Notes on Differential Topology

George Torres

Last updated January 4, 2019

Contents
1 Smooth manifolds and Topological manifolds 2
1.1 Smooth Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Product Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Calculus on smooth manifolds 3

2.1 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.2 Defining df . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 The Inverse Function Theorem 5

4 Immersions and Submersions 6

4.1 Immersions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.2 Submersions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

5 Regular Values and Implicit Manifolds 8

5.1 Transversality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.2 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

6 Sard’s Theorem 10
6.1 Proof of Sard’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

7 Whitney’s Embedding Theorem 12

7.1 Partitions of Unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7.2 Proof of Whitney . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

8 Manifolds with Boundary 15

8.1 Facts about Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
8.2 Transversality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
8.3 Sard’s Theorem for Manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
8.4 Classification of 1 dimensional Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

9 Transversality (Revisited) 18
9.1 Transversality and Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
9.2 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

10 Intersection Theory (Mod 2) 20

1
CONTENTS CONTENTS

11 Intersection Theory 22
11.1 Orientation of Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
11.2 Oriented Intersection Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
11.3 More General Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

12 Lefschetz Fixed Point Theory 26

12.1 Two dimensional maps and the Euler Characteristic . . . . . . . . . . . . . . . . . . . . . . . . . 27
12.2 Non-Lefshetz fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

13 Vector Fields and Poincaré-Hopf 28

13.1 The Euler Characteristic and Triangulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

14 Integration Theory 29
14.1 Differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
14.2 Integrating Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
14.3 Familiar examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
14.4 The Exterior Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
14.5 Stokes’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

——————————–

Note to the reader : These are lecture from Harvard’s 2014 Differential Topology course Math 132 taught by Dan
Gardiner and closely follow Guillemin and Pollack’s Differential Topology.

2
 1 SMOOTH MANIFOLDS AND TOPOLOGICAL MANIFOLDS

1 Smooth manifolds and Topological manifolds

Definition: A topological manifold X is a locally Euclidean space that is Hausdorff and second countable.
B To discuss calculus on topological manifolds, they must be equipped with a smooth structure. To define this,
we must define the concept of smoothness and smooth manifold.
Definition: A smooth map between Euclidean spaces must first be defined on open sets, then on arbitrary
sets:
• A smooth map between open subsets of Euclidean spaces U ⊂ Rn and V ⊂ Rm is a function f : U → V
that is C ∞ , or infinitely differentiable.
• A smooth map from an arbitrary subset A ⊂ Rn is a function f : A → Rn that extends locally to a smooth
function on open sets. Namely, around every a ∈ A there is an open set U ⊂ Rn and a smooth map
F : U → Rm such that: 
F =f

U ∩A

Definition: Let X ⊂ Rn and Y ⊂ Rm . A function is a diffeomorphism (or smooth equivalence) if f is smooth,

bijective and f −1 is smooth. The sets X and Y are said to be diffeomorphic if such a function exists between
them.
Definition: A smooth k manifold is a subset X ⊂ Rn such that for every point p ∈ X, there is an open1 subset
V ⊂ X containing p together with a diffeomorphism φ : U → V , where U ⊂ Rk is some open subset.
The maps φ : U → V are called parameterizations, and the maps φ−1 : V → U are called coordinate systems.
Together, the collection of all such triples {(U, V, φ)} is called an atlas of charts.
Example (S 1 ): The 1 sphere S 1 = {(x, y) | x2 + y 2 = 1} ⊂ R2 is a smooth 1 manifold.
To show this, we cover it with four open sets: the top half, the bottom half, the left half and the right
half. Each one can be shown to be diffeomorphic to (−1, 1) ⊂ R. For example, let U1 be the top half and
define:  p 
φ1 : (−1, 1) → U1 : x 7→ x, 1 − x2

This is smooth because it is a composition of continuous and C ∞ functions. The inverse is simply the
projection (x, y) 7→ x, which is also smooth (as the projection map on all of R2 is smooth). Thus φ is
a diffeomorphism. The same process for all other halves of the circle covers all points in S 1 , so it is a 1
manifold.
Definition: A smooth structure on a topological k-manifold X is a smooth k-manifold Y ⊂ Rn together with a
homeomorphism X → Y .

1.1 Smooth Structures

Several questions arise when discussing smooth structures:

Q: Does every topological manifold admit a smooth structure?
A: No. For example, there are 4 manifolds with no smooth structure. These aren’t immediately intuitive.
Q: If a topological manifold admits a smooth structure, is it unique (up to diffeomorphism)?
A: No. For example, the 7 sphere has 28 smooth structures, which form a group under connect sums.
1 Open in the subspace topology

3
1.2 Product Manifolds 2 CALCULUS ON SMOOTH MANIFOLDS

1.2 Product Manifolds

Proposition: Let X ⊂ Rn and Y ⊂ Rm be smooth manifolds. Then X × Y ⊂ Rn+m is also a smooth

manifold.
Proof:
Let (x, y) ∈ X × Y . Since X is a manifold, there is a local parameterization φ : U → X around X. There is
similarly a local parameterization around Y , say ψ : V → Y . If we take the map φ × ψ : U × V → X × Y
given by:
φ × ψ(u, v) = (φ(u), ψ(v))
It is easy to check that this is a parameterization for X × Y around (x, y).


2 Calculus on smooth manifolds

The primary use of smooth manifolds is calculus, and the fundamental component of calculus is the derivative as
a linear approximation. Namely, if f : Rn → Rm is a C ∞ function, then the derivative of f at ~c ∈ Rn , denoted
by dfc is a linear map satisfying:
f (~x) = f (~c) + dfc (~x − ~c) + err(~x)
where err(~x) is an error term sufficiently small near ~c.

2.1 Derivatives

The goal of this section is to define df , the best linear approximation to a smooth map f : X → Y between
smooth manifolds (possibly of different dimension). This definition will come out of the chain rule between X, Y
and their tangent spaces.

2.1.1 Tangent spaces

Definition: The tangent space of X at a point x, denoted Tx (X), is the image of dφ~0 , where φ is a local
paramaterization φ : U → X with φ(~0) = x.
B Note that Tx (X) ⊂ RN is a k subspace. It is useful to think of it as a copy of Rk centered at x on the
manifold.
Example: Take S 1 ⊂ R2 . What is T(0,1) (S 1 )?
√
Take the parameterization φ : (−1, 1) → S 1 that sends x 7→ (x, 1 − x2 ). Then:
 
1
dφ = √ −x
1−x2

At ~0, dφ = (1, 0), so the tangent space is the subspace spanned by (1, 0).

4
2.1 Derivatives 2 CALCULUS ON SMOOTH MANIFOLDS

B We now will show that Tx (M ) is well defined:

Proposition: The tangent space Tx (M ) doesn’t depend on the choice of parameterization.
Proof:
Suppose we have distinct parameterizations of X:

φ:U →X φ(0) = x

ψ:V →X ψ(0) = x

Since U and V have a nontrivial intersection and they are open, we may shrink them to assume U = V . Now
define h = ψ −1 ◦ φ. This is a map U → U . Since φ and ψ are diffeomorphisms, so is h. The chain rule now
says:
dφ0 = dψ0 ◦ dh0
This implies that the image of dφ0 is restricted to that of dψ0 , or Im(dφ0 ) ⊂ Im(dψ0 ). The same argument
can be made for h0 = φ−1 ◦ φ, so Im(dψ0 ) ⊂ Im(dφ0 ), and we have equality


2.1.2 Defining df

Now we return to attempting to define the derivative of a smooth map f : X → Y between manifolds. Intuitively,
the derivative at a point x should be a map between the tangent spaces Tx (X) → Tx (Y ).
We can parameterize around x and f (x) in X and Y respectively using:

φ:U →X φ(0) = x

ψ:V →Y ψ(0) = f (x)

If we define h = ψ −1 ◦ f ◦ φ, we require the following diagram to commute, so long as U is chosen small

enough:
f
X Y
φ ψ
h
U V

Similarly, we require the derivatives to commute. The derivatives of φ and ψ also map between tangent spaces
T0 (U ) ∈ Rk and T0 (V ) ∈ R` :
dfx
Tx (X) Tf (x) (Y )
dφ0 dψ0
dh0
T0 (U ) T0 (V )

Since dφ0 is an isomorphism (because φ is), dfx is uniquely determined because require the derivatives to
commute. Thus dfx = dψ0 ◦ dh0 ◦ (dφ0 )−1 .

5
 3 THE INVERSE FUNCTION THEOREM

B The derivative also satisfies the chain rule; that is, given:
f g
X Y Z

we have d(f ◦ g)x = dfx ◦ dg(f (x) . The proof of this is straightforward.
———————
Example: Let S 1 ⊂ C be the set {z ∈ C | |z| = 1}. We can define F : S 1 → S 1 by sending z 7→ z 2 . What is df
at i?
We can associate S 1 with the unit circle in R2 via a + bi ∼ (a, b); in which case the map F is:

(x, y) 7→ (x2 − y 2 , 2xy)

Now we seek df(0,1) . To do this, we need local parameterizations around (0, 1) and F (0, 1) = (−1, 0). One
possibility is: p
φ : (−1, 1) → S 1 x 7→ (x, 1 − x2 )
p
ψ : (−1, 1) → S 1 x 7→ (− 1 − x2 , x)

Now we need h = ψ −1 ◦ f ◦ φ. We note that ψ −1 is the projection onto the y axis, so:
p
h(x) = 2x 1 − x2

We can easily compute dh0 to be 2. df takes an element of the tangent space and maps it to another in
the tangent space of the image. We already saw that Tx (S 1 ) is the span of (1, 0). So df takes a vector
(a, 0) and returns:     
a a
df(0,1) = dψ0 ◦ dh0 ◦ (dφ0 )−1 = 2dψ0 (dφ0 )−1
0 0

We can compute this to be:    

a 0
df(0,1) =
0 2a

3 The Inverse Function Theorem

The question naturally arises: What information does the derivative contain? The Inverse Function Theorem
tells us that it can be used to determine if f is a local diffeomorphism, which we define below.
Definition: A smooth function f : X → Y between manifolds is a local diffeomorphism at x ∈ X if ∃ an open
set U ⊂ X containing x and an open set V ⊂ Y containing f (x) such that f : U → V is a diffeomorphism.
Theorem (Inverse Function): f : X → Y is a local diffeomorphism at x if and only if dfx is an isomor-
phism.
In other words, the question of whether f is a diffeomorphism locally comes down to showing det(dfx ) 6= 0.

