Ludwig-Maximilians-Universität München

Prof. Dr. Thomas Vogel

Lecture in the summer term 2017

Riemannian geometry
Please note: These notes summarize the content of the lecture, many details and
examples are omitted. Sometimes, but not always, we provide a reference for proofs,
examples or further reading. Some proofs were done in two lectures although they
appear in a single lecture in these notes. Changes to this script are made without
further notice at unpredictable times. If you find any typos or errors, please let me
know.

1. Lecture on May, 4 – Geodesics and the exponential map

• Let (M, g) be Riemannian and ∇ the Levi-Civita connection. A smooth curve
∇
γ in M is a geodesic if dt γ̇(t) = 0 and γ is not constant. It is common to write
∇γ̇ γ̇ = 0.
• Example: Let M ⊂ (Rn , gstandard ) be a submanifold with the induced metric.
Then a non-constant curve γ : (a, b) −→ M is a geodesic if and only if γ̈(t) is
orthogonal to Tγ(t) M .
• Fact: If γ is a geodesic, then |γ̇| is constant.
• Lemma: Let x1 , . . . , xn be coordinates on U and γ : (a, b) −→ U ⊂ M be
smooth. Then γ = (γ 1 , . . . , γ n ) is a geodesic if and only if
X
γ̈ k (t) + γ̇ i (t)γ̇ j (t)Γkij (γ(t)) = 0.
i,j

• Proof: Direct computation in coordinates.

• Fact: Standard ODE theory implies that this second order differential equa-
tion has maximal solutions γ(t, p, v) which are uniquely determined by initial
conditions γ(0) = p and γ̇(0) = v ∈ Tp M . The solutions depend smoothly on
initial conditions and parameters.
It is not true in general that γ exists for all times. However this is the case
when g is positive definite and M is compact. A more general statement is the
theorem of Hopf-Rinow below.
• Definition/Theorem: There is a unique vector field on T M whose trajectory
through v ∈ Tp M is
t 7−→ (γ(t, p, v), γ̇(t, p, v)).
The flow of this vector field is the geodesic flow. Its flow preserves level sets of
k · k2 on T M .
• Remark: If γ : (−δ, δ) −→ M is a geodesic, then the same is true of γa :
(−δ/2, δ/a) −→ M defined by γa (t) = γ(at) and γ(t, p, av) = γa (t, p, v).
• Definition/Theorem: Let p ∈ (M, g). Then there is a neighbourhood U ⊂
Tp M of 0 such that
expp : Tp M ⊃ U −→ M
v 7−→ γ(1, p, v)
is well defined and smooth. This map is called the exponential map at p.
2

• Theorem: There is a neighbourhood U ⊂ Tp M such that expp is defined on U

and expp : U −→ expp (U ) is a diffeomorphism.
• Proof: By definition of expp the differential D0 expp : T0 Tp M ' Tp M −→ Tp M
is the identity. The theorem follows from the inverse function theorem.
• Definition: Let A ⊂ R2 be open and f : A −→ M be smooth. A vectorfield
along f is a smooth map V : A −→ T M such that V (s, t) ∈ Tf (s,t) M . We will
∇
write ∂s V for the covariant derivative of the vector field V (·, t = t0 ) along the
curve f |{t=t0 } etc.
• Lemma: In the present situation

∇ ∂f ∇ ∂f
(1) = .
∂s ∂t ∂t ∂s

• Proof: Near f (s, t) one chooses local coordinates and expresses V and f, ∂f , ∂f
∂s ∂t
in terms of these coordinates. Since ∇ is assumed to be torsion free, ∇∂i ∂j =
∇∂j ∂i for all i, j. Using this the claim follows.
• Gauß-Lemma: Let v ∈ Tp M and w ∈ Tv Tp M ' Tp M orthogonal to v. Then




(2) g Dv expp (w), Dv expp (v) = 0.

• Proof: For ε > 0 small enough the map

f : (−ε, ε) × [0, 1] −→ M
(s, t) 7−→ expp (t(v + sw))

is well defined and smooth. We want to compute

 
∂f ∂f



g , (0, 1) = g Dv expp (w), Dv expp (v) .
∂s ∂t

Because g is a metric connection, γs (·) = f (s, ·) is a geodesic and by (1)

   
∂ ∂f ∂f ∇ ∂f ∂f  ∂f ∇ ∂f 
g , (0, t) = g , (0, t) + g , (0, t)
∂t ∂s ∂t ∂t ∂s ∂t ∂s |∂t{z∂t}
=0
since γs is geodesic
 
∇ ∂f ∂f
=g ,
∂s ∂t ∂t
 
1 ∂ ∂f ∂f
= g ,
2 ∂s ∂t ∂t
= 0( when s = 0)

∂f ∂f


The last equality holds since kγ̇s k2 = (kvk2 + s2 kwk2 ). Hence g ,
∂s ∂t
(0, t) is
independent of t. For t = 0 we get
 
∂f ∂f
g , (0, 0) = 0.
∂s ∂t
 3

2. Lecture on May, 8 – Minimizing properties of geodesics

• Proposition: Let p ∈ M and δ > 0 so that expp is defined on a neighbourhood
of Bδ (0). Let v ∈ Tp M be a unit vector and
γ : [0, δ] −→ M
t 7−→ expp (tv) = γ(1, p, tv).

The for every other piecewise smooth path c in M from p to γ(δ)

l(c) ≥ l(γ) = δ
with equality if and only of c is a reparametrisation of γ.
• Proof: It is enough to consider curves c whose image is contained in exp(Bδ (0))
and such that c−1 (p) = 0. For t 6= 0 there are unique functions which are smooth
(because expp is a local diffeomorphism)
r : (0, 1] −→ (0, δ]
α : (0, 1) −→ ∂B1 (0) ⊂ Tp M

with the property c(t) = expp (r(t)α(t)). Whenever c is smooth

2
kċ(t)k2 = Dr(t)α(t) expp (ṙ(t)α(t) + r(t)α̇(t))
2
= (ṙ(t))2 + Dr(t)α(t) expp (r(t)α̇(t))
≥ (ṙ(t))2

with equality if and only if α̇(t) = 0. The second equality uses the Gauß-Lemma.
For all 0 < ε < 1
Z 1 Z 1
l(c|[ε,1] ) = kċ(t)kdt ≥ |ṙ(t)|dt
ε ε
Z 1
≥ ṙ(t)dt = r(1) − r(ε) → δ = l(γ)
ε

when ε → 0. It is now clear how to characterize the equality case.

• Definition: Let (M, g) be a connected Riemannian manifold. The metric in-
duced by g is
d(p, q) = inf{l(c) | c is a piecewise smooth path from p to q}.
The previous proposition implies that d(p, q) = 0 if and only if p = q. The
other properties of a metric are easily verified.
• Fact: By the proposition a metric δ-ball around p ∈ M is the diffeomorphic
image of a δ-ball in Tp M under the exponential map for sufficiently small δ > 0.
This justifies the notation Bδ (p) = expp (Bδ (0)) and shows that the topology
induced by the metric coincides with the manifold topology of M . In particular,
d(p, ·) respectively d(·, ·) define continuous functions on M respectively M ×M .
Moreover, for all q ∈ Bδ (p) with δ sufficiently small there is a unique piecewise
smooth curve from p to q whose length is d(p, q) and this curve is a geodesic.
We did not yet show that curves with minimal length connecting two points
in (M, g) are always reparametrisations of geodesics.
4

3. Lecture on May, 11 – Convex neighbourhoods

• Theorem: For each p ∈ M there is δ > 0 and a neigbourhood W such that for
all q ∈ W the exponential map expq is defined on Bδ (0) ⊂ Tq M and expq (Bδ (0))
contains W .
For all p, q in W there is a unique shortest geodesic from p to q (it is shorter
than δ) and it depends smoothly on p, q.
• Proof: Let δ 0 > 0 so that expp is defined on B2δ0 (0). There is a neighbourhood
U of p such that for all q ∈ U the exponential map expq is defined on Bδ0 (0).
We denote v ∈ Tq M by (q, v) to keep track of base points. One applies the
inverse function theorem to the map
[
ψ: (Bδ0 (0) ⊂ Tq M ) −→ M × M
q∈U

(q, v) 7−→ (q, expq (v)).

By definition D(p,0) ψ(v, 0) = (v, v) and D(p,0) ψ(0, w) = (0, w). Hence ψ is a local
diffeomorphism of a neighbourhood V ⊂ T M of (p, 0) onto a neighbourhood of
(p, p) in M × M . There is a number 0 < δ < δ 0 and a neighbourhood V 0 of p in
M such that
[
(Bδ (0) ⊂ Tq M ) ⊂ V ⊂ T M.
q∈V 0

Let W be neighbourhood of p so that W × W is contained in

!
[
ψ (Bδ (0) ⊂ Tq M ) .
q∈V 0

Then δ and W have the desired properties.

• Corollary: Let c be a piecewise smooth curve from p to q in M which is
parametrized by arc length so that l(c) = d(p, q). Then c is a geodesic.
• Proof: If c is not smooth, then this curve can be shortened using the previous
theorem. The same is true if c is not a geodesic.
• Proposition: Let p ∈ M . There is a number δ > 0 such that for all 0 < δ 0 < δ
and every geodesic γ : (−ε, ε) −→ M which is tangent to ∂Bδ0 (p) in γ(0) maps
a neighbourhood of 0 to a set which meets Bδ0 (p) in γ(0) only.
• Proof: Let δ 00 > 0 and W be a neighbourhood with the properties from the pre-
vious theorem and 0 < δ 0 < δ 00 . We will assume that γ is a geodesic parametrized
by arc length which is tangent to ∂Bδ0 (p) in γ(0). Let Fγ (t) = k exp−1 2
p (γ(t))k .
Then
 
dFγ d −1 −1
(t) = 2gp expp (γ(t)), expp (γ(t))
dt dt
=0

by the Gauß-Lemma. Thus 0 is a critical point of Fγ when γ̇(0) is tangent to

∂Bδ0 (p).
We now consider Fγ for all geodesics γ parametrized by arc length with
γ(0) ∈ Bδ (p) and abbreviate uγ (t) = exp−1
p (γ(t)). If γ is a geodesic with γ(0) =
 5

p then uγ (0)
2
d2 Fγ d2
 
duγ
(0) = 2gp uγ (0), uγ (0) + 2 (0)
dt2 dt2 dt
2
duγ
=2 (t) = 2.
dt
Recall that the germ of a geodesic γ on (M, g) is uniquely determined γ(0)
and γ̇(0) ∈ Tγ(0) M and that the space of tangent vectors of length 1 in Tq M is
compact for all q. Thus there is 0 < δ < δ 0 such that
d2 F γ
(0) > 0
dt2
for all geodesics (parametrized by arc length) such that d(p, γ(0)) < δ.
• Corollary: For all p ∈ M there is δ > 0 such that for 0 < δ 0 < δ the δ 0 -ball
around p is strictly convex, i.e. for all p, q in Bδ0 (p) there is a unique minimal
geodesic from p to q which is contained in Bδ0 (p).
• Proof: Let p ∈ M and δ̂ > 0 the number from the previous proposition.
Then δ = δ̂/2 has the desired property: Let p, q ∈ Bδ0 (p) with δ 0 < δ and
γ : [0, 1] −→ M the unique minimal geodesic from p to q. This geodesic is
shorter than 2δ, so it cannot start at p, leave Bδ0 (p) and return to q. Then the
function d(p, γ(t))2 has a global maximum at t0 ∈ (0, 1). This means that γ is
tangent to the boundary of some ball B0 around p which is contained in Bδ0 (p)
and γ is completely contained in B0 . This is a contradiction, thus γ never leaves
Bδ0 (p).

