ANALYSIS AND GEOMETRY ON MANIFOLDS

ULRICH PINKALL

Contents

1. n-Dimensional Manifolds 1
2. Tangent Vectors 8
3. The tangent bundle as a smooth vector bundle 11
4. Vector bundles 13
5. Vector fields as operators on functions 15
6. Connections on vector bundles 16
7. Wedge product 21
8. Pullback 23
9. Curvature 24
10. Fundamental theorem for flat vector bundles 28
11. Affine connections 30
12. Flat Riemannian manifolds 31
13. Geodesics 33
14. The exponential map 36
15. Complete Riemannian manifolds 38
16. Sectional curvature 40
17. Jacobi fields 41
18. Second variational formula 42
19. Bonnet-Myers’s theorem 43

1. n-Dimensional Manifolds

1.1. Introduction. Informally, an n-dimensional manifold is a ”space” which locally (when

looked at through a microscope) looks like ”flat space” Rn .
Many important examples of manifolds M arise as certain subsets M ⊂ Rk , e.g.:
(1) n-dimensional affine subspaces M ⊂ Rk ,
(2) Sn = {x ∈ Rn+1 | x20 + · · · x2n = 1},
Date: January 23, 2018.
1
2 ULRICH PINKALL

(3) compact 2-dimensional submanifolds of R3

(4) SO(3) = {A ∈ R3×3 | At A = Id} is a 3-dimensional submanifold of R9 .
Flat spaces (vector spaces ∼ = Rn ) are everywhere. Curved manifolds come up in Stochastics,
Algebraic Geometry, . . . , Economics and Physics – e.g. as the configuration space of a pendulum
(S2 ), a double pendulum (S2 × S2 ) or rigid body motion (SO(3)), or as space time in general
relativity (the curved version of flat special relativity).

1.2. Crash Course in Topology.

Definition 1 (Topological space). A topological space is a set M together with a subset O ⊂
P(M) (the collection of all ”open sets”) such that:
(1) ∅, M ∈ O,
(2) Uα ∈ O, α ∈ I ⇒ ∪α∈I Uα ∈ O,
(3) U1 , . . . , Un ∈ O, ⇒ U1 ∩ · · · ∩ Un ∈ O.
Remark 1: Usually we suppress the the collection O of open sets and just say M is a topological
space. If several topologies and spaces are involved we use an index to make clear which topology
corresponds to which space.
Some ways to make new topological spaces out of given ones:
a) Let X be a topological space, M ⊂ X, then OM := {U ∩ M | U ∈ OX } defines a topology on
M – called ”induced topology” or ”subspace topology”.
b) Let X be a topological space, M be a set, and π : X → M a surjective map. Then there is
a bijection between M and the set of equivalence classes of the equivalence relation on X
defined by
x ∼ y ⇔ π(x) = π(y).
In other words: M can be identified with the set of equivalence classes. Conversely, given an
equivalence relation ∼ on a topological space X we can form the set of equivalence classes
M = X/∼ . The canonical projection π : X → M is the surjective map which sends x ∈ X to
the corresponding equivalence class [x]. The quotient topology
OM = {U ⊂ M | π −1 (U ) ∈ OX }
turns M into a topological space. By construction π is continuous.
Exercise 1 (Product topology). Let M and N be topological spaces and define B := {U × V |
U ∈ OM , V ∈ ON }. Show that O := {∪U ∈A U | A ⊂ B} is a topology on M × N.
Definition 2 (Continuity). Let M, N be topological spaces. Then f : M → N is called continuous
if
f −1 (U ) ∈ OM for all U ∈ ON .
Definition 3 (Homeomorphism). A bijective map f : M → N between topological spaces is
called a homeomorphism if f and f −1 are both continuous.
Remark 2: If f : M → N is a homeomorphism, then for U ∈ OM ⇔ f (U ) ∈ ON . So two
topological spaces are topologically indistinguishable, if they are homeomorphic, i.e. if there
exists a homeomorphism f : M → N.
Definition 4 (Hausdorff). A topological space M is called Hausdorff if for all x, y ∈ M with
x 6= y there are open sets Ux , Uy ∈ O with Ux ∩ Uy = ∅.
Example: The quotient space M = R/ ∼ with x ∼ y ⇔ x − y ∈ Q is not Hausdorff.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 3

Definition 5 (Second axiom of countability). A topological space M is said to satisfy the second
axiom of countability (or is called second countable), if there is a countable base of topology, i.e.
there is a sequence of open sets U1 , U2 , U3 , . . . ∈ O such that for every U ∈ O there is a subset
I ⊂ N such that U = ∪α∈I Uα .
Example: The balls of rational radius with rational center in Rn form a countable base of
topology, i.e. Rn is 2nd countable.
Remark 3: Subspaces of second countable spaces are second countable. Hence all subsets of Rn
are second countable. A similar statement holds for the Hausdorff property.
Example: M = R2 with O = {U × {y} | y ∈ R, U ∈ OR } is not second countable.
Definition 6 (Topological manifold). A topological space M is called an n-dimensional topo-
logical manifold if it is Hausdorff, second countable and for every p ∈ M there is an open set
U ∈ O with p ∈ U and a homeomorphism ϕ : U → V , where V ∈ ORn .
Remark 4: A homeomorphism ϕ : U → V as above is called a (coordinate) chart of M.
Exercise 2. Let X be a topological space, x ∈ X and n ≥ 0. Show that the following statements
are equivalent:
i) There is a neighborhood of x which is homeomorphic to Rn .
ii) There is a neighborhood of x which is homeomorphic to an open subset of Rn .
Exercise 3. Show that a manifold M is locally compact, i.e. each point of M has a compact
neighborhood.
Exercise 4 (Connectedness). A topological space M is connected if the only subsets of X which
are simultaneously open and closed are X and ∅. Moreover, X is called path-connected if any
two points x, y ∈ X can be joined by a path, i.e. there is a continuous map γ : [a, b] → X such
that γ(0) = x and γ(1) = y. Show that a manifold is connected if and only it is path-connected.

Given two charts ϕ : U → Rn and ψ : V → Rn , then the map f : ϕ(U ∩ V ) → ψ(U ∩ V ) given
by f = ψ ◦ ( ϕ|U ∩ V )−1 is a homeomorphism, called the coordinate change or transition map.
Definition 7 (Atlas). An atlas of a manifold M is a collection of charts {(Uα , ϕα )}α∈I such
that M = ∪α∈I Uα .
Definition 8 (Compatible charts). Two charts ϕ : U → Rn , ψ : V → Rn on a topological
manifold M are called compatible if f : ϕ(U ∩ V ) → ψ(U ∩ V ) is a diffeomorphism, i.e. f and
f −1 both are smooth.
Example: Consider M = Sn ⊂ Rn+1 . Let B = {y ∈ Rn | kyk ≤ 1}. Define charts as follows:
For i = 0, . . . , n,
Ui± = {x ∈ S2 | ±xi > 0}, ϕ± ±
i , : Ui → B, ϕ±
i (x0 , . . . , xn ) = (x0 , . . . , x
“i , . . . , xn ),

where the hat means omission. Check that ϕi are homeomorphisms. So: (Since Sn as a subset
of Rn+1 is Hausdorff and second countable) Sn is an n-dimensional topological manifold. All
ϕ± n
i are compatible, so this atlas turns S into a smooth manifold.

An atlas {(Uα , ϕα )}α∈I of mutually compatible charts on M is called maximal if every chart
(U, ϕ) on M which is compatible with all charts in {(Uα , ϕα )}α∈I is already contained in the
atlas.
Definition 9 (Smooth manifold). A differentiable structure on a topological manifold M is a
maximal atlas of compatible charts. A smooth manifold is a topological manifold together with
a maximal atlas.
4 ULRICH PINKALL

Exercise 5 (Real projective space). Let n ∈ N and X := Rn+1 \ {0}. The quotient space
RPn = X/∼ with equivalence relation given by
x ∼ y :⇐⇒ x = λy, λ∈R
is called the n-dimensional real projective space. Let π : X → RPn denote the canonical projec-
tion x 7→ [x].
For i = 0, . . . , n, we define Ui := π({x ∈ X | xi 6= 0}) and ϕi : Ui → Rn by
[x0 , . . . , xn ] 7→ (x0 /xi , . . . , x“i , . . . , xn /xi ).
Show that
a) π is an open map, i.e. maps open sets in X to open sets in RPn ,
b) the maps ϕi are well-defined and {(Ui , ϕi )}i∈I is a smooth atlas of RPn ,
c) RPn is compact. Hint: Note that the restriction of π to Sn is surjective.
Exercise 6 (Product manifolds). Let M and N be topological manifolds of dimension m and
n, respectively. Show that their Cartesian product M × N is a topological manifold of dimension
m + n. Show further that, if {(Uα , ϕα )}α∈A is a smooth atlas of M and {(Vβ , ψβ )}β∈B is a
smooth atlas of N, then {(Uα × Vβ , ϕα × ψβ )}(α,β)∈A×B is a smooth atlas of M × N. Here
ϕα × ψβ : Uα × Vβ → ϕα (Uα ) × ψβ (Vβ ) is defined by ϕα × ψβ (p, q) := (ϕα (p), ψβ (q)).
Exercise 7 (Torus). Let Rn /Zn denote the quotient space Rn /∼ where the equivalence relation
is given by
x ∼ y :⇔ x − y ∈ Zn .
Let π : Rn → Rn /Zn , x 7→ [x] denote the canonical projection. Show:
a) π is a covering map, i.e. a continuous surjective map such that each point p ∈ Rn /Zn has
a open neighborhood V such that π −1 (V ) is a disjoint union of open sets each of which is
mapped by π homeomorphically to V .
b) π is an open map.
c) Rn /Zn is a manifold of dimension n.
d) {( π|U )−1 | U ⊂ Rn open, π|U : U → π(U ) bijective} is a smooth atlas of Rn /Zn .
Definition 10 (Smooth map). Let M and M̃ be smooth manifolds. Then a map f : M → M̃ is
called smooth if for every chart (U, ϕ) of M and every chart (V, ψ) of M̃ the map
ϕ(f −1 (V ) ∩ U ) → ψ(V ), x 7→ ψ(f (ϕ−1 (x)))
is smooth.
Definition 11 (Diffeomorphism). Let M and M̃ be smooth manifolds. Then a bijective map
f : M → M̃ is called a diffeomorphism if both f and f −1 are smooth.

One important task in Differential Topology is to classify all smooth manifolds up to diffeomor-
phism.
Example: Every connected one-dimensional smooth manifold is diffeomorphic to R or S1 . Ex-
amples of 2-dimensional manifolds: [pictures missing: compact genus 0,1,2,... Klein bottle,
or torus with holes (non-compact)] - gets much more complicated already. For 3-dimensional
manifolds there is no list.
Exercise 8. Show that the following manifolds are diffeomorphic.
a) R2 /Z2 .
b) the product manifold S1 × S1 .
 ANALYSIS AND GEOMETRY ON MANIFOLDS 5

c) the torus of revolution as a submanifold of R3 :

¶ ©
T = ((R + r cos ϕ) cos θ, (R + r cos ϕ) sin θ, r sin ϕ) | ϕ, θ ∈ R .

1.3. Submanifolds.
Definition 12 (Submanifold). A subset M ⊂ M̃ in a k-dimensional smooth manifold M̃ is
called an n-dimensional submanifold if for every point p ∈ M there is a chart ϕ : U → V of M̃
with p ∈ U such that
ϕ(U ∩ M) = V ∩(Rn × {0}) ⊂ Rk .

Let us briefly restrict attention to M̃ = Rk .

Theorem 1. Let M ⊂ Rk be a subset. Then the following are equivalent:
a) M is an n-dimensional submanifold,
b) locally M looks like the graph of a map from Rn to Rk−n , which means: For every point
p ∈ M there are open sets V ⊂ Rn and W ⊂ M, W 3 p, a smooth map f : V → Rk−n and a
coordinate permutation π : Rk → Rk , π(x1 , ..., xk ) = (xσ1 , ..., xσk ) such that
π(W ) = {(x, f (x)) | x ∈ V },
c) locally M is the zero set of some smooth map into Rk−n , which means: For every p ∈ M
there is an open set U ⊂ Rk , U 3 p and a smooth map g : U → Rk−n such that
M ∩ U = {x ∈ U | g(x) = 0}
and the Jacobian g 0 (x) has full rank for all x ∈ M,
d) locally M can be parametrized by open sets in Rn , which means: For every p ∈ M there are
open sets W ⊂ M, W 3 p, V ⊂ Rn and a smooth map ψ : V → Rk such that ψ maps V
bijectively onto W and ψ 0 (x) has full rank for all x ∈ V .

Proof. First, recall two theorems from analysis:

• The inverse function theorem: Let U ⊂ Rn be open, p ∈ U , f : U → Rn continuously
differentiable, f 0 (p) 6= 0. Then there is a an open’ subset Ũ ⊂ U , Ũ 3 p and an open
subset V ⊂ Rn , V 3 f (p) such that
(1) f |Ũ : Ũ → V is bijective,
(2) f −1 : V → Ũ is continuously differentiable.
We have (f −1 )0 (q) = f 0 (f −1 (q))−1 for all q ∈ V . We in fact need a version where
’continuously differentiable’ is replaced by C ∞ . Let us prove the C 2 version. Then all
the partial derivatives of first order for f −1 are entries of (f −1 )0 . So we have to prove
that q 7→ (f −1 )0 (q) = (f 0 )−1 (f −1 (q)) is continuously differentiable. This follows from
the smoothness of the map GL(n, R) 3 A 7→ A−1 ∈ GL(n, R) (Cramer’s rule), the chain
rule and the fact that f 0 : Ũ → Rn×n is continuously differentiable. The general case
can be done by induction.
• The implicit function theorem (C ∞ − version): Let U ⊂ Rk be open, p ∈ U , g : U →
Rk−n smooth, g(p) = 0, g 0 (p) is surjective. Then, after reordering the coordinates
of Rk , we find open subsets V ⊂ Rn , W ⊂ Rn−k such that (p1 , . . . , pn ) ∈ V and
(pn+1 , . . . , pk ) ∈ W and V × W ⊂ U . Moreover, there is a smooth map f : V → W such
that {q ∈ V × W | g(q) = 0} = {(x, f (x)) | x ∈ V }.
Proof of Theorem 1. b) ⇒ a): Let p ∈ M. By b) after reordering coordinates in Rk we find
open sets V ∈ Rn , W ⊂ Rk−n such that p ∈ V × W and we find a smooth map f : V → W
6 ULRICH PINKALL

such that V × W ∩ M = {(x, f (x)) | x ∈ V }. Then ϕ : V × W → Rk , (x, y) 7→ (x, y − f (x)) is a

diffeomorphism and ϕ(M ∩(V × W )) ⊂ Rn × {0}. [picture missing]
a) ⇒ c): Let p ∈ M. By a) we find an open U ∈ Rk , U 3 p and a diffeomorphism ϕ : U →
Û ⊂ Rk such that ϕ(U ∩ M) ⊂ Rn × {0}. Now define g : U → Rk−n to be the last k − n
component functions of ϕ, i.e. ϕ = (ϕ1 , . . . , ϕn , g1 , . . . , gk−n ). Then M ∩(V × W ) = g −1 ({0}).
For q ∈ V × W we have
 
∗
 .. 

 . 

0
 ∗ 
ϕ (q) =  0 .
 
 g1 (q) 
..
 
 
 . 
0
gk−n (q)
Hence g 0 has rank k − n. c) ⇒ b) is just the implicit function theorem. Let us look at b) ⇒ d).
Let p ∈ M. After reordering the coordinates by b) we have an open neighborhood of p of the
form V × W and a smooth map f : V → W such that M ∩(V × W ) = {(x, f (x)) | x ∈ V }. Now
define ψ : V → Rk by ψ(x) = (x, f (x)). Then ψ is smooth
Ç å
0 IdRn
ψ (x) =
f 0 (x)

So ψ 0 (x) has rank n for all x ∈ V . Moreover, ψ(V ) = M × (V × W ). d) ⇒ b): Let p ∈ M.

Then by d) there are open sets V̂ ⊂ Rn , U ⊂ Rk , U 3 p and a smooth map ψ : V̂ → Rk such
that ψ(V̂ ) = M ∩ U such that rank ψ 0 (x) is n for all x ∈ V̂ . After reordering the coordinates
on Rk we can assume that ψ = (φ, fˆ)t with φ : V̂ → Rn with det φ0 (x0 ) 6= 0, where ψ(x0 ) = p.
Passing to a smaller neighborhood V ⊂ V̂ , V 3 p, we then achieve that φ : V → φ(V ) is a
diffeomorphism (by the inverse function theorem). Now for all y ∈ φ(V ) we have
φ(φ−1 (y)
Ç å Ç å
−1 y
ψ(φ (y)) = =:
fˆ(φ−1 (y)) f (φ−1 (y))


1.4. Examples of submanifolds in Rk .

a) Sn = {x ∈ Rn+1 | x21 +x22 +· · ·+x2n+1 = 1} is an n-dimensional submanifold (a hypersurface) of

Rn+1 , because Sn = {x ∈ Rn+1 | g(x) = 0}, where g : Rn+1 → R, g(x) = x21 +x22 +· · · x2n+1 −1.
We have to check that g 0 (x) has rank 1 on g −1 ({0}): We have g 0 (x) = 2x 6= 0 for x 6= 0.
2 2
b) O(n) ⊂ Rn×n = Rn , O(n) = {A ∈ Rn×n | At A = I} is a submanifold of Rn of dimension
n(n − 1)/2. Define g : Rn×n → Sym(n) = Rn(n−1)/2 by g(A) = At A − I.
a11 a12 · · · a1n
á ë
∗ a22 · · · a1n
.. .. ..
. . .
∗ ∗ ··· ann
Entries above and including the diagonal: n + (n + 1) + · · · + 2 + 1 = n(n − 1)/2. Need to
check that g 0 (A) : Rn×n → Sym(n) is surjective for all A ∈ O(n).
Interlude: Derivatives of maps f : U → Rm , where U ⊂ Rk open. f 0 (p) : Rk → Rm linear.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 7

How to calculate f 0 (p)X for X ∈ Rk ? Choose smooth γ : (−ε, ε) → Rk such that γ(0) = p
and γ 0 (0) = X. Then by the chain rule
(f ◦ γ)0 (0) = f 0 (γ(0))γ 0 (0) = f 0 (p)X.
So let A ∈ O(n), X ∈ Rn×n , B : (−ε, ε) → Rn×n with B(0) = A, B 0 (0) = X (e.g. B(t) =
A + tX). Then
g 0 (A)X = d
dt t=0 g(B(t)) = d t
dt t=0 [B(t) B(t) − I]
t 0 t 0 t t
= (B ) (0)B(0) + B (0)B (0) = X A + A X.
To check that g 0 (A) is surjective, let Y ∈ Sym(n) be arbitrary. So Y ∈ Rn×n , Y t = Y . There
is X ∈ Rn×n with X t A + At X = Y , e.g. X = 12 AY :
X t A + At X = 21 (Y t At A + At AY ) = Y.

