Differential Geometry

by Johan L. Dupont
Contents
Chapter 0 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 1 DIFFERENTIABLE MANIFOLDS . . . . . . . . . . . . . . . . 7
Appendix A PARTITION OF UNITY . . . . . . . . . . . . . . . . . . . . . 19
Chapter 2 CONNECTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Chapter 3 GEODESICS AND THE EXPONENTIAL MAP . . . . . . . 35
Appendix B THE TANGENT BUNDLE . . . . . . . . . . . . . . . . . . . 56
Appendix C DIFFERENTIABILITY OF THE EXPONENTIAL MAP . . 59
Chapter 4 THE CURVATURE TENSOR AND THE STRUCTURAL
EQUATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Chapter 5 THE SECTIONAL CURVATURE . . . . . . . . . . . . . . . 71
Chapter 6 CURVATURE FOR SUBMANIFOLDS OF EUCLIDEAN
SPACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Chapter 7 THE GAUSS-BONNET THEOREM . . . . . . . . . . . . . . 89
Chapter 8 LOCALLY AND GLOBALLY SYMMETRIC SPACES . 105
Chapter 9 LIE GROUPS AND LIE ALGEBRAS . . . . . . . . . . . . 119
Chapter 10 THE ISOMETRY GROUP OF A SYMMETRIC SPACE . 141
Chapter 11 SYMMETRIC SPACES AND ORTHOGONAL
INVOLUTIVE ALGEBRAS . . . . . . . . . . . . . . . . . . 151
Chapter 12 SEMI-SIMPLE LIE ALGEBRAS AND LIE GROUPS . . 161
Chapter 13 THE STRUCTURE OF ORTHOGONAL INVOLUTIVE LIE
ALGEBRAS . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
References A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

iii
Chapter 0 INTRODUCTION
Differential Geometry concerns geometric concepts studied by means of differential
and integral calculus. Its origin goes far back in the history of Mathematics, and an
important step in the development was the investigation by C.F. Gauss in 1827 of
surfaces in Euclidean 3-space. However, the real break-through in the theory was made
by B. Riemann in 1854 in his famous inaugural lecture “Über die Hypothesen, welche
der Geometrie zu Grunde liegen”: “On the basic hypotheses underlying Geometry”.
When you read this lecture you will be surprised to find that he here not only made
the foundations of Differential Geometry, but also that of other branches of Mathematics,
which were not existing at the time: Set-Theory and Topology. In his theory Riemann
wanted to create a framework, which on one hand included both Euclidean and the
non Euclidean geometries (which were new at the time) and on the other hand would
generalize the Gaussian theory of surfaces to higher dimensions. Furthermore, he
expected that his theory would be useful in formulating various parts of mathematical
physics, a subject he was also studying at about the same time.
As a starting point Riemann took the basic concepts of Euclidean Geometry:
“points”, “lines” and “distance”. First of all the “points” should constitute the
domain of his geometry and this should locally be described by real parameters,
( the dimension), so for this purpose he introduced the concept of an 
dimensional
manifold.
As an example we can think of a submanifold in Euclidean space, e.g. a surface
in 3-space

1
 
or even more concretely the 
sphere in   -space

But the main point made by Riemann was that a manifold is an intrinsically defined
object disregarded from the surrounding Euclidean space.
The other two concepts “lines” and “distance” are closely related since in Euclid s
“Elements” lines are characterized as the set of points lying “straight” between two
given points, i.e. it is a curve, which realizes the shortest distance between two points.
So how do we measure “distances” or rather “arc length” for curves in a manifold?
Let us look at our example with a submanifold in Euclidean space

γ
γ

and let    be a differentiable curve. Now the arc length in  is well-
known and is given by the formula
$
%& ' ',
 +* *
!#" )(

2
where 
 is the field of tangents along and  defines the norm of the
vector   Now again Riemann
s point is that this formula should be intrinsically
defined in  , that is, the norm of tangent vectors to  should be part of the structure;
furthermore this norm should come from an inner product in each tangent space. Thus
we are led to the following definition.

Definition 0.1
A Riemannian manifold is a smooth   manifold  together with a Riemannian
metric, that is, for each point !"# there is given an inner product in the tangent
space $&%'( i.e. a symmetric positive definite bilinear form.
) %*,+-.+/102$&%' 34$5%2 6 
Furthermore ) is smooth in the sense that for any two smooth vector fields 7498 in 
the function !:6 ) &
% ;7<%=85%2 is a smooth function.

Remark
For >?@$5%2 we shall often write ) %=>?=A B;>?=C %  B;>?=C depending on
the context.

Exercise 0.2
Let  D  be a submanifold of   Then for each !(E the tangent space
$&%' is naturally identified with a subspace of  and thus inherits the Euclidean inner
product from  . Show that this defines a Riemannian metric on  in the sense of
Definition 0.1.

Return to  a general manifold with Riemannian metric ) . The following definition

now makes good sense:

Definition 0.3
i) Let 0GFIH9JK?6  be a smooth curve. Then the arc length of is defined by

L LON RN
;GM P ; GQ S
TVU2 where W ) %&;X= for Y$&%'(
P

ii) Let Z0F H9J K6  be piecewise smooth, i.e. is continuous and there is a
subdivision H[ ]\# ^_ `.`.` bac J such that ed-F gfh ^ibgfjK is smooth for
k
mln.`.`.`o9pq Then define
L L N r a O L v-w
;Ge P  GM v w/xzyi{| }
ftsu^

3
iii) Assume is connected and let
 Then the distance between
and  is
defined
   !#" ()" (

   piecewise smooth, $%'&  *+
 $,-/.

Remark

We shall see later that defines a metric in in the usual sense and that this
metric defines the same topology as the given one ( being a manifold is a Hausdorff
space to begin with!)

Coming back to the geometric concept of a “line” as defined in Euclid0 s book, we

would again like to define a line in a Riemannian manifold as a set of points such that
the arc length between any two points on the line equals the distance in the sense of
Definition 0.3. However, this raises the following two questions:

Question 1

Given two points
123 ; does there exist an arc, called a geodesic curve,
connecting the two points and whose arc length equals the distance?

Question 2

Is the curve in Question 1, if it exists, uniquely determined?

Unfortunately the answers to both questions are “No”:

Examples 0.4

8 8
i) Let  8 4657  8 9. with the Riemannian metric inherited from 4 Let

: 5<; => ?;  

(-1,0) (0,0) (1,0)

Clearly, as seen on the figure, @ ABC1DFEGIHKJ However, a geodesic in L would also
be a geodesic in MNC and since that is unique it would contain the point A)OC1OPE , which
is not allowed, so there is no geodesic curve connecting B and D inside L .

ii) Let L GRQ*MTS UCVB6G North pole, DWG South pole

4
 p

The shortest length of an arc from to
is realized by any half great circle through
and
 Hence there are infinitely many geodesic curves from to
.
In spite of these examples it turns out that geodesic curves do exist “locally”, that
is, if the endpoints and
are not too far apart, and also in this case the geodesic curves
are unique. To show these facts will be our first task in the development of Riemannian
Geometry. However, before we get that far, we need some preparations.

5
Chapter 1 DIFFERENTIABLE MANIFOLDS
In this chapter we review the basic notions of differentiable manifolds

Definition 1.1
Let be a Hausdorff space with a countable basis for the topology. Let

(or zero).
i) A chart or coordinate system on is a pair   open , 
  a homeomorphism onto an open set.

ii) Two charts  and ! are said to have smooth overlap if
#"$&%(')*+,!$-*#.
and
/"!%(')$0-*+ 1-*#.
are smooth mappings (hence diffeomorphisms).

iii) An atlas for a differentiable structure on is a set 2 35)&678679 6;:=< of charts

4
with pairwise smooth overlap, such that covers 5)&6>9
i.e. ? 3 @ 6;:=< &68A Two
2
atlasses and 2 
define the same differentiable structure if 2B@C2  is again an atlas.


iv) A differentiable manifold of dimension is a Hausdorff space as above with
a differentiable structure, i.e. with an atlas A chart 2/A
belongs to the&
differentiable structure for ?
if it belongs to an atlas for this, i.e. if it has
smooth overlap with any chart in 2/A
We shall often write or to denote 
the differentiable manifold.

Notation
If is a chart we can write 3EDF ' HGIGIGH F KJ A Then FMLN are called
the local coordinates for the manifold.

Definition 1.2
Let O 
be a differentiable manifold. A submanifold ?PQRO  is a set such that
for each STU there exist charts  
around in S O
such that

*V?3XWZY[\^]] F P_ 'I`Y) 3aF P_(b `Y)c3dGIGHGe3aF  `Y) 3gf>h

7
 Exercise 1.3
a) Show

that  for


 a submanifold, the set of charts
 where   are charts in  as in definition 1.2,
makes into a manifold.

b) Show that 
 !#" $%'& !(*),+ is a submanifold of $%.-

&

c) Let
/ be the subset


  ! !  & ! 10 ! $2) + 
 / &  /
cover by the sets
43 
5 6 ! ! 7"8 & ! 3(:; +<>=%2)?A@B
3  3  / &9
and let 5DC 5 F E be defined by
 
5 ! ! G! ( 5/ ! ! H! -
 / /  / 

 3 3
Show that I 5 J  5 K + 3ML is an atlas for a differentiable structure on ; but
N /
x2

x1

is not a submanifold.

8
 Definition 1.4
A continuous mapping 
 between two manifolds is called smooth or
differentiable if for every chart  on  and  on
 !
"!#$% &'(*)+-,./0
(*) )!1 )!1

is smooth.

Notation
The set of smooth functions 2
3  is denoted 465 '798

The Tangent Space

Let  be an :; dimensional differentiable manifold and let <3=> .

Definition 1.5
i) A tangent vector at < is an equivalence class of pairs ?@BAC where $ is a chart
around and< AD= a vector.

E@BACF G2?HI-
if J #    ,LKNM GPO EI6QRA JS%Q differential of T8
U @BAWV G denotes an equivalence class, A0QXYA &Z[Z&Z[BA  are called the coordinates for
the tangent vector in the coordinate system 'B8

ii) The set of tangent vectors in < is denoted \ G  and is called the tangent space.

The following is an easy consequence of the definition

Proposition 1.6
] ) \ G  is an :; dimensional vector space over .
^ ) If _`ba x) is a chart around c , then there is a natural isomorphism d%e-fhgjiWk l m
given by

d%eWn daBoWpqi6ro

9
 ) If (y
 is some other chart around  we have a commutative diagram
 

 

   &  "! #%$ 

'
We can now talk about tangent vectors to arcs in

Definition 1.7
' (*) & (*)
Let
,+-

+.0/ be a differentiable curve and let 13254
6( ) (

+. with 132798;:=< The tangent >@?BABC3D-EGFIH JLK is defined by M,N ABC3DPERQ>@?BABC3D7E9Q
S TVUWLX W TRY U MO
J where Q M A >ZE[ for ( a x) any chart around : .
M O
 [
[ O]\^O`_
(Note: this is well-defined!)

Remark
T
For K Q a b c an open set and Q dfe we get a natural isomorphism
l\
d]e@ghiHjJja k cm<
We shall usually identify HjJja with c via this isomorphism. More
generally for K b c a submanifold (see definition 1.2) the tangent space HjJK for
:nFGK naturally identifies with a linear subspace of c , namely the subspace of tangent
vectors to curves through : lying entirely in K .

Another way of characterizing tangent vectors is in terms of the associated direc-

tional derivation:

Definition 1.8
S TVUWqX
Let opJnQ J r
F HjJK and let stFtuwv9AxKyEz< Then the directional derivative of
s with respect to opJ is
W7| }~€‚3ƒ „†ˆ‡Z‰ŠŒ‹Ž‚3ƒ„†ˆ‡Z‰zŠ’‘ }~-€“B”–•
opJ"A6sE{Q

(Note: again this is well-defined.)

Notation
“3—™˜ 
Let x)
‹
be a chart around š and let › ‰œœœ ›Ÿž be the canonical basis vectors,
“6¡"˜ œ.œœ ˜,¡"˜.¢ ˜ œœœ ˜¡j•
that is, ›’ ~¤ . Then we will denote the corresponding tangent
vectors in £
¥ § ¥ § ¥ §
§ ˜ ˜ § ˜ ˜ § I ‹ ¨† ˜ ~
¥@¦ ‰ §~ œœ.œ ¥–¦  §~ that is ¥–¦ §~ ›’ª©
§ § §

10
 Remark

  
  3"%$ &4' ) 
If then the derivation corresponding to is given by
 
    ! #"%$&('*),+.- 0/ 21   5   +.- 0/6

The directional derivative 87 9  4;:<=>;7 is a derivation in the following
sense:

? :@  >A7
Definition 1.9
is a derivation at B if ? is C linear and
? EDFG ? HDFIJBILKMNJB@ ? OF5
for all PQFR
S   T 6
? U: < >37
Theorem 1.10
B 9 
TV 
?If

that W9 ! DJ 6
is a derivation at then there is a unique such

For the proof we need a few lemmas.

Lemma 1.11
Let T
><=YXZ[PEX open neighbourhood of B
> 6 Thenfhgtherei is \T
>E>
such that and \ agree in a neighbourhood ] ^ _a`cbId e outside a bigger
neighbourhood j kmlonpj n j nq_%r

stG_vgou ~ _xwz{ y | r
Proof
_
By taking smaller we can suppose is the domain of a chart _
g }Ti t;_ w u k } }
l wU{ y _ w
Now choose a “bump-function” that is, is smooth, on some
zs €@‚;ƒ„l w } j w l w k j w n†_ w r
i
with and outside Then put
f e ‡ˆ‚ {h‰ outside j { Ns Š4‹jRwŒ‚
f
 }#%sŽ‡ˆ‚‚H 2‡ˆ‚ ‡‘’ƒ j

f ƒ”“Z•=Y_Z‚–k˜—N™5š*›Qš>!_%ksœ‚
Lemma 1.12
€(r sŽ{ J€@‚ ~ { i ` bIdžsz!_Z‚Ÿn
| Let
i
is a chart around
is an open ball with centre . Then there exist
Assume
0¡ƒ’“ • !_Z‚%k„¢ k¤£££,kQ¥Nk such that
f 2‡ˆ‚ { f ¦€@‚N§ ¨ | ©®2¯±°³²´ © ®2¯±° µL¯”¶R·
1) ©Œª«­¬
2) ©
´ ®¸I°¹ ºº¼½¤» ¾ ®¦¸@°

11
 
   and  the identity chart.
Proof

 
 % we% have
It suffices to take an open ball,
Then we must show that for defined in


 
 
 
 
 
  
 
 
 
 !
 #" $%  ' ( )*  
(*)
%,+ % &
for some ' .- with '  /0 11342 5  ! .
To show (*) consider 67 *6789 6;:< 8=> 
DE F
? *  A@B!C F 6  67 *67  G96
% %
$ %  DE  H I 67 *67   GJ6
& %%
 $% &  '   
with
% D E % D E % F %
' )*    H  67 *67   G96K ' / H /LG96M F  !ON 

+RQ,S
P  TUV- 
Proof of theorem 1.10
Let be a derivation at .
I. PLW=XYZPL=A[J=XYZPLW=X\[]=^"_=`[PWW=XON Hence PL=XabN By linearity PL(cdaZ for c
any constant.
Q,S
II. If  is zero in a neighbourhood of  then PW IefN In fact as in the proof of
lemma 1.11 we can find gh:  TUO such that gjiZ= near  and gkil outside
the neighbourhood where  vanishes. Hence m[9gnio so that
,pPL q[g Q,S pPL Ir[J=s"tuvr[wPL(gvMxPL7I
We conclude that if d ' : (TU and yi ' in a neighbourhood of  then
'
PL IzPL . Q,S
III. By step II we can extend P to all
S
Qbourhood of  , simply by setting PL(I,{Pw|~#}  where  is defined near €  } :
functions, which are just defined in some neigh-

(TU and  and } agree in a neighbourhood of  . Given  } exists by lemma

1.11. Extended in this way P is still a derivation on the set of all such locally
defined functions.

12
IV. Now let ( 
x) be a chart around  
as in lemma 1.12. Then

  
 
 
(**)

"
In fact by that lemma, !  #
and since $%&'(*)
0

+,-.)/   1 # 23)
0 
  
 40 5 76

Now the right hand side of (**) is the derivation by a tangent vector, which shows
the theorem.

