Differential Geometry - Christian Bär

Christian Bär
GEOMETRY IN POTSDAM
Differential Geometry
Summer Term 2013
Version of August 26, 2013

The titlepage was created using 3D-XplorMath and gimp.
Contents
Preface iii
1 Manifolds 1
1.1 Topological manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Differentiable manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Tangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Directional derivatives and derivations . . . . . . . . . . . . . . . . . . . 23
1.5 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 Semi-Riemannian Geometry 35
2.1 Bilinear forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2 Semi-Riemannian metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Differentiation of vector fields . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.4 Vector fields along maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.5 Parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.6 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3 Curvature 79
3.1 The Riemannian curvature tensor . . . . . . . . . . . . . . . . . . . . . . . 79
3.2 Sectional curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.3 Ricci- and scalar curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.4 Jacobi fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4 Submanifolds 107
4.1 Submanifold of differentiable manifolds . . . . . . . . . . . . . . . . . . . 107
4.2 Semi-Riemannian submanifolds . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3 Totally geodesic submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.4 Hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.5 Trigonometry in spaces of constant curvature . . . . . . . . . . . . . . . . 133
5 Riemannian Geometry 143

5.1 The Riemannian distance function . . . . . . . . . . . . . . . . . . . . . . 143
5.2 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.3 The second variation of the energy . . . . . . . . . . . . . . . . . . . . . . 156
5.4 The Bonnet-Myers theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Literature 163
i
Contents
Index 165
ii
Preface
These are the lecture notes of an introductory course on differential geometry that I
gave in 2013. It introduces the mathematical concepts necessary to describe and ana-
lyze curved spaces of arbitrary dimension. Important concepts are manifolds, vector
fields, semi-Riemannian metrics, curvature, geodesics, Jacobi fields and much more.
The focus is on Riemannian geometry but, as we move along, we also treat more gen-
eral semi-Riemannian geometry such as Lorentzian geometry which is central for ap-
plications in General Relativity. We also make a connection to classical geometry when
we apply differential geometry to derive the laws of trigonometry on spaces of constant
curvature. One fundamental result of Riemannian geometry that we show towards the
end of the course is the Bonnet-Myers theorem. It roughly states that the larger the
curvature of a space, the smaller the space itself must be.
The lecture course did not require prior attendance of a course on elementary differ-
ential geometry treating curves and surfaces but such a course would certainly help to
develop the right intuition.
It is my pleasure to thank all those who helped to improve the manuscript by sugges-
tions, corrections or by work on the LATEX code. My particular thanks go to Andrea
Röser who wrote the first version in German language and created many pictures in
wonderful quality, to Volker Branding who translated the manuscript into English and
to Ramona Ziese who improved the layout.
Potsdam, August 2013
Christian Bär
iii
1 Manifolds
1.1 Topological manifolds
Reminder. Let M be a set. A system of sets O ⊂ P(M ) is called a topology on M , if

1. ∅, M ∈ O.
S
2. If Ui ∈ O, i ∈ I, then also Ui ∈ O.
i∈I
3. If U1 , U2 ∈ O, then also U1 ∩ U2 ∈ O.
The pair (M, O) is called a topological space. By abuse of language, one often speaks
about the topological space M rather than (M, O).
A subset U ⊂ M is called open in M if U ∈ O. A subset A ⊂ M is called closed if

M \ A ∈ O.
If both (M, OM ) and (N, ON ) are topological spaces, a map f : M → N is called con-
tinuous, if
f −1 (V ) ∈ OM for all V ∈ ON .
In other words, preimages of open sets have to be open. A bijective continuous map
f : M → N , whose inverse f −1 is also continuous, is called a homeomorphism. Two
topological spaces M and N are called homeomorphic, if there exists a homeomor-
phism between them.
Definition 1.1.2. Let M be a topological space with topology O. Then M is called an

n-dimensional topological manifold, if the following holds:
1. M is Hausdorff, that is, for all p, q ∈ M with p 6= q there exist open sets U, V ⊂
M with p ∈ U , q ∈ V and U ∩ V = ∅.
V
U b q
b p
M
1
1 Manifolds
2. The topology of M has a countable basis, that is, there exists a countable subset
B ⊂ O, such that for every U ∈ O there are Bi ∈ B, i ∈ I with
[
U= Bi .
i∈I
3. M is locally homeomorphic to Rn , that is, for all p ∈ M exists an open subset

U ⊂ M with p ∈ U , an open subset V ⊂ Rn and a homeomorphism x : U → V .
≈
x V ⊂ Rn
U
b p
M
Remark 1.1.3. The first two conditions in the definition are more of a technical nature
and are sometimes neglected. The important fact is that a topological manifold is locally
homeomorphic to Rn . Loosely speaking manifolds look locally like Euclidean space. If
the topology on M is induced by a metric, then the first condition is satisfied automat-
ically. If M is given as a subset of RN with the subset topology, then both conditions 1
and 2 are satisfied automatically.
Example 1.1.4. (1) Euclidean space M = Rn itself is an n-dimensional topological ma-

nifold:
(i), (ii) Obvious.
(iii) Holds true with U = M , V = Rn and x = id.
(2) Let M ⊂ Rn be an open subset. Then M is an n-dimensional topological manifold.
(i), (ii) Obvious.
(iii) Holds true with U = M , V = M and x = id.
(3) The standard sphere M = S n = { y ∈ Rn+1 : kyk = 1 } is an n-dimensional topo-

logical manifold.
(i), (ii) Obvious, since S n is a subset of Rn+1 .
2
(iii) We construct two homeomorphisms with the help of the stereographic projec-
tion.
We define U1 := S n \ {SP } with
SP := (−1, 0, . . . , 0) ∈ Rn+1 and set
x(y) N P b
V1 := Rn . Furthermore, we define b
x : U1 −→ V1 , b y
2
y = (y 0 , y 1 , . . . , y n ) 7−→ x(y) = · ŷ.
1 + y0
b
| {z } SP
=:ŷ
The map x is continuous and bijective. The inverse map y is given by
y : V1 −→ U1 ,
1
x 7−→ y(x) = (4 − kxk2 , 4x),
4 + kxk2
and is also continuous. Hence, x is an homeomorphism.
Analogously, we define the homeomorphism, which omits the north pole:

Let now U2 := S n \ {N P } with N P := (1, 0, . . . , 0) ∈ Rn+1 and V2 := Rn .
Then
x̃ : U2 −→ V2 ,
2
y 7−→ x̃(y) = x̃(y 0 , y 1 , . . . , y n ) = · ŷ.
| {z } 1 − y0
=:ŷ
We have seen that the sphere S n is an n-dimensional topological manifold.
(4) All n-dimensional submanifolds of RN in the sense of Analysis 3 are n-dimensional

topological manifolds.
(5) Non-Example. We consider M := { (y 1 , y 2 , y 3 ) ∈ R3 | (y 1 )2 = (y 2 )2 + (y 3 )2 }, the

double cone.
Since M ⊂ R3 , both (i) and (ii) are
satisfied.
M b x(q1 )
But M is not a 2-dimensional manifold. If q1 b
it were, then there would exist an open ≈ c x(0)

b
0 x
b
subset U ⊂ M with 0 ∈ U , an open subset b
V ⊂ R2 and a homeomorphism x : U → q2 b
x(q2 )
V with x(0) = 0. V
3
1 Manifolds
W. l. o. g. assume V = Br (x(0)) with r > 0. Choose q1 , q2 ∈ U with q11 > 0 and

q21 < 0. Furthermore, choose a continuous path c : [0, 1] → V with
c(0) = x(q1 ), c(1) = x(q2 ) and c(t) 6= x(0) for all t ∈ [0, 1].
Define the continuous path c̃ := x−1 ◦ c : [0, 1] → U . Then
c̃(0) = q1 , c̃(1) = q2 ,
that is, we have c̃1 (0) > 0 while c̃1 (1) < 0. Applying the mean value theorem we
find, that there exists a t ∈ (0, 1) with c̃1 (t) = 0. Then c̃(t) = (0, 0, 0) and conse-
quently c(t) = x(c̃(t)) = x(0), which contradicts the choice of c. Hence, M is not a
2-dimensional topological manifold.
Definition 1.1.5. If M is an n-dimensional topological manifold, the homeomor-

phisms x : U → V are called charts (or local coordinate systems) of M .
≈
U x V ⊂ Rn
After choosing a local coordinate system x : U → V every point p ∈ U is uniquely

characterized by its coordinates (x1 (p), . . . , xn (p)).
In a 0-dimensional manifold M every point p ∈ M has an open neighborhood U , which

is homeomorphic to R0 = {0}. Consequently {p} = U is an open subset of M for all
p ∈ M , that is, M carries the discrete topology. Since there exists a countable basis for
the topology on M and the topology is discrete in addition, M has to be countable itself.
Thus we get:
Proposition 1.1.6
A topological space M is a 0-dimensional topological manifold, if and only if M is countable
and carries the discrete topology.
4
Definition 1.1.7. We call a topological manifold M connected, if for every two points
p, q ∈ M there exists a continuous map c : [0, 1] → M with c(0) = p and c(1) = q.
Given two points, there has to be a continuous curve in M which connects both. Usu-
ally, in Topology one calls this path-connected, which is in the case of manifolds equiva-
lent to being connected. We do not want to go deeper into this subject at this point.
Remark 1.1.8. Following Proposition 1.1.6 every connected 0-dimensional manifold M

is given by a single point: M = {point}.
In dimension 1 there are only a few connected manifolds:
Proposition 1.1.9
Every connected 1-dimensional topological manifold is homeomorphic to R or to S 1 .
A proof of this fact can be found in the appendix of [M65]. Thus, the only compact,
connected topological manifold of dimension 1 is S 1 .
Theorem 1.1.10
Let M and A be sets. For all α ∈ A assume that Uα ⊂ M and Vα ⊂ Rn are subsets and that
xα : Uα → Vα are bijective maps. Suppose the following holds:
[
(i) Uα = M ,
α∈A
(ii) xα (Uα ∩ Uβ ) ⊂ Rn is open for all α, β ∈ A and
(iii) xβ ◦ xα −1 : xα (Uα ∩ Uβ ) → xβ (Uα ∩ Uβ ) is continuous for all α, β ∈ A.
Then M carries a unique topology for which all Uα are open sets and all xα are homeomor-
phisms.
5
1 Manifolds
Uβ
xβ
xα
Vα Uα M Vβ
xβ ◦ xα −1
Proof. We first show uniqueness:

Let O be a topology on M containing the Uα and such that the xα are homeomorphisms.
If W ∈ O, then also W ∩ Uα ∈ O and xα (W ∩ Uα ) is open for all α ∈ A.
Conversely, if W ⊂ M is a subset such that xα (W ∩ Uα ) ⊂ Rn is open for all α ∈ A, then
W ∩ Uα is also
S open in Uα for all α. Since Uα is open in M , the set W ∩ Uα is open in M .
By (i), W = α∈A (W ∩ Uα ) is also open in M . We have shown that W ∈ O if and only
if xα (W ∩ Uα ) is open in Rn for all α,
O = {W ⊂ M | xα (W ∩ Uα ) ⊂ Rn open for all α ∈ A}.
Now we show existence:

We use the criterion for openness derived in the uniqueness part of the proof to define
the topogoly. We set:
O := {W ⊂ M | xα (W ∩ Uα ) ⊂ Rn open for all α ∈ A}.
Now we have to check that this O is a topology and that it has the desired properties:
(a) O is a topology because

(i) The empty set ∅ is open in M because xα (∅ ∩ Uα ) = xα (∅) = ∅ is open in Rn
for all α. Observe that the case α = β in (ii) shows that Vα = xα (Uα ) is open in
Rn . Now we see that M ∈ O because xα (M ∩ Uα ) = xα (Uα ) = Vα is open in
Rn for all α.
S
(ii) Assume Wi ∈ O for i ∈ I. Then i∈I Wi ∈ O because
! ! !
[ [ [
xα W i ∩ Uα = x α (Wi ∩ Uα ) = xα (Wi ∩ Uα )
| {z }
i∈I i∈I i∈I open in Rn
is open in Rn for all α ∈ A.

(iii) The conclusion W1 , W2 ∈ O ⇒ W1 ∩ W2 ∈ O follows similarly.
6
(b) We have to show Uβ ∈ O for all β ∈ A. This is obvious because xα (Uβ ∩ Uα ) ⊂ Rn

is open for all α ∈ A by assumption.
(c) The map xβ is continuous for all β ∈ A because:

Let Y ⊂ Vβ be open. Then we have for all α ∈ A:
xα (xβ −1 (Y ) ∩ Uα ) = xα (xβ −1 (Y ∩ xβ (Uα ∩ Uβ )))

= (xα ◦ xβ −1 ) (Y ∩ xβ (Uα ∩ Uβ )) is open in Rn .
| {z } | {z }
=(xβ ◦xα −1 )−1 open
continuous | {z }
open
Thus xβ −1 (Y ) ∈ O.
(d) The map xβ −1 is continuous because:

Let W ⊂ Uβ be open. Then W ∈ O. For all α ∈ A the set xα (W ∩ Uα ) is open, in
particular for α = β
(xβ −1 )−1 (W ) = xβ (W ) = xβ (W ∩ Uβ ) is an open set.
Example 1.1.11 (Real-projective space). We define the real-projective space by
M = RPn := P(Rn+1 ) := {L ⊂ Rn+1 | L is one-dimensional vector-subspace }.
We will use Theorem 1.1.10 to equip RPn with the structure of an n-dimensional topo-
logical manifold. We set
A := {affine-linear embeddings α : Rn → Rn+1 with 0 6∈ α(Rn )}.
Since α is affine-linear there exist a B ∈ Mat(n × (n + 1), R) and a c ∈ Rn+1 such that
α(x) = Bx + c
for all x ∈ Rn . Since α is an embedding, B has maximal rank, rank(B) = n.
α L
0 b c b
bc
α(Rn )
Rn
Rn+1
b
0
Consequently, α(Rn ) is an affine-linear hyperplane. Set
Uα := {L ∈ RPn | L ∩ α(Rn ) 6= ∅}.
7
1 Manifolds
For L ∈ Uα the space L ∩ α(Rn ) consists of exactly one point, because otherwise we
would have L ⊂ α(Rn ) and hence 0 ∈ α(Rn ), a contradiction. Moreover, we have
RPn \ Uα = {L | L ⊂ B(Rn ) one-dimensional subspace} (1.1)
where α(x) = Bx + c. For α ∈ A set Vα := Rn and
x α : U α → Vα , xα (L) := α−1 (L ∩ α(Rn )).
Then xα is a bijective map and we have
xα −1 (v) = R · α(v).
In the following we check the assumptions of Theorem 1.1.10:

S
(i) We show: Uα = M .
α∈A
To this end, let e0 , . . . , en ∈ Rn+1 be the standard basis. For j = 1, . . . , n we define:
αj (v) := v 1 e0 + . . . + v j ej−1 + ej + v j+1 ej+1 + . . . + v n en .
Assume there existed an

n
[ n
\
L ∈ RPn \ Uαj = (RPn \ Uαj ).
j=0 j=0 b
ej b
Then
n
\
L⊂ ej ⊥ = {0}.
j=0 ej ⊥
n
S S
This is a contradiction, consequently Uαj = RPn and hence Uα = RPn .
j=0 α∈A
(ii) We observe that xα (Uα ∩ Uβ ) is α(Rn )

the complement of an affine-linear B(Rn )
subspace in Rn . More precisely, by b
0
(1.1), xα (Uα ∩ Uβ ) = α−1 (α(Rn ) \
B(Rn )) where we have written
β(x) = Bx + c. Since affine-linear
subspaces are closed, xα (Uα ∩ Uβ )
is open.
(iii) We show that xβ ◦ xα −1 : v 7→ β −1 (R · α(v) ∩ β(Rn )) is continuous for all α, β ∈ A.
8
α(Rn )
α α(v) bc
xβ ◦ xα −1 (v)
b
Vα
b
0
v b
Vβ
bc
β
β(Rn )
Write α(v) = Bv + c and β(w) = Dw + f . Now w = xβ ◦ xα −1 (v) is equivalent

to x−1 −1
β (w) = xα (v), hence to R · β(w) = R · α(v). Therefore w = xβ ◦ xα (v)
−1
is equivalent to the existence of t ∈ R such Dw + f = t · (Bv + c). For the

that
w
left-hand-side we write Dw + f = (D, f ) . Note that (D, f ) is an invertible
1
(n+1)×(n+1)-matrix because otherwise we could write f as a linear combination
of the columns of D and hence 0 would lie in the image of β. Thus we get

w
= t · (D, f )−1 · (Bv + c). (1.2)
1
Taking the scalar product with en+1 = (0, · · · , 0, 1)⊤ yields
D E
w
1 = en+1 , = t · hen+1 , (D, f )−1 · (Bv + c)i. (1.3)
1
Inserting (1.3) into (1.2) gives us

w
= hen+1 , (D, f )−1 · (Bv + c)i−1 · (D, f )−1 · (Bv + c). (1.4)
1
This shows that the components of w are rational functions of the components of
v. In particular, they are continuous.
By Theorem 1.1.10, RPn has exactly one topology for which the Uα are open and the
xα are homeomorphisms. We still need criteria ensuring that this topology is Haus-
dorff and has a countable basis. Once we know this, we have turned RPn into an n-
dimensional topological manifold.
Proposition 1.1.12 (First Addition to Theorem 1.1.10)

If in Theorem 1.1.10 there exists a countable subset A1 ⊂ A with
[
Uα = M
α∈A1
then the resulting topology has a countable basis.
9
1 Manifolds
Example 1.1.11 continued. For RPn the finite set A1 := {α0 , . . . , αn } does the job. Con-
sequently, the topology of RPn has a countable basis.
Proof of Proposition 1.1.12. The topology resulting from A has all the properties of the
topology resulting from A1 . Since the topology is unique, A and A1 give the same
topology on M .
Without loss of generality we may therefore assume that A1 = A is countable. Now the
topology of each Vα ⊂ Rn has aScountable basis Bα . Then xα −1 (Bα ) is a countable basis
of the topogoly of Uα . Finally, α∈A xα −1 (Bα ) is a countable basis of M .
Proposition 1.1.13 (Second Addition to Theorem 1.1.10)

If in Theorem 1.1.10 for any two points p, q ∈ M there is an α ∈ A such that p, q ∈ Uα , then
the topology of M is Hausdorff.
Example 1.1.11 continued. For L1 , L2 ∈ RPn there exists an affine-linear hypersurface

E with L1 ∩ E 6= ∅ and L2 ∩ E 6= ∅. L2 L1
By Proposition 1.1.13, RPn is Haus- bc
bc
dorff. Summarizing, we see that E

RPn is a n-dimensional topological
manifold.
b
0
Proof of Proposition 1.1.13. Let p, q ∈ M with p 6= q. Choose an α ∈ A with p, q ∈ Uα .

Since Rn is Hausdorff, we can choose V1 , V2 ⊂ Vα open with xα (p) ∈ V1 , xα (q) ∈ V2 and
V1 ∩ V2 = ∅. Then xα −1 (V1 ) and xα −1 (V2 ) separate p and q.
xα
b p q b
Uα
M
V1 b b
V2
R n ⊃ Vα
We summarize:
10
1.2 Differentiable manifolds
Corollary 1.1.14
Let M and A be sets and let A1 ⊂ A be a countable subset. For all α ∈ A assume that
Uα ⊂ M and Vα ⊂ Rn are subsets and that xα : Uα → Vα are bijective maps. Suppose the
following holds:
[
(i) Uα = M ;
α∈A1
(ii) xα (Uα ∩ Uβ ) ⊂ Rn is open for all α, β ∈ A;
(iii) xβ ◦ xα −1 : xα (Uα ∩ Uβ ) → xβ (Uα ∩ Uβ ) is continuous for all α, β ∈ A;
(iv) for any two points p, q ∈ M there is an α ∈ A such that p, q ∈ Uα .
Then M carries a unique topology which turns M into an n-dimensional topological manifold
such that the xα : Uα → Vα are charts.
Example 1.1.15 (Complex-projective space). In complete analogy to the real-projective

space we define complex-projective space by
CPn := P(Cn+1 ) := {L ⊂ Cn+1 | L is one-dimensional complex subspace }.
Like in the real case we obtain charts xα : Uα → Cn = R2n . This turns CPn into a
2n-dimensional topological manifold.

For a topological manifold M , like for any topological space, it makes sense to speak
about continuous functions f : M → R. In a course on differential geometry we will
certainly need to differentiate functions. But what does differentiability of f mean?
Attempt of a definition. The function f is called differentiable at p ∈ M if for some chart

x : U → V with p ∈ U the function f ◦ x−1 : V → R is differentiable in x(p).
x V
p b
x(p)
U b
M
f
R f ◦ x−1
This is, in principle, a very reasonable definition. It means that f is differentiable on M

if it is differentiable on Rn when expressed in coordinates. But there is a problem with
11
1 Manifolds
this definition. If y : Ũ → Ṽ is another chart with p ∈ Ũ , then near y(p) we have
f ◦ y −1 = (f ◦ x−1 ) ◦ (x ◦ y −1 ) .
| {z } | {z }
diff’able only
at x(p) continuous
This concept of differentiability depends on the choice of chart x and this is really bad
because on a general topological manifold there are no preferred coordinate systems.
The sad truth is that there is no reasonable concept of differentiable functions on a
topological manifold.
But there is one thing we can do, we can refine the notion of a manifold. If x ◦ y −1 were
a diffeomorphism and not only a homeomorphism, then the differentiability of f ◦ x−1
would imply the differentiability of f ◦ y −1 . We enforce this by making the following
definition.
Definition 1.2.1. Let M be an n-dimensional topological manifold. Two charts

x : U → V and y : Ũ → Ṽ of M are called C ∞ -compatible if
y ◦ x−1 : x(U ∩ Ũ ) → y(U ∩ Ũ )
is a C ∞ -diffeomorphism.
Ũ
y
x
V U M Ṽ
x(U ∩ Ũ ) y(U ∩ Ũ )
y ◦ x−1
Definition 1.2.2. A set of charts xα : Uα → Vα of M , α ∈ A, is called atlas of M , if

[
Uα = M.
α∈A
An atlas A is called a C ∞ -atlas if any two charts in A are C ∞ -compatible.
12
Example 1.2.3
(1) Let M = U ⊂ Rn be open. Then A := {id : U → U } is a C ∞ -atlas.
(2) Let M = S n and A := {(x : U1 → V1 ), (x̃ : U2 → V2 )}, where U1 := S n \ {SP },
U2 := S n \ {N P } and V1 := V2 := Rn , compare Example 1.1.4.3. Furthermore, let
2 0

x(y) = ŷ, where y = y , ŷ ∈ Rn+1 ,
1 + y0
1 2

y(x) = 4 − kxk , 4x and
4 + kxk2
2
x̃(y) = ŷ.
1 − y0
Then we have for v ∈ x(U1 ∩ U2 ) = x(S n \ {SP, N P }) = Rn \ {0}:

−1 4 − kvk2 4v
x̃ ◦ x (v) = x̃ ,
4 + kvk2 4 + kvk2
2 4v
= 2
4 − kvk 4 + kvk2
1−
4 + kvk2
8v
=
4 + kvk − 4 + kvk2
2
4v
= .
kvk2
Hence x̃ ◦ x−1 is C ∞ on Rn \ {0} = x(S n \ {SP, N P }) = x(U1 ∩ U2 ). Similarly one
sees that x ◦ x̃−1 is smooth. This shows that x and x̃ are C ∞ -compatible. Hence A
is a C ∞ -atlas.
(3) Let M = RPn , A := {xα : Uα → Rn | xα is an affine-linear embedding Rn → Rn+1
of maximal rank and 0 6∈ α(Rn )}, compare Example 1.1.11. All changes of charts
xβ ◦ xα −1 are rational functions and hence C ∞ . Therefore A is a C ∞ -atlas.
(4) Analogously, for M = CPn as in Example 1.1.15, the resulting atlas is also a C ∞ -
atlas.
Remark 1.2.4. If A is a C ∞ -atlas of M then

Amax := {charts x of M | x is C ∞ -compatible with all charts in A}
also is a C ∞ -atlas of M . The reason is this:

If x and x̃ are two charts of M , which are C ∞ -compatible with all charts in A, then also x and
x̃ are C ∞ -compatible with each other.
Namely, for any p ∈ x(U ∩ Ũ) there exists a chart y : Ũ ˜ → Ṽ˜ in A with x−1 (p) ∈ Ũ ˜ . Near
p we then have:
x̃ ◦ x−1 = x̃ ◦ y −1 ◦ y ◦ x−1 .
| {z } | {z }
C∞ C∞
13
1 Manifolds
Hence x̃ ◦ x−1 is C ∞ and similarly for x ◦ x̃−1 .
Definition 1.2.5. An C ∞ -atlas Amax is called maximal (or also differentiable struc-
ture), if every chart that is C ∞ -compatible with all charts in Amax , is already con-
tained in Amax .
According to Remark 1.2.4, every C ∞ -atlas A is contained in exactly one maximal C ∞ -

atlas Amax .
Definition 1.2.6. A pair (M, Amax ), where M is an n-dimensional topological mani-

fold and Amax a differentiable structure on M , is called an n-dimensional differen-
tiable manifold.
Definition 1.2.7. Let M and N be differentiable manifolds, let p ∈ M and let

k ∈ N ∪ {∞}.
A continuous map f : M → N is called k-times continuously differentiable (or C k )
near p, if for one (and therefore for every other) chart
(x : U → V ) ∈ Amax (M ) with p ∈ U
and for one (and therefore for every other) chart
(y : Ũ → Ṽ ) ∈ Amax (N ) with f (p) ∈ Ũ
there exists a neighborhood W ⊂ x(f −1 (Ũ ) ∩ U ) of x(p), such that

y ◦ f ◦ x−1 : x f −1 (Ũ ) ∩ U → Ṽ
is C k on W .
14
p b
f f (p) b
U Ũ N
f −1 (Ũ ) ∩ U
M x y
y◦f ◦ x−1
V Ṽ
x(p)
b
−1
W x(f (Ũ ) ∩ U )
Example 1.2.8
(1) Let M = S n with the differentiable structure given by
A = {(x : U1 → V1 ), (x̃ : U2 → V2 )}
as in Example 1.2.3.2. We show that
f : Sn → Sn, f (y) = −y,
is C ∞ near N P . In fact, f is C ∞ on all of S n . We compute

n x−1 −1 4 − kvk2 4v
R ∋ v 7−→ x (v) = ,
4 + kvk2 4 + kvk2

f 4 − kvk2 −4v
7−→ − ,
4 + kvk2 4 + kvk2
x̃ 2 4v 8v
7−→ − 2 · 2
=− = −v
4 − kvk 4 + kvk 8
1+
4 + kvk2
Consequently, x̃ ◦ f ◦ x−1 (v) = −v and in particular x̃ ◦ f ◦ x−1 is C ∞ on Rn . Thus,

we may consider W = Rn .
This argument shows that f is smooth near all points except SP because SP is the
only point not contained in the chart U1 . Interchanging the two charts one sees
similarly that f is also smooth near SP . Hence f is smooth on all of Sn .
(2) We consider the atlases A1 := {x = id : R → R} on M = R with differentiable

structure A1,max and A2 := {x̃ : R → R} with x̃(t) = t3 and differentiable structure
A2,max .
√
Now x̃ ◦ x−1 (t) = t3 is C ∞ , but x ◦ x̃−1 (t) = 3 t is not.
Consequently, x and x̃ are not C ∞ -compatible and therefore the differentiable struc-
tures are different:
A1,max 6= A2,max .
15
1 Manifolds
• Is id : (R, A1,max ) → (R, A2,max ) a C ∞ -map?

id
R R
x=id x̃ ⇒ x̃ ◦ id ◦ x−1 is C ∞ and therefore also id.
t7→t3
R R
• Is id : (R, A2,max ) → (R, A1,max ) a C ∞ -map?

id
R R
x̃
√
x=id ⇒ x ◦ id ◦ x̃−1 is not C ∞ and the same holds true for id.
3
t7→ t
R R
Summarizing we see that id is a homeomorphism from (R, A1,max ) to (R, A2,max )

which is smooth but its inverse is not.
Definition 1.2.9. Let M and N be differentiable manifolds. A homeomorphism

f : M → N is called a C k -diffeomorphism, if f and f −1 are both C k . Instead of C ∞ -
diffeomorphism we simply say diffeomorphism. If there exists a diffeomorphism
f : M → N , we say that M and N are diffeomorphic.
Example 1.2.8.2 continued. Let M = (R, A1,max ) with A1,max = {x = id : R → R} and

N = (R, A2,max ) with A2,max = {x̃ : R → R, x̃(t) =
√ t3 }. We have seen that id : M → N
is not a diffeomorphism. But f : M → N, f (t) = 3 t is a diffeomorphism because
f
M N
x x̃
id
R R
Thus M and N are diffeomorphic.
Question. Is every differentiable structure on Rn diffeomorphic to the standard struc-

ture Amax , the one induced by A = {x = id : Rn → Rn }?
The answer is quite surprising. For n = 0, 1, 2, 3 and also for n ≥ 5 it is Y ES. But for
n = 4 it turns out to be N O. There exist uncountably many differentiable structures on
R4 which are pairwise not diffeomorphic (so-called exotic structures). The proof of these
facts is far beyond the scope of our lecture course.
Remark 1.2.10. In 1956 John Milnor showed that there exist exotic n-dimensional
spheres for n ≥ 7. These are differentiable manifolds which are homeomorphic to
Sn but not diffeomorphic. But in every dimension there are only finitely many.
16
1.3 Tangent vectors
1.3 Tangent vectors

Question. What is the derivative at a point of a differentiable map between differenti-
able manifolds?
The vague answer is: It is the linear approximation of the map at that point. But what
do we mean by the linear approximation in a point of a differentiable manifold? For this
to make sense we first need a concept of “linear approximation” of a manifold at a given
point.
Definition 1.3.1. Let M be a differentiable manifold and p ∈ M .

A tangent vector on M at the point p is an equivalence class of differentiable curves
c : (−ε, ε) → M with ε > 0 and c(0) = p, where two such curves c1 : (−ε1 , ε1 ) → M
and c2 : (−ε2 , ε2 ) → M are called equivalent, if for a chart x : U → V with p ∈ U we
have:
d d
(x ◦ c1 )|t=0 = (x ◦ c2 )|t=0 .
dt dt
Remark 1.3.2. This definition does not depend on the choice of the chart x : U → V .
Namely, if y : Ũ → Ṽ is another chart with p ∈ Ũ then we get by the chain rule

d d d
(y ◦ c)|t=0 = y ◦ x−1 ◦ (x ◦ c) |t=0 = D y ◦ x−1 |x(p) (x ◦ c)|t=0 . (1.5)
dt dt dt
Therefore the condition

d d
(x ◦ c1 )|t=0 = (x ◦ c2 )|t=0
dt dt
is equivalent to the condition
d d
(y ◦ c1 )|t=0 = (y ◦ c2 )|t=0 .
dt dt
Notation 1.3.3. We denote the equivalence class of c by ċ(0).
17
1 Manifolds
Definition 1.3.4. The set
Tp M := {ċ(0) | c : (−ε, ε) → M differentiable with c(0) = p}
is called tangent space of M at the point p.
Lemma 1.3.5
Let M be an n-dimensional differentiable manifold, let p ∈ M and let x : U → V be a chart
of M with p ∈ U . Then the map
d
dx|p : Tp M → Rn , ċ(0) 7→ (x ◦ c)|t=0 ,
dt
is well defined and bijective.
Proof. Well-definedness and injectivity are clear from to the definition of the equiv-
alence relation that defines ċ(0). To show surjectivity let v ∈ Rn and set
−1
c(t) := x (x(p) + tv). Choose ε > 0 so small that x(p) + tv ∈ V whenever |t| < ε.
Then we have
d d
dx|p (ċ(0)) = x ◦ x−1 x(p) + tv |t=0 = x(p) + tv |t=0 = v.
dt dt
x
b
x(p)
b p v
U
M V
This shows surjectivity and concludes the proof.
18
1.3 Tangent vectors
Definition 1.3.6. We equip Tp M with the unique vector space structure for which
dx|p becomes a linear isomorphism. In other words, for a, b ∈ R and
c1 : (−ε1 , ε1 ) → M , c2 : (−ε2 , ε2 ) → M we set:

a · ċ1 (0) + b · ċ2 (0) := (dx|p )−1 a · dx|p ċ1 (0) + b · dx|p ċ2 (0) .
Lemma 1.3.7
The vector space structure on Tp M does not depend on the choice of chart x : U → V .
Proof. Let y : Ũ → Ṽ be another chart with p ∈ Ũ . We have to show that the map
dy|p : Tp M → Rn is also linear with respect to the vector space structure induced by x.
This holds true since by (1.5)

dy|p = D y ◦ x−1 |x(p) ◦ dx|p
| {z } |{z}
linear linear
is the composition of two linear maps.
We may think of the tangent space Tp M as the linear approximation to M at p. Now we

can define the differential of a differentiable map between manifolds.
Lemma 1.3.8
Let M and N be differentiable manifolds, let p ∈ M , and let f : M → N be differentiable
near p. Then the map
df |p : Tp M → Tf (p) N, ċ(0) 7→ (f ◦ c)˙(0),

is well defined and linear.
ċ(0) f df |p (ċ(0))
b
c p f (p)b
f ◦c
M
N
19
1 Manifolds
Proof. We choose a chart x : U → V of M with p ∈ U and a chart y : Ũ → Ṽ of N with

f (p) ∈ Ũ . We compute, using the chain rule,
dy|f (p) ((f ◦ c)˙(0)) = (y ◦ f ◦ c)˙(0)

˙
= y ◦ f ◦ x−1 ◦ (x ◦ c) (0)

= D y ◦ f ◦ x−1 |x(p) · (x ◦ c)˙(0)

= D y ◦ f ◦ x−1 |x(p) · dx|p (ċ(0)).
Consequently, we have
−1
df |p = dy|f (p) ◦ D(y ◦ f ◦ x−1 )|x(p) ◦ dx|p .
In particular, df |p is well defined (independently of the choice of c) and linear.
Definition 1.3.9. The map df |p is called the differential of f at the point p.
Remark 1.3.10. If U ⊂ M is an open subset, then the differential of the inclusion map
ι : U ֒→ M is the canonical isomorphism dι : Tp U → Tp M , given by
ċ(0) 7→ (ι ◦ c)· (0) = ċ(0).
We will identify tangent spaces via this isomorphism and simply write Tp U = Tp M .
Remark 1.3.11. If M is an n-dimensional R-vector

space, then M and Tp M are canonically isomorphic via
b
M → Tp M, v p
v 7→ ċp,v (0), b cp,v
0
where cp,v (t) := p + tv.
Remark 1.3.12. For a chart x : U → V the differential dx|p has two meanings which are
20
1.3 Tangent vectors
related by this canonical isomorphism. The following diagram commutes:

ċ(0) (x ◦ c)˙(0)
∈
∈
dx|p
Tp U Tx(p) V = Tx(p) Rn
∼
dx|p ∼
= =
al
onic
d can rphism
dt (x ◦ c)|t=0 ∈ Rn isom
o
Theorem 1.3.13 (Chain Rule)

