Manifolds. Lecture Notes - I. Marcut.

Download as pdf or txt
Download as pdf or txt
You are on page 1of 149

Manifolds.

Lecture Notes - Fall 2017


Ioan Mărcut,

1
Contents
Manifolds. Lecture Notes - Fall 2017 1

Some words about this course 6

Lecture 1. 9
1.1. Brief review of notions from Topology and Analysis 9
1.2. Manifolds 12
1.3. Exercises 17

Lecture 2. 21
2.1. Smooth maps 21
2.2. How many smooth structures? 22
2.3. C k -manifolds 23
2.4. Embedded manifolds in Rn 24
2.5. Exercises 25

Lecture 3. 27
3.1. The topology is determined by the atlas 27
3.2. The real projective space 28
3.3. Smooth manifolds as quotients of group actions 30
3.4. Manifolds of dimensions 1 and 2 32
3.5. Exercises 33

Lecture 4. 37
4.1. Bump functions 37
4.2. Partitions of unity 37
4.3. Proof of Theorem 4.2.4 39
4.4. Corollaries of Theorem 4.2.4 41
4.5. Exercises 42

Lecture 5. 43
5.1. The tangent space 43
5.2. Notations for tangent vectors 45
5.3. The commutative algebra C ∞ (M ) 46
5.4. Tangent vectors as derivations 46
5.5. Exercises 49

Lecture 6. 53
6.1. The inverse function theorem 53
6.2. Immersions and submersions 53
3
4 IOAN MĂRCUT
, , MANIFOLDS

6.3. Embedded submanifolds 55


6.4. Preimages of submanifolds 57
6.5. Exercises 59

Lecture 7. 61
7.1. Immersed submanifolds 61
7.2. A weak version of Whitney’s Embedding Theorem 65
7.3. Exercises 66

Lecture 8. 69
8.1. The tangent bundle 69
8.2. Vector bundles 70
8.3. Sections of vector bundles 71
8.4. Frames 72
8.5. Vector fields as sections of the tangent bundle 73
8.6. Exercises 75

Lecture 9. 77
9.1. Vector fields as derivations 77
9.2. The Lie bracket 78
9.3. Related vector fields 79
9.4. Change of coordinates rule for vector fields 81
9.5. Integral curves of a vector field 81
9.6. Exercises 83

Lecture 10. 85
10.1. The flow of a vector field 85
10.2. Complete vector fields 87
10.3. The flow as the exponential of the Lie derivative 88
10.4. The Lie bracket as infinitesimal commutator of flows 89
10.5. Exercises 90

Lecture 11. 93
11.1. Lie groups 93
11.2. Exercises 97

Lecture 12. 101


12.1. The exterior algebra 101
12.2. The cotangent bundle 107
12.3. The exterior powers of the cotangent bundle 107
12.4. Differential forms 108
12.5. The pullback of differential forms 109
12.6. Exercises 110

Lecture 13. 113


13.1. The exterior derivative 113
13.2. De Rham cohomology 116
13.3. Cartan calculus 121
13.4. The algebra of graded derivations 122
13.5. Exercises 123
IOAN MĂRCUT
, , MANIFOLDS 5

Lecture 14. 127


14.1. Volume elements 127
14.2. The density bundle 128
14.3. Densities 129
14.4. Integration of densities 130
14.5. Orientations 132
14.6. Exercises 133

Lecture 15. 137


15.1. Manifolds with boundary 137
15.2. Integration of differential forms on oriented submanifolds 141
15.3. Stokes’ Theorem 143
15.4. Exercises 145

Bibliography 147
Exercises and assignments 149
Problems for exercise classes 149
Homework assignments 149
6 IOAN MĂRCUT
, , MANIFOLDS

Some words about this course


The goal of this course is to introduce and to study basic properties of smooth
manifolds. The easiest way to picture smooth manifolds is as generalizations of
smooth curves (1-dimensional) and surfaces (2-dimensional) in R3 to higher dimen-
sions (and in Rn , instead of R3 ). However, the fact that these spaces lie in R3
is merely a tool for us to picture them; the axioms of a smooth manifold do not
refer to an ambient space (surprisingly, these axioms recover precisely the intuitive
notion of a manifold in Rn ; this is the content of Whitney’s Embedding Theorem,
to be discussed in the lectures).
Manifolds are the mathematical objects that are used to model the abstract
shapes of “physical spaces”. A d-dimensional manifold is a topological space that
locally looks like Rd . For example, the surface of the Earth looks locally flat, like
a piece of the plane, but globally its topology is that of a sphere. The Universe is
modeled by a 3-dimensional manifold because locally it looks like a piece of R3 , but
its global topology might be more complicated. Space-time is a 4-dimensional man-
ifold. The space of possible positions of a ball rolling on a plane is a 5-dimensional
manifold.
Smooth manifolds are not just topological spaces; they are endowed with a
so-called differentiable structure, which allows one to build classical multi-variable
calculus on smooth manifolds. In particular, one defines smooth maps, derivatives,
integration etc. This forces one to treat many standard constructions from calculus
in a rather abstract way, which yields a more conceptual understanding of such
operations. On the other hand, using manifolds to model physical systems, the
differentiable structure allows one to write the laws of Physics, which are usually
expressed as differential equations.
Smooth manifolds are used as the “background space” in many branches of
modern geometry: Riemannian geometry = manifold + smooth distance, complex
geometry = holomorphic manifolds, symplectic geometry = manifold + symplectic
structure, Lie groups = manifold + smooth group multiplication etc.

So, let’s begin!


SOME WORDS ABOUT THIS COURSE 7

Some typographical conventions


• When a section is colored in blue, it indicates that the section was not
covered in the lectures.

• The symbol ♣ next to the statement of a theorem indicates that the proof
will not be given in this course, mostly because it is too involved or too
long.
LECTURE 1

1.1. Brief review of notions from Topology and Analysis


Did I use to know all these things? Was the summer so long?

We start by recalling some notions from Analysis and Topology, which will be
used throughout the course.

Some Topology
Definition 1.1.1. A topology on a set X is a set T of subsets of X, whose
elements are called open subsets, satisfying the axioms:
[
∅, X ∈ T , U, V ∈ T ⇒ U ∩ V ∈ T , (Ui ∈ T , ∀i ∈ I) ⇒ Ui ∈ T .
i∈I

The pair (X, T ) is called a topological space.


Definition 1.1.2. Let Y be a subset of a topological space (X, T ). The induced
topology on Y , also called the subspace topology, is given by:
T |Y := {U ∩ Y : U ∈ T }.
Definition 1.1.3. Let (X, TX ) and (Y, TY ) be topological spaces. A function
f : X −→ Y
−1
is called continuous if f (U ) is open in X for every open set U in Y .
The function f is called a homeomorphism if it is continuous, bijective and
its inverse f −1 : Y → X is also continuous.
Definition 1.1.4. A topological space (X, T ) is called Hausdorff if every two
distinct points can be separated by two open sets; in other words, for every x, y ∈ X,
with x 6= y, there exist U, V ∈ T such that x ∈ U , y ∈ V and U ∩ V = ∅.
Definition 1.1.5. A topological space (X, T ) is called second countable if there
exists a countable family of open sets {Ui }i∈N with the following
S property: for every
open set U there exists an index set I ⊂ N such that U = i∈I Ui . In other words,
the topology has a countable basis.
Next, we recall the definition of the Euclidean topology on
Rn = {x = (x1 , . . . , xn ) : x1 , . . . , xn ∈ R}.
9
10 IOAN MĂRCUT
, , MANIFOLDS

Definition 1.1.6. (1) The open ball in Rn of radius ε > 0 and center x0 ∈ Rn
is the set
Bε (x0 ) = {x ∈ Rn : |x − x0 | < ε},
where | · | is the usual Euclidean length:
p
|x − y| := (x1 − y 1 )2 + . . . + (xn − y n )2 .
(2) A subset U ⊂ Rn is called open if for every x0 ∈ U there exists a positive
number ε > 0 such that Bε (x0 ) ⊂ U . The collection of all such sets forms the
so-called Euclidean topology on Rn .

Smooth functions
Let U ⊂ Rn be an open set. A function f : U → Rm is called differentiable at a
point x ∈ U , if there exists a linear map
dx f : Rn −→ Rm
such that:
|f (x + h) − f (x) − dx f (h)|
lim = 0.
|h|→0 |h|
The map dx f is called the differential of f at x. If dx f exists, then it is also
uniquely determined, since it can be described as the directional derivative of f :
f (x + εv) − f (x)
dx f (v) = lim , v ∈ Rn .
ε→0 ε
Denote the components of f by f = (f 1 , . . . , f m ). The matrix of dx f in the standard
bases of Rn and Rm is given by the Jacobian matrix:
∂f 1 ∂f 1 ∂f 1
 
∂x12 ∂x22 . . . ∂x n
 ∂f ∂f ∂f 2 
dx f =  ∂x1
 ∂x2 . . . ∂x n
.

 ... ... ... ... 
∂f m ∂f m m

∂x1 ∂x2 . . . ∂f
∂xn
Recall also the chain rule for the differential:
Proposition 1.1.7 (The chain rule). Consider two maps f : U → Rm , g : V → U ,
where U ⊂ Rl and V ⊂ Rn are open sets. If g is differentiable at x ∈ V and f
is differentiable at g(x) ∈ U then f ◦ g is differentiable at x, and its differential is
given by
dx (f ◦ g) = (dg(x) f ) ◦ dx g : Rn −→ Rm .
In terms of partial derivatives the chain rule takes the form:
l
∂(f ◦ g) X ∂f ∂g j
i
(x) = j
(g(x)) i (x), for all 1 ≤ i ≤ n.
∂x j=1
∂y ∂x

The first, and most general class of maps are continuous maps; these are called
also functions of class zero, or just C 0 -functions.
Denote the space of all linear maps from Rn to Rm by Lin(Rn , Rm ). Writing
the linear maps in the standard bases, this space can be identified with the space of
n×m-matrices, hence with Rn×m . Using this identification, we endow Lin(Rn , Rm )
with the Euclidean topology on Rn×m .
LECTURE 1. 11

Definition 1.1.8. Let U ⊂ Rn be an open set. A map f : U → Rm is said to


be a C 1 -function if f is differentiable at every x ∈ U , and its differential is a
continuous function from U to Lin(Rn , Rm ):

df : U → Lin(Rn , Rm ), x 7→ dx f ∈ Lin(Rn , Rm ).

Functions of class C k are defined inductively. We say that f : U → Rm is


a C -function, for k ≥ 2, if it is differentiable and its differential is a C k−1 -
k

function.

Let us recall some standard properties of C k -functions:

Proposition 1.1.9. Let U ⊂ Rn be an open set, and let

f = (f 1 , . . . , f m ) : U −→ Rm

be a function.
(1) If f is a C k -function then it is also a C k−1 -function.
(2) The function f is of class C k if and only if its partial derivatives of order ≤ k:

∂lf i ∂f i
  
∂ ∂
= . . . . . . : U → R,
∂xj1 ∂xj2 . . . ∂xjl ∂xj1 ∂xj2 ∂xjl

for 1 ≤ i ≤ m, 0 ≤ l ≤ k, 1 ≤ j1 , . . . , jl ≤ n
exist and are continuous on U .
(3) If f is of class C k then its partial derivatives “commute”, meaning that:

∂lf i ∂lf i
= ,
∂xj1 ∂xj2 . . . ∂xjl ∂xjσ(1) ∂xjσ(2) . . . ∂xjσ(l)
for 1 ≤ i ≤ m, 1 ≤ l ≤ k, 1 ≤ j1 , . . . , jl ≤ n, and every permutation σ of the
set {1, . . . , l}.

We will mostly work with the following type of maps:

Definition 1.1.10. A smooth function, also called a C ∞ -function, is a function


f : U → Rm , where U is an open subset of Rn , which is of class C k for all k ≥ 0.
Equivalently, f is smooth if and only if its partial derivatives of all orders exist and
are continuous.

The following lemma, which will be repeatedly used throughout this course,
illustrates the richness of the class of smooth functions.

Lemma 1.1.11. Fix a point a ∈ Rm , and positive numbers 0 <  < δ. There exists
a smooth map χ : Rm → R such that
• 0 ≤ χ(x) ≤ 1, for all x ∈ Rm ,
• χ(x) = 1, for all x ∈ B (a),
• χ(x) = 0, for all x ∈
/ Bδ (a).

The functions constructed in the lemma are usually called bump-functions


(think about their graph in Rm+1 ). In Exercise 1.2 we show how to construct such
functions and how to prove the lemma.
12 IOAN MĂRCUT
, , MANIFOLDS

Analytic functions and Borel’s Lemma


For completeness, we recall also the class of analytic functions. Let f : U → Rm
be a smooth function defined on an open set U ⊂ Rn . The formal Taylor series
of f at a point a ∈ U is given by:
X 1 ∂ i1 +...+in f
Pfa (T1 , . . . , Tn ) := (a)T1i1 . . . Tnin .
i1 ! . . . in ! (∂x )i1 . . . (∂xn )in
1
i1 ,...,in ≥0

This expression belongs to the vector space of formal power series in the Ti ’s with
coefficients in Rm , denoted:
 
 X 
Rm [[T1 , . . . , Tn ]] := ci1 ,...,in T1i1 . . . Tnin : ci1 ,...,in ∈ Rm .
 
i1 ,...,in ≥0

Definition 1.1.12. The function f is called analytic around a if there exists  > 0
such that for all x ∈ B (a), the Taylor series evaluated on T = x − a converges
absolutely to f (x):
X 1 ∂ i1 +...+in f
f (x) = (a)(x1 − a1 )i1 . . . (xn − an )in .
i1 ! . . . in ! (∂x )i1 . . . (∂xn )in
1
i1 ,...,in ≥0

Finally, f is called analytic on U , or of class C ω , if it is analytic around every


point in U .
In this course we will develop the theory of smooth manifolds starting from
the class of smooth functions. There are theories of manifolds which are devel-
oped based on other classes of functions. There are two main reasons why smooth
functions are usually preferred: first, smooth functions are closed under several
operations (derivatives of smooth functions are smooth, solutions to ODE’s with
smooth coefficients are smooth etc.), and second, the class of smooth function is
very rich and very flexible (especially when compared to analytic functions; see
Exercise 1.3).
Also to show the contrast with analytic functions, we state the following (for a
proof, solve Exercises 1.4 and 1.5):
Theorem ♣ 1.1.13 (Borel’s Lemma). For any formal power series
X
Q(T1 , . . . , Tn ) := ci1 ,...,in T1i1 . . . Tnin ∈ R[[T1 , . . . , Tn ]],
i1 ,...,in ≥0

there exists a smooth function f : Rn → R which has Q as its Taylor series at 0.

1.2. Manifolds
Topological manifolds
Definition 1.2.1. An m-dimensional topological manifold is a Hausdorff, sec-
ond countable topological space, for which every point has an open neighborhood
homeomorphic to an open set in Rm .
The last condition means that a topological manifold looks locally like Rm .
The following terminology is used:
LECTURE 1. 13

Definition 1.2.2. Let M be a topological space. An m-dimensional chart (U, ϕ)


on M consists of a homeomorphism ϕ : U → U e from an open set U ⊂ M to an
e ⊂ Rm . The components of ϕ are denoted by
open set U
ϕ = (x1ϕ , . . . , xm
ϕ ), xiϕ : U −→ R, for 1 ≤ i ≤ m,
and are called the local coordinates on M corresponding to the chart (U, ϕ). The
inverse map ϕ−1 : U e → U is called a local parameterization of M .

Non-mathematically, an atlas is a book with charts. Here is an atlas of Earth:

Definition 1.2.3. Let M be a topological space. An m-dimensional topological


atlas on M (or C 0 -atlas on M ) consists of a collection of m-dimensional charts
A = {(Uα , ϕα )}α∈I covering M , i.e. M = ∪α∈I Uα .
With this terminology, note that a Hausdorff, second countable topological
space is an m-dimensional topological manifold if and only if it admits an m-
dimensional topological atlas.

Smooth manifolds
Let us recall:
Definition 1.2.4. A map f : U → V between open sets U, V ⊂ Rm is called a
diffeomorphism if f is bijective and both f and f −1 are smooth.
Note that (as in the case of homeomorphisms) the condition that f −1 be smooth
is not automatically satisfied. The standard example of a smooth bijection which
is not a diffeomorphism is
f : R −→ R, f (t) = t3 .
In a local chart a topological manifold is described as an open piece of Rm . To
develop analysis on manifolds, one needs to introduce derivatives and integration of
functions, notions which, locally in charts, should coincide with those from multi-
variable calculus. However, a function that is differentiable in one chart might fail
to be differentiable in different chart! To circumvent this, on a smooth manifold
one only works with mutually compatible charts.
Definition 1.2.5. Let M be a topological space. Let (U, ϕ) and (V, ψ) be two m-
dimensional charts on M . The map
ϕ ◦ ψ −1 : ψ(U ∩ V ) −→ ϕ(U ∩ V )
14 IOAN MĂRCUT
, , MANIFOLDS

is a called the change of coordinates map or the transition map between the
two charts. The two charts are said to be compatible if the transition map is a
diffeomorphism.
Definition 1.2.6. An m-dimensional topological atlas A on an m-dimensional
topological manifold M is said to be an m-dimensional smooth atlas (or C ∞ -
atlas) if every two charts in A are compatible.
Here are some simple examples of C ∞ -atlases:
Example 1.2.7. (1) On Rm there is a smooth atlas with only one chart
A = {(Rm , idRm )}.
Another smooth atlas is the collection of all diffeomorphisms between open
subsets of Rm :
B = { (U, ϕ) : U, V ⊂ Rm are open and ϕ : U → V is a diffeomorphism}.
In fact, B consists of all charts compatible with (Rm , idRm ).
(2) Consider the m-dimensional sphere with the induced topology from Rm+1 :
S m = (x0 , x1 , . . . , xm ) : (x0 )2 + (x1 )2 + . . . + (xm )2 = 1 ⊂ Rm+1 .


We construct an atlas on S m with only two charts. The “south pole” and the
“north pole” of S m are the points
s := (0, 0, . . . , 0, −1) ∈ S m , resp. n := (0, 0, . . . , 0, 1) ∈ S m .
The stereographic projection through the north pole is the map
πn : S m \{n} −→ Rm
which sends a point p ∈ S m , p 6= n, to the point q = πn (p) which is the
intersection of the plane xm = 0 and the line through n and p.
n
Sm

q
xm = 0
p

Explicitly, for p = (x0 , . . . , xm ), the point q must satisfy q = (y, 0) = (1−t)n+tp


1
for some t ∈ R. The last coordinate gives 1 − t + txm = 0, hence t = 1−x m,

and so,
1
πn (x0 , . . . , xm ) = (x0 , . . . , xm−1 ).
1 − xm
Similarly, we have the stereographic projection through the south pole:
1
πs : S m \{s} → Rm , πs (x0 , . . . , xm ) = (x0 , . . . , xm−1 ).
1 + xm
We claim that the following is a smooth atlas on S m :
A = {(S m \{n}, πn ), (S m \{s}, πs )}.
LECTURE 1. 15

First we show that the stereographic projections are homeomorphisms. Note


that the formula defining πn makes sense on the open set U = Rm+1 \{xm = 1},
and that it defines a continuous map U → Rm . Thus πn is the restriction to
U ∩ S m of a continuous map and, using the definition of the induced topology,
this implies that πn is continuous. The same argument implies that πs is
continuous. Finally, let us determine the inverse of πn . We have that p =
πn−1 (y) is the point at the intersection of S m and the line through q = (y, 0)
and n, and which is not n. So p = t(y, 0) + (1 − t)n, with t ∈ R. Writing
|p|2 = 1, we obtain: t2 |y|2 + (1 − t)2 = 1; or t2 (|y|2 + 1) = 2t. The solution
2
t = 0 corresponds to n; thus t = 1+|y| 2 . Using this, we obtain that the inverse

of πn is given by:
πn−1 : Rm −→ S m \{n},
2y 1 2y 2 2y m |y|2 − 1
 
−1 1 2 m
πn (y , y , . . . , y ) = , ,..., 2 , .
|y|2 + 1 |y|2 + 1 |y| + 1 |y|2 + 1
All functions appearing in this expression are continuous, and this proves that
πn is a homeomorphism. A similar calculation shows that also πs is a homeo-
morphism.
Finally, we need to check that the charts are C ∞ -compatible. Since πn (s) =
πs (n) = 0, the transition map is defined as follows:
πs ◦ πn−1 : Rm \{0} −→ Rm \{0},
and using the formulas above, we obtain that this map is given by:
|y|2 − 1
 
2y 1 2y y
πs ◦ πn−1 (y) = πs 2
, 2
= |y|2 −1 2
= 2.
|y| + 1 |y| + 1 1 + 2 |y| + 1 |y|
|y| +1

This map is clearly smooth. Note that the map satisfies (πs ◦ πn−1 )2 = idRm \{0} .
This shows that it equals its own inverse, hence it is a diffeomorphism.
We introduce a relation on atlases:
Definition 1.2.8. Let M be an m-dimensional topological manifold. Consider the
following relation on m-dimensional C ∞ -atlases on M :
def.
A1 ∼ A2 ⇐⇒ A1 ∪ A2 is a C ∞ − atlas.
As expected:
Proposition 1.2.9. The relation ∼ is an equivalence relation on the set of m-
dimensional C ∞ -atlases of M .
Proof. Reflexivity and symmetry of the relation ∼ are obvious. We will check that
transitivity holds. Consider three m-dimensional C ∞ -atlases on M : A1 , A2 and
A3 such that A1 ∼ A2 and A2 ∼ A3 . To check that A1 ∼ A3 we need to show
that each pair of charts (U1 , ϕ1 ) ∈ A1 and (U3 , ϕ3 ) ∈ A3 are compatible, i.e. that
the map
(1) ϕ1 ◦ ϕ−1
3 : ϕ3 (U1 ∩ U3 ) −→ ϕ1 (U1 ∩ U3 )

is a diffeomorphism. This map is clearly a bijection, with inverse


(2) ϕ3 ◦ ϕ−1
1 : ϕ1 (U1 ∩ U3 ) −→ ϕ3 (U1 ∩ U3 ).
16 IOAN MĂRCUT
, , MANIFOLDS

So, it suffices to prove that the maps (1) and (2) are smooth. Let p ∈ U1 ∩ U3 .
Since A2 is an atlas, there exists a chart (U2 , ϕ2 ) ∈ A2 such that p ∈ U2 . Since
A1 ∼ A2 and A2 ∼ A3 is follows that the following maps are diffeomorphisms:
ϕ1 ◦ ϕ−1
2 : ϕ2 (U1 ∩ U2 ) −→ ϕ1 (U1 ∩ U2 ),

ϕ2 ◦ ϕ−1
3 : ϕ3 (U2 ∩ U3 ) −→ ϕ2 (U2 ∩ U3 ).
In particular, the restriction of their composition is a diffeomorphism:
ϕ1 ◦ ϕ−1
3 : ϕ3 (U1 ∩ U2 ∩ U3 ) −→ ϕ1 (U1 ∩ U2 ∩ U3 ).
Hence, the map (1) and its inverse (2) are smooth when restricted to the open
neighborhoods ϕ3 (U1 ∩ U2 ∩ U3 ) and ϕ1 (U1 ∩ U2 ∩ U3 ), respectively, of ϕ3 (p) and
ϕ1 (p), respectively. Since p was chosen arbitrary in U1 ∩ U3 , it follows that (1) and
(2) are smooth everywhere. This finishes the proof. 
Next, we define:
Definition 1.2.10. An m-dimensional C ∞ -atlas A on the topological manifold M
is said to be maximal if
A ∼ B =⇒ B ⊂ A

for any m-dimensional C -atlas B on M .
We have that:
Proposition 1.2.11. Any equivalence class of smooth atlases has a unique maximal
representative. The maximal C ∞ -atlas equivalent to the C ∞ -atlas A is given by:
Amax = {(U, ϕ) : A ∪ {(U, ϕ)} is a C ∞ -atlas} .
Proof. Let A be an m-dimensional C ∞ -atlas on M . It is clear that there exists
at most one maximal smooth atlas which is equivalent to A, because if there were
two, A1max and A2max , then by transitivity of the relation ∼ we would have that
A1max ∼ A2max , and since both atlases are maximal, this implies that
A1max ⊂ A2max ⊂ A1max ;
thus A1max = A2max .
To prove existence of a maximal C ∞ -atlas equivalent to A, let Amax be the set
constructed in the statement. We check that Amax is indeed a C ∞ -atlas. First,
since A ⊂ Amax , it follows that the open sets in Amax cover M . Second, we check
that each pair of charts (U, ϕ) and (V, ψ) in Amax are compatible. Clearly A ∼
A ∪ {(U, ϕ)} and A ∼ A ∪ {(V, ψ)}, therefore, by transitivity of ∼, A ∪ {(U, ϕ)} ∼
A ∪ {(V, ψ)}, hence A ∪ {(U, ϕ)} ∪ {(V, ψ)} is a C ∞ -atlas. This implies that the
charts (U, ϕ) and (V, ψ) are compatible, and proves that Amax is indeed a C ∞ -atlas.
Finally, we prove that Amax is maximal. Note first that A ∼ Amax . Let B be
a C ∞ -atlas such that B ∼ Amax . Then, by transitivity of ∼, B ∼ A, which implies
that every chart in B is compatible with every chart in A; hence A ∪ {(U, ϕ)} is an
atlas for every (U, ϕ) ∈ B. The definition of Amax implies now that B ⊂ Amax . 
Definition 1.2.12. An m-dimensional differentiable structure (or smooth
structure) on a topological manifold M is an equivalence class of m-dimensional
C ∞ -atlases on M . Equivalently, by Proposition 1.2.11, a differentiable structure is
the same as a maximal atlas on M .
We are ready to give the general definition of a smooth manifold:
LECTURE 1. 17

Definition 1.2.13. An m-dimensional smooth manifold is a topological manifold


endowed with an m-dimensional differentiable structure.
Example 1.2.14. The topological space Rm and S m are smooth manifolds when
endowed with the differentiable structure corresponding to the C ∞ -atlases con-
structed in Example 1.2.7. Note that B Example 1.2.7 (1) is the maximal atlas on
Rm . The obtained differentiable structures on Rm and S m are called the standard
differentiable structures. A differentiable structure which is not equivalent to the
standard one is called an exotic differentiable structure. These do exist in certain
dimensions; see Section 2.3 for more details.
Remark 1.2.15. The axiom of being second countable insures that manifolds are
not “too big”. Let us mention that there are examples of Hausdorff topological
spaces, endowed with a smooth atlas, but which are not second countable. An
easy example is the disjoint union of an uncountable collection of manifolds of the
same dimension; e.g. an uncountable collection of points with the discrete topology
is not second countable, and has a 0-dimensional atlas. There are also connected
examples, but these are more difficult to describe; a famous one is the Long Line
(see e.g. [17, 18]).

1.3. Exercises
Exercise 1.1. (a) Show that Rn is Hausdorff and second countable.
(b) Show that if X is a Hausdorff topological space and Y ⊂ X is a subset, then
the induced topology on Y is Hausdorff.
(c) Show that if X is a second countable topological space and Y ⊂ X is a subset,
then the induced topology on Y is second countable.
Exercise 1.2. (a) Prove that the following function is smooth but is not analytic

0, x ≤ 0;
f : R −→ R, f (x) =
e−1/x , x > 0.
(b) Show that g : R → R, g(x) := f (x)f (1 − x) is a smooth function which is
positive on (0, 1) and zero elsewhere. Show also that the function h : R → R,
Rx
g(y)dy
h(x) := R01
0
g(y)dy
is a smooth and it satisfies: h(x) = 0 for x < 0, 0 < h(x) < 1 for 0 < x < 1,
and h(x) = 1 for x > 1.
(c) Use (a) and (b) to prove Lemma 1.1.11.
Exercise 1.3. Prove that there are no analytic functions χ : Rm → R with the
properties from Lemma 1.1.11.
The following exercise is used to prove Borel’s Lemma:
Exercise 1.4. Let Cbk (Rn ) denote the set of C k -function f : Rn → R which have
all partial derivatives up to order k bounded. On this space we define the so-called
C k -norm:
∂ i1 +...+in f
 
: x ∈ Rn , i1 + . . . + in ≤ k .

kf kk := sup (x)
(∂x1 )i1 . . . (∂xn )in
Let Cb∞ (Rn ) denote the set of smooth maps with all partial derivatives bounded.
18 IOAN MĂRCUT
, , MANIFOLDS

(a) Prove that (Cbk (Rn ), k · kk ) is a Banach space, i.e. prove that: every sequence
of functions in Cbk (Rn ) which is Cauchy with respect to the norm k · kk is in
fact convergent to an element of Cbk (Rn ).
(b) Prove that Cb∞ (Rn ) endowed with the family of norms {k · kk }k≥0 is a Fréchet
space, i.e. prove that: every sequence of functions in Cb∞ (Rn ) which is Cauchy
with respect to all the norms {k · kk }k≥0 is in fact convergent with respect to
all the norms {k · kk }k≥0 to an element in Cb∞ (Rn ).
In the next exercise you are asked to prove Borel’s Lemma; note that it uses
the previous exercise.
Exercise 1.5. Consider a formal power series
X
Q(T1 , . . . , Tn ) := ci1 ,...,in T1i1 . . . Tnin ∈ R[[T1 , . . . , Tn ]].
i1 ,...,in ≥0
n
Let χ : R → R be a smooth function such that χ(x) = 1, for x ∈ B1 (0), and
χ(x) = 0 for x ∈
/ B2 (0) (by Lemma 1.1.11, such functions exist).
For k ≥ 0, define the following function depending on a positive number k > 0:
 
X x
fk (x) := χ · ci1 ,...,in (x1 )i1 . . . (xn )in .
k
i1 +...+in =k

(a) Prove that fk ∈ Cb∞ (Rn ).


(b) For k ≥ 1, prove that one can choose k > 0 such that
kfk kk−1 < 2−k .
(c) Let k > 0 be as in (b). Show that the series:
X
fk
k≥0

converges to a smooth function f ∈ Cb∞ (Rn ).


(d) Prove that the Taylor series of f at 0 is Q.
Exercise 1.6. Consider the map
f : R −→ R, f (t) = t3 .
(a) Show that f is a smooth homeomorphism of R.
(b) Show that, there are no open neighborhoods U and V of 0 in R such that
f |U : U → V is a diffeomorphism.
(c) On the topological manifold R consider the atlases
A1 = {(R, idR )} and A2 = {(R, f )}.
Is A1 ∼ A2 ?
Exercise 1.7. On the unit sphere S m ⊂ Rm+1 consider the following collection of
2(m + 1) maps:
B = {(Ui , ϕi ) : 0 ≤ i ≤ m,  = ±1},
Ui = {(x0 , . . . , xm ) ∈ S m :  · xi > 0},
ϕ (x0 , . . . , xm ) = (x0 , . . . , xbi , . . . , xm ),
i
where the circumflex b indicates that the underlying term has been deleted.
(a) Prove that the pairs (Ui , ϕi ) are indeed charts, and prove that B is a smooth
atlas on S m .
LECTURE 1. 19

(b) Prove that the atlas A from Example 1.2.7 (2) and the atlas B define the same
differentiable structure on S m .
LECTURE 2

2.1. Smooth maps


We introduce smooth maps between smooth manifolds. First, let us define:
Definition 2.1.1. A smooth chart on a smooth manifold M is a chart (U, ϕ)
which belongs to the maximal C ∞ -atlas of M .
Definition 2.1.2. Let M and N be smooth manifolds of dimension m and n,
respectively. A continuous map f : M → N is said to be smooth (or C ∞ ), if for
every smooth chart (U, ϕ) on M and every smooth chart (V, ψ) on N , the map:
ψ ◦ f ◦ ϕ−1 : ϕ(U ∩ f −1 (V )) −→ ψ(V )
is smooth. The map ψ ◦ f ◦ ϕ−1 , which is defined between open sets in Rm and Rn
is called the local representation of f in the charts (U, ϕ) and (V, ψ).
The space of smooth maps from M to N is denoted by C ∞ (M, N ). If N = R
(as a smooth manifold) we simply write C ∞ (M ) := C ∞ (M, R).
The condition that f be continuous in the previous definition is necessary to
insure that U ∩ f −1 (V ) is open.
Remark 2.1.3. Smoothness can be checked on an open cover. Namley, a contin-
uous map f : M → N is smooth iff for each p ∈ M there exists a smooth chart
(U, ϕ) on M , with p ∈ U , and a smooth chart (V, ψ) on N with f (p) ∈ V such that
the local representation of f in these charts is smooth. This follows because any
two local representative around p and f (p) are related by composing on both sides
with local diffeomorphisms.
Composition of smooth maps is smooth:
Proposition 2.1.4. Let f : M → N and g : N → P be smooth maps between
smooth manifolds. Then g ◦ f : M → P is a smooth map.
Next, we extend also the notion of a diffeomorphism:
Definition 2.1.5. A map f : M → N between smooth manifolds M and N is
called a diffeomorphism if f is bijective and both f and f −1 are smooth. If there
exists a diffeomorphism f : M → N , then the manifolds M and N are said to be
diffeomorphic.
21
22 IOAN MĂRCUT
, , MANIFOLDS

Let us mention here that there is a product operation on manifolds:


Proposition 2.1.6. Let M and N be smooth manifolds of dimensions m, respec-
tively n. Then the product M × N endowed with the product topology is a smooth
m + n-dimensional manifold. A smooth atlas on M × N is
AM ×N := {(U × V, ϕ × ψ) : (U, ϕ) ∈ AM , (V, ψ) ∈ AN } ,
where AM is a smooth atlas on M and AN is a smooth atlas on N . Moreover, the
projection maps
pr1 : M × N → M, pr1 (x, y) = x, pr2 : M × N → N, pr2 (x, y) = y.
are smooth.

2.2. How many smooth structures?


This section is mostly for the general knowledge of the reader, and will not be used
in further lectures. The results stated here are not proven in this course, as they
are beyond the scope of these lectures; for more details, see the references in the
Wikipedia article [16].
Let us state one of the problems which has been of central interest since the
beginning of the theory of smooth manifolds:

Let M be a topological manifold. Can M be made into a smooth manifold? If


so, in how many non-equivalent ways?

Let us be a bit more precise about the second part of the questions. Assume that
M is a smooth manifold in two different ways, corresponding to the differentiable
structures D1 and D2 . We regard these differentiable structures are equivalent, if
there exists a diffeomorphism relating the two structure:
ϕ : (M, D1 ) −→ (M, D2 ).
Thus, the second part of the questions asks how many non-equivalent such differ-
entiable structures exist.
For small dimension, we have existence and uniqueness:
Theorem ♣ 2.2.1. For m ≤ 3, every m-dimensional topological manifold has a
smooth structure, and any two such smooth structures are equivalent (for m = 1, 2
Radó, 1920s, for m = 3 Moise, 1953, [9]).
However, in larger dimension:
Theorem ♣ 2.2.2. There exist compact topological manifolds of dimension ≥ 4
which do not admit a differentiable structure (Kervaire 1960, [6] first example in
dimension 10; nowadays many examples are known even of dimension 4).
Finally, also uniqueness fails in dimension m ≥ 4; there are topological man-
ifolds which admit non-equivalent smooth structures. The non-standard ones are
usually called exotic. The first examples of non-equivalent smooth structures on a
topological manifold were Milnor’s exotic 7-spheres:
Theorem ♣ 2.2.3. The topological manifold S 7 admits exactly 28 non-equivalent
smooth structures (Milnor 1956, [7]).
LECTURE 2. 23

However, it is not known whether S 4 has exotic smooth structures! Here is a


table with the number of exotic structure for spheres of dim ≤ 20:

dimension 1 2 3 4 5 6 7 8 9 10
C ∞ − str.’s 1 1 1 ? 1 1 28 2 8 6

dimension 11 12 13 14 15 16 17 18 19 20
C ∞ − str.’s 992 1 3 2 16256 2 16 16 523264 24
Even more surprisingly are the exotic R4 ’s:
Theorem ♣ 2.2.4. For m 6= 4, the topological manifold Rm admits a unique
smooth structure (Stallings, 1962 [12]). The topological manifold R4 admits un-
countably many non-equivalent smooth structures (Taubes 1987, [13]).

2.3. C k -manifolds
Throughout this section we fix k which is either an integer k ≥ 1, or k ∈ {∞, ω}.
We briefly introduce the notion of C k -manifolds.
All notions discussed so far can be extended directly to the C k -setting, by
simply replacing the requirement that a map be smooth with that of being C k .
For example, a C k -diffeomorphism is a map f : U → V between open sets
U, V ⊂ Rm which is bijective and both f and f −1 are C k -maps. Note the following:
Lemma 2.3.1. If f : U → V is a C 1 -diffeomorphism and a C k -map, then f is a
C k -diffeomorphism.
Proof. We need to check that the inverse of f , denoted by g : V → U , is also of
class C k . For k = ω, we omit the proof (one should write the power series expansion
of the equation f ◦ g = id, and determine a term-by-term formula for g, and then
show that this is convergent). Let k 6= ω. Differentiating the relation f ◦ g = id, by
the chain rule, we obtain that
dg(x) f ◦ dx g = id.
This gives that the differential of g is the inverse of the differential of f :
dx g = (dg(x) f )−1 .
This relation shows that if g is C l and f is C l+1 then g is C l+1 (the right hand side
is a composition of C l -maps, and the smooth (!) operation of taking the inverse of
a matrix). An argument by induction, for 1 ≤ l ≤ k, proves the statement. 
One introduces C k -compatible charts, C k -atlases, C k -differentiable structures,
and finally, C k -manifolds, all in perfect analogy with the smooth setting.

Let l, k ∈ {0, 1, 2, . . . , ∞, ω} l < k (where ∞ < ω). A C k -differentiable struc-


ture Dk on M is said to extend a given C l -differentiable structure Dl on M if every
C k -atlas representing Dk , when regarded as a C l -atlas it represents Dl . Two exten-
sions D1k and D2k of Dl are said to be equivalent, if there exists a C k -diffeomorphism
relating them:
ϕ : (M, D1k ) −→ (M, D2k ).
The following theorem says that C l -differentiable structures, with 0 < l < ∞,
are essentially the same as smooth structures (or even analytic structures).
24 IOAN MĂRCUT
, , MANIFOLDS

Theorem ♣ 2.3.2. If 1 ≤ l < k, with l, k ∈ {1, 2, . . . , ∞, ω}, then any C l -


differentiable structure can be extended to a C k -differentiable structure, and any
two extensions are equivalent (Whitney, 1936 [15]).
Whitney’s result also explains why the problem discussed in the previous section
is only interesting for topological manifolds.

2.4. Embedded manifolds in Rn


In this section we introduce embedded manifolds in Rn . These are subsets of Rn
which, locally, can be straighten out smoothly to look like an m-dimensional linear
space. This gives a large class of examples of smooth manifolds, and in fact, Whit-
ney’s theorem (which will be discussed later on) states that any smooth manifold
can be regarded as an embedded manifold in some Rn .
Definition 2.4.1. A subset M ⊂ Rn is called an m-dimensional embedded man-
ifold in Rn if around every point in M there exists an open set U ⊂ Rn and there
exists a diffeomorphism f : U → V , where V ⊂ Rn is open, such that:
f (M ∩ U ) = (Rm × {0}) ∩ V.
We will call a diffeomorphism (U, f ) as above a chart adapted to M .
Example 2.4.2. Let us check that the circle
S 1 ⊂ R2 , S 1 = {(x, y) ∈ R2 : x2 + y 2 = 1}
is a 1-dimensional embedded manifold in R2 . Consider the map
f (x, y) := (x, x2 + y 2 − 1).
Note that f is a diffeomorphism between the open sets
U := {(x, y) ∈ R2 : y > 0}, V := {(u, v) ∈ R2 : v > u2 − 1},
with inverse given by
p
f −1 : V −→ U, f −1 (u, v) = (u, v + 1 − u2 ).
Note that f (x, y) ∈ R × {0} if and only if (x, y) ∈ S 1 ; thus, f satisfies:
f (S 1 ∩ U ) = (R × {0}) ∩ V.
The open set U covers the upper half of the circle. On the open sets U1 := −U ,
U2 := {(x, y) : x > 0} and U3 := −U2 , similar diffeomorphisms can be constructed:
f1 := f |U1 , f2 := g|U2 , f3 := g|U3 , where g(x, y) := (y, x2 + y 2 − 1).
We will use the following notations for the canonical projections and inclusions,
respectively, between Rm and Rn , with m ≤ n:
prnm : Rn −→ Rm , prnm (x1 , . . . , xn ) := (x1 , . . . , xm )
im m n m 1 m 1 m
n : R −→ R , in (x , . . . , x ) := (x , . . . , x , 0, . . . , 0).