Definition 1.13
Let 98;: <>= be a differentiable mapping between differentiable manifolds. The
differential of  or the tangent mapping @?@BACDFE+
  at the point HG9: is defined by
@?;I $J
LK;M  ON2PQ
QRSPUT(VTW$X Y[Z \^]7_`ba0c;d Ze\^]

where ( fQg x) and ( h-g y) are charts around i and j _i a respectively.

(Note: This is well-defined.)

Remark
For k lmfon pmgrq lshtn u open sets, the differential of jwv!frx h at
iHyf identifies with the usual differential z{j \ v p|x u via the natural isomorphisms
}\ } d Ze\^]
f~l pg h>l u€

The following is straight forward:

Proposition 1.14
} } d Ze\ ]
i) jƒ‚„v \ k x q is linear.
}
ii) Let † \ y \ k‡€ Then jƒ‚4† \ corresponds to the derivation

_ j@‚† \ aˆ_Š‰a l‡† \ _Š‰Œ‹ j a ‰ y„Ž _ q a

13
iii) The identity 
    

   and if    and

  ! are differentiable mappings we have the “chain-rule”
" "
$#  # '#  #
%    &$ that is, (% )  +*&,-/.  01

*43
 2   *&,-/. 

"
5# 
(% 76 8 

 ,:9<;+*/.),-/. !

Smooth Vector Fields

Definition 1.15
A family of tangent vectors =/>5?  @2A >5CBD , is called a smooth vector field
"FE EJI K I OP<Q5R
if for every chart +GH% the mapping given by LNM is differentiable.

Remark
K Q

Q Y
For I
S TVUXW
K
we can identify a smooth
R
vector
K]
field with a smooth mapK
U using the natural identification Z U\[T Q
More generally for S W
aQ submanifold
Y I K
we shall identify
Q^
a R smooth vector
Kd
field on S with a smooth map
S such that L`_bacZ S W for all La S (cf. remark following
definition 1.7).

Example 1.16
^ d+O djikiki<dmlHdVnpo d
For U _ a chart and efThg okqkr$s ` R t
R/u v is a smooth vector field in U
s
o ] s
which we of course denote okq r

Q
If now is any vector field in U , then by
x~definition
K x
w xzy|{D} ~€
Q
T
x

} Y I Q
where U are smooth iff is smooth.

Exercise 1.17
Q5R R d d
A family of tangent
^
vectors a‚Z S
Q^
LDaS defines a smooth vector field
iff for every ƒ„a†ˆ‡ SJ_ the function ƒ_ defined by
Q^ ^ Q5R‰^
ƒ(_ L`_HT ƒ(_ LŠa‹S

is smooth.

14

 

Next let us define the Lie bracket of two smooth vector fields and on
For each define

  

    "! #$

Proposition 1.18
 %
&  (')*+! # ,.-

i) The mapping is a derivation, hence defines a tangent
vector at .

ii) The family

 %/
0  1234 defines a smooth vector field.
Proof
Let 5678! # :9 Then

.';6<=
.>';6@?A*'B
.6C
Hence

 
.';6<(D4  
EFG'B65HIJ?K
 >'B  6CI?L  >'M
 6CI?ADHI';  
.N6C(

Similarly

  .';6<DO
 N >'B65H5J?K  'M
 N6P5?A
 >'B  6CI?ADH5Q
 N 6CF

Subtracting we get
 R
 .';6<  %
&  ';65)IJ?SGHI  R
 N6P
which shows i).
ii) is obvious in view of exercise 1.17.

Definition 1.19
R
 %
0  %
0 


For smooth vector fields the vector field given by above is
called the Lie bracket of and .

Notation
Sometimes we shall write
T=U 
VD  %/
9

15
 Remark
If 
 is any chart then the vector fields satisfy
 
    "!

The following proposition is straight forward

Proposition 1.20
The Lie bracket satisfies
i) # $%'&)(+*-, . linear in both $ and & ,

ii) /1024356&7  $843915"&;:<3=/10>4&?)@3BABC)DEGFH ,

iii) (Jacobi identity)

# $%I#J&KMLN(O(:P# &Q#JLR'$S(O(T:U#VLRI# $WX&)(O( @ ,

iv) equivalently
/ 0 X#J&KMLY(Z  #/ 0 4&?'MLN(T:P#J&KX/ 0 4L[G( ,

v) # $%'&)(  .?#J&$W( .

Now let us return to the notion of a Riemannian metric:

Recall that a Riemannian metric on F is a collection of inner products
\^] ] ]
_5- 5`baOc F dec F f

such that for $WX& smooth vector fields on F the function gWhf \^] Z$ ] X& ]  is smooth.
Now suppose (
i x) is a chart; then in particular the functions \ j  \lk>m m n oip m m n'qOr are
smooth and for each sStvu the symmetric matrix w6xIyjz|{}sT~' determines the inner product
x^€ in ‚€"ƒ<„ In fact, if p‡† tˆ‰€Oƒ have coordinates

yŒ zŒ
2ŠH‹ yŽ † Š;‹ † zSŽ
y Œ Ž€ z Œ Ž€
Ž Ž
Ž Ž
Then
y“Œ zˆŒ
x € {Z p† ~Šx €R‘’ ‹ y[Ž p ‹ † zSŽ
y Œ 
 Ž€ z Œ Ž €O”•
Ž Ž
Ž Ž
y z Œ Œ
Š ‹ † x €e˜ y[Ž p zSŽ
y—– z Œ Ž€ Œ Ž €Q™
Ž Ž
y z Ž Ž
Š ‹ † xIyjz|{šs›~
y—– z

16
 is, 
is the
That  bilinear form given by the matrix    with respect to the basis
 
  Notice that since 
is symmetric and positive definite, the matrix
 

 
!
 "#$ % is a positive symmetric matrix.


Remark
Notice that the Euclidean metric in &(' ) with coordinates *,+.- +/)10 corre-
sponds to the matrix 

687:9 ;
(***) 2
43 5
..
.
? @
; 9=<8>
Therefore if ( & x) is a chart in a manifold A with metric such that (***) holds,
then we shall say that the metric with respect to B& is Euclidean or flat. We shall
see later that given a metric we can not always find C a local chart, which makes the
metric Euclidean.

Theorem 1.21
Every differentiable manifold has at least one Riemannian metric.

Proof
Let D be an arbitrary manifold. To begin with let ( EGF x) be a chart and observe
that at least in E we can find a Riemannian metric simply by choosing the Euclidean
one with respect to x, i.e. such that (***) holds. Hence if we cover D by coordinate
charts HJIBELKMFONPKRQ%S KT U then in each ELK we can find a metric
VXW KZY
I [ Q\I^]_F]`Qba cd D efcgdJD h F [MijEkK.l

Now we can choose a partition of unity subordinate HmEkKnS K1T U F that is, a family
H#oRK S KgTmU of smooth functions oRKbapD
h F such that

i) qsrtouKvr w

ii) xgyXzXz{ouK}|~EkK

iii) For each [bi=D there is a neighbourhood  , which intersects x1yzXz€ouK for only
finitely many  and ‚ oR„ K ƒ wml
K

Then we can define a metric V by

V V W KJY
d1I‡†uF%ˆ!Q‰ƒ{Š oRKnI [ Q‹] d I‡†uF%ˆ!Q\l
K
V
In fact for fixed [.F d is a finite convex combination of positive symmetric bilinear
functions and hence is a positive definite symmetric bilinear function. Furthermore V
is smooth:

17
 Given and

smooth vector fields the function 
   
   


is locally a finite sum of smooth functions. This proves the theorem.

18
Appendix A PARTITION OF UNITY


    a covering of 
Definition A1
Let be a topological space and
A covering    is called a refinement of
if, for every ! , there
is an "#!$ such that  &% ' .
1)

A set of subsets )(*   of

neighbourhood  such that ,+-(.02 / 1 for at most finitely many "34$5
2) is called locally finite if every point of has a

6
 7 '   an open covering of 6 . A

Definition A2.

98    of :<; functions on 6 such that

Let be a smooth manifold and
partition of unity subordinate is a family
the following holds:
i) =?>853>A@ and B CDED*8F % '  3G for all "HI$FG
ii)  B CDD.8 KJ "LM$  is locally finite,
J
iii) for all N!O6 we have P
  F
8 R
 S
Q 5
N T U
 @

V7'W   6X Then there exists

Theorem A3

Y
Let be an open covering of a smooth manifold
a partition of unity subordinate

For the proof we need a few preparations:

Z[= \
Notation
For put
k'l
Q]Z^T_V`5aMb c dd
Wd e fge hji
We shall use without proof the existence of “bump functions” (see e.g. Warner [ ]):

m
Lemma A4
There exists a non-negative smooth function on c such that
m
e n3oqpsrut p l
i)

vwxxymz
n0o|{r
ii)

19
 Next we prove:

Lemma A5
Let 
   be an open covering of an  dimensional smooth manifold 
Then there exists an atlas

 !#" $&%('*),+.-'
with J countable, such that
a) /0 132 is a locally finite refinement of ,
b)  
465879 '
c) If we put :;
< = > 58465@?/9A9 'CBEDGFIHKJ
L/:;0 132 is a covering of  .

Proof
OM N 'PH.+ MQN "
MSR N/TU VXY/W Z U[S\ Y^]_a` _
1. First we construct a sequence of compact subsets , such that

_ b ]dcfe USgIhShShIg e Y gShShShji eY

and In fact, since is locally compact with a countable basis,

k Y k ]ml [ U ] e U `
we can cover by countably many open sets such that is
[
k [Y e UnV6hIhSh3V e Ypo [ Y ` Y
compact for every . Then we define by induction: For put For
[
b q rdk^s l
higher suppose that is found such that Then since is
compact and is a covering we can find an such that

[ ^Y o e UtV;hShShfV evuw`
Then we put [ Y/x U ] e UQV6hShSh V e u `

2. Now fix kzy ` Then we have (putting [O{ ] [!| U ]~} )

 ] [ Y/x U€ [ \ Y o [ \ Y/xƒ‚ € [ Y | U ]a„
 is compact and „ is open. Around every point of  we can now choose a
chart ‡†Gˆm‰ ˆ!Š ]‹ŒŽ such that for some ‘zy“’ we have ˆ o „<”;•–—` Let ˜ ]
where

| U Œ™‹6ŒEl/Eš` Since  can be covered by finitely many such ˜›Š s we obtain a finite set
of charts

(A.6) Œ ˆ Yœ – gž Yœ –3Ÿ gShShShg3¡ˆ Yœ –¢£¥¤X¦—§¨ ©ª£¬«

such that
­ §/®¯° ­²± §´³¶µ §¨ ©·¸;¹S¹I¹/¸ µ §¨ ©ª £»º
Let ¼ be the atlas consisting of all the charts ½¾À¿Á ¤A¦ Á0ÂÄà ÁÆÅ3Ç ÈÊÉ
º Then clearly ËLÌͽ/¿Á0à Á1ÅÇ is a refinement of Î º
in (A.6) for all

20
 
By construction there are at most finitely many 
 s inside  hence is locally
finite. also since
        
      
"!
it is clear that the 
 s cover This proves the lemma since # is clearly countable.

Proof of theorem A3.

&%(')+* )-,(.
Let $ be an open covering of and choose an atlas / 65
as6in
7 lemma
4
A5. Choosing a “bump function” as in lemma A4 we define
7 1
0 3

2 # by
&= 0?>A@B
C8;:D< : 7 

01
98;:+< E !
: F 

?
Then clearly 0
is smooth and satisfies
EHG G I
i) 0

ii) JKLLNM1OQPSR3OW
TBU VYXZV\[^]9]`_

iii) Mba cdO`e fhg

Note that ikjAl m defined by
i Von+]qpsr MO Von+]
Out^v
is well-defined and smooth since w is locally finite. Since w is a refinement of x , we
can choose a function yzjA{"m | such that }O~PS€‚ Oƒ for all †
„ ‡{g Now we define
for each ˆ‰&| Š ‹
‹ 1
M O oV n+]CŽ i Vn+]A_
Von+]qp r n l
Š ‹ 
€ ‚ OƒŒ
(and p’‘ if y V „ ]`p “ ˆ for all „”s{ D
Š ‹ ). Note that since the cO• s cover l we have
i Von+]—– f for all n ˜l _ and also is well-defined and smooth on l . Clearly
‹ M1O
‘š™ r ™›r MO p i
Š ‹ €‚ OƒŒ Ot^v
so that ‘œ™ ™ f(g AlsoŠ ‹ ‹
-ž ‹ -ž ‹
LL P Ÿ LLMOqP Ÿ }O3P 
ŠD‹D¢ ‹ €‚ OƒŒ €‚ OƒŒ
-ž
and ¡ LL t(£ is locally finite. In fact, suppose  is a neighbourhood of n such that
¹›º only if »½¼
¾À¿Á ÂCæ}Ä O¤¦¥¨§¨§©§©¿1¥«Á ªÂƬ;Å­ ÄÈhas
only Ç non-empty intersection with ® . Then ¯-°±±³²D´µ¶®¸·
¥¨§©§¨§©¥ which is a finite set. Finally we clearly have
É Á;Í+Ä ¹ É É Á;Í+ÄÆÓ(ÔÕÁ;ÍDÄ
²D´ ¬
´ËÊ(Ì ´-ÊhÌ ÎÏ ¬ЍÑB´6Ò
É Á;Í+Ä×Ó(ÔÕÁ;Í+Ä ¹ÙØhÚ
¹ ¬
¬uÊ^Ö Ò
¾ Ç
Hence ²D´ ´-Ê^Ö is a partition of unity subordinate Û Ú

21
Chapter 2 CONNECTIONS
In the introduction we mentioned the problem of finding the curve between two given
points on a Riemannian manifold such that the arc length is minimal. This is actually a
classical problem in Variational Calculus and for a solution curve there is a necessary
condition known as “the Euler differential equation”. It turns out that this is most
conveniently expressible in terms of a first order differential operator called a “linear
connection” associated with the Riemannian metric.
We therefore start by studying the formal properties of such a linear connection.

Definition 2.1
Let be a smooth manifold. A linear connection in is an operator (“dell”),
which to any two smooth vector fields
and  on associates a third  such that
1)    is smooth.

2)        

3)     
4) !   "$#   #&%('*)+,
5)   #- ./#   01
2 #345 #6%6' ) 7,98
Proposition 2.2
  :<; depends only on
>= (and  ), that is, if
>=?@
B=A CEDGFH   :I;>
KJ  :<;98
Proof
First assume that
and
A agree in some neighbourhood of ; , and choose a “bump
function” # on , which is 1 near ; and 0 outside a neighbourhood in which
and

A agree. Then #
L#
A so that by 4)
#    !   ! MJ  (# N J   98
Evaluating at ; yields
  : <; MJ  9I;O8
Hence   9I; makes sense just
T is defined in a neighbourhood of ;
U V>\
. Now choose
a local chart ( PRQ x) and write
S VXW3Y[Z \]^ in _R` Then by 2) and 4)

abcde g f ^nm b cde r q Z V m ^ b c d

VIW0Y mpo
^ihkjl mpo ^
23
Hence 


 

   

 
 
where the right hand side only involves
  !#"$&%'%&%&)( , that is, the coordinates
of * + . 

Notation
We shall write -,.
/ 0

 

Remark
The Lie-derivation 1
 does not define a connection since it does not satisfy
4) in the definition 2.1.