4. Lecture on May, 15 – Completeness, Theorem of Hopf-Rinow

• Theorem (Hopf-Rinow, first part): Let (M, g) be a connected Riemannian
manifold and p ∈ M such that
expp : Tp M −→ M
is well defined. Then for all q there is a geodesic γ from p to q whose length is
d(p, q).
• Proof idea: To understand the idea, assume that the theorem is true. Let
0 < δ < d(p, q) be so small that Bδ (p) = expp (Bδ (0) ⊂ Tp M ) is geodesically
convex and γ a minimal geodesic from p to q. Then γ intersects ∂Bδ (p) exactly
once, we call the intersection point p0 . Then p0 has to be the point on ∂Bδ (p)
which is closest to q. Thus, to find a candidate for a geodesic from p to q it
is natural to consider the closest point p0 on ∂Bδ (p) from q and to extend the
unique geodesic from p to p0 in the convex ball and then to show that this
extension has the desired properties.
• Proof: Let δ be as above and p0 ∈ ∂Bδ (p) such that d(p0 , q) = d(Bδ (p), q) =
min{d(x, q)|x ∈ Bδ (p)}. Then
(3) d(p, q) = d(p, p0 ) + d(p0 , q)
follows by considering a path from p to q, and decomposing this path at its first
intersection point with ∂Bδ (p) (and the triangle inequality,choice of δ, choice
of p0 ).
Let γ be the radial unit speed geodesic from p to p0 . By assumption γ(t) is
defined for all t. Consider I = {t ∈ [0, d(p, q)] | d(γ(t), q) = d(p, q) − t}. Then
6

δ ∈ I since γ(δ) = p0 . Let 0 < T = sup I. Then T ∈ I by continuity and we

assume T < d(p, q) (otherwise we are done).
Let p0 = γ(T ) and 0 < δ 0 < d(p0 , q) so that expp0 is defined in Bδ0 (0). Consider
p00 on ∂Bδ0 (p0 ) which is closest to q. Then in analogy to (3)
(4) d(p0 , q) = d(p0 , p00 ) + d(p00 , q).
Let γ 0 be the radial unit speed geodesic from p0 to p00 of length d(p0 , p00 ). By (4),
the definition of T and the triangle inequality
d(p, q) = d(q, γ(T )) + T
= d(γ(T ), p00 ) + d(p00 , q) + T
≥ d(p00 , q) + d(p, p00 )
≥ d(p, q).
Hence all inequalities are equalities and the concatination of γ|[0,T ] with the
minimal geodesic from γ(T ) to p00 has length d(p, p00 ). Hence it is a geodesic
and p00 = γ(T + δ 0 ) lies on γ and also in I. But then T is not maximal in I.
This contradiction shows T = d(p, q).
• Definition: (M, d) is geodesically complete if expp is defined on Tp M for all
p ∈ M.
• Theorem (Hopf-Rinow, part 2): The following conditions are equivalent.
(a) expp is defined on Tp M for some p ∈ M .
(b) K ⊂ M is compact if and only if K is closed and bounded.
(c) M is complete as a metric space.
(d) M is geodesically complete.
(e) There are compact sets Kn , n ∈ N such that ∪n Kn = M , Kn+1 ⊃ Kn
and if qn 6∈ Kn , then limn→∞ d(qn , p) = ∞ for some p ∈ M .
• Proof: The proof of this is not very difficult, an important observation is that
if (a) holds, then one can use the first part of Hopf-Rinow. Bn (0) ⊂ Tp M is
compact, so its image under the exponential map is also compact. For notions
like metric completeness and its relationship with compactness one can consult
[?] or [Q].
• Example: In the open unit interval every pair of points is connected by a
geodesic. But it is not complete with the standard metric.

5. Lecture on May, 18 – Riemann curvature tensor

• Definition: Let (M, g) be a Riemannian manifold and ∇ the Levi-Civita
connection. The curvature tensor of (M, g) is
R : Γ(T M ) × Γ(T M ) × Γ(T M ) −→ Γ(T M )
(X, Y, Z) 7−→ R(X, Y )Z := ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z
• Warning: A sign convention which replaces R by −R is common. Depending
on which convention is used some of the definitions of objects induced by R
have to be changed, too.
• Lemma: R is C ∞ (M )-linear in all three variables. In particular, it is a tensor.
• The proof of this is a direct computation. The lemma means that the value of
R(X, Y )Z at p ∈ M depends only on X(p), Y (p), Z(p).
• Lemma: The Riemann curvature tensor has the following algebraic properties:
1. R(X, Y )Z + R(Y, X)Z = 0
 7

2. R(X, Y )Z + R(Y, Z)X + R(Z, X)Y = 0

3. g(R(X, Y )Z, W ) + g(R(X, Y )W, Z) = 0
4. g(R(X, Y )Z, W ) = g(R(Z, W )X, Y ).
• Proof: All identities can be proved by computation. It is convenient to assume
that all vector fields commute with each other (this is possible since R is tensor).
For a nicely organized proof of the last identity see Chapter 9 of [Mi].
• Definition: A trilinear map on a Euclidean vector space which satisfies the
properties of the previous lemma is called a curvature tensor.
• Definition: Let (M, g) be a Riemannian manifold and σ ⊂ Tp M a two-
dimensional subspace. The sectional curvature K(σ) of (M, g) is
g(R(X, Y )Y, X)
K(σ) =
|X|2 |Y |2 − g(X, Y )2
• Example: Let K ∈ R. Then the map
R(X, Y )Z = K(g(Y, Z)X − g(X, Z)Y )
is a curvature tensor whose sectional curvature is K. We did not yet show that
R arises as the curvature tensor of a Riemannian metric, but this is the case
(consider spheres of some radius R, Euclidean space and rescalings of hyperbolic
space).
• Lemma: Let f : R2 ⊃ A −→ M be smooth and V a vector field along f . Then
 
∇∇ ∇∇ ∂f ∂f
(5) V (s, t) − V (s, t) = R , V (s, t).
∂t ∂s ∂s ∂t ∂t ∂s
• Proof: Choose local coordinates near f (s, t) and expand all vector fields along
f in terms of the coordinate basis. Then compute.

6. Lecture on May, 22 – Jacobi fields

• Let γ = expp (tv), t ∈ [0, 1] be a geodesic. The goal is to better understand
Dv expp : Tv Tp M −→ Texpp (v) M.
If v = 0, then D0 expp = IdTp M . Now consider a curve w : (−ε, ε) −→ Tp M
representing w ∈ Tv Tp M and
f : (−ε, ε) × [0, 1] −→ M
(s, t) 7−→ expp (tw(s)).
∇ ∂f
Because f (s, ·) parametrizes a geodesic for fixed s we have ∂t ∂t
= 0 for all s,
hence
∇ ∇ ∂f
= 0.
∂s ∂t ∂t
Using (5) and (1) this implies the Jacobi equation for J(t) = ∂f
∂s
(0, t) along γ:
∇∇
J(t) + R(J(t), γ̇(t))γ̇(t) = 0.
∂t ∂t
This is a second order linear differential equation for J, in particular J is
˙
determined by J(0) and J(0), the domain of maximal solutions coincides with
the domain of γ and the solutions of the Jacobi equation form a vector space
whose dimension is 2dim(M ), its solutions are called Jacobi fields.
• Note: The proof of the Gauß-Lemma is based on a similar idea.
8

• Remark: It is important to note that in the derivation of the Jacobi equation

we used only the fact that f (s, ·) is a geodesic for fixed s, we did not use
f (s, 0) = 0 for all s. Thus
• Lemma: Let γs : [0, a] −→ M be a smooth family of geodesics. Then J(t) =
∂f
∂s
(0, t) is a Jacobi field along γ0 .
• Lemma: Conversely, let γ : [0, a] −→ M be a geodesic and J a Jacobi field
along γ. Then there is a smooth family of geodesics γs such that γ0 = γ and
J(t) = ∂γ s
∂s s=0
(t).
• Proof: Let U ⊂ M be a geodesically convex neighbourhood of p, i.e. for all
q, q 0 ∈ U there is a unique geodesic from q to q 0 which is contained in U (this
geodesic has length d(q, q 0 )). Choose δ > 0 so small that γ([0, δ]) ⊂ U and
curves α0 , αδ in U representing the tangent vectors J(0), J(δ). Let γs be the
unique geodesic such that γs (0) = α0 (s) and γs (δ) = αδ (s). Then γ0 = γ and
this is a smooth family of geodesics (uniqueness of γs and the first theorem in
Lecture 3). For |s| sufficiently small γs can be extended to [0, a] and γ0 = γ.
The variation vector field ∂γ s
∂s s=0
(t) is a Jacobi field and it satisfies
∂γs ∂γs
(0) = J(0) (δ) = J(δ).
∂s s=0 ∂s s=0
We have shown that
ev : {Jacobi fields along γ} −→ Tγ(0) M × Tγ(δ) M
H 7−→ (H(0), H(δ))
γs
is surjective. For dimension reasons ev is also injective, hence ∂s (s = 0, t) =
J(t).
• Example: If γ is a geodesic, then γ̇(t) and tγ̇(t) are Jacobi fields along γ.
• Fact: g(J(t), γ̇(t)) is an affine function in t when J is a Jacobi field along the
geodesic γ. Adding multiples of γ̇(t), tγ̇(t) to J one can obtain Jacobi fields
which are everywhere orthogonal to γ.
• Example: Let M, g be a manifold of constant sectional curvature K (such
space exist), γ a geodesic, W a parallel vector field along γ which is orthogonal
to γ with |W (t)| ≡ 1. Then the following vector fields are Jacobi fields along γ:

 tW (t)
 √
if K = 0
sin(t K)
(6) J(t) = √
K
W (t) if K > 0
√
 sinh√t −K W (t) if K < 0.

