VU University, Amsterdam

Bachelorthesis

De Rham Cohomology of smooth

manifolds

Supervisor:
Author:
Prof. Dr. R.C.A.M.
Patrick Hafkenscheid
Vandervorst
Contents
1 Introduction 3

2 Smooth manifolds 4
2.1 Formal definition of a smooth manifold . . . . . . . . . . . . . . . 4
2.2 Smooth maps between manifolds . . . . . . . . . . . . . . . . . . 6

3 Tangent spaces 7
3.1 Paths and tangent spaces . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Working towards a categorical approach . . . . . . . . . . . . . . 8
3.3 Tangent bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Cotangent bundle and differential forms 12

4.1 Cotangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Cotangent bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.3 Smooth vector fields and smooth sections . . . . . . . . . . . . . 13

5 Tensor products and differential k-forms 15

5.1 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Symmetric and alternating tensors . . . . . . . . . . . . . . . . . 16
5.3 Some algebra on r (V ) . . . . . . . . . . . . . . . . . . . . . . . 16
5.4 Tensor bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

6 Differential forms 19
6.1 Contractions and exterior derivatives . . . . . . . . . . . . . . . . 19
6.2 Integrating over topforms . . . . . . . . . . . . . . . . . . . . . . 21

7 Cochains and cohomologies 23

7.1 Chains and cochains . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.2 Cochains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
7.3 A few useful lemmas . . . . . . . . . . . . . . . . . . . . . . . . . 25

8 The de Rham cohomology 29

8.1 The definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.2 Homotopy Invariance . . . . . . . . . . . . . . . . . . . . . . . . . 30
8.3 The Mayer-Vietoris sequence . . . . . . . . . . . . . . . . . . . . 32

9 Some computations of de Rham cohomology 35

10 The de Rham Theorem 40

10.1 Singular Homology . . . . . . . . . . . . . . . . . . . . . . . . . . 40
10.2 Singular cohomology . . . . . . . . . . . . . . . . . . . . . . . . . 41
10.3 Smooth simplices . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
10.4 De Rham homomorphism . . . . . . . . . . . . . . . . . . . . . . 42
10.5 de Rham theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

1
11 Compactly supported cohomology and Poincar e duality 47
11.1 Compactly supported de Rham cohomology . . . . . . . . . . . . 47
11.2 Mayer-Vietoris sequence for compactly supported de Rham co-
homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
11.3 Poincare duality . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

12 Conclusion 53

13 Bibliography 54

2
1 Introduction
When I was told I had to write a thesis at the end of my bachelorphase I
thought long and hard on the subject. In the short period I have spend studying
mathematics I have always enjoyed both topology and analysis, as such my idea
was to write my thesis on a subject in either field. Unfortunately choosing turned
out to be a difficult task. Though, I remembered a course in my second year
that seemed to mix both fields into one subject, namely the theory of smooth
manifolds and differential geometry. Since it has always fascinated me I decided
I would write about something related to this. I chose the subject of de Rham
cohomology because it is very obvious that it relies heavily on both topology
as well as analysis. One might even say it creates a natural bridge between
the two. Of course I realized I wanted my thesis to be readable by others that
may not have the prerequired knowledge of manifolds or differential forms, that
is why the first part of this thesis (the first 5 chapters) is an introduction to
smooth manifolds and differential forms. In the chapter after that I wrote a
short introduction to algebraic topology, in here I define chains/cochains and
basic exactness properties. Very important is the Zig-zag Lemma that will be
used a lot in the later chapters. Then finally in chapter 7, de Rham cohomology
groups are defined, as well as some basic properties proved. I dedicated chapter
8 to some examples of calculation of the de Rham groups. An important lemma
is Poincares lemma that calculates the de Rham groups of contractible spaces.
Chapter 9 starts with a short introduction to singular cohomology, and goes
on with a proof of de Rhams theorem, which states that for smooth manifolds
singular cohomology is identical to de Rham cohomology. A similar proof is used
in chapter 10, where I proved Poincare duality, which gives a relation between
de Rham cohomology and de Rham cohomology with compact support. That
is as far as this thesis will go, so I hope youll find this an interesting read!

3
2 Smooth manifolds
To understand the ideas behind the de Rham cohomology it is first important
to understand the types of spaces we will be working with. Initially the types of
spaces we will be working with will be smooth manifolds but later on we will also
consider submanifolds of any Rn . A smooth manifold can best be described as a
topological space that is locally very much like the Euclidian space of a certain
dimension. The smooth part of the name will relate to the differentiability of
the maps that connect our space to the matching Euclidian space.

2.1 Formal definition of a smooth manifold

If we want to define what a smooth manifold is we first need to look at what a
not-necessarily-smooth manifold is,
Definition 2.1. A topological space M is called a manifold of dimension n if:
M is Hausdorff (points can be seperated by open sets).
M is second countable (M has a countable topological base)
For all p M there is an open neighbourhood U M such that U is
homeomorphic to an open subset V of Rn .

These properties may look like they narrow down the amount of spaces we
could work with, but in fact, in order to produce some of the theory this thesis
will discuss we will need a stricter definition. In order to achieve this we will
need the concept of an atlas. As we have seen in the definition of a manifold we
need for each point in M a neighbourhood U that is homeomorphic to an open
subset of Rn . We can make this more rigorous by the following definition;
Definition 2.2. Let M be a manifold of dimension n. A pair (U, ), where
U M is open and : U V Rn a homeomorphism to some open V , is
called a chart.
Remark 2.1. I will abuse notation in this thesis by saying that p (U, ) if
(U, ) is a chart of M and p U .
We can thus rewrite our third condition from Definition 2.1 as;
For all p : p (U, ) for some chart.
The collection of charts such that each p M is in a chart is called an atlas.
It is important to realise that an atlas characterizes a manifold. Now that we
have found a way to describe manifold with a collection of sets and maps we
can add the additional requirement of smoothness.
Definition 2.3. An atlas A = {(U , )}I is called smooth, if for all i , j
we have that 2 1
1 is a diffeomorphism between 1 (U1 U2 ) 2 (U1 U2 ).

4
 We already know that all the s are homeomorphisms, a smooth atlas
only adds a certain degree of smooth transitioning to the equation. Now there
is one more problem with a proper definition of a smooth manifold, and that
is the fact that a manifold is not generated by a unique atlas; there could be
multiple different atlasses that produce the same manifold. This is why we need
to introduce the concept of a maximal atlas. Before we can formally introduce
this we need the concept of compatibility of atlasses.
Definition 2.4. Atlasses A and A# are called compatible if A A# (contains
any union of charts from A and A# ) is again a smooth atlas.
Remark 2.2. The relation A B A and B compatible forms an equivalence
relation on smooth atlasses.
Definition 2.5. A maximal atlas for a manifold M is the union
! of all smooth
atlasses in one equivalence class. In other words; Amax = {B : B [A]} for
some A is a maximal atlas.
The maximal atlas of a manifold M is also called the differentiable structure
of M . Now we can finally define what a smooth manifold is.
Definition 2.6. A smooth manifold is a pair (M, Amax ) where M is a manifold
and Amax a differentiable structure of M .
While this is the formal definition, it usually suffices to find any smooth
atlas for a manifold M to determine smoothness.
Remark 2.3. Rn is an n-dimensional smooth manifold with atlas (Rn , id).
We need one more definition, mostly because we also want a working defini-
tion for manifolds that have some sort of boundary, if we take for example the
half-sphere (p1 , p2 , p3 ) R3 : p21 + p22 + p23 = 1 and p3 0, we can see that our
definition of a smooth manifold doesnt work since we cannot find an (open)
chart around our points on the boundary of the half-sphere.
Definition 2.7. M is called a manifold with boundary of dimension n if:
M is Hausdorff (points can be seperated by open sets).
M is second countable (M has a countable topological base)
For all p M there is an open neighbourhood U M such that U is
homeomorphic to an open subset V of Hn .

The ressemblences to our previous definition of a manifold are clear, the

difference is that open subsets of the upper-half plane Hn can contain some-
thing that could pass as a boundary. More general we might need corners,
for instance think of a tetrahedron, which in itself is locally very much like a
three dimensional space but the corners make it so that there are no diffeomor-
phisms to a Euclidian space. However if we replace Hn in definition 2.7 with
Rn+ = {(x1 , . . . , xn ) : xi 0} we get what is called a manifold with corners.

5
2.2 Smooth maps between manifolds
Now that weve defined what a smooth manifold is exactly, we want some sort
of definition for a smooth map between manifolds. Of course each manifold
is a topological space, so maps have to be at least continuous. It would be a
nice idea to have a sort of differentiability of a map between manifolds, however
differentiability is merely a concept of functions to Euclidian spaces. That is
why we define a smooth maps as follows;
Definition 2.8. Let M be an m-dimensional and N an n-dimensional smooth
manifold. A continuous map f : M N is called smooth if for all p M ,
p (U, ) there is a chart (V, ) of N such that;
f (U ) V
f := f 1 : Rm Rn is infinitely differentiable.

Remark 2.4. The previously mentioned function f is called the coordinate

representation of f
Remark 2.5. If f : M N is a smooth homeomorphism and f 1 is also
a smooth map we call f a diffeomorphism, and we say that M and N are
diffeomorphic.
We can now also take a more categorical approach to smooth manifolds,
which is illustrated by the following lemma.
Proposition 2.1. The collection M = {M : M is a smooth manifold } to-
gether with F = {f : M1 M2 |M1 , M2 M and f smooth }, defines the
category Man where Ob(Man) = M and Mor(Man) = F.
Proof. All we need to show is that there is a smooth identity function, and
that the composition of two smooth maps is smooth again.
The identity function from any manifold to itself is smooth since idM (p) =
p, so we can take only one chart (U, ) for both p and idM (p). Now it
"
follows that id M = idm
1
= idRm where m is the dimension of M .
And we know that the identity function on Euclidian spaces is infinitely
differentiable.
Let M , N and L be smooth manifolds of dimension m, n and l respectively.
Consider two smooth maps f : M N , g : N L. We want to show
that g f : M L is a smooth map. So take p (U, ) and (V, ) such
that f (U ) V and f infinitely differentiable. The same we can do for
f (p) (V # , # ) and a chart (W, ) of L. Now consider V V # with the
coordinate . We know coincides with # on V V # . Now consider;
f g = f 1 g .
Now the latter is a composition of infinite differentiable functions, and as
such so is the left side. This proves that f g is smooth.

