Sei sulla pagina 1di 98

Differentiable Manifolds and Lie

Groups
Course notes for MA338, IISc Bangalore

Thomas Richard

January 24, 2014


These are the notes for the MA338 course I’ve given in 2013 on “Differentiable man-
ifolds and Lie groups” at IISc Bangalore. These are far from being as polished and
complete as I would have liked, but could still be of some use. Please feel free to report
typos, misspellings, confusing statement at thomas.richard2@laposte.net.
About the exercises, the *’s are a rough indication of difficulty.
The prerequisites are some linear algebra, some multivariate calculus (the inverse
function theorem will be of constant use), and some point set topology (metric spaces,
topological spaces, compactness, connectedness).
I don’t claim any novelty in the treatment I have made of this classical subject. I’ve
been working mainly with the textbooks [Lee02] and [Laf96] (sorry the second one is
in French !). Other useful references (more advanced, and biased towards Riemannian
geometry) : [GHL04] and [Jos11]. About Lie groups, [Sti08] is really good, although
we won’t follow the same approach. An impressing collection of exercises can be found
in [nMM13]. The first volume of [Spi99] is also interesting, it covers all the topics we’ll
discuss here and takes great care to show how the various approaches (coordinate based
and geometric) link together. Although [Boo86] may feel slightly outdated, it is still a
worthwhile reference.
I am really thankful to the students who followed this course, there questions and
reactions during the lectures had a big influence on these notes.

2
Contents

1 Smooth manifolds, tangent spaces and smooth maps 5


1.1 Quick review of differential calculus . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 The differential of a map . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 The inverse function theorem and its consequences . . . . . . . . 6
1.2 Smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 The Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.2 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.3 Product manifolds, tori . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.4 Quotients, tori (again) and projective spaces . . . . . . . . . . . . 13
1.4 Smooth maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4.1 Definition and first examples . . . . . . . . . . . . . . . . . . . . . 19
1.4.2 Building smooth maps . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 The tangent space at a point . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.1 Some words about the tangent space to a submanifold in Rn . . . 22
1.5.2 Tangent vectors at a point to a manifold . . . . . . . . . . . . . . 23
1.5.3 Tangent vectors in coordinates . . . . . . . . . . . . . . . . . . . . 24
1.6 Differentiating maps between manifolds . . . . . . . . . . . . . . . . . . . 25
1.6.1 Definitions, first properties . . . . . . . . . . . . . . . . . . . . . . 25
1.6.2 Tangent vectors as derivations . . . . . . . . . . . . . . . . . . . . 27
1.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2 Tangent bundle, Vector fields and Lie brackets 34


2.1 Vector fields in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.1 General theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.2 Linear vector fields and matrix exponential . . . . . . . . . . . . . 36
2.2 Cut-off functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 From the tangent spaces to the tangent bundle . . . . . . . . . . . . . . . 38
2.3.1 T M as smooth manifold . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.2 T M as a vector bundle . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4 Vector fields on manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Vector fields and global derivations . . . . . . . . . . . . . . . . . . . . . 44
2.6 Lie bracket of vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3
3 Lie groups and homogenous spaces 52
3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.1 Abelian examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.2 The 3-sphere as a Lie group . . . . . . . . . . . . . . . . . . . . . 52
3.2.3 Matrix groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.4 Transformation groups . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3 Lie morphisms, Lie subgroups . . . . . . . . . . . . . . . . . . . . . . . . 53
3.4 Left invariant vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5 The Lie algebra of a Lie group . . . . . . . . . . . . . . . . . . . . . . . . 56
3.6 Homegeneous spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.6.1 Observations on topological groups . . . . . . . . . . . . . . . . . 61
3.6.2 Quotient of Lie groups . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6.3 Homogeneous spaces . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4 Tensors, differential forms and Stokes Theorem 68


4.1 The cotangent bundle and 1-forms . . . . . . . . . . . . . . . . . . . . . . 68
4.1.1 The cotangent bundle . . . . . . . . . . . . . . . . . . . . . . . . 68
4.1.2 Differential 1-forms . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 A little tensor algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.1 Tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.2 Alternating forms and wedge products . . . . . . . . . . . . . . . 70
4.3 Differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3.1 The differential of a p-form. . . . . . . . . . . . . . . . . . . . . . 74
4.3.2 Lie derivative and Cartan’s formula . . . . . . . . . . . . . . . . . 76
4.3.3 Closed forms, exact forms and Poicaré lemma . . . . . . . . . . . 78
4.4 Integration of differential forms . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.1 Orientability and volume forms . . . . . . . . . . . . . . . . . . . 80
4.4.2 The integral of an n-form . . . . . . . . . . . . . . . . . . . . . . 83
4.5 Stokes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5.1 Regular domains . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5.2 Stokes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.5.3 Some vector calculus revisited . . . . . . . . . . . . . . . . . . . . 90
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4
1 Smooth manifolds, tangent
spaces and smooth maps
1.1 Quick review of differential calculus
1.1.1 The differential of a map
Definition 1.1.1. Let (E, k kE ) and (F, k kF ) be two normed vector spaces, U ⊂ E an
open set, and f : E → F . f is said to be differentiable at a ∈ U if there exists a
(continuous) linear application L : E → F such that :

f (a + h) = f (a) + Lh + o(h)

where ko(h)k
khkE
F
goes to 0 as h goes to zero.
The linear map L is the differential of f at a, and will be denoted by Da f .

Remark 1.1.2. If E and F are finite dimensional, all linear maps are continuous and all
norms are equivalent, so the choice of the norms doesn’t matter. Although differential
calculus in infinite dimension has a number of nice application, we won’t use it, except
maybe in a few optional exercises.
From now on, E and F are assumed to be finite dimensional.
Remark 1.1.3. The definition implies that if f is differentiable at a then f is continuous
at a.
To make the link with the derivative of function of a real variable, consider a function
f : Rn → R which is differentiable at a ∈ Rn , and curve c : R → Rn , then the derivative
of f ◦ c at 0 can be computed in the following way :

f (c(h)) − f (c(0)) 1 
= Da f (c(h) − c(0)) + o(c(h) − c(0))
h h  
c(h) − c(0) 1 
= Da f + o c(h) − c(0)
h h
t→0
−−→ Da f (c0 (0)) since kc(h) − c(0)k ≤ Ch.

So (f ◦ c)0 (0) = Da f (c0 (0)). In other words, when one evaluates the differential of v on
some vector v, one gets the variation of f along a curve that has v as a tangent vector.
Example 1.1.4. Any (continuous) linear map L : E → F is differentiable, and for any
a ∈ E, Da f = L.

5
Example 1.1.5. Let B : E1 × E2 → F be a bilinear map, then B is differentiable and
Da1 ,a2 B(h1 , h2 ) = B(a1 , h2 ) + B(h1 , a2 ). Similar formulas hold for multilinear maps.
Example 1.1.6. det : Mn (R) → R is differentiable, and :

DA det(H) = trace(ÃH)

where à is the cofactor matrix of A.


Example 1.1.7. inv : GLn (R) → GLn (R) is differentiable and :

DA inv(H) = −A−1 HA−1 .

Assume that E = Rn and F = Rm and write f = (f 1 , · · · , f m ) the components of


f (each of the fi is a function of (x1 , · · · , xn ) ∈ Rn ). Then the matrix of Da f in the
canonical basis of Rn and Rm is :
 ∂f 1 ∂f 1

1 (a) · · · ∂xn
(a)
 ∂x .. ..  .
 . . 
∂f m ∂f m
∂x1
(a) · · · ∂xn (a)

The matrix above is called the Jacobian matrix of f at a. In order to keep notation
j
light, we will write ∂i f j for ∂f
∂xi
.
Assume that f : U → F is differentiable at every x ∈ U . Since L(E, F ) is again
a normed vector space, it makes sense to ask wether the application Df : U 3 x 7→
Dx f ∈ L(E, F ) is continuous. If this holds f is said to be C 1 . Moreover, one can ask
if Df itself is differentiable. If Df is differentiable at every x ∈ U , it defines a map
D2 f : U → L(E, L(E, F )), if this map is continuous, f is said to be C 2 . Similarly, one
defines the notion of C k function for any integer k, and a function is C ∞ (or smooth) if
it is C k for every k.
One has the following useful characterization of C k maps :
Proposition 1.1.8. f : Rm ⊃ U → Rn is C k is and only if it has C k partial derivatives.
The differential of functions built by composition is easily computed using the chain
rule :
Theorem 1.1.9. Let f : Rn ⊃ U → Rm differentiable at a ∈ U , g : Rm ⊃ U 0 → Rp
differentiable at f (a). Then x 7→ (g ◦ f )(x) is differentiable at a and :

Da (g ◦ f ) = Df (a) g ◦ Da f.

1.1.2 The inverse function theorem and its consequences


Definition 1.1.10. Let U ⊂ Rn , V ⊂ Rm be open sets, f : U → V is a C k -diffeomorphism
(k ≥ 1) if it is a C k bijection and it inverse f −1 : V → U is C k on V .
Proposition 1.1.11. If f : U → V is a C k -diffeomorphism (k ≥ 1) , then for any
a ∈ U Da f : Rn → Rm is an invertible linear map. In particular n = m.

6
Proof. Let g = f −1 , then g ◦ f = idRn . Since g and f are differentiable, we can apply
the chain rule for any a ∈ U :

idRn = Da (g ◦ f ) = Df (a) g ◦ Da f,

thus Da f is invertible.
Remark 1.1.12. This also shows that we could have required in the definition of a C k
diffeomorphism f −1 to be merely differentiable instead of C k . Let us show this for k =
1. Set g = f −1 , since g is differentiable, it is continuous, hence g is a homeomorphism.
Its differential is given by : Db g = (Dg(b) f )−1 . Since f is C 1 , a 7→ Da f is C 0 . Since
g is continuous, b 7→ Dg(b) f is continuous. And using that taking the inverse of an
invertible linear map is a smooth operation, b 7→ Db g = (Dg(b) f )−1 is continuous,
hence g is C 1 .
The inverse function theorem is a partial converse to this statement :
Theorem 1.1.13. Let f : U → Rn be a C k function and a ∈ U such that Da f is
invertible. Then there exists an open neighborhood of a U 0 ⊂ U such that f : U 0 → f (U 0 )
is a C k -diffeomorphism.
It has the drawback of being a local statement. We say that f : U → Rn is a local
diffeomorphism if any x ∈ U has an open neighborhood Ux such that f : Ux → f (Ux )
is a diffeomorphism. The inverse function theorem implies the following :
Theorem 1.1.14. f : U → Rn is a local diffeomorphism if and only if for any a ∈ U ,
Da f is invertible.
The following corollary is useful in the application :
Corollary 1.1.15. f : U → V is a bijection and for any a ∈ U , Da f is invertible, then
f is a diffeomorphism.
It is legitimate to ask wether something still holds when one replaces the hypothesis,
Da f invertible by Da f injective or surjective. A C k map f : Rn ⊃ U → Rm is called a
C k submersion if at any a ∈ U , Da f is surjective. A C k map f : Rn ⊃ U → Rm is
called a C k immersion if at any a ∈ U , Da f is injective.
Theorem 1.1.16. Let f : Rn ⊃ U → Rm be a C k submersion, then for any a ∈ U ,
one can find an neighborhood of a U 0 ⊂ U , a open set V ⊂ Rn and a diffeomorphism
ψ : V → U 0 such that, for any x ∈ V :

f (ψ(x1 , · · · , xn )) = (x1 , · · · , xm ).

In other words, up to a change of coordinates in the source, a submersion is locally


a linear projection. Since linear projection are open maps, and being an open map is a
local property property, the previous theorem shows that any submersion is open1 .
1
a map f between topological space is open if the image of any open subset of the source under f is
an open subset of the target.

7
Theorem 1.1.17. Let f : Rn ⊃ U → Rm be a C k immersion, then for any a ∈ U , one
can find an neighborhood of a U 0 ⊂ U , a open set W ⊃ f (U 0 ) and a diffeomorphism
ϕ : W → ϕ(W ) ⊂ Rm such that, for all x ∈ U 0 :

ϕ(f (x1 , · · · , xn )) = (x1 , · · · , xn , 0, · · · , 0).

In this case, f is a linear injection up to a change of coordinates in the image space.


A d-dimensional submanifold of Rn is a set that locally looks like a d-dimensional
linear subspace of Rn :

Definition 1.1.18. S ⊂ Rn is a C k submanifold of dimension d if, for any a ∈ S,


one can find a neighborhood U of a in Rn and C k diffeomorphism Φ : U → V ⊂ Rn such
that :

1. Φ(a) = 0,

2. ψ(S ∩ U ) = Rd ∩ V .

S Rd

V ⊂ Rn
Φ
n
U ⊂R Rn−d

Figure 1.1: The definition of a submanifold

Submanifolds can be characterized in terms of immersions or submersions in the fol-


lowing way :

Theorem 1.1.19. Let S ⊂ Rn . Then the following statements are equivalent :

1. S is a submanifold of dimension d.

2. for any a ∈ S, there exists an open set U ⊂ Rn which contains a and a submersion
f : U → Rn−d such that U ∩ S = f −1 (0).

3. for any a ∈ S, there exists an open set U ⊂ Rn which contains a, an open set
W ⊂ Rd and an immersion g : W → U such that U ∩S = f (W ) and g : W → U ∩S
is an homeomorphism.

The second property says that S is locally the set of solutions of a system of d equa-
tions, the third one says that S admits local parametrizations.

8
1.2 Smooth manifolds
Definition 1.2.1. An n-dimensional topological manifold is topological space M
which satisfies :

1. M is Hausdorff (for any two points x, y ∈ M , one can find disjoint open sets Ux
and Uy such that x ∈ Ux and y ∈ Uy ),

2. M is second countable (it is covered by a countable family of compact subsets),

3. each point m ∈ M has a neighborhood homeomorphic to a open set in Rn (M is


said to be locally euclidean).

A local chart of M is an open subset U ⊂ M together with an homeomorphism


ϕU : U → U 0 = φ(U ) where U 0 is an open subset of Rn . One can write φU in coordinates
φU = (x1 , · · · , xn ), each xi is a function from U to R, the xi ’s are coordinate functions
on U . U is sometimes called a coordinate patch. S
An atlas is a collection of local charts A = (Uα , ϕUα )α∈I such that α∈I Uα = M .
Let (U, ϕU ) and (V, ϕV ) be two local charts which overlap, that is U ∩ V is not empty.
One can then define a map :

ϕU V : ϕU (U ∩ V ) → ϕV (U ∩ V )
x 7→ ϕV (ϕ−1
U (x)).

This map is called the coordinate change, or transition function from U to V .

Definition 1.2.2. Two local charts (U, ϕU ) and (V, ϕV ) are said to be C k compatible
if the transition function ϕU V is a C k diffeomorphism.
An atlas A = (Uα , ϕUα ) is said to be C k if any two charts (U, ϕU ) and (V, ϕV ) in A
are C k compatible.
Two C k atlases A and A0 are said to be compatible if there union is a C k atlas.

ϕV
V

U
ϕU V = ϕV ◦ ϕ−1
U
M

ϕU

Figure 1.2: Transition functions

9
Example 1.2.3. (R, idR ) is a smooth atlas of R (as soon as an atlas has only one chart,
it is automatically smooth). (R, x 7→ x3 ) is also a smooth atlas of R. However, these
two atlas are C 0 compatible but not C k compatible for k ≥ 1.
Example 1.2.4. Let U ⊂ Rn be an open set, then (U, idU ) is a smooth atlas of U . (U, idU )
is not the only smooth atlas that one can use to define a smooth manifold structure
on U , any open covering of U would do. However, note that all these atlases are
compatible.
This raises the question : When should two smooth atlases A and A0 on a given
topological manifold M be considered the same ? The answer lies in the following
theorem :

Theorem 1.2.5. Any C k atlas A is contained in a unique maximal C k atlas Ā. More-
over, if A and A0 are C k compatible, then Ā = Ā0 .

Proof. Consider the set of all local charts (U, phU ) which are C k compatible with A,
this is, by definition, a topological atlas Ā. It is then a routine check to verify that Ā
is indeed a C k atlas. Ā is C k compatible with A by construction, and another routine
check shows that Ā is C k compatible with A0 . More details can be found in [Boo86].
This gives rise to the following definition :

Definition 1.2.6. Given a topological manifold M , a C k structure on M is a maximal


C k atlas Ā. A C k manifold is a topological manifold with a C k structure.

The previous theorem implies that given one C k atlas on M it defines a unique C k
structure on M , even though the particular atlas that we chose at the beginning is far
from being unique.
Remark 1.2.7. Just by changing the requirements on the transition functions, one can
define other structures on topological manifolds : there are real-analytic, holomor-
phic, Lipschitz manifolds...
As we will see, any construction or property which is local and well behaved under
diffeomorphism can be defined for smooth manifolds. For instance:

Definition 1.2.8. Let M be a smooth n-manifold and S ⊂ M , then S is a submanifold


of dimension k in M if and only if for any chart (U, ϕU ), ϕU (S ∩ U ) is a submanifold
of dimension k in ϕ(U ) ⊂ Rn .

1.3 Examples
1.3.1 The Sphere
The sphere Sn is defined by :

Sn = x ∈ Rn+1 kxk2 = 1 .


10
The north pole N and the south pole S are the points with coordinates (0, · · · , 0, 1) and
(0, · · · , 0, −1), we set UN = Sn \{N } and US = Sn \{S}. Remark that Sn = UN ∪ US .
For convenience we will write generic elements of Rn+1 as x = (x̄, xn+1 ) with x̄ ∈ Rn .
The (equatorial) stereographic projection with respect to N is the map : gN : UN → Rn

xn+1
N

O gN (x)
• • x̄
gS (x)

Figure 1.3: Stereographic projections

defined by the following relation, gN (x) is the unique (check it!) intersection of the line
(N x) with the hyperplane H = {x|xn+1 = 0} ' Rn . Similarly, gS : US → RN is defined
by the relation that gS (x) is the intersection of the line (Sx) with H. One can show
that :
x̄ x̄
gN (x) = n+1
, gS (x) = ,
1−x 1 + xn+1
which imply that gN and gS are homeomorphims.
Then we show (by a direct calculation or by remarking that the triangles OgS (x)S
and ON gN (x) are homothetic) :

−1 x̄
gS ◦ gN (x̄) =
kx̄k2

which is a smooth diffeomorphism of Rn \{0}. This shows that (UN , gN ) and (US , gS )
form a smooth atlas of Sn .
−1
In fact gN and gS−1 are parametrization of Sn as a submanifold of Rn+1 (they satisfy
the third point of Theorem 1.1.19 ). This in fact a general phenomenon, any submanifold
as a canonical manifold structure, as we will show in the next section.

1.3.2 Submanifolds
Proposition 1.3.1. A d-dimensional submanifold S ⊂ Rn has a natural manifold struc-
ture.

11
Proof. The topology on S induced by the standard topology on Rn makes S a topological
manifold.
Let us show that the inverse of the parametrizations of S provide a smooth atlas. Let
U, V be two open sets of Rn and US = U ∩ S, VS = V ∩ S and assume we have two
parametrizations ψU : U 0 → US ⊂ U and ψV : V 0 → VS ⊂ V with U 0 and V 0 two opens
sets in Rd . We need two show that the two charts (U, ψU−1 ) and (V, ψV−1 ) are compatible.
We have that ψV−1 ◦ ψU is an homeomorphism from ψU−1 (US ∩ VS ) to ψV−1 (US ∩ VS ), we
need to show it is a C k diffeomorphism. By the inverse function theorem, it is enough
to show that ψV−1 ◦ ψU is invertible at each point.
Let m ∈ US ∩ VS , W a neighborhood of m in Rn and Φ : W → W 0 ⊂ Rn a diffeomor-
phism such that
Φ(S ∩ W ) = {x ∈ Rn |xd+1 = · · · = xn = 0}.
Consider the map Φ ◦ ψU as map from U to Rn . It is an immersion (the differential of
Φ is bijective, the differential of ψU is injective, so their composition is injective). Now
Φ ◦ ψU is indeed valued in Rd , this implies that the differential of Φ ◦ ψU is in fact a
linear map from Rd to Rd . Since it is injective and the dimensions are equal, it is a linear
isomorphism, and thus Φ ◦ ψU is a diffeomorphism by the inverse function theorem.
Writing ψV− 1 ◦ ψU = (Φ ◦ ψU )−1 ◦ (Φ ◦ ψU ) gives that the coordinate change ψV−1 ◦ ψU
is a C k diffeomorphism.

1.3.3 Product manifolds, tori


Proposition 1.3.2. Let M1n and M2m be two smooth manifolds2 , then the product M1 ×
M2 is an n + m smooth manifold whose smooth structure is given by the charts (U1 ×
U2 , ϕU1 × ϕU2 ) where (Ui , ϕUi ) is a chart of Mi , and
ϕU1 × ϕU2 (m1 , m2 ) = (ϕU1 (m1 ), ϕU2 (m2 )) ∈ Rn × Rm .
Proof. Let’s first check that this atlas makes M1 × M2 a topological manifold. Pick
any (m1 , m2 ) ∈ M1 × M2 , and charts (Ui , ϕUi ) around mi on Mi . The U1 × U2 is
a neighborhood of (m1 , m2 ) which is homeomorphic to ϕU1 (U1 ) × ϕU2 (U2 ) which is a
product of an open set in Rn with an open set of Rm , hence an open set in Rn+m .
We now check that the atlas is smooth, let U = U1 × U2 and V = V1 × V2 . First, since
(U1 × U2 ) ∩ (V1 × V2 ) = (U1 ∩ V1 ) × (U2 ∩ V2 ), we only have something to check when
U1 ∩ V1 6= ∅ and U2 ∩ V2 6= ∅. In such a case, the transition map ϕU V : Rn+m → Rn+m is
just given by :
ϕU V (x, y) = (ϕU1 V1 (x), ϕU2 V2 (y)).
It is easy to see that ϕU V is smooth, bijective and has a smooth inverse ϕV U , thus ϕU V
is a smooth diffeomorphism.
An important example is the n-dimensional torus Tn , which is the product of n copies
of S1 .
For n = 2, the smooth manifold T2 is actually homeomorphic to the surface of a
doughnut in R3 .
2
Writing M n is a shorthand to indicate that M has dimension n.

12
1.3.4 Quotients, tori (again) and projective spaces
The quotient topology
First, let’s talk a little bit about the quotient topology.
Recall that a relation R on a set X is said to be an equivalence if is :

• reflexive: xRy ⇔ xRy.

• symmetric: ∀x ∈ X xRx.

• transitive: xRy and yRz ⇒ xRz.

The equivalence class of x ∈ X is the set [x] = {y ∈ X|xRy}. The set of all equivalence
classes forms a partition of X is called the quotient space XR and the quotient map is
just the map π : X → X/R which maps x to its class [x].

Definition 1.3.3. Let X be a topological space and R be an equivalence relation on X,


the quotient topology on X/R is the finest topology on X/R that makes the canonical
projection π : X → X/R continuous.

In other words, U ⊂ X/R is open if and only if π −1 (U ) ⊂ X is open.


If we want to build manifolds using this construction, we need to ensure that X/R is
Hausdorff. This is not automatic : take R and define xRy if x and y are both positive.
Then any neighborhood of the class of 1 in R/R contains 0. However we have the simple
criterion :

Proposition 1.3.4. If π : X → X/R is open and the graph of R3 is closed then X/R
is Hausdorff.

Proof. Let x̄ = π(x) and ȳ = π(y) be two distinct points in X/R. Since π(x) 6= π(y),
(x, y) ∈
/ Γ(R) and, since Γ(R) is closed, we can find Ux and Uy , open sets containing x
and y such that Ux × Uy ∩ Γ(R = ∅. This is shows that Ux̄ = π(Ux ) and Uȳ = π(Uy ) are
disjoint sets in X/R which contain x̄ and ȳ. Since π is open, the sets are open. Thus
X/R is Hausdorff.
Remark 1.3.5. Since the image of compact set under a continuous map between Haus-
dorff spaces is compact, if, under the assumptions of the previous proposition, X is
moreover second countable, X/R is automatically second countable.

Back to tori
Consider the relation on Rn given by :

xRy ⇔ x − y ∈ Zn .

3
defined by Γ(R) = {(x, y) ∈ X|xRy}

13
x2

Z2

x1

Figure 1.4: Rn /Zn

Then M = Rn /R is also the quotient of M = Rn /Zn of the group Rn by its subgroup


Zn . We will show that M is a topological manifold and can be endowed with a natural
smooth structure. We will see in Section 1.4 that M is diffeomorphic to Tn .
But first let’s try to get a more “hands on” feeling of the construction. Look at Figure
1.4. First note that π is a bijection when restricted to [0, 1)n , because quotienting by Zn is
like taking the floor of each coordinate. Consider the closed cube K = [0, 1]n , we already
know that that π is a bijection when restricted to the interior of K (the dashed square
on the figure). Now consider the ’left’ face {x ∈ K|x1 = 0}, any point (0, x2 , · · · , xn ) in
the left face will have the same image under π as the point (1, x2 , · · · , xn ) which lie on
the ’right’ face {x ∈ K|x1 = 1}. In other words the left and right faces of K are glued
by π. This also works along other coordinates xi . One can then say that M is obtained
from K by gluing opposite faces (the identification being given by translations of length
1 along the coordinate axis), this is represented by the arrow tips on the side K on the
figure.

Proposition 1.3.6. M is a topological manifold.

Proof. We first show that M is Hausdorff and second countable. Since Rn is second
countable, we only need to show that the hypothesis of the previous proposition are
satisfied. To see that π : Rn → M is open, consider an arbitrary open set U in Rn . By
definition of the quotient topology, π(U ) is open if and only if π −1 (π(U )) is open in Rn .
We have that : [
π −1 (π(U )) = τp (U )
p∈Zn

14
where τp is the translation by p in Rn . Since each of the τp (U ) open (translations are
homeomorphism), π −1 (π(U )) is open as a union of open sets. We now check that the
graph of R is closed. We have :

Γ(R) = {(x, y) ∈ R2n |x − y ∈ Zn }.

