Barilari GeoDiff

Lecture notes on
Differential Geometry
davide barilari
March 2, 2023
University of Padova, Academic Year 2020/21
Contents
Contents 2
1 Smooth manifolds and smooth maps 7

1.1 Topological and smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Some fundamental examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Smooth maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 Bump functions, partition of unity and paracompactness . . . . . . . . . . . . . . . . 19
1.5 Appendix: Brower invariance of domain . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 Tangent space and differentials 25

2.1 Tangent space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 A bit of geometry: tangent vector and curves . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Immersions, embeddings. Submanifolds 33

3.1 Immersions, submersions, embeddings . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 The constant rank theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Tangent bundle and vector fields 43

4.1 The tangent bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.4 Lie brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5 Integral curves and flows 51

5.1 Integral curves and flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2 Lie derivatives and Lie brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Lie brackets and commutativity of flows . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Left-invariant vector fields on a Lie group . . . . . . . . . . . . . . . . . . . . . . . . 58
6 Vector distributions: integrability vs non-integrability 61

6.1 Diffeomorphisms built with flows of vector fields . . . . . . . . . . . . . . . . . . . . 61
6.2 Rectification of vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2
6.3 Frobenius theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.4 Rashevski-Chow theorem: local version . . . . . . . . . . . . . . . . . . . . . . . . . 66
7 Tensors and Differential forms 71

7.1 Cotangent space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.2 Tensors and tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.3 Differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.4 Lie derivatives of tensors and differential forms . . . . . . . . . . . . . . . . . . . . . 85
8 Orientation, Integration on manifolds 89

8.1 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.2 Integration on manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.3 Manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
8.4 Stokes theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9 Riemannian manifolds 109

9.1 Riemannian structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
9.2 The metric structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
9.3 Length-minimizers and geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.4 Appendix: on the Euler-Lagrange equations . . . . . . . . . . . . . . . . . . . . . . . 119
10 Connections, parallel transport and curvature 121

10.1 Affine connections and parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . 121
10.2 The Levi-Civita connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
10.3 The Riemann curvature tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
A Problems and Exercises 133
Bibliography 145
3
Prelude
Differential Geometry is a vast subject, whose very first goal is to introduce instruments to develop
differential and integral calculus on manifolds, i.e., smooth spaces which are not necessarily Rn , or
embedded in some Euclidean space.
Indeed when starting to study the subject, one often has already encountered smooth subsets
of Rn , called submanifolds, such as surfaces of R3 like the sphere, the torus, etc. To say what
smooth means in this context one uses the notion of smooth function in Rn . In this approach, the
regularity properties of the object are extrinsic, i.e., related with the ambient space. In geometry
it is preferable to have an approach to objects that permits to work with them using intrinsic
properties, i.e., not referring to (or independent from) external objects or structures.
The goal of the course is to introduce the language of basic differential geometry. For this
reason, a major part consists in building correct notions and acquire the right flexibility in order
to work with them. A recurrent paradigm in differential geometry is the duality between the two
viewpoints: intrinsic vs in coordinates. Several instances of this question are already present in
linear algebra courses, comparing presentation of abstract vector spaces and their coordinate version
when choosing a basis. For instance one might define what is an eigenvalue for a square matrix and
then discover that similar matrices have the same eigenvalues. This might seem at first a beautiful
coincidence or only the result of a surprisingly short proof. However, when one understands that
the two matrices represent the same endomorphism in different coordinates, the proof is even no
more needed! Indeed, since the notion of an eigenvalue of an endomorphism is independent on any
choice of basis, the statement is a coordinate–invariance property and must be true.
La géométrie n’est pas vraie, elle est avantageuse.1

La Science et l’Hypothèse, 1902
Henrı̀ Poincaré, 1854 – 1912
These lecture notes presents some of the material taught by the author in the Master Degree
of Mathematics at Università degli Studi di Padova, in the course of Differential Geometry.
Padova, Academic years 20/21-21/22-22/23
1
Transl. Geometry is not true, it is advantageous.
5
Chapter 1
Smooth manifolds and smooth maps
La notion générale de variété est assez difficile à définir avec précision.1

Leçons sur la Géométrie des espaces de Riemann, 1946
Élie Cartan, 1869 – 1951
The basic idea to build the definition of abstract manifold, is the one of a space that “looks
locally like Rn ”. A naive approach might be dangerous (cf. É. Cartan). In what follows we formalize
this idea.
1.1 Topological and smooth manifolds

Definition 1.1. Let M be a topological space, U ⊂ M open. Let ϕ : U → ϕ(U ) = V ⊂ Rn
be an homeomorphism onto an open set V of Rn . The pair (U, ϕ) is called a chart. The inverse
ϕ−1 : V → U is a local parametrization.
A chart (U, ϕ) gives local coordinates to points of U . Namely, for q ∈ U ⊂ M we assign n

coordinates to it, i.e., ϕ(q) = (x1 , . . . , xn ) ∈ Rn . Here n is the number of scalar information needed
to identify a point on the space M when looking in a (small) region U .
We stress that there is no regularity on the space M up to now. Moreover, notice that the
number n is attached to a single chart and might a priori depend on the chart itself.
Remark 1.2. If we have two charts ϕ1 : U1 → ϕ(U1 ) = V1 ⊂ Rn1 and ϕ2 : U2 → ϕ(U2 ) = V2 ⊂ Rn2 ,
with U1 ∩ U2 6= 0, we can consider
ϕ2 ◦ ϕ−1
1 : ϕ1 (U1 ∩ U2 ) → ϕ2 (U1 ∩ U2 )
which is an homeomorphism from an open set of Rn1 and Rn2 . It follows that n1 = n2 (cf. Appendix
of this chapter). We deduce that on connected components of M the number n is independent on
the chart, this is the dimension of the connected component of M .
Definition 1.3. A topological manifold is a topological space together with an atlas, i.e., a family
{(Ui , ϕi )}i∈I of charts such that M is covered by the open sets, M = ∪i∈I Ui .
1
Transl. The general notion of manifold is quite hard to define precisely.
7
By Remark 1.2, if M is connected we have a well-defined notion of dimension of a topological
manifold. For simplicity we develop the theory for connected spaces and we will say M is a
n-dimensional topological manifold when charts take values in Rn .
Remark 1.4. As a consequence of the definition, the topology of a differentiable manifold is always
locally compact, locally connected and locally path connected (prove this as an exercise!). But this
does not impose topological properties at the global level.
Sometimes in the literature the definition of topological manifold contains further (global) topo-
logical assumptions, which we will impose now.
Standing assumptions: From now on we will always assume for a topological manifold M :
(C) M is connected
(H) M is Hausdorff (points are separated)
(SC) M is second countable (countable basis for the topology of M )
Assumption (C) could be replaced by asking that all connected components of a topological
manifold have the same dimension. Since all the theory is local, there is no restriction to assume
connectedness.
Assumption (H) permits to avoid the example of the line with two origins, which is locally
Euclidean but not Hausdorff and geometrically not so “smooth” !
Assumption (SC) is related to paracompactness, which is needed for partition of unity, cf.
Section 1.4.
Functions
Let M be a n-dimensional topological manifold. Since M is a topological space, we can say what
is a continuous function f : M → R.
On the other hand up to now we can not say what does it mean “f : M → R is of class C ∞ ”. It
turns out that defining smooth (C ∞ ) functions correspond exactly to introduce a notion of smooth
structure on M .
Let f : M → R be continuous. For each i of an open cover we define2
fbi := f ◦ ϕ−1 n
i : ϕi (Ui ) ⊂ R → R.
Notice that fbi is the f “read in coordinates” on the open set Ui . Hence with f we build a collection
{fbi }i∈I of functions from (open subsets of) Rn to R. It is easy to notice that f is continuous if and
only if each fbi is continuous.
We can say what does it mean for a single fbi to be C ∞ , but this does not define a good notion
on f since
fbi = fbj ◦ ϕj ◦ ϕ−1
i = fj ◦ ηij
b
where ηij := ϕj ◦ ϕ−1

i is the transition function. If fbj is C ∞ we can deduce that fbi is C ∞ only if
−1
ϕj ◦ ϕi is C ∞ as well. We stress that the change of charts is defined on the following sets
ϕj ◦ ϕ−1
i : ϕi (Ui ∩ Uj ) → ϕj (Ui ∩ Uj ).
2
technically one should write f |Ui ◦ ϕ−1
i
8
This motivates the following definitions (and proves that they are well-posed).
Definition 1.5. A differentiable manifold of class C k and dimension n is a n-dimensional topo-
logical manifold M together with a smooth atlas, i.e. an atlas {(Ui , ϕi )}i∈I of charts such that all
transition functions ϕj ◦ ϕ−1
i are of class C k .
Notice that here k may takes value in {0, 1, 2, . . . , ∞, ω} where ω stands for analytic. In what
follows we focus on C ∞ case and smooth means always C ∞ , unless specified. A smooth atlas is
what defines on M a so-called differentiable structure.
Definition 1.6. Let M be endowed with the smooth structure given by the atlas {(Ui , ϕi )}i∈I . A
funtion f : M → R is smooth if and only if fbi := f ◦ ϕ−1
i is smooth for every i ∈ I.
It is a direct consequence of the definition but it is a good idea to get convinced of the following
fact, saying that it is enough to check smoothness locally.
Exercise 1.7. Prove that f is smooth if and only if for every point x there exists j = j(x) ∈ I
such that x ∈ Uj and fbj := f ◦ ϕ−1
j is smooth.
Remark 1.8. 1. The Euclidean space Rn has its canonical differentiable structure covered by a
single chart (U, ϕ), with U = Rn and ϕ(x) = x, the identity map. With this differentiable structure
smooth functions with respect to the smooth structure are just the usual C ∞ functions in Rn , i.e.,
those for which partial derivatives of every order are continuous.
2. On R we can define a differentiable structure with a single chart (R, ψ), ψ : R → R, ψ(x) = x3 .
In this case the function f : R → R given by f (x) = x (the identity function) is in coordinates
√ √
fb = f ◦ ψ −1 (x) = f ( 3 x) = 3 x
which is not smooth! The standard differentiable structure and this new one are not equal, in the
sense that they define different smooth functions.
3. On the other hand one can consider on R the atlas {(Ix,r , ϕx,r )}x∈R,r>0 where Ix,r = (x −
r, x + r) is the open interval and ϕx,r (y) = y for every y ∈ Ix,r . This is the standard structure due
to locality of the notion of smoothness for real functions.
Two smooth atlases A, A0 are said compatible if every change of chart is smooth, i.e., if the
union A ∪ A0 is still a smooth atlas. A smooth atlas is called maximal if it is not contained in any
strictly larger compatible smooth atlas.
A chart is compatible with a smooth atlas A if adding the chart to the atlas A one gets a
smooth atlas. In this case we say that the chart is a smooth chart.
Proposition 1.9. Let M be a topological manifold. Then
(i) every smooth atlas for M is contained in a unique maximal smooth atlas
(ii) two smooth atlases for M determine the same maximal smooth atlas if and only if they are
compatible
In what follows we simply say smooth manifold for differentiable manifold of class C ∞ and we
assume that smooth atlas are maximal.
All this sounds very difficult to use in concrete situations, especially because one has to start
with a topological structure on the set before defining a smooth structure. Here is a construction
lemma for smooth manifolds which permits to do both steps together.
9
Proposition 1.10 (construction lemma). Let M be a set and {Uα }α∈A a collection of subsets
together with ϕα : Uα → Rn injective maps such that
1. ϕα (Uα ) is open in Rn ,
2. ϕα (Uα ∩ Uβ ) and ϕβ (Uα ∩ Uβ ) are open in Rn
3. ϕα ◦ ϕ−1
β is a diffeomorphism
4. M is covered by countably many Uα
5. if x 6= y either they belong to the same Uα or to two disjoint ones
Then there exists a unique smooth manifold structure such that (Uα , φα ) are charts.
Proof. Consider the topology generated by ϕ−1 n

α (V ), with V open set in R . (Exercice: prove that
this is a topology!). Then check that all other conditions included (H) and (SC) are satisfied by
construction.
1.2 Some fundamental examples

Note. Very often one might read “M is a smooth manifold” meaning that M can be endowed with
a smooth structure manifold.
• Rn is a n-dimensional manifold. Every open set U ⊂ Rn is a n-dim manifold. Union of

countably many points are 0-dimensional manifolds.
• if M is m-dimensional and N is n-dimensionale then M × N is (m + n)-dim manifold.
Spheres
We want to prove that the unit sphere S n has the structure of smooth n-dimensional manifold. Let
us consider
S n = {x = (x1 , . . . , xn+1 ) ∈ Rn+1 : kxk2 = 1} ⊂ Rn+1
with the topology induced by Rn+1 . We will build an atlas A = {(UN , ϕN ), (US , ϕS )} with two
charts where
UN = S n \ {N }, US = S n \ {S},
with N and S the north and south pole respectively. The maps ϕN , ϕS are stereographic projections.
For instance let us construct ϕN : given P = (x1 , . . . , xn+1 ) in UN we consider ϕN (P ) as the
intersection of the segment P N with the hyperplane H = {xn+1 = 0}. Identifying H with Rn (just
by removing the last component of the vector) we get a map
1
ϕN : UN → Rn , ϕN (P ) = (x1 , . . . , xn ),
1 − xn+1
which is well defined since on UN we have xn+1 6= 1. Similarly one can define and compute
1
ϕS : US → Rn , ϕS (P ) = (x1 , . . . , xn ).
1 + xn+1
10
A computation shows that the inverse has the following form
1
ϕ−1 n
N : R → UN , ϕ−1
N (u) = (2u1 , . . . , 2un , kuk2 − 1),
kuk2 + 1
and similar results hold for ϕ−1

S . We invite the reader to check the details and to compute that
y
ϕS ◦ ϕ−1 n n
N : R \ {0} → R \ {0}, y 7→ .
kyk2
Notice that with the above identifications ϕN (UN ∩ US ) = ϕS (UN ∩ US ) = Rn \ {0}.

Remark 1.11. The atlas A has two charts. This is the minimum. Indeed if we could build an atlas
with one chart S n would be homeomorphic to an open set of Rn , which is not possible since S n is
compact.
Exercise 1.12. Consider the function f : S 2 → R given by f (x, y, z) = z if (x, y, z) ∈ S 2 . Clearly

f is the restriction of a smooth function of R3 to the sphere. Write f in coordinates and show that
f is also smooth with respect to the differential structure just introduced.
Exercise 1.13. Consider the atlas A0 = {(Ui± , ϕ±

i )}i=1,...,n+1 defined as follows
Ui+ = S n ∩ {xi > 0}, Ui− = S n ∩ {xi < 0}
with maps
−
ϕ+
i (x1 , . . . , xn+1 ) = ϕi (x1 , . . . , xn+1 ) = (x1 , . . . , x
bi , . . . , xn+1 )
Show that A and A0 are compatible, i.e., change of charts between elements of the two atlases are
smooth.
Projective spaces
The usual definition of Pn (R) is as the quotient (endowed with quotient topology):
Pn (R) = (Rn+1 \ {0})/ ∼
with the equivalence relation x ∼ y in Rn+1 \ {0} if x = λy for some λ 6= 0. Denoting
x = (x0 , . . . , xn ) ∈ Rn+1 \ {0}
we denote
[x] = [x0 , . . . , xn ] ∈ Pn (R)
the equivalence class. We define Ui = {[x] ∈ Pn (R) : xi 6= 0} for i = 0, . . . , n. It is easy to see that
these are well defined and that
[n
n
P (R) = Ui .
i=0
We set ϕi : Ui → Rn as follows

x0 xi−1 c
xi xi+1 xn
ϕi ([x]) = ,..., , , ,..., ∈ Rn
xi xi xi xi xi
11
where the “hat” stands for having removed that component (so it remains a n-vector). Geometri-
cally ϕi ([x]) are the coordinates of the intersection of the line represented by [x] with the hyperplane
{xi = 1} (after removing the i-th coordinate equal to 1). The map ϕi is an homeomorphism onto
Rn with inverse
ϕ−1 n
i : R → Ui , y 7→ [y1 , . . . , yi−1 , 1, yi , . . . , yn ]
One can check that (we set for instance3 j < i) the change of charts is smooth

−1 n n y0 yj−1 1 yj+1 yi−1 c
yi yi+1 yn
ϕi ◦ϕj : R \{yi = 0} → R \{yj = 0} → y 7→ ,..., , , ,..., , , ,...,
yi yi yi yi yi yi yi yi
where we should notice that Rn \ {yi = 0} = ϕj (Ui ∩ Uj ), and similarly for the other one. It follows
that Pn (R) is a differentiable manifold of dimension n.
The set of affine lines in the plane

Consider the set AL(R2 ) of affine lines in R2 . This space has no a priori a topology, we use
Proposition 1.10 to build the differential structure.
Given a line ` of equation ax + by = c we define the sets
Ua = {` ∈ AL(R2 ) | a 6= 0}, Ub = {` ∈ AL(R2 ) | b 6= 0},
and the charts ϕi : Ui → R2 defined for i = a, b by

b c a c
ϕa (`) = − , , ϕb (`) = − , ,
a a b b
It is not difficult to check that if (u, v) denote the coordinate on R2 then

−1
ϕb ◦ ϕ−1
a (Ua ) = {(u, v) | u 6= 0} = ϕa ◦ ϕb (Ub )
and on this set

1 v
ϕb ◦ ϕ−1
a (u, v) = ,−
u u
which is smooth on {(u, v) | u 6= 0}.
The atlas A = {(Ui , ϕi )}i=a,b gives the structure of smooth manifold to AL(R2 ).
Remark 1.14. The charts corresponds to writing the line ax + by = c in the form
y = mx + p, x = ny + q
Indeed if the line is neither horizontal nor vertical we have

b c a c
(m, p) = − , , (n, q) = − ,
a a b b
Exercise 1.15. Let o be the origin of R2 , for every affine line ` ∈ AL(R2 ) define f (`) := dist2 (o, `),
where dist denotes the Euclidean distance in R2 from a point to a line. Prove that the function f
is C ∞ with respect to the smooth structure of AL(R2 ) .
3
the other case is similar but pay attention to indices if you write it!
12
Level sets
Let U ⊂ Rn be open and F : U → Rm be a smooth map. We say that x ∈ U is critical point if
DF (x) is not surjective, x is a regular point otherwise.
A point y ∈ Rm is said to be a regular value if every x ∈ F −1 (y) is a regular point.
Theorem 1.16. Let F : U ⊂ Rn → Rm be a smooth map such that y0 ∈ Rm is a regular value for
F . Then F −1 (y0 ) is a smooth manifold of dimension n − m.
Proof. It is not restrictive to assume y0 = 0. Fix x0 ∈ U ∩ F −1 (0). Since rank(DF (x0 )) = m up

to reordering variables we can split the space as x = (x0 , x00 ) ∈ Rn−m × Rm in such a way that
the m × m block ∂F/∂x00 is invertible. Then applying the classical implicit function theorem there
exists a neighborhood of x0 , which we can take of the form U 0 × U 00 with U 0 ⊂ Rn−m , U 00 ⊂ Rm ,
and a smooth function f : U 0 → U 00 such that
F (x) = 0, x ∈ U 0 × U 00 ,
is equivalent to
x00 = f (x0 ), x0 ∈ U 0 .
Then defining ψ : U 0 ⊂ Rn−m → Rn as ψ(x0 ) = (x0 , f (x0 )) we have proved that
(U 0 × U 00 ) ∩ F −1 (0) = ψ(U 0 )
with ψ invertible onto its image. Denoting ϕ = ψ −1 we have that ϕ defines a chart on the open set
(U 0 × U 00 ) ∩ F −1 (0).
Exercise 1.17. As an exercise check that the change of two such charts is smooth. Notice that
the operation of renaming variables (used at the very beginning of the proof) corresponds to apply
an invertible linear transformation to the space.
We will see how this generalizes to manifolds later.
Exercise 1.18. Discuss for which c ∈ R the following subset of R2 is a smooth manifold
x3 + xy + y 3 = c (1.1)
Exercise 1.19. Discuss whether the following subset of R3 is a smooth manifold

(
x2 + y 2 + z 2 = 1
(1.2)
x2 + y 2 − x = 0
This subset of R3 is also known as “Viviani’s window” (but check that on the web only after having
tried to solve the exercise!).
13
Grassmannians
Let us consider the Grassmannian of k planes in a n-dim vector space V .
Gk (V ) = {W ⊂ V | dim(W ) = k}.
We want to show this is a k(n − k) dimensional manifold. The idea behind the construction is
that k-dimensional subspaces can be described as the graph of linear maps Rk → Rn−k , hence
parametrized by k(n − k) matrices.
Charts are build as follows: for every U ⊂ V with dim U = n − k we consider
U t = {W ∈ Gk (V ) | W ⊕ U = V } = {W ∈ Gk (V ) | W ∩ U = 0}.
Of course Gk (V ) is covered by such open sets

[
Gk (V ) = U t,
U
since every k-dimensional set is transversal to some (n − k)-dim one. To define charts let us fix an
element Z ∈ U t , which will play the role of the origin of the chart.
The following lemma of linear algebra holds.
Lemma 1.20. Every W ∈ U t is the graph of a unique linear map AW : Z → U .
Choosing basis for Z and U we can easily build a bijective map
ϕU : U t → Mk,n−k (R) ' Rk(n−k) , ϕU (W ) = AW .
Notice that ϕU (Z) = 0 (i.e., the Z we have chosen is the origin).
Exercise 1.21. Let U1 , U2 be two (n − k) dimensional subspaces and fix Zi in such a way that
Zi ⊕ Ui = V . Compute
ϕU1 ◦ ϕ−1
U2 : R
k(n−k)
→ Rk(n−k)
and prove it is a diffeomorphism. Use the construction lemma (Proposition 1.10) to complete the
proof that this gives Gk (V ) the structure of a manifold.
Examples in the space of matrices

The space Mn (R) of square matrices with real entries can be endowed with a structure of smooth
manifold of dimension n2 , since it is a vector space. Let us consider some subset of it and show
that they are actually smooth manifolds.
• the space GLn (R) = {A ∈ Mn (R) | det(A) 6= 0} is an open set within Mn (R), since det is a
continuous function, so it is a smooth manifold of dimension n2 .
• the space SL(n) is a smooth manifold of dimension n2 − 1. Indeed we have
SL(n) = {A ∈ GLn (R) | det(A) = 1} = det−1 (1).
14
If we prove that F := det : GLn (R) → R is smooth and the differential is surjective at every
point, we are done. It is smooth because it is a polynomial map. Moreover we have
DF (I)H = trace(H) (1.3)
Equation (1.3) can be proved by linearity of the differential and applying to a basis. This
implies
det(I + H) = 1 + trace(H) + o(kHk)
Using the properties of the determinant, for A ∈ GLn (R) and H ∈ Mn (R)
det(A + H) = det(A) + det(A)trace(A−1 H) + o(kHk)
Which in turns gives for A ∈ SLn (R)
DF (A)H = trace(A−1 H)
Notice that that DF (A) : Mn (R) → R is surjective as a linear map for every A (this means
for every A there exists H such that trace(A−1 H) 6= 0, and this is clearly true).
• the space On (R) is a smooth manifold of dimension 21 n(n − 1). Recall that
On (R) = {A ∈ Mn (R) | AT A = I} = F −1 (I)
where F : On (R) → Symn (R) is given by F (A) = AT A. We have restricted the target space
in order for F to have surjective differential. In fact, it is easy to see that
DF (A)H = H T A + AT H
where DF (A) : Mn (R) → Symn (R). To prove that it is surjective for every S ∈ Symn (R)
choose H = 21 AS and for such H we have DF (A)H = S. Notice that
n(n + 1) n(n − 1)
dim On (R) = n2 − = .
2 2
Exercise 1.22. What happens in the above argument if one starts with the space of special
orthogonal matrices SOn (R) = {A ∈ Mn (R) | AT A = I, det = 1} = On (R) ∩ SLn (R)?
1.3 Smooth maps

We want to define what is a smooth map F : M → N between two smooth manifolds.
Definition 1.23. Let M, N be two smooth manifolds and F : M → N be continuous. We say that
F is of class C k around a point q0 ∈ M if there exists charts (U, ϕ) around q0 and (V, ψ) around
F (q0 ) such that the following map is of class C k
Fb := ψ ◦ F ◦ φ−1 : ϕ(U ) ⊂ Rn → ψ(V ) ⊂ Rn .
Exercise 1.24. Show that the previous definition one can replace “if there exists charts (U, ϕ)
around q0 and (V, ψ) around F (q0 )” with “for every charts (U, ϕ) around q0 and (V, ψ) around
F (q0 )”
15
Definition 1.25. Let M, N be two smooth manifolds. A continuous map F : M → N is a smooth
diffeomorphism if F is bijective with F and F −1 of class C ∞ .
We say that F is a local smooth diffeomorphism around q0 ∈ M if there exists U neighborhood
of q0 such that F (U ) is open in N and F |U : U → F (U ) is a diffeomorphism.
As before we focus on smooth maps (i.e., of class C ∞ ) but one can define C k diffeomorphisms
as well. We stress that a diffeomorphism is more than a smooth homeomorphism. The inverse
should also be smooth. Recall that a continuous injective map is a homeomorphism onto its image
by Theorem 1.54.
The function x 7→ x3 is an example of a map which is a smooth homeomorphism but not a
diffeomorphism from R to R (when both are endowed with the standard structure, cf. Example 1.28.
Exercise 1.26. Prove that if F : M → N and G : N → P are smooth then G ◦ F is smooth. If F

and G are diffeomorphisms then G ◦ F is also a diffeomorphism. This implies that being smoothly
diffeomorphic defines an equivalence relation between smooth manifolds.
Example 1.27. 1. The (open) unit ball B is diffeomorphic to Rn through (write explicitly the
inverse of F )
x
F : B → Rn , F (x) = p .
1 − kxk2
2. A trivial but important observation for what comes later is that if (U, ϕ) is a smooth chart
on a smooth manifold then ϕ : U → ϕ(U ) ⊂ Rn is a diffeomorphism when regarded as a map from
the smooth manifold U and the open subset ϕ(U ) in Rn with the standard structure.
3. The two smooth manifolds S 1 and SO(2) endowed with their standard smooth structures
(defined as in the previous section) are diffeomorphic through the bijective smooth map

x y
F : S 1 → SO(2), (x, y) 7→
−y x
where x2 + y 2 = 1. It is left to the reader to check that both F and F −1 are smooth.
Exercise 1.28. Recall that (R, ϕ) with the ordinary smooth structure given by the identity chart
ϕ(x) = x, and (R, ψ) with the chart ψ : R → R given by ψ(x) = x3 , are not defining the same
smooth structure.
Nevertheless these two structures are equivalent up to diffeomorphism. Consider F : (R, ϕ) →
√
(R, ψ) the map F (x) = 3 x which is an homeomorphism. Its coordinate presentation Fb = ψ ◦ F ◦
ϕ−1 (x) = x so it is smooth (and the same is true for the inverse).
Remark 1.29. We state here some general fact without proofs.
• If a smooth manifold M has dimension dim M ≤ 3, then all smooth structures on M are
equivalent up to diffeomorphisms.
• There exists topological manifolds which admit no smooth structure. In particular Michael
Freedman in 1982 found an example of a 4-dimensional topological manifold (called E8 man-
ifold) which can be proved to admit no smooth structure (using Rokhlin’s theorem, or Don-
aldson’s theorem).
16
• On Rn , for n 6= 4, all smooth structures are equivalent up to diffeomorphisms. R4 admits
an uncountable number of smooth structures that are non equivalent up to diffeomorphisms.
These are called fake R4 ’s. (DeMichelis Freedman, 1982)
• John Milnor first proves the existence of a smooth structure on S 7 which is not diffeomorphic
to the standard one (1956). There are actually 28 non equivalent (1963). These smooth
structures are also called exotic spheres.
Lie groups
Another key notion in geometry is the one of Lie group. This is both a group and a manifold and
the group law is well-behaved with respect to the smooth structure.
Definition 1.30. We say that G is a Lie group if G is a smooth manifold and a group where the
maps multiplication and inverse
m : G × G → G, m(x, y) = xy
i : G → G, i(x) = x−1
are smooth with respect to the differentiable structure.
Indeed one can check that it is enough to verify that the map G × G → G that (x, y) 7→ xy −1
is smooth. Then we can define right and left translations
Rg : G → G, Rg (h) = hg
Lg : G → G, Lg (h) = gh
which are diffeomorphisms since they are smooth (composition of injection in a product and mul-
tiplication) with inverses L−1 −1
g = Lg −1 and Rg = Rg −1 .
→ examples of Lie groups: S 1 , T n , GLn (R), SOn (R), SUn (C).
Coverings
Another class of smooth maps that are local diffeomorphisms are smooth coverings.
f → M is a smooth covering if π is surjective, of class C ∞ , and
Definition 1.31. We say that π : M
for every q ∈ M has a connected neighborhood U such that for every connected component U e of
−1
π (U ) we have π|Ue : U → U is a diffeomorphism.
e
Another way to say that: for every q ∈ M has a connected neighborhood U such that
[
π −1 (U ) = Vα
α∈A
where Vα are open set diffeomorphic to U through π. A covering is said to be universal covering if
M
f is simply connected. The universal covering is unique up to diffeomorphism.
17
Remark 1.32. Notice that the set A depends on q. If the cardinality of A is finite, then it is locally
constant (prove it!), so that if M is connected and the cardinality is finite at one point, then it is
everywhere constant. In this case we say that the covering has m = card(A) sheets.
Example 1.33. example R → S 1 with t 7→ e2πit . Infinitely many sheets
Example 1.34. example S n → Pn with x 7→ [x]. Two sheets. For n = 3 this has a Lie group
incarnation: SU (2) → SO(3). Consider S 3 ⊂ R4 ' C2 such that (u, v) ∈ C2 is in S 3 if |u|2 +|v|2 = 1.
Then
3 u v
S → SU (2), (u, v) 7→
−v u
is a bijective map, which is indeed a diffeomorphism with respect to the smooth structures.
The fact that SO(3) ' P 3 (R) is also diffeomorphic to P 3 (R) is interesting. Every matrix A
in SO(3) different from the identity represents a rotation around some axis Rv and of some angle
θ ∈ [−π, π]. Then one can build the map SO(3) → P 3 (R) given by A 7→ θv
Another way to see that S 3 is a group, is to identify it to unit quaternions
q = a + bi + cj + dk
where i, j, k are the imaginary elements.
Example 1.35. The covering π : S 1 → S 1 where z 7→ z n is a covering with n sheets.
Proper maps
We begin with some definitions
Definition 1.36. A continuous map f : M → N between smooth manifold is proper if f −1 (K) is

compact in M for any K compact in N .
Definition 1.37. A continous (resp. smooth) map f : M → N between smooth manifold is a

continuous (resp. smooth) covering map if for every y ∈ N , there exists an open neighborhood V of
y, such that f −1 (V ) is a union of disjoint open sets in M , each of which is mapped homeomorphically
(resp. diffeomorphically) onto V .
Proposition 1.38. Any proper continuous map f : M → N between smooth manifold is closed,
i.e., f (C) is closed in N for every closed set C ⊂ M .
Proof. Take a sequence yn in f (C) with yn → y. Then yn = f (xn ) and xn ∈ C. For n large
enough, we can assume yn ∈ K where K compact neighborhood of y. So that xi ∈ f −1 (K) which
is compact. hence xn → x for some x (up to subsequences) and f continuous so that f (x) = y.
f (C) is closed.
Theorem 1.39. Let M, N be smooth connected manifold and f : M → N be a proper local

diffeomorphism. Then f is a smooth covering map.
Proof. Since f is a local diffeomorphism, it is open. Since f is proper, it is closed. Hence f (M )

is open and closed in N and, by connectedness, f is surjective. Fix y ∈ N . Since π is a local
diffeomorphism, each point of f −1 (y) has a neighborhood on which f is injective, so f −1 (y) is a
18
discrete set. Since the singleton {y} is compact and f is proper, then f −1 (y) is compact, hence
finite. Set f −1 (y) = {x1 , . . . , xk }. Fix Ui a neighborhood of xi where f is a diffeomorphism.
It is not restrictive to suppose that Ui ∩ Uj = ∅ for i 6= j. Set V = ∩ki=1 f (Ui ). Since each
f (Ui ) is a neighborhood of y, V is a neighborhood of y also. By replacing V with the connected
component of V \ f (M \ ∪i Ui ) (which is open since f is closed) containing y, we can moreover
assume that V is connected and f −1 (V ) ⊂ ∪i Ui . Hence if one set U i := Ui ∩ f −1 (V ) one can
check that f −1 (V ) = ∪i U i , dijoint union of its connected components, and that f : U i → V is a
diffeomorphism, as desired.
One it is known that the map f is a covering map, to show that it is injective one should prove
that it is a 1-sheet covering, i.e., the preimage of each point is a single point. The following corollary
provides a criterium.
Corollary 1.40. Under the previous assumptions, if N is simply connected, then f is a diffeomor-
phism.
It is enough to show that the map f is injective. Let x1 6= x2 in M such that f (x1 ) = f (x2 ).
Take a continuous curve α : [0, 1] → M such that γ(0) = x1 and γ(1) = x1 homotopic to a point.
Its image γ := f ◦ α : [0, 1] → N is a closed loop in N such that γ(0) = γ(1) = y. Since N is simply
connected there exists a continous map
Γ : [0, 1] × [0, 1] → N
such that Γ(0, t) = y and Γ(1, t) = γ(t). For s sufficiently closed to 0 the curve γs (t) = Γ(s, t) stays
in the set V where f is a covering hence f −1 (γ) is the union on k closed loop and it should be
homotopic to a point. This gives a contradiction.
1.4 Bump functions, partition of unity and paracompactness

In this section we discuss partition of unity and its relation with paracompactness. For what
described in this section is really important to assume the standing assumptions in Section 1.1, in
particular the assumption (SC).
We start with the following auxiliary lemma.
Lemma 1.41. There exists a C ∞ function Φ : Rn → R such that
(i) 0 ≤ Φ ≤ 1
(ii) Φ ≡ 1 on B(0, 1/2) = {x ∈ Rn | kxk < 1/2},
(iii) Φ ≡ 0 on B(0, 1) = {x ∈ Rn | kxk > 1}.
Proof. We start by recalling that the following function h : R → R is of class C ∞
(
0, if t ≤ 0
h(t) = −1/t
e , if t > 0.
It is enough then to define Φ : Rn → R as follows
h(1 − kxk2 )
Φ(x) = ,
h(1 − kxk2 ) + h(kxk2 − 1/4)
19
It is easily seen that Φ is C ∞ and satisfies the properties above.
The function Φ is what is called a bump function. To build bump functions on smooth manifolds
we first need the following observation.
Proposition 1.42. Every topological manifolds admits a countable cover of precompact coordinate
balls.
Proof. Asssume M is covered by one chart ϕ : M → ϕ(M ) = V ⊂ Rn . Consider the set B of balls B
in Rn with rational center and rational radius, whose closure is contained in V . This is a countable
basis for the topology of V . Then consider the topology on M generated by the open sent ϕ−1 (B)
for B ∈ B. The general case follows using (SC) assumption. (Details left to the reader).
Proposition 1.43. Let K ⊂ V ⊂ M with K compact and V open subset of a smooth manifold M .
There exists a smooth function f : M → R such that f ≡ 1 on K and f ≡ 0 on M \ V .
Proof. Use an open cover with values in precompact balls. More precisely for every q in K take a
chart ϕq : Uq → B(0, 2) with U q ⊂ V . By compactness take a finite number (relabeled ϕi : Ui →
B(0, 2) for i = 1, . . . , N ) of such charts such that
N
[ N
[
K⊂ ϕ−1
i (B(0, 1/2)) ⊂ Ui ⊂ V
i=1 i=1
Then set (
Φ(ϕi (q)), q ∈ Ui ,
fi : Ui → R, fi (q) =
0, q∈/ Ui .
Notice that fi is smooth since ϕ−1

i (B1/2 ) ⊂ Ui . We define f : M → R as
N
Y
f (q) = 1 − (1 − fj (q))
i=1
which is clearly smooth. Moreover if q ∈ K then at least one fj is 1, hence f (q) = 1. If q ∈