6
3 Tangent spaces
As we have seen in the previous sections smooth manifolds can be characterized
by an atlas, however these atlasses can be very complex and hard to under-
stand, so ideally we would like a simple characteristic of manifolds that we can
easily work with. This section deals with one of these simpeler characteristics,
namely the one of linear approximation. Specifically this can be viewed as a
generalizations of the linear approach to the graph of a function (tangent line
of the graph of a function can be viewed as a linear approximation). There are
a few different definitions, we will use the one that uses paths in our space.

3.1 Paths and tangent spaces

Definition 3.1. A path in a manifold M is an smooth function : (1, 1)
M.
Now to turn the previous definition into vectors that are tangent to our
manifold were going to need some sort of differentiation. Of course we cannot
take the derrivative of our path, since its image is contained in M and not in a
Euclidian space, hence we need some way to link the image of the path to some
Euclidian space. Of course we have such a tool, namely the charts of our M .
We might not be able to differentiate on M , but we can define for each point p
and a chart (U, ) around it the following vector in Rdim(M ) ;
# $
d
( ) (0),
dt

if of course we take a path with (0) = p. Now this isnt exactly ideal to
work with, thus we will define an additional structure on it through means of
identification of paths that look linearly equal at a given point p M .
Definition 3.2. For a p M with p (U, ), we define an equivalence relation
on paths with (0) = (0) = p as follows;
( )# (0) = ( )# (0)
Proposition 3.1. Definition 3.2 does not depend on your choice of chart.
Proof. Let p (U, ) and p (V, ) for some manifold M , and let be a path
defined on M with (0) = p. Consider now U V , which is an open set in M ,
we know by the fact that we are dealing with a manifold that on U V , and
are the same. So if we restrict ourselves to this smaller open set which contains
our point p we can differentiate and get the same resulting vector.

Now surprisingly we can recognize an additional structure on the tangent

space Tp M := { : (0) = p and is a path }/ .
Lemma 3.1. Tp M is a real vectorspace with;

7
 c[] := [c], c R
[] + [] = [] for some with ( )# (0) + ( )# (0) = ( )# (0).

I will not present the proof, as it is just an exercise in checking all the axioms
of a vectorspace. Now as one would expect, the dimension of this vector space
is finite, which of course means it is isomorphic to a Euclidian space.
Theorem 3.1. Let M be an m dimensional smooth manifold, and p M then;
Tp M
= Rm .
Proof. The following map is an isomorphism for any chart (U, ), : Tp M
Rm , [] , ( )# (0).
Injectivity follows from the definition of the class [].
Surjectivity follows from considering curves i := 1 (x + tei ).
The group actions follow from the definition of the vector operations.

3.2 Working towards a categorical approach

We have seen before that the smooth manifolds with smooth maps form a cate-
gory. Now categories come with functors, and the objective of this subsection is
to make the first step towards the Tangentbundle-functor of manifolds. Func-
tors work on spaces but also on the morphisms of a space to another. Therefore
we need to construct a special linear map for each smooth function. We will do
this as follows;
Definition 3.3. Let f : M N where M and N are smooth manifolds. The
pushforward of f , or f : Tp M Tf (p) N is defined as [] , [f ].
Remark 3.1. Using the following definition we can now define a proper (canon-
ical) basis for Tp M namely by using a chart at p, (U, ) and then defining
1
xi |p := ( ) (ei ). Note that this definition still heavily relies on the chosen
chart, hence the (canonical) part.
Now in order for the pushforward to be a proper functorial result we need
the following lemma.
Lemma 3.2. Let f : M N , g : N P smooth functions and p M . Then,
(i) f is a linear map.
(ii) (g f ) = g f .
(iii) (idM ) = idTp M .
(iv) If f is a diffeomorphism then f is an isomorphism (of vectorspaces).

8
3.3 Tangent bundles
Now we have defined at each point of a manifold what the tangent space is, we
want some sort of way to have one structure for the entire manifold. This is
why we define the vectorbundle.
Definition 3.4. (E, M, ) is called a vectorbundle of rank k if E, M are topo-
logical spaces, and : E M a continous surjection and,
1 (p) has a k-dimensional real vectorspace structure for all p.

For all p M there is an open neighbourhood U and a homeomorphism

: 1 (U ) U Rk . Such that on 1 (U ) we have that M = ,
where M is the projection of M Rk to M . And also for each q in U we
have that the restiction of to 1 (q) gives an isomorphism from {q}Rk
to Rk .

Remark 3.2. We call a local trivialisation.

Remark 3.3. We call a vectorbundle E smooth if both E and M are smooth
manifolds and is in fact a smooth map.
Basically we can see the vectorbundle as attaching a vectorspace Rk to
each of our points in the base M . Now naturally we didnt choose the name
of our base without reason, in fact we will apply the idea of a vectorbundle to
manifolds. Namely, we will attach to each point p M the tangent space Tp M .
Definition 3.5. Let M be a smooth manifold then the tangent bundle T M is
defined as,
%
T M := Tp M.
pM

Now we want to add some additional structure to our tangent bundle, for
one we want to see that it is in fact a smooth vectorbundle, but even more than
that we would like to see that this makes it a
Theorem 3.2. Let M be an m-dimensional smooth manifold with smooth atlas
(U , )I . Then T M is a smooth vector bundle of rank m over M with

: T M M, Tp M , p.

Proof. Since we know that 1 (p) = Tp M we know by definition that it has

an m-dimensional real vectorspace structure for all p, all that is left to prove is
to find local trivialisations for all the charts, and of course to prove smoothness.

9
 Let (U, ) be a chart, define : 1 (U ) U Rm as
& (
' ) *
Xi |p := p, X 1 , . . . , X m .
i
x i

Clearly this satisfies M = . Now to prove smoothness we need to

find a smooth atlas for T M . Take a chart (U, ), define the following function
: 1 (U ) R2m as;
' ) *

( Xi |p ) = (p), X 1 , . . . , X m .
i
xi

Now the image of this map is (U ) Rm which is an open subset of R2m . These
maps will be our charts. Also notice that;
'
1 (x1 , . . . xm , v1 , . . . vm ) = vi | 1 .
i
xi (x)
Now to check smoothness take two different charts of M , (U, ) and (V, )
this corresponds to charts on T M , ( 1 (U ), ) and ( 1 (V ), ).
Now we would
like to see what these maps do on ( 1 (U ) 1 (V )) = (U V ) Rm and
1 (U ) 1 (V )) = (U V ) Rm so first consider
(
& (
'
1 1 m 1 m
(x , . . . , x , v , . . . v ) = vi i |1 (x)
i
x
) 1
*
= (x), v 1 , . . . v m .

This map is just 1 idRm which is clearly a diffeomorphism by the

fact that and came from a smooth atlas. + ,
Thus T M is a smooth manifold with atlas 1 (U ), . (To prove
I
that T M is a (not smooth) manifold is a matter of simple point-set topology).

Now that we know that T M has a known structure we can look a little closer
at smooth vectorbundles, and see what maps between spaces are.
Definition 3.6. We call a pair (f# , f ) a smooth bundle map between (E, M, )
and (E # , M # , # ) if,
f : M M # and f# : E E # are smooth maps.
f# |Ep : Ep Ef# (p) is a linear map for all p.

# f# = f .

Remark 3.4. The class of all smooth vectorbundles with smooth bundle maps
forms a category Bund.

10
Theorem 3.3. T : Man Bund is a covariant functor. With T (M ) := T M and
T (f ) := (f# , f ) such that f# |Ep = f : Tp M Tf (p) M # is the pushforward of
f.
Proof. Note that Ep = Tp M .

(i) First let us consider what T does on the identity map, We know by Lemma
3.2 that id = id, and by the definition of T we get that for all Ep f# |Tp M =
id thus f# = id. And obviously (id, id) is the identity map on the smooth
vectorbundles.

(ii) Next we will show that T (g f ) = T (g) T (f ); proof by commuting

diagram.

M
f
! M# g
! M ##

T T T

" (f# ,f ) " (g# ,g) "

TM ! TM# ! T M ##

11
4 Cotangent bundle and differential forms
4.1 Cotangent spaces
In the previous section we have seen that tangent spaces are naturally endowed
with a vectorspace structure. In linear algebra it is common to speak of the dual
space of a vectorspace, namely all the linear functions from the vectorspace to a
field. In this thesis we will take the field to be R. We thus obtain the following
definition,

Definition 4.1. The cotangent space of a manifold M at a point p is defined

as, Tp M := (Tp M ) , in other words it is the dual space of the vector space
Tp M .
Remark 4.1. The vectors in the dual space are commonly called covectors.

Now a general result in finite dimensional linear algebra is that the dual
space is again a vector space with the same dimension as the original space.
Another important result is the following lemma,
Lemma 4.1. Let V be an n-dimensional vectorspace and {v1 , . . . vn } a basis
for V , then the covectors {i } such that i (vj ) = i,j , form a basis for the dual
space V of V .
Remark 4.2. We can apply the above definition to our vector spaces Tp M and
Tp M as such,

{ x |p }i=1...n is a basis for Tp M and {dxj |p }j=1...n is the basis for Tp M such
i + ,

that, dxj |p x j
| p = i,j .

We call dxi |p differentials.

Analogue to the case where we determined the maps that are induced by the
tangent functor we need some sort of pushforward for the cotangent spaces.
We can almost do this, however the direction is reversed and we call this a
pullback.
Definition 4.2. Let M and N be smooth manifolds, and f : M N a smooth
map. Then the pullback of f denoted as f is a function from Tf(p) N Tp M
which does the following
& ( & (
' '
f Xi |f (p) = f Xi |p .
i
xi i
xi

Which is well defined since lives in the cotangent bundle on N and f

in the cotangent bundle of M .
Example 4.1. Let f : M R a smooth map, then on p M we have a
function f : Tp M R which is clearly an element of Tp M thus we can write

12
 -n
it as i=1 i dxi |p . Now we can see what f does on a vector xj |p :

' n

f |p = i dxi |p | p = j .
xj i=1
x j

f
However we also have that f x j
|p = f [1 (x + tei )] = [f 1 (x + tei )] = xi .
So all together we get,

'n
f i
f = dx |p .
i=1
xi

4.2 Cotangent bundle

In the last section we have tried to put, for an m-dimensional smooth manifold
M , all tangent spaces into one bigger space of dimension 2m. We can do exactly
the same with the cotangent spaces.

Definition 4.3. The Cotangent bundle of a smooth manifold M is defined as,

%
T M := Tp M.
pM

Proposition 4.1. For M an m-dimensional smooth manifold we have that

T M is a smooth vectorbundle of rank m.
Proof. See the proof of Theorem 3.2
Proposition 4.2. T : Man Bun is a contravarient functor with, T (M ) =
T M and T (f ) = (f# , f ) where f# |Tp M = f .

Proof. The proof is identical to that of the theorem about the functoriality of
the functor T .

