TOPICS ON SMOOTH

MANIFOLDS AND
DIFFERENTIAL TOPOLOGY

WRITTEN BY

BAZINGA

PUBLISHED IN THE WILD

Cover picture: Homology cycles on a torus, on a two-holed toroid, and on a Klein
bottle, created by Steelpillow.
A remark on notation:
In this book we use the so called Einstein summation convention implicitly
without mention. This convention states that when an index variable appears twice
in a single term, once upstair and once downstair, and is not otherwise defined, it
implies summation of that term over all the values of the index. For example, if
µ ∈ {1, ..., n}, then
aµ bµ := a1 b1 + a2 b2 + · · · + an bn .
P
This is the same thing as nµ=1 aµ bµ . Note that the dummy index µ does not matter
and we can rename it. That is, we can use ak bk or aν bν or aα bα to express the same
quantity.
One important remark is that an upper index in the denominator is to be taken
as a lower index. For example,
∂F µ ∂F 1 ∂F
µ
dx := 1
dx + · · · + n dxn .
∂x ∂x ∂x
Sometimes another way of expressing the same quantity above is ∂µ F dxµ .
Preface

This is a set of notes that I have written on basics of differentiable and smooth
manifolds, and some other topics that I liked a lot while learning, specifically de
Rham cohomology theory and its applications in differential topology. I was deeply
amazed by the beauty of these elegant results in differential topology, and the con-
crete geometric (in contrast to homological algebraic) approach to cohomology is
what I believe to be the most pedagogical approach to a first exposure in algebraic
topology.
The content in the majority of the text come from some of my old notes on differ-
ential manifolds that I wrote while self-studying and while taking graduate courses
on relevant topics. So technically it does not follow any textbooks or standard ref-
erences. However, the style of my writing was greatly influenced by [1], a series of
wonderful lectures in the WE-Heraeus International Winter School on Gravity and
Light by Frederic P. Schuller, and [2]. The readers are encouraged to read these
references if needed, since they certainly do a better job than what I did in these
notes.
The reader hoping to use these notes should have a solid foundation in multi-
variable calculus and linear algebra, and a certain level of mathematical maturity.
Ideally, they should also have a good background in point set topology, as I will only
review the relevant concepts very quickly in the beginning. Readers who does not
have relevant background mentioned above may find themselves having difficulties
to comprehend the text, and may want to do some extra reading on the background
material to catch up along the way.
I am deeply indebted to my professors: Xianzhe Dai, Guofang Wei, Rugang
Ye, Fedor Manin, and Daryl Cooper, for teaching me the theory of differentiable
manifolds, Riemannian geometry, and differential (algebraic) topology. I would also
like to thank my friends Merrick Hua and Benjamin Chung, for taking graduate
topology classes with me and doing a group reading on Bott and Tu’s book. Our
collective notes inspired the majority of Chapter 6.
Contents

I Basic Theory 7
0 Review of Basic Topology 8
0.1 Basic Concepts, Continuity . . . . . . . . . . . . . . . . . . . . . . . . 8
0.2 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
0.3 Product Topology, Connectedness . . . . . . . . . . . . . . . . . . . . 11

1 Manifolds 13
1.1 Topological Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Smooth Manifolds and Smooth Maps . . . . . . . . . . . . . . . . . . 15
1.3 Examples of Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . 17
1.4 Partition of Unity and Applications . . . . . . . . . . . . . . . . . . . 19

2 Tangent and Cotangent Spaces 26

2.1 Tangent Space of a Manifold . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Pushforward of a Smooth Map . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4 Cotangent Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Vector Bundles 37
3.1 Basic Definitions and Examples . . . . . . . . . . . . . . . . . . . . . 37
3.2 Tangent Bundle and Cotangent Bundle . . . . . . . . . . . . . . . . . 38
3.3 Vector, Covector, and Tensor Fields . . . . . . . . . . . . . . . . . . . 41

4 Differential Forms 43
4.1 Symmetric and Alternating Tensors, Wedge Product . . . . . . . . . 43
4.2 Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Orientation on Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Pullback and Exterior Differentiation . . . . . . . . . . . . . . . . . . 50

5 Integration on Manifolds 55
5.1 Basic Definitions and Properties . . . . . . . . . . . . . . . . . . . . . 55
5.2 Manifolds with Boundary and Boundary Orientation . . . . . . . . . 59
5.3 Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.4 Applications of Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . 64
CONTENTS 6

II Cohomology and Its Applications to Differential Topol-

ogy 67
6 Cohomology 68
6.1 Category Theory, de Rham Cohomology . . . . . . . . . . . . . . . . 68
6.2 Computations: First Installments . . . . . . . . . . . . . . . . . . . . 76
6.3 Some Homological Algebra . . . . . . . . . . . . . . . . . . . . . . . . 77
6.4 the Mayer-Vietoris Sequence . . . . . . . . . . . . . . . . . . . . . . . 80
6.4.1 Mayer-Vietoris Sequence for Compact Supports . . . . . . . . 81
6.4.2 Cohomology of S 1 . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.5 Homotopy Invariance and Applications . . . . . . . . . . . . . . . . . 84
6.6 Poincaré Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.6.1 Poincaré Lemma and Calculation of H ∗ (Rn ) . . . . . . . . . . 88
6.6.2 Poincaré Lemma for Compactly Supported Cohomology and
Calculation of Hc∗ (Rn ) . . . . . . . . . . . . . . . . . . . . . . 90
6.7 The Mayer-Vietoris Argument . . . . . . . . . . . . . . . . . . . . . . 95
6.8 The Thom Isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.8.1 Poincaré Duals and the Thom Class . . . . . . . . . . . . . . . 112

7 Differential Topology 116

7.1 Degree Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.2 The Brouwer Fixed Point Theorem . . . . . . . . . . . . . . . . . . . 119
7.3 The Borsuk-Ulam Theorem . . . . . . . . . . . . . . . . . . . . . . . 121
7.4 Intersection Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.5 Lefschetz Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . 127
7.6 Euler Characteristics and Poincaré-Hopf Theorem . . . . . . . . . . . 133
7.7 Euler Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

Index 139

Bibliography 142
 Part I

Basic Theory
0 | Review of Basic Topology

In this chapter we review the basic idea of topology and continuity defined on general
topological spaces. Here is the definition of a topology.

0.1 Basic Concepts, Continuity

Definition 0.1. Given a set M , a set O ∈ P(M ) , where P(M ) is the power set of
M , is called a topology of M if

1. ∅, M ∈ O;

2. U, V ∈ O implies U ∩ V ∈ O;
S
3. If Ui ∈ O for some i ∈ A, then i∈A Ui ∈ O.

One example. Let M = {1, 2, 3}. Let OM = {∅, {1, 2, 3}}. You can immediately
verify that this is indeed a topology on M . We have two trivial topologies: the
smallest possible topology ∅, and the biggest possible topology P(M ), which is the
set of all subsets of M .
We introduce the standard topology on Rn .

Definition 0.2. A set U ∈ OS , where U ⊂ Rn , implies that for all p ∈ U , there

exists a positive real r such that B(p, r) ∈ U .

You should verify that this topology OS on Rn satisfies the axioms above, thus
is indeed a topology. This topology is the the set of open sets in Euclidean space.

Definition 0.3. Suppose some set M is given some topology O. Together, (M, O)
is called a topological space.

Now we defined the continuity using topology. Given a map f : M → N , suppose

V is in the range of f . Recall the preimage of V , denoted by f −1 (V ), or sometimes
Pref (V ), is defined by
f −1 (V ) := {m : f (m) ∈ V }.

Definition 0.4. Let (M, OM ) and (N, ON ) be topological spaces. Then a map
f : M → N is continuous if for all V ∈ ON , we have f −1 (V ) ∈ OM . A continuous
bijection with a continuous inverse is called a homeomorphism.

Now comes the general definition of open and closed sets.

9 CHAPTER 0. REVIEW OF BASIC TOPOLOGY

Definition 0.5. Given a topological space (M, O), a set U ⊂ M is open in M if

U ∈ O. A set U is closed in M if M \ U ∈ O.
If we are in Euclidean space, then the definition of open sets and closed sets can
be equivalently given by the familiar δ-ε language: However, we wish to generalize
this statement into abstract topological spaces. For general functions, we use the
above definition of continuity rather than the δ-ε statement.
Theorem 0.6. Let (M, OM ), (N, ON ), (P, OP ) be topological spaces. Let f : M →
N and g : N → P be continuous. Then g ◦ f : M → P is also continuous.
Proof. Take some V ∈ OP . We have

(g ◦ f )−1 (V ) = {m ∈ M : g ◦ f (m) ∈ V }
= {m ∈ M : f (m) ∈ g −1 (V )}
= f −1 (g −1 (V )).

Since g is continuous, we must have g −1 (V ) ∈ ON . But then f −1 (g −1 (V )) ∈ OM by

the continuity of f . This completes the proof.
Definition 0.7. Let f : M → N be continuous mapping and M, N be topologi-
cal spaces with given topology OM , ON , respectively. Given S ⊂ M , defined the
restricted map f |S : S → N . We say f |S as f restricted to S.
Defined the subspace topology O|S to be

O|S := {U ∩ S : U ∈ OM }.

The definition above says that every element in O|S must be in the form U ∩ S,
where U ∈ OM . The good thing about this definition is shown below.
Theorem 0.8. Given topological spaces (S, O|S ) and (N, ON ), the restricted map
f |S : S → N is continuous.
Proof. Let V ∈ ON . Since f : M → N is continuous, we have Pref (V ) ∈ OM .
Consider

f −1 (V ) ∩ S = {m ∈ M : f (m) ∈ V } ∩ {m ∈ S : S ⊂ M }
= {m ∈ S : f (m) ∈ V } = (f |S )−1 (V ).

Therefore, we conclude that (f |S )−1 (V ) ∈ O|S . This completes the proof.

Theorem 0.9. A function f : Rn → Rm is continuous in the δ-ε sense iff it is
continuous in the sense that we defined above.
Proof. We prove the case where n = m = 1, the general case follows exactly the
same.
Let U ⊂ R be open. First we showSthat if f is continuous in the δ-ε sense,
then f −1 (U ) is open. By definition, U = p∈U B(p, εp ) where εp is chosen such that
B(p, εp ) ⊂ U . Since [
f −1 (U ) = f −1 (B(p, εp ))
p∈U
0.2. COMPACTNESS 10

it suffices to show that f −1 (B(p, ε)) is open for any p ∈ R and ε > 0.
Let x ∈ f −1 (B(p, ε)). Then f (x) ∈ B(p, ε), or equivalently |f (x) − p| < ε. Set
η = ε − |f (x) − p| > 0.
Since f is continous at x, we may choose δ > 0 such that |x0 − x| < δ implies
|f (x0 ) − f (x)| < η. In other words f (B(x, δ)) ⊂ B(f (x), η). Then by triangle
inequality:
|f (x0 ) − p| ≤ |f (x0 ) − f (x)| + |f (x) − p| < η + |f (x) − p| < ε.
Thus f (x0 ) ∈ B(p, ε) for all x0 ∈ B(x, δ). Hence we have shown that there exists
δ > 0 such that B(x, δ) ⊂ f −1 (B(p, ε)). This shows that f −1 (B(p, ε)) is open.
Conversely, we show that a continuous function f : R → R between topological
spaces is continuous in the δ-ε sense. Fix ε > 0, p ∈ R. Since f −1 (B(f (p), ε)) is an
open set (by assumption) that contains p, we can choose δ > 0 such that B(p, δ) ⊂
f −1 (B(f (p), ε)). Thus f (B(p, δ)) ⊂ B(f (p), ε). This is exactly the definition of δ-ε
continuity.

0.2 Compactness
Definition 0.10. A topological
S space X is called compact if for any open cover
{Uα }α∈A of X (i.e., X = α∈A Uα ), there exists a finite subcover. That is, there
exists a finite subcollection {U1 , ..., Un } ⊂ {Uα }α∈A such that X = U1 ∪ · · · ∪ Un .
In Rn , we know exactly what compact sets look like:
Theorem 0.11 (Heine-Borel). A set K ⊂ Rn is compact iff it is closed and bounded,
i.e., K ⊂ B(0, r) for some sufficiently large r > 0.
Example 0.12. Here are some examples of compact spaces:
• By Heine-Borel theorem, any closed interval [a, b] ⊂ R is compact. So is finite
union of them.
• Let X = {p1 , ..., pn } be a finite set. Let T be the collection of all subsets of
X. Then with respect to this topology, X is compact.
• Let X be a compact space and f : X → Y a continuous map between topo-
logical spaces. Then f (X) is a compact subspace of Y (subspace topology).

Proof. Let {Uα }α be an open cover of f (X). Then

[ [
X = f −1 (f (X)) = f −1 ( Uα ) = f −1 (Uα ).
α α

Since f is continous, f −1 (Uα ) is open for each α. Since X is compact, there

exists a finite collection such that
X = f −1 (U1 ) ∪ · · · ∪ f −1 (Un ).
Therefore f (X) = U1 ∪ · · · ∪ Un .
11 CHAPTER 0. REVIEW OF BASIC TOPOLOGY

• A curve in a topological space X is the image of a continuous map γ : [0, 1] →

X. By the last example, a curve is compact in X.
• A closed subspace K of a compact space X is compact.

Proof. Let {Uα }α be an open cover of K. Then {Uα }α ∪ {X \ K} is an open

cover of X. Since X is compact, we have can choose a finite subcollection
X = (X \ K) ∪ U1 ∪ · · · ∪ Un .
Then U1 ∪ · · · ∪ Un = K. This is a finite subcollection of {Uα }α that covers
K.

• The most important example is still compact subsets in Rn , which is equivalent

to closed and bounded subsets in Rn by the HeineBorel theorem. This is
probably one of the only few examples you need to remember as manifolds are
locally Euclidean.
As compactness is some sort of boundedness on a space, functions on that space
are also bounded as a result:
Theorem 0.13. Let f : X → R be a continuous real-valued function. Then f is
bounded. Moreover, f obtains its maximal and minimal values.
Proof. Apply Heinel-Borel: Since f (X) is compact in R, clearly f is bounded. Since
f (X) is bounded, inf x∈X f (x) and supx∈X f (x) exists. Since f (X) ⊂ R is closed,
the inf and sup are contained in f (X). Suppose f (x0 ) = inf x∈X f (X) and f (x1 ) =
supx∈X f (x). Then f obtains minimum at x0 and maximum at x1 .

0.3 Product Topology, Connectedness

Definition 0.14. Let (X, O) be a topological space. A base (or basis) for the
topology O is a subcollection B ⊂ O of open sets in X, such that every open sets U
in X is a union of open sets in B.
Definition 0.15. Let (X, OX ), (Y, OY ) be two topological spaces. Let X × Y be
the set theoretic product of X and Y . We define a topology OX×Y on X × Y as
follows: The basis B of OX×Y consists of all open sets of the form
B = {U × V : U ∈ OX , V ∈ OY }.
Example 0.16. Rn is the set theoretic product of n copies of R. Convince yourself
that the basis topology of Rn coincides with the standard topology on Rn . It is not
hard to show that any open sets U in Rn (with respect to the standard topology)
can be written as a union of open boxes, and every open boxes in Rn can be written
as a union of open balls.
Definition 0.17. A topological space X is said to be path-connected if for every
two points p, q, there exists a continuous curve γ : [0, 1] → X, such that γ(0) =
p, γ(1) = q.
0.3. PRODUCT TOPOLOGY, CONNECTEDNESS 12

Definition 0.18. A topological space X is said to be disconnected, if there exists

nonempty disjoint open sets U, V such that U ∪ V = X. Otherwise, X is said to be
connected. Note that the definition of X being connected is equivalent to saying
that the only closed and open subset in X is either ∅ or X.

Lemma 0.19. Let X be a connected topological space, and f : X → Y is a contin-

uous map between topological spaces that is locally constant, i.e., for every p ∈ X,
there exists an open set Up ⊂ X such that f is constant on Up . Then f is globally
constant.

Proof. Let p ∈ X be any point and let s = f (p) ∈ Y . Then consider the set

S := {q ∈ X : f (q) = s}.

Since S = f −1 (q) and f is continuous, we see that S is closed. On the other hand,
for every q ∈ S, by locally constant hypothesis, there exists an open set Uq ⊂ X
such S that f (q 0 ) = s for all q 0 ∈ Uq . That is, f is constant on Uq with value s. Thus
S = q∈S Uq . Hence S is open. Since S is closed and open at the same time, and by
assumption p ∈ S, so S 6= ∅. Then since X is connected, we conclude that S = X,
i.e., f is globally constant.

Theorem 0.20. The interval [0, 1] with subspace topology of R, is connected. Sim-
ilarly, any interval [a, b] is connected.

Proof. Suppose [0, 1] is disconnected and U, V be two disjoint nonempty open subsets
such that U ∪ V = [0, 1]. Define a function f : [0, 1] → {±1} via
(
+1 x ∈ U
f (x) =
−1 x ∈ V.

Then f is well-defined, since U, V are disjoint, and is nonconstant, since U, V are

both nonempty. Note that f is continuous since preimages of open sets in {±1} can
only be one of ∅, U, V, [0, 1], which are both open. But f is clearly locally constant
by construction. This contradicts Lemma 0.19.

Theorem 0.21. A path connected space is connected. (The converse fails, but in
the case of smooth manifolds, it is true)

Proof. Suppose X is path connected but disconnected. Then we can find nonempty
open sets U, V such that U ∪ V = X. Since both are nonempty, take p ∈ U, q ∈ V .
Since X is path connected, we may find a path γ : [0, 1] → X with γ(0) = p, γ(1) = q.
Since U, V are disjoint, γ −1 (U ), γ −1 (V ) are disjoint open sets that covers [0, 1]. But
[0, 1] is connected, contradiction.
1 | Manifolds

1.1 Topological Manifolds

Definition 1.1. A topological space (M, O) is called a topological manifold of
dimension n, if it is:

• Hausdorff: For every pairs of distinct points p, q ∈ M , there are disjoint

open sets U, V ⊂ M such that p ∈ U and q ∈ V .

• Second-countable: There exists a countable basis for the topology of M .

• Locally Euclidean: For every p ∈ M , there exists an open set U containing

p, and a homeomorphism φ : U → φ(U ) ⊂ Rn from U onto an open subset of
Rn .

Often we simply write M n , to denote a manifold M of dimension n.

Example 1.2. An open disk S 1 ⊂ R2 is a 1-dimensional topological manifold. For

each p ∈ U ⊂ S 1 , we can project it smoothly onto R1 by letting φ be the projection
map φ : R2 → R.
Similarly, a sphere S 2 ⊂ R3 is a 2-dimensional topological manifold. For each
p ∈ S 2 , rotate the hemisphere it is on to the top and make it the upper hemisphere.
Let φ : R3 → R2 be the projection map onto R2 plane, provided that U is contained
completely in the hemisphere. Or, just define the open set containing p be the
hemisphere it is on without the equator. Then this is clearly an open set containing
p. Its projection onto the R2 plane is just a disk without boundary, which is also an
open set.
This generalizes to S n ⊂ Rn+1 .

Definition 1.3. An open subset U ⊂ M and a homeomorphism φ : U → φ(U ) ⊂

Rn , together, (U, φ) is called a chart of (M, O).

S 1.4. A set A = {(Uα , φα ) : α ∈ A} of charts is called an atlas of (M, O)

Definition
if M = α∈A Uα . That is, if the union of elements in A covers M . The maps φα are
called chart maps.

Consider two overlapping charts (Uα , φα ), (Uβ , φβ ) on a manifold M . By over-

lapping charts we mean Uα ∩ Uβ 6= ∅. Then one can define a map between subsets
of Rn
τα,β ≡ φβ ◦ φ−1
α : φα (Uα ∩ Uβ ) → φβ (Uα ∩ Uβ ).
1.1. TOPOLOGICAL MANIFOLDS 14

This is called a chart transition map, for the obvious reason that it serves as a
transition map that transform one set of coordinates of points in Uα ∩ Uβ to another
set. Note that τα,β is continuous. See Figure 1.1 for illustration.

Figure 1.1: Chart transition maps τα,β , τβ,α .

Often, there exists a curve in real world M , defined by γ ⊂ U ⊂ M . As an

example, for a physicist, γ can be a particle trajectory changing with respect to
time, i.e. γ : R → U . But since there exists a map φ : U → Rn , the composition
φ ◦ γ : R → Rn is the trajectory of particle in Euclidean space that we know very
well.
But there is one potential danger: the chart φ is arbitrary. This depends on
what coordinate system a physicist chooses to use (Cartesian, polar, etc). But γ
is independent of the chart we choose. Since physics should not depend on the
choice of coordinate, we want to make sure that the defined property in one chart,
φ, does not change when we switch to another chart, say ψ. Specifically, we have a
commutative diagram
ψ(U )
ψ◦γ
ψ
γ
R U
φ
φ◦γ

φ(U )
For example, say if we have φ ◦ γ be a continuous map, then we better hope ψ ◦ γ
to be also continuous. Now,
ψ ◦ γ = ψ ◦ (φ−1 ◦ φ) ◦ γ = (ψ ◦ φ−1 ) ◦ (φ ◦ γ).
Since φ ◦ γ is continuous, and ψ ◦ φ−1 is continuous (since both ψ and φ−1 are
continuous), we have ψ ◦ γ be a continuous map.
However, if φ ◦ γ is differentiable, it need not be true that ψ ◦ γ is also differ-
entiable. This is due to the fact that ψ ◦ φ−1 only needs to be continuous. The
composition of a continuous map and a differentiable map is not necessarily differ-
entiable. Thus we need stronger condition for differentiability on manifolds, which
we will come back in the next section.
15 CHAPTER 1. MANIFOLDS

1.2 Smooth Manifolds and Smooth Maps

In the previous section, we tried to appeal to intuition that physicists use by viewing
a manifold as the world we live in, and charts as coordinate systems in our maps. We
have seen that for a continuous curve γ : R → M , what physicists have in their map
φ ◦ γ : R → Rn is also continuous. But we cannot ensure that the differentiability
of the curve that physicists describe is independent of the chart we choose without
adding further assumptions. In this section we try to solve this issue.
Let A∞ be the maximum atlas, the set of all possible atlas we can find, where
each atlas covers our manifold M . Let our desired property be ♥. Then define two
maps to be ♥-compatible as follows:

Definition 1.5. Let (U, φ), (V, ψ) ∈ A ⊂ A∞ be two charts of (M, O). We say
these two charts are ♥-compatible if either U ∩ V = ∅, or the chart translation
maps ψ ◦ φ−1 and (ψ ◦ φ−1 )−1 = φ ◦ ψ −1 have property ♥.

For example, if the two chart translation maps are at least once differentiable
(C ), then we say the two charts (U, φ), (V, ψ) are C 1 -compatible.
1

Definition 1.6. An atlas A♥ is ♥-compatible if any two charts (U, φ), (V, ψ) ∈ A♥
are ♥-compatible.

Definition 1.7. We say that A♥ is a maximal ♥ atlas, if it is not properly

contained in any larger ♥-compatible atlas. In other words, A♥ is maximal iff any
chart that is ♥-compatible with every chart in A♥ is already in A♥ .

Definition 1.8. A ♥-manifold is a triple (M, O, A♥ ), where A♥ is a maximal

atlas. The atlas A♥ is then called a ♥ structure.

Now, we can solve the problem mentioned above. Let ♥ be the property that
the maps are infinitely differentiable, or C ∞ . Then we restrict our choice of charts
by only choosing charts in some AC ∞ . Then the chart transition map ψ ◦ φ−1 is
differentiable. Thus ψ ◦ γ = (ψ ◦ φ−1 ) ◦ (φ ◦ γ) is differentiable given that φ ◦ γ is
differentiable.
For the rest of this notes, we will always focus on smooth manifolds (M, O, AC ∞ ).
Smooth manifolds are sometimes also called C ∞ -manifolds. Usually, we will just
write M , assuming the information O, AC ∞ is clear. Usually we just refer to smooth
manifolds as manifolds to be concise.
It is generally not easy to construct a maximal smooth atlas on a topological
manifold. However, the next theorem saves us and suggests that constructing a
smoothly compatible atlas is sufficient to determine a smooth structure:

Theorem 1.9. Every smooth atlas A is for a topological manifold M is contained

in a unique maximal smooth atlas A∗ .

Proof. Let A be a smooth atlas for M , and let A∗ be the set of all charts that are
smoothly compatible with every chart in A. We need to show:
1.2. SMOOTH MANIFOLDS AND SMOOTH MAPS 16

• A∗ is a smooth atlas: Let (U, φ), (V, ψ) ∈ A∗ be intersecting charts, we would

like to show ψ◦φ−1 : φ(U ∩V ) → ψ(U ∩V ) is smooth. Let x = φ(p) ∈ φ(U ∩V )
be any point. Because open sets in A covers M , there exists (W, θ) ∈ A with
p ∈ W . Since charts in A∗ are smoothly compatible with every chart in
A, we know θ ◦ φ−1 and ψ ◦ θ−1 are smooth where they are defined. Since
p ∈ U ∩ V ∩ W , it follows that
ψ ◦ φ−1 = (ψ ◦ θ−1 ) ◦ (θ ◦ φ−1 )
is smooth in a neighborhood of x. This proves the claim.
• A∗ is maximal: Any chart that is smoothly compatible with every chart in
A∗ must be in particular smoothly compatible with every chart in A, so it is
already in A∗ .
• A∗ is the unique maximal smooth atlas containing A: If B is another maximal
smooth atlas containing A, each of its charts is smoothly compatible with each
chart in A, so B ⊂ A∗ . By maximality of B, we have B = A∗ .
This completes the proof.
Now we define a smooth map between (smooth) manifolds.
Definition 1.10. A map f : M m → N n between manifolds is said to be smooth,
if for every p ∈ M , there exists a chart (U, φ) of M containing p, a chart (V, ψ) of
N containing f (U ), such that the map
ψ ◦ f ◦ φ−1 : φ(U ) → ψ(V )
is a smooth map (i.e., all partial derivative exists) from an open subset φ(U ) of Rm
to an open subset ψ(V ) of Rn .
Definition 1.11. A smooth map f : M → N with a smooth inverse f −1 : N → M
is called a diffeomorphism.
Note that one can view a chart φ : U → φ(U ) ⊂ Rn as a map between manifolds:
Here U is given the subspace topology and charts on U are given by charts on M
restricted to open subsets of U ; we may view Rn as a manifold with chart just being
identity; then φ(U ) is also a manifold in the same way that U is a manifold. Thus,
φ : U → φ(U ) is a diffeomorphism, since the induced maps are just identity maps
between open subsets of Euclidean spaces, which is clearly smooth.
Finally, we remark that the notion of smoothness of f : M → N is independent
of choices of parameterizations φ, ψ in the chart (U, φ), (V, ψ) if the open sets U, V
are the same. Suppose (U, φ̃) and (V, ψ̃) are some other choice of charts, then from
the commutative diagram below

ψ̃◦f ◦φ̃−1
φ̃(U ) ψ̃(V )
φ̃ ψ̃
f
φ̃◦φ−1 U V ψ̃◦ψ −1

φ ψ

φ(U ) ψ(V )
ψ◦f ◦φ−1
17 CHAPTER 1. MANIFOLDS

we see that
ψ̃ ◦ f ◦ φ̃−1 = (ψ̃ ◦ ψ −1 ) ◦ (ψ ◦ f ◦ φ−1 ) ◦ (φ̃ ◦ φ−1 )−1
is a composition of smooth maps in Euclidean spaces, hence is also smooth.

1.3 Examples of Smooth Manifolds

In this section, we study some examples of smooth manifolds. Our strategy to show
something is a smooth manifold can be summarized as follows: First, we start with
a topological space M and check that it is a topological manifold (by constructing
a chart on M ); Then, we show that the charts are smoothly compatible; Finally, we
use Theorem 1.9 and show that such a smooth atlas defines a smooth structure on
M , showing that M is indeed a smooth manifold.
Example 1.12 (0-dimensional Manifolds). A zero dimensional topological manifold
M is just a discrete sets of points. For each point p ∈ M , the only neighborhood of
p homeomorphic to an open set of R0 is {p}. There is exactly one coordinate map
φ : {p} → R0 . Thus the set of all charts on M is trivially smoothly compatible.
Hence M is a smooth manifold.
Example 1.13 (Rn ). Rn is an n-dimensional smooth manifold with the smooth
atlas consisting of a single chart (Rn , idRn ). This is called the standard smooth
structure on Rn .
Example 1.14 (S n ). For each n ≥ 0, the n-sphere defined by
S n := {x ∈ Rn+1 : kxk = 1}
is an n-dimensional smooth manifold: We construct an atlas on S n as follows:
AS n = {(Ui± , φ±
i ) : i = 1, ..., n + 1}

where
Ui+ = {(x1 , ..., xn+1 ) ∈ S n : xi > 0} Ui− = {(x1 , ..., xn+1 ) ∈ S n : xi < 0}
and the homeomorphisms φ± ±
i : Ui → B(0, 1) ⊂ R is given by
n

φ± 1
i (x , ..., x
n+1
) = (x1 , ..., x̂i , ..., xn+1 ).
It is easy to check that the defined maps are indeed homeomorphisms, with inverse
p
(φ± −1 1
i ) (a , ..., a ) = (a , ..., ±
n 1
1 − |a|2 , ai , ..., an ).
Hence we have a chart on S n , making it a topological manifold. Now to check
that AS n defines a smooth structure, we need to check that the transition maps are
smooth. In the case where i < j, we have
p
φ± ± −1 1
i ◦ (φj ) (u , ..., u ) = (u , ..., û , ..., ±
n 1 i 1 − |u|2 , ..., u)
which is clearly smooth on B(0, 1). A similar formula holds for i > j. When i = j,
the transition map is just identity on B(0, 1). Hence charts in AS n are smoothly
compatible, hence defining a smooth structure on S n . Thus S n is a smooth manifold
of dimension n.
1.3. EXAMPLES OF SMOOTH MANIFOLDS 18

Example 1.15 (Vector Spaces). Let V n be an n-dimensional real vector space. We

claim that V is an n-dimensional smooth manifold. Let e1 , ..., en be a basis of V .
We first show that

Lemma 1.16. Any two norms on a finite-dimensional real vector space is equivalent,
so in particular they all give the same topology to V .
P P
Proof. Let k · k be the norm on V given by k i ai ei k = ( i |ai |2 )1/2 . Let F :
V → R≥0 be any norm. It suffices to prove that F is equivalent to k · k. Define
µ := max1≤i≤n F (ei ) < ∞. Then by Cauchy-Schwarz inequality one has
X X X √
F (x) = F ( ai e i ) ≤ |ai |F (ei ) ≤ µ |ai | ≤ µ nkxk. (1.1)
i i i

Since by the reverse triangle inequality one has F (x−y) ≥ F (x)−F (y) and similarly
F (x − y) = F (y − x) ≥ F (y) − F (x), one has
√
|F (x) − F (y)| ≤ |F (x − y)| ≤ µ nkx − yk.

Hence F is continuous on (V, k · k). Then F attains a minimum on the compact set

K = {x ∈ V : kxk = 1}

by Theorem 0.13. Say the minimum is at p ∈ K. Note that since p 6= 0 ∈ V , one

has F (p) > 0. Thus

F (x) = kxkF (x/kxk) ≥ F (p) · kxk. (1.2)

Combining (1.1) and (1.2), we conclude that F and k · k are equivalent.

Now we consider V to be a topological space with topology given by any norm.

Then we get a homeomorphism φ : Rn → V by

φ(x1 , ..., xn ) = x1 e1 + · · · + xn en .

Hence we see that (V, φ−1 ) is a chart. That is, the coordinates on V are given by
components of a vector with respect to the basis eα ’s. If e01 , ..., e0n is another basis
for V and φ0 : Rn → V is another chart that is defined by the same recipe but with
respect to the basis e01 , ..., e0n , then there is an invertible matrix T = (Tβα ) such that
X
eβ = Tβα e0α .
β

Thus we see that the chart transition map φ0 ◦ φ is given by the linear map Rn → Rn
which is exactly (Tβα ). Since this is an invertible linear map, it is a diffeomorphism
from Rn to itself. Thus φ, φ0 are smoothly compatible. The collection of all such
charts thus defines a smooth structure, called the standard smooth structure
on V .
19 CHAPTER 1. MANIFOLDS

Example 1.17 (Open subsets of a manifold). Let U be an open subset of a smooth

manifold M . Then a smooth atlas is defined by

AU = {(V, φ) ∈ AM : V ⊂ U }.

This clearly is a smoothly compatible atlas and gives U a smooth structure. Thus,
we call any open subset of a smnooth manifold M an open submanifold of M .

Example 1.18 (Matrix groups). Let M (n, R) denote the space of n × n matrices
with real coefficients. This is a real vector space of dimension n2 , hence a smooth
manifold of the same dimension by our example above. Let GL(n, R) denote the set
of invertible n × n matrices with real entries. Then GL(n, R) is an open subset of
M (n, R) because GL(n, R) = det−1 (R∗ ) where det : M (n, R) → R is the determi-
nant function (which is continuous because it only involves polynomial operations).
Hence GL(n, R) is a smooth manifold of dimension n2 as well.

Example 1.19 (Product Manifolds). Let M, N be two smooth manifolds. Then

M × N is a topological space with the product topology. We can define a smoothly
compatible atlas on M × N via

AM ×N = {(U × V, φ × ψ) : (U, φ) ∈ AM , (V, ψ) ∈ AN }.

Hence M × N is again a smooth manifold of dimension dim M + dim N .

1.4 Partition of Unity and Applications

Partition of unity is a tool for us to patch smooth functions defined locally to a global
smooth function. This is probably one of the most widely used tools in the theory
of smooth manifolds. Despite its importance, proof of the existence of partition of
unity is very hard and technical. Readers who do not want to get into such technical
intricacies can just learn the definition of a partition of unity, believe in its existence,
and move on.

Definition 1.20. Let M be a smooth manifold, and f : M → R a smooth function.

The support of f , denoted by Supp f , is the closure of the set {p ∈ M : f (p) 6= 0}.

Definition 1.21. Let M be a smooth manifold, and {Uα } an open cover of M . A

partition of unity subordinate to the cover {Uα } is a collection of smooth
functions {ρα } defined on the whole manifold M so that

1. 0 ≤ ρα ≤ 1 for all α;

2. Supp(ρα ) ⊂ Uα for all α;

3. Each p ∈ M has a neighborhood which intersects only finitely many Supp(ρα )’s;
P
4. α ρα (p) = 1 for all p ∈ M . By the previous condition, at every point this is
only a finite sum, hence is well-defined.
1.4. PARTITION OF UNITY AND APPLICATIONS 20

Our goal is to prove that for any open cover U of M , there exists a partition of
unity subordinate to U. However, the proof requires several very technical lemmas
in advance. Since the existence of partition of unity subordinate to any open cover
of a manifold is such an important result, we will present the full proof for future
references. However, this is a good point for impatient readers to stop for now, and
come back to it later when needed. We now begin the technical preparation required
for the proof.
Construction 1.22. Consider the function f1 : R → R given by
(
e−1/x x > 0
f1 (x) =
0 x ≤ 0.
It is a standard exercise in calculus to check that f1 is smooth (check all derivative
exists by L’Hospital rule). One sees that 0 ≤ f1 (x) ≤ 1 when x > 0 and f1 (x) = 0
when x ≤ 0. See Figure 1.2a.
Next, we consider f2 : R → R defined by
f1 (x)
f2 (x) = .
f1 (x) + f1 (1 − x)
Then one sees that f2 is a nonnegative smooth function, such that f2 (x) = 0 when
x ≤ 0, 0 < f2 (x) < 1 when 0 < x < 1, and f2 (x) = 1 when x ≥ 1. See Figure 1.2b.
Finally, consider the function f3 : Rn → R defined by
f3 (x) = f2 (2 − |x|).
Then one sees that f3 is a bump nonnegative function that is 1 on B(0, 1), and
decays smoothly until it vanishes outside of B(0, 2). See Figure 1.2c.

Lemma 1.23. Let M m be a smooth manifold, A ⊂ M a compact subset, and U ⊂ M

an open subset that contains A. Then there is a bump function ϕ ∈ C ∞ (M ) such
that 0 ≤ ϕ ≤ 1, ϕ|A ≡ 1, and Supp(ϕ) ⊂ U .
Proof. For each q ∈ A, one can find1 a chart (Uq , φq ) near q such that Uq ⊂ U and
B(0, 3) ⊂ φq (Uq ). Let Uq0 = φ−1
q (B(0, 1)), and let
(
f3 (φq (p)) p ∈ Uq
fq (p) =
0 p∈/ Uq .
Here, f3 is defined in Construction 1.22. Then clearly fq ∈ C ∞ (M ), Supp(fq ) ⊂ Uq ,
and f |Uq′ ≡ 1. Now the family {Uq0 }q∈A is an open coverPof A. Since A is compact,
we can choose a finite subcover Uq01 , ..., Uq0n . Let ψ := ni=1 fqi . Then one can see
that ψ ∈ C ∞ (M ) such that ψ ≥ 1 on A and Supp(ψ) ⊂ U . It follows that
ϕ(p) := f2 (ψ(p))
satisfies all the conditions we want, where f2 is defined in Construction 1.22.
1
We can find such a chart because we may take any chart (U, φ) containing q, and a smooth
dialation λ : Rm → Rm that scales φ(U ) so that it contains B(0, 3). Then denote Uq = U, φq = λ◦φ.
Clearly (Uq , φq ) is smoothly compatible with any chart in the atlas of M , hence already contained
in the atlas because a smooth structure is maximal by definition.
21 CHAPTER 1. MANIFOLDS

(a) Plot of f1 . (b) Plot of f2 .

(c) Plot of f3

Figure 1.2: The plot of the three functions defined in Construction 1.22.

Lemma 1.24. For any topological manifold M , there exists a countable collection
of open sets {Xi }, which is called an exhaustion of M , so that
1. For each j, the closure X j is compact;

2. For each j, we have X j ⊂ Xj+1 ;

S
3. M = j Xj .
Proof. Since M is second countable, there is a countable basis B of the topology of
M . Out of this countable collection of open sets, we pick those that have compact
closures, and denote them by Y = {Y1 , Y2 , ...}.
We claim that Y is in fact an open cover of M . To see this, let p ∈ M be an
arbitrary point. We can take a chart (U, φ) such that p ∈ U , and choose ε > 0 small
enough, and denote B := B(φ(p), ε), such that

φ(p) ∈ B ⊂ φ(U )

and an element S in the basis B such that p ∈ S ⊂ φ−1 (B). Then we see that

S ⊂ φ−1 (B) = φ−1 (B)

is a closed subset of a compact set, hence is compact. Therefore S ∈ Y . Hence Y is

indeed an open cover of M .
Now we let X1 = Y1 . Since Y is an open cover of X 1 which is compact, there
exists finitely many open sets Yi1 , ..., Yik such that

X 1 ⊂ Yi 1 ∪ · ∪ Yi k .

Now define
X 2 = Y 2 ∪ Yi 1 ∪ · ∪ Yi k .
1.4. PARTITION OF UNITY AND APPLICATIONS 22

Clearly, X 2 is compact and X 1 ⊂ X2 . By repeating this procedure we get a sequence

of open
Ssets X1 , X2 , ... satisfying the
S requirements in the statement. Note that since
Xk ⊃ j=1 Yj , we have that M = k Xk .
k

Lemma 1.25. For any open covering U = {Uα } of a topological manifold M , one
can find two countable family of open covers V = {Vj } and W = {Wj } such that
(i) For each j, V j is compact and V j ⊂ Wj ;

(ii) W is a refinement of U : For each j, there is an α = α(j) so that Wj ⊂ Uα ;

(iii) W is locally finite: Any p ∈ M has a neighborhood W such that W ∩ Wj 6= ∅

for only finitely many Wj ’s.
Proof. The proof is in fact very geometric and ingenious: Let {Xj } be a countable
exhaustion of M as in Lemma 1.24. For each p ∈ M , there is an j and an α(p) so
that p ∈ X j+1 − Xj and p ∈ Uα(p) . Since M is locally Euclidean, one can choose
open neighborhoods Vp , Wp of p such that V p is compact and (See Figure 1.3):

V p ⊂ Wp ⊂ Uα(p) ∩ (Xj+2 − X j−1 ).

Uα(p)
M
Wp

Vp p

Xj+2 Xj+1 Xj Xj−1

Figure 1.3: Since Uα(p) ∩ (Xj+2 − X j−1 ) is the intersection of two ooen sets contain-
ing p, it is open. Intersecting this open set with a chart near p and map it down to
Rm under the homeomorphism φ, we get an open set U 0 ⊂ Rm under φ containing
φ(p). Then choose V 0 , W 0 ⊂ U 0 such that φ(p) ∈ V 0 ⊂ W 0 . Then set Vp = φ−1 (V 0 )
and Wp = φ−1 (W 0 ).

