Differential geometry

Lecture 17: Geodesics and the exponential map

David Lindemann

University of Hamburg
Department of Mathematics
Analysis and Differential Geometry & RTG 1670

3. July 2020

David Lindemann DG lecture 17 3. July 2020 1 / 44

1 Geodesics

2 The exponential map

3 Geodesics as critical points of the energy functional

4 Riemannian normal coordinates

5 Some global Riemannian geometry

David Lindemann DG lecture 17 3. July 2020 2 / 44

Recap of lecture 16:
constructed covariant derivatives along curves
defined parallel transport
studied the relation between a given connection in the
tangent bundle and its parallel transport maps
introduced torsion tensor and metric connections,
studied geometric interpretation
defined the Levi-Civita connection of a
pseudo-Riemannian manifold

David Lindemann DG lecture 17 3. July 2020 3 / 44

 Geodesics

Recall the definition of the acceleration of a smooth curve

γ : I → Rn , that is γ 00 ∈ Γγ (T Rn ).
Question: Is there a coordinate-free analogue of this construc-
tion involving connections?
Answer: Yes, uses covariant derivative along curves.

Definition
Let M be a smooth manifold, ∇ a connection in TM → M,
and γ : I → M a smooth curve. Then ∇γ 0 γ 0 ∈ Γγ (TM) is
called the acceleration of γ (with respect to ∇).

Of particular interest is the case if the acceleration of a curve

vanishes, that is if its velocity vector field is parallel:

Definition
A smooth curve γ : I → M is called geodesic with respect to a
given connection ∇ in TM → M if ∇γ 0 γ 0 = 0.

David Lindemann DG lecture 17 3. July 2020 4 / 44

 Geodesics

In local coordinates (x 1 , . . . , x n ) on M we obtain for any smooth

curve γ : I → M the local formula
n n
!
0
X ∂2γk X ∂γ i ∂γ j k ∂
∇γ 0 γ = + Γij .
∂t 2 i,j=1
∂t ∂t ∂x k
k=1

As a consequence we obtain a local form of the geodesic equa-

tion:
Corollary
γ is a geodesic if and only if in all local coordinates covering a
nontrivial subset of the image of γ it holds that
n
∂2γk X ∂γ i ∂γ j k
2
+ Γij = 0 ∀1 ≤ k ≤ n,
∂t i,j=1
∂t ∂t

where one usually writes Γkij instead of Γkij ◦ γ.

Alternative notation: ẍ k + ẋ i ẋ j Γkij = 0 ∀1 ≤ k ≤ n.

David Lindemann DG lecture 17 3. July 2020 5 / 44

 Geodesics

Example
 n
 the Levi-Civita connection ∇ of
Geodesics w.r.t.
n P i 2
R , (du ) are affine lines of constant speed. For any
i=1
arbitrary curve γ : I → Rn we have

∂2γ1 ∂2γn
 
∇γ 0 γ 0 = γ 00 = γ, , . . . , .
∂t 2 ∂t 2

If a curve is a geodesic with respect to a metric connection, e.g.

the Levi-Civita connection, it automatically has the following
property:

Lemma
Let γ : I → M be a geodesic on a pseudo-Riemannian manifold
(M, g ) with respect to a metric connection ∇. Then
g (γ 0 , γ 0 ) : I → R is constant.
0 0
Proof: ∂(g (γ∂t,γ )) = ∇γ 0 (g (γ 0 , γ 0 )) = (∇γ 0 g )(γ 0 , γ 0 ) +
2g (∇γ 0 γ 0 , γ 0 ) = 0
David Lindemann DG lecture 17 3. July 2020 6 / 44
 Geodesics

Corollary
A geodesic γ in w.r.t. the Levi-Civita connection ofp
a
Riemannian manifold (M, g ) has constant speed g (γ 0 , γ 0 ).

Next, we need to study if geodesics always exist and determine

their uniqueness.

Proposition A
Let M be a smooth manifold and ∇ a connection in
TM → M. Let further p ∈ M and v ∈ Tp M. Then there
exists ε > 0 and a smooth curve γ : (−ε, ε) → M,
γ(0) = p, γ 0 (0) = v , such that γ is a geodesic.
If γ1 : I1 → M and γ2 : I2 → M are geodesics on M such
that I1 ∩ I2 6= ∅ and for some point t0 ∈ I1 ∩ I2 ,
γ1 (t0 ) = γ2 (t0 ) and γ10 (t0 ) = γ20 (t0 ), then γ1 and γ2
coincide on I1 ∩ I2 , i.e. γ1 |I1 ∩I2 = γ2 |I1 ∩I2 .