So O(n) is a submanifold dimension n2 − n(n+1)

2 = n(n−1)
2 .
c) Consider the set Gk (R ) the set of k-dimensional linear subspaces of Rn . We represent a
n

linear subspace U ⊂ Rk by the orthogonal projection PU ∈ Rn×n onto U . The map PU is

defined by
(1.1) PU |U = IdU , PU |U ⊥ = 0.
PU has the following properties:
PU2 = PU , PU∗ = PU , tr PU = dim U.
In the decomposition Rn = U ⊕ U ⊥ , we have
Ç å
IdU 0
PU = .
0 0
Conversely: If P ∗ = P , then there is an orthonormal basis of Rn with respect to which P is
diagonal.
λ1
Ö è
.. .
.
λn
If further P 2 = P , then λ2i = λi ⇔ λi ∈ {0, 1} for all i ∈ {1, . . . , n}. After reordering the
basis we have Ç å
Ik 0
0 0
for some k < n. So P is the orthogonal projection onto a k-dimensional subspace with
k = tr P . Thus we have
Gk (Rn ) = {P ∈ End(Rn ) | P 2 = P, P ∗ = P, trace P = k}.
We fix a k-dimensional subspace V and define
WV := {L ∈ End(Rn ) | PV ◦ L|V invertible}.
Since WV is open, the intersection Gk (Rn ) ∩ WV is open in the subspace topology.
Fix a k-dimensional subspace V ⊂ Rn . Then a k-dimensional subspace U ⊂ Rn ’close’ to
V is the graph of a linear map Y ∈ Hom(V, V ⊥ ): With respect to the splitting Rn = V ⊕V ⊥ ,
Ç å
IdV
U = Im = {(x, Y x) | x ∈ V }.
Y
8 ULRICH PINKALL

The orthogonal complement U ⊥ of U is then parametrized over V ⊥ by (−Y ∗ , IdV ⊥ ): For

x ∈ V and y ∈ V ⊥ we have
−Y ∗ y ∂
Ç å Ç å
¨ x
, = hx, −Y ∗ yi + hx, Y yi = 0.
Yx y
Since rank (−Y, IdV ⊥ ) is n − k, we get
−Y ∗
Ç å
⊥
U = Im .
IdV ⊥
Further, since the corresponding orthogonal projection PU is symmetric we can write
A B∗
Ç å
PU = ,
B C
with A∗ = A, B ∗ = B. Explicitly A = PV ◦ S|V , B = PV ⊥ ◦ S|V and C = PV ⊥ ◦ S|V ⊥ .
From Equation (1.1) we get
A + B∗Y −Y ∗ −AY ∗ + B ∗
Ç å Ç å Ç å Ç å Ç å
IdV IdV
= PU = , 0 = PU = .
Y Y B + CY IdV ⊥ −BY ∗ + C
In particular, Y ∗ = A−1 B ∗ and, since A is self-adjoint,
(1.2) Y = BA−1 .
If we plug this relation into the equation IdV = A + B ∗ Y we get IdV = A(IdV + Y ∗ Y ).
Since hY ∗ Y x, xi = hY x, Y xi ≥ 0 the map IdV + Y ∗ Y is always invertible. This yields
A = (IdV + Y ∗ Y )−1 . In particular, PU ∈ WV ∩ Gk (Rn ). Further, since AY ∗ = B ∗ , we get
that B = Y (IdV + Y ∗ Y )−1 and, together with C = BY ∗ , C = Y (IdV + Y ∗ Y )−1 Y ∗ . Hence
(IdV + Y ∗ Y )−1 (IdV + Y ∗ Y )−1 Y ∗
Ç å
(1.3) PU = ∈ WV ∩ Gk (Rn ).
Y (IdV + Y ∗ Y )−1 Y (IdV + Y ∗ Y )−1 Y ∗
Equation (1.3) actually defines a smooth map φ : Hom(V, V ⊥ ) → WV ∩ Gk (Rn ) with left
inverse given by Equation (1.2), which is smooth on WV . Hence φ is surjective and has full
rank. Thus Gk (Rn ) is locally parametrized by Hom(V, V ⊥ ) ∼
= Rk·(n−k) .
Theorem 2. The Grassmannian Gk (Rn ) of k-planes in Rn (represented by the orthogonal
projection onto these subspaces) is a submanifold of dimension k(n − k).
Exercise 9. Show that G1 (R3 ) ⊂ Sym(3) is diffeomorphic to RP2 .
Exercise 10 (Möbius band). Show that the Möbius band (without boundary)
M = ((2 + r cos ϕ2 ) cos ϕ, (2 + r cos ϕ2 ) sin ϕ, r sin ϕ2 ) | r ∈ (− 21 , 12 ), ϕ ∈ R
¶ ©

is a submanifold of R3 . Show further that for each point p ∈ RP2 the open set RP2 \ {p} ⊂ RP2
is diffeomorphic to M.

2. Tangent Vectors

Let M be an n-dimensional smooth manifold. We will define for each p ∈ M an n-dimensional

vector space Tp M, the tangent space of M at p.
Definition 13 (Tangent space). Let M be a smooth n-manifold and p ∈ M. A tangent vector
X at p is then a linear map
X : C ∞ (M) → R, f 7→ Xf
 ANALYSIS AND GEOMETRY ON MANIFOLDS 9

such that there is a smooth curve γ : (−ε, ε) → M with γ(0) = p and

Xf = (f ◦ γ)0 (0).
The tangent space is then the set of all tangent vectors Tp M := {X | X tangent vector at p}.

Let ϕ = (x1 , . . . , xn ) be a chart defined on U 3 p. Let f˜ = f ◦ ϕ−1 , γ̃ = ϕ ◦ γ and p̃ = ϕ(p).

Then
Ö 0
γ̃1 (0)
è

Xf = (f ◦ γ)0 (0) = (f˜ ◦ γ̃)0 (0) = ∂1 f˜(p̃), . . . , ∂n f˜(p̃)

Ä ä .. .
.
0
γ̃n (0)
So tangent vectors can be parametrized by n numbers αi = γ̃i0 (0):
Xf = α1 ∂1 f˜(p̃) + · · · + αn ∂n f˜(p̃).
Exercise 11. Within the setup above, show that to each vector α ∈ Rn , there exists a curve
γ : (−ε, ε) → M such that γ(0) = p and (f ◦ γ)0 (0) = α1 ∂1 f˜(p̃) + · · · + αn ∂n f˜(p̃).
Definition 14 (Coordinate frame). If ϕ = (x1 , . . . , xn ) is a chart at p ∈ M, f ∈ C ∞ (M). Then
∂
f := ∂i (f ◦ ϕ−1 )(ϕ(p)), i = 1, . . . , n.
∂xi p

Interlude: How to construct C ∞ functions on the whole of M? Toolbox: f : R → R with

®
0 for x ≤ 0,
f (x) =
e−1/x for x > 0.
is C ∞ and so is then g(x) = f (1 − x2 ) and h(x) = 0x g. From h this we can build a smooth
R

function ĥ : R → [0, 1] with ĥ(x) = 1 for x ∈ [− 14 , 14 ] and ĥ(x) = 0 for x ∈ R \ (−1, 1). Then we
can define a smooth function h̃ : Rn → R by h̃(x) = ĥ(x21 + · · · x2n ) which vanishes outside the
unit ball and is constant = 1 inside the ball of radius 12 .
Theorem 3. Let M be a smooth n-manifold, p ∈ M and (U, ϕ) a chart with U 3 p. Let
a1 , . . . , an ∈ R. Then there is f ∈ C ∞ (M) such that
∂
f = ai , i = 1, . . . , n.
∂xi p

Proof. We define g̃ : Rn → R, g̃(x) = h̃(λ(x − ϕ(p))) with λ such that g̃(x) = 0 for all x 6∈ ϕ(U ).
Then let f˜: Rn → R, f˜(x) := g̃(x)(a1 x1 + a2 x2 + · · · + an xn ). Then f : M → R given by
f˜(ϕ(q))
®
for q ∈ U,
f (q) =
0 for q ∈
6 U
is such a function. 
∂ ∂
Corollary 1. ∂x1 p , . . . , ∂xn p are linearly independent.

Corollary 2. Tp M ⊂ C ∞ (M)∗ is an n-dimensional linear subspace.

Proof. Follows from the last corollary and from Exercise 11, which shows that Tp M is a subspace
spanned by ∂x∂ 1 , . . . , ∂x∂n . 
p p
10 ULRICH PINKALL

Theorem 4 (Transformation of coordinate frames). If (U, ϕ) and (V, ψ) are charts with p ∈
U ∩ V , ϕ|U ∩ V = Φ ◦ ψ|U ∩ V . Then for every X ∈ Tp M,
X ∂ X ∂
X= ai = bi ,
∂xi p ∂yi p

where ϕ = (x1 , . . . , xn ), ψ = (y1 , . . . , yn ), we have

a1 b1
Ö è Ö è
.. = Φ0 (ψ(p)) .. .
. .
an bn

Proof. Let γ : (−ε, ε) → M such that Xf = (f ◦ γ)0 (0). Let γ̃ = ϕ ◦ γ and γ̂ = ψ ◦ γ, then
a = γ̃ 0 (0), b = γ̂ 0 (0).
Let Φ : ψ(U ∩ V ) → ϕ(U ∩ V ) be the coordinate change Φ = ϕ ◦ ψ −1 . Then
γ̃ = ϕ ◦ γ = φ ◦ ψ ◦ γ = Φ ◦ γ̂.
In particular,
a = γ̃ 0 (0) = (Φ ◦ γ̃)0 (0) = Φ0 (ψ(p))γ̂ 0 (0) = Φ0 (ψ(p))b.

Definition 15. Let M and M̃ be smooth manifolds, f : M → M̃ smooth, p ∈ M. Then define a
linear map dp f : Tp M → Tf (p) M̃ by setting for g ∈ C ∞ (M̃) and X ∈ Tp M
dp f (X)g := X(g ◦ f ).
Remark 5: dp f (X) is really a tangent vector in Tp M̃ because, if X corresponds to a curve
γ : (−ε, ε) → M with γ(0) = p then
d d d
dp f (X)g = (g ◦ f ) ◦ γ = g ◦ (f ◦ γ ) = g ◦ γ̃.
dt t=0 dt t=0 | {z } dt t=0
=:γ̃

Notation: The tangent vector X ∈ Tp M corresponding to a curve γ : (−ε, ε) → M with

γ(0) = p is denoted by X =: γ 0 (0).
Theorem 5 (Chain rule). Suppose g : M → M̃, f : M̃ → M̂ are smooth maps. Then
dp (f ◦ g) = dg(p) f ◦ dp g.
Definition 16 (Tangent bundle). TM := tp∈M Tp M is called the tangent bundle of M. The
map π : TM → M, Tp M 3 X → p is called the projection map. So Tp M = π −1 ({p}).

Elegant version of the chain rule: If f : M → M̃ is smooth, then df : M → M̃ where df (X) =

dπ(X) f (X). With this notation,
d(f ◦ g) = df ◦ dg.
Theorem 6. If f : M → M̃ is a diffeomorphism then for each p ∈ M the map dp f : Tp M →
Tf (p) M̃ is a vector space isomorphism.

Proof. f is bijective and f −1 is smooth, IdM = f −1 ◦ f . For all p ∈ M,

IdTp M = dp (IdM ) = df (p) f −1 ◦ dp f.
So dp f is invertible. 
 ANALYSIS AND GEOMETRY ON MANIFOLDS 11

Theorem 7 (Manifold version of the inverse function theorem). Let f : M → M̃ be smooth,

p ∈ M with dp f : M → M̃ invertible. Then there are open neighborhoods U ⊂ M of p and V ⊂ M̃
of f (p) such that f |U : U → V is a diffeomorphism.

Proof. The theorem is a reformulation of the inverse function theorem. 

Theorem 8 (Manifold version of the implicit function theorem - ’submersion theorem’). Let
f : M̃ → M̂ be a submersion, i.e. for each p ∈ M̃ the derivative dp f : Tp M̃ → Tf (p) M̂ is
surjective. Let q = f (p) be fixed. Then
M := f −1 ({q})
is an n-dimensional submanifold of M̃, where n = dim M̃ − dim M̂.

Proof. Take charts and apply Theorem 1. 

Theorem 9 (Immersion theorem). Let f : M → M̃ be an immersion, i.e. for every p ∈ M the
differential dp f : Tp M → Tf (p) M is injective. Then for each p ∈ M there is an open set U ⊂ M
with p ∈ M such that f (U ) is a submanifold of M̃.

Proof. Take charts and apply Theorem 1. 

Is there a global version, i.e. without passing to U ⊂ M? Assuming that f is injective is not
enough.
Exercise 12. Let f : N → M be a smooth immersion. Prove: If f is moreover a topological
embedding, i.e. its restriction f : N → f (N) is a homeomorphism between N and f (N) (with its
subspace topology), then f (N) is a smooth submanifold of M.
Exercise 13. Let M be compact, f : M → M̃ an injective immersion, then f (M) is a submani-
fold.
Exercise 14. Let X := C2 \{0}. The complex projective plane is the quotient space CP1 = X/∼ ,
where the equivalence relation is given by
ψ ∼ ψ̃ :⇔ λψ = ψ̃, λ ∈ C.
Consider S3 ⊂ R4 ∼
= C2 , then the Hopf fibration is the map
π : S3 → CP1 , ψ 7→ [ψ].
Show: For each p ∈ CP1 the fiber π −1 ({p}) is a submanifold diffeomorphic to S1 .

3. The tangent bundle as a smooth vector bundle

Let M be a smooth n-manifold, p ∈ M. The tangent space at p is an n-dimensional subspace of

(C ∞ (M))∗ given by
Tp M = {X ∈ (C ∞ (M))∗ | ∃γ : (−ε, ε) → M, γ(0) = p, Xf = (f ◦ γ)0 (0) = Xf, ∀f ∈ C ∞ (M)}.
The tangent bundle is then the set
TM = t Tp M
p∈M
and comes with a projection π : TM → M, Tp M 3 X 7→ p ∈ M. The set π −1 ({p}) = Tp M is
called the fiber of the tangent bundle at p.
Goal: We want to make TM into a 2n-dimensional manifold.
12 ULRICH PINKALL

∂ ∂
If ϕ = (x1 , . . . , xn ) be a chart of M defined on U 3 p. Then we have a basis ∂x1 p , . . . , ∂xn p of
Tp M. So there are unique y1 (X), . . . , yn (X) ∈ R such that
X ∂
X= yi (X) .
∂xi p

Let {(Uα , ϕα )}α∈A be a smooth atlas of M. For each α ∈ A we get an open set Ûα := π −1 (Uα )
and a function yα : Ûα → Rn which maps a given vector to the coordinates with respect to the
frame defined by ϕα , yα = (yα,1 , . . . , yα,n ). Now, we define ϕ̂α : π −1 (U ) → Rn × Rn = R2n by
ϕ̂α = (ϕα ◦ π, yα ).
For any two charts we have a transition map φαβ : ϕα (Uα ∩ Uβ ) → ϕβ (Uα ∩ Uβ ) such that
ϕβ |Uα ∩ Uβ = φαβ ◦ ϕα |Uα ∩ Uβ . The chain rule yields:

yβ (X) = φ0αβ (ϕα (π(X)))yα (X).

Hence we see that ϕ̂β ◦ ϕ̂−1

α is a diffeomorphism.
Topology on TM:

OTM := W ⊂ TM | ϕ̂α (W ∩ Ûα ) ∈ OR2n for all α ∈ A .

¶ ©

Exercise 15. a) This defines a topology on TM.

b) With this topology TM is Hausdorff and 2nd-countable.
c) All ϕ̂α are homeomrophisms onto their image.

Because coordinate changes are smooth, this turns TM into a smooth 2n-dimensional manifold.
Definition 17 (Vector field). A (smooth) vector field on a manifold M is a smooth map
X : M → TM with π ◦ X = IdM , i.e. X(p) ∈ Tp M for all p ∈ M. Usually we write Xp
instead of X(p). If X is a vector field and f ∈ C ∞ (M), then Xf ∈ C ∞ (M ) is given by
(Xf )(p) = Xp f . Read: ”X differentiates f ”.
Exercise 16. Show that each of the following conditions is equivalent to the smoothness of a
vector field X as a section X : M → TM:
a) For each f ∈ C ∞ (M), the function Xf is also smooth.
∂
in a coordinate chart ϕ = (x1 , . . . , xn ) defined on U ⊂ M, then
P
b) If we write X|U =: vi ∂x i
the components vi : U → R are smooth.
Exercise 17. On S2 = {x = (x0 , x1 , x2 ) | kxk = 1} ⊂ R3 we consider coordinates given by the
stereographic projection from the north pole N = (1, 0, 0):
x1 x2
y1 = 1−x0 , y2 = 1−x0 .

Let the vector fields X and Y on S2 \ {N } be defined in these coordinates by

X = y2 ∂y∂ 1 − y1 ∂y∂ 2 , Y = y1 ∂y∂ 1 + y2 ∂y∂ 2 .

Express these two vector fields in coordinates corresponding to the stereographic projection from
the south pole S = (−1, 0, 0).
Exercise 18. Prove that the tangent bundle of a product of smooth manifolds is diffeomorphic
to the product of the tangent bundles of the manifolds. Deduce that the tangent bundle of a
torus S1 × S1 is diffeomorphic to S1 × S1 × R2 .
 ANALYSIS AND GEOMETRY ON MANIFOLDS 13

4. Vector bundles

Definition 18 (Vector bundle). A smooth vector bundle of rank k is a triple (E, M, π) which
consists of smooth manifolds E and M and a smooth map π : E → M such that for each p ∈ M
the fiber Ep := π −1 ({p}) has the structure of a k-dimensional vector space and each p ∈ M has
an open neighborhood U ⊂ M such that there exists a diffeomorphism
φ : π −1 (U ) → U × Rk
such that πU ◦φ = π and for each p ∈ M the restriction πRk ◦ φ|Ep is a vector space isomorphism.
Definition 19. Let E be a smooth vector bundle over M. A section of E is a smooth map
ψ : M → E such that π ◦ ψ = IdM . Γ(E) := {ψ : M → E | ψ section of E}.
Example: a) We have seen that the tangent bundle TM of a smooth manifold is a vector
bundle of rank dim M. Its smooth sections were called vector fields.
b) The product M×Rk is called the trivial bundle of rank k. Its smooth sections can be identified
with Rk -valued functions. More precisely, if π2 : M × Rk → Rk , then
Γ(M × Rk ) 3 ψ ←→ f := π2 ◦ ψ ∈ C ∞ (M).
From now on we will keep this identification in mind.

4.1. Ways to make new vector bundles out of old ones. General principle: Any linear
algebra operation that given new vector spaces out of given ones can be applied to vector
bundles over the same base manifold.
Example: Let E be a rank k vector bundle over M and F be a rank ` vector bundle over M.
a) Then E ⊕ F denotes the rank k + ` vector bundle over M the fibers of which are given by
(E ⊕ F)p = Ep ⊕ Fp .
b) Then Hom(E, F) denotes the rank k·` vector bundle over M with fiber given by Hom(E, F)p :=
{f : Ep → Fp | f linear}.
c) E∗ = Hom(E, M × R) with fibers (E∗ )p = (Ep )∗ .
Let E1 , . . . .Er , F be vector bundles over M.
d) Then a there is new vector bundle E∗1 ⊗ · · · ⊗ E∗r ⊗ F of rank rankE1 · · · rankEr · rankF with
fiber at p given by E∗1p ⊗ · · · ⊗ E∗rp ⊗ Fp = {β : E1p × . . . × Erp → Fp | β multilinear}.
Exercise 19. Give an explicit description of the (natural) bundle charts for the bundles (written
down as sets) in the previous example.

