Differential Geometry

Based on lectures by Marco Zambon in 2019–20

Notes taken by Gilles Castel, proofread by Marco Zambon
Contents

1 Differentiable manifolds 3
1.0 Topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Differentiable manifolds . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Differentiable map . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Partition of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Tangent vectors 11
2.1 Tangent vectors, tangent spaces . . . . . . . . . . . . . . . . . . . 11
2.2 The derivative of a map . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 The regular level set theorem . . . . . . . . . . . . . . . . . . . . 14
2.4 Tangent vectors as derivations . . . . . . . . . . . . . . . . . . . . 17

3 Vector fields 19
3.1 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Integral curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 The Lie bracket of vector fields . . . . . . . . . . . . . . . . . . . 21
3.4 Interpretation of the Lie bracket . . . . . . . . . . . . . . . . . . 23

4 Bundles 26
4.1 Fiber bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2 Vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Differential forms and integration 31

5.1 Forms on vector spaces . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Differential forms on manifolds . . . . . . . . . . . . . . . . . . . 32
5.3 Orientation and volume forms . . . . . . . . . . . . . . . . . . . . 34
5.4 Integration on manifolds . . . . . . . . . . . . . . . . . . . . . . . 35

6 Exterior derivative and Stokes theorem 37

6.1 The exterior derivative in Rm . . . . . . . . . . . . . . . . . . . . 37
6.2 The exterior derivative on manifolds . . . . . . . . . . . . . . . . 38
6.3 Manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . 39
6.4 Stokes’ theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

7 De Rham Cohomology 42
7.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.2 Homotopic maps and cohomology . . . . . . . . . . . . . . . . . . 44
7.3 The Mayer–Vietoris theorem . . . . . . . . . . . . . . . . . . . . 46

1
8 Foliations 47
8.1 Immersed submanifolds . . . . . . . . . . . . . . . . . . . . . . . 47
8.2 Distributions and involutivity . . . . . . . . . . . . . . . . . . . . 48
8.3 The Frobenius theorem . . . . . . . . . . . . . . . . . . . . . . . 48
8.4 Foliations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

9 Lie groups and Lie algebras 52

9.1 Lie groups, Lie subgroups . . . . . . . . . . . . . . . . . . . . . . 52
9.2 From Lie groups to Lie algebras . . . . . . . . . . . . . . . . . . . 55
9.3 The exponential map . . . . . . . . . . . . . . . . . . . . . . . . . 57
9.4 From Lie algebras to Lie groups . . . . . . . . . . . . . . . . . . . 58

CONTENTS 2
Chapter 1

Differentiable manifolds

1.0 Topological spaces

Definition 1.1 (Topological space). A topological space is (X, σ) where X
is a set and σ a family of subsets of X, called open sets, such that:

• ∅, X ∈ σ
• i∈I Ui ∈ σ whenever Ui ∈ σ for all i
S

• i<n Ui ∈ σ whenever Ui ∈ σ for all i

T

Let (X, σ) be a topological space.

Definition 1.2 (Open neighbourhood). An open subset that contains p ∈ X

is called a (open) neighbourhood of p.

Definition 1.3 (Subspace topology). If Y ⊂ X then (Y, σY ) is a topological

space, where
σY = {U ∩ Y | U ∈ σ}.
We call σY the subspace topology.

Example. Endowing R2 with the Euclidean topology, the subspace topology on

R × {0} ⊂ R2 is also the Euclidean topology. 

Definition 1.4 (Quotient topology). Let ∼ be an equivalence relation on X.

Consider π : X → X/ ∼. Then X/ ∼ is a topological space, where the
open sets are by definition sets such that π −1 (U ) is open in X.

Definition 1.5 (Continuous functions). A function f : X1 → X2 is called

continuous iff ∀U ∈ σ2 : f −1 (U ) ∈ σ1 .

3
 Definition 1.6. A topological space is called Hausdorff iff ∀x, y ∈ X, there
exist neighbourhoods U of x, V of y such that U ∩ V = ∅.

Example. Endow R2 \ {0} with the equivalence relation given by the thick lines
and the two half lines in the following figure. That is:
(
0 0 x = x0 if x 6= 0
(x, y) ∼ (x , y ) ⇔
yy 0 > 0 if x = 0.

Then the quotient topology on (R2 \ {0})/ ∼ is not Hausdorff. 

X = R2 \ {0}

X
∼

Figure 1.1: Example of a topology which is not Hausdorff

Definition 1.7 (Basis for a topology). A basis for a topology is S ⊂ σ such

that every open set of X is a union of elements of S.

Definition 1.8 (C2). A space (X, σ) is second countable if there exists a

countable basis.

Example. Rn is second countable. Indeed {B m1 (x) | x ∈ Qn , m ∈ N} is a

countable basis for the topology. Here Br (x) is the open ball with radius r
around x. 

1.1 Differentiable manifolds

Definition 1.9 (Topological manifold). A topological manifold M of dimen-
sion m is a second countable, Hausdorff topological space which is locally
homeomorphic to Rm .

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 4

Remark. ‘Locally homeomorphic to Rm ’ means that ∀p ∈ M, there exists a
neighborhood U of p and a homeomorphism φ : U → V ⊂ Rm . Recall that
open
homeomorphism means: bijective map that is continuous in both directions.

Definition 1.10 (Chart). The pair (U, φ) is called a chart.

Remark.
• Any subset of a Hausdorff space is Hausdorff

• Any subset of a C2 space is C2.

Definition 1.11 (Compatibility). Two charts (U1 , φ1 ) and (U2 , φ2 ) are called
smoothly compatible if

φ2 ◦ (φ1 )−1 |φ1 (U1 ∩U2 ) : φ1 (U1 ∩ U2 ) → φ2 (U1 ∩ U2 )

is a diffeomorphism (i.e. bijective, differentiable and inverse differentiable,

where “differentiable” means that all partial derivatives exist).

U2
U1

φ1
φ2

V1
Rm V2
Rm
φ2 ◦ φ−1
1

Figure 1.2: Compatible charts

Definition 1.12 (Smooth atlas). S A smooth atlas for M is called a collection

of charts {(Uα , φα )} such that α Uα = M and any two charts are smoothly
compatible.

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 5

 Definition 1.13 (Maximal smooth atlas). A smooth atlas A is maximal if:
whenever B is a smooth atlas and A ⊂ B then B = A.

Definition 1.14 (Differentiable manifold ). A differentiable manifold (also

called a smooth manifold) is a topological manifold M together with a
maximal smooth atlas.

Remark. Given a smooth atlas A on a topological manifold M , there exists a

unique maximal smooth atlas containing it, namely

{(V, ψ) | (V, ψ) is smoothly compatible with all charts of A}.

Example. Let U ⊂ Rn . It’s a smooth manifold: an atlas is {(U, Id)}. Take the
maximal smooth atlas containing it. 
Example. Let S n := {x ∈ Rn+1 | kxk = 1}. The sphere S n with the subspace
topology is Hausdorff and C2, simply because Rn+1 is. Two charts are given by
the stereographic projections from the Northpole N and Southpole S:

(x1 , . . . , xn )
φN : S n \ {N } → Rn : (x1 , . . . , xn+1 ) 7→
1 − xn+1
(x1 , . . . , xn )
φS : S n \ {S} → Rn : (x1 , . . . , xn+1 ) 7→ .
1 + xn+1

Now, φN and φS are homomorphisms. Furthermore, kφN (p)k · kφS (p)k = 1,

which allows us to calculate the inverse of φN . Hence
y
(φS ◦ φ−1 n n
N )|φN (S n \{N,S}) : R \ {0} → R \ {0} : y 7→ ,
kyk2

so φN and φS are smoothly compatible. Take the maximal smooth atlas con-
taining φN and φS . 
Remark. We could have started with other points P, Q ∈ S n instead of N, S.
The smooth atlases {φP , φQ } and {φN , φS } would be different, but they define
the same maximal smooth atlas.

1.2 Differentiable map

Let M be a smooth manifold.

Definition 1.15 (Smooth function). A function f : M → R is differentiable

(or smooth) at p ∈ M iff ∃ a chart (U, φ) around p such that f ◦ φ−1 is
differentiable in φ(p).

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 6

 M
f
U R
p

φ(U )

Rn

Figure 1.3: Smooth function from M to R

Remark. If f ◦ φ−1 is differentiable at p for a chart (U, φ), then f ◦ ψ −1 is also

differentiable at p, for any other chart (V, ψ) (in the maximal atlas of M ).

Proof. We want to argue that f ◦ ψ −1 is smooth.

f ◦ ψ −1 = (f ◦ φ−1 ) ◦ (φ ◦ ψ −1 ) .
| {z } | {z }
C∞ C∞

Notation. We write C ∞ (M ) to denote all smooth functions M → R.

Definition 1.16 (Smooth map). f : M → N is differentiable at p ∈ M iff

• it is continuous
• there exists charts (UM , φM ) around p and (UN , φN ) around f (p)
such that φN ◦ f ◦ φ−1
M is differentiable at φM (p)

Remark. The map φN ◦f ◦φ−1 M is defined on φM (UM ∩f

−1
(UN )). The continuity
of f ensures that this is an open neighborhood of φM (p) in Rm , hence it makes
sense to talk about the differentiability of the above map at φM (p).
Remark. A map f being a differentiable and a homeomorphism does not im-
ply that f is a diffeomorphism (which also include the differentiability of the
inverse). For√instance, f : R → R, x 7→ x3 is not a diffeomorphism, because the
inverse x 7→ 3 x is not differentiable at zero.

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 7

 M N
UM f UN
p f (p)

φM
φN

φN ◦ f ◦ φ−1
M
Rm Rn

Figure 1.4: Smooth function from M to N

1.3 Partition of unity

Definition 1.17 (Partition of unity). A partition of unity is a family {eα }α∈A
of smooth functions eα : M → [0, 1] such that
• For all p ∈ M, there exists a neighborhood U of p such that the set
{α ∈ A : eα |U 6≡ 0} is finite.
•
P
eα ≡ 1

Definition 1.18 (Subordinate). Let {Uα }α∈A be an open cover of M . A

partition of unity {eα }α∈A is subordinate to the cover ⇔ supp(eα ) ⊂ Uα ,
where
supp(eα ) = {p ∈ M : eα (p) 6= 0}.

Proposition 1.19. Let {Uα }α∈A be an open cover of M . Then there exists
a partition of unity subordinate to it.

This is useful in the following way: we can define functions, or vector fields
sα on open subsets of M which locally look like Rn . Then we create a partition
of
P unity subordinate to this cover, and paste the functions smoothly together:
eα · sα . This gives a smooth function (or vector field) on the whole of M .
The proof uses that the topology of M is second countable, which is one of the
reasons of requiring C2 in the definition of a manifold.

Proof. For all q ∈ M , choose α ∈ A such that q ∈ Uα . Let ψq be a functiona

such that ψq (q) = 1 supported in Uα . This is always possible, since the
manifold locally looks like Rn .
Since M is compact, the open cover {(ψq )−1 (R>0 )}q∈M has a finite

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 8

 Pm
subcover. So ∃q1 , . . . , qm ∈ M such that ψ = i=1 ψqi > 0 on M . Now,
ψ
define φi = ψqi , which satisfies φi = 1 and supp(φi ) = supp(ψqi ) ⊂ Uα
P
for some α ∈ A. Now, one can rearrange these functions such that they are
indexed by A.
a Such functions are sometime called bump functions.

1.4 Submanifolds
Let N be a smooth manifold of dimension n.

Definition 1.20 (Submanifold). A subset M ⊂ N is a submanifold of dimen-

sion m if ∀p ∈ M there exists a chart (U, φ) such that φ(U ) ∩ (Rm × {0}) =
φ(U ∩ M ).

p M {0} × Rn−m
φ

Rn φ(U )

Rm × {0}

Figure 1.5: Definition of a submanifold

Remark. A chart as above is called adapted to M .