Proof: (next page)

David Lindemann DG lecture 17 3. July 2020 7 / 44

 Geodesics

(continuation of proof)
suffices to prove this proposition in local coordinates
the differential equation for a geodesic in local
coordinates ẍ k + ẋ i ẋ j Γkij = 0, 1 ≤ k ≤ n, is a nonlinear
system of second order ODEs
turn this system of n second order ODEs into a
system of 2n first order system of ODEs
system of equations

ẋ k = v k , v̇ i = −ẋ i ẋ j Γkij ∀1 ≤ k ≤ n

with fitting initial values

first thing in need of clarification: the symbols v k
the v k are precisely the induced coordinates on
TU ⊂ TM, so that v k (V ) = V (x k ) for all V ∈ Tq M with
q∈U
in the above eqn., the x k and v k are, however, to be read
as components of a curve
(x = x(t), v = v (t)) : I → TM
(continued on next page)
David Lindemann DG lecture 17 3. July 2020 8 / 44
 Geodesics

(continuation of proof)
in local coordinates x k , v k , the loc. geod. eqn. can be
viewed as integral curve of a vector field on TU,
G ∈ X(TU), that is a smooth section in TTU → TU
to see this first observe that since the x k and v k are
coordinate functions on TU, they induce coordinates
on TTU
the corresponding local frame in TTU → TU is given by
 ∂
∂x 1
, . . . , ∂x∂ n , ∂v∂ 1 , . . . , ∂v∂ n

one can imagine each ∂x∂ k as being “horizontal” and each

∂
∂v k
as being “vertical”
using Einstein summation convention, G is given by

G = v k ∂x∂ k − v i v j Γkij ∂v∂ k

(continued on next page)

David Lindemann DG lecture 17 3. July 2020 9 / 44
 Geodesics

(continuation of proof)
by covering M with charts and thus TM with induced
charts, G extends to a vector field on TM, G ∈ X(TM)
since any integral curve of G , (x, v ) : I → TM,
t 7→ (x(t), v (t)), fulfils ẋ = v , it is precisely the velocity
vector field of the curve x : I → M, t 7→ x(t)
hence, the projection of any integral curve of G to M via
the bundle projection π : TM → M is a geodesic
by existence and uniqueness properties of integral
curves of vector fields, the statement of this proposition
follows

Definition
The (local) flow of the vector field G ∈ X(TM), locally given
by G = v k ∂x∂ k − v i v j Γkij ∂v∂ k as in Proposition A, is called
geodesic flow with respect to ∇.

David Lindemann DG lecture 17 3. July 2020 10 / 44

 Geodesics

Local uniqueness of geodesics allows us to define a maximality

property for geodesics:

Definition
A geodesic γ : I → M is called maximal if there exists no
strictly larger interval Ie ⊃ I and a geodesic γ
e : Ie → M, such
that γe|I = γ. This means that γ cannot be extended to a
larger domain while still keeping its geodesic property. A
smooth manifold with connection ∇ in TM → M is called
geodesically complete if every maximal geodesic is defined on
I = R. A pseudo-Riemannian manifold (M, g ), respectively the
metric g , is called geodesically complete if its Levi-Civita
connection is complete.

David Lindemann DG lecture 17 3. July 2020 11 / 44

 Geodesics

Question: Which reparametrisation of geodesics are allowed so

that the result is still a geodesic?
Answer:
Lemma
Let γ : I → M be a geodesic with non-vanishing speed with
respect to ∇ and f : I → I 0 a diffeomorphism. Then γ ◦ f is a
geodesic with non-vanishing speed if and only if f is
affine-linear, that is of the form f (t) = at + b for a ∈ R \ {0},
b ∈ R.

Proof:
local formula & chain rule

∇(γ◦f )0 (γ ◦ f )0 = f 00 · γ 0 ◦ f + (f 0 )2 · (∇γ 0 γ 0 ) ◦ f = f 00 · γ 0 ◦ f ,

where the last equality comes from assumption that γ is a

geodesic
hence, γ ◦ f is a geodesic if and only if f 00 = 0, that is if
f = at + b with a ∈ R \ {0}, b ∈ R

David Lindemann DG lecture 17 3. July 2020 12 / 44

 Geodesics

An immediate consequence is:

Corollary A
Maximal geodesics are unique up to affine reparametrisation.
If γ : I → M is a geodesic with initial value γ(0) = p,
γ 0 (0) = v ∈ Tp M, t 7→ γa (t) := γ(at) is a geodesic with initial
value γa (0) = p, γa0 (t) = av for all a ∈ R. If a = 0, the domain
of γa is R. If a 6= 0, the domain of γa is a−1 · I .