Now for each j, since the strip X j+1 −Xj is compact, one can choose finitely many
points p1,j , ..., pkj ,j so that Vp1,j , ..., Vpkj ,j is an open cover of X j+1 − Xj . Denote all
these Vpk,j ’s by V1 , V2 , V3 , ..., and the corresponding Wpk,j ’s by W1 , W2 , W3 , .... Then
V = {Vj } and W = {Wj } clearly satisfies (i) and (ii) as in the statement of the
lemma. For (iii), the local finiteness property of W follows from the fact that there
are only finitely many Wk ’s intersecting Xj+1 − X j−1 .
23 CHAPTER 1. MANIFOLDS

Lemma 1.26. Closure of unions of locally finite sets in a topological space is the
union of closure of these sets. In other let {Si : i ∈ I} be a collection of
S words, S
locally finite sets on M , then we have i∈I Si = i∈I S i .
S S S S
Proof. Since Si ⊂ i∈I Si , we have S i ⊂ i∈I Si , hence i∈I S i ⊂ i∈I Si . So the
nontrivial part is the inclusion in the other direction.
S To prove the other inclusion, by the definition of closure it suffices to prove S that
i∈I S i is closed. To this end, we show its complement is open: Take x ∈ / i∈I S i .
Since we have a locally finite collection, we may find an open neighborhood U of x
that only intersects finitely many Si ’s. Say U intersects Si1 , ..., Sin . Now consider
the new open set
U 0 := U \ (S i1 ∪ · · · ∪ S in ).
Note that U 0 is still an open set containing x, and it does not intersect any of the Si ’s.
We claim that this implies U 0 does not intersect any of the Si ’s either: Indeed, since
the complement (U 0 )c of U 0 is a closed set containing all the Si ’s, by Sminimality of
the closure, this means it also contains all the S i ’s. Hence x ∈ U 0 ⊂ ( i∈I S i )c .

Now we are finally ready to prove the big theorem:

Theorem 1.27. Let M be a smooth manifold, and {Uα } an open cover of M . Then
there exists a partition of unity {ρα } subordinate to {Uα }.

Proof. We choose {Vj }, {Wj } as in Lemma 1.25. Since V j is compact and Vj ⊂ Wj ,

by Lemma 1.23 we can find nonnegative functions ϕj ∈ C ∞ (M ) such that

0 ≤ ϕj ≤ 1 ϕj | V j ≡ 1 Supp(ϕj ) ⊂ Wj .
P
Since W is a locally finite covering, the function ϕ = j ϕj is a well-defined smooth
function on M . Since each ϕj is nonnegative, and V is a covering of M , ϕ is strictly
positive on M . PIt follows that the functions ψj = ϕj /ϕ are smooth and satisfy
0 ≤ ψj ≤ 1 and j ψj = 1.
Next, we need to re-index the family {ψj } to get the required partition of unity
subordinate to {Uα }. For each j, we fix an index α(j) so that Wj ⊂ Uα(j) , and define
X
ρα = ψj .
{j:α(j)=α}

Note that the right hand side is a finite sum near each point, so it does define a
smooth function. Moreover, By Lemma 1.26, we have
[ [ [
Supp(ρα ) = Supp(ψj ) = Supp(ψj ) = Supp(ψj ) ⊂ Uα .
{j:α(j)=α} {j:α(j)=α} {j:α(j)=α}

This concludes the construction.

Partition of unity has countless applications in the theory of smooth manifolds.

Here is one example: we can approximate continuous functions on a manifold by
smooth functions, and the proof uses Theorem 1.27. We first start with a definition:
1.4. PARTITION OF UNITY AND APPLICATIONS 24

Definition 1.28. Let M be a smooth manifold, g : M → R a function on M ,

A ⊂ M . We say that g is smooth on A if there exists open sets U ⊃ A and smooth
function g0 on U , such that g0 = g on A.
Theorem 1.29 (Whitney Approximation Theorem). Let M be a smooth manifold,
let A be a closed subset of M . For any continuous function g : M → R that is
smooth on A, and for any continuous δ : M → R>0 that is everywhere positive,
there exists a smooth function f : M → R such that
• For any p ∈ A, we have f (p) = g(p);

• For any p ∈ M , we have |f (p) − g(p)| < δ(p).

In particular, take A = ∅ and δ(p) = ε everywhere for a positive constant ε > 0,
we see that any continuous function g : M → R can be approximated by a smooth
function f : M → R, such that |f (p) − g(p)| < ε.
Proof. By definition, there exists open set U ⊃ A and smooth function g0 on U such
that g0 = g on A. Let

U0 = {p ∈ U : |g0 (p) − g(p)| < δ(p)}.

Then U0 is open in M and U0 ⊃ A. Next we construct a collection of open cover of

M − A. For any q ∈ M − A, define

Uq = {p ∈ M − A : |g(p) − g(q)| < δ(p)}.

Then {Uq : q ∈ M − A} is an open cover of M − A. Since S M is second countable,

we can find countable Uqi , where i = 1, 2, ..., such that i Uqi = M − A.
Now let {ρ0 , ρi } be a partition of unity subordinate to the open cover {U0 , Uqi }
of M . We define X
f (p) = ρ0 (p)g0 (p) + ρi (p)g(qi ).
i

Since the sum is locally finite, f ∈ C ∞ (M ). Moreover, by definition we have that

on A, f = g0 = g. Moreover, for any p ∈ M , we have
X X
|f (p) − g(p)| = ρ0 (p)g0 (p) + ρi (p)g(qi ) − ρi (p)g(p)
i i
X
≤ ρ0 (p)|g0 (p) − g(p)| + ρi (p)|g(qi ) − g(p)|
i
X
< ρ0 (p)δ(p) + ρi (p)δ(p)
i
< δ(p),

where the last inequality is obtained by the following argument: If p ∈ U0 , then by

definition ρ0 (p)|g0 (p) − g(p)| < ρ0 (p)δ(p), and if p ∈
/ U0 , we have ρ0 (p) = 0; similarly
for every i, if p ∈ Uqi , then we get the inequality ρi (p)|gi (q) − gi (p)| < ρi (p)δ(p), and
if p ∈
/ Uqi , we get zero. But at least for one i, we get strict inequality. Hence the
last inequality follows, and the theorem is proven.
25 CHAPTER 1. MANIFOLDS

By considering each components, obviously we get the generalization for Rk -

valued continuous functions:

Theorem 1.30. Let M be a smooth manifold, let A be a closed subset of M . For

any continuous function g : M → Rk that is smooth on A, and for any continuous
δ : M → R>0 that is everywhere positive, there exists a smooth function f : M → Rk
such that

• For any p ∈ A, we have f (p) = g(p);

• For any p ∈ M , we have |f (p) − g(p)| < δ(p).

In particular, take A = ∅ and δ(p) = ε everywhere for a positive constant ε > 0,

we see that any continuous function g : M → Rk can be approximated by a smooth
function f : M → Rk , such that |f (p) − g(p)| < ε.
2 | Tangent and Cotangent Spaces

2.1 Tangent Space of a Manifold

Usually we have a curve, i.e. a smooth map γ : R → M (or γ : (a, b) → M where
the domain of the curve is an interval. But an interval is clearly homeomorphic to
R, so we can use these definitions interchangably), we want to know what is the
velocity of the curve γ at some point p.
Definition 2.1. Let (M, O, A) be smooth manifold. The curve γ : R → M is at
least C 1 with respect to charts of the smooth atlas A. Suppose γ hits a point p ∈ M
at λ0 ∈ R, i.e. γ(λ0 ) = p. Then the velocity of γ at p is the linear map
∼
vγ,p : C ∞ (M ) −
→ R,

where C ∞ (M ) is the set of all smooth functions f : M → R equipped with addition

and multiplication:

(f ⊕ g)(p) = f (p) + g(p);

(λ g)(p) = λ · g(p).

More specifically, the velocity is defined by the rule

vγ,p (f ) := (f ◦ γ)0 (λ0 ),

that is, the derivative of f ◦ γ at that point p where γ takes λ0 as an input.

Example 2.2. Let γ : R → M be the path you take when you are running in real
world. Let the temperature be f : M → R. Then as you run, you plot f ◦ γ, which
is the temperature with respect to time. Then (f ◦ λ)0 (t) is the rate of change of the
temperature at that time t, which you are so used to.
Definition 2.3. For each point p ∈ M , define the tangent space to M at p,
denoted by Tp M , as

Tp M := {vγ,p : γ is a smooth curve}.

That is, we take all possible curves that passes p, and find their velocity at point
p. These velocity vector spans a imaginary plane that forms the tangent space Tp M .
Now it is natural to ask whether Tp M , if it is like any plane in Euclidean space,
forms a vector space. We define the addition and multiplication on Tp M and show
that it is indeed a vector space.
27 CHAPTER 2. TANGENT AND COTANGENT SPACES

Definition 2.4. Define the addition ⊕ on Tp M to be

⊕ : Tp M × Tp M → Hom(C ∞ (M ), R)

with the rule

(vγ,p ⊕ vδ,p )(f ) := vγ,p (f ) + vδ,p (f ),
where ⊕ happens in Tp M and + happens in R. Similarly, define the multiplication
structure

: R × Tp M → Hom(C ∞ (M ), R)
to be
(λ vγ,p )(f ) := λ · vγ,p (f ),
where happens in Tp M and · happens in R.

Theorem 2.5. (Tp M, ⊕, ) is a vector space.

Proof. Suppose γ(λ0 ) = p and δ(λ1 ) = p. We prove the theorem by showing

1. There exists a curve σ such that vγ,p ⊕ vδ,p = vσ,p ;

2. There exists a curve τ such that α vγ,p = vτ,p for α ∈ R.

We prove the second claim first. We claim that the curve

τ (λ) := γ(αλ + λ0 ) = (γ ◦ µα )(λ)

works, where µα (λ) = αλ + λ0 . Notice that τ (0) = γ(λ0 ) = p. Then by definition,

vτ,p (f ) : = (f ◦ τ )0 (0) = (γ ◦ µα )0 (0) = (f ◦ γ)0 (λ0 ) · α = α · vγ,p (f ),

where the third step follows from the chain rule. This finished the second half of the
proof. Now, for the first part, make a choice of chart (U, φ) where p ∈ U . Define
the curve σ : R → M to be

σ(λ) = φ−1 ((φ ◦ γ)(λ0 + λ) + (φ ◦ δ)(λ1 + λ) − (φ ◦ γ)(λ0 )).

Notice that

σ(0) = φ−1 ((φ ◦ γ)(λ0 ) + (φ ◦ δ)(λ1 ) − (φ ◦ γ)(λ0 )) = δ(λ1 ) = p.

Now,

vσ,p (f ) : = (f ◦ σ)0 (0)

= (f ◦ φ−1 ◦ φ ◦ σ)0 (0)
= ((f ◦ φ−1 ) ◦ (φ ◦ σ))0 (0)
2.1. TANGENT SPACE OF A MANIFOLD 28

Notice that f ◦φ−1 : Rm → R, and φ◦σ : R → Rm are functions that are very easy to
differentiate by using the multidimensional chain rule. We continue the calculation
and use chain rule to get

vσ,p (f ) = ((f ◦ φ−1 ) ◦ (φ ◦ σ))0 (0)

Xm
d(φ ◦ σ)i ∂(f ◦ φ−1 )
= · i
(φ(σ(0)))
i=1
dλ ∂x
X  d(φ ◦ γ)i d(φ ◦ δ)i

∂(f ◦ φ−1 )
= (λ0 ) + (λ1 ) · i
(φ(p))
i
dλ dλ ∂x
X d(φ ◦ γ)i ∂(f ◦ φ−1 ) X d(φ ◦ δ)i ∂(f ◦ φ−1 )
= (λ0 ) · i
(φ(p)) + (λ 1 ) · i
(φ(p))
i
dλ ∂x i
dλ ∂x
= (f ◦ γ)0 (λ0 ) + (f ◦ δ)0 (λ1 )
= vγ,p (f ) + vδ,p (f ).

And we are done.

In the remaining of the text we will not use ⊕, and will switch to ordinary
addition and multiplication notation for adding and scalar multiplying vectors in
Tp M for convenience.
Now let us consider a point p ∈ M and a chart (U, φ) containing p. Let γ : R →
M such that γ(0) = p. Using the same method of calculation in our proof above,
we have that for f ∈ C ∞ (M ),
∂(f ◦ φ−1 ) d(φ ◦ γ)i
vγ,p (f ) = (f ◦ γ)0 (0) = (φ(p)) · (0).
∂xi dλ
Now we make two very important notational convention:
dγ i d(φ ◦ γ)i
(0) := (0)
 dλ  dλ
∂ ∂(f ◦ φ−1 )
f := (φ(p)).
∂xi p ∂xi

Then we have that

X dγ i  
∂
vγ,p (f ) = (0) · f.
i
dλ ∂xi p

Usually, we use Einstein’s summation convention, which dictates that if a repeated

index (like i in the equation above) appears twice, once upstairs and once downstairs,
will be automatically summed over, with no summation notation explicitly written
out. So the compact form of the equation above is
 
dγ i ∂
vγ,p (f ) = (0) · f.
dλ ∂xi p

From now on, we will use this convention throughout this notes, to be concise.
The notation (∂/∂xi )p looks deceptively simple and familar (after all it is just
like the partial derivative), but it is fundamentally different from the familiar partial
29 CHAPTER 2. TANGENT AND COTANGENT SPACES

derivative that we know. So, from now on, whenever it appears, what we mean is
the operation
 
∂ ∞ ∼ ∂(f ◦ φ−1 )
: C (M ) −
→R f 7→ (φ(p))
∂xi p ∂xi

which is linear over C ∞ (M ). Hence (∂/∂xi )p may be viewed as an element in Tp M .

In fact, since any tangent vector v ∈ Tp M can be expressed as a linear combination
of these partials, a curve with tangent vector (∂/∂xi )p at p is the one with

dγ j d(φ ◦ γ i )
(0) = =0
dλ dλ
unless j = i. That is, a curve γ such that its image φ ◦ γ in Rm is constant for every
j 6= i. The existence and uniqueness theorem of ODEs guarantee that this curve is
unique at least in a neighborhood of p. See Figure 2.1.

M
∂

x2
φ(U )
x2 ∂x1 p
φ x1
x 1 p
φ(p)
γ φ◦γ
U

Figure 2.1: A curve γ on M having (∂/∂x1 )p as its tangent vector at p.

Note that the notion of (∂/∂xi )p only makes sense if a chart (U, φ) around p is
specified, as φ is explicitly needed in its definition. But overall we would like to
show the following:

Theorem 2.6. Let (U, φ) ∈ AC ∞ be a chart containing p. Then the vectors

   
∂ ∂
,··· , ∈ Tp M
∂x1 p ∂xm p

consitute a basis for Tp M , where m is the dimension of the chart map φ : U →

φ(U ) ⊂ Rm , or the dimension of the manifold M . We call this basis a chart induced
basis for Tp M .

Proof. We have already shown that any vector vγ,p ∈ Tp M can be written as a linear
combination of these vectors in the discussion above. So we only need to show that
these vectors are linearly independent. We will show that
 
i ∂
λ = 0 implies λi = 0 for all i.
∂xi p
2.2. PUSHFORWARD OF A SMOOTH MAP 30

Now, consider a family x1 , ..., xm ∈ C ∞ (U ), defined as follows: For every p ∈ U , we

define xi (p) to be the ith component of φ(p) ∈ Rm . Note that xj ◦ φ−1 : φ(U ) ⊂
Rm → R is simply the map (a1 , ...., am ) 7→ aj , hence we have
  −1
i ∂(x ◦ φ )
j
∂
λ i
x j
= λ (φ(p)) = λi δij = λj = 0,
∂xi p ∂xi

which does imply λj = 0 for all j.

Corollary 2.7. dim Tp M = dim M.

Proof. This is immediate since given any chart (U, φ) around p, we have shown that
(∂/∂x1 )p , ..., (∂/∂xm )p is a basis.

Example 2.8. Let p = (p1 , ..., pn ) ∈ Rn . The tangent space Tp Rn consists of all
linear maps C ∞ (Rn ) → R defined by C 1 -curves γ : R → Rn . Let x1 , ..., xn be the
standard coordinate on Rn . A basis for Tp Rn consists of tangent vectors of the
constant curves η i (λ) = (p1 , ..., pi + λ, ..., pn ) at λ = 0. It is easy to see that the
associated basis (∂/∂xi )p turns out to be exactly the directional derivatives in the
ith direction, evaluated at p. Let e1 = (1, ..., 0), ..., en = (0, ..., 1) be the standard
basis for Rn . Then the identification
X n   Xn
∂ ∼
=
v= v i
i
−
→ v i ei ∈ Rn
i=1
∂x p i=1

gives an isomorphism Tp Rn ∼
= Rn of vector spaces.

2.2 Pushforward of a Smooth Map

Definition 2.9. Let F : M → N be a smooth map between manifolds. For each
p ∈ M we define a map
dFp : Tp M → TF (p) N
as follows: for any given f ∈ C ∞ (N ) and v ∈ Tp M , we define dFp (v) ∈ TF (p) (N ) via

dFp (v)(f ) := v(f ◦ F ).

Note that this definition makes sense, because f ◦ F ∈ C ∞ (M ), and v acts linearly
on C ∞ (M ). This map dFp is called the differential or pushforward of F at p.
See Figure 2.2 for a more geometric description of the pushforward.

Proposition 2.10. Let M, N, P be smooth manifolds, F : M → N, G : N → P be

smooth maps, and p ∈ M .

1. dFp : Tp M → TF (p) N is linear;

2. d(G ◦ F )p = dGF (p) ◦ dFp ;

3. d(IdM )p : Tp M → Tp M is exactly idTp M ;

31 CHAPTER 2. TANGENT AND COTANGENT SPACES

M N
F (γ)
v
dFp (v)
γ F
F (p)
p

Figure 2.2: If v is the tangent vector of a curve γ at p and γ(0) = p, then for
any g ∈ C ∞ (M ), we have v(g) = (g ◦ γ)0 (0). Then using the definition dFp (v)(f ) =
v(f ◦ F ) = (f ◦ F ◦ γ)0 (0). Therefore, the geometric picture of dFp can be viewed
as follows: it maps v, the tangent vector of γ ⊂ M at p, to the tangent vector of
F ◦ γ ⊂ N at F (p).

4. If F is a diffeomorphism, then dFp : Tp M → TF (p) N is an isomorphism of

vector spaces, and (dFp )−1 = d(F −1 )F (p) .

Proof. 1. This is clear from definition, since v is linear; 2. This is also clear because
for f ∈ C ∞ (P ) and v ∈ Tp M , we have

dGF (p) ◦ dFp (v)(f ) = dFp (v)(f ◦ G) = v(f ◦ G ◦ F ) = d(G ◦ F )p (v);

3. This is clear from definition; 4. This follows from 2 and 3.

Example 2.11 (Tangent space of a finite-dimensional vector space). Let V be a

n-dimensional real vector space, and let α1 , ..., αn be a basis for V . Any norm on V
makes V into a topological space with the same Ptopology (see Example 1.15). There
is a map ϕ : V → R which identifies w = i w αi ∈ V with (w1 , ..., w n ) ∈ Rn .
n i

This map is a diffeomorphism (in fact a chart) by Example 1.15. Therefore, we can
make sense of dϕw : Tw V → Tϕ(w) Rn . Then the three isomorphisms in the diagram
below defines an identification isomorphism θ : V → Tw V :

θ
V Tw V
ϕ dϕw

Rn ∼
=
Tϕ(w) Rn

Hence we have shown that for any finite-dimensional vector space V ∼

= Tw V .

Example 2.12 (Another way of viewing basis for Tp M ). Let (U, φ) be a chart
around p ∈ M . Then we may view φ−1 : φ(U ) → U a smooth map (in fact a
diffeomorphism) between manifolds. Note that because curves on U are curves on
M as well, and curves on M restricts to curves on U , this gives an identification
Tp U ∼
= Tp M .Since φ−1 is a diffeomorphism, the pushforward

d(φ−1 )φ(p) : Tφ(p) φ(U ) → Tp U ∼

= Tp M
2.2. PUSHFORWARD OF A SMOOTH MAP 32

is an isomorphism between vector spaces. We have that, for any f ∈ C ∞ (M )

  !  
−1 ∂ ∂ −1

∂
d(φ )φ(p) (f ) = f ◦ φ (φ(p)) = f.
∂xi φ(p) ∂xi ∂xi p

Hence we see that the basis for Tp M induced by the chart (U, φ) is just the image
of the standard basis in Tφ(p) φ(U ) ⊂ Rn under the isomorphism d(φ−1 )φ(p) of vector
spaces.
We conclude this section by studying the expression of the differential dF :
Tp M → TF (p) N in coordinates.
Theorem 2.13. With respect to the basis for Tp M induced by a chart (U, φ) around p
and the basis for TF (p) N induced by a chart (V, ψ) of N containing F (U ), the matrix
representation of the linear map dFp : Tp M → TF (p) N is given by the Jacobian of
F̂ := ψ ◦ F ◦ φ−1 : φ(U ) → ψ(V ) evaluated at φ(p).
Proof. Let (∂/∂x1 )p , ..., (∂/∂xm )p a basis for Tp U induced by the chart (U, φ), and
similarly let (∂/∂y 1 )F (p) , ..., (∂/∂y n )F (p) a basis for TF (p) V induced by the chart
(V, ψ). We would like to study dFp in matrix form with respect to these base.
Now let f ∈ C ∞ (N ) be arbitrary, by chain rule we have
 !
∂ ∂

dFp i
f= i
f ◦ F ◦ φ−1 (φ(p))
∂x p ∂x
∂

= i
f ◦ ψ −1 ◦ ψ ◦ F ◦ φ−1 (φ(p))
∂x
∂(f ◦ ψ −1 ) ∂(ψ ◦ F ◦ φ−1 )j
= (ψ(F (p))) · (φ(p))
∂y j ∂xi
 
∂ F̂ j ∂
= (φ(p)) · f
∂xi ∂y j F (p)

where F̂ := ψ ◦ F ◦ φ−1 : φ(U ) → ψ(V ).

The result above also answers the question that how does the component of a
vector change under change of coordinates. Let p ∈ M and let (U, φ) and (V, ψ) be
two overlapping charts whose intersection contains p. Then the change of coordinate
map is just ψ◦φ−1 : φ(U ∩V ) → ψ(U ∩V ). Note that in this case we have F = idU ∩V .
Then by the result above we have
   
∂ ∂(ψ ◦ φ−1 ) ∂
= (φ(p)) · .
∂xi p ∂xi ∂y j p
Usually, we use a set of coordinate (x1 , ..., xm ) to reprensent points in φ(U ∩ V ),
where xi ∈ R. Similarly, we use (y 1 , ..., y m ) to represent points in ψ(U ∩V ). One can
view y 1 , ..., y m as functions of x1 , ..., xm , i.e., for a point p ∈ U ∩ V , its y j coordinate
is the j-th component of ψ ◦ φ−1 (x1 , ..., xm ), where x1 , ..., xm is the coordinate of p
in φ(U ∩ V ). Then the formula above can be conveniently written as
∂ ∂y j ∂
= .
∂xi ∂xi ∂y j
33 CHAPTER 2. TANGENT AND COTANGENT SPACES

This is easy to remember because of its resemblance to the chain rule formula in
multivariable calculus. This also generalizes to arbitrary tangent vectors v ∈ Tp M .
We first write v as a linear combination of basis vectors, then by above we get
j
∂ i ∂y ∂ ∂
v = vi i
= v i j
= ṽ j j ,
∂x ∂x ∂y ∂y
or simply
∂y j
ṽ j = v i .
∂xi
Example 2.14 (Tangent space of the circle). In this example, we explicitly calculate
the tangent space of S 1 at a point (a, b).
Rotating the circle if necessary, we may assume (a, b) lies on the upper half circle,
i.e. b > 0. Then a local parametrization is given by ψ : (−1, 1) → S 1 , where
√
ψ(x) = (x, 1 − x2 ).

Then its derivative is given by

 
1
dψx = .
− √1−x
x
2

Note that since dψx : R → R2 , this map should take a real number and
√ gives a vector,
which is done by scalar multiplication. Now if we let x = a, then 1 − x2 = b since
(a, b) is on the upper half circle. Therefore
     
1 1 −b
T(a,b) (S ) = dψa (R) = R
1
= R(−b) =R .
−a/b −a/b a

This is what we expect: the tangent space of the circle at the point p = (a, b) should
be spanned by the vector (−b, a), which is perpendicular to the radial vector (a, b)
at the point p.

2.3 Tensors
We provide a section that briefly reviews the concept of tensors.
Definition 2.15. Let (V, +, ×) be a finite dimensional vector space. An (r, s)
tensor T is a multilinear map
∗ ∗ ∗
T :V · · · × V} × V
| × V {z | × V {z· · · × V } → R.
r s

Multilinear means the tensor is linear in each slots. Say if T is a (1, 1) tensor.
Let φ, ψ ∈ V ∗ , let v, w ∈ V , and let λ ∈ F (some field, e.g. real or complex). Then

T (v, φ + ψ) = T (v, φ) + T (v, ψ)

T (v, λφ) = λT (v, φ).

Similarly for the first slot.

2.3. TENSORS 34

Example 2.16. Let P(R) be the vector space of polynomials on R and let g :
P(R) × P(R) → R be defined as
Z 1
g(p, q) = p(x)q(x)dx.
−1

Then you can verify that g is a (2, 0) tensor. From linear algebra, you know this is
an inner product. More generally speaking, inner products are (2, 0) tensors.
Of course, φ : V → R is a (1, 0) tensor.
Vectors are (0, 1) tensors. This is because, for finite dimensional vector space,
∗ ∗ ∼
(V ) = V . The identification is given by the correspondence that for any v ∈ V
and α ∈ V ∗ , we may define v(α) := α(v) ∈ R, hence viewing v as an element in V ∗ .
Similar to choosing a basis in vector space V , we can choose a basis for V ∗ .
Definition 2.17. Let V be a finite dimensional vector space with basis e1 , · · · , en .
Define ε1 , · · · , εn ∈ V ∗ where
εa (eb ) = δba .
ε1 , · · · , εn are uniquely determined by the choice of e1 , · · · , en . It is easy to verify
that ε1 , ..., εn form a basis for V ∗ . We call ε1 , · · · , εn the dual basis for V .
Example 2.18. Consider the vector space P3 (R), the set of real polynomials with
degree no more than 3. Let
e0 (x) = 1 e1 (x) = x e2 (x) = x2 e3 (x) = x3 ,
Let
1 a
εa = ∂
a! x=0
a
Then ε (eb ) = δba is a dual basis.
Definition 2.19. Let T be an (r, s) tensor on a finite dimensional vector space V
(dim V = n < ∞). Let e1 , · · · , en be the chosen basis for V . We can define the
components of T by
Tji11,··· ,js = T (ej1 , · · · , ejs , ε , · · · , ε ),
,··· ,ir i1 ir

where i1 , · · · , ir , j1 , · · · , js ∈ {1, 2, · · · , dim V }.

Example 2.20. Let T be a (1, 1) tensor. Then Tji = T (ej , εi ). Therefore,

XV
dim XV
dim
T (v, φ) = T ( v j ej , φ i εi )
j=1 i=1
XX
= φi v T (ej , εi )
j

i j
XX
= φi v j Tji
i j

= φi v Tji ,
j

where the second step is by the multilinear property of T . In the last step we use
Einstein summation convention and contract the indices.
35 CHAPTER 2. TANGENT AND COTANGENT SPACES

2.4 Cotangent Spaces

Definition 2.21. The dual space of Tp M , or the cotangent space, is defined by
Tp∗ M = {linear maps α : Tp M → R}.
Definition 2.22. The gradient of f ∈ C ∞ (M ) at a point p, denoted by dfp :
Tp M → R, is given by the rule
v 7→ dfp (v) := v(f ).
We can now compute the component of the gradient. Its jth component is given
by gradient acting on the jth basis for Tp M . That is,
   
∂ ∂ ∂(f ◦ φ−1 )
(df )p = f = (φ(p)).
∂xj ∂xj p ∂xj
Now this looks like the gradient that we are familiar with in multivariable calculus!
Another note on our choice of notation: Recall from the previous section that
given smooth function f : M → R, we can define a differential
dfp : Tp M → Tf (p) R.
∼
If we use the identification Tf (p) R = R by identifying ∂x f (p) ∈ Tf (p) R associated
∂

to the chart induced by the identity map on R with 1 ∈ R, then the two definitions
coincide, as the formula above is exactly Theorem 2.13.
Theorem 2.23. Let (U, φ) be a chart and xi : U → R be the smooth function defined
for any p ∈ U by giving the ith component of φ(p) ∈ Rm . Then the gradient vectors
(dx1 )p , (dx2 )p , · · · , (dxm )p ∈ Tp∗ M
form a basis for Tp∗ M .
Proof. By the calculation above, it is easy to see that
 !
∂ ∂(xi ◦ φ−1 ) ∂xi 1
(dxi )p = (φ(p)) = (x , ..., xm ) = δji .
∂xj p ∂xj ∂xj


That is, (dxi )p ’s are the dual basis to the basis ∂x∂ j p ’s in Tp M . Hence it is a basis
for Tp∗ M .
Now we can talk about the change of components of a covector under a change of
charts. Similar to our discussion on change of components of vectors, let (U, φ), (V, ψ)
be two overlapping charts containing p, and let xi , y j be the respective coordinates.
Then we have
(dxi )p vγ,p = vγ,p (xi ) = (xi ◦ γ)0 (λ0 ) = (xi ◦ ψ −1 ◦ ψ ◦ γ)0 (λ0 )
∂(xi ◦ ψ −1 )
= (y j ◦ γ)0 (λ0 ) · (ψ(p))
∂y j
∂(xi ◦ ψ −1 )
= vγ,p (y j ) · (ψ(p))
∂y j
∂(xi ◦ ψ −1 )
= (ψ(p)) · (dy j )p vγ,p .
∂y j
2.4. COTANGENT SPACES 36

Note that xi ◦ ψ −1 : ψ(U ∩ V ) → R is the function that gives the ith component
of the map φ ◦ ψ −1 : ψ(U ∩ V ) → φ(U ∩ V ). Similar to our previous argument on
change of components of tangent vectors, we may view xi as a function of y 1 , ..., y m ’s,
so that xi (y 1 , ..., y m ) is the ith component of φ ◦ ψ −1 (y 1 , ..., y m ). So the equation
above can be conveniently written as
 i
i ∂x
(dx )p = (dy j )p .
∂y j p

This formula is very easy to remember as it resembles the differentials we learned

in multivariable calculus.
This also works for general covectors, i.e., general elements in Tp∗ M , other than
basis vectors. Given α ∈ Tp∗ M , write

α = αi (dxi )p .

Then under a change of coordinate from xi ’s to y j ’s on the level of manifolds, this

induces a change of basis in Tp M , which then gives a new representation of α in
terms of (dy j )0 s by
 i
i ∂x
α = αi (dx )p = αi (dy j )p = α̃j (dy j )p ,
∂y j p

or simply
∂xi
α̃j = αi .
∂y j
Remark 2.24. Coming back to gradient dfp of a smooth function f ∈ C ∞ (M ). At
a local chart around p we may write dfp as a linear combination dfp = βi (dxi )p . To
figure out what the coefficient βi is, we note that
   
∂ i ∂
dfp = βi dx = βi δji = βj .
∂xj ∂xj

Then by the previous calculation we know

∂(f ◦ φ−1 )
βj = (φ(p)).
∂xj
Note that in the case where M = Rn , and φ is the identity map, we have exactly
∂f
dfp = j
(dxj )p .
∂x
Therefore, differentials should really be viewed as elements in the cotangent space
Tp∗ Rn . There are higher dimensional generalizations of them called differential forms,
which we will get to later.
3 | Vector Bundles

3.1 Basic Definitions and Examples

Definition 3.1. A bundle is a triple (E, M, π), where E is a smooth manifold
(called the total space), so is M (called the base space). And π : E → M a smooth
and surjective map.

• Let π : E → M be a bundle. Let p ∈ M . The fiber over p is defined to be

the preimage of p with respect to π, i.e. π −1 ({p}), frequently we simply write
π −1 (p).

• A section of a bundle σ is a map σ : M → E such that π ◦ σ = idM . If σ is

smooth (in the sense of smooth maps between manifolds), we call σ a smooth
section.

There is also the notion of a topological bundle, where the definition is the same
as above, except that we only require E, M to be topological spaces, and π : E → M
be a continuous and surjective map.
Note that from the definition above, in particular the surjectivity of π, we see
that as a set, [
E= π −1 (p)
p∈M

is a union of its fibers.

Example 3.2. Let M = S 1 be the circle, and E = S 1 × R be the cylinder (See

Figure 3.1). The projection π : E → M is defined by π(p, λ) = p, where p ∈ S 1 ,
and λ ∈ R. A section σ : M → E can be thought of a map σ(p) = (p, λ(p)), for all
p ∈ S 1 , and λ : S 1 → R is some function. If λ varies smoothly with respect to p,
then we expect to see a smooth section.

Definition 3.3. Let M be a smooth manifold. A (real) smooth vector bundle

of rank k over M is a bundle with the following two extra conditions:

(i) For each p ∈ M , the fiber Ep = π −1 ({p}) over p is endowed with the structure
of a k-dimensional real vector space.

(ii) For each p ∈ M , there exists a neighborhood U of p in M and a diffeomorphism

Φ : π −1 (U ) → U × Rk (called the local trivialization of E over U ),
satisfying the following three conditions:
3.2. TANGENT BUNDLE AND COTANGENT BUNDLE 38

σ(M )
p
M

π −1 (p)

Figure 3.1: An illustration of E = S 1 × R. For a concrete example, imagine a

particle is moving along a circle M with varying speed, and at every point p ∈ M ,
we associate its speed at that point v(p). Since v should depend smoothly on p, we
get a smooth section defined by σ(p) = (p, v(p)).

• πU ◦ Φ = π where πU : U × Rk → U is the projection.

• For each q ∈ U , the restriction of Φ to Eq is a vector space isomorphism
from Eq to {q} × Rk ∼ = Rk .
• Let U ∩ V 6= ∅ on M and Ψ : π −1 (V ) → V × Rk be a trivilization. The
map
Ψ ◦ Φ−1 : (U ∩ V ) × Rk → (U × V ) × Rk
are vector-space automorphisms of Rk in each fiber and hence give rise
to maps gU V : U ∩ V → GL(n, R) defined by gU V (p) = Ψ ◦ Φ−1 |{p}×Rk .

3.2 Tangent Bundle and Cotangent Bundle

We will now construct the tangent bundle of a smooth manifold. This is the most
important example of a vector bundle. The construction is rather complicated when
one first encounter it, but it is worth going through the whole process, because it
will be the guiding examples of how one can construct vector bundles of various
ranks over a smooth manifold M (as you will see, there is basically only one way,
which is achieved by mimicking our construction of the tangent bundle). Now that I
have convinced you that this construction is important, let us start the work. First,
let (M m , O, A) be a smooth manifold.

Step 1
Define the tangent boundle (on the level of set) as
a
T M := Tp M,
p∈M

which is the disjoint union of all the tangent space Tp M . Here, the disjoint union is
the set defined by a
Tp M = {(p, v) : p ∈ M, v ∈ Tp M }.
p∈M
39 CHAPTER 3. VECTOR BUNDLES

Step 2
Define our map π : T M → M such that
π : (p, v) 7→ p,
where p ∈ M, v ∈ Tp M . By the definition of T M , since we unite over all p ∈ M , we
know this map π will hit all points on M . Hence π is a surjective map. Now we have
π : T M → M where T M only has a set structure and we don’t know whether it is
smooth or not. π is only a surjective map for now. We don’t know if it is smooth
either. M is a smooth manifold by assumption. If we want to make this triple a
bundle, then we have to make T M a smooth manifold.

Step 3
Now, T M is just a set. We first make T M a topological manifold first by giving
it a proper topology. Let us require OT M to be the coarsest topology to make π
continuous. We define
OT M := {π −1 (U ) : U ∈ O}.
We claim that this is indeed a topology.
Proof. We verify that this axiom satisfies the definition of a topology, i.e., Definition
0.1:
1. Note that ∅ ∈ O, and π −1 (∅) = ∅ since π is surjective, we have ∅ ∈ OT M . Also,
M ∈ O and π −1 (M ) = T M since π surjective, we have T M ∈ O.

2. Let U 0 , V 0 ∈ OT M . Then there exists some U, V ∈ O such that π(U 0 ) = U

and π(V 0 ) = V . Then one can easily check that U 0 ∩ V 0 = π −1 (U ∩ V ). Since
U ∩ V ∈ O, we have U 0 ∩ V 0 ∈ OT M by definition.

3. Let Ui0 ∈ OT M for some i ∈ I. Then there exists Ui ∈ O such that we have
Ui0 = π −1 (Ui ) for the same i ∈ I. Then
[ [
Ui0 = π −1 (Uj ).
i∈I j∈I
S
But since Uj ∈ O, we have i∈I Ui0 ∈ OT M .
This concludes the proof.

Step 4
Now with a given topology, we construct a C ∞ atlas from A on M . Define
AT M := {(π −1 (U ), ξφ ) : (U, φ) ∈ A},
where ξφ : π −1 (U ) → R2m defined by the rule
ξφ : (p, v) 7→ (x1 (p), ..., xm (p), (dx1 )p (v), ..., (dxm )p (v)).
3.2. TANGENT BUNDLE AND COTANGENT BUNDLE 40

Recall that xi (p) is the ith coordinate of φ(p) ∈ Rm . All this is saying is that given
a chart (U, φ) containing p, we take the first m coordinate of (p, v) ∈ T M under ξφ
to be the coordinates of p under φ; then for the second m coordinates, express v as
a linear combination of basis tangent vectors in Tp M induced by the chart φ (see
the previous chapter),  
j ∂
v=v .
∂xj p
Then our second m coordinates will be v 1 , ..., v m . It is easy to see that the inverse
of ξφ is given by
 !
∂
ξφ−1 : (α1 , ..., αm , β 1 , ..., β m ) 7→ φ−1 (α1 , ..., αm ), β j ∈ T M.
∂xj p

Step 5
To show that we have constructed a smooth atlas, we still need to show that the
chart transition maps are smooth. Hence, let (U, φ), (V, ψ) be two overlapping charts
on M , and xi , y j be the respective coordinates. Then
 !
∂
ξψ ◦ ξφ−1 (α1 , ..., αm , β 1 , ..., β m ) = ξψ φ−1 (α1 , ..., αm ), β j
∂xj p
k
 !
∂y ∂
= ξψ φ−1 (α1 , ..., αm ), β j j
∂x ∂y k p
 1 m

−1 j ∂y j ∂y
= ψ ◦ φ (α , ...α ), β
1 m
, ..., β .
∂xj ∂xj

This is clearly smooth with respect to α1 , ..., αm , β 1 , ..., β m , because recall that
∂y k /∂xj = ∂(ψ ◦ φ)/∂xj by our definition, and since A is a smooth atlas, we know
that ψ◦φ−1 ∈ C ∞ . By Theorem 1.9, we see that AT M determines a smooth structure
on T M , making T M a smooth manifold.

Conclusion
This completes the construction. See what have we got: we defined a smooth
manifold T M , and a smooth and surjective map π : T M → M . And finally, we get
a bundle T M , which is called the tangent bundle of M , as it is constructed by
gluing pieces of tangent spaces of M . Concretely, we have constructed a space that
contains the information of all points of M along with all possible tangent spaces of
M . To a physicist, M can be thought of as a space where the particle of interests
move along, and the particle, when at a point p ∈ M , has a velocity v ∈ Tp M . As
the particle moves along M , physicists may want to write down the pair (p, v), for
p ∈ M all points the particle was at, and v the corresponding velocity at p. What
physicist will get eventually is a curve in T M .
Remark 3.4 (The Cotangent Bundle). One can construct cotangent bundle of a
smooth manifold M m in basically the same way: In step 1, we define the cotangent
41 CHAPTER 3. VECTOR BUNDLES

`
bundle T ∗ M = p∈M Tp∗ M ; The projection map π and the topology OT ∗ M is defined
exactly as in Step 2 and Step 3 above; The chart construction (π −1 (U ), ξφ ) for a
chart (U, φ) is similar to that in step 4: the first m coordinates of ξφ (p, α) for
p ∈ U, α ∈ Tp∗ M is still given by the coordinates φ(p) ∈ Rm , and then express
α = αi (dxi )p , then the last m coordinates will be α1 , ..., αm ; Similar to Step 5, one
can show that this indeed defines a smooth atlas.
Example 3.5 (Tangent Bundle of S 1 ). The tangent bundle T S 1 of the circle S 1 is
diffeomorphic to S 1 × R, which is shown in Figure 3.1. To see this, note that for
any p = (a, b) ∈ S 1 ⊂ R2 , we have dim Tp S 1 = 1, as S 1 is one dimensional. In fact,
the calculation in Example 2.14 shows that T(a,b) S 1 is spanned by a single tangent
vector ∂/∂θ|(a,b) = (−b, a). Hence we can define a map ψ : T S 1 → S 1 × R via
ψ(p, λ · ∂/∂θ|p ) = (p, λ). It is easy to check that ψ is a diffeomorphism.

