Differential Geometry

Lecture Notes1

written by Clemens Koppensteiner

Version 1.0
June 10, 2010

1 Following

the Part III lecture of the same name by Dr. J. Ross. Michaelmas term 2009, Cambridge.

These notes are not endorsed by Julius Ross. They have not been carefully proofread and might
therefore contain lots of errors. Please notify me of any you find at [email protected].
The most current version of this document is available from http://www.caramdir.at/math.

This work is licensed under the Creative Commons Attribution-Share Alike 3.0 Austria License. To
view a copy of this license, visit http://creativecommons.org/licenses/by-sa/3.0/at/ or send
a letter to
Creative Commons
171 Second Street, Suite 300
San Francisco
California, 94105
USA.

Contents
1 Manifolds

2 Tangent space

3 The tangent bundle

4 Lie groups

5 Vector fields and flows

10
5.1 The Lie derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6 Submanifolds

13

7 Integral manifolds

15

8 Vector bundles
8.1 Vector bundle constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17
18

9 Tensors and exterior algebra

20

10 Differential forms

23

11 De Rham cohomology

24

12 Integration

26

13 Connections
13.1 Connections as derivations . . . . . . .
13.2 Coordinate description of connections
13.3 Connections and parallel transport . .
13.4 Horizontal splitting . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
. .

28
28
29
30
31

14 Curvature

31

15 Linear connections

33

16 Geodesics and the exponential map

36

17 Curvature of Riemannian manifolds

39

Preliminaries
Notation The symbol A o B means A is an open subset of B. This notation is not used in the
lecture, but is useful to avoid the constant reoccurring where U is an open subset of Rn .

1 Manifolds
Definition 1.1. A chart on a set M is a bijection x : U Rn from U M to some open set
x(U ) Rn .

An atlas on M is a collection of charts (x , U ) such that M is covered by the U and for all , :
x(U U ) is open in Rn and the transition map
x x1
: x (U U ) x (U U )
is smooth (i.e. C ).
Two atlases are compatible, if their union is an atlas. This gives an equivalence relation on the set
of atlases. An equivalence class of this relation is called a differentiable structure. A manifold is a set
with a differentiable structure.
Remarks 1.2.
Given a chart (x, U ) and a function f : M R. We abuse notation and confuse f |U with
f x1 . E.g. if x(p) = (x1 , . . . , xn ) we write f (x1 , . . . , xn ) for f (p).
If x(p) = 0, we say the chart (x, U ) is centred at p.
A differentiable structure defines a topology on M by V M is open if and only if x (V U )
is open for all charts (x , U ). (In particular, all U are open.) We always assume that this
topology is Hausdorff and second countable (i.e. has a countable basis).
Given charts x : U Rn and x : U Rn with U U 6= , the transition map gives
a homeomorphism from some nonempty subset of Rn to the same in Rm . By a theorem in
topology this implies m = n. Hence there is a well-defined notion of dimension dimp M . If M
is connected, then dim M is well-defined.
Examples 1.3.
M = Rn with the single chart x : Rn Rn , x(t) = t.
M = R. Let A be the atlas given by the usual differential structure and let the atlas A0 be
1
given by the single chart y : R R, y(t) = t3 . Since t t 3 is not smooth, the atlases A and
A0 are not compatible.
M = S n with stereographic projection.
If U is an open subset of any manifold, then U is a manifold with charts U U for charts U
of M .
Mn = {n n-matrices with real values}
= Rn ;
GLn = {A Mn : det A 6= 0} is open in Mn and so is a manifold.
2

Real projective space RPn with the usual charts.

C
= R2 is a manifold as is Mn (C), GLn (C) and PCn .
Definition 1.4. Let M and N be manifolds. A function F : M N is smooth at p M , if there
exists a chart (x, U ) of M with p U and a chart (y, V ) of N with F (p) V such that y F x1 is
smooth at x(p). We set
C (M ) = {f : M R smooth}.
A function F : M N is a diffeomorphism if it is smooth, bijective and has a smooth inverse. We
set
Diff(M ) = {F : M M diffeomorphism}.
Note that if a function is smooth with respect to some charts, it is smooth with respect to every
chart (by the chain rule).
Example 1.5. Let M = R with the usual differential structure and N = R with the atlas A0 from
above (i.e. given by y(t) = t3 ). Then F : M N , F (t) = t1/3 is a diffeomorphism.

2 Tangent space
Notation. For U o Rn , f : U R, p U and v Rn the directional derivative is
Dv |p (f ) = lim

h0

f (p + vh) f (p)
.
h

Definition 2.1. Let M be a manifold and p M . A tangent vector to M at p is a derivation on

the algebra of germs of M at p.

d
Example 2.2. Suppose : (, ) M is smooth with (0) = p. Define (0)(f

) = dt
f . Then
t=0
(0)

is a tangent vector at p.
Notation. Let x1 , . . . , xn be coordinates around p (i.e. x = (x1 , . . . , xn ), (x, U ) is a chart). Set


f 

:= Dei |x(p) (f x1 ).
(f ) =
xi p
xi p
(This is the above for (t) = (0, . . . , 0, t, 0, . . . , 0).)
Definition 2.3. The set of tangent vectors at p M forms a vector space, called the tangent space
to M at p and denoted Tp M .
Definition 2.4. Let F : M N be a smooth map and p M . The derivative of F at p is the linear
map
DF |p : Tp M TF (p) M given by DF |p (v)(f ) = v(f F ).
It is also denoted by (F )|p .
The derivative indeed is well-defined (exercise).
Theorem 2.5 (Chain Rule). If f : M M 0 and G : M 0 N are smooth functions of manifolds,
then
D(G F )|p = DG|F (p) DF |p .
Proof. Let v Tp M
D(G F )|p (v)(f ) = v(f G F ) = DF |p (v)(f G) = DG|F (p) (DF |p (v))(f ).

Corollary 2.6. If F is a diffeomorphism, then DF |p is an isomorphism with (DF |p )1 = DF 1 |F (p) .





Theorem 2.7. Given coordinates x1 , . . . , xn around p, x
1 p , . . . , xn p are a basis of Tp M .
Proof. First, we consider the case that p = 0 U o Rn .
Claim: Suppose that U is convex. Given f : U R smooth, there exist hi : U R such that
P
f (r1 , . . . , rn ) = f (0, . . . , 0) + ri hi (r1 , . . . , rn ) and hi (0) = Dei f |0 .
Indeed, fix r U and set G(t) = f (tr), t [0, 1]. Then, by the fundamental theorem of calculus
and the chain rule,
Z
f (r) f (0) = G(1) G(0) =
0

Set hi (r) =

R 1 Pn
0

i=1

G0 (t)dt =

Z
0

n
1X

Dei f (tr)ri dt.

i=1

Dei f (tr)dt.

Next we will show that De1 |0 , . . . , Den |0 is a basis for T0 Rn . Let v T0 Rn . Given f , write
P
f (r) = f (0) + i ri hi (r) as above. Then
X
X
X
ri hi (r) =
v(ri )|0 hi (0) + ri |0 v(hi ) =
v(ri )|0 Dei f |0 .
v(f ) = v(f (0)) +v
| {z }
i

=0, since f (0) is constant

P
Set ai = v(ri )|0 . Then v =
ai Dei |0 and hence the Dei |0 span the tangent space. Because of
Dei |0 (rj ) = ij , they are also linearly independent.


For the general case we have to show that x
are a basis of Tp M . Consider the diffeomorphism
i
p
x : U x(U ). Without loss of generality x(p) = 0. Then Dx|p : Tp M T0 R is an isomorphism and


one checks that Dx|p ( x
i p ) = Dei |p .


P
P


Concretely this means that for v Tp M , v = i v(xi ) x
. [If v = i ai x
, then v(xj ) =
i
i
p
p
P i
j
i a ij = a .]
Corollary 2.8 (Transformation Law). Suppose y 1 , . . . , y n are another set of coordinates at p. Then

X xi  


 .
=
j  xi 
y j p
y
p
p
i
In particular, for v Tp M with v =
Proof.

ai


xi p

bj

ai

xi 

i y j  xi p
p


xi p

bj


y j p ,

we have ai =

bj

xi 
y j p .

From any linear function f : U R (p U o M ) we get a linear map

Df |p : Tp M Tf (p) R
=R
and hence an element the dual space Tp M = (Tp M ) . It is denoted df |p .
Lemma 2.9.

1. dx1 p , . . . , dxn |p is the dual basis to

P f 
i
2. df |p = i x
i  dx p .


x1 p

,...,


xn p .

Hence df |p carries the same information as the gradient.

Proof.
1. Consider g : R R, t 7 t:








i
i
dx p
(g) =
(g x ) =
(xi ) = ij .
xj p
xj p
xj p
2. df |p


xj p


xj p

(f ) =

f 
xi p .

3 The tangent bundle

Definition
` 3.1. Let M be a manifold. Define the tangent bundle T M of M as follows:
` As a set
T M = pM Tp M . For every chart (x, U ) of M define a chart (tU , T U ) on T M by T U = pU Tp M
and
tU : T U x(U ) Rn R2n ,
vp Tp M 7 (x(p), (a1 (p), . . . , an (p))t ),

P

where vp = i ai (p) x
.
i
p

This gives a differential structure on T M , which makes T M into a manifold of dimension 2 dim M :
Given another chart (y, V ) the transition map is given by
n
n
tU t1
V : y(U V ) R x(U V ) R ,

(, (b1 , . . . , bn )t ) 7 (xy 1 (), (a1 , . . . , an )t ),

j
where ai = j bj x
i y . So they are smooth. Also this gives a topology on T M that is Hausdorff
and second countable (exercise).
Lemma 3.2. The map : T M M mapping vp Tp M to p is smooth.
Proof. Given p M and a chart (x, U ) around p we have
1
n t
x t1
U (, (a , . . . , a ) ) = .

Consider a function X : U T U with X = idU . Then X =

see that X is smooth if and only if all ai are smooth.

i
ai x
i for some a : U R. We

Definition 3.3. A smooth map X : U T M with X = idU is called a vector field on U .

We will often write Xp for X(p). The C (U )-module of vector fields on U is denoted Vect(U ).
Here multiplication is pointwise, i.e. (f X)(p) = f (p)Xp . Further, Vect(U ) acts on C (U ) by
P g
(Xg)(p) = Xp (g)(p) =: Xp (g). (Locally Xg = ai x
i , so it is smooth.)
Definition 3.4. Let F : M N be smooth. Then define
DF : T M T N,

vp Tp M 7 DF |p (vp ) TF (p) N.

Proposition 3.5. DF is smooth and

TM

DF

TN

commutes.
F (p) N .
Proof. Let dim M = m and dim N = n. Pick charts (x, U ) around p M and
(y, V ) around
P j 

1
n
i
i 1
m
Say F = (F , . . . , F ) where F = F (x , . . . , x ) are smooth. Then DF |p xi p = j a yj 
F (p)








F j 
j
j
j
a = DF |p
(y
)
=
y

F
=
.
xi p
xi p
xi p




In other words, with respect to the basis x
of Tp M and x
of TF (p) N , DF |p is given by
i
j
p
F (p)


j
the matrix F
. [By abuse of notation this matrix is also denoted DF |p .] Thus
xi 

with

j,i

1
n t
1
tV DF t1
(), DF |p (b1 , . . . , bn )t ,
U (, (b , . . . , b ) ) = yF x
| {z }
the matrix

which is smooth.
Warning. While it is possible to push forward individual tangent vectors via DF , for a vector
field X Vect(M ), DF (X) does not make sense in general, as for q N there might be more than
one point in F 1 (q).