Exercise 2.3
Let 2 43 5
be a submanifold, and as usual identify 67+82 / :9 2 , with a
9 let < < +
subspace in
5;
(See remark following definition 1.7.) For
denote the orthogonal projection of onto . Now define for 67+
2   * 5 + 9 76 +82   =,
as follows: 
 2 ?> 2 @
Identify , a smooth vector field on , with a map
5 and define

,8
 <BADC ,8
DE +

where C ,
 C

* +  A *F+ A   E8&%'%'%& *F+ A  5 EGE is the directional derivative
of 
in the direction * + . Show that .,
 defines a connection in 2 in the sense
of definition 2.1.

Notation
For 2an open set in H the connection in Exercise 2.3 is called the
Euclidean or flat connection.

Next let us express a connection in local coordinates:

Let A HI)J  '%'%'%K)J E
be a local coordinate system on and let us put 2 L7M YX
&N NOGP ) QR M
"$'%'%'%S)( for short. Then T
(restricted to ) is determined by H
smooth functions (VU W
given by
X YX

(2.4) N 
L Z WM LM
M
24
    
In fact let 
 
 be smooth vector fields on  then
 
     !
#"     %$& "    (')
 $&  (- .     /')

*"   " ,+  
  0- 1.  32 74  4/8

*"  "   +  "  "    "546
<;  4>= .  $& ?4   ')A@B 4

*" 4 9: "    "   "  6 
That is

C 2  4D 8 4
 
#" 4 "   
4 D   4 . ?4  

 " 6 
(2.5)

We summarize in

E F GIH \7JL] KNMOJ(POP(POJLKRQTS a local chart in U . Then VV H J is

Proposition 2.6

WYX Z3[ in ^`_badcTe . Furthermore given WfX such functions

Let be a connection in and

g H ^h_bajike and this defines a connection in H a

determined by the smooth functions
we can define in by the formulas

g
Proof

nl mpoqA^Hera
It remains to prove that given by (2.5) satisfies definition 2.1. The only non-trivial
point is equation 5). So let Then

s ^`lNt!evu#w w \zy \<{ l%| [>} J \~€ [

[ x
where { ‚l | [O} J \ u €5\ { l‚| [O}Cƒ w Z \?[ ] lf| ] u €k\ ^`l„e3O| [†ƒ ‡l >| [ J \
]
Hence

s ^`lNt!eˆu w w \ y \ 5€ \ `^ l„e3>| [ ~ € [ ƒ w w \ y \ lf| [ J \ ~ € [ Šu ‰‹^l„eŒt ƒ l s ^t!ea

[ x [ x
25
From this we conclude

Corollary 2.7
Every smooth manifold has a connection.

Proof
The proof is similar to the proof of the existence of a Riemannian metric:  Cover

   which
 In each
 choose any connection, say put 
by local coordinate charts.
defines a connection in
 Now choose a partition of unity   subordinate

   , and put
!#" 


that is
$&%('*) ! "   %+',)
 $

One checks easily that this defines a connection in .

Parallel Transport
Now consider a differentiable manifold with given connection - . The reason why
- is called a “connection” is that it provides a way of “connecting” the tangent space
at one point . by the tangent space at another point / . We cannot expect to find a
canonical isomorphism 0213 ! 4 056 for any two points unless is parallellizable
(which not all manifolds are). However, given a smooth curve
798:<; >=>?@7 %; ) ! . 7 %=) ! /
we can parallel transport a tangent vector ACBD01 along 7 to a vector in 05 % ).

Let 7E8F: ; >=G?H@ be a smooth curve. By a smooth vector field along 7 we mean
a family IJK JLM NPO QLR of tangent vectors SUTWVYXUZ3[ T]\(^ which is smooth in the following
sense:
Suppose _a`cbGdfegbghghghibGdkjl are local coordinates
v near mHn]oqpsr and
v*{{
uv j x {
S T t ]n oGrzy Z3[ T|\ v
w d
e y
for o in an interval around oqp ; then we require the functions x n]oGr to be smooth (note
that this is independent of choice of chart).

Example.
Z
The velocity vector field }T is a smooth vector field along m .
}
26
 dγ
p
dt

Now our connection makes it possible to make the covariant derivative of a vector
field along :

Lemma 2.8
There is a uniquely determined operator in the set of vector fields along

 


 
   


called covariant derivation along such that

a)       #"

! $  

,.-0/ 21 

b) For %'&)(+*  34 5   3 " % ! 

-
c) If  7
 698  6 a vector field on : , then
 

        >
 
;  < 6
=

Proof
It is clearly->BDenough to prove this locally,
C4-FEFEGEG-0B9HJIK-
so we can assume that
,M- / NPO -
there are local


coordinates ?A@ such that $L @RQ Put S;T  U and write 
X T - T 1 ,M- /[N  B -`_a-cb -FUFEFV4EFW Ef->g
TZY S;T Y L smooth, and 9\  \]^ ed Q

27
 Uniqueness: Using a) – c) we obtain

    
 
         
 ! ,+.-/
   " "*) 
  
"$#&#&% '0 (
  2
 1 -/
    0
    0 " 
 8
0 5 ( 5 -/
    043 5 21
  5  0 76 
 8
0 5 -/ 5
5    890 :1
   6  
That is given ; and
1=<?>AB4@C depends only on

< which proves uniqueness.

Existence simply follows by defining

>ABD@C by the above formula.

Definition 2.9
A vector field

along
1 is called parallel if
>ABD@CFEHG

Proposition 2.10
Given a smooth curve IKJML NOQPSRUT V and a vector WXZY\[]:^ X`_ Vba Then there is a
unique parallel vector field Wdc along e extending WdX .
Proof
f8g O4hjikOmlml`l`OQhdnpo
By subdivision it is clearly enough to do this for lying inside a coordinate chart e
|
. Looking at the proof of lemma 2.8 we see that we must solve the
differential equations q:rts q 8|s z r z€‚ „ƒ †
quwvyx{z9| q e u~}
Omlmlml‡O4ˆ

28
for and initial condition  

    "! .
Now this is a differential equation of the form

#$&%'( )+*,-$  /. 0
for %12  an !435! matrix-valued function. From the general theorem for existence and
uniqueness of differential equations we know that such solutions exist at least in a small
interval. However, since the equation is linear the solution exists over all of  $  6 In
fact suppose 87:9; is the supremum of end points of intervals  $ <= on which solutions
do exist. Now in a small interval around  7 we can find solutions >    8
that for each in the interval ?>@2     A  is a basis for 0 6 Now given B choose
0 such

8CD9E 7 in this interval so that we have0 a solution defined on   <F , GCH9I<&9E 7 .
Then BA C FKJ/> ?>  C MLN 8LOJ P C  QJ
and J/> ?>/LN  GLRJ 0 0 is a solution in
0 0
a small interval around  7 extending 6 This is a contradiction so  7S .
Furthermore the solution is unique which proves the proposition.

Definition 2.11
TGU *WVYXZ . V[ Z
given by T8U ]\  : ^\M_ for \$` a parallel field along a is called
the parallel transport along a .

Remark
TGU *bVcX?Z . V[Z is a linear isomorphism.

Example 2.13
Z  0 with the Euclidean connection. Then T *dVYX 0 . V[ 0 is the usual
parallel translation.

Exercise 2.12
a) Let eFf:g h be the unit sphere with connection induced from h as in exercise 2.3
Let a be the horizontal circle parametrized by
abA =i]jlknmoMpqjlortsupd<v  wE9xjy9z|{l9;<O9}
w~€q~Kƒ‚4 j f LN< f 

„ 2
 )
  
 †
{ ‡
o ˆ
r u
s p ‰ 
k Š
m M
o p ‰ c
w y
  ‹ > Œ U ` be the normalized velocity field and
 A |{<lkmŠoMp|{Ž<ortsuGGj be the orthogonal
Let
Πfield pointing to the north pole.
Show that for any angle  7‘
S wGƒ‚1 the field ’ given by

’“” Kkmo•P 7u{–<W—„ 2 ,L˜ortsl 7u{–</"  A  p
N wG@‚,

is parallel along a .

29


     "!$#&%
b) Let be the cone

       

#(' '*) )-,
    +    0# the
induced connection from  and let .(
/ be the ray through the point   .
where are fixed and satisfy Again let have

Finally let  have the Euclidean connection and let 12
 be the sector given
 546078/ #9'$79'(:<; %
by polar coordinates

13  
=> 54?A@-BC79D4 B EGFH780,

Show that there is a diffeomorphism IKJL1NM POQ. given by
where

546078 4 ?A@-B 7 4 BSETF 7 4

M R     VU
(Hint: Express the Euclidean connection in  by polar coordinates, i.e., find WVY[X Z ).
and that it preserves the connections.

0d0e
Let \^] \
 _ be two submanifolds which “touch”>d0e5, along a curve `aJ b c M
\K]Dfg\  that is, hji8kmlonp\q]rshji8kml np\  for all tvu(bwc Show that in the induced
c)

connections from _ parallel transport along ` in \q] is the same as parallel

transport along ` in \ .

Give a “geometric proof of a), using c) with \q]x y  and \ a cone in 
d)
touching y  along the circle `


M2

M1

We now again turn to a manifold with a Riemannian metric, and we want to

show that there is a canonical connection associated to it, the so-called “Riemannian

30
connection” or the “Levi-Civita connection”.

  
Proposition 2.13
Let be a Riemannian manifold with metric a connection in
the following are equivalent:
i) For all smooth vector fields  
    !  " $# % & 
For every smooth curve ')(+*-, .0/ 1 2 the parallel transport 3546(87 4:9<;>= 1
7 4?9A@B= is a linear isometry.
ii)

Proof
First let us show that i) is equivalent to


i ) For 'C( *-, ./D1 G any smooth curve and EF vector fields along '
G:HJI EFLK NMO G?EH F)P #QM ER O G?F H PS
In fact clearly i ) implies i). On the other hand, if i ) is true for ERFC and TU(V* , ./V1
is smooth, then it is easy to see that i ) is true for E and F replaced by TWE and F or by
E and TWF . Now if i) is true for then, it is clearly true locally, hence in particular
it follows for any  and   X Y[Z]\ ^ in some local coordinate system. Now
since acb!d ebgfhe only depends on i$bgfje we
>
X ^>_` thus have ik ) valid for l \m n ^ \p^
restricted to q . And since any vector field along q is a linear combination^>_[ofo these ^>_ `
where the coefficients are smooth functions, ik ) follows by our previous remark. Thus
i) is equivalent to ik ).
Now rseut rvrse0w In fact by ik )x for l and x m parallel along q we have since
y{lz|~} \ :my z |~}
that  l
€[ \mL‚ƒn } \ that is,  l \m„‚ is constant along q and in particular
 l† \m ‚Rn  lˆ‡ \m ‡ ‚
To prove rgre t rse or rather rvre t r k e first choose parallel fields along
which proves ii).

q \Š‰‹>\Œ>Œ>Œ5\‰Ž\ such z 
‰ ‹ z b‘†e \Œ>Œ>Œ%\‰Ž b‘ˆe is an orthonormal basis in ’”“?• –B— w Then zS˜
by assumption b e b e is again an orthonormal basis for ’ “?• – — for each
that
™  w Now for ‰l]‹ b z e \>Œ>Œ>Œ5\‰Ž ž b z e b z e b z e £ b z e b z e any €two vector fields
\š0›
along q we then have
nœŸžŠ ‰ž \¡m n¢œ£5¤ ‰W£
 l \m„‚¥n§¦ ž ž ¤ ž \
31
  
hence   
 

  
  


But since    $
  $
 

 
"! 
#$ 
%!

'& &
(because the ! s are parallel) this equation is just i ) which proves the proposition.

Definition 2.14
A connection ( is called symmetric or torsion free if for all smooth vector fields
)
and *
+-, */.10 ) )
2 , . 43  *65

Exercise 2.15
)
Define for  * as above the torsion
7 ,)  )
*8. +9, */.10 2 , .10 3 )  *:5
7 7 , )  */. <, ; . depends only on ):= and * = 
a) Show that is a tensor, that is,
7 , )  7 ,)  7 ) 
(Hint: Show that for >@?BA6C ,ED .  > *9. >F*/. > , */. and
compare the proof of proposition 2.2)
U$
b) If GIH KJMLNPONONOPU$ KJQR is$I:aXFlocal
Y coordinate system and 1
S T are defined by (2.4), then
7 WV
iff S T S T [Z

Theorem 2.16
Let D be a Riemannian manifold. Then there is precisely one connection ( such that
a) is symmetric.

b) Either i) or ii) in proposition \ ^]_ is fulfilled.

Notation
The connection given by theorem \ <]P` is called the Riemannian connection or the
Levi-Civita connection for the Riemannian metric.

32
 Proof of theorem 2.16.
Uniqueness. It is enough to do it locally. Thus let 
    be a coordinate
chart and as before let       , and !"$#%&'()*#"+, where &-./-0+ is the
Riemannian metric. Now by assumption

using proposition 2.13 i)
6 6
(2.17)   1! #2 34  &' # ) 2 +(5&  #87 9 2 +:;&< #   2 7 +
   

and as in (2.4) let

@
6 @ED
=# 7 ?>A@CB $# 
 

Furthermore put
@
6 @
(2.18) F HGJILKA& =# 7  2 +M?> @ B $# ! 2
 

Then (2.17) reads

(2.19)N ! #2 OF PGJILK:;F 1IAQGK

and by cyclic permutations

(2.19)NRN *#S! 2 TF GLIAUKL:VF GW JIXK

(2.19)NRNYN  2 !8$#/OFZI QGKL:;F[I=G'K

6']L^
Now since the connection is symmetric F \GJILKAOF GW JIXK and we obtain from _ 7 NR`.N[N[Na
@
@

F \GJIKA ] b! #2 :c=#/! 2,d  2 !8$# 3;>A@eB $# ! 2
(2.20)
 1GJIf? 

g
Now introducing the inverse matrix
$#Lh
! ji"!"$#8k `

we can write (2.20) in the form

@ @
2
(2.21) B $# V> F \GJILK.! lQG,m? 
2
@
^
so that B $# are uniquely determined by i"!8$#8k and the derivatives of i!"$#8k This clearly
shows uniqueness.

Existence now follows @ locally by simply defining in a chart a connection corre-

sponding to the functions B $# given by (2.21). On overlapping charts the corresponding

33
connections will agree by the uniqueness. This ends the proof.

Notation
The functions
  and  are called Christoffel-symbols (of the first respectively
the second kind) for the connection, and the equations (2.20) and (2.21) are called the
Christoffel identities.

Exercise 2.22
Let    be a submanifold and let  be the connection in  induced from
 as in exercise 2.3.
i) Show that  is symmetric.

ii) Show that  is the Riemannian connection associated to the Riemannian metric
induced from  (cf. exercise 0.2).

34
Chapter 3 GEODESICS AND THE
EXPONENTIAL MAP

Let be a manifold with a linear connection
.

Definition 3.1
A smooth curve     is called a geodesic if the velocity vector field  
 

is parallel along , that is if

   

.02/ 1
Let us express this condition in local coordinates: let 0 "!3 #57$&6 %''$ ()0"('3 ()3 $+5 *-5 , be98 6;the: '()local
4 ('(< =
?
6 >
coordinates and let
so that 
be the Christoffel symbols. Let
0 A@ 0ED Now in the proof of proposition

  
 C
 B
field along  given in local coordinates by F
G
6 > 
0 H 0 0 2.10 we found that a vector

H/ 1 06  B is parallel iff it satisfies

JILK 0M1  . 1"/ 0 H  N 6O: '('()(<=QP

W^2U _ ST ^ SW T _
Hence is a geodesic iff it satisfies
SR TVU
(3.2) WX"Y[Z]\"^2_a` WX W XGbc debgfh'i)i'i)hjQk

l bnmpoVqrh'i'i)i<hotsvuxw sChzy|{ Y s
This is a system of 2nd order differential equations of the following form: Let
an open set with coordinates } l h ~€h‚„ƒ yO† s
is smooth, and the equation is
WYl l Wl
WX"Y‡b‚‡ˆ h WŠX ‰
In this situation we need the following

l }l q h~ q
Theorem 3.3
Given
for each l} }  h q ~ h ~  ‘q #ww‹Œ y
 there exists a neighbourhood
the differential equation
Œ of and ŽLc such that

WYl Wl
l
WX"Y b‚ ˆ h WXŠ‰
X ’ l } X  defined for “ X “V”g and satisfying
L
has a unique solution †
Wl
l }•c–—b l  WX™˜˜  bž~  k
˜˜›šœ
35
Furthermore the solution depends smoothly on the initial data 
 

“Proof”
This follows from the corresponding existence and uniqueness theorem for the 1st
order equations

 


 
   


with initial condition
 
     See e.g. Lang [ 4].