Starting from TM:

a) T∗ M := (TM)∗ is called the cotangent bundle.
b) Bundles of multilinear forms with all the E1 , . . . , Er , F copies of TM, T∗ M or M × R are
called tensor bundles. Sections of such bundles are called tensor fields.
Example: We have seen that Gk (Rn ) = {Orthogonal projections onto k-dim subspaces of Rn }
is an (n − k)k-dimensional submanifold of Sym(n). Now, we can define the tautological bundle
as follows:
E = {(P, v) ∈ Gk (Rn ) | P v = v}.
W is an open neighborhood of PV as described in the Grassmannian example. Then for (PU , v) ∈
E define φ(PU , v) ∈ W × V ∼
= W × Rk by φ(Pu , v) = (PU , PV v). Check that this defines a local
trivialization.
14 ULRICH PINKALL

Exercise 20. Let M ⊂ Rk be a smooth submanifold of dimension n. Let ι : M ,→ Rk denote

the inclusion map. Show that the normal bundle NM = tp∈M (Tp M)⊥ ⊂ ι∗ TRk ∼
= M × Rk is a
smooth rank k − n vector bundle over M.
Definition 20 (pullback bundle). Given a smooth map f : M → M̃ and a vector bundle E → M̃.
Then the pullback bundle f ∗ E is defined as the disjoint union of the fibers (f ∗ E)p = Ef (p) , i.e.
f ∗ E = t Ef (p) ⊂ M × E.
p∈M

Exercise 21. f ∗E is a smooth submanifold of M × E.

Definition 21 (Vector bundle isomorphism). Two vector bundles E → M, Ẽ → M are called
isomorphic if there is a bundle isomorphism f : E → Ẽ, i.e. π̃ ◦ f = π (fibers to fibers) and
f |Ep : Ep → Ẽp is a vector space isomorphism.

Fact (without proof): Every rank k vector bundle E over M is isomorphic to f ∗ Ẽ, where Ẽ is
the tautological bundle over Gk (Rn ) (some n) and some smooth f : M → Gk (Rn ).
Definition 22. A vector bundle E → M of rank k is called trivial if it is isomorphic to the
trivial bundle M × Rk .
Remark 6: If E → M is a vector bundle of rank k then, by definition, each point p ∈ M has an
open neighborhood U such that the restricted bundle E|U := π −1 (U ) is trivial, i.e. each bundle
is locally trivial.
Definition 23 (Frame field). Let E → M be a rank k vector bundle, ϕ1 , . . . , ϕk ∈ Γ(E). Then
(ϕ1 , . . . , ϕk ) is called a frame field if for each p ∈ M the vectors ϕ1 (p), . . . , ϕk (p) ∈ Ep form a
basis.
Proposition 1. E is trivial if and only if E has a frame field.

Proof. ”⇒”: E trivial ⇒ ∃F ∈ ΓHom(E, M × Rk ) such that Fp : Ep → {p} × Rk is a vector

space isomorphism for each p. Then, for i = 1, . . . , k define ϕi ∈ Γ(E) by ϕp = F −1 ({p} × ei ).
”⇐”: (ϕ1 , . . . , ϕk ) frame field define F ∈ ΓHom(E, M × Rk ) as the unique map such that
Fp (ϕi (p)) = {p} × ei for each p ∈ M. F is a bundle isomorphism. 

From the definition of a vector bundle: Each p ∈ M has a neighborhood U such that E|U has
a frame field.
Theorem 10. For each p ∈ M there is an open neighborhood U and ϕ1 , . . . , ϕk ∈ Γ(E) such
that ϕ1 |U , . . . , ϕk |U is a frame field of E|U .

Proof. There is an open neighborhood Ũ of p such that E|Ũ is trivial. Thus there is a frame
field ϕ̃1 , . . . , ϕ̃k ∈ Γ( E|Ũ ). There is a subset U ⊂ Ũ , a compact subset C with U ⊂ C ⊂ Ũ and
a smooth function f ∈ C ∞ (M) such that f |U ≡ 1 and fM\C ≡ 0. Then, on Ũ , we define
ϕi (q) = f (q)ϕ̃i (q), i = 1, . . . , n,
and extend it by the 0-vector field to whole of M, i.e. ϕi (q) = 0 ∈ Eq for q ∈ M \ Ũ . 

Example: A rank 1 vector bundle E (a line bundle) is trivial ⇔ ∃ nowhere vanishing ϕ ∈ Γ(E).
Example: M ⊂ R` submanifold of dimension n rank ` − n vector bundle NM (the normal
bundle of M) is given by Np M = (NM) = (Tp M)⊥ ⊂ Tp R` = {p} × R` . Fact: The normal
bundle of a Moebius band is not trivial.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 15

Example: The tangent bundle of S2 is not trivial - a fact known as the hairy ball theorem:
Every vector field X ∈ Γ(T S2 ) has zeros.
Exercise 22. Show that the tangent bundle TS3 of the round sphere S3 ⊂ R4 is trivial.
Hint: Show that the vector fields ϕ1 (x1 , x2 , x3 , x4 ) = (−x2 , x1 , x4 , −x3 ), ϕ2 (x1 , x2 , x3 , x4 ) = (x3 , x4 , −x1 , −x2 )
and ϕ3 (x1 , x2 , x3 , x4 ) = (−x4 , x3 , −x2 , x1 ) form a frame of TS3 .

5. Vector fields as operators on functions

Let X ∈ Γ(TM), f ∈ C ∞ (M). Then Xf : M → R, p 7→ Xp f , is smooth. So X can be viewed as

a linear map C ∞ (M) → C ∞ (M),
f 7→ Xf.
Theorem 11 (Leibniz’s rule). Let f, g ∈ C ∞ (M), X ∈ Γ(TM), then X(f g) = (Xf )g + f (Xg).
Definition 24 (Lie algebra). A Lie algebra is a vector space g together with a skew bilinear
map [., .] : g × g → g which satisfies the Jacobi identity,
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
Theorem 12 (Lie algebra of endomorphisms). Let V be a vector space. End(V) together with
the commutator [., .] : End(V) × End(V) → End(V), [A, B] := AB − BA forms a Lie algebra.

Proof. Certainly the commutaor is a skew bilinear map. Further,

[A, [B, C]] + [B, [C, A]] + [C, [A, B]] = A(BC − CB) − (BC − CB)A + B(CA − AC)
− (CA − AC)B + C(AB − BA) − (AB − BA)C,
which is zero since each term appears twice but with opposite sign. 

Theorem 13. For all f, g ∈ C ∞ (M), X, Y ∈ Γ(M), [f X, gY ] = f g[X, Y ] + f (Xg)Y − g(Y f )X.
∂
Lemma 1 (Schwarz lemma). Let ϕ = (x1 , . . . , xn ) be a coordinate chart. Then [ ∂x , ∂ ] = 0.
i ∂xj

Exercise 23. Prove Schwarz lemma above.

P ∂ P ∂
Thus, if X = i ai ∂xi and Y = j bj ∂xj , we get
X ∂ ∂ XÄ ∂bj ∂ ∂ai ∂ ä XÄ ∂bi ∂ai ä ∂
[X, Y ] = [ai , bj ]= ai − bj = aj − bj .
i,j
∂xi ∂xj i,j
∂xi ∂xj ∂xj ∂xi i,j
∂xj ∂xj ∂xi

Thus [X, Y ] ∈ Γ(TM). In particular, we get the following theorem.

Theorem 14. Γ(TM) ⊂ End(C ∞ (M)) is a Lie subalgebra.
Exercise 24. Calculate the commutator [X, Y ] of the following vector fields on R2 \ {0}:
x ∂ y ∂ ∂ ∂
X=p +p 2 , Y = −y +x .
x2 2
+ y ∂x 2
x + y ∂y ∂x ∂y
Write X and Y in polar coordinates (r cos ϕ, r sin ϕ) 7→ (r, ϕ).
Definition 25 (Push forward). Let f : M → N be a diffeomorphism and X ∈ Γ(TM). The
push forward f∗ X ∈ Γ(TN) of X is defined by f∗ X := df ◦ X ◦ f −1 .
Exercise 25. Let f : M → N be a diffeomorphism, X, Y ∈ Γ(TM). Show: f∗ [X, Y ] =
[f∗ X, f∗ Y ].
16 ULRICH PINKALL

6. Connections on vector bundles

Up to now we did basically Differential Topology. Now Differential Geometry begins, i.e. we
study manifolds with additional (”geometric”) structure.
Definition 26 (Connection). A connection on a vector bundle E → M is a bilinear map
∇ : Γ(TM) × Γ(E) → Γ(E) such that for all f ∈ C ∞ (M), X ∈ Γ(TM), ψ ∈ Γ(E),
∇f X ψ = f ∇X ψ, ∇X f ψ = (Xf )ψ + f ∇X ψ.

The proof of the following theorem will be postponed until we have established the existence of
a so called partition of unity.
Theorem 15. On every vector bundle E there is a connection ∇.
Definition 27 (Parallel section). Let E → M be a vector bundle with connection ∇. Then
ψ ∈ Γ(E) is called parallel if ∇X ψ = 0 for all X ∈ TM.
˜ be two connections on E. Define A : Γ(TM) × Γ(E) → Γ(E) by AX ψ = ∇
Let ∇, ∇ ˜ X ψ − ∇X ψ.
Then A satisfies
˜ f X ψ − ∇f X ψ = f AX ψ
Af X ψ = ∇
and
AX (f ψ) = · · · = f AX ψ.
Suppose we have ω ∈ ΓHom(TM, End E). Then define B : Γ(TM) × Γ(E) → Γ(E) by
(BX ψ)p = ωp (Xp )(ψp ) ∈ Ep .
Then
Bf X ψ = f BX ψ, BX (f ψ) = f BX ψ.
Theorem 16 (Characterization of tensors). Let E, F be vector bundles over M and A : Γ(E) →
Γ(F) linear such that for all f ∈ C ∞ (M), ψ ∈ Γ(E) we have
A(f ψ) = f A(ψ).
Then there is ω ∈ ΓHom(E, F) such that (Aψ)p = ωp (ψp ) for all ψ ∈ Γ(E), p ∈ M.

Proof. Let p ∈ M, ψ̃ ∈ Ep . Want to define ω by saying: Choose ψ ∈ Γ(E) such that ψp = ψ̃.
Then define ωp (ψ̃) = (Aψ)p . Claim: (Aψ)p depends only on ψp , i.e. if ψ, ψ̂ ∈ Γ(E) with ψp = ψ̂p
then (Aψ)p = (Aψ̂)p , or in other words: ψ ∈ Γ(E) with ψp = 0 then (Aψ)p = 0. To check this
choose a frame field (ψ1 , . . . , ψk ) on some neighborhood and a function f ∈ C ∞ (M) such that
f ψ1 , . . . , f ψk are globally defined sections and f ≡ 1 near p. Let ψ ∈ Γ(E) with ψp = 0.
ψ| U = a1 ψ1 + · · · + ak ψk with a1 , . . . , ak ∈ C ∞ (U ). Then
f 2 Aψ = A(f 2 ψ) = A((f a1 )(f ψ1 ) + · · · + (f ak )(f ψk )) = (f a1 )A(f ψ1 ) + · · · + (f ak )A(f ψk )).
Evaluation at p yields then (Aψ)p = 0. 
Remark 7: In the following we keep this identification be tensors and tensorial maps in mind
and just speak of tensors.
Thus the considerations above can be summarized by the following theorem.
Theorem 17. Any two connections ∇ and ∇ ˜ on a vector bundle E over M differ by a section
of Hom(TM, End E):
˜ − ∇ ∈ ΓHom(TM, End E).
∇
 ANALYSIS AND GEOMETRY ON MANIFOLDS 17

Exercise 26 (Induced connections). Let Ei and F denote vector bundles with connections ∇i
and ∇, respectively. Show that the equation
ˆ X T )(Y1 , . . . , Yr ) = ∇X (T (Y1 , . . . , Yr )) −
X
(∇ T (Y1 , . . . , ∇iX Yi , . . . , Yr )
i
ˆ on the bundle
for T ∈ Γ(E∗1 ⊗ · · · ⊗ E∗r ⊗ F) and vector fields Yi ∈ Γ(Ei ) defines a connection ∇
∗ ∗
of multilinear forms E1 ⊗ · · · ⊗ Er ⊗ F.
Remark 8: Note that, since an isomorphism ρ : E → Ẽ between vector bundles over M maps
for each p ∈ M the fiber of Ep linearly to the fiber Ẽp , the map ρ can be regarded as a section
ρ ∈ ΓHom(E, Ẽ). If moreover E is equipped with a connection ∇ and Ẽ is equipped with a
connection ∇ ˜ we can speak then of parallel isomorphisms: ρ is called parallel if ∇ρ
ˆ = 0, where
∇ˆ is the connection on Hom(E, Ẽ) induced by ∇ and ∇ ˜ (compare Example 26 above).
Definition 28 (Metric). Let E → M be a vector bundle and Sym(E) be the bundle whose fiber
at p ∈ M consists of all symmetric bilinear forms Ep × Ep → R. A metric on E is a section h., .i
of Sym(E) such that h., .ip is a Euclidean inner product for all p ∈ M. A vector bundle together
with a metric (a pair (E, h., .i)) is called Euclidean vector bundle.
Definition 29 (Metric connection). Let (E, h., .i) be a Euclidean vector bundle over M. Then
a connection ∇ is called metric if for all ψ, ϕ ∈ Γ(E) and X ∈ Γ(TM) we have
Xhψ, ϕi = h∇X ψ, ϕi + hψ, ∇X ϕi.
Exercise 27. Let ∇ be a connection on a direct sum E = E1 ⊕ E2 of two vector bundles over
M. Show that Ç 1 å
∇ A
∇= ,
à ∇2
where à ∈ Ω1 (M, Hom(E1 , E2 )), A ∈ Ω1 (M, Hom(E2 , E1 )) and ∇i are connections on the bun-
dles Ei .

Recall: A rank k vector bundle E → M is called trivial if it is isomorphic to the trivial bundle
M × Rk . We know that
E trivial ⇔ ∃ϕ1 , . . . , ϕk ∈ Γ(E) : ϕ1 (p), . . . , ϕk (p) linearly independent for all p ∈ M.
The trivial bundle comes with a trivial connection ∇trivial : Γ(M × Rk ) 3 ψ ↔ f = π2 ◦ ψ ∈
C ∞ (M, Rk ), then ∇trivial
X ψ ↔ dX f = Xf , X ∈ Γ(TM). More precisely,
∇trivial
X ψ = (π(X), Xf ).
This clarified in the following the trivial connection often will be denoted just by d.
Every vector bundle E is locally trivial, i.e. each point p ∈ M has an open neighborhood U such
that E|U is trivial.
Definition 30 (Isomorphism of vector bundles with connection). An isomorphism between
˜ is a vector bundle isomorphism ρ : E → Ẽ,
vector bundles with connection (E, ∇) and (Ẽ, ∇)
which is parallel, i.e. for all X ∈ Γ(TM), ψ ∈ Γ(E),
˜ X (ρ ◦ ψ) = ρ ◦ (∇X ψ).
∇
Two vector bundles with connection are called isomorphic if there exists an isomorphism between
them. A vector bundle with connection (E, ∇) over M is called trivial if it is isomorphic to the
trivial bundle (M × Rk , d).
Remark 9: Note that ψ ∈ Γ(M × Rk ) is parallel if π ◦ ψ is locally constant.
18 ULRICH PINKALL

Theorem 18. A vector bundle E with connection is trivial iff there exists a parallel frame field.

Proof. ”⇒”: Let ρ : M × Rk → E be a bundle isomorhism such that ρ ◦ d = ∇ ◦ ρ. Then

φip := ρ(p, ei ), i = 1, . . . , k, form a parallel frame. ”⇐”: If we have a parallel frame field
ϕi ∈ Γ(E), then define ρ : M × Rk → E, ρ(p, v) := vi ϕi (p). It is easily checked that ρ is the
P

desired isomorphism. 
Definition 31 (Flat vector bundle). A vector bundle E with connection is called flat if it is
locally trivial as a vector bundle with connection, i.e. each point p ∈ M has an open neighborhood
U such that E|U (endowed with the connection inherited from E) is trivial. In other words: If
there is a parallel frame field over U .
Definition 32 (Bundle-valued differential forms). Let E → M be a vector bundle. Then for
` > 0 an E-valued `-form ω is a section of the bundle Λ` (M, E) whose fiber at p ∈ M is the
vector space of multilinear maps Tp M × · · · × Tp M → Ep , which are alternating, i.e. for i 6= j
ωp (X1 , . . . , Xi , . . . , Xj , . . . , X` ) = −ωp (X1 , . . . , Xj , . . . , Xi , . . . , X` ).
Further, define Λ0 (M, E) := E. Consequently, Ω0 (M, E) := Γ(E).
Remark 10: Each ω ∈ Ω` (M, E) defines a tensorial map Γ(TM)` → Γ(E) and vice versa.
Definition 33 (Exterior derivative). Let E → M be a vector bundle with connection ∇. For
` ≥ 0, define the exterior derivative d∇ : Ω` (M, E) → Ω`+1 (M, E) as follows:
d∇ ω(X0 , . . . , X` ) =
X
(−1)i ∇Xi (ω(X0 , . . . , X̂i , . . . , X` ))
i
X
i+j
+ (−1) ω([Xi , Xj ], X0 , . . . , X̂i , . . . , X̂j , . . . , X` ), X0 , . . . , X` ∈ Γ(TM).
i<j

Proof. Actually there are two things to be verified: d∇ ω is tensorial and alternating. First let
us check it is tensorial:
d∇ ω(X0 , . . . , f Xk , . . . , X` ) =
X
(−1)i ∇Xi ω(X0 , . . . , X̂i , . . . , f Xk , . . . , X` )
i<k

+ ∇f Xk ω(X0 , . . . , X̂k , . . . , X` )
X
+ (−1)i ∇Xi ω(X0 , . . . , f Xk , . . . , X̂i , . . . , X` )
i>k
X
+ (−1)i+j ω([Xi , Xj ], . . . , X̂i , . . . , f Xk , . . . , X̂j , . . . , X` )
i<j,i6=k,j6=k
X
+ (−1)i+k ω([Xi , f Xk ], . . . , X̂i , . . . , X̂k , . . . , X` )
i<k
X
+ (−1)k+i ω([f Xk , f Xi ], . . . , X̂k , . . . , X̂i , . . . , X` )
k<i
∇
= f d ω(X0 , . . . , f Xk , . . . , X` )
X
+ (−1)i (Xi f )ω(X0 , . . . , X̂i , . . . , X` )
i6=k
X
+ (−1)i+k ω((Xi f )Xk , . . . , X̂i , . . . , X̂k , . . . , X` )
i<k
X
− (−1)k+i ω((Xi f )Xk , . . . , X̂k , . . . , X̂i , . . . , X` )
k<i
 ANALYSIS AND GEOMETRY ON MANIFOLDS 19

= f d∇ ω(X0 , . . . , f Xk , . . . , X` ).