On the other hand, if F is a diffeomorphism, then DF (X) makes sense:

(F X)|q = DF (X)|q = DF |F 1 (q) (XF 1 (q) ).
Remark 3.6. We can think of vector fields a derivations on the space of smooth functions:


X(gh) (p) = Xp (gh) = g(p)Xp (h) + h(p)Xp (g),
i.e. X(gh) = gX(h) + hX(g).
Proposition 3.7. Given any derivation X : C (M ) C (M ), there exists a unique vector field
X on M such that X (g)(p) = Xp (g) for all g C (M ). This defines an isomorphism between
derivations on C (M ) and Vect(M ).
For the proof the following construction is needed, that will also come in handy later on.
Lemma 3.8 (Bump Functions). Let p U o M . Then there exists an open subset p U 0 U and
a function C (M ) such that 1 on U 0 and supp() U .
Proof. Define:

(t) =

e1/t , t > 0
0,
t0

(t)
(t) + (1 t)
(t) = (2 t)(2 + t).

(t) =

Given > 0, set h (x) =

kxk
, x Rn . If (x, U ) is a chart, set

(q) =

h (x(q)), q U
0,
q
/U

for some small enough.

Proof of proposition. Suppose f |W = g|W for some p W o M . We need to show that X (p)(f ) =
X (p)(g). Pick a bump function such that 1 on some p W 0 and supp W . Then f = g
on M . Hence
X (f ) = X (g)
X (f ) + f X () = X (g) + gX ().
Thus X (f )(p) = X (g)(p) and so Xp is well-defined on germs. As X is a derivation, Xp is one. So
Xp Tp M .

P i

. Then
To check that p 7 Xp is smooth, introduce local coordinates Xp =
(p) x
i
p

i (p) = Xp (xi ) = X (xi )p
for a suitable bump function . So i and hence X is smooth.
A section X : M T M is smooth if and only if X(f ) is smooth for all f C (M ).
If X, Y Vect(M ), then XY does not need to be a vector field as
X(Y (gf )) = X(gY (f ) + f Y (g)) = X(g)Y (f ) + gX(Y (f )) + X(f )Y (g) + f X(Y (g))
does not give a derivation. However, XY Y X does give a derivation.

Definition 3.9. Let X and Y be vector fields on M . Then the Lie bracket of X and Y is the vector
field [X, Y ] = XY Y X.
Proposition 3.10.
1. [, ] makes Vect(M ) into a (possibly infinite dimensional) Lie algebra, i.e.
a) [, ] is bilinear and anticommutative ([X, Y ] = [Y, X]);
b) [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0 ( Jacobi identity)
2. If f, g C (M ), then [f X, f Y ] = f g[X, Y ] + f (Xg)Y g(Y f )X.

3. If x1 , . . . , xn are coordinates, then [ x

i , xj ] = 0 for all i, j.

4 Lie groups
Definition 4.1. A Lie group G is a manifold G that is also a group such that the maps
m : G G G, (g, h) 7 gh and i : G G, g 7 g 1
are smooth.
Example 4.2. GL(n, R), S 1 C and S 3 H (where H are the quaternions and S 3 is thought of as
the unit quaternions) are Lie groups.
Given g G, consider Lg : G G, h 7 gh, which is smooth with smooth inverse Lg1 . Then
DLg |e : Te G Tg G is an isomorphism.
Definition 4.3. The Lie algebra of the Lie group G is g = Te G.
To justify this definition, we need to define a Lie algebra structure on g. Recall that if F : M M
is a diffeomorphism and X a vector field on M , then we can define another vector field on M by
(F X)|p = DF |F 1 (p) ( X|F 1 (p) ) Tp M .
Lemma 4.4. [F X, F Y ] = F [X, Y ].
Proof. exercise
Definition 4.5. A vector field X on a Lie group G is called left-invariant if (Lg ) X = X for all
g G. The space of all left-invariant vector fields on G is denoted VectL (G).
By the lemma, the Lie bracket of two left-invariant vector fields is again left-invariant.
Proposition 4.6. For g set define X by (X )g = DLg () Tg G. The map 7 X gives a
isomorphism g VectL (G).
Proof. X is a vector field: Given f C (X):
(X )g (f ) = DLg ()(f ) = (f Lg ).
Since m is smooth, this is smooth with respect to g (exercise).
Because of


(Lg ) X h = DLg |g1 h ( X |g1 h ) = DLg |g1 h ( DLg1 h e ()) = DLh |e () = X |h
(chain rule), X is left-invariant.
Conversely, if X VectL (G), let = Xe . Then X = X .
This gives a Lie algebra structure on g by X[,] = [X , X ].

5 Vector fields and flows


d
) where
Recall: Given a smooth curve : (, ) M , we set (t
0 ) = D|t0 ( dt
t0

d
dt t0

Tt0 R.

Definition 5.1. Let X be a vector field on M .

1. An integral curve of X through p M is a smooth curve : T M (with I o R an interval
with 0 I) such that (0) = p and (t)

= X(t) for all t I.

2. A flow of X is a smooth : I U M (with 0 I o R and U o M ) such that for all p U ,
t 7 (t, p) is an integral curve of X through p.
P i
Remark 5.2. Suppose that in local coordinates x1 , . . . , xn , X =
a xi and p = (p1 , . . . , pn ). Then
= (1 , . . . , n ) is an integral curve if and only if i (0) = pi and i (t) = ai ((t)) for all i.
Theorem 5.3 (Existence and Uniqueness of Solutions of ODEs). Let 0 U o Rm and X : U Rm
Lipschitz. Then there exist > 0 and a > 0 such that there exists a unique flow : (, ) Ba (0)
d
(t, p) = X((t, p)), (0, p) = p for all p Ba (0).) Furthermore, if X is smooth,
Rm of X (i.e. dt
then so is (i.e. the solutions of an ODE depend smoothly on the initial conditions).
Proof. See Lang, Analysis II, chapter VI.
Theorem 5.4. Let X be a vector field on M and p M . Then there exists a unique maximal integral
curve : I M of X through p (i.e. if
e : Ie M is a another integral curve through p, then Ie I
and |Ie =
e).
S
Proof. Let I = Ic be the union over all integral curves c : Ic M of X through p. By the
existence
theorem I =
6 . We want to show uniqueness (even outside of a chart around p). Let S =

t > 0 : 1 |(0,t) = 2 |(0,t) 6= , S I1 I2 . Set t0 = sup S and assume t0 < min {sup I1 , sup I2 }.
By continuity, 1 (t0 ) = 2 (t0 ). The curves t 7 1 (t + t0 ), t 7 2 (t + t0 ) are integral curves
through 1 (t0 ) = 2 (t0 ). By uniqueness (in the ODE theorem), there is a neighborhood of 0 such
that 1 (t + t0 ) and 2 (t + t0 ) agree. Hence there is a neighborhood of t0 with 1 = 2 , contradicting
the choice of t0 .
Hence t0 = min {sup I1 , sup I2 }. An analogous statement holds for negative time. So we can
patch the curves together to give the maximal curve .

Example 5.5. Let M = R2 and consider X = x

. Let p = (a, b) M . For a curve = (1 , 2 ) with
(0) = (a, b) to be an integral curve it must fulfill





d 
d1 
d2 
= D
=
+
,
x (t)
dt t
dt x (t)
dt y (t)

and hence

So (t) = (a + t, b).
For M = R2 \ {0} and X =
defined for t < .

d1
= 1,
dt

x ,

d2
= 0.
dt

is not defined on all of R. In fact at p = (, 0), it is only

Example 5.6. Let M = R, X = x x

, (0) = p and an integral curve. Then (t)

= (t) and so
t
(t) = pe . Notice that if p = 0, then (t) 0 is constant.

Theorem 5.7 (Existence of Flows). Let X be a vector

T field on M and for p M let p : Ip M be
the maximal integral curve of X through p. Set I = pM Ip (possibly I = {0}) and : T M M ,
(t, p) 7 p (t). Write t (p) = (t, p). Then:
1. 0 = idM ;

10

2. is smooth;
3. t+s = t s wherever this is defined.
is called the flow of X.
Proof.
1. Obvious as p (0) = p.
2. Comes from the last part of the ODE theorem.
3. Fix s. Then both t 7 t s (p) and t 7 t+s (p) are the unique integral curve through s (p).
Hence they are equal.
Definition 5.8. We say that a vector field X is complete if for every point p M the maximal
integral curve through p has domain R.
Theorem 5.9. If supp(X) is compact (e.g. if M is compact), then X is complete.
Proof. For each p M there is an open interval Ip around 0 and an open neighborhood Up M of p
such that there is a flow cq (t) on Ip Up .
Say supp(X) = T
{p M : Xp 6= 0} K compact. Then there exist p1 , . . . , pm M such that Upi
cover K. Set I = m
i=0 Ipi (is an open interval). For t I define

t (q) =

cpi (t)
0

if q supp(X), q Upi
if q
/ supp(X)

So t is a flow defined on all of M , but only for t I.

For the extension to all t R fix t and pick N N such that

t
N

I. Set

N
t = Nt Nt = Nt
.
|
{z
}
N times

This does not depend on the choice of N and gives a flow, i.e. s+t = s t (exercise).

5.1 The Lie derivative

Consider C (M ) as a vector space. Then Diff(M ) acts on C (M ) by (, f ) 7 f := f . Let
X Vect(M ) be a vector field with flow t (t defined in some interval around 0). Then t 7 t f is a
path in C (M ) through the point f C (M ) at t = 0.
Definition 5.10. The Lie derivative of f along X is

d 
t f f

LX f :=

f
=
lim
.
t
t0
dt t=0
t
Proposition 5.11. LX f = Xf .
Proof. At p M consider the integral curve cp (t) = t (p) through p. By definition
Xcp (0) = Xp . Thus


d 
d 
f

(p)
=
f cp (t) = Xp (f ).
LX f (p) =
t
dt t=0
dt t=0

d
dt t=0 cp (t)

11

Now consider Vect(X) as a vector

 space and the action of Diff(M ) on Vect(M ) given by (, Y ) 7
Y = (1 ) Y . We have D1 (p) : T(p)M Tp M so that

( Y )|p (g) = D1 (p) (Y )(g) = Y(p) (g 1 )

(1)

Definition 5.12. The Lie derivative of Y Vect(M ) is


d 
LX Y =
Y.
dt t=0 t
Proposition 5.13. LX Y = [X, Y ].
Proof. Put g = f in (1):
Y |p (f ) = Y(p) (f ) =: h((p)),
i.e. set h = Y (f ). We have
LX Y |p (f ) = lim

t0

= lim
t0
|

t Y |p (f ) Yp (f )
t
t Y |p (f ) t Y |p (f t )
{z t

=:A

+ lim
t0
} |

t Y |p (f t ) Yp (f )
{z t

=:B

By linearity,
Y |p
z }| { f f
t
A = lim t Y |p
= Y |p (X(f )).
t0
t
|
{z
}
d
dt
|t=0 t (f )=X(f )

B = lim

t0

h(t (p)) h(p)

= LX h|p = Xp (h) = Xp (Y (f )).
t

So LX Y = XY Y X.
Corollary 5.14.
1. LX Y is linear with respect to both X and Y .
2. LX Y = LY X.
3. LX X = 0. If Xp = 0 and Yp = 0, then LX Y |p = 0.
Lemma 5.15.
1. Let : M N be a diffeomorphism and X Vect(M ) with flow t . Then the flow of X is
t 1 .
2. Let : M M be a diffeomorphism and X Vect(M ) with flow t . Then X = X if and
only if t = t .
Proof. exercise.
Proposition 5.16. Let X and Y be vector fields on M with flows t and s respectively. Then
[X, Y ] = 0 if and only if t s = s t for all s, t.