In the case of pseudo-Riemannian manifolds, we have the fol-

lowing additional result:

Corollary
A geodesic in a pseudo-Riemannian manifold with respect to
the Levi-Civita connection with nonvanishing velocity can
always be parametrised to be of unit speed, that is either
g (γ 0 , γ 0 ) ≡ 1 or g (γ 0 , γ 0 ) ≡ −1.

David Lindemann DG lecture 17 3. July 2020 13 / 44

 Geodesics

Example
Each maximal geodesics of Rn equipped with the
canonical connection with initial condition
γ(0) = p ∈ Rn , γ 0 (0) = v ∈ Tp Rn ∼
= Rn , is of the form

γ : R → Rn , t 7→ p + tv .

This in particular means that the canonical connection on

Rn is geodesically complete.
Consider S n ⊂ Rn+1 with induced metric
g = h·, ·i|TS n ×TS n , where h·, ·i denotes the standard
Riemannian metric on Rn . The maximal geodesics of
(S n , g ) with respect to the Levi-Civita connection are
great circles, that is

γ : R → S n, t 7→ e At p

for γ(0) = p ∈ S n , γ 0 (0) = Ap, A ∈ Mat(n × n) skew.

This, again, means that the Levi-Civita connection of
(S n , g ) is geodesically complete.
David Lindemann DG lecture 17 3. July 2020 14 / 44
 The exponential map

Next, we will construct the exponential map of a given connec-

tion in the tangent bundle of a smooth manifold. This requires
some technical tools and auxiliary results.

Definition
Let M be a smooth manifold with connection ∇ in TM → M.
An open neighbourhood of the zero section in TM → M is
an open set V ⊂ TM such that for all p ∈ M, Vp := Tp M ∩ V
is an open neighbourhood of the origin 0 ∈ Tp M. Note that
the smooth manifold structure and topology on Tp M are
induced by the local trivialisations of TM → M and the
corresponding fibrewise isomorphisms Tp M ∼ = Rn .

Lemma
Let ∇ be a connection in TM → M. Then there exists an
open nbhd. of the zero section V ⊂ TM, such that for all
v ∈ Vp ⊂ V , the geodesic γv with initial condition γv (0) = p,
γv0 (0) = v , has domain containing the compact interval [0, 1].

Proof: (next page)

David Lindemann DG lecture 17 3. July 2020 15 / 44
 The exponential map

(continuation of proof)
Corollary A implies that if γv is defined on at least [0, 1],
γrv for r ∈ [0, 1] is also defined on [0, 1]
construction in proof of Proposition A geodesics can be
viewed as projections of integral curves of a vector field
on TM
hence, by identifying M with the image of the zero
section in TM and using Corollary A, it follows that in
order to prove this proposition it suffices to show that for
all p ∈ M ⊂ TM we can find εp > 0 and an open
neighbourhood Wp of p in TM (not a subset of the fibre
Tp M), such that all integral curves of
G = v k ∂x∂ k − v i v j Γkij ∂v∂ k starting in Wp are defined on at
least [0, εp ]
this follows from the fact that G is a smooth vector field
if εp < 1, we rescale Wp fibrewise with scaling factor εp ,
so that we can assume w.l.o.g. that all integral curves of
G starting in Wp are defined on at least [0, 1]
(continued on next page)
David Lindemann DG lecture 17 3. July 2020 16 / 44
 The exponential map

(continuation of proof)
repeating this procedure for all p ∈ M ⊂ TM, we obtain
our desired open neighbourhood V ⊂ TM of the zero
section in TM → M by setting
[
V := Wp
p∈M

Now we have all tools at hand to define the exponential map:

Definition
Let V ⊂ TM be an open neighbourhood of the zero section
in TM → M such that for all v ∈ V , the unique maximal
geodesic γv with respect to ∇ with initial condition γv (0) = p,
γv0 (0) = v , is defined on [0, 1]. The exponential map with
respect to ∇ is defined as

exp : V → M, v 7→ γv (1).

