Géométrie Semi-Riemannienne: Loubeau
Géométrie Semi-Riemannienne: Loubeau
Éric Loubeau
Master 2 de mathématiques fondamentales · Université de Rennes
Notes prises par Téofil Adamski (version du 5 avril 2023)
i
Sommaire
1 Differentiable manifolds
1.1 Differentiable manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Tangent spaces and tangent bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Semi-riemannian manifolds
2.1 First definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Connection and Levi-Civita connection . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Curvature and Ricci tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Killing vector field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Geodesics
3.1 First definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4 Examples
5 Calculus of variations
1
Chapitre 1
Differentiable manifolds
where the maps xi are called the coordinates functions of ξ and we will denote ξ = (x1 , . . . , xn ).
Two charts ξ and η of dimension n intersect in a smooth manner if the maps ξ ◦ η −1 and η ◦ ξ −1
are of classe C ∞ .
An atlas is a collection of charts of dimension n such that
An atlas is complete if it contains all the charts of S which intersect in a smooth manner. Any atlas
admits a completion.
Définition 1.1. A differentiable manifold is a topological space equipped with a complete atlas.
If f ∈ F (M ), we denote V (f )(p) := Vp (f ).
with v i ∈ R. We consider
π −1 (U ) ⊂ TM −→ R2n ,
η̃ :
(p, u) 7−→ (x1 (π(p, u)), . . . , xn (π(p, u)), v 1 , . . . , v n ).
where v i = v(xi ) =: ẋi (u). This defines a atlas on TM . If (u1 , . . . , u2n ) are the coordinates on R2n ,
the transition functions are given by
ui ξ˜ ◦ η̃ −1 = xi ◦ π ◦ η̃ −1 (a, b) = xi η −1 (a),
X ∂xi
ui ξ˜ ◦ η̃ −1 = ẋi ◦ η̃ −1 (a, b) = bk k (η −1 (a)).
∂y
So the map ξ˜ ◦ η̃ −1 are of class C ∞ .
We have
dγ
= γ̇(0) ∈ Tp S2
dt t=0
and all tangent vector can be obtained this way. As |γ| = 1, we find
d
|γ(t)|2 = 0 = 2⟨p, γ̇(0)⟩.
dt
Thus we conclude
Tp S2 = {X ∈ R3 | ⟨X, p⟩ = 0}.
Exemples. Open subsets of Rn are differentiable manifolds. The half-plane H2 := R × R∗+ has the
tangent space Tp H2 = R2 and so its tangle bundle is TH2 = H2 × R2 .
– The map
R −→ R2 ,
t 7−→ (t3 − 4t, t2 − 4)
is differentiable and an immersion, but there is a self-intersection sot it is not an embedding.
– The map t 7−→ (t, sin(1/t)) is an immersion with no self-intersecting point, but it is not an
embedding.
– The cone x2 + y 2 − z 2 = 0 is not a submanifold of R3 for connectivity reasons.
1.3. Tensors
Let V be a module over a ring K.
Définition 1.10. Let r, s ∈ N be integers with rs > 0. A tensor of type (r, s) is a K-multilinear
function
(V ∗ )r × V s −→ K.
We denote Tr,s (V ) the set of tensors of type (r, s).
A tensor field is a tensor on the ring X (M ) which denotes the set of vectors field on a
differentiable manifold M . The set X (M ) is a module on the ring F(M ) of functions on M . So a
tensor field of type (r, s) is a F(M )-linear map
A : X ∗ (M )r × X (M )s −→ F(M ).
Démonstration. We show that, if θi0 (p) = 0 or Xi0 (p) = P 0, then A(θ1 , . . . , θr , X1 , . . . , Xs )(p) = 0.
n
1
Let (U, (x , . . . , x )) be a chart. Then we can write Xj0 = X i ∂i . Let f be a bump function on U
i 2
with f (p)P= 1. We have Xj0 (p) = 0 ⇔ X (p) = 0, ∀i and f Xj0 is a vector field and we can write
f 2 Xj0 = f X i (f ∂i ). So
f 2 A(θ1 , . . . , θr , X1 , . . . , Xs ) = A(θ1 , . . . , θr , X1 , . . . , f 2 Xj0 , . . . , Xs )
X
= f xI A(θ1 , . . . , θr , X1 , . . . , f ∂i , . . . , X s )
i
1 r
and A(θ , . . . , θ , X1 , . . . , Xs )(p) = 0. ⋄
Chapitre 1. Differentiable manifolds 5
and we have X
A= Aij11,...,i i1 ir
,...,js ∂ji ⊗ · · · ⊗ ∂js ⊗ dx ⊗ · · · ⊗ dx .