Remark. The set M itself is a smooth manifold, with smooth atlas given by
{(U ∩ M, φ|U ∩M ) : (U, φ) is a chart of N adapted to M }.
Remark. Submanifolds of the same dimension as M are exactly the open sets
of M , and those with dimension 0 are exactly the points.
Example. The union of the axes {0} × R ∪ R × {0} is not a submanifold of R2 .
The problematic point is the origin (the “cross”). 
Example. The unit circle S 1 ⊂ R2 is a submanifold of R2 . The idea is that, for
each point of S 1 , we need to find a chart that ‘flattens’ a part of the circle to a
line. Define
φa : R2 − (R≥0 × {0}) −→ (0, 2π) × R+ φb : R2 − (R≤0 × {0}) −→ (−π, π) × R+
   
r cos θ r cos θ
7−→ (θ, r) 7−→ (θ, r).
r sin θ r sin θ
Note that 0 is not included in either of the charts, but that is not a problem.
We have that φa (S 1 − {(1, 0)}) = (0, 2π) × {1} and φb (S 1 − {(−1, 0)}) =
(−π, π) × {1}. This is already flat, but per definition of submanifold, we need
to move this to zero. So instead of {φa , φb }, use {φa − (0, 1), φb − (0, 1)}.

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 9

 R2
1
φa

0 2π

1 R2

φb

−π 0 π

Figure 1.6: A circle is a submanifold of the plane.


Example. Consider the torus, S × S , where S = {z ∈ C : |z| = 1} Fix α ∈ R.
1 1 1

Now consider
M = {(e2πit , e2πiαt ) : t ∈ R},
a subset of the torus. This is a submanifold of the torus iff α ∈ Q (this happens
exactly when the spiral closes up). When α ∈ / Q, M is dense in the torus. 

Figure 1.7: Two examples of submanifolds of the torus. On the left, α = 7

2 and
on the right α = 4.

CHAPTER 1. DIFFERENTIABLE MANIFOLDS 10

Chapter 2

Tangent vectors

2.1 Tangent vectors, tangent spaces

Note. This and the next section do not follow neither Lee’s nor Tu’s book.
Remark. Given a submanifold M of Rn , one can define tangent vectors to M
at some point p ∈ M as the collection of dt
d
0
γ(t) where γ : (−ε, ε) → Rn is
smooth and Im γ ⊂ M , γ(0) = p. This uses the ambient space, hence to define
tangent vectors to manifolds we cannot proceed in the same way.
Let M be a smooth manifold of dimension m.

Definition 2.1 (Tangent vector). A tangent vector of M at p is an equiva-

lence class [γ] of smooth curves γ : (−ε, ε) → M with γ(0) = p, where

γ1 ∼ γ2 ⇔ ∃ a chart (U, φ) containing p s.t. (φ ◦ γ1 )0 (0) = (φ ◦ γ2 )0 (0).

Lemma 2.2. If (φ ◦ γ1 )0 (0) = (φ ◦ γ2 )0 (0) for a chart φ, then the same holds
for all charts.

Proof. The linear map Dφ(p) (ψ ◦ φ−1 ) : Rm → Rm sends (φ ◦ γi )0 (0) to

(ψ ◦ γi )0 (0) because of the chain rule, for i = 1, 2. Now, as (φ ◦ γ1 )0 (0) =
(φ ◦ γ2 )0 (0), the vectors obtained applying Dφ(p) must also be the same.

Definition 2.3 (Tangent space). ∀p ∈ M, Tp M is the set of all tangent

vectors at p, which we call the tangent space at p.

11
 V
U
p

φ
ψ

φ(U ) ψ ◦ φ−1 ψ(V )

Rm Rm

Figure 2.1: Charts in the proof of Lemma 2.2

Proposition 2.4. Tp M is a vector space, and its dimension is the same as

the dimension of the manifold.

Proof. Let (U, φ) be a chart. We obtain a map

Tp M → Rm , [γ] 7→ (φ ◦ γ)0 (0).

It’s well-defined and injective by the definition of tangent vector. It is

also surjective. Indeed, given v ∈ Rm , take the straight line t 7→ φ(p) + tv
in Rm and then apply φ−1 , to obtain the following curve on M :

γ(t) = φ−1 (φ(p) + tv).

Now [γ] maps to v.

Now that we’ve proved that this is a bijection, we immediately get a
vector space structure on Tp M by “transporting” the one on Rm . Suppose
we had started with another chart (V, ψ), then we would have obtained the
same vector space structure on Tp M , because

Dφ(p) (ψ ◦ φ−1 ) : Rm → Rm .

is a linear isomorphism making this diagram commute:

[γ] ∈ Tp M

Dφ(p) (ψ◦φ−1 )
(φ ◦ γ)0 (0) ∈ Rm (ψ ◦ γ)0 (0) ∈ Rm

CHAPTER 2. TANGENT VECTORS 12

 When you choose a chart around p ∈ M , you get a basis of Tp M .

Definition 2.5 (Basis of Tp M induced by a chart). Let p ∈ M , (φ, U ) a chart

with p ∈ U whose components we denote by x1 , . . . , xm (hence xi : U → R).
By means of the isomorphism of vector spaces

Tp M → Rm : [γ] 7→ (φ ◦ γ)0 (0),

the standard basis of Rm induces a basis of Tp M , which we denote by

∂
∂x1 ,..., ∂
∂xm .
p p

Remark. Let W ⊂ Rm open, q ∈ W . There is a canonical linear isomorphism

Tp W → Rm , [γ] → γ 0 (0). To see this, take φ = Id in the proof of the previous
lemma.

2.2 The derivative of a map

Definition 2.6 (Derivative of a smooth map). If f : M → N is differentiable,
then its derivative at p is

(f∗ )p : Tp M → Tf (p) N : [γ] 7→ [f ◦ γ].

Proposition 2.7. (f∗ )p is well defined and linear.

Proof. Let φM be a chart for M near p. Let φN be a chart for N near f (p).
Consider the commutative diagram

(f∗ )p
[γ] ∈ Tp M Tf (p) N 3 [f ◦ γ]

DφM (p) (φN ◦f ◦φ−1

M )
(φM ◦ γ)0 (0) ∈ Rm Rn 3 (φN ◦ (f ◦ γ))0 (0)

As the two vertical maps are linear isomorphisms (as we saw in the proof
of Proposition 2.4) and the horizontal map is linear (being the derivative of
a smooth map between open subsets of Euclidean space), the composition
(f∗ )p is also linear.

f g
Proposition 2.8 (Chain rule). If M → N → L are smooth maps, then ∀p ∈
M we have (g ◦ f )∗ p = (g∗ )f (p) ◦ (f∗ )p : Tp M → T(g◦f )(p) L.

Remark. Let p ∈ M and (U, φ) a chart around p. One can show: φ : U → φ(U )
is a diffeomorphism of manifolds (Notice that since U is an open set of M
and φ(U ) is an open set of Rn , both carry manifold structures). We have a

CHAPTER 2. TANGENT VECTORS 13

commutative diagram of isomorphisms.

[γ] ∈ Tp M
(φ∗ )p

(φ ◦ γ)0 (0) ∈ Rm ∼
= Tφ(p) (φ(U ))

In other words, the map Tp M → Rm from the proof of Proposition 2.4, under
the identification given in the previous remark, is (φ∗ )p . In particular

∂
∈ Tp M
∂xi p

and i-th standard basis vector of Tφ(p) φ(U ) ∼

= Rm correspond under (φ∗ )p .

∂
is basis of Tp U
∂xi p

Basis of Tφ(p) φ(U )

Rn

Figure 2.2: The i-th standard basis vector and ∂

∂xi p correspond under (φ∗ )p .

2.3 The regular level set theorem

Remark. Let f : U → V be a diffeomorphism between open subsets of Rn . Then
∀q ∈ U , Dq f : Rn → Rn is an isomorphism, because its inverse is Df (q) f −1 .
Conversely,

Lemma 2.9 (Inverse function theorem in Rn ). Let U ⊂ Rn open, f : U → Rn

smooth s.t. Dq f : Rn → Rn is an isomorphism for some q ∈ U . Then
there exists a neighbourhood V ⊂ U of q such that f |V : V → f (V ) is a
diffeomorphism.

CHAPTER 2. TANGENT VECTORS 14

 Corollary 2.10 (Inverse function theorem for manifolds). Let f : M → N be
a smooth map p ∈ M , such that f∗ (p) : Tp M → Tf (p) N is an isomorphism.
Then there exists a neighbourhood W of p s.t. the map f |W : W → f (W )
is a diffeomorphism.

Idea of the proof: this is a local statement, and by means of charts, locally every
manifolds can be identified with an open subset of Rn .
Example. Consider f : R → S 1 ⊂ C, t 7→ e2πit . This is not a diffeomorphism.
But f∗ (t) is an isomorphism for all t ∈ R. So locally f restricts to a diffeomor-
phism onto it image. 
Given k ≥ n, we denote

π : Rn × Rk−n → Rn , (v, w) 7→ v.

Lemma 2.11 (Submersion theorem in Rn ). Let U be a neighbourhood of the

origin 0 in Rn × Rk−n and f : U → Rn smooth such that

(D0 f )|Rn ×{0} : Rn × {0} → Rn

is an isomorphism. Then there exists a diffeomorphism τ between neigh-

bourhoods in Rn × Rk−n such that f ◦ τ −1 = π.

This theorem states that precomposing f with a diffeomorphism, we can

arrange that it becomes the projection on the first components.

Rn × Rk−n

f
Rn π

Figure 2.3: The submersion theorem. The dotted lines denote the preimages of
points of Rn under f and π.

Proof. Consider τ := (f1 , . . . fn , xn+1 , . . . , xk ) : U → Rn × Rk−m , which we

CHAPTER 2. TANGENT VECTORS 15

 can write concisely as (f, IdRk−n ). Consider its derivative at the origin
 
(D0 f )|Rn ×{0} ∗
D0 τ = .
0 IdRk−n

This matrix is invertible, because it is an upper block matrix, (D0 f )|Rn ×{0}
is invertible, and IdRk−n is invertible. This means that τ is a diffeomorphism
near 0, by the inverse function theorem. We have π ◦ τ = f .

Definition 2.12 (Regular value). Given a smooth map f : M → N , a point

c ∈ N is a regular value iff ∀p ∈ f −1 (c), (f∗ )p : Tp M → Tc N is surjective.

Theorem 2.13. Let f : M k → N n be a smooth, and let c ∈ N be a regular

value s.t. f −1 (c) 6= ∅. Then
• f −1 (c) is a submanifold of M with dimension k − n
• ∀p ∈ f −1 (c) : Tp (f −1 (c)) = Ker(f∗ (p))

The first item states that the codimension is preserved when taking inverse
images.
Example. Consider the “height function” f : S 2 → R : (x, y, z) 7→ z. Note that
−1 and 1 are not regular values. 

Figure 2.4: The “height function” f : S 2 → R. The derivative of f vanishes at

the Northpole and Southpole.

Proof. Fix p ∈ f −1 (c). Take charts (U, φM ) near p, and (V, φN ) near f (p) =
c, chosen so that φN (c) = 0. We know that Dφ(p) (φN ◦ f ◦ φ−1 M ):R →R .
k n

is surjective (because c is a regular value). We can assume that

Dφ(p) (φN ◦ f ◦ φ−1

M )
Rn ×{0}

CHAPTER 2. TANGENT VECTORS 16

 is an isomorphism. (When we restrict a surjective linear map to a subspace
transverse to the kernel, it becomes an isomorphism. If Rn × {0} is not
transverse to the kernel of Dφ(p) (φN ◦ f ◦ φ−1
M ), we can change the chart
φN by composing it with e.g. a rotation in Rn .)
By the last lemma, there exists a diffeomorphism τ such that (φN ◦ f ◦
φ−1
M )◦τ
−1
= π, which can be rewritten as

φN ◦ f ◦ (τ ◦ φM )−1 = π.