12

Consider = s t s t . Then the proposition says that (p) = p for all p and all s, t if
and only if [X, Y ] = 0.
flow along X for t sec

flow along Y for s sec

flow along Y for s sec

(p)
p

flow along X for t sec

Proof. If and commute then by the second point of the lemma, t Y = Y and hence [X, Y ] =
LX Y = 0.
Y ( ) Y
Suppose now that [X, Y ] = 0. So LX Y = limh0 q hh q = 0. Fix p M and set c(t) =
(t ) Y |p . Then
1
(c(t + h) c(t))
h

1
= lim
(t+h ) Y |p (t ) Y |p
h0 h

(h ) Y |p Y |p
= lim (t )
h0
h
{z
}
|

c0 (t) = lim

h0

= (t ) (0) = 0.
Thus c is constant, i.e. c(t) = c(0) for all t. So (t ) Y |p = Y |p . This is true for all p, so that
(t ) Y = Y for all t.
By the lemma this implies that t and s commute.

6 Submanifolds
Definition 6.1. A submanifold of a manifold M is a manifold N together with a smooth injection
: N M such that
1. for all p N , D|p : Tp N T(p) M is injective; and
2. (N ) has the subspace topology.
If only 1 holds, then we say that N is an immersed submanifold and an immersion. A submanifold
is sometimes called an embedded submanifold.
Example 6.2.
M
is an immersion, but not a submanifold.

13

Theorem 6.3. Let F : M N be smooth. Let c N and set Z = F 1 (c). Suppose rank DF |p =
dim N for all p Z. Then Z is a submanifold of M (or Z = ) of dimension dim Z = dim M dim N .

If Z , M is the inclusion, then D|p : Tp Z

ker( DF |p : Tp M TF (p) N ).
Theorem 6.4 (Inverse Function Theorem). Let U o RN and G : U RN smooth such that DG|p
is an isomorphism at some point p U . Then G is a local diffeomorphism around p (i.e. there exist
neighborhoods p V o U and G(p) W such that G induces a diffeomorphism G : V W ).
Proof. handout.
Proof of the submanifold theorem. Let n = dim N , dim M = n + m. Fix p Z and local coordinates
x1 , . . . , xm+n on U around p and y 1 , . . . , y n around F (p) = c. Say F = (F1 , . . . , Fn ), DF |p =


Fi 
and c = (c1 , . . . , cn ) with respect to these coordinates.
xj 
p

1in, 1jm+n

As DF |p is a surjection, it has a non-singular n n minor. Without loss of generality assume that


Fi 
A = x
is non-singular. Set
j
p

1i,jn

G : U Rn+m ,

(x1 , . . . , xn+m ) 7 (F1 (x1 , . . . , xn+m ), . . . , Fn (x1 , . . . , xn+m ), xn+1 , . . . , xn+m ).

Then DG|p = ( A0 I ) is an isomorphism, so by the inverse function theorem there are neighborhoods

p U 0 U , G(p) V 0 such that G|U 0 : U 0

V 0.
0
= U Z. Then G| has image {(c1 , . . . , cn , xn+1 , . . . , xn+m )}
Consider U
= {(xn+1 , . . . , xn+m )}
e
U
Rm . This gives a chart on Z and the transition functions are smooth since they are given by a
composition of smooth functions.
Example 6.5. F : Rn+1 R, (x0 , . . . , xn ) 7 x20 + x2n gives S n = F 1 (1). For a fixed p S n ,
P
DF |p (v0 , . . . , vn ) = 2 i pi vi . So DF |p is onto (i.e. has rank 1) for all p S n . Thus S n is a
submanifold of Rn+1 and
Tp S n = {(v0 , . . . , vn ) : DF |p (v0 , . . . , vn ) = 0} = {v : v p = 0}.
Example 6.6. The symmetric matrices Symn = {A Mn : A = AT } are a vector space (and thus a
manifold) of dimension n(n+1)
. Consider F : GLn Symn , A 7 AAT . So O(n) = F 1 (I).
2
For B O(n), the map RB : GLn GLn , A 7 AB is a diffeomorphism and F RB (A) =
ABB T AT = AAT = F (A), i.e. F RB = F . Thus DF |I = DF |B DRB |I . Hence DF |B is
surjective if and only if DF |I is surjective. From F (I + tH) = I + t(H + H T ) + O(t2 ), we see that
DF |I (H) = H + H T is surjective.
Therefore O(n) is a manifold of dimension n2 n(n+1)
= n(n1)
.
2
2
Definition 6.7. Let : N M be a smooth embedding. Two vector fields X Vect(N ) and
Y Vect(M ) are called -related, if D|p (Xp ) = Y(p) for all p N .
The condition is equivalent to Y(p) (g) = D|p (Xp )(g) = Xp (g ) for all p and g and hence to
Y (g) = X(g ).
Proposition 6.8. If Xj Vect(N ) are -related to Yj Vect(M ) for j = 1, 2, then [X1 , X2 ] is
-related to [Y1 , Y2 ].
Proof.
[Y1 , Y2 ](g) = (Y1 (Y2 (g)) Y2 (Y1 (g)))
= X1 (Y2 (g) ) X2 (Y1 (g) )
= X1 (X2 (g )) X2 (X1 (g )) = [X1 , X2 ](G ).

14

Example 6.9. A Lie subgroup of a Lie group G is a subgroup H G, that is an immersed submanifold.
Say : H G. Let h and g be the corresponding Lie algebras. Then D = : h g. We will show
that is a map of Lie algebras, so that h is a subalgebra of g.
e a : G G be left multiplication. Then L
e a = La . Fix h which
Let La : H H and L
gives us X VectL (H). We have to show that X VectL (G):

e a X = (La ) X = X .
L

Also for h, [X , X ] is related to [ X , X ] so that preserves the Lie bracket.

In fact, the converse is also true: Given a Lie subalgebra h g, the exists a Lie subgroup H G
with Lie algebra h.

7 Integral manifolds
Definition 7.1.
1. A k-dimensional distribution on M is a choice of k-dimensional subspace p Tp M for each
p M.
2. Say is smooth if for all p M there exists an open p U and X1 , . . . , Xk Vect(U ) such
that spanR { X1 |q , . . . , Xk |q } = q for all q U .
3. An immersion : N M is an integral manifold of , if Tq N = q for all q N .
Suppose that is a distribution on M and X1 , X2 Vect(M ) such that X1 |p , X2 |p p .

j 
Suppose further that there exists an integral manifold : N , M of with p N . Define X
q

j  ) = Xj | for all q N , j = 1, 2. (This is possible as (Tq N ) = q and is injective.)
by ( X
q
q

j is -related to Xj , so that [X
1 , X
2 ] is -related to
Then Xj Vect(N ) (exercise). By definition, X
[X1 , X2 ]. Thus [X1 , X2 ]p p . This motivates the following definition:
Definition 7.2. Let be a smooth distribution on M .
1. Say X Vect(M ) belongs to if Xp p for all p M .
2. Say is integral (or integrable or involutive) if whenever X, Y belong to , then [X, Y ] belongs
to .
Theorem 7.3 (Frobenius Integrability Theorem). If is integral if and only if there exists an
integral manifold of through every point of M .
Lemma 7.4. Suppose X1 , . . . , Xk Vect(U ) are linearly independent and [Xi , Xj ] = 0 for all i, j.

Let p M . Then there exist coordinates y 1 , . . . , y n around p such that Xi = y

i for i = 1, . . . , k near
p.
n
Proof. by working locally, we
 may assume M = R , p = 0. Further, by a linear change of coordinates

we may assume Xi |0 = xi 0 for i = 1, . . . , k.
(t)

Say Xi has flow i . We will construct the first k coordinates using these flows. Consider
F : Rn Rn ,

(1)

(2)

(k)

F (t1 , . . . , tn ) = t1 t2 tk (0, . . . , 0, tk+1 , . . . , tn ).

Using the inverse function theorem we want to show that F is a local diffeomorphism. For this we
show that DF |0 = I.

15

Fix a = (a1 , . . . , an ) Rn . Then






(g) =
DF
(g F )
x1 a
t1 a
g(F (a1 + h, a2 , . . . , an )) g(F (a))
= lim
h0
h
(1)

(2)

(k)

g(a1 +h a2 ak (0, . . . , 0, ak+1 , . . . , an )) g(F (a))

h0
h

= lim

(1)

g(h (F (a))) g(F (a))

h0
h
= LX1 g(F (a)) = X1 (g)|F (a) .
= lim

Thus



= X1 |F (a) ,
x1 a





=
X
|
=
X
|
=
.
DF
1 F (0)
1 0
x1 0
x1 0
DF

Since [Xi , Xj ] = 0, the flows commute, so that the same calculation works to show



DF
= Xi |F (a) ,
xi a





DF
=
X
|
=
i
0
xi 0
xi 0

(2)
(3)

for i = 1, . . . , k. For i > k we have






DF
(g F )
(g) =
xi 0
ti 0
g(F (0, . . . , 0, h, 0, . . . , 0)) g(F (0))
= lim
h0
h
g(0, . . . , 0, h, 0, . . . , 0) g(0)
= lim
h0
h

g 
=
.
xi 0
Thus DF |0 = I and by the inverse function theorem F is a local diffeomorphism. Hence y = F 1

is a coordinate system and by (2), y

i = Xi for i = 1, . . . , k.
Proof of Frobenius
Theorem.
We may assume that we are working on Rn with p = 0 and 0 =




spanR { x1  , . . . , xk  }. Consider
0

: Rn Rk ,

(x1 , . . . , xn ) 7 (x1 , . . . , xk ).

So D|0 : T0 Rk is an isomorphism. So for q near 0, D|q : q T(q) Rk is an isomorphism



and in particular injective. Thus there exist X1 , . . . , Xk such that Xj (q) = x
for q near

j
(q)

0 and j = 1, . . . , k.

By assumption [Xi , Xj ] . So [Xi , Xj ] = [ x

, ] = 0. By the lemma there exist coordinates
i xj
y 1 , . . . , y n with Xi =

16

y i

for i = 1, . . . , k. So N = {y k+1 = = y n = 0} is an integral manifold of

 near p = 0.
Remark 7.5. With a little more work one can show that there exists a unique maximal integral
manifold through p.
Example 7.6. Let G be a Lie group and h a Lie subalgebra of g. We will show that there exists a Lie
group H G with Lie group h.
For g G set g = spanR { X |g : h}. This is a smooth distribution and as h is closed under the
Lie bracket, is integrable. So by the Frobenius theorem there exists a maximal integral manifold
H through e G.
For g G note (Lg ) g0 = (Lgg0 ) e , so (Lg ) takes to itself. One can check that this implies
that Lg takes the integral manifold through e to that through g. If h H, then Lh1 takes H to the
unique maximal integral manifold through Lh1 (h) = e, i.e. Lh1 takes H to H so that H is closed
under multiplication. So it is a subgroup. One can show that multiplication is smooth on H.