/ V then
all fj are 0 hence f (q) = 0.
Of course one can exchange zero with one and reduce the compact to a point.
Corollary 1.44. For every q ∈ M and U open neighborhood of q, there exists a smooth function
such that f (q) = 0 and f ≡ 1 outside U .
Another such a property. Here by f smooth on a compact set means f coincides with the
restriction to K of a smooth function defined on some open neighborhood of K.
Proposition 1.45. Let K ⊂ V ⊂ M with K compact and V open subset of a smooth manifold M .
If g : K → RN is smooth then there exists a smooth extension gb : M → RN with supp gb ⊂ V .
Recall that for f ∈ C ∞ (M, RN ) we denote supp f = {x ∈ M | f (x) 6= 0}.
20
Proof. Let U be the neigh of K where there exists the extension so that g = ge|K for ge : U → RN .
Set W = U ∩ V which is still a neighborhood of K and apply the proposition to the pair K ⊂ W ,
i.e., find a smooth function f such that f ≡ 1 on K and f ≡ 0 outside W . Then consider:
(
g (q), q ∈ W,
f (q)e
gb(q) =
0, q∈/ W.
which satisfies the requirements.
Paracompactness and partition of unity

Given an open cover U = {Ui }i∈I of M we say that an open cover V = {Vj }j∈J is a refinement of
U if for every j ∈ J one has Vj ⊂ Ui for some i ∈ I.
An open cover U = {Ui }i∈I of M is locally finite if each point q ∈ M has a neighborhood V
that V ∩ Ui 6= ∅ only for a finite number of i ∈ I.
Definition 1.46. A smooth manifold is said to be paracompact if from every open cover we can
extract a refinement which is locally finite.
Proposition 1.47. Every smooth manifold is paracompact.
Proof. For a complete proof we refer to [Lee13]. Here we only prove a key step which is given in
the following Lemma.
Lemma 1.48. Every topological manifold has a countable and locally finite open cover by precom-
pact sets.
Sketch of the proof. Start from {Bi }i∈N countable open cover by precompact balls. Make it locally
finite as follows: set U1 = B1 and then define iteratively Uj as follows: if Uj is defined then set mj
such that
U j ⊂ B1 ∪ . . . ∪ Bmj
which is finite by compactness. Set Uj+1 = B1 ∪ . . . ∪ Bmj . Notice that necessarily mj+1 > mj in
this construction. Then define Vj = Uj+2 \ Uj . This is locally finite.
We stress that here the second countable assumption (SC) we added is crucial! This proves the
existence of the partition of unity.
Definition 1.49. A partition of unity is a family of smooth functions {ψi }i∈I such that
(i) 0 ≤ ψi ≤ 1
(ii) {supp ψi }i∈I is a locally finite cover of M

P
(iii) i∈I ψi = 1
A partition of unity {ψi }i∈I is subordinated to an open cover U = {Ui }i∈I if supp ψi ⊂ Ui for
every i ∈ I. Notice that the sum in (iii) is finite at every point by condition (ii).
Theorem 1.50. Every open cover U = {Ui }i∈I admits a partition of unity {ψi }i∈I which is subor-
dinated to U.
21
Proof. See [Lee13].
Notice that the initial open cover in the previous theorem is not necessarily locally finite. Thanks
to partition of unity it is very easy to prove the existence of bump function even for closed sets
(not only compact as in Proposition 1.43).
Proposition 1.51. Let C ⊂ V ⊂ M with C closed and V open subset of a smooth manifold M .
There exists a smooth function a : M → R such that f ≡ 1 on C and f ≡ 0 on M \ V .
Proof. Let us consider the open cover U1 = V and U2 = M \ C. Let ψ1 , ψ2 a partition of unity
subordinated to the open cover. We have ψ2 = 0 on C i.e., outside U2 = M \ C. Then ψ1 = 1 on
C and has support inside U1 = V , i.e., satisfies the requirements.
Similarly, but with little more work, one proves the corresponding extension lemma.
Proposition 1.52. Let C ⊂ V ⊂ M with C closed and V open subset of a smooth manifold M . If
f : C → RN is smooth then there exists a smooth extension fb : M → RN with supp fb ⊂ V .
Proof. See [Lee13].
To end the section let us prove another application: every noncompact manifold can be approx-
imated via a family of compact subsets. Recall that an exhaustion function for a smooth manifold
is a f ∈ C ∞ (M ) such that the sublevel set {f ≤ c} is compact for every c ∈ R.
Proposition 1.53. Every smooth manifold admit a smooth exhaustion function.
Proof. Since the sublevel sets {f ≤ c} are closed and monotone with respect to c ∈ R, it is enough
to prove that {f ≤ N } is compact for every N ∈ N.
Consider a countable open cover by precompact open sets {Vj }j∈N and a partition of unity
{ψj }j∈N subordinated to it. Define
X∞
f (q) := jψj (q)
j=1
Notice that the sum is finite at every point since {ψj }j∈N is a partition of unity (the supports define
a locally finite cover). Fix N ∈ N and notice that if q ∈/ ∪Nj=1 V j then ψj (q) = 0 for 1 ≤ j ≤ N . We
can estimate
X∞ X∞ X∞
f (q) = jψj (q) > N ψj (q) = N ψj (q) = N
j=N +1 j=N +1 j=1
It follows that the sublevel set satisfies for every N ∈ N.
{f ≤ N } ⊂ ∪N
j=1 V j
which is a closed set contained in a compact set, hence compact.
22
1.5 Appendix: Brower invariance of domain
The following theorem is purely topological.
Theorem 1.54 (invariance of domain, Brower 1912). Let U ⊂ Rn be open and f : U ⊂ Rn → Rn

injective and continuous. Then f (U ) is open and f is a homeomorphisms between U and f (U ).
The proof of this theorem is not trivial.4 It has the following
Corollary 1.55. Let n 6= m. No non-empty open subset of Rn can be homeomorphic to any open
subset of Rm .
Proof. Assume m < n and consider a homeo h : U ⊂ Rn → V = h(U ) ⊂ Rm . Then compose with
i : Rm → Rn canonical injection. We have that i ◦ h is continuous and injective but not open in Rn
since i ◦ h(U ) = i(V ) ⊂ Rm × {0}, which is in contradiction with the invariance of domain.
This corollary is intuitively obvious, but note that topological intuition is not always rigorous.
For instance, it is intuitively plausible that there should be no continuous surjection from Rm to Rn
for n > m, but such surjections always exist, thanks to variants of the Peano curve construction.
4
It is strongly related to Brower fixed point theorem. For an interesting discussion on this
topic one can see for instance the Terence Tao’s blog https://terrytao.wordpress.com/2011/06/13/
brouwers-fixed-point-and-invariance-of-domain-theorems-and-hilberts-fifth-problem/
23
Chapter 2
Tangent space and differentials
The formal definitions in the preceding section do not help much

to an understanding of what the notion of a vector really is.
“The mathematical notion of a vector”1 , 1923
Sir Arthur Stanley Eddington, 1882 – 1944
For submanifolds of Rn the tangent space is naturally defined as a vector space of the Rn
itself, for smooth manifolds we cannot use the ambient space and we have to define an abstract
notion. There are several approaches to define tangent space: historically one of the first is the
characterisation we will see in Exercice 2.13, which is by the way a non evident definition of
(tangent) vector at a first sight (cf. Sir Arthur Stanley Eddington).
2.1 Tangent space

A natural approach to tangent space is through (equivalence classes of) smooth curves. This has
some drawbacks in particular when proving that the tangent space is a vector space and prove that
the differential is linear.
Here we’ll use the approach through derivations (defined on C ∞ (M ) and not on germs, exploit-
ing bump functions). The reader is invited to have a look to the different presentations one can
find in different books.
Definition 2.1. Let M be a smooth manifold. A tangent vector X at a point q in M is a linear

map X : C ∞ (M ) → R which is a derivation, i.e., X satisfies for f, g ∈ C ∞ (M )
X(f g) = X(f )g(q) + f (q)X(g)
The tangent space Tq M is the set of all tangent vectors at q.
Notice that Tq M has the natural structure of vector space. This is not related to any external
structure, which has an advantage: no verification is needed. Notice also that, at the moment, it
is not at all clear what is the dimension of Tq M .
Lemma 2.2. We have X(c) = 0 for every constant function c, morever we have that X(f g) = 0
whevever f (q) = g(q) = 0.
1
a section in the book “The Mathematical Theory of Relativity” by the same author
25
Proof. The second is trivial by definition of derivation. The first proof is also easy since
X(1) = X(1 · 1) = 2X(1).
Hence X(1) = 0 and by linearity X(c) = cX(1) = 0.

Let us prove locality of tangent vectors. (Here it is crucial to work in the C ∞ category and not
in the analytic one)
Proposition 2.3. Let f and g agree on a neighborhood U of a point q. Then X(f ) = X(g) for
every X ∈ Tq M .
Proof. It is enough to show that if h = 0 on a neighborhood U of q then X(h) = 0. Indeed by
linearity then we obtain the statement by choosing h = f − g.
Let ψ be a bump function such that ψ ≡ 0 on U and ψ ≡ 1 on supp h, whose existence is
guaranteed by Proposition 1.51. Then h = ψh by construction (where h is non zero, ψ is 1) and
we have
X(h) = X(ψh) = X(ψ)h(q) + ψ(q)X(h) = 0.
2.2 Differential
Next we move to the definition of differential. This is a linear map between vector spaces. If
one thinks in terms of smooth curves the following definition is natural, otherwise it is a little bit
abstract.
Definition 2.4. Let F : M → N be a smooth map between smooth manifolds and q ∈ M . We
define the differential F∗ : Tq M → TF (q) N as follows: for every v ∈ Tq M we define F∗ v ∈ TF (q) N
(F∗ v)(g) = v(g ◦ F )
Easy properties of the differential follow from this definition:

Proposition 2.5. Let F : M → N be a smooth map between smooth manifolds and q ∈ M
(i) F∗ : Tq M → TF (q) N is a linear map
(ii) (G ◦ F )∗ = G∗ ◦ F∗
(iii) (idM )∗ = idTq M
(iv) if F is a diffeo then F∗ is an isomorphism

Proof. The proofs are easy: (i) is linearity of derivations : for every f
F∗ (v + w)(f ) = (v + w)(f ◦ F ) = v(f ◦ F ) + w(f ◦ F ) = F∗ (v)(f ) + F∗ (w)(f )
(ii) is the definition again
((G ◦ F )∗ v)f = v(f ◦ (G ◦ F )) = v((f ◦ G) ◦ F ) = (F∗ v)(f ◦ G) = (G∗ F∗ v)f
(iii) follows by definition. (iv) follows by G = F −1 and (iii).
26
Combine the locality property with the previous Proposition 2.5 to prove.
Exercise 2.6. Let F : M → N is a local diffeo around q ∈ M . Then F∗ : Tq M → TF (q) N is

an isomorphism. In particular show that if U ⊂ M is open and i : U → M is the inclusion then
i∗ : Tq U → Tq M is an isomorphism.
Consider now a chart (U, ϕ) on M , let us write ϕ(q) = (x1 , . . . , xn ). Sometimes we also write
(U, {xi }) to denote the chart. This defines n derivations associated with the charts2 which we will
denote
∂ ∂
,...,
∂x1 q ∂xn q
acting as follows: for any f : U → R we set
∂ ∂(f ◦ ϕ−1 ) ∂ fb
f := = (2.1)
∂xj q ∂xj ϕ(q) ∂xj ϕ(q)
where fb = f ◦ ϕ−1 is the coordinate representation of f and the derivatives on the right are in Rn .
Remark 2.7. This notation will make confusion at first. Do not be scared, it is a good idea to keep
this notation because when you will be used to that you will just identify the two objects!
Proposition 2.8. Let M be a smooth manifold and q ∈ M . If (U, {xi }) is a chart near q, then the
coordinate vectors
∂ ∂
,...,
∂x1 q ∂xn q
define a basis for Tq M . In particular dim Tq M = dim M .
Proof. To prove this, we notice that by definition we have

!
∂ ∂
= ϕ−1
∗
∂xj q ∂xj ϕ(q)
where the derivative in the right hand side is the usual derivative with respect to the j-th variable
in Rn . Since ϕ−1
∗ is an isomorphism (by construction ϕ is a smooth local diffeo), it is enough to
prove that the family { ∂x∂ j ϕ(q) } is a basis for Tϕ(q) Rn , which is proved in the following lemma.
Lemma 2.9. Let f : Rn → R be smooth, then there exist smooth funcions gi : Rn → R for
i = 1, . . . , n such that gi (0) = 0 and
n n
X ∂f X
f (x) = f (0) + (0)xi + gi (x)xi .
∂xi
i=1 i=1
2
meaning that these depends on the choice of the charts
27
Proof. We have
Z 1 n Z 1
d X ∂f
f (x) − f (0) = f (sx)ds = xi (sx)ds
0 ds i=1 0 ∂xi
n n Z 1
X ∂f X ∂f ∂f
= xi (0) + xi (sx) − (0) ds.
∂xi 0 ∂xi ∂xi
i=1 i=1
R 1 ∂f ∂f

and it is sufficient to set gi (x) = 0 ∂xi (sx) − ∂xi (0) ds.
∂ ∂
Lemma 2.10. Let x0 ∈ Rn . Then ∂x1 x0 , . . . , ∂xn x0 is a basis for Tx0 Rn .
Proof. We have to prove that the space of derivations at x0 in Rn is generated by ∂x∂ 1 x0 , . . . , ∂x∂n x0
and that these derivations are independent. It is enough to prove this statement for x0 = 0.
(a). Assume that v = ni=1 αi ∂x∂
= 0. This means that for every smooth function f ∈ C ∞ (Rn )
P
i 0
n
X ∂
0 = vf = αi f
∂xi 0
i=1
By choosing f = xj we get αj = 0 for every j = 1, . . . , n, hence the derivations are independent.

(b). Now we want to show that every derivation v at 0 can be written as ni=1 αi ∂x ∂
P
i 0
for some
Pn ∂
αi . Let us choose αj := v(xj ) and set Xv = i=1 αi ∂xi 0 for this choice. We show v = Xv . Using
Lemma 2.9
n n
X ∂f X
f (x) = f (0) + (0)xi + gi (x)xi ,
∂xi
i=1 i=1
for some smooth funcions gi such that gi (0) = 0. Then
n n
X ∂f X ∂f
v(f ) = 0 + v(xi ) (0) + 0 = αi (0) = Xv (f ).
∂xi ∂xi
i=1 i=1
The differential in coordinates

Let F : M → N be smooth, let us compute the representation of F∗ in coordinates. Recall that if
we have charts (U, ϕ) and (V, ψ) around q and F (q) respectively then F in coordinates read
Fb = ψ ◦ F ◦ ϕ−1 : ϕ(U ) ⊂ Rn → ψ(V ) ⊂ Rm
Denoting (x1 , . . . , xn ) and (y1 , . . . , ym ) the corresponding coordinates on M and N we want to

understand what is the matrix A = (aij ) such that
m
∂ X ∂
F∗ = aij
∂xi q ∂yj F (q)
j=1
28
It is not difficult: we have to sit down, apply to a function f the left hand side, apply definitions
and see what happens:
!
∂ ∂
F∗ f= (f ◦ F )
∂xi q ∂xi q
∂ ∂
= [f ◦ F ◦ ϕ−1 ] = [f ◦ ψ −1 ◦ ψ ◦ F ◦ ϕ−1 ]
∂xi ϕ(q) ∂xi ϕ(q)
∂
= [fb ◦ Fb]
∂xi ϕ(q)
hence applying the chain rule in Rn we get

m m
!
∂ X ∂ fb ∂ Fbj X ∂ fb ∂ Fbj
F∗ f= =
∂xi q ∂yj Fb(ϕ(q)) ∂xi ϕ(q) ∂yj ψ(F (q)) ∂xi ϕ(q)
j=1 j=1
m
X ∂ Fbj ∂
= f
∂xi ϕ(q) ∂yj F (q)
j=1
∂
where we have used ψ ◦ F = Fb ◦ ϕ (pay attention to ∂xi when they act on Rn or M ).
Proposition 2.11. Let F : M → N be smooth map between manifolds and q ∈ M . Given charts
(U, ϕ) and (V, ψ) around q and F (q) respectively, then we have that
m
∂ X ∂Fj ∂
F∗ =
∂xi q ∂xi q ∂y j F (q)
j=1
In particular the matrix representing the differential F∗ in coordinates is the Jacobian matrix
DFb(ϕ(q)) of the coordinate representation Fb = ψ ◦ F ◦ ϕ−1
 b 
∂ F1 ∂ Fb1
∂x1 . . . ∂x n
 . .. 
DFb = 
 .
. .


∂ Fbm ∂ Fbm
∂x1 . . . ∂xn .
Recall that if F : Rn → Rm is a map and F (ei ) = aij fj
P
Remark 2.12. There are several other common notations for the differential of a map at q
F∗,q , F 0 (q), DF (q), Dq F, dq F, dF (q), Tq F
Exercise 2.13 (Change of coordinates for vectors). Given two coordinate sets (U, {xi }) and
(U 0 , {x0i }) with q ∈ U ∩ U 0 show that if v ∈ Tq M writes as
n n
X ∂ X ∂
v= vi = vj0 .
∂xi q ∂x0j q
i=1 j=1
then
n
X ∂x0j
vj0 = vi .
∂xi
i=1
∂x0j
(Here, if ϕ, ϕ0 denotes the coordinate maps, the quantity ∂xi is the jacobian of ϕ0 ◦ ϕ−1 ).
29
Use what said for Rn to prove the following.
Exercise 2.14. Let V be a vector space and v ∈ V . Prove that Tv V is canonically isomorphic to
V (i.e., the isomorphism is independent on the choice of a basis in V ).
2.3 A bit of geometry: tangent vector and curves

Now that we have a differential, given a smooth curve γ : I → M defined on an open interval and
t0 ∈ I we can define simply !
d
γ̇(t0 ) := γ∗ ∈ Tγ(t0 ) M
dt t=t0
where d/dt|t=t0 is the standard basis for the one-dimensional space Tt0 I ' Tt0 R. By definition
d
γ̇(t0 )f = (f ◦ γ)
dt t=t0
which is indeed coherent with the intuitive notion of differentiating along a curve.
Remark 2.15. Notice that if γ(t) = ϕ−1 (b
γ (t)) for some chart (U, ϕ) centered at q = γ(0) we have
n n
d d X ∂ fb 0 X ∂
γ̇(t0 )f = (f ◦ γ) = (fb ◦ γ
b) = γ
bi (t0 ) = bi0 (t0 )
γ f
dt t=t0 dt t=t0 ∂xi 0 ∂xi q
i=1 i=1
This means that the coordinate components of the vector can be computed in any coordinate set!
Lemma 2.16. Let q ∈ M . Every tangent vector in Tq M is the tangent vector to a smooth curve.
Proof. Take a smooth chart (U, ϕ) centered at q and write v ∈ Tq M as v = ni=1 vi ∂x ∂

P
i q
. Then
−1
consider γ(t) = ϕ (tv1 , . . . , tvn ). This curve is smooth and is tangent to v at t = 0 by the above
computation.
We can now formally prove the property which was at the basis of our motivation for the
definition of differential.
Proposition 2.17. Let F : M → N and let v ∈ Tq M such that v = γ̇(t0 ) for some γ : I → M .
Then we have
d
F∗ (v) = (F ◦ γ).
dt t=t0
Proof. The proof is just a collage of definitions: define the curve η := F ◦ γ

!
d d d
η̇(t0 ) = η∗ = (F ◦ γ)∗ = F∗ γ∗ = F∗ γ̇(t0 )
dt t=t0 dt t=t0 dt t=t0
This permits the geometric interpretation of the differential: given v ∈ Tq M to compute F∗ v it

is enough to compute the tangent vector to the curve F ◦ γ where γ is a curve tangent to v.
30
2.4 Examples
1. (Tangent space to a submanifold of Rn .) Consider a submanifold S of Rn defined by
S = {x ∈ Rn | F (x) = y0 },
where F : Rn → Rm has surjective differential at every point of S = F −1 (y0 ), for y0 ∈ Rm . Then

we have that for x0 ∈ S
Tx0 S = ker DF (x0 ) = {v ∈ Rn | DF (x0 )v = 0}.
Indeed for every curve γ : I → S such that x0 = γ(0) and v = γ̇(0) we have that F (γ(t)) = y0 for
every t and by differentiating at t = 0 we have
d
0= F (γ(t)) = DF (γ(0))γ̇(0) = DF (x0 )v.
dt t=0
2. (Tangent space to SL(n) and SO(n) at the identity.) Let us compute the tangent space to
these matrix groups. Consider
SL(n) = {A ∈ Mn (R) | det(A) = 1} = det−1 (1).
We have proved that this is a (n2 − 1)-dimensional manifold. Using the previous fact we have that,
denoting I the identity matrix,
sl(n) := TI SL(n) = ker D det(I) = {H ∈ Mn (R) | trace(H) = 0}.
Similarly one proves that for
O(n) = {A ∈ Mn (R) | AT A = I} = F −1 (I)
where F : O(n) → Sym(n) is given by F (A) = AT A we have
so(n) := TI O(n) = ker DF (I) = {H ∈ Mn (R) | H + H T = 0}.
where we used that DF (A)H = AT H + H T A.

3. (Tangent space on Lie groups.) If G is a Lie group then its tangent space to the identity
e ∈ G is a vector space which is called its Lie algebra and denoted by g, namely
g := Te G
If g ∈ G then the left translation Lg : G → G is a diffeomorphisms hence Lg∗ : Te G → Tg G is

a diffeomorphism, hence one can recover the tangent space to every point just by applying the
differential of the left translation
Tg G = Lg∗ (Te G) = Lg∗ (g).
Similar relations are of course true also for right translations.
31
Exercise 2.18. Prove that if G is a Lie group of matrices, i.e., G is a subgroup of GLn (R) (or
GLn (C)) with the product of matrices as operation, then if g ∈ G is a matrix then Tg G = gTe G in
the sense that
Tg G = {gv | v ∈ Te G},
where now g and v are two matrices and gv is their product in GLn (R).
Exercise 2.19 (Tangent space to the Grassmannian.). Let Gk (V ) be the Grassmannian of k-di
subspaces of a n-dim vector space V .Given W ∈ Gk (V ), prove that
TW Gk (V ) ' Hom(W, V /W ).
32
Chapter 3
Immersions, embeddings.
Submanifolds
Geometry is the science of correct reasoning on incorrect figures

George Pólya, 1887 – 1985
In this chapter we discuss the notion of submanifold. The basic idea, thinking to a manifold
as a set which “locally looks like Rn ” for some n, is that a submanifold should be a subset of a
manifold which “locally looks like Rk × {0} ⊂ Rn ” for some 0 ≤ k ≤ n. In this context we also
have a “bifurcation” of notions of submanifold, namely immersed and embedded submanifold.
To introduce these concepts, let us first start by some further considerations on smooth maps
between manifolds.
3.1 Immersions, submersions, embeddings

Definition 3.1. Let F : M → N be smooth. The rank of F at q is the rank of the linear map
F∗ : Tq M → TF (q) N , i.e., the dimension of im F∗ .
(i) F is an immersion if F∗ is injective at every point q ∈ M .
(ii) F is a submersion if F∗ is surective at every point q ∈ M .
(iii) F is an embedding if F is an immersion and F : M → F (M ) ⊂ N is an homeomorphism onto

its image endowed with the subspace topology of N .
Later we introduce immersed (resp. embedded ) submanifolds. As we will see, these are exactly
images F (M ) of a smooth manifold M under an injective immersion (resp. embedding) F : M → N .
Example 3.2. 1. A local diffeomorphism F : M → N is both an immersion and a submersion.
2. Projections from products πi : M1 × . . . × Mn → Mi are submersions. Injections in products

ii : Mi → M1 × . . . × Mn are embedding.
3. A smooth curve γ : I → M is an immersion if and only if γ̇(t) 6= 0 for every t ∈ I.
33
4. The curve γ1 : R → R2 given by γ(t) = (t2 , t3 ) is injective but not an immersion since
γ̇(t) = (2t, 3t2 ) vanish at t = 0. Notice that the image of γ is contained in (in fact it coincides
with) the set {(x, y) ∈ R2 | x3 = y 2 }.
1.0
0.5
0.2 0.4 0.6 0.8 1.0
-0.5
-1.0
Figure 3.1: The curve γ1
5. The curve γ2 : R → R2 given by γ2 (t) = (t3 − 4t, t2 − 4) is an immersion but is not injective
since γ̇2 (t) = (3t2 − 4, 2t) 6= (0, 0) but γ2 (2) = γ2 (−2). It is not an embedding (not injective
hence not homeo). Notice that the image of γ2 is contained in (in fact it coincides with) the
set {(x, y) ∈ R2 | x2 = (y + 4)y 2 }.
2
-6 -4 -2 2 4 6
-1
-2
-3
-4
6. The curve γ3 : I =] − π/2, 3π/2[→ R2 given by γ3 (t) = (sin(2t), cos(t)) is an injective immer-
sion. But it is not an embedding since γ3 (I) ∩ B(0, r) is not homeomorphic to an interval for
any r > 0! (make a picture!). Otherwise show that γ3 (I) is closed and bounded in R2 , hence
compact, while I is not! One can observe that the image of the curve is contained in (in fact
it coincides with) the set {(x, y) ∈ R2 | x2 = 4y 2 (1 − y 2 )}.
7. Let c ∈ R \ Q. The curve γ : R → S 1 × S 1 given by γ(t) = (e2πit , e2πict ) is an injective

immersion. If one writes S 1 = [0, 1]/ ∼ then γ : R → [0, 1]2 / ∼ becomes γ(t) = (t, ct) mod 1.
The curve γ is not an embedding since the closure of γ(R) is S 1 × S 1 , i.e., the curve γ has
dense image. See Appendix of the chapter.
In general it is not easy to check whether an injective immersion is an embedding. This is a

particular case when it is possible.
Proposition 3.3. Let F : M → N be an injective immersion. If M is compact then F is an

embedding.
34
1.0
0.5
-1.0 -0.5 0.5 1.0
-0.5
-1.0
Proposition 3.3 follows from purely topological considerations: if f : X → Y is continuous

and bijective from X compact topological space and Y Hausdorff topological space, then f is an
homeomorphism. The reader is invited to check the details.
Exercise 3.4. Recall that F : M → N is proper if and only if F −1 (K) is compact for every K ⊂ N
compact. Prove that if F is proper then F is a closed map (i.e., F (C) is closed for any closed set
C ⊂ M ) and then show that the assumption M compact in Proposition 3.3 can be replaced by F
proper.
3.2 The constant rank theorem

Our perspective is mainly local so we are now interested in the description of local properties of
immersions or submersions. We first prove an important result which is a consequence of the inverse
function theorem in Rn .
Theorem 3.5. Let F : Rn → Rm be a smooth map with constant rank equal to r in a neighborhood
of x0 ∈ Rn . Then there exists
(i) a local diffeomorphism ϕ : U → U0 from a neigh U of x0 to a neigh U0 of 0
(ii) a local diffeomorphism ψ : V → V0 from a neigh V of F (x0 ) to a neigh V0 of 0
such that ϕ(x0 ) = 0, ψ(F (x0 )) = 0 and
ψ ◦ F |U ◦ ϕ−1 : U0 → V0 , ψ ◦ F |U ◦ ϕ−1 (x1 , . . . , xn ) = (x1 , . . . , xr , 0, . . . , 0).
Proof. We split the space (x, y) ∈ Rn = Rr × Rn−r and (u, v) ∈ Rm = Rr × Rm−r in such a way
that we write
F (x, y) = (F1 (x, y), F2 (x, y))
with F1 : Rr × Rn−r → Rr and F2 : Rr × Rn−r → Rm−r . It is not restrictive to assume that
x0 = (0, 0) and F (x0 ) = (0, 0). In this notation, we have to prove the existence of ψ and ϕ such
that ψ ◦ F ◦ ϕ−1 (x, y) = (x, 0) for all (x, y) close to (0, 0).
We can assume that rank DF (x, y) = r at every point (x, y) ∈ Rn and (up to reordering
variables) that the r × r matrix Dx F1 (0, 0) invertible. Let us set
ϕ : Rn → Rn , ϕ(x, y) = (F1 (x, y), y)
35
By construction Dϕ(0, 0) is invertible, hence ϕ is a local diffeo on a neighborhood U of (0, 0) (since
we assume that x0 = (0, 0) in the proof it is not restrictive to take U = U0 ). It is a direct check to
see that the composition F ◦ ϕ−1 has the form
F ◦ ϕ−1 (x, y) = (x, Fe2 (x, y))
for a suitable map Fe2 : Rr × Rn−r → Rm−r . Since a local diffeomorphism does not change the rank,
it holds rank D(F ◦ ϕ−1 )(x, y) = r on U . Writing down explicitly D(F ◦ ϕ−1 )(x, y), one easily see
that this implies that Fe2 does not depend on y, i.e.,
F ◦ ϕ−1 (x, y) = (x, Fe2 (x)).
By setting ψ : Rm → Rm as ψ(u, v) = (u, v − Fe2 (u)) (notice that x and u both belong to Rr , the
space where Fe2 is defined) then
ψ ◦ F ◦ ϕ−1 (x, y) = ψ(x, Fe2 (x)) = (x, Fe2 (x) − Fe2 (x)) = (x, 0)
which is equivalent to the statement.
Remark 3.6. The statement is local. In particular it can be applied to a smooth map defined on
an open set Ω ⊂ Rn .
Remark 3.7. Notice the following two particular cases of Theorem 3.5: if F is an immersion then
r = n and there exists local diffeomorphisms ϕ, ψ such that
ψ ◦ F |U ◦ ϕ−1 (x1 , . . . , xn ) = (x1 , . . . , xn , 0, . . . , 0). (3.1)
If F is a submersion then r = m and there exist local diffeomorphisms ϕ, ψ such that
ψ ◦ F |U ◦ ϕ−1 (x1 , . . . , xn ) = (x1 , . . . , xm ). (3.2)
Notice that (3.1) is the canonical linear immersion of Rn into Rn × Rm−n (for n < m), and (3.2) is
the canonical linear projection of Rn−m × Rm into Rm (for n > m).
We get the following theorem for manifolds just by applying the previous discussion.
Theorem 3.8 (Rank theorem for manifolds). Let F : M → N be smooth map between manifolds
of constant rank r in a neighborhood of a point q. Then there exist a chart (U, ϕ) around q a chart
(V, ψ) around F (q) such that the coordinate representation Fb = ψ ◦ F ◦ ϕ−1 of F is given by
Fb(x1 , . . . , xn ) = (x1 , . . . , xr , 0, . . . , 0).
Proof. Let us consider coordinates (U, ϕ) centered at q and (V, ψ) centered at F (q). Then Fb := ψ ◦
F |U ◦ ϕ−1 : φ(U ) ⊂ Rn → ψ(V ) ⊂ Rm by definition is a smooth map such that rank(DFb(0, 0)) = r.
Applying Theorem 3.5 to Fb there exists local diffeo ϕ0 , ψ 0 such that
ψ 0 ◦ Fb ◦ (ϕ0 )−1 (x1 , . . . , xn ) = (x1 , . . . , xr , 0, . . . , 0).
Writing
ψ 0 ◦ Fb ◦ (ϕ0 )−1 = ψ 0 ◦ ψ ◦ F |U ◦ ϕ−1 ◦ (ϕ0 )−1
and using as charts Ψ := ψ 0 ◦ ψ and Φ = ϕ0 ◦ ϕ the statement is proved.
36
A series of consequences descend from this result. We can say that immersions are locally injec-
tive, submersions are locally surjective, where the word “locally” should be properly understood.
The following is a more formal statement and the reader is invited to check the details.
Proposition 3.9. If F : M → N is an immersion then every point q ∈ M has a neighborhood

such that F |U : U → N is an embedding. If F : M → N is a submersion then F is an open map.
Combining the above considerations, also the following corollary is immediate.
Corollary 3.10 (Inverse function theorem for manifolds). Let F : M → N be smooth and q ∈ M
such that F∗ : Tq M → TF (q) N is an isomorphism. Then F is a local diffeomorphism at q.
The following result also holds (but it works only for constant rank maps!)
Corollary 3.11. Let F : M → N be smooth map of constant rank r between manifolds . Then
(i) if F is injective then it is an immersion.
(ii) if F is surjective then it is a submersion.
The reader is invited to deduce (i) from the rank theorem. The proof of (ii) requires the
definition of a “set of measure zero” on a manifold, it is not difficult but at the moment it is
omitted.
3.3 Submanifolds
We start with the following observation: an injective immersion F defines a smooth structure on
the image of F . The proof is left to the reader.
Lemma 3.12. Let F : M → N be an injective immersion. Given a smooth atlas {(Ui , ϕi )}i∈N of
M prove that {(F (Ui ), ϕi ◦ F |−1
Ui )}i∈N is a smooth atlas for F (M ).
Notice that with this smooth structure F : M → F (M ) is a diffeomorphism.
Definition 3.13. An immersed submanifold of a smooth manifold M is a subset S ⊂ M such

that S is endowed with a smooth manifold structure such that the canonical injection i : S → M
is a smooth immersion. The dimension of an immersed submanifold S is the dimension of S as a
smooth manifold.
More or less by definition we have the following property.
Proposition 3.14. Immersed submanifold are precisely images of smooth injective immersions.
Proof. If S is immersed submanifold then S is the image of the canonical injection i : S → M ,

which is a smooth injective immersion by definition.
We have to prove that given F : M → N an injective immersion, the canonical inclusion
i : F (M ) → N is an immersion, where F (M ) is endowed with the smooth structure of Lemma 3.12.
But this is the composition of the map F −1 |F (M ) : F (M ) → M which is a smooth diffeomorphism
and F : M → N which is an injective immersion.
37
A question is then whether the structure of manifold F (M ) has from the injective immersion
“agrees” with the smooth structure F (M ) might inherit from N .
Another simple but crucial observation before moving to the following definition.
Lemma 3.15. Let F : M → N be an embedding. For every point F (q) in F (M ) there exists a
chart (V, ψ) of N with V neighborhood of F (q) such that
ψ(V ∩ F (M )) = ψ(V ) ∩ (Rk × {0})
Proof. Let U ⊂ M be an open set containing q. Since F embedding then (crucial!) we have that
F (U ) = V ∩ F (M ) for some V open in N .
Since F is an immersion there exists charts (U, ϕ) and (V, ψ) such that
Fb = ψ ◦ F ◦ ϕ−1 : ϕ(U ) → ψ(V ), Fb(x1 , . . . , xk ) = (x1 , . . . , xk , 0 . . . , 0)
In particular we have ψ(F (U )) = ψ(V ∩ F (M )) = ψ(V ) ∩ (Rk × {0}).
Remark 3.16. If the inclusion i : S → M is an embedding, then renaming charts and applying the
above consideration: for every q ∈ S there exist a chart ϕ satisfying
ϕ(U ∩ S) = ϕ(U ) ∩ (Rk × {0})
This is the existence of a “slice chart” for S at every point.
Definition 3.17. An embedded submanifold of a smooth manifold M is a subset S ⊂ M such that

S is endowed with a smooth manifold structure such that the canonical injection i : S → M is a
smooth embedding.
Similarly as in Proposition 3.14 we have the following.
Proposition 3.18. Embedded submanifolds are precisely images of smooth embeddings.
Remark 3.19. (again on the difference immersed vs embedded) It is important to stress that given
S ⊂ M an immersed submanifold Proposition 3.9 guarantees that for every q ∈ M there exists a
neighborhood V of q in S such that V is an embedded submanifold. But in general it is not true
that there exists a neighborhood U of M such that U ∩ S is an embedded submanifold.
An easy consequence of the existence of slice charts is the following fact.
Proposition 3.20. Let f : M → R be smooth and let S be an embedded submanifold of M . Then

the restriction f |S : S → R is smooth.
We can also characterize the tangent space to embedded submanifolds from the algebraic view-
point.
Proposition 3.21. Let S ⊂ M be an embedded submanifold, i : S → M the inclusion and q ∈ S.