We shall now “translate” this theorem into a statement concerning geodesics in a

manifold. This involves the tangent bundle of the manifold considered as a manifold.
Thus
!#" ! "
 $ %
% &'

is given the structure of a smooth manifold determined by the local charts given as
"
follows: Let as usual ()+*-,.0/1/1/0+*3254 be local coordinates in and 67  8 the vector
! ! " 819 :
fields in ) . Then in )  ; % every tangent vector is uniquely expressible in
% &<

the form

> 2
= % = 7
 67BA
A%
7@? A
,

so
= %DCE = =
( * , GFH+I/1/1/I* 2 GFH  , 1/1/1/0 2 4KJ L 2

!#"
gives a chart in . For details we refer to appendix B (theorem B.4).

"
Now suppose is given a Riemannian metric (always possible by theorem 1.21).
!
Then a neighbourhood of the 0 vector in %1M+N OQPIR1SIO contains a neighbourhood of

the form

TU U
RWVYX#N Z V]\DO_^ R`^.acbd
Z.[

for some \fegN neighbourhood of [ih and some bkjlPim Again we refer to appendix
B for a proof (see proposition B.5).

36
 0p
0

U
( )

p
0

Again let be a connection in
(not necessarily the Riemannian one). Then we

can reformulate theorem 3.3 applied to the equations (3.2):

Corollary 3.4
For every point 
there is a neighbourhood  of  and real numbers 
such that: for each  and 
with  there is a unique geodesic
"!$#&%('*) + ) ,.-

such that
4 "!
"!"% /,1023 "4 5 % 6 ,70&8
5 5
Furthermore "!6% , depends smoothly on and 9:;<:*>= ?  @AB8
=
Remarks
1. In this corollary can be taken to be 1 by replacing by DC . In fact if
6!2#%E'*)  ) <,- 5 5 5

is a geodesic then "F ! % ,:0 "!6% ,3  %('D) ) ,G is also a
geodesic.
2. By the usual arguments for solutions to differential equations there is a unique
5
maximal geodesic "!6% , (i.e. defined on the largest possible interval %('*H JIK, , H JIDL6, .

Definition 3.5
5 5
Suppose "!6% , is defined for 0NM then put
O9PRQ  % S,70 6!"% TM ,

Remarks
1. By Remark 1 above O PQ  % , is defined for ULR<
of length VWX in a
neighbourhood of   8 Furthermore the map ZY- O PRQ  % , is smooth on this set. We
shall prove a more global statement in appendix C.

37
 2. Note also that 
  by the same argument.

Example 3.6
!
Consider  !  * with the induced connection. Then  "#%$
*  
 is given by  '& # +-,). In fact let /0 1-,2 then 43-56
&)( 
,1-,-) and 798;: 9<= is the orthogonal projection onto >=@?BA C; 3 =D of 3E3 <F
&)(
G +, which is clearly 0, so that is a geodesic.

Next let us study the local properties of geodesics. Again suppose H has a metric
and for I0JKH let, as in Corollary 3.4, L be a neighbourhood of I and let MONQP be
chosen such that RS- exists for TUJ6LVBWXYW[Z\M]V and so that it depends smoothly
on TV . Let ^_ La` H be the set of bJ R H V"TOJcL with WdW;ZeM and define
f
^g$ H hH

by !
f
 R i TjVk R  # .
f f
Then clearly is smooth, and lP  mnoIpVqIr.

Proposition 3.7
f9s
is non-singular at P  .

Proof
!
| Assume LV=tuvVvwvwxwvVt@y# are local coordinatesŠ‹ around I and as usual let z{c
|x}[~ Then any €‚F€„ƒX† is of the form €;ƒ‡‰ˆŠ]€ ŠŒ ƒ and we have a local coordinate
Œ
system on † given by

€;ƒFŽ l‘p’x“l”–•)—x˜x˜v˜v—‘™š“-”•"—k€’x—x˜v˜x˜v—€™› 

Now œ 0œ has a local coordinate system Ÿž\ž\—k‘ ’ —v˜x˜x˜v—=‘ ™ =— ‘ ’ —x˜v˜x˜v—‘ ™ › around
’ ’
“¢¡p—q¡š• . The differential £S¤ at the point ¥¦ is given by
§
‹ ‹ ‹
Œ Œ Œ
£S¤ ‹ Š Œ  ‹ Š Œ ‹ Œ
‘ Œ
©
Π
¨ v
ª « ‘ ŒŒ ¦+¬ ‘ Š Œ¦
Œ
§ ’
‹ ‹
Œ Œ
£S¤ ‹ Š Œ  ‹ Š Œ 
€ Œ ‘ ŒŒ ¦
Œ¨ ª «
¥  Q.E.D.
So £S¤ has matrix ­@®
® ®¯
Hence £ is a diffeomorphism in a neighbourhood of ¥v¦ . Again suppose this
neighbourhood is of the form

†°²±€;ƒ‚Œ O” † ³"€;ƒj³;´¶µB·
Œ

38
Choose neighbourhood of
such that

Then we have actually proved:

Theorem 3.8
Let M be a Riemannian manifold and a connection in M (not necessarily the
Riemannian connection). For each
 there is a neighbourhood and a real
number 
such that
1)  unique !"$#&%    with (' !)'+*, such that - ./ 20 143 %  -5!  is a
geodesic from q to +76

2) The map 8 / #9 given by  :  ./ ! is smooth.

3) For each ;<=>021?3 %A@CB !DE#F% GG '>!H'+*IKJD/  is a diffeomorphism onto an
open set L % =6
Notation
A neighbourhood of the form LM% as in 3) above is called a normal neighbourhood
of . 
Remarks
1. Notice that in theorem 3.8 can be chosen as a normal neighbourhood of .


2. One can actually prove that for small a normal neighbourhood of radius 
is actually convex, that is, the geodesic between any two points in stays inside .
(See e.g. Helgason, [2, Chapter I § 6].)

Geodesics for Riemannian connections.

So now assume that is the Riemannian connection. Recall that for N @O P RQST/U
the length V
is given by

V N XW VZ[Y  N \W ] Y ` ` N ` -

[_^^ - ^^
a
and consider the arc length function
^^ ^^
 - bW VX[c  N XW ] c ``eNd `ed 6
[ ^^ ^^
Now if N is a geodesic then
^^ ^^
` ` N f WhgFikj lo leo
`- `- lnm lemqp lnmsrut)v
^^ ^^
^^ ^^
39




   
so that is constant, in fact if     then
  so that  
 "!
  for $#%'&)( *+,( *-/.  0 Hence we have shown that a geodesic is parametrized
proportionally to arc length.

Notation
 
If 12-)4365789 with  ;: then  is called a normalized geodesic and  is
just the arc length.

Remark
If 3<57>=?@>BA9C then    is the length of the geodesic 3D57>=/E> from A to A9C ,
hence 1) in theorem 3.8 can be expressed by saying that up to parametrization by a
constant there is a unique geodesic from A to A9C of length FBG60

We shall now show that locally geodesics are the curves of shortest length:

Theorem 3.9
Let  be a Riemannian manifold and H the Riemannian connection. Let I and
G be as in theorem 3.8 and J#LK MNO:<P,.  a geodesic of length QR@S$FTG joining two
points U@MV+WAXYUZ:[+;A C \A A C ]IX0
Then for any path ^ in  joining A and A C we have QR@S8_`QR^a0

Furthermore = holds iff ^ and  agree after reparametrization.

For the proof we need several lemmas. First a small technical one.

Lemma 3.10
Let bc#>d].  Rd`e f with parameters 2' , be a smooth map. Then at every
point of d
g i g i
b b
h i  h i
   

Proof
Clearly it is enough to prove this locally, so assume b,2d8,ekj`+lZj,mon9qpOpOpqm>r>s is
a local chart. As usual let tou1wyv x z be the Christoffel symbols for the connection. Recall
from lemma 2.8 that |{ } is given by

~ € ˆ Œ@”• Œ
ŒyN‘“’ ˆ ˆ Œ

€-ƒ‚`„ †‡ € Š‰‹„ZŒŽ €  ‘ ‚–„>Œ ‘

40



  
    ! 
where , and a vector field along fixed. In particular for

& &
"%$& (' *) $
 12 &43
#  0

& & ,+ -
/. 

0
Now since 5 is symmetric, i.e. -

. .
in  and , which proves the statement.
, the above expression is completely symmetric

Next we have the famous

Lemma 3.11 (Gauss).

3
Let 687:9<; be a normal neighbourhood of = Then the geodesics through = are
orthogonal trajectories to the hypersurfaces
> 7?@A CBEDGFIH KJLNM JPORQ*SAT UWVYXZV\[^]_:UY]a`NbAc
7 MM
Proof
We shall show that tangents to a curve in d*S?e]Af are orthogonal to the radial
geodesics, i.e. the geodesics emanating from g . So let XhejiflkmQnSoTpUilkrqtsLUuGvwU be a
curve with VlXhejifxV\[zyc We shall show that
{
{ i:|G}~ S e]Xhejif€fa
ƒ‚ „E‚‡†
is orthogonal to
{
4{ ˆ |G}I~ S e ˆ hX eKiŠ‰ff  c
t‹ „Œ

q
Sq (c)

So consider Ž%‘€’a“\”G“–•˜—š™ ›L”œŸž given by

ŽK¡”¢•¤£p¥G¦I§¨!w¡ª©hj¢•€•
41
 


      "!$#&%
and let us prove



Now

'
 +* '


 *



( )


( )
-,
(

)

 *

  ,
 
) 
.
0/1  2  "
since
 in view of the previous lemma. Furthermore
is a geodesic and
33 3
33
 3353 4 76896 ;:
so


 >*

 4
 *

< =
 
. 
 
? ,
 
)
. @BADCEGF$%
= =
It follows that

'



 
)H I
    R  G $ 
so that JLK5M T(K5 M is constant  in$.V However, for have 
S  for all KON KOP)"Q !"# so that K5M U UW hence J K5M  K5M X we for %
KOP KON KOP)Q all and
\[ 
^In] the_ next lemma again YHZ is a normal neighbourhood of radius around
S

Lemma _ab 3.12

"
 "
! # 1  c
d
Let ` YHZ LS k Ue f$be a lpiecewise
f$ $$  smoothV curve andUfwrite
$ m  %
` 0
 i
g U
h j 5
A 8 
6 9
8 6 
:  X
 m
Z
Then npo  rts G ! auGk)vs
q `
8
and = holds iff is constant and monotone.

r(b)

r(a)
q

42
 Proof
Again consider

  

    "!# $% &(')

Then by the chain rule *,+

* -./
 0 0
 21 
0 0
. 7
and by Gauss s lemma 3 4
*,
5 354 hence
3+ 6 398;:
< < = < < = < < =?>
< * <  .
/  = < 0 < < 0 <  .
/  =
<  < <  < 1 <  <
< < < 0 < < 0 <

@
and = holds iff 354 (where this makes sense).
398
It follows that
*,+
+ B > B >NM M
E < < E
ACB
 < * <  . HJIK9
/L 
&9POQ
R$ )
D <  <
< <
D DGF F
F F
Furthermore =M holds iff M
HSITKU
,V 
/&UWOX
$;  
1) i.e. is monotone,
and
Y Z  
/ 
2) 354 “almost everywhere”, i.e., since is a diffeomorphism, is constant.
398
This proves the lemma.

Proof of theorem 3.9

 ^]  .
Again
Z 
S;
let [

be the normal neighbourhood of \ of radius
a`bcd-Y)
and \
>
with _ _ We shall show that if e is any piecewise smooth
. A
 ) d f 
curve from
 hg
RfJ
\ to \
g
,
then e
f
Now

let and consider the two spheres in
[ and of radius and respectively.

S(δ)

q
S(r) ω

43

  
Then there is some segment
  ofsometime
connecting and and lying totally in


between. In fact must

  
cross by

continuity and hence there
is  a “last”
   .
point   
  with

      and since !


and


a “first” point    
  with 
Clearly   #" $&%('*),+ - we conclude from the
previous lemma that

21
  . 0/ 3 '  for all
043576
Hence 
 8396 Now suppose   ;:<96 Then again for = 4>?435 we conclude from
  and  . In fact the segment
the
  to   has  
@3 '  A but also B:   8   .C ED  
FD   HG 
previous lemma that contains a ray connecting
from
: .CJIK
IKHG and   .C >  by the first part of the lemma, so that



  @L3 '   i.e.   M:< ' N6 For different 
where
 choices of these rays either have the
same direction or are disjoint, so for small they must have the same direction, and
thus, if we reparametrize by arc length, it contains the ray OQP" RTSVU OXW 
  5YL O L
KZTW[Z :\ where W is the common direction of the rays. Therefore coincides after
reparametrization with this geodesic.

Now recall the definition of distance

]F2^ + M:<_a`Vb ),  cdNe,c a piecewise smooth curve from
^ to +f- 6

Corollary 3.13
]
i) is a metric.
ii) The topology of g agrees with the metric topology given by
]6

Proof
First notice that for = small the normal neighbourhood $h% around + of radius = is
just the set

$h% : i +
 g j ] /k+l+
21Qm =,n 6

j q 5 \
:
In fact clearly o holds m since O?" RTSpU % OXW 0O
    is t a geodesic from + to +
K


RrSpU %  W
 of length ZTWsZ = 6 And other hand m if + $&% then a curve from
+  toB+ umustvwcutm the sphere   x  onofthe
 any radius

]  =

B 6 so by theorem 3.9 has
and
 =r hence  =? that is, ++ =
] ] ++
M:y5(z + : +
6
To see that is a metric
By the above remark +


h
$
the
%
only
so that
+

:
non-trivial
r
R p
S U  W 
statement is
T
Z
that
s
W Z :{\  5|L m = 6 Now
%
}4~5 then any curve from + to + has   ( so that ] +l+
€45  which
if
contradicts the assumption. Hence
J:‚5 , that is, + : +
. It is straight forward to prove
the triangle inequality:
]a^ + @L<] / ^  +
a1 Dƒ] / +
+ 1 Q
v ^  +l+
6

44
For this choose 
 and a curve  joining  and   of length   and
a curve   joining   and  of length       !#" Then  followed by   gives
a curve from  to  of length $%  &'  (&*)+ so that

*-,. %/ 0&,1   / 2)3 45+"

Hence the result.
For the second part we just observe that the topology of 6 has a basis consisting
of the neighbourhoods 7&891 for small. However, this is exactly the topology defined
by the metric by our first statement.

We can now prove a global version of the last part of theorem 3.9:

Corollary 3.14
Let ;:=< >?@BA 6 be a piecewise smooth curve parametrized by arc length and
suppose that  has length less than or equal to the length of any other curve from ='C
to ='?' . Then  is a geodesic (and in particular a smooth curve).

Proof
It is clearly enough to show locally that  is a geodesic curve. So consider a
segment EDF< GH.I@ contained in an open set J as in theorem 3.8. Then by theorem 3.9
EDF< GHI.@ must agree with a geodesic (ED%< G.IK@ clearly has smaller length than any other
curve from ='G> to =LIM ).

Definition 3.15
A geodesic realizing the distance between two points is called a minimal geodesic.

