Introduction to Differential Topology Fundamentals

Introduction to Differential topology
Eulershi
2021-01-24
2
Contents
1 Preliminary 5
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Differential Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Derivatives and Tangents . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Second-countable, Paracompactness . . . . . . . . . . . . . . . . . . 9
2
1.4.1 Second-countable—C . . . . . . . . . . . . . . . . . . . . . . 9
1.4.2 Paracompactness . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Appendix A: Tangent Spaces, Tangent Vectors, Function Germs and
Cotangent Spaces (abstract) . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Appendix B: More Topological Properties . . . . . . . . . . . . . . . 10
2 Local Classification of Smooth Mappings 11

2.1 The Inverse Function Theorem . . . . . . . . . . . . . . . . . . . . . 11
2.2 Immersions and Embeddings . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Submersions and Regularity . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Transversality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Homotopy and Stabilty . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Sard’s Theorem and Morse Functions . . . . . . . . . . . . . . . . . 20
2.6.1 Sard’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6.2 Morse Functions . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 Whitney’s Embedding Theorem and Partition of Unity . . . . . . . 23
3
4 CONTENTS
3 Transversality and Intersection 29
3.1 Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 One-Manifolds and Some Consequences . . . . . . . . . . . . . . . . 32
3.3 Transversality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Intersection Theory Mod 2 . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Winding numbers and the Jordan-Brouwer Separation Theorem . . 43
3.6 The Borsuk-Ulam Theorem . . . . . . . . . . . . . . . . . . . . . . . 47
4 Oriented Intersection Theory 53
4.1 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 Oriented Intersection Number . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Lefschetz Fixed-point Theory . . . . . . . . . . . . . . . . . . . . . . 62
4.4 Vector Fields and the Poincare-Hopf Theorem . . . . . . . . . . . . 68
4.5 The Hopf Degree Theorem . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 The Euler Characteristic and Triangulations . . . . . . . . . . . . . . 73

Chapter 1
Preliminary
Braving the wind and the waves may hang straightly sometimes,
to say that sail helps the sea.
—Bai Li
1.1 Introduction
The lecture note is prepared for the course Intrdouction to Differential topology.
The main subject is the manifold theory and smooth mappings. The original inspi-
ration for this note was V. Guillemin and A. Pollack’s lovely Differential topology.
1.2 Differential Manifolds

Beautiful, deep insights into the structure and properties of many geometric
spaces can be developed intuitively with the aid of a few tools from elememtary
calculus. Because calculus is bulit on the local geometry of Euclidean space, it
most naturally adapts itself to spaces that locally "look the same" as some Eu-
clidean space. We call such objects manifolds, spaces in which the environment of
each point is "just like" a small piece of Euclidean space.
5
6 CHAPTER 1. PRELIMINARY
Definition 1.2.0.1. Suppose M is a topological space. We say that M is a topological

manifold of dimension n or a topological n-manifold if it has the following properties:
• M is Hausdorff;
• M is C 2 ;
• M is locally Euclidean of dimension n: each point of M has a neighborhood that is
homeomorphic to an open subset of Rn ;
In Differential topology, our subject is mainly differnetial(smooth) manifolds, which

need a "diffeomorphism".
Definition 1.2.0.2. Suppose M is a topological manifold in RN , we say that M is a k-

dimensional (smooth) manifold(k ≤ N ) if it’s locally deffeomorphic to an open subset
of Rk .
Remark 1. • A smooth map f : X → Y of subsets of two Euclidean spaces is a diffeo-

morphism if it’s one-to-one and onto, and the inverse map f −1 : Y → X is also smooth.
• A diffeomorphism φ : U → V is called a parametrization of the neighborhood
V .(V = Ve ∩ X, Ve is open in RN . U is open in Rk .)
• The inverse diffeomorphism φ−1 : V → U is called a coordinate system on V .
Now we want to manufacture new manifolds from known ones: two useful
methods are considering the subspace and product space:
Definition 1.2.0.3. If X and Z are both manifolds in RN and Z ⊂ X, then Z is a

submanifold of X. In paricular, X is itself a submanifold of RN .
Suppose that X and Y are manifolds inside RN and RM respectively, so that

X × Y is a subset of RN × RM = RN +M . If dimX = k and x ∈ X, we can find an
open set W ⊂ Rk and a local parametrization ψ : W → X around x. Similarly, if
dimY = l and y ∈ Y , we can find an open set U ⊂ Rl and a local parametrization
φ : U → Y around y.
Define the map φ × ψ : W × U → X × Y by the formula
φ × ψ(w, u) = (φ(w), ψ(u)).

1.3. DERIVATIVES AND TANGENTS 7
of course φ × ψ is a local parametrization of (x, y) in X × Y . Thus we have proved:
Theorem 1.2.0.1. If X and Y are manifolds, so is X × Y , and dimX × Y = dimX +

dimY .
1.3 Derivatives and Tangents

In basic calculus we have known the derivative of f in the direction h, taken
at the point x, which is defined by the conventional limit
f (x + th) − f (x)
dfx (h) = lim
t→0 t
where dfx : Rn → Rm . And the Chain Rule says that for each x ∈ U ,
d(g ◦ f )x = dgf (x) ◦ dfx .
The derivative of a mapping is its best linear approximation. Suppose that X

sits in RN and that φ : U → X is a local parametrization around x, where U is
an open set in Rk , and assume that φ(0) = x for convenience. The best linear
approximation to φ : U → X at 0 is the map
u → φ(0) + dφ0 (u) = x + dφ0 (u).
define the tangent space of X at x to be the image of the map dφ0 : Rk → RN . So the
tangent space, which we denote Tx (X), is a vector subspace of RN whose parallel
translate x + Tx (X) is the closest flat approximation to X through x.
A tangent vector to X ⊂ RN at x ∈ X is a point v of RN that lies in the vector
subspace Tx (X) of RN .
∂ ∂ ∂
Remark 2. The basis of Tx (X) is denoted by 1, 2,··· , k.
∂x ∂x ∂x
Another method to define tangent space and tangent vector is through differ-
ential curves.
Definition 1.3.0.1. A differential curve passing through x is a differential mapping

γ : (−, ) → M satisfying γ(0) = x. the set of all differential curves passing through x
is denoted Mx .
Also, the set of all differnetial real functions around x is denoted Fx . For
γ ∈ Mx and f ∈ Fx , let
d
hf, γix = f ◦ γ(t)|t=0 .
dt
Definition 1.3.0.2. introducing tangency relation ∼ on Mx :
γ1 ∼ γ2 ⇔ hf, γ1 ix = hf, γ2 ix , ∀f ∈ Fx .
each equivalence class [γ] is called a tangent vector at x of X. the set of all tangent
vectors Mx = Mx / ∼ is called the tangent space at x of X, denoted Tx (X).
For tangent vector [γ] ∈ Tx (X), define
hf, [γ]i := hf, γix .
which means the direction derivative of f through [γ].

The dimension of the vector space Tx (X) is the dimension k of X. The prroof
is obvious.
Next we consider the best approximation of a smooth map of arbitary mani-
folds f : X → Y at a point x. If f (x) = y, to define dfx : Tx (X) → Ty (Y ) we need
our parametrization.
f
XO / YO
φ ψ
−1
h=ψ ◦f ◦φ
U /V
suppose that φ : U → X parametrizes X about x and ψ : V → Y parametrizes

Y about y, where U ⊂ Rk , V ⊂ Rl , and , say, φ(0) = x, ψ(0) = y. If U is small
enough, then we can draw the commutative square above. Thus we can define
dfx = dψ0 ◦ dh0 ◦ dφ−1
0 .
1.4. SECOND-COUNTABLE, PARACOMPACTNESS 9
We can easily verify that all definitions we have defined above does not de-
pend on parametrization, and our deritative truly conform to the chain rule:
f g
XO / YO / ZO
φ ψ η
−1 −1
h=ψ ◦f ◦φ j=η ◦g◦ψ
U /V / W
f g
Theorem 1.3.0.1. (Chain Rule) If X → Y → Z are smooth maps of manifolds, then
d(g ◦ f )x = dgf (x) ◦ dfx .
Another method to define the derivative of smooth map is simple:

f
Definition 1.3.0.3. Suppose X → Y , the differential dfx of f at x is defined as dfx :
Tx (X) → Tf (x) (Y ), where dfx ([γ]) = [f ◦ γ].
Remark 3. Another definition of the above is used in our definition of abstract differential
manifolds. The details can be found in the Appendix A.
1.4 Second-countable, Paracompactness
1.4.1 Second-countable—C 2
Since we have known the definition of C 2 , we just give a theorem:
Theorem 1.4.1.1. (Lindelöf ) Any open cover of an C 2 topological space X has a

countable subcover.
1.4.2 Paracompactness
Definition 1.4.2.1. Suppose X is a set, V and W is two covers of X. If for any V ∈ V
there exists W ∈ W such that V ⊂ W , we say V subdivide W , denoted V ≺ W .
Definition 1.4.2.2. Suppose X is Hausdorff, we say X is paracompact if for any open

cover O of X, there exist a locally finite open cover O 0 , such that O 0 ≺ O.
1.5 Appendix A: Tangent Spaces, Tangent Vectors, Func-

tion Germs and Cotangent Spaces (abstract)
1.6 Appendix B: More Topological Properties

Chapter 2
Local Classification of Smooth

Mappings
Healthy and strong, rather white in the first shift of the heart;
The poor and benefit kin, do not fall Albatron ambition.
—Bo Huang
Before we really begin to discuss the topology of manifolds, we must study the
local behavior of smooth maps.
2.1 The Inverse Function Theorem

If X and Y are smooth manifolds of the same dimension, then the simplest
behavior a smooth map f : X → Y can possibly exhibit around a point x is to
carry a neighborhood of x diffeomorphically onto a neighborhood of y = f (x). In
such an instance, we call f a local diffeomorphism at x.
A necessary condition for f to be a local diffeomorphism at x is that its deriva-
tive mapping dfx : Tx (X) → Ty (Y ) be an isomorphism. The fact that this linear
condition is just the inverse function theorem:
11
12 CHAPTER 2. LOCAL CLASSIFICATION OF SMOOTH MAPPINGS
Theorem 2.1.0.1. (The Inverse Function Theorem) Suppose that f : X → Y is a

smooth map whose derivative dfx at the point x is an isomorphism. Then f is a local
diffeomorphism at x.
The proof of the theorem is omitted here and can refer to Rudin’s Mathematical
Analysis Arinciples or Zhusheng Zhang’s a New Lecture on Differential Topology.
Remember that the Inverse Function Theorem is purely a local result. We can
suggestively reformulate the inverse function theorem by using local coordinates:
Theorem 2.1.0.2. (The Inverse Function Theorem) If dfx is an isomorphism, then

one can choose local coordinates around x and y so that f appears to be the identity,
f (x1 , · · · , xk ) = (x1 , · · · , xk ).
In general, we shall say that two maps f : X → Y and f 0 : X 0 → Y 0 are

equivalent if there exist diffeomorphism α and β completing a commutative square:
f
XO / YO
α β
0
f
X0 /Y0
In this terminology, the inverse function theorem says that if dfx is an isomor-
phism, then f is locally equivalent, at x, to the identity.
2.2 Immersions and Embeddings

For the inverse function theorem, the dimensions of X and Y must be equal.
What’s the best local behavior a map can have when dimX < dimY ? The best
we can demand is that dfx : Tx (X) → Ty (Y ) be injective. If so, f is said to be an
immersion at x.
Definition 2.2.0.1. If dfx : Tx (X) → Ty (Y ) is injective, then f is called an immersion

at x. If f is an immersion at every point, it’s simply called an immersion.
2.2. IMMERSIONS AND EMBEDDINGS 13
The canonical immersion is the standard inclusion map of Rk into Rl for l ≥ k,

where (a1 , · · · , ak ) maps to (a1 , · · · , ak , 0, · · · , 0). In fact, up to diffeomorphism,
this is locally the only immersion.
Theorem 2.2.0.1. (Local Immersion Theorem) Suppose that f : X → Y is an immer-

sion at x and y = f (x). Then there exist local coordinates around x and y such that
f (x1 , · · · , xk ) = (x1 , · · · , xk , 0, · · · , 0).
In other words, f is locally equivalent to the canonical immersion near x.
The proof is left as an exercise for the readers.

However, in order to force an immersion to exhibit desirable global proper-
ties, we must append topological conditions to the local differential data.
Of particular interest is the image of an immersion. In fact, the image of an
arbitrary immersion f : X → Y need not be a submanifold of Y . An example is
twisting a circle in R2 .
A map f : X → Y is called proper if the preimage of every compact set in Y is
compact in X.
Definition 2.2.0.2. An immersion that is injective and proper is called an embedding.
Having subjoined appropriate global topologucal constraints to the local im-

mersion condition, we may now prove a reasonable global extension of the local
immersion theorem:
Theorem 2.2.0.2. An embedding f : X → Y maps X diffeomorphically onto submani-

fold of Y .
Proof. In order to prove f (X) is a manifold, it suffices to show that the image of
any open set W of X is an open subset of f (X). If f (W ) is not open in f (X), then
there exists a sequence of points yi ∈ f (X) that do not belong to f (W ) but that
converge to a point y in f (W ). As the set {yi , y} is compact, its preimage in X
must be compact. Each point yi has precisely one preimage point xi in X, and y
possesses precisely one preimage point x, which must belong to W . Since {xi , x}
is compact, by passage to a subsequence we must assume that xi converges to
z ∈ X. Thus f (xi ) → f (z), we get x = z. We conclude that, for large i, xi ∈ W .
Contradiction!
It’s trivial to show that f : X → f (X) is a diffeomorphism. We have known
that f is a local diffeomorphism, since it’s bijective, the inverse f −1 is well-defined.
And locally we have known that f −1 is smooth. Thus the proof is complete.
In particular, if X itself is compact, every map f : X → Y is proper. Thus for

X, embeddings are just one-to-one immersions.
2.3 Submersions and Regularity

Now we consider the case when dimX ≥ dimY . If f : X → Y carries x to
y, the strongest condition we can impose on its derivative dfx : Tx (X) → Ty (Y ) is
surjectivity. If dfx is surjective, f is called a submersion at x.
Definition 2.3.0.1. If dfx : Tx (X) → Ty (Y ) is surjective, then f is called a submersion

at x. If f is a submersion at every point, it’s simply called a submersion.
The canonical submersion is the standard projection of Rk onto Rl for k ≥ l, in

which (a1 , · · · , ak ) → (a1 , · · · , al ). As in the case of immersion, every submersion
is locally canonical, up to a diffeomorphism.
Theorem 2.3.0.1. (Local Submersion Theorem) Suppose that f : X → Y is a sub-

mersion at x and y = f (x). Then there exist local coordinates around x and y such that
f (x1 , · · · , xk ) = (x1 , · · · , xl ). That is, f is locally equivalent to the canonical submersion
near x.
The proof is left as an exercise for the readers.