Next we want to see that d∇ ω is alternating. Since d∇ ω is tensorial we can test this on
commuting vector fields, i.e [Xi , Xj ] = 0. With this we get for k < m that

d∇ ω(X0 , . . . , Xm , . . . , Xk , . . . , X` ) =
X
(−1)i ∇Xi ω(X0 , . . . , X̂i , . . . , Xm , . . . , Xk , . . . , X` )
i<k

+ (−1)k ∇Xm ω(X0 , . . . , X̂m , . . . , Xk , . . . , X` )

X
+ (−1)i ∇Xi ω(X0 , . . . , Xm , . . . , X̂i , . . . , Xk , . . . , X` )
k<i<m
m
+ (−1) ∇Xk ω(X0 , . . . , Xm , . . . , X̂m , . . . , X` )
X
+ (−1)i ∇Xi ω(X0 , . . . , Xm , . . . , Xk , . . . , X̂i , . . . , X` )
i>k
X
=− (−1)i ∇Xi ω(X0 , . . . , X̂i , . . . , Xk , . . . , Xm , . . . , X` )
i<k

+ (−1)k+(m−k−1) ∇Xm ω(X0 , . . . , Xk , . . . , X̂k , . . . , X` )

X
− (−1)i ∇Xi ω(X0 , . . . , Xk , . . . , X̂i , . . . , Xm , . . . , X` )
k<i<m
m+(m−k−1)
+ (−1) ∇Xk ω(X0 , . . . , X̂k , . . . , Xm , . . . , X` )
X
i
− (−1) ∇Xi ω(X0 , . . . , Xk , . . . , Xm , . . . , X̂i , . . . , X` )
i>k
X
= (−1)i ∇Xi ω(X0 , . . . , X̂i , . . . , X` )
i<k
= −d∇ ω(X0 , . . . , Xk , . . . , Xm , . . . , X` ),

where the second equation follows by successively shifting the vector fields Xm resp. Xk to the
right resp. left. 

1-forms: Let E → M be a vector bundle with connection ∇, then Λ1 (M, E) = Hom(TM, E).
We have Ω0 (M, E) = Γ(E). We obtain a 1-forms by applying d∇ :

Ω0 (M, E) 3 ψ 7→ d∇ ψ = ∇ψ ∈ Ω1 (M, E).

As a special case we have E = M × R. Then Γ(M × R) ↔ C ∞ (M) and Λ1 (M, M × R) =

Hom(TM, M × R) ↔ Hom(TM, R) = T∗ M. So in this case Ω1 (M, M × R) ∼ = Γ(T∗ M) = Ω1 (M)
(ordinary 1-forms are basically sections of T∗ M). For M = U ⊂ Rn (open) we have the standard
∂
coordinates xi : U → R (projection to the i-component) dxi ∈ Ω1 (M). Let Xi := ∂x i
∈ Γ(TU )
n
which as R -valued functions is just the canonical basis Xi = ei . Then X1 , . . . , Xn is a frame:
We have dxi (Xj ) = δij , thus dx1 , . . . , dxn is the frame of T∗ U dual to X1 , . . . , Xn . So every
1-form is of the form:

ω = a1 dx1 + · · · + an dxn , a1 , . . . , an ∈ C ∞ (U ).
∂f
If f ∈ C ∞ (U ), then Xi f = ∂xi . With a small computation we get

∂f ∂f
df = dx1 + · · · + dxn .
∂x1 ∂xn
20 ULRICH PINKALL

`-forms: Let M ⊂ Rn be open and consider again E = M × R. Then for i1 , . . . , i` define

dxi1 ∧ · · · ∧ dxi` ∈ Ω` (M) by
dxi1 (X1 ) · · · dxi1 (X` )
Ö è

dxi1 ∧ · · · ∧ dxi` (X1 , . . . , X` ) := det .. .. .. .

. . .
dxi` (X1 ) · · · dxi` (X` )
Note: If iα = iβ for α 6= β, then dxi1 ∧ · · · ∧ dxi` = 0. If σ : {1, . . . , `} → {1, . . . , `} is a
permutation, we have
dxiσ1 ∧ · · · ∧ dxiσ` = sign σ dxi1 ∧ · · · ∧ dxi` .
Theorem 19. Let U ⊂ Rn be open. The `-forms dxi1 ∧ · · · ∧ dxi` for 1 ≤ i1 < · · · < i` ≤ n are
a frame field for Λ` (U ), i.e. each ω ∈ Ω` (U ) can be uniquely written as
X
ω= ai1 ···i` dxi1 ∧ · · · ∧ dxi`
1≤i1 <···<i` ≤n

with ai1 ···i` ∈ C ∞ (U ). In fact,

Ä ∂ ∂ ä
ai1 ···i` = ω ,..., .
∂xi1 ∂xi`

Proof. For uniqueness note that

δi1 j1 ··· δi1 j`
Ö è
®
Ä ∂ ∂ ä .. .. 1 if {i1 , . . . , i` } = {j1 , . . . , j` },
dxi1 ∧· · ·∧dxi` ,..., = det .. =
. . . 0 else.
∂xj1 ∂xj`
δi` j1 ··· δi` j`
Existence we leave as an exercise. 
Theorem 20. Let U ⊂ Rn be open and ω = dxi1 ∧ · · · ∧ dxi` ∈ Ω` (U ), then
P
1≤i1 <···<i` ≤n ai1 ···i`
n
X X ∂ai1 ···i`
dω = dxi ∧ dxi1 ∧ · · · ∧ dxi` .
1≤i1 <···<i` ≤n i=1
∂xi

Proof. By Theorem 19 it is enough to show that for all 1 ≤ j0 < · · · < j` ≤ n

Ä ∂ n
∂ ä X X ∂ai1 ···i` Ä ∂ ∂ ä
dω ,..., = dxi ∧ dxi1 ∧ · · · ∧ dxi` ,...,
∂xj0 ∂xj` ∂xi ∂xj0 ∂xj`
1≤i <···<i1 ` ≤n i=1
` ∂a Ä ∂
X j0 ···ĵk ···j` ∂ ä
= dxjk ∧ dxj0 · · · ∧ dx
‘ jk · · · ∧ dxj` ,...,
k=0
∂xjk ∂xj0 ∂xj`
` ∂aj
0 ···jk ···j`
X
= (−1)k .
“
k=0
∂xjk
But we also get this sum if we apply the definition and use that [ ∂x∂ k , ∂x∂m ] = 0. 
Example: Let M = U ⊂ R3 be open. Then every σ ∈ Ω2 (M) can be uniquely written as
σ = a1 dx2 ∧ dx3 + a2 dx3 ∧ dx1 + a3 dx1 ∧ dx2 .
Let σ = dω with ω = v1 dx1 + v2 dx2 + v3 dx3 . Then
Ä ∂ ∂ ä ∂ Ä ∂ ä ∂ Ä ∂ ä ∂vj ∂vi
dω , = ω − ω = − .
∂xi ∂xj ∂xi ∂xj ∂xj ∂xi ∂xi ∂xj
Thus we get that a = curl(v).
 ANALYSIS AND GEOMETRY ON MANIFOLDS 21

The proofs of Theorem 19 and Theorem 20 directly carry over to bundle-valued forms.
Theorem 21. Let U ⊂ Rn be open and E → U be a vector bundle with connection ∇. Then
ω ∈ Ω` (U, E) can be uniquely written as
X
ω= ψi1 ···i` dxi1 ∧ · · · ∧ dxi` , ψi1 ···i` ∈ Γ(E).
1≤i1 <···<i` ≤n
Moreover,
n Ä
d∇ ω =
X X ä
∇ ∂ ψi1 ···i` dxi ∧ dxi1 ∧ · · · ∧ dxi` .
∂xi
1≤i1 <···<i` ≤n i=1

Exercise 28. Let M = R2 . Let J ∈ Γ(EndTM) be the 90◦ rotation and det ∈ Ω2 (M) denote
the determinant. Define ∗ : Ω1 (M) → Ω1 (M) by ∗ω(X) = −ω(JX). Show that
2 2
a) for all f ∈ C ∞ (M), d ∗ df = (∆f ) det, where ∆f = ∂x ∂ ∂
2 f + ∂y 2 f ,

b) ω ∈ Ω1 (M) is closed (i.e. dω = 0), if and only if ω is exact (i.e. ω = df ).

7. Wedge product

Let U, V, W be vector bundles over M. Let ω ∈ Ωk (M, U), η ∈ Ω` (M, V). We want to define
ω ∧ η ∈ Ωk+` (M, W). Therefore we need a multiplication ∗ : Up × Vp → Wp bilinear such that
for ψ ∈ Γ(U), φ ∈ Γ(V) such that ψ ∗ φ : p 7→ ψp ∗p φp is smooth, i.e. ψ ∗ φ ∈ Γ(W). In short,
∗ ∈ Γ(U∗ ⊗ V∗ ⊗ W).
Example: a) Most standard case: U = M × R = V, ∗ ordinary multiplication in R.
b) Also useful: U = M × Rk×` , V = M × R`×m , W = M × Rk×m , ∗ matrix multiplication.
c) Another case: U = End(E), V = W = E, ∗ evaluation of endomorphisms on vectors, i.e.
(A ∗ ψ)p = Ap (ψp ).
Definition 34 (Wedge product). Let U, V, W be vector bundles over M and ∗ ∈ Γ(U∗ ⊗V∗ ⊗W).
For two forms ω ∈ Ωk (M, U) and η ∈ Ω` (M, V) the wedge product ω ∧ η ∈ Ωk+` (M, W) is then
defined as follows
1 X
ω ∧ η(X1 , . . . , Xk+` ) := sgn σ ω(Xσ1 , . . . , Xσk ) ∗ η(Xσk+1 , . . . , Xσk+` ).
k!`! σ∈S
k+`

Example (Wedge product of 1-forms): For ω, η ∈ Ω1 (M) we have ω(X)η(Y ) − ω(Y )η(X).
Theorem 22. Let U, V, W be vector bundles over M, ∗ ∈ Γ(U∗ ⊗ V∗ ⊗ W), ˜∗ ∈ Γ(V∗ ⊗ U∗ ⊗ W)
∗ ψ for all ψ ∈ Γ(U) and φ ∈ Γ(V), then for ω ∈ Ωk (M, U), η ∈ Ω` (M, V)
such that ψ ∗ φ = φ ˜
we have
ω ∧ η = (−1)k` η ∧ ω.

Proof. The permutation ρ : {1, . . . , k + `} → {1, . . . , k + `} with (1, . . . , k, k + 1, . . . , k + `) 7→

(k + 1, . . . , k + `, 1, . . . , k) needs k` transpositions, i.e. sgn ρ = (−1)k` . Thus
1 X
ω ∧ η(X1 , . . . , Xk+` ) = sgn σ ω(Xσ1 , . . . , Xσk ) ∗ η(Xσk+1 , . . . , Xσk+` )
k!`! σ∈S
k+`

1 X
= sgn σ η(Xσk+1 , . . . , Xσk+` ) ˜∗ ω(Xσ1 , . . . , Xσk )
k!`! σ∈S
k+`

1 X
= sgn (σ ◦ ρ) η(Xσρk+1 , . . . , Xσρk+` ) ˜∗ ω(Xσρ1 , . . . , Xσρk )
k!`! σ∈S
k+`
22 ULRICH PINKALL

(−1)k` X
= sgn σ η(Xσ1 , . . . , Xσk ) ˜∗ ω(Xσk+1 , . . . , Xσk+` )
k!`! σ∈S
k+`
k`
= (−1) η ∧ ω(X1 , . . . , Xk+` ).

Remark 11: In particular the above theorem holds for symmetric tensors ∗ ∈ Γ(U∗ ⊗ U∗ ⊗ V).
Theorem 23. Let E1 , . . . , E6 be vector bundles over M. Suppose that ∗ ∈ Γ(E∗1 ⊗ E∗2 ⊗ E∗4 ),
˜∗ ∈ Γ(E∗4 ⊗ E∗3 ⊗ E5 ), ˚
∗ ∈ Γ(E1 ⊗ E6 ⊗ E5 ) and ˆ∗ ∈ Γ(E∗2 ⊗ E∗3 ⊗ E6 ) be associative, i.e.
(ψ1 ∗ ψ2 ) ˜
∗ ψ3 = ψ1 ˚
∗ (ψ2 ˆ
∗ ψ3 ), for all ψ1 ∈ Γ(E1 ), ψ1 ∈ Γ(E2 ), ψ1 ∈ Γ(E3 ).
Then for ω1 ∈ Ωk1 (M, E1 ), ω2 ∈ Ωk2 (M, E2 ) and ω3 ∈ Ωk3 (M, E3 ) we have
ω1 ∧ (ω2 ∧ ω3 ) = (ω1 ∧ ω2 ) ∧ ω3 .

Proof. To simplify notation: E1 = · · · = E6 = M × R with ordinary multiplication of real

numbers. ω1 = α, ω2 = β, ω3 = γ, k1 = k, k2 = `, k3 = m.
1 X
α ∧ (β ∧ γ)(X1 , . . . , Xk+`+m ) = sgn σ α(Xσ1 , . . . , Xσk )
k!(` + m)! σ∈S
k+`+m

1 X
· sgn ρ β(Xσk+ρ1 , . . . , Xσk+ρ` )γ(Xσk+ρ`+1 , . . . , Xσk+ρ`+m )
m! ρ∈S
`+m

Observe: Fix σ1 , . . . , σk . Then σk+1 , . . . , σk+`+m already account for all possible permutations
of the remaining indices. In effect we get the same term (` + m)! (number of elements in S`+m )
many times. So:
1 X
α ∧ (β ∧ γ)(X1 , . . . , Xk+`+m ) = sgn σ α(Xσ1 , . . . , Xσk )
k!`!m! σ∈S
k+`+m

·β(Xσk+1 , . . . , Xσk+` )γ(Xσk+`+1 , . . . , Xσk+`+m ).

Calculation of (α ∧ β) ∧ γ gives the same result. 

Important special case: On a chart neighborhood (U, ϕ) of M with ϕ = (x1 , . . . , dxn ) we have
X
dxi1 ∧ · · · ∧ dxik (Y1 , . . . , Yk ) = sgn σ dxi1 (Yσ1 ) · · · dxik (Yσk ) = det(dxij (Yk ))j,k ,
σ∈Sk

as was defined previously. In particular, for a bundle-valued form ω ∈ Ω` (M, E) we obtain with
Theorem 21 that
X
ω|U = ψi1 ···i` dxi1 ∧ · · · ∧ dxi` , ψi1 ···i` ∈ Γ( E|U ),
1≤i1 <···<i` ≤n

and
(d∇ ω) d∇ ψi1 ···i` ∧ dxi1 ∧ · · · ∧ dxi` .
X
=
U
1≤i1 <···<i` ≤n

Theorem 24. Let E1 , E2 and E3 be vector bundles over M with connections ∇1 , ∇2 and ∇3 ,
respectively. Let ∗ ∈ Γ(E∗1 ⊗ E∗2 ⊗ E3 ) be parallel, i.e. ∇3 (ψ ∗ ϕ) = (∇1 ψ) ∗ ϕ + ψ ∗ (∇2 ϕ) for
all ψ ∈ Γ(E1 ) and ϕ ∈ Γ(E2 ).. Then, if ω ∈ Ωk (M, E1 ) and η ∈ Ω` (M, E2 ), we have
3 1 2
d∇ (ω ∧ η) = (d∇ ω) ∧ η + (−1)k ω ∧ (d∇ η).
 ANALYSIS AND GEOMETRY ON MANIFOLDS 23

Proof. It is enough to show this locally. For ω = ψ dxi1 ∧ · · · ∧ dxik , η = ϕ dxj1 ∧ · · · ∧ dxj` ,
3 3
d∇ (ω ∧ η) = d∇ (ψ ∗ ϕ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxj` )
3
= d∇ (ψ ∗ ϕ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxj`
1 2
= ((d∇ ψ) ∗ ϕ + ψ ∗ d∇ ϕ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxj`
1
= (d∇ ψ) ∗ ϕ ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxj`
2
+ ψ ∗ (d∇ ϕ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxj`
1
= d∇ ψ ∧ dxi1 ∧ · · · ∧ dxik ∧ ϕ dxj1 ∧ · · · ∧ dxj`
2
+ (−1)k ψ ∧ dxi1 ∧ · · · ∧ dxik ∧ d∇ ϕ ∧ dxj1 ∧ · · · ∧ dxj`
1 2
= (d∇ ω) ∧ η + (−1)k ω ∧ (d∇ η).
Since d∇ is R-linear and the wedge product is bilinear the claim follows. 

8. Pullback

Motivation: A geodesic in M is a curve γ without acceleration, i.e. γ 00 = (γ 0 )0 = 0. But what

a map is γ 0 ? What is the second prime? γ 0 (t) ∈ Tγ(t) M. Modify γ 0 slightly γ“0 (t) = (t, γ 0 (t))
γ“0 ∈ Γ(γ ∗ TM). Right now γ ∗ TM is just a vector bundle over (−ε, ε). If we had a connection
∇ then we can define
“
γ 00 = ∇
“∂γ“0 .
∂t

Definition 35 (Pullback of forms). Let ω ∈ Ωk (M̃, E). Then define f ∗ ω ∈ Ωk (M, f ∗ E) by

(f ∗ ω)(X1 , . . . , Xk ) := (p, ω(df (X1 ), . . . , df (Xk )))
for all p ∈ M, X1 , . . . , Xk ∈ Tp M. For ψ ∈ Ω0 (M̃, E) we have f ∗ ψ = (Id, ψ ◦ f ).

For ordinary k-forms ω ∈ Ωk (M̃) ∼

= Ωk (M̃, M̃×R): (f ∗ ω)(X1 , . . . , Xk ) = ω(df (X1 ), . . . , df (Xk )).
Let E → M̃ be a vector bundle with connection ∇,˜ f : M → M̃.
∗˜ ∗
Theorem 25. There is a unique
Ä connection ä ∇ =: f ∇ on f E such that for all ψ ∈ Γ(E),
∗ ˜
X ∈ Tp M we have ∇X (f ψ) = p, ∇df (X) ψ . In other words

(f ∗ ∇)(f
˜ ∗
ψ) = f ∗ (∇ψ).
˜

Proof. For uniqueness we choose a local frame field ϕ1 , . . . , ϕk around f (p) defined on V ⊂ N
and an open neighborhood U ⊂ M of p such that f (U ) ⊂ V . Then for any ψ ∈ Γ( (f ∗ E)|U )
there are g1 , . . . , gk ∈ C ∞ (U ) such that ψ = j gj f ∗ ϕj . If a connection ∇ on f ∗ E has the
P

desired property then, for X ∈ Tp M,

(Xgj )f ∗ ϕj + gj ∇X (f ∗ ϕj ) = (Xgj )f ∗ ϕj + gj (p, ∇
XÄ ä XÄ ä
∇X ψ = ˜ df (X) ϕj )
j j
∗
XÄ X ä XÄ X ä
= (Xgj )f ϕj + gj (p, ωjk (X)ϕk ) = (p, (Xgj )ϕj ◦ f + gj ωjk (X)ϕk ◦ f ) ,
j k j k

˜ df (X) ϕj =
where ∇
P
k ωjk (X)ϕk ◦ f , ωjk ∈ Ω1 (U ). For existence check that this formula defines
a connection. 
24 ULRICH PINKALL

Theorem 26. Let ω ∈ Ωk (M, U), η ∈ Ω` (M, V) and ∗ ∈ Γ(U∗ ⊗ V∗ ⊗ W). Then
f ∗ (ω ∧ η) = f ∗ ω ∧ f ∗ η.