3.3 Vector, Covector, and Tensor Fields

Definition 3.6. A smooth vector field X on M is a smooth section X : M → T M
of the tangent bundle, i.e., π ◦ X = idM .
Definition 3.7. We collect all smooth sections (vector fields) on a manifold M and
denote it by
Γ(T M ) := {X : M → T M smooth sections}
with addition and multiplication structure

(X ⊕ X̃)f = X(f ) + X̃(f ),

7 X(p)f . The second addition + takes place in C ∞ (M ). And for

where X(f ) : p →
multiplication : C ∞ (M ) × Γ(T M ) → Γ(T M ) is defined by

(g X)f = g · X(f ),

where the second multiplication · takes place in C ∞ (M ).

This set with addition and multiplication defined as above (Γ(T M ), ⊕, ⊗) is a
∞
C vector space (verify) over a ring. This is because Γ(T M ) does not satisfy the
inverse of multiplication axiom and is, therefore, not a field, but a ring. We call
such structure a module. Thus (Γ(T M ), ⊕, ⊗) is a C ∞ (M )-module.
A vector space over a ring, or a module, may not have all the properties a
vector space have. For example, we cannot find vector fields, for any manifold M ,
X1 , · · · , Xd ∈ Γ(T M ), such that for every vector field X ∈ Γ(T M ), X = ai Xi .
Although this can always be done locally in a chart (π −1 (U ), ξφ ) using the smooth
vector fields ∂/∂xi , where ∂/∂xi |p is the ith basis vector induced by the chart φ on
Tp M .
Definition 3.8. Similarly, we could define the covector fields as

Γ(T ∗ M ) := {α : M → T ∗ M smooth sections}.

This is also a C ∞ (M )-module.

3.3. VECTOR, COVECTOR, AND TENSOR FIELDS 42

Then we can introduce the definition of a tensor field.

Definition 3.9. An (r, s) tensor field T is a multilinear map

T : Γ(T M ) × · · · × Γ(T M ) × Γ(T ∗ M ) × · · · × Γ(T ∗ M ) → C ∞ (M )

| {z } | {z }
r s

where we used C ∞ (M ) rather than R because the tensor field need not to be a
constant. Instead, it varies from point to point.

Remark 3.10. Notice that, for every p ∈ M , and X ∈ Γ(T M ), we have X(p) ∈
Tp M . Hence, at a point, a vector field is just a vector in the tangent space. Similarly
for α ∈ Γ(T ∗ M ), we have α(p) ∈ Tp∗ M , hence at a point a coverctor field is just a
covector. Similarly, at a point p, a tensor field is just a tensor in the sense of Section
2.3, where V = Tp M .
4 | Differential Forms

4.1 Symmetric and Alternating Tensors, Wedge

Product
Denote
∗
| ⊗ ·{z
T r,s (V ) = V · · ⊗ V} ⊗ V · · ⊗ V }∗ .
| ⊗ ·{z
r s
∗
Q
Then T (V ) = ⊗ V is the set of multilinear maps k V → R, defined via α1 ⊗
0,k k

· · · αk : (v1 , ..., vk ) 7→ α1 (v1 )α2 (v2 ) · · · αk (vk ). Then we have two types of important
tensors: If α ∈ T 0,k (V ), then α is said to be symmetric if
α(v1 , ..., vk ) = α(vσ(1) , ..., vσ(k) )
Q
for any (v1 , ..., vk ) ∈ k V and σ ∈ Sk . Equivalently, one can define α to be the
unchanged by interchanging any pair of arguments. We say α is alternating if
α(v1 , ..., vk ) = sgn(σ)α(vσ(1) , ..., vσ(k) )
Q
for any (v1 , ..., vk ) ∈ k V and σ ∈ Sk . Equivalently, one can define α to be the
tensor that takes the negative of the original value by interchanging any pair of
arguments.
Now, given any α ∈ T 0,k (V ), we can define its symmetrization S (α) to be
1 X
S (α)(v1 , ..., vk ) = α(vσ(1) , ..., vσ(k) )
k! σ∈S
k

It is clear that S (α) is a symmetric tensor. One can also define the antisym-
metrization A (α) to be
1 X
A (α)(v1 , ..., vk ) = sgn(σ)α(vσ(1) , ..., vσ(k) ).
k! σ∈S
k

Proposition 4.1. For any α ∈ T 0,k

(V ), its antisymmetrization A (α) is alternating.
Q
Proof. Let τ ∈ Sk and (v1 , ..., vk ) ∈ k V . Then we calculate
1 X
A (α)(vτ (1) , ..., vτ (k) ) = sgn(σ)α(vσ◦τ (1) , ..., vσ◦τ (k) )
k! σ∈S
k

1 X sgn(η)
= α(vη(1) , ..., vη(k) )
k! η∈S sgn(τ )
k

= sgn(τ )A (α)(vτ (1) , ..., vτ (k) )

4.1. SYMMETRIC AND ALTERNATING TENSORS, WEDGE PRODUCT 44

where we made the substitution η = σ ◦ τ and we used sgn(η) = sgn(σ) sgn(τ ). But
since sgn(τ ) = ±1, so dividing by sgn(τ ) is the same as multiplying by it.

Now we may define

∧k V ∗ = A (⊗k V ∗ ) = {σ ∈ ⊗k V ∗ : σ alternating}.

If α ∈ ∧k V ∗ , then we define deg α := k to be the degree of α.

Definition 4.2. Let V n be a vector space, and ε1 , ..., εn any basis for V ∗ . Let
I = (i1 , ..., ik ) be a multi-index of length k, where 1 ≤ is ≤ n. We define the
so called elementary alternating tensor εI to be the tensor such that for any
v1 , ..., vk ∈ V ,
εi1 (v1 ) · · · εi1 (vk )
εI (v1 , ..., vk ) := .. ..
. ··· . .
εik (v1 ) · · · εik (vk )
By definition, εI is alternating because the determinant is, hence εI ∈ ∧k V ∗ .

Lemma 4.3. Let e1 , ..., en be a basis for V , and ε1 , ..., εn be the dual basis for V ∗ .
Let εI be defined as above where I = (i1 , ..., ik ).

(1) If I has an repeated index, then εI = 0.

(2) If J = Iσ := (iσ(1) , ..., iσ(k) ) where σ ∈ Sk , then εJ = (sgn σ)εI .

(3 We have, for a multi-index J = (j1 , ..., jk )

δji11 · · · δji1k
εI (ej1 , ..., ejk ) = ... · · · .. =: δ I .
. J
δjik1 · · · δjikk

Proof. For (1), if I has an repeated index, then the determinant in the definition
of εI will have two identical rows, hence is equal to zero; Next, (2) is immediate
because interchanging I by σ amounts to interchanging columns of the determinant
in the definition of εI by σ; Finally, (3) is just obvious from the definition of εI .

Theorem 4.4. Let V n be a vector space with basis e1 , ..., en and ε1 , ..., εn the dual
basis for V ∗ . Then, for each positive k ≤ n, the collection of εI with increasing
multi-index of length k, i.e.,

E := {εI : I = (i1 , ..., ik ), i1 < i2 < · · · < ik }

is a basis for ∧k V ∗ . Therefore, dim ∧k V ∗ = nk . Specifically, dim ∧k V ∗ = 0 if
k > n.

Proof. To show that E spans ∧k V ∗ : For each multi-index I = (i1 , ..., ik ), define
αI := α(ei1 , ..., eik ). Since α is alternating, αI = 0 if I has an repeated index and
45 CHAPTER 4. DIFFERENTIAL FORMS

αJ = (sgn σ)αI if J = Iσ . Thus the collection of real numbers αI with I increasing

determines α completely. We claim that
X
α= α I εI .
I increasing

To see this, let J = (j1 , ..., jk ) be any multi-index set. Note that by the definition
of δJI above, we can see that δJI = 0 if I or J has an repeated index or if J is not a
permutation of I; and δJI = sgn σ if neither I nor J has a repeated index and J = Iσ
for some σ ∈ Sk . Then by Lemma 4.3,
X X
αI εI (ej1 , ..., ejk ) = αI δJI = αJ = α(ej1 , ..., ejk ).
I increasing I increasing

This proves the claim, showing that E spans ∧k V ∗ .

To show that E is linearly independent: Suppose that for some collections of
coefficients αI we have X
σ := αI εI = 0.
I increasing

Let J be any increasing multi-index. Then again we have

X X
0= αI εI (ej1 , ..., ejk ) = αI δJI = αJ = 0.
I increasing I increasing

Hence αJ = 0 for all coefficients αJ . Thus σ ≡ 0. This proves linear independence.

Definition 4.5. The wedge product ∧ is a bilinear map ∧ : ∧r V ∗ × ∧s V ∗ →

∧r+s V ∗ by
(r + s)!
∧(α, β) = α ∧ β = A (α ⊗ β).
r!s!
Theorem 4.6. Let V n be a vector space with basis e1 , ..., en and ε1 , ..., εn the dual
basis for V ∗ . For any multi-index I = (i1 , ..., ik ), J = (j1 , ..., jl ), define IJ :=
(i1 , ..., ik , j1 , ..., jl ). Then we have

εI ∧ εJ = εIJ .

Proof. By linearity, it suffices to show that εI ∧ εJ and εIJ give the same result for
any sequence (ep1 , ..., epk+l ) of basis vectors. Denote P = (p1 , ..., pk+l ).
We first reduce the problem by restricting our attention to nontrivial multi-index
P in which the equation is not obviously true. Note that:

• If P contains an index that does not appear in either I or J, then by Lemma

4.3, both sides are zero.

• If P contains a repeated index, then since both sides are clearly alternating
tensors, both sides are again zero.

• We will discuss the case where P = IJ and has no repeated index later.
4.1. SYMMETRIC AND ALTERNATING TENSORS, WEDGE PRODUCT 46

• If P is a permutation of IJ and has no repeated indices, and if εI ∧ εJ =

εIJ is true on eP where P = IJ, then say P = (IJ)σ for some σ ∈ Sk+l .
Again because both sides are alternating, we see that the previous case implies
(sgn σ)(εI ∧ εJ ) = (sgn σ)εIJ , which proves what we want.
Therefore, we only need to check the case where P = IJ and has no repeated indices.
Direct computation using the definition shows that
(k + l)!
εI ∧ εJ (ep1 , ..., epk+l ) = A (εI ⊗ εJ )(ep1 , ..., epk+l )
k!l!
1 X
= (sgn σ)εI (epσ(1) , ..., epσ(k) )εJ (epσ(k+1) , ..., epσ(k+l) ).
k!l! σ∈S
k+l

By Lemma 4.3 again, the only terms in the sum above that is nonzero are the ones
in which σ permutes the first k indices and the last l indices of P separately. In
other words, σ = τ η where τ ∈ Sk permutes the index 1, ..., k, and η ∈ Sl permutes
the index k + 1, ..., k + l. Since sgn(τ η) = (sgn τ )(sgn η), we have
1 X
εI ∧ εJ (ep1 , ..., epk+l ) = (sgn τ )(sgn η)εI (epτ (1) , ..., epτ (k) )εJ (epk+η(1) , ..., epk+η(l) )
k!l! τ ∈S ,η∈S
k l

1 X 1X
= (sgn τ )εI (epτ (1) , ..., epτ (k) ) · (sgn η)εJ (epk+η(1) , ..., epk+η(l) )
k! τ ∈S l! η∈S
k l

= A (ε )(ep1 , ..., epk ) · A (ε )(epk+1 , ..., epk+l )

I J

= εI (ep1 , ..., epk ) · εJ (epk+1 , ..., epk+l )

= 1,

where the second to last inequality is because εI , εJ are already alternating, and
the last inequality is because of the assumption P = IJ. Note that since P = IJ,
clearly εIJ (eP ) = 1 by Lemma 4.3, so indeed εI ∧ εJ = εIJ .
Proposition 4.7. The wedge product is bilinear, associative, and anticommutative,
i.e., ω ∧ η = (−1)deg ω deg η η ∧ ω.
Proof. Bilinearity is immediate from definition of wedge product, because both the
tensor product and alternation are bilinear. To prove associativity, note on basis
elements associativity holds:

(εI ∧ εJ ) ∧ εK = εIJ ∧ εK = εIJK = εI ∧ εJK = εI ∧ (εJ ∧ εK ).

The general case follows from bilinearity. To show anticommutativity, note that
again this holds on basis elements:

εI ∧ εJ = εIJ = (−1)|I|·|J| εJI = (−1)|I|·|J| εJ ∧ εI .

The general case follows again from bilinearity.

Corollary 4.8. Let I = (i1 , ..., ik ) be an multi-index, then εi1 ∧ · · · ∧ εik = εI .
Proof. This is immediate from Theorem 4.6 and induction.
47 CHAPTER 4. DIFFERENTIAL FORMS

Definition 4.9. Let V n be a vector space and L : V → V be a linear map. Then

this induces a map L(v1 ∧ · · · ∧ vn ) = (Lv1 ) ∧ · · · ∧ (Lvn ).
P
Now, suppose that Lvi = j aji vj . Then we see that
X
L(v1 ∧ · · · ∧ vn ) = (aj11 · · · ajnn )(vj1 ∧ · · · vjn )
J

where J = (j1 , ..., jn ) runs over all n-tuples where each entry takes value between
1, ..., n. However, the previous theorem tells us that if J contains any repeated
indices, then vj1 ∧ · · · ∧ vjn = 0. Therefore we can in fact take J to run over all
permutations of (1, ..., n). Therefore the equation above actually becomes
X σ(1)
L(v1 ∧ · · · ∧ vn ) = (a1 · · · anσ(n) )(vσ(1) ∧ · · · vσ(n) ) (4.1)
σ∈Sn
X σ(1)
= sgn(σ)(a1 · · · anσ(n) )(v1 ∧ · · · ∧ vn ) (4.2)
σ∈Sn

= (det L)(v1 ∧ · · · ∧ vn ) (4.3)

Definition 4.10. Let V n be a vector space and let v1 , ..., vn and w1 , ..., wn be two
basis for V . Then these two basis are said to be equivalent if the linear map
A : V → V such that wj = Avj has positive determinant, i.e., det A > 0.

Corollary 4.11. The two basis v1 , ..., vn and w1 , ..., wn of V are equivalent iff

w1 ∧ · · · ∧ wn = λ(v1 ∧ · · · ∧ vn )

for some λ > 0.

4.2 Differential Forms

Now we generalize the notion of alternating k-tensors to alternating k-tensor fields
on manifolds, these are known as differential k-forms. Pointwise speaking, at every
point p ∈ M on a manifold, we take V = Tp M in the previous section. But we
need to define differential forms globally instead of just at every point. Therefore,
we first need to construct a bundle ∧k T ∗ M . This will be almost identical to the
construction of the tangent bundle and the cotangent bundle in section 3.2.
First, we define, on the level of sets, that
a
∧k T ∗ M = ∧k Tp∗ M.
p∈M

There is an obvious projection map π : (p, α) 7→ p where p ∈ M, α ∈ ∧k Tp∗ M . We

make this set into a topological space by defining O∧k T ∗ M = {π −1 (U ) : U ∈ OM }.
It remains to define an atlas on ∧k T ∗ M and check that it is smooth. We define

A∧k T ∗ M = {(π −1 (U ), ξφ ) : (U, φ) ∈ AM }

4.2. DIFFERENTIAL FORMS 48

where ξφ is defined as follows: Let p ∈ U and dx1 , ..., dxm be the induced basis for
Tp∗ M via the chart map φ. Then any α ∈ ∧k Tp∗ M can be written as a summand
X
α= ai1 ,...,ik dxi1 ∧ · · · ∧ dxik
1≤i1 <···<ik ≤m

m


where the summand contains k
terms (some could be zero) by Theorem 4.4 and
Corollary 4.8. Then define

ξφ : (p, α) 7→ (φ(p), aI ) ∈ Rm × R( k )
m

where the first m coordinates are the coordinates of φ(p), and the last m k
coor-
dinates are the coefficients ai1 ,...,ik . Clearly this map is invertible by definition. To
check that this is a smooth atlas, i.e., the transition maps are smooth, let (V, ψ) be
an overlapping chart with (U, φ) and let dy 1 , ..., dy m be the induced basis for Tp∗ M
via the chart map ψ. Then
X  i1   ik 
∂x ∂x
α= ai1 ,...,ik j1
dy j1
∧ ··· ∧ jk
dy jk

1≤i1 <···<ik ≤m
∂y ∂y
X  i1 
∂x ∂xik
= ai1 ,...,ik j
· · · j dy j1 ∧ · · · ∧ dy jk ,
1≤i <···<i ≤m
∂y 1 ∂y k
1 k

where the indices j1 , ..., jk are summed over using the summation convention. Since
∂xis /∂y js are smooth functions because φ, ψ are smoothly compatible charts, we see
that from definition ξψ ◦ ξφ−1 is smooth. This concludes the construction.
Definition 4.12. A section of ∧k T ∗ M is called a differential k-form. The integer
k is called the degree of the form. We say α is a smooth k-form is as a section
α : M → ∧k T ∗ M is a smooth map between manifolds. We denote the space of
smooth k-forms by Ωk (M ).
Proposition 4.13. A k-form α can be written, locally in a chart (U, φ), as a sum-
mand X
α= ai1 ,...,ik dxi1 ∧ · · · ∧ dxik .
1≤i1 <···<ik ≤m

Then α is smooth iff the functions ai1 ,...,ik ∈ C ∞ (U ), for every open set U in the
atlas AM .
Proof. First, assume α : M → ∧k T ∗ M is a smooth section, then for every chart
(U, φ) in the atlas, the component functions are just the components of the com-
position of smooth maps ξφ ◦ α ◦ φ−1 , which are of course smooth. The converse
implies the smoothness of α : M → ∧k T ∗ M as a section follows immediately from
Definition 1.10.
Corollary 4.14. If (U, φ, xi ) and (V, ψ, y i ) are two overlapping smooth coordinate
charts on M n , then the following identity holds on U ∩ V :
 i
∂y
dy ∧ · · · ∧ dy = det
1 n
dx1 ∧ · · · ∧ dxn = [J(ψ ◦ φ−1 ) ◦ φ]dx1 ∧ · · · ∧ dxn .
∂xj
Proof. This follows immediately from (4.3).
49 CHAPTER 4. DIFFERENTIAL FORMS

4.3 Orientation on Manifolds

Definition 4.15. A smooth atlas A of a smooth manifold M is called an orien-
tation atlas if the following holds true: For all (U, φ, xi ) and (V, ψ, y j ) in A such
that U ∩ V 6= ∅, we have
 i
−1 ∂y
J(ψ ◦ φ ) = det >0
∂xj

for all x ∈ U ∩ V. A maximal orientation atlas is called an orientation. A manifold

M is said to be orientable if an orientation exists. An oriented manifold is a
manifold with a chosen orientation.

Theorem 4.16. An n-dimensional manifold M is orientable iff M admits a nowhere

vanishing n-form ω.

Proof. Let ω be a nowhere vanishing n-form on M . Then for each chart (U, φ, xi )
there exists a nowhere vanishing function f 6= 0 such that ω = f dx1 ∧ · · · ∧ dxn .
We can always take such a chart near each point so that f > 0, otherwise we can
replace x1 by −x1 . Now suppose (Uα , φα ) and (Uβ , φβ ) are two such charts. Then
suppose ω = f dx1α ∧ · · · ∧ dxnα = gdx1β ∧ · · · ∧ dxnβ where f, g > 0. Then

0 < g = ω(∂1β , ..., ∂nβ ) = J(φβ ◦ φ−1 −1

α )ω(∂1 , ...∂α ) = J(φβ ◦ φα )f.
α n

So J(φβ ◦ φ−1
α ) = g/f > 0.
Conversely, let A be an orientation. For each local chart Uα , we consider the
form ωα = dx1α ∧ · · · ∧ dxnα . Pick a partition of unity ρα suboridnate to the open
cover Uα . Then consider the n-form
X
ω= ρα ωα .
α

P vanishing. For each p ∈ M , there

We claim that ω is nowhere P is a neighborhood U
of p such that the sum α ρα ωα reduces to a finite sum ki=1 ρi ωi . It follows that
near p, one has X
ω(∂11 , ..., ∂n1 ) = J(φi ◦ φ−1
1 )ρi > 0.
i

This shows that ω is nowhere vanishing.

With this theorem in mind, we can formulate the following (equivalent) defini-
tion:

Definition 4.17. We shall say M is orientable if it is possible to define a smooth

n-form ω on M which is nonvanishing, in which case M is said to be oriented by the
choice of ω in the following sense: Two nonvanishing n-forms ω and η on M induces
two orientations as shown by the previous theorem. We call those two orientations
equivalent iff ω = f η for some smooth function f that is positive everywhere. A
nowhere vanishing form on M is usually called an orientation form, for the
obvious reason.
4.4. PULLBACK AND EXTERIOR DIFFERENTIATION 50

Theorem 4.18. Every connected orientable smooth manifold has exactly two ori-
entations.

Proof. Let M be a connected smooth orientable manifold. By the previous theorem,

there exists a nonvanishing n-form ω on M . Any other choice of orientation will
induce a nonvanishing n-form η on M . Since ω = f η for some smooth function
f : M → R such that f (p) 6= 0 for all p ∈ M , then f (M ) ⊂ R is a connected subset
which does not contain 0, hence f is either always positive or always negative. If
f > 0, then ω and η induces the same orientation at Tp M for each p ∈ M , hence
they induce the same orientation. Therefore assume f < 0 and ω, η induces different
orientations on M . Then any other orientation will induce a non-vanishing n-form
τ such that τ = gω for some smooth function g ∈ C ∞ (M ). So g > 0 or g < 0, in
which case τ induces the same orientation as ω or η does.

4.4 Pullback and Exterior Differentiation

Let M n be a smooth manifold, and let Ωk (M ) denote the set of smooth k-form on
M , or equivalently Ωk (M ) = Γ(∧k T ∗ M ), which is the set of all smooth sections of
∧k T ∗ M.

Definition 4.19. Let F : M → N be a smooth map. Recall that dFp : Tp M → Tp N

is given by dFp (v)(f ) = v(f ◦ F ) for v ∈ Tp M, f ∈ C ∞ (N ). Then its dual map
Fp∗ : TF∗ (p) N → Tp∗ M is called the pullback of F at p. If ω ∈ Ωk (N ), then we can
also define the pullback F ∗ ω ∈ Ωk (M ) given by

(F ∗ ω)p (v1 , ..., vk ) = ωF (p) (dFp (v1 ), ..., dFp (vk )).

By definition, if ω ∈ Ω0 (N ) is a smooth function, then F ∗ ω := ω ◦ F .

Using this definition, we have the following easy observation:

Lemma 4.20. Suppose F : M → N is smooth. Then

(a) F ∗ : Ωk (N ) → Ωk (M ) is linear over R.

(b) F ∗ (ω ∧ η) = (F ∗ ω) ∧ F ∗ (η).

(c) In any smooth chart, we have that

!
X X
F∗ ωI dy i1 ∧ · · · ∧ dy ik = (ωI ◦ F )d(y i1 ◦ F ) ∧ · · · ∧ d(y ik ◦ F ).
I I

The exterior derivative d is a map d : Ωk (M ) → Ωk+1 (M ) for each k, con-

structed as follows: Let ω ∈ Ωk (M ). Then in a local chart (U, φ), we may write
X
ω= ωI dxI
I:i1 <...<ik
51 CHAPTER 4. DIFFERENTIAL FORMS

where ωI ’s are smooth functions U → R. Then we define

!
X X X ∂ωI
dω = dωI ∧ dxI = j
dxj ∧ dxI (4.4)
I:i <...<i I:i <...<i j
∂x
1 k 1 k
!
X X ∂ωi ◦ φ−1
1 ···ik
= dxj ∧ dxi1 ∧ · · · ∧ dxik .
i1 <...<ik j
∂xj
(4.5)
Since partial derivatives commutes, using the summation convention we see that
!
X ∂ωI
d(dω) = d j
dxj ∧ dxI
I,j
∂x
X ∂ 2 ωI
= j ∂xk
dxk ∧ dxj ∧ dxI
I,j,k
∂x
X ∂ 2 ωI X ∂ 2 ωI
= j k
dx ∧ dx ∧ dx +
k j I
j k
dxk ∧ dxj ∧ dxI
k<j
∂x ∂x k>j
∂x ∂x
X ∂ 2 ωI X ∂ 2 ωI
= j ∂xk
dx k
∧ dx j
∧ dx I
− j ∂xk
dxk ∧ dxj ∧ dxI = 0.
k<j
∂x k<j
∂x

Therefore, we see that d ◦ d : Ωk (M ) → Ωk+2 (M ) is identically the zero map, i.e.,

d ◦ d = 0. More frequently this is compactly written as d2 = 0.
Note that this definition (potentially) depends on the chart we choose. But it
turns out that this definition is independent of charts.
Proposition 4.21. Let d : Ωk (M ) → Ωk+1 (M ) be an operator that satisfies the
following properties:
(i) d is linear over R.
(ii) For any ω ∈ Ωk (M ), η ∈ Ωl (M ), we have d(ω ∧ η) = dω ∧ η + (−1)k ω ∧ dη.
(iii) d2 = 0.
(iv) For f ∈ Ω0 (M ) = C ∞ (M ), df is the differential of f , given by df (X) = Xf.
Then d exists and is unique. Moreover, in any smooth coordinate chart, d is given
by (4.4).
Proof. First, suppose M = Rn . Then for ω ∈ Ωk (Rn ) we define its exterior derivative
to be
! !
X X X X ∂ωJ
d ωJ dxJ = dωJ ∧ dxJ = i
dxi ∧ dxj1 ∧ · · · ∧ dxjk .
J J J i
∂x

Then one can check that d defined in Rn satisfies the properties (i)-(iv). Moreover,
one can check that if F : U → V is a smooth map in Euclidean space, then we have
F ∗ (dω) = d(F ∗ ω) (4.6)
4.4. PULLBACK AND EXTERIOR DIFFERENTIATION 52

the detailed calculation is in [1, Proposition 14.23]. Now for a general manifold M ,
we first show the existence of d. Let ω ∈ Ωk (M ), and we define d on chart (U, φ) by

dω = φ∗ d(φ−1∗ ω).

This is well-defined, since for any other chart (V, ψ), the map φ ◦ ψ −1 is a diffeomor-
phism. Hence by (4.6) we have

(φ ◦ ψ −1 )∗ d(φ−1∗ ω) = d((φ ◦ ψ −1 )∗ φ−1∗ ω) = d(ψ −1∗ ω).

Therefore, multiplying both sides by ψ ∗ , we get that φ∗ d(φ−1∗ ω) = ψ ∗ d(ψ −1∗ ω). It
follows quite easily from the calculation in Rn that this definition satisfies (i)-(iv).
Next, we prove uniqueness. Suppose d is any operator satisfying (i)-(iv). We
claim that if ω1 , ω2 ∈ Ωk (M ) such that ω1 = ω2 on an open set U ⊂ M , then
dω1 = dω2 on U . To see this, let p ∈ U and η = ω1 − ω2 . Let ψ ∈ C ∞ (M ) be a
bump function that is identically 1 on some neighborhood of p and supported in U .
Then ψη is identically zero. So (i)-(iv) implies that

0 = d(ψη) = dψ ∧ η + ψdη.

Evaluating this at p using ψ(p) = 1 and dψp = 0, we have that dω1 |p − dω2 |p =
dηp = 0.
P let ωI ∈ Ω (M ) be arbitrary, and (U, φ) be a smooth chart on M . Write
k
Now
ω = I ωI dx on U . For any p ∈ U , by means of a bump function we can construct
global smooth functions ω̂I , x̂i that agrees with ωI , xi in a neighborhood of p. By
virtual of (i)-(iv), we see that dω has to have the expression of (4.4) at p: More
detailed explanation can be found in [6, p.221]. Since p is arbitrary, this d must be
everywhere equal to the one we defined locally by (4.4).

Proposition 4.22. If F : M → N is a smooth map, then for each k, the pullback

map F ∗ : Ωk (N ) → Ωk (M ) commutes with d: F ∗ (dω) = d(F ∗ ω). In other words,
the diagram below commutes:

F∗
Ωk (N ) Ωk (M )
d d

Ωk+1 (N ) F∗
Ωk+1 (M )

Proof. We first claim that for any f ∈ C ∞ (N ), we have F ∗ (df ) = d(f ◦ F ). This is
a simple calculation:

Fp∗ (dfF (p) )(v) = dfF (p) (dFp (v)) = (dFp (v))(f ) = v(f ◦ F ) = d(f ◦ F )p (v).

Next, since d is local, for any ω ∈ Ωk (N ), we can restrict to local charts (V, ψ, xi )
of M and (U, φ, y j ) of N such that F (V ) ⊂ U . Let
X
ω= ωj1 ···jk dxj1 ∧ · · · ∧ dxjk .
J
53 CHAPTER 4. DIFFERENTIAL FORMS

Then X
dω = dωj1 ···jk ∧ dxj1 ∧ · · · ∧ dxjk .
J

Hence
X
F ∗ (dω) = F ∗ (dωj1 ···jk ) ∧ F ∗ (dxj1 ) ∧ · · · ∧ F ∗ (dxjk )
J
X
= d(ωj1 ···jk ◦ F ) ∧ d(xj1 ◦ F ) ∧ · · · ∧ d(xjk ◦ F ).
J

Next,
X
F ∗ω = (ωj1 ···jk ◦ F ) ∧ F ∗ dxj1 ∧ · · · ∧ F ∗ dxjk
J
X
= (ωj1 ···jk ◦ F ) ∧ d(xj1 ◦ F ) ∧ · · · ∧ d(xjk ◦ F ).
J

Therefore,
X
d(F ∗ ω) = d(ωj1 ···jk ◦ F ) ∧ d(xj1 ◦ F ) ∧ · · · ∧ d(xjk ◦ F ) = F ∗ (dω).
J

This completes the proof.

Theorem 4.23. Let F : N n → M n be a smooth map, and let ω ∈ Ωn (M ). Let
(U, φ, xi ) and (V, ψ, y j ) be two charts. Suppose ω = hdy 1 ∧ · · · ∧ dy n on (V, y j ).
Then the following holds on U ∩ F −1 (V ):

F ∗ ω = F ∗ (udy 1 ∧ · · · ∧ dy n ) = (u ◦ F )(det dF )dx1 ∧ · · · ∧ dxn .

Proof. We have shown in the previous proof that

F ∗ (udy 1 ∧ · · · ∧ dy n ) = (u ◦ F )d(y 1 ◦ F ) ∧ · · · ∧ d(y n ◦ F ).

Now how does the one form d(y j ◦ F ) relate to dxi ’s? By the way how 1-forms
transform (p.281 of Lee’s Smooth manifold), we have that

∂(y j ◦ ψ ◦ F ◦ φ−1 ) i ∂ F̂ j i
d(y j ◦ F ) = d(y j ◦ ψ ◦ F ) = dx = dx .
∂xi ∂xi
Therefore, we see that
!
1 n
∂ F̂ ∂ F̂
F ∗ (udy 1 ∧ · · · ∧ dy n ) = (u ◦ F ) · · · jn dxj1 ∧ · · · ∧ dxjn
∂xj1 ∂x
!
∂ F̂ j
= (u ◦ F ) det dx1 ∧ · · · ∧ dxn
∂xi
= (u ◦ F )(det dF )dx1 ∧ · · · ∧ dxn .

This completes the proof.

4.4. PULLBACK AND EXTERIOR DIFFERENTIATION 54

We conclude with a coordinate invariant formulation of exterior differentiation:

Proposition 4.24. Let ω ∈ Ω1 (M ). Then for smooth vector fields X, Y ∈ Γ(T M ),

dω(X, Y ) = Xω(Y ) − Y ω(X) − ω([X, Y ]). (4.7)

Proof. Since ω is local and linear and the commutator is multilinear, it P

suffices to
prove the statement for X = ∂/∂x , Y = ∂/∂x on a chart. Write ω = k fk dxk ,
i j

then
X X ∂fk
dω = dfk ∧ dxk = l
dxl ∧ dxk .
k k,l
∂x

Therefore,
  X ∂fk  
∂ ∂ ∂ ∂
dω i
, j = dx ∧ dx
l k
, j
∂x ∂x k,l
∂xl i
∂x ∂x
X ∂fk
= (δil δjk − δik δji )
k,l
∂xl
∂fj ∂fi
= i
− j.
∂x ∂x
One can check that this is equal to the right hand side of 4.7 applies to the vector
fields X = ∂i , Y = ∂j . This concludes the proof.
5 | Integration on Manifolds

5.1 Basic Definitions and Properties

Let M n be an oriented smooth manifold (i.e. orientable and has a chosen orienta-
tion). We want to define integral of an n-form on M . First, assume ω is a compactly
supported n-form, i.e., ω ∈ Ωnc (M ). Further assume that the support of ω is con-
tained in a single chart (U, φ) of M . Then we define the integral of ω over M
as Z Z
ω=± (φ−1 )∗ ω (5.1)
M φ(U )
where the expression on the right hand side is in Rn . And if φ is positively ori-
ented, i.e., the coordinate frame (∂/∂xi ) is a positively oriented basis for each tan-
gent space, then we choose the plus sign in the expression above. If φ is negatively
oriented, then we choose the minus sign in the expression above.
We should be more explicit about the definition above. Let η ∈ Ωn (D) where
D ⊂ Rn . Then using the standard coordinate (x1 , ..., xn ) of Rn , one can write
η = g(x1 , ..., xn )dx1 ∧ · · · ∧ dxn
for some continuous function g : D → R. Then we define the integral of η over
D to be
Z Z Z
η= g(x , ..., x )dx ∧ · · · ∧ dx :=
1 n 1 n
g(x1 , ..., xn )dx1 · · · dxn .
D D D

Then, we can now write (5.1) more explicitly. Suppose that on the chart (U, φ, xi ),
we have that
ω = f dx1 ∧ · · · ∧ dxn
for some smooth function f : Supp ω ⊂ M → R. Then the pullback formula gives
(recall that on manifolds, the coordinate function xi is really xi ◦ φ):
(φ−1 )∗ ω = (φ−1 )∗ (f dx1 ∧ · · · ∧ dxn )
= (f ◦ φ−1 )d(x1 ◦ φ ◦ φ−1 ) ∧ · · · ∧ d(xn ◦ φ ◦ φ−1 )
= (f ◦ φ)dx1 ∧ · · · ∧ dxn
where the last expression is a differential form in a subset of Rn , and xi ’s are coor-
dinate functions in Rn . Therefore, using the definition of integral in Rn above, we
can rewrite (5.1) as
Z Z Z
−1 ∗
ω=± (φ ) (f dx ∧ · · · ∧ dx ) := ±
1 n
(f ◦ φ−1 )dx1 · · · dxn .
M M φ(U )
5.1. BASIC DEFINITIONS AND PROPERTIES 56

One needs to check that this definition is indepent of charts chosen. First, recall
the change of variable formula in Rn :
Lemma 5.1. Let F : U → F (U ) be a diffeomorphism between subsets of Rn . Let
J(F ) = (∂F i /∂xj ) denote the Jacobian of F . Let (y j ) denote the coordinate systems
in F (U ) and let (xi ) denote that of U . Then we have, for an integrable function f ,
Z Z
f dy · · · dy = (f ◦ F )|J(F )|dx1 · · · dxn .
1 n
F (U ) U

Proposition 5.2. The definition of integral above does not depend on which chart
we choose.
Proof. We only need to check the case for positively oriented charts, since the other
cases will follow exact same calculation, except we will get a minus sign when com-
puting Jacobian determinants. So suppose (U, φ, xi ) and (V, ψ, y j ) are two intersect-
ing charts such that Supp ω ⊂ U ∩ V. And suppose
ω = f dx1 ∧ · · · ∧ dxn = hdy 1 ∧ · · · ∧ dy n .
Then the transformation for dxi into dy j gives the relation
ω = hdy 1 ∧ · · · ∧ dy n
 1 
∂y ∂y n
=h · · · in ◦ φdxi1 ∧ · · · ∧ dxin
∂xi1 ∂x
 i
∂y
= h det ◦ φdx1 ∧ · · · ∧ dxn
∂xj
where in the last step, we reorder every dxi1 ∧· · · dxin into dx1 ∧· · ·∧dxn , at a cost of a
sgn(σ) where σ(i1 , ..., in ) = (1, ..., n). But that with the partial derivatives combined
is exactly the determinant of the matrix (∂y i /∂xj ). Therefore, the expression above
gives  i
∂y
f = det ◦ φ.
∂xj
Now, since ψ ◦ φ−1 is a positively oriented chart, we have that J(ψ ◦ φ−1 ) > 0.
Therefore, we have that
Z Z
−1
(h ◦ ψ )dy · · · dy =
1 n
h ◦ ψ −1 ◦ (ψ ◦ φ−1 )|J(ψ ◦ φ−1 )|dx1 · · · dxn
ψ(V ) φ(U )
Z
= h ◦ ψ −1 ◦ (ψ ◦ φ−1 )J(ψ ◦ φ−1 )dx1 · · · dxn
φ(U )
Z   i 
∂y
= h · det j
◦ φ ◦ φ−1 dx1 · · · dxn
∂x
Zφ(U )
= (f ◦ φ−1 )dx1 · · · dxn .
φ(U )

Therefore, this definition is indeed independent of coordinate charts. Note that we

may assume (ψ ◦ φ−1 )(φ(U ∩ V )) = ψ(V ) since Supp ω is contained in U ∩ V , so
outside of ψ(U ∩ V ) the integration gives zero.
57 CHAPTER 5. INTEGRATION ON MANIFOLDS

Now suppose that ω ∈ Ωnc (M ). Let (Ui , φi ) be positive charts and let (ρi ) be a
partition of unity suboridnate to the open cover (Ui ) of M . Then we may write
X X
ω= ρi ω =: ωi .
i i

Since Supp ω is compact, it intersects finitely many open cover (Ui )’s. Hence we
really have a finite sum
Xm
ω= ωi .
i=1
Then we define the integral of ω over M to be
Z Xm Z
ω= ωi .
M i=1 M

Proposition 5.3. This definition is independent of which partition of unity we

choose.
Proof. Let
P (ηj ) be a partition of unity subordinate
P to another set of charts (Vj , ψj ).
So ω = ηj ω. Note that we can write ωρi = j (ωρi )ηj . Thus we can swap the
sums since they are finite. Therefore, we have that
!
XZ XXZ XZ X XZ
ωρi = ωρi ηj = ωηj ρi = ωηj .
i M i j M j M i j M

Hence this definition is indeed independent of the choice of partition of unity.

Some properties are immediate:
R R R
Lemma 5.4. M (aω +R bη) = a M ω + b M η. Also, if ω is a positively oriented
orientation form, then M ω > 0.
Proof. See [1], Proposition 16.6.
Theorem 5.5 (Integration and Orientations). Let M n be a smooth manifold and
ω ∈ Ωnc (M ).
(a) If −M denotes M with the opposite orientation, then
Z Z
ω=− ω.
−M M

(b) If F : N → M is a diffeomorphism, then

Z Z
ω=± F ∗ω
M N

where the ± depends on whether F is orientation preserving or reversing.

We say F is orientation preserving if for each p ∈ N , the isomorphism
F∗p : Tp N → Tp M takes positively oriented bases of Tp N to positively oriented
bases of Tp M , and say F is orientation reversing if it is not orientation
preserving.
5.1. BASIC DEFINITIONS AND PROPERTIES 58

Proof. (a) Let (U, φ, xi ) be a positively oriented chart of M containing Supp ω.

Then since −M has opposite orientation of M , this would be a negatively
oriented chart. A positively oriented chart on −M containing Supp ω would
be (U, ψ, y j ), where ψ = T ◦φ and T : Rn → Rn is defined as T (x1 , x2 , ..., xn ) =
(x2 , x1 , ..., xn ). Then suppose ω = f dx1 ∧ · · · ∧ dxn = hdy 1 ∧ · · · dy n . By the
previous calculation, we have that
 i    i
∂y ∂(ψ ◦ φ−1 )i ∂T
f = h · det j
◦ φ = h · det j
◦ φ = h · det ◦ φ = −h.
∂x ∂x ∂xj

Then by the change of variable formula, one sees that

Z Z
ω= (ψ −1 )∗ ω
−M
Zψ(U )
= (hdy 1 ∧ · · · dy n )
Zψ(U )
= (h ◦ ψ −1 )dy 1 · · · dy n
Zψ(U )
= (h ◦ ψ −1 ◦ ψ ◦ φ−1 )|J(ψ ◦ φ−1 )|dx1 · · · dxn
Zφ(U )
= (h ◦ φ−1 )dx1 · · · dxn
φ(U )
Z
=− (f ◦ φ−1 )dx1 · · · dxn
Zφ(U )
=− (φ−1 )∗ ω
φ(U )
Z
=− ω.
M

Note that we have used the fact that |J(ψ ◦ φ−1 )| = |J(T )| = 1. This proves
(a).