Remark
Thus corollary 3.14 says that any curve realizing the distance between its endpoints
is (after reparametrization) a minimal geodesic. Also by theorem 3.9 small segments
of a geodesic are minimal.

We shall now find conditions which ensure the existence of minimal geodesics
(although they are not unique).

Definition 3.16
A connection N on a manifold 6 is called geodesically complete if any geodesic
can be extended infinitely in both directions, i.e., if every maximal geodesic is defined
on "

45
 Theorem 3.17 (Hopf-Rinow)
Let be a Riemannian manifold. Then the following are equivalent:
a) The Riemannian connection is geodesically complete.

b)
 is a complete metric space (i.e., every Cauchy sequence converges to some
point).

If case a) and b) holds we have furthermore

c) Any two points of can be joined by a minimal geodesic.

Proof
!"
 Let 
 !
 be a maximal geodesic and suppose  . Let
be an increasing sequence  $# Since  is parametrized proportionally to arc

length let us assume without loss of generality that is the arc length. Then
!('*) ! !2'*)43!
%
&
+,-
,/.10 0 56,798 #
!  ! " !
Hence, since :; 
 is a Cauchy sequence. Let by assumption <>= 2D?AF @CB E -


and choose a neighbourhood G around < and an H>IKJ as in theorem 3.8.

!  !
For large 7 we have 
L8MG and -
 is the minimal geodesic from 

!2'*)
 to
-
 so that
!('*) ! !2'*)43!
%
&
/N
,O= #
! 3S!   ! !2'*)2U
Letting 5P  N%
&<Q3V


!
,O=R # Therefore 
N
3!2'*)
8T is a curve from

the sphere of radius  to the sphere of radius  centered at < , so that

γ(t n )

γ(t n+p )

WMX
XOY Z![9\]Z![2^*_2` is just a radial geodesic by lemma 3.12, that is
Wa a,a&jlk jlkVno opj
Z+bOcRdfehg*i Zb!mbN\ Z

W
for some mrqVs it \vuwmxuyc{z}| But then we can simply extend by
Wa aajNk j; o‚j„ƒ n
Z+bOcMd~ehg*i Z+b€m*bl\ Z |

46
This contradicts the maximality of so that
 . Similarly  
In order to prove   we shall prove
   
I. To see that any Cauchy sequence converges it is enough to see that any bounded
set is contained in a compact set. (Since a Cauchy sequence is bounded it then has
a converging subsequence, which in turn implies that the original sequence converges.)
To see this choose !#"$ and suppose %'&(!)*!,+-/. 0 12!3+4"$5 But by c)

! + 7698;:;<&>=?@)@="2A <CB
and by a) 6D8E:'<GFHA <,B I B is always defined and continuous (cf. appendix C), so
in particular

! + " K J MLN6PO36D8E:?<FQ=2RR = 0XW IYB WZ

RTS SVU
which is compact. It thus remains to prove
II a)  c). Consider [T)*!\" B )]%^&_[T)*!N`a Choose a normal bc ball dTe around
[ and let fg&ihMjkde be the shell of radius hl.mb . Since f/&(hN is compact there exists
[?n"fg&ihM with %^&_[?n)*!o minimal. Put [?np698E: e &(hq=?/) = srP)`=t"$Ae B
Claim: !u expeE&va=w . S S
We shall prove this in the following way. We put H&>xyz698E: e &>x{=?g)@x/" (which
is well-defined by a)), and we shall show

(*) %'&>|&}xy*)*!M~7a]x 1lx€"‚ƒh,)yaq„v

In particular for x †a (*) is the claim.
First notice that we always have

%'&>|&}xy*)*!M€‡ˆa‰x
in fact; aŠ‹%^&_[T)*!M %'&_[T)yH&}xyŒ~7%'&vH&>xyy)*!M x~%'&>H&}xyD)*!M* Next observe that (*)
holds for x\h . InU fact U
aŽˆ‘;’ QT“ &vT RR arc from [ to ! W
J R
‡ •–N—?”˜š ™›œ>'œ”žTŸ*M¡¢'œ(NŸ*£N¡¡¥¤z¦§¢¨'œ”ž?©Ÿ*£N¡*ª
hence 'œ”ž;©MŸ*£M¡/«z¬­¨¦,Ÿ which proves (*) for ® ¤m¦ . Also notice that if (*) holds for ®
then it also holds for all smaller ®K¯g° ¦ since then

H±(²€± ® ¯”³ Ÿy£ ³ «´|±(²€± ® ¯”³ Ÿy²Hœ ® ¡ ³ ¢¨'œ}²Hœ ® ¡*Ÿy£o¡€« ® ­ ® ¯ ¢µ¬‰­ ® ¤ ¬‰­ ® ¯>¶
By continuity we can now find a maximal ® © such that (*) holds for ® « ®©. Suppose
® ©2·¸¬ and we want to get a contradiction:
47
 Do as above: Take a small shell 
 of radius  around
 . Let  again be
of minimal distance from  . Then as above
     

     ! "
 
Then      $# 
% &      '()(  ( !  
  
*,+- "
But the curve from  to
' followed by the geodesic ray from .
' to   is a
curve of length /+0 and hence must be the (unbroken) geodesic from  to 1 . Since
it agrees with on the first part, it must be equal to , that is, 1 2
,+- and so
.  ,+-   '     '(  ,+-   "
 
Hence (*) holds for 345+ contradicting the maximality of .

p
0

γ( t 0 )

p
0
p

Examples of geodesics

Example 3.18
i) 6 with the Euclidean metric has as Riemannian connection the Euclidean con-
nection (exercise 2.22) which has 7.9;8 :=<?>@ Hence the differential equation for a
geodesic is just ACB
AE DB
<2>/F DHGJI K FMLONQP 6 @
The solutions are of course just the straight lines in 6 @ Notice that 6 is complete.

48

 

 
ii) an open subset with the Euclidean metric has of course again straight


lines as geodesics. However, unless will never be complete. Notice,
however, that if is convex any two points can be joined by a minimal geodesic
so that c) in theorem 3.17 is not equivalent to completeness.

Example 3.19
   

The sphere with the induced metric has as geodesics the great circles,
that is, the intersections of with the 2-planes through 0.

       

 i.e.,
To see this let be such a 2-plane and let be the
isometry reflecting in
%  ' &(
 "!$# )  ' &(+*
Then clearly ', - ,  has ./#0213,  as fixed point set. We shall show that .
is a geodesic. Clearly it suffices to show that if 4657&8. are close, then the smaller
segment of . joining them is the connecting minimal geodesic (which exists by theorem
3.9). So let 9 be the minimal geodesic joining  and 5 . Then  9:! is also a minimal
geodesic joining  and 5; Hence if 9 is parametrized by arc length we get <9"!=#>9 ,
that is, 9/?.@;

Example 3.20. The Poincaré upper halfplane

A-B
#/CED+#FHGJIK5L& MM 5LNPORQ
Let

be given the metric

M
 SUTWV KDX!6! #ZY []O  \ [ O  \:^

Proposition 3.21
i) The group _a`cbKdfe gihkjlnmo p o p
q rts/uu rwv qyx{zR|
uu
49
 acts on 
as a group of isometries by
     


ii) The geodesics in 

are the half circles and half lines perpendicular to the line
In particular 

is complete.

Proof
i) Identifying a tangent vector in
 ! 

by a complex number we first notice that the
inner product at a point is given by

" #%$'&(*)  + 31 2  # &  $ #%$'& 
 ,.-/0

4
Now for a matrix  65 consider the map 798:

given by 7 ;  
<=>?*<@   The differential at a point  is given by multiplication by the com-
plex derivative

7BA <   >?C<@.E

DF   G 
0
<  H

where H  =IDJ
is the determinant.
Also we compute
4   
,.-K 7 ; '  NL + M = > ? D    5  NL + M <=O?PQ >?U 0 R
DS<  PT
U

 U H  ,.-V U 


Therefore
" 7NW #%$ 7XW &(Y[Z )Q\  132^ # & _
; ,.- 7 +   ]
7 A <= 7 A < 
U U
 ;,.- 7 A ;    ]
 132  # &   `,.-a+ 0 1T2  # &   ";#%$'&( )
7


which proves that  acts as an isometry.
ii) First let us show that the half  D axis,
b dc M ee IfShg
e
is a geodesic. For this we use the reflection ij8k

given by ;lm M    DTln M  $
which is easily seen to be an isometry (not covered by i)) with as fixed point set. The
b i
50
argument of example 3.19 then shows that is a geodesic. If we write 

 and
parametrize by arc length then must satisfy
 
 
 
  
 


 

that is 
 
 Therefore considered as a geodesic through the point  is given
by

 !" #%$ &'(& $)

in particular it is defined for all * In order to get geodesics in all directions through the
point " let + in i) be the isometry given by the rotation matrix
,.-0/2143 1"57683
# 3=<
1"57693 -!/21:3;

Then clearly 
> and the differential is given by multiplication by


+ 
 1"57683@?A-0/B143 EGF CHJI 

DC

Hence3 +:
KK
is the geodesic through pointing in the direction determined by the angle
L
#NM  In order to get geodesics through any other point just notice that O9PK
QMR S acts
?YX
C
transitively on T C  In fact if UV W ZW\[^] then U_ +:
 for + given by the matrix
, X
W
;  `aEW\[b]c
] 

In all cases the geodesics are the images of under some Möbius transformation as in
i). These always take circles (or lines) to circles (or lines) and also preserve angles.
Therefore, and since the de# axis is mapped onto itself, the images will always be either
a circle perpendicular to the de# axis or a line perpendicular to the de# axis. (One might
think that it is enough that only “one end” of the geodesic is perpendicular to the de# axis,
but by first rotating by 180 f we see that also the other end must be perpendicular to
the de# axis).

Exercise 3.22 The disc model for the hyperbolic plane.

< 
Let g h be the unit disc gikjU ml U l &Ecn with the Riemannian metric
p
given by op

q%
q%rs8 w t_u@v 
@# l U l C!x C

i) Show that the Cayley transform y=zm{}| ~€ given by

ƒ‰Š'ˆ
y‚KƒR„‡†‰ˆ
ƒ‹†Œˆ

is an isometry.

51
  
ii) Show that for 
with      
the transformation
 
 


is an isometry of  onto itself.

iii) Show that the geodesics through 0 are the Euclidean straight lines and that the
distance from 0 to any other point is given by

      

!#"$
 %  ' &

iv) Show that all other geodesics are circular arcs perpendicular to the boundary circle

  ( and that for any two points *)
  the distance is given by

- ,)   
+ , )  

#"$ ,)  ).    )0/


 
where ) and are the points of intersection of the joining geodesic circular arc

and the boundary (cf. the figure).

b2

z2
z1

b1

Example 3.23. The hyperbolic n-space

In 1*243 let 5 be the bilinear form

57698;:=<?>A@CB 8EDF<GDIHJ8 3 < 3 H!KLKLKMHJ8 1 < 1

and consider N

1 @POE8RQ 1,243TS U 5 6V8;:W8X>A@CBY[Z

S
S

N
Proposition 3.24
N N N N
i) 1 is an n-dimensional submanifold of 1,243 with two connected components,
1 @ 1 ^ 1 : where _ D @`6aY*:=b[:=b[:cKLKLKd:=be>Q 1
2]\ 2gf
52
ii) For 
 the tangent space   is naturally identified with
 
 
   

and restricted to  $ #&  %  is positive  $# so that   gets a natural Riemannian
$# definite,
metric defined by "!  !  !
'  )(
$- -4  
 
iii) Let * /.1032  be the subgroup of linear maps A satisfying
,+  + 7 7 7
  :
;<
$- 
 6
 ?5 - 598  8 8
and let *   .>*  consist of those satisfying further
=+ =+ 
GF  (
?- @A5CBED BED
Then *   is a subgroup acting $- transitively on-    as a group of isometries. QPEPQPE
H
 +
Furthermore the subgroup of * 
 fixing is O  acting on I$JLKMON  (
H+ BED   B B SR
iv) The geodesics in  are all curves of the form   . < is a 2-plane
 U
 
T V V
XW V
through 0 such that is non-degenerate of type (1,1).

v)  is complete.

vi) There is an isometry of D in exercise 3.22 onto  Y given by
4 Z G_C`ba Z G _ ced Zf
Z\[] ^ + W W Y Z Y
+/ghW W
Proof
i) That  
is a submanifold follows easily from the Implicit Function Theorem. To   
 k
 QPQPQPo? :
l 
nmDji D
see that it has at least 2 components observe that , so it
 qrp
s 
 Yl
suffices to see that, say , is connected. But if then
4 PQPQPt4  
m p   Y km D p Y 
   QPEPQPE? g'+vu
( Hence m  p lies on a sphere and thus for
-lw
and = holds only for
E
B D
it can be connected to the point xy z"{}|t~,€ƒ‚&„†q„Q‡E‡Q‡E„ˆ‰rŠ Hence we can
+
suppose that { Œ‹ {  ‹ ‡Q‡Q‡ ‹ { Ž ‹ qŠ Thus we are reduced to the case  ‹ ‚
where  Ž consists of two hyperbolas.

ii) Let ‘ ‹“’ ‘ | „Q‡Q‡Q‡o„ˆ‘ Ž”X•  Ž Š By differentiation of –@z6{7„?{—~ ‹ €v‚ it follows easily
that ˜™
 Ž ‹›š { • Žœžž –z"{7„ˆ‘q~ ‹ Ÿ
ž
53
  
which is clearly a 
dimensional vector space. To see that is positive
 
definite let us assume  (otherwise

   "!#!#!"$ %
 &  .- !#!#! -
which  is  a
)( + * ,
 
trivial case). Then for &  ' & #!#!#!" & &  &
so that
12 122
//   //  //   - !#!#! - //0 3 2 4 4
& & & &798;: 3 <7;89:
65 65
>=  &  &?8 - &  8 : =  8 :
A@CB
5 5 5
Hence

//  // 0 =   -  = 8 :
A@
& & &?8 & 8: 5   
5 5
/  /  
and since = :
D@FE / /HG @ this is only possible if either &  JI or & &?8LK I .
  MI N , O 5
But if &  MI and & &?8 then clearly & MI B Hence is positive
definite.
5

 *  T S G I
iii) To prove that O @ #P8!#!#R"! Q  isU( a subgroup 
first notice that if T  S  then 8 B
In fact  ' 5  SW ' S #!#!#!"TS U(  with X S K Q I and  5  Y8 
V
 FS XS  5
8
Z@<B Then
5
 TS    - 4
8
[ S 7S7 G I
5 
since
// // 122 122
// 4 7 S 7 // 0 3
4
<7 R 9
8 : 3
4 S 798;: ]=  8 :
A@ = S  8 :
A@ G //  S  // B
/  / \5 >5 5 5
*a`  Q     cSd>_e 
Therefore, for ^ T_ @ b8 ^ satisfies
5
 ' _gf    $   (     h_i  G I
^ ^ 8
5
 *j`
that is, _ f ^ @  P8 Q B To see that O @  P89Q acts transitively
5 5 * on Q is now easily
proved using the usual Gram-Schmidt procedure  to extend Q to an orthonormal
basis for Q (orthonormal with respect to ).
The remaining statements we leave as

Exercise 3.25
Prove all unproven statements in proposition 3.24.

54
 n
H+

e0

e1

55
Appendix B THE TANGENT BUNDLE
Let be a differentiable manifold. We will show how to make the disjoint union



 


into a manifold. This manifold is called the tangent bundle for . The first step is to


make into a topological space.
More generally consider a set together with a covering by subsets    that
is,     Suppose further that for each  there is associated a topological
space  and a bijection !" $#   In this situation we have:

Lemma B.1
Assume that the following holds for each  &% ' such that )($+*-/
, .0
1. 2134(5+*6798):<;>=@?$7A?CB 

2.  *ED F"G  1 ( *H6 # *  is continuous.