One of the most valuable implications of the local classification theorem con-
cerns the geometric nature of solutions of the equation f (x) = y comprise a subset
2.3. SUBMERSIONS AND REGULARITY 15
of X that is called the preimage of y, denoted f −1 (y). For general maps, the set
f −1 (y) need not be at all reasonable as a geometric object. But suppose that f is a
submersion at a point x ∈ f −1 (y). Select local coordinates around x and y so that
f (x1 , · · · , xk ) = (x1 , · · · , xl ).
and y corresponds to (0, · · · , 0). Thus near x, f −1 (y) is just the set of points
(0, · · · , 0, xl+1 , · · · , xk ). More precisely, let V denote the neighborhood of x on
which the coordinate system (x1 , · · · , xk ) is defined. Then f −1 (y) ∩ V is the set
of points where x1 = 0, · · · , xl = 0. The functions xk+1 , · · · , xl therefore form a
coordinate system on the set f −1 (y) ∩ V , which is a open subset of f −1 (y).
We are led to another definition:
Definition 2.3.0.2. For a smooth map of manifolds f : X → Y , a point y ∈ Y is called

a regular value for f if dfx : Tx (X) → Ty (Y ) is surjective at every point x such that
f (x) = y. A point y ∈ Y that is not a regular value of f is called a critical value.
The argument above just given proves that
Theorem 2.3.0.2. (Preimage Theorem) If y is a regular value of f : X → Y , then the

preimage f −1 (y) is a submanifold of X, with dimf −1 (y) = dimX − dimY .
When dimX > dimY , the regularity of a value y means that f is a submersion
at each preimage point x ∈ f −1 (y).
It’s simpler to create submanifolds by the use of the preimage theorem. For
instance, consider the map f : Rk → R defined by
f (x) = |x|2
we have S k−1 is a k − 1 dimensional manifold.

A more powerful application of the theorem is provided by the orthogonal
group O(n). Consider the map f : M (n) → S(n) defined by f (A) = AAt . Thus
O(n) = f −1 (I), where I is a regular value of f . Thus O(n) is a submanifold of
M (n) as well as a group.
Definition 2.3.0.3. A group that is a smooth manifold, and whose group operations are
smooth, is called a Lie group.
Now consider something that will be useful in the next section. Here is a
practical sort of problem: suppose g1 , · · · , gl are smooth, real-valued functions on
a manifold X of dimension k ≥ l. Under what condition is the set Z of common
zeros a reasonable geometric object?
Definition 2.3.0.4. If the l functionals d(g1 )x , · · · , d(gl )x are linearly independent on

Tx (X), we say that the l functions g1 , · · · , gl are independent at x.
By preimage theorem, we have
Proposition 2.3.0.1. If the smooth, real-valued functions g1 , · · · gl on X are independent

at each point where they all vanish, then the set Z of common zero is a submanifold of X
with dimension equal to dimX − l.
It’s convenient here to define the codimension of an arbitrary submanifold Z

of X by the formula codimZ = dimX − dimZ. Thus l independent functions on X
cut out a submanifold of codimension l.
However, the converse is not true. That is, not every submanifold Z of X
can be cut out by independent functions. However, we can get two useful partial
converses:
Partial Converse 1. If y is a regular value of a smooth map f : X → Y , then the

preimage submanifold f −1 (y) can be cut out by independent functions.
The proof is obvious.
Partial Converse 2. Every submanifold of X is locally cut out by independent functions.
The proof is based on a claim: there exist l independent functions g1 , · · · , gl

defined on some open neighborhood W of z in X such that Z ∩ W is the common
vanishing set of the gi . The detail is left to the readers.
2.4. TRANSVERSALITY 17
In particular, taking Z to be an arbitrary Euclidean space, we note that every

manifold is locally definable by a collection of independent functions in Euclidean
space.
In the end, it’s useful to note the following
Proposition 2.3.0.2. Let Z be the preimage of a regular value y ∈ Y under the smooth
map f : X → Y . Then the kernel of the derivative dfx : Tx (X) → Ty (Y ) at any point
x ∈ Z is precisely the tangent space to Z, Tx (Z).
Remark 4. (Stack of Records Theorem) Suppose that y is a regular value of f : X →

Y , where X is compact and has the same dimension as Y , then f −1 (y) is a finite set
{x1 , · · · , xN }. There exists a neighborhood U of y in Y such that f −1 (U ) is a disjoint
union V1 ∪ · · · ∪ VN , where Vi is an open neighborhood of xi and f maps each Vi diffeo-
morphically onto U .
2.4 Transversality
We have observed that the solution of f (x) = y form a smooth manifold,
provided that y is a regular value of f .
Now assume Z to be a submanifold of Y , and examine the set of solutions of
the relation f (x) ∈ Z. This lead a new differential property.
For if y = f (x), we may write Z in a neighborhood of y as the zero set of
a collection of independent functions g1 , · · · , gl (by the claim in 2.3), l being the
codimension of Z in Y . Then near x, the preimage f −1 (Z) is the zero set of the
function g1 ◦ f, · · · , gl ◦ f . We may apply the results already obtained: (g ◦ f )−1 (0)
is guaranteed to be a manifold when 0 is a regular value of g ◦ f . Thus we get the
condition:
Image(dfx ) + Ty (Z) = Ty (Y ).
Definition 2.4.0.1. The map f is said to be transversal to the submanifold Z, abbrevi-

ated f > Z, if the equation following holds true at each point x in the preimage of Z.
Image(dfx ) + Ty (Z) = Ty (Y ).
the argument above has proved:
Theorem 2.4.0.1. If the smooth map f : X → Y is transversal to a submanifold Z ⊂ Y ,

then the preimage f −1 (Z) is a submanifold of X. Moreover, the codimension of f −1 (Z) in
X equals the codimension of Z in Y .
It is easy to prove that the above results are completely consistent with the
conclusion in 2.3 when Z is just a single point.
We consider a special situation that the transversality of the inclusion map
i of one submanifold Z ⊂ Y . To say a point x ∈ X belongs to the preimage
i−1 (Z) simply means that x beolongs to the intersection X ∩Z. Also, the derivative
dix : Tx (X) → Tx (Y ) is merely the inclusion map of Tx (X) into Tx (Y ). so i > Z if
and only if for every x ∈ X ∩ Z,
Tx (X) + Tx (Z) = Tx (Y ).
the equation is symmetric in X and Z. Thus when it holds, we say that the two
submanifolds X and Z are transversal, and write X > Z.
Theorem 2.4.0.2. The intersection of two transversal submanifolds of Y is again a sub-

manifold. Moreover,
codim(X ∩ Z) = codimX + codimZ.
Notice that if the dimensions of X and Z do not add up to at least the dimen-
sion of Y , then they can only intersect transversally by not intersecting at all.
2.5 Homotopy and Stabilty

The definition of homotopy is very clear and will not be repeated here.
2.5. HOMOTOPY AND STABILTY 19
Now we will think about the stability of all the differential properties that
we talked earlier. The only physically meaningful properties of a mapping, con-
sequently, are those that remain vaild when the map is sightly deformed. Such
properties are stable properties, and a collection of maps that possess a particular
stable properties may be referred to as a stable class of maps.
Definition 2.5.0.1. A properties is stable provided that whenever f0 : X → Y possesses
the property and ft : X → Y is a homotopy of f0 , then, for some > 0, each ft with t <
also possesses the property.
This definition brings us immediately to the tube lemma. Thus we usually
need X is compact.
Lemma 2.5.0.1. Let X and Y be topological spaces, and assume that Y is compact. If
x ∈ X, and U is an open set in X ×Y containing {x}×Y , then there exists a neighborhood
W of x in X such that W × Y ⊂ U .
The proof can be seen in Colin Adams and Robert Franzosa’s Introduction to
Topology—pure and applied.
Now we give that all the differential properties of maps X → Y discussed so
far are stable, provided that X is compact.
Theorem 2.5.0.1. The following classes of smooth maps of a compact manifold X into a
manifold Y are stable classes:
(a) local diffeomorphisms;
(b) immersions;
(c) submersions;
(d) maps transversal to any specified closed submanifold Z ⊂ Y ;
(e) embeddings;
(f) diffeomorphisms.
The proof is left to the readers.
2.6 Sard’s Theorem and Morse Functions
2.6.1 Sard’s Theorem

We may think that the condition is so strong that regular values occur too
rarely for our preimage theorem to be of much use. In fact, precisely the opposite
is true.
Theorem 2.6.1.1. (Sard’s Theorem) If f : X → Y is any smooth map of manifolds, then

almost every point in Y is a regular value of f . In other words, the set of critical values
has measure zero.
The measure 0 of the set in Y is obviously defined by local coordinates, which

is well-defined since
Lemma 2.6.1.1. Let U be an open set of Rn , and let f : U → Rn be a smooth map. If

A ⊂ U is of measure zero, then f (A) is of measure zero.
the proof of Sard’s Theorem:

We give some lemmas firstly without proofs:
Lemma 2.6.1.2. The natural copy of Rn−1 inside Rn -namely, {(x1 , · · · , xn−1 , 0)} has
measure 0.
Lemma 2.6.1.3. (Fubini Theorem) Let A be a closed subset of Rn such that A ∩ Vc has
measure 0 in Vc for all c ∈ Rk (Vc is the vertical slice {c} × Rl , n = k + l). Then A has
measure 0 in Rn .
By lemma 2.6.1 and lemma 2.6.3 we can prove that
Lemma 2.6.1.4. (Mini Sard) Let U be an open subset of Rn , and let f : U → Rm be a

smooth map. Then if m > n, f (U ) has measure 0 in Rm .
now we are ready for the task of proving Sard’s theorem:

By C 2 we can find a countable collection of open sets (Ui , Vi ), Ui open in X and
Vi open in Y such that Ui0 s cover X, f (Ui ) ⊂ Vi , and Ui0 s and Vi0 s are diffeomorphic
to open sets in Rn . Therefore it suffices to prove
2.6. SARD’S THEOREM AND MORSE FUNCTIONS 21
Theorem 2.6.1.2. Suppose that U is open in Rn and f : U → Rp is a smooth map. Let C

be the set of critical points of f . Then f (C) is of measure zero in Rp .
We prove it step by step:
(a) the theorem is certainly true for n = 0, by induction, we assume it’s true for
n − 1 and prove it for n.
(b) Partition C into a sequence of nested subsets C ⊃ C1 ⊃ C2 ⊃ · · · , where C1

is the set of all x ∈ U such that (df )x = 0, and Ci for i ≥ 1 is the set of all x
such that the partial derivatives of f of order ≤ i vanish at x.
(c) The image f (C − C1 ) measures zero.
(d) The image f (Ck − Ck+1 ) measures zero for k ≥ 1.
(e) For k > n/p − 1, f (Ck ) is of measure zero.
Thus the proof is complete.

Let’s illustrate the use of Sard’s theorem with a typical application. We con-
sider only smooth functions on a manifold X. for any x ∈ X, f is either regular or
dfx = 0. If it’s regular, then we can choose a coordinate system around x so that f
is simply the first coordinate function (local submersion theorem). But what can
we say at critical points?
We consider the Hessian matrix of second partials
∂ 2f
H=( )
∂xi ∂xj
suppose it’s nonsingular at x, we say that x is a nondegenerate critical point of f .
Definition 2.6.1.1. Suppose x is a critical point of f and the Hessian matrix at x is

nonsingular, we say that x is a nondegenerate critical point of f .
Proposition 2.6.1.1. Nondegenerate critical points are isolated.

∂f ∂f
use g(x) = ( ∂x 1
, · · · , ∂xk
) we can easily prove that.
2.6.2 Morse Functions

Lemma 2.6.2.1. (Morse Lemma) Suppose that the point a ∈ Rk is a nondegenerate
critical point of the function f , and
∂ 2f
(hij ) = ( (a))
∂xi ∂xj
is the Hessian of f at a. Then there exists a local coordiniate system (x1 , · · · , xk ) around
a such that
X
f = f (a) + hij xi xj
near a.
thus every function near a nondegenerate critical point is locally equivalent

to a quadratic polynomial.
The concept of nondegeneracy also make sense on manifolds by local coordi-
nates. We just need:
Lemma 2.6.2.2. Suppose that f is a function on Rk with a nondegenerate critical point at

0, and ψ is a diffeomorphism with ψ(0) = 0. Then f ◦ ψ also has a nondegenerate critical
point at 0.
Thus the Morse lemma complete describes the behavior of functions at non-
degenerate points.
Definition 2.6.2.1. the function whose critical points are all nondegenerate is called a
Morese function.
Details on the Morse functions are available in J.Minlor’s Morse Theory.