−K

7. Lecture on May, 29 – Conjugate points, Theorem of

Hadamard-Cartan
• Definition: Let γ : [0, a] −→ M be a geodesic and 0 < t0 ≤ a. Then γ(t0 ) is
conjugate to γ(0) along γ if there is a non-trivial Jacobi field J along γ such
that J(0) = J(t0 ) = 0. The multiplicity of γ(t0 ) as a conjugate point is the
dimension of the vector space of such Jacobi fields.
• Remark: Because J(t) = tγ̇(t) is a Jacobi field along γ which vanishes for
t = 0 but nowhere else, the multiplicity of conjugate points is bounded by
dim(M ) − 1. This bound is realized for the round metric on the sphere. If γ(t0 )
is not conjugate to γ(0) along γ then expγ(0) is a local diffeomorphism near
t0 γ̇(0).
 9

• Lemma: Let f : (M, g) −→ (N, h) be a smooth map between connected Rie-

mannian manifolds such that
(i) f is surjective,
(ii) (M, g) is complete (as metric space),
(iii) kDf (v)k ≥ kvk for all v ∈ T M .
Then f is a covering.
• Reminder: If you are not familiar with coverings then you should read Chapter
5 from [Ma] or Chapter 9 from [J]. We will use covering spaces in several places.
• Proof: We show that f has the unique path lifting property for all starting
points, i.e. for all smooth γ : [0, 1] −→ N and q ∈ f −1 (γ(0)) there is a unique
smooth map γ e : [0, 1] −→ M such that γe(0) = q and f ◦ γ
e = γ.
The assumptions imply that for all initial points γ(0) of γ, the set f −1 (γ(0))
is not empty. Let q ∈ f −1 (γ(0)). Moreover, f is a local diffeomorphism ((iii)
implies that Df is injective, since f is surjective it follows that Df is also
surjective everywhere (see for example [BJ], Theorem 5.4). The set

I = {t ∈ [0, 1] | γk[0,t] has the unique path lifting property for initial point q}

is not empty and open. We show that I is closed, hence I = [0, 1]. Let ti be
a monotone sequence converging to sup(I). Let i ≤ j. Since γ(ti ) converges to
γ(sup(I)) it is a Cauchy sequence. Because the path lifts with fixed starting
point are unique the lift of γ|[0,ti ] is the restriction of the lift of γ|[0,tj ] to [0, ti ].
Hence we get a well defined sequence γ e(ti ) in M . Then by (iii)


d(e e(tj )) ≤ length γ
γ (ti ), γ e|[ti ,tj ]


≤ length γ|[ti ,tj ]

is a Cauchy sequence in M which converges because M is complete. Let q∞ =

limi γ
e(ti ) (hence f (q∞) = γ(sup(I))) and U a neighbourhood of q∞ such that
f |U is a diffeomorphism onto its image. For ε > 0 small enough γ([sup(I) −
ε, sup(I) + ε]) ⊂ f (U ). Because f |U is a diffeomorphism onto f (U ) there is
a unique lift of γ|[sup(I)−ε,sup(I)+ε] through a point γ e(ti ) when for all j ≥ i
e(tj )) ∈ U . Thus I is closed. Since [0, 1] is connected I = [0, 1].
γ
• Definition: Let (M, g) be a Riemannian manifold and p ∈ M . p is a pole of no
geodesic γ through p contains a point which is conjugate to p along γ.
• Theorem: Let (M, g) be a complete Riemannian manifold and p ∈ M a pole.
Then expp : Tp M −→ M is a covering.
• Proof: By assumption expp is a local diffeomorphism. Therefore h = exp∗p g is
a Riemannian metric on Tp M . The geodesics through the origin are the rays
through the origin since these get mapped to geodesics in M . In particular,
the exponential map at 0 ∈ Tp M of the Riemannian manifold (Tp M, h) is well
defined. By the theorem of Hopf-Rinow (Tp M, h) is complete! By definition
expp : (Tp M, h) −→ (M, g) is a local isometry. By the previous lemma expp is
a covering.
• Lemma: Let (M, g) be a complete Riemannian manifold such that K(σ) ≤ 0
for all planes σ ⊂ T M . Then every p ∈ M is a pole.
• Proof: Let p ∈ M , γ a geodesic with γ(0) = p and J a non-trivial Jacobi field
along γ with J(0) = 0. Because expp is a local diffeomorphism at p ∈ 0 there is
a neighbourhood of 0 ∈ R where 0 is the only zero of J. In particular, there is
a sequence 0 < ti → 0 such that dtd g(J(ti ), J(ti )) > 0. By the Jacobi equation
10

and since K(σ) ≤ 0

d2 ¨ J(t)) + 2g(J(t), ˙ ˙
g(J(t), J(t)) = 2g(J(t), J(t))
dt2
≥ g(−R(J(t), γ̇(t))γ̇(t), J(t))
≥ 0.
Hence dtd g(J(t), J(t)) > 0 for all t > 0 and J(t) 6= 0 if t 6= 0.
• Corollary: Let (M, g) complete with non-positive sectional curvature and di-
mension n. Then expp : Rn ' Tp M −→ M is a universal covering.
• Corollary: If (M, g) is complete, connected, simply connected and has non-
positive sectional curvature, then for all p, q there is a unique geodesic from p
to q (up to reparametrisation).

8. Lecture on May, 29 – Cartan criterion for local isometries

• We consider the following setting. Let (M, g), (M , g) be Riemannian manifolds
of the same dimension n, p ∈ M, p ∈ M and I : Tp M −→ Tp M ) an isometry.
R respectively R is the curvature tensor of g respectively g.

φ = expp ◦I exp−1
p
defines a local diffeomorphism of a neighbourhood of p to a neighbourhood of
p.
For a geodesic γ : [0, a] −→ M we write Pγ for the parallel transport
Tγ(0) M −→ Tγ(t) M along γ, Pγ has the analogous meaning for geodesics γ
in γ. Given a geodesic γ through p we define γ as the unique geodesic through
p with γ̇(0) = I(γ̇(0)). For t ∈ [0, a] let
Iγ = Pγ ◦ I ◦ Pγ−1 : Tγ(t) M −→ Tγ(t) M .
Strictly speaking one should write Iγ (t), but enough is enough.
• Theorem (Cartan): φ is a local isometry if and only if for all geodesics
γ : [0, t] −→ M starting at p and contained in a convex neighbourhood of p
(7) g(R(X, Y )Z, W ) = g(R(Iγ X, Iγ Y )Iγ Z, Iγ W )
for all X, Y, Z, W ∈ Tγ(t) M .
• Proof: Let γ be a geodesic through p parametrized by arc length and e1 , . . . , en−1 , en =
γ̇ a parallel orthonormal frame of T M along γ. We fix a Jacobi field J along
γ with J(0) = 0. Since Iγ is a composition of isometries kJ(t)k2 = kIγ J(t)k2 .
P ei i = Iγ ei . This is a parallel
Let P i orthonormal frame along γ. Then J(t) =
i y (t)ei (t) and Iγ J(t) = i y (t)ei (t). Using (7) one shows that Iγ is a Jaco-
bi field which vanishes at t = 0.
Since ei and ei are parallel it follows that I J(0) ˙ = dtd t=0 (Iγ J(t)). Now
˙


J(t) = Dten expp (J(0))
˙


Iγ J(t) = Dten expp (I J(0)).
Hence Dγ(t) φ(J(t))k2 = kJ(t)k2 , i.e. φ is an isometry on the domain where it is
a well defined diffeomorphism.
• Corollary: All manifolds with the same constant (sectional) curvature are
locally isomteric to each other.
 11

• Corollary (Riemann): A Riemannian manifold (M, g) is locally isometric to

the Euclidian space of the same dimension if and only if R ≡ 0.

9. Lecture on June, 8 – Cartan-Ambrose-Hicks theorem

• We generalize the main theorem from the previous section to a more global
statement. For this extend the above construction to broken geodesics.
• Definition: A piecewiese smooth curve γ : [0, l] −→ M is a broken geodesic if
there are t0 = 0 < t1 < . . . < tn−1 < tn = l such that γ|[ti ,ti+1 ] is a geodesic for
all i = 0, . . . , n − 1. We write i γ for γ|[0,ti ] .
• We consider complete Riemannian manifolds M, M and an isometry I : Tp M −→
Tp M . For a broken geodesic γ in M with γ(0) = p we define a broken geodesic
γ in M inductively as follows: 1 γ is the geodesic with domain [t0 = 0, t1 ] and
1 γ̇(0) = I(γ̇(0)). We define

1I : Tγ(t1 ) M −→ T1 γ(t1 )
v 7−→ P1 γ ◦ I ◦ P1 γ (v).
If i γ and i I are defined (with i < n), then let we set

i γ(t) t ≤ ti
i+1 γ(t) =
expi γ(ti ) ((t − ti ) i I(γ̇(ti ))) ti ≤ t ≤ ti+1
where γ̇(ti ) is the derivative of γ|[ti ,ti+1 ] . We define also
Ii+1 γ : Tγ(ti+1 ) M −→ Ti+1 γ(ti+1 )
v 7−→ Pi+1 γ|[ti ,ti+1 ] ◦ i I ◦ Pγ|−1[t ,t +1] (v).
i i

We will write I(l) when we do not have to fix the number of breaking points
of the broken geodesic.
• Theorem (Cartan-Ambrose-Hicks): Assume in addition that M is simply
connected and for all broken geodesics γ : [0, l] −→ M



I(l) R(X, Y )Z = R I(l)(X), I(l)(Y ) I(l)(Z)
for all X, Y, Z ∈ Tγ(l) M . Then for any to broken geodesics γ0 , γ1 : [0, l] −→ M
with γ0 (0) = γ1 (0) = p and γ0 (l) = γ1 (l) the transferred geodesics γ 0 and γ 1
have the same endpoints, i.e.
γ 0 (l) = γ 1 (l).
• Proof: This is an adaptation of [CE], p. 37f. Let γ0 , γ1 as above. By introducing
artificial breaking points to γ0 , γ1 where these curves are actually smooth we
may assume that these two broken geodesics have the same break points 0 =
t0 < t1 < . . . < tn = l.
We will first assume that γ0 , γ1 are close to each other in the sense that for all
i = 0, . . . , n the points γ0 (ti−1 ), γ0 (ti ), γ0 (ti+1 ), γ1 (ti−1 ), γ1 (ti ), γ1 (ti+1 ) together
with the (unique) minimal geodesics connecting them lie in convex balls around
γ0 (ti ) (together with all minimal geodesic connecting these points). We make
the analogous assumption for the break points of γ 0 and γ 1 . If necessary one
introduces additional (and artificial) break points to achieve this.
We construct a chain of local isometries i φ0 from neighbourhoods of γ0 (ti )
to neighbourhoods of γ 0 (ti ) such that the domains of i−1 φ0 intersect in convex
sets containing the image of γ0 |[ti−1 ,ti ] and γ1 |[ti−1 ,ti ] . We define i φ0 to be the
local isometry obtains in Cartan theorem such that i φ0 (γ(ti )) = γ 0 (ti ) and
12

D i φ0 =i I (i.e. the differential at γ0 (ti ). Analogously we construct a chain of

local isometries i φ1 .
We can assume that the domain of i φ0 and i φ1 contains all minimal geodesic
segment connecting any two of the points γ0 (ti−1 ), γ0 (ti ), γ0 (ti+1 ), γ1 (ti−1 ), γ1 (ti ), γ1 (ti+1 )
because all these points are supposed to lie in convex balls around γ0 (ti ) and
γ1 (ti ). Recall that the proof of Cartan’s theorem yields local isometries whose
domain is any neighbourhood of a point m to a neighbourhood of a point m
where expm and expm are diffeomorphisms.
By induction we now show that i φ0 and i φ1 coincide on the intersection on
their domain. For i = 0 we have γ0 (t0 ) = γ1 (t0 ) = p and 0 I0 = I =0 I1 .
Therefore 0 φ0 =0 φ1 on the (convex) intersection of their domains.
For i = 1 we obtain local isometries. The domain of 1 φ0 contains p and the
differential of 1 φ0 at that point is I since p = expγ0 (t1 ) (−(t1 − t0 )γ̇0 (t1 )) and the
proof of Cartan’s theorem computes the differential of 1 φ0 at p as follows:
Dp 1 φ0 = P1−1
γ 0 ◦1 I0 ◦ P1 γ0 = I.