s
The contraction of A on the indices i and j is the tensor field Cji A of type (r − 1, s − 1) which is
the composition of C and the tensor
(θ, X) 7−→ A(θ1 , . . . , θ, . . . , θr , X1 , . . . , X, . . . , Xs ).
The component of Cji A are Aij11,...,m,...,i
,...,m,...,js with m ∈ {1, . . . , n}.
r
For a function f ∈ F(M ) ⊂ T0,0 (M ), we set f ⊗ A = f A and we have D(f A) = f DA + (Df )A.
The derivation D is a derivation of functions so there exists a V ∈ X (M ) such that Df = V (f ).
The chain rule becomes
D(A(θ1 , . . . , θr , X1 , . . . , Xs )) = (DA)(θ1 , . . . , θr , X1 , . . . , Xs )
X r Xs
+ A((θ1 , . . . , Dθi , . . . , θr , X1 , . . . , Xs )) + (θ1 , . . . , θr , X1 , . . . , DXj , . . . , Xs ).
i=1 i=1
Théorème 1.14. Given a vector field V and an R-linear map δ : X (M ) −→ X (M ) such that
δ(f X) = V (f )W + f (δX),
there exists a unique derivation of tensors which equals to δ on X (M ) and V on F(M ).
Définition 1.15. Let V ∈ X (M ). Then we set the derivation LV as
LV (f ) := V (f ) et LV (X) := [V, X]
for all f ∈ F(M ) and X ∈ X (M ). It is called the Lie’s derivation.
Définition 1.16. Let V be a vector space. The index of a bilinear form b is the dimension of the
largest subspace W ⊂ V such that the restriction b|W ×W is negative definite.
A vector v ∈ V is null or isotropic if v ̸= 0 and b(v, v) = 0.
Lemme 1.17. Let V and W be two linear spaces of the same dimension. Then they are equipped
with inner products with the same indices if and only if the exists a linear isometry V −→ W .
7
Chapitre 2
Semi-riemannian manifolds
Proposition 2.6. Let (M, g) be a semi-riemannian manifold. Let V ∈ X (M ) a vector field. Let V ∗
be the 1-form defined by
V ∗ (X) := g(V, X).
Then the map V 7−→ V ∗ is a F(M )-linear isomorphism.
Exemples. – We take the sphere S2 . For a point p ∈ S2 and two tangent vectors X, Y ∈ Tp S2 ,
we can define
gp (X, Y ) := ⟨X, Y ⟩
where the notation ⟨·, ·⟩ is the standard inner product on R3 . This gives a riemannian metric
on S2 . If we replace the inner product ⟨·, ·⟩ by another semi-riemannian metric on R3 , then
this metric is no longer semi-riemannian in general.
– If g is a riemannian metric, then a another riemannian metric is given by
g̃p (X, Y ) := ef (p) gp (x, y)
for a smooth function f ∈ C ∞ (M, R).
– There exists three vector fields Ei on S3 which form a orthonormal family where the semi-
riemannian is the same as the first example. Then we can define a new semi-riemannian
metric by
◦ ⟨Ei , Ej ⟩ = 0 for all i ̸= j ;
◦ |E1 |2 = −1 and |E2 |2 = |E3 |2 = 1.
