1 ∧ · · · ∧ ωp (v1 , . . . vp ) = det  .. .. .. .
 . . ... . 
ωp (v1 ) ωp (v2 ) . . . ωp (vp )

Proposition 1.6. Let ωi ∈ Λpi V ∗ , for i = 1, . . . , 3. Then the following properties hold:
• (Associativity) ω1 ∧ (ω2 ∧ ω3 ) = (ω1 ∧ ω2 ) ∧ ω3 .
• (Distributivity) if p2 = p3 then ω1 ∧ (ω2 + ω3 ) = ω1 ∧ ω2 + ω1 ∧ ω3 .
• (Super-commutativity) ω1 ∧ ω2 = (−1)p1 ·p2 ω2 ∧ ω1 .
Definition 1.7. Let Φ : V → W be a linear morphism between vector spaces. Let ω ∈
Λp W ∗ .
Then the pull-back Φ∗ ω ∈ Λp V ∗ of ω is defined by, for all v1 , . . . , vp ∈ V ,
(Φ∗ ω) (v1 , . . . vp ) := ω (Φv1 , . . . Φvk ) .
DIFFERENTIAL TOPOLOGY 3

Proposition 1.8. Let Φ : V → W and Ψ : W → Z be linear maps between vector spaces.


Then the following properties hold:
• the pull-back Φ∗ : Λp W ∗ → Λp V ∗ , ω 7→ Φ∗ ω is linear and it preserves the exterior
product, i.e.
Φ∗ (ω1 ∧ ω2 ) = Φ∗ ω1 ∧ Φ∗ ω2 .
• (Ψ ◦ Φ)∗ ω = Φ∗ Ψ∗ ω.
• Assume V = W and p = dim V . Then
Φ∗ ω = det(Φ)ω.
1.2. Differential forms on a manifold. We now assume that M is a (smooth) manifold
of dimension n. For all x ∈ M , the tangent space Tx M is a vector space of dimension n.
Notation 1.9. We denote
Λp Tx∗ M := Λp (Tx M )∗ .
The set a
Λp T ∗ M := Λp Tx∗ M
x∈M
is called the p-th exterior bundle of M . Note that there exists a natural map
π : Λp T ∗ M → M
such that, for all x ∈ M , the fibre π −1 (x) is the vector space Λp Tx∗ M . It can be checked
that Λp T ∗ M is a vector bundle and, in particular, it is a manifold. The projection π is a
smooth morphism.
Example 1.10. We trivially have Λ0 T ∗ M = M ×R. On the other hand, Λ1 T ∗ M is called
the cotangent bundle of M .
Definition 1.11. A differential p-form ω on M is a smooth section of π, i.e. a smooth
morphism
ω : M → Λp T ∗ M
such that π ◦ ω = IdM .
Note that, for all x ∈ M , ω(x) is an alternating p-form on Tx M .
Notation 1.12. We denote
Ωp (M ) = {ω = differential p-form on M }
and n
M
Ω• M := ΩP (M ).
p=0

Example 1.13. We have


Ω0 (M ) ' {f : M → R is a smooth morphism}.
If M has dimension n then Ωp (M ) = 0 for all p ≥ n + 1.
4 DIFFERENTIAL TOPOLOGY

Definition 1.14. Given ω1 ∈ Ωp M and ω2 ∈ Ωq M , we can define ω1 ∧ ω2 ∈ Ωp+q M , by,


for all x ∈ M ,
(ω1 ∧ ω2 )(x) := ω1 (x) ∧ ω2 (x) ∈ Λp+q Tx∗ M.
Note that associativity, distributivity and super-commutativity hold as in Proposition
1.6.
Assume that M and N are manifolds and let F : M → N be a smooth morphism. Then
for all x ∈ M , we have that the differential of F at x is a linear map
DFx : Tx M → TF (x) N.
Thus, we can define
F ∗ : Λp TF∗ (x) N → Λp Tx∗ M
by, for all ω ∈ Λp TF∗ (x) and v1 , . . . , vp ∈ Λp Tx∗ ,

F ∗ ω(x)(v1 , . . . , vp ) := ω(F (x))(DFx v1 , . . . , DFx vp ).


Thus, given a differential p-form ω on N , we have a differential p-form F ∗ ω on M defined
by
F ∗ ω : M → Λp T ∗ M
such that F ∗ ω(x) = ω(F (x))(DFx v1 , . . . , DFx vp ). The p-form F ∗ ω is called the pull-back
of ω.
Note that given F : M → N , we have defined
F ∗ : Ωp (N ) → Ωp (M ).
Moreover, we have that for ω1 ∈ Ωp (N ) and ω2 ∈ Ωq (N ),
F ∗ (ω1 ∧ ω2 ) = F ∗ (ω1 ) ∧ F ∗ (ω2 ).
If G : N → P is also a smooth morphism between manifolds, we have
(G ◦ F )∗ ω = F ∗ G∗ ω.
Both these properties follow from Proposition 1.8.
1.3. Local description of p-forms. Let M be a manifold of dimension n and let x0 ∈ M .
Let x ∈ U be a local chart with local coordinates (x1 , . . . , xn ) (so that x0 corresponds
to (0, . . . , 0)). Then { ∂∂x , . . . , ∂x∂ } is a basis of Tx0 M and if we define dxi ∈ Tx∗0 M by
1 n

dxi ( ∂∂ xj ) = δi,j then {dx1 , . . . , dxn } is a basis of Tx∗0 M .


A basis of Λp Tx∗0 M is then given by
{dxi1 ∧ · · · ∧ dxip | 1 ≤ i1 < i2 < . . . ik ≤ n}
and a differential p-form on M is locally given by
X
ω= fI dxi1 ∧ · · · ∧ dxip
|I|=p

where fI : U → R is a C ∞ -function for all I = (i1 , . . . , ip ).


DIFFERENTIAL TOPOLOGY 5

Example 1.15. Let F : M → N be a smooth morphism between manifold of dimension


n. Let ω be a n-form on N . Then, locally, we may write ω = f dy1 ∧ · · · ∧ dyn where f
is a smooth function. Then Proposition 1.8 implies that the n-form F ∗ ω on M is locally
given by
F ∗ ω(x) = f ◦ F (x) · det DFx · dx1 ∧ · · · ∧ dxn .

Note that, given a smooth function f : M → R (i.e. f ∈ Ω∗ (M )), then the differential
of f is locally defined by
n
X ∂f
df = dxi ,
i=1
∂xi
or, in other words df (X) = X(f ). Note that df = f ∗ dx, where dx is a 1-form on R.
More in general, let ω ∈ Ωp (M ) be a p-form, locally given by
X
ω= fI dxi1 ∧ · · · ∧ dxip .
|I|=p

Then the differential of ω is locally given by


X
dω = dfI ∧ dxi1 ∧ · · · ∧ dxip .
|I|=p

Note that dω does not depend on the chart U , and in particular we have dω ∈ Ωp+1 (M )
is a (p + 1)-form. The induced map
d : Ωp (M ) → Ωp+1 (M )
is called the de Rham differential of M .
Proposition 1.16. The following properties hold:
• (Leibnitz rule) if ω1 ∈ Ωp1 (M ) and ω2 ∈ Ωq2 (M ), then
d (ω1 ∧ ω2 ) = dω1 ∧ ω2 + (−1)p ω1 ∧ dω2 .
• For all p ≥ 0, we have that the map d ◦ d : Ωp (M ) → Ωp+2 (M ) is the zero map,
i.e. d ◦ d = 0.
• If F : M → N is a smooth map between manifolds, then for any ω ∈ Ωp (M ), we
have
F ∗ dω = d(F ∗ ω).
Definition 1.17. Let ω ∈ Ωp (M ) be a differential p-form. Then
• ω is said to be closed if dω = 0.
• ω is said to be exact if there exists ω 0 ∈ Ωp−1 (M ) such that dω 0 = ω.
Note that it follows immediately from the proposition above that any exact p-form ω
is closed. We will see that the opposite is not true.
6 DIFFERENTIAL TOPOLOGY

1.4. Integrations on manifolds. Let M be a manifold of dimension n and let F : M →


M be a smooth morphism. Recall that if ω ∈ Ωn (M ) is a differential p-form then, for all
x ∈ M,
(F ∗ ω)(x) = det(DF )x · ω(F (x)).
Locally, if (y1 , . . . , yn ) are local coordinates of M , then we may write
ω = f dy1 ∧ · · · ∧ dyn
where f is a C ∞ -function. Assume that {(Uα , φα )} is a smooth atlas on M so that
φα : Uα ⊂ M → Vα ⊂ Rn .
For all α, β, we denote
hα,β := φβ ◦ φ−1 n n
α : φα (Uα ∩ Uβ ) ⊂ R → φβ (Uα ∩ Uβ ) ⊂ R ,

so that hα,β (x1 , . . . , xn ) = (y1 , . . . , yn ). Then, if


ω = f dy1 ∧ · · · ∧ dyn ,
as in Example 1.15 we have
h∗αβ ω(x) = (f ◦ hαβ ) (x) det (Dhαβ (x)) dx1 ∧ · · · ∧ dxn .
Recall that if D ⊂ Rn is a compact subset with border of measure zero and if h : D →
h(D) ⊂ Rn is a diffeomorphism, then
Z Z
f dy1 ∧ . . . ∧ dyn = f ◦ h · | det dh|dx1 ∧ . . . ∧ dxn .
h(D) D

Thus, if ω ∈ Ωn (D) is a differential n-form given by ω = f dx1 ∧ · · · ∧ dxn , then we can


define Z Z
ω := f dx1 ∧ . . . ∧ dxn .
D D

Definition 1.18. Let U ⊂ Rn be an open set. Then the support of a differential p-form
ω ∈ Ωp (U ) is
Supp(ω) := {x ∈ U | ω 6= 0}.
If D = Supp(ω) is compact then
Z Z
ω := ω
U D

is well defined. If Φ : V → U is a diffeomorphism such that det(DΦ)x > 0 for all x ∈ V ,


then Z Z
ω= Φ∗ ω.
U V
DIFFERENTIAL TOPOLOGY 7

1.5. Orientation. Let V be a vector space over R of dimension n and let B1 = (v1 , . . . , vn )
and B2 = (w1 , . . . , wn ) be ordered basis of V . Then B1 and B2 have the same orientation
if det T > 0, where T : V → V is the linear map defined by T (vi ) = wi for all i = 1, . . . , n.
Fix ω ∈ Λn V ∗ . Then an orientation Λ of V is the set of all ordered basis (v1 , . . . , vn )
of V such that ω(v1 , . . . , vn ) > 0. Note that, by Proposition 1.8, if B1 = (v1 , . . . , vn ) and
B2 = (w1 , . . . , wn ) are ordered basis of V with the same orientation then ω(v1 , . . . , vn ) > 0
if and only if ω(w1 , . . . , wn ) > 0.
Let Φ : V → W be an isomorphism of vector spaces with orientations Λ1 and Λ2 on V
and W respectively. Then Φ is said to be orientation preserving if an ordered basis of
V in Λ1 induces an ordered basis of W in Λ2 .
Consider V = Rn . Then the orientation Λ+ , defined by the natural ordered basis
e1 , . . . , en , where ei = (0, . . . , 0, 1, 0, . . . , 0) for i = 1, . . . , n, is called the positive orien-
tation of V .
Now let M be a manifold of dimension n. We want to define an orientation ΛM on
M , i.e. an orientation Λx on Tx M for each x ∈ M , in such a way that it is compatible
with the structure of the manifold M . First assume that M = U ⊂ Rn is an open set of
Rn . Then there is a natural isomorphism Tx U ' Rn for each x ∈ U . Thus, the positive
orientation Λ+ +
U on U is the collection of orientations Λx on Tx V for each x ∈ U , in such
a way that the isomorphism Tx U ' Rn induces the positive orientation Λ+ on Rn .
Now, let M be any manifold and let (Uα , φα ) be a chart. The positive orientation Λ+ α on
(Uα , φα ) is the collection of orientations on Tx M , for each x ∈ Uα , such that, considering
the positive orientation on φα (Uα ), the isomorphism
Dφα,x : Tx Uα → Tφα (x) (φα (Uα ))
is orientation preserving.
Then M is called orientable if there exists an atlas {(Uα , φα )} of positively oriented
charts, such that, for x ∈ Uα ∩ Uβ , the orientation induced by (Uα , φα ) coincides with the
orientation induced by (Uβ , φβ ). Note that, in such a case, all the transition functions
have differential with positive determinant.
Let F : M → N be a smooth morphism between oriented manifold. Then F is said
to be orientation preserving if for any x ∈ M , the linear map DFx : Tx M → TF (x) N
is orientation preserving, with respect to the orientations induced on Tx M and TF (x) N
respectively.

Notation 1.19. For any integer p ≥ 0, we denote


Ωpc (M ) := {ω ∈ Ωp (M ) | ω has compact support}.

Assume that ω ∈ Ωnc (M ) and that Supp(ω) ⊂ U where (U, φ) is a positively oriented
chart of M . Then
∗
φ−1 ω ∈ Ωnc (φ(U ))
8 DIFFERENTIAL TOPOLOGY

and
Z Z
∗
(1) ω := φ−1 ω
M φ(U )

is well defined.
Indeed, let (U, φ) and (U , φ) be positively oriented charts containing Supp(ω). Then
the differential of
φ̄ ◦ φ−1 : φ(U ∩ Ū ) → φ̄(U ∩ Ū )
has positive determinant. Thus,
Z Z Z
−1 ∗ −1 ∗
∗ −1 ∗
φ̄ ◦ φ−1
 
φ̄ ω= φ̄ ω= φ̄ ω
φ̄(Ū ) φ̄(U ∩Ū ) φ(U ∩Ū )
Z
∗ ∗
= φ−1 φ̄∗ φ̄−1 ω
φ(U ∩Ū )
Z Z
−1 ∗
∗
φ−1 ω.

= φ ω=
φ(U ∩Ū ) φ(U )

1.6. Partitions of unity. Now we are finally able to integrate over an entire manifold.
The idea is to use a partition of unity in order to use the previous definition on each chart
and then “glue” it together.
Definition 1.20. Let M be a manifold and let U = {Uα } be an open covering. A parti-
tion of unity with respect to U is a collection of smooth functions fα : M → [0, 1] such
that
Supp(fα ) ⊂ Uα for all α (in particular, fα = 0 outside Uα ),
(1) P
(2) α fα (x) = 1 for all x ∈ M , and
(3) for all x ∈ M , there exists an open nieghbourhood V of x such that Supp(fα )∩V 6= 0
for only finitely many α.
Example 1.21. Let S 1 ⊂ R2 be the unit circle. Let U1 = S 1 \ {(−1, 0)} and U2 =
S 1 \ {(1, 0)}. Let fi : M → [0, 1] be defined by
1 1 1 1
f1 (cos(θ), sin(θ)) =+ cos(θ), f2 (cos(θ), sin(θ)) = − cos(θ).
2 2 2 2
It is easy to check that f1 and f2 define a partition of unity with respect to {Ui }.
We will omit the proof of the following
Proposition 1.22. If M is a manifold, any open cover of M induces a partition of unity.
Proposition 1.23. Let M be a manifold of dimension n. Then M is orientable if and
only if there exists a non-vanishing n-form ω ∈ Ωn (M ). ω is called volume form.
DIFFERENTIAL TOPOLOGY 9

Proof. Assume that ω is a non-vanishing n-form on M and let {(Uα , φα )} be an atlas on


M . For each α, we may write
(φ−1 ∗
α ) ω = gα dx1 ∧ · · · ∧ dxn ,

for some smooth function gα on φα (Uα ). Note that since ω is non-vanishing, it follows that
gα is either positive or negative on φα (Uα ). After possibly replacing φα by the composition
t ◦ φα where
t : Rn → Rn (x1 , . . . , xn ) 7→ (−x1 , x2 , . . . , xn ),
we may assume that gα is positive for all α, and we consider the orientation on Uα so that
(Uα , φα ) is positively oriented. By Proposition 1.8, if x ∈ Uα ∩ Uβ for some α and β, then
(Uα , φα ) and (Uβ , φβ ) define the same orientation on x. Thus, M is orientable.
Viceversa, assume that M is orientable and let U = (Uα , φα ) be an atlas of positively
oriented charts. Let fα be a partition of the unity with respect to U and define ωα =
φ∗α (dx1 ∧· · ·∧dx Pn ). Let ω̃α be the n-form defined by ωα inside Uα and equal to zero outside
Uα . Let ω := α fα ω̃α . Note that, for each x ∈ M and for each positively oriented basis
v1 , . . . , vp , we have that ω(x)(v1 , . . . , vp ) > 0. Thus, ω is a volume form on M . 

We can now define the integral of a n-form on a n-dimensional manifold:


Definition 1.24. Let M be a oriented manifold. Let ω ∈ Ωnc (M ) and let (Uα , φα ) be
an atlas of M with positively oriented smooth charts. Let fα be a partition of unity with
respect of Uα .
Then the integral of ω over M is defined by
Z XZ
(2) ω= fα ω.
M α M

Remark 1.25. Note that for each α, we have that the support of fα ω is contained in Uα
and therefore each term of the sum is well-defined as in (1). Indeed, we have
Z XZ XZ ∗
ω= fα ω = φ−1
α fα .
M α M α φα (Uα )
R
Lemma 1.26. The integral M ω does not depend on the choice of the atlas (Uα , φα ) and
on the choice of the partition of the unity fα .
Proof. We have already shown that, for a n-form whose support is contained in a smooth
chart, the integral in (1) does not depend on the chosen chart. Let (Uα , φα ) and Ūα , φα
be atlases on M defined by positively oriented charts and let fα and f α be partition of
unity with respect to these coverings. Then, since by definition,
X X
fα = f¯α = 1,
α α
10 DIFFERENTIAL TOPOLOGY

we have !
Z Z X XZ
fα ω = f¯β fα ω = f¯β fα ω,
M M β β M

Thus,
XZ XZ
fα ω = f¯β fα ω
α M α,β M
!
XZ X XZ
= f¯β fα ω= f¯β ω.
β M α β M
R P R
Thus, the integral M
ω= α M
fα ω does not depend on the choice of the partition of
unity. 