For Dp 1 φ0 we obtain the same result. Therefore these local isometries coincides
on the intersection of their domains (recall that this intersection is convex, hence
connected) which includes γ0 (t1 ). It follows that
2 γ 0 (t2 ) =1 φ0 (γ0 (t2 )) =1 φ1 (γ0 (t2 ))
and that the differentials of 1 φ0 and 1 φ1 coincide at that point.
It now follows in the same way that 2 φ0 and 2 φ1 coincide at γ0 (t1 ) and that
both local isometries have the same differential there (namely 1 I0 ). Therefore
the coincide on the intersection of their domains. Hence
3 γ 0 (t3 ) =2 φ0 (γ0 (t3 )) =2 φ1 (γ0 (t3 ))
and that the differentials of 2 φ0 and 2 φ1 coincide at that point. Now one iterates
this argument.
We obtain two chains of local isometries, i φ0 respectively i φ1 centered around
γ0 (ti ) respectively γ1 (ti ) for i = 0, . . . , n. The domains of i φ0 and i φ1 intersect in
a non-empty convex set and there these two isometries coincide. In particular,
n φ0 and n φ1 coincide near γ0 (tn ) = γ1 (tn ) and by definition n φ0 (γ0 (tn )) =
γ 0 (tn ) = γ 1 (tn ) =n φ1 (γ1 (tn )).
In order to conclude the proof one has to eliminate the closeness argument.
Let γ0 and γ1 be two broken geodesics with the same endpoints as in the
theorem. Since π1 (M ) = {1}, there is a continuous family γs interpolating
between γ0 and γ1 which has fixed endpoints. For a sufficiently fine subdivision
the points γsi (tj ) and γsi+1 (tj ) satisfy the closeness assumption made above.
Now the local isometry obtained as above near γ0 (l) from γ0 (0) is the same as
the one obtained from γs1 on a neighbourhood of γ0 (l). And the latter is the
same as the local isometry obtained from γs2 on a neighbourhood near γ0 (l)
etc. Thus the germ at γ0 (l) of the local isometry obtained as above is the same
for γ0 and γ1 .
• Theorem: Let M, M be complete Riemannian manifolds with constant sectio-
nal curvature K which are both simply connected. Then these manifolds are
isometric. Moreover, the isometry group of M acts transitively on M and for all
isometries I : Tp M −→ Tp0 M there is an isometry φ of M such that φ(p) = p0
and Dp φ = I.
 13

• Fact: For K ∈ R the following Riemannian manifolds are simply connected,

complete and have constant sectional curvature K.


 (Rn , gstandard )  if K = 0
  n
MK = SK = {kxk2 = K1 } ⊂ Rn+1 , g = gstandard S n if K > 0
 P K

 {kxk2 < −4 } ⊂ Rn , g(v, w) = i vi w

Pi 2 if K < 0

K 1+(K/4) x i i

• Definition: A complete Riemannian manifold with constant sectional curva-

ture is called a space form.
• Fact: When M is a space form with sectional curvature K, then π1 (M ) acts
properly discontinuously on the universal cover MK by isometries. Conversely,
any such action induces a space form on the quotient manifold.
• Reminder: A group action of G on M is properly discontinuous if for all p ∈ M
there is an open neighbourhood U such that gU ∩ U = ∅ unless g = 1.
• Proposition: Assume that M n is a space form with K = +1 and n is even.
Then M is either the sphere S n or the projective space S n /{±1}.
• Proof: If g ∈ Isom(Sn ) = O(2n + 1) has det(g) = 1, then 1 is an eigenvalue of
g (the determinant is the product of all eigenvalues, a complex number ζ is an
eigenvalue of g if and only if ζ is an eigenvalue of g). The eigenvector is a fixed
point of g. Because the action is properly discontinuous, g = id. If det(g) = −1
then g 2 = id. Hence all eigenvalues of g are ±1. Thus all eigenvalues have to
be −1, and hence g = −id
• Example: For n odd there are many space forms with positive sectional cur-
vature. For a prime p and q ∈ {1, . . . , p − 1} consider

Z/pZ × S 3 ⊂ C2 −→ S 3
([k], (z1 , z2 )) 7−→ e2πi/p z1 , e2πqi/p z2 .

This is a properly discontinuous action by isometries, the quotient is the lens

space L(p, q) which has a metric of constant positive sectional curvature and
π1 (L(p, q)) = Z/pZ. A comprehensive reference is [Wo].

10. Lecture on June, 12 – First and second variation of energy

• Let γ : [a, b] −→ M be a piecewise smooth curve and α : (−ε, ε) × [a, b] −→ M
a piecewise smooth variation of γ, i.e. α is continuous, here are t0 = a < t1 <
. . . < tk = b such that α|(−ε,ε)×[ti ,ti+1 ] is smooth and α(0, t) = γ(t). The variation
∂
vectorfield is W (t) = ∂s s=0
α(s, t). This is a continuous vector field along γ.
• Definition: The energy E(γ) of γ is
Z b
E(γ) = kγ̇(t)k2 dt.
a
Rb
• Remark: Let L(γ) = a kγ̇(t)kdt be the length of γ. By the Cauchy-Schwarz
inequality L(γ)2 ≤ (a − b)E(γ) with equality if and only if kγ̇(t)k is constant.
Given p, q in a complete manifold there is a minimal geodesic γ0 from p to q.
All other piecewise smooth curves γ from p to q have E(γ) ≥ E(γ0 ).
14

• Theorem (First variation of energy): In the situation above with [a, b] =

[0, 1]

n−1 1  
1 ∂E ◦ α ∇
X Z
(8) =− g (W (ti ), γ̇(ti , +) − γ̇(ti , −)) − g W (t), γ̇(t) dt.
2 ∂s s=0 i=1 0 dt

• Here γ̇(ti , +) respectively γ̇(ti , +) is the derivative of γ at ti from the right

respectively the left.
• The proof is a direct computation.
• Corollary: If γ is not a smooth geodesic, then there is always a variation vector
field W (vanishing at end points) such that a variation α of γ in direction W
with fixed end points decreases the energy.
• We will now consider a smooth geodesic γ and a two-parameter variation
α : (−ε, ε)2 × [0, 1] −→ M which is piecewise smooth (defined as above).
The variation vector field Wj = ∂α sj
for j = 1, 2, here s1 , s2 are the cartesian
2
coordinates on (−ε, ε) . Wj is a piecewise smooth vector field along the smooth
curve γ.
• Theorem (Second variation of energy): In the situation above

n−1 
1 ∂ 2 (E ◦ α)

X ∇ ∇
=− g W2 (t), W1 (ti , +) − W1 (ti , −)
2 ∂s1 ∂s2 s1 =s2 =0 i=1
dt dt
(9) Z 1 
∇2

− g W2 (t), 2 W1 (t) + R(W1 (t), γ̇(t))γ̇(t) dt
0 dt

• The proof is a direct computation starting with (8). You should think that this
is familiar, see for example the lecture on May, 22.

11. Lecture on June, 19 – Minimizing geodesics, more on variation

formulas
• Lemma: Let γ : [0, T ] −→ M be a geodesic such that γ(0) is conjugate to γ(t0 )
for t0 ∈ (0, T ). Then γ is not a minimal geodesic from γ(0) to γ(T ). In other
words: Minimizing geodesics contain no conjugate points in their interior.
• Remark: Theorem 15.1 (Index Theorem) in [Mi] is a similar (but much stron-
ger) statement.
• Proof: Let J : [0, t0 ] −→ T M be a non-trivial Jacobi field which vanishes at
∇
γ(0) and γ(t0 ). Let X0 := dt J(t0 ).
We extend J by 0 to a continuous, piecewise smooth vector field along γ and
we choose a smooth vector field X along γ such that X(t0 ) = 0 and X vanishes
at the endpoints of γ. For c > 0 we apply the second variation formula to a
variation αs of γ with variation vector field cJ −c−1 X and with fixed endpoints.
 15

By the second variation formula and because J is a piecewise Jacobi field

 
1 d2  −1 ∇ 
E(α(s)) = −g  cJ(t0 ) − c X(t 0 ), −c J(t0 , −)
2 ds2 s=0
 dt
| {z }

=X0
 T  2 
∇
Z
−1 −1
− g cJ(t) − c X(t), −c X + R(X, γ̇)γ̇ dt
0 dt2
1 T ∇2
Z  
= −g(X0 , X0 ) − 2 g X, 2 X + R(X, γ̇)γ̇ dt
c 0 dt
Z t0  2

∇
+ g J, 2 X + R(X, γ̇)γ̇ dt
0 dt
1 T ∇2
Z  
2
= −kX0 k − 2 g X, 2 X + R(X, γ̇)γ̇ dt
c 0 dt
Z t0    
∇ ∇
− g J, X + g(R(J, γ̇)γ̇, X) dt
0 dt dt
1 T ∇2
Z  
2
= −kX0 k − 2 g X, 2 X + R(X, γ̇)γ̇ dt
c 0 dt
 
Z t0 t0
 ∇2
 
∇
− g X, 2 J + R(J, γ̇)γ̇  dt − g J, X

0 |dt {z } dt t=0
=0

1 T ∇2
Z  
2
= −2kX0 k − 2 g X, 2 X + R(X, γ̇)γ̇ dt.
c 0 dt

In this computation we (once) used the symmetry properties of the curvature

tensor (cf. Lecture of May, 18) and (twice) the fact that ∇ is metric.
Since J 6≡ 0 implies X0 6= 0 the first summand is negative. The second
summand can then be ignored if c is big enough. Thus there is a variation of γ
with fixed endpoints such that curves near γ have smaller energy than γ. Hence
γ is not minimizing.
• Remark: So far we have considered only proper variations of curves/geodesics,
i.e. variations with fixed end points. Similar computations as above lead to the
following results.
• First variation of energy: αs is a piecewise smooth, maybe not proper va-
riation of a piecewise smooth curve γ : [0, a] −→ M with variation vector field
W along γ. Then

a  
∇
Z
1 d X
E(αs ) = − g W, γ̇ dt − g (W (ti ), γ̇(ti , +) − γ̇(ti , −))
2 ds s=0 0 dt t break points
i