Now suppose that the manifold X sits in RN , and let x1 , · · · , xN be the usual
coordinate functions on RN . If f is a function on X and a = (a1 , · · · , aN ), we define
a new function fa = f + a1 x1 + · · · + aN xN . then we have
Theorem 2.6.2.1. No matter what the function f : X → R is , for almost every a ∈ Rn

the function fa is a Morse function on X.
2.7. WHITNEY’S EMBEDDING THEOREM AND PARTITION OF UNITY 23
First we prove in Rk :
Lemma 2.6.2.3. Let f be a smooth function on an open set U of Rk . Then for almost all a
in Rk , the function fa is a Morse function on U .
This is easily proved by the above conclusions.

proof of theorem: suppose that x is any point in X and that x1 , · · · , xN are
the standard coordinate functions on RN . Then the restrictions of some of these
coordinate functions xl1 , · · · , xlk to X constitute a coordinate system in a neigh-
borhood of x.
Therefore we cover X with open set Uα such that on each α some k of the
functions x1 , · · · , xN form a coordinate system. Moreover, by Lindelof Theorem we
may assume there are only countably many Uα .
Now suppose that Uα is one of these sets, and for convenience, assume that
(x1 , · · · , xk ) is a coordinate system on Uα , For each N − k tuple c = (ck+1 , · · · , cN ),
consider
f(0,c) = f + ck+1 xk+1 + · · · + cN xN .
by the lemma , for almost b ∈ Rk , the function
f(b,c) = f(0,c) + b1 x1 + · · · + bk xk .
is a Morse function on Uα . Suppose Sα be the set of points a in RN such that fα is

not a Morse function on Uα . We have shown that Sα ∩ Rk × {c} measures zero. By
S
Fubini Theorem we have Sα measures zero. Thus S = Sα measures zero. Thus
α
the proof is complete.
2.7 Whitney’s Embedding Theorem and Partition of

Unity
A k-dimensional manifold X has been defined as a subset of some Euclidean
space Rn that may be enormous compared to X. This ambient Euclidean space is
rather arbitrary when we consider the manifold X as an abstract object. How large
N must be in order that RN contain a diffeomorphic copy of every k-dimensional
manifold. Whitney’s preliminary answer was that N = 2k + 1; this is the result
we shall prove. After a great deal of hard work, Whitney improved his result by
one, establishing that every k-dimensional manifold actually embeds in R2k . And
the Klein bottle illustrates that his result is optimal.
Figure 2.1: an immersion of Klein bottle in R3
A useful object in proving the theorem is the tangent bundle of a manifold X

N
in R .
Definition 2.7.0.1. For a manifold X in RN , its tangent bundle T (X) is a subset of

X × RN is defined by
T (X) = {(x, v) ∈ X × RN : v ∈ Tx (X)}.
T (X) contains a natural copy X0 of X, consisting of the points (x, 0). In the
direction perpendicular to X0 , it contains copies of each tangent space Tx (X), em-
bedded as the sets {(x, v) : with x fixed}.
Any smooth map f : X → Y induces a global derivative map: df : T (X) →
T (Y ), defined by f (x, v) = (f (x), dfx (v)). We can easily find that df is smooth and
T (X) is a manifold in R2N . Diffeomorphic manifolds have diffeomorphic tangent
bundles; As a result, T (X) is an object intrinsically associated to X.
Proposition 2.7.0.1. The tangent bundle of a manifold is another manifold, and dimT (X) =
2dimX.
We first prove a version of Whitney’s result.
Theorem 2.7.0.1. Every k-dimensional manifold admits a one-to-one immersion in R2k+1 .
Proof. If X ⊂ RN is k-dimensional and N > 2k + 1, we shall produce a linear

projection RN → R2k+1 that restricts to a one-to-one immersion of X.
Proceeding inductively, we prove that if f : X → RM is an injective immer-
sion with M > 2K + 1, then there exists a unit vector a ∈ RM such that the compo-
sition of f with the projection map carrying RM onto the orthogonal complement
of a is still an injective immersion. The complement H = {b ∈ RM : b ⊥ a} is an
M − 1 dimensional vector space of RM , hence isomorphic to RM −1 ; thus we obtain
an injective immersion into RM −1 .
Define a map h : X × X × R → RM by h(x, y, t) = t[f (x) − f (y)]. Also, define a
map g : T (X) → RM by g(x, v) = dfx (v). Since M > 2k + 1. Sard’s theorem implies
that there exists a point a ∈ RM belonging to neither their image; note that a 6= 0,
since 0 belongs to both images.
Let π be the projection, certainly π ◦ f : X → H is injective immersion. Thus
the proof is complete.
Thus for compact manifold we have proved the theorem. In general, we must
modify the immersion to make it proper, which is a topological thing.
The fundamental trick used for such generalizations is the following theorem.
Theorem 2.7.0.2. Let X be an arbitary subset of RN . For any covering of X by open

subsets {Uα }, there exists a sequence of smooth functions {θ i } on X, called a partition
of unity subordinate to the open cover {Uα }, with the following properties:
(a) 0 ≤ θi (x) ≤ 1 for all x ∈ X and all i;
(b) Each x ∈ X has a neighborhood on which all but finitely many functions θi are
identically zero;
(c) Each function θi is identically zero except on some closed set contained in one of the
Uα ;
P
(d) For each x ∈ X, θi (x) = 1.
i
Proof. Each Uα may be written as X ∩ Wα for some open set Wα in the ambient
Euclidean space RN . Set W =
S
Wα , and let {Kj } be any nested sequence of
α
compact sets exhausting the open set W . That is
∞
[
Kj = W.
j=1
and Kj ⊂ Int(Kj+1 ). The collection of all open balls of RN whose closures belong
to at least one Wα is an open cover of W . Select a finite number of such balls that
cover the set K2 . Each ball selected we may find a smooth nonnegative function
on RN that is identically one on that ball and zero outside a closed set contained
in one of the Wα . Call these functions η1 , · · · , ηr .
We continue building a sequence of functions inductively. For each j ≥ 3,
the compact set Kj − Int(Kj−1 ) is contained inside the open set W − Kj−2 .The
collection of all open balls small enouth to have their closures contained both in
W − Kj−2 and in some Wα forms an open cover of Kj − Int(Kj−2 ). Extract a
finite subcover, and then add to our sequence {ηi } one function for each ball; the
function is to be equal to one on the ball and zero outdside a closed set contained
in both W − Kj−2 and in one of the Wα .
By construction, for each j only finitely many functions ηi fail to vanish on
Kj . Thus since every point of W belongs to the interior of some Kj , the sum
∞
X
ηj
j=1
is actually finite in a neighborhood of every point of W . Furthermore, at least one

term is nonzero at any point of W . Therefore
ηi
∞
P
ηj
j=1
is well-defined and smooth. If we let θi be the restriction of this to X, then we are

done.
Corollary 2.7.0.1. On any manifold X there exists a proper map ρ : X → R.
∞
P
the proof is simply consider the map ρ = iθi , where {Uα } is the collection
i=1
of open subsets of X that have compact closure, θi is a subordinate partition of
unity.
Then we can prove the theorem:
Theorem 2.7.0.3. (Whitney Theorem) Every k-dimensional manifold embeds in R2k+1 .
Proof. Begin with a one-to-one immersion of X into R2k+1 . Composing with any
diffeomorphism of R2k+1 into its unit ball-say, z → z/(1 + |z|2 ) we obtain an injec-
tive immersion f : X → R2k+1 such that |f (x)| < 1 for all x ∈ X. Let ρ : X → R
be a proper function, and define a new injective immersion F : X → R2k+2 by
F (x) = (f (x), ρ(x)). Now drop back down to R2k+1 as in the earlier theorem by
composing F with an orthogonal projection π : R2k+1 → H, where H is the linear
space perpendicular to a suitable unit vector, a in R2k+2 .
Since the map π ◦ F : X → H is still an injective immersion for almost every
a ∈ S 2k+1 , so we may pick an a that happens to be neither of the sphere’s two
poles. Now π ◦ F is easily seen to be proper.
In fact, given any bound c, we claimed that there exists another number d
such that the set of points x ∈ X where |π ◦ F (x)| ≤ c is contained in the set where
|ρ(x)| ≤ d.
If the claim is false, then there exists a sequence of points {xi } in X for which
|π ◦ F (xi )| < c but ρ(xi ) → ∞. By definition, for each z ∈ R2k+2 the vector z − π(z)
is a multiple of z. Thus F (xi ) − π ◦ F (xi ) is a multiple of a for each i, and hence so
is the vector
1
wi = [F (xi ) − π ◦ F (xi )].
ρ(xi )
Consider as i → ∞,
F (xi )
→ (0, · · · , 0, 1).
ρ(xi )
because |f (xi )| < 1 for all i. The quotient
π ◦ F (xi )
ρ(xi )
has norm ≤ c/ρ(xi ), so it converges to zero. Thus wi → (0, · · · , 0, 1), a contradic-

tion! Thus the proof is complete.
Remark 5. • (The Whitney Immersion Theorem) Every k-dimensional manifold X

may be immersed in R2k .
• (The Smooth Urysohn Theorem) If A and B are disjoint, smooth, closed subsets
of a manifold X, then there is a smooth function φ on X such that 0 ≤ φ ≤ 1 with φ = 0
on A and φ = 1 on B.
Chapter 3
Transversality and Intersection
O peak of peaks, how high it stands!

One boundless green o’erspreads two States.
A marvel done by Nature’s hands,
O’er light and shade it dominates.
Clouds rise therefrom and lave my breast;
My eyes are strained to see birds fleet.
I must ascend the mountain’s crest,
It dwarfs all peaks under feet.
—Fu Du
In this chapter we will continue to discuss transversality, the most important topic
in differential topology. And we’ll also introduce the theory of intersection, an
interesting part in manifolds.
3.1 Manifolds with Boundary

We now enlarge the class of geometric objects under study by allowing our
manifolds to possess boundaries.
29
30 CHAPTER 3. TRANSVERSALITY AND INTERSECTION
Definition 3.1.0.1. A subset X of RN is a k-dimensional manifold with boundary if

every point of X possesses a neighborhood diffeomorphic to an open set in the space Hk .
As before, such a diffeomorphism is called a local parametrization of X. The boundary
of X, denoted ∂X, consists of those points that belong to the image of the boundary of Hk
under some local parametrization. Its complement is called the interior of X, Int(X) =
X − ∂X.
Remark 6. • The boundary or interior of X we define here is not always the topo-
logical boundary or interior of X as a subset of RN . The notions often agree when
dimX = N , they definitely do not correspond when dimX < N .
• The product of two manifolds with boundary is not generally another manifold with
boundary, as the square [0, 1] × [0, 1] illustrates.
But at least the following proposition is true:
Proposition 3.1.0.1. The product of a manifold without boundary X and a manifold with
boundary Y is another manifold with boundary. Furthermore,
∂(X × Y ) = X × ∂Y,
and
dim(X × Y ) = dimX + dimY.
tangent spaces and derivatives are still defined in the setting of manifolds
with boundary.
If X is a manifold with boundary, Int(X) is automatically a boundaryless
manifold of the same dimension as X. The reason is that any interior point is in
the range of a local parametrization whose domain is an open set of Hk contained
entirely in Int(Hk ) and therefore an open set of Rk . And more interesting,
Proposition 3.1.0.2. If X is a k-dimensional manifold with boundary, then ∂X is a

(k − 1) dimensional manifold without boundary.
3.1. MANIFOLDS WITH BOUNDARY 31
Observe that if x ∈ ∂X, then the tangent space to the boundary Tx (∂X) is a
linear subspace of Tx (X) with codimension 1. For any smooth map f defined on
X, let us introduce the notation ∂f for the restriction of f to ∂X. The derivative of
∂f at x is just the restriction of dfx to the subspace Tx (∂X).
We would like conditions that would guarantee that if f : X → Y encoun-
ters a submanifold Z of Y , then f −1 (Z) is a manifold with boundary. We also
want ∂f −1 (Z) = f −1 (Z) ∩ ∂X. Unfortunately, the transversality of f alone doesn’t
guarantee this. We need to add more conditions.
Theorem 3.1.0.1. Let f be a smooth map of a manifold X with boundary onto a bound-
aryless manifold Y , and suppose that both f : X → Y and ∂f : ∂X → Y are transversal
with respect to a boundaryless submanifold Z in Y . Then the preimage f −1 (Z) is a mani-
fold with boundary
∂[f −1 (Z)] = f −1 (Z) ∩ ∂X.
and the codimension of f −1 (Z) in X equals the codimension of Z in Y .
The proof turns out to be very technical, and I’ll talk about it in more detail
in class.
The generalization of Sard’s theorem to manifolds with boundary is more
straightforward.
Theorem 3.1.0.2. (Sard’s theorem) For any smooth map f of a manifold X with bound-
ary into a boundaryless manifold Y , almost every point of Y is a regular value of f : X →
Y and ∂f : ∂X → Y .
Remark 7. There are precisely two unit vectors in Tx (X) that are perpendicular to
Tx (∂X) and that one lies inside Hx (X), the other outside. The one in Hx (X) is called
the inward unit normal vector to the boundary, and the other is the outward unit
normal vector to the boundary.
3.2 One-Manifolds and Some Consequences