6
 4 IMMERSIONS AND SUBMERSIONS

4 Immersions and Submersions

B Before, we saw that f : X → Y is a local diffeomorphism if and only if dfx is an isomorphism. If we
pair this with f being bijective, it is enough to claim that f is a (global) diffeomorphism. In this spririt,
we seek to characterize functions between spaces that aren’t of the same dimension (where dfx can’t be an
isomorphism).
Definition: f : X → Y is an immersion if dfx is injective. It is a submersion if dfx is surjective. The former case
can only happen when dim(X) ≤dim(Y ). The latter can only happen when dim(Y ) ≤dim(X). An immersive
submersion is a local diffeomorphism by the Inverse Function Theorem.

4.1 Immersions

B There is a canonical immersion defined to be:

C : R k → R`

(x1 , ..., xk ) 7→ (x1 , ..., xk , 0, ..., 0)

where k ≤ `.
The Fundamental theorem of immersions tells us about how C relates to general immersions:
Theorem (Immersions): If f is an immersion at x, then f is locally equivalent to the canonical immersion.
Specifically, this means there exist parameterizations φ, ψ so that g = C in the diagram below.
f
X Y
φ ψ
g
U V

Proof:
Start with arbitrary parameterizations φ and ψ. Define G : U × R`−k → R` as:

G(x, z) = g 0 (x) + h(z)

where g 0 : U → R` and h : R`−k → R` are:

g 0 (x) = (g(x), 0)

h(z) = (0, z)
As a fact of linear algebra, it is possible to choose coordinates in R` such that dg00 looks like:
 
0 Ik 0
dg0 =
0 0

Where Ik is a k identity matrix, and the 0 are zero matrices of appropriate dimensions. It is easy then to see
that dG0 looks like:    
0 Ik 0 0 0
dG0 = dg0 + dh0 = + =I
0 0 0 I`
Therefore dG0 is bijective, and hence an isomorphism. The Inverse Function Theorem then says G is a local
diffeomorphism around 0. This means we have a parameterization ψ ◦ G of Y locally, up to shrinking U and
V . This is shown below.

7
4.2 Submersions 4 IMMERSIONS AND SUBMERSIONS

f
X Y
φ ψ◦G
g̃
U V
The map g̃ is induced by the other three maps. From our first diagram, we also had:
f ◦φ=ψ◦g
But since g = G ◦ C, we have f ◦ φ = (ψ ◦ G) ◦ C. Now looking at the second diagram, this tells us g̃ = C,
as desired.


B Images of immersions have good properties, as we will see below.

Definition: A function is proper if f −1 (K) is compact wherever K is compact
Theorem: The image of a proper, injective immersion is a submanifold (such a map is also called an embed-
ding).
Example: Consider the map that sends S 1 to the figure eight shown below.

This is a subset of R2 and we may ask if the image of this map is a submanifold. It is not because the
map is not injective (the origin is mapped to twice). However, even if we remove the redundant point in
the domain, the map still wouldn’t be proper because the preimage of a closed subset of the image near
the origin wouldn’t be closed in the domain.

4.2 Submersions

B Just as with immersions, there is a canonical submersion defined to be:

C s : Rk → R`
(x1 , ..., xk ) = (x1 , ..., xk−` )
where k ≥ `. This is simply truncation of the coordinates to R` .
There is also a similar theorem about submersions:
Theorem: If f : X → Y is a submersion at x ∈ X, then there are coordinates such that f is equivalent to the
canonical submersion; that is there exist parameterizations φ, ψ such that:
f
X Y
φ ψ
Cs
U V

Proof: Found on homework 2.

8
 5 REGULAR VALUES AND IMPLICIT MANIFOLDS

5 Regular Values and Implicit Manifolds

The submersion theorem above lets us prove useful ways of finding manifolds as the zero locus of a system of
equations.
Definition: Given a map f : X → Y , a point y ∈ Y is a regular value if dfx is surjective for all x ∈ f −1 (y). If
y∈
/ f (X), then it is a regular value vaccuously.
Theorem: If f : X → Y is a smooth map with regular value y ∈ Y , then f −1 (y) is a submanifold of X.
Proof:
Let x ∈ f −1 (y) be a point in the preimage. Since y is regular, dfx is surjective, and thereforea submersion.
By the Submersion Theorem, we can choose coordinates around x and y such that y = (0, ..., 0) and f is
equivalent to the canonical submersion (i.e. f (x1 , ..., xk ) = (x1 , ..., x` )). So, in coordinates, f −1 (y) looks like:

(0, . . . , x`+1 , ..., xk )

| {z }
parameterize f −1 (y) locally

Therefore x can be parameterized, making f −1 (y) a manifold.



This opens up a very useful way of showing objects are a manifold: showing it is the preimage of a point.
Example: Show the sphere S n is a manifold.
Define the map f : Rn → R that takes x 7→ |x|2 . The derivative at a ∈ Rn is:

dfa = 2(a1 , ..., an )

Consider the preimage f −1 (1). Since every vector of norm 1 must have a nonzero component, there is
some ai 6= 0. This means dfa surjects onto R, and therefore 1 is a regular value. This means:

{x ∈ Rn | |x|2 = 1} = S n−1

is a manifold.
Corollary: Any `-manifold M can be written locally as the zero set of a collection of functions. That is, for a
subset U ⊂ X, there exist g1 , .., g` , with gi : U → R, such that M ∩ U = {x ∈ X | g1 (x) = g2 (x) = ...g` (x) = 0}.
2

5.1 Transversality

B We saw that when y is a regular value, f −1 (y) is a manifold. This can be generalized to entire subsets Z of
the codomain, as long as a condition called transversality is satisfied.
Definition: A smooth function f : X → Y is transverse to a submanifold Z ⊂ Y at x if:

Im(dfx ) + Tf (x) (Z) = Tf (x) (Y )

Meaning every element of Tf (x) (Y ) can be written as a sum of elements in Im(dfx ) and Tf (x) (Z).
2 This isn’t necessarily true globally, though sometimes it is, as the sphere example showed

9
5.2 Stability 5 REGULAR VALUES AND IMPLICIT MANIFOLDS

Example: Taking X = R, Y = R2 , Z = h1, 0i, and f (t) = (0, t), we get a transverse mapping. This transverse
because Im(df ) = h0, 1i, and Z + Im(df ) = span(e1 , e2 ) = R2 .
B An important special case is when X, Z ⊂ Y are submanifolds, and f : X → Y is taken to be inclusion. The
transversality condition means:
Tp (X) + Tp (Z) = Tp (Y )
This means that, wherever Z and X intersect, their tangent spaces at that point must span that of Y . We say
that two manifolds intersect transversally Z t X if this condition is satisfied.
Theorem: If f : X → Y is transverse to a submanifold Z ⊂ Y , then f −1 (Z) is a submanifold.
Proof:
Recall that we can write Z locally as the zero set of functions g1 , ..., g` that are independent. Let U be a
neihborhood of f (x) in Y , and define g : U → R` as g = (g1 , ..., g` ). We can look at the map g ◦ f : X → R` .
In particular, the zero set of g ◦ f is the set of points that map to Z, which is precisely f −1 (Z). Therefore,
we need only show that 0 ∈ R` is a regular value of g ◦ f to show that f −1 (Z) is a manifold. By the chain
rule we have:
dfx dgf (x)
Tx (X) Tf (x) (Y ) R`

d(g◦f )x

We wish to show this is a surjection. Since the gi were chosen to be independent, dg is surjective. Therefore
d(g ◦ f ) is surjective if and only if df carries its domain beyond the kernel of dg. Not only must the image of
df be more than the kernel of dg, it together with the kernel of dg must span Tf (x) (Y ) (otherwise g wouldn’t
necessarily hit everything in R` ). In other words, we must have:

Im(dfx ) + ker(dgf (x) ) = Tf (x) (Y )

But ker(dg) = Tf (x) (Z), so we must have:

Im(dfx ) + Tf (x) (Z) = Tf (x) (Y )

This is guaranteed by the transversality of f .



5.2 Stability

B An important part of the study of properties of maps is how stable these properties are under small deforma-
tions. In order to make this rigorous, we define homotopy and stability:
Definition: A pair of maps f0 , f1 : X → Y are smoothly homotopic if there is some F : X × [0, 1] → Y such
that F (x, 0) = f0 (x) and F (x, 1) = f1 (x).
Definition: A property P of maps is called stable if it is preserved under small deformation. Specifically, if
f0 : X → Y satisfies a stable property P, and F is some homotopy with F (x, 0) = f0 , then there extists  > 0
such that F (x, δ) = fδ satisfies P for all δ < .
Theorem: The following are stable properties of maps on manifolds:
(a) local diffeomorphism
(b) immersion

10
 6 SARD’S THEOREM

(c) submersion
(d) transverse to some submanifold Z
(e) embedding
(f) diffeomorphism
Proof: See G.P. pp 36

6 Sard’s Theorem
B Recall that the preimage of any regular value y of a smooth mapping f : X → Y is a smooth manifold. This
gives rise to the natural question:
Q: How much of Y is made up of regular values?
A: The answer to this is called Sard’s theorem. In order to state this, we need to define the concept of measure
on a manifold.
Definition: A subset A ⊂ R` has measure zero if A can be covered by a countable number of rectangles with
arbitrarily small volume.
Definition: A subset A ⊂ M of a manifold has measure zero if and only if:
1. For every parameterization φ : U → M , the preimage φ−1 (A) has measure zero as a subset of R` .
2. There exists a covering of M by charts (φα , Uα ) such that φ−1 `
α (A) has measure zero as a subset of R for
all α.
Theorem (Sard’s): If f : X → Y is a smooth map of manifolds, then almost every point in Y is regular. That
is, if C is the set of critcal points of f , then f (C) has measure zero.