Then
i∗ (Tq S) = {v ∈ Tq M | vf = 0, ∀ f ∈ C ∞ (M ), f |S = 0}. (3.3)
Identifying elements of Tq S with i∗ (Tq S) ⊂ Tq M we can see Tq S as a vector subspace of Tq M .
38
Proof. Let us first show the inclusion ⊂ in (3.3). Let w ∈ Tq S let us prove that v = i∗ w satisfies
vf = (i∗ w)f = 0 for every f ∈ C ∞ (M ), f |S = 0. Indeed
(i∗ w)f = w(f ◦ i) = w(0) = 0.
Conversely let v ∈ Tq M such that vf = 0 for all f ∈ C ∞ (M ), f |S = 0. Then we want to show that
v = i∗ w for some w ∈ Tq S. It is easy to show that the set
W = {v ∈ Tq M | vf = 0, ∀ f ∈ C ∞ (M ), f |S = 0}
is a vector subspace of Tq M . Let q ∈ S and choose coordinates (U, ϕ) such that ϕ(S ∩ U ) =
ϕ(U ) ∩ (Rk × {0}). Then we have
ϕ(S ∩ U ) = {(x1 , . . . , xn ) ∈ ϕ(U ) | xk+1 = . . . = xn = 0}.
which can be expressed by saying that locally in coordinates
S = {(x1 , . . . , xn ) | xk+1 = . . . = xn = 0}.
which more formally means

Hence f |S = f (x1 , . . . , xk , 0, . . . , 0) and W is spanned by ∂/∂xj for j = 1, . . . , k and one sees in
coordinates that dim W = k = dim S. Hence W = Tq S.
This characterization is interesting from an abstract viewpoint, but not so convenient for com-
putations. Indeed what one does is to show that embedded submanifold can be always locally
described as regular level set of some map.
How to compute the tangent space to an immersion If F : M → N is an immersion then

locally we can look at it as F : Rk → Rn assume F (0) = x0 and let Σ = F (U ) so that
Tx0 Σ = F∗ T0 Rk
this is generated by the vectors {F∗ ∂ui }i=1,...,k . But we have
∂ ∂F
F∗ = (3.4)
∂ui ∂ui
Indeed
∂ ∂ X ∂g ∂Fj
F∗ = (g ◦ F ) =
∂ui ∂ui ∂yj ∂ui
j
this means exactly (3.4) as a vector in Rn
Level sets
The following results can be obtained simply by corresponding results in Rn and using charts, as
in the proof of Theorem 3.8.
Proposition 3.22. Let F : M → N be a smooth map of constant rank equal to k. Then each level
set F −1 (q), for q ∈ N , is a smooth submanifold of M of dimension n − k.
39
We have this corollary, which is the previous one with k = dim N .
Corollary 3.23. Let F : M → N be a smooth submersion. Then each level set F −1 (q), for q ∈ N ,
is a smooth submanifold of M of dimension dim M − dim N .
Notice that this result can be strenghtened considerably, since actually we only need to check
the assumption on the level set. Following the notation already introduced.
Definition 3.24. Let F : M → N be smooth. We say that q ∈ M is critical point if F∗ , q is not

surjective, q is a regular point otherwise. A point y ∈ N is said regular value if every q ∈ F −1 (y)
is a regular point.
Proposition 3.25. Let F : M → N be smooth. Assume q ∈ N is a regular value for F , then

F −1 (q) is a smooth submanifold of M of dimension dim M − dim N .
Proposition 3.26. Let S ⊂ M be an embedded submanifold of dimension k and q ∈ S. There

exists U ⊂ M neighborhood of q and Φ : U → Rn−k such that S ∩ U is a regular level set U ∩ Φ−1 (y)
for some regular value y in N
Proof. Work as in the proof of Lemma 3.15 and subsequent Remark. For every q ∈ S ⊂ M of
dimension k, there exists a chart (U, ϕ) such that ϕ(S ∩ U ) = ϕ(U ) ∩ (Rk × {0}).
Consider the map π : Rn → Rn−k which forgets the first k variables. Then π ◦ ϕ : U → Rn−k
and

π ◦ ϕ(S ∩ U ) = π ϕ(U ) ∩ (Rk × {0}) = {0}.
The map π is a submersion and ϕ a local diffeomorphism, so that Φ := π ◦ ϕ is a submersion.
We say that Φ : U → N is a local map defining S near q ∈ S if S ∩ U is a regular level set

U ∩ Φ−1 (y) for some regular value y in N . By the previous exercise there always exists local maps
defining S.
Proposition 3.27. Let S ⊂ M be an embedded submanifold. If Φ : U → N is any local map

defining S near q ∈ S then
Tq S = ker Φ∗ : Tq M → TΦ(q) N
Proof. Identifying as before Tq S with i∗ Tq S ⊂ Tq M under the inclusion map i : S → M . Notice

that Φ ◦ i is a constant map on S ∩ U , by definition. Hence Φ∗ ◦ i∗ = 0. Then imi∗ ⊂ ker Φ∗ . On
the other hand a dimensional count says that
dim ker Φ∗ = dim Tq M − dim TΦ(q) N = dim Tq S = dim im i∗ .
40
How to compute the tangent space of a submersion Add discussion here
Example 3.28 (Matrices of fixed rank). Recall that the set Mm,n k (R) of matrices of size m × n
and of rank k is an open set of Mm,n (R) when k = min{m, n} hence a submanifold of dimension
mn, or of codimension 0.
k (R) is an embedded
The goal of this exercise is to prove that for 0 < k < min{m, n} the set Mm,n
submanifold of Mm,n (R) of codimension (m − k)(n − k).
Let us consider the open subset of Mm,n (R) given by

A B
U= M= , det(A) 6= 0
C D
where the upper left A is of size k × k and the other ones accordingly to the fact that M is of size
m × n. Clearly matrices in U have rank ≥ r so that we are intersted in describing U ∩ Mm,n k (R).
Following the proof of the constant rank theorem: we consider the inverse of the map ϕ(x, y) =
(Ax + By, y). This means x0 = Ax + By and y 0 = y. Inverting, then y = y 0 and x = A−1 (x0 − By 0 ).
Then consider −1
−A−1 B

A
P =
0 I
and note that
I 0
MP =
CA−1 D − CA−1 B
Since M has rank k the same is true for M P being P invertible, hence D − CA−1 B should be the
zero matrix. This suggests to consider the map
Φ : U ⊂ Mm,n (R) → Mm−k,n−k (R), Φ(M ) = D − CA−1 B
and to observe that U ∩ Mm,nk (R) = Φ−1 (0). The map Ψ is a submersion if and only if for every
matrix X ∈ Mm−k,n−k (R) there exists a curve γ(t) ∈ U such that γ(0) = M0 and (Φ ◦ γ)0 (0) = X.
This is easily done by taking
A B
γ(t) =
C D + tX
Adjusting the proof for every minor (i.e., applying linear inverible maps to the open set U ), one
k (R) is a smooth submanifold of dimension
gets that that Mm,n
k
dim Mm,n (R) = nm − (n − k)(m − k).
An example to understand Let us consider S 2 (as a subset of R3 ) and consider the map
F : S 2 → R given by
F (x, y, z) = x
Find all regular values y of F and for every such y the set F −1 (y) is a circle. Notice that there are
two critical points in S 2 .
Exercise 3.29. Prove that the map G : R2 → R3 defined by
G(ϕ, θ) = ((2 + cos φ) cos θ, (2 + cos φ) sin θ, sin φ)
41
is an immersion. Show that G(R2 ) is contained in the set
T = {(x, y, z) | ((x2 + y 2 )1/2 − 2)2 + z 2 − 1 = 0}
Indeed G(R2 ) = T . Let Φ(x, y, z) = ((x2 + y 2 )1/2 − 2)2 + z 2 − 1. Then
T = {(x, y, z) | ((x2 + y 2 )1/2 − 2)2 + z 2 − 1 = 0} = Φ−1 (0)
and it is easy to see that DΦ(x, y, z) has maximal rank if (x, y) 6= (0, 0), which is true on T . Hence
T is an embedded 2-dimensional submanifold.
Let F : T → R defined by F (x, y, z) = x for every (x, y, z) ∈ T . Prove that F is smooth. Find
the critical points of F .
Exercise 3.30. Consider the function F : R2 → R defined by F (x, y) = x3 + 3xy + y 3 . For which
values of c ∈ R the set {(x, y) ∈ R2 | F (x, y) = c} is a smooth embedded submanfold?
3.4 Appendix
Proposition 3.31. Let α ∈ R \ Q. The curve γ : R → S 1 × S 1 given by γ(t) = (e2πit , e2πiαt ) has
dense image.
Proof. If we identify S 1 × S 1 with [0, 1]2 / ∼ with the identification of the boundary then γ is
rewritten as
γ(t) = (t, αt) mod 1.
It is not difficult to see that γ has dense image in [0, 1]2 / ∼ if and only if the sequence
(αn )n∈N ⊂ [0, 1], αn := nα mod 1
is dense in [0, 1].

Let us prove this fact: first of all we notice that αn 6= αm for n 6= m otherwise we would have
(n − m)α = k for some integer k, which is a contradiction with α ∈ R \ Q.
Moreover we notice that for n > m we have αn − αm = αn−m .
Let us show that 0 is an accumulation point for the sequence: for every N there exists n ∈ N such
that |αn | < N1 . Indeed consider the first N +1 element of the sequence α1 , . . . , αN +1 . By pigeonhole
principle there exists two integers n, m such that αn , αm ∈ [ Nk , k+1
N ] for the same k ∈ {0, . . . , N −1}.
Then the element x := αn−m = αn − αm belongs to [0, N1 ].
Then one can notice that for arbitrary p ∈ N we have pαn−m = αp(n−m) and then we can find
p = p(k) such that px stays in [ Nk , k+1
N ].
42
Chapter 4
Tangent bundle and vector fields
A smooth assignment of a tangent vector to each point of a manifold is called a vector field. This
seems very clear, up to the moment you start to ask yourself: what does it mean smooth here?
These kind of situations puzzle the modern mathematician which then feel more confortable
with the following neat definition:
A smooth vector field is a smooth section of the tangent bundle.
The next pages are devoted to give a meaning to the last sentence.
4.1 The tangent bundle

The tangent bundle is defined as the disjoint union
[
TM = Tq M
q∈M
endowed with the natural projection
π : T M → M, π(v) = q, if v ∈ Tq M
Proposition 4.1. T M has the natural structure of smooth manifold with dim T M = 2 dim M .
Proof. Consider charts {(Ui , ϕi )} on M . Then π −1 (Ui ) is an open set of T M and the reunion of
these sets is an open cover of T M . We define coordinates
ϕi : π −1 (Ui ) → ϕi (Ui ) × Rn
as follows
ϕi (v) = (x1 , . . . , xn , v1 , . . . , vn )
if π(v) = q with ϕi (q) = (x1 , . . . , xn ), and moreover
n
X ∂
v= vi
∂xi q
i=1
43
Notice that if Ui ∩ Uj 6= ∅ then π −1 (Ui ) ∩ π −1 (Uj ) 6= ∅ and
ϕi ◦ ϕ−1 n
j : ϕj (Ui ∩ Uj ) × R → ϕi (Ui ∩ Uj ) × R
n
where ϕi ◦ ϕ−1 0 0
j (x, v) = (x , v ) if (cf. with Exercise 2.13)
∂(ϕi ◦ ϕ−1
j )
x0 = ϕi ◦ ϕ−1
j (x), v0 = v (4.1)
∂x
which completes the proof.
Notice that the projection π : T M → M is smooth in the atlas defined above since in coordinates
it is just the linear projection (x1 , . . . , xn , v1 , . . . , vn ) 7→ (x1 , . . . , xn ).
We can treat the collection of the differentials at different points as a single map F∗ : T M → T N .
The previous computations shows that.
Corollary 4.2. Let F : M → N be smooth. Then F∗ : T M → T N is smooth.
If in coordinates F : M → N is smooth and X = ni=1 Xi ∂x

∂
P
i
then
n n
X ∂(g ◦ F ) X X m n
X ∂g ∂g ∂Fj
Yj = Xi = Xi
∂yj ∂xi ∂yj ∂xi
j=m i=1 j=1 i=1
so that the statement follows from formulas

n n n
! !
X ∂ X X ∂Fj ∂
F∗ Xi = Xi
∂xi ∂xi ∂yj
i=1 j=1 i=1
Write down the details of the corollary as an exercise if you feel that is not clear.
Example 4.3 (Tangent bundle T S 1 ). We prove that the tangent bundle T S 1 to the circle S 1 is
diffeomorphic to S 1 × R. Let us identify
S 1 = {z ∈ C : |z| = 1}
and first notice that T1 S 1 = iR. This is a simple consequence of the following computation: let
γ(t) = eiθ(t) for some smooth function θ with θ(0) = 0. Then γ̇(0) = iθ̇(0) ∈ iR.
Given v ∈ Tz S 1 we notice that z −1 v ∈ iR. Indeed let γ(t) = eiθ(t) for some smooth function θ
with θ(0) = θ0 with z = γ(0) and v = γ̇(0). Then
v = γ̇(0) = iθ̇(0)eiθ(0) = iθ(0)z, ⇒ z −1 v ∈ iR.
It follows that we can define the map (identifying iR ' R)
Ψ : T S 1 → S 1 × R, v 7→ (z, z −1 v)
where z = π(v). The verification that Ψ is a diffeomorphism is left to the reader.
Notice that S 1 is a Lie group. Where did we use the Lie group structure here?
44
4.2 Vector bundles
We start by the definition of vector bundle.
Definition 4.4. Let M be a smooth n-dimensional manifold. A smooth vector bundle of rank
k over M is a smooth manifold E of dimension n + k together with a smooth surjective map
π : E → M such that
(i) for each q ∈ M the set Eq := π −1 (q) is a vector space of dimension k
(ii) for every q ∈ M there exists a neighborhood U ⊂ M and a diffeomorphism Φ : π −1 (U ) →

U × Rk such that1 π = pr1 ◦ Φ and Φ|Eq : Eq → {q} × Rk ' Rk is a linear isomoprhism.
We say that E is the total space, M the base of the vector bundle, and π the projection. The
diffeomorphism Φ : π −1 (U ) → U × Rk is called local trivialization. If one can choose U = M then
E is diffeomorphic to M × Rk and we say that the vector bundle is trivial
Technically one should write that a vector bundle is a triple (E, M, π). For simplicity we will
also say that π : E → M (or even E → M ) is a vector bundle.
Lemma 4.5. Let M be a smooth manifold. Then T M is a vector bundle of rank equal to dim M .
Proof. Given any smooth chart (U, ϕ) for M with coordinates {xi } we set Φ : π −1 (U ) → U × Rn as
n
!
X ∂
Φ vi = (q, v1 , . . . , vn )
∂xi q
i=1
We notice that (ϕ × id) ◦ Φ = ϕ where ϕ × id : U × Rn → ϕ(U ) × Rn is defined as (q, v) 7→ (ϕ(q), v).

Since both ϕ × id and ϕ are diffeomorphisms, it follows that Φ is a diffeomorphism as well. We
have build local trivializations. The reader is invited to check all others requirements.
If a bundle is not trivial then we need more than one local trivialization.
Proposition 4.6. Let π : E → M be a smooth vector bundle of rank k and let Φ1 : π −1 (U1 ) →
U1 × Rk and Φ2 : π −1 (U2 ) → U2 × Rk be two local trivialization with U1 ∩ U2 6= ∅. Then
Φ2 ◦ Φ−1 k
1 : (U1 ∩ U2 ) × R → (U1 ∩ U2 ) × R
k
writes as Φ2 ◦ Φ−1
1 (q, v) = (q, τ (q)v) where τ : (U1 ∩ U2 ) → GLk (R) is smooth.
Proof. Note that by construction pr1 ◦ Φ2 ◦ Φ−1 1 = pr1 hence Φ2 ◦ Φ−1

1 (q, v) = (q, σ(q, v)) with
k k
σ : (U1 ∩ U2 ) × R → R smooth. For a fixed q, the map v 7→ σ(q, v) is linear. Hence σ(q, v) = τ (q)v
for some matrix τ (q). We have just to prove that q 7→ τ (q) is smooth. But the coordinates
τ (q) = (τij (q)) satisfy τij (q) = pri (σ(q, ej )) hence they are smooth2 .
The map τ is called transition function. Like for smooth manifolds we have a construction
lemma for vector bundles.
1
given a product X × Y , we denote pr1 : X × Y → X the projection onto the first factor.
2
here pri : Rn → R is the projection onto the i-th factor pri (x1 , . . . , xn ) = xi
45
Proposition 4.7. Let M be smooth manifold. Assume for every q ∈ M we have a k-dimensional
vector space Eq and define [
E= Eq , π:E→M
q∈M
where π is the natural projection. Then assume that we have
(i) an open cover {Ui }i∈N of M
(ii) diffeomorphisms Φi : π −1 (Ui ) → Ui × Rk that are linear isomorphisms on fibers
such that
(a) for every Ui ∩ Uj 6= ∅ there exists τij : Ui ∩ Uj → GLk (R) such that
Φj ◦ Φ−1
i (q, v) = (q, τij (q)v).
Then there exists a unique smooth structure on E such that π is smooth, E is a smooth vector
bundle of rank k over M , and {Φi }i∈N is a local trivialization. ***cocycle property***
Given a smooth vector bundle E → M , a (local) section of E is a map σ : U ⊂ M → E such
that π ◦ σ = idU . If U can be chosen equal to M , then σ is a global section.
Remark 4.8. Recall that given a vector bundle there exists always a particular section, which is the
zero global section. This is the section ζ : M → E defined as ζ(q) = 0q where 0q ∈ Eq is the origin
of the vector space.
Remark 4.9. The space of smooth functions C ∞ (M ) can be identified with the space of smooth
sections of the trivial vector bundle M × R of rank 1.
Definition 4.10. We say that a family of sections σ1 , . . . σr : U → E are independent if {σi (q)}i=1,...,r
are linearly independent as vectors in Eq for every q ∈ U . When r = rank(E) then we say that
{σ1 , . . . σr } is a local frame.
If Φ : π −1 (U ) → U × Rk is a local trivialization we can always define a local frame on U as
follows
σi (q) = Φ−1 (q, ei )
where ei is the canonical basis of Rk . Conversely if {σ1 , . . . σk } is a local frame on U then we can
build a local trivialization as follows
k
X
Ψ : U × Rk → π −1 (U ), Ψ(q, v) = vi σi (q).
i=1
We have proved the following fact.
Corollary 4.11. Every local frame for a smooth vector bundle is associated to a local trivialization.
A smooth vector bundle is trivial if and only if it admits a smooth global frame.
We observe that a local chart (U, ϕ) on M together with a local frame {σ1 , . . . σk } gives coor-
dinates to E
k
!
X
ϕ vi σi (q) = (ϕ(q), v1 , . . . , vk ) = (x1 (q), . . . , xn (q), v1 , . . . , vk ).
i=1
Remark 4.12. We observe that T S 1 is trivial, while the Möbius band is not.
46
4.3 Vector fields
Definition 4.13. A smooth vector field on an open set U ⊂ M is a smooth section of the tangent
bundle T M , i.e., a smooth map X : U → T M such that π ◦ X = idU .
The value at a point q of a vector field X is denoted X(q) or Xq and is an element of Tq M .
The set of all smooth vector fields in M is denoted Vec(M ).
Given cooordinate open set (U, ϕ) with coordinates {xi }, we notice that for every q ∈ U we
have a basis of the tangent space
∂ ∂
,...,
∂x1 q ∂xn q
and a smooth vector field X on U is decomposed along the canonical basis
n
X ∂
X(q) = Xi (q)
∂xi q
i=1
where Xi : U ⊂ M → R are functions defined on U .

Lemma 4.14. X is smooth in U if and only if Xi are smooth functions.
Proof. Taking the standard chart on π −1 (U ) ⊂ T M we have
ϕ(X(q)) = (x1 , . . . , xn , X
b1 (x), . . . , X
bn (x))
bi = Xi ◦ ϕ−1 is the coordinate representation of Xi .
where X
We can use bump functions to prove that
Exercise 4.15. Let q ∈ M and v ∈ Tq M . Then there exists X in Vec(M ) such that X(q) = v.
Given f ∈ C ∞ (M ) and X ∈ Vec(M ) we notice that f X is a new smooth vector field such that
(f X)(q) = f (q)X(q). Vec(M ) is a C ∞ (M )-module, i.e., a module over the ring C ∞ (M ).
Another way to “couple” vector fields and functions is to use X to differentiate smooth functions:
for f ∈ C ∞ (M ) we can set
Xf (q) = X|q f
where in the right hand side the vector differentiate the function.
Corollary 4.16. X is smooth if and only if Xf is smooth for every f ∈ C ∞ (M ).
Proof. Let
n
X ∂
X(q) = Xi (q)
∂xi q
i=1
If we have a chart (U, ϕ) and q = ϕ−1 (x) we have
n n
d(x) = Xf (ϕ−1 (x)) = Xϕ−1 (x) f =
X
−1 ∂ X
bi (x) ∂ f
b
Xf Xi (ϕ (x)) f= X
∂xi ϕ−1 (x) ∂xi x
i=1 i=1
Then if f is smooth clearly Xf is smooth. If we know that Xf is smooth for every f smooth, take
f = xj and you get that Xj (x) smooth.
47
Remark 4.17. We have proved exactly that the identity of vector fields
n
X ∂
X= Xi
∂xi
i=1
implies the identity of functions (in coordinates)

n
X ∂f
Xf = Xi
∂xi
i=1
where we have removed the “hat” in the notation. In what follows we keep this spirit.
A vector field hence induces a linear map X : C ∞ (M ) → C ∞ (M ) which satifies
X(f g) = f · Xg + g · Xf
where · here is usual multiplication of functions. Every vector field is a derivation of the algebra
C ∞ (M ). Indeed one can prove the converse: every derivation of the algebra C ∞ (M ) is a vector
field!
Example 4.18. Graphical example of T S 1 and a vector field on it
Vector fields and smooth maps

If F : M → N is a smooth map and X is a vector field on M then for every q ∈ M we can define
F∗ Xq which is an element of TF (q) N . This in general does not define a vector field on N . (Think
for instance the case when F is not injective)
If X ∈ Vec(M ) and Y ∈ Vec(N ) we say that Y is F -related to X if F∗ Xq = YF (q) for every
q ∈ M.
Exercise 4.19. Prove that F : R → R2 given by F (t) = (cos t, sin t) then ∂/∂t is F -related to
−y ∂/∂x + x ∂/∂y.
Lemma 4.20. Let F : M → N be a smooth map, X ∈ Vec(M ), Y ∈ Vec(N ). Then Y is F -related
to X if for every g ∈ C ∞ (N )
Y g ◦ F = X(g ◦ F ).
When F is a diffeomorphism then F∗ X is a well defined smooth vector field on N , called the
push-forward of X via F , and satisfies the following identity
(F∗ X)g = X(g ◦ F ) ◦ F −1
Notice that, at the level of vectors, we have (F∗ X)q g = XF −1 (q) (g ◦ F ).
n n n
! !
X ∂ X X ∂Fj ∂
F∗ Xi = Xi
∂xi ∂xi ∂yj
i=1 j=1 i=1
Exercise 4.21. Let F : Rx → R+ x

y F (x) = e then F∗ (∂/∂x) = y∂/∂y
48
4.4 Lie brackets
If one thinks at a vector field as a first order differential operator, one might ask if (or when) a
Schwartz-like formula holds: give two vector fields X, Y in Vec(M ), when is it true that XY f =
Y Xf for every f ∈ C ∞ (M )?.
Remark 4.22. Notice that in general the map D : C ∞ (M ) → C ∞ (M ) defined by D = XY is not a
derivation. A simple check is with M = R2 , X = ∂/∂x, Y = ∂/∂y and f (x, y) = x, g(x, y) = y.
Exercise 4.23. Let X, Y in Vec(M ), prove that the operator D : C ∞ (M ) → C ∞ (M ) defined by

D = XY − Y X is a derivation of C ∞ (M ) hence cooresponds to a vector field.
The first think we can do is to define the Lie bracket
Definition 4.24. We define the Lie bracket between X, Y in Vec(M ) as the vector field corre-
sponding to the derivation [X, Y ] := XY − Y X
Lemma 4.25. Let (U, {xi }) be a coordinate set and X, Y in Vec(M ). Then if on U we have
n n
X ∂ X ∂
X= Xi (x) , Y = Yi (x)
∂xi ∂xi
i=1 i=1
we have
n
X ∂Yj ∂Xj ∂
[X, Y ] = Xi (x) (x) − Yi (x) (x)
∂xi ∂xi ∂xj
i,j=1
Proof. Exercise
Exercise 4.26. Compute the Lie bracket [X, Y ] between the two vector fields in R3 with coordi-
nates (x, y, z)
∂ y ∂ ∂ x ∂
X= − , Y = +
∂x 2 ∂z ∂y 2 ∂z
and
∂ ∂ ∂ ∂
X = cos z + sin z , Y = − sin z + cos z
∂x ∂y ∂x ∂y
The Lie bracket is clearly a bilinear skew-symmetric form on Vec(M ) (as R-vector space). It
enjoys also more interesting properties related to the fact that Vec(M ) is a module over C ∞ (M )
(recall that C ∞ (M ) is an associative R-algebra).3
Proposition 4.27. The Lie bracket satisfies for X, Y, Z in Vec(M ) and f in C ∞ (M )
(i) [X, Y ] = −[Y, X]

3
recall that an associative algebra A is endowed with two compatible operations addition, multiplication (assumed
to be associative), and a scalar multiplication by elements in some field K. The addition and multiplication operations
together give A the structure of a ring; the addition and scalar multiplication operations together give A the structure
of a vector space over K. A commutative algebra is an associative algebra that has a commutative multiplication, or,
equivalently, an associative algebra that is also a commutative ring.
49
(ii) [X + Y, Z] = [X, Z] + [Y, Z]
(iii) [f X, Y ] = f [X, Y ] − (Y f )X
(iv) [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0
Proof. (i) and (ii) are direct consequences of the definition.

(iii) It is enough to prove the statement by applying both sides to every smooth g ∈ C ∞ (N )
and then apply (i) and (ii) to get the general statement. For every g ∈ C ∞ (M ) we have
[f X, Y ]g = f · XY g − Y (f · Xg) = f · XY g − Y f · Xg − f · Y Xg = f [X, Y ]g − Y f · Xg
which proves the statement. We used a f · g as product between smooth functions f and g. (iv)
The proof is a simple check applying the whole expansion to a smooth function f and expanding,
it is left as an exercise.
Remark 4.28. Collecting the different properties we have also for every X, Y in Vec(M ) and every
f, g in C ∞ (M )
[f X, gY ] = f g[X, Y ] − g(Y f )X + f (Xg)Y. (4.2)
Lemma 4.29. Let F : M → N be a diffeomorphism. Then for every X1 , X2 in Vec(M ) we have
F∗ [X1 , X2 ] = [F∗ X1 , F∗ X2 ]. (4.3)
Proof. This is a particular case of the more general fact: if F is smooth and X1 , X2 are F -related
with Y1 Y2 then [X1 , X2 ] is F -related to [Y1 , Y2 ]. Then the statement follows when F is a diffeomor-
phism and Yi = F∗ Xi for i = 1, 2. To prove this claim recall that by Lemma 4.20 we have
Yi g ◦ F = Xi (g ◦ F ), i = 1, 2,
hence
Xi Xj (g ◦ F ) = Xi (Yj g ◦ F ) = Yi Yj g ◦ F, i = 1, 2,
by linearity
[X1 , X2 ](g ◦ F ) = [Y1 , Y2 ]g ◦ F
which is the conclusion.
Exercise 4.30. Prove that if X(x) = Ax and Y (x) = Bx are linear vector fields in Rn then
[X, Y ](x) = [A, B]x where [A, B] = AB − BA is the commutator of matrices.
50
Chapter 5
Integral curves and flows
Knowing what is big and what is small is more important

than being able to solve differential equations
Stanislaw Ulam (1909–1984)
In this chapter we discuss integral curves and flows of vector fields, thanks to which we can give
a more geometric interpretation of the Lie bracket.
5.1 Integral curves and flows

Definition 5.1. Let M be a smooth manifold and X ∈ Vec(M ). An integral curve of the vector
field X is a smooth curve γ : J → M , where J ⊂ R is an open interval, such that
γ̇(t) = X(γ(t)), ∀ t ∈ J. (5.1)
Take a vector field X defined on M and a chart (U, ϕ). Write X in coordinates, i.e., consider
the vector field ϕ∗ X
n
bi (x) ∂
X
X
b = ϕ∗ X = X
∂xi
i=1
b in Rn
Consider a solution x(t) to the ODE associated to X
ẋ = X(x)
b
defined on some open interval J containing 0. Recall that this means for every t ∈ J the curve x(t)
satisfies in Rn the system of autonomous differential equations
ẋi (t) = X
bi (x1 (t), . . . , xn (t)), i = 1, . . . , n
Then it is easy to see that γ(t) = ϕ−1 (x(t)) is an integral curve of X. Indeed

γ̇(t) = ϕ−1 −1 b
∗ (ẋ(t)) = ϕ∗ X(x(t)) = (ϕ−1 −1
∗ X)(ϕ (x(t))) = X(γ(t)).
b
A consequence of the classical existence uniqueness theorem for solution of ODEs ensures that,
for every initial condition, there exists a unique integral curve of a smooth vector field, defined on
some open interval.
51
Theorem 5.2. Let X ∈ Vec(M ) and fix t0 ∈ R, q0 ∈ M . Then there exists a unique integral curve
γ : I → M of X such that γ(t0 ) = q0 defined on some maximal open interval I containing t0 .
Remark 5.3. Notice that if γ : I → M is an integral curve then γc : Ic → M defined by γc (t) =
γ(t − c) is also an integral curve (defined for t ∈ Ic := I + c).
Hence we can can always shift initial time and consider integral curves γ : I → M where I is
an open interval containing 0 and γ(0) = q0 . Notice that this is a consequence of the fact that
the vector field is autonomous, i.e., the right hand side of the corresponding ODE does not depend
explicitly on t.
Since vector fields under consideration are smooth, the corresponding ODEs in Rn have smooth
coefficients. Thus we have not only existence and uniqueness of solutions but also smoothness with
respect to initial data. This is translated into the following result.
Theorem 5.4. Let X ∈ Vec(M ). There exists an open set U ⊂ R × M containing {0} × M and a
map ΦX : U → M of class C ∞ such that
(a) for q ∈ M the set I q = {t ∈ R | (t, q) ∈ U} open neighborhood of 0,
(b) for q ∈ M the curve γ q : I q → M given by γ q (t) = ΦX (t, q) integral curve of X, γ q (0) = q,
X (s,q)
(c) for every s ∈ I q and t ∈ I Φ we have t + s ∈ I q and
ΦX (t, ΦX (s, q)) = ΦX (t + s, q)
We denote by ΦX X
t := Φ (t, ·) the flow of X at time t, which is a map defined on the open set
U t = {q ∈ M | (t, q) ∈ U}. Then the property (c) is rewritten as ΦX X X
t ◦ Φs = Φt+s .
Remark 5.5. It will be also convenient to use the exponential notation ΦX tX
t = e , in such a way
that the property (c) is written as
etX ◦ esX = e(t+s)X .
Notice that by definition for every t ∈ I q we have1
d tX d
e (q) = X(etX (q)), X(q) = etX (q).
dt dt t=0
Remark 5.6. When X(x) = Ax is a linear vector field on Rn , where A is a n × n matrix, the
corresponding flow is the matrix exponential ΦX tA
t (x) = e x.
Example 5.7. Let us compute the integral curves and flow of the vector field
X = x∂x + y∂y .
Given (x0 , y0 ) in R2 the integral curve which passes through (x0 , y0 ) at t = 0 is the solution

ẋ = x

ẏ = y

x(0) = x0 , y(0) = y0

The solution is easily computed (x(t), y(t)) = (x0 et , y0 et ). If (x0 , y0 ) = (0, 0) the vector field is zero
at that point and the corresponding integral curve is constant.
1 d tX
notice the formula dt
e = XetX (this should be compared with the more formal (5.9))
52
Appendix: on completeness
A vector field X ∈ Vec(M ) is called complete if, for every q0 ∈ M , the maximal solution of the
equation is defined on I = R.
The classical theory of ODE ensures completeness of the vector field X ∈ Vec(M ) in the
following cases:
(i) M is a compact manifold,
(ii) X has compact support in M ,
(iii) M = Rn and X has sub-linear growth at infinity, i.e., there exists C1 , C2 > 0 such that
kX(x)k ≤ C1 kxk + C2 , ∀ x ∈ Rn .
where k · k denotes the Euclidean norm in Rn .
When we are interested in the behavior of the trajectories of a vector field X ∈ Vec(M ) in a
compact subset K of M , the assumption of completeness is not restrictive.
Indeed consider an open neighborhood U of a compact K in M . There exists a smooth bump
function ψ : M → R that is identically 1 on K, and that vanishes out of U . Then the vector field
ψX is complete, since it has compact support in M . Moreover, the vector fields X and ψX coincide
on K, hence their integral curves coincides on K as well.
Example 5.8. Let f : M → R be a smooth function and X ∈ Vec(M ) be a complete vector field.
Denote by γ an integral curve of X and let ϕf : R → R be a solution of the differential equation
ϕ̇(t) = f (γ(ϕ(t)). Prove that the curve defined by γf (t) = γ(ϕf (t)) is an integral curve of f X.
Pushforward and flows

If F : M → N is a diffeomorphisms and X ∈ Vec(M ), then F∗ X is the vector field whose integral
curves are the image under F of integral curves of X. The vector field F∗ X is also called the
pushforward of X through F .
Lemma 5.9. Let F : M → N be a diffeomorphisms, X ∈ Vec(M ). Then
etF∗ X = F ◦ etX ◦ F −1 , (5.2)
Proof. Given y ∈ N consider the curve η(t) = F ◦ etX ◦ F −1 (y), we want to prove that η(t) is an
integral curve of F∗ X. Indeed
d d
η̇(t) = F ◦ etX ◦ F −1 (q) = F ◦ eεX ◦ etX ◦ F −1 (q) =
dt dε ε=0
−1
= F∗ (X(e tX
◦F (q))) = (F∗ X)(F ◦ etX ◦ F −1 (q)) = (F∗ X)(η(t))
This proves that
F ◦ etX ◦ F −1 (q) = η(t) = etF∗ X (q)
for every q ∈ N , which is the statement.
Along the lines we also proved that for q ∈ N
d d
(F∗ X)(q) = etF∗ X (q) = F ◦ etX ◦ F −1 (q), (5.3)
dt t=0 dt t=0
53
5.2 Lie derivatives and Lie brackets
In this section we introduce the Lie derivative of Y in the direction of X.
Definition 5.10. Let X, Y ∈ Vec(M ). We define the Lie derivative of Y wrt X as the vector field
∂
LX Y := e−tX
∗ Y (5.4)
∂t t=0
More precisely, this means that for every q ∈ M we define
1 −tX
LX Y (q) = lim e∗ (YetX (q) ) − Yq
t→0 t
Notice that for every t the vector e−tX

∗ (YetX (q) ) belong to Tq M hence the limit makes sense in Tq M .
Remark 5.11. The geometric meaning of the Lie bracket can be understood by writing explicitly
∂ ∂ ∂
LX Y q
= e−tX
∗ Y q
= e−tX
∗ (Y etX (q)
)= e−tX ◦ esY ◦ etX (q). (5.5)
∂t t=0 ∂t t=0 ∂s∂t t=s=0
Remark 5.12. We have used that for X, Y ∈ Vec(M ) we can reinterpret the pushforward of Y with
respect to X as follows:
d
(etX = etX etX ◦ esY ◦ e−tX (q).

∗ Y) q ∗ Y e−tX (q)
= (5.6)
ds s=0
where we used that

d
Y (e−tX (q)) = esY ◦ e−tX (q)
ds s=0
and the definition of pushforward.

