Riemannian Geometry

Romaric Kana Nguedia ([email protected])

African Institute for Mathematical Sciences (AIMS)
Cameroon

Supervised by: Nicolau Sarquis Aiex (University of Auckland, New Zealand)

and Diletta Martinelli (Universiteit Van Amsterdam, Netherlands )

07 May 2020
Submitted in Partial Fulfillment of a Structured Masters Degree at AIMS-Cameroon
Abstract
In this work we are going to provide some basic concepts in Riemannian Geometry, we will see how
to define a metric on a differentiable manifold, that we call a Riemannian metric, and thus define a
structure of a Riemannian manifold. Having done so, we introduce the notion of covariant derivative,
to talk about one of the fundamental theorems of Riemannian geometry that shows the existence of
a Levi-Civita connection which is useful to develop two fundamental notions of Riemannian Geometry:
geodesics and curvature. We will present numerous examples.

Declaration

I, the undersigned, hereby declare that the work contained in this essay is my original work, and that
any work done by others or by myself previously has been acknowledged and referenced accordingly.

i
Contents

Abstract i

1 Introduction 1

2 Preliminary Results 2
2.1 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Differentiable manifolds and tangent space . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Riemannian manifolds and Levi-Civita connection 8

3.1 Riemannian metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Connection and covariant derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.3 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 Curvature in Riemannian manifold 21

4.1 Curvature tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Sectional curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3 Ricci curvature and scalar curvature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5 Conclusion 31

A Some additional data 32

Acknowledgements 37

References 38

ii
1. Introduction
Introduced by Bernhard Riemann around 1854 [7], Riemannian geometry is a branch of differential
geometry. Riemann developed the fundamental notions of geometric shape and curvature in order to
generalize the traditional geometry that was limited to the Euclidean space of dimension 3. This field un-
derwent several evolutions due to the hard work of mathematicians like Greogorio Ricci-Curbastro [9],
Tullio Levi-Civita [1] etc. and in the mid-1930’s this research was concluded by the establishment of
the Whitney plunge Theorem in 1936, that allowed to define a formal map of the Riemannian geometry.
The basic objects of Riemannian geometry are gradually finding many applications also outside of
Riemannian geometry itself, for instance, in the study of metric spaces. Gromov for example, defines
notions in geometric group theory, such as the hyperboloic group. In the same way, Villani [3], Lott
[6] and Sturm [11] introduced in 2010 an extented ”synthetic” vision of the notion of Ricci curvature
minus a formulation of optimal transport on a metric space with a measure.
The main characters of differential geometry are objects called manifold, they can be seen locally as
an n-dimensional subspace of the Euclidean space, we define a differentiable structure on them, to turn
them into differentiable manifolds. A Riemannian manifold, the main focus of Riemannian geometry, is
a differentiable manifold endowed with a Riemannian metric. This metric is a tool which is useful to
calculate distances, measure angles, evaluate volume and so on.
In recent years, many notions of Riemannian geometry have proven to be useful in different research
areas: in graph theory, for instance, where geodesics are very useful to determine the shortest path
between two vertexes, in neural networks and so on.
The main objective of this thesis is to study the main basic concepts in Riemannian geometry focusing
in particular on the two fundamentals notions of geodesic and curvature, with many examples.
This thesis is structured in the following way: In Chapter 2, we start introducing some basic notions in
differential geometry. In Chapter 3, we introduce the notion of Riemannian metric and we then use it
to define Riemannian manifolds and the Levi-Civita connection. We conclude presenting geodesics. In
Chapter 4, we discuss the notion of curvature. We conclude with a summary of the results.

1
2. Preliminary Results
2.1 Manifolds

Before defining the notion of an n-dimensional manifold M , we need to introduce some key notions to
have a better understanding for what we are going to do.

2.1.1 Definition. Given a topological space M , we will say that it is a Hausdorff space if given two
points p1 , p2 in M , we can find two open sets U1 and U2 such that pi ∈ Ui , i = 1, 2, and U1 ∩ U2 = ∅.

2.1.2 Definition. A n-dimensional chart of M is the couple (U, ϕ), where U is and open set in the
topological space M and ϕ is a homeomorphism ϕ : U −→ ϕ(U ) ⊂ Rn .

2.1.3 Examples.

- The couple (Rn , idRn ) is a chart of Rn at each points.

- Let U be an open set of Rn , and j : U ,→ Rn an injection function, then (U, j) is a chart for all
points in U .

2.1.4 Definition. Given two charts (U, ϕ) and (V, ψ) such that U ∩ V 6= ∅, the transition function is
the map ϕ ◦ ψ −1 : ψ(U ∩ V ) −→ ϕ(U ∩ V ). If this map is a homeomorphism we will say that the two
charts are compatible. See figure in the Appendix A.1b

2.1.5 Definition. An atlas of M is a family {(Ui , ϕi )}i∈I of charts, such that (Ui )i∈I defines a cover
of M , and any two charts (Ui , ϕi ) and (Uj , ϕj ) for i, j ∈ I (i 6= j) are compatibles.

Now what is an n-dimensional manifold?

2.1.6 Definition. A topological space M is called n-dimensional manifold, if the following three
conditions are satisfied:

1. M is a Hausdorff space;

2. For any p ∈ M , there exists an open subset U that contains p, which is homeomorphic to an open
subset V ⊂ Rn . This means that we can find ψ : U −→ V such that ψ is a homeomorphism;

3. M has a countable basis of open sets.

From Definition 2.1.6 it turns out that a manifold is locally connected , locally compact due to condition
2, it is also the union of countably many compact subsets. The local coordinates around a point p ∈ M ,
is defined by taking a chart (Ui , ϕi ) coming from an atlas of M , such that Ui is an open set that contains
p and ϕi : Ui −→ Rn homeomorphism, ϕi (p) ∈ Rn . This means that ϕi (p) = (x1i (p), · · · , xni (p)), with
xki : Ui −→ R are differentiable functions, so that the family (x1i , · · · , xni ) is the local frame in the chart
(Ui , ϕi ).

2
Section 2.2. Differentiable manifolds and tangent space Page 3

2.1.7 Examples.

1. if we take the atlas (Rn , IdRn ), then Rn endowed with its usual topology, (Rn , IdRn ) is an n-
dimensional manifold.

2. The unit sphere S n of Rn+1 , is homeomorphic to Rn . So it is a topological space which is

separable. Next we define our atlas by {(U1 , ϕ1 ), (U2 , ϕ2 )} with U1 = S n − {N } and U2 =
S n − {S}, where we denote by N = (0, 0, · · · , 1) the North point and by S = (0, 0, · · · , −1) the
south point. we define by ϕ1 the stereographic projection into the intersection of the hyperplane
xn+1 = 0, and the line that passes through p and N . We denote by ϕ2 the stereographic projection
onto the intersection of the hyperplane xn+1 = 0, the line that passes through p ∈ S n and S. one
can show that S n = U1 ∪ U2 . Let any p ∈ S n , then p ∈ U1 or p ∈ U2 , and according to the open
in which belongs, p we take the stereographic projection ϕ1 or ϕ2 which is an homeomorphism.

Knowing that M is a topological space and separable in the sense of Hausdorff and has a basis of
countable open sets, we want now to define a structure of differentiable manifold on M .

2.2 Differentiable manifolds and tangent space

We start by recalling the definition of regular surface that we will then generalize to the concept of
differentiable manifold. Let S ⊂ R3 , S is a regular surface if for all p ∈ S, there exists an open set V
that contains p and h : U ⊂ R2 −→ V ∩ S, such that:

- h is a diffeomorphism.

- The differential of h at a point q dhq : R2 −→ R3 is injective for all q ∈ U .

2.2.1 Definition. A differentiable manifold of dimension n, is a set M , with a family of injective

mapping Xα : Rn ⊃ Uα −→ M , such that:

1. ∪α∈I Xα (Uα ) = M .

2. Given two mappings such that Xα (Uα ) ∩ Xβ (Uβ ) 6= ∅, the transition map Xβ ◦ Xα−1 is a differ-
entiable function.

3. {(Uα , Xα )}α∈I is a maximal atlas(see Proposition 6.5 on [5]).

Hence every family {(Uα , Xα )}α∈I that satisfies 1 and 2 of the Definition 2.2.1 is called differentiable
structure on M .
2.2.2 Example.

1. M = Rn
Considering the trivial atlas {(Rn , IdRn )}, hence Rn has a structure of differential manifold.
n+1
X
2. S n = {(x1 , ..., xn+1 )| x2i = 1} ⊂ Rn+1 .
i=1
Section 2.2. Differentiable manifolds and tangent space Page 4

as in example 2.1.7 consider the two stereographic projections, to be in phase with the definition
of differential manifold let us define the atlas. The stereographic projection is defined by the
following formula:
1
ϕ1 (x1 , x2 , ..., xn+1 ) = (x1 , ..., xn )
1 − xn+1
Similar formula holds for ϕ2 . Note that ϕ1 is differentiable, injective and maps S n − {N }
to Rn . The same property holds for ϕ2 . Hence ϕ1 and ϕ2 are invertible. Thus the set
{(Rn , ϕ−1 n −1
1 ), (R , ϕ2 )} defines a differentiable structure on S .
n

In what follows, we will denote by M n a n-dimensional differentiable manifold. The notion of equivalence
for two differentiable manifolds is the notion of diffeomorphism between them, the question is how can
we define differentiable maps between two differentiable manifolds?
2.2.3 Definition (differentiable maps between two manifolds). Let M1n and M2m two differentiable
manifolds. Let us consider the map f : M1 −→ M2 , and p ∈ M1

• f is differentiable at a point p if there exists a chart (U, ϕ) of M1 at p, and a chart (V, φ) of N

at f (p) such that f (U ) ⊂ V and the map φ ◦ f ◦ ϕ−1 : ϕ(U ) −→ φ(V ) is differentiable at ϕ(p).
• f is differentiable on M1 if it is the case at each point of M1 .

2.2.4 Remark.

- The function φ ◦ f ◦ ϕ−1 is a local expression for f in relation to the local charts (U, ϕ) and (V, φ).
- this local expression does depend on the choice of the two charts.
2.2.5 Definition. The tangent space at a point p ∈ M n , denoted by Tp M , is defined as follows:
Tp M = {v : D(M ) −→ R| ∃γ : I ⊂ R −→ M }
where I is an interval of R which contains the origin 0, and γ is a differentiable curve such that γ(0) = p
and γ 0 (0) = v and satisfies
d(f ◦ γ)
γ 0 (0)f = |t=0 .
dt
Let D(M ) the set of differentiable functions defined on M . Take a parametrization X : U −→ M n ,
such that X(0) = p, and f ∈ D(M ) defined in the neighborhood of p, and consider γ : I 3 0 −→ M .
In this parametrization we have:
f ◦ X(q) = f (x1 , ..., xn ), q = (x1 , ..., xn ) ∈ U
and
X −1 ◦ γ(t) = (x1 (t), ..., xn (t))
Indeed
d(f ◦ γ) df (x1 (t), ..., xn (t))
γ 0 (0)f = t=0
this implies γ 0 (0)f =
dt dt
n
X ∂ ∂xi (t)
γ 0 (0)f = f |t=0
∂xi dt
i=1
n   !
0
X
0 ∂
γ (0)f = xi (0) f ∀f ∈ D(M );
∂xi |t=0
i=1
Section 2.3. Vector Bundles Page 5

Pn    
Then γ 0 (0) = 0
i=1 xi (0)
∂
∂xi , we can conclude that the basis of Tp M is given by ∂
∂xi .
|t=0

2.2.6 Remarks.

- The point p of M is called the origin of all the tangent vectors at p.

