Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

240 Bnotes

Download as pdf or txt
Download as pdf or txt
You are on page 1of 112

Math 240B: Differentiable Manifolds and

Riemannian Geometry

Simon Rubinstein–Salzedo

Winter 2006
0.1 Introduction
These notes are based on a graduate course on differentiable manifolds and Rieman-
nian geometry I took from Professor Doug Moore in the Winter of 2006. The textbook
was Riemannian Geometry by Manfredo Perdigão do Carmo. Many other books are
also mentioned in the notes. Since the professor handed out very good notes, I have
made very few changes to these notes.

1
Chapter 1

January 10, 2006

Let M n be a smooth manifold.

A Riemannian metric on M is a function h , i which assigns to each p ∈ M a


positive-definite symmetric bilinear form

h·, ·ip : Tp M × Tp M → R

which varies smoothly with p ∈ M .

The last condition means that if (U, (x1 , . . . , xn )) is a smooth coordinate system on
M , then each of the functions gij : U → R defined by
* +  
∂ ∂ ∂ ∂
gij (p) = , or gij = ,
∂xi p ∂xj p ∂xi ∂xj

is smooth. We sometimes write


n
X
h , i |U = gij dxi ⊗ dxj .
i,j=1

A Riemannian manifold is a pair (M, h , i), where M is a smooth manifold and


h , i is a Riemannian metric on M .

2
1.1 Geometry of a Riemannian Manifold
One can use the Riemannian manifold to define lengths of vectors and lengths of
curves. Thus if v ∈ Tp M , the length of v is given by the formula
q
|v| = hv, vip .

If γ : [a, b] → M is a smooth curve, the length of γ is


Z bq
L(γ) = hγ 0 (t), γ 0 (t)iγ(t) dt.
a

If v, w ∈ Tp M , we can define the angle θ between v and w by

hv, wip
cos θ = .
|v| |w|

We can also use Riemannian metrics to calculate areas and volumes.

Simplest Example. M = Rn with standard coordinates (x1 , . . . , xn ) and Rieman-


nian metric h , i defined by
* n n + n
X ∂ X j ∂ X
ai i
, b j
= ai b j
i=1
∂x p j=1 ∂x p i=1

= (a1 , . . . , an ) · (b1 , . . . , bn ),

where · stands for the dot product. In this case,


(
1 if i = j,
gij = δij =
0 if i 6= j.

If γ : [a, b] → Rn is a smooth curve,


Z bp
L(γ) = γ 0 (t) · γ 0 (t) dt.
a

3
An Important Class of Examples. Suppose M n is a smooth manifold and F :
M n → RN is an imbedding. We then define the induced Riemannian metric on M
by
hv, wip = F∗p (v) · F∗p (w), for v, w ∈ Tp M.
We can identify TF (p) RN with RN . Then
!
∂ ∂F
F∗p i
= (p),
∂x p ∂xi

where on the right, F is regarded as a vector-valued function. Then


 
∂ ∂ ∂F ∂F
i
, j = · ,
∂x ∂x ∂xi ∂xj
which is clearly smooth.

Note that if h , i is the Riemannian metric on M induced by F : M n → RN and


γ : [a, b] → M is a smooth curve, then
Z bq
L(γ) = hγ 0 (t), γ 0 (t)iγ(t) dt
a
Z bq
= F∗γ(t) (γ 0 (t)) · F∗γ(t) (γ 0 (t)) dt
a
Z bp
= (F ◦ γ)0 (t) · (F ◦ γ)0 (t) dt
a
= L(F ◦ γ).

Thus the length of a curve γ in M is the length of its image in RN .

In the case of Euclidean space Rn , with coordinates (x1 , . . . , xn ) and the Riemannian
metric
h , i = dx1 ⊗ dx1 + · · · + dxn ⊗ dxn ,
the curves of smallest length joining p and q are straight lines.

What are the curves of shortest length in a Riemannian manifold?

4
This question can be formulated with the calculus of variations.

Let Ω(M ; p, q) = {smooth paths γ : [0, 1] → M | γ(0) = p and γ(1) = q}. Define
L : Ω(M ; p, q) → R by L(γ) =length of γ. The function L : Ω(M ; p, q) → R is difficult
to deal with because

1. There is a troublesome square root in the integrand.

2. If γ
e is a reparametrization of γ, L(e
γ ) = L(γ).

Fortunately, there is a function J : Ω(M ; p, q) → R, closely related to L, which does


not share these problems. This function is defined by

1 1 0
Z
J(γ) = hγ (t), γ 0 (t)i dt
2 0

and is called the action (or sometimes the energy).

Proposition. (L(γ))2 ≤ 2J(γ), with equality holding if and only if γ is of constant


speed; that is, hγ 0 (t), γ 0 (t)iγ(t) is constant.

Proof. It follows from the Cauchy-Schwarz inequality that


Z 1 2
2
p
(L(γ)) = hγ 0 (t), γ 0 (t)i dt
0
Z 1 Z 1
≤ 1 dt hγ 0 (t), γ 0 (t)i dt
0 0
= 2J(γ),

with equality holding if and only if the functions 1 and hγ 0 (t), γ 0 (t)i are linearly inde-
pendent. 

Proposition. If γ ∈ Ω minimizes L and has constant speed, it minimizes J.

Proof. If λ ∈ Ω,
2J(γ) = [L(γ)]2 ≤ [L(λ)]2 ≤ 2L(γ). 

5
Proposition. If γ ∈ Ω minimizes J, it minimizes L and has constant speed.

Sketch of Proof. Suppose we can reparametrize γ so that it has constant speed.


Let γ
e be such a reparametrization. Then

2J(e γ )]2 = [L(γ)]2 ≤ 2J(γ).


γ ) = [L(e

If γ minimizes J, we must have equality, and by the first proposition, γ has constant
speed. Suppose that λ ∈ Ω and L(λ) < L(γ). If λ has a unit speed reparametrization
λ,
e then

2J(λ) e 2
e = (L(λ))
= (L(λ))2
< (L(γ))2
= 2J(γ),

contradicting the fact that γ minimizes J. 

There is only one problem with this sketch: it assumes we can reparametrize γ and
λ to have constant speed. This can always be done, so long as γ 0 (t) and λ0 (t) never
vanish.

If dim M ≥ 2, we can complete the sketch by approximating γ and λ by immersions


(possible by an approximation theorem of Whitney, proven in Hirsch, Differential
Topology, Theorem 2.12).

But there is also an easier argument we will be able to give later. The upshot is that
γ ∈ Ω minimizes J iff γ ∈ Ω minimizes L and has constant speed. This motivates
looking at the calculus of variations problem for

J : Ω(M ; p, q) → R.

We think of Ω(M ; p, q) as an “infinite-dimensional manifold.” Let γ ∈ Ω. A variation


of γ is a map ᾱ : (−ε, ε) → Ω such that

6
1. ᾱ(0) = γ, and
2. the map α : (−ε, ε) × [0, 1] → M defined by α(s, t) = ᾱ(s)(t) is smooth.

We can think of a variation of γ as a “smooth path” ᾱ : (−ε, ε) → Ω with ᾱ(0) = γ.

Definition. We say that γ ∈ Ω is a critical point for J if



d
(J(ᾱ(s))) =0
ds s=0

for every variation ᾱ of γ.

Note that if γ ∈ Ω minimizes J, it must be a critical point for J.

Definition. An element γ ∈ Ω is a (constant speed) geodesic if it is a critical


point for J.

Theorem. Suppose that F : M n → RN is an imbedding and M n is given the induced


Riemannian metric. Then γ ∈ Ω is a geodesic iff

[(F ◦ γ)00 (t)]T = 0,

where [·]T is the orthogonal projection from TF (γ(t)) RN to Tγ(t) M ⊆ TF (γ(t)) RN .

To simplify notation, we often identify p ∈ M with F (p) ∈ RN . In this case, the


condition that γ be a geodesic is just

[γ 00 (t)]T = 0,

where [·]T is the orthogonal projection to the tangent space.

Proof. We consider a variation ᾱ : (−ε, ε) → Ω(M ; p, q) of γ. Then

α : (−ε, ε) × [0, 1] → M ⊆ RN

7
is a smooth map. Dropping F , we can regard α as a smooth map

α : (−ε, ε) × [0, 1] → RN

whose image lies in the submanifold M . Note that

1 1 ∂ ∂α
Z  
d ∂α
(J(ᾱ(s))) = (s, t) · (s, t) dt
ds 2 0 ∂s ∂t ∂t
Z 1 2
∂ α ∂α
= (s, t) · (s, t) dt
0 ∂s∂t ∂t
Z 1
∂α ∂2α
=− (s, t) · 2 (s, t) dt.
0 ∂s ∂t
The last step follows by integration by parts, since
∂α ∂α
(s, 0) ≡ 0 = (s, 1).
∂s ∂s
Thus
Z 1
∂2α

d ∂α
(J(ᾱ(s)))
=− (0, t) 2 (0, t) dt
ds s=0 0 ∂s ∂t
Z 1
=− V (t) · γ 00 (t) dt,
0

where V (t) = ∂α∂s


(0, t) ∈ Tγ(t) (M ), since α takes values in M ⊆ RN . We can construct
a variation ᾱ(s) such that
∂α
V (t) = (0, t)
∂s
is any smoothly varying family of tangent vectors to M such that V (0) = 0 = V (1).
Thus
(γ 00 (t))T = 0. 

8
Chapter 2

January 12, 2006

Let M n ⊆ RN with the induced Riemannian metric.

Motivated by the theorem we proved last time, we say that a smooth map γ : (a, b) →
M n ⊆ RN is a constant speed geodesic if it satisfies the condition

(γ 00 (t))T = 0 for all t ∈ (a, b).,

where (·)T is the orthogonal projection to the tangent space Tγ(t) (M ) ⊆ Tγ(t) RN .
Thus γ : (a, b) → M n ⊆ RN is a constant speed geodesic if the tangential component
of its acceleration is zero.

Thus, for example, if (e1 , e2 , e3 ) is an orthonormal basis for R3 , S 2 is the unit sphere
in R3 , then the great circle γ : R → S 2 ⊆ R3 ,

γ(t) = cos t e1 + sin t e2

is a constant speed geodesic, since

γ 00 (t) = − cos t e1 − sin t e2 = −γ(t),

which is perpendicular to Tγ(t) S 2 . Indeed, it has unit speed (γ 0 (t) · γ 0 (t) = 1) and is
a smooth closed geodesic, since γ(t + 2π) = γ(t) for all t ∈ R.

By a similar argument, we can show that if

M 2 = {(x, y, z) ∈ R3 : x2 + y 2 = 1},

9
the right circular cylinder, then the curve

γ : R → M, γ(t) = (cos(at), sin(at), bt)

is a constant speed geodesic for all choices of constants a and b. Thus the geodesics
on the right circular cylinder are helices and degenerate cases of these (circles and
lines).

Recall that any Riemannian manifold (M n , h , i) arises from some imbedding F :


M n → RN :

Nash Imbedding Theorem. (1956) If (M n , h , i) is any Riemannian manifold,


there is an imbedding F : M n → RN for some N such that h , i is the metric on M n
induced by F .

This is a very famous theorem by the mathematician John Nash who received a Nobel
Prize in Economics. (See the movie A Beautiful Mind.) However, given (M n , h , i),
it is usually quite difficult to construct F , and in any case, F is by no means unique!

Thus we would like an intrinsic definition of geodesics, one that does not require
an “isometric” imbedding F : M n → RN .

Problem. If (M, h , i) is a Riemannian manifold and γ : (a, b) → M , can we define


the acceleration of γ? (Geodesics would then, one would hope, be curves of “zero
acceleration.”)

In Euclidean space Rn with Euclidean coordinates (x1 , . . . , xn ) and Euclidean metric

h , i = dx1 ⊗ dx1 + · · · + dxn ⊗ dxn ,

we can set
n n  2 i
d2 (xi ◦ γ)

X ∂ X dx ∂
acceleration of γ at t = (t) i = (t). (∗)
i=1
dt2 ∂x γ(t) i=1
dt2 ∂xi

10
However, under a change to curvilinear coordinates,
d2 xi X ∂xi d2 y j X ∂ 2 xj dy j dy k
= + ,
dt2 ∂y j dt2 ∂y j ∂y k dt dt
so (∗) does not hold in curvilinear coordinates. We need a more subtle definition of
acceleration, based upon the notion of connection.

Definition. Let X(M n ) be the set of smooth vector fields on M and F (M n ) the set
of smooth real-valued functions on M . A connection (in the tangent bundle T M to
M ) is a map
∇ : X(M n ) × X(M n ) → X(M n )
(we write ∇X Y for ∇(X, Y )) such that
1. ∇f X+gY Z = f ∇X Z + g∇Y Z,
2. ∇Z (f X + gY ) = (Zf )X + f ∇Z X + (Zg)Y + g∇Z Y
for X, Y, Z ∈ X(M n ), f, g ∈ F (M n ).

Lemma. A connection ∇ is local: if U is an open subset of M ,

X |U ≡ 0 ⇒ (∇X Y ) |U ≡ 0 and (∇Y X ≡ 0)

for all Y ∈ X(M n ).

Proof. Let p ∈ U and choose f : M → R such that f ≡ 0 on a neighborhood of p


and f ≡ 1 outside U . Then X |U = 0 implies that f X = X, and hence

(∇X Y )(p) = (∇f X Y )(p)


= f (p)(∇X Y )(p)
= 0,
(∇Y X)(p) = (∇Y (f X))(p)
= (Y f )(p)X(p) + f (p)∇Y X(p)
= 0.

Since p is an arbitrary point of U , (∇X Y ) |U ≡ 0 and (∇Y X)(p) = 0. 

11
The lemma can be restated as

supp(∇X Y ) ⊆ supp(X), supp(∇Y X) ⊆ supp(X).

It follows from the lemma that a connection ∇ restricts to a well-defined connection


on any coordinate neighborhood U for local coordinate systems (x1 , . . . , xn ).

If (U, (x1 , . . . , xn )) is such a local coordinate chart, we can define smooth functions
Γkij : U → R by
n
∂ X ∂
∇ ∂i j = Γkij k .
∂x ∂x ∂x
k=1

If X = ni=1 f i ∂x∂ i and Y = nj=1 g j ∂x∂ j , then


P P

n  
X
j ∂
∇X Y = ∇f i ∂ g
i,j=1
∂xi ∂xj
n  j

X
i j ∂ i ∂g ∂
= f g ∇ ∂i j + f (∗)
i,j=1
∂x ∂x ∂xi ∂xj
n
" n #
X X ∂g i X ∂
= fj j + Γijk f j g k .
i=1 j=1
∂x ∂xi

Lemma. (∇X Y )(p) depends only on X(p) and on Y on some curve tangent to X(p).

Proof. Immediate from (∗). 

Thus if v ∈ Tp M and X ∈ X(M n ), we can define ∇v X ∈ Tp M . In particular, if


γ : (a, b) → M is any smooth curve, we can define the acceleration of γ (with
respect to ∇) to be ∇γ 0 (t) γ 0 (t), also written as (∇γ 0 γ 0 )(t).

12
In terms of local coordinates,
n
d2 (xi ◦ γ)

0
X ∂
(∇γ 0 γ )(t) = (t) i
i=1
dt2 ∂x γ(t)
n i
◦ γ) d(xi ◦ γ)

X d(x ∂
+ Γkij (γ(t)) (t) (t) k .
i,j,k=1
dt dt ∂x γ(t)

We have defined acceleration in terms of a connection, but we really want to define


acceleration in terms of a Riemannian metric. To do this we use the Levi-Civita
connection.

Fundamental Theorem of Riemannian Geometry. If (M, h , i) is a Riemannian


manifold, there is a unique connection ∇ on T M (called the Levi-Civita connection)
such that

A. ∇ is symmetric: ∇X Y − ∇Y X = [X, Y ].

B. ∇ is metric: XhY, Zi = h∇X Y, Zi + hY, ∇X Zi.

We will prove this next time and show that for this connection, if γ : (a, b) → M n ,

(∇γ 0 γ 0 )(t) = 0 ⇐⇒ (γ 00 (t))T = 0

whenever M n ⊆ RN with the induced Riemannian metric.

13
Chapter 3

January 17, 2006

Recall that a connection on the tangent bundle T M to a smooth manifold M is a


map
∇ : X(M n ) × X(M n ) → X(M n ),
where X(M n ) is the set of smooth vector fields on M n , which satisfies the following
axioms (with ∇X Y = ∇(X, Y )):
1. ∇X+Y Z = ∇X Z + ∇Y Z,
2. ∇Z (X + Y ) = ∇Z X + ∇Z Y ,
3. ∇f X Y = f ∇X Y ,
4. ∇X (f Y ) = (Xf )Y + f ∇X Y ,
for X, Y, Z ∈ X(M n ) and f a smooth real-valued function on M n .

Last time we saw that a connection is local; in other words,


supp(∇X Y ) ⊆ supp(X) ∩ supp(Y ).
Thus we can restrict a connection ∇ to the domain U of a coordinate system (x1 , . . . , xn ).
With respect to these coordinates, we can define the components of the connection
as the functions
n
∂ X ∂
Γkij :U →R defined by ∇ ∂i j = Γkij k .
∂x ∂x ∂x
k=1

14
Fundamental Theorem. If (M, h , i) is a Riemannian manifold, there is a unique
connection ∇ on T M which satisfies the conditions

A. ∇X Y − ∇Y X = [X, Y ],

B. XhY, Zi = h∇X Y, Zi + hY, ∇X Zi

for X, Y, Z ∈ X(M n ).

We can write these conditions in local coordinates as

A0 . Γkij = Γkji .
∂gij Pk Pn
B0 . ∂xk
= `=1 gi` Γ`jk + `=1 gj` Γ`ik .

For A0 , note that


  n
∂ ∂ ∂ ∂ X ∂
i
, j = 0 ⇒ ∇ ∂ i j − ∇ ∂j i = 0 ⇒ (Γkij − Γkji ) k = 0.
∂x ∂x ∂x ∂x ∂x ∂x
k=1
∂x

We leave the verification of B0 to the reader.

It follows from A0 and B0 that


n n
∂gjk X `
X
= g Γ
`j ik + g`k Γ`ij
∂xi `=1 `=1
n n
∂gik X `
X
= g`i Γjk + g`k Γ`ij
∂xj `=1 `=1
n n
∂gij X
`
X
− k =− g`i Γjk − g`j Γ`ik ,
∂x `=1 `=1

and so
n
∂gjk ∂gik ∂gij X
+ − = 2 g`k Γ`ij .
∂xi ∂xj ∂xk `=1

15
If (g mk ) = (gmk )−1 , we conclude that
n  
1 X mk ∂gjk ∂gik ∂gij
Γm
ij = g + − k . (∗)
2 `=1 ∂xi ∂xj ∂x

This establishes uniqueness of the connection satisfying A and B.