Proposition 2.4.3. Let M be an m-dimensional embedded manifold in Rn . Then


M , endowed with the induced topology, is a smooth m-dimensional manifold with
differentiable structure represented by the atlas
A = {(M ∩ U, prnm ◦ f |M ∩U ) : (U, f ) is a chart adapted to M }.
LECTURE 2. 25

Proof. According to Exercise 1.1, the induced topology is Hausdorff and second
countable. We need to check that A is a smooth atlas on M . By definition, M is
covered by adapted charts. Let (U, f ) be an adapted chart, and denote V := f (U )
and ϕ := prnm ◦ f |M ∩U . First, note that M ∩ U is open in the induced topology
on M , and the map ϕ is continuous for the induced topology, because it is the
restriction of a continuous map on U . We have that
ϕ(M ∩ U ) = prnm ◦ f (M ∩ U ) = prnm ((Rm × {0}) ∩ V ) ⊂ Rm .
Since prnm restricts to a homeomorphism between Rm × {0} (with the induced
topology) and Rm , we have that ϕ(M ∩ U ) is open in Rm . Finally, the map ϕ is a
homeomorphism onto its image, because its inverse is given by as a composition of
continuous maps:
ϕ−1 : ϕ(M ∩ U ) −→ M ∩ U, ϕ−1 = f −1 ◦ im
n |ϕ(M ∩U ) .
This proves that (M ∩ U, ϕ) is an m-dimensional chart on M . Thus, M is a
topological m-dimensional manifold.
To show that A is a smooth atlas, we need to check that any two charts are
C ∞ -compatible. Let f : U → V and f 0 : U 0 → V 0 be two adapted charts, inducing
the following charts in A:
(M ∩ U, ϕ := prnm ◦ f |M ∩U ), (M ∩ U 0 , ϕ0 := prnm ◦ f 0 |M ∩U 0 ).
The corresponding transition map is smooth since it is given by:
ϕ0 ◦ ϕ−1 : ϕ(M ∩ U ∩ U 0 ) −→ ϕ0 (M ∩ U ∩ U 0 ), prnm ◦ f 0 ◦ f −1 ◦ im
n.
By interchanging the roles of f and f 0 we see that its inverse is also smooth. Hence
any two charts in A are C ∞ -compatible. 
The converse of this result also holds:
Theorem ♣ 2.4.4 (Whitney’s Embedding Theorem). Any m-dimensional smooth
manifold, for m ≥ 1, is diffeomorphic to an embedded manifold in R2m , which is a
closed subset.

2.5. Exercises
Exercise 2.1. Prove Proposition 2.1.6.
Exercise 2.2. Let M be an embedded manifold in Rn . Show that the inclusion
map M ,→ Rn , x 7→ x, is smooth.
Exercise 2.3. (a) Prove that S m is an embedded manifold in Rm+1 (as in Defini-
tion 2.4.1).
(b) Prove that the smooth structure on S m resulting from Proposition 2.4.3 co-
incides with the smooth structure on S m coming from the smooth atlas in
Example 1.2.7.
LECTURE 3

3.1. The topology is determined by the atlas


It is important to notice that an atlas determines the topology uniquely. This
remark allows one to describe smooth manifolds by giving a collection of charts on
a set. We state this in the following proposition.
Proposition 3.1.1. Let M be a set and let A be a collection of maps
A = {(Uα , ϕα ) : α ∈ I}, Uα ⊂ M, ϕα : Uα −→ Rm ,
which satisfy the following conditions:
(1) ∪α∈I Uα = M ;
(2) for each α ∈ I, ϕα is injective;
(3) for each α, β ∈ I, ϕβ (Uα ∩ Uβ ) is open in Rm ;
in particular, for α = β, ϕα (Uα ) is open in Rm ;
(4) for each α, β ∈ I, the map ϕα ◦ ϕ−1 β : ϕβ (Uα ∩ Uβ ) → ϕα (Uα ∩ Uβ ) is a
diffeomorphism.
There exists a unique topology TA on M such that, for all α ∈ I, the pair (Uα , ϕα )
is a chart on M . Moreover,
• If I is at most countable, then TA is second countable.
• If for every two distinct points p, q ∈ M there exist α, β ∈ I and open sets
Oα and Oβ in Rm such that
−1 −1
p ∈ ϕ−1 −1
α (Oα ), q ∈ ϕβ (Oβ ), ϕα (Oα ) ∩ ϕβ (Oβ ) = ∅,

then TA is Hausdorff.
• If TA is Hausdorff and second countable, then M has an m-dimensional
differentiable structure for which A is a C ∞ -atlas.
Proof. First, we show that there is at most one topology TA on M for which A is
an atlas: since {Uα }α∈I is an open cover of M w.r.t. TA , we have that

V ∈ TA ⇐⇒ ∀ α ∈ I : V ∩ Uα ∈ TA ;
and since ϕα : Uα → ϕα (Uα ) is a homeomorphism, we have that
V ∈ TA ⇐⇒ ∀ α ∈ I : ϕα (V ∩ Uα ) is open in Rm ).
27
28 IOAN MĂRCUT
, , MANIFOLDS

The above defines TA uniquely. Next we show that TA is indeed a topology. Clearly,
∅ ∈ TA , and by (3), M ∈ TA . Let V1 , V2 ∈ TA . For every α ∈ I, we have that
(2)
ϕα (Uα ∩ V1 ∩ V2 ) = ϕα ((Uα ∩ V1 ) ∩ (Uα ∩ V2 )) = ϕα (Uα ∩ V1 ) ∩ ϕα (Uα ∩ V2 ),
which shows that ϕα (Uα ∩ V1 ∩ V2 ) is open. This implies that V1 ∩ V2 ∈ TA . Finally,
consider an arbitrary family {Vλ }λ∈Λ ⊂ TA . Then, for each α ∈ I,
ϕα (Uα ∩ (∪λ∈Λ Vλ )) = ϕα (∪λ∈Λ Uα ∩ Vλ ) = ∪λ∈Λ ϕα (Uα ∩ Vλ ),
which shows that ϕα (Uα ∩ (∪λ∈Λ Vλ )) is open. Hence, ∪λ∈Λ Vλ ∈ TA . We conclude
that TA is a topology.
Let us see that (Uα , ϕα ) is indeed a chart on M for the topology TA . By (3),
Uα is open. Next, we show that ϕα : Uα → ϕα (Uα ) is a homeomorphism. For any
open set O ⊂ ϕα (Uα ), we have that ϕβ (Uβ ∩ ϕ−1 −1
α (O)) = ϕβ ◦ ϕα (O) which is open
because, by (4), ϕβ ◦ ϕα is a homeomorphism. Since β is arbitrary, ϕ−1
−1
α (O) ∈ TA .
So, ϕα is continuous. By definition of TA , we have that also ϕ−1 α : ϕα (Uα ) → M
is continuous. Thus, the maps ϕα : Uα → ϕα (Uα ) are homeomorphisms, i.e. charts
on M for the topology TA .
Assuming that I is at most countable, we prove that TA is second countable.
Let {Ok }k∈N be a countable basis for the topology of Rm , i.e. every open set U ⊂ Rm
can be represented as U = ∪k∈JU Ok , for a subset JU ⊂ N. We claim that the family
of open sets {ϕ−1
α (Ok )}α∈I,k∈N is a basis for TA . Since I is at most countable, the
index set I × N is countable. Let V ∈ TA be an open set. First, decompose
V = ∪α∈I (V ∩ Uα ). Since ϕα (V ∩ Uα ) is open in Rm , there exists a subset Jα ⊂ N
such that ϕα (V ∩ Uα ) = ∪k∈Jα Ok . Thus we have that
V = ∪α∈I V ∩ Uα = ∪α∈I ∪k∈Jα ϕ−1
α (Ok ).

This shows that TA is second countable.


The criterion for being Hausdorff is clear: the points p and q are separated by
−1
the open sets ϕ−1
α (Oα ) and ϕβ (Oβ ).
By (4) the charts in A are pairwise compatible, which makes the final conclusion
obvious. 

3.2. The real projective space


Proposition 3.1.1 allows us to describe the structure of a smooth manifold on a set
in terms of a collection of local maps to Rm satisfying certain conditions. Here is
an important example:
Example 3.2.1. The m-dimensional real projective space is defined as the set
of all lines through the origin in Rm+1 :
Pm (R) := { l : l ⊂ Rm+1 is a line with 0 ∈ l}.
Note first that a non-zero vector in Rm+1 belongs to a unique line in Pm (R), and
two vectors lie on the same line Pm (R) iff one is a scalar multiple of the other. This
allows us to identify
Pm (R) = Rm+1 \{0} / ∼,


where ∼ is the equivalence relation on Rm+1 \{0} given by


v∼w ⇐⇒ v = tw, for some 0 6= t ∈ R.
LECTURE 3. 29

We denote the equivalence class of v = (x0 , x1 , . . . , xm ) 6= 0 by:


[v] = [x0 : x1 : . . . : xm ] = (tx0 , tx1 , . . . , txm ) : 0 6= t ∈ R .


Consider the following m + 1 subsets of Pm (R):


Ui := [x0 : x1 : . . . : xm ] : xi 6= 0 ⊂ Pm (R), 0 ≤ i ≤ m.


and the maps:


ϕi : Ui −→ Rm ,
 0
xi−1 xi+1 xm

0 i−1 i i+1 m x
ϕi ([x : . . . : x :x :x : . . . : x ]) := ,..., i , i ,..., i .
xi x x x
Note that the pair (Ui , ϕi ) is well-defined: if (x0 , . . . , xm ) ∼ (y 0 , . . . , y m ) then
j j yj
y j = txj , for some t 6= 0. So, xi 6= 0 iff y i 6= 0, and xxi = tx
txi = y i .
We claim that Pm (R) is a smooth m-dimensional manifold with C ∞ -atlas
A = {(Ui , ϕi ) : 0 ≤ i ≤ m}.
We verify the condition (1)-(4) in Proposition 3.1.1. Since each point in Rm+1 \{0}
has at least one non-zero coordinate, it follows that each point in Pm (R) lies in one
of the sets Ui , hence condition (1) is satisfied
[
Pm (R) = Ui .
0≤i≤m
m
Condition (2) follows since ϕi : Ui → R is a bijection with inverse
ϕ−1
i :R m
−→ Ui , ϕ−1 1 m
i (y , . . . , y ) = [y 1 : . . . : y i : 1 : y i+1 : . . . : y m ].
For (3), note that

 {y ∈ Rm : y j+1 6= 0}, if i > j
ϕi (Ui ∩ Uj ) = {y ∈ Rm : y j 6= 0}, if i < j
 m
R , if i = j
Finally, (4) follows by calculating the transition maps explicitly; for i < j:
ϕj ◦ ϕ−1 i : {y ∈ Rm : y j 6= 0} −→ {y ∈ Rm : y i+1 6= 0}
 1
y i 1 y i+1 y j−1 y j+1 ym

−1 1 m y
ϕj ◦ ϕi (y , . . . , y ) = ,..., j, j, j ,..., j , j ..., j ,
yj y y y y y y
and
ϕi ◦ ϕ−1 j : {y ∈ R : y
m i+1
6= 0} −→ {y ∈ Rm : y j 6= 0}
 1
y i y i+2 yj 1 y j+1 ym

y
ϕi ◦ ϕ−1 j (y 1
, . . . , y m
) = , . . . , , , . . . , , , , . . . , .
y i+1 y i+1 y i+1 y i+1 y i+1 y i+1 y i+1
By Proposition 3.1.1, Pm (R) has a unique topology for which A is an m-
dimensional atlas. Since the atlas is finite, this topology is second countable.
It remains to check that the topology is Hausdorff. Consider two distinct
points p = [p0 : . . . : pm ], q = [q 0 : . . . : q m ] ∈ Pm (R). Assume first that there exists
0 ≤ i ≤ m such that p, q ∈ Ui . If Op and Oq are two open sets in Rm separating
ϕi (p) and ϕi (q), then p and q can be separated by the open sets ϕ−1 i (Op ) and
ϕ−1
i (O q ). Assume now that there is no 0 ≤ i ≤ m such that Ui contains both p and
q. By (1) there exist 0 ≤ i, j ≤ m such that p ∈ Ui and q ∈ Uj . Since p ∈ / Uj and
q∈ / Ui , we have that
pi 6= 0, pj = 0, q i = 0, q j 6= 0.
30 IOAN MĂRCUT
, , MANIFOLDS

We claim that the following two sets are open and separate p and q:
U (p) := {[x0 : . . . : xm ] : |xi | > |xj |}, U (q) := {[x0 : . . . : xm ] : |xi | < |xj |}.
Note first that these sets are well-defined. Clearly, p ∈ U (p), q ∈ U (q) and U (p) ∩
U (q) = ∅. Next, note that U (p) ⊂ Ui and that ϕi (U (p)) is open in Rm . Since ϕi
is a homeomorphism, we conclude that U (p) is open. Similarly, U (q) is open. This
proves that Pm (R) is indeed Hausdorff.

3.3. Smooth manifolds as quotients of group actions


New examples of manifolds can be constructed by taking the quotient of a manifold
under a group action. We begin with some definitions.
Definition 3.3.1. Consider an action of a group G on a smooth manifold M :
a : G × M −→ M, a(g, x) = g · x.
(1) The action is called free if, for all x ∈ M , the equality g · x = x with
g ∈ G implies that g = e, where e denotes the identity element in G.
(2) The action is called smooth (resp. continuous) if, for each g ∈ G, the
action by g is smooth (resp. continuous):
ag : M −→ M, x 7→ g · x
(3) A continuous action of G on M is called proper if every two points x, y ∈
M have open neighborhoods Ux and Uy such that there are only finitely
many g ∈ G for which Ux ∩ g · Uy 6= ∅.
(4) The quotient space M/G is defined as the set of orbits

M/G := [x] : x ∈ M , [x] := G · x.
This set comes with a canonical projection
π : M −→ M/G, π(x) = [x].
(5) The quotient topology on M/G is defined by
U ⊂ M/G is open ⇐⇒ π −1 (U ) ⊂ M is open.
In other words, the quotient topology is the largest topology (in the sense
that it has the most open sets) for which the projection π is still continuous.
In the following result we construct a smooth manifold structure on the quotient
space of a smooth, free and proper action.
Theorem 3.3.2. Consider a smooth, free and proper action of group G on a smooth
m-dimensional manifold M . Then M/G has a unique m-dimensional differentiable
structure for which the canonical projection π : M → M/G is a local diffeomor-
phism.
In the statement we have used the following concept:
Definition 3.3.3. We say that a smooth map f : M → N is a local diffeomor-
phism if each p ∈ M has an open neighborhood U ⊂ M such that f (U ) ⊂ N is
open and f |U : U → f (U ) is a diffeomorphism.
LECTURE 3. 31

Proof of Theorem 3.3.2. We consider the quotient topology on M/G. First note
that π is an open map, i.e. π(V ) is open M/G for every open set V ⊂ M . This
holds because π −1 (π(V )) is the union of the open sets g · V , g ∈ G.
To see that M/G is second countable, let {Uk }k∈N be a countable basis for the
topology of M . Then the collection {π(Uk )}k∈N is a basis for the topology of M/G
(since π is an open map, these sets are indeed open). Let V ⊂ M/G be any open
set. Then π −1 (V ) = ∪k∈I Uk , for some subset I ⊂ N, and therefore V = ∪k∈I π(Uk ).
Next, we show that M/G is Hausdorff. Let [x], [y] ∈ M/G with [x] 6= [y]. Since
the action is proper, the points have neighborhoods Ux and Uy , respectively, such
that Ux and g · Uy intersect only for a finite number of elements g ∈ G; call these
elements g1 , . . . , gk . Since [x] 6= [y] it follows that gi y 6= x, for 1 ≤ i ≤ k; thus, since
M is Hausdorff, we can find open neighborhoods Ui of gi y and Ux0 ⊂ Ux of x such
−1
0 0
that Ui ∩ Ux = ∅. Let Uy := ∩i=1 gi Ui ∩ Uy . Then the open neighborhoods Ux0
k


and Uy0 of x and y, respectively, have the property that Ux0 ∩ gUy0 = ∅ for all g ∈ G.
This implies that gUx0 ∩ hUy0 = ∅ for all g, h ∈ G; or equivalently, π(Ux0 ) ∩ π(Uy0 ) = ∅.
Since π is an open map, the sets π(Ux0 ) and π(Uy0 ) are disjoint open neighborhoods
of [x], respectively of [y]. Thus M/G is Hausdorff.
Next, we construct charts on M/G. Consider [x] ∈ M/G. Since the action is
proper, there is an open neighborhood U of x such that g · U intersects U only for
a finite number of elements g ∈ G; call these elements g0 := e, g1 , . . . , gk . Since the
action is free, gi · x 6= x, for 1 ≤ i ≤ k; thus we can find open neighborhoods Ui of
gi · x and U 0 ⊂ U of x such that Ui ∩ U 0 = ∅. Let V := ∩ki=1 gi−1 Ui ∩ U 0 . This


open neighborhood of x satisfies V ∩ g · V = ∅, for all g 6= e. This implies that


π|V : V → π(V ) is injective. By shrinking V , we may assume that it is the domain
of a chart (V, ϕ) on M . We define the following chart on M/G:
ϕ ◦ (π|V )−1 : π(V ) −→ Rm .
Since π is open, it follows that π|V : V → π(V ) is a homeomorphism; hence
(π(V ), ϕ ◦ (π|V )−1 ) is indeed a chart on M/G. Consider the collection of all charts
obtained as above:
A := (π(V ), ϕ ◦ (π|V )−1 ) : (V, ϕ) is a smooth chart on M s.t. π|V is injective .


Above we have constructed a chart around each [x] ∈ M/G; thus the open sets in
A cover M/G. Finally, we need to prove that the transition maps are smooth. Let
(V, ϕ) and (W, ψ) be two smooth charts on M such that π|V and π|W are injective.
We need to show that the following map is a diffeomorphism:
ϕ ◦ (π|V )−1 ◦ (π|W ) ◦ ψ −1 : ψ(V ∩ π −1 (π(W ))) −→ ϕ(π −1 (π(V )) ∩ W ).
Let p ∈ π(V ) ∩ π(W ), and let x ∈ V and y ∈ W such that [x] = p = [y]. We show
that the transition map is smooth in a neighborhood of ψ(y). Since [x] = [y], there
exists g ∈ G such that x = g · y. Let U := g −1 · V ∩ W . On U , we claim that the
following equality holds:
(π|V )−1 ◦ π|U = ag |U : U −→ V.
By composing on both sides with π|V , the equality clearly holds; and so the claim
follows by injectivity of π|V . Hence, on ψ(U ), the transition map is a composition
of diffeomorphisms
ϕ ◦ ag ◦ ψ −1 : ψ(U ) −→ ϕ(g · U );
32 IOAN MĂRCUT
, , MANIFOLDS

therefore it is itself a diffeomorphism. This finishes the proof that A is a smooth


atlas, and so M/G is a smooth m-dimensional manifold.
Finally, let us show that π : M → M/G is a local diffeomorphism. As we have
seen above, around any point in M we can find a chart (V, ϕ) such that π|V is
injective. The local representation of π in the charts (V, ϕ) and (π(V ), ϕ ◦ (π|V )−1 )
is simply given by the identity map of the set ϕ(V ). This implies that π is smooth,
and π|V is a diffeomorphism; hence π is a local diffeomorphism.
Denote by T the quotient topology and by D the differentiable structure con-
structed above. Next, we show that D and T are unique such that π is a local
diffeomorphism. Consider a second topology T 0 and a second differentiable struc-
ture D0 on M/G such that π is a local diffeomorphism.
First we show that T = T 0 . Since π : M → M/G is continuous, it follows that
T ⊂ T . Since local diffeomorphisms are open maps, π(U ) ∈ T 0 for any open set
0

U in M . Thus, if V ∈ T then π −1 (V ) is open in M , hence V = π(π −1 (V )) ∈ T 0 .


This proves that T ⊂ T 0 ; hence the two topologies coincide.
Since π is a local diffeomorphism for D0 , every point has an open neighborhood
V such that π(V ) is open and π|V : V → π(V ) is a diffeomorphism. By shrinking
V , we may assume that it is the domain of a smooth chart (V, ϕ) on M . Hence,
ϕ ◦ (π|V )−1 : π(V ) → Rm is a diffeomorphism for D0 , and so (π(V ), ϕ ◦ (π|V )−1 )
is a smooth chart for D0 . The collection of such charts covers M/G, hence it is an
atlas A0 representing D0 . On the other hand, A0 ⊂ A, and so D = D0 . 

3.4. Manifolds of dimensions 1 and 2


Without providing proofs, in this section we discuss some classification results for
manifolds in dimension one and two.
Theorem ♣ 3.4.1. Any connected 1-dimensional manifold is diffeomorphic to
either R or S 1 .
Two-dimensional smooth manifolds, also called surfaces, are more compli-
cated. For example every open subset U of R2 is a surface; just think about the
cases U = R2 \Z2 , or U is the complement of the Cantor set.
Nevertheless, compact surfaces have a simple classification; we introduce these
basic examples in a less-rigorous fashion.
The genus-g surface, denoted by Σg , is a surface which looks like the one in
the picture, where g ≥ 0 stands for the number of holes:

Σg
For g = 0, we obtain the 2-sphere S 2 and for g = 1 the 2-torus T 2 :

Σ0 ∼
= S2 Σ1 ∼
= T2
LECTURE 3. 33

Place Σg in R3 such that it is symmetric with respect to the three coordinate


planes; here is the picture for Σ3 :

x
Σ3
In this position, the group Z2 = {0̂, 1̂} acts freely on Σg :
0̂ · (x, y, z) = (x, y, z), 1̂ · (x, y, z) = (−x, −y, −z).
The quotient space Σg /Z2 inherits therefore a smooth structure. For g = 0, S 2 /Z2
is the projective space (see Exercise 3.7), and for g = 1, T 2 /Z2 is the Klein bottle
(described also in Exercise 3.9).
We are ready to state the classification theorem for compact surfaces:
Theorem ♣ 3.4.2. Any compact connected 2-dimensional manifold is diffeomor-
phic to either Σg or Σg /Z2 , for some g ≥ 0.
The manifolds in the list Σg , Σg /Z2 , for g ≥ 0, are non-diffeomorphic. Let us
mention that, the surfaces Σg are the orientable ones, and the surfaces Σg /Z2 are
the non-orientable ones; the notion of orientable manifold will be introduced in a
later lecture.

3.5. Exercises
In the following exercise we construct a non-example of a manifold; a non-Hausdorff
second countable topological space with a smooth 1-dimensional atlas.
Exercise 3.1 (Two intervals glued along an open subinterval). Consider the set:
M = (−1, 1) ∪ [2, 3).
Let U = (−1, 1) and V = (−1, 0) ∪ [2, 3). On these subsets of M define the maps
ϕ : U → R, ϕ(t) = t, and

t, t ∈ (−1, 0)
ψ : V → R, ψ(t) =
t − 2, t ∈ [2, 3)
Show that A = {(U, ϕ), (V, ψ)} satisfies the conditions (1)-(4) of Proposition 3.1.1.
Show that the topology TA on M is non-Hausdorff.
Exercise 3.2. Let L be the set of all lines in R2 . For a, b, c ∈ R, with a 6= 0 or
b 6= 0, denote by l(a, b, c) the line given by the equation
ax + by + c = 0.
Let U ⊂ L be the set of lines l(a, b, c) with a 6= 0, and let V ⊂ L be the set of lines
l(a, b, c) with b 6= 0. Define the maps
ϕ : U → R2 , ϕ(l(a, b, c)) = (b/a, c/a), ψ : V → R2 , ψ(l(a, b, c)) = (a/b, c/b).
34 IOAN MĂRCUT
, , MANIFOLDS

Prove that L is a smooth 2-dimensional manifold with atlas A = {(U, ϕ), (V, ψ)}.
Exercise 3.3. Let S 1 denote the unit circle in C
p
S 1 := {z = x + iy : |z| = x2 + y 2 = 1} ⊂ C.
Let M be the collection of subsets of S 1 with 2 elements:
M := {{z, w} : z, w ∈ S 1 , z 6= w}.
(Pay attention to the fact that {z, w} = {w, z}!!)
(a) Consider the set
V := {(α, β) : 0 < α < β < 2π} ⊂ R2 ,
and for k ∈ {0, 1, 2} consider the map ψk : V → M given by
2πk 2πk
ψk (α, β) := {ei(α+ 3 ) , ei(β+ 3 ) }.
Prove that ψk is injective.
(b) Let Uk := ψk (V ), and let ϕk := ψk−1 : Uk → V . Prove that M has the structure
of a smooth 2-dimensional manifold, for which
A := {(Uk , ϕk ) : k ∈ {0, 1, 2}}
is a C ∞ -atlas. (Recall that eiα = cos(α) + i sin(α)).
Exercise 3.4. Show that the manifolds L and M constructed in Exercise 3.2 and
3.3, respectively, are diffeomorphic. Note that both are a Möbius band:

The following exercise generalizes the construction of the projective space.


Exercise 3.5. The Grassmannian Gr(k, n) is the set of all k-dimensional lin-
ear subspaces of Rn . We construct a smooth manifold structure on Gr(k, n) of
dimension k · (n − k).
Let {ei : 1 ≤ i ≤ n} denote the standard basis of Rn . Consider a subset I
of {1, 2, . . . , n} with k elements. Denote the elements of I and of its complement
increasingly as follows:
I = {i1 < i2 < . . . < ik }, {1, . . . , n}\I := {j1 < j2 < . . . < jn−k }.
For each such subset I we construct the following map
χI : Rk×(n−k) −→ Gr(k, n)
LECTURE 3. 35

D n−k
X E
χI (A) = Span eiu + au,v ejv : 1 ≤ u ≤ k , for A = (au,v )1≤u≤k,1≤v≤n−k ,
v=1
where Span denotes the linear span of a subset of Rn .
(a) Prove that χI is injective.
(b) Let UI := χI (Rk×(n−k) ) and let ϕI := χ−1
I : UI → R
k×(n−k)
.
Prove that Gr(k, n) is a smooth k × (n − k)-dimensional manifold with atlas:
A = {(UI , ϕI ) : I = {1 ≤ i1 < i2 < . . . < ik ≤ n}}.
(c) Prove that the manifolds Gr(k, n) and Gr(n − k, n) are diffeomorphic.
The following example is from [14].
Exercise 3.6. A continuous action of a group G on a manifold M is called wan-
dering if every point has a neighborhood U such that gU ∩ U = ∅, for all g 6= e.
Consider the action of (Z, +) on M := R2 \{0} given by
n · (x, y) = (2n x, 2−n y).
(a) Prove that this action is wandering, but not proper.
(b) Consider the quotient topology on M/Z. Prove that M/Z is second countable,
but not Hausdorff.
(c) Prove that M/Z has a smooth 2-dimensional atlas, i.e. a collection of charts A
satisfying conditions (1)-(4) from Proposition 3.1.1.
In the following exercise we construct the projective plane as a quotient of the
sphere.
Exercise 3.7. Consider the following action of Z2 = {0̂, 1̂} on S m :
0̂ · x = x, 1̂ · x = −x.
Prove that the action is smooth, proper and free. Prove that S m /Z2 is diffeomorphic
to Pm (R).
In the following exercise we construct the m-dimensional torus.
Exercise 3.8. Consider the action of the group (Zm , +) on Rm given by:
(k 1 , . . . , k m ) · (x1 , . . . , xm ) := (k 1 + x1 , . . . , k m + xm ).
Prove that the action is smooth, proper and free. The quotient manifold
T m := Rm /Zm = {x + Zm : x ∈ Rm }
is called the m-dimensional torus. Prove that T m is diffeomorphic to the product
of m-copies on S 1 (see Proposition 2.1.6 for the construction of the product).
By using actions on R2 , in the following exercise we construct several 2-dimensional
manifolds: the cylinder, the Möbius band, the torus and the Klein bottle.
Exercise 3.9. Consider the following two diffeomorphisms of R2 :
a, b : R2 −→ R2 , a(x, y) = (x, y + 1), b(x, y) = (x + 1/2, −y).
Let Diff(R2 ) be the group of diffeomorphisms of R2 . Let Ga , Gb , Ga,b2 and Ga,b ,
respectively, denote the subgroups of Diff(R2 ) generated by the sets {a}, {b}, {a, b2 }
and {a, b}, respectively.
36 IOAN MĂRCUT
, , MANIFOLDS

(a) Prove that the following relation holds: ba = a−1 b. Using this relation, show
that every element in Ga,b can be represented as an bm , for unique n, m ∈ Z.
Prove that the action of all four groups Ga , Gb , Ga,b2 and Ga,b on R2 is smooth,
free and proper.
(b) Prove that the quotient R2 /Ga is diffeomorphic to the cylinder S 1 × R.
(c) Prove that R2 /Ga,b2 is diffeomorphic to the 2-torus T 2 (from Exercise 3.8).
(d) Prove that R2 /Gb is diffeomorphic to the manifold from Exercise 3.2. (A man-
ifold diffeomorphic to R2 /Gb is called a Möbius band; see also Exercise 3.4.)
(e) Prove that Ga,b2 is an index two subgroup of Ga,b , i.e. Ga,b /Ga,b2 ∼= Z2 . Using
this, define a smooth, free and proper action of Z2 on T 2 such that T 2 /Z2 is
diffeomorphic to R2 /Ga,b . (A manifold diffeomorphic to R2 /Ga,b is called a
Klein bottle.)
Exercise 3.10. Let a : G × M → M be a proper smooth action, and assume that
M is compact. Prove that G is finite.
LECTURE 4

4.1. Bump functions


Recall the following topological notion:
Definition 4.1.1. Let X be a topological space, and let χ : X → R be a function.
The support of χ is defined as the closure of the set where χ is not zero:
supp(χ) := {x ∈ X : χ(x) 6= 0}.
Lemma 1.1.11 shows that around any point in Rn we can find a smooth bump
function, i.e. a function with support in a fixed neighborhood of the point, and
which is equal to the constant function 1 on a smaller neighborhood. This can be
easily generalized to manifolds.
Lemma 4.1.2. Let M be a smooth manifold, let p ∈ M and let U ⊂ M be an open
neighborhood of p. Then there exists a smooth function χ : M → [0, 1] such that
supp(χ) ⊂ U and χ|V = 1, where V is some neighborhood of p.
Proof. Let (W, ϕ) be a chart with p ∈ W ⊂ U , and let  > 0 be such that the
closed ball B  around ϕ(p) is included in ϕ(W ). By Lemma 1.1.11, there exists a
smooth function θ : Rm → [0, 1] such that θ|B/2 = 1 and supp(θ) ⊂ B  . Define
χ : M → [0, 1], χ|W = θ ◦ ϕ, χ|M \ϕ−1 (B  ) = 0.

Note that χ is well-defined: on the overlap of the sets W and M \ϕ−1 (B  ), we have:
θ ◦ ϕ|W \ϕ−1 (B  ) = 0.
Since χ is smooth on these two open sets that cover M , it follows that χ is smooth
on M . The conclusion holds with V := ϕ−1 (B/2 ). 

4.2. Partitions of unity


Partitions of unity play an important technical role in the theory of smooth mani-
folds, because they allow to build global objects by gluing together locally defined
objects. In this section we prove the existence of partitions of unity.
We start by introducing the necessary terminology.
Definition 4.2.1. Let X be a topological space.
37
38 IOAN MĂRCUT
, , MANIFOLDS

(1) An open cover of a subset Y ⊂ X is a family of open subsets of X


U = {Ui }i∈I , such that Y ⊂ ∪i∈I Ui ,
where I is an index set.
(2) A subcover of an open cover U of Y is a subfamily V ⊂ U which is still a
cover of Y .
(3) A subset Y ⊂ X is called compact, if any open cover of Y has a finite subcover.
(4) A refinement of an open cover U of Y is a second open cover V of Y such
that for every V ∈ V there exists U ∈ U such that V ⊂ U .
(5) A family S = {Si }i∈I of subsets of X is called locally finite if every x ∈ X
has an open neighborhood V which intersects only finitely many elements of S.
(6) A family of functions {ρi : X → R}i∈I is called locally finite if the family of
supports {supp(ρi )}i∈I is locally finite.
The following will be often used:
Lemma 4.2.2. Let M be a smooth manifold, and consider a locally finite family
of smooth functions on M , {ρi ∈ C ∞ (M )}i∈I . Since at every point x ∈ M there
are only a finite number of i ∈ I such that ρi (x) 6= 0, we can define:
X
ρ= ρi .
i∈I
The function ρ is smooth.
Proof. Let x ∈ M . Since the family is locally finite, x has an open neighborhood U
which intersects only the supports of a finite number of functions, say ρi1 , . . . , ρin .
Therefore, on U we have that:
X n
ρ|U = ρik |U .
k=1
The sum of a finite number of smooth functions is smooth, hence ρ|U is smooth. So,
every point has a neighborhood on which ρ is smooth; thus ρ is smooth on M . 
In order to define the function ρ in the lemma above, it suffices that at every
point only a finite number of functions are non-zero. However, this condition does
not insure smoothness of the resulting function; in Exercise 4.1, you are asked to
provide an example.
The main notion of this section is the following:
Definition 4.2.3. A partition of unity on a smooth manifold M is a locally
finite family of smooth functions
{ρi }i∈I , ρi ∈ C ∞ (M ), for i ∈ I,
such that, for all x ∈ M , 0 ≤ ρi (x) ≤ 1 and
X
ρi = 1.
i∈I

The partition of unity {ρi }i∈I is said to be subordinate to an open cover


U = {Ui }i∈I if supp(ρi ) ⊂ Ui for all i ∈ I.
The following is the main result about partitions of unity:
Theorem 4.2.4. Any open cover of a smooth manifold has a partition of unity
subordinate to it.
LECTURE 4. 39

4.3. Proof of Theorem 4.2.4


The proof is divided into three main steps. First, we show that it suffices to prove
the result for a refinement of the cover (Lemma 4.3.1); next, we show that open
covers of manifolds admit locally finite refinements whose elements have compact
closures (Lemma 4.3.4; this result implies that manifolds are paracompact, see
Remark 4.3.5); and finally, we prove Theorem 4.2.4 for open covers with these
properties.
Lemma 4.3.1. Let V be a refinement of the open cover U. If V has a partition of
unity subordinate to it, then so does U.
Proof. Denote V = {Vi }i∈I and U = {Uj }j∈J . The fact that V is a refinement
of U implies that there exists a function f : I → J (axiom of choice) such that
Vi ⊂ Uf (i) . Let {ρi }i∈I be a partition of unity subordinate to V. Since
Pthe family
of supports is locally finite, the following functions are smooth: χj := i∈f −1 (j) ρi
(if f −1 (j) = ∅, we set χj = 0). We claim that {χj }j∈J is a partition of unity
subordinate to U. First, note that:
[ [
(*) supp(χj ) ⊂ supp(ρi ) ⊂ Vi ⊂ Uj .
i∈f −1 (j) i∈f −1 (j)

Next, we show next that the family {χj }j∈J is locally finite. Let p ∈ M , and
consider O an open neighborhood of p such that O intersects only a finite number
of the sets supp(ρi ); denote these indexes by i1 , . . . , in ∈ I. If q ∈ supp(χj ), then
by (*), q ∈ supp(ρi ) for some i satisfying f (i) = j. Therefore, O intersects only the
sets supp(χj ), for j = f (i1 ), . . . , f (ik ). Finally, it is clear that:
X X X X
χj = ρi = ρi = 1.
j∈J j∈J i∈f −1 (j) i∈I


The following two results are used in the proof of Lemma 4.3.4.
Lemma 4.3.2. The topology of a manifold has a countable basis all of whose ele-
ments have compact closures.
Proof. Let B be a countable basis for M , and let Bc be the collection of sets in B
with compact closure. Clearly, Bc is countable. We prove that Bc is a basis. Let
O ⊂ M be an open set. For p ∈ O, let Kp ⊂ O be a compact neighborhood (e.g. let
Kp be the preimage by a chart around p of a small closed ball). Since B is a basis,
there exists Up ∈ B such that p ∈ Up ⊂ int(Kp ) ⊂ Op . In particular: Up ∈ Bc and
Up ⊂ O. Thus, we can write O = ∪p∈O Up , which shows that Bc is a basis. 
Lemma 4.3.3. Any manifold M has an open cover {Gk }∞
k=1 such that, for k ≥ 1,
Gk is compact and Gk ⊂ Gk+1 .
Proof. By Lemma 4.3.2, there exists a countable basis B = {Bn }n≥1 for the topol-
ogy of M such that B n is compact for all n ≥ 1. Let Om := ∪m n=1 Bn . Note that
Om ⊂ Om+1 , Om = ∪m n=1 B n is compact, and that ∪m≥1 Om = M . Therefore,
for any m ≥ 1 there is a smallest integer f (m) > m such that Om ⊂ Of (m) .
Let m1 := 1, and define inductively for k ≥ 2, mk := f (mk−1 ). The sequence
Gk := Omk satisfies all requirements. 
40 IOAN MĂRCUT
, , MANIFOLDS

Lemma 4.3.4. Any open cover U of a manifold M has a refinement V which is at


most countable, locally finite, and whose elements have compact closures.
Proof. Denote U = {Ui }i∈I . Consider a sequence of open sets {Gk }∞
k=1 as in
Lemma 4.3.3, and denote G−1 = G0 = ∅. For each k ≥ 1, the family
{Ui ∩ (Gk+1 \Gk−2 )}i∈I
is an open cover of the compact set Gk \Gk−1 , therefore it has a finite subcover Vk .
Let V := ∪k≥1 Vk . We prove that V has the required properties. Clearly, V is an
open cover because it covers Gk \Gk−1 , and it is a refinement of U because each
element is contained in some Ui . Since, for each k ≥ 1, Vk contains only a finite
number of open sets, V is at most countable. Every set in Vk has compact closure
because it is contained in the compact Gk+1 . Finally, let p ∈ M and l ≥ 1 be such
that p ∈ Gl . Then Gl intersects at most elements in Vk with k ≤ l + 1. This shows
that V is locally finite. 
Let us make a general remark about what we have proven so far.
Remark 4.3.5. A simple analysis of the proofs of the previous three Lemmas
(4.3.2,4.3.3,4.3.4) shows that we have used only the following topological properties
of a manifold:
• second countable;
• Hausdorff (in particular, compact sets are closed);
• locally compact (i.e. every point has a compact neighborhood).
A topological space for which every open cover has a locally finite refinement is
called paracompact. Thus we have proven that: Any topological space which is
second countable, Hausdorff and locally compact is paracompact.
Finally, we show that open covers which have the properties from Lemma 4.3.4
admit partitions of unity, and therefore we complete the proof of Theorem 4.2.4.
Lemma 4.3.6. Let V be a locally finite open cover of M whose elements have
compact closures. Then there exists a partition of unity subordinate to V.
Proof. Let V ∈ V. By Lemma 4.1.2, for each p ∈ V there is a smooth function
ψpV : M → [0, 1] such that ψpV (p) = 1 and supp(ψpV ) ⊂ V . Denote WpV := {x ∈ M :
ψpV (x) > 0}. Then the collection W := {WpV : p ∈ V ∈ V} is an open cover of M
which is a refinement of V. By Lemma 4.3.4 W has a locally finite refinement O.
So, for O ∈ O there is a smooth function ψ O : M → [0, 1] and V ∈ V such that
ψ O |O > 0 and supp(ψ O ) ⊂ V.
We show that an element V ∈ V intersects only finitely many elements in O. Since
O is locally finite, V can be covered by open sets which intersect only finitely many
elements in O; and by compactness of V , we may extract a finite subcover. Let U
be the union of the elements of this subcover. Then V ⊂ U and U intersects only
finitely many elements in O.
In particular, for any V ∈ V the set
OV := {O ∈ O : supp(ψ O ) ⊂ V }
is finite. Consider the family of smooth maps
X
ψ V := ψ O : M → [0, ∞), V ∈ V.
O∈O V
LECTURE 4. 41

Because it is subordinate to V, it follows that the family is locally finite; and by


the properties ofPO, at every point in M at least one of these functions is positive.
Therefore ψ := V ∈V ψ V > 0. The family of smooth functions ρV := ψV /ψ, V ∈ V
is a partition of unity subordinate to V. 

4.4. Corollaries of Theorem 4.2.4


For covers with two elements, Theorem 4.2.4 gives a smooth Urysohn’s lemma:
Corollary 4.4.1. Let A, B ⊂ M be two disjoint closed sets. There exists a smooth
function f : M → [0, 1] such that f |A = 0 and f |B = 1.
In particular, M is a normal topological space.
Proof. Consider the open cover {M \A, M \B}. If {f, g} is a partition of unity
subordinated to it, then f satisfies the condition from the statement. Finally, M
is normal, because A and B can be separated by the open sets {x : f (x) < 1/2}
and {x : f (x) > 1/2}, respectively. 

Theorem 4.2.4 can be used to extend smooth functions defined on closed em-
bedded submanifolds in Rn .
Corollary 4.4.2. Let M ⊂ Rn be an embedded manifold in Rn which is closed as
a subset. For every f ∈ C ∞ (M ) there exists fe ∈ C ∞ (Rn ) such that fe|M = f .
Proof. Recall that around every point in M there is a diffeomorphism ϕ : U → V ,
with V, U ⊂ Rn open, such that
ϕ(M ∩ U ) = (Rm × {0}) ∩ V.
We cover M by a family of such charts {(Ui , ϕi )}i∈I . Consider the following open
cover: {Rn \M } ∪ {Ui }i∈I , and let {ρ} ∪ {ρi }i∈I be a subordinate partition of unity.
For each i ∈ I, consider the following local extension of f :
fei := f ◦ (ϕ−1 n ∞
i ◦ prm ◦ ϕi ) ∈ C (Ui ),

and consider also the function:



ρi (x)fei , x ∈ Ui
gi ∈ C ∞ (M ), gi (x) :=
0, x∈/ supp(ρi )
P
The family {gi }i∈I is locally finite, and it is easy to see that the sum fe := i∈I gi
satisfies fe|M = f . 