– On the half-plane H2 , we can define the metric
dx2 + dy 2
g := .
y2
Question. When can we equip M with a semi-riemannian metric ? It is not always the case for a
semi-riemannian metric with strictly positive index. But it is always the cases for a riemannian
metric. There is two ways to do that :
– by using the Withney’s theorem : the exists an immersion ι : M −→ RN for a large enough
integer N and we take the pullback of the euclidean matric on RN , that is
gp (X, Y ) := ⟨dιp (X), dιp (Y )⟩;
– if (Ui , xi ) are an atlas of M , we define
X
gp := αi x∗i ⟨·, ·⟩Rn .
i
Théorème 2.7. Let (M, g) be a semi-riemannian manifold. Then there exists a unique connection D
such that, for all vector fields V and W , we have
Chapitre 2. Semi-riemannian manifolds 9
– [V, W ] = DV W − DW V ;
– Xg(V, W ) = g(DX V, W ) + g(V, DX W )
Moreover, the connection D is characterized by the Koszul formula
2g(DV W, X) = V (g(X, W )) + W (g(X, V )) − Xg(V, W )
− g(V, [W, X]) + g(W, [X, V ]) + g(X, [V, W ]).
It is called the Levi-Civita connection.
Démonstration. Let D be a connection satisfying these two points. In the right-hand side of the
Koszul formula, using the two points, we obtain 2g(DV W, X). This proves the uniqueness because
of the one-to-one correspondance between vector fields and 1-forms.
Let proves the existence. Let F (V, W, X) the right-hand side of the Koszul formula. Then if we
take two vector fields V and W , then the map F (V, W, ·) : X (M ) −→ R is F(M )-linear. So it is
a 1-form. Thus there exists a unique vector field DV W such that
g(DV W, X) = F (V, W, X), ∀X ∈ X (M ).
This show the Koszul formula and that the map D is a connection. With this formula, we can prove
the two points. ⋄
Notation. We will write ∇ for the Levi-Civita connection. With this notation and g(·, ·) = ⟨·, ·⟩,
the two points of the theorem are
X⟨Y, Z⟩ = ⟨∇X Z, Y ⟩ + ⟨∇X Y, Z⟩,
[X, Y ] = ∇X Y − ∇Y X.
Définition 2.8. The Christoffel symbols for the chart (U, xi ) are the functions on U given by
n
X
D∂i (∂j ) = Γki,j ∂k .
k=1
Recall that [∂i , ∂j ] = 0 = D∂i (∂j ) − D∂j (∂i ) by the Schwarz theorem, so Γki,j = Γkj,i . Moreover,
P j
if W = W ∂j on U , then
X X
D∂i (W ) = (∂i (W j )∂j + W j Γki,j ∂k )
j k
X X
= (∂i (W j ) + W j Γki,j )∂k
k j
Démonstration. We need to check that R(f X, Y )Z = f R(X, Y )Z and R(X, Y )(f Z) = f R(X, Y )Z.
We have [f X, Y ] = f XY − Y (f )X − f Y X and
∇[f X,Y ] Z = f ∇X ∇Y Z
which prove the formula. ⋄
Remarque. The map R is a tensor. For X, Y, Z ∈ X (M ) and p ∈ M , the quantity (R(X, Y )Z)p
only depend on the values X(p), Y (p) and Z(p). So we can define Rp (u, v)w for u, v, w ∈ Tp M .
Proposition 2.11. Let X, Y and Z be three vector fields. Then
(∇Z R)(X, Y ) + (∇X R)(Y, Z) + (∇Y R)(Z, X) = 0.
Remarque. We have
(∇X R)(Y, Z)W = ∇X (R(Y, Z)W ) − R(∇X Y, Z)W − R(Y, ∇X Z)W − R(Y, Z)∇X W.
Moreover, we have
(∇X g)(Y, Z) = X(g(Y, Z)) − g(∇X Y, Z) − g(X, ∇X Z) = 0.
Let p ∈ M be a point and Π be a plane in Tp M . For two tangent vectors v, w ∈ Tp M not null
and not colinear, we define
Q(v, w) := ⟨v, v⟩⟨w, w⟩ − ⟨v, w⟩2 .
Chapitre 2. Semi-riemannian manifolds 11
where (Ei ) is a orthonormal frame (whe g(Ei , Ej ) = 0 if i ̸= j and g(Ei , Ei ) = εi = ±1). This
defines a symmetric form.