For existence, use (∗) to define Γkij and hence ∇ on a coordinate chart (U, (x1 , . . . , xn )).
Check that ∇ satisfies A and B.

Local uniqueness implies that the locally defined ∇’s fit together to give a globally
defined connection ∇ on T M which satisfies A and B. 

The connection ∇ whose existence is guaranteed by the preceding theorem is called


the Levi-Civita connection.

Pn Pn
If X = i=1 f i ∂x∂ i and Y = i=1 g i ∂x∂ i , then
n n
X ∂g j ∂ X ∂
∇X Y = fi
i i
+ f i g j Γkij k
i,j=1
∂x ∂x i,j,k=1
∂x
n
" n
#
X
k
X
i j k ∂
= X(g ) + f g Γij .
k=1 i,j=1
∂xk

This formula shows that (∇X Y )(p) depends only on X(p) and on Y along some
curve tangent to X(p). Thus if γ : (a, b) → M is a smooth curve, we can define
∇γ 0 γ 0 (t) ∈ Tγ(t) M .

Definition. A smooth curve γ : (a, b) → M is called a (constant speed) geodesic


if ∇γ 0 γ 0 = 0.

16
We need to check that this agrees with the definition previously given when M n ⊆ RN
has the induced metric. To do this, we first note that the Levi-Civita connection for
Euclidean space Rn with Euclidean coordinates (x1 , . . . , xn ) and Riemannian metric
n
(
X
i j 1 if i = j,
gij dx ⊗ dx , gij = δij =
i,j=1
0 if i 6= j.
Pn
is given by Γkij = 0. (This follows from (∗).) Thus if X and Y = i=1 g i ∂x∂ i are vector
fields on Rn , the Levi-Civita connection for Euclidean space is
n
X ∂
DX Y = X(g i ) .
i=1
∂xi

Condition B can be stated as

X(Y · Z) = (DX Y ) · Z + Y · (DX Z).

Suppose now that M n ⊆ RN is an imbedded submanifold with the induced metric: if


X, Y ∈ X(M n ),

hX, Y i(p) = X(p)


e · Ye (p), for p ∈ M n,

e and Ye are extensions of X and Y to vector fields on RN .


whenever X

We claim that the Levi-Civita connection ∇ on M n is given by

(∇X Y )(p) = ((DXe Ye (p)))T for X, Y ∈ X(M n ), p ∈ M n , (∗∗)

where (·)T is the orthogonal projection into the tangent space. One verifies that
(∗∗) defines a connection on T M which satisfies A0 and B0 . Thus (∗∗) defines the
Levi-Civita connection on T M . In particular,

(∇γ 0 γ 0 )(t) = (Dγ 0 γ 0 (t))T = (γ 00 (t))T ,

and a geodesic γ : (a, b) → M n ⊆ RN is simply a curve which has tangential acceler-


ation zero.

17
Chapter 4

January 19, 2006

Suppose that (M, h , i) is a Riemannian manifold with


n
X
h, i= gij dxi ⊗ dxj .
i,j=1

Last time we saw that there is a unique connection ∇, the so-called Levi-Civita
connection, on T M such that

A. ∇X Y − ∇Y X = [X, Y ],

B. XhY, Zi = h∇X Y, Zi + hY, ∇X Zi, for X, Y, Z ∈ X(M ).

In terms of local coordinates,


n
∂ X ∂
∇ ∂i j = Γkij k ,
∂x ∂x ∂x
k=1

where n  
1 X k` ∂gi` ∂gj` ∂gij
Γkij = g + − .
2 `=1 ∂xj ∂xi ∂x`

In the special case where M n ⊆ RN and h , i is the metric induced by the Euclidean
dot product on RN ,
∇X Y = (DX Y )T ,

18
where (·)T is the orthogonal projection to T M and D is the usual directional derivative
on RN . In terms of Euclidean coordinates (u1 , . . . , uN ) on RN ,
N
! N
I ∂ ∂
X X
DX f I
= (Xf I ) I .
I=1
∂u I=1
∂u

The Levi-Civita connection is used to make the following key definition:

Definition. A smooth curve γ : (a, b) → M is a (constant speed) geodesic if


∇γ 0 γ 0 (t) ≡ 0.

In terms of local coordinates (x1 , . . . , xn ) on M ,


n i

X d(x ◦ γ) ∂
γ 0 (t) =

(t) i ,
i=1
dt ∂x γ(t)
n !
X X d(xi ◦ γ) ∂
(∇γ 0 γ 0 )(t) = γ 0 (t)

(t) i
i=1
dt ∂x γ(t)
n j
◦ γ) d(xk ◦ γ)

X d(x ∂
+ Γijk (γ(t)) (t) (t) i
i,j,k=1
dt dt ∂x γ(t)
n
" n
#
d(xi ◦ γ) i k

X X
i d(x ◦ γ) d(x ◦ γ) ∂
= (t) + Γjk (γ(t)) (t) (t) .
i=1
dt2 j,k=1
dt dt ∂xi γ(t)

We write xi (t) for (x◦ γ)(t). Then γ : (a, b) → M is a geodesic if and only if
n
d2 xi X
i dxj dxk
(t) + Γjk (γ(t)) (t) (t) = 0. (∗)
dt2 j,k=1
dt dt

dxi
If we let ẋi (t) = dt
(t), we can rewrite (∗) as
( i
dx
dt
= ẋi ,
i (∗∗)
dẋ
= − nj,k=1 Γijk ẋj ẋk .
P
dt

19
Suppose p ∈ M , v ∈ Tp M . It follows from the fundamental existence and uniqueness
theorem for ordinary differential equations that for some ε > 0, there is a unique
solution xi (t), −ε < t < ε to (∗) such that
dxi
xi (0) = p, (0) = ẋi (0) = dxi |p (v).
dt
Thus there is a unique geodesic γv : (−ε, ε) → M such that

γv (0) = p, γv0 (0) = v,

for any p ∈ M , v ∈ Tp M .

Definition. Let (M, h , i) be a Riemannian manifold. A diffeomorphism F : M → M


is an isometry of F ∗ h , i = h , i, or in other words, if for every p ∈ M ,

v ∈ Tp M ⇒ |Fp∗ (v)| = |v|.

Isometries preserves lengths of curves and take geodesics to geodesics.

It is usually quite difficult to find explicit solutions to (∗). The explicit solutions
usually occur only when M has lots of isometries.

Examples.

M n = Rn with the Euclidean coordinates (x1 , . . . , xn ) and the Riemannian metric


n
(
X 1 if i = j,
h, i= dxi ⊗ dxi , gij = δij =
i=1
0 if i 6= j.
2 i
In this case, Γkij = 0, and (∗) becomes ddtx2 = 0, which has xi (t) = ai t + bi as its
solutions, where ai and bi are constants. Thus the geodesics in Euclidean space are

20
just the constant speed straight lines.

Define F : R2 → R2 by
 1     1  1
x ◦F cos θ − sin θ x c
= + 2 ,
x2 ◦ F sin θ cos θ x2 c
where θ, c1 , and c2 are constants. Then F is an isometry. One can construct an
isometry which takes any straight line into any other straight line.

Suppose next that M 2 = S 2 = {(x, y, z) ∈ R3 : x2 + y 2 + z 2 = 1}, and give S 2 the


Riemannian metric it inherits from R3 . We have seen that if (e1 , e2 , e3 ) is a positively
oriented orthonormal basis, then the constant speed great circle,
γ : R → S 2, γ(t) = cos(at)e1 + sin(at)e2
is a geodesic in S 2 .

Let SO(3) = {A ∈ GL(3, R) : AT = A, det A = 1}. If A ∈ SO(3), define


   
x x
2 2
FA : S → S by FA y = A y  .
  
z z
Then FA is an isometry. As we saw in 240A, SO(3) is a three-dimensional manifold.
Thus we have a three-dimensional family of isometries of S 2 which take any great
circle to any other great circle.

Suppose next that H2 = {(x, y) ∈ R2 : y > 0}, the so-called Poincaré upper
half-plane, with the Riemannian metric
1
h , i = 2 (dx2 + dy 2 ).
y
We can calculate the geodesics by brute force. Let x1 = x, x2 = y. Then a straight-
forward calculation gives
Γ111 = Γ122 = Γ121 = Γ112 = 0
1
Γ222 = Γ121 = Γ112 = −
y
2
Γ11 = y.

21
Hence the equations for geodesics are
( 2
d x dy
dt2
− y2 dx
dt dt
=0
d2 y 1 dy 2 dx 2
− y dt + y1
 
dt2 dt
= 0.

It dx
dt
= 0 at one point, it follows from the first equation that dx
dt
≡ 0 and from the
second that  2
d2 y 1 dy
− = 0.
dt2 y dt
dy
Reduce the order, setting z = dt
, so
(
dy
dt
=z dz z
dz
⇒ = .
dt
= y1 z 2 dy y

dz dy dy
Hence z
= y
, log |z| = log |y| + C, z = (±ec )y = by, dt
= by, y = aebt . Thus

x = 0, y = aebt

is a geodesic for all a, b.

dx
If dt
6= 0 and y is a function of x,

d2 y
 
d dy dx
=
dt2 dt dx dt
d dy dx dy d2 x
 
= +
dt dx dt dx dt2
2
d2 y dx dy d2 x
 
= 2 + ,
dx dt dx dt2

and hence
2 2 2
d2 y d2 x
    
1 dy 1 dx dx dy
− = 2 +
y dt y dt dx dt dx dt2
2
d2 y

dx 2 dy dx dy
= 2 + .
dx dt y dx dt dt

22
dx 2

Dividing by dt
allows us to eliminate t:
2  2
d2 y 2 dy

1 dy 1
− = 2+ ,
y dx y dx y dx
 2
d2 y dy
y 2+ = −1,
dx dx
 
d dy
y = −1,
dx dx
dy
y = x + 2a,
dx
y 2 = −x2 + ax + b.

Thus we finally obtain the equation

y 2 + x2 + ax + b = 0.

These are the equations of circles with centers on the x-axis.

Conclusion. Geodesics in the Poincaré upper half-plane are suitably parametrized


vertical lines and semicircle centered on the x-axis.

Of course, we still need to find the constant speed parametrizations of the semicircles.
These could be determined by exploiting the three-dimensional group of isometries
which operates on H2 . Let
  
a b
SL(2, R) = : ad − bc = 1 ,
c d
 
a b
the so-called special linear group. If A = ∈ SL(2, R), define FA : H2 → H2
c d
by
az + b
x ◦ F1 + iy ◦ Fa = , z = x + iy.
cz + d
It is an exercise to show that each FA is an isometry. Moreover, given any semicircle,
there is an A ∈ SL(2, R) such that FA takes a vertical line to that semicircle.

23
The three Riemannian manifolds

R2 with the Euclidean metric,


S 2 with the metric induced by the inclusion S 2 ⊆ R3 ,
H2 with the Poincaré half-plane metric

have three-dimensional groups of isometries. Along with RP 2 , they are the only Rie-
mannian manifolds with three-dimensional groups of isometries.

Theorem. (Hilbert, 1901) There is no imbedding F : H2 → R3 which makes the


Poincaré half-plane metric the induced metric on H2 .

24
Chapter 5

January 24, 2006

Let F0 : R2 → R3 be the parametrized surface defined by

F0 (t, s) = (t, s, 0).

The Riemannian metric that F0 induces on R2 is

h , i = dt ⊗ dt + ds ⊗ ds,

which is, of course, just the Euclidean metric.

But there are many other imbeddings from R2 into R3 which induce exactly the same
metric. Consider, for example, the catenary z = cosh y. We can parametrize the
catenary by arc length; one can check that
√ √
y = s1 + 1, z = log(s + s2 + 1)

is such a parametrization. We can then construct a cylinder over the catenary F1 :


R2 → R 3 , √ √
F1 (t, s) = (t, s2 + 1, log(s + s2 + 1)).
A direct, if slightly tedious, calculation shows that the induced metric on R2 is once
again
h , i = ds ⊗ ds = dt ⊗ dt.

We say that the parametrized surfaces F0 and F1 have the same intrinsic geometry
because the Riemannian metrics induced on R2 are the same, but different extrinsic

25
geometry because the shapes of the images F0 (R2 ) and F1 (R2 ) are different.

How can we recognize when the intrinsic geometry of a surface is not that of Euclidean
space?

More generally, we say that a Riemannian manifold (M, h , i) is locally isometric


to Euclidean space (or locally Euclidean) if given any p ∈ M , there is an open
neighborhood U of p such that (U, h , i |U ) is isometric to an open subset of Euclidean
space.

How do we recognize when a given Riemannian manifold (M, h , i) is locally Eu-


clidean?

The invariant that answers this question is the Riemann-Christoffel curvature tensor.

Recall that associated to a Riemannian metric


n
X
h, i= gij dxi ⊗ dxj
i,j=1

is a Levi-Civita connection ∇,
n n  
∂ X 1 X k` ∂gi` ∂gj` ∂gij
∇ ∂i j = Γkij , Γkij = g + − .
∂x ∂x
k=1
2 `=1 ∂xj ∂xi ∂x`

Definition. The Riemann-Christoffel curvature tensor is the map R : X(M ) ×


X(M ) × X(M ) → X(M ) defined by

R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z for X, Y, Z ∈ X(M ).

26
Just like ∇, R is a local operator and it is clearly linear over the field R of real num-
bers. However, it is also linear over the ring F (M ) = {smooth real-valued functions
f : M → R}.

Proposition.

(a) R(f X, Y )Z = f R(X, Y )Z,

(b) R(X, f Y )Z = f R(X, Y )Z,

(c) R(X, Y )f Z = f R(X, Y )Z.

We prove only (c); the other proofs are quite similar.

R(X, Y )f Z = ∇X ∇Y (f Z) − ∇Y ∇X (f Z) − ∇[X,Y ] (f Z)
= ∇X ((Y f )Z + f ∇Y Z) − ∇Y ((Xf )Z + f ∇X Z) − ([X, Y ](f )Z + f ∇[X,Y ] Z)
= XY (f ) + (Y f )∇X Z + (Xf )∇Y Z + f ∇X ∇Y Z
− Y X(f ) − (Xf )∇Y Z − (Y f )∇X Z − f ∇Y ∇X Z
− [X, Y ](f )Z − f ∇[X,Y ] Z
= f (∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z)
= f R(X, Y )Z,

as desired. 

`
The Proposition justifies defining the components Rkij of R by
  n
∂ ∂ ∂ X
` ∂
R , = R kij .
∂xj ∂xj ∂xk `=1
∂x `

Proposition.
n
` ∂ ` ∂ `
X
Rkij = (Γ ) − j (Γki ) + (Γm ` m `
jk Γim − Γik Γjm ).
∂xi kj ∂x m=1

27
Proof.
n  
X
` ∂ ∂ ∂ ∂
Rkij =R , j
`=1
∂x` ∂x ∂x ∂xk
i
   
∂ ∂
= ∇ ∂ i ∇ ∂j k − ∇ ∂j ∇ ∂ i k
∂x ∂x ∂x ∂x ∂x ∂x

m
! m
!
X ∂ X ∂
= ∇ ∂i Γ`jk ` − ∇ ∂ j Γ`ik `
∂x
`=1
∂x ∂x
`=1
∂x
n  
X ∂ ` ∂ `
X
` m ` m ∂
= i
(Γjk ) − j (Γik ) + (Γim Γjk − Γjm Γik ) ,
`=1
∂x ∂x ∂x`

as desired. 

If M is Euclidean space,
(
1 if i = j,
gij = δij = ⇒ Γ`ij = 0 ⇒ Rkij
`
= 0.
0 if i 6= j

Thus any locally Euclidean Riemannian manifold must have R = 0.

Proposition. The curvature tensor R satisfies the following symmetries:


1. R(X, Y )Z = −R(Y, X)Z.
2. R(X, Y )Z = R(Y, Z)X + R(Z, X)Y = 0.
3. hR(X, Y )Z, W i = −hR(X, Y )W, Zi.
4. hR(X, Y )Z, W i = hR(Z, W )X, Y i.

∂ ∂
Note. It suffices to prove these symmetries for coordinate fields X = ∂xi
, Y = ∂xj
,
Z = ∂x∂ k , and W = ∂x∂ ` .

28
Proof of 1.
 
∂ ∂ ∂ ∂ ∂
R i
, j k
= ∇ ∂ i ∇ ∂j k − ∇ ∂j ∇ ∂ i k
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x
 
∂ ∂ ∂
= −R , i .
∂x ∂x ∂xk
j

Proof of 2.
     
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
R i
, j k
+R j
, k i
+R , i
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂xj
k

∂ ∂
= ∇ ∂ i ∇ ∂j k − ∇ ∂j ∇ ∂ i k
∂x ∂x ∂x ∂x ∂x ∂x

∂ ∂
+ ∇ ∂j ∇ ∂ i
− ∇ ∂ ∇ ∂j i
∂x ∂x ∂x
k ∂x k ∂x ∂x

∂ ∂
+ ∇ ∂ ∇ ∂i j − ∇ ∂i ∇ ∂ = 0,
∂x k ∂x ∂x ∂x ∂x ∂xj
k

∂ ∂
where we have used the fact that ∇ ∂
∂xj
=∇ ∂
∂xi
.
∂xi ∂xj

Proof of 3. It suffices to show that


   
∂ ∂ ∂ ∂
R , , = 0.
∂xi ∂xj ∂xk ∂xk

But
∂2 ∂2
  
∂ ∂
0= − j i ,
∂xi ∂xj ∂x ∂x ∂xk ∂xk
   
∂ ∂ ∂ ∂ ∂ ∂
= 2 i ∇ ∂j k , k − 2 j ∇ ∂ i k , k
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x
   
∂ ∂ ∂ ∂
= 2 ∇ ∂ i ∇ ∂j k , k + 2 ∇ ∂j k , ∇ ∂ i k
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x
   
∂ ∂ ∂ ∂
− 2 ∇ ∂j ∇ ∂ i k , k − 2 ∇ ∂ i k , ∇ ∂j k
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x
   
∂ ∂ ∂ ∂
=2 R , , .
∂xi ∂xj ∂xk ∂xk

29
Proof of 4.
hR(X, Y )W, Z) = −hR(Y, X)W, Zi
= hR(X, W )Y, Zi + hR(W, Y )X, Zi
= −hR(X, Y )Z, W i
= hR(Y, Z)X, W i + hR(Z, X)Y, W i,
so
2hR(X, Y )W, Zi = hR(X, W )Y, Zi+hR(W, Y )X, Zi+hR(Y, Z)X, W i+hR(Z, X)Y, W i.
Interchanging (X, Y ) and (W, Z) gives
2hR(W, Z)X, Y i = hR(W, X)Z, Y i+hR(X, Z)W, Y i+hR(Z, Y )W, Xi+hR(Y, W )Z, Xi.
The terms on the right sides are equal in pairs, so
2hR(X, Y )W, Zi = 2hR(W, Z)X, Y i. 