Recall the following notion:


Definition 4.4.3. A continuous function f : X → Y between topological spaces X
and Y is called proper, if for any compact subset K ⊂ Y , we have that f −1 (K) ⊂ X
is compact.
Corollary 4.4.4. On any manifold M there exists a proper smooth function h :
M → [0, ∞).
Proof. Let {ρk }k≥1 be a partition of unity subordinated to a locally finite, at most
P cover {Vk }k≥1 , with V k compact (as in Lemma 4.3.4). Then the function
countable
h := k≥1 kρk is a positive proper function. 
42 IOAN MĂRCUT
, , MANIFOLDS

4.5. Exercises
Exercise 4.1. Construct a sequence of functions ρk ∈ C ∞ (R), k = 1, 2, . . ., such
P x ∈ R, ρk (x) 6= 0 only for a finite number of k’s, but such that the
that at every
function k≥1 ρk is not smooth.
Exercise 4.2. Let U = {Ui }i∈I and V = {Vj }j∈J be two open covers of M . Show
that W := {Ui ∩ Vj }(i,j)∈I×J is an open cover of M . If {ρi }i∈I is a partition of
unity subordinate to U and {χj }j∈J is a partition of unity subordinate to V, prove
that {ρi χj }(i,j)∈I×J is a partition of unity subordinate to W.
The locally finite cover found in Lemma 4.3.4 is at most countable. However,
this fact is automatic.
Exercise 4.3. Let U be a locally finite open cover of a manifold M .
(a) If M is compact, prove that U has a finite number of elements.
(b) In general, prove that U is at most countable.
Hint: choose a second cover V, as in Lemma 4.3.4, and show that every element
in V hits only a finite number of elements of U.
Exercise 4.4. Let {ρi }i∈I be a locally finite family of smooth functions on M .
(a) If M is compact, prove that only a finite number of functions are not identically
zero.
(b) In general, prove that the set of functions which are not identically zero is at
most countable.
LECTURE 5

Remark. For simplicity, from now on we will usually suppress the adjective smooth,
by implicitly assuming smoothness of the objects involved: by manifold we mean
smooth manifold, by chart we mean smooth chart, etc.

5.1. The tangent space


Intuitively, the tangent space of a smooth manifold M at a point p, is a vector
space denoted by Tp M of dimension m = dim(M ), which represents all velocities
of curves passing through p.
Example 5.1.1. For example, the tangent space to the 2-sphere S 2 at a point
p ∈ S 2 can be identified with the set of vectors perpendicular to p: Tp S 2 ∼ = {v ∈
R3 : v · p = 0}. Namely, any smooth curve γ : (−, ) → S 2 (with γ(0) = p)
satisfies γ(t) · γ(t) = 1, hence taking the derivative, γ 0 (t) · γ(t) = 0; in particular
v · p = 0, where v = γ 0 (0). On the other hand, given a local parameterization
f : O ⊂ R2 → S 2 (i.e. the inverse of a local chart) with f (q) = p, we can describe
the tangent space as the image of the differential of f :
Tp S 2 ∼
= (dq f )(R2 ).
Example 5.1.2. The tangent space of Rm at p ∈ Rm can be canonically identified
Tp Rm ∼ = Rm : any smooth curve γ : (−, ) → Rm with γ(0) = p induces the
tangent vector γ 0 (0) ∈ Rm . Conversely, any v ∈ Rm is the speed of some curve
through p; for example, γ(t) = p + tv. We think about Tp Rm as being “vectors
based at p”. Let us see what happens when we change coordinates on Rm , i.e.
consider a diffeomorphism χ : U − ∼
→ V , between open sets U, V ⊂ Rm . Then any
smooth curve γ : (−, ) → U , with γ(0) = p, can be transported to a smooth curve
χ ◦ γ : (−, ) → V , and their speeds are related by the chain rule:
(χ ◦ γ)0 (0) = (dp χ)γ 0 (0).
Thus, the differential of the change of coordinate matrix gives a natural identifica-
tion between the tangent spaces.
On a general manifold, a fixed chart induces a model for the tangent space (i.e.
just Rm ), and we use the differential of the change of coordinate maps to identify
these various models.
43
44 IOAN MĂRCUT
, , MANIFOLDS

Definition 5.1.3. Let M be a smooth manifold of dimension m. A tangent vector


at p ∈ M is a map v which associates to each chart (U, ϕ) around p an element
vϕ ∈ Rm satisfying the following rule
vψ = dχ(p) (ψ ◦ χ−1 )(vχ ),
for any two charts (O, χ) and (V, ψ) around p. The set of all tangent vectors at p
is called the tangent space of M at p and is denoted by Tp M .
Since the differential of the change of coordinates map is a linear map, it follows
that Tp M is a vector space with operations
(λv)ϕ = λvϕ , (v + w)ϕ = vϕ + wϕ , ∀v, w ∈ Tp M, λ ∈ R.
We have that:
Lemma 5.1.4. The tangent space of M at p is a vector space of dimension m =
dim(M ). Moreover, for any chart (U, ϕ) around p, the map
Tp M → Rm , v 7→ vϕ
is a linear isomorphism.
Proof. That the map is linear is clear. The kernel of the map is zero: if vϕ = 0
then, by linearity of the differential, vψ = dϕ(p) (ψ ◦ϕ−1 )(vϕ ) = 0 for any other chart
(V, ψ), thus v = 0. Thus the map is injective. To show surjectivity, let u ∈ Rm ,
and define v so that
vψ := dϕ(p) (ψ ◦ ϕ−1 )(u),
for a chart (V, ψ). That v is indeed a tangent vector follows from the chain rule:
for any two charts (V, ψ) and (O, χ), we have
vψ = dϕ(p) (ψ ◦ ϕ−1 )(u) = dϕ(p) (ψ ◦ χ−1 ◦ χ ◦ ϕ−1 )(u) =
= dχ(p) (ψ ◦ χ−1 ) ◦ dϕ(p) (χ ◦ ϕ−1 )(u) = dχ(p) (ψ ◦ χ−1 )(vχ ).
Clearly, vϕ = u; thus the assignment is indeed onto. 
m m
Using the standard chart (R , id) on R , the lemma above allows us to identify
(∗) Tp Rm → Rm , v 7→ vid .

5.1.5. The differential of a smooth map


Next, we extend the notion of the differential of a smooth map. We will use the
same notation as for maps between open sets in Euclidean space; the standard
notion is obtained by using the identification (∗) above.
Definition 5.1.6. Let f : M → N be a smooth map. The differential of f at
p ∈ M is the linear map
dp f : Tp M → Tf (p) N
dp f (v) ψ = dϕ(p) (ψ ◦ f ◦ ϕ−1 )(vϕ ),


where (U, ϕ) is a chart around p and (V, ψ) is a chart around f (p).


To show that the above assignment ψ 7→ (dp f (v))ψ yields indeed a tangent
vector, one needs to show that it transforms accordingly to the rule from Definition
5.1.3; however, this is a straightforward application of the chain rule, which we omit
(as in the proof of Lemma 5.1.4).
LECTURE 5. 45

With the canonical identification of the tangent space of Rm , note that the
differential of a chart (U, ϕ) around p ∈ M becomes
dp ϕ : Tp M → Tϕ(p) Rm , dp ϕ(v) = vϕ .
The chain rule also extends to the setting of smooth manifolds, and its proof
is again a straightforward application of the ‘classical’ chain rule:
Lemma 5.1.7. Let f : M → N and g : N → P be smooth maps between smooth
manifolds. The differentials satisfy the chain rule:
dp (g ◦ f ) = df (p) g ◦ dp f, ∀ p ∈ M.

5.2. Notations for tangent vectors


Let M be a smooth manifold and consider a chart (U, ϕ) around p ∈ M inducing
local coordinates
xiϕ : U → R, ϕ = (x1ϕ , . . . , xm
ϕ ).

Let e1 , . . . , em be the standard basis of Rm . The corresponding basis of Tp M will


be denoted by
∂ ∂ ∂
, , . . . , ∈ Tp M,
∂x1ϕ p ∂x2ϕ p ∂xm

ϕ p
 ∂   ∂ 
= d ϕ = ei , 1 ≤ i ≤ m.

p
∂xiϕ p ϕ ∂xiϕ p

To explain the notation, let f ∈ C ∞ (M ) be a smooth function on M . Then its


differential is given in this standard basis by:
 ∂ 
−1 i ∂f ◦ ϕ−1
dp f = d (f ◦ ϕ )(e ) = (ϕ(p));

ϕ(p)
∂xiϕ p ∂xi

in other words the differential of f on the standard basis is given by the partial
derivatives of the local expression f ◦ ϕ−1 of f .
More generally, let f : M → N be a smooth map, and consider a smooth chart
(V, ψ) around q = f (p), with local coordinates ψ = (yψ1 , . . . , yψn ).The matrix of dp f
in the bases
  ( )
∂ ∂
: 1≤i≤m and : 1≤j≤n
∂xiϕ p ∂yψj q
of Tp M and Tq N , respectively, is the Jacobian matrix of the local representation
ψ ◦ f ◦ ϕ−1 (x) = (f 1 (x), . . . , f n (x))
of f in these charts:
∂f 1 ∂f 1 ∂f 1
 
∂x1 ∂x2 ... ∂xm
  ∂f 2 ∂f 2 ∂f 2

dp f = 
 ∂x1 ∂x2 ... ∂xm  (ϕ(p));

 ... ... ... ... 
∂f n ∂f n ∂f n
∂x1 ∂x2 ... ∂xm
n
 ∂  X ∂f j ∂
dp f = (ϕ(p)) j .

i
∂xϕ p

∂x i
j=1 ∂yψ q
46 IOAN MĂRCUT
, , MANIFOLDS

5.3. The commutative algebra C ∞ (M )


The “cleanest” way to introduce the tangent space is by using the commutative
R-algebra of smooth functions:
(C ∞ (M ), +, ·), (f + g)(x) := f (x) + g(x), (f · g)(x) := f (x) · g(x).
Note that smooth maps induce algebra homomorphisms:
Proposition 5.3.1. Any smooth map ϕ : M → N induces a homomorphism of
algebras, called the pullback along ϕ:
ϕ∗ : C ∞ (N ) → C ∞ (M ), ϕ∗ (f ) = f ◦ ϕ.
The more surprising result is the converse of this Proposition, which implies
that the commutative algebra C ∞ (M ) encodes algebraically the entire information
about the manifold M :
Theorem ♣ 5.3.2. Any algebra homomorphism s : C ∞ (N ) → C ∞ (M ) is of the
form s = ϕ∗ for a unique smooth map ϕ : M → N .
In particular, this result implies that two manifolds are diffeomorphic if and
only if their commutative algebras of smooth functions are isomorphic. A proof of
Theorem 5.3.2 is outlined in the exercises at the end of this lecture.

5.4. Tangent vectors as derivations


Recall that on Rm the differential of a function f : U ⊂ Rm → Rn is defined using
directional derivatives:
d
dp f (v) = f (p + tv) t=0 .
dt
We introduce the analog terminology on manifolds:
Definition 5.4.1. Let M be a smooth manifold. The derivative along v ∈ Tp M
is the map
Dv : C ∞ (U ) → R, Dv (f ) := dp f (v).
The following holds:
Lemma 5.4.2. For all f, g ∈ C ∞ (M ), and v ∈ Tp M we have that:
Dv (f · g) = f (p)Dv (g) + Dv (f )g(p).
Proof. Using a chart, it suffices to prove this on open sets in Rm . Here it follows
from the usual Leibniz rule (α · β)0 = α · β 0 + α0 · β applied to the functions
α(t) = f (p + tv) and β(t) = g(p + tv) at t = 0. 

Definition 5.4.3. A derivation of C ∞ (M ) at p ∈ M is a linear map


D : C ∞ (M ) −→ R
satisfying the derivation rule:
D(f · g) = f (p)D(g) + D(f )g(p), ∀ f, g ∈ C ∞ (M )
The space of derivations of C ∞ (M ) at p will be denoted by Derp (C ∞ (M )).
LECTURE 5. 47

Note that the space of derivations is a vector space: for all λ, µ ∈ R and
D1 , D2 ∈ Derp (C ∞ (M )),
λD1 + µD2 : C ∞ (M ) → R, (λD1 + µD2 )(f ) := λD1 (f ) + µD2 (f )
is also derivation of C ∞ (M ) at p.
The following result shows that all derivations come from derivatives along
tangent vectors.
Theorem 5.4.4. Let M be a smooth manifold, and let p ∈ M . The map
Tp M → Derp (C ∞ (M )), v 7→ Dv
is a linear isomorphism.
To prove the theorem, we first show that derivations of C ∞ (M ) at p are local,
in the sense that they depend only on the behavior of the functions around p. The
following terminology will be therefore rather useful:
Definition 5.4.5. Let M be a manifold and p ∈ M . Two functions f, g ∈ C ∞ (M )
are said to have the same germ at p, if there is an open neighborhood W of p such
that f |W = g|W .
Lemma 5.4.6. Let f, g ∈ C ∞ (M ) have the same germ at p ∈ M , then for any
D ∈ Derp (C ∞ (M )), we have that D(f ) = D(g).
Proof. Let h := f − g; then h|W = 0, for some neighborhood W of p. We show
that D(h) = 0. Consider a bump function χ ∈ C ∞ (M ) such that χ(p) = 1 and
supp(χ) ⊂ W . Then, χh = 0. Therefore:
0 = D(0) = D(χh) = h(p)D(χ) + χ(p)D(h) = D(h). 


Lemma 5.4.7. Let U ⊂ M be an open neighborhood of p. Then the map


I : Derp (C ∞ (U )) → Derp (C ∞ (M )), I(D)(g) := D(g|U )
is a linear isomorphism.
Proof. Let χ ∈ C ∞ (M ) be a bump function such that χ|W = 1, on some neigh-
borhood W ⊂ U of p and such that supp(χ) ⊂ U . We define the map
E : C ∞ (U ) → C ∞ (M ), E(f )|U := χf, E(f )|M \supp(χ) := 0.
Note that f and E(f )|U = χf have the same germ at p. Consider the map
R : Derp (C ∞ (M )) → Derp (C ∞ (U )), R(D0 )(f ) := D0 (E(f )).
Since E(f g) and E(f )E(g) have the same germ at p, R(D0 ) is indeed a derivation:
R(D0 )(f g) =D0 (E(f g)) = D0 (E(f )E(g)) = E(f )(p)D0 (E(g))
+ E(g)(p)D0 (E(f )) = f (p)R(D0 )(g) + g(p)R(D0 )(f ).
Note that I ◦ R(D0 )(f ) = D0 (E(f )|U ) = D0 (f ), hence I ◦ R(D0 ) = D0 ; and similarly,
that R ◦ I(D)(g) = D(E(g|U )) = D(g), hence R ◦ I(D) = D. We obtain that I is
invertible with I −1 = R. 

For the proof, we will need the following version of Taylor’s theorem:
48 IOAN MĂRCUT
, , MANIFOLDS

Lemma 5.4.8. Let B ⊂ Rm be a convex open neighborhood of 0. For any f ∈


C ∞ (B) there exist smooth functions gi ∈ C ∞ (B), 1 ≤ i ≤ m, such that
m
X
f (x) = f (0) + gi (x)xi ,
i=1
∂f
and moreover, gi (0) = ∂xi (0).

Proof. Fix x ∈ B, and consider the function h : [0, 1] → R, h(t) = f (tx). Clearly,
h is smooth, therefore the fundamental theorem of calculus applies:
Z 1
h(1) − h(0) = h0 (t)dt.
0
This equation gives:
Z 1 Z 1Xm m
d ∂f X
f (x) − f (0) = f (tx1 , . . . , txm )dt = i
(tx)xi
dt = gi (x)xi ,
0 dt 0 i=1 ∂x i=1
R 1 ∂f ∞ ∂f
where gi (x) = 0 ∂xi (tx)dt. Clearly, gi ∈ C (B) and gi (0) = ∂xi (0). 
Proof of Theorem 5.4.4. Consider a chart (U, ϕ) on M around p, with coordi-
nates ϕ = (x1ϕ , . . . , xm
ϕ ) such that ϕ(p) = 0 and ϕ(U ) = B is convex.
By Lemma 5.4.7, it suffices to show that the map
Tp M → Derp (C ∞ (U )), v 7→ Dv
is an isomorphism.
Let v ∈ Tp M and denote vϕ = (v 1 , . . . , v m ). By the formulas in Section 5.2,
m
X ∂xi
Dv (xiϕ ) = dp xiϕ (v) = vj (0) = v i ;
j=1
∂xj
thus, the map is injective.
Let D ∈ Derp (C ∞ (U )). We show that D vanishes on constant functions. Since
D(1) = D(1 · 1) = 1 · D(1) + D(1) · 1 = 2 · D(1),
we obtain that D(1) = 0. By linearity, D(c) = D(c · 1) = c · D(1) = 0, for all c ∈ R.
Let v ∈ Tp M be the vector with vϕ = (v 1 , . . . , v m ) ∈ Rm , where v i := D(xiϕ ).
We show that D = Dv . Let f ∈ C ∞ (U ). Applying Lemma 5.4.8, we find functions
gi ∈ C ∞ (U ), for 1 ≤ i ≤ m, such that
Xm  ∂ 
f = f (p) + xiϕ gi , gi (p) = dp f .

∂xi p
i=1 ϕ

Applying D, we obtain:
m
X
D(f ) = D(f (p)) + D(xiϕ gi ) =
i=1
m
X
0 · D(gi ) + D(xiϕ ) · gi (p) =

=0+
i=1
m
X  ∂ 
= vi dp f = dp f (v) = Dv (f ).

i=1
∂xiϕ p
Thus, D = Dv . 
LECTURE 5. 49

5.5. Exercises
Exercise 5.1. Let M ⊂ Rn be a manifold embedded in Rn , as in Definition 2.4.1,
and let p ∈ M . Let Vp ⊂ Rn consist of all vectors γ 0 (0) ∈ Rn , where γ : (−, ) → M
is a smooth curve such that γ(0) = p. Show that Vp is a linear subspace of Rn , and
that there is a linear isomorphism Vp − ∼→ Tp M , such that γ 0 (0) ∈ Vp corresponds
d
to d0 γ( dt |t=0 ) ∈ Tp M .
The following exercise gives an alternative definition of the tangent space:
Exercise 5.2. Let M be a manifold and let p ∈ M . Denote by Ip ⊂ C ∞ (M ) the
ideal of functions vanishing at p:
Ip = {f ∈ C ∞ (M ) : f (p) = 0}.
Denote by Ip2 the square of this ideal, i.e. Ip2 consists of finite sums of the form:
k
X
f= gj hj , gj , hj ∈ Ip , 1 ≤ j ≤ k.
j=1

(a) Prove that Ip /Ip2 has dimension m = dim(M ).


Hint: use Lemma 5.4.8 and Lemma 4.1.2.
(b) Show that dp f (v) = 0 for all f ∈ Ip2 and v ∈ Tp M .
(c) Prove that the following map is a linear isomorphism:
E : Tp M → (Ip /Ip2 )∗ , E(v)(f + Ip2 ) := dp f (v), f ∈ Ip .
We introduce some terminology for the following exercises.
Definition 5.5.1. An algebra over R is a real vector space A endowed with an
associative, bilinear multiplication · : A × A → A, and with a unit 1A ∈ A; i.e.
(f · g) · h = f · (g · h), 1A · f = f = f · 1A ,
(λf + µg) · h = λ(f · h) + µ(g · h), f · (λg + µh) = λ(f · g) + µ(f · h),
for all f, g, h ∈ A and all λ, µ ∈ R. The algebra A is called commutative if
f · g = g · f for all f, g ∈ A.
A homomorphism between the R-algebras A and B is a linear map s : A → B,
which preserves multiplication and units:
s(f · g) = s(f ) · s(g) and s(1A ) = 1B .
Definition 5.5.2. Let A and R be commutative algebras over R, and let s : A → R
be an algebra homomorphism. A derivation at s of A is a linear map
D : A −→ R s.t. D(u · v) = s(u)D(v) + D(u)s(v), for all u, v ∈ A.
We denote the space of all derivations at s by:
Ders (A, R).
Definition 5.5.3. Let A be a commutative algebra over R. A character of A is
an algebra homomorphism χ : A → R.
Definition 5.5.4. For a commutative algebra R, let R[X] denote the polynomial
algebra with coefficients in R, i.e. each element in p ∈ R[X] can be written as a
finite sum:
p = u0 + u1 X + . . . + uk X k ,
50 IOAN MĂRCUT
, , MANIFOLDS

for unique u0 , . . . , uk ∈ R; the algebraic operations are defined as usually for poly-
nomials. Consider the following algebra:
R[] := R[X]/(X 2 ),
where (X 2 ) ⊂ R[X] is the ideal generated by X 2 , i.e. each element p ∈ R[] can be
uniquely written as
p = u0 + u1 ,
where  := X + (X 2 ) ∈ R[X]/(X 2 ); and the multiplication is such that 2 = 0:
(u0 + u1 ) · (v0 + v1 ) = u0 v0 + (u0 v1 + u1 v0 ).
Exercise 5.3. (a) Let s : A → R be an algebra homomorphism, and let D ∈
Ders (A, R). Prove that the map
s + D : A −→ R[], (s + D)(u) := s(u) + D(u)
is an algebra homomorphism.
(b) Prove that any algebra homomorphism σ : A → R[] is of the form described
at (a).
Exercise 5.4. Let s : A → R be a homomorphism of algebras.
(a) Prove that a homomorphism of algebras t : B → A induces a linear map
between the derivation spaces, defined as follows:
t∗s : Ders (A, R) −→ Ders◦t (B, R), t∗s (D) := D ◦ t.
(b) If t is surjective, prove that t∗s is injective.
(c) Prove that the following “chain rule” holds: if t : B → A and r : C → B are
algebra homomorphisms, then
(t ◦ r)∗s = rs◦t

◦ t∗s : Ders (A, R) −→ Ders◦t◦r (C, R).
Exercise 5.5. Let X be a topological space, and let C(X) be the algebra of
continuous maps X → R. Consider a subalgebra A ⊂ C(X), with unit 1A the
constant map 1A (p) = 1, and which satisfies the condition:
(*) (f ∈ A, f (x) 6= 0, ∀ x ∈ X) =⇒ 1/f ∈ A.
We denote by A∨ the set of characters. For χ ∈ A∨ , let Iχ denote the kernel of χ:
Iχ := {f ∈ A : χ(f ) = 0}.
(a) Let p ∈ X. Prove that the following map gives a character of A:
χp : A −→ R, χp (f ) := f (p).
We denote Ip := Iχp .
(b) For χ ∈ A∨ , and p ∈ X, prove that Ip ⊂ Iχ , implies χ = χp .
(c) Prove that if f ∈ Iχ , then there exists p ∈ X such that f (p) = 0.
(d) Let χ ∈ A∨ be a general character, and assume that χ 6= χp for every p ∈ X.
Prove that for any point p ∈ X there exists f ∈ Iχ such that f (p) > 0.
(e) Prove that if X is compact, then every χ ∈ A∨ is of the form χ = χp for some
p ∈ X.
(f) Assume that there exists f ∈ A such that f −1 (λ) ⊂ X is compact for any
λ ∈ R. Show that every χ ∈ A∨ is of the form χ = χp for some p ∈ X.
LECTURE 5. 51

Exercise 5.6. Prove that for any algebra homomorphism χ : C ∞ (M ) → R (i.e.


any character of C ∞ (M )) there exists a unique point p ∈ M such that:
χ(f ) = f (p), for all f ∈ C ∞ (M ).
Hint: use Exercise 5.5 and Corollary 4.4.4.
Exercise 5.7. Let M be a manifold. Show that there is a one-to-one correspon-
dence between tangent vectors on M and algebra homomorphisms C ∞ (M ) → R[].
Hint: use Exercises 5.3 and 5.6.
The following exercise proves Theorem 5.3.2.
Exercise 5.8. Let s : C ∞ (N ) → C ∞ (M ) be an algebra homomorphism.
(a) Prove that there exists a unique function ϕ : M → N such that s(f ) = f ◦ ϕ
for all f ∈ C ∞ (N ). Hint: use Exercise 5.6.
(b) Prove that for any smooth function f ∈ C ∞ (N ), the set
Uf := {p ∈ N : f (p) 6= 0}
is open, and that these sets forms a basis for the topology of N .
Hint: use Lemma 4.1.2.
(c) Prove that ϕ is continuous.
(d) Prove that ϕ is smooth.
LECTURE 6

6.1. The inverse function theorem


The inverse function theorem gives a very simple condition to check that a map is
locally a diffeomorphism, namely, it suffices that its differential be invertible. For
a proof of this classical result, see for example [11]. This can be directly extended
to smooth manifolds:
Theorem 6.1.1 (The inverse function theorem). Let f : M → N be a smooth map
and let p ∈ M . If dp f : Tp M → Tf (p) N is a linear isomorphism, then there exists
an open neighborhood O ⊂ M of p, such that f (O) is open and f restricts to a
diffeomorphism:
f |O : O −∼→ f (O).
Proof. Let (U, ϕ) be a chart around p and (V, ψ) be a chart around f (p). By
shrinking U , we may assume that f (U ) ⊂ V . By the chain rule, we have that
−1
dϕ(p) (ψ ◦ f ◦ ϕ−1 ) = df (p) ψ ◦ dp f ◦ (dp ϕ) .
Since the charts are diffeomorphisms, their differentials are linear isomorphisms;
therefore, the right hand side is the composition of linear isomorphisms. We obtain
that also dϕ(p) (ψ ◦f ◦ϕ−1 ) is a linear isomorphism. The map ψ ◦f ◦ϕ−1 is a smooth
map between open sets in Rm , and its Jacobian matrix at ϕ(p) coincides with its
differential, thus it is invertible. By the standard inverse function theorem (see e.g.
[11]), there exists an open neighborhood W of ϕ(p) such that ψ ◦ f ◦ ϕ−1 (W ) is
open and ψ ◦ f ◦ ϕ−1 |W : W − ∼→ ψ ◦ f ◦ ϕ−1 (W ) is a diffeomorphism. Since the
charts are diffeomorphisms, this implies that conclusion with O := ϕ−1 (W ). 

6.2. Immersions and submersions


The following two classes of maps play an important role:
Definition 6.2.1. Let f : M → N be a smooth map, and let p ∈ M .
(1) f is called an immersion at p if dp f : Tp M → Tf (p) N is injective. If this
holds at all points in M , then f is called an immersion.
(2) f is called a submersion at p if dp f : Tp M → Tf (p) N is surjective. If this
holds at all points in M , then f is called a submersion.
53
54 IOAN MĂRCUT
, , MANIFOLDS

The most basic example of an immersion is the linear inclusion:

ikl : Rk −→ Rl , ikl (x1 , . . . , xk ) := (x1 , . . . , xk , 0, . . . , 0), for k ≤ l,

and the most basic example of a submersion is the linear projection

prlk : Rl −→ Rk , prlk (x1 , . . . , xl ) := (x1 , . . . , xk ), for k ≤ l.

The following two theorems say that locally, up to a change of coordinates,


these are the only examples:

Theorem 6.2.2 (The local immersion theorem). Let f : M → N be a smooth map.


If f is an immersion at p ∈ M , then there are charts (U, ϕ) and (V, ψ) around p
and f (p), respectively, such that f (U ) ⊂ V and, in these coordinates, f becomes
the map imn ; i.e.

ψ ◦ f ◦ ϕ−1 = im n
n |ϕ(U ) : ϕ(U ) −→ R ,

ψ ◦ f ◦ ϕ−1 (x1 , . . . , xm ) = (x1 , . . . , xm , 0, . . . , 0).

Theorem 6.2.3 (The local submersion theorem). Let f : M → N be a smooth


map. If f is a submersion at p ∈ M , then there are charts (U, ϕ) and (V, ψ) around
p and f (p), respectively, such that f (U ) ⊂ V and, in these coordinates, f becomes
the map prmn ; i.e.

ψ ◦ f ◦ ϕ−1 = prm n
n |ϕ(U ) : ϕ(U ) −→ R ,

ψ ◦ f ◦ ϕ−1 (x1 , . . . , xm ) = (x1 , . . . , xn ).

The proofs of these two results are completely analogous, therefore, we only
give the details for the second result; the proof of the local immersion theorem is
left as an exercise.

Proof of the local submersion theorem. Let (U e , ϕ)


e and (V, ψ) be two charts
around p and f (p), respectively, such that f (U ) ⊂ V . Denote the local expression
e
of f by h := ψ ◦ f ◦ ϕ e−1 . We are going to “complete h to a diffeomorphism”.
By applying Exercise 6.1 to the linear map A = dϕ(p) e h (which, by the chain rule,
is surjective), we obtain a linear map B : Rm → Rm−n such that (dϕ(p) e h, B) :
Rm → Rm is a linear isomorphism. This map is precisely the differential at ϕ(p) e
of the map (h, B) : ϕ( eUe ) → Rm . Therefore, by the inverse function theorem,
there is a neighborhood O ⊂ ϕ( eU e ) of ϕ(p)
e such that W := (h, B)(O) is open and

(h, B) : O −→ W is a diffeomorphism; denote its inverse by g := (h, B)−1 : W − ∼
→ O.
Then we have that (h(g(x)), B(g(x))) = x for all x ∈ W , hence h ◦ g = prm n |W .
Equivalently, ψ ◦ f ◦ ϕ−1 = prmn |W , where

ϕ := g −1 ◦ ϕ e−1 (O);
e : U −→ Rm , U := ϕ
e = (h, B) ◦ ϕ

hence, the charts (U, ϕ) and (V, ψ) satisfy the conditions from the statement. 

Remark 6.2.4. In the local submersion theorem, the chart on the codomain (V, ψ)
can be chosen arbitrary; this follows from the proof. Similarly, in the local immer-
sion theorem the chart (U, ϕ) on the domain can be chosen arbitrary.
LECTURE 6. 55

6.3. Embedded submanifolds


The following generalizes the notion of embedded manifolds in Rn from Section 2.4:
Definition 6.3.1. A subset S ⊂ M of a smooth m-dimensional manifold M is
called a k-dimensional embedded submanifold if for every point p ∈ S there
exists a chart (U, ϕ) around p, such that
ϕ(S ∩ U ) = Rk × {0} ∩ ϕ(U ) ⊂ Rm .

(*)
A chart (U, ϕ) with these property is called a chart adapted to S. The codimen-
sion of S is the number m − k = dim(M ) − dim(S).
Next, we show that embedded submanifolds are indeed smooth manifolds:
Theorem 6.3.2. Let S ⊂ M be a k-dimensional embedded submanifold. Then S,
endowed with the subset topology, has a k-dimensional differentiable structure, with
atlas obtained by restricting adapted charts:
AS = {(U ∩ S, prm
k ◦ ϕ|U ∩S ) : (U, ϕ) is a chart adapted to S}.
Moreover, the inclusion map i : S ,→ M , i(x) = x is a smooth immersion.
Proof. The proof that S, endowed with the subset topology, has a k-dimensional
differentiable structure for which AS is a smooth atlas can be proven exactly as in
the case of embedded manifolds in Rn (see Proposition 2.4.3.
Next, since S has the induced topology, it follows that i is continuous. Let
(U, ϕ) be an adapted chart around p ∈ S. Then, note that in the pair of charts
(U ∩ S, prmk ◦ ϕ|U ∩S ) and (U, ϕ) the local expression of i is the canonical immersion
ikm : Rk → Rm . This proves that i is a smooth immersion. 
In Definition 6.3.1 above, denote that the coordinates by ϕ = (x1ϕ , . . . , xm
ϕ ).
Then condition (*) means that S is locally described by the l = m − k equations:
S ∩ U = {q ∈ U : xk+1 k+2 m
ϕ (q) = 0, xϕ (q) = 0, . . . , xϕ (q) = 0}.
The converse also holds, and turns out to be a very useful criterion: any subset
which is locally described as the zeroes of by l independent equations is an embedded
submanifold of codimension l (here, independent means that the equations form a
submersion):
Proposition 6.3.3. A subset S ⊂ M is an embedded submanifold of codimension
l if and only if for every p ∈ S there exists an open set p ∈ U ⊂ M and smooth
map h : U → Rl which is a submersion at p, such that S ∩ U = h−1 (0).
Proof. Assume first that S is an embedded submanifold of codimension l. Let
p ∈ S and let (U, ϕ) be a chart adapted to S with p ∈ U . Define
h : U −→ Rl , h := prm
l ◦ ϕ,
where prm l : R
m
→ Rl denotes the projection onto the last l coordinates. Then h
is a submersion and S ∩ U = h−1 (0).
Conversely, assume that S ⊂ M is a subset such that, around any p ∈ S, we can
find an open set U and a smooth map h : U → Rl which is a submersion at p and
satisfies U ∩ S = h−1 (0). We apply the local submersion theorem to h. By Remark
6.2.4, we do not need to change the coordinates on Rl . Thus, after shrinking U , we
may assume that there is a chart (U, ϕ) on M such that
h ◦ ϕ−1 (x1 , . . . , xm ) = (xk+1 , . . . , xm ), where k = m − l;
56 IOAN MĂRCUT
, , MANIFOLDS

or equivalently, in the usual notation for coordinates ϕ = (x1ϕ , . . . , xm


ϕ ), we have:

h(q) = (xk+1 m
ϕ (q), . . . , xϕ (q)), for q ∈ U.
Thus, for q ∈ U : h(q) = 0 iff ϕ(q) ∈ Rk × {0}, and since U ∩ S = h−1 (0), we obtain:
ϕ(S ∩ U ) = (Rk × {0}) ∩ ϕ(U ). This concludes the proof. 

Remark 6.3.4. Consider an embedded submanifold S ⊂ M . By Theorem 6.3.2


the inclusion map i : S ,→ M is a smooth immersion. Therefore if p ∈ S the map
dp i : Tp S → Tp M is injective, and so it sends Tp S isomorphically onto a linear
subspace of Tp M . We will identify these spaces:
Tp S = dp i(Tp S) ⊂ Tp M.
Assume now that h : U → Rl is a local submersion around p such that S = h−1 (0).
We claim that
(*) Tp S = ker(dp h) = {v ∈ Tp M : dp h(v) = 0}.
Since h ◦ i(U ∩ S) = 0, it follows that dp h ◦ dp i(Tp S) = 0. Hence Tp S ⊂ ker(dp h).
The other inclusion follows because dp h is surjective, and so these vector spaces
have the same dimension:
dim ker(dp h) = dim Tp M − dim im(dp h) = dim M − codim S = dim S = dim Tp S.
Let us introduce some standard terminology:
Definition 6.3.5. A regular value of a smooth map f : M → N is a point q ∈ N
such that f is a submersion at every point in f −1 (q). A point q ∈ N which is not
a regular value is called a critical value for f .
Corollary 6.3.6 (Preimage Theorem). Let f : M → N be a smooth map. If q ∈ N
is a regular value f , then f −1 (q) is an embedded submanifold of M with
codim(f −1 (q)) = dim(N ), Tp (f −1 (q)) = ker(dp f ), for p ∈ f −1 (q).
Proof. Take a chart (U, ϕ) around p with ϕ(q) = 0. Then h := ϕ ◦ f : V → Rn is
a submersion, and f −1 (q) = h−1 (0). Proposition 6.3.3 implies the result. 

Let us also mention a deep result of mathematical analysis, which states that
most points are regular values (note that points that are not in the image are always
regular values). Without introducing all notions involved, we state this result below
(for details, the reader is encouraged to consult [8] or [4]).
Theorem ♣ 6.3.7 (Sard’s Theorem). The set of critical values of a smooth map
f : M → N has Lebesgue measure zero. In particular, the set of regular values of f
is dense in N .
Remark 6.3.8. If dimM < dimN , then every point in M is a critical point of
f . In this situation, Sard’s Theorem says that f (M ) has measure zero in N ; and
therefore N \f (M ) is dense in N . In particular, one cannot have a surjective map
smooth f : M → N if dimM < dimN .
For this property, smoothness is essential. In the continuous setting, there are
the co-called space-filling curves (e.g. Peano’s curve [21]). For example, one can
prove that any connected, nonempty manifold is the image of a continuous curve!!
LECTURE 6. 57

6.4. Preimages of submanifolds


The following extends the notion of a regular value to submanifolds.
Definition 6.4.1. A smooth map f : M → N is said to be transverse to an
embedded submanifold Q ⊂ N , if
Tf (p) N = Tf (p) Q + dp f (Tp M ), ∀ p ∈ f −1 (Q).
Remark 6.4.2. Note that the sum in the decomposition is not assumed to be
a direct sum, i.e. the vector spaces Tf (p) Q and dp f (Tp M ) may have a nontrivial
intersection. For example, a submersion is transverse to any embedded submanifold
of the codomain.
Theorem 6.4.3. Let f : M → N be a smooth map that is transverse to the
embedded submanifold Q ⊂ N . Then P := f −1 (Q) is an embedded submanifold of
M with codim P = codim Q and
−1
Tp P = (dp f ) (Tf (p) f −1 (Q)), ∀p ∈ P.
Before giving the proof, let us consider an example.
Example 6.4.4. Consider the orthogonal projection from the 2-sphere to R2 ,
f : S 2 −→ R2 , f (x, y, z) := (x, y).
Note that f is a submersion outside of the equatorial circle, which we denote by Z.
Consider two embedded submanifold Q1 and Q2 in R2 which are indicated on the
picture below:

f −1 (Q1 )
2
S

f −1 (Q2 )
p2
p

p0 Z p1

f (p2 )
f (p) Q1 Q2

R2
f (p0 ) f (Z) f (p1 )

Note that f is not transverse to Q1 , because at the point of intersection f (p0 ) of


f (Z) and Q1 , we have that Tf (p0 ) Q1 = dp0 f (Tp0 S 2 ). Note also that f −1 (Q1 ) is not
58 IOAN MĂRCUT
, , MANIFOLDS

an embedded submanifold of S 2 , because it has a singularity which looks like the


intersection of two lines.
On the other hand, f is transverse to Q2 ; outside of Z this is clear because f
is a submersion, and at the two points p1 , p2 ∈ Z ∩ f −1 (Q2 ) note that
dpi f (Tpi S 2 ) + Tf (pi ) (Q2 ) = Tf (pi ) R2 , i = 1, 2.
As predicted by the Theorem 6.4.3, f −1 (Q2 ) is an embedded submanifold of S 2 ; in
fact it is diffeomorphic to a circle.
Proof of Theorem 6.4.3. We check that P satisfies the condition in Proposition
6.3.3. Consider p ∈ P , and denote q := f (p) ∈ Q. Since Q is an embedded
submanifold there exists an open neighborhood U of q and a smooth map h : U →
Rl , where l = codimQ, such that h is a submersion at q and h−1 (0) = U ∩ Q. On
the open neighborhood f −1 (U ) of p, consider the smooth map h ◦ f : f −1 (U ) → Rl .
We show that this map satisfies the conditions from Proposition 6.3.3. Note that
(h ◦ f )−1 (0) = f −1 (h−1 (0)) = f −1 (U ∩ Q) = f −1 (U ) ∩ f −1 (Q) = f −1 (U ) ∩ P.
By Remark 6.3.4, Tq Q = ker dq h; hence dq h(Tq Q) = 0. Using this, that h is a
submersion at q, that f is transverse to Q, and the chain rule, we obtain:
T0 Rl = dq h (Tq N ) = dq h (Tq Q + dp f (Tp M )) = dq h ◦ dp f (Tp M ) = dp (h ◦ f )(Tp M ).
Thus h◦f is a submersion at p. So Proposition 6.3.3 implies that P is an embedded
manifold of codim P = l = codim Q. Finally, again by (*), we have that
Tp P = ker dp (h ◦ f ) = ker (dq h ◦ dp f ) = (dp f )−1 (ker dq h) = (dp f −1 )(Tq Q).

Remark 6.4.5. Note that the empty set ∅ is a smooth manifold of any dimension
d ≥ 0: it has a (unique) topology; it is locally homeomorphic to the empty set in Rd
(which is open); and the only transition map is the id∅ , which is smooth, when we
regard ∅ ⊂ Rd (prove this “by contradiction”!). In the setting of Theorem 6.4.3, it
could happen that f (M ) ∩ Q = ∅; in this case f is transverse to Q and f −1 (Q) = ∅.
Theorem 6.4.3 can be used to intersect submanifolds. First, we need the fol-
lowing definition:
Definition 6.4.6. Two embedded submanifold R, S ⊂ M are said to intersect
transversely, if for every p ∈ R ∩ S we have that
Tp M = Tp R + Tp S.
This condition is equivalent to the inclusion i : R ,→ M being transverse to S.
Note also that
i−1 (S) = R ∩ S, (dp i)−1 (Tp R) = Tp R ∩ Tp S, p ∈ R ∩ S.
Hence, Theorem 6.4.3 implies the following:
Corollary 6.4.7. The intersection R ∩ S of two embedded submanifolds R, S ⊂ M
which intersect transversally is an embedded submanifold with
codim R ∩ S = codim R + codim S;
moreover, for each p ∈ R ∩ S we have that
Tp (R ∩ S) = Tp R ∩ Tp S.
LECTURE 6. 59

6.5. Exercises
Exercise 6.1. Let k ≤ n, and let A : Rn → Rk be a surjective linear map, and let
C : Rk → Rn be an injective linear map.
(a) Prove that there exists a linear map B : Rn → Rn−k such that the map
(A, B) : Rn → Rk × Rn−k = ∼ Rn , (A, B)(x) := (A(x), B(x)),
is a linear isomorphism.
(b) Prove that there exists a linear isomorphism G : Rn → Rn such that
A ◦ G = prnk .
(c) Prove that there exists a linear map D : Rn−k → Rn such that the map
C + D : Rk × Rn−k ∼ = Rn → Rn , (C + D)(y, z) := C(y) + D(z),
is a linear isomorphism.
(d) Prove that there exists a linear isomorphism H : Rn → Rn such that
H ◦ C = ikn .
Exercise 6.2. Let a ∈ R. Show that the following map is an immersion:
fa : R −→ S 1 × S 1 , fa (t) := (eit , eita ).