In other words, Γ(R) is the inverse image under the continuous application (x, y) 7→ x−y
from R2n → Rn of Zn , which is closed. Hence Γ(R) is closed. This shows that M is
Hausdorff and second countable. In fact we have more than that : since M is the image
by π of K, M is compact.
We can now show that M is locally Euclidean. Consider x̄ = π(x) ∈ M , and let
B ⊂ Rn be an open ball of radius less than 21 centered at x. Then the restriction π|B of
π to B is injective, and since π is open, π|B is also open and hence an homeomorphism
on its image π(B), which is an open set containing x̄.
We now build an atlas of M . For all x ∈ Rn , denote by Bx of radius 12 centered
at x. What we did in the previous paragraph shows that, setting ϕx = (π|Bx )−1 and
Ux = π(Bx ), the collection (Ux , ϕx )x∈X is an atlas of M .
Proposition 1.3.7. The atlas (Ux , ϕx )x∈X is smooth.
Proof. Consider x, y ∈ Rn such that Ux ∩ Uy 6= ∅. We have the following fact : if
z̄ ∈ Ux ∩ Uy there exists zx ∈ Bx such that π(zx ) = z̄ and zy ∈ Bx such that π(zy ) = z̄.
Since π(zx ) = π(zy ), p = zx − zy ∈ Zn .
Consider the coordinate change from Bx to By ϕUx Uy = ϕUy ◦ ϕ−1 Ux . We have that
ϕUx Uy (zx ) = zy . Moreover, since Bx and By are open, we can find a small ε such that
B(zx , ε) ⊂ Bx and B(zy , ε) ⊂ By . We will show that ϕUx Uy is equal to τp on B(zx , ε).
Let m ∈ B(zx , ε), then m0 = ϕUx Uy (m) = (π|By )(π(m)). So m0 is characterized by :

m0 − m ∈ Zn and m0 ∈ By .

Since p + m satisfy these conditions, m0 = p + m. Thus ϕUx Uy is equal to τp on B(zx , ε),


and is therefore smooth.

The projective spaces RPn and CPn


If E is a vector space, the projectivization P(E) of E is the space of vectorial lines4 in E.
Here we will see that the projectivizations RPn and CPn of Rn+1 and Cn+1 have natural
smooth manifold structures.
To build RPn as quotient, we define, on X = Rn+1 \{0} an equivalence relation R by :

xRy ⇔ x and y are on the same (vectorial) line ⇔ ∃λ ∈ R∗ y = λx.

Then the equivalence class of any point x ∈ X is the vectorial line through x (minus the
origin), so that X/R is the set RPn of vectorial line. The projection π : X → RPn is just
4
A caveat here : a line is a 1-dimensional vector subspace of E, in particular, if one consider complex
vector spaces, a line has real dimension 2.

15
the map which sends x to the line Rx (minus its origin). It is customary in projective
geometry to denote the the image of x = (x1 , · · · , xn+1 ) in RPn by [x] = [x1 : · · · : xn+1 ],
the expression on right hand side of the previous inequality is called the homogenous
coordinates representation of the point [x] ∈ RPn . Note that [x1 : · · · : xn+1 ] and
[λx1 : · · · : λxn+1 ] (λ 6= 0) represent the same point in RPn .

Proposition 1.3.8. RPn is Hausdorff and compact.

Rx

Ry
x

y
O

Figure 1.5: The complement of Γ(R) is open.

Proof. We want to check that RPn with the quotient topology is Hausdorff and second
countable. Like for the case of the torus, for any open set U ⊂ X :
[
π −1 (π(U )) = λU.
λ∈R∗

Thus π −1 (π(U )) is open as union of open sets (homotethies are homeomorphisms), and
π is open.
We want to see that the graph of R is closed. We have :

Γ(R) = (x, y) ∈ X 2 ∃λ ∈ R∗ y = λx .


We show that Γ(R) is closed by showing that its complement is open. Let (x, y) ∈ / Γ(R).
This means that the lines Rx and Ry intersect only at the origin. It is clear that one
can find two neighborhoods Ux and Uy of x and y such that for any x0 ∈ Ux and y 0 ∈ Uy ,
x0 and y 0 are not on the same vectorial line (look at Figure 1.5) . Then Ux × Uy does
not intersect Γ(R). Hence Γ(R) is closed and RPn is Hausdorff.
Since any vectorial line intersects the sphere Sn+1 , RPn = π(Sn+1 ) is compact.

16
We will now build a smooth atlas for RPn . Set, for any integer between 1 and n + 1 :

Ui = [x] = [x1 : · · · : xn+1 ] ∈ RPn xi 6= 0


and :
Hi = x = (x1 , · · · , xn+1 ) ∈ Rn+1 xi = 1 .


Ui is open as it is the image under π of the open set {x ∈ Rn+1 |xi 6= 0} ⊂ X under π
which is open. Define ψi = π|Hi .

Proposition 1.3.9. For each i, ψi is a homeomorphism from Hi to Ui .

Proof. Set ϕi : Ui → Hi by :

x1 xi−1 xi+1 xn
 
1 n+1
ϕi ([x : · · · : x ]) = , · · · , , 1, , · · · , .
xi xi xi xi

ϕi is well defined because ϕi ([λx]) = ϕi ([x]) and has values in Hi . Moreover, it is easy
to see that ϕi ◦ ψi = idHi and ψi ◦ ϕi = idUi . Hence ψi is a continuous bijection, and
since π is open, ψi is open and hence an homeomorphism.
We identify each Hi with Rn by 1 i−1 i+1 n+1
S x 7→ (x ,n· · · , x , x , · · · , x ). This makes each
(Ui , ϕi ) a local chart. Moreover i Ui = RP , so we have an atlas A = {(Ui , ϕi )|1 ≤ i ≤
n + 1}.

Proposition 1.3.10. A is a smooth atlas.

Proof. Let (Ui , ϕi ) and (Uj , ϕj ) be two charts, by permuting the coordinates, we can
assume that i = 1 and j = 2. We have that :

U1 ∩ U2 = [x] = [x1 : · · · : xn+1 ] ∈ RPn x1 6= 0, x2 6= 0 ,


hence:
ϕ−1 2
1 (U1 ∩ U2 ) = {x = (x , · · · , x
n+1
) ∈ Rn |x2 6= 0}
and:
ϕ−1 1 3
2 (U1 ∩ U2 ) = {x = (x , x , · · · , x
n+1
) ∈ Rn |x1 6= 0}.
Now :

ϕ12 (x2 , · · · , xn+1 ) =ϕ2 (ϕ−1 2


1 (x , · · · , x
n+1
))
2 n+1
=ϕ2 ([1 : x : · · · : x ])
1 x3 xn+1
 
= , ,··· , 2
x2 x2 x

which is easily checked to be a smooth diffeomorphism whenever x2 6= 0.

17
Remark 1.3.11. The same construction works to define a smooth manifold structure on:
CPn = {L|L ⊂ Cn+1 complex line},
seen as the quotient of Cn+1 \{0} by the equivalence :
xRy ⇔ ∃λ ∈ C∗ y = λx.
In fact we get more, the coordinate change will have the same form as the one in
the proof of Proposition 1.3.10, except that the real variables xi will be replaced by
complex variables z i , and the coordinate change will therefore be a biholomorphism.
CPn is the first example of a complex manifold.

z
R2 R2 ' {z = 1} ⊂ RP2
L L

L0 L0
L ∩ L0 ∈ RP2

Figure 1.6: There are no parallels in RP2 .

Remark 1.3.12. The embedding Hn+1 ' Rn → RPn allows one to see the n-dimensional
affine space Rn as an open set in RPn . The points that are missed by the embedding
are called points at infinity of the image of Rn in RPn . In this sense, RPn is a
geometric compactification of the affine space Rn .
A projective line L in RP2 is the image under π of a vectorial plane P in Rn+1 .
A point p in RP2 belongs to L if, as a line in Rn+1 , p is included in P . Similarly,
one can define a projective subspace of dimension k as the image under π of a k + 1
dimensional vectorial subspace of Rn+1 . It is easy to see that each k-dimensional
affine subspace of Rn define a unique k-dimensional projective subspace of RPn .
Consider the case n = 2. Then line and points in RP2 satisfy the following prop-
erties:
1. Given two distinct points, there exist a unique line which contains both.
2. Two distinct lines always intersect at a unique point.
This a geometry without parallels ! In fact, two affine lines which are parallel in R2
don’t intersect in R2 but have a unique intersection “at infinity” in RP2 . See Figure
1.6.
These are the first steps in projective geometry, one of the first example of non-
euclidean geometry. A lot more can be found for instance in [Aud03] and [Ber09].

18
1.4 Smooth maps
1.4.1 Definition and first examples
The first advantage of smooth manifolds over topological manifolds is that they allow
us to use calculus, but before really using calculus, we need to say what smooth maps
between manifolds are.
Definition 1.4.1. Let M n and N p be two smooth manifolds, and f : M → N a map. f
is said to be smooth if for any chart (U, φU ) of M , any chart (V, ϕV ) of N , ϕV ◦f ◦(ϕU )−1
is smooth as function from the open set of ϕU (f −1 (V ) ∩ U ) ⊂ Rn to Rk .
In other words, f is smooth if in the following commutative diagram:

f
M ⊃U V ⊂N

ϕU ϕV

Rn ⊃ ϕU (U ) ϕV (V ) ⊂ Rn
ϕV ◦ f ◦ ϕ−1
U

the bottom arrow is smooth. Note that the diagram slightly abuses notations because
ϕV ◦ f ◦ ϕ−1
U doesn’t have U as a source, but ϕU (f
−1
(V ) ∩ U ). We will frequently make
this sort of abuse of to avoid cumbersome notations.
Remark 1.4.2. It is easy to check that if f is smooth for charts in some smooth atlases
AM and AN on M and N , then it is smooth for every other choice of atlas. It is
enough to check the definition for two smooth atlases AM and AN .
Remark 1.4.3. If N is Rk with its standard smooth structure, it is enough to check that
for every chart (U, ϕU ) of M , f ◦ (ϕU )−1 is smooth from U to Rk .
We have the useful notion :
Definition 1.4.4. f : M → N is a diffeomorphism if f is bijective and f −1 : N → M
is smooth.
Any local property of smooth maps between open sets in Euclidean spaces can be
defined through charts, as a property of smooth maps between manifolds:
Definition 1.4.5. f : M → N is a submersion (resp. immersion) (resp. local
diffeomorphism) if for any chart (U, φU ) of M , any chart (V, ϕV ) of N , ϕV ◦f ◦(ϕU )−1
is submersion (resp. immersion) (resp. local diffeomorphism) as a map between open
sets in Euclidean spaces.
With these definitions, we have the following corollaries of Theorem 1.1.19:
Proposition 1.4.6. If f : M → N is a submersion, then, for any x ∈ N , f −1 (x) is a
submanifold of M , whose codimension is the dimension of N .

19
Proposition 1.4.7. If f : M → N is an immersion and an homeomorphism from M
to f (M ), then f (M ) is a smooth submanifold of N , which is diffeomorphic to M .

Remark 1.4.8. In the last proposition, f is said to be a smooth embedding.


We leave the proof as exercises, the guiding line should be: “ using properties of
submersion or immersion in charts, the theorem is reduced to the case where N is a
vectorial subspace of a vector space M , in which case it is trivial”.
We now present some examples.
Example 1.4.9. The projection M1 × M2 → M1 is a smooth submersion.
Example 1.4.10. The projections Rn → Rn /Zn and Rn+1 → RPn built in Section 1.3.4
are also smooth submersions. In fact Rn → Rn /Zn is a local diffeomorphism.
Example 1.4.11. The inclusion M1 → M1 × M2 is a smooth embedding.

1.4.2 Building smooth maps


Proposition 1.4.12. if f : M → N and g : N → P are smooth, then g ◦ f : M → P is
smooth.

Proof. Let a ∈ M and (U, ϕU ), (V, ϕV ), (W, ϕW ) be charts on M , N and P respectively


such that a ∈ U , f (a) ∈ V and g(f (a)) ∈ W . We want to show that ϕW ◦ (g ◦ f ) ◦ ϕ−1
U
is smooth. We have the following commutative diagram :

f g
U V W

ϕU ϕV ϕW

ϕU (U ) ϕV (V ) ϕV (W )
ϕV ◦ f ◦ ϕU−1 ϕW ◦ f ◦ ϕ−1
V

ϕW ◦ (g ◦ f ) ◦ ϕ−1
U

Thus ϕW ◦ (g ◦ f ) ◦ ϕ−1 −1
U is smooth (where it is defined) as ϕV ◦ f ◦ ϕU and ϕW ◦
f ◦ ϕ−1
V are. A caveat here : what we have said only shows that ϕW ◦ (g ◦ f ) ◦ ϕU
−1
−1 −1 −1
is smooth on ϕU (U ∩ f (g (W ) ∩ V )), rather than ϕU ((g ◦ f ) (W ) ∩ U ). However,
since smoothness is local, it is enough to check it in a neighborhood of a for every a ∈ M ,
and U ∩ f −1 (g −1 (W ) ∩ V ) is a neighborhood of a.

Proposition 1.4.13. If N ⊂ M is a submanifold, then the inclusion N → M is a


smooth immersion.

Proposition 1.4.14. If f : M → N is smooth, P ⊂ N is a submanifold of N and


f (M ) ⊂ P then f is smooth as a map from M to P .

20
Submersions are powerful tools to build smooth maps. If π : M → P is a submersion,
the submanifold π −1 (p) is called the fiber of π over p.
Theorem 1.4.15. Let π : M → P be a surjective submersion, and f : M → N be a
smooth map which is constant on the fibers of π, then there exist a unique smooth map
f¯ : P → N such that f = f¯ ◦ π.
f
M N

π

P

Proof. Denote by dM , dP and dN the dimensions of M , N and P .


First, pick p ∈ P and some m ∈ M such that π(m) = p. Since f is constant on π −1 (p),
the formula f¯(p) = f (m) gives a well defined map f¯ : P → N .
We now need to show it is smooth. First pick a chart (Ũm , ϕ̃m ) in M around m,
and a chart (Up , ϕp ) around p, then ϕp ◦ π ◦ ϕ−1
m is a submersion. Thus we can find a
diffeomorphism ψ : O → ϕ̃m (Um ) around ϕ̃m (m) in RdM such that :
ϕp ◦ π ◦ ϕ̃−1 1
m ◦ ψ(x , · · · , x
dM
) = (x1 , · · · , xdP ).
By setting ϕm = ϕ̃m ◦ ψ −1 , we define a new chart (ϕm , Um ) around m such that :
ϕp ◦ π ◦ ϕ−1 1
m (x , · · · , x
dM
) = (x1 , · · · , xdP ).
In these coordinates, the fiber in Um over some point ϕ−1 1 dP
p (x , · · · , x ) in P is just :

ϕ−1 1 dP dM −dP

m {(x , · · · , x )} × R .
Pick any chart (UN , ϕN ) around f (m) = f¯(p) in N . The fact that f is constant on
the fibers thus translates as that ϕN ◦ f ◦ ϕm doesn’t depend on xi for i > dP . Hence
there exist a smooth function f˜ of dP variables with values in RdN such that:
ϕN ◦ f ◦ ϕ−1 (x1 , · · · , xdM ) = f˜(x1 , · · · , xdP ).
m

Moreover we have that:


ϕN ◦ f¯ ◦ ϕ−1 1 dP −1 1
p (x , · · · , x ) = ϕN ◦ f ◦ ϕm (x , · · · , x
dM
) = f˜(x1 , · · · , xdP ).
This shows that f¯ is smooth around p.
Example 1.4.16. The complex exponential map ϕ : R → S1 defined by ϕ(t) = eit is a
local diffeomorphism. Consider now f : R → R a 2π periodic smooth function. f is
constant over the fibers of ϕ, and thus yields a smooth function f¯ from S1 to R.
f
R R

ϕ

S1

21
Example 1.4.17. We consider here the projective space RPn with its projection π :
Rn+1 \{0} → RPn which is a submersion.
Consider an invertible linear map L : Rn+1 → Rn+1 , since L is invertible, Lv 6= 0
when v 6= 0. Therefore we can build the composition L0 = π ◦ L : Rn+1 \{0} → RPn
which is smooth as a composition of smooth maps. Moreover, it is easy to see that
π(Lv) = π(Lλv) for any nonzero λ. This implies that L0 is constant on the fibers of
π, so it defines a unique smooth map : L̃ = RPn → RPn .

L
Rn+1 \{0} Rn+1 \{0}

L0
π π

RPn RPn

In fact, L̃ is a diffeomorphism, as is seen by making the same construction with L−1


and checking that the map we obtain is the reciprocal of L̃.
These transformations are called homographies of RPn , and are the natural trans-
formations of projective geometry.

1.5 The tangent space at a point


In the preceding section, we have defined what smooth maps are, but haven’t say what
their derivatives are. Before being able to do so, we need another construction.
The big idea of differential calculus is to approximate a map at a point by a linear
map. When f goes from Rn → Rk its differential at a point will be a linear map from
Rn to Rk . However, when f is a map between two manifolds, if we want to approximate
f by a linear map, we first need to understand between which vector spaces should be
the source and the target of such a linear map.
Remember that in the first section it was possible to compute the differential of a
map: f : Rn → Rk at a by computing the derivative of the one variable functions f ◦ c
for every smooth curve c that pass through a. This is the idea we will follow to build
the tangent space.

1.5.1 Some words about the tangent space to a submanifold in


Rn
Consider S ⊂ Rn a k-dimensional submanifold and m ∈ S. We define the tangent
space Tm S to S at m is defined in the following way :

C∞
n o
Tm S = v ∈ Rn ∃c : R −−→ S, c(0) = m, ċ(0) = v .

22
Lemma 1.5.1. Tm S is a vectorial subspace of Rn .
Proof. To see that v ∈ Tm S ⇒ λv ∈ Tm S, we consider c(t) ∈ S such that c(0) = m and
ċ(0) = v, and define c̃(t) = c(λt) ∈ S. Then c̃(0)˙ = λv.
It remains to show that v1 , v2 ∈ Tm S ⇒ v1 + v2 ∈ Tm S. For this, we consider a local
parametrization of S ψ : Rk → S such that ψ(0) = m. Let c1 and c2 be two curve in S
such that ċ1 (0) = v1 and ċ2 (0) = v2 , and c̃i = ψ −1 ◦ ci . The chain rule implies :

D0 ψ c̃˙i (0) = ċi (0) = vi .




Consider the curve c(t) = ψ(c̃1 (t) + c̃2 (t)). This is a curve in S that goes through m,
and :
ċ(0) = D0 ψ c̃˙1 (0) + c̃˙2 (0) = v1 + v2 .


Thus v1 + v2 ∈ Tm S.
Remark 1.5.2. This definition is not the most practical to use in applications. How-
ever it is easily seen to be equivalent to the following more computation friendly
characterizations :
• If f : Rn → Rn−k is a local equation of S around m, then Tm S = ker(Dm f ).

• If ψ : Rk → S is a local parametrization of S such that ψ(0) = m, then Tm S =


D0 ψ(Rk ).
Example 1.5.3. The tangent space to Sn at v is the orthogonal to v.

1.5.2 Tangent vectors at a point to a manifold


Let M be a smooth manifold and m ∈ M , we define CM (m) to be the set of smooth
curves c : R → M such that c(0) = m.
Two curves c1 , c2 ∈ CM (m) are said to be tangent at m if there exist a chart (U, ϕU )
around M such that :
(ϕU ◦ c1 )0 (0) = (ϕU ◦ c2 )0 (0).
This doesn’t depend on the chart we used. Indeed, if (V, ϕV ) is another chart around
m, then :

(ϕV ◦ c1 )0 (0) = (ϕU V ◦ ϕU ◦ c1 )0 (0)


= DϕU (m) ϕU V ((ϕU ◦ c1 )0 (0))
= DϕU (m) ϕU V ((ϕU ◦ c2 )0 (0))
= (ϕV ◦ c2 )0 (0).

Moreover, the relation “c1 is tangent to c2 at m” is an equivalence on CM (m). We will


denote it c1 Rc2 .
Definition 1.5.4. The set Tm M of tangent vectors to M at m is the quotient of
CM (m) by R.

23
The equivalence class of a curve c ∈ CM (m) (its image under the quotient map
CM (m) → Tm M ) is called the tangent vector to c at 0, and is denoted by c0 (0).
Example 1.5.5. The set of tangent vectors at a point m ∈ U ⊂ Rn is naturally isomorphic
to Rn . The isomorphism is given mapping a curve c through m to c0 (0) ∈ Rn and
passing to the quotient.
Proposition 1.5.6. Tm M has a natural vector space structure.
Proof. Given a chart (U, ϕU ) around m, we define a map θϕU : Tm M → Rn by :
θϕU (ξ) = (ϕU ◦ c)0 (0)
where c is a representative of ξ in Tm M . This is an injection by definition. Moreover, if
v ∈ Rn and c(t) = ϕ−1 U (tv) then the image of the class of c by θϕU is v. Thus θϕU is a
bijection from Tm M to Rn .
Moreover, if (V, ϕV ) is another chart, one can check that :
θϕV ◦ θϕ−1U = DϕU (m) ϕU V .
Hence θϕV ◦ θϕ−1U is a linear isomorphism of Rn .
Thus we can define a vector space structure on Tm M in the following way:
ξ + η = θϕ−1U (θϕU (ξ) + θϕU (η)) , λξ = θϕ−1U (λθϕU (ξ)),
the preceding remark show that this structure is independent of the chart we used.
Definition 1.5.7. Tm M with the natural vector space structure built in the preceding
proposition is called the tangent space to M at m.
Example 1.5.8. Assume that S ⊂ Rn is a submanifold. Then its tangent space as a
submanifold is naturally isomorphic to its tangent space as a manifold.

1.5.3 Tangent vectors in coordinates


Let M be a manifold and (U, ϕU ) a chart around m ∈ M . Then the map θϕU : Tm M →
Rn is (by construction) a linear isomorphism. We will denote the coordinates on ϕU (U )
by (x1 , · · · , xn ). Let BU = ∂x∂ 1 m , · · · , ∂x∂n m be the image under θϕU of the standard
basis of Rn (for a justification of the notation, see section 1.6.2). BU is a basis of Tm M
and thus any vector v ∈ Tm M can be written as :
n
i ∂
X
v= v i .
i=1
∂x m

1 n
Now consider another chart (V, ϕV) centered
at m and denote by (y , · · · , y ) the
coordinates on ϕV (V ) and by BV = ∂y∂ 1 , · · · , ∂y∂n the corresponding basis (via

m m
θϕV ) of Tm M . Then :
n
X ∂
j
v= ṽ .
j=1
∂y j m

24
The law according to which the coordinates v i of v with respect to BU change to the
coordinates ṽ i is given by θV ◦ θU−1 = DϕU (m) ϕU V . If we see the change of coordinates
ϕU V as a n-tuple of functions (y 1 (x1 , · · · , xn ), · · · , y n (x1 , · · · , xn )), then the matrix of
DϕU (m) ϕU V with respect to the basis BU and BV has coefficients :
∂y j
.
∂xi
Therefore : n
j
X ∂y j
ṽ = vi.
i=1
∂xi
This is the way the tangent space used to be defined in old text books : a tangent
vector was an n-tuple of numbers associated to a coordinate system which changed
according to the above law when one changed coordinates.

1.6 Differentiating maps between manifolds


1.6.1 Definitions, first properties
Definition 1.6.1. Let f : M → N be a smooth map, then the tangent map (or
differential) of f at m ∈ M is the map : Tm f : Tm M → Tf (m) N defined by :
Tm M 3 v 7→ Tm f (v) = (f ◦ c)0 (0) ∈ Tf (m) N.
To see that this definition makes sense, remark, using a chart around m and a chart
around f (m), that if c1 and c2 are tangent at m, then f ◦ c1 and f ◦ c2 are curves which
are tangent at f (m).
Another way to phrase the definition is to say that Tm f : Tm M → Tf (m) N is built
from the map f˜ : CM (m) → CN (f (m)), c 7→ f ◦ c by quotienting by the tangency relation
at the source and the target space.
Example 1.6.2. By definition, the tangent map of ϕU : U → ϕU (U ) ⊂ Rdim M is the map
θϕU defined in the previous section.
Example 1.6.3. Assume that M and N are open sets in euclidean space, then Tm f is
just the usual differential.
Just like the differential, the tangent map satisfies the chain rule :
Proposition 1.6.4. If f : M → N and g : N → P are smooth, then, for any m ∈ M :
Tm (g ◦ f ) = Tf (m) g ◦ Tm f.
Proof. Let cM be a curve through m. Then cN = f ◦ c is a curve through f (m) and
cP = g ◦ f ◦ c is a curve through g(f (m)). Denote by ξM , ξN and ξP the tangent vectors
to cM , cN and cP . We have that, by definition of the differential :
ξN = Tm f (ξM ), ξP = Tf (m) g(ξN ), ξP = Tm (g ◦ f )(ξM ).
This ends the proof.

25
Proposition 1.6.5. Tm f : Tm M → Tf (m) N is linear.
Proof. Consider charts (U, ϕU ) around m and (V, ϕV ) around f (m). We have the fol-
lowing commutative diagram :

f
U V

ϕU ϕV

ϕU (U ) ϕV (V )
ϕV ◦ f ◦ ϕ−1
U

All the arrows of this diagram are smooth maps, using the chain rule, we then get the
following diagram between tangent spaces :

Tm f
Tm M Tf (m) N

θϕU θϕV

Rdim M Rdim N
TϕU (m) (ϕV ◦ f ◦ ϕ−1
U )

Since the arrows on the side are linear isomorphism, and the arrow on the bottom is
linear, Tm f is linear.
From the proof of the previous proposition, we see that requiring some property of
DϕU (m) (ϕV ◦ f ◦ ϕ−1
U ) which which can be expressed independently of a choice of basis
in Rdim M and Rdim N is the same as requiring this property on Tm f , thus we have :
Proposition 1.6.6. f : M → N is an immersion (resp. submersion) (resp. local
diffeomorphism) if and only if at each point m ∈ M , Tm f is injective (resp. surjective)
(resp. an isomorphism).
Example 1.6.7. Consider the canonical projection π : Rn+1 \{0} → RPn and f the re-
striction of π to Sn . We will show that f is a local diffeomorphism. Denote by
ι : Sn → Rn+1 the inclusion. The differential of ι at v ∈ Sn is just the inclusion
Tι : v ⊥ = Tv Sn → Rn+1 .
We then have that : Tv f = Tv π ◦Tv ι. We want to show that Tv f is an isomorphism.
We already know that Tv π is surjective and Tv ι, Thus we only need to show that
im Tv ι ∩ ker Tv π = {0}.
For dimensional reason, ker Tv π has dimension π. Moreover, if we set c(t) = (1+t)v,
we have that π(c(t)) ∈ RPn is constant. Beware here that v is viewed simultaneously
as a point in Rn+1 \{0} and as a tangent vector to Rn+1 at v, this makes sense only
because Rn+1 is an open set on a vector space. This shows that Tv π(v) = 0, thus
KerTv π = Rv. In particular im Tv ι ∩ ker Tv π = {0}.

26
1.6.2 Tangent vectors as derivations
This section gives another view on the tangent space at a point. The only point we will
use is the notation defined in the next paragraph. More details (including proofs !) can
be found in [Lee02].
Let p ∈ M and Vp ∈ Tp M . We define an action of Vp on smooth real valued functions
f : U → R on any neighborhood U of p by :

Vp · f = Tp f (Vp ) ∈ R.