- Let M be a differentiable manifold of dimension n. For all p ∈ M , Tp M is a real vector space of
dimension n.
2.2.7 Definition. Let M m and N n be two differentiable manifolds and f : M −→ N a diffeomorphism.
We define the differential map of f at p, like a linear map denoted by dfp , defined as follows:
dfp : Tp M −→ Tf (p) N
u −→ dfp (u)
d
with dfp (u) = dt f (γ(t)) t=0 , where γ(0) = p and γ̇(0) = u.
This notion of differential map is very useful to define the notions immersion, and embedding.
2.2.8 Definition. Let M m , N n two differentiable manifolds, f : M −→ N , a differentiable map of
classe C r .

- f is an immersion if and only if the differential map dfp is injective for all p ∈ M
- f is an embedding if and only is an immersion that in addition is an homeomorphism defined
from M into f (M ).

2.3 Vector Bundles

The aim of this section is to introduce the notion of vector bundles. Given a differential manifold M n ,
take p ∈ M and define the tangent space at p by Tp M . If we consider the union of all the tangents
spaces defined for all the points in M , we obtain the tangent bundle denoted by T M = ∪p∈M (Tp M ).
An element of T M is a couple (p, v) such that p ∈ M and v ∈ Tp M . This notion of tangent bundle
will be useful to define the vector fields on the differentiable manifold M . Consider M and E two
differentiable manifolds, and Π : E −→ M a surjective projection of class C ∞ .
2.3.1 Definition. E is a real vector bundle of rank n over M when:

1. For all p ∈ M , Ep := Π−1 (p) has a structure of real vector space of dimension n.
2. Local trivialization.
For all p ∈ M , there exists an open neighborhood U of p and a diffeomorphism
φ : Π−1 (U ) −→ U × Rn
such that the following diagram commutes:

;U c
Π P r1

Π−1 (U ) / U × Rn
φ

The restriction of φ to Ep defined by φ|Ep : Ep −→ {p} × Rn is a linear isometric.

Section 2.3. Vector Bundles Page 6

2.3.2 Remarks.

- In this Definition 2.3.1 for all p ∈ M , Ep ∼

= V, where V is a real vector space of dimension n.
- We write (E, M, Π, V) to denote that E is a real vector bundle over M .

- M is the base space and E total space. For this Definition 2.3.1 V = Rn .

Let (E, M, Π, V) be a vector bundle, and U an open subset of M . A local section C ∞ , s of E over U
is an application defined by: s : U −→ E such that Π ◦ s = idu . Knowing that a differentiable manifold
M n induces a differentiable manifold T M . In Definition 2.3.1 for E = T M , we have: Π : T M −→ M ,
that assigns (p, v) ∈ T M to p ∈ M . Π is a subjective projection C ∞ . Let us show that (T M, M, Π, Rn )
is a real vector bundle.

1. Let p ∈ M we have:

Ep = Π−1 (p) = {(q, v) ∈ T M Π(q, v) = p} =⇒ Ep = {(q, v) ∈ T M | q = p} = {p} × Tp M

Since Tp M ∼
= Rn
=⇒ Ep = {p} × Rn ∼
= Rn .

Then Ep is a real vector space of dimension n.

2. Local trivialization
Let p ∈ M , there exists a chart(U, ψ) coming from an atlas of M such that U 3 p, this implies
(T ψ, ψ(U ) × Rn ) is a chart of T M , where

T ψ : Π−1 (u) −→ ψ(U ) × Rn

 
i ∂
q, v |q −→ (ψ(U ), v 1 , ..., v n ); (2.3.1)
∂xi

and (x1 , ..., xn ) is the local coordinates in the chart (U, ψ). Let us define φ. Having T ψ and
ψ −1 × idRn : ψ(U ) × Rn −→ U × Rn ; we can write: φ = (ψ −1 × idRn ) ◦ T ψ.
 
∂
Let us show now the commutativity. We need to show that P r1 ◦ φ = Π. Let q, v i ∂x i
|q in
Π−1 (U ), we have:
   
i ∂ −1 i ∂


P r1 ◦ φ q, v |q = P r1 ◦ (ψ × idRn ) ◦ T ψ q, v |q
∂xi ∂xi
= P r1 ψ −1 (ψ(q)), idRn (v 1 , ..., v n )

= P r1 q, v 1 , ..., v n = q


   
i ∂ i ∂
P r1 ◦ φ q, v |q = Π q, v |q =⇒ P r1 ◦ φ = Π
∂xi ∂xi
So it commutes.
Now let us show that φ|Ep : Π−1 (q) −→ {q} × Rn is a linear isomorphism. We have:
 
i ∂
φ q, v |q = (q, (v 1 , ..., v n ))
∂xi
So φ|Ep is a linear isomorphism. Which ends the local trivialization.
Section 2.3. Vector Bundles Page 7

We can conclude that (T M, M, Π, Rn ) is a vector bundle of T M over M . Its global sections are defined
as follows:

s:M −→ T M
p −→ (p, s(p)) such that s(p) ∈ Tp M

This is because we have Π ◦ s(p) = Π(p, s(p)) = p ie Π ◦ s = idM . Therefore sections of T M above
M are vector fields. Hence this gives rise to the following definition.

2.3.3 Definition. A vector field is a map X : M −→ T M , which associates to a point p ∈ M an

element X(p) ∈ Tp M , such that: Π ◦ X = idM .

2.3.4 Remark.

1. The vector field is said to be differentiable if X is differentiable.

2. Defining F(M ) to be the set of function defined on M , a vector field can also be viewed as a
map X : D(M ) −→ F(M ) which maps f ∈ D(M ) to Xf ∈ F(M ). Considering a local chart
(U, ρ) of a point p ∈ M , with is local coordinates (x1 , ..., xn ) we have:

∂f
(Xf )(p) = ai (p) (p);
∂xi

where ai : U ⊂ Rn −→ R are differentiable functions in sense to make X to be differentiable.

3. Let f, g ∈ D(M ) we have: X(f g) = X(f )g + f X(g).

2.3.5 Definition. Let X and Y be two differentiable vector fields. The lie bracket of X and Y is
denoted by [X, Y ] and is defined as follows: [X, Y ] = XY − Y X

This lie bracket has the following properties. For any three vector fields X, Y, Z and f, g ∈ D(M ) we
have:

1. [X, Y ] = −[Y, X] (antisymmetry),

2. [aX + bY, Z] = a[X, Z] + b[Y, Z], a, b ∈ R(linearity),

3. [[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0 (Jacobi identity),

4. [X, Y ](f g) = f [X, Y ](g) + [X, Y ](f )g,

5. [X, gY ] = [X, Y ]g + X(g)Y.

3. Riemannian manifolds and Levi-Civita
connection
3.1 Riemannian metrics

3.1.1 Definition. A Riemannian metric on a differentiable manifold M is a correspondence which

associates to each point p of M an inner product <, >p on the tangent space Tp M . It is Differentiable in
the sense that given a local coordinate around p, X : U ⊂ Rn −→ M , with X(x1 , ..., xn ) = q ∈ X(U )
∂
and ∂x i
(q) = dXq (0, ..., 1, 0, ..., 0) with 1 on the ith position, then:

∂ ∂
< (q), (q) >= gij (q) (3.1.1)
∂xi ∂xj

The gij in (3.1.1) defines differentiable functions. We have gij = gji due to the symmetry of the inner
product in the coordinate X. A Riemannian manifold is a differentiable manifold, equipped with a
Riemannian metric.
Since two differentiable manifolds M m and N n , are equivalent if we can define a diffeomorphism between
them, this gives rise to the question: How can we say that two Riemannian manifolds are equivalents?.
3.1.2 Definition. Let M and N be two Riemannian manifolds, and f : M −→ N a diffeomorphism.
The map f is an isometry if it preserves the inner product:

< u, v >p =< dfp (u), dfp (v) >f (p) , ∀p ∈ M, u, v ∈ Tp M (3.1.2)

Hence two Riemannian manifolds; M and N , are locally isometric at p; if we can find an open
neighborhood U ⊂ M which contains p and f : U −→ f (U ) is a diffeomorphism; that satisfies (3.1.2).
3.1.3 Some Examples of Riemannian manifolds.

1. Euclidian space M = Rn ,
We know that (Rn , Id) is a differential manifold, and since we have Tp M ∼ = Rn , ∀p ∈ Rn , we
Xn
can define the usual inner product < u, v >p = ui vi , where u = (u1 , ..., un ), v = (v1 , ..., vn ) ∈
i=1
Tp M . Then M with this metric, that we denote by (M, <, >), is a Riemannian manifold.
2. Immersed manifolds.
Let us consider two differentiable manifolds M n , N n+k ,n, k ∈ N∗ and f : M n −→ N n+k , an
immersion. If N is a Riemannian manifold, then M is a Riemannian manifold equipped with the
inner product defined by :

< u, v >p =< dfp (u), dfp (v) >f (p) . (3.1.3)

Since M n is a differentiable manifold, we just need to prove that <, >p , defined in (3.1.3), is an
inner product on Tp M
• It is evident that <, >p is bilinear and symmetric due to the fact that <, >f (p) is an inner
product on N .

8
Section 3.1. Riemannian metrics Page 9

• Positive definite. Let u ∈ Tp M , let us prove that <, > is Positive definite, which means that

< u, u >p ≥ 0 (3.1.4)

and
< u, u >p = 0 =⇒ u = 0Rn . (3.1.5)

We have:

< u, u >p = 0 =⇒ < dfp (u), dfp (u) >f (p) = 0

=⇒ dfp (u) = 0 because <, >f (p) is an inner product on Tf (p) N
=⇒ u ∈ ker dfp
=⇒ u = 0Rn since dfp is injective

which prove (3.1.5). By definition we have

< u, u >p = < dfp u, dfp u >f (p) ≥ 0 , because <, >f (p) is an inner product on Tf (p) N

Which prove (3.1.4)

So we can conclude that <, > is an inner product on Tp M , and then M equipped with this
metric is a Riemannian manifold.
3. The unit sphere S n−1 of Rn , n ≥ 2
Let us prove that it is a Riemannian manifold with a metric that we are going to define on it.
S n−1 is an embedded manifold because it is the pre-image of a regular value by the norm square,
hence it is a Riemannian manifold, equipped with the metric induced by Rn on S n−1 .
3.1.4 Proposition. Given a differentiable manifold M , we can equip it with a Riemannian metric.

Proof. The proof can be found in Do Carmo’s Book [2]

Up to now we have seen how we can define a Riemannian metric on a differentiable manifold M to make
it a Riemannian manifold then how can we calculate the length of a curve over M ?. Let c : I ⊂ R −→ M
be a differentiable map, which means that c is a curve on M .
3.1.5 Definition. A vector field along a curve c, is a tangent vector on c; that we can see as a map
V : I −→ T M , such that the folllowing diagram commutes:

I
c V
~ !
Mo TM
Π

ie: Π ◦ V = c

Now given a segment on c, which is a restriction of c to a closed interval [a, b] ⊆ I; a, b ∈ R(a ≤ b),
and a Riemannian manifold M . we define the length of the curve c along the segment [a, b], denoted
by lab (c) as follows:
Z b
b dc dc 1
la (c) = (< , >) 2 dt. (3.1.6)
a dt dt
Section 3.2. Connection and covariant derivative Page 10

To conclude this section, we can resume that given a differentiable manifold M , we can always define
a Riemannian metric on it by using the Proposition 2.1.5, and then define at each point p ∈ M a
measurement of the length of the tangent vector, also the angles between them. In the following
section we are going to adopt the following notation:

X(M ) = Set of all vector fields of class C ∞ on M

D(M ) = the ring of real valued function C ∞ on M

In what follows we will work with an n−dimensional Riemannian manifold M , and its metric <, > that
we will denote by (M n , <, >), n ≥ 2.