The situation is summarized by the following diagram:

f
U V 3c
τ ◦φM φN .
π
open ⊂ R k
open ⊂ R 3 0
n

Hence τ ◦ φM is a chart of M adapted to f −1 (c).

For the second part, let p ∈ f −1 (c), and take a path γ : (−ε, ε) → f −1 (c)
with φ(0) = p. Then

(f∗ )p [γ] = [f ◦ γ] = 0 ∈ Tc N

as f ◦ γ ≡ c. This shows that Tp (f −1 (c)) ⊂ Ker(f∗ )p . Both vector spaces

have the same dimension, which gives the equality.

π
0

(τ ◦ φM )(f −1 (c) ∩ U )

Figure 2.5: The chart τ ◦φM straightens out f −1 (c) (intersected with the domain
of the chart).

2.4 Tangent vectors as derivations

Fix M = Rn . Recall that a tangent vector at p is an element of Tp Rn ∼
= Rn .

Definition 2.14 (Derivation). A derivation at a point p ∈ Rn is a linear map

D : C ∞ (Rn ) → R satisfying the Leibniz rule:

D(f g) = D(f )g(p) + f (p)D(g).

CHAPTER 2. TANGENT VECTORS 17

Example. ∀v ∈ Rn , the directional derivative

C ∞ (Rn ) −→ R
X ∂f
f 7−→ (dp f )(v) = vi
i
∂xi

is a derivation. This follows from the fact that partial derivatives obey the
Leibniz rule. 
Remark. The derivation of a constant function, Dc, is always 0. Because D is
linear, it is enough to show this for c = 1. We have D1 = (D1)1 + 1(D1) =
2(D1), so D1 = 0.

Proposition 2.15. The map

φ : Tp Rn −→ Derivations at p
 X ∂f 
v 7−→ f 7→ vi (p)
i
∂xi

is an isomorphism of vector spaces.

Proof.
• This formula really defines a derivation, as we’ve showed in the pre-
vious example.
• It is clearly linear.
• P
To show that it’s injective, we check that the kernel is 0. If for all f ,
i vi ∂xi = 0, then it is particular true for the functions xj , so vj = 0
∂f
P ∂x
for all j. In formulae: vj = i vi ∂xji = 0.

• Surjectivity. Let D be a derivation at p. For all f ∈ C ∞ (Rn ), we

have X
f (x) = f (p) + (xi − pi )gi (x),
i

where gi (x) is a smooth function with gi (p) = ∂x ∂f

i
(p). (This is a
version of Taylor’s theorem, in which the o(x − p) terms are absorbed
in g.) Then
X
Df = 0 + D(xi )gi (p) + 0
i
X ∂f
= D(xi ) (p).
i
∂xi

So v = (D(x1 ), . . . , D(xn )) maps to D.

CHAPTER 2. TANGENT VECTORS 18

Chapter 3

Vector fields

3.1 Vector fields

DefinitionS3.1 (Vector field). A vector field on a manifold M is a map
X : M → p∈M Tp M , such that
• X(p) ∈ Tp M

• X satisfies the following smoothness condition: for any chart (U, φ),
writing X(p) = i ai (p) ∂x ∂
, all the coefficients ai : M → R are
P
i
p
smooth.

Notation. We denote the set of all vector fields on M with X(M ).

Example. Let (U, φ) be a chart on M . Then ∂
∂xi is a vector field on U , for all
i = 1, . . . , dim(M ). 
Example. If U ⊂ R2 open, using the chart Id, ∂
∂x1 is just the vector field with
unit vectors pointing in the x1 direction. 

3.2 Integral curves

Definition 3.2 (Integral curve). Let X ∈ X(M ). A smooth curve γ : (a, b) →
M is an integral curve of X iff

γ̇(t) = X|γ(t)

for all t ∈ (a, b).

Remark. Here γ̇(t) is defined as (γ∗ )t (1), where (γ∗ )t : Tt (a, b) → Tγ(t) M , and
Tt (a, b) ∼
= R, so using 1 as an input is valid. Another way to look at it, in terms
of tangent vectors as equivalence classes of curves: γ̇(t) equals [s 7→ γ(s + t)].

19
 Proposition 3.3. Let X be a vector field, p ∈ M . Then there exists a
neighborhood U of p, an ε > 0 and a unique smooth map

F : U × (−ε, ε) → M

s.t. for all q ∈ U , the curve γq defined by γq (t) = F (q, t) is an integral curve
of X with γq (0) = q.

Proof. Fix a chart (φ, V ) near p. In these coordinates, we have X =

i for some smooth functions fi . We need to show that there exists
∂
fi (x) ∂x
a neighboorhood W ⊂ φ(V ) of p, ∃ε > 0 and ∃! y : W × (−ε, ε) → φ(V )
such that (
∂
∂t y(x, t) = f (y(x, t))
y(x, 0) = x
for all x. This holds by the fundamental theorem of ODE’s. (It says that
for each initial value x, there is a unique solution defined on a small interval
(−ε, ε), and the solution varies smoothly with x.)

Remark. The map F in the above proposition is called flow.

Remark. In particular, for all p ∈ M , there exists an ε > 0 and a unique integral
curve γp of X defined on (−ε, ε), starting at point p.
Example. Sometimes, we cannot continue the curve. For example, look at R2 \
{(0, 0)} and X = ∂x ∂
1
. Then the integral curve starting at (−2, 0) is defined
only up to time t = 2. 
Assume, for the sake of simplicity, that the flow of X is defined on M × R.
Then F is a 1-parameter group of diffeomorphisms, as a consequence of the
uniqueness in the above proposition.

Definition 3.4 (1-parameter group of diffeomorphisms). A 1-parameter group

of diffeomorphisms on M is a smooth map F : M ×R → M such that, using
the notation Ft (p) = F (p, t) (think of it as fixing t and varying the point),
one has

• Fs ◦ Ft = Fs+t ∀s, t
• F0 = Id
(It then follows that Ft is a diffeomorphism for all t.)

Remark. There is a bijection between

• vector fields on M whose flow is defined on the biggest possible domain
M × R, and
• 1-parameter groups of diffeomorphisms on M .

CHAPTER 3. VECTOR FIELDS 20

The bijection reads:

X flow F as above
X given by X(q) = d
dt 0 Ft (q) F : M × R → R.

Example. On R2 take X = ∂x1 . Then

∂
the flow is Ft (x1 , x2 ) = (x1 + t, x2 ). 
Example. On R2 let X = ∂
x ∂y ∂
− y ∂x .Then the images of integral curves are
circles. The flow is given by Ft (x, y) = Rt (x, y), where Rt denotes the rotation
by the angle t. 

3.3 The Lie bracket of vector fields

We denote C ∞ (M ) = {smooth functions from M to R}.
Remark. C ∞ (M ) is an algebra: it is a vectorspace, but we can also multiply
two functions.

Definition 3.5 (Derivation of C ∞ (M )). A derivation of C ∞ (M ) is a linear

map D : C ∞ (M ) → C ∞ (M ) such that

D(f g) = D(f )g + f D(g).

Remark. If D1 , D2 are derivations of C ∞ (M ), then the commutator D1 ◦ D2 −

D2 ◦ D1 is also a derivation. But D1 ◦ D2 on its own is not a derivation!

Proposition 3.6. There is a linear map

Φ : X(M ) −→ Derivations of C ∞ (M )
X 7−→ (f 7→ f∗ (X) ∈ C ∞ (M )).

Here f∗ (X) is the function on M given by (f∗ (X))p = (f∗ )p (X|p ) ∈

Tf (p) R ∼
= R. We denote f∗ (X) =: X(f ).

Proof. We’ll show that ∀X ∈ X(M ), Φ(X) is a derivation. Let f, g ∈

C ∞ (M ), p ∈ M . Let γ be a curve in M such that γ(0) = p, [γ] = Xp .
Then

(f g)∗p (Xp ) = [(f g) ◦ γ] ∈ T(f g)(p) R

∼
= ((f g) ◦ γ)0 (0) ∈ R
= [(f ◦ γ) · (g ◦ γ)]0 (0)
= (f ◦ γ)0 (0) · g(γ(0)) + f (γ(0)) · (g ◦ γ)0 (0)
= (f∗ )p (X|p ) · g(p) + f (p) · (g∗ )p (X|p ).

In the second-last equality we applied the product rule of calculus.

Remark. Φ is an isomorphism. This can be showed using the material in §2.4.

CHAPTER 3. VECTOR FIELDS 21

 Definition 3.7 (Lie bracket of vector fields). The Lie bracket of two vector
fields X, Y ∈ X(M ) is

[X, Y ] := X ◦ Y − Y ◦ X,

using the identification between X(M ) and the derivations of C ∞ (M ).

Definition 3.8 (Lie algebra). A Lie algebra is a vector space g with a bilinear
map [·, ·] : g × g → g such that
• [X, Y ] = −[Y, X] Skew symmetric
• [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0 Jacobi Identity

Example. (X(M ), [·, ·]) is a Lie algebra. 

Example. The square matrices (M (n, R) with [A, B] := AB − BA is a Lie alge-
bra. The Jacobi identity holds as a consequence of the associativity of matrix
multiplication. 
Remark. Let (U, φ) be a chart on M . Then we get vector fields ∂
∂xi on U ⊂ M .

• The vector field ∂x ∂

i
, seen as a derivation, maps xj ∈ C ∞ (U ) to δij

• If X = i ai ∂x∂
, Y = i bi ∂x ∂
, with ai , bi ∈ C ∞ (U ), then
P P
i i

X X !
∂bj ∂aj ∂
[X, Y ] = ai − bi .
j i
∂xi ∂xi ∂xj

hThis can ibe seen by applying [X, Y ] to the functions xi . In particular

∂xi , ∂xj = 0, because the ai and bi are constants here.
∂ ∂

Definition 3.9 (F -related vector fields). Let F : M → N be a smooth map

and X ∈ X(M ), Y ∈ X(N ). We say that X and Y are F -related iff
(F∗ )p (Xp ) = YF (p) for all points p ∈ M .

Remark. Equivalently: X and Y are F -related iff ∀g ∈ C ∞ (N ) we have X(F ∗ g) =

F ∗ (Y (g)). Here F ∗ is the pullback of functions, i.e. F ∗ (g) = g ◦ F .

Proposition 3.10. Suppose Xi is F -related to Yi for i = 1, 2. Then [X1 , X2 ]

is F -related to [Y1 , Y2 ].

Proof. Use X1 (X2 (F ∗ (g)) = X1 (F ∗ (Y2 (g))) = F ∗ (Y1 (Y2 (g))).

CHAPTER 3. VECTOR FIELDS 22

 p

Figure 3.1: Two p-related vector fields, where p : R2 → R is the first projection.

3.4 Interpretation of the Lie bracket

Definition 3.11 (Pushforward of a vector field by a diffeomorphism). Given a
diffeomorphism φ : M → N and X ∈ X(M ), we denote by φ∗ X the unique
vector field on N such that X is φ-related to φ∗ X.

Explicitly, we have (φ∗ X)φ(p) = (φ∗ )p (Xp ) for all p ∈ M .

Lemma 3.12. Let φ : M → N be a diffeomorphism, let X ∈ X(M ) and

Y ∈ X(N ). Then Y = φ∗ (X) iff

FtY ◦ φ = φ ◦ FtX for all t s.t. FtX is defined.

Here F X denotes the flow of X, and similarly for Y .