Proof. Trivial. 
Theorem 27. Let E be a vector bundle with connection ∇ over M̃, f : M → M̃, ω ∈ Ωk (M̃, E).
Then
∗
df ∇ (f ∗ ω) = f ∗ (d∇ ω).

Proof. Without loss of generality we can assume that M̃ ⊂ Rn is open and that ω is of the form
X
ω= ψi1 ···ik dxi1 ∧ · · · ∧ dxik .
1≤i1 <···<ik ≤n

Then
f ∗ω = (f ∗ ψi1 ···ik )f ∗ dxi1 ∧ · · · ∧ f ∗ dxik ,
X

1≤i1 <···<ik ≤n
∇
X
d ω= ∇ψi1 ···ik ∧ dxi1 ∧ · · · ∧ dxik .
1≤i1 <···<ik ≤n

Hence
f ∗ d∇ ω = f ∗ (∇ψi1 ···ik ) ∧ f ∗ dxi1 ∧ · · · ∧ f ∗ dxik
X

1≤i1 <···<ik ≤n

(f ∗ ∇f ∗ ψi1 ···ik ) ∧ dxi1 ◦ df ∧ · · · ∧ dxik ◦ df

X
=
1≤i1 <···<ik ≤n
∗∇
f ∗ ψi1 ···ik ) ∧ d(xi1 ◦ f ) ∧ · · · ∧ d(xik ◦ f )
X
= (df
1≤i1 <···<ik ≤n
∗∇
= df (f ∗ ω).

Exercise 29. Consider the polar coordinate map f : {(r, θ) ∈ R2 | r > 0} → R2 given by
f (r, θ) := (r cos θ, r sin θ) = (x, y). Show that
f ∗ (x dx + y dy) = r dr and f ∗ (x dy − y dx) = r2 dθ.
Theorem 28 (Pullback metric). Let E → M̃ be a Euclidean vector bundle with bundle metric g
and f : M → M̃. Then f ∗ E there is a unique metric f ∗ g such that (f ∗ g)(f ∗ ψ, f ∗ φ) = f ∗ g(ψ, φ)
and f ∗ g is parallel with respect to the pullback connection f ∗ ∇.
Exercise 30. Prove Theorem 28.

9. Curvature

Consider the trivial bundle E = M×Rk , then f ∈ C ∞ (M, Rk ) ↔ ψ ∈ Γ(E) by f ↔ ψ = (IdM , f ).

On E we have the trivial connection ∇:
ψ = (IdM , f ) ∈ Γ(E), X ∈ Γ(TM) ∇X ψ := (IdM , Xf ).
This ∇ satisfies for all X, Y ∈ Γ(TM), ψ ∈ Γ(E):
∇X ∇Y ψ − ∇Y ∇X ψ = ∇[X,Y ] ψ.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 25

Proof. ∇Y ψ = (IdM , Y f ), ∇X ∇Y ψ = (IdM , XY f ), ∇X ∇Y ψ − ∇Y ∇X ψ = (IdM , [X, Y ]f ) =

∂
∇[X,Y ] ψ. In case M ⊂ Rn open, X = ∂x i
, Y = ∂x∂ j [X, Y ] = 0 and the above formula says
∇ ∂ ∇ ∂ ψ=∇ ∂ ∇ ∂ ψ.
∂xi ∂xj ∂xj ∂xi

The equation ∇X ∇Y ψ − ∇Y ∇X ψ − ∇[X,Y ] ψ = 0 reflects the fact that for the trivial connection
partial derivatives commute. Define a map R̃∇ : Γ(TM) × Γ(TM) × Γ(E) → Γ(E) by
(X, Y, ψ) 7→ R̃∇ (X, Y )ψ := ∇X ∇Y ψ − ∇Y ∇X ψ − ∇[X,Y ] ψ.
Theorem 29. Let E be a vector bundle with connection ∇. Then for all X, Y ∈ Γ(TM) and
ψ ∈ Γ(E) we have
R̃∇ (X, Y )ψ = d∇ d∇ ψ(X, Y ).

Proof. In fact,
d∇ (d∇ ψ)(X, Y ) = ∇X (d∇ ψ(Y ))−∇Y (d∇ ψ(X))−d∇ ψ([X, Y ]) = ∇X ∇Y ψ−∇Y ∇X ψ−∇[X,Y ] ψ.

Theorem 30 (Curvature tensor). Let ∇ be a connection on a vector bundle E over M. The
map R̃∇ is tensorial in X, Y and ψ. The corresponding tensor R∇ ∈ Ω2 (M, EndE) such that
[R̃∇ (X, Y )ψ]p = R∇ (Xp , Yp )ψp is called the curvature tensor of ∇.

Proof. Tensoriality in X and Y follows from the last theorem. Remains to show that R̃∇ is
tensorial in ψ:
R̃∇ (X, Y )(f ψ) = ∇X ∇Y (f ψ) − ∇Y ∇X (f ψ) − ∇[X,Y ] (f ψ)
= ∇X ((Y f )ψ + f ∇Y ψ) − ∇Y ((Xf )ψ + f ∇X ψ))ψ − (([X, Y ]f )ψ + f ∇[X,Y ] ψ)
= X(Y f )ψ + (Y f )∇X ψ + (Xf )∇Y ψ + f ∇X ∇Y ψ − Y (Xf )ψ
− (Xf )∇Y ψ − (Y f )∇Y ψ − f ∇Y ∇X ψ − ([X, Y ]f )ψ − f ∇[X,Y ] ψ
= f R̃∇ (X, Y )ψ.

Exercise 31. Let E → M be a vector bundle with connection ∇, ψ ∈ Γ(E) and f : N → M.
Then
∗
(f ∗ R∇ )(f ∗ ψ̃) = f ∗ (R∇ ψ) = Rf ∇ f ∗ ψ.
Lemma 2. Given X̂1 , . . . , X̂k ∈ Tp M, then there are vector fields X1 , . . . , Xk ∈ Γ(TM) such
that X1p = X̂1 , . . . , Xkp = X̂k and there is a neighborhood U 3 p such that [Xi , Xj ]|U = 0.

Proof. We have already seen that we can extend coordinate frames to the whole manifold. This
yields n vector fields Yi such that [Yi , Yj ] vanishes on a neighborhood of p. Since there Yi form a
frame. Then we can build linear combinations of Yi (constant coefficients) to obtain the desired
fields. 
Theorem 31. Let E → M be a vector bundle with connection ∇. For each ω ∈ Ωk (M, E)
d∇ d∇ ω = R∇ ∧ ω.
26 ULRICH PINKALL

Proof. Let p ∈ M, X̂1 , . . . , X̂k+2 ∈ Tp M. Choose X1 , . . . , Xk+2 ∈ Γ(TM) such that Xip = X̂i
and near p we have [Xi , Xj ] = 0, i, j ∈ {1, . . . , k + 2}. The left side is tensorial, so we can use
X1 , . . . , Xk+2 to evaluate d∇ d∇ ω(X̂1 , . . . , X̂k+2 ). Then ij ∈ {1, . . . , k + 2}
k
d∇ ω(Xi0 , . . . , Xik ) =
X
(−1)j ∇Xij ω(Xi0 , . . . , X̂ij , . . . , Xik ).
j=0

Then
d∇ d∇ ω(X1 , . . . , Xk+2 ) =
X
(−1)i+j ∇Xi ∇Xj ω(X1 , . . . , X̂i , . . . , X̂j , . . . , Xk+2 )
i<j
X
+ (−1)i+j+1 ∇Xi ∇Xj ω(X1 , . . . , X̂j , . . . , X̂i , . . . , Xk+2 )
j<i
X
= (−1)i+j (∇Xi ∇Xj − ∇Xj ∇Xi )ω(X1 , . . . , X̂i , . . . , X̂j , . . . , Xk+2 )
i<j

(−1)i+j R∇ (Xi , Xj )ω(X1 , . . . , X̂i , . . . , X̂j , . . . , Xk+2 ).

X
=
i<j

On the other hand

1
R∇ ∧ ω(X1 , . . . , Xk+2 ) = sgn σR∇ (Xσ1 , Xσ2 )ω(Xσ3 , . . . , Xσk+2 ).
X
2 · k! σ∈S
k+2

For i, j ∈ {1, . . . , k + 2}, i 6= j define

¶ ©
A{i,j} := σ ∈ Sk+2 | {σ1 , σ2 } = {i, j} .
ij
For i < j define σ ij ∈ Sk+2 by σ1ij = i and σ2ij = j, σ3ij < · · · < σk+2 , i.e.
σ ij = (i, j, 3, . . . , î, . . . , ĵ, . . . , k + 2).
In particular we find that sgn σ ij = (−1)i+j . Further
A{i,j} = {σ ij ◦ ρ | ρ ∈ Sk+2 , ρ1 = 1, ρ2 = 2} ∪ {σ ij ◦ ρ | ρ ∈ Sk+2 , ρ1 = 2, ρ2 = 1} .
| {z } | {z }
=:A+
{i,j}
=:A−
{i,j}

Note, sgn(σ ij ◦ ρ) = (−1)i+j sgn ρ. With this we get

1 X X
R∇ ∧ ω(X1 , . . . , Xk+2 ) = sgn σR∇ (Xσ1 , Xσ2 )ω(Xσ3 , . . . , Xσk+2 )
2 · k! i<j σ∈A
{i,j}
Ö
1 X
sgn σR∇ (Xσ1 , Xσ2 )ω(Xσ3 , . . . , Xσk+2 )
X
=
2 · k! i<j
σ∈A+
{i,j}
è

sgn σR∇ (Xσ1 , Xσ2 )ω(Xσ3 , . . . , Xσk+2 )

X
+
σ∈A−
{i,j}
Ñ
1 X
(−1)i+j sgn ρR∇ (Xi , Xj )ω(Xσij , . . . , Xσij
X
= )
2 · k! i<j ρ∈Sk+2 ,ρ1 =1,ρ2 =2
ρ3 ρk+2

(−1)i+j sgn ρR∇ (Xj , Xi )ω(Xσij , . . . , Xσij

X
+ )
ρ3 ρk+2
ρ∈Sk+2 ,ρ1 =2,ρ2 =1
 ANALYSIS AND GEOMETRY ON MANIFOLDS 27
Ñ
1 X
(−1)i+j sgn ρR∇ (Xi , Xj )sgn ρ ω(Xσij , . . . , Xσij )
X
=
2 · k! i<j ρ∈Sk+2 ,ρ1 =1,ρ2 =2
3 k+2

(−1)i+j sgn ρR∇ (Xj , Xi )(−sgn ρ) ω(Xσij , . . . , Xσij

X
+ )
ρ3 ρk+2
ρ∈Sk+2 ,ρ1 =2,ρ2 =1

(−1)i+j R∇ (Xi , Xj )ω(X1 , . . . , X̂i , . . . , X̂j , . . . , Xk+2 )

X
=
i<j

Lemma 3. Let E → M be a vector bundle, p ∈ M, ψ̃ ∈ Ep , A ∈ Hom(Tp M, Ep ). Then there is
ψ ∈ Γ(E) such that ψp = ψ̃ and ∇X ψ = A(X) for all X ∈ Tp M.

Proof. Choose a frame field ϕ1 , . . . , ϕk of E near p. Then we have near p

k k k
βi ϕip for β ∈ (Tp M)∗ ,
X X X
∇ X ϕi = αij (X)ϕj , for αij ∈ Ω1 (M), A(X) = ψ̂ = ai ϕip .
j=1 i=1 i=1

requirements on fi . Certainly fi (p) = ai . Further, for X ∈ Tp M,

P
Ansatz: ψ = i fi ϕi near p
X XÄ X ä
βi (X)ϕip = ∇X ψ = dfi (X)ϕip + fi (p) αij (X)ϕjp .
i j

With fi (p) = ai , X
βi = dfi + aj αji .
j
Such fi are easy to find. 
Theorem 32 (Second Bianchi identity). Let E be a vector bundle with connection ∇. Then its
curvature tensor R∇ ∈ Ω2 (M, End(E)) satisfies
d∇ R∇ = 0.

Proof 1. By the last two lemmas we can just choose X0 , X1 , X3 ∈ Γ(TM) commuting near p
and ψ ∈ Γ(E) with ∇X ψ = 0 for all X ∈ Tp M. Then near p
R∇ (Xi , Xj )ψ = ∇Xi ∇Xj ψ − ∇Xj ∇Xi ψ
and thus
[d∇ R∇ (X0 , X1 , X3 )]ψ = (∇X0 R∇ (X1 , X2 ))ψ + (∇X1 R∇ (X2 , X0 ))ψ + (∇X2 R∇ (X0 , X1 ))ψ
= ∇X0 ∇X1 ∇X2 ψ − ∇X0 ∇X2 ∇X1 ψ + ∇X1 ∇X2 ∇X0 ψ
− ∇X1 ∇X0 ∇X2 ψ + ∇X2 ∇X0 ∇X1 ψ − ∇X2 ∇X1 ∇X0 ψ
= R∇ (X0 , X1 )∇X2 ψ + R∇ (X2 , X0 )∇X1 ψ + R∇ (X1 , X2 )∇X0 ψ,
which vanishes at p. 

Proof 2. We have (d∇ R∇ )ψ = d∇ (R∇ ψ) − R∇ ∧ d∇ ψ = d∇ (d∇ d∇ ψ) − d∇ d∇ (d∇ ψ) = 0. 

Exercise 32. Let M = R3 . Determine which of the following forms are closed (dω = 0) and
which are exact (ω = dθ for some θ):
a) ω = yz dx + xz dy + xy dz,
b) ω = x dx + x2 y 2 dy + yz dz,
c) ω = 2xy 2 dx ∧ dy + z dy ∧ dz.
28 ULRICH PINKALL

If ω is exact, please write down the potential form θ explicitly.

Exercise 33. Let M = Rn . For ξ ∈ Γ(TM), we define ω ξ ∈ Ω1 (M) and ?ω ξ ∈ Ωn−1 (M) as
follows:
ω ξ (X1 ) := hξ, X1 i, ?ω ξ (X2 , . . . , Xn ) := det(ξ, X2 , . . . , Xn ), X1 , . . . , Xn ∈ Γ(TM).
Show the following identities:
df = ω gradf , d ? ω ξ = div(ξ) det,
and for n = 3,
dω ξ = ?ω rotξ .

10. Fundamental theorem for flat vector bundles

Let E → M be a vector bundle with connection ∇. Then

E trivial ⇐⇒ ∃ frame field Φ = (ϕ1 , . . . , ϕk ) with ∇ϕi = 0, i = 1, . . . , k
and
E flat ⇐⇒ E locally trivial, i.e. each point p ∈ M has a neighborhood U such that E|U is trivial.
Theorem 33 (Fundamental theorem for flat vector bundles). (E, ∇) is flat ⇐⇒ R∇ = 0.

Proof. ”⇒”: Let (ϕ1 , . . . , ϕk ) be a local parallel frame field. Then we have for i = 1, . . . , k
R∇ (X, Y )ϕi = ∇X ∇Y ϕi − ∇Y ∇X ϕi − ∇[X,Y ] ϕi = 0.
Since R∇ is tensorial checking R∇ ψ = 0 for the elements of a basis is enough.
”⇐”: Assume that R∇ = 0. Locally we find for each p ∈ M a neighborhood U diffeomorphic
to (−ε, ε)n and a frame field Φ = (ϕ1 , . . . , ϕk ) on U . Define ω ∈ Ω1 (U, Rk×k ) by
k
X
∇ϕi = ϕj ωji .
j=1

With ∇Φ = (∇ϕ1 , . . . , ∇ϕk ), we write

∇Φ = Φω.
Similarly, for a map F : U → Gl(k, R) define a new frame field:
Φ̃ = ΦF −1
All frame fields on U come from such F . We want to choose F in such a way that ∇Φ̃ = 0. So,
!
0 = ∇Φ̃ = ∇(ΦF −1 ) = (∇Φ)F −1 + Φd(F −1 ) = (∇Φ)F −1 − ΦF −1 dF F −1 = Φ(ω − F −1 dF )F −1 ,
where we used that d(F −1 ) = −F −1 dF F −1 . Thus we have to solve
dF = F ω.
The Maurer-Cartan Lemma (below) states that such F : U → Gl(k, R) exists if and only if the
integrability condition (or Maurer-Cartan equation)
dω + ω ∧ ω = 0
 ANALYSIS AND GEOMETRY ON MANIFOLDS 29

is satisfied. We need to check that in our case the integrability condition holds: We have
0 = R∇ (X, Y )Φ = ∇X ∇Y Φ − ∇Y ∇X Φ − ∇[X,Y ] Φ
= ∇X (Φω(Y )) − ∇Y (Φω(X)) − Φω([X, Y ])
= Φω(X)ω(Y ) + Φ(Xω(Y )) − Φω(Y )ω(X) − Φ(Y ω(X)) − Φω([X, Y ])
= Φ(dω + ω ∧ ω)(X, Y ).
Thus dω + ω ∧ ω = 0. 
˜ as
Exercise 34. Let M ⊂ R2 be open. On E = M × R2 we define two connections ∇ and ∇
follows:
Ç å Ç å
0 −x dy ˜ 0 −x dx
∇=d+ , ∇=d+ .
x dy 0 x dx 0
˜ and
Show that (E, ∇) is not trivial. Further construct an explicit isomorphism between (E, ∇)
the trivial bundle (E, d).
Lemma 4 (Maurer-Cartan). Let U := (−ε, ε)n , ω ∈ Ω1 (U, Rk×k ), F0 ∈ Gl(k, R). Then
∃F : U → Gl(k, R) : dF = F ω, F (0, . . . , 0) = F0 ⇐⇒ dω + ω ∧ ω = 0.
Remark 12: Note that dω + ω ∧ ω automatically vanishes on 1-dimensional domains.