3.2 Connection and covariant derivative

3.2.1 Definition. Given a Riemannian manifold M ;a connection ∇ is a correspondence defined as

follows:

∇ : X(M ) × X(M ) −→ X(M )

(X, Y ) −→ ∇X Y (3.2.1)

with the following properties:

i) ∇X (Y + Z) = ∇X Y + ∇X Z

ii) ∇(f X+gZ) Y = f ∇X Y + g∇Z Y

iii) ∇X (f Y ) = X(f )Y + f ∇X Y

∀ X, Y, Z ∈ X(M ), f, g ∈ D(M ).

In this definition, the connection ∇X Y is the derivative of Y in the direction of X .

3.2.2 Remark. From the Definition 3.2.1

1. The second property satisfied, means that the connection ∇ is linear in X on the space X(M ).

2. The third one shows us that with respect to Y , it acts like a derivation.

The usual derivative dv(t)

dt of a vector field along a curve v(t) does not belong to T M in general, hence
d
the notion dt is not an intrinsic property. To avoid this, we introduce the notion of the covariant
derivative (that we can geometrically see like the orthogonal projection of the usual derivative of v(t)
over T M ).

3.2.3 Definition. A Covariant-derivative is a correspondence D, which associates to a vector field V

along c : I ⊂ R −→ M (curve on M ) a vector field DVdt on T M , such that:

D(V +W ) DV DW
- dt = dt + dt ,
D(f V ) Df
- dt = dt V +f DV
dt ,
Section 3.2. Connection and covariant derivative Page 11

- Given V vector feild along c, such that it induce by Y ∈ V ect(M ), which means that ∃t ∈ I such
that V (t) = Y (c(t)), we have :
DV DY (c(t))
= = ∇ dc(t) Y. (3.2.2)
dt dt dt

This last property makes sense, since ∇X Y (p) depends on the value of X(p) and the value of Y
along the curve, tangent to X at p .This last property shows us also that the connection allows us to
differentiate a vector along a curve.
3.2.4 Lemma. The connection is an intrinsic property, it depends on the ambient space.

Consider the local coordinate around a point p ∈ M , p = (x1 , ..., xn ), the Components of connection
can be express in terms of Christoffel symbols:

∇Xi Xj = Γkij Xk . (3.2.3)

where Γkij are differentiable functions.

For the proof of Lemma 3.2.4 see Appendix A.0.4. Having the Definition 3.2.3, we can define some key
concepts of parallel vector, and parallel transport.
3.2.5 Definition. Let (M, <, >) be a Riemannian manifold, with its connection ∇; a vector field V
along the curve c, such that DV
dt ≡ 0 ∀t ∈ I is called a parallel vector along c.

3.2.6 Lemma. Let M 2 ⊂ R3 be a surface in R3 with the induced Riemannian metric. Let c : I −→ M
a differential curve on M and V a vector field tangent to M along c. Then V is parallel iff dV
dt is
3
perpendicular to Tc(t) M ⊂ R .

Proof. Suppose that V is parallel. Let ω ∈ Tc(t) M , consider the orthogonal projection of dV
dt on Tc(t) M ,
> ⊥ >
we know that dV (Tc(t) M )⊥ ; then dV dV
+ dV ; where dV
L



dt ∈ Tc(t) M dt = dt dt dt ∈ Tc(t) M and
dV

⊥ ⊥
dt ∈ (Tc(t) M ) . Recall The covariant derivative is just the projection of the usual derivative in

>
the ambient space. In this case the ambient space is Tc(t) M , the covariant derivative is dVdt

We have:
 >  ⊥  >  ⊥
dV dV dV dV dV dV
< , ω >=< + , ω > ⇐⇒ < , ω >=< + ,ω >
dt dt dt dt dt dt
 >  ⊥
dV dV dV
⇐⇒ < , ω >=< ,ω > + < ,ω >
dt dt dt

⊥
⊥
Since < dV
dt , ω >= 0, because dV
dt ∈ (Tc(t) M )⊥
> ⊥
dV >
   
dV dV dV
< + , ω > ⇐⇒ < , ω >=< ,ω >
dt dt dt dt
dV > DV
 
by hypothesis = =0
dt dt
dV
⇐⇒ < , ω >= 0.
dt
Section 3.2. Connection and covariant derivative Page 12

3.2.7 Example. Let us consider S 2 ⊆ R3 the sphere of R3 ; with the metric induced by the metric of
R3 on S 2 . The velocity field along great circles, parametrized by arc length, is a parallel field. How to
see it? Consider the parametrization of the great circle define by
γ : [0; 2π] −→ S 2
t −→ (cost, sint, 0). (3.2.4)
The velocity is given by V (t) = γ̇(t) = (−sint, cost, 0), hence dV 2 ⊥
dt = γ̈(t) ∈ (Tγ(t) S ) , so it is
orthogonal to Tγ(t) S 2 . By Lemma 3.2.6, we conclude that V (t) is a parallel field. For the sketch see in
the Appendix figure A.1a, our curve is in red and the parallel field is arrow.
The definition of a parallel vector transport comes directly, from the following proposition.
3.2.8 Proposition. Consider a Riemannian metric (M n , <, >), and a connection ∇. Taking V0 ∈
Tc(t0 ) M ; there exists a unique parallel vector along c. V ; such that V (t0 ) = V0 .

Proof. Confer Do Carmo [2]

This unique vector field defined in Proposition 3.2.8 is called the parallel transport along the curve c;
which could be also viewed like a vector field V along c
V : I0 −→ T M
t −→ V (t), (3.2.5)
Where I0 is an open interval which contains t0 , such that at t = t0 ; V (t0 ) = V0 .
Before we reach to the main theorem of this section 2 which is a Fundamental theorem of Riemannian
Geometry, let us introduce two key notions of symmetry of a connection ∇ and the compatibility of a
connection with respect to the metric on a Riemannian manifold. Let us consider a Riemannian manifold
M with its metric <, >.
3.2.9 Definition. A connection ∇ on M , is compatible to the metric <, >. When given a curve c and
two parallel vector fields P, P 0 along the curve c we have:
< P, P 0 >= constante (3.2.6)
3.2.10 Proposition. Considering the same Riemannian manifold as in Definition 3.2.9. A connection
∇ on M is compatible with the metric if and only if for any two vectors W1 , W2 field along the curve
c we have:
d DW1 DW2
< W1 , W2 >=< , W2 > + < W1 , > ∀t ∈ I. (3.2.7)
dt dt dt

Proof. Confer chapter 2 in Do Carmo’s Book [2]

3.2.11 Remarks.

1. The compatibility of the connection allows us to differentiate the inner product, by using the
product rule.
2. We can also say that a connection ∇ on a Riemannian manifold M n , is compatible if it preserves
the metric, which means that:
X < Y, Z >=< ∇X Y, Z > + < Y, ∇X Z >; X, Y, Z ∈ X(M ).
Section 3.2. Connection and covariant derivative Page 13

3.2.12 Definition. A connection ∇ is said to be symmetric, when it is torsion free, which means

∇X Y − ∇Y X = [X, Y ] ∀X, Y ∈ X(M ). (3.2.8)

3.2.13 Theorem (Levi-Civita connection). Given a Riemannian manifold M , we can find a unique
connection ∇ on M , which satisfies:

1. ∇ is symmetric.

2. ∇ is compatible with the Riemannian metric on M .

3.2.14 Lemma (Koszul formula). Given a connection ∇ that is symmetric and compatible with the
metric, we have the following formula:

2 < ∇X Y, Z > = X(< Y, Z >) + Y (< X, Z >) − Z(< X, Y >)− < Y, [X, Z] > − < X, [Y, Z] >
− < Z, [Y, X] > . (3.2.9)

For the prove see appendix A.0.1. Now let us prove Theorem 3.2.13

Proof.

• Assuming the existence of this connection ∇, let us show that it is unique. Since ∇ is symmetric,
by lemma it satisfies (3.2.9), by the uniqueness of the inner product, we conclude directly that ∇
is unique.

• Let us prove the existence of this connection which satisfies the two conditions of the theorem.
Let us define ∇ like in (3.2.9), and let us show that it verified the 3 conditions of a connection.

– The first property comes, using the fact that taken Y1 , Y2 ∈ X(M ),we have,
[Y1 + Y2 , Z] = [Y1 , Z] + [Y2 , Z] and also by the linearity of <, > we get the answer.
– For second one we use the same trick.
– The property of derivative. Let f ∈ D(M ), in (3.2.9) replacing Y by f Y we have :

2 < ∇X (f Y ), Z > = X(< f Y, Z >) + f Y (< X, Z >) − Z(< X, f Y >)− < f Y, [X, Z] >
− < X, [f Y, Z] > − < Z, [f Y, X] > .

Since

X(< f Y, Z >) = X(f < Y, Z >) = X(f ) < Y, Z > +f X < Y, Z >
Z(< X, f Y >) = Z(f ) < X, Y > +f Z < X, Y >
[f Y, Z] = f Y Z − Z(f Y ) = f Y Z − f ZY − Z(f )Y = f [Y, Z] − Z(f )Y
[f Y, X] = f [Y, X] − X(f )Y.
Section 3.3. Geodesics Page 14

After putting those expressions together, we can write

2 < ∇X (f Y ), Z > = f (X(< Y, Z >) + Y (< X, Z >) − Z(< X, Y >)

− < Y, [X, Z] > − < X, [Y, Z] − < Z, [Y, X] >)
−Z(f ) < X, Y > +Z(f ) < X, Y >
+X(f ) < Y, Z > +X(f ) < Y, Z >
= 2f < ∇X Y, Z > +2X(f ) < Y, Z >
2 < ∇X (f Y ), Z > = 2 < f ∇X Y, Z > +2 < X(f )Y, Z >
2 < ∇X (f Y ), Z > = 2 < f ∇X Y + X(f )Y, Z >
2 < ∇X (f Y ) − f ∇X Y − X(f )Y, Z > = 0 implies ∇X (f Y ) = f ∇X Y + X(f )Y

Then ∇ acts like a derivative

This ends the proof of Theorem 3.2.13

3.2.15 Remark. The connection defined in Lemma 3.2.14, is called the levi Civita-connection on the
Riemannian manifold M.

3.2.16 Corollary. For the Levi-Civita connection of M , considered at a point p ∈ M an open neigh-
borhood U ⊂ M containing p, and the local coordinates (x1 , ..., xn ), with the basis (X1 , ..., Xn ), where
∂
Xi = ∂x i
(p), we have:

1. Γm m
ij = Γji

2. The relation between the local representation of a Riemannian metric (gij ) and the Christoffel
symbols are given by
1 ml
Γmij = g (gjl,i + gil,j − gij,l ) ; (3.2.10)
2
∂
where ∂k gij = ∂xk (gij ) = gij,k

For the demonstration see appendix A.0.3. Now we are going to introduce one of the fundamental
notions in Riemannian geometry, the geodesic which comes from geodesy, the science of measuring the
size and shape of the earth. In what follows, we will work with the Levi-Civita connection and thus, for
D
every curve, γ on M , let dt be the associated covariant derivative along c, ∇γ 0 .