− g(W (0), γ̇(0)) + g(W (a), γ̇(a))

• Second variation of energy: αs is a piecewise smooth, maybe not proper

variation of a geodesic γ : [0, a] −→ M with variation vector field W along γ.
16

Then
a
1 d2 ∇2
Z  
E(αs ) = − g W (t), 2 W + R(W, γ̇)γ̇ dt
2 ds2 s=0 0 dt
 
X ∇ ∇
− g W (ti ), W (ti , +) − W (ti , −)
ti interior break points
dt dt
(10)    
∇ ∂αs ∇ ∂αs
−g (0, 0), γ̇(0) + g (0, a), γ̇(a)
∂s ∂s ∂s ∂s
   
∇ ∇
− g W (0), W (0) + g W (a), W (a) .
dt dt
∇ ∇
• Reformulation of (10): Using dtd g(X, Y ) = g dt



X, Y + g X, dt Y (10) can
be written in a more compact form:
Z a 
1 d2
 
∇W ∇W
E(αs ) = g , − g(R(W, γ̇)γ̇, W ) dt
2 ds2 s=0 0 dt dt
   
∇ ∂αs ∇ ∂αs
−g (0, 0), γ̇(0) + g (0, a), γ̇(a) .
∂s ∂s ∂s ∂s
• Remark: For proper variations the right hand side depends only on the va-
riation vector field. If the variation is not proper, then the precise form of the
variation at the end points of γ matters.
• Definition: Let γ : [0, a] −→ M be a smooth geodesic and W a piecewise
smooth variation vector field (maybe not vanishing at endpoints). Then we
write
Z a   
∇W ∇W
Ia (W, W ) := g , − g(R(W, γ̇)γ̇, W ) dt.
0 dt dt

• Definition: Let (M, g) be a Riemannian manifold with dim(M ) > 1 and R

its curvature tensor. Let Ei , i = 1, . . . , n be an orthonormal basis of Tp M and
X, Y, V ∈ Tp M .
1 X
Ric(X, Y ) = g (R(X, Ei )Ei , Y )
n−1 i
1 X
Ric(V ) = g (R(V, Ei )Ei , V )
n−1 i

are both called the Ricci curvature of M . The scalar curvature is

1X
scal(p) = Ric(Ei )
n i
1 X
= K(Ei , Ej ).
n(n − 1) i6=j

• Remark: Ric(X, Y ) = Ric(Y, X) by the symmetry properties of the curvature

tensor. Hence the Ricci tensor can be compared with the metric tensor (which
is also symmetric and bilinear).
 17

12. Lecture on June, 22 – Theorems of Bonnet-Myers, Weinstein-Synge

• Theorem (Bonnet-Myers): Let M be a complete Riemannian manifold such
that there is a constant ρ > 0 with

kvk2
(11) Ricp (v) ≥ .
ρ2

Then diam(M ) ≤ πρ. In particular, M is bounded and hence compact.

• Proof: Let p, q ∈ M we will show that d(p, q) = l > πρ leads to a contradiction.
Let γ be a geodesic from p to q with length l, parametrized by arc length and
e1 , . . . , en−1 , en = γ̇ parallel orthonormal vector fields along γ. Let Wi (t) =
sin(πt/l)ei (t). Applying the second variation formula to a variation αi (s, ·) with
variation vector field Wi we get
l
1 d2
Z
sin2 (πt/l) π 2 /l2 − K(en , ej ) dt


E(αj (s, ·)) =
2 ds s=0 0

Summing this over j = 1, . . . , n − 1 we get

n−1 Z l
π2
 
1 X 00 2
E (0) = sin (πt/l) (n − 1) 2 − (n − 1)Ricγ(t) (en (t)) dt
2 j=1 j 0 l
Z l  2 
2 π 1
< (n − 1) sin (πt/l) − dt = 0
0 (πρ)2 ρ2

by (11) and our assumption l < πρ. Then there is j such that Ej00 (0) < 0
and γ is not the geodesic with minimal energy/length from p to q. This is a
contradiction.
• Corollary: Let (M, g) be as in the theorem. Then |π1 (M )| < ∞.
• Proof: The universal cover of M with the the pull back metric is complete and
satisfies (11). Hence the universal cover of M is compact, i.e. π1 (M ) is finite.
• Remark: The round sphere shows that the bound on the diameter is sharp.
• Example: No torus admits a Riemannian metric with positive Ricci-curvature.
• Theorem (Weinstein-Synge): Let M be compact, connected, orientable,
Riemannian with K(σ) > 0 for all planes σ in T M . Then every isometry f of
M has a fixed point provided f is
orientation preserving if dim(M ) is even
orientation reversing if dim(M ) is odd.
• Proof: Assume that f has no fixed point. By compactness, there is p0 such that
0 < l = d(p0 , f (p0 )) ≤ d(p, f (p)) for all p ∈ M . There is a minimal geodesic
γ : [0, l] −→ M from p to f (p).
If γ̇(l) 6= Df (γ̇(0)) then the distance between γ(l/2) and f (γ(l/2)) is smaller
than l because there is a broken geodesic connecting these points whose length
is l. Hence the map Df : (γ̇(0))⊥ ⊂ Tγ(0) M −→ (γ̇(l))⊥ ⊂ Tγ(l) M is well
defined. The parallel transport Pγ along the geodesic γ defines a linear map
between the same spaces.
By the orientation assumptions on M the map Pγ−1 ◦ Df has 1 as an ei-
genvalue, hence there is a unit vector V which is orthogonal to γ̇(0) such that
Pγ (V ) = Df (V ). Let V (t) be the parallel extension of V (0) = V and consider
18

the variation
α : (−ε, ε) × [0, l] −→ M
(s, t) 7−→ expγ(t) (sV (t))
of γ. By the second variation formula we get
Z l
1 d2
E(α(s, ·)) = − g(R(V (t), γ̇(t))γ̇(t), V (t))dt
2 ds2 s=0 0
Z l
=− K(V (t), γ̇(t))dt < 0
0

where K(V (t), γ̇(t)) is the sectional curvature of the plane spanned by V (t)
and γ̇(t). For small |s| =
6 0 the energy of α(s, ·) is smaller than the energy of γ.
By the choice of V we have Df (V (0)) = V (l) and hence


α(s, l) = expγ(l) (sDf (V (0))) = f expγ(0) (sV (0)) = f (α(s, 0).
Since L(α(s, ·))2 ≤ lE(α(s, ·)) < lE(γ) = L(γ)2 the path α(s, ·) is shorter than
γ. This is a contradiction to the choice of p0 (and γ).

13. Lecture on June, 26 – Corollaries of Weinstein-Synge, Index Lemma

• Corollary: Let M be an orientable, compact, connected Riemannian mani-
fold of even dimension with positive sectional curvature. Then M is simply
connected.
• Proof: The sectional curvature of M is bounded from below by κ > 0. Then
Ricp (v) ≥ κkvk2 . The universal cover of M is complete, by the theorem of
Bonnet-Myers, the universal cover M f −→ M of M is compact and the deck
transformations are orientation preserving. Hence they all have fixed points,
i.e. M itself is simply connected.
• Remark: Compare with the Proposition in the lecture of June, 8 (on p. 13).
• Corollary: Let M be an orientable, compact, connected Riemannian manifold
of odd dimension with positive sectional curvature. Then M is orientable.
• Proof: As before, the universal cover of M is compact. Because it is simply
connected, it is orientable (one can use parallel transport to coherently orient
all tangent spaces using some orientation of Tp M f for some p ∈ M f). All deck
transformations preserve the orientation, therefore the quotient of M f by the
deck group is orientable, but this quotient is M .
• Remark: All space forms with K > 0 and odd dimension are orientable.
• Index-Lemma: Let γ : [0, a] −→ M be a geodesic such that γ(t), t ∈ (0, a]
is not conjugate to γ(0) along γ. Let J be a Jacobi field along γ such that
∇
J(0) = 0 and dt J(0) ⊥ γ̇(0) (hence J(t) ⊥ γ(t) for all t). Let V be a piecewise
smooth vector field along γ such that V (0) = 0 and V (t0 ) = J(t0 ) for some
t0 ∈ (0, a]. Then
!
Z t0 2
∇
It0 (V, V ) = V − g (R(V, γ̇)γ̇, V ) dt ≥ It0 (J, J)
0 dt

with equality if and only if V (t) = J(t) for all t ∈ [0, a].
 19

14. Lecture on June, 29 – Proof of the Index Lemma, Rauch

Comparison theorem
• Proof of the Index Lemma: Let n = dim(M ). Pick a basis J1 , . . . , Jn−1 of
the space of Jacobi fields along γ which vanish at γ(0) and are orthogonal to
γ. Because there are no conjugate points to γ(0) along γ, J1 (t), . . . , Jn−1 (t) is a
basis of (γ̇(t))⊥ ⊂ Tγ(t) M for 0 < t ≤ a. Hence there are functions fi : (0, a] −→
R and constants αi such that
X X
V (t) = fi (t)Ji (t) J(t) = αi Ji (t).
i i
The functions fi extend to piecewise smooth functions on the closed interval.
For this notice that there are vector fields Ai along γ such that tAi (t) = Ji (t)
∇ ⊥
P Ai (0) = dt Ji (0). A1 (t), . . . , An−1 (t) is a basis of γ̇(t) for all t, i.e. V (t) =
and
i hi (t)Ai (t) with hi (0) = 0, i.e. there are functions gi such that tgi (t) = hi (t).
Hence X X X
V = hi Ai = tgi Ai = gi Ji .
i i i
Therefore gi is the desired extension of fi .
On an interval where V is smooth the following identity holds
 
∇ ∇
g V (t), V (t) − g (R(V (t), γ̇(t))γ̇(t), V (t)) =
dt dt
(12) ! !
X X d X X ∇
g fi0 (t)Ji (t), fj0 (t)Jj (t) + g fi (t)Ji (t), fj (t) Jj (t) .
i j
dt i j
dt
Integrating this identity (summing over smooth intervals), using fi (t0 ) = αi
and applying the same identity to J one obtains the desired inequality since
Z t0 !
X X
0 0
g fi (t)Ji (t), fj (t)Jj (t) dt ≥ 0.
0 i j

In the equality case one has fi0 ≡ 0 for i = 1, . . . , n − 1, hence fi = αi and

V (t) = J(t). P
To prove (12), one expands V (t) = i fi (t)Ji (t) everywhere on both sides of
the identity and apply the product rule. Most terms cancel, one of them can be
treated using the symmetries of the curvature tensor and the Jacobi equation.
Finally, one uses the the fact that for all i, j
   
∇ ∇
h(t) = g Ji (t), Jj (t) − g Ji (t), Jj (t) .
dt dt
vanishes. To see this evaluate at zero and show that h0 ≡ 0 (this uses again
the Jacobi equation and the symmetries of the curvature tensor, c.f. Lecture
on May, 18).
• Rauch comparison theorem: Let (M, g), (M f, ge) be Riemannian manifolds
such that dim(M f) ≥ dim(M ). Let J : [0, a] −→ T M respectively Je : [0, a] −→
TMf Jacobi fields along the geodesic γ : [0, a] −→ M respectively γ e : [0, a] −→
M such that
f
– J(0) = 0, J(0)e = 0,
 
e˙ (0) (we write J 0 instead of dt
– g(J 0 (0), γ̇(0)) = ge Je0 (0), γ ∇
J etc.)
20

– kJ 0 (0)k = kJe0 (0)k

– kγ̇(t)k = kγe˙ (t)k =
6 0.
We assume moreover that no point γ(t)
g is conjugate to γ
e(0) along γ
e and that

K(V, γ̇(t)) ≤ K( e˙ (t))

e Ve , γ

e˙ (t)⊥ . (Here K resp. K

for all V ∈ γ̇(t)⊥ and Ve ∈ γ e denote sectional curvatures
of planes in T M resp. T Mf.)
Then kJ(t)k
e ≤ kJ(t)k and there are no conjugate points to γ(0) along γ. If
there is equality, then K(J(t), γ̇(t) = K(e J(t),
e γ e˙ (t)) for t ∈ [0, a].