The first of the above propositions shows that R restricts to a multilinear map
R : Tp M × Tp M × Tp M → Tp M
for each p ∈ M . We can define
R : Tp M × Tp M × Tp M × Tp M → R
by R(x, y, z, w) = hR(x, y)w, zi. Then the last of the above propositions implies that
R satisfies the symmetries
R(x, y, z, w) = −R(y, x, z, w) = −R(x, y, w, z) = R(z, w, x, y)
and
R(x, y, z, w) + R(y, z, x, w) + R(z, x, y, w) = 0.

Reference: Milnor, Morse Theory, §9.

30
Chapter 6

January 26, 2006

A pseudo-Riemannian metric h , i on M is a smoothly varying family p 7→ h , ip


of nondegenerate symmetric bilinear forms on Tp M :

h , ip : Tp M × Tp M → R.

The metric is Riemannian if each h , ip is positive definite. In local coordinates,


n
X
h, i= gij dxi ⊗ dxj .
i,j=1

Fundamental Theorem. Associated to each pseudo-Riemannian metric is a Levi-


Civita connection ∇, a connection which satisfies

A. ∇X Y − ∇Y X = [X, Y ],

1. XhY, Zi = h∇X Y, Zi + hY, ∇X Zi.

 

Pn k ∂ k 1
Pn k` ∂gi` ∂gj` ∂gij
In local coordinate, ∇ ∂
∂xj
= k=1 Γij ∂xk , where Γij = 2 `=1 g ∂xj
+ ∂xi
− ∂x`
.
∂xi

The connection in turn defines the Riemann-Christoffel curvature tensor R:

R(X, Y )Z = ∇X ∇Y Z = ∇Y ∇X Z − ∇[X,Y ] Z.

31
Examples. First, suppose M N = RN with coordinates (x1 , . . . , xN ) and metric

h , i = dx1 ⊗ dx1 + · · · + dxN ⊗ dxN .

Then Γkij = 0, and the Levi-Civita connection D is defined by

N
! N
X ∂ X ∂
DX fI I = X(f I ) .
I=1
∂x I=1
∂xI

Next, suppose that M n ⊆ RN , a submanifold with the induced metric. Vector fields
X, Y on M n can be extended to vector fields X, Y on RN . Note that if p ∈ M n ,
(DX Y )(p) depends only on X(p) and on Y along some curve tangent to X(p). Thus
(DX Y )(p) depends only on the values of X and Y along M n itself.

If p ∈ M n and v ∈ Tp RN , we can write v = v T + v ⊥ , where v T is tangent to M and


v ⊥ is perpendicular to M . IF X : M n → RN is a smooth vector-valued function on
M n , X T is a smooth vector field on M , while X ⊥ is a smooth “section” of the normal
bundle G
N M = {Np M : p ∈ M },
Np M being the orthogonal complement to Tp M in Tp RN ∼
= RN .

We can define ∇ : X(M n ) × X(M n ) → X(M n ) by

(∇X Y )(p) = (DX Y (p))⊥ ,

and check that it satisfies the axioms for a connection on T M . Moreover, ∇ satisfies
axioms A and B. It suffices to check A when X = ∂x∂ i and Y = ∂x∂ j , where (x1 , . . . , xn )
are curvilinear coordinates on M which extend to curvilinear coordinates (x1 , . . . , xN )
on RN , and then
 T
∂ ∂ ∂ ∂
∇ ∂ i j − ∇ ∂j i = D ∂ i j − D ∂j i = 0.
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x

32
Similarly, axiom B holds: if X, Y, Z are tangent to M n ,

XhY, Zi = X(Y · Z)
= (DX Y ) · Z + Y · (DX Z)
= h∇X Y, Zi + hY, ∇X Zi.

Thus ∇ is the Levi-Civita connection on M .

Let Γ(N M ) = {smooth maps X : M → RN : X(p) is perpendicular to Tp M for


p ∈ M } = {smooth sections of the normal bundle N M }.

Definition. The second fundamental form of M n in RN is the map

α : X(M ) × X(M ) → Γ(N M ), α(X, Y )(p) = (DX Y (p))⊥ .

Proposition. α is bilinear over functions and symmetric:


(a) α(f X, Y ) = f α(X, Y ) = α(X, f Y ),
1. α(X, Y ) = α(Y, X).

Proof of (a).

α(f X, Y ) = (Df X Y )⊥
= f (DX Y )⊥
= f α(X, Y ).
α(X, f Y ) = (DX (f Y ))⊥
= ((Xf )Y + f DX Y )⊥
= f (DX Y )⊥
= f α(X, Y ).

33
∂ ∂
Proof of (b). We can take X = ∂xi
and Y = ∂xj
. Then
 ⊥
∂ ∂
α(X, Y ) − α(Y, X) = D ∂ i j − D ∂j i = 0,
∂x ∂x ∂x ∂x

as desired. 

If γ : (a, b) → M n ⊆ RN is a unit-speed curve, the acceleration γ 00 (s) is also called


the curvature of γ at s.

κ(s) = curvature of unit speed γ at s = γ 00 (s).

We set

κg (s) = (γ 00 (s))T
= (∇γ 0 γ 0 )(s)
= geodesic curvature of γ at s,
κn (s) = (γ 00 (s))⊥
= α(γ 0 (s), γ 0 (s))
= normal curvature of γ at s.

The Proposition implies that α restricts to a symmetric bilinear map

α : Tp M × Tp M → Np M

with the following interpretation: If v ∈ Tp M has unit length, α(v, v) is the normal
curvature of a unit-speed curve γ : (−ε, ε) → M with γ(0) = p and γ 0 (0) = v.

Theorem. Suppose M n ⊆ RN with the induced Riemannian metric. Then

hR(x, y)w, zi = α(x, z) · α(y, w) − α(x, w) · α(y, z)

for x, y, z, w ∈ Tp M .

34
This is known as the Gauß equation. It provides an often quite effective means of
calculating the curvature of M n .

Proof. Since Euclidean space has zero curvature, DX DY W −DY DX W −D[X,Y ] W = 0


for X, Y, W ∈ X(M ), and hence if Z ∈ X(M ),

0 = (DX DY W − DY DX W − D[X,Y ] W ) · Z
= X(DY X · Z) − (DY W ) · (DX Z)
− Y (DX W · Z) + (DX W · DY Z) − h∇[X,Y ] W, Zi
= Xh∇Y W, Zi − h∇Y W, ∇X Zi − α(Y, W ) · α(X, Z)
− Y h∇X W, Zi + h∇X W, ∇Y Zi + α(X, W ) · α(Y, Z) − h∇[X,Y ] W, Zi
= h∇X ∇Y W, Zi − α(Y, W ) · α(X, Z)
h∇Y ∇X W, Zi + α(X, W ) · α(Y, Z) − h∇[X,Y ] W, Zi
= hR(X, Y )W, Zi − α(Y, W ) · α(X, Z) + α(X, W ) · α(Y, Z),

as desired. 

Suppose, for example, that


( n+1
)
X
M n = S n (a) = (x1 , . . . , xn+1 ) ∈ Rn+1 : (xI )2 = a2 ,
I=1

the sphere of radius a. If γ : (−ε, ε) → S n (a) is a unit speed great circle


s s
γ(s) = a cos e1 + a sin e2 ,
a a
where e1 and e2 are unit-length vectors perpendicular to each other, then

00 1 s 1 s 1
γ (s) = − cos e1 − sin = − N (γ(s)),
a a a a a
where N (p) is the outward-pointing unit normal to S n (a) at p. Then
1
α(γ 0 (s), γ 0 (s)) = − N (γ(s)),
a
and
1
α(v, v) = − N (p), when v ∈ Tp M, hv, vi = 1.
a

35
By polarization,
1
α(x, y) = − hx, yiN (p) for x, y ∈ Tp M.
a
It therefore follows from the Gauß equation that
1
hR(x, y)w, zi = [hx, zihy, wi − hx, wihy, zi]
a2
for x, y, z, w ∈ Tp M .

36
Chapter 7

January 31, 2006

There is one case in which the curvature of a Riemannian manifold is easy to calcu-
late:

Theorem. Suppose M n is a submanifold of Rn and is given the induced Riemannian


metric h , i. Then the curvature R of M is given by the formula

hR(x, y)w, zi = α(x, z) · α(y, w) − α(x, w) · α(y, z)

for x, y, z, w ∈ Tp M , the dot on the right denoting the usual dot product in RN .

Proof. If D is the Levi-Civita connection on RN ,

DX DY W − DY DX W − D[X,Y ] W = 0 for X, Y, W ∈ X(M n ).

Hence if Z ∈ X(M ),

0 = (DX DY W − DY DX W − D[X,Y ] W ) · Z
= X(DY W · Z) − (DY W ) · (DX Z) − Y (DX W · Z) + (DX W ) · (DY Z) − (D[X,Y ] W ) · Z
= Xh∇Y W, Zi − h∇Y W, ∇X Zi − α(Y, W ) · α(X, Z)
− Y h∇X W, Zi + h∇X W, ∇Y Zi + α(X, W ) · α(Y, Z)
− hD[X,Y ] W, Zi
= h∇X ∇Y W, Zi − α(Y, W ) · α(X, Z)−i∇Y ∇X W, Zi
+ α(X, W ) · α(Y, Z) − h∇[X,Y ] W, Zi
= hR(X, Y )W, Zi − α(X, Z) · α(Y, W ) + α(X, W ) · α(Y, Z),

37
as desired. 

As we saw last time, this enables us to calculate the curvature of

S n (a) = {(x1 , . . . , xn+1 ) ∈ Rn+1 : (x1 )2 + · · · + (xn+1 )2 = a2 },

the result being


1
hR(x, y)w, zi = [hx, zihy, wi − hx, wihy, zi]
a2
for x, y, z, w ∈ Tp S n (a).

The above Theorem can be extended to submanifolds of Minkowski space-time,


the arena for special relativity. Minkowski space-time is Rn+1 with coordinates
(t, x1 , . . . , xn ) and pseudo-Riemannian metric

h , i = −dt ⊗ dt + dx1 ⊗ dx1 + · · · + dxn ⊗ dxn


 
dt
1
1 n  dx 
 
= dt dx · · · dx J  ..  ,
 . 
dxn

where  
−1 0 ··· 0
0 1 · · · 0
J =  .. ..  .
 
..
 . . .
0 0 · 1
Let O(1, n) = {A ∈ GL(n + 1, R) : AT JA = J}, a subgroup of the general linear
group called the Lorentz group. If
   
t t
e
 x1  x 1
e 
= A  ..  ,
 
 .. 
. .
n
x en
x

then A ∈ O(1, n) if and only if

h , i = −de
t ⊗ de x1 ⊗ de
t + de x1 + · · · + de
xn ⊗ de
xn .

38
Elements of O(1, n) are called Lorentz transformations.

We will sometimes identify Rn+1 with its tangent space and denote the metric h , i
on Minkowski space-time by a dot: X · Y = hX, Y i.

The Levi-Civita connection D of Minkowski space-time is defined by


n
! n
∂ X ∂ ∂ X ∂
DX f + f i i = (Xf ) + X(f i ) i .
∂t i=1 ∂x ∂t i=1 ∂x

Of course, the curvature of Minkowski space-time is zero.

Analogous to S n (a) is the submanifold

H n (a) = {(t, x1 , . . . , xn ) : t > 0 and t2 − (x1 )2 − · · · − (xn )2 = a2 }


= {p = (t, x1 , . . . , xn ) : t > 0 and p · p = −a2 }.

The submanifold H n (a) inherits an induced metric from Rn+1 , and we claim that it
is positive definite. Indeed, in terms of the coordinates (x1 , . . . , xn ), we can write
p
t = a2 + (x1 )2 + · · · + (xn )2 ,

and hence Pn i i
i=1 x dx
dt = p ,
a2 + (x1 )2 + · · · + (xn )2
and hence the induced metric is
Pn i j i j
i,j=1 x x dx ⊗ dx
h, i=− 2 + dx1 ⊗ dx1 + · · · + dxn ⊗ dxn .
a + (x1 )2 + · · · + (xn )2
Thus
xi xj
gij = δij − ,
a2 + (x1 )2 + · · · + (xn )2
and one readily checks that the matrix (gij ) is positive definite.

If p ∈ H n (a), and γ : (−ε, ε) → H n (a) is a smooth curve with γ(0) = p, then


γ(t) · γ(t) ≡ −a2 implies that γ 0 (0) · p = 0, so p is perpendicular to H n (a).

39
Lemma. If p ∈ H n (a), there is an element A ∈ O(1, n) such that
 
a
0
p = A  ..  .
 
.
0

Proof. Suppose    
t a cosh α
 x1   a sinh α u1 
p =  ..  =  ,
   
..
.  . 
xn a sinh α un
where  
u1
 .. 
.
un
is a unit-length vector in Rn . Let B be a rotation matrix of Rn such that
 
 
1 1
u 0
 .. 
 .  = B  ..  .
 
.
un
0
Then
 
  cosh α sinh α 0 · · · 0   
1 0 ··· 0   a cosh α
  sinh α cosh α 0 · · · 0 a

0
 ..  a sinh α u1 
 

 ..
 0
 0 1 ··· 0   0 . =  .. ,
. B   .. .. .. ..   . 
 . . . . 0
0 a sinh α un
0 0 0 ··· 1
so we can take
 
  cosh α sinh α 0 · · · 0
1 0 ··· 0 
0   sinh α cosh α 0 · · · 0 
A =  ..
  0
 0 1 ··· 0 ,
. B   .. .. .. .. 
 . . . .
0
0 0 0 ··· 1

40
as desired. 

Lemma. If p ∈ H and v ∈ Tp Hn (a), there is an element A ∈ O(1, n) such that


 
  0
1 c
0  
p = A , v = A 0 for some c ∈ R.
   
···  .. 
.
0
0

Proof. Exercise. 

We can now calculate the geodesics of H n (a). Define γ : R → H n (a) by


 s s 
γ(s) = a cosh , a sinh , 0, . . . , 0 . (∗)
a a
Then γ 00 (s) = a1 cosh as , sinh as , 0, . . . , 0 is a scalar multiple of γ(s); hence (γ 00 (s))T =
  

0. Thus γ is a geodesic. The other geodesics of γ are just the curves

γ
e(s) = Aγ(s),

where A ∈ O(1, n) and A(H n (a)) ⊆ H n (a). These are just parametrizations of inter-
sections of H n (a) with planes passing through the origin.

The Levi-Civita connection ∇ on H n (a) is defined by

∇X Y = (DX Y )T for X, Y ∈ X(H n (a)),

while the second fundamental form α is defined by

α(X, Y ) = (DX Y )⊥ for X, Y ∈ X(H n (a)).

If γ is defined by (∗),
 s s 
γ 0 (s) = sinh , cosh , 0, . . . , 0
a a

41
and
1
α(γ 0 (s), γ 0 (s)) = (γ 00 (s))⊥ = γ(s).
a2
Thus if x ∈ Tp H n (a) and hx, xi = 1,
1
α(x, x) = N (p),
a
where N is the future-pointing normal vector at p with N (p) · N (p) = −1. By
polarization,
1
α(x, y) = hx, yiN (p) for x, y ∈ Tp H n (a).
a
Since N (p) · N (p) = −1, it follows from the theorem presented at the beginning of
the lecture that
1
hR(x, y)w, zi = − [hx, zihw, yi − hx, wihy, zi].
a2

Definition. If σ ⊆ Tp M is a two-dimensional subspace, the sectional curvature of


σ is
hR(x, y)y, xi
K(σ) = 2 2 ,
|x| |y| − hx, yi2
whenever x, y is a basis for σ.

E n has K(σ) ≡ 0 for all 2-planes. S n (a) has K(σ) ≡ a12 for all 2-planes. H n (a) has
K(σ) ≡ − a12 for all 2-planes. These are the spaces of constant curvature, the
Euclidean and non-Euclidean geometries in n dimensions.

Proposition. Let

R, S : Tp M × Tp M × Tp M × Tp M → R

be two quadrilinear functionals satisfying the curvature symmetries. If

R(x, y, x, y) = S(x, y, x, y) for all x, y ∈ Tp M,

then R = S.

42
Thus the sectional curvatures determine the full curvature tensor.

Proof. Let T = R − S. Then T satisfies the curvature symmetries, and

T (x, y, x, y) = 0 for all x, y ∈ Tp M.

Hence

0 = T (x, y + z, x, y + z)
= T (x, y, x, y) + T (x, y, x, z) + T (x, z, x, y) + T (x, z, x, z)
= 2T (x, y, x, z),

and

0 = T (x + z, y, x + z, w) = T (x, y, z, w) + T (z, y, x, w),


0 = T (x + w, y, z, x + w) = T (x, y, z, w) + T (w, y, z, w).

Thus

0 = 2T (x, y, z, w) + T (z, y, x, w) + T (w, y, z, x)


= 2T (x, y, z, w) − T (y, z, x, w) − T (z, x, y, w)
= 3T (x, y, z, w),

as desired. 

43
Chapter 8

February 2, 2006

Let (M, h , i) be a pseudo-Riemannian manifold. A map ϕ : M → M is said to be


an isometry if ϕ∗ h , i = h , i.

Definition. A vector field Z on M n is a Killing field if


h∇X Z, Y i + hX, ∇Y Zi = 0 for all X, Y ∈ X(M n ),
or equivalently
h∇X Z, Xi = 0 for all X ∈ X(M n ).

Note that h∇X Z, Y i(p) depends only on X(p) and Y (p). Thus to show that Z is
Killing it suffices to show that
h∇X Z, Y i(p) + hX, ∇Y Zi(p) = 0
for all vector fields X and Y such that hX, Xi, hY, Y i, and hX, Y i are constant.

Proposition. Let Z be a smooth vector field on a pseudo-Riemannian manifold


(M, h , i) with one-parameter group {ϕt : t ∈ R}. Then if each ϕt is an isometry, Z
is a Killing field.

Remark. The converse is also true, but we will not prove it.