3.3 Geodesics

3.3.1 Definition. Let (M n , < , >) be a Riemannian manifold. A curve, γ : I ⊆ R −→ M is a

0
geodesic if and only if Dγ 0 0
dt = ∇γ γ = 0; which means γ is a parallel vector along the curve γ.
0

If M was embedded in Rd , a geodesic would be curve γ, with zero acceleration vector which means
0
that γ 00 = Dγ 0
dt ≡ 0. A geodesic is said to be normalized if |γ (t)| = 1.
Section 3.3. Geodesics Page 15

3.3.2 Geodesic equation. Given a Riemannian manifold (M n , < , >), and a geodesic γ, taking the
local chart at p = γ(t0 ) ∈ M , (U, ϕ), where U is an open that contains p, with its local coordinates
(x1 , ..., xn ),let us
 find the equation satisfied by the geodesic in this local coordinates. the local frame
∂
is given by ∂xi at the given point p. γ(t) = (x1 (t), ..., xn (t)), so
dγ 0
γ 0 (t) = ∂
= (ẋ1 (t), ..., ẋn (t)) = ẋi ∂x
dt i
. Since γ is a geodesic, we have Dγ 0
dt = ∇γ γ = 0. Then:
0

 
0 j ∂
∇γ γ = 0 =⇒ ∇ẋi ∂ ẋ
0 =0
∂xi ∂xj
 
i ∂ j ∂ i j ∂
=⇒ ẋ (ẋ ) + ẋ ẋ ∇ ∂ =0
∂xi ∂xj ∂xi ∂xj
d k ∂ ∂
=⇒ (ẋ ) + ẋi ẋj Γkij =0
dt ∂xk ∂xk
 2 k 
d x ∂
=⇒ 2
+ ẋi ẋj Γkij =0
dt ∂xk
d 2 xk
∇γ 0 γ 0 = 0 =⇒ + ẋi ẋj Γkij = 0
dt2
Therefore the equation satisfied by the geodesic in the local frame is given by:
d2 xk
+ ẋi ẋj Γkij = 0; (3.3.1)
dt2
which is a second-order non linear differential equation in xi .
3.3.3 Examples of geodesics.

• On Rn geodesics are straights lines, with constant speed.

this is because if we consider a lines γ(t) = (γ1 (t), ..., γn (t)) with constant speed it will means
that γ 0 (t) = (γ10 (t), ..., γn0 (t)), where each components are constants, so γ 00 (t) ≡ 0, then γ is a
geodesic.
• On S n (r)the sphere of radius r is a Riemannian manifold.
Since S n (r) is a Riemannian manifold of Rn+1 with the induced metric of Rn+1 . Let us denote
by ∇ the Levi-Civita connection of Rn+1 and by ∇ the Levi-Civita connection of S n (r). Let us
find the Levi-Civita connection of S n (r). Using the fact that S n (r) is embedding in Rn+1 , let
call h : S n (r) −→ Rn+1 the immersion, we know that S n (r) is a Riemannian manifold with the
metric induced by the euclidean metric on S n (r), then considering two vectors fields X and Y on
S n (r), which are also vector fields in Rn+1 , and then (∇X Y )(p) is just the tangent component of
(∇X Y )(p). Hence we can write: ∇X Y =Projection (∇X Y ) on the ambient space. We therefore
have:
∇X Y = (∇X Y )⊥ + (∇X Y )>
Since (∇X Y )> = ∇X Y
∇X Y = (∇X Y )⊥ + ∇X Y =⇒ ∇X Y = ∇X Y − (∇X Y )⊥ .
Knowing that the normal vector of S n (r) at a point p is the simply the line generated by p, let
p p p
us take the unit normal one |p| and hence we have (∇X Y )⊥ =< ∇X Y, |p| > |p| . Then the
n
Levi-Civita connection of S (r), is given by:
p p
∇X Y = ∇X Y − < ∇X Y, > (3.3.2)
|p| |p|
Section 3.3. Geodesics Page 16

Let γ : I ⊂ R −→ S n (r), γ is a geodesic implies that ∇γ 0 γ 0 = 0, using (3.3.2) it means that:

γ(t) γ(t) γ(t) γ(t)

∇γ 0 (t) γ 0 (t)− < ∇γ 0 (t) γ 0 (t), > =0 =⇒ ∇γ 0 (t) γ 0 (t) =< ∇γ 0 (t) γ 0 (t), >
|γ(t)| |γ(t)| |γ(t)| |γ(t)|
1
=⇒ γ 00 (t) = < γ 00 (t), γ(t) > γ(t).
r2
Hence geodesics on the sphere S n (r) are curve γ such that:
1
γ 00 (t) = < γ 00 (t), γ(t) > γ(t).
r2

• If we consider a particular case of the unit sphere of R3 , geodesics are the great circles. Let us
show it.
Considering the parametrization in polar coordinates as follows:

h : (0, π) × [0, 2π] −→ S 2

(ϕ, θ) −→ (sinϕcosθ, sinϕsinθ, cosϕ).

Using the following result that we are going to show in Chapter 4, we have the Christophel
symbols which are given by: Γ111 = Γ112 = Γ121 = Γ211 = Γ222 = 0 and Γ122 = − 21 sin(2ϕ);
Γ212 = Γ2 21 = cot(ϕ). Then using (3.3.1), and setting x1 = ϕ; x2 = θ we have the following
equation satisfied by the geodesic:

d2 ϕ dθ dθ
2
+ Γ122 =0 (3.3.3)
dt dt dt
d2 θ dϕ dθ
2
+ 2Γ212 =0 (3.3.4)
dt dt dt
Changing t as a function of θ as follows: t = aθ; a 6= 0, so there is not dependence on ϕ, so ϕ is
d2 θ
viewed like a constant it means that dϕ dθ 1
dt = 0; t = aθ =⇒ dt = a and then dt = 0. By applying
those results on the above equations, (3.3.4) is always true because the right hand side vanishes.
(3.3.3) implies that, 0 − 21 sin(2ϕ) × a1 = 0, so sin(2ϕ) = 0 then ϕ = kπ 2 ; k ∈ Z. Since
π
ϕ ∈ (0, 2π) =⇒ 0 < k < 2 =⇒ k = 1, then ϕ = 2 . Hence our geodesic is given by

h π2 : [0, 2π] −→ S 2
π π π
θ −→ (sin( )cosθ, sin( )sinθ, cos( ) = (cosθ, sinθ, 0).
2 2 2
This is a great circle. If we give another initial conditions such that the initial point does not
belong to the equator, we can make some transformation like a rotation to end at the solution.
Since on the sphere, it is the only circle which is invariant by such kind of isometry we will reach
again to another great circle. Then we can conclude that geodesics are great circles on the sphere.

Having (3.3.1), we could determine geodesics by solving it, which is not an easy task since it is a non
linear differential equation. We Know that every second order differential equation can be transformed
into a system of first-order differential equations, except to work on M , we will work on T M . setting
dxk k dy k i j k
dt = y , then we have dt = −ẋ ẋ Γij which lead to the following system of differential equations:
( k
dx k
dt = y
dy k
i j k
(3.3.5)
dt = −ẋ ẋ Γij
Section 3.3. Geodesics Page 17

k = 1, ..., n. The solutions of this equation will be seen like a trajectory passing through a point q ∈ U ,
with velocity vq ∈ Tq U in T M . Hence we are interested in the uniqueness and the existence of the
solution of (3.3.5) with initial condition, at time t0 it passes through a point p with a velocity vp ∈ Tp U .
having a curve γ : I ∈ R −→ M , we can define curve β on T M as follows:

β : I −→ T M
t −→ (γ(t), γ 0 (t))

Hence if γ is a geodesic, then β defined like above should be a geodesic on T M .

3.3.4 Proposition. Let p ∈ M , v ∈ Tp M , there exists a unique geodesic γv satisfying:

γv (t0 ) = p, γ̇v (t0 ) = v

Proof. Assuming the existence, let us first prove the uniqueness of γv . Consider the coordinates (U, ϕ)
on M , the trajectories of γv on TU are defined as follows t −→ (γv (t), γ̇v (t)), where γv is a geodesic.
It follows that γv satisfies (3.3.5), with initial condition γv (t0 ) = p, γ̇v (t0 ) = v. Using the Cauchy-
Lipschitz theorem, if γv exists then it is unique. The existence comes directly if we define a system of
equation which verifies (3.3.5) with the initial condition mentioned in Proposition 3.3.4 using the same
Cauchy-Lipschitz theorem the existence comes.

Hence the above Proposition 3.3.4 can be also be rewritten as follows by using theorem 2.2 chapter 3
of Do Carmo [2].

3.3.5 Proposition (Do Carmo [2]). Let (M n , < , >) be a Riemannian manifold, given p ∈ M , there
exist V ⊂ M , V and open that contains p; δ > 0, 1 > 0 and C +∞ mapping φ : (−δ, δ) × U −→ M ;
U = {(q, v) ∈ Tp M ; p ∈ V ; v ∈ Tq M, |v| < 1 } such that the curve t −→ γ(t, q, v), t ∈ (−δ, δ) is the
unique geodesic of M which at t = 0, passes through q = γ(0, q, v) with velocity v = γ̇(0, q, v) for all
q and v element of the tangent space at p, such that |v| < 1 .

3.3.6 Lemma. (Homogeneity of a geodesic) If the
geodesic γ(t, q, v) is defined on the interval (−δ, δ),
Then γ(at, q, v) is a geodesic defined on − aδ , aδ and γ(t, q, av) = γ(at, q, v). Notice that if a < 0 we
just swap the bounds of the interval.

Proof. Let h : − aδ , aδ −→ h(t) = γ(at, q, v) be a curve. h is well defined, since at ∈ (−δ, δ). We

have h(0) = γ(0, q, v) = q and dh dh

dt (t) = aγ̇(at, q, v) implies that
dt (0) = aγ̇(0, q, v) = av. Let us
D dh 0
check if h satisfies the property of a geodesic, we have: dt dt (t) = ∇h0 (t) h (t) using the expression
D dh 2


of the derivative of h we get dt dt (t) = ∇aγ̇(at,q,v) (aγ̇(at, q, v))
= a ∇γ̇(at,q,v) (γ̇(at, q, v)). Since
D dh
at ∈ (−δ, δ) we obtain ∇γ̇(at,q,v) (γ̇(at, q, v)) = 0, then dt dt (t) = 0. So h is a geodesic which
passed through q with a velocity av, using the uniqueness of geodesic we have h(t) = γ(at, q, v), so
γ(at, q, v) = γ(t, q, av).

Now let us introduce a little bit the notion of exponential map, and some examples of them before
ending this section by the properties of minimizing geodesic.
Section 3.3. Geodesics Page 18

3.3.7 Exponential map. The interest in defining this notion is due to the fact that it helps to parametrize
a Riemannian manifold M , locally near any p ∈ M in terms of a map from the tangent space Tp M to
the manifold. This notion of the exponential map is being defined in terms of geodesics.
3.3.8 Definition. Let P ∈ M , and let U ⊂ T M be an open set given like in Proposition 3.3.5, then
the map exp : U −→ M , given by:

exp(q, v) = γ(1, q, v), (q, v) ∈ U ;

is called the exponential map on U .

1 v
Using the Lemma 3.3.6, we can see that taken a = |v| we can write : exp(q, v) = γ(|v|, q, |v| )
in what follows we are going to define the exponential map on an open ball that contains the origin 0
of Tq M , with radius  in Tq M denoted B (0) as follows:

expq : B (0) −→ M.

3.3.9 Proposition. The exponential map is a local diffeomorphism.

Proof. Let us find d(expq )0 . we have:

d
d(expq )0 v = (expq (tv))|t=0 = v; by the definition of the exponential map;
dt
d(expq )0 = id.

Hence d(expq ) has a maximal rank, using the inverse theorem function there exists an open ball in
Tq M of radius  with center 0, and an open set V ⊂ M such that expq : B (0) −→ V ⊂ M is a
diffeomorphism.

3.3.10 Remarks.

• The map exp is differentiable since γ is differentiable.

• Given a point q ∈ M , expq is diffeomorphic from an open subset of T M to an open subset of M .

• The exponential map is not necessarilly defined on all of the tangent space.