15. Lecture on July, 3 – Proof of the Rauch Comparison theorem,

simple applications
∇ ∇
• Proof: Using the assumptions on J(0), dt J(0), J(0), J(0) one can reduce to
e e e
dt
the case when both Jacobi fields are orthogonal to the geodesics they are asso-
ciated with.
We define l(t) = g(J(t), J(t)), ˜l(t) = g̃(J(t),
˜ J(t))
˜ and we will show l
l(t)
First, by l’Hopitals rule we extend l̃(t) to t = 0

l(t) 2g(J 0 (t), J(t)) g(J 0 (t), J 0 (t)) + g(J 00 (t), J(t))
lim = lim = lim
t→0 ˜
l(t) t→0 2g̃(J˜0 (t), J(t))
˜ t→0 g̃(J˜0 (t), J˜0 (t)) + g̃(J˜00 (t), J(t))
˜
=1

∇
where we write J 0 for dt
J etc.
0 0
d l(t)
We will show ≥ 0, or equivalently ll(t)
dt l̃(t)
(t)
≥ l̃l̃(t)
(t)
. Let t0 ∈ (0, a], we put
U (t) = J(t)/kJ(t0 )k (the notation U does not the reflect the dependence on t0 )
and
Z t0 2
l0 (t0 ) 0 d d
= 2g(U (t0 ), U (t0 )) = g(U (t0 ), U (t0 )) = g(U (t), U (t))dt
l(t0 ) dt 0 dt2
(13) Z t0
=2 (g(U 0 (t), U 0 (t)) − g(R(U, γ̇)γ̇, U )) dt = 2It0 (U, U ).
0

In order to apply the index lemma we want to transplant U to M f and com-

pare the result to U ˜ J(t
e = J/k ˜ 0 )k. Let E1 = γ̇/kγ̇k, E2 , . . . , En be a paral-
lel orthonormal frame along γ such that E2 (t0 ) = U (t0 ). We also fix E e1 =
γ
e/kγe˙ k, E2 , . . . , E
en+k a parallel orthonormal frame along γe such that E e2 (t0 ) =
U
e (t0 ). Let

φ : {vector fields along γ} −→ {vector fields along γ

e}
X X
V (t) = fi (t)Ei (t) 7−→ fi (t)E
ei (t).
i i
 21
∇ ∇
This map is C ∞ linear,


φ(V ) = φ dt V and g̃(φ(W1 )(t), φ(W2 )(t)) = g(W1 (t), W2 (t)).
e
dt
By the curvature assumption
Z t0
It0 (U, U ) = (g(U 0 (t), U 0 (t)) − g(R(U, γ̇)γ̇, U )) dt
0
Z t0     
≥ g̃ (φ(U )0 (t), φ(U )0 (t)) − g̃ R
e φ(U ), γe˙ γe˙ , φ(U ) dt
0
= It0 (φ(U ), φ(U ))
≥ It0 (U
e, U
e)

where in the last step we used the Index Lemma (by construction U e is a Jacobi
field such that Ue (0) = 0 and Ue (t0 ) = U (t0 )). By the analogous computation to
(13) for ˜l we obtain
l0 (t0 ) ˜l0 (t0 )
≥
l(t0 ) ˜l(t0 )
for (arbitrary) t0 ∈ (0, a]. This implies kJ(t)k ≥ kJ(t)ke for all t. In the case of
equality kJ(t0 )k = kJ(t
e 0 )k for t0 ∈ (0, a] we have It0 (U
  ) = It
e, U
e 0 (φ(U ), φ(U )).
Then φ(U ) is a Jacobi field and g(R(U, γ̇)γ̇, U ) = g̃ R e φ(U ), γe˙ γ e˙ , φ(U ) .
• Applications: The following two propositions are simple applications of Rauch’s
theorem.
• Proposition: Let M be a complete manifold with sectional curvature K(σ)
satisfying 0 < L ≤ K(σ) ≤ H for constants H, L. Let γ be a geodesic in M .
Then the distance d (measured along γ) between γ(0) and the next conjugate
point along γ satisfies
π π
√ ≤d≤ √ .
H L
• Remark: The upper bound can be obtained as in the theorem of Bonnet Myers.
For the lower bound compare M with M f = S n (H), the sphere with sectional
curvature H (cf. p. 13).
• Proposition: Let M, M f be complete Riemannian manifolds of the same di-
mension such that for all planes σ ⊂ T M and σ
e ⊂ TM f we have K(σ) ≤ K(e e σ ).
Let r > 0 be such that
expp : Br (0) ⊂ T M −→ M is a diffeo. onto its image
expp̃ : Br (0) ⊂ T M
f −→ M
f is non-singular,

c : [0, 1] −→ expp (Br (0)) ⊂ M a smooth curve and I : Tp M −→ Tp̃ M f an

−1
c(t) = expp̃ ◦I ◦ expp (c(t)). Then l(e
isometry. Define e c) ≤ l(c).
• Remark: The previous proposition may help to verify that a given r works.
Even more favorable circumstances are non-positive curvature and simple connec-
tivity.

16. Lecture on July, 6 – Isometric immersions, Moore’s theorem

(non-existence of isom. immersions)
• Remark: Let M ⊂ M be a submanifold in a Riemannian manifold with Levi-
Civita connection ∇. Then the Levi-Civita connection of the induced metric
22

on M is

∇X Y (p) = pr ∇X Y
where X, Y are extensions of X, Y to a neighbourhood of p and pr : Tp M −→
Tp M is the orthogonal projection (similar notation in what follows).
• Notation: B(X, Y ) = ∇X Y − ∇X Y .
• Proposition: B(·, ·) is well defined, symmetric, C ∞ -linear.
• Proof: elementary, for symmetry use that the connections are torsion free.
• Definition: Let N ∈ Tp M ⊥ . The second fundamental form IIN on Tp M is the
quadratic form
IIN (X) := HN (X, X) := g(B(X, X), N ).
To HN corresponds the symmetric endomorphism SN of Tp M defined by g(SN (X), Y ) =
HN (X, Y ).
• Proposition: Let p ∈ M, X ∈ Tp M, N ∈ Tp M ⊥ . (We denote any extension of
N to a local vectorfield orthogonal to M by the same letter.) Then


SN (X) = −pr ∇X N .
• Proof: Computation:


g (SN (X), Y ) = g (B(X, Y ), N ) = g ∇X Y − ∇X Y, N



= g ∇X Y , N = g Y, −∇X N .
• Theorem (Gauß): Let X, Y ∈ Tp M be orthonormal. Then the sectional cur-
vatures K and K of the X, Y -plane are related as follows:
(14) K(X, Y ) − K(X, Y ) = g(B(X, X), B(Y, Y )) − kB(X, Y )k2 .
• Proof: Computation:



K(X, Y ) − K(X, Y ) = g ∇X ∇Y Y − ∇Y ∇X Y − ∇X ∇X Y − ∇Y ∇X Y , X (p)
 
+ g ∇[X,Y ] Y − ∇[X,Y ] Y, X (p).
The second summand vanishes because the first entry in g is orthogonal to Tp M .
Let E1 , . . . , Em be a collection of pairwise orthonormal vector fields normal to
M on a neighbourhood of p (with m = dim(M ) − dim(M ). Then
X
B(X, Y ) = HEi (X, Y )Ei .
i

Express ∇Y Y in these terms, then express ∇X ∇Y Y and proceed in the same

way for the other summand.
• Theorem (Moore): Let M be complete, simply connected with sectional
curvature K(σ) ≤ b ≤ 0 for all planes σ ⊂ T M .
Let M ⊂ M be compact submanifold (maybe immersed) with sectional cur-
vature K such that K − K ≤ −b. Then dim(M ) ≥ 2 dim(M ).
• Discussion: The Clifford torus in R4 is a flat T 2 . Closed surfaces in R3 have
positive curvature somewhere. The flat 3-torus R3 /Z3 contains a flat T 2 , so the
π1 -assumption is needed.
• Proof: By contradiction. Let p ∈ M \ M and p ∈ M such that
(15) l = d(p, q) ≥ d(p, p)for all q ∈ M.
There is a unique minimal geodesic from p to p with speed 1. By the first
variation formula γ is perpendicular to Tp M .
 23

Let V ∈ Tp M of unit length and c(s) a curve in M representing V . Let

c̃(s) = expp−1 (c(s)) and
f : (−ε, ε) × [0, l] −→ M
 
tc̃(s)
(s, t) 7−→ expp .
l
The variation vector field V (t) is a Jacobi field along γ since f is a variation
through geodesics (with only one fixed endpoint). Moreover, V (0) = 0 and
V (l) = V .
We now compare M with the complete, simply connected Riemannian mani-
fold Mf(b) with sectional curvature b ≤ 0. There are no conjugate points along
any geodesic. Since K ≤ K e =b
Il (V , V ) ≥ Il (Ve , Ve )
where Ve = φ(V ) with φ from the proof of the Rauch comparison theorem. We
can estimate Il (Ve , Ve ) from below using the unique Jacobi field Je along γ
e which
coincides with V at t = 0 and t = l.√Recall (6) on p. 8
e
A computation shows Il (J, e J) e > −b and by the Index Lemma we get
√
Il (V , V ) ≥ Il (Ve , Ve ) ≥ Il (J, e > −b.
e J)
By the second variation formula
1 d2

E(f (s, ·)) = Il (V , V ) + g Sγ̇(l) V (l), V (l) .
2 ds2 s=0
and since E(γs ) ≤ E(γ0 = γ) we know

√

0 ≥ Il (V , V ) + g Sγ̇(l) V (l), V (l) > −b + g Sγ̇(l) V (l), V (l)
Recall that V = V (l) was arbitrary (of unit length) and that g(Sγ̇(l) (V (l)), V (l)) =
g(B(V (l)), V (l)), γ̇(l)). Thus
√
(16) kB(V, V )k > −b
Now let V, W ∈ Tp M be orthonormal. Then by the Gauß-formula (14)
(17) K(V, W ) − K(V, W ) = g(B(V, V ), B(W, W )) − kB(V, W )k2 ≤ −b.
In the next lecture we will see how these inequalities interact with the dimension
assumption to finish the proof.