Then there is a unique topology on such that for all I' the following holds:
1.  is open in

2. JKL#  is a homeomorphism.

Proof
First notice that it follows from i) and ii) that furthermore * 1 4(5+* 6 is open
in * and

(B.2)  * D  FMG N 1 O( *>6 #  *21 P(5 *H6

is a homeomorphism. Now define a topology on by stipulating RQ is open

iff for all ST  "UVL(5W is open in XZY[ . This is easily seen to define a
topology on \ and clearly 1) is fulfilled. To show 2) we fix ]_^a` and consider
bdc b b)g
X [ . If is open in \ then clearly e ["f is open in X [ Y by the definition of
b)g
the topology. We must show the converse, i.e., that if e [f is open in X Y[ then
b bon
is open, that is, for every hJ^i`'jkeHl2m X+lp is open in XYl . for this notice that
bqn c n brn g
X l X [ X l and since the map in (B.2) is a homeomorphism, clearly e l f X l is
n bsn n c
open in eHl m X [ X+l p iff e [ m X+l p is open in e [ m X [ X+l p Xt[
Y u But this obviously
b4g
follows from the assumption that e M
[ f is open in XY[ , which proves the lemma.

Now return to v an 2 w x dimensional manifold and cover it by coordinate charts

g c
f X [ j<y [ j@]z^/`Jjqy [J{ }m |"[ ~ j€€j |‚[ pƒ X C
[ „ X [Y ‚ u Then tv is covered by

56
the sets

  

and for each  we have a bijection
 

  

   

given by
 "!$# &% ('*),+ !.- %0/2131213/ )  !.- %4/ #5+ /3121213/ # 76
(B.3)
9<;
 9
#   89 # ;
where
-
,?
) >=  & / 
: + == 
 =
CB 
It is easily checked that @
  together with the covering A D and the
bijections , E / satisfy lemma B.1, so that becomes a topological space.
Furthermore the following is straightforward.

Theorem B.4
!
 /is a % differentiable manifold of dimension FHG with local coordinate charts given
by as defined by (B.3).

Now let be equipped with a Riemannian metric IKJ / JML / and let NOJPN be the
associated norm in the tangent spaces. The following is used in connection with
corollary 3.4:

Proposition B.5
Let Q SRUTWVYX3Z0[ be the \^] vector and let _ `
V [ be an open neighbourhood of
a
\XZ in Va[ . Then there is an open neighbourhood bc`d_ of the form
bdegfh XiTjVa[ kkHl TnmoPp h Xqpartsvu
for a suitable open neighbourhood m of l5w in [ and syx \^z

Proof
|
Let { m}|~o4 o3€2€2€So4‚7ƒ be a coordinate chart around l5w and let the metric be given
by the matrix „HH†ˆ‡v‰ o i.e.

v†ˆ‡ Š lq‹ eŒ7Ž† kk X o Ž ‡ kk X5 o l Tnm}| z

That is if h X e‘’†“h † Ž† kk X then

p h Xqp•” e– v†ˆ‡™Š lq‹ h † h ‡ z

†˜— ‡

57
Since this is clearly a differentiable expression as a function in the coordinates of 
it

follows that a set as defined in the proposition is an open set. Let   
be the chart

  !"
$#$#%#%
&  '("
&
$#$#$#
& *)

/32 /54
where +,.-0/1 4 76 Since  is a homeomorphism onto its image we can clearly

choose a neighbourhood 18:9 of ;%< of the form
 4
>=?A@BDC 4 ,@ 
FE +(EHGJI ?K

is a neighbourhood of ML and

where
 8
E ENPO    Q #$#$# Q R   

is the usual Euclidean norm. Now let S 8  be a compact neighbourhood of ML

of the form
4
ST = U@  4 WE VX (ZY[V\]^L_%E`ba^K

for some adcP; 6 We claim that there is an I cP; such that


e egf
+ I  E E

for all  satisfying h@bS . In that case

4

ji!  @BDC 4 WE VX (ZYkVX  L $EHGlam
e  e GnIpoq

4

where I5rI I , is a neighbourhood of ;% < of the required forme and
e 18:9 . That

I  exists follows from the continuity of the function  s + restricted to the
compact set t
4
>=u+A@BDC 4 ,@7Sv
FE !ENTw K 6

This ends the proof of the proposition.

58
Appendix C DIFFERENTIABILITY OF THE EXPONENTIAL MAP
In the proof of Hopf-Rinow s theorem (3.17) it is used that for a geodesically complete




Riemannian manifold the exponential map  is smooth.
In this appendix we shall prove this statement using the corresponding local


statement in corollary 3.4. Thus let be an dimensional manifold with connection

and metric 

Theorem C.1


Let  be the domain for the exponential map, that is, !#"$ iff there


exists a geodesic curve %'&(*)+-,./0 such that %1&2)43506879;:=< )@3506ABC,ED


:?>
3EFHGJIE/EDLK' Then MN is an open set and .OP is a smooth map.

For this we need the following:

Lemma C.2


Let Q be a manifold and let %( R /?S=TVU;Q be a smooth map. Suppose there
is a coordinate chart WXY[Z]\^`_`_`_a=ZbPc such that %d)@/EU;Q;0XY For ef"fQ let
g
<.hji'kl1monqp r!sut vwkl1myx+p rqsut
be the parallel transport along z|{4}~€]Y~‚}„ƒ†ˆ‡~?‰?ŠŒ‹ Then the mapping 
i Ž‘’vwk“t
given by ž@
›œŸž ž*¡
¡
”{@•–~€] —™˜ • ~£•¤—¦¥§•¨^~`©`©`©a~=• ~(«ƒ ~
l.š l'monp r!s@¢ ŽOª 
is a smooth map.

Proof
It clearly suffices to prove this locally; hence we can assume z|{ ‡–~=‰?Š ®­°¯Ÿ‹
¬
Now for fƒ let ± be a parallel field along |
z {³}~€´*~Ÿ}ƒµˆ‡~?‰=Š and write
 mo²4p r!s ž ¡
J· ž ¡
¡
± —¶ ž • {4}~€] ¡ ‹
mo²4p rs
l1mo²@p rqs
That is, ¥ • ¨ ~`©^©`©^~=• are solutions to thež linear system of differential equations
Ž ª
¸ »
•5¹ ž¼» z
¸ ¶ ž¼»¾½ ¹ • —À¿¤~ÂÁ×Ä'~`©^©`©^~=Åd‹
} º }

Since these solutions depend smoothly on the initial values ¥ • ¨ {@‡–~Ɓ=~`©`©`©.~=• {Œ‡~€´ and
ž ž Ž ª
since ž ž
”{³•J~]‚— ¶ ž • {Œ‰q~€] for •”— ¶ ž • @{ ‡~€] ~

59
the lemma clearly follows.

Proof of theorem C.1

Suppose 
 is in the domain for   and put    
 !#"%$'&
We must show that exp is defined and smooth on an open subset of ( containing the
line segment )*
,+ -./"#$102& By compactness of the interval we can find a subdivision
+
34
657895;:/:/:!51<=>"

and for each ? a neighbourhood @BA of 9'AC and an DEAGFH as in theorem 3.8 such
that 9I.!%"#$JLKM@N
OP:/:/:*OQ@ < & By possibly subdividing further we can assume that
both 9'AR and 9'ATSVUI lie in @GAW?XY!%:/:/:%Z\[4" & By induction we shall now show
that exp is defined and is smooth on an open set containing ) 
]+ -.^'A $_0a` The
+
case ?Q\ is obvious by theorem 3.8. So assume the statement holds for ? and we
8
shall prove it for ?cbd" & First we choose a coordinate chart e_f^g /:/:%:#gihj around k
and notice that (by possibly making f smaller) we can assume that l!2monL is defined
tvw m 8
and smooth for pq4f and nrMsutvn tlx  where e n /:%:/:%nh j 7y for some open
set yzK h containing the segment
{
8 t w
|eT
/ :/:/:#
h j ++ } - !^'AI$^~N where
€
t ++ &
+ t +

Also, we can assume that l!2monL‚=@N
9Oa:/:/:ƒO„@GA& Now put Z  f‡†ay and let
ˆ `‰.!%"%$†ŠZ ‹ be defined by
ˆ TŒ Ipn Y  m TŒ/n6 ŒL-Ž/"#$o
8 tw
where n  e n /:/:%:%nh j and nd€sutin tlx m & Now let r` h†QZ‘‹ 9 be defined
as in lemma C.2. Then clearly

(C.3) 2m’'ŒRn=l!i“'”l•#–—™˜›šiœ^œo9ž Ÿ|¡¢Xœo£ ¤I¥ƒ¤^£ ¡_¡

is defined for a¦§Ÿ as long as

œ'¥¤£ u¡ ¨N©B¤ and

(C.4) ª ª‰« ¬I­ ­ ³
œ'ž4Ÿ|¡E¢Xœo£ ¤I¥ƒ¤^£ ¡ ¤ for #® ¯°!±’œ£6¡¨Q²

Furthermore the expression in (C.3) is smooth as a function of ¥ƒ¤^£ , and  . The set ´6µ›¶
·¸
of vectors ¹,ºYR£ satisfying (C.4) is clearly an open subset,
ª
and thus exp is smooth ªŠ«
on
¬
this set by (C.3). Since »9œ½¼'¾¿WÀ%¡¨\²B¾-¤ we have that œ¼Áža¼'¾C¡1¢Xœ¹ Â|¤'Û¤¹ ÂR¡
¾ for ¼B¨ÅÄ ¼'¾ ¤_¼'¾^¿ÁÀ^Æ and
³
»9œJ¼¡ºY®l¯!°!Ç —ÉÈÉÊ^š œ_œ¼Áža¼'¾C¡1¢Xœo¹  ¤TÛ¤¹  ¡^¡
³
Hence ËR¼¹ ƒ
Â Ì ¼}¨-Ä ¼'¾|¤^¼'¾^¿ÁÀÆ̈́¶Î´6µ Since already ËR¼¹ ÂÌ ¼¨-Ä.Ϥ^¼'¾ Æö́¶Î´Lµ we have com-
pleted the induction.

60
Chapter 4 THE CURVATURE TENSOR AND
THE STRUCTURAL EQUATIONS
In this section we are concerned with the problem of determining when two Riemannian
manifolds are isometric. In particular when a Riemannian manifold is isometric to the
Euclidean space with the usual Euclidean metric. We shall see that already locally
this presents a non-trivial problem. First a few elementary facts about isometries:

Definition 4.1
A diffeomorphism

   of Riemannian manifolds
 and
 is called an

     
isometry if the differential

is a linear isometry for each 

 , i.e. if


"!$#  %&! '()  +*

We have already seen examples of isometries in the previous chapter §§ 3.19 - 3.24,
where we have implicitly used the following


Exercise 4.2

,- 
a)
Let
If

and
is an isometry then


be Riemannian manifolds and let be a diffeomorphism.
preserves arc length and distances, and maps geodesics
to geodesics.

b) If


is distance preserving, i.e., if ./
/102 3
4/ 52%2 # ./10  62 ' 0    then
is an
isometry.
(Hint: first show that geodesics are mapped to geodesics).



Let us return to the problem of determining when two Riemannian manifolds
and are locally isometric. More precisely choose points , and suppose 70   90 8: 
; 5-<  -<>=1
we have given a linear isometry

and we ask for an isometry

of a neighbourhood of 0 onto a neighbourhood of 098

/?02 # 0 8 3
 < # ; *
such that

With these data there is at most one choice for

@
61
 To see this choose 
 so that the balls    and   of radius
are mapped diffeomorphically under the exponential maps   and   onto open
sets  and   respectively. Suppose now that  "!   is an isometry and
#$&%(' . Then for )"*+, the curve
-/.1032 %4657  1. 0 ) 298 : 0;:=<+>
is clearly a geodesic with

-?.  2 " - .  2 %E' . ) 2

% @BA C D 0
D
so that /
- .1032 F
%   .G0 ' . ) 2H2 and hence
I5J  . ) 298 K
%   . ' . ) 2H2ML
Hence we have proved

Proposition 4.3
The isometry N?  !   is uniquely determined by
O%K ,P ' P R QTS II!   L

However,  defined by this formula is not always an isometry, and we shall give
rather complicated looking necessary and sufficient conditions for this.
As usual the first step is to reduce the problem to a question about connections.
So now let and  have linear connections (denoted by U and U A respectively) and
suppose VW !  is a diffeomorphism.

Notation:
Given a vector field X on we have the ZY transformed vector field X\[ on 
defined by

X ] [  %E#^5_X [`badc ] fe 8

Definition 4.4
g !h is called an affine transformation if for all ikj vector fields lnm3o on p
qrtsu r
owvRxny{z z7o}|d| v~
The following exercise was implicitly used in exercise 2.12:

62
 
     be a diffeomorphism.
Exercise 4.5

 & * let   be the corre-

Let
a) For "
 !
a coordinate   %
$#&0 ! &10 *
chart' (
& )
!
for
on "+ Show that ,

is an affine transformation iff -/. -.
sponding coordinate chart

b) Show that affine transformations preserve parallel transport and geodesics.

,
 2
Proposition 4.6

3546 87 :98;=<
Let be a diffeomorphism of connected Riemannian manifolds. Suppose
that at some point is an isometry. Then is an isometry iff is an affine
transformation with respect to the Riemannian connections.

>
Proof
?  . Then the formula
@ 7BA ; !DC F@ EG AIHKJML HONQP SR TA vector fields on 
Let be the Riemannian connection on

is easily seen to define a connection on  , and also it is easily seen that it is symmetric
and Riemannian. Hence ? is the Riemannian connection on  , hence is an affine
U
Y
:Z\[^] Let_ VW`4a]  beandthe choose a piecewise smooth curve X from 3 to V . Let
transformation.

parallel transport along b . Since c is affine it preserves

parallel transport and so d
cfe8g aFh,i'j\k cfel[ kiMmKn
d
where i j is parallel transport along c k b . Now since i j and i are both isometries, we
conclude that also cfeogBa is an isometry.

Now let ] be a manifold with connection p . We have already defined the torsion
tensor field ` for p (see d qs rTt 2.15)dlt by the formula
exercise dlq qyrTt{z
(4.7) q t ` g h u g/v w g/v"x
qyrTtr~}
where and are smooth vector fields on ] . Similarly we define the curvature tensor
field | , which to any dlqyr3t smooth
} vectordB} fields dB} associates adB} 4th by the formula
(4.8) | g h u w g/v w u g/v  u€ wK‚ gƒ

63
 Exercise 4.9
Show that (4.7) and (4.8) define tensors, that is, their values at a point 

depend only on .    
We now want to show that in some sense and determine the connection. For  
this purpose we shall use a local moving frame, that is, an open neighbourhood and 
 !""#$%&')(
a set of smooth vector fields  
on , such that for every point  , the set 

is a basis for . %"

*)+#,   , .- *)/1002 3 ""40025 -

Example 4.10
If is a local coordinate system then is a local
moving frame.

Notice that the connection restricted to  is determined by the 687 functions 9;:=>@< ?BA
given by

(4.11) CEDFG ? EHJI : @>< ?  !< K

<
Similarly the tensors  and  are determined by the functions 9 L>@< ? A and 9 MN< >O? A
respectively given by

(4.12) P& >   ? QHRI  >O< ?  <

<
(4.13) & >   ? S N HJI  N< >@?  < K
<
The so-called structural equations are really just a reformulation of these equations in
terms of differential forms.
First let us recall the basic facts of the calculus of differential forms (see e.g. Spivak
[6, I, chapter 7] or Warner [7, chapter II] ):

Differential forms
T U
A differential form of degree associates to smooth vector fields U ""  <
a real valued smooth function TLG"  <
such that it has the “tensor property” (see
exercise 4.9 above), and such that it is multilinear and alternating in     <!K
For TV WYX
an form and a =T Z U[X
form, the product is the T\ ]TQZ
form given by  U;^WX
TV]_TEZ &  <a` N
H UP^cb WY d Ifehgjilknmpoqsr tVu'vSw$xBy ujzj{|||{ wx.y~} zS\€‚ vSw$xBy~}@ƒ uFz){| |"|"{ wxBy~}@ƒ „ zS
64

 
where runs through all permutations of This product is associative and


graded commutative, i.e.   !"# Furthermore there is an exterior
 $%
&
differential d, which to any  form  associates 2
a   form ' given by
0 -  7
2 6 6 2

1 243
 , 
& -  :9 
(' *)+ ) -   ./
5 )  )8 ) ) ;- =<><
2 A
6DC 2 A;E  2 A
1 2@?BA
& -   F9 ;.9 
5  ) ) )+ ) ) ) ;-  <HG

where the “hat” means that the term is left out.