Remark 1.6.8. This gives a justification of the notation we introduced in section 1.5.3.

With the notations of section 1.5.3, any chart (U, ϕ) around p defines vectors ∂x∂ i p ∈
Tp M , one then has :

∂(f ◦ ϕ−1 ) −1


· f = (ϕ (p)).
∂xi p ∂xi

Proposition 1.6.9. f 7→ Vp · f has the following properties :

• it is local : if f and g are equal on a neighborhood of p, Vp · f = Vp · g.

• (Vp , f ) 7→ Vp · f is R-bilinear.

• it satisfies Leibnitz rule : Vp · (f g) = f (p)(Vp · f ) + g(p)(Vp · f ).

f 7→ Vp · f defines a map from the algebra C ∞ (M ) of smooth functions on M to R.


The third property is the defining property of a derivation at p.

Definition 1.6.10. A derivation at p ∈ M is a linear map δ : C ∞ (M ) → R such that


for any f, g ∈ C ∞ (M ) :
δ(f g) = f (p)δ(g) + δ(f )g(p).

Let us denote by Dp the set of derivations on Cp . It is naturally a vector space. Then


:

Proposition 1.6.11. The map L : Tp M → Dp defined by Vp 7→ (f 7→ Vp · f ) is a linear


isomorphism.

The proof of this proposition uses a tool we haven’t introduced yet: cut-off functions,
we’ll come back to it later.
This provides yet another definition of the tangent space, perhaps less intuitive than
the one we have used. It has however the advantage that the vector space structure is
obvious when one uses this definition.

27
1.7 Exercises
Exercise 1.1. Let U ⊂ Rn be an open convex set. f is said to be convex if, for any
x, y ∈ U , any t ∈ [0, 1] :

f ((1 − t)x + ty) ≤ (1 − t)f (x) + tf (y).

Assume that f is C 1 . Show that f is convex if and only if, for any x, y ∈ U :

f (y) − f (x) ≥ Dx f (y − x).

Exercise 1.2. Consider f : Mn (R) → Mn (R) defined by f (A) = A2 .

1. Compute, for any A ∈ Mn (R), DA f .

2. Use the inverse function theorem to prove that any matrix which is close enough
to the identity has a square root.

Exercise 1.3. Using the inverse function theorem we give a proof of Theorem 1.1.16.
f : Rn → Rm will be a smooth function whose differential at 0 is surjective.

1. write f = (f 1 (x1 , · · · , xn ), · · · , f m (x1 , · · · , xn )). Show that, up to a permutation


of the coordinates in Rn , one can assume that the square m × m matrix whose
coefficients are ∂i f j (0) for i, j ≤ m is invertible.

2. Consider the function F : Rn → Rn defined by :

F (x1 , · · · , xn ) = (f 1 (x1 , · · · , xn ), · · · , f m (x1 , · · · , xn ), xm+1 , · · · , xn ).

Show that F is a local diffeomorphism at 0.

3. Show that this implies Theorem 1.1.16.

Exercise 1.4. Let f : Rn → R and g : Rn → R be two smooth function. Define


Σ = g −1 (0). Assume that x ∈ Σ is a minimum of f |Σ . Show that there exist Λ ∈ R such
that Dx f = λDx g.
Hint: First prove, using the submersion theorem, that whenever Dx f 6= 0, for any
v ∈ ker Dx f , there is a smooth curve γ in Σ such that γ(0) = x and γ̇(0) = v.
Exercise 1.5. * The goal of this exercise is to compute the differential of det : Mn (R) →
R.

1. Show that det is smooth. Hint: written in coordinates, it is a polynomial.

2. Let I be the identity matrix, show that : DI det H = trace(H). Hint : Use the
multilinearity of the determinant.

3. Show that whenever A is invertible, DA det(H) = det(A) trace(A−1 H).

28
4. Show that for any A ∈ Mn (R), DA det(H) = trace(ÃH), where à is the cofactor
matrix of A.
Exercise 1.6. * We compute the differential of inv : GLn (R) → GLn (R)
1. Show that GLn (R) is open.

2. Using that (I + H)−1 = k≥1 (−H)k for kHk ≤ 1, compute DI inv.


P

3. Show that DA inv(H) = −A−1 HA−1 .

4. Show that inv is smooth without actually computing its differential (use that
A−1 = det(A)−1 Ã). Compute the differential of inv by differentiating the relation
A × inv(A) = idRn .
Exercise 1.7. Recall that, for M ∈ Mn (K) (K is R or C) :
X Mk
exp M =
k=0
k!

1. Show that exp is C 1 .

2. Show that D0 exp = idMn (K) .


Exercise 1.8. ** We will use the inverse function theorem to prove that exp : Mn (C) →
GLn (C) is surjective.
1. Let G be a connected topological group5 , and H be a subgroup of G which contains
a neighborhood of the neutral element of G. Show that G = H Hint : H and its
complement are open.

2. Show that (Mn (C), +) and (GLn (C), ×) are topological groups.

3. Let A be a fixed element of GLn (C), and define :

C[A] = {P (A) | P ∈ C[X]

Show that +, × and exp preserve C[A]. Hint: for exp, show that if M ∈ C[A],
then exp M is a limit of element of C[A].

4. Show that if M ∈ C[A] is invertible, then M −1 ∈ C[A]. Hint: Cayley-Hamilton


can help.

5. Let C[A]× = C[A] ∩ GLn (C). Show that exp : (C[A], +) → C[A]× is topological
group morphism.

6. Show that exp is a diffeomorphism on a neighborhood of 0.


5
A group G is a topological group if it is equipped with a topology which makes the group law and
the inverse continuous.

29
7. Conclude that exp(C[A]) = C[A]× , and that exp : Mn (C) → GLn (C) is surjective.

The next exercises give important example of submanifolds.


Exercise 1.9. Show that Sn is a submanifold of Rn+1 .
Exercise 1.10. For which values of c is Σc = {(x, y, z) ∈ R3 |x2 + y 2 = z 2 + c} a
submanifold of R3 ?
Exercise 1.11. Show that the matrix groups SL(n, R), O(n, R) and SO(n, R) are sub-
manifolds of Mn (R) ' Rn×n .
Starting from now, we get to the core of the chapter, manifolds, smooth maps...
Exercise 1.12. Let E be a n dimensional vector space. For any basis B = (b1 , · · · , bn )
of E, we define a chart (E, ϕB ) on E in the following way: for x ∈ E, ϕB (x) ∈ Rn is the
vector formed by the coordinates of x in basis B.
Show that A = {(E, ϕB )|B is basis of E} is a smooth atlas of E.
Exercise 1.13. Consider X = R × {0, 1}. Define an equivalence relation on R on X
by :
(x, a)R(y, b) ⇔ x 6= 0 and x = y.
X/R with the quotient topology is called the line with two origins.

1. Show that X is not Hausdorff.

2. Show that each point of X has a neighborhood homeomorphic to an open interval


in R.

Exercise 1.14. Let M1 be R with the atlas (R, idR ) and M2 be R with the atlas (R, x 7→
x3 ).

1. Show that these two atlases are not C k compatible for any k ≥ 1.

2. Describe the smooth maps from M1 to M2 and from M2 to M1 .

3. Show that M1 and M2 are diffeomorphic as smooth manifolds.

Exercise 1.15. Consider the sphere Sn ⊂ Rn+1 . Set, for 1 ≤ i ≤ n + 1, Ui+ =


{x ∈ Sn |xi > 0} and Ui− = {x ∈ Sn |xi > 0}.
Define ϕ+ + n + 1
i : Ui → R by ϕi (x , · · · , x
n+1
) = (x1 , · · · , xi−1 , xi+1 , · · · , xn+1 ), and ϕ−
i :
− n
Ui → R by the same formula.
− −
1. Show that the charts (Ui+ , ϕ+
i ) and (Ui , ϕi ) form a smooth atlas A of S
n

2. Show that this atlas define the same smooth structure on Sn as the atlas defined
by the stereographic projections.

Exercise 1.16. 1. Show that RP1 is diffeomorphic to S1 .

30
2. Consider the Riemann sphere Ĉ = C ∪ {∞}. Consider the following atlas on Ĉ:

(C, z 7→ z) (Ĉ\{0}, z 7→ 1/z)

with the convention that 1/∞ = 0 and 1/0 = ∞. Show that this atlas is smooth.

3. Show that Ĉ is diffeomorphic to CP1 and S2 .


Exercise 1.17. Consider ψ : R2 → C2 ' R4 defined by :
1
ψ(t, s) = √ (eit , eis ).
2
1. Show that S = ψ(R2 ) is a submanifold of R4 .

2. Show that S is in fact a submanifold of S3 .

3. Show that S is diffeomorphic to T2 .

Remark 1.7.1. S is called the Clifford torus. It has been recently proven by Coda-
Marques and Neves that, in some precise sense, it is the least twisted torus in S3 ,
solving a conjecture known as the Willmore conjecture.
Exercise 1.18. ** Let G be a discrete group which acts on a smooth manifold M . We
assume that:

• x 7→ g · x is smooth (the action is smooth).

• For any compact subsets K, L ⊂ M , #{g ∈ G|g · K ∩ L 6= ∅} is finite (the action


is proper).

• For any g ∈ G\{e}, for any x ∈ M , g · x 6= x (the action is free).

We will show that the quotient6 M/G has a natural smooth manifold structure.

1. Show that the quotient map M → M/G is open. Hint: Argue as in section 1.3.4,
only the smoothness of the action matters here.

2. Show that M/G is Hausdorff. Hint: the proof is similar to section 1.3.4, properness
must be used.

3. Show that any point x ∈ M admits a neighborhood V such that, for any g ∈ G\{e},
g · V ∩ V = ∅. Deduce that πV is an homeomorphism from V to π(V ).

4. Since M is a manifold, one can find a chart (U, ϕU ) of M around x such that
U ⊂ V . Then, setting Ū = π(U ) and ϕ̄U = ϕU ◦ π|−1
U , we have that (Ū , ϕ̄U is a
chart around π(x).
Show that the charts obtained by these process form a smooth atlas of M/G.
6
the equivalence induced by the group action is xRy if and only if y = g · x for some g ∈ G.

31
Exercise 1.19. * The goal of this exercise is to build a smooth embedding of RP2 in
R4 .
Consider the bilinear map B : R3 × R3 → R5 defined by :
0
 
   0  xx
x x  xy 0 + x0 y 
0
 0 0 0 

B  y  ,  y  =  xz + yy + x z .
z z0  yz 0 + y 0 z 
zz 0

1. Show that B(v, v 0 ) = 0 if and only of v = 0 or v 0 = 0.


Hint: Show that, when v 6= 0, the linear map v 0 7→ B(v, v 0 ) is injective using that
an n × m matrix has a non trivial kernel if and only if all its m × m minors vanish.

2. Consider G : R3 \{0} → S4 defined by:


B(v, v)
F (v) = .
kB(v, v)k
Show that F (v) = F (v 0 ) if and only if v 0 = λv.
Hint: g(v) = g(v 0 ) ⇒ B(v, v) = λ2 B(v 0 , v 0 ) ⇒ B(v + λv 0 , v − λv 0 ) = 0.

3. Show that there exist a unique smooth f : RP2 → S4 such that f = F ◦ π where
π : R3 \{0} → RP2 is the usual projection.

4. Show that f is a smooth embedding.


Hint: To see that f is an immersion, show that ker Tv F = ker Tv π. The following
commutative diagram may help:

R3 \{0} R5

π F

RP2 S4
f

/ f (RP2 ) and use it to define a smooth embedding of RP2


5. Show that (0, 1, 0, 0, 0) ∈
4
into R .
Exercise 1.20. Let M and N be two compact manifolds of the same dimension and
f : M → N a smooth map. y ∈ N is called a regular value of f if for any x ∈ f −1 ({y}),
Tx f is invertible. Let y be a regular value of f .
1. Show that f −1 ({y}) is finite.

2. Show that the cardinal of f −1 ({y 0 }) is constant for y 0 is some neighborhood of y.

32
Exercise 1.21. * The goal of this problem is to give a proof of the fundamental theorem
of algebra with the tools introduced in this chapter.
Consider P ∈ C[X] a non constant complex polynomial.

1. Show that the function P̂ : Ĉ → Ĉ defined by :


(
P̂ (z) = z if z ∈ C
P̂ (∞) = ∞.

is smooth.

2. Let F = {y ∈ Ĉ|y is not a regular value of P̂ .}. Show that F is finite.

3. Show that the function z 7→ #f −1 ({z}) is constant on Ĉ\F .

4. Show that P̂ is surjective and that this imply the fundamental theorem of algebra.

33
2 Tangent bundle, Vector fields and
Lie brackets
2.1 Vector fields in Rn.
This part is quite fast and sketchy, a more detailed exposition can be found in [HS74].

2.1.1 General theory


Definition 2.1.1. Let U ⊂ Rn be an open set. A smooth vector field on U is a
smooth map V : U → Rn .
Vector fields on U ⊂ R2 are usually plotted in the following way: at each point x, one
draws the vector V (x) with its “foot” set at x. A high level reason to do that is that one
usually sees the vector V (x) as an element of Tx R2 ' R2 , a tangent vector to R2 at x.
We will also sometimes write the value of V at x as Vx .

Figure 2.1: The vector field V (x, y) = (−y/2, x/2) in R2

The Cauchy problem for V with initial condition x0 ∈ U is the following problem:
find a function x(t) for t ∈ I ⊂ R some interval containing 0 such that:
(
x0 (t) = V (x(t)) for all t ∈ I
(2.1)
x(0) = x0 .

34
A solution to (2.1) is a curve t ∈ I 7→ x(t) ∈ U which satisfies (2.1).
Remark 2.1.2. If t 7→ x(t) is a solution, then x̃(t) = x(t + t0 ) is a solution with initial
condition x̃(0) = x(t0 ).
A maximal solution of (2.1) is a solution x : I → U of (2.1) which cannot be extended
further: if x̃ : I˜ → U is another solution of (2.1) such that I˜ ⊃ I and x̃|I = x, then
˜
I = I.
We have the following important theorem:

Theorem 2.1.3. Let V be a smooth vector field on U ⊂ Rn , then for any x0 ∈ U the
Cauchy problem has a unique maximal solution.

The following property of maximal solutions is useful in the applications:

Proposition 2.1.4. Let x : [0, T ) → U be a solution to (2.1), with T < ∞ then either
for any compact K ⊂ U there exists tK ∈ [0, T ) such that x(tK ) ∈
/ K or x can be extended
to a solution x̃ : [0, T + ε) → U .

Proof. Assume that there exist a compact K ⊂ U such that x(t) ∈ K for all t ∈ [0, T ),
R T to show thatRxT is not maximal. Since K is compact, V is bounded on K. Thus
we want
x1 = 0 V (x(t))dt = 0 x0 (t)dt exists and limt→T x(t) = x1 .
Let y(t)t∈[0,T 0 ) be the maximal solution to the Cauchy problem y(0) = x1 , y 0 = V (y)
and set: (
x(t) if t < T ,
x̄(t) =
y(t − T ) if t ∈ [T, T + T 0 ).
Then x̄ is C 0 and satisfy x̄0 = V (x̄) and x̄(0) = x0 . Thus x is not maximal.
Example 2.1.5. If one can show that any solution to (2.1) stays in a compact set, then
the maximal solution is defined for all t ∈ R.

Definition 2.1.6. V is said to be complete if for any x0 ∈ U the maximal solution


x(t) to (2.1) with initial condition x0 is defined for all t ∈ R.

We now assume that V is a smooth vector field. We define O ⊂ U × R by (x, t) ∈ O if


and only if the maximal solution to the Cauchy problem y 0 = V (y), y(0) = x is defined
at t, if V is complete, then O = U × R. For (x, t) ∈ O, we define ϕt (x) = y(t) where
y(t) is the solution to y 0 = V (y), y(0) = x. (x, t) 7→ ϕt (x) is called the local flow of V .

Proposition 2.1.7. ϕt has the following property :

1. (x, t) 7→ ϕt (x) is smooth.

2. ϕt+s (x) = ϕt (ϕs (x)) whenever that makes sense. Moreover, if V is complete then
for any t, ϕt : U → U is a smooth and satisfies ϕt+s = ϕt ◦ ϕs . In this case ϕt is
a smooth diffeomorphism of U .
d
3. ϕ (x)
dt |t=0 t
= Vx .

35
2.1.2 Linear vector fields and matrix exponential
Definition 2.1.8. A vector field V on Rn is linear if and only if V (x) = Ax for some
linear map A from Rn to Rn .
Definition 2.1.9. For any A ∈ Mn (R), the exponential eA of A is defined by :

X Ak
eA = .
k=0
k!

It is easy to see that, if Mn (R) is equipped with an operator norm:


∞ k ∞
X A X kAkk

k! ≤ ekAk ,
k=0 k=0
k!

so this series converge absolutely and thus eA is well defined. Moreover if we interpret
the serie as a serie of functions from Mn (R) to Mn (R), then the previous inequality says
that this serie is normally convergent on every compact set, in particular : A 7→ eA is
continuous.
Just like functions Ceαt are all solutions to the (scalar!) differential equation x0 = αx,
we have :
Proposition 2.1.10. Consider the following Cauchy problem:
(
x0 = Ax
x(0) = x0 .

The maximal solution is given by :

x(t) = etA x0 t ∈ R.

Proof. The vector field x 7→ Ax is smooth, so the previous section tells us that there
is a unique maximal solution, so we only need to check that the formula defining x(t)
actually yields a solution.
k
To see this, set fk (t) = (tA)
P
k!
x0 . Then x(t) = k≥0 fk (t). We want to differentiate the
serie term by term, we can do it because:
X X tk−1 Ak x0 X (tA)k−1

0 tkAk
|fk (t)| = (k − 1)! ≤
kAk
(k − 1)! |x0 | ≤ kAke
|x0 |,
k≥0 k≥1 k≥1

and the serie of the derivatives normally converges on compacts of R.


Then:
!
X X tk−1 Ak x0 X (tA)k−1
x0 (t) = fk0 (t) = =A x0 = AetA x0 = Ax(t).
k k≥1
(k − 1)! k≥1
(k − 1)!

36
This shows that the flow of V (x) = Ax is particularly simple in this case, we have:
ϕt (x) = etA x.
The matrix exponential have the following properties:
−1
Proposition 2.1.11. 1. If P is invertible, eP AP = P eA P −1 .

2. If AB = BA, eA+B = eA eB .

3. e−A = (eA )−1 .

4. det eA = etrace A .

2.2 Cut-off functions


The goal of the section it to prove the following proposition:

Proposition 2.2.1. Let M be a smooth manifold, O ⊂ M an open set and K ⊂ O a


compact set. Then there exist a smooth function η : M → R such that:

• 0 ≤ η ≤ 1.

• for all x ∈ K, η(x) = 1.

• for x ∈
/ O, η(x) = 0.

Proof.
Claim 2.2.2. The proposition is true for M = Rn , K = B(0, r) and O = B(0, R) (r < R).
First consider the function f : R → R defined by:
(
0 if t ≤ 0,
f (t) =
e−1/t if t > 0.

It is a classical fact that f is smooth. Setting g(t) = f (t)f (1 − t) give aR nonnegative


smooth function with support [0, 1], scaling g if necessary, we can assume R g(t)dt = 1.
Rt
Then G(t) = −∞ g(s)ds is nondecreasing, equal to 0 for t < 0 and equal to 1 for t > 1.
Then for any r, R > 0, one can find parameters α and β such that Gr,R (t) = G(αt + β)
is smooth, nonincrreasing, equal to 1 for t < r and equal to 0 for t > R. Finally, setting
η(x) = Gr,R (|x|) does the job.
Claim 2.2.3. The proposition is true if O is included in U for some chart (U, ϕU ).
Reducing O, we can assume the closure of O is in U . We will in fact build η on
U , we can then extend it to 0 outiside U .Since charts are smooth diffeomorphsims,
we can work in ϕU (U ). So we might as well assume that O ⊂ Rn . Since K ⊂ O,
for each p ∈ K we can find rp such that B(p, rp ) ⊂ O. Now, the B(p, rp /2) form an
open cover of K which is compact and we can find finitely many points p1 , · · · , pN such

37
that K ⊂ N n
S
i=1 B(pi , rpi /2). Let ηi : R → [0, 1] be a smooth function which is 1 on
B(pi , rpi /2) and 0 outiside of B(pi , rpi ). Finally we set:

η = 1 − (1 − η1 ) · · · (1 − ηN ),

and it works.
We can now prove the general result. Since K is compact, we can find S a finite number
of charts (U1 , ϕ1 ), · · · , (UN , ϕN ) such that each Ui ⊂ O and K = i Ki where each
Ki ⊂ Ui is compact. We apply the previous claim for each pair Ki ⊂ Ui to get ηi , and
set :
η = 1 − (1 − η1 ) · · · (1 − ηN ).

Using these cutoff functions, we can build partitions of unity:

Proposition 2.2.4. Let M be a smooth manifold and (Oi )i∈I be a locally finite1 open
cover of M . Then there exists smooth functions ηi : M → R such that:

• the support of ηi is included in Oi .


P
• i∈I ηi = 1.

Proof. First we can find an open cover Vi of M such that for each i, the closure Ki
of Vi is included in Ui . We then consider the cutoff functions η̄i associated to the pair
Ki ⊂ Ui , and set:
η̄i (x)
ηi (x) = P .
j∈I η̄j (x)

2.3 From the tangent spaces to the tangent bundle


2.3.1 T M as smooth manifold
We want to study vector fields on manifolds. In order to do this we first need to define
smooth vector fields. A vector field V on a smooth n-manifold M S should associate to
each point p ∈ MS a vector V (p) in Tp M . In particular V : M → p∈M Tp M . The only
problem is that p∈M Tp M doesn’t have a smooth structure,Sso we cannot talk of the
smoothness of V . The goal of this section is to endow T M = p∈M Tp M with a natural
smooth structure.
First we define π : T M → M by π(v) = p if v ∈ Tp M , in other words π(v) is the foot
of v. Remark that a vector field V : M → T M should satisfy V ◦ π = idM (V (p) should
be a vector at p, not at some other point).

1
any compact K ⊂ M intersects a finite number of Oi ’s.

38
S
Now let’s define charts on T M . Consider a chart (U, ϕU ) of M . Let T U = p∈U Tp M ,
and define ΦU : T U → ϕ(U ) × Rn by:
ΦU (ξp ) = (ϕ(p), Tp ϕ(ξ)) if ξp ∈ Tp M .
We saw in the previous chapter that Tp ϕ : Tp M → Rn is a linear isomorphism when
(U, ϕU ) is a smooth chart. This shows that ΦU : T U → ϕ(U ) × Rn is a bijection.
We haven’t defined a topology on T M yet. We will do it in the following way: a set
O ⊂ T M is said to be open if for any chart (U, ϕU ) of M , Φ(T U ∩ O ⊂ ϕ(U ) × Rn is
open.
Proposition 2.3.1. The statement above defines a topology on T M for which the col-
lection of all (T U, ΦU ) is a smooth atlas.
Proof. The fact that weS have definedSa topology on T M is easy : if Oi is a family
n
of open
sets in T M , then ΦUS( i Oi ∩ T U ) = i ΦU (Oi ∩ T U ), which is open in U × R as a union
of open sets. Thus i Oi is open. The same argument works for finite interesections.
We now show that ΦU : T U → U × Rn is an homeomorphism. Since we already know
that ΦU is a bijection, this is equivalent to the following claim:
Claim 2.3.2. For any O ⊂ T U , then O is open in T M if and only if ΦU (O) is open in
U × Rn .
To see this, consider another chart (V, ϕV ) of M , and the transition function ΦU V =
ΦV ◦ Φ−1 n n
U from ϕU (U ∩ V ) × R to ϕV (U ∩ V ) × R . We have :
 
ΦU V (x, v) = ΦV (Tϕ−1 (x) ϕU )−1 (v)
U

= ΦV Tx (ϕ−1

U )(v)
 
= ϕV (ϕ−1U (x)), T −1 ϕ
ϕ (x) V Tx (ϕ−1
U )(v)
U

= (ϕU V (x), Tx ϕU V (v)) .


Since ϕU V is smooth, (x, v) 7→ Tx ϕU V (v) = Dx ϕU V (v) is smooth and ΦU V is smooth.
Moreover, the fact that ΦU V ◦ ΦV U = id implies, since ΦV U is also smooth, that ΦU V
is a diffeomorphism. In particular ΦU V is a homeomorphism. Thus if ΦU (O) is open,
ΦV (O ∩ T V ) = ΦU V (ΦU (O) ∩ (ϕU (V ∩ U ) × Rn )) is also open. The claim is proved.
Moreover, we have already seen that the transition functions ΦU V are diffeomorphism.
Thus T M is a smooth manifold.
Remark 2.3.3. The dimension of T M is twice the dimension of M .
Remark 2.3.4. The formula for the transition functions shows that if M is a C k manifold,
then T M is a C k−1 manifold.
Proposition 2.3.5. π : T M → M is a submersion.
Proof. Let (U, ΦU ) be a chart if M and (T U, ΦU ) be the associated chart of T M . Then:
ϕU ◦ π ◦ Φ−1
U (x, v) = x

which is a submersion from R2n → Rn .

39
2.3.2 T M as a vector bundle
The charts ΦU : T U → U × Rn are more than just diffeomorphisms, they also behave
well with respect to the vector space structure of Tp M . This can be summarized in the
following proposition, which is a direct consequence of the definition of ΦU .

Proposition 2.3.6. π : T M → M is the natural projection. We have the following


properties :

• for any p ∈ M , π −1 (p) is a d-dimensional vector space.

• for any p ∈ M , there exists a neighborhood U of p such and a diffeomorphism


ΦU : π −1 (U ) → U × Rn such that the following diagram is commutative:

ΦU
π −1 (U ) U × Rn

π
π1
U

where π1 is the projection (p, v) 7→ p.

• for each p ∈ U , ΦU restricted to Tp M = π −1 (p) is a linear isomorphism onto


{p} × Rn .

From this situation we extract the definition of a vector bundle on a manifold, which
is a family of vector spaces smoothly attached above each point of a manifold M .

Definition 2.3.7. Let E and M be two smooth manifolds. A smooth d-dimensional


vector bundle above M is a smooth map π : E → M such that:

• for any p ∈ M , Ep = π −1 (p) is a d-dimensional vector space.

• for any p ∈ M , there exists a neighborhood U of p such and a diffeomorphism


ΦU : π −1 (U ) → U × Rd such that the following diagram is commutative:

ΦU
π −1 (U ) U × Rd

π
π1
U

where π1 is the projection (p, v) 7→ p.