We simply assert:
Theorem 3.2.0.1. (The Classification of One-Manifolds) Every compact, connected,

one-dimensional manifold with boundary is diffeomorphic to [0, 1] or S 1 .
Our proof is based on Morse functions. For a proof based on arc length in-
stead, see Milnor’s Topology from a Differential Viewpoint.
Lemma 3.2.0.1. (Smoothing Lemma) Let g be a function on [a, b] that is smooth and
has positive derivative everywhere except at one interior point c. Then there exists a glob-
ally smooth function ge that agrees with g near the endpoints and has positive derivative
everywhere.
Proof. Let ρ be a smooth negative function that vanishes outside a compact subset
Rb
of (a, b), which equals 1 near c, and which satisfies a ρ = 1. Define
Z x
ge(x) = g(x) + [kρ(s) + g 0 (s)(1 − ρ(s))]ds,
a
where the constant

Z b
k = g(b) − g(a) − g 0 (s)(1 − ρ(s))ds.
a
then ge is what we need.
to prove the theorem, choose a Morse function f on X. Let S be the union of

the critical points of f and the boundary points of X. As S is finite, X − S consists
of a finite number of connected one-manifolds L1 , · · · , LN .
The rest parts we will simply list:
Proposition 3.2.0.1. f maps each Li diffeomorphically onto an open interval in R1 .
Lemma 3.2.0.2. Let L be a subset of X diffeomorphic to an open interval of R1 , where

dimX = 1. Then its closure L̄ contains at most two points not in L.
3.2. ONE-MANIFOLDS AND SOME CONSEQUENCES 33
We will call a sequence L1 , · · · , Lk a chain, provided that each consecutive pair

L̄j and L̄j+1 have a common boundary point pj , j = 1, · · · , k − 1. Let p0 denote the
other boundary point L1 , and pk the other boundary point for Lk . Since there are
only finite many Li altogether, there clearly exists a maximal chain, a chain that
cannot be extended by appending another Li . We finish our proof by establishing
the
Claim. If L1 , · · · , Lk is a maximal chain, then it contains every Li . If L̄k and L̄0 have
a common boundary point, then X is diffeomorphic to the circle; if not, X is diffeomorphic
to a closed interval.
As every compact one-manifold with boundary is the disjoint union of finitely
many connected components, we obtain a trivial corollary that will have surpris-
ingly nontrivial applications.
Corollary 3.2.0.1. The boundary of any compact one-dimensional manifold with bound-
ary consists of an even number of points.
the first use is
Theorem 3.2.0.2. (Retraction Theorem) If X is any compact manifold with boundary,

then there exists no smooth map g : X → ∂X such that ∂g : ∂X → ∂X is the identity.
That is, there is no "retraction" of X onto its boundary.
Proof. Suppose that such a g exists, and let z ∈ ∂X be a regular value. Then g −1 (z)
is a submanifold of X with boundary. As the codimension of g −1 (z) in X equals
the codimension of {z} in ∂X, namely dimX − 1, g −1 (z) is one dimensional and
compact. But since ∂g =identity,
∂g −1 (z) = g −1 (z) ∩ ∂X = {z},
contradicting the corollary.
We now prove a famous theorem of Brouwer:

Theorem 3.2.0.3. (Brouwer Fixed-Point Theorem) Any smooth map f of the closed
unit ball B n ⊂ Rn into itself must have a fixed point; that is, f (x) = x for some x ∈ B n .
Proof. Suppose that there exists an f without fined points, and we shall construct
a retraction g : B n → ∂B n . Since f (x) 6= x, the two points x and f (x) determine
a line. Let g(x) be the point where the line segment starting at f (x) and passing
through x hits the boundary. If x ∈ ∂B n already, g(x) ≡ x. Thus g : B n → B n is
the identity on ∂B n . We only need show that g is smooth to obtain a contradiction
to the retraction theorem and thereby complete the proof. Since x is in the line
segment between f (x) and g(x), we may write the vector g(x) − f (x) as a multiple
t times the vector x − f (x), where t ≥ 1. Thus g(x) = t(x) + (1 − t)f (x). If t
depends smoothly on x, then g(x) is smooth. Take the dot product of both sides
of this formula. Because |g(x)| = 1, we have
t2 |x − f (x)|2 + 2tf (x) · [x − f (x)] + |f (x)|2 − 1 = 0.
solve the equation we have that t is smooth to x. Thus the proof is complete.
3.3 Transversality
In this section we will go into more detail about transversality. Earlier we
proved that transversality is a property that is stable under small perturbations,
at least for maps with compact domains. From Sard’s theorem we shall deduce
the much more subtle map f : X → Y , no matter how bizarre its behavior with
respect to a given submanifold Z in Y , may be deformed by an arbitrary small
amount into a map that is transversal to Z.
The central theorem is
Theorem 3.3.0.1. (The Transversal Theorem) Suppose that F : X ×S → Y is a smooth

map of manifolds, where only X has boundary, and let Z be any boundaryless submanifold
of Y . If both F and ∂F are transversal to Z, then for almost every s ∈ S, both fs and ∂fs
are transversal to Z.
Proof. The preimage W = F −1 (Z) is a submanifold of X × S with boundary ∂W =

W ∩ ∂(X × S). Let π : X × S → S be the natural projection map. We shall show
that whenever s ∈ S is a regular value for the restriction map π : W → S, then
fs > Z, and whenever s is a regular value for ∂π : ∂W → S, then ∂fs > Z. By Sard’s
theorem, almost every s ∈ S is a regular value for both maps, so the theorem
follows.
In order to show that fs > Z, suppose that fs (x) = z ∈ Z. Because F (x, s) = z
and F > Z, we know that
dF(x,s) T(x,s) (X × S) + Tz (Z) = Tz (Y ).
that is, given any vector a ∈ Tz (Y ), there exists a vector b ∈ T(x,s) (X × S) such that
dF(x,s) (b) − a ∈ Tz (Z).
We want to exhibit a vector v ∈ Tx (X) such that dfs (v) − a ∈ Tz (Z). Now
T(x,s) (X × S) = Tx (X) × Ts (S).
so b = (w, e) for vector w ∈ Tx (X) and e ∈ Ts (S). If e = 0 we would be done, for

since the restriction of F to X × {s} is fs , it follows that
dF(x,s) (w, 0) = dfs (w).
Although e need not be zero, we may use the projection π to kill it off. As
dπ(x,s) : Tx (X) × Ts (S) → Ts (S)
is just projection onto the second factor, the regularity assumption that dπ(x,s) map
T(x,s) (W ) onto Ts (S) tells us that there is some vector of the form (u, e) in T(x,s) (W ).
But F : W → Z, so dF(x,s) (u.e) ∈ Tz (Z). Consequently, the vector v = w − u ∈
Tx (X) is our solution. For
dfs (v) − a = dF(x,s) [(w, e) − (u, e)] − a = [dF(x,s) (w, e) − a] − dFs (u, e).
and both of the latter vectors belong to Tz (Z).

Precisely the same argument shows that ∂fs > Z when s is a regular value of
∂π. Thus the proof is complete.
The transversal theorem easily implies that transversal maps are generic when
the target manifold Y is a Euclidean space RM .
Theorem 3.3.0.2. If f : X → RM is any smooth map, then f can be deformed into a

transversal map.
Proof. Take S to be an open ball of RM itself and then define F : X × S → RM by

F(x,s) = f (x) + s. For any fixed x ∈ X, F is a translation of the ball S, obviously a
submersion. So, of course, F is a submersion of X × S and therefore transversal
to any submanifold Z of RM . By Transversal Theorem we have for almost every
s ∈ S, the map fs (x) = f (x) + s is transversal to Z. Thus f may be deformed into
a transversal map by the simple addition of arbitrary small quantity s.
For an arbitrary, boundaryless, target manifold Y , we must understand a little

of the geometry of Y with respect to its environment. As usual, the compact case
is clearest.
Theorem 3.3.0.3. (-Neighborhood Theorem) For a compact boundaryless manifold Y

in RM and a positive number , let Y be the open set of points in RM with distance less
than from Y . If is sufficiently small, then each point w ∈ Y possesses a unique closest
point in Y , denoted π(w). Moreover, the map π : Y → Y is a submersion. When Y
is not compact, there still exists a submersion π : Y → Y that is the identity on Y ,
but now must be allowed to be a smooth positive function on Y , and Y is defined as
{w ∈ RM : |w − y| < (y) for some y ∈ Y }.
We postpone the proof for a moment.
Corollary 3.3.0.1. Let f : X → Y be the smooth map, Y being boundaryless. Then there
is an open ball S in some Euclidean space and a smooth map F : X × S → Y such that
F (x, 0) = f (x), and for any fixed x ∈ X the map s → F (x, s) is a submersion S → Y .
In particular, both F and ∂F are submersions.
Proof. just let F (x, s) = π[f (x) + (f (x))s].
The transversality is generic follows directly.
Theorem 3.3.0.4. (Transversality Homotopy Theorem) For any smooth map f : X →

Y and any boundaryless submanifold Z of the boundaryless manifold Y , there exists a
smooth map g : X → Y homotopic to f such that g > Z and ∂g > Z.
Proof. For the family of mappings F of the corollary, the Transversality Theorem
implies that fs > Z and ∂fs > Z for almost all s ∈ S. But each fs is homotopic to f ,
the homotopy X × I → Y being (x, t) → F (x, ts).
To prove the -Neighborhood Theorem, we introduce an artifice similar to the

tangent bundle.
Definition 3.3.0.1. For each y ∈ Y , define Ny (Y ) the normal space of Y at y to be the

orthogonal complement of Ty (Y ) in RM . The normal bundle N (Y ) is then defined to be
the set
{(y, v) ∈ Y × RM : v ∈ Ny (Y )}.
Note that unlike T (X), N (Y ) is not intrinsic to the manifold Y but depends
on the specific relationship between Y and the surrounding RM . There is a natural
projection map σ : N (Y ) → Y defined by σ(y, v) = y.
In order to show that N (Y ) is a manifold, we must recall an elementary fact
from linear algebra.
Lemma 3.3.0.1. Suppose that A : RM → Rk is a linear map. If A is surjective, then At

maps Rk isomorphically onto the orthogonal complement of the kernel of A.
Proposition 3.3.0.1. If Y ⊂ RM , then N (Y ) is a manifold of dimension M and the

projection σ : N (Y ) → Y is a submersion.
proof of the -Neighborhood Theorem.

Let h : N (Y ) → RM be h(y, v) = y + v. Notice that h is regular at every
point of Y × {0} in N (Y ), for through (y, 0) there pass two natural complementary
manifolds of N (Y ), i.e., Y × {0} and {y} × Ny (Y ). The derivative of h at (y, 0)
maps the tangent space of Y × {0} at (y, 0) onto Ty (Y ) and maps the tangent space
of {y} × Ny (Y ) at (y, 0) onto Ny (Y ). Therefore it maps onto Ty (Y ) + Ny (Y ) = RM .
Since h mps Y × {0} diffeomorphically onto Y and is regular at each (y, 0), it
must map a neighborhood of Y ×{0} diffeomorphically onto a neighborhood of Y
in RM . Now any neighborhood of Y contains some Y ; this point is obvious when
Y is compact and easy to show in general. Thus h−1 : Y → N (Y ) is defined, and
π = σ ◦ h−1 : Y → Y is the desired submersion. Thus the proof is complete.
We will need a somewhat stronger from the Transversality Homotopy Theo-
rem.
Definition 3.3.0.2. A map f : X → Y is transversal to Z on a subset C of X if the

transversality condition
dfx Tx (X) + Tf (x) (Z) = Tf (x) (Y )
is satisfied at every point x ∈ C ∩ f −1 (Z).
Theorem 3.3.0.5. (Extension Theorem) Suppose that Z is a closed submanifold of Y ,

both boundaryless, and C is a closed subset of X. Let f : X → Y be a smooth map with
f > Z on C and ∂f > Z on C ∩∂X. Then there exists a smooth map g : X → Y homotopic
to f , such that g > Z, ∂g > Z, and on a neighborhood of C we have g = f .
Lemma 3.3.0.2. If U is an open neighborhood of the closed set C in X, then there exists
a smooth function γ : X → [0, 1] that is identity equal to one outside U but is zero on a
neighborhood of C.
Proof. Let C 0 be any closed set contained in U that contains C in its interior, and let
{θi } be a partition of unity subordinate to the open cover {U, X − C 0 } of X. Then
just take γ to be the sum of those θi that vanish outside of X − C 0 .
3.4. INTERSECTION THEORY MOD 2 39
the proof of theorem is left to the readers.