3.3.11 Example of exponential map on some manifold.

• The euclidean space Rn

We know that for a given p ∈ Rn we have Tp Rn ∼
= Rn , and since geodesic are straight line we
have:

expp : Rn −→ Rn .
v −→ p + v

• The unit sphere S 2 of R3

Since that geodesics are great circles, defining exponential map at a point (N, v) ∈ TS2 , such that
N is the north pole, we will have: expN (2πv) = N . It is easier to view that defining all on TN S 2
it is not injective because expN ( π2 v) = expN (− 3π
2 v). And to make it to be injective we can define
it on Bπ (0) to S 2 − {S} the south pole.
Section 3.3. Geodesics Page 19

3.3.12 Geodesic coordinates. Having the notion of exponential map, it could be useful if we can
define a local coordinates which could simplify some computations. A geodesic coordinates is local
coordinates on Riemannian manifold with a symmetry connection defined by the exponential map on
an open ball containing the origin 0 in Tp M and an isomorphism E : Rn −→ Tp M given by any
orthonormal basis of the tangent space at a fixed point p ∈ M . It means that in this coordinate the
Christophel symbols Γm ij (p) will vanish since p is identified by (0, 0, .., 0), and this will induce the fact
that the covariant derivative of a tensor will coincide with the normal derivative.
Hence working on a geodesic coordinates around a point p will induce that (3.3.1) of geodesic in this
local frame will be simply reduce to
d2 xk
=0
dt2
which implies that: xk = β k t. It is coming down that there is a correspondence between the straight
lines xk in the domain where the geodesic coordinates are defined an curves γ(t). To conclude this
having the exponential map we can also define the notion of normal neighborhood V ⊂ M , which
is the image by the exponential map of an open U ⊂ Tp M , such that the restriction of expp|U is a
diffeomorphism into V . If U is a open ball on Tp M is hold that V = expp (U ) is a normal ball.

3.3.13 Proposition. Let p ∈ M , we can find a normal neighborhood of p, V , and a constant α > 0
such that for every q ∈ V , the exponential map defined on the open ball Bα (0) ⊂ Tq M is diffeomorphic
to an open of M and we have V ⊂ expq (Bα (0))

Proof. We use the Proposition 3.3.5 and the Proposition 3.3.9. For more detail see Do.Carmo book’s [2]

3.3.14 Minimizing properties of geodesics. The idea behind this notion is to show that the geodesic
which connects two sufficiently close points is the shortest path between them.

3.3.15 Definition. A piecewise differential curve β defined on [a, b] is a continuous function, such
that it exists a partition of the interval [a, b] into interval [ti ; ti+1 ], i ∈ {0, ..., n − 1} with t0 = a, tn = b,
such that β|[ti ;ti+1 ] are differentiable.

Considering the segment of geodesic γ : [a, b] −→ M , denoted by P the set of piecewise curves joining
γ(a) and γ(b).

3.3.16 Definition. A segment of the geodesic γ defined like above, is called minimizing if:

lab (γ) ≤ lab (β) where β ∈ P.

Hence we can see that geodesics locally minimize the arc length. This hold that every small segment of
geodesic is minimal.

3.3.17 Proposition (proposition 3-6 Do.Carmo [2]). Let p ∈ M , U a normal neighborhood of p, and
B ⊂ U a normal ball of center p. Let γ : [0, 1] −→ B be a geodesic segment with γ(0) = p. If
c : [0, 1] −→ M is any piecewise differentiable curve joining γ(0) to γ(1) then l01 (γ) ≤ l01 (c) and if
equality holds then γ([0, 1]) = c([0, 1]).

3.3.18 Proposition. Let M be a Riemannian manifold, and γ a geodesic on M , then γ is regular.

Section 3.3. Geodesics Page 20

Proof. Let γ be a geodesic, and a point p ∈ γ, hence p ∈ M . Defining the exponential map on Tp M ,
is locally diffeomorphic to and open V ⊂ M due to Proposition 3.3.9, then expp : U ⊂ Tp M −→ V
is a diffeomorphism, setting V to be a convex neighborhood. Then taking another point q such that
q ∈ V and belong to γ, hence γ can be viewed as a map [t0 , t1 ] −→ M such that γ(t0 ) = p, γ(t1 ) = q,
and γ([t0 , t1 ]) ⊂ V , γ|[t0 ,t1 ] is the unique geodesic that connects p and q . Consider

c : [t0 , t1 ] −→ V
 
t − t0
t −→ expp v .
t − t1

it comes that c is a differentiable map like composition of two differentiable maps, so c is a curve, we
have c(t0 ) = expp (0) = p, c(t1 ) = expp (v) = q, using Proposition 3.3.17, ltt01 (γ) ≤ ltt01 (c). c is regular.
By using the definition of the exponential map and Lemma 3.3.6; we find out that c is a geodesic which
connect p and q, then γ|[t0 ,t1 ] ≡ c so γ is regular.

3.3.19 Proposition. If a piecewise differentiable curve γ : [a, b] −→ M , with parameter proportional

to arc length, has length less or equal to the length of any other piecewise differentiable curve joining
γ(a), γ(b) then γ is a geodesic.

Proof. Let t ∈ [a, b] and W a normal neighborhood of γ(t), since γ(t) ∈ M , hence there exists
I0 ⊂ [a, b], a closed interval such that γ|I0 is a piecewise function defined from I0 −→ w joining two
points of the normal ball. Using the previous proposition, and due to the fact that γI0 is parametrized
by arc length since l(γ|I0 ) is equal to the radial geodesic; we can conclude that γ|I0 is a geodesic on I0 ,
hence it could be extended for all t ∈ [a, b]

This proposition holds to that, if a piecewise differentiable curve c is minimizing we can show that c is
a geodesic, and this lead to the fact that geodesic are regular.
4. Curvature in Riemannian manifold
The curvature on a Riemannian manifold was introduced by Riemann (see Riemann [8]), the objective
was to generalize the notion of Gaussian curvature that we know for surfaces. In what follows we will
work with the Riemannian manifold (M n , <, >), with its Levi-Civita connection.

4.1 Curvature tensor

Before reaching to the definition of curvature tensor, let us make a quick recap on the notion of tensor
on a Riemannian manifold.
4.1.1 Definition. A tensor T of order r on a Riemannian manifold is a D(M )-multilinear mapping,
defined as follows:
T : X(M ) × X(M ) × .... × X(M ) −→ D(M ). (4.1.1)
| {z }
r−factors

This means that given Z1 , ..., Zr , T (Z1 , ..., Zr ) is a differentiable function over M , and T is linear with
respect to each variables.
4.1.2 Definition. The curvature denoted by R, is a mapping:
R : X(M ) × X(M ) × X(M ) −→ X(M )
(X, Y, Z) −→ R(X, Y )Z
defined as follows:
R(X, Y )Z = ∇Y ∇X Z − ∇X ∇Y Z + ∇[X,Y ] Z; (4.1.2)
where ∇ is the Riemannian connection of M mentioned in the previous chapter 3 (Levi Civita connec-
tion).

This definition can be different from one book to another like in [4], the one we use here is the one
defined in [2]. But the idea behind this notion still the same. Let us find the curvature of the Riemannian
manifold when M = Rn : Z ∈ X(Rn ), it can be looked like Z = (z1 , ..., zn ) then from (4.1.2) we have :
∇Y ∇X Z = (Y Xz1 , ..., Y Xzn ) (4.1.3)
∇X ∇Y Z = (XY z1 , ..., XY zn ) (4.1.4)
Hence
∇Y ∇X Z − ∇X ∇Y Z = (Y Xz1 − XY z1 , ..., Y Xzn − XY zn )
= ((Y X − XY )z1 , ..., (Y X − XY )zn )
∇Y ∇X Z − ∇X ∇Y Z = −∇[X,Y ] Z.
Then
R(X, Y )Z = −∇[X,Y ] Z + ∇[X,Y ] Z
R(X, Y )Z = 0 ∀ X, Y, Z ∈ X(M ) =⇒ R ≡ 0. (4.1.5)
Hence due to (4.1.5), we see the curvature as a tool to measure how much a Riemannian manifold
deviate from being euclidean.

21
Section 4.1. Curvature tensor Page 22

4.1.3 Proposition. A curvature defined above has the property that it is a D(M )-tri-linear map. Hence
we can see it as follows:

• R is bilinear in X(M ) × X(M ).

• For all X, Y ∈ X(M ), R(X, Y ) linear from X(M ) −→ X(M ).

For the prove See Appendix A.0.2.

4.1.4 Remark. The appearance of ∇[X,Y ] is very relevant to make R(X, Y ) linear.

4.1.5 Definition. The curvature tensor is the 4−tensor, defined as follows:

(X, Y, Z, T ) =< R(X, Y )Z, T > ∀ X, Y, Z, T ∈ X(M ). (4.1.6)

NB It comes that (X, Y, Z, T ) is linear with respect to T , and since R(X, Y )Z is tri-linear, we then
conclude that (X, Y, Z, T ) is a 4− tensor. This curvature tensor satisfies some, properties that we are
going to see in the following proposition.

4.1.6 Proposition. The curvature tensor (X, Y, Z, T ) is:

1. Skew-symmetric with respect to the two first entries and the two last, which means:

(X, Y, Z, T ) = −(Y, X, Z, T ) (4.1.7)

(X, Y, Z, T ) = (Y, X, T, Z). (4.1.8)

2. Symmetric with respect to the two first entries and the two last one, which means:

(X, Y, Z, T ) = (T, Z, X, Y ). (4.1.9)

For the prove see Appendix A.0.2

4.1.7 Expression of the curvature and curvature tensor in terms of Christophel
 symbol Γij . Let
1 n ∂
us consider the chart (U, φ), with local coordinate (x , ..., x ), and ∂xi = Xi the basis. We
i=1,...,n
have:
R(Xi , Xj )Xk = ∇Xj ∇Xi Xk − ∇Xi ∇Xj Xk , since [Xi , Xj ] = 0.
Due to the fact that R(Xi , Xj )Xk ∈ X(M ), in term of local coordinate we have:
l
R(Xi , Xj )Xk = Rijk Xl (4.1.10)

Since ∇Xi Xk = Γsik Xs , ∇Xj Xk = Γsjk Xs , we have

R(Xi , Xj )Xk = ∇Xj (Γsik Xs ) − ∇Xi (Γsjk Xs )

Since
∇Xj (Γsik Xs ) = Xj (Γsik )Xs + Γsik ∇Xj Xs = Xj (Γsik )Xs + Γsik Γljs Xl
∇Xi (Γsjk Xs ) = Xi (Γsjk )Xs + Γsjk ∇Xi Xs = Xi (Γsjk )Xs + Γsjk Γlis Xl
Section 4.1. Curvature tensor Page 23

Then

R(Xi , Xj )Xk = Xj (Γsik )Xs + Γsik Γljs Xl − Xi (Γsjk )Xs − Γsjk Γlis Xl
= Xj (Γlik )Xl + Γsik Γljs Xl − Xi (Γljk )Xl − Γsjk Γlis Xl
 
R(Xi , Xj )Xk = Xj (Γlik ) + Γsik Γljs − Xi (Γljk ) − Γsjk Γlis Xl (4.1.11)
 
∂
Hence using (4.1.10) and (4.1.11) we get the components of curvature in ∂xi , which are given by:

l ∂ ∂
Rijk = Γsik Γljs − Γsjk Γlis + (Γl ) − (Γl ) (4.1.12)
∂xj ik ∂xi jk

Using (4.1.12), let us write the expression of the curvature tensor, we have:
l
< R(Xi , Xj )Xk , Xt > = < Rijk Xl , Xt >
l l
= Rijk < Xl , Xt >= Rijk glt
 
∂
Hence the components of the curvature tensor in ∂x i
are Rijkl , define by:

l
Rijkt = Rijk glt (4.1.13)

4.1.8 Remark.

1. The Proposition 4.1.6 in local coordinate can be written , as follow : -

• Skew-symmetric or antisymmetric : Rijkt = −Rjikt = Rjitk

• Symmetric : Rijkt = Rtkji

2. Due to the antisymmetric, we can see that if the two first entries or the two last entries are
identical, the corresponding components of the curvature tensor will vanish.ie

if i = j or k = t =⇒ Riikt = 0 or Rijkk = 0

3. this Proposition 4.1.6 is very useful when we want to calculate the components of the bending
tensor because it allows us to reduce the number of operations to perform and to deduce the
other components using the antisymmetric and symmetric properties of the bending tensor.