4.3 Smooth vector fields and smooth sections

From calculus we know what a vectorfield is, namely a function that sents every
point in some Euclidian space to a vector. It is possible to do exactly that with
manifolds and tangent bundles.
Definition 4.4. A smooth vectorfield is a smooth function X : M T M such
that X = idM .

Remark 4.3. It follows from the definition that X sents each point p in M to
a tangentvector in the tangentspace Tp M . In this way it relates to what was
previously mentioned about Euclidian vector fields.

13
 Of course now that we know that the cotangent spaces also form a vector-
bundle we can repeat this definition. However, it turns out we will use this
definition far more often and thus it gets a different name.
Definition 4.5. A differential 1-form is a smooth vector field from M to
T M .
Remark 4.4. The space of all differential 1-forms is usually denoted 1 (M )
The fact that there is a 1 in the denotion of all the differential 1-forms makes
it seem as though there is something as a differential k-form. Indeed we will
see in the next chapter that these exist and are of great importance to the de
Rahm Cohomology.
It is clear that the differential 1-forms and the vector fields have a lot in com-
mon, after all they were defined in almost exactly the same way. However there
is one big difference, namely that of the pullback-property of the differential
1-forms.

Definition 4.6. Let f : M N a smooth map and X(p) a vectorfield, then

we define the pullback on differential 1-forms as such, for 1 (N ):

f (X(p)) = (f (X(p))) 1 (M ).

Now if we attempt to do the same with vector fields and pushforwards we

have a problem, namely if we have a function f : N M we get that a vector
field on N (which sents each point of N to a vector) might not be injective, and
thus we get on the pushed-forward vector field that one point in M admits to
multiple vectors, which does not create a vectorfield. Also, if f does create a
vector field on M , all points in M should be hit by f , after all a vectorfield is
a function from ALL of M to T M . Thus we need a smooth bijection in order
for a vector field to be pushed forward, this is of course only the case if f is a
diffeomorphism. While on differential 1-forms no restrictions are necessary.

14
5 Tensor products and differential k-forms
As mentioned before, we will now generalize the notion of a 1-form. We can see
the 1-forms as something one dimensional. It turns out differential forms are
very natural to integrate over. However we can only integrate maximum forms,
for instance if we want to integrate a form over the sphere, it turns out we will
need a 2-form, 2 being the dimension of the sphere. However we will need a few
more definition before we can see what a differential k-form is.

5.1 Tensors
Definition 5.1. A covariant k-tensor on V T is a (multi)linear function
T : V V R. Where V is a vectorspace.
. /0 1
k times

Remark 5.1. k in the previous definition is called the rank of the tensor.
Remark 5.2. If we take V in the definition to be a dual vector space we call
the tensor contravariant on V .
We denote the collection of all covariant r-tensors on V , T r (V ) and the set
of all contravariant r-tensors Tr (V ). These spaces are vectorspaces. Now we
would like to find a way to multiply two tensors to make a new one. Naturally
this is hard to do in vectorspaces in general. However in R multiplication is as
easy as it gets. Thus the tensorproduct is defined as follows,

Definition 5.2. Let T : V n R and S : W m R be two contravariant n-

respectively m-tensors. Then
T S(v1 , . . . vn , w1 , . . . wm ) := T (v1 , . . . vn )S(w1 , . . . wm ).
Remark 5.3. It is easy to see that now the tensorproduct of two covariant
tensors on V is again a covariant tensor on V . With the rank of the new tensor
the sum of the ranks of the original tensors.
Remark 5.4. Do notice that taking the tensor product is not a commutative
action.
The tensorspaces of any vector space V are, as noted before, vectorspaces.
These vectorspaces are nr dimensional, where r is the rank of the tensors in
the space and n the dimension of V . But since this is a finite dimensional
vectorspace we can find a basis for T r (V ),
Theorem 5.1. Let V be a vectorspace with dual basis {1 , . . . n } then,

{i1 ir : ij {1, . . . n},

is a basis for T r (V ).

15
5.2 Symmetric and alternating tensors
The fact that tensors are multilinear functions does not mean we can interchange
any arbitrary argument with another. However if the tensor can we call it
symmetrical. If we can interchange any arbitrary argument with another and
the result is -1 times the unchanged tensor, we call the tensor alternating.
Now clearly not every tensor is symmetrical or alternating, however we can
always make a tensor symmetrical or alternating in the following way,
1
-
Definition 5.3. Sym T = r! Sr T .

Now this definition needs some explaination, first of all what T is. Basically
it is T where we interchange the arguments according to the cycle . We sum
over all the different cycles and we then divide by the amount of cycles in Sr .
In a way what we are doing is taking the average over all permutated tensors.
Now it follows that any interchanging of arguments, does nothing on the tensor
Sym T , because (ij)Sr = Sr .
In the same way we can define a function that makes a tensor alternating.
1
-
Definition 5.4. Alt T = r! Sr ,() T.

Where , is the signfunction for permutations. Now that we have the Alt
function we can define one of the most important operations in differential
geometry.
Definition 5.5. Let and be tensors on V , then
= Alt .
This operation is called the wedge product. The set of all alternating r-tensors
on V is usually denoted r (V )

5.3 Some algebra on r (V )

Since we will be working with the space of alternating tensors a lot we will use
this section to do some computations that will make it easier to work with it.
Lemma 5.1. Let T , S, R be t-, s- respectively r-tensors then,
(i) T (S R) = (T S) R.
(ii) (T + S) R = T R + S R.
(iii) T S = (1)ts S T.
(iv) T T = 0.

Proof. We will proof only the most important part of the lemma, namely
T T = 0; For one forms this is trivial since = = 0.
Now for higher forms the same principle applies, for every permutation we get
in the sum there is the antipodal permutation which we can get to by switching
around arguments (and since its alternating tensors we just multiply the result
by -1). This will end up analogously to the result for 1 forms with 0.

16
Lemma 5.2. For V a vectorspace with covector basis {1 , . . . , n }, r (V ) is a
(sub)vectorspace (of T r (V )) with basis

{i1 ir : 1 i1 < < ir n}.

Remark 5.5. Notice that there is a strict inequality in the definition of the
basis. This is, of course, because of property (iv) of Lemma 5.1.
Remark 5.6. With Lemma 5.1 and the basis defined as above we can see that
for a vectorspace with dimension n we get that n+i (V ) = 0 for all i N.

5.4 Tensor bundles

Now naturally, our next step is to apply the tensor theory to vectorspaces we
know and have worked with, namely tangent spaces of manifolds. We of course
take a look at the covariant r-tensors of the tangentspace at a point p M ;
T r (Tp M ). As we know these spaces are vectorspaces themselves and as such it
is possible to define a vectorbundle;
2
Definition 5.6. T r M := pM T r (Tp M )

As with the tangent bundle we can also show that this space is in fact a
smooth vector bundle of rank m! over M . We call it the covariant r-tensor
bundle. A useful identity is T 1 M = T M . We can now, as we did before define
tensor field, which are basically smooth sections in T r M .
Definition 5.7. A smooth tensor field is a smooth function : M T r M
such that = idM .
If we restrict ourselves to all the alternating tensors, rather than all of them
we end up with the tensorbundle r M which is a subvectorbundle of T r M and
also our main objective in this section.

Definition 5.8. %
r M = r (Tp M )
pM

Just like we can pullback smooth vector fields we can also pullback tensor-
fields.

Definition 5.9. Let f : M N be smooth and Xi Tp M , we have f :

T k (Tf (p) N ) T k (Tp M ) defined as,

f (X1 , . . . , Xk ) = (f X1 , . . . , f Xk ).

Remark 5.7. This new notion of pullback is more general than the one from
Definition 4.6 since T (Tp M ) = T 1 (Tp M ).

17
Lemma 5.3. Let f : M N , g : N P be smooth functions. p M .
T k (Tf (p) N ) and T l (Tf (p) N ). Then,

(i) f is a linear map.

(ii) f ( ) = f f .
(iii) (g f ) = f g .

(iv) id = id.
(v) f induced a smooth bundle map in the obvious way.

Now finally we have the means to define differential r-forms,

Definition 5.10. A differential r-form on a manifold M is a smooth section on
r M . And the space of all differential r-forms is called r (M ).

18
6 Differential forms
This section is devoted to differential forms and operations on them. As men-
tioned before differential forms are crucial in studying the de Rham Cohomology
of a manifold. We have already seen that differential forms are smooth sections
of covariant tensorbundles but they also have a very physical intepretation,
namely they will turn out to be very natural to integrate over a manifold.

6.1 Contractions and exterior derivatives

The first operation we will take a look at is contracting a differential form, this
operations sends a differential k-form to a differential k 1 form, in a quite
natural way.
Definition 6.1. The contraction (with X) of a differential k-form , is a linear
function iX : k k1 where X is a smooth section M T M defined by,

iX := (X, , , )
This basically means we fix the first coordinate of our smooth section and
thus leave k 1 free arguments. The contraction admits to a few properties,
Remark 6.1. Let r (M ), s (M ) and X : M T M a smooth section.
Contracting is also lineair in its smooth section argument.
iX iX = 0.
iX ( ) = (iX ) + (1)r (iX )

Remark 6.2. This last property is also called anti-derivation, for reasons that
will soon be made clear.
The next operation on differential is probably the most important for the
De Rham Cohomology. We will formulate the existence of these operators as a
theorem. But first a bit of notation.

I dxI = i1 ...m dxi1 dxim , I = (i1 . . . ik ).

Theorem 6.1. Let M be a smooth manifold. Then there are linear maps
dk : k (M ) k+1 (M ) for all k 0 such that:
(i) If f 0 (M ) then d0 f = f .
(ii) If k (M ) and l (M ) then

dk+l ( ) = dk + (1)k dl .

(iii) di+1 di = 0 i 0.

19
Proof. We will proof this theorem in two steps.
1 First assume M has an atlas with exactly one chart, and let (x1 , . . . , xm ) be
its coordinates. We can now define dk as such;
' m
'' J
dk ( J dxj1 dxjk ) = dxi dxj1 dxjk .
i=1
xi
J J
I I
Or: d(f dx ) = df dx . However this only works for increasing indices I since
they form the basis of our k (M ). This map is-clearly linear (by linearity of
m
partial derivation), and it satisfies (i) by f = i=1 f i
x dx .
Now d(f dxI ) = df dxI works for not only increasing indices but, in fact, for
all indices. To see this consider the permutation that send an index J to
an increasing index I. We thus get
d(f dxJ ) = ,()d(f dxI ) = ,()df dxI = df dxJ .
Now we can prove (ii), let = f dxI , = gdxJ
d( ) = d(f gdxI dxJ )
= d(f g) dxI dxJ
= (gdf + f dg) dxI dxJ
= (df dxI ) (gdxJ ) + (1)k (f dxI ) (dg dxJ )
= d + (1)k d

By linearity we can extend this to any differential form.