Proposition 1.27. Let M and N be oriented manifolds of dimension n and let ω, η ∈


Ωnc (M ).
Then the following properties hold:
R R R
• (Linearity) M (aω + bη) = a M ω + b M η for all a, b ∈ R.
• (Orientation reversal) Let M denote the manifold M with opposite orientation.
Then Z Z
ω=− ω.
M M
R
• (Positivity) Let ω be a volume form on M , then M ω > 0.
• (Diffeomorphism invariance) Let F : N → M be an orientation preserving diffeo-
morphism. Then
Z Z
ω= F ∗ ω.
M N

Proof. The first two properties follow directly from the definition.
Let ω be a volume form. Then for each atlas U = (Uα , φα ) and for each partition of the
unity fα with respect to U, the function φ−1 α fα ω is locally of the form gdx1 ∧ · · · ∧ dxn for
some local coordinate (x1 , . . . , xn ), where g is a smooth function such that g > 0. Hence
each
R term in the sum (2) is non-negative and at least one term is strictly positive. Thus,
M
ω > 0.
Now, let F : N → M be an orientation preserving diffeomorphism. We may assume
that the support of ω is contained in a unique positively oriented chart (U, φ).
By assumption, (F −1 (U ), φ ◦ F ) is also a positively oriented chart with contains the
support of F ∗ ω. Thus,
Z Z
ω= F ∗ ω.
U F −1 (U )
DIFFERENTIAL TOPOLOGY 11

Thus, by taking the sum as in (2), it follows that


Z Z
ω= F ∗ ω,
M N
as claimed. 
1.7. Manifolds with boundary. We consider the set
Rn+ := {(x1 , . . . , xn ) ∈ Rn | xn ≥ 0}
with the topology induced by Rn . The boundary of Rn+ is the set
∂Rn+ := {(x1 , . . . , xn ) ∈ Rn | xn = 0}.
Definition 1.28. A manifold with boundary of dimension n is a Hausdorff topo-
logical space M with an atlas {(Uα , φα )}, such that the functions φα : Uα → Rn>0 is a
homeomorphism for any α and, for each α and β, the induced morphism
φα ◦ φ−1
β : φβ (Uα ∩ Uβ ) → φα (Uα ∩ Uβ )

is a smooth morphism.
The boundary ∂M of M is defined as
∂M := {x ∈ M | φα (x) ∈ ∂Rn }.
The interior of M is the set int(M ) := M \ ∂M .
Remark 1.29.
• If M is a manifold with boundary then the boundary of M is a closed subset of M .
In particular if M is compact then also ∂M is compact.
• If M is a manifold with boundary of dimension n then the interior of M is a
manifold of dimension n.
• Manifolds are defined intrinsically, meaning that they are not defined as subsets
of another topological space, therefore, the notion of boundary will differ from the
usual boundary of a topological subset.
Example 1.30.
• M = [0, 1] ⊂ R is a manifold with boundary and ∂M = {0, 1}.
• The closed disc
D := {x ∈ Rn | |x| ≤ 1}
is a manifold with boundary and
∂M = {x ∈ Rn | |x| = 1} =: S n−1 .
• The cylinder defined as the product of a real interval with the circle M = [0, 1]×S 1
is a manifold with boundary and the boundary ∂M is the union of two circles, i.e
a manifold with two components.
Remark 1.31.
12 DIFFERENTIAL TOPOLOGY

• The notions of the tangent space, tangent map, tensors and tensor fields, ori-
entability, differential forms,. . . are defined for manifolds with boundary in exactly
the same way as for ordinary manifolds.
• If M is an oriented manifold then ∂M is also oriented. Recall that if M = Rn ,
the volume form dx1 ∧ · · · ∧ dxn defines the positive orientation on M .
Similarly, for each point in the interior of Rn+ , the volume form dx1 ∧ · · · ∧ dxn
defines a positive orientation which, as a convention, induces the positive volume
form on ∂M given by (−1)n dx1 ∧ · · · ∧ dxn .

1.8. Stokes’ Theorem. Let M be a manifold with boundary M and let U = {Uα } be
an open cover of M . A partition of the unity with respect to U can be defined exactly
as in Definition 1.20. As in Proposition 1.22, any manifold with boundary M , with open
cover U = {Uα }, admits a partition of the unity with respect to U.
n
R Similarly, if M is a manifold with boundary of dimension n and ω ∈ Ωc then the integral
M
ω of ω on M is defined as in Definition 1.24.
Theorem 1.32 (Stokes’ Theorem). Let M be a smooth oriented manifold with boundary
and let ω ∈ Ωn−1
c (M ).
Then Z Z
dω = ω.
M ∂M
Proof. Let {(Uα , φα )} be an atlas on M and let fα be a partition of the unity with respect
to this covering. Then
Z XZ XZ ∗
dω = d (fα ω) = φ−1
α d (fα ω) .
M α M α φα (Uα )
By Proposition 1.16, we have that
(φ−1 ∗ −1 ∗
α ) d (fα ω) = d(φα ) fα ω.
∗˜
The (n − 1)-form (φ−1 n
α ) fα ω in R+ can be writen in local coordinates on φα (Uα ) as
X n
fα ωj dx1 ∧ · · · ∧ dx
cj ∧ . . . dxn
j

where ωj is a smooth function on φα (Uα ), f˜α := fα ◦ φ−1 α and dxj means that we omit the
c
component dxj . The exterior derivative of this form is therefore given locally by
 
X n
!
Xn ∂ ˜
f ω
α j
d f˜α ωj dx1 ∧ · · · ∧ dx
cj ∧ . . . dxn = dxj ∧ dx1 ∧ · · · ∧ dx
cj ∧ . . . dxn
j=1 j=1
∂x j
  
X n ∂ f˜α ωj
=  (−1)j−1  dx1 ∧ · · · ∧ dxn .
j=1
∂x j
DIFFERENTIAL TOPOLOGY 13

Thus, we have
 
Z XZ +∞ Z +∞ Z ∞ Xn ∂ f˜α ωj
dω = ··· (−1)j−1 dxn . . . dx1
M α −∞ −∞ 0 j=1
∂x j
| {z }
n−1 times

where the i-th integral of the (n − 1)-first integrals correspond to the integration over dxi ,
for i = 1, . . . , n − 1, and the last integral corresponds to the integration over dxn . We can
apply Fubini’s theorem in order to switch the order of the integrals and integrate over dxi
first. Thus, by the fundamental theorem of calculus we get
Z XX n Z +∞ Z[+∞ Z +∞ Z ∞  
dω = ... ... (−1)j−1 f˜α ωj dxn . . . dx
cj . . . dx1
M α j=1 −∞ −∞ −∞ 0
at boundary

where we have to evaluate the result of each integral at the boundary of the domain of
integration. But on this boundary if Uα is a chart of the interior of M , then fα has to
be zero and on the boundary charts, only the part of the boundary of the chart which
touches the boundary of M (i.e. where xn = 0) will have fα 6= 0. The extra minus sign
comes from the fact that the non-vanishing term of the integral is its lower bound.
Thus, by our convention as in Remark 1.31, we have
Z X Z +∞ Z +∞ Z

dω = ... (−1) fα (x1 , . . . , xn−1 , 0)ωn dx1 ∧ . . . ∧ dxn−1 = ω
M α −∞ −∞ ∂M

as claimed. 

We now see some applications of Stokes’ theorem.


Proposition 1.33 (Integration by parts). Let M be a manifold with boundary of dimen-
sion n and let p ≥ 0. Let ω ∈ Ωpc (M ) and η ∈ Ωn−p−1
c (M ).
Then Z Z Z
ω∧η = dω ∧ η + (−1)p ω ∧ dη.
∂M M M

Proof. By Stokes’ theorem and by Leibnitz rule (cf. Remark 1.16), we have
Z Z Z Z
p
ω∧η = d(ω ∧ η) = dω ∧ η + (−1) ω ∧ dη. 
∂M M M M

Theorem 1.34 (Brouwer’s fixed point theorem). Let


D := {x ∈ Rn | |x| ≤ 1}
be the closed disc and let f : D → D be a smooth morphism.
Then f admits a fixed point, i.e. there exists x ∈ D such that f (x) = x.
14 DIFFERENTIAL TOPOLOGY

Proof. Suppose by contradiction that f (x) 6= x for all x ∈ D. Then the ray beginning at
f (x) and passing through x is uniquely defined. Let g(x) be the point that the ray meets
at the boundary of D. Then g : D → ∂D is a smooth function such that
g(x) = x for all x ∈ ∂D.
Note that ∂D is the (n − 1)-dimensional sphere, which is orientable. Thus, by Proposition
1.23, it admits a volume form ω ∈ Ωn−1 (∂D). Thus,
Z Z Z Z
∗ ∗
0< ω= g ω= d (g ω) = g ∗ dω
∂D ∂D D D
where the first equality comes from the fact that g is the identity on ∂D, the second from
Stokes’ theorem and the third follows by Proposition 1.16. But dω is a n-form on the
(n − 1)-dimensional manifold ∂D and, thus, it is zero (see Example 1.13). Thus, we get
a contradiction. 

Example 1.35. Recall that an exact form on a manifold M (cf. Definition 1.17) is always
closed. We now show that the opposite does not hold. Let M = R2 \ {(0, 0)} and let
x y
ω= 2 2
dy − 2 dx.
x +y x + y2
Then    
∂ x ∂ y
dω = dx ∧ dy − dy ∧ dx
∂x x2 + y 2 ∂y x2 + y 2
.
(x2 + y 2 ) − x · 2x (x2 + y 2 ) − y · 2y
= dx ∧ dy − dy ∧ dx = 0
(x2 + y 2 )2 (x2 + y 2 )2
Thus, ω is closed.
Let
γ : [0, 2π] → M γ(θ) = (cos θ, sin θ) ∈ M.
Then Z 2π Z 2π
cos2 θ − sin θ(− sin θ)
Z Z

ω= γ ω= dθ = dθ = 2π.
S1 [0,2π] 0 cos2 θ + sin2 θ 0
Assume by contradiction that ω is exact. Then there exists a 0-form, i.e. a smooth
function (cf. Example 1.13), f on M such that ω = df . But as S 1 has no boundary,
Stokes’ theorem imply Z Z Z
ω= df = f = 0.
S1 S1 ∂S 1
Hence, we get a contradiction.

As in the previous example, Stokes’ theorem imply:


Proposition 1.36. Let ω ∈ Ωnc (M ) be an exact form on an oriented manifold M of
dimension
R n without boundary.
Then M ω = 0.
DIFFERENTIAL TOPOLOGY 15

Similarly, Stokes’ theorem immediately implies:


Proposition 1.37. Let M be an oriented manifold of dimension n with boundary and let
ω ∈ Ωn−1
c R(M ) be a closed form.
Then ∂M ω = 0.

Let M be an oriented manifold of dimension n and let N ⊂ M be a oriented submanifold


of dimension k. Let ω ∈ Ωkc (M ). Then we define the integral of ω on N as
Z Z
ω := i∗ ω
N N

where i : N → M is the embedding map. Note that i∗ ω ∈ Ωk (N ). We denote


ω|N := i∗ ω.

Proposition 1.38. Let M be an oriented manifold of dimension n and let ω ∈ Ωkc . Let
R ⊂ M be a compact oriented submanifold of dimension k without boundary and suppose
S
S
ω 6= 0.
Then
(1) ω is not exact on M and ω|S is not exact on S.
(2) S is not the boundary of a compact oriented submanifold N ⊂ M of dimension
k + 1.
Proof. We first prove (1). Assume ω|S is exact on S. Then there exists η ∈ Ωk−1 (M ) such
that dη = ω. Thus, by Proposition 1.36, we have
Z Z
0 6= ω= dη = 0,
S S

a contradiction. Hence ω|S is not exact on S and, consequently, ω is not exact on M .


We now prove (2). Assume that N ⊂ M is a submanifold with boundary of dimension
k + 1 such that ∂N = S. Then, since ω is closed, Proposition 1.37 implies
Z Z Z
0= dω = ω= ω,
N ∂N S

a contradiction. 

2. de Rham Cohomology
Definition 2.1. Let M be a manifold and let p ≥ 0. Two closed forms ω, ω 0 ∈ Ωp (M ) are
said to be cohomologous if their difference ω − ω 0 is exact, i.e. ω − ω 0 = dη for some
η ∈ Ωp−1 (M ).
In particular, a closed form is cohomologous to zero if it is exact.
16 DIFFERENTIAL TOPOLOGY

Note that the set of closed p-forms coincides with the kernel of the de Rham differential
d : Ωp (M ) → Ωp+1 (M ).
We will denote it by Z p (M ). Thus,
Z p (M ) := Ker d : Ωp (M ) → Ωp+1 (M ) ⊂ Ωp (M ).


Similarly, if p ≥ 1, then the set of exact p-forms coincides with the image of
d : Ωp−1 (M ) → Ωp (M ).
We will denote it by B p (M ). Thus,
B p (M ) := Im d : Ωp−1 (M ) → Ωp (M ) ⊂ Ωp (M ).


By convention, we denote B 0 (M ) = 0.
Both Z p (M ) and B p (M ) are vector spaces over R. Furthermore, Proposition 1.16
implies that for each p ≥ 1, we have
B p (M ) ⊂ Z p (M ).
Note that two closed p-forms ω, ω 0 ∈ Z p (M ) are cohomologous if and only if their
difference lies in B p (M ).
Definition 2.2. The quotient space
Z p (M )
H p (M ) :=
B p (M )
is called the p-th de Rham cohomology group of M .
For each closed p-form ω ∈ Z p (M ), the equivalence class
[ω] := {ω 0 ∈ Z p (M ) | ω − ω 0 ∈ B p (M )} ∈ H k (M )
is called a p-th de Rham cohomology class.
Note that H p (M ) is a vector space over R. If M is a non-trivial manifold then the
vector spaces Z p (M ) and B p (M ) are infinitely dimensional. On the other hand, we will
see that if M is compact then H p (M ) is finitely dimensional. Thus, we define:
Definition 2.3. The dimension of the p-th de Rham cohomology group
bp (M ) := dim H p (M )
is called the p-th Betti number of M .
Proposition 2.4. Let M be a connected manifold. Then H 0 (M ) = R and, in particular,
b0 (M ) = 1.
Proof. As in Example 1.13, we have Ω0 (M ) = C ∞ (M ). Let f ∈ Ω0 (M ). Locally, we may
write n
X ∂f
df = dxi .
j=1
∂xi
DIFFERENTIAL TOPOLOGY 17

∂f
Thus, if f is closed then ∂x i
= 0 for all i = 1, . . . , n. It follows that f is locally constant
and, since M is connected, it is constant. Therefore Z 0 (M ) = R. Since B 0 (M ) = 0, the
result follows. 
More in general, the same proof implies that if M admits m connected components
then b0 (M ) = m.
Proposition 2.5. Let M be a manifold of dimension n.
Then H p (M ) = 0 for all p ≥ n + 1.
Proof. As in Example 1.13, Ωp (M ) = 0 for all p ≥ n + 1. Thus, H p (M ) = 0. 
Proposition 2.6. Let M be a compact orientable manifold of dimension n and without
boundary.
Then H n (M ) 6= 0.
R
Proof. By Proposition 1.23, M admits a volume form ω ∈ Ωn (M ). In particular, M ω > 0.
Since dω is a (n+1)-form, it must vanish (cf. Example 1.13), i.e. ω is closed. In particular,
[ω] ∈ H n (M ).
By Proposition 1.36, it follows that ω is not exact, i.e. [ω] 6= 0. Thus, the claim
follows. 
Proposition 2.7. Let M and N be manifolds and let G : M → N be a smooth morphism.
Then, for all p ≥ 0,
G∗ : Ωp (N ) → Ωp (M )
takes closed forms to closed forms and exact forms to exact forms.
Proof. The result follows immediately from Proposition 1.16. 
From the Proposition, it follows that if G : M → N is a smooth morphism between
manifolds and p ≥ 0, then the linear map
H p (N ) → H p (M ) [ω] 7→ [G∗ ω]
is well defined. We will still denote this map by
G∗ : H p (N ) → H p (M ).
Corollary 2.8. If M and N are diffeomorphic then their cohomology groups H p (M ) and
H p (N ) are isomorphic for all p ≥ 0.
Proof. Let F : M → N be a diffeomorphism. By Proposition 2.7, F and F −1 induce linear
maps
F ∗ : H p (N ) → H p (M ) and (F −1 )∗ : H p (M ) → H p (M ).
Since F ◦ F −1 = IdN and F −1 ◦ F = IdM , it follows by Proposition 1.8 that (F −1 )∗ ◦ F ∗ =
IdH p (N ) and F ∗ ◦ (F −1 )∗ = IdH p (M ) . Thus, the claim follows. 
18 DIFFERENTIAL TOPOLOGY