The main goal of the section is to prove the following result, which is not evident at a first
glance.
LX Y = [X, Y ] (5.7)
Before going into the proof of the theorem we need some preliminary observations.
Vector fields as operators on functions

The action of a vector field X ∈ Vec(M ) on the algebra C ∞ (M ) of smooth functions on M can be
rewritten as follows
d
Xf (q) = f (etX (q)), q ∈ M. (5.8)
dt t=0
In other words Xf is the derivative of the function f along the integral curves of X.
Notice that given a point q there exists an open neighborhood U of (0, q) ∈ R × M where the
function f (t, q) := f (etX (q)) is defined and smooth.
The next statement makes more precise in which sense Xf is the first order term in the expansion
of f ◦ etX with respect to t.
54
Lemma 5.13. Let us denote f : U → R the function f (t, q) := f (etX (q)). Then we can write
f (t, q) = f (q) + g(t, q)t
where g : U → R is smooth (both in q and t) and satisfies g(0, q) = Xf (q).
Proof. Let us fix (t, q) ∈ U and consider F : R → R defined by F (s) := f (st, q). The fundamental
theorem of calculus Z 1
d
F (1) − F (0) = F (s)ds
0 ds
is rewritten using that d

ds F (s) = t ∂f
∂t (st, q) as follows
Z 1
∂f
f (t, q) = f (0, q) + t (st, q)ds
0 ∂t
R1 ∂f
which proves the statement with g(t, q) := 0 ∂t (st, q)ds which is smooth. Clearly we have also
g(0, q) = ∂f
∂t (0, q) = Xf (q).
Remark 5.14. The previous result intuitively says that t 7→ ft = f ◦ etX is “smooth” (but formally
we should give a smooth structure to C ∞ (M )) and states that Xf represents the first order term
in the expansion of ft when t → 0.
We can state this in the following manner: for q ∈ M one has
ft (q) = f (q) + t (Xf )(q) + r1 (t, q)
where supq∈U t−1 |r1 (t, q)| → 0 for t → 0 (and the same is true also for every spatial derivative).
We can write this as
ft (q) = f (q) + t (Xf )(q) + o(t).
Since the remainder is locally uniform, we can interpret this as the identity of functions
ft = f + t (Xf ) + o(t)
Exercise 5.15. Let f ∈ C ∞ (M ) and X ∈ Vec(M ), and denote ft = f ◦ etX . Prove the following
formulas2
d
ft = Xft , (5.9)
dt
t2 t3 tk
ft = f + t Xf + X 2 f + X 3 f + . . . + X k f + o(tk+1 ). (5.10)
2! 3! k!
where as before o(tk ) is a function rk (t, q) such that supq∈U t−k |rk (t, q)| → 0 for t → 0.
2
we can think to the formula
t2 2 t3 3 tk

f ◦ etX = Id + t X + X + X + . . . + X k + o(tk ) f
2! 3! k!
55
Theorem 5.16. The Lie derivative LX Y is a smooth vector field and, as derivations on functions,
it satisfies
LX Y = [X, Y ]. (5.11)
Proof. We want to prove that for every q ∈ M
∂
LX Y (q) = (e−tX
∗ Y )(q) = [X, Y ](q).
∂t t=0
Set g = f ◦ e−tX for f smooth and we have 3
(e−tX
∗ Y )f = Y (f ◦ e−tX ) ◦ etX = Y g ◦ etX = Y g + t(XY g) + o(t)
Now use that g = f − t(Xf ) + o(t) and
Y g = Y f − tY Xf + o(t)
XY g = XY f + o(1)
hence collecting the results we have
(e−tX
∗ Y )f = Y f + t[X, Y ]f + o(t)
and the statement is proved.
5.3 Lie brackets and commutativity of flows

Lemma 5.17. Let X, Y ∈ Vec(M ) be complete. Then the two properties are equivalent
(i) [X, Y ] = 0
(ii) for every t ∈ R we have e−tX

∗ Y = Y.
Proof. If e∗−tX Y = Y then by definition of Lie brackets clearly [X, Y ] = 0.

Assume now that [X, Y ] = dtd
e−tX Y = 0 and we want to prove that
t=0 ∗
d −tX
dt e∗ Y = 0 for all
t ∈ R. Indeed we have
d −tX d −(t+ε)X d
e Y = e∗ Y = e−tX e−εX Y
dt ∗ dε ε=0 dε ε=0
∗ ∗
d
= e−tX
∗ e−εX
∗ Y = e−tX
∗ [X, Y ] = 0,
dε ε=0
It follows that e−tX

∗ Y does not depend on t, hence it coincides with its value at t = 0 which is
e∗−0X Y = id∗ Y = Y .
Remark 5.18. Notice that since [X, Y ] = −[Y, X], then [X, Y ] = 0 is also equivalent to e−tY
∗ X=X
for every t ∈ R.
3
Setting h(t, q) := g(−t, q) we have f (e−tX (q)) = f (q) − h(t, q)t, with h smooth and h(0, q) = Xf (q).
56
To end this section, we show that the Lie bracket of two vector fields is zero (i.e., they commute
as operator on functions) if and only if their flows commute. We state this for complete vector fields
but indeed the result is local so we can apply the argument in the “Appendix: on completeness”
and the result is indeed general.
Proposition 5.19. Let X, Y ∈ Vec(M ) be complete. The following properties are equivalent:
(i) [X, Y ] = 0,
(ii) etX ◦ esY = esY ◦ etX , ∀ t, s ∈ R.

Proof. (i)⇒(ii). Fix t ∈ R. Let us show that φs := e−tX ◦ esY ◦ etX is the flow generated by Y .
Indeed we have
∂ ∂
φs (q) = e−tX ◦ e(s+ε)Y ◦ etX (q)
∂s ∂ε ε=0
∂
= e−tX ◦ eεY ◦ etX ◦ e|−tX ◦ {z
esY ◦ etX}(q)
∂ε ε=0 φs
= e−tX
∗ Y ◦ φs (q) = Y ◦ φs (q)
where in the last equality we used the previous lemma. Using uniqueness of the flow generated by
a vector field we get
e−tX ◦ esY ◦ etX = esY , ∀ t, s ∈ R,
which is equivalent to (ii).
(ii)⇒(i). Using that [X, Y ] = LX Y and the characterization (5.5) we have
∂ ∂
[X, Y ](q) = e−tX ◦ esY ◦ etX (q) = e−tX ◦ etX ◦ esY (q)
∂s∂t t=s=0 ∂s∂t t=s=0
∂
= esY (q) = 0.
∂s∂t t=s=0
Exercise 5.20. Let X, Y ∈ Vec(M ) and q ∈ M . Consider the curve on M
γ(t) = e−tY ◦ e−tX ◦ etY ◦ etX (q).
Prove that for every f ∈ C ∞ (M ) we have
f (γ(t)) = f (q) + t2 [X, Y ]f (q) + o(t2 )

√
This can be interpreted in the following way: the curve t 7→ γ( t) is differentiable at t = 0 (but
not smooth!), and its tangent vector at t = 0 is [X, Y ](q).
Exercise 5.21 (Another proof of Jacobi identity). Prove that the Lie bracket satisfies the Jacobi
identity:
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0. (5.12)
by differentiating the identity etX tX tX
∗ [Y, Z] = [e∗ Y, e∗ Z] with respect to t at t = 0.
57
Exercise 5.22. Let X, Y ∈ Vec(M ). Using the semigroup property of the flow, prove that
d −tX
e Y = e−tX [X, Y ]. (5.13)
dt ∗ ∗
Deduce the following formal series expansion

∞ n
X t
e−tX
∗ Y = (ad X)n Y (5.14)
n!
n=0
t2 t3
= Y + t[X, Y ] + [X, [X, Y ]] + [X, [X, [X, Y ]]] + . . .
2 6
where we have introduced the notation (ad X)Y := [X, Y ].
5.4 Left-invariant vector fields on a Lie group

A Lie algebra is a vector space g endowed with an operation
[·, ·] : g × g → g
that is bilinear, skew-symmetric and satisfies the Jacobi identity
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0
for every X, Y, Z ∈ g.
Remark 5.23. Given an associative algebra A with multiplication (x, y) 7→ xy denoted with juxta-
position, we can always give a Lie algebra structure to A by defining the Lie bracket
[x, y] = xy − yx
Indeed in this case the Jacobi identity is trivially satisfied.

The set of smooth vector fields Vec(M ) on a smooth manifold M is naturally a Lie algebra
using as operation the Lie brackets of vector fields.
If M is a Lie group we can give the following definition.
Definition 5.24. Let G be a Lie group and X a vector field on G. We say that X is left-invariant
if for every g ∈ G we have (Lg )∗ X = X, i.e., we have
(Lg )∗ Xh = Xgh , ∀ g, h ∈ G
Let us denote by VecL (G) the subset of left-invariant vector fields on G.
Proposition 5.25. The Lie bracket of left-invariant vector fields is left-invariant. Hence VecL (G)
is a Lie algebra.
Proof. If X anf Y are left-invariant then (Lg )∗ X = X and (Lg )∗ Y = Y for every g ∈ G. Hence
(Lg )∗ [X, Y ] = [(Lg )∗ X, (Lg )∗ Y ] = [X, Y ]
for every g, which proves that [X, Y ] is left-invariant.
58
Notice that a left-invariant vector field in particular satisfies the following property
Xg = (Lg )∗ Xe , ∀g ∈ G
where e is the indentity of the group G. Indeed this property characterizes left-invariant vector
fields since if this is true we have
(Lg )∗ Xh = (Lg )∗ (Lh )∗ Xe = (Lg ◦ Lh )∗ Xe = (Lgh )∗ Xe = Xgh
This says that a left-invariant vector field is characterized by its value at one point, for instance
the origin. Hence we have
Proposition 5.26. The map ε : VecL (G) → Te G given by ε(X) = Xe is an isomorphisms of vector
spaces. In particular dim(VecL (G)) = dim G.
The two spaces VecL (G) and Te G are then identified and we denote by g both the set of left-
invariant vector fields of G or, equivalently, the tangent space to the identity.
Remark 5.27. Analogously we can introduce right-invariant vector fields VecR (G). This space can
be also identified through right translations with the tangent space to the identity but notice that
in general a left-invariant vector field is not right-invariant.
It is a convention that the Lie algebra of a Lie group is the one of left-invariant vector fields.
Exercise 5.28. Consider in M = R2 × S 1 with coordinates (x, y, θ) the two vector fields
X = cos θ ∂x + sin θ ∂y , Y = ∂θ
1. Prove that [X, Y ] = sin θ ∂x − cos θ ∂y and give a geometrical interpretation of this fact in
terms of the Exercice 5.20.
2. Prove that X, Y, [X, Y ] are linearly independent at every point of R2 × S 1
3. Prove that the following is a group law in M
(x, y, θ) · (x0 , y 0 , θ0 ) = (x + (cos θ)x0 − (sin θ)y 0 , y + (sin θ)x0 + (cos θ)y 0 , θ + θ0 )
and that M is a Lie group.
4. Prove that X, Y are left-invariant.
59
Chapter 6
Vector distributions: integrability vs

non-integrability
If a notion bears a personal name, then this name

is not the name of the discoverer.
The Arnold Principle.
V.I. Arnold (1937–2010)
In this chapter, we discuss some results about integrability of vector distributions. The most
classical result is Frobenius theorem, stating necessary and sufficient conditions for a vector distri-
bution to be integrable, i.e., to admit a tangent submanifold.
We then discuss also distributions which satisfy an “opposite” assumption than the one of
Frobenius, i.e., bracket-generating distributions and the corresponding Chow theorem.
Despite being named for Ferdinand Georg Frobenius work in 1877, the two implications of
Frobenius theorem were proven by Feodor Deahna, 1840, and Alfred Clebsch, 1866.
Similarly Chow theorem is named after Wei-Liang Chow who proved it in 1939, but Petr Kon-
stanovich Rashevskii proved it independently in 1938. (cf. The Arnold principle)
6.1 Diffeomorphisms built with flows of vector fields

Most of the results of this chapter are based on local diffeomorphisms built with compositions of
flows of vector fields. We start with a proposition computing the differential of such a map.
Proposition 6.1. Let M be a smooth n-dimensional manifold and X1 , . . . , Xn be linearly indepen-
dent vector fields at a point q0 ∈ M . Then the map
ψ : Rn → M, ψ(t1 , . . . , tn ) = et1 X1 ◦ . . . ◦ etn Xn (q0 ),
is a local diffeomorphism at 0. Moreover we have, denoting t = (t1 , . . . , tn ),
∂ψ
t X

(t) = et∗1 X1 · · · e∗i−1 i−1 Xi (ψ(t)). (6.1)
∂ti
Proof. The map ψ is clearly smooth. It is easy to compute the differential of the map at t = 0.
∂ψ d d
(0) = ψ(0, . . . , 0, s, 0, . . . , 0) = esXi (q0 ) = Xi (q0 ) (6.2)
∂ti ds s=0 ds s=0
61
Hence the partial derivatives of ψ are linearly independent, thus ψ is a local diffeo. For t in
a neighborhood of 0 it is less trivial to compute the differential. First let us observe that by
definition of the map ψ we have that
e−ti Xi . . . ◦ e−t1 X1 (ψ(t)) = eti+1 Xi+1 . . . ◦ etn Xn (q0 ) (6.3)
Hence we have
∂ψ d
(t) = ψ(t1 , . . . , ti + s, . . . , tn )
∂ti ds s=0
d
= et1 X1 ◦ . . . ◦ e(ti +s)Xi ◦ . . . ◦ etn Xn (q0 )
ds s=0
d
= et1 X1 ◦ . . . ◦ eti Xi ◦ esXi ◦ eti+1 Xi+1 . . . ◦ etn Xn (q0 )
ds s=0
and using (6.3)
∂ψ d
(t) = et1 X1 ◦ . . . ◦ eti Xi ◦ esXi ◦ e−ti Xi . . . ◦ e−t1 X1 (ψ(t))
∂ti ds
s=0

t1 X1 tX
· · · et∗i Xi Xi (ψ(t)) = et∗1 X1 · · · e∗i i−1 Xi (ψ(t)).

= e∗
where in the last identity we used (5.6) and the fact that the differential of a composition is the
composition of differentials. Notice that etX
∗ X = X for every X.
6.2 Rectification of vector fields

As a direct consequence of the previous result, every vector field locally around a point which is
not singular can be “rectified”, in the sense that its flow in coordinates is given by straight lines.
Corollary 6.2 (Rectification of a vector field). Let X be a vector field on M and q0 ∈ M such
that X(q0 ) 6= 0. Then there exists coordinates (x1 , . . . , xn ) defined by ϕ : U ⊂ M → Rn on a
neighborhood U of q0 such that ϕ∗ X = ∂/∂x1 .
Proof. Let X1 := X be the first element of a family of vector fields X1 , . . . , Xn that are linearly
independent at q0 ∈ M . Notice that the existence of such a family is guaranteed by Exercise 4.15.
Let now ψ : Rn → M be the map build in Proposition 6.1 associated with X1 , . . . , Xn . Let
V ⊂ Rn be the neighborhood of q0 where ψ is a local diffeomorphism on U = ψ(V ). Notice that
the identity (6.1) for i = 1 gives
∂ψ
(t) = X1 (ψ(t)). (6.4)
∂t1
It follows that
∂
ψ∗ = X1 .
∂t1
Using ϕ = ψ −1 : U ⊂ M → V ⊂ Rn as a coordinate map we have ϕ∗ X1 = ∂/∂t1 .
62
Another consequence of Proposition 6.1 is a characterization of families X1 , . . . , Xn of vector
fields that can be simoultaneously “rectified”, i.e., if they appear as a family of coordinate vector
fields.
Theorem 6.3. Let M be a smooth n-dimensional manifold and X1 , . . . , Xn be linearly independent

vector fields at a point q0 ∈ M . Then there exists local coordinates ϕ : U ⊂ M → Rn in a
neighborhood U of q0 such that
∂
ϕ∗ Xi = , i = 1, . . . , n, (6.5)
∂xi
if and only if [Xi , Xj ] = 0 for every i, j = 1, . . . , n.
Proof. (i) If a local coordinate map ϕ : U → Rn satisfying (6.5) exists then

∂ −1 ∂ ∂ ∂
[Xi , Xj ] = ϕ−1
∗ , ϕ = ϕ−1
, =0
∂xi ∗ ∂xj ∗
∂xi ∂xj
since the coordinate vector fields commute.

(ii). Let us consider the map ψ associated to X1 , . . . , Xn and defined by
which is a local diffeomorphism at 0. For t = (t1 , . . . , tn ) small we have
∂ψ
t X

(t) = et∗1 X1 · · · e∗i−1 i−1 Xi (ψ(t)). (6.6)
∂ti
Remember that [X, Y ] = 0 implies etX ∗ Y = Y (cf. Lemma 5.17). Hence if we assume [Xi , Xj ] = 0
for every i, j = 1, . . . , n, then we have
∂ψ
(t) = Xi (ψ(t)). (6.7)
∂ti
and the statement is proved using ϕ = ψ −1 .
Example 6.4. Let us consider X = x∂x + y∂y and find coordinates that rectify X around (1, 0).
Notice that Y = −y∂x + x∂y is linearly independent from X at (1, 0).
Following the construction a map that rectify X is the inverse of the map
ψ : R2 → M, ψ(t1 , t2 ) = et1 X ◦ et2 Y (1, 0)
It is easy to compute
et2 Y (1, 0) = (cos t2 , sin t2 ), et1 X (x0 , y0 ) = (et1 x0 , et1 y0 )
Hence
ψ(t1 , t2 ) = (et1 cos t2 , et1 sin t2 )
One might check that ψ∗ ∂/∂t1 = X. It also holds ψ∗ ∂/∂t2 = Y . This last fact is a consequence of
[X, Y ] = 0.
63
6.3 Frobenius theorem
In this section we prove Frobenius theorem about vector distributions.
Definition 6.5. Let M be a smooth manifold. A vector distribution D of rank m on M is a family
of vector subspaces Dq ⊂ Tq M where dim Dq = m for every q.
A vector distribution D is said to be smooth if, for every point q0 ∈ M , there exists a neighbor-
hood U of q0 and a family of vector fields X1 , . . . , Xm defined on U such that
Dq = span{X1 (q), . . . , Xm (q)}, ∀ q ∈ U. (6.8)
Definition 6.6. A smooth vector distribution D (of rank m) on M is said to be involutive if for
every point q0 ∈ M , there exists a neighborhood U of q0 and a family of vector fields {X1 , . . . , Xm }
satisfying (6.8) such that
m
X
[Xi , Xk ] = akij Xj , ∀ i, k = 1, . . . , m. (6.9)
j=1
for some smooth functions akij on M .
Exercise 6.7. Prove that a smooth vector distribution D is involutive if and only if for every point
q0 ∈ M there exists a neighborhood U of q0 and for every family of vector fields {X1 , . . . , Xm }
satisfying (6.8) we have
m
X
[Xi , Xk ] = akij Xj , ∀ i, k = 1, . . . , m. (6.10)
j=1
for some smooth functions akij on M .

Definition 6.8. A smooth vector distribution D on M is said to be flat if for every point q0 ∈ M
there exists a neighborhood U of q0 and a local diffeomorphism ϕ : U → Rn such that ϕ∗,q (Dq ) =
Rm × {0} for all q ∈ U .
To prove Frobenius theorem we need a lemma
Lemma 6.9. Assume that D is involutive and locally spanned by X1 , . . . , Xm . Then for every
k = 1, . . . , m, we have etXk
∗ D = D.
Notice that this means that for every k = 1, . . . , m and q ∈ M we have
etX
∗ (Dq ) = DetXk (q)
k
Proof of Lemma 6.9. Let us define the time-dependent vector fields
Yik (t) := etX

∗ Xi .
k
Using (6.10) and (5.13) we compute

m
X Xm
Ẏik (t) = etX
∗ [Xi , Xk ] =
k
etX
∗
k
a k
X
ij j = akij (t)Yjk (t),
j=1 j=1
64
where we set akij (t) = akij ◦ e−tXk . More explicitly, this means that for every q ∈ M
m
d tXk X
(e∗ Xi )(q) = akij (e−tXk (q))(etX
∗ Xj )(q),
k
dt
j=1
Notice that for every fixed q ∈ M and k ∈ {1, . . . , m} this is a system of non-autonomous linear
differential equations in the vector space Tq M of the form
k
d X
xi (t) = aij (t)xj (t)
dt
j=1
Thus denote by Ak (t) = (akij (t))m

i,j=1 and consider for every k the unique solution to the matrix
Cauchy problem
Ṁ (t) = Ak (t)M (t), M (0) = I. (6.11)
which we denote by M k (t) = (mkij (t))m
i,j=1 . Then we have
m
X
Yik (t) = mkij (t)Yjk (0),
j=1
since Yjk (0) = Xj , this implies for every i, k = 1, . . . , m,

m
X
etX
∗ Xi =
k
mkij (t)Xj ,
j=1
which proves the claim.
Now we can prove the main result of this section.
Theorem 6.10 (Frobenius Theorem). A smooth distribution is involutive if and only if it is flat.
Proof. The statement is local, hence it is sufficient to prove the statement on a neighborhood U of
an arbitrary point q0 ∈ M . We can assume the vector fields are complete.
(i). Assume first that the distribution is flat. Then there exists a diffeomorphism ϕ : U → Rn
such that Dq = ϕ−1 m
∗,q (R × {0}). It follows that for all q ∈ U we have

−1 ∂
Dq = span{X1 (q), . . . , Xm (q)}, Xi (q) := ϕ∗,q .
∂xi
and we have for i, k = 1, . . . , m,

−1 ∂ −1 ∂ −1 ∂ ∂
[Xi , Xk ] = ϕ∗,q , ϕ∗,q = ϕ∗,q , = 0.
∂xi ∂xk ∂xi ∂xk
(ii). Let us now prove that if D is involutive then it is flat. As before, it is not restrictive to
work on a neighborhood U where
Dq = span{X1 (q), . . . , Xm (q)}, ∀ q ∈ U. (6.12)
65
and (6.10) are satisfied. Complete the family X1 , . . . , Xm to a basis of the tangent space
Tq M = span{X1 (q), . . . , Xm (q), Zm+1 (q), . . . , Zn (q)},
in a neighborhood of q0 and set ψ : Rn → M defined by
ψ(t1 , . . . , tm , sm+1 , . . . , sn ) = et1 X1 ◦ . . . ◦ etm Xm ◦ esm+1 Zm+1 ◦ . . . ◦ esn Zn (q0 ).
By construction ψ is a local diffeomorphism at (t, s) = (0, 0) and for (t, s) close to (0, 0) we have
that (cf. Proposition 6.1) for every i = 1, . . . , m
∂ψ
(t, s) = et∗1 X1 . . . e∗ti Xi Xi (ψ(t, s)),

∂ti
These vectors are linearly independent and, thanks to Lemma 6.9, belong to D. Hence

∂ ∂
Dq = ψ∗ span ,..., , q = ψ(t, s),
∂t1 ∂tm
and the claim is proved by considering ϕ := ψ −1 .
Reformulating Frobenius theorem in terms of submanifolds, one immediately obtains the fol-
lowing corollary.
Corollary 6.11. Let D be an involutive distribution of rank m on a smooth manifold M of dimen-

sion n ≥ m. Then, for every q ∈ M , there exists a (locally defined) submanifold S of dimension m
passing through q and that is tangent to D at every point, i.e., Tx S = Dx for every x ∈ S.
This in particular says that if we move from a given point q0 ∈ M only with curves that are
tangent to D, then we are confined in a m-dimensional submanifold S of M .
The converse (integrability implies involutivity) is also true thanks to the following lemma.
Lemma 6.12. Let N ⊂ M be an embedded submanifold and X, Y in Vec(M ) be vector fields

tangent to N , i.e., X(q), Y (q) ∈ Tq N , for every q ∈ N . Then [X, Y ] is tangent to N .
Proof. A vector field X in M is tangent to N if and only it is i-related to a vector field on N ,

i.e., there exists V such that X|N = i∗ V for V ∈ Vec(N ) where here i : N → M is the canonical
inclusion. If X|N = i∗ V and Y |N = i∗ W then [X, Y ]|N = [i∗ V, i∗ W ]|N = (i∗ [V, W ])|N = i∗ [V, W ]
hence [X, Y ] is also tangent to N .
6.4 Rashevski-Chow theorem: local version

We now state a theorem about start by introducing bracket-generating family of vector fields.
Definition 6.13. Let M be a smooth manifold and D a smooth vector distribution. D is said
bracket generating if for every local basis in a neighborhood U of q we have
Dq = span{X1 (q), . . . , Xm (q)}, ∀ q ∈ U. (6.13)
and
span{[Xi1 , . . . , [Xij−1 , Xij ]](q) : 1 ≤ i` ≤ m, ` ≤ j, j ∈ N} = Tq M (6.14)
66
We denote the left hand side by
Lieq D := span{[Xi1 , . . . , [Xij−1 , Xij ]](q) : 1 ≤ i` ≤ m, ` ≤ j, j ∈ N}
these are all possible Lie brackets of the basis evaluated at q.
Exercise 6.14. Show that Lieq D is well defined, i.e., does not depend on the choice of the basis.
A direct consequence of Lemma 6.12 is the following
Corollary 6.15. Let N ⊂ M be an embedded submanifold and D be a smooth vector distribution
such that Dq ⊂ Tq N for every q ∈ N . Then for every q ∈ N we have dim Lieq D ≤ dim N .
We can now prove the main technical lemma of this section
Lemma 6.16 (Rashevski-Chow lemma). Let M be an n-dimensional manifold and D locally
spanned by X1 , . . . , Xk be a bracket generating distribution of rank k.
Then for every q0 ∈ M and every neighborhood V of the origin in Rn there exist τ = (τ1 , . . . , τn ) ∈
V , and a choice of n vector fields Xi1 , . . . , Xin , such that τ is a regular point of the map
ψ : Rn → M, ψ(t1 , . . . , tn ) = etn Xin ◦ · · · ◦ et1 Xi1 (q0 ).
Remark 6.17. Observe that, if k < n, then τ 6= 0. Indeed, the image of the differential of ψ at t = 0
is
im ψ∗,0 = spanq0 {Xij | j = 1, . . . , n} ⊆ spanq0 {Xi | i = 1, . . . , k} = Dq0
and the differential of ψ cannot be surjective if Dq0 6= Tq0 M . In particular, in the choice of
Xi1 , . . . , Xin , the same vector field is allowed to appear more than once.
The following proof indeed works for any family of vector fields F = {X1 , . . . , XN } such that
Lieq F = Tq M , without necessarily asking that the vector fields are linearly independent.
Proof of Lemma 6.16. We prove the lemma by steps. Let F = {X1 , . . . , Xm }.
1. There exists a vector field Xi1 ∈ F such that Xi1 (q0 ) 6= 0, otherwise all vector fields in F
vanish at q0 and dim Lieq0 F = 0, which contradicts the bracket-generating condition. Then,
for |s| small enough, the map
φ1 : s1 7→ es1 Xi1 (q0 ),
is a local diffeomorphism onto its image Σ1 . If dim M = 1 the Lemma is proved.
2. Assume dim M ≥ 2. Then there exist t11 arbitrarily close to 0, and Xi2 ∈ F such that, if we
1
denote by q1 = et1 Xi1 (q0 ), the vector Xi2 (q1 ) is not tangent to Σ1 . Otherwise, by Lemma
6.12, dim Lieq0 F = 1, which contradicts the bracket-generating condition. Then the map
φ2 : (s1 , s2 ) 7→ es2 Xi2 ◦ es1 Xi1 (q0 ),
is a local diffeomorphism near (t11 , 0) onto its image Σ2 . Indeed the vectors
∂φ2 ∂φ2
∈ Tq1 Σ1 , = Xi2 (q1 ),
∂s1 (t11 ,0) ∂s2 (t11 ,0)
are linearly independent by construction. If dim M = 2 the Lemma is proved.
67
3. Assume dim M ≥ 3. Then there exist (t12 , t22 ) arbitrarily close to (t11 , 0), and Xi3 ∈ F such
2 1
that, if q2 = et2 Xi2 ◦ et2 Xi1 (q0 ) we have that Xi3 (q2 ) is not tangent to Σ2 . Otherwise, by
Lemma 6.12, dim Lieq1 D = 2, which contradicts the bracket-generating condition. Then the
map
φ3 : (s1 , s2 , s3 ) 7→ es3 Xi3 ◦ es2 Xi2 ◦ es1 Xi1 (q0 ),
is a local diffeomorphism near (t12 , t22 , 0). Indeed the vectors
∂φ3 ∂φ3 ∂φ3

, ∈ Tq2 Σ2 , = Xi3 (q2 ),
∂s1 (t12 ,t22 ,0) ∂s2 (t12 ,t22 ,0) ∂s3 (t12 ,t22 ,0)
are linearly independent since the last one is transversal to Tq2 Σ2 by construction, while the
first two are linearly independent since φ3 (s1 , s2 , 0) = φ2 (s1 , s2 ) and φ2 is a local diffeomor-
phisms at (t12 , t22 ) which is close to (t11 , 0).
Repeating the same argument n times (with n = dim M ), the lemma is proved.
Corollary 6.18. The map ψb : Rn → M defined by
b 1 , . . . , tn ) = e−τ1 Xi1 ◦ · · · ◦ e−τn Xin ◦ ψ(t1 , . . . , tn ),

ψ(t
c of τ ∈ V to a neighborhood of ψ(b
is a diffeomorphism from a neighborhood W b s ) = q0 .
Here sb = (b
s1 , . . . , sbn ))
Thanks to Lemma 6.16 there exists a neighborhood Vb ⊂ V of sb such that ψ is a diffeomorphism
from Vb to ψ(Vb ). We stress that in general q0 = ψ(0) does not belong to ψ(Vb ), cf. Remark 6.17.
Corollary 6.19. Let M be a smooth manifold and D be a bracket generating distribution.

Then D is totally non-integrable, i.e., there is no submanifold N of M such that Dq = Tq N for
every q in a neighborhood of a point.
This means that if we move from a given point q with curves that are tangent to D, then we
can reach an open neighborhood in M .
We have that D bracket-generating implies D non-integrable. The converse in full generality is
not true.
Example 6.20. Let us consider the smooth function

( 2
e−1/t , t 6= 0
φ(t) =
0, t=0
This is a C ∞ function that vanish at t = 0 with all derivatives equal to zero at that point.
Consider the rank 2 distribution D in M = R3 given by D = span{X, Y } with
X = ∂x , Y = ∂y + φ(x)∂z
It is easy to see that

[X, Y ] = φ0 (x)∂z , [Y, [X, Y ]] = 0
68
while, more in general,
[X, . . . , [X , Y ]] = φ(j) (x)∂z
| {z }
j
Hence the structure is not bracket generating on the plane P = {x = 0}. However, D is totally
non integrable since if N ⊂ R3 is an integral manifold then it should be contained in P = {x = 0}.
But if x = 0 then X is not tangent to P .
So non-integrable does not imply bracket generating at every point. One can prove the following
partial converse
(a) if D is non integrable then D is bracket-generating on an open dense set of M
(b) if M and D are analytic, D is non-integrable implies D bracket-generating at every point.
Exercise 6.21 (Sphere rolling on a plane).
69
Chapter 7
Tensors and Differential forms
Duality in mathematics is not a theorem, but a “principle”.

Duality in Mathematics and Physics, 2007.
Sir M.F. Atiyah (1929–2019)
7.1 Cotangent space

Covectors are dual object to vectors, i.e., linear functionals defined on the tangent space. The space
of all covectors at a point q ∈ M , is called cotangent space.
Definition 7.1. Let M be a n-dimensional smooth manifold. The cotangent space at a point
q ∈ M is the set
Tq∗ M := (Tq M )∗ = {λ : Tq M → R, λ linear}.
If λ ∈ Tq∗ M and v ∈ Tq M , we will denote by hλ, vi := λ(v) the evaluation of the covector λ on the
vector v.
Covectors are not just abstract object but corresponds to differential of scalar functions.
Example 7.2. Let f : M → R be a smooth function and q ∈ M . The differential f∗,q : Tq M → R

of the scalar function f at q will be denoted dq f or df q , it satisfies
d
hdq f, vi := f (γ(t)), v ∈ Tq M, (7.1)
dt t=0
where γ is any smooth curve such that γ(0) = q and γ̇(0) = v, is an element of Tq∗ M .
Recall that if (U, ϕ) is a chart and (x1 , . . . , xn ) is the associated coordinate system if
n
X ∂
v= vi
∂xi q
i=1
then (7.1) satisfies

n n
d X ∂ X ∂ fb
hdq f, vi := f (γ(t)) = vi f= vi (x) (7.2)
dt t=0 ∂xi q ∂xi
i=1 i=1
71
where fb = f ◦ ϕ−1 and x = ϕ(q). If we choose as function f the coordinate function xi and as a
∂
vector v the vector ∂xj we have
q
* +
∂ ∂
dq xi , = xi = δij
∂xj q ∂xj q
which proves that dq x1 , . . . , dq xn is a basis of Tq∗ M that is dual to the coordinate basis of Tq M .
Exercise 7.3 (Change of coordinates for covectors). Given two coordinate sets (U, {xi }) and
(U 0 , {x0i }) with q ∈ U ∩ U 0 show that if λ ∈ Tq∗ M writes as
n
X n
X
λ= pi dxi q
= p0j dx0j q .
i=1 j=1
then
n
X ∂x0j
pi = p0j .
∂xi
j=1
∂x0j
(Here, if ϕ, ϕ0 denotes the coordinate maps, the quantity ∂xi is the jacobian of ϕ0 ◦ ϕ−1 ).
The differential of a smooth map yields a linear map between tangent spaces. The dual of the
differential gives a linear map between cotangent spaces.
Definition 7.4. Let F : M → N be a smooth map and q ∈ M . The pullback of F at point F (q),
where q ∈ M , is the map
F ∗ : TF∗ (q) N → Tq∗ M, 7 F ∗ λ,
λ→
defined by duality in the following way
hF ∗ λ, vi := hλ, F∗ vi , ∀ v ∈ Tq M, ∀ λ ∈ TF∗ (q) N.
Notice that we can also write F ∗ λ = λ ◦ F∗ .
The cotangent bundle

The cotangent bundle is defined as the disjoint union
[
T ∗M = Tq∗ M
q∈M
endowed with the natural projection
π : T ∗ M → M, π(λ) = q, if λ ∈ Tq∗ M
Proposition 7.5. T ∗ M has the natural structure of smooth vector bundle of rank n = dim M .
72
Proof. The proof is similar to the argument for T M , Given charts {(Ui , ϕi )} on M , then π −1 (Ui ) is
an open set of T ∗ M and the reunion of these sets is an open cover of T ∗ M . We define coordinates
ϕi : π −1 (Ui ) → ϕi (Ui ) × Rn
as follows
ϕi (λ) = (x1 , . . . , xn , p1 , . . . , pn )
if π(λ) = q with ϕi (q) = (x1 , . . . , xn ), and moreover
n
X
λ= pi dxi q
i=1
Notice that if Ui ∩ Uj 6= ∅ then π −1 (Ui ) ∩ π −1 (Uj ) 6= ∅ and

ϕi ◦ ϕ−1 n
j : ϕj (Ui ∩ Uj ) × R → ϕi (Ui ∩ Uj ) × R
n
where ϕi ◦ ϕ−1 0 0
j (x, p) = (x , p ) if (cf. with Exercise 7.31)
∂(ϕi ◦ ϕ−1
j )
x0 = ϕi ◦ ϕ−1
j (x), p= p0
∂x
which completes the proof.
Corollary 7.6. Let F : M → N be smooth. Then F ∗ : T ∗ N → T ∗ M is smooth.
Differential 1-forms
Definition 7.7. A differential 1-form on a smooth manifold M is a smooth section of T ∗ M i.e., a
map
ω : q 7→ ω(q) ∈ Tq∗ M,
that associates with every point q in M a cotangent vector at q. We denote by Λ1 (M ) the set of
differential forms on M .
In coordinates a differential form is written as
Xn
ω= ωi dxi
i=1
where ωi are smooth functions. Since differential 1-forms are dual objects to vector fields, it is well
defined the action of ω ∈ Λ1 M on X ∈ Vec(M ) pointwise, defining a function on M .
hω, Xi : q 7→ hω(q), X(q)i . (7.3)
Exercise 7.8. Prove that a differential form ω is smooth if and only if, for every smooth vector
field X ∈ Vec(M ), the function hω, Xi ∈ C ∞ (M ).
Definition 7.9. Let F : M → N be a smooth map and g : N → R be a smooth function. The
pullback F ∗ g is the smooth function on M defined by
F ∗ g = g ◦ F, q ∈ M.
If ω is a 1-form on N . The pullback F ∗ ω is the 1-form on M defined by
F ∗ ω = ω ◦ F∗
73
The differential of a function
The differential of a function automatically defined a 1-form. We want to find the expression in
coordinates, i.e. the functions ωi such that
n
X
df = ωi dxi
i=1
By definition of dual basis

∂ ∂f
ωj = df, =
∂xj ∂xj
where the last element of the identity is written in coordinates. Then
n
X ∂f
df = dxi
∂xi
i=1
Remark 7.10. We might think to our background in calculus and to the gradient
n
X ∂f ∂
∇f = (7.4)
∂xi ∂xi
i=1
but this quantity indeed is not well defined. Show as an exercise (7.4) is not invariant by change
of coordinates
Some properties of differential and pullback
Proposition 7.11. Let F : M → N be smooth and g smooth on M then
F ∗ dg = d(g ◦ F ) = dF ∗ g (7.5)
∗ ∗
F (gω) = (g ◦ F )F ω (7.6)
Remark 7.12. Let F : M → N be smooth and ω ∈ Λ1 (N ). Then if in coordinates

m
X
ω= ωj dyj
j=1
we have using (7.5)

m
X m
X
∗ ∗
F ω= (ωj ◦ F )F dyj = (ωj ◦ F )dFj
j=1 j=1
where Fj := yj ◦ F denotes the j-th component of F in coordinates.
Exercise 7.13. Let F : R3 → R2 be the function F (x, y, z) = (x2 y, y sin z) and ω = udv + vdu in
Λ1 (R2 ). Compute F ∗ ω.
Remark 7.14. Let T : V → W be a linear map. The dual map is T ∗ : W ∗ → V ∗ and its inverse
(T ∗ )−1 : V ∗ → W ∗ . Fix basis {e1 , . . . , en } for V and {f1 , . . . , fm } for W . If T is represented by
the matrix A with respect to these basis then (T ∗ )−1 is represented by (AT )−1 with respect to the
dual basis.
74
7.2 Tensors and tensor fields
Definition 7.15. A (covariant) k-tensor on a vector space V is a multilinear map
T : |V × ·{z
· · × V} → R.
k times
The space of all such tensors is denoted T k (V ).
Exercise 7.16. Basic examples are: linear maps (1-tensors), bilinear maps (2-tensors), determinant
on Rn (n-tensor).
If we have η1 , η2 ∈ V ∗ which means η1 , η2 : V → R linear maps (or 1-tensors) we can produce a

2-tensor η1 ⊗ η2 as follows
η1 ⊗ η2 : V × V → R, η1 ⊗ η2 (v, w) := η1 (v)η2 (w)
As a preliminary observation note that the operation is not commutative in general since η1 ⊗ η2 6=
η2 ⊗ η1 , but it is associative since (η1 ⊗ η2 ) ⊗ η3 = η1 ⊗ (η2 ⊗ η3 ).
Hence one can define multiple products η1 ⊗ . . . ⊗ ηk . It is well defined the tensor product of a
finite number k of elements in V ∗ defining an element of T k (V ).
Let V be a vector space and e1 , . . . , en be a basis and e∗1 , . . . , e∗n the dual basis. Every bilinear
form B : V × V → R can be written in an unique way as follows
n
X
B= Bij e∗i ⊗ e∗j
i,j=1
where Bij = B(ei , ej ) (check the details as an exercise!). Hence T 2 (V ) has dimension n2 . More in
general we have the following.
Lemma 7.17. Let V be a vector space and e1 , . . . , en be a basis and e∗1 , . . . , e∗n the dual basis. Then
the set
{e∗i1 ⊗ · · · ⊗ e∗ik | 1 ≤ i1 , . . . , ik ≤ n}
is a basis for the space of k tensors T k (V ). In particular dim T k (V ) = nk .
Notice that if T ∈ T k (V ) then we can then write

X
T = Ti1 ...ik e∗i1 ⊗ · · · ⊗ e∗ik (7.7)
1≤i1 ,...,ik ≤n
where Ti1 ...ik = T (ei1 , · · · , eik ) are the components of the tensor. Through its coordinates a tensor
is represented by a sort of generalized matrix with k indices.
Exercise 7.18. Prove that if T is written as in (7.8) with respect to a basis e1 , . . . , en and
X
T = Tj01 ...jk fj∗1 ⊗ · · · ⊗ fj∗k (7.8)
1≤i1 ,...,ik ≤n
75
where f1 , . . . , fn is a basis satisfying ei = aij fj , then we have
X
Tj01 ...jk = Ti1 ...ik ai1 j1 · · · aik jk
1≤i1 ,...,ik ≤n
where we denoted by aij the elements of the inverse of the matrix with elements aij , namely
jl
P
j aij a = δil .
Generalizing the above procedure, if T1 and T2 are tensor of rank k, l on V we can define the
tensor product T1 ⊗ T2 a tensor of rank k + l on V .
T1 ⊗ T2 (v1 , . . . , vk , w1 , . . . , wl ) = T1 (v1 , . . . , vk )T2 (w1 , . . . , wl )
As before, the tensor product just defined is associative but not commutative. One can consider
the associative graded (not commutative) algebra
M
T k (V ).
k≥0
Tensor product of vector spaces

Following the previous construction we have seen that the space T k (V ) is generated by elements of
the form e∗i1 ⊗ · · · ⊗ e∗ik so that it is suggestive to write
∗
T k (V ) = V · · ⊗ V }∗
| ⊗ ·{z (7.9)
k times
but what is V ∗ ⊗· · ·⊗V ∗ ?