• for each p ∈ U , ΦU restricted to Ep M = π −1 (p) is a linear isomorphism onto


{p} × Rd .

40
Remark 2.3.8. The dimension of E is d + dim M and π is a submersion.
Example 2.3.9. T M → M is an dim M -dimensional vector bundle over M .
Example 2.3.10. M × Rd → M is a vector bundle, a vector bundle is said to be trivial
if it is isomorphic to it.
Proposition 2.3.11. Consider a smooth f : M → N . Define T f : T M → T N by
T f (ξ) = Tp f (ξ) if ξ ∈ Tp M . Then :
1. T f is smooth.

2. T f makes the following diagram commutative:

Tf
TM TN

πM πN

M N
f

3. for each p ∈ M , the restriction of T f to Tp M is linear from Tp M to Tf (p) N .


These properties say that T f is a vector bundle morphism.
Definition 2.3.12. Let π1 : E1 → M1 and π2 : E2 → M2 be two smooth vector bundle.
A vector bundle morphism between E1 and E2 is a couple (f, g) of smooth maps
f : M1 → M2 and g : E1 → E2 such that :
1. f and g make the following diagram commute:
g
E1 E2

π1 π2

M1 M2
f

2. for each p ∈ M1 , the restriction of g to (E1 )p = π1−1 ({p}) is linear from (E1 )p to
(E2 )f (p) = π2−1 ({f (p)}).
Vector bundle morphism can be composed. A vector bundle isomorphism from E1
to E2 which has an inverse, it is equivalent to require g to be diffeomorphism. A d-
dimensional vector bundle E → M is said to be trivial if it is isomorphic to M ×Rd → M .
Definition 2.3.13. A smooth section of a vector bundle π : E → M is a map
σ : M → E such that π ◦ σ = idE . A section of the tangent bundle T M of M is called a
vector field on M .

41
Remark 2.3.14. The set Γ(E) of smooth sections of E is a vector space. On can show
using bump functions on M that it has infinite dimension. In other words, vector
bundles on smooth manifolds have a lot of sections. This isn’t the case for complex
manifolds: finding holomorphic sections of complex vector bundles is in general a
hard task.
With this definition a smooth vector field on M is a smooth section of T M → M .
The following charcterization of trivial bundles is useful:

Proposition 2.3.15. An n-dimensional vector bundle π : E → M is trivial if and only


if there exist n sections σ1 , · · · , σn such that at each p ∈ M , (σ1 (p), · · · , σn (p)) is a basis
of Ep .

Proof. Assume first that E → M is trivial, then we have an isomorphism (f, g) from
E → M to Ẽ = M × Rn → M . Let us define smooth sections σ̃i of Ē by σ̃i (p̃) = (p̃, ei )
where ei is the canonical basis of Rn , and p̃ ∈ M . We then define σi (p) = g −1 (σ̃i (f (p))),
these are smooth sections of E. Since g|Ep is a linear isomorphism, (σi (p))i=1,··· ,n is a
basis of Ep .
For the converse, let (σ1 , · · · , σn ) be n linearly independent sections E and
P define a
i
bundle morphism (f, g) from Ē → M to E → M by f = idM and g(p, x) = i x σi (p).
Then (f, g) is the required isomorphism.
Remark 2.3.16. In particular, if a vector bundle is trivial, it has a nowhere vanishing
section. This can be used to show that the tangent bundle to S2 is not trivial, since
any vector field on S2 has a zero (the so-called hairy ball theorem), a theorem we
will prove later using differential forms and Stokes theorem.

2.4 Vector fields on manifolds


All the facts recalled in section 2.1 extends to manifolds. Namely:

Definition 2.4.1. Let M be a smooth manifold and V be a smooth vector field on M .


A smooth curve x : I → M is said to be a solution to the Cauchy problem:
(
x0 = V (x)
x(0) = x0 .

if it satisfies x(0) = x0 and x0 (t) = V (x(t)) ∈ Tx(t) M for any t ∈ I.

Theorem 2.4.2. The Cauchy problem:


(
x0 = V (x)
x(0) = x0 .

has a unique maximal solution x : I → R.

42
Definition 2.4.3. A vector field V is said to be complete if for any x0 ∈ M the maximal
solution to the Cauchy problem with initial condition x0 is defined for I = R.

Proposition 2.4.4. If a vector field on a manifold has compact support, then it is


complete.
If M is compact, then any vector field on M is complete.

If V is a smooth complete vector field on M , we can define the ϕt : M → M which


will be a one parameter group of diffeomorphisms of M . If V is not complete, then the
flow (t, x) 7→ ϕt (x) is only defined on

The maximal solution to x0 = V (x)
 
O = (y, t) ∈ M × R .
with x(0) = y is defined at t.

Definition 2.4.5. Let f : M → N be a diffeomorphism, we define the push forward


of V by f (denoted by f∗ V ) by f∗ V (y) = Tf −1 (y) f (Vf −1 (y) ), this is a smooth vector field
on N .

Let u be a smooth function on N , then, as a derivation on C ∞ (N ), acts by:

f∗ V · u = (V · (u ◦ f )) ◦ f −1 .

Proposition 2.4.6. x : I → M is an integral curve of V if and only if f ◦ x : I → N


is an integral curve of f∗ V .

Proof. Set x̃ = f ◦ x then:


x̃0 (t) = Tx(t) f (x0 (t)) .
Thus x̃ is an integral curve of f∗ V if an only if

Tx(t) f (x0 (t)) = f∗ Vx̃(t) = Tf −1 (x̃(t)) f Vf −1 (x̃(t))




which is equivalent to x0 (t) = Vx(t) .


A useful observation that follows from this is that, if we denote by ϕt the flow of V
and ψt the flow of f∗ V , then ψt = f ◦ ϕt ◦ f −1 . This property actually characterizes f∗ V .

Proposition 2.4.7. Assume that V is a complete vector filed on M . Let ϕs be the flow
of V , then (ϕs )∗ V = V .

Proof. Let p ∈ M , then: ((ϕt̃ )∗ V )p = D(ϕs )−1 (p) V(ϕs )−1 (p) .. Consider the solution x to the
Cauchy problem x0 (t) = V (t) with initial condition x(0) = (ϕs )−1 (p). Then by definition
of the flow x(s) = p and x0 (s) = Vp . Moreover x̃(t) = ϕs (x(t)) = x(s + t) is an integral
curve of V which satisfies x̃(0) = p. Then

Vp = x̃0 (0) = Dx(0) ϕs (x0 (0)) = D(ϕs )−1 (p) ϕs V(ϕs )−1 (p) = ((ϕs )∗ V )p .

43
An equilibrium (or stationary) point of a vector field V is a point p ∈ M such that
Vp = 0. Let us consider the constant vector field ∂1 in Rn . The following theorem tells
us that around a non-stationnary point, any vector field looks like ∂1 .
Theorem 2.4.8. Let V be a vector field on M and p ∈ M such that V (p) 6= 0. Then
there exists a neighborhood U of p and a diffeomorphism ψ : U → Ũ ⊂ Rn such that
ψ∗ V = ∂1 .
Proof. Since ce we are looking for a local result, we can only consider V in a coordinate
chart around p. This way we can assume that V is complete Consider a chart (U, xU =
(x1 , · · · , xn )) around p (we assume that xU (p) = 0) and the flow Φt of V . For t small,
we can assume that Φt is a diffeomorphism from V ⊂ U to Φt (V ) ⊂ U .
Define a map ρ on a neighborhood of p by ρ(x1 , · · · , xn ) = ϕx1 (0, x2 , · · · , xn ).
The differential of ρ at p is such that Tp ρ(∂1 ) = V (p) and Tp ρ(∂i ) = ∂i for i ≥ 2. By
permuting coordinates if necessary, we can assume the the ∂1 component of V (p) is not
0. This implies that Tp ρ is invertible, thus ρ is a diffeomorphism on some neighborhood
of p.
We now set ψ = ρ−1 . Now the flow ϕ̃t of ψ∗ V is given by ψ ◦ ϕt ◦ ψ −1 and we can
show that :

ϕ̃t (x1 , · · · , xn ) = ψ ◦ ϕt ◦ ρ(x1 , · · · , xn )


= ψ ϕt (ϕx1 (0, x2 , · · · , xn ))


= ψ ϕt+x1 (0, x2 , · · · , xn )


= (t + x1 , x2 , · · · , xn ),

which is the flow of ∂1 , since the flow determines the vector filed, this shows that ψ∗ V =
∂1 .

2.5 Vector fields and global derivations


We have mentionned in the first chapter that tangent vectors at a point p ∈ M can be
viewed as derivations at p. Here we will prove this result and a global version of it :
vector fields on M can be viewed as derivations on M .
Let’s first recall the definition of a derivation at p ∈ M
Definition 2.5.1. A linear map δ : C ∞ (M ) → R is a derivation at p ∈ M if:

δ(f g) = δ(f )g(p) + f (p)δ(g).

The set of derivation ot p is denoted by Dp .


Proposition 2.5.2. A derivation at p satisfies the following properties:
1. If f is constant, then δ(f ) = 0.

2. If f and g coincide on a neighborhood of p, then δ(f ) = δ(g).

44
Proof. We first prove the first statement. By linearity, we can assume that f constant
equal to 1, thus f 2 = f . This implies :

δ(f ) = δ(f × f ) = δ(f )f (p) + f (p)δ(f ) = 2δ(f ),

and δ(f ) = 0.
For the second statement, using linearity it is enough to show that that δ(f − g) = 0.
set h = f − g and notice that h vanishes on a neighborhood U of p. Consider a compact
neighborhood K of p which is included in U , and a bump function η which is equal to
1 on K and equal to 0 outside of U . This implies to ηh = 0 on M . Thus:

0 = δ(ηh) = δ(η)h(p) + η(p)δ(h) = δ(h).

This proposition shows that derivations are local, if U is a neighborhood of p, we can


define δ for smooth functions f on U by setting δU (f ) = δ(f˜) where f˜ = ηf on U with
η a bump function like in the previous proof, and extended to M by 0.
We can now prove:

Theorem 2.5.3. The map L : Xp 7→ (f 7→ Xp · f )) from Tp (M ) to Dp is a linear


isomorphism.

Proof. Let (U, ϕU ) be a chart around p, we assume that ϕU (p) = 0 ∈ U . Consider the
coordinate functions xi : U → R.
Let f : U → R be a smooth function. We make the following claim:
Claim 2.5.4. There exists smooth function hi : U → R such that f (x) = f (p) +
i
P
i x (x)hi (x), moreover hi (p) = ∂i f (p).
For the proof of the claim, we work on ϕU (U ) ⊂ Rn . Using bump function, it is
enough to prove the claim in a convex neighborhood of 0. We thus have:
Z 1 Z 1X X Z 1
d i
f (x) = f (0) + (f (tx))dt = f (0) + x ∂i f (tx)dt = f (0) + xi ∂i f (tx)dt.
0 dt 0 i i 0

R1
Setting hi (x) = 0 xi ∂i f (tx)dt does the job. The claim is proved.
Now consider δ a derivation at p acting on f , we have:
X  X
δ(f ) = δ(xi )hi (p) + xi (p)δ(hi ) = δ(xi )∂i f (p).
i i

And if we define V = δ(xi )∂i , we have that δ = L(V ). Thus L is surjective.


To show that L is injective, assume that L(V ) = 0, then we have that V · xi = 0, since
V = (V · xi )∂i , this implies that V = 0.
We are now going to globalize this construction:

45
Definition 2.5.5. A (global) derivation on M is a linear map δ : C ∞ (M ) → C ∞ (M )
such that:
δ(f g) = δ(f )g + f δ(g).
We denote by D(M ) the space of derivations at M .

Example 2.5.6. Let V be a smooth vector field on M , then we define X · f by:

(X · f )(p) = Tp f (Xp ).

f 7→ X · f is then a derivation on M
As in the case of pointwise derivations, Leibnitz implies the locality of derivations.

Proposition 2.5.7. Let δ be a derivation on M .

1. If f = g on some open set U ⊂ M , then (δf )|U = (δg)|U .

2. For any open set U ⊂ M , there exist a unique derivation δ|U such that, for any
f ∈ C ∞ (M ), δ|U (f |U ) = (δf )|U .

Proof. For the first point, linearity implies that we just need to to show that if f |U = 0,
then (δf )|U = 0. Let V ⊂ U be any open subset whose closure is contained in U .
Consider a bump function η which is 1 on V and 0 outside U . Then f η = 0 on M and:

0 = δ(ηf ) = δ(η)f + ηδ(f ),

moreover δ(ηf )|V = δ(η)|V f |V + η|V δ(f )|V = δ(f )|V , thus δ(f ) is 0 on any open set of
U whose closure is contained in V , hence (δf )|U = 0.
For the second point, the only point is to define δ|U for smooth functions f ∈ C ∞ (U )
who don’t extend to smooth functions on the whole of M . To do this we consider V and
η as above and set δ|U (f )|V = δ(ηf ), where ηf is extended by 0 to M . The previous
statement shows that this definition doesn’t depend on the choice of η.

Theorem 2.5.8. The map L : V 7→ (f 7→ V · f ) from Γ(T M ) to D(M ) is a linear


isomorphism.

Proof. First L is injective. If L(V ) is not zero, there is a function f such that V · f is
not zero. If p ∈ M is a point where (V · f ) 6= 0, then Tp f (Vp ) 6= 0, in particular Vp is
not 0 and V 6= 0.
Let us now show that L is surjective. Let δ be a derivation. For any p ∈ M , f 7→
(δf )(p) is a derivation at p, so there exists Vp ∈ Tp M such that (δf )(p) = Vp · f . We
now only need to show that p 7→ Vp is a smooth section. This follows from the fact that,
in any chart (U, ϕU ): X
V (p) = δ(xi )∂i
i
i
and δ(x ) is smooth.

46
2.6 Lie bracket of vector fields
Consider two vector fields V and W on M , we then have the corresponding derivations
f 7→ V · f and f 7→ W · f .
Proposition 2.6.1. The map L : f 7→ V · (W · f ) − W · (V · f ) from C ∞ (M ) to C ∞ (M )
is a derivation.
Proof. The fact that L is linear is obvious, we just need to show it satisfies Leibnitz rule:
L(f g) = V · (W · (f g)) − W · (V · (f g))
= V · ((W · f )g + f (W · g)) − W · ((V · f )g + f (V · g))
= (V · (W · f ))g + (W · f )(V · g) + (V · f )(W · g) + f (V · (W · g))
− (W · (V · f ))g − (V · f )(W · g) − (W · f )(V · g) − f (W · (V · g))
= (Lf )g − f (Lg).

Since L is a derivation, there exist a vector field X on M such that Lf = X · f for


any f . X is called the Lie bracket of V and W and is usually denoted by [V, W ]. It
has the following properties:
Proposition 2.6.2. • (V, W ) 7→ [V, W ] is bilinear.
• [V, W ](p) only depens on the value of V and W on a neighborhood of p
• [V, W ] = −[W, V ].
• [V, [W, X]] + [X, [V, W ]] + [W, [X, V ]] = 0.
• if ϕ : M → N is a diffeomorphism, [ϕ∗ V, ϕ∗ W ] = ϕ∗ [V, W ].
Example 2.6.3. For any coordinate chart (U, ϕU ), the coordinate vector fields satisfy
[∂i , ∂j ] = 0, since mixed partial derivatives commute.
Let’s work out a formula for computing Lie brackets in coordinates: we can assume
that V = V i ∂i and W = W i ∂i in some coordinate charts.
[V, W ] · f = V · (W · f ) − W · (V · f )
= V · (W i ∂i f ) − W · (V i ∂i f )
= (V · W i )∂i f + W i (V · ∂i f ) − (W · V i )∂i f − V i (W · ∂i f )
= (V j ∂j W i )∂i f + W i V j ∂j ∂i f − (W j ∂j V i )∂i f − V i W j ∂i ∂j f
= (V j ∂j W i − W j ∂j V i )∂i f
thus we get
[V, W ] = (V j ∂j W i − W j ∂j V i )∂i .
The following proposition gives an interpretation of the Lie bracket in term of flow,
informally, it shows that [W, V ] is the infinitesimal push forward of W along the flow if
V.

47
Proposition 2.6.4. Let V and W be vector fields on M , and ϕt be the flow of W . Then:
d
[W, V ] = (ϕt )∗ W.
dt |t=0
Remark 2.6.5. There is something unclear here. If we fix p ∈ M , then t 7→ ((ϕt )∗ W )p
is a map from R → T M , therefore its derivatve should be an element of T (T M ),
the double tangent bundle. However we claim that it is [W, V ]p , an element of Tp M .
What happens here is that (ϕt )∗ W is a time dependent vector field, and since vector
fields on M form a vector space we can define :
d (ϕt )∗ W − W
(ϕt )∗ W = lim .
dt |t=0 t→0 t
Proof. We work with derivations. We have that ((ϕt )∗ W ) · f = (W · (f ◦ ϕt )) ◦ ϕ−t .
Consider the one parameter family of smooth functions g(x, t) = f ◦ϕt (x)−f (x). Arguing
as in the proof of Claim 2.5.4, since g(x, 0) = 0, one can find a one parameter family of
smooth functions h(x, t) such that g(x, t) = th(x, t) and h(x, 0) = dtd f (x, 0) = V · f .
Then :

((ϕt )∗ W ) · f = (W · (f + th(., t))) ◦ ϕ−t


= (W · f ) ◦ ϕ−t + t(W · h(., t)) ◦ ϕ−t

We now differentiate at t = 0. The first term gives −V · (W · f ) and the second gives
W · (h(., 0)) = W · (V · f ).
The Lie bracket has the following interpretation: it measures the lack of commutativity
of the flows of V and W .
Proposition 2.6.6. Let V and W be two vector fields on M and denote by ϕt and ψt
their flows. Then ψt ◦ ϕs = ϕs ◦ ψt if and only if [V, W ] = 0.
Proof. We first assume that the flows commute. We have that ψt = ϕs ◦ψt ◦(ϕs )−1 which
implies that (ϕs )∗ W = W , taking the derivative at s = 0 we get, using the previous
proposition, that [V, W ] = 0.
We now assume that [V, W ] = 0. We only need to show that (ϕs )∗ W = W . To see it
we compute, using that (ϕs )∗ V = V :

0 = [V, W ] = [(ϕs )∗ V, (ϕs )∗ W ] = [V, (ϕs )∗ W ].

Thus:
d d d
0 = [V, (ϕs )∗ W ] = (ϕt )∗ (ϕs )∗ W = (ϕt+s )∗ W = (ϕs )∗ W.
dt |t=0 dt |t=0 ds
And, since ϕ0 = idM , (ϕ0 )∗ W = 0. With the previous equality this implies that
(ϕs )∗ W = 0.

48
2.7 Exercises
Exercise 2.1. Consider polar coordinates (r, θ) defined by x = r cos θ, y = r sin θ for
(x, y) ∈ R2 \{0}. Write the vector fields ∂r and ∂θ in term of ∂x and ∂y .
Exercise 2.2. Let ψ : (R, +) → (GLn (R), ×) be a continuous group morphism.

1. Show that there exists t0 > 0 and A0 ∈ Mn (R) such that ψ(t0 ) = exp(A0 ). Hint:
use that exp : Mn (R) → GLn (R) is a diffeomorphism on a neighborhood of 0.

2. Show that for any r ∈ Q, ψ(rt0 ) = erA0 .

3. Show that there exist a unique A ∈ Mn (R) such that ψ(t) = etA .

Exercise 2.3. Let (x, t) 7→ ϕt (x) be a smooth map from O × R → O such that ϕt+s =
ϕt ◦ ϕs . Show that ϕt is the flow of a vector field on O. Apply this result to
  
cos t sin t x
ϕt (x, y) = .
− sin t cos t y

Exercise 2.4. Let V : Rn → Rn be a vector field. Assume that there is a constant


C > 0 such that | hx, V (x)i | ≤ Ckxk2 . Show that V is complete.
Exercise 2.5. * Consider the vector field V (x, y) = (ax−bxy, −cy +dxy), for a, b, c, d >
0 on U = {(x, y)|x > 0, y > 0}. The differential equation (x0 , y 0 ) = V (x, y) is called
the Lotka-Volterra equation, it models the evolution of an two interacting populations
of preys (x(t)) and predators (y(t)).

1. Consider the function: f : U → R given by f (x, y) = dx − c log x + by − d log y.


Show that f is proper.

2. Show that f (x(t), y(t)) is constant when (x(t), y(t)) is an integral curve of V .

3. Show that V is complete.

4. Show that there exists a unique (x∗ , y∗ ) ∈ U such that V (x∗ , y ∗ ) = 0.

5. Show that (x∗ , y∗ ) is the unique maximum of f .

6. Show that (x∗ , y∗ ) is a stable equilibrium of V .


This means that for any neighborhood V of (x∗ , y∗ ), there is a neighborhood V 0
of (x∗ , y∗ ) such that for any integral curve (x(t), y(t)) satifying (x(0), y(0)) ∈ V 0 ,
(x(t), y(t)) ∈ V for t ≥ 0.

Exercise 2.6. ** Consider f : Rn → Rn a smooth map. We will show that if f is a


local diffeomorphism and is proper, then f is a (global) diffeomorphism. We assume
that f is a proper local diffeomorphism.

1. Show that f (Rn ) ⊂ Rn is open and closed. Use it to prove that f is surjective.

49
2. Show that for any y ∈ Rn , f −1 ({y}) is finite.
We want to show that f −1 ({y}) is in fact a single point (to see that f is injective).
Replacing f by f − y, we see that it is enough to show that f −1 ({0}) is a single
point, the next questions give a proof of this fact.

3. Consider the vector field V (x) = (−Dx f )−1 f (x), and denote by ϕt his flow. Show
that f (ϕt (x)) = e−t f (x) and use it to prove that V is complete.

4. Let f −1 ({0}) = {x1 , · · · , xk } denote the equilibrium points of V . Show that each
xi is an asymptotically stable equilibrium point.
This means that there exist a neighborhood V of xi such any integral curve x(t)
which satisfies x(0) ∈ V satisfies limt→∞ x(t) = xi .

Let Oi = {x0 ∈ Rn | limt→∞ ϕt (x) = xi }. Show that the Oi are disjoint and that
5. S
n
i Oi = R .

6. Show, using the smoothness of ϕt , that Oi is open. Conclude that f −1 ({0}) is


reduced to a single point.

Exercise 2.7. Use the Stone-Weierstrass theorem and bump functions to show that
that on a compact manifold M , C ∞ (M ) is dense in C 0 (M ) (equipped with the sup
norm).
Exercise 2.8. * Let M n be a compact smooth manifold, (Ui , ϕi )i=1,...N be a finite atlas
of M , and for each i let Vi be relatively compact open subset of Ui such that the Vi ’s
cover M . Let ηi be be smooth function with support in Ui which is equal to 1 on Vi .
Let fi : M → Rn be defined by fi (x) = ηi (x)ϕi (x) for x ∈ Ui and fi (x) = 0 for x ∈
/ Ui .
N (n+1)
Define F : M → R by:

F (x) = (f1 , . . . , fN , η1 , . . . , ηN ).

Show that F is a smooth embedding.


Using more elaborated tools, one can show that M embeds in R2n .
Exercise 2.9. ** Let M be a smooth manifold. Using the flow of compactly supported
vector fields, show that the group of diffeomorphism of M acts transitively on M .
First consider the case of balls in Rn , then use a covering of M by open sets diffeo-
morphic to open balls to globalize.
Exercise 2.10. 1. Let π1 : E1 → M1 and π2 : E2 → M2 be two vector bundles.
Show that π1 × π2 : E1 × E2 → M1 × M2 is a smooth vector bundle.

2. Let π : E → M be a smooth vector bundle and f : N → M be a smooth map.


Set:
f ∗ E = {(p, v) ∈ N × E|f (p) = π(v)}
and letπf ∗ E be the restriction to f ∗ E of the first projection N × E → N . Show
that πf ∗ E : f ∗ E → N is a smooth vector bundle.

50
The construction in the first question is called the product of E1 and E2 , the construc-
tion in the second question is the pullback of E along f . An important example is when
π1 : E1 → M and π2 : E2 → M are two vector bundles over the same base M . On can
the consider the diagonal map ∆ : M → M × M , and the bundle ∆∗ (E1 × E2 ) is called
the direct sum of E1 and E2 . As a set it can be desribed as:

E1 ⊕ E2 = {(p, v1 , v2 ) ∈ M × E1 × E2 |π1 (v1 ) = π2 (v2 ) = p}.

Exercise 2.11. Let S k ⊂ Rn be a smooth submanifold. The normal space at p ∈ S is


defined by:
Np S = {v ∈ Rn |v ⊥ Tp S}.

1. Show that the normal bundle N S = {(p, v)|p ∈ S, v ∈ Np S} is a vector bundle of


rank n − k.

2. Show that N Sn is a trivial vector bundle.

3. Show that N S ⊕ T S is a trivial vector bundle.

Exercise 2.12. Consider the skew field of quaternions H = {x+iy +jz +kt|(x, y, z, t) ∈
R4 }. Recall that ij = −ji = k, jk = −kj = l and ki = −ik = j. The modulus of
q = x + iy + jz + kt is defined by |q| = x2 + y 2 + z 2 + t2 and |qq 0 | = |q||q 0 |. Consider
S3 = {q ∈ H||q| = 1}.

1. Show that z ∈ S3 7→ iz defines a smooth nowhere vanishing vector field on S3 .

2. Show that T S3 is trivial.

Exercise 2.13. Let A, B ∈ Mn (R) and consider the two vector fields on Rn defined
by VA (x) = Ax and VB (x) = Bx. Compute their Lie bracket and relate it to the
commutator of A and B.
Exercise 2.14. Let V (x, y) = (V x (x, y), V y (x, y)) be a vector field on R2 . Show that
the flow of V commutes with rotations around the origin if and only if:
(
V x + y∂x V y = 0
V y − x∂y V x = 0.

Hint: find a vector field on R2 whose flow acts by rotations around the origin.

51
3 Lie groups and homogenous
spaces
3.1 Definitions
Definition 3.1.1. A Lie group is a group G with a smooth manifold structure such
that the group law (g, h) 7→ gh from G × G to G and the inverse g 7→ g −1 from G to G
are smooth.

Remark 3.1.2. One can also define topological groups, by requiring G to be equipped
with a topology such that the group law and the inverse are continuous.