4.1.9 Examples.

1. The curvature in euclidean geometry is equal to zero, since R ≡ 0

2. The unit sphere S 2 of R3 , let us compute the components of the curvature tensor

S 2 = {(x, y, z) ∈ R3 |x2 + y 2 + z 2 = 1}

Let us consider the parametrization in polar coordinates:

h : (0, π) × [0, 2π] −→ S 2

(ϕ, θ) −→ (sinϕcosθ, sinϕsinθ, cosϕ)
Section 4.1. Curvature tensor Page 24

 
∂ ∂ gϕϕ gϕθ
For a given point p, we have Tp S 2 =span{ ∂ϕ (p), ∂θ (p)}, and then noted by g = the
gϕθ gθθ
∂ ∂ ∂ ∂
matrix associate to the metric at p, then we have gϕϕ =< ∂ϕ (p), ∂ϕ (p) >, gθθ =< ∂θ (p), ∂θ (p) >
∂ ∂
, gϕθ = gθϕ =< ∂ϕ (p), ∂θ (p) >. We have:

∂
(p) = (cosϕcosθ, cosϕsinθ, −sinϕ)
∂ϕ
∂
(p) = (−sinϕsinθ, sinϕcosθ, 0)
∂θ
   
2 1 0 −1 1 0
Hence gϕϕ = 1, gθθ = sin ϕ, gϕθ = 0, so g = and then g = 1 .
0 sin2 ϕ 0 sin2 ϕ

∂ ∂ ∂ ∂
. In what follow let us identified ∂ϕ (p) = ∂x1 (p), ∂θ (p) = ∂x2 (p), gϕϕ = g11 , gθθ = g22 , gϕθ = g12 .

Using (3.2.10) of chapter 2 section 2, let us find the christophel symbols Γ111 , Γ122 , Γ112 , Γ212 , Γ211 , Γ222
associate to the connection of S 2 :
1 1l
Γ111 = g (g1l,1 + g1l,1 − g11,l ) = 0 = Γ112 = Γ121 Since g11 = 1, g 12 = 0, g21 = 0
2
1 2l
Γ211 = g (g1l,1 + g1l,1 − g11,l ) = 0 Since g11 = 1, g 21 = 0
2
1 2l
Γ222 = g (g2l,2 + g2l,2 − g22,l ) = 0 Since g22 depend on ϕ, g 12 = 0
2
1 1l 1 1 ∂ 1
Γ122 = sin2 ϕ = − sin(2ϕ)


g (g2l,2 + g2l,2 − g22,l ) = − g22,1 = −
2 2 2 ∂ϕ 2
1 2l 1 1
Γ212 = g (g2l,1 + g1l,2 − g12,l ) = g 22 g22,1 = 2sinϕcosϕ = cotϕ = Γ221
2 2 2sin2 ϕ
l g . They are 16 terms to find, due to remark4.1.8 we
Components of the curvature tensor Rijkt = Rijk lt
just need to find 8 of them. Using the same remark they are 7 terms many which vanish, except R2121
1 2 1
R2121 = R212 g11 + R212 g21 = R212 g11 since g21 = 0

1
Using (4.1.12), let us find R212

1 ∂ ∂
R212 = Γs22 Γ11s − Γs12 Γ12s + (Γ122 ) − (Γ1 )
∂x1 ∂x2 12
∂
= Γ122 Γ111 + Γ222 Γ112 − Γ112 Γ121 − Γ212 Γ122 + (Γ1 )
∂x1 22
Since Γ111 = Γ112 = 0
1 ∂
R212 = −Γ212 Γ122 + (Γ1 )
∂x1 22
   
sin2ϕ cosϕ ∂ sin2ϕ
= − − + −
2 sinϕ ∂x1 2
= cos2 ϕ − cos2ϕ = cos2 ϕ − cos2 ϕ + sin2 ϕ
1
R212 = sin2 ϕ

So R2121 = sin2 ϕ
Section 4.1. Curvature tensor Page 25

∂ ∂
Hence the components of the curvature tensor in local frame ∂ϕ (p), ∂θ (p) are :

R2121 = sin2 ϕ, R1221 = −sin2 ϕ, R2112 = −sin2 ϕ, R1212 = sin2 ϕ

and all others terms vanish.

4.1.10 Proposition. The curvature and the curvature tensor are invariant under local isometries. Which
means that given two Riemannian manifolds M, M 0 and φ : M −→ M 0 , a local isometries, we have:

(φ∗ X, φ∗ Y, φ∗ Z, φ∗ T ) = (X, Y, Z, T ) ∀X, Y, Z, T ∈ X(M ) (4.1.14)

where φ∗ denoted the differential map between X(M ) and X(M 0 )

4.1.11 Lemma. Let M and M 0 be two Riemannian manifolds and φ : M −→ M 0 an isometric, and
∇ a Riemannian connection on M 0 such that the following diagram commutes:

X(M ) × X(M )
∇ / X(M )

φ∗ ×φ∗ φ∗
 
X(M 0 ) × X(M 0 ) / X(M 0 )
∇

then ∇ is the Riemannian connection on M, and

φ ∗ ∇ X Y = ∇ φ∗ X φ ∗ Y ∀X, Y ∈ X(M ) (4.1.15)

(4.1.15) come to the fact that the above diagram is commutative, because: φ∗ ∇ = ∇φ∗ φ∗ , and then
taken X, Y ∈ X(M ) we have:

• For the right side of the equality: φ∗ ∇(X, Y ) = φ∗ ∇X Y .

• For left side we have: ∇φ∗ φ∗ (X, Y ) = ∇(φ∗ X, φ∗ Y ) = ∇φ∗ X φ∗ Y .

Hence equal the right path and the left path we have (4.1.15). Let now prove proposal 4.1.10.

Proof. We have:

(φ∗ X, φ∗ Y, φ∗ Z, φ∗ T ) = < R(φ∗ X, φ∗ Y )φ∗ Z, φ∗ T >

Let us evaluate the quantity R(φ∗ X, φ∗ Y )φ∗ Z:

R(φ∗ X, φ∗ Y )φ∗ Z = ∇φ∗ Y ∇φ∗ X (φ∗ Z) − ∇φ∗ X ∇φ∗ Y (φ∗ Z) + ∇[φ∗ X,φ∗ Y ] φ∗ Z

Then using (4.1.15) twice on the expressions ∇φ∗ Y ∇φ∗ X (φ∗ Z), ∇φ∗ X ∇φ∗ Y (φ∗ Z) we have :

∇φ∗ Y ∇φ∗ X (φ∗ Z) = ∇φ∗ Y (φ∗ ∇X Z) = φ∗ ∇Y ∇X Z

∇φ∗ X ∇φ∗ Y (φ∗ Z) = ∇φ∗ X (φ∗ ∇Y Z) = φ∗ ∇X ∇Y Z

Section 4.2. Sectional curvature Page 26

So

R(φ∗ X, φ∗ Y )φ∗ Z = φ∗ ∇Y ∇X Z − φ∗ ∇X ∇Y Z + ∇[φ∗ X,φ∗ Y ] φ∗ Z

= φ∗ ∇Y ∇X Z − φ∗ ∇X ∇Y Z + ∇φ∗ [X,Y ] φ∗ Z
= φ∗ ∇Y ∇X Z − φ∗ ∇X ∇Y Z + φ∗ ∇[X,Y ] Z
R(φ∗ X, φ∗ Y )φ∗ Z = φ∗ R(X, Y )Z (4.1.16)

Hence using (4.1.16), we get

(φ∗ X, φ∗ Y, φ∗ Z, φ∗ T ) = < φ∗ R(X, Y )Z, φ∗ T >

= < R(X, Y )Z, T > , since φ is an isometric
(φ∗ X, φ∗ Y, φ∗ Z, φ∗ T ) = (X, Y, Z, T )

4.2 Sectional curvature

This notion is likely correlated to the notion of curvature tensor because, given the sectional curvature
for all σ ⊂ Tp M , (σ a subspace of dimension2), we can define curvature tensor as we are going to see.
Let us consider a vector space V, given x, y ∈ V, the area of the parallelogram generated by x, y, can
be express as follow: p
|x ∧ y| = |x|2 .|y|2 − < x, y >2 . (4.2.1)
In fact (4.2.1) come directly from the definition.
4.2.1 Definition. Given a point p ∈ M , a two dimensional subspace σ ⊂ Tp M , generated by x, y two
linearly independent tangent vectors in Tp M . The sectional curvature of σ at p; is a real number
denoted K(p, σ) defined as follow:
(x, y, x, y)
K(p, σ) = K(x, y) = . (4.2.2)
|x ∧ y|2
4.2.2 Proposition. K(p, σ) = K(x, y) does not depend on the choice of the basis which generated σ.
 
a b
Proof. Let {x0 , y 0 }
another basis of σ, so it exist a transition matrix P = (inversible ma-
c d
(
x0 = ax + by
trix) between the basis {x, y} and {x0 , y 0 } in such a way that ; let us show :
y 0 = cx + dy
K(x0 , y 0 ) = K(x, y). We have
(ax + by, cx + dy, ax + by, cx + dy)
K(x0 , y 0 ) =
|(ax + by) ∧ (cx + dy)|2
Let’s decompose the denominator,

|(ax + by) ∧ (cx + dy)|2 = |ax ∧ (cx + dy) + by ∧ (cx + dy)|2

= |ad(x ∧ y) + bc(y ∧ x)|2 ; since ax ∧ cx = 0 = by ∧ dy
= |ad(x ∧ y) − bc(x ∧ y)|2
|(ax + by) ∧ (cx + dy)| = (ad − bc)2 |x ∧ y|2 . (4.2.3)
Section 4.2. Sectional curvature Page 27

Let’s decompose the numerator, by using the fact that the curvature tensor is a 4− tensor, which as
some properties (see Proposition 4.1.6 )

(ax + by, cx + dy, ax + by, cx + dy) = (ax, cx + dy, ax + by, cx + dy) + (by, cx + dy, ax + by, cx + dy)
= (ax, dy, ax + by, cx + dy) + (by, cx, ax + by, cx + dy)
= (ax, dy, ax, cx + dy) + (ax, dy, by, cx + dy) + (by, cx, ax, cx + dy)
+(by, cx, by, cx + dy)
= (ax, dy, ax, dy) + (ax, dy, by, cx) + (by, cx, ax, dy) + (by, cx, by, cx)
= (ax, dy, ax, dy) − (ax, dy, cx, by) − (cx, by, ax, dy) + (cx, by, cx, by)
= (ad)2 (x, y, x, y) − 2adcb(x, y, x, y) + (cb)2 (x, y, x, y)
= ((ad)2 − 2adcb + (cb)2 )(x, y, x, y)
(ax + by, cx + dy, ax + by, cx + dy) = (ad − cb)2 (x, y, x, y). (4.2.4)

Hence using (4.2.3) and (4.2.4), we have:

(ad − cb)2 (x, y, x, y) (x, y, x, y)

K(x0 , y 0 ) = 2 2
= = K(x, y).
(ad − bc) |x ∧ y| |x ∧ y|2

4.2.3 Remark.

1. The sectional curvature at a point p is not a 2− tensor, because we have:

K(p, σ) = K(2x, y) = K(x, y) 6= 2K(x, y)

2. Definition 4.2.1, allows us to define at a point p ∈ M the sectional curvature of a plane Π

spanX, Y in Tp M as the Gaussian curvature, that we are going to define now.