Let M , N and P be differentiable manifolds and let p ∈ M . Assume f : M → N and
g : N → P are differentiable near p and near f (p), respectively. Then the following holds:
d(g ◦ f )|p = dg|f (p) ◦ df |p .
Proof. For a curve c : (−ε, ε) → M with c(0) = p we have:

d
d(g ◦ f )|p (ċ(0)) = (g ◦ f ) ◦ c |t=0
dt
d
= g ◦ (f ◦ c) |t=0
dt
= dg|f (p) (f ◦ c)˙(0)

= dg|f (p) df |p ċ(0) .
This proof of the chain rule was very simple. One may wonder why the proof of the
chain rule that one remembers from one’s course on calculus of several variables re-
quired a lot more work. The reason for the simplicity here is that one has already built
the chain rule into the definition of the differential of a map.
Definition 1.3.14. Let M and N be differentiable manifolds. Let k ∈ N ∪ {∞}. A

surjective C k -map f : M → N is called a local C k -diffeomorphism, if for all p ∈ M
there exists an open neighborhood U of p in M and an open neighborhood V of f (p)
in N , such that
f |U : U → V
is a C k -diffeomorphism.
21
1 Manifolds
Example 1.3.15. Let f : R → S1 , f (t) = eit . R

Then f is not injective (in particular, not a dif- b f (t0 )
U t0
b
feomorphism), but it is a local diffeomorphism: S1

f b
For t0 ∈ R choose U := (t0 − π, t0 + π) and 0
V := S1 \ {−f (t0 )}. −f (t0 ) bc V
Remark 1.3.16. If f : M → N is a local C k -diffeomorphism, then

df |p : Tp M → Tf (p) N
is an isomorphism. In particular, we have dim(Tp M ) = dim(Tf (p) N ) and therefore also

dim M = dim N .
Proof. W.l o.g. let f be a C k -diffeomorphism. For a curve c : (−ε, ε) → M with c(0) = p
we have:
d(idM )|p ċ(0) = (idM ◦ c)˙(0) = ċ(0)
and hence
d(idM )p = idTp M .
Applying the chain rule we find:

idTp M = d(idM )|p = d f −1 ◦ f |p = df −1 |f (p) ◦ df |p .
Analogously, we can derive df |p ◦ df −1 |f (p) = idTf (p) N . Therefore we get:

df −1 |f (p) = (df |p )−1 .
The converse of the last statement is also true:
Theorem 1.3.17 (Inverse Function Theorem)

Let M and N be differentiable manifolds and let p ∈ M . Let f : M → N be a C k -map,
k ≥ 1.
If df |p : Tp M → Tf (p) N is an isomorphism, then there exists an open neighborhood U of p
in M and an open neighborhood Ũ of f (p) in N , such that
f |U : U → Ũ
is a C k -diffeomorphism.
22
1.4 Directional derivatives and derivations
Proof. Choose a chart x : U1 → V1 of M with p ∈ U1 and a chart y : U2 → V2 of N with

f (p) ∈ U2 .
U p b f f (p) Ũ b
U2 N
U1 f −1 (U2 )
x y
M
y ◦ f ◦ x−1 |V
V1 Ṽ V2
x(p)
b
b
−1
y(f (p))
V x(f (U2 ) ∩ U1 )
On x(U1 ∩ f −1 (U2 )) the map y ◦ f ◦ x−1 is defined. Since df |p is invertible, we also have
that D(y ◦ f ◦ x−1 )|x(p) is invertible.
The ”classical” inverse function theorem says that there exists an open neighborhood
V ⊂ x(U1 ∩ f −1 (U2 )) of x(p) and an open neighborhood Ṽ ⊂ V2 of y(f (p)), such that
y ◦ f ◦ x−1 |V : V → Ṽ
is a C k -diffeomorphism. With U := x−1 (V ) and Ũ := y −1 (Ṽ ) it follows that

f |U : U → Ũ is a C k -diffeomorphism.
Definition 1.4.1.
Let M be a differentiable manifold, let p ∈ M and and
let ċ(0) ∈ Tp M . For a function f : M → R, differenti- ċ(0) b
able near p, we call p

c
d
∂ċ(0) f := df |p (ċ(0)) = (f ◦ c)|t=0 ∈ R M
dt
the directional derivative of f in the direction ċ(0).
23
1 Manifolds
Notation 1.4.2. For U ⊂ M open and k ∈ N ∪ {∞}, we write
C k (U ) := {f : U → R | f is C k }.
For α ∈ R, f ∈ C k (U ) and g ∈ C k (Ũ ) we set
α · f ∈ C k (U ), (α · f )(q) := α · f (q)
k
f + g ∈ C (U ∩ Ũ ), (f + g)(q) := f (q) + g(q)
f · g ∈ C k (U ∩ Ũ ), (f · g)(q) := f (q) · g(q)
and [
C∞
p := C ∞ (U ).
U open
p∈U
Definition 1.4.3. A map ∂ : Cp∞ → R is called derivation at p if the following condi-

tions are satisfied:
(i) Locality: If Ũ ⊂ U is open, p ∈ Ũ , f ∈ C ∞ (U ), then
∂f = ∂(f |Ũ ).
(ii) Linearity: If α, β ∈ R, f, g ∈ Cp∞ , then
∂(αf + βg) = α∂f + β∂g.
(iii) Leibniz Rule: For f, g ∈ Cp∞ we have
∂(f · g) = ∂f · g(p) + f (p) · ∂g.
Example 1.4.4
∂
(1) Let M = Rn and p ∈ M . Then ∂ = ∂xi p
is a derivation.
(2) Let M be an arbitrary differentiable manifold, let p ∈ M and ċ(0) ∈ Tp M . Then ∂ċ(0)
is a derivation. We check (iii):
d
∂ċ(0) (f · g) = (f · g) ◦ c |t=0
dt
24
d
= (f ◦ c) · (g ◦ c) |t=0
dt
d d
= (f ◦ c)|t=0 · g c(0) + f (c(0)) · (g ◦ c)|t=0
dt dt
= ∂ċ(0) f · g(p) + f (p) · ∂ċ(0) g.
The other two conditions are even simpler to verify.
p ) of all derivations at p forms an R-vector space via

Remark 1.4.5. The set Der(C ∞
(α∂1 + β∂2 )(f ) = α∂1 f + β∂2 f.
Lemma 1.4.6
The map ∂· : Tp M → Der(Cp∞ ), ċ(0) 7→ ∂ċ(0) , is linear.
Proof. Let x : U → V be a chart of M with p ∈ U . By the definition of the vector space

structure on Tp M , we have to show that ∂˙ ◦ (dx|p )−1 is linear. Assume v ∈ Rn and put
c(t) := x−1 x(p) + tv . We find:

(∂· ◦ (dx|p )−1 (v))(f ) = df |p (dx|p )−1 (v)

= df |p ċ(0)
d
= f ◦ c(t)) |t=0
dt
d
= f ◦ x−1 (x(p) + tv) |t=0
dt
= grad f ◦ x−1 |x(p) , v .
This expression is linear in v.
Remark 1.4.7. Let e1 , . . . , en be the standard basis of Rn . Then

(dx|p )−1 (e1 ), . . . , (dx|p )−1 (en ) form a basis of Tp M . We find
∂(f ◦ x−1 ) ∂f
∂(dx|p )−1 (ej ) (f ) = grad f ◦ x−1 |x(p) , ej = =:
∂xj x(p) ∂xj p
25
1 Manifolds
≈
U x V ⊂ Rn
(dx|p )−1 (e 2)
e2
p b
(dx|p )−1 (e1 ) b
e1
M x(p)
For every chart x we have the derivations

∂ ∂
,..., .
∂x1 p ∂xn p
Proposition 1.4.8
Let M be a differentiable manifold and let p ∈ M . Then the map
∂· : Tp M → Der(Cp∞ ), ċ(0) 7→ ∂ċ(0) ,
is an isomorphism. In particular, every derivation is a directional derivative and for every

chart x : U → V with p ∈ U
∂ ∂
1
,...,
∂x p ∂xn p
is a basis of Der(Cp∞ ).
Proof. It suffices to show that the derivations

∂ ∂
,...,
∂x1 p ∂xn p
form a basis of Der(Cp∞ ). Namely, then we know that the linear map ∂· maps the basis
(dx|p )−1 (e1 ), . . . , (dx|p )−1 (en ) of Tp M onto the basis ∂x∂ 1 p , . . . , ∂x∂ n p of Der(Cp∞ ) and is
hence an isomorphism.
n
X ∂
a) Linear Independence: Let αi = 0. We have to show: α1 = . . . = αn = 0.
∂xi p
i=1
Choose f = xj . Then
n
X ∂xj
0= αi = αj for j = 1, . . . , n.
∂xi p
i=1 | {z }
=δij
26
b) Generating Property: Let δ ∈ Der(Cp∞ ). Set αj := δ(xj ) for j = 1 . . . , n. We will show

that
n
X ∂
δ= αj · .
∂xj p
j=1
b1) We have
(iii)
δ(1) = δ(1 · 1) = δ(1) · 1 + 1 · δ(1) = 2δ(1)
and hence δ(1) = 0. Now let α ∈ R. Then we find
(ii)
δ(α) = δ(α · 1) = α · δ(1) = 0.
Consequently, derivations vanish on all constant functions.
b2) Let f ∈ Cp∞ , more precisely f ∈ C ∞ (Ũ ) with p ∈ Ũ open. Choose a neighbor-
˜ of p with Ũ
hood Ũ ˜ ⊂ Ũ ∩ U and x(Ũ
˜ ) = B(x(p), r).
x
Ũ
b
x(p)
˜ p
b
U Ũ B(x(p),r)
M V
x(U ∩ Ũ )
Lemma 1.4.9 (see below) with h = f ◦ x−1 says that there exist g1 , . . . , gn ∈
C ∞ (B(x(p), r)) such that
n
X

f ◦ x−1 (x) = f ◦ x−1 (x(p)) + xi − xi (p) · gi (x) and
i=1
∂(f ◦ x−1 )
i
(x(p)) = gi x(p) .
∂x
It follows that
27
1 Manifolds
(i)
δ(f ) = δ(f | ˜ )
Ũ
n
X

= δ f (p) + xi − xi (p) (gi ◦ x)
i=1
(ii) n
X
(b1)
= δ xi − xi (p) (gi ◦ x)
i=1
n
X
(iii)
= δ xi − xi (p) · gi x(p) + xi − xi (p) |p δ(gi ◦ x)
i=1
| {z }
=0
(ii) n
X
(b1)
= δ xi gi x(p)
i=1
n
X ∂f
= αi · .
∂xi p
i=1
Lemma 1.4.9
Let h ∈ C ∞ (B(q, r)). Then there exist g1 , . . . , gn ∈ C ∞ (B(q, r)) with
n
X
(i) h(x) = h(q) + xi − q i gi (x) and
i=1
∂h
(ii) (q) = gi (q).
∂xi
Proof. For x ∈ B(q, r) set wx : [0, 1] → R, wx (t) := h(tx + (1 − t)q). It follows that
h(x) − h(q) = wx (1) − wx (0)
Z1
= ẇx (t) dt
0
Z1 X
n
∂h
= · (xi − q i ) dt
∂xi tx+(1−t)q
0 i=1
n Z1
X
i i ∂h
= (x − q ) dt
∂xi tx+(1−t)q
i=1 0
| {z }
=: gi (x)
28
With this definition of the gi , (i) holds. Moreover, (ii) follows from (i) by differentiation
at q.
At this point we have the following situation for a differentiable manifold:

∂· ∂
(dx|p )−1 (ej ) ∈ Tp M Der(Cp∞ ) ∋
∼
= ∂xi p
∼
= ∼
=
dx|p ∂ ◦ (dx|p )−1
Rn
|{z}
∋ ej
depend on
the choice of x
From now on we identify Tp M with Der(Cp∞ ) via the isomorphism ∂· . For example, we
write for ξ ∈ Tp M
n
X ∂
ξ= ξi
∂xi p
i=1
n
P n
P
∂
instead of ∂ξ = ξ i ∂x i p and ξ = ξ i (dx|p )−1 (ei ) where (ξ 1 , . . . , ξ n )⊤ = dx|p (ξ).
i=1 i=1
Question. How do the coefficients ξ 1 , . . . , ξ n of a tangent vector change, if we replace

the chart x by another chart y?
Let ξ ∈ Tp M , let x and y be charts, both containing p. We express ξ with respect to both
charts
n n
i ∂ ∂
X X
ξ= ξ i
= ηj .
∂x p ∂y j p
i=1 j=1
Now we want to compute the coefficients ξ i in terms of the η j and vice versa. Using the
Chain Rule (Theorem 1.3.13) we compute
    1   1
ξ1 η η
 ..   −1  ..  −1  .. 
 .  = dx|p (ξ) = (dx|p ) (dy|p )  .  = D(x ◦ y )|y(p)  .  .
ξn ηn ηn
! !
η1 ξ1
Interchanging the roles of x and y, we also get .. = D(y ◦ x−1 )|x(p) .. . Thus
.n .n
η ξ
n
X ∂(y j ◦ x−1 )
ηj = · ξi (1.6)
∂xi x(p)
i=1
29
1 Manifolds
In the physics literature this transformation rule is put at the heart of the definition of a
tangent vector, then usually called a contravariant vector. For a physicist, a contravariant
vector is a vector (ξ 1 , . . . , ξ n ) associated to a chart which transforms as in (1.6) when the
chart is changed. We have now understood that this vector is the coefficient vector of
an (abstractly defined) tangent vector with respect to the basis ∂x∂ 1 p , . . . , ∂x∂ n p of Tp M
induced by the chart x.
∂ 1 n ⊤ = e . By (1.6), we get
Let us look at the special case ξ = ∂x i p , that is, (ξ , . . . , ξ ) i
n
∂ X ∂
= ηj
∂xi p ∂y j p
j=1
n Xn
X ∂(y j ◦ x−1 ) ∂
= ξk ·
∂xk x(p) ∂y j p
j=1 k=1
n
X ∂(y j ◦ x−1 ) ∂
= · ,
∂xi x(p) ∂y j p
j=1
hence
n
∂ X ∂(y j ◦ x−1 ) ∂
= · (1.7)
∂xi p ∂xi x(p) ∂y j p
j=1
In the physics literature it is customary to use the Einstein summation convention mean-
ing that when an index appears twice in an expression, once as an upper index and
once as a lower index, then summation over this index is understood. So (1.7) would
be written as
∂ ∂(y j ◦ x−1 ) ∂
i
= i
· j
∂x p ∂x x(p) ∂y p
or even shorter as
∂ ∂y j ∂
= · .
∂xi ∂xi ∂y j
This makes formula (1.7) easy to memorize; we simply cancel ∂y j . In these lecture notes
we will not use the Einstein summation convention unless explicitly stated otherwise.
But when you do computations for yourself, the Einstein summation convention can
be quite convenient and is recommended as long as you are aware of it.
30
1.5 Vector fields
1.5 Vector fields

Next we want to introduce vector fields. Vector fields are maps
which associate to each point of a manifold a tangent vector
in the corresponding tangent space. Hence the target space
is varying and depends on the point. For this reason we first
need to introduce the tangent bundle.
1
Definition 1.5.1. Let M be a differentiable manifold. Then we call

[
T M := Tp M
p∈M
the tangent bundle of M .
We equip T M with the structure of a differentiable manifold. Denote the differentiable

structure of M by AM,max . Let π : T M → M , π(ξ) = p for ξ ∈ Tp M be the “footpoint
map”. For every chart x : U → V in AM,max we construct a chart Xx : Ux → Vx of T M
as follows: We set
Ux := π −1 (U ) ⊂ T M,
Vx := V × Rn ⊂ R2n and

Xx (ξ) := x π(ξ) , dx|π(ξ) (ξ) .
Then we have Xx −1 (v, w) = (dx|x−1 (v) )−1 (w).

Schematic picture:
TM Ux
Xx Vx
U V
M x
By construction we have:
[
Ux = T M.
(x:U →V )
∈AM,max
1
Source: http://www.weatheronline.co.uk
31
1 Manifolds
Let x and y be charts on M . Then we have:

Xy ◦ Xx −1 (v, w) = Xy (dx|x−1 (v) )−1 (w)

= y π (dx|x−1 (v) )−1 (w) , dy|π((dx|x−1 (v) )−1 (w)) (dx|x−1 (v) )−1 (w)

= y x−1 (v) , dy|x−1 (v) (dx|x−1 (v) )−1 (w)

= y ◦ x−1 (v), D y ◦ x−1 |v ·w .
| {z } | {z }
C∞ C∞
Hence Xy ◦ Xx −1 is a C ∞ -diffeomorphism, in particular, it is a homeomorphism. By

Theorem 1.1.10, T M carries exactly one topology, for which the Xx are homeomor-
phisms.
We show: The topology of T M has a countable basis. Since the topology of M has a count-
able basis, M has a countable C ∞ -atlas. Then the corresponding (countably many)
charts of T M suffice to cover T M . By Proposition 1.1.12 the topology of T M has a a
countable basis.
We show: T M is Hausdorff. Let ξ, η ∈ T M with ξ 6= η. We consider two cases.
Case 1: π(ξ) 6= π(η).
TM
Since M is Hausdorff there exists an open neighbor-
hood U1 of π(ξ) and an open neighborhood U2 of π(η) ξ U1
bc
such that U1 ∩ U2 = ∅. The sets π −1 (U1 ) and π −1 (U2 ) U2

are open neighborhoods of ξ and η with η bc
M
π −1 (U1 ) ∩ π −1 (U2 ) = ∅.
Case 2: π(ξ) = π(η). TM

Let x : U → V be a chart of M with π(ξ) = π(η) ∈ ξ bc
U . Then we have ξ, η ∈ π −1 (U ) = Ux . The proof of

Proposition 1.1.13 shows that we can separate ξ and U bc
η
M
η.
We summarize: The tangent bundle T M carries a unique topology turning it into a

2n-dimensional topological manifold with atlas
AT M = {Xx : Ux → Vx | (x : U → V ) ∈ AM,max }.
Since the changes of charts Xx ◦ Xy −1 are not only homeomorphisms but C ∞ -

diffeomorphisms, we find that AT M is a C ∞ -atlas. Hence (T M, AT M,max ) becomes a
2n-dimensional differentiable manifold.
Remark 1.5.2. The footpoint map π : T M → M is expressed in the charts x : U → V of
M and Xx : Ux → Vx of T M by
x ◦ π ◦ Xx−1 : V × Rn → V, (v, w) 7→ v.
In particular, π is a smooth map.
32
1.5 Vector fields
Definition 1.5.3
A map ξ : M → T M is called a vector field on M , ξ(p)
if for all p ∈ M we have p b
M

π ξ(p) = p.
Remark 1.5.4. Let x : U → V be a chart of M . A vector field ξ on U is characterized by

coefficient functions
ξ1, . . . , ξn : V → R
for which
n
X ∂
ξ(p) = ξ i x(p) .
∂xi p
i=1
Since a vector field is a map from the differentiable manifold M to the differentiable
manifold T M we know what it means that the vector field is C k . We investigate how
this can be characterized in terms of the coefficient functions. For the chart x of M we
consider the corresponding chart Xx on T M . The commutative diagram
ξ
M TM
∪ ∪
−1 ξ|U
x (v) ∈ U Ux ∋ ξ(x−1 (v))
x Xx
=x−1 (v)
z }| {
v∈ V V × Rn ∋ x π ξ(x−1 (v)) , dx| ξ(x−1 (v))
ξ(x−1 (v))
1 n
π
∩ ∩ = v, ξ (v), . . . , ξ (v)
Rn R2n

shows that ξ corresponds in these coordinates to the map v 7→ v, ξ 1 (v), . . . , ξ n (v) .
Thus ξ is C k on U if and only if the coefficient functions ξ 1 , . . . , ξ n are C k on V .
Example 1.5.5.
We consider M = R2 with polar coordinates. For ϕ0 ∈ R we set U :=
cos ϕ0
R2 \ R≥0 · , V := (0, ∞) × (ϕ0 , ϕ0 + 2π) and y : U → V such that
sin ϕ0
y −1 (r, ϕ) := (r cos ϕ, r sin ϕ).
∂
On U the vector field ξ := r is defined. Using (1.7) we express this vector field in
∂r
33
1 Manifolds
terms of Cartesian coordinates, i.e., with respective to the chart x = id : R2 → R2 :
∂
ξ = r
∂r
1
∂x ∂ ∂x2 ∂
= r +
∂r ∂x1 ∂r ∂x2

∂(r cos ϕ) ∂ ∂(r sin ϕ) ∂
= r +
∂r ∂x1 ∂r ∂x2

∂ ∂
= r cos ϕ 1 + sin ϕ 2
∂x ∂x
∂ ∂
= x1 1 + x2 2 .
∂x ∂x
In Cartesian coordinates:
ξ 1 (x1 , x2 ) = x1 ,
ξ 2 (x1 , x2 ) = x2 .
b
In polar coordinates:
η 1 (r, ϕ) = r,
η 2 (r, ϕ) = 0.
∂
Similarly, we can express the vector field
∂ϕ
in Cartesian coordinates:
∂ ∂x1 ∂ ∂x2 ∂
= + b
∂ϕ ∂ϕ ∂x1 ∂ϕ ∂x2
∂ ∂
= −r sin ϕ 1 + r cos ϕ 2
∂x ∂x
2 ∂ 1 ∂
= −x +x .
∂x1 ∂x2
34
2 Semi-Riemannian Geometry
On topological manifolds one can consider continuous maps. In order to be able to
define differentiable maps we had to add structure to a topological manifold which
gave rise to differentiable manifolds. We were then able to define linear approxima-
tions to manifolds (tangent spaces) and and to maps (the differential). The concept of a
differentiable manifold is what one needs to do analysis.
In order to do geometry we need to enrich our manifolds once more. We want to mea-
sure lengths of and angles between tangent vectors. This requires scalar products on
the tangent spaces and leads to the concept of a Riemannian manifold.
2.1 Bilinear forms

We start by recalling some facts about bilinear forms from linear algebra.
Definition 2.1.1. Let V be an n-dimensional R-vector space. A symmetric bilinear

form is a map g : V × V → R with
(i) g(αv + βw, z) = αg(v, z) + βg(w, z) for all v, w, z ∈ V and α, β ∈ R and
(ii) g(v, w) = g(w, v) for all v, w ∈ V .
We call g non-degenerate if g(v, w) = 0 for all w ∈ V implies v = 0.
For a basis (b1 , . . . , bn ) of V we set
gij := g(bi , bj ) ∈ R
for i, j = 1, . . . , n. Then (gijP Pn n j× n-matrix. From (gij )i,j=1,...,n we

)i,j=1,...,n is a symmetric
n i
can reconstruct g: For v = i=1 α bi and w = j=1 β bj we have:
n
X n
X n
X
i j
g(v, w) = g α bi , β bj = αi β j gij .
i=1 j=1 i,j=1
35
Notation 2.1.2. Let b∗1 , . . . , b∗n the dual basis of the dual space V ∗ = {linear maps V →
R} of b1 , . . . , bn , that is b∗i (bj ) = δij . Often, we write
n
X
g= gij b∗i ⊗ b∗j .
i,j=1
The insertion of v, w ∈ V then means the following:

n
X n
X
g(v, w) = gij b∗i (v) · b∗j (w) = gij αi β j .
i,j=1 i,j=1
Transformation of principal axes. Let g be a non-degenerate symmetric bilinear form

on V . Then there exists a basis e1 , . . . , en of V , such that
(
0 i 6= j
g(ei , ej ) = ,
εi ∈ {±1} i=j
in other words,
 
−1 0
..

 . 

(gij )i,j=1,...,n =  −1 . (2.1)
1
 .. 
.
0 1
Such a basis is called a generalized orthonormal basis. We the number of −1’s oc-
curring in (2.1) the index of g and denote it by Index(g). We observe that for a non-
degenerate symmetric bilinear form the following are equivalent:
(1) g is a Euclidean scalar product;
(2) g is positive definite;
(3) Index(g) = 0.
If B = (b1 , . . . , bn ) and B̃ = (b̃1 , . . . , b̃n ) are two bases of V , we define the transformation
matrix T = (tji )i,j=1,...,n by
n
X
b̃i = tji bj .
j=1
36
2.2 Semi-Riemannian metrics
Then the representing matrix of g transforms as follows:

(B̃)
gij = g b̃i , b̃j
X n n
X
k l
= g t i bk , t i bl
k=1 l=1
n
X
= tki tlj · g(bk , bl )
k,l=1
Xn
(B)
= tki tlj · gkl (2.2)
k,l=1
Let V and W be two finite-dimensional R-vector spaces. Let g be a symmetric bilinear

form on V and Φ : W → V be a linear map. Then we can pull back g via Φ to W , that
is, we can define a symmetric bilinear form Φ∗ g on W by

Φ∗ g (w1 , w2 ) := g Φ(w1 ), Φ(w2 ) .
Remark 2.1.3. If g is positive definite, then Φ∗ g is positive semidefinite. Namely:
(Φ∗ g)(w, w) = g(Φ(w), Φ(w)) ≥ 0 ∀ w ∈ W.
If furthermore Φ is injective, then Φ∗ g is also positive definite. Namely:
(Φ∗ g)(w, w) = 0 =⇒ Φ(w) = 0 =⇒ w = 0.
Definition 2.1.4. Let gV and gW be symmetric bilinear forms on V and W , respec-

tively. We call a bijective linear map Φ : W → V an isometry, if

gV Φ(w1 ), Φ(w2 ) = gW (w1 , w2 ), ∀ w1 , w2 ∈ W,
that is, if Φ∗ gV = gW .

Let M be a differentiable manifold. We consider maps g which assign to every point
p ∈ M a non-degenerate symmetric bilinear form g|p on Tp M . If x : U → V is a chart of
37
(x)
M , we define gij = gij : V → R by
!
∂ ∂
gij (v) := g|x−1 (v) , .
∂xi x−1 (v) ∂x j
x−1 (v)
Definition 2.2.1. Such a map g is called a semi-Riemannian metric on M , if the map

depends smoothly on the base point in the following sense:
For every chart x : U → V of M the gij : V → R are C ∞ -functions.
Remark 2.2.2. Note the similarity of the definition of smoothness of g with the charac-
terization of smoothness of vector fields in Remark 1.5.4. We express the vector field
or semi-Riemannian metric with respect to the basis ∂x∂ 1 , . . . , ∂x∂ n of the tangent space
induced by a chart and then require smoothness of the coefficient functions.
Transformation by change of charts

Let x : U → V and y : Ũ → Ṽ be two charts of M with p ∈ U ∩ Ũ . By (1.7),
n
X ∂(xj ◦ y −1 )
∂ ∂
= ·
∂y i p ∂y i y(p) ∂x j
p
| {z } j=1 | {z } | {z }
= b̃i = tji = bj
Inserting this into (2.2) yields

n
(y)
X ∂(xk ◦ y −1 ) ∂(xl ◦ y −1 ) (x)
gij (y(p)) = · · gkl (x(p)).
∂y i y(p) ∂y j y(p)
k,l=1
For all v ∈ y(U ∩ Ũ )) we hence have
n
(y)
X ∂(xk ◦ y −1 ) ∂(xl ◦ y −1 ) (x)
gij (v) = · · gkl (x ◦ y −1 )(v) (2.3)
∂y i v ∂y j v
k,l=1
In the physicist’s short notation this formula reads as
(y) ∂xk ∂xl (x) −1

gij = · gkl ◦ x ◦ y .
∂y i ∂y j
38
Consequence. The condition that g is smooth does not have to be checked for all charts,
if suffices to check it for a subatlas of Amax (M ) which covers M .
Remark 2.2.4. Recall that dx|p : Tp M → Rn is a linear isomorphism for any chart
x : U → V with p ∈ U . In particular, dx1 |p , . . . , dxn |p ∈ (Tp M )∗ .
Definition 2.2.5. The dual space (Tp M )∗ =: Tp∗ M is called cotangent space of M
at p.
Lemma 2.2.6
∂ ∂
The dx1 |p , . . . , dxn |p form the dual basis of ,..., .
∂x1 p ∂xn p

∂
Proof. Since dx|p = ei we have dxj |p ∂
∂xi p
= δij for i = 1, . . . , n.
∂xi p
According to Notation 2.1.2 we may also write:
n
X
g|p = gij x(p) · dxi |p ⊗ dxj |p
i,j=1
In the physics literature you will find the following short version of this equation:
g = gij · dxi · dxj
n
P
If one changes the basis of a vector space by the transformation b̃i = tji bj , then we
j=1
get
n
X
b∗i = tij b̃∗j .
j=1
39
Namely, denote the transformation matrix by T = (tji ). Then we find:

n
X n
X n
X
tij b̃∗j (bk ) = tij b̃∗j (T −1 )lk b̃l
j=1 j=1 l=1
X n
= tij (T −1 )lk b̃∗j (b̃l )
j,l=1
| {z }
=δjl
n
X
= tij (T −1 )jk
j=1
= δki ,
hence
n
X
tij b̃∗j = b∗i .
j=1
For b∗1 = dx1 |p , . . . , b∗n = dxn |p this means:
n
X ∂(xi ◦ y −1 )
dxi |p = · dy j |p
∂y j y(p)
j=1
or, in the physicist’s short notation
∂xi j
dxi = dy
∂y j
If you have forgotten the transformation formula (2.3), you can quickly deduce it in
“physics style” as follows:
(y) (x)
gkl · dy k · dy l = gij · dxi · dxj
i j
(x) ∂x k ∂x l
= gij · · dy · · dy
∂y k ∂y l
∂xi ∂xj (x)
= · ·g · dy k · dy l .
∂y k ∂y l ij
Comparing the coefficients in the blue boxes yields (2.3).
Example 2.2.7. Let M ⊂ Rn be open. Let β be a non-degenerate symmetric bilinear

form on Rn . For every p ∈ M let Φp : Tp M → Rn the canonical isomorphism. Set
40
g|p := Φ∗p β. We check the smoothness of g in Cartesian coordinate, i.e., in the chart
x = id : U = M → V = M .

∂ ∂ ∗
∂ ∂
g|p , = Φp β ,
∂xi p ∂xj p ∂xi p ∂xj p

∂ ∂
= β Φp , Φp
∂xi p ∂xj p

∂ ∂
= β dx|p , dx|p
∂xi p ∂xj p
= β(ei , ej ).
Consequently, the gij are constant, hence C ∞ . In this manner, we can equip M with a
semi-Riemannian metric with arbitrary index.
Example 2.2.8. Let M ⊂ Rn+k be an n-dimensional submanifold. Then there exists a

canonical injective map Φp : Tp M → Rn+k , defined by
d
ċ(0) 7→ c|t=0
dt
equivalence class of the derivative of

curve c : (−ε, ε) → M c : (−ε, ε) → Rn+k
Pn+k i i
Then define g|p := Φ∗p h·, ·i, where hx, yi = i=1 x y is the usual Euclidean scalar
product, x = (x1 , . . . , xn+k )T , y = (y 1 , . . . , y n+k )T . Since the Euclidean scalar product
is positive definite and Φp is injective, we conclude that g|p is also positive definite
for all p ∈ M . The semi-Riemannian metric on M defined in this way is called first
fundamental form.
The charts of submanifolds correspond to local parametrizations of M , i.e. to maps
F : V → M with V ⊂ Rn open, where
x = F −1 : U = F (V ) → V
is a chart of M . In addition, we have with p = x−1 (v):

∂ ∂
gij (v) = g|p ,
∂xi p ∂xj p

∗
∂ ∂
= Φp h·, ·i ,
∂xi p ∂xj p
* +
∂ ∂
= Φp , Φp
∂xi p ∂xj p

d d
= F (v + t · ei )|t=0 , F (v + t · ej )|t=0
dt dt
41

∂F ∂F
= (v), j (v) .
∂xi ∂x
∂F ∂F
Hence gij = ,
∂xi ∂xj
, in particular, the gij are smooth.
Definition 2.2.9. A semi-Riemannian metric g, for which g|p is always positive de-
finite, is called Riemannian metric. A pair (M, g), consisting of a differentiable
manifold M and a (semi-)Riemannian metric g on M is called (semi-)Riemannian
manifold.
A semi-Riemannian metric g is called Lorentz metric, if g|p has always index 1. The
pair (M, g) is then called Lorentz manifold.
Example 2.2.10. The first fundamental form of a submanifold M ⊂ Rn+k is a Rieman-

nian metric. For example, for S n ⊂ Rn+1 we call the first fundamental form the standard
metric gstd of S n .
We express the standard metric of S 2 in the coordinates given by stereographic projec-
tion from the “south pole” (−1, 0, 0). Recall from Example 1.1.4 that the inverse of this
chart map is given by
1
F : R2 → S 2 ⊂ R3 , F (x) = (4 − kxk2 , 4x).
4 + kxk2
One computes
∂F 1
1
= (−16x1 , 4(4 − (x1 )2 + (x2 )2 ), −8x1 x2 ),
∂x (4 + kxk2 )2
∂F 1
2
= (−16x2 , −8x1 x2 , 4(4 + (x1 )2 − (x2 )2 )).
∂x (4 + kxk2 )2
Moreover,
∂F ∂F 16
g11 = , =
∂x1 ∂x1 (4 + kxk2 )2
and similarly for the other gij . The metric in these coordinates turns out to be

16 1 0
(gij ) = .
(4 + kxk2 )2 0 1
42
Example 2.2.11. Let M ⊂ Rn+1 be open. The Minkowski scalar product hh·, ·ii on Rn+1 has
index 1, where
hhx, yii = −x0 y 0 + x1 y 1 + · · · + xn y n
for x = (x0 , x1 , . . . , xn ) and y = (y 0 , y 1 , . . . , y n ). If Φp : Tp M → Rn+1 is the canonical

isomorphism, we can define a Lorentz metric on M by
gMink |p := Φp ∗ hh·, ·ii .
The Lorentz manifold (Rn+1 , gMink ) is called Minkowski space. The four-dimensional
Minkowski space is the mathematical model for spacetime in special relativity.
Example 2.2.12. We express the Euclidean metric geucl = dx1 ⊗ dx1 + dx2 ⊗ dx2 of R2 in
polar coordinates. Here x1 and x2 are the Cartesian coordinates. With x1 = r cos ϕ and
x2 = r sin ϕ we then find:
∂x1 ∂x1
dx1 = dr + dϕ = cos ϕ dr − r sin ϕ dϕ
∂r ∂ϕ
∂x2 ∂x2
dx2 = dr + dϕ = sin ϕ dr + r cos ϕ dϕ.
∂r ∂ϕ
Thus
geucl = (cos ϕ dr − r sin ϕ dϕ) ⊗ (cos ϕ dr − r sin ϕ dϕ)
+(sin ϕ dr + r cos ϕ dϕ) ⊗ (sin ϕ dr + r cos ϕ dϕ)
= cos2 ϕ dr ⊗ dr − r cos ϕ sin ϕ dr ⊗ dϕ − r sin ϕ cos ϕ dϕ ⊗ dr + r 2 sin2 ϕ dϕ ⊗ dϕ
+ sin2 ϕ dr ⊗ dr + sin(ϕ)r cos ϕ dr ⊗ dϕ + r cos ϕ sin ϕ dϕ ⊗ dr + r 2 cos2 ϕ dϕ ⊗ dϕ
= dr ⊗ dr + r 2 dϕ ⊗ dϕ
and hence
Polar 1 0
gij =
0 r2
This matrix tells us: ∂
∂ ∂ϕ
• has length 1,
∂r
b
∂ b
• has length r, ∂
∂ϕ
∂r
∂ ∂
• and are orthogonal to each other
∂r ∂ϕ
In Cartesian coordinates we have: ∂
∂x2
1 0
Cartes
gij = b
0 1 ∂
∂x1
43
Definition 2.2.13. Let (M, gM ) and (N, gN ) be semi-Riemannian manifolds. A local

diffeomorphism ϕ : M → N is called local isometry, if
dϕ|p : (Tp M, gM |p ) → (Tϕ(p) N, gN |ϕ(p) )
for all p ∈ M is a linear isometry.