44
Proof. We can assume that hX, Y i is constant. Then hϕt∗ (X), ϕt∗ i is constant, and

hX(p), Y (p)i = hϕt∗p (X(p)), ϕt∗p (Y (p))i


= h(ϕt∗ X)(ϕt (p)), (ϕt∗ Y )(ϕt (p))i
= h(ϕt∗ X)(p), (ϕt∗ Y )(p)i,

so

d
0 = h(ϕt∗ X)(p), (ϕt∗ Y )(p)i
dt
  t=0 
d d
= (ϕt∗ X(p)) Y (p) + X(p), (ϕt∗ Y )(p)

dt t=0 dt t=0
= −h[Z, X], Y i(p) − hX, [Z, Y ]i(p),

and we conclude that


h[Z, X], Y i + hX, [Z, Y ]i = 0. (1)
But ZhX, Y i = 0, so
h∇Z X, Y i + hX, ∇Z Y i = 0. (2)
Subtracting (1) from (2) yields

h∇X Z, Y i + hX, ∇Y Zi = 0. 

Proposition. Suppose that γ : (a, b) → M is a geodesic in the pseudo-Riemannian


manifold (M, h , i). Then

(a) hγ 0 , γ 0 i is constant (geodesics have constant speed).

(b) If Z is a Killing field, hZ, γ 0 i is constant.

Proof.

(a)
d 0 0
hγ , γ i = 2h∇γ 0 γ 0 , γ 0 i = 0;
dt

45
(b)
d
hZ, γ 0 i = h∇γ 0 Z, γ 0 i + hZ, ∇γ 0 γ 0 i = 0. 
dt

This proposition can be used to study qualitative behavior of geodesics on surfaces


of revolution, a surface M 2 ⊆ R3 invariant under the rotation
ϕθ (x, y, z) = (cos θ x − sin θ y, sin θ x + cos θ y, z).
If γ : (a, b) → M 2 ⊆ R3 is the path of a particle of mass m and Z has one-parameter
group {ϕθ : θ ∈ R},
1
mhγ 0 , γ 0 i = kinetic energy, mhγ 0 , Zi = angular momentum.
2
Thus the above proposition states that kinetic energy and angular momentum are
constant.

Thus suppose that F : (a, b) × R → R3 is an immersion of the form


F (u, θ) = (r(u) cos θ, r(u) sin θ, z(u)),
where r > 0, and that (r0 (u))2 + (z 0 (u))2 = 1. For example, if (a, b) = R and
r(u) = 2 + cos u, z(u) = sin u, F describes a torus of revolution. The Riemannian
metric on M is
h , i = F ∗ (dx) ⊗ F ∗ (dx) + F ∗ (dy) ⊗ F ∗ (dy) + F ∗ (dz) ⊗ F ∗ (dz)
= (d(r(u) cos θ))2 + (d(r(u) sin θ))2 + d(z(u))2
= (r0 (u) cos θ du − r(u) sin θ dθ)2 + (r0 (u) sin θ du + r(u) cos θ dθ)2 + (z 0 (u))2 (du)2
= (r0 (u)2 + z 0 (u)2 ) du ⊗ du + (r(u))2 dθ ⊗ dθ
= du ⊗ du + (r(u))2 dθ ⊗ dθ.

Z= ∂θ
is a Killing field on M 2 ⊆ R3 in this case.

If γ(t) = (u(t), θ(t)) is a geodesic on M , 12 hγ 0 , γ 0 i = E, γ 0 , ∂θ





= L, where E and L
are constants. Hence
(
1 du 2
2
+ 12 (r(u))2 dθ

2 dt dt
= E,
2 dθ
(r(u)) dt = L.

46

Eliminating dt
yields
2
L2

1 du
+ = E. (∗)
2 dt 2(r(u))2
This is first order equation for u in terms of t, and the solution can be reduced to an
integration.

8.1 Qualitative Behavior of Solutions to (∗)


L2
We call VL (u) = 2(r(u))2
the “effective potential energy.” We can then interpret (∗) as

kinetic energy + potential energy = constant.

For example, in the case of the torus,

L2
VL (u) = .
2(2 + cos u)2
L2 L2
If 18
<E< 2
, the motion will be confined to one of the potential wells.

One can also calculate the curvature of surfaces of revolution quite easily.

Definition. Suppose that M 2 ⊆ R3 is a surface. The Gaussian curvature of M 2


is the function K : M 2 → R defined by

K(p) = sectional curvature of Tp M.

Thus if M 2 is the image of F : (a, b) × R → R3 defined by

F (u, θ) = (r(u) cos θ, r(u) sin θ, z(u)),

47
where (r0 (u))2 + (z 0 (u))2 = 1, a direct calculation gives

∂F
= (r0 (u) cos θ, r0 (u) sin θ, z 0 (u)),
∂u
∂F
= (−r(u) sin θ, r(u) cos θ, 0),
∂θ
N (u, θ) = unit normal
∂F ∂F
∂u
× ∂θ
= ∂F ∂F

∂u
× ∂θ
= (−z 0 (u) cos θ, −z 0 (u) sin θ, r0 (u)).

Thus
 
∂ ∂ ∂
α , ·N =D∂ ·N
∂u ∂u ∂u ∂u
∂2F
·N
=
∂u2
= −r00 (u)z 0 (u) + r0 (u)z 00 (u),
∂2F
 
∂ ∂
α , ·N = ·N
∂u ∂θ ∂u∂θ
= 0,
∂2F
 
∂ ∂
α , ·N = ·N
∂θ ∂θ ∂θ2
= r(u)z 0 (u).

Hence the Gaussian curvature of M 2 is given by the formula


∂ ∂ ∂ ∂ ∂ ∂ ∂
, ∂
   
α ∂u , ∂u · α ∂θ , ∂θ − α ∂u , ∂θ ·α ∂u ∂θ
K(p) =
∂ ∂
∂ ∂
∂ ∂ 2
,
∂u ∂u
,
∂θ ∂θ
− ∂u , ∂θ
0
z (u) 0
= (r (u)z 00 (u) − z 0 (u)r00 (u)).
r(u)

If r(u) = 2 + cos θ and z(u) = sin θ, we find that


cos u
K(u, θ) = .
2 + cos u

48
Chapter 9

February 7, 2006

Definition. A Lie group is a group G which is also a manifold such that the maps

µ:G×G defined by µ(σ, τ ) = σ · τ

and
ν:G→G defined by ν(σ) = σ −1
are smooth.

Example. GL(n, R) = {n × n real matrices A : det A 6= 0} with µ(A, B) = A · B


(matrix multiplication), ν(A) = A−1 (matrix inverse).

If G is a Lie group and σ ∈ G, define diffeomorphisms Lσ , Rσ : G → G by Lσ (τ ) = σ·τ


and Rσ (τ ) = τ · σ. We call Lσ and Rσ left and right translation by σ.

From Math 240A, Lecture 12, we know that a diffeomorphism F : M → M induces


a map F∗ : X(M ) → X(M ),

F∗ (X)(p) = F∗F −1 (p) (X(F −1 (p))),

and that F∗ ([X, Y ]) = [F ∗ (X), F ∗ (Y )].

If G is a Lie group, a vector field X ∈ X(G) is said to be left invariant if (Lσ )∗ (X) =
X for all σ ∈ G.

49
Let g = {X ∈ X(G) : (Lσ )∗ (X) = X for all σ ∈ G}. g is called the Lie algebra of
G. If X, Y ∈ g, then

(Lσ )∗ ([X, Y ]) = [(Lσ )∗ (X), (Lσ )∗ (Y )] = [X, Y ] for σ ∈ G,

so [X, Y ] ∈ g.

We claim that g is finite-dimensional and that in fact g ∼


= Te G, where e is the identity
element of G. Indeed, we can define linear maps

α : g → Te G, β : Te G → g

by α(X) = X(e), β(x)(σ) = (Lσ )∗e (x). One easily checks that α ◦ β = id, β ◦ α = id,
so α and β are real vector space isomorphisms.

Definition. A Lie group homomorphism is a homomorphism F : G → H be-


tween Lie groups which is smooth.

If F : G → H is a Lie group homomorphism, F induces a linear map F∗ : Te G → Te H


and hence a linear map F∗ : g → h, where g and h are the Lie algebras of G and H.
Note that

F∗ (X)(F (σ)) = (LF (σ) )∗ (F∗ (X)(e)) = F∗σ (Lσ∗ (X(e))) = F∗ (X(σ)),

because LF (σ) ◦ F = F ◦ Lσ , and hence X is F -related to F∗ (X) and F∗ ([X, Y ]) =


[F∗ (X), F∗ (Y )].

More generally, a Lie algebra is a real vector space g together with an operation
[ , ] : g × g → g such that

A. [X, Y ] = −[Y, X],

B. [[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0.

50
A Lie algebra homomorphism between Lie algebras g and h is a linear map
f : g → h such that
f ([X, Y ]) = [f (X), f (Y )].

We have defined a “functor” which assigns to each Lie group G its Lie algebra g and
to each Lie group homomorphism F : G → H the corresponding Lie algebra homo-
morphism F∗ : g → h. This is called the Lie group–Lie algebra correspondence.

Let G be a Lie group. A one-parameter subgroup of G is a Lie group homomor-


phism θ : R → G. If θ : R → G is a one-parameter subgroup of G, then

{ϕt = Rθ(t) : t ∈ R}

is a one-parameter group of diffeomorphisms of G. It therefore generates a vector


field X. Moreover, Lσ ◦ Rθ(t) = Rθ(t) ◦ Lσ , so (Lσ )∗ (X) = X for σ ∈ G. Thus a
one-parameter subgroup θ : R → G generates a corresponding left invariant vector
field Xθ ∈ g. Conversely,

Proposition. If X is a left invariant vector field on a Lie group G, then X is com-


plete, and the integral curve θX : R → G such that θX (0) = e is a one-parameter
subgroup of G.

Proof. Let θX : (a, b) → G be the maximal integral curve for X such that θX (0) = e.
Assume b < ∞. If σ ∈ G,

(σ · θX )0 (t) = (Lσ ◦ θX )0 (t) = Lσ∗ (θX


0
(t)) = (Lσ )∗ (X(θX (t))) = Xσ·θX (t) .

Hence σ · θX = Lσ ◦ θX : (a, b) → G is an integral curve for X such that σ · θX (0) = σ.


In particular, if t0 ∈ (a, b), t0 > 0,

θe : (a + t0 , b + t0 ) → G defined by e = θX (t0 )θX (t − t0 )


θ(t)

is an integral curve for X satisfying θ(t


e 0 ) = θX (t0 ). By uniqueness of integral curves,
θX = θe on (a + t0 , b). Hence θX can be extended to (a, b + t0 ), contradicting maxi-
mality of b. Hence b = ∞. Similarly, a = −∞, and X is complete.

51
The one-parameter group of diffeomorphisms corresponding to X is now seen to be
{ϕt = RθX (t) : t ∈ R}, and

ϕt ◦ ϕs = ϕt+s ⇒ RθX (t) ◦ RθX (s) = RθX (t+s) ⇒ θX (s) · θX (t) = θX t + s,

so θX : R → G is a one-parameter subgroup of G. 

Definition. Suppose that G is a Lie group. A biinvariant pseudo-Riemannian


metric on G is a Riemannian metric h , i such that Lσ and Rσ are isometries for all
σ ∈ G.

For example, we have seen in 240A that

O(n) = {A ∈ GL(n, R) : AT A = I}
2
is a submanifold of GL(n, R) ⊆ Rn . It inherits a Riemannian metric h , i which we
claim is biinvariant. Indeed, if we define
 
a11 · · · a1n
xij : O(n) → R by xij  ... ..  = ai ,

. j
a1 · · · ann
n

then n
X
h, i= dxij ⊗ dxij .
i,j=1

Moreover,
n
X X
xij (AB) = aik xkj (B) ⇒ xij ◦ La = aik xkj
k=1
X X
⇒ L∗A (xij ) = aik xkj ⇒ L∗A (dxij ) = aik dxkj .

52
Hence
n
X
L∗A h , i = L∗A (dxij ) ⊗ L∗A (dxij )
i,j=1
Xn
= aik ai` dxkj ⊗ dx`j
i,j,k,`=1
X n
= dxkj ⊗ dxkj
j,k=1

= h , i,

so LA is an isometry. Similarly, RA is an isometry.

Theorem. Let G be a Lie group with a biinvariant pseudo-Riemannian metric h , i.


Then

(a) geodesics passing through the origin are just the one-parameter subgroup of G.

(b) the Levi-Civita connection on G is defined by


1
∇X Y = [X, Y ] for X, Y ∈ g.
2

(c) the curvature tensor R is given by

hR(X, Y )W, Zi = h[X, Y ], [Z, W ]i for X, Y, Z, W ∈ g.

Proof. Let X be a left-invariant vector field with corresponding one-parameter group


θX . Since each RθX (t) is an isometry, X is a Killing field. Hence

h∇Y X, Zi + hY, ∇Z Xi = 0 for Y, Z ∈ g. (∗)

But since hX, Xi is constant, 0 = Y hX, Xi = h2∇Y X, Xi. Hence it follows from (∗)
that h∇X X, Y i = 0, which implies ∇X X = 0. But then

0 = ∇X+Y (X + Y ) = ∇X Y + ∇Y X,

53
and
1
∇X Y − ∇Y X = [X, Y ] ⇒ ∇X Y = [X, Y ].
2
0
This gives (b), and ∇X X = 0 ⇒ ∇θX0 (t) θX (t) = 0, so θX : R → G is a geodesic,
yielding (a). Finally

R(X, Y )W = ∇X ∇Y W − ∇Y ∇X W − ∇[X,Y ] W
1 1 1
= [X, [Y, W ]] − [Y, [X, W ]] − [[X, Y ], W ]
4 4 2
1 1 1
= [X, [Y, W ]] + [Y, [W, X]] − [[X, Y ], W ]
4 4 2
1 1
= − [W, [X, Y ]] − [[X, Y ], W ]
4 2
1
= − [[X, Y ], W ].
4
If X, Y, Z ∈ g,

0 = 2XhY, Zi = 2h∇X Y, Zi + 2hY, ∇X Zi = h[X, Y ], Zi + hY, [X, Z]i,

and hence
1
hR(X, Y )W, Zi = − h[[X, Y ], W ], Zi
4
1
= h[W, [X, Y ]], Zi
4
1
= − h[X, Y ]i, [W, Z]i
4
1
= h[X, Y ], [Z, W ]i,
4
finishing (c). 

54
Chapter 10

February 9, 2006

If A is an n × n real matrix, let


1 2 1 3 1
eA = I + A + A + ! A + · · · + An + · · · .
2! 3 n!
It is not difficult to see that this infinite series converges.

Let θA (t) = etA . Then θA (t) is a real analytic function of t, and


d tA
(e ) = AetA . (∗)
dt
Indeed,
t2 2 t3 3 tk k
 
d tA d
(e ) = I + tA + A + A + · · · + A + · · ·
dt dt 2! 3! k!
2 k−1
t t
= A + tA2 + A3 + · · · + Ak + · · ·
2! (k − 1)!
t2 2 tk−1
 
k−1
= A I + tA + A + · · · + A + ···
2! (k − 1)!
= AetA .

Thus if b ∈ Rn ,
dx
x(t) = etA · b is a solution to = Ax.
dt
The matrix exponential θA (t) = etA thus yields the solutions to systems of homoge-
neous linear first order ordinary differential equations with constant coefficients.

55
Proposition. etA esA = e(t+s)A .

Proof. X1 (t) = etA · esA and X2 (t) = e(t+s)A are both solutions to the initial value
problem
d(X(t))
= A · X(t), X(0) = esA ,
dt
so they must be equal by uniqueness of solutions to systems of ordinary differential
equations. 

In particular, etA · e−tA = I, so etA ∈ GL(n, R) for all t.

Conclusion. If A is an n × n real matrix, θA (t) = etA is a one-parameter subgroup


of GL(n, R).

Let xij : GL(n, R) → R by


 
a11 · · · a1n
xij  ... ..  = ai .

. j
a1 · ann
n

   
a11 · · · a1n b11 · · · b1n
If A =  ... ..  and B =  .. .. , then

. . .
an1 · · · an n
b1 · · · bnn
n

n
X
xij (B · θA (t)) = xik (B)xkj (θA (t)),
k=1
n
d i X
(xj (B · θA (t))) =
xik (B)xkj (A)
dt t=0 k=1
n
X
= xik (B)akj .
k=1

Hence the tangent vector to the curve


t 7→ BθA (t) = LB (θA (t)) at t=0

56
is n
X ∂
X |B = xik (B)akj .
i,j,k=1
∂xij B

In other words, if A = (aij )


is an n×n real matrix defining the one-parameter subgroup
θA (t), the corresponding left invariant vector field is
n
X ∂
XA = xik akj .
i,j,k=1
∂xij

If n
X ∂
XB = xik bkj ,
i,j,k=1
∂xij
then a direct calculation yields
  XX  
i k ∂ λ ν ∂ λ ν ∂ i k ∂
XX
[XA , XB ] = x k aj i x ν b µ µ − x ν b µ µ x k aj i
i,j,k λ,µ,ν
∂xj ∂xλ λ,µ,ν i,j,k
∂xλ ∂xj
 
X ∂ ∂
= xik akj bj` ` − xik bkj aj` `
i,j,k,`
∂xi ∂xi
X ∂
= xik (xkj (AB) − xkj (BA)) i
i,j,k
∂xj
= X[A,B] ,
where [A, B] = AB −BA. Hence we see that the Lie algebra of GL(n, R) is gl(n, R) =
{n × n real matrices} with bracket [A, B] = AB − BA. Most important Lie groups
are subgroups of GL(n, R), and their Lie algebras are subalgebras of gl(n, R).

T
Thus θA (t) = etA takes values in O(n) iff etA · etA = I. Differentiating and setting
t = 0 yields the relation

d tA tAT
(e · e ) = A + AT = 0.
dt t=0

Thus the Lie algebra of O(n) is


o(n) = {A ∈ gl(n, R) : A + AT = 0}.

57
Proposition. Suppose (M, h , i) is a pseudo-Riemannian manifold, p ∈ M n . Then
there is an open neighborhood V of 0 in Tp M such that if v ∈ Tp M , the unique
geodesic γv which satisfies the initial conditions

γv (0) = p, γv0 (0) = v

is defined for t ∈ [0, 1].