6.1 Proof of Sard’s Theorem

B The proof of Sard’s Theorem involves reducing this to a local fact, and treating it like a multivariable calculus
problem that uses a version Fubini’s Theorem.
Definition: Suppose n = k + `. Then Rn = Rk × R` . For c ∈ Rk , let vc be the vertical slice {c} × R` in Rn . A
subset of vc has measure zero if it has measure zero on R` .
Theorem (Fubini): If a closed subset A ⊂ Rn such that A ⊂ vc has measure zero for all c, then A has measure
zero.
B The proof of Sard’s theorem follows from two lemmas:
Lemma: Let f : U → Rp be a smooth map on an open set U ⊂ Rn , and let C be the set of critical points of U .
Further, define a filtration of C as:
C1 = {x ∈ U | dfx = [0]}
  ∂j f 
Ci = x ∈ U 

= ~0 ∀m ∈ {1, ..., p}, ∀j ≤ i
∂xjm
Intuitively, Ci is the set of points for which all partials of order ≤ i are zero. Also note C ⊃ C1 ⊃ Ck . Then, we
have:
1. f (C \ C1 ) has measure 0

11
6.1 Proof of Sard’s Theorem 6 SARD’S THEOREM

2. f (Ck \ Ck+1 ) has measure 0 for all k ≥ 1

n
3. f (Ci ) has measure 0 for k > p − 1.
Proof:
∂fi
1. Let x ∈
/ C1 . Then dfx 6= [0]. This means there exists some partial partial derivative xj 6= 0. Assume
∂f1 n
without loss of generality that ∂x1 = a 6= 0, and define h : U → R as:

h(x) = (f1 (x), x2 , x3 , ..., xn )

Then dhx has the form:  

a
 
dhx =  
 In−1 

This must be invertible. Therefore, by the Inverse Function Theorem, h is a local diffeomorphism.
Specifically, there is some open V containing x which is mapped diffeomorphically by h to some V 0 ⊂ Rn .
Now we consider g = f ◦h−1 : V 0 → Rp . Since dh is invertible, the critical values (points whose preimages
are critical points) of f and g are the same. Therefore we seek to show that the critical values of g have
measure 0.
By construction, g maps points as:

(t, x2 , ..., xn ) 7→ (t, y2 , ..., yp )

Namely, it is the identity on the first component. Therefore for any t ∈ R, g induces a map g t :
{t} × Rn−1 ∩ V 0 → {t} × Rp−1 . Now we look at dg and how it relates to dg t :
 
1 0
 
dg =  
 dg t 

We see that any point of {t} × Rn−1 is critical for g if and only if it is critical for g t . Therefore if the
collection of critical points for g t is of measure zero in the ’slice‘ {t} × Rn−1 , then the entire collection
in Rn has measure zero by Fubini’s Theorem. Inducting on n, with the clear base case of n = 0, we see
that f (V ∩ C) has measure zero. Since we can cover C \ C1 with countably many such V , we have that
the measure of f (C \ C1 ) is also zero.
2. The proof for this is very similar to that for part 1. See G.P. for complete version.
3. Fix δ > 0, and consider a cube S of side length δ contained in U . We will first show that f (Ck ∩ S) has
measure 0 for sufficiently large k. To show this, subdivide S into rn cubelets each of side √ length
δ/r.
Let S1 be a cubelet that intersects Ck . Any point in S1 can be written as x + h with |h| < n rδ and
x ∈ Ck .
Now we use Taylor’s Theorem, which states:

f (x + h) = f (x) + R(x, h)

Where R(x, h) are remainder terms depending on higher derivatives. Since all derivatives are assumed
to be zero for i ≤ k, we have |R(x, h)| < a|h|k+1 . In this context Taylor’s Theorem tells us that f (S1 )
b
is contained in a cube around f (x) with sides of length rk+1 .

12
 7 WHITNEY’S EMBEDDING THEOREM

b

p
Therefore, f (Ck ∩ S) is contained in a union of cubelets with total volume rn rk+1 = bp rn−(k+1)p . If
n p −N
we assume k > p − 1, then f (Ck ∩ S) is contained in a rectangle of volume b r where N is positive.
As r → ∞, this gets arbitrarily small. Therefore f (Ck ∩ S) has measure 0.
n
Now, we cover Ck by countably many such S and we get Ck having measure 0 as well so long as k > p −1.

Lemma: If f : U → Rp is a smooth map on an open set U ⊂ Rp , and C is the set of critical points of U , then
f (C) has measure zero in Rp .
Proof:
By the additivity of measure and the first lemma, we know:

m(f (C)) = m (f (C \ C1 ) ∪ f (C1 )) = m(f (C \ C1 )) + m(f (C1 )) = m(f (C1 ))

n
By the same reasoning, m(f (C1 )) = m(f (C2 )), and so on. Since m(f (Ck )) = 0 for k > p − 1, we have:

m(f (C)) = m(f (C1 )) = m(f (C2 ) = ... = m(f (Ck )) = 0

Proof (of Sard’s Theorem):

Let C be the set of critical points of f : X → Y . Define the critical values of f to be f (C). For any x ∈ X,
parameterize by (φ, U ), with U ⊂ R` . Further, we can parameterize around y ∈ f (φ(U )) so that (ψ, U 0 ) is a
chart around y and U 0 ⊂ f (φ(U )). Composing these mappings gives:

ψ −1 ◦f ◦φ
U Rp

This is a smooth mapping, and its cricial values are exactly those of ψ −1 ◦f because φ is a local diffeomorphism.
By the lemma above, we know that the crical values of ψ −1 ◦ f ◦ φ have measure zero. Therefore the critical
values of ψ −1 ◦f have measure zero in Rp . Since having Lebesque measure 0 is invariant under diffeomorphism,
we then have:
0 = m(ψ −1 ◦ f (C)) = m(ψ(ψ −1 ◦ f (C))) = m(f (C))


7 Whitney’s Embedding Theorem

We showed on homework 1 that every manifold can be embedded into RN for N sufficiently large. The natural
question then arises:
Q: What is the smallest N such that M embeds into RN ?
A: The answer depends on the manifold, but Whitney’s Embedding Theorem gives us an upper bound based
on the dimension of M

13
7.1 Partitions of Unity 7 WHITNEY’S EMBEDDING THEOREM

Theorem (Whitney): If X is a k manifold, then X embeds into R2k+1 . In fact, it embeds into R2k , but we
won’t prove this.
B The proof of this involves the definition of a tangent bundle, partitions of unity and a lemma.
Definition: The tangent bundle T (M ) of a manifold M is the disjoint union of its tangent spaces. In other
words:
T (M ) = {(x, v) ∈ X × Rk | v = Tx (M )}
It can be shown that this is a 2k dimensional manifold.

7.1 Partitions of Unity

Definition: Let X ⊂ Rn be an arbitrary set. For any covering of X by open sets {Uα }, a partition of unity is
a sequence {θi } of smooth functions θi : X → R satisfying:
1. 0 ≤ θi ≤ 1
2. For any x, there extists an open neighborhood of x on which all but finitely many θi are zero.
3. Each θi is zero except on some closed set contained in Uα for some α.
P
4. θi (x) = 1 for all x.
It can be shown that any set X admits a partition of unity (see GP).
B We can use partitions of unity to prove a lemma useful in proving Whitney’s theorem:
Lemma: On any manifold X, there exists a smooth proper map ρ : X → R.
Proof:
Let {Uα } the collection of open sets in X with compact closure. Take a partition of unity {θi } subodinate
to {Uα }. Now define:
X∞
ρ= i · θi (x)
i

For any x, this is a finite sum because the θi are locally finite. Now observe if ρ(x) ≤ j, then one of
θi , ..., θj 6= 0 at x. Then the preimage satisfies:
j
[
ρ−1 ([−j, j]) ⊂ {x | θi (x) 6= 0}
i=1

For each i, we see that the set { x | θi (x) 6= 0} is that which is guaranteed for θi in property 3 of the definition
of partition of unity. Therefore each of these is contained in Uα for some α. Since we picked the cover so
that Uα has compact closure, we must have that each set {x | θi (x) 6= 0} is bounded. Therefore the union is
bounded. Since ρ is continuous, the preimage of a closed set [−j, j] is also closed. Therefore it is compact,
as it is contained in a bounded set. Hence. ρ is proper because any compact set on R is contained in some
[−j, j].


14
7.2 Proof of Whitney 7 WHITNEY’S EMBEDDING THEOREM

7.2 Proof of Whitney

B The proof can be broken down into two smaller chunks: first showing that any k manifold has an injective
immersion into R2k+1 , then showing that this can be a proper map (and hence an embedding).
Proof (of Whitney):
1. Let X ⊂ RN be a manifold. If f : X → RM is an injective immersion with M > 2k + 1, then we claim
we can find a 6= 0 ∈ RM such that π ◦ f : X → H is an injective immersion, where H is the orthogonal
complement of a. To find such an a, define h : X × X × R → RM and g : T (X) → RM as:

h(x, y, t) = t(f (x) − f (y))

g(x, v) = dfx (v)

In particular, the dimensions of the domains of h and g are less than thier codomains (as M > 2k + 1),
so Sard’s theorem tells us that there is a nonzero a not contained in Im(h) and Im(g). Now we look at
π ◦ f , the composition of f and the projection π onto the orthogonal complement of a. If this map wasn’t
injective, then there are distinct x and y such that