If a local isometry is also injective, that is, if it is a diffeomorphism, we call it an
isometry.
Definition 2.2.14. If ϕ : M → N is a local diffeomorphism and g a semi-Riemannian

metric on N , then we call the semi-Riemannian metric ϕ∗ g on M given by
(ϕ∗ g)|p := (dϕ|p )∗ (g|ϕ(p) ),
the pullback of g. In other words, we have for ξ, η ∈ Tp M :

(ϕ∗ g)|p (ξ, η) = g|ϕ(p) dϕ|p (ξ), dϕ|p (η) .
Remark 2.2.15. The metric ϕ∗ g is the unique semi-Riemannian metric on M , for which
ϕ is a local isometry.
Definition 2.2.16. Let (M, g) be a semi-Riemannian manifold. Then we call
Isom(M, g) := {ϕ : M → M isometry}
the isometry group of M .
Remark 2.2.17. The set Isom(M, g) is a group with respect to composition of maps. The
neutral element is idM .
44
Example 2.2.18. We look for the isometries of (Rn , geucl ). Let

ϕ : Rn → Rn , ϕ(x) = Ax + b,
be an affine map with A ∈ O(n) and b ∈ Rn . Such a ϕ is called a Euclidean motion. We
check that every Euclidean motion is an isometry of (Rn , geucl ): Let Φp : Tp M → Rn
be the canonical isomorphism; for ξ = Φ−1 p (X) ∈ Tp M this means that ξ = ċ(0) where
c(t) = p + tX. Similarly, η = Φ−1
p (Y ) = ˙
c̃(0) ∈ Tp M with c̃(t) = p + tY . We compute:
ϕ∗ (geucl |p )(ξ, η) = geucl |p (dϕ|p (ξ), dϕ|p (η))

= hΦp (dϕ|p (ξ)), Φp (dϕ|p (η))i
= hΦp ((ϕ ◦ c)· (0)), Φp ((ϕ ◦ c̃)· (0))i
= hΦp ((A(p + tX) + b)· (0)), Φp ((A(p + tY ) + b)· (0))i
= hΦp (Ap + b + tAX)· (0)), Φp (Ap + b + tAY )· (0))i
= hAX, AY i
= hX, Y i
= hΦp (ξ), Φp (η)i
= geucl (ξ, η).
Hence ϕ∗ (geucl |p ) = geucl showing that ϕ is a local isometry. Since ϕ is bijective, it is an
isometry. Summarizing, we have shown
{Euclidean motions} ⊂ Isom(Rn , geucl ).
We will see later that the inverse conclusion also holds; the isometries of (Rn , geucl ) are
precisely the Euclidean motions.
Example 2.2.19. To find isometries of Minkowski space (M, g) = (Rn+1 , gMink ) we de-
fine

O(n, 1) := A ∈ Mat((n + 1) × (n + 1), R) | hhAy, Azii = hhy, zii ∀ y, z ∈ Rn+1

= A ∈ Mat((n + 1) × (n + 1), R) | A⊤ I1,n , A = I1,n
where  
−1 0 ··· 0
 .. .. 
 0 1 . .
I1,n = ..
.
 .. .. 
 . . . 0
0 ··· 0 1
Now affine transformations ϕ : →Rn+1 Rn+1 ,
ϕ(x) = Ax + b with A ∈ O(n, 1) and
b ∈ R n+1 , are called Poincaré transformations. The same discussion as for Euclidean
space shows
{Poincaré transformations} ⊂ Isom(Rn+1 , gMink ).
45
Again, we will see later that equality holds; the isometries of Minkowski space are
precisely the Poincaré transformations.
Example 2.2.20. To find isometries of the sphere (M, g) = (S n , gstd ) let A ∈ O(n + 1).
We set ϕ := A|S n : S n → S n . Let Φp : Tp S n → Rn+1 be as in Example 2.2.8. Then the
diagram
dϕ|p
Tp S n Tϕ(p) S n
Φp Φϕ(p)
A
Rn+1 Rn+1
commutes because:
ċ(0) (ϕ ◦ c)· (0) = (A ◦ c)· (0)
d d d
c|t=0 A· c|t=0 = (A · c)|t=0
dt dt dt
Therefore
gstd (dϕ|p (ξ), dϕ|p (η)) = hΦϕ(p) (dϕ|p (ξ)), Φϕ(p) (dϕ|p (η))i
= hAΦp (ξ), AΦp (η)i
= hΦp (ξ), Φp (η)i
= gstd (ξ, η).
This shows that ϕ is an isometry. Hence
O(n + 1) ⊂ Isom(S n , gstd ).
Again, it will turn out that equality holds.
2.3 Differentiation of vector fields

We know how to differentiate functions on a manifold. We also know what differenti-
able vector fields are. But: How do we differentiate a vector field? What is the differen-
tial of a vector field at a point in the manifold?
First attempt. Let M be a differentiable manifold and let p ∈ M . Let ξ ∈ Tp M and let η
be a differentiable vector field on M . We try to define the derivative of η in the direction
ξ.
To this
P extent,∂ we choose → V on M with p ∈ U . We write ξ ∈ Tp M as
a chart x : U P
ξ = ni=1 ξ i ∂x i p with ξ i ∈ R and η = n j ∂ j
i=j η ∂xj where the η are smooth functions
near x(p).
46
The first idea that comes to one’s mind is to differentiate the coefficient functions η j in
the direction ξ. This would yield the expression
n
X ∂η j ∂
ξi · ·
∂xi x(p) ∂xj p
i,j=1
for the derivative of η in direction ξ.

Problem. This “definition” depends on the choice of chart x.
Example 2.3.1. Let M = R2 . In polar coordinates (r, ϕ) we set
∂
ξ=η= .
∂ϕ
Then the derivative of η in direction ξ equals 0 because the coefficient functions η j are
constant. On the other hand, in Cartesian coordinates (x1 , x2 ) we get
∂ ∂
ξ = η = −x2 1
+ x1 2 .
∂x ∂x
For the derivative of η in direction ξ we would then find

∂ ∂ ∂ ∂ ∂ ∂
−x2 1 + x1 2 (−x2 ) 1 + −x2 1 + x1 2 (x1 ) 2
∂x ∂x ∂x ∂x ∂x ∂x
∂ ∂ ∂
= −x1 1
− x2 2 = −r 6= 0.
∂x ∂x ∂r
We see that the idea of simply differentiating the coefficient functions was to naive.
Since we do not know how to come up with a better definition we follow an axiomatic
approach similar to the concept of derivations, except that this time we differentiate
vector fields rather than functions.
Notation 2.3.2. Let M be a differentiable manifold and let k ∈ N∪{∞}. For any open
subset U ⊂ M we put
C k (U, T M ) := {C k -vector fields, defined on U }.
For p ∈ M we set [
Ξp := C ∞ (U, T M ).
U ⊂M open
with p∈U
47
Now we list the properties that the derivative of vector fields should have. Differenti-
ation takes a tangent vector ξ ∈ Tp M and a smooth vector field η defined near p and
gives us a tangent vector in Tp M as a result. Hence it is a map Tp M × Ξp → Tp M .
Definition 2.3.3. Let (M, g) be a semi-Riemannian manifold and p ∈ M . A map
∇ : Tp M × Ξp → Tp M
is called Levi-Civita connection (at p), if the following holds:

(i) Locality
For all ξ ∈ Tp M , for all η ∈ C ∞ (U, T M ) and for all Ũ ⊂ U with p ∈ Ũ we have:
∇ξ η = ∇ξ (η|Ũ ).
(ii) Linearity in the first argument

For all ξ1 , ξ2 ∈ Tp M , for all α, β ∈ R and for all η ∈ Ξp we have:
∇αξ1 +βξ2 η = α∇ξ1 η + β∇ξ2 η.
(iii) Additivity in the second argument

For all ξ ∈ Tp M and for all η1 , η2 ∈ Ξp we have:
∇ξ (η1 + η2 ) = ∇ξ η1 + ∇ξ η2 .
(iv) Product rule I

For all f ∈ Cp∞ , for all η ∈ Ξp and for all ξ ∈ Tp M we have:
∇ξ (f · η) = ∂ξ f · η|p + f (p) · ∇ξ η.
(v) Product rule II

For all ξ ∈ Tp M and for all η1 , η2 ∈ Ξp we have:
∂ξ g(η1 , η2 ) = g|p (∇ξ η1 , η2 |p ) + g|p (η1 |p , ∇ξ η2 ).
(vi) Torsion-freeness
For all charts x : U → V of M with p ∈ U we have:
∂ ∂
∇ ∂ j
=∇ ∂ i
∂xi p ∂x ∂xj p ∂x
for all i and j.
48
Remark 2.3.4
(1) From (iii) and (iv) we get the R-linearity in the second argument. Let α, β ∈ R:
(iii)
∇ξ (α η1 + β η2 ) = ∇ξ (α η1 ) + ∇ξ (β η2 )
(iv)
= ∂ξ α ·η1 |p + α∇ξ (η1 ) + ∂ξ β ·η2 |p + β∇ξ (η2 )
|{z} |{z}
=0 =0
= α∇ξ (η1 ) + β∇ξ (η2 ).
(2) If (vi) holds in a chart x, then it also holds in every other chart y containing p.
n
!
∂ X ∂xk ∂
∇ ∂ j
= ∇ ∂
∂y i p ∂y ∂y i p ∂y j p ∂xk
k=1
(iii) n
!
(iv) X ∂ 2 xk ∂ ∂xk ∂
= · + ∇
∂y i ∂y j y(p) ∂xk ∂y j y(p) ∂y∂ i p ∂xk
k=1
n n
(ii) X ∂ 2 xk ∂ X ∂xk ∂xl ∂
= · k+ · ∇ ∂
∂y i ∂y j y(p) ∂x ∂y j y(p) ∂y i y(p) ∂xl p ∂xk
k=1 k,l=1
The first summand is symmetric in i and j due to Schwarz’ Theorem. Concerning

the second summand we have:
n n
X ∂xk ∂xl ∂ (vi) X ∂xk ∂xl ∂
j
· i
∇ ∂ k
= · ∇ ∂
∂y y(p) ∂y y(p) ∂xl p ∂x ∂y y(p) ∂y y(p) ∂xk p ∂xl
j i
k,l=1 k,l=1
n
change of
X ∂xl ∂xk ∂
= · ∇ ∂
indices ∂y j y(p) ∂y i y(p) ∂xl p ∂xk
l,k=1
Hence the second summand is also symmetric in i and j.

(3) In general, for non-coordinate fields ξ and η we have
∇ξ η 6= ∇η ξ.
∂ ∂
As an example we can choose ξ = ∂x1 and η = f · ∂x1 with ∂ξ f 6= 0.
Definition 2.3.5. Let x : U → V be a chart. Write

n
∂ X ∂
∇ ∂ j
= Γkij x(p) · (2.4)
∂xi p
∂x ∂xk p
k=1
The Γkij are called Christoffel symbols.
49
Pn i ∂
Remark 2.3.6. The Christoffel symbols determine ∇. Namely, let ξ = i=1 ξ ∂xi p
∈
P
Tp M and η = nj=1 η j ∂x∂ j ∈ Ξp . Then we compute:
 
n (ii) n
X
j ∂  (iii) X i j ∂
∇Pn ξ i ∂  η = ξ∇ ∂ η
i=1 ∂xi p ∂xj ∂xi p ∂xj
j=1 i,j=1
n
!
(iv) X ∂η j ∂ ∂
= ξi · + η j |x(p) · ∇ ∂
∂xi x(p) ∂x
j
p ∂xi p
∂xj
i,j=1
n n
!
X
i ∂η j ∂ j
X ∂
= ξ · + η |x(p) · Γkij (x(p)) ·
∂xi x(p) ∂x j
p ∂xk p
i,j=1 k=1
 
n n
X ∂η k X ∂
= ξi  + η j |x(p) · Γkij (x(p)) (2.5)
∂xi x(p) ∂xk p
i,k=1 j=1
Remark 2.3.7. Torsion freeness is equivalent to the Christoffel symbols being symmet-
ric in the two lower indices:
Γkij = Γkji for all i, j, k.
Theorem 2.3.8
Let (M, g) be a semi-Riemannian manifold and let p ∈ M . Then there is exactly one Levi-
Civita connection at p.
Proof. Uniqueness: Let x : U → V be a chart of M with p ∈ U . We compute, using the

Einstein summation convention:

∂gij ∂ ∂ ∂
= g ,
∂xk ∂xk ∂xi ∂xj

(v) ∂ ∂ ∂ ∂
= g ∇ ∂ ,
i ∂xj
+g ,∇ ∂
∂xk ∂x ∂xi ∂xk ∂x
j

∂ ∂ ∂ ∂
= g Γlki l , j + g i
, Γlkj j
∂x ∂x ∂x ∂x

l ∂ ∂ l ∂ ∂
= Γki · g , + Γkj · g ,
∂xl ∂xj ∂xi ∂xj
= Γlki · glj + Γlkj · gil .
50
Renaming the indices we get the equations:
∂gij
= Γlki · glj + Γlkj · gil (2.6)
∂xk
i→i ∂gik
j →k = Γlji · glk + Γljk · gil (2.7)
k→j ∂xj
i →k ∂gkj
j →j = Γlik · glj + Γlij · gkl (2.8)
k→ i ∂xi
Equation (2.6) − (2.7) + (2.8) together with the symmetry of the Christoffel symbols in
the lower indices yields:
∂gij ∂gik ∂gkj

− + = 2 Γlki · glj .
∂xk ∂xj ∂xi
Let (gij )i,j=1,...,n be the inverse matrix of (gij )i,j=1,...,n . This matrix exists because g|p is
non-degenerate. In other words, we have:
g ij · gjk = δki .
Therefore

∂gij ∂gik ∂gkj
k
− j
+ gjm = 2 Γlki · glj · gjm = 2 Γlki · δlm = 2 Γm
ki
∂x ∂x ∂xi
and hence
1 ∂gij ∂gik ∂gkj
Γm
ki = k
− j
+ gjm .
2 ∂x ∂x ∂xi
Renaming indices (k → j, m → k, j → m) we obtain:
n
1 X mk ∂gim ∂gjm ∂gij
Γkij = g + − m (2.9)
2 ∂xj ∂xi ∂x
m=1
Consequently, the Christoffel symbols are uniquely determined and hence ∇ is

uniquely determined by the components of the semi-Riemannian metric and its first
derivatives.
Existence: Define Γkij by equation (2.9) and ∇ by equation (2.5). Then conditions (i), (ii),
(iii), and (vi) of the Levi-Civita connection are obvious. For the first product (iv) rule
we have:
k

i ∂(f · η ) j k ∂
∇ξ (f η) = ξ i
+ f η Γij
∂x ∂xk
k
∂η ∂ ∂f ∂
= f · ξi + η j Γkij + ξ i i ηk k
∂xi ∂xk ∂x ∂x
= f · ∇ξ η + ∂ξ f · η
51
We check the second product rule (v), using the Einstein summation convention and
occasional renaming of indices:
∂ζ g(ξ, η) − g(∇ζ ξ, η) − g(ξ, ∇ζ η)

i j
k ∂ ∂ξ ∂η
i j
k l i j i k l j
=ζ gij ξ η − gij ζ + ξ Γlk η − gij ξ ζ + η Γlk
∂xk ∂xk ∂xk
gij
= ζ k k ξ i η j − gij ζ k ξ l Γilk η j − gij ξ i ζ k η l Γjlk
∂x
i j k gij l l
=ξη ζ − g lj Γ ik − g il Γ jk
∂xk

(2.9) i j k gij 1 ml ∂gim ∂gkm ∂gik
= ξη ζ − glj g + − m
∂xk 2 | {z } ∂xk ∂xi ∂x
=δjm
!
1 ∂gjm ∂gkm ∂gjk
− gil gml + − m
2 | {z } ∂xk ∂xj ∂x
=δim
!
gij 1 ∂gij ∂gkj ∂gik 1 ∂gji ∂gki ∂gjk
= ξ i ηj ζ k k
− k
+ i
− − + −
∂x 2 ∂x ∂x ∂xj 2 ∂xk ∂xj ∂xi
= 0.
Remark 2.3.9. For any chart x : U → V on (M, g) the Christoffel symbols are smooth
functions
n
k 1 X mk ∂gim ∂gjm ∂gij
Γij = g · + − m : V → R.
2 ∂xj ∂xi ∂x
m=1
Remark 2.3.10. Our naive ansatz to differentiate vector fields by simply differentiat-
ing the coefficient functions corresponds to formula (2.5) with Γkij = 0. The problem
was that this depends on the choice of coordinates. When we use formula (2.5) with
the correct definition (2.9) for the Christoffel symbols, then we get the uniquely deter-
mined Levi-Civita connection. In particular, this kind of differentiating vector fields is
independent of the choice of chart.
Note however, that the Levi-Civita connection depends on the semi-Riemannian met-
ric. This cannot only seen from (2.9) but also from the second product rule (v) in Defi-
nition 2.3.3 which involves the metric. There is nothing we can do about this; different
semi-Riemannian metrics will in general lead to different Levi-Civita connections.
So the situation is somewhat curious: Differentiability and the derivative of a function
are well defined on a differentiable manifold. Differentiability of a vector field is also
well defined on a differentiable manifold. But in order to define the derivative of a
vector field we need a semi-Riemannian metric.
52
Definition 2.3.11. Let (M, g) be a semi-Riemannian manifold and let ∇ be its Levi-
Civita connection. Let p ∈ M , let ξ ∈ Tp M and let η ∈ Ξp . Then
∇ξ η ∈ Tp M
is also called the covariant derivative of η in direction ξ.
Example 2.3.12. Let (M, g) = (R2 , geucl ) be the 2-dimensional Euclidean space. In
Cartesian coordinates x1 , x2 the gij = δij are constant. Therefore Γkij = 0. In this case,
covariant differentiation is indeed given by differentiation of the coordinate functions.
For example,

∂ 2 ∂ 1 ∂
∇∂ = ∇−x2 ∂ +x1 ∂ −x +x
∂ϕ ∂ϕ ∂x1 ∂x2 ∂x1 ∂x2

2 ∂ 1 ∂ 2 ∂ 2 ∂ 1 ∂ ∂
= −x 1
+x 2
(−x ) 1 + −x 1
+x 2
(x1 ) 2
∂x ∂x ∂x ∂x ∂x ∂x
∂ ∂ ∂
= −x1 1 − x2 2 = −r
∂x ∂x ∂r
In polar coordinates r, ϕ we have

1 0 ij 1 0
(gij )(r, ϕ) = and (g )(r, ϕ) = 1 .
0 r2 0 r2
The Christoffel symbols with respect to polar coordinates are given by
1
Γ111 = 1 · (0 + 0 − 0) + 0 · . . . = 0.
2
and similarly
Γ111 = Γ211 = Γ112 = Γ121 = Γ222 = 0.
Moreover:

1 1 ∂g12 ∂g22 ∂g12 1
Γ212 = Γ221 = + − + 0 · ... = and Γ122 = −r.
2 r2 ∂ϕ ∂r ∂ϕ r
Thus
∂ ∂ ∂ ∂
∇ ∂ = Γ122 + Γ222 = −r .
∂ϕ ∂ϕ ∂r ∂ϕ ∂r
Indeed, we obtained the same result for both computations, one in Cartesian and one
in polar coordinates.
53
Remark 2.3.13. We defined ∇ pointwise, i.e., as a map Tp M × Ξp → Tp M . We may also

consider ∇ as a map
∇ : Ξ(M ) × Ξ(M ) → Ξ(M ),
where Ξ(M ) denotes the set of all smooth vector fields defined on all of M . Namely, we
put
(∇ξ η)(p) := ∇ξ(p) η.
We know
∇ξ (α1 η1 + α2 η2 ) = α1 ∇ξ η1 + α2 ∇ξ η2
for α1 , α2 ∈ R and
∇f1 ξ1 +f2 ξ2 η = f1 ∇ξ1 η + f2 ∇ξ2 η
for f1 , f2 ∈ C ∞ (M ). This means that ∇ξ η is C ∞ (M )-linear in ξ but only R-linear in η.
Remark 2.3.14. To compute ∇ξ η with ξ = ċ(0) we only need to know η along the curve
c. Namely,
   
n n
X ∂  ∂ X
∇ċ(0)  ηj = ∇Pn ċi (0) ∂  ηj j 
∂xj i=1 ∂xi ∂x
j=1 j=1
n
j ∂
X
i
= ċ (0) ∇ ∂ η
∂xi ∂xj
i,j=1
n n
!
X ∂η j ∂ X ∂
= ċi (0) i j
+ η j Γkij
∂x ∂x ∂xk
i,j=1 k=1
n n
X d j ∂ X ∂
= η ◦ c |t=0 j + ċi (0) η j (c(0))Γkij (c(0)) k .
dt ∂x ∂x
j=1 i,j,k=1
2.4 Vector fields along maps
Definition 2.4.1. Let M and N be differentiable manifolds and ϕ : N → M a map.

Then a map ξ : N → T M is called a vector field along ϕ, if
πM ◦ ξ = ϕ.
holds. Here πM : T M → M is the “footpoint map”.
54
Example 2.4.2
(1) Vector fields along curves. Let N = I ⊂ R be an open interval and c = ϕ : N = I → M
be a curve.
c
ξ
N
M
An important special case is given by ξ(t) = ċ(t) := ċt (0) where ct (s) := c(t + s).
This is the velocity field of c.
c
ξ
N
M
(2) If N = M and ϕ = id then a vector field along id is just a vector field in the usual
sense.
(3) Let ϕ be constant, i.e., ϕ(x) = p for all x ∈ N . Then a vector field along ϕ is a map
N → Tp M .
(4) Let ϕ be differentiable and let ξ be a vector field on N . Then

p 7→ dϕ|p ξ(p) ∈ Tϕ(p) M
is a vector field along ϕ.
(5) If ξ is a vector field on M then
p 7→ ξ ϕ(p)
is a vector field along ϕ.
Definition 2.4.3. Let N be a differentiable manifold and (M, g) a semi-Riemannian

manifold. Let ϕ : N → M be a differentiable map and η : N → T M a differentiable
vector field along ϕ. For p ∈ N and ξ ∈ Tp N we define the covariant derivative
∇ξ η ∈ Tϕ(p) M as follows:
Choose a chart x : U → V of M with ϕ(p) ∈ U and write
n
X ∂
η(q) = η j (q) ·
∂xj ϕ(q)
j=1
55
with differentiable functions η 1 , . . . , η n defined on ϕ−1 (U ). In addition, choose a

curve c : (−ε, ε) → N with ċ(0) = ξ and set
 
n n
X d X d ∂
∇ξ η :=  η k ◦ c t=0 + η j (p) ϕi ◦ c t=0 Γkij x ϕ(p) 
dt dt ∂xk ϕ(p)
k=1 i,j=1
 
n n
X
∂ξ η k +
X ∂
= η j (p) dϕ(ξ)i Γkij x ϕ(p)  .
∂xk ϕ(p)
k=1 i,j=1
Proposition 2.4.4
Let N be a differentiable manifold, (M, g) a semi-Riemannian manifold and ϕ : N → M a
differentiable map. Let η, η1 , η2 be differentiable vector fields along ϕ. Let α1 , α2 ∈ R and
f : N → R be a differentiable function. Furthermore, let p ∈ N and ξ, ξ1 , ξ2 ∈ Tp N .
Then the covariant derivative ∇ξ η is defined independently of the choice of chart x and the
choice of curve c with ċ(0) = ξ and we have:
(i) If η is the form η = ζ ◦ ϕ where ζ is a differentiable vector field on M , then we have
∇ξ η = ∇dϕ|p (ξ) ζ.
(ii) Locality: If η1 and η2 coincide on a neighborhood of p, then ∇ξ η1 = ∇ξ η2 .

(iii) Linearity in the first argument:
∇α1 ξ1 +α2 ξ2 η = α1 ∇ξ1 η + α2 ∇ξ2 η.
(iv) Linearity in the second argument:

∇ξ (α1 η1 + α2 η2 ) = α1 ∇ξ η1 + α2 ∇ξ η2 .
(v) Product rule I:

∇ξ (f · η) = ∂ξ f · η(p) + f (p)∇ξ η.
(vi) Product rule II:

∂ξ g(η1 , η2 ) = g|ϕ(p) ∇ξ η1 , η2 (p) + g|ϕ(p) η1 (p), ∇ξ η2 .
(vii) Torsion freeness: For all charts y of N and all i, j = 1, . . . , dim(N ) we have:

∂ ∂
∇ ∂ dϕ = ∇ ∂ dϕ .
∂y i ∂y j ∂y j ∂y i
56
Proof. The assertions follow directly from the definition and the corresponding state-
ments for the Levi-Civita connection.
Notation 2.4.5. For local coordinates y on N we write

 
n k n i
∇η
 ∂η ∂ϕ ∂
X X
(p) := ∇ ∂ η = l
+ l
(p) · η j y(p) · Γkij x ϕ(p)  k (p).
∂y l
∂y l p ∂y y(p) ∂y ∂x
k=1 i,j
If N is one-dimensional, we also write

∇η ∇η
=: .
∂t dt
Remark 2.4.6. For a vector field along a curve c : I → M we have the following formula
in local coordinates on M :
 
n n
∇η X
η̇ k (t) +
X ∂
(t) = ċi (t) · η j (t) · Γkij x c(t)  .
dt ∂xk c(t)
k=1 i,j
In particular, for the velocity field we get

 
n n
∇ċ X
c̈k (t) +
X ∂
(t) = ċi (t) · ċj (t) · Γkij x c(t)  .
dt ∂xk c(t)
k=1 i,j
Example 2.4.7. Let (M, g) = (Rn , geucl ) or (M, g) = (Rn , gMink ). Then the gij are con-
stant in Cartesian coordinates. Consequently, the Christoffel symbols with respect to
Cartesian coordinates vanish, Γkij = 0.
P
For a C 1 -curve c : I → M and a C 1 -vector field ξ along c with ξ(t) = nj=1 ξ j (t) ∂x∂ j |c(t)
we have:
n
∇ X ∂
ξ(t) = ξ̇ j (t) .
dt ∂xj c(t)
j=1
Hence, in this case, covariant differentiation just consists of differentiation of the coef-
ficient functions. Note however, that this is no longer true in other coordinate systems
such as polar coordinates.
57
Example 2.4.8. In the Euclidean plane (M, g) = (R2 , geucl ) we consider the circle line
c(t) = (cos(t), sin(t)) and its velocity field
∂ ∂
ξ(t) = ċ(t) = − sin(t) + cos(t) .
∂x1 c(t) ∂x2 c(t)
In Cartesian coordinates we get by the previous example
∇ ∇ ∂ ∂ ∂
ξ(t) = ċ(t) = − cos(t) − sin(t) =− .
dt dt ∂x1 c(t) ∂x2 c(t) ∂r c(t)
For the fun of it, let us also carry out the calculation in polar coordinates (r, ϕ). Now
∂
c1 (t) = r(t) = 1, c2 (t) = ϕ(t) = t and ξ(t) = ∂ϕ c(t) , i.e., ξ 1 (t) = 0 and ξ 2 (t) = 1. This
time there are no derivatives of the coefficients of ξ but we have to take the Christoffel
symbols into account. Recall from Example 2.3.12 that there are three non-vanishing
Christoffel symbols for polar coordinates,
1
Γ212 = Γ221 = , Γ122 = −r.
r
Therefore we get
2 2
∇ X ∂ X ∂
ξ(t) = i
ċ (t) ξ j
(t) Γ1ij
r(t), ϕ(t) + ċi (t) ξ j (t) Γ2ij r(t), ϕ(t)
dt ∂r c(t) ∂ϕ c(t)
ij=1 ij=1

∂ 1 1 ∂
= ċ2 (t) ξ 2 (t) − r(t) + ċ1 (t) ξ 2 (t) + ċ2 (t) ξ 1 (t)
∂r c(t) r(t) r(t) ∂ϕ c(t)
∂ ∂
= 1 · 1 · (−1) + (0 · 1 · 1 + 1 · 0 · 1)
∂r c(t) ∂ϕ c(t)
∂
= − .
∂r c(t)
So indeed, we have obtained the same result.
2.5 Parallel transport
Definition 2.5.1. Let (M, g) be a semi-Riemannian manifold and c : I → M be a

C 1 -curve. A C 1 -vector field ξ along c is called parallel, if
∇
ξ ≡ 0.
dt
58
Pn Let j(M, g)
Example 2.5.2. = (Rn , geucl ) or (Rn , gMink ). In Cartesian coordinates, a vector
∂
field ξ(t) = j=1 ξ (t) ∂xj c(t) along a curve c is parallel if and only if ξ̇ j (t) = 0 for all
t ∈ I, i.e., if and only if the ξ j are constant.
Example 2.5.3. Let (M, g) = (R2 , geucl ). Recall from Example 2.3.12 that the Christoffel
symbols in polar coordinates (r, ϕ) are given by:
1
Γ111 = Γ211 = Γ112 = Γ121 = Γ222 = 0, Γ212 = Γ221 = , Γ122 = −r.
r
∂ ∂
Thus ξ = ξ 1 ∂r + ξ 2 ∂ϕ is parallel along a curve c if and only if
∇
0 = ξ
dt
∂ ∂ ∂ ∂
= ξ̇ 1 + ξ 1 ∇ċ1 ∂ +ċ2 ∂ + ξ˙2 + ξ 2 ∇ċ1 ∂ +ċ2 ∂
∂r ∂r ∂ϕ ∂r ∂ϕ ∂r ∂ϕ ∂ϕ

1 ∂ 1 1 2 1 ∂ ˙2 ∂ 2 1 1 ∂ 2 1 ∂
= ξ̇ + ξ ċ · 0 + ċ · 1 +ξ + ξ ċ 1 + ċ (−c )
∂r c ∂ϕ ∂ϕ c ∂ϕ ∂r
∂ 2 1

ċ ċ ∂
= ξ̇ 1 − c1 ċ2 ξ 2 + ξ̇ 2 + 1 ξ 1 + 1 ξ 2 .
∂r c c ∂ϕ
This is equivalent to:
ċ2 ċ1
ξ˙1 − c1 ċ2 ξ 2 = 0, ξ˙2 + 1 ξ 1 + 1 ξ 2 = 0,
c c
that is 1 1
ξ˙ 0 c1 ċ2 ξ
= 2 1 .
ξ˙2 − ċ1 c
− cċ1 ξ2
This is a system of linear first order ordinary differential equations for (ξ 1 , ξ 2 ).
M
Proposition 2.5.4
ξ0
Let (M, g) be a semi-Riemannian manifold
and c : I → M be a C 1 -curve and t0 ∈ I.
For any ξ0 ∈ Tc(t0 ) M there exists exactly one ξ
b
c
c(t0 )
parallel vector field ξ along c with ξ(t0 ) = ξ0 .
59
Proof. Case 1: Let c(I) be contained in one chart and let x : U → V be such a chart.
∇
Then the condition dt ξ = 0 is equivalent to
n
X
ξ˙k = − Γkij ◦ x ◦ c ċi · ξ j ,
i,j=1
which is a system of linear ordinary equations of first order. Hence there exists a unique
solution with initial condition

ξ 1 (t0 ), . . . , ξ n (t0 ) = (ξ01 , . . . , ξ0n ).
Since the system is linear, the solution is defined on all of I.
Case 2: Suppose c(I) is not contained in one chart.