8 Vector bundles
The idea of a vector bundle is to give a vector space on each point of M , varying smoothly.
Definition 8.1. A (smooth) vector bundle E of rank r on a manifold M consists of a smooth
manifold E and a smooth map : E M such that
1. for all p M , Ep = 1 (p) is an r-dimensional real vector space;
2. for all p M there exists an open neighborhood U and a diffeomorphism tU : E|U U Rn
(where E|U = 1 (U )) such that
a) tU |Eq : Eq {q} Rn is a linear isomorphism and
b) The diagram
E|U
|U

tU

U Rr
p

commutes.
E is called the total space, M the base space, Ep is the fiber of E over p and tU is a trivialization of
E over U . If r = 1, then E is called a line bundle.
1
n
Example
8.2. The

 tangent bundle T M is a vector bundle: Given a coordinate chart x , . . . , x , then


x1 q , . . . , xn q are a basis for Tq M . Set

tU : T M |U = T U U Rn ,



n 
where vq = a1 x
+

a
Tq M .
1
n
x
q
q

vq 7 (q, (a1 , . . . , an )),

Example 8.3. E = M Rn is called the trivial bundle.

Suppose that tU and tV are trivializations of E and consider
r
r
tU t1
V : (U V ) R (U V ) R ,

(p, w) 7 (p, w0 ).

The map w 7 w0 is a linear isomorphism, i.e. tU t1

V (p, w) = (p, U V (p)(w)) with U V : U V
GLr (R). The map U V is called the transition function.
Remark 8.4. tU t1
V is smooth if and only if U V is smooth.
Observe:

17

1. U U = I (as tU t1
U = id);
1 1
2. U V = 1
= tV t1
V U (as (tU tV )
U );
1
1
3. U V V W = U W (as tU t1
V tV tW = tU tW ).

Conditions 13 are called the cocycle condition. Given an open cover {U } and functions { : U
U GLr (R)}, call { } a cocycle if = I, = 1
and = .

8.1 Vector bundle constructions

Proposition
8.5. Suppose for every point p M we have a vector space Ep of dimension r. Set
`
E = pM Ep and : E M sending vp Ep to p. Suppose further that {U } is an open cover by
charts and that for all we have a bijective map tU : E|U U Rr such that
1. The diagram
tU

E|U
|U

U Rr
p

commutes;
2. tU : Eq
= {q} Rn for all q U ; and
3. tU t1
U (p, v) = (p, (p)v) with smooth.
Then there exists a unique manifold structure on E making it into a vector bundle over M with a
cocycle { }.
Proof. Suppose our charts are (x , U ). Charts on E are given by
tU
x
E|U
U Rr
x (U ) Rr
= Rn+r .

Exercise: The transition functions are smooth. The induced topology is Hausdorff and second
countable.
`
Example 8.6. Let E and F be vector bundles on M . Define EF as follows: Set EF = pM Ep Fp .
We can choose an open cover of M by U such that there are trivialization tU and sU of E and
0
F . Consider E F |U U Rr+r given by these functions. One can check that if E has
transition functions
and F has transition functions , then E F has transition functions

0
=
0 , which is smooth.
Example 8.7. Let M = RPn . Define a line bundle L on M by
L[v] = line in Rn+1 through v = {v : R}.
L is called the Hopf line bundle or the tautological bundle. It is indeed a vector bundle:
a
L=
L[v] .
[v]RPn

Consider the charts Ui = {vi 6= 0} RPn , Ui

= Rn , and set
tUi : L|Ui Ui R,

18

(v0 , . . . , vn ) 7 ([v0 : : vn ], vi ) .

This is linear on L[v] , so gives a trivialization. We have

tUi t1
Uj : (Ui Uj ) R (Ui Uj ) R,
So Ui Uj =

vi
vj

([v], ) 7 ([v],

vi
).
vj

GLn (R) = R is smooth.

`
Example 8.8 (Dual bundle). Suppose E M is a vector bundle of rank r and set E = pM Ep .
To make E into a bundle, take a trivialization tU : E|U U Rr , say tU (vp ) = (p, Fp (vp )) with

Fp : Ep Rr linear. Then Fp1 : Ep (Rr )

= Rr . So consider
e
tU : E |U U Rr ,

Ep 7 p, Fp1 (p ) .

If E has smooth transition functions U V : U V GLr , then E has transition functions 1

UV
which are also smooth. So E is a vector bundle, called the dual of E.

T

Definition 8.9. The cotangent bundle to M is T M = (T M ) .

Remark 8.10. Say we have coordinates x1 , . . . , xn on U . There are maps


U T M |U ,
p 7
.
xi p
There are also maps
U T M |U ,


p 7 dxi p .

i

Both of them are smooth (exercise). Recall: {dxi } are dual to { x
i }, i.e. dx p
xj p = ij .

Definition 8.11. Given a bundle E

M , a section over U o M is a smooth map s : U E
such that s = idU . The space of sections of E over U is denoted C (U, E) or (U, E) and is a
C (M )-module.
Example 8.12. C (M, T M ) = Vect(M ) and C (M, M R) = C (M ).
Definition 8.13. 1 (M ) = C (M, T M ) are the 1-forms of M .
Remark 8.14. Observe the pairing
1 (M ) Vect(M ) C (M ),

(, X) 7 (p 7 p (Xp )) .

If f C (M ), then Df |p : Tp M R is an element of Tp M . So we get df 1 (M ), defined by

P f i
df |p (Xp ) = X(f )|p . In coordinates this is df =
xi dx .
Definition 8.15. Let : E M and 0 : E 0 M 0 be vector bundles. A bundle map from E to E 0
consists of a pair (F, f ), where f : M M 0 and F : E E 0 are smooth such that
1. The diagram
F

E0

M0

commutes.
2. F : Ep Ef0 (p) is linear for all p M .

19

Example 8.16. If f : M M 0 is smooth, then Df : T M T M 0 forms a bundle map.

Definition 8.17. Given E M and E 0 M , a bundle map over M is a bundle map (F, idM ). A
bundle isomorphism is a bundle map with an inverse that is also a bundle map.
Remark 8.18. Given a bundle map F : E E 0 over M and U o M we get an induced map
: C (M, E) C (M, E 0 ),

s 7 F s,

i.e. (s)(p) = F (s(p)). The map is C (U )-linear: For g C (U ) we have

(gs)(p) = F ((gs)(p)) = F (g(p)s(p)) = g(p)F (s(p)),
i.e. (gs) = g(s).
Proposition 8.19. Suppose that : C (M, E) C (M, E 0 ) is C (M )-linear. Then there exists
a unique bundle map (over M ) F : E E 0 such that (s) = F s.
Proof. Example sheet.

9 Tensors and exterior algebra

Definition 9.1. Let V and W be vector spaces. The tensor product of V and W is a vector space
V W and a bilinear map : V W V W such that if : V W U is a bilinear map to
some vector space U , then there exists a unique linear map
: V W U making the diagram

V W

V W

U
!

commute.

Assume that the spaces are finite dimensional. For U = R the diagram says that (V W ) =
Bilinear(V W, R). Thus we can set V W = Bilinear(V W, R) , which shows existence of the
tensor product.
For v V and w W we write v w for (v, w). Note that is in general not surjective!
Proposition 9.2. The tensor product is functorial, i.e. if f : V V 0 and g : W W 0 are linear,
then there exists a unique linear map f g : V W V 0 W 0 such that
V W

V W

f g

f g

V 0 W0
V 0 W0

commutes. The following isomorphisms hold and are natural, i.e. compatible with the diagram above:
1. V W
=W V;
2. V (U W )
= (V U ) W ;

3. (V W )
= V W .

20

Lemma 9.3. If {v1 , . . . , vn } and {w1 , . . . , wm } are bases for V and W , then
{vi wj : i = 1, . . . , n, j = 1, . . . , m}
is a basis for V W .
Remark 9.4. Given V W , say =

: V W R,

i,j

i j , we can think of as a bilinear map

(v, w) =

i (v)j (w).

i,j

Remark 9.5. Using the vector bundle construction, if E M and F M are vector bundles, we
have a bundle E F whose fibers are (E F )p = Ep Fp (exercise).
k

Definition 9.6. Set T (k,l) M = (T M )

are called mixed tensors.

(T M )l . The elements of T (k,l) (U ) = C (U, T (k,l) U )

Example 9.7. T (1,0) (U ) = Vect(U ) and T (0,1) (U ) = 1 (U ).

Remark 9.8. Given coordinates x1 , . . . , xn on U and T (k,l) (U ), we can write
=

k
ji11,...,i
,...,jl

i dxj1 dxjl
i
1
x
x k

i
with ji11,...,i
,...,jl C (U ). There is a change of coordinates formula similar to that for xi and dx .
Remark 9.9. V W
= Hom(W, V ) via : V W Hom(W, V ) given by (v )(w) = (w)v
and linear extension.

Remark 9.10. Say V W W U . Then we can form the contraction of , C() V U by

C(v w u) = (w)v u and linear extension.
Example 9.11. Let T (k,l) (U ) be written in local coordinates as above. Then we can form the
a-b-contraction
!

a
i
j1
j
jb

l
j
b dx
Cb () =
j

dx

dx

dx
.
xi1
xia
xik
xia
|
{z
}
=ia ,jb

Definition 9.12. Set T (V ) =

k0

V k (with V 0 = R). The maps

V k V l V (k+l) ,

(v1 , v2 ) 7 v1 v2

make T (V ) into an algebra, called the tensor algebra of V .

Definition 9.13. Let I be the two-sided ideal of T (V ) generated by elements v v, v V . The
exterior algebra of V is V = T (V )/I. Let : T (V ) V be the projection. Then r V = (V r )
is the r-th exterior power of V .
Notation. The multiplication on T (V ) gives a multiplication on V , denoted .
Remark 9.14. For v1 , v2 V : v1 v2 = v2 v1 .
The idea is to consider r V as the space Altr (V ) of alternating forms V V R. Say
1 r r V for i V . Then
X
(1 r )(v1 , . . . , vr ) =
()(1) (v1 ) (r) (vr ),
Sr

where () is the sign of the permutation .

21

Lemma 9.15.
1. If p V , q V , then = (1)

pq

2. dim 0 V = 1; 1 V = V , dim n V = 1, where n = dim V .

3. If v1 , . . . , vn is a basis for V , then B = {vi1 vir : i1 < < ir } is a basis for r V .
Proof.
1. If v V , then v v = 0, so for v1 , v2 V : 0 = (v1 + v2 ) (v1 + v2 ) = v1 v2 + v2 v1 . So
v1 v2 = v2 v1 .
Hence swapping two adjacent elements in v1 vr changes the sign. Thus
v1 vp w1 wq = (1)pq w1 wq v1 vp
and clearly any element p V (resp. q V ) is a linear sum of vectors of the form
v1 vp (resp. w1 wq ).
2. 0 V = (V 0 ) = R, 1 V = (V ) = V .
Let v1 , . . . , vn be a basis of V . Then n V is clearly spanned by v1 vn , so dim n V 1.
As the determinant is a nontrivial element of Altn (V ), we have dim n V 1.
P
3. Clearly B spans r V . To show linear independence, suppose I aI vI = 0 (where I runs over
all (i1 < < ir ) and vI = vi1 vir ). Fix I 0 and consider J = I 0c . Then
!
X
X
0=
aI vI vJ =
aI vI vJ = aI 0 .
I

So aI 0 = 0, proving linear independence.