– The scalar curvature is the function
n
X
scal := εm Ric(Em , Em ).
m=1
X
= εm εj g((∇X R)(Em , Ej )Ej + R(∇X Em , Ej )Em + R(Em , ∇X Ei )Em + R(Em , Ej )∇X Em , Em )
m,j
X
= εm εj g((∇X R)(Em , Ej )Ej , Em )
m,j
X
= εm εj [−g((∇Ej R)(X, Em )Ej , Em ) − g((∇Em R)(Ej , X)Ej , Em )]
m,j
X
=− εm εj [(∇Em R)(Ej , X, Ej , Em ) + ∇Ej R(X, Em , Ej , Em )]
m,j
X
= εm εj [(∇Ej R)(Ej , Em , Em , X) + ∇EM R(Em , Ej , Ej , X)]
m,j
X
=2 εm εj (∇Ej R)(Ej , Em , Em , X)
m,j
X
=2 εm εj ∇Ej (R(Ej , Em , Em , X)) because ∇g = 0
m,j
X
=2 εm εj ∇Ej (Ric(Ej , X))
m,j
X
=2 εm εj ∇Ej (Ric(X, Ej ))
m,j
X
=2 εm εj (∇Ej Ric)(Ej , X)
m,j
= 2 div(Ric)X. ⋄
Démonstration. The idea is to take the divergence of the equation (∗). We have
X
div(f g)(X) = εi (∇Ei (f g))(Ei , X)
i
X
= εi [Ei (f g(Ei , X)) − f g(∇Ei Ei , X) − f g(Ei , ∇Ei X)]
i
X
= εi [Ei (f )g(X, Ei ) + f (Ei g(X, Ei )) − f g(∇Ei Ei , X) − f g(Ei , ∇Ei X)]
i
X
= εi Ei (f )g(X, Ei )
i
X
= εi g(X, Ei )Ei (f )
i
= X(f ) = df (X)
and so div(f g) = df . Taking the divergence of the equation (∗), we obtain div(Ric) = df and, by
the last proposition, we have df = dscal/2. Taking its trace, we get scal = (dim M )f . Then we have
scal = 2f + K for a constant K. As dim M ⩾ 2, the function scal is a constant and so does the
function f . ⋄
Chapitre 2. Semi-riemannian manifolds 13
Proposition 2.18. A vector field X is a Killing vector field if and only if its flow is by isometries.
Let Fp (t) = dΨ−t (WΨt (p) . The right-hand side of the equation (2) is exactly Fp′ (0). Assume Vp ̸= 0.
Let (xi ) a system of local coordinates such that ∂/∂xi = V . Then locally
x1 (Ψt (q)) = x1 (q) + t et xj (Ψt (q)) = xj (q), j⩾2
and
∂
X ∂Ψj ∂ ∂
dΨt = = (Ψt ).
∂xi j
∂xi ∂xj ∂x i
∂
W i ∂x
P
Let W = i i
. Thus
X ∂
Fp (t) = W i (Ψt (p)) .
i
∂xi
and so
X d(W i ◦ Ψt (p)) ∂
Fp′ (0) =
i
dt t=0 ∂xi
X ∂
= Vp (W i )
i
∂xi
X ∂W i
= ∂i .
i
∂x1
So
[V, W ] = (∂1 , W )
X
= [∂1 , W i ∂i ]
X
= ∂1 (W 1 )∂i + W i [∂1 , ∂i ]
X ∂X i
= ∂i = Fp′ (0). ⋄
∂x1
Lemme 2.19. Let X be a Killing vector field on a connected manifold M such that, for all
point p ∈ M , we have
X(p) = 0 et (∇X)p = 0.
Then X = 0 on M .
Remarque. Klling vector fields form a Lie algebra of finite dimension because [L X , L Y ] = L [X,Y ] .
Moreover, the dimension is less then n(n + 1)/2 where n = dim M .
Proposition 2.20. Let X be a Killing vector field and f := |X|2 /2. Then
∆f = − Ric(X, X) + |∇X|2 .
Théorème 2.23 (Berger). Let M a compact riemannian manifold of even dimension with positive
sectional curvature. The any Killing vector field has a zero.
Démonstration. Let X be a vector field. Let f := |X|2 /2. Then grad f = −∇X X. If X has no
zero, then f has a positive minimum at a point p ∈ M . Then Hess f (p) ⩾ 0. Let V be a vector field.