Existence: The interval I can be open, closed or half-open. We restrict ourselves to open
intervals, the other cases being slightly simpler. Write I = (a, b) where −∞ ≤ a < b ≤
∞. Choose a < ai < t0 < bi < b with ai → a and bi → b monotonically. Then c([ai , bi ]) is
compact and can be covered by finitely many charts x1 : U1 → V1 , . . . , xN : UN → VN .
W.l.o.g. we assume that Ui ∩ c([a1 , b1 ]) is connected.
ξ0
Ui bc
bc
bc
c(t1 ) c(b1 )
c(t0 )
c
bc
c(a1 )
Not something like this!
∇
W.l.o.g. let c(t0 ) ∈ U1 , otherwise renumber the charts. We solve the equation ξ=0
dt
as in Case 1 with ξ(t0 ) = ξ0 in U1 .
If the solution is not defined on the whole of [a1 , b1 ], we choose t1 ∈ (a1 , b1 ) with
c(t1 ) ∈ U1 ∩ U2 . Then we solve the equation in the chart x2 with the initial condition
ξ(t1 ), given by the previous solution.
Due to uniqueness in Case 1 both parallel vector fields coincide on U1 ∩U2 . After finitely
many steps we get a parallel vector field which is defined on [a1 , b1 ].
The same holds true for the next compact subinterval [a2 , b2 ] and we obtain a parallel
vector field on [a2 , b2 ] which extends the one on [a1 , b1 ]. By induction, we then find a
parallel vector field on every [ai , bi ] extending the one on the smaller interval[ai−1 , bi−1 ].
S
Since N i=1 [ai , bi ] = (a, b) we obtain a parallel vector field ξ on (a, b) with ξ(t0 ) = ξ0 .
˜ 0 ) = ξ0 .
Uniqueness: Let ξ and ξ̃ be two parallel vector fields along c with ξ(t0 ) = ξ(t
Write I = Igood ⊔ Ibad where

˜
Igood = t ∈ I | ξ(t) = ξ(t)

˜
Ibad = t ∈ I | ξ(t) 6= ξ(t)
60
Since ξ and ξ˜ are continuous, Igood is closed in I. For t1 ∈ Igood choose a chart x : U → V
which contains c(t1 ). By uniqueness in Case 1 we then have ξ(t) = ξ̃(t) for all t ∈ I with
c(t) ∈ U . Therefore a neighborhood of t1 is contained in Igood . Hence Igood is open in I.
We have seen that Igood is open and closed in I. It is also non-empty because t0 ∈ Igood .
Since I is connected, we have I = Igood and therefore ξ(t) = ξ(t) ˜ for all t ∈ I.
Definition 2.5.5. Let M be a semi-Riemannian manifold and let c : I → M be a

C 1 -curve. Let t0 , t1 ∈ I. The map
Pc,t0 ,t1 : Tc(t0 ) M → Tc(t1 ) M,

ξ0 7→ ξ(t1 ),
is called parallel transport along c. Here ξ(t) is the parallel vector field along c with
ξ(t0 ) = ξ0 .
Proposition 2.5.6
Let M , c, t0 , and t1 as in Definition 2.5.5 and let t2 ∈ I. Then we have:
(a) Pc,t0 ,t1 : (Tc(t0 ) M, g|c(t0 ) ) → (Tc(t1 ) M, g|c(t1 ) ) is a linear isometry;
(b) Pc,t0 ,t2 = Pc,t1 ,t2 ◦ Pc,t0 ,t1 .
Proof. (a) Let ξ0 , η0 ∈ Tc(t0 ) M . Let ξ, η the corresponding parallel vector fields along c.
Then
d ∇ ∇
g(ξ, η) = g ξ , η + g ξ, η = 0.
dt dt
|{z} dt
|{z}
=0 =0
Therefore g(ξ, η) is constant, hence

g Pc,t0 ,t1 (ξ0 ), Pc,t0 ,t1 (η0 ) = g ξ(t1 ), η(t1 )

= g ξ(t0 ), η(t0 )
= g(ξ0 , η0 ).
This proves that parallel transport is a linear isometry.
61
(b) is obvious.
Remark 2.5.7. For ξ0 ∈ Tc(t0 ) M the parallel vector field ξ with ξ(t0 ) = ξ0 is given by
ξ(t) = Pc,t0 ,t1 (ξ0 ).
We can reconstruct the Levi-Civita connection ∇ from parallel transport:
Proposition 2.5.8
Let (M, g) be a semi-Riemannian manifold, let c : I → M be a C 1 -curve, and let t0 ∈ I.
Then for every C 1 -vector field ξ along c we get:

∇ Pc,t,t0 ξ(t) − ξ(t0 )
ξ = lim .
dt t0 t→t0 t − t0
Proof. Let e1 (t0 ), . . . , en (t0 ) be a basis of Tc(t0 ) M . Let e1 (t), . . . , en (t) be the correspond-
ing parallel vector fields along c.
By Proposition 2.5.6 P(a), we know that e1 (t), . . . , en (t) form a basis of Tc(t) M for every
t ∈ I. Write ξ(t) = nj=1 ξ j (t)ej (t). Then
=ej (t0 )
Pn z }| { P
Pc,t,t0 (ξ(t)) − ξ(t0 ) j=1 ξ j (t) P
c,t,t0 (ej (t)) − nj=1 ξ j (t0 )ej (t0 )
=
t − t0 t − t0
n
X ξ j (t) − ξ j (t0 )
= ej (t0 )
t − t0
j=1
n
X
t→t0
−→ ξ̇ j (t0 )ej (t0 ).
j=1
On the other hand, we have

n
X
∇ ∇ j
ξ|t0 = ξ ej
dt dt t=0
j=1
n
X
˙j ∇j
= ξ (t0 )ej (t0 ) + ξ (t0 ) ej |t0
j=1 |dt {z }
=0
62
2.6 Geodesics
n
X
= ξ˙j (t0 )ej (t0 ).
j=1
We have the following scheme of geometric structures:
semi-Riemannian covariant parallel

metric derivative ∇ transport P
Remark 2.5.9. If ψ : M → M̃ is a local isometry and if c : I → M is a C 1 -curve, consider

the image curve c̃ := ψ ◦ c. Then we have for every C 1 -vector field ξ along c:
ξ parallel along c ⇐⇒ ξ̃ := dψ ◦ ξ parallel along c̃.
In particular, the following diagram commutes:

Pc,t0 ,t1
Tc(t0 ) M Tc(t1 ) M
dψ|c(t0 ) dψ|c(t1 )
Pc̃,t0 ,t1
Tc̃(t0 ) M̃ Tc̃(t1 ) M̃
Remark 2.5.10. In general, parallel transport depends on the curve joining two given
points. This means, in general we have Pc,t0 ,t1 6= Pĉ,s0,s1 if c and ĉ are two curves in M
with c(t0 ) = ĉ(s0 ) and c(t1 ) = ĉ(s1 ). In this respect, Euclidean space is not typical.
Example 2.5.11. Curve dependence of the parallel transport on (M, g) = (S 2 , gstd ) is

illustrated at: http://www.math.uiuc.edu/˜jms/java/dragsphere/
2.6 Geodesics
Definition 2.6.1. Let (M, g) be a semi-Riemannian manifold and c : [a, b] → M a C 1 -

curve. Then we call
Zb
1
E[c] := g ċ(t), ċ(t) dt
2
a
the energy of c.
63
Remark 2.6.2. If (M, g) is Riemannian, then g(ċ, ċ) ≥ 0 and therefore E[c] ≥ 0 (and
equal to 0 if and only if c is constant).
Question. Are there curves with minimal energy joining two given endpoints? More
generally, are there curves with “stationary energy”?
Definition 2.6.3. Let M be a differentiable manifold and c : [a, b] → M a smooth

curve. A variation of c is a smooth map
c : (−ε, ε) × [a, b] → M
with c(0, t) = c(t) for all t ∈ [a, b]. If c(s, a) = c(a) and c(s, b) = c(b) for all s ∈ (−ε, ε)
then we call c(s, t) a variation with fixed endpoints.
c(s, t) b
ξ
c(t) b b
c b
c
M M
∂c
The vector field ξ(t) := (0, t) is called the variational vector field.
∂s
Remark 2.6.4. The variational vector field ξ of a variation with fixed endpoints satisfies
ξ(a) = 0 and ξ(b) = 0.
Theorem 2.6.5 (First variation of the energy)

Let (M, g) be a semi-Riemannian manifold, let c : [a, b] → M be a smooth curve and let
c : (−ε, ε) × [a, b] → M be a variation of this curve. Let ξ be the variational vector field.
Write cs (t) = c(s, t). Then
Zb
d ∇
E[cs ] s=0
= − g ξ(t), ċ(t) dt + g ξ(b), ċ(b) − g ξ(a), ċ(a) .
ds dt
a
64
2.6 Geodesics
Proof. We compute:
Zb
d 1 d
E[cs ] s=0
= g ċs (t), ċs (t) dt
ds 2 ds s=0
a
Zb
1 ∂ ∂c ∂c
= g (s, t), (s, t) dt
2 ∂s s=0 ∂t ∂t
a
Zb
1 ∇ ∂c ∂c ∂c ∇ ∂c
= g (0, t), (0, t) + g (0, t), (0, t) dt
2 ∂s ∂t ∂t ∂t ∂s ∂t
a
Zb
∇ ∂c ∂c
= g (0, t), (0, t) dt
∂s ∂t ∂t
a
Zb
(∗) ∇ ∂c ∂c
= g (0, t), (0, t) dt
∂t ∂s ∂t
a
Zb
∇
= g ξ(t), ċ(t) dt
dt
a
Zb
d ∇
= g(ξ(t), ċ(t)) − g ξ(t), ċ(t) dt
dt dt
a
Zb
∇
= g ξ(b), ċ(b) − g ξ(a), ċ(a) − g ξ(t), ċ(t) dt.
dt
a
Equality (∗) holds because of torsion-freeness of the Levi-Civita connection.
Corollary 2.6.6
If the variation has fixed endpoints then
Zb
d ∇
E[cs ] s=0
= − g ξ(t), ċ(t) dt.
ds dt
a
65
Lemma 2.6.7
Let c : [a, b] → M be a smooth curve and ξ a smooth vector field along c. Then there exists a
variation c of c with variational vector field ξ. If ξ(a) = 0 and ξ(b) = 0, then we can choose
the variation with fixed endpoints.
Proof. a) We first consider the case that supp(ξ) is contained in a chart x : U → V , i.e.,
c(t) ∈ U whenever ξ(t) 6= 0.
ξ
x
U (c1 , . . . , cn )
c M
Rn ⊃ V
n
X ∂
We write ξ(t) = ξ j (t) and we set
∂xj c(t)
j=1
(
x−1 c1 (t), . . . , cn (t) + s ξ 1 (t), . . . , ξ n (t) , c(t) ∈ U
c(s, t) :=
c(t), c(t) ∈
6 U
Then we have for the corresponding variational vector field:
j
∂c j ∂c
(0, t) = dx (0, t)
∂s ∂s
∂(xj ◦ c)
= (0, t)
∂s

∂ cj (t) + sξ j (t)
=
∂s
s=0
= ξ j (t).
Hence the variation c has the variational vector field ξ. Moreover, if ξ vanishes at the
endpoints, then c has fixed endpoints.
b) In the general case, cover the compact set c([a, b]) with finitely many charts and
construct the variation piecewise.
Remark 2.6.8. Later, when we have the Riemannian exponential map at our disposal,
we will be able to directly write down a suitable variation without usage of charts.
66
2.6 Geodesics
Notation 2.6.9. Let M be a differentiable manifold and p, q ∈ M . Then we set

Ωp,q (M ) := smooth curves c : [a, b] → M with c(a) = p and c(b) = q .
Corollary 2.6.10
Let (M, g) be a semi-Riemannian manifold and c ∈ Ωp,q (M ). Then the following are equiv-
alent:
(i) The curve c is a “critical point” of the energy functional, i.e.,
d
E[cs ] s=0
=0
ds
for all variations cs of c with fixed endpoints;
(ii) For all t we have

∇
ċ(t) = 0.
dt
Proof. The implication “(ii)⇒(i)” is directly clear by Corollary 2.6.6. We show “(i)⇒(ii)”.
Let [a, b] be the parameter interval of c. Assume there exists a t0 ∈ (a, b) with
∇
dt ċ(t0 ) 6= 0. Then there exists a ξ0 ∈ Tc(t0 ) M with

∇
g ξ0 , ċ(t0 ) > 0
dt
because g is non-degenerate. Let ξ˜ be the parallel vector field along c with ξ(t
˜ 0 ) = ξ0 .
By continuity there exists an ε > 0 such that (t0 − ε, t0 + ε) ⊂ (a, b) and

∇
g ξ̃(t), ċ(t) > 0
dt
holds for all t ∈ (t0 − ε, t0 + ε). We choose a

̺
smooth function ̺ : [a, b] → R with ̺(t) > 0 for
all t ∈ (t0 − ε, t0 + ε) and ̺(t) = 0 otherwise.
a t0 b
67
Set ξ(t) := ̺(t) · ξ̃(t). Then we have:

(
∇ ˜ ∇ > 0 for t ∈ (t0 − ε, t0 + ε)
g ξ(t), ċ(t) = ̺(t) · g ξ(t), ċ(t) .
dt dt = 0 otherwise
By Lemma 2.6.7 we can choose a variation of c with fixed endpoints and variational
vector field ξ. Then we have for this variation
Zb
d ∇
E[cs ] s=0
= − g ξ(t), ċ(t) dt < 0
ds dt
a
∇
which contradicts the assumption. Hence we have dt ċ = 0 on (a, b) and by continuity
also on the whole of [a, b].
∇
Definition 2.6.11. A smooth curve c with dt ċ = 0 is called a geodesic.
Example 2.6.12. Let (M, g) = (Rn , geucl ) or (Rn , gMink ). In Cartesian coordinates
x1 , . . . , xn we have:
∇
ċ = 0 ⇐⇒ c̈1 = 0, . . . , c̈n = 0
dt
⇐⇒ cj (t) = pj + tv j
⇐⇒ c(t) = p + tv.
Hence geodesics are straight lines, parametrized with constant speed.
Lemma 2.6.13
For any geodesic c the quantity g ċ, ċ is constant.
Proof. We compute
d
∇
dt g ċ, ċ = 2 · g dt ċ , ċ = 0.
|{z}
=0
68
2.6 Geodesics
Definition 2.6.14. A smooth curve c is called
• parametrized by arc-length , if g(ċ, ċ) ≡ 1,
• parametrized by proper time , if g(ċ, ċ) ≡ −1
• parametrized proportional to arc-length , if g(ċ, ċ) ≡ α > 0,
• parametrized proportional proper time , if g(ċ, ċ) ≡ −α < 0 and
• a null curve, if g(ċ, ċ) ≡ 0.
Theorem 2.6.15 (Existence and uniqueness of geodesics)

Let (M, g) be a semi-Riemannian manifold.
For any p ∈ M and ξ ∈ Tp M there exists an open ξ

p c
interval I with 0 ∈ I and a geodesic c : I → M with b
c(0) = p and ċ(0) = ξ. M
If c : I → M and c̃ : I˜ → M are two such geodesics with c(0) = c̃(0) and ċ(0) = c̃(0),
˙ then
˜
c and c̃ coincide on their common domain I ∩ I.
Proof. In a chart x : U → V in p we consider the equation for a geodesic

n
∇ X
ċ = 0 ⇐⇒ k
c̈ + Γkij c1 , . . . , cn · ċi · ċj = 0
dt
i,j=1
for k = 1, . . . , n and ck = xk ◦ c. This is a system of ordinary differential equations of

second order. By the Theorem of Picard-Lindelöf the we get the assertion.
Remark 2.6.16. The system of differential equations is non-linear. Therefore we do not

have a-priori control over the maximal domain of definition I of the geodesic.
Remark 2.6.17. If ψ : M → M̃ is a local isometry, then

c : I → M is a geodesic ⇐⇒ ψ ◦ c : I → M̃ is a geodesic.
69
Example 2.6.18. Let M = (R2 \ {0}, geucl ) be the Euclidean plane with the origin re-
moved and let M̃ = {(x, y, z) ∈ R3 | x2 + y 2 = z 2 /3, z < 0)} be a cone with the cone
tip removed and equipped with √ the first fundamental form g̃. Now ψ : M → M̃ ,
ψ(u, v) = 2√u12 +v2 (u2 − v 2 , 2uv, − 3(u2 + v 2 )), can be checked to be a local isometry.
Hence ψ maps straight lines in M onto geodesics in M̃ .
bc
bc
ψ
R2 \ {0}
Definition 2.6.19. Let ψ : M → M be a diffeomorphism. Then we call
Fix(ψ) := {p ∈ M | ψ(p) = p}
the fixed point set of ψ.
Proposition 2.6.20
Let (M, g) be a semi-Riemannian manifold and ψ ∈ Isom(M, g).
Then for any p ∈ Fix(ψ) and any ξ ∈ Tp M with dψ|p (ξ) = ξ the geodesic c : I → M with
c(0) = p and ċ(0) = ξ
is entirely contained in Fix(ψ), i.e., for all t ∈ I we have c(t) ∈ Fix(ψ).
Proof. Set c̃(t) := ψ ◦ c(t). Since ψ is an isometry, c̃ is also a geodesic. Furthermore, we
70
2.6 Geodesics
have:

c̃(0) = ψ c(0) = ψ(p) = p = c(0) and

˙
c̃(0) = dψ|c(0) ċ(0) = dψ|p (ξ) = ξ = ċ(0).
Applying the uniqueness part of Theorem 2.6.15 we get for all t ∈ I:

c(t) = c̃(t) = ψ c(t) .
This means c(t) ∈ Fix(ψ) for all t.
Φp (ξ)
Example 2.6.21
We use Proposition 2.6.20 to determine the geodesics
of the sphere (S n , gstd ). Let p ∈ S n and ξ ∈ Tp S n . Let p
E ⊂ Rn+1 be the two-dimensional vector subspace
b
spanned by p and Φp (ξ). Let A : Rn+1 → Rn+1 be the Sn bc
0
reflection about E. Then A ∈ O(n + 1). Hence
ψ := A|S n ∈ Isom(S n , gstd ).
Then Fix(A) = E and therefore Fix(ψ) = E ∩ S n is a

E
great circle.
Proposition 2.6.20 implies that c(t) ∈ E ∩ S n for all t. Since geodesics on a Rie-
mannian manifold are parametrized proportional to arc-length we seek an arc-length
parametrization of this great circle:
Φp (ξ)
c(t) = p · cos(αt) + · sin(αt).
||Φp (ξ)||
We have to satisfy the initial conditions:
c(0) = p is satisfied.
d Φp (ξ)
c(0) = · α and therefore α = ||Φp (ξ)|| = ||ξ|| .
dt ||Φp (ξ)||
d
Then we get dt c(0) = Φp (ξ), i.e., ċ(0) = ξ. Thus the geodesic c with initial conditions
c(0) = p and ċ(0) = ξ is given by
Φp (ξ)
c(t) = p · cos ||ξ|| t + · sin ||ξ|| t .
||ξ||
Remark 2.6.22. Let (M, g) be a semi-Riemannian manifold and p ∈ M . For ξ ∈ Tp M let

cξ be the geodesic with
cξ (0) = p and ċξ (0) = ξ.
71
For α ∈ R set c̃(t) := cξ (αt). Then

∇ ˙ ∇ ∇
c̃(t) = α · ċξ (αt) = α2 ċξ (αt) = 0.
dt dt dt
Hence c̃ is also a geodesic. Since its initial conditions are
c̃(0) = cξ (0) = p,
˙
c̃(0) = α · ċξ (0) = αξ,
we conclude c̃ = cαξ . In particular, cξ (α) = cαξ (1).
Definition 2.6.23. Let M be a semi-Riemannian manifold and p ∈ M . For ξ ∈ Tp M

set
expp (ξ) := cξ (1)
if the maximal domain of the geodesic cξ contains 1. Furthermore, set
Dp := {ξ ∈ Tp M | 1 is contained in the maximal domain of cξ }.
Then we call expp : Dp → M the Riemannian exponential map (at the point p).
Remark 2.6.24
(1) By Remark 2.6.22 we know expp (t · ξ) = ctξ (1) = cξ (t). Thus t 7→ expp (tξ) is the
geodesic with initial values p and ξ.
(2) For any p ∈ M we have expp (0) = p because c0 is the constant curve c0 (t) = p.
(3) Let ξ ∈ Dp . Then cξ is defined on [0, 1]. Let

0 ≤ α ≤ 1. From cαξ (t) = cξ (αt) we see that Dp
cαξ is defined on 0, α1 ⊃ [0, 1]. Therefore b
0
αξ ∈ Dp . This shows that Dp is star-shaped
with respect to 0 ∈ Tp M . Tp M
S
(4) Set D := p∈M Dp ⊂ T M and exp : D → M , exp(ξ) := expπ(ξ) (ξ). The theory of
ordinary differential equations implies that D is open and that exp is a smooth map
(smooth dependence of solutions of the initial values). In particular, Dp = D ∩ Tp M
is open in Tp M .
72
2.6 Geodesics
Example 2.6.25
(1) Let (M, g) = (Rn , geucl ) or (Rn , gMink ). Then we have:
expp (ξ) = p + 1 · Φp (ξ) = p + Φp (ξ).
Here Dp = Tp Rn .
(2) Let (M, g) = (R2 \ {0}, geucl ). Then ξ

b
p
Dp = Tp M \ − t · Φp −1 (p) | t ≥ 1 . 0 bc
R2
Tp M
Dp
(3) Let (M, g) = (S n , gstd ). Then we have Dp = ξ
b b
Tp M and M p
Φp (ξ)
expp (ξ) = p · cos (||ξ||) + · sin (||ξ||).
||ξ||
Lemma 2.6.26
The differential of the map expp : Dp → M at 0 is given by the canonical isomorphism
d expp |0 = Φ0 : T0 Dp = T0 Tp M → Tp M.
Proof. Let ξ ∈ Tp M . Then we have:

−1
d d
d expp 0
Φ0 (ξ) = d expp 0
(tξ) t=0
= expp (tξ) t=0
= ξ.
dt dt
In the literature Lemma 2.6.26 is sometimes formulated slightly imprecisely as follows

idTp M = d expp 0
: Tp M → Tp M.
Corollary 2.6.27
For p ∈ M there exists an open neighborhood Vp ⊂ Dp ⊂ Tp M of 0, such that
expp |Vp : Vp → expp (Vp ) =: Up
73
is a diffeomorphism.
Proof. By Lemma 2.6.26 d expp |0 is invertible. The inverse function theorem yields the
claim.
Remark 2.6.28. In general, expp : Dp → expp (Dp ) ⊂ M is not a diffeomorphism because

expp is not injective in general. Moreover, d expp |ξ is not necessarily invertible for ξ 6= 0.
Example 2.6.29. Let (M, g) = (S n , gstd ). For p ∈ S n we have Dp = Tp M and
Φp (ξ) Tp S n
expp (ξ) = p · cos ||ξ|| + · sin ||ξ|| . p
||ξ|| b
b
{ξ ∈ Tp M | ||ξ|| = π}
In particular, for any ξ ∈ Tp M with ||ξ|| = π
we have Sn
b
−p
expp (ξ) = p · cos(π) = −p.
For ξ ∈ Tp M with ||ξ|| = π the differential d expp |ξ has the (n − 1)-dimensional kernel
{η ∈ Tξ Tp S n | Φξ (η) ⊥ ξ}.
Now we construct coordinates which are well adapted to the geometry and to this end
we choose a generalized orthonormal basis E1 , . . . , En of Tp M regarding g|p , that is
g|p (Ei , Ej ) = εi δij , εi ∈ {±1}.

n
P
We get a linear isomorphism A : Rn → Tp M , (α1 , . . . , αn ) 7→ αi Ei .
i=1
expp
Tp M ⊃ Vp ≈
Up ⊂ M
∼
= A
≈
R n ⊃ Vp
We put Vp := A−1 (Vp ). Then expp ◦A : Vp → Up is a diffeomorphism. Set

x := (expp ◦A)−1 . Then x : Up → Vp is a chart.
74
2.6 Geodesics
Definition 2.6.30. The coordinates we just defined are called Riemannian normal
coordinates around the point p.
In which sense are these coordinates well adapted to the geometry?
Proposition 2.6.31
Let (M, g) be a semi-Riemannian manifold and p ∈ M . Let gij : Vp → R be the metric
coefficients and Γkij : Vp → R be the Christoffel symbols in Riemannian normal coordinates
around p. Then we have:
x(p) = 0, gij (0) = εi δij , Γkij (0) = 0.

Proof. a) Clearly, we have x(p) = A−1 expp −1 (p) = A−1 (0) = 0.
b) Let e1 , . . . , en be the standard basis of Rn . Then

gij (0) = g|p dx−1 |0 (ei ), dx−1 |p (ej )

= g|p d(expp ◦A)|0 (ei ), d(expp ◦A)|0 (ei )

= g|p d expp |0 (Ei ), d expp |0 (Ej )
L. 2.6.26
= g|p (Ei , Ej )
= εi δij
c) Let v = (v 1 , . . . , v n ) ∈ Rn . Then c(t) = x−1 (tv) = expp (tAv) is a geodesic with

c(0) = p and ċ(0) = Av. In Riemannian normal coordinates the equation for a
geodesic is in this case
n
X
0 = c̈k (t) + Γkij c1 (t), . . . , cn (t) · ċi (t) · ċj (t).
i,j=1
Here ck (t) = xk (c(t)) = tv k , ċk (t) = v k and c̈k (t) = 0. For t = 0 we get
n
X
0=0+ Γkij (0, . . . , 0) · v i · v j .
i,j=1
75
Pn
For each k we define a bilinear form β k on Rn by β k (y, z) := k i j
i,j=1 Γi,j (0) y z .
These bilinear forms are symmetric because:
n
X n
X n
X
k
β (z, y) = Γkij (0) z i j
y = Γkji (0) z j i
y= Γkij (0) y i z j = β k (y, z).
i,j=1 j,i=1 i,j=1
Exchanging ∇ free of
indices torsion
Since we know that β k (v, v) = 0 for all v ∈ Rn , polarization yields β k (y, z) = 0 for
all y, z ∈ Rn . This means Γkij (0) = 0 for all i, j, k.
straight lines in Rn through 0 0 b
Tp M
M p b
Geodesic
in M
through p
Example 2.6.32. Let (M, g) = (Rn , geucl ) or (Rn , gMink ) and p ∈ M . Choose
A = Φp = canonical isomorphism Rn → Tp Rn .
Then we have expp (Av) = p + v, thus Riemannian normal coordinates around p are
given by
x : Rn → Rn , x(q) = q − p.
Up to translation by −p, Riemannian normal coordinates coincide with Cartesian coor-

dinates.
Corollary 2.6.33
In Riemannian normal coordinates we have for the Taylor expansion of gij : Vp → R
around 0:
gij (x) = εi δij + O(||x||2 ).
76
2.6 Geodesics
Proof. Expanding gij into a Taylor series at 0 yields

n
X ∂gij k 2
gij (x) = gij (0) + k
(0) · x + O ||x|| .
∂x
k=1
In the proof of Theorem 2.3.8 we found

n
∂gij X
l l
(0) = Γki (0) glj (0) + Γkj (0) gil (0)
∂xk
l=1
which is zero in our situation because the Christoffel symbols vanish at 0.
77
3 Curvature
We now come to one of the central concepts of differential geometry, that of curvature.
We will see that there are various inequivalent notions of curvature. We start with the
most basic one.
3.1 The Riemannian curvature tensor
Definition 3.1.1. Let (M, g) be a semi-Riemannian manifold and p ∈ M .

Let ξ ∈ Tp M and η, ζ ∈ Ξp (M ). Then we have ∇η ζ ∈ Ξp (M ) and
∇2ξ,η ζ := ∇ξ ∇η ζ − ∇∇ξ η ζ ∈ Tp M
is called the second covariant derivative of ζ in the direction ξ and η.
Lemma 3.1.2
The second covariant derivative ∇2ξ,η ζ depends on η only via η|p , i.e., if η, η̃ ∈ Ξp (M ) with
η|p = η̃|p then
∇2ξ,η ζ = ∇2ξ,η̃ ζ.
Proof. We choose Riemannian normal coordinates x around p. In these coordinates we

write (using the Einstein summation convention) the vector fields locally as:
∂ ∂ ∂
ξ = ξi , η = ηj , ζ = ζk .
∂xi p ∂xj ∂xk
Since the Christoffel symbols vanish at 0 we get
j
j ∂ i ∂η ∂
∇ξ η = ∇ξ i ∂ η j
= ξ i
(0)
∂xi p ∂x ∂x ∂xj p
79
3 Curvature
and therefore
j
i ∂η k ∂ ∂η j ∂ζ k ∂
∇∇ξ η ζ = ξ i
(0)∇ ∂ ζ = ξi (0) (0) . (3.1)
∂x ∂xj p ∂xk ∂x i ∂x j ∂xk p
Moreover,
k ∂ j ∂ζ k ∂ ∂
∇η ζ = ∇η j ∂ ζ =η + ζ k Γm
jk
∂xj ∂xk j
∂x ∂x k ∂xm
and hence (again using that the Christoffel symbols vanish)
k ∂
j ∂ζ j k m ∂
∇ξ ∇η ζ = ∇ξ i ∂ η + η ζ Γjk m
∂xi p ∂xj ∂xk ∂x
∂η j ∂ζ k ∂ ∂2ζ k ∂ ∂Γm
jk ∂
= ξi i
(0) j
(0) + ξ i η j (0) (0) + ξ i η j (0)ζ k (0) .
∂x ∂x ∂xk p
j
∂x ∂x i ∂xk p ∂xi ∂xm p
(3.2)
Subtracting (3.1) from (3.2) we see that the terms containing a derivative of the η j cancel
and we are left with

2 i j ∂2ζ k i j m ∂Γkim ∂
∇ξ,η ζ = ξ η (0) i j (0) + ξ η (0)ζ (0) i
(0) . (3.3)
∂x ∂x ∂x ∂xk p
This expression depends on η only via the η j (0) which are the coefficients of η|p .
Consequence. The express ∇2ξ,η ζ is well defined for ξ, η ∈ Tp M and ζ ∈ Ξp .
Lemma 3.1.4
For ξ, η ∈ Tp M and ζ ∈ Ξp (M )
R(ξ, η)ζ := ∇2ξ,η ζ − ∇2η,ξ ζ
depends only on ζ via ζ|p . Thus R(ξ, η)ζ ∈ Tp M is well defined for ξ, η, ζ ∈ Tp M .
Proof. Again we choose Riemann normal coordinates around p and recall (3.3):
!
2 i j ∂2ζ k m
∂Γkjm ∂
∇ξ,η ζ = ξ (0)η (0) i j
(0) + ζ (0) i
(0) .
80
Relabeling summation indices and using the Schwarz theorem we get
R(ξ, η)ζ
!
k
∂2ζ k ∂Γ jm ∂
= ξ i (0)η j (0) − ξ j (0)η i (0) i j
(0) + ζ m (0) i
(0)
!
i j ∂2ζ k ∂2ζ k ∂Γkjm ∂Γkim ∂
= ξ (0)η (0) (0) − (0) + ζ m (0) (0) m
− ζ (0) (0)
∂xi ∂xj ∂xj ∂xi ∂xi ∂x j ∂xk p
!
∂Γkjm ∂Γkim ∂
= ξ i (0)η j (0)ζ m (0) (0) − (0) .
∂xi ∂xj ∂xk p
Definition 3.1.5. The map
R : Tp M × Tp M × Tp M → Tp M
(ξ, η, ζ) 7→ R(ξ, η)ζ
is called the Riemann curvature tensor at the point p.
Representation in local coordinates.

Let x : U → V be a chart on M . Then R is determined on U by smooth functions
l
Rkij : V → R, defined by
n
∂ ∂ ∂ X
l ∂
R i
, j k
= Rkij . (3.4)
∂x ∂x ∂x ∂xl
l=1
As we have already seen, we have in Riemann normal coordinates:
l
∂Γljk ∂Γlik
Rkij (0) = (0) − (0)
∂xi ∂xj
Remark 3.1.6. One can check (not difficult but tedious) that we have in arbitrary coor-
dinates
n
l
∂Γljk ∂Γlik X
Rkij = − + (Γm l m l
kj Γmi − Γki Γmj )
∂xi ∂xj
m=1
In particular, if the curvature tensor R : Tp M × Tp M × Tp M → Tp M does not vanish at

the point p, then there does not exist a chart containing p for which Γkij ≡ 0.
81
3 Curvature
Proposition 3.1.7 (Symmetries of the curvature tensor)

Let (M, g) be a semi-Riemannian manifold, p ∈ M and ξ, η, ζ, ν ∈ Tp M . Then we have:
(1) R : Tp M × Tp M × Tp M → Tp M is trilinear;
(2) R(ξ, η)ζ = −R(η, ξ)ζ;
(3) g|p (R(ξ, η)ζ, ν) = −g|p (R(ξ, η)ν, ζ);
(4) First Bianchi-identity:

R(ξ, η)ζ + R(η, ζ)ξ + R(ζ, ξ)η = 0:
(5) g|p (R(ξ, η)ζ, ν) = g|p (R(ζ, ν)ξ, η).
Proof.
(1) is obvious because ∇2ξ,η ζ is already R-linear in ξ, η and ζ.
(2) is also clear by definition.
(3) We choose Riemannian normal coordinates around p and consider the special case
∂ ∂ ∂ ∂
ξ= , η= , ζ= , ν= .
∂xi p ∂xj p ∂xk p ∂xl p
Then we find
! !
∂ ∂ ∂ ∂
g|p R(ξ, η)ζ, ν = g|p R , ,
∂xi p ∂x
j
p ∂xk p ∂x
l
p
n
!
X
m ∂ ∂
= g|p Rkij (0) ,
∂xm p ∂xl p
m=1
n
!
X
m ∂ ∂
= Rkij (0) · g|p ,
∂xm p ∂x
l
p
m=1
Xn
m
= gml (0) · Rkij (0).
m=1
From the proof of Theorem 2.3.8 we recall

n
∂gij X
= (gmj Γm m
ki + gmi Γkj )
∂xk
m=1
82
and thus, in Riemannian normal coordinates,
n
∂ 2 gij X ∂Γm
ki
∂Γmkj
(0) = gmj (0) (0) + gmi (0) (0) .
∂xk ∂xl m=1
∂xl ∂xl
Thus
∂ 2 gij ∂ 2 gij
0= (0) − (0)
∂xk ∂xl ∂xl ∂xk
n
X ∂Γmki
∂Γmkj ∂Γm
li
∂Γm
lj
= gmj (0) (0) + gmi (0) (0) − gmj (0) k (0) − gmi (0) k (0)
m=1
∂xl ∂xl ∂x ∂x
n
X
m m
= gmj (0)Rilk (0) + gmi (0)Rjlk (0)
m=1
Renaming the indices via l 7→ i, k 7→ j, i 7→ k, j 7→ l leads to
n
X
m m
0= gml (0)Rkij (0) + gmk (0)Rlij (0)
m=1
and therefore
n
X n
X
m m
gml (0)Rkij (0) = − gmk (0)Rlij (0).
m=1 m=1
This proves the assertion for coordinate fields ξ, η, ζ, ν of Riemannian normal coor-
dinates.
By multilinearity the assertion follows for general ξ, η, ζ and ν.
(4) The first Bianchi-identity is equivalent to
l l l
Rkij + Rijk + Rjki = 0.
We check this in Riemann normal coordinates:
l l l
Rkij (0) + Rijk (0) + Rjki (0)
∂Γljk ∂Γlik ∂Γlki ∂Γlji ∂Γlij ∂Γlkj

= (0) − (0) + (0) − (0) + (0) − (0)
∂xi ∂xj ∂xj ∂xk ∂xk ∂xi
= 0.
83
3 Curvature
(5) Proof by an explicit calculation:

(4)
0 = g|p (R(η, ζ)ξ, ν) + g|p (R(ζ, ξ)η, ν) + g|p (R(ξ, η)ζ, ν)
+ g|p (R(ζ, ξ)ν, η) + g|p (R(ξ, ν)ζ, η) + g|p (R(ν, ζ)ξ, η)
+ g|p (R(ξ, ν)η, ζ) + g|p (R(ν, η)ξ, ζ) + g|p (R(η, ξ)ν, ζ)
+ g|p (R(ν, η)ζ, ξ) + g|p (R(η, ζ)ν, ξ) + g|p (R(ζ, ν)η, ξ)
(2),(3)
= 2g|p (R(ξ, η)ζ, ν) + 2g|p (R(ζ, ν)η, ξ)
= 2(g|p (R(ξ, η)ζ, ν) − g|p (R(ζ, ν)ξ, η)) .
Example 3.1.8. Let (M, g) = (Rn , geucl ) or (Rn , gMink ). In Cartesian coordinates we have
Γkij = 0. Thus we get Rkij
l = 0 for all i, j, k, l and therefore R ≡ 0.
Definition 3.1.9. A semi-Riemannian manifold with R ≡ 0 is called flat.
Warning. In the literature there are two different sign conventions for R: For example,
our R is the negative of the curvature tensor as defined in [ON83].
Lemma 3.1.11
Let (M, g) and (M̃ , g̃) be semi-Riemannian manifolds and ψ : M → M̃ a local isometry. Let
p ∈ M . Then the curvature tensors R of M at p and R̃ of M̃ at ψ(p) are related by:

dψ|p R(ξ, η)ζ = R̃ dψ|p (ξ), dψ|p (η) dψ|p (ζ)
for all ξ, η, ζ ∈ Tp M .
Proof. Let x : U → V be a chart on M with p ∈ U . By making U smaller if necessary we

can assume that ψ : U → Ũ := ψ(U ) is a diffeomorphism. Then x̃ := x ◦ ψ −1 : Ũ → V
is a chart on M̃ .
84
Since ψ is a local isometry, it follows that gij = g̃ij : V → R, where the gij are the
components of g w.r.t. x and the g̃ij are the components of g̃ w.r.t. x̃. Therefore the
Christoffel symbols coincide, Γkij = Γ̃kij , hence so do the components of the curvature
l
tensors, Rkij l . From (3.4) we conclude
= R̃kij
X n ∂
∂ ∂ ∂ l
dψ R , = R kij dψ
∂xi ∂xj ∂xk ∂xl
l=1
n
X
l ∂
= Rkij
∂ x̃l
l=1
n
X
l ∂
= R̃kij
∂ x̃l
l=1

∂ ∂ ∂
= R̃ ,
∂ x̃i ∂ x̃j ∂ x̃k
∂ ∂
∂
= R̃ dψ , dψ dψ .
∂xi ∂xj ∂xk
∂
This proves the lemma for the coordinate basis tangent vectors ∂xi
. By trilinearity of R
it follows for all tangent vectors.
Alternatively one can define the curvature tensor as a multilinear map R : Tp M ×Tp M ×
Tp M × Tp M → R by
R(ξ, η, ζ, ν) = g R(ξ, η)ζ, ν .
In this version, R is known as the Riemannian (4, 0)-curvature tensor. In local coordinates
x : U → V around p, we define Rijkl : V → R by
!
∂ ∂ ∂ ∂
Rijkl (x(p)) := R , , , .
∂xi p ∂xj p ∂xk p ∂xl p
Then we have
Prop.
3.1.7(5)
Rijkl = Rklij