Remark 9.16. Given a linear map f : V W , we get a map f : T (V ) T (W ), which descends to

give a map r V r W , called r f or just f . It is given by
r f (v1 vr ) = f (v1 ) f (vr )
and linear extension.
Lemma 9.17. If dim V = dim W = n, then n f is multiplication by det f , i.e. if v1 , . . . , vn is a
basis for V and w1 , . . . , wn one for W , then
f (v1 vn ) = det(f )w1 wn .
Proof. Say f (vi ) =

Aji wj . Then

f (v1 ) f (vn ) =
=

Aj1 wj
Ajn wj

A(1)1 w(1) A(n)n w(n)

Sn

()A(1)1 A(n)n w1 wn .

Sn

22

{z

=det(f )

Remark 9.18. Given a vector bundle E M , we have vector bundles r E such that (r E)p = r Ep
for all p M .
Definition 9.19. The bundle of p-forms on M is p T M . A section C (M, p T M ) is called
a p-form on M . p (U ) = C (U, p T M ) is the space of p-forms on U .
Remark 9.20. 0 (U ) = C (U ), 1 (U ) = C (U, T M ).
If we have coordinates x1 , . . . , xn on U , then p (U ) can be written uniquely as
X
=
ai1 ,...,ir dxi1 dxir
i1 <<ir

for ai1 ,...,ir in C (U ).

Definition 9.21. Let f : M N be a smooth map. The pullback map f : p (N ) p (M ) is
given by
f |x = p ( Df |x ) ( |f (x) )
for x M , p (N ).
Example 9.22. f is smooth.
Remark 9.23. For x M :
Df |x : Tx M Tf (x) N
( Df |x ) : Tf(x) N Tx M
p ( Df |x ) : p Tf(x) N p Tx M
Proposition 9.24.
1. f (1 + 2 ) = f (1 ) + f (2 );
2. f (1 2 ) = f (1 ) f (2 );
3. (g f ) = f g .
Proof. not given

10 Differential forms
Remark 10.1. If q (U ) and x1 , . . . , xn are coordinates on U , then
X
ai1 ,...,iq dxi1 dxiq ,
=
ai1 ,...,iq C (U ).
i1 <<iq

If y 1 , . . . , y n are another set of coordinates on U , then dxik =

=

ai1 ,...,iq

i1 <<iq 1j1 ,...,jq n

xik
j=1 y j

Pn

dy j , so

xi1
xiq j1

dy dy jq .
j
y 1
y jq

Remark 10.2. If f C (M ) = 0 (M ), then df 1 (M ). Locally: df =

f
1
x1 dx

+ +

f
n
xn dx .

Theorem 10.3. There exist unique natural linear maps

d : q (M ) q+1 (M )
(q 0), called the exterior derivative, such that

23

1. if f 0 (M ), then df is the derivative of f as defined previously;

2. d2 = 0;
3. if p (M ) and q (M ), then d( ) = d + (1)p d.
P
Proof. Say = i1 <<iq ai1 ,...,iq dxi1 dxiq locally. Set
d =

dai1 ,...,iq dxi1 dxiq =

n
X
ai1 ,...,iq

i1 <<iq j=1

i1 <<iq

Then
d2 =

n
X
ai1 ,...,iq

i1 <<iq j=1

xj xk

xj

dxj dxi1 dxiq .

dxj dxk dxi1 dxiq = 0

as the partial derivatives commute. Further suppose = f dxI , = gdxJ for some multiindices I, J
with |I| = p, |J| = q. Then
d( ) = d(f gdxI dxJ ) = d(f g) dxI dxJ
= gdf dxI dxJ + f d(g) dxI dxJ = d + (1)p d.
So linearity gives 3.
P
For uniqueness and well-defined: Say = i1 <<iq bi1 ,...,iq dy i1 dy iq for another set of
i1
iq
e =P
coordinates. Set d
i1 <<iq dbi1 ,...,iq dy dy . Then
3

d =

e
dbi1 ,...,iq dy i1 dy iq + bi1 ,...,iq d(dy i1 dy iq ) = d.
{z
}
|

i1 <<iq

=0 by 2 and 3

Definition 10.4. Let q (M ) and let X Vect(M ) with flow t . The Lie derivative of is

d 
.
LX =
dt t=0 t
Here t : M M and hence t : q (M ) q (M ).
Proposition 10.5.
1. LX (1 + 2 ) = LX 1 + LX 2 ;
2. LX (f ) = X(f ) + f LX ();
3. LX ((Y )) = (LX )(Y ) + (LX Y ) for 1 (M ) and Y Vect(M ).
Proof. exercise.

11 De Rham cohomology
Definition 11.1. The q-th de Rham cohomology group of M is
q
HdR
(M ) =

Remarks 11.2.

24

ker(d : q q+1 )
,
im(d : q1 q )

q 0.

q
1. HdR
(M ) is an R-vector space.
0
(M ) = ker(d : C (M ) 1 (M )) is the space of locally constant functions. If M is
2. HdR
0
connected, then HdR
(M ) = R.

3. If d = 0, the is called closed.

4. If = d, the is called exact.
p
q
p+q
(M ) and [] HdR
(M ), define [] [] = [ ] HdR
(M ). (This is well-defined,
5. If [] HdR
p
0
since d( ) = d + (1) d = 0 and ( + d ) = + d(0 ) by adding
(1)p1 0 d = 0.)

6. Functoriality: Suppose F : M N is smooth. Then d F = F d, so if d = 0, then

dF = 0. Hence we can define
q
q
F : HdR
(N ) HdR
(M ),

[] 7 [F ].

q
q
Theorem 11.3 (de Rham). There exists a natural isomorphism HdR
(M )
= Hsing (M ).

Proof. sketch later.

Definition 11.4. Let F and G be smooth maps M N . Then F and G are smoothly homotopic
if there is a smooth map H : M [0, 1] N such that H0 = F and H1 = G (where we write
Ht (x) = H(x, t)).
q
q
(N ) HdR
(M ) for all
Theorem 11.5. If F and G are smoothly homotopic, then F = G : HdR
q 0.

Proof. Let p (N ) with d = 0. Consider H on M [0, 1]. We can write

H = + dt ,
where t is the coordinate on [0, 1], = (t) p (M ) and p1 (M ).
We will first show that (t) = Ht : Fix t0 and consider the embedding : M {t0 } , M [0, 1].
Then
Ht0 = (H ) = H = ( + dt ) = + |{z}
dt .
=d t=dt0 =0

As d = 0,
0 = H d = dH = dM [0,1] ( + dt ) = dM +
Hence

dt dt d.
t

= d and

G F = H1 H0 = (1) (0) =

Z
0

dt =
t

Z
ddt = d

dt =: d.
0

Therefore [F ] = [G ], i.e. F = G .
Example 11.6. Consider M = N = Rn and Ht (x) = tx, i.e. H1 = id and H0 0. Then H0 =
p
p
p
H1 : HdR
(Rn ) HdR
(Rn ). But for p > 0, H0 = 0 and H1 = id. Hence HdR
(Rn ) = 0 for p > 0.
p
Corollary 11.7 (Poincare Lemma). If M is a star-shaped domain, then HdR
(M ) = 0 for all p > 0,
p
p1
i.e. if (M ) with d = 0, then there is a (M ) with = d.

25

12 Integration
Definition 12.1. An orientation on a manifold M is an equivalence class of charts (U , x ) such
that J(x x1
) > 0 for all , . A manifold M is orientable if an orientation exists. An oriented
manifold is a manifold with a chosen orientation.
Example 12.2. Rn and S n are orientable. Note that
Rn Rn : (x1 , . . . , xn ) 7 (x1 , . . . , xn )
Rn Rn : (x1 , . . . , xn ) 7 (x1 , . . . , xn )
give two orientations on Rn .
The M
obius band is not orientable.
Definition 12.3. Let be a section of n T M (not necessarily smooth) and M oriented. If supp()
is compact and contained in some chart U , = f dx1 . . . dxn (f not necessarily smooth), define
Z
Z
= f (x1 , . . . , xn ) dx1 dxn ,
U

x(U )

if the latter is defined.

(This is well-defined as supp f (x1 , . . . , xn ) is compact and so the change of variable formula for
can be used and works because of the fixed orientation.)

Definition 12.4. Let {U } be an open cover of M . A collection of i C (M ) is called a partition

of unity (p.o.u.) subordinate to {U } if
1. For all i there exists (i) such that supp i U(i) ;
2. the supp i are locally finite, i.e. for all p M , j (p) = 0 for all but finitely many j; and
P
3.
i i 1 and each i 0.
Theorem 12.5. Partitions of unity exist.
Proof. Handout (only examinable if M is compact).
Definition 12.6. Let be an n-form on M (dim M = n, not necessarily smooth). Cover M by
charts (U ) and pick a subordinate partition of unity such that each i has compact support. Define:
Z
X Z
=
i .
i U
(i)

Theorem 12.7 (Stokes). If M is compact and n1 (M ), then

Z
d = 0.
M

Proof. See the more general version below.

Proposition 12.8. A manifold M is orientable of and only if there exists n (M ) with p =
6 0
for all p M .

26

Proof. First suppose that M is orientable. Given any chart (U, x) consider U = dx1 dxn on
U . If y 1 , . . . , y n is another chart of the same orientation, then U = f dy 1 dy n with f positive.
Cover M by oriented charts (U , x ) and choose a partition of unity i subordinate to this cover. Set
X
=
i U(i) ,
i

which is positive everywhere.

Conversely let n (M ) be nowhere vanishing. For a connected chart (U, x) we have =
f dx1 dxn on U with f nowhere vanishing. Thus f > 0 or f < 0. In the first case, take
(U, (x1 , . . . , xn )) as an oriented charts, otherwise take (U, (x1 , . . . , xn1 , xn )).
n
Corollary 12.9. If M is compact and orientable, then HdR
(M ) 6= 0.

Proof. Let n (M ) be nowhere vanishing. Without loss of generality > 0. Clearly d = 0. If

= d , then
Z
Z
0 < = d = 0
M
n
by Stokes theorem. This is absurd. Hence [] 6= 0 in HdR
(M ).

Definition 12.10. Let H+ = {(x1 , . . . , xn ) Rn : xn 0}. A chart-with-boundary on M is a

bijection x : U V H+ , U M , V H+ open. Two charts are compatible if x y 1 is smooth.
A differentiable structure is an equivalence class as before. This gives a manifold-with-boundary. Set
M = {p M : xn (p) = 0 for some chart (x1 , . . . , xn ) around p}.

Example 12.11. M = B1 (0) Rn , M = S n1 .

Fact 12.12. If M is a manifold-with-boundary of dimension n, then M is a manifold of dimension
n 1 and M = M \ M is a manifold of dimension n.
Definition 12.13. Suppose M is oriented, with the orientation given by some nowhere vanishing
n (M ). Then we get the induced orientation on M as follows: Say x1 , . . . , xn is a chart such
that M = {xn = 0}. If = f dx1 dxn , we set f dx1 dxn1 to be the (n 1)-form on
M defining the induced orientation.
Example 12.14. Open cylinder: M = S 1 q S 1 :

Theorem 12.15 (Stokes). If n1 (M ), then

Z
Z
d = .
M

Sketch of proof. By linearity we may assume that supp( ) is in some single chart and has compact
support. So this reduces to Stokes theorem in Rn .

27

Facts 12.16.
R
R
1. M = M .
2. Suppose M = U V , where U and V are open submanifolds with U V = and has
measure 0 in M . Then
Z
Z
Z
= + .

3. If M Rn is a closed subset that is a manifold-with-boundary, n (Rn ), = f dx1 dxn ,

then
Z
Z
= f dx1 dxn ,
M

where the second integral denotes the usual integral for subsets of Rn .