(b) Suppose F is orientation preserving, and (U, φ) is a positive chart of M

such that Supp ω ⊂ U . Then we see that (F ∗ ω)x (v1 , ..., vn ) 6= 0 implies
ωF (x) (F∗ (v1 ), ..., F∗ (vn )) 6= 0, which implies that ω 6= 0 at F (x). Hence
F (Supp F ∗ ω) ⊂ Supp ω ⊂ U . Therefore Supp F ∗ ω ⊂ F −1 (U ). Since F is
orientation preserving, then (F −1 (U ), φ ◦ F ) is a positively oriented chart of
N containing Supp F ∗ ω. Thus
Z Z
∗
F ω= [(φ ◦ F )−1 ]∗ F ∗ ω
N φ(U )
Z Z Z
−1 −1 ∗ −1 ∗
= (F ◦ F ◦ φ ) ω = (φ ) ω = ω.
φ(U ) φ(U ) M

If F is orientation reversing, then (F −1 (U ), φ ◦ F ) is a negatively oriented

chart of N containing Supp F ∗ ω. Hence by the definition of the integral, we
59 CHAPTER 5. INTEGRATION ON MANIFOLDS

have
Z Z Z Z
∗ −1 ∗ ∗ −1 ∗
F ω=− [(φ ◦ F ) ] F ω = − (φ ) ω = − ω
N φ(U ) φ(U ) M

from the previous calculation. This proves (b).

Remark 5.6. RThe previous Rtheorem says that if F : N → M is a diffeomorphism,

then we have N F ∗ ω = ± M ω. Using the pullback formula we obtained from
Theorem 4.23, we see that when N and M are Euclidean spaces, then this statement
is just a reformulation of the change of variable formula Lemma 5.1. This formulation
is more concise and elegant. However, we should not think that Lemma 5.1 as a
consequence of this statement. Remember that the whole theory of integration on
manifolds is built on integration on Rn at the first place. Also, recall that in the
proof of the previous theorem, we used the change of variable formula. It would
be useful, however, to use this more concise statement as an aid to remember the
change of variable formula in Rn .

5.2 Manifolds with Boundary and Boundary Ori-

entation
We denote the n-dimensional upper half-space H n = {(x1 , ..., xn ) ∈ Rn : xn ≥ 0}.
An n-dimensional manifold with boundary is a second-countable Hausdorff space
M in which every point has a neighborhood U homeomorphic to an open subset in
H n via a chart map φ : U → φ(U ) ⊂ H n . A point p ∈ M is called an interior
point of M if it is in the domain of some chart (U, φ) such that φ(U ) ∩ ∂H n = ∅.
We call such a chart an interior chart. We say p ∈ M is a boundary point if it
is in the domain of some boundary chart (V, ψ), i.e., a chart with ψ(V )∩∂H n 6= ∅.
The boundary ∂M of M consists of all boundary points of M .

Figure 5.1: A manifold with boundary. An interior point and a boundary point
are shown.
5.2. MANIFOLDS WITH BOUNDARY AND BOUNDARY ORIENTATION 60

A smooth structure on a manifold M with boundary is defined similarly to our

previous construction of smooth structure on a manifold. A map f : U → Rk where
U is a subset of H n is said to be smooth if it admits a smooth extension to some
open subset V ⊂ Rn that contains U . We define a smooth structure on M to be
a maximal smooth atlas, smooth in the sense that the chart transition maps are
smooth as defined above.
The following theorem is Theorem 5.11 of [1], whose proof is omitted here:
Theorem 5.7. If M is a smooth manifold with boundary, then with subspace topol-
ogy, ∂M is a topological (n − 1)-manifold without boundary, and has a smooth struc-
ture such that it is a properly embedded submanifold of M .
Definition 5.8. Let M be a manifold with boundary and let p ∈ ∂M . A vector
v ∈ Tp M is said to be inward-pointing if v ∈ / Tp (∂M ), and there exists some
ε > 0, a smooth curve γ : [0, ε] → M such that γ(0) = p and γ 0 (0) = v. A vector
v ∈ Tp M is said to be outward-pointing if −v is inward pointing.

Figure 5.2: An inward-pointing vector N ∈ Tp M .

Lemma 5.9. Suppose M is a smooth manifold with boundary,p ∈ ∂M , and (U, φ, xi )

is a chart containing p. Then the inward-pointing vectors v = v i ∂/∂xi in Tp M are
precisely those with v n > 0, the outward-pointing ones are those with v n < 0.
Proof. Let p ∈ ∂M . Suppose v ∈ Tp M is an inward-pointing vector. Then v ∈ /
Tp (∂M ) and there exists a curve γ : [0, ε] → M such that γ(0) = p, γ 0 (0) = v. Then
under the coordinate basis, we have that

∂ 0 d(φ ◦ γ)i ∂
v = vi i
= γ (0) = (0) i .
∂x dt ∂x
We claim that since v ∈ / Tp (∂M ), then v n 6= 0. In fact, we will show that v ∈ Tp (∂M )
iff v n = 0: Suppose v ∈ Tp (∂M ). Then there exists a curve η : [0, ε] → M such
that η ∈ ∂M and η 0 (0) = v. Then under the chart (U, φ), we see that a portion of
η([0, ε]) near p is mapped to ∂H n . Hence (φ ◦ η)n (t) = 0 for sufficiently small t by
definition. Therefore we see that v n = d(φ ◦ η)n /dt|t=0 = 0. Conversely, if v n = 0,
then consider η̂ : [0, ε] → H n defined by η̂ i (t) = v i (t) + φn (p). Then it is easy to
check that the curve η = φ−1 ◦ η̂ : [0, ε] → M satisfies η(0) = p, η 0 (0) = v. Since
η̂ ⊂ ∂H n , we have η ⊂ ∂M . Therefore v ∈ Tp (∂M ).
61 CHAPTER 5. INTEGRATION ON MANIFOLDS

Now that we argued since v ∈ / Tp (∂M ), we have v n 6= 0, then we must have

(φ ◦ γ) (t) > 0 for all t > 0, otherwise d(φ ◦ γ)n (t)/dt|t=0 = v n will be zero, which
n

is a contradiction. But then this shows that v n > 0 since

d(φ ◦ γ)n (t) (φ ◦ γ)n (t) − φn (p) (φ ◦ γ)n (t)
|t=0 = lim+ = lim+ > 0.
dt t→0 t t→0 t
Conversely, suppose that v = v i ∂/∂xi with v n > 0. Then by our argument above
v∈ / Tp (∂M ). Consider the curve γ̂ : [0, ε] → H n defined by γ̂ i (t) = v i t+φi (p). Then
it is easy to check that γ = φ−1 ◦ γ̂ : [0, ε] → M satisfies γ(0) = p, γ 0 (0) = v. Hence
v is inward-pointing. The outward-pointing statement follows immediately from
definition.

Theorem 5.10. If M is any smooth manifold with boundary, there is a smooth

outward pointing vector field along ∂M .

Proof. Cover a neighborhood of ∂M by smooth boundary charts (Uα , φα ). In each

such chart,
∂
να = − n
∂x ∂M ∩Uα
is a smooth vector field along ∂M ∩ Uα , and is outward-pointing by the previous
lemma. Let (ρα ) be a partition of unity subordinate to the cover (∂M ∩ Uα )α of
∂M , and define a global vector field ν along ∂M by
X
ν= ρα ν α .
α

Clearly ν is a smooth vector field along ∂M . To show that ν is outward-pointing,

let (y j ) be any smooth boundary coordinate in a neighborhood of p ∈ ∂M . Because
each να is outward pointing, it satisfies dy n (vα ) < 0. Hence
X
ν n = dy n (ν) = ρα dy n (να ) < 0.
α

This concludes the proof.

Now we define, for an oriented manifold M , its boundary orientation on
∂M . In non-rigorous terms, an orientation on ∂M is a consistent choice of orien-
tation at each Tp (∂M ). Let N ∈ Tp M be an outward pointing vector at ∂M . Let
(E1 , ..., En−1 ) be an ordered basis for Tp (∂M ). We define this basis to have positive
orientation iff (N, E1 , ..., En−1 ) is a positively oriented basis for Tp M .
More rigorously, let ω ∈ Ωn (M ), and let N be an outward-pointing vector field
on ∂M . Then define i∗∂M (N ¬ ω) ∈ Ωn−1 (∂M ) by

i∗∂M (N ¬ ω)p (v1 , ..., vn−1 ) = ω(di∂M (Np ), di∂M (v1 ), ..., di∂M (vn−1 ))

where i∂M : ∂M → M is the inclusion map, and v1 , ..., vn−1 ∈ Tp (∂M ).

Proposition 5.11. If ω is an orientation form on M , then i∗∂M (N ¬ ω) is an orien-

tation form on ∂M .
5.3. STOKES’ THEOREM 62

Proof. Denote σ = i∗∂M (N ¬ ω). Then σ ∈ Ωn−1 (∂M ). It will follow that σ is an
orientation form on ∂M if we can show it never vanishes. Given a basis (E1 , ..., En−1 )
of Tp (∂M ), the fact that Np ∈ / Tp (∂M ) implies (Np , E1 , ..., En−1 ) is a basis for Tp M .
Therefore
σp (E1 , ..., En−1 ) = ωp (Np , E1 , ..., En−1 ) 6= 0
since ωp is non-vanishing at every p, and if ωp is zero on a basis, then a multilinear-
argument would show that ωp ≡ 0. We also used the fact that i∗∂M is just restriction
to vectors tangent to ∂M . Since σp does not vanish and p is arbitrary, σ is an
orientation form. Moreover, since σ(E1 , ...En−1 ) > 0 iff ωp (Np , E1 , ..., En−1 ) > 0, the
orientation determined by σ is exactly as described in our informal statement.
It remains to show that this orientation is independent of the choice of N . Let
(x ) be coordinate of a boundary chart near p ∈ ∂M . If N and Ñ are two different
i

outward-pointing vector field on ∂M , then both (Np , ∂1 , ..., ∂n−1 ) and (Ñp , ∂1 , ..., ∂n−1 )
are bases for Tp M , and the transition matrix between them has determinant given
by N n (p)/Ñ n (p) > 0. Thus both bases determine the same orientation for Tp M .
Hence both N, Ñ determines the same orientation for Tp (∂M ).
Corollary 5.12. Let ∂H n−1 be equipped with the induced orientation when H n itself
has the standard orientation inherited from Rn . Then we can identify ∂H n with
Rn−1 . But the induced orientation agrees with the standard orientation on Rn−1
only when n is even. That is,
∂H n = (−1)n Rn−1 .
Proof. The standard orientation on Rn−1 is positively oriented for ∂H n iff the or-
dered basis [−∂n , ∂1 , ..., ∂n−1 ] is a positively oriented basis for Rn , where −∂n is an
outward-pointing vector field of H n . Note that
[−∂n , ∂1 , ..., ∂n−1 ] = −[∂n , ∂1 , ..., ∂n−1 ] = (−1)n [∂1 , ..., ∂n ].
This concludes the proof.

5.3 Stokes’ Theorem

Theorem 5.13 (Stokes’ Theorem). Let M n be an oriented manifold with boundary,
and ω ∈ Ωn−1
c (M ). Then let ∂M to have the induced boundary orientation. We then
have that Z Z
dω = i∗∂M ω,
M ∂M
where i∂M : ∂M → M is the inclusion map. More concisely, we can write
Z Z
dω = ω. (5.2)
M ∂M

Proof. First suppose M = H n . Since ω is of compact support, choose R > 0 such

that Supp ω ⊂ [−R, R]n−1 × [0, R] =: A. Write ω in standard coordinate as
X
n
ω= ci ∧ · · · ∧ dxn
ωi dx1 ∧ · · · ∧ dx
i=1
63 CHAPTER 5. INTEGRATION ON MANIFOLDS

where the hat means that dxi is omitted. Therefore,

X
n
dω = ci ∧ · · · ∧ dxn
dωi ∧ dx1 ∧ · · · ∧ dx
i=1
X n
∂ωi j
= ci ∧ · · · ∧ dxn
dx ∧ dx1 ∧ · · · ∧ dx
∂x j
i,j=1
X
n
∂ωi 1
= (−1)i−1 dx ∧ · · · ∧ dxn .
i=1
∂xi

We can change the order of integration in each term so as to do the xi integration

first. By the fundamental theorem of calculus, the terms for which i 6= n reduce to

X
n−1 Z R Z R Z R
∂ωi
(−1) i−1
··· (x)dx1 · · · dxn
i=1 0 −R −R ∂xi
X
n−1 Z RZ R Z R
∂ωi ci · · · dxn
= (−1) i−1
··· (x)dxi dx1 · · · dx
i=1 0 −R −R ∂xi
X
n−1 Z RZ R Z R
···
i
ci
(−1)i−1 =−R dx · · · dx · · · dx = 0,
[ωi (x)]xxi =R 1 n
=
i=1 0 −R −R

because we have chosen R large enough such that ω = 0 when xi = ±R. The only
term that might not be zero is the one for which i = n. For that term we have
Z Z R Z RZ R
∂ωn
dω = (−1) n−1
··· n
(x)dxn dx1 · · · dxn−1
Hn −R −R 0 ∂x
Z R Z R
n
= (−1)n−1 ··· =0 dx · · · dx
[ωn (x)]xxn =R 1 n−1
−R −R
Z R Z R


= (−1)n ··· ωn x1 , . . . , xn−1 , 0 dx1 · · · dxn−1
−R −R

because ωn = 0 when xn = R. To compare this to the other side of (5.2), we

compute as follows:
Z XZ

ω= ci ∧ · · · ∧ dxn .
ωi x1 , . . . , xn−1 , 0 dx1 ∧ · · · ∧ dx
∂H n i A∩∂H n

Because xn vanishes on ∂H n , the pullback of dxn to the boundary is identically zero.

Thus, the only nonzero term is the one for which i = n, which becomes
Z Z


ω= ωn x1 , . . . , xn−1 , 0 dx1 ∧ · · · ∧ dxn−1
∂H n A∩∂H nZ

= (−1)n ωn (x1 , ..., xn−1 , 0)dx1 · · · dxn−1

A∩∂H n
R
which is equal to Hn
dω. Note that in the last step we used Corollary 5.12.
5.4. APPLICATIONS OF STOKES’ THEOREM 64

Next we consider another special case: M = Rn . In this case, the support of ω

is contained in a cube of the form A = [−R, R]n . Exactly the same computation
goes through, except that in this case the i = n term vanishes like all the others, so
the left-hand side of (5.2) is zero. Since M has empty boundary in this case, the
right-hand side is zero as well.
Now let M be a smooth manifold with boundary, ω ∈ Ωn−1 c (M ) whose support
is contained in a single chart (U, φ). WLOG assume this chart is positively oriented.
The definition yields
Z Z Z Z
−1 ∗ −1∗
dω = (φ ) dω = d(φ ω) = (φ−1 )∗ ω
M Hn Hn ∂H n

from the calculation above, where ∂H n has the induced orientation. Since φ∗ takes
outward-pointing vectors on ∂M to outward-pointing vectors on H n , it follows that
R preserving diffeomorphism on φ(U ) ∩ ∂H , and the expres-
n
φ|U ∩∂M is an orientation
sion above is equal to ∂M ω. For an interior chart, we get the same computations
with H n replaced by Rn . This proves the theorem in this case.
Finally, let ω ∈ Ωn−1
c (M ) be arbitrary. Choose a cover of Supp ω by finitely
many domains of positively oriented smooth charts (Ui )i∈I , and choose a subordinate
smooth partition of unity (ρi )i∈I . We can apply the argument to ρi ω for each i and
obtain
Z XZ XZ XZ
ω= ρi ω = d (ρi ω) = dρi ∧ ω + ρi dω
∂M ∂M M M
Z
i
! i
Z !i Z
X X
= d ρi ∧ω+ ρi dω = 0 + dω
M i M i M

P
because i ψ1 ≡ 1. This completes the proof of Stokes’ theorem.

5.4 Applications of Stokes’ Theorem

In this section, we use Stokes’ theorem to deduce all the major theorems in elemen-
tary single and multivariable calculus (with probably more restrictive condition on
singularities). In fact, as we will see, these theorems are all just special cases of the
Stokes’ theorem (Theorem 5.2) that we obtained in the previous section, which is
what makes Stokes’ theorem so powerful and elegant. The computations in these
examples are very straightforward, and the more challenging part (which serves as
good practice) is to figure out the orientation and the boundary orientation.

Corollary 5.14 (The fundamental theorem of calculus). Let f : [a, b] → R be a

differentiable function, then we have
Z b
f 0 (x)dx = f (b) − f (a).
a

Proof. Take M = [a, b] in Stokes’ theorem. We define on orientation form on [a, b]

to be dx and a basis v of Tp [a, b] is positively oriented if dx(v) > 0 and negatively
65 CHAPTER 5. INTEGRATION ON MANIFOLDS

oriented otherwise. Notice that ∂[a, b] has only two points a, b. Hence the integral
degenerates into a sum. Since the outward-pointing normal at a is a negatively
oriented basis for Ta [a, b], we assign a minus sign to a. Since the outward-pointing
normal at b is a positively oriented basis for Tb [a, b], we assign a plus sign to b, i.e.,
∂[a, b] = b − a.
Now take ω = f (x)dx. Then by Stokes’ theorem,
Z Z Z b
f (b) − f (a) = ω= dω = f 0 (x)dx.
∂[a,b] [a,b] a

Corollary 5.15 (The fundamental theorem of line integral). Let f : R3 → R be a

smooth function, and let γ : [a, b] → R3 be a smooth nonsingular curve, i.e., γ is a
diffeomorphism onto its image. Then
Z b
∇f · γ 0 dλ = f (γ(b)) − f (γ(a)).
a

This version is slightly more restrictive since we do not allow singularity on γ.

Proof. We define the orientation form on γ([a, b]) to be (γ −1 )∗ (dx) where dx is the
orientation form on [a, b]. Therefore, similar to the previous argument, ∂γ([a, b]) =
γ(b) − γ(a). Now take the zero form ω = f ∈ Ω(γ([a, b])). We have
Z Z Z Z Z
∗ ∗ ∂f dγ i
dω = γ dω = d(γ ω) = d(f ◦ γ) = i
dλ
γ([a,b]) [a,b] [a,b] [a,b] [a,b] ∂x dλ
Z b
= ∇f · γ 0 dλ.
a

while Z
ω = f (γ(b)) − f (γ(a))
∂γ([a,b])
as discussed.
Corollary 5.16 (Green’s theorem). Let D ⊂ R2 be a region and ∂D is smooth.
If P, Q are functions defined on an open region containing D and have continuous
partial derivatives there, then
Z Z  
∂Q ∂P
P dx + Qdy = − dx ∧ dy.
∂D D ∂x ∂y
Proof. Choose ω = P dx + Qdy ∈ Ω1 (D), then the claim immediately follows from
Stokes’ theorem.
Corollary 5.17 (Divergence theorem). Let S ⊂ R3 be a region with smooth bound-
ary ∂S. Let F = (Fx , Fy , Fz ) : R3 → R3 be a continuously differentiable vector field.
Then Z Z
(∇ · F)dV = F · dS
S ∂S
where dV = dx∧dy ∧dz is the volume form on S, and dS = (dy ∧dz, dz ∧dx, dx∧dy)
is the area element. One can check that dS(v, w) = v × w for two vectors v, w in
R3 .
5.4. APPLICATIONS OF STOKES’ THEOREM 66

Proof. This is just a direct application of Stokes theorem with

ω = F · dS = Fx dy ∧ dz + Fy dz ∧ dx + Fz dx ∧ dy.

Simple calculation gives

 
∂Fx ∂Fy ∂Fz
dω = + + dx ∧ dy ∧ dz = (∇ · F )dV.
∂x ∂y ∂z

Corollary 5.18 (Classical Stokes’ theorem). Let Σ ⊂ R3 be an oriented surface

with smooth boundary ∂Σ (which is a curve). Let F = (Fx , Fy , Fz ) be a vector field
that is continously differentiable in a region containing S. Then
Z Z
(∇ × F) · dS = F · dl
Σ ∂Σ

where dS = (dy∧dz, dz ∧dx, dx∧dy) is again the area element, and dl = (dx, dy, dz)
is the curve element.

Proof. This again follows from straightforward calculation with suitable choice of
form ω = Fx dx + Fy dy + Fz dz.
 Part II

Cohomology and Its Applications

to Differential Topology
6 | Cohomology

6.1 Category Theory, de Rham Cohomology

In this chapter, we introduce a powerful tool that is used to study topology of
manifolds, called cohomology. The idea of differential and algebraic topology is to
associate, to each topological space X, some algebraic objects A(X). These objects
are usually groups or vector spaces. But we do not just want any algebraic objects,
we want some algebraic objects that are topological invariants in the following sense:
Suppose X, Y are topological spaces that are homeomorphic, then we
want A(X), A(Y ) to be isomorphic algebraic objects.
The motivation behind this request is simple: Since we are studying topology, and
the existence of a homeomorphism between X, Y means that these two objects are
indistinguishable from a topological point of view, then any information we obtain
from them, including algebraic information A(X), A(Y ), should be indistinguishable
in their respective sense. So now the goal of the game is clear: Given topological
spaces, we want to produce algebraic objects that are invariant under homeomor-
phisms. But this can be a difficult task. To accomplish this goal, we first need to
introduce our playgrounds for the game, categories:
Definition 6.1. A category C consists of the following data:
• A class of objects, which we denote by Ob(C);

• For each pair of objects X, Y ∈ Ob(C), a class of morphisms HomC (X, Y );

• For each triple X, Y, Z ∈ Ob(C), a composition map HomC (Y, Z)×HomC (X, Y ) →
HomC (X, Z), denoted (f, g) 7→ f ◦ g.
These data must satisfy the following axioms:
(C1) For every X ∈ Ob(C), there exists a morphism idX ∈ HomC (X, X), such that
f ◦ idX = f and idX ◦g = g whenever these compositions make sense.

(C2) Composition is associative. That is, (f ◦ g) ◦ h = f ◦ (g ◦ h) whenever these

compositions make sense.
Example 6.2. There are many examples of categories:
• Consider the category Set, where the objects are sets, and for A, B ∈ Ob(Set),
let HomSet (A, B) be the set of all maps from A to B.
69 CHAPTER 6. COHOMOLOGY

• Consider the category Top, where the objects are topological spaces, and for
X, Y ∈ Ob(Top), let HomTop (X, Y ) be the set of all continuous maps from
X to Y .

• Consider the category Vec, where the objects are vector spaces, and for V, W ∈
Ob(Vec), let HomVec (V, W ) be the set of all linear maps from V to W .

• Consider the category Grp, where the objects are groups, and for G, H ∈
Ob(Grp), let HomGrp (G, H) be the set of all group homomorphisms from G
to H.

• Consider the category Mfd, where the objects are smooth manifolds, and for
M, N ∈ Ob(Mfd), let HomMfd (M, N ) be the set of all smooth maps from M
to N .

Recall that when we first learned what a set is, we then went on to study functions
between sets; similarly, now we study functions between categories, called functors:

Definition 6.3. Let A, B be two categories. A (covariant) functor F : A → B

contains the following data:

• A map F : Ob(A) → Ob(B), denoted by A 7→ F (A).

• For every pair X, Y ∈ Ob(A), a map F : HomA (X, Y ) → HomB (F (X), F (Y )),
denoted by ϕ 7→ F (ϕ).

These data should be compatible with identity and composition in the following
sense:

1. We must have F (idX ) = idF (X) .

2. For every composable pair (f, g), we have F (f ◦ g) = F (f ) ◦ F (g). Note that
the composition f ◦ g is composition of morphisms in A, while F (f ) ◦ F (g) is
composition of morphisms in B.

One can visualize a functor F : A → B with the following picture: we have a

sequence of objects and morphisms

f g
X−
→Y −
→Z

in A. Then we imagine having a machine F that takes objects in A to objects in

B, and morphisms in A to morphisms in B. Applying this machine to the sequence
above gives a new sequence

F (f ) F (g)
F (X) −−→ F (Y ) −−→ F (Z)

of objects and morphisms in B.

6.1. CATEGORY THEORY, DE RHAM COHOMOLOGY 70

Example 6.4 (the tangent functor). The description above might seem very ab-
stract. But really this is just a fancier language that can be used to describe many
examples we have already seen. One might think of an example where, for any
smooth manifold M and a point p ∈ M , its tangent space Tp M is a vector space,
and for F : M → N smooth maps, we have a linear map dFp : Tp M → TF (p) N . And
what’s luck is that associativity is satisfied because d(G ◦ F )p = dGF (p) ◦ dFp . Does
this mean we have a functor Mfd → Vec? Almost, but not quite: This is base-point
dependent. However, we can introduce the category of pointed manifolds, de-
noted by Mfd∗ , where the objects are ordered pair (M, p), where M is a manifold
and p ∈ M is a point, and the morphisms are pointed maps F : (M, p) → (N, p0 ),
i.e., maps F : M → N with p0 = F (p). Then we get a functor Mfd∗ → Vec because
we can define the functor to take the sequence of pointed manifolds and morphisms

→ (N, p0 ) −
→ (Q, p00 )
F G
(M, p) −

to a sequence of vector spaces and linear maps

dFp dGp′
Tp M −−→ Tp′ N −−→ Tp′′ Q.

But note that given a smooth map F : M → N , we can define its global differen-
tial dF : T M → T N , via (p, X) 7→ (F (p), dFp (X)). This is a smooth map between
tangent bundles. So we get a functor Mfd → Mfd, by sending the sequence
F G
M−
→N −
→Q

to the sequence
dF dG
T M −→ T N −→ T Q.
Now we define what is called a contravariant functor. It almost just behaves like
a functor, but it reverses the direction of the morphism:
Definition 6.5. Let A, B be two categories. A contravariant functor F : A → B
contains the following data:
• A map F : Ob(A) → Ob(B), denoted by A 7→ F (A).

• For every pair X, Y ∈ Ob(A), a map F : HomA (X, Y ) → HomB (F (Y ), F (X)),

denoted by ϕ 7→ F (ϕ).
These data should be compatible with identity and composition in the following
sense:
1. We must have F (idX ) = idF (X) .

2. For every composable pair (f, g), we have F (f ◦ g) = F (g) ◦ F (f ).

Again, one should have the following picture in mind to visualize a contravariant
functor F : Suppose we have a sequence of objects and morphisms
f g
X−
→Y −
→Z
71 CHAPTER 6. COHOMOLOGY

in A. Then we imagine having a machine F that takes objects in A to objects in

B, and morphisms in A to morphisms in B, but with the direction of the arrow
reversed. Applying this machine to the sequence above gives a new sequence
F (g) F (f )
F (Z) −−→ F (Y ) −−→ F (X).
Example 6.6 (pullback functor). Note that for any smooth map F : M → N , the
pullback F ∗ : C ∞ (N ) → C ∞ (M ), f 7→ F ∗ (f ) = f ◦F defines a contravariant functor
Mfd → Vec by sending the sequence
F G
M−
→N −
→Q
to the sequence
G∗ F∗
C ∞ (Q) −→ C ∞ (N ) −→ C ∞ (M ).
But note that we can also define pullbacks of differential forms. Hence we get
another contravariant functor Mfd → Vec by sending the same sequence to
G∗ F∗
Ω∗ (Q) −→ Ω∗ (N ) −→ Ω∗ (M ).
But how can this abstract category theory nonsense help us achieve our goal
of constructing topological invariants? The insight is that instead of just thinking
about getting an algebraic object from each topological space, we should really be
thinking about constructing functors Top → Grp.
Proposition 6.7. Let A : Top → Grp be a functor between categories, either
covariant or contravariant. Then if X, Y are homeomorphic topological spaces, then
A(X), A(Y ) are isomorphic groups.
Proof. This is almost immediate from all definitions. Homeomorphic assumption
means that there exist continuous maps f : X → Y, g : Y → X such that f ◦ g = idY
and g ◦ f = idX . Then since A is a (covariant) functor, it respects composition,
meaning we get group homomorphisms A(f ) : A(X) → A(Y ) and A(g) : A(Y ) →
A(X), such that A(f ) ◦ A(g) = idA(Y ) and A(g) ◦ A(f ) = idA(X) . This is exactly the
definition of groups A(X), A(Y ) being isomorphic. The argument works similarly
for contravariant functors.
Remark 6.8. The argument above also works if we have a functor Top → Vec.
Then homeomorphic spaces gives isomorphic vector spaces (existence of linear iso-
morphisms).
From the example above, one might immediately think we have already accom-
plished our task: we already have a contravariant functor Ω∗ : Mfd → Vec. But
as it turns out, the vector space Ω∗ (M ) is too large to be useful: Just imagine zero
forms on R, i.e., Ω0 (R) = C ∞ (R). This is an infinite-dimensional real vector space.
But we have the right idea, we just need to reduce Ω∗ (M ) to something smaller.
This motivates the definition of cohomology groups.
In the previous chapter, we have defined exterior differentiation operator d :
Ω (M ) → Ωk+1 (M ), we recall that d2 = 0. This means that
k

im d|Ωk+1 (M ) ⊂ ker d|Ωk (M )

6.1. CATEGORY THEORY, DE RHAM COHOMOLOGY 72

for all k. Therefore, we have a sequence of cochain complex, i.e. a sequence of

abelian groups with ascending index with a sequence of maps d connecting them
such that d2 = 0:
d d d d d
Ω0 (M ) −
→ Ω1 (M ) −
→ Ω2 (M ) −
→ · · · Ωk (M ) −
→ Ωk+1 (M ) −
→ ···

Definition 6.9. We define the kth (de Rham) cohomology group with real
coefficients of M by the quotient of vector spaces (in particular it is a group under
vector addition)
ker d|Ωk (M ) Z k (M )
H k (M ) = = k .
im d|Ωk−1 (M ) B (M )
Therefore, this group is a quotient of closed k-forms Z k (M ) , i.e., dω = 0, by the
exact k-forms B k (M ), i.e., ω = dη for some η ∈ Ωk−1 (M ). Then
M
H(M ) := H k (M ).
0≤k≤dim M

is a ring with wedge product as multiplication operation.

Definition 6.10. Let M be a manifold and ω ∈ Ωk (M ). We define the support

Supp(ω) to be the closure of the set {p ∈ M : ω(p) 6= 0}. Denote Ωkc (M ) to be the
space of smooth k-forms with compact support. Note that

Ωkc (M ) ∼
= Cc∞ (M ) ⊗ Ωk (M ),

where Cc∞ (M ) = Ω0 (M ) is the set of smooth functions on M with compact support.

Define the kth cohomology group with compact support to be the quotient
vector spaces
ker d|Ωkc (M ) Z k (M )
H k (M ) = = ck .
im d|Ωk−1
c (M ) B c (M )
and M
Hc (M ) := Hck (M ).
0≤k≤dim M

In the remaining part of this chapter, we will first develop the basic theory of de
Rham cohomology. In the next chapter, we will see some beautiful applications of
cohomology and then we can appreciate how powerful it is: we can actually deduce
topological information about manifolds from cohomology.

Theorem 6.11. Given a smooth map F : M → N , the induced pullback map on

forms F ∗ : Ωk (N ) → Ωk (M ) induces a well-defined linear map

F ∗ : H k (N ) → H k (M ).

This makes H k into a contravariant functor Top → Grp (in fact it is also a con-
travariant functor Top → Vec).
73 CHAPTER 6. COHOMOLOGY

Proof. By Proposition 4.22, we know that the pullback commutes with d. This
means that if α ∈ Ωk (N ) is a closed form, i.e., dα = 0, then F ∗ α ∈ Ωk (M ) is also
closed, because d(F ∗ α) = F ∗ (dα) = 0. Therefore F ∗ (Z k (N )) ⊂ Z k (M ). Similarly,
if β ∈ Ωk (N ) is an exact form, i.e., β = dη, then F ∗ β = F ∗ (dη) = d(F ∗ η) ∈ Ωk (M )
is also exact, hence F ∗ (B k (N )) ⊂ B k (M ).
Now, we define F ∗ : H k (N ) → H k (M ) as follows: For every [α] ∈ H k (N ), it is
represented by some closed form α ∈ Ωk (N ). We define

F ∗ ([α]) = [F ∗ α] ∈ H k (M ).

We must check that F ∗ is well-defined. To this end, let α0 ∈ Ωk (N ) be another

(closed) representative of [α]. Then by definition, there exists some β ∈ Ωk−1 (N )
such that α − α0 = dβ. Then we see that

F ∗ [α] − F ∗ [α0 ] = [F ∗ α] − [F ∗ α0 ] = [F ∗ (α − α0 )] = [F ∗ (dβ)] = [d(F ∗ β)] = 0

because exact form are quotient to zero in H k (M ). Hence F ∗ : H k (N ) → H k (M )

is indeed well-defined. Clearly F ∗ is a linear map, because [α] + [β] := [α + β]
by definition and F ∗ : Ωk (N ) → Ωk (M ) is linear, which means it is also a group
homomorphism.
We now introduce some important concepts in topology.

Definition 6.12. Let M, N be smooth manifolds and F0 , F1 : M → N be two

smooth maps.

• A homotopy from F0 to F1 is a smooth map H : M × [0, 1] → N . Equiv-

alently, it is a family of smooth maps Ht : M → N , with t ∈ [0, 1], such that
H0 = F0 and H1 = F1 . If there exists a homotopy from F0 to F1 , we say that
F0 , F1 are homotopic, written F0 ' F1 .

• Let A ⊂ M be a subset. Then F0 , F1 are said to be homotopic relative to

A (or simply homotopic rel A), if there exists a homotopy H from F0 to F1 ,
with the property that H(p, t) = F0 (p) = F1 (p) for all p ∈ A. That is, Ht fixes
every point in A for all t.

• Two paths γ0 , γ1 : [0, 1] → M are said to be path-homotopic, if they are

homotopic rel {0, 1}. The notation is γ0 ' γ1 .

• A retraction from M onto a subspace A is a map r : M → M such that

r(M ) = A and r|A = idA .

• A deformation retraction of M onto a subspace A is a homotopy from

idM to a retraction of M onto A. If there exists a deformation retraction of M
onto A, we say M deformation retracts onto A, or A is a deformation
retract of M .

• A smooth map F : M → N is called a homotopy equivalence if there is a

smooth map G : N → M such that F ◦ G ' idN and G ◦ F ' idM . In this
case we say that M, N are homotopy equivalent.
6.1. CATEGORY THEORY, DE RHAM COHOMOLOGY 74

• A space X is said to be contractible if it is homotopy equivalent to a point. In

particular, if X deformation retracts to a point p ∈ X, then X is contractible.

Remark 6.13. Equivalently, a homotopy from F0 to F1 can be think of a map

H : M × R → N such that H(x, t) = F1 (x) for t ≥ 1 and F (x, t) = F0 (x) for
t ≤ 0, and for all x ∈ M . Sometimes one is more convenient than the other when it
comes to solving certain problems. We will use these two definitions interchangeably
throughout the text.

We have already seen that cohomology group H ∗ (M ) is a topological invariant

(i.e., invariant under homeomorphisms) simply because it is functorial. What is
surprising is that it is invariant under homotopy equivalence (note that homotopic
equivalence is weaker than being homeomorphic). We will prove this claim later,
but for now, let us conclude this section by proving homotopy invariance of integral
of a closed form along curves.

Theorem 6.14. Let γ0 , γ1 : [0, 1] → M be two (piecewise) differentiable paths that

are path-homotopic. Let ω ∈ Ω1 (M ) be a closed 1-form (i.e., dω = 0). Then
Z Z
ω= ω.
γ0 γ1

Proof. Here we will only prove the claim for differentiable paths. Piecewise differ-
entiable path can be handled similarly by doing each pieces at a time and adding
them up.
Let H : [0, 1] × [0, 1] → M be a path-homotopy from γ0 to γ1 . Denote I 2 :=
[0, 1] × [0, 1]. Since exterior derivative commutes with pullback by Theorem 4.22,
and the two vertical edges Γ2 , Γ3 of ∂I 2 are crushed into two points γ0 (0), γ0 (1) via
H (See Figure 6.1), Stokes’ theorem gives
Z Z Z
∗ ∗
H ω= d(H ω) = H ∗ (dω) = 0
∂I 2 I2 I2

since ω is closed. But on the other hand (See Figure 6.1 for the orientation of
integration),
Z 3 Z
X Z Z Z Z
∗ ∗ ∗ ∗
0= H ω= H ω= H ω− H ω= ω− ω.
∂I 2 i=0 Γi Γ1 Γ3 γ0 γ1

This is what we want to prove.

Example 6.15 (Winding number). Consider ω ∈ Ω1 (R2 − 0) defined by

1 xdy − ydx
ω := .
2π x2 + y 2
Simple calculation suggests that dω = 0. Hence ω is a closed form. For each integer
k, let Ck : [0, 1] → R2 − 0 be the curve defined by Ck (t) = (cos 2πkt, sin 2πkt), i.e.,
a circle that warps around the origin k times. One can calculate and verify that
75 CHAPTER 6. COHOMOLOGY

Γ1
M
γ1
H
Γ3 I2 Γ2
γ0

Γ0
Figure 6.1: The homotopy H maps the square I 2 to the loops enclosed by γ0 , γ1 ,
maps Γ0 to γ0 and Γ1 to γ1 , collapses Γ2 to γ0 (1) = γ1 (1), and collapses Γ3 to
γ0 (0) = γ1 (0).

R
ω = k. Hence, the form ω, integrated along Ck , detects the number of times the
Ck
curve Ck goes around the origin.
Now, consider an arbitrary R piecewise differentiable loop γ : [0, 1] → M with
γ(0) = γ(1) = (1, 0). What is γ ω? It is a result in algebraic topology that γ ' Ck
for some unique integer k (which is exactlyR the number of times γ warps around the
origin). Hence by Theorem 6.14, we have γ ω = k for that k. Therefore, ω is called
the winding number form, because it detects the number of times any γ winds
around the origin.
This example is also very instructive, because of the following reason: We know
that d2 = 0. If ω is exact, i.e., ω = dη for some η ∈ Ω0 (R2 − 0), then dω = d2 η = 0,
as expected. So there are two related questions one can ask:
• Does there exists η ∈ Ω0 (R2 − 0) such that ω = dη?

• Is it true that on R2 − 0, every closed 1-form is exact? This is equivalent to

asking whether H 1 (R2 − 0) = 0 is true.
The answers are no to both the questions above. For the first question, suppose
ω = dη for some 0-form η. Then just take k a nonzero integer and consider the loop
Ck . Since ∂Ck = 0 (because it is a loop), Stokes’ theorem gives a contradiction
Z Z Z
k= ω= dη = η = 0.
Ck Ck ∂Ck

Note that the use of Stokes’ theorem here is legit because η, although not necessarily
having compact support on R2 − 0, does have compact support on Ck because Ck
is compact. In fact, here we are really taking about i∗Ck η, where iCk : Ck → R2 − 0
is the inclusion map.
Therefore, immediately we know the answer to the second question is no, with
ω being a counterexample. In fact, we have H 1 (R2 − 0) ∼ = R, because R2 − 0
1 ∼
deformation retracts onto S and H (S ) = R, as we will see later.
1 1

Definition 6.16. A path-connected manifold M is said to be simply connected

if every path γ : [0, 1] → M is path-homotopic to the constant loop based at γ(0).
Proposition 6.17. Let M be a simply connected manifold. Then H 1 (M ) = 0.
6.2. COMPUTATIONS: FIRST INSTALLMENTS 76

Proof. This is immediate from the definition and Theorem 6.14.

Remark 6.18. From the previous proposition and Example 6.15, we can conclude
that R2 −0 is not simply connected. Indeed, the definition of being simply connected
means that one can continuously (even smoothly) shrink any loop γ to its basepoint
at γ(0). However, this clearly cannot be done for a loop in R2 − 0 containing
the missing origin, since there is no continuous shrinking possible because of the
missing hole. This is the reason why R2 − 0 is not simply connected, and the
reason why H 1 (R2 − 0) 6= 0. We have come to a very important observation which
shows the geometric meaning of cohomology H k (M ): the kth cohomology detects
k-dimensional holes in spaces. We will not go on and discuss why a missing point
is a 1-dimensional hole since it would be too much of a detour.

6.2 Computations: First Installments

Our next main goal is to develop results and tools to eventually calculate basic
cohomology groups, for example, H k (S n ), H k (Rn ), Hck (Rn ), and so on. The process,
although worth our effort, will be technical. For now, we dedicate this section to
computing some examples directly from definitions.

Example 6.19. If M is an m-dimensional manifold, by definition H k (M ) = 0 for

all k > m.

Example 6.20 (Simply connected manifold). We already learned in the previous

section that if M is simply connected, then H 1 (M ) = 0.
`
Example 6.21 (Cohomology of disjoint unions). Let M = Q α Mα is a countable
k ∼
unions of disjoint manifolds Mα . Then we claim that H (M ) = α H k (MQ α ). This is
because on the level of forms, we already have an isomorphism Ωk (M ) → α Ωk (M )
via ω 7→ (i∗α ω)α .. Hence, the corresponding cochain complex are isomorphic, which
gives isomorphic cohomology groups.
This shows that we only need to focus on computing cohomology groups of
connected manifolds.

Example 6.22 (Cohomology of degree 0). Let M be a connected manifold. Since

there are no (-1)-forms, hence B 0 (M ) = 0. Thus H 0 (M ) = Z 0 (M ). A closed 0-form
is the one such that df = (∂f /∂xi )dxi = 0 on a local chart. Hence f must be a locally
constant function. But on a connected topological space, locally constant functions
are globally constant by Lemma 0.19. Hence Z 0 (M ) ∼ = R. Thus H 0 (M ) ∼
= R.