∂ ∂ ∂ ∂
= g R , ,
∂xk ∂xl ∂xi ∂xj
n
!
X
m ∂ ∂
= g Rikl ,
m=1
∂xm ∂xj
n
X
m ∂ ∂
= Rikl g ,
∂xm ∂xj
m=1
Xn
m
= gmj Rikl .
m=1
85
3 Curvature
Hence we have
n
X
m
Rijkl = gmj Rikl
m=1
We have lowered the upper index. On the other hand we have

n
X n
X
l l m
Rkij = δm Rkij = g al gma Rkij
m
,
m=1 a,m=1
hence
n
X
l
Rkij = gal Rkaij
a=1
In this case we have raised the index.
Proposition 3.1.12
Let (M, g) be a semi-Riemannian manifold. In Riemannian normal coordinates we have:
n
1 X
gij (x) = εi δij + Rikjl (0)xk xl + O ||x||3 .
3
k,l=1
Proof. We already know that gij (x) = εi δij + O (||x||2 ) by Corollary 2.6.33. In the fol-
lowing we will use the Einstein summation convention and the following abbreviations
∂f ∂2f
f,k := and f,kℓ =
∂xk ∂xk ∂xℓ
for the first and the second partial derivatives. In the proof of Theorem 2.3.8 we have
seen that
gij,k = Γm m
ki gmj + Γkj gmi .
We differentiate this equation with respect to xℓ , evaluate at 0 and use that the Christof-
fel symbols vanish at 0:
gij,kℓ (0) = Γm m
ki,ℓ (0) · gmj (0) + Γkj,ℓ (0) gmi (0) . (3.5)
Claim:
Γkij,ℓ (0) + Γkℓi,j (0) + Γkjℓ,i (0) = 0 . (3.6)
86
Proof of the claim: In normal coordinates the straight lines t 7→ t · x give geodesics. The
equation for geodesics then looks like:
0 = Γkij (t · x) xi xj .
We differentiate this with respect to t and evaluate at t = 0:
d
0= Γkij (tx) xi xj = Γkij,ℓ (0) xℓ xi xj .
dt t=0
Thus we have for every k a polynomial of third degree in x, namely

P k (x) := Γkij,ℓ (0) xi xj xℓ , which vanished identically. Thus for every monomial
xα xβ xγ the sum of coefficients Γkij,ℓ (0) with xi xj xℓ = xα xβ xγ has to vanish. The
six permutations of the three lower indices yield
Γkij,ℓ (0) + Γkℓi,j (0) + Γkjℓ,i (0) + Γkji,ℓ (0) + Γkiℓ,j (0) + Γkℓj,i (0) = 0.
The symmetry of the Christoffel symbols in their two lower indices implies the claim.✓
ℓ (0) = Γℓ ℓ
From Rkij jk,i (0) − Γik,j (0) we conclude:

Rkℓij (0) = Γm m
jk,i (0) − Γik,j (0) gmℓ (0)
(3.6)
= − Γm m m
ij,k (0) + Γki,j (0) + Γik,j (0) gmℓ (0)

= − Γm m
ij,k (0) + 2Γki,j (0) gmℓ (0) . (3.7)
Thus we have:
Prop. 3.1.7
2Rikjℓ (0) xk xℓ = (−Rkijℓ (0) − Rℓjik (0)) xk xℓ
(3.7)
= Γm m k ℓ
jℓ,k (0) + 2Γkj,ℓ (0) gmi (0) x x

+ Γm m k ℓ
ik,ℓ (0) + 2Γℓi,k (0) gmj (0) x x
(∗)
= Γm m k ℓ
jℓ,k (0) + 2Γkj,ℓ (0) gmi (0) x x

+ Γm m ℓ k
iℓ,k (0) + 2Γki,ℓ (0) gmj (0) x x
(3.5)
= gij,ℓk (0) + 2gij,kℓ (0) · xk xℓ
= 3gij,kℓ (0) · xk xℓ .
At equality (∗) we renamed the summation parameter k to ℓ and vice versa. Thus we
get for the second term in the Taylor expansion
1 1
gij,kℓ (0) xk xℓ = Rikjℓ (0) · xk xℓ .
2 3
87
3 Curvature
3.2 Sectional curvature

The Riemannian curvature tensor contains the full curvature information of a Rieman-
nian manifold but for many applications other curvature entities are more suitable. We
will introduce the sectional curvature, Ricci curvature and scalar curvature in this and
the following sections.
We start with some linear algebra. Let V be a finite dimensional real vector space with
a non-degenerate symmetric bilinear form h·, ·i. Later we will apply this to V = Tp M
and h·, ·i = g|p (·, ·).
Definition 3.2.1. A subvector space U ⊂ V is called non-degenerate, if

h·, ·i |U ×U : U × U → R is a non-degenerate bilinear form on U . We define:

Gk V, h·, ·i := k-dimensional, non-degenerate subvector spaces of V .
Note that every subvector space is non-degenerate if h·, ·i is definite. We set

Q : V × V → R, Q(ξ, η) := hξ, ξi hη, ηi − hξ, ηi2 .
Lemma 3.2.2
For two-dimensional subvector spaces E ⊂ V the following assertions are equivalent:
(i) E ∈ G2 (V, h·, ·i);
(ii) there exists a basis ξ, η of E with Q(ξ, η) 6= 0;
(iii) for all basis ξ, η of E we have Q(ξ, η) 6= 0.
Proof. With respect to any basis ξ, η of E, the bilinear form h·, ·i |E×E is represented by
the matrix
hξ, ξi hη, ξi
Aξ,η := .
hξ, ηi hη, ηi
Then we have Q(ξ, η) = det Aξ,η which proves the lemma.
88
Remark 3.2.3
(a) If h·, ·i is positive definite, then
p
Q(ξ, η) = area of the parallelogram spanned by ξ and η.
(b) The two-dimensional subvector space E ⊂ V is degenerate if and only if there exists
a basis ξ, η of E with hξ, ξi = hξ, ηi = 0. Namely,
“⇐”: Q(ξ, η) = hξ, ξi hη, ηi − hξ, ηi2 = 0.
| {z } | {z }
=0 =0
“⇒”: Let E be degenerate, i.e., h·, ·i |E×E is degenerate. Then there exists a
ξ ∈ E \ {0} with hξ, ζi = 0 for all ζ ∈ E. Now complete ξ by some η to
a basis of E. ✓
Example 3.2.4
Let V = R3 with the Minkowski product
x0
hhξ, ηii = −ξ 0 η 0 + ξ 1 η 1 + ξ 2 η 2 .
b
Consider the lightcone

x1 , x2
C := {ξ ∈ R3 \ {0} | hhξ, ξii = 0}.
C
Then the plane E ⊂ R3 is degenerate if and only if
E = Tp C for some p ∈ C.
Namely, assume c : (−ε, ε) → C is a smooth curve with c(0) = p and ċ(0) = ξ. Then we
have:
hhc(t), c(t)ii = 0 ∀ t ∈ (−ε, ε)
d
⇒ 0= hhc(t), c(t)ii = 2 hhċ(0), c(0)ii = 2 hhξ, pii
dt t=0
⇒ Tp C ⊂ p⊥ , where both are two-dimensional subvector spaces of R3
⇒ Tp C = p⊥
⇒ for ξ = p and any η ∈ Tp C which is not a multiple of ξ we obtain a basis of Tp C

with hhξ, ξii = hhξ, ηii = 0
⇒ Tp C is degenerate.
Conversely, if E is degenerate, then we choose a basis ξ, η of E such that hhξ, ξii =

hhξ, ηii = 0. We put p := ξ. Clearly p ∈ C. Now we have E ⊂ p⊥ = Tp C. Since both E
and Tp C are two-dimensional we conclude E = Tp C. ✓
89
3 Curvature
p
b b b
p⊥
degenerate non-degenerate non-degenerate

(indefinite) (definite)
Lemma 3.2.5
Let V be a finite-dimensional real vector space with non-degenerate symmetric bilinear
form h·, ·i. Let R : V × V × V × V → R be multilinear with
R(ξ, η, ζ, ν) = −R(η, ξ, ζ, ν) = −R(ξ, η, ν, ζ)
for all ξ, η, ζ, ν ∈ V . Then for E ∈ G2 (V, h·, ·i) and any basis ξ, η of E the expression
R(ξ, η, η, ξ)
K(E) :=
Q(ξ, η)
does not depend on the choice of the basis ξ, η of E, but only on E itself.
Proof. Let µ, ν be another basis of E with µ = aξ + bη and ν = cξ + dη. Then we have:

R(µ, ν, ν, µ) = R(aξ + bη, cξ + dη, cξ + dη, aξ + bη)
= adcb · R(ξ, η, ξ, η) + adda · R(ξ, η, η, ξ) + bccb · R(η, ξ, ξ, η)
+ bcda · R(η, ξ, η, ξ)

= − abcd + a2 d2 + b2 c2 − abcd · R(ξ, η, η, ξ)
= (ad − bc)2 · R(ξ, η, η, ξ) (3.8)
The map R1 : V × V × V × V → R, defined by
R1 (ξ, η, ζ, ν) := hξ, νi hη, ζi − hξ, ζi hη, νi
has all the symmetries of the curvature tensor as in Proposition 3.1.7. Hence we get
R1 (µ, ν, ν, µ) = (ad − bc)2 R1 (ξ, η, η, ξ) . (3.9)
| {z } | {z }
= Q(µ, ν) = Q(ξ, η)
90
Dividing (3.8) by (3.9) proves the lemma.
[
Set G2 (M, g) := G2 (Tp M, g|p ).
p∈M
Definition 3.2.6. The function K : G2 (M, g) → R, defined by
R(ξ, η, η, ξ)
K(E) := ,
Q(ξ, η)
where ξ, η is a basis of E, is called sectional curvature of (M, g). Here R is the Rie-
mannian (4, 0)-curvature tensor.
Remark 3.2.7. The sectional curvature is only defined for manifolds of dimension at
least 2. If dim(M ) = 1, then R(ξ, η, ζ, ν) = 0 for all ξ, η, ζ, ν ∈ Tp M due to the skew-
symmetry in ξ and η.
Definition 3.2.8. If (M, g) is a two-dimensional semi-Riemannian manifold, then we

call
K : M → R, K(p) := K(Tp M )
the Gauß curvature of M .
Remark 3.2.9. The sectional curvature determines the curvature tensor, as can be seen
by
6R(ξ, η, ζ, ν) = K(ξ + ν, η + ζ)Q(ξ + ν, η + ζ) − K(η + ν, ξ + ζ)Q(η + ν, ξ + ζ)

− K(ξ, η + ζ)Q(ξ, η + ζ) − K(η, ξ + ν)Q(η, ξ + ν)
− K(ζ, ξ + ν)Q(ζ, ξ + ν) − K(ν, η + ζ)Q((ν, η + ζ)
+ K(ξ, η + ν)Q(ξ, η + ν) + K(η, ζ + ξ)Q(η, ζ + ξ)
+ K(ζ, η + ν)Q(ζ, η + ν) + K(ν, ξ + ζ)Q(ν, ξ + ζ)
+ K(ξ, ζ)Q(ξ, ζ) + K(η, ν)Q(η, ν) − K(ξ, ν)Q(ξ, ν) − K(η, ζ)Q(η, ζ)
91
3 Curvature
for all ξ, η, ζ, ν ∈ Tp M , for which the corresponding sectional curvatures are

defined. The set of quadruples (ξ, η, ζ, ν), that satisfies this, is open and
dense in Tp M × Tp M × Tp M × Tp M . By continuity this determines R on all of
Tp M × Tp M × Tp M × Tp M .
Special case: If K(E) only depends on p but not on the particular plane E ⊂ Tp M (satis-
fied automatically if dim(M ) = 2, but not in general if dim(M ) ≥ 3), then:

R(ξ, η, ζ, ν) = K(p) hη, ζi hξ, νi − hξ, ζi hη, νi .
Moreover, we always have: K = 0 ⇔ R = 0.
3.3 Ricci- and scalar curvature

The Riemann curvature tensor and sectional curvature can be computed from one an-
other. They contain the same amount of information. Both are rather complicated ob-
jects. In this section we introduce two simplified curvature concepts which however
contain less information than the full curvature tensor.
Let (M, g) be a semi-Riemannian manifold and p ∈ M . The Riemann curvature tensor
at the point p ∈ M is a multilinear map
R : Tp M × Tp M × Tp M → Tp M.
For fixed ξ, η ∈ Tp M we get a linear map

R(ξ, ·)η : Tp M → Tp M, ζ 7→ R(ξ, ζ)η.
Definition 3.3.1. The map ric : Tp M × Tp M → R,
ric(ξ, η) := −tr(R(ξ, ·)η) = tr(R(·, ξ)η)
is called the Ricci curvature at the point p.
Remark 3.3.2. Let V be a n-dimensional R-vector space with non-degenerate symmet-

ric bilinear form g and E1 , . . . , En be a generalized orthonormal basis of (V, g), that is
g(Ei , Ej ) = εi δi,j with εi = ±1. Then for every endomorphism A : V → V we have
n
X
tr(A) = εi · g(A(Ei ), Ei ). (3.10)
i=1
92
Why? If we define ωi : V → R by ωi (ξ) := εi · g(ξ, Ei ) then ω1 , . . . , ωn is the dual basis

of V ∗ to E1 , . . . , En . Hence
n
X n
X
tr(A) = ωi (A(Ei )) = εi · g(A(Ei ), Ei ).
i=1 i=1
The local description of Ricci curvature is similar to that of the semi-Riemannian metric
itself: For any chart x : U → V of M we define the functions

∂ ∂
ricij : V → R, ricij (x(p)) := ric , .
∂xi p ∂xj p
Lemma 3.3.3 (Properties of the Ricci curvature)
(i) The map ric is bilinear and symmetric on Tp M .
(ii) For any generalized orthonormal basis E1 , . . . , En of (Tp M, g|p ) we have:

n
X
ric(ξ, η) = εi · g(R(ξ, Ei )Ei , η).
i=1
n
X
k
(iii) We have: ricij = Rikj .
k=1
Proof. (i) Bilinearity of ric follows directly from trilinearity of R. We show symmetry
of ric:
n
X
ric(η, ξ) = εi · g R(η, Ei )Ei , ξ
i=1
Prop.
Xn
3.1.7(5)
= εi · g R(Ei , ξ)η, Ei
i=1
Prop.
3.1.7 n
X
(2),(3)
= εi · g R(ξ, Ei )Ei , η
i=1
= ric(ξ, η).
(ii) is clear from (3.10).
93
3 Curvature
∂ ∂
∂
∂
(iii) We fix i and j and we have ricij = ric ∂x i , ∂xj = tr ζ 7→ −R ∂x i , ζ ∂xj . W.r.t.
∂
the basis ∂x∂ 1 , . . . , ∂x∂ n the endomorphism ζ 7→ −R ∂x ∂
i , ζ ∂xj has the matrix rep-
resentation
l l
(−Rjik )kl = (Rjki )kl .
Pn k
Thus we get that ricij = k=1 Rjki and because of (i) we have ricij = ricji , which
yields the assertion.
We defined Ricci curvature using the Riemann curvature tensor. Since the curvature
tensor and sectional curvature contain the same information, Ricci curvature should
also be computable in terms of sectional curvature. Indeed, Ricci curvature can can be
computed by averaging the sectional curvature of certain planes.
Lemma 3.3.4
Let (M, g) be a semi-Riemannian manifold and p ∈ M . If ξ ∈ Tp M with g(ξ, ξ) 6= 0 and if
E2 , . . . , En is a generalized orthonormal basis of ξ ⊥ , then
n
X E3
ric(ξ, ξ) = g(ξ, ξ) · K(span{ξ, Ej })
j=2
| {z } ξ
This is essentially the
mean value of K on all
planes containing ξ.
E2
Proof. W.l.o.g. let g(ξ, ξ) = ±1. Write ξ =: E1 . Then E1 , . . . , En forms a generalized

orthonormal basis of Tp M . Therefore
n
X
ric(ξ, ξ) = g(Ei , Ei ) · g R(ξ, Ei )Ei , ξ
i=1
n
X
= g(Ei , Ei ) · g R(ξ, Ei )Ei , ξ
i=2
n
X
= g(Ei , Ei ) · K span{ξ, Ei } · g(ξ, ξ)g(Ei , Ei ) − g(ξ, Ei )2
| {z }
i=2
=0
n
X
= g(ξ, ξ) · K span{ξ, Ei } .
i=2
94
Remark 3.3.5. Lemma 3.3.4 expresses ric(ξ, ξ) in terms of sectional curvatures provided
g(ξ, ξ) 6= 0. Since g is non-degenerate the set of vectors ξ ∈ Tp M with g(ξ, ξ) 6= 0 is
dense in Tp M . By continuity, ric(ξ, ξ) is determined for all ξ ∈ Tp M . By polarization,
this determines the values of ric(ξ, η) for all ξ, η ∈ Tp M via
1

ric(ξ, η) = 2 ric(ξ +η, ξ +η) − ric(ξ, ξ) − ric(η, η) .
Remark 3.3.6. Both maps ric : Tp M × Tp M → R and g : Tp M × Tp M → R are bilinear

and symmetric. The second map g is in addition non-degenerate. Thus there exists a
unique endomorphism Ric : Tp M → Tp M such that

ric(ξ, η) = g Ric(ξ), η
for all ξ, η ∈ Tp M .
In local coordinates: For any chart x : U → V we get functions Ricji : V → R by:

! n
∂ X ∂
Ric = Ricji (x(p))
∂xi p ∂xj p
j=1
We compute:

∂ ∂
ricij = ric ,
∂xi ∂xj

∂ ∂
= g Ric , j
∂xi ∂x
n
!
k ∂ ∂
X
= g Rici k , j
∂x ∂x
k=1
n
X
k ∂ ∂
= Rici · g , .
∂xk ∂xj
k=1
We have shown:
n
X
ricij = Ricki · gkj
k=1
The functions ricij are obtained from the functions Ricki by lowering the upper index.
Similarly, the Ricki can be obtained from the ricij by raising one index.
95
3 Curvature
Definition 3.3.7. The map scal : M → R defined by
scal(p) := tr(Ric|p )
is called the scalar curvature of M .
Lemma 3.3.8
(i) In local coordinates we have
n
X n
X

scal(p) = Ricii x(p) = ricij x(p) · gij x(p) .
i=1 i,j=1
(ii) For a generalized orthonormal basis E1 , . . . , En of Tp M we have

n
X
scal(p) = εi · ric(Ei , Ei ).
i=1
Proof. Clear.
Remark 3.3.9. Let us consider the special case when dim(M ) = 2. Let K be the Gauß
curvature, i.e., K(p) = K(Tp M ). Then the curvature tensor is given by

R(ξ, η, ζ, ν) = K(p) g(η, ζ)g(ξ, ν) − g(ξ, ζ)g(η, ν) .
Thus we get for the Ricci curvature
2
X
ric(ξ, η) = εi · R(ξ, Ei , Ei , η)
i=1
2
X
= K(p) εi g(Ei , Ei )g(ξ, η) − g(ξ, Ei )g(Ei , η)
i=1

= K(p) 2g(ξ, η) − g(ξ, η)
= K(p) · g(ξ, η).
96
3.4 Jacobi fields
This shows
ric = K · g
and
scal = 2K.
In the case of surfaces the Riemann curvature tensor, sectional curvature (Gauß curva-
ture), Ricci curvature and scalar curvature all determine each other. In higher dimen-
sions this is no longer so.
Remark 3.3.10. The following table shows how the different notions of curvature de-
pend on each other:
dim M 2 3 ≥4
R
m m m
K
m m ⇓
ric
m ⇓ ⇓
scal
Remark 3.3.11. In the physics literature the following notation in local coordinates is
often used:
l
• for R and R ones writes: Rijk and Rijkl (as here),
• for Ric and ric one write: ricij = Rij and Ricji = Rij ,
• for scal one write: scal = R.
3.4 Jacobi fields

In order to better understand the behavior of geodesics we will linearize the geodesic
equations. This leads to the Jacobi fields and relates geodesics and curvature.
Definition 3.4.1. Let M be a semi-Riemannian manifold. A variation of curves

c : (−ε, ε) × I → M is called a geodesic variation if for every s ∈ (−ε, ε) the curve
t 7→ cs (t) := c(s, t)
is a geodesic.
97
3 Curvature
∂
Let ξ(t) := c(0, t) be the corresponding variational vector field. Then we have:
∂s
2
∇ ∇∇ ∂
ξ(t) = c(s, t)|s=0
dt ∂t ∂t ∂s
∇∇ ∂
= c(s, t)|s=0
∂t ∂s ∂t
∇ ∇ ∂ ∂c ∂c ∂c
= c(s, t) |s=0 + R (0, t), (0, t) (0, t)
∂s |∂t ∂t{z } ∂t ∂s ∂t
≡0 since cs geodesic
= R(ċ0 (t), ξ(t))ċ0 (t)
Definition 3.4.2. The equation for vector fields ξ along a geodesic c0

2
∇
ξ = R(ċ0 , ξ)ċ0
dt
is called the Jacobi equation. Its solutions are called Jacobi fields.
The above computation shows that the variational vector field of a geodesic variation
is a Jacobi field.
Proposition 3.4.3
Let M be a n-dimensional semi-Riemannian manifold, c : I → M a geodesic and t0 ∈ I.
For all ξ, η ∈ Tc(t0 ) M there exists a unique Jacobi field J along c with
∇
J(t0 ) = ξ and J(t0 ) = η.
dt
The set of Jacobi fields along c forms a 2n-dimensional vector space.
Proof. Let E1 (t0 ), . . . , En (t0 ) be a basis of Tc(t0 ) M . By parallel transport

Pn along c we
obtain a basis E1 (t), . . . , En (t) of Tc(t) M for all t ∈ I. Write J(t) = v j (t)E (t).
j=1 j
98
3.4 Jacobi fields
∇ 2
Pn j (t)E (t)
Then dt J(t) = j=1 v̈ j and
n
X
R(ċ(t), J(t))ċ(t) = v j (t)R(ċ(t), Ej (t))ċ(t).
j=1
Pn k
Write R(ċ(t), Ej (t))ċ(t) = k=1 aj (t)Ek (t). Then J is a Jacobi field if and only if
n
X n
X
k
v̈ Ek = akj v j Ek ,
k=1 j,k=1
hence if and only if

n
X
v̈ k = akj v j for all k = 1, . . . , n
j=1
This is a linear system of ordinary differential equations of second order. Thus solutions
exist (on all of I) and are uniquely determined by the initial data v k (t0 ) and v̇ k (t0 ), i.e.,
∇
by J(t0 ) and dt J(t0 ).
The linearity of the Jacobi equation implies that its solution space forms a vector space.
∇
The map {Jacobi fields} → Tc(t0 ) M ⊕ Tc(t0 ) M , J 7→ (J(t0 ), dt J(t0 )) is a vector space
isomorphism. In particular, the dimension of the space of Jacobi fields along c equals
2n.
Example 3.4.4. If M is flat then the equation for Jacobi fields is simply given by
2
∇
J ≡ 0.
dt
Hence
{Jacobi fields} = ξ(t) + t η(t) | ξ, η parallel .
Example 3.4.5. Let c be a geodesic in an arbitrary semi-Riemannian manifold. Then the

vector field J(t) := (a + bt)ċ(t) is a Jacobi field for any a, b ∈ R. Namely, we have:
2
∇
J(t) = 0, and R(ċ, J)ċ = (a + bt)R(ċ, ċ)ċ = 0.
dt
Such a J is the variational vector field of the geodesic variation

c(s, t) = c(t + s(a + bt)) = c (1 + sb)t + sa .
This is a variation of c which is obtained by simply reparametrizing the geodesic. It

contains no geometric information. Therefore such a Jacobi field is uninteresting. Thus
there is a two-dimensional space of uninteresting Jacobi fields.
99
3 Curvature
Remark 3.4.6. If a Jacobi field J : I → T M satisfies:

∇
J(t0 ) ⊥ ċ(t0 ) and J(t0 ) ⊥ ċ(t0 ) for a t0 ∈ I,
dt
then we have
∇
J(t) ⊥ ċ(t) and J(t) ⊥ ċ(t) for all t ∈ I.
dt
Namely,
2
d ∇ ∇ ∇ ∇
J, ċ = J, ċ + J, ċ = hR(ċ, J)ċ, ċi = 0
dt dt dt dt |{z}
dt
=0
∇
implies dt J, ċ ≡ 0 and from

d ∇
hJ, ċi = J, ċ ≡ 0
dt dt
we see that hJ, ċi ≡ 0.
Consequence. Let c be non light-like. In this case we have Tc(t) M = Rċ(t)⊕ ċ(t)⊥ . Then
∇
{Jacobi fields along c} = R · ċ ⊕ R · t ċ ⊕ {Jacobi fields J along c | J ⊥ ċ, dt J ⊥ ċ} .
| {z } | {z }
uninteresting interesting Jacobi fields
Jacobi fields
Remark 3.4.8. For light-like geodesics c this is not true because ċ ⊥ ċ.
Example 3.4.9
c
Let (M, g) = (R2 , gMink ), let c be a light-like geodesic and
let ξ be a light-like parallel vector field along c which is
linearly independent of ċ. ξ
Since ξ is parallel and R = 0, the vector field ξ is also a C
Jacobi field and we have:
{Jacobi field along c} = R · ċ ⊕ R(tċ) ⊕ R ξ ⊕ R(tξ)
| {z }
∇
= Jacobi field J along c | J ⊥ ċ, dt J ⊥ ċ
100
3.4 Jacobi fields
Definition 3.4.10. For κ ∈ R the generalized sine and cosine function sκ , cκ : R → R

are defined by
 √  √
√1 sin( κ · r), κ>0
cos( κ · r), κ>0

 κ
 
sκ (r) := r, κ = 0 and cκ (r) := 1, κ=0
 1
p  p
 √ sinh( |κ| · r), κ < 0
 
cosh( |κ| · r), κ < 0
|κ|
respectively.
It is easy to check that
κ s2κ + c2κ = 1,
s′κ = cκ and sκ (0) = 0,
c′κ = −κ sκ and cκ (0) = 1.
Example 3.4.11. Let (M, g) be a Riemannian manifold with constant sectional curvature
K ≡ κ. Let c be a geodesic, parametrized by arc-length. Let ξ be a parallel vector field
along c with ξ ⊥ ċ. Set

J(t) := a sκ (t) + b cκ (t) ξ(t) with a, b ∈ R.
Then
2
∇
J = (a s̈κ + b c̈κ )ξ = −κ(a sκ + b cκ )ξ = −κJ.
dt
For the curvature tensor we here have R(ξ, η)ζ = κ(hη, ζi ξ − hξ, ζi η). Thus

R(ċ, J)ċ = (a sκ + b cκ ) · κ hξ, ċi ċ − hċ, ċi ξ = −κ a sκ + b cκ ξ = −κJ.
| {z } | {z }
=0 =1
Hence J is a Jacobi field and
n ∇ o
Jacobi fields along c | J ⊥ ċ, J ⊥ ċ
dt
= (a sκ + b cκ ) ξ | a, b ∈ R, ξ parallel along c, ξ ⊥ ċ .
101
3 Curvature
κ>0 J
b b b b
κ=0
J
J
b b
κ<0
Proposition 3.4.12
Let M be a semi-Riemannian manifold and c : [a, b] → M a geodesic. Let ξ be a smooth
vector field along c. Then
ξ is a Jacobi field ⇐⇒ ξ is the variational field of a geodesic variation.
Proof. The implication “⇐” is already known. We show “⇒”.

ξ(t0 )
Let ξ be a Jacobi field along c. Choose a t0 ∈ [a, b]
and choose a smooth curve γ : (−ε, ε) → M with ξ b
γ(0) = c(t0 ) and γ̇(0) = ξ(t0 ). Let η1 be the c(t0 )
parallel vector field along γ with η1 (0) = ċ(t0 ). c
Let η2 be the parallel vector field along γ with γ η1
∇
η2 (0) = dt ξ(t0 ).
η
Set η(s) := η1 (s) + sη2 (s) and

c(s, t) := expγ(s) (t − t0 )η(s) .
c
Since the domain of definition of exp is open, γ
c(s, t) is defined for |s| sufficiently small and for
all t ∈ [a, b]. Then we have
102
3.4 Jacobi fields

c(0, t) = expγ(0) (t − t0 )η(0) = expc(t0 ) (t − t0 )ċ(t0 ) = c(t)
∂c
Hence c(s, t) is a geodesic variation of c(t). Let J(t) := ∂s (0, t) be the corresponding
variational field. Then J is a Jacobi field. We show:
∇ ∇
ξ(t0 ) = J(t0 ) and ξ(t0 ) = J(t0 ).
dt dt
Then we get ξ = J because Jacobi fields are uniquely determined by their initial data
and hence ξ is the variational field of the geodesic variation c(s, t).
We calculate
∂c d d
J(t0 ) = (0, t0 ) = expγ(s) (0) = γ(s) = γ̇(0) = ξ(t0 )
∂s ds s=0 ds s=0
and
∇ ∇ ∂c ∇ ∂c ∇ ∇
J(t0 ) = (0, t0 ) = (0, t0 ) = η(0) = η2 (0) = ξ(t0 ).
dt ∂t ∂s ∂s ∂t ds dt
We are now able to generalize Lemma 2.6.26 and can identify the differential of the
exponential at arbitrary points in its domain.
Proposition 3.4.13
Let M be a semi-Riemannian manifold, p ∈ M and ξ ∈ Tp M . We assume that the geodesic
γ(t) := expp (tξ) is defined on [0, 1], i.e., ξ lies in the domain of expp .
For η ∈ Tp M (∼ ∇
= Ttξ Tp M ) let J be the Jacobi field along γ with J(0) = 0 and dt J(0) = η.
Then we have for all t ∈ (0, 1]:
J(t)
d expp |tξ (η) = .
t
Proof. Consider the geodesic variation

c(s, t) := expp t(ξ + sη) .
∂c
Let ζ := ∂s |s=0 be the corresponding variational Jacobi field. Then we have
∂c d
ζ(0) = (0, 0) = expp |s=0 (0) = 0 = J(0)
∂s ds
and
103
3 Curvature
Tp M
0
b
η
ξ
expp
p ξ
M
∇ ∇ ∂c ∇ ∂c ∇ ∇
ζ(0) = (0, 0) = (0, 0) = (ξ + sη)|s=0 = η = J(0)
dt dt ∂s ∂s ∂t ds dt
Hence ζ = J. Now we compute for fixed t ∈ (0, 1]:
∂ ∂ s 1 1
d expp |tξ (η) = expp (tξ + s η)|s=0 = expp t ξ + η = ζ(t) = J(t).
∂s ∂s t s=0 t t
Corollary 3.4.14
Let M be a semi-Riemannian manifold, let p ∈ M and let ξ be in the domain of expp . Then

ker(d expp |ξ ) ∼
= Jacobi field along γ(t) = expp (tξ) | J(0) = 0, J(1) = 0 .
Definition 3.4.15. Let M be a semi-Riemannian manifold and γ : I → M a geodesic.

Then t1 , t2 ∈ I, t1 6= t2 are called conjugated points along γ, if there exists a non-
trivial Jacobi field J along γ with J(t1 ) = 0 and J(t2 ) = 0.
Consequence. d expp |ξ is non-invertible if and only if 0 and 1 are conjugated points

along γ(t) = expp (tξ).
104
3.4 Jacobi fields
Example 3.4.17. Let M be a Riemannian manifold with constant sectional curvature

K ≡ κ.
Case 1: κ ≤ 0 . Every Jacobi field has at most one zero.

⇒ There are no conjugated points.
⇒ d expp |ξ is invertible for all ξ ∈ Dp .
⇒ The map expp : Dp → M is a local diffeomorphism.
Case 2: κ > 0 .
For a geodesic parametrized by arc-length, the ker d expp |ξ
conjugate points belonging to t0 are the points pb
ξ
t0 + m √πκ for m ∈ Z \ {0}. Considering the case ξ′
m = 1 we have Tp S n

expp {ξ ∈ Tp S n | ||ξ|| = π} = {−p}.
−pb
Sn
For ||ξ|| = π we obtain
ker d expp |ξ = ξ ⊥ .
Proposition 3.4.18
Let M be a semi-Riemannian manifold and let c : [t0 , t1 ] → M be a geodesic. Let t0 and t1
be not conjugated with each other along c.
Then for ξ ∈ Tc(t0 ) M and η ∈ Tc(t1 ) M there exists exactly one Jacobi field J along c with
J(t0 ) = ξ and J(t1 ) = η.
ξ
J
c(t1 )
b b b
c(t0 )
η
105
3 Curvature
Proof. The linear map
2n-dimensional (n + n)-dimensional
z }| { z }| {
{Jacobi field along c} → Tc(t0 ) M ⊕ Tc(t1 ) M ,

J 7→ J(t0 ), J(t1 ) ,
is injective since t0 and t1 are not conjugated to each other along c. For dimensional
reasons, this map is an isomorphism.
Proposition 3.4.18 means that in the non-conjugate case we can also characterize Jacobi
fields by the boundary values J(t0 ) and J(t1 ) instead of the initial values J(t0 ) and
∇
dt J(t0 ). In the conjugate case this is certainly wrong.
Example 3.4.19. Let c be a geodesic emanating b

p = c(t0 )
from p ∈ S n which is parametrized by arc-length.
The set of η ∈ T−p S n for which exists a Jacobi field
J along c with J(0) = 0 and J(π) = η is given by c
{η = α · ċ(π) | α ∈ R}.
b −p = c(t1 )
106
4 Submanifolds
4.1 Submanifold of differentiable manifolds
Definition 4.1.1. Let M be an m-dimensional differentiable manifold. A subset

N ⊂ M is called an n-dimensional submanifold if for every p ∈ N there exists a
chart x : U → V of M with p ∈ U such that
x(N ∩ U ) = V ∩ (Rn × {0}).
x {0} × Rm−n
N V ⊂ Rn
U
x(p)
p b
b
Rn × {0}
M
Such a chart is called submanifold chart of N . The number m − n is called codimen-

sion of N in M .
Example 4.1.2
(1) Codimension n = 0: A subset N ⊂ M is a submanifold of codimension 0 if and
only if N is open subset of M .
(2) Dimension n = 0: A subset N ⊂ M is a submanifold of dimension 0 if and only if

N is a discrete subset of M .
(3) Affine subspaces: Let N ⊂ M = Rm be an affine subspace, i.e., N is of the form

N = N ′ + p, where N ′ ⊂ Rm is an n-dimensional vector subspace and p ∈ Rm fixed.
Choose A ∈ GL(m) with AN ′ = Rn × {0}. Then x : U = Rm → V = Rm , given by
x(q) := A(q − p),
is a submanifold chart.
107
4 Submanifolds
(4) Graphs: Let M1 and M2 be differentiable manifolds and let f : M1 → M2 be a

smooth map. Set M = M1 × M2 and

N = Γf = (ξ, η) ∈ M1 × M2 | η = f (ξ) .
U1 × U2 p Γx2 ◦f ◦x1 −1
b
U2 x 1 × x 2 b
Ψ b
M2
M1 U1 V1 × V2
Choose charts xi : Ui → Vi of Mi with p ∈ U1 × U2 . W.l.o.g. let f (U1 ) ⊂ U2 . For

w ∈ V1 and z ∈ V2 set

ψ(w, z) := w, z − x2 ◦ f ◦ x1 −1 (w) .
Then x := ψ ◦ (x1 × x2 ) is a submanifold chart, defined on U1 × U2 .
Theorem 4.1.3
Let M be an m-dimensional differentiable manifold. Let N ⊂ M be a subset. Then the
following assertions are equivalent:
(i) N is an n-dimensional submanifold.
(ii) For every p ∈ N there exists an open neighborhood U of p and smooth functions
f1 , . . . , fm−n : U → R such that
(a) N ∩ U = {q ∈ U | f1 (q) = . . . = fm−n (q) = 0};
(b) The differentials df1 |p , . . . , dfm−n |p ∈ Tp∗ M are linearly independent.
(iii) For every p ∈ N there exists an open neighborhood U of p, an (m − n)-dimensional

differentiable manifold R and a smooth map f : U → R with
(a) N ∩ U = f −1 (q) where q = f (p);
(b) df |p : Tp M → Tq R has maximal rank.
Proof. “(i)⇒(ii)”: Let p ∈ N and let x : U → V be a submanifold chart for N with p ∈ U .