π ◦ f (x) = π ◦ f (y) =⇒ π ◦ (f (x) − f (y)) = 0

This means f (x) − f (y) is a scalar multiple of a, as its projection onto the orthogonal complement is zero.
So:
f (x) − f (y) = ta
We see that t 6= 0 because f is injective. But then this means h(x, y, 1/t) = a, which is a contradiction.
Therefore π ◦ f is injective.
Now assume π ◦ f is not an immersion; this implies there is some v 6= 0 in Tx (X) that is contained in the
kernel of d(π ◦ f )x . Namely:

d(π ◦ f )x (v) = 0 =⇒ π ◦ dfx (v) = 0 =⇒ dfx (v) = ta

But t 6= 0 because f is an immersion. This implies g(x, v/t) = a, a contradiction. Thus we have found an
injective immersion X → H ∼ = RM −1 .
Now, we proceed inductively. If we start with f : X → RN an embedding, we can compose this with
a series of injective immersions constructed as above taking X → RN1 → RN −2 → ... → R2k+1 . Once
this point is reached, we cannot use Sard’s theorem as above, so the projection concludes. Therefore this
composition is an injective immersion of X into R2k+1 .
2. Now given an injective immersion f : X → R2k+1 , we can ensure that |f (x)| < 1 by composing this with
the diffeomorphism of R2k+1 onto its unit ball. Choose a proper function ρ : X → R (using the lemma
partitions of unity lemma), and define F : X → R2k+2 given by F (x) = (f (x), ρ(x)). This is an injective
immersion because f is. As with the previous part, we can compose this with an orthogonal projection
π to R2k+1 . In other words, we have an injective immersion π ◦ F : X → H, where H is the orthogonal
complement of some choice of a 6= 0.
Pick a to be neither of the poles of S 2k+1 (we can do this because π ◦ F is an injective immersion at almost
every a). We now claim that π ◦ F is proper. To show this, first we claim that for any c, there is some d
such that:
|π ◦ F (x)| ≤ c =⇒ |ρ(x)| < d ∀x
To see this, assume it is false; that there is some sequence xi ∈ X such that |π ◦F (xi )| ≤ c but |ρ(xi )| → ∞.
Consider now the vector:
1
wi = [F (xi ) − π ◦ F (xi )]
ρ(xi )

15
 8 MANIFOLDS WITH BOUNDARY

This is a vector that takes off the projection of F (xi ) onto H; hence its projective component on H is zero,
and so it is a multiple of a. As i → ∞, we have:
 
F (xi ) f (xi )
= , 1 → (0, 0, ..., 1)
ρ(xi ) ρ(xi )

Further, we also have:

π ◦ F (xi )
→ (0, 0, ..., 0)
ρ(xi )
because π ◦ F (xi ) is bounded by c. Therefore wi → (0, 0, ..., 1). Since a is a multiple of wi , we have that a
is on either of the poles of S 2k+1 , a contradiction. Therefore |ρ(x)| ≤ d for sufficiently large d.
Now, consider the preimage of a closed ball Bc in H, or the set {x | |π ◦ F (x)| ≤ c}. We saw above that
this is a subset of {x | |ρ(x)| ≤ d}, the preimage under ρ of the closed ball of radius d. Since ρ is proper,
we have this preimage being compact. This means (π ◦ F )−1 (Bc ) is bounded. It is also closed, as π ◦ F
is continuous. Thus we have the preimage is compact, and so this is a proper map as any compact set is
contained in some Bc . So π ◦F : X → R2k+1 is injective, proper and is an immersion, so it is an embedding.


8 Manifolds with Boundary

We can slightly generalize the definition of a manifold by allowing to have a boundary.
Definition: A subset of RN is a manifold with boundary if every point in X has an open neighborhood which
is diffeomorphic to an open subset of the upper half plane H k .
Examples of open subsets in H 2 are shown in the figure. Note how some aren’t open in R2 .
B Note that, under this definition, all manifolds are
manifolds with boundary. This just means the ones
we’ve studied so far are simply ones with empty
boundary.

Terminology:
• The diffeomorphisms from the definition of man-
ifold with boundary are still called local param-
eterizations.
• The boundary of X, denoted ∂X, is the set of
points that land on the boundary hyperplane
Figure 1: Open sets in H 2 xn = 0 of H k undser some coordinate chart.

8.1 Facts about Manifolds with Boundary

B Many things about manifolds carry over when they have boundary. In particular, derivatives can be defined
in almost the same way, and therefore transversality as well.
B The first fact is that ∂X is a manifold with empty boundary.

16
8.2 Transversality 8 MANIFOLDS WITH BOUNDARY

Derivatives
Let U be an open subset of H k , and let g : U → R` be smooth. If x is in the interior of U , the derivative is
defined as it is on an open set of Rk . If x ∈ ∂U , then by the definition of g being smooth, there is an extension
g̃ defined on an open neighborhood of x. Then we define dgx ≡ dg̃x . This can be shown to be indepdendent of
extension g̃, so it is indeed well defined. Because of these extensions, the chain rule still holds.
Definition: The tangent space on a manifold with boundary is defined the same way it is on manifolds without
boundary; namely it is the image of dgx , where g is a chart of X at x. Keep in mind that on the boundary, the
tangent space is not a half hyperplane; it is a hyperplane just like everywhere else.
Definition: The derivative between manifolds with boundary is defined in the same way again, namely the
unique map induced between the tangent spaces.
Product Manifolds
Beware! The product of two manifolds with boundary isn’t necessarily a manifold with boundary. For example,
the product [0, 1] × [0, 1] is the unit square. This square isn’t a manifold because the corners aren’t diffeomorphic
to Hk (see Homework 4).
However if Y has boundary and X has empty boundary, then X × Y is a manifold with boundary.

8.2 Transversality

B The notion of transversality can be extended to manifolds with boundary, but with a few extra constraints.
Theorem: Let f : X → Y be a smooth function between X, a manifold with boundary, and Y , a manifold
without boundary. Further, let Z ⊂ Y be a submanifold without boundary. If f is transverse to Z and
∂f = f ∂X → Y is transverse to Z, then f −1 (Z) is a manifold with boundary. Further, we have:

∂(f −1 (Z)) = f −1 (Z) ∩ ∂X

codim(X, f −1 (Z)) = codim(Y, Z)

Proof: See G.P. Very similar to the proof for the first version.
Example:
Let Y = R2 , X be the closed solid disk of radius 1 (f taken to be inclusion). Lets consider two one
dimensional submanifolds Z1 and Z1 detailed below.

In both cases, f is transverse to Z, since the tangent space of the disk is all of R2 . Further, ∂f is transverse
to Z1 , the vertical line, because the tangent spaces of the ∂X and Z1 are independent vectors. Thus Z1 ∩X
is a submanifold with boundary (namely, a closed interval). However, ∂f is not transverse to Z2 because
the tangent space of ∂X and Z2 coincide to only form a 1 dimensional subspace.

17
8.3 Sard’s Theorem for Manifolds with boundary 8 MANIFOLDS WITH BOUNDARY

8.3 Sard’s Theorem for Manifolds with boundary

B Sard’s theorem has an equivalent formulation as well (with similar additinal constraints as above):
Theorem: Let f : X → Y be smooth, and X be a manifold with boundary. Further, assume Y has no boundary.
Then almost every point in Y is a regular value of f and of ∂f .

8.4 Classification of 1 dimensional Manifolds

Theorem: Every compact, connected, 1-manifold with boundary is diffeomorphic to a circle or an interval.
Proof: See G.P. appendix.
Corollary: The boundary points of any compact 1 manifold with boundary has an even number of points.
B An interesting application of this corollary is showing there is no retraction of a manifold to its boundary:
Theorem: If X is a compact manifold with boundary. Then there is no smooth contraction of X to its boundary.
Namely, there does not exist g : X → ∂X such that g restricted to the boundary is the identity.
Proof:
Assume such a g exists, and let z ∈ ∂X be a regular value (exists by Sard’s theorem). By above, we
have that g −1 (z) is a compact submanifold with boundary. By the transversality theorem, we have that
codim(X, g −1 (z)) = codim(∂X, {z}). But the latter is just the dimension of ∂X, which is n − 1. Therefore
g −1 (z) is a 1 manifold because its codimension in X is n − 1. Also by the transversality condition, we have:
∂g −1 (z) = g −1 (z) ∩ ∂X
But ∂g is the identity as z ∈ ∂X, so the boundary of g −1 (z) is just the point z. This contradicts the corollary
above, as this is an odd number. Therefore no such g exists.


B As a corollary of this, we can prove Brouwer’s Fixed Point Theorem:

Corollary: Let B n be the closed unit ball in Rn . If F : B n → B n is smooth map, then there is a fixed point
x ∈ B n , so that F (x) = x.
Proof:
We proceed by contradiction; that its, let F : B n → B n be a map with no fixed point. Then define g(x) to
be the intersection of the line from f (x) to x (note that g(x) is necessarily on the boundary of B n ). This is
well defined for all x because F (x) 6= x for all x by assumption. The explicit equation for the line along x
and f (x) is clearly:
L(t) = tx − (1 − t)F (x)
We note that when |L(t)| = 1, we have L = g(x) because g maps to the boundary. If we enforce L · L = 1,
we get:
L · L = 1 =⇒ t2 |x − F (x)|2 − 2tF (x) · (x − f (x)) + |F (x)|2 = 1
We can invert this relation for t; call it T (x). Since this is a quadratic in t, we note T (x) must be smooth. Then
we know g(x) = L(T (x)). Since L is smooth and T is smooth, so it g. However, we note that g : B n → ∂B n
is a retraction to the boundary because g(x) = x for x ∈ ∂B n . Therefore we have a contradiction by the
above theorem.