Ten
Hess f (V, V ) := ⟨∇V (∇f ), V ⟩ = ⟨−∇V ∇X X, V ⟩
= ⟨R(V, X)X, V ⟩ + ⟨∇V X, ∇V X⟩.
But B : X 7−→ ∇X X is skew symmetric, so (∇X X)(p) = (grad f )p = 0, so B admits λ = 0 as
an eigenvalue with X(p) as a eigenvector. As dim M is even, there exists another eigenvector V
corresponding to λ = 0. ⋄
17
Chapitre 3
Geodesics
Exemple. The map t 7−→ (γ(t), γ ′ (t)) is a vector field along the curve γ.
Proposition 3.2. Let M a semi-riemannian manifold et γ : I −→ M a curve. Then there exists a
unique R-linear operator
D
: {vector fields along γ} −→ {vector fields along γ}
dt
such that
D df D
– dt (f X) = dt X + f dt X;
D
– If X(t) = Y (γ(t)), then dt X = (∇γ Y ) ◦ γ.
Démonstration. Let t0 ∈ I. Let (U, x) a chart on M and J ⊂ I an interval such that γ(J) ⊂ U .
Let Xi := ∂/∂xi . If Y is a vector field along γ, we have
X
Tγ(t) M ∋ Y (t) = αj (t)(Xj )γ(t).
j
Démonstration. Let (U, x) be a chart such that γ(t0 ) ∈ U . Let Xi := ∂/∂xi . Let J ⊂ I be a
interval such that γ(J) ⊂ U . We denote
X X
γ̇(t) = γ̇ i (t)Xi (γ(t)) et Y (t) = αj (t)Xj (γ(t)).
j
Then
DY Xh X i
(t) = α̇k (t) + αj (t)γ̇i (t)Γki,j (γ(t)) Xk (γ(t))
dt i,j
k
and so
DY X
(t) = 0 ⇐⇒ ∀k, α̇k (t) + αj (t)γ̇i (t)Γki,j (γ(t)) = 0. (∗)
dt i,j
Lemme 3.6. Let X and Y be two parallel vector field along a curve γ. The the map
t 7−→ gγ(t) (X(t), Y (t))
a constant. For X = Y = γ̇, if γ is a geodesic, then g(γ̇, γ̇) = |γ|2 is constant.
Remarque. So causal type of geodesics is preserve on frame (Xi ).
Théorème 3.7. Let M be a semi-riemannian manifold. Let p ∈ M and v ∈ Tp M . Then there exists
an open interval I and a unique geodesic γ : I −→ M such that
γ(0) = p et γ̇(0) = v.
Démonstration. Let (U, x) be a chart such that γ(t0 ) ∈ U . Let Xi := ∂/∂xi . Let J ⊂ I be a
interval such that γ(J) ⊂ U . We write
X
γ̇ = γ̇i (Xi ◦ γ).
i
We have Xh X i
∇γ̇ γ̇ γ̈k (t) + γ̇j (t)γ̇j (t)Γki,j ◦ γ Xk (γ(t)).
k i,j
if on only if its components satisfy the systems of second order nonlinear ordinary differential
equation. Existence is given, for any initial data p and v, by the Picard-Lindelöf-Cauchy theorem. ⋄
19
Chapitre 4
Examples
Exemple. The euclidean space Rn is a semi-riemannian manifold. The geodesics are straight lines.
Indeed, we have Γki,j = 0 and a path γ must verify the equation
γ̈ k + Γki,j γ̇ i γ̇j = 0
Exemple. The hyperbolic space is Hm := R∗+ × Rm−1 . Its tangent spaces are Tp Hm ≃ Rm . We
equip this manifold with the metric
⟨X, Y ⟩
g(X, Y ) = .
x21
It is a riemannian manifold with sectional curvature equal to −1.
We can choose others models of the hyperbolic space such as
Hm = {(x0 , . . . , xm ) ∈ Rm+1 | x0 > 0, −x20 + x21 + · · · + x2m = −1}.
Equipped with the induce metric, it is a riemannian manifold. An other model is the Poincaré
model
Dm := {x ∈ Rm | |x| < 1}
with the metric
4
g(X, Y ) = ⟨X, Y ⟩.
(1 − |x|2 )2
The sectional curvature is also equal to −1.