Example 6.23 (Cohomology of zero-manifolds). Let M be a manifold of dimension

zero, i.e., a discrete set of n points. Then H 0 (M ) ∼
= Rn by the previous two examples.
Since dim Tp∗ M = 0 for every p ∈ M , we have Ωk (M ) = 0 for k > 0. Hence
H k (M ) = 0 for all k > 0. Similarly, we have Hc0 (R0 ) = R and Hck (R0 ) = 0 for k > 0.

Example 6.24. In this example we compute Hc∗ (R). Since there are no k-forms for
k > 1 on R, we have Hck (R) = 0 for k > 2. So we only need to compute Hc0 (R) and
Hc1 (R).
77 CHAPTER 6. COHOMOLOGY

Since there is no constant function on R with compact support, Hc0 (R) = 0. We

show that Hc1 (R) = R via first R isomorphism theorem: Let I : Ωc (R) → R denote
1

the integration map, I(ω) = R ω.

Clearly I is linear.
I is surjective since if we pick a positive function
R g ∈ Cc∞ (R), which amounts
to picking ω = g(x)dx ∈ Ω1c (R), then I(ω) = R ω < ∞. Thus, we can hit any real
number by considering I(λω), for arbitrary λ ∈ R.
We claim that ker I = Bc1 (R), the set of exact 1-forms with compact support on
R. If df ∈ Bc1 (R) for some f ∈ Cc∞ (R), then say Supp f ⊂ [a, b], then we see that
Z Z
df = df = f (b) − f (a) = 0 − 0 = 0.
R [a,b]

Thus Bc1 (R) ⊂ ker I. Conversely, if g(x)dx ∈ ker I ⊂ Ω1c (R), then it is easy to check
that Z x
f (x) = g(u)du ∈ Cc∞ (R)
−∞

and df = g(x)dx. Thus ker I ⊂ Bc1 (R). Hence we conclude that Bc1 (R) = ker I.
We also note that Zc1 (R) = Ω1c (R), i.e., every 1-form on R is closed. One inclusion
is obvious, and to see Ω1c (R) ⊂ Zc1 (R), let g(x)dx ∈ Ω1c (R). Then

d(g(x)dx) = g 0 (x)dx ∧ dx = 0.

Finally, by the first isomorphism theorem, we know that

Ω1c (R) Z 1 (R)

R∼
= = c1 = Hc1 (R).
ker I Bc (R)

As a final remark, note that I : Ω1c (R) → R descends to a map I : Hc1 (R) → R, still
defined by literal integration. The result ker I = Bc1 (R) suggests that the descended
integration map I : Hc1 (R) → R is an isomorphism.

Example 6.25. As a generalization to the example above, we have Hck (Rn ) = R if

k = n and H k (Rn ) = 0 otherwise. This is called the Poincaré lemma for cohomology
with compact support, which we will prove later.

6.3 Some Homological Algebra

Although it is possible to compute the cohomology groups of many spaces directly
from definition, as we did in the previous section, it should be apparent now that this
is no easy task. Fortunately, there is an algebraic tool, called the exact sequence,
that makes computations of unknown cohomology of spaces using known spaces
significantly easier, and in fact almost like a dull procedure. However, such a tool
does not come for free. We have to develop some technical algebraic machinery to
be able to use it, and this is the goal of this and the next section.
6.3. SOME HOMOLOGICAL ALGEBRA 78

Definition 6.26. A (cochain) complex is a sequence of vector spaces (or abelian

groups) and linear maps (group homomorphisms):
dq−1 dq
· · · → Aq−1 −−→ Aq −
→ Aq+1 → · · ·

such that dq ◦ dq−1 = 0 for all q. We denote a chain complex like this by A∗ .

We have already seen an example of a chain complex, that is Ω∗ (M ) with the

exterior derivative operator. The requirement that dq ◦ dq−1 = 0, or simple d2 = 0, is
to ensure that im dq−1 ⊂ ker dq for every q, hence the quotient space ker dq / im dq−1
is well-defined.

Definition 6.27. We define the kth cohomology of the cochain complex A∗ by

H k (A) := ker dq / im dq−1 .

Definition 6.28. Let A∗ , B ∗ be cochain complexes. A cochain map F : A∗ →

B ∗ is a collection of linear maps F q : Aq → B q such that the following diagram
commutes for each q:

dqA
··· Aq Aq+1 ···
Fq F q+1

··· Bq B q+1 ···

dqB

Remark 6.29 (Cochain maps induces maps on cohomology groups). An example of

cochain map is the pullback: Let F : M → N be a smooth map, then we get a chain
map defined by pullback F ∗ : Ωq (N ) → Ωq (M ). We know that dF ∗ = F ∗ d from
Proposition 4.22. Hence F ∗ is a cochain map. In exactly the same way we prove
Theorem 6.11, one can show that a chain map F : A∗ → B ∗ induces a morphism on
cohomology groups F q : H q (A) → H q (B) for each q.

Definition 6.30. A sequence

dq−1 dq
· · · → Aq−1 −−→ Aq −
→ Aq+1 → · · ·

is said to be exact at Aq if one has ker dq = im dq−1 . A sequence is said to be

exact if it is exact at every Aq . Note that for a cochain complex, we only require
dq ◦ dq−1 = 0, i.e., im dq−1 ⊂ ker dq . So an exact sequence is a more restrictive type
of cochain complex.

Remark 6.31. It should be clear from definition that for an exact sequence A∗ , one
has H k (A∗ ) = 0 for all k.

The following lemma is immediate from definition:

Lemma 6.32. Let A, B, C be vector spaces (or abelian groups).

f
1. The sequence 0 → A −
→ B is exact iff f is injective.
79 CHAPTER 6. COHOMOLOGY

g
2. The sequence B −
→ C → 0 is exact iff g is surjective.
f g
3. The sequence 0 → A −→B− → C → 0 is exact iff f is injective, g is surjective,
and im f = ker g. Such a sequence is called a short exact sequence.

Theorem 6.33 (Snake lemma). Let f : A∗ → B ∗ , g : B ∗ → C ∗ are cochain maps

such that for each q, the maps

fq gq
0 → Aq −→ B q −
→ Cq → 0

is a short exact sequence. That is, we have the following big commutative diagrams
with exact rows:

.. .. ..
. . .
dq+1
A dq+1
B dq+1
C

f q+1 g q+1
0 Aq+1 B q+1 C q+1 0
dqA dqB dqC
fq gq
0 Aq Bq Cq 0
dq−1
A dq−1
B dq−1
C

f q−1 g q−1
0 Aq−1 B q−1 C q−1 0

.. .. ..
. . .

then this induces a (long) exact sequence of cohomology groups:

f∗ g∗ d∗
· · · → H q (A) −→ H q (B) −
→ H q (C) −
→ H q+1 (A) → · · ·

Proof Sketch. Detailed proof (e.g., check exactness) can be found in [3], though quite
unreadable. We only define the maps f ∗ , g ∗ , d∗ here and leave the rest to the readers
who are interested. Here f ∗ : H q (A) → H q (B) is induced by the map f q : Aq → B q
and similarly g ∗ : H q (B) → H q (C) is induced by g q : B q → C q , according to Remark
6.29. We briefly review the construction of the map d∗ : H q (C) → H q+1 (A): Let
c ∈ C q with dqC (c) = 0, i.e., c represents a class in H q (C), then since g q is surjective,
there exists b ∈ B q such that g q (b) = c. Now

g q+1 dqB (b) = dqC g q (b) = dqC (c) = 0.

Therefore dqB (b) ∈ ker g q+1 = im f q+1 . Thus there exists a ∈ Aq+1 such that dqB (b) =
f q+1 (a). Thus we define
d∗ [c] = [a].
6.4. THE MAYER-VIETORIS SEQUENCE 80

Theorem 6.34 (The Five Lemma). Given a commutative diagram of Abelian groups
and group homomorphisms
f1 f2 f3 f4
··· A B C D E ···
α β γ δ ε

··· A0 f1′
B0 f2′
C0 f3′
D0 f4′
E0 ···

in which the rows are exact, if the maps α, β, δ and ε are isomorphisms, then so is
the middle one γ.
Proof. First, we prove that γ is injective. Suppose for c ∈ C we have γ(c) = 0.
Then f30 (γ(c)) = 0, and by commutativity of the diagram δ(f3 (c)) = 0. Since δ is
an isomorphism, f3 (c) = 0, and thus c ∈ ker f3 = im f2 . It follows that there exists
b ∈ B such that f2 (b) = c. Now, γ(f2 (b)) = γ(c) = 0, and by commutativity of
the diagram f20 (β(b)) = 0. So, β(b) ∈ ker f20 = im f10 , and thus there exists a0 ∈ A0
such that β(b) = f10 (a0 ). Since α is an isomorphism, there exists a ∈ A such that
α(a) = a0 . We then have β(b) = f10 (α(a)) = β(f1 (a)) by commutativity, and since
β is an isomorphism we have b = f1 (a). Now, we see that c = f2 (f1 (a)) = 0 by
exactness, so γ is indeed surjective.
Now, we prove that γ is surjective. Let c0 ∈ C 0 be arbitrary. By exact-
ness f40 (f30 (c0 )) = 0. Since δ, ε are isomorphisms, f40 = ε ◦ f4 ◦ δ −1 , so we have
ε(f4 (δ −1 (f30 (c0 )))) = 0. Since ε is an isomorphism, f4 (δ −1 (f30 (c0 ))) = 0, and thus
δ −1 (f30 (c0 )) ∈ ker f4 = im f3 , and thus there exists c ∈ C such that δ −1 (f30 (c0 )) =
f3 (c). Now, consider γ(c). We have f30 (γ(c)) = δ(f3 (c)) = f30 (c0 ), and thus c0 −γ(c) ∈
ker f30 = im f20 . Therefore, there exists b0 ∈ B 0 such that c0 − γ(c) = f20 (b0 ). Since β
is an isomorphism, there exists b ∈ B so that b0 = β(b). It follows by commutativity
of the diagram that c0 − γ(c) = f20 (β(b)) = γ(f2 (b)), and thus c0 = γ(c + f2 (b)). This
proves surjectivity, and we are done.

6.4 the Mayer-Vietoris Sequence

Suppose M = U ∪ V where U, V are two open sets, then we have a sequence of
inclusions
iU `
U ∩V U V i
M.
iV
` `
Here, the map U V → M is ` just the obvious map, sending (u, U ) ∈ U ⊂ U V
to u ∈ M and (v, V ) ∈ V ⊂ U V to v ∈ M . Applying the contravariant functor
Ω∗ gives a sequence
i∗U
∗ i∗ ∗ ∗
Ω (M ) Ω (U ) ⊕ Ω (V ) Ω∗ (U ∩ V ).
i∗V

Here, the first map is given by i∗ (ω) = (i∗U ω, i∗V ω). Then we obtain the Mayer-
Vietoris sequence
i∗ φ
0 → Ω∗ (M ) −
→ Ω∗ (U ) ⊕ Ω∗ (V ) −
→ Ω∗ (U ∩ V ) → 0
81 CHAPTER 6. COHOMOLOGY

where the map φ : Ω∗ (U ) ⊕ Ω∗ (V ) → Ω∗ (U ∩ V ) is given by

φ(ω, τ ) = i∗U ∩V τ − i∗U ∩V ω = τ − ω.
Proposition 6.35. The Mayer-Vietoris sequence is exact.
Proof. We first show i∗ is injective: If i∗ (ω) = 0, then it means i∗U ω = i∗V ω = 0.
Thus ω ≡ 0 on U and on V , so it is zero on the entire M = U ∪ V .
Now we show ker φ = im i∗ . Suppose (ω, τ ) ∈ ker φ, then τ − ω ≡ 0 on U ∩ V .
Thus τ = ω on U ∩ V . Therefore we may define η ∈ Ω∗ (M ) by setting η = ω on U
and η = τ on V . Then i∗ (η) = (ω, τ ).
Finally we show φ is surjective. Let ω ∈ Ω∗ (U ∩ V ). Let ρU , ρV be a partition of
unity subordinate to U, V . Then we claim that φ(−ρV ω, ρU ω) = ω. Indeed,
φ(−ρV ω, ρU ω) = ρU ω − (−ρV ω) = (ρU + ρV )ω = ω.
Lastly, notice that (−ρV ω, ρU ω) ∈ Ω∗ (U ) ⊕ Ω∗ (V ).
By snake lemma (Theorem 6.33), the Mayer-Vietoris sequence above induces the
following long exact sequence (which is also called Mayer-Vietoris sequence):
d∗
· · · H q (M ) → H q (U )⊕H q (V ) → H q (U ∩V ) −
→ H q+1 (M ) → H q+1 (U )⊕H q+1 (V ) → · · ·
Remark 6.36 (Explicit description of d∗ ). Recall how we diagram chase the snake
lemma in the proof sketch of Theorem 6.33: Let ω ∈ Ωq (U ∩V ) be such that dω = 0.
Then by exactness of the row there is ξ ∈ Ωq (U ) ⊕ Ωq (V ) such that φ(ξ) = ω. One
can see that
ξ = (−ρV ω, ρU ω).
Since the snake diagram commutes and dω = 0, we know dξ = 0 on Ωq+1 (U ∩ V ).
That is, −d(ρV ω) ≡ d(ρU ω) on U ∩ V . Hence dξ is the image of an element in
Ωq+1 (M ) which is closed and well-defined. Explicitly one sees that
(
[−d(ρV ω)] on U
d∗ [ω] =
[d(ρU ω)] on V.

6.4.1 Mayer-Vietoris Sequence for Compact Supports

If j : U → M is the inclusion of an open subset, then we define the induced map
j∗ : Ω∗c (U ) → Ω∗c (M ) to be the map which extends a form on U by zero to a form
on M . Then Ω∗c is a covariant functor under inclusion of open sets.
The sequence of inclusions
iU `
U ∩V U V i
M.
iV

again gives rise to a sequence of forms with compact support

0 → Ω∗c (U ∩ V ) −
→ Ω∗c (U ) ⊕ Ω∗c (V ) −
→ Ω∗c (M ) → 0
δ +

where
δ(ω) = (−(jU )∗ ω, (jV )∗ ω).
6.4. THE MAYER-VIETORIS SEQUENCE 82

Proposition 6.37. The Mayer-Vietoris sequence for compact support above is ex-
act.

Proof. The map δ is injective because if δ(ω) = (−(jU )∗ ω, (jV )∗ ω) = (0, 0), then
(jU )∗ ω = 0, so ω ≡ 0 on U , in particular on U ∩ V .
Clearly, we have +(−(jU )∗ ω, (jV )∗ ω) = −(jU )∗ ω + (jV )∗ ω = 0. Thus im δ ⊂
ker +. Conversely, let (ω, τ ) ∈ Ω∗c (U ) ⊕ Ω∗c (V ) such that ω + τ = 0 on M . Then
ω = −τ on U ∩ V . Thus if we define η = ω on U ∩ V , then we have δ(η) = (ω, τ ) =
(−τ, τ ) = (−j∗ ω, j∗ ω). So ker + ⊂ im δ.
Finally, let ω ∈ Ω∗c (M ). Then ω = +(ρU ω, ρV ω). Moreover, since

Supp(ρU ω) = Supp ρU ∩ Supp ω

which is a closed set (Supp ρU ) intersected with a compact set (Supp ω), which is
compact in a Hausdorff space, we have ρU ω ∈ Ω∗c (U ), and similarly ρV ω ∈ Ω∗c (V ).
So + is surjective. This concludes the proof.
Again by snake lemma (Theorem 6.33), we get a long exact sequence, also called
the Mayer-Vietoris sequence for compact supports, from the short exact sequence
of cochain complexes above:
∗ d
· · · → Hcq (U ∩ V ) → Hcq (U ) ⊕ Hcq (V ) → Hcq (M ) −
→ Hcq+1 (U ∩ V ) → · · ·

Remark 6.38 (Explicit description of d∗ ). Just like what we did in the ordinary
MV sequence, we give an explicit description of d∗ here. Let τ ∈ Ωqc (M ) be a closed
form representing [τ ] ∈ Hcq (M ). By exactness of the MV sequence for Ωqc ’s, there
exists an element in Ωqc (U ) ⊕ Ωqc (V ) that sums to τ , namely (ρU τ, ρV τ ). Note that
ρU τ, ρV τ indeed have compact supports (why?). Then under exterior differentiation
c (U ) ⊕ Ωc (V ). Since dτ = 0 in Ωc (U ∩ V ) by
it goes to (d(ρU τ ), d(ρV τ )) ∈ Ωq+1 q+1 q+1

assumption, exactness of the MV sequence for Ωq+1 c ’s implies there is an element

d∗ τ ∈ Ωc (U ∩ V ), such that
q+1

(−(jU )∗ (d∗ τ ), (jV )∗ (d∗ τ )) = (d(ρU τ ), d(ρV τ )).

6.4.2 Cohomology of S 1
We now calculate H ∗ (S 1 ), H ∗ (S 1 ) using the Mayer-Vietoris sequence. We do the
regular cohomology first.
By Example 6.22, and the fact that S 1 is connected, we see that H 0 (S 1 ) = R. By
Example 6.19, we know that H k (S 1 ) = 0 for k > 1. Hence it remains to calculate
H 1 (S 1 ). We cover S 1 by two open sets U, V as suggested by Figure 6.2:
Now the Mayer-Vietoris sequence reads
δ d∗
H 0 (S 1 ) → H 0 (U ) ⊕ H 0 (V ) −
→ H 0 (U ∩ V ) −
→ H 1 (S 1 ) → H 1 (U ) ⊕ H 1 (V ) → · · ·
`
Since U, V ∼ = R, U ∩ V ∼= R R, using H 0 (R) = R, H 1 (R) = 0, we get that the MV
sequence reads:
δ d∗
R→R⊕R− →R⊕R− → H 1 (S 1 ) → 0.
83 CHAPTER 6. COHOMOLOGY

U S1

V
Figure 6.2: Choose U to be an open subset of S 1 that contains the upper hemi-
sphere, and V an open subset that contains the lower hemisphere, such that U ∩ V
is homeomorphic to two disjoint copies of intervals.

Here, note that δ is induced by the map φ : Ω0 (U ) ⊕ Ω0 (V ) → Ω0 (U ∩ V ), where

φ(ω, τ ) = τ − ω. Thus δ is the map δ : (ω, τ ) 7→ (τ − ω, τ − ω). Hence dim im δ = 1.
By exactness of the sequence, we know d∗ is surjective. Hence

R⊕R R⊕R ∼
H 1 (S 1 ) = ∗
= = R.
ker d im δ

Remark 6.39 (A generator for H 1 (S 1 )). We now find an explicit generator of the
group H 1 (S 1 ) ∼
= R. Since H 1 (S 1 ) = R2 / ker d∗ , anything that is not killed by d∗ in
H 0 (U ∩ V ) will represent a generator. Hence, let α ∈ Ω0 (U ∩ V ) be a closed 0-form
(i.e., locally constant function on U ∩ V ) that is not in im δ = ker d∗ , then d∗ α will
represent a generator of H 1 (S 1 ). For example, if we take α to be a locally constant
function that is 1 on one connected piece of U ∩ V , and 0 on the other, then it will
represent the element (1, 0) ∈ H 0 (U ∩ V ), which is clearly not in im δ.
Now by Remark 6.36, we have
(
−d(ρV α) on U
d∗ α =
d(ρU α) on V

where ρU , ρV is a partition of unity subordinate to U, V . Note that this form d∗ α is

well-defined, since on the intersection

−d(ρV α) − d(ρU α) = −d((ρU + ρV )α) = −dα = 0

since α is closed by assumpution.

Hence we see that 


R k = 0
H (S ) = R k = 1
k 1


0 k ≥ 2.
Next, we compute Hc∗ (S 1 ). Obviously, since S 1 is compact, Hc∗ (S 1 ) = H ∗ (S 1 ), so
we already have the answer above. But let us compute it using the MV sequence as
an illustration. Again Hck (S 1 ) = 0 for k ≥ 2, so we only worry about the zeroth and
6.5. HOMOTOPY INVARIANCE AND APPLICATIONS 84

first cohomology groups with compact support. We still choose U, V as in Figure

6.2. Then the MV sequence reads

Hc0 (U ∩ V ) → Hc0 (U ) ⊕ Hc0 (V ) → Hc0 (S 1 ) →

δ
→ Hc1 (U ) ⊕ Hc1 (V ) → H 1 (S 1 ) → 0.
→ Hc1 (U ∩ V ) −
`
By Example 6.24 and Since U, V ∼ = R, U ∩V ∼= R R, we see that the first two terms
of the sequence are zero, and Hc1 (U ∩ V ) ∼
= R ⊕ R. Recall that the map δ is induced
by the extension map Ω1c (U ∩ V ) → Ω1c (U ) ⊕ Ω1c (V ), where α 7→ (−(jU )∗ α, (jV )∗ α).
Hence the map δ is given by

δ : Hc1 (U ∩ V ) → Hc1 (U ) ⊕ Hc1 (V )

ω 7→ (−(jU )∗ ω, (jV )∗ ω).

Proposition 6.40. dim im δ = 1.

Proof. By Example 6.24, for any open set WR homeomorphic to R, we have Hc1 (W ) ∼ =
R, with the isomorphism defined by α 7→ W α. In other words, while making R R the
identification H 1 (U ∩V ) ∼ = R2 , we are really using the isomorphism ω 7→ ( A ω, B ω),
where A, B are the two connected components
R Rof U ∩ V . Now Rthe image
R δ(ω) =
(−(jU )∗ ω, (jV )∗ ω) is identified with ( U −(jU )∗ ω, V (jV )∗ ω) = (− U ∩V ω, U ∩V ω). in
R2 . Now putting all the identification together, and viewing δ as a map R2 → R2 ,
the map is actually
δ : (λ, µ) 7→ (−λ − µ, λ + µ)
which makes it clear that dim im δ = 1.

By exactness, we see that the map ϕ : Hc1 (S 1 ) → Hc1 (U ∩ V ) in the MV sequence

is an injection, since the previous term in the sequence is 0. Hence

Hc0 (S 1 ) ∼
= im ϕ = ker δ ∼
=R

by the previous proposition and the rank-nullity theorem. And since Hc1 (U ) ⊕
Hc1 (V ) → H 1 (S 1 ) is a surjection by exactness,

R⊕R ∼
Hc1 (S 1 ) = = im δ ∼
=R
ker δ
by the 1st isomorphism theorem and the previous proposition.

6.5 Homotopy Invariance and Applications

We now prove that de Rham cohomology is a homotopy invariant, and use this
result to compute H ∗ (Rn ), H ∗ (S n ), and some other examples.

Theorem 6.41. Homotopic maps induce the same map in cohomology.

85 CHAPTER 6. COHOMOLOGY

Proof. Let f, g : M → N be homotopic. By Remark 6.13, there exists H : M × R →

N such that H(x, t) = f (x) for t ≥ 1 and H(x, t) = g(t) for t ≤ 0. Equivalently let
s0 , s1 : M → M × R be the 0-section and 1-section, then f = H ◦ s1 and g = H ◦ s0 .
Then f ∗ = s∗1 ◦ H ∗ and g ∗ = s∗0 ◦ H ∗ .
Now, let xi : M × R → R with i = 1, ..., n + 1, and xn+1 = t. Then we see that
xi ◦ s : M → M × R → R is given in local coordinate by

xi ◦ s0 (x1 , ..., xn ) = xi (x1 , ..., xn , 0) = xi

for i 6= n + 1 and 0 when i = n + 1. Thus s∗0 (dxi ) = d(xi ◦ s0 ) = dxi when i 6= n + 1

and s∗0 (dt) = 0. Similarly we see s∗1 is given by the same expression. Thus s∗0 = s∗1 .
Thus f ∗ = g ∗ .
Corollary 6.42. If M ' N , then H ∗ (M ) ∼
= H ∗ (N ).
Corollary 6.43. If A is a deformation retract of M , then H ∗ (M ) ∼
= H ∗ (A).
Proof. A deformation retraction is a homotopy H from idM to a retraction r : M →
A of M onto A, where r|A = idA . Let i : A → M be the inclusion map, then ri = idA
and ir ' idM via H. Thus i∗ , r∗ are inverses of each other by homotopy invariance,
which establishes an isomorphism from H ∗ (M ) to H ∗ (A).
Example 6.44 (Cohomology of Rn ). Since Rn deformation retract onto a point via
H(x, t) = tx, for t ∈ [0, 1], we see that H ∗ (Rn ) ∼
= H ∗ (pt), hence
(
R k=0
H k (Rn ) =
0 k > 0.

Example 6.45 (S 1 ' R2 − 0). Viewing S 1 ⊂ R2 , we can define a retraction r :

R2 − 0 → S 1 , where r(x) := x/kxk. We can also find a deformation retraction
from R2 − 0 onto S 1 : This is simply a straight line homotopy H(x, t) := tx/kxk.
In particular, we see that r ◦ i = idS 1 and i ◦ r ' idR2 −0 via H. Hence S 1 and
R2 − 0 is homotopy equivalent. Hence by the homotopy invariance of cohomology,
we immediately get that
H ∗ (R2 − 0) ∼
= H ∗ (S 1 )
and we know H ∗ (S 1 ) from the Mayer-Vietoris calculation. Thus H k (R2 − 0) ∼
= R if
k = 1 and is zero otherwise.
We will briefly sketch the construction of an explicit isomorphism
∼
=
ψ : H 1 (R2 − 0) −
→ R.

Let [α] ∈ H 1 (R2 − 0), which is represented by a closed form α ∈ Ω1 (R2 − 0). Take
any smooth closed loop γ in R2 − 0 and define
Z
ψ([α]) := α.
γ

We note that ψ is well-defined by Stokes’ theorem, and ψ is surjective because of

the existence of the winding number form defined in Example 6.15. To check that
6.5. HOMOTOPY INVARIANCE AND APPLICATIONS 86

ψ is Rinjective, it is equivalent to showing that if α ∈ Ω1 (R2 − 0) such that dα = 0,

and γ α = 0, then there exists F ∈ Ω0 (R2 − 0) such that dF = α. We define F
as follows: Fix a base point p ∈ R2 − 0. For any x ∈ R2 R− 0, pick a smooth path
η such that η(0) = p and η(1) = x. Then define F (x) := η α. One can check that
R
F is well-defined by Theorem 6.14 and the assumption that γ α = 0. Finally, one
can check that dF = α by a simple δ − ϵ limit argument (see the proof of Morera’s
theorem, for example).

Example 6.46. By the example above, we may consider another interesting appli-
cation of the Mayer-Vietoris sequence. The problem is to calculate H ∗ (Ps ), where

Ps := R2 − {p1 , ..., ps }

is the plane with s distinct points removed. Clearly H k (Ps ) = 0 for k > 2 because
of the dimension, and H 0 (Ps ) = R because Ps is connected. Hence the problem
is really about the calculation of H 1 (Ps ). We choose U, V in the Mayer-Vietoris
sequence as Figure 6.3 suggests:

U∼
= Ps−1 V ∼
= P1

p1 p2
ps

ps−1

Figure 6.3: The relative position of p1 , ..., ps does not matter since cohomology
is invariant under homeomorphisms. So we may choose U, V as suggested, with
U∼ = Ps−1 , V ∼
= P1 , U ∩ V ∼
= R2 .

We claim that H 1 (Ps ) ∼

= Rs , for any integer s > 0. We argue by induction. By
the previous remark we already know this is true for s = 1. Now given s > 1, we
may assume H 1 (Ps−1 ) ∼
= Rs−1 inductively. Now the Mayer-Vietoris sequence reads
δ d∗
· · · H 0 (U ) ⊕ H 0 (V ) −
→ H 0 (U ∩ V ) −
→ H 1 (Ps ) → H 1 (U ) ⊕ H 1 (V ) → H 1 (U ∩ V ) · · ·

which by what we discussed and Poincaré lemma, becomes

δ d∗ α
··· → R ⊕ R −
→R−
→ H 1 (Ps ) −
→ Rs−1 ⊕ R → 0 → · · ·

where δ(a, b) = a − b, which implies dim im δ = 1. Hence dim ker d∗ = dim im δ = 1.

By exactness, we know α is a surjection, hence

H 1 (Ps )/ ker α ∼
= Rs

by the first isomorphism theorem. Again by exactness, ker α = im d∗ . And since

dim ker d∗ + dim im d∗ = 1

87 CHAPTER 6. COHOMOLOGY

we conclude that dim im d∗ = 0. Hence dim ker α = 0. This implies that ker α = 0
and
H 1 (Ps ) ∼
= Rs
which proves the claim. Similar to the previous remark, we can construct an explicit
isomorphism ψ : H 1 (Ps ) → Rs , by choosing s smooth closed loops γ1 , ..., γs , each
only contains p1 , ..., ps Rin their Rinterior (ref. Jordan curve theorem), respectively.
Then define ψ([α]) = ( γ1 α, ..., γs α).

Example 6.47 (Cohomology of S n ). Cover S n by two open sets U and V where U is

slightly larger than the northern hemisphere and V slightly larger than the southern
hemisphere (See Figure 6.2 for the case where n = 1). Then U ∩ V is diffeomorphic
to S n−1 × R1 , which is homotopic to S n−1 via a deformation retraction. We claim
that (
R k = 0, n
H k (S n ) =
0 k 6= 0, n
for n > 0 by induction. We already proved the statement for n = 1 in Section 6.4.2.
Assuming the statement holds for S n−1 . Then the Mayer-Vietoris sequence reads
`
Sn U V U ∩V

H n+1 0 0 0

Hn ? 0 0

H n−1 ? 0 R

H n−2 ? 0 0

.. .. .. ..
. . . .

H1 ? 0 0

H0 ? i
R⊕R δ
R

Exactness alone gives

H n (S n ) = R,
H k (S n ) = 0 for k = 2, · · · , n − 1,

And H 0 (S n ) = R by connectedness, so all that remains is to compute H 1 (S n ). To

do this, note that that ker δ = im i is one-dimensional by injectivity of i. Thus, im δ
6.6. POINCARÉ LEMMAS 88

is also one-dimensional, so δ is surjective, and H 1 (S 1 ) = 0, which completes the

induction.
Example 6.48. Let M be a manifold, we claim that H ∗ (M × R) ∼ = H ∗ (M ). Clearly
M × R deformation retracts onto its zero section M × {0} via the homotopy Ht :
(p, x) 7→ (p, tx) for p ∈ M, x ∈ R, t ∈ [0, 1]. By the same construction, we have
H ∗ (M × I) ∼= H ∗ (M ).

6.6 Poincaré Lemmas

6.6.1 Poincaré Lemma and Calculation of H ∗ (Rn )
In the previous section, we already used homotopy invariance to deduce H ∗ (Rn )
and H ∗ (M × R). This section will present an alternative way of arriving at the
same conclusion, via Poincaré Lemma. The proofs are more involved, but some
readers might find it more concrete. More importantly, it introduces the tool of a
chain homotopy operator. Since the important calculation are already done in
the previous section, we encourage the reader to skip this section first and return
to it later if they are not interested at the moment (We will need this result in the
proof of Poincaré duality later).
Theorem 6.49 (Poincaré Lemma). Let π : Rn+1 → Rn be the projection onto the
first factor, and let s : Rn → Rn+1 be the zero section. That is,

π(x1 , ..., xn , t) = (x1 , ..., xn ) s(x1 , ..., xn ) = (x1 , ..., xn , 0).

Then the induced map on cohomology π ∗ : H ∗ (Rn ) → H ∗ (Rn+1 ) and s∗ : H ∗ (Rn+1 ) →

H ∗ (Rn ) are mutual inverses. Therefore, H ∗ (Rn+1 ) ∼
= H ∗ (Rn ).
Therefore, by induction, and the computation of H ∗ (R), one immediately con-
clude that H k (Rn ) = 0 for all k 6= 0, and H 0 (Rn ) = R, for any n.
Proof of Poincaré Lemma. By Theorem 6.11, we see that π ∗ , s∗ are well-defined
morphisms on cohomology. Since π ◦ s = idRn , by functoriality, we see that s∗ π ∗ =
idH ∗ (Rn ) . Therefore, all it remains is to show that π ∗ s∗ = idH ∗ (Rn+1 ) . This is more
difficult: note that even on the level of forms, π ∗ s∗ 6= id.
To show the desired result, we shall construct, for each q, an operator K q :
Ωq (Rn+1 ) → Ωq−1 (Rn+1 ), such that

id −π ∗ s∗ = (−1)q−1 (dK q − K q+1 d) on Ωq (Rn+1 ). (*)

Such a family of operators is called a chain homotopy operator. Note that (*)
implies that π ∗ s∗ is equal to identity on H q (Rn+1 ), although it is not identity on
Ω∗ (Rn+1 ). Because any element [α] ∈ H q (Rn+1 ) is represented by a closed form α,
and exact forms are set to be zero in H q (Rn+1 ). Hence the two terms on the right
hand side of (*) will both be zero when we descend to H q (Rn+1 ). This will finish
the proof.
It remains to construct the family of operators K q : Ωq (Rn+1 ) → Ωq−1 (Rn+1 )
described above. If we let x1 , ..., xn , t be the coordinate on Rn+1 , the we note that
89 CHAPTER 6. COHOMOLOGY

for any form in Ωq (Rn+1 ), it either contains the differential of the last coordinate dt
in the wedge, or it does not. We define K q linearly as follows: If the input does not
contain dt, i.e., the input is of the form

α = f (x, t)dxi1 ∧ · · · ∧ dxiq ,

then define
K q (α) := 0.
If the input contains dt, i.e., the input is of the form

β = f (x, t)dxi1 ∧ · · · ∧ dxiq−1 ∧ dt,

then define Z t
K (β) := dx ∧ · · · ∧ dx
q i1 iq−1
· f (x, λ)dλ.
0

We now check that (*) holds for forms α, β respectively. Since elements in Ωq (Rn+1 )
are linear combinations of elements of the form α, β, we are done since K q ’s are
defined linearly. We first check (*) holds for α. By definition of π, we have s∗ (f ) =
f ◦ s, and π ∗ (dxa ) = dxa , s∗ (dxa ) = dxa , and so on. Therefore,

(id −π ∗ s∗ )α = α − π ∗ (f (x, 0)dxi1 ∧ · · · ∧ dxiq )

= α − f (x, 0)dxi1 ∧ · · · ∧ dxiq
= [f (x, t) − f (x, 0)]dxi1 ∧ · · · ∧ dxiq .

On the other hand, we have

(dK q − K q+1 d)α = −K q+1 α

 
q+1 ∂f
= −K dt ∧ dx ∧ · · · ∧ dx
i1 iq
∂t
 
q ∂f
= −K q+1
(−1) dx ∧ · · · ∧ dx ∧ dt
i1 iq
∂t
Z t
∂f
= (−1) dx ∧ · · · ∧ dx ·
q+1 i1 iq
(x, λ)dλ
0 ∂t
= (−1)q−1 [f (x, t) − f (x, 0)]dxi1 ∧ · · · ∧ dxiq

by the fundamental theorem of calculus. This proves that (*) holds on forms of type
α. Now we check that (*) holds on forms of type β. Note that since s∗ dt = d(t ◦ s) =
d(0) = 0, we have
(id −π ∗ s∗ )β = β.
Now, we have
 
∂f a
K q+1
dβ = K q+1
dx ∧ dx ∧ · · · ∧ dx
i1 iq−1
∧ dt
∂xa
Z t
∂f
= (−1) dx ∧ · · · ∧ dx
q−1 i1 iq−1
∧ dx ·
a
a
(x, λ)dλ
0 ∂x
6.6. POINCARÉ LEMMAS 90

and
 Z t 
dK β = d dx ∧ · · · ∧ dx
q i1 iq−1
· f (x, λ)dλ
0
Z t
∂
= f (x, λ)dλ · dxa ∧ dxi1 ∧ · · · ∧ dxiq−1
∂xa 0
Z
∂ t
+ f (x, λ)dλ · dt ∧ dxi1 ∧ · · · ∧ dxiq−1
∂t 0
Z t
∂f
= a
dt · (−1)q−1 dxi1 ∧ · · · ∧ dxiq−1 ∧ dxa + f (x, t)dt ∧ dxi1 ∧ · · · ∧ dxiq−1
∂x
Z0 t
∂f
= a
dt · (−1)q−1 dxi1 ∧ · · · ∧ dxiq−1 ∧ dxa + (−1)q−1 β.
0 ∂x

Therefore,
dK q β − K q+1 dβ = (−1)q β
which again proves (*).

Therefore, as we discussed, the Poincaré lemma implies that

(
R k=0
H k (Rn ) =
0 k > 0.

Remark 6.50 (General Poincaré Lemma). In fact, let M be any smooth manifold,
and π : M ×R → M be the projection, and s : M → M ×R be the zero section. The
proof of the Poincaré lemma carries over and shows that π ∗ , s∗ gives an isomorphism

H ∗ (M × R) ∼
= H ∗ (M ).

6.6.2 Poincaré Lemma for Compactly Supported Cohomol-

ogy and Calculation of Hc∗ (Rn )
One last important computational task we are yet to finish is computing Hc∗ (Rn ).
This is a consequence of a more general statement:

Theorem 6.51 (Poincaré Lemma for Compactly Supported Cohomology). Let M

be a smooth manifold. Then there is an isomorphism Hc∗+1 (M × R) → Hc∗ (M ).

The construction of the isomorphisms are more difficult, hence we will talk
through the proof for the remainder of this section.
To prove this theorem, again consider π : M × R → M , the projection map. But
π ω ∈ Ω∗ (M × R) clearly has noncompact support, for example for a compactly
∗

supported f ∈ Ω0c (M ), we have π ∗ f (x, t) = f (π(x, t)) = f (x), so Supp(π ∗ f ) =

Supp(f )×R is noncompact. Therefore we cannot hope π ∗ to send Ω∗c (M ) to Ω∗c (M ×
R). However, we may define an integration along fiber map

π∗ : Ω∗c (M × R) → Ω∗−1
c (M )
91 CHAPTER 6. COHOMOLOGY

by again only caring about forms with dt in it: Give M × R the product charts of M
and R, with t being the coordinate on R. Every ω ∈ Ωqc (M × R) is of the following
two types: Either
ω = (π ∗ ϕ)f (x, t) ϕ ∈ Ωqc (M ), f ∈ Cc∞ (M × R)
or
ω = (π ∗ ϕ) ∧ f (x, t)dt ∞
c (M ), f ∈ Cc (M × R).
ϕ ∈ Ωq−1
Then we define π∗ as follows, for two types of forms respectively:
π∗ : (π ∗ ϕ)f (x, t) 7→ 0
Z ∞
∗
π∗ : (π ϕ) ∧ f (x, t)dt 7→ ϕ f (x, t)dt.
−∞
We will need to show that π∗ d = dπ∗ . So that this gives a well-defined map
π∗ : Hc∗ (M × R) → Hc∗−1 (M ).
Lemma 6.52. π∗ d = dπ∗ .
Proof. To prove the equality, we just need to show that it holds for both types of
forms. First let ω ∈ Ωck (M × R) be of “type 1”, i.e. ω = (π ∗ ϕ)f for some ϕ ∈ Ωk (M )
and some compactly supported smooth f : M × R → R. Then
d(π∗ ω) = d(0) = 0,
and
π∗ (dω) = π∗ (d(π ∗ ϕ) ∧ f + (−1)k (π ∗ ϕ) ∧ df )
= π∗ (d(π ∗ ϕ) ∧ f ) + (−1)k π∗ ((π ∗ ϕ) ∧ df )
= π∗ (π ∗ (dϕ) ∧ f ) + (−1)k π∗ ((π ∗ ϕ) ∧ df )
= (−1)k π∗ ((π ∗ ϕ) ∧ df )
where the last equality holds by definition of π∗ . Therefore, we just need to show
that π∗ ((π ∗ ϕ) ∧ df ) = 0. In local coordinates (x1 , · · · , xn , t), we have
 
∗ ∗ ∂f 1 ∂f n ∂f
π ϕ ∧ df = π ϕ ∧ dx + · · · + n dx + dt
∂x1 ∂x ∂t
 
∗ ∗ ∂f
=π ϕ∧ π η+ dt
∂t
∂f
= π ∗ ϕ ∧ π ∗ (η) + π ∗ ϕ ∧ dt
∂t
∂f
= π ∗ (ϕ ∧ η) + π ∗ ϕ ∧ dt,
∂t
where η ∈ Ω1 (M ) is given by η = d(f ◦ s), with s : M → M × R being the zero
section. Clearly η is globally defined, and (∂f /∂t)dt is globally defined when a
product atlas is used. Noting that f is compactly supported, we then have
Z ∞
∗ ∂f
π∗ (π ϕ ∧ df ) = ϕ dt = ϕ[f (x, ∞) − f (x, −∞)] = 0
−∞ ∂t
6.6. POINCARÉ LEMMAS 92

as desired. This establishes the equality dπ∗ ω = π∗ dω in the case that ω is of type
1.
Now assume ω is of type 2, i.e. ω = (π ∗ ϕ)f dt for some ϕ ∈ Ωk−1 (M ) and
some compactly supported smooth f : M × R → R. Then, in local coordinates
(x1 , ..., xn , t), we have
 Z ∞ 
d(π∗ ω) = d ϕ f dt
−∞
Z ∞ Z ∞ 
= dϕ ∧ k−1
f dt + (−1) ϕd f dt
−∞ −∞
Z ∞  Z ∞ 
∂f
= dϕ ∧ f dt + (−1) ϕ ∧ dx
k−1 a
a
dt
−∞ −∞ ∂x

while
  
∗ ∗ ∂f a
π∗ (dω) = π∗ d(π ϕ) ∧ f dt + (−1) (π ϕ) ∧
k−1
dx ∧ dt
∂xa
Z ∞    
k−1 ∗ ∂f a
= dϕ ∧ f dt + π∗ (−1) π ϕ ∧ a dx ∧ dt
−∞ ∂x
Z ∞  
k−1 ∗ a ∂f
= dϕ ∧ f dt + π∗ (−1) π (ϕ ∧ dx ) a dt
−∞ ∂x
Z ∞ Z ∞
∂f
= dϕ ∧ f dt + (−1)k−1 (ϕ ∧ dxa ) a
dt
−∞ −∞ ∂x
Z ∞  Z ∞ 
∂f
= dϕ ∧ f dt + (−1) ϕ ∧ dx
k−1 a
a
dt
−∞ −∞ ∂x
= d(π∗ ω).