Consider p ∈ M , and X, Y ∈ Tp M , since Tp M ∼ ∂

= Rn , an knowing that Tp M =Span{ ∂x i
}, we have
i ∂ j ∂
X = x ∂xi , Y = y ∂xj , and then we get:

(X, Y, X, Y )
K(X, Y ) =
|X ∧ Y |2
(X, Y, X, Y )
=
< X, X > . < Y, Y > − < X, Y >2
 
∂ j ∂ , xk ∂ , y l ∂
xi ∂x i
, y ∂xj ∂xk ∂xl
= ∂
< xi ∂x i
, xk ∂x∂ k > . < y j ∂x∂ j , y l ∂x
∂
l
∂
> − < xi ∂x i
, y j ∂x∂ j > . < xk ∂x∂ k , y l ∂x
∂
l
>
Rijkl xi y j xk y l
=
xi xk gik y j y l gjl − xi y j xk y l gij gkl
Rijkl xi y j xk y l
K(X, Y ) = . (4.2.5)
(gik gjl − gij gkl )xi y j xk y l

Then (4.2.5) give us the sectional curvature of Π.

Section 4.2. Sectional curvature Page 28

4.2.4 Remarks.

1. From (4.2.5), we can see that the sectional curvature of a plan generated by two independent
linear vectors is a ratio of two tensors.

2. Directly for (4.2.5), if we multiply the metric by a constant α, the Christophel symbol will remain
the same because it also involves the inverse of the matrix associate to the metric, hence by the
formula 4.1.12 we see that the components of the curvature will not change, and due to formula
4.1.13, we can see that the components of the curvature will be multiply be this α, since the
denominator of (4.2.5) will be scale by α2 , then the sectional curvature of the plane generated by
X, Y will be scale by α−1 .

3. Knowing that the denominator is the area of the parallelogram spanned by X, Y , which is twice
the area of the triangle with vertex 0, X, Y . The sectional curvature expresses the deviation of
the area of a triangle in the manifold from that in the tangent space.

4. The sectional curvature K(X, Y ) can then be shown to be the Gauss curvature of the surface Π
generated by X, Y with the induced metric.

4.2.5 Example. Computation of the sectional curvature

• Sphere S 2
Given a point p ∈ S 2 , Tp S 2 ∼
= R2 , hence taken σ ⊂ Tp S 2 , spanned by X, Y , element of Tp S 2 .
So σ is a subspace of 2-dimension include in a 2-dimension vector space, then σ = Tp M , which
is generated by e1 , e2 canonical basis of R2 . Then:
R1212
K(p, Tp S 2 ) = K(e1 , e2 ) =
g11 g22 − g12 g12
sin2 ϕ
=
sin2 ϕ
K(p, Tp M ) = 1

• The upper half-plane

R2+ = {(x, y)R2 |; y > 0}
The metric here is the metric of Lobatchevski’s non-euclidean geometry, given by
1
g11 = g22 = , g12 = g21
y2
After performing operations we obtain the Christophel symbol of the connection is given by
1 1
Γ111 = Γ212 = Γ221 = Γ122 = 0, Γ211 = , Γ112 = Γ121 = Γ222 = −
y y
The components of the curvature tensor all vanish, except R1212 = R2121 = −R2112 = −R1221 .
Section 4.2. Sectional curvature Page 29

We have:
1 2
R2121 = R212 g11 + R212 g21
1
= R212 g11 since g21 = 0
∂ 1 ∂ 1
= Γs22 Γ11s − Γs12 Γ12s + (Γ22 ) − (Γ )
∂x ∂y 12
∂ 1
= Γ122 Γ111 + Γ222 Γ112 − Γ112 Γ121 − Γ212 Γ122 − (Γ )
∂y 12
∂ 1
= Γ222 Γ112 − Γ112 Γ121 − (Γ )
∂y 12
1
R2121 = − .
y4

At a point p ∈ R2+ , Tp R2+ = R2 , then:

R1212
K(p, Tp R2+ ) =
g11 g22 − g12 g12
− y14
= 1
y4
K(p, Tp R2+ ) = −1

From Example 4.2.5 , we can see that both of those Riemannian manifolds have their sectional curvatures
are constants, but it is not all Riemannian manifold with their Levi-Civita whose as this property, now
we are looking for properties which could be useful to know such types of Riemannian manifolds.

4.2.6 Proposition. Consider M a Riemanman manifold and p a point of M , and a tri-linear mapping
R0 : Tp M × Tp M × Tp M −→ Tp M , such that

(R0 (X, Y, W ), Z) = (X, Y, W, Z)0 =< X, W >< Y, Z > − < Y, W >< X, Z > (4.2.6)

for all X, Y, W, Z ∈ Tp M . Then M has constant sectional curvature equal to K0 if and only if R = K0 R0 ,
where R is the curvature of M .

First of all by the definition of R0 , If in (4.2.6):

• We change the role of X and Y we get the antisymmetric of 4−tensor ( , , , )0 respect to the
two first entries

(Y, X, W, Z) =< Y, W >< X, Z > − < X, W >< Y, Z >= −(X, Y, W, Z)0

• We change the role of W and Z, we get the antisymmetric of 4−tensor ( , , , )0 respect to the
two last entries.

(X, Y, Z, W )0 =< X, Z >< Y, W > − < Y, Z >< X, W >= (X, Y, W, Z)0

• We use the symmetric of the inner product we have the symmetric of 4−tensor ( , , , )0 .

Hence ( , , , )0 satisfies proposition A.0.5, so it a curvature tensor on M .

Section 4.3. Ricci curvature and scalar curvature. Page 30

Proof. We have : (X, Y, X, Y )0 =< X, X >< Y, Y > − < X, Y >2 , suppose M has a constant
sectional curvature equal to K0 , let p ∈ M , and σ ∈ Tp M generated by X, Y , then

< R(X, Y )X, Y >

K(p, σ) = K0 = =⇒ < R(X, Y )X, Y >= K0 (X, Y, X, Y )0
|X|2 |Y |2 − < X, Y >2

Since knowing the sectional curvature for all σ ∈ Tp M , permit us to define the curvature tensor, then
for all X, Y, W, Z we have

(R(X, Y )W, Z) = K0 (R0 (X, Y, W ), Z) =⇒ R(X, Y )W = K0 R0 (X, Y, W )

then R = K0 R0 . the converse it obvious it come directly from the calculation.

To make a fee summary about sectional curvature, we can see that this notion given at a point p doesn’t
depends on the basic which generated the plane include in the tangent space at p, Hence if we change
the plane we also change the sectional curvature. Knowing that if dim(Tp M ) = n there are n(n−1) 2
independent planes of dimension two, so n(n−1)2 sectional curvatures to compute at point p. Hence
some of them appear with such frequencies that they deserve special names, which reach us to the
notion of Ricci curvature and scalar curvature.

4.3 Ricci curvature and scalar curvature.

4.3.1 Definition. Let X = Zn be a unit vector in Tp M ; we take an orthonormal basis Z1 , Z2 , ..., Zn − i

of the hyperplane in Tp M orthogonal to X, the Ricci curvature in the direction X at p denoted
Ricp (X), is the mean of the tensor curvature (X, Zi , X, Zi )i∈1,...,n−1 , it is given by:

n−1
1 X
Ricp (X) = (X, Zi , X, Zi ) (4.3.1)
n−1
i=1

4.3.2 Definition. The scalar curvature at p denoted by k(p) is the mean of the Ricci curvature at p
in the directions Zj j=1,...,n at p :
n
1X
k(p) = Ricp (Zj ) (4.3.2)
n
j

The definition 4.3.1 can be also write in function of sectional curvature,

n−1
1 X
Ricp (X) = Kp (X, Zi ).|X ∧ Zi |
n−1
i=1

then it is invariant in the choose of orthogonal basis.

5. Conclusion
Our main purpose in this work was to provide some basic concepts in Riemannian geometry. For that we
have defined first some basic notions on differential geometry, followed by the definition of Riemannian
manifold. we have proved the existence of a unique connection on Riemannian manifold: the Levi-Civita
connection. Then we have introduced one of the fundamental notions of the Riemannian geometry:
geodesics. we have proved that geodesics are solutions of an ODE of second-order, and that the solution
is unique if we have specified the initial condition. We also showed some properties of minimizing
geodesic and then proved the regularity of geodesics. We ended introducing the second fundamental
notion: curvature. We proved that defining the sectional curvature on all subspace of dimensional two,
it can be sufficient to define the curvature tensor. Therefore, we moved to the properties of minimizing
geodesic that we can use to define a distance on the Riemannian manifold. We conclude studying the
topology induced by the distance and introducing the Hopf-Rinow Theorem.

31
Appendix A. Some additional data
A.0.1 Proof of Koszul formula. Lemma 3.2.14

Proof.

X(< Y, Z >) = < ∇X Y, Z > + < Y, ∇X Z > (A.0.1)

Y (< X, Z >) = < ∇Y X, Z > + < X, ∇Y Z > (A.0.2)
Z(< X, Y >) = < ∇Z X, Y > + < X, ∇Z Y > (A.0.3)

Hence adding (A.0.1) with (A.0.2) and subtract it to (A.0.3), we have:

X(< Y, Z >) + Y (< X, Z >) − Z(< X, Y >) = < ∇X Y, Z > + < Y, ∇X Z > + < ∇Y X, Z >
− < Z, ∇X Y > + < Z, ∇X Y > + < X, ∇Y Z >
− < ∇Z X, Y > − < X, ∇Z Y >
= < ∇X Y, Z > + < Y, ∇X Z > − < ∇Z X, Y >
+ < X, ∇Y Z > − < X, ∇Z Y > + < ∇Y X, Z >
− < Z, ∇X Y >
Due to the symmetry of <, >;
X(< Y, Z >) + Y (< X, Z >) − Z(< X, Y >) = 2 < ∇X Y, Z > + < Y, ∇X Z > − < Y, ∇Z X >
+ < X, ∇Y Z > − < X, ∇Z Y > + < Z, ∇Y X >
− < Z, ∇X Y >
X(< Y, Z >) + Y (< X, Z >) − Z(< X, Y >) = 2 < ∇X Y, Z > + < Y, ∇X Z − ∇Z X >
+ < X, ∇Y Z − ∇Z Y > + < Z, ∇Y X − ∇X Y >

Since ∇ is symmetric; ∇X Z − ∇Z X = [X, Z] , ∇Y Z − ∇Z Y = [Y, Z] , ∇Y X − ∇X Y = [Y, X]

X(< Y, Z >) + Y (< X, Z >) − Z(< X, Y >) = 2 < ∇X Y, Z > + < Y, [X, Z] > + < X, [Y, Z] >
+ < Z, [Y, X] >

Then

2 < ∇X Y, Z >= X(< Y, Z >)+Y (< X, Z >)−Z(< X, Y >)− < Y, [X, Z] > − < X, [Y, Z] > − < Z, [Y, X] >

This ends the proof of the lemma.

A.0.2 Proof of Proposition 4.1.6 .

Proof. 1. Skew-symmetric respect to the two first entries and the two last entries.

• Skew-symmetric respect to the two first entries is obvious, since R(X, Y )Z = −R(Y, X)Z
then (X, Y, Z, T ) =< R(X, Y )Z, T >=< −R(Y, X)Z, T >= −(Y, X, Z, T ), which shows
(4.1.7).