18
 9 TRANSVERSALITY (REVISITED)

9 Transversality (Revisited)
B Now we revisit the idea of transversality. Specifically, we’ll discuss it as a stable property of maps.
Definition: A family of maps parameterized by a boundary-less manifold S is a smooth map F : X × S → Y ,
and each family member, denoted fs , is F : X × {s} → Y .
B The parametric transversality theorem (below)3 tells us about the transversality of families of functions, which
will come in handy for a future discussion.
Theorem: Suppose F : X × S → Y is a family of maps and Z ⊂ Y are boundary-less manifolds, and F, ∂F are
transverse to Z. Then almost every fs , ∂fs is transverse to Z.
Proof:
Since F is transverse, F −1 (Z) = W is a submanifold of Y and ∂W = W ∩ ∂(X × S). Now consider the
standard projection π : X × S → S restricted to W . We now claim the following:
1. If s is a regular value of π : W → S, then fs t Z
2. If s is a regular value of π : ∂W → S, then ∂fs t Z.
4
Once we have shown these claims, then the theorem is true by Sard. We will prove (1). . Suppose
F (x, s) = z ∈ Z; since F is transverse to Z, we have:

Im(dFx,s ) + Tz (Z) = Tz (Y )

Namely, for any a ∈ Tz (Y ), there is some b ∈ Tx,s (X × S) such that:

dFx,s (b) − a ∈ Tz (Z)

We note that Tx,s (X ×S) = Tx (X)×Ts (S). This means we can write b = (w, e) for w ∈ Tx (X) and e ∈ Ts (S).
Now we look at the derivative of π : W → S;

dπx,s : Tx (X) × Ts (S) → Ts (S)

This derivative is simply [0, ..., 1]. When applied to a particular vector, it projects that vector onto its Ts (S)
component. Since we assumed a was a regular value of π, this projection is surjective. In particular, we
can find some c = (u, e) ∈ Tx,s (W ). But since F maps into Z, we have dFx,s (u, e) ∈ Tz (Z). Now, since
dFx,s (v, 0) = dfs (v), we have:

dfs (w − u) − a = dFx,s (w − u, 0) − a = dFx,s [b − c] − a = dFx,s (b) − a − dFs (c)

| {z } | {z }
∈Tx (Z) ∈Tz (Z)

Therefore dfs (w − u) − a ∈ Tz (Z). Thus, we have found a vector in the image of dfs (namely dfs (w − u)), and
a vector in Tz (Z) (namely dfs (w − u) − a) whose difference is a ∈ Tz (Y ). Since this is true for any a ∈ Tz (Y ),
we have that fs is transverse to Z.


3 This is the strongest transversality theorem we have encountered thus far

4 The proof for (2) is almost identical

19
9.1 Transversality and Stability 9 TRANSVERSALITY (REVISITED)

9.1 Transversality and Stability

B We will see here that we can use the above stated transversality theorem to construct a homotopy of any
non-transverse map and a transverse map. To begin, we start with a simplified case:
Simple Case: let f : X → Y with Y = RM . Take S to be an open ball in RM and define the family
F : X × S → RM by F (x, s) = f (x) + s. For any fixed x, the map F simply translates the ball S, which
is a submersion. Therefore Im(dFx,s ) = RM , which means F is transverse to any submanifold of Y = RM .
Appealing to the transversality theorem, we have that fs (x) = f (x) + s is transverse to Z for almost all s ∈ S.
We can take s to be arbitrarily small, so F is a homotopy of f and a transverse map fs .
B To generalize this to any target manifold Y , we need a way to ensure that f (x) + s remains inside Y . The
general strategy is to thicken Y by  so that Y  has zero codimension, and write a similar homotopy F composed
with a projection back down onto Y . To do this, we first need to state the - neighborhood Theorem:
Theorem: For a compact boundary-less manifold Y ⊂ RM and a positive number , let Y  be the open set of
points of distance <  from Y . If  is sufficiently small, then each w ∈ Y  has a unique closest point given by
the map π : Y  → Y , which is a submersion.
Remark: There is an equivalent formulation of this for non-compact Y , where  is actually a function of y ∈ Y .
Proceeding forward we assume Y is compact implicitly, but rest assured that even if it isn’t, the −neighborhood
theorem still applies.
Proof: See G.P. pg 71.
Corollary: Let f : X → Y be smooth, and Y be boundary-less. Then there is an open ball S in some RM and
a smooth map F : X × S → Y such that F (x, 0) = f (x) and F and ∂F are submersions.
Proof:
Let Y be in RM and let S be the unit ball in RM . Define:
F (x, s) = π[f (x) + s]
where  and π are the number/map given by the -neighborhood theorem. Firstly, note that the map
(x, s) → f (x) + s is a submersion because Ts (S) has greater (or equal) dimension than Y . Since π is also a
submersion, the composition F is also a sumbersion. The same applies to ∂F .
Further, we see that F (x, 0) = π(f (x)) = f (x).


B Finally, we can state the general Transversality Homotopy Theorem, which follows nearly immediately from
the above corollary.
Theorem: For any smooth map f : X → Y with Z ⊂ Y both boundary-less, there exists a smooth map
g : X → Y homotopic to f such that g t Z and ∂g t Z.
Proof:
Define the homotopy F as from the corollary:
F (x, s) = π[f (x) + s]
Since this is a submersion, the image of dF is all of Tx (X) × Ts (S) = Tx (S) × RM . Since Y has dimension
less than or equal to M , we have that F is transverse to any Z ⊂ Y . The transversality theorem states that
fs t Z and ∂fs t Z for almost all s ∈ S. Let g be any such fs . Then g ∼ f by the homotopy:
G(x, t) = F (x, ts)

20
9.2 Extensions 10 INTERSECTION THEORY (MOD 2)

9.2 Extensions

Here we will give a slightly stronger statement of transversality not necessarily everywhere, but on a subset of
X.
Definition: A map f : X → Y is transversal to Z on a subset C ⊂ X if the transversality condition:

Im(dfx ) + Tf (x) (Z) = Tf (x) (Y )

is satisfied on every point x ∈ C ∩ F −1 (Z).

Theorem: Let Z ⊂ Y both be without boundary, and let C ⊂ X be closed. If f : X → Y is transverse to Z
on C and ∂f is transverse to Z on ∂X ∩ C, then there is a smooth map g homotopic to f for which g t Z and
∂g t Z. Further, on a neighborhood of C we have f = g.
B Intuitively, this means we can take a map that is only transverse on a subset of X, and morph it into some g
transverse everywhere on X with the condition that f = g locally somewhere.
Proof: See G.P. pp 72.
Corollary: If f : X → Y is transverse to some Z when restricted to ∂X, then there is a map g : X → Y that
is transverse to Z, homotopic to f and ∂g = ∂f .
Proof: This is a special case of the above theorem, since the boundary is always closed in X.

10 Intersection Theory (Mod 2)

B We return to manifolds without boundary. Two submanifolds X and Z of Y are of complementary dimension
if dim(X) + dim(Z) = dim(Y ). Note that this means that X and Z intersect on a set of codimension dim(Y )
(i.e. a zero dimensional set). If X is compact, this is a finite set.
Definition: Let f : X → Y be smooth, transverse to Z ⊂ Y which is closed, and X be compact. Further,
suppose X and Z have complementary dimension to Y , and suppose X, Y and Z are boundaryless. Define the
mod 2 intersection number of f with Z to be the number of points in f −1 (Z) mod 2, denoted I2 (f, Z).
The first fact about the intersection number is that it is homotopy invariant:
Theorem 10: If f0 ∼ f1 are transverse to Z, then I2 (f0 , Z) = I2 (f1 , Z).
Proof:
Let F : X × I → Y be a homotopy of f0 and f1 . By the transversality homotopy theorem, we may
assume that F is transverse to Z (otherwise we may perturb it a small amount so that it is). Further, since
∂(X × I) = X × {0} ∪ X × {1}, and ∂F = f0 on X × {0} and ∂F = f1 on X × {1}, we have that ∂F t Z
since f0 , f1 t Z. By the transversality theorem for manifolds with boundary, F −1 (Z) is a one dimensional
submanifold with boundary and:

∂F −1 (Z) = F −1 (Z) ∩ ∂(X × I) = f0−1 (Z) × {0} ∪ f1−1 (Z) × {1}

Our classification of 1 manifolds tells us that ∂F −1 (Z) has an even number of boundary points. Then:

|∂F −1 (Z)| = 2n = |f0−1 (Z)| + |f1−1 (Z)| =⇒ |f0−1 (Z)| = |f1−1 (Z)| (mod 2)

21
 10 INTERSECTION THEORY (MOD 2)

This allows us to more generally define the intersection number:

Definition: The intersection number of an arbitarary smooth map g : X → Y with Z is the interseciton number
of f where f ∼ g and f t Z. Since homotopy holds transitively, we have that any two homotopic maps have the
same intersection number.

Special Cases
B If X ⊂ Y , we can define the intersection number of two manifolds Z and X to be I2 (X, Z) ≡ I2 (i, Z) where
i : X → Y is inclusion. Note that this isn’t always the number of intersection points of Z and X. For example,
if X didn’t intersect Z transversely, the intersection number would actually be the intersection number of some
deformation of X (given by deforming i homotopically). If
B If I2 (X, Z) 6= 0, then X cannot be “pulled apart” from Z no matter how it is deformed.
Example:
Consider the torus T2 = S 1 × S 1 , and let X and Z be the independent circles S 1 × {0} and {0} × S 1 ,
shown below.

We see that Z and X intersect transversely, which means their intersection number is the number of
intersection points, which is 1. We also see that, no matter how you deform X, there is no way to pull it
apart from Z so it has no intersection.
B A third special case is when dim X = 21 dim Y . We can then define the self intersection number of X with
itself. This tells you how many intersections X has with a (transversely intersecting) deformation of itself.
B If X happens to be the boundary of some bigger manifold W , then we have what is known as the boundary
theorem:
Theorem: Let X be the boundary of a compact manifold W and let g : X → Y be a smooth map. If g can be
extended to W , then I2 (g, Z) = 0 for any closed Z ⊂ Y of complementary dimension.
Proof:
Let G : W → Y be an extension of g (that is, ∂G = g). From the transversality homotopy theorem, we can
find some F : W → Y homotopic to G that is transverse to Z and has ∂F t Z. Let f = ∂F . Since G ∼ F ,
the boundary maps are also homotopic (i.e. g ∼ f ). This means I2 (g, Z) = |f −1 (Z)| mod 2. We also know
that F −1 (Z) is a manifold in W . The codimension of F −1 (Z) is equal to the codimension of Z in Y , which
is dim(X) by assumption. Since W has dimension dim(X) + 1, the dimension of F −1 (Z) is 1. Therefore
it is a compact 1 manifold with boundary. In particular, from our classification of such objects, we know
|∂F −1 (Z)| = |f −1 (Z)| = even. Therefore I2 (f, Z) = I2 (g, Z) = 0.


22
 11 INTERSECTION THEORY

B Finally, the case where dim(X) = dim(Y ), we get a homotopy invariant of maps between X and Y . Because
any submanifold Z in this discussion must be complementary to X, and dim(X) = dim(Y ), we must have
dim(Z) = 0. This leads us to defining:
Definition: The mod 2 degree of f : X → Y , where Y is connected and dim(X) = dim(Y ), is I2 (f, {y}),
denoted deg2 (f ).
Theorem: Mod 2 degree is well defined, and homotopic maps have the same mod 2 degree.
Proof: For the first statement, see G.P. pp 80. The second is a corollary of Theorem 10.
Theorem: If X = ∂W and f : X → Y may be extended to all of W , then deg2 (f ) = 0.
Proof: This is a special case of the theorem before last.