The exponential map at p ∈ M expp : Vp → M is the

restriction of exp to Vp .
David Lindemann DG lecture 17 3. July 2020 17 / 44
 The exponential map

We want to use the exponential map to construct certain nice

local coordinates. In order to do so, we need:

Proposition B
Let M be a smooth manifold and ∇ a connection in TM → M.
For all p ∈ M, the exponential map at p is a local
diffeomorphism near 0 ∈ Tp M.

Proof:
suffices to show d expp = idTp M , which together with
theorem about local invertibility will complete the proof
let γv denote the maximal geodesic with chosen initial
value γv (0) = p, γv0 (0) = v for v ∈ Tp M
∂
Corollary A implies d expp (v ) = ∂t t=0
expp (tv ) =
∂ ∂ 0
∂t t=0
γ tv (1) = ∂t t=0
γ v (t) = γ v (0) = v
since v ∈ Tp M was arbitrary the claim follows
Note: Strictly speaking, we identified T0 Tp M with Tp M for the
domain of d expp via the canonical isomorphism (0, v ) = v .

David Lindemann DG lecture 17 3. July 2020 18 / 44

 The exponential map

Remark
If ∇ is geodesically complete, exp is defined on TM. This
however does not mean that there exists p ∈ M, such that
expp is a diffeomorphism.

Exercise
Show that for any p ∈ Rn , expp defined on Tp Rn with
respect to the canonical connection is a diffeomorphism.
Show that if M is compact and ∇ is any connection in
TM → M, expp is never a diffeomorphism for all
p ∈ M, independent of its domain Vp ⊂ Tp M.

David Lindemann DG lecture 17 3. July 2020 19 / 44

 The exponential map

Aside from geodesics, it is also common to study curves fulfilling

a similar but weaker requirement:

Definition
A smooth curve γ : I → M is called pregeodesic with respect
to a connection in TM → M if it has a reparametrisation as a
geodesic, that is if there exists a diffeomorphism f : I 0 → I ,
such that γ ◦ f is a geodesic.

Pregeodesics fulfil the following equation similar to the geodesic

equation:

Lemma
Any given pregeodesic γ : I → M with respect to a
connection ∇ in TM → M fulfils ∇γ 0 γ 0 = F γ 0 for some
smooth function F : I → R.

Proof: Follows from writing out ∇(γ◦f )0 (γ ◦ f )0 using our local

formula for γ ◦ f a geodesic.

David Lindemann DG lecture 17 3. July 2020 20 / 44

 Geodesics as critical points of the energy functional

We have not yet given a geometric reason as to why one would

study geodesics in the first place. To do so we introduce the
energy functional for curves with compact domain.

Definition
Let (M, g ) be a pseudo-Riemannian manifold and
γ : [a, b] → M be a smooth curve. The energy functional
evaluated at γ, or simply energy of γ, is given by
Zb
1
E (γ) := g (γ 0 , γ 0 )dt.
2
a

Note:
Compare the above with the definition of the length L(γ)
of γ for (M, g ) Riemannian!
E (γ) can also be defined for piecewise smooth curves,
see discussion in lecture notes.
We did not specify a structure on the domain of E . This
is, in general, a very difficult task.
David Lindemann DG lecture 17 3. July 2020 21 / 44
 Geodesics as critical points of the energy functional

For our purposes we need to understand how to perturb a given

curve γ : [a, b] → M in a “good” way, so that we can study the
infinitesimal change in E (γ).

Definition
Let γ : [a, b] → M be a smooth curve and ε > 0. A smooth
map η : (−ε, ε) × [a, b] → M is called variation of γ if
η(0, t) = γ(t) for all t ∈ [a, b]. η is called variation with fixed
endpoints of γ if η(s, a) = γ(a) and η(s, b) = γ(b) for all
s ∈ (−ε, ε). The vector field V along γ,
Vγ(t) = ∂η
∂s
(0, t) ∈ Tγ(t) M, is the variational vector field
corresponding to η.

Note: Geometrically, we understand η as a smooth family of

curves containing γ. Furthermore, independently of the chosen
variation with fixed endpoints, Vγ(a) = Vγ(b) = 0.