Definition 3.13 (Lie derivative). Let X, Y ∈ X(M ). The Lie derivative of

Y in the direction of X is the vector field
d X
LX Y = (F−t )∗ Y.
dt t=0

Remark. (LX Y )p = dt d X
[(F−t )∗ Y ]p = d
dt
X
[(F−t )∗ YFt (p) ]. These are all tan-
t=0 t=0
gent vectors in Tp M .
Remark. One can show that LX Y = [X, Y ].

CHAPTER 3. VECTOR FIELDS 23

 Proposition 3.14. Let X, Y ∈ X(M ). The following are equivalent:
a) LX Y = 0

b) (FtX )∗ Y = Y for all t

c) The flows of X and Y commute: FtX ◦ FsY = FsY ◦ FtX for all t, s.

Proof. • b) ⇒ a). LX Y = d X
dt (F−t )∗ Y = d
dt Y = 0.
• a) ⇒ b). Fix a point p. Consider t 7→ (F−t
X
)∗ YFt (p) . This is a curve in
Tp M , which at t = 0 equals Yp . We will show that this is a constant
curve by taking the derivative. For all t0 :
d X d X
(F−t )∗ (YFt (p) ) = (F−t ) (YFtX+s (p) )
0 −s ∗
dt t0 ds s=0 0
d 
X X
= (F−t )∗ (F −s )∗ YF s (F t (p))
 ds s=0
0 0

X
= (F−t0 )∗ (LX Y )Ft0 (p) = 0,

where in the first equality we set t = t0 + s.

• b) ⇔ c). For all t, apply the last lemma to FtX : M → M .

Corollary 3.15. [X, Y ] = 0 iff the flows commute.

h i
Example. On R2 , we know ∂x ∂
, ∂ = 0. Indeed, the flows given by (x, t) 7→
1 ∂x2

x + (t, 0) and (x, t) 7→ x + (0, t) commute. 

Proposition 3.16. Let V1 , . . . , Vk linearly independent vector fields on M

such that [Vi , Vj ] = 0. Then for every p ∈ M there is a chart (U, (s1 , . . . , sn ))
centered at p such that Vi = ∂s ∂
i
for i = 1, . . . , k.

Proof. Since this is a local statement, let’s assume that we’re working in
Rn , and p = 0. Assume that spani {Vi |0 } ⊕ ({0} × Rn−k ) = Rn . Denote
by θi the flow of Vi and let Ω be a small neighborhood of the origin in
{0} × Rn−k . Define

Φ : (−ε, ε)k × Ω −→ Rn
(s1 , . . . , sk , sk+1 , . . . , sn ) 7−→ (θ1 )s1 ◦ · · · ◦ (θk )sk (0, . . . , 0, sk+1 , . . . , sn )T .

Then ∂
∂xi is Φ-related to Vi for i ≤ k. Indeed for for i ≤ k you can compute
that  ∂ 
(D(s1 ,...,sn ) Φ) = Vi |Φ(s1 ,...,sn )
∂si

CHAPTER 3. VECTOR FIELDS 24

 by using the curve ε 7→ (s1 , . . . , si + ε, . . . , sn ) and using the fact that the
flows θi commute by the last corollary. At the point p = 0 we clearly
have (D0 Φ)( ∂s∂
i
∂
) = ∂x i
|0 for i > k, hence d0 Φ is an isomorphism. By the
inverse function theorem, there exists a neighbourhood V of 0 such that
Φ|V : V → Φ(V ) =: U is a diffeomorphism. The desired chart is the inverse
of this diffeomorphism.

(θ2 )s2 ◦ (θ1 )s1 (0, 0, s3 )

(θ2 )s2 ◦ (θ1 )s1 (0, 0)
(θ1 )s1 (0, 0, s3 )
(θ1 )s1 (0, 0) (0, 0, s3 )

(0, 0)

(0, 0, 0)

Figure 3.2: Two examples illustrating the proof. On the left, n = k = 2. On

the right, n = 3 and k = 2.

CHAPTER 3. VECTOR FIELDS 25

Chapter 4

Bundles

4.1 Fiber bundles

Definition 4.1 (Fiber bundle). Let F be a manifold. A fiber bundle with
fiber F is a smooth surjective map

π:E→B

between manifolds s.t. for all x ∈ B, there exists an open neighbourhood U

and a diffeomorphism ψ : π −1 (U ) → U × F making this diagram commute:

ψ
π −1 (U ) U ×F
π
π1

E = S 1 × (−ε, ε)

π −1 (U )
π

B = S1
U

Figure 4.1: Example of a product fiber bundle

Remark. The map π : E → B is a submersion, i.e. (π∗ )e is surjective in all

points e of E. Indeed: locally, the map π looks like the projection onto the
first factor π, as ψ is a diffeomorphism. So we can work with π1 , which is a
submersion.

26
Remark. For all x ∈ B, the fiber π −1 (x) is diffeomorphic to F . Indeed, π −1 (x) ≈
π1−1 (x) = {x} × F ≈ F . So we have a family of manifolds (the fibers) which all
are diffeomorphic to F , and this family is parametrized by B.
E is called the total space, B the base space and ψ a local trivilization. A
section is a smooth f : B → E such that π ◦ f = IdB .

E = S 1 × (−ε, ε)
Image of a section

B = S1

Figure 4.2: Definition of a section

π
Example. Given F, B the product bundle is B × F −→
1
B. 
Example. Let F be a manifold, φ : F → F diffeomorphism. Define the manifold
E := ([0, 1] × F )/ ∼ and B = S 1 , where ∼ is given by

(0, p) ∼ (1, φ(p)).

Then π : E → B is a fiber bundle with typical fiber F . For instance, take

F = (−ε, ε), φ = − Id. Then E is the Möbius strip.

Figure 4.3: Möbius strip


Example. The first projection π1 : R \{0} → R, (x, y) 7→ x is not a fiber bundle.
2

For instance, not all fibers are diffeomorphic. 

CHAPTER 4. BUNDLES 27
 4.2 Vector bundles
Definition 4.2 (Vector bundle). Fix n ∈ N≥0 . A vector bundle of rank n is
a smooth surjective map π : E → B between manifolds E, B such that
• Ep = π −1 (p) is a n-dimensional vector space for all p ∈ B
• For all p ∈ B, there exist a neighborhood U of p and a diffeomorphism
ψ : E|U := π −1 (U ) → U × Rn such that the following diagram
commutes
ψ
π −1 (U ) U × Rn
π
π1

U
and ψ|Eq : Eq → {q} × Rn is a linear isomorphism, for all q ∈ U .

image of the zero section B E

Figure 4.4: Definition of a vector bundle

Remark. Equivalently, a vector bundle of rank n is a fiber bundle s.t. each fiber
as an n-dimensional vector space and trivializations are linear in each fiber.
Remark. • There exists a canonical section B → E, q 7→ (zero vector in Eq ).
It is called zero section.

• Denote by Γ(E) the set of all sections of E, it is a vector space. Note that
we can also multiply a function B → R with a section B → E. Hence
Γ(E) is a module over the algebra C ∞ (B).
π
Example. Given a manifold, take B×Rn −→
1
B. Notice Γ(B×Rn ) = C ∞ (B, Rn ).


CHAPTER 4. BUNDLES 28
 Proposition 4.3. Let B be a manifold of dimension n. Then
G
T B := Tp B
p∈B

is naturally a vector bundle of rank n, called the tangent bundle of B

Proof. Define π : T B → B, v ∈ Tp B 7→ p. Notice that π −1 (p) = Tp B is a

vector space.
We show that T B is a manifold. Consider a chart φ : U → φ(U ) of B,
denote its components by (x1 , . . . , xn ). It induces a bijection

ψ : (T B)|U −→ U × Rn
∂
ai 7−→ (p, ai ).
∂xi p

Now take a cover of B by charts (Uα , φα )α∈A . Notice that ψβ ◦ ψα−1 is a

smooth map from (Uα ∩ Uβ ) × Rn to itself. Consider the topology on T B
with basis ψα−1 (σ) for σ ⊂ Uα × Rn open and α ∈ A. Then {((T B)|Uα ), ψa }
is aa smooth atlas. Hence T B is a smooth manifold. For all α ∈ A, the
chart ψα : (T B)|Uα → Uα × Rn is a diffeomorphism and linear on every
fiber.
a Strictly speaking, ψ n is not an open subset of
α is not a chart, because Uα × R
Rn × Rn . But we can identify Uα with φα (Uα ), which is an open subset of Rn .

Remark. Γ(T M ) = X(M )

Remark. Let π1 : E1 → B and π2 : E2 → B be a vector bundle with the same
base. Their direct sum (Whitney sum) is a vector bundle E1 ⊕ E2 → B with
fiber over p ∈ B given by (E1 ⊕ E2 )p = (E1 )p ⊕ (E2 )p . Trivialisations are given
by taking the direct sum of the trivializations of E1 and of E2 .

Definition 4.4 (Vector subbundle). Let π : E → B be a vector bundle.

A vector subbundle is a subset D ⊂ E s.t. π|D : D → B, with the
induced smooth structure and vector space structure on the fibers, is a
vector bundle.

CHAPTER 4. BUNDLES 29
 Definition 4.5 (Vector bundle morphism). Let πi : Ei → Bi be vector bun-
dles for i = 1, 2. A smooth map F : E1 → E2 is a vector bundle morphism
if there exists a smooth map f : B1 → B2 such that the following diagram
commutes
F
E1 E2
π1 π2
f
B1 B2
and such that for all p ∈ B1 , the map F |(E1 )p : E1,p → (E2 )f (p) is linear.
We call F an isomorphism if F is invertible and F, F −1 are both vector
bundle morphisms. (Or equivalently, if F is a diffeomorphism.)

Definition 4.6 (Frame). Given a vector bundle E → B, a frame is a collec-

tion of sections that form a basis at every point.

Remark. E admits a frame iff E is isomorphic to vector bundle B × Rn . In this

case we call E a trivial vector bundle.
Example. Let S n ⊂ Rn+1 be the unit sphere. The tangent bundle T S n and
its orthogonal (T S n )⊥ are vector bundles over S n . (T S n )⊥ is a trivial bundle,
indeed, we can choose the normal vector at each point as a frame. However,
T S n is not a trivial bundle in general. (T S n only admits a frame when n = 1, 3.
The fact that T S 2 does not admit a frame is a consequence of the Hairy ball
theorem.)
The Whitney sum T S n ⊕ (T S n )⊥ is isomorphic to Rn+1 × S n , which is a
trivial vector bundle. So this is an example of a sum of a trivial and a non-trivial
vector bundle being trivial. 

S2

Figure 4.5: The Whitney sum of T S 2 ⊕ (T S 2 )⊥ is a trivial vector bundle.

CHAPTER 4. BUNDLES 30
Chapter 5

Differential forms and

integration

5.1 Forms on vector spaces

Let V be a finite dimensional real vector space.

Definition 5.1. For every k ≥ 1, we define

k
^
V ∗ := {ω : V × · · · × V → R | multilinear and skew symmetric}
| {z }
k times

Skew symmetric means: ω(. . . , v, . . . , w, . . .) = −ω(. . . , w, . . . , v, . . .). We

V0 ∗
define V =R
V1
Example. V∗ =V∗ 
V2
Example. V = bilinear maps V × V → R that are skew symmetric.
∗


Definition 5.2 (Wedge product). ∀`, k ≥ 0, the wedge product is

k
^ `
^ `+k
^
∧: V∗× V ∗ −→ V∗
1 X
(ω ∧ τ )(v1 , . . . , vk+` ) 7−→ (−1)σ ω(vσ(1) , . . . , vσ(k) ) · τ (vσ(k+1) , . . . , vσ(k+`) ).
k!`!
σ∈Sk+`

Vk V`
Remark. ω ∧ τ = (−1)k` τ ∧ ω, where ω ∈ V ∗ and τ ∈ V ∗.