3.2 Examples
3.2.1 Abelian examples
The group (R, +) is a Lie group. More generally any finite dimensional vector space E
with the group structure given by vector addition is a Lie group.
The circle S1 is a Lie group, there are two different ways to see its group structure.
The first one is to see S 1 as the quotient space R/Z, since Z is a normal subgroup of
R, R/Z has a natural group structure. The smoothness of the group structure can be
shown using Theorem 1.4.15.
Another way to go is to see S1 as the group of modulus 1 complex numbers. The
group law on S1 is then the restriction of the usual product C × C → C, and hence is
smooth. The same holds for inversion.
Using products, the torus Tn of dimension n is then a Lie group. All these examples
are abelian.

3.2.2 The 3-sphere as a Lie group


Consider S3 as the unit sphere in R4 and identify R4 with the skew field of quaternions
H. Then S3 = {q ∈ H| |q| = 1}. Moreover, since |qq 0 | = |q||q 0 |, S3 is stable under
quaternion multiplication. The inverse of a quaternion q is given by q̄/|q|2 , so S3 is also
stable by inversion. So S3 is a subgroup of H∗ , since the product and inverse are smooth
on H∗ , they are smooth when restricted to the smooth submanifold S3 . Hence S3 is a
Lie group.

52
3.2.3 Matrix groups
Gln (R) is at the same time a smooth manifold (as an open subset of Mn (R)) and a group.
It is a classical that matrix product and inversion are smooth maps. Thus Gln (R) is a
Lie group. The same holds for Gln (C).
Assume G ⊂ Gln (R) is at the same time a subgroup and a submanifold, then G is
automatically a Lie group (group law and inverse are smooth as restrictions of smooth
maps). We have already seen examples Sln (R), O(n, R) and SO(n, R). We can similarly
show that Sln (C), U (n, C) and SU (n, C) are Lie groups.

3.2.4 Transformation groups


When one works with a geometric structure on some set X, it is often useful to consider
the group G of bijections of X which preserve this structure. In a lot of cases, this turn
out to be a Lie group.
Consider Rn as an affine space, any affine bijection f : Rn → Rn can be uniquely
written as f (x) = Ax + b for some A ∈ Gln (R) and some b ∈ Rn . These form a group,
the affine group Af f (Rn ) which can also be characterized as the group of continuous
bijections of Rn which maps triplets of aligned points to triplets of aligned points. As a
set Af f (Rn ) is just Gln (R) × Rn , which gives it a smooth structure. On can check that
multiplication and inversion of affine maps are smooth maps, and this gives Af f (Rn )
the structure of a Lie group.
We have defined homographies of RPn in chapter one. They are diffeomorphisms of
RPn given by fA ([x]) = [Ax] for some A ∈ Gln+1 (R). It is easy to see that A 7→ fA define
a group morphism ϕ from Gln+1 (R) to the group of diffeomorphisms of RPn . The kernel
of this morphism is the normal subgroup H = R∗ I where I is the identity. The projective
group P Gln (R) (or group of homographies) is the image of ϕ. It is isomorphic to
Gln+1 (R)/H. Arguing as in the construction of RPn in chapter 1, one can can show that
that there is a unique smooth structure on P Gln (R) such that ϕ : Gln+1 (R) → P Gln (R)
is a smooth submersion. Once this is done it is easy to show that multiplication and
inversion in P Gln (R) are smooth. Thus P Gln (R) is a Lie group.

3.3 Lie morphisms, Lie subgroups


Definition 3.3.1. Let G and H be two Lie groups. f : G → H is a Lie group
morphism if it is smooth and a group morphism.

Example 3.3.2. The standard submersion π : Rn → Tn is a Lie group morphism.


Example 3.3.3. The determinant is a morphism from any matrix group to the Lie group
(R, +).
Example 3.3.4. To each affine transformation L(x) = Ax + b one can associate its linear
part A, this defines a Lie morphism from Af f (Rn ) to Gln (R).

53
Definition 3.3.5. H ⊂ G is a Lie subgroup if it is at the same time a subgroup and
a submanifold of G.
Remark 3.3.6. This definition is not universal, some author define a Lie subgroup to be
an injective immersion of H into G which is a group morphism. When the image of
this immersion is moreover a submanifold, H is said to be a proper Lie subgroup.
For instance matrix groups are Lie subgroups of Gln (R).
The kernel of a Lie group morphism is always a Lie subgroup, this follows from the
fact that a closed subgroup of a Lie group is a Lie subgroup: this is a hard theorem that
we won’t prove here (a proof can be found in [Lee02]). However this is not always true
for the image.
Example 3.3.7. Consider the Lie group morphism f : R → R2 given by f (t) = (t, αt)
where α is some real number. Consider F = π ◦ f , where π is the projection from
R2 to the torus T2 , it is a Lie group morphism from R → T2 .
If α is rational equal to p/q, then F (t) = 0 if and only if t ∈ qZ. This implies that
F pass through the quotient as a smooth group morphism F̃ from R/qZ to T2 , which
is moreover an injective immersion. So F (R) = F̃ (R/qZ) is a submanifold of T2 . So
F (R) is a Lie subgroup of T2 which is moreover isomorphic to S1 as a Lie group.
Assume now that α ∈ / Q. Then it is a classical fact that the set {bkαc|k ∈ Z} is
dense in [0, 1]. This can be used to show that F (R) is in fact dense in T2 , thus if
F (R) cannot be a submanifold of dimension 1, if it was a submanifold of dimension
2, it would be an open set, hence it would be the whole of T2 since it is dense. So
F (R) is ot a submanifold of T2 and hence not a Lie subgroup.

3.4 Left invariant vector fields


An important feature of Lie groups is that they have distiguished diffeomorphisms. We
set:
Lg : G → G Rg : G→G
h 7→ gh h 7→ hg,
these are diffeomorphisms (since Lg ◦ Lg−1 = Lg−1 ◦ Lg = idG ) called left and right
translations. The associativity of G translate into the fact that Lg ◦ Rh = Rh ◦ Lg .
Another important diffeomorphism is the inversion I : G → G.
Let X be a vector field on G.
Definition 3.4.1. X is said to be left invariant (resp. right invariant) if for any
g ∈ G, (Lg )∗ X = X (resp. (Rg )∗ X = X).
We will develop the theory of left invariant vector fields below, the case of right
invariant vector fields is parallel and we will omit it.
The set of left invariant vector fields is in fact linear subspace of the space of vector
fields on G. A left invariant vector field is determined by its value at the neutral element
e ∈ G.

54
Proposition 3.4.2. The map X 7→ Xe from the set of left invariant vector fields on G
to Te G is a linear isomorphism.
Proof. If X is left invariant, then by definition of the push forward Xg = Te Lg (Xe ) for
any g ∈ G. So if Xe = 0, X = 0 and X 7→ Xe is injective.
Let us now fix v ∈ Te G. We set Xg = Te Lg (v), then:
(Lh )∗ Xg = Th−1 g Lh ◦ Te Lh−1 g (v) = Te (Lh ◦ Lh−1 g )v = Te Lg (v).
So we only need to show that this define a smooth vector field. In order to do this we
will show it defines a derivation on C ∞ (G). Let f be a smooth function, we need to
show that g 7→ Tg f (Xg ) is smooth. Consider a smooth curve c(t) such that c(0) = e and
c0 (0) = v. Set F (t, g) = f (gc(t)) = f (Lg (c(t))), then g 7→ (t 7→ F (g, t))0 (0) is smooth.
To conclude remark that:
d
|t=0 F (g, t) = Tg f (Te Lg (v)) = Tg f (Xg ).
dt

Example 3.4.3. Let us consider the case of G = GLn (R). Then Lg (h) = gh is linear in
h, so Te Lg = Lg . And if Xe ∈ Te G ' Mn (R), then the Left invariant extension X of
Xe will be given by Xg = gXe .
This in particular implies that T G is a trivial bundle: pick a basis (e1 , . . . , en ) of Te G
and consider the left invariant vector fields E i such that Eei = ei , these are vector fields
which form a basis of Tg G at each g.
Proposition 3.4.4. 1. If X and Y are left invariant, so is [X, Y ].
2. If X is left invariant, then I∗ X is right invariant.
3. If X is left invariant, so is (Rg )∗ X.
Proof. The first point comes from the fact that (Lg )∗ [X, Y ] = [(Lg )∗ X, (Lg )∗ Y ] = [X, Y ].
For the second, notice that, since (gx)−1 = x−1 g −1 , I ◦ Lg = Rg−1 ◦ I. So, if X is left
invariant:
(Rg )∗ I∗ X = (Rg ◦ I)∗ X = (I ◦ Lg−1 )∗ X = I∗ (Lg−1 )∗ X = I∗ X.
The third one comes from the fact that Lg ◦ Rh = Rh ◦ Lg .
Definition 3.4.5. A one parameter subgroup of a Lie group G is a smooth morphism
h from R to G.
Example 3.4.6. We have seen in the examples that the one parameter subgroups of
Gln (R) are of the form h(t) = etA .
If G is a Lie subgroup of Gln (R) and h is a one parameter subgroup of G, then h is
a one parameter subgroup of Gln (R), so h(t) = etA for some A ∈ Mn (R). Moreover,
since h(t) ∈ G, A = h0 (0) ∈ TI G. We will see that any A ∈ TI G actually gives rise
to a one paramter subgroup of G.

55
From a one parameter subgroup of G, we get a one parameter subgroup of diffeomor-
phisms of G by setting:
ϕt (x) = xh(t) = Rh(t) (x).

Proposition 3.4.7. ϕt is the flow of a left invariant vector field X. Conversely, the
flow of any left invariant vector field X is of the form ϕt (x) = xh(t) for some uniquely
determined one parameter subgroup h.

Proof. We know that ϕt is the flow of the vector field:


d
Xp = ϕt (p).
dt |t=0

We want to show that (Lg )∗ X = X, but the flow of (Lg )∗ X is given by

Lg ◦ ϕt ◦ L−1 −1
g = Lg ◦ Rh(t) ◦ Lg = Lg ◦ Lg −1 ◦ Rh(t) = ϕt .

So X and (Lg )∗ X have the same flow, hence they are equal.
Conversly, let h(t) be the solution to h0 = X(h), h(0) = e. We have, since (Lg )∗ X = X,
that ϕt ◦ Lg = Lg ◦ ϕt , hence:

ϕt (g) = ϕt (Lg (e)) = Lg (ϕt (e)) = gh(t).

So we only need to show that h(t) is a one parameter subgroup, the fact that h(t + s) =
h(s)h(t) comes from the similar property of ϕt . We need to show that h is in fact defined
on the whole of R. We know that h(t) exists for t ∈ (−ε, ε), now if h is defined up to
T > 0, it is easy to see that h(t)h(eps/2) yiedls a solution defined until T + ε/2. So h
is defined for all t ≥ 0. The same argument works backward in time, so h(t) is defined
for any real t.

3.5 The Lie algebra of a Lie group


Definition 3.5.1. A Lie algebra on a field K is a K-vector space V endowed with an
antisymmetric K-bilinear map [ , ] : V × V → V which satisfy the Jacobi identity:

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

Example 3.5.2. The vector space of vector fields on a manifold M , with the Lie bracket
of vector fields is a Lie algebra.
Example 3.5.3. If G is a Lie group, then the space of left invariant vector fields with
the Lie bracket is a Lie algebra. Since there is a canonical isomorphism between left
invariant vector fields and Te G (given by X 7→ Xe ), this gives a canonical lie algebra
structure on Te G. This Lie algebra is called the Lie algebra of G and usually
denoted by g.

56
Example 3.5.4. Let us specialise the previous discussion to G = GLn (R). Consider
A, B ∈ Te G ' Mn (R), the associated one parameter subgroups are etA and etB while
the flows are ϕA
t : M 7→ M e
tA
and ϕB tB
t : M 7→ M e . To compute the Lie bracket of
A and B, let us consider the left invariant extensions XA (g) = gA and XB (g) = gB.
We know that:
d
[XA , XB ] = (ϕB )∗ XA
dt |t=0 t
but:
−tA
(ϕB
t )∗ XA (e) = TϕB ϕB XA (ϕB
−t (e) t −t (e)) = e BetA .
Using that etA = I + tA + . . . , we get that [XA , XB ](e) = BA − AB, thus the Lie
algebra structure on Mn (R) is just given by the commutator [A, B] = BA − AB.
Actually, for any associative algebra A, the commutator defines a Lie bracket on
A.

Definition 3.5.5. A Lie subalgebra of V is a linear subspace of V which is stable


under [ , ].

Definition 3.5.6. A Lie algebra morphism is a linear map A from (V, [ , ]V ) to


(V 0 , [ , ]V 0 ) such that:
[A(x), A(y)]V 0 = A([x, y]V ).

A Lie algebra isomorphism is a bijective Lie algbra morphism, it is then automatic


that its inverse is also a Lie algebra morphism.

Definition 3.5.7. The exponential map of a Lie group G is the map exp : g → G
which associates to each Xe ∈ g ϕ1 (e) where ϕt is the flow of the left invariant vector
field X̄ such that X̄e = Xe .

One can check that this generalises the matrix exponential: if A ∈ Mn (R) (which is
the Lie algebra of Gln (R), the X : g 7→ gA is the left invariant vector field such that
XI = A and c : t 7→ etA statisfies c0 = Xc and c(0) = I.

Proposition 3.5.8. For any one parameter subgroup h(t) of G, there exist a unique
Xe ∈ g such that h(t) = exp(tX)

Proof. We have seen that for any one parameter subgroup h(t) there is a left invariant
vector field X such that h(t) = ϕt (e) where ϕt is the flow of X. This comes from the fact
that if c(t) is a solution to c0 = X(c), then cα (t) = c(αt) is a solution to c0α = αX(cα ).
In particular, the flow of a left invariant veector field X is given by: ϕt (x) = x exp(tXe ).

Proposition 3.5.9. exp : g → G is smooth and a local diffeomorphism on a neighbor-


hood of 0 ∈ g.

57
Proof. The smoothness of exp comes from the fact that if Xα , where α is a parameter
in an open set of a vector space, is a smooth family of vector fields then the map
(x, t, α) 7→ ϕαt (x) where ϕαt is the flow of Xα is smooth. (In this α ∈ g.)
We now just need to compute the differential of exp at 0. To see this consider the
curve c : t 7→ tXe in g. Then one shows that exp(tXe ) is a solution to c0 = X(c) with X
the left invariant extension of Xe , thus c0 (0) = Xc(0) , thus:

Te exp(Xe ) = Xe

which shows that Te exp is invertible.


We have seen that the pushforward by Rg of a left invariant vector field is again left
invariant. The following proposition shows how this translates under the identification
of the space of left invariant vector fields with Te G.

Proposition 3.5.10. Let X be a left invariant vector field, then:


d
(Rg )∗ Xe = (g −1 exp(tXe )g).
dt |t=0

Proof. The flow of X is ϕt (x) = x exp(tXe ), thus the flow of (Rg )∗ X is ϕ̃t = Rg ◦ϕt ◦Rg−1 .
Then:
d d
(Rg )∗ Xe = ϕ̃t (e) = (g −1 exp(tXe )g).
dt |t=0 dt |t=0

We define, for every g ∈ G a linear map Adg : g → g by:

d
Adg (Xe ) = (g exp(tXe )g −1 ) = (Rg−1 )∗ Xe .
dt |t=0

It follows from the properties of pushforwards of vector fields that Ad(g1 g2 ) = Ad(g1 ) ◦
Ad(g2 ). This shows that Ad : G → Gl(g) is a linear representation of G. Moreover:

[Adg X, Adg Y ] = Adg [X, Y ],

in other words, Adg is a Lie algebra morphism.


Ad is called the adjoint representation of G.

Proposition 3.5.11.
d
(Adexp(tX) Y ) = [X, Y ].
dt |t=0
Proof. We have that Adexp(tX) Y = (Rexp(−tX) )∗ Y , but Rexp(−tX) is the flow of −X,
proposition 2.6.4 then gives the answer.
We this we can show the following fundamental property of Lie group morphisms:

58
Proposition 3.5.12. Let f : G → H be a Lie group morphism, then Te f : g → h is a
Lie algebra morphism.

Proof. Let Xe ∈ g and h(t) be the one parameter subgroup of G generated by Xe , then
k(t) = f (g(t)) is a one parameter subgroup of H, thus corresponds to a unique Ze ∈ h.
To find this Ze , we consider the flow of the left invariant extension Z of Ze , we know it
is given by: ϕt : x 7→ xk(t), so Ze = dtd |t=0 ϕt (e) = dtd |t=0 k(t) = Te f (Xe ). In particular
f (exp(tXe )) = exp(tZe ).
Now for any g ∈ G, we have f (g exp(tXe )g −1 ) = f (g) exp(tZe )f (g −1 ). Taking the
derivative of the last equality with respect to t at t = 0, Proposition 3.5.10 yields:

Te f (Adg Xe ) = Adf (g) Ze = Adf (g) Te f (Xe ).

We now take g to be exp(sYe ) and differentiate with respect to s at s = 0, the last


proposition gives:
Te f ([Xe , Ye ]) = [Te f (Xe ), Te f (Ye )].

Example 3.5.13. Consider the determinant det : Gln (R) → R∗ , this is a Lie group
morphism. So Te det = trace is a Lie algbra morphism from Mn (R) to R, so
trace([A, B]) = [trace(A), trace(B)] = 0 (since the Lie bracket on R is trivial).

Corollary 3.5.14. Let H ⊂ G be a Lie subgroup of G. Then its Lie algebra h is a Lie
subalgebra of g.

Proof. The injection H → G is a Lie group morphism.


Example 3.5.15. Let us consider G ⊂ Gln (R) a Lie subgroup. Then g is a Lie subalgebra
of Mn (R), and the Lie bracket on g is just the restriction of the Lie bracket of Mn (R),
which is the matrix commutator.
For instance if G = O(n), then g = TI O(n) is the space of antisymmetric matrices.
We recover in this way that a commutator of antisymmetric matrices is antisymmet-
ric.
The following theorem is an example of the rigidity Lie groups enjoy:

Theorem 3.5.16. Let f : G → H be a continuous group morphism between two Lie


groups, then f is smooth.

The proof of this theorem will require the following Lemmas:

Lemma 3.5.17. Let f : G → H be a group morphism which is smooth on a neighborhood


of e ∈ G, then f is smooth on G

Proof. Pick any g ∈ G, then: Lf (g) ◦ f ◦ Lg−1 = f , the left hand side is smooth on a
neighborhood of g since Lg−1 (g) = e and left multiplication is smooth.

59
Remark 3.5.18. We can differentiate this identity to get:
Te Lf (g) Te f ◦ Tg Lg−1 = Tg f.
Since left multiplication is a diffeomorphism, it follows that if f is a Lie group mor-
phism, then f is an immersion (resp. submersion, local diffeomorphism) if and only
if Te f is injective (resp. surjective, bijective).
Lemma 3.5.19. Let (X1 , . . . Xn ) be a basis of g then the map:
ϕ : (t1 , . . . , tn ) → exp(t1 X1 ) · · · exp(tn Xn )
is a diffeomorphism on a neighborhood of 0.
Proof. ϕ is smooth. Moreover we have:
d
T0 ϕ(Xi ) = exp tXi = Xi .
dt |t=0
Thus T0 ϕ : g → g is the identity.
Lemma 3.5.20. Let h : R → H be continuous group morphism, then there exist Y ∈ h
such that h(t) = exp(tY ).
Proof. Consider a ball B centered at 0 in h such that exp is a diffeomorphism from B
to U = exp B/2. Then f −1 (U ) contains the interval I = (−ε, ε) for ε > 0 small enough.
Pick t0 ∈ I, then there exist Ỹ ∈ B/2 such that f (t0 ) = exp(Ỹ ). Similarly, there is
Ỹ 0 ∈ B/2 such that f (t0 /2) = exp(Ỹ 0 ). Then we have that:
exp(2Ỹ 0 ) = f (t0 /2)2 = f (t0 ) = exp(Ỹ ).
Since 2Ỹ 0 and Ỹ belong to B where exp is a diffeomrphism, Ỹ 0 = Ỹ /2. Repeating
the following argument, we get that f (t0 /2k ) = exp(Ỹ /2k ), and the group morphism
property of f implies that f (nt0 /2k ) = exp(nỸ /2k ), since f is continuous we must have
f (st0 ) = exp(sỸ ), which implies that f (s) = exp(sY ) for Y = Ỹ /t0 .
Proof (of Theorem 3.5.16). Consider a basis (X1 , . . . , Xn ) of g, then f (exp(tXi )) is a
continuous morphism from R to H, hence f (exp(tXi )) = exp(tYi ). Consider ϕ as in
Lemma 3.5.19, then:
f ◦ ϕ(t1 , . . . , tn ) = f (exp(t1 X1 ) · · · exp(tn Xn )) = exp(t1 Y1 ) · · · exp(tn Yn )
is smooth from Rn to H. Since ϕ is a diffeo from a neighborhood of 0 ∈ Rn to a
neighborhood of e ∈ G; f is smooth on a neighborhood of e, and hence smooth.

3.6 Homegeneous spaces


The goal of this section is to show that the quotient space G/H where H is a Lie subgroup
of G has a natural smooth structure. A lot of the manifolds we have encountered can
actually be described as quotients G/H, this manifolds are called (smooth) homogeneous
spaces.

60
3.6.1 Observations on topological groups
We will need some topological properties of Lie groups which hold more generally for
topological groups. In this subsection, G will be a topological group and H will be a
closed subgroup of G.

Proposition 3.6.1. Let U be a neighborhood of e ∈ G, then there exist a neighborhood


V ⊂ U of e such that V −1 = V and a neighborhood W of e such that W 2 = gh|g, h ∈ W ⊂
U.

Proof. Since the inversion is continuous, U −1 is open, then V = U ∩ U −1 fits.


For the second point, the multiplication µ : G × G → G is continuous, O = µ−1 (U ) is
open in G × G and contains (e, e), in particular O contains an open set W × W where
W is a neighborhood of e.
For any g, h 7→ gh is an homeomorphism of G. In particular, any neighborhood of g
is of the form gU for some neighborhood U of e. Using right multiplication instead, any
neighborhood of g is also of the form U 0 g.
In a topological group, there is a nice interplay between the group structure and
connectedness.

Proposition 3.6.2. Any open subgroup G0 of G is closed, in particular if G is connected


then G0 = G.

Proof. We want to show that the complement of G0 is open. Consider an open neigh-
borhood U of e included in G0 , then if g ∈/ G0 , gU is an open neighborhood of g which
doesn’t intersect G . (If gu ∈ G , then g ∈ G0 u−1 = G0 .) Thus G\G0 is open.
0 0

Proposition 3.6.3. The connected component of e ∈ G, denoted by G0 is an open


normal subgroup of G.

Proof. Consider V a connected open neighborhood of e such that V −1 = V . (To find


one, just take the connected component of e in a neighborhood which is stable under
inversion.) We will prove the following claim:
[
G0 = V n.
n

Since V is connected, V nSis connected and n V n is also connected, thus the inclusion
S
⊃ is proved. Moreover n V n S is a subgroup of G, since if g ∈ V n and h ∈ V k then
gh ∈ V n+k and g −1S∈ V n . Now n V n contains a neighborhood of the identity S
(V ), and
is a subgroup, so n V is an open subgroup of G, and thusSis closed. Since n V n is
n

connected, it is one of the connected component of G, hence n V n = G0 .


To see that G0 is a normal subgroup, just remark that gG0 g −1 is connected and
contains e, and is thus included in G0 .

61
When H is a subgroup of G, we have the right action of H on G given by h · x = xh =
Rh (x). The relation ∼ defined by x ∼ y if and only if x = yh(= h · y) for some h ∈ H
is an equivalence on G. The quotient of G by this equivalence is denoted by G/H. Its
elements are of the form gH. As usual π : G → G/H will denote the quotient map.
The left action of G on itself g · x = gx = Lg (x) goes down to G/H: if x = yh, then
gx = (gy)h. This action is continuous and transitive. We will again denote it by Lg , it
satisfies: Lg ◦ π = π ◦ Lg .
We now show some topological properties of G/H endowed with the quotient topology.
Proposition 3.6.4. 1. π : G → G/H is open.
2. G/H is Hausdorff.
Proof. U ⊂ G/H is open if and only if π −1 (U ) is open in G. Let O ⊂ G be open, then
Õ = π −1 (π(O)) can be written as:
[
Õ = gO.
g∈G

Since each gO is open, Õ is open.


For the second point, let x̄ = π(x) and ȳ = π(y) be two distinct point in G/H. Then
x doesn’t belong to yH, which is closed in G. Thus there is a neighborhood U of e such
that U x∩yH =. This implies that U xH ∩yH =: if uxh0 = yh, then ux = y(hh0−1 ) ∈ yH.
Consider a neighborhood V ⊂ U of e such that V 2 ⊂ U and V −1 = V , then V 2 xH ∩
yH =, which implies that V xH ∩ V yH =. This says that π(V x) and π(V y) are disjoint,
since π is open we are done.

3.6.2 Quotient of Lie groups


We now get back to the case where G is a Lie group and H is a Lie subgroup of G.
This whole section is devoted to the proof of the following result:
Theorem 3.6.5. There exist a unique smooth structure on G/H such that π : G → G/H
is a smooth submersion.
We already know that G/H is Hausdorff, the second countable property then follows.
We set M = G/H.
We now need to build charts on M , for this we will use the exponential map. Consider
a suppelment m to h in g. We have g = m ⊕ h.
Lemma 3.6.6. The map ψ : m ⊕ h → G defined by:

(X, Y ) 7→ exp(X) exp(Y )

is a local diffeomorphism on a neighborhood of 0.


Proof. One sees easily that T0 ψ((X, 0)) = (X, 0) and T0 ψ((0, Y )) = (0, Y ). Thus T0 ψ is
an isomorphism from m ⊕ h to g, the inverse function theorem applies.