Proof. According to the theory of ordinary differential equations applied to


n
d2 xi X j
i dx dx
k
+ Γjk = 0,
dt2 j,k=1
dt dt

there is a neighborhood W of 0 in Tp M and ε > 0 such that the geodesic γw is defined


on [0, ε] for all w ∈ W . Let V = εW . If v ∈ V , v = εw for some w ∈ W and

d
γw (εt) = εγw0 (0) = εw = v,

dt t=0

so γv (t) = γw (εt). γV is defined on [0, 1]. 

Define expp : V → M by expp (v) = γv (1). expp is called the exponential map.

Note that if v ∈ V , t 7→ expp (tv) is a geodesic (because expp (tv) = γtV (1) = γv (t)),
and hence expp takes line segments through the origin in Tp M to geodesic segments
in M . For example, suppose M is a Lie subgroup G of GL(n, R) with a biinvariant
metric. (For example, G = O(n).) Let p = I, the identity matrix and v = A ∈ g ⊆
gl(n, R). Then the geodesic γA satisfying the initial conditions

γA (0) = I, γA0 (0) = A

is just the one-parameter group θA (t) corresponding to A:

t2 2
γA (t) = θA (t) = etA = I + tA + A + ··· .
2!
Thus expI (A) = eA in this case.

58
Proposition. There is an open neighborhood U e of 0 ∈ Tp M which expp maps dif-
feomorphically onto an open neighborhood U of p in M .

Proof. By the inverse function theorem, it suffices to show that

(expp )∗0 : T0 (Tp M ) → Tp M

is an isomorphism. We identify T0 (Tp M ) with Tp M . If v ∈ Tp M , define λv : R → Tp M


by λv (t) = tv. Then
expp ◦λv (t) = expp (tv) = γv (t),
so
(expp )∗0 (v) = (exp p)∗0 (λ0v (0)) = γv0 (0) = v,
and (expp )∗0 is indeed an isomorphism. 

59
Chapter 11

February 14, 2006

Let (M, h , i) be a Riemannian manifold, expressed in terms of local coordinates


(x1 , . . . , xn ) as
X n
h, i= gij dxi ⊗ dxj .
i,j=1

Idea. Find coordinates near a point p which make the gij ’s as simple as possible.

To do this we use the exponential map

expp : Tp M → M

described last time.

We can always choose local coordinates (y 1 , . . . , y n ) on a neighborhood of p such that


* +
∂ ∂
y i (p) = 0 and , = δij .
∂y i p ∂y j p

These allow us to define coordinates


!
X ∂
ẋ1 , . . . , ẋn : Tp M → R by ẋi aj j
= ai .
∂y p

60
(ẋ1 , . . . , ẋn ) is a global coordinate system on Tp M .

Suppose that expp maps the open neighborhood U e of 0 in Tp M diffeomorphically


only the open neighborhood U of p in M . Define coordinates (x1 , . . . , xn ) on U by
xi = ẋi ◦ exp−1 1 n
p . The coordinates (x , . . . , x ) are called normal coordinates at p.

p
Let r : U → R be defined by r = (x1 )2 + · · · + (xn )2 , and define a radial vector
field R on U − {p} by
n
X xi ∂
R= .
i=1
r ∂xi

Gauß Lemma. dr = hR, ·i on U − {p}; thus the radius field R is perpendicular to


the level sets r =constant.

Definition. If f is any real-valued function on a Riemannian manifold (M, h , i),


the gradient of f is the vector field grad f defined by
df = hgrad f, ·i.
Thus the Gauß Lemma states that R = grad r.

To prove the Gauß Lemma, we make use of the rotation vector field
∂ ∂
Eij = xi j
− xj i .
∂x ∂x

Lemma 1 [Eij , R] = 0.

Proof. The radial vector field R is invariant under rotation. Thus if {ϕt : t ∈ R} is
the one-parameter group corresponding to Eij ,

d
[Eij , R] = − (ϕt∗ (R)) = 0. 
dt t=0

61
Lemma 2 If ∇ is the Levi-Civita connection on M , ∇R R = 0.

Pn
Proof. If (a1 , . . . , an ) are constants with i=1 (a
i 2
) = 1, then the curve γ defined by

xi (γ(t)) = ai t

is an integral curve for R as one verifies by direct calculation. On the other hand,
!
X
i ∂
γ(t) = expp a t i ,
∂x p

so γ is a geodesic. We conclude that the integral curves to R are geodesics, so


∇R R = 0. 

Lemma 3 hR, Ri ≡ 1 on U − {p}.

Proof. Suppose γ is an integral curve for R as above. Then dtd hγ 0 (t), 0


P γi (t)i =
0 0 0 0 0
2h∇γ 0 (t) γ (t), γ (t)i = 0, so γ (t) has constant length. But hγ (0), γ (0)i = (a )2 = 1,
so hγ 0 (t), γ 0 (t)i ≡ 1. It follows that hR, Ri ≡ 1. 

Lemma 4 hR, Eij i = 0.

Proof. We calculate the derivative in the radial direction:

RhR, Eij i = h∇R R, Eij i + hR, ∇R Eij i


= hR, ∇Eij Ri
1
= Eij hR, Ri
2
= 0.

Then hR, Eij i is constant along the geodesic rays emanating from p. Let kXk =
p
hX, Xi. Then
|hR, Eij i| ≤ kRk kEij k ≤ kEij k → 0

62
as r → 0. It follows that hR, Eij i ≡ 0. 

Proof of Gauß Lemma. First note that R(r) = 1, Eij (r) = 0, by direct calculation.
Suppose that X is a smooth vector field on part of U − {p} of the form
n
∂ X
X=f + fij Eij ,
∂r i,j=1

where f and fij are smooth. Then


X
dr(X) = X(r) = f R(r) + fij Eij (r) = f,
D X E
hR, Xi = R, f R + fij Eij = f hR, Ri = f.
Thus dr = R. 

Theorem. Suppose that (M, h , i) is a Riemannian manifold and that U e is an open


ball of radius ε centered at 0 in Tp M such that expp maps diffeomorphically onto an
open neighborhood U of p in M . Suppose that v ∈ U e and that

γ : [0, 1] → M by γ(t) = expp (tv).

Let q = expp (v). If λ : [0, 1] → M n is any smooth curve such that λ(0) = p and
λ(1) = q, then

(i) L(λ) ≥ L(γ), with equality holding only if λ is a reparametrization of γ.

(ii) J(λ) ≥ J(γ) with equality holding only if λ = γ.

Proof. We use normal coordinates (x1 , . . . , xn ) defined on U . Note that L(γ) =length
of γ = r(q). Suppose that λ : [0, 1] → M is a smooth curve with λ(0) = p and
λ(1) = q. We need to show that L(λ) ≥ L(γ).

63
Case I. λ does not leave U . Then
Z 1 p
L(λ) = hλ0 (t), λ0 (t)i dt
0
Z 1
= kλ0 (t)k dt
Z0 1
≥ hλ0 (t), R(λ(t))i dt
Z0 1
= dr(λ0 (t)) dt
0
= r(λ(1)) − r(λ(0))
= r(q)
= L(γ).

Moreover, equality holds if and only if λ0 is a nonnegative multiple of R ◦ λ, i.e. if and


only if λ is a reparametrization of γ.

Case II. λ leaves U at some time t0 < 1. Then


Z 1p
L(λ) = hλ0 (t), λ0 (t)i dt
Z0 t0 p
≥ hλ0 (t), λ0 (t) dt
0
Z t0
≥ hλ0 (t), R(λ(t))i dt
Z0 t0
= dr(λ0 (t)) dt
0
= (r ◦ γ)(t0 )

> L(γ).

This finishes the proof for L. The proof for J is quite similar. 

Recall that if (M, h , i) is a connected Riemannian manifold, we can define a map


d : M × M → R by

d(p, q) = inf{L(γ) : γ : [0, 1] → M is a smooth map with γ(0) = p, γ(1) = q}.

64
The only difficult part in showing that d is a metric is the implication

d(p, q) = 0 ⇒ p = q.

But if q lies in a normal coordinate system centered at p, this implication follows from
the preceding theorem, while if q does not lie in such a coordinate system, the proof
of the theorem shows that d(p, q) > ε for some ε > 0.

65
Chapter 12

February 16, 2006

We can give a small extension of the results established in Lecture 10. Namely, on
an open neighborhood U of the zero-section in T M , we can define
exp : U → M × M by exp(v) = (p, expp (v)),
for v ∈ Tp M .

Proposition. The exponential map makes a small neighborhood V of the zero-


section in T M (with V ⊆ U ) diffeomorphically onto a neighborhood of the diagonal
∆ ⊆ M × M.

The proof is a straightforward application of the inverse function theorem.

It follows from this Proposition and the results of Lecture 11 that any point p has a
neighborhood U with the following property: if q1 , q2 ∈ U , q1 and q2 can be joined by
a unique length-minimizing geodesic (which depends smoothly on (q1 , q2 ) ∈ M × M ).

Recall that a Riemannian manifold (M, h , i) can be made into a metric space (M, d)
by setting
d(p, q) = inf{length of γ | γ : [a, b] → M is a smooth curve with γ(a) = p, γ(b) = q}.
The proof of the theorem given last time shows that if γ : [a, b] → M is a smooth
curve with γ(a) = p and γ(b) = q and length of γ is d(p, q), then γ is a reparametriza-
tion of a constant speed geodesic.

66
Definition. A curve γ : [a, b] → M is a minimal geodesic if it is a constant speed
geodesic and its length is d(p, q).

Question. Given p, q ∈ M , does there exist a minimal geodesic from p to q?

Not Always. If M = R2 − {(0, 0)} with the standard Euclidean metric, p = (−1, 0)
and q = (1, 0), there is no minimal geodesic from p to q.

Definition. A Riemannian manifold (M, h , i) is geodesically complete if geodesics


in M can be extended indefinitely.

Examples of geodesically complete Riemannian manifolds include E n (Rn with the


Euclidean metric), S n (a), and H n (a).

Theorem 1 If (M, h , i) is geodesically complete, then any two points p, q ∈ M n can


be connected by a minimal geodesic.

Theorem 2 (Hopf and Rinow) Let (M n , h , i) be a connected Riemannian manifold.


The following conditions are equivalent:
(a) (M n , h , i) is geodesically complete.
(b) expp : Tp M → M is globally defined for some p ∈ M .
(c) (M, d) is complete as a metric space.

Proof of Theorem 1 We will in fact prove the following assertion (also needed in
the proof of Theorem 2): If p is a point in a Riemannian manifold (M n , h , i) such
that expp : Tp M → M is globally defined, then p can be joined to any point q ∈ M
by a minimal geodesic.

67
A candidate for the minimal geodesic: Let B eε be a closed ball of radius ε centered
at 0 ∈ Tp M lying in an open set which is mapped diffeomorphically by expp onto an
open neighborhood of p.

eε , S = expp (Seε ). S is a compact subset of M , so there is a point m ∈ S


Let Seε = ∂ B
of minimal distance from q. m = expp (εv) where v is a unit-length element of Tp M .
If a = d(p, q), define γ : [0, a] → M by γ(t) = expp (tv). γ is a candidate for the
minimal geodesic from p to q.

It suffices to show that γ(a) = q. Thus it suffices to establish

d(γ(t), q) = a − t (∗)

for all t ∈ [0, a].

Note first that d(γ(t), q) ≥ a−t (because if d(γ(t), q) < a−t then d(p, q) ≤ d(p, γ(t))+
d(γ(t), q) < t+a−t = a). Next note that if (∗) holds to t0 ∈ [0, a] it holds for t ∈ [0, t0 ],
because if t ∈ [0, t0 ],

d(γ(t), q) ≤ d(γ(t), γ(t0 )) + d(γ(t0 ), q) ≤ (t0 − t) + (a − t0 ) ≤ a − t.

Let t0 = sup{t ∈ [0, a] | (∗) holds for t}. By continuity, (∗) also holds for t0 . We will
show

(1) t0 ≥ ε,

(2) 0 < t0 < a leads to a contradiction.

To prove (1), note that

d(p, q) = inf{d(p, r) + d(r, q) : r ∈ S}


= inf{ε + d(r, q) : q ∈ S}
= ε + d(m, q).

68
Hence a = ε + d(m, q), a − ε = d(m, q) = d(γ(ε), q), and (∗) holds for t = ε.

To prove (2), construct a sphere S about γ(t0 ) as we did for p, and let m be a point in
S of minimal distance from q. Then d(γ(t0 ), q) = inf{d(γ(t0 ), r) + d(r, q) | r ∈ S} =
ε + d(m, q), so a − t0 = ε + d(m, q), so d(m, q) = a − (t0 + ε).

Note that d(p, m) ≥ t0 + ε, because otherwise d(p, q) < t0 + ε + a − (t0 + ε) = a.


The broken path from p to γ(t0 ) to m has length t0 + ε. If this broken path had a
corner, one could use the Proposition at the beginning of the lecture to construct a
shorter curve from p to m, contradicting d(p, m) = t0 + ε. Hence m must lie on γ,
m = γ(t0 + ε), and t0 + ε must satisfy (∗), contradicting maximality of (∗). Hence
t0 = a, d(γ(a), q) = 0 and γ(a) = q. 

Proof of Theorem 2. (a)⇒(b): Trivial.

(b)⇒(c): We assume that expp is globally defined for some p ∈ M . If A is a closed


bounded subset of M , let K = sup{d(p, q) : q ∈ A}. It follows from the proof of
Theorem 1 that
A ⊆ expp {v ∈ Tp M : kvk ≤ K}.
A is a closed bounded set of a compact set and hence A must be compact.

If {p1 , p2 , . . . , pi , . . .} is a Cauchy sequence in M n , its closure is contained in a closed


bounded subset of M , hence a compact subset of M by the above argument. Thus
{p1 , p2 , . . . , pi , . . .} must possess a convergent subsequence. This shows that (M, d) is
complete as a metric space.

(c)⇒(a): This follows from the theory of ordinary differential equations, as we will
see next time.

69
Chapter 13

February 21, 2006

To complete the proof of the Hopf-Rinow theorem, we need to show that if (M, d) is
a complete metric space, then geodesics on M can be extended indefinitely.

The equation for geodesics


n
d2 xi X j
i dx dx
k
+ Γjk =0
dt2 j,k=1
dt dt

can be regarded as a first-order system in the variables xi and ẋi :


( i
dx
dt
= ẋi ,
dẋi
= − nj,k=1 Γijk ẋj ẋk .
P
dt

This first-order system corresponds to the vector field


n n
X ∂ X
j k ∂
X= ẋi − Γi
jk ẋ ẋ
i=1
∂xi i,j,k=1 ∂ ẋi

on the tangent bundle T M . Since geodesics do not depend upon the choice of coor-
dinates on M , neither can X. X is called the geodesic spray.

Let T 1 M = {v ∈ T M : kvk = 1}. T 1 M is a codimension one submanifold of M called


the unit tangent bundle. The geodesic spray is tangent to T 1 M because geodesics
have constant speed. Thus we can regard X as a vector field on T 1 M .

70
Let {ϕt : t ∈ R} be the local one-parameter group of diffeomorphisms corresponding
to X. It follows from the theory of ordinary differential equations (see 240A, Lecture
10) that given any v0 ∈ T 1 M , there is an open neighborhood U of v0 ) in T 1 M and
an ε > 0 such that

ϕt (v) is defined for v ∈ V and t ∈ (−ε, ε).

Indeed, since X is never zero, it follows from a Lemma in Lecture 11 that there are
coordinates (y 1 , . . . , y 2n−1 ) on a neighborhood U of v0 such that


X= on U.
∂y 1
From this we see that U is decomposed into a disjoint union of integral curves for X.

Suppose now that γ : [0, a) → M is a unit-speed geodesic which cannot be extended


to a, a < ∞. Let t0 , t1 , . . . , ti , . . . be a sequence of real numbers with ti → a. Then
γ(t0 ), γ(t1 ), . . . , γ(ti ), . . . is a Cauchy sequence in M (since d(γ(ti ), γ(tj )) ≤ |ti − tj |),
and a subsequence must converge to a limit point p0 ∈ M .

After possibly passing to a subsequence, we can assume that γ 0 (ti ) → v0 ∈ T 1 M .


The curve γ 0 : [0, a) → T 1 M enters a coordinate neighborhood (U, (y 1 , . . . , y 2n−1 )) as
described above with X = ∂y∂ i , and γ 0 (ti ) must lie on the orbit through v0 . Hence γ
can be extended so that γ 0 (a) = v0 , a contradiction. 

Suppose that γ : [a, b] → M is a smooth curve. A smooth vector field along γ is a


smooth function X : [a, b] → T M such that X(t) ∈ Tγ(t) M .

Alternatively, we can define γ ∗ T M by

γ ∗ T M = {(t, v) ∈ [a, b] × T M : v ∈ Tγ(t) M }.

We can regard γ ∗ T M as the total space of a vector bundle over [a, b], with projection
π : γ ∗ T M → [a, b] defined by π(t, v) = t. Then a vector field X along γ is a section
of γ ∗ T M ; that is, a smooth map X : [a, b] → γ ∗ T M such that π ◦ X = id.

71
We say that a vector field X along γ is parallel if

∇γ 0 X ≡ 0.

In terms of local coordinates (x1 , . . . , xn ), we can write


X ∂ X dxi ∂
X= fi , γ0 = ,
∂xi dt ∂xi
and " #
n n
X df i X dx k

∇γ 0 X = + Γijk f j i
.
i=1
dt j,k=1
dt ∂x

Proposition. If γ : [a, b] → M is a smooth curve, t0 ∈ [a, b], v ∈ Tγ(t0 ) M , then there


is a unique vector field X along γ such that

∇γ 0 X = 0 and γ(t0 ) = v. (∗)

P i ∂
Proof. Suppose v = a ∂xi . Then (∗) is equivalent to the linear initial value
problem
df i X i dxj k
+ Γjk f = 0, f i (t0 ) = ai ,
dt dt
which has a unique solution on [a, b] by the theory of ordinary differential equations.


If γ : [a, b] → M is a smooth curve, we can define a vector space isomorphism

τγ : Tγ(a) M → Tγ(b) M

by τγ (v) = X(b), where X is the unique parallel vector field along γ such that
X(a) = v. Similarly, we can define τγ if γ is only piecewise smooth. We call τγ
parallel translation along γ.

If X and Y are parallel vector fields along γ, then

γ 0 hX, Y i = h∇γ 0 X, Y i + hX, ∇γ 0 Y i = 0,

72
so hX, Y i is constant along γ. Thus τγ : Tγ(a) M → Tγ(b) M is an isometry of inner
product spaces.

Parallel translation around a geodesic triangle on S 2 (with its metric of constant


curvature one) rotates through an angle ϕ. (It can be shown that the angle is equal
to the area of the triangle.) By contrast, parallel translation around any closed path
in Euclidean space is the identity map.