David Lindemann DG lecture 17 3. July 2020 22 / 44

 Geodesics as critical points of the energy functional

Question: Can every vector field along a curve γ with compact

domain be realised as the variational vector field of a variation
of γ?
Answer: Yes!
Lemma
Let γ : [a, b] → M be a smooth curve and V ∈ Γγ (TM). Then
there exists a variation η of γ, such that V is the variational
vector field of η. If Vγ(a) = Vγ(b) = 0, η can be chosen to be
variation with fixed endpoints.

Proof:
fix a Riemannian metric g on M with Levi-Civita
connection ∇
let exp : V → M denote the corresponding exponential
map
we now define a variation of γ via

η : (−ε, ε) × [a, b] → M, η(s, t) := exp(sVγ(t) )

for ε > 0 small enough

David Lindemann DG lecture 17 3. July 2020 23 / 44
 Geodesics as critical points of the energy functional

(continuation of proof)
we can always find such an ε by the compactness of
[a, b] and the smoothness of V
if V vanishes at γ(a) and γ(b), η has the property
η(s, a) = γ(a) and η(s, b) = γ(b) for all s ∈ (−ε, ε)
we check with a calculation as the one in the proof of
Proposition B
∂η
(0, t) = Vγ(t)
∂s
for all t ∈ [a, b]
hence, η fulfils the required properties of this lemma

David Lindemann DG lecture 17 3. July 2020 24 / 44

 Geodesics as critical points of the energy functional

We can now can describe the infinitesimal change in E (γ) with

respect to a variational vector field:
Lemma
Let (M, g ) be a pseudo-Riemannian manifold with Levi-Civita
connection ∇. Then the first variation of the energy at a
smooth curve γ : [a, b] → M with respect to a given
variational vector field V ∈ Γγ (TM) with a choice of
corresponding variation of γ, η : (−ε, ε) × [a, b] → M,
∂
η : (s, t) 7→ η(s, t), is given by ∂s s=0
E (η(s, ·)) and fulfils

∂
E (η(s, ·))
∂s s=0
Zb
=− g (V , ∇γ 0 γ 0 )dt + g (Vγ(b) , γ 0 (b)) − g (Vγ(a) , γ 0 (a)).
a

In the special case that V vanishes at the start- and

end-point of γ, we have have
Rb
∂
∂s s=0
E (η(s, ·)) = − g (V , ∇γ 0 γ 0 )dt.
a

David Lindemann DG lecture 17 3. July 2020 25 / 44

 Geodesics as critical points of the energy functional

Proof:
we will use the Einstein summation convention
let η 0 = η 0 (s, t) = ∂η
∂t
denote the velocity vector field of
the family of smooth curves η for s fixed
we calculate
Zb
∂ 1 ∂
E (η(s, ·)) = g (η 0 , η 0 )dt
∂s s=0 2 ∂s s=0
a
Zb Zb
1
= ∇V (g (η 0 , η 0 ))dt = g (γ 0 , ∇V η 0 )dt
2
a a

for the last equality we have used that ∇ is metric

for the next step we need to prove that ∇V η 0 = ∇γ 0 V
use local coordinates, that is fix p ∈ γ([a, b]) and
choose local coordinates (x 1 , . . . , x n ) on an open
neighbourhood of p ∈ M
(continued on next page)
David Lindemann DG lecture 17 3. July 2020 26 / 44
 Geodesics as critical points of the energy functional

(continuation of proof)
∂η k ∂
denote V s = Vη(s,t)
s
:= ∂s ∂x k
, so that V 0 = V
suffices to show that (∇V s η 0 )|s=0 = ∇γ 0 V
using our local formula we find
 2 k
∂η i ∂η j k

∂ η ∂
∇V s η 0 = + Γij
∂s∂t ∂s ∂t ∂x k
and
∂ 2 ηk ∂η i ∂η j k
 
∂
∇γ 0 V = + Γij .
∂t∂s ∂t ∂s s=0 ∂x k

∇ being torsion-free is equivalent to the Christoffel

symbols being symmetric in the lower indices
hence, the above local formulas for (∇V s η 0 )|s=0 and
∇γ 0 V indeed coincide
since p ∈ γ([a, b]) was arbitrary we deduce that the
equality holds for all t ∈ [a, b]
(continued on next page)
David Lindemann DG lecture 17 3. July 2020 27 / 44
 Geodesics as critical points of the energy functional

(continuation of proof)
hence we obtain with partial integration
Zb
g (γ 0 , ∇V η 0 )dt
a
Zb
= g (γ 0 , ∇γ 0 V )dt
a
Zb  
∂
= g (γ 0 , V ) − g (∇γ 0 γ 0 , V ) dt
∂t
a
Zb
= g (Vγ(b) , γ 0 (b)) − g (Vγ(a) , γ 0 (a)) − g (V , ∇γ 0 γ 0 )dt
a

reordering the above equation finishes the proof

David Lindemann DG lecture 17 3. July 2020 28 / 44

 Geodesics as critical points of the energy functional

Corollary
Geodesics defined on a compact interval with respect to the
Levi-Civita connection of a pseudo-Riemannian are critical
points of the energy functional in the sense that the first
variation of the energy with respect to variations with fixed
end points vanishes.