Make a sketch of fa (R) for a = 0, 1/2, 1, 2, 3. For which a ∈ R is fa injective?
Exercise 6.3. Let f : P2 (R) → R3 be the map defined by
1
f ([x, y, z]) = 2 (yz, xz, xy).
x + y2 + z2
Show that f is smooth and show that it only fails to be an immersion at 6 points.
Make a sketch of the image of f .
Being an immersion/submersion is an open condition:
Exercise 6.4. If a smooth map f : M → N is an immersion (resp. a submersion)
at p ∈ M prove that there is an open neighborhood V ⊂ M of p such that f |V is
an immersion (resp. a submersion).
Do not use the local immersion/submersion theorem!
Exercise 6.5. Prove the local immersion theorem.
Exercise 6.6 (Alternative definition of immersions and submersions). Let f : M →
N be a smooth map, and let p ∈ M . Prove the following:
(a) f is an immersion at p iff it has a local left inverse, i.e. there exist open neigh-
borhoods U ⊂ M of p and V ⊂ N of f (p) and a smooth map σ : V → U such
that f (U ) ⊂ V and
σ(f (x)) = x, for all x ∈ U.
(b) f is a submersion at p iff it has a local right inverse, i.e. there exist open
neighborhoods U ⊂ M of p and V ⊂ N of f (p) and a smooth map j : V → U
such that j(f (p)) = p and
f (j(y)) = y, for all y ∈ V.
Exercise 6.7. Let f : M → N be a submersion. Prove that f is an open map (i.e.
for every open set U ⊂ M , f (U ) is open in N ).
60 IOAN MĂRCUT
, , MANIFOLDS

Exercise 6.8. Let f : M → N be a smooth map, and let K ⊂ M be a compact


subset. If f |K : K → N is injective, and dp f : Tp M → Tf (p) N is a linear isomor-
phism for every p ∈ K, prove that there is an open set U such that K ⊂ U , f (U )
is open in N and f : U → f (U ) is a diffeomorphism.
Exercise 6.9. Use the Preimage Theorem (Corollary 6.3.6) to prove that S n is an
embedded submanifold of Rn+1 .
Exercise 6.10. Let Sb1 ⊂ R3 be the circle in the plane z = 0 of radius b > 0 and
with center at the origin. The 2-dimensional torus of radii 0 < a < b, denoted by
2
Ta,b , is the set of points in R3 at distance a from Sb1 . Use the Preimage Theorem
2
(Corollary 6.3.6) to prove that Ta,b is a 2-dimensional manifold (Find an explicit
equation describing it! )
Exercise 6.11. Let M (n) denote the space of all real n × n matrices. Let S(n)
denote the space of symmetric matrices:
S(n) := {B ∈ M (n) : B = B t }.
Prove that I is a regular value of the map
f : M (n) −→ S(n), f (A) = AAt .
Using the Preimage Theorem (Corollary 6.3.6), prove that the set of orthogonal
matrices
O(n) := {A ∈ M (n) : AAt = I},
is a compact embedded submanifold of M (n) of dimension n(n − 1)/2.
2
Hint: For compactness, show that O(n) ⊂ S n −1 .
LECTURE 7

7.1. Immersed submanifolds


There are two ways to describe submanifolds: either by giving (local) equations
that define them, or by giving a parameterization. As discussed in the previous lec-
ture, the first class corresponds to embedded submanifolds. The second class, called
immersed submanifolds, is more general; it incorporates several natural “subman-
ifolds” which are not embedded, but play an important role in various geometric
settings: Lie subgroups of Lie groups, orbits of Lie group actions, leaves of folia-
tions, etc.

Definition 7.1.1. Let M be a smooth manifold. An immersed k-dimensional


submanifold of M is a subset S ⊂ M endowed with a topology and a k-dimensional
differentiable structure such that the inclusion map i : S ,→ M , i(x) = x is a smooth
immersion.

Here are some obvious examples:

Example 7.1.2. (1) An embedded submanifold S ⊂ M with the differentiable


structure from Theorem 6.3.2 is also an immersed submanifold.
(2) If f : N → M is an injective immersion then S := f (N ) is an immersed
submanifold, with differentiable structure and topology such that f : N → S is
a diffeomorphism (this condition defines a unique topology and differentiable
structure on S!). In fact, every immersed submanifold S is of this type: just
take S = N , f = i.

Remark 7.1.3. It is important to note that:


• The topology of an immersed submanifold is not necessarily the induced
topology (!)
• There can be several immersed submanifolds with the same underlying
set S ⊂ M , i.e. there can exist several differentiable structures on the set
S, such that S is an immersed submanifold (!)

Here are some examples illustrating these remarks:

Example 7.1.4. Denote the six curves represented below by Ca , Cb , . . . , Cf :


61
62 IOAN MĂRCUT
, , MANIFOLDS

a. Topologist’s sine curve b. Figure 8 c. Shade of a trefoil knot

d. Dutch figure 8 e. Alien creature f. Non-analytic bifurcation

All these curves are immersed submanifolds in R2 ! Here are some of their properties:
a. The red line segment that Ca approaches is not part of Ca . An injective immer-
sion that parameterizes Ca (or a curve resembling Ca ) is:

j : (0, ∞) −→ R2 , j(t) = (t, sin(1/t)).

Note that Ca is in fact an embedded submanifold of R2 : this follows by Propo-


sition 6.3.3, because Ca is the 0-set of the submersion:

h : (0, ∞) × R −→ R, h(x, y) = y − sin(1/x).

The differentiable structure on Ca coming from j coincides with the differentiable


structure of Ca regarded as an embedded submanifold; in fact, we will prove that,
in general, on an embedded submanifold there is a unique differentiable structure
which makes it into an immersed submanifold; namely, the one constructed in
Theorem 6.3.2. Note that Ca is the only embedded submanifold among the 6
examples.
b. The curve Cb is the image of the immersion:

j : R −→ R2 , j(t) = (sin(2t), sin(t)).

Note that j is injective on either of the two intervals I1 := (−π, π) and I2 :=


(0, 2π); and that j(I1 ) = Cb = j(I2 ). Therefore, each of the maps

j1 := j|I1 : I1 −→ Cb , and j2 := j|I2 : I2 −→ Cb

endows Cb with the structure of an immersed submanifold. Note that the re-
sulting topologies are different! In fact, these are the only two differentiable
structures on Cb making Cb an immersed submanifold. Note that (0, 0) = j1 (0)
LECTURE 7. 63

and (0, 0) = j2 (π). At this point, the tangent spaces corresponding to the dif-
ferent smooth structures are different:
D ∂ ∂ E
T(0,0) Cb = d0 j1 (T0 R) = 2 (0,0) +
∂x ∂y (0,0)
D ∂ ∂ E
T(0,0) Cb = dπ j2 (Tπ R) = 2 (0,0) − .
∂x ∂y (0,0)
c. The curve Cc has 23 = 8 differentiable structures; at each intersection point there
are 2 options how to separate the curve. Also, the 8 topologies are different!
d. The curve Cd has a unique differentiable structure as an immersed submanifold;
but note that it is not an embedded submanifold!
e. The curve Ce is a closed version of the curve Ca ; the behavior at the two “ends”
is similar to that of the Topologist’s sine curve. This curve has a unique dif-
ferentiable structure as an immersed submanifold, but note that it is not an
embedded submanifold!
f. The curve Cf represents two curves that are “infinitely tangent”. There are two
differentiable structure on Cf which make it into an immersed submanifold. For
example, as a model one can take the following two decompositions of Cf :
Cf = {(x, 0) : x ∈ R} ∪ {(x, e−1/x ) : x > 0};
Cf = {(x, 0) : x > 0} ∪ {(x, f (x)) : x ∈ R},
where f is the smooth function: f (x) = 0, if x ≤ 0 and f (x) = e−1/x , if x > 0,
which give different decompositions of Cf into connected components. Note that
both smooth structures have the same tangent space at (0, 0):
D ∂ E
T(0,0) Cf = .
∂x (0,0)
The following is the main tool for comparing different immersed submanifolds
with the same underlying subset.
Lemma 7.1.5. Let f : N → M be an injective immersion. Consider a smooth
map ϕ : X → M such that ϕ(X) ⊂ f (N ), and denote by ϕ e : X → N the unique
map satisfying f ◦ ϕ
e = ϕ. If ϕ
e is continuous then it is smooth.
X
ϕ
ϕ
 "
e

N
f
/M

Proof. Let x ∈ X, and denote y := ϕ(x)


e ∈ N and z := f (y) = ϕ(x) ∈ M . Since
f is an immersion, by Exercise 6.6, there exists a smooth map σ : V → U , where
U ⊂ N is an open neighborhood of y and V ⊂ M is an open neighborhood of
z such that f (U ) ⊂ V and σ ◦ f |U = idU . Since ϕ e is continuous, we have that
W := ϕ e−1 (U ) ⊂ X is an open neighborhood of x.
x∈W
ϕ
ϕ
 (/
e
f
y∈U o V 3z
σ

Note that on W , the map ϕ


e is a composition of smooth functions:
e W = σ ◦ ϕ|W ,
ϕ|
64 IOAN MĂRCUT
, , MANIFOLDS

e is smooth on W . So, around every point x ∈ X we have found an open


hence ϕ
neighborhood on which ϕ
e is smooth; this implies that ϕ
e is smooth. 
Corollary 7.1.6. Let fi : Ni → M , i = 1, 2, be two injective immersions from
manifolds of the same dimension dim(N1 ) = dim(N2 ) which have the same image
f1 (N1 ) = f2 (N2 ) = S. If the map f2−1 ◦ f1 : N1 → N2 is continuous, then it is
a diffeomorphism. Hence f1 and f2 induce the same structure of a k-dimensional
immersed submanifold on S (recall Example 7.1.2 (2)).
Proof. By Lemma 7.1.5, χ := f2−1 ◦ f1 is smooth. Since f1 = f2 ◦ χ, it follows
that χ is an immersion between manifolds of the same dimension, hence it is a local
diffeomorphism. Since χ is a bijection, it is a diffeomorphism. 
Remark 7.1.7. The condition dim(N1 ) = dim(N2 ) in Corollary 7.1.6 superfluous.
Namely, since χ : N1 → N2 is onto, it follows by Sard’s Theorem (see Remark 6.3.8)
that dim(N1 ) ≥ dim(N2 ), and the equality follows because χ is an immersion.
As a consequence of the previous result, we obtain:
Corollary 7.1.8. The differentiable structure on an embedded k-dimensional sub-
manifold S ⊂ M constructed in Theorem 6.3.2 is the unique k-dimensional differ-
entiable structure on S for which the inclusion map is an immersion.
Proof. Let f : N → M be an injective immersion with f (N ) = S, which induces a
second k-dimensional differentiable structure on S. As an embedded submanifold,
S has the induced topology from M ; hence f : N → S is continuous, and so
by Corollary 7.1.6, f is a diffeomorphism. Thus, the two differentiable structure
coincide. 
Let us introduce the following notion:
Definition 7.1.9. (1) Let X and Y be topological spaces. An injective continuous
map f : X → Y is called a topological embedding if f : X → f (X) is a
homeomorphism, where f (X) is endowed with the induced topology from Y .
(2) Let N and M be smooth manifolds. A smooth map f : N → M is called a
smooth embedding if f is an injective immersion, and f : N → f (N ) is a
homeomorphism, where f (N ) is endowed with the induced topology.
Embedded submanifolds are precisely the image of smooth embeddings:
Proposition 7.1.10. Let S ⊂ M be a subset. The following are equivalent:
(1) S is an embedded k-dimensional submanifold.
(2) S is an immersed k-dimensional submanifold, whose topology is the induced
topology from M (see Definition 1.1.2).
(3) There exists a k-dimensional manifold N and a smooth embedding f : N → M
such that S = f (N ).
Proof. (1) ⇒ (2) follows because the topology on an embedded submanifold is the
induced topology. (2) ⇒ (3) follows with N := S, f := i : S → M . Finally, we
prove (3) ⇒ (1). Let f : N → M be a smooth embedding. Note the following
consequence of the Local Immersion Theorem 6.2.2: every point in N has an open
neighborhood U ⊂ N such that f (U ) is an embedded k-dimensional submanifold.
Since f is a topological embedding, f (U ) = V ∩ f (N ), for some open set V ⊂ M .
We conclude that every point in S := f (N ) has an open neighborhood V ⊂ M
LECTURE 7. 65

such that S ∩ V is a k-dimensional embedded submanifold. This implies that S is


a k-dimensional embedded submanifold. 
Next, recall that a continuous map is called proper if the preimages of compact
subsets are compact (see Definition 4.4.3). We note the following topological result:
Lemma 7.1.11. Let f : X → Y be a continuous map. Assume that f is proper
and injective and that Y is Hausdorff and locally compact (i.e. every point has a
compact neighborhood). Then f is a topological embedding and f (X) is closed.
Proof. We show that f is a closed map, i.e. the image of any closed set is closed.
Let C ⊂ X be a closed set, and let y ∈ Y \f (C). Since Y is locally compact, y
has an open neighborhood V with V compact. Since f is proper, E := f −1 (V ) is
compact in X. Thus, f (E) is compact. Since Y is Hausdorff, f (E) is closed. Let
W = V \f (E). One easily checks that W is a neighborhood of y disjoint from f (C),
as desired. In particular, f (X) ⊂ Y is closed. Therefore, f : X → f (X) is a closed
continuous bijection; which implies that it is a homeomorphism. 
Finally, Lemma 7.1.11 and Proposition 7.1.10 give:
Corollary 7.1.12. The image of a proper injective immersion is a closed embedded
submanifold. In particular, the image of an injective immersion from a compact
manifold is an embedded submanifold.

7.2. A weak version of Whitney’s Embedding Theorem


As an application of the existence of partitions of unity, and of Corollary 7.1.12, we
prove of a weak version of Whitney’s Embedding Theorem 2.4.4.
Theorem 7.2.1. A manifold which admits a finite atlas can be embedded in Rn ,
for large enough n. In particular, a compact manifold can be embedded in some Rn .
Proof. Consider a finite atlas {(Ui , ϕi )}ki=1 on a manifold M . Let {ρi }ki=1 be a
partition of unity subordinated to the cover {Ui }ki=1 . For each i, define the function

0, x∈ / Ui ;
ei : M → Rm , ϕ
ϕ ei (x) =
ρi (x)ϕi (x), x ∈ Ui .
ei is smooth on the open sets Ui and M \supp(ρi ), it is smooth on M . Let
Since ϕ
f : M → Rm(k+1) , f (x) = (ρ1 (x), . . . , ρk (x), ϕ
e1 (x), . . . , ϕ
ek (x)).

P that f is an injective immersion. Let x, y ∈ M such that f (x) = f (y).


We prove
Since i ρi (x) = 1, there is i such that ρi (x) > 0. Since f (x) = f (y), we have that
ρi (x) = ρi (y) > 0 and ϕ ei (y). In particular, x, y ∈ Ui , and therefore:
ei (x) = ϕ
ϕ
ei (x) ϕ
ei (y)
ϕi (x) = = = ϕi (y).
ρi (x) ρi (y)
Since ϕi is injective, we conclude that x = y. To see that f is an immersion,
consider x ∈ M and v ∈ Tx M such that dx f (v) = 0. This implies that dx ρi (v) = 0
ei (v) = 0. Let i be such that ρi (x) > 0. Then x ∈ Ui , and note that
and dx ϕ
dx ϕ
ei (v) = dx (ρi ϕi )(v) = dx ρi (v)ϕi (x) + ρi (x)dx ϕi (v).
Thus, we obtain that dx ϕi (v) = 0. Since ϕi is a diffeomorphism, v = 0.
By Corollary 7.1.12, if M is compact then f is an embedding. In the general
case consider a proper smooth function h : M → R, which exists by Corollary 4.4.4.
66 IOAN MĂRCUT
, , MANIFOLDS

Then the map (f, h) : M → Rm(k+1)+1 is a proper injective immersion, thus, by


Corollary 7.1.12 an embedding. 

7.3. Exercises
Exercise 7.1. If X is compact and Z is Hausdorff, prove that any continuous
bijection f : X → Z is a homeomorphism. Do not use Lemma 7.1.11!
Definition 7.3.1. An immersed submanifold S ⊂ M is called an initial subman-
ifold, if for every smooth map ϕ : X → M , such that ϕ(X) ⊂ S, we have that the
induced map ϕ e : X → S, ϕ(x)
e = ϕ(x) is smooth.
X
ϕ
ϕ
 "
e

S
i /M
Exercise 7.2. Prove that embedded submanifolds are initial. Hint: Use Lemma
7.1.5.
Exercise 7.3. Prove that an initial submanifold has a unique differentiable struc-
ture such that the inclusion map is an immersion. Hint: Use Corollary 7.1.6.
Exercise 7.4. (a) Construct a smooth map χ : (−, 1 + ) → [0, 1] such that
χ|(−,0] = 0, χ|[1,1+) = 1, χ(0, 1) = (0, 1),
and χ : (0, 1) → (0, 1) is a diffeomorphism.
(b) Let M be a manifold and consider two smooth curves γ1 , γ2 : [0, 1] → M such
that γ1 (1) = γ2 (0). Show that the following curve is smooth:

γ1 (χ(t)), 0≤t≤1
γ : [0, 2] −→ M, γ(t) =
γ2 (χ(t − 1)), 1 ≤ t ≤ 2
Recall that a map from a closed interval f : [a, b] → M is smooth iff there is a
smooth map fe : (a − , b + ) → M , for some  > 0, such that fe|[a,b] = f .
(c) Consider a subset in R2 which looks like the letter T :
S = {(0, t) : t ∈ R} ∪ {(t, 0) : t ∈ [0, ∞)}.
Prove that S is an immersed submanifold of R2 .
(d) Using item (b) prove that S is not an initial submanifold of R2 .
Exercise 7.5. Consider the six curves from Example 7.1.4. Show that Cb , Cc , Cd ,
and Cf are not initial submanifold. Explain why Ca and Ce are initial manifolds
(for Ce , you do not need to give a rigorous proof). Show that Ce is not an embedded
submanifold (note that Ce is a compact subset).
Exercise 7.6. Let M be a smooth manifold, and let X ⊂ M be a subset which is
at most countable. Prove that X is an initial submanifold of M . Prove that X is
an embedded submanifold iff it is discrete, i.e. for every point x ∈ X has an open
neighborhood U ⊂ M such that X ∩ U = {x}.
Exercise 7.7. For a ∈ R\Q, consider the map from Exercise 6.2:
fa : R −→ S 1 × S 1 , fa (t) := (eit , eita ).
Show that fa (R) is an initial submanifold of the 2-torus S 1 × S 1 .
LECTURE 7. 67

For the following exercises, we consider another class of submanifolds. This


notion is not standard; however, it seems useful because it is less general than
initial submanifold, but it includes the interesting examples of leaves of foliations,
Lie subgroups and orbits of smooth actions of Lie groups.
Definition 7.3.2. A subset S ⊂ M is called a leaf-like submanifold of codimen-
sion l, if for every p ∈ S there exists a submersion f : U → Rl , where U is an
open neighborhood of p in M , and there exists a subset Λ ⊂ Rl which is at most
countable such that
U ∩ S = f −1 (Λ).
Exercise 7.8. Prove that the initial submanifold from Exercise 7.7 is a leaf-like
submanifold of codimension one.
Exercise 7.9. Prove that an embedded submanifold S ⊂ M of codimension l is a
leaf-like submanifold of codimension l.
Exercise 7.10. Show that the initial submanifold Ce from Example 7.1.4 is not a
leaf-like submanifold.
Exercise 7.11. Prove that any leaf-like submanifold is an initial submanifold.
Don’t forget second countability!
Exercise 7.12. Prove that a connected leaf-like submanifold S ⊂ M which is
closed (as a subset of M ) is also embedded.
The following exercise represents the converse of Corollary 4.4.2, and gives an
algebraic description of closed embedded submanifolds (recall also Theorem 5.3.2).
Exercise 7.13. Let ϕ : S → M be a smooth map such that the pullback map
ϕ∗ : C ∞ (M ) −→ C ∞ (S), ϕ∗ (f ) = f ◦ ϕ,
is surjective. The goal of this exercise is to show that ϕ is a closed embedding, and
so, by Proposition 7.1.10 and Corollary 7.1.12, ϕ(S) ⊂ M is a closed embedded
submanifold and ϕ : S → ϕ(S) is a diffeomorphism. Prove this by showing that
the following hold:
(a) ϕ is injective.
(b) ϕ is an immersion (argue “by contradiction”: if it is not an immersion prove
that there is x ∈ S and a vector v ∈ Tx S such that v 6= 0 but v(g) = 0 for all
g ∈ C ∞ (S)).
(c) ϕ is an embedding (show first that open sets of the form f −1 (0, ∞), f ∈ C ∞ (S)
form a basis for the topology of S; then show that the image via ϕ of such open
sets is open in ϕ(S)).
(d) ϕ(S) is closed.
(e) Show that ϕ(S) is a closed embedded submanifold (apply Corollary 7.1.12).
LECTURE 8

8.1. The tangent bundle


Definition 8.1.1. The tangent bundle of a smooth manifold M is the set of all
tangent vectors: G
T M := Tp M.
p∈M
The canonical projection is the map
π : T M −→ M, v ∈ Tp M =⇒ π(v) = p.
Theorem 8.1.2. Let M be a smooth manifold of dimension m. The tangent bundle
T M has a unique 2m-dimensional differentiable structure such that, for any chart
(U, ϕ) on M , the pair (T U, dϕ) is a smooth chart on T M , where
dϕ : T U −→ T ϕ(U ) = ϕ(U ) × Rm , v ∈ Tp M 7→ (ϕ(p), dp ϕ(v)).
Moreover, the canonical projection π : T M → M is a surjective submersion.
Proof. Let A = {(Uα , ϕα )}α∈I be a maximal atlas on M . We will apply Propo-
sition 3.1.1 to the collection Ae := {(T Uα , dϕα )}α∈I . Note that: (1) the collection
{T Uα }α∈I covers T M ; (2) the maps dϕα : T Uα → ϕα (Uα ) × Rm are bijections; (3)
dϕα (T Uα ∩ T Uβ ) = ϕα (Uα ∩ Uβ ) × Rm
are open sets; (4) the transition maps of Ae are smooth because they are the differ-
entials of the transition maps of A:
dϕβ ◦ (dϕα )−1 = d(ϕβ ◦ ϕ−1 m m
α ) : ϕα (Uα ∩ Uβ ) × R −→ ϕβ (Uα ∩ Uβ ) × R .
So the conditions (1)-(4) from Proposition 3.1.1 are satisfied, therefore T M has a
unique topology for which the maps (T Uα , dϕα ) are (topological) charts. For this
topology, the map π : T M → M is continuous: for any open set V ⊂ M , we have
that π −1 (V ) is open:
[ [
π −1 (V ) = π −1 (V ∩ Uα ) = (dϕα )−1 (ϕα (Uα ∩ V ) × Rm ) .
α∈I α∈I
To show that the topology is second countable, by Lemma 4.3.4, we can consider
a countable refinement {Vk }k≥1 of the cover {Uα }α∈I , i.e. Vk ⊂ Uik for some ik ∈ I.
If {Bl }l≥0 is a countable basis of Rm × Rm , then {π −1 (Vk ) ∩ (dϕik )−1 (Bl )}k,l≥1 is
a countable basis for the topology on T M .
69
70 IOAN MĂRCUT
, , MANIFOLDS

Next, we show that the topology is Hausdorff. Let v, w ∈ T M , with v 6= w. If


π(v) = π(w), consider a chart (Uα , ϕα ) such that π(v) ∈ Uα . Then v, w ∈ T Uα .
Since dϕα : T Uα → ϕα (U ) × Rm is a homeomorphism, and the second space is
Hausdorff, v and w can be separated by open sets in T Uα . If π(v) 6= π(w), let
V, W ⊂ M be open sets separating π(v) and π(w). Since π is continuous, π −1 (V )
and π −1 (W ) are open sets separating v and w.
Proposition 3.1.1 implies now that T M has a unique 2m-dimensional differen-
tiable structure such that Ae is a smooth atlas. Finally, we check that π : T M → M
is a submersion. Note that in the pair of charts (T Uα , dϕα ) and (Uα , ϕα ), π becomes
the canonical projection:
ϕα ◦ π ◦ (dϕα )−1 = pr2m m
m : ϕα (U ) × R −→ ϕα (U ).

Thus π is locally a submersion; hence it is a submersion. 

8.2. Vector bundles


Note that the tangent bundle T M of a manifold M has more structure: it is a
collection of vector spaces Tp M depending smoothly on p ∈ M . Such structures
arise in many natural constructions, therefore we introduce the following concept:
Definition 8.2.1. A smooth vector bundle of rank r consists of a surjective
submersion π : E → M satisfying
• For each p ∈ M , the fibre of π over p, denoted by
Ep := π −1 (p),
is endowed with the structure of a real vector space of dimension r.
• Around every point in M , there is an open set U and a diffeomorphism
Φ : π −1 (U ) −→ U × Rr ,
such that the following diagram commutes:

π −1 (U )
Φ / U × Rr pr1 ◦ Φ = π,
π pr1
 
U
idU
/U

where pr1 (x, v) = x, and such that for every p ∈ U , the restriction
pr2 ◦ Φ|Ep : Ep −→ Rr
is a linear isomorphism.
Here are some examples:
Example 8.2.2. The trivial vector bundle of rank r over a manifold M is
pr1 : M × Rr −→ M,
where the vector space structure on {p} × Rr is the obvious one.
Example 8.2.3. The tangent bundle T M of a manifold M is a vector bundle of
rank(T M ) = dim(M ).
LECTURE 8. 71

Example 8.2.4. The group ({±1}, ·) acts on S 1 × R via


±1 · (z, x) = (±z, ±x).
The action is free, and the quotient E := (S 1 × R)/{±1} is diffeomorphic to the
Möbius band. Then E is a vector bundle over S 1 with surjective submersion:
π : E −→ S 1 , π([z, x]) = z 2 .
For w ∈ S 1 , fix z ∈ S 1 one of the two solutions of the equation z 2 = w. The vector
space structure on the fibre Ew = {[z, x] : x ∈ R} is given by:
a[z, x] + b[z, y] = [z, ax + by], a, b ∈ R, [z, x], [z, y] ∈ Ew .
Next, we introduce vector bundle morphisms:
Definition 8.2.5. A morphism between two vector bundles πE : E → M and
πF : F → N is a pair of smooth maps Φ : E → F and ϕ : M → N such that for all
p ∈ M we have Φ(Ep ) ⊂ Fϕ(p) and that the induced map
Φ|Ep : Ep −→ Fϕ(p)
is linear. If Φ and ϕ are diffeomorphisms, then (Φ, ϕ) is said to be a vector bundle
isomorphism, and the two bundles are said to be isomorphic.
Remark 8.2.6. Note that the condition Φ(Ep ) ⊂ Fϕ(p) in the definition above is
equivalent to the commutativity of the following:

E
Φ /F πF ◦ Φ = ϕ ◦ πE
πE πF
 
M
ϕ
/N

Example 8.2.7. If ϕ : M → N is a smooth map, then (dϕ, ϕ), with df : T M →


T N , is a vector bundle morphism.
Definition 8.2.8. The restriction of a vector bundle π : E → M to an open set
U ⊂ M is the vector bundle over U
E|U := π −1 (U ), π|E|U : E|U −→ U.
Definition 8.2.9. A vector bundle isomorphic to a trivial vector bundle is called
trivializable.
Remark 8.2.10. Every vector bundle π : E → M is locally trivializable. This
holds by definition, since M can be covered by oven subsets U such that there
exists a vector bundle isomorphism Φ : E|U − ∼
→ U × Rr covering idU . Such a local
isomorphism to the trivial bundle is called a local trivialization.

8.3. Sections of vector bundles


Definition 8.3.1. A smooth section of a vector bundle π : E → M is a smooth
map σ : M → E such that π ◦ σ = idM ; in other words: σ(p) ∈ Ep , for all p ∈ M .
We denote the space of all sections of π : E → M by Γ(E).
Note that every vector bundle π : E → M has the zero-section 0 ∈ Γ(E)
which sends a point p ∈ M to the origin of the vector space Ep : p 7→ 0p ∈ Ep .
72 IOAN MĂRCUT
, , MANIFOLDS

The space of sections Γ(E) forms a module over the commutative algebra of
smooth functions C ∞ (M ), with the point-wise defined operations:
+ : Γ(E) × Γ(E) −→ Γ(E), (σ1 + σ2 )(p) := σ1 (p) + σ2 (p) ∈ Ep ,
· : C ∞ (M ) × Γ(E) −→ Γ(E), (f · σ)(p) := f (p)σ(p) ∈ Ep ,
where, on the right hand side, the vector space operations on Ep are used. The
zero-section plays the role of the zero-element of Γ(E).
Consider the case of a trivial bundle M × Rr → M . Its sections are smooth
maps of the form:
σ : M −→ M × Rr , σ(p) = (p, (f1 (p), . . . , fr (p))) ∈ M × Rr .
Thus, we have a one-to-one correspondence between Γ(M × Rr ) and C ∞ (M ; Rr ).
Using the structure of a C ∞ (M )-module of Γ(M × Rr ), we can decompose any
section σ uniquely as
σ = f1 σ1 + . . . + fr σr ,
where, for 1 ≤ i ≤ r, σi ∈ Γ(M × Rr ) represents the section
σi (p) := (p, (0, . . . , 1, . . . , 0)) ∈ M × Rr ,
with 1 is on the i-th position. In other words:
Γ(M × Rr ) is a free C ∞ (M )-module with basis σ1 , . . . , σr .
Since a general vector bundle π : E → M is locally trivializable, there is a cover
of M by open sets U such that Γ(E|U ) is a free C ∞ (U )-module of rank r.

8.4. Frames
Definition 8.4.1. Let π : E → M be a vector bundle. A local frame over U ⊂ M
is a smooth map
γ : E|U −→ Rr ,
such that, for all p ∈ U , γ|Ep : Ep → Rr is a linear isomorphism.
Lemma 8.4.2. Let π : E → M be a vector bundle, and let U ⊂ M be an open set.
There is a one-to-one correspondence between the following objects:
(1) local trivializations Φ : E|U − ∼→ U × Rr ;
r
(2) local frames γ : E|U → R ;
(3) sections σ1 , . . . , σr ∈ Γ(E|U ) such that, for every p ∈ U , σ1 (p), . . . , σr (p) forms
a basis of Ep .
This correspondence is such that, given a local frame γ, the corresponding local
trivialization is Φ := π × γ : E|U → U × Rr , and the corresponding set of sections
σ1 , . . ., σr is given by σi (p) := γp−1 (ei ), where e1 , . . . , er is the standard basis of Rr .
Proof. By passing to the restriction E|U , we may assume that U = M .
First, we explain the equivalence between (a) and (b). Consider a trivialization
Φ:E− ∼
→ M × Rr . Then, by the definition of Φ, we have that Φ = π × γ, where
γ : E → Rr is a frame. Conversely, let γ : E → Rr be a frame on E over M . We
need to show that Φ = π × γ : E → M × Rr is a diffeomorphism. Since Φ is a
bijection, and the manifolds have the same dimension, invoking the inverse function
theorem, it suffices to show that π × γ is an immersion. Let v ∈ Te E be such that
0 = de (π × γ) (v) = (de π(v), de γ(v)).
LECTURE 8. 73

By Corollary 6.3.6,ker(de π) = Te (Eπ(e) ). Therefore, v ∈ Te (Eπ(e) ), and so 0 =


de γ(v) = de γ|Eπ(e) (v). Since γ|Eπ(e) is a diffeomorphism, we have that v = 0.
We explain now the equivalence between (b) and (c). Let γ : E → Rr be a
frame on E. By the first part, Φ = π × γ : E → M × Rr is a diffeomorphism,
therefore the maps:
σi : M → E, σi (p) := Φ−1 (p, (0, . . . , 1, . . . , 0))
are smooth, and therefore σi ∈ Γ(E). Now, for all p ∈ M , γp (σi (p)) = ei , therefore
σ1 (p), . . ., σr (p) form a basis of Ep . Conversely, consider σ1 , . . ., σr ∈ Γ(E) which
form a point-wise basis of E. Consider the map
Ψ : M × Rr −→ E, Ψ(p, (a1 , . . . , ar )) := a1 σ1 (p) + . . . + ar σr (p).
Clearly, Ψ is a vector bundle morphism. An argument similar to the proof of
the first part shows that Ψ is also a diffeomorphism; hence Ψ is a vector bundle
isomorphism. The corresponding frame is the map γ : E → Rr determined by
Ψ−1 = π × γ. 
The Lemma implies that:
Corollary 8.4.3. For a vector bundle π : E → M , the following are equivalent:
(1) E is trivializable;
(2) E admits a global frame γ : E → Rr ;
(3) there are smooth sections σ1 , . . . , σr ∈ Γ(E) such that, for all p ∈ M , their
values at p form a basis of Ep .
Definition 8.4.4. Let π : E → M be a vector bundle of rank r. Let γ : E|U → Rr
be a local frame over the open set U ⊂ M , and let σ1 , . . . , σr ∈ Γ(E|U ) denote the
corresponding sections which, at every p ∈ U , form a basis σ1 (p), . . . , σr (p) of Ep .
The local representation of a section σ ∈ Γ(E) in this local frame is the
decomposition of σ|U ∈ Γ(E|U ) as a linear combination
σ|U = f1 σ1 + . . . + fr σr ,
with coefficients fi ∈ C ∞ (U ). The coefficients are determined by
(f1 , . . . , fr ) = γ ◦ σ|U : U −→ Rr .
Let us mention that vector bundles are determined by their module of smooth
sections; in fact, one can characterize algebraically which modules arise from vector
bundles. Without introducing the necessary terminology, we state the following:
Theorem ♣ 8.4.5 (Serre-Swan). Let M be a smooth manifold. The correspon-
dence E 7→ Γ(E) is a one-to-one correspondence between vector bundles over M
(up to isomorphism) and finitely generated projective modules over C ∞ (M ) (up to
isomorphism).

8.5. Vector fields as sections of the tangent bundle


Definition 8.5.1. A vector field on a smooth manifold M is a smooth section of
the tangent bundle π : T M → M ; in other words it is a smooth map
X : M −→ T M such that Xp ∈ Tp M, for all p ∈ M.
We use the notation Xp rather than X(p) for the value of X at p ∈ M .
74 IOAN MĂRCUT
, , MANIFOLDS

The space of all smooth vector fields is denoted by


X(M ) := Γ(T M ).
Let (U, ϕ) be a chart on M inducing coordinates ϕ = (x1ϕ , . . . , xm i
ϕ ), with xϕ ∈

C (U ). The chart induces a canonical frame on T M over U :
∂ ∂
, ..., ∈ X(U ).
∂x1ϕ ∂xm
ϕ

The local representation of a vector field X ∈ X(M ) is given by:


m
X ∂
X|U = Xi , X i ∈ C ∞ (U ),
i=1
∂xiϕ
where the coefficients are smooth functions on U .
The notion of trivializable specializes to:
Definition 8.5.2. If the tangent bundle of a manifold is trivializable, then the
manifold is said to be parallelizable.
Example 8.5.3. Clearly, Rm is parallelizable. A global frame on T Rm is given by

the vector fields ∂x i , with 1 ≤ i ≤ m.

Example 8.5.4. The circle S 1 is also parallelizable: if θ ∈ R/(2πZ) denotes the


“angle coordinate” on S 1 , then a global frame is given by the vector field

∈ X(S 1 ).
∂θ
Even though the θ makes sense as a coordinate only on sets of the form S 1 \{eiβ },

the vector field ∂θ makes sense globally, as it can be defined as the derivative of
rotation:
∂ d iθ
:= e θ=α .
∂θ eiα dθ
Example 8.5.5. Products of parallelizable manifolds are parallelizable. In par-
ticular the m-dimensional torus T m = (S 1 )m is parallelizable. Using the angle
coordinates (θ1 , . . . , θm ) ∈ (R/(2πZ))m , we obtain the global frame
∂ ∂
, . . . , m ∈ X(T m ).
∂θ1 ∂θ
The 2-dimensional sphere S 2 is not parallelizable. This is a consequence of the
following result (see Corollary 8.4.3):
Theorem ♣ 8.5.6 (The Hairy Ball Theorem). Every smooth vector field X ∈
X(S 2 ) has at least one zero, i.e. a point p ∈ S 2 such that Xp = 0.
Here is an interesting application of this result. Thinking about the surface of
the Earth as a manifold diffeomorphic to S 2 , and assuming that the velocity of the
wind can be represented by a smooth vector field X ∈ X(S 2 ), this result implies
that there always exists a point on Earth where the wind is not blowing (most likely
not in this part of the World, but you can check it out here: [22]).
It is known precisely which spheres are parallelizable:
Theorem ♣ 8.5.7. The only parallelizable spheres are S 0 , S 1 , S 3 and S 7 (i.e.
c
S 2 −1 , with c = 0, 1, 2, 3).
LECTURE 8. 75

You can check that this is a consequence of the following classical theorem:
Theorem ♣ 8.5.8. On the sphere S n the maximum number k = k(n) of vector
fields X1 , . . . , Xk such that at every p ∈ S n the vectors X1,p , . . . , Xk,p ∈ Tp S n are
linearly independent is given as follows: write n + 1 = (2a + 1)24b+c , with a, b ∈ N
and c ∈ {0, 1, 2, 3}, then k(n) = 8b + 2c − 1.
If n is even then k(n) = 0, so every vector field on S n has a zero. In general,
note that k(n) is quite small compared to n; in particular, k(n) ≤ 2 log2 (n + 1) + 1.

8.6. Exercises
Exercise 8.1. Prove that T S 1 is isomorphic to the trivial vector bundle S 1 × R.
Exercise 8.2. Let f : M → N be a smooth map. Prove that df : T M → T N is a
smooth vector bundle morphism.
Exercise 8.3. Let M be a smooth manifold. Prove that there exists a vector field
V on the manifold T M (i.e. V ∈ X(T M ) = Γ T (T M ) ) satisfying:
dv π(Vv ) = v, ∀ v ∈ T M.
Hint: construct V locally, and use a partition of unity.
Exercise 8.4. Recall that the m-dimensional real projective space is the space of
all lines through the origin in Rm+1 , with the manifold structure from Example
3.2.1:
Pm (R) := {l : l ⊂ Rm+1 is a line with 0 ∈ l}.
The tautological line bundle over Pm (R) is the rank 1 vector bundle π : L(m) →
Pm (R) whose fiber over a line l is the line l:
L(m) := {(l, v) : l ∈ Pm (R), v ∈ l}.
(a) Sketch a picture of L(1).
(b) Show that π : L(m) → Pm (R) is indeed a vector bundle of rank one, such that
the following map is morphism of vector bundles:

L(m)
Φ / Pm (R) × Rm+1 , Φ(l, v) := (l, v) ∈ Pm (R) × Rm+1 .
pr

π
 1
m
P (R)
id / Pm (R)

(c) Sketch a picture of the maps above for m = 1.


(d) Prove that L(1) is not trivializable (use Corollary 8.4.3).
(e) Prove that L(m) is not trivializable, by using the morphism of vector bundles:

L(1)
Ψ / L(m) ,
π
 pr
 1
P1 (R) / Pm (R)

Ψ(l, v) := (l × {0}, (v, 0)) ∈ Pm (R) × Rm+1 .


The following exercise gives a proof of the Hairy Ball Theorem 8.5.6. Any such
proof requires some knowledge of algebraic topology. We will give a proof using
only the winding number of a curve, which should be familiar from a standard
course in one-variable complex analysis.
76 IOAN MĂRCUT
, , MANIFOLDS

Definition 8.6.1. Let γ : S 1 → C∗ be a smooth closed curve, where C∗ := C\{0}.