Since ∂X is always closed in X, we obtain the special case:
Corollary 3.3.0.2. If, for f : X → Y , the boundary map ∂f : ∂X → Y is transversal to

Z, then there exists a map g : X → Y homotopic to f such that ∂g = ∂f and g > Z.
the corollary may be interpreted in another useful form:
Corollary 3.3.0.3. Suppose h : ∂X → Y is a map transversal to Z. Then if h extends to

any map of the whole manifold X → Y , it extends to a map that is transversal to Z on all
of X.
Remark 8. (General Position Lemma) Let X and Y be submanifolds of RN , then for

almost every a ∈ RN the translate X + a intersect Y transversally.
3.4 Intersection Theory Mod 2

In this section we will use the transversality lemma and the other results to
develop a simple intuitive invariant for intersecting manifolds.
Two submanifolds X and Z inside Y have complementary dimension if dimX +
dimZ = dimY . If X > Z, the dimension condition makes their intersection X ∩ Z
a zero-dimension manifold. If we further assume that both X and Z are closed
and that at least one of them, say X is compact, then X ∩ Z must be a finite set
of points. Provisionally, we might refer to the number of points in X ∩ Z as the
"intersection number" of X and Z, indicated by #(X ∩ Z).
How can we define the intersection number of the compact X with an arbi-
trary closed Z of complementary dimension? Without transversality, X ∩ Z may
be some frowzy. But we can wiggle X, deforming it ever so slightly and make
it transversal to Z, and simply define the intersection number of X and Z to be
the intersection numbers we obtain after the plastic surgery. The difficulty is that
different alterations of X can produce different intersection numbers.
X 00 Z X Z X0 Z
Figure 3.1: two deformations of X, with different intersection numbers
Happily the idea is savable nonetheless, because as we shall see, the intersec-
tion numbers obtained with different deformations at least agree mod 2. Thus a
mod 2 intersection number of X and Z can be meaningfully defined.
We must deal with the necessary of deforming X in a mathematically pre-
cise manner. Considering X as an abstract manifold and its inclusion mapping
i : X ,→ Y simply as an embedding, we know using homotopy to deform i. Since
embedding form a stable class of mappings, any small homotopy of i gives us
another embedding X → Y and thus produces an image manifold that is a diffeo-
morphic copy of X adjacent to the original.
Suppose that X is any compact manifold, not necessarily inside Y , and f :
X → Y is a smooth map transversal to the closed manifold Z in Y , where dimX +
dimZ = dimY . Then f −1 (Z) is a closed zero-dimensional submanifold of X, hence
a finite set. Define the mod 2 intersection number of the map f with Z, I2 (f, Z),
to be the number of points in f −1 (Z) modulo 2. For an arbitrary smooth map
g : X → Y , select any map f that is homotopic to g and transversal to Z, and
define I2 (g, Z) = I2 (f, Z). The ambiguity in the definition is remedied by:
Theorem 3.4.0.1. If f0 , f1 : X → Y are homotopic and both transversal to Z, then

I2 (f0 , Z) = I2 (f1 , Z).
Proof. Let F : X × I → Y be a homotopy of f0 and f1 . By the Extension Theorem,

we may assume that F > Z. Since ∂(X × I) = X × {0} ∪ X × {1} and ∂F is f0 on
X ×{0} and f1 on X ×{1}, ∂F > Z. Then F −1 (Z) is a one-dimensional submanifold
of X × I with boundary
∂F −1 (Z) = F −1 (Z) ∩ ∂(X × I) = f0−1 (Z) × {0} ∪ f1−1 (Z) × {1}.

3.4. INTERSECTION THEORY MOD 2 41
From the classification of one-manifolds, ∂F −1 (Z) must have an even number of

points, so #f0−1 (Z) = #f1−1 (Z) mod 2.
thus we immediately obtain
Corollary 3.4.0.1. If g0 , g1 : X → Y are arbitrary homotopic maps, then we have

I2 (g0 , Z) = I2 (g1 , Z).
Definition 3.4.0.1. If X is a compact submanifold of Y and Z a closed submanifold of

complementary dimension, we define the mod 2 intersection number of X with Z by
I2 (X, Z) = I2 (i, Z), where i : X ,→ Y is the inclusion.
when X > Z, then I2 (X, Z) is just #X ∩ Z mod 2. If I2 (X, Z) 6= 0, then no

matter how X is deformed it cannot be pulled entirely away from Z. For example,
consider S 1 × {0} and {0} × S 1 in S 1 × S 1 .
A curious situation prevails when dimX = 21 dimY , for then we may consider
I2 (X, X) the mod 2 self-intersection number of X. For example , suppose X is the
central curve of Mobius band.
If X happens to be the boundary of some W in Y , then I2 (X, Z) = 0.
Theorem 3.4.0.2. (Boundary Theorem) Suppose that X is the boundary of some com-
pact manifold W and g : X → Y is a smooth map. If g may be extended to all of W , then
I2 (g, Z) = 0 for any closed submanifold Z in Y of complementary dimension.
Proof. Let G : W → Y extend g; that is, ∂G = g. From the Transversal Ho-

motopy Theorem, we obtain a homotopic map F : W → Y , with F > Z and
f = ∂F > Z. Then f ∼ g, so I2 (g, Z) = #f −1 (Z) mod 2. But F −1 (Z) is a compact
one-dimensional manifold with boundary, so #∂F −1 (Z) = #∂f −1 (Z) is even.
Theorem 3.4.0.3. If f : X → Y is a smooth map of a compact manifold X into a

connected manifold Y and dimX = dimY , then I2 (f, {y}) is the same for all points
y ∈ Y . This common value is called the mod 2 degree of f , denoted deg2 (f ).
Proof. Given any y ∈ Y , alter f homotopically, if necessary, to make it transversal

to {y}. By the Stack of Records Theorem we can find a neighborhood U of y such that
the preimage f −1 (U ) is a disjoint union V1 ∪ · · · Vn , where each each Vi is an open
set in X mapped, by f diffeomorphically onto U . We conclude that I2 (f, {z}) = n
mod 2 for all z ∈ U . Consequently, the function defined on Y by y → I2 (f, {y}) is
locally constant. Since Y is connected, it must be globally constant.
and we immediately obtain
Theorem 3.4.0.4. Homotopic maps have the same mod 2 degree.
Theorem 3.4.0.5. If X = ∂W and f : X → Y may be extended to all of W , then

deg2 (f ) = 0.
Proposition 3.4.0.1. Suppose p is a smooth complex function and W is a smooth compact

region in the plane, a two-dimensional manifold with boundary. If the mod 2 degree of
p/|p| : ∂W → S 1 is nonzero, then the function p has a zero inside W .
Proof. by theorem 3.4.5.
Theorem 3.4.0.6. (One-half Fundamental Theorem of Algebra) Every complex poly-

nomial of odd degree has a root.
Proof. Suppose that

p(z) = z m + a1 z m−1 + · · · + am
is a monic complex polynomial, and define a homotopy by
pt = tp(z) + (1 − t)z m
= z m + t(a1 z m−1 + · · · + am ).
It’s easily seen that if W is a closed ball of sufficiently large radius, then none of
the pt have zeros on ∂W . For
pt (z) 1 1
m = 1 + t(a1 + · · · + am m )
z z z
3.5. WINDING NUMBERS AND THE JORDAN-BROUWER SEPARATION THEOREM43
and the term in parenthesis→ 0 as z → ∞. Thus the homotopy pt /|pt | : ∂W → S 1

is defined for all t, so we conclude that deg2 (p/|p|) = deg2 (p0 /|p0 |). The polynomial
p0 is just z m ; therefore every point in S 1 has precisely m preimage points in ∂W
under p0 /|p0 |, thus deg2 (p0 /|p0 |) = m mod 2. Thus the proof is complete.
3.5 Winding numbers and the Jordan-Brouwer Sepa-

ration Theorem
We have known that the Classical Jordan Curve Theorem says that every sim-
ple closed curve in R2 divides the plane into two pieces, the "inside" and "outside"
of the curve. Now we want to generalize this to n dimension.
Definition 3.5.0.1. A hypersurface in a manifold is a submanifold of codimension 1.
Begin with a compact, connected manifold X and a smooth map f : X → Rn .

Suppose that dimX = n − 1, so that, in particular, f might be the inclusion map of
a hypersurface into Rn . We wish to study how f wraps X around in Rn , so take
any point z of Rn not lying in the image f (X). To see how f (x) winds around z,
we inquire how often the unit vector
f (x) − z
u(x) =
|f (x) − z|
which indicates the direction from z to f (x), points in a given direction. From
intersection theory, we know that u : X → S n−1 hits almost every direction vector
the same number of times mod 2, namely, deg2 (u) times. So seize this invariant
and define the mod 2 winding number of f around z to be W2 (f, z) = deg2 (u).
To establish a generalized version of the Jordan curve theorem, we need a
preliminary theorem, whose proof introduces a beautifully simple technique.
Theorem 3.5.0.1. Suppose that X is the boundary of D, a compact manifold with bound-
ary, and let F : D → Rn extending f ; that is, ∂F = f . Suppose that z is a regu-
lar value of F that does not belong to the image of f . Then F −1 (z) is a finite set, and
W2 (f, z) = #F −1 (z) mod 2. That is, f winds X around z as often as F hits z, mod 2.
Figure 3.2: W2 (f, z) = #F −1 (z) mod 2
The proof is broken down into the following steps:
(1) If F does not hit z, then W2 (f, z) = 0 (If u extends to D, then deg2 (u) = 0).
(2) Suppose that F −1 (z) = {y1 , · · · , yl }, and around each point yi , let Bi be a ball.
Demand that the balls be disjoint from one another and from X = ∂D. Let
fi : ∂Bi → Rn be the restriction of F , and
W2 (f, z) = W2 (f1 , z) + · · · + W2 (f, z) mod 2.

l
(If u extends to D0 = D −
S
Int(Bi ), then deg2 (u) = 0).
i=1
(3) Use the regularity of z to choose the balls Bi so that W2 (fi , z) = 1, and thus
prove the theorem.
Now assume that X is actually is a compact, connected hypersurface in Rn . If

X really does separate Rn into an inside and an outside, then it should be the
boundary of a compact n-dimensional manifold with boundary-namely, its inside.
In this case, the theorem above tells us that if z ∈ Rn is any point not on X, then
W2 (X, z) must be 1 or 0, depending on whether z lies inside or outside of X.
Now we give the main theorem:
Theorem 3.5.0.2. (The Jordan-Brouwer Separation Theorem) The complement of

the compact, connected hypersurface X in Rn consists of two connected open sets, the
3.5. WINDING NUMBERS AND THE JORDAN-BROUWER SEPARATION THEOREM45
"outside" D0 and the "inside" D1 . Moreover, D̄1 is a compact manifold with boundary
∂ D̄1 = X.
The proof is broken down into the following steps:
(1) Let z ∈ Rn −X, if x is any point of X and U any neighborhood of x in Rn , then

there exists a point of U that may be joined to z by a curve not intersecting
X (Show that the points x ∈ X for which the statement is true constitute a
nonempty, open, and a closed set).
(2) Rn − X has, at most, two connected components (use (1)).
(3) If z0 and z1 belong to the same connected component of Rn −X, then W2 (X, z0 ) =
W2 (X, z1 ).
(4) Given a point z ∈ Rn − X and a direction vector ~v ∈ S n−1 , consider the ray r
emanating from z in the direction of ~v ,
r = {z + t~v : t ≥ 0}.
the ray r is transversal to X if and only if ~v is a regular value of the direction

map u : X → S n−1 . In particular, almost every ray from z intersects X
transversally.
To prove this we need

f g
Lemma 3.5.0.1. Let X −→ Y −→ Z be a sequence of smooth maps of manifolds,
and assume that g is transversal to submanifold W of Z. Show f > g −1 (W ) if and
only if g ◦ f > W .
and let g : Rn − {z} → S n−1 is g(y) = y − z/|y − z|.
(5) Suppose that r is a ray emanating from z0 that intersects X transversally in

a nonempty (necessarily finite) set. Suppose that zl is any other point on r
Figure 3.3: W2 (X, z0 ) = W2 (X, z1 ) + 3 mod 2
(but not on X), and let l be the number of times r intersects X between z0
and z1 . Then W2 (X, z0 ) = W2 (X, z0 ) + l mod 2.
(by (4), ~v is a regular value for both u0 and u1 , but
#u−1 v ) = #u−1
0 (~ 1 (~
v ) + l.
(6) Rn − X has precisely two components:
D0 = {z : W2 (X, z) = 0}
and
D1 = {z : W2 (X, z) = 1}
(use (2), (3), (5)).
(7) If z is very large, then W2 (X, z) = 0 (since X is compact, when |z| is large the
image u(x) on S n−1 lies in a small neighborhood of z/|z|).

3.6. THE BORSUK-ULAM THEOREM 47
Also by the proof we have actually derived a simple procedure for determining
whether a given point z lies inside or outside of X:
Proposition 3.5.0.1. Given z ∈ Rn − X, let r be the ray emanating from z that is

transversal to X, then z is inside X if and only if r intersects X in an odd number of
points.
Proof. By (5)and(7).
3.6 The Borsuk-Ulam Theorem

We will prove our winding number apparatus to prove another famous the-
orem from topology, the Borsuk-Ulam theorem.
Theorem 3.6.0.1. (Borsuk-Ulam Theorem) Let f : S k → Rk+1 be a smooth map whose

image does not contain the origin, and suppose that f satisfies the symmetry condition
f (−x) = −f (x) for all x ∈ S k ,
Then W2 (f, 0) = 1.
Informally, any map that is symmetric around the origin must wind around it an odd
number of times.
Proof. Proceed by induction on k. For k = 1, it’s obvious.