13 Connections
13.1 Connections as derivations
Let E be a vector bundle on M . Set p (E) = C (M, p T M E), the E-valued p-forms.
Definition 13.1. A connection on E is a linear map : 0 (E) 1 (E) such that for all f C (M )
and all s 0 (E) = C (E),
(f s) = df s + f s.
This gives a way to differentiate sections of E along vector fields: For X Vect(M ) set
X (s) = (s, X) := C(s X),
where the contraction is between the 1-form part of s and the vector field. So X (f s) = (df
s, X) + (f s, X) and hence we have
X (f s) = X(f )s + f X s
gX (s) = gX (s)
X1 +X2 (s) = X1 s + X2 s.
Example 13.2. Consider U o Rn and E = U Rm . Define sections s1 , . . . , sm by si (x) =
(x, (0, . . . , 1, . . . , 0)) and define by requiring si = 0 for all i. For any section s C (E)
P
write s =
ai si uniquely for some ai C (M ). Then
X
s =
dai si + ai si .
| {z }
=0

P
X(ai )si . We write s = (a1 , . . . , an ) and X (s) = (X(a1 ), . . . , X(am )).
If X Vect(U ), then X (s) =
Exercise 13.3. Show that every vector bundle admits at least one connection.
Lemma 13.4. If 1 and 2 are connections on E, then 1 2 1 (M, EE ) = 1 (Hom(E, E)).
Proof. For any s C (E) and g C (M ):
(1 2 )(gs) = 1 (gs) 2 (gs) = dg s + g1 s dg s g2 s = g(1 2 )s.
So 1 2 : C (E) C (T M E) is C (M )-linear. Hence it is induced by a bundle map
E T M E, i.e. given by an element of 1 (M, E E ).

28

Remark 13.5. Given a connection on E, we get an induced map

d : p (E) p+1 (E),

d ( s) = d s + (1)p s

for p (M ) and s C (E), called the covariant derivative. One checks that for q (E),
d ( ) = d + (1)p d .

13.2 Coordinate description of connections

Definition 13.6. Sections s1 , . . . , sm 0 (U, E) are a frame for E if for all p U , s1 (p), . . . , sm (p)
is a basis of Ep .
Suppose we P
have a frames e1 , . . . , em for E and 1 , . . . , l for p T M on U . Given s p (E), we
can write s = i,j fji j ei with fji C (U ).
P
Exercise 13.7. The quantities i = j fji j are independent of the chosen frame.
P
Hence we can write s = i i ei with i p (U ), ei 0 (U, E) on U .
P
Definition 13.8. Given a connection , write ej = i ij ei for some ij 1 (U ). The matrix
= (ij )i,j of 1-forms is called the connection matrix of with respect to the frame e1 , . . . , em .
Lemma 13.9. Let s 0 (U, E), s =
e = (e1 , . . . , em )T . Then

si ei . Write s = (s1 , . . . , sm )T , ds = (ds1 , . . . , dsm )T and

s = (ds + s) e.
Proof.
s =

(si ei )

(linearity)

dsi ei + si ei

(Leibnitz)

!
=

ds ei + s

ji ej

!
=

X
j

dsj +

si ji

ej

Proposition 13.10. Suppose that e1 , . . . , em and e01 , . . . , e0m are frames of E over U . Let be a
connection that has corresponding connection matrices and 0 . Let : U GLm (R) such that
e0 = e. Then
0 = (d) 1 + 1 .
Proof.
e0 = (e) = (d)e + e
= (d)e + (e)
= (d) 1 e0 + 1 e0 .

29

Remarks 13.11.
1. If = d + , then d : p (U, E) p+1 (U, E) can also be written as d = d + , i.e. d s =
ds + s for s p (U, E).
2. For a given frame e1 , . . . , em for E over U and coordinates x1 , . . . , xn on U , we can write
P
ij = k ijk dxk for some ijk C (U ) (1 k n, 1 i, j m) called Christoffel symbols of
.

13.3 Connections and parallel transport

Definition 13.12. A smooth function s : [0, 1] E is a section of E along a curve c : [0, 1] M , if
s(t) Ec(t) for all t. The space of all sections along c is denoted C (c, E).
If s 0 (E), then se(t) = s(c(t)) is a section along c. We will also write s(t) for this section. Note
that not all elements of C (c, E) look like this.
Example 13.13. c(t)
C (c, T M ), but if c has a node, then this cannot be the restriction of any
s 0 (T M ).
Proposition 13.14. There exists a unique covariant derivative map
D
: C (c, E) C (c, E),
dt
depending on c, such that
1.

D
dt

is linear;

2. if f : [0, 1] R is smooth and s C (c, E), then

3. if s 0 (E), then

D
dt (s(t))

D
dt (f s)

df
dt s

+ f Ds
dt .

= c(t)
s.

Proof. Work locally, say with

and coordinates x1 , . . . , xn on U . If s
Pa iframe e1 , . . . , em of E
D

i
C (c, E), we can write s =
exists, we
s (t) ei |c(t) for some s C ([0, 1]). Assuming that dt
calculate

D 1. X D i
s=
s (t) ei |c(t)
dt
dt
i

X dsi
2.
iD
ei |c(t) + s (ei c)
=
dt
dt
i

X dsi
3.
i
=
ei |c(t) + s c(t)
ei |c(t) .

dt
i
Let c = (c1 , . . . , cn ). Then c =

cj


xj c(t)

X


c(t)
cj
ei c(t) =
j

Hence
X
D
(s) =
dt
k

determines

30

D
dt

uniquely.

So,

xj



ei 

=
c(t)

X
j,k



cj kij 

c(t)

ek |c(t) .

!
dsk X i j k
+
s c (ij c) ek |c(t)
dt
i,j

(4)

Definition 13.15. We say s C (c, E) is parallel, if Ds

dt = 0 for all t [0, 1].
By (4), this condition is a system of linear ODEs for si . So if v Ec(0) is an initial condition, the
theory of ODEs says that there is a unique solution s C (c, E) with Ds
dt 0 and s(0) = v. This s
is called the parallel transport of v along c.
If we set a = c(0) and b = c(1), we have linear isomorphisms
c
Pa,b
: Ea Eb ,

c
Pa,b
(v) = s(1),

where s C (c, E) is a parallel transport of v.

c
, we can recover :
Given Pa,b

= v Tp M and write Pt =
Proposition 13.16. Let c : (, ) M with c(0) = p, c(0)

c
Pp,c(t)
: Ep
Ec(t) . Then, for s 0 (E),
v s = lim

Pt1 ( s|c(t) ) s|p

t

t0

Proof. Let e1 , . . . , em be a basis for Ep . Define eei (t) = Pt (ei ) Since Pt is an isomorphism
for all
P i
t, ee1 , . . . , eem are a frame along c (called a parallel frame). Given s, we write s =
ei (t) and
s (t)e
calculate
Pt1 ( s|c(t) ) s|p
t

P
=

si (t)Pt1 (e
ei (t)) si (0)e
ei (0)
t
X
X si (t) si (0)
t0
ei
s i (0)ei = c(0)
s = v s.
=

t
i
i

13.4 Horizontal splitting

See notes.

14 Curvature
Definition 14.1. The curvature of a connection is the map
R = d d : 0 (E) 2 (E).
Lemma 14.2. R is C (M )-linear.
Proof. Let f C (M ), s 0 (E).
R(f s) = d d (f s) = d (df s + f d s) = d2 f s df d s + df d s + f d d s = f R(s).
|{z}
=0

Remark 14.3. So R induces a bundle morphism E E 2 T M , and thus a section

R 0 (E E 2 T M ) = 2 (E E ) = 2 (End(E)).
We can write this as R(s) = R s 2 (E), where we contract the endomorphism part with s.

31

Remark 14.4. In coordinates: Let e1 , . . . , em be a frame for E over U and let have connection
P
matrix = (ij )i,j (ij 1 (U )) defined by ej = i ij ei . Then the curvature matrix of with
respect to this frame is defined by
X
R(ej ) =
ij ei ,
ij 2 (M ).
i

Lemma 14.5. = d + .
Proof.
!
X

R(ej ) = d d (ej ) = d

ij

ei

dij ej ij d (ei )

!
dij

ej

ij

ki

ek

!
=

dkj

ij

ki

ek .

Proposition 14.6 (Bianchi identity). d = .

Proof.
d = d(d + ) = |{z}
d2 +d d
=0

= ( ) ( ) = .
Definition 14.7. For X, Y Vect(M ), define
R(X, Y ) : 0 (E) 0 (E),

s 7 C(R(s) X Y ),

where the 2-form part of R(s) is contracted with X and Y .

Lemma 14.8. R(X, Y ) = X Y Y X [X,Y ] .
Proof. We will show that given 1 (E),
(d )(X, Y ) = X ((Y )) Y ((X)) ([X, Y ]).
Then with = d (s) (thus (Y ) = Y (s)):
R(X, Y )(s) = d d (s)(X, Y ) = d ()(X, Y ) = X Y s Y X s [X,Y ] s.
For the claim, we may assume by linearity that = s with 1 (M ), s 0 (E). Then
d ()(X, Y ) = d ( s)(X, Y ) = (d s s)(X, Y ) = d(X, Y ) s (X)Y s + (Y )X s.
On example sheet 2: d(X, Y ) = X(Y ) Y (X) ([X, Y ]). Thus
d ()(X, Y ) = X((Y ))s Y ((X))s ([X, Y ])s (X)Y s + (Y )X s.

32

Also
X ((Y )) = X (( s)(Y ))
= X ((Y )s) + (((Y )s), X)
= (d((Y )) s + (Y )s, X)
= X((Y ))s + (Y )X s
Y ((X)) = Y ((X))s + (X)Y s
([X, Y ]) = ([X, Y ])s.
Example 14.9. Let M = U , P
E = U Rm with frame e1 , . . P
. , em . Define a connection on E by requiring
ei = 0 for all i. So if s = j sj ej 0 (E), then s = j dsj ej . Further, if X Vect(M ), then
P
X (s) = j X(sj )ej . So clearly (X Y Y X [X,Y ] )(s) = 0.
Hence we can interpret R as a measure to which extent a vector bundle E is trivial.
Remark 14.10. Let e1 , . . . , em and e01 , . . . , e0m be frames on U with connection and curvature matrixes
, and 0 , 0 . Let e0 = e for some : U GLm . Then 0 = (d) 1 + 1 and
R0 (e0 ) = R(e) = R(e) = ( e) = 1 e0 .
Thus
0 = 1 .
In particular, tr() 2 (E) is independent of the chosen frame.
Facts 14.11.
2
1. d tr() = 0. Set c1 := [tr()] HdR
(M ).

2. c1 is independent of .

15 Linear connections
Definition 15.1. A connection on T M is called a linear connection or a Koszul connection or a
connection on M .
Remark 15.2. Thus a linear connection is a map Vect(M ) Vect(M ) Vect(M ) sending (X, Y ) to
X Y . Remember that we have f X (Y ) = f X (Y ) and X (f Y ) = X(f )Y + f Y X.
Definition 15.3. A Riemannian metric on M is a g C (T M T M ) such that
1. Symmetry: g(Xp , Yp ) = g(Yp , Xp ) for all Xp , Yp Tp M .
2. Non-degeneracy: If g(Xp , Yp ) = 0 for all Xp , then Yp = 0.
3. Positive definite: g(Xp , Xp ) 0 for all Xp .
1

Write hX, Y i = g(X, Y ). Given Xp Tp M , set |Xp | = hXp , Xp i 2 . Given c : [0, 1] M , set
Z

|c(t)|dt.

length(c) =
0

Given a, b M , set
d(a, b) = inf{length(c) : c : [0, 1] M with c(0) = a, c(1) = b}.
Exercise 15.4. If M is connected, d gives a metric.