The converse also holds true:

Exercise
A curve in a pseudo-Riemannian manifold defined on a
compact interval is a geodesic with respect to the Levi-Civita
connection if it is a critical point of the energy functional.

Furthermore, you should try to solve:

Exercise A
Find a formula for the first variation of the length of a curve
in a Riemannian manifold. Are geodesics also critical points
of the length functional in our sense?

David Lindemann DG lecture 17 3. July 2020 29 / 44

 Geodesics as critical points of the energy functional

Remark
In Riemannian geometry, one can show that geodesics with
respect to the Levi-Civita connection and with compact domain
are not just critical points of the energy and length functional,
but also local minimisers. This means that for every variation
with fixed endpoints η : (−ε, ε) × [a, b] → M of a geodesic
γ : [a, b] → M in (M, g ), E (η(s, ·)) ≥ E (γ) for ε small enough.

References:
J.M. Lee, Riemannian Manifolds – An Introduction to
Curvature, Springer GTM 176 (1997)
C. Bär, Differential Geometry, lecture notes (2013)
O. Goertsches, Differentialgeometrie, lecture notes (2014)
(in German)

David Lindemann DG lecture 17 3. July 2020 30 / 44

 Riemannian normal coordinates

The exponential map at a fixed point can be used to construct

coordinates on a given manifold with particularly nice proper-
ties if expp comes from the Levi-Civita connection.

Definition
Let M be a smooth manifold with connection ∇ in its tangent
bundle. Suppose that V ⊂ Tp M is a star-shaped open
neighbourhood of the origin, such that expp : V → expp (V )
is a diffeomorphism. Then U = expp (V ) is an open
neighbourhood of p ∈ M and is called normal neighbourhood
of p ∈ M. Let U ⊂ M be such a normal neighbourhood of a
fixed p ∈ M. Then the exponential map at p can be used to
define local coordinates (x 1 , . . . , x n ) near p as follows:
Choose a basis {v1 , . . . , vn } of Tp M and define coordinates
implicitly via !
X n
i
expp x (q)vi = q
i=1

for all q ∈ U. This just means that the x i are the prefactor
functions of exp−1
p written in the basis {v1 , . . . , vn }.
(continued on next page)
David Lindemann DG lecture 17 3. July 2020 31 / 44
 Riemannian normal coordinates

Definition (continuation)
Smoothness of the x i follows from the implicit function
theorem. If (M, g ) is a pseudo-Riemannian manifold with
Levi-Civita connection ∇, normal coordinates at p ∈ M with
respect to an orthonormal basis {v1 , . . . , vn } of Tp M are
called Riemannian normal coordinates at p ∈ M. If (M, g ) is
Riemannian and V = Br (0) = {v ∈ Tp M | gp (v , v ) < r } for
some r > 0, the corresponding domain of the Riemannian
normal coordinates Brg (p) := expp (Bε (0)) is called geodesic
ball of radius r centred at p in M. The upper index g
indicates the corresponding Riemannian metric.

One application of Riemannian normal coordinates is:

Lemma
Any two points of a connected pseudo-Riemannian manifold
(M, g ) can be connected by a piecewise smooth curve, such
that every smooth segment of that curve is a geodesic.

Proof: Exercise!
David Lindemann DG lecture 17 3. July 2020 32 / 44
 Riemannian normal coordinates

Apart from connecting points with piecewise geodesics, Rieman-

nian normal coordinates have the property that certain geomet-
ric quantities are, at the reference point, of a very simple form.