17. Lecture on July, 10 – Conclusion of the proof of Moore’s theorem

• The following Lemma shows that (16) and (17) contradict each other when
dim(M ) < 2dim(M ). In the Lemma below Rn corresponds to Tp M and Rk
corresponds to Tp M ⊥ . The two quantitative conditions on B correspond to
(16) and (17).
• Lemma (Otsuki): Let B : (Rn , g) × Rn −→ (Rk , g) be a symmetric bilinear
form such that for some b ≤ 0:
g(B(V, V ), B(W, W )) − kB(V, W )k2 ≤ −b
√
kB(V, V )k > −b
for all g-orthonormal pairs V, W . Then k ≥ n.
24

• Proof: We assume k < n. The function f : S n−1 −→ R, f (V ) = kB(V, V )k2

attains its minimum somewhere, we call that point V0 . Because of the second
assumption on B in the Lemma we know B(V0 , V0 ) 6= 0.
Let W ∈ S n−1 be orthogonal to V0 , so it can be thought of as element of
TV0 S n−1 . Because V is a critical point of f :
(18) 0 = (DV0 f )(W ) = 4g(B(V0 , V0 ), B(W, V0 )).
Since V0 is a minimum:
0 ≤ (DV2 0 f )(W, W )
(19)
= 8kB(W, V0 )k2 − 4g(B(V0 , V0 ), B(V0 , V0 )) + 4g(B(W, W ), B(V0 , V0 )).
Now assume that k < n. Then the linear map
TV0 S n−1 −→ (B(V0 , V0 ))⊥ ⊂ Rk
W 7−→ B(V0 , W )
is well defined by (18) and it has non-trivial kernel if k < n since (B(V0 , V0 ))⊥
has dimension k − 1 while dim(TV0 S n−1 ) = n − 1. Let W0 be a unit vector in
that kernel. Using (19) with V = V0 , W = W0 and (18) we get
0 ≤ 8kB(W, V0 )k2 + 4g(B(W0 , W0 ), B(V0 , V0 )) − 4g(B(V0 , V0 ), B(V0 , V0 ))
  

= 4 g(B(W0 , W0 ), B(V0 , V0 )) − kB(V0 , W0 )k2  − g(B(V0 , V0 ), B(V0 , V0 ))

| {z }
=0
< −b − (−b) = 0
using the assumptions on B in the lemma. Thus the assumption k < n leads
to a contradiction.

18. Lecture on July, 13 – Focal points

• Today N ⊂ M is a submanifold.
• Lemma: Let N ⊂ M be a submanifold, p ∈ N , q ∈ M and γ : [0, l] −→ M a
geodesic from p to q parametrized by arc length. Let f : (−ε, ε) × [0, l] −→ M
be a variation of γ such that
– α(s) : f (s, 0) ∈ N ,
– t 7−→ f (s, t) is a geodesic,
– A(s) = ∂f ∂t
(s, 0) ∈ Tα(s) N ⊥ .
Let J(t) = ∂f∂s
(0, t) be the variation vector field of f . Then J is a Jacobi field,
M
J(0) ∈ Tp N and ∇dt J(0) + Sγ̇(0) (J(0)) ∈ Tp N ⊥ .
• Proof: Only the last statement is non trivial. Let V a local vector field near p
in M tangent to N at points in N . Then
 M   M 
∇ ∇ d
f (0, 0), V = g (∇M


g J(0), V = g J(0) A)(0), V
dt ds dt

tang 
M


= g (∇J(0) A)(0) , V = g −SA(0) J(0), V
by the definition of S (on p. 22). V (p) ∈ Tp N is arbitrary.
• Lemma: Let γ : [0, l] −→ M be a geodesic and J a Jacobi field along γ such
that
– γ(0) ∈ N , γ̇(0) ∈ Tp N ⊥ ,
 25

– J(0) ∈ Tp N , and
– J 0 (0) + Sγ̇(0) J(0) ∈ Tp N ⊥ .
Then there is a variation f of γ as in the previous lemma.
• Proof: Pick a curve α : (−ε, ε) −→ N representing J(0) and a vector field
W along α such that W (0) = γ̇(0) and ∇W ds
(0) = J 0 (0). Decompose W (s) =
U (s) + V (s) with U (s) ∈ T N and V (s) ∈ T N ⊥ . Define
f : (−ε, ε) × [0, l] −→ M
(s, t) 7−→ expα(s) (tV (s)).
∂f
This is a variation through geodesics perpendicular to N and ∂s
(0, 0) = J(0).
Note that
 
∇ ∂f ∂f
(0, 0) + Sγ̇(0) (0, 0) ∈ Tp N ⊥
∂t ∂s ∂s
∇J
(0) + Sγ̇(0) (J(0)) ∈ Tp N ⊥
∂t
by the first lemma/assumption. Hence
∇J ∇ ∂f ∇W ∇ ∂f ∇U
(0) − = (0) − = (0) ∈ Tp N ⊥
dt ∂t ∂s ds ∂s ∂t ds
U is tangent to N along α. Let X be a vector field orthogonal to N along α.
Then
 
 
d ∇U ∇X 
0= g(U (s), X(s)) = g (0), X(0) + g U (0), (0) .
ds s=0 ds | {z } ds
=0

∇ ∂f
Therefore ∂t ∂s
(0, 0) = J 0 (0) since X is arbitrary. Since J(0) = ∂f
∂s
(0, 0) = J(0)
∂f
it follows that ∂s (s, t) = J(t) (both are Jacobi fields).
• Definition: Let N ⊂ M be a submanifold of M . q ∈ M is a focal point of N if
there is a geodesic γ : [0, l] −→ M such that γ(l) = q and a non-trivial Jacobi
field along γ with the properties as in the previous lemma.
• Notation: Let T N ⊥ = {V ∈ Tp N ⊥ , p ∈ N } be the normal bundle in M .
This is a smooth submanifold of the restriction of T M to N , its dimension is
the dimension of M . There is a neighbourhood of the zero section of T N ⊥ on
which the exponential map is defined. There is no harm in assuming that M is
complete, then exp⊥ is defined on T N ⊥ .
• Proposition: Let q ∈ M and N ⊂ M a submanifold and M complete. Then
q ∈ M is a focal point if and only of q is a critical value of exp⊥ .
• Proof: Assume that q is a focal point. Let γ, J be a corresponding geodesic/non-
trivial Jacobi field, f the variation provided by the previous lemma and A(s) =
∂f
∂t
(s, 0). Then the path s 7−→ w(s) = (f (s, 0), lA(s)) is a smooth path in T N ⊥
representing a tangent vector. exp⊥ (f (0, 0), lA(0)) = q and (D exp⊥ )w(0) (w0 (0)) =
(D exp)lA(0) (lJ 0 (0)) = J(l) = 0. Since w represents a non-zero tangent vec-
tor (multiplication with real numbers induces an automorphism of T N ⊥ and
s 7−→ (f (s, 0), cA(s)) maps to non-zero tangent vectors for many c ∈ R because
J(c) does not always vanish).
Conversely one easily constructs a geodesic/Jacobi field as in the definition
of focal points when q is a critical value of exp⊥ .
26

19. Lecture on July, 17 – Rauch theorem for submanifolds, Toponogov’s

theorem, number of generators of M1 (M ) with lower curvature bounds
• Definition: Let γ : [0, a] −→ M be a geodesic. Then γ is focal point free  if
⊥
there is ε such that γ contains no focal point for expγ(0) Bε (0) ⊂ (γ̇(0)) .
• Index Lemma for focal points: Let γ : [0, l] −→ M be a focal point free
geodesic and J a Jacobi field along γ which is orthogonal to γ̇, J 0 (0) = 0, and
V a piecewise smooth vector field along γ and J(t0 ) = V (t0 ) for t0 ∈ (0, l].
Then It0 (V, V ) ≥ It0 (J, J) with equality if and only if V ≡ J on [0, t0 ].
• Rauch theorem, vers. 2: Let γ : [0, a] −→ M and γ : [0, a] −→ M be two
geodesics with Jacobi fields J, J such that
– dim(M ) ≤ dim(M ), and kγ̇k = kγ̇k,
0
– J 0 (0) = 0 = J (0), kJ(0)k = kJ(0)k, and
– g(J(0), γ̇(0)) = g(J(0), γ̇(0)).
Assume that K(X, γ̇) ≤ K(X, γ̇) for all X ∈ Tγ(t) M and X ∈ Tγ(t) M . Then

kJ(t)k ≤ kJ(t)k.

Equality for some t0 ∈ (0, a] implies K(J(t), γ̇(t)) = K(J(t), γ̇(t)) for all t ∈
[0, t0 ].
• Definition: A geodesic triangle in M is a set of three geodesics γ1 , γ2 , γ3 pa-
rametrized by arc-length such that γi (li ) = γi+1 (0) (indices are taken mod 3)
and li + li+1 ≥ li+2 . The angle between −γ̇i+1 (li+1 ) and γ̇i+2 (0) is denoted by
0 ≤ αi ≤ π. The length of γi is li .
• Notation: M (H) is the complete, simply connected Riemannian manifold of
dimension 2 and constant sectional curvature (unique up to isometry).
• Theorem (Toponogov): Let M be a complete Riemannian manifold with √
sectional curvature KM ≥ H ∈ R. If H ≤ 0 the conditions pertaining to π/ H
are vacuous.
(A) Let (γ1 , γ2 , γ√ 3 ) be a geodesic triangle in M such that γ1 , γ3 are minimal
and li ≤ π/ H. In M (H) there is a geodesic triangle (γ 1 , γ 2 , γ 3 ) such
that l(γ i ) = li and αi ≤ αi for i = 1,√3. This triangle in M (H) is unique
up to isometry except when li = π/ H for some i.
(B) Let γ1 , γ2 be geodesic segments (param. by arclength) in M such that
γ1 (l1 ) = γ2 (0). We denote the angle between −γ̇1 (l1 ) and γ̇2 (0) by α
(the configuration (γ1 , √γ2 , α) is called a geodesic hinge. Assume that γ1
is minimal and l2 ≤ π/ H. Let (γ 1 , γ 2 , α) be a geodesic hinge in M (H)
with l(γ i = li for i = 1, 2. Then

d(γ1 (0), γ2 (l2 )) ≤ d(γ 1 (0), γ 2 (l2 )).