Proof. ”⇒”: Let F : U → Gl(k, R) solve the initial value problem dF = F ω, F (0, . . . , 0) =
F0 . Then 0 = d2 F = d(F ω) = dF ∧ ω + F dω = F ω ∧ ω + F dω = F (dω + ω ∧ ω). Thus
dω + ω ∧ ω = 0. ”⇐ (Induction on n)”: Let n = 1. We look for F : (−ε, ε) → Gl(k, R) with
dF = F ω, F (0, . . . , 0) = F0 ∈ Gl(k, R). With ω = A dx, this becomes just the linear ODE
F 0 = F A,
which is solvable. Only thing still to check that F (x) ∈ Gl(k, R) for initial value F0 ∈ Gl(k, R).
But for a solution F we get (det F )0 = (det F ) trA. Thus if (det F )(0) = det F0 6= 0 then
det F (x) 6= 0 for all x ∈ (−ε, ε). Now let n > 1 and suppose that the Maurer-Cartan lemma
holds for n − 1. Write ω = A1 dx1 + · · · + An dxn with Ai : (−ε, ε)n → Rk×k . Then
X 
∂
X ∂Aj ∂Ai
(dω+ω∧ω)( ∂x , ∂ )=
i ∂xj
dAα ∧dxα + ∂
Aα Aβ dxα ∧dxβ ( ∂x , ∂ )=
i ∂xj ∂xi − ∂xj +Ai Aj −Aj Ai .
α α,β

∂ F̂
By induction hypothesis there is F̂ : (−ε, ε)n−1 → Gl(k, R) with ∂x i
= F̂ Ai , i = 1, . . . , n − 1,
and F̂ (0) = F0 . Now we solve for each (x1 , . . . , xn−1 ) the initial value problem
F̃x0 1 ,...,xn−1 (xn ) = F̃x1 ,...,xn−1 (xn )An (x1 , . . . , xn ), F̃x1 ,...,xn−1 (0) = F̂ (x1 , . . . , xn−1 ).
∂F
Define F (x1 , . . . , xn ) := F̃x1 ,...,xn−1 (xn ). By construction ∂xn = F An and with dω + ω ∧ ω = 0,
∂ ∂F ∂ ∂F ∂ ∂ ∂
∂xn ( ∂xi − F Ai ) = ∂xi ∂xn − ∂xn (F Ai ) = ∂xi (F An ) − ∂xn (F Ai )
∂F ∂F ∂An ∂Ai
= ∂xi An − ∂xn Ai + F ( ∂xi − ∂xn )
∂F
= ∂xi An − F An Ai + F (An Ai − Ai An )
∂F
= F ( ∂xi
− F Ai )An .
∂F ∂F
Thus t 7→ ( ∂xi
− F Ai )(x1 , . . . , xn−1 , t) solves a linear ODE. Since ∂xi
− F Ai = 0 on the slice
∂F
{x ∈ (−ε, ε) | xn = 0}, we conclude ∂xα − F Aα for all α ∈ {1, . . . , n} on whole of (−ε, ε)n . 
n
30 ULRICH PINKALL

Exercise 35. Let M ⊂ R be an interval and consider the vector bundle E = M × Rk , k ∈ N,

equipped with some connection ∇. Show that (E, ∇) is trivial. Furthermore, show that any
vector bundle with connection over an intervall is trivial.

11. Affine connections

Definition 36. A connection ∇ on the tangent bundle is called an affine connection.

Special about the tangent bundle is that there exists a canonical 1-form ω ∈ Ω1 (M, TM), the
tautological form, given by
ω(X) := X.
Definition 37 (Torsion tensor). If ∇ is an affine connection on M, the TM-valued 2-form
T ∇ := d∇ ω is called the torsion tensor of ∇. ∇ is called torsion-free if T ∇ = 0.
Example 1: Let M ⊂ Rn open. Identify TM with M × Rn by setting (p, X)f = dp f (X).
On M × R use the trivial connection: All X ∈ Γ(M × R) are of the form X = (Id, X̂) for
X̂ ∈ C ∞ (M, Rn ).
(∇X Y )p = (p, dp Ŷ (X)).
Remark (engineer notation): ∇X Y = (X · ∇)Y , with ∇ = ( ∂x∂ 1 , ∂x∂ 2 , ∂x∂ 3 )t and X = (x1 , x2 , x3 )
∂ ∂ ∂
X · ∇ = x1 + x2 + x3 .
∂x1 ∂x2 ∂x3
Define a frame field X1 , . . . , Xn on M of ’constant vector fields’ Xj = (p, ej ). Then with ∇
denoting the trivial connection on TM = M × Rn we have
T ∇ (Xi , Xj ) = ∇Xi Xj − ∇Xj Xi − [Xi , Xj ] = 0.
Theorem 34 (First Bianchi identity). Let ∇ be a torsion-free affine connection on M. Then
for all X, Y, Z ∈ Γ(TM) we have
R∇ (X, Y )Z + R∇ (Y, Z)X + R∇ (Z, X)Y = 0.

Proof. For the tautological 1-form ω ∈ Ω1 (M, TM) and a torsion-free connection we have 0 =
d∇ d∇ ω(X, Y, Z) = R∇ ∧ ω(X, Y, Z) = R∇ (X, Y )Z + R∇ (Y, Z)X + R∇ (Z, X)Y . 
Theorem 35. If ∇ is a metric connection on a Euclidean vector bundle E → M then we have
for all X, Y ∈ Γ(TM) and ψ, ϕ ∈ Γ(E)
hR∇ (X, Y )ψ, ϕi = −hψ, R∇ (X, Y )ϕi,
i.e. as a 2-form R∇ takes values in the skew-adjoint endomorphisms.

Proof. The proof is straightforward. We have

0 = d2 hψ, ϕi
= dhd∇ ψ, ϕi + dhψ, d∇ ϕi = hd∇ d∇ ψ, ϕi − hd∇ ψ ∧ d∇ ϕi + hd∇ ψ ∧ d∇ ϕi + hψ, d∇ d∇ ϕi
= hd∇ d∇ ψ, ϕi + hψ, d∇ d∇ ϕi.
With d∇ d∇ = R∇ this yields the statement. 
Definition 38 (Riemannian manifold). A Riemannian manifold is a manifold M together with
a Riemannian metric, i.e. a metric h., .i on TM.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 31

Theorem 36 (Fundamental theorem of Riemannian geometry). On a Riemannian manifold

there is a unique affine connection ∇ which is both metric and torsion-free. ∇ is called the
Levi-Civita connection.

Proof. Uniqueness: Let ∇ be metric and torsion-free, X, Y, Z ∈ Γ(TM). Then

XhY, Zi + Y hZ, Xi − ZhX, Y i = h∇X Y, Zi + hY, ∇X Zi + h∇Y Z, Xi
+ hZ, ∇Y Xi − h∇Z X, Y i − hX, ∇Z Y i
= h∇X Y + ∇Y X, Zi + hY, ∇X Z − ∇Z Xi + h∇Y Z − ∇Z Y, Xi
= h2∇X Y − [X, Y ], Zi + hY, [X, Z]i + h[Y, Z], Xi.
Hence we obtain the so called Koszul formula:
1Ä ä
h∇X Y, Zi = XhY, Zi + Y hZ, Xi − ZhX, Y i + h[X, Y ], Zi − hY, [X, Z]i − h[Y, Z], Xi .
2
So ∇ is unique. Conversely define ∇X Y by the Koszul formula (for this to make sense we need
to check tensoriality). Then check that this defines a metric torsion-free connection. 
Exercise 36. Let (M, g) be a Riemannian manifold and g̃ = e2u g for some smooth function
u : M → R. Show that between the corresponding Levi-Civita connections the following relation
holds:
˜ X Y = ∇X Y + du(X)Y + du(Y )X − g(X, Y )grad u.
∇
Here grad u ∈ Γ(TM) is the vector field uniquely determined by the condition du(X) = g(grad u, X)
for all X ∈ Γ(TM).
Definition 39 (Riemannian curvature tensor). Let M be a Riemannian manifold. The curva-
ture tensor R∇ of its Levi-Civita connection ∇ is called the Riemannian curvature tensor.
Exercise 37. Let (M, h., .i) be a 2-dimensional Riemannian manifold, R its curvature tensor.
Show that there is a function K ∈ C ∞ (M) such that
Ä ä
R(X, Y )Z = K hY, ZiX − hX, ZiY , for all X, Y, Z ∈ Γ(TM).
Exercise 38. Let h., .i be the Euclidean metric on Rn and B := {x ∈ Rn | |x|2 < 1}. For
k ∈ {−1, 0, 1} define
4
gk |x := h., .i.
(1 + k|x|2 )2
Show that for the curvature tensors Rk of the Riemannian manifolds (B, g−1 ), (Rn , g0 ) and
(Rn , g1 ) and for every X, Y ∈ Rn the following equation holds:
Ä ä
gk (Rk (X, Y )Y, X) = k gk (X, X)gk (Y, Y ) − gk (X, Y )2 .

12. Flat Riemannian manifolds

The Maurer-Cartan-Lemma states that if E → M is a vector bundle with connection ∇ such

that R∇ = 0 then E is flat, i.e. each p ∈ M has a neighborhood U and a frame field ϕ1 , . . . , ϕk ∈
Γ( E|U ) with ∇ϕj = 0, j = 1, . . . , k. In fact if we look at the proof we see that given a basis
ψ1 , . . . , ψk ∈ Ep the frame ϕ1 , . . . , ϕk can be chosen in such a way that ϕj (p) = ψj , j = 1, . . . , k.
Suppose E is Euclidean with compatible ∇ then choose ψ1 , . . . , ψk to be an orthonormal basis.
Then for each X ∈ Γ(TU) we have Xhϕi , ϕj i = 0, i, j = 1, . . . , k, i.e. (assuming that U
is connected) ϕ1 , . . . , ϕk is an orthonormal frame field: hϕi , ϕj i(q) = δij for all q ∈ U . We
summarize this in the following theorem.
32 ULRICH PINKALL

Theorem 37. Every Euclidean vector bundle with flat connection locally admits an orthonormal
parallel frame field.
Definition 40 (Isometry). Let M and N be Riemannian manifolds. Then f : M → N is called
an isometry if for all p ∈ M the map dp f : Tp M → Tf (p) N is an isometry of Euclidean vector
spaces. In other words, f is a diffeomorphism such that for all p ∈ M, X, Y ∈ Tp M we have
hdf (X), df (Y )iN = hX, Y iM .

Intiution: n-dimensional Riemannian manifolds are ”curved versions of Rn ”. Rn = ”flat space”.

The curvature tensor R∇ measures curvature, i.e. deviation from flatness.
The following theorem states that any Riemannian manifold with curvature R = 0 is locally
isometric to Rn .
Theorem 38. Let M be an n-dimensional Riemannian manifold with curvature tensor R = 0
and let p ∈ M. Then there is a neighborhood U ⊂ M of p, an open set V ⊂ Rn and an isometry
f: U →V.

Proof. Choose Ũ ⊂ M open, p ∈ Ũ then there is a parallel orthonormal frame field X1 , . . . , XN ∈

Γ(TŨ ). Now define E := TM ⊕ (M × R) = TM ⊕ R. Any ψ ∈ Γ(E) is of the form
Ä ä
ψ= Y
g

with Y ∈ Γ(TM) and g ∈ C ∞ (M). Define a connection ∇

˜ on E as follows
Ä ä Ä ∇ Y −gX ä
˜X
∇ Y
g := X
.
Xg
˜ is a connection. Now
It is easy to see that ∇
˜ ∇[X,Y ] Z−g[X,Y ]
Ä ∇ Z−gY ä Ä ∇ Z−gX ä  
R∇ (X, Y )
Ä ä
Z
g
˜X
=∇ Y ˜Y
−∇ X
−
Yg Xg [X,Y ]g
∇[X,Y ] Z−g[X,Y ]
Ä ∇ ∇ Z−(Xg)Y −g∇ Y −(Y g)X ä Ä ∇ ∇ Z−(Y g)X−g∇ X−(Xg)Y ä  
= X Y X
XY g
− Y X Y
Y Xg
− [X,Y ]g
Ä ä
R(X,Y )Z
= 0
= 0.
˜ = 0. Then ψ = (Y, g) with
Now choose Û ⊂ Ũ , p ∈ Û and ψ ∈ Γ( E|Û ) with ψp = (0, 1), ∇ψ
Pn
Y = j=1 fj Xj and
P
dfj (X)Xj −gX
Ä ∇ Y −gX ä 
( 00 ) = X
Xg = j
Xg
.
In particular, g = 1. If we define f : Û → Rn by f = (f1 , . . . , fn ) then
X
hdf (X), df (Z)i = hdfj (X), dfj (Z)i = hgX, gY i = hX, Y i.
j

In particular, dp f is bijective. The inverse function theorem then yields a neighborhood U of p

such that f |U : U → V ⊂ Rn is a diffeomorphism and hence an isometry. 
Exercise 39. Let M and M̃ be Riemannian manifolds with Levi-Civita connections ∇ and ∇, ˜
˜ f X f∗ Y .
respectively. Let f : M → M̃ be an isometry and X, Y ∈ Γ(M). Show that f∗ ∇X Y = ∇ ∗

Remark 13: With the last exercise follows that a Riemannian manifold M has curvature R = 0
if and only if it is locally isometric to Rn .
Exercise 40. a) Show that hX, Y i := 21 trace(X̄ t Y ) defines a Riemannian metric on SU(2).
b) Show that the left and the right multiplication by a constant g are isometries.
c) Show that SU(2) and

the 3-sphere S3 ⊂ R4 (with induced metric) are isometric.
a b
Hint: SU(2) = { −b̄ ā | a, b ∈ C, |a|2 + |b|2 = 1}.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 33

13. Geodesics

Let M be a Riemannian manifold, ∇ the Levi-Civita connection on TM, γ : [a, b] → M, Y ∈

Γ(γ ∗ TM). Then, for t ∈ [a, b] we have Yt ∈ (γ ∗ TM)t = {t} × Tγ(t) M ∼
= Tγ(t) M. Y is called a
0 ∗ dY
vector field along γ. Now define (Y )t = (γ ∇) ∂ | Y =: ds (t).
∂s t

Definition 41 (Geodesic). γ : [a, b] → M is called a geodesic if γ 00 = 0.

Exercise 41. Let f : M → M̃ and g : M̃ → M̂ be smooth. Show that f ∗ (g ∗ TM̂) ∼
= (g ◦ f )∗ TM̂
and
ˆ = f ∗ (g ∗ ∇)
(g ◦ f )∗ ∇ ˆ
ˆ on M̂. Show further that, if f is an isometry between Riemannian
for any affine connection ∇
manifolds, γ is curve in M and γ̃ = f ◦ γ, then
γ̃ 00 = df (γ 00 ).
Exercise 42. Let M be a Riemannian manifold, γ : I → M be a curve which is parametrized
with constant speed, and f : M → M be an isometry which fixes γ, i.e. f ◦ γ = γ. Furthermore,
let
ker(id − dγ(t) f ) = Rγ̇(t), for all t.
Then γ is a geodesic.
Definition 42 (Variation). A variation of γ : [a, b] → M is a smooth map α : (−ε, ε)×[a, b] → M
such that γ0 = γ, where γt : [a, b] → M such that γt (s) = α(t, s). The vector field along γ given
d
by Ys := dt α(t, s) is called the variational vector field of α.
t=0
Definition 43 (Length and energy of curves). Let γ : [a, b] → M be a smooth curve. Then
Z b
L(γ) := |γ 0 | is called the length of γ,
a
Z b
E(γ) := 1
2 |γ 0 |2 is called the energy of γ.
a
Theorem 39. Let γ : [a, b] → M be a smooth curve. Let ϕ : [c, d] → [a, b] be smooth with
ϕ0 (t) > 0 for all t ∈ [c, d], ϕ(c) = a and ϕ(d) = b. Then
L(γ ◦ ϕ) = L(γ).

Rd Rd R ϕ(d) Rb
Proof. L(γ ◦ ϕ) = c |(γ ◦ ϕ)0 | = c |(γ 0 ◦ ϕ)|ϕ0 = ϕ(c) |γ 0 | = a |γ 0 | = L(γ). 

Theorem 40. E(γ) ≥ 1

2(b−a) L(γ)
2 (equality if and only if |γ 0 | is constant).

Rb
Proof. The Cauchy-Schwarz inequality yields L(γ)2 ≤ 2E(γ) a 1 = 2(b − a)E(γ). 

Theorem 41. Let γ : [a, b] → M be a smooth curve such that γ 0 (t) 6= 0 for all t ∈ [a, b]. Then
there is a smooth function ϕ : [0, L(γ)] → [a, b] with ϕ0 (t) > 0 for all t, ϕ(0) = a and ϕ(L(γ)) = b
such that γ̃ = γ ◦ ϕ is arclength parametrized, i.e. |γ̃ 0 | = 1.

Proof. If ϕ0 = 1/|γ 0 ◦ϕ|, then |γ̃ 0 | = |(γ 0 ◦ϕ)ϕ0 | = 1. Define ψ : [a, b] → [0, L(γ)] by ψ(t) = at |γ 0 |.
R

Then ψ 0 (t) > 0 for all t, ψ(a) = 0 and ψ(b) = L(γ). Now set ϕ = ψ −1 . Then ϕ0 = 1/|γ 0 ◦ ϕ|. 
34 ULRICH PINKALL

Theorem 42. Let M̃ be a manifold with torsion-free connection ∇. ˜ Let f : M → M̃ and let
˜ = f ∇ be the pullback connection on f TM̃. Then, if X, Y ∈ Γ(TM) we have df (X), df (Y ) ∈
∇ ∗ ∗

Γ(f ∗ TM) and

˜ X df (Y ) − ∇
∇ ˜ Y df (X) = df ([X, Y ]).

˜ ˜
Proof. Let ω denote the tautological 1-form on TM̃. Then d∇ ω = T ∇ = 0 and f ∗ ω = df . Thus
˜
0 = f ∗ d∇ ω = d∇ f ∗ ω = d∇ df.
Thus 0 = d∇ df (X, Y ) = ∇X df (Y ) − ∇Y df (X) − df ([X, Y ]). 
∂ ∂ ∂f ∂f
Example 2: Let M ⊂ Rn be open, X = ∂xi and Y = ∂xj . Then df (X) = ∂xi and df (Y ) = ∂xj .
∂
We have [ ∂x , ∂ ] = 0. Hence
i ∂xj
∂f ∂f
∇ ∂
∂xj =∇ ∂
∂xi .
∂xi ∂xj

Theorem 43 (First variational formula for energy). Suppose α : (−ε, ε) × [a, b] → M is a

variation of γ : [a, b] → M with variational vector field Y ∈ Γ(γ ∗ TM). Then
Z b
d 0 b
E(γt ) = hY, γ i − a hY, γ 00 i.
dt t=0 a

Proof.
Z b Z b Z b
d
dt t=0 E(γt ) = d 1
dt t=0 2 |γt0 |2 = 1
2
d ∂α 2
dt t=0 | ∂s | = h (α∗ ∇) ∂ ∂α ∂α
∂s (0,s) , ∂s i
a a a ∂t
Z b Z b Z b
= h (α∗ ∇) ∂ ∂α ∂α
∂t (0,s) , ∂s i = ∂ ∂α ∂α
∂s (0,s) h ∂t , ∂s i − h ∂α ∗
∂t , (α ∇) ∂s
∂
∂α
∂s (0,s) i
a ∂s a a
Z b Z b Z b
0 b
= d
ds hY, γ i − hY, γ i = hY, γ 0 i a −
00
hY, γ 00 i.
a a a

Corollary 3. If α is a variation of γ with fixed endpoints, i.e. α(t, a) = γ(a) and α(t, b) = γ(b)
d
for all t ∈ (−ε, ε), and γ is a geodesic, then dt E(γt ) = 0.
t=0

Later we will see the converse statement: If γ is a critical point of E, then γ is a geodesic.
Existence of geodesics: Let ∇ be an affine connection on an open submanifold M ⊂ Rn . Let
∂
Xi := ∂x i
. Then there are functions Γkij , called Christoffel symbols of ∇, such that
X
∇Xi Xj = Γkij Xk
k
Let γ = (γ1 , . . . , γn ) be a smooth curve in M. Then γ 0 = 0 ∗ By definition of γ ∗ ∇,
P
i γi (γ Xi ).

(γ ∗ Xj )0 = (γ ∗ ∇) ∂ γ ∗ Xj = ∇γ 0 Xj = γi0 γ ∗ (∇Xi Xj ) = γi0 (Γkij ◦ γ)γ ∗ Xk .