W.l.o.g. let
108
(1) x(p) = 0 ∈ Rm (otherwise compose x with a suitable translation);
(2) V = V1 × V2 where V1 ⊂ Rn and V2 ⊂ Rm−n (otherwise make U smaller).
Now fj : U → R, fj := xn+j , do the job (j = 1, . . . , m − n).

“(ii)⇒(iii)” is obvious. Simply set R := Rm−n and f := (f1 , . . . , fm−n ).
“(iii)⇒(i)”: R
N U f
p b b
q
Ũ
M ϕ̃
ϕ
ϕ(p)
b
Ṽ b
ϕ̃(q)
ϕ̃ ◦ f ◦ ϕ−1
V
Choose a chart ϕ : U → V of M around p and a chart ϕ̃ : Ũ → Ṽ of R around q := f (p).
W.l.o.g. we assume that f (U ) ⊂ Ũ . Since ϕ and ϕ̃ are diffeomorphisms, we have
rank D(ϕ̃ ◦ f ◦ ϕ−1 )|ϕ(p) = rank df |p = m − n.
The implicit function theorem yields: One can shrink V and U to smaller neighbor-
hoods of q and p, respectively, such that V = V1 × V2 and one can find a smooth map
g : V1 → V2 such that
(ϕ̃ ◦ f ◦ ϕ−1 )−1 (ϕ̃(q)) = (f ◦ ϕ−1 )−1 (q) = Γg .
If we compose ϕ with a submanifold chart for graphs as in the Example 4.1.2 (4) then
we get a submanifold chart for N in M around p.
Definition 4.1.4. Let M and R be differentiable manifolds and let f : M → R be

smooth. A point p ∈ M is called a regular point of f if df |p has maximal rank.
Otherwise p is called a critical point of f .
A point q ∈ R is called a regular value of f if all p ∈ f −1 (q) are regular points.
Otherwise q is called a critical value of f .
109
4 Submanifolds
Example 4.1.5. Let M = R = R and f (t) = t2 . We R

have
critical
df |t (ξ) = f ′ (t) · ξ. point critical
value
Hence t is a critical point of f if and only if f ′ (t) = M b b
0
regular values
0. In this example t = 0 is the only critical point
regular points Punkte
and f (0) = 0 is the only critical value.
R
Example 4.1.6. Let M = R = R and
2
f (t) = t2 (t − 1). Here f has the critical M 0b
3
b b
points t = 0 and t = 32 . The critical values b
4
are f (0) = 0 and f ( 32 ) = − 27 .
Example 4.1.7. Let M = R = R and

f (t) = 0. In this case all t ∈ R are critical M b
points but 0 is the only critical value.
The examples indicate that there may be many critical points but there are always only
few critical values. This is true in general:
Theorem 4.1.8 (Sard)

Let M and R be differentiable manifolds and let f : M → R be smooth. Then the set of
critical values of f is a null set in R. In other words, for every chart x : U → V of R the set
x(U ∩ {critical values of f }) ⊂ V is a null set (in the sense of Lebesgue measure theory).
For a proof see [M65, Chapter 3].
Corollary 4.1.9
If f : M → R is smooth and if q ∈ R is a regular value of f , then N = f −1 (q) is empty or a
submanifold of M with codim(N ) = dim(R).
110
Proof. This follows directly from Criterion (iii) in Theorem 4.1.3.
Example 4.1.10. Let M = Rn+1 and R = R. Let f : Rn+1 → R be given by f (x) =

||x||2 = (x0 )2 + . . . + (xn )2 . Then S n = f −1 (1) and for any x ∈ Rn+1 we have
Df |x = (2x0 , . . . , 2xn ).

1, x 6= 0
⇒ rank(Df |x ) =
0, x = 0
⇒ For all x ∈ f −1 (1) we have rank(Df |x ) = 1.

⇒ 1 is a regular value of f .
4.1.9
⇒ S n ⊂ Rn+1 is a submanifold of codimension 1.
Remark 4.1.11. In this example all q ∈ R \ {0} are regular values. We have
n √
−1 S ( q), q > 0
f (q) =
∅ ,q<0
For the critical value q = 0 we have that f −1 (0) = {0} is also (by coincidence) a subma-
nifold, but of the wrong codimension n + 1. In general, the preimage of a critical value
is not a submanifold.
Remark 4.1.12. Sometimes the set f −1 (q) is a submanifold with codimension dim R
even if q is a critical value.
Example 4.1.13. Let g : Rn+1 → R, g(x) = (||x||2 − 1)2 . Then 0 is critical value of g but
S n = g−1 (0) is a submanifold of codimension 1.
Remark 4.1.14. Submanifolds of differentiable manifolds are themselves differentiable

manifolds. Namely:
Let N ⊂ M be a submanifold and p ∈ N and x : U → V a submanifold chart with
x = (x1 , . . . , xn , xn+1 , . . . , xm ), then
(x1 , . . . , xn ) : U ∩ N → V ∩ Rn
is a chart of N . The set of charts of N obtained in this manner by restricting submanifold

charts to N is a C ∞ -atlas for N .
111
4 Submanifolds
Theorem 4.1.15
Let N ⊂ M be a submanifold. Let ι : N ֒→ M be the inclusion map, ι(p) = p. Then we
have:
(i) ι is smooth and dι|p : Tp N → Tp M is injective.
(ii) If f : M → P is smooth then f |N : N → P is also smooth.
(iii) If g : Q → M is smooth with g(Q) ⊂ N then g : Q → N is also smooth.
Proof. (i) Let x = (x1 , . . . , xm ) be a submanifold chart of N in M and x̃ = (x1 , . . . , xn )

the corresponding chart of N . The following diagram commutes:
ι
N ⊃ U ∩N U ⊂M
x̃ x
ξ7→(ξ,0)
V ∩ Rn V
Obviously, ξ 7→ (ξ, 0) is smooth. Since this map is linear, it coincides with its
differential, such that the differential is in particular injective.
(ii) The function f |N = f ◦ ι is the composition of two smooth maps and therefore
again smooth.
(iii) Let q ∈ Q and x = (x1 , . . . , xm ) be a submanifold chart of M around g(q). Since

g is smooth the functions g i := xi ◦ g are also smooth. From g(Q) ⊂ N we see
that (g 1 , . . . , gm ) = (g 1 , . . . , gn , 0, . . . , 0). Now (g1 , . . . , g n ) is smooth and thus also
g : Q → N.
Remark 4.1.16. One identifies Tp N with dι|p (Tp N ) and thinks of it as a vector subspace
of Tp M .
Tp M
b Tp N
p
N
M
112
4.2 Semi-Riemannian submanifolds
Remark 4.1.17. If M = Rm , i.e., N ⊂ Rm , then one often considers Tp N as a vector

∼
=
subspace of Rm via Tp N ⊂ Tp Rm −→
canon.
Rm .
isom.
Example 4.1.18. For N = S n ⊂ Rn+1 we have Tp S n = p⊥ .
Definition 4.2.1. Let (M̄ , ḡ) be a semi-Riemannian manifold. A submanifold M ⊂ M̄

is called a semi-Riemannian submanifold, if for all p ∈ M
(ḡ|p ) Tp M ×Tp M
=: g|p
is non-degenerate.
Example 4.2.2. If (M̄ , ḡ) is Riemannian, then every submanifold is a semi-Riemannian

submanifold.
Example 4.2.3. Let (M̄ , ḡ) = (R2 , gMink ) be 2-dimensional Minkowski space, i.e.,
gMink = −dx0 ⊗ dx0 + dx1 ⊗ dx1 . Then
N1 = {(x0 , 0) | x0 ∈ R} is semi-Riemannian (with negative-definite metric).
N2 = {(0, x1 ) | x1 ∈ R} is semi-Riemannian (with positive-definite metric).
N3 = {(t, t) | t ∈ R} is not semi-Riemannian, since the restriction of gMink on Tp N3

vanishes.
N4 = S 1 has 4 points at which the restriction of gMink degenerates.
x0 M̄ + M̄
N4 b b
−
N2 b
x1 b
− b b
+
N3 N1
113
4 Submanifolds
Definition 4.2.4. Let M ⊂ M̄ be a semi-Riemannian submanifold. Then we call

Np M := Tp M ⊥ = ξ ∈ Tp M̄ | ḡ|p (ξ, η) = 0 ∀ η ∈ Tp M
the normal space of M at the point p.
Tp M̄
b Tp M
p
Np M
M
M̄
Remark 4.2.5. We have Tp M̄ = Tp M ⊕ Np M since
(1) dim Np M ≥ dim Tp M̄ − dim Tp M .
(2) If there existed a ξ ∈ Tp M ∩ Np M with ξ 6= 0, then we would have ξ ∈ Tp M with

ḡ|P (ξ, η) = 0 for all η ∈ Tp M . Then (ḡ|p )|Tp M ×Tp M would be degenerate, which is a
contradiction.
Let M ⊂ M̄ be a semi-Riemannian submanifold and p ∈ M . Let
tan :Tp M̄ → Tp M,
nor :Tp M̄ → Np M,
be the orthogonal projections. Both M and also M̄ have, when seen as semi-Riemannian
¯ respectively. Now we
manifolds in their own rights, a Levi-Civita connection ∇ and ∇,
want to investigate, how we can determine ∇ directly from ∇.¯
(M̄ , ḡ) ¯ Levi-Civita connection

∇
?
(M, g) ∇ Levi-Civita connection
Here g|p := (ḡ|p )Tp M ×Tp M .
114
Let p ∈ M , ξ ∈ Tp M and η ∈ C ∞ (U, T M ), where

U ⊂ M is an open neighborhood of p. Choose a
M U
smooth extension η̄ of η to an open neighborhood
¯ ξ η̄ ∈ Tp M̄ does not depend
Ū of p in M̄ . Then ∇ Ū
η̄
on the choice of continuation η̄.
Namely: the tanget vector ξ ∈ Tp M is of the form M̄
¯ ξ η̄ η
ξ = ċ(0) with a curve c : (−ε, ε) → M . Hence ∇
depends on η̄ only along c, that is, only on η.
We can also write:
¯ ξ η := ∇
∇ ¯ ξ η̄.
Example 4.2.6 η
Let M̄ = (R2 , geucl ) and M = S1. Set
η(x , x ) = (−x , x ). For c : R → S 1 with c(t) =
1 2 2 1
(cos(t), sin(t)) we have b
ċ(t) = η(c(t)).
¯ ηη
∇
Then we get:
¯
¯ η η = ∇ ċ = c̈ = − cos(t), − sin(t)

∇
dt
which is not tangent to S 1 .
¯ ξ η).
We set ∇ξ η := tan(∇
Theorem 4.2.7
Let (M̄ , ḡ) be a semi-Riemannian manifold and M ⊂ M̄ a semi-Riemannian submanifold
with induced semi-Riemannian metric g. Let ∇ ¯ be the Levi-Civita connection of (M̄ , ḡ).
Then
∇ξ η = tan ∇¯ ξη
is the Levi-Civita connection of (M, g).
115
4 Submanifolds
Proof. We check that ∇ satisfies the axioms of the Levi-Civita connection for (M, g). By
the uniqueness statement in Theorem 2.3.8, ∇ must then be the Levi-Civita connection
of (M, g).
¯ is local.
(i) Locality is clear because ∇
¯ is linear in ξ.
(ii) Linearity in ξ is clear because tan is linear and ∇
(iii) Linearity in η is clear by a similar argument.
(iv) Product rule I: Let f ∈ C ∞ (U ) and η ∈ C ∞ (U, T M ), where U ⊂ M is an open

neighborhood of p and ξ ∈ Tp M . Let η̄ and f¯ be smooth extensions of η and f to
an open neighborhood Ū of p in M̄ . Then

∇ξ (f · η) = tan ∇ ¯ ξ f¯ · η̄

= tan ∂ξ f¯ · η̄|p + f¯(p) · ∇
¯ ξ η̄

= tan ∂ξ f · η̄|p + f (p) · ∇¯ ξ η̄

= ∂ξ f · tan η̄|p + f (p) · tan ∇ ¯ ξ η̄
= ∂ξ f · η|p + f (p)∇ξ η.
(v) Product rule II: Let ξ ∈ Tp M and η1 , η2 ∈ C ∞ (U, T M ). Choose smooth extensions
η̄1 , η̄2 ∈ C ∞ (Ū , T M̄ ). Then

∂ξ g(η1 , η2 ) = ∂ξ ḡ η̄1 , η̄2

= ḡ|p ∇¯ ξ η̄1 , η̄2 |p + ḡ|p η̄1 |p , ∇ ¯ ξ η̄2
|{z}
=η2 |p , in particular tangent to M

= g|p tan ∇ ¯ ξ η̄1 , η2 |p + g|p η1 |p , tan ∇ ¯ ξ η̄2

= g|p ∇ξ η1 , η2 |p + g|p η1 |p , ∇ξ η2 .
(vi) Freeness of torsion: Let x1 , . . . , xm , xm+1 , . . . , xm̄ be submanifold coordinates on M̄ .

Here x1 , . . . , xm are coordinates on M . For 1 ≤ i, j ≤ m:
! !
∂ ¯ ∂ ∂ ¯ ∂ ∂ ∂
∇ ∂ j
= tan ∇ j
= tan ∇ i
=∇ ∂ i
.
∂xi p ∂x ∂xi p ∂x ∂xj p ∂x ∂xj p ∂x
Example 4.2.8. Let M = S 1 ⊂ M̄ = R2 with ḡ = geucl . Set η(c(t)) = ċ(t) where

c(t) = (cos(t), sin(t)). Then

∇η η = tan ∇¯ ηη = tan(−p) = 0.
p
Hence c is a geodesic in S 1 (but not in R2 ).
116
Lemma 4.2.9
Let ξ ∈ Tp M and η ∈ C ∞ (U, T M ), where U ⊂ M is an open neighborhood of p. Then
nor(∇¯ ξ η) ∈ Np M only depends η via η|p .
Proof. Let x1 , . . . , xm , xm+1 , . . . , xm̄ be submanifold coordinates on M̄ around p. Let

Γkij : U → R be the Christoffel symbols of ∇, 1 ≤ i, j, k ≤ m, and Γ̄kij : U → R be the
Christoffel symbols of ∇, ¯ 1 ≤ i, j, k ≤ m̄. On U we write η = Pm η j ∂ j and we define
j=1 ∂x
on Ū : (
η j (x1 , . . . , xm ) for j = 1, . . . , m
η̄ j (x1 , . . . , xm̄ ) := .
0 for j = m + 1, . . . , m̄
P Pm i ∂
Set η̄ := m̄ j ∂
j=1 η̄ ∂xj . Furthermore, write ξ = i=1 ξ ∂xi p . Then we have:
¯ ξ η) = nor(∇
nor(∇ ¯ ξ η̄)
¯ ξ η̄ − ∇ξ η
= ∇
 
m m̄ k m̄
X X ∂ η̄ X ∂
= ξi 
i
+ Γ̄kij |x(p) · η̄ j |x(p) 
∂x x(p) ∂xk p
i=1 k=1 j=1
 
m m k m
 ∂η ∂
X X X
− ξi i
+ Γkij |x(p) · η j |x(p) 
∂x x(p) ∂xk p
i=1 k=1 j=1
m m m̄ m
!
X X X ∂ X ∂
= ξi η j |x(p) Γ̄kij |x(p) − Γkij |x(p) .
∂xk p ∂xk p
i=1 j=1 k=1 k=1
This only depends on η j |x(p) , i.e., only on η|p .
Definition 4.2.10. The map II : Tp M × Tp M → Np M , given by

¯ ξη ,
II(ξ, η) = nor ∇
is called the second fundamental form of M in M̄ (at the point p ∈ M ).
117
4 Submanifolds
Lemma 4.2.11
The second fundamental form II is bilinear and symmetric.
Proof. In the previous proof we have shown that

m m̄ m
!
X X ∂ X ∂
II(ξ, η) = Γ̄kij − Γkij ξ i ηj .
∂xk p ∂xk p
i,j=1 k=1 k=1
Clearly, II is bilinear. By the symmetry of the Christoffel symbols in the lower indices,
II is also symmetric.
Example 4.2.12. Let M = S 1 ⊂ M̄ = R2 and η as in Example 4.2.8. The second funda-

mental form is then given by II(η, η) = −p.
Conclusion. The equation

¯ ξ η = ∇ξ η + II(ξ, η|p )
∇
¯ ξ η into its tangential and normal parts.
is the splitting of ∇
Notation 4.2.13. For better readability we will from now on write hξ, ηi instead of
g(ξ, η) or ḡ(ξ, η).
Since one can compute the Levi-Civita connection ∇ of the submanifold M from the
Levi-Civita connection ∇¯ of M̄ , one should also be able to compute the curvature tensor
R of M from that of M̄ . Indeed this is possible.
Theorem 4.2.14 (Gauß Formula)

Let M ⊂ M̄ be a semi-Riemannian submanifold and p ∈ M . Let ξ, η, ζ, ν ∈ Tp M . Then we
have
hR(ζ, ν)ξ, ηi = R̄(ζ, ν)ξ, η + hII(ν, ξ), II(ζ, η)i − hII(ζ, ξ), II(ν, η)i .
118
Proof. Let x1 , . . . , xm be coordinates of M around p coming from a submanifold chart

x1 , . . . , xm , xm+1 , . . . , xm̄ . By multilinearity, it suffices to show the assertion for ξ =
∂
∂xi p
, η = ∂x∂ j p , ζ = ∂x∂ k p and ν = ∂x ∂
l p . We have
* +
¯ ζ∇
¯ ∂ ∂ ¯¯ ∂ ¯ ¯ ∂ ¯¯ ∂
R̄(ζ, ν)ξ, η = ∇ i
−∇ ∇ ∂ | ∂ l ∂xi − ∇ν ∇ ∂ k ∂xi + ∇∇ ∂
i
,η
∂xl ∂x p ∂xk ∂x
∂ |
k p ∂x ∂x
∂x ∂xl

¯ ¯ ∂ ¯ ¯ ∂
torsion freeness = ∇ζ ∇ ∂ − ∇ν ∇ ∂ ,η
∂xl ∂x
i ∂xk ∂x
i

¯ ∂ ¯ ∂ ∂ ¯ ∂ ¯ ∂ ∂
= ∇ζ ∇ ∂ i
+ ∇ζ II , − ∇ν ∇ ∂ − ∇ν II , ,η
∂xl ∂x ∂xl ∂xi ∂xk ∂x
i ∂xk ∂xi

∂ ∂
= ∇ζ ∇ ∂ i
− ∇ν ∇ ∂ i
,η
∂xl ∂x ∂xk ∂x
≡0
z
}| {
∂ ∂ ∂ ∂ ∂ ¯ ∂
+ ∂ζ II , , j − II , , ∇ζ j
∂xl ∂xi ∂x ∂xl ∂xi ∂x
≡0
z
}| {
∂ ∂ ∂ ∂ ∂ ¯ ∂
− ∂ν II , , j + II , , ∇ν j
∂xk ∂xi ∂x ∂xk ∂xi ∂x
= hR(ζ, ν)ξ, ηi + hII(ζ, ξ), II(ν, η)i − hII(ν, ξ), II(ζ, η)i .
Corollary 4.2.15
If E ⊂ Tp M is a non-degenerate plane with basis ξ, η, then we have
hII(ξ, ξ), II(η, η)i − hII(ξ, η), II(ξ, η)i

K(E) = K̄(E) + .
hξ, ξi hη, ηi − hξ, ηi2
Proof. This follows directly from the definition of sectional curvature and the Gauß
formula.
Lemma 4.2.16
Let M ⊂ M̄ be a semi-Riemannian submanifold. Let ϕ : M̄ → N̄ be a local isometry. Set
119
4 Submanifolds
ϕ(M ) =: N . For ξ, η ∈ Tp M we have

IIN dϕ|p (ξ), dϕ|p (η) = dϕ|p IIM (ξ, η) .
¯ Since II is the difference of ∇ and ∇

Proof. Local isometries preserve ∇ and ∇. ¯ we get
the assertion.
4.3 Totally geodesic submanifolds

Let M ⊂ M̄ be a semi-Riemannian submanifold and c : I → M a smooth curve. Let ξ
be a smooth vector field at M along c. Then the splitting in tangential and normal parts
of the covariant derivative is given by
¯
∇ ∇
ξ= ξ + II(ξ, ċ).
dt dt
In particular, we have for ξ = ċ
¯
∇ ∇
ċ = ċ + II(ċ, ċ).
dt dt
Therefore the curve c is a geodesic in M if and only if
¯
∇ ¯
∇
ċ = II(ċ, ċ), i.e., if ċ(t) ∈ Nc(t) M for all t ∈ I.
dt dt
Example 4.3.1. Let M = S n ⊂ M̄ = Rn+1 with Euclidean metric. Let c : I → S n be a

great circle,
c(t) = cos(t) · p + sin(t) · ξ.
with p ∈ S n , ξ ∈ Tp S n ⊂ Rn+1 and kξk = 1. From this we get
¯
∇
ċ(t) = c̈(t) = − cos(t) · p − sin(t) · ξ = −c(t) ∈ Nc(t) S n .
dt
Hence c is a geodesic in S n .
120
4.3 Totally geodesic submanifolds
Definition 4.3.2. A semi-Riemannian submanifold is called totallygeodesic if II ≡ 0.
Theorem 4.3.3
For a semi-Riemannian submanifold M ⊂ M̄ the following statements are equivalent:
(i) M ist totally geodesic.
(ii) Every geodesic in M is also a geodesic in M̄ .
(iii) For any p ∈ M and ξ ∈ Tp M there exists an ε > 0 such that the M̄ -geodesic
c : (−ε, ε) → M̄ with c(0) = p and ċ(0) = ξ lies in M , i.e., c(t) ∈ M for all
t ∈ (−ε, ε).
(iv) Let c : I → M be a smooth curve. Then the parallel transport along c w.r.t. M and
w.r.t. M̄ coincide (for tangent vectors of M ).
Proof. “(ii)⇒(iii)”: Let p ∈ M and ξ ∈ Tp M . Let c be the M̄ -geodesic with c(0) = p and
ċ(0) = ξ. Let c̃ be the M -geodesic with c̃(0) = p and c̃(0)˙ = ξ. By (ii), c̃ is also geodesic
˙
in M̄ . Since we have c̃(0) = c(0) and c̃(0) = ċ(0), the two M̄ -geodesics must coincide,
c = c̃ on (−ε, ε) for a ε > 0. In particular, c lies in M .
“(iii)⇒(i)”: Let p ∈ M and ξ ∈ Tp M . Let cξ be the M̄ -geodesic with cξ (0) = p and
ċξ (0) = ξ. By (iii), cξ lies in M for t ∈ (−ε, ε) with suitable ε > 0. On (−ε, ε) we get:
¯
∇ ∇
0= ċξ = ċξ + II(ċξ , ċξ )
dt dt
|{z} | {z }
normal
tangential
In particular, we have

II ċξ (t), ċξ (t) = 0 for all t ∈ (−ε, ε).
For t = 0 this means that II(ξ, ξ) = 0. Since ξ is arbitrary, polarization yields II ≡ 0.

∇ ¯
∇
“(i)⇒(iv)”: We have dt ξ = dt ξ. Hence ξ is parallel in M if and only if ξ is parallel in M̄ .
“(iv)⇒(ii)”: Let c be a geodesic in M .
⇒ ċ is parallel in M .
(iv)
⇒ ċ is parallel in M̄ .
121
4 Submanifolds
⇒ c ist geodesic in M̄ .
Example 4.3.4. Let M ⊂ M̄ = Rn be an affine subspace where Rn is equipped with geucl

or gMink . Criterion (iii) shows that M ⊂ Rn is totally geodesic.
Example 4.3.5. Let M̄ be an arbitrary semi-Riemannian manifold.
(1) All 0-dimensional submanifolds are totally geodesic.
(2) Every submanifold of codimension 0, i.e., every open subset of M̄ , is totally

geodesic.
(3) Let M = {c(t)|t ∈ I}, where c : I → M̄ is a geodesic. If M is a semi-Riemannian

submanifold (has no self-intersection, for instance), then M is totally geodesic.
Remark 4.3.6. Most semi-Riemannian manifolds M̄ do not have totally geodesic sub-
manifolds of dimension m ∈ {2, . . . , m̄ − 1}.
Theorem 4.3.7
Let M ⊂ M̄ be a semi-Riemannian submanifold. Assume that there exists an isometry
ϕ ∈ Isom(M̄ ), such that M is a connected component of
Fix(ϕ) = {p ∈ M̄ | ϕ(p) = p}.
Then M is totally geodesic.
Proof. We check Criterion (iii) in Theorem 4.3.3. Let p ∈ M and ξ ∈ Tp M . We first show
that
dϕ|p (ξ) = ξ.
Namely, let γ : J → M be a smooth curve with γ(0) = p and γ̇(0) = ξ. Then
d
dϕ|p (ξ) = dϕ|p (γ̇(0)) = (ϕ ◦ γ) |t=0 = γ̇(0) = ξ. ✓
dt | {z }
= γ , since
M ⊂Fix(ϕ)
By Proposition 2.6.20, c lies in Fix(ϕ). Since c(I) is connected, c remains in M .
122
4.4 Hypersurfaces
Example 4.3.8. Let M̄ = S n . Let W ⊂ Rn+1 be a

subvector space. Let A ∈ O(n + 1) be the reflection
about W . Sn
⇒ ϕ := A|S n ∈ Isom(S n )
⇒ Fix(ϕ) = W ∩ S n is totally geodesic
Hence all “great subspheres” in S n are totally W

geodesic submanifolds. In particular, S n admits
totally geodesic submanifolds of every codimen-
sion.
The Gauß Formula (Theorem 4.2.14) tells us that if M ⊂ M̄ is totally geodesic, then
R(ξ, η)ζ = R̄(ξ, η)ζ for all p ∈ M and ξ, η, ζ ∈ Tp M,

K(E) = K̄(E) for all non-degenerate planes E ⊂ Tp M.
4.4 Hypersurfaces
Definition 4.4.1. A semi-Riemannian submanifold M ⊂ M̄ is called a semi-

Riemannian hypersurface if codimM = 1.
The signature of M is ε = +1 if (ḡ|p )|Np M ×Np M is positive definite, and ε = −1 if
(ḡ|p )|Np M ×Np M is negative definite.
Remark 4.4.2. For ε = +1 we have Index(M̄ , ḡ) = Index(M, g) while for ε = −1 we get
Index(M̄ , ḡ) = Index(M, g) + 1.
Notation 4.4.3. For ξ ∈ Tp M we write

p
|ξ| := | hξ, ξi |.
123
4 Submanifolds
Caution! This does not define a norm unless h·, ·i is definite. In particular, it can occur
that |ξ| = 0 even if ξ 6= 0.
Gradient of a differentiable function
Let (M, g) be a semi-Riemannian manifold of dimension n. Let f : M → R be differen-

tiable and p ∈ M . Then df |p ∈ Tp∗ M . In coordinates we have
n
X ∂f
df = dxi .
∂xi
i=1
Since g|p is non-degenerate there exists exactly one ξ ∈ Tp M such that
df |p (η) = g|p (ξ, η) for all η ∈ Tp M.

Pn i ∂
Write ξ =: gradf |p . In local coordinates, write gradf = i=1 α ∂xi . Then we have:
n n
!
∂f X ∂f i ∂ ∂ X
i ∂ ∂
= dx = df = g α ,
∂xj ∂xi ∂xj ∂xj ∂xi ∂xj
i=1 i=1
n n
X ∂ ∂ X
= αi g , = αi gij .
∂xi ∂xj
i=1 i=1
n
i
X ∂f
Matrix multiplication with (gij )ij yields α = gij , thus
∂xj
j=1
n
X ∂f ij ∂
gradf = g
∂xj ∂xi
i,j=1
Lemma 4.4.4
Let M̄ be a semi-Riemannian manifold and f : M̄ → R smooth and c ∈ R be a regular value
of f . Then M := f −1 (c) ⊂ M̄ is a semi-Riemannian hypersurface of signature ε, if
hgradf, gradf i · ε > 0.
gradf |p
Moreover, we have ν := ∈ Np M and hν, νi = ε.
|gradf |p |
124
4.4 Hypersurfaces
Proof. Since c is a regular value, M is a hypersurface. The lemma follows once we show
gradf |p ⊥ Tp M.
Let ξ ∈ Tp M . We choose γ : I → M with γ̇(0) = ξ and we obtain:
d
0= f γ(t) |t=0 = df |p (ξ) = hgradf |p , ξi .
dt | {z }
≡c
Definition 4.4.5 Np M
Let M ⊂ M̄ be a semi-Riemannian hypersurface ν
and p ∈ M . Let ν ∈ Np M with |ν| = 1.
bc
p
M
The linear map Sν : Tp M → Tp M , characterized
by
hSν (ξ), ηi = hII(ξ, η), νi for all ξ, η ∈ Tp M,
is called the Weingarten map.
Lemma 4.4.6
The Weingarten map Sν is self-adjoint.
Proof. This is clear because II is symmetric.
Remark 4.4.7. We have S−ν = −Sν . Without specifying the choice of ν, the Weingarten
map is only determined up to a sign.
125
4 Submanifolds
Lemma 4.4.8
Let M ⊂ M̄ be a semi-Riemannian hypersurface and ν
p ∈ M . Let U ⊂ M be an open neighborhood of p and bc
ν ∈ C ∞ (U, N M ) with |ν| = 1. Then we have p

M U
¯ ξ ν.
Sν (ξ) = −∇
Proof. For all η ∈ C ∞ (U, T M ) we have:

¯ ξ η), ν = ∇
hSν (ξ), ηi = hII(ξ, η), νi = nor(∇ ¯ ξ η, ν
¯ ξν = − ∇
= ∂ξ hη, νi − η, ∇ ¯ ξ ν, η .
| {z }
=0
Gauß formula:
Let M ⊂ M̄ be a semi-Riemannian hypersurface with signature ε. Let ξ, η, ζ ∈ Tp M .
Then:

R(ξ, η)ζ = R̄(ξ, η)ζ + ε hSν (η), ζi Sν (ξ) − hSν (ξ), ζi Sν (η) .
For any non-degenerate plane E ⊂ Tp M we have
hSν (ξ), ξi hSν (η), ηi − hSν (ξ), ηi2

K(E) = K̄(E) + ε ·
hξ, ξi hη, ηi − hξ, ηi2
where ξ, η is a basis of E.
Pseudospheres and pseudo-hyperbolic spaces

Pk−1 i P
Now consider M̄ = Rn+1 with ḡ = − i=0 dx ⊗ dxi + ni=k dxi ⊗ dxi in Cartesian
coordinates x0 , . . . , xn . Then (M̄ , ḡ) is a semi-Riemannian manifold of index k. For
k = 0 we have the Euclidean metric and for k = 1 the Minkowski metric. For general k
the representing matrix of ḡ in Cartesian coordinates is given by
 
−1 .
 . . −1 0 
(ḡij ) =  1.  .
0 ..
1
In particular, all ḡij are constant. Hence all Christoffel symbols vanish in Cartesian
coordinates. Therefore the curvature vanishes:
R̄ ≡ 0, K̄ ≡ 0, ric ≡ 0 and scal ≡ 0.
126
4.4 Hypersurfaces
Now consider the function

k−1
X n
X n
X
f : Rn+1 → R, f (x0 , . . . , xn ) = − (xi )2 + (xi )2 = εi (xi )2 .
i=0 i=k i=0
For the gradient we get

n
X ∂f ∂
gradf |x = i
(x) gij
∂x |{z} ∂xj
i,j=0 =εi δij
n
X ∂f ∂
= εi (x) i
∂xi ∂x
i=0
n
X ∂
= 2 εi · εi xi
∂xi
i=0
n
X ∂
= 2 xi .
∂xi
i=0
Thus gradf |x = 0 if and only if x = 0. Consequently, the only critical point of f is x = 0

and 0 = f (0) is the only critical value of f . If c 6= 0 then M := f −1 (c) therefore defines
a differentiable submanifold of codimension 1. We compute:
* n n
+
i ∂ i ∂
X X
hgradf |x , gradf |x i = 4 x , x
∂xi ∂xi
i=0 i=0
n
X
= 4 xi xj gij
i,j=0
X n
= 4 εj (xj )2
j=0
= 4f (x).
Hence for c > 0 we have that f −1 (c) is a semi-Riemannian hypersurface of signature
ε = +1, for c < 0 it is a semi-Riemannian hypersurface of signature ε = −1.
Definition 4.4.9. Let r > 0. The semi-Riemannian hypersurface

Skn (r) := f −1 (r 2 )
of (Rn+1 , ḡ) is called the pseudo-sphere of radius r and with index k. The semi-
Riemannian hypersurface
n
Hk−1 (r) := f −1 (−r 2 ),
is called the pseudo-hyperbolic space of index k − 1.
127
4 Submanifolds
Example 4.4.10. Let k = 0 and ḡ = geucl . Example 4.4.11. The case k = 1 and ḡ =
Then S0n (r) = S n (r) is the standard gMink is also of great interest.
sphere of radius r.
f −1 (0)
bc
H0n (r)
r
S1n (r)
S n (r)
Definition 4.4.12. The hypersurface H n := {x ∈ H0n (1) | x0 > 0} together with the
induced Riemannian metric ghyp is called the n-dimensional hyperbolic space.
Definition 4.4.13. The hypersurface S14 (r) together with the induced Lorentzian met-
ric is called deSitter spacetime and H14 (r) is called anti-deSitter spacetime.
Remark 4.4.14. The pseudo-sphere Skn (r) is diffeomorphic to Rk × S n−k while the
pseudo-hyperbolic space Hkn (r) is diffeomorphic to S k × Rn−k . See the exercises or
[ON83, page 111] for a proof of this fact.
We determine the curvature of these hypersurfaces. For M = f −1 (c) with c 6= 0 we

recall
hgradf |x , gradf |x i = 4f (x) = 4c,
hence
n
gradf |x gradf |x 1X i ∂
ν|x := p = = x .
|4c| 2r r ∂xi
i=0
128
4.4 Hypersurfaces
By Lemma 4.4.8 we get

1
Sν = − id
r
Now the Gauß formula yields
ε ε
R(ξ, η)ζ = (hη, ζi ξ − hξ, ζi η) and K≡
r2 r2
We compute
n
X
ric(ξ, η) = εi hR(ξ, ei )ei , ηi
i=1
n
ε X
= εi hei , ei i ξ − hξ, ei i ei , η
r2 | {z }
i=1 =εi
ε
= (n hξ, ηi − hξ, ηi),
r2
thus
ε(n − 1) ε(n − 1)n
ric = g and scal =
r2 r2
Remark 4.4.15. For the Einstein tensor of S14 (r) or H14 (r) we get
1 3ε 1ε·3·4 ε
G = ric − scal · g = 2 g − g = −3 2 g.
2 r 2 r2 r
Thus deSitter and anti-deSitter spacetime are vacuum solutions of the Einstein field
equations with cosmological constant Λ = r32 and Λ = − r32 , respectively.
Next we determine the geodesics of the pseudo-spheres and pseudo-hyperbolic spaces.