Lemma 5.3. Suppose θ1 , · · · , θk ∈ V ∗ and v1 , . . . , vk ∈ V . We have

(θ1 ∧ · · · ∧ θk )(v1 , . . . , vk ) = det(θi (vj )).

31
 Lemma 5.4. Let m = dim V . Suppose θ1 , . . . , θm is a basis of V ∗ . Then
∀k ≥ 1,
{θi1 ∧ · · · ∧ θik | 1 ≤ i1 < . . . < ik ≤ m}
Vk ∗
is a basis of V .

Example. Suppose dim V = 3 and (θ1 , θ2 , θ3 ) is a basis of V ∗ , then

V0 ∗
• 1 is a basis of V =R
V1 ∗
• {θ1 , θ2 , θ2 } is basis of V =V∗
V2 ∗
• {θ1 ∧ θ2 , θ2 ∧ θ3 , θ1 ∧ θ3 } is a basis of V
V3 ∗
• θ1 ∧ θ2 ∧ θ3 is a basis of V .


Vk
Remark. If k > m, the dimension of the space, then V ∗
is the zero vector
space.

Definition 5.5 (Pullback). Let f : V → W be a linear map. Then the

dual map is f ∗ : W ∗ → V ∗ , given by (f ∗ θ)(v) = θ(f (v)). More generally,
∀k ≥ 1, the pullback by f is
k
^ k
^
f∗ : W ∗ −→ V∗
ω 7−→ f ∗ ω

where
(f ∗ ω)(v1 , . . . , vk ) = ω(f (v1 ), . . . , f (vk )).

5.2 Differential forms on manifolds

Let M be a manifold, f ∈ C ∞ (M ). For all p ∈ M , define

(df )p := (f∗ )p : Tp M → Tf (p) R = R

We notice that (df )p ∈ (Tp M )∗ =: Tp∗ M

Let (U, φ = (x1 , . . . , xn )) be a chart.

Lemma 5.6. The set {dx1 |p , dx2 |p , . . . , dxk |p } is the basis of Tp∗ M , which
is dual to ∂x
 ∂
i p

Proof. We have that ∂

∂xi p = (φ−1 )∗,φ(p) ei , where ei is the i-th standard

CHAPTER 5. DIFFERENTIAL FORMS AND INTEGRATION 32

 basis vector. Hence, for every j,
 
∂
= dxj |p (φ−1 )∗,φ(p) ei = (xj ◦ φ−1 )∗,φ(p) ei .


dxj |p i
∂x p

Since xj ◦ φ−1 is the jth projection, this expression is is the jth component
of ei , which is δij .

Example. Consider R2 with standard coordinates x1 , x2 : dxi |p : Tp R2 → R2 is

given by ∂x∂
1
|p 7→ 1 and ∂x
∂
2
|p 7→ 0. 
Vk ∗
Let k ≥ 0. For every p ∈ M , consider the vector space Tp M .

Definition 5.7 (Differential form). A k-form on M is a map

k
G ^
α:M → Tp∗ M
p∈M

such that
• α(p) ∈ Λk Tp∗ M for each p
• For any chart (U, φ), writing
X
α(p) = ai1 ...ik dxi1 ∧ · · · ∧ dxik |p ,
1≤i1 <...<ik ≤m

the coefficients ai1 ...ik : U → R are smooth.

Vk
Remark. In other words: a k-form is a section of the vector bundle T ∗M .
We denote Ωk (M ) = {k-forms on M }. Notice that Ω0 (M ) = C ∞ (M ).
Remark. Given α ∈ Ωk (M ) and vector fields X1 , . . . , Xk ∈ X(M ), we define
α(X1 , . . . , Xk ) ∈ C ∞ (M ) by

(α(X1 , . . . , Xk ))(p) := α(p)(X1 (p), . . . , Xk (p)).

Notice that for all f ∈ C ∞ (M ), α(f X1 , . . . Xk ) = f α(X1 , . . . , Xk ).

Definition 5.8. Let F : M → N be a smooth map. The pullback of k-forms

is F ∗ : Ωk (N ) → Ωk (M ) :

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

where p ∈ M , vi ∈ Tp M .

We now study the pullback of differential forms on Rn .

CHAPTER 5. DIFFERENTIAL FORMS AND INTEGRATION 33

 Proposition 5.9. Let U ⊂ Rm open, V ⊂ Rn open, G : U → V smooth.
Denote with xj the standard coordinates on Rm and by yi those on Rn .
Then
Pm ∂Gi
1) G∗ (dyi ) = j=1 dxj ∈ Ω1 (U )
∂xj
2) If m = n, then

G∗ (f (y)dy1 ∧ . . . ∧ dym ) = (f ◦ G) det(Jac G)dx1 ∧ . . . ∧ dxm .

Above Jac G denotes the Jacobian of G, i.e. the matrix representing the deriva-
tive DG of G.
   
• G∗ (dyi ) ∂ ∂ ∂Gk ∂ ∂Gi
P
Proof. ∂xj = dyi G∗ ∂x j
= dyi ( k ∂xj ∂yk ) = ∂xj

• G∗ (f dy1 ∧ . . . ∧ dym ) = (G∗ f )(G∗ dy1 ∧ . . . ∧ G∗ dym ). Evaluating on

∂x1 , . . . , ∂xm , we get
∂ ∂

  
∗ ∗ ∂
(G f ) det G (dyi ) = (f ◦ G) det(Jac G).
∂xj
| {z }
∂Gi
= by above
∂xj

5.3 Orientation and volume forms

Definition 5.10 (Oriented atlas). An oriented atlas is a smooth atlas (Uα , φα )
α ) > 0 for all α, β.
s.t. det D(φβ ◦ φ−1

Definition 5.11 (Orientation). An orientation is a choice of maximal ori-

ented atlas.

Remark. Not all manifolds are orientable, e.g. the Möbius band and RP2 are
not.
Remark. Let V be vector space. On {ordered bases of V }, there is an equiva-
lence relation (v1 , . . . , vm ) ∼ (w1 , . . . , wm ) if change of basis has det > 0. An
orientation of V is by definition a choice of one of the two equivalence classes.
An ordered basis of an oriented vector space is positive if it belongs to the
equivalence class that gives the orientation.
Now let M be a manifold. An orientation on M induces and orientation on
each tangent space Tp M .

Definition 5.12 (Volume form). A volume form on M m is an m-form Ω such

that Ω(p) 6= 0 for all p ∈ M .

CHAPTER 5. DIFFERENTIAL FORMS AND INTEGRATION 34

Remark. If Ω is a volume form, then an other volume form looks like f Ω, where
f : M → R \ {0}.

Proposition 5.13. M is orientable iff there exists a volume form Ω

Proof. We denote by Vol := dx1 ∧ . . . ∧ dxm the standard volume form

on Rm .
⇐: choose an atlas consisting of charts such that (φ−1 α ) Ω = fα Vol,
∗

where fα is a positive function. Then det D(φβ ◦ φ−1α ) > 0 because of item
2 of the last proposition.
⇒: given an oriented atlas {(UjP
, φj )}, choose a partition of unity {eα }
subordinate to it, and define Ω = α eα · (φα )∗ Vol. Then Ω is a volume
form, because on Uα ∩ Uβ we have

α ) Vol = (a positive function) · (φα ) Vol.

(φβ )∗ Vol = (φα )∗ (φβ ◦ φ−1 ∗ −1 ∗
| {z }
det(D(φβ ◦φ−1
α ))·Vol

5.4 Integration on manifolds

Let M m be a oriented manifold. We want to define ω, where ω ∈ Ωm (M )
R
M
has compact support.

Step 1. Assume the support supp(ω) := {p ∈ M : ω(p) 6= 0} is contained in

one chart (U, φ) of the oriented atlas.. Write (φ−1 )∗ ω = f dx1 ∧ . . . ∧ dxm for
some f ∈ C ∞ (φ(U )).

Definition 5.14 (Integral of top form supported in a chart).

Z Z Z
ω := (φ−1 )∗ ω := f (x) dx1 dx2 · · · dxm ,
M φ(U ) φ(U )

where on the right hand side is the multiple Riemann integral of the func-
tion f over φ(U ).

Recall the transformation rule in Rm . Let U, V ⊂ Rm and θ : V → U a

diffeomorphism. Let f ∈ C ∞ (U ) be integrable. Then
Z Z
f (x) dx1 · · · dxm = (f ◦ θ)(y)| det(Jac θ)| dy1 · · · dym .
U V

ω is independent of the choice of chart in the oriented

R
Lemma 5.15. M
atlas.

CHAPTER 5. DIFFERENTIAL FORMS AND INTEGRATION 35

 Proof. Let (V, ψ) be another chart as above, and denote θ := φ ◦ ψ −1 .

V
U
supp(ω)

φ ψ
Rm
φ(U ) θ ψ(V )
Rm

We want to find g ∈ C ∞ (ψ(V )) such that (ψ −1 )∗ ω = g(y)dy1 dy2 · · · dym .

To do so we compute

(ψ −1 )∗ ω = (φ−1 ◦ θ)∗ ω = θ∗ ((φ−1 )∗ ω) = (f ◦ θ) det(Jac θ) dy1 ∧ . . . ∧ dym

| {z }
so this is g

using a Proposition from §5.2 in the last equality. By the transformation

rule
Z Z
f (x) dx1 dx2 · · · dxm = (f ◦ θ)(y)| det(Jac θ)| dy1 · · · dym ,
φ(U ) ψ(V ) | {z }
=g

finishing the proof. (The function on the right hand side equals g since the
absolute values can be removed, due to the fact that ψ and φ lie in the
oriented atlas of M .)

Step 2. For any ω ∈ Ωm (M ) with compact support, let {(Uα , φα )} be an

oriented atlas such that supp(ω) ∩ Uβ 6= ∅ for finitely many β. PLet eβ be
aPpartition of unity subordinate to this cover. Notice that ω = ( eβ )ω =
(eβ ω).

Definition 5.16 (Integral of top form).

Z XZ
ω= eβ ω.
M β

Notice that each summand eβ ω was defined in Step 1, since supp(eβ ω) ⊂ Uβ .

R

Remark. One can show: this definition is independent of the choice of oriented
atlas and partition of unity.

CHAPTER 5. DIFFERENTIAL FORMS AND INTEGRATION 36

Chapter 6

Exterior derivative and Stokes

theorem

6.1 The exterior derivative in Rm

Let U ⊂ Rm open. Denote by x1 , . . . , xm the standard coordinates on Rm .
Remark. If f ∈ C ∞ (U ), then df = f∗ lies in Ω1 (U ), because for all p ∈ U we
have (f∗ )p ∈ Tp∗ M . Expressing df in terms of dxi , we can write We have
X ∂f
df = dxi ,
i
∂xi
 
since (f∗ )p ∂
∂xi = (Dp f )(ei ) = ∂f
∂xi (p).
p
Example. On R , we have d(xy) = ydx + xdy.
2


Definition 6.1 (Exterior derivative on Rm ). The exterior derivative (or de

Rham differential)
d : Ωk (U ) −→ Ωk+1 (U )
is defined as follows: for k = 0, as above. For k > 0
 
X X
d ai1 ...ik dxi1 ∧ · · · ∧ dxik  := (dai1 ...ik )∧dxi1 ∧· · ·∧dxik
1≤i1 <...<ik ≤m

Notice that above ai1 ...ik ∈ C ∞ (U ).

Example. On R3 consider (x1 )3 dx2 ∧ dx3 , then

dω = 3(x1 )2 dx1 ∧ dx2 ∧ dx3 .