For (iii) consider first, for a 0-form f :

k '
' k
2f
d(df ) = i xj
dxi dxj
i=1 j=1
x
' # 2f 2f
$
= i xj
j xi
dxi dxj = 0.
i<j
x x

Now by (ii) we get that

' '
d(d) = d( dJ dxJ ) = d(dJ ) dxJ = 0.
J J

2 Now let M be any manifold. We have now shown that we can define a
differential operator for all charts. Now we would like to see that they coincide
on overlapping charts, so let U and V be two charts. My claim is that
dU |U V = dV |U V , thus we can find a single d that satisfies the properties
mentioned above for the whole M rather than just a single chart. Namely by
(d)p = (dU |U )p ) p U.

20
Lemma 6.1. The above mentioned differential operator is unique.
Proof. Let d and d# be two operators that satisfy the conditions of Theorem 6.1.
Note that by (i) we get that on 0-forms d = d# , after all df = f = d# f . Now by
induction assume that the property holds for differential n 1 forms. Look now
at d(f dxi1 |p dxin |p ) = df dxi1 |p dxin |p = d# f dxi1 |p dxin |p =
d# (f dxi1 |p dxin |p ). Now by linearity this extends to arbitrary differential
forms.

Proposition 6.1. The differential operator commutes with pullbacks of maps,

let f : M N
f d = df .
for k (N )
Proof. Let k (N ), we need only check the above locally since d works
locally. So take p N Now by linearity of both f and d we need only check
this property for gdxi1 |p dxik |p . Now,
) * ) *
f d gdxi1 |p dxik |p = f dg dxi1 |p dxik |p
= d(g f ) d(xi1 f )|p d(xik f )|p
) *
= d (g f )d(xi1 f )|p d(xik f )|p
) *
= df gdxi1 |p dxik |p .

6.2 Integrating over topforms

One of the (possibly) surprising facts about differential forms is, that theres a
very natural way to integrate over them. In fact the construction of differential
forms started all because of the need for a unique way to integrate over a man-
ifold. Normally integrating with two different parametrizations could result in
two different integrals, something you would like to avoid.
Definition 6.2. Let M be an n-dimensional smooth manifold, and let
{E1 , . . . En } and {E1# , . . . En# } be two bases for some p M . We say that two
basis are consistently oriented if the transitionmatrix between them has positive
determinant. This forms an equivalence relation, and the equivalence classes are
called orientations. We call a manifold with a given basis for all p an oriented
manifold. All the bases that are in the equivalence class of the given basis are
called positively oriented and those that are not are called negatively oriented.
Remark 6.3. A manifold M is called orientable if there exists an orientation
for it. There are manifolds that are non-orientable, such as the M
obius Strip.

21
 An important property of orientable manifolds is that there always exists a
non-vanishing topform that is positively oriented at each point. (The value of
the form evaluated at a positively oriented basis is > 0). We call such a form
an orientation form.
We need the orientation to determine the sign of the integral, after all we
would like integration over the top half of any euclidian space of a positive
function to be positive.
Definition 6.3. Let U be an open subset of some euclidian space and f dx1
dxm a (compactly supported) topform on U . Then we say;
3 3
f dx1 dxm := f dx1 dxm
U U

i.e. normal Lebesgue integration.

Definition 6.4. Let M be an n-dimensional oriented manifold with one coor-
dinate chart (U, ), and let n (M ). Then we say
3 3
= (1 ) .
M (U )

Where we use the previous definition to evaluate the integral over a differential
form on a subset of a Euclidian space.
Remark 6.4. Note that the previous definition does not rely on the choice of
coordinate (We will not proof this as I do not think it gives great insight).
We can now define the integral over any manifold of a differential form with
compact support. Still we need the compact support as will become clear in the
definition. Note that if {(U , )} is an atlas for M this is an open cover for
every subset of M , as such also for the compact support of a differential form.
But by compactness only a finite amount of charts is needed to cover it. As such
consider (Ui , i )N
i=1 to be this finite collection. And {i } to be its partition of
unity.
Definition 6.5. Let M, be as above. Then
3 'N 3
= i .
M i=1 M

Now later on we will need integration in a more general sense, namely in-
tegrating forms over manifolds with boundary or even manifolds with corners.
In these cases not a lot changes apart from the open sets (in Euclidian spaces)
that could differ when dealing with boundaries or corners.
Now a final lemma which will be used later on;
Lemma 6.2. Let M be an orientable manifold, and 0 an orientation form.
3
0 > 0.
M

22
7 Cochains and cohomologies
This chapter is all about basic algebraic topology, since the de Rham Cohomol-
ogy will turn out to be a cohomology theory it is important to see what that
means. Therefore we will work towards this concept step by step.

7.1 Chains and cochains

In this subsection we will not be talking about manifolds at all. Mostly we will
talk of modules (over a general ring R or Z).
Definition 7.1. A sequence C = (Cn , n | n Z) of modules Cn and homo-
morphisms n : Cn Cn1 is called a chain complex, if for all n Z we have
that n1 n = 0 holds.
Remark 7.1. The n functions are usually called the boundary operators or
differentials.
Remark 7.2. A chain complex is usually visualised in a diagram as such,

! Cn
n+1 n
! Cn1 !
n1

An important observation is that since n1 n = 0 we immediately get

that Im(n ) Ker(n1 ). Note that these are both submodules of Cn
Definition 7.2. The n-cycles of a chain complex C is;

Zn (C ) = { Ker(n )}.

Definition 7.3. The n-boundaries of a chain complex C is;

Bn (C ) = { Im(n+1 )}.

Definition 7.4. The n-th homology module of a chain complex C is;

Hn (C ) = Zn /Bn .

Definition 7.5. Let C = (Cn , cn ) and D = (Dn , dn ) be chain complexes.

Then a chainmap f : C D is a sequence of homomorphisms fn : Cn Dn
such that;
Cn
cn
! Cn1

fn fn1

" "
Dn ! Dn1
dn

commutes for all n.

Now naturally it is possible that after taking the homology quotient nothing
is left. This is a special kind of chaincomplex,

23
Definition 7.6. We call a chaincomplex(sequence) exact if Zn = Bn for all n
Remark 7.3. An exact sequence has only trivial homology modules.
Remark 7.4. The homology module is nothing more than a quotient, so a
chainmap can be defined on homology level as fn ([c]) = [fn (c)].
Because we will naturally apply the above definition to topological spaces,
specifically manifolds we would like homologies to be invariant under homotopy.
The first step into this might seem a bit weird but it turns out this definition is
what were looking for.

7.2 Cochains
Sometimes working with decreasing indices doesnt quite cut it. Therefore we
define a cochain complex to be a chaincomplex with increasing indices. Ana-
loguous you can define a cochain to consist of the modules of R-linear maps from
your modules to R, together with special (boundary) maps dn : C n C n+1 .
Remark 7.5. The co part of cochain can be related to the dual of a module in
the same way a vector space relates to its dual. This, together with a fitting
boundary operator would also give us a cochain. For now however the definition
as above is more useful and easier to work with.
Definition 7.7. A chain homotopy s from f : C D to g : C D is a
sequence sn : C n Dn1 such that, if c and d are the boundary maps for
C and D respectively, then;

dn1 sn + sn+1 cn = g n f n .

Remark 7.6. f and g are called (chain)homotopic maps.

Theorem 7.1. If f and g are homotopic cochainmaps they are equal on the
homology modules of the domaincomplex.
Proof. To see that this is satisfied consider [] Hn (C ), now

f n [] g n [] = (f n g n )[]
) *
= dn1 sn + sn+1 cn []
= [dn1 sn ] + [sn+1 cn ]
= 0 + [sn+1 0] = 0.

The first term is zero since it is mapped into the image of dn1 which on
homology level is 0. The second part is 0 because is in the Ker(cn ) by
definition. Thus we conclude that f and g are equal on homology level.

24
7.3 A few useful lemmas
Before we get into the De Rham Cohomology it may be useful to state a
few lemmas regarding (co-)chains. These will occur often in proofs and it is
therefore essential that they are mentioned.

First, a few observations regarding injectivity and surjectivity of maps in a

chain. Consider the following exact sequence,
0! a!
0 A B
from the fact that this sequence is exact we can deduct that a is an injective
function, after all Ker(a) = Im(0), and R-linear maps with trivial kernel are
injective. In the same way you can prove that for
b! 0!
B C 0
it can be deducted in the same way as before that b has to be surjective if the
sequence is exact.
Remark 7.7. It can also be proven that the reverse holds, if a injective or b
surjective then the sequences are exact.
Remark 7.8. The 0-map will from hereon be omitted since it is the only
possible map there could be in that spot.
Combining the above two sequences into one gives us a special kind of exact
sequence.
Definition 7.8. A exact sequence of the form,

0 ! A a!
B
b!
C ! 0

is called a short exact sequence, often abbreviated by ses.

Remark 7.9. We can extend the concept of a short exact sequence of modules
to a short exact sequence of (co)chain complexes, namely by saying

0 ! A ! B
a b
! C ! 0

is short exact if,

0 ! An ! Bn
an bn
! Cn ! 0

is short exact for all n.

Another important sequence is the following.We will omit the proof and
merely refer to the previous results.

0 ! D =!
E ! 0

Now we will discuss some important lemmas in algebraic topology which will
be useful in the future.

25
Lemma 7.1. (The Five Lemma) Consider the following commutative diagram
with exact rows,
!
! C
! D ! E
A B

a b c d e
" " # " # " "
#!
! C#
! D# #!
A# B# E#
If a, b, d and e are isomorphisms then so is c.
Proof. First we will prove injectivity of c. Consider x C such that c(x) = 0.
Then # c(x) = d (x) = 0. This implies that (x) Ker(d), by injectivity of
d we now get that (x) = 0 thus x Ker() = Im(). Thus there is a y such
that x = (y). Now c (y) = # b(y) = 0, thus b(y) Ker( # ) = Im(# ).
Now we get that b(y) = # (z # ), by surjectivity of a we obtain b(y) = # a(z) for
a unique z A. By commutativity, b(y) = b(y). Thus finally x = (z) = 0
by exactness.
Now for surjectivity consider x# C # . Look at # (x# ) by surjectivity of d we
get that there is a y D such that d(y) = # (x# ) now by exactness # d(y) = e
(y) = 0. By injectivity of e we thus get that (y) = 0 and y Ker() = Im().
Thus there exist an x C such that (x) = y. Now by # c(x) = d(x) = d(y).
Now c(x) and x# both map to d(y) under # , we can thus consider the difference,
# (c(x) x# ) = 0 to obtain (c(x) x# ) Ker( # ) = Im( # ). Thus there is a
v # B # such that # (v # ) = c(x) x# , by surjectivity of b we get a v B such
that b(v) = v # thus, # b(v) = c (v) = c(x) x# c(x (v)) = x# and
thus there is an element of C such that it maps to x# for all x# C # , this proves
surjectivity.