62
Let V ⊂ h and U ⊂ m be neighborhoods of 0 such that ψ is a diffeomorphism from
U × V to its image.
Lemma 3.6.7. There exists a neighborhood U 0 ⊂ U of 0 ∈ m such that for any g ∈ G,
the map ϕg : U 0 → M defined by:
ϕg (X) = π(g exp(X))
is an homeomorphism onto its image.
Proof. Since ϕg = Lg ◦ ϕe and Lg : G/H → G/H is an homeomorphism, we only need
to work with ϕe .
We will first restrict U in the following way: first choose W a neighborhood of e ∈ G
such that W 3 ⊂ ψ(U × V ) and W −1 = W , and choose U 0 and V 0 neighborhoods of 0 in
h and m such that U 0 × V 0 ⊂ ψ −1 (W ).
We now show that ϕe is injective on U 0 . Pick X and X 0 in U 0 such that ϕe (X) =
ϕe (X 0 ). This means that there exists h ∈ H such that exp(X) = exp(X 0 )h, moreover
h = (exp X)−1 exp(X) ∈ W 2 , thus exp(X 0 )h ∈ W 3 ⊂ ψ(U × V ). We can then apply the
previous lemma to conclude that X = X 0 .
To prove that ϕe is an homeomorphism on its image, we only need to show that ϕe is
open. To see this, just remark that if O ⊂ U 0 is open, then O × V 0 is open and ψ(O × V 0 )
is also open since ψ is a diffeomorphism. But ϕe (O) = π(ψ(O × V 0 )) and π is open, so
ϕe (O) is open.
Proof (of Theorem 3.6.5). The (π(g exp(U 0 )), (ϕg )−1 ) from the previous lemma form a
topological atlas M , to show that it is smooth we need to show that ϕ−1
g2 ◦ ϕg2 is smooth.
0 0
Asssume that O = π(g1 exp(U )) ∩ π(g2 exp(U )) is not empty, then for any x ∈ O
there exist unique X1 and X2 such that:
x = π(g1 exp(X1 )) = π(g2 exp(X2 )),
and we need to show X1 7→ X2 is a diffeomorphism.
Fix some x̂ ∈ O and let X̂1 and X̂2 such that ϕg1 (X̂1 ) = ϕg2 (X̂2 ), then there exist
ĥ ∈ H such that g1 exp(X̂1 )h = g2 exp(X̂2 ).
Consider the neighborhood W of e from the proof of the previous lemma. Then
g2 exp(X̂2 ) = g1 exp(X̂1 )ĥ ∈ g2 W , so g2−1 g1 exp(X̂1 )ĥ ∈ W . Hence for X1 in some neigh-
borhood Û of X̂1 , g2−1 g1 exp(X1 )ĥ ∈ W . Hence there exist X̃ ∈ m and Ỹ ∈ h, depending
smoothly on X1 ∈ Û , such that: g2−1 g1 exp(X1 )h = ψ(X̃, Ỹ ) = exp(X̃) exp(Ỹ ). So:
g2 exp(X̃) exp(Ỹ ) = g1 exp(X1 )ĥ.
Moreover, we have:
ϕg2 (X̃) = π(g2 exp(X̃)) = π(g2 exp X̃ exp Ỹ )
= π(g1 exp(X1 )ĥ) = π(g1 exp(X1 ))
= ϕg1 (X1 ) = x
= ϕg2 (X2 ).

63
So X̃ = X2 since ϕg2 is injective. Since X̃ depends smoothly on X1 , we have defined a
smooth structure on M .
We now prove that π : G → G/H is a submersion. Consider the charts of G given by
the inverse of the maps ψg : (X, Y ) 7→ gψ(X, Y ), in these charts arguing as before we
have ϕ−1
g ◦ π ◦ ψg (X, Y ) = X, which is a submersion.
For the uniqueness part, the structure of topological manifold on M has to be given
by the atlas we considered (since the toplogy on M is the quotient topology), and the
requirement that π is smooth forces the maps ϕg to be smooth.

3.6.3 Homogeneous spaces


On M = G/H, since π is a submersion, the right action Rg of G on itself goes through
the quotient to a smooth action of G on M which is transitive. A manifold on which
a Lie group acts smoothly transitively is called a smooth G-homeneous space. The
following theorem says that all such manifolds are of the form G/H.

Theorem 3.6.8. Let M be a smooth G-homogeneous space and x ∈ M , set Gx = {g ∈


G|g · x = x} (the stabilizer of x), then Gx is a Lie subgroup of G and M is diffeomorphic
to G/Gx .
The diffeomorphism is built by quotienting the map F (g) = g · x.

The proof uses some tools that we haven’t introduced and won’t be given here. How-
ever a lot of the examples of manifolds we have seen so far can be seen in this way.
Example 3.6.9. Consider the sphere Sn in Rn+1 . The Lie group G = SO(n + 1) acts
transitively on Sn . The stabilizer of (1, 0, . . . , 0) is given by the subgroup H consisting
of matrices of the form:  
1 0
0 h
where h ∈ SO(n). We then have a diffeomorphsim from G/H to Sn .

64
3.7 Exercises
Exercise 3.1. Consider Cn with its usual hermitian structure (hz1 , z2 i =
P i i
i z1 z̄2 ).
Recall the adjoint M ∗ of M ∈ Mn (R) is defined by M ∗ = M̄ T .
The unitary group U (n) is the group of matrices M ∈ GLn (C) such that M M ∗ = I,
the special unitary group SU (n) is the intersection of U (n) with SLn (C).
Show that U (n) and SU (n) are Lie subgroups of GLn (C), and describe their Lie
algebras.
Exercise 3.2. Let E be a finite dimensional (real or complex) vector space and q be a
non degenerate quadratic form on E, set:

O(q) = {A ∈ Gl(E)|∀x ∈ E, q(Ax) = q(x)}

We will denote by b : E × E → R the associated symmetric bilinear form.


The goal of this exercise is to prove that O(q) is a Lie group.

1. Show that for any M ∈ L(E), there exist a unique M ∗ ∈ L(E) such that b(M x, y) =
b(x, M ∗ y) for any x, y ∈ E. Show that M 7→ M ∗ is linear (and hence smooth).

2. Show that M ∈ O(q) if and only if M M ∗ = I.

3. Let ϕ : M 7→ M M ∗ and S = {M ∈ L(E)|M = M ∗ }. Show that ϕ(M ) ∈ S and


show that for any M ∈ O(q), TM ϕ : E → S is surjective.

4. Show that O(q) is a Lie subgroup of GL(E) and compute its tangent space at I.
n
Nondegenerate quadratic forms on RP are classified
Pn by their signature, they are all
p i 2 i 2
congruent to one of the forms q(x) = i=1 (x ) − i=p+1 (x ) . The orthogonal group
of this particular quadratic form is denoted by O(p, n − p).
Exercise 3.3. Let E be the space of 2 symmetric matrices.

1. Show that det is a quadratic form on E of signature (2, 1).

2. For any A ∈ SL2 (R), define ϕA : E → E by ϕA (S) = ASAT . Show that A 7→ ϕA


is a Lie group morphism from SL2 (R) to O(det).

3. Show that O0 (2, 1) is isomorphic to SL2 (R)/{I, −I}

Exercise 3.4. Show that a vector field on (Rn , +) is left invariant if and only if it is
constant. Describe left invariant vector fields of GLn (R).
Exercise 3.5. Show that the vector fields on S3 ⊂ H given by X1 : x 7→ ix, X2 : x 7→ jx
and X3 : x 7→ kx and form a basis of the space of left invariant vector fields. Compute
[X1 , X2 ].

65
Exercise 3.6. * Consider the affine group G = Af f (Rn ). We define a map ϕ : G →
GLn+1 (R) by sending the affine map f (x) = Ax + b to the matrix:
 
1 0
.
b A

Show that ϕ is an injective Lie group morphism and that ϕ(G) is a Lie subgroup of
GLn (R). Use it to describe the Lie bracket on Te G in terms of A and b.
Exercise 3.7. Show that on G is a matrix subgroup, then for g ∈ G and X ∈
g,Adg (X) = gXg −1 . Prove directly that, for [X, Y ] ∈ g:

d
Adexp(tY ) X = [Y, X].
dt
Exercise 3.8. Consider the set of matrices in A = M2 (C) of the form:
 
α β
−β̄ ᾱ

1. Show that A is a complex subalgebra of M2 (C).

2. Consider the map f : A → H:


 
α β
7→ α + βj
−β̄ ᾱ

Show that f is a C-linear isomorphism. (H is viewed as a C vector space with the


left multiplication.)

3. Show that f is moreover an algebra isomorphism.

4. Use this isomorphism to show that S3 is isomorphic to SU (2).

Exercise 3.9. For any unit quaternion q ∈ S3 , consider the the transformation ϕq of H
given by: ϕq (x) = qxq −1 .

1. Show that ϕq is an isometry for the scalar product hq1 , q2 i = <(q1 q¯2 ).

2. Show that ϕq preserves the space of purely imaginary quaternions.

3. Use this fact to build a Lie morphism S3 → SO(3, R), what is its kernel ?

4. Show that SO(3) is diffeomorphic to RP3 .

Exercise 3.10. For q1 , q2 ∈ S3 × S3 , consider the transformation of H defined by


ϕq1 ,q2 (x) = q1 xq2−1 .

1. Use it to define a surjective Lie group morphism S3 × S3 → SO(4), compute its


kernel.

66
2. Show that SO(4) is not simple. (It has non trivial normal subgroups.)

Exercise 3.11. * Let H ⊂ G be a normal Lie subgroup. Show that h is an ideal of g,


that is for any X ∈ g, Y ∈ h, [X, Y ] ∈ h. Hint: "differentiate" the relation ghg −1 ∈ H.
Exercise 3.12. ** Let G be an Abelian connected Lie group.

1. Show that for any X, Y ∈ g, [X, Y ] = 0.

2. Show that exp : (g, +) → (G, ×) is a Lie group morphism.

3. Use the fact that exp is a local diffeomrphism and that G is connected to prove
that exp is surjective.

4. Show that the kernel K of exp is a discrete subgroup of (g, +).

Discrete subgroup of vector spaces are of the form Ze1 ⊕ · · · ⊕ Zek where the ei ’s are
linearly independant, from this one can deduce that G ' g/K is actually isomorphic to
(S1 )k × Rn−k .
Exercise 3.13. * For each of the sets below, find a transitive Lie group action and use
it to describe it as a homegeneous space:

1. The projective space RPn .

2. The set of affine lines in Rn .

3. The set of k dimensional vector spaces in Rn (this set is called the Grassmannian
of k-planes).

4. The of flags in a real vector space E of dimension n. (A flag in E is a sequence


E0 , E1 , . . . , En of linear subspaces of E such that Ei ⊂ Ei+1 (stricly), in particular,
dim(Ei ) = i.)

67
4 Tensors, differential forms and
Stokes Theorem
4.1 The cotangent bundle and 1-forms
One of the most important construction in linear algebra is duality, it also has its im-
portance in differential geometry. Recall that if E is a real vector space, its dual E ∗ is
just the vector space of linear maps from E to R.
Also recall that if f : E → F is a linear map, it induces a linear map f ∗ : F ∗ → E ∗ ,
called the transpose of f by setting, for α ∈ F ∗ , f ∗ α : x ∈ E 7→ α(f (x)) ∈ R, or more
concisely f ∗ α = α ◦ f . f ∗ is an isomorphism if and only if f is, and (f ◦ g)∗ = g ∗ ◦ f ∗ .

4.1.1 The cotangent bundle


The cotangent bundle of a smooth manifold M , denoted by T ∗ M , is, as a set the
union of the duals to the tangent spaces:
[
T ∗M = (Tp M )∗ .
p∈M

An element of T ∗ M is thus a linear form on some Tp M .


In the same way as we did for the tangent bundle, we can define a smooth structure on
T M such that the natural projection π : T ∗ M → M defines a vector bundle structure

on T ∗ M .
Recall that the charts on T M are given ΦU : T U → U → Rn where (U, ϕU ) is a chart
of M and:
ΦU : v ∈ Tp U → (ϕU (p), Tp ϕU (v)).
Let us define T ∗ U = p∈U (Tp M )∗ and:
S

ΨU : α ∈ Tp U → (ϕU (p), (Tp ϕ−1 ∗ n ∗


U ) (α)) ∈ U × (R ) .

Since Tp ϕU is an isomorphism, ((Tp ϕU )−1 )∗ is an isomorphism and ΨU is a bijection. the


topology on T ∗ M is built in the same way the topology on T M was. The ΨU form a
topological atlas for this topology.
We then compute the transition function given by ΨU V = ΨV ◦ Ψ−1 U , if ΨV comes from
a smooth chart (V, ϕV ):
 
−1 ∗ −1 ∗ −1

ΨU V (x, α) = ϕU V (x), (TϕU (x) ϕV ) (TϕU (x) ϕU ) α
= ϕU V (x), ((Tx ϕU V )∗ )−1 α .


68
These are smooth diffeomorphisms. Moreover, ΨU restricted to (Tp M )∗ is a linear iso-
morphism from (Tp M )∗ to (Rn )∗ .
Thus we have shown:
Proposition 4.1.1. T ∗ M with the projection π : T ∗ M → M is a smooth vector bundle
over M .

4.1.2 Differential 1-forms


Definition 4.1.2. A (smooth) differential 1-form on M is a smooth section of T ∗ M .
The vector space of smooth 1-forms on M is usually denoted by Ω1 (M ).
Example 4.1.3. Let f be a smooth function on M , then for each p ∈ M , Tp f is a linear
form on Tp M . Thus T f : p 7→ Tp f can be seen as a section of T ∗ M , which is usually
denoted by df .
To see that it is smooth we compute:
ΨU ◦ df ◦ ϕ−1
U (x) = ψU (Tϕ−1 f)
U (x)
 
= x, ((Tϕ−1 (x) ϕU )−1 )∗ Tϕ−1 (x) f
U U
 
= x, Tϕ−1 (x) f ◦ Tx ϕ−1
U
U
−1

= x, Tx (f ◦ ϕU )
which is smooth since f ◦ ϕ−1
U is.
1 n
To express differential forms
 in coordinates  ϕU (p) = (x (p), . . . , x (p)), we have a
natural basis for each Tp M : ∂x∂ 1 |p , . . . , ∂x∂ 1 |p . This gives us a dual basis of Tp∗ M , that
we denote by (dx1|p , . . . , dxn|p ). p 7→ dxi|p is a actually the differential of the coordinate
function p 7→ xi (p).
From this we deduce that any smooth one form α can be written as α = i αi dxi ,
P
where the αi ’s are smooth P functions on U . Moreover if α = df for some smooth function
f : M → R, then df = i (∂i · f )dxi , just as in the case of functions on an open set of
Rn .
Let f : M → N be a smooth map and α ∈ Ω1 (N ).
Definition 4.1.4. The pullback of α by f is the smooth 1-form on M defined by:
f ∗ α : v ∈ Tp M 7→ α(Tp f (v)).

4.2 A little tensor algebra


4.2.1 Tensor product
Let E and F be two finite dimensional real vector spaces. We will define tensor prod-
ucts only for dual spaces, which is enough for our purpose. However this restriction is
unnecessary.

69
Definition 4.2.1. The tensor product E ∗ ⊗ F ∗ is the vector space of bilinear maps from
E × F to R.
There is a natural bilinear map: E ∗ × F ∗ → E ∗ ⊗ F ∗ defined by:

(f, g) ∈ E ∗ × F ∗ 7→ (f ⊗ g : (u, v) 7→ f (u)g(v)).


i
Assume that E has a basis (ei )i and F has a basis (f j )j , and denote by (e )i and (f j )j
the corresponding dual basis. Let b ∈ E ∗ ⊗ F ∗ , u = i ui ei ∈ E and v = j v j fj ∈ F ,
P P

then: b(u, v) = i,j ui v j b(ei , fj ). Recall that by definition of dual basis: ei ⊗ f j (u, v) =
P

ei (u)f j (v) = ui v j . This show that: b = i,j bij ei ⊗ f j with bij = b(ei , fj ). In particular
P

the (ei ⊗ f j ) form a basis Nof E ∗ ⊗ F ∗ , so dim(E ∗ ⊗ F ∗ ) = dim(E ∗ ) × dim(F ∗L ).


Similarly one defines ni=1 Ei∗ as the vector space of n-linear maps from ni=1 Ei to
R. It has dimension dim(E1∗ ) × · · · × dim(En∗ ), and a basis is given by elements of the
ei11 ⊗ · · · ⊗ einn if (eikk )ik is a basis of (Ek )∗ .
Of particular interest to us is the tensor product k E ∗ of E ∗ with itself k times, it is
N
the vector space of k-linear maps from E k to R (also called k-linear forms). From now
on, n will denote the dimension of E
One can define the tensor product of α ∈ k E ∗ and β ∈ l E ∗ by:
N N

α ⊗ β(v1 , . . . , vk+l ) = α(v1 , . . . , vk )β(vk+1 , . . . , vk+l ).

α ⊗ β is thus an element of k+l E ∗ . Note that this tensor product is associative but
N
not commutative. L Nk
The tensor product turns T (E ∗ ) = k into an associative algebra called the

tensor algebra of E .

4.2.2 Alternating forms and wedge products


Nk
Definition 4.2.2. An n-linear form α ∈ E ∗ is said to be alternating if for any
permutation σ ∈ Sk :

α(vσ(1) , . . . , vσ(k) ) = sign(σ)α(v1 , . . . , vk ),

where sign(σ) is the signature of the permutation σ.


Proposition 4.2.3. Let α ∈ k E ∗ ,
N

1. α is alternating if and only if α(v1 , . . . , vk ) = 0 as soon as two of the vi ’s are equal.

2. If α is alternating then α(v1 , . . . , vk ) = 0 as soon as the vk ’s are linearly dependant.


Proof. For the first part, since transpositions generate Sk as a group, it is enough to show
that exchanging two of the vk ’s in α(v1 , . . . , vk ) changes the sign, since the signature of a
transposition is −1. So we only need to show the condition in (1) is equivalent to the fact
that β : (vi , vj ) 7→ α(v1 , . . . , vk ) is antisymmetric, which is equivalent to β(vi , vj ) = 0 if
vi = vj .

70
For the second part, we might as well assume that v1 = i≥2 λi vi . Then:
P

X
α(v1 , . . . , vk ) = λi α(vi , v2 , . . . , vk ),
i≥2

and each term of the sum vanishes since vi appears two times in it.
We will denote by Λk E ∗ the space of alternating k-forms on E. Any 1-form is alter-
1 ∗
nating, so Λ E = E ∗ , we will set (it is purely a convention) Λ0 E ∗ = R. We will also
set ΛE ∗ = k Λk E ∗ .
L

Proposition 4.2.4. Let (ei ) be a basis of E and α ∈ Λk E ∗ , then:


 i i1

v 1
1
· · · v k
det  ... ..  α(e , . . . , e )
X
α(v1 , . . . , vk ) =

.  i1 ik
1≤i1 <···<ik ≤n ik ik
v1 · · · vk

where vl = i vli ei .
P

Proof. The k-linearity gives:


X
α(v1 , . . . , vk ) = v1i1 v2i2 · · · vkik α(ei1 , . . . , eik ).
1≤i1 ,...,ik ≤n

The alternating property gives that we can only consider k-tuples (i1 , . . . , ik ) where
all the il are different. We then get the required formula by regrouping the k-tuples
that corresponds to the same subsets of {1, . . . , n} and applying the formula for the
determinant of a matrix in term of its coefficients.
 
k ∗ n
Corollary 4.2.5. dim(Λ E ) = .
k
In particular Λk E ∗ is non trivial only for k between 0 and n, also we get that
dim(ΛE ∗ ) = 2n .
Definition 4.2.6. Alt : k E ∗ → Λk E ∗ is defined by:
N

1 X
Alt(α)(v1 , . . . , vk ) = sign(σ)α(vσ(1) , . . . , vσ(k) ).
k! σ∈S
k

We have that α is alternating if and only if Alt(α) = α.


Example 4.2.7. If β = 2 E ∗ , then:
N

1
Alt(β)(v, w) = (β(v, w) − β(w, v)).
2
We want to turn ΛE ∗ into an algebra. Unfortunately the tensor product of two
alternating forms is not alternating in general, so we need to define a new product,
called the wedge product:

71
Definition 4.2.8. Let α ∈ Λk E ∗ and β ∈ Λl E ∗ . We define α ∧ β ∈ Λk+l E ∗ by:

(k + l)!
α∧β = Alt(α ⊗ β).
k!l!
Example 4.2.9. If α, β ∈ Λ1 E ∗ , then:

α ∧ β(v, w) = α(v)β(w) − α(w)β(v).

The wedge product has the following properties:

Proposition 4.2.10. 1. (α, β) 7→ α ∧ β is bilinear.

2. (α ∧ β) ∧ γ = α ∧ (β ∧ γ).

3. If α ∈ Λk E ∗ and β ∈ Λl E ∗ , then α ∧ β = (−1)kl β ∧ α.

Example 4.2.11. One can show that if (ei ) is a basis of E and ei is the dual basis of E ∗ ,
then:  i 
v11 · · · vki1
ei1 ∧ · · · ∧ eik (v1 , . . . vk ) = det  ... .. 

. 
v1ik · · · vkik
where vl = vli ei .
The first and second properties show that the wedge product can be extended on the
whole ΛE ∗ , and turns it into an associative algebra.
Let f : F → E be a linear map. And α ∈ Λk E ∗ then f ∗ α is the element of Λk F ∗
defined by:
f ∗ α(u1 , . . . , uk ) = α(f (u1 ), . . . , f (uk )).
It is easy to check that f ∗ (α ∧ β) = f ∗ α ∧ f ∗ β and that (f ◦ g)∗ = g ∗ ◦ f ∗ .
Example 4.2.12. Consider the one dimensional vector space Λn E ∗ , then for any f ∈
L(E), f ∗ defines a linear map from Λn E ∗ to itself. Thus there exists λ ∈ R such
that f ∗ ω = λω for any ω ∈ Λn E ∗ . One can actually show that λ = det(f ). This is a
totally coordinate free definition of the determinant. The fact that (f ◦ g)∗ = g ∗ ◦ f ∗
gives that det(f ◦ g) = det(g) det(f ).

4.3 Differential forms


One defines the bundle of p-form Λp T ∗ M as a set by the formula:
[
Λp T ∗ M = Λp (Tm M ∗ ),
m∈M

it comes with the natural projection π : Λp T ∗ M → M which sends each Λp (Tm M ∗ ) to


m.

72
Then for any chart of (U, ϕU ) of M , one defines a chart on Λp T ∗ U = Λp (Tm M ∗ )
S
m∈U
by setting:
ΨU (α) = (π(α), (Tp ϕ−1 ∗
U ) α).
As in the case of the cotangent bundle, one easily checks that it defines a smooth
structure on Λp T ∗ M such that π : Λp T ∗ M → M is a smooth vector bundle over M .
Definition 4.3.1. A smooth (differential) p-form on M is a smooth section Λp T ∗ M .
The vector space (in fact it isLalso a C ∞ (M ) module) of sections of Λp T ∗ M is denoted
by Ωp M , and Ω(M ) denotes p Ωp M .
Note that Ω0 M is just the vector space of smooth functions. If α ∈ Ωp M , we will
denote by αm its value on m ∈ M (which is an alternating linear form on Tm M ).
Example 4.3.2. Assume M is an open set U in Rn , then the bundle Λp T ∗ U is trivial,
and the p-forms:
m ∈ U 7→ dxi1 ∧ · · · ∧ dxip
for p-tuples such that i1 < · · · < ip gives a basis of Ωp U as a C ∞ (U )-module. In
particular any p-form α on U can be written as:
X
α= αi1 ...ip dxi1 ∧ · · · ∧ dxip ,
i1 <···<ip

where the αi1 ...ip are smooth functions on U . This shows that ΩU is generated by
smooth functions and their differential as en algebra.
If f : M1 → M2 is a smooth map and α ∈ Ωp M2 , we define the pull back of α by f by
the formula:
(f ∗ α)m1 = (Tm1 f )∗ αf (m1 ) .
f ∗ α is a smooth p-form on M2 .
Moreover, f ∗ (α ∧ β) = f ∗ α ∧ f ∗ β. If η ∈ Ω0 M = C ∞ (M ), f ∗ η = η ◦ f .
Remark 4.3.3. When computing in coordinates, the pull back is actually just a “change
of variables”. Assume that f : Rn → Rm and maps x = (x1 , . . . , xn ) to y(x) =
(y 1 (x1 , . . . , xn ), . . . , y m (x1 , . . . , xn )). Then a p-form α on Rm can be written as:
X
αy = αi1 ...ip (y 1 , . . . , y m )dy i1 ∧ · · · ∧ dy ip .
0≤i1 <···<ip ≤m

Then f ∗ α is easilyP computed by changing y i to y i (x1 , . . . , xn ) and computing the


differentials dy i as i ∂j y i dxj .
For instance consider f (r, θ, φ) = (r sin θ cos φ, r sin θ sin φ, r cos θ) (spherical coor-
dinates in R3 ), and α = dx ∧ dy ∧ dz, then:
f ∗ α =d(r sin θ cos φ) ∧ d(r sin θ sin φ) ∧ d(r cos θ)
=((sin θ cos ϕ)dr + (r cos θ cos ϕ)dθ − (r sin θ sin ϕ)dϕ)
∧ ((sin θ sin ϕ)dr + (r cos θ sin ϕ)dθ + (r sin θ cos ϕ)dϕ)
∧ (cos θdr − r sin θdθ)
=r2 sin θdr ∧ dθ ∧ dϕ.

73
Proposition 4.3.4. Let M be a compact manifold, then ΩM is generated by smooth
functions on M and their differential as an algebra.

Proof. We only need to show that any p-form α can be written as a sum of forms of the
type gdf1 ∧ · · · ∧ dfp . Consider a finite cover of M by charts (Ui ,P ϕi ), and ηi a partition
of unity subordinated to the cover Ui . Then if we set αi = ηi α, i αi = α. So we only
need to check the property for each αi . We know that the property is true on open sets
of Rn , so, in ϕi (Ui ) :
X
α̃i = (ϕ−1
i )∗
α i = (α̃i )k1 ...kp dxk1 ∧ · · · ∧ dxkp .
k1 ,...,kp

Thus on Ui : X
(α̃i )k1 ...kp ◦ ϕi ϕ∗i dxk1 ∧ · · · ∧ ϕ∗i dxkp .

αi =
k1 ,...,kp

But if we write ϕi = (ϕ1i , . . . , ϕni ), then ϕ∗i dxk = dϕki . So:


X k
(α̃i )k1 ...kp ◦ ϕi dϕki 1 ∧ · · · ∧ dϕi p .

αi =
k1 ,...,kp

We are almost done. The only problem here is that that the ϕki are only defined on Ui .
To overcome this, pick a smooth ψi which is 0 outside of Ui and 1 on the support of
ηi , then since αi vanishes outside of the support of ηi , we can safely replace the dϕki by
d(ψi ϕki ) and ψi ϕki extended by 0 outside of Ui is a smooth function on the whole M .
This result is useful in the following situation: assume we want to prove a linear for-
mula about differential forms, and that this formula behaves well under wedge products,
then it is enough to check it for smooth functions f and their differentials df .

4.3.1 The differential of a p-form.


Theorem 4.3.5. There exists a unique R-linear map: d : ΩM → ΩM such that:

1. d(Ωp M ) ⊂ Ωp+1 M ,

2. if f ∈ Ω0 M is a smooth function, then df coincides with the differential of f ,

3. if α ∈ Ωp M , d(α ∧ β) = dα ∧ β + (−1)p α ∧ dβ,

4. d ◦ d = 0.

We will prove the theorem in several steps.

Lemma 4.3.6. The theorem is true if M is an open set U ⊂ Rn .