X X
∂s
i i,k

Thus γ is a geodesic of ∇ if and only if

0 = γ 00 = γj00 γ ∗ Xj + γj0 γi0 (Γkij ◦ γ)γ ∗ Xk .
XÄ X ä

j i,k

Since γ ∗ Xi form a frame field we get n equations:

0 = γk00 + γi0 γj0 Γkij ◦ γ.
X

i,j
 ANALYSIS AND GEOMETRY ON MANIFOLDS 35

This is an ordinary differential equation of second order and Picard-Lindelöf assures the exis-
tence of solutions.
Theorem 44 (First variational formula for length). Let γ : [0, L] → M be arclength parametrized,
i.e. |γ 0 | = 1. Let t → γt for t ∈ (−ε, ε) be a variation of γ with variational vector field Y . Then
Z L
d 0 L
L(γt ) = hY, γ i 0 − hY, γ 00 i.
dt t=0 0

Proof. Almost the same as for the first variational formula for energy. 

Theorem 45. Let γ : [a, b] → M be a geodesic. Then |γ 0 | = constant.

Proof. We have hγ 0 , γ 0 i0 = 2hγ 0 , γ 00 i = 0. 

Definition 44 (Killing fields). Suppose t → gt for t ∈ (−ε, ε) is a 1-parameter family of

isometries of M, i.e. each gt : M → M is an isometry. Then the vector field X ∈ Γ(TM),
d
Xp = dt gt (p) is called a Killing field of M.
t=0
Theorem 46. Let X ∈ Γ(TM) be a Killing field and γ : [a, b] → M be a geodesic. Then
hX, γ 0 i = constant.

Proof. Let γt := gt ◦ γ. Then Ys = Xγ(s) and L(γt ) = L(γ) for all t. Thus
Z b
b b
0= d
dt t=0 L(γt ) = hXγ , γ 0 i a − hXγ , γ 00 i = hXγ , γ 0 i a .
a

Thus hXγ(a) , γ 0 (a)i = hXγ(b) , γ 0 (b)i. 

Example 3 (Surface of revolution and Clairaut’s relation): If we have a surface of revolution in

Euclidean 3-space, then the rotations about the axis of revolution are isometries of the surface.
This yields a Killing field X such that X is orthogonal to the axis of revolution and |X| = r,
where r denotes the distance to the axis. From the last theorem we know that if γ is a geodesic
parametrized with unit speed then r cos α = hγ 0 , Xi = c ∈ R. Thus r = c/ cos α and, in
particular, r ≥ c Thus, depending on the constant c, geodesics cannot pass arbitrarily thin
parts.
Example 4 (Rigid body motion): Let M = SO(3) ⊂ R3×3 , q1 , . . . , qn ∈ R3 , m1 , . . . , mn > 0.
Now if t → A(t), t ∈ (−ε, ε), B = A(0), X = A0 (0). Then define
n
X
hX, Xi = 1
2 mi |Xqi |2 ,
i=1
d
where Xqi = dt |t=0 (A(t)qi ).
hX, Xi is called the kinetic energy at time 0 of the rigid body that
undergoes the motion t → A(t). The principle of least action then says: When no forces act on
the body, it will move according to s → A(s) ∈ SO(3) which is a geodesic. For all G ∈ SO(3) the
left multiplication A 7→ GA is an isometry. Suitable families t 7→ Gt with G0 = I then yields
the conservation of angular momentum. We leave the details as exercise.
Theorem 47 (Rope construction of spheres). Given p ∈ M, for t ∈ [0, 1] let γt : [0, 1] → M such
that γt (0) = p for all t. Let X(t) ∈ Tp M such that X(t) = γt0 (0), |X| = v ∈ R, η : [0, 1] → M,
η(t) = γt (1). Then for all t we have
hη 0 (t), γt0 (1)i = 0.
36 ULRICH PINKALL

Proof. Apply the Rfirst variational formula to γ = γt : Then we have Y0 = 0 and Y1 = η 0 . Since
L(γt ) = 0 |γ | = 01 |X(0)| = v, we have
R1 0

0= d
dt t=t0 L(γt ) = hη 0 (t0 ), γt00 (1)i − h0, γt00 (0)i = hη 0 (t0 ), γt00 (1)i.


14. The exponential map

Theorem 48. For each p ∈ M there is a neighborhood U ⊂ M and ε > 0 such that for all
X ∈ Tq M, q ∈ U , with |X| < ε there is a geodesic γ : [0, 1] → M such that γ(0) = q, γ 0 (0) = X.

Proof. Picard-Lindelöf yields a neighborhood W̃ ⊂ TM of 0 ∈ Tp M and ε1 > 0 such that for

X ∈ W̃ , X ∈ Tq M, there is a geodesic γ : [−ε1 , ε1 ] → M such that γ(0) = q and γ 0 (0) = X.
Choose U ⊂ M open, ε2 > 0 such that W := {X ∈ Tq M | q ∈ U, |X| ≤ ε2 } ⊂ W̃ . Now
set ε = ε1 ε2 . Let q ∈ U , X ∈ Tq M with |X| < ε and define Y := ε11 X. Then |Y | < ε2 , i.e.
Y ∈ W ⊂ W̃ . Thus there exists a geodesic γ̃ : [−ε1 , ε1 ] → M with γ̃ 0 (0) = Y . Now define
γ : [0, 1] → M by γ(s) = γ̃(ε1 s). Then γ is a geodesic with γ 0 (0) = ε1 γ̃ 0 (0) = ε1 Y = X. 
Definition 45 (Exponential map). Ω := {X ∈ TM | ∃ γ : [0, 1] → M geodesic with γ 0 (0) = X}.
Define exp : Ω → M by exp(X) = γ(1), where γ : [0, 1] → M is the geodesic with γ 0 (0) = X.
Lemma 5. If γ : [0, 1] → M is a geodesic with γ 0 (0) = X then γ(t) = exp(tX) for all t ∈ [0, 1].

Proof. For t ∈ [0, 1] define γt : [0, 1] → M by γt (s) = γ(ts). Then γt0 (0) = tX , γt (1) = γ(t), γt
is a geodesic. So exp(tX) = γt (1) = γ(t). 
Exercise 43. Show that two isometries F1 , F2 : M → M which agree at a point p and induce
the same linear mapping from Tp M agree on a neighborhood of p.
Theorem 49. Let p ∈ M. Then there is ε > 0 and an open neighborhood U ⊂ M of p such that
Bε := {X ∈ Tp M | |X| < ε} ⊂ Ω and exp|Bε : Bε → U is a diffeomorphism.

Proof. From the last lemma we get d0p exp(X) = X. Here we used the canonical identification
between Tp M and T0p (TM) given by X 7→ (t 7→ tX). The claim then follows immediately from
the inverse function theorem. 
Definition 46 (Geodesic normal coordinates). ( exp|Bε )−1 : U → Bε ⊂ Tp M ∼
= Rn viewed as a
coordinate chart is called geodesic normal coordinates near p.
Exercise 44. Let M be a Riemannian manifold of dimension n. Show that for each point p ∈ M
there is a local coordinate ϕ = (x1 , . . . , xn ) at p such that
Ä
∂ ∂ ∂
g ∂xi , ∂xj ) p = δij , ∇ ∂ ∂xj = 0.
∂xi p
Theorem 50 (Gauss lemma). exp|Bε maps radii t 7→ tX in Bε to geodesics in M. Moreover,
these geodesics intersect the hypersurfaces Sr := {exp(X) | X ∈ Bε , |X| = r} orthogonally.

Proof. This follows by the last lemma and the rope construction of spheres. 
Definition 47 (Distance). Let M be a connected Riemannian manifold. Then for p, q ∈ M
define the distance d(p, q) by
d(p, q) = inf{L(γ) | γ : [0, 1] → M smooth with γ(0) = p, γ(1) = q}.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 37

Exercise 45. a) Is there a Riemannian manifold (M, g) which has finite diameter (i.e. there
is an m such that all points p, q ∈ M have distance d(p, q) < m) and there is a geodesic of
infinite length without self-intersections?
b) Find an example for a Riemannian manifold diffeomorphic to Rn but which has no geodesic
of infinite length.
Definition 48 (Metric space). A metric space is a pair (X, d) where X is a set and d : X×X → R
a map such that

a) d(p, q) ≥ 0, d(p, q) = 0 ⇔ p = q,
b) d(p, q) = d(q, p),
c) d(p, q) + d(q, r) ≥ d(p, r).

Is a Riemannian manifold (with its distance) a metric space. Symmetry is easy to see: If
γ : [0, 1] → M is a curve from p to q, then γ̃(t) := γ(1−t) is a curve from q to p and L(γ̃) = L(γ).
For the triangle inequality we need to concatenate curves. So let γ : [0, 1 → M] be a curve from p
to q and γ̃ : [0, 1] → M be a curve from q to r. Though the naive concatenation is not smooth we
can stop for a moment and then continue running: Let ϕ : [0, 1] → [0, 1] be smooth monotone
function such that ϕ(0) = 0, ϕ(1) = 1 and ϕ0 vanishes on [0, ε) ∪ (1 − ε, 1] for some ε > 0
sufficiently small. Then define for γ from p to q and γ̃ from q to r
®
γ(ϕ(2t)), for t ∈ [0, 1/2),
γ̂(t) =
γ̃(ϕ(2t − 1)), for t ∈ [1/2, 1].
Then L(γ̂) = L(γ) + L(γ̃). For every ε > 0 we find γ and γ̃ such that
L(γ) ≤ d(p, q) + ε, L(γ̃) ≤ d(q, r) + ε.
Thus by concatenation we obtain a curve γ̂ from q to r such that L(γ̂) ≤ d(p, q) + d(q, r) + 2ε.
Thus d(p, r) ≤ d(p, q) + d(q, r). Certainly, L(γ) ≥ 0 d(p, q) ≥ 0 and d(p, p) = 0. So the only
part still missing is that p = q whenever d(p, q) = 0.
Theorem 51. Let p ∈ M and f : Bε → U ⊂ M be geodesic normal coordinates at p. Then
d(p, exp(X)) = |X|, for |X| ≤ ε.
Moreover, for q 6∈ U , d(p, q) > ε.

Proof. Choose 0 < R < ε. Take γ : [0, 1] → M with γ(0) = p, γ(t) := exp(tX) with |X| = R. Let
q := γ(1) = exp(X). Then L(γ) = R. In particular, d(p, q) ≤ R. Now, choose 0 < r < R and
let γ : [0, 1] → M be any curve with γ(0) = p and γ(1) = q. Define a to be the smallest t ∈ [0, 1]
such that there is Y such that γ(t) = exp(Y ), |Y | = r. Define b to be the smallest t ∈ [0, 1],
a < b, such that there is Z such that γ(b) = exp(Z), |Z| = R. Now find ξ : [a, b] → TM such
that r < |ξ(t)| < R for all t ∈ (a, b), |ξ(a)| = r, |ξ(b)| = R and exp(ξ(t)) = γ(t) for all t ∈ [a, b].
Define ρ : [a, b] → M by ρ := |ξ| and ν : [a, b] → M by ξ =: ρν. Claim: L( γ|[a,b] ) ≥ R − r.
Afterwards: L(γ) ≥ R − r for all such r > 0. Hence L(γ) ≥ R and thus d(p, q) = R. Let us
prove the claim: For all t ∈ [a, b] we have
γ 0 (t) = d exp(ξ 0 (t)) = d exp(ρ0 (t)ν(t) + ρ(t)ν 0 (t)) = ρ0 (t)d exp(ν(t)) + ρ(t)d exp(ν 0 (t)).
By the Gauss lemma we get then
|γ 0 (0)|2 = |ρ0 (t)d exp(ν(t))|2 + |ρ(t)d exp(ν 0 (t))|2 ≥ |ρ0 (t)|2 |d exp(ν(t))|2 = ρ0 (t)2 .
| {z }
=1
38 ULRICH PINKALL

Thus we have Z b Z b
L( γ|[a,b] ) = 0
|ρ | ≥ ρ0 = ρ|ba = R − r.
a a
Certainly, we can have equality only for ν 0 = 0. This yields the second part. 
Corollary 4. A Riemannian manifold together with its distance function is a metric space.
Corollary 5. Let γ : [0, L] → M be an arclength-parametrized geodesic. Then there is ε > 0
such that d(γ(0), γ(t)) = t for all t ∈ [0, ε].

The first variational formula says: If γ : [a, b] → M is a smooth length-minimizing curve, i.e.
L(γ) = d(γ(a), γ(b)), then γ is a geodesic. To see this, choose a function ρ : [a, b] → R with
ρ(s) > 0 for all s ∈ (a, b) but ρ(a) = 0 = ρ(b). Then there is ε > 0 such that α : (−ε, ε)×(a, b) →
M, α(t, s) = exp(tρ(s)γ 00 (s)). Without loss of generality we can assume that |γ 0 | = 1. Then
Z b Z b
d 0 00 b 00 00
0= L(γ) = hγ , ργ i a− hργ , γ i = − ρ|γ 00 |2
dt t=0 | {z } a a
=0
for all such ρ. Thus we conclude γ 00
= 0 and so γ is a geodesic. We need a slightly stronger
result. For preparation we give the following exercise:
Exercise 46. d(p, q) = inf{L(γ) | γ : [a, b] → M piecesewise smooth, γ(a) = p, γ(b) = q}.
Theorem 52. Let γ : [0, L] → M be a continuous piecewise-smooth curve with |γ 0 | = 1 (when-
ever defined) such that L(γ) = d(γ(0), γ(L)). Then γ is a smooth geodesic.

Proof. Let 0 = s0 < · · · < sk = L be such that γ|[si−1 ,si ] is smooth, i = 1, . . . , k. The above
discussion then shows that the parts γ|[si−1 ,si ] are smooth geodesics. We need to show that
there are no kinks. Let j ∈ {1, . . . , k − 1} and X := γ 0 |[sj−1 ,sj ] (sj ), X̃ = γ 0 |[sj ,sj+1 ] (sj ).
Claim: X = X̃. Define Y = X̃ − X and choose any variation γt of γ which does nothing on
[0, sj−1 ] ∪ [sj+1 , L]. Then
d X d
0= L(γt ) = L( γt |[sj−1 ,sj ] ) = hY, Xi − hY, X̃i = |X̃ − X|2 .
dt t=0 j
dt t=0

Thus X̃ − X = 0. 

15. Complete Riemannian manifolds

Definition 49 (Complete Riemannian manifold). A Riemannian manifold is called complete

if exp is defined on all of TM, or equivalently: every geodesic can be extended to R.
Theorem 53 (Hopf and Rinow). Let M be a complete Riemannian manifold, p, q ∈ M Then
there is a geodesic γ : [0, L] with γ(0) = p, γ(L) = q and L(γ) = d(p, q).

Proof. Let ε > 0 be such that exp|Bε is a diffeomorphism onto its image. Without loss of
generality, assume that δ < d(p, q). Let 0 < δ < ε and set S := exp(Sδ ), where Sδ = ∂Bδ . Then
f : S → R given by f (r) = d(r, q) is continuous. Since S is compact, there is r0 ∈ S where f
has a minimum, i.e.
d(r0 , q) ≤ d(r, p) for all r ∈ S.
Then r0 = γ(δ), where γ : R → M with γ(0) = p. Define
d(S, q) := inf{d(r, q) | r ∈ S}.
 ANALYSIS AND GEOMETRY ON MANIFOLDS 39

Then d(S, q) = d(r0 , q). Every curve η : [a, b] → M from p to q has to hit S: There is t0 ∈ [a, b]
with η(t0 ) ∈ S. Moreover,
L(η) = L( η|[a,t0 ] ) + L( η|[t0 ,b] ) ≥ δ + d(S, q) = δ + d(r0 , q).
So d(p, q) ≥ δ + d(r, q). On the other hand, the triangle inequality yields d(p, q) ≤ d(p, r0 ) +
d(r0 , q) = δ + d(r0 , q). Thus d(γ(δ), q) = d(p, q) − δ.
Define statement A(t): ”d(γ(t), q) = d(p, q) − t.” So we know A(δ) is true. We want to show
that also A(d(p, q)) is true. Define
t0 := sup{t ∈ [0, d(p, q)] | A(t) true}.
Assume that t0 < d(p, q). Claim: A(t0 ) is true. This is because there is a sequence t1 , t2 , . . . ,
with limn→∞ tn = t0 and A(tn ) true, i.e. f (tn ) = 0 where f (t) = d(γ(t), q) − (d(p, q) − t).
Clearly, f is continuous. Thus f (t0 ) = 0, too.
Now let γ̃ be a geodesic constructed as before but emanating from γ(t0 ). With the same
argument as before we then get again
d(γ̃(δ̃), q) = d(γ̃(0), q) − δ̃.
Now, since A(t0 ) is true, we have
d(p, q) ≤ d(p, γ̃(δ̃)) + d(γ̃(δ̃), q) = d(p, γ̃(δ̃)) + d(γ̃(0), q) − δ̃ = d(p, γ̃(δ̃)) + d(p, q) − t0 − δ̃.
There obviously is a piecewise-smooth curve from p to γ̃(δ̃) of length t0 + δ̃. So d(p, γ̃(δ̃)) ≤ t0 + δ̃.
Hence d(p, γ̃(δ̃)) = t0 + δ̃. Hence this piecewise-smooth curve is length minimizing and in
particular it is smooth, i.e. there is no kink and thus we have γ̃(δ̃) = γ(t0 + δ̃).
Now we have d(γ(t0 + δ̃), q) = d(γ(t0 ), q) − δ̃ = d(p, q) − (t0 + δ̃). Thus A(t0 + δ̃) is true, which
contradicts the definition of t0 . So A(d(p, q)) is true. 
Theorem 54. For a Riemannian manifold M the following statements are equivalent:
a) M is complete Riemannian manifold.
b) All bounded closed subsets of M are compact.
c) (M, d) is a complete metric space.

Proof. a) ⇒ b): Let A ⊂ M be closed and bounded, i.e. there is p ∈ M and c ∈ R such that
d(p, q) ≤ c for a all p, q ∈ A. Look at the ball Bc ⊂ Tp M. Hopf-Rinow implies then that
A ⊂ exp(Bc ). Hence A is a closed subset of a compact set and thus compact itself. b) ⇒ c) is
a well-known fact: Any Cauchy sequence {pn }n∈N is bounded and thus lies in bounded closed
set which then is compact. Hence {pn }n∈N has a convergent subsequence which then converges
to the limit of {pn }n∈N . c) ⇒ a): Let γ : [0, `] → M be a geodesic.
T := sup{t ≥ ` | γ can be extended to [0, T ]}.
We want to show that T = ∞. Define pn := γ(T − n1 ). Then {pn }n∈N defines a Cauchy sequence
which thus has a limit point p := limn→∞ pn . Thus γ extends to [0, T ] by setting γ(T ) := p.
Thus γ extends beyond T . which contradicts the definition of T . 
Exercise 47. A curve γ in a Riemannian manifold M is called divergent, if for every compact
set K ⊂ M there exists a t0 ∈ [0, a) such that γ (t) 6∈ K for all t > t0 . Show: M is complete if
and only if all divergent curves are of infinite length.
Exercise 48. Let M be a complete Riemannian manifold, which is not compact. Show that
there exists a geodesic γ : [0, ∞) → M which for every s > 0 is the shortest path between γ (0)
and γ (s).
40 ULRICH PINKALL

Exercise 49. Let M be a compact Riemannian manifold. Show that M has finite diameter, and
that any two points p, q ∈ M can be joined by a geodesic of length d(p, q).