Lemma 7.2. (Splitting Lemma)

Let 0 ! A a!
B
b!
C ! 0 be a short exact sequence. Then
equivalent are,
(i) The image of a is a direct summand of B.
(ii) There is a homomorphism r : B A such that r a = id.
(iii) There is a homomorphism s : C B such that b s = id.
We call such a sequence a splitting sequence.
Lemma 7.3. (Zigzag Lemma) Given a ses of cochains

0 ! A a!
B
b!
C ! 0

then for all n there exists a connecting homomorphism : H n (C ) H n+1 (A ),

such that the following sequence is exact,
an bn an+1
H n (A ) H n (B ) H n (C ) H n+1 (A )

26
Proof. Consider the following commuting diagram with exact rows:

0 ! An an ! Bn bn ! Cn ! 0

d d d

" " "

0 ! An+1 !
an+1
B n+1 !
bn+1
C n+1 ! 0

d d d

" " "

0 ! An+2 !
an+2
B n+2 !
bn+2
C n+2 ! 0

) *1 1
We would like to make the map an+1 d (bn ) , which works on the
level of cohomologies due to the fact that a and b commute with d and thus
send boundaries to boundaries and cycles to cycles (more on that later). Now
the question is of course if this map as we defined above is well defined, and
independant of choice.

So take an element n C n , because bn is surjective there exists a n

such that bn n = n . But since we are only interested in working on the coho-
mology level it suffices to consider n such that dn = 0. Hence by commutivity
we get that, dbn n = bn+1 dn = 0. Thus dn Ker(bn+1 ) = Im(an+1 ), thus
there exists a (unique by injectivity) n+1 such that an+1 n+1 = dn . Now we
get by commutivity of the diagram that an+2 dn+1 = dan+1 n+1 = ddn = 0,
thus dn+1 ker an+2 but since an+2 is injective it follows that dn+1 = 0
which of course means that the procedure weve followed ends up at a repre-
sentative of a cohomology class.

However we are not completely done, we have to show that the output
doesnt depend on the choice we made for n (upto an element of the form
dn# ). Furthermore it is not yet clear that respects the homology structure.
So let us start with choosing a different n# .

Consider n n# since both map to the same point under bn we get that
bn (n n# ) = 0. By exactness there exists a n such that an n = n n# .
Now by commutativity d(n n# ) = an+1 dn . By previous results we
know there exist x, x# such that dn = an+1 x and dn# = an+1 x# . Now
consider; an+1 (x x# dn ) = 0, by injectivity we then get x x# = dn .
Thus on homology level x x# = 0, and thus maps to 0.
Now take an element n = dn Cn . Now by surjectivity of bn1 we
obtain a n1 such that bn1 n1 = n1 . Now by commutativity of the

27
 diagram we get that dn1 = n . Thus dn = 0. Now if we follow the
defining process of we at some point obtain an+1 n+1 = dn = 0, by
injectivity this means that n+1 = 0. Thus we see that all elements in
[n ] (whose difference is a boundary) are mapped onto the same [n+1 ]
which implies well definedness of .

Now last of all we need to prove exactness of the sequence in the lemma,

an bn an+1
H n (A ) H n (B ) H n (C ) H n+1 (A )

As for exactness of the sequence we will suffice with proving exactness at H n (A),
as the proof for H n (B) is trivial and the proof at H n (C) similar.
Take an element [c] H n (A) and apply an to this element. By following the
defining process we see that an [c] = [dn1 ] = [0]. Thus Im() Ker(an ).
The other way around let [n ] be an element in Ker(an ). Thus we have that
an n = dn1 . We can (as a sketch) inverse the boundary operator to see;
1 ) *1 1
((an ) d bn1 ) = bn1 d1 an .

Now since we know that an n = dn1 we can find that the image we seek is
bn1 n 1. And thus finally we get Ker(an ) Im().

28
8 The de Rham cohomology
We are now in a postition to define the de Rham Cohomology.

8.1 The definition

Now we want to apply the previous chapter of basic homologytheory to mani-
folds. For this we need, as seen before, R-modules. We will use for R-modules
in this case real vectorspaces, namely the vector spaces or differential k-forms
k (M ), for all k = 1 . . . m, if m is the dimension of the manifold.
Definition 8.1. A differential form is called closed if d = 0. And a differ-
ential k-form is exact if there exist a differential (k 1)-form such that
= d.
Remark 8.1. Note that because d d = 0 we have that all exact forms are also
closed.
Lemma 8.1. Let M be a smooth manifold, the following diagram is a cochain;

d ! n1 (M ) d!
n (M )
d!
n+1 (M )
d !

Remark 8.2. The n-cycles are exactly the closed n-forms on M . And the
n-boundaries are the exact n-forms on M .
Remark 8.3. We also use the fact that n = 0 for n > dim(M ) and n < 0.
Proof. Note that n (M ) is a real vector space, thus an R-module. And d is a R-
linear map. Furthermore dd = 0 which is the same as saying Im(d) Ker(d).
Thus we are dealing with a cochain.

Definition 8.2. The p-th de Rham cohomology group is equal to the p-th coho-
p
mology groups of the cochain in Lemma 8.1. This is usually denoted HdR (M ).

Lemma 8.2. Let M and N be smooth manifolds and f : M N a smooth map

then the pullback sends closed forms on N to closed forms on M and exact forms
p p
on N to exact forms on M . Thus f has a pullback f : HdR (N ) HdR (M ).
Proof. Let be a closed form on N then d = 0. Now by Proposition 6.1 we
have
0 = f d = df
thus f is closed. Now let be an exact form, thus = d then we get, again
by Proposition 6.1,
f = f d = df
and thus f is closed.

29
 p
Remark 8.4. The pullback as defined above is sometimes denoted HdR (M )(f ).
With the remark we can now prove the functoriality of the de Rham Coho-
mology functor;
p
Theorem 8.1. HdR () is a contravariant functor from MAN to R-MOD.
Remark 8.5. Remember that R-MOD is the category of all modules over the
ring R.

Proof. It will be sufficed to prove that (F G) = G F and id = id. The
first part follows almost immediatly from the commuting property of pullbacks
with the differential operator. The second part will follow later on in the chapter
of homotopy invariance.

8.2 Homotopy Invariance

The definition of the de Rham cohomology groups may be clear now but it is
not clear yet how to compute them or even work with them. That is why we
will use the next sections to make it easier to compute these groups. First of
all we will prove homotopy invariance.
We have seen before, that chainmaps that are homotopic induce the same
maps on homology level. So what we want to find for each homotopy

H : M I N,

a (chain) homotopy operator (hn ) : n (N ) n1 (M ). This will suffice be-

cause of Theorem 7.1. Often we shall omit the subscript as with the differential
operator to make it a little less confusing.
Definition 8.3. Two smooth maps f, g : M N are called smoothly homotopic
if there excists a smooth map H : M I N such that H(x, 0) := H0 (x) = f (x)
and H(x, 1) := H1 (x) = g(x).

Lemma 8.3. Let M be a smooth manifold and i0 : M / M I the inclusion

on the zero level and i1 : M / M I the inclusionmap on the 1 level. Then
there exists a chainhomotopy h between i0 and i1 .
Proof. We shall define the chainhomotopy as such, let k (M I), and let
X1 , . . . , Xk1 Tp M ;
3 1 # $

(h)p = , X1 , . . . , Xk1 dt
0 t
Which basically boils down to integrating over a coefficient function. Now
to check that h(d) + d(h) = i1 i0 . Consider two cases, one in which there
is a dt term in the differential form, and one where there isnt.

30
 Case 1; = f (x, t)dt dxi1 dxik1 .

##3 1 $ $
i1 ik1
d(h) = d f (x, t)dt dx dx
0
'# 3 1 $
= j
f (x, t)dt dxj dxi1 dxik1
j
x 0
#3 1 $
f
= j
(x, t) dxj dxi1 dxik1 .
0 x

Now for the other term because dt dt = 0;

# $
f j i1 ik1
h(d) = h dx dt dx dx
xj
3 1
f ) *
= j
(x, t)i dxj dt dxi1 dxik1 dt
0 x
t
#3 1 $
f
= j
(x, t)dt dxj dxi1 dxik1
0 x
= d(h)

Now it follows that the sum of the terms above adds up to zero. But since
i1 dt = i0 dt = 0 the homotopic equivalence relation holds.
Case 2; = f (x, t)dxi1 dxik . Now because i = 0 we get that
t
h(d) = 0, as for the other part,
# $
f
h(d) = h dt dxi1 dxik + some terms without dt
t
#3 1 $
f
= (x, t)dt dxi1 dxik
0 t
= (f (x, 1) f (x, 0))dxi1 dxik .

Now since f (x, s)dxi1 dxik = is we get that;

(f (x, 1) f (x, 0))dxi1 dxik = i1 i0 .

Which proves the homotopic equivalence relation h(d) + d(h) = i1

i0 .

31
Theorem 8.2. Let M and N be smooth manifolds and f, g : M N be
smoothly homotopic maps. Then,

f = g : HdR
n n
(N ) HdR (M ).

Proof. First note that if it : M / M I then H i0 = f and H i1 = g, and

let h be the homotopy operator of Lemma 8.3. Define

= h H : n (N ) n1 (M ).
h
For any n (N ) we have,

h(d) = h(H d) + d(hH ) = hd(H ) + dh(H )
+ d(h)
= i1 H i0 H
= (H i1 ) (H i0 ) = G F .

Now by Lemma 7.1 we get the desired result.

Definition 8.4. Let f : M N and g : N M be smooth maps between

smooth manifolds, then if f g is smoothly homotopic to the identitymap on N
and g f is smoothly homotopic to the identity map on M , we say that M and
N are (smoothly) homotopic equivalent, sometimes denoted as M 2 N .
n
Theorem 8.3. Let M and N be manifolds such that M 2 N . Then HdR (M )
=
n
HdR (N ) for all n.
Proof. M 2 N implies that there are f : M N and g : N M such that
f g 2 id and g f 2 id. But since we know that homotopic equivalent maps
induce the same map on de Rham cohomology level, we get

f g = (g f ) = id = id

And also
g f = (f g) = id = id
So clearly these maps are eachothers inverse which implies that f is an
isomorphism.