Example.
d((x1 )2 dx1 ∧ dx2 ) = 0


37
 Proposition 6.2. The exterior derivative satisfies the following:
i) d is R-linear
ii) d(α ∧ β) = dα ∧ β + (−1)k α ∧ dβ, where α ∈ Ωk (U ).

iii) d2 = 0.
iv) If F : U → V smooth, then d(F ∗ ω) = F ∗ (dω).

Proof. iii) We prove it only for g ∈ C ∞ (U ).

 X ∂g 
d(dg) = d dxj
j
∂xj
X  ∂g 
= d ∧ dxj
j
∂xj
X ∂2g
= dxi ∧ dxj .
j,i
∂xi ∂xj

2
Now dxi ∧ dxj is anti-symmetric in i and j, while ∂x∂i ∂x
g
j
is symmetric
in i and j (partial derivatives commute!), so d = 0.
2

iv) We prove it only for g ∈ C ∞ (U ). Let p ∈ U , v ∈ Tp U = Rm . Then

(F ∗ (dg))v = dg((F∗ )p v) = ((g ◦ F )∗ )p v = (F ∗ g)∗ v = d(F ∗ g)v,

using the chain rule in the second equality.

6.2 The exterior derivative on manifolds

Let M be a manifold.

Definition 6.3 (Exterior derivative on manifolds). Let ω ∈ Ωk (M ), then dω ∈

Ωk+1 (M ) is defined as follows: for all charts (U, φ),
  
(dω)|U = φ∗ d (φ−1 )∗ ω .

Notice that (φ−1 )∗ ω is a k-form on an open subset of Rm .

Remark. The above is well defined because if (V, ψ) is another chart, then the
map (ψ ◦ φ−1 )∗ commutes with d by part iv) of the previous proposition.

CHAPTER 6. EXTERIOR DERIVATIVE AND STOKES THEOREM 38

6.3 Manifolds with boundary
Definition 6.4. Hm = {(x1 , . . . , xm ) ∈ Rm | x1 ≤ 0}

Definition 6.5 (Differentiable map on a open subset of Hm ). Let U ⊂ Hm be

open. Then f : U → Rk is differentiable iff there exists an open Ũ ⊂ Rm
such that U = Ũ ∩ Hm , and there exists f˜ : Ũ → Rm differentiable such
that f˜|U = f . In that case, for all p ∈ U , Dp f = Dp f˜.

Definition 6.6 (Manifold with boundary). A manifold with boundary of di-

mension m consists of topological space M that is second countable and
Hausdorff, together with a maximal smooth atlas.
A smooth atlas consists of an open cover Uα and homeomorphisms φα
from Uα to open subsets of Hm , such that φβ ◦ φ−1α is differentiable for
all α, β.

Remark. In particular, all manifolds are manifolds with boundary.

Definition 6.7 (Boundary). The boundary of M is

∂M = {p ∈ M : ∃ chart (U, φ) : φ(p) ∈ ∂Hm },

where ∂Hm = {0} × Rm−1 .

Remark. If one charts satisfies this property, all charts satisfy this property.
Example. Consider. M = {v ∈ Rm : kvk ≤ 1}. Then the boundary of M is
∂M = {v ∈ Rm : kvk = 1}. 
One can show the following

Proposition 6.8. ∂M is a manifold, of one dimension less than M .

Proposition 6.9. An orientation on M induces an orientation on ∂M , as

follows:

∀p ∈ ∂M : (v1 , . . . , vm−1 ) is a positive basis of Tp ∂M

⇔ (e, v1 , . . . , vm−1 ) is a positive basis of Tp M ,

where e ∈ Tp M is “outward pointing”.

Example. Note that R2 has a standard ordered basis, hence the unit disk D ⊂ R2
too. The induced orientation on the circle ∂D is the “anticlockwise” one. To see
this: (e, v1 ) as in the figure is a positive basis of R2 . Hence the orientation that
D induces on ∂D is the one for which v1 is a positive basis.

CHAPTER 6. EXTERIOR DERIVATIVE AND STOKES THEOREM 39

 v1
e

Figure 6.1: Orientation of the boundary of a disk

6.4 Stokes’ theorem

Rb Rb
We generalize the fundamental theorem of calculus: a
df = a
f 0 (x)dx = f |ba .

Theorem 6.10 (Stokes). Let M m be an oriented manifold with boundary

and ω ∈ Ωm−1 (M ) with compact support. Then
Z Z
dω = i∗ ω,
M ∂M

where i : ∂M → M is the inclusion and ∂M has the induced orientation of

M.

Remark. Note that since ω has compact support, so does dω.

The idea of the proof is to use a partition of unity to reduce this to the case
where supp(ω) is contained in a chart, i.e. to reduce to M = Hm . There one
can apply suitably the fundamental theorem of calculus.
Example. Let D = {v ∈ R2 : kvk ≤ 1}. We have
Z Z Z
i∗ (xdy − ydx) = d(xdy − ydx) = 2 dx ∧ dy = 2π.
S1 D D

Corollary 6.11. Let M be a manifold (without boundary) and ω ∈ Ωm−1 (M )

with compact support. Then
Z
dω = 0.
M

CHAPTER 6. EXTERIOR DERIVATIVE AND STOKES THEOREM 40

Corollary 6.12. If M is compact, orientable manifold with boundary, then
there is no smooth f : M → ∂M such that f |∂M = Id∂M .

Proof. Since M is orientable, ∂M is too. Therefore, there exists volume

form on the boundary, call it ω. Suppose f exists. Then d(f ∗ ω) = f ∗ dω =
0, because dω = 0 by degree reasons. So
Z Z Z Z
0= d(f ∗ ω) = i∗ f ∗ ω = (f ◦ i)∗ ω = ω 6= 0,
M ∂M ∂M ∂M

because ω is a volume form.

CHAPTER 6. EXTERIOR DERIVATIVE AND STOKES THEOREM 41

Chapter 7

De Rham Cohomology

7.1 Basic definitions

Let M be a manifold of dimension m.

Definition 7.1 (Closed forms). ω ∈ Ωk (M ) is called closed iff dω = 0.

Definition 7.2 (Exact forms). ω ∈ Ωk (M ) is called exact iff dα = ω for some

α ∈ Ωk−1 (M ).

We have
d0 d1 dm−1
0 Ω0 (M ) Ω1 (M ) ··· Ωm (M ) 0

with dk ◦ dk−1 = 0, so in particular the image of dk−1 is included in the kernel

of dk . This is an example of a cochain complex. Cochain because d increases
the degree.

Ωk−1 Ωk Ωk+1

exact

closed
dk−1 dk

Figure 7.1: Visualization of the de Rham cochain complex. Exact k-forms lay
per definition in the image of dk−1 , and closed k-forms in the kernel of dk . As
d2 = 0, the exact k-forms form a subspace of the closed k-forms.

42
 Definition 7.3 (De Rham cohomology). The k-th de Rham cohomology
group of M is
Ker dk closed k-forms
H k (M ) := HdR
k
(M ) := = .
Im dk−1 exact k-forms

If ω ∈ Ωk (M ) is closed, we denote by [ω] ∈ H k (M ) its class.

Ωk−1 Ωk Ωk+1

dk−1 dk
π

Hk

Figure 7.2: Definition of the De Rham cohomology.

Remark. Each H k (M ) is a vector space. Only when 0 ≤ k ≤ m, H k (M ) can

be different from zero. If M is compact, one can show that H k (M ) is finite
dimensional for all k. This is not trivial, because the dimension of Ker dk is
normally ∞-dimensional.

Proposition 7.4. If M is connected, then H 0 (M ) ∼

= R.

Proof. For all f ∈ Ω0 (M ), f is closed if df = 0, so f is constant (M is

connected and f is smooth). f is exact iff it is zero, since Ω−1 (M ) =
{0}.

Remark. If not connected, we get Rk , with k being the number of connected

components.

Theorem 7.5. If M m is compact, connected and orientable, then H m (M ) ∼

=
R.

Proof. Forms of ‘top degree’ are always closed. The idea is that we con-
struct a map from the m-forms to R and then show that the kernel are

CHAPTER 7. DE RHAM COHOMOLOGY 43

 exactly the exact m-forms. Consider

I : Ωm (M ) −→ R
Z
ω 7−→ ω.
M

ThenR I is surjective: since M is oriented, there exists a volume form, Ω,

and M Ω 6= 0. We now argue that ker(I) = {exact m-forms}.
Consider an exact m-form ω = dα. Then by Stokes’ theorem
Z Z
dα = α = 0.
M ∂M

Idea for the other inclusion: Assume M ω = 0, then we need to show that
R

ω = dα for some α. Reduce to forms with support (necessarily compact)

contained in a chart. Then this reduces to the following: if f Ris a function
x
on R with compact support and R f (x)dx = 0, then F (x) = −∞ f (t)dt is
R

a primitive of f with compact support (because whenRx is really small, this

integral is zero, and when x is large enough, F (x) = R f (x)dx = 0).
This shows that
Ωm (M )
= H m (M ).
Ker(I)
But this quotient is isomorphic to the image of I, which is R.

7.2 Homotopic maps and cohomology

Lemma 7.6. Let f : N → M be a smooth map. Then
• If ω ∈ Ω(M ) is closed, then f ∗ ω is also closed.
• If ω ∈ Ω(M ) is exact, then f ∗ ω is also exact.

Proof. We only prove the first part. d(f ∗ ω) = f ∗ dω = f ∗ 0 = 0.

Given f : N → M , consider the pullback of differential forms f ∗ : Ωk (M ) →

Ω (N ). This restricts to a map between closed forms, and also between exact
k

forms, so it induces a map on the level of cohomologies:

H(f ) : H k (M ) −→ H k (N )
[ω] 7−→ [f ∗ ω].

If f is a diffeomorphism, then it’s clear that H(f ) is an isomorphism. But there

are more maps that induces an isomorphism in cohomology!

Definition 7.7 (Smooth homotopy). Let f, g : N → M smooth maps. A

smooth homotopy between f and g is a smooth map h : N × [0, 1] → M
such that h|N ×{0} = f and hN ×{1} = g .

In other words, h is a smooth family of maps that interpolate between f and g.

CHAPTER 7. DE RHAM COHOMOLOGY 44

 One can prove:

Proposition 7.8. If f, g : N → M are smoothly homotopic, then f ∗ and g ∗

are cochain homotopic, i.e. there exists a linear map

K : Ω• (M ) → Ω•−1 (M ),

such that d ◦ K + K ◦ d = f ∗ − g ∗ .

d d
Ωk−1 (M ) Ωk (M ) Ωk+1 (M )
f ∗ −g ∗
K K
k−1 d k d
Ω (N ) Ω (N ) Ωk+1 (N )

Theorem 7.9. If f, g are smoothly homotopic, then H(f ) = H(g).

Proof. Let K = Ω• (M ) → Ω•−1 (M ) a cochain homotopy between f ∗ and

g ∗ . Let ω ∈ Ωk (M ) be closed. Then

f ∗ ω − g ∗ ω = d(Kω) + K(|{z}
dω )
=0

is exact. Hence [f ∗ ω] = [g ∗ ω].

Corollary 7.10. Let M and N be homotopy equivalent, i.e. ∃α : N → M

and β : M → N such that α ◦ β ' 1N and β ◦ α ' 1M , where ' means
smoothly homotopic. Then H(α) : H(M ) → H(N ) is an isomorphism with
inverse H(β).

Proof. H(α)◦H(β) = H(β ◦α) = H(1N ), because of the previous theorem,

and H(1N ) = 1H(N ) . You can do the same for the reverse order.

Remark (Poincaré Lemma). Let U ⊂ Rn be an open subset which is star shaped1 .