If we accept identity (7.9) for a moment, using the canonical isomorphisms
between the vector space V and its bidual V ∗∗ given by
β : V → V ∗∗ , β(v)[η] := η(v), η ∈V∗
one can similarly build the set of contravariant l-tensors
Tl (V ) := T l (V ∗ ) = V ∗∗ ⊗ · · · ⊗ V ∗∗ ' V ⊗ · · · ⊗ V.
Formally T ∈ Tl (V ) is a multilinear map
∗
· · × V }∗ → R.
| × ·{z
T :V
l times
Pushing the idea, we can define also mixed tensors of type (k, l) as follows
∗
Tlk (V ) := V . . ⊗ V }∗ ⊗ V
| ⊗ .{z | ⊗ ·{z
· · ⊗ V}
k times l times
as the set of multilinear maps

∗
| × ·{z
T :V · · × V} × V · · × V }∗ → R.
| × ·{z
k times l times
If T ∈ Tlk (V ) then we can write along a basis
Tij11...i
...jl ∗
X X
T = k
ei1 ⊗ · · · ⊗ e∗ik ⊗ ej1 ⊗ · · · ⊗ ejl
1≤i1 ,...,ik ≤n 1≤j1 ,...,jl ≤n
where Tij11...i
...jl
k
are the components of the tensor.
76
Remark 7.19. Here there is a good reason to start also to use the “physicists convention” about
the position of indexes
Tij11...i
...jl i1
X X
T = k
e ⊗ · · · ⊗ eik ⊗ ej1 ⊗ · · · ⊗ ejl
1≤i1 ,...,ik ≤n 1≤j1 ,...,jl ≤n
by using upper indices on vectors for dual elements and writing the dual basis as {e1 , . . . , en } instead
of {e∗1 , . . . , e∗n } (notice the corresponding position for components). So, with this convention, a
vector and a covector should be written in components as follows
n
X n
X
i
v= v ei , η= ηi ei
i=1 i=1
or sometimes, even more shortly using Einstein summation, v = v i ei , η = ηi ei .

Finally, we can also define the space of all tensors on V .
M
T (V ) := Tlk (V ).
k,l≥0
Notice that (T (V ), ⊗) becomes a non-commutative associative algebra (only now the product of
elements of the space belong to the space). It is a graded algebra in the sense that the product is
0 k+k0
compatible with the grading as follows. If T ∈ Tlk (V ) and S ∈ Tlk0 (V ) then T ⊗ S ∈ Tl+l 0 (V ).
On tensor product of spaces

Similarly to the previous construction, one can define also tensor products of different vector spaces
V ⊗ W , bilinear maps V ∗ × W ∗ → R, or also V ∗ ⊗ W , bilinear maps V × W ∗ → R, etc.
Remark 7.20 (An abstract way to define V ⊗ W ). Consider the free vector space R hV × W i over
V × W , i.e. the vector space where every element (v, w) ∈ V × W is an element of a basis. Then
we introduce the equivalence relation (where k is an arbitrary scalar)
k(v, w) ∼ (kv, w) ∼ (v, kw),

(v1 + v2 , w) ∼ (v1 , w) + (v2 , w),
(v, w1 + w2 ) ∼ (v, w1 ) + (v, w2 ).
Then we set V ⊗ W = R hV × W i / ∼ and we denote by v ⊗ w the equivalence class [(v, w)]∼ . We

have automatically satisfied
k(v ⊗ w) = kv ⊗ w = v ⊗ kw,
(v1 + v2 ) ⊗ w = v1 ⊗ w + v2 ⊗ w
v ⊗ (w1 + w2 ) = v ⊗ w1 + v ⊗ w2 .
By definition every element of V ⊗ W is a linear combination of elements of the form v ⊗ w, but it

is not true in general that every element of V ⊗ W is written as v ⊗ w for some v, w.
Exercise 7.21. Let e1 , . . . , e4 be a basis of R4 . Prove that e1 ⊗ e2 + e3 ⊗ e4 cannot be written as

v ⊗ w for some vectors v, w.
77
Exercise 7.22. Prove that there is a unique isomorphisms between (V ⊗ W ) ⊗ Z and V ⊗ (W ⊗ Z)
sending (v ⊗ w) ⊗ z and v ⊗ (w ⊗ z). In particular we can write V ⊗ W ⊗ Z.
As a final example let us prove.
Lemma 7.23. Let V, W be finite dim vector spaces. Then V ∗ ⊗ W is isomorphic to Hom(V, W )
Proof. Let us build a linear isomorphisms L : V ∗ ⊗ W → Hom(V, W ). Of course it is enough to

define it on elements of the form η ⊗ w for η ∈ V ∗ and w ∈ W .
We set L(η ⊗ w) ∈ Hom(V, W ) as follows L(η ⊗ w)[v] = η(v)w. Then it is very easy to show
that this map is injective. Indeed if L(η ⊗ w) is the zero map then we want to show that η ⊗ w = 0.
This means that either w = 0 or η = 0. But if L(η ⊗ w) is the zero map then η(v)w = 0 for every
v. If w 6= 0 then this implies η(v) = 0 for all v and the statement is proved.
Remark 7.24. More in general an element of

∗
Tlk (V ) := V . . ⊗ V }∗ ⊗ V
| ⊗ .{z | ⊗ ·{z
· · ⊗ V}
k times l times
can also be thought as a linear map V ⊗k → V ⊗l where V ⊗j := V

| ⊗ ·{z
· · ⊗ V}.
j times
Symmetric and Alternating tensors

We focus again on tensors of type (k, 0). Given a permutation σ ∈ Sk we set
T σ (X1 , . . . , Xk ) = T (Xσ(1) , . . . , Xσ(k) )
Definition 7.25. Given a tensor T of type (k, 0) we say that
• T is symmetric if for every σ we have T = T σ
• T is alternating if for every σ we have T = sgn(σ)T σ
Given T we can define its symetrization and skew-symetrization as

1 X σ 1 X
Sym(T ) = T , Alt(T ) = sgn(σ)T σ
k! k!
σ∈Sk σ∈Sk
Exercise 7.26. Prove that given a tensor T of type (k, 0)
• Sym(T ) is symmetric, Alt(T ) is skew-symmetric,
• T is symmetric if and only if T = Sym(T )
• T is skew-symmetric if and only if T = Alt(T ).
It is meaningful to speak about symmetric and skew-symmetric tensor and tensor fields. More
about alternating tensors.
78
Lemma 7.27. T is alternating if and only if
(a) T (X1 , . . . , Xk ) = 0 whenever X1 , . . . , Xk are linearly dependent.
(b) T (X1 , . . . , Xk ) = 0 whenever two elements are equal
Given T and S we define the skew-symmetric product

(k + l)!
T ∧S = Alt(T ⊗ S)
k!l!
For instance we have, given a basis e1 , . . . , en of V
ei ∧ ej = ei ⊗ ej − ej ⊗ ei
We denote by Λk (V ) the set of alternating tensors of type (k, 0) and

M
Λ(V ) = Λk (V )
k≥0
Lemma 7.28. We have the following properties in Λ(V )
(i) ∧ is R-bilinear
(ii) T ∧ S = (−1)rs S ∧ T if T, S of order r, s
(iii) (T ∧ S) ∧ R = T ∧ (S ∧ R)
Notice that (ii) follows from
λ1 ∧ . . . ∧ λr ∧ η = (−1)r η ∧ λ1 ∧ . . . ∧ λr
Notice also that to prove (iii) one has to prove that
Alt (Alt(T ⊗ S) ⊗ R) = Alt (T ⊗ Alt(S ⊗ R))
so there is something to check. If that is true then we have well-defined

(k + l + r)!
T ∧S∧R= Alt(T ⊗ S ⊗ R)
k!l!r!
Lemma 7.29. Given covectors λ1 , . . . , λk and vectors v1 , . . . , vk then
λ1 ∧ · · · ∧ λk (v1 , · · · , vk ) = det(hλi , vj i)
The proof is (we use (iii))

k! X
λ1 ∧ · · · ∧ λk (v1 , · · · , vk ) = sgn(σ)λσ(1) ⊗ · · · ⊗ λσ(k) (v1 , · · · , vk )
k!
σ∈Sk
X
= sgn(σ) λσ(1) , v1 · · · λσ(k) , vk
σ∈Sk
= det(hλi , vj i)
79
n
Lemma 7.30. Λk (V ) is a vector subspace of T k (V ). We have dim Λk (V ) =

k and a basis is
given by
{e∗i1 ∧ · · · ∧ e∗ik | 1 ≤ i1 < . . . < ik ≤ n}. (7.10)
Proof. If two elements are repeated in e∗i1 ∧ · · · ∧ e∗ik then this is zero since alternating. If we
exchange position, we have a minus sign by (ii). Hence the only elements which might be lin.ind,
up to reordering, are those listed in (7.10). The proof that they are linearly independent follows
by noticing that
e∗i1 ∧ · · · ∧ e∗ik (ej1 , · · · , ejk ) = δIJ
where δJI is the Kronecker delta for multindex I = (ii , . . . , ik ) and J = (j1 , . . . , jk ) which is equal
to the sgn(σ) if J = σ(I) for some permutation σ, and zero otherwise.
Tensor fields
To give a concrete content to this theory let us consider tensor fields. We define the tensor bundle
as [
Tlk (M ) := Tlk (Tq M )
q∈M
We have clear identifications

T00 M = M × R
T01 M = T ∗ M, T10 M = T M
These are vector bundles. We can consider smooth tensor fields, i.e., smooth sections of these vector
bundles which are denoted by Tlk (M ) or also Γ(Tlk (M )). If
σ : M → Tlk (M )
is a smooth tensor field we can write locally in coordinates on an open set U
∂ ∂
σij11...i
...jl
X X
σ= dxi1 ⊗ · · · ⊗ dxik ⊗ ⊗ ··· ⊗
k ∂xj1 ∂xjl
1≤i1 ,...,ik ≤n 1≤j1 ,...,jl ≤n
We recover smooth functions for (k, l) = (0, 0), differential 1-forms for (k, l) = (1, 0) and vector
fields for (k, l) = (0, 1).
Exercise 7.31 (Change of coordinates for tensor fields). Given two coordinate sets (U, {xi }) and
(U 0 , {x0i }) with q ∈ U ∩ U 0 show that if
∂ ∂
σij11...i
...jl
X X
σ= dxi1 ⊗ · · · ⊗ dxik ⊗ ⊗ ··· ⊗
k ∂xj1 ∂xjl
1≤i1 ,...,ik ≤n 1≤j1 ,...,jl ≤n
and
X X ∂ ∂
σ= (σ 0 )h`11...`
...hl
dx0`1 ⊗ · · · ⊗ dx0`k ⊗ 0 ⊗ ··· ⊗ 0
k ∂xh1 ∂xhl
1≤`1 ,...,`k ≤n 1≤h1 ,...,hl ≤n
Write the change of coordinates between σij11...i

...jl
k
and (σ 0 )h`11...`
...hl
k
.
80
From now on we focus on tensors of the form (k, 0), with special emphasis on alternating ones.
The following lemma should be now well understood, but we invite the reader to check details
Lemma 7.32. Let M be smooth manifold and σ : M → T k (M ) a smooth tensor field. Then σ is
smooth if and only if one of the following equivalence conditions is satisfied
• in every coordinate charts the coefficients σi1 ...ik are smooth functions,
• for every smooth vector fields X1 , . . . , Xk we have that
σ(X1 , . . . , Xk )(q) = σq (X1 |q , . . . , Xk |q )
is a smooth function.
Given two tensor fields σ and τ and f ∈ C ∞ (M ) then f σ and σ ⊗ τ are also tensor fields.
Exercise 7.33. Write the coordinate representation of f σ and σ ⊗ τ in terms of the ones of σ, τ .
If F : M → N is a smooth map we can pull-back tensors and tensor fields of type (k, 0)
F ∗ : T k (TF (q) N ) → T k (Tq M )
by applying duality to every element for tensors
(F ∗ T )(v1 , . . . , vk ) = T (F∗ v1 , . . . , F∗ vk )
and tensor fields
(F ∗ σ)(X1 , . . . , Xk )|q = σ(F∗ X1 , . . . , F∗ Xk )|F (q)
Some properties which we leave as exercises.
Proposition 7.34. We have the following properties for F : M → N a smooth map
(i) F ∗ is R-linear on sections
(ii) F ∗ (σ ⊗ τ ) = F ∗ σ ⊗ F ∗ τ
(iii) F ∗ (gσ) = (g ◦ F )F ∗ σ
Notice that in general there is neither push-forward nor pull-back of mixed tensor fields through
smooth maps (but if F is a diffeomorphism one can define both).
7.3 Differential forms

Consider the set of alternating tensors of type (k, 0) on Tq M and build the vector bundle
[
Λk (M ) := Λk (Tq M )
q∈M
We set Ωk (M ) the smooth sections of Λk (M ), called differential k-forms. In coordinates ω ∈ Ωk (M )

writes as
Xn
ω= ωi1 ...ik dxi1 ∧ . . . ∧ dxik
i1 ,...,ik =1
where ωi1 ...ik are smooth functions on M .
81
Proposition 7.35. Suppose F : M → N is smooth. We have the following properties
(i) F ∗ : Ωk (N ) → Ωk (M ) is R-linear
(ii) F ∗ (ω ∧ η) = F ∗ ω ∧ F ∗ η
(iii) F ∗ (gω) = (g ◦ F )F ∗ ω
Remark 7.36. In particular we have a formula to compute the pull-back: if on N with coordinates
{yj } we have
Xn
ω= ωj1 ...jk dyj1 ∧ . . . ∧ dyjk
j1 ,...,jk =1
where ωj1 ...jk are smooth functions on N . Then we get
n
X
F ∗ω = (ωj1 ...jk ◦ F )d(yj1 ◦ F ) ∧ . . . ∧ d(yjk ◦ F ),
j1 ,...,jk =1
which means, denoting Fj the j-th coordinate of F

X
F ∗ω = (ωj1 ...jk ◦ F )dFj1 ∧ . . . ∧ dFjk ,
j1 ...jk
Exercise 7.37. Consider the map

F :]0, +∞[×]0, 2π[→ R2 , F (r, θ) = (r cos θ, r sin θ)
Prove that F ∗ (dx ∧ dy) = r dr ∧ dθ
Notice that Ωn (M ) on an n-dimensional manifold has dimension 1, hence every top dimensional
form is a smooth multiple of dx1 ∧ . . . ∧ dxn in coordinates.
Proposition 7.38. Let F : M → N is smooth map between n-dimensional manifolds.
F ∗ (g dy1 ∧ . . . ∧ dyn ) = (g ◦ F )(det DF )dx1 ∧ . . . ∧ dxn
where DF is the Jacobian matrix of F in the corresponding coordinates.
Proof. By the previous result
F ∗ (g dy1 ∧ . . . ∧ dyn ) = (g ◦ F )dF1 ∧ . . . ∧ dFn
so it is enough to prove
dF1 ∧ . . . ∧ dFn = (det DF )dx1 ∧ . . . ∧ dxn
It is enough to check the last identity on a basis. But from Lemma 7.29

∂ ∂ ∂
dF1 ∧ . . . ∧ dFn ,..., = det dFi = det DF.
∂x1 ∂xn ∂xj
so that dF1 ∧ . . . ∧ dFn = (det DF )dx1 ∧ . . . ∧ dxn .
Exercise 7.39. Let {xi } and {x0j } be two set of coordinates. Prove that
0
∂xj
dx01 ∧ . . . ∧ dx0n = det dx1 ∧ . . . ∧ dxn .
∂xi
82
Exterior derivative. We introduce the exterior derivative d. This is a linear map d : Ωk (M ) →
Ωk+1 (M ) for every k ≥ 0. Recall that Ω0 (M ) = C ∞ (M ).
Theorem 7.40. There exists a unique linear map d : Ωk (M ) → Ωk+1 (M ) for every k ≥ 0 such
that
(i) if f ∈ Ω0 (M ) = C ∞ (M ), then df ∈ Ω1 (M ) is its differential,
(ii) if ω ∈ Ωk (M ) and η ∈ Ωl (M ),
d(ω ∧ η) = dω ∧ η + (−1)k ω ∧ dη,
(iii) d ◦ d = 0,
Moreover
(iv) d is local: if ω = ω 0 on U ⊂ M then dω = dω 0 on U ,
(v) d commutes with restrictions dω|U = d(ω|U ),
(vi) d in coordinates express as

X X ∂ωi ...i
1 k
dω = dωi1 ...ik ∧ dxi1 ∧ . . . ∧ dxik = dxj ∧ dxi1 ∧ . . . ∧ dxik . (7.11)
∂xj
i1 ...ik i1 ...ik ,j
We denote by dxI = dxi1 ∧ . . . ∧ dxik if I = (i1 , . . . , ik ) increasing multiindex.

Proof. Assume M has only one chart. Then use the formula (7.11) to define d. Clearly d is linear,
local and commutes with restrictions and satiesfies (i). We have to check only (ii) and (iii).
First notice that d(f dxI ) = df ∧ dxI for every multiindex I, even if I is not increasing. Then
to check (ii) by linearity it is enough to check for ω = f dxI and η = gdxJ for I, J increasing
multiindex. We have
d(ω ∧ η) = d(f gdxI ∧ dxJ )

= (f dg + gdf ) ∧ dxI ∧ dxJ
= (−1)k f dxI ∧ dg ∧ dxJ + df ∧ dxI ∧ gdxJ
= (−1)k ω ∧ dη + dω ∧ η.
For (iii) since

X
d(dω) = d(dωi1 ...ik ) ∧ dxi1 ∧ . . . ∧ dxik ,
i1 ...ik
it is enough to prove that d(df ) = 0. We have

n n
!
X ∂f X ∂2f
d(df ) = d dxi = dxj ∧ dxi
∂xi ∂xi xj
i=1 i,j=1
X ∂2f
= dxj ∧ dxi = 0.
∂xi xj
1≤i<j≤n
83
Let us now prove that if there exists an operator D satisfying (i)-(iii) then in coordinates it satisfies
(vi). From this everything follows. (We leave other details to the reader.)
Let D be such an operator and apply to a k-form in coordinates, from property (ii)
X
Dω = Dωi1 ...ik ∧ dxi1 ∧ . . . ∧ dxik +
i1 ...ik
X
+ ωi1 ...ik ∧ D(dxi1 ∧ . . . ∧ dxik )
i1 ...ik
The first line is (vi) since D = d on functions. The second line is zero since by (ii)
D(dxi1 ∧ . . . ∧ dxik ) = Ddxi1 ∧ (dxi2 . . . ∧ dxik ) − dxi1 ∧ D(dxi2 . . . ∧ dxik )
and Ddxi1 = DDxi1 = 0 by (i), and the other terms is also zero iterating same argument.
Exercise 7.41. The differential of an arbitrary 1-form in R3 . If a, b, c are smooth functions and
ω = a dx + b dy + c dz
then we have
dω = da ∧ dx + db ∧ dy + dc ∧ dz
that is
∂c ∂a ∂b ∂a ∂c ∂b
dω = − dx ∧ dz + − dx ∧ dy + − dy ∧ dz
∂x ∂z ∂x ∂y ∂y ∂z
Remark 7.42 (An algebraic comment). If A = ⊕k Ak is a graded associative algebra (not necessarily
commutative), a linear map D : A → A has degree m if D(Ak ) ⊂ Ak+m for each k. It is a derivation
(resp. antiderivation) if
D(xy) = (Dx)y + x(Dy), (resp. D(xy) = (Dx)y + (−1)k x(Dy) )
whenever x ∈ Ak and y ∈ Al .
We can summarize the above properties of the differential d on C ∞ (M ) saying that it extends
to a unique antiderivation on Ω(M ) of degree 1 and such that d squares to zero.
Proposition 7.43. Let F : M → N be smooth and ω ∈ Ωk (N ). Then
F ∗ (dω) = d(F ∗ ω)
Proof. For k = 0 setting F ∗ f = f ◦ F this is item (i) in Proposition 7.11. For k ≥ 1 we reduce to
it. Since d is local it is sufficient to prove the formula in local coordinates. Since in coordinates
X
ω= ωi1...ik dxi1 ∧ · · · ∧ dxik
it is sufficient to prove for ω of the form f dxi1 ∧ · · · ∧ dxik . But F ∗ (ω ∧ η) = F ∗ ω ∧ F ∗ η so it is

enough to prove for ω = f dg. We have since d2 = 0
dω = df ∧ dg + f d2 g = df ∧ dg
84
and
F ∗ (dω) = F ∗ df ∧ F ∗ dg = dF ∗ f ∧ F ∗ dg = d(f ◦ F ) ∧ F ∗ dg
On the other hand
d(F ∗ ω) = d((f ◦ F )F ∗ dg) = d(f ◦ F ) ∧ F ∗ dg + (f ◦ F )dF ∗ dg = d(f ◦ F ) ∧ F ∗ dg
where we used Proposition 7.11 and dF ∗ dg = ddF ∗ g = 0 since d2 = 0.
Exercise 7.44. Consider the map
F :]0, +∞[×]0, 2π[→ R2 , F (r, θ) = (r cos θ, r sin θ)
We proved that F ∗ (dx ∧ dy) = r dr ∧ dθ. On the other hand

1 1 2
dx ∧ dy = d xdy − ydx , r dr ∧ dθ = d r dθ
2 2
and F ∗ (xdy − ydx) = r2 dθ.
7.4 Lie derivatives of tensors and differential forms

Definition 7.45. Let X ∈ Vec(M ) and ω ∈ Λk M , where k ≥ 0. We define the Lie derivative of τ
covariant tensor with respect to X as the operator
d
LX : T k (M ) → T k (M ), LX τ = (etX )∗ τ. (7.12)
dt t=0
Notice that
d 1 tX ∗
LX τ = (etX )∗ τ = lim (e ) (τetX (q) ) − τq
dt t=0 t→0 t
We define the Lie derivative of ω ∈ Ωk (M ) with respect to X as the operator
d
LX : Λk M → Λk M, LX ω = (etX )∗ ω. (7.13)
dt t=0
We stress that the Lie derivative of a k-form along a vector field defines a new k-form.
For k = 0 this definition coincides with the Lie derivative of smooth functions, LX f = Xf , for
f ∈ C ∞ (M ). From the previous properties, one easily deduces the following properties of the Lie
derivative:
(i) LX (ω1 ∧ ω2 ) = (LX ω1 ) ∧ ω2 + ω1 ∧ (LX ω2 ),
(ii) LX (f ω) = (Xf )ω + f LX ω
Property (i) can be also expressed by saying that LX is a derivation of the exterior algebra of
k-forms. Since d commutes with F ∗ and is linear we immediately have also
Lemma 7.46. LX ◦ d = d ◦ LX .
85
Given a k-form and a vector field, one can also introduce their inner product, defining a (k − 1)-
form as follows.
Definition 7.47. Let X ∈ Vec(M ) and ω ∈ Λk M , with k ≥ 1. We define the inner product of ω
and X as the operator iX : Λk M → Λk−1 M , such that
(iX ω)(Y1 , . . . , Yk−1 ) := ω(X, Y1 , . . . , Yk−1 ), Yi ∈ Vec(M ). (7.14)
The operator iX is an anti-derivation, in the following sense:
iX (ω1 ∧ ω2 ) = (iX ω1 ) ∧ ω2 + (−1)k1 ω1 ∧ (iX ω2 ), ωi ∈ Λki M, i = 1, 2. (7.15)
We end this section proving two classical formulas, usually referred as Cartan’s formulas.
Proposition 7.48 (Cartan’s formula). Let X ∈ Vec(M ). The following identity holds true
LX = iX ◦ d + d ◦ iX . (7.16)
Proof. Set DX := iX ◦ d + d ◦ iX . It is easy to check that DX is a derivation on the algebra of

k-forms, since iX and d are anti-derivations. Let us show that DX commutes with d. Indeed, using
the fact that d2 = 0, one gets
d ◦ DX = d ◦ iX ◦ d = DX ◦ d.
P
Since any k-form can be expressed in coordinates as ω = ωi1 ...ik dxi1 . . . dxik , it is sufficient to
prove that LX coincide with DX on functions. This last property is easily verified, since
DX f = iX (df ) + d(iX f ) = hdf, Xi = Xf = LX f.

| {z }
=0
Corollary 7.49. Let X, Y ∈ Vec(M ) and ω ∈ Λ1 M , then
dω(X, Y ) = X hω, Y i − Y hω, Xi − hω, [X, Y ]i . (7.17)
Proof. On one hand Definition 7.45 implies, by Leibniz rule1

d
hLX ω, Y iq = (etX )∗ ω, Y q
dt t=0
d
= ω, etX
∗ Y etX (q)
dt t=0
= X hω, Y i − hω, [X, Y ]i .
On the other hand, Cartan’s formula (7.16) gives
hLX ω, Y i = hiX (dω), Y i + hd(iX ω), Y i

= dω(X, Y ) + Y hω, Xi .
Comparing the two identities one gets (7.17).

1
Let f (x, y) be a function of two variables. Then
d ∂f ∂f d d
f (t, t) = (0, 0) + (0, 0) = f (t, 0) + f (0, t)
dt t=0 ∂x ∂y dt t=0 dt t=0
86
Exercise 7.50. Prove that for a k form ω we have
LY (ω(X1 , . . . , Xk )) = (LY ω)(X1 , . . . , Xk ) + ω(LY X1 , . . . , Xk )

+ . . . + ω(X1 , . . . , LY Xk )
which also means
(LY ω)(X1 , . . . , Xk ) = Y (ω(X1 , . . . , Xk )) − ω([Y, X1 ], . . . , Xk )

− . . . − ω(X1 , . . . , [Y, Xk ])
Exercise 7.51. Prove the following Leibniz rule formula: for X ∈ Vec(M ), ω ∈ Λk M , and
f ∈ C ∞ (M )
Lf X ω = f LX ω + df ∧ iX ω (7.18)
87
Chapter 8
Orientation, Integration on manifolds
Stokes’ theorem, first appeared in print in 1854.

George Stokes had for several years been setting the Smith’s Prize Exam at Cambridge,
in February, 1854, examination, question #8 is the following: [a version of Stokes’ theorem]
V. Katz, The History of Stokes’ Theorem
Mathematics Magazine, Vol. 52, No. 3 (May, 1979)
8.1 Orientation
Let V be a vector space. An orientation of V is an equivalence class of ordered basis. Two basis
e1 , . . . , en and e01 , . . . , e0n belong to the same equivalence class if the matrix of the change of basis
has positive determinant.
n
X
e0i = aij ej , det(aij ) > 0
j=1
Of course this is an equivalence relation and we have two equivalence classes.

We can think to a orientation also as a choice of an “orthogonal vector” to the vector space
“up” or “down”. It is better formalized as follows.
Lemma 8.1. Let V be a vector space of dim n ≥ 1. A non zero element Ω of Λn (V ) defines an
orientation for V : all ordered basis e1 , . . . , en such that Ω(e1 , . . . , en ) > 0.
The same can be done on manifolds, with top dimensional differential forms.
Definition 8.2. Two charts (U, ϕ) and (U 0 , ϕ0 ) are equioriented if det D(ϕ0 ◦ϕ−1 ) > 0 on ϕ(U ∩U 0 ).
An atlas is oriented if every pair of charts are equioriented. A manifold M is orientable if M admits
an oriented atlas. An orientation of M is a choice of an oriented atlas of M .
Lemma 8.3. If M is covered by two charts (U, ϕ) and (U 0 , ϕ0 ) with U ∩ U 0 connected, then M is
orientable.
Proof. Notice that by assumption det D(ϕ0 ◦ ϕ−1 ) is not zero since it is a local diffeo, hence its sign
is constant on every connected set, hence on U ∩ U 0 . If det D(ϕ0 ◦ ϕ−1 ) > 0 ok, otherwise choose
ϕ0 and ϕ = (x1 , . . . , xn , xn−1 ) which then are equioriented.
89
Corollary 8.4. S n is orientable for every n ≥ 1.
For n > 1 it is given by the previous Lemma 8.3. For n = 1 one can check it explicitly.
Exercise 8.5 (The Mobius band is not orientable). Let us consider the Mobius band M as the
infinite strip M := R × [0, 1]/ ∼ with the identitication (x, 0) ' (−x, 1), endowed with the quotient
topology. We denote by [x, y] = π(x, y) the image of a point under the canonical projection.
Consider on M the two open sets
U1 = π(R×]0, 1[), U2 = π(R × [0, 1/2[∪]1/2, 1])
Define the charts

(
(x, y), 0 ≤ y < 1/2
ϕ1 ([x, y]) = (x, y), ϕ2 ([x, y]) =
(−x, y − 1), 1 ≥ y > 1/2
It is easy to see that ϕ2 is well defined on M since ϕ2 ([x, 0]) = (x, 0) = ϕ2 ([−x, 1]). These are
homeomorphism onto their images and the change of charts satisfies
(
(x, y), 0 < y < 1/2
ϕ2 ◦ ϕ−1
1 (x, y) =
(−x, y − 1), 1 > y > 1/2
with det D(ϕ2 ◦ ϕ−1 −1

1 ) = 1 > 0 on 0 < y < 1/2 and det D(ϕ2 ◦ ϕ1 ) = −1 < 0 on 1 > y > 1/2.
This is a not oriented atlas. Indeed all atlas need to show this behaviour. To be more formal, let
(Ui , ϕi )i∈I be an oriented atlas on M . Define a “sign” function σ : [0, 1] → {−1, 1} as follows. For
y ∈ [0, 1] we consider some i = i(x) such that [0, y] ∈ Ui and set σ(y) = sgn det D(ϕ−1 i ◦ π)(0, y).
It is easy to check that σ is well-defined and locally constant, hence constant on [0, 1] which is
connected. On the other hand the identification [x, y] = [−x, y + 1] imposes σ(0) = −σ(1), which
gives a contradiction.
Exercise 8.6. Let F : M → N be a local diffeomorphism between oriented manifolds. Then F

preserves orientations if and only if for every pair of oriented charts (U, ϕ) and (V, ψ) such that
F (U ) ⊂ V we have det(DFb) > 0 on ϕ(U ), where as usual Fb = ψ ◦ F ◦ ϕ−1 .
Definition 8.7. A volume form on a n-dimensional smooth manifold M is a non vanishing n-form
ν ∈ Ωn (M ).
Notice that given a volume form every other n-form ω is written as ω = f ν with f ∈ C ∞ (M ).
If f 6= 0 then ω is also a volume form.
Lemma 8.8. For a volume form ν and a change of charts
∂x0j

∂ ∂ ∂ ∂
ν ,..., 0 = det ν ,...,
∂x01 ∂xn ∂xi ∂x1 ∂xn
90
Proof. In coordinates ν = f dx01 ∧ . . . ∧ dx0n for some f 6= 0. Hence it is enough to prove
0
∂xj
dx01 ∧ . . . ∧ dx0n = det dx1 ∧ . . . ∧ dxn (8.1)
∂xi
But this is a consequence of the formula
n
X ∂x0j
dx0j = dxi
∂xi
i=1
and Lemma 7.29.