Proposition C
Let (M, g ) be a pseudo-Riemannian manifold with
Levi-Civita connection ∇ and let ϕ = (x 1 , . . . , x n ) be
Riemannian normal coordinates near p ∈ M corresponding
to a choice of orthonormal basis {v1 , . . P
. , vn } of Tp M. Then
the prefactors of g written locally as gij dx i dx j fulfil

gij (p) = εij

for all 1 ≤ i, j ≤ n, where εij = g (vi , vj ). The Christoffel

symbols of ∇ and all partial derivatives of the local smooth
functions gij vanish at p, that is
∂gij
Γkij (p) = 0, (p) = 0
∂x k
for all 1 ≤ i, j, k ≤ n. (continued on next page)

David Lindemann DG lecture 17 3. July 2020 33 / 44

 Riemannian normal coordinates

Proposition C (continuation)
If γw : (−ε, ε) → M, γw (0) = p, γw0 (0) = w ∈ Tp M, is a
geodesic starting at p ∈ M such that its image is contained in
the domain of ϕ, ϕ ◦ γw is of the form

ϕ(γw (t)) = tw

for all t ∈ (−ε, ε).

Proof: (next page)

David Lindemann DG lecture 17 3. July 2020 34 / 44

 Riemannian normal coordinates

(continuation of proof)
∂
for gij (p) = εij we show that vk = ∂x k p
for all 1 ≤ k ≤ n
x k (p) = 0 for all 1 ≤ k ≤ n by construction implies
(after, as before, identifying T0 Tp M ∼
= Tp M)
n
X
d expp 0
(vk ) ⊗ dx k |p = idTp M . (1)
k=1

on the other hand we know by Proposition B that

d expp |0 (vk ) = vk for all 1 ≤ k ≤ n
∂
applying both sides of (2) to ∂x k p
proves our claim and,
hence, gij (p) = εij
∂g
next, note that ∂x ijk (p) = 0 implies with the help of
explicit local formula for Christoffel symbols that all
Christoffel symbols at p must also vanish
∂g
In order to prove ∂x ijk (p) = 0 we first show that the local
form of geodesics ϕ ◦ γw is of the claimed form
(continued on next page)
David Lindemann DG lecture 17 3. July 2020 35 / 44
 Riemannian normal coordinates

(continuation of proof)
by construction of the exponential map have
γw (t) = expp (tw ) for all t ∈ (−ε, ε)
n
w k vk , we have by definition of
P
writing w =
k=1
Riemannian normal coordinates

γw (t) = expp (tw )

n
! n
!
X k
X k
= expp tw vk = expp x (γw (t))vk ,
k=1 k=1

showing that ϕ(γw (t)) = tw for all t ∈ (−ε, ε) as claimed

writing down the geodesic equation for γw in our local
2 k
coordinates at p with ẍ k (0) = ∂ (x∂t(γ
2
w ))
(0) = 0 and
k k
ẋ (0) = w for all 1 ≤ k ≤ n shows that
n
X
Γkij (p)w i w j = 0 ∀1 ≤ k ≤ n
i,j=1

(continued on next page)

David Lindemann DG lecture 17 3. July 2020 36 / 44
 Riemannian normal coordinates

(continuation of proof)
this holds for arbitrary initial condition for the geodesic
γw0 (0) = w ∈ Tp M, this proves that for each fixed
1 ≤ k ≤ n, (Γkij (p))ij viewed as symmetric bilinear form
on Tp M × Tp M must identically vanish
hence, Γkij (p) = 0 for all 1 ≤ i, j, k ≤ n
for the vanishing of the partial derivatives of each gij at
p, observe that ∇ being metric implies
  
∂gij ∂ ∂ ∂
= g ,
∂x k ∂x k ∂x i ∂x j
n
! n
!
X ` ∂ ∂ X ` ∂ ∂
=g Γki , +g Γkj ,
∂` ∂j ∂` ∂i
`=1 `=1

for all 1 ≤ k ≤ n
evaluating the above equation at p and using that all
Christoffel symbols vanish at p yields the desired
result

David Lindemann DG lecture 17 3. July 2020 37 / 44

 Riemannian normal coordinates

Warning
In Proposition C we have seen that with the right choice of
coordinates, any pseudo-Riemannian metric and Levi-Civita
connection can be brought to a very simple form at a chosen
point. While this works of course for every point in the
manifold, this does not mean that every pseudo-Riemannian
metric is locally of the form gij = εij on some open
neighbourhood of our chosen reference point, this can in
general only be achieved at said point! Otherwise, every
manifold would be flat, and comparing with the upcoming
lecture about curvature shows that this is clearly not the case.