The winding number of γ is defined by the formula:
I
1 dz
W (γ) := ∈ Z.
2πi γ z
This number is an integer, and counts how many of times γ goes around 0 ∈ C in
the positive direction, minus the number of times it goes around 0 in the negative
direction.
Note that the winding number is a homotopy invariant: if γt : S 1 → C∗ is a
family of closed curves depending continuously on t ∈ [0, 1], then W (γ0 ) = W (γ1 ).
This is because W (γt ) has integer values, and depends continuously on t; thus
W (γt ) must be constant.
∂ ∂
Exercise 8.5. For a vector field V = a(x, y) ∂x + b(x, y) ∂y ∈ X(R2 ), we denote by
1
γV : S → C the closed curve
γV (cos(θ), sin(θ)) := a(cos(θ), sin(θ)) + ib(cos(θ), sin(θ)) ∈ C.
(a) Let V ∈ X(R2 ) be a vector field which is nowhere zero, i.e. V(x,y) 6= 0 for all
(x, y) ∈ R2 . Construct a continuous family of closed curves γt : S 1 → C∗ , with
t ∈ [0, 1], such that γ1 = γV and γ0 = λ, where λ ∈ C∗ is a constant. Conclude
that W (γV ) = 0 (Hint: look at the vector field V on circles of radius t).
(b) Consider the stereographic atlas on S 2 with two charts from Example 1.2.7.
Let X ∈ X(S 2 ) be a vector field, and consider its representatives in the these
two charts:
∂ ∂
V s = a(x, y) + b(x, y) ∈ X(R2 ),
∂x ∂y
∂ ∂
V n = c(x, y) + d(x, y) ∈ X(R2 ).
∂x ∂y
Show that, at (x, y) 6= (0, 0), these local representatives are related by:
   
2 2 x y x y
c(x, y) = (y − x )a , − 2xy · b , ,
x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2
   
x y 2 2 x y
d(x, y) = −2xy · a , + (x − y )b , .
x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2
Conclude the following relation between the corresponding curves:
γV n = e2iθ (−γ V s ).
(c) Prove that every vector field X ∈ X(S 2 ) has at least one zero. Hint: assume
that a nowhere vanishing vector field X exists. By applying (a) to V s , and then
(b), show that γV n is homotopic to the curve e2iθ . Deduce that W (γV n ) = 2,
which contradicts (a).
LECTURE 9

9.1. Vector fields as derivations


Vector fields have a natural action on smooth functions:
Definition 9.1.1. The derivative along a vector field X ∈ X(M ) of a smooth
function f ∈ C ∞ (M ) is the function X(f ) ∈ C ∞ (M ), M 3 p 7→ Xp (f ).
The fact that X(f ) is indeed a smooth function can be checked locally. Repre-
sent X in a local chart (U, ϕ = (x1ϕ , . . . , xm
ϕ )) on M ,
m
X ∂
X|U = Xi , X i ∈ C ∞ (U ),
i=1
∂xiϕ
then we have the explicit formula:
m m −1
i ∂ i ∂f ◦ ϕ
X X
X(f )|U = X (f |U ) = X ◦ϕ
i=1
∂xiϕ i=1
∂xi
which shows that X(f )|U is smooth; and since (U, ϕ) is arbitrary, X(f ) is smooth.
The resulting operation f 7→ X(f ) on C ∞ (M ) has the following property:
Definition 9.1.2. A derivation of C ∞ (M ) is a linear map
D : C ∞ (M ) −→ C ∞ (M )
satisfying the rule
D(f · g) = f · D(g) + D(f ) · g, for all f, g ∈ C ∞ (M ).
Proposition 9.1.3. Every derivation D of C ∞ (M ) is the derivative along a unique
vector field X ∈ X(M ).
Proof. Let X ∈ X(M ). Because Xp is a derivation for all p ∈ M :
Xp (f · g) = f (p)Xp (g) + Xp (f )g(p),
we obtain that the derivative along X is indeed a derivation of C ∞ (M ):
X(f · g) = f · X(g) + X(f ) · g.
Assume that X, Y ∈ X(M ) induces the same derivation, i.e. X(f ) = Y (f ) for all
f ∈ C ∞ (M ). Hence Xp (f ) = Yp (f ), for all f ∈ C ∞ (M ) and p ∈ M , and so
Xp = Yp for all p ∈ M . Thus X = Y , which shows the uniqueness assertion.
77
78 IOAN MĂRCUT
, , MANIFOLDS

Finally, consider a derivation D of C ∞ (M ). Evaluating the equality D(f g) =


f D(g) + D(f )g at a point p ∈ M , we obtain that the linear map
Xp : C ∞ (M ) → R, Xp (f ) := D(f )(p)
is a derivation at p. Thus, we obtain a map X : M → T M , with Xp ∈ Tp M , which
satisfies that, for all f ∈ C ∞ (M ), the map M 3 p 7→ Xp (f ) is smooth (because
this map is D(f )). We show that this implies smoothness of X. Consider a chart
(U, ϕ = (x1ϕ , . . . , xm
ϕ )) around a point p ∈ M . Note that, even if X is not smooth,
we can still decompose X|U in the corresponding frame:
m
X ∂
X|U = Xi i ,
i=1
∂x ϕ

just that the coefficients X i : U → R might not be smooth. For 1 ≤ i ≤ m,


eiϕ ∈ C ∞ (M ) have the same germ as xiϕ at p, i.e. x
let x eiϕ |V = xiϕ |V for some
neighborhood V ⊂ U of p. Then
xiϕ )|V = X|V (xiϕ |V ) = X i |V .
X(e
xiϕ ) ∈ C ∞ (M ), we have shown that X i |V is smooth, hence X|V : V → T M
Since X(e
is smooth. So X is smooth because it is smooth around any p ∈ M . This concludes
the proof, because D and X act the same on C ∞ (M ). 
From now on, we will identify the space of vector fields with the space of
derivations of C ∞ (M ).

9.2. The Lie bracket


Let X, Y ∈ X(M ) be two vector fields, which we regard as derivations:
X, Y : C ∞ (M ) −→ C ∞ (M ).
In general, their composition is not a derivation. The failure for this to hold is
(X ◦ Y )(f · g) − (X ◦ Y )(f ) · g − f · (X ◦ Y )(g) = X(f ) · Y (g) + Y (f ) · X(g).
Note that the right hand side is symmetric in X and Y . Therefore, if we subtract
from this the same equation with X and Y interchanged, we obtain:
(X ◦ Y − Y ◦ X)(f · g) = (X ◦ Y − Y ◦ X)(f ) · g + f · (X ◦ Y − Y ◦ X)(g).
Thus, X ◦ Y − Y ◦ X is again a derivation of C ∞ (M ), and so, by Proposition 9.1.3,
again a vector field.
Definition 9.2.1. The Lie bracket is the following operation on vector fields:
[·, ·] : X(M ) × X(M ) → X(M ), [X, Y ] := X ◦ Y − Y ◦ X.
The Lie bracket [X, Y ] is also called the commutator of X and Y (because it
measures whether the derivations X and Y commute, i.e. whether X ◦ Y = Y ◦ X).
The Lie bracket has the following properties (the proof is left to the reader):
Proposition 9.2.2. The Lie bracket
(1) is skew-symmetric:
[X, Y ] = −[Y, X];
(2) is R-bilinear:
[X, aY + bZ] = a[X, Y ] + b[X, Z];
LECTURE 9. 79

(3) satisfies the Jacobi identity:


[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0;
(4) satisfies the Leibniz rule:
[X, f Y ] = f [X, Y ] + X(f )Y ;
for all X, Y, Z ∈ X(M ), a, b ∈ R and f ∈ C ∞ (M ).
Remark 9.2.3. A vector space V endowed with an operation [·, ·] : V × V → V
satisfying (1), (2) and (3) is called a Lie algebra. This notion will be discussed in
Lecture 11.
Let us write the Lie bracket [X, Y ] of two vector fields X, Y ∈ X(M ) in a local
chart (U, ϕ = (x1ϕ , . . . , xm
ϕ )) on M . Denote
m m
X ∂ X ∂
X|U = Xi , Y |U = Yi .
i=1
∂xiϕ i=1
∂xiϕ

Then, for any f ∈ C (U ) we have that:
m m m
X ∂f  X X i ∂  j ∂f 
X ◦ Y (f )|U = X Yj j = X Y =
j=1 ∂xϕ i=1 j=1
∂xiϕ ∂xjϕ
m X
m m X m
X ∂Y j ∂  X ∂2f
= Xi (f ) + X i j
Y .
i=1 j=1
∂xiϕ ∂xjϕ i=1 j=1 ∂xiϕ ∂xjϕ
Note that there are second order derivatives of f appearing, and this is the reason
2 2
why X ◦Y is not a vector field. However, since ∂xi∂∂xj = ∂xj∂∂xi , these terms cancel
ϕ ϕ ϕ ϕ
in the commutator [X, Y ]|U . We obtain the local expression of the Lie bracket:
m m
!
j j
i ∂Y i ∂X ∂
X X
[X, Y ]|U = X −Y .
j=1 i=1
∂xϕi ∂xϕ ∂xjϕ
i

9.3. Related vector fields


Vector fields cannot be transported via smooth maps (only via diffeomorphisms).
However, the following relation makes sense in general:
Definition 9.3.1. Let ϕ : M → N be a smooth map. A vector field X ∈ X(M ) is
said to be ϕ-related to a vector field Y ∈ X(N ) if, for all p ∈ M ,
dp ϕ(Xp ) = Yϕ(p) .
Recall that a smooth map ϕ : M → N induces a pullback map on smooth
functions, denoted:
ϕ∗ : C ∞ (N ) → C ∞ (M ), ϕ∗ (f ) := f ◦ ϕ.
Here is an alternative description of related vector fields in terms of the pullback:
Lemma 9.3.2. Let ϕ : M → N be a smooth map. Then X ∈ X(M ) if ϕ-related to
Y ∈ X(N ) iff
X(ϕ∗ (f )) = ϕ∗ (Y (f )), ∀ f ∈ C ∞ (N ).
80 IOAN MĂRCUT
, , MANIFOLDS

Proof. At p ∈ M , the left-hand-side reads:


X(ϕ∗ (f ))(p) = Xp (f ◦ ϕ) = (dp ϕ(Xp )) (f ),
and the right-hand-side:
ϕ∗ (Y (f ))(p) = Yϕ(p) (f ).
The conclusion follows by noting that two vectors in Tϕ(p) N are equal iff they act
the same on smooth functions. 
The following is a very useful result:
Lemma 9.3.3. Let ϕ : M → N be a smooth map, and let X 1 , X 2 ∈ X(M ) and
Y 1 , Y 2 ∈ X(N ) be four vector fields. If X 1 is ϕ-related to Y 1 and X 2 is ϕ-related
to Y 2 , then also [X 1 , X 2 ] ∈ X(M ) is ϕ-related to [Y 1 , Y 2 ] ∈ X(N ).
Proof. Using Lemma 9.3.2, this result is straightforward:
ϕ∗ ([Y 1 , Y 2 ](f )) = ϕ∗ Y 1 ◦ Y 2 (f ) − ϕ∗ Y 2 ◦ Y 1 (f ) =
 

= X 1 ϕ∗ (Y 2 (f )) − X 2 ϕ∗ (Y 1 (f )) =
 

= X 1 ◦ X 2 (ϕ∗ f ) − X 2 ◦ X 1 (ϕ∗ f ) =
= [X 1 , X 2 ](ϕ∗ f ).

Vector fields can always be transported along diffeomorphisms.
Definition 9.3.4. The pushforward of a vector field X ∈ X(M ) via a diffeomor-

phism ϕ : M −→ N is the vector field
ϕ∗ X = dϕ ◦ X ◦ ϕ−1 ∈ X(N ),
more explicitly:
(ϕ∗ X)q := dp ϕ(Xp ), where p = ϕ−1 (q).
Note that ϕ∗ X is the unique vector field on N such that X and ϕ∗ X are ϕ-related.
The pullback of a vector field Y ∈ X(N ) via the diffeomorphism ϕ : M −∼
→N
is the vector field
ϕ∗ Y := (ϕ−1 )∗ Y = (dϕ)−1 ◦ Y ◦ ϕ ∈ X(M ).
Note that ϕ∗ Y is the unique vector field on M such that ϕ∗ Y and Y are ϕ-related.
Example 9.3.5. Let ι : M ,→ N denote the inclusion of the embedded submanifold
M of N . Then X ∈ X(M ) is ι-related to Y ∈ X(N ) iff
Yp = Xp ∈ Tp M ⊂ Tp N, ∀ p ∈ M
For example, consider ι : S 1 ,→ R2 , and the following vector field on R2
∂ ∂
Y := x −y ∈ X(R2 ).
∂y ∂x
First, note that Y is tangent to S 1 because, for (u, v) ∈ S 1 , we have that
n ∂ ∂ o
T(u,v) S 1 := ker(d(u,v) (x2 + y 2 − 1)) = a (u,v) + b (u,v) : ua + bv = 0 .
∂x ∂y
Consider the “angle coordinate” θ ∈ R/(2πZ) on S 1 (see Example 8.5.4),
θ 7→ (x(θ), y(θ)) := (cos(θ), sin(θ)).
LECTURE 9. 81

∂ ∂
We claim that Y is ι-related to X := ∂θ , i.e. Y |S 1 = ∂θ . We will use Lemma 9.3.2.
For any f ∈ C ∞ (R2 ), we have that
∂f ∂f
Y (f ) = x (x, y) − y (x, y),
∂y ∂x
and so
∂f ∂f
ι∗ (Y (f )) = cos(θ) (cos(θ), sin(θ)) − sin(θ) (cos(θ), sin(θ)).
∂y ∂x
On the other hand, by the chain rule, we have that:

X(ι∗ f ) = (f (cos(θ), sin(θ))) =
∂θ
∂f ∂ cos(θ) ∂f ∂ sin(θ)
= (cos(θ), sin(θ)) + (cos(θ), sin(θ)) =
∂x ∂θ ∂y ∂θ
∂f ∂f
= cos(θ) (cos(θ), sin(θ)) − sin(θ) (cos(θ), sin(θ)).
∂y ∂x

9.4. Change of coordinates rule for vector fields


Let X be a smooth vector field on M , and let (U, ϕ = (x1ϕ , . . . , xm
ϕ )) be a chart.
Denote the local expression of X by:
m
X ∂
X|U = X i i , X i ∈ C ∞ (U ).
i=1
∂x ϕ

Consider a second chart (V, ψ = (yψ1 , . . . , yψm )) on M , and denote the corre-
sponding local expression of X by
m
X ∂
X|V = Y i i , Y i ∈ C ∞ (V ).
i=1
∂yψ

Thus, on U ∩ V , we have that:


m m
X ∂ X ∂
Xi = Yi i .
i=1
∂xiϕ i=1
∂yψ

Applying this equality to the coordinate function yψj , we obtain

j
m
X
i
∂yψj
Y = X .
i=1
∂xiϕ
This expression represents the rule of change of coordinates for vector fields, i.e. it
gives the local expression of X in the chart (V, ψ) in terms of the local expression
of X in the chart (U, ϕ).

9.5. Integral curves of a vector field


Let I ⊂ R be an open interval, and consider a smooth curve γ : I → M . Recall
that the derivative of γ is the following curve in T M :
dγ ∂
(t) := dt γ ∈ Tγ(t) M.
dt ∂t
We will consider curves whose derivatives are given by a fixed vector field:
82 IOAN MĂRCUT
, , MANIFOLDS

Definition 9.5.1. Let X ∈ X(M ) be a vector field. An integral curve of X is a


smooth curve γ : I → M , where I ⊂ R is an open interval, which satisfies:

(t) = Xγ(t) , for all t ∈ I.
dt
The integral curve γ is said to start at a point p ∈ M , if 0 ∈ I and γ(0) = p.
Let (U, ϕ = (x1ϕ , . . . , xm
ϕ )) be a chart on M . Denote the local expression of a
vector field X ∈ X(M ) in this chart by:
m
X ∂
X|U = Xi , X i ∈ C ∞ (U ).
i=1
∂xiϕ

We will write locally what it means for a curve γ : I → U to be an integral curve


of X. Denote the local expression of γ in this chart by:
ϕ ◦ γ(t) = (γϕ1 (t), . . . , γϕm (t)), t ∈ I.
Then, as explained in Subsection 5.2, we have that
m
dγ ∂  X dγϕi ∂
(t) = dt γ = (t) i γ(t) .
dt ∂t i=1
dt ∂xϕ

Therefore, the condition that γ be an integral curve of X is equivalent to the


following system of ordinary differential equations (ODE’s)
dγϕi
(t) = X i ◦ ϕ−1 (γϕ1 (t), . . . , γϕm (t)), for 1 ≤ i ≤ m,
dt
on curves (γϕ1 , . . . , γϕm ) : I → ϕ(U ). We write this system in a more compact way:
dz
(t) = F (z(t)), z : I → V,
dt
where V = ϕ(U ) ⊂ Rm and
z = (γϕ1 , . . . , γϕm ) : I → V, F = (X 1 , . . . , X m ) ◦ ϕ−1 : V → Rm .
Such systems are studied using standard techniques from the theory of ODE’s. We
will use the following result (e.g. [1, Appendix C]):
Theorem 9.5.2. Let F : V → Rm be a smooth map, where V ⊂ Rm is an open
set. For some p ∈ V , consider the ordinary differential equation
dz
(t) = F (z(t)), z(0) = p.
dt
Every point in V has an open neighborhood W and there exists a smooth map
y : (−, ) × W −→ V,
for some  > 0, such that the curve t 7→ y(t, p) is a solution of the ODE, i.e.
dy
(t, p) = F (y(t, p)), y(0, p) = p.
dt
Moreover, the solution is unique, in the sense that if p ∈ W and t 7→ ye(t, p) is a
second solution defined for t around 0, then ye(t, p) = y(t, p) for t around 0.
LECTURE 9. 83

9.6. Exercises
Exercise 9.1. Consider the polar coordinates (U, r, θ) on R2 , where
U = R2 \ (−∞, 0] × {0} , x = r cos(θ), y = r sin(θ), (r, θ) ∈ (0, ∞) × (−π, π).


Write in polar coordinates the expression of an arbitrary vector field on R2 :


∂ ∂
X = a(x, y) + b(x, y) ∈ X(R2 ).
∂x ∂y
Exercise 9.2. Let ϕ : M → N be a local diffeomorphism (recall Definition 3.3.3).
Define the pullback ϕ∗ X of a vector field X ∈ X(N ) along ϕ (your definition should
generalize Definition 9.3.4). Show that this operation preserves the Lie bracket:
ϕ∗ [X, Y ] = [ϕ∗ X, ϕ∗ Y ], for all X, Y ∈ X(N ).
Exercise 9.3. Calculate the Lie bracket of the vector fields:
∂ ∂ ∂ ∂
X = sin(x + y) + exy , Y = y 2 + x2 .
∂x ∂y ∂x ∂y
Exercise 9.4. Consider the map
ϕ : R2 −→ R2 , ϕ(x, y) := (ex cos(y), ex sin(y)),
i.e. ϕ is the complex exponential written in real coordinates.

(a) Find two vector fields X, Y ∈ X(R2 ) such that X is ϕ-related to ∂x and Y is

ϕ-related to ∂y . Are these vector fields uniquely determined by this condition?
Calculate [X, Y ].
∂ ∂
(b) Find two vector fields U, V ∈ X(R2 ) such that ∂x is ϕ-related to U and ∂y is
ϕ-related to V . Are these vector fields uniquely determined by this condition?
Calculate [U, V ].
Exercise 9.5. Let ϕ : M → N and ψ : N → P be smooth maps. If X ∈ X(M )
is ϕ-related to Y ∈ X(N ) and if Y is ψ-related to Z ∈ X(P ) show that X is
ψ ◦ ϕ-related to Z.
Exercise 9.6. Let M ⊂ N be an embedded submanifold. A vector field X ∈ X(N )
is said to be tangent to M , if Xp ∈ Tp M for all p ∈ M . If X ∈ X(N ) and Y ∈ X(N )
are tangent to M show that also [X, Y ] ∈ X(N ) is tangent to M .
Exercise 9.7. (a) Let X ∈ X(M ) be a vector field. Show that a smooth curve

γ : I → M is an integral curve of X iff ∂t ∈ X(I) is γ-related to X.
(b) Let ψ : M → N be a smooth map, and assume that X ∈ X(M ) is ψ-related to
Y ∈ X(N ). If γ : I → M is an integral curve of X, prove that ψ ◦ γ : I → N is
an integral curve of Y .
Exercise 9.8. Calculate the integral curves of the following vector fields on R2
∂ ∂ ∂ ∂ ∂ ∂
X= + , Y =y − x , Z = x2 +y .
∂x ∂y ∂x ∂y ∂x ∂y
LECTURE 10

10.1. The flow of a vector field


Let us fix a vector field X on a smooth manifold M . In this section we will describe
the integral curves of X. Most results discussed here are in essence consequences
of Theorem 9.5.2. First let us remark:
Lemma 10.1.1. If γ : (a, b) → M is an integral curve of X then, for any s ∈ R,
also σ : (a − s, b − s) → M , σ(t) := γ(t + s) is an integral curve of X.
Proof. Let τs (t) := t + s. Then σ = γ ◦ τs , and
d(γ ◦ τs ) ∂ 
= (dt+s γ) ◦ dt (τs ) ∂ =
 
(t) =dt (γ ◦ τs )
dt ∂t t ∂t t
 ∂  dγ
=(dt+s γ) ◦ = (t + s) = Xγ◦τs (t) .
∂t t+s dt

Lemma 10.1.2. For any p ∈ M there exists an integral curve
γp : Ip → M
of X starting at p which is maximal in the sense that, any other integral curve
γ : I → M of X starting at p satisfies
I ⊂ Ip and γ = γp |I .
Proof. Consider two integral curves starting at p, denoted γ i : I i → M , i = 1, 2.
By continuity of the curves, the set
J := {t ∈ I 1 ∩ I 2 : γ 1 (t) = γ 2 (t)}
is closed in I 1 ∩ I 2 . Let s ∈ J. Note that σ i (t) := γ i (t + s), i = 1, 2, are two integral
curves which start at q := γ i (s); hence, in a chart around q, they are solutions to
the same ODE with the same initial condition. Thus, by the uniqueness assertion
in Theorem 9.5.2, we have that σ 1 = σ 2 in a neighborhood of 0; in other words
J contains a neighborhood of s. We have shown that J is both open and closed
in the open, non-empty inteval open I 1 ∩ I 2 , and since 0 ∈ J, this implies that
J = I 1 ∩ I 2 . In other words any two integral curves starting at p coincide on the
intersection of their domains.
85
86 IOAN MĂRCUT
, , MANIFOLDS

By Theorem 9.5.2, there is at least one integral curve γ : I → M starting at p.


Let Ip be the union of all intervals I for which there is an integral curve γ : I → M
starting at p. By the above, for any t ∈ Ip all integral curves starting at p and
defined at t ∈ Ip take the same value at t; we define γp (t) to be this common value.
Clearly, γp : Ip → M satisfies the maximality property. 

Note that the maximality property of the integral curve γp : Ip → M implies


its uniqueness. This allows us to define:
Definition 10.1.3. Let X be a vector field on a manifold M . The maximal
domain (or just the domain) of the flow of X is the set
D(X) := {(t, p) : p ∈ M, t ∈ Ip } ⊂ R × M,
and the maximal flow (or just the flow) of X is the map
φX : D(X) → M, (t, p) 7→ φtX (p) := γp (t).
For some t ∈ R, the flow at time t is the map
φtX : Dt (X) → M, p 7→ φtX (p),
where its domain is given by
Dt (X) := {p ∈ M : t ∈ Ip } ⊂ M.
Lemma 10.1.4. For s ∈ Ip , we have that t ∈ IφsX (p) iff t + s ∈ Ip ; and moreover,
φtX (φsX (p)) = φt+s
X (p) ∀ t ∈ IφsX (p) .
Proof. Denote q := φsX (p) and τs (t) := t + s. Then γp ◦ τs : τ−s (Ip ) → M is an
integral curve starting at q, therefore τ−s (Ip ) ⊂ Iq and γp ◦ τs = γq |τ−s (Ip ) ; i.e.
t+s
φX (p) = φtX (φsX (p)), ∀ t ∈ τ−s (Ip ).
Similarly, γq ◦ τ−s : τs (Iq ) → M is an integral curve starting at p, hence τs (Iq ) ⊂ Ip .
We obtain that Iq = τ−s (Ip ) , and this concludes the proof. 

We have that
Theorem 10.1.5. Let X be a vector field on M . Then the domain D(X) of X is
open in R × M and the flow of X is a smooth map
φX : D(X) → M, (t, p) 7→ φtX (p).
Proof. Theorem 9.5.2 implies that for every point p ∈ M there exists an open
neighborhood W ⊂ M and there exists  > 0 such that (−, ) × W ⊂ D(X), and
φX is smooth on (−, ) × W .
Let (t, p) ∈ D(X). Denote the image of the integral curve from p and φtX (p) by
C := {φsX (p) : s ∈ [0, t] (or s ∈ [t, 0])}.
Since C is compact, the argument above implies that there is some open set C ⊂
W ⊂ M and some  > 0 such that (−, ) × W ⊂ D(X) and φX is smooth on
t/n
(−, ) × W . Let n > 0 be such that t/n ∈ (−, ); then φX : W → M is smooth.
Consider the open sets {Wi }ni=0 defined inductively by
 −1
t/n
Wn := W, Wi := φX |W (Wi+1 ), 0 ≤ i ≤ n − 1.
LECTURE 10. 87

Since C ⊂ W , it follows, by induction on j = 0, . . . , n and by Lemma 10.1.4 that


t n−j
φX n (p) ∈ Wn−j ;
therefore, W0 is not empty, because it contains p. Also, directly by the definition
of the sets Wi and by Lemma 10.1.4, it follows that
ti
φXn (W0 ) ⊂ Wi .
In particular, we obtain that {t} × W0 ⊂ D(X). Moreover, φtX is smooth on W0 ,
because it is given by the composition of smooth maps:
t/n t/n
φtX |W0 = (φX |Wn ) ◦ . . . ◦ (φX |W0 ) : W0 −→ M.
On the other hand, since φX is smooth on (−, ) × W (in particular continuous),
there is an open set p ∈ V ⊂ W and there is 0 < δ <  such that φX ((−δ, δ) × V ) ⊂
W0 . Hence, (t − δ, t + δ) × V ⊂ D(X) is an open neighborhood of (t, p) on which
φX is smooth, since it is given by:
s−t
φsX |V = (φtX |W0 ) ◦ (φX |V ), s ∈ (t − δ, t + δ).
This finishes the proof. 
The following is a rather direct consequence of Theorem 10.1.5 and Lemma
10.1.4; the proof is left to the reader.
Corollary 10.1.6. For t ∈ R, Dt (X) is open in M and φtX is a diffeomorphism

between φtX : Dt (X) −→ D−t (X) with inverse φ−t ∼
X : D−t (X) −→ Dt (X).

10.2. Complete vector fields


Definition 10.2.1. A vector field X ∈ X(M ) is called complete, if its integral
curves are defined at all times; in other words, if D(X) = R × M .
Given a complete vector field X ∈ X(M ), for each t ∈ R, φtX is a diffeomorphism
of M . Moreover, this assignment satisfies
φsX ◦ φtX = φs+t
X ;
in other words, we obtain a group homomorphism:
(R, +) −→ (Diff(M ), ◦), t 7→ φtX .
Equivalently, one can think about the flow of a complete vector field as a smooth
action of the group (R, +) on M :
φX : R × M −→ M, (t, p) 7→ φtX (p).
The family {φtX }t∈R is also called a one-parameter group of diffeomor-
phisms, in the sense that it is a one-dimensional subgroup of the infinite-dimensional
group of diffeomorphisms.
The following result is useful for checking completeness:
Lemma 10.2.2. Let X ∈ X(M ) be a vector field. If there is  > 0 such that
(−, ) × M ⊂ D(X), then X is complete.
Proof. Let p ∈ M , and let t ∈ Ip . Then (−, ) ⊂ IφtX (p) and so, by Lemma 10.1.4,
(t − , t + ) ⊂ Ip . Hence, Ip = R. 
A similar argument implies the following behavior of integral curves:
88 IOAN MĂRCUT
, , MANIFOLDS

Lemma 10.2.3. Let X ∈ X(M ) be a vector field, and let p ∈ M . If there is a


compact set K ⊂ M such that φtX (p) ∈ K for all t ∈ Ip , then Ip = R (in other
words, an integral curve is either complete, or it leaves every compact subset).
Proof. Since K is compact, there is  > 0 such that (−, ) × K ⊂ D(X). So, for
q ∈ K, (−, ) ⊂ Iq . Thus if t ∈ Ip , then (−, ) ⊂ IφtX (p) , which by Lemma 10.1.4
implies that (t − , t + ) ⊂ Ip . We conclude that Ip = R. 
Either of the two lemmas, implies the following:
Corollary 10.2.4. Any vector field on a compact manifold is complete. More
generally, any compactly supported vector field on a manifold is complete.

10.3. The flow as the exponential of the Lie derivative


Definition 10.3.1. Let X ∈ X(M ) be a vector field. The Lie derivative along X
of a function f ∈ C ∞ (M ) is the smooth function
d
LX f := t=0 (φtX )∗ f ∈ C ∞ (M ).
dt
The Lie derivative along X of a vector field Y ∈ X(M ) is the vector field
d
LX Y := t=0 (φtX )∗ Y ∈ X(M ).
dt
Remark 10.3.2. Let us be more precise about the definitions above. First, recall
that the flow of X at time t is defined as a map φtX : Dt (X) → M and that Dt (X)
is an open subset of M . Thus, for f ∈ C ∞ (M ) and Y ∈ X(M ), we have
(φtX )∗ f = f ◦ φtX ∈ C ∞ (Dt (X)) and (φtX )∗ Y = (dφtX )−1 (YφtX ) ∈ X(Dt (X)).
Let p ∈ M . The maps Ip 3 t 7→ ((φtX )∗ f )(p) and Ip 3 t 7→ ((φtX )∗ Y )p are smooth
maps into the vector spaces R and Tp M , respectively. Therefore their derivatives
at t = 0 ∈ Ip are well-defined, and these give (LX f )(p) and (LX Y )p , respectively.
The following shows that the operators (φtX )∗ play the role of the exponential
of the operators tLX .
Proposition 10.3.3. Let X ∈ X(M ). For f ∈ C ∞ (M ) and Y ∈ X(M ), we have:
d t ∗
(φ ) f = (φtX )∗ ◦ X(f ) = X ◦ (φtX )∗ f ∈ C ∞ (Dt (X)),
dt X
d t ∗
(φ ) Y = (φtX )∗ [X, Y ] = [X, (φtX )∗ Y ] ∈ X(Dt (X)).
dt X
In particular,
LX (f ) = X(f ), LX Y = [X, Y ].
Proof. The first equality is equivalent to
d
f (φtX (p)) = XφtX (p) (f ), for all p ∈ Dt (X),
dt
which holds because φtX (p) is an integral curve of X. Using that φt+s t s
X = φX ◦ φX ,
and applying the first equality for s = 0, we obtain the second equality:
d t ∗ d ∗ d
(φX ) f = (φt+s
X ) f = (φs )∗ ((φtX )∗ f ) = X((φtX )∗ f ).
dt ds s=0 ds s=0 X
LECTURE 10. 89

Next, note that, for any g ∈ C ∞ (M ) and Z ∈ X(M ), we have that:

((φtX )∗ Z)g = (dφ−t −t t ∗ −t ∗



(*) X ) ◦ ZφtX g = ZφtX (g ◦ φX ) = (φX ) ◦ Z ◦ (φX ) g.

Taking the derivative of this equality, for Z = Y , and using the first part, we obtain:

d d
((φtX )∗ Y )g = (φtX )∗ ◦ Y ◦ (φ−t ∗

X ) g =
dt dt
= (φtX )∗ ◦ (X ◦ Y − Y ◦ X) ◦ (φ−t ∗
X ) g =
= (φtX )∗ ◦ [X, Y ] ◦ (φ−t ∗ t ∗
X ) g = ((φX ) [X, Y ])g,

where in the last step we have used again (∗) for Z = [X, Y ]. Since, g is arbitrary, we
obtain the third equality dt d
(φtX )∗ Y = (φtX )∗ [X, Y ]. Since [X, X] = 0, this implies
that (φtX )∗ X = X. Next, note that the vector fields (φtX )∗ X = X, (φtX )∗ Y and
(φtX )∗ [X, Y ] are φtX -related to the vector fields X, Y and [X, Y ], respectively. Thus,
by Lemma 9.3.3, and by the fact that φtX is a local diffeomorphism, (φtX )∗ [X, Y ] =
[X, (φtX )∗ Y ]. This implies also the last equality. 

10.4. The Lie bracket as infinitesimal commutator of flows


The Lie bracket [X, Y ] of two vector fields X, Y ∈ X(M ) can be regarded as the
infinitesimal obstruction for the flows of the vector fields to commute. Around
every point in M we can find an open set W and  > 0 such that the expression:

σp (t, s) := φ−s −t s t
Y ◦ φX ◦ φY ◦ φX (p) ∈ M,

is defined for (t, s, p) ∈ (−, ) × (−, ) × W . If X and Y are complete, then


σ(t, s) ∈ Diff(M ) is the group-theoretic commutator of the diffeomorphisms φtX
and φsY . Geometrically, σp (t, s) is obtained as follows: try to construct a “curved
parallelogram” starting at p with one pair of sides pq and nm in the direction of
X and of “length” t, and the other pair of sides qm and np0 parallel to Y and of
“length” s, as in the picture below:

m
n

p
q
p0

The black and red curves represent the flow lines of X and Y , respectively, and

q = φtX (p), m = φsX (q), n = φ−t 0 −s


X (m), p = φX (n) = σp (t, s).

In general, the curved parallelogram does not close up, i.e. p0 = σp (t, s) 6= p.
90 IOAN MĂRCUT
, , MANIFOLDS


On the other hand, since σ(t, 0) = idM , the derivative t 7→ ∂s σp (t, 0) is a
smooth vector field on M , which is given by:
d d −s
φY ◦ φ−t s t

σp (t, s) s=0 = X ◦ φY ◦ φX (p) s=0 =

ds ds
= −Yp + (dφ−t
X )(YφtX (p) ) =
t ∗

= (φX ) Y − Y p .

Applying Proposition 10.3.3, we obtain the following description of the Lie bracket:
Proposition 10.4.1. With the notation from above, we have that:
d d
t=0
σ(t, s) = [X, Y ].
dt ds s=0

10.5. Exercises
Exercise 10.1. Let ψ : M → N be a smooth map. Consider X ∈ X(M ) which is
ψ-related to Y ∈ X(N ). Prove that their flows are ψ-related in the following sense:
for every p ∈ M , Ip ⊂ Iψ(p) , and for all t ∈ Ip we have that

ψ(φtX (p)) = φtY (ψ(p)).


In other words (idR × ψ)(D(X)) ⊂ D(Y ), and the following diagram commutes:
idR ×ψ
D(X) / D(Y )
φX φY
 
M
ψ
/N

Exercise 10.2. Let X ∈ X(M ). For λ ∈ R, prove that the flow of λX is given by:
φtλX (p) = φλt
X (p),

and moreover that


(t, p) ∈ D(λX) ⇐⇒ (λt, p) ∈ D(X).
Exercise 10.3. Let X be a vector field on M . Let p ∈ M be such that Xp = 0.
Prove that Ip = R and that φtX (p) = p for all t ∈ R.
Exercise 10.4. On the m-dimensional sphere
m+1
X
S m = {(x1 , . . . , xm+1 ) ∈ Rm+1 : (xi )2 = 1},
i=1

consider the stereographical atlas:


A = {(U− , ϕ− ), (U+ , ϕ− )},

where U± = S m \{(0, . . . , 0, ±1)} and ϕ± : U± −→ Rm induce the coordinates
1 m
ϕ± = (y± , . . . , y± ):
i xi
y± := , 1 ≤ i ≤ m.
1 ∓ xm+1
LECTURE 10. 91

(a) Show that there exists a smooth vector field X on S m which in the coordinates
i
y+ is given by
m
i ∂
X
X|U+ = y+ i
∈ X(Rm ),
i=1
∂y +

i
and write the expression of X in the coordinates y− .
(b) Determine the flow of X in both charts.
e ∈ X(Rm+1 ) satisfying
(c) Prove that there exists a vector field X

X
ep = Xp , ∀ p ∈ Sm.

Exercise 10.5. (a) For any vector field X = f (x) ∂x ∈ X(R), prove that the vector

field sin(x)f (x) ∂x ∈ X(R) is complete (Hint: use Lemma 10.2.3).
(b) Prove that any vector field on R can be written as the sum of two complete
vector fields on R (Hint: use part (a)).
Pm ∂
Exercise 10.6. Let X = i=1 X i (x) ∂x i ∈ X(R
m
) be a bounded vector field, i.e.
there is a constant C > 0 such that |X (x)| < C for all x ∈ Rm and all 1 ≤ i ≤ m.
i

Prove that X is complete (Hint:


√ first, show that its flow at time t can displace a
point only by a distance of C m|t|; second, use an argument as in Lemma 10.2.3).
Exercise 10.7. For f ∈ C ∞ (M ) and X ∈ X(M ), prove that the Taylor series of
(φtX )∗ f at t = 0 is given by:
X Tn
exp(T LX )f := L n (f ) ∈ C ∞ (M )[[T ]],
n! X
n≥0

where C ∞ (M )[[T ]] denotes the ring of formal power series in T with coefficients in
C ∞ (M ).
Exercise 10.8. Let A be a real vector space endowed with a bilinear map
∗ : A × A −→ A.
We are not assuming any other axiom for the operation ∗ (like associativity, com-
mutativity, Jacobi identity etc.).
An automorphism of (A, ∗) is a linear isomorphism g : A → A satisfying
g(a ∗ b) = g(a) ∗ g(b).
Denote the space of all automorphisms by Aut(A, ∗).
A derivation of (A, ∗) is a linear map D : A → A satisfying:
D(a ∗ b) = D(a) ∗ b + a ∗ D(b), for all a, b ∈ A.
Denote the space of all derivations by Der(A, ∗).
(a) Show that (Aut(A, ∗), ◦) is a group.
(b) Show that, if D1 , D2 ∈ Der(A, ∗), then
[D1 , D2 ] := D1 ◦ D2 − D2 ◦ D1 ∈ Der(A, ∗).
(c) Prove that the operation [·, ·] on Der(A, ∗) satisfies the condition (1),(2) and
(3) of Proposition 9.2.1; i.e. (Der(A, ∗), [·, ·]) is a Lie algebra.
92 IOAN MĂRCUT
, , MANIFOLDS

For (d) and (e), assume A to be finite dimensional1 Hence, we can talk about
smooth curves g : (−, ) → LinR (A, A), where LinR (A, A) is the (finite dimensional)
vector space of linear maps from A to A. Identifying Tg(t) LinR (A, A) ∼= LinR (A, A),
we regard dg
dt (t) ∈ LinR (A, A).
(d) Consider a smooth curve g : (−, ) → LinR (A, A), such that g(0) = idA and
g(t) ∈ Aut(A, ∗), for all t ∈ (−, ).
dg
Prove that ∈ Der(A, ∗).
dt (0)
(e) Let D ∈ Der(A, ∗). Show that the following series converges absolutely to an
automorphism of (A, ∗)
X 1
exp(D) = Dn ∈ Aut(A, ∗).
n!
n≥0

For (f) and (g) assume (A, ∗) to be associative and commutative, however,
A can be infinite dimensional.
(f) For a ∈ A and D ∈ Der(A, ∗), prove that a ∗ D ∈ Der(A, ∗), where a ∗ D is
defined in the obvious way:
(a ∗ D)(b) := a ∗ (D(b)).
(g) Prove that the following Leibniz rule holds:
[D1 , a ∗ D2 ] = D1 (a) ∗ D2 + a ∗ [D1 , D2 ].

1This assumption simplifies the discussion; however, the results hold also in an infinite dimensional
setting, e.g. for bounded operators on a Banach space.
LECTURE 11

11.1. Lie groups


Definition 11.1.1. A Lie group is a manifold G endowed with a smooth group
structure, i.e. the multiplication and the inversion are smooth maps:
µ : G × G −→ G, (g, h) 7→ g · h,

ι : G −→ G, g 7→ g −1 .
A Lie subgroup of G is a subgroup H ⊂ G which is an immersed submanifold.
Examples 11.1.2. (1) Every at most countable group is a 0-dimensional Lie group.
(2) The additive group (Rn , +) is an n-dimensional Lie group.
(3) The multiplicative group (C∗ , ·) is a 2-dimensional Lie group. The following
are 1-dimensional subgroups R∗ ⊂ C∗ and S 1 ⊂ C∗ .
(4) The general linear group. The group GL(n) of linear automorphisms of
Rn is an n2 -dimensional Lie group. Its smooth structure comes by regarding
GL(n) as an open subset in the n2 -dimensional vector space M (n) of linear
maps from Rn to Rn .
(5) The special linear group. The group SL(n) is the subgroup of GL(n) con-
sists of linear automorphisms of Rn which have determinant 1.
(6) The orthogonal group. The group O(n) of orthogonal matrices (i.e. A ∈
M (n) such that AAt = I) is a compact Lie group of dimension n(n − 1)/2 (see
Exercise 6.11).
(7) There are complex versions of the groups above: GL(n, C) the group of in-
vertible complex matrices, with subgroup SL(n, C) consisting of matices with
determinant 1; O(n, C) the group of complex matrices satisfying AAt = I.
(8) The product G1 × . . . × Gk of Lie groups is a Lie group. In particular, the
k-torus T k := S 1 × . . . × S 1 is a k-dimensional compact Lie group.
(9) If G is any Lie group, then the connected component of the identity, denoted
by G◦ ⊂ G, is a Lie subgroup (which is both open and closed). In fact, G◦ is
a normal subgroup.
(10) The group GL(n)◦ consists of matrices with positive determinant.
(11) The special orthogonal group is the Lie group SO(n) := O(n)◦ .
(12) If G is a Lie group, then T G is a Lie group. Its structure maps are the differ-
entials of the structure maps of G.
93
94 IOAN MĂRCUT
, , MANIFOLDS

Definition 11.1.3. The left translation by g ∈ G is the diffeomorphism of G


λg : G −→ G, λg (h) := gh.
The right translation by g ∈ G is the diffeomorphism of G
ρg : G −→ G, ρg (h) := hg.
Using left translations one obtains the following:
Proposition 11.1.4. Every Lie group is a parallelizable manifold.
Proof. Using left translations, we construct a global frame on T G:
θL : T G −→ Te G, θL (X) := dg λg−1 (X), X ∈ Tg G.
By Corollary 8.4.3, the existence of θL1 implies that G is parallelizable. 
Definition 11.1.5. Let G be a Lie group. A vector field X ∈ X(G) is called
left invariant if X is λg -related to X, for all g ∈ G. We denote the space of
left invariant vector fields by XL (G). Similarly, one defines the space of right
invariant vector fields XR (G).
Explicitly, the condition that X ∈ X(G) is left invariant can be written as:
dh λg (Xh ) = Xgh , for all g, h ∈ G.
Letting h = e, this formula implies that X is determined by its value at the identity:
Xg = de λg (Xe ).
Conversely, the left invariant extension of v ∈ Te G is the vector field:
v L ∈ X(G), vgL := de λg (v).
That v L is left invariant follows by the chain rule:
dh λg (vhL ) = dh λg ◦ de λh (v) = de (λg ◦ λh )(v) = de λgh (v) = vgh
L
.
We conclude with the following:
Proposition 11.1.6. (1) The following map is a linear isomorphism:
X (G) − ∼
→ T G, X 7→ X , with inverse v 7→ v L .
L e e

(2) If X, Y ∈ XL (G), then [X, Y ] ∈ XL (G).