Proposition 8.9. A smooth manifold M is orientable if and only if M admits a volume form
Proof. Fix a volume form ν ∈ Ωn (M ), never vanishing. We build the atlas A of all positive charts,
i.e., charts (U, ϕ) such that U connected and (notice that the sign is constant on a connected U )

∂ ∂
ν ,..., > 0.
∂x1 ∂xn
The atlas A is oriented since for any two charts (U, ϕ) and (U 0 , ϕ0 ) we have that property
0
∂xj

∂ ∂ ∂ ∂
ν , . . . , 0 = det ν ,...,
∂x01 ∂xn ∂xi ∂x1 ∂xn
hence every change of charts must have positive determinant since ν is positive on oriented frames..
Conversely given an atlas A of equioriented charts we take a partition of unity {ψα } subordinated
to it and we set X
ν= ψα dxα1 ∧ . . . ∧ dxαn
α
This is well defined on all M (the sum is finite at every point). We have to show that ν is never
vanishing. But this follows from the formula (8.1). Fix a point q and a chart (U, ϕ) containing that
point. In a neighborhood we read
α
X ∂xj
ν= ψα det dx1 ∧ . . . ∧ dxn
α
∂xi
Notice that all terms are ≥ 0 an at least one is > 0. Hence ν does not vanish.
Example 8.10. We know that S n is orientable. We can find a volume form on S n as follows.
Consider in Rn+1 the n-form
n
X
ω= (−1)i xi dx0 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxn
i=0
where the “hat” stands for removing the corresponding term. The restriction of this n form to S n
is still a n-form which is never vanishing since for every v1 , . . . , vn ∈ Tx S n we have
ωx (v1 , . . . , vn ) = det(x, v1 , . . . , vn ) 6= 0
interpreting x and the vi as elements of Rn+1 (check as exercice!). For instance in S 2 this gives the
volume form in coordinates (x, y, z) in R3
ω = xdy ∧ dz − ydx ∧ dz + zdx ∧ dy
91
Exercise 8.11. Let F : M → N be a smooth map between manifolds of the same dimension,
and let ω be a volume form on N . Prove that if F ∗ ω is a volume form on M then F is a local
diffeomorphism.
Recall that a manifold M is said paralelizable if T M is trivial.
Proposition 8.12. Every paralelizable manifold is orientable
Proof. Let X1 , . . . , Xn be n vector fields that are linearly independent everywhere. Consider the
dual basis η1 , . . . , ηn of differential 1-forms. Then Ω := η1 ∧ . . . ∧ ηn is a never vanishing volume
form.
Corollary 8.13. Every Lie group G is orientable.
Proof. Every Lie group is paralelizable since T G is diffeomorphic to G × g hence to G × Rn where

n = dim G. In other words one can choose v1 , . . . , vn in g and build n left-invariant vector fields
X1 , . . . , Xn where Xi (g) = Lg∗ vi that are automatically everywhere linearly independent. The dual
basis η1 , . . . , ηn satisfies L∗g ηi = ηi for every i = 1, . . . , n. Hence Ω = η1 ∧ . . . ∧ ηn satisfies L∗g Ω = Ω
for every g.
Proposition 8.14. The projective spaces Pn are orientable if and only if n is odd.
Proof. We prove that Pn cannot be orientable if n even. We consider Pn as the quotient of S n with
the group {1, i} where 1(x) = x and i(x) = −x. If we denote by p : S n → Pn the covering. Assume
we have a volume form ν for Pn then η = p∗ ν is a volume form for S n . Hence η = p∗ ν = f ω where
ω is the volume form of the Example 8.10 and f is a never vanishing function.
Notice that p ◦ i = p hence p∗ = i∗ p∗ : applied to ν this gives i∗ η = η. But we have also
i∗ ω = (−1)n+1 ω = −ω if n even. So
η = i∗ η = i∗ (f ω) = −(f ◦ i)ω
This shows that η changes sign, and being smooth it vanishes at some point. Contradiction.
To compete the proof one can either (a) prove by hand that the altlas is orientable if n odd (b)
prove that the form ω we have defined on the sphere descends to a volume form on the projective
space for even dimensional spaces (cf. the more general Exercise 8.15).
Exercise 8.15. Let M be manifold and G discrete subgroup acting properly and smoothly on M .
let p : M → M/G. If ω is a differential form on M such that g · ω = ω for every g ∈ G then there
exists on M/G a unique differential form η such that p∗ η = ω. If ω is a volume form then η is a
volume form as well.
8.2 Integration on manifolds

We start by integrating n forms on open sets of Rn .
Let U ⊂ Rn open and η = f (x)dx1 ∧ . . . ∧ dxn be a n-form with compact support in U . Then
we set Z Z Z
η= f (x)dx1 ∧ . . . ∧ dxn = f (x)dx
U U U
92
where dx denotes here the Lebesgue measure.
Let now M be a manifold of dimension n and assume (U, ϕ) is a chart with ω ∈ Ωn (M ) with
compact support in U . It would be natural to set (recall that ϕ : U ⊂ M → ϕ(U ) ⊂ Rn )
Z Z
ω= (ϕ−1 )∗ ω
U ϕ(U )
with the right hand side a well-defined integral of the n-form η = (ϕ−1 )∗ ω with compact support
in ϕ(U ). is this definition well posed? Let ω be a n-form with compact support in the intersection
U ∩ U 0 of two charts (U, ϕ) (U 0 , ϕ0 ). On one side we are setting
Z Z
ω= (ϕ−1 )∗ ω
U ϕ(U )
while on the other one Z Z

ω= (ϕ0−1 )∗ ω
U ϕ0 (U 0 )
We claim that we have equality between the two definitions if and only if the charts are equioriented.
Indeed let us write
(ϕ−1 )∗ ω = f (x)dx, (ϕ0−1 )∗ ω = g(y)dy
then writing F (x) = y (i.e., we are setting F := ϕ0 ◦ ϕ−1 ) we have by the formula (??)
f (x)dx = F ∗ (g(y)dy) = (g ◦ F ) det(DF )dx
by our previous considerations. On the other hand the change of variable formula says
Z Z
g(y)dy = (g ◦ F (x))| det(DF )|dx
U0 U
with the absolute value! It works if and only if the charts are equioriented. We have proved
Proposition 8.16. Let M be smooth manifold oriented and ω an n-dim form with compact support.
Let (U, ϕ) (U 0 , ϕ0 ) two equioriented charts such that the support of ω is contained in U ∩ U 0 . Then
Z Z
(ϕ0−1 )∗ ω = (ϕ−1 )∗ ω
ϕ0 (U 0 ) ϕ(U )
In particular the integral is independent on the chart and we can set on an oriented manifold
Z Z
ω := (ϕ−1 )∗ ω
U ϕ(U )
Proposition 8.17. Let M be smooth manifold oriented and ω an n-dim form with compact support.
Let {(Ui , ϕi )} be an oriented atlas and {ψi } partition of unity subordinated to it. Then the quantity
XZ
ψi ω
i∈N M
is well defined and independent on atlas and partition of unity.
93
Proof. Since supp(ω) is compact and the {ψi } partition of unity then the sum is finite at every
point and every form ψi ω has compact support in his Ui . If we take two partition of unity {ψi }
and {ψj0 } we can compare
XZ XZ
ψi ω, ψj0 ω
i∈N M j∈N M
since both sums can write as

XZ
ψi ψj0 ω
i,j∈N M
and each terms has support in Ui ∩ Uj0 hence independent by previous results.
Definition 8.18. The integral of ω ∈ Ωn (M ) on M is then defined as follows
Z XZ
ω := ψi ω
M i∈N M
where {ψi } is an arbitrary partition of unity.

Remark 8.19. It is easy to see that if M is oriented and if we denote by M the manifold M oriented
with the opposite orientation then for every volume form ω we have
Z Z
ω=− ω.
M M
Moreover if M is oriented with the orientation induced by the volume form ν, then
Z
ν > 0.
M
Remark 8.20. Given a volume form ν it makes sense to define

Z
fν
M
R
If ν is the orientation
R of M sometimes we write by abuse of notation M f . In particular if M is
compact volν (M ) = M ν.
Another immediate consequence of our construction we have the following version of the change
of variable formula.
Proposition 8.21. Let F : M → F (M ) ⊂ N be an orientation preserving diffeomorphism of M
onto its image F (M ). Then Z Z
ω= F ∗ω
F (M ) M
It is enough to consider differential forms with compact support in one chart and then one is
reduced to the case of differential forms in Rn where this is the change of variable formula for the
Lebesgue integral. Concretely this is what one does.
Proposition 8.22. Let M be a smooth oriented manifold. Suppose the support of ω is contained
in ∪i Ui with Ui compact and let Vi be compact domain in Rn with Fi : Vi → Ui parametrizations
such that
94
• Fi (Vi ) = Ui and Fi preserves the orientation (in the interior of the compact sets)
• Ui and Uj intersect only on the boundary
Then Z XZ
ω= Fi∗ ω
M i Vi
Exercise 8.23. Let us compute the integral over S 2 of
ω = xdy ∧ dz − ydx ∧ dz + zdx ∧ dy
This is the volume of the sphere and gives 4π. Let us consider the parametrization
F (ϕ, θ) = (sin ϕ cos θ, sin ϕ sin θ, cos ϕ)
defined on the (open) rectangle D = (0, π) × (0, 2π). We have that the chart F is positively
oriented with respect to ω (or that F preserves the orientation if we orient D ⊂ R2 with standard
orientation). It is enough to check at one point: for instance F (π/2, 0) = (1, 0, 0)
ω(1,0,0) = dy ∧ dz
and
d d
F∗ (∂ϕ |(π/2,0) ) = F (π/2 + t, 0) = (0, cos(t), − sin(t)) = (0, 0, −1) := v1
dt t=0 dt t=0
d d
F∗ (∂θ |(π/2,0) ) = F (π/2, t) = (cos(t), sin(t), 0) = (0, 1, 0) := v2
dt t=0 dt t=0
and
dy ∧ dz (v1 , v2 ) = 1.
We have also
F ∗ ω = sin ϕ dϕ ∧ dθ
and (removing a zero measure set1 )
Z Z Z π Z 2π
∗
ω= F ω= sin ϕ dθdϕ = 4π
S2 D 0 0
Notice that we can integrate n-forms on n-dim manifolds, but for every n!. Hence in an n-
dimensional manifold we can integrate k-forms on k-dimensional submanifolds. But we need to
speak about orientation of submanifolds.
Remark 8.24. If M is not orientable one can still integrate densities. These are not tensors but
they behave well (with absolute value) under change of charts. See [Lee13].
1
R
one can prove using a change of charts that zero-measure set are well defined on a manifold and that A
ω for
every zero measure set A in M and ω volume form.
95
The Hairy Ball theorem
We end with an important application of the integration theory, known as Hairy Ball theorem. The
proof we present here is due to J. Milnor.
Theorem 8.25 (Hairy Ball theorem). There exists no never vanishing smooth vector field on S n
when n is even.
Proof. Let us think to S n as subset of Rn+1 . Then a vector field on S n is a map X : Rn+1 → Rn+1
such that x · X(x) = 0 where the dot denotes the scalar product in Rn+1 . Assume there exists
a never vanishing vector field. Then we can consider X(x)/kX(x)k and assume without loss of
generality it has norm equal to 1 everywhere.
Let us consider the following family of maps for ε > 0
p
fε : S n (1) → S n ( 1 + ε2 ), fε (x) = x + εX(x)
We first need an auxiliary lemma.

Lemma 8.26. For ε > 0 small enough the map fε is a global diffeomorphism.
Proof of the Lemma. Let us consider the n-form in Rn+1 (which we recall it restricts to a volume
form on every sphere)
Xn
ω= (−1)i xi dx0 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxn
i=0
√
and let us consider fε∗ ω the pull-back on S n (1) of the restriction of ω to S n (r) with r = 1 + ε2 .
Using the formula
i
X
fε∗ ω = (−1)n (xi ◦ fε )d(x0 ◦ fε ) ∧ . . . ∧ d(x
\ i ◦ fε ) ∧ . . . ∧ d(xn ◦ fε )
i=0
it is not difficult to see that fε∗ ω is a polynomial with respect to ε of degree ≤ n + 1 and being f0
equal to the identity map we can indeed write
fε∗ ω = ω + εηε
where ηε is a family of smooth n forms which is polynomial with respect to ε fo degree ≤ n. In

particular, since we speak about volume forms, there exists a family of functions gε such that
ηε = gε ω and
fε∗ ω = (1 + εgε )ω
Since spheres are compact, this implies that for ε > 0 small enough fε∗ ω is a volume form on S n .
In particular fε is a local diffeomorphism for ε > 0 small enough thanks to Exercise 8.11.
Let us show that for ε > 0 small enough the map fε is also injective. If this is not the case
we would have a sequence εk → 0 and sequences of distincts points xk , yk ∈ S n (1) such that
fεk (xk ) = fεk (yk ) which can be rewritten as
xk − yk X(xk ) − X(yk )
= −εk
kxk − yk k kxk − yk k
96
This is a contradiction since the left hand side has norm 1 while the norm of the right hand side
tends to zero thanks to the inequality
kX(xk ) − X(yk )k ≤ Ckxk − yk k
which holds2 since X is a smooth vector field and S n (1) is compact.

We have proved that for ε > 0 small enough the map fε is an injective local diffeomorphism.
Since it is a local diffeomorphism, it is open. Since the source space is compact, it is a closed
map.3 Hence the image is open and closed and, by connectedness, it is surjective. A bijective local
diffeomorphism is by construction a global diffeomorphism.
To end the proof, let us now compute in two different ways the integral
Z
ω
S n (r)
One one hand using the change of variables y = F (x) = rx we have that F ∗ ω = rn+1 ω and
Z Z Z Z
ω= ω= F ∗ ω = rn+1 ω = cn rn+1
S n (r) F (S n (1)) S n (1) S n (1)
√
On the other hand using fε as a change of variables (recall r = 1 + ε2 )
Z Z Z
ω= ω= fε∗ ω = P (ε)
S n (r) fε (S n (1)) S n (1)
where P is some polynomial in ε. This implies for ε > 0 small enough

n+1
cn (1 + ε) 2 = P (ε)
which is a contradiction for n even.
Remark 8.27. Notice that on odd dimensional spheres there always exists a non vanishing vector
field. Indeed consider for n ≥ 1 the vector field in R2n
X = x2 ∂x1 − x1 ∂x2 + . . . + x2n ∂x2n−1 − x2n−1 ∂x2n
It is easy to see that this vector field is never vanishing and is tangent to (i.e., restricts to a well
defined vector field on) the sphere S 2n−1 ⊂ R2n .
2
one can take C = sup{kDX(x)k, x ∈ S n (1)}
3
Recall that any proper continuous map f : X → Y between smooth manifold is closed. If X is compact,
every continuous map f : X → Y is proper. Indeed in Y , compact sets are closed (assuming Y is Hausdorff). f
is continuous, so the inverse image of a closed set is closed. But a closed subset of a compact (Hausdorff) space is
compact. So the inverse image of a compact set is compact.
97
The Haar measure on compact Lie groups
CorollaryR 8.28. If G is a compact Lie group there exists a unique left-invariant volume form Ω
such that G Ω = 1, called the Haar measure on G.
A Lie group G is called unimodular if all left-invariant volume forms Ω are also right-invariant.
Remark 8.29. Some consequences: S n is not paralelizable (and T S n is not trivial) for n even, . In
particular there cannot exists a Lie group structure on S n for n even. Indeed one can prove that
the only spheres which carry a Lie group structure are S 1 and S 3 (together with S 0 ' Z2 ).
Proof. It is enough to take any volume form ν and then set Ω = c−1 ν with c = G ν. Notice that c
R
is finite since G is compact.
We stress that on any Lie group it is possible to speak about Haar measures (plural) for left
invariant volume forms, but there is not the possibility of fixing a normalization in general since
the volume might be infinite.
8.3 Manifolds with boundary

We define the half space in Rn
Hn = Rn+ := {x ∈ Rn | xn ≥ 0}.
Notice that ∂Hn = {xn = 0}

A boundary chart for a topological space M is a pair (U, ϕ) where U ⊂ M open and ϕ : U → V
homeomorphism on an open set V of Hn with the subspace topology Hn ⊂ Rn . In other terms
V = A ∩ Hn for some open set A in Rn .
Repeating all the theory and replacing “charts” with “boundary charts” we have a definition
of smooth manifold with boundary. This is a topological space M which is second countable and
Hausdorff, locally homemorphic at every point to an open set of Hn .
The boundary ∂M of a manifold with boundary M is the set of points of M for which there
exists a chart (U, ϕ) containing this point belong to ϕ−1 (∂Hn ∩ ϕ(U )). This is well defined in
the sense that if given q ∈ M there exists a chart (U, ϕ) such that q ∈ ϕ−1 (∂Hn ∩ ϕ(U )), then
q ∈ (ϕ0 )−1 (∂Hn ∩ ϕ0 (U 0 )) for all other charts (U 0 , ϕ0 ) containing q.
Lemma 8.30. If M is a n-dimensional manifold with boundary then ∂M is a (n − 1)-dimensional

manifold without boundary.
The property just stated of the boundary says ∂(∂M ) = ∅, hence we can say ∂ ◦ ∂ = 0. As the
exterior derivative d, the operator ∂ squares to zero. The Stokes theorem will add one more reason
to feel that d and ∂ are related one to each other (in a suitable sense, they are dual operators).
Exercise 8.31. Let f : Rn → R be a smooth function. Assume that 0 is a regular value for f ,
hence f −1 (0) is a smooth (n − 1)-dim manifold. Prove that M = {x ∈ Rn | f (x) ≥ 0} is a manifold
with boundary ∂M = f −1 (0).
For instance B n = {x ∈ Rn : kxk ≤ 1} is a manifold with boundary of dimension n and

∂B n = S n−1 , which is correctly a (n − 1)-dimensional manifold without boundary.
98
Induced orientation on the boundary
A first result is a result about orientation of hypersurfaces. Recall that given a n-form ω on M and
a vector field X we can define a (n − 1)-form
iX ω(Y1 , . . . , Yn−1 ) = ω(X, Y1 , . . . , Yn )
Given a hypersurface S ⊂ M , we say that a vector field is transverse to S if X(q) + Tq S = Tq M

for every q ∈ M
Proposition 8.32. Let M be an oriented smooth n-dim manifold and S be an hypersurface of M .

Let X be a transverse vector field to S.
(a) there exists a unique orientation on S induced by X and compatible with M , in the sense that
Y1 , . . . , Yn−1 is positive on S if and only if X, Y1 , . . . , Yn−1 is positive on M .
(b) this orientation is induced by the volume form iX ω|S , where ω is an orientation for M .
Example 8.33. S n is an hypersurfaces of Rn+1 which can be oriented compatibly with the standard
orientation of Rn+1 induced by the transverse vector field
n
X ∂
X= xi
∂xi
i=0
This orientation is the one induced by iX ω where ω = dx1 ∧ . . . ∧ dxn hence
iX ω = iX (dx0 ∧ . . . ∧ dxn )
Xn
= (−1)i dx0 ∧ . . . ∧ iX dxi ∧ . . . ∧ dxn
i=0
n
X
= (−1)i xi dx0 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxn
i=0
which is exactly the volume form we used before on the sphere.
Not every smooth hypersurface admits an everywhere transverse vector field (this is indeed
equivalent to the existence of a nonzero section of a suitably defined normal bundle!). But we have
the following important result.
Proposition 8.34. Let M be a smooth oriented manifold with boundary. Then there exists a
smooth transverse vector field X to ∂M .
Proof. Let q ∈ ∂M . We say that v ∈ Tq M is inward pointing if v is the tangent vector to a curve
γ : [0, ε] → M that is contained in M , outward pointing otherwise (that is if −v is inward pointing).
In charts these vectors are exactly those with xn > 0 (resp. xn < 0).
Cover now a neighborhood of ∂M by smooth boundary charts {(Ui , ϕi )}. In each chart fix
Ni = −∂/∂xn . Then set with a partition of unity
X
X= ψi Ni
i∈N
99
This is a smooth vector field on ∂M . Let us check it is outward pointing. Fix a point q and a chart
(y1 , . . . , yn ) at this point. Each Ni defined at q is outward pointing hence dyn (Ni ) < 0. Hence we
have X
dyn (X)|q = ψi (q)dyn (Ni )|q < 0
i∈N
because all terms are ≤ 0 an at least one is < 0.
Corollary 8.35. Let f : Rn → R be a smooth function. Assume that 0 is a regular value for
f , hence f −1 (0) is a smooth (n − 1)-dim manifold. Then f −1 (0) = ∂M is orientable, where
M = {x ∈ Rn | f (x) ≥ 0}.
This could also more easily be done via some Riemannian metric defining a transverse vector
field.
Definition 8.36 (Stokes orientation of the boundary). Let M be oriented smooth manifold. Then
∂M is orientable and the orientation induced by any tranverse vector field outward pointing is
well-defined. This is called the Stokes orientation of ∂M .
We have to consider these three main examples
• The case of a domain Ω in R2 and its boundary.
• The sphere S n is the boundary of the ball B n . Its Stokes orientation is exactly the one already
considered.
• The boundary ∂Hn of Hn . Notice that N = −∂xn is an outward pointing vector. Hence the
boundary is positively oriented by the chart (x1 , . . . , xn−1 ) 7→ (x1 , . . . , xn−1 , 0) only if the
family of vectors
(−∂xn , ∂x1 , . . . , ∂xn−1 )
is positively oriented in Rn , which gives (−1)n , in the sense that, the induced orientation by
the standard chart coincides with the Stokes orientation only if n is even, and its opposite if
n is odd.
8.4 Stokes theorem

We can now prove the main theorem of the chapter.
Theorem 8.37. Let M be a smooth oriented n-dim manifold with boundary and let ω be a compactly
supported smooth (n − 1)-dim form on M . Then
Z Z
dω = ω.
M ∂M
Here ∂M is understood with the induced orientation given by M . If M has no boundary then
the right hand side is zero.
100
Proof. We split the proof into three part (i) M = Hn (ii) M is covered by a single chart (iii) the
general case.
(i) The fact that ω has compact support means that
n
X
ω= ωi dx1 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxn
i=1
with supp(ω) contained in a rectangle A := [−R, R]n−1 × [0, R]. We have

n
X
dω = dωi ∧ dx1 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxn
i=1
hence
n
X ∂ωi
dω = (−1)i−1 dx1 ∧ . . . ∧ dxi ∧ . . . ∧ dxn
∂xi
i=1
We start the integration over Hn

and we do it in such a way that (a) we split the part of the sum
which goes normal to the boundary with the others
Z Z n−1
X ∂ωi
dω = (−1)i−1 dx1 ∧ . . . ∧ dxi ∧ . . . ∧ dxn
Hn Hn i=1 ∂xi
Z
∂ωn
+ (−1)n−1 dx1 ∧ . . . ∧ dxi ∧ . . . ∧ dxn
Hn ∂xn
and (b) we integrate first along the variable xi in the first addend of the sum, and wrt xn in the
last one
Z n−1 Z RZ R Z R
X ∂ωi
dω = (−1)i−1 ... dxi dx1 . . . dx
ci . . . dxn
Hn 0 −R −R ∂xi
i=1
Z R Z R Z R
n−1 ∂ωn
+ (−1) ... dxn dx1 . . . dxn−1 (8.2)
−R −R 0 ∂xn
The first addend of the sum is zero since for every i = 1, . . . , n − 1

Z R
∂ωi
dxi = ωi (R) − ωi (−R) = 0.
−R ∂xi
RR n
Similarly, the last one gives 0 ∂ω
∂xn dxn = −ωn (x1 , . . . , xn−1 , 0) since at xn = R it vanishes. Thus
(8.2) reduces to
Z Z R Z R
n
dω = (−1) ... ωn (x1 , . . . , xn−1 , 0)dx1 . . . dxn−1
Hn −R −R
(Notice that this term is zero in the case the support of ω does not intersect the boundary). The
other side of the equality is
Z n Z
X
ω= ωi (x1 , . . . , xn−1 , 0)dx1 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxn
∂Hn i=1 ∂Hn
101
Since xn is constantly equal to zero on ∂Hn all the terms containing dxn vanish, and only the i = n
term survives Z Z
ω= ωn (x1 , . . . , xn−1 , 0)dx1 ∧ . . . ∧ dxn−1
∂Hn ∂Hn
Considering the fact that the coordinates (x1 , . . . , xn−1 ) are positively oriented on ∂Hn if n is even
and negatively if n odd we have
Z Z R Z R
n
ω = (−1) ... ωn (x1 , . . . , xn−1 , 0)dx1 . . . dxn−1
∂Hn −R −R
which proves the statement if M is the half space.

(ii) If M is covered by a single chart (U, ϕ) (which we assume to be an oriented chart). We have
M = ϕ(U ) and ∂M = ϕ−1 (∂Hn ∩ ϕ(U ))
Z Z Z
−1 ∗
dω = (ϕ ) dω = d(ϕ−1 )∗ ω
M ϕ(U ) ϕ(U )
now we use the result in Hn and we have (notice ∂ϕ(U ) = ϕ(U ) ∩ ∂Hn )
Z Z Z Z
−1 ∗
dω = (ϕ ) ω = ω= ω
M ∂ϕ(U ) ϕ−1 (∂Hn ∩ϕ(U )) ∂M
(iii) Assume now that supp(ω) is compact in M and cover it with finitely
PN many oriented charts
{(Ui , ϕi )}N
i=1 and let ψi a subordinate smooth partition of unity, with i=1 ψi = 1. Then we have
Z N Z
X N Z
X
ω= ψi ω = d(ψi ω)
∂M i=1 ∂M i=1 M
XN Z
= (dψi ∧ ω + ψi dω)
i=1 M
N N
Z ! Z ! Z
X X
= d ψi ∧ω+ ψi dω = dω
M i=1 M i=1 M
Remark 8.38. The assumption of compact support for ω can clearly be removed if M is compact.
If M is non compact the theorem does not hold for arbitrary non compactly supported forms, as
there can be evident integrability issues for both terms of the equality to be defined. Even if both
terms are defined,
R they can be Rdifferent. For instance if M = [0, +∞[ and f = 1 is a 0-form (a
function) then M df = 0 while ∂M f = −1.
There are some special cases. For instance in R2 on a region D we have the following
Corollary 8.39. (Green formula in R2 ) Let D be a regular domain in R2 and ω = P dx + Qdy.
Then Z Z
∂Q ∂P
− dxdy = P dx + Qdy
D ∂x ∂y ∂D
As a particular case if D is a region enclosed by a curve γ oriented in the counter clockwise sense
Z
1
Area(D) = xdy − ydx
2 γ
102
Remark 8.40. Notice that if D is a region enclosed by a curve γ : [0, 1] → R2 oriented in the counter
clockwise sense and we define in R2 the vector field X(x, y) = (Q(x, y), −P (x, y)) we can rewrite
as a divergence Theorem (check the next chapter)
Z Z
div(X)dxdy = X · ν
D γ
There are similar result in R3 . The reader is invited to write down explicit formulas and find
back calculus formulas.
Another example: the Curl Theorem. The line integral of a vector field over a loop γ is equal
to the flux of its curl through the surface S whose boundary is ∂S = γ.
Z Z
(∇ × F )dA = F · ds
S ∂S
where dA is the area element on S and ds is the lin element on γ = ∂S. Recall that ∇×F = curl(F )
is equal to  
∇ × F = det  
Remark 8.41. Exactly as one can define smooth manifolds with boundary by locally modeling
topological spaces (with suitable assumptions) on open sets of Hn , one can define “manifold with
corners” by locally modeling on open sets of
Rn+ = {(x1 , . . . , xn ) | xi ≥ 0}.
There holds a version of Stokes theorem for manifold with corners. In particular this allows for
curves that are piecewise smooth.
Line integrals
We can integrate 1-forms ω ∈ Ω1 (M ) along smooth curves γ : [a, b] → M . By definition of integral
we can use γ as a chart and we have
Z Z
ω := γ ∗ ω.
γ [a,b]
Recall that if in coordinates ω = i=1 ωi dxi then γ ∗ ω = ni=1 ωi (γ(t)) hdxi , γ̇(t)i hence we get the
Pn P
following explicit formula for the integral
Z Z b
ω= ωγ(t) , γ̇(t) dt. (8.3)
γ a
Clearly the integral is linear and independent on the parametrization. If γ : [a, b] → M is piecewise
smooth we can extend the definition (8.3) by adding the pieces
Proposition 8.42 (Fundamental theorem of calculus). Let f : M → R and γ : [a, b] → M piecewise

smooth. Then Z
df = f (γ(b)) − f (γ(a))
γ
103
Proof. Indeed we have
Z Z b Z b
df = dfγ(t) , γ̇(t) dt = (f ◦ γ)0 (t)dt = f (γ(b)) − f (γ(a))
γ a a
which can be also seen as a consequence of the Stokes theorem.

R
Notice in particular that the value of the integral γ df is independent on the path joining
x = γ(a) and y = γ(b). In particular if the curve γ is closed but non trivial, the integral is always
zero. We can be more precise.
Proposition 8.43. Let ω ∈ Ω1 (M ). The two following properties are equivalent

R
(a) γ ω = 0 for every closed curve γ in M ,
(b) ω = df for some smooth f : M → R. (i.e., ω is exact)
Proof. Only (a) implies (b) is needed. Fix an arbitrary point q0 ∈ M , we want to prove that the
function f satisfying is Z q
f (q) = ω
q0
where the integral is computed along any curve joining q0 with q (well-defined thanks to (a)).
Notice that if we replace q0 by any other point q1 in M we get a different function fe, which differs
by a constant, hence dfe = df . Hence it is sufficient to prove that in coordinates we have
∂f
(x) = ωi (x)
∂xi
Notice that if ω = df then in coordinates

n
X ∂f
ω= ωi dxi , ωi =
∂xi
i=1
Hence
∂ωi ∂2f ∂2f ∂ωj
= = =
∂xj ∂xj ∂xi ∂xi ∂xj ∂xi
This means that ω is closed. Hence exact implies closed. Closed differential 1-forms are not
necessarily exact.
Example 8.44. Let us consider the 1-form in R2
xdy − ydx
ω=
x2 + y 2
This is closed but not exact on R2 \ {0} since the integral over the circle of radius 1 parametrized
as t 7→ (cos t, sin t) is 2π (the length of the circle!). It is exact on every ball not containing zero
since ω = dθ where θ = arctan(y/x).
104
On closed and exact forms
Definition 8.45. We say that a differential k-form ω ∈ Ωk (M ) is closed if dω = 0, we say that ω
is exact if ω = dη for some η ∈ Ωk−1 (M ).
Every exact form is closed since if ω = dη then dω = d2 η = 0. The converse is not true.
Corollary 8.46. Let M be a compact smooth manifold without boundary. Then the integral over
M of an exact form is zero.
Proof. If ω = dη we have since ∂M = ∅

Z Z Z
ω= dη = η=0
M M ∂M
Hence we have a way to check whether a closed form is exact or not by restating the previous
results.
Corollary 8.47. Let M be a smooth manifold

R and ω be a closed k-form. If there exists S compact
submanifold without boundary such that S ω 6= 0 then ω is not exact.
Example 8.48. Let us consider the 1-form in R2 \ {0}
xdy − ydx
ω=
x2 + y 2
This 1-form is not exact on R2 \ {0} since the integral over the circle of radius 1 parametrized as
t 7→ (cos t, sin t) is 2π (it is the length of the circle!). The form ω is closed since

∂ y ∂ x
dω = + dx ∧ dy = 0
∂y x2 + y 2 ∂x x2 + y 2
Corollary 8.49. Let M be a compact smooth manifold with boundary. Then the integral over ∂M
of a closed form is zero.
Proof. If dω = 0 we have Z Z
ω= dω = 0
∂M M
Example 8.50. The circle of radius 1 is not a boundary R2 \ {0}! Indeed if you want to see it as
a boundary in R2 \ {0} you must consider it as a connected component of the boundary ∂D of an
annulus type domain D.
Homotopy invariance
Definition 8.51. Two smooth paths γ0 , γ1 : [0, 1] → M are smoothly homotopic if there exists a
smooth map H : [0, 1] × [0, 1] → M such that H(0, t) = γ0 (t) and H(1, t) = γ1 (t).
Thanks to Stokes theorem (version with corners!) we have
105
Theorem 8.52. Let ω be a closed 1-form on a smooth manifold M , i.e., dω = 0. Let γ1 , γ2 be two
homotopic path with the same endpoints. Then
Z Z
ω= ω
γ0 γ1
Proof. Since ω is closed we have

Z Z
∗
d(H ω) = H ∗ dω = 0
[0,1]2 [0,1]2
On the other hand by Stokes theorem (version with corners!)

Z Z
∗
d(H ω) = H ∗ω
[0,1]2 ∂[0,1]2
and
Z 4 Z
X
∗
H ω= ω
∂[0,1]2 i=1 H◦Γi
where Γi for i = 1, . . . , 4 are counter clockwise parametrization of the 4 edges of the square [0, 1]2 .
Analyzing the terms we have
4 Z
X Z Z Z Z
ω= ω− ω= ω− ω
i=1 H◦Γi H◦Γ1 H◦Γ3 γ0 γ1
Remark 8.53. One can adapt the proof to just a continuous homotopy between two curves that are
piecewise smooth.
The difference between closed and exact forms is topological, hence global. In the same spirit
we have this important result, which we will only sketch for the moment.
Theorem 8.54 (Poincare Lemma for 1-forms). Let U ⊂ Rn be open starshaped, then any closed
1-form ω is exact on U .
Proof. Fix x0 ∈ U and defines f (x) as the line integral of ω along the line segment from x0 ∈ U
to x. Using the fact that ω is closed (and differentiation under the integral sign) one shows that
df = ω.
Indeed the same is true for k-forms.
Theorem 8.55 (Poincare Lemma for k-forms). Let U ⊂ Rn be open starshaped, then any closed
k-form ω is exact on U .
This means that closed is equivalent to locally exact. We end this chapter by defining the de
Rham cohomology groups
106
Definition 8.56. Let M be a smooth manifold, let Z k (M ) be the set of closed k-forms and B k (M )
the set of exact k-forms. We set
k Z k (M )
HdR (M ) = k
B (M )
where ω1 ∼ ω2 in Z k (M ) if ω1 = ω2 + dη for some η ∈ Ωk−1 (M ).
Thanks to the Poincaré Lemma we can conclude that

0
HdR (Rn ) = R, k
HdR (Rn ) = 0, k≥1
Exercise 8.57. Prove that if F : M → N is a smooth map then F ∗ maps closed forms into closed
forms and exact forms into exact forms. Hence induces a well-defined map F ∗ : HdR
k (N ) → H k (M ).
dR
Prove that two diffeomorphic manifold have isomorphic de Rham cohomology groups
More in general we have

0 (M ) ' R. If M is a connected, orientable and
Proposition 8.58. If M is connected, then HdR
n
compact manifold. Then HdR (M ) ' R.
Indeed the de Rham cohomology groups are not only preserved by diffeomorphisms but also
homotopies. The de Rham theorem states that these groups coincide with the singular cohomology
groups that one can define with algebraic topology approach. These results goes beyond the scope
of these lecture notes and we invite the interested reader to have a look to for further results in
this directions.
107
Chapter 9
Riemannian manifolds
The investigation of this more general kind would require no really different principles [. . . ]
I restrict myself, therefore, to those manifolds in which the line element
is expressed as the square root of a quadratic differential.
Ueber die Hypothesen, welche der Geometrie zu Grunde liegen, 1867
Bernhard Riemann, 1826 – 1866
A (covariant) 2-tensor T : V × V → R on a finite dimensional vector space V is said to be

non-degenerate if T (v, w) = 0 for all w ∈ V implies v = 0. This is equivalent to ask one of the
following two equivalent conditions:
• the map L : V → V ∗ given by L(v) = T (v, ·) ∈ V ∗ is an isomorphism,
• for every basis e1 , . . . , en of V the matrix (T (ei , ej ))i,j is invertible.
Endowing a vectors space with a non-degenerate 2-tensor hence defines a natural isomorphism
V → V ∗ induced by T . This is what happens for instance given an inner product on V , which
can be thought as a non degenerate symmetric (and positive) 2-tensor h·, ·i : V × V → R. The
above isomorphism is just a finite dimensional version of the so-called Riesz representation theorem
x ∈ V 7→ φx : V → R such that φx (v) = hv, wi.
Extending this idea to manifolds, every non-degenerate 2-tensor field T on M will induce a
natural isomorphism from T M to T ∗ M producing a one to one correspondence between differential
one forms and vector fields. This is at the base of the definition of gradient ∇f of a smooth
function f on a Riemannian manifold, and also of Hamiltonian vector field ~h associated to a
smooth function h on a symplectic manifold N . Both objects are vector fields associated to the
differentials of the corresponding functions with respect to a tensor field, which is symmetric positive
and non-degenerate in the case of the Riemannian manifold (the Riemannian inner product), and
skew-symmetric in the case of the symplectic manifold (the symplectic form).
We will illustrate similarities and differencies in what follows, and show how these notions
naturally meet when considering the geodesic flow on a Riemannian manifold, which is also the
flow of a Hamitonian vector field.
109
9.1 Riemannian structure
A Riemannian metric on a smooth manifold M is a covariant 2-tensor field g (i.e., a section of the
tangent bundle of covariant 2 tensor on M ) that is symmetric and positive definite. This means
that for every q ∈ M we have
g : Tq M × Tq M → R
which is a symmetric and positive definite bilinear form, i.e., an inner product. Sometimes the
inner product is also denoted by hv | wiq (or simply hv | wi when the notation is clear), the vertical
bar distinguishing the inner product of vectors from the duality product of a covector on a vector.
This permits to define norm of vectors and angle between vectors as usual
g(v, w)
kvk2 = g(v, v), cos(vw)
c = .
kvkkwk
A Riemannian metric in coordinates on a chart (U, ϕ) is written as

n
X
g= gij (x) dxi ⊗ dxj
i,j=1
where (gij (x)) is a symmetric and positive definite matrix whose entries are smooth functions
gij ∈ C ∞ (U ). Notice that in terms of the standard basis induced by coordinates

∂ ∂
gij = g ,
∂xi ∂xj
Proposition 9.1. Every smooth manifolds admits a Riemannian metric
Proof. Take a covering of the manifold {Uα , ϕα } and a partition of unity {ψα } and set
X
g= ψα gα
α
where gα = ϕ∗α g is the pullback in Uα of the Euclidean metric on Rn . It is an easy check that all
conditions are satisfied.
Given a Riemannian metric g on a smooth manifold M we say that (M, g) is a Riemannian

manifold.
Definition 9.2. An isometry between two Riemannian manifolds (M, g) and (M 0 , g 0 ) is a smooth
diffeomorphism F : M → M 0 such that F ∗ g 0 = g. This implies that for every q ∈ M and v, w ∈ Tq M
gq (v, w) = gF0 (q) (F∗ v, F∗ w)
Using local diffeomorphisms we can define local isometries. Notice that in this definition isome-
tries are smooth. Isometries preserves length of vectors and angles between them.
Definition 9.3. An conformal tranformation between two Riemannian manifolds (M, g) and
(M 0 , g 0 ) is a smooth diffeomorphism F : M → M 0 such that F ∗ g 0 = ef g for some f ∈ C ∞ (M ).
110
Two metrics that are conformal defines the same angles but do not define the same length.
Exercise 9.4. Let π : S n → Rn be the stereographic projection from the north pole N ∈ S n .
Prove that if gS n is the standard metric on S n and ḡ the Euclidean metric on Rn then on S n \ {N }
we have gS n = ef π ∗ ḡ.
Example 9.5. Metric on immersed submanifolds of RN .
Remark 9.6. On a Riemannian manifold (M, g) we can always define locally an orthonormal frame
X1 , . . . , Xn of vector fields such that
g(Xi , Xj ) = δij
Indeed it is enough to apply the Gram Schmidt algorithm to the smooth family of vectors given
by the frame. In particular given a non vanishing vector field X we can find a local orthonormal
frame X1 , . . . , Xn such that X = X1 .
The metric g induces an isomoprhism between tangent and cotangent spaces
I : Tq M → Tq∗ M, v 7→ g(v, ·).
In the standard basis induced by coordinates the matrix associated to I is computed as follows
X n
∂
g ,· = aij dxj
∂xi
j=1
∂
and it is easy to see that aij = gij by applying both sides to ∂xj .
Remark 9.7. The isomorphisms
I : Tq M → Tq∗ M, I −1 : Tq∗ M → Tq M,
which are represented respectively by G = (gij ) and G−1 = (g ij ) in coordinates, are also called
musical isomorphism because using proper indices notation I(v) = v [ lower indices and I −1 (η) = η ]
raises indices. Indeed, with Einstein summation notation
∂
v = vi , v [ = vj dxj , vj = gij v i .
∂xi
∂
η = ηj dxj , η] = ηi , ηi = g ij ηj .
∂xi
Definition 9.8. Let f ∈ C ∞ (M ), the Riemannian gradient of f is the vector field ∇f = I −1 (df ),
that is the vector field satisfying
g(∇f, v) = df (v), ∀v ∈ TM
In coordinates it is simple to check that

n
X ∂f ∂
∇f = g ij
∂xi ∂xj
i,j=1
111
Exercise 9.9. Let X1 , . . . , Xn be local o.n frame for the metric and let η1 , . . . , ηn be the dual basis.
Prove that
Xn
g= ηi ⊗ ηi
i=1
Prove that the differential of a function f and its Riemannian gradient are written as
n
X n
X
df = (Xi f )ηi , ∇f = (Xi f )Xi .
i=1 i=1
Riemannian volume
If M is orientable Riemannian manifold we can define a natural volume, called Riemannian volume
on M that is the volume dV that has value 1 on some (equivalently, every) positive orthonormal
basis.
Proposition 9.10. The Riemannian volume is written in coordinates
p
dVg = det gij dx1 ∧ . . . ∧ dxn
Proof. We write
dV = f dx1 ∧ . . . ∧ dxn
and we look for the function f . Consider a local orthonormal frame X1 , . . . , Xn
n
X ∂
Xi = bij
∂xj
j=1
Then we have by Lemma 7.29

∂ ∂
1 = dV (X1 , . . . , Xn ) = det(B) dV ,..., = det(B)f
∂x1 ∂xn
If g(Xi , Xk ) = δij we have that
 
n n n
X ∂ X ∂  X
δik = g bij , bkl = bij gjl bkl
∂xj ∂xl
j=1 l=1 j,l=1
that means the matrix identity BGB T = I. In particular det(B)2 = det(G)−1 , which completes
the proof.
Definition 9.11. Let ω ∈ Ωn (M ) be a volume form. Then for a vector field X on M we denote
by divω X the function such that LX ω = (divω X)ω
Exercise 9.12. Prove that given f > 0 smooth positive function
divf ω X = divω X + Xf
In particular if dq f = 0 at some point, the divergence of f at q is independent with respect to the

volume form.
112
On a Riemannian manifold we simply write div X for the divergence with respect to the Rie-
mannian volume dVg .
P
Exercise 9.13. Prove that the Riemannian divergence of X = i Xi ∂/∂xi is written in coordi-
nates as
n
1 X ∂ p
div(X) = p det gij Xk
det gij k=1 ∂xk
where we assume orientability such that det gij > 0.
Exercise 9.14. Let X1 , . . . , Xn be local o.n frame for the metric and let η1 , . . . , ηn be the dual
basis. Prove that
dVg = η1 ∧ . . . ∧ ηn
For a vector field Y compute first LY (dVg ) and then div(Y ) in terms of a local o.n. frame assuming
that
n
X
[Xi , Xj ] = ckij Xk .
k=1
We can give the following statement.