This concludes the proof of the lemma.

Now we need to define another map e∗ in the reverse direction, and show that
they are mutual inverses in the cohomology level. We define, on forms,

e∗ : Ω∗c (M ) → Ω∗+1
c (M × R)
R
as follows: Let e = e(t)dt ∈ Ω1c (R) such that R e = 1. Then define e∗ : Ω∗c (M ) →
Ω∗+1
c (M × R) by
e∗ : ϕ 7→ (π ∗ ϕ) ∧ e.
It is easy to see that de∗ = e∗ d. Thus e∗ descends to a well-defined map on the
cohomology groups. It follows directly from definition that

π∗ e∗ = idΩ∗c (M ) .

But again, e∗ π∗ 6= idΩ∗c (M ×R) . So we need to construct a chain homotopy K :

Ω∗c (M × R) → Ω∗−1
c (M × R) again. This time, we define

K : (π ∗ ϕ)f (x, t) 7→ 0
93 CHAPTER 6. COHOMOLOGY

and
Z t Z ∞ 
∗ ∗
K : (π ϕ) ∧ f (x, t)dt 7→ π ϕ f (x, λ)dλ − A(t) f (x, λ)dλ
−∞ −∞

where Z Z t
A(t) = e= e(λ)dλ.
(−∞,t) −∞

Finally, we conclude the proof of Theorem 6.51 because this shows that e∗ , π∗
are isomorphisms that are inverses of each other on the level of cohomology groups,
proving the Poincaré lemma for compactly supported cohomology.

Theorem 6.53. We have

id −e∗ π∗ = (−1)q−1 (dK − Kd) on Ωqc (M × R).

Proof. Again we check the statement on the two types of forms. First, let

ω = π ∗ ϕ · f (x, t) ϕ ∈ Ωqc (M ), f ∈ Cc∞ (M × R)

be of type 1. Then (id −e∗ π∗ )ω = ω − 0 = ω. On the other hand, we have

(dK − Kd)ω = −Kdω

 
∗ ∂f a ∂f
= −Kπ dϕ · f (x, t) + (−1) ϕ ∧ a dx + (−1) ϕ ∧
q q
dt
∂x ∂t
 
∗ ∂f
= (−1) K π ϕ ∧
q−1
dt
∂t
Z t Z ∞ 
q−1 ∗ ∂f ∂f
= (−1) π ϕ (x, λ)dλ − A(t) (x, λ)dλ
−∞ ∂t −∞ ∂t
= (−1)q−1 π ∗ ϕf (x, t) = (−1)q−1 ω

because ∂f /∂t(x, ±∞) = 0 by the assumption that f ∈ Cc∞ (M × R). Therefore, the
claim holds for forms of type 1. Now let

ω = π ∗ ϕ ∧ f (x, t)dt ϕ ∈ Ωq−1 ∞

c (M ), f ∈ Cc (M × R)

be a form of type 2. Now

 Z ∞   Z ∞ 
∗
(id −e∗ π∗ )ω = ω − e∗ ϕ · f (x, t)dt =ω− π ϕ· f (x, t)dt ∧ e.
−∞ −∞

Now we calculate (dK −Kd)ω, which is a bit tedious because it involves cancellation
of many terms. First, for simplicity, we put
Z t Z ∞
β(x, t) := f (x, λ)dλ − A(t) f (x, λ)dλ.
−∞ −∞
6.6. POINCARÉ LEMMAS 94

Then we have
dKω = d(π ∗ ϕβ(x, t))
 
∗ ∂β a ∂β
= π dϕ · β(x, t) + (−1) ϕ ∧ a dx + (−1) ϕ ∧
q−1 q−1
dt
∂x ∂t
Z t Z ∞
∗ ∗
= π dϕ f (x, λ)dλ − π dϕA(t) f (x, λ)dλ
−∞ −∞
Z t Z ∞
q−1 ∗ a ∂ q−1 ∗ a ∂
+ (−1) π ϕ ∧ dx f (x, λ)dλ − (−1) π ϕA(t)dx f (x, λ)dλ
∂xa −∞ ∂xa −∞
Z t Z ∞ 
q−1 ∗ ∂
+ (−1) π ϕ ∧ dt f (x, λ)dλ − A(t) f (x, λ)dλ .
∂t −∞ −∞

Furthermore,
 
∗ q ∗ ∂f a
Kdω = K π dϕ ∧ f (x, t)dt + (−1) π ϕ ∧ a dx ∧ dt
∂x
Z t Z ∞ 
∗
= π dϕ f (x, λ)dλ − A(t) f (x, λ)dλ
−∞ −∞
Z t Z ∞ 
q−1 ∗ ∂f ∂f
+ (−1) π ϕ ∧ dx a
a
(x, λ)dλ − A(t) a
(x, λ)dλ
−∞ ∂x −∞ ∂x
Z t Z ∞
∗ ∗
= π dϕ f (x, λ)dλ − π dϕA(t) f (x, λ)dλ
−∞ −∞
Z t Z ∞
q−1 ∗ a ∂ q−1 ∗ a ∂
+ (−1) π ϕ ∧ dx f (x, λ)dλ − (−1) π ϕA(t)dx f (x, λ)dλ.
∂xa −∞ ∂xa −∞
Hence we see that
Z t Z ∞ 
q−1 ∗ ∂
(dK − Kd)ω = (−1) π ϕ ∧ dt f (x, λ)dλ − A(t) f (x, λ)dλ
∂t −∞
Z−∞∞
= (−1)q−1 π ∗ ϕ ∧ f (x, t)dt − (−1)q−1 π ∗ ϕ ∧ dt · f (x, λ)dλ · e(t)
−∞
 Z ∞ 
∗
= (−1) ω − (−1)
q−1 q−1
π ϕ· f (x, t)dt ∧ e = (−1)q−1 (id −e∗ π∗ )ω.
−∞

As a simple consequence of Theorem 6.51, Example 6.24, and induction, we see

that (
R k=n
Hck (Rn ) =
0 else.
Recall in Example 6.24, the isomorphism Hc1 (R) ∼ = R is achieved by integration.
By the proof of Theorem 6.51, the isomorphism Hcn (Rn ) ∼ = R is given by iterated
n ∼
applications of π∗ . Hence, the isomorphism H (R ) = R is also given by integration.
n

Therefore, a generator of HcnR(Rn ) is a bump n-form α = f (x1 , ..., xn )dx1 ∧ · · · ∧ dxn

with f ∈ Cc∞ (Rn ) such that Rn α = 1. We may take Supp(f ) to be as small as we
like.
95 CHAPTER 6. COHOMOLOGY

6.7 The Mayer-Vietoris Argument

Definition 6.54. An open cover U = (Uα ) of a manifold M n is called a good
cover if all nonempty finite intersections Uα0 ∩ · · · ∩ Uαp are diffeomorphic to Rn .
A manifold which has a finite good cover is said to be of finite type.

The proof of the following theorem requires knowledge in Riemannian geometry,

which we will not cover. Hence, the reader can simply accept the statement of the
theorem.

Theorem 6.55. Every manifold has a good cover. If the manifold is compact, then
the cover may be chosen to be finite.

Proof. Endow a smooth manifold with a Riemannian metric, which is possible for
any smooth manifold: see [1], Proposition 13.3. Then every point on the manifold
has a geodesically convex neighborhood, and intersection of geodesically convex
neighborhoods is again geodesically convex (see [4], p.72, Theorem 3.7 and p.76,
Proposition 4.2).

Let U = (Uα )α∈I , U0 = (Uβ0 )β∈J be two open covers of M . If for all β ∈ J,
Uβ0 ⊂ Uα for some α ∈ I, then we say U0 is a refinement of U, and we write U < U0 .

Corollary 6.56. Every open cover on a manifold has a refinement which is a good
cover.

Definition 6.57. A directed set is a set I with a relation < such that

(a) a < a for all a ∈ I;

(b) If a < b and b < c then a < c;

(c) For any a, b ∈ I, there exists c ∈ I such that a < c and b < c.

Lemma 6.58. The set of open covers on a manifold is a directed set.

Definition 6.59. A subset J of a directed set I is cofinal in I if for every i ∈ I

there exists j ∈ J such that i < j. (Clearly such J will also be a directed set.)

Corollary 6.60. The good covers are cofinal in the set of all covers of a manifold
M.

The existence of good covers on manifolds gives a powerful tool of proving many
results by using the Mayer-Vietoris sequence and five lemma, inducting on the car-
dinality of a good cover. This proof method is called the Mayer-Vietoris argument.
We will prove several statements using the Mayer-Vietoris argument. But in the
proof we will always assume that our manifold is of finite type, i.e., admits a fi-
nite good cover (for example, take any compact manifolds). Many statements that
we will prove are much more general and holds for all smooth manifolds (and even
topological spaces), but then the proof will require much more work. Here, we are
willing to sacrifice generality for more concise and elegant proofs.
6.7. THE MAYER-VIETORIS ARGUMENT 96

Finite Dimensionality of de Rham Cohomology

Proposition 6.61. If M has a finite good cover, then H ∗ (M ) are all finite dimen-
sional.
Proof. The Mayer-Vietoris sequence says
d∗ r
· · · → H q−1 (U ∩ V ) −
→ H q (U ∪ V ) −
→ H q (U ) ⊕ H q (V ) → · · ·
And we have an exact sequence 0 → ker r → H q (U ∪ V ) → im r → 0. Since
im r ∼
= Rn for some n, it is free, so the exact sequence splits, therefore
H q (U ∪ V ) ∼
= ker r ⊕ im r ∼
= im d∗ ⊕ im r
where the last isomorphism is by the exactness of the MV sequence. This says that
if H q (U ), H q (V ), H q−1 (U ∩ V ) are finite dimensiobal, then so is H q (U ∪ V ).
If M is diffeomorphic to Rn , then Poincaré lemma proves that H ∗ (M ) is finite di-
mensional. Now induct on the cardinality of a good cover. Suppose the cohomology
of any manifold having good cover with at most p open sets is finite dimensional.
Let M be a manifold with p + 1 good cover U0 , ..., Up . Now (U0 ∪ · · · ∪ Up−1 ) ∩ Up is
a manifold with a good cover consisting of p open sets U0 ∩ Up , U1 ∩ Up , ..., Up−1 ∩ Up .
Thus we know that H ∗ (Up ), H ∗ (U0 ∪ · · · ∪ Up−1 ), H ∗ ((U0 ∪ · · · ∪ Up−1 ) ∩ Up ) are all
finite dimensional. Then the claim that H ∗ (M ) is finite dimensional follows from
the previous paragraph.
Similarly,
Proposition 6.62. If M has a finite good cover, then Hc∗ (M ) are all finite dimen-
sional.

Poincaré Duality on an Orientable Manifold

Definition 6.63. A pariting between two finite-dimensional vector spaces h , i :
V ⊗ W → R is said to be nondegenerate if hv, wi = 0 for all w ∈ W implies
v = 0 and for all v ∈ V implies w = 0. Equivalently, the map v 7→ hv, i defines an
injection V → W ∗ and the map w 7→ h , wi defines an injection W → V ∗ .
Lemma 6.64. The pairing between two finite-dimensional vector spaces h , i : V ⊗
W → R is nondegenerate iff the map v 7→ hv, i defines an isomorphism V → W ∗ .
Proof. Suppose the pairing is nondegenerate. Then the map v 7→ hv, i defines an
injection V → W ∗ and the map w 7→ h , wi defines an injection W → V ∗ . So we
have
dim V ≤ dim W ∗ = dim W ≤ dim V ∗ = dim V.
Thus dim V = dim W and the injection v 7→ hv, i is actually an isomorphism.
Conversely, suppose the map φ : v 7→ hv, i defines an isomorphism V → W ∗ .
Suppose hv, wi = 0 for all w ∈ W . Then denote the map φ(v)(w) := hv, wi = φv (w).
So φv (w) = 0 for all w ∈ W , which means φv ≡ 0, so v = 0 since φ is an isomorphism
(hence injective). Now suppose hv, wi = 0 for all v ∈ V . Then pick ψ ∈ W ∗ such
that ψ : W → R is injective. Since φ is an isomorphism, ψ = φu for some u ∈ V .
By assumption, ψ(w) = φu (w) = hu, wi = 0. Hence w = 0 since ψ is injective.
97 CHAPTER 6. COHOMOLOGY

Now let M n be an oriented smooth manifold with no boundary (For manifold

with boundary, Poincare duality does not hold, see this post. There is a gener-
alization called Lefschetz duality for manifold with boundary). We consider the
pairing
Z
: H q (M ) ⊗ Hcn−q (M ) → R
Z
α ⊗ β 7→ α ∧ β.
M

Here, we are slightly abusing notation and use α instead of [α] to denote a cohomol-
ogy class in H q (M ). This map is well-defined: If α = α0 ∈ H q (M ), i.e., α − α0 = dη
for some η ∈ Ωq−1 (M ), then
Z Z Z Z
0
(α − α ) ∧ β = dη ∧ β = d(η ∧ β) − (−1) deg η
η ∧ dβ
M M Z M Z M

= η ∧ β − (−1) deg η
η ∧ dβ = 0.
∂M M

In the second to last equality, we used Stokes’ theorem in the first term, which we
can because η ∧ β ∈ Ωn−1
c (M ), and since ∂M = ∅, the first term is zero. The second
term is also zero because β is closed since it represents a cohomology class. Similarly
one can show that if β = β 0 ∈ Hcn−q (M ), the resulting pairing is also well-defined.
Theorem 6.65 (Poincaré
R Duality for Oriented Manifold). Let M n be an
R oriented
manifold. The paring : H (M ) ⊗ Hc (M ) → R defined by α ⊗ β 7→ M α ∧ β is
q n−q

nondegenerate. Equivalently, we have an isomorphism

∼
→ (Hcn−q (M ))∗
=
H q (M ) −
R
Given by α 7→ M
α ∧ (•).
The proof of the Poincaré duality theorem relies on the following key technical
lemma:
Lemma 6.66. The following diagram obtained by putting the MV exact sequence for
U, V on top, and the MV exact sequence for cohomology of U, V with compact support
below and dualizing, has exact rows and is sign commutative (i.e., commutative up
to a plus or minus sign):

r − d∗
· · · → H q (U ∪ V ) H q (U ) ⊕ H q (V ) H q (U ∩ V ) H q+1 (U ∪ V ) → · · ·

∫ ∫ ∫ ∫ ∫
α7→ U ∪V
α∧(•) (σ,τ )7→ U
σ∧(•)+ V
τ ∧(•) ω7→ U ∩V
ω∧(•) η7→ U ∪V
η∧(•)

+∗ δ∗ (d∗ )∗
· · · → Hcn−q (U ∪ V )∗ Hcn−q (U )∗ ⊕ Hcn−q (V )∗ Hcn−q (U ∩ V )∗ Hcn−q−1 (U ∪ V )∗ → · · ·

Here, r and − are restriction and difference maps defined in the MV sequence,
and d∗ is the boundary map defined in the MV sequence. The map δ : Hcn−q (U ∩
V ) → Hcn−q (U ) ⊕ Hcn−q (V ) is the signed inclusion δ(ω) = (−j∗ ω, j∗ ω), and + :
Hcn−q (U ) ⊕ Hcn−q (V ) → Hcn−q (U ∪ V ) is the summation map.
6.7. THE MAYER-VIETORIS ARGUMENT 98

Proof. We check the (sign) commutativity of each squares:

• For the first square, let α ∈ RH q (U ∪ V ). We first go down and then go left.
Denote the linear functional U ∪V α ∧ (•) by α̂. By definition of dualization of
a linear map,
Z
∗
+ (α̂)(β1 , β2 ) = α̂(+(β1 , β2 )) = α ∧ (β1 + β2 )
U ∪V

where β1 ∈ Hcn−q (U ), β2 ∈ Hcn−q (V ). Clearly we get the same result by first

going left and then go down in the first square. Note that the integral on
U ∪ V indeed splits into an integral on U plus an integral on V , because β1 is
supported in U , and similarly for β2 .

• For the second square, we again first go down and then go left. Let (σ, τ ) ∈
H q (U ) ⊕ H q (V ). The linear functional we get in Hcn−q (U ∩ V )∗ , recalling the
definition of dualization, will be defined as follow: for any α0 ∈ Hcn−q (U ∩ V ),
it gets mapped to
Z Z Z Z
0 0 0
σ ∧ (−j∗ α ) + τ ∧ (j∗ α ) = − σ ∧ α + τ ∧ α0 .
U V U V

Since the map H q (U ) ⊕ H q (V ) → H q (U ∩ V ) is defined by (σ, τ ) 7→ τ − σ, we

clearly get the same result when we go down.

• For the third square, let ω ∈ H q (U ∩ V ). Recall that d∗ ω ∈ H q+1 (U ∪ V ) is

defined by d∗ ω|U = −d(ρV ω) and d∗ ω|V = d(ρU ω). Also recall the definition
of d∗ from Remark 6.38: For τ ∈ H n−q−1 (U ∪ V ), we showed that d∗ τ is a
form in Hcn−q (U ∩ V ) such that

(−(jU )∗ (d∗ τ ), (jV )∗ (d∗ τ )) = (d(ρU τ ), d(ρV τ )).

Note that since τ is represented by a closed form, we have d(ρV τ ) = (dρV ) ∧ τ .

Similarly, d(ρV ω) = (dρV ) ∧ ω.
Now we first go down and go left in the third square of the diagram. The
result will be
Z Z Z
ω ∧ d∗ τ = ω ∧ (dρV ) ∧ τ = (−1) deg ω
(dρV ) ∧ ω ∧ τ.
U ∩V U ∩V U ∩V

Now if we first go left and go down in the third square, we would get (note
that Supp(ω) ⊂ U ∩ V ):
Z Z
∗
d ω∧τ =− (dρV ) ∧ ω ∧ τ
U ∪V U ∩V

Hence we see that the result we get from two different paths only differ by a
sign, i.e., differ by (−1)deg ω+1 .
99 CHAPTER 6. COHOMOLOGY

Remark 6.67. There are still two quick remarks I want to make about the proof
of the previous lemma, and how it helps us to prove Poincaré duality:
• There is one small detail: We know MV sequence for Hc∗ is exact, but we do
not know yet when we dualize everything the sequence remain exact. Luckily
in this case, it is.
Reason: The functor HomR (−, M ) is exact iff M is an injective R-module.
Given a field k, every k-vector space Q is an injective k-module. See details
in the example section of Wikipedia. In particular, HomR (−, R) is exact.

• So we know from lemma 6.66 that the diagram there is sign commutative.
But to prove Poincaré duality we want to use the Five Lemma. But the
Five Lemma requires the diagram to be commutative instead of just sign
commutative. The trick here is to fix the vertical map a little bit so that
everything becomes commutative: Suppose we have a diagram with exact
rows and sign commutative squares, with vertical maps being isomorphisms
as below, but say the first square is only commutative up to a minus sign:
f g
··· A B C ···
α β γ

··· A0 f′
B0 g′
C0 ···

That is, βf = −f 0 α. Then replacing β by −β : B → B 0 gives again an

isomorphism, and this time the square commutes. Now if the next square is
again sign commutative only, we replace γ by −γ. And we did not change the
rows, so exactness is preserved. After fixing the diagram, we may apply the
Five Lemma.
Finally we present the proof of the Poincaré duality for an oriented manifold with
a finite good cover (this is necessary for induction). Although this is a weaker version
of the Poincaré duality theorem, but its proof is extremely concise and elegant.
Proof of Poincaré Duality. When M is diffeomorphic to Rn , then Poincaré dual-
ity follows from the Poincaré lemma, i.e., R the calculation of H ∗ (Rm ) and Hc∗ (Rm ).
We need to show that the integration : H (R ) → Hcn−q (Rn )∗ is actually an
q n

isomorphism. When q 6= 0, by Poincaré lemma H q (Rn ) = 0 = Hcn−q (Rn )∗ . So

any map between the trivial groups is an isomorphism. When q = 0, by Poincaré
lemma
R we have H 0 (Rn ) = Hcn (Rn ) ∼ = R. Therefore Hcn (Rn )∗ ∼ = R as well, so
n ∗
: H (R ) → Hc (R ) can be viewed as a linear map between R. We claim that
0 n n

this map is Rsurjective: Take any φ ∈ Hcn (Rn )∗ and ω ∈ Hcn (Rn )Ra generator of R the
group, i.e., Rn ω = 1. Then say φ(ω) = λ ∈ R. Then the map (λ) : α 7→ λ Rn α
is also a linear functional on Hcn (Rn ) that agrees with φ on the generator ω, hence
they are equal. This shows the integration map is a surjective linear map between
vector spaces of same dimension, hence is an isomorphism.
Now induct on the cardinality of a good cover. Suppose Poincaré duality holds
for any manifold having a good cover with at most p. Let M be a manifold with
p + 1 good cover U0 , ..., Up . Now (U0 ∪ · · · ∪ Up−1 ) ∩ Up is a manifold with a good
6.7. THE MAYER-VIETORIS ARGUMENT 100

cover consisting of p open sets U0 ∩ Up , U1 ∩ Up , ..., Up−1 ∩ Up . Use MV sequence on

the open sets (U0 ∪ · · · ∪ Up−1 ) ∩ Up and Up and Lemma 6.66. This completes the
inductive step.
Corollary 6.68. If M is a connected oriented manifold of dimension n, then Hcn (M ) ∼
=
R. In prticular if M is compact, oriented, and connected, then H n (M ) ∼
= R.

Künneth Formula
Theorem 6.69 (Künneth Formula). Let M, F be two manifolds, then for every
nonnegative integer n, one has
M
n
H (M × F ) ∼
n
= H p (M ) ⊗ H n−p (F ).
p=0

Or written more compactly, we have

H ∗ (M × F ) ∼
= H ∗ (M ) ⊗ H ∗ (F ),
L
where H ∗ (M ) = ∞ p ∗
p=1 H (M ) denote the cohomology ring, and similarly for H (F )
and H ∗ (M × F ).
Proof. We again prove this theorem for manifolds with finite good cover, using the
beautiful MV argument. There are two obvious projection maps

π :M ×F →M
ρ : M × F → F.

These two maps induces a map in cohomology rings

ψ : H ∗ (M ) ⊗ H ∗ (F ) → H ∗ (M × F )
ω ⊗ ϕ 7→ π ∗ ω ∧ ρ∗ ϕ

for ω ∈ H p (M ), ϕ ∈ H n−p (N ), and for any 0 ≤ p ≤ n. We shall prove Künneth

formula by showing that ψ defined above is an isomorphism.
For ML= Rm this is simply the Poincaré lemma. But we still need to show
that ψ : np=0 H p (Rm ) ⊗ H n−p (F ) → H n (Rm × F ) induces an isomorphism using
the Poincaré lemma. If p 6= 0 then H p (Rm ) = 0, so ψ is really just a map ψ : R ⊗
H n (F ) → H n (Rm ×F ). Recall in the proof of the Poincaré lemma, we constructed an
isomorphism π ∗ : H n (Rm−1 ×F ) → H n (Rm ×F ) from the projection π : (Rm−1 ×F )×
R1 → Rm−1 × F . Therefore by inductively applying projection map and eliminating
one coordinate of Rm , we get an isomorphism (π ∗ )n : H n (F ) → H n (Rm × F ), where
π n : Rm × F → F . But then π n is just ρ : Rm × F → F . Thus ψ can actually be
written as a composition of two isomorphisms
ψ
R ⊗ H n (F ) H n (Rm × F )
ϵ:a⊗ω7→aω
ρ∗
n
H (F )
101 CHAPTER 6. COHOMOLOGY

Here, ϵ is an isomorphism just by the property of tensor product, and ρ∗ = (π ∗ )n

is an isomorphism because it is a composition of π ∗ ’s, each is an isomorphism by
Poincaré lemma. The diagram obviously commutes, so we have ψ = ρ∗ ◦ ε which is
an isomorphism.
For general M , let U, V be two open sets in M . Then U × F, V × F are two open
sets in M × F . Since tensoring with a vector space preserves exactness, we tensor
the MV sequence of the pair (U, V ) with H n−p (F ) and get an exact sequence
· · · → H p (U ∪ V ) ⊗ H n−p (F ) → (H p (U ) ⊗ H n−p (F )) ⊕ (H p (V ) ⊗ H n−p (F ))
→ H p (U ∩ V ) ⊗ H n−p (F ) → · · ·
And direct summing also preserves exactness, so we get an exact sequence
M M
··· → H p (U ∪ V ) ⊗ H n−p (F ) → (H p (U ) ⊗ H n−p (F )) ⊕ (H p (V ) ⊗ H n−p (F ))
p∈Z p∈Z
M
→ H p (U ∩ V ) ⊗ H n−p (F ) → · · ·
p∈Z

The trick is to put this exact sequence on top, and the MV exact sequence of the
pair (U × F, V × F ) on the bottom, match them up using ψ componentwise, and
show that we have a commutative diagram
Ln Ln Ln
p=0 H p (U ∪ V ) ⊗ H n−p (F ) p=0 (H
p
(U ) ⊕ H p (V )) ⊗ H n−p (F ) p=0 H p (U ∩ V ) ⊗ H n−p (F )
ψ ψ ψ

H ((U ∪ V ) × F )
n
H (U × F ) ⊕ H (V × F )
n n
H ((U ∩ V ) × F )
n

L Ln
Note that here instead of writing p∈Z we write p=0 . This is because when
p < 0 or p > n, one of the two groups we are tensoring is the trivial group, so
the direct sum becomes a finite direct sum with only nontrivial cohomology groups
tensored together.
The commutativity of the first and the second square commutes is left as an
exercise. The first one is easy once all the relevant maps are written out and we
can just chase the diagram. The second square commutes because we are using the
isomorphism (See Remark 6.70)
M
n
∼
=
M
n
(H (U ) ⊕ H (V )) ⊗ H
p p n−p
(F ) −
→ (H p (U ) ⊗ H n−q (F )) ⊕ (H p (V ) ⊗ H n−q (F ))
p=0 p=0
n−p ∼
=
(αp , β p ) ⊗ τ −
→ (αp ⊗ τ n−p , β p ⊗ τ n−p )
for αp ∈ H p (U ), β p ∈ H p (V ), τ n−p ∈ H n−p (F ).
Accepting this, then the commutativity of the second square can be verified
easily. Now we verify commutativity of the third square
Ln ∗ Ln
p=0 H (U ∩ V ) ⊗ H (F ) d (U ∪ V ) ⊗ H n−p (F )
p n−p p+1
p=−1 H

ψ ψ

d∗
H n ((U ∩ V ) × F ) H n+1 ((U ∪ V ) × F )
6.7. THE MAYER-VIETORIS ARGUMENT 102

L L
Note that in the second term on the right, the sum p∈Z actually becomes np=−1 .
It suffices to verify the commutativity of the diagram for pure tensors (Remark
6.71). Let ω ⊗ ϕ ∈ H p (U ∩ V ) ⊗ H n−p (F ). Then

ψd∗ (ω ⊗ ϕ) = π ∗ (d∗ ω) ∧ ρ∗ ϕ
d∗ ψ(ω ⊗ ϕ) = d∗ (π ∗ ω ∧ ρ∗ ϕ).

Recall that if ρU , ρV is a partition of unity subordinate to U, V , then d∗ ω|U =

−d(ρV ω) and d∗ ω|V = d(ρU ω). Note that π ∗ ρU , π ∗ ρV form a partition of unity on
(U ∪ V ) × F subordinate to the cover U × F, V × F , and since ϕ is closed, on
(U ∩ V ) × F we have

d∗ (π ∗ ω ∧ ρ∗ ϕ) = d((π ∗ ρU )π ∗ ω ∧ ρ∗ ϕ)
= (dπ ∗ (ρU ω)) ∧ ρ∗ ϕ
= π ∗ (d∗ ω) ∧ ρ∗ ϕ.

Hence the diagram is indeed commutative.

Finally to finish off the proof of Künneth theorem: By the Five Lemma, if the
theorem is true for U, V, U ∩ V , then it is true for U ∪ V . Use the MV argument and
induct on the cardinality of a good cover.

Remark 6.70. tensor product distributes over direct sum is a standard result in
algebra. A detailed proof of this isomorphism can be found in this post. To get an
idea, we can do a simple case and prove (A ⊕ B) ⊗ C ∼ = (A ⊗ C) ⊕ (B ⊗ C). The
isomorphism can be constructed via φ : (A ⊕ B) ⊗ C → (A ⊗ C) ⊕ (B ⊗ C) where

φ : (a, b) ⊗ c 7→ (a ⊗ c, b ⊗ c).

The inverse of this map is given by ψ : (A ⊗ C) ⊕ (B ⊗ C) → (A ⊕ B) ⊗ C where

ψ : (a ⊗ c1 , b ⊗ c2 ) 7→ (a, 0) ⊗ c1 + (0, b) ⊗ c2 .

One can verify that both φ, ψ are well-defined homomorphisms that are inverses of
each other. For example

φψ(a⊗c1 , b⊗c2 ) = φ((a, 0)⊗c1 +(0, b)⊗c2 ) = (a⊗c1 , 0)+(0, b⊗c2 ) = (a⊗c1 , b⊗c2 ).

The key here is to realize that (A ⊕ B) ⊗ c does not only consists of elements of the
form (a, b) ⊗ c, but linear combinations of elements like this.

Remark 6.71. Finally, I shall remark that for our purpose, it suffices to check the
diagram above commutes for elements with the same coefficient in H n−q (F ), i.e.,
elements of the form (αp ⊗ τ n−p , β p ⊗ τ n−p ). This is because for any random element,
it can be decomposed into elements with the same H n−q (F ) coefficients:

(αp ⊗ σ n−p , β p ⊗ τ n−p ) = (αp ⊗ σ n−p , 0 ⊗ σ n−p ) + (0 ⊗ τ n−p , β p ⊗ τ n−p ).

Therefore by linearity, it suffices to check the diagram commutes for elements with
the same H n−q (F ) coefficients.
103 CHAPTER 6. COHOMOLOGY

Example 6.72. To demonstrate the power of Künneth theorem Q with an example,

let us prove that the cohomology of the n-fold torus T n := ni=1 S 1 = S 1 × · · · S 1
satisfies
H k (T n ) = R(k )
n

for all k. We prove this statement by induction on n. First, the claim obviously
holds for the base case where n = 1, by Section 6.4.2. Next, we assume that
H k (T n−1 ) = R( )
n−1
k

holds for all k. Now by Theorem 6.69, we have that

M
H k (T n ) = H k (T n−1 × S 1 ) = H m (T n−1 ) ⊗ H n (S 1 )
m+n=k
k
= (H (T n−1
) ⊗ R) ⊕ (H k−1 (T n−1 ) ⊗ R)
∼
= H k (T n−1 ) ⊕ H k−1 (T n−1 )
∼
= R( k ) ⊕ R(k−1) ∼
n−1 n−1
= R( k )
n

where one can check the identity

     
n−1 n−1 n
+ =
k k−1 k
easily with some algebra. This proves the claim. But one can also repeatedly apply
Theorem 6.69 n times on individual factors S 1 in the product T n = S 1 × · · · S 1 and
get M
H k (T n ) = H i1 (S 1 ) ⊗ · · · ⊗ H in (S 1 ).
i1 +···+in =k

Now, H (S ) = 0 unless ip = 0, 1 for all 1 ≤ p ≤ n. If we have H ip (S 1 ) = 0, then

ip 1

the term in the direct sum containing it will be the trivial group, which does not
contribute to H k (T n ). And since any finite tensor products of R is isomorphic to R,
the number of nontrivial factors of H k (T n ) in the direct sum above corresponds to
the number of distinct ways one
n

can write i1 + · · · + in = k, where i1 , ..., in ∈ {0, 1}.
And the answer of course is k . This is why the binomial coefficient appears in this
problem, because it is just a combinatorial problem in disguise.

Poincaré Dual of a Closed Oriented Submanifold

Let M n be an oriented manifold and S k a closed (as subspace
R ∗of M ) submanifold
∗
of M . Let ω ∈ Ωc (M ). Then iS ω ∈ Ωc (S). So the integral S iS ω is defined.
k k
R ∗ Then
integration on S induces a linear functional on Hc (M ) via IS : ω 7→ S iS ω. By
k

Poincaré duality:
(Hck (M ))∗ ∼= H n−k (M ).
R
Recall that the isomorphism H n−k (M ) → (Hck (M ))∗ is given by η 7→
R M (•) ∧ ω up
to a sign. Therefore there exists a unique ηS ∈ H (M ) such that M (•) ∧ ηS = IS .
n−k

That is, there exists unique ηS ∈ H n−k (M ) such that

Z Z
∗
iS ω = ω ∧ ηS .
S M
We call ηS the Poincaré dual of S.
6.8. THE THOM ISOMORPHISM 104

Proposition 6.73. When S is compact, its Poincaré dual may be taken to have
compact support.

Proof. Now Rsuppose M n has a finite good cover, and S k is compact. R Then since S
is compact, S i∗S ω < ∞ for any ω ∈ Ωk (M ). Thus this time ω 7→ S i∗S ω is a map
IS : H k (M ) → R. Now since M has a finite good cover, its de Rham cohomology
(with compact support) are finite-dimensional vector spaces, so Poincaré duality
H k (M ) ∼= (Hcn−k (M ))∗ implies that Hcn−k
R (M ) ∼
= (H k (M ))∗ . The isomorphism
Hcn−k (M ) → (H k (M ))∗ is given by η 7→ M (•) ∧ η Rup to a sign. Therefore, there
exists a unique ηS0 ∈ Hcn−k (M ) such that IS (ω) = M ω ∧ ηS0 , for all ω ∈ H k (M ).
That is, Z Z
i∗S ω = ω ∧ ηS0 .
S M

6.8 The Thom Isomorphism

Let M m be a smooth manifold, and let E be a rank n vector bundle over M (recall
Definition 3.3). Throughout this section, we consider the situation where (Uα )α is a
family of open covers of M , and let ϕα : π −1 (Uα ) → Uα × Rn be the corresponding
family of trivilization maps on E. The maps

ϕα ◦ ϕ−1
β : (Uα ∩ Uβ ) × R → (Uα ∩ Uβ ) × R
n n

are vector space automorphisms of Rn in each fiber and hence give rise to maps

gαβ : Uα ∩ Uβ → GL(n, R) gαβ (x) = ϕα ◦ ϕβ |{x}×Rn .

Definition 6.74. A trivilization (Uα , ϕα )α∈I on E is said to be oriented if for every

α, β, the transition function gαβ has positive determinant everywhere on Uα ∩ Uβ .
Two trivilizations (Uα , ϕα ), (Vβ , ψβ ) are equivalent if for every x ∈ Uα ∩ Vβ , the
function ϕα ◦ ψβ−1 (x) : Rn → Rn has positive determinant, written as (Uα , ϕα ) ∼
(Vβ , ψβ )

One can easily check that

Proposition 6.75. ∼ is an equivalence relation on the set of oriented trivilizations

on E.

This immediately gives

Proposition 6.76. On a connected maniflod M the equivalence relation ∼ parti-

tions all the oriented trivilizations of E into two equivalence classes. Either equiva-
lence class is called an orientation on the vector bundle E.

Lemma 6.77. An orientable vector bundle E over an orientable manifold M is an

orientable manifold.
105 CHAPTER 6. COHOMOLOGY

Proof. Let (Uα , ψα )α be an oriented atlas for M with transition functions hαβ =
ψα ◦ ψβ−1 . Let ϕα : π −1 (Uα ) → Uα × Rn be the family of local trivilizations for E
with transition functions gαβ . Then the composition
ϕα ψα ×id
π −1 (Uα ) −→ Uα × Rn −−−→ Rm × Rn

gives an atlas for E. The transition function of this atlas is given by

σαβ : (ψα × id) ◦ gαβ ◦ (ψβ−1 × id) : Rm × Rn → Rm × Rn

which is given by
σαβ (x, y) = (hαβ (x), gαβ (ψβ−1 (x))y).
Then its Jacobian is given by
 
J(hαβ ) ∗
J(σαβ ) =
0 gαβ (ψβ−1 (x))

which clearly has positive determinant.

Remark 6.78 (Co-orientation). Let E be an oriented vector bundle of rank n over
an oriented manifold M . The orientation given by Lemma 6.77 induces a natural
orientation on the fibers. We may orient each fiber Ep = π −1 (p) as follows: Near
any point p ∈ M , let x1 , ..., xm be a local coordinate of M determined by the open
set Uα , the zero section in E, and t1 , ..., tn be a fiber coordinate of E. Since near p
we have π −1 (Uα ) ∼
= Uα × Rn via the local trivilization ϕα . This means that at the
point q with π(q) = p, we have Tq E ∼ = Tp M ⊕ Tq Ep , where we identify M with its
zero section in E. Then we orient the fiber Ep such that the direct sum orientation
Tp M ⊕ Tq Ep agrees with that of Tq E. This is called the co-orientation.
Here is a result that we will need, but we will not prove it here. For a proof, see
[2] Theorem 6.8 and Corollary 6.9 on p.57-59.
Lemma 6.79. If M is contractible and E is any rank n vector bundle over M , then
E is trivial: E ∼
= M × Rn .
Corollary 6.80. If M has a good cover U = (Uα )α∈I , then π −1 (U) is a good cover
on E, where π : E → M is a vector bundle. In particular, if M is of finite type,
then so is E.
Proof. Let Uα ⊂ U be a good cover for M . Then π −1 (Uα ) is a vector bundle over
U , which is a contractible manifold. Hence by Lemma 6.79, π −1 (Uα ) is trivial,
i.e., there are diffeomorphisms ϕα : π −1 (Uα ) → Uα × Rn ∼ = Rm × Rn . Moreover,
= (Uα ∩ Uβ ) × Rn ∼
π −1 (Uα ) ∩ π −1 (Uβ ) = π −1 (Uα ∩ Uβ ) ∼ = Rm × Rn . Hence we see
that π −1 (U) is indeed a good cover for E.
We now start our computation of H ∗ (E), Hc∗ (E) using the information of coho-
mologies of M . First of all, we have
Proposition 6.81. Let E be any vector bundle over M . Then we have H ∗ (E) ∼
=
∗
H (M ).
6.8. THE THOM ISOMORPHISM 106

Proof. Consider the zero section s0 : M → E defined by s0 (x) = (x, 0) ∈ E. Then

clearly π ◦ s0 = idM . We claim that s0 ◦ π ' idE . The homotopy is given by
H : E × R → E where H(x, v, t) = (x, tv). Hence by homotopy invariance, we see
that E ' M (In fact, the zero section M × {0} ⊂ E is a deformation retract of
E and is diffeomorphic to M ). Thus by homotopy invariance we get the desired
result.
However, it is not true in general that Hc∗ (E) ∼
= Hc∗−n (M ), despite it being true
for E = M × R according to Poincaré Lemma for compactly supported cohomology.
For example, the open Möbius strip E satisfies Hck (E) = 0 for all k, but E is a rank
one vector bundle over S 1 and we don’t have H 2 (E) ∼ = H 1 (S 1 ). But our prediction
is indeed true for the following special cases:
Theorem 6.82 (Thom isomorphism). Let M be an oriented manifold of finite type,
and E an oriented rank n vector bundle over M . Then for all k, we have

Hck+n (E) ∼
= Hck (M ).

Proof. By Lemma 6.77, we know that E is also an oriented manifold, and by Corol-
lary 6.80, we know that E is of finite type. Then the statement can be proven by
using Poincaré duality twice, and H ∗ (E) ∼
= H ∗ (M )

Hck+n (E) ∼
= (H m+n−k (E))∗ ∼
= (H m+n−k (M ))∗ ∼
= Hck−n (M ).

This proves the claim.

This is Thom isomorphism, in its simplest form. However, there is a more general
version of Thom isomorphism, which we will now study. The proof is very simliar
to the Poincaré Lemmas. We need some definitions first:
Definition 6.83. Let E be a vector bundle of rank n over a smooth manifold M m .
A differential k-form ω ∈ Ωk (E) is said to have compact vertical support if for
every compact subset K ⊂ M , the subset Supp ω ∩ π −1 (K) is compact in E. Define

Ωkcv (E) := {ω ∈ Ωk (E) : ω has compact vertical support}.