Proof. Part (1) was discussed above; part (2) follows from Lemma 9.3.3. 
Definition 11.1.7. A Lie algebra is a vector space g endowed with an operation:
[·, ·] : g × g −→ g,
called the Lie bracket, which
(1) is skew-symmetric:
[u, v] = −[v, u], u, v ∈ g,
(2) is R-bilinear:
[u, av + bw] = a[u, v] + b[u, w], a, b ∈ R, u, v, w ∈ g,
(3) and satisfies the Jacobi identity:
[u, [v, w]] + [v, [w, u]] + [w, [u, v]] = 0, u, v, w ∈ g.
1The map θ L is called the left Maurer-Cartan form of G.
LECTURE 11. 95

Definition 11.1.8. The Lie algebra of a Lie group G is the vector space
g := Te G
with Lie bracket:
[·, ·] : g × g → g, [v, w] := [v L , wL ]e .
By Proposition 9.2.1, (g, [·, ·]) satisfies indeed the axioms of a Lie algebra.
Remarks 11.1.9. (1) The Lie algebra g of a Lie group G plays the role of a
“linearized version” of G; it encodes infinitesimal information about G. In fact,
if G is simply connected, it can be fully recovered as a Lie group from the
structure of g (see for example [2]).
(2) In general Lie groups are denoted by capital Latin letters: G, H, K, L, P ,
Q, etc., while the corresponding Lie algebras are denoted by the corresponding
small Fraktur letters (also called Gothic letters): g, h, k, l, p, q etc. (in LaTeX
use \mathf rak{..}).
(3) One can define the Lie algebra of G also using right invariant vector fields;
nevertheless, one obtains an isomorphic object. The push-forward along the
inversion map gives an isomorphism between left invariant and right invariant
vector fields. Note however that, if both Lie algebras are identified with Te G
(by evaluating vector fields at e), one obtains opposite Lie brackets on Te G,
because the differential of the inversion map at e is −IdTe G .
Proposition 11.1.10. Left invariant vector fields are complete. Moreover,
φtX (gh) = gφtX (h), for all X ∈ XL (G), g, h ∈ G, t ∈ R.
Proof. Let X ∈ XL (G) and let g, h ∈ G. As in Lecture 10, let Ih denote the
maximal interval on which the flow line starting at h is defined. For t ∈ Ih , we
have:
d d 
gφtX (h) = dφtX (h) λg φtX (h) = dφtX (h) λg XφtX (h) = XgφtX (h) ,
 
dt dt
where we have used that X is left invariant. This shows that
Ih 3 t 7→ gφtX (h)
is an integral curve of X. Since the curve starts at gh, by Lemma 10.1.2, we
conclude that Ih ⊂ Igh and that gφtX (h) = φtX (gh). In particular, Ie ⊂ Ig for all
g ∈ G, and so, by Lemma 10.2.2, X is complete. 
The Proposition above allows us to define:
Definition 11.1.11. The exponential map of a Lie group G with Lie algebra g
is the map:
exp : g −→ G, v 7→ φ1vL (e).
Exercise 11.1 guides you through a possible proof that the exponential map is
indeed smooth. In fact, the exponential map is a local diffeomorphism around 0.
Proposition 11.1.12. The exponential map exp : g → G satisfies:
d0 exp = Idg ,
where we identify T0 g = g = Te G. Hence, exp is a diffeomorphism between a
neighborhood of 0 in g and a neighborhood of e in G.
96 IOAN MĂRCUT
, , MANIFOLDS

Proof. The result follows from the formula φtsvL (g) = φst
v L (g) of Exercise 10.2:

d d d
d0 exp(v) = =0
exp(v) = =0 φ1vL (e) = =0 φvL (e) = veL = v.
d d d


Next, we discuss maps between Lie groups and Lie algebras


Definition 11.1.13. (1) A Lie group homomorphism is a smooth group ho-
momorphism
F : (G, ·) −→ (H, ·).
(2) A homomorphism of Lie algebras is a linear map
f : (g, [·, ·]) −→ (h, [·, ·])
satisfying
f ([u, v]) = [f (u), f (v)], for all u, v ∈ g.
Lie group homomorphisms induce Lie algebra homomorphisms:
Proposition 11.1.14. Let (G, ·) and (H, ·) be two Lie groups with corresponding
Lie algebras (g, [·, ·]) and (h, [·, ·]). Consider a Lie group homomorphism:
F : (G, ·) −→ (H, ·).
Since F (e) = e, define:
f := de F : g = Te G −→ h = Te H.
Then f is a homomorphism of Lie algebras, and the following diagram commutes:

GO
F /H
O
exp exp

g
f
/h

i.e. F (exp(v)) = exp(f (v)) for all v ∈ g.


Proof. For v ∈ g, we show that v L and f (v)L are F -related. Since F is a Lie group
homomorphism, we have F ◦ λg = λF (g) ◦ F . Therefore,

dg F (vgL ) = dg F ◦ de λg (v) = de (F ◦ λg )(v) = de (λF (g) ◦ F )(v) =


= de λF (g) ◦ de F (v) = de λF (g) (f (v)) = f (v)L
F (g) .

So, for any w ∈ g, wL and f (w)L are also F -related, and so, by Lemma 9.3.3,
[v L , wL ] is F -related to [f (v)L , f (w)L ]. Evaluating this identity at the unit, implies
that f is a Lie algebra homomorphism:
[f (v), f (w)] = [f (v)L , f (w)L ]e = de F ([v L , wL ]e ) = f ([v, w]).
Finally, using again that v L is F -related to f (v)L , and Exercise 10.1, we obtain the
commutativity of the diagram:
F (exp(v)) = F (φ1vL (e)) = φ1f (v)L (F (e)) = exp(f (v)).

LECTURE 11. 97

11.2. Exercises
Exercise 11.1. Let G be a Lie group with Lie algebra g. Denote by v L ∈ XL (G)
the left invariant extension of a vector v ∈ g.
(a) Show that the following formula defines a smooth vector field on G × g:
ξ ∈ X(G × g), ξ(g,v) := (vgL , 0) ∈ Tg G ⊕ Tv g ∼
= T(g,v) (G × g).
(b) Prove that the flow of ξ is given by
φtξ (g, v) = (φtvL (g), v),
and conclude that ξ is complete.
(c) Use (b) to prove that the exponential map exp : g → G is smooth.
Exercise 11.2. Prove that left invariant vector fields commute with right invariant
vector fields:
X ∈ XL (G), Y ∈ XR (G) =⇒ [X, Y ] = 0.
Exercise 11.3. For k ≥ 1, denote by T k the k-dimensional torus, i.e.
T k = (S 1 )k = {(z1 , . . . , zk ) ∈ Ck : |z1 | = . . . = |zk | = 1}.
(a) Prove that left invariant vector fields on the k-torus are the same as right
invariant vector fields:
XL (T k ) = XR (T k ).
Moreover, if θi , 1 ≤ i ≤ k denote the “angle coordinates” on T k (see Example
8.5.5), prove that the following is a basis of XL (T k ):
∂ ∂
,..., ∈ X(T k ).
∂θ1 ∂θk
(b) Calculate the exponential map:
exp : Te T k −→ T k .
Hint: use the basis from (a) for Te T k .
(c) Consider an n × m matrix with integer coefficients A = {ai,j ∈ Z}1≤i≤n,1≤j≤m .
Show that the map
FA : (T m , ·) −→ (T n , ·),
a a1,m a an,m
FA (z1 , . . . , zm ) = (z1 1,1 . . . zm , . . . , z1 n,1 . . . zm )
is a Lie group homomorphism.
(d) Show that any Lie group homomorphism F : T m → T n is of the form described
in (c). Hint: use Proposition 11.1.14.
Exercise 11.4. Consider the Lie group GL(n) of invertible n×n-matrices. Denote
the space of all n × n-matrices by gl(n). Since GL(n) is an open subset of gl(n),
we identify:
Tg GL(n) = gl(n), for all g ∈ GL(n).
(a) Prove that, with the above identification, the differential of left translation is
simply given by matrix multiplication:
dh λg : Th GL(n) = gl(n) −→ Tgh GL(n) = gl(n),
dh λg (v) = g · v.
98 IOAN MĂRCUT
, , MANIFOLDS

(b) Prove that the left invariant extension of v ∈ TI GL(n) = gl(n) is the vector
field
v L ∈ XL (GL(n)), vgL = g · v ∈ Tg GL(n) = gl(n).
(c) Prove that the flow of the left invariant extension v L ∈ XL (GL(n)) of v ∈
TI GL(n) = gl(n) is given by
φtvL (g) = g · etv ,
n
where ew = n≥0 wn! is the matrix exponential.
P

(d) Using Proposition 10.4.1 and (c) prove that the Lie bracket on gl(n) is the
commutator of matrices:
[v, w] = v · w − w · v, v, w ∈ gl(n).
Exercise 11.5. The quaternions are the real 4-dimensional associative algebra
H with basis 1, i, j, k, where the multiplication is determined by the rules:
i2 = j 2 = k 2 = −1, ij = k, jk = i, ki = j, ji = −k, kj = −i, ik = −j.
One can also represent H as the space of all 2 × 2-matrices with complex entries of
the form:  
z w
, z, w ∈ C.
−w z
The correspondence between the two descriptions is given by
     
i 0 0 1 0 i
i ←→ , j ←→ , k ←→ .
0 −i −1 0 i 0
(a) Prove that every element in H∗ = H\{0} is invertible, and conclude the H∗ is
a 4-dimensional Lie group.
(b) Prove that the map following map is a Lie group homomorphism:
F : (H∗ , ·) −→ (R>0 , ·), F (a + bi + cj + dk) = a2 + b2 + c2 + d2 .
(c) Prove that G := F −1 (1) is a Lie subgroup of H∗ which as a manifold is diffeo-
morphic to S 3 .
(d) Identifying T1 H∗ = H, prove that the exponential map of H∗ is given by the
usual exponential series:
X An
exp : H −→ H∗ , A 7→ .
n!
n≥0

(e) Prove that the Lie algebra of G is given by


g = T1 G = {bi + cj + dk : b, c, d ∈ R}.
(f) Consider the set of “square roots of the unity” in H:
Σ := {X ∈ H : X 2 = −1}.
Prove that Σ ⊂ g, and that Σ is diffeomorphic to S 2 .
(g) Prove that every element in A ∈ H can be decomposed as
A = λ + µX, λ, µ ∈ R, X ∈ Σ.
(h) In the decomposition A = λ + µX from (g), prove that A ∈ G iff λ2 + µ2 = 1.
(i) In the decomposition A = λ + µX from (g), prove that
exp(A) = eλ (cos(µ) + sin(µ)X).
LECTURE 11. 99

(j) On g we consider the metric such that Σ is the unit sphere around the origin,
and denote by Br ⊂ g the open ball of radius r > 0. Using the previous items,
prove the following about the exponential map exp : g → G:
(i) exp is a diffeomorphism between Bπ and G\{−1};
(ii) exp sends the spheres of radii 2kπ to 1 ∈ G, and the spheres of radii
(2k + 1)π to −1 ∈ G;
(iii) exp restricts to a diffeomorphism between B(k+1)π \B kπ and G\{1, −1}.
(k) Show that every flow line of a left invariant vector field on G is a circle.
LECTURE 12

12.1. The exterior algebra


There are (at least) two important algebras associated to any vector space V . First,
the algebra S(V ∗ ) of polynomial functions on V , which is also called the symmetric
algebra of V ∗ . One can construct S(V ∗ ) as the algebra generated by the elements
of V ∗ subject to the relations that these generators commute (in fact this gives
a presentation of S(V ∗ ) in the realm of unital associative algebras). The second
algebra, which will beVconstructed in thisVsection, is called the exterior algebra of
V ∗ and is denoted by V ∗ . The algebra V ∗ is the algebra generated by elements
of V ∗ where these generators are subject to the condition that they anti-commute,
i.e. αβ = −βα.
Definition 12.1.1. An alternating k-form on a vector space V is a map
ω : V × . . . × V −→ R
| {z }
k × V
which is multilinear
ω(v1 , . . . ,vi−1 , av + bw, vi+1 , . . . , vk ) =
= aω(v1 , . . . , vi−1 , v, vi+1 , . . . , vk ) + bω(v1 , . . . , vi−1 , w, vi+1 , . . . , vk ),
and satisfies
ω(v1 , . . . , vi , vi+1 , . . . , vk ) = −ω(v1 , . . . , vi+1 , vi , . . . , vk ).
Vk ∗
The space of alternating k-forms on V is denoted by V and is called the

V1 ∗ ∗
V0 ∗
k-th exterior power of V . We have that V = V , and we define V := R.
Vk ∗
Fixing v ∈ V to be the first argument of ω ∈ V , we obtain an alternating
Vk−1 ∗
k − 1 form, denoted by ιv ω ∈ V ,
(ιv ω)(v1 , . . . , vk−1 ) := ω(v, v1 , . . . , vk−1 ).
The operation ω 7→ ιv ω is called the interior product by v, or the contraction
V0 by
v. For k = 1 and η ∈ V ∗ , we have that ιv η = η(v). For k = 0 and a ∈ R = V ∗,
we set ιv a := 0.
101
102 IOAN MĂRCUT
, , MANIFOLDS

Remark 12.1.2. The symmetric group Sk consists of permutations of the set


{1, . . . , k}. Recall that Sk is generated by the transpositions i ↔ (i + 1). If σ ∈ Sk
can be written as a product of p transpositions, then the parity of p is independent
of this decomposition, and the sign of σ is defined as sign(σ) = (−1)p . This implies
that an alternating k-from ω satisfies
ω(vσ(1) , . . . , vσ(k) ) = sign(σ)ω(v1 . . . , vk ),
for all σ ∈ Sk . In particular, ω returns 0 whenever two entries of coincide:
ω(. . . , v, . . . , v, . . .) = 0.
Exercise 12.2 gives a more general version of this last observation.
Vk ∗ Vl ∗
The tensor product of α ∈ V and β ∈ V is the multilinear map:
α ⊗ β : V × . . . × V −→ R,
| {z }
(k + l) × V
(α ⊗ β)(v1 , . . . , vk ,w1 , . . . , wl ) := α(v1 , . . . , vk )β(w1 , . . . , wl ).
Note that α ⊗ β is not alternating: if we permute the vi ’s among themselves,
or the wj ’s among themselves, the sign does change correspondingly, however, if
the vi ’s and the wj ’s get mixed up, one obtains a completely different outcome.
For example, if k = l = 1, then α(v)β(w) and α(w)β(v) can be quite different.
Nevertheless, note that their difference is alternating:
(α ⊗ β − β ⊗ α)(v, w) = α(v)β(w) − α(w)β(v).
This operation can be extended to all degrees k, l ≥ 1, by considering the sum (with
signs) of all possible ways the vi ’s and the wj ’s can mix. To make this precise we
introduce the following notion:
Definition 12.1.3. A (k, l)-shuffle is a permutation σ ∈ Sk+l which is increasing
on the first k integers and on the last l integers:
σ(1) < σ(2) < . . . < σ(k), σ(k + 1) < σ(k + 2) < . . . < σ(k + l).
We denote the set of all (k, l)-shuffles by S(k, l).

A (k, l)-shuffle for k + l = 52.


Definition 12.1.4. The wedge product (or exterior product) is the following
operation on alternating forms:
k
^ l
^ k+l
^
∗ ∗
∧: V × V −→ V ∗,
LECTURE 12. 103
X
(α ∧ β)(v1 , . . . , vk+l ) = sign(σ)α(vσ(1) , . . . , vσ(k) )β(vσ(k+1) , . . . , vσ(k+l) ).
σ∈S(k,l)
V0 Vk
For a ∈ R = V ∗ and α ∈ V ∗ note that:
a ∧ α = aα = α ∧ a.
It is not obvious from the definition that α ∧ β is alternating; this will be shown
this in the proof of the following:
Theorem 12.1.5. The wedge product has the following properties:
(1) the interior product with any v ∈ V is a graded derivation:
ιv (α ∧ β) = (ιv α) ∧ β + (−1)k α ∧ (ιv β);
(2) the wedge product is graded commutative:
α ∧ β = (−1)kl β ∧ α,
(3) the wedge product is associative:
(α ∧ β) ∧ γ = α ∧ (β ∧ γ),
Vk Vl Vm ∗
for all α ∈ V ∗, β ∈ V ∗ and γ ∈ V .
Moreover, if n = dim(V ∗ ) and {e1 , . . . , en } is a basis of V ∗ , then the set
 i1
e ∧ ei2 ∧ . . . ∧ eik : 1 ≤ i1 < i2 < . . . < ik ≤ n

Vk ∗
forms a basis of V . Hence
k   n = n!
k!(n−k)! , 0 ≤ k ≤ n
^
∗ k
dim V =
0, n < k.
Proof. First we prove (1). Note that the set of (k, l)-shuffles admits the disjoint
decomposition S(k, l) = S 0 (k, l) t S 00 (k, l), where
S 0 (k, l) = {σ ∈ S(k, l) : σ(1) = 1} S 00 (k, l) = {σ ∈ S(k, l) : σ(k + 1) = 1}.
Therefore,
(ιv1 (α ∧ β))(v2 , . . . , vk+l ) = (α ∧ β)(v1 , . . . , vk+l ) =
X
= sign(σ)α(vσ(1) , . . . , vσ(k) )β(vσ(k+1) , . . . , vσ(k+l) ) =
σ∈S(k,l)
X
= sign(σ 0 )(ιv1 α)(vσ0 (2) , . . . , vσ0 (k) )β(vσ0 (k+1) , . . . , vσ0 (k+l) )+
σ 0 ∈S 0 (k,l)
X
+ sign(σ 00 )α(vσ00 (1) , . . . , vσ00 (k) )(ιv1 β)(vσ00 (k+2) , . . . , vσ00 (k+l) ).
σ 00 ∈S 00 (k,l)

Next, note that there are one-to-one correspondences


θ ∈ S(k − 1, l) ↔ σ 0 ∈ S 0 (k, l) η ∈ S(k, l − 1) ↔ σ 00 ∈ S 00 (k, l)
 00
0 σ (i) − 1, 1≤i≤k
θ(i) = σ (i + 1) − 1 η(i) =
σ 00 (i + 1) − 1, k + 1 ≤ i ≤ k + l − 1
104 IOAN MĂRCUT
, , MANIFOLDS

and moreover, that sign(σ 0 ) = sign(θ) and sign(σ 00 ) = (−1)k sign(η). Using this
fact, the above formula becomes:
(ιv1 (α ∧ β))(v2 , . . . , vk+l ) =
X
= sign(θ)(ιv1 α)(vθ(1)+1 , . . . , vθ(k)+1 )β(vθ(k+1)+1 , . . . , vθ(k+l)+1 )+
θ∈S(k−1,l)
X
(−1)k sign(η)α(vη(1)+1 , . . . , vη(k)+1 )(ιv1 β)(vη(k+1)+1 , . . . , vη(k+l−1)+1 ) =
η∈S(k,l−1)

= (ιv1 α ∧ β)(v2 , . . . , vk+l ) + (−1)k (α ∧ ιv1 β)(v2 , . . . , vk+l ).


This proves (1).
Vk ∗
Next, we prove by induction on n := k + l ≥ 0 that, for all α ∈ V
Vl ∗
and β ∈ V , α ∧ β is alternating. Clearly, this holds for n = 0. Assuming the
Vn−1 ∗
induction hypothesis for n−1, it follows that (ιv1 α)∧β+(−1)k α∧(ιv1 β) ∈ V .
Hence, if we switch vi and vi+1 in (α ∧ β)(v1 , . . . , vk+l ), for 1 < i < n − 1, the result
changes the sign. For v1 and v2 , we need to check that
ιv1 ιv2 (α ∧ β) = −ιv2 ιv1 (α ∧ β).
For this, we apply the derivation rule twice, and use that α and β are alternating:
ιv2 ιv1 (α ∧ β) =ιv2 (ιv1 α) ∧ β + (−1)k α ∧ (ιv1 β) =


=(ιv2 ιv1 α) ∧ β + (−1)k−1 (ιv1 α) ∧ (ιv2 β)+


+ (−1)k (ιv2 α) ∧ (ιv1 β) + α ∧ (ιv2 ιv1 β) =
= − (ιv1 ιv2 α) ∧ β − (−1)k (ιv1 α) ∧ (ιv2 β)+
− (−1)k−1 (ιv2 α) ∧ (ιv1 β) − α ∧ (ιv1 ιv2 β) =
= − ιv1 ιv2 (α ∧ β),
where in the last step we have performed first two steps backwards, with v1 and
v2 interchanged. This finishes the inductive argument, and proves that the wedge
product of alternating forms is indeed alternating.
Properties (2) and (3) are proven similarly, by induction on n = k + l and
n = k + l + m, respectively, and by using the derivation rule. We will prove only
(2), and leave it to the reader to check (3). Clearly, (2) holds for n = k + l = 0.
Vk ∗ Vl ∗
Assume that it holds for n − 1. Let α ∈ V and β ∈ V . For v ∈ V , we have
ιv (α ∧ β) = (ιv α) ∧ β + (−1)k α ∧ (ιv β) =
= (−1)(k−1)l β ∧ (ιv α) + (−1)k+k(l−1) (ιv β) ∧ α =
= (−1)kl (−1)l β ∧ (ιv α) + (ιv β) ∧ α =


= (−1)kl ιv (β ∧ α),
where we have used the induction hypothesis in the second line. Since this holds
for all v, (2) follows.
For the last part, let e1 , . . . , en ∈ V be the dual basis, i.e. these elements are
defined by ei (ej ) = δji , where δji = 1 if i = j and δji = 0 if i 6= j. By repeatedly
applying the derivation rule, one can easily show that, for 1 ≤ i1 < . . . < ik ≤ n
and 1 ≤ j1 < . . . < jk ≤ n, the following holds:
ei1 ∧ . . . ∧ eik (ej1 , . . . , ejk ) = δji11 . . . δjikk ,

LECTURE 12. 105

in other words the result is zero unless ip = jp , for all 1 ≤ p ≤ k, and then the
result is 1.
Consider now a linear combination:
X k
^
ω= ai1 ...ik ei1 ∧ . . . ∧ eik ∈ V ∗.
i1 <...<ik

Then for 1 ≤ j1 < . . . < jk ≤ n, we have that ω(ej1 , . . . , ejk ) = aj1 ...jk . Thus if
ω = 0, then all coefficients aj1 ...jk = 0. This shows that the k-forms ei1 ∧ . . . ∧ eik
are linearly independent.
Vk ∗ Vk ∗
To show that these forms span V , let ω ∈ V . Consider the k-form:
X k
^
η := ω − ω(ei1 , . . . , eik )ei1 ∧ . . . ∧ eik ∈ V ∗.
i1 <...<ik

Then η(ej1 , . . . , ejk ) = 0 for all 1 ≤ j1 < . . . < jk ≤ n. Since η is alternating,


this implies that, η(el1 , . . . , elk ) = 0, for all l1 , . . ., lk ∈ {1, . . . , n} (see Remark
Pn Finally, consider k vectors v1 , . . . , vk ∈ V . By writing these in the basis
12.1.2).
vi = l=1 vil el , and using that η is multilinear, we obtain:
n
X n
X
η(v1 , . . . , vk ) = ... v1l1 . . . vklk η(el1 , . . . , elk ) = 0.
l1 =1 lk =1

Thus η = 0, and therefore:


X
ω= ω(ei1 , . . . , eik )ei1 ∧ . . . ∧ eik .
i1 <...<ik

We conclude that the set {ei1 ∧ . . . ∧ eik : 1 ≤ i1 < . . . < ik ≤ n} forms a basis of
Vk ∗
V . Note that sequences of the form 1 ≤ i1 < . . . < ik ≤ n are in one-to-one
correspondence with subset I = {i1 , . . . , ik } of {1, . . . , n} with exactly k elements;
Vk
this implies that dim( V ∗ ) = nk .


As a consequence of the graded commutativity, we obtain:
Vk ∗
Corollary 12.1.6. If k is odd and ω ∈ V then ω 2 = ω ∧ ω = 0.
Putting the exterior powers of V ∗ together, one obtains the exterior algebra:
Definition
V ∗ 12.1.7. Let V be a vector space. The exterior algebra of V ∗ , denoted

by ( V , +, ∧), is the direct sum of the exterior products of V
^ k
M^
V ∗ := V ∗,
k≥0

endowed with the bilinear extension of the wedge product


^ ^ ^
∧: V ∗ × V ∗ −→ V ∗.
The exterior algebra is an associative unital algebra of dim( V ∗ ) = 2dim(V ) .
V
V0 ∗
Remark 12.1.8. The unit of V ∗ is just 1 ∈ R =
V
V . TheV “bilinear extension
of the wedge product” is meant in the obvious way: if α, β ∈ V ∗ decompose as
i
^
α = α0 + α1 + . . . + αk , αi ∈ V ∗,
106 IOAN MĂRCUT
, , MANIFOLDS

j
^
β = β 0 + β 1 + . . . + βl , βj ∈ V ∗,
then their wedge product is given by:
X s
^
α ∧ β = γ0 + γ1 + . . . + γk+l , γs := αi ∧ βj ∈ V ∗.
i+j=s

Definition 12.1.9. Let V and W be vector space, and let A : V → W be a linear


map. The pullback map induced by A is the algebra homomorphism
^  ^ 
A∗ : W ∗ , +, ∧ −→ V ∗ , +, ∧ ,

defined by
k
^ k
^
A∗ : W ∗ −→ V ∗ , (A∗ ω)(v1 , . . . , vk ) := ω(Av1 , . . . , Avk ),
V0 V0
for k ≥ 1, and for a ∈ W ∗ = R, A∗ a = a ∈ V ∗ = R.

It can be easily checked that A∗ is indeed an algebra homomorphism:


A∗ (α ∧ β) = (A∗ α) ∧ (A∗ β).
Vdim(V ) ∗
The top exterior power of V ∗ , i.e. the 1-dimensional vector space V ,
is also called the determinant of V ∗ , because of the following:
Vdim(V ) ∗
Proposition 12.1.10. For any linear map A : V → V and any ω ∈ V :
A∗ (ω) = det(A)ω;
explicitly, for v1 , . . . , vn ∈ V , where n = dim(V ),
ω(Av1 , . . . , Avn ) = det(A)ω(v1 , . . . , vn ).

Remark 12.1.11. Note that (k, l)-shuffles are in one-to-one correspondence with
pairs of subsets (A, B) of the set {1, . . . , k + l}, such that:
{1, . . . , k + l} = A ∪ B, |A| = k, |B| = l.
The element σ ∈ S(k, l) corresponds to the pair
A = {σ(1), . . . , σ(k)}, B = {σ(k + 1), . . . , σ(k + l)}.
Conversely, given such a pair (A, B), one first orders the elements of the two sets:
A = {a1 < a2 < . . . < ak }, B = {b1 < b2 < . . . < bl },
and then the corresponding σ is given by:
σ(i) := ai , 1 ≤ i ≤ k; σ(k + j) := bj , 1 ≤ j ≤ l.
In particular, we have that
 
k+l (k + l)!
|S(k, l)| = = .
k k! l!
LECTURE 12. 107

12.2. The cotangent bundle


Definition 12.2.1. Let M be a manifold. The cotangent space at p ∈ M , denoted
by Tp∗ M , is defined as the linear dual of the tangent space Tp M ; i.e. Tp∗ M consists
of all linear maps α : Tp M → R. Elements of Tp∗ M are called cotangent vectors.
All the cotangent spaces together form the cotangent bundle of M :
G
T ∗ M := Tp∗ M.
p∈M

Example 12.2.2. Let f ∈ C ∞ (M ) be a smooth function. The differential of f at


p ∈ M is a cotangent vector dp f ∈ Tp∗ M at p
dp f : Tp M −→ R, dp f (v) := v(f ),
where, as usually, we identify Tf (p) R = R.
Let (U, ϕ = (x1 , . . . , xm )) be a coordinates chart on M 1. At p ∈ U , we have:
 ∂  ∂xi 
i i 1, if i = j,
dp x = (p) = δj =
∂xj p ∂xj 0, if i 6= j.
This shows that dp x1 , . . . , dp xm is a basis of the cotangent space Tp∗ M dual to the
∂ ∂
basis ∂x 1 |p , . . . , ∂xm |p of Tp M .

Theorem 12.2.3. The cotangent bundle π : T ∗ M → M is a vector bundle over M


of rank equal to m = dim(M ). The vector bundle structure is uniquely determined
by the condition that, for every chart (U, ϕ = (x1 , . . . , xm )) on M , the differentials
of the coordinate functions:
dx1 , . . . , dxm : U −→ T ∗ M |U , p 7→ dp xi ∈ Tp∗ M,
form a local frame on T ∗ M .
The above result can be proven exactly like Theorem 8.1.2, therefore we omit
the proof. However, it is instructive to write down the transition functions from
one frame to another. Given a second set of coordinates (V, ψ = (y 1 , . . . , y m )), for
p ∈ U ∩ V , we have that:
m
X ∂xi
dxi |p = j
(p)dy j |p ,
j=1
∂y


which can be easily seen by applying to both sides the basis vectors ∂y k p
.

12.3. The exterior powers of the cotangent bundle


Next, we consider the exterior powers of the cotangent spaces.
Definition 12.3.1. Let M be a manifold, and let p ∈ M . The k-th exterior
Vk ∗
power of the cotangent space at p is denote by Tp M . The collection of all these
vector spaces is the k-th exterior power of the cotangent bundle:
k
^ k
G ^
T ∗ M := Tp∗ M.
p∈M

1From this lecture on, we use simplified notation: instead of xi we denote the coordinates by xi .
ϕ
108 IOAN MĂRCUT
, , MANIFOLDS

Consider a chart (U, ϕ = (x1 , . . . , xm )) on M . By Theorem 12.1.5, for each


Vk ∗
p ∈ U , a basis of Tp M is given by the elements:
dxi1 ∧ . . . ∧ dxik |p , 1 ≤ i1 < . . . < ik ≤ m.
Generalizing the case k = 1 of the cotangent bundle, when choosing a second chart
(V, ψ = (y 1 , . . . , y m )), on U ∩ V the frame changes as follows:
X ∂xi1 ∂xik l1
dxi1 ∧ . . . ∧ dxik = . . . dy ∧ . . . ∧ dy lk ,
∂y l1 ∂y lk
l1 ...lk

where the sum runs over all 1 ≤ l1 , . . . , lk ≤ m. After ordering the indices l1 , . . . , lk
increasingly, and dropping the terms where some indexes are repeated (are therefore
are zero), we can rewrite:
X
dxi1 ∧ . . . ∧ dxik = cji11,...,i j1 jk
,...,jk dy ∧ . . . ∧ dy ,
k

j1 <...<jk

where the coefficients are given explicitly by


X ∂xi1 ∂xik ∂(xi1 . . . xik )
cij11,...,ik
,...,jk = sign(σ) jσ(1) . . . jσ(k) = ,
∂y ∂y ∂(y j1 . . . y jk )
σ∈S k

where the notation on the right is the standard notation for the minor consisting of
∂xi m
the rows i1 , . . . , ik and columns j1 , . . . , jk in the Jacobian matrix ( ∂y j )i,j=1 of the

change of coordinates map ψ ◦ ϕ−1 , i.e.


∂(xi1 . . . xik )  ∂xia k 
= det .
∂(y j1 . . . y jk ) ∂y jb a,b=1
The precise formula is not that important; what matters is that these coefficients
are smooth, because this allows us to apply the same argument as in the proof of
Theorem 8.1.2, and conclude:
Theorem 12.3.2. The k-th exterior power of the cotangent bundle
k
^
π: T ∗ M −→ M
is a vector bundle over M of rank equal to m

k , where
Vk ∗
m = dim(M ). A chart
1 m
(U, ϕ = (x , . . . , x )) on M induces a local frame on T M over U
dxi1 ∧ . . . ∧ dxik , 1 ≤ i1 < . . . < ik ≤ m;
moreover, these frames define the vector bundle structure uniquely.

12.4. Differential forms


Definition 12.4.1. A differential form of degree k (or a k-form) on M is a
Vk ∗
section T M . The space of differential k-forms is denoted by
k
^ 
k
Ω (M ) := Γ T ∗M .

For k = 0, Ω0 (M ) = C ∞ (M ). The direct sum of these vector spaces is denoted by


M
Ω(M ) := Ωk (M ).
k≥0
LECTURE 12. 109

A differential form on M is an element


ω ∈ Ω(M ), ω = ω0 + . . . + ωm , ωi ∈ Ωi (M ),
where m = dim(M ).
The pointwise wedge product induces the structure of an associative unital
algebra on the space of differential forms:
(Ω(M ), +, ∧), (α ∧ β)p := αp ∧ βp ,
which is graded commutative:
α ∧ β = (−1)kl β ∧ α, α ∈ Ωk (M ), β ∈ Ωl (M ).
The interior product with a vector field X ∈ X(M ) is the linear operator:
ιX : Ωk (M ) −→ Ωk−1 (M ), (ιX ω)p := ιXp (ωp ),
which is a graded derivation for the wedge-product:
ιX (α ∧ β) = (ιX α) ∧ β + (−1)k α ∧ (ιX β),
for α ∈ Ωk (M ) and β ∈ Ω(M ).
Given a local chart (U, ϕ) with coordinates ϕ = (x1 , . . . , xm ), a differential
k-form ω ∈ Ωk (M ) can be written as
X
ω|U = ωi1 ...ik dxi1 ∧ . . . ∧ dxik ,
i1 <...<ik

with smooth coefficients ωi1 ...ik ∈ C ∞ (U ). The notation where the indexes are not
ordered is also useful:
X 1
ω|U = ωi1 ...ik dxi1 ∧ . . . ∧ dxik ,
i ...i
k!
1 k

where the coefficients are defined to be fully skew-symmetric:


for all σ ∈ Sk .
ωiσ(1) ...iσ(k) = sign(σ)ωi1 ...ik ,
Pm ∂
In this notation the interior by a vector fields X|U = i=1 X i ∂x i is given by

X 1
(ιX ω)|U = X i1 ωi1 ...ik dxi2 ∧ . . . ∧ dxik .
i1 ...ik
(k − 1)!

12.5. The pullback of differential forms


Differential forms can be pulled back via any smooth maps:
Definition 12.5.1. The pullback along a smooth map f : M → N is defined by:
f ∗ : Ωk (N ) −→ Ωk (M ), (f ∗ ω)p := (dp f )∗ ωf (p) ,
that is, for v1 , . . . , vk ∈ Tp M :
(f ∗ ω)p (v1 , . . . , vk ) = ωf (p) (dp f (v1 ), . . . , dp f (vk )).
The pullback satisfies the following:
110 IOAN MĂRCUT
, , MANIFOLDS

Proposition 12.5.2. The pullback along a smooth map f : M → N is an algebra


homomorphism:
f ∗ : (Ω(N ), +, ∧) −→ (Ω(M ), +, ∧) ,
i.e. for all α, β ∈ Ω(N ) and all a, b ∈ R it satisfies:
f ∗ (α ∧ β) = f ∗ (α) ∧ f ∗ (β),
f ∗ (aα + bβ) = af ∗ (α) + bf ∗ (β).
Moreover, if g : N → P is also a smooth map, then f ∗ ◦ g ∗ = (g ◦ f )∗ .

12.6. Exercises
Exercise 12.1. Prove the following formulas for the wedge product:
1 X
(α ∧ β)(v1 , . . . , vk+l ) = sign(σ)α(vσ(1) , . . . , vσ(k) )β(vσ(k+1) , . . . , vσ(k+l) ).
k!l!
σ∈Sk+l

Exercise 12.2. Let v1 , . . . , vp be p linearly dependent vectors in V . Show that for


Vk ∗
any alternating form ω ∈ V , we have that
ιv1 . . . ιvp ω = 0.
Exercise 12.3. Let α1 , . . . , αk ∈ V ∗ , and v1 , . . . , vk ∈ V . Prove that:
k
(α1 ∧ . . . ∧ αk )(v1 , . . . , vk ) = det (αi (vj ))i,j=1 .
Exercise 12.4. Prove Proposition 12.1.10.
Exercise 12.5. Denote the product of k-copies of V by V ×k . For a k × k-matrix
R = {rij }ki,j=1 , consider the following map:
k
X k
X 
RV : V ×k −→ V ×k , RV (v1 , . . . , vk ) := r1j vj , . . . , rkj vj .
j=1 j=1
×k
Prove that a map ω : V → R is an alternating k-form if and only if, for every
k × k-matrix R:
ω ◦ RV = det(R) · ω.
V2 ∗
Exercise 12.6. The rank of a 2-form ω ∈ V is the largest even integer
2r such that ω r 6= 0. Prove that ω has rank 2r if and only if V ∗ has a basis
e1 , . . . , er , f 1 , . . . , f r , g 1 , . . . , g c such that
ω = e1 ∧ f 1 + . . . + er ∧ f r .
Exercise 12.7. Let V be a vector space of dimension n. Let v ∈ V and ξ ∈ V ∗
such that ξ(v) = 1. Consider the operators: interior product by v
^ ^
ιv : V ∗ −→ V ∗ , ω 7→ ιv ω,
and exterior products by ξ
^ ^
eξ : V ∗ −→ V ∗ , ω 7→ ξ ∧ ω.
(a) Prove that these operators satisfy:
eξ ◦ ιv + ιv ◦ eξ = IdV V ∗ .
(b) For ω ∈ V V , prove ιv ω = 0 if and only if there α ∈ V V ∗ such that ω = ιv α.
V ∗ V
(c) For ω ∈ V ∗ , prove ξ ∧ω = 0 if and only if there β ∈ V ∗ such that ω = ξ ∧β.
LECTURE 12. 111

dimension n. An element α ∈ V ∗
V
Exercise 12.8. Let V be a vector space V of
is called invertible if there exists β ∈ V ∗ such that α ∧ β = 1. Prove that an
element
^ i
^
α = α0 + α1 + . . . + αn ∈ V ∗ , αi ∈ V∗
1
is invertible iff α0 6= 0 (Hint: think about the geometric series 1−q = 1+q +q 2 +. . .).
Exercise 12.9. Consider the differential forms, the vector field and the map below:
2
α = cos(y)dx + dz − e−w dw ∈ Ω1 (R4 ),
β = x2 dx ∧ dy + w7 dw ∧ dz ∈ Ω2 (R4 ),
γ = w2 dz + dz ∧ dy ∧ (dw + ydw ∧ dx) ∈ Ω(R4 ),
∂ ∂ ∂
X=x −y + y3 ,
∂y ∂z ∂w
f : R2 → R4 , f (u, v) = (u2 , euv , u − v 4 , sin(v)).
(a) Calculate the nine exterior products:
α∧α α∧β α∧γ
β∧α β∧β β∧γ .
γ∧α γ∧β γ∧γ
(b) Calculate the three interior products: ιX α, ιX β, ιX γ.
(c) Calculate the pullback along f of the three forms: f ∗ (α), f ∗ (β), f ∗ (γ).
(d) Construct, or prove that it is impossible to do so, two one-forms η, θ ∈ Ω1 (R4 )
such that β = η ∧ θ.
LECTURE 13

13.1. The exterior derivative


Recall that, with our usual identification Tt R ∼
= R, the differential at p ∈ M of a
smooth map a ∈ C ∞ (M ) can be viewed as a covector
dp a ∈ Tp∗ M, dp a(v) = v(a), v ∈ Tp M.
Moreover, the assignment p 7→ dp a defines a one-form da ∈ Ω1 (M ). Globally, we
obtain a linear maps from functions to one-forms:
d : C ∞ (M ) −→ Ω1 (M ), a 7→ da.
In local coordinates (U, ϕ = (x1 , . . . , xm )) the differential is given by
m
X ∂a i
da = i
dx .
i=1
∂x
The exterior derivative is an operator which extends the differential of functions
to forms of higher degree, and is uniquely determined by the properties below.
Theorem 13.1.1. For any manifold M there is a unique linear map
d : Ωk (M ) −→ Ωk+1 (M ), k ≥ 0
satisfying the following properties:
(1) on zero-forms Ω0 (M ) = C ∞ (M ) it is the differential of functions;
(2) it squares to zero: d2 α = 0, for all α ∈ Ω(M );
(3) it is a graded derivation for the wedge product, i.e.
d(α ∧ β) = dα ∧ β + (−1)k α ∧ dβ,
for all α ∈ Ωk (M ) and all β ∈ Ω(M ).
Moreover, the following holds:
(4) if f : M → N is a smooth map, then df ∗ (α) = f ∗ (dα) for all α ∈ Ω(N ).
Definition 13.1.2. The operator d : Ω(M ) → Ω(M ) from Theorem 13.1.1 is called
the exterior derivative.
Remark 13.1.3. It is important to note that (2), (3) and (4) are essentially equiv-
alent to the following properties of smooth functions (this will be seen in the proof):
(2) partial derivatives commute;
113
114 IOAN MĂRCUT
, , MANIFOLDS

(3) the Leibniz rule;


(4) the chain rule.
We first prove Theorem 13.1.1 for open subsets of the Euclidean space.
Lemma 13.1.4. Theorem 13.1.1 holds for open subsets M ⊂ Rm and N ⊂ Rn .
Proof. Let us deduce a formula for d by using properties (1)-(3); this argument
will imply uniqueness. Let α ∈ Ωk (M ). To calculate dα we decompose α in the
standard basis:
X
(*) α= ai1 ...ik dxi1 ∧ . . . ∧ dxik , ai1 ...ik ∈ C ∞ (M ).
i1 <...<ik

Using linearity of d and the derivation rule (3), we obtain


X
dai1 ...ik ∧ dxi1 ∧ . . . ∧ dxik + ai1 ...ik ∧ d dxi1 ∧ . . . ∧ dxik ,

dα =
i1 <...<ik

where by (1) the first term is determined since dai1 ...ik is the usual differential of
the smooth map ai1 ...ik . For the second term, by applying (3), we obtain:
d dxi1 ∧ . . . ∧ dxik = d dxi1 ∧ dxi2 ∧ . . . ∧ dxik − dxi1 ∧ d dxi2 ∧ . . . ∧ dxik .
  