• The proof of this theorem is lengthy but not very difficult, see for example
Chapter 2 of [CE]. Instead of dealing with the proof we present an application
of part (B) of Toponogov’s theorem (following [Me]).
• Consequence of Rauch/Toponogov: Let (a, b, c) be a non-degenerate geo-
desic triangle. If K < 0, then the sum of the interior angles is in (0, π).
• Theorem (Gromov): Let M be a complete manifold with non-negative sec-
tional curvature and dimension n. There is a number C(n) such that π1 (M )
can be generated by C(n) elements.
 27

• Proof: Consider the universal covering pr : M

f −→ M , the action of π1 (M ) on
f (by isometries) and x0 ∈ M
M f. Define the displacement function

| · | : π1 (M ) −→ R
g 7−→ d(x0 , gx0 ).

Since Mf is complete, closed balls in M are compact. Hence if there were infi-
nitely many g ∈ π1 (M ) with |g| < R for fixed R this would contradict the fact
that pr is a covering. Thus we can pick generators of π1 (M ) according to the
following rules.
1. Pick g1 so that 0 < |g1 | ≤ |g| for all g 6= 1.
2. Assume g1 , . . . , gk are already fixed. If the smallest subgroup hg1 , . . . , gk i ⊂
π1 (M ) containing g1 , . . . , gk is not the entire group, pick gk+1 so that
|gk+1 | ≤ |g| for all g ∈ π1 (M ) \ hg1 , . . . , gk i. Otherwise stop (it is not yet
clear that this algorithm really stops).
Then for i < j we have li = |gi | ≤ |gj | = lj and
lij = d(gi x0 , gj x0 ) ≥ lj .
Otherwise, |gi−1 gj | < lj and we should have chosen gi−1 gj instead of gj as j-th
generator. For all i fix a minimal geodesic γi (parametrized by arc length) in
M
f from x0 to gi x0 . Let αij be the angle between γ̇i (0) and γ̇( 0).
We apply part (B) of Toponogov’s theorem with H = 0. By the law of cosines
lij ≤ li2 + lj2 − 2li lj cos(αij ). Hence if i < j
2

2
li2 + lj2 − lij li2 + lj2 − lj2 1
(20) cos(αij ) ≤ ≤ 2
=
2li lj 2li 2
Therefore αij ≥ π3 . Thus the π/6-balls around γ̇i (0), i = 1, . . . are pairwise
disjoint. γ̇i (0) are unit vectors in Tx0 M
f. Hence the number of generators is
bounded from above by
vol(S n−1 )
2vol(π/6 − ball around a point in S n−1 )
(the additional factor in the denominator comes from considering π/6-balls
around ±γ̇i (0). The denominator can be roughly estimated from below by the
volume of the Euclidean sin(π/6)-ball in n − 1-dimensional Euclidean space.
π (n−1)/2
The volume of the r-ball in Euclidean n − 1-space is rn−1 Γ((n+1)/2) , the volume
nrπ n/2
of the 1-sphere in Euclidean n-space is Γ((n+2)/2)
, see [Wa] on p. 254f.
nπ n/2 Γ((n+1)/2)
Hence the algorithm above stops after at most 2Γ((n+2)/2)π (n−1)/2 (1/2)n−1
=
√ Γ((n+1)/2)
πn2n−1 2Γ((n+2)/2) = C(n) steps.
• Theorem (Gromov): Let M be complete, n-dimensional with diameter diam(M ) =
D and sectional curvature bounded from below by −λ2 . Then there is a function
C(n, D, λ) such that π1 (M ) can be generated by C(n, D, λ) elements.
• Remark: Since M is complete and bounded in the above theorem, M is com-
pact (unlike in the previous theorem).
• Proof: The following modifications in the proof of the previous theorem are
needed:
28

1. The law of cosines in the plane with sectional curvature −λ2 in the above
notation is
cosh(λlij ) = cosh(λli ) cosh(λlj ) − sinh(λli ) sinh(λlj ) cos(αij ).
2. Let ε > 0 be arbitrary. Then π1 (M ) is generated by elements gi with
|gi | < 2D + ε. To see this consider a loop σ representing some element h
of π1 (M ). Decompose γ into consecutive segments σi of length ≤ ε. For
each segment σi pick a path βi from x0 to the starting point of σi such
that the length of βi is at most D. For the first segment choose β1 to
be the constant path at x0 . Then βi ∗ σi ∗ βi+1 (βi+1 denotes orientation
reversal of βi+1 , ∗ the concatination of paths) has length at most 2D + ε.
The product of these loops represents h. Hence li ≤ 2D since ε was
arbitrary.
A computation similar to (20) above shows that αij , i < j, is bounded from
below (independent of i, j, of course) by α(D, λ) > 0. At the end of the previous
proof replace π/3 by this α(D, λ).

20. Lecture on July, 20 – Closed geodesics, translations

• Fact: Let γ0 , γ1 represent elements of π1 (M, x0 ). Then γ0 and γ1 are freely
homotopic (i.e. homotopic without fixed base point) if and only if γ0 is conjugate
to γ1 in π1 (M, x0 ). In particular, γ0 is freely null homotopic if and only if
γ0 = 1 ∈ π1 (M, x0 ).
• Theorem: Let M be compact and γ0 : S 1 −→ M represents a nontrivial
element of pi1 (M, x0 ). Then there is a closed geodesic γ : S 1 −→ M (i.e. γ̇1 is
parallel) which is freely homotopic to γ0 .
• Proof: Let d := inf{l(γ) | γ a loop freely homotopc to γ0 }. Show that there is a
sequence of piecewise smooth geodesics γj which are parametrized by arclength
such that limj l(γj ) = d. Use the Arzelà-Ascoli Theorem to extract a convergent
subsequence and show that the limit γ is a closed geodesic which is freely
homotopic to γ0 .
• Reference: For the theorem of Arzelà-Ascoli see Theorem II.3.4 in [We]. For
the relationship between the characteristic subgroup of a covering and the group
of decktransformations see Chapter 9.7 of [J].
• Definition: Let f : M f −→ M f be an isometry without fixed points. f is a
translation if there is a geodesic γ : R −→ Mf such that f (γ(R)) = γ(R). (Then
f is a translation along γ and γ is the axis of f .)
• Proposition: Let M be compact, M f −→ M
f a covering and f : M f a non-trivial
covering transformation. Then f is a translation along a geodesic.
• Proof: Let xe∈M f be a basepoint and γ a closed path from x e to f (e
x). γ projects
to a closed loop in M which is homotopically non-trivial and therefore freely
homotopic to a closed geodesic. f is then a translation along a certain lift of γ
to M
f.

21. Lecture on July, 24 – Preissman’s theorem, topology of compact

manifolds with K < 0
• Lemma: Let M f be complete, simply connected with K < 0 and f a fixed point
free isometry along γ. Then γ is the unique axis of f .
 29

• Proof: By contradiction. If there are two axis one finds a rectangle with two
opposite sides on the two axis and interior angle sum = 2π. This leads to
a contradiction to the fact that the angle sum of a non-degenerate geodesic
triangle is in (0, π) when K < 0.
• Lemma: Let M f be as above, f, g two commuting isometries along axis. Then
their axis coincide.
• Theorem (Preissman): Let M be a compact manifold with K < 0 and A an
Abelian subgroup of π1 (M ). Then A is trivial or infinite cyclic.
• Proof: Consider the action of π1 (M ) on the universal covering. The elements of
A act by isometries and properly discontinuously on a geodesic γ(R). Moreover,
two deck transformations coincide when the coincide on a point. Hence A is
cyclic.
• Corollary: Let M be compact with π1 (M ) 6= {1}. Then M × S 1 does not
admit a metric of negative sectional curvature.
• Lemma: Let M be a complete manifold with K < 0 and γ e an axis in the
universal cover such that all elements of the Deck group preserve γ e. Then M
is not compact.
• Proof: Let x0 be a point on γ e and βe a unit speed geodesic through β(0)e = x0 ,
orthogonal to γ e and parametrized by arc length. Let αt be a minimal geodesic
in M connecting pr(β(t))
e to pr(x0 ). Then l(αt ) ≤ t. If one lifts αt to M
f with
starting point β(t)
e one arrives at a point g(x0 ) on γ
e with g a deck transformation
(because all decktransformations preserve γ e. If g 6= id, then one obtains a non-
degenerate geodesic triangle in Mf and since K < 0

t2 ≥ l(lift of αt )2 > l(β([0, t]))2 + d(x0 , g(x0 ))2 ≥ t2 .

Hence g(x0 ) = x0 and the lift of αt is the unique geodesic from β(t)
e to x0 . Then
l(αt ) = t and M cannot be bounded.
• Corollary: If M is compact and K < 0, then π1 (M ) is not Abelian, i.e.
π1 (M ) 6' Z.
• Definition: Let H be a group. Then H is solvable if there are subgroups
H = H0 ⊃ H1 ⊃ . . . Hk−1 ⊃ Hk = {1} such that Hi is a normal subgroup in
Hi−1 and Hi−1 /Hi is Abelian.
• Theorem (Byers): Let M be as in Preissman’s theorem and H a solvable
subgroup of π1 (M ). Then H is trivial or infinite cyclic.
• Proof: Hk−1 is Abelian, hence cyclic and there is an axis γ in the universal
covering on which Hk−1 acts. Let 1 6= a ∈ Hk−2 and 1 6= b ∈ Hk−1 . Then
aba−1 b−1 ∈ Hk−1 . Hence aba−1 b−1 preserves γ as set, b also preserves γ, hence
aba−1 preserves γ as set. Then b preserves a−1 (γ). Since b has a unique axis
a−1 (γ) = γ. Thus Hk−2 acts freely by isometries on γ. Hence Hk−2 is Abelian,
etc.

Literatur
[BJ] T. Bröcker, K. Jänich, Einführung in die Differentialtopologie, Springer 1990.
[dC] M. do Carmo, Riemannian geometry, Birkäuser.
[CE] J. Cheeger, D. Ebin, Comparison theorems in Riemannian geometry, North-Holland 1975.
[J] K. Jänich, Topologie, Springer (1996).
[Ma] W. Massey, A basic course in Algebraic topology, Graduate texts in Mathematics 127, Sprin-
ger 1991.
30

[Me] W. Meyer, Toponogov’s theorem and applications, lecture notes, available at:
https://www.math.upenn.edu/∼wziller/math660/TopogonovTheorem-Myer.pdf
[Mi] J. Milnor, Morse theory, Annals of Math. Studies Vol 51, Princeton Univ. Press.
[Q] B. von Querenburg, Mengentheoretische Topologie, Springer.
[Wa] W. Walter, Analysis 2, Springer Verlag, 3. Auflage 1991.
[We] D. Werner, Funktionalanalysis, Springer 1995.
[Wo] J. Wolf, Spaces of constant curvature, McGraw-Hill.