2.1. Homotopy invariance.


Definition 2.9. Let M0 and M1 be manifolds. Let f0 , f1 : M0 → M1 be smooth morphisms.
Then f0 and f1 are said to be smoothly homotopic if there exists a smooth map
H : M0 × [0, 1] → M1 ,
such that
H(x, i) = fi (x) for any x ∈ M0 and i = 0, 1.
The morphism H is called smooth homotopy between M0 and M1 .
Notation 2.10. Under the assumptions above, we will denote f0 ∼ f1 and
ft (x) = H(x, t) for all x ∈ M0 , t ∈ [0, 1].
Definition 2.11. Let M0 and M1 be manifolds. A smooth morphism f : M0 → M1 is
called a smooth homotopy equivalence if there exists a smooth map g : M1 → M0
such that
g ◦ f ∼ IdM0 and f ◦ g ∼ IdM1 .
In this case, M0 and M1 are called homotopy equivalent and g is the homotopy
inverse of f .
Example 2.12.
• We want to show that M = Rn is homotopy equivalent to a point {x0 }. We may
assume that x0 is the origin 0 ∈ Rn . Let i : {0} → M and π : M → {0} be the
natural maps. Then π ◦ i = Id{0} . Let
H : M × [0, 1] → M (x, t) 7→ t · x.
Then H(x, 0) = i◦π and H(x, 1) = IdM . Thus, i◦π ∼ IdM , and the claim follows.
• We want to show that M = R2 \ {0} is homotopy equivalent to the unit circle
N = S 1 . Let
x
f: M →N x 7→ ,
|x|
and let i : N → M be the inclusion. Then f ◦ i = IdN . Let
x
H : M × [0.1] → M (x, t) 7→ (1 − t) + tx.
|x|
Then H(x, 0) = i ◦ f and H(x, 1) = IdM . Thus, i ◦ f ∼ IdM , and the claim follows.
Proposition 2.13. Let M and N be manifolds and let H : M × [0, 1] → N be a smooth
map. For any t ∈ [0, 1], denote
ft = H(·, t) : M → N.
Then, for all p ∈ [0, 1], the pull-back
ft∗ : H p (N ) → H p (M )
does not depend on t ∈ [0, 1].
DIFFERENTIAL TOPOLOGY 19

Proof. Let t1 , t2 ∈ [0, 1] such that t1 < t2 . Let


ik : M → M × [0, 1] ik (x) = (x, tk ) for k = 1, 2.
We want to show that there exists a linear map
h : Ωp (M × [t1 , t2 ]) → Ωp−1 (M )
such that, for all ω ∈ Ωp (M × [t1 , t2 ]), we have
(3) d(h(ω)) + h(dω) = i∗2 ω − i∗1 ω.
Indeed, if η ∈ Z p (N ) is a closed p-form, then d(H ∗ η) = 0 and (3) implies
[ft∗1 η] = [i∗1 H ∗ η] = [i∗2 H ∗ η] = [ft∗2 η] in H p (M ).
Thus, the Proposition follows.
We define h by, for all x ∈ M and v1 , . . . , vp−1 ∈ Tx M ,
Z t2  

h(ω)(x)(v1 , . . . , vp−1 ) := ω(x, t) , v1 , . . . , vp−1 dt.
t1 ∂t
It follows easily that h : Ωp (M × [t1 , t2 ]) → Ωp−1 (M ) is linear. In order to show (3), it is
enough to show that it holds locally. Locally, around a point (x, t) ∈ M × [0, 1], we can
choose coordinates (x1 , . . . , xn , t) and we can write ω as
X X
gI (x, t)dxi1 ∧ · · · ∧ dxip + gJ (x, t)dxj1 ∧ · · · ∧ dxjp−1 ∧ dt
|I|=p |J|=p−1

where gI and gJ are smooth functions. By linearity, it is enough to prove (3) for each
term of the sum.
Thus, we can distinguish two cases. We first assume that
ω(x, t) = g(x, t)dxi1 ∧ · · · ∧ dxip .


Then ω(x, t) v1 , . . . , vp−1 , ∂t = 0 and, in particular, h(ω) = 0. We have
 
∂g
h(dω) = h dt ∧ dxi1 ∧ · · · ∧ dxip + terms without dt
∂t
Z t2 
∂g
= dt dxi1 ∧ · · · ∧ dxip =
t1 ∂t
= (g(x, t2 ) − g(x, t1 ))dxi1 ∧ · · · ∧ dxip
= i∗2 ω − i∗1 ω.
Thus, the claim follows.
Now we assume that
ω(x, t) = g(x, t)dxj1 ∧ · · · ∧ dxjp−1 ∧ dt.
First, note that i∗1 ω = i∗2 ω = 0, since t ◦ i1 and t ◦ i2 are constant functions.
20 DIFFERENTIAL TOPOLOGY

We also have
Z t2  
p−1
d(h(ω)) = (−1) ·d g(x, t)dt dxj1 ∧ · · · ∧ dxjp−1
t1
n Z t2 
p−1
X ∂
= (−1) · g(x, t)dt dxj ∧ dxj1 ∧ · · · ∧ dxjp−1
j=1
∂xj t1
n Z t2 
p−1
X ∂
= (−1) · g(x, t)dt dxj ∧ dxj1 ∧ · · · ∧ dxjp−1 .
j=1 t1 ∂xj

On the other hand,


n
!
X ∂
h(dω) = h g(x, t)dxj ∧ dxj1 ∧ · · · ∧ dxjp−1 ∧ dt
j=1
∂xj
n Z t2 
p
X ∂
= (−1) · g(x, t)dt dxj ∧ dxj1 ∧ · · · ∧ dxjp−1
j=1 t1 ∂xj

= −d(h(ω)).
Thus, also in this case, (3) holds. 
Corollary 2.14. Let M and N be manifolds and let f : M → N be a smooth homotopy
equivalence.
Then, for all p ≥ 0, the pull-back f ∗ : H p (N ) → H p (M ) is an isomorphism.
Proof. By assumption there exists g : N → M such that f ◦ g ∼ IdN and g ◦ f ∼ IdM . By
Proposition 1.8 and Proposition 2.13, it follows that
g ∗ ◦ f ∗ = (f ◦ g)∗ = (IdN )∗ = IdH p (N ) .
Similarly f ∗ ◦ g ∗ = IdH p (M ) . Thus, the claim follows. 
Definition 2.15. A manifold M is (smoothly) contractible if it is homotopy equivalent
to a point.
As in Example 2.12, M = Rn is contractible. More in general, any convex submanifold
M ⊂ Rn is contractible. Recall that M is said to be convex if, for any two points x, y ∈ M ,
the segment joining x and y is contained in M .
Theorem 2.16 (Poincaré’s lemma). If M ⊂ Rn is a contractible submanifold then
H 0 (M ) = 0 and H p (M ) = 0 for any p ≥ 1.
Proof. By Corollary 2.14, H p (M ) = H p ({x0 }) for any p ≥ 0, where x0 ∈ M is a point.
By Proposition 2.4, we have that H 0 ({x0 }) = R and by Proposition 2.5, we have that
H p ({x0 }) = 0 for all p ≥ 1. 
DIFFERENTIAL TOPOLOGY 21

Corollary 2.17. Let M be a manifold and let ω ∈ Ωp (M ) be a closed form for some
p ≥ 1.
Then, for any x ∈ M , the form ω is exact on some open neighbourhood U of x.
Proof. For any x ∈ M , there exists an open neighbourhood U of x which is diffeomorphic
to a ball B ⊂ Rn . Since B is convex, it is also contractible and by Proposition 2.16, we
have that H p (B) = 0 for all p ≥ 1. By Corollary 2.8, it follows that H p (U ) = 0 for all
p ≥ 1. Thus, [ω] = 0 ∈ H p (U ), i.e. ω is exact on U . 
Recall that, as in Example 1.35, closed forms are not always exact.

Definition 2.18. Let M be a manifold and let x, y ∈ M be points. A homotopy of


paths from x to y is a continuous map F : [0, 1] × [0, 1] → M such that
F (0, t) = x and F (1, t) = y for all t ∈ [0, 1].
In this case, the paths F (·, 0), F (·, 1) : [0, 1] → M are said to be homotopic paths from
x to y.
Proposition 2.19. Let M be a manifold and let ω ∈ Ω1 (M ) be a closed form. Let
γ1 , γ2 : [0, 1] → M be smooth paths from x to y which are homotopic.
Then Z Z
ω= ω.
γ1 γ2

The proof follows essentially from Stokes’ Theorem. The problem is that [0, 1] × [0, 1]
is not a manifold with boundary and therefore it requires a bit more work, which we omit
here.
Proposition 2.20. Let M be a simply connected oriented manifold (i.e. π1 (M ) = 0).
Then H 1 (M ) = 0.
R
Proof. We first show that a 1-form ω ∈ Ω1 (M ) is exact if and only if γ ω = 0 for any
smooth closed path γ : [0, 1] → M . Assume that ω is exact. Then ω = df for some
smooth function f : M → R. Since γ is closed, we have γ(0) = γ(1) =: x. Then, by the
fundamental theorem of calculus, we have
Z Z
ω = df = f (x) − f (x) = 0.
γ γ
R
Viceversa, fix x ∈ M and assume that γ ω = 0 for any smooth closed path γ : [0, 1] →
Ry
M . For any y ∈ M , define f (y) = x ω. The vanishing implies that f (y) does not depend
on the path γ. Moreover, df = ω. Thus, ω is exact.
R
In order to prove the result, we need to show that γ ω = 0 for all 1-forms. Since M
is simply connected, any closedR loop γ is path homotopic to the constant loop c. By
Proposition 2.19, we then have γ ω = 0 and ω is exact. 
22 DIFFERENTIAL TOPOLOGY

2.2. Some homological algebra. In order to prove Mayer-Vietoris theorem, we intro-


duce some basic notions in homological algebra.
Definition 2.21. A cochain complex (C • , d• ) of vector spaces (resp. abelian groups,
modules) is a sequence of vector spaces (resp. abelian groups, modules) C k , with k ∈ Z,
and homomorphisms dk : C k → C k+1 such that dk+1 ◦ dk = 0 for each k. The map dk is
called differential.
Notation 2.22. Assume that C := (C • , d• ) is a cochain complex of vector spaces. The
elements in
Z k (C) := Ker(dk ) ⊂ C k
are called cocycles. The elements in
B k (C) := Im(dk−1 ) ⊂ C k
are called coboundaries. Note that B k (C) ⊂ Z k (C) for all k. The quotient
Z k (C)
H k (C) =
B k (C)
is called the k-th cohomology group of C.
Definition 2.23. Let (C • , d• ) and (D• , d• ) be two cochain complexes. A cochain map
f : (C • , d• ) → (D• , d• ) is a collection of homomorphisms f k : C k → Dk such that f k+1 ◦
dk = dk ◦ f k for all k ∈ Z.
Proposition 2.24. A cochain map f : (C • , d• ) → (D• , d• ) between cochain complexes
induces a homomorphism of cohomology groups H k (C) → H k (D) for all k.
Proof. Let ω ∈ Z k (C). Then dk ω = 0. Thus, dk f k ω = f k+1 dk ω = 0, which implies that
f k ω ∈ Z k (D). Similarly, if ω ∈ B k (C), then ω = dk−1 η for some η ∈ C k−1 . It follows that
f k ω = f k dk−1 η = dk−1 f k−1 η ∈ B k (D).
Thus, the claim follows. 
Definition 2.25. A sequence of homomorphisms
f1 f2 fn
C 1 −→ C 2 → . . . −→ C n
between vector spaces (resp. abelian groups, modules) is called exact if Im(f k ) = Ker(f k+1 )
for all k = 1, . . . , n − 1.
Example 2.26.
f
• The sequence 0 → C 1 → C 2 is exact if and only if f is injective.
f
• The sequence C 1 → C 2 → 0 is exact if and only if f is surjective.
DIFFERENTIAL TOPOLOGY 23

• A short exact sequence is a a sequence


f1 f2
0 → C 1 −→ C 2 −→ C 3 → 0.
Note that C 1 = Kerf 2 and C 3 = Cokerf 1 . In particular, if W is a subspace of the
vector space V then we have a short exact sequence
0 → W → V → V /W → 0.
Lemma 2.27 (Snake Lemma). Consider the following commutative diagram
f1 f2
C 1 −−−→ C 2 −−−→ C 3 −−−→ 0
  
α1 y α2 y α3 y
  

g1 g2
0 −−−→ D1 −−−→ D2 −−−→ D3
where the horizontal lines are exact.
Then there exists a homomorphism δ : Ker(α3 ) → Coker(α1 ) such that
δ
Ker(α1 ) → Ker(α2 ) → Ker(α3 ) −→ Coker(α1 ) → Coker(α2 ) → Coker(α3 )
is exact.
Moreover, if we consider the diagram
f1 f2
0 −−−→ C 1 −−−→ C 2 −−−→ C 3 −−−→ 0
  
α1 y α2 y α3 y
  

g1 g2
0 −−−→ D1 −−−→ D2 −−−→ D3 −−−→ 0
where the horizontal lines are exact then
δ
0 → Ker(α1 ) → Ker(α2 ) → Ker(α3 ) −→ Coker(α1 ) → Coker(α2 ) → Coker(α3 ) → 0
is exact.
Idea of the proof. We just show how to construct the homomorphism δ. Let x ∈ Ker(α3 ) ⊂
C 3 . Since f 2 is surjective, there exists y ∈ C 2 such that f 2 (y) = x. By commutativity,
we have
g 2 ◦ α2 (y) = α3 ◦ f 2 (y) = α3 (x) = 0.
Thus, α2 (y) ∈ Ker(g 2 ) = Im(g 1 ), and therefore there exists z ∈ D1 such that g 1 (z) =
α2 (y). We define δ(x) to be the image of z by D1 → Coker(α1 ).
It is easy to check that δ is well defined and that the long sequences in the Lemma are
exact. 
24 DIFFERENTIAL TOPOLOGY

2.3. The Mayer-Vietoris sequence. Let M be a manifold of dimension n such that


M = U ∪ V , where U, V ⊂ M are open subsets and U ∩ V is not empty. Denote by
iU : U → M, iV : V → M, jU : U ∩ V → U, jV : U ∩ V → V
the inclusion maps. These maps induce pull-backs i∗U , i∗V , jU∗ and jV∗ on the space of
differential p-forms, for each p ≥ 0. Let
f := (i∗U , i∗V ) : Ωp (M ) → Ωp (U ) ⊕ Ωp (V ) ω 7→ (ω|U , ω|V )
and
g := jV∗ − jU∗ : Ωp (U ) ⊕ Ωp (V ) → Ωp (U ∩ V ), (ωU , ωV ) 7→ ωV |U ∩V − ωU |U ∩V
Proposition 2.28. With the same notation as above, the sequence
f g
0 → Ωp (M ) → Ωp (U ) ⊕ Ωp (V ) → Ωp (U ∩ V ) → 0
is exact.
Proof. We first show that f is injective. Let ω ∈ Ωp (M ) such that f (ω) = 0. Then
i∗U ω = i∗V ω = 0,
i.e. the restriction of ω to both U and V is zero. Since X = U ∪ V , it follows that ω = 0.
Thus, f is injective.
Now we show that Ker(g) = Im(f ). If ω ∈ Ωp (M ), then
g(f (ω)) = i∗V (ω)|U ∩V − i∗U (ω)|U ∩V = 0.
Thus, Im(f ) ⊂ Ker(g). Now assume that (ω1 , ω2 ) ∈ ker(g). Then
ω1 |U ∩V = ω2 |U ∩V
and, in particular, there exists ω ∈ Ωp (M ) such that
(
ω1 on U
ω=
ω2 on V.
Thus, (ω1 , ω2 ) = f (ω) ∈ Im(f ). It follows that Ker(g) = Im(f ).
It remains to show that g is surjective. Let η ∈ Ωp (U ∩ V ). Let {fU , fV } be a partition
of the unity with respect to the open cover {U, V }, and whose existence is guaranteed by
Proposition 1.22. Then, there exist η̃1 ∈ Ωp (U ) and η̃2 ∈ Ωp (V ) such that
( (
fV · η on U ∩ V fU · η on U ∩ V
η̃1 = and η̃2 =
0 on M \ Supp(fV ) 0 on M \ Supp(fU ).
We then have
g(−η̃1 , η̃2 ) = η̃2 |U ∩V + η̃1 |U ∩V = fU · η + fV · η = η.
Thus, g is surjective. 
DIFFERENTIAL TOPOLOGY 25

Theorem 2.29 (Mayer-Vietoris). With the same notation as above, for each p ≥ 0, there
exists a linear map
δ : H p (U ∩ V ) → H p+1 (M )
such that the following sequence is exact:
δ f g δ f
. . . → H p (M ) → H p (U ) ⊕ H p (V ) → H p (U ∩ V ) → H p+1 (M ) → . . .
Proof. By Proposition 2.28 and the fact that the de Rham differential commutes with the
pull-back (cf. Proposition 1.16), the following diagram
f g
0 → Ωp (M ) −−−→ Ωp (U ) ⊕ Ωp (V ) −−−→ Ωp (U ∩ V ) → 0
  
dpM y
 (dp ,dp ) dp
y U V y U ∪V
f g
0 → Ωp+1 (M ) −−−→ Ωp+1 (U ) ⊕ Ωp+1 (V ) −−−→ Ωp+1 (U ∩ V ) → 0
is commutative.
We get a new commutative diagram
f g
Coker(dp−1 ) −−−→ Coker(dUp−1 , dp−1 p−1
V ) −−−→ Coker(dU ∩V ) → 0
 M  
dpM y
 (dp ,dp ) dp
y U V y U ∩V
f g
0 → Ker(dp+1 p+1 p+1
M ) −−−→ Ker(dU , dV ) −−−→ Ker(dp+1
U ∩V )
where the horizontal sequences are exact by the Snake lemma (cf. Lemma 2.27), and the
existence of the vertical arrows follow easily from the fact that d ◦ d = 0 (cf. Proposition
1.16).
We now apply the Snake Lemma again and we obtain a long exact sequence
δ
Ker(dpM ) → Ker(dpU , dpV ) → Ker(dpU ∪V ) −→ Coker(dpM ) → Coker(dpU , dpV ) → Coker(dpU ∩V )
It is easy to check that
Ker dpM : Coker(dM
p−1
) → Ker(dp+1 p

M ) = H (M )
and
Coker dpM : Coker(dp−1 p+1
= H p+1 (M ).

M ) → Ker(d M )
The same equalities hold for the cohomology groups of U, V and U ∩V . Thus, the Theorem
follows. 
Example 2.30 (Circle). Let M = S 1 be the circle and consider the cover
U = S 1 \ {(−1, 0)} V = S 1 \ {(1, 0)}.
Then, by Theorem 2.29 and Proposition 2.5, we have the long exact sequence
δ
0 → H 0 S 1 → H 0 (U )⊕H 0 (V ) → H 0 (U ∩V ) → H 1 S 1 → H 1 (U )⊕H 1 (V ) → H 1 (U ∩V ) → 0.
 