Chapitre 5
Calculus of variations
Facts.
D
1. For all X0 ∈ Tγ(0) M , there exists Y ∈ Aγ such that Y (0) = X0 and dt Y = 0.
D
2. For all X0 ∈ Ta M , there exists t0 > 0 and γ : [0, t0 [ −→ M such that γ(a) = X0 and dt γ = 0.
Such a γ is called a geodesic.
3. We also write dtD
Y = Y ′ = Ẏ = ∇γ̇ Y .
4. If γ is a geodesic, then
d
⟨γ̇, γ̇⟩ = ⟨γ̈, γ̇⟩ = 0.
dt
Définition 5.1. Let M be a semi-Riemannian manifold. A variation of a function α : [a, b] −→ M
of class C ∞ is a map x : [a, b] × ]−δ, δ[ −→ M of class C ∞ with δ > 0 such that x(u, 0) = α(u).
The variation vector field is the vector field V such that
∂x
V (u) := (u, 0).
∂v
The length of α is
Z b
L(α) := |α′ (s)| ds
a
p
where |·| = |⟨·, ·⟩|. The length of V is
Z b
∂x
L(v) = Lx (v) = (s, v) ds.
a ∂u
We consider curves such that |γ ′ (t)| > 0, called regular curves of space-like. We denote ε the
sign of ⟨α′ , α′ ⟩.
Lemme 5.2. If x is a variation of α with |α′ | > 0, then
Z b ′
α (u)
L′x (0) = ε ⟨ ′ , V ′ (u)⟩ du.
a |α (u)|
22
∂x
Démonstration. With xu = ∂u , we have
Z b
L(u) := |xu (u, v)| du.
a
We have α′ = xu (u, 0)′ . So for δ small enough, we have |xu (u, v)| > 0 for u ∈ ]−δ, δ[. So
Z b
′ d
L (0) = |xu | dt.
a du u=0
But we get
d 1 ε⟨xu , xuv ⟩
|xu | = (ε⟨xu , xu ⟩)−1/2 2ε⟨xu , xuv ⟩ = .
du 2 ⟨xu , xu ⟩
Take u = 0, we get xu (u, 0) = α′ (0) and xv (u, 0) = V (u) and xuv (u, 0) = V ′ (u). ⋄
Proposition 5.3 (first variation). Let α : [a, b] −→ M be a continuous and smooth curve piece-wise
of constant speed c > 0 and of sign ε. Let x be a variation of α. Then
n
ε b ′′
Z
εX ε
L′ (0) = − ⟨α , V ⟩ du − ⟨∆α′ (Ui ), V (Ui )⟩ + ⟨α′ , V ⟩|ba
c a c i=1 c
with U1 < · · · < Uk are points where α is not C ∞ and
∆α′ (Ui ) = α′ (Ui+ ) − α′ (Ui− ) ∈ Tα(Ui ) M.
Démonstration. We have
α′ 1
⟨ ′
, V ⟩ = ⟨α′ , V ′ ⟩.
|α | c
On ]Ui , Ui+1 [, we have
d ′
⟨α′ , V ′ ⟩ = ⟨α , V ⟩ − ⟨α′′ , V ⟩.
du
So Z Ui+1 Z Ui+1
U
⟨α′ , V ′ ⟩ du = ⟨α′ , V ⟩Ui+1
i
− ⟨α′′ , V ⟩ du.
Ui Ui
We sum up to obtain the desired formula. ⋄
Corollaire 5.4. A piece-wise smooth curve α with constant speed c > 0 is a geodesic if and only if
the first variation of L is zero for any variation with fixed ends.
Remarque. Fixed ends imply that V is zero at a and b and
ε ′
⟨α , V ⟩|ba = 0.
2
Démonstration. Suppose that α is a geodesic, that is α′′ = 0. Then α is smooth, so ∆α′ (Ui ) = 0.
In particular, we get V (a) = V (b) = 0 and so L′ (0) = 0.