11 Intersection Theory
Mod 2 intersection theory, while helpful, doesn’t contain very much information about intersecting manifolds of
complementary dimension. Should we wish to develop a more general intersection number, we need to develop
the theory of manifold orientation.

11.1 Orientation of Manifolds

To define an orientation of a manifold, we must first define it on vector spaces.

Vector Spaces Manifolds

Orientations of manifolds are based on orientations of
Let {vi } and {ui } be ordered bases for a vector space vector spaces:
V . The change of basis matrix P maps one basis to Definition: An orientation of a smooth manifold with
another. This induces an equivalence relation on the boundary is a smooth choice of orientations on each
set of ordered bases of V , namely two bases β, β 0 are Tx (X). ‘Smooth’ here means around every x ∈ X,
equivalent if det(P ) > 0. there should exist a parameterization h : U → X such
that dh : Rk → Th(u) (X) is orientation preserving,
where Rk has the usual orientation.
Definition: An orientation of a vector space is a map
B → {±1}, where B is the set of ordered bases, such Remark: Not all manifolds admit an orientation (for
that equivalent bases have the same sign. We say an example the Möbius Strip). If a manifold X has an
isomorphism A of vector spaces is orientation preserv- orientation, it automatically has another orientation,
ing if β ∼ β 0 ⇒ Aβ ∼ Aβ 0 (otherwise it is orientation denoted −X, which is the opposite choice of basis at
reversing). every point.

Induced Orientations
Product Orientation
B Let X × Y be a product of oriented manifolds, with Y boundaryless. This has a natural orientation. Namely,
give T(x,y) (X × Y ) the basis γ = {α × 0, β × 0}, where α is an ordered basis for Tx (X) and β is an ordered basis

23
11.1 Orientation of Manifolds 11 INTERSECTION THEORY

for Ty (Y ). Then the orientation of X × Y is determined by the orientations of X and Y as:

sign(γ) = sign(α)sign(β)

Boundary Orientation
B The boundary of an oriented manifold also comes with a natural orientation. To define this, we introduce the
normal vectors on the boundary. Since ∂X is of codimension 1, there are two unit vectors in Tx (X) for x ∈ ∂X
that are perpendicular to Tx (∂X). One of these maps into the inward part of Hk , and the other maps out of
Hk . Denote ηx to be the outward unit normal. Now, we can define boundary orientation. Let {v1 , ..., vk−1 } be
a basis for Tx (∂X); its sign under the induced orientation is:

sign({v1 , ..., vk−1 ) = sign({ηx , v1 , ...vk−1 })

when thought of as a basis of Tx (X).

Example: Let’s look at orienting the boundary of a compact 1-manifold:
If X is a 1 compact manifold, then it is diffeomorphic to a union of intervals and/or circles. The boundary
of X is a collection of points, which are the boundary points of the intervals. No matter how X is oriented,
every interval’s boundary points will be a + and a −. Therefore the signed sum of the boundary points
is 0, as there are an even number of them.
Preimage Orientation
B The final instance of an induced orientation is when we have f : X → Y with f and ∂f transverse to
Z ⊂ Y , which are both boundaryless. If X, Y and Z are all oriented, there is an induced orientation for the
preimage f −1 (Z), which is a manifold. In the same way as before, we first look at this from a linear algebra
perspective:

Vector Spaces Manifolds

−1
Let S = f (Z), and let Nx (S, X) be the orthogonal
complement of Tx (S). Then the following determine
the orientation of S:

dfx (Nx (S, X)) ⊕ Tz (Z) = Tz (Y )

Nx (S, X) ⊕ Tx (S) = Tx (X)

Namely, an orientation of Tx (S) is induced by the orien-
Let V = V1 ⊕ V2 , and let β1 and β2 be bases of V1 and tations of Nx (S, X) and Tx (X). The former is oriented
V2 . Further let β = (β1 , β2 ) be a basis of V . Then, if by the first equation since dfx restricted to Nx (S, X) is
we require sign(β) = sign(β1 )sign(β2 ), an orientation of an isomorphism. The exact reasoning for this can be
any two determines an orientation of the third. found in G.P. pp 100.

One might notice that the boundary of a preimage S = f −1 (Z) as formulated above now has two induced
orientations: one as the boundary of an oriented manifold S, and one as the preimage of the map ∂f : ∂X → Y
of Z. These (sometimes) differ by a sign. Namely:
Proposition: If ` = codim(Z), then ∂[f −1 (Z)] = (−1)` (∂f )−1 (Z).
Proof: See G.P. pp 101.

24
11.2 Oriented Intersection Theory 11 INTERSECTION THEORY

11.2 Oriented Intersection Theory

We can now formulate intersection theory for manifolds of complementary dimension.

Definition: Let f : X → Y be smooth, Z ⊂ Y be closed with complementary dimension. Further suppose that
X, Y and Z are boundaryless and are oriented. If f is transverse to Z, then f −1 (Z) is a finite set, each point
having an orientation ±1. The intersection number of f and Z, denoted I(f, Z) is the signed count of these
points.
Proposition: If X = ∂W and f : X → Y extends to W , then I(f, Z) = 0 for any closed manifold of
complementary dimension when W is compact.
Proof:
Let F be an extension of f ; then f = ∂F . Since f is transverse to Z, there is a map G homotopic to F that
is transverse to Z by our corollary of the Extension Theorem. This means F −1 (Z) is a compact oriented 1
manifold whose boundary is f −1 (Z), which is oriented and finite. We showed that the sum of orientation
numbers of the boundary is zero in a previous example. Thus I(f, Z) = 0.


A special case of this proves:

Proposition: If f1 ∼ f0 are transverse to Z, then I(f1 , Z) = I(f0 , Z).
Proof:
Let F : X × I → Y be a homotopy between f0 and f1 . It can be shown that ∂(I × X) = X1 − X0 . Since
F (x, 0) = f0 and F (x, 1) = f1 , the preimage ∂F −1 (Z) is the union of the preimages of f1 and f2 (with proper
orientations). Thus:
∂F −1 (Z) = f1−1 (Z) − f0−1 (Z)
Therefore the intersection number of F −1 (Z) is:

I(∂F, Z) = I(f1 , Z) − I(f0 , Z)

But from the previous proposition, we have that I(∂F, Z) = 0 because f1 ∪ f0 : X1 − X0 → Y extends to
F : X × I → Y . Thus I(f1 , Z) = I(f0 , Z).


B We can also define the intersection number of a nontransverse map to be the intersection number of a homotopic
map that is transverse. Further, we can define the intersection number I(X, Z) of two submanifolds as well as
the degree of a map when dim(X) = dim(Y ) in the same way as before in the mod 2 case (except now it is
signed count).

11.3 More General Perspective

B It will prove useful to to think of intersection theory in the more general setting of having two maps f : X → Y
and g : Z → Y , instead of Z necessarily being a submanifold of Y .
Definition: To maps f : X → Y and g : Z → Y of boundaryless manifolds with X, Z compact are transverse
if:
dfx (Tx (X)) + dgz (Tz (Z)) = Ty (Y )

25
11.3 More General Perspective 11 INTERSECTION THEORY

where f (x) = g(z) = y. This is denoted f t g. When dim(X) + dim(Z) = dim(Y ), the derivatives must be
injective and so the sum is direct:
dfx (Tx (X)) ⊕ dgz (Tz (Z)) = Ty (Y ) (1)
Since these are injections, they isomorphisms onto their images. This means if any two are oriented, the third
can be oriented in the usual way of orienting vector spaces.
———————
B If X, Y and Z are oriented, we can think of the intersection number of f and g. Their “intersection” will be
the set:
I = {f (x) = g(z)} ⊂ X × Z
Any point in this set is oriented via the direct sum in the above definition. That is, a point (x, z) in the
intersection will have sign + if the left orientation of (1) agrees with the right orientation of (1), and sign −
otherwise.
Definition: The intersection number of transverse maps f : X → Y and g : Z → Y , where X, Z are of
complementary dimension, is the signed count of the intersection set I, denoted I(f, g).
Remark: Two things must be verified for the above definition to make sense: the set I must be finite, and
I(f, g) should be a homotpy invariant (of both f and g). These follow from a proposition:
5
Proposition: f t g if and only if f × g t 4 ⊂ X × Z and:

I(f, g) = (−1)dim(Z) I(f × g, 4)

In particular, I(f, g) is well defined and homotopy invariant because I(f × g, 4) is.
Proof: See G.P. pp 114.
Proposition: The intersection number is sometimes antisymmetric (depending on the dimension of X and
Z):
I(f, g) = (−1)dim(X)·dim(Z) I(g, f )

Proof: See G.P. pp 115.

———————
B The intersection number in this formulation has some useful applications in orientability and detecting invari-
ants of Y .
Example: Let dim(Y ) = 2 dim(X) and dim(X) is odd. Then the intersection of X with itself inside Y ,
which is I(i, i) = I(X, X), must be zero because I(X, X) = −I(X, X). This means I2 (X, X) = 0. Since
mod 2 intersection number can be compueted without knowing the orientation of Y , if we find X such that
I2 (X, X) = 1, then Y must be non-orientable by the contrapositive. This is a helpful way of showing a manifold
is not orientable. For example, the diametrical circle in the Möbius strip has self intersection number 1, so the
strip must be nonorientable, which is the case.
Example/Definition: Consider a compact boundaryless orientable manifold Y , and consider the submanifold
4 ⊂ Y × Y . This has dimension half that of Y × Y , so we can compute I(4, 4). We define this to be the Euler
Characteristic χ(Y ), an important invariant of Y . This is a bit like the number of intersections that Y has with
itself inside of Y × Y . Just as before, we note that if Y is odd dimensional, then I(4, 4) = χ(Y ) = 0.
5 the diagonal in X × Z

26
 12 LEFSCHETZ FIXED POINT THEORY

12 Lefschetz Fixed Point Theory

B Let f : X → X be a map of a compact boundaryless manifold to itself. A particular important question
might be: Are there fixed points of f ? How many are there? The technique of Lefschetz Fixed Point theory is
approaching this question from the perspective of intersection theory.
That is, consider the submanifolds 4 and Γ(f ), the graph of f , inside X × X. Then the fixed points of f is the
intersection of these manifolds.
Definition: For a map f : X → X of a compact boundaryless manifold to itself, the Lefschetz number of f ,
denoted L(f ), is I(4, Γ(f )). This is a homotopy invariant by our discussion of intersection theory.
Example: Suppose f : X → X has no fixed point and f ∼ id. Then χ(X) = 0, because I(4, 4) = I(4, Γ(f )) =
0.
B The work we’ve done so far in intersection theory gives us an immediate theorem about fixed points:
Theorem: If f : X → X is a smooth map on a compact orientable manifold. If L(f ) 6= 0, then f has a fixed
point.
Proof: By contrapositive, if f doesn’t have a fixed point, then 4 and Γ(f ) have no intersection, and so L(f ) = 0
trivially.