Since d extends the usual derivation (1), and since d2 = 0 (2), the first term is zero.
By repeatedly applying this argument, the entire expression is zero.
We conclude that there is a unique linear operator d : Ω(M ) → Ω(M ) satisfying
the conditions (1)-(3), and that on a form α ∈ Ωk (M ) as in (*) it acts by:
X
(**) dα = dai1 ...ik ∧ dxi1 ∧ . . . ∧ dxik =
i1 <...<ik
X ∂ai
1 ...ik
X
= dxi ∧ dxi1 ∧ . . . ∧ dxik .
i1 <...<ik i
∂xi
Next, we check that this linear operator satisfies indeed the properties from
Theorem 13.1.1. Property (1) clearly holds.
We check (2)-(4) first on smooth functions. For a ∈ C ∞ (M ), we have that:
n
! n n X n
∂2a
 
2
X ∂a i X ∂a i
X
d a=d i
dx = d i
∧ dx = i ∂xj
dxj ∧ dxi =
i=1
∂x i=1
∂x i=1 j=1
∂x
n
X ∂2a X ∂2a
dxi ∧ dxi + dxj ∧ dxi + dxi ∧ dxj = 0,

= i 2 i j
i=1
(∂x ) i<j
∂x ∂x
where we have used that the partial derivatives commute, and the following alge-
braic properties of the exterior algebra:
dxi ∧ dxi = 0 and dxi ∧ dxj = −dxj ∧ dxi .
The derivation rule (3), for a, b ∈ C ∞ (M ), is immediate:
d(ab)(v) = v(ab) = v(a)b + av(b) = (da · b + a · db)(v), ∀ v ∈ T M.
Similarly, for f : N → M and a ∈ C ∞ (M ), (4) holds by the chain rule:
f ∗ (da)(v) = da(df (v)) = d(a ◦ f )(v) = d(f ∗ a)(v), ∀ v ∈ T N.
Next, we show that (3) holds in all degrees. Since the wedge product is bilinear,
and d is linear, it suffices to show (3) for elements of the form α = adxI ∈ Ωk (M )
LECTURE 13. 115

and β = bdxJ ∈ Ωl (M ), where a, b ∈ C ∞ (M ) and, for I = {i1 < . . . < ik } and


J = {j1 < . . . < jl }, we have denoted:
dxI := dxi1 ∧ . . . ∧ dxik , dxJ := dxj1 ∧ . . . ∧ dxjl ,
Note that d(dxI ∧ dxJ ) = 0. Using (3) for functions, we obtain (3) in all degrees:
d(α ∧ β) = d(abdxI ∧ dxJ ) = d(ab) ∧ dxI ∧ dxJ = ((da)b + a(db)) ∧ dxI ∧ dxJ =
= da ∧ dxI ∧ bdxJ + (−1)k adxI ∧ db ∧ dxJ = dα ∧ β + (−1)k α ∧ dβ.
Again, by linearity of d, it suffices to prove (2) for α = adxI . Using (3) and
(2) for functions, we obtain (2) in general:
d2 (α) = d da ∧ dxI = (d2 a) ∧ dxI − da ∧ d(dxI ) = 0.


By linearity of d and of the pullback map f ∗ , it suffices to check (4) for α =


I
adx . By (4) for functions, we have that
f ∗ (dxI ) = df ∗ (xi1 ) ∧ . . . ∧ df ∗ (xik ),
and therefore, applying (2) and (3) repeatedly,
k
X
df ∗ (dxI ) = (−1)p+1 df ∗ (xi1 ) ∧ . . . ∧ d2 f ∗ (xip ) ∧ . . . ∧ df ∗ (xik ) = 0.
p=1

Using this, and again (4) for functions, we obtain that (4) holds in general:
df ∗ (α) = d f ∗ (a)f ∗ (dxI ) = d(f ∗ a) ∧ f ∗ (dxI ) + f ∗ (a)df ∗ (dxI ) =


= f ∗ (da) ∧ f ∗ (dxI ) = f ∗ (da ∧ dxI ) = f ∗ (dα).



Next, we prove existence of the operator d for any manifold:
Lemma 13.1.5. For any manifold M there exists an operator d : Ω(M ) → Ω(M )
with the properties (1)-(4) from Theorem 13.1.1.
Proof. For α ∈ Ω(M ), let dα be such that, for any chart (U, ϕ) on M ,
(dα)|U = ϕ∗ ◦ dR ◦ (ϕ−1 )∗ (α|U ),


where on the right-hand-side we have denoted by dR the operator constructed in


Lemma 13.1.4 for forms on ϕ(U ) ⊂ Rm . For this definition we need to check that,
given a second chart (V, ψ) we obtain the same value for dα on U ∩ V . Consider
the change of coordinates diffeomorphism between open sets in Rm :

f : ψ(U ∩ V ) −→ ϕ(U ∩ V ), f = ϕ ◦ ψ −1 .
By Lemma 13.1.4, we have that the operator dR satisfies property (4); therefore:
dR ◦ f ∗ (ϕ−1 )∗ (α|U ∩V ) = f ∗ ◦ dR (ϕ−1 )∗ (α|U ∩V ) .
 

Writing f ∗ = (ψ −1 )∗ ◦ ϕ∗ , and composing both sides with ψ ∗ on the left, we obtain:


ψ ∗ ◦ dR ◦ (ψ −1 )∗ (α|U ∩V ) = ϕ∗ ◦ dR ◦ (ϕ−1 )∗ (α|U ∩V ).
 

In other words, the two definition of dα coincide on U ∩ V . We conclude that d is


well-defined. Clearly, it suffices to check properties (1)-(4) locally; and these follow
from the respective properties of dR which were proven in Lemma 13.1.4. This
concludes the proof. 
116 IOAN MĂRCUT
, , MANIFOLDS

Next, we show that any operator satisfying (3) is local.


Lemma 13.1.6. Any linear operator D : Ω(M ) → Ω(M ) satisfying (3) is local,
i.e. Dα|U depends only on α|U for any α ∈ Ω(M ) and any open set U ⊂ M .
Proof. We need to show that, if α1 , α2 ∈ Ω(M ) satisfy α1 |U = α2 |U , then also
D(α1 )|U = D(α2 )|U . Since D is linear, by taking the difference of these forms,
it suffices to show that α|U = 0 implies that Dα|U = 0. Let p ∈ U , and let
χ ∈ C ∞ (M ) be a smooth function such that χ(p) = 1 and supp(χ) ⊂ U (this exists
by Lemma 4.1.2). Then, χα = 0. Applying D, we obtain:
0 = D(χα) = Dχ ∧ α + χDα.
Evaluating at p, and using that αp = 0 and χ(p) = 1, we obtain that (Dα)p = 0. 
The following concludes the proof of Theorem 13.1.1
Lemma 13.1.7. The operator d : Ω(M ) → Ω(M ) constructed in Lemma 13.1.5 is
the unique linear map satisfying (1)-(3).
Proof. Assume D : Ω(M ) → Ω(M ) is a second linear map satisfying (1)-(3). As
in the proof of Lemma 13.1.4, (1)-(3) imply that D = d on forms of the form
α = a0 da1 ∧ . . . ∧ dak , where ai ∈ C ∞ (M ). Consider now a general α ∈ Ω(M ). Let
p ∈ M and choose a chart (U, (x1 , . . . , xm )) around p, and write
X
α|U = ai1 ...ik dxi1 ∧ . . . ∧ dxik .
i1 <...<ik

By Lemma 4.1.2, there are function x ai1 ...ik ∈ C ∞ (M ) and an open set p ∈ V ⊂ U
ei , e
i i
e |V = x |V and e
such that x ai1 ...ik |V = ai1 ...ik |V . Consider the form:
X
αe := ai1 ...ik de
e xi1 ∧ . . . ∧ de
x ik .
i1 <...<ik

By the above, de
α = De e|V , Lemma 13.1.6 gives:
α. Since α|V = α
(Dα)|V = (De
α)|V = (de
α)|V = (dα)|V .
Thus, Dα and dα are equal in a neighborhood of p. We conclude that D = d. 

13.2. De Rham cohomology


Differential forms can be used to extract topological information about manifolds
through the so-called de Rham cohomology.
Definition 13.2.1.
• A k-form ω ∈ Ωk (M ) satisfying dω = 0 is called closed.
• A k-form ω ∈ Ωk (M ) for which there exists α ∈ Ωk−1 (M ) such that
dα = ω is called exact.
Since d2 = 0, every exact form is closed.
Definition 13.2.2.
• The k-th de Rham cohomology group of M is the quotient of the space
of closed forms modulo the space of exact forms:
k closed k-forms ker(d : Ωk (M ) → Ωk+1 (M ))
HdR (M ) := = .
exact k-forms im(d : Ωk−1 (M ) → Ωk (M ))
LECTURE 13. 117

• The de Rham cohomology of M is the direct sum of all these spaces:


M
k closed forms ker d
HdR (M ) := HdR (M ) = = .
exact forms im d
k≥0

• The cohomology class of a closed k-form ω is the equivalence class:


[ω] = ω + dΩk−1 (M ) ∈ HdR
k
(M ).
• The k-th Betti number of M is the dimension of the k-th de Rham
cohomology group of M (provided this vector space is finite dimensional):
k

bk (M ) := dim HdR (M ) .
Remark 13.2.3. The general algebraic framework in which cohomology groups
are considered is that of a chain complex (C • , δ), i.e. a sequence of vector spaces
{C k }k∈Z together with linear maps δ k : C k → C k+1 satisfying δ k+1 ◦ δ k = 0. The
cohomology of the chain complex are the quotient spaces:
ker(δ k )
H k (C • , δ) := .
im(δ k−1 )
Remark 13.2.4. We have that HdR 0
(M ) is the space of function a ∈ C ∞ (M )
satisfying da = 0. In a chart, this condition is equivalent to all partial derivatives
of a being zero. Therefore, da = 0 is equivalent to a being a locally constant
function. Such functions are uniquely determined by their values on the connected
0
components of M ; thus, HdR (M ) = Maps(π0 (M ), R) where π0 (M ) denotes the set
of connected components of M . So, if finite, b0 (M ) counts connected components
of M .
Remark 13.2.5. Let M be a connected manifold of dimension m. The top degree
cohomology has the following interpretation: if M is compact and orientable (to be
m m
defined in the following lecture), then HdR (M ) = R, otherwise, HdR (M ) = 0.
Example 13.2.6. Since S 1 is connected, we have that HdR
0
(S 1 ) ∼
= R. Let 0 ≤ θ ≤
1 1
2π be the “angle coordinate” on S . Every one-forms on S can be written as
f (θ)dθ ∈ Ω1 (S 1 ), f ∈ C ∞ (S 1 ).
The notation dθ is a bit misleading, because dθ is not an exact 1-form (θ is not a
smooth function on S 1 ); in fact, let us prove that [dθ] is a basis of HdR
1
(S 1 ), i.e.
1
HdR (S 1 ) = R[dθ] 6= 0.
Clearly, every one-form is closed. The exact one-forms have the form
∂g
f (θ)dθ = d(g(θ)) = (θ)dθ,
∂θ
for some g(θ) ∈ C ∞ (S 1 ). Note that an exact one-form integrates to 0 when viewed
as a smooth function on [0, 2π]
Z 2π
f (τ )dτ = g(2π) − g(0) = 0.
0

Conversely, given f ∈ C (S 1 ), we regard f as a 2π-periodic function on R. Then
Z x
g(x) := f (τ )dτ,
0
118 IOAN MĂRCUT
, , MANIFOLDS

defines a smooth function g : R → R, which satisfies


Z 2π
g(x + 2π) = g(x) + f (τ )dτ.
0
R 2π R 2π
Thus, g ∈ C ∞ (S 1 ) iff 0 f (τ )dτ = 0; and so, f (θ)dθ is exact iff 0 f (τ )dτ = 0.
In particular, [dθ] 6= 0 in H 1 (S 1 ). One the other hand, for f (θ)dθ ∈ Ω1 (S 1 ), we
c
R 2π c
have that (f (θ) − 2π )dθ has integral 0 for c = 0 f (τ )dτ , and so [f (θ)dθ] = 2π [dθ].
We conclude that:
H 1 (S 1 ) = R[dθ] 6= 0.
Remark 13.2.7. Assume that M is connected, and let p ∈ M . It can be proven
1
that HdR (M ) is isomorphic to the vector space of group homomorphisms from the
fundamental group (π1 (M, p), ∗) to (R, +):
1
HdR (M ) ∼
= Hom(π1 (M, p); R).
Even if the reader is not familiar with the notion of fundamental group, it is im-
portant to note that this notion is purely topological; it uses only the topology of
M and not its differentiable structure. In fact, this is true in all degrees; namely,
even though the groups HdR (M ) are constructed using smooth forms, they encode
information only about the topology of M . We state the precise result below:
k
Theorem ♣ 13.2.8 (De Rham). We have that HdR (M ) ∼
= Hom(Hk (M, Z); R)1.
Next, we give some important properties of de Rham cohomology.
Proposition 13.2.9. The wedge product on differential forms Ω(M ) induces a
product on de Rham cohomology HdR (M ):
[α] ∧ [β] = [α ∧ β],
which turns (HdR (M ), +, ∧) into an associative algebra.
Proof. If α and β are closed forms, using that d is a derivation for the wedge
product (Theorem 13.1.1 (3)), we have that α ∧ β is closed; hence the class [α ∧ β]
exists. Next, we show that [α ∧ β] depends only on [α], [β] ∈ HdR (M ), and not on
the particular representatives α, β ∈ Ω(M ). Any other representatives are of the
form α0 = α + dη and β 0 = β + dθ. Again by the derivation rule, we have that:
α0 ∧ β 0 = α ∧ β 0 + d(η ∧ β 0 ) = α ∧ β + d((−1)k α ∧ θ + η ∧ β 0 ),
where k is the degree of α. Thus [α0 ∧ β 0 ] = [α ∧ β]. This proves that the product
is well-defined. Clearly, associativity and the other algebraic properties (unital,
graded commutativity) are inherited from the wedge product on Ω(M ). 
Proposition 13.2.10. The pullback of differential forms along any smooth map
f : M → N induces an algebra homomorphism in de Rham cohomology
f ∗ : HdR (N ) −→ HdR (M ), f ∗ [ω] := [f ∗ ω].
Moreover, if g : N → P is a second smooth map, then
f ∗ ◦ g ∗ = (g ◦ f )∗ : HdR (P ) −→ HdR (M ).
1The group H (M, Z) is called the k-th singular homology of M ; it will not be introduced in this
k
course, but it is important to note that it is defined purely in topological terms
LECTURE 13. 119

Proof. By item (4) of Theorem 13.1.1, we have that f ∗ : Ω(N ) → Ω(M ) sends
closed forms to closed forms, and exact forms to exact forms; thus it induces indeed
a map f ∗ : HdR (N ) → HdR (M ). The fact that f ∗ is an algebra homomorphism on
differential forms, implies that it is an algebra homomorphism in cohomology, and
similarly, the equality f ∗ ◦ g ∗ = (g ◦ f )∗ holds already at the level of forms. 

An immediate consequence of the Proposition, is that de Rham cohomology is


an invariant of a manifold:
Corollary 13.2.11. If M and N are diffeomorphic manifolds, then HdR (M ) and
HdR (N ) are isomorphic algebras.
Theorem 13.2.12 (Homotopy invariance of de Rham cohomology). Let ft : M →
N be a family of smooth maps depending smoothly on t ∈ [0, 1], then
f0∗ = f1∗ : HdR (N ) −→ HdR (M ).
Proof. We can assume that ft is defined for all t ∈ R (first, extend the family to a
small open interval containing [0, 1], and then use a diffeomorphism of this interval
to R which fixes [0, 1]). The family can be regarded as a smooth map
f : R × M −→ N.
Differential forms on R × M can be uniquely decomposed as
dt ∧ αt + βt ∈ Ω(R × M ),
where αt , βt are smooth families of differential forms on M . We denote the exterior
derivative of forms on R × M by dtot , and the exterior derivatives on M and N as
usually by d. In the above decomposition, dtot becomes:
d
dtot (dt ∧ αt + βt ) = −dt ∧ dαt + dt ∧ βt + dβt .
dt
Note that the pullback of ω ∈ Ωk (N ) by f decomposes as follows:
f ∗ (ω) = dt ∧ Ht (ω) + ft∗ (ω),
where Ht (ω) ∈ Ωk−1 (M ) is a smooth family of k−1-forms on M . Next, we calculate
both sides of the equality dtot f ∗ (ω) = f ∗ (dω) (Theorem 13.1.1 (4)):
 d 
dtot dt ∧ Ht (ω) + ft∗ (ω) = dt ∧ − dHt (ω) + ft∗ (ω) + dft∗ (ω)

dt
f ∗ (dω) = dt ∧ Ht (dω) + ft∗ (dω).
Using that also ft∗ commutes with the exterior derivative, and the uniqueness of
the decomposition of differential forms on R × M , we obtain:
d ∗
f (ω) = Ht (dω) + dHt (ω).
dt t
Integrating from 0 to 1:
(*) f1∗ (ω) − f0∗ (ω) = h ◦ d(ω) + d ◦ h(ω),
where h denotes the following linear operator:
Z 1
k k−1
h : Ω (N ) −→ Ω (M ), h(η) := Ht (η)dt.
0
120 IOAN MĂRCUT
, , MANIFOLDS

This operator is called a homotopy operator, and relation (*) is all what is needed
k
to conclude the proof. Namely, if [ω] ∈ HdR (M ), then dω = 0, and so
f1∗ ([ω]) = [f1∗ (ω)] = [f0∗ (ω) + d ◦ h(ω)] = f0∗ ([ω]).
Thus f1 and f0 induce the same map in cohomology. 
Remark 13.2.13. The operator Ht : Ωk (N ) → Ωk−1 (M ) can be given explicitly:
Ht (ω) = ft∗ (ι d ft ω).
dt

Therefore, the homotopy operators are given by


Z 1
h(ω) = ft∗ (ι d ft ω)dt.
dt
0

Homotopy invariance can be used to calculate the cohomology of Rn .


Corollary 13.2.14 (The Poincaré Lemma). We have that

k n 0, k 6= 0;
HdR (R ) =
R, k = 0.
Proof. Consider the family of maps ft : Rn → Rn , ft (x) := tx. Then f1 = id,
and so f1∗ [ω] = [ω] for all [ω] ∈ HdR
k
(Rn ). On the other hand, f0 (x) = 0 for all

n
x ∈ R , and so f0 [ω] = 0 for all [ω] ∈ HdR k
(Rn ) and k > 0, and f0∗ (c) = c, for
0 n
c ∈ R = H (R ). Applying Theorem 13.2.12, we obtain the result. 
More generally, we have that:
Corollary 13.2.15. Let π : E → M be a vector bundle. Then π ∗ : HdR (M ) →
HdR (E) is a linear isomorphism.
Example 13.2.16. We show that
1
HdR (S n ) = 0, n ≥ 2.
Consider a closed one-form α ∈ Ω1 (S n ). We will prove that α is exact. Decompose
S n = UN ∪ US , where UN is the complement of the north pole, and US is the
complement of the south pole. Since UN ∼ = Rn we have that [α|UN ] = 0 ∈ HdR 1
(UN ).

Therefore, there exists fN ∈ C (UN ) such that dfN = α|UN . Similarly, we find
fS ∈ C ∞ (US ) such that dfS = α|US . On the sphere without the two poles, we have
that
d(fN |UN ∩US − fS |UN ∩US ) = α|UN ∩US − α|UN ∩US = 0.
This implies that fN |UN ∩US −fS |UN ∩US is a constant function c ∈ R (this is because
UN ∩US is connected; here we are using that n ≥ 2). Therefore, the following defines
a smooth function on S n :
f : S n −→ R, f |UN := fN − c, and f |US = fS ,
which clearly satisfies df = α.
The general structure of the cohomology of spheres is given below (the proof
is not difficult; see e.g. [3]):
Theorem ♣ 13.2.17. The cohomology of spheres is given by:

k 0, 0 < k < n;
HdR (S n ) =
R, k = 0, n.
LECTURE 13. 121

13.3. Cartan calculus


In this section we introduce the Lie derivative of a differential form, and we prove
several relations between this operator, the exterior derivative and the interior
product. This entire algebraic machinery, including the definition of the exterior
derivative, was developed by Élie Cartan, and therefore it is commonly called Car-
tan calculus.
Definition 13.3.1. The Lie derivative of differential forms along a vector field
X ∈ X(M ) is the operator
LX : Ωk (M ) −→ Ωk (M )
d 1
LX α := (φtX )∗ (α) t=0 = lim ((φtX )∗ (α) − α).

dt t→0 t

Theorem 13.3.2. The following relations hold:


(1) LX (α ∧ β) = LX α ∧ β + α ∧ LX β
(2) LX ◦ d = d ◦ LX
(3) LX ◦ ιY − ιY ◦ LX = ι[X,Y ]
(4) LX = d ◦ ιX + ιX ◦ d (Cartan’s magic formula)
(5) LX ◦ LY − LY ◦ LX = L[X,Y ]
for all X, Y ∈ X(M ) and all α, β ∈ Ω(M ).
Proof. Since the pullback by the flow of X preserves the wedge product, we obtain:
(φtX )∗ (α ∧ β) = (φtX )∗ (α) ∧ (φtX )∗ (β).
Taking the derivative at t = 0, and using that φ0X = id, we obtain (1).
For (2), we use that the pullback commutes with the exterior derivative:
(φtX )∗ (dα) = d(φtX )∗ (α),
and take the derivative at t = 0.
Formula (3) is a version of (1) for the interior product instead of the exterior
product. Namely, directly from the definitions of the pullback, it follows that
(φtX )∗ (ιY α) = ι(φtX )∗ (Y ) (φtX )∗ (α).
Taking the derivative at t = 0, we obtain that:
LX (ιY α) = ιLX Y α + ιY (LX α).
Using that LX Y = [X, Y ] (see Proposition 10.3.3), we obtain (3).
Denote the operator in (4) by DX := d ◦ ιX + ιX ◦ d. By using the derivation
rule for d and ιX , one easily checks that DX satisfies the same derivation rule as
LX (this follows also from the comments in the previous subsection)
DX (α ∧ β) = DX α ∧ β + α ∧ DX β.
2
Using that d = 0, we obtain that DX commutes with d:
DX ◦ d = (d ◦ ιX + ιX ◦ d) ◦ d = d ◦ ιX ◦ d = d ◦ (d ◦ ιX + ιX ◦ d) = d ◦ DX .
Next, note that DX and LX coincide on functions a ∈ C ∞ (M ):
DX (a) = (d ◦ ιX + ιX ◦ d)(a) = ιX da = LX (a).
Finally, since both operators LX and DX are local and linear, it suffices that show
that they coincide on forms of the form a0 da1 ∧ . . . ∧ dak , with ai ∈ C ∞ (M ). This
122 IOAN MĂRCUT
, , MANIFOLDS

follows by applying the derivation rule, that both commute with d, and that both
act the same on functions.
Formula (5) follows by using (2)-(4):
L[X,Y ] = d ◦ ι[X,Y ] + ι[X,Y ] ◦ d =
= d ◦ (LX ◦ ιY − ιY ◦ LX ) + (LX ◦ ιY − ιY ◦ LX ) ◦ d =
= LX ◦ d ◦ ιY − d ◦ ιY ◦ LX + LX ◦ ιY ◦ d − ιY ◦ d ◦ LX =
= LX ◦ (d ◦ ιY + ιY ◦ d) − (d ◦ ιY + ιY ◦ d) ◦ LX =
= LX ◦ LY − LY ◦ LX .

A differential k-form ω ∈ Ωk (M ) can be viewed also as a C ∞ (M )-multilinear
k-form on the C ∞ (M )-module X(M ) with values in C ∞ (M ):
ω : X(M ) × . . . × X(M ) −→ C ∞ (M )
| {z }
k × X(M )
ω(X1 , . . . , Xk )(p) := ωp ((X1 )p , . . . , (Xk )p ),
for X1 , . . . , Xk ∈ X(M ). This allows for a coordinate-free definition of the exterior
derivative:
Theorem 13.3.3. The exterior derivative satisfies:
k
X
dω(X0 , . . . , Xk ) = (−1)i LXi (ω(X0 , . . . , X
bi , . . . , Xk ))
i=0
X
+ (−1)i+j ω([Xi , Xj ], X0 , . . . , X
bi , . . . , X
bj , . . . , Xk ),
0≤i<j≤k
k
for all ω ∈ Ω (M ) and all X0 , . . . , Xk ∈ X(M ), where the notation X
bi means that
the vector field Xi is being skipped.
The proof is left as an exercise. In low degrees the formula gives:
da(X0 ) =LX0 (a),
dθ(X0 , X1 ) =LX0 (θ(X1 )) − LX1 (θ(X0 )) − θ([X0 , X1 ]),
dω(X0 , X1 , X2 ) =LX0 (ω(X1 , X2 )) − LX1 (ω(X0 , X2 )) + LX2 ω(X0 , X1 )
− ω([X0 , X1 ], X2 ) + ω([X0 , X2 ], X1 ) − ω([X1 , X2 ], X0 ),

for a ∈ C ∞ (M ), θ ∈ Ω1 (M ) and ω ∈ Ω2 (M ).

13.4. The algebra of graded derivations


We explain the algebraic framework in which the relations of Theorem 13.3.2 fit
best. A graded algebra is an algebra (A, +, ·) which decomposes as a direct sum
into vector space: M
A= Ak ,
k∈Z
and satisfies
a · b ∈ Ak+l , for all a ∈ Ak , b ∈ Al .
LECTURE 13. 123

For this discussion, it is not important whether A is unital, associative, or (graded)


commutative. A derivation of degree p ∈ Z of A is a linear map D : A → A,
satisfying D(Ak ) ⊂ Ak+p , and the derivation rule:
D(a · b) = D(a) · b + (−1)kp a · D(b), for all a ∈ Ak , b ∈ A.
Let us denote by Derp (A) the space of all derivations of degree p of A. The graded
commutator of D1 ∈ Derp (A) and D2 ∈ Derq (A) is the map
[D1 , D2 ]c := D1 ◦ D2 − (−1)pq D2 ◦ D1 : A −→ A.
It can be easily checked that [D1 , D2 ]c is a derivation of degree p + q. Consider
M
Der(A) := Derp (A).
p∈Z

This vector space endowed with the bilinear extension of the commutator is the
graded algebra of derivations of A:
(Der(A), +, [·, ·]c ).
The commutator satisfies the following relations:
[D1 , D2 ]c = −(−1)pq [D2 , D1 ]c ,
[D1 , [D2 , D3 ]c ]c = [[D1 , D2 ]c , D3 ]c + (−1)pq [D2 , [D1 , D3 ]c ]c ,
for all D1 ∈ Derp (A), D2 ∈ Derq (A) and D3 ∈ Der(A). These relations are called
graded commutativity and the graded Jacobi identity, respectively. Note
that the second relation is equivalent to the operator [D1 , ·]c being a derivation of
degree p of Der(A), i.e. [D1 , ·]c ∈ Derp (Der(A)). A graded algebra which satisfies
these two axioms is called a graded Lie algebra.
In our case, A = Ω(M ), with multiplication the usual wedge product. In the
language of graded algebras, the derivation rule for the interior product and the
exterior derivative become:
ιX ∈ Der−1 (Ω(M )) and d ∈ Der1 (Ω(M )).
Note that we have the relations d2 = 0 and ιX ◦ ιY = −ιY ◦ ιX can be expressed
using the graded commutator:
[d, d]c = 2d2 = 0 and [ιX , ιY ]c = ιX ◦ ιY + ιY ◦ ιX = 0.
Moreover, Theorem 13.3.2 can be restated as follows:
(1) LX ∈ Der0 (Ω(M ))
(2) [LX , d]c = 0
(3) [LX , ιY ]c = ι[X,Y ]
(4) LX = [ιX , d]c
(5) [LX , LY ]c = L[X,Y ]

13.5. Exercises
The exterior derivative of a function a ∈ C ∞ (M ) can be calculated as follows:
d
da(v) = a(γ(t)) t=0 ,
dt

where t 7→ γ(t) is a curve on M , with dt (0) = v. The following exercise extends
this result to higher degrees:
124 IOAN MĂRCUT
, , MANIFOLDS

Exercise 13.1. Let v1 , . . . , vk ∈ Tp M be k-vectors. A small parallelotope with


directions v1 , . . . , vk is a smooth map (t1 , . . . , tk ) 7→ γ(t1 , . . . , tk ) defined on an open
neighborhood of 0 in Rk such that
∂γ
γ(0, . . . , 0) = p, (0, . . . , 0) = vi .
∂ti
Prove that for any k − 1-form ω ∈ Ωk−1 (M ), the exterior derivative is given by:
k
X ∂  ∂γ ∂γ
d ∂γ 
dω(v1 , . . . , vk ) = (−1)i−1 ω ,..., ,..., .
i=1
∂ti ∂t1 ∂ti ∂tk (t1 ,...,tk )=(0,...,0)

Hint: Calculate dγ ∗ (ω), and use item (4) from Theorem 13.1.1.
Exercise 13.2. Let U ⊂ R3 be an open set. Recall the standard operators from
Calculus: the gradient, the curl and the divergence:
grad = ∇ : C ∞ (U ) −→ X(U ),
curl = ∇× : X(U ) −→ X(U ),
div = ∇· : X(U ) −→ C ∞ (U ),

which for f ∈ C ∞ (U ) and F = Fx ∂x


∂ ∂
+ Fy ∂y ∂
+ Fz ∂z ∈ X(U ) are given by:
∂f ∂ ∂f ∂ ∂f ∂
∇f = + + ,
∂x ∂x ∂y ∂y ∂z ∂z
 ∂F ∂Fy  ∂  ∂F ∂Fz  ∂  ∂F ∂Fx  ∂
z x y
∇×F = − + − + − ,
∂y ∂z ∂x ∂z ∂x ∂y ∂x ∂y ∂z
∂Fx ∂Fy ∂Fz
∇·F = + + .
∂x ∂y ∂z
Denote by h·, ·i the standard dot-product on R3 = Tp U .
(a) Show that the following give C ∞ (U )-linear isomorphisms:

A : X(U ) −→ Ω1 (U ), A(F ) := hF, ·i

B : X(U ) −→ Ω2 (U ), B(F ) := ιF (dx ∧ dy ∧ dz)
∞ ∼
C : C (U ) −→ Ω3 (U ), C(f ) := f dx ∧ dy ∧ dz.
(b) Prove that the following diagram commutes:

C ∞ (U )
∇ / X(U ) ∇×
/ X(U ) ∇· / C ∞ (U ) .
 id A B C
C ∞ (U )
d / Ω1 (U ) d / Ω2 (U ) d / Ω3 (U )

(c) Deduce the classical relations:


curl(grad(f )) = 0 and div(curl(F )) = 0,
and show that under the map A conservative vector fields correspond to exact
1-forms, irrotational vector fields correspond to closed 1-forms; and under the
map B divergence free vector fields correspond to closed 2-forms. (Hint: open
a calculus book.)
LECTURE 13. 125

Exercise 13.3. (a) For ω ∈ Ωk (M ) and X0 , . . . , Xk ∈ X(M ), prove that:


k
X
LX0 (ω(X1 , . . . , Xk )) = LX0 (ω)(X1 , . . . , Xk ) + ω(X1 , . . . , [X0 , Xi ], . . . , Xk ).
i=1
Hint: either directly from the definitions, or use item (3) from Theorem 13.3.2.
(b) Prove Theorem 13.3.3.
Hint: argue by induction; use (a) and item (4) from Theorem 13.3.2.
Exercise 13.4. Let U ⊂ C be an open set, and consider a smooth function
f : U → C, f (z) = u(z) + iv(z), u(z), v(z) ∈ R.
Consider the one-forms θ1 , θ2 ∈ Ω1 (U ):
θ1 (x, y) := v(x + iy)dx + u(x + iy)dy, θ2 (x, y) := u(x + iy)dx − v(x + iy)dy.
Prove that f is a holomorphic map if and only if both θ1 and θ2 are closed.
Exercise 13.5. Prove Corollary 13.2.15.
Exercise 13.6. (a) For 1 ≤ k ≤ n, given a closed k-form on Rn :
X
ω= ωi1 ,...,ik (x1 , . . . , xn )dxi1 ∧ . . . ∧ dxik ∈ Ωk (Rn )
i1 <...<ik

give an explicit formula for a k − 1-form α ∈ Ωk−1 (Rn ) such that dα = ω.


(Hint: extract a formula from the proofs of Theorem 13.2.12 and of Corollary
13.2.14, and then check directly that it works.)
(b) Prove Corollary 13.2.14 directly, by using (a).
Exercise 13.7. Calculate the de Rham cohomology of the Möbius band.
LECTURE 14

In this lecture we construct the density bundle of a smooth manifold. The


density bundle and smooth densities give a precise geometric meaning to the notion
of “volume element” dx1 . . . dxn which appears when writing integrals
Z
f (x1 , . . . , xn )dx1 . . . dxn .

We will show that smooth densities, i.e. smooth section of the density bundle, can
be canonically integrated. The discussion below follows [10].

14.1. Volume elements


Let us review some elementary facts about volumes of n-dimensional parallelotope,
i.e. the n-dimensional generalization of the notions of: line segment, parallelogram,
parallelepiped. Let P ⊂ Rn be an n-dimensional parallelotope which has a corner in
the origin of Rn , and is spanned by n linearly independent vectors v1 , . . . , vn ∈ Rn :
P = {t1 v1 + . . . + tn vn : 0 ≤ ti ≤ 1}.
Note that, up to a permutation, P determines the basis (v1 , . . . , vn ). Writing vj =
(a1j , . . . , anj ), we parameterize P by the standard n-cube [0, 1]n
n

X
A : [0, 1]n −→ P, (t1 , . . . , tn ) 7→ (x1 , . . . , xn ), xi = aij tj .
j=1

The volume of P is simply the absolute value of the determinant of A:


Z Z
Vol(P ) = dx1 . . . dxn = |det(A)|dt1 . . . dtn = |det(A)|,
P [0,1]n

where we used the change of coordinate formula for multiple integrals. More gen-
erally, the same argument applies to the following situation: if Q is any other
n-dimensional parallelotope spanned by w1 , . . . , wn , and B : Rn → Rn is linear
isomorphism satisfying B(vi ) = wi , for 1 ≤ i ≤ n (i.e. B is the transition matrix
between the two bases) then we have that:
Vol(Q) = |det(B)|Vol(P ).
The following is the abstract notion of a volume of parallelotopes:
127
128 IOAN MĂRCUT
, , MANIFOLDS

Definition 14.1.1. Let V be an n-dimensional vector space. A volume element


on V is a map
µ : V × . . . × V −→ R
| {z }
n × V
such that, for all linear maps A : V → V and all v1 , . . . , vn ∈ V :
µ(Av1 , . . . , Avn ) = |det(A)|µ(v1 , . . . , vn ).
We denote the space of volume elements on V by Ve(V ).
Up to a constant, there is a unique non-zero volume element:
Lemma 14.1.2. Let e1 , . . . , en be a basis of the vector space V . Then there is a
unique volume element µ0 ∈ Ve(V ) such that µ0 (e1 , . . . , en ) = 1. Moreover, for any
other µ ∈ Ve(V ) there is a unique c ∈ R such that µ = cµ0 ; in other words the
volume elements on V form a 1-dimensional vector space with basis µ0 .
Proof. Let v1 , . . . , vn ∈ V , and define µ0 (v1 , . . . , vn ) = |det(A)|, where A : V → V
is the linear map determined by Aei = vi , for 1 ≤ i ≤ n. Note that µ0 is indeed a
volume element: for any linear map B : V → V , we have that
µ0 (Bv1 , . . . , Bvn ) = µ0 (BAe1 , . . . , BAen ) = |det(BA)| =
= |det(B)||det(A)| = |det(B)|µ0 (v1 , . . . , vn ).
Consider an arbitrary µ ∈ Ve(V ), and denote c := µ(e1 , . . . , en ). Let v1 , . . . , vn ∈
V , and let A : V → V be such that Aei = vi , for 1 ≤ i ≤ n. Then, we have that
µ(v1 , . . . , vn ) = |det(A)|µ(e1 , . . . , en ) = cµ0 (v1 , . . . , vn ).
Hence µ = cµ0 . This shows that Ve(V ) is indeed 1-dimensional, and also that µ0
is the unique volume element such that µ0 (e1 , . . . , en ) = 1. 

14.2. The density bundle


Let M be a manifold. For p ∈ M , denote the space of volume elements on Tp M by
Dp (M ) := Ve(Tp M ).
The density bundle of M is the union of all these vector spaces:
G
D(M ) := Dp (M ).
p∈M
1 m
Let (U, ϕ = (x , . . . , x )) be a chart around a point p ∈ M . Lemma 14.1.2 shows
that there exists a unique volume element
|dx1 ∧ . . . ∧ dxm |p ∈ Dp (M )
such that  ∂ ∂ 
|dx1 ∧ . . . ∧ dxm |p |p , . . . , |p = 1,
∂x1 ∂xm
1 m
and moreover, that |dx ∧ . . . ∧ dx |p forms a basis of Dp (M ).
Consider a second chart (V, ψ = (y 1 , . . . , y m )), with p ∈ U ∩ V . Then,
m
X ∂xj
∂ ∂
i p
= i
(p) j p ,
∂y j=1
∂y ∂x
LECTURE 14. 129

and so, by Lemma 14.1.2, this implies that:


∂  ∂(x1 , . . . , xm )
 ∂
1 m
|dx ∧ . . . ∧ dx |p ,..., m p = (p) ,
∂y 1 p ∂y ∂(y 1 , . . . , y m )
∂(x1 ,...,xm )
where we have denoted by ∂(y 1 ,...,y m ) the determinant of the Jacobian matrix of
ϕ ◦ ψ −1 , i.e.
∂(x1 , . . . , xm ) −1
 ∂xj 
(p) = det(d p ϕ ◦ (dp ψ) ) = det (p) .
∂(y 1 , . . . , y m ) ∂y i
Thus we have obtained the following change of coordinate formula:
∂(x1 , . . . , xm )

1 m
|dx ∧ . . . ∧ dx |p =
(p) · |dy 1 ∧ . . . ∧ dy m |p .
∂(y 1 , . . . , y m )
Using this, the result below can be proven exactly like Theorem 8.1.2:

Theorem 14.2.1. The density bundle π : D(M ) → M is a smooth vector bundle


of rank 1. The vector bundle structure is uniquely determined by the condition that,
for every chart (U, ϕ = (x1 , . . . , xm )) on M , |dx1 ∧ . . . ∧ dxm | : U → D(M ) is a
smooth section.

14.3. Densities
Definition 14.3.1. The space of sections of the density bundle will be denoted by

D(M ) := Γ(D(M )).

Sections of the density bundle are called smooth densities (or 1-densities, or just
densities) on M . A density µ ∈ D(M ) is said to be positive, if µp (v1 , . . . , vm ) > 0
for every p ∈ M and every basis v1 , . . . , vm of Tp M .

Given a chart (U, ϕ = (x1 , . . . , xm )) on M , the standard volume element defines


a positive density over U

|dx1 ∧ . . . ∧ dxm | ∈ Γ(D(M )|U ) = D(U ).

Any other density µ ∈ D(U ) over U is of the form

µp = f (p)|dx1 ∧ . . . ∧ dxm |p ,

for a unique smooth map f ∈ C ∞ (U ).


Next, we show that D(M ) is a trivializable vector bundle (however, the trivi-
alization is not canonical).

Proposition 14.3.2. On any manifold M there exists a positive density µ ∈ D(M ).


In particular, the density bundle D(M ) is trivializable.

Proof. Consider an atlas on M , {(Ui , ϕi )}i∈I , and let {ρi }i∈I be a subordinate
partition of unity. Let µi = |dx1ϕi ∧ . . . ∧ dxm ϕi | ∈ D(Ui )Pbe the local density
corresponding to the chart ϕi . It is easy to check that µ := i∈I ρi µi is a positive
density on M . 
130 IOAN MĂRCUT
, , MANIFOLDS

14.4. Integration of densities


We define the pullback of densities under diffeomorphisms
Definition 14.4.1. The pullback of a smooth density µ ∈ D(N ) along a local
diffeomorphism ϕ : M → N is the smooth density ϕ∗ (µ) ∈ D(M ) defined by:
ϕ∗ (µ)p (v1 , . . . , vm ) := µϕ(p) (dp ϕ(v1 ), . . . , dp ϕ(vm )).
Let us write the pullback of densities in local coordinates. Let U ⊂ Rm be an
open set, and let µ ∈ D(U ) be a density on U . We can write:
µ = f (x1 , . . . , xm )|dx1 ∧ . . . ∧ dxm |, f ∈ C ∞ (U ).
Consider a diffeomorphism ϕ : V − ∼→ U , where V ⊂ Rm is another open set. Then,
 i 
∗ 1 m ∂ϕ
· |dy 1 ∧ . . . ∧ dy m |.

ϕ (µ) = f (ϕ(y , . . . , y )) det
∂y j
Note that this is precisely the change of variables formula in multiple integrals!
More precisely, let us recall the following:
Theorem 14.4.2 (Substitution rule for multiple integrals). Let U, V ⊂ Rm be
open sets, and let ϕ : V → U be a diffeomorphism. If f ∈ C ∞ (U ) is a compactly
supported function on U , then we have that
Z Z  i 
1 m 1 m 1 m ∂ϕ 1
dy . . . dy m .

f (x , . . . , x )dx . . . dx = f (ϕ(y , . . . , y )) det
U V ∂y j
Let Dc (M ) denote the space of compactly supported densities on M . Com-
pactly supported densities can be integrated canonically:
Theorem 14.4.3. For each manifold M , there is a linear map
Z
: Dc (M ) −→ R,
M
which is uniquely determined by the following conditions:
(1) It extends the usual Riemann/Lebesgue integration of compactly supported func-
tions on Rm : if µ ∈ Dc (Rm ) is a compactly supported density, then
Z Z
µ := f (x1 , . . . , xm )dx1 . . . dxm ,
Rm Rm

where µ = f (x , . . . , x )|dx ∧ . . . ∧ dxm |.1


1 m 1

(2) It is local in the sense that: if µ ∈ Dc (M ) has support inside the open U , then
Z Z
µ= µ.
U M

(3) It is invariant under diffeomorphisms: if ϕ : M −→ N is diffeomorphism, and
µ ∈ Dc (N ), then Z Z
µ= ϕ∗ (µ).
N M
1This property looks quite tautological. However, note that the right hand side of the equation is
the integral of the smooth compactly supported function f on U (the Riemann/Lebesgue integral),
and dx1 . . . dxm plays a formal notational role (the “volume element” which one often forgets to
write), whereas µ = f (x1 , . . . , xm )|dx1 ∧ . . . ∧ dxm | is an “actual volume element”, as defined in
Definition 14.1.1.
LECTURE 14. 131

We will make the following identification: if U ⊂ M is an open set, we will


identify Dc (U ) with elements in Dc (M ) with support in U . This is used implicitly
in writing condition (2), and will be used throughout the proof.