Now assume the theorem true for k − 1, and let f : S k → Rk+1 − {0} be
symmetric. Consider S k−1 to be the equator of S k , embedded by (x1 , · · · , xk ) →
(x1 , · · · , xk , 0). We will compute W2 (f, 0) by counting how often f intersects a line
l in Rk+1 . By choosing l disjoint from the image of the equator, we can use the
inductive hypothesis to show that the equator winds around l an odd number of
times. Finally, it’s easy to calculate the intersection of f with l once we know the
behavior of f on the equator.
Denote the restriction of f to the equator S k−1 by g. In choosing a suitable

line l, use Sard to select a unit vector ~a that is a regular value for both maps
g
: S k−1 → S k
|g|
and
f
: Sk → Sk
|f |
From symmetry, it’s clear that −~a is a regular value for both maps. By dimen-
sional comparison, regularity for g/|g| simply means that g/|g| never hits ~a or −~a;
consequently, g never intersects the line l = R · ~a. It’s easy to verify that regularity
for f /|f | is equivalent to the condition f > l.
Now by definition,
f f
W2 (f, 0) = deg2 ( ) = #( )−1 (~a) mod 2.
|f | |f |
and f /|f | hits ~a precisely as often as it hits −~a, due to symmetry. Thus
f −1 1
#( ) (~a) = #f −1 (l).
|f | 2
We can calculate this result on the upper hemisphere alone. Let f+ be the restric-
tion of f to the upper hemisphere S+k , we know that
1
#f+−1 (l) = #f −1 (l).
2
We conclude that W2 (f, 0) = #f+−1 (l) mod 2.
The upper sphere is a manifold with boundary, and on its boundary ∂S+k =
S k−1 we can use the inductive hypothesis. Let V be the orthogonal complement
of l, and let π : Rk+1 → V be orthogonal projection. Since g is symmetric and π is
linear, the composite π ◦ g : S k−1 → V is symmetric; moreover, π ◦ g is never zero.
Thus W2 (π ◦ g, 0) = 1.
Since f+ > l,
π ◦ f+ : S k → V
is transversal to 0, so
W2 (π ◦ g, 0) = #(π ◦ f+ )−1 (0).
But
(π ◦ f+ )−1 (0) = f+−1 (l).
so
W2 (f, 0) = #f+−1 (l) = W2 (π ◦ g, 0) = 1 mod 2.
we can easily have
Theorem 3.6.0.2. If f : S k → Rk+1 −{0} is symmetric about the origin, then f intersects
every line through 0 at least once.
Proof. Suppose f never hits l, the use this l in the proof, obtaining the contradic-
tion:
1
W2 (f, 0) = #f −1 (0) = 0.
2
Also, we can give some surprising consequences:
Theorem 3.6.0.3. Any k smooth functions f1 , · · · , fk on S k that all satisfy the symmetry
condition fi (−x) = −fi (x), i = 1, · · · , k, must possess a common zero.
Proof. If not, apply the corollary to the map
f (x) = (f1 (x), · · · , fk (x), 0)
taking the xk+1 axis for l.
Theorem 3.6.0.4. For any k smooth functions g1 , · · · , gk on S k there exists a point p ∈

S k such that
g1 (p) = g1 (−p), · · · , gk (p) = gk (−p).
Proof. just let

fi (x) = gi (x) − gi (−x).
A meteorological formulation of this result is that at any given time, there

are two places in the world, at opposite ends of the earth from each other, having
exactly the same weather.
Another verbalization of the theorem is: if a balloon is deflated and laid on
the floor, two antipodal points end up over the same point of the floor.
The most famous formulation is the Ham Sandwich Theorem:
Theorem 3.6.0.5. (Ham Sandwich Theorem) If there exist n connected compact set
A1 , · · · , An in Rn , then there exists a hyperplane of dimension n − 1 such that it bisects
each Ai .
Figure 3.4: sandwich
Proof. Take a ball B n with a sufficient radius in Rn such that it contains all Ai , with
o as its centre. x is any point on ∂B n = S n−1 , with antipode point −x. Then there
exists unique hyperplane πAi bisects Ai and perpendicular to the diameter x(−x).
Suppose XAi is point of intersection of πAi and x(−x). If xAi is between x and o,
define gAi (x) = |xAi − o|; if not, gAi (x) = −|xAi − o|.
Then we define
f : S n−1 → Rn−1 , f (x) = (gA1 (x) − gA2 (x), · · · , gA1 (x) − gAn (x)).
By theorem 3.63, there exists a point x such that f (x) = 0, which means that
gA1 (x) = · · · = gAn (x).
Thus the hyperplane perpendicular to x(−x) and whose directed distance

from o is gA1 (x) bisects all Ai . The proof is complete.
Chapter 4
Oriented Intersection Theory
Where hills bend,streams wind and the pathway seems to end,

past dark willows and flowers in bloom lies another vil...
—You Lu
4.1 Orientation
First we consider finite-dimensional vector spaces.
Definition 4.1.0.1. Suppose that V is a finite-dimensional real vector space and β =
{v1 , · · · , xvk } is an ordered basis. If β 0 = {v10 , · · · , vk0 } is another ordered basis, then
there is a unique linear isomorphism A : V → V such that β 0 = Aβ. We shall say that
β and β 0 are equivalently oriented if the determinant of the linear transformation A is
positive.
Definition 4.1.0.2. An orientation of V is an arbitrary decision to affix a positive sign to
the elements of one equivalence class and a negative sign to the others. The sign given an
ordered basis β is called its orientation, so β is either positively oriented or negatively
oriented.
A separate definition is required for the zero-dimensional vector space. Here
an orientation is just a choice of sign +1 or −1.
53
54 CHAPTER 4. ORIENTED INTERSECTION THEORY
If A : V → W is an isomorphism of vector spaces. If both V and W are

oriented, meaning that an orientation is specified for both, the sign of Aβ is either
always the same as the sign of β or always opposite. That is, A either preserves or
reverses the orientation.
Now we return to manifolds.
Definition 4.1.0.3. An orientation of X, a manifold with boundary is a smooth choice

of orientations for all the tangent spaces Tx (X). The smoothness condition is to be in-
terpreted in the following sense: around each point x ∈ X there must exist a local
parametrization h : U → X, such that dhu : Rk → Th(u) (X) preserves orientation at
each point u of the domain U ⊂ Hk . (The orientation on Rk is implicitly assumed to be
the standard one).
A map like h whose derivative preserves orientations at every point is simply called
an orientation-preserving map.
Not all manifolds possess orientations, the most famous example being the
Mobius strip.
Figure 4.1: Mobius strip
For zero-dimensional manifolds, orientations are very simple. To each point

x ∈ X we simply assign an orientation number +1 and −1.
Definition 4.1.0.4. X is orientable if it may be given an orientation.
Proposition 4.1.0.1. A connected, orientable manifold with boundary admits exactly two
orientations.
4.1. ORIENTATION 55
Proof. We just need show the set of points at which two orientations agree and the
set where they disagree are both open.
By an oriented manifold, we mean a manifold together with a specified smooth

orientation. If X is oriented, we shall symbolically denote the orientated manifold
obtained by reversing the orientation on X as −X.
Now we consider the product orientation.
Definition 4.1.0.5. If X and Y are oriented and one of them is boundaryless, then at each
point (x, y) ∈ X × Y ,
T(x,y) (X × Y ) = Tx (X) × Ty (Y ).
Let α = {v1 , · · · , vk } and β = {w1 , · · · , wl } are ordered bases for Tx (X) and Ty (Y ), and
denote by (α × 0, 0 × β) the ordered basis {(v1 , 0), · · · , (vk , 0), (0, w1 ), · · · , (0, wl )} of
Tx (X) × Ty (Y ). Define the orientation on Tx (X) × Ty (Y ) by setting
sign(α × 0, 0 × β) = sign(α)sign(β).
an orientation of X naturally induces a boundary orientation on ∂X. At every

point x ∈ ∂X, Tx (∂X) has codimension 1 in Tx (X). Therefore there are precisely
two unit vectors in Tx (X). We denote the outward unit normal vector at x by nx .
Now orient Tx (∂X) by declaring the sign of any ordered basis β = {v1 , · · · , vk−1 }
to be the sign of the ordered basis {nx , β} for Tx (X).
A particularly important example is ∂(I × X) = X1 − X0 .
If dimX = 1, then ∂X is zero dimensional. The orientation of the zero-
dimensional vector space Tx (∂X) is equal to the sign of the basis {nx } for Tx (X).
Now let X be any compact oriented one-manifold with boundary, we have
Proposition 4.1.0.2. The sum of the orientation numbers at the boundary points of any
compact oriented one-dimensional manifold with boundary is zero.
Now we put orientations on transversal preimages. To do so, we shall use

L
the following simple observation. Suppose that V = V1 V2 is a direct sum.
Then orientations on any two of these vector spaces automatically induces a direct
sum orientation on the third. Choose ordered bases β1 and β2 for V1 and V2 , let
β = (β1 , β2 ) be the combined ordered basis for V . Simply demand that sign(β) =
sign(β1 ) · sign(β2 ).
Remark 9. The order of the summands V1 and V2 is crucial, for (β1 , β2 ) may not be
equivalent in orientation to (β2 , β1 ).
Now let f : X → Y be a smooth map with f > Z and ∂f > Z, where X, Y, Z

are all oriented and the last two are boundaryless. We define a preimage orientation
on the manifold with boundary S = f −1 (Z). If f (x) = z ∈ Z, then Tx (S) is the
preimage of Tz (Z) under the derivative map dfx : Tx (X) → Tx (S). Let Nz (S; X) be
the orthogonal complement to Tx (S) in Tx (X). Then
M
Nx (S; X) Tx (S) = Tx (X).
so we need only choose an orientation on Nx (S; X) to obtain a direct sum orienta-

tion on Tx (S).
Because
dfx Tx (X) + Tz (Z) = Tz (Y ).
and Tx (S) is the entire preiamge of Tz (Z), we get a direct sum

M
dfx Nx (S; X) Tz (Z) = Tz (Y ).
Thus the orientations on Z and Y induce a direct image orientation on dfx Nx (S; X).
But Tx (S) contains the entire kernel of the linear map dfx , so dfx must map Nx (S; X)
isomorphically onto its image. Therefore the induced orientation on dfx Nx (S; X)
defines an orientation on Nx (S; X) via the isomorphism dfx .
Conclusion. There are two direct sum equations that define the preimage
orientation: M
dfx Nx (S; X) Tz (Z) = Tz (Y )
M
Nx (S; X) Tx (S) = Tx (X).
4.1. ORIENTATION 57
Remark 10. Our use of Nx (S; X), the orthogonal complement of Tx (S) in Tx (X), was
only for convenient definiteness. In fact, if H is any other subspace of Tx (X) complemen-
tary to Tx (S), then the two direct sums
M
dfx H Tz (Z) = Tz (Y )
M
H Tx (S) = Tx (X).
define the same orientation on Tx (S).
We have notice that: suppose that we have a map f : X → Y as above, where

f > Z, ∂f > Z, the manifolds are oriented, and only X has boundary. Then the
manifold ∂f −1 (Z) acquires two orientations-one as the preimage of Z under the
map ∂f : ∂X → Y , and a second as the boundary of the manifold f −1 (Z). It turns
out that the these orientations differ by the sign (−1)codimZ .
Proposition 4.1.0.3.
∂[f −1 (Z)] = (−1)codimZ (∂f )−1 (Z).
Proof. Denote f −1 (Z) by S. Let H be a subspace of Tx (∂X) complementary to

Tx (∂S), so
M
H Tx (∂S) = Tx (∂X).
Note that H is also complementary to Tx (S) in Tx (X); simply compare dimen-

sions, noting that H must be disjoint from Tx (S) because
Tx (S) ∩ Tx (∂X) = Tx (∂S).
so we may use H to define the orientation of both S and ∂S at x. Since H ⊂

Tx (∂X), the map dfx and d(∂f )x agree on H. Thus H is assigned the same orienta-
L
tion under the two maps, via the direct sum dfx H Tz (Z) = Tz (Y ). Now that H
L
is oriented, the orientation of S induced by f is defined by H Tx (S) = Tx (X),
L
and the orientation of ∂S induced by ∂f is defined by H Tx (∂S) = Tx (∂X).
Let nx be the outward unit vector to ∂S in S, and let R · nx represent the one-
dimensional subspace spanned by nx , oriented so that {nx } is a positively oriented
basis. nx need not be perpendicular to all of Tx (∂X). Nonetheless, we claim that

the orientations of Tx (∂X) and Tx (X) are related by the direct sum
M
R · nx Tx (∂X) = Tx (X).
into the equation

M
Tx (X) = R · nx Tx (∂X).
insert the formulas for the preimage orientations of S and ∂S, obtaining
M M M
H Tx (S) = R · nx H Tx (∂S).
thus
l = dimH = codimS = codimZ.
we are done.
4.2 Oriented Intersection Number

Now we are prepared to reconstruct intersection theory. X, Y and Z are
boundaryless manifolds, X is compact, Z is a closed submanifold of Y , and dimX+
dimY = dimZ. However, we shall work exclusively with oriented manifolds.
Definition 4.2.0.1. If f : X → Y is transversal to Z, then f −1 (Z) is a finite number of

points, each with an orientation number ±1 provided by the preimage orientation. Define
the intersection number I(f, Z) to be the sum of these orientation numbers.
For if f (x) = z ∈ Z, then transversality plus dimensional complementarity

give a direct sum
M
dfx Tx (X) Tz (Z) = Tz (Y ).
now dfx must be an isomorphism onto its image. So the orientation number at x is
+1 if the orientations on dfx Tx (X) and Tz (Z) "add up" to the prescribed orientation
on Y , and −1 if not.
4.2. ORIENTED INTERSECTION NUMBER 59
Proposition 4.2.0.1. If X = ∂W and f : X → Y extends to W , then I(f, Z) = 0 (W

compact).
Proposition 4.2.0.2. Homotopic maps always have the same intersection numbers.
Given any g : X → Y , select a homotopic map f that is transversal to Z, and

define I(g, Z) = I(f, Z).
When Y is connected and has the same dimension as X, we define the degree
of an arbitrary smooth map f : X → Y to be the intersection number of f with
any point y, deg(f ) = I(f, {y}). Our proof that I2 (f, {y}) is the same for all points
y ∈ Y works perfectly well for the oriented theory, so deg(f ) is accepted defined.
Since degree is defined as an intersection number, it must automatically be a
homotopy invariant.
A class of interesting maps of the circle involves the restrictions of the com-
plex monomials z → z m . Its degree is m.
Proposition 4.2.0.3. Suppose that f : X → Y is a smooth map of compact oriented

manifolds having the same dimension and that X = ∂W (W compact). If f can be
extended to all of W , then deg(f ) = 0.
We now complete an item left unfinished by the mod 2 theory:
Theorem 4.2.0.1. (The Fundamental Theorem of Algebra) Every nonconstant com-

plex polynomial has a root.
Proof. For any complex polynomial p(z) of order m, we showed earlier that on
a circle S of sufficiently large radius r in the plane,p(z)/|p(z)| and z m /|z m | are
homotopic. Thus p/|p| has degree m. When m > 0 we have p/|p| cannot extend to
the whole disk, thus p must have a zero inside the disk.
Proposition 4.2.0.4. Let W be a smooth compact region in C whose boundary contains

no zeros of the polynomial p. Then the total number of zeros of p inside W , counting
multiplicities, is the degree of the map p/|p| : ∂W → S 1 .
Definition 4.2.0.2. When X also happens to be a submanifold of Y , then, as in the mod