———————
B Functions for which Γ(f ) t 4 are called Lefschetz maps. In this case, since Γ(f ) and 4 are of complementary
dimension in X × X, their intersection must be finite if they are transverse.
Proposition: Every map f : X → X is homotopic to a Lefschetz map.
Proof: See G.P. pp 120. This is an indirect consequence of the Transversality Theorem.
Q: What does a fixed point of a Lefschetz map look like?
A: The tangent space of Γ(f ) can be computed by parameterizing in the obvious way: φ(x) = (x, f (x)). Then
the derivative is the diagonal matrix with Ik and dfx as the blocks. Thus the tangent space is Γ(dfx ). The
tangent space of 4 is the diagonal of Tx (X) × Tx (X), denoted 4x . Then transversality means:

Γ(dfx ) + 4x = Tx (X) × Tx (X)

As these are vector spaces of complementary dimension, they fill out Tx (X) × Tx (X) if and only if Γ(dfx ) ∩
4x = 0. This means dfx has no fixed point. What is more, it has no eigenvalue +1.
B The above characterization can sometimes apply to fixed points of non-Lefschetz maps. That is, if a fixed
point of any smooth map X → X has the property that dfx has no +1 eigenvalue, then we call it a Lefschetz
fixed point. Intuitively, the Lefschetz fixed points are the isolated fixed points of a map.
B The Lefschetz points have an orientation, which is either ±1 depending on the preimage orientation. This is
called the local Lefschetz number, denoted Lx (f ). A more explicit characterization comes from considering the
map dfx − I. At a Lefschetz point, this map must have no kernel. Thus it is an isomorphism of Tx (X). Therefore
we can consider the sign of the determinant:
Proposition: The local Lefschetz number Lx (f ) is the sign of determinant of dfx − I.
Proof: G.P. pp 121.

27
12.1 Two dimensional maps and the Euler Characteristic 12 LEFSCHETZ FIXED POINT THEORY

12.1 Two dimensional maps and the Euler Characteristic

B It is instructive to look at the two dimensional case, where f : R2 → R2 . Assuming that f fixes the origin,
and let A = df0 so that f (x) = Ax + (x) in the usual. Also assuming that A has two independent eigenvectors,
we get a diagonalization of A:  
α1 0
A=
0 α2
We can analyze the possible fixed point behavior by looking at different values of α1 , α2 (when they are positive).
These are summarized below.

These correspond to α1 , α2 > 1, α1 , α2 < 1 and α1 < 1 < α2 , respectively. We can use this to characterize the
euler characteristic of compact, oriented, boundaryless manifolds.
Proposition: The surface of genus k (torus with k holes) has a Lefshetz map homotopic to the identity, with
one source, one sink and 2k saddles. Consequently, the Euler characteristic is 2 − 2k.
The important image to keep in mind here is the “hot fudge” analogy. This map looks like the flow of a fluid
poured over the k torus that is vertically oriented. The source fixed point is the top, the sink is the bottom, and
the saddles occur near the tops/bottoms of the holes.

12.2 Non-Lefshetz fixed points

Consider a map that is not Lefshetz, but still has isolated fixed points. We can also assign a meaningful value
to these points that will in some sense be the same as what we defined for Lefshetz points.
Definition: Let x be an isolated fixed poinf of f : X → X, and let B be a ball around x so that B only contains
the one fixed point. Then we have a map ∂B → S k−1 given by:

f (x) − x
g(x) =
|f (x) − x|

The local Lefshetz number for x, denoted Lx (f ), is then deg(g).

Proposition: For a Lefshetz fixed point x, our two definitions of Lx (f ) agree.
Proof: They key idea is to write f (x) = dfx (x) + (x) and look at the homotopy ft (x) = dfx (x) + t(x). Then
define:
ft (x) − x
Ft (x) =
|ft (x) − x|
We then have Lx (f ) = deg(F1 ) = deg(F0 ). Thus, we just need to verify that the degree of the map:

dfx − I
F0 =
|dfx − I|

is +1 whenever dfx − I has positive determinant, and is −1 whenever dfx − I has negative determinant. This
follows as a fact of linear algebra (proved on Hw 6).


28
 13 VECTOR FIELDS AND POINCARÉ-HOPF

Corollary: The Lefshetz number, as defined below, is well-defined and is a homotopy invariant for smooth maps
f : X → X having finitely many fixed points.
X
L(f ) = Lx (f )
f (x)=x

13 Vector Fields and Poincaré-Hopf

Definition: Let X ⊂ Rn be a compact boundaryless manifold. A vector field is a smooth map v : X → Rn
such that v(x) ∈ Tx (X).
B Vector fields carry useful information near points where v(x) = 0. We assign a local invariant around an
isolated zero of a vector field v called the index. It is defined by taking a ball B around a zero containing no
other zeros finding the degree of the map D : ∂B → S k−1 defined by:

v(x)
x 7→
|v(x)|

Example:
Consider a vector field v : R2 → R2 with a zero that looks like the drawing below.

The degree of D : S 1 → S 1 is the number of clockwise rotations the vectors make taken with sign. In this
case, this is 1. So ind(x) = 1.
Theorem (Poincaré-Hopf ): Let v be a vector field on a compact boundaryless manifold X with finitely many
fixed points. Then: X
ind(x) = χ(X)
v(x)=0

Proof: See G.P. pp 135.

Corolloary: S k for k even does not admit a non-vanishing vector field.
Proof: If such a vector field existed, then v(x)=0 ind(x) = 0 trivially. But we proved on Hw 10 that χ(S k ) = 2
P
for k even. This is a contradiction by Poincaré-Hopf.

B Another relevant result named after Hopf is:
Theorem (Hopf degree): Two maps of a compact connected, oriented k manifold X into S k are homotopic
if and only if they have the same degree.
B One direction we know to be true as a fact of our definition of degree as an intersection number.

29
13.1 The Euler Characteristic and Triangulations 14 INTEGRATION THEORY

13.1 The Euler Characteristic and Triangulations

B We defined before the Euler characteristic of a manifold to the its self intersection number. There is an
equivalent formulation using triangulations. Informally, break down a manifold into disjoint simplices (triangles),
and look at the number of j dimensional faces. Then:
X
χ(X) = (−1)j # of j-dimensional faces
j

We can show this is in fact true by constructing a vector field whose fixed points are in correspondence to the
faces of the simplices. For two dimensions, we can visualize such a vector field explicitly (below)

We see that each vertex corresponds to a +1 index, each edge corresponds to a −1 index and each face corresponds
to a +1 inex. Thus in two dimensions the Euler characteristic is F − E + V by Poincaré-Hopf.

14 Integration Theory
B Integration on manifolds takes place in the language of differential forms. Namely, we integrate differential
forms on manifolds. In order to understand and define integration of differential forms, we must first develop
the language of tensors in linear algebra.

14.1 Differential forms

Definition: Let V be a vector space of finite dimension. A p tensor on V is a map T : V × V × ... × V → R

that is multilinear.
B For example, a 2 tensor is the dot product. A k tensor can be the determinant of [v1 , ..., vk ]. We denote
T p (V ∗ ) to be the vector space of p tensors, where V ∗ is the dual of V . We can also define a product of two
tensors. Namely, let m ∈ T p (V ∗ ) and n ∈ T q (V ∗ ). Then:

m ⊗ n(v1 , ..., vp+q ) := m(v1 , ..., vp ) · n(vp+1 , ..., vp+q )

Note that m ⊗ n 6= n ⊗ m.
B Given a basis of V ∗ , we can find write a baisis of T p (V ∗ ), Namely, if {φ1 , ..., φk } is a basis for V ∗ , then:
 
φi1 ⊗ ... ⊗ φik  = 1 ≤ i1 , ..., ip ≤ k

30
14.1 Differential forms 14 INTEGRATION THEORY

is a basis for T p (V ∗ ). The justification for this can be found in G.P. pp 154.
B It is instructive to restrict our scope to alternating tensors, or antisymmetric tensors. These are relevant in
the study of differential forms, as we want to keep track of signed areas. We then denote Λp (V ∗ ) to be the subset
of alternating tensors. It is very possible that this is not closed under ⊕; so we have to define another law of
multiplication (called exterior product, or wedge product).
Definition: We construct a map Alt : T p (V ∗ ) → Λp (V ∗ ) that, in some sense, forces a p tensor to be alternat-
ing:
1 X
Alt(T ) := sgn(σ)T σ
p!
σ∈Sp
σ
Where Sn is the symmetric group on n letters and T is T applied to the vectors {vσ(1) , ..., vσ(n) }.
Definition: If T and S are alternating p and q tensors, then their wedge product:
T ∧ S := Alt(T ⊗ S)
is an alternating form satisfying T ∧ S = (−1)pq S ∧ T . We note that for p = q = 1, we get the important relation
T ∧ T = −T ∧ T = 0. Just as before, we can take a basis of Λ1 (V ∗ ) = V ∗ and write a basis of Λp (V ∗ ):