73
Chapter 14

February 23, 2006

Linear ordinary differential equations are easier to solve than nonlinear ordinary dif-
ferential equations. Given a solution to a nonlinear ordinary differential equation,
one often studies nearby solutions by means of “linearization.” The linearization of
the nonlinear geodesic equation is the Jacobi equation.

Suppose that Map([a, b], M ) denotes the space of smooth maps γ : [a, b] → M and
that
Geod([a, b], M ) = {γ ∈ Map([a, b], M ) | γ is a geodesic}.
By completing with respect to a suitable Banach space norm, we can make Map([a, b], M )
into an infinite-dimensional smooth manifold.

If we are lucky, Geod([a, b], M ) will be a smooth manifold.

The tangent space of Map([a, b], M ) is the space of sections of γ ∗ T M , and we ask:
What is the tangent space to Geod([a, b], M )? To answer this question, we study
“deformations through geodesics.”
Let α : (−ε, ε) → M be a smooth map. then for each s ∈ (−ε, ε), we have a smooth
curve ᾱ(s) : [a, b] → M defined by ᾱ(s)(t). We use the notation ∂α∂s
(s0 , t0 ) = (tangent
∂α
vector to s 7→ α(s, t0 ) at s = s0 ), ∂t (s0 , t0 ) = (tangent vector to t 7→ α(s0 , t) at
t = t0 ). Although ∂α  ∂αand
∂s
∂α
∂t
are only defined along the image of α, their integral
∂α

curves commute on ∂s , ∂t along the image of α. We can let

α∗ T M = {(s, t, v) ∈ (−ε, ε) × [a, b] × T M | v ∈ Tα(s,t) M },

74
which is the total space of a vector bundle over (−ε, ε) × [a, b]. If we let Γ(α∗ T M )
denote the space of sections of α∗ T M , then
∂α ∂α
, ∈ Γ(α∗ T M ).
∂s ∂t
We are interested in the case where γ(t) = α(0, t) is a geodesic. We let
∂α
X(t) = (0, t) = deformation vector field.
∂s
Note that X is a vector field along γ, so X ∈ Γ(γ ∗ T M ).

Proposition. If each ᾱ(s) is a geodesic, then X will satisfy the Jacobi equation

∇γ 0 ∇γ 0 X + R(X, γ 0 )γ 0 = 0.

Proof. ᾱ(s) is a geodesic for every s implies that ∇ ∂α ∂α


∂t
≡ 0, so ∇ ∂α ∇ ∂α ∂α = 0, so
∂t ∂s ∂t ∂t
∂α ∂α ∂α ∂α ∂α ∂α ∂α ∂α
 
∇ ∂ ∇ ∂ ∂t + R ∂s , ∂t ∂t = 0. Thus ∇ ∂α ∇ ∂α ∂s + R ∂s , ∂t ∂t = 0. Now evaluate
∂t ∂s ∂t ∂t
along γ. 

The proof suggests a broader definition of connection. Let (E, π, M ) be a smooth


vector bundle over M , Γ(E) the space of smooth sections of E.

Definition. A connection in the vector bundle E is a map ∇ : X(M )×Γ(E) → Γ(E)


(with ∇X σ denoting ∇(X, σ), for X ∈ X(M ), σ ∈ Γ(E)) such that
1. ∇f X+Y σ = f ∇X σ + ∇Y σ,

2. ∇X (f σ + τ ) = f ∇X σ + (Xf )σ + ∇X τ
for X, Y ∈ X(M ), σ, τ ∈ Γ(E), f ∈ F (M ).

Given a deformation α such as above, we can define a connection ∇ on α∗ T M by

∇ ∂ σ = ∇ ∂α σ, ∇ ∂ σ = ∇ ∂α σ.
∂s ∂s ∂t ∂t

75
This is called the pullback connection.

Solutions to the Jacobi equation are called Jacobi fields.

Suppose γ : [a, b] → M is a geodesic parametrized by arclength and that (e1 , . . . , en )


is an orthonormal basis for Tγ(a) M with e1 = γ 0 (a). We can parallel transport these
vectors to vector fields (E1 , E2 , . . . , En ) along γ so that E10 = γ. We can define the
i
component functions Rjk` (t) by
X
i
R(Ek , E` )Ej = Rjk` Ei .
Pn
If X = i=1 f i Ei , then Jacobi’s equation becomes
n
d2 f i X
i
(s) + R1j1 (t)f j (t) = 0,
dt2 j=1

a first order linear system. This system has a linear space of solutions of dimension
2n. The Jacobi fields which vanish at a given point form a linear subspace of dimen-
sion n.

Example. If M n has “constant sectional curvature”; that is,

R(X, Y )W = k[hY, W iX − hX, W iY ],


Pn
then X = i=1 f i Ei will be a Jacobi field iff

∇γ 0 ∇γ 0 X + R(X, γ 0 )γ 0 = k[hX, γ 0 iγ 0 − hγ 0 , γ 0 iX],

or equivalently, (
d2 f 1
dt2
= 0,
d2 f i
dt2
= −kf for 2 ≤ i ≤ n.
The solutions are

f 1 (t) = a1 + b1 t,

i
√ i

a cos( kt) + b sin( kt)
 if k > 0,
i i i
f (t) = a + b t if k = 0,

 i √ i

a cosh( −kt) + b sinh( −kt) if k < 0.

76
Here (a1 , b1 , . . . , an , bn ) are constants of integration to be determined by the initial
conditions.

If a = 0 and X(a) = 0, the solutions are

f 1 (t) = b1 t,

i

b sin( kt)
 if k > 0,
i
f (t) = bi t if k = 0,

 i √
b sinh( −kt) if k < 0.

Definition. Suppose that γ : [a, b] → M is a geodesic in a Riemannian manifold


with γ(a) = p and γ(b) = q. We say that p and q are conjugate along γ if there is a
nonzero Jacobi field X along γ which vanishes at p and q.

Example. Antipodal points on S n (1) are conjugate along the geodesics which join
them, while H n (1) and E n do not have conjugate points. Great circles which start
at he north pole p focus at the south pole q.

Suppose that M is geodesically complete, p ∈ M , v ∈ Tp M . We have a geodesic


γv : [0, 1] → M defined by γv (t) = expp (tv). We say that v belongs to the conjugate
locus in Tp M if γv (0) and γv (1) are conjugate along γv .

77
Chapter 15

February 28, 2006

Suppose that (M, h , i) is a complete Riemannian manifold, p ∈ M . If v ∈ Tp M , we


let γv : R → M by γv (t) = expp (tv). We say that v belongs to the conjugate locus
in Tp M if γv (0) and γv (1) are conjugate along γv .

Proposition. A vector v ∈ Tp M belongs to the conjugate locus iff expp∗ is singular


at v.

Proof. ⇐: If expp∗ is singular at v, there is a nonzero vector w ∈ Tv (Tp M ) such that


expp∗v (w) = 0. Define α : (−ε, ε) × [0, 1] → M by
α(s, t) = expp (t(v + sw)).
If X(t) = ∂α
∂t
(0, t), X is a Jacobi field along γv which satisfies the condition X(0) = 0.
Moreover,
X(1) = expp∗v (w) = 0.
Hence X is a nonzero Jacobi field which vanishes at γv (0) and γv (1), so γv (0) and
γv (1) are conjugate along γv .

⇒: Conversely, if X is a nonzero Jacobi field along γv with X(0) = 0, then (∇γ 0 X)(0) =
w for some nonzero w ∈ Tp M , and
∂α
X(t) = (0, t),
∂s
where α : (−ε, ε) × [0, 1] → M is the deformation defined by
α(s, t) = expp (t(v + sw)).

78
∂α
Then expp∗v (w) = ∂s
(0, 1) = X(1) = 0, so expp∗ is singular at v. 

Recall that if F : M1 → M2 is a smooth map, a point q ∈ M2 is a regular value for


F if
p ∈ F −1 (q) ⇒ dFp is onto.
Otherwise it is a critical value. According to Sard’s Theorem (240A, Lecture 9),
the set of critical values has measure zero. It therefore follows from the preceding
Proposition that

Corollary. If p ∈ M , the set of q ∈ M which are conjugate to p along some geodesic


has measure zero.

Suppose that (M n , h , i) is a complete Riemannian manifold. A point p ∈ M is a


pole if the conjugate locus in Tp M is empty.

Proposition. If (M, h , i) is a complete Riemannian manifold whose curvature tensor


satisfies
hR(x, y)y, xi ≤ 0 for all x, y ∈ Tq M, q ∈ M,
then any point p ∈ M is a pole.

Proof. Let p ∈ M , v ∈ Tp M , γ(t) = expp (tv). We need to show that p = γ(0) and
q = expp (v) = γ(1) are not conjugate along γ.

Suppose, on the contrary, that X is a nonzero Jacobi field along γ which vanishes at
γ(0) and γ(1). Then
∇γ 0 ∇γ 0 X + R(X, γ 0 )γ 0 = 0,
h∇γ 0 ∇γ 0 X, Xi = −hR(X, γ 0 )γ 0 , Xi ≥ 0.
Pn i
Write X = i=1 f Ei , where (E1 , E2 , . . . , En ) is a parallel orthonormal from field
along γ. Then the above inequality becomes
n
X d2 f i
fi ≥ 0.
i=1
dt2

79
Since X(0) = X(1) = 0, f i (0) = f i (1) = 1, and we can integrate by parts
Z 1 X d2 f i
Z 1 X  df i 2
0≤ fi 2 dt = − dt le0.
0 dt 0 dt
df i
We conclude that dt
≡ 0, f i ≡ 0, and X ≡ 0. 

Theorem. (Hadamard-Cartan) If (M n , h , i) is a complete Riemannian manifold


whose curvature R satisfies the inequality

hR(x, y)y, xi ≤ 0 for all x, y ∈ Tq M, q ∈ M, (∗)

then for any p ∈ M , expp : Tp M → M is a smooth covering.

Recall that a smooth map π : M f → M between connected manifolds is a smooth


covering, and each p ∈ M lies in an open set U which is evenly covered by π; that is,
π −1 (U ) is a disjoint union of open sets, each of which is mapped diffeomorphically by
π onto U .

Corollary. If (M, h , i) is a complete simply connected Riemannian manifold whose


sectional curvatures satisfy (∗), then M n is diffeomorphic to Rn .

f → M is a smooth covering of connected


This follows from the fact that if π : M
f → M is a diffeomorphism.
manifolds and M is simply connected, then π : M

The Hadamard-Cartan Theorem follows from

Proposition. If (M n , h , i) is a complete Riemannian manifold and p ∈ M is a pole,


then expp : Tp M → M is a smooth covering.

We will prove this next time.

80
Chapter 16

March 2, 2006

One of the most studied problems in Riemannian geometry is the relationship be-
tween curvature and topology. To discuss this problem requires an understanding of
some of the basic topological invariants: the cohomology groups H k (M ; R) that we
saw in 240A and the homotopy groups πn (M ).

16.1 The Fundamental Group


Most of you have seen this invariant described in other courses, so our presentation
will be brief.

Let X be a metric space, x0 , x1 ∈ X. We let

Ω(X, x0 , x1 ) = {continuous paths γ : [0, 1] → M | γ(0) = x0 , γ(1) = x1 }.

Suppose γ, λ ∈ Ω(X, x0 , x1 ). We say γ and λ are homotopic and write γ ' λ if


there is a continuous map α : [0, 1] × [0, 1] → X such that α(s, 0) = x0 , α(s, 1) = x1
for all s ∈ [0, 1], α(0, t) = γ(t), α(1, t) = λ(t) for all t ∈ [0, 1].

It is not hard to check that ' is an equivalence relation: γ ' γ, γ ' λ ⇒ λ ' γ,
γ ' λ and λ ' µ ⇒ γ ' µ. Let π(X, x0 , x1 ) denote the set of equivalence classes, [γ]
the equivalence class of γ.

81
If [γ] ∈ π(X, x0 , x1 ) and [λ] ∈ π(X, x1 , x2 ), we define

[γ] · [λ] ∈ π(X, x0 , x2 )

to be the equivalence class of γ · λ, where


(
for t ∈ 0, 12 ,
 
γ(2t)
γ·λ=
λ(2t − 1) for t ∈ 12 , 1 .
 

One can check that this product is well-defined on equivalence classes: γ ' γ 0 and
λ ' λ0 implies γ · λ ' γ 0 · λ0 .

The fundamental groupoid of X is the “category” whose objects are the points of
X and whose morphisms from x0 to x1 are the elements of π(X, x0 , x1 ). The funda-
mental group of X at x0 is π(X, x0 ) = π(X, x0 , x0 ).

To check that the fundamental groupoid is really a category, we need to verify asso-
ciativity:
([γ] · [λ]) · [µ] = [γ] · ([λ] · [µ]) (∗)
and existence of an identity [εx0 ] ∈ π(X, x0 ), for each x0 ∈ X, such that

[εx0 ] · [γ] = [γ], [λ] · [εx0 ] = [λ], (∗∗)

when the products are defined. To establish (∗), we need to show that

(γ · λ) · µ ' γ · (λ · µ),

while to establish (∗∗), we need to verify that

εx0 · γ ' γ, λ · εx0 ' λ.

Explicit homotopies are given in Spanier, Algebraic Topology, Chapter 1, and many
other places.

To make π(X, x0 ) into a group, we need to define an inverse to [γ] ∈ π(X, x0 ). This
is the equivalence class of γ −1 ∈ Ω(X, x0 , x0 ), where

γ −1 (t) = γ(1 − t).

82
One verifies that γ · γ −1 ' εx0 and γ −1 · γ ' εx0 , so [γ] · [γ −1 ] = [εx0 ] = [γ −1 ] · [γ].

A continuous map F : X → Y induces a covariant functor F∗ from the fundamental


groupoid of X to the fundamental groupoid of Y by

F∗ ([γ]) = [F ◦ γ] ∈ π(Y, F (x0 ), F (x1 )) for [γ] ∈ π(X, x0 , x1 ).

This of course defines a homomorphism of fundamental groups

F∗ : π(X, x0 ) → π(Y, F (x0 )).

We have a covariant functor from pointed topological spaces to groups such that

(X, x0 ) 7→ π(X, x0 ),
(F : (X, x0 ) → (Y, F (x0 ))) 7→ (F∗ : π(X, x0 ) → π(Y, F (x0 ))).

If X is arcwise connected and x0 , x1 ∈ X and [γ] ∈ π(X, x0 , x1 ), we can define an


isomorphism

h[γ] : π(X, x0 ) → π(X, x1 ) by h[γ] [λ] = [γ] · [λ] · [γ −1 ].

The fundamental group π1 (X, x0 ) is actually the first of a series of homotopy groups
πn (X, x0 ), n = 1, 2, 3 . . . To define the other homotopy groups, one makes

Ω(X, x0 ) = Ω(X, x0 , x0 )

into a metric space by setting

d(γ, λ) = sup{d(γ(t), λ(t)) : t ∈ [0, 1]}.

One then defines πn (X, x0 ) to be πn−1 (Ω(X, x0 ), εx0 ). This gives a series of groups
which are usually very difficult to calculate. Thus it is still not known how to calcu-
late πn (S m , x0 ) for all values of n and m, for example.

Definition. A metric space X is simply connected if it is arcwise connected and


π1 (X, x0 ) ∼
= {e}, the trivial group.

83
Let X
e and X be metric spaces. A covering from X e to X is a surjective continuous
map π : Xe → X such that any p ∈ X lies in an open neighborhood U which is evenly
covered; even covered means that π −1 (U ) is a disjoint union of open sets each of
which is mapped homeomorphically by π onto U .

Example. Define π : R → S 1 = {z ∈ C : |z| = 1} by π(t) = e2πit .

The first two key facts about coverings are:

Proposition 1 (Unique path lifting) If π : Xe → X is a covering of arcwise connected


spaces, γ : [0, 1] → X is a continuous path and pe ∈ π −1 (γ(0)), then there exists a
unique path γe : [0, 1] → X e(0) = pe and π ◦ γ
e such that γ e = γ.

To prove this, we cover the image of γ with a finite collection of evenly covered open
sets U1 , U2 , . . . , Un . We then lift γ inductively over U1 , U2 , . . . using the defining prop-
erty of evenly covered sets.

Proposition 2 (Unique homotopy lifting) If π : X e → X is a covering of arcwise


connected spaces, α : [0, 1] × [0, 1] → X is a continuous map, and σ e : [0, 1] → X
e is a
continuous path such that α(s, 0) = π ◦ σ e(s) for s ∈ [0, 1], then there is a unique lift
e : [0, 1] × [0, 1] → X
α e such that π ◦ αe = α and αe(s, 0) = σe(s) for s ∈ [0, 1].

The proof is quite similar to that of the previous proposition.

Proposition 3 Suppose π : X e → X is a covering of arcwise connected spaces. If X


is simply connected, π is a homeomorphism.

Proof. We need only show that π is one-to-one. If xe, ye ∈ X


e and π(ex) = π(e
y ), then
since X e : [0, 1] → X
e is arcwise connected, there is a smooth path γ e with σ
e(0) = x e,
e(1) = ye. Let γ = π ◦ γ
σ e. If x = γ(0) = γ(1), γ ∈ Ω(X, x0 ). Since X is simply

84
connected, there is a homotopy α : [0, 1] × [0, 1] → X such that

α(s, 0) = γ(s), α(s, 1) = x, α(0, t) = α(1, t) = x.

By unique homotopy lifting, there is an α e : [0, 1] × [0, 1] → X such that π ◦ αe = α,


α e(s). The path t 7→ α
e(s, 0) = γ e(0, t) projects to a constant and hence must be con-
stant by unique path lifting. Similarly, t 7→ α
e(1, t) and s 7→ αe(s, 1) are constant. It
follows that x
e=α e(0, 0) = α
e(1, 0) = ye. 

If M and Mf are smooth manifolds, a smooth covering from M f to M is just a smooth


map which is also a covering. If π : M → M is a smooth covering and M is simply
f
connected, it follows from Proposition 3 that π is a diffeomorphism, a fact we cited
in the previous lecture.

We now complete the proof of the Hadamard-Cartan Theorem. This will follow from:

Proposition. If (M n , h , i) is a complete Riemannian manifold and p ∈ M is a pole,


then expp is a smooth covering.

Proof of Proposition. Since expp is nonsingular at each v ∈ Tp M , we can define a


Riemannian metric hh , ii on Tp M by

hhx, yii = hexpp∗v (x), expp∗v (y)i for x, y ∈ Tv (Tp M ).