33

We want to show how g gives rise to a unique connection on M with nice properties.
n
n
Example 15.5. Let M = Rn with global coordinates x1 , . . . , xn . Then
P Ti Mi = R R with
P global

frame x1 , . . . , xn . Define a Riemannian metric by g(X, Y ) = i X Y , where X =

X i x
i,
P i
P

Y xi . We also have a linear connection given by xi = 0 for all i, i.e. X Y =

X(Y i ) x
Y =
i.
We observe
X hY, Zi = hX Y, Zi + hY, X Zi
for all X, Y, Z,

where we set X (f ) = X(f ).

Definition 15.6. A linear connection is compatible with a Riemannian metric g, if
X hY, Zi = hX Y, Zi + hY, X Zi

for all X, Y, Z Vect(M ).

Remark 15.7. Observe on Rn : X Y Y X = [X, Y ].

Definition 15.8. Given a linear connection on M , set the torsion of to be
(X, Y ) = X Y Y X [X, Y ].
Say is symmetric (or torsion-free) if (X, Y ) = 0 for all X, Y Vect(M ).
Theorem 15.9 (Fundamental theorem of Riemannian geometry). Let M be a manifold with Riemannian metric g. Then there exists a unique linear connection that is symmetric and compatible
with g. It is called Levi-Civita connection.
Proof. First show uniqueness: Suppose is symmetric and compatible with g:
X(hY, Zi) = hX Y, Zi + hY, X Zi
= hX Y, Zi + hY, Z Xi + hY, [X, Z]i

(5)

Y (hZ, Xi) = hY Z, Xi + hZ, X Y i + hZ, [Y, X]i

(6)

Z(hX, Zi) = hZ X, Y i + hX, Y Zi + hX, [Z, Y ]i

(7)

(5) + (6) (7):

X(hY, Zi) + Y (hZ, Xi) Z(hX, Y i) = 2 hX Y, Zi + hY, [X, Z]i + hZ, [Y, X]i hX, [Z, Y ]i .
Rearranging:
hX Y, Zi =

1
(X(hY, Zi) + Y (hZ, Xi) Z(hX, Y i) hY, [X, Z]i hZ, [Y, X]i + hX, [Z, Y ]i) . (8)
2

The right hand side is independent of , hence it defines X Y uniquely in terms of the metric.
For existence use (8) to define X Y and check that it is a linear connection with the desired
properties. (See handout.)
Example 15.10. Suppose M is embedded into RN via such that D preserves the metric, i.e.
hD(Xp ), D(Yp )iRN = hXp , Yp ig

for all p M and Xp , Yp Tp M.

and set : Tp RN Tp M to be the orthogonal projection

Denote the standard connection on RN by
N
with respect to the metric on R . Define on M by
Y

X Y := (
e ),
X

Y
Vect(RN ) extending X, Y.
X, Y Vect(M ), X,

Exercise: This gives a well-defined connection that is symmetric and compatible with g.

34

Definition 15.11. Let E M , E 0 M be two vector bundles with connections and 0

respectively. Get induced connections

s0 ) = s s0 + s 0 s0 ;
1. on E E 0 by (s
s0 ) = (s, 0 s0 );
2. on E E 0 by (s,
3. on E : for C (E ): ()(s) = d((s)) (s), s C (E).
Exercise 15.12. Check that these definitions indeed give connections. For the connection on the dual
bundle:
(f )(s) = d(f (s)) f (s)
= (df )(s) + f d((s)) f (s)
= (df )(s) + f (s)

s C (E).

Thus (f ) = df + f .
So given a connection on T M , we get induced connections on T (k,l) M = (T M )k (T M )l with
the convention that the induced connection on T (0,0) M = M R is f = df , f C (M ).
Remark 15.13. For X Vect(M ):
1. X (f ) = X(f ).
0

2. X (F G) = F X G + X F G, for F T (k,l) M and G T (k ,l ) M .

3. d((, Y )) = ()(Y ) + (Y ) for q (M ), Y Vect(M );
X((Y )) = (X )(Y ) + (X Y ).
4. X (, Y ) = (X , Y ) + (, X Y ) for p (M ) and Y T (p,0) M .
Proposition 15.14. Let g C (T M T M ) be a metric on M and a connection on M . The
following are equivalent:
1. is compatible with g.
2. g = 0.
3. Is V, W are vector fields along a curve c, then

d
dt

hV, W i =

dt V, W

D 
+ V, dt
W .

4. Suppose V , W are parallel along a curve c, Then hV, W i is constant along c. Equivalently |V |
is constant along c.
c
5. Parallel transport is an isometry: Given c : [0, 1] M , c(0) = p, c(1) = q, then Pp,q
: Tp M
Tq M is an isometry of vector spaces.

Proof. The identity

(X g)(Y, Z) = X (g(Y, Z)) = g(X Y, Z) g(Y, X Z)
implies the equivalence of 1 and 2. For 1 to 3 apply 1 to c(t):

d
hV, W i = c(t)
hV, W i = hc V, W i + hV, c W i =

dt
The implications 3 to 4 to 5 are obvious since parallel means
exercise.

D
dt V

D
V, W
dt

D
+ V, W .
dt

= 0. The implication 5 to 1 is an

35

16 Geodesics and the exponential map

= 0 for all t [0, 1].
Definition 16.1. A geodesic on M is a curve c : [0, 1] M such that c(t)
c(t)
D
Equivalently c(t)
is parallel along c; equivalently dt
c(t)
= 0.
Proposition 16.2. Geodesics exist locally and are unique given initial conditions c(0), c(0).

Proof. Recall the discussion of parallel transport, in particular formula (4). In this case E = T M ,

the frame is given by x

So c is a geodesic exactly if
i and s = c.
!

X
X

k
j
i
k
c (t)
c (t)c (t) ij 
= 0.
k
c(t)
x
i,j
k

Hence we need to solve the system

k (t) = ck (t)

P

k (t) + i,j j (t)i (t) kij 

c(t)

=0

for k = 1, . . . , n. This is a system of ODEs and hence locally has a unique solution.
Definition 16.3. Fix a point p M . For v Tp M let v be the unique geodesic through p with
v (0) = v. Define exp : S T M M by exp(v) = v (1) for v Tp M , where S T M is the set
where v (1) is defined.
Fact 16.4. From the theory of ODEs: S is open and exp is smooth. Clearly 0 S.
Lemma 16.5. exp(tv) = v (t).
Proof. Fix t and set c(s) = v (st). By the chain rule, c is a geodesic with c(0)

= t v (0) = tv. So by
uniqueness of geodesics, exp(tv) = c(1) = v (t).
Proposition 16.6. Consider exp restricted to Tp M and let S Tp M be its domain. Then exp is a
local diffeomorphism, i.e. there exist open neighborhoods U of 0 in S and U 0 of p in M such that exp
is an isomorphism from U to U 0 .
Proof. By the inverse function theorem is suffices to prove that
D exp : T0 Tp M
= Tp M Tp M
is an isomorphism. We will even show that D exp = I. Fix v Tp M and consider the curve : t tv
in Tp M . It has (0) = v and



d 
d 
d 
D exp(v) =
exp( (t)) =
exp(tv) =
v (t) = v.
dt t=0
dt t=0
dt t=0
Let e1 , . . . , en be an orthonormal basis of Tp M . Let the coordinatization with respect to this
basis be , i.e.

Rn

Tp M
exp

U0

U0 M

Set x = 1 exp1 on U 0 . This is a coordinate chart on U such that

36

1. x(p) = 0;
2. geodesics through p are straight lines, i.e. the geodesic through p with initial vector v =
is mapped to the curve v (t) = (v 1 t, . . . , v n t);
3. If g =

v i x
i

gij dxi dxj , then gij |p = ij ;

Definition 16.7. such a chart x is called a system of normal coordinates (or geodesic coordinates)
on M centered at p.
With r =

pP
(xi )2 define

P xi

r xi

on U 0 \ {p}.

Proposition 16.8. If |v| = 1, then v (t) = exp(vt) is a unit speed geodesic with v (t) =


r (t) .

Proof. As v is parallel, | v (t)| is independent of t and | v (0)| = |v| = 1. By 2 above, v (t) =

j

(xj ) = xr , so
i.e. v (xj ) = v j . Also r

v i x
i,


j 
tv j
x 
= P
= vj

2 1/2
r (t)
2
i
t v
as |v| = 1.

We want to study how geodesics change when the initial vector changes. For this we need study
variations. Fix a point p M and a vector v Tp M . Consider a curve in Tp M through v,
i.e. : (, ) Tp M with (0) = v. Set
(s, t) = exp(t(s)).
This is the geodesic starting at (s) at time t. We have (0, t) = exp(tv) and (s, 0) = p. Write
s (t) = (s, t) = t (s) (the first is considered as a curve in t with fixed s and the second as a curve
in s with fixed t). Define
d
d
t = s and s =
t .
dt
ds
s

t (s)
s (t)

v
t

Lemma 16.9. Ds t = Dt s , where Ds =

D
ds

with respect to s .

37

Proof. Say = (f 1 (s, t), . . . , f n (s, t)) locally. Then

t =

X f k
,
t xk
k

Ds t =

Dt s =

2f k
st
2f k
ts

X f k
,
s xk
k
!
X f i f j

k
+
i,j
t s
xk
i,j
!
X f i f j

+
ki,j
k
s
t
x
i,j
s =

So we have to show that kij = kji . By definition,

xi

xi

( x
j) =

kij x k and by symmetry of

)
=
,
= 0.
(
xj
xj
xi
xi xj

Thus kij = kji .

Definition 16.10. Fix a point p and consider exp : Tp M M . Let BR (0) = {v Tp M : |v| < R}
and SR (0) = {v TP M : |V | = R}. A geodesic ball of radius R around p is exp(BR (0)) and a
geodesic sphere around of radius R around p is exp(SR (0)).

Theorem 16.11 (Gau lemma). Let U be a geodesic ball around p. Then

geodesic spheres in U \ {p} (here r is with respect to normal coordinates).

is orthogonal to

Proof. Fix q =
6 p in U and let v Tp M with q = exp(v). Further
let X Tq M be tangential to the


geodesic sphere through q. We have to show that X and r
are
orthogonal. The geodesic from p
q
to q is given by (t) = exp(tv), t [0, 1], which is radial. So (t)

is proportional to
that (1)

is orthogonal to X.

r .

We will show

Pick w Tq M such that (D exp) : Tv Tp M

= Tp M Tq M has D exp(w) = X. Pick a curve
Tp M with (0) = v, (0)

= w and || R (this is possible as |v| = R and exp(w) = X is

tangential to the geodesic sphere with radius R. Set
(s, t) = exp(t(s)),

S = s ,

T = t .

Then
S(0, 0) = 0 as 0 (s) p;

d 
exp((s)) = D exp((0)) = D exp(w) = X;
S(0, 1) =
ds s=0
T (0, 0) = v as (0, t) = (t);
T (0, 1) = (1).

Trivially, S(0, 0) is orthogonal to T (0, 0). We want to show that S(0, 1) is orthogonal to T (0, 1). Note
that Dt t = 0 since as t is parallel as is given by a geodesic for fixed s.