By Proposition 2.4, since S 1 , U and V are connected, have that


H 0 (S 1 ) = H 0 (U ) = H 0 (V ) = R.
26 DIFFERENTIAL TOPOLOGY

On the other hand, U ∩ V has two connected components, and Proposition 2.4 implies
H 0 (U ∩ V ) = R ⊕ R.
On the other hand, U and V are contractible and, Poincaré’s lemma (cf. Theorem 2.16)
implies
H 1 (U ) = H 1 (V ) = 0.
Similarly, U ∩ V is homotopy equivalent to the union of two points and Corollary 2.14
implies
H 1 (U ∩ V ) = 0.
Thus the long exact sequence becomes:
δ
0 → R → R ⊕ R → R ⊕ R → H 1 S 1 → 0 → 0 → 0.


Hence δ is surjective. Its kernel is moreover equal to the image of the map from H 0 (U )⊕
H (V ) to H 0 (U ∩ V ). But recall that this map is just the subtraction map, and since the
0

difference of two constant functions is a constant function, its image is the set of constant
functions, which is isomorphic to R. Therefore H 1 (S 1 ) = R.
Remark 2.31. In the previous example, instead of looking explicitly at the map from
H 0 (U ) ⊕ H 0 (V ) to H 0 (U ∩ V ), we could have used the following fact. If
f1 f2 f n−1
0 → C 1 −→ C 2 −→ . . . −→ C n → 0
is an exact sequence of vector spaces, then it is easy to check that
X n
(−1)i dim C i = 0.

i=1
The alternating sum is called the Euler Characteristic of the sequence.
Example 2.32 (Sphere). We want to show that

p n R if p = 0 or n
H (S ) =
0 otherwise
We proceed by induction on n. For n = 1 the claim is true by the previous example.
Assume now that the claim is true for S n . We now want to compute H p (S n+1 ). Let
U = S n+1 \ {N } and V = S n+1 \ {S} were N = (0, . . . , 0, 1) and S = (0, . . . , 0, −1). By
Theorem 2.29 and Proposition 2.5, we have the long exact sequence
δ δ
0 → H 0 (S n+1 ) → H 0 (U ) ⊕ H 0 (V ) → H 0 (U ∩ V ) → . . . → H p (S n+1 ) → H p (U ) ⊕ H p (V )
δ δ
→ H p (U ∩ V ) → · · · → H n (U ∩ V ) → H n+1 (S n+1 ) → H n+1 (U ) ⊕ H n+1 (V ) → H n+1 (U ∩ V ) → 0
Since S n+1 is connected, as in the previous example we have H 0 (S n+1 ) = R.
By the Stereographic projection, both U and V are diffeomorphic to Rn+1 . Thus, as in
the previous example, we have
H 0 (U ) = H 0 (V ) = R and H p (U ) = H p (V ) = 0 for p ≥ 1.
DIFFERENTIAL TOPOLOGY 27

On the other hand, it is easy to check that U ∩ V is homotopic equivalent to the sphere
S n . Thus, by induction and Corollary 2.14, we have

p R if p = 0 or n
H (U ∩ V ) =
0 otherwise.

Thus, the long exact sequence becomes


δ δ
0 → R → R ⊕ R → R → H 1 (S n+1 ) → 0 → · · · → 0 → H p (S n+1 ) → 0
δ δ
→ 0 → . . . 0 → R → H n+1 (S n+1 ) → 0 → 0 → 0
The claim then follows immediately.

Example 2.33 (Torus). Let T = S 1 × S 1 be the torus. Then



p 2
 R if p = 0 or 2
H T =
R ⊕ R if p = 1.

We leave the proof as an exercise.

Definition 2.34. Let M be a manifold. An open cover {Ui } of M is called a good cover
if any finite intersection Ui1 ∩ · · · ∩ Uik is either empty or contractible.

Lemma 2.35. Let M be a connected manifold of dimension n which admits a finite good
cover. Then H p (M ) is a finite dimensional vector space for all p ≥ 0.

Proof. Let {Ui }i=1,...,k be a finite good cover of M . We will prove the result by induction
over the number k of the open sets in the cover. If k = 1 then M is contractible and the
result holds by Poincaré Lemma (cf. Theorem 2.16).
Assume now that k > 1. Let
k−1
U = ∪i=1 Ui V = Uk .

Obviously, M = U ∪ V . Moreover, the set {Ui ∩ V }i=1,...,k−1 is a good cover for U ∩ V


and {Ui }i=1,...,k−1 is a cover of U . Thus, by induction, it follows that H p (U ), H p (V ) and
H p (U ∩ V ) are finite dimensional for all p. The result then follows immediately from
Theorem 2.29. 

It can be proven that any manfold admits a good cover1. Thus, the previous theorem
implies the following:

Theorem 2.36. The cohomology groups of a connected compact manifold are finite di-
mensional.

1For example, see [Pet16, pag. 386]


28 DIFFERENTIAL TOPOLOGY

2.4. Complactly supported de Rham cohomology. Let M be a manifold. We can


define de Rham cohomology for compactly supported forms in the same way we defined
the usual de Rham cohomology. Indeed if ω is a compactly supported form on M , then
dω is compactly supported as well and we can consider the subcomplex of the de Rham
cochain complex:
d d d
Ω0c (M ) → Ω1c (M ) → . . . → Ωnc (M ).
Note that for each p ≥ 0, Ωpc (M ) is a vector subspace of Ωp (M ).
Definition 2.37. The quotient
ker (d : Ωpc (M ) → Ωp+1
c (M )) Zcp (M )
Hcp (M ) := =:
Im d : Ωp−1 p
Bcp (M )

c (M ) → Ωc (M )

is called the p-th compactly supported de Rham cohomology group of M .


Clearly, if M is compact, then Hcp (M ) = H p (M ) for all p. On the other hand, we have:
Proposition 2.38. Let M be a connected non-compact manifold. Then Hc0 (M ) = 0.
Proof. As in the proof of Proposition 2.4, we have that Hc0 (M ) is represented by the class
of constant functions on M . On the other hand, if f : M → R is a constant function on
M with compact support, then f must be zero. Thus, the claim follows. 
Remark 2.39. Let f : M → N be a smooth map between non-compact manifolds and let
ω ∈ Ωpc (M ). Then
Supp(f ∗ ω) ⊂ f −1 (Supp(ω))
and f ∗ ω does not necessarily have compact support. Thus, we cannot in general define
the pull-back map between compactly supported de Rham cohomology groups, unless f is
a proper morphism, i.e. f −1 (K) is compact for any K ⊂ Y compact subset. In this case,
the pull-back of a compactly supported form on N is a compactly supported form on M
and the map
f ∗ : Hcp (N ) → Hcp (M )
is well defined. In particular, if M and N are diffeomorphic, then Hcp (M ) and Hcp (N ) are
isomorphic, for all p ≥ 0.

Proposition 2.38 implies that compactly supported cohomology groups are not homo-
topy invariant, unless we assume that the homotopy morphism is proper. More precisely,
we define (compare with Definition 2.9):
Definition 2.40. Let M0 and M1 be manifolds. Let f0 , f1 : M0 → M1 be smooth mor-
phisms. Then f0 and f1 are said to be properly (smoothly) homotopic if there exists
a smooth proper morphism
H : M0 × [0, 1] → M1 ,
such that
H(x, i) = fi (x) for any x ∈ M0 and i = 0, 1.
DIFFERENTIAL TOPOLOGY 29

The morphism H is called a properly smooth homotopy between M0 and M1 .


A smooth morphism f : M0 → M1 is called a properly smooth homotopy equiva-
lence if there exists a smooth map g : M1 → M0 such that
g ◦ f ∼ IdM0 and f ◦ g ∼ IdM1 .
Remark 2.41. Note that, in the definition above, it is not enough to assume that H(·, t) is
proper for all t ∈ [0, 1]. Indeed, it is easy to construct a smooth function H : R×[0, 1] → R
such that H(·, t) is proper for each t but H −1 (0) is not compact.
As in Corollary 2.14, we have
Proposition 2.42. Let f : M0 → M1 be a properly smooth homotopy equivalence.
Then, for each p ≥ 0, the pull-back map
f ∗ : Hcp (M1 ) → Hcp (M0 )
is an isomorphism.

Although we cannot always define the pull-back for compactly supported forms, one of
their interesting properties is that they can be pushed forward in the following way. Let
M be a manifold and U ⊂ M be an open set, with inclusion i : U ,→ M . A compactly
supported form on U can be extended to a compactly supported form on M by setting it
equal to zero on M \ U . The induced map will be denoted
i∗ : Ωc (U ) → Ωpc (M )
and it is called the push-forward map.
Lemma 2.43. Let U be an open subset of a manifold M with inclusion i : U ,→ M .
Then, for all p ≥ 0, the map i∗ : Ωpc (U ) → Ωpc (M ) commutes with the de Rham differ-
ential.
In particular, i∗ takes closed forms to closed forms and exact forms to exact forms,
inducing a linear map
i∗ : Hcp (U ) → Hcp (M ).
Proof. If ω is a compactly supported p-form on U then dω is a compactly supported
(p + 1)-form on U . Hence, by definition,

dω on U
i∗ dω =
0 on M \ U
and 
dω on U
di∗ ω =
d0 = 0 on M \ U
Thus, i∗ commutes with the de Rham differential. The rest of the Lemma just follow from
the definitions. 
30 DIFFERENTIAL TOPOLOGY

Proposition 2.44 (Punctured manifolds). Let M be a manifold of dimension n and let


p ≥ 2.
Then, for any point x ∈ M , the inclusion i : M \ {x} → M induces an isomorphism
i∗ : Hcp (M \ {x}) → Hcp (M ).
Moreover, if M is compact, then equality also holds for p = 1.
Proof. We first prove injectivity. Let ω be a closed form on M \ {x} and assume that
[i∗ ω] = 0, i.e. i∗ ω is exact. Thus, there exists η ∈ Ωp−1
c (M ) such that i∗ ω = dη. We want
to find a (p − 1)-form η 0 , which is cohomologous to η and whose support is contained
in M \ {x}. Note that, since the support of ω is a compact set contained in M \ {x},
there exists an open neighbourhood U of x such that i∗ ω = dη = 0 on U . By Poincaré
Lemma (cf. Corollary 2.17), after possibly shrinking U , we may assume that η is exact
on U . Thus, there exists a (p − 2)-form σ ∈ Ωp−2 (U ) such that η = dσ on U . Let
{fU , fM \{x} } be a partition of the unity with respect to the cover {U, M \ {x}} and let
η = η − d(fU · j∗ (σ)) where j : U → M is the inclusion. Then η = 0 in a neighbourhood of
x and, in particular, it induces a (p − 1)-form on Ωp−1 c (M \ {x}) such that dη = dη = ω.
It follows that also ω is exact, i.e. [ω] = 0. Thus, i∗ is injective.
Assume now that M is compact and that p = 1. Let ω be a closed 1-form on M \ {x}
and assume that [i∗ ω] = 0, i.e. i∗ ω is exact. As above, there exists a function η ∈ Ω0 (M )
such that i∗ ω = dη and there exists an open neighbourhood U of x such that i∗ ω = dη = 0
on U . Thus, η is constant on U , say η|U = c ∈ R and, since M is compact, the function
η = η −c is still compactly supported and η = 0 in U . Thus, η ∈ Ω0c (M \{x}) and dη = ω.
It follows that, also in this case, i∗ is injective.
Now we prove surjectivity for all p ≥ 1. Let [ω] ∈ Hcp (M ). As above, by Corollary
2.17, we may find a neighbourhood of x in M such that ω is exact on U , i.e. there
exists a (p − 1)-form η ∈ Ωp−1 (U ) such that ω = dη on U . As above, let j : U → M be
the immersion and let {fU , fM \{x} } be a partition of the unity with respect to the cover
{U, M \ {x}}. Let ω = ω − d(fU · j∗ η). Then [ω] = [ω] and the support of ω is contained
in M \ {x}. Thus, the claim follows. 
Example 2.45. We want to show that

R, if p = n
Hcp n
(R ) =
0, if p < n
If p = 0, then the claim follows from Proposition 2.38. Recall that Rn is diffeomorphic to
the punctured sphere S n \ {x} where x ∈ S n . Thus, by Proposition 2.44, for all p ≥ 1, we
have
Hcp (Rn ) = Hcp (S n \ {x}) = Hcp (S n ) = H p (S n ).
Thus, the claim follows from Example 2.32.
DIFFERENTIAL TOPOLOGY 31

Similarly to the Mayer-Vietoris sequence as in Section 2.3, there exists a compactly


supported version of Mayer-Vietoris sequences, defined with pushforwards instead of pull-
backs in the following way.
Let M be a manifold and let U and V be open sets of M such that M = U ∪ V and
U ∩ V = ∅. Consider again the inclusions
iU : U → M, iV : V → M, jU : U ∩ V → U, jV : U ∩ V → V
together with the corresponding push-forward:
iU∗ : Ωpc (U ) → Ωpc (M ), iV ∗ : Ωpc (V ) → Ωpc (M )
jU∗ : Ωpc (U ) ∩ Ωpc (V ) → Ωpc (U ), jV ∗ : Ωpc (U ∩ V ) → Ωpc (V ).
Similarly as in Proposition 2.28, we have the following:
Proposition 2.46. With the same notation as above, the sequence
i j
0 ← Ωpc (M ) ← Ωpc (U ) ⊕ Ωpc (V ) ← Ωpc (U ∩ V ) ← 0
is exact, where i := iU∗ + iV∗ and j := (−jU∗ , jV∗ ).
Note that due to the use of pushforwards instead of pullbacks, the arrow go in the
direction opposite to the direction of the usual Mayer-Vietoris sequence. This sequence
is called the compactly supported Mayer-Vietoris short exact sequence.
Proof. We first show injectivity of j. Let ω ∈ Ωpc (U ∩ V ) be such that j(ω) = 0. Then
jU∗ ω = jV∗ ω = 0 and, in particular ω = 0. Thus, j is injective.
We now show that Ker(i) = Im(j). It is easy to check that Im(j) ⊂ Ker(i). Now, let
ω = (ωU , ωV ) ∈ Ker(i). Then iU∗ ωU + iV∗ ωV = 0, which implies that the supports of ωU
and ωV are contained in U ∩ V . Thus, ω ∈ Im(g) and the claim follows.
Finally, we show that i is surjective. Let {fU , fV } be a partition of the unity with
respect to the cover {U, V }. Let ω ∈ Ωpc (M ) and define ωU := fU · ω|U and ωV = fV · ω|V .
Then we have
i∗ (ωU , ωV ) = fU ω|U + fV ω|V = (fU + fV ) ω = ω.
Thus, i is surjective. 

Since the push-forward commutes with the de Rham differential (cf. Lemma 2.43), we
obtain the following commutative diagram
j i
0 → Ωpc (U ∩ V ) −−−→ Ωpc (U ) ⊕ Ωpc (V ) −−−→ Ωpc (M ) → 0
  
 (d,d) 
dy y yd
j i
0 → Ωp+1 p+1 p+1 p+1
c (U ∩ V ) −−−→ Ωc (U ) ⊕ Ωc (V ) −−−→ Ωc (M ) → 0.

Using the Snake Lemma and a similar argument as in Theorem 2.29, we obtain:
32 DIFFERENTIAL TOPOLOGY

Theorem 2.47. With the same notation as above, for each p ≥ 0, there exists a linear
map
δ : Hcp (M ) → Hcp+1 (U ∩ V )
such that the following sequence is exact:
δ j i δ j
. . . → Hcp (U ∩ V ) → Hcp (U ) ⊕ Hcp (V ) → Hip (M ) → Hcp+1 (U ∩ V ) → . . .
2.5. Poincaré duality. We begin with the following
Proposition 2.48. Let M be a manifold. Then the cup product
∪ : H p (M ) × H q (M ) → H p+q (M ) ([ω], [η]) 7→ [ω ∧ η].
is well defined and it is such that, for any ω ∈ Ωp (M ) and η ∈ Ωq (M ) closed forms, we
have
[ω] ∪ [η] = (−1)p·q · [η] ∪ [ω].
Proof. We leave the first part of the Proposition as an exercise, whilst the second part of
the Proposition follows immediately from Proposition 1.6. 

Lemma 2.49. Let M be an oriented manifold of dimension n (without boundary).


Then the integration map
Z
n
IM : Hc (M ) → R, IM ([ω]) := ω
M

is a well-defined surjective linear map.


Proof. Consider the integration map on n-forms given by
Z
n
Ωc (M ) → R, I(ω) := ω
M
R
By Proposition 1.36, if ω is exact then M ω = 0. Hence the linear map IM is well defined.
Let ω ∈ ΩnM be a volume form on M , whose existence is guaranteed by Proposition 1.23.
Let fR: M → R≥0 be a smooth function with compact support on M . Then, f ·ω ∈ Ωnc (M )
and M f ω 6= 0. Since Ωn+1 (M ) = 0, it follows that f · ω is closed. It follows that IM is
surjective. 
R
Example 2.50. Let M = S n and let ω ∈ Ωn (M ) be such that M ω = 0. We want to
show that ω is exact. As in Example 2.32, we have that
Hcn (M ) = H n (M ) = R.
By Lemma 2.49, the map IM : Hcn (M ) → R is surjective and, therefore, it is an isomor-
phism. Thus, [ω] = 0 and the claim follows.
DIFFERENTIAL TOPOLOGY 33

We will see later that IM is actually always an isomorphism for oriented connected
manifolds without boundary.
Note that give two forms on a manifold M of dimension n, if at least one has compact
support, then their wedge product also has compact support. Therefore, for any p, q ≥ 0,
the cup product induces a pairing on compactly supported cohomology groups
∪ : H p (M ) × Hcq (M ) → Hcp+q (M ), [ω1 ] ∪ [ω2 ] := [ω1 ∧ ω2 ] .
If q = n − p then, by composing with the integration, we obtain a bilinear map
Z
p n−p
H (M ) × Hc (M ) → R, (ω1 , ω2 ) 7→ ω1 ∧ ω2 .
M

The map is called the Poincaré pairing on M .