8.3 The Mayer-Vietoris sequence

Another important tool we may use in finding cohomology groups of manifolds
is the Mayer-Vietoris sequence. This tool looks a bit like the Van Kampen
theorem in homotopy theory and is used in a similar way. For the prove of
existence we will use the Zigzag Lemma (7.3)

32
Theorem 8.4. (Mayer-Vietoris) Let M be a smooth manifold, and U and V
open subsets of M such that U V = M , then for all n there is a connecting
n n+1
homomorphism : HdR (U V ) HdR (M ) such that the following sequence
is exact:
n k l
n n n i j n+1 k l
HdR (M ) HdR (U )HdR (V ) HdR (U V ) HdR (M )

where i, j, k and l are all inclusion maps as such,

U V
i ! U n (M )
k ! n (U )

j k l i
" " " j
"
V
l ! M n (V ) ! n (U V )

it may be useful to see that the pullback of inclusions is nothing but the
restriction of a differentialform.
Now before we start the proof of this powerful tool we need an additional
powerful tool. Namely the existence of a partition of unity for every open cover
of a manifold.
Definition 8.5. Let (U )I be an open cover of a smooth manifold M . A
collection (smooth) functions { : M R}I is called a (smooth) partition
of unity if:
(i) 0 1 for all .
(ii) The support of is contained in U .
(iii) Each point p M has a neighbourhood that intersects only a finite number
of supp( ).
-
(iv) I (x) = 1 for all p M .

The following theorem is most important, but the proof is rather technical
so we will suffice with just mentioning the theorem.
Theorem 8.5. If M is a smooth manifold, any open cover induces a smooth
partition of unity.
Now we can prove the Mayer-Vietoris theorem.
Proof. By the zigzag lemma it will suffice to prove that the following is short
exact:
k l i j
0 n (M ) n (U ) n (V ) n (U V ) 0
First we will prove exactness at n (M ), which means we have to show that
k l is injective. So take such that (k l ) = (|U , |V ) = (0, 0)
but since U V = M this implies that = 0 and this proves injectivity.

33
 To prove exactness at n (U ) n (V ) first consider
(i j ) (k l ) = (i j ) (|U , |V ) = |U V |U V = 0.
Thus Im(k l ) Ker(i j ).
Now for the other side consider (, # ) Ker(i j ) then i = j #
and thus |U V = # |U V this means there exists a form on M such
that |U = and |V = # now clearly (, # ) = (k l ) and thus
Im(k l ) Ker(i j ). This proves exactness.
Next is exactness at n (U V ) which translates to nothing but proving
that i j is surjective. So let n (U V ). Since {U, V } is an open
cover of M there exists a smooth partition of unity {, }. Now define
n (U ) as
4
on U V ;
=
0 on U supp .
4
on U V ;
# =
0 on U supp .

Then we obtain (i j )(, # ) = i j # = () = .

34
9 Some computations of de Rham cohomology
In this section we will calculate the de Rham cohomology groups of a few well
known manifolds. And some simple corollaries that follow from these facts.
A point as a topological space is usually denoted by *. If we want to de-
n
termine HdR () the first step is finding the spaces n (). We know that since
* is 0-dimensional that n () = 0 for all n > 0. As for 0 () this consists of
all functions R and as such 0 () = R. Now we can determine all the de
Rham groups;
Proposition 9.1. 4
n R n = 0;
HdR () =
0 n > 0.

Proof. Consider the chaincomplex of n ()s.

0 ! R ! 0 ! ...

The homology groups now become Ker (R0)

Im(0R) = R for n = 0 and 0 for all the
other n.

Lemma 9.1. (Poincare) Let M be a contractible manifold.

4
n R n = 0,
HdR (M ) =
0 n > 0.

Proof. If M is a contractible manifold we have that M 2 {} now by Theorem

n
8.3 we obtain HdR (M ) n
= HdR ().

Now the next thing to find is the de Rham groups of all the spheres. In
order to obtain this we will need a lemma which is in a way overkill for what
we will use it for. Nonetheless it is an important result that needs mentioning.
From topology we know what a disjoint union of sets 5 is. The disjoint union
of manifolds M1 and M2 is commonly denoted M1 M2 , and is called the
coproduct of M1 and M2 , important to realize is that this coproduct is again a
manifold. For we can take the union of any smooth atlas of M1 with a smooth
atlas of M2 , this new atlas is again smooth since there is no overlap of elements
due to M1 M2 = . Now this relates to de Rham groups in the following way;
Proposition 9.2. De Rham Cohomology is additive, which means for {Mi }iI
smooth manifolds that;
& (
6 7
H n
dR Mi = H n (Mi ). dR
iI iI

35
Proof. Let i be the inclusion maps of Mi / M then the isomorphism is given
by
, (1 , 2 , . . . ) = (|M1 , |M2 , . . . ).
Injectivity and surjectivity follow almost instantly.
Corollary 9.1. The de Rham groups of a space M that is smoothly homotopic
to two points are; 4
n R R n = 0,
HdR (M ) =
0 n > 0.

Proposition 9.3. Let M be a connected smooth manifold, then;

0
HdR (M ) = R.
) * ) *
Proof. HdR 0
(M ) = Z 0 ( (M )) / B 0 ( (M )) . Now 1 (M ) = 0, so
B 0 ( (M )) = 0. Thus,
0
HdR (M ) = Z 0 .
So all we need to look at is functions f : M R such that df = 0. Now since
M is connected it follows that df needs to be a constant function, thus Z 0
= R.

Now for a last lemma;

Lemma 9.2. Let

0 ! A0 !
d0
! Am
dm1
! 0

be an exact sequence of finite vectorspaces. Then

m
'
(1)i dim(Ai ) = 0.
i=0

Remark 9.1. This is a special case of the Euler characteristic of exact se-
quences.
Proof. Since we are working with finite vectorspaces we know that for all i,

dim Im(di ) + dim Ker(di ) = dim Ai .

Furthermore we know that, since this is an exact sequence

dim Im(di ) = dim Ker(di+1 ).

Now telescoping gives us the desired result.

36
 Now we have enough tools to compute the de Rham groups of a circle.
Proposition 9.4. 4
n R n = 0, 1;
HdR (S1 ) =
0 n > 1.

Proof. Denote N as the northpole of S1 and S as the southpole of S1 . Let

U = S1 N and V = S1 S. We can now apply Mayer-Vietoris to obtain;

0 ! H 0 (S1 ) ! H 0 (U ) H 0 (V ) ! H 0 (U V )
dR dR dR dR

i j
"
0 # 1
HdR (U V ) # 1
HdR 1
(U ) HdR (V ) # 1
HdR (S1 )

As an exact sequence. Since S1 is connected, and U and V are contractible, and

U V is homotopic equivalent to two points we can fill out the diagram as such.

0 ! R ! RR f
! RR

"
0 # 0 # 0 # 1
HdR (S1 )

1
Now by Lemma 9.2 we can compute the dimension of HdR (S1 ), namely we get;

1 + 2 2+ ? 0=0
1
Thus we obtain dim HdR (S1 ) = 1 and as such HdR
1
(S1 )
= R.

Now that we know what the de Rham groups of a circle are we can inductively
compute the groups of all spheres.
Theorem 9.1. For n > 0;
4
q R q = 0, n;
HdR (Sn ) =
0 other.

Proof. The main idea of the proof is that we assume the theorem holds for
n 1 and show that it then also holds for n. By Proposition 9.4 we already
know this holds for n = 1.
So assume the theorem holds for m = n 1 then consider N to be the
northpole of Sn and S to be the southpole. Now define U = Sn \S and V =

37
Sn \N . Since these two open sets cover Sn we can apply Mayer-Vietoris. Once
more U and V are contractible spaces. However this time U V is not merely
two points, it is in fact homotopic equivalent to Sn1 . Now by the induction
assumption we know what the de Rham groups of this sphere is. Additionally
we know that Sn is connected thus has HdR 0
(Sn )
= R.
Now at the parts of the Mayer-Vietoris sequence with 0 < q < n 1 we get
that it has the following shape;

0 ! H q (Sn ) ! 0 ! 0.
dR

q
Thus we get that for these values HdR (Sn )
= 0. So all we need to look at is the
following part of the Mayer-Vietoris sequence;

0 ! H n1 (Sn ) ! H n1 (U ) H n1 (V ) ! H n1 (U V )
dR dR dR dR

"
0 # n
HdR (U V ) # n
HdR n
(U ) HdR (V ) # n
HdR (Sn )

which filled in becomes;

0 ! H n1 (Sn ) ! 0 ! R
dR

"
0 # 0 # 0 # n
HdR (Sn )

n
This gives us the desired result HdR (Sn ) n1 n
= R and HdR (S ) = 0. Now the
last space for which we will compute the De Rham groups will give us a peculiar
conclusion.
Proposition 9.5.
4
q R q = 0, n 1;
HdR (Rn \{0}) =
0 other.

Proof. We have that Rn \{0} 2 Sn1 , we can for instance use the the map,
# $
1
(x, t) , (1 t) + t x.
7x7

So now when we use Theorem 8.3 and Theorem 9.1 the desired result follows.

38
Remark 9.2. Instead of 0, we could remove any point and this result would
still hold.
Now what we can conclude by this result is that by removing a point from
a space we have created differential forms that are exact but not closed. So in
a way this is addition by substraction.
We can also compute the top cohomology group of an orientable manifold.
This is in fact where the De Rham cohomology works a lot better than singular
cohomology (which will be introduced later).
Proposition 9.6. Let M be a smooth, connected, orientable and compact n-
manifold. Then
n
I : HdR (M ) R,
the integration map is an isomorphism.
Proof. First of all we need to show that this map 8is well defined. 8 In other
words we want for a closed differential form that M + d = M for all
(dimensionmatching)
8 differential forms . But by linearity of the integral and
Stokes theorem M d = 0. As such the identity holds for all (closed) differential
forms .
Next there is the issue of surjectivity, but since we know that when 8 M is
orientable there exists an orientation form 0 with the property that M 0 =
b > 0 (see Lemma 6.2). We can use linearity of integration I(a0 ) = aI(0 ) =
ab, and since b is non-zero we can make any real number in such a way.
For injectivity we use same the proof as in Corollary 11.1.

Remark 9.3. It turns out that if M is a smooth, connected, compact and

NON-orientable manifold the top cohomology group will be 0. Thus we have
found a way to check if a smooth, connected and compact manifold is orientable
or not, by using de Rham groups.