One can easily see that exterior differentiation preserves forms with vertical compact
support. Hence we define the kth cohomology of E with compact vertical
support
k ker(d : Ωkcv (E) → Ωk+1
cv (E))
Hcv (E) := .
im(d : Ωcv (E) → Ωkcv (E))
k−1

Now we define a map, called integration along fiber, similar to what we did
in order to prove Poincaré Lemma.

cv (E) → Ω (M ) as follows:
Definition 6.84. Define π∗ : Ωk+n k

• First suppose E = M × Rn . Let x1 , ..., xm be a coordinate on the base M and

t1 , ..., tn be a coordinate on the fiber. Then any ω ∈ Ωk+n (E) has the form

ω = (π ∗ ϕ)f (x, t)dt1 ∧ · · · ∧ dtn + ζ

107 CHAPTER 6. COHOMOLOGY

where ϕ ∈ Ωk (M ), f has compact support for each fixed x on M , and ζ is a

form with less than n fiber differential in its wedges. We then define
Z
(π∗ ω)(p) = ϕ(p) · f (x, t)dt1 ∧ · · · ∧ dtn .
Ep

Here, we orient Ep by co-orientation as in Remark 6.78. In other words,

Z
π∗ (ω) = ϕ f (x, t1 , ..., tn )dt1 · · · dtn .
Rn

• Now suppose E be an arbitrary oriented vector bundle, with oriented trivi-

lizations ϕα : π −1 (Uα ) → Uα × Rn of E. Write ωα := ω|π−1 (Uα ) . Suppose on
π −1 (Uα ) we have coordinates (x1 , ..., xm , t1 , ..., tn ), and

ωα = (π ∗ ϕ)f (x1 , ..., xm , t1 , ..., tn )dt1 ∧ · · · ∧ dtn + ζα

where ζα again contains less than n fiber differential in its wedges. Then define
−1
cv (π (Uα ) → Ω (Uα ) via
π∗ : Ωk+n k

Z
(π∗ ωα )(p) = ϕ(p) · f (x, t)dt1 ∧ · · · ∧ dtn
Ep

for any p ∈ Uα . Again, we give Ep the co-orientation. In other words,

Z
π∗ (ωα ) := ϕ f (x, t1 , ..., tn )dt1 · · · dtn .
Rn

Proposition 6.85. If E is an oriented vector bundle and ω ∈ Ωk+n cv (E) as in the

second case in the definition above, then π∗ ωα = π∗ ωβ on the intersection. Hence
(π∗ ωα )α pieces together to give a global form π∗ ω on M . Moreover, this definition is
independent of the choice of oriented trivilization for E. Hence we get a well-defined
map
cv (E) → Ω (M ).
π∗ : Ωk+n k

Proof. First, suppose

ωβ = (π ∗ τ )g(y 1 , ..., y m , u1 , ..., un )du1 ∧ · · · ∧ dun + ζβ

for another set of coordinate. Suppose

ωα = π ∗ (ρdxi1 ∧ · · · ∧ dxik )f dt1 ∧ · · · dtn + ζα

for ρ ∈ C ∞ (Uα ) and f has compact support on every fiber of E. Then on π −1 (Uα ) ∩
π −1 (Uβ ), note that since π ∗ dxi = dxi , one has

ωα = (π ∗ ρ)f dxi1 ∧ · · · ∧ dxik ∧ dt1 ∧ · · · ∧ dtn

∂xi1 ∂xik ∂t1 ∂tn p1
= (π ∗ ρ)f p1 · · · p · · · dy ∧ · · · dy pn ∧ duq1 ∧ · · · duqn .
∂y ∂y k ∂uq1 ∂uqn
6.8. THE THOM ISOMORPHISM 108

Hence on π −1 (Uα ) ∩ π −1 (Uβ ), we should have

∂xi1 ∂xik p1
τ =ρ · · · dy ∧ · · · ∧ dy pn
∂y p1 ∂y pk
and
∂t1 ∂tn q1
gdu1 ∧ · · · ∧ dun = f · · · du ∧ · · · duqn
∂uq1 ∂u 
qn

∂tµ
= f det du1 ∧ · · · ∧ dun
∂uν

where in the second equality, we used the argument in the proof of (4.3). Now for a
fixed point p on the intersection Uα ∩ Uβ , we have, by the change of variable formula
of integration,
Z
π∗ ωα (p) = ρdx ∧ · · · ∧ dx (p) ·
i1 ik
f dt1 ∧ · · · ∧ dtn
E
Z p
= ρdxi1 ∧ · · · ∧ dxik (p) · f ◦ ϕ−1α (p, t)dt · · · dt
1 n
R
 µ
n
Z
∂xi1 ∂xik p1 −1 ∂t
= ρ p1 · · · p dy ∧ · · · ∧ dy (p) · pn
f ◦ ϕβ (p, u) det ν
du1 · · · dun
∂y ∂y k R ∂u
Z
n

= τ (p) · g ◦ ϕ−1
β (p, u)du · · · du
1 n

ZR
n

= τ (p) · gdu1 ∧ · · · ∧ dun = π∗ ωβ (p)

Ep

where (p, u) = ϕβ ◦ ϕ−1

α (p, t) and the Jacobian of (ϕβ ◦ ϕα )(p) : R → R is
n n

 µ
∂t
det = det(gβα ) > 0
∂uν
by assumption, hence this gives the 4th equality. It remains to prove that π∗ is
independent of trivilizations. Let

ϕ0β : π −1 (Vβ ) → Vβ × Rn

be another family of oriented trivilizations for E. We would like to show π∗ (ϕ)ω =

π∗ (ϕ0 )ω on M . For any p ∈ M , choose α, β such that p ∈ Uα ∩ Vβ . If ϕ0β ◦ ϕ−1α
has positive determinant everywhere on Uα ∩ Vβ , then repeat the proof above where
we replace ϕβ by ϕ0β . Otherwise, we would get a minus sign in the 4th equality
above, but the 5th equality still holds because this is how we define integration on
an orientation reversing chart.

Proposition 6.86. Integration along fiber π∗ commutes with exterior differentiation

d.

Proof. Let (Uα , ϕα ) be a trivilization for E, and let ρα be a partition of unity sub-
ordinate to Uα (note that this time Uα ’s are actually on E instead of M ). Since for
109 CHAPTER 6. COHOMOLOGY

P
any ω ∈ Ωk+n
cv (E), we have ω = α ρα ω, and both π∗ and d are linear, it suffices to
prove the proposition for ρα ω ∈ Ωk+n
cv (Uα ). This means that we may as well assume
that E = M × R is trivial. We consider two situations: First, if ω is a top degree
n

form in the fiber coordinate:

ω = (π ∗ ϕ)f (x, t)dt1 ∧ · · · ∧ dtn

for some ϕ ∈ Ωk (M ), then we have

 Z 
dπ∗ ω = d ϕ f (x, t)dt · · · dt
1 n
Rn
Z Z
∂f 1
= (dϕ) f (x, t)dt · · · dt + (−1) ϕ ∧ dx
1 n k µ
µ
dt · · · dtn
∂x
 R R

n n

∗ k ∗ ∂f
= π∗ (π dϕ)f dt ∧ · · · ∧ dt + (−1) π ϕ ∧ µ dx ∧ dt ∧ · · · ∧ dt
1 n µ 1 n
∂x
= π∗ dω.

Now suppose ω does not have top degree in the fiber coordinate:

ω = (π ∗ ϕ)f (x, t)dti1 ∧ · · · ∧ dtir

where ϕ ∈ Ωk+n−r (M ). Then by definition dπ∗ ω = 0. On the other hand, we have

 
∗ ∂f
π∗ dω = (−1) k+n−r
π∗ π ϕ µ dt ∧ dt ∧ · · · ∧ dt
µ i1 ir
∂t

which is zero if dtµ ∧ dti1 ∧ · · · ∧ dtir 6= ±dt1 ∧ · · · ∧ dtn . Otherwise, we note that
Z
∂f µ
µ
dt ∧ dti1 ∧ · · · ∧ dtir = 0
Rn ∂t
−1
cv (M ), hence for each fixed x ∈ M , we have π (x) ∩ Supp ω is
because ω ∈ Ωk+n
compact. Hence ω has compact support on each fiber for a fixed x. Thus
Z +∞
∂f µ
µ
dt = f (..., +∞, ...) − f (..., −∞, ...) = 0.
−∞ ∂t

This concludes the proof.

Since π∗ d = dπ∗ , it is a cochain map. A cochain map descends to a well-defined
map on cohomology, which can be proven using exactly the same argument as in
the proof of Theorem 6.11. Hence, we get a well-defined map
k+n
π∗ : Hcv (E) → H k (M )

by π∗ [ω] = [π∗ ω], where on the right hand side the π∗ is the integration along fiber
cv (E) → Ω (M ).
map π∗ : Ωk+n k

Theorem 6.87 (Projection formula). Let π : E → M be an oriented real rank

cv (E) → Ω (M ) be the
n vector bundle over a smooth manifold M m and π∗ : Ωn+k k

integration along fiber map. Then:

6.8. THE THOM ISOMORPHISM 110

(a) For any ω ∈ Ωℓ (M ) and τ ∈ Ωn+k

cv (E), we have

π∗ (π ∗ ω ∧ τ ) = ω ∧ π∗ τ.

(b) Suppose in addition that M is oriented, and τ ∈ Ωk+n

cv (E), and ω ∈ Ωc
m−k
(M ).
If we orient E as in Lemma 6.77, then we have
Z Z
∗
π ω∧τ = ω ∧ π∗ τ.
E M

Proof. (a). Since two forms are the same iff they are the same at every point, we
may treat this problem locally and assume E = M × Rn . If

τ = π ∗ ϕf (x, t)dti1 ∧ · · · dtir (6.1)

with ϕ ∈ Ωn+k−r (M ) and r < n, then we see that

π∗ (π ∗ ω ∧ τ ) = π∗ (π ∗ (ω ∧ ϕ)f (x, t)dti1 ∧ · · · ∧ dtir ) = 0 = ω ∧ π∗ τ.

Now suppose
τ = (π ∗ ϕ)f (x, t)dt1 ∧ · · · ∧ dtn (6.2)
a top degree form in fiber coordinate with ϕ ∈ Ωk (M ). Then
Z
∗
π∗ (π ω ∧ τ ) = ω ∧ ϕ f (x, t)dt1 · · · dtn = ω ∧ π∗ τ.
Rn

(b). Let (ρα , Uα ) be a partition of unity for M , with ϕα : π −1 (Uα ) → Uα × Rn

oriented
P trivilizations. Then (π ∗ ρα , π −1 (Uα )) is a partition of unity for E. Then
τ = α (π ∗ ρα )τ . Therefore,
Z XZ
∗
π ω∧τ = π ∗ ω ∧ (π ∗ ρα )τ
E α π −1 (Uα )

and
Z XZ XZ XZ
∗
ω ∧ π∗ τ = ω ∧ π∗ ((π ρα )τ ) = ω ∧ ρα π ∗ τ = ω ∧ π∗ ((π ∗ ρα )τ )
M α M α M α Uα

by part (a). Hence again we reduced the problem to a local trivilization. That is, it
R that E = M × R again. With this setup, first suppose τ is given
n
suffices to assume
by (6.1). Then M ω ∧ π∗ τ = 0 because π∗ τ = 0. On the other hand,
Z Z
∗
π ω∧τ = π ∗ (ω ∧ ϕ) ∧ f (x, t)dti1 ∧ · · · ∧ dtir = 0
E E

because

deg(ω ∧ ϕ) = deg(ω) + deg(ϕ) = (m − k) + (k + n − r) = m + n − r > m

111 CHAPTER 6. COHOMOLOGY

because
P r < n by assumption. P Now assume τ is given by (6.2). Assume that
ω = |I|=m−k ωI dxI and ϕ = |J|=k ϕJ dxJ with ωI , ϕJ functions on M . Then we
have
Z Z Z
ω ∧ π∗ τ = (ωI dx ∧ ϕ) ·
I
f (x, t)dt1 · · · dtn
ZM R
Z
M n

= ωI ϕJ dx ∧ dx ·
I J
f (x, t)dt1 · · · dtn
ZM Rn

= ωI ϕJ dxI ∧ dxJ f (x, t)dt1 ∧ · · · ∧ dtn

ZM ×R
n

= π ∗ (ωI ϕJ dxI ∧ dxJ )f (x, t)dt1 ∧ · · · ∧ dtn

ZM ×R
n

= π ∗ (ωI dxI ) ∧ π ∗ (ϕJ dxJ )f (x, t)dx1 ∧ · · · dtn

ZM ×R
n

= π ∗ ω ∧ τ.
E

This completes the proof of the projection formula.

First, we utilize our previous work in the proof of Poincaré Lemma and get
Proposition 6.88. Integration along the fiber gives an isomorphism
k+n
π∗ : Hcv (M × Rn ) → H k (M ).

Proof. This is exactly like the proof of the Poincaré Lemma for compactly supported
cohomology (Theorem 6.51). We have already constructed R π∗ and showed that it
commutes with d. Now we define e ∈ Ωc (R ) with Rn e = 1, and define e∗ :
n n
∗
Ωk (M ) → Ωk+n
cv (M × R ) via e∗ : ϕ 7→ (π ϕ) ∧ e. Then we construct the chain
n

homotopy operator just like we did previously, and proof of Theorem 6.53 works
identically.
Now we are ready to prove the Thom isomorphism theorem in its most general
form.
Theorem 6.89 (General Thom isomorphism). Let E be an orientable vector bundle
of rank n over an orientable manifold of finite type. Then integration along fiber
gives an isomorphism
n+k
π∗ : Hcv (E) → H k (M ).
Proof. Note that E is an orientable manifold of finite type by Theorem 6.77 and
Corollary 6.80. The proof of Thom isomorphism is another example of the Mayer-
Vietoris argument. Let U, V be open cover of M . Using a partition of unity from
the base M , we obtain a partition of unity on E, subordinate to U 0 = π −1 (U ), V 0 =
π −1 (V ). We then see that the sequence

0 → Ω∗cv (E|U ′ ∪V ′ ) → Ω∗cv (E|U ′ ) ⊕ Ω∗cv (E|V ′ ) → Ω∗cv (E|U ′ ∩V ′ ) → 0

is exact. So this induces a long exact sequence of compact vertical cohomology

groups. We consider the diagram
6.8. THE THOM ISOMORPHISM 112

d∗
··· k+n
Hcv (E|U ′ ∪V ′ ) k+n
Hcv (E|U ′ ) ⊕ Hcv
k+n
(E|V ′ ) k+n
Hcv (E|U ′ ∩V ′ ) k+n+1
Hcv (E|U ′ ∪V ′ ) ···
π∗ π∗ π∗ π∗

··· H k (U ∪ V ) H k (U ) ⊕ H k (V ) H k (U ∩ V ) H k+1 (U ∪ V ) ···

d∗

The first two squares of this diagram are clearly commutative. We check the com-
mutativity for the third square. Recalling from Remark 6.36, we have that
π∗ d∗ ω = π∗ d((π ∗ ρU )ω) = π∗ ((π ∗ dρU ) ∧ ω) = dρU ∧ π∗ ω = d∗ π∗ ω
where we used the projection formula, i.e., part (a) of Theorem 6.87, and the fact
that both ω and dω are closed since they represent cohomology classes. So the
diagram is commutative.
Note that if U, V are good covers, then E|U ′ , E|V ′ , E|U ′ ∩V ′ are trivial and U 0 , V 0
are good covers on E, by Lemma 6.79 and Corollary 6.80. So in these cases, π∗ are
isomorphisms by Proposition 6.88. By the Five Lemma if the Thom isomorphism
holds for U 0 , V 0 , U 0 ∩ V 0 , then it holds for U 0 ∪ V 0 . The proof now proceeds by
induction on the cardinality of a good cover for the base (which is also the cardinality
of the corresponding good cover for E).
Definition 6.90 (Thom class). Let E be an orientable rank n vector bundle over an
n
orientable manifold M of finite type. The Thom isomorphism π∗ : Hcv (E) → H 0 (M )
determines a cohomology class
τ (E) = (π∗ )−1 (1) ∈ Hcv
n
(E)
called the Thom class of E. It is characterized by the property that π∗ τ (E) = 1.
That is, the Thom class τ (E), when integrated on each fiber (with co-orientation)
gives unit.
Proposition 6.91. The inverse Thom isomorphism (π∗ )−1 : H k (M ) → Hcv
n+k
(E) is
given by
(π∗ )−1 (α) = π ∗ α ∧ τ (E).
Proof. Since π∗ τ (E) = 1, by the projection formula (Theorem 6.87), one has
π∗ (π ∗ α ∧ τ (E)) = α ∧ π∗ τ (E) = α ∧ 1 = α.
This proves the claim.

6.8.1 Poincaré Duals and the Thom Class

Let S k be an oriented submanifold of an oriented manifold M m . Recall that the
Poincaré dual ηS is the unique cohomology class in H m−k (M ) characterized by
Z Z
ω= ω ∧ ηS
S M

for any ω ∈ Hck (M ).

It turns out that the Poincaré dual of a submanifold relates
to the Thom class of a bundle (normal bundle of S). We conclude this section by
explaining this connection.
113 CHAPTER 6. COHOMOLOGY

Definition 6.92. Let E, E 0 , E 00 be a sequence of vector bundles over M . A sequence

of vector bundles 0 → E → E 0 → E 00 is said to be exact if at every p ∈ M , the
sequence of vector spaces 0 → Ep → Ep0 → Ep00 → 0 is exact.
Definition 6.93 (Normal bundle). Let S k ⊂ M m be defined as above. Then the
normal bundle of S is the vector bundle N S over S defined by the exact sequence

0 → T S → T M |S → N S → 0

where T M |S is the restriction of the tangent bundle of M to S.

Np S S

Tp S
p

Figure 6.4: One should think of the normal bundle N S of S literally as being the
bundle whose fiber at each point p ∈ S is tangent to S, i.e., orthogonal to Tp S.
At every p ∈ S we have an exact sequence of vector spaces 0 → Tp S → Tp M →
Np S → 0, which means that Np S ∼ = Tp M/Tp S, i.e., Np S is isomorphic to the vector
space that is the complement of Tp S in Tp M , which is isomorphic to the orthogonal
complement of Tp S in Tp M , if we have a Riemannian metric on M .

By the tubular neighborhood theorem, which we will not prove, every submani-
fold S has a tubular neighbohood of T (S), and in fact T (S) is diffeomorphic to an
open neighborhood the normal bundle N S. The reader is encourage to refer to [1,
Theorem 6.24] and [7, p.67] for a proof, but we will not dive into this rabbit hole.
The orientation of T (S) and N S, when viewed as subsets of M , is given by the
induced orientation of M . However, we may orient the fiber of these bundles again
using co-orientation. For example, the normal bundle is oriented in such a way
that T S ⊕ N S has the orientation of T M |S .
Now, we apply the Thom isomorphism theorem to the normal bundle N S. Then
we get a sequence of maps
(π∗ )−1 j∗
H 0 (S) −−−−→ Hcv
m−k
(N S) −
→ H m−k (M )

where j∗ is extension by zero. Here, j∗ is defined because every form τ ∈ Hcv

m−k
(N S)
−1
has compact support on each fiber since π (p) ∩ Supp τ is compact for every p ∈ S.
Hence τ would vanish near the boundary of N S. Under the above sequence, we get
an element j∗ τ (N S).
6.8. THE THOM ISOMORPHISM 114

Theorem 6.94. Let S k ⊂ M m be a compact oriented submanifold of an oriented

manifold M . The Poincaré dual ηS ∈ H m−k (M ), which may be taken to have com-
pact support by Proposition 6.73, can be chosen to be the Thom class of the normal
bundle N S, i.e.,
ηS = j∗ τ (N S).
Proof. We only need to show that j∗ τ (N S) satisfies the defining property of Poincaré
dual. Let ω ∈ Ωkc (M ) be any closed form with compact support on M . Now,
Z Z
ω ∧ j∗ τ (N S) = ω ∧ τ (N S) (6.3)
M NS

because j∗ τ (N S) has support in N S. Let i : S → N S be the inclusion, regarded as

the zero section of the bundle π : N S → S. Since π is a deformation retraction of
N S onto S, we know that π ∗ , i∗ are inverse isomorphisms in cohomology H ∗ (N S) ∼ =
H ∗ (S). Hence, if we view ω (or rather its restriction) as a form in N S, then ω =
π ∗ i∗ ω + dα, i.e., ω differ from π ∗ i∗ ω by an exact form α ∈ Ωk−1 (N S). Hence
Z Z
ω ∧ τ (N S) = (π ∗ i∗ ω + dα) ∧ τ (N S). (6.4)
NS NS
m−k
Note that since S is compact, Hcv (N S) = Hcm−k (N S). So in fact τ (N S) has
compact support in N S. Applying Stokes’ theorem we see that
Z Z Z
dα ∧ τ (N S) = d(α ∧ τ (N S)) = α ∧ τ (N S) = 0
NS NS ∂(N S)

because ∂(N S) = ∅. Note that the dimension in the equation above works out:
dim ∂(N S) = dim(N S)−1 = m−1, and deg(α∧τ (N S)) = (k−1)+(m−k) = m−1.
Hence (6.4) reads
Z Z Z Z
∗ ∗ ∗
ω ∧ τ (N S) = π i ω ∧ τ (N S) = i ω ∧ π∗ τ (N S) = i∗ ω (6.5)
NS NS S S

by part (b) of the projection formula (Theorem 6.87), and by the definition of Thom
class that π∗ τ (N S) = 1. Putting (6.3), (6.4), and (6.5) together, we see that
Z Z
ω ∧ j∗ τ (N S) = i∗ ω.
M S

This is exactly the characterizing property of the Poincaré dual. Hence we conclude
that ηS = j∗ τ (N S).
Corollary 6.95. Suppose π : E → M is an oriented vector bundle over an oriented
compact manifold. Then the Poincaré dual η of the zero section of E can be chosen
to be the Thom class τ (E) of E.
Proof. Note that M embeds diffeomorphically as the zero section of E, hence there
is an exact sequence of vector bundles
0 → T M → T E|M → E → 0.
In other words, the normal bundle of M in E is E itself. Then the claim follows
immediatly from Theorem 6.94.
115 CHAPTER 6. COHOMOLOGY

Corollary 6.96. The support of the Poincaré dual ηS of a compact submanifold S

can be shrunk into any tubular neighborhood of S.

Proof. This is because the normal bundle of S in M is diffeomorphic to any tubular

neighborhood of S by the tubular neighborhood theorem.
Theorem 6.94, Corollary 6.95, and Corollary 6.96 will play crucial roles in the
foundation of intersection theory, as we will see in the next chapter.
7 | Differential Topology

The goal of this chapter is to prove some of the most important results in differential
topology using de Rham cohomology theory. The main tools that we will introduce
are degree theory and intersection theory.
The readers may recognize that degree and intersection theory are also the main
tools used to prove these results in standard differential topology textbooks like [7]
and [8]. But we want to stress that these tools arises naturally under the framework
of de Rham cohomology, whereas in the texts mentioned above it is not obvious
how this is true. Moreover, topological invariants like Euler characteristics and Lef-
schetz number are defined using intersection numbers directly in these texts without
involving (co)homology groups, thus readers who choose to continue their study in
algebraic topology may find it difficult to see why these definitions are equivalent
when they read a textbook like [3]. Hence the goal of this writing is to bridge the
gap between existing literature and hopefully provide a smooth transition between
the frameworks of differential and algebraic topology.
Unfortunately, the prerequisites needed to understand the proofs are by no ways
trivial. It would be ideal if the reader knows proofs of results in differential topol-
ogy like regular value theorem, Thom’s transversality theorem, Sard’s theorem, and
tubular neighborhood theorem. This is well introduced in Chapter 6 of [1] or in
standard texts like [7] and [8]. However, it is sufficient to just know the results
without knowing the proof. Still, readers might find it difficult to understand this
chapter because we quickly introduce relevant concept and omit the proofs of many
important results. This is completely normal, since our only goal here is to provide a
rapid introduction to this vast and deep subject. My goal here is not trying to make
the readers understand everything in this chapter, but to show them many interest-
ing and beautiful results. If you are fascinated by some of these results despite not
understanding a considerable amount of material, then I have successfully fulfilled
my duty. In that case, grab a book on differential topology and start reading! And
don’t worry: it will certainly be more well-written than anything you read here!

7.1 Degree Theory

Definition 7.1. A continuous map f : X → Y between topological spaces is said
to be proper if for any K ⊂ Y compact, f −1 (K) is a compact subset of X.
If f : M → N is a smooth map between oriented smooth manifolds, and ω ∈
Ωkc (N ),then f ∗ ω need not have compact support. For example, consider the pullback
of functions under the projection N × R → N . Hence, Ω∗c is not a functor on
117 CHAPTER 7. DIFFERENTIAL TOPOLOGY

the category of manifolds and smooth maps. However, if f is proper, then since
Supp f ∗ ω = f −1 (Supp ω), then we see that f ∗ ω ∈ Ωkc (M ). Moreover, one can show
that

Proposition 7.2. Ω∗c (and also Hc∗ ) is a contravariant functor in the category of
proper smooth manifolds (where the objects are smooth manifolds and the mor-
phisms are proper smooth maps) to the category of vector spaces.

Note that we know from Section 6.4.1 that Ω∗c is a covariant functor under
inclusions of open sets.
Now suppose that M m is a connected oriented manifold. It was shown in Corol-
lary 6.68 that Hcn (M ) ∼
= R.

Definition 7.3. If both M m , N m are oriented and connected manifolds of the same
degree, and f : M → N is a proper map, then f ∗ : Hcm (N ) → Hcm (M ) will be
a linear map between R, i.e., is multiplication by a number. This real number is
denoted by deg(f ), called the degree of f .

One can also remove the proper restriction from the definition above, but then
we must assume that both M and N are compact manifolds. But of course in this
case, Hck (M ) = H k (M ) since M is compact.

Remark 7.4. Note that the Poincaré duality theorem states that the isomorphism
Hcn (M ) → R is given by integration. Therefore, we have that for any ω ∈ Ωnc (M ),
Z Z
∗
f ω = (deg f ) · ω.
M N

We first prove a lemma on proper mappings:

Lemma 7.5. Let f : X → Y be a proper map between topological spaces. Assume Y

is locally compact and Hausdorff (in particular this holds if Y is a manifold). Then
f is a closed map.

Proof. Let C ⊂ X be closed. We want to show Y \ f (C) is open. So we let

y ∈ Y \ f (C). Then by assumption y has an open neighborhood V with compact
closure. Then f −1 (V ) is compact since f is proper. Let E = C ∩ f −1 (V ). Then E
is compact, hence f (E) is compact. Since Y is Hausdorff, f (E) is closed. Now let
U = V \ f (E). Then we claim that U is an open neighborhood of y disjoint from
f (C), which will complete the proof.
To show the claim, we note that because y ∈ V \ f (C), and f (E) ⊂ f (C) ∩ V ⊂
f (C), we conclude that y ∈ / f (E). Hence y ∈ V \ f (E) = U . And clearly U is open.
Finally we show U ∩ f (C) = ∅. Otherwise suppose y 0 ∈ U ∩ f (C), then there
exists x ∈ C such that f (x) = y 0 , and f (x) ∈ U ⊂ V ⊂ V , i.e., x ∈ E = C ∩ f −1 (V ).
But then y 0 = f (x) ∈ f (E), which is a contradiction for y 0 ∈ U = V \ f (E).
Then we can show:

Corollary 7.6. If M m , N m are orientable smooth manifolds and f : M → N is a

proper map that is not surjective, then deg(f ) = 0.
7.1. DEGREE THEORY 118

Proof. Let q ∈
/ f (M ) in N . By the lemma above, f (M )c is open, so we choose an
Ropen neighborhood V of q such that f (M ) ∩ V = ∅. Choose ω ∈ Ωm c (V ) such that
∗
N
ω = 1. But f ω = 0. Hence deg(f ) = 0.
Definition 7.7. Let F : M → N be a smooth map between manifolds. A point
q ∈ N is called a regular value of f if for every p ∈ F −1 (q) ⊂ M , the differential
dFp : Tp M → Tq N is surjective. If q is not a regular value of F , it is called a
singular value of F .
It is a deep theorem of Sard, proven in Chapter 6 of [1], that the set of critical
values in N has measure zero. Hence, almost every point in N is a regular value. In
particular, it is always possible to find a regular value of any smooth map F : M →
N on N .
Now we show how the definition of degree (usually defined in algebraic topology)
relates the degree in differential topology:
Theorem 7.8. Assume f : M m → N m a proper map between orientable manifolds,
and q ∈ N is a regular value of F , then we have
X
deg(f ) = sign(dfx ).
x∈f −1 (q)

Proof. By [1] Corollary 5.14, f −1 (q) is a properly embedded submanifold of M of

dimension 0. Since f is proper, f −1 (q) is compact. Hence it consists of finitely many
points (A discrete compact set cannot be infinite. Otherwise, because each singleton
in the set is open, we get an open cover of the set with no finite subcover, which
contradicts compactness). Say f −1 (q) = {p1 , ..., pk }. Then by the inverse function
theorem, there exists open neighborhoods V of q and Ui of qi for i = 1, ..., k, such
that
• Ui ∩ Uj = ∅ if i =
6 j;
`
• f −1 (V ) = ki=1 Ui ;
• f |Ui : Ui → V is a diffeomorphism.
R
Then pick ω ∈ Ωm c (V ) such that N ω = 1. In other words, ω represents a generator
in Hcm (N ). Then
Z X k Z X k Z X
∗ ∗
f ω= f ω= sign(dfx ) ω= sign(dfx ).
M i=1 Ui i=1 V x∈f −1 (q)

This proves the claim.

Example 7.9. Let i 6= m + 1 and consider the reflection
ri (x1 , ..., xm+1 ) = (x1 , ..., −xi , ..., xm+1 )
on S m . The north pole x = (0, ..., 0, 1) ∈ S m is a fixed point of ri . Since ri is a
diffeomorphism on a neighborhood of the north pole, x is a regular point of ri . Let
φ : B(0, 1) ∼
= Rm → S m be the chart


φ(x1 , ..., xm ) = x1 , ..., xm , (1 − x21 − · · · − x2m )1/2 .
119 CHAPTER 7. DIFFERENTIAL TOPOLOGY

By definition of orientation on S m , we know φ is orientation preserving. So by

definition, sign(dri )x = sign(dψ0 ) where ψ = φ−1 ◦ ri ◦ φ : B1 (0) → B1 (0) is the
induced map such that the diagram
ri
Sm Sm
φ φ

B1 (0) ψ
B1 (0)

commutes. By straightforward computation, ψ(x1 , ..., xm ) = (x1 , ..., −xi , ..., xm ).

Hence dψ0 = diag(1, ..., −1, ..., 1), which has determinant −1. Thus deg(ri ) = −1.
Now as a result of the above example, we prove the famous
Theorem 7.10 (Hairy Ball Theorem). There does not exist nowhere vanishing vec-
tor fields on S 2n .
Proof. We view S 2n ⊂ R2n+1 . Let X be a vector field on S 2n that is nonvanishing, for
the sake of contradiction. Then we may normalize at every point (i.e., by multiplying
a smooth function) and assume WLOG that |Xp | = 1 for all p ∈ S 2n . View both
p ∈ S 2n and Xp as vectors in R2n+1 . Consider the map

F : S 2n × [0, 1] → S 2n
(p, t) 7→ p cos(πt) + Xp sin(πt).

Note that the target is indeed in S 2n by straightforward computation using inner

product because |p| = |Xp | = 1, and p ⊥ Xp . Then F (·, 0) = idS 2n , while on the
other hand F (p, 1) = −p is the antipodal map. Thus we get a homotopy from the
identity to the antipodal map. Identity has degree 1, while the antipodal map is
a composition of reflection, and by the example above, has degree (−1)2n+1 = −1.
This is a contradiction because degree is a homotopy invariant by definition.

7.2 The Brouwer Fixed Point Theorem

In this section we use the tool of degree theory to prove the famous Brouwer fixed
point theorem. We start with a useful lemma:
Lemma 7.11. Let M m be a connected, orientable, compact manifold with boundary,
and suppose ∂M is also connected. Let X m−1 be a connected, oriented manifold
without boundary. If a smooth map f : ∂M → X can be extended to a smooth map
g : M → X, then deg(f ) = 0.
Proof.R Let i : ∂M → M be the inclusion. Then f = g ◦ i. Let ω ∈ Ωm−1 (X) such
that X ω = 1. Then dω = 0 since ω is already a top degree form on X. Then
Z Z Z Z Z
∗ ∗ ∗ ∗
deg(f ) = deg(f ) ω= f ω= ig ω= d(g ω) = g ∗ (dω) = 0
X ∂M ∂M M M

by Stokes’ theorem. This concludes the proof.

7.2. THE BROUWER FIXED POINT THEOREM 120

Corollary 7.12. Let M m be a connected, orientable, compact manifold with bound-

ary, and suppose ∂M has k connected component ∂j M , where j = 1, ..., k. Let X m−1
be a connected, oriented manifold without boundary. If a smooth map f : ∂M → X
can be extended to a smooth map g : M → X, then if we denote fj = f |∂j M , then

X
k
deg(fj ) = 0.
j=1

Proof. The proof is completely analogous to the proof of the previous lemma. Let
iRj : ∂j M → M be the inclusion. Then fj = g ◦ ij . Let ω ∈ Ωm−1 (X) such that
X
ω = 1. Then dω = 0 since ω is already a top degree form on X. Then

X
k X
k Z k Z
X
deg(fj ) = deg(fj ) ω= fj∗ ω
j=1 j=1 X j=1 ∂j M

k Z
X Z Z
= i∗j g ∗ ω = ∗
d(g ω) = g ∗ (dω) = 0
j=1 ∂j M M M

by Stokes’ theorem. This concludes the proof.

Now we prove

Theorem 7.13 (Brouwer Fixed Point Theorem). Let Dm ⊂ Rm be the closed unit
disk. Then every continuous map f : Dm → Dm has a fixed point.

Proof. We argue by contradiction. Suppose F0 : Dm → Dm has no fixed point.

Then since F0 is continuous, we have inf p∈Dm |p − F0 (p)|/3 > 0. Choose a positive
number r such that 0 < r < inf p∈Dm |p − F0 (p)|/3 > 0. By Whitney approximation
theorem (Theorem 1.30), there exists a smooth map F1 : Dm → Rm such that
|F0 (p) − F1 (p)| < r for all p ∈ Dm . Now consider the map F : Dm → Dm given by

F1 (p)
F (p) = .
1+r
Then F maps Dm to Dm by triangle inequality, and F is a smooth map from Dm
to itself. Moreover,

|F (p) − F0 (p)| ≤ |F (p) − F1 (p)| + |F1 (p) − F0 (p)|

< |F (p)|r + r
≤ 2r
< |F0 (p) − p|.

Thus |F (p) − p| ≤ |F (p) − F0 (p)| + |F0 (p) − p| < 2|F0 (p) − p|. Hence F : Dm → Dm
is smooth and does not have a fixed point.
Now consider the smooth map G : Dm → S m−1 defined by

p − F (p)
G(p) = .
|p − F (p)|
121 CHAPTER 7. DIFFERENTIAL TOPOLOGY

And we put g = G|S m−1 : S m−1 → S m−1 . Clearly G is an extension of g. So by

Lemma 7.11, we have deg(g) = 0.
On the other hand, consider the map

H : S m−1 × [0, 1] → S m−1

p − tF (p)
(p, t) 7→
|p − tF (p)|

is well-defined, because |F (p)| ≤ 1 and p 6= F (p), and moreover H is a homotopy

from the identity on S m−1 to g. Thus deg(g) = deg(IdS m−1 ) = 1. This is the desired
contradiction.

7.3 The Borsuk-Ulam Theorem

The goal of this section is to prove the Borsuk-Ulam theorem. The proof, however,
requires the following technical lemma. In fact, the proof of this lemma will probably
be the most technical (also the most clever) proof of this whole section (one can argue
that this is even the case for the whole book):

Lemma 7.14. Suppose f : S m → S m is a smooth odd map, which means for any p
we have f (−p) = −f (p), then deg(f ) is an odd number.

Proof. We induct on the number m.

For the base case m = 1, let f : S 1 → S 1 be an odd map. For simplicity, we
denote the upper, lower, left, right hemisphere of S 1 by S∧1 , S∨1 , S<1 , S>1 , respectively.
WLOG 1 that N = (0, 1) is a regular value of f . Then f −1 (±N ) is closed in S 1 ,
hence compact, thus has finitely many points. Again we may assume WLOG that
f ((±1, 0)) 6= N .2 Hence f ((±1, 0)) 6= −N as well by oddity. WLOG we assume
f ((1, 0)) ∈ S>1 .
−1
We claim that it suffices
P to show f (N ) has odd number of points: By Theorem
7.8 we have deg(f ) = x∈f −1 (N ) sign(dfx ). If the sum has more +1’s than −1’s,
cancel them out in pairs and we get a sum with only +1’s. If |f −1 (N )| is odd, then
after canceling points in pairs we get deg(f ) is odd. If the sum has more −1’s than
+1’s, we consider − deg(f ) and use the same argument. Hence our goal now is to
show that |f −1 (N )| is an odd number.
Since f is an odd map, we have that
1
|f −1 (N )| = |f −1 (±N )| = |f −1 (±N ) ∩ S∧1 |,
2
where the first equality is by symmetry of odd maps, and the second equality is
again by symmetry and the assumption that f ((±1, 0)) 6= N .
Now, points in f −1 (±N ) separates S∧1 into finitely many arcs (or segments). We
claim that adjacent segments must be sent to S>1 and S<1 respectively by f : To see
1
Otherwise, we may compose f with a rotation which is an orientation preserving diffeomor-
phism, then the claim follows from deg(f g) = deg(f ) deg(g) and Sard’s theorem.
2
Rotation argument again!
7.3. THE BORSUK-ULAM THEOREM 122

this, note that since N is a regular value of f , so for any x ∈ f −1 (±N ) ∩ S∧1 , we
have dfx : Tx S 1 → Tf (x) S 1 is surjective. Then for any vector v ∈ Tx S 1 , there exists
a curve γ such that γ 0 (0) = v and γ(0) = x. If f sends adjacent arcs separated by x
to the same hemisphere, then3 dfx (v) = (f ◦ γ)0 (0) = 0, since f ◦ γ changes direction
at f (x), i.e., at t = 0. Thus dfx is not surjective.
Therefore, adjacent arcs goes to different hemisphere S>1 or S<1 . Also note that
f ((1, 0)) and f (0, 1) belong to S>1 and S<1 (or the other way around) separately.
So we must start with arc mapping into S>1 , and then arc mapped into S<1 , and
alternate... and eventually end with arc mapping into S<1 . So there must be an even
number of arcs, i.e., odd number of endpoints. So |f −1 (±N ) ∩ S∧1 | = |f −1 (N )| is
odd. This proves the base case.
Now we prove the inductive case. To this end, suppose f : S m → S m is an odd
map. WLOG N = (0, ..., 0, 1) is a regular value of f . Denote S m−1 the equator of S m
by setting xm+1 = 0. By the same argument, we have f −1 (±N ) ∩ S∧m = {p1 , ..., pk }
and
1
|f −1 (N )| = |f −1 (±N )| = |f −1 (±N ) ∩ S∧m |.
2
By the inverse function theorem and possibly shrinking neighborhoods, we can
choose neighborhoods U of N , disjoint Ui of pi for i = 1, ..., k, such that f sends
each Ui diffeomorphically to U or −U , and such that Ui ⊂ S∧m for all i. Now put

a
k
M= S∧m − Ui
i=1

and
g = r ◦ π ◦ f |M
where π(x1 , ..., xm+1 ) = (x1 , ..., xm ) is the projection and r(x) = x/|x| retract from
B1 (0) \ {0} to S m−1 . We set g0 = g|S m−1 . Then by straightforward calculation, g0 is
an odd map. Now consider gi = g|∂Ui . Since we have a sequence of diffeomorphisms
f π r
∂Ui −
→ ±∂U −
→ π(∂U ) −
→ S m−1

we see that gi is a diffeomorphism for each i. Now:

• By construction, g0 : S m−1 → S m−1 is odd. So by inductive hypothesis,

deg(g0 ) is odd.

• gi : ∂Ui → S m−1 is a diffeomorphism for each i, so deg(gi ) = ±1 for each i.

Pk
• By Corollary 7.12, we have i=0 deg(gi ) = 0, since clearly g is a smooth
extension of gi ’s.

Thus we conclude that k is an odd number. This concludes the proof.

We conclude this section by proving
3
Note that since f is continuous, f will first map one arc with endpoint x to a segment on
a hemisphere, and then map the adjacent arc back to a portion of the same segment with the
opposite orientation.
123 CHAPTER 7. DIFFERENTIAL TOPOLOGY

Theorem 7.15 (Borsuk-Ulam Theorem). Let f : S m → Rm be continuous. Then

there exists a point p0 ∈ S m such that f (−p0 ) = f (p0 ).