Let p ∈ M where M = Skn (r) or M = Hk−1 n (r) and let ξ ∈ T M ⊂ T Rn+1 ∼ Rn+1 with
p p =
ξ 6= 0. What is the geodesic c with c(0) = p and ċ(0) = ξ?
Note that p 6= 0. Then p and ξ, considered as vectors in Rn+1 , are linearly independent
because ξ ∈ Tp M and p ∈ Np M . Let E ⊂ Rn+1 be the plane spanned by p and ξ. If
ξ is space-like or time-like, then E is non-degenerate for ḡ. Then the reflection (w.r.t.
ḡ) about E is an isometry of (Rn+1 , ḡ). The restriction of the reflection to M yields
an isometry ϕ of M , see the discussion of isometries below. Now E ∩ M is the fixed
point set of ϕ, hence a 1-dimensional totally geodesic submanifold. In other words, if
we parametrize the connected component of E ∩ M containing p proportionally to arc-
length or eigentime, respectively, in such a way that ċ(0) = ξ, then it is the geodesic c
we are after.
129
4 Submanifolds
E
E
M M M
If ξ is light-like, then E is degenerate. But now E ∩ M consists of two parallel straight

lines. If we take any affine parametrization of the straight line containing p, then we get
a geodesic in (Rn+1 , ḡ) which contains p and lies entirely in M . Thus it is also a geodesic
in M . When choose the affine parametrization such that c(0) = p and ċ(0) = ξ, then we
found the right geodesic also in the light-like case.
In order to determine the isometry group of pseudo-spheres and pseudo-hyperbolic
spaces we define
O(n + 1 − k, k) := {A ∈ GL(n + 1) | hAx, Ayi = hx, yi ∀ x, y ∈ Rn+1 }.
Pk−1 j j Pn
Here hx, yi = − j=0 x y + j=k xj y j . We have O(n + 1, 0) = O(n + 1) and O(n, 1) is
the Lorentz group. For any A ∈ O(n + 1 − k, k) we have
A(Skn (r)) = Skn (r) and n

A(Hk−1 n
(r)) = Hk−1 (r).
Since the semi-Riemannian metric of M is obtained by restricting ḡ to M , the restriction

of isometries of (Rn+1 , ḡ) to M are isometries of M . We have constructed an injective
group homomorphism
O(n + 1 − k, k) → Isom(M ),
A 7→ A|M .
Next we show that this homomorphism is also surjective.
Proposition 4.4.16
Let M be a semi-Riemannian manifold, let p ∈ M and ϕ, ψ ∈ Isom(M ) with ϕ(p) = ψ(p)
and dϕ|p = dψ|p .
Then ϕ and ψ coincide on the set of all points which can be joined with p by a geodesic.
130
4.4 Hypersurfaces
Proof: Let c : [0, 1] → M be a geodesic with c(0) = p q dϕ|p (ċ(0))

b
and c(1) = q. Then c̃ := ϕ ◦ c and ĉ := ψ ◦ c are also c ϕ(q) ϕ ◦ c b
geodesics and we have c̃(0) = ϕ(p) = ψ(p) = ĉ(0)

b
ϕ(p)
ċ(0)
˙
and c̃(0) ˙
= dϕ|p (ċ(0)) = dψ|p (ċ(0)) = ĉ(0). Therefore b p
c̃ = ĉ. In particular, ϕ(q) = c̃(1) = ĉ(1) = ψ(q).
M
Corollary 4.4.17
If all points of M can be joined by geodesics with p, then every isometry ϕ of M is uniquely
determined by ϕ(p) and dϕ|p .
Example 4.4.18. Let M = (Rn , geucl ). We already know

{Euclidean motions} ⊂ Isom(M ),
where a Euclidean motion ϕ : Rn → Rn has the form ϕ(x) = Ax + b with A ∈ O(n) and
b ∈ Rn . We can now use Proposition 4.4.16 to show that there are no further isometries
of Euclidean space. Let ϕ ∈ Isom(M ). Put b := ϕ(0) and A := dϕ|0 ∈ O(n). Then
the Euclidean motion ϕ̃(x) := Ax + b satisfies ϕ̃(0) = b = ϕ(0) and dϕ̃|0 = A = dϕ|0 .
Since any two points in Euclidean space can be joined by a straight line we can apply
Corollary 4.4.17 and conclude ϕ = ϕ̃. This proves
{Euclidean motions} = Isom(M ).
Similarly one can show
Isom(Rn , gMink ) = Poincaré group.
Remark 4.4.19. The assumption that the points can be joined with p by geodesics is
necessary for the statement of Corollary 4.4.17.
Example 4.4.20. Let M = {p1 , p2 , p3 } be a 0-dimensional mani-

fold consisting of 3 points. On a 0-dimensional manifold all tan-
gent spaces are trivial so g = 0 is a Riemannian metric. All bi-
b
jective maps M → M are isometries. Consider the following two p3
maps: 
 p1 7→ p1 p1 p2
b b
ϕ1 := id, and ϕ2 := p2 7→ p3

p3 7→ p2
Then ϕ1 6= ϕ2 despite ϕ1 (p1 ) = ϕ2 (p1 ) and dϕ1 |p1 = 0 = dϕ2 |p1 .
131
4 Submanifolds
Example 4.4.21. Here is a 1-dimensional example. Let M =

{(x, y) ∈ R2 | |y| = 1} = M + ⊔ M − where M ± = {(x, y) ∈
R2 | y = ±1}. Let p = (0, 1) ∈ M . We provide M with the
+
Riemannian metric induced by the Euclidean metric on R2 . M b
p
Put

(x, y) on M +
ϕ1 := id, and ϕ2 (x, y) :=
(−x, y) on M − M−
Both ϕ1 and ϕ2 are isometries. Now ϕ1 (p) = ϕ2 (p) and
dϕ1 |p = dϕ2 |p but ϕ1 6= ϕ2 .
Lemma 4.4.22
On M = Skn (r) (where 0 ≤ k ≤ n − 1), on M = Hkn (r) (where 1 ≤ k ≤ n) and on
M = H n (r) any two points can be joined by a geodesic.
Proof. W.l.o.g. we assume n ≥ 2. Let p, q ∈ M . Since M is connected we can choose

a continuous curve c : [0, 1] → M with c(0) = p and c(1) = q. W.l.o.g. we assume
c(t) 6∈ {p, −p} for all t ∈ (0, 1). Then p and c(t) are linearly independent for all t ∈ (0, 1)
and span a unique plane E(t).
For any t ∈ (0, 1) the intersection M ∩ E(t) consists of an ellipse or a pair of hyperbolas
or a pair of straight lines. For t → 0 the points c(t) converges to p; hence the points p
and c(t) lie on the same connected component of M ∩ E(t) for sufficiently small t.
For t ∈ (0, 1) we choose X(t) ∈ Rn+1 depending continuously on t, which spans
E(t) together with p and which, w.r.t. to the Euclidean skalar product h·, ·ieukl , satis-
fies X(t) ⊥ p and ||X(t)||eukl = 1. Since S n is compact there is a sequence ti ∈ (0, 1) with
ti → 1 such that X(ti ) converges. Put lim X(ti ) =: X(1). By continuity, X(1) ⊥ p and
i→∞
||X(1)||eukl = 1. Hence p and X(1) are linearly independent and span a plane E(1). By
continuity, p, q ∈ M ∩E(1) and they lie in the same connected component of M ∩E(1).
Theorem 4.4.23
Restriction yields isomorphisms
Isom(Skn (r)) ∼
= O(n + 1 − k, k) for 0 ≤ k ≤ n − 1
Isom(H (r)) ∼
n
k = O(n − k, k + 1) for 1 ≤ k ≤ n
Isom(H (r)) ∼
n
= SO(n, 1) := {A ∈ O(n, 1) | A00 > 0}.
132
4.5 Trigonometry in spaces of constant curvature
Proof. Put M = Skn (r), M = Hkn (r) or M = H n (r) and G = O(n + 1 − k, k), G =
O(n − k, k + 1), or G = SO(n, 1), respectively. We need to show: Every isometry of M
is of the form
ϕ = A|M with A ∈ G.
a) We first show that G acts transitively on M . This means that for all p, q ∈ M there
exists an A ∈ G with Ap = q.
Namely: W.l.o.g. let p = r · e0 = (r, 0, . . . , 0)T . From hq, qi = ±r 2 we see that b0 := 1r q
satisfies hb0 , b0 i = ±1. We extend b0 to a generalized eigenbasis b0 , b1 , . . . , bn of Rn+1 .
Now A := (b0 , b1 , . . . , bn ) ∈ G and Ap = r Ae0 = rb0 = q.
b) Next we show: For any linear isometry B : Tp0 M → Tp0 M where p0 = re0 , there exists
an A ∈ G such that ϕ = A|M satisfies ϕ(p0 ) = p0 and dϕ|p0 = B. Namely:
 
1 0...0
 0 
A :=   ...

B 
0
does the job.
c) Let now ϕ ∈ Isom(M ). Put p1 := ϕ(p0 ) where p0 = re0 . By a) there exists an A1 ∈ G
with A1 p0 = p1 . Hence ψ := A−1
1 |M ◦ ϕ is an isometry of M with ψ(p0 ) = p0 .
Moreover, B := dψ|p0 : Tp0 M → Tp0 M is a linear isometry. By b) there is an A2 ∈ G
such that χ := A2 |M satisfies dχ|p0 = B. Lemma 4.4.22 and Corollary 4.4.17 imply
χ = ψ. Thus
ϕ = A1 |M ◦ ψ = A1 |M ◦ χ = A1 |M ◦ A2 |M = (A1 ◦ A2 ) |M .
| {z }
∈G

We want to extend the classical trigonometry of the Euclidean plane to 2-dimensional
model spaces of constant curvature. This means that we investigate length- and angular
relations in geodesic triangles.
Notation 4.5.1. The model space Mnκ is defined as



 S n ( √1κ ) if κ > 0,

Mnκ := Rn if κ = 0,
 n 1
H ( √ ) if κ < 0.

|κ|
133
4 Submanifolds
Thus Mnκ is an n-dimensional Riemannian manifold with the constant sectional cur-
vature κ.
Remark 4.5.2. Since for any given three points there exists a two-dimensional subma-
nifold of Mnκ which contains these points, it suffices to consider the case n = 2.
Define the bilinear form on R3
hx, yiκ := x0 y 0 + κ(x1 y 1 + x2 y 2 ).
Set M̂κ := {x ∈ R3 | hx, xiκ = 1} and put

(
M̂κ if κ > 0, M̂−1
Mκ := 0 M̂1
{x ∈ M̂κ | x > 0} if κ ≤ 0. M̂0
1
In the case κ 6= 0, the metric h·, ·iκ on
κ R3
induces
a Riemannian metric on Mκ . In particular,
(
S 2 if κ = 1,
Mκ =
H 2 if κ = −1.
In the case κ = 0, every bilinear form on R3 of the form λ · x0 y 0 + x1 y 1 + x2 y 2 induces

the same Euclidean metric on M0 , independent of λ ∈ R. We choose λ = 0 and in the
case κ = 0 we make the definition:
1
κ hx, yiκ := x1 y 1 + x2 y 2 .
Lemma 4.5.3
For every κ ∈ R, the isometry group of Mκ contains the subgroup

Gκ := ϕ | ϕ = A|Mκ where A ∈ GL(3) with hAx, Ayiκ = hx, yiκ ,
1 1
hAx, Ayiκ = hx, yiκ ∀ x, y ∈ R3 and A(Mκ ) = Mκ .
κ κ
Remark 4.5.4. In the case κ 6= 0 the conditions hAx, Ayiκ = hx, yiκ and κ1 hAx, Ayiκ =
1
κ hx, yiκ are of course equivalent and we could omit one of them. But in the case κ = 0
we need both of them.
134
From hAx, Ayiκ = hx, yiκ it already follows that A(M̂κ ) = M̂κ . In the case κ ≤ 0, A
could possibly exchange the two connected components of M̂κ . This is ruled out by the
condition A(Mκ ) = Mκ . In the case κ > 0 we could omit this condition.
Proof of the Lemma. Let A ∈ Gκ . Since ϕ = A|Mκ is the restriction of the linear map A,
we get that for p ∈ Mκ the differential dϕ(p) : Tp Mκ → Tϕ(p) Mκ also is the restriction of
A,
dϕ(p) = A|Tp Mκ .
Here, the tangent spaces of Mκ are viewed as subvector spaces of R3 . Since A respects
the bilinear form κ1 h·, ·iκ , the differential dϕ(p) is a linear isometry for every p ∈ Mκ .
Thus ϕ is an isometry of Riemannian manifolds.
Remark 4.5.5. Indeed, we have Isom(Mκ ) = Gκ but we will not need this fact.
For κ = 1 we have
Gκ ∼
= {A ∈ GL(3) | hAx, Ayi = hx, yi ∀x, y ∈ R3 } = O(3)
the group of orthogonal transformations. For κ = −1, Gκ is the group of time-

orientation preserving Lorentz transformations.
In case κ = 0, we have:
G0 = {A ∈ GL(3) | hAx, Ayi0 = hx, yi0 , 01 hAx, Ayi0 = 1
0 hx, yi0 ∀ x, y ∈ R3 , AM0 = M0 }
   
 1 0 0 
1 1 2
= A=  b  b , b ∈ R, B ∈ O(2)
 B 
b2
This holds true since for any A ∈ G0 ,
x0 y 0 = (Ax)0 (Ay)0 = (A00 x0 + A01 x1 + A02 x2 )(A00 y 0 + A01 y 1 + A02 y 2 )
Thus 
0 2 A(M0 )=M0
 For x = y =e0 : 1 = (A0 )

 ⇒ A00 = ±1 ⇒ A00 =1.
 For x = y =e1 : 0 = (A01 )2 ⇒ A01 =0.



For x = y =e2 : 0 = (A02 )2 ⇒ A02 =0.
For x̂, ŷ ∈ R2 we have with x = (0, x̂)⊤ and y = (0, ŷ)⊤ :

1 1 1 0 0
hx̂, ŷiR2 = 0 hx, yi0 = 0 hAx, Ayi0 = 0 , = hB x̂, B ŷiR2 .
B x̂ B ŷ 0
135
4 Submanifolds
Hence B ∈ O(2) and therefore

1 0 2
G0 ⊂ A= b ∈ R , B ∈ O(2) .
b B
The other inclusion ”‘⊃”’ follows by a direct computation.
We now analyze, how G0 acts, if we identify M0 with R2 .

10

bB
R2 → M0 −→ M
0 → R2
1 10 1 1
x̂ 7→ 7→ = 7→ b + B x̂
x̂ bB x̂ b + B x̂
Hence the group G0 acts like the group of Euclidean motions.
As seen in the last paragraph, the geodesics in Mκ ,

viewed as a set of points, equal the sets of the form e0
b
M0
Mκ ∩ E, b
0
E
where E ⊂ R3 is a two-dimensional subvector space.
Lemma 4.5.6
The geodesics parametrized by arc-length γ : R → Mκ with γ(0) = e0 are then given by
 
cκ (r)
γ(r) =  sκ (r) · sin(ϕ) 
sκ (r) · cos(ϕ)
where ϕ ∈ R is fixed.
Proof. The curve γ stays in M̂κ because

hγ(r), γ(r)iκ = cκ (r)2 + κ sκ (r)2 sin(ϕ)2 + s2κ cos(ϕ)2 = cκ (r)2 + κ sκ (r)2 = 1.
Since γ(0) = e0 ∈ Mκ and γ is continuous, γ remains in Mκ . Moreover, γ lies in the plane

E, which is spanned by e0 and (0, sin(ϕ), cos(ϕ))⊤ . Hence γ is contained in Mκ ∩ E. In
addition, γ is parametrized by arc-length because
* −κ s (r)   −κ s (r) +
κ κ
1 1
κ hγ̇(r), γ̇(r)i κ = κ
 cκ (r) sin(ϕ)  ,  cκ (r) sin(ϕ) 
cκ (r) cos(ϕ) cκ (r) cos(ϕ) κ
136

1
= κ κ2 sκ (r)2 + κ cκ (r)2 sin(ϕ)2 + cκ (r)2 cos(ϕ)2
= κ sκ (r)2 + cκ (r)2
= 1.
The generalized sine and cosine functions allow us to explicitly write down many
isometries in Gκ .
Example 4.5.7. Rotations about the e0 -Axis are isometries,
 
1 0 0
0 cos(ϕ) − sin(ϕ) ∈ Gκ
0 sin(ϕ) cos(ϕ)
for any ϕ and any κ ∈ R. Using κ s2κ + c2κ = 1 one easily checks that
 
cκ (r) −κ sκ (r) 0
sκ (r) cκ (r) 0 ∈ Gκ
0 0 1
for all r ∈ R. In the case κ = 1 this is a rotation about the e2 -axis. For κ = 0 this is
the identity, hence uninteresting. In the case κ = −1 such isometries are called Lorentz
boosts. Similarly, one sees that
 
cκ (r) 0 κ sκ (r)
Lr :=  0 1 0  ∈ Gκ .
sκ (r) 0 −cκ (r)
Before using these isometries we observe that
 
cκ (r)
Lr e0 =  0 
sκ (r)
and
    
cκ (r) cκ (r) 0 κsκ (r) cκ (r)
Lr  0  =  0 1 0  0 
sκ (r) sκ (r) 0 −cκ (r) sκ (r)
 2 2

cκ (r) + κsκ (r)
= 0 
sκ (r)cκ (r) − cκ (r)sκ (r)
 
1
= 0

0
= e0 .
Thus Lr interchanges the points e0 and (cκ (r), 0, sκ (r))⊤ .
137
4 Submanifolds
Definition 4.5.8.
C
b
Let M be a Riemannian manifold. A geodesic triangle γA
is a 6-tupel γB
(A, B, C, γA , γB , γC ),
b
B
where A, B, C ∈ M are pairwise disjoint points, γA , γB b
γC
and γC geodesic segments with endpoints B and C, C A
and A or A and B, respectively.
The points A, B and C are the vertices, the geodesic segments γA , γB and γC are the
sides of the geodesic triangle. The angle at a vertex is defined to be the angle of the
tangent vectors of the sides at that vertex.
Cb
γ a
Let (A, B, C, γA , γB , γC ) a geodesic triangle in Mκ . The sides b
have the lengths a, b and c, respectively, and the angles are
denoted by α, β and γ, respectively. β b
α B
b
c
A
Here the length of a geodesic segment γ is defined as the length of the parameter in-
terval × the norm of γ̇, which is constant. A more general definition of the length of a
differentiable curve in a Riemannian manifold will be introduced later. Since the isom-
etry group of Mκ acts transitively, we can assume w.l.o.g. that
 
1
A = e0 =  0  .
0
Applying an isometry of the form
 
1 0 0
 0 cos(ϕ) − sin(ϕ) 
0 sin(ϕ) cos(ϕ)
(rotation about the e0 -axis) we can rotate B in the e0 -e2 -plane without moving A = e0 .
The formula from Lemma 4.5.6 for the geodesic γC with ϕ = 0 and r = c then tells us
 
cκ (c)
B= 0 
sκ (c)
138
Lemma 4.5.6 for the geodesic γB with ϕ = α and r = b yields

 
cκ (b)
C =  sκ (b) sin(α)  .
sκ (b) cos(α)
Lc C b
γ b
B = Lc A
Hence the isometry Lc interchanges the points A and
a
B and we obtain a new geodesic triangle. On the one
hand one can compute Lc C similarly as C itself and α b
one obtains β b
c
A = Lc B
 
cκ (a)
Lc C =  sκ (a) sin(β)  .
On the other hand sκ (a) cos(β)
  
cκ (c) 0 κsκ (c) cκ (b)
Lc C =  0 1 0   sκ (b) sin(α) 
sκ (c) 0 −cκ (c) sκ (b) cos(α)
 
cκ (c)cκ (b) + κsκ (c)sκ (b) cos(α)
= sκ (b) sin(α) 
sκ (c)cκ (b) − cκ (c)sκ (b) cos(α)
Thus we obtain the equations:
cκ (a) = cκ (c)cκ (b) + κ sκ (c)sκ (b) cos(α) (Law of Cosines) (1)
sκ (a) sin(β) = sκ (b) sin(α)
sκ (a) sκ (b)
= (Law of Sines) (2)
sin(α) sin(β)
sκ (a) cos(β) = sκ (c)cκ (b) − cκ (c)sκ (b) cos(α) (3)
Equation (3) with the roles of B and C interchanged yields

sκ (a) cos(γ) = sκ (b)cκ (c) − cκ (b)sκ (c) cos(α) (4)
Equation (3) · cos(α) − (2) · sin(α)2 · sin(β) then yields
sκ (a) cos(β) cos(α) − sκ (a) sin(β) sin(α)
= sκ (c)cκ (b) cos(α) − cκ (c)sκ (b) cos(α)2 − sκ (b) sin(α)2
Hence
sκ (a)(cos(α) cos(β) − sin(α) sin(β))
(4)
= sκ (b)cκ (c) − sκ (a) cos(γ) − sκ (b)cκ (c) cos(α)2 − sκ (b) sin(α)2
= sκ (b)cκ (c) sin(α)2 − sκ (a) cos(γ) − sκ (b) sin(α)2
(2)
= sκ (a)cκ (c) sin(α) sin(β) − sκ (a) cos(γ) − sκ (a) sin(α) sin(β)
139
4 Submanifolds
and thus cos(α) cos(β) = cκ (c) sin(α) sin(β) − cos(γ), hence
cos(γ) = cκ (c) sin(α) sin(β) − cos(α) cos(β) (Cosine Rule for Angles).
We have proved
Theorem 4.5.9
Let κ ∈ R. For a geodesic triangle in Mκ with the side lengths a, b, c and the angles α, β, γ
we have
(1) Law of Sines:

sκ (a) sκ (b) sκ (c)
= = .
sin(α) sin(β) sin(γ)
(2) Law of Cosines (Cosine Rule for Sides):
cκ (a) = cκ (b)cκ (c) + κsκ (b)sκ (c) · cos(α),

cκ (b) = cκ (a)cκ (c) + κsκ (a)sκ (c) · cos(β),
cκ (c) = cκ (a)cκ (b) + κsκ (a)sκ (b) · cos(γ).
(3) Cosine Rule for Angles:
cos(α) = cκ (a) sin(β) sin(γ) − cos(β) cos(γ),

cos(β) = cκ (b) sin(α) sin(γ) − cos(α) cos(γ),
cos(γ) = cκ (c) sin(α) sin(β) − cos(α) cos(β).
Now we analyze the sum of angles in the model space of constant curvature.
Theorem 4.5.10
Let κ ∈ R. For the sum of angles α + β + γ of a geodesic triangle in Mκ with the inner angles
0 < α, β, γ < π we have 
 > π, if κ> 0
α+β+γ = π, if κ= 0

< π, if κ< 0
140
b b b
b b b
b b b
κ>0 κ=0 κ<0
Proof. W.l.o.g. we assume that α ≥ β. For this proof we will use the notation “⋚” for
“<”, if κ > 0, for “=”, if κ = 0, and for “>”, if κ < 0. We have −κ ⋚ 0, for instance.
If is κ > 0, then Mκ is the sphere of radius √1κ . Thus in this case the side lengths have
2π
to be < √
κ
. In the case κ ≤ 0, we do not have any bounds on the side lengths. We use
the convention √1 = ∞, if κ ≤ 0. With this convention we have in all cases
κ
cκ ⋚ 1
2π
in the interval (0, √κ
). Since sin is positive on (0, π) the Cosine Rule for Angles yields
cos(α) = cκ (a) sin(β) sin(γ) − cos(β) cos(γ)

⋚ sin(β) sin(γ) − cos(β) cos(γ)
= − cos(β + γ)
= cos(π − (β + γ))
= cos(β + γ − π).
Since 0 < β, γ < π we have −π < π − (β + γ) < π.
First case: π − (β + γ) ≥ 0.
Since cos is strictly monotonically decreasing on [0, π], the relation
cos(α) ⋚ cos(π − (β + γ)) yields π − (β + γ) ⋚ α and thus π ⋚ α + β + γ. This
is what we wanted to show.
Second Case: π − (β + γ) < 0.

If κ > 0, we obtain π < β + γ < α + β + γ directly, which proves the claim. Hence, let
κ ≤ 0. Then from cos(α) ≥ cos(β + γ − π) we may deduce that α ≤ β + γ − π. Since
α ≥ β and γ < π this implies
α < α + π − π = α,
giving a contradiction.
Remark 4.5.11. Since the inner angles are < π, we always have for the sum of angles in
a geodesic triangle α + β + γ < 3π. It is easy to see that for Mκ with κ > 0 the sum of
141
4 Submanifolds
angles of a geodesic triangle can take all values in (π, 3π). For Mκ with κ < 0 all values
of the interval (0, π) occur.
142
5 Riemannian Geometry
From now on we concentrate on Riemannian geometry, that is, on semi-Riemannian
manifolds whose metric is positive definite and hence defines a Euclidean scalar prod-
uct on each tangent space. One special feature of the Riemannian case is that each
connected Riemannian manifold naturally becomes a metric space.
5.1 The Riemannian distance function

General Assumption. Let M be a connected Riemannian manifold and let h·, ·i denote
the Riemannian metric.
Definition 5.1.1. Let c : [a, b] → M be a continuous piecewise C 1 -curve. Then we

call
Zb
L[c] := ||ċ(t)|| dt
a
the length of c.
Remark 5.1.2. The length does not depend on the parametrization of the curve.
Namely, if ϕ : [a, b] → [α, β] is a parameter transformation, then we have
Zb
d
L[c ◦ ϕ] = (c ◦ ϕ)(t) dt
dt
a
Zb
= ||ċ(ϕ(t))|| · |ϕ̇(t)| dt
a
Zβ
Substitution
s = ϕ(t) = ||ċ(s)|| ds
α
= L[c].
143
Definition 5.1.3. Let p, q ∈ M . Then we call

d(p, q) = inf L[c] | c : [a, b] → M piecewise C 1 -curve with c(a) = p, c(b) = q
the Riemannian distance of p and q.
Remark 5.1.4. The infimum is, in general, not a minimum. In other words, there need
not exist a shortest curve connection p and q.
Example 5.1.5 Rn
b
p
M = Rn \ {0} and p = −q. We have d(p, q) = 2 ||p||, but 0 bc
every curve c from p to q has length L[c] > 2 ||p||. q

b
Theorem 5.1.6 (Gauß lemma)

Let p ∈ M and ξ ∈ Tp M . The geodesic γ(t) = Tp M η
expp (tξ) is supposed to be defined on [0, b]. ξ
ξ
b
Then expp is defined on an open neighborhood of

b
0 b
tξ
{tξ | 0 ≤ t ≤ b} ⊂ Tp M and we have
expp
(i) d expp |tξ (ξ) = γ̇(t).
d expp |tξ (η)
(ii) For η ∈ Ttξ Tp M ∼
= Tp M we have b
γ b
γ̇(t)b
p γ(t)
d expp |tξ (η), γ̇(t) = hη, ξi .
M
In particular, d expp |tξ (η) ⊥ γ̇(t), if η ⊥ ξ.
d d
Proof. (i) We compute d expp |tξ (ξ) = expp (t ξ + s ξ)|s=0 = γ(t + s)|s=0 = γ̇(t).
ds ds
(ii) By (i) it suffices to consider the case η ⊥ ξ. Let J be the Jacobi field along γ with
∇
J(0) = 0 and dt J(0) = η. Proposition 3.4.13 yields
J(t)
d expp |tξ (η) = for t > 0.
t
144
∇
Since both J and dt J are perpendicular to γ̇ at t = 0, this holds for all t. We conclude

J(t)
d expp |tξ (η), γ̇(t) = , γ̇(t) = 0 = hη, ξi .
t
We now consider the diffeomorphism
x
Φ : Tp M \ {0} −→ (0, ∞) × S n−1 , x = t · y 7→ (t, y) = (kxk, ),
kxk
where S n−1 ⊂ Tp M is the unit sphere in the tangent space. There exists an r > 0, such
that expp maps B(0, r) ⊂ Tp M diffeomorphically onto a neighborhood U of p in M .
Then the map
(0, r) × S n−1 → U \ {p}, (t, y) 7→ expp (ty),
is a diffeomorphism. Now let y 2 , . . . , y n be local coordinates on an open set U1 ⊂ S n−1 .

Then the coordinates given by the diffeomorphism
expp (ty) 7→ (t, y 2 , . . . , y n ),
are called geodesic polar coordinates.
Tp M
0 S n−1
b
Φ
B(0, r) ≈
expp U1
p
b
0 r
U
M
The Gauß lemma says that in such coordinates the Riemannian metric takes the form
 
1 0 ··· 0
 0 
 ...
(gij ) =  
∗ 
0
145
Corollary 5.1.7
Let r > 0 so small that expp |B̄(0,r) is a diffeomorphism on its image. Let c : [a, b] → M be a
piecewise C 1 -curve with c(a) = p and c(b) 6∈ expp (B(0, r)). Then L[c] ≥ r.
Proof. Let β ∈ (a, b) be minimal such that c(β) ∈ ∂ expp (B(0, r)) = expp (S n−1 (r)). Let
α ∈ [a, β) maximal such that c(α) = p. Now it is ensured that for τ ∈ (α, β) the curve
c(τ ) lies in expp (B(0, r)) \ {p}. For τ ∈ (α, β] we write
c̃(τ ) := expp −1 (c(τ )) = t(τ ) · y(τ )

c̃(τ ) n−1 . Let ξ˜ be the unit vector
where t(τ ) := kc̃(τ )k ∈ (0, r] and y(τ ) := kc̃(τ )k ∈ S
x
field on Tp M \ {0} which corresponds to ξ̃(x) = kxk under the canonical isomorphism
Tx Tp M ∼= Tp M . Using the diffeomorphism expp we transport this vector field to the
manifold, that is, on expp (B̄(0, r)) \ {p} we set

ξ(q) := d expp ξ˜ expp −1 (q) .
The first part of the Gauß lemma implies ||ξ|| ≡ 1. Because of
d dt dy
c̃(τ ) = · y(τ ) +t(τ ) · (τ )
dτ dτ |{z} |dτ{z }
= ξ̃(c̃(τ ))
⊥ ξ̃(c̃(τ ))
part (ii) of the Gauß lemma yields
D E D E
˙ ) = dt .

ξ c(τ ) , ċ(τ ) = d expp ξ˜ c̃(τ ) , d expp c̃(τ
˙ ) = ξ̃ c̃(τ ) , c̃(τ
dτ
146
Thus we get
L[c] ≥ L[c|[α,β] ] Tp M
Zβ 0 ξ̃
b
= ||ċ(τ )|| dτ
B(0, r)
α
Cauchy
Zβ expp

Schwarz ≥ ξ c(τ ) , ċ(τ ) dτ
inequality
α ξ
Zβ p
b
dt U
= dτ
dτ
α M
= t(β) − t(α)
= r − 0 = r.
Theorem 5.1.8
(M, d) is a metric space.
Proof. a) Obviously we have d(p, q) ≥ 0 and d(p, p) = 0 because the constant curve has
length 0. Now let p 6= q. We have to show d(p, q) > 0. Choose r > 0 such that expp |B(0,r)
is a diffeomorphism and q 6∈ expp (B(0, r)). Then by Corollary 5.1.7 every curve from p
to q has length r at least. Hence d(p, q) ≥ r > 0.
b) Symmetry d(p, q) = d(q, p) is clear. Simply traverse the curves in the opposite direc-
tion.
c) It remains to show the triangle inequality d(p, q) ≤ d(p, r) + rd(r, q). b
Let ε > 0. Choose a continuous piecewise C 1 -
curves c1 from p to r with L[c1 ] ≤ d(p, r) + ε and c2
c2 from r to q with L[c2 ] ≤ d(r, q) + ε. Now concate- c1
nate c1 and c2 to a continuous piecewise C 1 -curve c b
q
from p to q. Then we have b
p
d(p, q) ≤ L[c] = L[c1 ] + L[c2 ] ≤ d(p, r) + ε + d(r, q) + ε.
Taking the limit ε ց 0 yields the assertion.
147
Notation 5.1.9. For p ∈ M and r > 0 set
B(p, r) := {q ∈ M | d(p, q) < r},

B̄(p, r) := {q ∈ M | d(p, q) ≤ r},
S(p, r) := {q ∈ M | d(p, q) = r}.
Definition 5.1.10. For p ∈ M

injrad(p) := sup{r | expp |B(0,r) : B(0, r) → expp B(0, r) is diffeomorphism}
is called the injectivity radius of M at p.
Example 5.1.11. The injectivity radius depends on p.
here injrad is large
here injrad is small
Remark 5.1.12. For 0 < r < injrad(p) we have expp (B(0, r)) = B(p, r). Namely:
“⊂”: Let q = expp (ξ) with ||ξ|| < r. Then t 7→ expp (tξ), t ∈ [0, 1], is a curve from p to q
with length ||ξ|| < r. Hence d(q, p) < r, i.e., q ∈ B(p, r).
“⊃”: Corollary 5.1.7.
Theorem 5.1.13
The metric d induces the original topology on M .
148
Proof. For the moment we denote the open subsets w.r.t. d of M as “d-open”. We have
to show: d-open = open.
a) Claim: Every d-open set is open.
Let U ⊂ M be d-open. For every p ∈ U there exists a r(p) > 0, such that B(p, r(p)) ⊂ U .
W.l.o.g. let r(p) < injrad(p). Then B(p, r(p)) = expp (B(0, r(p))) is the diffeomorphic
| {z }
open in Tp M
S
image of an open subset of Tp M . Hence it is open itself. Therefore U = B p, r(p)
p∈M
is the union of open subsets of M and thus open.
b) Claim: Every open set is d-open. The proof is similar.
Corollary 5.1.14
The map d : M × M → R is continuous.
If Φ ∈ Isom(M ).
Remark 5.1.15. Then we have L[Φ ◦ c] = L[c] and thus also
d Φ(p), Φ(q) = d(p, q).
Rb
Recall that E[c] = 1
2 ||ċ(t)||2 dt is the energy of c.
a
Proposition 5.1.16
Let M be a Riemannian manifold and c : [a, b] → M be a continuous, piecewise C 1 -curve.
Then we have
L[c]2 ≤ 2(b − a) · E[c].
Equality holds if and only if c is parametrized proportional to arc-length.
Proof. With the Cauchy Schwarz inequality for the L2 -scalar product we obtain:
 b 2
Z Zb Zb
2
L[c]2 =  ||ċ(t)|| · 1 dt ≤ ||ċ(t)|| dt · 12 dt = 2E[c] · (b − a).
a a a
149
Equality holds if and only if ||ċ|| and 1 (as a function) are linearly dependent. This means
that ||ċ|| is constant, i.e., that c is parametrized proportionally to arc-length.
Corollary 5.1.17
A curve c minimizes the energy in the set of all continuous piecewise C 1 -curves connecting p
and q if and only if c minimizes the length and is parametrized proportionally to arc-length.
Remark 5.1.18. By Corollary 2.6.10 energy minimizing curves are geodesics.
Corollary 5.1.19
Every shortest curve from p to q is a geodesic up to parametrization. It is a geodesic if and
only if it is parametrized proportionally to arc-length.
Caution! The converse is not true. Not every geodesic is a shortest curve connecting its
endpoints.
Example 5.1.20. Great circles on S n are geodesics connecting points to themselves. But
the only shortest curves connecting a point to itself are constant curves which have
length 0.