David Lindemann DG lecture 17 3. July 2020 38 / 44

 Riemannian normal coordinates

Proposition C implies that we have some control over the Taylor

expansion of a pseudo-Riemannian metric at a chosen point.

Corollary
The Taylor expansion of gij in Riemannian normal
coordinates (x 1 , . . . , x n ) at their reference point p ∈ M is of
the form !
X n
k 2
gij = εij + O (x ) .
k=1

David Lindemann DG lecture 17 3. July 2020 39 / 44

 Some global Riemannian geometry

Lastly, we will discuss some advanced results in this setting that

we will not prove, but you should still have heard about.

Remark
We have seen that geodesics with compact domain in
Riemannian manifolds are critical points of the energy
functional, and solving Exercise A shows that they are in
fact also critical points of the length functional. One
can, however, show more and prove that they are not just
any type of critical point but local minimisers, meaning
that for any variation η of γ with fixed endpoints,
E (η(s, ·)) ≥ E (γ) and L(η(s, ·)) ≥ L(γ) for s small
enough. Reference: Chapter 6 in J.M. Lee, Riemannian
Manifolds – An Introduction to Curvature, Springer GTM
176 (1997)
(continued on next page)

David Lindemann DG lecture 17 3. July 2020 40 / 44

 Some global Riemannian geometry

Remark (continuation)
An other way to study Riemannian manifolds is in the
context of metric geometry. In fact, every Riemannian
metric g on a smooth manifold M induces the structure of
a metric space on M, which in turn induces a topology
on M. It turns out that the induced topology on M
coincides, independently of the Riemannian metric g ,
with the initial topology on M.
We have interpreted variations of curves as a family of
curves depending on one parameter. In the case that
γ : I → M is a geodesic and η : (−ε, ε) × I → M is a
variation of γ, are there choices for η, such that every
η(s, ·) : I → M is a geodesic, not just η(0, ·) = γ? The
answer is yes, and the corresponding variational vector
fields are called Jacobi fields.

David Lindemann DG lecture 17 3. July 2020 41 / 44

 Some global Riemannian geometry

An extremely useful theorem in global Riemannian geometry is

the following:

Theorem (Hopf-Rinow)
Let (M, g ) be a Riemannian [Link] the following are
equivalent:
(M, g ) is geodesically complete.
M with the induced metrica from the Riemannian metric
g is complete as a metric space.
Every closed and boundedb subset of M is compact.
a
As in metric space.
b
W.r.t. the induced metric.

Proof: See Ch. 5, Thm. 21 in B. O’Neill, Semi-Riemannian Ge-

ometry With Applications to Relativity (1983), Pure and Applied
Mathematics, Vol. 103, Academic Press, NY.

David Lindemann DG lecture 17 3. July 2020 42 / 44

 Some global Riemannian geometry

The Hopf-Rinow Theorem can be used to prove the following

lemmata about geodesic completeness. Note that they are in
practice actually useful.

Lemma
A Riemannian manifold (M, g ) is geodesically complete if and
only if every curve with image not contained in any compact
set has infinite length.

Proof:
if (M, g ) is geodesically complete, a curve γ that is not
contained in any compact set is by the Hopf-Rinow
Theorem in particular not contained in the closure of
the geodesic ball Brg (γ(t0 )) for any t0 in the domain of γ
and any r > 0
hence, γ has infinite length
(continued on next page)

David Lindemann DG lecture 17 3. July 2020 43 / 44

 Some global Riemannian geometry

(continuation of proof)
if (M, g ) is geodesically incomplete, we can find an
inextensible geodesic γ : [0, a) → M, a > 0, of unit
speed [note: L(γ) = a < ∞]
suppose that γ([0, a)) is contained in a compactum
K ⊂M
then γ converges in K and can thus be extended as a
geodesic, which is a contradiction

Lemma
Let M be a smooth manifold and g , h Riemannian
metrics on M. Assume that for all p ∈ M and all
v ∈ Tp M, hp (v , v ) ≥ gp (v , v ), or h ≥ g for short. If
(M, g ) is geodesically complete, (M, h) is also
geodesically complete.
Let (M, g ) be a Riemannian manifold. If there exists
R > 0, such that BRg (p) is compactly embedded in M for
all p ∈ M, then (M, g ) is geodesically complete.

Proof: Exercise!
David Lindemann DG lecture 17 3. July 2020 44 / 44
 END OF LECTURE 17

Next lecture:
curvature

David Lindemann DG lecture 17 3. July 2020 45 / 44