Thus, we can define
µM : H p (M ) → Hcn−p (M )∗ = Hom Hcn−p (M ), R


by, for each [ω1 ] ∈ H p (M ),


Z
µM ([ω1 ]) : Hcn−p (M ) →R [ω2 ] 7→ ω1 ∧ ω2 .
M

We will show that µM is an isomorphism for any oriented connected manifold M without
boundary. We first consider the following example:
Example 2.51. Let U ⊂ Rn be an open subset which is diffeomorphic to Rn . We want to
show that µU is an isomorphism. By Poincaré Lemma (cf. Theorem 2.16), we have that
(
R if p = 0
H p (U ) =
0 otherwise.
As in Example 2.45, we have

R, if p = n
Hcp (U ) =
0, if p < n.
Thus, we just need to show that µU : H 0 (U ) → Hcn (U )∗ is an isomorphism. Note that
H 0 (U ) is generated by constant real functions. Thus, in this case, the Poincaré pairing is
just the multiplication by a constant function c, i.e.
Z
[ω] 7→ c · ω.
U

If c 6= 0, then this map is not zero. It follows that µU is injective and, therefore, an
isomorphism, as claimed.

Similarly to Lemma 2.35, we will assume that the manifold admits a finite cover whose
intersections are either empty or diffeomorphic to Rn (compare with Definition 2.34).
34 DIFFERENTIAL TOPOLOGY

Theorem 2.52 (Poincaré duality). Let M be an oriented manifold of dimension n without


boundary. Assume that M admits a finite cover {Ui }i=1,...,q such that any intersection
Ui1 ∩ · · · ∩ Uik is either empty or diffeomorphic to Rn .
Then the map
µM : H p (M ) → Hcn−p (M )∗
is an isomorphism for every p ≥ 0.
Note that the Theorem holds without assuming the existence of such a cover. Moreover,
as in the case of good covers, it is possible to show that any compact manifold admits a
cover as in the assumption of the Theorem.
We first recall two results from linear algebra and homological algebra.
Lemma 2.53. Let
f1 f2
C 1 −→ C 2 → C 3
be an exact sequence sequence of finite dimensional vector spaces.
Then there exists an induced exact sequence between the dual vector spaces:
f 2∗ f 1∗
C 3∗ −→ C 2∗ → C 1∗ .
Proof. Recall that the dual vector space of C i is C i∗ := Hom(C i , R) and the induced
map f i∗ : C (i+1)∗ → C i∗ is defined by f i∗ (φ) := φ ◦ f i . We want to show that Im(f 2∗ ) =
Ker(f 1∗ ).
Let φ ∈ Im(f 2∗ ). Then there exists ψ ∈ C 3∗ such that f 2∗ (ψ) = φ. Hence, since
Im(f 1 ) ⊂ Ker(f 2 ), we have
f 1∗ (φ) = f 1∗ f 2∗ (ψ) = ψ ◦ f 2 ◦ f 1 = 0.


Thus, φ ∈ Ker(f 1∗ ).
Now let φ ∈ Ker(f 1∗ ). Then φ ◦ f 1 = 0. It follows that
Ker(f 2 ) = Im(f 1 ) ⊂ Ker(φ).
Thus, the induced map φ : C 2 /Ker(f 2 ) → R is well-defined. Since C 2 /Ker(f 2 ) ' Im(f 2 ) ⊂
C 3 , there exists a linear map ψ : C 3 → R such that ψ ◦ f 2 = φ, i.e. f 2∗ (ψ) = φ, and, in
particular, φ ∈ Im(f 2∗ ) as claimed. 
Lemma 2.54 (Five Lemma). Let
f1 f2 f3 f4
C 1 −−−→ C 2 −−−→ C 3 −−−→ C 4 −−−→ C 5
    
α1 y α2 y α3 y α4 y α5 y
    

g1 g2 g3 g4
D1 −−−→ D2 −−−→ D3 −−−→ D4 −−−→ D5
be a commutative diagram of vector spaces, where the horizontal lines are exact. Assume
that
(1) α1 is surjective,
(2) α2 and α4 are isomorphisms, and
DIFFERENTIAL TOPOLOGY 35

(3) α5 is injective.
Then α3 is an isomorphism.
Proof. We show that α3 is injective. The proof of surjectivity is similar and it is left as
an exercise.
Assume that x ∈ C 3 is such that α3 (x) = 0. Then
α4 ◦ f 3 (x) = g 3 ◦ α3 (x) = 0.
Since α4 is an isomorphism, it follows that f 3 (x) = 0. Thus, by exactness, there exists
y ∈ C 2 such that f 2 (y) = x. We have
0 = α3 (x) = α3 ◦ f 2 (y) = g 2 ◦ α2 (y).
Thus, by exactness, there exists z ∈ D1 such that g 1 (z) = α2 (y). Since α1 is surjective,
there exists w ∈ C 1 such that α1 (w) = z.
It follows that
α2 ◦ f 1 (w) = g 1 ◦ α1 (w) = g 1 (z) = α2 (y)
and since α2 is an isomorphism, it follows that f 1 (w) = y. Thus,
x = f 2 (y) = f 2 ◦ f 1 (w) = 0,
which implies that α3 is injective. 

We now proceed with the proof of Poincaré duality.


Proof of Theorem 2.52. We first assume that M = U ∪V where U and V are open subsets
of M such that U , V and U ∩ V are diffeomorphic to Rn . By Theorem 2.29, the Mayer-
Vietoris sequence is exact:
δ f g δ f
. . . → H p (M ) → H p (U ) ⊕ H p (V ) → H p (U ∩ V ) → H p+1 (M ) → . . . .
Similarly, for the compact supported cohomology, by Theorem 2.47 we have the following
Mayer-Vietoris exact sequence
δ j i δ j
c
... → Hcn−p (U ∩ V ) → Hcn−p (U ) ⊕ Hcn−p (V ) → Hcn−p (M ) →
c
Hcn−p+1 (U ∩ V ) → . . .
By Lemma 2.53, by taking the dual of this sequence, we obtain
δ∗ i∗ j∗ δ∗ i∗
... →c
Hcn−p (M )∗ → Hcn−p (U )∗ ⊕ Hcn−p (V )∗ → Hcn−p (U ∩ V )∗ →
c
Hcn−p+1 (M )∗ → . . . .
We claim that the Poincaré pairing µ yields a commutative diagram
g δ f
H p−1 (U ) ⊕ H p−1 (V ) −−−−→ H p−1 (U ∩ V ) −−−−→ H p (M ) −−−−→ H p (U ) ⊕ H p (V )
   
np−1 µU ⊕µV y
 np−1 µU ∩V 
y np µM 
y np µU ⊕µV y

n−(p−1) n−(p−1) j∗ n−(p−1) δ∗ i∗


Hc (U )∗ ⊕ Hc (V )∗ −−−−→ Hc (U ∩ V )∗ −−−c−→ Hcn−p (M )∗ −−−−→ Hcn−p (U )∗ ⊕ Hcn−p (V )∗
where n0 = 1 and np = (−1)p−1 · np−1 .
By assumption, the open sets U , V and U ∩ V are diffeomorphic to Rn . Thus, as in
Example 2.51, we have that the maps µU ⊕ µV and µU ∩V are isomorphisms. Thus, in
36 DIFFERENTIAL TOPOLOGY

order to show that µM is an isomorphism, by the Five Lemma 2.54, it is enough to show
that the diagram is commutative.
We will prove that the diagram is commutative in general, without our assumption on
the open set U and V . We begin by considering the first square. Let ωU ∈ Ωp−1 (U ) and
ωV ∈ Ωp−1 (V ) be closed form. We want to show that
µU ∩V ◦ g([ωU ], [ωV ]) = j ∗ (µU ([ωU ]), µV ([ωV ]))
n−(p−1)
as elements in Hc (U ∩ V )∗ .
n−(p−1)
Thus, for any closed form η ∈ Ωc (U ∩ V ), we want to show that
µU ∩V ◦ g([ωU ], [ωV ])([η]) = j ∗ (µU ([ωU ]), µV ([ωV ]))([η]).
Since j := (−jU∗ , jV∗ ) (cf. Proposition 2.46), by definition of µ, this is equivalent to prove
Z Z Z
g (ωU , ωV ) ∧ η = − ωU ∧ jU∗ η + ωV ∧ jV∗ η.
U ∩V U V
Recall that g = jV∗ − jU∗ (cf. Section 2.3). Thus, the equality holds and the first square is
commutative.
We now consider the second square. To this end, we have to study explicitly the linear
maps δ and δc , whose existence is guaranteed by the Snake Lemma (cf. Lemma 2.27).
Let ω ∈ Ωp (M ) be a closed form and let {fU , fV } be a partition of the unity with respect
to the cover {U, V }. Let
ωU = fU · ω|U and ωV = fV · ω|V .
Then (ωU , ωV ) ∈ Ωpc (U ) ⊕ Ωpc (V ) and
i(ωU , ωV ) = ω where i := iU∗ + iV∗
as in Proposition 2.46. Since ω is closed, Lemma 2.43 implies that
i(dωU , dωV ) = 0.
Thus,
(dωU , dωV ) ∈ Ker(i) ⊂ Ωp+1 p+1
c (U ) ⊕ Ωc (V ).
By Proposition 2.46, it follows that Ker(i) = Im(j) and that j is injective. Thus, there
exists a unique element δc (ω) ∈ Ωp+1
c (U ∩ V ) such that j(δc (ω)) = (dωU , dωV ). Note that

dfU = d(fU + fV ) − dfV = −dfV .


Thus,
j(δc ω) = (dωU , dωV ) = (dfU ∧ ω|U , dfV ∧ ω|V )
= (−dfV ∧ ω|U , dfV ∧ ω|V )
= j(dfV ∧ ω|U ∩V ).
Thus, the connecting linear map
δc : Ωpc (M ) → Ωp+1
c (U ∩ V )
DIFFERENTIAL TOPOLOGY 37

is given by
δc (ω) = dfV ∧ ω|U ∩V = −dfU ∧ ω|U ∩V .
In particular, for any form η we have
δc (dη) = dfV ∧ dη|U ∩V = −d(dfV ∧ η|U ∩V ) = −dδc η.
Hence, δc maps closed forms to closed forms and exact forms to exact forms. Thus, it
induces a linear map
δc : Hcp (M ) → Hcp+1 (U ∩ V ).
Similarly, we can show that the connecting linear maps in the usual Mayer-Vietoris long
exact sequence is given by

p p+1 −dfU ∧ ω (= dfV ∧ ω) , on U ∩ V
δ : H (U ∩ V ) → H (M ) δ(ω) =
0 on M \ U ∩ V.
We can now show the commutativity of the second square. Let ω1 ∈ Ωp−1 (U ∩ V ) be a
closed form. We want to show that
µM ◦ δ([ω1 ]) = (−1)p−1 δc∗ ◦ µU ∩V ([ω1 ]).
This is equivalent to show that, for any closed form ω2 ∈ Ωcn−p (M ), we have
Z Z
p−1
δω1 ∧ ω2 = (−1) ω1 ∧ δc ω2 .
M U ∩V
Using the explicit description of the connecting homomorphisms discussed above, we get
Z Z Z Z
p−1 p−1
δω1 ∧ ω2 = dfV ∧ ω1 ∧ ω2 = (−1) ω1 ∧ dfV ∧ ω2 = (−1) ω1 ∧ δc ω2
M U ∩V U ∩V U ∩V
which shows the results.
We now show the commutativity of the third square. Let ω ∈ Ωp (M ) be a closed form.
Then, similarly as above, for any closed form ηU ∈ Ωn−p (U ) and ηV ∈ Ωn−p (V ), we have
Z Z Z
ω|U ∧ η + ω|V ∧ η = ω ∧ i (ηU , ηV )
U V M
Thus, the claim follows.
It follows that the Theorem holds under the assumption that q = 2, i.e. the open
covering of M is given by two open subsets of M . In general, we proceed by induction.
If q = 1 then M is diffeomorphic to Rn and the result follows immediately as in Example
2.51. If q > 1, then, similarly to the proof of Lemma 2.35, we define
U := U1 ∪ · · · ∪ Uq−1 and V := Uq .
Clearly M = U ∪ V and by induction, we may assume that µU , µV and µU ∩V are isomor-
phisms. By using the same diagram as above, the Five Lemma (cf. Lemma 2.54) implies
the Theorem. 

As an immediate consequence of Poincaré’s Theorem and Proposition 2.44, we obtain:


38 DIFFERENTIAL TOPOLOGY

Corollary 2.55. Let M be a connected compact oriented manifold without boundary of


dimension n.
Then H n (M ) = R. Moreover, for any x ∈ M , we have Hcn (M \ {x}) = R and
Hc (M \ {x}) = H p (M ) for all p < n.
p

Poincaré Duality allows to deduce informations about the Euler characteristic of M:


Definition 2.56. Let M be a manifold. The number
X X
χ(M ) := (−1)p dim H p (M ) = (−1)p bp (M )
p p

is called the Euler characteristic of M .


Corollary 2.57. Let M be a connected compact oriented manifold without boundary of
dimension n. Assume that n is odd.
Then χ(M ) = 0.
Proof. Since M is compact, Poincaré duality implies that bp (M ) = bn−p (M ) for all p.
Thus, the claim follows easily. 

2.6. Degree of a morphism. Let M and N be connected oriented manifolds of dimen-


sion n. Let f : M → N be a smooth proper morphism. As in Remark 2.39, for all p ≥ 0,
we can consider the pull-back
f ∗ : Hcp (N ) → Hcp (M ).
In particular, if p = n, then by Poincaré duality (cf. Theorem 2.52) and Proposition 2.4,
we have
Hcn (M ) = Hcn (N ) = R.
Thus f ∗ induces a linear map R → R which is of the form v 7→ c · v for some constant
c ∈ R, called the degree of f . We will denote it by deg f .
Thus, for all ω ∈ Ωnc (N ), we have
Z Z

f ω = deg f · ω.
M N

Proposition 2.58. Let M , N and P be connected oriented manifolds of dimension n.


(1) Let f : M → N and g : N → P be proper smooth morphisms. Then
deg(g ◦ f ) = deg f · deg g.
(2) Let f : M → M be a diffeomorphism. Then

1 if f is orientation preserving
deg f =
−1 if f is orientation reversing
(3) If f, g : M → N are properly homotopic smooth morphisms then deg f = deg g.
DIFFERENTIAL TOPOLOGY 39

Proof. (1) follows from Proposition 1.8. (2) follows from Proposition 1.27. (3) follows
from Proposition 2.42. 
Theorem 2.59 (Mapping degree theorem). The degree deg f of a proper smooth surjective
morphism f : M → N between connected oriented manifolds is an integer.
Before proving the result, we recall some definitions and results we need.
Definition 2.60. Let f : M → N be a smooth morphism between manifolds. A regular
value of f is a point y ∈ N such that for each x ∈ f −1 (y), the differential Dfx has
maximal rank.
Theorem 2.61 (Preimage Theorem). Let f : M → N be a smooth morphism between
manifolds and let y ∈ f (M ) be a regular value.
Then f −1 (y) is a manifold of dimension dim M − rkDfx , where x ∈ f −1 (y).
Theorem 2.62 (Implicit Function Theorem). Let f : M → N be a smooth morphism
between manifolds. Let x ∈ M and assume that DFx is an isomorphism.
Then there exists an open neighbourhood U of x such that f |U : U → f (U ) is a diffeo-
morphism.
Theorem 2.63 (Sard’s Lemma). Let f : M → N be a smooth morphism between mani-
folds.
Then the subset Z ⊂ N of regular values for f is such that Z ∩ f (M ) is dense in f (M ).

We now proceed with the proof of Theorem 2.59.


Proof of Theorem 2.59. By Sard’s Lemma, since dim M = dim N , we may assume that
there exists a regular value y ∈ N such that Dfx has rank n for every x ∈ f −1 (y), as
otherwise deg f = 0 and there is nothing to prove. Since f is proper, it follows that f −1 (y)
is compact. By the Preimage Theorem it follows that f −1 (y) is a finite set, say
f −1 (y) = {x1 , . . . , xk }.
By the Implicit Function Theorem, there exists a connected open neighbourhood U of y
such that f −1 (U ) is the disjoint union of open subsets Ui 3 xi such that f |Ui : Ui → U is
a diffeomorphism. R
Let ω ∈ Ωnc (N ) be a form supported in U and such that N ω = 1. Then, as in Section
1.4, we have Z Z

(f |Ui ) ω = sgn (det Dfxi ) ω = sgn (det Dfxi ) .
Ui U
Since ω is supported in U , we have that f ω is supported in f −1 (U ) = ∪Ui . Thus,

Z Z Xk Z k
X
∗ ∗ ∗
f ω= f ω= (f |Ui ) ω = sgn (det Dfxi ) ∈ Z.
M f −1 (U ) i=1 Ui i=1
It follows that the degree of f is an integer. Note that it does not depend on the choice
of the regular value y ∈ N . 
40 DIFFERENTIAL TOPOLOGY

Example 2.64. Let


A : Sn → Sn x 7→ −x
be the antipodal map. We want to show that
deg A = (−1)n+1 .
Let i : S n → Rn+1 be the inclusion and let
ω̃ := x1 dx2 ∧ · · · ∧ dxn+1 ∈ Ωn (Rn+1 ).
Let ω := i∗ ω̃ ∈ Ωn (S n ). By Stokes’ Theorem (cf. Theorem 1.32), we have
Z Z Z Z

ω= i ω̃ = dω̃ = dx1 ∧ dx2 ∧ · · · ∧ dxn 6= 0
Sn Sn Dn+1 Dn+1

Let à : Rn+1 → Rn+1 be the extension of A to Rn+1 . Then à ◦ i = i ◦ A and Ã∗ ω̃ =


(−1)n+1 ω̃. Thus,
A∗ ω = A∗ i∗ ω̃ = (i ◦ A)∗ ω̃ = (Ã ◦ i)∗ ω̃ = i∗ Ã∗ ω̃ = (−1)n+1 i∗ ω̃ = (−1)n+1 ω.
Thus, Z Z Z
∗ n+1
deg A ω= A ω = (−1) ω,
Sn Sn Sn
and the claim follows.