Definition 5.1.21. A geodesic γ : [a, b] → M with L[γ] = d γ(a), γ(b) is called mini-
mal.
150
5.2 Completeness
5.2 Completeness
General Assumption. Throughout this section let M be a connected Riemannian ma-
nifold.
Definition 5.2.1. Let p ∈ M . Then M is called geodesically complete at p if expp is

defined on all of Tp M , i.e., if all geodesics through p are defined on all of R.
Theorem 5.2.2 (Hopf-Rinow)

Let M be a connected Riemannian manifold and p ∈ M . Then the following assertions are
equivalent:
(1) M is geodesically complete at p.
(2) M ist geodesically complete at all q ∈ M .
(3) The closed balls B̄(p, r) are compact for all r > 0.
(4) The closed balls B̄(q, r) are compact for all r > 0 and all q ∈ M .
(5) (M, d) is complete as a metric space, i.e., all d-Cauchy sequences converge.
All of these conditions imply in addition
(6) Every point q can be joined with p by a minimal geodesic.
Remark 5.2.3. Assertion (6) is weaker than (1) through (5). bc
Example 5.2.4. Let M = {x ∈ Rn | ||x|| < 1} with the Euclidean bc
metric. Then M satisfies (6), but not (1)–(5).
Definition 5.2.5. If the equivalent conditions (1)–(5) in Theorem 5.2.2 hold, then one
calls M a complete Riemannian manifold.
151
Corollary 5.2.6
Every compact Riemannian manifold is complete.
Proof of Corollary 5.2.6. We check condition (3) in the Hopf-Rinow theorem. Indeed,
B̄(p, r) ⊂ M is a closed subset of the compact space M and thus compact itself.
Proof of Theorem 5.2.2. We will prove this theorem in five steps. The structure of the
proof is as follows:
(a) trivial (e)
(5) (2) (1) (6)
| {z }
(b) (d)
(4) (3)
(c)
a) Let γ : (α, β) → M be a geodesic with maximal domain of definition. W.l.o.g. we

assume that γ is parametrized by arc-length.
We assume β < ∞ (the case α > −∞ is analogous). Then we have for a sequence
i→∞
ti ∈ (α, β) with ti −→ β, that

d γ(ti ), γ(tj ) ≤ L[γ|[ti ,tj ] ] = |ti − tj |.
Hence (γ(ti ))i∈N is a d-Cauchy-sequence. Since (M, d) is complete there exists a

i→∞
q ∈ M with γ(ti ) −→ q.
1. Claim: The limit point q does not depend on the choice of the sequence (ti )i∈N with
i→∞
ti −→ β.
Proof. If (t′i )i∈N is another such sequence with q ′ = limi→∞ γ(t′i ), then also (t′′i )i∈N is
such a sequence, where
′′ tj , i = 2j
ti :=
t′j , i = 2j + 1
The sequence (γ(t′′i ))i∈N is a d-Cauchy-sequence with accumulation points q and q ′ .
We thus have q = q ′ . This proves the first claim. ✓
Thus we obtain a continuous continuation γ̄ : (α, β] → M of γ by

γ(t), t ∈ (α, β)
γ̄(t) =
q ,t=β
2. Claim: The velocity field γ̇ also has a continuous extension to (α, β].
152
5.2 Completeness
Proof. Let x : U → V be a chart of M around q with x(q) = 0. Choose r > 0 such that
B̄(0, r) ⊂ V . Since B̄(0, r) is compact, there exist constants C1 , C2 , C4 > 0 with
• Γkij (y) ≤ C1 for all y ∈ B̄(0, r).
n
P
• ||a||max ≤ C2 aj ∂x∂ j x−1 (y) for all a = (a1 , . . . , an ) ∈ Rn and y ∈ B̄(0, r).
j=1
g
∂Γkij
• ∂xl
(y) ≤ C4 for all y ∈ B̄(0, r).
Choose ε > 0 small enough so that γ(t) ∈ x−1 (B̄(0, r)) for t ∈ (β − ε, β). Write
γ k := xk ◦ γ and ak := γ̇ k . By the equations for geodesics we obtain:
n
X n
X
ȧk = γ̈ k = − Γkij (γ 1 , . . . , γ n ) · γ̇ i γ̇ j = − Γkij (γ 1 , . . . , γ n ) ai aj
i,j=1 i,j=1
and hence
ȧk ≤ n2 · C1 · ||a||2max .
This implies
ȧ max
≤ n2 C1 · ||a||2max ≤ n2 C1 · C2 2 ||γ̇||g 2 = n2 C1 C2 2 =: C3 .
| {z }
=1
We get
Ztj Ztj
||a(ti ) − a(tj )||max = ȧ(t) dt ≤ ||ȧ(t)||max dt ≤ C3 |ti − tj | .
ti max ti
Thus the a(ti ) form a Cauchy sequence in Rn and hence converge to some A ∈ Rn .
i→∞
As before A is independent of the special choice of the sequence (ti )i∈N with ti −→
β. Thus we obtain a continuous extension of a by
(
a(t), t ∈ (β − ε, β)
ā(t) :=
A, t=β
Hence the velocity field γ̇ is extended continuously to t = β. This shows that the
extension γ̄ of γ is C 1 . ✓
Differentiation of the geodesics equations yields
n n
!
X X ∂Γkij
äk = − al ai aj + 2 Γkij ȧi aj
∂xl
i,j=1 l=1
153
This implies
||ä||max ≤ n3 C4 ||a||3max + 2n2 C1 ||ȧ||max ||a||max

≤ n3 C4 C2 3 + 2n2 C1 C3 C2
=: C5
As before we see that (ȧ(ti ))i∈N forms a d-Cauchy-sequence in Rn . This shows that
the extension γ̄ is even a C 2 -curve. By continuity it satisfies the geodesic equation
also at t = β.
˙
Now let γ̂ : (β − δ, β + δ) → M be the geodesic with γ̂(β) = γ̄(β) and γ̂(β) ˙
= γ̄(β).
Since geodesics are uniquely determined by their initial values, γ̂ and γ̄ coincide on
their common domain of definition. This yields a continuation of γ as a geodesic on
(α, β + δ). This contradicts the maximality of β and thus shows (2).
b) Let all closed balls in M be compact. Let (pi )i∈N be a Cauchy sequence in M . Since
Cauchy sequences are bounded, there exists a R > 0 such that pi ∈ B̄(p, R) for all i ∈
N. Since B̄(p, R) is compact, the Cauchy sequence (pi )i∈N has an accumulation point
q ∈ B̄(p, R). Since accumulation points of Cauchy sequences are unique, (pi )i∈N
converges to q.
c) Let all B̄(p, r) be compact for all r > 0. Let q ∈ M
and let R > 0. Set r := R + d(p, q). Then
q R b
B̄(q, R) ⊂ B̄(p, r), r b
p
because for x ∈ B̄(q, R) we have
d(x, p) ≤ d(x, q) + d(q, p) ≤ R + d(q, p) = r.

Hence B̄(q, R) is a closed subset of the compact set B̄(p, r) and therefore it is compact
itself.
d) Let (pi )i∈N be a sequence in B̄(p, r). We have to show that (pi )i∈N possesses a con-
vergent subsequence.
By (6) there exist minimal geodesics γi with γi (0) = p and γi (ti ) = pi for suitable ti .
W.l.o.g. let γi be parametrized by to arc-length. Then ti =
L[γi ] = d(p, pi ) ≤ r. pi
b
b
b
The γ̇i (0) are unit vectors in Tp M . Since S n−1 (1) ⊂ Tp M is b
γ̇i (0) b
b
compact we have, after passing to a suitable subsequence, b b
..
p .
X b
i→∞ q
γ̇i (0) −→ X ∈ Sn − 1(1) ⊂ Tp M.
B̄(p, r)
i→∞
The ti lie in the compact interval [0, r]. After passing to a subsequence again, ti −→
T ∈ [0, r] converges too. Set q := expp (T · X). This definition is possible because of
154
5.2 Completeness
(1). We now have

lim pi = lim expp ti · γ̇i (0) = expp lim ti γ̇i (0) = expp (T X) = q.
i→∞ i→∞ i→∞
This proves (3).

e) Let q ∈ M . We already know, that we can find minimal geodesics from p to q, if
q ∈ B(p, injrad(p)).
Let ck be continuous piecewise C 1 -curves
from p to q with L[ck ] = d(p, q) + εk with qk ck
b
εk ց 0. b
b
b
We assume q 6∈ B(p, injrad(p)) because oth- p b

b
b
q̄ γ
erwise we are finished. Choose 0 < r0 < b
q
injrad(p). Then B̄(p, r0 )

S(p, r0 ) = expp S n−1 (r0 )
is compact. Let qk be the first intersection point of ck with S(p, r0 ). After passing to
a subsequence, qk possesses a limit q̄ ∈ S(p, r0 ). We have
d(p, q) ≤ d(p, qk ) + d(qk , q) ≤ L[ck ] ≤ d(p, q) + εk
k→∞
⇒ d(p, q) ≤ d(p, q̄) + d(q̄, q) ≤ d(p, q)
⇒ d(p, q) = d(p, q̄) + d(q̄, q)
Let γ be the unique minimal geodesic that connects p with q̄. We parametrize γ by
arc-length. With (1) we can extend γ to [0, d(p, q)].
It remains to show that γ : [0, d(p, q)] → M is a minimal geodesic from p to q. Set

I := t ∈ [0, d(p, q)] | d(p, γ(t)) = t and d(p, γ(t)) + d(γ(t), q) = d(p, q) .
We have seen that [0, r0 ] ⊂ I. Set t0 := sup(I). We have to show that t0 = d(p, q)
because then

d γ(t0 ), q = d(p, q) − d γ(t0 ), p = d(p, q) − t0 = 0,
which implies γ(t0 ) = q and that γ is a minimal geodesic from p to q.

We therefore assume that t0 < d(p, q) from which
we have to derive a contradiction. Set q ′ := γ(t0 ).
Choose 0 < r1 < d(p, q) − t0 such that B(q ′ , r1 ) is a p ′
r1 γ q̄
b
normal coordinate neighborhood. As above, there 1

b
γ b
q
exists a q̄ ′ ∈ ∂B(q ′ , r1 ) with
b
γ(t ) = q ′ 0
d(q ′ , q̄ ′ ) + d(q̄ ′ , q) = d(q ′ , q).
Now let γ1 be a minimal geodesic, parametrized by arc-length, with γ1 (t0 ) = q ′ and

γ(t0 + r1 ) = q̄ ′ .
155
⇒ d(p, q̄ ′ ) ≤ d(p, q ′ ) + d(q ′ , q̄ ′ )

= d(p, q ′ ) + d(q ′ , q) − d(q̄ ′ , q)
= d(p, q) − d(q ′ , q) + d(q ′ , q) − d(q̄ ′ , q)
= d(p, q) − d(q̄ ′ , q)
≤ d(p, q̄ ′ )
⇒ d(p, q̄ ′ ) = d(p, q ′ ) + d(q ′ , q̄ ′ )
⇒ The curve γ|[0,t0 ] ∪ γ1 |[t0 ,t0 +r1 ] is a shortest.
⇒ t0 + r1 ∈ I. This contradicts the maximality of t0 . We have proved (6).
5.3 The second variation of the energy
We recall: If cs is a C 2 -variation of c : [a, b] → M with vari-

ξ
ational field ξ, then the first variation formula (Theorem
b
2.6.5) says:
b
c
Rb cs
d ∇
ds E[cs ]|s=0 = − ξ, dt ċ dt + hξ, ċi |ba .
a
ċ(t+
c(ti+1 )
If cs is continuous and only piecewise that is, C 2, i )
there exists a partition a = t0 < t1 < · · · < tN = b, c
such that (s, t) 7→ cs (t) is continuous on (−ε, ε)×[a, b] b
c(ti )
b
and C 2 on (−ε, ε) × [ti−1 , ti ], then we have c(ti−1 ) ċ(t−

i )
N
d d X
E[cs ]|s=0 = E[cs |[ti−1 ,ti ] ]|s=0
ds ds
i=1
N Z ti !
X ∇ − +
= − ξ, ċ dt + ξ(ti ), ċ(ti ) − ξ(ti−1 ), ċ(ti )
ti−1
dt
i=1
Z b N
∇ X
= − ξ, ċ dt + ξ(b), ċ(b− ) − ξ(a), ċ(a+ ) + ξ(ti ), ċ(t− +
i ) − ċ(ti )
a dt
i=1
d
Question: If c is a continuous and only piecewise C 2 -curve with ds E[cs ]|s=0 = 0 for all
continuous, piecewise C 2 -variations cs with fixed endpoints, does c then have to be a
156
5.3 The second variation of the energy
geodesic (and thus in particular C ∞ )?

b
Answer: Yes. Namely: First of all, consider only such varia-

tions with ξ(ti ) = 0 for all i ∈ {0, . . . , N }, then it follows as
∇
in the proof of Corollary 2.6.10 that dt ċ ≡ 0 on every [ti−1 , ti ]
for i = 1, . . . , N . b b
⇒ The curve c is piecewise a geodesic. b
If ċ(t− +
i ) 6= ċ(ti ) for an i ∈ {1, . . . , N − 1} then we can choose an η ∈ Tc(ti ) M with
η, ċ(t− +
i ) − ċ(ti ) > 0.
Now continue η via parallel transport along c. Choose a smooth function ϕ : R → R

with ϕ(ti ) = 1 and ϕ ≡ 0 on R \ (ti−1 , ti+1 ). Set ξ(t) := ϕ(t)η(t). Then we have
ξ(a) = ξ(b) = 0 and thus
d
0= E[cs ]|s=0 = ξ(ti ), ċ(t− + − +
i ) − ċ(ti ) = η, ċ(ti ) − ċ(ti ) > 0.
ds
This is a contradiction. We summarize:
Theorem 5.3.1
Let M be a semi-Riemannian manifold and c : [a, b] → M be a continuous, piecewise C 2 -
curve. Then for every continuous piecewise C 2 -variation cs of c with variational field ξ we
have
Zb b N −1
d ∇ X
E[cs ] =− ξ, ċ dt + hξ, ċi + ξ(ti ), ċ(t− +
i ) − ċ(ti ) ,
ds s=0 dt a
a i=1
where a = t0 < t1 < · · · < tN = b is a partition for which both c and cs are C 2 on the
intervals [ti−1 , ti ], i = 1, . . . N .
The curve c is a geodesic if and only if for all such variations with fixed endpoints we have
d
E[cs ] = 0.
ds s=0
To investigate the minima of the energy, we have to consider the second variation of
the energy.
157
Theorem 5.3.2 (Second Variation of the energy)

Let M be a semi-Riemannian manifold. Let c : [a, b] → M be a geodesic. Let cs be a C 3 -
variation of c with variational field ξ and fixed endpoints. Then we have
Zb
d2 ∇ ∇
E[cs ] = ξ, ξ − hR(ξ, ċ)ċ, ξi dt.
ds2 s=0 dt dt
a
Proof. In the proof of Theorem 2.6.5 we have already shown that

Zb
d ∇ ∂cs ∂cs
E[cs ] = , dt
ds ∂t ∂s ∂t
a
holds for all s, not just for s = 0. Therefore
Zb
d2 ∇ ∇ ∂cs ∇ ∇ ∂cs
E[cs ] = , ċ + ξ, dt
ds2 s=0 ∂s ∂t ∂s s=0 dt ∂s ∂t s=0
a
Zb Zb Zb
∇ ∇ ∂cs ∇ ∇
= , ċ dt + hR(ξ, ċ)ξ, ċi dt + ξ, ξ dt.
∂t ∂s ∂s s=0 dt dt
a a a
The assertion follows from

Zb Zb * +!
∇ ∇ ∂cs ∂ ∇ ∂cs ∇ ∂cs ∇
, ċ dt = , ċ − , ċ dt
∂t ∂s ∂s s=0 ∂t ∂s ∂s s=0 ∂s ∂s dt
s=0 |{z}
a a
=0
b
∇ ∂cs
= , ċ = 0,
∂s ∂s s=0 a
because cs is a variation with fixed endpoints.
5.4 The Bonnet-Myers theorem
Definition 5.4.1. Let M be a connected Riemannian manifold. Then we call

diam(M ) := sup{d(p, q) | p, q ∈ M } ∈ (0, ∞]
the diameter of M .
158
Example 5.4.2. For M = S n equipped with the standard metric g = gstd we have
diam(S n ) = π. For M = Rn with the Euclidean metric g = geucl and for hyperbolic
space M = H n with g = ghyp we have diam(Rn ) = diam(H n ) = ∞.
Remark 5.4.3. If M is complete then
diam(M ) < ∞ ⇔ M is compact.
Namely:
“⇐”: M is compact ⇒ M × M is compact ⇒ d : M × M → R is bounded and attains

its maximum C ⇒ diam(M ) = C < ∞.
“⇒”: If diam(M ) =: R < ∞, then for arbitrary p ∈ M we have M = B̄(p, R). Hence
M is compact by the Hopf-Rinow Theorem 5.2.2.
Theorem 5.4.4 (Bonnet-Myers)

Let M be a complete connected Riemannian manifold of dimension n. Assume there exists a
κ > 0 such that
ric ≥ κ(n − 1)g.
This means that ric(ξ, ξ) ≥ κ(n − 1)g(ξ, ξ) for all ξ ∈ T M . Then M is compact and we
have:
π
diam(M ) ≤ √ .
κ
Example 5.4.5
(1) Let M = S n with g = α2 · gstd where α is a positive constant. Then we have
1 n−1
diam(M ) = α π, K≡ , ric ≡ g
α2 α2
π 1
⇒ diam(M ) = √ and ric = κ(n − 1)g with κ = .
κ α2
This shows that the estimate in the Bonnet-Myers theorem is optimal and cannot be
improved.
(2) Now let M = RPn with g = gstd . Since RPn is locally isometric to S n , we have as for
the sphere ric = (n − 1)g. But diam(RPn ) = π2 . Here we find diam(M ) < √πκ where
κ = 1.
159
Proof of Theorem 5.4.4. Let p, q ∈ M with p 6= q. Set δ := d(p, q). Since M is complete,
there exists a minimal geodesic from p to q by the Hopf-Rinow theorem. W.l.o.g. let
γ : [0, δ] → M be parametrized by arc-length with γ(0) = p and γ(δ) = q.
Let e ∈ Tp M with e ⊥ γ̇(0) and ||e|| = 1. Let e(t) be the vector field along γ obtained
from e by parallel transport. Set
π
ξ(t) := sin t · e(t). ξ
δ
Let γs (t) be a variation of γ with fixed endpoints and vari- b
ational field ξ, for example e q

γ
γs (t) = expc(t) (s · ξ(t)).
b
p
Since γ is a minimal geodesic, we have
d
0= E[γs ]|s=0
ds
and d2
0 ≤ 2 E[γs ]|s=0
ds
Zδ !
∇ 2
= ξ − hR(ξ, γ̇)γ̇, ξi dt
dt
0
Zδ π π 2
π 2
= cos t e(t) − sin t hR(e, γ̇)γ̇, ei dt
δ δ δ
0
Zδ
π2 π 2 π 2
= cos t · 1 − sin t K(e, γ̇) dt.
δ2 δ δ
0
If e1 , . . . , en−1 is a orthonormal basis of γ̇(0)⊥ , we obtain with e = ei and summation

over i:
Zδ
π2 π 2 π 2
0≤ (n − 1) 2 cos t − sin t ric(γ̇, γ̇) dt
δ δ δ | {z }
0 ≥ (n−1)κ·1
Zδ π 2
π2 π 2
≤ (n − 1) cos t − sin t · κ dt
δ2 δ δ
0
1 π 2 − κδ2
= (n − 1) · .
2 δ
Therefore 0 ≤ π 2 − κδ2 and hence δ ≤ √πκ . Since this holds for all choices of p and q we
conclude
π
diam(M ) ≤ √ .
κ
160
By Remark 5.4.3, M is compact.
The theorem tells us that the larger the Ricci curvature of a Riemannian manifold, the
smaller the manifold.
Note that the following general implications hold:
K≥κ ⇒ ric ≥ (n − 1) κ · g ⇒ scal ≥ n(n − 1) κ. (1)
Thus the Bonnet-Myers theorem also holds if the sectional curvature is bounded from
below by a postive constant, K ≥ κ > 0. Does the Bonnet-Myers theorem also hold
under the weaker condition scal ≥ n(n − 1)κ?
The answer is “no” as we see by the following counterex- M2
ample. If M1 and M2 are Riemannian manifolds and if b
p
(p1 , p2 ) 2
b
M := M1 × M2 carries the product metric, then

T(p1 ,p2 ) M
gM (ξ1 + ξ2 , η1 + η2 ) = gM1 (ξ1 , η1 ) + gM2 (ξ2 , η2 ).
| {z }
b
∈ Tp1 M1 ⊕Tp2 M2
p1 M1
=T(p1 ,p2 ) M

RM1 (ξ1 , η1 ) 0
⇒ RM (ξ 1 + ξ2 , η1 + η2 ) = M
0 R (ξ2 , η2 )
2

M ricM1 0
⇒ ric =
0 ricM2
⇒ scalM = scalM1 + scalM2 .
For n ≥ 3 we obtain with M = S n−1 × R that
scalM = (n − 1)(n − 2) + 0 = (n − 1)(n − 2),
but diam(M ) = ∞. Thus the Bonnet-Myers theorem does not hold under the weaker
condition scal ≥ n(n − 1)κ if n ≥ 3.
For n = 2 on the other hand, the three conditions in (1) are equivalent.
161
Bibliography
[M65] J. M ILNOR : Topology from the differential viewpoint. University Press of Virginia,
Charlottesville 1965.
[ON83] B. O’N EILL : Semi-Riemannian geometry. Academic Press, New York 1983.
[S96] T. S AKAI : Riemannian geometry. Amer. Math. Soc., Providence 1996.
[W83] F. W. WARNER : Foundations of Differential Manifolds and Lie Groups. Springer-

Verlag, New York-Berlin-Heidelberg 1983.
163
Index
A, atlas, 12 H14 (r), anti-deSitter spacetime, 128

Amax , maximal atlas, 13, 14 n (r), pseudo-hyperbolic space, 127
Hk−1
B(p, r), 148 II, second fundamental form, 117
B̄(p, r), 148 Index(g), 36
CPn , complex-projective space, 11 injrad(p), injectivity radius in p, 148
C k (U ), 24 Isom(M, g), isometry group of M , 44
C k (U, T M ), 47 K, Gauß-curvature, 91
Cp∞ , 24 K(E), sectional curvature, 90, 91
cκ , generalized cosine function, 101 L[c], length of the curve c, 143
ċ(0), 17 Lr , 137
cs (t), 64, 97 Mnκ , model space, 133
c(s, t), variation of the curve c, 64 Mκ , 134
Der(Cp∞ ), 25 M̂κ , 134
Dp , 72 Np M , normal space, 114
∂, 24 ∇, connection, 48
∂ċ(0) f , directional derivative of f , 23 ∇η
, 57
∂y l
df |p , differential of f at the point p, 20 ∇η
diam(M ), diameter, 158 dt , 57
dx|p , 18 ∇ξ η, 55
E[c], energy of the curve c, 63 ν, 124
ε, Signum, 123 O, topology, 1
expp (ξ), exponential map, 72 Ωp,q (M ), 67
Fix(ψ), fixed set of ψ, 70 Pc,t0 ,t1 , parallel transport, 61
g, metric, 35, 38 ϕ∗ g, pullback of g, 44
G2 (M, g), 91 Q, 88
gij , coefficients of g, 35, 38 R, Riemannian curvature tensor, 81
gMink , Minkowski metric, 43 RPn , real-projective space, 7
ghyp , 128 (Rn+1 , gMink ), Minkowski space, 43
GK (V, h·, ·i), 88 Ric, Ricci-tensor, 95
gstd , standard metric of S n , 42 ric, Ricci curvature, 92
Gκ , 135 ricij , 93
Γkij , Christoffel symbols, 49 S(p, r), 148
Gκ , 135 S14 (r), deSitter spacetime, 128
gradf , 124 Skn (r), pseudosphere, 127
H n , hyperbolic space, 128 Sν , Weingarten map, 125
165
Index
sκ , generalized Sinfunction, 101 Cosine Rule for Sides, 140

scal, Scalar curvature, 96 Cosines, Law of, 140
T M , tangent bundle, 31 cotangent space, 39
Tp M , tangent space of M in the point p, countable basis, 2
18 covariant derivative, 55
Tp∗ M , cotangent space, 39 covariant derivative of a vector field, 53
Tp U , 20 covariant derivative, second, 79
tan, 114 critical point, 67, 109
V ∗ , 36 critical value, 109
Ξ(M ), 54 curvature tensor, Riemann, 81
Ξp , 47
ξ, variational vector field, 64 d(p, q), distance of the points p and q,
h·, ·iκ , 134 144
h·, ·i, 118, 143 derivation, 24
hh·, ·ii, Minkowski-scalar produkt, 43 deSitter spacetime, 128
| · |, 123 diameter of a Riemannian manifold, 158
diffeomorphic, 16
Angles, Cosine Rule for Angles, 140 diffeomorphism, 16
anti-deSitter spacetime, 128 diffeomorphism, local, 21
atlas, 12 differentiable structure, 14
atlas, maximal, 14 differential of a map, 20
directional derivative, 23
Bianchi-identity, first, 82
distance, Riemannian, 144
bilinear form, 35
double cone, 3
Bonnet-Myers, Theorem of, 159
by arc-length, parametrized, 69 Einstein summation convention, 30
Einstein tensor, 129
Chain Rule, 21
energy functional
chart, 4
critical point, 67
Christoffel symbols, 49
energy of a curve, 63
C ∞ -compatibility of charts, 12
exponential map, Riemannian, 72
C ∞ -atlas, 12
C k -diffeomorphism, 16 first fundamental form, 41
C k -map, 14 fixed point set, 70
closed, 1 flat manifold, 84
codimension, 107 fundamental form, first, 41
complete manifold, 151 fundamental form, second, 117
complex-projective space, 11
conjugated points, 104 Gauß curvature, 91
connected, 5 Gauß formula, 118
continuous map, 1 Gauß formula for hypersurfaces, 126
contravariant vector, 30 Gauß lemma, 144
coordinate system, local, 4 generalized cosine function, 101
cosine function, generalized, 101 generalized orthonormal basis, 36, 74
Cosine Rule for Angles, 140 generalized sine function, 101
166
Index
geodesic, 68 manifold, (semi-)Riemannian, 42

geodesic polar coordinates, 145 manifold, differentiable, 14
geodesic triangle, 138 manifold, Riemannian, 42
geodesic variation, 97 manifold, topological, 1
geodesic, minimal, 150 map, differentiable, 14
geodesically complete, 151 maximal atlas, 14
geodesics metric, Lorentz-, 42
existence and uniqueness of, 69 metric, Riemannian, 42
Gradient of a differentiable function, metric, semi-Riemannian, 38
124 minimal geodesic, 150
Minkowski scalar product, 43
homeomorphic, locally, 2 Minkowski space, 43
homeomorphism, 1 model space, 133
Hopf-Rinow, Theorem of, 151 Modelspace of constant curvature, 133
hyperbolic space, 128 motion, Euclidean, 45
hypersurface, semi-Riemannian, 123
non-degenerate bilinear form, 35
index, 36 non-degenerate subvector space, 88
injectivity radius, 148 normal coordinates, Riemannian, 75
inverse function theorem, 22 normal space, 114
isometry, 44 null curve, 69
isometry group, 44
open, 1
isometry, linear, 37
orthonormal basis, generalized, 36, 74
isometry, local, 44
parallel transport, 61
Jacobi equation, 98 parallel transport on the sphere, web-
Jacobi field, 98 site, 63
interesting, 100 parallel vector field along a curve, 58
uninteresting, 99 parametrization of a curve
by arc-length, 69
Law of Cosines, 140 by proper time, 69
Law of Sines, 140 proportional to arc-length, 69
length of a curve, 143 proportional to proper time, 69
Levi-Civita connection, 48 Poincaré transformations, 45
lightcone, 89 polar coordinates, 33
local coordinate system, 4 polar coordinates, geodesic, 145
local diffeomorphism, 21 proper time, 69
locally homeomorphic, 2 pseudo-hyperbolic space, 126, 127
Lorentz boost, 137 pseudo-sphere, 127
Lorentz group, 130 Pseudosphere, 126
Lorentz manifold, 42 pullback of a metric, 44
Lorentz metric, 42
Lorentz transformations, time- real-projective space, 7
orientation preserving, 135 regular point, 109
167
Index
regular value, 109 velocity field, 55

Ricci curvature, 92 vertex of a geodesic triangle, 138
Riemann curvature tensor, 81
Riemannian (4, 0)-curvature tensor, 85 Weingarten map, 125
Riemannian distance, 144
Riemannian exponential map, 72
Riemannian metric, 42
Riemannian normal coordinates, 75
Sard, theorem of, 110

scalar curvature, 96
second covariant derivative, 79
second fundamental form, 117
sectional curvature, 91
semi-Riemannian metric, 38
side of a geodesic triangle, 138
signature of a hypersurface, 123
sine function, generalized, 101
Sines, Law of, 140
standard metric, 42
stereographic projection, 3
submanifold chart, 107
submanifold, differentiable, 107
submanifold, semi-Riemannian, 113
submanifold, totally geodesic, 121
symmetric bilinear form, 35
tangent bundle, 31
tangent space, 18
tangent vector, 17
topological space, 1
topology, 1
torsion freeness, 50, 56
torsion-free, 48
totally geodesic submanifold, 121
transformation of principal axes, 36
triangle, geodesic, 138
variation of a curve, 64
variation of the energy, second, 157
variation of the energy, first, 64
variation, geodesic, 97
variational vector field, 64
vector field, 33
vector field along a map, 54
168

Differential Geometry - Christian Bär

Uploaded by

Copyright:

Available Formats

Differential Geometry - Christian Bär

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Differential Geometry - Christian Bär

Uploaded by

Copyright:

Available Formats

Christian Bär

Version of August 26, 2013

5 Riemannian Geometry 143

Potsdam, August 2013

Reminder. Let M be a set. A system of sets O ⊂ P(M ) is called a topology on M , if

A subset U ⊂ M is called open in M if U ∈ O. A subset A ⊂ M is called closed if

Definition 1.1.2. Let M be a topological space with topology O. Then M is called an

3. M is locally homeomorphic to Rn , that is, for all p ∈ M exists an open subset

Example 1.1.4. (1) Euclidean space M = Rn itself is an n-dimensional topological ma-

(iii) Holds true with U = M , V = Rn and x = id.

(2) Let M ⊂ Rn be an open subset. Then M is an n-dimensional topological manifold.

(i), (ii) Obvious.

(iii) Holds true with U = M , V = M and x = id.

(3) The standard sphere M = S n = { y ∈ Rn+1 : kyk = 1 } is an n-dimensional topo-

(i), (ii) Obvious, since S n is a subset of Rn+1 .

The map x is continuous and bijective. The inverse map y is given by

and is also continuous. Hence, x is an homeomorphism.

Analogously, we define the homeomorphism, which omits the north pole:

(4) All n-dimensional submanifolds of RN in the sense of Analysis 3 are n-dimensional

(5) Non-Example. We consider M := { (y 1 , y 2 , y 3 ) ∈ R3 | (y 1 )2 = (y 2 )2 + (y 3 )2 }, the

it were, then there would exist an open ≈ c x(0)

subset U ⊂ M with 0 ∈ U , an open subset b

W. l. o. g. assume V = Br (x(0)) with r > 0. Choose q1 , q2 ∈ U with q11 > 0 and

Define the continuous path c̃ := x−1 ◦ c : [0, 1] → U . Then

Definition 1.1.5. If M is an n-dimensional topological manifold, the homeomor-

After choosing a local coordinate system x : U → V every point p ∈ U is uniquely

In a 0-dimensional manifold M every point p ∈ M has an open neighborhood U , which

Remark 1.1.8. Following Proposition 1.1.6 every connected 0-dimensional manifold M

In dimension 1 there are only a few connected manifolds:

(ii) xα (Uα ∩ Uβ ) ⊂ Rn is open for all α, β ∈ A and

(iii) xβ ◦ xα −1 : xα (Uα ∩ Uβ ) → xβ (Uα ∩ Uβ ) is continuous for all α, β ∈ A.

Proof. We first show uniqueness:

O = {W ⊂ M | xα (W ∩ Uα ) ⊂ Rn open for all α ∈ A}.

Now we show existence:

O := {W ⊂ M | xα (W ∩ Uα ) ⊂ Rn open for all α ∈ A}.

(a) O is a topology because

is open in Rn for all α ∈ A.

(b) We have to show Uβ ∈ O for all β ∈ A. This is obvious because xα (Uβ ∩ Uα ) ⊂ Rn

(c) The map xβ is continuous for all β ∈ A because:

xα (xβ −1 (Y ) ∩ Uα ) = xα (xβ −1 (Y ∩ xβ (Uα ∩ Uβ )))

(d) The map xβ −1 is continuous because:

Example 1.1.11 (Real-projective space). We define the real-projective space by

M = RPn := P(Rn+1 ) := {L ⊂ Rn+1 | L is one-dimensional vector-subspace }.

A := {affine-linear embeddings α : Rn → Rn+1 with 0 6∈ α(Rn )}.

for all x ∈ Rn . Since α is an embedding, B has maximal rank, rank(B) = n.

Consequently, α(Rn ) is an affine-linear hyperplane. Set

Uα := {L ∈ RPn | L ∩ α(Rn ) 6= ∅}.

RPn \ Uα = {L | L ⊂ B(Rn ) one-dimensional subspace} (1.1)

where α(x) = Bx + c. For α ∈ A set Vα := Rn and

x α : U α → Vα , xα (L) := α−1 (L ∩ α(Rn )).

Then xα is a bijective map and we have

In the following we check the assumptions of Theorem 1.1.10:

To this end, let e0 , . . . , en ∈ Rn+1 be the standard basis. For j = 1, . . . , n we define:

αj (v) := v 1 e0 + . . . + v j ej−1 + ej + v j+1 ej+1 + . . . + v n en .

Assume there existed an

(ii) We observe that xα (Uα ∩ Uβ ) is α(Rn )

(iii) We show that xβ ◦ xα −1 : v 7→ β −1 (R · α(v) ∩ β(Rn )) is continuous for all α, β ∈ A.

Write α(v) = Bv + c and β(w) = Dw + f . Now w = xβ ◦ xα −1 (v) is equivalent

is equivalent to the existence of t ∈ R such  Dw + f = t · (Bv + c). For the

Proposition 1.1.12 (First Addition to Theorem 1.1.10)

is equivalent to the existence of t ∈ R such Dw + f = t · (Bv + c). For the