39
10 The de Rham Theorem
One of the most common homology theories in algebraic topology is the Singu-
lar Homology, this homology is often the first step into the world of algebraic
topology and hence the first thing learned in most basic courses of algebraic
topology. The central question in this section will be How does singular homol-
ogy relate to de Rham cohomology?, we will answer this question in steps. First
a brief recap of what singular homology (and cohomology) is, then additional
preperation working towards the main theorem of this section the de Rham
Theorem, which will give a direct isomorphism between Singular Cohomology
groups and the de Rham groups.

10.1 Singular Homology

Singular homology is an homology theory based on so called singular simplices.
Definition 10.1. Let R have the basis e0 , e1 , . . . . The standard p-simplex is
9 p :
' '
p = i e i : i = 1, 0 i 1
i=0

Remark 10.1. The i s are called the barycentric coordinates.

Definition 10.2. For - . , vn Rq , [v0 , . . . , vn ] is the map n Rq
given v0 , . .-
which works as such; i ei , i vi . This is called an affine singular n-
simplex.
The next definition gives a notation for mapping a lower standard simplex
into one of higher dimension.
Definition 10.3. The ith facemap Fip is the map [e0 , . . . , ei , . . . , ep ] : p1
p where the hat on a vector means you leave it out.
Definition 10.4. Let X be a topological space, a continous map p : p X
is called a singular p-simplex. The free abelian group over all p-simplices is
denoted Sp (X) and called the singular p-chain group.
Remark 10.2. So a p-chain of X is a formal sum of p-simplices.
Now as before, we can turn the graded groups p into a chaincomplex by
defining a differential p : p (X) p1 (X).
Definition 10.5. Let p : p (X) p1 (X) be the homomorphism that
works on the basis elements of p (X) as such, let be a p-simplex. And
(i) := Fip .
p
'
p := (1)i (i) .
i=0

Lemma 10.1. (p , p )p0 is a chaincomplex.

40
Remark 10.3. The homology groups of this chaincomplex are called the sin-
gular homology groups.
Do note the distinct difference between the de Rham cohomology and singu-
lar homology that the de Rham complex is a complex over R, since were talking
about real vectorspaces. On the other side singular homology are free abelian
groups, so modules over Z. So in order to properly link these two we need to
make the singular chains work over the field R (among other things).

10.2 Singular cohomology

Two very common functors in algebraic topology are the A functor and
the hom(, A) functors. These functors, when applied to the singular chain
groups, for A for instance an abelian group, or R-module, sometimes make the
homology groups easier to compute. We will focus on the hom(, A) functor
now to define the singular cohomology.
Lemma 10.2. hom(, A) is a functor from Ab to Ab.
Now we want to turn the chaincomplex of singular chains into a cochain,
but in order to do this we need a differential map.
Definition 10.6. dp : hom(Sp (X), A) hom(Sp+1 (X), A) which works as
such;
dp f = f p+1 .
Lemma 10.3. (hom(Sp (X), A), dp )p0 is a cochain complex.
Remark 10.4. The (co)homology groups of this cochain complex are called
the singular cohomology groups.

10.3 Smooth simplices

So far we have considered singular simplices as continuous maps from the stan-
dard simplices into our topological space. However since we are only considering
smooth manifolds in this thesis we will need to add differentiability requirements
for the maps.
Definition 10.7. Cp (X) is the set of all p-simplices.
Definition 10.8. Cp (X) is the set of all smooth p-simplices.
Now repeating the process of the first subsection of this chapter we can
define in the natural way, the smooth singular homology groups Hp (M ). Now
it would be preferable if these two homologies coincide for any smooth manifold.
Thankfully we can make such an identification. This identification uses a main
theorem in differential topology by Whitney.
Theorem 10.1. (Whitney Approximation Theorem) Let M and N be smooth
manifolds and F : M N be a continous map. Then F is homotopic to a
smooth map F : M N .

41
10.4 De Rham homomorphism
De Rham theorem states that de Rham cohomology groups are isomorphic to
singular cohomology with coefficients in R. We will prove this theorem in two
steps. First we will define the de Rham homomorphism which maps the de
Rham groups to the singular cohomology groups. Then we will show that this is
a isomorphism if the space is a smooth manifold. (Note that singular homology
can be computed for all topological spaces, while de Rham cohomology is only
defined for manifolds)
Definition 10.9. Let be a (smooth) singular p-simplex and a closed p-form,
then we define; 3 3
:= .
p

The latter being the integral over a submanifold with corners of a differential-
form on Rp .
Remark 10.5. This integral is called the intergral of over .
Remark 10.6. We can also extend Definition 10.9 to any (smooth) p-chain by
3 k
' 3
!k
:= ci .
i=1 c i i i=1 i

Now as with the regular integration over manifolds we want an analogous

result of Stokes Theorem for chains.
Theorem 10.2. (Stokes theorem) Let c be a smooth q-chain in M a smooth
manifold. And let be a smooth differential (q 1)-form on M . Then;
3 3
= d.
c c

Remark 10.7. Note that this is basically proves commutativity of differential

operators.
Proof. We will prove it only for simplices, as it just extends by linearity to
chains.

Let be a p-simplex. Then with Definition 10.9 and commutativity of

pullbacks and differential operators(by Proposition 6.1) we get;
3 3 3 3

d = d = d = .
q q q
-q
Now since we know that q = i=0 (1)i q Fi,q we can deduce,
3 3
= ! .
q i
q i=0 (1) q Fi,q

42
Which by the remark of Definition 10.9 and parametrization is equal to;
'q 3 'q 3 q
' 3
(1)i = (1)i
Fi,q = (1)i ( Fi,q ) ,
i=0 q Fi,q i=0 q1 i=0 q1

now once again using the definition we get

'q 3 3
i
= (1) = .
i=0 Fi,q

p
Definition 10.10. Let [] HdR (M ) and c [c] Hp (M ). The de Rham
p
homomorphism I : HdR (M ) H (M ; R) is the following map;
p

3
I[][c] = .
c

Before we check that this is in fact an isomorphism we should first see if its
even well-defined.
Lemma 10.4. I is well-defined in [] and [c].
Proof. Let = + d. Then;
3 3 3 3 3
I[][c] = = + d = + d = = I[][c],
c c c c c

the integral over d is 0 because c is in the singular p-cycles of M , thus

c = 0,
it now follows that the integral is 0 with Stokes theorem.

Next let c# = c + d. Then again by linearity over the chains of the integral
and applying Stokes theorem analogously we get again that I[][c] = I[][c# ].

In the coming section we will also need naturality of the de Rham homo-
morphism. Now in algebraic topology a natural transformation is a function
between functors, it involves a commuting diagram which will show up in the
lemma.
Lemma 10.5. (Naturality of the de Rham homomorphism) Let F : M N be
a smooth function. The following diagram commutes;
p F! p
HdR (N ) HdR (M )

I I

" "
F !
H p (N ; R) H p (M ; R)

43
Proof. The lemma comes down to proving;

I(F [])[] = I[](F []).

Now just writing out the definitions;

3 3

I(F [])[] = F = (F ) = I[][F ] = I[]F [].
p p

10.5 de Rham theorem

Before we put down the de Rham theorem first a few useful definitions and
lemmas.
Definition 10.11. A smooth manifold M is called a de Rham manifold if the
de Rham homomorphism is an isomorphism on it.
Lemma 10.6. Every open convex subset U of Rn is de Rham.

Proof. By the Poincare lemma for de Rham cohomology and singu-

q
lar cohomology we get that HdR (U ) = H q (U ; R) = 0 for n > 0 and
HdR0
(U ) = H 0 (M ; R) = R. So we need only show that I : HdR
0
(U ) H 0 (U ; R)
is an isomorphism. Take 0 = {0} (really any point would do)

But since we know that HdR0

(U ) consists of all constant functions f : M R
we can calculate 3
I[f ][] = f = (f )(0) = f.
0

thus since f R, U is de Rham.

5 Let {Ui }iI be a collection of open, disjoint, de Rham subsets

Lemma 10.7.
of M then iI Ui is de Rham.
p 5 5
Proof.
5 We know that
5 HdR ( iI Ui ) p
= iI HdR (Ui ), the same can be said for
p
H ( iI Ui ; R) = iI H (Ui ; R). Now by Lemma 10.5 the following diagram
p

commutes;
p
6 6 p

=!
HdR ( Ui ) HdR (Ui )
iI iI

I
=
" "
6 6
p p
H ( Ui ; R)
=! H (Ui ; R)
iI iI

44
 Now one final lemma before we can finally prove the de Rham theorem.
Lemma 10.8. Let U and V be open subsets of a smooth manifold M , then if
U , V , and U V are de Rham. U V is also de Rham.
Proof. Consider the following Mayer-Vietoris sequences for de Rham cohomol-
ogy and singular cohomology respectively;

.. ..
. .

" "
p
HdR (U V )
I ! H p (U V ; R)

"
"
p p =!
HdR (U ) HdR (V ) H p (U ; R) H p (V ; R)

"
"
p
HdR (U V )
= ! H p (U V ; R)

" "
p+1
HdR (U V )
I ! H p+1 (U V ; R)

" "
.. ..
. .

Now since we have a 1-2-1-2 pattern of isomorphisms we can apply the Five
p
Lemma to see, that HdR (U V )
= H p (U V ; R). Thus showing that U V is
de Rham.

45
 Now all that is left to prove is the following lemma from Bredons Topology
and Geometry;
Lemma 10.9. Let M be a smooth n-manifold. Suppose that P (U ) is a state-
ment about open subsets of M , satisfying the following three properties:

(1) P (U ) is true for U diffeomorphic to a convex open subset of Rn ;

(2) P (U ), P (V ), P (U V ) P (U V ); and
!
(3) {U } disjoint, and P (U ), all P ( U ).

Then P (M ) is true.
Proof. Since we know M has a countable basis, elements of which are are
diffeomorphic to some open subset of Rn . All we really need to show is that
P (U ) for any open subset of Rn . But since U is open in a Euclidian space it can
be written as a countable union of open sets each of which are diffeomorphic
to open balls (Rn is second countable with open balls). Now for all open balls
W we have P (W ) since these are convex, furthermore the intersection of two
open balls is convex once more. Thus by (2) the union of all these open balls is
convex, so P (U ), which implies P (M ).

p
Theorem 10.3. (de Rham) I : HdR (M ) H p (M ; R) is an isomorphism.
Proof. We need to show that all smooth manifolds are de Rham. But by
Lemma 10.6, 10.8 and 10.7 we get that being de Rham satisfies the conditions
of Lemma 10.9. And thus we can conclude that every smooth manifold is de
Rham.