Then U is homotopy equivalent to a point {p}, via the inclusion α : {p} → U
and the constant map β : U → {p}. Hence,
(
k ∼ k ∼ R if k = 0
H (U ) = H ({p}) =
0 if k > 0
Notice that Rn is star-shaped, and every point of a manifold has a star-shaped
neighborhood.
Example. If E → M is a vector bundle, then it is homotopic equivalent to M ,
as we can shrink the fibers (which are vector spaces) to the zero section. So the
cohomology of M is the same as that of E: H(E) ∼ = H(M ). 
1 ∃p such that for all u ∈ U the line segment between u and p lies in U .

CHAPTER 7. DE RHAM COHOMOLOGY 45

Remark. Given manifolds M, N , if H(M ) 6= H(N ), then the manifold are not
homotopic equivalent.

7.3 The Mayer–Vietoris theorem

Theorem 7.11. Suppose a manifold admits a cover by 2 open sets U, V , i.e
M = U ∪ V . Then

··· H k−1 (U ∩ V )

Ak Bk
H k (M ) H k (U ) ⊕ H k (V ) H k (U ∩ V )
δk

H k+1 (M ) ···

is a long exact sequence, i.e. the image of any arrow is the kernel of the
next one. Here,

Ak [ω] := [ω|U ] ⊕ [ω|V ]

Bk (α ⊕ β) := β|U ∩V − α|U ∩V .

Remark. The proof uses the smooth maps

iU ∩V,V
U ∩ V −−−−→ U t V → M.
iU ∩V,U

Example. Let’s consider the sphere S m with m ≥ 2. Write

Sm = U ∪ V

where U = S m \ {North pole}, V = S m \ {South pole}. Notice that

• U ' {p},
• V ' {p},

• U ∩ V ' S m−1 .
Applying the theorem and doing some diagram-chasing, we get
(
k m ∼ R if k = 0, m
H (S ) =
0 otherwise.

CHAPTER 7. DE RHAM COHOMOLOGY 46

Chapter 8

Foliations

8.1 Immersed submanifolds

Definition 8.1 (Immersion). An immersion is a smooth map F : H → M
such that (F∗ )p is injective for all p ∈ H.

Definition 8.2 (Immersed submanifold). An immersed submanifold of M is

a subset H ⊂ M with a topology (not necessarily the subspace topology)
and a smooth manifold structure such that the inclusion i : H → M is an
immersion.

Remark. In that case, for all p ∈ H, the map (i∗ )p is an isomorphism from Tp H
to (i∗ )p (Tp H) ⊂ Tp M .
Remark. All submanifolds are immersed submanifolds. Immersed submanifolds
are not necessarily submanifolds.
Remark. If H is an immersed submanifold, then for all p ∈ H, there exists an
open neighbourhood U in H (“open” w.r.t. the topology of H) such that U is a
submanifold of M .
Example. For all λ ∈ R consider F : R → S 1 × S 1 : t 7→ (e2πit , e2πiλt ). If λ ∈ Q,
then F (R) is a submanifold of S 1 × S 1 . If λ 6∈ Q, then F (R) is just an immersed
submanifold with the smooth structure given by the bijection R ∼ = F (R). The
topology on F (R) differs from the subspace topology induced by S 1 × S 1 .

S1 × S1
F i

R H


47
8.2 Distributions and involutivity
Definition 8.3 (Distribution). A distribution D on M is a subbundle of T M .

For all p ∈ M , we get a subspace Dp ⊂ Tp M of the same dimension varying

smoothly with p.

Definition 8.4 (Involutive distribution). A distribution is involutive iff for all

X, Y ∈ Γ(D), we have [X, Y ] ∈ Γ(D)

Remark. D is involutive iff for all p ∈ M , there exists a U ⊂ M and a frame

X1 , . . . , Xk ∈ Γ(D|U ) such that [Xi , Xj ] ∈ Γ(D|U ).

Definition 8.5 (Integral manifold). An integral manifold is a (non-empty)

immersed submanifold H of M such that Tp M = Dp for all p ∈ H.

Example. Let X be a nowhere vanishing vector field on M . Then D given

by Dp = span Xp is a rank 1 distribution, thus involutive. The image of any
integral curve of X is an integral manifold of D. 
n o
Example. On R , D = span ∂x , ∂y is an involutive rank-2 distribution. The
3 ∂ ∂

planes {z = c} ⊂ R3 are integral manifolds. 

n o
Example. On R3 , D = span ∂x ∂ ∂
, ∂y ∂
− x ∂z is not involutive. Indeed,
 
∂ ∂ ∂ ∂
, −x =− 6∈ D.
∂x ∂y ∂z ∂z


Proposition 8.6. Let D be a distribution. Suppose that every point of M

is contained in an integrable manifold of D. Then D is involutive.

Proof. Let X, Y ∈ Γ(D), p ∈ M . Let H be an integral manifold through p.

Then X|H , Y |H are tangent to H, so [X, Y ]|H = [X|H , Y |H ] is tangent to
H. (To see this equality, use that there is a neighborhood of p in H which
is a submanifold of M .) In particular, [X, Y ]p ∈ Tp H = Dp .

8.3 The Frobenius theorem

Let D be a rank k distribution on M m .

Definition 8.7 (Flat chart for D).A chart (U, φ) is flat for D if φ(U ) =
I1 × · · · × In ⊂ Rn and D = span ∂x∂ ∂
1 , . . . , ∂xk on U . Here, Ii are open
intervals.

CHAPTER 8. FOLIATIONS 48
 Definition 8.8 (Completely integrable distribution). D is completely inte-
grable if for all p ∈ M , there exists a flat chart containing p.

Theorem 8.9 (Frobenius). D is completely integrable iff D is involutive.

Notice that the condition on the left is a local one, while the condition on the
right is an infinitesimal one (thus easier to check).

Proof. .
⇒ At each p ∈ M , take a flat chart. Then D = span ∂ ∂
and

 ∂ ∂x1 , . . . , ∂xk
∂

∂xi , ∂xj = 0 ∈ Γ(D).

⇐ Sketch: Show that near every point p ∈ M , D has a frame of vector

fields s.t. [Xi , Xj ] = 0. Then apply Prop. 3.16.

One can prove:

Proposition 8.10. Let D be an involutive rank-k distribution, H and inte-

gral manifold. For any flat chart (U, (x1 , . . . , xn )) for D, H ∩ U is the union
of countably many open subset of slices {xi = const} (i > k) of U .

Example. On the torus S 1 × S 1 , let D = span{ ∂x∂

1
∂
+ λ ∂x 2
}, where λ ∈ R \ Q.
Every line with slope λ in S 1 × S 1 is an integral manifold, and the countable
condition holds. 

8.4 Foliations
Let M n be a smooth manifold.

Definition 8.11 (Foliation). A rank k foliation is a collection {Lα }α∈A of

connected immersed submanifolds of M such that
• M = α∈A Lα (disjoint union)
F

• For all p ∈ M , there exists a chart such that U ∩ Lα is a countable

union of slices {xk+1 = const, . . . , xn = const}, or empty. The Lα are
called leaves.

CHAPTER 8. FOLIATIONS 49
 Rn−k

φ
M
Rk

Figure 8.1: Definition of a foliation

Example. On R20 , circles with radius r > 0 form a foliation. 

Example. On R , the following is not a foliation. (It’s impossible to straighten
2

all leaves nearby the x-axis.)

Figure 8.2: Not a foliation.

CHAPTER 8. FOLIATIONS 50
 The following theorem (sometimes called “Global Frobenius Theorem”) gives
a bijection between global objects on M and infinitesimal objects.

Theorem 8.12. Let M be a manifold. There is a bijection

{Foliations} ↔ {Involutive distributions}

{Lα } 7→ D s.t. Dp = Tp (leaf through p)

The inverse map reads

D 7→ {maximal connected integral manifolds of D}.

Proof. We show that both maps are well-defined. It is easy to see that they
are inverses of each other.

→: D, as defined above, is a distribution. Through every point of M there

passes an integral manifold, hence D is involutive by Proposition 8.6.
←: By the Frobenius theorem, for every point of M there is an integral
manifold through it.

One can show that through every p ∈ M there passes a maximal

connected integral manifold, because the union of all integral manifolds
through p is again an integral manifold. The decomposition of M by the
above maximal connected integral manifolds is a foliation, by Proposition
8.10.

CHAPTER 8. FOLIATIONS 51
Chapter 9

Lie groups and Lie algebras

9.1 Lie groups, Lie subgroups

Definition 9.1 (Lie group). A Lie group is a group which is a manifold, such
that the multiplication m : G×G → G and the inversion i : G → G : g 7→ g −1
are differentiable.

Definition 9.2 (Lie group morphism). A Lie group morphism is a group

morphism which is smooth.

Example. (Rn , +) is a Lie group. Check, for instance, that (x, y) 7→ x + y is

smooth. 
Example. We check that

GL(n, R) = {A ∈ Mat(n, R) : A invertible}

is a Lie group. Here Mat(n, R) denotes the real n×n matrices. The set GL(n, R)
is open in the vector space Mat(n, R) ∼
2
= Rn , so it’s a manifold. It is also a group,
and the multiplication is smooth, because
X
(AB)ij = Ajk Bki
k

is a polynomial in the entries of A and B. The inversion is also smooth: for

n = 2, for instance,
 −1  
a b 1 d −b
= ,
c d ad − bc −c a
and similarly for arbitrary n. 
Remark. GL(n, R) has 2 connected components: det > 0 and det < 0.
Example. We check that

SL(n, R) = {A ∈ Mat(n, R) : det(A) = 1}

is a Lie group.

52
 The map det : Mat(n, R) → R is smooth, because det(A) is a polynomial
in the entries of A. We check that 1 ∈ R is a regular value of det, i.e. for all
A ∈ SL(n, R), the derivative dA det : TA Mat(n, R) = Mat(n, R) → T1 R = R is
surjective:
 
d d
(dA det) (t + 1)A = det((t + 1)A)
dt 0 dt 0
d
= (t + 1)n det A
dt 0
= n det(A) = n 6= 0.

Hence SL(n, R) = det−1 ({1}) is a submanifold of Mat(n, R), by the regular value
theorem, and thus a submanifold of its open set GL(n, R). The multiplication
and inversion of SL(n, R) are smooth, being the restrictions of those of GL(n, R).

Example. The following are all examples of Lie groups:

• O(n), defined as matrices with A−1 = AT , which has two components.

• SO(n), defined as matrices with A−1 = AT and det(A) = 1. It is the
connected component of the identity of O(n).
T
• U(n), defined as matrices A ∈ Mat(n, C) such that AA = 1.
T
• SU(n), defined as matrices A ∈ Mat(n, C) such that AA = 1, det A = 1.

Definition 9.3 (Left translation, left invariant vector field). Let G be a Lie
group. For all g ∈ G, the diffeomorphism

Lg : G → G : h 7→ gh

is called left translation. A vector field X ∈ X(G) is left invariant iff

∀g ∈ G : (Lg )∗ X = X.

Notice that Lg is the restriction to to {g} × G of the multiplication map m :

G × G → G.
Remark. There is a linear isomorphism

Te G −→ X(G)L := {left-invariant vector fields}

v 7−→ ←
−
v where (←
−
v ) = ((L ) ) v.
g g ∗ e

The idea is that if you have a left invariant vector field, then it is determined by
its value at any point, for example e, so that Xg = [(Lg )∗ X]g = ((Lg )∗ )e (Xe ) ∈
Tg G.

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 53

 It follows that:
• The tangent bundle T G is a trivial vector bundle. Indeed,

G × Te G −→ T G
(g, v) 7−→ ((Lg )∗ )e v.

is a vector bundle isomorphism.

• G admits a volume form, hence it’s orientable.

Example. Of all the spheres, only S 0 , S 1 , S 3 are Lie groups. (S 2 is not a Lie
group; for instance, the Hairy ball theorem implies that the tangent bundle is
not trivial.) 

Proposition 9.4. A connected Lie group is generated (as a group) by any

neighbourhood W of identity.