Proof. For m = 1, define g(p) = f (−p) − f (p). We want to show g has a zero
on S 1 . Note g(1, 0) = f (−1, 0) − f (1, 0) and g(−1, 0) = f (1, 0) − f (−1, 0). Hence
g(1, 0) = −g(−1, 0). If g(1, 0) = 0 we are done, otherwise g(S 1 ) is connected and
contains both positive and negative numbers on the real line, so 0 ∈ g(S 1 ). This
proves the base case.
Now for m 6= 1, suppose f : S m → Rm is continuous, and suppose f (−p) 6= f (p)
for all p ∈ S m . Let ε = inf p∈S m |f (−p) − f (p)| > 0. Then take an ε/4-smooth
approximation f 0 of f by Whitney approximation theorem (Theorem 1.30). If there
exists a p such that f 0 (−p) − f 0 (p) = 0, then
ε ε
|f (−p) − f (p)| ≤ |f (p) − f 0 (p)| + |f 0 (p) − f 0 (−p)| + |f 0 (−p) − f (−p)| ≤ + <ε
4 4
which contradicts the choice of ε. Therefore we have found a smooth map f 0 : S m →
Rm , such that f 0 (−p) 6= f (p) for all p ∈ S m . Now let g : S m → S m−1 be defined by

f 0 (p) − f 0 (−p)
g(p) = .
|f 0 (p) − f 0 (−p)|

By definition, g̃ = g|S m−1 is a map from S m−1 to itself, and is an odd map by
construction. Therefore by the lemma above, deg(g̃) is odd. Hence in particular
deg(g̃) 6= 0. Now
g|S∧m : S∧m → S m−1
is clearly an extension of g̃, as ∂S∧m = S m−1 . Thus by Lemma 7.11, we have deg(g̃) =
0, which is the desired contradiction.

7.4 Intersection Theory

Definition 7.16. Let S s , Lℓ be oriented submanifolds of an oriented manifold N n
with complementary dimension, i.e., s + ℓ = n. We say that S and L intersect
transversally, written S ⋔ L, if for every p ∈ S ∩ L, one has Tp S ⊕ Tp L = Tp M
as vector spaces (but we are not concerned with whether the orientations on both
sides of the equation agrees or not, we just need Tp S and Tp L to span Tp N ). See
Figure 7.1 for examples.

Now we introduce a notion that allows us to count signed intersections of two

submanifolds:

Definition 7.17. Let S s , Lℓ be oriented submanifolds of an oriented manifold N n ,

where n = s + ℓ and S ⋔ L.

• If p ∈ S ∩ L, we define the local intersection number ϵ(p, S, L) to be +1 if

the direct sum orientation Tp S ⊕ Tp L coincides with the orientation of Tp M ;
and we define ϵ(p, S, L) = −1 if otherwise.
7.4. INTERSECTION THEORY 124

L
L

Tp L

p S p S
Tp S Tp L Tp S

Figure 7.1: We have two examples of intersecting submanifolds S, L in R2 with

complementary dimension. On the left, S and L intersect transversally because their
tangent spaces span a two dimensional plane, which is exactly Tp R2 ; However, on
the right, S and L does not intersect transversally because their tangent spaces are
the same one dimensional subspaces, so the tangent spaces do not span the whole
Tp R2 . Thus one should think of having transversal intersection as the property of
having minimal dimension of intersection, because transversality is equivalent to
dim Tp (S ∩ L) = 0.

• If S is compact, then L ∩ S is a finite set of points (S ∩ L has dimension 0,

hence a set of points, which is a closed subset of a compact set S). We define
the intersection number of S and L to be
X
I(S, L) = ϵ(p, S, L).
p∈S∩L

Note that ϵ(p, S, L) = (−1)sℓ ϵ(p, L, S), and similarly I(S, L) = (−1)sℓ I(L, S).
Now we present the key result of this section:
Theorem 7.18. Suppose that S s , Lℓ are closed, oriented submanifolds of an oriented
manifold N n of finite type (i.e., has a finite good cover), such that
• S is compact;
• s + ℓ = n;
• L ⋔ S.
Now let ηS be the compactly supported closed form representing the Poincaré dual of
S (by Proposition 6.73) in N , and let ηL be a closed form representing the Poincare
dual of L in N . Then we have
Z
I(S, L) = ηS ∧ ηL .
M

Proof. For simplicity, we write ϵ(p) := ϵ(p, S, L). Since L ⋔ S, for all p ∈ S ∩ L,
we can find4 a local coordinate (x1 , ..., xn ) defined on an open neighborhood Up of p
such that
Up ∩ S = {xs+1 = · · · = xs+ℓ = 0}
4
Short proof: Since Tp L ⊕ Tp S = Tp N , we can choose v1 , ..., vℓ ∈ Tp L and vℓ+1 , ..., vℓ+s ∈ Tp S.
Then give N a Riemannian metric, and let the curve xi (λ) = expp (λvi ). Then the coordinates
x1 , ..., xs+ℓ will do the job. For a more detailed and rigorous proof, see Lemma 18 on p.119 of [12].
125 CHAPTER 7. DIFFERENTIAL TOPOLOGY

and
Up ∩ L = {x1 = · · · = xs = 0}.
The orientation from on S ∩ Up is given by dx1 ∧ · · · dxs , and the orientation form on
L ∩ Up is given by dxs+1 ∧ · · · ∧ dxn . Then the wedge dx1 ∧ · · · ∧ dxn will correspond
to direct sum orientation. Hence the orientation form on Up is given by

ϵ(p)dx1 ∧ · · · ∧ dxn . (7.1)

By shrinking Up ’s if necessary, we may assume Up ∩ Uq = ∅ if p 6= q in L ∩ S. Now

we construct a tubular neighborhood (See Figure 7.2) N of S in N such that for all
p ∈ L ∩ S, we have

Np = (L ∩ Up ) ∩ N =: Lp

N S

Up
N
L
Lp

Figure 7.2: Construction of the tubular

` neighborhood N of S. We may choose
N thin enough such that L ∩ N = p∈S∩L Np . This can be done because we can
construct N arbitrarily as long as it is diffeomorphic to the normal bundle of S. In
this case, N = M × M and S = ∆ the diagonal of M , where M is a manifold. But
the construction works in general as well.

where the equality above is to be understood as equality of sets. But as oriented

manifolds, they have different orientations: By definition Lp is a subset of L, hence
its orientation is the induced orientation from the manifold L. Np is the fiber of N .
By convention, the orientation of Np (namely the co-orientation) is induced by the
direct sum Tp N = Tp S ⊕ Np S where N S is the normal bundle of S. By the tubular
neighborhood theorem, there is an orientation preserving diffeomorphism from N S
to N . That is, Np has the orientation such that direct sum orientation of Tp S ⊕ Np
agrees with that of Tp N . Since by (7.1), the orientation of N is given by

Tp N = ϵ(p)Tp S ⊕ Tp L

we have that
Lp = ϵ(p)Np .
7.4. INTERSECTION THEORY 126

Now, by Corollary 6.96, we may assume ηS ∈ Ωℓc (N ). In addition, We may choose

N thin enough such that a
L∩N = Np .
p∈S∩L

Then we have, by the definition of Poincaré dual and discussion above, that
Z Z X Z X Z X
ηS ∧ ηL = ηS = ηS = ϵ(p) ηS = ϵ(p)
M L p∈S∩L Lp p∈S∩L Np p∈S∩L

where in the last equality we used Corollary 6.95.

Example 7.19. Let N = T 2 = S 1 × S 1 be a torus, and S = S 1 × b, L = a × S 1
where p = (a, b) ∈ M is a fixed point. See Figure 7.3.

ϕ
T2

φ
S
L

θ
2
[0, 2π)
Figure 7.3: Let S, L be two submanifolds of T 2 with orientations indicated as in
the figure, and let φ : [0, 2π)2 → T 2 be the diffeomorphism parameterizing the torus
as shown in the figure. It is easy to see that φ is orientation preserving, where we
define the orientation of T 2 as follows: View T 2 as a submanifold of R3 , and let
v, w ∈ Tp T 2 . We declear the ordered basis v, w ∈ Tp T 2 to be positive iff v, w, n is a
positive basis for R3 , where n is the normal to T 2 at p.

We see that I(S, L) = ϵ(p, S, L) = +1 from Figure 7.3 where x = (a, b). We
know that ηS satisfies Z
ηS = 1,
Np S

where N S is the fiber of a tubular neighborhood of S at p by Theorem 6.94. Let

φ : [0, 2π)2 → T 2 be an orientation preserving diffeomorphism parameterizing the
torus, as Figure 7.3 suggests. We take ρ : [0, 2π) → R to be a smooth compactly
supported function with Z 2π
ρ(λ)dλ = 1.
0
Then take
φ∗ ηS = ρ(ϕ)dϕ φ∗ ηL = −ρ(θ)dθ.
Then we see that
Z Z Z
∗ ∗
η S ∧ ηL = φ η S ∧ φ ηL = ρ(θ)ρ(ϕ)dθ ∧ dϕ = 1.
T2 [0,2π)2 [0,2π)2
127 CHAPTER 7. DIFFERENTIAL TOPOLOGY

Again this is consistent with the result we get from counting I(S, L).
One can check that our choice of ηS , ηL are indeed the Poincaré dual of S and L.
To see this, let α ∈ Ω1 (T 2 ) be any closed 1-form. Write φ∗ α = F (θ, ϕ)dθ+G(θ, ϕ)dϕ.
Then we see that φ∗ (α ∧ ηS ) = F (θ, ϕ)ρ(ϕ)dθ ∧ dϕ. Since Supp ηS can be chosen to
be sufficiently small as long as it is contained in a thin tubular neighborhood of S,
and ρ has mass 1, we see that we can make ηS more and more concentrated on S:
think of it as a dirac delta function in the limit. If we denote ϕ0 the ϕ coordinate of
φ−1 (S) in [0, 2π)2 , we can make ρ(ϕ) → δ(ϕ0 ), then it is clear that
Z Z 2π Z 2π Z 2π Z
∗ 0
φ (α ∧ ηS ) = F (θ, ϕ)ρ(ϕ)dθdϕ = F (θ, ϕ )dθ = φ∗ α.
[0,2π)2 0 0 0 φ−1 (S)

Under the diffeomorphism φ, we see that on T 2 the integral equality above becomes
Z Z
α ∧ ηS = α,
T2 S

which is exactly the defining integral equation of ηS , similarly we can check ηL is

indeed the Poincaré dual of L.

7.5 Lefschetz Fixed Point Theorem

In this section, our goal is to prove the Lefschetz fixed point theorem for compact
manifold. The proof of Lefschetz fixed point theorem is not hard using the tools
of intersection theory we developed in the previous section. However, we need a
preliminary result to prove the Lefschetz fixed point theorem, which involves some
technical calculations.
First, we recall the Künneth formula from Theorem 6.69.Künneth theorem says
that if we have M, N two orientable manifolds of finite type, then
M
H k (M × N ) ∼
= H p (M ) ⊕ H q (N ).
p+q=k

In particular, if we define πM : M ×N → M and πN : M ×N → N be the projection

maps onto each factors, then the map
ψ : H ∗ (M ) ⊗ H ∗ (N ) → H ∗ (M × N )
∗ ∗
ω ⊗ ϕ 7→ πM ω ⊗ πN ϕ
is an isomorphism of cohomology rings.
Now the preliminary result we want to study concerns the following situation:
Let M m be an oriented closed (compact and has no boundary) manifold, and let
f : M → M be a smooth map. We define the graph of f to be
Γf = {(x, f (x)) : x ∈ M } ⊂ M × M.
Then, what is the Poincaré dual ηΓf ∈ H m (M × M )? Note that since Γf is diffeo-
morphic to M via the projection onto M , codimension of Γf in M must be m.
Note that by the Mayer-Vietoris argument, if M m is of finite type, then H j (M )
is finite dimensional for each j by Proposition 6.61.
7.5. LEFSCHETZ FIXED POINT THEOREM 128

Proposition 7.20. Let βj = dim H j (M ) for each j. Let ωij be a basis of H j (M ),

where 1 ≤ i ≤ βj . Let θim−j be the dual basis of ωij for 1 ≤ i ≤ βm−j = βj under
Poincaré duality. Let πi : M × M → M be the projection onto the 1st and 2nd
factor where i = 1, 2. f ∗ : H j (M ) → H j (M ) be given by
X
f ∗ (ωij ) = Aki j
j ωk .
k

Then X
∗ j ∗ m−j
ηΓ f = j (π1 ωi ∧ π2 θk
(−1)j Aik ). (7.2)
i,j,k

Proof. Since all the cohomology groups are finite dimensional vector spaces, under
Poincaré duality and dual vector space isomorphism, we have

= Hcm−j (M )∗ ∼
H j (M ) ∼ = H m−j (M )∗ ∼
= H m−j (M ).

Under these isomorphisms, we see that we have the relation

Z
ωij ∧ θkm−j = δik
M

for all 1 ≤ i, k ≤ βj . By Künneth theorem, the set

{π1∗ ωij ∧ π2∗ θkm−j : 0 ≤ j ≤ m, 1 ≤ i, k ≤ βj }

is a basis of H m (M × M ).
Now we begin our calculation:
Z
(π1∗ ωij ∧ π2∗ θkm−j ) ∧ (π1∗ θsm−u ∧ π2∗ ωtu )
M ×M
Z
= (−1) m(m−j)
π1∗ (ωij ∧ θsm−u ) ∧ π2∗ (ωtu ∧ θkm−j )
ZM ×M Z
= (−1) m(m−j)
ωi ∧ θs
j m−u
ωtu ∧ θkm−j
MZ MZ

= (−1) m(m−j)
δju ωi ∧ θs
j m−j
ωtj ∧ θkm−j
M M
m(m−j)
= (−1) δju δis δtk .

In the calculation above, the first equality can be seen by swapping the 2nd term
in the wedge to the last; the 3rd equality can be seen by considering integrating on
local coordinate of product M × M ; the 4th equality can be seen by the following
argument: Since all forms
R are compactly supported, if one coordinate or more is
missing, say x , then φ(U ) (· · · )dxn = 0 where φ is some local chart and U ⊂ M is
n

some open set containing support of the form in question (if it is not contain in one
chart, use a partition of unity argument).
Now we write ηΓf as a linear combination of basis elements:
X
∗ j ∗ m−j
ηΓ f = j (π1 ωi ∧ π2 θk
cik )
i,j,k
129 CHAPTER 7. DIFFERENTIAL TOPOLOGY

and denote the expression on the right hand side by ω. For fixed u, s, t, by the
definition of Poincaré dual we have
Z Z
∗ m−u ∗ u
(π1 θs ∧ π2 ωt ) ∧ ω = i∗ (π1∗ θsm−u ∧ π2∗ ωtu ) (7.3)
M ×M Γf

where i : Γf → M × M denote the inclusion map. We now calculate the left hand
side and the right hand side of (7.3) separately. The left hand side is
X Z
(−1) m2
cik
j (π1∗ ωij ∧ π2∗ θkm−j ) ∧ (π1∗ θsm−u ∧ π2∗ ωtu )
i,j,k M ×M
2
X
= (−1)m cik
j δju δis δtk (−1)
m(m−j)

i,j,k
m2 +m(m−u)
= (−1) cst
u
= (−1)−mu cst
= (−1)mu cst
u.

Now recall that we set X

f ∗ (ωij ) = Aki j
j ωk .
k

Let ϕ : M → Γf be the orientation preserving diffeomorphism ϕ(x) = (x, f (x)).

Note that π1 ◦ i ◦ ϕ = Id and π2 ◦ i ◦ ϕ = f . Then the right hand side of (7.3) is
given by
Z Z
∗ ∗ ∗ m−u ∗ u
ϕ i (π1 θs ∧ π2 ωt ) = θsm−u ∧ f ∗ ωtu
M
X
M Z
= Au kt
θsm−u ∧ ωku
k M
X
= (−1)u(m−u) Akt
u δsk
k
u(m−u)
= (−1) Ast
u.

Equating the left hand side and right hand side of (7.3) gives

(−1)mu cst
u = (−1)
u(m−u) st
Au .

Hence
−u 2
cst
u = (−1) Ast u st
u = (−1) Au .

This proves (7.2).

As an easy corollary, we obtain the Poincare dual of the diagonal ∆ in M × M :
Corollary 7.21. Assuming conditions and notations in the previous proposition.
Let ∆ ⊂ M × M be the diagonal, i.e., ∆ = {(x, x) : x ∈ M }. Then
X
η∆ = (−1)j (π1∗ ωij ∧ π2∗ θim−j ). (7.4)
j
7.5. LEFSCHETZ FIXED POINT THEOREM 130

Proof. Use (7.2) where f : M → M is the identity. Then Γf = ∆. Therefore

f ∗ = id : H j (M ) → H j (M ) hence all Aki
j = δik .

Definition 7.22. Let f : M → M be a smooth map from a compact oriented

manifold M to itself. We define the Lefschetz number L(f ) of f to be
X

L(f ) = (−1)j Tr f ∗ : H j (M ) → H j (M ) .
j

Lemma 7.23. Let i : ∆ → M × M be the inclusion map, and f : M → M a smooth

map where M is an oriented compact manifold. Then
Z
ηΓf = L(f ).
∆

More specifically, we have

Z X

i∗ η Γ f = (−1)j Tr f ∗ |H j (M ) .
∆ j

Proof. Consider the orientation preserving diffeomorphism ϕ : M → ∆ where

ϕ(x) = (x, x). Then πi ◦ i ◦ ϕ : M → M are both identity map where i = 1, 2.
Hence by our previous calculation,
Z Z
∗
i ηΓ f = ϕ∗ i∗ ηΓf
∆
X
M
X Z
= (−1) j ik
Aj ϕ∗ i∗ π1∗ ωij ∧ ϕ∗ i∗ π2∗ θkm−j
j i,k M
X X Z
= (−1)j Aik
j ωij ∧ θkm−j
j ik M
X X
= (−1)j Aik
k δik
j i,k
X X
= (−1)j Aiij
j i
X

= (−1) Tr f ∗ |H j (M ) .
j

This concludes the proof.

Definition 7.24. Let f : M → M be a smooth map between compact manifolds.
Then f is said to be Lefschetz or a Lefschetz map if Γf ⋔ ∆ in M × M .
Now we are ready to prove
Theorem 7.25 (Lefschetz Fixed Point Theorem). Let M m be a compact, oriented,
smooth manifold, and f : M → M a smooth map. Then
Z
ηΓf ∧ η∆ = L(f ).
M ×M
131 CHAPTER 7. DIFFERENTIAL TOPOLOGY

Hence, if Γf ⋔ ∆ (i.e., f is a Lefschetz map), then

I(Γf , ∆) = L(f ).

In particular, if L(f ) 6= 0, then f has at least one fixed point.

Proof. By the definition of Poincaré dual, and the previous lemma, we have
Z Z
η Γ f ∧ η∆ = i∗∆ ηΓf = L(f ).
M ×M ∆

The second part of the theorem follows directly from Theorem 7.18. Finally, if f
has no fixed point, then Γf ∩ ∆ = ∅. Hence by the definition of intersection number,
I(Γf , ∆) = 0. Thus L(f ) = 0.

Remark 7.26. Therefore, we see an alternative definition of L(f ) for a smooth map
f : M → M between a compact oriented manifold, which is L(f ) := I(Γf , ∆). Note
that some books like [7] defined L(f ) = I(∆, Γf ). But we insist on defining L(f )
to be I(Γf , ∆) because it agrees with the cohomological definition, i.e., that L(f ) is
the alternating trace of f ∗ on cohomology groups, where the other convention is off
by a sign when dimension is odd.

We conclude this section with a final result, which relates Lefschetz number L(f )
with its local Lefschetz number Lx (f ) at every fixed point x.

Lemma 7.27. If f : M m → M m is a Lefschetz map between compact manifolds,

then at every fixed point x of f (guaranteed to exist at least one by Lefschetz fixed
point theorem), 1 is not an eigenvalue of dfx . In other words, det(I − dfx ) 6= 0.

Proof. Let (x, x) = p ∈ Γf ∩∆ ⊂ M ×M . Since f is Lefschetz, we have Tp Γf +Tp ∆ =

Tp (M × M ). If 1 is an eigenvalue of dfx , then say dfx (v) = v for some v ∈ Tx M .
Then (v, v) ∈ Tp Γf ∩ Tp ∆. But this contradicts the transersality condition since
dim(Γf ) = dim(∆) = m, hence dim(Tp Γf ∩ Tp ∆) = 0.

This motivates the following definition:

Definition 7.28. Let f : M → M be a Lefschetz map from a oriented, compact

manifold to itself. Let x be a fixed point of f . Then define the local Lefschetz
degree Lx (f ) of f at x by

Lx (f ) = sign det(I − dfx ).

We conclude this section by proving

Theorem 7.29. Let f : M m → M m be a Lefschetz map and M a compact, oriented

manifold. Then f has finitely many fixed points, and moreover,
X
L(f ) = Lp (f ).
p∈Fix(f )
7.5. LEFSCHETZ FIXED POINT THEOREM 132

Proof. We have dim(Γf ∩ ∆) = 0 by transversality. Since M is compact, Γf ∩

∆ is a compact 0-dimensional manifold, which must have finitely many points by
compactness argument again. This shows that Fix(f ) is finite. To prove the final
statement, note that by the Lefschetz fixed point theorem, we have
Z Z X
∗
L(f ) = i∆ ηΓf = ηΓf ∧ η∆ = I(Γf , ∆) = ϵ(p, Γf , ∆). (7.5)
∆ M ×M p∈Fix(f )

Hence it suffices to prove that ϵ(p, Γf , ∆) = Lp (f ) for every fixed point p. Let
v1 , ..., vm be a positively oriented basis of Tp M . Then v1 ⊕ v1 , ..., vm ⊕ vm is a
positively oriented basis of T(p,p) ∆, and v1 ⊕ dfp (v1 ), ..., vm ⊕ dfp (vm ) a positively
oriented basis of T(p,p) Γf . Then a positively oriented basis of T(p,p) Γf ⊕ T(p,p) ∆ is

v1 ⊕ dfp (v1 ), ..., vm ⊕ dfp (vm ), v1 ⊕ v1 , ..., vm ⊕ vm .

We now examine the orientation of this basis in T(p,p) (M × M ) Let ω be a (positive)

orientation form on M . Then let π1 : M × M → M and π2 : M × M → M be the
respective projections. Then a (positive) orientation form on M × M is given by
α = π1∗ ω ∧ π2∗ ω ∈ Ω2m (M ). Since α is alternating, we see that

α(v1 ⊕ dfp (v1 ), ..., vm ⊕ dfp (vm ), v1 ⊕ v1 , ..., vm ⊕ vm )

= α(0 ⊕ (dfp − I)(v1 ), ..., 0 ⊕ (dfp − I)(vm ), v1 ⊕ v1 , ..., vm ⊕ vm )
= α(0 ⊕ (dfp − I)(v1 ), ..., 0 ⊕ (dfp − I)(vm ), v1 ⊕ 0, ..., vm ⊕ 0)
2
= (−1)m α(v1 ⊕ 0, ..., vm ⊕ 0, 0 ⊕ (dfp − I)(v1 ), ..., 0 ⊕ (dfp − I)(vm ))
= (−1)m α(v1 ⊕ 0, ..., vm ⊕ 0, 0 ⊕ (dfp − I)(v1 ), ..., 0 ⊕ (dfp − I)(vm ))
= α(v1 ⊕ 0, ..., vm ⊕ 0, 0 ⊕ (I − dfp )(v1 ), ..., 0 ⊕ (I − dfp )(vm )),

where in the second equality we used the fact that I − dfp : Tp M → Tp M is an

isomorphism by the lemma above, hence vi ’s are linear combinations of (I−dfp )(vj )’s.
Then the second equality follows from multilinearity and alternating property of α.
Since we have the identification T(p,p) (M × M ) ∼= Tp M ⊕ Tp M and the first m
terms is a positive basis for Tp M , the final result depends on whether I − dfp is
orientation preserving or not. That is, we have that

ϵ(p, Γf , ∆) = sign det(I − dfp ) = Lp (f ).

This is what we wanted to prove.

By the way, this section provides a full solution to Exercise 11.26 in [2], in case
you haven’t noticed (although the original problem has a typo, as the next remark
suggests).

Remark 7.30. One would note that the local Lefschetz degree in our definition
Lx = sign det(I − dfx ) is off by a factor (−1)m compared to literature like [7],
where they define Lx = sign det(dfx − I). P Thisj is because we

want to stick to
∗
the definition where L(f ) is definedPto be j (−1) Tr f |H j (M ) , and we would also
like to have the relation L(f ) = x∈Fix(f ) Lx (f ). These two requirements forces
133 CHAPTER 7. DIFFERENTIAL TOPOLOGY

Lx (f ) = sign det(I − dfx ), which disagrees with the convention like [7] when the di-
mension is odd. These different sign conventions can be quite annoying, and I think
Bott and Tu missed this in their Exercise 11.26 on p.129 of [2]. In fact, in [12], it is
calculated on p.123 that the local Lefschetz degree must be Lx (f ) = sign det(I − dfx )
for Theorem 7.29 to hold. See also the discussion in this MSE post.

7.6 Euler Characteristics and Poincaré-Hopf The-

orem
Now we introduce the notion of self-intersection numbers:

Definition 7.31. Let N 2m be a smooth, oriented, compact manifold. For any

compact submanifold S m ⊂ N 2m of dimension m, we define the self-intersection
number I(S, S) to be Z
I(S, S) = ηS ∧ ηS .
N

Remark 7.32. Since the wedge product is alternating, one has dxi ∧ dxi = 0 for
all i. Therefore one may wonder why don’t we immediately get ηS ∧ ηS = 0. This
is not necessarily true. The reason is that ηS may not be a pure wedge, it can also
be linear combination of them. For example, take ηS = dx1 ∧ dx2 + dx3 ∧ dx4 , then
ηS ∧ ηS = 2dx1 ∧ dx2 ∧ dx3 ∧ dx4 .

By Thom transversality theorem, as proved in [7] or [1], in such a situation, one

can always find a smooth homotopy F : S × [0, 1] → M , such that

• For all t, Ft is an embedding;

• F0 = i : S → M is the inclusion map;

• Ft ⋔ S, i.e., Ft (S) ⋔ S.

We denote St = Ft (S). Then one can show

Proposition 7.33. ηSt = ηS in H m (N ).

Proof. We prove the following more general statement: Let S k , Lk ⊂ M m be sub-

manifolds such that there exists a smooth map F : S ×[0, 1] → M , with the property
that F0 (S) = S and F1 (S) = L. Then we claim that ηS = ηL .
To see this, take any representative ω of Hck (M ). Then dω = 0. Now by Stokes’
theorem, one has
Z Z Z Z
ω− ω=± ω=± dω = 0.
S L ∂F (S×[0,1]) F (S×[0,1])

Thus we have Z Z
i∗S ω = i∗L ω
S L
7.6. EULER CHARACTERISTICS AND POINCARÉ-HOPF THEOREM 134

for any representative ω in Hck (M ). Then by the definition of Poincaré dual:

Z Z Z
∗ ∗
iS ω = iL ω = ω ∧ ηL .
S L M

Since ω is arbitrary, we have ηS = ηL , as desired.

We can now define
X
I(S, S) = I(St , S) = ϵ(p, St , S)
p∈St ∩S

for every t ∈ [0, 1]. Hence our most important tool in intersection theory, Theorem
7.18, can be generalized to self-intersection number again.

Definition 7.34. Let M be a smooth manifold of finite type. Then the Euler
characteristic χ(M ) of M is defined to be the number
X X
χ(M ) = (−1)j rank H j (M ) = (−1)j βj .
j j

Then we have the following easy corollary:

Corollary 7.35. Let M be a compact manifold. Then we have I(∆, ∆) = χ(M ).

Proof. Take f : M → M to be idM in the Lefschetz fixed point theorem, and note
that
Z
I(∆, ∆) := η∆ ∧ η∆
M ×M
= L(idM )
X

= (−1)j Tr idM : H j (M ) → H j (M )
j
X
= (−1)j rank H j (M ).
j

where the 2nd equality is by the first part of the Lefschetz fixed point theorem. This
concludes the proof.
We immediately get

Corollary 7.36. If M 2n+1 is a compact manifold of odd dimension, then χ(M ) = 0.

Proof. Just note that I(∆, ∆) = (−1)(2n+1) I(∆, ∆) = −I(∆, ∆) if we swap the two
2

inputs by the definition of intersection number. Then χ(M ) = I(∆, ∆) = 0 by the

above corollary.

Remark 7.37. The compactness assumption in the corollaries above cannot be

removed, because in the proof of Corollary 7.35, we used Theorem 7.18, which
requires ∆ to be compact. As a counterexample, we note that χ(R1 ) = 1 which is
nonzero.
135 CHAPTER 7. DIFFERENTIAL TOPOLOGY

Remark 7.38. Textbooks like [7] uses I(∆, ∆) as the definition of the Euler char-
acteristic χ(M ). The discussion above gives the reason why this definition is com-
patible with the definition using cohomology, which is more common in algebraic
topology literature.
We conclude this section by giving a proof of the Poincaré-Hopf theorem, which
requires almost no additional work other than what we have already developed.
Let M be a compact manifold, X a smooth vector field on M . For sufficiently
small t, let φt : M → M be the flow generated by X. Then it is not hard to see that
x ∈ M is a zero of the vector field X iff φt fixes x. Then from dφtx : Tx M → Tx M
we get a linear map
d
Ax = (dφtx ) .
dt t=0

Definition 7.39. Let M, X be defined as above. If X(x) = 0, and det(Ax ) 6= 0,

then we call x a non-degenerate zero of X. We define the index ι(X, x) of X at
x to be
ι(X, x) = sign(det Ax ).
In Chapter 3 of [7], it is showed that ι(X, x) is the degree of the associated Gauss
map of X near the non-degenerate zero x. Now we prove the celebrated
Theorem 7.40 (Poincaré-Hopf Theorem). Let M be a compact oriented manifold
and X a smooth vector field on M that has only non-degenerate zeros x1 , ..., xk .
Then
Xk
ι(X, xj ) = χ(M ).
j=1

Proof. For t > 0 sufficiently small, we note that

det dφtx − I = det tAx + O(t2 ) = tm det(Ax ) + O(tm+1 ).
Hence Lx (φt ) = sign det(I − dφtx ) = (−1)m sign(det Ax ). Therefore, we have
X
k X
(−1)m sign(det Axj ) = Lx (φt ) = L(φt )
j=1 x∈Fix(φt )

by Theorem 7.29. But φt is obviously homotopic to idM . Hence L(φt ) = L(idM ) =

χ(M ). This gives
Xk
(−1)m sign(det Axj ) = χ(M ).
j=1

Now if m is even, then this is exactly the Poincaré-Hopf theorem. If m is odd, then
we already know χ(M ) = 0 by Corollary 7.36. Hence multiplying both sides of the
equation above by (−1)m we again get Poincaré-Hopf theorem as desired.
Note that one can use Poincaré-Hopf theorem to give an alternative proof of the
hairy ball theorem:
Proof of Hairy Ball Theorem. Since χ(S 2n ) = 2, we conclude that smooth vector
fields on S 2n must have a zero.
7.7. EULER CLASS 136

7.7 Euler Class

Suppose M is a connected, oriented, compact manifold. Let E be an oriented rank
r vector bundle over M . Let s : M → E be any section of E, i.e., any smooth
map s : M → E such that π ◦ s = idM where π : E → M is the projection map.
Let τ (E) ∈ Hcr (E) = Hcv
r
(E) denote the Thom class if E. Then we get an element
∗
s (τ (E)) ∈ H (M ).
r

Proposition 7.41. The cohomology class s∗ (τ (E)) is independent of the choice of

the section s : M → E.
Proof. Let s0 : M → E be the zero section. For any section s : M → E we have a
homotopy from s0 to s defined by

H : M × [0, 1] → E
(p, t) 7→ (p, ts(p)).

Therefore, s∗ (τ (E)) = s∗0 (τ (E)).

Definition 7.42. Let M be a connected, oriented, compact manifold, and E an
oriented rank r vector bundle over M . Then define the Euler class e(E) of E to
be
e(E) := s∗ (τ (E)) ∈ H r (M ).
Lemma 7.43. Let E1 an oriented rank r vector bundle over M1 , and E2 an oriented
rank r vector bundle over M2 , where both M1 , M2 are oriented, compact manifolds.
Suppose there exists an orientation preserving bundle isomorphism ψ : E1 → E2 that
descends to an orientation preserving diffeomorphism ψ0 : M1 → M2 , and sections
s1 : M1 → E1 , s2 : M2 → E2 , such that the following diagram commutes:
ψ
E1 E2
s1 s2

M1 ψ0
M2

Then we have Z Z
e(E1 ) = e(E2 ).
M1 M2

Proof. We have already shown that the Euler class is independent of the choice of
sections. The commutative diagram implies s∗1 ψ ∗ = ψ0∗ s∗2 . Hence we have

ψ0∗ e(E2 ) = ψ0∗ s∗2 τ (E2 ) = s∗1 ψ ∗ τ (E2 ).

But we claim that ψ ∗ τ (E2 ) = τ (E1 ). This is because the Thom class of E1 is
characterized by the property that fiber integration map applied to it gives unit:
π∗ τ (E1 ) = 1 by definition. But since ψ is an orientation preserving bundle isomor-
phism, it preserves fiber integration. Or in other words
Z Z Z
∗
ψ τ (E2 ) = τ (E2 ) = τ (E2 ) = 1
(E1 )p ψ((E1 )p ) (E2 )ψ0 (p)
137 CHAPTER 7. DIFFERENTIAL TOPOLOGY

for any p ∈ M . Hence indeed we have ψ ∗ τ (E2 ) = τ (E1 ). Therefore, we have

ψ0∗ e(E2 ) = s∗1 τ (E1 ) = e(E1 ).

This implies
Z Z Z Z
e(E1 ) = ψ0∗ e(E2 ) = e(E2 ) = e(E2 ).
M1 M1 ψ0 (M1 ) M2

This completes the proof.

Now we study another version of Poincaré-Hopf theorem, which explains the

mysterious name Euler class:

Theorem 7.44. Let M be a compact, oriented manifold. Then

Z
e(T M ) = χ(M ).
M

Proof. Let ∆ ⊂ M × M be the diagonal of M , and N ∆ ⊂ M × M the normal

bundle of ∆ in M × M . Let N (∆) be a tubular neighborhood of ∆. Then consider
the following commutative diagram

ψ1 ψ2 ψ3
TM T∆ N∆ N (∆)
s1 s2 s3 s4

M ∆ id∆
∆ id∆
∆

where ψ1 (x, v) = (x, x, v, v), ψ2 (x, x, v, v) = (x, x, −v, v), and ψ3 : N (∆) → N (∆)
is an orientation preserving bundle isomorphism, whose existence is guaranteed by
the tubular neighborhood theorem. It is easy to verify that each of the blocks in the
diagram above satisfies the conditions in the previous lemma. Hence by repeatedly
applying the lemma above, we get
Z Z Z Z
e(T M ) = e(T ∆) = e(N ∆) = e(N (∆)).
M ∆ ∆ ∆

Now, s4 : ∆ → N (∆) is homotopic to the zero section (inclusion), and τ (N (∆)) may
also be taken as the Poincaré dual η∆ of ∆ in N (∆) by Corollary 6.95. Therefore,
by the definition of the Poincaré dual, we get that
Z Z Z Z
∗ ∗
e(N (∆)) = i∆ τ (N (∆)) = i∆ η∆ = η∆ ∧ η∆ = χ(N (∆)) = χ(M )
∆ ∆ ∆ N (∆)

where the last equality is because N (∆) deformation retracts onto ∆, which is
diffeomorphic to M , and the Euler characteristic is a homotopy invariant.
This concludes the proof.
7.7. EULER CLASS 138

We conclude this writing with a story of René Thom (1923-2002), who won the
Fields medal in 1958 for his contribution to topology. We hope that this serves as a
belated celebration of his 100th birth anniversary:
Once Thom was joined by two anthropologists in an interview. One of the ques-
tions they discussed was why human ancestors wanted to preserve tinders (Small dry
sticks or finely-divided fibrous matter used to help light a fire). One anthropologist
said that because fire can help ancient human to stay warm, another said because
fire can help cook delicious food. Thom, instead, said that when the night comes,
the waving firelight is breathtakingly beautiful.
Beauty is the eternal pursuit of mathematicians.

Figure 7.4: René Thom (1923-2002).

Index

antisymmetrization, 43 cotangent spaces, 35

atlas, 13 curve, 26

base, 11 de Rham cohomology, 72

basis, 11 deformation
Borsuk-Ulam theorem, 123 retract, 73
Brouwer fixed point theorem, 120 retraction, 73
bundle, 37 degree
cotangent, 40 of a smooth map, 117
normal, 113 of a form, 48
tangent, 38 of an alternating tensor, 44
topological, 37 diffeomorphism, 16
vector, 37 differential, 30
differential k-form, 48
category, 68 with compact vertical support, 106
chain homotopy operator, 88 directed set, 95
chart, 13 disconnected, 12
boundary, 59
interior, 59 Einstein summation convention, 3
chart transition map, 14 elementary alternating tensor, 44
closed, 9 Euler
form, 72 characteristic, 134
co-orientation, 105, 113 class, 136
cochain exact
complex, 72, 78 form, 72
map, 78 sequence, 78
cofinal, 95 sequence of vector bundles, 113
cohomology, 72 exhaustion, 21
of cochain complex, 78 exterior derivative (differentiation), 50
with compact support, 72
with compact vertical support, 106 fiber, 37
compact, 10 field
compatible atlas, 15 tensor, 42
connected, 12 vector, 41
simply, 75 finite type, 95
continuous, 8 functor
contractible, 74 contravariant, 70
INDEX 140

covariant, 69 form, 49
preserving, 57
good cover, 95 reversing, 57
gradient, 35 oriented
hairy ball theorem, 119, 135 manifold, 49
Hausdorff, 13 positively, 55
homeomorphism, 8 trivilization, 104
homotopic, 73 outward-pointing, 60
path, 73
partition of unity, 19
homotopy, 73
path-connected, 11
relative to A, 73
Poincaré
equivalence, 73
dual, 103
equivalent, 73
duality, 97
rel A, 73
lemma, 88
index lemma for compactly supported co-
of a vector field, 135 homology, 90
integral, 55 Poincaré-Hopf theorem, 135
of top degree form, 57 point
integration boundary, 59
along fiber, 90, 106 interior, 59
intersection number, 124 pointed
local, 123 manifolds, 70
inward-pointing, 60 maps, 70
preimage, 8
Lefschetz proper map, 116
number, 130 pullback, 50
fixed point theorem, 130 pushforward, 30
map, 130
local Lefschetz degree, 131 refinement, 95
regular value, 118
manifold
retraction, 73
smooth, 15
topological, 13 Sard’s theorem, 118
with boundary, 59 second-countable, 13
maximal atlas, 15 section, 37
Mayer-Vietoris self-intersection number, 133
sequence for compact suppports, 82 short exact sequence, 79
argument, 95 simply connected, 75
sequence, 80, 81 singular value, 118
nondegenerate, 96 smooth
on a subset, 24
open, 9 maps, 16
orientable, 49 structure, 15
orientation, 49 snake lemma, 79
atlas, 49 Stokes’ theorem, 62
boundary, 61 subspace topology, 9
141 INDEX

support
of a function, 19
compact, 72
of a form, 72
symmetrization, 43

tangent space, 26
tensor, 33
alternating, 43
components, 34
symmetric, 43
Thom
isomorphism, 111
transversality theorem, 133
class, 112, 114
René, 138
topological space, 8
topology, 8
transversal intersection, 123
transversality, 124
trivialization, 37
trivilization
equivalent, 104
oriented, 104
tubular neighborhood theorem, 113

upper-half space, 59

velocity, 26

wedge product, 45
Whitney approximation theorem, 24
winding number, 75
form, 75
Bibliography

[1] John M. Lee, Introduction to Smooth Manifolds, Springer (2012).

[2] Raoul Bott, Loring W. Tu, Differential Forms in Algebraic Topology, Springer
(1982).

[3] Allen Hatcher, Algebraic Topology, Cambridge University Press (2001).

[4] Manfredo Perdigão do Carmo, Riemannian Geometry, Springer (1992).

[5] Glen E. Bredon, Topology and Geometry, Springer (2010).

[6] William Boothby, An Introduction to Differentiable Manifolds and Riemannian

Geometry, Academic Press (2003)

[7] Victor Guillemin, Alan Pollack. Differential Topology. American Mathematical

Soc., 2010

[8] John Milnor. Topology from the Differentiable Viewpoint. Princeton University
Press, 1997.

[9] Zuoqin Wang. Lecture Notes on Differentiable Manifolds. URL. [Chinese]

[10] Zuoqin Wang. Lecture Notes on Differentiable Manifolds. URL. [English]

[11] Liviu I. Nicolaescu. Intersection Theory. URL.

[12] Richard Koch. A Short Course on de Rham Cohomology. URL.

[13] Joel W. Robbin, Dietmar A. Salamon. Introduction to Differential Topology.

URL.