46
11 Compactly supported cohomology and
Poincar
e duality
With the de Rham cohomology we have found a diffeomorphism invariant of
smooth manifolds. However the de Rham cohomology does not give any kind
of differentiation between contractible manifolds. Since by the Poincare Lemma
their (de Rham) cohomology groups are always identical. For instance it does
not differentiate between Rn and Rm for n 8= m. This is why we will introduce a
slightly altered version of the de Rham cohomology. Furthermore we will write
about a famous relation between normal and compactly supported de Rham
cohomology, namely the Poincare duality.

11.1 Compactly supported de Rham cohomology

Definition 11.1. A function f : M R is said to have compact support if
supp(f ) := {p M : f (p) 8= 0} is compact.
Definition 11.2. nc (M ) n (M ) is the set of all n-forms that have compact
support(or with compactly supported coefficient functions).
Now as was the case with the normal (M ), c (M ) also forms a chaincom-
plex with the differential operator. We denote its n-th cohomology group with
Hcn (M ). This group is called the n-th compactly supported de Rham group.
But why is this definition so different from the normal de Rham cohomology?
After all, all compactly supported closed differential forms are also closed in
the set of differential forms. The main difference is in the fact that when a
differential form is exact, its anti-derivative need have compact support. And
this is where the difference is nested.

Also note that on compact smooth manifolds all differential forms have
compact support, therefore the definitions coincide as qc (M ) = q (M ).

Now you might wonder if compactly supported de Rham cohomology is

still homotopy invariant. This next lemma will disprove this notion.
Lemma 11.1. Hc1 (R)
= R.
8
Proof. Let I : Hc1 (R) R be the map , R . This map is well defined
8 df 8 R df
because dx dx = R dx dx = f (R) f (R), and since f is compactly
supported it is 0 outside of some compact set of R (compact sets are bounded in
Hausdorff spaces) thus f (R) = f (R) = 0 for R big enough. Next, surjectivity
boils down to finding a compactly supported smooth function that integrates to
1.
8 Finally all that is left is proving injectivity. What we need to prove is that if
R
= 0 then = d. So consider = f dx the following function;
3 x
F (x) = f (t)dt.

47
Now clearly dF = f dx = , though we still need to show that F (x) has compact
8R
support. Clearly if we choose R large enough we get F (R) = f (t)dt =
8
f (t)dt = 0. But also if we choose r < 0 large enough we get F (r) =
8
r 8r

f (t)dt = 0 = 0. And thus we get that F has compact support and we
have proven that I is an isomorphism.

Remark 11.1. Compactly supported de Rham cohomology is not homotopy

invariant.
Proof. We know that R 2 {} but since the point is compact its compactly
supported de Rham groups are the same as its regular de Rham groups. These
differ from the compactly supported groups of R and as such compactly sup-
ported de Rham is not homotopy invariant.

Remember that there are no differential 2-forms on a 1-dimensional smooth

manifold. As such the compactly supported de Rham cohomology groups of
degree higher than the dimension of the manifold are 0. Next will calculate all
topcohomology groups of Euclidian spaces.
Proposition 11.1. Hcn (Rn )
= R (n 0, since R0 is compact.)
Proof. We will use the same basic proof as the previous lemma, only now
we integrate the topform over Rn we will no longer prove surjectivity and well
definedness and focus merely on injectivity. This8boils down to saying if is
a compactly supported differential n-form where Rn = 0, then = d, with
compactly supported. Now we know that this is true for n = 1, now with
induction assume it holds for n = m 1. And consider a compactly supported
m-form on Rm that integrates to 0. Since is compactly supported and
Rn is a T4 -space we have open balls around the origin B and B # such that
supp() B B B # . Furthermore, by the Poincare lemma for regular de
Rham cohomology there is a 0 such that d0 = . Now consider;
3 3 3 3
0= = = d0 = 0 .
Rm B# B# B #

Now because Rm \B 2 Sm1 and 0 is an (m 1)-form that integrates to 0, we

know that 0 is exact on Rm \B by the inductionhypothesis. And as such we
can find a such that 0 = d. Where is an (m 2)-form on Rm \B. If we
now take to be a function that is 1 on R \B and compactly supported in
n #

Rm \B. We get that = 0 d() is a smooth funtion on Rm . and furthermore

satisfies d = d0 = . Since d() = d = 0 on Rm \B # we have that it is
compactly supported.

48
 Now that we know the top compactly supported de Rham groups we would
like to know the other ones. As noted before the groups with index higher than
the dimension of the manifold are 0. We shall see that all groups with index
smaller than the dimension of the manifold will also be 0.
Proposition 11.2. Hcp (Rn ) = 0 for all 0 p < n.
Proof. First consider p = 0. As we saw before the only differential 0-forms f
in normal de Rham cohomology that give df = 0 are the constant functions,
however only the function f 0 has compact support for n > 0. Thus Hc0 (Rn ) =
0. The basic idea is to make an isomorphism from Hcp (Rn ) Hc0 (Rnp ) by
integrating out p coordinates. I will not go into details but this is the main idea.
Note that you can also use this construction to proof the last proposition.

11.2 Mayer-Vietoris sequence for compactly supported de

Rham cohomology
In section 7.3 we deduced the Mayer-Vietoris sequence for de Rham cohomology,
this proved to be a valuable tool in computing the de Rham groups, so naturally
we want an analogue for the compactly supported case. However the compact
support part has a strange counterintuitive result, namely the existence of the
special inclusion map for U / M , i# : nc (U ) nc (M ). That sends a com-
pactly supported differential form on U to the same differential that is just 0
outside of U . (This can be done in a smooth way because of the compactly
support of the form and since U is open (and M is Hausdorff)).
Lemma 11.2. The map i# commutes with the differential.
Proof. We know that the differential sends nc (U ) to d n+1
c (U ). So
what (i# d) would be d extended to 0 outside of U . Now if we first apply
i# we get; 4
0 on M \U ;
i# =
on U.
Thus; 4
d0 = 0 on M \U ;
di# =
d on U.

Remark 11.2. i# induced a map i : Hcp (U ) Hcp (M ).

Theorem 11.1. (Mayer-Vietoris for compactly support de Rham) Let U, V
M open, such that U V = M . Then there exists a map such that the
following sequence is exact.
...
! Hcp (U V ) i (j
! ) Hcp (U ) Hcp (V ) k +l
!

i (j )
Hcp (M )

! H p+1 (U V ) ! ...
c

49
where i, j, k and l are inclusion maps.
Since the proof is so much like the original Mayer-Vietoris proof, and its
mostly diagram chasing we will omit the proof.
Remark 11.3. There is also an MVS for the dual space of the compactly
supported de Rham groups Hcp (M ) ;
( )
(i ) (j ) p
(l )
... # Hcp (U V ) # Hc (U ) Hcp (V ) #
(k )

(j )
Hcp (M ) # Hcp+1 (U V ) #
( ) (i )
...

11.3 Poincar
e duality
In this section we will proof an important relation between normal and com-
pactly supported de Rham groups. The proof will be similar to the proof of the
de Rham theorem.
Definition 11.3. Consider the following map PD : p (M ) np
c (M ) ;
3
PD()() = .
M

This is a linear map because the wedge product is bilinear, also this map
commutes with the differentials d and d# (dual differential) because;
3 3 3
PD(d)() = d = d( ) (1)p d =
M M M
PD()(1p d) = d# PD()(),

since d( ) is a (n + 1) form on an n-manifold. This results into that PD

induced the eponymous map on homology level. Now a few lemmas we will need
to prove Poincare duality. And a new definition.
Definition 11.4. A smooth (oriented) manifold is called Poincare if PD is an
isomorphism.
Lemma 11.3. Every open ball U of Rn is Poincare.
Proof. We know that since U is contractible we get that HdR 0
(U ) = R and
the other de Rham groups are 0. We also know that U is diffeomorphic to
Rn thus Hcn (U ) = R, and the other compactly supported de Rham groups are
0. Thus we need only check that PD : HdR 0
(U ) (Hcn ) is an isomorphism.
Surjectivity follows easily since PD is nothing but multiplication by a constant
function. And because U is oriented we have a non-vanishing topform. Basically
the same argument
8 as with the de Rham theorem. For injectivity consider f
such that M f = 0 for all , again by the existence of the non-vanishing form
it is required that f = 0.

50
Lemma 11.4. Let U, V be open sets such that U, V and U V are Poincare.
Then U V is Poincare.
Proof. Consider the following diagram with MVS as rows, and apply the Five
Lemma:
0
= ! 0

" "
0
HdR (U V )
PD ! H 0 (U V )
c

" "
p
=!
0
HdR (U ) HdR (V ) Hc0 (U ) Hp0 (V )

"
"
0
HdR (U V )
= ! H 0 (U V )
c

" "
0 0

Remark 11.4. Note that you will need to have that PD commutes with the
connecting homomorphism.
Lemma 11.5. Let F : M N , then the following diagram commutes.
p
HdR (N )
F ! H p (N )
dR

PD PD

" "
(F )
Hcnp (N ) ! H np (M )
c

The proof is analogous to that of Lemma 10.5. In the same way we can
prove the following lemma.

51
Lemma
5 11.6. Let {U } be a collection of open disjoint Poincare sets then
U
is Poincare.
Now we can use Lemma 10.9, to once again show that;

Theorem 11.2. Let M be a smooth, orientable n-manifold. Then;

p
HdR (M )
= Hcnp (M ) .

Corollary 11.1. Let M be an orientable smooth compact n-manifold. Then;

p np
dim HdR (M ) = dim HdR (M ).

Proof. Since M is compact we have that compactly supported de Rham coho-

mology is the same as normal de Rham cohomology. Furthermore since all de
Rham cohomology groups of a compact manifold are finite dimensional we get
that the dimension of a dual space of a finite dimensional vectorspace is equal
to the original dimension.

52
12 Conclusion
When I started writing this thesis I made a goal for myself that I wanted to
understand, and be able to explain what exactly de Rham cohomology is and
what its importance is to the world of mathematics. I can now say without
doubt that I have completed the first part, I understand the ideas behind the de
Rham groups and in what way they work. As for the importance of the theory
I will go back to my introduction where I stated that de Rham groups form a
perfect example of the interaction between analysis and topology. For instance,
if you know all about the differential forms of a manifold you can say something
non-trivial about its shape (is it diffeomorphic to a sphere, etc). Analogously,
if you know about the shape of a manifold you can often conclude something
relevant with respect to the functions on this manifold. I can certainly say that
I enjoyed working on this thesis and that I truly learned a lot.

53
13 Bibliography
Introduction to Smooth Manifolds, John M. Lee
From Calculus to Cohomology, Ib H. Madsen
Differential Forms in Algebraic Topology, Raoul Bott & Loring W. Tu

Topology and Geometry, Glen E. Bredon

Introduction to differentiable manifolds(lecture notes), Rob van der Vorst

54