Locally, expp is an isometry from (Tp M, hh , ii) to (M, h , i), and it takes lines through
the origin in Tp M to geodesics in M . Thus lines through the origin must be geodesics
in the Riemannian manifold (Tp M, hh , ii). It follows from the Hopf-Rinow Theorem
that (Tp M, hh , ii) is complete. Thus the Proposition will be a consequence of

f → M is a local isometry of connected Riemannian manifolds with


Lemma. If π : M
M complete, then π is a smooth covering.
f

Proof. One easily checks that π is surjective. Let q ∈ M . We need to show that
q possesses an open neighborhood U which is evenly covered by π. There exists
ε > 0 such that expq maps the open ball of radius 2ε diffeomorphically onto {r ∈

85
M | d(r, q) < ε}. Let {e qα : α ∈ A} = π −1 (q), U = {r ∈ M | d(r, q) < ε},
eα = {e
U r∈M f | d(e
r, qeα ) < ε}. We have a commutative diagram
π∗
Tqeα M
f / Tq M
isometry
exp qeα expq
 
/M
M
f π

(Note that exp qeα is defined on Beε = {ev ∈ Tqeα M


f | ke
v k < ε} by completeness of M
f.)
Since expqeα maps B eα and expq ◦π∗ is a diffeomorphism from B
eε surjectively onto U eε
onto U , π maps Ueα diffeomorphically onto U .

If re ∈ U
eα ∩ U
eβ , we would have geodesics γ eα , γ
eβ of length less than ε from qeα and qeβ
to re. These would project to geodesics γα and γβ from q to r = π(e r) of length less
than ε. By uniqueness of geodesics emanating from q in a normal coordinate chart,
we would have γα = γβ . Since π is a local isometry, γ eα and γeβ would satisfy the same
initial conditions at r. thus γ
eα = γ
eβ and qeα = qeβ . We have proven that U eα ∩ Ueβ 6= ∅,
so α = β.

If re ∈ π −1 (U ), r = π(e
r) ∈ U , and there exists a geodesic γ from r to q of length less
than ε. γ lifts to a geodesic from re to qeα of length less than ε for some qeα ∈ π −1 (q).
Hence re ∈ U eα for some α ∈ A, and π −1 (U ) = S{Uα : α ∈ A}.

We have shown that each q ∈ M lies in an open neighborhood U such that π −1 (U )


is a disjoint union of open sets, each of which is mapped diffeomorphically by π onto
U . This shows that π is a smooth covering. 

86
Chapter 17

March 7, 2006

17.1 Manifolds of Positive Ricci Curvature


Suppose (M, h , i) is a Riemannian manifold with curvature tensor R. The Ricci
curvature
Ric : Tp M × Tp M → R
is defined by

Ric(x, y) = trace of the linear map (v 7→ R(v, x)y).

Thus if (e1 , . . . , en ) is an orthonormal basis for Tp M ,


n
X n
X
Ric(e1 , e1 ) = hR(ei , e1 )e1 , ei i = K(σi ),
i=1 i=2

where σi is the two-plane spanned by e1 , ei . Hence if the sectional curvatures of M


satisfy the inequality
1
K(σ) ≥ 2 ,
a
then the Ricci curvature will satisfy
 
x x n−1 n−1
Ric , ≥ 2
or Ric(x, x) ≥ hx, xi
kxk kxk a a2

for x ∈ Tp M , x 6= 0.

87
The scalar curvature s : M → R is defined by
n
X
s(p) = Ric(ei , ei ),
i=1

whenever (e1 , . . . , en ) is an orthonormal basis for Tp M .

Myers’s Theorem. Let (M n , h , i) be a connected complete Riemannian manifold


whose Ricci curvature satisfies
n−1
Ric(x, x) ≥ hx, xi for x ∈ Tp M, (∗)
a2
where a is a nonzero real number. Then M n is compact and has finite fundamental
group. Moreover, d(p, q) ≤ πa for all p, q ∈ M .

To prove this theorem, we need to take the “second derivative” of the action function

J : Ω(M ; p, q) → R.

Here
Ω(M ; p, q) = {smooth paths γ : [0, 1] → M | γ(0) = p, γ(1) = q}
and Z 1
1
J(γ) = hγ 0 (t), γ 0 (t)i dt.
2 0

We think of J as a function on an “infinite-dimensional manifold” Ω(M ; p, q) and


apply the second-derivative test for local minima. To carry this out, it is convenient
to have a candidate for the tangent space to Ω(M ; p, q) at γ, and we take

Tγ Ω(M ; p, q) = {smooth sections X of γ ∗ T M | X(0) = 0 = X(1)}


= {smooth vector fields X along γ | X(0) = 0 = X(1)}.

The second derivative should then be a symmetric bilinear map

d2 E(γ) : Tγ Ω(M ; p, q) × Tγ Ω(M ; p, q) → R.

If ᾱ : (−ε, ε) → Ω(M ; p, q) is a curve such that ᾱ(0) = γ and the function

α : (−ε, ε) × [0, 1] → M defined by α(s, t) = ᾱ(s)(t)

88
is smooth, we say that ᾱ or α is a deformation of γ with deformation field
X(t) = ∂α
∂s
(0, t). The second derivative is defined by requiring
d2

2

d E(f )(X, X) = 2 J(ᾱ(s))
ds s=0

whenever ᾱ is a deformation with deformation field X.


d2
So we need to calculate ds J(ᾱ(s)) s = 0, and we find that

2

d2 d2 1 1 ∂α ∂α
 Z   

J(ᾱ(s)) = , dt
ds2
s=0 ds 2 2
0 ∂t ∂t
Z 1   s=0
∂ ∂α ∂α
= ∇ ∂α , dt
0 ∂s
∂s ∂t ∂t
Z 1   s=0  
∂α ∂α ∂α ∂α
= ∇ ∂α ∇ ∂α , + ∇ ∂α , ∇ ∂α dt
∂s ∂s ∂t ∂t ∂s ∂t ∂s ∂t
0
Z 1     s=0
∂α ∂α ∂α ∂α
= ∇ ∂α ∇ ∂α , + ∇ ∂α , ∇ ∂α dt
∂s ∂t ∂s ∂t ∂t ∂s ∂t ∂s
0
Z 1       s=0
∂α ∂α ∂α ∂α ∂α ∂α
= R , , + ∇ ∂α ∇ ∂α ,
0 ∂s ∂t ∂s ∂t ∂t ∂s ∂s ∂t
 
∂α ∂α
+ ∇ ∂α , ∇ ∂α dt
∂t ∂s ∂t ∂s
Z 1    s=0  
∂α ∂α ∂α ∂α ∂ ∂α ∂α
= R , , + ∇ ∂α ,
0 ∂s ∂t ∂s ∂t ∂t ∂s ∂s ∂t
   
∂α ∂α ∂α ∂α
− ∇ ∂α , ∇ ∂α + ∇ ∂α , ∇ ∂α dt .
∂s ∂s ∂t ∂t ∂t ∂s ∂t ∂s
s=0

The second term integrates to zero because


   
∂α ∂α
∇ ∂α (0, 0) = 0 = ∇ ∂α (0, 1).
∂s ∂s ∂s ∂s

The third term vanishes because γ is a geodesic. We are left with


Z 1
d2

[hR(X, γ 0 )X, γ 0 i + h∇γ 0 X, ∇γ 0 Xi] dt,

2
J(ᾱ(s)) =
ds s=0 0

and by “polarization,” we conclude that


Z 1
2
d E(γ)(X, Y ) = [h∇γ 0 X, ∇γ 0 Y i − hR(X, γ 0 )γ 0 , Y i] dt.
0

89
We can integrate by parts to obtain
Z 1
2
(d J)(γ)(X, Y ) = − h∇γ 0 ∇γ 0 X + R(X, γ 0 )γ 0 , Y i dt. (∗∗)
0

Proof of Myers’s Theorem. Suppose that p, q ∈ M and d(p, q) > πa. Let γ :
[0, 1] → M be a geodesic which minimizes J. Since |γ 0 | is constant, |γ 0 | = d(p, q).
1
Choose parallel orthonormal vector fields (E1 , . . . , En ) along γ such that E1 = d(p,q) γ0.
Let Xi (t) = (sin πt)Ei (t) for i = 2, . . . , n. Then Xi (0) = 0 = Xi (1), so Xi ∈
Tγ Ω(M ; p, q). Then

∇γ 0 Xi = −π cos πt Ei ,
∇γ 0 ∇γ 0 Xi = −π 2 sin πt Ei ,
h∇γ 0 ∇γ 0 Xi , Xi i = −π 2 sin2 πt.

On the other hand,

hR(Xi , γ 0 )γ 0 , Xi i = d(p, q)2 sin2 πthR(Ei , E1 )E1 , Ei i.

Substituting into (∗∗) yields


Z 1
2
d J(γ)(Xi , Xi ) = [π 2 sin πt − d(p, q)2 sin2 πt R(Ei , E1 , E1 , Ei )] dt
Z0 1
= sin2 πt[π 2 − d(p, q)2 R(Ei , E1 , E1 , Ei )] dt.
0

Hence
n
X Z 1
2
d J(γ)(Xi , Xi ) = sin2 πt[(n − 1)π 2 − d(p, q)2 Ric(E1 , E1 )] dt.
i=2 0

But

d(p, q) > πa ⇒ d(p, q)2 > π 2 a2
⇒ d(p, q)2 Ric(E1 , E1 ) > (n − 1)π 2 .
Ric(E1 , E1 ) ≥ n−1
a2

Hence n
X
d2 J(γ)(Xi , Xi ) < 0 and d2 J(γ)(Xi , Xi ) < 0
i=2

90
for some i, 2 ≤ i ≤ n. This contradicts minimality of the geodesic. Thus it must be
the case that d(p, q) ≤ πa.

It follows that M is compact because it lies in the image under the exponential map
of the ball of radius πa.

To prove that π1 (M ) is finite, we use the fact that any connected smooth manifold
M has a universal cover M f: a simply connected manifold such that π : M f → M is
a smooth covering. Give M f the metric π ∗ h , i, where h , i is the Riemannian metric
on M . Then Mf is also complete and has Ricci curvature satisfying (∗). Hence M f is
−1 −1
also compact, and if p ∈ M , π (p) must be a finite set. But the points of π (p) are
in one-to-one correspondence with the elements of π1 (M, p). Thus π1 (M ) is finite.


91
Chapter 18

March 9, 2006

In the proof of Myers’s Theorem, we used the following theorem from topology:

Theorem. Let M be a connected smooth manifold. Then there is a smooth manifold


M f → M. M
f which is simply connected and a smooth covering projection π : M f is
uniquely determined up to diffeomorphism.

We call M
f the universal cover of M .

If p ∈ M , pe ∈ Mf, and π(e p) = p, we can define a map h : π1 (M, p) → π −1 (p) by


h([γ]) = γ e : [0, 1] → M is the unique map such that π ◦ γ
e(1), where γ e = γ and
γ
e(0) = pe.

Proposition 2 from Lecture 16 shows that h is well-defined, i.e. γ


e(1) does not depend
on the choice of γ ∈ [γ]. We claim that h is a bijection. h is onto because M f is path
f is simply connected; hence if qe ∈ π −1 (p) and γ
connected. h is injective because M e, λ
e
are two curves in Mf with γ e(0) = λ(0)
e = pe and γe(1) = λ(1)
e = qe, then γe ' λ,
e and
hence π ◦ γ
e ' π ◦ λ.
e

Example. M = S 1 = {z ∈ C | |z| = 1} has universal cover M f = R with projection


2πit
π defined by π(t) = e . (This is actually a group homomorphism.) In this case,
h : π1 (S 1 , 1) → Z, and one can check that h is a group isomorphism.

92
This example shows that universal covers can sometimes be used to calculate funda-
mental groups.

Using them, one can show, for example, that π1 (S 1 × S 1 ) ∼= Z ⊕ Z. For n ≥ 2, S n


is simply connected, and R × S is the universal cover of S 1 × S n . It follows that
n

π1 (S 1 × S n ) ∼
= Z for n ≥ 2. Hence S 1 × S 2 does not have a metric of positive Ricci
curvature by Myers’s Theorem. It therefore does not have a metric of positive sec-
tional curvature.

Open Problem. Does S 2 × S 2 have a metric with positive sectional curvature?

Can we find further relationships between curvature and the fundamental group? The
answer is “yes,” and a famous example is

Synge’s Theorem. Let (M, h , i) be a compact Riemannian manifold with positive


sectional curvatures.

1. If M is even-dimensional and oriented, it must be simply connected.

2. If M is odd-dimensional, it must be orientable.

Recall that RP n is S n with antipodal points identified. There is a natural projection


π : S n → RP n which exhibits S n as the universal cover of RP n . Since π is two-to-one,
π1 (RP n ) ∼
= Z2 . Moreover, there is a Riemannian metric h , i on RP n such that π ∗ h , i
is the standard Riemannian metric on S n of constant curvature one.

If n is odd, RP n is orientable, in agreement with the second statement of Synge’s


theorem. If n is even, RP n is not orientable, hence not a counterexample to the first
statement of Synge’s theorem.

93
The proof of Synge’s theorem is based upon the notion of “free homotopy class.”
Let Map(S 1 , M ) = {continuous maps γ : S 1 → M }, where S 1 = [0, 1] with the
points 0 and 1 identified. We say that two elements γ, λ ∈ Map(S 1 , M ) are freely
homotopic, and write γ ' λ, if there is a continuous map
(
α(0, t) = γ(t),
α : [0, 1] × S 1 → M such that
α(1, t) = λ(t).

Free homotopy is just homotopy without regard to base points. It is easily verified
that ' is an equivalence relation. We let [S 1 , M ] be the set of equivalence classes and
write [γ] for the equivalence class of γ ∈ Map(S 1 , M ).

It is not too difficult to show that [S 1 , M ] is in one-to-one correspondence with the set
of conjugacy classes in π1 (M ), when M is a connected manifold. Thus a connected
manifold is simply connected iff [S 1 , M ] consists of a single point.

If M is connected, [S 1 , M ] can be regarded as the set of components of Map(S 1 , M ).


One of the components contains all of the constant paths
c = {γ : S 1 → M | γ(t) ≡ p for all t ∈ [0, 1]}.
M
Elements of the component of Map(S 1 , M ) which contains M
c are said to be null
homotopic.

We will establish the following theorem next time:

Lemma. If (M, h , i) is a compact Riemannian manifold and λ : S 1 → M is not freely


homotopic to a constant path, then there is a smooth closed geodesic γ : S 1 → M
which is freely homotopic to λ and minimizes the action integral J among all smooth
curves freely homotopic to γ.

Here the action integral J is defined by


1 1 0
Z
J= hγ (t), γ 0 (t)i dt.
2 0

94
Let us assume the Lemma for now and prove Synge’s Theorem with its help. The
proof uses the “second derivative” of J in a manner quite similar to that employed
in Myers’s Theorem.

We give a slightly new meaning to the symbol Map(S 1 , M ). We need to regard it


as the space of smooth maps γ : S 1 → M . We want to think of Map(S 1 , M ) as an
“infinite-dimensional smooth manifold” with tangent space

Tγ Map(S 1 , M ) = {smooth sections X of γ ∗ T M }.

A deformation of γ is a map ᾱ : (−ε, ε) → Map(S 1 , M ) such that ᾱ(0) = γ which


is smooth in the following sense: if we define

α : (−ε, ε) × S 1 → M by α(s, t) = ᾱ(s)(t),

then α is smooth.

To each such deformation is associated a deformation vector field


∂α
t 7→ X(t) = (0, t),
∂s
which is an element of Tγ Map(S 1 , M ). Conversely, given X ∈ Tγ Map(S 1 , M ), we
can construct a deformation ᾱ : (−ε, ε) → Map(S 1 , M ) which has deformation vector
field X.

As in the proof of Myers’s Theorem, we can calculate

d2


J(ᾱ(s)) = d2 J(γ)(X, X).
ds2
s=0

he steps are virtually the same, except that we do not have to worry about endpoints.
The result is
Z 1
2
d J(γ)(X, X) = [h∇γ 0 X, ∇γ 0 Xi − hR(X, γ 0 )γ 0 , Xi] dt. (∗)
0

95
If γ : S 1 → M minimizes in its free homotopy class, we must have d2 J(γ)(X, X) ≥ 0
for all X ∈ Tγ Map(S 1 , M ).

Let us now turn to part 1 of Synge’s Theorem. If M is not simply connected, the
Lemma says there is a smooth closed geodesic γ : S 1 → M which minimizes in its free
homotopy class. Let p = γ(0). Parallel translation around γ gives an isomorphism
τγ : Tp M → Tp M .

τγ is defined by τγ (x) = X(1), where X(t) is the unique section of γ ∗ T M → [0, 1] such
that ∇γ 0 X ≡ 0 and X(0) = x. Note that |τγ (x)| = |x|; that is, τγ is an orthogonal
transformation of Tp M . Since M is oriented, τγ preserves orientation.

We can choose an orthonormal basis (e1 , . . . , en ) for Tp M such that e1 = 1


L(γ)
γ 0. With
respect to such a basis, τγ is represented by an element of

SO(n) = {A ∈ GL(n, R) : AT A = I and det A = 1}.

A canonical form theorem for elements of SO(n) states that after conjugation, we can
arrange that τγ is represented by a matrix of the form

±1
 
 ±1 
..
 

 . 

±1
 
 
A= cos θ1 − sin θ1 .
 
sin θ1 cos θ1
 
 
 .. 

 . 

 cos θk − sin θk 
sin θk cos θk

Since the determinant of A is one, the number of (−1)’s is even. Since dim M is
even the number of (+1)’s is also even. There must be a +1 that corresponds to
1
e1 = L(γ) γ 0 , and hence there must be at least one more +1 corresponding to e2 . Thus
there is a unit-length vector field E2 along γ such that ∇γ 0 E2 ≡ 0, E2 (1) = E2 (0),

96
and E2 ⊥ E1 . Then
Z 1
2
d J(γ)(E2 , E2 ) = − hR(E2 , γ 0 )γ 0 , E2 i dt,
0

which is less than 0 since sectional curvatures are assumed positive. This yields a
contradiction. Hence M must be simply connected.

The proof of part 2 is similar; one uses a smooth closed geodesic γ : S 1 → M such
that parallel translation around γ is orientation-reversing. 

97
Chapter 19

March 14, 2006

To finish the proof of Synge’s Theorem, we need to establish

Theorem. Let (M, h , i) be a compact Riemannian manifold. If λ : S 1 → M is


a smooth map which is not freely homotopic to a constant, then there is a smooth
closed geodesic γ : S 1 → M freely homotopic to γ which minimizes the action integral
J among all smooth curves freely homotopic to λ.