3. Morse Theory
3.1. Introduction.
Definition 3.1. Let M be a manifold of dimension n and let f : M → R be a smooth
function. A point x ∈ M is called a critical point of f if Dfx = 0, or equivalently,
given local coordinates x1 , . . . , xn in a neighbourhood U of x, then

f (x) = 0 for i = 1, . . . , n
∂xi
A critical value is the image of a critical point.
A critical point x ∈ M is called non-degenerate if the Hessian matrix
∂2
 
Hf := f
∂xi ∂xj i,j=1,...,n

is invertible at x.
A Morse function on M is a function f : M → R such that all the critical points of
f are non-degenerate.
For any h ∈ R, we denote by S h the sublevel set
S t := {x ∈ M | f (x) ≤ h}.
Let Dn := {x ∈ Rn | |x| ≤ 1} denote the unit disc, so that ∂Dn = S n−1 and int(Dn ) =
Dn \ ∂Dn is the open disc.
DIFFERENTIAL TOPOLOGY 41

Definition 3.2. An n-cell (or cell of dimension n) is a topological space homeomorphic


to the open disk int(Dn ). A cell decomposition of a topological space M is a family
F = {ei }i∈I of pairwise disjoint subspaces of M such that

• each ei is a cell;
• the union tI ei = M .

If I is finite, then M is called a finite cell decomposition. The m-skeleton of M is


the subspace

G
skm (M ) := ei .
dim ei ≤m

Example 3.3. The circle S 1 can be thought as S 1 = e0 t e1 where e0 is a point of S 1 and


e1 is a 1-cell.

Notation 3.4. Let M be a topological space and let f∂ : S n−1 → M be a continuous map.
We consider the quotient

M ∪f∂ Dn := M t Dn / ∼

where ∼ is the relation given by

x ∼ f∂ (x) for all x ∈ ∂Dn .

We refer to M ∪f∂ Dn as the space obtained from M by attaching and n-cell and f∂ is
the attaching map.

Example 3.5. Let T = S 1 × S 1 be the torus embedded in R3 and resting upright on the
horizontal xy-plane. Consider the height function

f: T →R (x, y, z) 7→ z.

For each h ∈ R, we consider the sublevel set

S h := f −1 ((−∞, h]).
42 DIFFERENTIAL TOPOLOGY

The critical values for f are a, b, c and d as in the picture. We have


1) If h < a then S1 := S h = ∅;
2) if h ∈ (a, b) then S2 := S h is homotopic equivalent to a point, i.e. a 0-cell.
3) if h ∈ (b, c) then S3 := S h is a cylinder, which is homotopic equivalent to the space
obtained attaching a 1-cell to S2 .
4) If h ∈ (c, d) then S4 := S h is homotopic equivalent to the space obtained attaching
a 1-cell to S3 .
5) If h > d then S5 := S h = T and it can be obtained by attaching a 2-cell to S4 .
Definition 3.6. Let M be a manifold and let f : M → R be a smooth function. For any
non-degenerate critical point x ∈ M , let Eig− Hf (x) be the space spanned by eigenvectors
with negative eigenvalues for the Hessian of f at x. The index of f at x is the dimension
of Eig− Hf (x).
We omit the proof of the following:
Lemma 3.7 (Morse Lemma). Let M be a manifold and let f : M → R be a smooth
function. Let x0 ∈ X be a non-degenerate critical point of index λ.
Then there exist coordinates x1 , . . . , xn locally around x such that x0 = (0, . . . , 0) and
λ
X n
X
f (x) = f (x0 ) − x2i + x2i .
i=1 i=λ+1

Note that, in particular, Morse Lemma implies that non-degenerate critical points of a
smooth function f : M → R are isolated.
Example 3.8. Let f : T → R be as in the previous example. Then the critical points
a, b, c and d have index 0, 1, 1 and 2 respectively.
DIFFERENTIAL TOPOLOGY 43

3.2. CW complexes.
Definition 3.9. A topological space M is said to have a CW-structure if there are
subspaces
[
M (0) ⊆ M (1) ⊆ · · · ⊆ M = M (n)
n∈Z+
such that
(1) M (0) is a discrete set of points;
(2) M (n+1) is obtained from M (n) by attaching (n + 1)-cells for all n ≥ 0;
(3) a subset V ⊂ M is closed if and only if V ∩ M (n) is closed for all n ≥ 0.
Such a topological space is called a CW complex and the subspace M (n) is the n-skeleton
of M .
A finite complex is a CW complex with only finitely many cells.
Note that, for a finite complex, condition (3) is redundant.
An attaching map f∂ : S n−1 → M (n−1) extends to a map f : Dn → M (n) called the
characteristic map. The image of Dn under f is called a closed cell and the image of
int(Dn ) under f is called an open cell. Note that the open cell is open in M (n) but not
necessarily in M .
A subcomplex S of M is a closed subspace of M which is a union of cells. It is clear
that S is then itself a CW-complex.
Example 3.10.
(1) We can think of Rn as a CW complex obtained as the union of n-cubes whose
vertices have integer coefficients, so that the 0-cells are the the integer points, the
1-cells are the edges, etc...
(2) The sphere S n is a CW-complex consisting of a point, as a 0-cell and a n-cell.
(3) If n 6= 4, any manifold of dimension n is a CW complex. It is still unknown if
manifolds of dimension 4 all admit a CW complex structure.
Proposition 3.11. Let M be a CW complex. Then
(1) if K ⊂ M is a compact subset, then K is contained in a finite union of open cells;
(2) the closure of every cell of M is contained in a finite subcomplex of M .
Proof. We first prove (1). Let K ⊂ M be a compact subset. We want to show that K
only intersects finitely many cells of M . Assume by contradiction that there is an infinite
sequence of points S = {xj } ⊂ K all lying in distinct cells. We claim that S ∩ M (n) is
closed and discrete for all n ≥ 0. We proceed by induction on n. For n = 0, this follows
from the fact that M (0) is closed and discrete. Assume now that S ∩ M (n) is closed and
discrete. Then, if {ei }I are the (n+1)-cells, then the open cell corresponding to ei contains
at most one xj ∈ S. Thus S ∩ (∪i ei ) is closed and discrete. It follows that S ∩ M (n+1) is
closed and discrete, as claimed. Since S ⊂ K, it follows that S is finite, a contradiction.
44 DIFFERENTIAL TOPOLOGY

We now prove (2). To this end, we proceed by induction on the dimension n of the
cell. For n = 0, the result is clear. Assume now that the result is true for any m-cell with
m < n and let en be an n-cell. In particular, the border K of en is the image of S n−1 and
it is compact. Hence, it is contained in a finite union of open cells of dimension smaller
than n by (1). By induction, each of these cells is contained in a finite subcomplex. The
union of these subcomplexes is a finite subcomplex containing K. Hence attaching en
results in a finite subcomplex containing en . 
Corollary 3.12. Let M be a CW complex. Then any compact subset of M is contained
in a finite subcomplex.
Proof. Since a finite union of finite subcomplex is again a finite subcomplex, the result
follows immediately from the previous Proposition. 
3.3. CW-structure associated to a Morse function. We begin by recalling some
basic notions from differential geometry, which we will need later.
Definition 3.13. Let M be a manifold. A flow on M is a smooth one-parameter group
of diffeomorphisms φt : M → M , i.e. a smooth map
φ: R × M → M
such that, if φt (x) := φ(t, x), then
(1) φ0 = IdM ,
(2) for each t ∈ R, the function φt is a diffeomorphism, and
(3) for each t, s ∈ R, we have φt+s = φt ◦ φs .
In particular, if we fix x ∈ M , the map γx := φ(·, x) : R → M is called a flow line (or
integral curve).
For any x ∈ M , the flow line φ(·, x) passes through x and the tangent vector
d
γx (0) ∈ Tx M
dt
is called velocity vector.
Note that a flow induces a vector field on the manifold, i.e. a section of the tangent
bundle. Conversely, we have:
Lemma 3.14. Let M be a manifold and let X a smooth compactly supported vector field
on M .
Then X generates a unique one-parameter group of diffeomorphisms φt : M → M such
that, for all x ∈ M , we have
d
X ◦ γx (t) = γx (t).
dt
It can be shown that two distinct flow lines are disjoint. Thus, the manifold M decom-
poses into a disjoint union of flow lines.
DIFFERENTIAL TOPOLOGY 45

Definition 3.15. A Riemannian metric g on a manifold M is a family of positive


definite inner products gx : Tx M × Tx M → R, with x ∈ M , such that for any vector field
X and Y , we have that the function
M →R x 7→ gx (Xx , Yx )
is smooth.
A manifold M with a Riemannian metric g is called a Riemannian manifold and it
is denoted (M, g). It can be shown that any manifold admits a Riemannian metric. Since
a positive definite inner product on a vector space V defines an isomorphism of V with
its dual V ∗ , a Riemannian metric defines an isomorphism between the tangent and the
cotangent bundle of M . In particular, we can define:
Definition 3.16. Let (M, g) be a Riemannian manifold and let f : M → R be a smooth
function. The gradient vector field of f with respect to g is the unique smooth vector
field ∇f such that for any vector field X on M , we have
g(∇f, X) = Df (X) = X(f ),
(recall that X(f ) denotes the directional derivates of f along the vector field X).
In particular, we have
k∇f k2 = g(∇f, ∇f ) = Df (∇f )
Note that ∇f vanishes exactly at the critical points of f . Moreover, we can check that is
always orthogonal to the level sets f −1 (c) for all c ∈ R.
Lemma 3.14 implies that there exists a local flow φt : M → M generated by −∇f , i.e.
such that, for all x ∈ M , if as above we denote γx (t) = φt (x), then
d
γx (0) = (−∇f ) ◦ γx (0) γx (0) = x.
dt
The flow φt is called the gradient flow of f and the curves γt (x) are called the gradient
flow lines (or just gradient lines).
Proposition 3.17. Let M be a manifold and let f : M → R be a smooth function.
Then f decreases along its gradient lines.
Proof. Let γx (t) be the gradient line at x. Then
d d
f (γx (t)) = Dfγx (t) ( γx (t))
dt dt
= Dfγx (t) ((−∇f )(γx (t)))
= −k∇f (γx (t))k2 ≤ 0.
Thus, the claim follows. 
46 DIFFERENTIAL TOPOLOGY

Proposition 3.18. Let M be a compact manifold and let f : M → R be a Morse function.


Then, for every x ∈ M , the gradient flow line γx (t) begins and ends at a critical point,
i.e. the limits
lim γx (t) and lim γx (t)
t→−∞ t→∞

both exist and are critical points.


Proof. We first prove that if these limits exist then they are critical points. Since M is
compact, Lemma 3.14 implies that the gradient flow line γx (t) is defined for all t ∈ R.
Moreover, for every x ∈ M , the image of f (γx (t)) is a bounded set in R. Thus, as in the
proof of Proposition 3.17, since f (γx (t)) is decreasing, we have
d
(4) − lim k∇f (γx (t)k2 = lim f (γx (t)) = 0,
t→±∞ t→±∞ dt

and the claim follows.


Thus, we just need to check that the limits exist. Let U be the union of small disjoint
open balls around the critical points (recall that, by Lemma 3.7 the critical points of a
Morse function are discrete). Since M is compact, it follows that M \ U is also compact.
Therefore, for all x ∈ M , since there are no critical points inside M \ U , the function
k∇f (γx (t))k2 is bounded from below by a positive constant. By (4), it follows that for all
t sufficiently large, γx (t) ∈ U . Since the balls are disjoints and γx (t) is continuous, there
exists a critical point x0 ∈ M such that for any open ball around x0 , γx (t) is in that ball
for sufficiently large t and therefore limt→∞ γx (t) = x0 . Similarly limt→−∞ γx (t) exists and
it is a critical point. 

Definition 3.19. Let X be a topological space and let S ⊂ X be a subspace. A defor-


mation retraction of X onto S is a continuous map F : X × [0, 1] → X such that for
every x ∈ X and s ∈ S, we have
F (x, 0) = x F (x, 1) ∈ S and F (s, 1) = s
In particular, S is homotopy equivalent to M .
Theorem 3.20 (First Fundamental Theorem of Morse Theory). Let M be a manifold
and let f : M → R be a Morse function. Let a < b and suppose that f −1 ([a, b]) is compact
and contains no critical points. Let
Sh := f −1 ((−∞, h)) = {x ∈ M | f (x) ≤ h}.
Then Sa is diffeomorphic to Sb and Sb is a deformation retract of Sb .
The idea is to let the level set f −1 (b) flow down to the level set f −1 (a) along the gradient
flow lines orthogonal to the level sets of f .
DIFFERENTIAL TOPOLOGY 47

Proof. By Lemma 3.7, the critical points of f are isolated. Thus, since f −1 ([a, b]) does not
contain any critical points, it follows that for every  > 0 small enough, f −1 ([a − , b + ])
does not contain any critical points either. Define a smooth function ρ such that
, on f −1 [a, b]
 1
ρ= k∇f k2 .
0, on M \ f −1 ([a − ε, b + ε])
Then the support of ρ is compact and contained in f −1 ([a − , b + ]). Fix a Riemannian
metric g on M and define the vector field
X(x) = −ρ(x)∇f (x)
for all x ∈ M . Thus, by Lemma 3.14, it generates a flow φt on f −1 ([a − , b + ]). Let
γx (t) := φt (x). Then
d d
f (γx (t)) = Dfγx (t) ( γx (t))
dt dt
d
= g(∇f (γx (t)), γx (t))
dt
= g(∇f (γx (t)), −ρ∇f (γx (t))
= −ρ(γx (t))k∇f (γx (t))k2 .
Thus, it follows from the definition of ρ that
d
f (γx (t)) = −1
dt
for all t such that γx (t) ∈ f −1 ([a, b]). Moreover, since ρ ≥ 0, we have that
d
f (γx (t)) ≤ 0
dt
for all t ∈ R. Thus f (γx (t)) is decreasing.
By the fundamental theorem of calculus, we have that if f (γx (s)) ∈ [a, b] for all s ∈ [0, t]
then
Z t
d
f (γx (t)) − f (γx (0)) = f (γx (s))ds = −t.
0 ds

Since γx (0) = x, we have


f (γx (t)) = f (x) − t.
Thus, by taking t = b − a, we obtain:
(1) If f (x) ≤ b then f (φb−a (x)) ≤ a.
(2) if f (x) > b then f (φb−a (x)) > a.
(1) implies that φb−a maps S b to S a and, similarly, (2) implies that φa−b maps S a to S b .
Thus, it follows that φb−a is a diffeomorphism between S b and S a . For any t ∈ [0, 1] define
48 DIFFERENTIAL TOPOLOGY

Ft : S b → S a such that
(
x if f (x) ≤ a
Ft (x) =
φt(f (x)−a) (x) if a ≤ f (x) ≤ b.
Then, it is easy to check that F is a deformation retract and the theorem follows. 
Corollary 3.21 (Reeb’s Theorem). Let M be a compact manifold of dimension n without
boundary and let f : M → R be a Morse function admitting only two critical points.
Then M is homeomorphic to a sphere S n .
Proof. Since M is compact, f admits a maximum at xmax and a minimum at xmin and these
are the two critical points. Let hmax := f (xmax ) and hmin := f (xmin ). Then λ(xmax ) = n
and λ(xmin ) = 0. Morse Lemma (cf Lemma 3.7) implies that there exist local coordinates
x1 , . . . , xn in an open neighbourhood Umin of xmin such that
Xn
f = hmin + x2i .
i=1

For some a > hmin sufficiently close to hmin , we have


f −1 ([hmin , a]) = (x1 , . . . , xn ) |x21 + · · · + x2n ≤ a − hmin .


Hence f −1 ([hmin , a]) = f −1 ((−∞, a]) = S a is a closed n-cell Dn− . Similarly, for b < hmax ,
we have that f −1 ([b, hmax ]) is a closed n-cell Dn+ . By Theorem 3.20, S a is diffeomorphic
to S b , since there are no critical points between a and b. Hence M = f −1 ([hmin , hmax ]) is
the union of two n-cells attached at their common boundary S n−1 = ∂Dn+ = ∂Dn− .
Now we need to construct an homeomorphism between M and S n . The (closed) north-
ern hemisphere H + is diffeomorphic to Dn+ and the (closed) southern hemisphere H − to
Dn− . The only problem is that the two n-cells are not necessarily glued by the identity
map, but by a homeomorphism f : ∂H − → ∂H + . Let φ± : H ± → Dn± be the two home-
omorphisms and E = H + ∩ H − the equator so that φ− |E = φ+ |E ◦ f . We claim that
f : E → E extends to a homeomorphism F : H + → H + such that F |E = f . Assuming
the claim, we can define a homeomorphism φ̃ : S n → M as follows: we have
φ̃|H + = φ+ ◦ F and φ̃H − = φ− .
This is well-defined since
(φ+ ◦ F )|E = φ+ |E ◦ f = φ− |E .
It is continuous and bijective since it is restricted to each of Dn± . Thus, it is a homeomor-
phism.
It remains to prove the claim. Since H − is homeomorphic to the unit n-ball Dn it is
enough to show that every homeomorphism g : ∂Dn → ∂Dn extends to a homeomorphism
G : Dn → Dn . This is done via the Alexander trick, using concentric spheres: for v ∈ ∂D,
the unit (n − 1)-sphere, considered as a vector in Rn , we define G(tv) = tg(v) for all
DIFFERENTIAL TOPOLOGY 49

0 ≤ t ≤ 1. This is continuous and the same argument shows that g −1 extends to G−1 .
Thus, it is a homeomorphism, as claimed. 
Theorem 3.22 (Second Fundamental Theorem of Morse Theory). Let M be a manifold
of dimension n and let f : M → R be a smooth function with a non-degenerate critical
point x0 ∈ M of index λ. Let c = f (x0 ) and assume that f −1 [c − , c + ] is compact and
does not contain any critical point of f other than x0 , for some  > 0.
Then, if  is sufficiently small, S c+ is homotopy equivalent to S c− with a λ-cell at-
tached.
Proof. The difficult part of the Theorem is to prove it for critical points of index λ ∈
{1, . . . , n − 1}. Otherwise x0 is either a minimum or a maximum of the function f . If it is
a minimum, then the Theorem follows from the fact that a n-disc is homotopy equivalent
to a point. If it is a maximum, then we can use the same method as in the proof of Reeb’s
Theorem (cf. Corollary 3.21).
Thus, we will assume that λ ∈ {1, . . . , n − 1}. By Morse Lemma (cf. Lemma 3.7) there
exist a neighbourhood U of x0 and coordinates x1 , . . . , xn on U such that x0 = (0, . . . , 0)
and
λ
X n
X
(5) f (x) = f (x0 ) − x2i + x2i .
i=1 i=λ+1

Let  > 0 small enough so that U contains the closed ball


X
B√2 = {(x1 , . . . , xn ) | x2i ≤ 2}.
Define the λ-cell eλ by
eλ := (x1 , . . . , xn ) |x21 + · · · + x2λ ≤ ε, xλ+1 = · · · = xn = 0 .