Theorem 9.15. Let (M, g) be an orientable Riemannian manifold with boundary, dVg the Rie-
mannian volume and dSg the surface measure on ∂M with the Stokes orientation. Then for every
compactly supported vector field X we have
Z Z
(div X)dVg = g(X, N )dSg
M ∂M
where N is an outward unit normal vector field to ∂M .
Proof. Since X is compactly supported also div X is compactly supported and
Z Z Z Z
(div X)dVg = LX dVg = d(iX dVg ) = iX dVg
M M M ∂M
and the proof is completed by observing that iX dVg = g(X, N )dSg . Notice that we used LX dVg =
d(iX dVg ) + iX d(dVg ) but d(dVg ) = 0 since dVg is a top dimensional form.
Remark 9.16. Under the same assumptions if f is a compactly supported function we have
Z Z Z
f (div X)dVg + Xf dVg = f · g(X, N )dSg
M M ∂M
Notice that if M = Ω an open set of R3

with the standard inner product denoted by x1 · x2
3 Z Z Z
X ∂Xi ∂f
f (x) (x)dx + Xi (x) (x)dx = f (y)(X(y) · ν(y))dσ(y)
Ω ∂xi Ω ∂xi ∂Ω
i=1
This is nothing but the version “without” f
3 Z Z
X ∂Xi
(x)dx = (X(y) · ν(y))dσ(y)
Ω ∂xi
i=1 ∂Ω
applied to f X and Leibnitz rule.
113
9.2 The metric structure
We can introduce the length of a piecewise smooth curve γ : [0, T ] → M
Z T
L(γ) = kγ̇(t)kdt
0
The length is invariant by reparametrization. A curve that has finite length can always be
reparametrized by arc lenght parameter s defined by
Z t
s(t) = kγ̇(t)kdt
0
We can introduce the intrinsic distance (called Riemannian distance)
d(x, y) = inf{L(γ) | γ : [0, T ] → M, γ(0) = x, γ(T ) = y}
Theorem 9.17. Let (M, g) be a Riemannian manifold and d be the induced distance. Then (M, d)
is a metric space whose metric topology coincides with the manifold topology.
Proof. The fact that d is non negative and symmetric is easy. The fact that we have chosen as class
of curves those that are piecewise smooth helps to prove the triangular inequality which is trivial.
We have only to prove that x 6= y implies d(x, y) 6= 0, and that the two topologies coincides. But
the former fact is implied by the latter. We start with a lemma
Lemma 9.18. Let g be any Riemannian metric on Rn . The for every compact K ⊂ Rn there exists
a constant such that
c1 kvkRn k ≤ kvkg ≤ c2 kvkRn ,
for every x ∈ K and v ∈ Tx Rn .
Proof. The set of tangent vectors v to Rn of norm 1 and based in K is compact, and the norm
associated to g is smooth. Hence there exists constant such that c1 ≤ kvkg ≤ c2 for every such unit
v. By homogeneity one conclude for every v.
The previous lemma says that for every smooth curve γ sufficiently short to stay in a chart then
we can compare its length with the Euclidean length in the chart and
c1 LRn (γ) ≤ Lg (γ) ≤ c2 LRn (γ)
Now let U be open in the manifold topology, fix x ∈ U let us find a ball B(x, ε) contained in
U . It is enough to take a chart on a set V ⊂ U with compact closure and notice that if a smooth
curve is not entirely contained in V then Lg (γ) ≥ c1 LRn (γ) ≥ c1 δ. The converse implication says
that for ε = c1 δ we have1 B(x, ε) ⊂ V ⊂ U .
The other implication is similar and is left as an exercise for the reader (cf. Exercise 9.19).
Exercise 9.19. Prove the remaining implication: for every open metric ball B(x, ε) there exists
an open set U in the manifold topology contained in B(x, ε).
1
notice that this part of the argument in particular says that x 6= y implies d(x, y) 6= 0.
114
Example 9.20. Let us consider the group of positive affine transformation of the real line
f (t) = at + b, a > 0, b ∈ R
This is a group with the composition as a product
(a, b) · (a0 , b0 ) = (aa0 , ab0 + b)
Putting coordinates (x, y) = (b, a) we have coordinates on the upper half-plane H2 endowed with
the Lie group structure
(x, y) · (x0 , y 0 ) = (x + yx0 , yy 0 )
Notice that the neutral element of the group is e = (0, 1) and (x, y)−1 = (−xy −1 , y −1 ). The
left-invariant vector fields X, Y with coincide with ∂x , ∂y at the identity are
X = y∂x , Y = y∂y
and the left-invariant Riemannian metric

1
g= (dx2 + dy 2 )
y2
which is a model for the hyperbolic plane. Write in complex coordinates z = x + iy
dx2 + dy 2 dzdz
g= 2
= −4
y (z − z)2
Prove that all maps of the form T : z 7→ az+b

cz+d with real coefficients and ad−bc = 1 are isometries. In
particular all such transformations are obtained by composing the following: translations, dilations
of positive factor and inversion (with reflection) with respect to the unit circle
1
Tb : z 7→ z + b, Ta : z 7→ a2 z, T± : z 7→ ±
z
we have
1
T− Tb Ta : z 7→ −
a2 z +b
Exercise 9.21. Prove that a piece of vertical segment is length-minimizer in the example. What
about a piece of horizontal segment?
9.3 Length-minimizers and geodesics

Definition 9.22. Let γ : [0, T ] → M such that γ(0) = x and γ(T ) = y, we say that γ is a
length-minimizer if d(x, y) = `(γ).
The existence of length-minimizers is not guaranteed in general. Counterexamples are very easy
to build, such as R2 \ {0} with the Euclidean metric.
Theorem 9.23. Let (M, g) Riemannian manifold and assume (M, d) is a complete metric space.
Then for every x, y ∈ M there exists a length-minimizer joining x and y.
115
To prove this theorem one needs the following version of Arzela-Ascoli theorem in metric spaces.
Proposition 9.24 (Arzela-Ascoli). In a compact metric space, any sequence of curves with uni-
formly bounded lengths contains a uniformly converging subsequence.
For the moment we will not prove Theorem 9.23. A key property is the semicontinuity of the
length-functional: if γn is a sequence of curves with fixed endpoints which converges uniformly to
γ then
`(γ) ≤ lim inf `(γn )
n→∞
We discuss necessary conditions satisfied by length minimizers. We refer to [].
Necessary conditions
Let us discuss necessary conditions to be minimizers. Recall that every curve can be reparametrized
by constant speed.
Lemma 9.25. Let T > 0 be fixed. A piecewise smooth curve γ : [0, T ] → M is a length-minimizer
and has constant speed if and only if γ is a minimizer of the functional
1 T
Z
E(γ) = kγ̇(t)k2 dt
2 0
Proof. We can assume without loss of generality that γ is smooth since the integral is additive. By
the Cauchy-Schwartz inequality we have the general inequality
`(γ)2 ≤ 2E(γ)T (9.1)
Notice that Cauchy-Schwartz inequality also says that we have equality in (9.1) if and only if γ has
constant speed. The conclusion then easily follows.
We can now look to minimizers of the energy functional. Let us consider a short curve whose
support is contained in a single chart (U, ϕ) and let x(t) = ϕ(γ(t)). In coordinates
1 t
Z
E(γ) = gij (x(t))ẋi (t)ẋj (t)dt.
2 0
Lemma 9.26. A solution x : [0, T ] → Rn to the problem
1 t
Z
min gij (x(t))ẋi (t)ẋj (t)dt, x(0) = x, x(T ) = y
2 0
satisfies the system of equations
n
X
ẍk + Γkij ẋi ẋj = 0, k = 1, . . . , n. (9.2)
i,j=1
where Γkij are smooth functions defined by

n
1 X mk ∂gim ∂gjm ∂gij
Γkij = g + −
2 ∂xj ∂xi ∂xm
m=1
116
Remark 9.27. These equations makes sense only in local coordinates. For the moment no geo-
metric interpretation of the equation (9.2). We will solve this problem later with two different
interpretations.
Proof. We use the following fact from Calculus of Variation (see Appendix): solutions satisfy the
Euler-Lagrange equations for the functional
Z T
1
L(x(t), ẋ(t))dt, L(x, ẋ) = gij (x)ẋi ẋj
2 0
These equation are written as
d ∂L ∂L
− = 0, m = 1, . . . , n
dt ∂ ẋm ∂xm
∂L
We have ∂ ẋm = gim ẋi hence
d ∂L ∂gim 1 ∂gim 1 ∂gjm

= ẋi ẋj + gim ẍi = ẋi ẋj + ẋi ẋj + gim ẍi
dt ∂ ẋm ∂xj 2 ∂xj 2 ∂xi
where we used symmetry of the first term. Moreover
∂L 1 ∂gij
= ẋi ẋj
∂xm 2 ∂xm
We have
1 ∂gim ∂gjm ∂gij
gim ẍi + + − ẋi ẋj = 0
2 ∂xj ∂xi ∂xm
thus multiplying for g mk (which recall is the inverse of the metric)

1 mk ∂gim ∂gjm ∂gij
ẍk + g + − ẋi ẋj = 0
2 ∂xj ∂xi ∂xm
Notice that (9.2) can be rewritten as a first-order system in T M as follows

(
ẋk = vk ,
, k = 1, . . . , n (9.3)
v̇k = −Γkij (x)vi vj
We call geodesics curves that satisfy the necessary condition, i.e., critical points of the length
functional. It follows from the previous considerations that:
Corollary 9.28. For every q ∈ M and v ∈ Tq M there exists a unique geodesic γq,v : [0, T [→ M
defined on some open interval such that γ(0) = q and γ̇(0) = v.
Exercise 9.29. Assuming that both sides of the equality are defined, prove the following homo-
geneity property γq,v (ts) = γq,tv (s).
117
By standard continuity theorems for ODEs with respect to initial data, there exists an open
subset U in T M such that for v ∈ T M and q = π(v) the corresponding solution of the ODE is
defined for T ≥ 1.
Definition 9.30. We define the exponential map exp : U → M defined by
exp(q, v) := expq (v) = γq,v (1)
where γv is the unique geodesic such that γq,v (0) = q and γ̇q,v (0) = v.
The exponential map defines good coordinates in a neighborhood of the base point.
Proposition 9.31. Let q ∈ M . The map expq : Tq M → M is a local diffeomorphism at v = 0 and

d0 expq : Tq M → Tq M is the identity map.
Proof. We have
d d d
d0 expq (v) = expq (tv) = γq,tv (1) = γq,v (t) = v
dt t=0 dt t=0 dt t=0
Coordinates induced by expq are called normal coordinates. In these coordinates geodesics
starting from q becomes straight lines by construction. We then state two important theorem
without proofs.
Theorem 9.32. Assume that (M, d) is a complete metric space. Then exp is defined on whole
T M , i.e., all geodesics can be defined on [0, +∞[.
Proof. Let q0 ∈ M be arbitrary. It enough to show that any geodesic γv0 (t) starting from q0 ∈ M
with initial velocity v0 ∈ Tq0 M with kv0 k = 1 is defined for all t ∈ R. Recall that the pair
(x(t), v(t)), coordinates of (γv0 (t), γ̇v0 (t)), satisfy the geodesic equation in coordinates.
Let γv0 (t) be defined on [0, T [, and assume that it is not extendable to some interval [0, T + ε[.
For any sequence tj % T the sequence (γv0 (tj ))j is a Cauchy sequence on M since
d(γ(ti ), γ(tj )) ≤ |ti − tj |.
The sequence (γv0 (tj ))j∈N is then convergent to a point q1 ∈ M by completeness. Since (γ̇(tj ))j∈N
stays in a compact set, there exists a subsequence (which we denote by the same symbol) such that
(γv0 (tj ), γ̇v0 (tj )) → (q1 , v1 ). This contradicts the fact that (x(t), v(t)) is not extendable by standard
result of ODEs.
A crucial property of geodesics is that short arcs are global length-minimizers, even among
piecewise smooth curves.
Theorem 9.33. Let γ : [0, T ] → M be a geodesic. Then there exists ε > 0 such that γ|[0,ε] is the
unique length-minimizer among all piecewise-smooth curves joining γ(0) and γ(ε).
This has the following crucial implication.
Corollary 9.34. Every piecewise smooth curve γ : [0, T ] → M that is a length-minimizer is of

class C ∞ .
118
Proof. Write a proof one day
This is based on the Gauss Lemma
Lemma 9.35. Let q ∈ M and let v ∈ Tq M such that q 0 = expq (v) is defined. Let w ∈ Tq M '
Tv (Tq M ) then
g(dv expq (v), dv expq (w)) = g(v, w)
Remark 9.36. Notice that expq : Tq M → M is not a local isometry, i.e., not all geodesics in a
neighborhood of q becomes straight lines, but only those passing through q!
9.4 Appendix: on the Euler-Lagrange equations

Assume we are interested in minimizing the quantity
Z T
I(x(t)) = L(x(t), ẋ(t))dt
0
Then if x(t) is a minimizer we have I(x(t)) ≤ I(x(t) + εh(t)) for every ε ≥ 0 and arbitrary h(t).
Assuming some smoothness we need to have
d
I(x(t) + εh(t)) = 0
dε ε=0
We have
Z T
I(x(t) + εh(t)) = L(x(t) + εh(t), ẋ(t) + εḣ(t))dt
0
Z T Z T
∂L ∂L
= L(x(t), ẋ(t))dt + ε h(t) + ḣ(t)dt + o(ε)
0 0 ∂x ∂ ẋ
With an integration by parts

Z T Z T
∂L ∂L ∂L d ∂L
h(t) + ḣ(t)dt = − h(t)dt
0 ∂x ∂ ẋ 0 ∂x dt ∂ ẋ
hence the first order term in ε is zero if for every function h(t) we have
Z T
∂L d ∂L
− h(t)dt = 0
0 ∂x dt ∂ ẋ
which implies the integrand is zero

∂L d ∂L
− =0
∂x dt ∂ ẋ
119
Chapter 10
Connections, parallel transport and

curvature
Nous pensons que, après avoir surmonté les difficultés de initiation,

on se convaincra aisement que la généralité [...] contribue non seulement à l’élégance
mais aussi à l’agilité et à la perspicuité des demonstrations et des conclusions.1
Méthodes de calcul différentiel absolu et leurs applications, 1900
Gregorio Ricci-Curbastro 1853 – 1925
Tullio Levi Civita, 1873 – 1941
On a manifold, in general there is no canonical way to identify tangent spaces (or, more generally,
fibers of a vector bundle) at different points. Thus, one has to expect that a notion of derivative for
vector fields (or sections of a vector bundle), depends on a certain choice. The additional structure
required to correctly define these notions is the one of connection.
10.1 Affine connections and parallel transport

Recall that we had a way to differentiate vector fields along another one
LX Y = [X, Y ]
This corresponds to compute for γ(t) = etX (q) and Y (t) = Y (γ(t))
e−tX
∗ Y (t) − Y (0)
LX Y |q = lim
t→0 t
This seems a good object but it has a drawback:
• the map X 7→ LX Y it is not a tensor, i.e., it is not C ∞ (M ) linear since LX Y 6= f LX Y .
In this section M is just a smooth manifold, no metric g at the moment.

1
Transl. We think that after having overcome the difficulties at the beginning, one will easily convince himself
that the generality [...] contributes not only to the elegance but also to the agility and the perspicaciousness of the
proofs and the conclusions
121
Definition 10.1. A linear connection on T M is ∇ : Vec(M ) × Vec(M ) → Vec(M ) bilinear and
satisfying the following properties
(a) ∇f X (Y ) = f ∇X Y
(b) ∇X (gY ) = (Xg)Y + g∇X Y
Definition 10.2. Given a local frame E1 , . . . , En (not necessary o.n.) we define the Christoffel
symbols of ∇ associated to the frame as the set of functions Γkij satisfying
∇Ei Ej = Γkij Ek
As we will see the name is not occasional. The first observation is that ∇X Y is local.
Lemma 10.3. The value ∇X Y |q depends only on the value of X at q and on the values of Y on
a neighborhood of q.
Proof. Let E1 , . . . , En local frame (not necessary o.n.) and write
n
X n
X
X= fi Ei , Y = gj Ej
i=1 j=1
Here we treat fi and gi as functions defined on a neighborhood of a point q. Then by using the
rules
Xn
∇X Y = fi ∇Ei Y (10.1)
i=1
X n
= fi (Ei gj )Ej + fi gj ∇Ei Ej
i,j=1
X n
= (Xgk + fi gj Γkij )Ek (10.2)
i,j=1
Thanks to last formula (10.2) we can reinforce the previous locality statement
Corollary 10.4. The value of ∇X Y |q actually depends only on the value of X at q and on the
values of Y on a curve that is tangent to X at q.
We can also deduce existence
Proposition 10.5. Every smooth manifold admits linear connection on T M
Proof. The proof combines partition of unity with the following observation: given any local frame
E1 , . . . , En and a set of n3 functions Γkij the formula (10.2) defines a connection in the open set.
Example 10.6. In Rn we have a canonical connection ∇ defined as

n
X ∂
∇X Y = (XY i )
∂xi
i=1
the ordinary differentiation along X of the coefficients of Y (corresponds to the choice Γkij = 0).
122
Differentiation along curves and parallel transport
Hence given a linear connection ∇, a regular curve γ : [0, T ] → M (with γ̇ 6= 0) and a smooth
vector field Y defined only along γ, it is well defined the vector field ∇γ̇ Y along γ.
Remark 10.7 (coordinate formula). Consider a local frame E1 , . . . , En and write
n
X n
X
γ̇(t) = vi (t)Ei |γ(t) , Y (γ(t)) = yj (t)Ej |γ(t)
i=1 j=1
then from the previous formulae we get

 
n
X n
X
∇γ̇ Y |γ(t) = ẏk (t) + vi (t)yj (t)Γkij (γ(t))) Ek |γ(t) .
k=1 i,j=1
If there exists a smooth vector field X such that X|γ(t) = γ̇(t) then ∇γ̇ Y = ∇X Y on γ.
Definition 10.8. We say that a vector field Y along a smooth curve γ : [0, T ] → M is parallel
with respect to ∇ if ∇γ̇ Y = 0 along γ.
Proposition 10.9. Given any linear connection ∇. Let γ : [0, T ] → M be a smooth curve and
v0 ∈ Tγ(0) M . There exists a unique smooth vector field V along γ such that V |γ(0) = v0 and V
parallel along γ with respect to ∇.
Proof. We have to solve the non autonomous linear system of differential equations
n
X
ẏk (t) + Γkij (γ(t))γ̇i (t)yj (t) = 0, k = 1, . . . , n
i,j=1
Pn k
which can be written setting Ak,j (t) = i=1 Γij (γ(t))γ̇i (t) as
n
X
ẏk (t) + Ak,j (t)yj (t) = 0, k = 1, . . . , n
j=1
Since the differential equation is linear then the solution is global, i.e., defined on [0, T ].
The map which associates v0 with V |γ(t) is called the parallel transport of v0 along γ.
Proposition 10.10. Let γ : [0, T ] → M be a smooth curve and let v0 ∈ Tγ(0) M . Then the map
γ
P0,t : Tγ(0) M → Tγ(t) M defined by v0 7→ vt := V |γ(t) is a linear isomoprhism.
γ
Proof. The fact that P0,t is linear comes from the fact that the flow of a linear (nonautonomous)
equation is linear.
γ
Remark 10.11. Similarly we can define Ps,t for all s, t and, we have the relations
γ γ γ γ γ −1 γ
Ps,s = id, Pt,r ◦ Ps,t = Ps,r , (Ps,t ) = Pt,s .
γ
The connection ∇ and the parallel transport P0,t are intimately related.
123
Proposition 10.12. Let γ : [0, T ] → M be a smooth curve γ(0) = q and γ̇(0) = X0 . For Y vector
field along γ we have
γ −1
(P0,t ) Y (γ(t)) − Y (q)
∇X0 Y (q) = lim
t→0 t
γ
Proof. Fix a basis v1 , . . . , vn of tangent vectors at q = γ(0), and set Vi (t) := P0,t
P(vi ). By construc-
tion the vector fields Vi are parallel along γ hence ∇γ̇ Vi = 0 Write Y (γ(t)) = ni=1 yi (t)Vi (t). We
have
n
X
−1
P0,t Y (γ(t)) = yi (t)vi
i=1
We have γ −1 n
(P0,t ) Y (γ(t)) − Y (q) X yi (t) − yi (0)
lim = lim vi = ẏi (0)vi
t→0 t t→0 t
i=1
On the other hand
∇X0 Y (q) = (∇X0 yi (t))Vi |t=0 + yi (0)(∇X0 Vi )|t=0 = ẏi (0)vi
where we used that ∇X0 yi (t) = ẏi (t)
Covariant derivative
A connection ∇ permits to differentiate tensors T of type (k, l) giving a tensor ∇T of type (k + 1, l).
The formula for tensor of type (k, 0) is as follows
n
X
∇T (X1 , . . . , Xn , Y ) = Y (T (X1 , . . . , Xn )) − T (X1 , . . . , ∇Y Xi , . . . , Xn )
i=1
and the general case is similar. Given T and a vector field X we define the covariant derivative
(∇Y T )(X1 , . . . , Xn ) := ∇T (X1 , . . . , Xn , Y )
Notice that on 0-tensors (functions) we have ∇X f = Xf . For a covariant 1 tensor (differential

form) ω we have
(∇X ω)Y = ∇ω(Y, X) = Xω(Y ) − ω(∇X Y )
while for covariant 2 tensors we have
(∇X τ )(Y, Z) = ∇τ (Y, Z, X) = Xτ (Y, Z) − τ (∇X Y, Z) − τ (X, ∇X Z)
Notice that the compatibility with the metric is defined by ∇g = 0.
10.2 The Levi-Civita connection

Definition 10.13. A linear connection ∇ on a Riemannian manifold is said compatible with the
metric if
X g(Y, Z) = g(∇X Y, Z) + g(Y, ∇X Z). (10.3)
This is equivalent to ask that g is parallel, i.e., ∇g = 0.
124
Definition 10.14. The torsion of a connection is the (2, 1) tensor T : Vec(M )×Vec(M ) → Vec(M )
defined by
T (X, Y ) = ∇X Y − ∇Y X − [X, Y ]
Theorem 10.15. On a Riemannian manifold (M, g) there exists a unique linear connection that
is compatible with the metric and with zero torsion.
Proof. The proof is based on the following key fact, which is proved by the following algorithm
which we invite the reader to check:
(i) writing three times identity (10.3) for the ordered triples {X, Y, Z}, {Y, Z, X} and {Z, X, Y },
(ii) compute (1) + (2) − (3),
(iii) use the T = 0 identity ∇X Y − ∇Y X = [X, Y ],
Definition 10.16. The connection uniquely defined by Theorem 10.15 is called Levi-Civita con-
nection of the Riemannian manifold (M, g).
One obtains the following key formula.
Lemma 10.17 (Koszul formula). For a connection ∇ which is compatible with the metric g we
have the identity
2 h∇X Y | Zi = X hY | Zi + Y hZ | Xi − Z hX | Y i
+ h[X, Y ] | Zi − h[Y, Z] | Xi + h[Z, X] | Y i (10.4)
The Koszul formula says in particular that if ∇ exists it is unique. The existence is guaranteed
by formula (10.4) once one proves that the right hand side correctly defines a connection, which is
left as an exercice.
Remark 10.18. Two particular cases: Koszul formula to a frame that is commuting [Ei , Ej ] = 0
only the first line is non zero.
2 h∇X Y | Zi = X hY | Zi + Y hZ | Xi − Z hX | Y i (10.5)
On the other side, applying Koszul formula to a o.n. frame g(Xi , Xj ) = δi,j then only the commu-
tator shows up.
2 h∇X Y | Zi = h[X, Y ] | Zi − h[Y, Z] | Xi + h[Z, X] | Y i (10.6)
Exercise 10.19. The Christoffel symbols of the Levi Civita connection associated with the frame
∂
Ei = ∂x i
satisfy
n
k 1 X mk ∂gim ∂gjm ∂gij
Γij = g + −
2 ∂xj ∂xi ∂xm
m=1
These are the symbols found in the previous chapter.
This permits to interpret as geodesics as those curves which the acceleration in the sense of the
Levi Civita connection is zero.
125
Proposition 10.20. A smooth curve γ : [0, T ] → M parametrized by constant speed is a geodesic
if and only if ∇γ̇ γ̇ = 0
Of course we can also take a local o.n. frame and use it to build a connection.
Lemma 10.21. Let X1 , . . . , Xn be a local o.n. frame such that [Xi , Xj ] = ckij Xk Prove that the
Christoffel symbols of the Levi Civita connection associated to this frame ∇Xi Xj = Γkij Xk are
written as
1
Γkij = (ckij − cijk + cjki )
2
In particular we have the property Γkij = −Γjik .
Proof. Use the Koszul formula for o.n. frames
2 h∇X Y | Zi = h[X, Y ] | Zi − h[Y, Z] | Xi + h[Z, X] | Y i (10.7)
onto the basis X = Xi , Y = Xj and Z = Xk .

γ
Proposition 10.22. Let γ : [0, T ] → M be a smooth curve and v0 ∈ Tγ(0) M . The map P0,t :
Tγ(0) M → Tγ(t) M defined by v0 7→ V |γ(t) with the Levi-Civita connection is a linear isometry.
Proof. Let us rewrite the differential equation in terms of an orthonormal frame

n
X
ẏk (t) + Ak,j (t)yj (t) = 0, k = 1, . . . , n
j=1
setting Ak,j (t) = ni=1 Γkij (γ(t)))γ̇i (t). Since the frame is orthonormal the corresponding symbols
P
satisfy Γkij = −Γjik hence Ak,j (t) = −Aj,k (t) is skew-symmetric. The flow is defined by an orthogonal
γ
matrix, hence P0,t is an isometry.
Remark 10.23. It follows the geometric interpretation of the parallel transport on a 2D surface: a
field v(t) along a geodesic is parallel if and only if kv(t)k is constant and the angle between v(t)
and γ̇(t) is constant. If γ is not a geodesic we can use approximations.
Exercise 10.24. Compute the Christoffel symbols of the Levi Civita connection associated to the
left invariant frame on the hyperbolic plane. Prove that its geodesics are either vertical lines or
semicircles (centered on y = 0).
Remark 10.25. The Levi-Civita connection of the ordinary Rn is the canonical connection ∇ and
acts as
n
X ∂
∇X Y = XY i
∂xi
i=1
the ordinary differentiation along X of the coefficients of Y (corresponds to Γkij = 0)
126
Proposition 10.26. Let (M, g) be an embedded submanifold of Euclidean RN with the induced
metric. Then the Levi-Civita connection ∇ for M is written as
∇X Y = π ⊥ (∇X Y )
where for every x ∈ M we have considered the orthogonal projection
π ⊥ : RN ' Tx RN → Tx M
and ∇ is the canonical connection in Rn .
Proof. First observe that the formula correctly defines a connection on M (check is left to the
reader)
∇X Y = π ⊥ (∇X Y )
To prove that ∇ is torsion free, for every X, Y tangent to M we have
∇X Y − ∇Y X = π ⊥ (∇X Y − ∇Y X) = π ⊥ [X, Y ] = [X, Y ]
since [X, Y ] is also tangent to M . The fact that ∇ is metric: for every X, Y, Z tangent to M ,
extend them to RN and compute
Xg(Y, Z) = Xg(Y, Z) = g(∇X Y, Z) + g(Y, ∇X Z) (10.8)
but we have for X, Y, Z tangent to M , extend them to RN and compute
g(∇X Y, Z) = g(π ⊥ ∇X Y, Z) = g(π ⊥ ∇X Y, Z) (10.9)

= g(π ⊥ ∇X Y, Z) + g((1 − π ⊥ )∇X Y, Z) (10.10)
= g(∇X Y, Z) (10.11)
Corollary 10.27. Let (M, g) a Riemannian manifold which is an hypersurface in RN with the
induced metric. Then a smooth curve γ : [0, T ] → M parametrized by constant speed is a geodesic
on M if and only if γ̈(t) ⊥ Tγ(t) M .
Indeed on hypersurfaces of RN the equation ∇γ̇ γ̇ = 0 means π ⊥ (∇γ̇ γ̇) = 0, but since ∇γ̇ γ̇ = γ̈
this means γ̈ ⊥ Tγ(t) M when we see γ̈ as a vector of RN .
Exercise 10.28. Let f : M → R be a smooth function. Observe that the covariant derivative ∇f
of f with respect to the Levi-Civita connection is the Riemannian gradient. Indeed
(∇f )(X) = ∇X f = Xf.
The Riemannian Hessian of a function f is defined as ∇2 f = ∇(∇f ). We have
∇2 f (Y, X) = ∇X ((∇f )(Y )) = ∇X (Y f ) = X(Y f ) − (∇X Y )f
Notice that the Hessian is symmetric if and only if the torsion is zero
∇2 f (X, Y ) − ∇2 f (Y, X) = Y (Xf ) − (∇Y X)f − X(Y f ) + (∇X Y )f = T (X, Y )f
127
10.3 The Riemann curvature tensor
We first start by observing that the parallel transport (associated with the Levi-Civita connection)
along a closed curve is in general non zero. One can consider for instance a geodesic triangle on
the sphere S 2 .
We introduce the curvature (3,1) tensor R(X, Y ) also called Riemann tensor
R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z. (10.12)
Let us first check that R is indeed a tensor, i.e., the value of R(X, Y )Z at a point depends only on
the value of X, Y, Z at the point itself.
Proposition 10.29. R is skew-symmetric wrt X, Y and C ∞ (M )-linear in every variable.
Proof. The skew-symmetry is immediate from the formula. Next we prove that R is C ∞ (M )-linear.
By skew-symmetry, it is sufficient to prove that R is linear in the first argument, namely that
R(f X, Y )Z = f R(X, Y )Z, where f ∈ C ∞ (M ). (10.13)
Applying the definition of ∇ and the Leibniz rule for the Lie bracket one gets
R(f X, Y ) = ∇f X ∇Y Z − ∇Y ∇f X Z − ∇[f X,Y ] Z

= f ∇X ∇Y Z − ∇Y (f ∇X )Z − ∇f [X,Y ]−(Y f )X
= f ∇X ∇Y Z − (Y f )∇X Z − f ∇Y ∇X Z − f ∇[X,Y ] Z + (Y f )∇X Z
which proves R(f X, Y ) = f R(X, Y ).
Remark 10.30. Another observation proving that R is the (3,1) tensor is that
R(X, Y )Z = (∇2 Z)(Y, X) − (∇2 Z)(X, Y ). (10.14)
where ∇Z is the (1,1) tensor (∇Z)(X) = ∇X Z and ∇2 Z = ∇(∇Z).
Proposition 10.31. Let X1 , . . . , Xn local o.n. frame. We define the coefficients
l
Rijk = hR(Xi , Xj )Xk , Xl i
we have that
l
Rijk = Xi (Γljk ) − Xj (Γlik ) + Γlia Γajk − Γlja Γaik − caij Γlak
In particular if Γkij = 0 for all i, j, k, then R = 0.
Exercise 10.32. Prove a similar formula to the previous one but for the curvature coefficients
hR(∂i , ∂j )∂k , ∂l i in terms of the Christoffel symbols of the coordinate frame.
128
Flat manifolds
Theorem 10.33. The following conditions are equivalent:
(a) R(X, Y )Z = 0 for every X, Y, Z
(b) there exists a local orthonormal frame such that ∇Xi Xj = 0
(c) there exists a local orthonormal frame such that [Xi , Xj ] = 0.
Proof. The only non trivial fact is (a) implies (b). Indeed (b) implies (c) since the torsion free
condition gives
[Xi , Xj ] = ∇Xi Xj − ∇Xj Xi = 0
and (c) implies (a) since [Xi , Xj ] = 0 implies ckij = 0 hence Γkij = 0 for all i, j, k.
To prove (a) implies (b) we do in the case n = 2 (the general case is similar). Do as follows: it
is enough to prove that we can build a frame such that ∇∂i Xj = 0 in a neighborhood. Fix X1 , X2
as ∂1 , ∂2 at q and build the parallel transport on the segment (x1 , 0) by parallel transport along
the x1 axis and then at every (x1 , x2 ) by parallel transport along the second.
We have clearly ∇∂1 Xj |(x1 ,0) = 0 and ∇∂2 Xj |(x1 ,x2 ) = 0 for j = 1, 2. The second means
∇∂2 Xj = 0 locally. We do not know if ∇∂1 Xj |(x1 ,x2 ) is zero if x2 6= 0.
But since ∂1 , ∂2 is coordinate frame and ∇∂1 ∇∂2 = ∇∂2 ∇∂1 . Then
∇∂2 ∇∂1 Xj |(x1 ,x2 ) = ∇∂1 ∇∂2 Xj |(x1 ,x2 ) = ∇∂1 0 = 0
Hence ∇∂1 Xj |(x1 ,x2 ) is the parallel transport along x2 axis of ∇∂1 Xj |(x1 ,0) , which is zero.
With this charachterization we have immediately
Corollary 10.34. (M, g) admits a local orthonormal frame such that [Xi , Xj ] = 0, if and only if
it is locally isometric to Euclidean Rn .
It is enough to prove that if there exists a local orthonormal frame such that [Xi , Xj ] = 0 then
we can build the local isometry.
If [Xi , Xj ] = 0 then consider the map
which satisfies since [Xi , Xj ] = 0 (cf. Chapter ?)
∂
ψ∗ = Xi
∂ti
hence ψ is a local isometry.
129
Some curvature identities
The following identity, first discovered by G. Ricci-Curbastro, is known as the first Bianchi identity.
Proposition 10.35 (first Bianchi identity). For every X, Y, Z ∈ Vec(M ) the following identity
holds
R(X, Y )Z + R(Y, Z)X + R(Z, X)Y = 0. (10.15)
Proof. We will show that (10.15) is a consequence of the Jacobi identity for vector fields (5.12).
Using the fact that ∇ is a torsion-free connection we can write
[X, [Y, Z]] = ∇X [Y, Z] − ∇[Y,Z] X

= ∇X ∇Y Z − ∇X ∇Z Y − ∇[Y,Z] X,
[Z, [X, Y ]] = ∇Z ∇X Y − ∇Z ∇Y X − ∇[X,Y ] Z,
[Y, [Z, X]] = ∇Y ∇Z X − ∇Y ∇X Z − ∇[Z,X] Y,
Then, adding these identities and using (5.12), one gets
0 = [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]]

= ∇X ∇Y Z − ∇X ∇Z Y − ∇[Y,Z] X
+ ∇Z ∇X Y − ∇Z ∇Y X − ∇[X,Y ] Z
+ ∇Y ∇Z X − ∇Y ∇X Z − ∇[Z,X] Y
= R(X, Y )Z + R(Y, Z)X + R(Z, X)Y.
Exercise 10.36 (second Bianchi identity). Prove that for every X, Y, Z, W ∈ Vec(M ) one has
(∇X R)(Y, Z, W ) + (∇Y R)(Z, X, W ) + (∇Z R)(X, Y, W ) = 0.
(Hint: Expand the identity ∇[X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y ]] W = 0.)