Suppose that L′ (0) = 0. First we show that α is a geodesic on ]Ui , Ui+1 [, that is α′′ (t) = 0
for t ∈ ]Ui , Ui+1 [. Let y be in Tα(t) M and f a smooth function defined on [a, b] with supp f ⊂
[t − δ, t + δ] ⊂ ]Ui , Ui+1 [ and f ∈ [0, 1] and f = 1 on ]t − δ/2, t + δ/2[. Let Y be the vector field
D
obtained by parallel transport of y along α, that is dt Y = 0 and Y (t) = 0. Let V := f Y . Then
V (a) = V (b) = 0. Let exp be the exponential map, that is the map
expp : D ⊂ Tp M −→ M
with p ∈ M where
expp (v) = B(1)
where B is the geodesic starting at p with initial speed v and where
D = {v ∈ Tp M | B(1) exists}.
Let x(u, v) = expα(u) (vV (u)). Then x(u, v) is a variation of α with fixed ends. So L′ (0) = 0 and
then
Z b
0= ⟨α′′ , v⟩ du
a
Chapitre 5. Calculus of variations 23
Z t+δ
= ⟨α′′ , f Y ⟩.
t−δ
We will compute L′′ (0) if L′ (0) = 0. Any vector field Y along α decomposes as Y = Y T + Y ⊥
where Y T = ε⟨Y, α′ ⟩α′ =: f α′ and Y ⊥ is orthogonal to α′ . If α is a geodesic, then
Y ′ = f ′ α′ + (Y ⊥ )′ .
Moreover, we have (Y ′ )⊥ = (Y ⊥ )′ .
Théorème 5.5 (second variation). Let γ be a geodesic of constant speed c > 0 and of sign ε. If x
is a variation of γ, then
ε b ′⊥ ′⊥
Z
ε
L′′ (0) = ⟨V , V ⟩ − ⟨R(V, γ ′ )V, γ ′ ⟩ du + ⟨γ ′ , A⟩|ba
c a c
where V (u) = xv (u, 0) and A(u) = xvv (u, 0).
Let Ω(p, q) be the space of smooth piece-wise curves from [a, b] to M starting at p and ending
at q. The tangent space to Ω(p, q) at α is the set Tα Ω(p, q) of vector fields V along α with
V (a) = V (b) = 0. The index of σ ∈ Ω(p, q) is the bilinear symmetric form
Iσ : Tσ Ω −→ Tσ Ω
such that Iσ (V, V ) = Lx (σ) where x is a variation with fixed ends and variation vector V , that is
ε b ′⊥ ′⊥
Z
Iσ (V, W ) = ⟨V , V ⟩ − ⟨R(V, σ ′ )W, σ ′ ⟩ du.
c a
We have Iσ (V, W ) = Iσ (V ⊥ , W ⊥ ).
Lemme 5.6. Let σ be a non-null geodesic with sign ε. Let M be a semi-Riemannian manifold with
index ν. Then
1. if Iσ is semi-definite positive, then ν = 0 or n ;
2. if Iσ is semi-definite negative, then ν = 1 or n − 1.
Définition 5.7. Let γ be a geodesic. A vector field Y along γ is called Jacobi field if
Y ′′ = R(Y, γ ′ )γ ′ .
If x is a variation of γ such that
∀v, u 7−→ x(u, v) is a geodesic,
then the variation vector u 7−→ ∂x∂v (u, 0) is a Jacobi field.
For all v, w ∈ Tp M , there exists an unique Jacobi field Y along γ such that Y (0) = v and
Y ′ (0) = w.
Définition 5.8. Two points σ(a) and σ(b) with a ̸= b on a geodesic σ are conjugate if there exists
a nontrivial Jacobi field Y such that J(a) = J(b) = 0.
Then σ(a) and σ(b) are conjugate if and only if there exists a variation x of σ such that the map
∂x
u 7−→ x(u, v) is a geodesic, for all v, started from σ(a) such that ∂u (b, 0) = 0. This is equivalent
to the fact that the exponential map expp : Tp M −→ M is singular at bσ ′ (0), that is there is a
tangent vector x to p at bσ ′ (0) such that d(expp )bσ′ (0) (x) = 0.
24
Lemme 5.9. Let σ be a geodesic such that σ ⊥ (s) ∈ Tσ(s) M is space-like. If ⟨R(v, σ ′ )v, σ ′ ⟩ ⩽ 0 for
all v ⊥ σ ′ , then there is no conjugate points along σ.