We want to show that S c− ∪ eλ is a deformation retract of S c+ .


In order to illustrate the method, we first consider the case n = 2 and λ = 1. We have
• The preimage of c in U
f −1 (c) ∩ U = {x ∈ U | f = c} = {(x1 , x2 ) | x21 = x22 }
is the union of two lines.
• The level set
S c− = {x1 , x2 ) | x21 − x22 ≥ }
is represented by the area in red.
• If
S c+ = {x1 , x2 ) | x21 − x22 ≥ −}
then S c+ \ S c− is represented by the area in green.
50 DIFFERENTIAL TOPOLOGY

We define the 1-cell in U by


e1 = {(x1 , x2 ) | x21 ≤ , x2 = 0}.
√ √
Then e1 is attached to the boundary of S c− in ( , 0) and (− , 0).
The same construction works for higher dimensional manifold and index λ ∈ {1, . . . , n−
1}. Let
Xλ Xn
2
ξ= xi and η= x2i .
i=1 i=λ+1
We have that
f −1 (c) ∩ U = {x1 , . . . , xn | ξ = η}
is a double cone with vertex in x0 . We also have
S c+ε = {(x1 , . . . , xn ) |ξ − η ≥ −ε}
S c−ε = {(x1 , . . . , xn ) |ξ − η ≥ ε}
DIFFERENTIAL TOPOLOGY 51

and we define
(6) eλ := {(x1 , . . . , xn ) | ξ ≤ , xλ+1 = . . . , xn = 0}.
As in the previous case, eλ is attached to the boundary of S c− .
We now define a function g which coincides with the function f outside U , it is slightly
smaller than f on U , and it has the same critical point of f . To this end, let µ : R≥0 → R≥0
be a smooth function such that
(
> if t = 0
µ(t)
=0 if t ≥ 2
and
−1 < µ0 (t) ≤ 0 for all t ≥ 0.
We define (
f outside U
g :=
f − µ(ξ + 2η) inside U .
Note that g ≤ f and by (5), we have that if x ∈ U , then
(7) g(x) = c − ξ + η − µ(ξ + 2η).
Finally, since µ = 0 if r ≥ 2, we have that g = f outside the ellipsoid
E = {(x1 , . . . , xn ) | ξ + 2η ≤ 2} ⊂ B√2 ⊂ U
If y ∈ E then
3
η(y) − ξ(y) ≤  − ξ(y) ≤ .
2
Thus, E ⊂ S c+ . We are going to prove the following properties:
(i) g −1 ((−∞, c + ]) = f −1 ((−∞, c + ]) = S c+ .
(ii) f and g have the same critical points.
(iii) g −1 ((−∞, c − ]) is a deformation retract of S c+ .
(iv) S c− ∪ eλ is a deformation retract of g −1 ((−∞, c − ]).
We first show (i). Since g ≤ f , it is clear that
f −1 ((−∞, c + ]) ⊂ g −1 ((−∞, c + ]).
Assume that f (x) > c + . Then, since E ⊂ S c+ , we have that g(x) = f (x). Thus,
g(x) > c +  and (i) follows.
We now prove (ii). Outside U , we have that f = g and there is nothing to prove. By
(7), inside of U , we have
∂g
∂ξ
= −1 − µ0 (ξ + 2η) < 0
∂g
∂η
= 1 − 2µ0 (ξ + 2η) > 1
where the two inequalities come from the fact that, by construction, −1 < µ0 ≤ 0. Thus,
∂g ∂g
dg = dξ + dη = 0
∂ξ ∂η
52 DIFFERENTIAL TOPOLOGY

if and only if
dξ = dη = 0
which is satisfied if and only if x1 = · · · = xn = 0, i.e. x = x0 . Thus, (ii) follows.
We now prove (iii). By (ii), x0 is the only critical point of g and by (7), we have

g(x0 ) = c − µ(0) < c − .

Thus x0 ∈ / g −1 ([c−, c+]) and g −1 ([c−, c+]) does not contain any critical point. Thus,
(iii) follows by the First Fundamental Theorem of Morse Theory (cf. Theorem 3.20) and
by (i).
We finally show (iv). Let H denote the closure of g −1 ((−∞, c + ]) \ S c− , then

g −1 ((−∞, c − ]) = S c− ∪ H.

We first show that eλ ⊂ H. Indeed, since ∂g


∂ξ
< 0 for all x ∈ eλ , we have g(x) ≤ g(x0 ).
Moreover ξ(x0 ) = 0 ≤ ξ(x), hence g(x) < c −  and x ∈ g −1 ((−∞, c − ]). On the other
hand, by (5) and (6)., we have

f (x) = c − ξ + η > c − 

/ S c− and the claim follows.


for all x in the interior of eλ . Thus, x ∈
In order to construct the deformation retraction, we consider three cases.

Case I: Assume ξ < .


Define the deformation retract by

rt : (x1 , . . . xn ) 7→ (x1 , . . . , xλ , txλ+1 , . . . txn ) .


∂g
Thus, r1 = Id and r0 maps everything to eλ . Moreover, since ∂η
< 0, each rt maps
g −1 ((−∞, c − ]) into itself.

Case II: Assume  ≤ ξ ≤ η + .


Define the deformation retract by

rt : (x1 , . . . xn ) 7→ (x1 , . . . , xλ , lt xλ+1 , . . . lt xn )

where
s
ξ−
lt := t + (1 − t) .
η
Then it follows easily that rt is continuous. As in the previous case, we have r1 = Id and
r0 maps everything to S c− , since (x1 , . . . , xλ , l0 xλ+1 , . . . l0 xn ) is the level set f −1 (c − ).
Moreover, as in Case I, rt maps g −1 ((−∞, c − ]) to itself.
DIFFERENTIAL TOPOLOGY 53

Case III: We finally assume η +  ≤ ξ.


This is exactly the set S c− . Thus, we choose rt = Id.
Note that Case I, coincides with Case II when ξ =  and Case II coincides with Case
III when ξ = η + . Thus, we have shown (iv).
By (iii) and (iv), it follows that S c− ∪ eλ is a deformation retract of S c+ . Thus, the
Theorem follows. 
Remark 3.23. We can generalise the proof of the previous Theorem to the case of k
critical points x1 , . . . , xk such that f (x1 ) = · · · = f (xk ) = c, with indices λ1 , . . . , λk
respectively. In this case, the sublevel set S c+k is homotopically equivalent to S c− ∪f1
Dλ1 ∪ · · · ∪fk Dk for some attaching maps f1 , . . . , fk .
We can finally state the main Theorem of this Section.
54 DIFFERENTIAL TOPOLOGY

Theorem 3.24. Let M be a manifold and let f : M → R be a Morse function such that
the sublevel sets S t are compact for all t ∈ R.
Then M is homotopy equivalent to a CW complex with one λ-cell for each critical point
of index λ.

Before we proceed with the proof of the Theorem, we need to recall few facts from
Topology.
Definition 3.25. A continuos map f : X → Y between CW complexes is cellular if it
maps skeletons to skeletons (cf. Definition 3.9), i.e. f (X (n) ) ⊂ Y (n) for all n ≥ 0.
Theorem 3.26 (Cellular Approximation). 2 Let X and Y be CW-complexes and S ⊂ X
be a subcomplex. Assume that f : X → Y is a continuous morphism such that f |S is
cellular.
Then f is homotopic to a cellular map f˜: X → Y such that f˜|S = f |S .
The idea of the proof is to prove the claim on the skeleton M (n) by proceeding by
induction on n.
3
Theorem 3.27 (Whitehead). Let X be a topological space and let
φ0 , φ1 : ∂Dk → X
be homotopic attaching maps.
Then the identity maps of X extends to an homotopy equivalence
H : X ∪f0 Dk → X ∪f1 Dk
In particular, the Theorem implies that the homotopy of a space X ∪f Dk does not
depend on the way the cell Dk is attached if the attaching maps are homotopically equiv-
alent. Similarly, the following theorem implies that the homotopy of a space X ∪f Dk
does not depend on the homotopy class of X.
Theorem 3.28 (Hilton). 4 Let X be a topological space and let f : ∂Dk → X an attaching
map.
Then any homotopy equivalence h : X → Y extends to an homotopy equilvance
H : X ∪f Dk → X ∪h◦f Dk .

We can now proceed with the proof of Theorem 3.24.


Proof of Theorem 3.24. Let c0 , c1 , . . . be the critical values of f . Since the sublevel set S t
is compact for all t ∈ R, there are no accumulation points at the critical values. Moreover,
S t is empty, if t < c0 .
2E.g. see [Hat02, Theorem 4.8]
3E.g. see [Hat02, Theorem 4.6]
4E.g. See [Mil63, Lemma 3.7]
DIFFERENTIAL TOPOLOGY 55

We now prove by induction that for each i there exists i > 0 such that S ci +i is
homotopy equivalent to a CW complex with one λ-cell for each critical point x of index
λ and such that f (x) ≤ ci . If 0 > 0 is sufficiently small, then S c0 +0 is just the disjoint
union of contractible subsets containing all the points x ∈ M such that f (x) = c0 and
such that, for each such point, the index is zero. Thus, the claim follows.
We now assume that i > 0. By induction, there exists i−1 such that there is a homotopy
equivalence S ci−1 +i−1 → Xi−1 , where Xi−1 is a CW complex. Let x1 , . . . , xk ∈ M be the
critical points such that f (x1 ) = · · · = f (xk ) = ci and with indices λ1 , . . . , λk respectively.
By the second fundamental Theorem of Morse theory (cf. Theorem 3.22 and Remark
3.23), there exists i > 0 such that S ci +i has the homotopy type of
S ci −i ∪f1 Dλ1 ∪ · · · ∪fk Dk
for some attaching maps f1 , . . . , fk . Thus, for each k, we have a morphism
f gi−1 hi−1
k
∂Dλk −→ S ci −ε −→ S ci−1 +i−1 −→ Xi−1 ,
where the existence of the second morphism follows from the first fundamental Theorem
of Morse theory (cf. Theorem 3.20).
By the Cellular Approximation Theorem (cf. Theorem 3.26), the induced morphism
(λ )
fk ◦ gi−1 ◦ hi−1 is homotopic equivalent to a cellular map φk : ∂Dλk → Xi−1k .
By Whitehead Theorem and Hilton Theorem (cf. Theorem 3.27 and 3.28), it follows
that the CW complex
Xi := Xi−1 ∪ψ1 Dλ1 ∪ψ2 · · · ∪ψk Dλk
is homotopy equivalent to S ci +i , i.e. to
S ci −i ∪f1 Dλ1 ∪ · · · ∪fk Dλk .
Thus, the Theorem follows. 

3.4. Morse homology. Let (M, g) be a Riemannian manifold and let f : M → R be a


Morse function.
For every x ∈ M , let γx (t) be the gradient flow line with respect to f starting at x. For
any critical point x ∈ M , we may define the subsets
W s (x) := {x ∈ M | lim γx (t) = x}
t→∞

W u (x) := {x ∈ M | lim γx (t) = x}


t→−∞

called the stable and unstable manifold of c respectively. In addition, the sets W s (x)
(resp. W u (x)) are pairwise disjoint. Thus, by Proposition 3.18, it follows that they define
two partitions of M :
G G
M= W s (x) = W u (x).
c c
56 DIFFERENTIAL TOPOLOGY

It can be shown moreover that if λ is the index of the critical point c then W s (x)
and W u (x) are diffeomorphic to open discs of dimension n − λ and λ respectively. More
precisely, we have:
Theorem 3.29 (Stable Manifold Theorem). 5 Let (M, g) be a Riemannian manifold, let
f : M → R be a Morse function and x ∈ M be a critical point of f .
Then the tangent space of M at x can be decomposed as follows:
Tx M = Tx W u (x) ⊕ Tx W s (x).
Moreover, W s (x) and W u (x) are smooth submanifolds diffeomorphic to open discs with
dim W s (x) = λ and dim W u (x) = n − λ.
Example 3.30. Let M = S n and consider the height function f : M → R. The only
critical points of f are the points S and N corresponding to the minimum and maximum
of f respectively. Then
W u (N ) = S n \ {S} and W s (N ) = {N }.
Similarly,
W u (S) = S and W s (N ) = S n \ {S}.
Definition 3.31. Let M be a manifold and let K, L be submanifolds of M . Then K and
L are said to be transverse if for every point x ∈ M , we have
Tx K + Tx L = Tx M,
i.e. every vector in Tx M can be written as a sum of a vector in Tx K and a vector in Tx L.
Proposition 3.32. If K, L ⊂ M are transverse embedded submanifolds then K ∩ L is
also an embedded submanifold of dimension dim K + dim L − dim M .
Proof. Let k = dim M − dim K be the codimension of K and let x ∈ K ∩ L. Then there
exists a neighbourhood U of x in M such that K ∩ U = f −1 (0) where f : U → Rk is a
smooth function and 0 ∈ Rk is a regular value for f . Similarly if l = dim M − dim L is the
codimension of L, then there exists a smooth function g : U → Rl such that L∩U = g −1 (0)
and 0 ∈ Rl is a regular value for g.
Then K ∩ L ∩ U = F −1 (0), where F = (f, g) : U → Rk+l . Moreover, 0 ∈ Rk+l is a
regular value for F since
KerDFx = KerDfx ∩ KerDgx = Tx L ∩ Tx L
has codimension k + l by the transversality condition. 
Definition 3.33. Let (M, g) be a Riemannian manifold and let f : M → R be a Morse
function. Then f is said to be Morse-Smale if for all critical points x, y ∈ M for f , the
manifolds W u (x) and W s (y) are transverse.
Note that, the definition depends on the metric g.
5See [BH04, Theorem 4.2].
DIFFERENTIAL TOPOLOGY 57

Proposition 3.34. Let (M, g) be a compact Riemannian manifold of dimension n and


let f : M → R be a Morse-Smale function. Let x, y ∈ M be critical points for f of index
λx , λy respectively and such that
N := W u (y) ∩ W s (x) 6= ∅.
Then N is an embedded submanifold of dimension λy − λx .
Proof. By Proposition 3.32 and Theorem 3.29, we have that N is an embedded manifold
of dimension
dim N = dim W u (y) + dim W s (x) − n = λy + n − λx − n = λy − λx .
Thus, the claim follows. 
Notation 3.35. In the same set-up as in Proposition 3.34, we denote
W (x, y) := W u (x) ∩ W s (y).
Corollary 3.36. Let (M, g) be a compact Riemannian manifold of dimension n and let
f : M → R be a Morse-Smale function. Let x, y ∈ M be critical points for f of index
λx , λy respectively and such that W (x, y) 6= ∅.
Then λy > λx .
Proof. If W (x, y) 6= ∅ then it contains at least one flow line from y to x. Hence, by
Proposition 3.34, we have λy − λx = dim W (x, y) ≥ 1 and the claim follows. 
Example 3.37. Let M and f as in Example 3.5. Then there is a flow line connecting
two critical points both with index 1. Thus, f is not a Morse-Smale function. On the
other hand, this can be fixed by bending the torus a bit and leaving the height function
unchanged so that we dot have flow lines beginning and ending at critical points of the
same index.
58 DIFFERENTIAL TOPOLOGY

Corollary 3.38. Let (M, g) be a compact Riemannian manifold of dimension n and let
f : M → R be a Morse-Smale function. Let x, y ∈ M be critical points for f of index
λx , λy respectively, such that λy − λx = 1.
Then W (x, y) = W (x, y) ∪ {x, y} has finitely many components, i.e. the number of
gradient flow lines from y to x is finite.
Proof. Since λy − λx = 1, it follows that W (x, y) ∪ {x, y} is closed because there are no
other critical points between y and x as the index is strictly decreasing along flow lines.
Since M is compact, it follows that W (x, y) ∪ {x, y} is also compact.
The gradient flow lines form an open cover of W (x, y) which can be extended to an
open cover of W (x, y) by taking the union of each gradient flow line with small open
subsets in W (x, y) around x and y. Thus, by compactness, it follows that the number of
flow lines from y to x is finite. 