32
 Page 33

• Skew-symmetric respect to the two last entries

(X, Y, Z, T ) = (Y, X, T, Z) ⇐⇒ (X, Y, Z, T ) = −(X, Y, T, Z)

Taken T = Z
⇐⇒ (X, Y, Z, Z) = −(X, Y, Z, Z)
⇐⇒ (X, Y, Z, Z) = 0

Hence (4.1.8) ⇐⇒ (X, Y, Z, Z) = 0. We have:

(X, Y, Z, Z) = < R(X, Y )Z, Z >

= < ∇Y ∇X Z, Z > − < ∇X ∇Y Z, Z > + < ∇[X,Y ] Z, Z >

using the fact that the connection is compatible, we have

< ∇Y ∇X Z, Z > = Y < ∇X Z, Z > − < ∇X , ∇Y Z >

− < ∇X ∇Y Z, Z > = −X < ∇Y Z, Z > + < ∇Y Z, ∇X Z >
< ∇[X,Y ] Z, Z > = [X, Y ] < Z, Z > − < ∇[X,Y ] Z, Z > this implies
1
< ∇[X,Y ] Z, Z > = [X, Y ] < Z, Z >
2
Summing those expressions implies
1
(X, Y, Z, Z) = Y < ∇X Z, Z > −X < ∇Y Z, Z > + [X, Y ] < Z, Z > (A.0.4)
2
Since
1
< ∇X Z, Z > = X < Z, Z > − < Z, ∇X Z > =⇒ < ∇X Z, Z >= X < Z, Z >
2
1
< ∇Y Z, Z > = Y < Z, Z >
2
Replacing < ∇X Z, Z >, < ∇Y Z, Z > by their above expressions in equationA.0.4, we have:
1 1
(X, Y, Z, Z) = ((Y X − XY ) < Z, Z >) + [X, Y ] < Z, Z >
2 2
1 1
= − [X, Y ] < Z, Z > + [X, Y ] < Z, Z >
2 2
(X, Y, Z, Z) = 0.

Which ends the anti-symmetric respect to the last two entries.

2. Symmetric
Bianchi identity: R(X, Y )Z + R(Y, Z)X + R(Z, X)Y = 0. Using this identity we have directly
< R(X, Y )Z + R(Y, Z)X + R(Z, X)Y, T >= 0., and this implies (X, Y, Z, T ) + (Y, Z, X, T ) +
(Z, X, Y, T ) = 0. Then using this final expression by exchanging the position of the vector fields
X, Y, Z, T 4 times, and make the summation of the 4 equations that we will have, we will get
(4.1.9)

Prove of Proposition 4.1.3

 Page 34

• Bi-linearity in X(M ) × X(M ).

Let X1 , X2 , Y1 , Y2 ∈ X(M ), and f ∈ D(M ). We need to show that:

R(f X1 + X2 , Y1 ) = f R(X1 , Y1 ) + R(X2 , Y1 ) (A.0.5)

R(X1 , f Y1 + Y2 ) = f R(X1 , Y1 ) + R(X2 , Y1 ). (A.0.6)

We have:

R(f X1 + X2 , Y1 ) = ∇Y1 ∇f X1 +X2 − ∇f X1 +X2 ∇Y1 + ∇[f X1 +X2 ,Y1 ]

= ∇Y1 (f ∇X1 + ∇X2 ) − (f ∇X1 + ∇X2 ) ∇Y1 + ∇[f X1 +X2 ,Y1 ] .

Since [f X1 + X2 , Y1 ] = f [X1 , Y1 ] + [X2 , Y1 ] − Y1 (f )X1 ;

R(f X1 + X2 , Y1 ) = f ∇Y1 ∇X1 + Y1 (f )∇X1 + ∇Y1 ∇X2 − f ∇X1 ∇Y1 − ∇X2 ∇Y1
+∇f [X1 ,Y1 ]+[X2 ,Y1 ]−Y1 (f )X1
= f ∇Y1 ∇X1 + Y1 (f )∇X1 + ∇Y1 ∇X2 − f ∇X1 ∇Y1 − ∇X2 ∇Y1
+∇f [X1 ,Y1 ] + ∇[X2 ,Y1 ] − ∇Y1 (f )X1
= f ∇Y1 ∇X1 + Y1 (f )∇X1 − f ∇X1 ∇Y1 − ∇X2 ∇Y1 + ∇Y1 ∇X2
+f ∇[X1 ,Y1 ] + ∇[X2 ,Y1 ] − Y1 (f )∇X1
= f ∇Y1 ∇X1 − f ∇X1 ∇Y1 + f ∇[X1 ,Y1 ] + ∇Y1 ∇X2 − ∇X2 ∇Y1 + ∇[X2 ,Y1 ]
R(f X1 + X2 , Y1 ) = f R(X1 , Y1 ) + R(X2 , Y1 )

This shows relation (A.0.5). In this previous development, the sixth line comes by putting together
the first, third, and fifth terms of the expression in line fifth, and cancel the second terms with
the last one.
To show relation (A.0.6), we can see that R(X1 , f Y1 + Y2 ) = −R(f Y1 + Y2 , X1 ), and using the
previous result with X1 = X2 , we get

R(X1 , f Y1 + Y2 ) = −f R(Y1 , X1 ) − R(Y1 , X1 )) = f R(X1 , Y1 ) + R(X1 , Y1 ).

So R is Bi-lineary in X(M ) × X(M ).

• Let X, Y ∈ X(M ) and R(X, Y ) a linear map from X(M ) −→ X(M ).

Let Z1 , Z2 , Z ∈ X(M ), and f ∈ D(M ), we need to show that:

R(X, Y )(Z1 + Z2 ) = R(X, Y )(Z1 ) + R(X, Y )(Z2 ) (A.0.7)

R(X, Y )(f Z) = f R(X, Y )(Z). (A.0.8)

We have :

R(X, Y )(Z1 + Z2 ) = ∇Y ∇X (Z1 + Z2 ) − ∇X ∇Y (Z1 + Z2 ) + ∇[X,Y ] (Z1 + Z2 )

= ∇Y ∇X Z1 − ∇X ∇Y Z1 + ∇[X,Y ] Z1
+∇Y ∇X Z2 − ∇X ∇Y Z2 + ∇[X,Y ] Z2
R(X, Y )(Z1 + Z2 ) = R(X, Y )(Z1 ) + R(X, Y )(Z2 ).

This shows (A.0.7);

R(X, Y )(f Z) = ∇Y ∇X (f Z) − ∇X ∇Y (f Z) + ∇[X,Y ] (f Z).

 Page 35

Since:

∇Y ∇X (f Z) = Y X(f )Z + X(f )∇Y Z + f ∇Y ∇X Z + Y (f )∇X Z (A.0.9)

∇X ∇Y (f Z) = XY (f )Z + Y (f )∇X Z + X(f )∇Y Z + f ∇X ∇Y Z (A.0.10)
∇[X,Y ] (f Z) = [X, Y ](f )Z + f ∇[X,Y ] (Z) (A.0.11)
(A.0.9) − (A.0.10) = [Y, X](f )Z + f (∇Y ∇X − ∇X ∇Y )Z. (A.0.12)

Hence summing (A.0.12) with (A.0.11), we have :

R(X, Y )(f Z) = [Y, X](f )Z + f (∇Y ∇X − ∇X ∇Y )Z + [X, Y ](f )Z + f ∇[X,Y ] (Z)

= −[X, Y ](f )Z + f (∇Y ∇X − ∇X ∇Y )Z + [X, Y ](f )Z + f ∇[X,Y ] (Z)
Since the first and third terms will cancel in this previous expression;
= f (∇Y ∇X Z − ∇X ∇Y Z + ∇[X,Y ] Z)
R(X, Y )(f Z) = f R(X, Y )(Z).

This shows (A.0.8), and ends the proof of the proposition.

(a) Curve γ with the parallel field (b) Transition map [10].

A.0.3 Corollary 3.2.16 .

Proof.

1. Let us prove that Γm m

ij = Γji .
Since we are working with the Levi-Civita connection, using the first condition of Theorem 3.2.13,
with X = Xi , Y = Xj we have :

∇Xi Xj − ∇Xj Xi = [Xi , Xj ] . (A.0.13)

Let f ∈ D(M ), and consider the restriction of f to U , evaluating the right hand side of (A.0.13),
we have:

[Xi , Xj ] f = Xi Xj (f ) − Xj Xi (f )
∂ ∂f ∂ ∂f
= (p) − (p)
∂xi ∂xj ∂xj ∂xi
= 0 ,since f ∈ D(M ) we can swap the order of differentiation
[Xi , Xj ] f = 0 , for all f =⇒ [Xi , Xj ] = 0.

(A.0.13), becomes:

∇Xi Xj − ∇Xj Xi = 0 (A.0.14)

 Page 36

(A.0.14) =⇒ Γm m
ij Xm − Γji Xm = 0
=⇒ Γm m
ij = Γji .

This ends the proof.

2. Using (3.2.9) of Lemma 3.2.14, with X = Xi , Y = Xj , Z = Xl , we have

2 < ∇Xi Xj , Xl >= Xi (< Xj , Xl >) + Xj (< Xi , Xl >) − Xl (< Xi , Xj >), (A.0.15)

because [Xi , Xl ] = [Xj , Xl ] = [Xl , Xi ] = 0.

(A.0.15) =⇒ 2 < Γm ij Xm , Xl >= Xi (gjl ) + Xj (gil ) − Xl (gij ),

∂ ∂ ∂
=⇒ 2Γmij < Xm , Xl >= (gjl ) + (gil ) − (gij ),
∂xi ∂xj ∂xl
=⇒ 2Γmij gml = gjl,i + gil,j − gij,l ,
1 −1
=⇒ Γm
ij = (gml ) (gjl,i + gil,j − gij,l ) ,
2
1 ml
=⇒ Γm
ij = g (gjl,i + gil,j − gij,l ) ,
2
where g ml are components of the inverse matrix associate to inner product. This ends the proof

A.0.4 Proof of the Lemma 3.2.4.

Proof. We are going to use Einstein summation convention. Let p ∈ M , consider the local coordinate
∂
around p = (x1 , ..., xn ), and the local basis (Xi )i=1,..,n of Tp M , where Xi = ∂x i
. Taking X, Y ∈ Tp M ,
i j i j
we can write X = x Xi , Y = y Xj , where x , y are differentiable functions, let us introduce the
Christophel symbols by:

∇Xi Xj = Γkij Xk . (A.0.16)

where Γkij are differential function, which are components of the connection in the local basis, then we
have

∇X Y = ∇xi Xi Y
= xi ∇Xi Y
= xi ∇Xi (y j Xj )
= xi Xi (y j )Xj + xi y j ∇Xi Xj
∇X Y = xi ∇Xi (y j )Xj + xi y j Γkij Xk using (A.0.16).

Since xi ∇Xi (y j ) = X(y j )

∇X Y = X(y j )Xj + xi y j Γkij Xk = X(y k )Xk + Γkij Xk

 
∇X Y = X(y k ) + xi y j Γkij Xk . (A.0.17)

Hence by (A.0.17), we see that ∇X Y depend on the values of X(y k )(p) which is the derivative of y k
by X at the point p, and also xi (p), y j (p).
Acknowledgements
I cool not finish this without thank all my supervisors: Nicolau Sarquis, whose accept the follow me
during all this essay and also Dr Diletta .

37
References
[1] Aglio, L. D. and isreal, G. (1987). Stability and qualitative analysis. Pitagora, Bologna.

[2] do Carmo, M. (1992). Riemannian Geometry. Birkhauser Boston.

[3] Figaro, Alessio, V. (2011). Non linear PDE’s and application. Springer.

[4] Jost, J. (2017). Riemannian Geometry and Geometry Analysis. Springer International Publishing
AG 2017.

[5] Kriele, M. (1999). Analysis on manifolds. Spacetime: Foundations of General Relativity and Differ-
ential Geometry, pages 47–149.

[6] Lott, J. and Villani, C. (2009). Ricci curvature for metric-measure spaces via optimal transport.
Annals of Mathematics, pages 903–991.

[7] Petersen, P. (1998). Riemannian Geometry. Springer New York.

[8] RAUCH, H. (1959). Geodesies and Curvature in Differential Geometry in the Large. Yeshiva Uni-
versity, N.Y.

[9] Ricci-Curbastro, G., Hermann, R., Ricci, M., and Levi-Civita, T. (1975). Ricci and Levi-Civita’s
Tensor Analysis Paper: Translation, Comments, and Additional Material. Number 2. Math Science
Press.

[10] Siemssen, D. (2015). The semiclassical einstein equation on cosmological spacetimes. arXiv preprint
arXiv:1503.01826.

[11] Zettl, A. (2010). Sturm-liouville theory. Number 121. American Mathematical Soc.

38