16. Sectional curvature

Definition 50 (Sectional curvature). Let M be a Riemannian manifold, p ∈ M, E ⊂ Tp M,

dim E = 2, E = span{X, Y }. Then
hR(X, Y )Y, Xi
KE :=
hX, XihY, Y i − hX, Y i2
is called the sectional the sectional curvature of E.
Exercise 50. Check that KE is well-defined.
Theorem 55. Let M be a Riemannian manifold, p ∈ M, X, Y, Z, W ∈ Tp M. Then
hR(X, Y )Z, W i = hR(Z, W )X, Y i.

Proof. The Jacobi identity yields the following 4 equations:

0 = hR(X, Y )Z, W i + hR(Y, Z)X, W i + hR(Z, X)Y, W i,
0 = hR(Y, Z)W, Xi + hR(Z, W )Y, Xi + hR(W, Y )Z, Xi,
0 = hR(W, Z)X, Y i + hR(X, W )Z, Y i + hR(Z, X)W, Y i,
0 = hR(X, W )Y, Zi + hR(Y, X)W, Zi + hR(W, Y )X, Zi.


The following theorem tells us that the sectional curvature completely determine the curvature
tensor R.
Theorem 56. Let V be a Euclidean vector space. R : V × V → V bilinear with all the symme-
tries of the curvature tensor of a Riemannian manifold. For any 2-dimensional subspace E ⊂ V
with orthonormal basis X, Y define KE = hR(X, Y )Y, Xi. Let R̃ be another such tensor with
K̃E = KE for all 2-dimensional subspaces E ⊂ V . Then R̃ = R.

Proof. KE = K̃E implies hR(X, Y )Y, Xi = hR̃(X, Y )Y, Xi for all X, Y ∈ V . We will show that
we can calculate hR(X, Y )Z, W i for all X, Y, Z, W ∈ V provided we know hR(X, Y )Y, Xi for
all X, Y ∈ V . Let X, Y, Z, W ∈ V . Define f : R2 → R by
f (s, t) = hR(X + sW, Y + tZ)(Y + tZ), X + sW i − hR(X + sZ, Y + tW )(Y + tW ), X + sZi.
For fixed X, Y, Z, W this is polynomial in s and t. We are only interested in the st term: It is
hR(W, Z)Y, Xi + hR(W, Y )Z, Xi + hR(X, Z)Y, W i + hR(X, Y )Z, W i
− hR(Z, W )Y, Xi − hR(Z, Y )W, Xi − hR(X, W )Y, Zi − hR(X, Y )Z, W i
= 4hR(X, Y )Z, W i + 2hR(W, Y )Z, Xi − 2hR(Z, Y )W, Xi
= 4hR(X, Y )Z, W i + 2hR(W, Y )Z + R(Y, Z)W, Xi
= 4hR(X, Y )Z, W i − 2hR(Z, W )Y, Xi
= 6hR(X, Y )Z, W i.

 ANALYSIS AND GEOMETRY ON MANIFOLDS 41

Corollary 6. Let M be a Riemannian manifold and p ∈ M. Suppose that KE = K for all

E ⊂ Tp M with dim E = 2. Then
R(X, Y )Z = K(hZ, Y iX − hZ, XiY ).

Proof. Define R̃ by this formula. Then R̃(X, Y ) is skew in X, Y and

hR̃(X, Y )Z, W i = K(hY, ZihX, W i − hZ, XihY, W i)
is skew in Z, W . Finally,
R̃(X, Y )Z + R̃(Y, Z)X + R̃(Z, X)Y
= K(hZ, Y iX − hZ, XiY + hX, ZiY − hX, Y iZ + hY, XiZ − hY, ZiX) = 0.
and if X, Y ∈ Tp M is an orthonormal basis then
K̃E = KhhY, Y iX − hY, XiY, Xi = K.


17. Jacobi fields

Let γ : [0, L] → M be a geodesic and α : (−ε, ε) × [0, L] → M be a geodesic variation of γ, i.e.

γt = α(t, .) is a geodesic for all t ∈ (−ε, ε). Then the corresponding variational vector field
Y ∈ Γ(γ ∗ TM) along γ,
∂α
Ys = ,
∂t (0,s)
is called a Jacobi field.
Lemma 6. Let α be a variation of a curve, ∇ ˜ = α∗ ∇ and R̃ = α∗ R. Then

˜ ∂
˜ ∂∇ ∂ ∂ ∂ ∂ ˜ ∂∇ ˜ ∂ ∂ α.
∇ ∂t α = R̃( , ) α+∇
∂s ∂s ∂s ∂t ∂s ∂t ∂s
∂s

˜ ∂
Proof. Since ∇ is torsion-free we have ∇ ∂ ˜ ∂
= ∇ ∂
∂s
∂t α ∂t
∂s α. The equation then follows from
∂ ∂
[ ∂s , ∂t ] = 0. 
Theorem 57. A vector field Y ∈ Γ(γ ∗ TM) is a Jacobi field if and only if it satisfies
Y 00 + R(Y, γ 0 )γ 0 = 0.

Proof. ”⇒”: With the lemma above evaluated for (0, s) we obtain
∂ ∂ ∂
Y 00 = R̃( ∂s
∂ ∂ ∂
, ∂t ) ∂s α ˜ ∂∇
+∇ ˜ ∂ ∂
∂s α (0,s) = R̃( , ) α = R(γ 0 , Y )γ 0 .
(0,s) ∂t ∂s ∂s ∂t ∂s (0,s)

”⇐”: Suppose a vector field Y along γ satisfies Y + 00 R(Y, γ 0 )γ 0

= 0. We want to construct a
geodesic variation α such that γ0 = γ and with variational vector field Y . The solution of a
linear second order ordinary differential equation Y is uniquely prescribed by Y (0) and Y 0 (0).
In particular the Jacobi fields form a 2n-dimensional vector space. Denote p := γ(0). By the
first part of the prove it is enough to show that for each V, W ∈ Tp M there exists a geodesic
variation α of γ with variational vector field Y which satisfies Y (0) = V and Y 0 (0) = W : The
curve η : (−ε̃, ε̃) → M, η(t) = exp(tV ) is defined for ε̃ > 0 small enough. Define a parallel vector
field W̃ along η with W̃0 = W . Similarly, let Ũ be parallel along η such that Ũ0 = γ 0 (0). Now
define α : [0, L]×(−ε, ε) → M by α(s, t) = exp(s(Ũt +tW̃t )), for ε > 0 small enough. Clearly, α is
42 ULRICH PINKALL

a geodesic variation of γ. From Ũt , W̃t ∈ Tη(t) M we get γt (0) = η(t) and hence Y (0) = η 0 (0) = V .
Moreover, Y 0 (0) = α̇0 (0, 0) = ∇ ∂ α0 = ∇ ∂ (Ũt + tW̃t ) = W̃0 = W . 
∂t |(s,t)=(0,0) ∂t |t=0

Exercise 51. Show that, as claimed in the previous proof, there is ε > 0 such that for |t| < ε
the geodesic γt = α(., t) really lives for time L.

Trivial geodesic variations: γt (s) = γ(a(t)s + b(t)) with functions a and b such that a(0) = 1,
b(0) = 0. Then the variational vector field is just Ys = (a0 (0)s + b0 (0))γ 0 (s). Thus Y 0 = a0 (0)γ 0
and hence Y 00 = 0. Certainly also R(Y, γ 0 ) = 0. Thus Y is a Jacobi field.
Interesting Jacobi fields are orthogonal to γ 0 : Let Y be a Jacobi-field. Then f : [0, L] → R,
f = hY, γ 0 i. Then f 0 = hY 0 , γ 0 i and f 00 = hY 00 , γ 0 i = −hR(Y, γ 0 )γ 0 , γ 0 i = 0. Thus there are
a, b ∈ R such that f (s) = as + b. In particular, with V := Y (0) and W := Y 0 (0) we have
f (0) = hV, γ 0 i, f 0 (0) = hW, γ 0 (0)i. Then we will have f ≡ 0 provided that V, W ⊥ γ 0 (0). So
hY, γ 0 i ≡ 0 in this case. This defines a (2n − 2)-dimensional space of (interesting) Jacobi fields.
Example 5: Consider M = Rn . Then Y Jacobi field along s 7→ p + sv if and only if Y 00 ≡ 0,
i.e. Y (s) = V + sW for parallel vector fields V, W along γ (constant).
Example 6: Consider the round sphere Sn ⊂ Rn+1 and let p, V, W ∈ Rn+1 be orthonormal.
Define γt as follows
γt (s) := cos s p + sin s (cos t V + sin t W ).
Then Ys = sin s W is a Jacobi field and thus

− sin s W = Y 00 (s) = −R(Y (s), γ 0 (0))γ 0 (0) = − sin sR(W, γ 0 )γ 0 .

Thus W = R(W, γ 0 )γ 0 . Evaluation for s = 0 then yields W = R(W, V )V . In particular, if

E = span{V, W } ⊂ Tp Sn , then KE = hR(W, V )V, W i = 1.

18. Second variational formula

Theorem 58 (Second variational formula). Let α : (−ε, ε) × (−ε, ε) × [0, L] → M be a 2-

parameter variation of a geodesic γ : [0, L] → M, i.e. α(0, 0, s) = γ(s), with fixed endpoints, i.e.
α(u, v, 0) = γ(0) and α(u, v, L) = γ(L) for all u, v ∈ (−ε, ε). Let
∂α ∂α
Xs = , Ys = , γu,v (s) := α(u, v, s).
∂u (0,0,s) ∂v (0,0,s)

Then
∂2 L Z
E(γu,v )(0, 0) = − hX, Y 00 + R(Y, γ 0 )γ 0 i.
∂u∂v 0
Remark 14: Actually, that is an astonishing formula. Since the left hand side is symmetric in
u and v, the right hand side must be symmetric in X and Y . Let’s check this first directly: Let
X, Y ∈ Γ(γ ∗ TM) such that X0 = 0 = Y0 and XL = 0 = YL . Then with partial integration we
get
Z L Z L Z L Z L Z L
hX, Y 00 +R(Y, γ 0 )γ 0 i = hX, Y 00 i+ hX, R(Y, γ 0 )γ 0 i = − hX 0 , Y 0 i+ hX, R(Y, γ 0 )γ 0 i,
0 0 0 0 0

which is symmetric in X and Y .

 ANALYSIS AND GEOMETRY ON MANIFOLDS 43

Proof. First,
Z L Z L Z L
∂ ∂
E(γu.v ) = 21 ∂
h ∂s ∂
α, ∂s αi = ∂
h∇ ∂ ∂s ∂
α, ∂s αi = h∇ ∂ ∂ ∂
∂u α, ∂s αi
∂u ∂u 0 0 ∂u 0 ∂s
Z L Z L
∂ ∂ L ∂ ∂ ∂ ∂
= h ∂u α, ∂s αi − h ∂u α, ∇ ∂ ∂s αi =− h ∂u α, ∇ ∂ ∂s αi.
0 0 ∂s 0 ∂s

Now, let us take the second derivative:

∂2
Z L Z L Z L
∂ ∂ ∂ ∂ ∂ ∂ ∂
E(γu.v ) = − h ∂u α, ∇ ∂ ∂s =−
αi h∇ ∂ ∂u α, ∇ ∂ ∂s αi − h ∂u α, ∇ ∂ ∇ ∂ ∂s αi
∂u∂v ∂v 0 ∂s 0 ∂v ∂s 0 ∂v ∂s
Z L Z L
∂ ∂ ∂ ∂ ∂ ∂ ∂
= − h∇ ∂ ∂u α, ∇ ∂ ∂s αi − h ∂u α, ∇ ∂ ∇ ∂ ∂s α + R( ∂v , ∂s ) ∂s αi.
0 ∂v ∂s 0 ∂s ∂v

Evaluation at (u, v) = (0, 0) yields

∂2
Z L Z L
E(γu,v )(0, 0) = − ∂
h ∇ ∂ ∂u α , γ 00 i − hX, Y 00 + R(Y, γ 0 )γ 0 i.
∂u∂v 0 ∂v (u,v)=(0,0) 0

With γ 00 = 0 we obtain the desired result. 

19. Bonnet-Myers’s theorem

Missing: Integration on manifolds. Let M be compact Riemannian manifold, f ∈ C ∞ (M).

Then on can define Z
f ∈ R.
M
If M is any orientable manifold of dimension n and ω ∈ Ωn0 (M) then one can define
Z
ω.
M

There is an interesting relation between Topology and geometry (curvature).

Definition 51 (Simply connected). A manifold M is called simply connected if for every smooth
map γ : S1 → M, S1 = ∂D2 , there is a smooth map f : D2 → M such that γ = f |S1 .
Theorem 59. Let M be a simply connected complete Riemannian manifold with√constant sec-
tional curvature K > 0. Then M is isometric to a round sphere of radius r = 1/ K.

Without completeness: only a part of the sphere, without ’simply connected’: RPn has
also constant sectional curvature. Similar with lense spaces: Identify points on S3 ⊂ C2 that
differ by e2πi/n , M = S3 / ∼.
1
Theorem 60. M simply connected complete, for all sectional curvature KE we have 4 < KE ≤
1. Then M is homeomorphic to Sn .
1
Remark 15: For M = CPn one has 4 ≤ KE ≤ 1.
Theorem 61 (Gauss-Bonnet). Let M be compact of dimension 2. Then there is an integer
χ(M ) ≤ 2 such that
Z
K = 2πχ(M ).
M

If M, M̃ are orientable, then: χ(M) = χ(M̃) ⇔ M, M̃ diffeomorphic.

44 ULRICH PINKALL

Definition 52 (Scalar curvature). M Riemannian manifold, p ∈ M, G2 (Tp M) Grassmanian of

2-planes E ⊂ Tp M ( dim G2 (Tp M) = n(n − 1)/2). Then
Z
1
S̃ := vol(G2 (Tp M)) KE
G2 (Tp M)
is called the scalar curvature.
Definition 53 (Ricci curvature). M Riemannian, p ∈ M, X ∈ Tp M, |X| = 1, Sn−2 ⊂ X ⊥ ⊂
Tp M. Then Z
1
Ric(X,
g X) = vol(Sn−2 )
Kspan{X,Y } dY
Sn−2
is called Ricci curvature.

Let us try something simpler: Choose an orthonormal basis Z1 , . . . , Zn of Tp M with Z1 = X

and define
n
X n
X
1 1
Ric(X, X) := n−1 hR(Zi , X)X, Zi i = n−1 Kspan{X,Zi } .
i=1 i=2
Then with AZ := R(Z, X)X defines an endomorphism of Tp M and
n
X n
X
1 1 1
Ric(X, X) := n−1 hR(Zi , X)X, Zi i = n−1 hAZi , Zi i = n−1 tr(A).
i=1 i=1
Thus Ric(X, X) does not depend on the choice of the basis.
Definition 54.
1
Ric(X, Y ) = n−1 tr(Z 7→ R(Z, X)Y ).
Theorem 62. Ricp : Tp M × Tp M → R is symmetric.

Proof.
n
X n
X
1 1
Ric(X, Y ) = n−1 hR(Zi , X)Y, Zi i = n−1 hR(Zi , Y )X, Zi i = Ric(Y, X).
i=1 i=1


Now we have two symmetric bilinear forms on each tangent space, h., .i and Ric.
Theorem 63 (without proof ). Ric(X,
g X) = Ric(X, X).
Definition 55. Define ricp : Tp M → Tp M by hricp X, Y i := Ric(X, Y ).

Eigenvalues κ1 , . . . , κn of ricp (and eigenvectors) provide useful information.

Definition 56. Z1 , . . . , Zn orthonormal basis of Tp M. Then define
X
2
S(p) := n(n−1) hR(Zi , Zj )Zj , Zi i.
i<j

n
X n
X X n X
X
1 1 1 1
n Ric(Zj , Zj ) = n n−1 hR(Zi , Zj )Zj , Zi i = n(n−1) hR(Zi , Zj )Zj , Zi i
j=1 j=1 i6=j j=1 i6=j
X n
X
2 1
= n(n−1) hR(Zi , Zj )Zj , Zi i = n hricp Zj , Zj i = n1 tr(ricp ).
i<j j=1

Theorem 64 (without proof ). S̃(p) = S(p).

 ANALYSIS AND GEOMETRY ON MANIFOLDS 45

Definition 57 (Diameter). Let M be a Riemannian manifold. Then

diam(M) := sup{d(p, q) | p, q ∈ M} ∈ R ∪ {∞}
is called the diameter of M.
Theorem 65. M complete diam(M) < ∞ ⇔ M compact.

Proof. ”⇒”: diam(M) < ∞, then M closed and bounded, thus compact. ”⇐”: d : M × M → R
is continuous, thus takes its maximum. diam(M) < ∞. 
1
Theorem 66 (Bonnet-Myers). M complete Riemannian manifold, Ric(X, X) ≥ r2
hX, Xi all
X ∈ TM. Then diam(M) ≤ πr.

Proof. Choose p, q ∈ M. L := d(p, q) > 0. By Hopf-Rinow there is an arclength-parametrized

geodesic γ : [0, L] → M with γ(0) = p and γ(L) = q. Now choose a parallel orthonormal
frame field X1 , . . . , Xn along γ with X1 = γ 0 . Define Yi ∈ Γ(γ ∗ TM) by Yi (s) = sin( πs
L )Xi (s).
i
Define variations α̃i : (−ε, ε) × [0, L] → M of γ by αi (t, s) = exp(tYi (s)), γt = α̃i (t, .). With
αi (u, v, s) := α̃i (u + v, s) we have
∂αi ∂αi
= Yi (s) = .
∂u (0,0,s) ∂v (0,0,s)
Then we use the second variational formula of length: If g : (−ε, ε) → R, g(t) = L(γt ), then g
has a global minimum at t = 0, i.e. 0 ≤ g 00 (0). Thus
Z L
∂2
0 ≤ g 00 (0) = i
∂u∂v L(γu,v ) =− hYk , Yk00 + R(Yk , γ 0 )γ 0 i.
0
Since Yk (s) = sin( πs 00 π 2 πs
L )Xk , we have Yk (s) = −( L ) sin( L )Xk (s). Thus, for each k,
Z L Z L Z L Z L
( Lπ )2 sin2 ( Lπ ) =− hYk00 , Yk i ≥ 0 0
hR(Yk , γ )γ , Yk i = sin2 ( πs 0 0
L )hR(Xk , γ )γ , Xk i.
0 0 0 0
1
By assumption Ric(X, X) ≥ r2
hX, Xi. Thus summing over k = 2, . . . , n we get
Z L Z L
0 0
n−1
r2
sin2 ( πs
L) ≤ (n − 1) sin2 ( πs
L )Ric(γ , γ )
0 0
n Z L
0 0
X
= sin2 ( πs
L )hR(Xk , γ )γ , Xk i
k=2 0
Z L
≤ (n − 1)( Lπ )2 sin2 ( Lπ ).
0
RL
Then, since 0 sin2 ( πs
L ) > 0, we get L ≤ πr.