Here is S 1 is regarded as [0, 1] with endpoints identified, and

1 1 0
Z
J(γ) = hγ (t), γ 0 (t)i dt.
2 0

Let Map(S 1 , M ) = {piecewise smooth paths γ : S 1 → M }. Informally, we think of


Map(S 1 , M ) as a smooth manifold with tangent space

Tγ Map(S 1 , M ) = {piecewise smooth sections of γ ∗ T M }.

How do we make this picture rigorous? One approach is through finite-dimensional


approximations. More precisely, we let

Map(S 1 , M )c = {γ ∈ Map(S 1 , M ) | J(γ) ≤ c}.

The Cauchy-Schwarz inequality will enable us to approximate this space by a finite-


dimensional manifold.

98
Let N be a large integer, and for integers i such that 0 ≤ i ≤ N , let ti = Ni . Let
BGN (S 1 , M ) be the set of continuous maps γ : [0, 1] → M with γ(0) = γ(1) and
γ |[ti−1 ,ti ] is a smooth constant speed geodesic for 1 ≤ i ≤ N } and

BGN (S 1 , M )c = {γ ∈ BGN (S 1 , M ) | J(γ) ≤ c}.

We have a map
h : BGN (S 1 , M )c → M
| × ·{z
· · × M}
N

defined by h(γ) = (γ(t1 ), γ(t2 ), . . . , γ(tN )).

Recall that the Cauchy-Schwarz inequality asserts that


Z b 2 Z b Z b
2
f (t)g(t) dt ≤ (f (t)) dt (g(t))2 dt,
a a a

with equality holding if and only if f and g are linearly dependent.

p
We set f (t) = hγ 0 (t), γ 0 (t)i and g(t) = 1. Then
Z b 2 Z b Z b
p 0 0
hγ 0 (t), γ 0 (t)i dt ≤ hγ (t), γ (t)i dt a dt,
a a a

which can be written


(L(γ))2 ≤ 2(b − a)J(γ),
where Z bp Z b
1
L(γ) = hγ 0 (t), γ 0 (t)i dt, J(γ) = hγ 0 (t), γ 0 (t)i dt.
a 2 a

Thus if J(γ) ≤ c, √
L(γ) ≤ b − a · c. (∗)
By taking b − a sufficiently small, we can arrange that L(γ) < δ where δ > 0 is any
prescribed number. Since M is compact, there exists δ > 0 such that if p, q ∈ M ,
d(p, q) < δ, p and q are connected by a unique geodesic γp,q of length d(p, q) which
depends smoothly on p and q. Indeed, we can choose δ > 0 so that if Ω(M ; p, q)2δ =

99
{piecewise smooth γ : [0, 1] → M such that γ(0) = p, γ(1) = q, and L(γ) ≤ 2δ} with
the topology defined by the metric

d(γ1 , γ2 ) = sup{d(γ1 (t), γ2 (t)) : t ∈ [0, 1]},

we ca define a continuous map

H : Ω(M ; p, q) × [0, 1] → Ω(M ; p, q)

such that H(γ, 0) = γ and H(γ, 1) = γp,q .

q
1
We can choose N so that N
· c < δ, so that it follows from (∗) that
 
L γ |[ i−1 , i ] < δ.
N N

If γ ∈ Map(S 1 , M )c , we let g(γ) = γ e |[ i−1 , i ] is the unique geodesic from


e, where γ
N N

γ i−1 i
  1
N
to γ N
. This defines a map g : Map(S , M ))c → BGN (S 1 , M )c . In fact,
with the metric space topology defined above, the homotopy H shows that g is a
“homotopy equivalence.”

We call γ
e a broken geodesic and regard it as an approximation to γ.

Proof of the Theorem. Let

µ = inf{L(γ) : γ ∈ Map(S 1 , M ) and γ is freely homotopic to λ}.

Choose c > µ. Choose a sequence γi ∈ Map(S 1 , M )c such that J(γi ) → µ. Then


γ
ei = g(γi ) is a broken geodesic with J(e γi ) → µ. h(e
γi ) = (γi (t1 ), γi (t2 ), . . . , γi (tN )) ∈
| × ·{z
M · · × M}. But M ×^ · · · × M N is compact. Hence a subsequence of {h(e γi ) : i =
N
1, 2, . . .} will converge to a point (p1 , . . . , pN ) ∈ M × · · · × M . Let γ be the broken
geodesic such that γ |[ti−1 ,ti ] is a constant speed minimal geodesic from pi−1 to pi .
Then J(γ) = µ.

γ is piecewise smooth with possible corners at ti = Ni . But γ cannot have corners


because existence of a corner would enable us to construct a piecewise smooth path

100
b : S 1 → M with J(b
γ γ ) < µ by smoothing off the corner. Hence γ is a smooth closed
geodesic with J(γ) = µ. 

A similar argument could be used to give an alternate proof of the following theorem
from Lecture 12:

Theorem. Let (M, h , i) be a complete Riemannian manifold, p, q ∈ M . Then there


exists a smooth geodesic
γ : [0, 1] → M
such that γ(0) = p, γ(1) = q, and L(γ) = d(p, q).

101
Chapter 20

March 16, 2006

Let (M, h , i) be a smooth Riemannian manifold with


n
X n
X
i j
h, i= gij dx ⊗ dx = geαβ dy α ⊗ dy β ,
i,j=1 α,β=1

in two different coordinate systems (x1 , . . . , xn ) and (y 1 , . . . , y n ). Then


n
X ∂xi ∂xj
geαβ = gij ,
i,j=1
∂y α ∂y β

so 2
∂xi
 
det(e
gαβ ) = det(gij ) det ,
∂y α
so  i 
∂x
q q
det(e
gαβ ) = det(gij ) det .
∂y α
It follows that if M is oriented and the coordinate systems (x1 , . . . , xn ) and (y 1 , . . . , y n )
are positively oriented,
q p
det(gij ) dx1 ∧ · · · ∧ dxn = det geαβ dy 1 ∧ · · · ∧ dy n .

On a positively oriented Riemannian manifold (M, h , i), we can thus define a volume
form Θ by q
Θ= det(gij ) dx1 ∧ · · · ∧ dxn ,

102
whenever (x1 , . . . , xn ) are positively oriented coordinates on M and h , i = gij dxi ⊗
P
dxj .

If K is a compact region within M , then


Z
volume of K = Θ.
K

We can apply this to the case of a compact Lie group G.

If G is any Lie group, any inner product

h , i : Te G × Te G → R

extends to a unique left invariant Riemannian metric on G such that

hLσ∗ (x), Lσ∗ (y)iσ = hx, yie for x, y ∈ Te G,

where Lσ is left translation, defined by Lσ (τ ) = σ · τ .

Can we construct a metric which is also invariant under right translation, Rσ : G → G,


Rσ (τ ) = τ · σ? Note that Rσ (e) = e, so (Rσ )∗ : Te G → Te G. Thus given an inner
product h , i on Te G, we can define

f : G → L2 (Te G; R) by f (σ) = Rσ∗ h , i.

If G is compact, we can then set


R
f (σ) Θ
hh , ii = GR
∈ L2 (Te G; R).
G

This will be an inner product on Te G which extends to a biinvariant Riemannian


metric hh , ii on G.

The conclusion is that any compact Lie group has a biinvariant Riemannian metric.
For example, if G = O(n),

o(n) = TI O(n) ∼
= {n × n matrices A | A + AT = 0},

103
and hhA, Bii =trace(AT B) defines a biinvariant metric on O(n).

It follows from Lecture 9 that if G is a compact Lie group with a biinvariant Rieman-
nian metric h , i, then the curvature of G is given by
1
hR(X, Y )W, Zi = h[X, Y ], [Z, W ]i
4
for X, Y, Z, W ∈ g, the Lie algebra of G. Thus curvature is given by a simple explicit
formula in this case.

We can also determine curvatures of certain submanifolds of G. To do this, we need


to generalize the Gauß equation from Lecture 7. Suppose (M, h , i) is a Riemannian
manifold, M0 an imbedded submanifold. Let X(M0 ) be the set of smooth vector
fields on M0 and X(M0 )⊥ the set of smooth vector fields in M along M0 which are
perpendicular to M . Let i : M0 → M be the inclusion. Then

(i∗ T M )p = T pM0 ⊕ Np M0 ,

where Np M0 = {v ∈ Tp M : v ⊥ Tp M0 }. As p varies over M0 , we get a “direct sum”


of vector bundles i∗ T M = T M0 ⊕ N M0 and

X(M0 ) = Γ(T M0 ), X(M0 )⊥ = Γ(N M0 ).

Let D : X(M ) × X(M ) → X(M ) be the Levi-Civita connection on M . We can then


define

∇ : X(M0 ) × X(M0 ) → X(M0 ) by ∇X Y = (DX Y )T ,


α : X(M0 ) × X(M0 ) → X(M0 )⊥ by α(X, Y ) = (DX Y )⊥ ,

where (·)T and (·)⊥ are orthogonal projections to tangent and normal space, respec-
tively.

Lemma. ∇ is the Levi-Civita connection on M and α is symmetric and bilinear over


functions.

104
The proof is exactly the same as before (when M was Euclidean space).

The symmetric bilinear form α : X(M0 ) × X(M0 ) → X(M0 )⊥ is called the second
fundamental form of M0 in M . Note that α is defined pointwise and restricts to a
bilinear map
α : Tp M0 × Tp M0 → Np M0
for each p ∈ M0 .

Theorem. Suppose (M, h , i) is a Riemannian manifold, M0 is an imbedded sub-


manifold with the induced metric (also denoted by h , i). If R is the curvature of M
and R0 is the curvature of M0 , then

hR0 (x, y)w, zi = hR(x, y)w, zi + hα(z, z), α(y, w)i − hα(x, w), α(y, z)i, (∗)

for x, y, z, w ∈ Tp M .

Proof. Same as before. 

Definition. The submanifold M0 ⊆ M is said to be totally geodesic if α ≡ 0.

Note that if γ : (−ε, ε) → M0 ⊆ M , then

(Dγ 0 γ 0 )(t) = ((Dγ 0 γ 0 )(t))T + ((Dγ 0 γ 0 )(t))⊥


= ∇γ 0 γ 0 (t) + α(γ 0 (t), γ 0 (t)).

Thus if M0 is totally geodesic, Dγ 0 γ 0 = ∇γ 0 γ 0 .

Indeed, we easily see that M0 ⊆ M is totally geodesic iff geodesics in M0 are also
geodesics in M .

105
If M0 ⊆ M is totally geodesic, the Gauß equation (∗) becomes

hR0 (x, y)w, zi = hR(x, y)w, zi for x, y, z, w, ∈ Tp M0 .

How do we construct totally geodesic submanifolds?

Theorem. Let (M, h , i) be a compact Riemannian manifold, F : M → M an


isometry, M0 a connected component of

Fix(F ) = {p ∈ M | F (p) = p}.

Then M0 is an imbedded totally geodesic submanifold of M .

Proof. Suppose p ∈ M0 , and let

Bε = {v ∈ Tp M : kvk < ε}.

Choose ε > 0 sufficiently small that expp maps Bε diffeomorphically onto an open
subset U of p in M . Since F is an isometry, it takes geodesics to geodesics. Hence if
v ∈ Tp M , γ(t) = F (expp (tv)) is a geodesic in M and γ 0 (0) = F∗p (v). Thus

F (expp (tv)) = γ(t) = expp (tF∗p (v)),

and hence
F ◦ expp = expp ◦F∗p : Bε → U.
Let W = {v ∈ Tp M | F∗p (v) = v}, a linear subspace of Tp M . Moreover,

ϕ : exp−1
p : U → Tp M

maps U ∩ M0 bijectively onto Bε ∩ M0 if ε is sufficiently small. Thus (U, ϕ) is a


submanifold chart for M0 and M0 is indeed an imbedded submanifold of M . By
construction, geodesics through p in M0 are also geodesics through p in M . So M0 is
totally geodesic. 

We now return to the example of a compact Lie group with a biinvariant Riemannian
metric h , i. Can we construct isometries of G which have interesting fixed point

106
sets?

Note that ν : G → G define by ν(σ) = σ −1 takes biinvariant metrics to biinvariant


metrics and
ν∗e = −id : Te G → Te G,
so ν must be an isometry. Similarly, if s : G → G is an automorphism of G such that
(s∗ )id is an isometry, s will be an isometry.

Suppose now that s is an automorphism such that


(i) s∗ h , i = h , i and

(ii) s2 = id.
Let K = {σ ∈ G : s(σ) = σ}, a compact subgroup of G. Of course, K is a totally
geodesic submanifold of G, but we can construct an even more interesting one as
follows:

Define an isometry F : G → G by F (σ) = s(σ −1 ) and let

M0 = {σ ∈ G | F (σ) = σ} = {σ ∈ G | s(σ) = σ −1 },

a totally geodesic submanifold. Define

β : G × M0 → M0 by β(σ, τ ) = στ s(σ −1 ).

(Note that s(στ s(σ −1 )) = s(σ)s(τ ) · σ −1 = s(σ)τ −1 σ −1 = (στ s(σ −1 ))−1 , so β(σ, τ ) ∈
M0 .) The map τ 7→ β(σ, τ ) is an isometry for each σ ∈ G. Note that β(σ, e) = e iff
σs(σ −1 ) = e iff e ∈ K. This implies that β induces a map h : G/K → M0 by

h(σK) = σs(σ −1 ).

One can check that h is a bijection. Hence we havegiven a manifold structure to the
quotient G/K and have a totally geodesic imbedding of G/K in G.

Example. Let G = SO(p + q) and define s : G → G by S(A) = JAJ, where


 
−Ip 0
J= .
0 Iq

107
Then K = {A ∈ P (p) × O(q) : det A = 1} and G/K is the Grassmann manifold of
p-planes in Rp+q . Thus the Grassmann manifold is a totally geodesic submanifold of
SO(p + q).

Note that F∗I : TI SO(p + q) → TI SO(p + q) by F∗I (A) = JAJ and

TI M0 = {A ∈ gl(p + q, R) | A + AT = 0, JAJ = −A};

thus the matrices in TI M0 are of the form


 
0 X
.
−X T 0
   
0 X 0 Y
If x = and y = are elements of TI M0 , we find therefore
−X T 0 −Y T 0
that
1
hR0 (x, y)y, xi = h[x, y], [y, x]i,
4
or  
1 (XY T − Y X T )2 0
hR0 (x, y)y, xi = trace .
4 0 (X T Y − Y T X)2
This gives a very explicit formula for the curvature of the Grassmann manifold.

This construction gives a large class of manifolds for which we can actually calculate
the curvature quite explicitly.

108
Chapter 21

Problems

1. Let U = {(r, θ) : r > 0, 0 < θ < 2π}. Define F : U → R3 by
 √ 
r cos(√ 2θ)
1
F (r, θ) = √  r sin( 2θ)  .
2 r

(a) Show that the metric induced on U is

h , i = dr ⊗ dr + r2 dθ ⊗ dθ.

(b) Show that if we let x = r cos θ, y = r sin θ, the metric takes the form

h , i = dx ⊗ dx + dy ⊗ dy.

(c) Sketch the geodesics in U and their images under F in R3 .

2. Let M 2 be an orientable imbedded surface in Euclidean space R3 and N :


M 2 → R3 a choice of smoothly varying unit normals on M n . Then the second
fundamental form α : Tp M × Tp M → Np M satisfies

α(x, y) = h(x, y)N (p), for x, y ∈ Tp M,

where h : Tp M × Tp M → R is a real-valued symmetric bilinear form.

(a) Let (e1 , e2 ) be an orthonormal basis for Tp M . Use the Gauß equations to
show that
h(e1 , e1 ) h(e1 , e2 )
K(p) =
.
h(e2 , e1 ) h(e2 , e2 )

109
(b) By citing an appropriate theorem from linear algebra, show that there is
an orthonormal basis for Tp M such that h(e1 , e2 ) = 0.
(c) If (e1 , e2 ) is such an orthonormal basis, h(e1 , e1 ) and h(e2 , e2 ) are called
principal curvatures of M 2 at p, and are denoted by κ1 (p) and κ2 (p).
The Gaussian curvature is then given by the formula K(p) = κ1 (p)κ2 (p).
Suppose that F : R × S 1 → R3 by

F (u, eiv ) = (cosh u cos v, cosh u sin v, sinh u),

and let M 2 be the image of F . Find an orthonormal basis (e1 , e2 ) for Tp M


for each p ∈ M , determine the principle curvatures κ1 and κ2 , and find the
Gaussian curvature K : M 2 → R. M 2 is called the hyperboloid.

3. Suppose that (M, h i) is a connected Riemannian manifold which has the fol-
lowing property: Given p, q ∈ M n , there is an isometry F : M → M such that
F (p) = q. Show that (M n , h i) is geodesically complete.

4. Consider the paraboloid of revolution z = x2 +y 2 in R3 which can be parametrized


by
F (u, v) = (u, v, u2 + v 2 )
or in terms of polar coordinates u = r cos θ, v = r sin θ,

Fe(r, θ) = (r cos θ, r sin θ, r2 ).

(a) Show that the curves θ = constant are geodesics.



(b) Show that ∂θ
is a Jacobi field.
(c) Show that the point (0, 0, 0) on the paraboloid is a pole.

5. Let F : (0, ∞) × S 1 → R3 by
u √
 Z 
iv −u −u
F (u, e ) = e cos v, e sin v, 1 − e−2w dw .
1

Let M 2 be the image of F , a two-dimensional submanifold of R3 .

(a) Sketch M 2 .
(b) Calculate the coefficients of the metric induced by F on (0, ∞) × S 1 .
(c) Calculate the second fundamental form.
(d) Determine the Gaussian curvature κ.

110
6. (a) Let (M, h , i) be a Riemannian manifold, C a one-dimensional submani-
fold which is the fixed-point set of an isometry. Show that C consists of
geodesics.
(b) Suppose that M = {(x, y) ∈ R2 | x2 + y 2 < 1} and
4
h, i= (dx ⊗ dx + dy ⊗ dy).
(1 − x2 − y 2 )2

Find the geodesics which pass through the origin.


(c) Parametrize these geodesics by arclength.
(d) Show that (M, h , i) is geodesically complete.

7. Let G be a compact Lie group with a biinvariant Riemannian metric h , i.


Suppose that the Lie algebra g of G satisfies the condition

{X ∈ g : [X, Y ] = 0 for all Y ∈ g} = {0}.

Use Myers’s Theorem to show that G has finite fundamental group.

111

You might also like