2 case, we define its intersection number with Z, I(X, Z), to be the intersection number
of the inclusion map of X with Z.
If X > Z, then I(X, Z) is calculated by counting the points of X ∩ Z.
Remark 11. When X and Z are both compact, then they possess two intersection num-
bers, I(X, Z) and I(Z, X), and these numbers may be different.
Now we define an intersection number for a pair of arbitrary maps f : X →

Y, g : Z → Y . X and Z are both compact. First, consider the transversal case. Two
maps f and g are defined to be transversal, f > g, if
dfx Tx (X) + dgz Tz (Z) = Ty (Y ).
whenever
f (x) = y = g(z).
The dimensional complementarity further implies that the sum is direct and that
both derivatives dfx and dgz are injective. Define the local intersection number
L
at (x, z) to be +1 if the direct sum orientation of dfx Tx (X) dgz Tz (Z) equals the
given orientation on Ty (Y ), and −1 otherwise. Then I(f, g) is defined as the sum
of the local contributions from all pairs (x, z) at which f (x) = g(z). To show
that the sum is finite, we use a standard reformulation once more. If 4 denote
the diagonal of Y × Y , and f × g : X × Z → Y × Y is the product map, then
f (x) = g(z) precisely at pairs (x, y) in (f × g)−1 (4). Now dim(X × Y ) = codim4,
so if f × g > 4, then the preimage of 4 is a compact zero-dimensional manifold,
hence a finite set.
L
Lemma 4.2.0.1. Let U and W be subspace of the vector space V . Then U W = V if
L
and only if U × W 4 = V × V . Assume, also, that U and W are oriented, and give V
the direct sum orientation. Now assign 4 the orientation carried from V by the natural
isomorphism V → 4. Then the product orientation on V × V agrees with the direct sum
L
orientation from U × W 4 if and only if W is even dimensional.
4.2. ORIENTED INTERSECTION NUMBER 61
Substituting
U = dfx Tx (X), W = dgz Tz (Z), V = Ty (Y ).
the lemma translates as
Proposition 4.2.0.5. f > g if and only if f × g > 4, and then
I(f, g) = (−1)dimZ I(f × g, 4).
thus we can remove the transversality assumption on f and g. For arbitary

maps f : X → Y, g : Z → Y , we may simply define I(f, g) to be (−1)dimZ I(f ×
g, 4).
Proposition 4.2.0.6. If f0 and g0 are respectively homotopic to f1 and g1 , then I(f0 , g0 ) =

I(f1 , g1 ).
Corollary 4.2.0.1. If Z is a submanifold of Y and i : Z → Y is its inclusion map, then

I(f, i) = I(f, Z) for any map f : X → Y .
Corollary 4.2.0.2. If dimX = dimY and Y is connected, then I(f, {y}) is the same for
every y ∈ Y . Thus deg(f ) is well defined.
Proof. Since Y is connected, the inclusion maps i0 , i1 of any two points y0 , y1 ∈ Y

are homotopic. Therefore,
I(f, {y0 }) = I(f, i0 ) = I(f, i1 ) = I(f, {y1 }).
Proposition 4.2.0.7. I(f, g) = (−1)(dimX)(dimZ) I(g, f ).
Corollary 4.2.0.3. If X and Z are both compact submanifolds of Y , then
I(X, Z) = (−1)(dimX)(dimZ) I(Z, X).

In particular, suppose that dimY = 2dimX, then the self-intersection number

I(X, X) is defined. If X is odd dimensional, then I(X, X) = 0. Consequently,
I2 (X, X) = I(X, X) mod 2 vanishes as well.
Thus we may calculate I2 (X, X) for compact orientable submanifolds X of
half dimension in an arbitrary manifold Y . If one of these self-intersection num-
bers fails to vanish, then Y cannot be oriented. For example, consider the central
circle in the Mobius strip.
Definition 4.2.0.3. Suppose Y is compact, oriented manifold, its Euler characteristic

χ(Y ) is defined to be the self-intersection number of the diagonal 4 in Y × Y :
χ(Y ) = I(4, 4).
The Euler characteristic is a diffeomorphism invariant of compact manifolds

that plays a fundamental role in a variety of geometric and topological situations.
Remark 12. The Euler characteristic is well defined for nonorientable manifolds and that
it’s still a diffeomorphism invariant.
Proposition 4.2.0.8. The Euler characteristic of an odd-dimensional, compact, oriented

manifold is zero.
4.3 Lefschetz Fixed-point Theory

Now we use intersection theory to study the fixed points of a smooth map
f : X → X on a compact oriented manifold.
Note that x is a fixed point precisely when (x, f (x)) ∈ X × X belongs to the
intersection of graph(f ) with the diagonal 4. As the latter are submanifolds of
complementary dimension in X × X.
Definition 4.3.0.1. I(4, graph(f )) is called the global Lefschetz number of f , denoted
L(f ).
4.3. LEFSCHETZ FIXED-POINT THEORY 63
Of course, f may actually have an infinite number of fixed points.

We observe some immediate consequences of the intersection theory approach:
Theorem 4.3.0.1. (Smooth Lefschetz Fixed-point Theorem) Let f : X → X be a

smooth map on a compact orientable manifold. If L(f ) 6= 0, then f has a fixed point.
Proof. If f has no fixed points, then 4 and graph(f ) are disjoint, hence trivially
transversal. Thus
L(f ) = I(4, graph(f )) = 0.
Contradiction!
Proposition 4.3.0.1. L(f ) is a homotopy invariant.
We know the graph of the identity map is just the diagonal itself; thus L(identity) =
I(4, 4) = χ(X). So
Proposition 4.3.0.2. If f is homotopic to the identity, then L(f ) equals the Euler char-
acteristic of X. In particular, if X admits a smooth map f : X → X that is homotopic to
the identity and has no fixed points, then χ(X) = 0.
Definition 4.3.0.2. We say that the maps f : X → X is a Lefschetz map if graph(f ) > 4.
Since Lefschetz maps are defined by a transversality condition, no one should

be surprised to discover that "most" maps are Lefschetz:
Proposition 4.3.0.3. Every map f : X → X is homotopic to a Lefschetz map.
Proof. We can find an open ball S of some Euclidean space and a smooth map
F : X × S → X, such that F (x, 0) = f (x) and s → F (x, s) defines a submersion
S → X for each x ∈ X. Note that the map G : X ×S → X ×X, defined by G(x, s) =
(x, F (x, s) is also a submersion, hence G > 4. By the Transversality Theorem for
almost s the map X → X × X is > 4. That is x → F (x, s) is Lefschetz.
What does it mean for f to be Lefschetz? Suppose that x is a fixed point

of f . The tangent space of graph(f ) in Tx (X) × Tx (X) is the graph of the map
dfx : Tx (X) → Tx (X), and the tangent space of the diagonal 4 is the diagonal 4x
of Tx (X) × Tx (X). Thus graph(f ) > 4 at (x, x) if and only if
graph(dfx ) + 4x = Tx (X) × Tx (X).
As graph(dfx ) and 4x are vector subspaces of Tx (X) × Tx (X) with complemen-

tary dimension, they fill out everything precisely if their intersection is zero. But
graph(dfx ) ∩ 4x = 0 means just that dfx has no nonzero fixed point or dfx has no
eigenvector of eigenvalue +1.
Definition 4.3.0.3. A fixed point x is a Lefschetz fixed point of f if dfx has no nonzero
fixed point.
Proposition 4.3.0.4. f is a Lefschetz map if and only if all its fixed points are Lefschetz.
Definition 4.3.0.4. If x is a Lefschetz fixed point, we denote the orientation number ±1

of (x, x) in the intersection 4 ∩ graph(f ) by Lx (f ), called the local Lefschetz number
of f at x.
Thus for Lefschetz maps.

X
L(f ) = Lx (f ).
f (x)=x
It’s easy to identity Lx (f ) more explicitly. First, recognize that the Lefschetz con-
dition at x is equivalent to the requirement that dfx − I be an isomorphism of
Tx (X), for the kernel of dfx − I is the fixed-point set of dfx . Now you should not
be shocked to discover that Lx (f ) simply reflects whether dfx − I preserves or
reverses orientation.
Proposition 4.3.0.5. The local Lefschetz number Lx (f ) at a Lefschetz fixed point is +1

if the isomorphism dfx − I preserves orientation on Tx (X), and it is −1 if it reserves
orientation. That is, the sign of Lx (f ) equals the sign of the determinant of dfx − I.
Now we consider the two-dimensional case. Assume that f : R2 → R2 and f

fixes the origin. Set A = df0 , so that
f (x) = Ax + (x).
where → 0 rapidly as x → 0. Assume that A has two independent real eigenvec-

tors, thus by the proper choice of coordinates, its matrix is diagonal,
!
α1 0
A=
0 α2
Then
L0 (f ) = sign(α1 − 1)(α2 − 1).
Assume that both α1 , α2 are positive.
Case 1. Both α1 , α2 > 1. L0 (f ) = +1, and locally f is an "expanding map"
with source at the origin.
Case 2. Both α1 , α2 < 1, again L0 (f ) = +1, and locally f is a "contracting
map" with sink at the origin.
Case 3. α1 < 1 < α2 . Here L0 (f ) = −1, and the origin is a saddlepoint of f .
Thus L0 (f ) reports on the qualitative topological behavior behavior of f near
the fixed point x.
Proposition 4.3.0.6. The Euler characteristic characteristic of S 2 is 2.
Proof. Suppose that π : R3 − {0} → S 2 is the projection x → x/|x|, then let

1
f (x) = π(x + (0, 0, − ).
2
thus LN (f ) = +1 = Ls (f ), so
L(f ) = 2.
As f is homotopic to the identity via
t
ft (x) = π[x + (0, 0, − )].
2
we have L(f ) = χ(S 2 ).
Corollary 4.3.0.1. Every map of S 2 that is homotopic to the identity must possess a fixed
point. In particular, the antipodal map x → −x is not homotopic to the identity.
Theorem 4.3.0.2. (Classification of Two-Manifolds) Every compact oriented bound-

aryless two-manifold is diffeomorphic to one of the following:
• Surface of genus 0(sphere);
• Surface of genus 1(torus);
• Surface of genus n(n ≥ 2);
As we do above we have
Proposition 4.3.0.7. The surface of genus k admits a Lefschetz map homotopic to the
identity, with one source, one sink, and 2k saddles. Consequently, its Euler characteristic
is 2 − 2k.
Now we will prove an interesting fact: under an arbitrarily small perturba-

tion, the fixed points of any map f : X → X split into Lefschetz fixed points. If f
has isolated fixed points, we need only disrupt it locally.
Proposition 4.3.0.8. (Splitting Proposition) Let U be a neighborhood of the fixed point

x that contains no other fixed points of f . Then there exists a homotopy ft of f such that
f1 has only Lefschetz fixed points in U , and each ft equals f outside some compact subset
of U .
Definition 4.3.0.5. Suppose that x is an isolated fixed point of f in Rk . If B is a small

closed ball centered at x that contains no other fixed point, then the assignment
f (z) − z
z→ .
|f (z) − z|
defines a smooth map F : ∂B → S k−1 . We call the degree of this map the local Lefschetz
number of f at x, denoted Lx (f ).
Note that the choice of B is insignificant.
Proposition 4.3.0.9. At Lefschetz fixed points, the two definitions of Lx (f ) agree.
to prove this, we need a lemma:
Lemma 4.3.0.1. Suppose that E is a linear isomorphism of Rk that preserves orientation.

Then there exists a homotopy Et consisting of linear isomorphisms, such that E0 = E and
E1 is the identity. If E reverses orientation, then there exists such a homotopy with E1
equal to the reflection map
E1 (x1 , · · · , xk ) = (−x1 , x2 , · · · , xk ).
Not only does our new definition agree with the old for Lefschetz fixed points,
it’s also easily seen to be the correct "charge" invariant.
Proposition 4.3.0.10. Suppose that the map f in Rk has an isolated fixed point at x, and
let B be a closed ball around x containing no other fixed point of f . Choose any map f1
that equals f outside some compact subset of Int(B) but has only Lefschetz fixed points
in B. Then
X
Lx (f ) = Lz (f1 ).
f1 (z)=z
Now we extend this to manifolds. If f : X → X has an isolated fixed point

at x, choose any local diffeomorphism φ around x, and let g = φ−1 ◦ f ◦ φ on
Euclidean space. Suppose that φ(0) = x, and define Lx (f ) = L0 (g). Does this
definition depend on the choice of φ?
First we check Lefschetz fixed points. If x is Lefschetz, then Lx (f ) is positive
or negative, depending on whether dfx − I preserves or reverses orientation. But
dg0 − I = dφ−1
0 ◦ (dfx − I) ◦ dφ0 .
So dg0 − I is an isomorphism if and only if dfx − I is.