Proof. Let H be the subgroup generated by W , i.e. finite products of ele-

ments of W and W −1 .

H is open. Indeed W1 = W ∪ W −1 is open. W2 := W1 · W1 = g∈W1 gW1

S
is open as the left translation is smooth and any union of open subsets
is open. Similarly, for all k we get that Wk := W1 · Wk−1 is open.
Thus H = ∪k≥1 Wk is open.
H is non-empty , as e ∈ H.

H c is open. Indeed, [
G−H = gH
g6∈H

is also open, as the union of open sets.

Since G is connected, it follows that G − H is empty, i.e. H = G.

Definition 9.5 (Lie subgroup). Let G be a Lie group. A Lie subgroup H is

an (abstract) subgroup of G which is an immersed manifold, such that H
becomes a Lie group with the induced group and manifold structures.

Remark. H might not be a submanifold

Example. G = S 1 × S 1 is a Lie group, since S 1 = U (1), and the product of two
Lie groups is again a Lie group.
Let λ ∈ R, then
H = {(e2πit , e2πiλt ) : t ∈ R}.
is a subgroup of G. It is also a immersed submanifold (as we saw earlier). The
multiplication and inversion are smooth. (To check this: the smooth structure
on H is obtained from the one on R, and he induced multiplication is the
addition on R.) Conclusion: H is a Lie subgroup. 

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 54

 Proposition 9.6. Let G be a Lie group, let H be a subgroup and also a
submanifold. Then H is a Lie subgroup.

Proof. There is an obvious manifold structure on H. We have to check that

with the manifold structure induced by G, the multiplication m : H × H →
H and the inversion i : H → H are smooth. We just do the latter: the
inversion of G is smooth, so restricting it to a submanifold, we again get a
smooth map.
One can show:

Theorem 9.7. Let G be a Lie group, H a subgroup. If H ⊂ G is closed

in the topological sense, then H is a submanifold, and therefore a Lie
subgroup.

9.2 From Lie groups to Lie algebras

Recall:

Definition 9.8 (Lie algebra). A Lie algebra is a vector space g equipped

with a bilinear, skew symmetric map [·, ·] : g × g → g, such that

[x, [y, z]] + [z, [x, y]] + [y, [z, x]] = 0.

We now define morphisms and the “subobjects” of Lie algebras:

Definition 9.9 (Lie algebra morphism). A map φ : (g, [·, ·]g ) → (h, [·, ·]h ) is a
Lie algebra morphism if it preserves all the structure:
• the map φ is linear

• φ([x, y]g ) = [φ(x), φ(y)]h

Definition 9.10 (Lie subalgebra). A Lie subalgebra of g is a vector subspace

h, such that
[x, y] ∈ h ∀x, y ∈ h.

Proposition 9.11 (A). Let G be a Lie group, then g := Te G is a Lie algebra,

with the bracket
[v, w] := [←
−
v ,←
−]| .
w e

Proof. Recall the linear isomorphism Te G → X(G)L , v 7→ ← −v . The Lie

bracket of two left invariant vector fields is again a left invariant vector

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 55

 field:

(Lg )∗ [←
−
v ,←
−] = [(L ) ←
w − ←− ←− ← −
g ∗ v , (Lg )∗ w ] = [ v , w ].

So X(G)L is a Lie algebra. Now use the above linear isomorphism to trans-
port this Lie algebra structure to Te G.
Example. Consider GL(n, R). Then Te (GL(n, R)) is all n × n matrices, i.e.
Mat(n, R). So Mat(n, R) has an induced Lie algebra structure. Its bracket is

[A, B] = AB − BA.

(This is not at all obvious, it requires a bit of computation) 

Proposition 9.12 (B). Let Φ : G → H be a Lie group morphism (i.e. a

smooth group homomorphism). Then

de Φ := (Φ∗ )e : Te G → Te H

is a Lie algebra morphism.

←−−−−−
Proof. For all v ∈ Te G, the vector field ←
−
v is Φ-related to (de Φ)(v). Indeed,

(dg Φ)(←
−
v )g = (dg Φ)(Lg )∗ (v) = de (Φ ◦ Lg )(v) = de (LΦ(g) ◦ Φ)(v),

where the last equation is true because Φ is a homomorphism. Now, pulling

derivatives apart again, we see that the above equals
←−−−−
(LΦ(g) )∗ ((de Φ)v) = (de Φ(v))Φ(g)

Take v1 , v2 ∈ Te G. Applying the previous statement we obtain that [← v−1 , ←

v−2 ]
←−−−−− ←−−−−−
is Φ-related to [(de Φ)v1 , (de Φ)v2 ] (naturality of the Lie bracket). In partic-
ular, at g = e,
h←−−−−− ←−−−−−i
(de Φ)([←
v−1 , ←
v−2 ]e ) = (de Φ)v1 , (de Φ)v2 .
e

Proposition 9.13 (C). Let H be a Lie subgroup of G. Then Te H is a Lie

subalgebra of Te G.

Proof. We know that the inclusion is Lie group morphism. Therefore, its
derivative Te H → Te G is a Lie algebra morphism by the last proposition.

Example. SL(n, R) is a Lie subgroup of GL(n, R), hence the Lie algebra of
SL(n, R) is a Lie subalgebra of GL(n, R). It turns out that the Lie algebra
of SL(n, R) are exactly the traceless matrices. 

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 56

9.3 The exponential map
Let G be a Lie group, g = Te G.

Lemma 9.14. For all v ∈ Te G, there is a unique morphism of Lie groups

γv : (R, +) → G with the property that γv0 (0) = v.

Notice: there are lots of curves γ that have γ 0 (0) = v, but by the lemma there
exists only one which is also a group homomorphism, i.e. γv (s)·γv (t) = γv (s+t).

Proof. Existence: The left invariant vector field ← −

v is complete, i.e. integral
curves are defined for all times. Let γ : R → G be the integral curve starting
at e. Clearly γv0 (0) = v. γ is a group homomorphism, because for every
fixed s, we have that the curves t 7→ Lγ(s) γ(t), and t 7→ γ(s + t) agree. (At
t = 0, both of these go through γ(s), and they are also integral curvesa
of ←−
v .)
Uniqueness: let γ : R → G be a Lie group morphism with γ 0 (0) = v.
Since ∂t∂
is a left-invariant vector field on R, by the proof of Proposition 9.12
←−−−−−−− ←−−−
it is γ-related to (d γ)( ∂ | ) = γ 0 (0) = ←
0 ∂t 0
−
v . Hence for every t ∈ R: 0

∂
=←
−

(dt0 γ) v |γ(t0 ) .
∂t t0

The left hand side is just γ 0 (t0 ). Hence γ is the integral curve of ←
−
v starting
at e.
a For t 7→ γ(s + t), it’s trivial. For t 7→ L
γ(s) γ(t), we have that this is an integral
curve of (Lγ(s) )∗ ←
−
v , which is the same as ← −
v.

Definition 9.15 (Exponential map). Let G be a Lie group with Lie algebra g.
The exponential map is

exp : g → G, v 7→ γv (1).

Proposition 9.16. a) exp(tv) = γv (t) for all t ∈ R and v ∈ g.

b) There exists a neighborhood U of 0 ∈ g such that

exp |U : U → exp(U )

is a diffeomorphism onto an open subset of G. (This allows to study

G close to the identity element e by studying the Lie algebra. This
also defines a chart near e.)
c) If H ⊂ G is a Lie subgroup. Then expH : Te H → H is the restriction
of expG : Te G → G.

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 57

 Proof. a) The curves s 7→ γtv (s) and s 7→ γv (st) agree, as they are both
Lie group morphisma (R, +) → G with the same velocity vt at s = 0.
Take s = 1.
b) Idea: check that d0 exp : T0 g = g → Te G = g is the identity on g.
Then apply the inverse function theorem.

a The first by definition, and the second is a composition of two group morphisms.

Example. For GL(n, R), we have

exp : Te GL(n, R) = Mat(n, R) −→ GL(n, R)

∞
X An
A 7−→ eA := .
n=0
n!

Proof. For all A ∈ Mat(n, R), the curve

(R, +) → GL(n, R), t 7→ etA

is a group morphism, since etA · esA = e(t+s)A . Furthermore,

d d
etA = (I + tA + O(t2 )) = A.
dt 0 dt 0
So this curve is γA .

9.4 From Lie algebras to Lie groups

Let (g, [·, ·]) be a (finite dimensional) Lie algebra.

Definition 9.17 (Lie group integrating a given Lie algebra). A Lie group G
integrates g iff Te G is isomorphic to g (as Lie algebras).

We will show:

Theorem 9.18 (A). Up to isomorphism, there exists a unique simply con-

nected Lie group GSC integrating g.

Remark. All other connected Lie groups integrating g are quotients of GSC by
discrete normal subgroups.
Example. Let g = (R, [·, ·] = 0). Then GSC = (R, +). Note that U (1) = S 1 =
R/Z is also a Lie group that integrates g. 
Example. Let g = {A ∈ Mat(n, R) : A + AT = 0}. The Lie group SO(n)
integrates g. 

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 58

 Proposition 9.19 (C). Let G be a Lie group and h a Lie subalgebra of
g = Te G. There exists a unique connected Lie subgroup H whose Lie
algebra is h.

Example. Let G = U (1) × U (1), h = span(1, λ) ⊂ R2 = Te G, where λ ∈ R.

Then H = {(eit , eiλt ) : t ∈ R}. 

Proof. Sketch: Denote by D the distribution on G given by

Dg := (Lg )∗ h.

It is involutive because it is spanned by left-invariant vector fields and h is

a Lie subalgebra.
By the global Frobenius theorem, there is a (unique) foliation of G
whose leaves are tangent to D. Notice that the foliation is invariant under
left-translation: (Lg )(Sg0 ) = Sgg0 , where Sg0 denotes the leaf of D through
g0 .
One can show that Se , the leaf of the foliation through e, is a Lie
subgroup with Lie algebra h. (Clearly Se is an immersed manifold with
Te Se = h. To show that it is a subgroup, use the left-invariance of the
foliation. The multiplication and inversion are smooth.) Further, one can
show uniqueness.

Corollary 9.20. Given a Lie group G, there is a bijection

{Connected Lie subgroups of G} ↔ {Lie subalgebras of Te G}.

Remark. The nice bijection in this corollary justifies the (slightly involved) def-
inition of Lie subgroup.
Once can show:

Proposition 9.21 (B). Let G be a simply connected Lie group, H a Lie

group, and Ψ : Te G → Te H a Lie algebra morphism. Then there is a
unique Lie group morphism Φ : G → H such that de Φ = Ψ.

This proposition allows us to easily prove the uniqueness in Theorem (A):

Proof. Let G1 , G2 be simply connected Lie groups and Ψ : Te G1 → Te G2

a Lie algebra isomorphism. Then there exist
• a Lie group morphism Φ : G1 → G2 s.t. de Φ = Ψ,

• a Lie group morphism χ : G2 → G1 s.t. de χ = Ψ−1 .

Since χ ◦ Ψ and IdG1 are both Lie group morphisms with derivative IdTe G1 ,
they must agree. Similarly, Ψ ◦ χ = IdG2 . So Φ is an isomorphism.

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 59

 We sketch the proof of the existence in Theorem (A), i.e. given a Lie alge-
bra g, there exists a simply connected Lie group integrating it.

Proof. Idea of proof: g is isomorphic to a Lie subalgebra g0 of Mat(n, R)

for some n (Ado’s theorem).
Let G0 be the unique connected Lie subgroup of of GL(n, R) with Lie
algebra g0 . Take GSC to be the universal cover of G0 . (The universal cover
is again a Lie group, and simply connected.)

CHAPTER 9. LIE GROUPS AND LIE ALGEBRAS 60