Remark 10.37. The relations for the Christoffel symbols implies the following skew-symmetry prop-
erty: for X, Y, Z, W ∈ Vec(M )
hR(X, Y )Z | W i = − hR(X, Y )W | Zi ,
where h· | ·i denotes the Riemannian inner product.

Let us introduce the notation
R(X, Y, Z, W ) := hR(X, Y )Z | W i .
Then, the first Bianchi identity (10.15) can be rewritten as follows: for X, Y, Z, W ∈ Vec(M ) one
has
R(X, Y, Z, W ) + R(Z, X, Y, W ) + R(Y, Z, X, W ) = 0. (10.16)
Moreover, the skew-symmetry properties of the curvature tensor discussed in Proposition 10.29 and
Remark 10.37 can be rewritten as follows
R(X, Y, Z, W ) = −R(Y, X, Z, W ), R(X, Y, Z, W ) = −R(X, Y, W, Z). (10.17)
130
Proposition 10.38. For every X, Y, Z, W ∈ Vec(M ) we have R(X, Y, Z, W ) = R(Z, W, X, Y ).
Proof. Using (10.16) four times we can write the identities
R(X, Y, Z, W ) + R(Z, X, Y, W ) + R(Y, Z, X, W ) = 0,

R(Y, Z, W, X) + R(W, Y, Z, X) + R(Z, W, Y, X) = 0,
R(Z, W, X, Y ) + R(X, Z, W, Y ) + R(W, X, Z, Y ) = 0,
R(W, X, Y, Z) + R(Y, W, X, Z) + R(X, Y, W, Z) = 0.
Summing these identities and using (10.17), one gets R(X, Z, W, Y ) = R(W, Y, X, Z).
Proposition 10.39. Assume that R(X, Y, X, W ) = 0 for every X, Y, W ∈ Vec(M ). Then
R(X, Y, Z, W ) = 0 ∀ X, Y, Z, W ∈ Vec(M ).
Proof. By assumptions and the skew-symmetry properties (10.17) of the Riemann tensor we have
that R(X, Y, Z, W ) = 0 whenever any two of the vector fields coincide. In particular
0 = R(X, Y + W, Z, Y + W ) = R(X, Y, Z, W ) + R(X, W, Z, Y ). (10.18)
Notice that the two extra terms that should appear developing the left hand side vanish, by as-
sumptions. Then (10.18) can be rewritten as
R(X, Y, Z, W ) = R(Z, X, Y, W ).
This means that the quantity R(X, Y, Z, W ) is invariant by cyclic permutations of X, Y, Z. But the
cyclic sum of these terms is zero thanks to (10.16), hence R(X, Y, Z, W ) = 0.
From the properties of the Riemann curvature one obtains the following.
Corollary 10.40. There is a well defined map
R : ∧2 Tq M → ∧2 Tq M, R(X ∧ Y ) := R(X, Y ).
Moreover R is self-adjoint with respect to the scalar product on ∧2 Tq M induced by the Riemannian
scalar product, namely
R(X ∧ Y ) | Z ∧ W = X ∧ Y | R(Z ∧ W ) .
More curvatures and comparison

We define the sectional curvature
R(X, Y, X, Y )
Sec(X, Y ) =
kXk2 kY k2 − hX | Y i
which indeed depends only on the plane Π = span{X, Y }.

Remark 10.41. Indeed one can prove that Sec completely determines the Riemann tensor R.
131
We define the Ricci curvature which is a quadratic form computing an average of sectional
curvatures containing a fixed vector
n
X
Ric(V ) = Sec(V, Xi )
i=1
where X1 , . . . , Xn is an orthonormal basis.

This opens the big theory of comparison geometry, for instance let us state one single result.
Theorem 10.42 (Bonnet-Myers theorem, 1941). Let (M, g) be a complete Riemannian manifold
of dimension n whose Ricci curvature satisfies Ric(V ) ≥ (n − 1)KkV k2 for some K > 0. Then M
is compact and √
diam(M ) ≤ π/ k.
Bonnet proved the version with inequality on all sectional curvatures. Myers weaken the result
at Ricci curvatures.
In the previous theorem the equality case is attained for instance in the case of the sphere S n (r)
of radius r = 1/K. A striking result is the following rigidity result.
Theorem 10.43 (Cheng, 1975). Let (M, g) be a complete Riemannian manifold of dimension √n
whose Ricci curvature satisfies Ric(V ) ≥ (n − 1)KkV k2 for some K > 0. If diam(M ) = π/ k,
then M is isometric to the sphere S n (r) of radius r = 1/K.
132
Appendix A
Problems and Exercises
The following exercises are taken from exams from Academic Year 2020/21 and 2021/22. Solutions
are available on the MOODLE page of the course.
Each exam is composed by three exercises. There are 5 exams per year. So Exercise A.1–A.3
is Exam 1 of 2020/21,. . . , Exercise A.13–A.15 is Exam 5 of 2020/21, Exercise A.16–A.18 is Exam
1 of 2021/22, and so on.
Try to do the exercise before looking at the solution
Exercise A.1. Consider the two subsets of R3 defined by
H = {(x, y, z) ∈ R3 | x2 + y 2 − z 2 = 1}, E = {(x, y, z) ∈ R3 | x2 + y 2 + 3z 2 = 3}.
0. Prove that E and H are C ∞ submanifolds of R3
1. Is E ∩ H a smooth submanifold of R3 ? Is E ∩ H compact? Is E ∩ H connected?
Let now g : R3 → R defined by g(x, y, z) = x2 + y 2 + 3z 2 − 3 and let (a, b, c) ∈ R3 such that

g(a, b, c) > 0. (Notice that E = g −1 (0)).
2. Let C the subset of points p = (x, y, z) of E such that the affine tangent hyperplane Tp E
passes through (a, b, c). Prove that C is a smooth submanifold of R3 .
Exercise A.2. Let ω be the differential 1-form in R3
ω = dz + a(x, y)dx − b(x, y)dy
where a, b : R2 → R be smooth functions depending only on x, y. Consider the vector distribution

D = ker ω (i.e., Dq = ker ωq for every q = (x, y, z))
1. Find two everywhere linearly independent smooth vector fields such that D = span{X, Y }.
2. Compute [X, Y ] and give a necessary and sufficient condition (C1) on the functions a, b such
that D is integrable.
133
3. Compute the 2-form dω and give a necessary and sufficient condition (C2) on the functions
a, b such that ω ∧ dω is a volume form in R3 . What is the relation between (C1) and (C2)?
Exercise A.3. Let g denote the Euclidean metric in R3 . Let F : R+ ×]0, π[×]0, 2π[→ R3
F (ρ, θ, ϕ) = (ρ sin θ cos ϕ, ρ sin θ sin ϕ, ρ cos θ)
be the map defining spherical coordinates in R3 .
1. Compute the tensor F ∗ g, then deduce that the standard Riemannian metric gS 2 on the sphere
S 2 (which is the restriction of g to S 2 ) is written in the coordinates (θ, ϕ)
gS 2 = dθ2 + sin2 θdϕ2
2. compute the length of meridians (curves of the form ϕ = ϕ0 ) and parallels (curves of the form
θ = θ0 ) with respect to the Riemannian metric gS 2 .
3. compute the Riemannian volume form volg on S 2 in the coordinates (θ, ϕ) and deduce the
volume A of the sphere with respect to volg
Z
A= dvolg
S2
.
Exercise A.4. The sphere S 2 ⊂ R3 described as
S 2 = {x = (x1 , x2 , x3 ) ∈ R3 | x21 + x22 + x23 = 1}
can be endowed by the atlas A = {(UN , ϕN ), (US , ϕS )} where UP = S 2 \ {P } for P ∈ {N, S}

the north and south pole N = (0, 0, 1), S = (0, 0, −1), and ϕP : UP → R2 the corresponding
stereographic projection from P on the plane {x3 = 0}.
1. Find the explicit expression for ϕN and ϕS and check the smooth compatibility between
charts. Is it a maximal atlas?
2. Consider the map F : C → C, F (z) = z 2 + 1 thought as a smooth map R2 → R2 . Consider

the function F : S 2 → S 2 defined by
(
ϕ−1
N ◦ F ◦ ϕN (x), x 6= N
F (x) =
N, x=N
Is the function F smooth with respect to the C ∞ structure ?
Hint: it may be useful to write ϕN ◦ ϕ−1

S in complex coordinates.
134
Exercise A.5. Let ω be a differential k-form in U = Rn \ {0}. For t > 0, let Ht : U → U be the
map Ht (x) = tx. For p ∈ N, we say that ω is p-homogeneous if Ht∗ ω = tp ω.
1. Express the p-homogeneity of ω in terms of its coefficients in coordinates

X
ω= ωi1 ...ik dxi1 ∧ . . . ∧ dxik .
1≤i1 <...<ik ≤n
Pn ∂
2. Let X be the vector field X = i=1 xi ∂xi . Prove that if ω is p-homogeneous then LX ω = pω.
Exercise A.6. Consider the (2, 0) tensor in R3 defined by τ = dx ⊗ dx + dy ⊗ dy − dz ⊗ dz.

0. Define what is a Riemannian metric. Is τ a Riemannian metric on R3 ?
1. Let i : H ,→ R3 be the canonical inclusion of the surface
H = {(x, y, z) ∈ R3 | x2 + y 2 − z 2 = −1, z > 0} ⊂ R3 .
Prove that i∗ τ defines a Riemannian metric g on H
2. Compute the Riemannian length (with respect √ of curve defined by H ∩ {x =

√ to g) of the piece
0} and joining the two points P = (0, −1, 2) and Q = (0, 1, 2).
Exercise A.7. Denote S 2 the unit sphere of R3 and P2 (R) the real projective plane. Let F be the
map
F : R3 → R3 , F (x, y, z) = (2xz, 2yz, 1 − 2z 2 )
1. Show that F restricts to a map f from S 2 to S 2 . Is f of class C ∞ ?
2. Compute the linear map f∗ : Tp S 2 → Tf (p) S 2 for p = (1, 0, 0) ∈ S 2 (you might fix a basis of
your choice in the tangent spaces).
3. Is f a local diffeomorphism? If not, find all critical points and critical values of f .
4. Denote the canonical projection π : S 2 → P2 (R). Show that there exists a unique map
h : P2 (R) → S 2 of class C ∞ such that f = h ◦ π
Exercise A.8. Let X, Y be two vector fields on a smooth manifold M . Consider the following two
operators acting on differential forms on M
• LX the Lie derivative of a differential form with respect to a vector field X,
• iY the inner product of a differential form with respect to a vector field Y .

Prove the following identity on differential forms
LX ◦ iY − iY ◦ LX = i[X,Y ] (A.1)
Hint: start by proving the required identity for a 1-differential form ω
135
Exercise A.9. Let (M, g) be a Riemannian manifold.
1. Recall the definition of Levi-Civita connection ∇ defined on M , and then the explicit formula
for ∇X Y , where ∇ is the Levi-Civita connection in the Euclidean space (Rn , g).
2. In the Euclidean space R2 , compute ∇X X where X = x∂y − y∂x .
3. Prove that ∇Y Y = 0, where Y = X|S 1 and ∇ is the Levi Civita connection on S 1 considered
with the induced Riemannian metric of R2 .
Exercise A.10. Let In the identity matrix of size n and

0 In
J=
−In 0
The set Sp(2n) of symplectic matrices is the subset of square matrices of size 2n
Sp(2n) := {M ∈ M2n (R) : M T JM = J} ⊂ M2n (R).
1. For n = 1, describe explicitly Sp(2) and show it is a submanifold of M2 (R) ' R4 . Of which
dimension?
2. Prove that Sp(2n) is a submanifold of M2n (R) and compute its dimension.
3. Compute TI Sp(2n), the tangent space to Sp(2n) at identity.
Hint: consider the map F : M2n (R) → X, F (M ) = M T JM , for a suitable space X such that. . .
Exercise A.11. For which values of α ∈ R the following 2-form Ω in R3 \ {0} is closed?
Ω = (x2 + y 2 + z 2 )α (xdy ∧ dz + ydz ∧ dx + zdx ∧ dy)
Exercise A.12. Consider on R2 the operation
(x1 , x2 ) · (y1 , y2 ) = (x1 + y1 , y2 + x2 ey1 )
1. Prove that (R2 , ·) is a Lie group and find the identity e of the group (i.e., the neutral element)
2. Find the two left-invariant vector fields X1 , X2 satisfying Xi (e) = ∂xi for i = 1, 2
3. Compute X3 := [X1 , X2 ]. Is X3 left-invariant?
4. Compute the Riemannian metric g for which X1 , X2 is a global orthormal frame
5. Compute the Riemannian volume associated with g of the unit square Q = [0, 1] × [0, 1]
136
Exercise A.13. Consider the map F : C2 → C × R given by F (u, v) = (2uv, |u|2 − |v|2 ). Consider
S 3 as a subset of C2 ' R4 and similarly S 2 as a subset of C × R ' R3 .
1. Prove that π := F |S 3 is a well defined map from S 3 to S 2 .
2. Prove that π is C ∞ with respect to the smooth structures of S 3 and S 2 .
3. Prove that π −1 (x) is a submanifold of S 3 for every x ∈ S 2 . Of which dimension?
Exercise A.14. Consider the sphere S 2 embedded in R3 . Let X be the vector field
∂ ∂ ∂
X=x +y +z
∂x ∂y ∂z
1. Show that the restriction of the 2-form iX (dx ∧ dy ∧ dz) to the sphere is a volume form on
S 2 . We denote this form α.
2. Show that on the complement of the equator {z = 0} ∩ S 2 , we have
dx ∧ dy
α=
z
3. Let ϕN be the stereographic projection from the north pole N = (0, 0, 1) from S 2 \ {N } to
the plane {z = 0}. Write out ϕ−1 −1 ∗
N and calculate ω := (ϕN ) α.
R
4. Compute R2 ω
Exercise A.15. Let (M, g) be an orientable two dimensional Riemannian manifold. Let X1 , X2
be an orthonormal basis for a metric g on M . Assume that for α1 , α2 ∈ C ∞ (M ) we have on M
[X1 , X2 ] = α1 X1 + α2 X2 , (A.2)
1. Recall the definition of Riemannian volume. Prove that the Riemannian volume form can be
written as volg = ν1 ∧ ν2 where ν1 , ν2 are dual basis of X1 , X2 .
2. Recall the definition of divergence of a vector field with respect to the Riemannian volume.
Compute the divergence div(X) and div(Y ) with respect to volg in terms of α1 , α2 .
Exercise A.16. Let f : Rn → R be a C ∞ function and let M be its graph
M = {(x, f (x)) ∈ Rn+1 = Rn × R | x ∈ Rn } ⊂ Rn+1
1. Show that M is a smooth orientable manifold specifying an atlas and the dimension.
2. Is M an immersed/embedded submanifold of Rn+1 ?
3. Let i : M → Rn+1 be the canonical inclusion. Compute g := i∗ g the induced Riemannian

metric on M , where g is the Euclidean metric in Rn+1 .
137
4. Compute ΩM the Riemannian volume form of M
5. Compute U ΩM for the special case n = 2, f (x) = kxk2 and U = {(x, f (x)) ∈ M | kxk2 ≤ 1}.
R
Hint: one might use the formula det(I + vv T ) = 1 + kvk2 , for v ∈ Rn seen as a column vector,
where we denote I the identity matrix and v T the transpose of v.
Exercise A.17. Consider the 2-sphere S 2 ⊂ R3 endowed by the standard atlas1 A = {(UN , ϕN ), (US , ϕS )}.
1. for P ∈ S 2 describe TP S 2 as a subset of TP R3 and compute the linear map (ϕN )∗ : TP S 2 → R2
2. Compute explicitly Xi = (ϕ−1 2

N )∗ Yi defined on UN = S \ {N }, where Yi = ∂ui for i = 1, 2 is
2
the constant vector field on R , with coordinates (u1 , u2 ).
Hint: Solve the linear equation (ϕN )∗ Xi = Yi with unknown Xi tangent to S 2 .
3. Prove that Xi can be continuously extended to a C ∞ vector field X i on S 2
4. Compute the Lie bracket [X 1 , X 2 ].
Exercise A.18. Consider the set AL(R2 ) of affine lines in R2 . Given a line ` of equation ax+by = c
we define the sets U1 = {` ∈ AL(R2 ) | a 6= 0}, U2 = {` ∈ AL(R2 ) | b 6= 0} and the charts
ϕi : Ui → R2 defined for i = 1, 2 by

b c a c
ϕ1 (`) = , , ϕ2 (`) = ,
a a b b
The atlas {(Ui , ϕi )}i=1,2 gives the structure of smooth manifold to AL(R2 ).
1. Let o be the origin of R2 , for every affine line ` ∈ AL(R2 ) define f (`) := dist2 (o, `), where
dist denotes the Euclidean distance in R2 from a point to a line. Prove that the function f is
C ∞ with respect to the smooth structure of AL(R2 ) .
2. Find all critical points of f given at point 2. Discuss their nature.
Exercise A.19. Consider F : Rn \ {0} → Rn given by F (x) = x/kxk.
1. For x0 6= 0 and v ∈ Rn compute DF (x0 )v. Determine the rank of F and ker DF (x0 ).
Let now g : Rn → R of class C ∞ and M = {x ∈ Rn | g(x) = 0}. Assume that
(a) h∇g(x), xi =
6 0 for every x ∈ M , where h·, ·i is the Euclidean inner product.
(b) for all y ∈ Rn \ {0} there exists unique r > 0 such that ry ∈ M .
2. Prove that M is a C ∞ embedded submanifold of Rn \ {0}. Of which dimension?

1
this is A = {(UN , ϕN ), (US , ϕS )}, UP = S 2 \ {P }, N = (0, 0, 1), S = (0, 0, −1) where ϕP : UP → R2 is the
corresponding stereographic projection from P onto {x3 = 0} ' R2 .
138
3. Prove that the restriction of F to M , regarded as a map G := F |M : M → S n−1 , is an
injective immersion. Is G a local diffeomorphism?
Exercise A.20. Consider in R3 the differential 2-form ω = x dy ∧ dz − y dx ∧ dz + z dx ∧ dy and

∂ ∂ ∂
the vector field X = x ∂x + y ∂y + z ∂z .
1. Does there exist a differential 1-form ν in R3 such that ω = dν?
2. Compute LX ω and iX (dω), where LX denotes the Lie derivative and iX the interior product.
3. Compute the integral E LX ω, where E = {(x, y, z) ∈ R3 | x2 + y 2 + 4z 2 = 4}.

R
Exercise A.21. Consider the subsets of R3 given by
M = {(x, y, z) ∈ R3 | x2 + y 2 + 1 = z 2 , z > 0}, N = {(x, y, z) ∈ R3 | x + y + 4 = 4z},
1. Is M a smooth manifold? Is M ∩N an embedded smooth manifold of R3 ? Of which dimension?
2. Let C = M ∩ N . Compute the tangent spaces Tq M and Tq C for q = (0, 0, 1).
3. Find two vector fields X, Y tangent to M such that π∗ X = ∂x and π∗ Y = ∂y where π : R3 →

R2 is the projection onto the first two coordinates. Compute their Lie bracket [X, Y ].
4. Let g be the Riemannian metric induced on M by the restriction of the Euclidean metric g
in R3 . Compute ∇X Y and ∇Y X, where ∇ is the Levi Civita connection of (M, g).
Exercise A.22. Let us denote by P1 (R) the real projective line (points on P1 (R) are denoted
[x : y])
1. Recall the standard differential structure (i.e., the atlas) on P1 (R). Is P1 (R) orientable?
2. Establish an explicit smooth diffeomorphism between P1 (R) and S 1 .
3. Show that the projection π : R2 \ {0} → P1 (R) defined by π(x, y) = [x : y] is of class C ∞ .
4. Let P (x), Q(x) be two real polynomials with no real root in common. Prove that the following
map is smooth ( P (x)
[ : 1] if Q(x) 6= 0
F : R → P1 (R), F (x) = Q(x)
[1 : 0] if Q(x) = 0
Give an example, for some choice of non constant P (x) and Q(x), such that F is an immersion
and an example when F is not an immersion.
5. Show that G : P1 (R) \ {[1 : 0]} → P1 (R) defined by
G([x : 1]) = F (x)
admits a continuous extension to G : P1 (R) → P1 (R). Prove that G is smooth. Give an

example, for some choice of non constant P (x) and Q(x), such that G is an diffeomorphism.
139
Exercise A.23. Let f : R → R be smooth and consider the 1-differential form in R3 where we
denote points (x, y, z)

ω = 1 − f (x) dy − 1 + f (x) dz.
1.a. Find necessary and sufficient conditions (V) on f under which ω ∧ dω is a volume form on
R3 .
1.b. Assume conditions (V) holds. Find2 a smooth vector field Z such that iZ ω = 1 and iZ dω = 0.
1.c. For the vector field Z computed in 1.b., then compute LZ ω.
2. Find necessary and sufficient conditions (C) on f in such a way that the distribution D = ker ω
is involutive. Assuming (C), find the integral manifold through the origin.
Exercise A.24. Consider f : R →]0, +∞[ smooth positive and the subset of R3 given by
M = {(x, y, z) ∈ R3 | x2 + y 2 = f 2 (z)}
1. Prove that M is an embedded smooth manifold. Of which dimension?
2. Let i : M → R3 be the canonical inclusion. Compute g = i∗ g the restriction of the Euclidean

metric g in R3 to M .
3. Compute the length (with respect to the Riemannian metric g) of the curve γc given by the
intersection of M and the plane {z = c}.
4. For which c ∈ R the curve γc is a geodesic?
Exercise A.25. Consider the map F : R4 → R defined by
F (x, y, z, t) = x2 + y 2 + z 2 − t2
1. Prove that M := F −1 (1) is a regular submanifold of R4 .
2. Let π : R4 → R3 be the map π(x, y, z, t) = (x, y, z). Is π|M : M → R3 a submersion?
3. Prove that M is diffeomorphic to the product S 2 × R.
4. Does there exists a smooth vector field on M which is never vanishing?

(If yes, show an example. If not, justify your statement.)
2
here iX η (resp. LX η) denotes the interior product (resp. the Lie derivative) of a differential k-form η with
respect a vector field X.
140
Exercise A.26. Let Ω = {(x, y, z) ∈ R3 | x2 + y 2 6= 0} ⊂ R3 and ω the 1-form on Ω defined by
x y
ω= dy − 2 dx + zdz
x2 +y 2 x + y2
Moreover let i : R2 \ {(0, 0)} → Ω defined by

u v 2 2
i(u, v) = √ ,√ ,u + v
u2 + v 2 u2 + v 2
1. Prove that ω is closed
2. For I = [0, 2π] and γ : I → R2 given by γ(t) = (cos t, sin t) compute ◦ γ)∗ ω
R
I (i
3. Show that i∗ ω is not exact
Exercise A.27. Let M = {(x, y) ∈ R2 | x > 0} be endowed with the Riemannian metric g such
that the two vector fields X1 = ∂x and X2 = x∂y define an orthonormal basis for g.
1. Compute the Riemannian metric g and the corresponding Riemannian volume form volg .
2. Compute the length of the following curves for x0 > 0, y0 ∈ R and 0 < a < b
γ1 : [a, b] → R, γ1 (t) = (x0 , t),
γ2 : [a, b] → R, γ2 (t) = (t, y0 ),
3. Recall the definition of divergence of a vector field X with respect to the Riemannian volume
volg . Compute the divergence div(X1 ) and the divergence div(X2 ) .
4. Is γ2 a length-minimizer for the Riemannian metric g ?
Exercise A.28. Consider the map F : R3 → R2 defined by
F (x, y, z) = (x + y 2 , x + 2y 2 + z 2 )
1. Prove that M := F −1 (0, 1) is an embedded submanifold of R3 .
2. Prove that M is diffeomorphic to S 1 .
3. Let π : R3 → R2 be the map π(x, y, z) = (x, y). Is π|M : M → R2 an immersion?
4. Let g : M → R be defined as g(x, y, z) = x for every (x, y, z) ∈ M . Find critical points of g.
Exercise A.29. Consider the following vector fields in R3
X = ∂x − y∂z , Y = ∂y + x∂z , Z = ∂z
141
1. For i = 1, 2 discuss whether Di is a flat distribution.
D1 = span{X, Y }, D2 = span{X, Z}.
2. For flat distributions, find an integral manifold through a generic point (x0 , y0 , z0 ) ∈ R3 .
3. Find a non zero differential 1-form ω such that ω(X) = ω(Y ) = 0. Compute LZ ω and iZ dω.
Exercise A.30. Consider the 2-sphere S 2 ⊂ R3 endowed by the standard3 atlas A = {(UN , ϕN ), (US , ϕS )}
and the Riemannian metric gS 2 induced by the Euclidean metric g of R3 . Consider the inverse of
−1
the chart ϕN : R2 → S 2 \ {N }. After writing explicitly ϕN and ϕ−1
N
1. Compute the Riemannian metric g := (ϕ−1 ∗

N ) gS 2 defined on R
2
2. Compute the corresponding Riemannian volume volg associated with (R2 , g). Verify that
du ∧ dv
volg = 4
(1 + u2 + v 2 )2
R
3. Compute the integral R2 volg . Comment your result.
Exercise A.31. Consider S 1 ⊂ R2 and the map F : S 1 ×S 1 → R2 defined for every (x, y) ∈ S 1 ×S 1
F (x, y) = x + y
1. Prove that S 1 × S 1 is a smooth oriented manifold of dimension 2.
2. Show that F is a smooth map.
3. Is F a submersion? If not, find all critical points of F .
4. Is M = F −1 (S 1 ) a smooth manifold of S 1 × S 1 ? Of which dimension? Describe M .
Exercise A.32. Consider the differential one form in M = S 1 × R2 with coordinates (θ, x, y)
ω = cos θ dx + sin θ dy
1. Find a basis X1 , X2 of vector fields of the distribution D := ker ω. Is the distribution D flat?
2. Prove that ω ∧ dω is a volume form on M .
3. Show that there exists a unique vector field X0 such that4 iX0 ω = 1 and iX0 (dω) = 0.
4. For every t ∈ R, compute the flow etX0 of X0 , and show that (etX0 )∗ ω = ω .
3
here UP = S 2 \ {P } for P ∈ {N, S} the north and south pole N = (0, 0, 1), S = (0, 0, −1), and ϕP : UP → R2
the corresponding stereographic projection from P on the plane {x3 = 0} ' R2 .
4
here iX denotes the interior product with respect to a vector field X
142
Exercise A.33. Let us consider the unit open disk D = {(x, y) ∈ R2 | x2 + y 2 < 1} in the plane
R2 with the Riemannian metric
4
g= (dx2 + dy 2 )
(1 − x2 − y 2 )2
1. Find an orthonormal frame X1 , X2 for the Riemannian metric g on D.

R
2. Compute the Riemannian volume form Ω := volg and then the Riemannian area D Ω of D.
√
3. Prove that M = {(u, v, w) ∈ R3 | w = u2 + v 2 + 1} ⊂ R3 is a smooth manifold and given

3 2 u v
F : R \ {w = −1} → R , F (u, v, w) = ,
1+w 1+w
prove that Φ := F |M is a diffeomorphism between M and D. (Hint: write Φ−1 )
4. Compute the two form F ∗ Ω on R3 \ {w = −1}. Is Φ∗ Ω a volume form on M ?
Exercise A.34. In the real projective plane P2 (R) consider the subset M given by
M = {[x0 , x1 , x2 ] ∈ P2 (R) | x20 + x21 − x22 = 0}.
1. Recall the atlas defining the differentiable structure for the real projective plane P2 (R).
2. Is M an embedded submanifold of P2 (R)? Of which dimension? Is M connected?
3. Prove that the following function F : M → R is (well-defined) and smooth
x20 − x21 − x22

F ([x0 , x1 , x2 ]) =
x20 + x21 + x22
4. Is F a submersion? If not find all critical points of F .
Exercise A.35. Let D be the distribution in R3 spanned by the vector fields X and Y
∂ 2xz ∂ ∂
X= + , Y =
∂x 1 + x2 ∂z ∂y
1. Prove that D is flat and find a 1-form ω such that D = ker ω.
2. Compute explicitly the flow etX of the vector field X. Is it true that etX
∗ Y =Y?
3. Find an integral manifold S1 for D passing through (0, 0, 0) and give equation(s) for S1 .
4. Find an integral manifold S2 for D passing through (1, 1, 1) and give equation(s) for S2
143
Exercise A.36. Let us consider the unit sphere S 2 and its standard atlas (UN , ϕN ), (US , ϕS ) given
by the stereographic projections5
1. Write explicitly the chart ϕN : UN → R2
2. Compute the Riemannian metric on UN given by g := (ϕN )∗ g where g = dx2 + dy 2 is the

Euclidean metric on R2 .
3. Prove that there exists ψ ∈ C ∞ (UN ) such that g(v, w) = ψhv, wiR3 , where hv, wiR3 is the
inner product of R3 . Compute ψ. Can ψ be extended to a smooth function on S 2 ?
4. Compute the length with respect to g of parallels of S 2 \ {N } (i.e., intersections of S 2 with

the plane z = z0 , for −1 ≤ z0 < 1) .
5
recall that UP = S 2 \ {P } for P = N, S the north and south poles N = (0, 0, 1), S = (0, 0, −1)
144
Bibliography
[Lee13] John M. Lee. Introduction to smooth manifolds, volume 218 of Graduate Texts in Mathe-
matics. Springer, New York, second edition, 2013.
145

Barilari GeoDiff

Uploaded by

Copyright:

Available Formats

Barilari GeoDiff

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Barilari GeoDiff

Uploaded by

Copyright:

Available Formats

Lecture notes on

1 Smooth manifolds and smooth maps 7

2 Tangent space and differentials 25

3 Immersions, embeddings. Submanifolds 33

4 Tangent bundle and vector fields 43

5 Integral curves and flows 51

6 Vector distributions: integrability vs non-integrability 61

7 Tensors and Differential forms 71

8 Orientation, Integration on manifolds 89

9 Riemannian manifolds 109

10 Connections, parallel transport and curvature 121

A Problems and Exercises 133

La géométrie n’est pas vraie, elle est avantageuse.1

Padova, Academic years 20/21-21/22-22/23

Smooth manifolds and smooth maps

La notion générale de variété est assez difficile à définir avec précision.1

1.1 Topological and smooth manifolds

A chart (U, ϕ) gives local coordinates to points of U . Namely, for q ∈ U ⊂ M we assign n

where ηij := ϕj ◦ ϕ−1

2. ϕα (Uα ∩ Uβ ) and ϕβ (Uα ∩ Uβ ) are open in Rn

4. M is covered by countably many Uα

5. if x 6= y either they belong to the same Uα or to two disjoint ones

Proof. Consider the topology generated by ϕ−1 n

1.2 Some fundamental examples

• Rn is a n-dimensional manifold. Every open set U ⊂ Rn is a n-dim manifold. Union of

• if M is m-dimensional and N is n-dimensionale then M × N is (m + n)-dim manifold.

and similar results hold for ϕ−1

Notice that with the above identifications ϕN (UN ∩ US ) = ϕS (UN ∩ US ) = Rn \ {0}.

Exercise 1.12. Consider the function f : S 2 → R given by f (x, y, z) = z if (x, y, z) ∈ S 2 . Clearly

Exercise 1.13. Consider the atlas A0 = {(Ui± , ϕ±

Ui+ = S n ∩ {xi > 0}, Ui− = S n ∩ {xi < 0}

Pn (R) = (Rn+1 \ {0})/ ∼

with the equivalence relation x ∼ y in Rn+1 \ {0} if x = λy for some λ 6= 0. Denoting

x = (x0 , . . . , xn ) ∈ Rn+1 \ {0}

The set of affine lines in the plane

Ua = {` ∈ AL(R2 ) | a 6= 0}, Ub = {` ∈ AL(R2 ) | b 6= 0},

and the charts ϕi : Ui → R2 defined for i = a, b by

It is not difficult to check that if (u, v) denote the coordinate on R2 then

and on this set  

Indeed if the line is neither horizontal nor vertical we have

Proof. It is not restrictive to assume y0 = 0. Fix x0 ∈ U ∩ F −1 (0). Since rank(DF (x0 )) = m up

Then defining ψ : U 0 ⊂ Rn−m → Rn as ψ(x0 ) = (x0 , f (x0 )) we have proved that

We will see how this generalizes to manifolds later.

Exercise 1.19. Discuss whether the following subset of R3 is a smooth manifold

Of course Gk (V ) is covered by such open sets

Lemma 1.20. Every W ∈ U t is the graph of a unique linear map AW : Z → U .

Choosing basis for Z and U we can easily build a bijective map

ϕU : U t → Mk,n−k (R) ' Rk(n−k) , ϕU (W ) = AW .

Notice that ϕU (Z) = 0 (i.e., the Z we have chosen is the origin).

Examples in the space of matrices

• the space SL(n) is a smooth manifold of dimension n2 − 1. Indeed we have

SL(n) = {A ∈ GLn (R) | det(A) = 1} = det−1 (1).

DF (I)H = trace(H) (1.3)

det(A + H) = det(A) + det(A)trace(A−1 H) + o(kHk)

Which in turns gives for A ∈ SLn (R)

On (R) = {A ∈ Mn (R) | AT A = I} = F −1 (I)

1.3 Smooth maps

Fb := ψ ◦ F ◦ φ−1 : ϕ(U ) ⊂ Rn → ψ(V ) ⊂ Rn .

and on this set