' has the following properties:

i) ' is linear over

ii) 'I'J K
 
& 
iii) L
' M#NO  P('# Q  5 #RS' 

for # a  form,

iv) For T a smooth function U' T is the 1–form given by

   
V'UT V)  )X(T
W for ) a smooth vector field,

v) ' is local, that is, for any open set Y[Z]\

F^_Y%%K ` '.^aY/%K

(i.e. '.^aY depends only on .^aY ).

 ,, 
In a local coordinate system bYcd dBeBf any  form  has a unique presentation
2Mij?l1 k_kmk_?L24n
X
hg g psrz
x yy,y&x p4u
eNoqpsrtmtmt p4u;vIw v{w
where | are smooth functions on }~
‡‰ˆ
Finally u €‚
p r tmtmt p if ƒ „ is a smooth map of manifolds ‚ and „ and if † is a
‡‰ˆ ‡
form on „ then there is a unique induced form $Š=† on ‚ such that for any
vector fields ‹+Œ, =‹Ž on ‚
yyy
 Š& †R  ‹8Œ =‹   (‘  ’/†“ ”–•=—˜j ‹+Œ • ™ ‹  •,š ›œ‘ž %
‚ ~
Š Š
yyy yyy
 Š preserves and commutes with ~
x
v

65
 The structural equations
Now return to our manifold with connection
and consider a local moving
frame    which will be kept fixed in the following. In  we have the

dual 1–forms    defined by

 "!  

where !   is the Kronecker ! , that is, !  $# for %(

& ' and !  *),+ Dual to equation

(4.11) we consider the 1–forms
/
/
(4.14) -  (.0/$1   

/

so that    and 23-  54 together determine 1   and hence the connection.

Theorem 4.15 (The structural equations)

/
:@? /
:>=  =
i) 6   798;:<-   A 8 C/ B
  

: /
? /
:>=  =
ii) 6D-  E  798;: -  - E A 8 G/ F
HE   


Proof
We shall prove only i) since ii) is completely analogous. First define the functions
I
KJL by M

I
 J  LN  .  JL  +


Then we shall check i) by evaluating both sides on  J  LPO M

Q ) Q Q Q
6RS0 J  L   T  J R L USV7W L RXY J USZ7W  J  LN US

) I
 7 T  JL +

On the other hand

[
:,^ : :
:]= ) Q : :
79. : -\   J  L   7 T . : -\  J U  L _7`-\ Y L U  J XS

) Q ) Q
 7 T -  L  J _7`-  J Y L  S  7 T 1  JL 7a1  LJ S

and bc
/RdegfhiRjhlknm oqp ik v
) /
B  =
T . /    rPsut


66
But since


             

 
 
  "!
$# 

 



 
  "!
&% The equation i) now clearly follows from this.
we have
  
-
 , /. )
'
& /
. &
. "
( 1
 * 0 and let + be a “normal neighbourhood” around ' defined as
Theorem 4.15 becomes particularly useful in the following special moving frame:

( * be a basis and write any vector 45( 62 * in the form

4 87
4

% Choose 9;:=<32 so that the 9  ball
Consider a point
follows. Let

> @BA? 45( * C D 4
#/FGIJH K 93LBM *

2 62 CC
E NPO 32
maps diffeomorphically under QSRUT onto an open neighbourhood
2IW on * such that -, &././.& V0 + is (this
orthonormal at ' ).
is possible by

Now we get a moving frame X3Y&Z/Z&Z&Y[I] \ on ^ by parallel transporting [ X Y/Z/Z&Z&Y[ ]

theorem 3.8 by choosing any metric

along the radial geodesics, and we consider the structural equations with respect to this

Consider _aX ` d
b 3
c g
e f given by _Ehaikjml Ynporqq l n5sIt evuUw_ is an open set with
moving frame:
]
coordinates jxl Yn Y/Z/Z/ZYn o w Let y{z|_=} ^ be the map y~jxl Yn|o h€SU‚ e jxl npo w
Then we have

y S\ ƒ„ h 1„ n ‹„x Œl‡Ž † ƒ „ Y y \$ˆ‰„ Š h ˆ„ Š

Theorem 4.16 (The structural equations in polar coordinates)

where ƒ ˆ „ Š do not contain l$w

i)

ii) Furthermore ƒ
„‹Œ¢ ¤Ž ˆ „ ‘ satisfy the¥ © differential
¢
” ˜›šgœž Ÿ¡m¢£œ  ¢ Ÿ¡ ¥¢§¦¨¥¢ œ
equations:

a) ’ –/“•—™

– 6
ª « ¬ 
¥ ¢ ¥ ©¢
b) –/— ˜ ¥¢®­ °¯ œ
© V± ¤ ±
with initial conditions ± —³²µ´ ˜·¶U¸  ¯ ± —¹²º´ ˜¶U»

67
  "!  
    
Proof
i) Since
! 
and since # gives the $ th coordinate
with respect to %
'
& the first equation is obvious. Similarly
3 3/6 3 3:9
 *  *  *

) ( , + -
,021354 # +7- 18354 
- /. - /.
! * C BD
9
!  the fields
are parallel along the
However, ! *curve
M ;8< =>?@:A with  E?

FG
It follows that for each HJI <K'L
vanishes along ;:< , i.e.
\ _\ _
NPO QRTSVUJXW Y O[Z:\^] Z`_ba X UcW along d Q

or equivalently e _\ \
O 8Z \ a X ] along dQ for all fhgFi8j
This proves i).
ii) We prove only equation a); the other one is similar:
By i) we have _ _'o out _
k SFlWnm Y O k ] kprqsk8p t m q k8p j
p terms not involving

_ o{z w 4.15 we have_

_ using theorem \ \ o
\rƒ X o
Similarly
w
lW@S k m Y O^v Zxw y ] k8p|}q€~ Z ‚\  X ] k8p m q ] m X k8p}„ q terms not involving k8p j
_\ X
\_
Using X
 O†v  X and comparing the coefficients of k:p in the two equations we obtain
_ t _ w w_ _\ \
v k ] q t p m O[Z‡wu] y q Z \  X ] X m
X ze
which is just a). The initial conditions follows trivially from the fact that
l g ] | is
constantly equal to ˆ .

Corollary 4.17
A Riemannian manifold ‰ is locally isometric to Š with the Euclidean metric iff
the curvature tensor ‹ is constantly equal to 0.
e
Notation
A Riemannian manifold satisfying ‹Œ is called flat.

68
 Proof of 4.17
  

       and
is trivial.

 
   "! $ we have

 #
For a Riemannian connection so are the solutions

 ' 
&%   and (%   )  *
of a) and b) of theorem 4.16. In particular for

,
 +   ,12
Thus, if  is given the usual Euclidean metric in the coordinates - /.0./.0
clearly  is an affine transformation. Hence if 34 /././.5 3
1 are chosen orthonormal then

then  is an isometry by proposition 4.6.

More generally the uniqueness of the solutions to the equations a) and b) in theorem
4.16 is expressed as follows:

6 +  798;:=< ?6 >  @A7

Corollary 4.18
 CB  @DEBF@
G G @ HJI +  H @ IK>
Let be manifolds with connections and let and be

LM' '+ #  0N > $ $@

the torsion and curvature tensor fields for and respectively. Let and

O OP QSR'T
let be a linear isomorphism. Let and be normal neighbourhoods

QSR0UWV XY[Z]\^X`_a5b"c Redgfh \ iWjklX P _ab"c R U dnmofh YpZ P

around and of the same radius (with respect to a fixed basis of respectively its

q0r qrsP asb"c R dgtufvh^w jCxzy{jE|k~}€;ji‚ƒasb"c R U,dDt„m=fvhC

image in ). Furthermore for each
and let and be the parallel transports along
‹ XŒYZ m †ˆ‡Z ‰ ZŠP
Ž
Then extends to an affine transformation iff the following is satisfied:
the linear isomorphism

m r_qr U"‘ m ‘ q“r ’•” ‡'Qor/T ‰–Qor U V

satisfies Ž Ž Ž
a. Q Ps— m r d?fvh \ m r d?˜`hu™ _ m rQ d?f \ ˜(h
‹ f \ ˜ YŒQ=r0T
Ž Ž Ž Ž
b. šFP — m r d?fvh \ m r d?˜`h ™ m r d~›Sh _ m r5š d?f \ ˜(hu›
‹ f \ ˜ \ › YŒQor/T
Furthermore † is unique and is given by
†œ_ab"c R U ‘ m ‘ ab"cR ’•”

Remark
T V m ‡'QSR'T ‰
Q"R U V
In particular for Riemannian manifolds and and a linear isometry
, the above map is an isometry iff b) above is satisfied.

Proof of 4.18
Let ž ” 0\ Ÿ/Ÿ/Ÿ5\Ež¡¢Y QSR'T be a basis and ž¤”£ \/Ÿ/Ÿ0Ÿ5\Ež¤£
the associated moving frame
in Let Z and ¥¦¨§s© ªo«¬§ ­S® be the corresponding frame and connection forms. Put

69

    
 
    
 


 
  " !
$ &(' ) +-,
and similarly consider the frame in and
%
# $ (
& ' 
) * #
.
/3#
2 

the corresponding forms and Also let and
.0/1#%2 
 :=<3> ) 465 7?9 .
A@ 5
87
-B C:D<3> ) + @ 5
7
EB !
be as in theorem 4.16 and let and be the maps defined by
.465 8739 ;
F
C:=<1> ) +HG  G :=<1>J) I + 
Again we define

F
G .  .
G 
and observe that

!
where  /#%2K# is the map 465 87?9ML2 465   4 7?9A9 With respect to the above frames the
torsion and curvature tensors have components  '
 N   PO Q 8N  and  ' 8
N   JO
Q8N 

' 8 N  ' 
Q G  and O Q 8N  O
Q8N G 

which by our assumptions satisfy

!
in # Hence, in the notation of theorem 4.16, both

R
       and        
R  
satisfy the structural equations in polar coordinates. Since they also have the same
F
.   S @
  .   .  
R
 

 

 B 
initial values they must be the same set of 1–forms. Hence we conclude that

F
and
.    T  @ .
 
  B  .   
R  !
In particular, since .  U:D<1> ) is aF diffeomorphism, F we obtain the equations
(4.19)    and    
  !
  

  

4   W9 V  
 
By the first equation in (4.19) we get
X ZY  \[ !
Writing ]_^ @   B N  N ]
^ @ 
 B
N
N
`  aN c
b `+  aN c
U
N 
N F !
and

we conclude from the second equation in (4.19) that} c   c  G

]_^ } tw
Hence

4 ` 9Wdfehgjilm?k n oqpr stvx u w gyi mzs k n{rb|~} P€s m i s k r g x gjilm?k n‚n o„ƒ
That is, is an affine transformation.

Remark
€‡ o r €‰s ˆ‹Š Z r Œ ˆ ˆ8Ž ƒ ‘“’ ˆ
It follows from the proof that
”h• – r;† i — – k 1u ˜ • ˜ i —J– q™ k š r – ƒ Hence, for all

k
70
Chapter 5 THE SECTIONAL CURVATURE

In the previous section we saw that a Riemannian metric is locally determined by

 not sound as a great advance since
does
its curvature tensor field. This

 we need to
determine the functions instead of the  functions  However, the
curvature tensor field satisfies certain symmetry relations which enable us to cut down
on the number of invariants needed to determine the metric up to isometry.
If   is a Riemannian manifold with metric !#"%$&')( and curvature ten-
 field
sor we shall also consider the curvature tensor in 4 vector field variables
$ +*,.-/102.34(.5

Proposition 5.1
The curvature tensor of a Riemannian manifold with metric  satisfies:
 
i) 6*7.-#&098 :-;<*=10>">?
  
ii) 6*7.-#&0@8 :-;A0BC*D8 :02<*=E-F"4? EG'H6IKJML.NHCOQPRHTSUVJW6HXW:YZ
 
iii) $ +*7<-/&0A3[(\8]$ 6*7.-^13_A0(`"a?
 
iv) $ +*7<-/&0A3[(b"c$ :02A3[E*,.-/(

Proof
Recall the definition

6*7.-#&0d" e fg0ih f eB0ih j elk fnmpo2q
The skew-symmetry relation i) is obvious from this. ii) Since r is a tensor it suffices
to prove ii) for sut v t v o t v where &€ is a local chart
in ˆ . In this case all vxLie
wzyV{<| brackets
vxwA}{ ‰ s vxwz~ ‰ s { o ‰ o {<are
‚x{xƒxzero,
ƒxƒ„{<Z‡†so we must show
the identity
{.|Š‹{ { ŠŒ{ | { Š

(5.2) Ž  o‘  Ž o“’  ”•s  ” ;s ’ ” Ž  Ž ” t—–

| |
However, the symmetry of the connection implies

 o˜ ” t™‰ s o t9– ” s  Ž o t9– Ž   sšt—–

| { Š { { |
from which (5.2) clearly follows.
To prove iii) it clearly suffices to prove for all s oœ›
{.|{

r#ž6s o o¡ td– q
{.|#Ÿ {
71
Again we may assume
  so that we must show
     
(5.3)         
Now since  is Riemannian we have

      

and

           !     

     " 
  #$    &%

so that


'
()
*"   
we get
but since

    
+   ,$     

   
 
by symmetry in
and  . This proves (5.3) and thus iii).
iv) now follows completely algebraically from i)-iii):
In fact by i) and ii) we obtain

(5.4)
-/.
10  &2   -/.  3
40  &2   5-/. (

  06 &2 7-/. /06
(2 
and by iii) and ii)

(5.5)
5-/.
(310  2   8-/.
(/0924   -/.  &2:0;
( 7-/. 24<
40=  %
Adding (5.4) and (5.5) now gives
 -/.
>0  &2 
(5.6)
 -/.
(  0= &2 7-/. 806
&2 !-/.  &2:0;
( 7-/. 24<
?0=  %

In this interchange
and respectively  and 2 to obtain
-/-/.; . &2:06
 75-/.
(5.7)  -/. 3
? 0=24 75-/.
&2@0   7A5--//. . 2410  3
775--//.;.  B  09243


 06B&2 24<
40=   &2@06
 /06
&2
where we have used i) and iii) in each term. Since the right hand sides of (5.6) and
(5.7) are the same we have proved iv).

With these identities proved we can now show that the curvature tensor field
- in
some sense is determined by the so-called sectional curvature:

72
 Definition 5.8
Let be a Riemannian manifold of
   with metric and let be the
corresponding curvature tensor field. For  and    a 2–plane choose 
spanning  and define the sectional curvature by
&
 '()*),+
!#"%$ - -21
/.0
- -1 & & & 1
where 3.4 "  + )+!$ )+ is the square of the area of the parallelogram
spanned by  and  .

Proposition 5.9

567 does not depend on the choice of  and  .

Proof
Let

 87":9#8<;>=*8)
?87"@9 1 3;>= 1 5A

Then clearly
B B1
- -1 B 9C8 =*8 BB - -1
 8 .4 8 " B  .0 A
/
B9 1 = 1 B
&
Now by proposition 5.1 the curvature tensor 'EDF)G'IHJKL+ is alternating in DF)G as
well as HMNK , hence
B B
& B 9C8 =*8 BB &
'(8O)?8PQ,8R)?8S+#" B '(),8O)?8O+
B9 1 = 1 B
B B1
B 9C8 =*8 BB &
" B '()6*),+
B9 1 = 1 B
from which the proposition clearly follows.