Let (M, g) be a compact Riemannian manifold of dimension n and let f : M → R be a


Morse-Smale function. Let x, y ∈ M be critical points for f of index λx , λy . Then R acts
on W (x, y) by flowing along the flow lines for t ∈ R. The quotient
M (x, y) = W (x, y)/R
is called the moduli space of flow lines from y to x. By the previous Corollary, if
λy − λx = 1 then M (x, y) is a finite set with each point corresponding to a flow line from
y to x. We are now going to assign a sign ±1 to each of these point depending on their
orientation.
Lemma 3.39. Let (M, g) be a compact Riemannian manifold of dimension n and let
f : M → R be a Morse-Smale function. Let x, y ∈ M be critical points for f of index
λx , λy . Let z ∈ W (x, y) and let γ be the orbit of z under the R-action.
Then there are canonical isomorphisms:
(1) Tz W u (y) ' Tz W (x, y) ⊕ Tz M/Tz W s (x);
(2) Tz W (x, y) ' Tz M (x, y) ⊕ Tz γ;
(3) Tz W u (y) ' Tz M (x, y) ⊕ Tz γ ⊕ Tz M/Tz W s (x).
Proof. We first prove (1). By Proposition 3.34, the Riemannian metric induces the de-
composition
Tz W u (y) ' Tz W (x, y) ⊕ Tz W u (y)/Tz W (x, y).
Consider the map
φ : Tz W u (x) → Tz M/Tz W s (x)
obtained as the composition of the inclusion with the quotient map. Then,
Kerφ = Tz W (x, y).
By definition of Morse-Smale function, W u (x) and W s (y) are transverse. Thus,
Tz W u (y) + Tz W s (x) = Tz M.
DIFFERENTIAL TOPOLOGY 59

It follows that φ is surjective and


Tz W u (y)/Tz W (x, y) ' Tz M/Tz W s (x).
Thus, (1) holds.
We now prove (2). Let π : W (x, y) → W (x, y)/R be the quotient map. Then
Dπz : Tz W (x, y) → Tπ(z) W (x, y)/R
is a surjective linear map and the kernel of Dπz is exactly Tz γ. Thus, by the nullity
theorem, we get a canonical isomorphism
Tz W (x, y)/Tz γ ' Tπ(z) W (x, y)/R.
Using the metric, we obtain (2).
(3) follows immediately rom (1) and (2). 
Assume now that (M, g) is a compact oriented Riemannian manifold. In the set-up of
the Lemma above, we now want to choose an orientation which is compatible with the
flow. To this end, we first choose an orientation on M and an orientation on Tx W u (x)
for any critical point x. By the Stable Manifold Theorem (cf. Theorem 3.29), W u (x)
is an oriented manifold. Thus, the orientation on Tx W u (x) induces an orientation on
W u (x) for all z ∈ W u (x). Since, by transversality, we have Tx M = Tx W u (x) ⊕ Tx W s (x),
the orientation on Tz M and Tz W u (x) induces an orientation on Tx W s (x). By the Stable
Manifold Theorem, W s (x) is also oriented and so this induces an orientation on W s (x)
for all z ∈ W s (x).
Assume now that x, y ∈ M are critical points for f of index λx , λy respectively, such
that λy −λx = 1. By Corollary 3.38, it follows that M (x, y) is a finite set. Let z ∈ W (x, y).
Then the gradient flow with respect of f induces an orientation on Tz γ, given by −∇f (z).
The above orientation on W u (y), provides an ordered basis
(−∇f (z), B u (z))
of Tz W u (y). Similarly, the orientation on W u (x) induces an orientation on Tx W s (x) and
thus an ordered basis B s (z) of Tz W s (x). Since −∇f (z) spans Tz W u (y) ∩ Tz W s (x), it
follows that
(B u (x), B s (x))
is an ordered basis of Tz M . If this coincides with the orientation on Tz M then we assign
+1 as the orientation of the corresponding element in M (x, y), otherwise, we assign −1.
Note that the choice of these orientations is not unique, as the depends on the choice of
the manifold W u (x).
Thus, given a smooth compact oriented Riemannian manifold (M, g) with a Morse-
Smale function f : M → R, after we assign the orientation of the unstable manifolds of
f , we may define the number n(x, y) ∈ Z as the algebraic sum of signed flow lines from
y ∈ Crk (f ) to x ∈ Crk−1 (f ), where, for any k ≥ 0,
Crk (f ) := {critical point for f of index k}.
60 DIFFERENTIAL TOPOLOGY

Example 3.40. Let M = S 1 ⊂ R2 be the circe, let


f : S1 → R (x, y) | y
be the height function and let g be the metric on M induced by the Euclidean metric on
R2 . As in Example 3.30, f admits two critical points S and N and we have
W u (N ) = S n \ {S} W s (N ) = {N }

W u (S) = S W s (N ) = S n \ {S}.
Thus, the function f is Morse-Smale.
We now want to compute n(N, P ). We pick the “clockwise” orientation on S 1 . We now
have to pick an orientation for the unstable manifolds at the two critical points. We pick
the basis on TN W u (N ) to be the tangent vector “from left to right”. This determines the
orientation of Tx W u (N, P ) for all x ∈ W (S, N ). Since W u (S) is just a point, we assign
the orientation +1. This yields an orientation on all stable manifolds such that at the
critical points, the orientations of the unstable and stable manifolds are compatible with
the one of S 1 , i.e. for example the basis of TS W s (S) has to be the tangent vector “from
right to left”. Consider now the negative gradient −∇f (x). It agrees with the orientation
of Tx W u (N ) on the right side and has the opposite orientation on the left side. Since
Tx W u (N ) is one dimensional, B u (x) is just a sign which turns the basis −∇f (x) into
a positive basis of Tx W u (N ). Hence B u (x) = +1 on the right side and B u (x) = −1 on
the left side. Since B s (x) is a positive basis of Tx W s (S) by our choice, and W u (S) and
W s (S) are oriented compatibly with M at S, we see that the orientation +1 of TS W u (S)
together with a positive basis of TS W s (S) gives a positive basis of TS M . There also at x,
+1 together with a positive basis of Tx W s (p) yields a positive basis of Tx M . We conclude
finally that (B u (x), B s (x)) is positive on the right flow line and negative on the left. Thus,
n(S, N ) = 1 − 1 = 0.
DIFFERENTIAL TOPOLOGY 61

Let Ck (f ) := Z[Crk (f )] be the free abelian group generated by all the critical points of
index k and define the linear map
X
∂k : Ck (f ) → Ck−1 (f ) ∂k (y) := n(x, y)p.
x∈Crk−1 (f )

6
Theorem 3.41. Under the set-up above, the pair (C• (f ), ∂• ) is a chain complex. If we
denote
Zk (C• (f ), ∂• ) := Ker(∂k : Ck → Ck−1 )
and
Bk (C• (f ), ∂• ) := Im(∂k+1 : Ck+1 → Ck ).
The quotient
Zk (C• (f ), ∂• )
Hk (M, f ) :=
Bk (C• (f ), ∂• )
does not depend on f and it coincides with the singular homology group Hk (M ).
The pair (C• (f ), ∂• ) is called the Morse-Smale-Witten chain complex. The group
Hk (M, f ) is called the k-th Morse Homology group of M with respect to f .
Example 3.42. Consider again the example M = S 1 with f and g as in Example 3.40.
Since n(S, N ) = 0, we have ∂1 = 0 and the Morse-Smale-Witten chain complex is
1∂
C1 (f ) −→ C0 (f ) → 0,
where C1 (f ) = C0 (f ) = Z, since there is only one critical point of index 0 and one critical
point of index 1. Thus,
(
Z if p = 0 or 1
Hp (M, F ) =
0 otherwise.

4. Singular homology
Definition 4.1. A standard n-simplex is
( n
)
X
∆n := (t0 , . . . tn ) ∈ Rn+1 | ti = 1, ti ≥ 0 .
i=0

A k-face of ∆n is defined as [ei0 , . . . , eik ] with 0 ≤ i0 ≤ . . . ik ≤ n, where e0 , . . . , en denotes


the standard basis of Rn+1 .
The i-th face map of ∆n is defined to be the map
Fin := [e0 , . . . , êi , . . . en ] : ∆n−1 → ∆n .
6See “Lectures on Morse homology”, A. Banyaga and D. Hurtubise - Theorem 7.4.
62 DIFFERENTIAL TOPOLOGY

Example 4.2. The 1-faces of ∆2 are


[e0 , e1 ], [e1 , e2 ] and [e0 , e2 ].
Note that the positive orientation on ∆2 induced by R3 , yields an orientation on the 1-
faces so that [e0 , e1 ] and [e1 , e2 ] are positive oriented but [e0 , e2 ] is negatively oriented.
More in general, the positive orientation on ∆n induces the positive orientation on the
face Fin if i is even and negative orientation if i is odd.
Definition 4.3. Let X be a topological space. Then a singular n-simplex of X is a
continuous map
σ : ∆n → X.
The i-th face of σ is given by
σi := σ ◦ Fin
and the boundary of σ is defined as
n
X
∂n (σ) = (−1)i σi .
i=0

Note that the factor (−1)i takes into account the orientation of the i-th face as discussed
in Example 4.2.
Definition 4.4. Let X be a topological space. For each p ≥ 0, the singular p-chain
group Cp (X) is the free Abelian
P group generated by the singular p-simplices of X. A
p-chain is a formal finite sum nσ σ of p-simplices with integral coefficients nσ .
The p-boundary operator is defined as the linear map
∂p : Cp (X) → Cp−1 (X)
P
defined by, for any c = nσ σ,
X
∂p (c) := nσ ∂k (σ).
σ

Lemma 4.5. Let X be a topological space. Then ∂p−1 ◦ ∂p = 0 for all p ≥ 1.


Proof. ∂p (σ) = i (−1)i σ ◦ [e0 , . . . , êi , . . . , ep ] by definition.
P
Thus,
X
∂p−1 ◦ ∂p (σ) = (−1)i (−1)j−1 σ ◦ [e0 , . . . , êi , . . . , êj . . . , ep ]
i<j
X
+ (−1)i (−1)j σ ◦ [e0 , . . . , êi , . . . , êj . . . , ep ] .
i<j

where the first sum comes from deleting the entry ei first and then ej while the second
sum deletes the ej first and then the ei . Thus, the two sums cancel out and the claim
follows. 
DIFFERENTIAL TOPOLOGY 63

By the previous Lemma, we get a chain complex


∂p+1 ∂p ∂p −1
. . . −→ Cp (X) −→ Cp−1 (X) −→ . . . .
The Chain complex (C• (X), ∂• ) is called the singular chain complex of X. The Kernel
Zp (X) := Ker(∂p : Cp (X) → Cp−1 (X))
is called the group of singular k-cycles and the image
Bp (Z) := Im(∂p+1 : Cp+1 (X) → Ck (X))
is called the group of singular k-boundaries. By Lemma 4.5, we have Bp (Z) ⊂ Zp (X).
Thus we may define the quotient
Zp (X)
Hp (X) := ,
Bp (Z)
which we call the singular p-th homology group of X.
If f : X → Y is a continuous map between topological spaces then, for all p ≥ 0, there
is a natural linear map
f∗ : Cp (X) → Cp (Y )
such that, for any singular p-simplex σ : ∆n → X, we set
f∗ σ = f ◦ σ : ∆n → Y.
It follows easily that f∗ induces a map in homology:
Hp (X) → Hp (Y ).
For all p ≥ 0, we denote
C p (X, R) := Hom (Cp (X), R) .
The singular k-boundary operator is the linear map
∂ p : C p (X, R) → C p+1 (X, R)
such that, for all φ ∈ C p (X, R), we set
∂ p φ := φ ◦ ∂p+1 .
It follows immediately that ∂ p+1 ◦ ∂ p = 0.
The cochain complex (C • (X, R), ∂ • ) is called the singular cochain complex of X with
coefficients in R. The singular p-cocycles, the singular p-coboundaries and the singular
p-th cohomology groups H p (M, R) are defined in the same way as usual.
If f : X → Y is a continuous map between topological spaces then, for all p ≥ 0, there
is a natural linear map
f ∗ : C p (Y, R) → C p (X, R)
such that, for all φ ∈ C p (Y, R), we have
f ∗ φ(σ) = φ(f∗ σ).
64 DIFFERENTIAL TOPOLOGY

The map f ∗ induces a pull-back map in cohomology


f ∗ : H p (Y, R) → H p (X, R).

Example 4.6. Assume that X = {x} is a single point. Then there is only one map
∆n → {x} for each n. The boundary operators are therefore given by ki=0 (−1)i σi which
P
is zero when k is odd and equal to 1 when k is even. Hence we have the chain complex
0 1 0 0 1 0 0
. . . → Z → Z → . . . → Z → Z → Z → 0.
As a consequence, we have

Z/0 = Z if p = 0
 (
Z if p = 0,
Hp (X) = Z/Z = 0 if p is odd =

0/0 = 0 0 otherwise.
if p > 0 is even
Similarly, we have (
R if p = 0,
H p (X, R) =
0 otherwise.
Note that so far we have considered singular simplices as continuous maps from the
standard simplices into topological spaces. But since we are only considering manifolds
we want to restrict our study to smooth singular simplices. However we can apply our
previous discussion with the exact same definitions to this case. Moreover it can be shown
that the smooth singular (co)homology groups coincide on a manifold to the singular
(co)homology groups with have defined. Finally, similarly to de Rham cohomology we
have the following two results which we will need for the proof of de Rham theorem.
Similarly to Corollary 2.14, we have:
Theorem 4.7 (Homotopy equivalence for singular cohomology). Let M and N be man-
ifolds and let f : M → N be a smooth homotopy equivalence.
Then, for all p ≥ 0, the pull-back
f ∗ : H p (N ) → H p (M )
is an isomorphism.
Similarly to Theorem 2.29, we have
Theorem 4.8 (Mayer-Vietoris for singular cohomology). Let M be a manifold such that
M = U ∪ V , where U, V ⊂ M are open subsets and U ∩ V is not empty. Then for each
p ≥ 0, there exists a linear map
δ : H p (U ∩ V ) → H p+1 (M )
such that the following sequence is exact:
δ f g δ f
. . . → H p (M ) → H p (U ) ⊕ H p (V ) → H p (U ∩ V ) → H p+1 (M ) → . . .
DIFFERENTIAL TOPOLOGY 65

where if
iU : U → M, iV : V → M, jU : U ∩ V → U, jV : U ∩ V → V
denote the inclusion maps, then f := (i∗U , i∗V ) and g := jV∗ − jU∗ .
4.1. de Rham homomorphism. Let M be a manifold of dimension n and let σ : ∆p →
M be a singular p-simplex. If ω is a closed compactly supported p-forms on M , then the
integral of ω over σ is defined as
Z Z
ω := σ ∗ ω.
σ ∆k
P
The definition can be extended linearly to any smooth p-chain c = σ nσ σ by
Z X Z
ω := nσ · ω.
c σ σ

There is a version of Stokes Theorem (cf. Theorem 1.32) for k-chain:


Theorem 4.9. Let M be a manifold of dimension n and c a smooth p-chain on M , for
some p ≥ 0. Let ω be a compactly supported (p − 1)-form on M .
Then Z Z
ω= dω.
∂c c

Note that for each singular p-simplex σ, the integral over the boundary ∂σ respects the
orientation induced by σ.
We can now define a linear map between the space of forms and the smooth cochain
group in the following way. Let RM be a manifold of dimension n and let p ≥ 0. For each
ω ∈ Ωp (M ), we define the map ω by
Z Z
ω : Cp (M ) → R c 7→ ω.
c
R p
Thus, ω ∈ Hom (Cp (X), R) = C (M, R). Moreover, we have a linear map
 Z 
p p p
l : Ω (M ) → C (M, R) ω 7→ c 7→ ω
c
p
By Theorem 4.9, the map l commutes with the singular cohomology operator and the
de Rham differentials, i.e.
lp ◦ dp−1 = ∂ p−1 ◦ lp−1 .
Thus, by Proposition 2.24, lp induces a cochain map
lp : H p (M ) → H p (M, R)
between the de Rham cohomology and the singular cohomology groups. The map is called
de Rham homomorphism.
66 DIFFERENTIAL TOPOLOGY

Lemma 4.10. Let U be an open subset of Rn which is contractible. Then the de Rham
homomorphism
lp : H p (U ) → H p (U, R)
is an isomorphism for each p ≥ 0.
Proof. Since U is contractible, by Example 4.6, Theorem 4.7 and Theorem 2.16, it follows
that
H p (U ) = H p (U, R) for all p ≥ 0.
Thus, we just need to check that l is an isomorphism. The group H 0 (U ) is generated
0

by constant real functions on U . Similarly, the only 0-simplices are the maps {0} → U .
Integrating a function over a point gives the value of the function at that point. Hence l0
is an isomorphism, since it is not zero. 
Theorem 4.11. Let M be a compact manifold. Then the de Rham homomorphism
lp : H p (M ) → H p (M, R)
is an isomorphism for each p ≥ 0.
The proof is similar to the one we saw for the Poincaré Duality (cf. Theorem 2.52).
The idea is to pick a good cover {U } as in Definition 2.34 and then proceed by induction
on the number of open sets in the cover. By using Mayer-Vietoris Theorem both for the
de Rham cohomology and the singular cohomology (cf. Theorem 2.29 and Theorem 4.8),
we may apply the Five Lemma (cf. Lemma 2.54) and the induction hypothesis, to obtain
the desired isomorphism.

References
[BH04] Augustin Banyaga and David Hurtubise. Lectures on Morse homology, volume 29 of Kluwer Texts
in the Mathematical Sciences. Kluwer Academic Publishers Group, Dordrecht, 2004.
[Hat02] Allen Hatcher. Algebraic topology. Cambridge University Press, Cambridge, 2002.
[Lee13] John M. Lee. Introduction to smooth manifolds, volume 218 of Graduate Texts in Mathematics.
Springer, New York, second edition, 2013.
[Mil63] J. Milnor. Morse theory. Based on lecture notes by M. Spivak and R. Wells. Annals of Mathe-
matics Studies, No. 51. Princeton University Press, Princeton, N.J., 1963.
[Pet16] Peter Petersen. Riemannian geometry, volume 171 of Graduate Texts in Mathematics. Springer,
Cham, third edition, 2016.
[Tu11] Loring W. Tu. An introduction to manifolds. Universitext. Springer, New York, second edition,
2011.

You might also like