74
Proof. We need to have that d(dxi ) = 0 by the last property. Then property 3 and an
elementary induction imply that d(dxi1 ∧ · · · ∧ dxip ) = 0. Write any α ∈ Ωp U as:
X
α= αi1 ...ip dxi1 ∧ · · · ∧ dxip ,
i1 <···<ip

Then property 3 and linearity imply:


X
dα = dαi1 ...ip ∧ dxi1 ∧ · · · ∧ dxip ,
i1 <···<ip

but αi1 ...ip is a smooth function, so by property 2 dαi1 ...ip is just the usual differential.
So the formula above is the only possible definition for d.
It is then a routine check to see that it satisfies the 4 required properties. Let us see
that d ◦ d = 0 (assumeing the other ones). We have that:
X
d2 α = d2 αi1 ...ip ∧ dxi1 ∧ · · · ∧ dxip ,
i1 <···<ip

so it is enough to check that d2 f = 0 for any smooth function. We have then that:
X X
d2 f = d( ∂i f dxi ) = ∂ji f dxi ∧ dxj .
i i,j

This is zero because dxi ∧ dxi = 0 and dxi ∧ dxj = −dxj ∧ dxi .
The following property is central, and shows that d is "natural".
Proposition 4.3.7. Let dM and dN be linear maps on ΩM and ΩN which satisfy the
assumptions f the theorem, and f : M → N be a smooth map. Then:

f ∗ ◦ dN = dM ◦ f ∗ .

Proof. First the formula is true if α = η ∈ Ω0 N = C ∞ (N ) because:

f ∗ dN η = f ∗ T η = T η ◦ T f = T (η ◦ f ) = dM (η ◦ f ) = dM f ∗ η.

Assume now that α = dη = dN η ∈ Ω1 N (dN coincides with the usual differential d on


smooth functions). Then dM (f ∗ dη) = dM (d(f ∗ η)) because η is a smooth function, hence
dM (f ∗ dη) = d2M (f ∗ η) = 0 since d2M = 0. Moreover f ∗ (dN dη) =∗ (d2N η) = 0. So:

f ∗ dN (dη) = dM f ∗ (dη).

Now using the fact that ΩN is generated as an algebra by smooth functions and their
differential and the formulas for f ∗ (α ∧ β) and d(α ∧ β), one can prove the formula for
every p-form.
Considering the special case where U ⊂ M is open and f : U → M is the inclusion,
we get:

75
Lemma 4.3.8. Let U ⊂ M be an open set, and dU , dM be operators on ΩU and ΩM
satisfying the hypothesis of the theorem. Then for any α ∈ ΩM :

(dM α)|U = dU (α|U ).

We can now prove the theorem.


Proof. The uniqueness of dM comes from the uniqueness on Rn . For any α ∈ ΩM and
any chart (U, ϕ), we must have, since ϕ is a diffeomorphism:

(dM α)|U = ϕ∗ d((ϕ−1 )∗ α),

since d is unique on ϕ(U ) ⊂ Rn , the right hand side is the unique possible definition of
dM on U .
To get existence, we need to show that the definition we gave does not depend on
charts, but if (V, ψ) is another chart:

ψ ∗ d((ψ −1 )∗ α) = (ψ ◦ ϕ−1 ◦ ϕ)∗ d((ϕ−1 ◦ ϕ ◦ ψ −1 )∗ α)


= ϕ∗ (ψ ◦ ϕ−1 )∗ d(ϕ ◦ ψ −1 )∗ (ϕ−1 )∗ α
= ϕ∗ d((ϕ−1 )∗ α)

since (ψ ◦ ϕ−1 )∗ d = d(ψ ◦ ϕ−1 )∗ .


Definition 4.3.9. The operator d whose existence and uniqueness is given by the pre-
vious theorem is called the exterior differential.

4.3.2 Lie derivative and Cartan’s formula


Definition 4.3.10. Let X be a vector field on M and α be a p-form on M , the Lie
derivative of α along X is the p-form defined by:
d
LX α = ϕ∗ α,
dt |t=0 t
where ϕt is the flow of X.
Proposition 4.3.11. LX is charaterised by the following properties:
1. for η ∈ Ω0 M = C ∞ (M ), LX η = dη(X),

2. LX ◦ d = d ◦ LX ,

3. LX (α ∧ β) = LX α ∧ β + α ∧ LX β.
Proof. Properties 1,2 and 3 follow from taking the time derivative of the similar prop-
erties of the pullback: ϕ∗t η = η ◦ ϕt , ϕ∗t d = dϕ∗t and ϕ∗t (α ∧ β) = ϕ∗t α ∧ ϕ∗t β.
To see that these properties charcterize LX , we just observe that properties 1 and 2
define LX and functions and their differentials and property 3 tells us how to compute
the Lie derivative of a wedge propduct.

76
We can actually compute the Lie derivative in terms of the Lie Bracket.

Proposition 4.3.12. Let X be a vector field on M , and α be a p-form on M . Then for


any smooth vector fields X1 , . . . , Xp , we have:
X
LX α(X1 , . . . , Xp ) = X · (α(X1 , . . . , Xn )) + α(X1 , . . . , Xi−1 , [Xi , X], Xi+1 , . . . , Xp ).
i

Proof. Consider the smooth function: f : m ∈ M 7→ αm ((X1 )m , . . . , (Xp )m ). Let ϕt be


the flow of X. Then:

f ◦ ϕt (m) =αϕt (m) ((X1 )ϕt (m) , . . . , (Xp )ϕt (m) )


=ϕ∗t αm (Tϕt (m) ϕ−t (X1ϕt (m) ), . . . , Tϕt (m) ϕ−t (Xp ϕt (m)) )
=ϕ∗t αm ((ϕ−t ∗ (X1 ))m , . . . , (ϕ−t ∗ (Xp ))m ).

Thus:
X d
X · f (m) =LX α(X1 , . . . , Xp ) + α(X1 , . . . , Xi−1 , ϕ−t ∗ Xi , Xi+1 , . . . , Xp )
i
dt |t=0
X
=LX α(X1 , . . . , Xp ) + α(X1 , . . . , Xi−1 , [Xi , −X], Xi+1 , . . . , Xp )
i

Remark 4.3.13. If we set LX Y = [X, Y ], which is justified by the fact that [X, Y ] =
d
ϕ Y = dtd |t=0 ϕ∗t Y , we get that:
dt |t=0 −t ∗

X · (α(Y, Z)) = LX α(Y, Z) + α(LX Y, Z) + α(Y, LX Z).

which should be seen as another occurence of Leibnitz rule.


We will need a third operation on differential forms, the interior product:

Definition 4.3.14. Let X be a vector field on M and α be a p-form, the interior


product of α with X is the p − 1-form ιX α defined by:

(ιX α)m (v1 , . . . , vp−1 ) = αm (Xm , v1 , . . . , vp−1 )

for any m ∈ M and v1 , . . . , vp−1 ∈ Tm M . If η is a smooth function, we set ιX η = 0.

The interior product and the wedge product satisfy are related by the following for-
mula:
ιX (α ∧ β) = ιX α ∧ β + (−1)k α ∧ ιX β.
The Lie derivative, interior product and exterior differential are linked by Cartan’s
formula.

Theorem 4.3.15. LX = d ◦ ιX + ιX ◦ d.

77
Proof. Set L̃X = d ◦ ιX + ιX ◦ d. We will show that L̃X satisfy the charcteristic properties
of LX from the previous proposition. For smooth functions η:

L̃X η = ιX (dη) = dη(X) = LX η.

Now, since d2 = 0: d ◦ L̃X = d ◦ ιX ◦ d = L̃X ◦ d. To end the proof we use the formula
for ιX (α ∧ β) together with the similar identity for d, this implies that L̃X (α ∧ β) =
L̃X α ∧ β + α ∧ L̃X β.
This allows us to define recursively d in without any reference to coordinates. We will
work this out only for 1-forms, this works in full generality but the formula gets messier.
Consider a 1-form α and two vector fields X and Y , we have:

dα(X, Y ) =ιX dα(Y )


=LX α(Y ) − d(ιX α)(Y )
=X · (α(Y )) − Y · (α(X)) − α([X, Y ]).

4.3.3 Closed forms, exact forms and Poicaré lemma


Definition 4.3.16. A p-form α is said to be closed if dα = 0.
Definition 4.3.17. A p-form α is said to be exact if there exist a (p − 1)- form β such
that dβ = α.
Since d2 = 0, any exact form is closed. However, the converse is not true.
Example 4.3.18. Consider the 1-form α on R2 \{0} given by:
−ydx + xdy
α= .
x2 + y 2
We have that:
−(2dx ∧ dy)(x2 + y 2 ) − (−ydx + xdy) ∧ (2xdx + 2ydy)
dα = = 0.
(x2 + y 2 )2

So α is closed. Assume that there is some smooth function η on R2 \{0} such that
dη = α, and consider the standard inclusion: i : S1 = {x2 + y 2 = 1} → R2 \{0}.
Then i∗ α = i∗ dη = d(i∗ η) = d(η ◦ i). In particular, since S1 is compact, i∗ α has to
vanish at some point. But for any p = (x, y) ∈ S1 , vp = (−y, x) ∈ Tp S1 and:

i∗ α(vp ) = α(vp ) = 1,

so that i∗ α never vanishes. Hence α is not closed.


The relation between closed and exact forms can actually be read on the topology of
M , through what is called de Rahm cohomology. We won’t develop this theory here,
however let us mention Poincaré lemma, which says that if the topology of M is simple,
any closed differential form is exact.

78
Theorem 4.3.19. Let O ⊂ Rn be a starshaped1 open set, then any closed differential
form on O is exact.
For the proof we will use the following property:
Lemma 4.3.20. Let ϕt be the (local) flow of some vector field X on U and α ∈ Ωp (U )
be a closed p-form, then for any t1 , t0 :

ϕ∗t1 α − ϕ∗t0 α

is an exact form.
Proof. We first write: Z t1
d ∗
ϕ∗t1 α − ϕ∗t0 α = ϕ αdt.
t0 dt t
Then:
d d
ϕ∗s α = ϕ∗t+s α = LX ϕ∗t α.
ds s=t ds s=0
We now use Cartan’s formula: LX ϕt α = d(ιX ϕ∗t α) + ιX d(ϕ∗t α). Thus, since d(ϕ∗t α) =

ϕ∗t dα = 0, we get:
Z t1 Z t1 
∗ ∗ ∗ ∗
ϕt1 α − ϕt0 α = d(ιX ϕt α)dt = d (ιX ϕt α)dt .
t0 t0

We can now prove the Poincaré Lemma:


Proof. We assume that O is starshaped with respect to the origin in Rn , and let α be
a closed k-form on U . Consider the radial vector field Xp = p, and let ϕt be its flow.
Then for any t < 0, ϕt maps U inside U . Moreover, by the previous lemma:

α − ϕ∗t α = dωt

where: Z 0
ωt = (ιX ϕ∗t α)dt.
t
It is now just a matter of easy computations to check that as t goes to −∞, ϕ∗t α goes
to 0 and ωt converges to some smooth form ω−∞ such that:

α = dω−∞ .

Remark 4.3.21. The topologically minded reader will notice that ϕt is actually an ho-
motopy between the constant map (at t = −∞) and the identity (at t = 0).
1
A domain O is starshaped if there exist o ∈ O such that for any x ∈ O, [o, x] ⊂ O.

79
4.4 Integration of differential forms
4.4.1 Orientability and volume forms
Let us first consider the case of finite dimensional vector spaces. When E is a vector space
over R, the group Gl(E) has two connected components, and the connected component
of the identity is given by:
Gl+ (E) = {g ∈ Gl(E)| det g > 0}.
The group Gl(E) acts transitively on the set of (ordered) basis of E by g · (v1 , . . . , vn ) =
(gv1 , . . . , gvn ). Two basis v = (v1 , . . . , vn ) and ṽ = (ṽ1 , . . . , v˜k ) of E are said to define
the same orientation if ṽ = g · v for some g ∈ Gl+ (E). Defining the same orientation
is an equivalence relation on the set of basis of E, which has exactly two equivalence
classes. For this reason, linear maps of Gl+ (E) are said to be orientation preserving.
Definition 4.4.1. An orientation of E is the choice of one equivalence class of basis
under the "defining the same orientation" relation. A vector space on which an orien-
tation has been given is said to be oriented.
Given an orientation of E, a basis is said to be positive if it belongs to the equivalence
class of basis that defines the orientation.
Recall that the space of n-forms on E has dimension 1. Let ω be a non zero element
in Λn E ∗ , then the set of basis v = (v1 , . . . , vn ) such that ω(v1 , . . . , vn ) > 0 defines an
orientation on E. This comes from the fact that if v and ṽ satisfy ω(v1 , . . . , vn ) > 0 and
ω(v˜1 , . . . , ṽn ) > 0, and we define f : E → E by setting f (vi ) = ṽi , then:
ω(v˜1 , . . . , ṽn ) = ω(f (v1 ), . . . , f (vn ))
= f ∗ ω(v1 , . . . , vn )
= det(f )ω(v1 , . . . , vn ).
To summarize, orienting a vector space is equivalent to choosing a nonzero n-form up
to scaling by a positive real. Now whenever we encounter the vector space Rn , we will
endow it with its orientation coming from the standard basis, or equivalently from the
n-form given by the determinant.
We now move to manifolds. Let U be an open set in a vector space E, and f : U → E
be a diffeomorphism. f is said to be orientation preserving if for any x ∈ E, Tx f ∈
Gl+ (E). For instance if f : Rn → Rn , it is equivalent to the positivity of the determinant
of the Jacobian matrix of f . The connectedness of Gl+ (E) implies that it is enough to
check that Tx f ∈ Gl+ (E) at only one point x.
Definition 4.4.2. An orientation atlas of a smooth manifold is a smooth atlas such
that all transition functions are orientation preserving diffeomorphisms. A manifold
which admits an orientation atlas is said to be orientable.
Two orientation atlases (U, ϕ) and (V, ψ) are said to be compatible if their union
is again an orientation atlas. Compatibility is an equivalence relation on orientation
atlases, on a given connected manifold their are exactly two classes for this relation.

80
Example 4.4.3. On Sn consider the atlas given by the stereographic projections around
x
the north and south pole. The transition function is: x 7→ kxk 2 , which is orientation

reversing. In particular this is not an orientation atlas. However, if we compose the


stereographic projection around the south pole with an orientation reversing map of
Rn , we get an orientation atlas. Thus Sn is orientable.

Definition 4.4.4. An orientation on an orientable manifold M is an equivalence class


of compatible orientation atlases.
An orientation atlas of an oriented manifold is an atlas which belongs to the equiv-
alence class which defines the orientation of M .

Given an orientation of M , one defines an orientation on each Tp M in the following


way. One first pick some orientation atlas in the given equivalence class of orientation at-
lases of M , and for each p ∈ M , one defines an orientation on Tp M by picking
 a chart from

the orientation atlas (U, ϕ) around p and by declaring that the basis ∂x∂ 1 |p , . . . , ∂x∂n |p
is positive. The fact that the chosen atlas is an orientation atlas implies that this ori-
entation doesn’t depend on the chosen chart. Similarly, the orientation won’t change if
we replace (U, ϕ) by a compatible atlas.
A volume form ω on an n-manifold M n is a smooth n-form such that ωp 6= 0 for
any p ∈ M . It is clear that the pulback of a volume form by a local diffeomorphism
is a volume form. Since the bundle Λn T ∗ M has rank 1, the existence of volume form
is equivalent to the triviality of Λn T ∗ M . In particular, once a volume form is chosen,
every n-form can be written as f ω for some smooth function f .

Theorem 4.4.5. A compact manifold M is orientable if and only if it admits a volume


form.

Remark 4.4.6. The theorem is also valid for second countable manifolds, however the
proof requires some topological subtleties.
Proof. First assume that M is orientable and choose a finite orientation atlas (ϕi , Ui ),
consider on each Ui the n-form ωi = ϕ∗i (dx1 ∧ · · · ∧ dxn ). If Uj intersects Ui , then at
p ∈ Ui ∩ Uj :

(ωj )p = ϕ∗j (dx1 ∧ · · · ∧ dxn )


= (ϕi ◦ ϕ−1 ∗ 1 n
i ◦ ϕj ) (dx ∧ · · · ∧ dx )
= (ϕ−1 ∗
i ◦ ϕj ) (ωi )p
= fij (p)(ωi )p ,

where fij is smooth and positive on Ui ∩ Uj . The fact that the atlas we are working with
is an orientation atlas imply that f (p) > 0. We now choose a partition of unity ηi and

81
P
set ω = i ηi ωi . Then for any p ∈ Uj :
X
ωp = ηi ωi
i
X
= ηj ωj + ηi ωi
{i6=j|p∈Ui }
X
= ηj ωj + ηi fji (p)ωj
{i6=j|p∈Ui }
 
X
= ηj (p) + ηi (p)fji (p) ωi .
{i6=j|p∈Ui }

The coefficient in front of ωi is positive, so ω is not 0 and is thus a volume form.


We now assume that we are given a volume form ω on M . Let (Ui , ϕi ) be a finite
atlas of M , for any i, we have on Ui that:

ω = fi ϕ∗i (dx1 ∧ · · · ∧ dxn ).

Where the smooth function fi doesn’t change sign. Exchanging two of the coordinates
xi if necessary, we can assume that fi > 0. Now we write:
1 −1 ∗ ∗ 1 fi
dx1 ∧ · · · ∧ dxn = (ϕj ) fi ϕi (dx ∧ · · · ∧ dxn ) = (ϕi ◦ ϕ−1 ∗ 1 n
j ) (dx ∧ · · · ∧ dx ).
fj fj

Since on n-forms, (ϕi ◦ϕ−1 ∗ −1


j ) is just multiplication by the jacobian determinant of ϕi ◦ϕj ,
the jacobian determinant has to be fj /fi > 0.
Example 4.4.7. Let ν = dx1 ∧ · · · ∧ dxn+1 be the standard volume form on Rn+1 , and
consider α = ιX ν, the interior product of ν with the vector field X given by Xp = p.
Consider then the usual inclusion i : Sn → Rn+1 , and let ω0 = i∗ α. One can check
that ω0 is never zero, and is thus a volume form on Sn .
ω0 has a special property (which actually charcterise it up to scaling): ω0 is invari-
ant under the action of the special orthogonal group SO(n + 1, R), that is for any
g ∈ SO(n + 1, R), g ∗ ω0 = ω0 . To prove that we first remak that g ∗ ν = det(g)ν = ν.
Now pick p ∈ Sn , and v1 , . . . , vn ∈ Tp Sn :

(g ∗ ω0 )p (v1 , . . . , vn ) = (ω0 )g(p) (Tp g(v1 ), . . . , Tp g(vn ))


= (i∗ ιX ν)g(p) (gv1 , . . . , gvn )
= νg(p) (Xg(p) , gv1 , . . . , gvn )
= νg(p) (gXp , v1 . . . , vn )
= (g ∗ ν)p (Xp , v1 , . . . , vn )
= νp (Xp , v1 , . . . , vn )
= (i∗ ιX ν)p (v1 , . . . , vn ) = ω0 (v1 , . . . , vn ).

Similarly if g ∈ O(n + 1)\SO(n + 1), g ∗ ω0 = −ω0 .

82
Proposition 4.4.8. RPn is orientable if and only if n is odd.
Proof. For this proof we will see RPn as the quotient Sn /{I, σ}, where σ : x 7→ −x, and
denote by π : Sn → RPn the projection map, which is a local diffeomorphism.
Let us first assume that n is even and that ω is a volume form on RPn . Then π ∗ ω is a
volume form on Sn , therefore there is a nowhere vanishing smooth function f : Sn → R
such that:
π ∗ ω = f ω0 ,
where ω0 is the volume form built in the previous example.
But π ◦ σ = π, so σ ∗ (π ∗ ω0 ) = π ∗ ω0 , hence:
f ω0 = σ ∗ (f ω0 ) = (f ◦ σ)σ ∗ ω0 = −(f ◦ σ)ω0 ,
because σ ∈ O(n + 1)\SO(n + 1) since n + 1 is odd.
In particular, f (−x) = −f (x) for all x ∈ Sn , so f take both positive and negative
values, and must vanish at some point, which contradicts the fact that f doesn’t vanish.
We now assume that n is odd. Consider an open set U ⊂ Sn where π is a diffeomor-
−1 ∗
phism. On π(U ), we can define a volume form by setting ωU = (π|U ) ω0 . Now assume
that V is another open set such that: π(U ) ∩ π(V ) is not empty, and pick some x̄ in the
intersection, now either x̄ = π(x) for some x ∈ U ∩ V , or x ∈ U and −x ∈ V . In the
former case, we trivially have that (ωU )x̄ = (ωV )x̄ . In the latter case, since σ ∗ ω0 = ω0
when the dimension is odd, we also get that: (ωU )x̄ = (ωV )x̄ . This shows that the n-form
ω defined by setting ω|U = ωU is actually a well defined volume form on RPn .

4.4.2 The integral of an n-form


When one inspects the change of variable formula for integrals of several variables func-
tions, it is clear that this notion is not invariant by diffeomorphisms (because of the
Jacobian determinant term). Therefore one cannot integrate functions on a smooth
manifold in a meaningful way without some extra structure.
However consider an n-form α on Rn with compact support ({p|αp 6= 0} has compact
closure) in some U ⊂ Rn . Then there is a unique smooth function f with compact
support in U such that:
α = f dx1 ∧ · · · ∧ dxn .
We define: Z Z
α= f dx1 . . . dxn ,
U U
where the right hand side is just the usual Riemann or Lebesgue integral of f .
This definition may seem purely conventional and highly unimpressive, but it turns
out to have the following important property:
Proposition 4.4.9. Let ϕ : U ⊂ Rn → ϕ(U ) ⊂ Rn be an orientation preserving
diffeomorphism and α be differential form with support in ϕ(U ), then:
Z Z

ϕ α= α.
U ϕ(U )

83
Proof. Let us write α as f dx1 ∧ dxn , then:
ϕ∗ α = (f ◦ ϕ)ϕ∗ dx1 ∧ · · · ∧ dxn = (f ◦ ϕ) det(Dϕ)dx1 ∧ · · · dxn .
In particular: Z Z

ϕ α= (f ◦ ϕ) det(Dϕ)dx1 . . . dxn .
U U
On the other hand, the usual change of variable formula for integrals of functions on Rd
tells us that:
Z Z Z
α= f dx . . . dx = (f ◦ ϕ)| det(Dϕ)|dx1 . . . dxn .
1 n
ϕ(U ) ϕ(U ) U

Now since ϕ is orientation preserving, det(Dϕ) > 0 and thus det(Dϕ) = | det(Dϕ)|
which gives the required equality.
Now consider an oriented manifold M . We will assume for simplicity that M is
compact. Take a finite orientation atlas (Ui , ϕi ) of M , and a partition of unitity ηi
suborditated to the cover Ui .
We set, for any n-form α on M :
Z XZ
α= (ϕ−1 ∗
i ) (ηi α).
M i ϕi (Ui )

Proposition 4.4.10. The expression above doesn’t depend on the chosen orientation
atlas and partition of unity.
P
Proof. Let us first notice that α = i (ηi α), and that each αi = ηi α is compactly
supported in Ui .
Now let (Vj , ψj ) be another orientation atlas and νi and associated partition of unity.
We want to check that:
XZ XZ
−1 ∗
(ϕi ) (ηi α) = (ψj−1 )∗ (νj α).
i ϕi (Ui ) j ψj (Vj )

We have:
XZ XZ X
(ϕ−1 ∗
i ) (ηi α) = (ϕ−1
i )∗
(( νj )ηi α) (4.1)
i ϕi (Ui ) i ϕi (Ui ) j
XZ
= (ϕ−1 ∗
i ) (νj ηi α). (4.2)
i,j ϕi (Ui ∩Vj )

Now ϕi ◦ ψj−1 is an orientation preserving diffeomorphism, so:


Z Z
−1 ∗
(ϕi ) (νj ηi α) = (ϕi ◦ ψj−1 )∗ (ϕ−1 ∗
i ) (ηi νj α)
ϕi (Ui ∩Vj ) ψ (U ∩V )
Z j i j
= (ψj−1 )∗ (ηi νj α).
ψj (Ui ∩Vj )

84
Plugging this back in equation (4.1), we get:
XZ XZ
−1 ∗
(ϕi ) (ηi α) = (ψj−1 )∗ (ηi νj α).
i ϕi (Ui ) i,j ψj (Ui ∩Vj )
XZ X
= (ψj−1 )∗ (( ηi )νj α)
j ψj (Vj ) i
XZ
= (ψj−1 )∗ (νj α),
j ψj (Vj )

which is the required equality.


R
Definition 4.4.11. M α is the integral of the n-form α.
We will also need to integrate differential forms on compact subsets K ⊂ M . We will
define: Z XZ
α= (ϕ−1 ∗
i ) (ηi α).
K i ϕi (Ui ∩K)

One shows as before that this depends neither on the orientation atlas, nor on the
partition of unity that we have used.
Volume forms play the role of non vanishing functions:

RProposition 4.4.12. If ω is a volume form compatible with the orientation of M then


M
ω > 0.
Proof. In each chart (Ui , ϕi ) of an orientation atlas, we know that (ϕ−1 ∗ 1
i ) ω = fi dx ∧
· · · ∧ dxn for some smooth positive function fi on ϕi (Ui ), so:
XZ XZ
−1 ∗
(ϕi ) (ηi α) = ηi fi dx1 . . . dxn .
i ϕi (Ui ) i ϕi (Ui )

Each of the terms in the sum are nonnegative, and except if ηi is vanishing, they are
actually positive, since at least one of the ηi doesn’t vanish, the ntegral is positive.
This framework gives a really geometric interpretation of the divergence of a vector
field. Consider a manifold M with a volume form ω, and X a vector field on M . Consider
the flow ϕt of X. The question we want to answer is the following: how
R does the volume
of Kt = ϕt (K) evolves with respect to t ? Here by volume we mean Kt ω? To see this
we write: Z Z
ω= ϕ∗t ω.
Kt =ϕt (K) K

R the right hand side with respect to t at t = 0, differentiating


Now we want to differentiate
under the integral we get K LX ω. Now LX ω is an n form, so LX ω = f ω for some smooth
f , f is called the divergence of X (with respect to the volume form ω), in other words
LX ω = div(X)ω is the definition of the divergence in this context. From a more physical
point of view, div(X) measures the change of volume when one moves along X.

85
Let us now assume that M = Rn with the standard volume form ω = dx1 ∧ · · · ∧ dxn ,
and that X = X i ∂i is a vector field. We want to compute the divergence of X in terms of
its components X i . We use Cartan’s formula, which, since dω = 0 (dω is an n + 1-form!)
writes as: X
LX ω = d(ιX ω) = d(X i ι∂i ω).
i

Now:
ι∂i ω = (−1)i dx1 ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn ,
and d(X i ) = ∂j X i dxj . Using that dι∂i ω = 0, we get:
P
j
!
X
i
LX ω = d(ιX ω) = ∂i X ω.
i

∂i X i , the usual formula.