Proof. We will construct the integration map in several steps:


Step 1. Let U ⊂ Rm be an open set, and let µ = f |dx1 ∧ . . . ∧ dxm | ∈ Dc (U )
be
R a compactly supported density on U . Conditions (1) and (2) force us to define
1
. . . dxm .
R
U
µ = Rm f dx
Step 2. Consider two open sets U, V ⊂ Rm , and a diffeomorphism θ : V − ∼→ U . Let

µ ∈ Dc (U ) be a compactly supported density on U . Then θ (µ) is a compactly
supported density on V , and Theorem 14.4.2 can be restated as:
Z Z
µ= θ∗ (µ).
U V

This shows that (3) holds for open subsets of Rm .


Step 3. Let µ ∈ Dc (M ), and assume that there is a coordinate chart (U, ϕ) on M
such that supp(µ) ⊂ U . Then we define:
Z Z
µ= (ϕ−1 )∗ (µ).
M ϕ(U )

Note that this definition is forced by conditions (1)-(3). We prove that this def-
inition does not depend on the chart. Consider a second chart (V, ψ) such that
supp(µ) ⊂ V . Then, by applying Step 2 to the change of coordinates diffeomor-
phism θ = ϕ ◦ ψ −1 : ψ(U ∩ V ) → ϕ(U ∩ V ), we obtain
Z Z Z
(ϕ−1 )∗ (µ) = θ∗ (ϕ−1 )∗ (µ) = (ψ −1 )∗ (µ).
ϕ(U ∩V ) ψ(U ∩V ) ψ(U ∩V )

Step 4. Consider a cover of M by coordinate charts {(Ui , ϕi )}i∈I , and let {ρi }i∈I
be a partition of unity subordinate to it. For µ ∈ Dc (M ), define
Z XZ
µ := ρi µ.
M i∈I M
R
Since ρi µ is supported in the domain Ui of the chart ϕi , it follows that M ρi µ is
well-defined; since the support of µ is compact, and the family {ρi }i∈I is locally
finite, only for a finite Rnumber of i ∈ I we have that ρi µ 6= 0, therefore the sum
above is finite. Thus, M µ is well-defined. Moreover, the definition is forced by
linearity of the integral and by uniqueness of the integral on the domains of charts
(see Step 3).
Step 5. Next, we prove that the definition is independent on the atlas and on
the partition of unity. Consider a second atlas {(Vj , ψj )}j∈JP , with subordinate
partition of unity {σj }j∈J . For each i ∈ I, we have ρi µ = j∈J ρi σj µ is a finite
R
sum of densities supported in Ui . Since M : Dc (Ui ) → R is linear,
Z XZ
ρi µ = ρi σj µ.
M j∈J M

This implies that:


XZ XXZ
ρi µ = ρi σj µ.
i∈I M i∈I j∈J M
132 IOAN MĂRCUT
, , MANIFOLDS

Since all sums involved are finite, we can change the summation order on the
right hand side. Therefore, by applying the same argument with the two atlases
interchanged, we obtain:
XZ XZ
ρi µ = σj µ.
i∈I M j∈J M

This shows that the integral is independent on the choice of the atlas.
We have proven existence of the integral, and we have proven that the condi-
tions (1)-(3) imply uniqueness of the integral. We leave it to the reader to check
that conditions (1)-(3) are indeed satisfied for the integral we have constructed. 

14.5. Orientations
Definition 14.5.1. Let V be an n-dimensional vector space V , and let B(V ) denote
the space of all bases of V . An orientation on V is a map
o : B(V ) −→ {−1, 1},
such that, for all linear isomorphisms A : V −∼
→ A, it satisfies:
o(Av1 , . . . , Avn ) = sign(det(A))o(v1 , . . . , vn ).
We denote by Or(V ) the set of orientations on V . Given an orientation o on V a
basis v1 , . . . , vn of V is said to be a positive basis for o (resp. negative basis), if
o(v1 , . . . , vn ) = 1 (resp. o(v1 , . . . , vn ) = −1).
Clearly, every vector space has precisely two orientations. This is true even for
a 0-dimensional vector space, which has a unique basis, namely the empty set ∅.
The two orientations are o(∅) = 1 and o(∅) = −1; we simply write o = 1 or o = −1.
The standard orientation on Rn is the orientation for which the standard
basis e1 , . . . , en is positive. In general, a basis v1 , . . . , vn of Rn is positive with
respect to the standard orientation iff the matrix A = (vij )1≤i,j≤n has positive de-
terminant, where we have denoted vi = (vi1 , . . . , vin ).

Let M be a manifold. Putting together the orientations of all tangent spaces


Tp M one obtains the orientation bundle of M :
G
O(M ) := Or(Tp M ).
p∈M

Now O(M ) comes with a canonical projection π : O(M ) → M . For p ∈ M , π −1 (p)


is composed of the two orientations of Tp M . The orientation bundle is a smooth
manifold and π : O(M ) → M is a local diffeomorphism. To a chart (U, ϕ) on M
we associate two charts on O(M ):
(U + , ϕ+ ) and (U − , ϕ− ).
Let ϕ = (x1 , . . . , xm ). For each p ∈ U consider the orientation on Tp M denoted by

p ∈ Op (M ) = Or(Tp M )

which is determined by the condition that


∂ ∂
1 p
, . . . , m p
∂x ∂x
LECTURE 14. 133

forms a positive basis. Then oϕ : U → O(M )|U is a section of π, i.e. π ◦ oϕ = idU .


Let −oϕ : U → O(M )|U be the opposite orientation. The charts are given by:
U + := oϕ (U ), ϕ+ := ϕ ◦ π|U + ,
U − := (−oϕ )(U ) = −U + , ϕ− := ϕ ◦ π|U − .
We leave it to the reader to check that O(M ), endowed with these family of charts,
is a smooth manifold of the same dimension as M .
Example 14.5.2. The orientation bundle of the Möbius band is a cylinder; you
can see this here: [19].
Definition 14.5.3. An orientation on a manifold M is a smooth choice of ori-
entations on all the tangent spaces of M . In other words, an orientation of M is
a smooth section of the orientation bundle, i.e. a smooth map o : M → O(M ) such
that π ◦ o = idM .
A manifold M is said to be orientable if it admits an orientation.
An oriented manifold is a manifold M with a fixed orientation o.
If (M1 , o1 ) and (M2 , o2 ) are two oriented manifolds, then a local diffeomorphism
ϕ : M1 → M2 is said to be orientation preserving, if for every p ∈ M1 and every
positive basis v1 , . . . , vm of Tp M1 , we have that dp ϕ(v1 ), . . . , dp ϕ(vm ) is a positive
basis of Tϕ(p) M2 .
Proposition 14.5.4. A connected orientable manifold has two orientations.
Proof. Since M is orientable, there exists an orientation o : M → O(M ) on M .
Now, also −o : M → O(M ) is an orientation on M . Thus there are at least two
orientations. In general, given any other orientation, o1 : M → O(M ), we can look
at the quotient map o/o1 : M → {1, −1}, which returns 1 if the orientations agree,
and −1 if they disagree. This map is smooth; and therefore it is locally constant.
Since M is connected, o/o1 is constant. Therefore, either o1 = o or o1 = −o. 

14.6. Exercises
Exercise 14.1. (a) Let (V, h·, ·i) be an n-dimensional vector space together with
an inner-product. Prove that the following expression defines a positive volume
element on V : q
µ(v1 , . . . , vn ) := det (hvi , vj ii,j ).
(b) Let M ⊂ Rn be an embedded submanifold. For each p ∈ M , we regard Tp M as
a linear subspace of Tp Rn = Rn . The restriction of the standard dot-product on
Rn gives an inner-product on Tp M . Show that the following defines a smooth
density on M :
q
µMp (v 1 , . . . , v m ) := det (hvi , vj ii,j ),
where v1 , . . . , vm is a basis of Tp M .
n
(c) With the notation from (b), prove that µR = |dx1 ∧ . . . ∧ dxn |.
(d) For n = 3, show that the construction from (b) recovers the usual notions of
length of a curve and area of a surface, more precisely:
• Let γ : (a, b) ,→ R3 be a smooth embedding, and let C be the image of
γ. Prove that µC = ds, where, at p = γ(t), dsp denotes the “arc-length
134 IOAN MĂRCUT
, , MANIFOLDS


element”, dγ
dt (t) dt. Conclude that

Z Z b
C

µ = dt (t) dt,

C a

assuming that both sides are bounded (note that the right hand side is
the length of C).
• Let U ⊂ R2 be an open set, and let σ : U ,→ R3 be a smooth embedding
parameterizing the curve S := σ(U ). Prove that µSp = dSp , where dSp is
the “area form”:

∂σ ∂σ
dSp := (u, v) ×
(u, v) dudv, at p = σ(u, v).
∂u ∂v
Conclude that the following holds:
Z Z
S
∂σ ∂σ
µ = ∂u (u, v) × ∂v (u, v) dudv,

S U

assuming that both sides are bounded (note that the right hand side is
the area of S).
Exercise 14.2. Let G be a Lie group with Lie algebra g := Te G. A smooth density
µ ∈ D(G) is called left invariant, if λ∗g (µ) = µ for all g ∈ G (recall that λg denotes
left translation by g). Let DL (G) denote the space of left invariant densities.
(a) Prove that evaluation at the identity
DL (G) −→ Ve(g), µ 7→ µ|p
is a linear isomorphism.
(b) If G is compact, prove
R that there exists a unique left invariant density µG ∈
DL (G) such that G µG = 1 (The density µG is called the Haar measure of
G).
(c) For g ∈ G and µ ∈ DL (G), prove that ρ∗g (µ) ∈ DL (G) (recall that ρg denotes
right translation by g). Moreover, show that ρ∗g (µ) = u(g)µ, where u(g) ∈
R\{0} does not depend on µ 6= 0. Prove that the map u : G → R\{0},
g 7→ u(g) is a Lie group homomorphism.
(d) If G is compact, prove that u(g) = 1, for all g ∈ G.
(e) Consider the Lie group G = Aff(1) consisting of affine transformation of the real
line R, i.e. the elements of Aff(1) are maps f : R → R of the form f (x) = ax+b,
with a 6= 0, and the group structure is the composition of maps. Calculate the
map u : Aff(1) → R\{0}.
Exercise 14.3. Prove that a manifold M is orientable if and only if it admits
an atlas {(Ua , ϕa )}a∈I for which all transition maps ϕa ◦ (ϕb )−1 : ϕb (Ua ∩ Ub ) →
ϕa (Ua ∩ Ub ), for a, b ∈ I, have positive Jacobian determinant:
!
∂xia
det > 0,
∂xjb
where ϕa = (x1a , . . . , xm
a ).

Exercise 14.4. Let M be a connected manifold. Prove that M is orientable iff


O(M ) is disconnected.
LECTURE 14. 135

Exercise 14.5. Let (M, o) be a connected oriented manifold, and let ϕ : M → N be


a local diffeomorphism. Assume that ϕ is not injective, and consider p, q ∈ M , with
p 6= q, such that ϕ(p) = ϕ(q) =: z. Consider the orientation oz ∈ Or(Tz N ) for which
dp ϕ : (Tp M, op ) → (Tz N, oz ) is orientation preserving (i.e. it sends positive bases to
positive bases). If dq ϕ : (Tq M, oq ) → (Tz N, oz ) is not orientation preserving, prove
that N is not orientable.
Exercise 14.6. (a) Show that the n-sphere S n is orientable.
(b) Let τ : S n → S n be the antipodal map, τ (x) = −x. Prove that τ is orientation
preserving iff n is odd.
(c) Show that Pn (R) is orientable iff n is odd.
(d) If n is even, prove that O(Pn (R)) is diffeomorphic to S n .
LECTURE 15

15.1. Manifolds with boundary


The simplest example of a manifold with boundary is the upper half-space
Hn := {(x1 , . . . , xn ) ∈ Rn : xn ≥ 0}.
The boundary of Hn is the n − 1-dimensional space:
∂Hn := Rn−1 × {0} ⊂ Hn ,
and its complement is called the interior of Hn :
int(Hn ) := Rn−1 × (0, ∞) ⊂ Hn .
Clearly, these coincide with the topological notions of boundary and interior as
subsets in Rn , but soon we will introduce the more general class of manifolds
with boundary, which also are made of an “interior” and a “boundary” but which,
intrinsically, are not subsets of a larger space.
The boundary and interior of an open set U ⊂ Hn are defined as follows:
∂U := U ∩ ∂Hn int(U ) := U ∩ int(Hn ).
Note that already in this case, these sets do not coincide with the topological notions
of boundary and interior.
A map f : U → Rm defined on an open subset U ⊂ Hn is called smooth, if
there exists a smooth map fe : U e → Rm , where U e is an open set in Rn containing
U such that f |U = f . If U is already open as a subset of Rn , this is equivalent to
e
the usual smoothness condition on f . The differential of f at p ∈ U is defined as
dp f := dp fe : Rn −→ Rm ,

where fe is an extension of f . The differential does not depend on the chosen


extension of f : at p ∈ int(U ) this is clear; at p ∈ ∂U , this is because dfe is
continuous and int(U ) is dense in U , and therefore:
dp fe = lim
q→p
dq fe = lim
q→p
dq f,
q∈int(U ) q∈int(U )

and clearly the right hand side is independent of fe.


137
138 IOAN MĂRCUT
, , MANIFOLDS

Clearly the composition of smooth maps between open subsets of upper half-
spaces is again smooth (its extension is the composition of the extensions), and the
usual chain rule holds as well (it follows by the chain rule for the extensions).
The notion of a diffeomorphism generalizes accordingly: if U, V ⊂ Hn are
open subsets, then f : U → V is a diffeomorphism if and only if f is smooth,
bijective, and f −1 : V → U is also smooth.
The boundary and the interior are preserved under diffeomorphisms:
Lemma 15.1.1. Let U, V ⊂ Hn be open sets, and let f : U → V be a diffeomor-
phism. Then
f (∂U ) = ∂V f (int(U )) = int(V ).
Proof. Note that f |int(U ) : int(U ) → Rn is an injective local diffeomorphism;
hence f (int(U )) is an open set (as a subset of Rn ). This implies that f (int(U )) ⊂
int(V ). The same argument for f −1 implies that f −1 (int(V )) ⊂ int(U ). Since f is a
bijection, we conclude that f (int(U )) = int(V ). But then f restricts to a bijection
between the complements of these sets; thus, f (∂U ) = ∂V . 

Remark 15.1.2. In fact any homeomorphism f : U → V , where U, V ⊂ Hn are


open sets, satisfies the conclusions of the Lemma: f (∂U ) = ∂V and f (int(U )) =
int(V ). To show this, one needs to find a topological property which differentiates
points on the boundary from interior points. For example, a point p ∈ ∂U has a
basis of neighborhoods which consists of contractible open sets O such that O\{p}
is also contractible. This is not true for a point p ∈ int(U ); actually, if O is any
open neighborhood of p ∈ int(U ) then O\{p} then is not contractible (a small n−1-
sphere around p in O in a nontrivial element of Hn−1 (O; Z), because its image under
the map Hn−1 (O; Z) → Hn−1 (Rn \{p}; Z) ∼ = Z is the generator of this group).
In the usual definition of a smooth manifold, if one considers charts with values
in Hm , one obtains the concept of a manifolds with boundary. Let us give a brief
outline of this construction.
First, we consider atlases with values in Hm :
Definition 15.1.3. Let M be a topological space. A C ∞ -atlas with values in
Hm on M is a family A := {(Uα , ϕα )}α∈I , where
• {Uα }α∈I is an open cover of M ;
• ϕα : Uα → Hm are maps such that ϕα (Uα ) are open sets in Hm , and
ϕα : Uα → ϕα (Uα ) are homeomorphisms,
such that, for all α, β ∈ I, the transition map
ϕα ◦ ϕ−1
β : ϕβ (Uα ∩ Uβ ) → ϕα (Uα ∩ Uβ )

is a diffeomorphism between open sets in Hm .


Two C ∞ -atlases A and B with values in Hm on M are said to be equivalent,
if A ∪ B is again such a C ∞ -atlas with values in Hm .
Definition 15.1.4. A smooth m-dimensional manifold with boundary is a
Hausdorff, second countable topological space M endowed with an equivalence class
of C ∞ -atlases with values in Hm .
Note that a usual m-dimensional manifold can be regarded also as a manifold
with boundary, by choosing an atlas with values in int(Hm ) ∼
= Rm .
LECTURE 15. 139

Definition 15.1.5. Let M be a manifold with boundary. A boundary point of


M is a point p ∈ M such that there exists a chart ϕ : U → Hm around p with
ϕ(p) ∈ ∂Hm . By Lemma 15.1.1, the transition functions preserve ∂Hm ; therefore,
if p is a boundary point, then every chart around p sends p to a point on ∂Hm . The
set of boundary points of M is called the boundary of M is denote by ∂M .
An interior point of M is a point p ∈ M which is not a boundary point, i.e.
for every chart ϕ : U → Hm we have that ϕ(p) ∈ int(Hm ) (which is equivalent to
the existence of at least one such chart). The set of interior points of M is called
the interior of M and is denoted by int(M ).
Thus, any manifold with boundary M has a partition into two subsets M =
∂M t int(M ). Note that int(M ) is open in M and so ∂M is closed in M . These
subsets are manifolds (without boundary):
Proposition 15.1.6. Let M be an m-dimensional manifold with boundary. Then
int(M ) and ∂M are smooth manifolds without boundary of dimensions m and m−1,
respectively. Their differentiable structures are such that, if ϕ : U → Hm is a chart
on M , then
ϕ|int(U ) : int(U ) −→ int(Hm ) and ϕ|∂U : ∂U −→ Rm−1
are smooth charts on int(M ) and ∂M , respectively, where we have denoted int(U ) :=
U ∩ int(M ) and ∂U := U ∩ ∂M , and we have used the identification
∂Hm ∼ = Rm−1 , (x1 , . . . , xm−1 , 0) 7→ (x1 , . . . , xm−1 ).
Example 15.1.7. The intervals [0, 1], (0, 1], [0, 1) and (0, 1) are manifolds with
boundary; all have interior (0, 1), and their respective boundaries are:
∂[0, 1] = {0, 1}, ∂(0, 1] = {1}, ∂[0, 1) = {0}, ∂(0, 1) = ∅.
Example 15.1.8. The closed ball in Rn
B n := {(x1 , . . . , xn ) : (x1 )2 + . . . + (xn )2 ≤ 1}
is a manifold with boundary the n − 1-dimensional sphere and interior the open
ball in Rn :
∂B n = S n−1 , int(B n ) = B n .
This can be checked by applying Exercise 15.1 to the map
f : Rn −→ R, f (x1 , . . . , xn ) = 1 − ((x1 )2 + . . . + (xn )2 ).
All the construction we have performed for manifold without boundary gener-
alize to manifolds with boundary. We briefly enumerate what one needs to keep
track of. Let M and N be manifolds with boundary. Then
• The notion of a smooth map generalizes accordingly: f : M → N is
smooth iff it is continuous and
ϕ ◦ f ◦ ψ −1 : ψ(U ∩ f −1 (V )) −→ Hn
is smooth for every pair of charts chart ψ : U → Hm and ϕ : V → Hn on
M and N , respectively.
• The tangent space Tp M at a point p is defined as for usual manifolds. Even
at p ∈ ∂M , one has that dim(Tp M ) = m. The tangent bundle π : T M →
M is a 2m-dimensional manifold with boundary ∂T M = π −1 (∂M ). Note
that ∂T M 6= T ∂M : dim(∂T M ) = 2m − 1, but dim(T ∂M ) = 2m − 2. The
differential of smooth maps generalizes accordingly.
140 IOAN MĂRCUT
, , MANIFOLDS

• A vector bundle π : E → M is defined exactly as in Definition 8.2.1,


just that E is also a manifold with boundary ∂E = π −1 (∂M ).
• The density bundle D(M ), the orientation bundle O(M ), the cotangent
Vk ∗
bundle T ∗ M and its exterior powers T M are constructed exactly as in
the case of manifold without boundary (note that all these were obtained
by applying a linear algebra construction to Tp M for all p ∈ M ).
• The exterior derivative d : Ω(M ) → Ω(M ) is defined in the same way,
and has the usual properties. One can consider the associated de Rham
cohomology HdR (M ) (one can show that HdR (M ) ∼ = HdR (int(M ))).
• There is a canonical integration
R of compactly supported densities defined
as for usual manifolds M : Dc (M ) → R.
Vectors along the boundary, not tangent to the boundary, are of two types:
Definition 15.1.9. Let p ∈ ∂M . A vector v ∈ Tp M such that v ∈ / Tp ∂M is said
to point inwards if there exists a smooth curve
d
γ : [0, ) −→ M such that γ(0) = p and γ(t) t=0 = v,
dt
and v is said to point outwards, if −v points inwards.
In other words, the hyperplane Tp (∂M ) divides Tp M into two components:
vectors pointing inwards and vectors pointing outwards. Note that, for p ∈ ∂M ,
and v ∈ Tp M precisely one of the following situations holds: v points inwards,
v points outwards, or v ∈ Tp (∂M ). In a chart, ϕ : U → Hm around p, with
dp ϕ(v) = (v1 , . . . , vm ) ∈ Tϕ(p) Rm ∼
= Rm this can be checked as follows:
• v points inwards iff vm > 0 iff dp ϕ(v) ∈ int(Hm );
• v points outwards iff vm < 0 iff dp ϕ(v) ∈ / Hm ;
• v is tangent to ∂M iff vm = 0 iff dp ϕ(v) ∈ ∂Hm .
In other words, the one-dimensional vector space Tp M/Tp ∂M , which consists of
“directions transverse to ∂M ”, comes with a canonical orientation. This implies
that an orientable manifold with boundary has an orientable boundary, as the
following construction shows:
Definition 15.1.10. Let (M, o) be an oriented manifold with boundary. The
boundary orientation ∂o is the orientation on the manifold ∂M defined on the
basis v1 , . . . , vm−1 of Tp ∂M by:
∂o(v1 , . . . , vm−1 ) = o(v, v1 , . . . , vm−1 ),
where v ∈ Tp M is any outwards pointing vector, and m = dim(M ).
Example 15.1.11. Consider the manifold with boundary [0, 1]. Consider the ori-

entation o on [0, 1], such that ∂t ∈ Tt [0, 1] is a positive basis for all t ∈ [0, 1].
Outwards pointing vectors at the boundary ∂[0, 1] = {0} ∪ {1} are given by:
∂ ∂
− t=0 ∈ T0 [0, 1] and ∈ T1 [0, 1].
∂t ∂t t=1
Recall that an orientation of a 0-dimensional vector space is the same as the choice
of a sign (see the comments after Definition 14.5.1). The boundary orientation ∂o
consists of the following two signs:
 ∂  ∂ 
(∂o)0 = o − t=0 = −1 and (∂o)1 = o = +1.
∂t ∂t t=1
LECTURE 15. 141

The following result describes a manifold in a neighborhood of its boundary.


Theorem 15.1.12. (Collar Neighborhood Theorem) Let M be a manifold with
boundary. There exists a neighborhood U ⊂ M of ∂M and a diffeomorphism ϕ :

∂M × [0, 1) −→ U , such that ϕ(x, 0) = x for all x ∈ M .
For the proof of the above, we will use:
Lemma 15.1.13. Let M be a manifold with boundary. There exists a vector field
X ∈ X(M ) such that Xp points inwards, for all p ∈ ∂M .
Proof. Let A = {(Ui , ϕi )}i∈I be a smooth atlas on M with valued Hm . In each
chart, consider the inwards pointing vector field Xi := (ϕi )∗ ( ∂x∂m ) ∈ X(Ui ). Let

Pi }i∈I be a partition of unity subordinated to the cover {Ui }i∈I . Define X :=
i∈I ρi Xi ∈ X(M ). Since each Xi is pointing inwards, it follows that also X
points inwards (note that pointing inwards is a convex condition). 

We prove Theorem 15.1.12 only in the case of compact boundary:

Proof of Theorem 15.1.12 for compact ∂M . Let X ∈ X(M ) be a vector field


pointing inwards. Because X points inwards, there is a neighborhood U ⊂ [0, ∞) ×
M of {0} × M on which the flow is defined:
φX : U −→ M, (t, p) 7→ φtX (p).
Consider the restriction of the flow to the boundary of M :
ψ : U ∩ [0, ∞) × ∂M −→ M, ψ(t, p) := φtX (p).


For any p ∈ ∂M , we have that ψ(0, p) = p, hence d(0,p) ψ(0, v) = v for any v ∈
Tp ∂M . On the other hand, since the curve t 7→ ψ(t, p) is a flow line of X,
∂ 
d(0,p) ψ
, 0 = Xp .
∂t
Since Tp M = Tp ∂M ⊕ RXp , we conclude that d(0,p) ψ is a linear isomorphism. The
Inverse Function Theorem 6.1.1, implies that ψ is a diffeomorphism in a neigh-
borhood of p. On the other hand, since ∂M is compact and ψ(0, p) = p for all
p ∈ ∂M , Exercise 6.8 implies that ψ is a diffeomorphism between a neighborhood
V ⊂ U of {0} × ∂M and a neighborhood O ⊂ M of ∂M . On the other hand,
since ∂M is compact, there exists  > 0 such that [0, ) × ∂M ⊂ V . Finally, define
ϕ : [0, 1) × ∂M → M , ϕ(t, p) = ψ(t, p). 

15.2. Integration of differential forms on oriented submanifolds


The top power of V ∗ is closely related to volume elements on V . Proposition 12.1.10
implies the following:
Lemma 15.2.1. Let V be a vector space of dimension n. An orientation o ∈ Or(V )
induces a linear isomorphism:
n

^
V∗ −→ Ve(V ), ω 7→ oω,

(oω)(v1 , . . . , vn ) := o(v1 , . . . , vn )ω(v1 , . . . , vn ),


for all (v1 , . . . , vn ) ∈ B(V ).
142 IOAN MĂRCUT
, , MANIFOLDS

Consider an oriented manifold (M, o) of dimension m. The above lemma implies


that the orientation o gives a vector bundle isomorphism
m

^
T ∗M −→ D(M ), ω 7→ oω.

Definition 15.2.2. Let (M, o) be an oriented manifold of dimension m. The in-


tegral of a compactly supported top-form ω ∈ Ωm c (M ) on M is defined as the
integral of the compactly supported density oω ∈ Dc (M ):
Z Z
ω := oω.
(M,o) M

Remark 15.2.3. If the orientation changes, the sign of the integral changes:
Z Z
ω=− ω.
(M,−o) (M,o)

In notation, if the orientation o on M is fixed, it is common to omit the orienta-


tion under the orientation under the integral symbol; we also adopt this convention,
and write Z Z
ω instead of ω.
M (M,o)

Integration of forms is invariant under orientation preserving diffeomorphisms:


Proposition 15.2.4. Let M and N be connected oriented manifolds, and let ϕ :
M − ∼
→ N be a diffeomorphism. Then, for all compactly supported top-forms ω ∈
m
Ωc (N ), we have that
Z Z

ϕ (ω) =  ω,
M N
where  = 1 if ϕ is orientation preserving, and  = −1 if ϕ is orientation reversing.
Proof. Let oM and oN be the respective orientations. We have that following
equality of densities: ϕ∗ (oN ω) = oM ϕ∗ (ω). Using that integration of densities is
invariant under diffeomorphisms (see Theorem 14.4.3), we obtain the result:
Z Z Z Z Z
ϕ∗ (ω) = oM ϕ∗ (ω) =  ϕ∗ (oN ω) =  oN ω =  ω.
(M,oM ) M M N (N,oN )

Forms of lower degree can be integrated on oriented submanifolds:


Definition 15.2.5. Let M be an m-dimensional manifold. Consider an embedded
submanifold N ⊂ M of dimension n, which is oriented, and consider an n-form
η ∈ Ωn (M ). If supp(η) ∩ N is compact, we define the integral of η over N by:
Z Z
η := i∗ (η),
N (N,o)

where o is the orientation on N , and i : N → M denotes the inclusion map.


Note that supp(i∗ (η)) = supp(η) ∩ N ; thus the compactness assumption is
equivalent to i∗ (η) being compactly supported.
LECTURE 15. 143

15.3. Stokes’ Theorem


Stokes’ Theorem is the higher-dimensional version of the Fundamental Theorem of
Calculus:
Theorem 15.3.1 (Stokes’ Theorem). Let (M, o) be an oriented manifold with
boundary, and consider on ∂M the induced orientation ∂o. Then, for any com-
pactly supported m − 1-form ω ∈ Ωm−1
c (M ), where m = dim(M ), we have that:
Z Z
dω = ω.
(M,o) (∂M,∂o)
m
Proof. We first prove the result for H endowed with the standard orientation
ost . Consider ω ∈ Ωm−1
c (Hm ). Then ω can be written uniquely as the sum:
m
X
ω= fi (x)dx1 ∧ . . . ∧ dx
di ∧ . . . ∧ dxm ,
i=1

where the coefficients fi ∈ Cc∞ (Hm ) are compactly supported function on Hm , and
di denotes that dxi is missing in the product. The differential of ω is given by:
dx
m
!
i−1 ∂fi
X
dω = (−1) i
(x) dx1 ∧ . . . ∧ dxm .
i=1
∂x

Note that, under the standard orientation on Hm , we have that dx1 ∧ . . . ∧ dxm
corresponds to the density |dx1 ∧ . . . ∧ dxm |, and therefore:
Z m Z
X ∂fi
dω = (−1)i−1 i
(x)dx1 . . . dxm .
Hm
i=1 Hm ∂x
Choosing a large enough number N > 0 such that
supp(ω) ⊂ [−N, N ]m−1 × [0, N ].
Consider a term in the sum above with 1 ≤ i < m. By interchanging the order of
integration, we first calculate the integral with respect to dxi :
Z N
∂fi
i
(x)dxi = fi (x1 , . . . , xi−1 , N, xi+1 , . . . , xm )
−N ∂x
− fi (x1 , . . . , xi−1 , −N, xi+1 , . . . , xm ) = 0,
where we have used the Fundamental Theorem of Calculus and that supp(fi ) ⊂
[−N, N ]m−1 × [0, N ]. The term with i = m gives:
Z N
∂fm
m
(x)dxm = fm (x1 , . . . , xm−1 , N ) − fm (x1 , . . . , xm−1 , 0) =
0 ∂x
= −fm (x1 , . . . , xm−1 , 0).
We conclude that
Z Z
m
(*) dω = (−1) fm (x1 , . . . , xm−1 , 0)dx1 . . . dxm−1 .
Hm Rm−1
m
On the other hand, if i : ∂H → Hm denotes the inclusion, we have that:
i∗ (ω) = fm (x1 , . . . , xm−1 , 0)dx1 ∧ . . . ∧ dxm−1 .
144 IOAN MĂRCUT
, , MANIFOLDS

Note that the vector − ∂x∂m is pointing outwards, therefore, the boundary orienta-
tion on ∂Hm is such that
 ∂ ∂   ∂ ∂ ∂ 
(∂ost ) , . . . , = o st − , , . . . , = (−1)m .
∂x1 ∂xm−1 ∂xm ∂x1 ∂xm−1
We obtain that
(∂ost )dx1 ∧ . . . ∧ dxm−1 = (−1)m |dx1 ∧ . . . ∧ dxm−1 |,
and therefore
Z Z Z
∗ m
ω= (∂ost ) i (ω) = (−1) fm (x1 , . . . , xm−1 , 0)dx1 . . . dxm−1 .
∂Hm ∂Hm Rm−1
m
This and (∗) prove Stokes’ Theorem for H with the standard orientation.
Consider an oriented manifold (M, o) with boundary. Let ω ∈ Ωm−1 c (M ).
Assume first that the support of ω lies inside the domain of a coordinate chart
ϕ : U → Hm , such that U and ∂U := U ∩ ∂M are connected. By Proposition 15.2.4
and Stokes’ Theorem for Hm , we obtain that
Z Z Z Z Z
−1 ∗ −1 ∗
dω = dω =  (ϕ ) (dω) =  d(ϕ ) (ω) =  d(ϕ−1 )∗ (ω) =
M U ϕ(U ) ϕ(U ) Hm
Z Z
−1 ∗
= (ϕ ) (ω) =  (ϕ−1 )∗ (ω),
∂Hm ∂ϕ(U )

where  = 1 if ϕ : U → ϕ(U ) is orientation preserving, and otherwise  = −1. If


∂ϕ(U
R ) = ∅, then the last term is zero, but also U ∩ ∂M = ∅, and therefore also
∂M
ω = 0. On the other hand, if ∂ϕ(U ) 6= ∅, then note that ϕ : U → ϕ(U ) is
orientation preserving if and only if ϕ|∂U : ∂U → ∂ϕ(U ) is orientation preserving;
therefore, applying Proposition 15.2.4 to ϕ|∂U , we obtain that
Z Z Z
 (ϕ−1 )∗ (ω) = ω= ω.
∂ϕ(U ) ∂U ∂M
R R
In both cases, we obtain that M dω = ∂M ω.
In general, using partitions of unity, we can decompose any ω ∈ Ωcm−1 (M ) as a
finite sum ω = ω1 + . . . + ωl , with ωi ∈ Ωm−1
c (M ) having support inside the domain
of a coordinate chart with the properties from above. Applying Stokes’ Theorem to
each ωi , and summing up the resulting equalities, we obtain that Stokes’ Theorem
holds for ω. 
Example 15.3.2. Consider the closed interval [a, b] with the standard orientation.
Stokes’ Theorem in this case gives
Z b Z
dF = F, for all F ∈ Ω0 ([a, b]) = C ∞ ([a, b]).
a {a,b}
Rb
The left hand side is just a F 0 (t)dt, while the right hand side gives oa F (a) +
ob F (b) = F (b) − F (a), where oa , ob ∈ {−1, 1} are the induced orientations on the
boundary (see example 15.1.11). Thus, in this case, Stokes’ Theorem is equivalent
to the Fundamental Theorem of Calculus:
Z b
f (t)dt = F (b) − F (a), where f (t) = F 0 (t).
a

For manifolds without boundary, Stokes’ Theorem gives:


LECTURE 15. 145

Corollary 15.3.3. Let M be a compact oriented manifold without boundary of


m−1
R
dimension m. Then, for any m − 1 form ω ∈ Ω (M ), we have that M
dω = 0.
Equivalently, if µ ∈ Ωm (M ) satisfies M µ 6= 0, then its cohomology class [µ] ∈
R
m
HdR (M ) is non-trivial [µ] 6= 0.
We have that:
Corollary 15.3.4. Let M be a compact oriented manifold without boundary of
m
dimension m. Then HdR (M ) 6= 0.
Proof. Let p ∈ M , and consider an oriented chart (U, ϕ = (x1 , . . . , xm )), around p.
Let χ : M → [0, 1] be smooth function with compactR support in U , and such that
χ(p) > 0. Then µ = χ · dx1 ∧ . . . ∧ dxm satisfies M µ > 0, thus, by the previous
m
Corollary, [µ] 6= 0 in HdR (M ). 

15.4. Exercises
Exercise 15.1. Let M be a manifold without boundary, and let f : M → R be a
smooth function. If 0 ∈ R is a regular value of f , prove that N := f −1 ([0, ∞)) can
be given the structure of a manifold with boundary ∂N = f −1 (0) such that the
inclusion N → M is smooth. Hint: use the local submersion Theorem 6.2.3.
Exercise 15.2. Show that Stokes’ Theorem implies the following classical results
from Calculus:
(a) Green’s Theorem. Let Ω ⊂ R2 be an open set bounded by the simple closed
smooth curve C. For every smooth vector field F = (f1 , f2 ) on R2 , we have
that: I ZZ 
∂f2 ∂f1 
f1 dx + f2 dy = − dxdy.
C Ω ∂x ∂y
(b) Divergence Theorem. Let Ω ⊂ R3 be a bounded open set with smooth boundary.
Then for every smooth vector field F on R3 , we have that:
ZZZ ZZ
divF dv = F · n dA.
D ∂D
3
(c) Classical Stokes’ Theorem. Let S ⊂ R be a smooth bounded surface with
smooth boundary. Then for every smooth vector field F = (f1 , f2 , f3 ) on R3 ,
we have that:
I ZZ
f1 dx + f2 dy + f3 dz = curl(F) · n dA.
∂S S
Use Exercise 13.2 and a Calculus book (or Wikipedia) to make sense of all terms.
Exercise 15.3. Let U ⊂ C be an open set, and let f : U → C be a holomorphic
function. Consider a closed disk D ⊂ U with center a and boundary C. Prove
Cauchy’s integral formula:
I
1 f (z)
f (a) = dz.
2πi C z − a
Hint: Use Exercise 13.4 and Stokes’ Theorem.
BIBLIOGRAPHY

1. L. Conlon, Differentiable manifolds. Reprint of the 2001 second edition. Modern


Birkhäuser Classics. Birkhäuser Boston, Inc., Boston, MA, 2008.
2. J. J. Duistermaat, J. Kolk, Lie groups, Universitext, Springer-Verlag Berlin,
2000.
3. R. Loja Fernandes, Differentiable Manifolds I, http://www.math.illinois.
edu/~ruiloja/Math518/notes.pdf.
4. V. Guillemin, A. Pollack, Differential topology. Prentice-Hall, Inc., Englewood
Cliffs, N.J., 1974.
5. N. Hitchin, Differentiable manifold, lecture notes, http://people.
maths.ox.ac.uk/hitchin/hitchinnotes/Differentiable_manifolds/
manifolds2012.pdf
6. M. Kervaire, A manifold which does not admit any differentiable structure,
Coment. Math. Helv., (1960) 34, 257–270.
7. J.W. Milnor, On manifolds homeomorphic to the 7-sphere, Ann. of Math. (2)
64 (1956): 399–405.
8. J.W. Milnor, Topology from the differentiable viewpoint, Princeton University
Press, Princeton, NJ, 1997.
9. E. Moise, Affine structures in 3-manifolds. V. The triangulation theorem and
Hauptvermutung Ann. of Math. (2) 56 (1952), 96–114.
10. L. Nicolaescu, Lectures on the Geometry of Manifolds, http://www3.nd.edu/
~lnicolae/Lectures.pdf
11. W. Rudin, Principles of mathematical analysis. Third edition. International
Series in Pure and Applied Mathematics. McGraw-Hill Book Co., New York-
Auckland-Düsseldorf, 1976.
12. J. Stallings, The piecewise-linear structure of Euclidean space, Proc. Cambridge
Philos. Soc. 58 (3), 1962, 481–488.
13. C.H. Taubes, Gauge theory on asymptotically periodic 4-manifolds, J. of Diff.
Geom. 25 (3), (1987), 363–430.
14. W. Thurston, Three-dimensional geometry and topology. Vol. 1. Edited by
Silvio Levy. Princeton Mathematical Series, 35. Princeton University Press,
Princeton, NJ, 1997.
15. H. Whitney, Differentiable manifolds, Ann. of Math. (2) 37 (1936), no. 3, 645–
680.
16. https://en.wikipedia.org/wiki/Differential_structure
147
148 IOAN MĂRCUT
, , MANIFOLDS

17. https://en.wikipedia.org/wiki/Long_line_(topology)
18. http://pages.uoregon.edu/koch/math431/LongLine.pdf
19. https://commons.wikimedia.org/wiki/File:Orientation_cover_of_
Mobius_strip.webm
20. https://en.wikipedia.org/wiki/Prfer_manifold
21. https://en.wikipedia.org/wiki/Space-filling_curve
22. http://earth.nullschool.net/
EXERCISES AND ASSIGNMENTS

Problems for exercise classes


• Exercise class 11/9: 1.1, 1.2, 1.6, 1.7.
• Exercise class 18/9: 2.1, 2.2, 2.3.
• Exercise class 25/9: 3.2, 3.9.
• Exercise class 2/10: 4.1, 4.3.
• Exercise class 9/10: 5.1, 5.3.
• Exercise class 16/10: 6.3, 6.6.
• Exercise class 23/10: 6.5, 6.11.
• Exercise class 13/11: 8.1, 8.2, 8.3.
• Exercise class 20/11: 9.3, 9.7, 9.8.
• Exercise class 27/11: 10.2,10.3,10.4.
• Exercise class 4/12: 11.3.
• Exercise class 11/12: 12.3,12.9.
• Exercise class 18/12: 13.4, 13.5, 13.6.
• Exercise class 8/1: 14.1, 14.6.
• Exercise class 15/1: 15.2.

Homework assignments
• Homework for 15/9: 1.2, 1.7.
• Homework for 22/9: 2.3.
• Homework for 29/9: 3.9.
• Homework for 6/10: 4.3.
• Homework for 13/10: 5.1.
• Homework for 20/10: 6.3, 6.6.
• Homework for 27/10: 6.5, 6.11.
• Homework for 17/11: 8.1, 8.2, 8.3.
• Homework for 24/11: 9.3, 9.7, 9.8.
• Homework for 1/12: 10.2,10.3,10.4.
• Homework for 8/12: 11.3.
• Homework for 15/12:12.3,12.9.
• Homework for 22/12: 13.4, 13.5, 13.6.
• Homework for 12/1: 14.1, 14.6.
• Homework for 19/1: 15.2.

149

You might also like