{φi1 ∧ ... ∧ φip  1 ≤ i1 ≤ ... ≤ ip ≤ k}

From this it follows that dim(Λp (V ∗ )) = kp . In particular, if k = p, we have dim(Λp (V ∗ )) = 1. This means

every multilinear map in this space is a multiple of the determinant. Further, if p > k, then one of the indices in
the basis elements must be repeated, and thus the tensor must be the zero map. This means that the dimension
of Λp (V ∗ ) terminates at p = k; this allows us to form the exterior algebra:
Λ(V ∗ ) = Λ0 (V ∗ ) ⊕ Λ1 (V ∗ ) ⊕ ... ⊕ Λk (V ∗ )
with the convention Λ0 (V ∗ ) = R.
B If A : W → V is a linear map, there is an induced map A∗ : Λp (W ∗ ) → Λp (V ∗ ) given by:
A∗ (T )(v1 , ..., vp ) = T (Av1 , ..., Avp )
This is sometimes called the transpose of A (when A is a matrix, it is literally the transpose). The determinant
theorem tells us more when W = V :
Theorem: If A : V → V is a linear map, then A∗ T = det(A)T for all T ∈ Λk (V ) where k = dim(V ).
———————
B All of this work now allows us to define a differential form.
Definition: Let X be a smooth manifold. A differential p form on X is a function ω : X → Λp (Tx (X)∗ ). That
is, ω is a choice of p form on the tangent spaces.
B In the case of 0 forms, we have Λo (Tx (X)) = R, and these are just functions X → R. An example of a 1 form
might be x 7→ dφx , where φ : X → R is a smooth 0 form. This is called the differential of φ. In this way, the
coordinate functions x1 , ..., xk on Rk yield 1 forms dx1 , ..., dxk that project a tangent vector into its individual
components.
Proposition: Every p form on an open set U ⊂ Rk can be uniquely expressed as a sum I fI dxI over increasing
P
index sequences I = (i1 < ... < ip ) (here fI are real valued functions on U ), where dxI = dxi1 ∧ ... ∧ dxip .
Definition: If f : X → Y is a smooth map, and ω is a differential form on Y , we define the pullback of ω,
denoted f ∗ ω, to be the form on X given by:
f ∗ ω(x) = (dfx )∗ ω(f (x))

31
14.2 Integrating Differential Forms 14 INTEGRATION THEORY

Recall we defined (dfx )∗ as the induced map from Λp (Tx X) → Λp (Tf (x) Y ). The pullback has several proper-
ties:
f ∗ (ω1 + ω2 ) = f ∗ ω1 + f ∗ ω2
f ∗ (ω ∧ θ) = (f ∗ ω) ∧ (f ∗ θ)
(f ◦ h)∗ ω = h∗ f ∗ ω

Definition: Let ω be a p form on Rk ; writing it as

P
I aI dxI , we say ω is smooth if each fI is smooth.
3
Example: What do 2 forms on R look like?
They will be sums of wedge products of the elementary 1 forms:
ω = a(x)dx1 ∧ dx2 + b(x)dx1 ∧ dx3 + c(x)dx2 ∧ dx3
Keep in mind these have to be increasing in index.

14.2 Integrating Differential Forms

Integrating k forms on open sets

B Let a be a real valued function on an open set U ⊂ Rk . We can look at the k form ω = adx1 ∧ ...dxk , which
is in a sense a volume form. Then we define the integral of ω to be:
Z Z
ω= a dx1 . . . dxk
U U

in the usual multivariable sense. With this comes the change of varialbes formula, which makes use of pull-
back:
Change of Variables: Let f : V → U is an orientation preserving diffeomorphism of open sets in Rk (or Hk ),
and let ω be an integrable k form on U . Then:
Z Z Z
ω= f ∗ω = det(dfx ) ω ◦ f
U V V

If f is orientation reversing, then: Z Z

ω=− f ∗ω
U V
.
Integrating k forms on manifolds
B Let ω be a k form on a manifold M with compact support (i.e. the closure of the set of points where ω(x) 6= 0
is compact).
Warm-up case: Suppose the support of ω is contained entirely inside a parameterizable open set W ∈ M . Let
φ : U → W be a diffeomorphism that parameterizes W . Then φ∗ ω is a differential form on U , an open set in
Rk . We know how to integrate forms on open sets in Rk , so we define:
Z Z
ω := φ∗ ω
M U

Is this well defined? Let ψ : V → W be another parameterization of W . Then f = φ−1 ◦ ψ is an orientation

preserving diffeomorphism. Then by the change of variables formula, we have:
Z Z Z Z
φ∗ ω = f ∗ φ∗ ω = (φ ◦ f )∗ φ∗ ω = ψ∗ ω X
U V V V

32
14.3 Familiar examples 14 INTEGRATION THEORY

General case: Now we wish to integrate an arbitrary k form whose support may not be entirely contained in
a parameterizable open set. To deal with this, we take a partition of unity {ρi } of M subject to the set of all
parameterizable open sets in M . Then local finiteness implies all but finitely many ρi are zero on the support
of ω. Then finitely many of {ρi ω} are nonzero, so we can define the integral of ω:

Z XZ
ω := ρi ω
M i M

The right hand side makes sense because it is a finite sum of integrals of functions with support inside a
parameterizable open set, which we have already defined. Is this well defined? Let {ρ0j } be another partition of
unity; then: Z XZ
ρi ω = ρ0j ρi ω
M j M
Z XZ
ρ0j ω = ρi ρ0j ω
M i M

So we have:
XZ XXZ XZ
ρi ω = ρ0j ρi ω = ρ0j ω X
i M i j M j M

where we exchanged the summations. Thus this doesn’t depend on the partition.
Integrating ` < k forms on submanifolds
B Perhaps we might want to integrate an ` form on M , where ` < dim(M ). We can’t integrate over M , because
the dimensions don’t add up, but we can integrate over an ` dimensional submanifold. There is a natrual
definition of this via the inclusion map i : Z → M :
Z Z
ω= i∗ ω
Z Z

14.3 Familiar examples

B Now we look at a few exampeles for M = R3 . If ω is a 1 form, then it looks like:

ω = f1 dx1 + f2 dx2 + f3 dx3

Consider integrating ω on a curve γ : [0, 1] → R3 , written as hγ1 , γ2 , γ3 i. Then we compute γ ∗ ω by computing

on the components:
dγi
γ ∗ dxi = dγi = dt
dt
γ∗f = f ◦ γ
So:
Z 3 Z 1
X dγi
ω= f ◦ γ(t) dt
C i=1 0 dt

We recognize this to be the line integral of the vector field hf1 , f2 , f3 i. If ω is a 2 form, we similarly get the
standard flux integral of the same vector field.

33
14.4 The Exterior Derivative 14 INTEGRATION THEORY

14.4 The Exterior Derivative

B We will define a map d : {k-forms} → {k + 1-forms}, called the exterior derivative. We have already encoun-
tered this for k = 0. Namely, if f is a 0 form (a function), then the differential df is a 1 form. To generalize this,
we start with the Euclidean case.
Euclidean case: Say ω is a p form on U ⊂ Rk . Then we can write ω =
P
aI dxI , where I is defined as in
section 14.1. Then, we define dω to be exactly what we might expect:
X
dω = daI ∧ dxI

This formulation satisfies three important properties:

1. Linearity: d(ω1 + ω2 ) = dω1 + dω2 .
2. Product rule: d(α ∧ β) = dα ∧ β + (−1)p α ∧ dβ, where α is a p form.
3. d(dω) = 0.
This last property is a consequence of the equality of mixed partials (see G.P. pp 175). Alternatively, we can
define d as the unique operator that satisfies these three properties. Another property of d is that, if g : V → U
is a diffeomorphism, then d(g ∗ ω) = g ∗ (dω). Namely, it commutes with pullbacks.
General case: The exterior derivative d for forms on an arbitrary manifold is defined locally in a natural way.
That is, if φ : U → V is a local parameterization, define dω on φ(U ) to be (φ−1 )∗ d(φ∗ ω). This formulation of d
has the following properties:
1. d(ω1 + ω2 ) = dω1 + dω2
2. d(ω ∧ θ) = (dω) ∧ θ + (−1)p ω ∧ dθ, where ω is a p form
3. d(dω) = 0
4. If g : X → Y is a smooth map of manifolds with boundary, then d(g ∗ ω) = g ∗ (dω).
Example:
Consider differentiating 0, 1, 2 and 3 forms on R3 :
• If ω = f is a 0 form, it is a function and df = g1 dx1 + g2 dx2 + g3 dx3 , where g1 = ∂f
∂xi . We often call
hg1 , g2 , g3 i = ∇f the gradient of f .
• If ω = f1 dx1 + f2 dx2 + f3 dx3 is a 1 form, then dω = g1 dx2 ∧ dx3 + g2 dx3 ∧ dx1 + g3 dx1 ∧ dx2 , where:
∂f3 ∂f2 ∂f1 ∂f3 ∂f2 ∂f1
g1 = − , g2 = − , g3 = −
∂x2 x3 ∂x3 ∂x1 ∂x1 ∂x2
We often call hg1 , g2 , g3 i = ∇ × F the curl of F = hf1 , f2 , f3 i.
• If ω = f1 dx2 ∧ d3 + f2 dx3 ∧ dx1 + f3 dx1 ∧ dx2 is a 2 form, then dω = F dx1 ∧ dx2 ∧ x3 , where:
∂f1 ∂f2 ∂f3
F = + +
∂x1 ∂x2 ∂x3
This is sometimes called the divergence.
• If ω is any 3 form, then dω = 0 because at least one of the indices must repeat in each term of the
sum.

34
14.5 Stokes’s Theorem 14 INTEGRATION THEORY

14.5 Stokes’s Theorem

B One of the most fundamental results of integration theory is Stokes’s Theorem, which relates differentiation
and integration:
Theorem (Stokes): If X is a k dimensional manifold with boundary, and ω is a smooth k − 1 form on X
(extending to the boundary), then: Z Z
dω = ω
X ∂X

This has various names for small k. If k = 1, this is is the Fundamental Theorem of Calculus. For k = 2, this is
the Classical Stokes’s Theorem (or Green’s Theorem). For k = 3, this is the divergence theorem.
Proof: See G.P. pp 183-185.

35