P
Hence div(X) = i

4.5 Stokes Theorem


4.5.1 Regular domains
R R
Stokes theorem reads as “ D dω = ∂D ω”, before getting to this, we need to understand
who D and ∂D are.
When D is a subset of M , we will denote by D̄ it closure, D̊ its interior, and ∂D = D̄\D̊
its boundary.

Definition 4.5.1. A connected compact D in a manifold M n is said to be a regular


domain if it is equal to the closure of its interior, and its boundary ∂D is a smooth
submanifold of dimension n − 1 in M .

Example 4.5.2. If M is a compact smooth manifold, then M itself is a regular domain


(with empty boundary).
The unit ball in Rn is a regular domain with boundary the unit sphere Sn−1 .
The annulus Ar1 ,r2 = {x ∈ Rn |r1 ≤ kxk ≤ r2 } is a regular domain, its boundary is
the union of the spheres of radius r1 and r2 .
The square in the plane is not a regular domain: its boundary is not a smooth
submanifold.

Proposition 4.5.3. For each p ∈ ∂D there is a chart (U, ϕ) of M such that ϕ(U ∩∂D) =
{x ∈ ϕ(U )|x1 = 0} and ϕ(U ∩ D) = {x ∈ ϕ(U )|x1 ≤ 0}.

Proof. The first point is just the fact that ∂D is submanifold of M . We may assume
that ϕ(U ) is the unit ball B in Rn . Consider the open sets O = ϕ(U ∩ D̊) and O0 =
B\{x1 = 0}. We have that O ⊂ O0 and, since D is a regular domain, the frontier of O0
as a subset of O0 is empty. Thus O is closed and open in O0 . Hence O has to be a union
of connected components of O0 , which are B ∩ {x1 < 0} and B ∩ {x1 > 0}.

86
However, we know that the frontier of O in B is not empty (and equal to {x1 = 0}),
so it has to be one of the connected components. Therefore, composing ϕ with the
symmetry with respect to {x1 = 0} if necessary, we get that: ϕ(U ∩ D̊) = B ∩ {x1 < 0}.
Taking the closure we have ϕ(U ∩ D) = B ∩ {x1 ≤ 0}.
We want to integrate n-forms on D ⊂ M , so we need M to be orientable. We also
want to integrate n − 1 forms on ∂D, so we also need ∂D to be orientable. This is
actually automatic.

Proposition 4.5.4. Let D be a regular domain in an orientable manifold, then ∂D is


orientable.

Proof. We consider a chart (U, ϕ) as in the previous lemma. Changing the x2 coordinate
to −x2 if necessary, we can ensure that (U, ϕ) is compatible with orientation of M .
We denote by ϕ = (x1 , . . . , xn ) the coordinates of ϕ. We consider the smooth map:
ϕ̃ : Ũ = ∂D ∩ U → Rn−1 given by ϕ̃(p) = (x2 (p), . . . , xn (p)). We will show that these ϕ̃
form an orientation atlas of ∂D.
Consider another orientation chart of M from the lemma (V, ψ), and denote by
ψ = (y 1 , . . . , y n ) the coordinates of ψ. The change of chart ψ ◦ϕ−1is then the map
j
(x1 , . . . , xn ) 7→ (y 1 , . . . , y n ). Its Jacobian matrix is thus given by ∂y
∂xi
.
ij
Moreover, since ϕ(∂D∩U ) = {x1 = 0} and ψ(∂D∩V ) = {y 1 = 0}, ψ◦ϕ−1 (0, x2 , . . . , xn ) =
1
(0, y 2 , . . . , y n ), so ∂y
∂xi
(0, x2 , . . . , xn ) = 0 for i ≥ 2. Also, using that ϕ(D ∩ U ) = {x1 ≤ 0}
and ψ(D ∩ U ) = {y ≤ 0}, we get that y 1 (x1 , . . . , xn ) ≥ 0 when x1 ≥ 0, in particular
1
∂y 1
(using that ψ ◦ ϕ−1 is a diffeomorphism) we get that ∂x 2 n
1 (0, x , . . . , x ) > 0.

In particular tthe Jacobian matrix of ψ ◦ ϕ−1 is of the form:


 ∂y1 
∂x 1 0 · · · 0
 ∂y2 ∂y2 · · · ∂y2 
 ∂x1 ∂x2 ∂xn 
 . . .. 
 .. .. . 
∂y n ∂y n ∂y n
∂x1 ∂x2
··· ∂xn
 
∂y j
and at points (0, x2 , . . . , xn ), the determinant of the submatrix ∂xi
is positive.
i,j≥2
To end the proof, we just remark that ψ̃ ◦ ϕ̃−1 maps (x2 , . . . , xn ) to (y , . . . , y n ), and
2

we just observed that the Jacobain of this map is positive.


The proof not only shows that ∂D is orientable, but also gives it an orientation (the
one defined by the atlas of all the (Ũ , ϕ̃). This orientation is the one coming from an
"outward normal". To see this we work in one of the chart of M (U, ϕ) considered in
the proof. A basis of the tangent space at each p ∈ U , and we consider the vector field
X on U given by (ϕ−1 ∂
∗ ∂x1 , notice that on ∂D, X is pointing towards the outside of D.
The orientation on ∂D can then be defined in the following way: A basis (v2 , . . . , vn ) of
Tp ∂D is positive if and only (Xp , v2 , . . . , vn ) is a positive basis of Tp M .

87
D

Figure 4.1: Orienting ∂D.

For example if D is the unit disk, the oriention on the boundary S1 is the "anti
clockwise" one. Furthermore, if we remove a disk from the interior of D, the orientation
of the inner component of the boundary will be reversed.
An important particular case is when D = [a, b] ⊂ R, then ∂[a, b] = {a} ∪ {b}, a
union of two 0-dimensional submanifolds. A 0-form on a point is just a constant, so
a orientation is just the choice of a sign (+ or −). On [a, b], consider the orientation
given by the 1-form dx. The vector X+ = ∂x is outward pointing at b, the proof above
shows that the orientation on the boundary is given by the volume form ιX ω, where X
is an outward pointing vector and ω is a volume form in M . So the orientation on {b}
is given by ιX dx = 1, so is +. Similarly, since −∂x is an outard pointing vector at a, the
orientation on {a} is −. This explains the notation ∂[a, b] = {b} − {a}.

4.5.2 Stokes Theorem


Theorem 4.5.5. Let D be a regular domain in a smooth oriented n-manifold M and α
be an n − 1-form on M . We denote by i : ∂D → M the inclusion. Then:
Z Z
dα = i∗ α.
D ∂D

This is a generalization of the fundamental theorem of calculus. Let f be a smooth


function on R and α = f (x) seen as a 0 form. Then:
Z Z Z
0
f (x)dx = dα = α = f (b) − f (a).
[a,b] [a,b] ∂[a,b]

Another important case is the case when D = M where M is a compact manifold,


thus the boundary of D is empty and:
Z
dα = 0.
M

88
In other words the integral of an exact n-form on a compact manifold is zero. As a
corollary, we get that a volume form is never exact.
Proof. Using a partition of unity, it is enough to show the result for compactly supported
forms whose support is included in some chart.
We have three cases to deal with. First if the support of α doesn’t meet D, the result
is trivial. R
R of α is included in the interior D. Then ∂D α = 0,
Second, assume that the support
and we only need to show that D dα = 0. Let (U, ϕ) be a chart such that the support
of α is included in U , then:
Z Z Z
dα = dα = d(ϕ−1 )∗ α
D U Rn

Since (ϕ−1 )∗ α has support in ϕ(U ). We set β = (ϕ−1 )∗ α.


Let η k = dx1 ∧ · · · ∧ dxk−1 ∧ dxk+1 ∧ · · · dxn . The η k form a basis
P of Ωk
n−1 n
R as a
∞ n
C (R ) module. So there are smooth functions fk such that β = k fk η . Moreover
each fk is compactly supported and:
X
dβ = ( (−1)k+1 ∂k fk )dx1 ∧ · · · ∧ dxn .
k
Z XZ
dβ = (−1)k+1 ∂k fk dx1 . . . dxn .
{x1 ≤0} k {x1 ≤0}

Now for any k:


Z Z Z
1 n
∂k fk dx . . . dx = ( ∂k fk dxk )dx1 . . . dxk−1 dxk+1 . . . dxn = 0,
Rn Rn−1 R

since fk is compactly supported. This comes form the fundamental theorem of calculus:
Z Z
1 n k
∂k fk (x , . . . , x )dx = ∂k fk (x1 , . . . , xn )dxk
R [−A,A]

=fk (x1 , . . . , xk−1 , A, xk+1 , . . . , xn )


− fk (x1 , . . . , xk−1 , −A, xk+1 , . . . , xn )
=0.
R
for A big enough. Thus Rn dβ = 0.
The last case is the case when the support of α meets ∂D. Using another partition of
unity, we can assume that the support of α is contained in a chart given by Proposition
4.5.3. Let (U, ϕ) be such a chart. We can assume withouth loss of generality that ϕ(U )
is included the unit ball B in Rn and that ϕ(D) = {x1 ≤ 0}. Then:
Z Z Z
−1 ∗
dα = (ϕ ) dα = d(ϕ−1 )∗ α.
D ϕ(D) {x1 ≤0}

89
Similarly, using the orientation charts for ∂D built in the previous section:
Z Z Z
∗ −1 ∗ ∗
i α= (ϕ̃ ) i α = ī∗ (ϕ−1 )∗ α
∂D {x1 =0} {x1 =0}

whereR ī is the inclusion of {x1 = 0} in Rn . Setting β = (ϕ−1 )∗ α, we only need to show


that {x1 ≤0} dβ = {x1 =0} ī∗ β.
R

With the same notation as the previous case, we write β = k fk η k . We get:


P

Z XZ
dβ = (−1)k+1 ∂k fk dx1 . . . dxn .
{x1 ≤0} i {x1 ≤0}

If k 6= 1:
Z Z Z
1 n
∂k fk dx . . . dx = ( ∂k fk dxk )dx1 . . . dxk−1 dxk+1 . . . dxn = 0,
{x1 ≤0} Rn−1 R

since fk is compactly supported.


Now for k = 1, we have:
Z Z Z
1 n
∂1 f1 dx . . . dx = ( ∂1 f1 dx1 )dx2 . . . dxn
{x1 ≤0} Rn−1 {x1 ≤0}
Z Z
= ( ∂1 f1 dx1 )dx2 . . . dxn
Rn−1 [−A,0]
Z
= (f1 (0, x2 , . . . , xn ) − f1 (−A, x2 , . . . , xn ))dx2 . . . dxn
n−1
ZR
= f1 (0, x2 , . . . , xn )dx2 . . . dxn
n−1
ZR
= ī∗ (f1 η1 ).
{x1 =0}

Moreover {x1 =0} fk ī∗ ηk = 0 for k 6= 1, because ī∗ ηk = 0 in this case. Putting all these
R

together we finally get: Z Z


dβ = ī∗ β.
{x1 ≤0} {x1 =0}

4.5.3 Some vector calculus revisited


Let us derive a couple of consequences. First we have the Green Riemann formula:
Corollary 4.5.6. Let D ⊂ R2 be a regular domain and P, Q be smooth function on R2 .
Then: Z Z
∂Q ∂P
− dxdy = P dx + Qdy.
D ∂x ∂y ∂D

90
Proof. We just compute:
∂Q ∂P
d(P dx + Qdy) = dP ∧ dx + dQ ∧ dy = − dx ∧ dy
∂x ∂y

Let us look at the divergence theorem.


Corollary 4.5.7. Let X be a vector field on a compact oriented manifold M and ω be
a volume form, then: Z
div(X)ω = 0.
M
Proof. By definition, div(X)ω = LX ω. Now, using Cartan’s formula:
LX ω = d(ιX ω) + ιX (dω) = d(ιX ω).
In particular div(X)ω is an exact n-form, so its integral on M is zero by Stokes Theorem.

What happens when we integrate on a regular domain D ? Stokes theorem gives:


Z Z
div(X)ω = ιX ω.
D ∂D
n
Let us assume that D is a regular domain in R and that ω is the usual volume form
dx1 ∧ · · · ∧ dxn . The left hand side of the integral is just the integral of the real valued
function div(X) on D. To understand the right hand side, we need a definition:
Definition 4.5.8. Let D ⊂ Rn be a regular domain. The canonical volume form of
∂D is the the n − 1-form on ∂D given by σ = i∗ ιN ω, where N is a unit outward normal
vector field on ∂D and i : ∂D → Rn is the inclusion.
Remark 4.5.9. We have used in a crucial way the euclidean structure here, so this defini-
tion doesn’t make sense for smooth manifolds, however it makes sense for Riemannian
manifolds.
Using this canonical volume form, we can integrate function on ∂D. In R2 , ∂D is an
oriented curve and we get the concept of line integral, in R3 , ∂D is an oriented surface
and we get the concept of surface integral.
Now, let us consider p ∈ ∂M and v1 , . . . , vn−1 be vectors in Tp ∂D, then:
(ιX ω)p (v1 , . . . , vn−1 ) = ωp (Xp , v1 , . . . , vn−1 )
= ω(hXp , Np i Np + (Xp − hXp , Np i Np ), v1 , . . . , vn−1 )
= hXp , Np i ω(Np , v1 , . . . , vn−1 )
= hXp , Np i σ(v1 , . . . , vn−1 )
since (Xp −hXp , Np i Np ) belong to Tp ∂D, the space of the vi ’s. So that i∗ ιX ω = hX, N i σ.
To summarize we get the equality:
Z Z
div(X)ω = hX, N i σ.
D ∂D
The right hand is called the flux of X through ∂D.

91
4.6 Exercises
Exercise 4.1. Let M be a smooth manifold. Recall that the space of vector fields
Γ(T M ) is a C ∞ (M ) module. Consider a map µ : Γ(T M ) → C ∞ (M ) which is C ∞ M
linear (µ(f X + gY ) = f µ(X) + gµ(Y ) for f, g ∈ C ∞ (M ) and X, Y ∈ Γ(T M )). Show
that there exist a unique 1-form α such that for any vector field X:

µ(X)(p) = αp (Xp ).

consider the case where M is


First use a partition of unity to show that it is enough toP
some open set in R . Second decompose any X as a sum i X i ∂i to see what α should
n

be.
Exercise 4.2. Let α = xdy + ydx, show that α is exact. Show that β = xdy − ydx is
not exact.
Exercise 4.3. * Consider a one form α on Rn . Write α = i gi dxi and assume that
P
for i 6= j, ∂i gj = ∂j gi . Set:
X Z 1
f (x) = xi gi (tx)dt.
i 0

Show that df = α. R1 d
Show that ∂j f (x) = 0 dt
(tgj (tx))dt.
Exercise 4.4. Consider a smooth curve c : [a, b] → M , and α a 1-form on M , we define
the integral of α along c by:
Z Z b
α= αc(t) (c0 (t))dt.
c a

1. Let ϕ : [c, d] → [a, b] be a smooth increasing diffeomorphism and c̃ = c ◦ ϕ. Show


that: Z Z
α = α.
c̃ c

2. Assume that α = df for some smooth function f on M , show that:


Z
α = f (c(b)) − f (c(a)).
c

3. Consider the smooth 1-form on R2 \{0} given by:


ydx − xdy
α= .
x2 + y 2
Show that α is not exact on R2 \{0}.

4. Show that however α is exact on U = {(x, y)|y > 0}. (Hint:α is sometimes denoted
by dθ.)

92
Exercise 4.5. An alternating p-form α ∈ Λp E ∗ is said to be simple if it can be written
as the wedge product of p linear forms.
1. Show that any n or n − 1 alternating for is simple.
2. Show that α = θ ∧ α0 for some linear form θ if and only if θ ∧ α = 0.
3. Let α, β, γ, δ be linearly independent linear forms. Show that ω = α ∧ β + γ ∧ δ is
not simple.
4. Show that ω ∧ ω 6= 0.
∗ ∗
Exercise 4.6. ** Let E = Λ2 R4 . And let ω be a non-zero element in Λ4 R4 .

1. For any u, v ∈ E, u ∧ v ∈ Λ4 R4 , hence there exist B(u, v) ∈ R such that: u ∧ v =
B(u, v)ω. Show that (u, v) 7→ B(u, v) is a symmetric bilinear form. What is it
signature ?
2. For any g ∈ Sl4 (R), show that B(u, v) = B(g ∗ u, g ∗ v).
3. Show that O0 (3, 3) is isomorphic to a quotient of Sl4 (R).
Exercise 4.7. Any 2-form ω ∈ Λ2 E ∗ defines a map ϕ : E → E ∗ by ϕ(x) = (y 7→
ω(x, y)). ω is said to be non degenerate if ϕ is an isomorphism.
1. Show that if E has odd dimension, then any 2 form is degenerate.
2. On E = R2n , set:
α = dx1 ∧ dxn+1 + · · · + dxn ∧ dx2n .
Show that α is nondegenerate.
3. Set Sp(n, R) = {u ∈ Gl2n (R)|∀x, y ω(x, y) = ω(u(x), u(y))}. Show that u ∈
Sp(n, R) if and only if its matrix A satisfies J = AT JA, where:
 
0 In
J= .
−In 0

4. Show that Sp(n, R) is a Lie group, and determine its Lie algebra.
A non degenerate linear 2-form is called a (linear) symplectic form, the one we have
studied here happens to be the only one (up to a change of basis). Sp(n, R) is called the
symplectic group.
Exercise 4.8. Let F (r, θ) = (r cos θ, r sin θ), compute F ∗ (dx ∧ dy), F ∗ (xdx + ydy) and
F ∗ (ydx − xdy).
Exercise 4.9. Let α be a 1-form on a smooth manifold M , X be a smooth vector field
and ϕ : M → M be a diffeomorphism. Show that:
ϕ∗ (α(X)) = ϕ∗ α(ϕ∗ X),
where we have set ϕ∗ X = (ϕ−1 )∗ X.

93
Exercise 4.10. We work on M = R2 . A differential form α on R2 is said to be SO(2, R)
invariant if g ∗ α = α for any g ∈ SO(2, R)

1. Let α = xdx+ydy and β = ydx−xdy. Show that α and β are SO(2, R)) invariant.

2. Let γ be an SO(2) invariant 1-form on R2 \{0}. Show that there exist two smooth
functions f, g : R∗+ → R such that:

γ = f (r)α + g(r)β,
p
where r = x2 + y 2 .

3. What are the SO(2) invariant 1-forms on R2 .

Exercise 4.11. For E a vector space, we denote by S 2 E ∗ the space of symmetric bilinear
forms on E. If f ∈ L(E, F ) and B ∈ S 2 F ∗ we set:

f ∗ B(u, v) = B(f (u), f (v)),

so that f ∗ B ∈ S 2 E ∗ .
As in the case of say 2-forms, the set formed by the union of the S 2 (Tp M )∗ is a smooth
vector bundle whose charts are given by:

B ∈ S 2 (Tp M )∗ 7→ (ϕ(p), ((Tp ϕ)−1 )∗ B).

A Riemannian metric on M is a smooth section g of S 2 T ∗ M such that gp is an


inner product on Tp M for every p, a manifold with a Riemannian metric is called a
Riemannian manifold. Let (M, g) be a Riemannian manifold.

1. Show that if f : N → M is an immersion, then f ∗ g is a Riemannian metric on N .

2. Show that every compact manifold admits a Riemannian metric.

3. Show that for any smooth vector field X on M , αp : vp ∈ T M 7→ gp (Xp , vp ) is a


smooth 1-form on M (usually denoted by X [ )

4. Show that for any 1-form α on M , there exist a smooth vector field α# on M such
that for any vp ∈ Tp M :
gp (α# , vp ) = α(X).
When α = df , df # is called the gradient of f and usually denoted by ∇f .

Exercise 4.12. We work on R3 , with the volume form ω = dx ∧ dy ∧ dy, and inner
product h , i, which actually defines a Riemannian metric on R3 .

1. Let X = i X i ∂i be smooth vector field on R3 . Compute X [ in terms of X i and


P
dxi .

94
2. Let X be a vector field on R3 . Show that rot(X) is charcterized by:

ιrot(X) ω = d(X [ ),

and that div(X) is characterized by:

div(X)ω = d(ιX ω).

3. Show, using the properties of d, that rot(∇f ) = 0 and div(rot X) = 0.


Exercise 4.13. Show that on p-forms:

LX (LY α) − LY (LX α) = L[X,Y ] α

and:
LX (ιY α) − ιY (LX α) = ι[X,Y ] α.
Exercise 4.14. Let S be an hypersurface in U ⊂ Rn which is given by S = f −1 (0)
where f : U → R is a submersion. Show that S is orientable.
Show that at any point p ∈ S, ∇f is orthogonal to Tp S. Use it to build a volume form
on S from a volume form on U .
Exercise 4.15. * Let G be a connected Lie group. We denote by (Ωp (G))` the vec-
tor space of p-forms α on G such that )Lg )∗ α = α for every g ∈ G (or equivalently
(Rg−1 )∗ α = α). Those p-forms are called left invariant.
1. Show that α ∈ (Ωp (G))` 7→ αe ∈ Λp g is a vector space isomorphism.

2. Show that any Lie group is orientable.

3. Let α ∈ Ω1 (G)` . Show that for any two left invariant vector fields:

dα(X, Y ) = α([X, Y ]).

Use that p 7→ αp (Xp ) is constant.

4. Derive a similar formula for left invariant 2-forms.


What we see here is that on Left invariant forms, the d operator can be recovered from
the Lie bracket, and thus from the data of the Lie algebra of G only.
Exercise 4.16. * We use the same notations as in the previous exercise.
1. Let ω be a volume form in (Ωn (G))` . Show that, for any compactly supported f
and any g ∈ G: Z Z
(f ◦ Lg )ω = f ω.
G G

2. Show that for any g ∈ G, (Rg )∗ ω ∈ (Ωn (G))` , and that there exists a real mod(g)
such (Rg )∗ ω = mod(g)ω.

95
3. Show that mod is morphism from G to (R∗ , ×).
4. Show that if G is compact and connected, for any g ∈ G, Rg ∗ ω = ω.
ω is thus left and right invariant, it is called bi-invariant.
 
a b
5. By considering the subgroup of Gl2 (R) of matrices of the form , show that
0 1
this property can fail when G is not compact.
6. Consider the morphism G → Gl(G) given by g 7→ Adg , where Adg (X) = (Rg )∗ X.
Show that ω is right invariant if and only if det(Adg ) = 1 for every g.
7. ** Show that G admits a bi-invariant volume form if and only the linear map
adX : Y ∈ g 7→ [X, Y ] ∈ g has zero trace for any X ∈ g.
The left invariant volume form on G is unique up to scaling, it is called the Haar
measure on G. The last question show that the property that G admits a bi-invariant
volume form can actually be read from the Lie algebra of G.
Exercise 4.17. Let G be a compact Lie subgroup of Gln (R). Let ω be a Haar measure
on G.
1. Show that, for v, w ∈ Rn :
Z
hv, wiG = hgv, gwi ω
G

defines an inner product on Rn .


2. Show that for any g ∈ G:
hgv, gwiG = hv, wiG .

3. Show that there exist h ∈ Gln (R) such that:


hGh−1 ⊂ O(n)

Exercise 4.18. Consider a surface S in R3 and Σ be a regular domain in S. We denote


by C the boundary of Σ.
1. Let c : I → R3 be a smooth curve. Show that:
Z Z
[
Xc(t) , c0 (t) dt.


X =
c I
H
The right hand side is called the circulation of X around c, denoted by c
X.
2. Show the following corollary of Stokes theorem:
Z I
hrot(X), N i σ = X,
Σ c

Namely, the flux of rot(X) along Σ is equal to the circulation of X along ∂Σ

96
Exercise 4.19. We want to compute the volumes of spheres and balls. We consider
Sn ⊂ Rn+1 . Let ω = dx1 ∧ · · · ∧ dxn+1 and σ = ιX ω|Sn be the canonical volume form on
Sn , where Xp = p is the
R radial vector field.
n
We set Voln (S ) = Sn σ.

1. Let F : R∗+ × Sn → Rn+1 \{0} be defined by: F (ru) = ru. Show that F ∗ ω =
rn dr ∧ σ.
2
2. By computing Rn+1 e−|x| dx1 . . . dxn in two different ways, show that:
R

Z n Z
−t2 2
e dt = Voln S n
e−r rn dr.
R R+

3. Show that: n+1


n π 2
Voln S = 2 n+1
,
Γ 2

e−s ts−1 ds.


R
where Γ(t) R+

4. Show that (n + 1)ω = d(ιX ω).

5. Use Stokes theorem to show that:


Z
1
ω = Vol B n+1 = Vol Sn ,
B n+1 n+1

where B n+1 is the unit ball in Rn+1 .

97
Bibliography
[Aud03] Michèle Audin. Geometry. Transl. from the French. Berlin: Springer, 2003.

[Ber09] Marcel Berger. Geometry. I, II. Transl. from the French by M. Cole and S.
Levy. Corrected 4th printing. Berlin: Springer, 2009.

[Boo86] William M. Boothby. An introduction to differentiable manifolds and Rieman-


nian geometry. 2nd ed. , 1986.

[GHL04] Sylvestre Gallot, Dominique Hulin, and Jacques Lafontaine. Riemannian ge-
ometry. 3rd ed. Berlin: Springer, 2004.

[HS74] Morris W Hirsch and Stephen Smale. Differential equations, dynamical sys-
tems, and linear algebra, 1974.

[Jos11] Jürgen Jost. Riemannian geometry and geometric analysis. 6th ed. Berlin:
Springer, 2011.

[Laf96] Jacques Lafontaine. Introduction to differentiable manifolds. (Introduction


aux variétés différentielles.). Grenoble: Presses Universitaires de Grenoble;
Les Ulis: EDP Sciences, 1996.

[Lee02] John M. Lee. Introduction to smooth manifolds. New York, NY: Springer,
2002.

[nMM13] Pedro M. Gadea; Jaime Mu noz Masqué and Ihor V. Mykytyuk. Analysis and
algebra on differentiable manifolds: a workbook for students and teachers. 2nd
revised ed. London: Springer, 2nd revised ed. edition, 2013.

[Spi99] Michael Spivak. A comprehensive introduction to differential geometry. Vol.


1-5. 3rd ed. with corrections. Houston, TX: Publish or Perish, 1999.

[Sti08] John Stillwell. Naive Lie theory. New York, NY: Springer, 2008.

98

Potrebbero piacerti anche