38

d
hS, T i = hDt S, T i + hS, Dt T i
dt
= hDt s , T i + hS, Dt t i
= hDt s , T i = hDs T, T i
1 d
1 d
1 d
hT, T i =
|T | =
tR = 0
=
2 ds
2 ds
2 ds
So hS, T i = constant = 0 and hence X is orthogonal to (1).

Theorem 16.12. Let x1 , . . . , xn be normal coordinates centered at p on U M . For any point

q U , the radial geodesic from p to q is (up to reparametrization) the unique curve of shortest length
from p to q.
Proof. Let U = exp(BR (0)) and let v Tp M with q = exp(v). Set = |v| < R. Then (v) =
exp(tv) = (tv 1 , . . . , tv n ) is the radial geodesic from p to q.
Z

|(t)|dt

= .

l() =
0

Let : [0, 1] U with (0) = p and (1) = q. First, we want to show that l() . Write

= f (t) r
+ Z(t) with Z(t) tangent to the geodesic sphere through (t). By Gau lemma,

d
2
|(t)|

= f (t)2 + |Z(t)|2 f (t)2 as | r

| = 1. Also dt
r((t)) = (t)(r)

= f (t) + Z(t)(r). So
| {z }
=0

Z
|(t)|dt

l() =
0

Z
f (t)dt =

d
r((t)) = r((1)) r((0)) = .
|{z}
|{z}
dt
q

Now, if l() = , we must have Z(t) 0, i.e. (t)

= f (t) r
. If we reparametrize to have unit
speed, f (t) 1, then is a radial geodesic.

Definition 16.13. A curve is locally minimizing if whenever p = (t0 ) for some t0 , there exists a
neighborhood U of p such that if qi = (ti ) U (i = 1, 2), then is the shortest path from q1 to q2 .
Theorem 16.14. A curve is a geodesic if and only if it is locally minimizing.
Proof. handout (non-examinable).

17 Curvature of Riemannian manifolds

Let M be a manifold, g a Riemannian metric and a Levi-Civita connection. Recall that the
curvature is represented by a form R 2 (End(T M )). So for X, Y Vect(M ), we have R(X, Y )
0 (End(T M )) given by R(X, Y ) = X Y Y X [X,Y ] . Given a third vector field Z, we set
R(X, Y )Z = X Y Z Y X Z [X,Y ] Z.
This is a tensor of type (3, 1):
R(X, f Y )Z = X f Y Z f Y X Z [X,f Y ] Z =
X(f )Y Z + f X Y Z f Y X Z f [X,Y ] Z X(f )Y Z = f R(X, Y )Z.

39

Similarly it is C (M )-linear in the other variables. If x1 , . . . , xn are coordinates,

R=

l
Rijk
dxi dxj dxk

i,j,k,l

l
Rijk

xl

with

=R
,
.
xl
xi xj xk

Definition 17.1. The Riemannian tensor is Rm (X, Y, Z, W ) = hR(X, Y )Z, W ig .

It can be written as
Rm =

Rijkl dxi dxj dxk dxl ,

Rijkl =

m
glm Rijk
.

Lemma 17.2. The Riemannian tensor has the following symmetries:

1. Rm (X, Y, Z, W ) = Rm (Y, X, Z, W );
2. Rm (X, Y, Z, W ) = Rm (X, Y, W, Z);
3. Rm (X, Y, Z, W ) = Rm (W, Z, X, Y );
4. Rm (X, Y, Z, W ) + Rm (Y, Z, X, W ) + Rm (Z, X, Y, W ) = 0;
These say
Rijkl = Rjikl ,

Rijkl = Rijlk ,

Rijkl = Rklij ,

Rijkl + Rjkil + Rkijl = 0.

Definition 17.3. A manifold M with Riemannian metric g is called flat if R 0.

Theorem 17.4. A manifold M with Riemannian metric g is called flat if and only if it is locally
isometric to Rn , i.e. if around each point p M there
P exists a chart (x, U ) such that Dx is an
isometry (where Rn is endowed with the usual metric i,j ij dxi dxj ) or equivalently if there exist
P
local coordinates such that g = i,j i,j dy i dy j on U .

List of Notation
[X, Y ]
R

Lie bracket of vector fields

integral of an n-form

26

a connection

28

X s = (s, X) C(s X)

28

open subset of

product in the exterior algebra

21

Altr (V )

alternating r-forms over V

21

BR (0)

ball of radius R centered at 0

38

contraction

21

Cba

a-b-contraction

21

40

C (c, E)

sections of E along c

C (M )

smooth functions M R

C (M, E)

sections of E over M

19

exterior derivative

23

d(a, b)

distance between two points

33

P xi

37

D
dt

covariant derivative along a curve c

30

r xi

30
4

tangent vector a p in coordinate direction

DF

derivative of F : M N

df

1-form given by the derivative of f

DF |p

Jacobian matrix of F : M N

DF |p

derivative of F : M N at p M

df |p

differential of f : M R at p; same as Df |p

DF (X)

push forward of a vector field

Diff(M )

group of diffeomorphisms M M

boundary manifold

27

covariant derivative

29

dual bundle

19

EF

direct sum of vector bundles

18

Ep

stalk at p for a vector bundle : E M

17

()

sign of the permutation

21

E|U

1 (U ) for a vector bundle : E M

17

EF

tensor product of vector bundles

21

exp

exponential map

36

pullback of a form

23

F X

push forward of a vector field

Lie algebra of the Lie group G

a Riemannian metric

(0)

tangent vector associated to a curve

ijk

Christoffel symbols

30

(M, E)

sections of E over M

19

19

33
5

41

H+

the half-space {(x1 , . . . , xn ) Rn : xn 0}

27

q
HdR
(M )

q-th de Rham cohomology group

24

Ht (x)

H(x, t) for a smooth homotopy

25

r E

exterior power of a vector bundle

23

r f

induced map r V r W

22

r V

r-fold exterior power of V

21

exterior algebra of V

21

Lg

left translation in a Lie group

LX f

Lie derivative of f C (M ) along X Vect(M )

11

LX

Lie derivative of a differential form

24

LX Y

Lie derivative of Y Vect(M ) along X Vect(M )

12

M \ M

27

p (E)

E-valued p-forms, C (M, p T M E)

28

p (U )

p-forms on U

23

O(n)

orthogonal real n n-matrices

14

c
Pa,b

parallel transport map

31

pullback of a function by a diffeomorphism

11

pullback of a vector field by a diffeomorphism

12

t (p)

flow through p at time t

10

: E M

a vector bundle over M

17

curvature

31

element of 2 (End(E)) representing R

31

Rm

Riemannian tensor

40

R(X, Y )

s 7 C(R(s) X Y )

32

Sr

symmetric group

21

SR (0)

sphere of radius R centered at 0

38

s(t)

section along c(t) for a section s

30

Symn

symmetric real n n-matrices

14

T M

cotangent bundle to M

19

(X, Y )

torsion

34

= (ij )i,j

curvature matrix

32

= (ij )i,j

connection matrix

29

42

T (k,l) M

(T M )

(T M )l

21

T (k,l) (U )

mixed tensors, C (U, T (k,l) U )

21

TM

tangent bundle to M

Tp M

dual space to Tp M

Tp M

tangent space at p M

T (V )

tensor algebra of V

VectL (G)

left-invariant vector field on G

Vect(M )

space of vector fields

X(g)

vector field acting on a function

|Xp |

hXp , Xp i 2

Xp

X(p) for X Vect(M ) and p M

Xp (g)

short for Xp (g)(p) = X(g)(p)

hX, Y i

same as g(X, Y )

21

33

33

43

Index
1-form, 19, see also form, 23
alternating form, 21
atlas, 3
compatible, 4
base space, 17
Bianchi identity, 32
bump function, 8
bundle isomorphism, 20
bundle map, 19
over M , 20
centred chart, 4
chain rule, 5
chart, 3
centred, 4
with boundary, 27
Christoffel symbol, 30
closed form, 25
cocycle, 18
condition, 18
cohomology
de Rham, see de Rham cohomology
singular, see singular cohomology
compatible atlases, 4
complete vector field, 11
connection, 28
matrix, 29
on M , see linear connection
contraction, 21
cotangent bundle, 19
covariant derivative, 29, 30
curvature, 31
matrix, 32
curve
associated tangent vector, 5
locally minimizing, 39
de Rham cohomology, 24
de Rham theorem, 25
derivative, 5, 23
covariant, see covariant derivative
diffeomorphism, 4
differentiable structure, 4, 27
dimension, 4
distance, 33
distribution, 15
integral, 15

44

smooth, 15
dual bundle, 19
embedded submanifold, see submanifold
exact form, 25
exponential map, 36
exterior algebra, 21
exterior derivative, 23
exterior power, 21
of a vector bundle, 23
fiber, 17
flat, 40
flow, 10, 11
existence, 10
form, 23
pullback, 23
vector field valued, 28
frame, 29
Frobenius integrability theorem, 15
fundamental theorem of Riemannian geometry,
34
Gau lemma, 38
geodesic, 36
ball, 38
sphere, 38
geodesic coordinates, 37
Hopf line bundle, 18
immersed submanifold, 13
immersion, 13
induced connection, 35
integrable distribution, 15
integral, 26
distribution, 15
manifold, 15
supported in a chart, 26
integral curve, 10
maximal, 10
inverse function theorem, 14
Jacobi identity, 9
Koszul connection, see linear connection
left translation, 9
left-invariant vector field, 9

length, 33
Levi-Civita connection, 34
Lie
algebra, 9
bracket, 9
derivative
of a form, 24
of a function, 11
of a vector field, 12
group, 9
subalgebra, 15, 17
subgroup, 15, 17
line bundle, 17
linear connection, 33
compatible with a metric, 34
symmetric, 34
torsion, 34
torsion-free, 34
locally minimizing, 39
manifold, 4
topology, 4
with boundary, 27
maximal integral curve, 10
mixed tensor, 21
normal coordinates, 37
ODE
existence and uniqueness of solutions, 10
for geodesics, 36
for parallel transport, 31
orientable, 26
orientation, 26
on the boundary manifold, 27
oriented manifold, 26
orthogonal group, 14
p.o.u., see partition of unity
parallel
frame, 31
section, 31
transport, 31
partition of unity, 26
Poincare Lemma, 25
projective space, 4
pullback
of a function, 11
of a vector field, 12
of forms, 23
r-form, 23, see also form
rank

of a vector bundle, 17
Riemannian
metric, 33
tensor, 40
section, 19
along a curve, 30
singular cohomology, 25
smooth, 4
distribution, 15
vector bundle, see vector bundle
smoothly homotopic, 25
Stokes theorem, 26, 27
submanifold, 13
symmetric, 34
tangent
bundle, 6, 17
space, 5
vector, 5
associated to a curve, 5
tautological bundle, 18
tensor algebra, 21
tensor product, 20
functoriality, 20
of vector bundles, 21
topology, 4
torsion, 34
torsion-free, 34
total space, 17
transformation law
for tangent vectors, 6
transition function, 17
transition map, 3
for the tangent bundle, 7
trivial bundle, 17
trivialization, 17
variations, 37
vector bundle, 17
direct sum, 18
dual, 19
exterior power, 23
morphism of, see bundle map
rank, 17
tensor product, 21
vector field, 7
as derivation, 8
belonging to a distribution, 15
complete, 11
left-invariant, see left-invariant vector field
pullback, see pullback of a vector field

45