If x is an arbitrary fixed point, use the proposition above we can get the con-
clusion.
Theorem 4.3.0.3. (Local Computation of the Lefschetz Number) Let f : X → X be

any smooth map on an compact manifold, with only finitely many fixed points. Then the
global Lefschetz number equals the sum of the local Lefschetz numbers:
X
L(f ) = Lx (f ).
f (x)=x
4.4 Vector Fields and the Poincare-Hopf Theorem

Definition 4.4.0.1. A vector field on a manifold X in RN is a smooth assignment of
a vector tangent to X at each point x- that is, a smooth map ~v : X → RN such that
~v ∈ Tx (X) for every x.
the following definition is equivalent:
Definition 4.4.0.2. A vector field ~v on X is a cross section of T (X)- that is, a smooth
map ~v : X → T (X) such that p ◦ ~v equals the identity map of X, where p is the projection
map such that p(x, v) = x.
all the interesting behavior of ~v occurs around its zeros, the points x ∈ X
where ~v (x) = 0. The field may circulate around x; it may have a source, sink,
or saddle; it may spiral in toward x or away; or it may form a more complicated
pattern.
In order to investigate the relation between ~v and the topology of X, we must
quantify the directional change of ~v around its zeros. First, assume that we are in
Rk and that ~v has an isolated zero at the origin. Thus the direction variation of ~v
around 0 is measured by the map x → ~v /|~v | carrying any small sphere S around
0 into S k−1 . Choose the radius so small that ~v z has no zeros inside S except the
origin. We define the index of ~v at 0:
Definition 4.4.0.3. the index of ~v at 0, ind0 (~v ), is the degree of this direction map S →
S k−1 .
4.4. VECTOR FIELDS AND THE POINCARE-HOPF THEOREM 69
Figure 4.2: some zeros
For example, the indices of the five vector fields in the figure is a : +1, b :
+1, c : +1, d : −1, e : +1, f : +2).
To define the index on manifold is easy, we just use local parametrization.
Suppose that φ : U → X is a local parametrization carrying the origin of Rk to x.
There is a natural way to pull back the vector field ~v from X to make a vector field
on the open subset U ⊂ Rk . For each u ∈ U , the derivative dφu is an isomorphism
of Rk with the tangent space of X at φ(u). We simply define the pullback vector
field, denoted φ∗~v , by assigning to u the vector that corresponds to the value of ~v
at φ(u):
φ∗~v (u) = dφ−1
u ~v (φ(u)).
Now if ~v has an isolated zero at x, φ∗~v has an isolated zero at the origin, and we
define indx (~v ) = ind0 (φ∗~v ).
The nature of the topological limitation on ~v is explained by the following
celebrated theorem.
Theorem 4.4.0.1. (Poincare-Hopf Index Theorem) If ~v is a smooth vector field on the

compact, oriented manifold X with only finitely many zeros, then the global sum of the
indices of ~v equals the Euler characteristic of X.
to prove the theorem without differential equations, we need a definition:
Definition 4.4.0.4. Suppose {ft } is any homotopic family of transformations of X with

f0 = identity. We shall say that {ft } is tangent to the vector field ~v at time zero if,
for each fixed x ∈ X, the vector ~v (x) is tangent to the curve ft (x) at time zero.
Proposition 4.4.0.1. Suppose, for t 6= 0, that the maps ft have no fixed point in the set
U except the origin and that the vector field ~v vanishes only at 0. If {ft } is tangent to ~v at
time zero, then the local Lefschetz number of each ft at 0 equals the index of ~v :
ind0 (~v ) = L0 (ft ).
Proof of Poincare-Hopf. Given a vector field ~v defined on the compact man-

ifold X, we shall produce a globally defined family maps ft that is tangent to ~v
at time zero and that possesses two important properties: the fixed points of each
ft for t > 0 are precisely the zeros of ~v , and f0 is the identity map of X. From
the proposition, it then follows that the sum of the indices of ~v equals the sum of
the local Lefschetz numbers of ft - that is, the global Lefschetz number. But since
f0 is the identity, L(ft ) = L(f0 ) is the Euler characteristic of X. Thus the proof is
complete.
Corollary 4.4.0.1. (Hairy ball Theorem) There is no smooth nonzero vector field on S 2 .
4.5 The Hopf Degree Theorem

We have know that the degree of a smooth is a homotopy invariance. In this
section you will know that two maps f0 , f1 : X → S k are homotopic if and only if
they have the same degree.
And now we need a new topological term:
4.5. THE HOPF DEGREE THEOREM 71
Figure 4.3: hairy ball
Definition 4.5.0.1. An isotopy is a homotopy in which each map ht is a diffeomorphism,

and two diffeomorphisms are isotopy if they can be joined by an isotopy.
Definition 4.5.0.2. An isotopy is compactly supported if the maps ht are equal to the
identity map outside some fixed compact set.
Lemma 4.5.0.1. (Isotopy Lemma) Given any two points y and z in the connected man-
ifold Y , there exists a diffeomorphism h : Y → Y such that h(y) = z and h is isotopic to
the identity. Moreover, the isotopy may be taken to be compactly supported.
Corollary 4.5.0.1. Suppose that Y is a connected manifold of dimension greater than 1,

and let y1 , · · · , yn and z1 , · · · , zn be two sets of distinct points in Y . Then there exists a
diffeomorphism h : Y → Y , isotopic to the identity, with
h(y1 ) = z1 , · · · , h(yn ) = zn .
Moreover, the isotopy may be taken to be compactly supported.

Proof. Argue inductively, for the theorem gives n = 1. Assuming the corollary
for n − 1, obtain a compactly supported isotopy h0t of the punctured manifold
Y − {yn , zn } such that h01 (yi ) = zi for i < n and h00 = identity. Here we use the hy-
pothesis dimY > 1, which guarantees that the punctured manifold is connected.
The compact-support provision implies that the h0t are all equal to the identity
near yn and zn so they extend to diffeomorphisms of Y that fix those two points.
Similarly, applying the theorem to the punctured manifold
Y − {y1 , · · · , yn−1 , z1 , · · · , zn−1 }.
we obtain a compactly supported isotopy h00t on Y such that h001 (yn ) = zn , h000 =
identity, and all h00t fix the points yi and zi , i < n. So ht = h00t ◦ h0t is the required
isotopy.
Now we begin the proof of Hopf’s theorem:
1 Let f : U → Rk be any smooth map defined on an open subset U of Rk , and

let x be a regular point, with f (x) = z. Let B be a sufficiently small closed
ball centered at x, and define ∂f : ∂B → Rk to be the restriction of f to
the boundary of B. Then W (∂f, z) = +1 if f preserves orientation at x and
W (∂f, z) = −1 if f reserves orientation at x.
2 Let f : B → Rk be a smooth map defined on some closed ball B in Rk .

Suppose that z is a regular value of f that has no preimages on the boundary
sphere ∂B, and consider ∂f : ∂B → Rk . Then the number of preimages of z,
counted with our usual orientation convention equals the winding number
W (∂f, z).
3 Let B be a closed ball in Rk , and let f : Rk − Int(B) → Y be any smooth

map defined outside the open ball Int(B). If the restriction ∂f : ∂B → Y is
homotopic to a constant, then f extends to a smooth map defined on all of
Rk into Y .
4.6. THE EULER CHARACTERISTIC AND TRIANGULATIONS 73
4 (Special Case) Any smooth map f : S l → S l having degree zero is homo-

topic to a constant map.
And we have the corollary:
Corollary 4.5.0.2. Any smooth map f : S l → Rl+1 − {0} having winding number
zero with respect to the origin is homotopic to a constant.
5 Hopf’s theorem in its full generality is essentially a particular instance of the

following theorem:
Theorem 4.5.0.1. (Extension Theorem) Let W be a compact, connected, oriented

k + 1 dimensional manifold with boundary, and let f : ∂W → S k be a smooth map.
Then f extends to a globally defined map F : W → S k , with ∂F = f , if and only
the degree of f is zero.
6 Now we conclude the theorem:
Theorem 4.5.0.2. (The Hopf Degree Theorem) Two maps of a compact, con-
nected, oriented k-manifold X into S k are homotopic if and only if they have the
same degree.
in fact, set W = X × I, then use the Extension Theorem we can prove the
theorem.
As an application:
Theorem 4.5.0.3. A compact, connected, oriented manifold X possesses a nowhere-vanishing

vector field if and only if its Euler characteristic is zero.
4.6 The Euler Characteristic and Triangulations

In differential geometry, we have known that the Euler characteristic equals
to F − E + V for a triangulation for 2 dimensional. In fact, there exists a vector
field on X that has a source in each polygonal face, a saddle on each edge, a sink
at each vertex, and no other zeros. Thus we obtain the theorem F − E + V = χ(X).
The same proof outline is valid in higher dimensions. If X is k dimensional,
then one use k-dimensional generalizations of polygons and obtains the Euler
characteristic of X as the alternating sum
k
X
(−1)j · (number of faces of dimension j).
j=0
This is amazing!

Introduction to Differential Topology Fundamentals

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction to Differential Topology Fundamentals

Uploaded by

Copyright:

Available Formats

Introduction to Differential topology

2 Local Classification of Smooth Mappings 11

3 Transversality and Intersection 29

3.1 Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . . . . 29

3.2 One-Manifolds and Some Consequences . . . . . . . . . . . . . . . . 32

3.4 Intersection Theory Mod 2 . . . . . . . . . . . . . . . . . . . . . . . . 39

3.5 Winding numbers and the Jordan-Brouwer Separation Theorem . . 43

3.6 The Borsuk-Ulam Theorem . . . . . . . . . . . . . . . . . . . . . . . 47

4 Oriented Intersection Theory 53

4.2 Oriented Intersection Number . . . . . . . . . . . . . . . . . . . . . . 58

4.3 Lefschetz Fixed-point Theory . . . . . . . . . . . . . . . . . . . . . . 62

4.4 Vector Fields and the Poincare-Hopf Theorem . . . . . . . . . . . . 68

4.5 The Hopf Degree Theorem . . . . . . . . . . . . . . . . . . . . . . . . 70

4.6 The Euler Characteristic and Triangulations . . . . . . . . . . . . . . 73

1.2 Differential Manifolds

Definition 1.2.0.1. Suppose M is a topological space. We say that M is a topological

In Differential topology, our subject is mainly differnetial(smooth) manifolds, which

Definition 1.2.0.2. Suppose M is a topological manifold in RN , we say that M is a k-

Remark 1. • A smooth map f : X → Y of subsets of two Euclidean spaces is a diffeo-

Definition 1.2.0.3. If X and Z are both manifolds in RN and Z ⊂ X, then Z is a

Suppose that X and Y are manifolds inside RN and RM respectively, so that

φ × ψ(w, u) = (φ(w), ψ(u)).

of course φ × ψ is a local parametrization of (x, y) in X × Y . Thus we have proved:

Theorem 1.2.0.1. If X and Y are manifolds, so is X × Y , and dimX × Y = dimX +

1.3 Derivatives and Tangents

d(g ◦ f )x = dgf (x) ◦ dfx .

The derivative of a mapping is its best linear approximation. Suppose that X

u → φ(0) + dφ0 (u) = x + dφ0 (u).

Definition 1.3.0.1. A differential curve passing through x is a differential mapping

For tangent vector [γ] ∈ Tx (X), define

hf, [γ]i := hf, γix .

which means the direction derivative of f through [γ].

suppose that φ : U → X parametrizes X about x and ψ : V → Y parametrizes

d(g ◦ f )x = dgf (x) ◦ dfx .

Another method to define the derivative of smooth map is simple:

1.4 Second-countable, Paracompactness

Theorem 1.4.1.1. (Lindelöf ) Any open cover of an C 2 topological space X has a

Definition 1.4.2.2. Suppose X is Hausdorff, we say X is paracompact if for any open

1.5 Appendix A: Tangent Spaces, Tangent Vectors, Func-

1.6 Appendix B: More Topological Properties

Local Classification of Smooth

2.1 The Inverse Function Theorem

Theorem 2.1.0.1. (The Inverse Function Theorem) Suppose that f : X → Y is a

Theorem 2.1.0.2. (The Inverse Function Theorem) If dfx is an isomorphism, then

In general, we shall say that two maps f : X → Y and f 0 : X 0 → Y 0 are

2.2 Immersions and Embeddings

Definition 2.2.0.1. If dfx : Tx (X) → Ty (Y ) is injective, then f is called an immersion

The canonical immersion is the standard inclusion map of Rk into Rl for l ≥ k,

Theorem 2.2.0.1. (Local Immersion Theorem) Suppose that f : X → Y is an immer-

f (x1 , · · · , xk ) = (x1 , · · · , xk , 0, · · · , 0).

In other words, f is locally equivalent to the canonical immersion near x.

The proof is left as an exercise for the readers.

Definition 2.2.0.2. An immersion that is injective and proper is called an embedding.

Having subjoined appropriate global topologucal constraints to the local im-

Theorem 2.2.0.2. An embedding f : X → Y maps X diffeomorphically onto submani-

In particular, if X itself is compact, every map f : X → Y is proper. Thus for

2.3 Submersions and Regularity

Definition 2.3.0.1. If dfx : Tx (X) → Ty (Y ) is surjective, then f is called a submersion

The canonical submersion is the standard projection of Rk onto Rl for k ≥ l, in

Theorem 2.3.0.1. (Local Submersion Theorem) Suppose that f : X → Y is a sub-

The proof is left as an exercise for the readers.

Definition 2.3.0.2. For a smooth map of manifolds f : X → Y , a point y ∈ Y is called