We can now reformulate corollary 4.18 in the Riemannian case in terms of the
sectional curvature:

Theorem 5.10
&VU U &[U U
Let and T be Riemannian manifolds with metrics  +XW?YZ  +Q\ respec-
 
tively and let  \ be the associated sectional curvatures. Let ]^:_ \ ^T
and `ba< c Rdfe be a linear isometry. Let g and gMh be normal neighbour-
hoods around i and i h of the same radius. Furthermore for each jlkmg nXj%o

73

 
 ! #"%$&')(!*+-,./0,. 
be the parallel transports along

  213 54)687:9;68<:13= 6?>A@ 
  )2B1 C BD

Then  extends to an isometry EGF $IH $J

iff the following is satisfied:
K L"M$
N
the linear isometry
!.O,. P  P ,R. QTS .?V H . W
FU U
N
satisfies X X
.  ZY R[]\^ ._ [` K [ba .?V 'dcde-fgh')(i
U .
Furthermore E is unique and given by
jZ
  P  P Z
i QTS
E

Proof
By corollary 4.18 it suffices to show for all kmlOn that
t u)v2wyx:z !
t u{v |;x3} !
t u~€ t!u_v vAwz| x3~x ‚ƒwz„|Rz:~ uˆ‡%‰
(5.11) oqpZrCs s s s o l†

tŒ t;u†zŽ Žu

In the following let us fix kŠl‹n and write s s etc. Then (5.11) is
equivalent to
t;wz g
t |‘} g
t ~z g
t ’€“ p  tCv vAwz|!x”~‘x:z g
t ’5“ p
(5.12) o p r s s s s  s o s

t
or, since s is an isometry
t | } t!~z R
t;wz g t ’ “ p –• v2wz| x”~z’#— ‚˜wz„|Rz:~z’ uˆ‡j‰
(5.13)  oqp r s s s s o l†

t
Again since s is an isometry clearly
™ ™
™ t!|#š tR~ ™ ž |5šm~i ‚ƒ|gz:~ u?‡
™ s s ™œ›
›
l†

 tR]}–¡v¢£xdz¤‚¥%¦ u8‡Oz
so our assumption p rŸs is equivalent to

t ~ } t;|gz g
t!|Rz ! t ~ “ p §• v2|Rz~‘x”|gz:~‘— ‚˜|gz:~ u?‡O‰
(5.14)  o p r s s s s o l¨

Thus we want to conclude (5.13) from (5.14). For this we write

© v2wz„|Rz:~z’ªx« t v |;x } tg~z !
t!wz C t ’ “ pA¬ • v2wz| x3~z’#—Zz
 oqp rCs s s s o
wz„|Rz:~z’ u?‡O‰
l†

and we notice that

© v |gz:~z„|Rz:~xj­ ‚Ž|gz:~ uˆ‡j‰
(5.15) l†

74
Now by proposition 5.1 we have

   
        
so in particular

!  " is symmetric in  and # By (5.15) and the “parallelogram
identity” we obtain

!$%&')( * +, 
or

(5.16)

  "-/.0
1  $  2 *$ , 3#
Now using ii) in proposition 5.1 we obtain

(5.17)

  &546
7   &546
1   $&-( #
But

  %2-8.9
7   "
 $:2
and

7:  2-8.9
$  "
 $:2
so that by (5.17) we obtain ;

!$ "*<( for all    =, 3# This proves
the theorem.

Remark
  +> ? @
A B C #
In case of a surface i.e. dim there is only one plane at each point
so the sectional curvature in this case is just a function

Next let us see how to calculate the sectional curvature: First a trivial lemma:

Lemma 5.18
 D E %\
FHG  I5JKMLML L IN O
Let be a manifold with connection and curvature tensor
 S W
 X
V Z
 [
Y  M
L M
L L and let
be a local coordinate system. Also with ` `9d PRQ SMMT U ` ` , write

S U
P^] -_[`ba Qc] P f\ e E
P[Q  ^P ]  Phg 8_[` E g Qc] P #

`
E g Qi] ` `
` k `k k ` [k n
Then are given by
aP g P a
E g Qi]  P I ] Q . P I Q] g 4j_Kk Kl a ] g a Q m
. a a] #
Qg

75
 Proof
Since 
    is a coordinate chart we have   for all  "! Hence
#%$ $ 8 $ 8
8 <
" 8&<
&'(*) +, -/. -0 +1)32 -40 -.   +5)
63798;: 63798;:
8 8 @
 - .
/ 8>  +=  2 -0 8 @ +
7 8 : 7 8?:
9 7"@A:
8 8 @
 &  +  8  + 8  @
7 8 : 7 8 : 7 @ :
@ @ 8 @ 8 @
2  
( + 2 + @ =   8 8 @
7 @CB : : 7 @ 7 8 B : : : :

 &  + 2D  +FE   +  2 +  E 

which proves the lemma.

Now if G is a Riemannian manifold with metric H4IJ>ILK and we again look at the
#%$NM
curvature tensor H 'OP)QR/STK in the local coordinate system
Z
   U   then
@ @ @ Z
it is determined by the functions
7*Z\[
# #%$ #
(5.19) + V WH &'(*)+X KY + V !

In general this formula gives too messy calculations for practical use even for surfaces.
However, for G ]G_^ a surface one can choose the coordinates in a convenient way
to simplify the calculations: Thus consider `baG and choose an orthonormal basis
c(dfe g/d h*i
for jlk"mon Then in a normal neighbourhoodg'{"|
prqsm centered{ at t of radius
uwv?x x~} }u
we have the geodesic polar coordinates yNz , z , in an interval
}€f‚
of length , such that
ƒ „o†‡‰ˆ ƒ | { ƒ |dfeŽ { ƒ |d h || ƒ“’ i
k yNzy yNŠ ‹(Œ y Œ‘ y p•”b–t n

Notice that by the Gauss lemma (3.11) the vector fields — and — are perpendicular,
— ˜ —9™
and since
†/‡ˆ {&dfeŽ {d h*||
z“š k yNzyNŠU‹&Œ Œ4‘

is a normalized geodesic, — has length 1. Therefore the metric › is given in the

g'{"| — ˜ iŸg¡4g ¢ „¤£ g €¦¥
coordinates yNz by the matrix –*›9œž
e4e „W£ g eh „ h'e „x g hh „§ g
(5.20) › › › ›

§?„s¨ g vWx œVl«

where — — n The inverse matrix ª"› is then
— ™
9 — ™‰©
9 ee eh h'e h4h
„¬£ g „ „x g „T£*­§ g
(5.21) › › › ›

76
and the Christoffel symbols (2.20) and (2.21) are easily calculated:



     
 



(5.22)    



   

 

  
 




 
         ! 

   
(5.23)    
              "#

  
 

Exercise 5.24
Verify (5.22) and (5.23).

We now have

Proposition 5.25 (Gauss)

  '&
Let $ be a surface with geodesic polar coordinates % and metric given by (5.20).
Then the sectional curvature is the function ( given by
 + * -,

) #
( *  

Proof
Since at every point . and . spans the tangent space, the sectional curvature is
. / .0
given by
21354   1354   3
 . . . .  . . . .   
( .99 / . 0 6 9.9 / . 087 . / . 0:6 .0 . /7
9 9
9 9

where by (5.19) and lemma 5.18

 =    A A
3 <; 3    "3       A
4     A
!
       >@?A   
6
          J
  GF

 
CB
ED
  H IB ED

 * + * , #
 

This proves the proposition.

77
 Example 5.26
Let us calculate the curvature of the hyperbolic plane using the disc model in
exercise 3.22. By (ii) in that exercise the geodesic polar coordinates at the point 0 is
given by
  where   and  
!"#$%
&(')#$ or
 '+
* .-
 ,

Then B
132 2 B B
/  
D E'FG
D 3,.
,0 2  2 ,6 798;:=<5> B B C6 B
B B
 54 8 :;? >A@
(') 
H  I. 'I
H  'F.
B B 7 7
L'+M B @ @
,6KJ B ,N=OQPSR  -

6  

Therefore by proposition 5.25 the sectional curvature is given by

2 B
T 
 ' 2 B
VN&OQPSR9WXY'Z -
NOUPSR9 
T
Notice that the fact that is constant follows without any calculations just from the
fact that the isometry group acts transitively on the hyperbolic plane.

Exercise 5.27
B]\ ^
For the 2–sphere [ of radius 1 show that the metric in geodesic polar
coordinates is given by
B
B B B&B
0 _0 `0 )abc0 IN&OQP 

>=> > T >

and show that the sectional curvature is d .

Now return to a Riemannian manifold of dimension egf+h and let us show that the
computation of sectional curvatures can be reduced to surfaces:

Proposition 5.28
Let i be a Riemannian manifold, let jlknm and let oqpCr%sWm be a 2-plane.
Choose tCprus m a ball such that v w%x sKy t{z |}pnm is a diffeomorphism onto a
normal neighbourhood, and consider the surface ~ €vw%x s(‚ t)ƒ„o† Let ~ be given
the metric induced from m . Then the sectional curvatures ‡‰ˆ and ‡‰Š of ~ and m
respectively satisfy:

‡ sˆ ‚ o!‹)‡ sŠ ‚ oE

78
 Proof
It is convenient to use what is called a Riemannian normal coordinate sys-
tem around defined as follows: Let
     be an orthonormal basis for
   
then, since   is a diffeomorphism, we have a local coordinate
  #
system  !   " determined by
*.-
*21
) ,
* +
$&%   (' $0/  43 $65
87

-
In this coordinate system the Christoffel symbols satisfy:
*=<
/>%? 7
(5.29) 9;: 36@ACBED
 # 5
In fact for F % F  "F the curve GIH
J   J  #R# - TS J S small,
K  L F NMOP*MQF *V U *
J / % J
is a geodesic through with coordinates G % GIH W F  so by the differential
J X% - ?  * <
equation (3.2) we have for *Y<
) *Y< 7
(5.30) 9;: / F F %Z? D %\[  ]^ !_

For fixed D this is a quadratic form (since the connection is symmetric) which is
*Y<
constantly zero, hence the coefficients are all zero, which proves (5.29). The components
`
:Ka for the curvature tensor field in this coordinate system are now given by lemma
5.18 which at the point reduces - to < *
*Y< *ed <d
` 9 : a d 9 d
/N% c d : a d
(4.31) :ba d gf c d
c c
and hence by (5.19) - - - k -
*=< * < k *Y<
` ` )Rkml `
a /4%ih  E /  a E k % / /
a k
: - < 0
: j : - *
k * d k < d
)Pk l d )Pk l d
% / c 9 a d / c 9 a d
d gf d
: k : k
c k c
(5.32) * < 1 dd < k * 1 dd
) k l d ) k l d
% c ' 9 a d f c ' 9 a d
- : d - : d
c *on <sn c
% c BopC Drq / f c @LpC Doq / 7
c c

Now choose the basis
    t in such that uwvmxAytz0{}|~R€ ~R‚oƒ Clearly for
„†ˆ‡ the curve ‰‹Š ŒKyIŽ‹‰ „t‘ is a geodesic in ’ so that “’†€"”  €A”  ‘ is a Riemannian
normal coordinate system for ’ . It follows that the metric in ’ with respect to this

79
coordinate system is just given by the matrix 
    where 
 is the matrix
defining the metric in  in the coordinate system     "!#
Now clearly by (5.32) and (2.20)
$&%('*),+.-0/2143 35'768+9-:1 3 53 ';68+9- < = '768+A/ < 3 = ';68+
5>@? >>@?
   <   < 
is the same for  and B which proves the proposition.
/
As an application let us calculate the sectional curvatures for the C sphere and
/
hyperbolic C space (see example 3.23):

Proposition 5.33
)
i. For E ! D !GFH the sphere of radius 1 the sectional curvature is constantly equal 1,
$ %'),+4-
& 6KJL) ) %O'*) + P
that is, >5 I !  DNM !
/
ii. For Q ! the hyperbolic C space the sectional curvature is constantly equal to –1,
F $&%(',
) +2-R/ 6SJ ) % P
that is, >5 I Q !  DNM Q ! #
F F

Proof
Let us prove ii), the proof of i) is entirely similar. Since the isometry group
T ' + $U%('*),+
>5 C F acts transitively on Q ! (proposition 3.24 iii)) it suffices to calculate
6V-XW YZ-[' + F WY
for >5]\(  ]\  and since the isometry group fixing is the orthogonal
$U%('*),+
group we get the same sectional curvature in all plane directions. Thus at least
$_^a`'),+ )b-dc e"fhg W W 3 ^k` P
is constant and we just need to calculate for 3  iDjM aQ ! #
e ^a` '),+.- c e"fh g W Y W W 3 F
Now clearly by proposition 3.24 iv) we have l@m 3 Q D    so by
$^a`'*)+ P F $_^a `'),+9-n/
proposition 5.28 can be calculated in Q Hence by example 5.26 >
F
which proves ii).

Exercise 5.34
Prove proposition 5.33 i).

Exercise 5.35
i. Let oqpr\ be a positive real number and let s0tu v B be a diffeomorphism of
Riemannian manifolds with metrics wx"yx{z and wkx"|xz~} respectively such that
€‚„ƒA„€†„‡‰ˆ*Š‰‹LŒhhŽƒ ‡"ˆ ‘ƒ ‡“’•”u–— .‘˜™’i—›š

€
Show that — is an affine transformation and deduce that the sectional curvatures œ
Š
and œ for and  respectively are related by
– ¡ ¢ €† £,¤.‹ ¥ – ¢ £,¤ £§¦¨”u– —›
œž Ÿ Œ  œ for all 2-planes
‘˜™’i—›š

80
ii. For 
 let  
  be the sphere of radius  i.e.,
  
      "!!!#  $#  &%
 
with the usual induced Riemannian metric. Show that the sectional curvature ' of
(  is constant and ' )+*, -
iii. For .
/ let 01    be the hyperbolic space of “radius 23 ”, i.e., 01  
is a connected component of the submanifold

  8  
0 4 65  
7 
     "!!9!;: <   >=
 7
with the metric induced from the bilinear form ? as in example 3.23. Show that
the sectional curvature ' of 01  is constant and that ' ) A@ -
7
Corollary 5.36
Let B be a Riemannian manifold of constant sectional curvature '
i. If ' C

 then B (E
 DGFIHIJ  * ,LK ' -
is locally isometric to

ii. If '   then B is locally isometric to  -

iii. If ' MN then B is locally isometric to 01  DGFIHOJ  * , K ' -

7
Proof
This is immediate from theorem 5.10.

81
Chapter 6 CURVATURE FOR SUBMANIFOLDS
OF EUCLIDEAN SPACE

  
In this section we shall study the curvature for an 
dimensional submanifold
 
with the Riemannian metric in induced from the Euclidean metric in
 which we 
shall denote by    As usual we shall identify the tangent space    
 
with the corresponding subspace of so that a vector field  on is identified
   
with a function    In the Riemannian connection is just given by
 
the directional derivative   for  a vectorfield on and !" any (tangent)
  
vector. We shall often consider vector fields along M, i.e. functions  
 
and for these the directional derivative #$% is defined for !&' ()   In
particular, as we have seen in exercises 2.3 and 2.22, the Riemannian connection in
 
is given for  a tangent field to and !*+, by

(6.1)  .-/#021 3546  .-780:9 

 
where 3; for ;  is the orthogonal projection of ; onto  . Now we shall
< 
also need the normal space  of vectors perpendicular to , and for ; 
 <
 we denote ; 1 ;
3; the projection of ; onto the normal space at  . Thus as
 
in (6.1) let  be a tangent field to and !=>, and consider
<
(6.2) ? -@!ABC0D1E4F#$G-/#0H9 
Then we have

Proposition 6.3
1. ?I-@,B#0 does only depend on ,, that is, ? defines a bilinear map
 K    <
?JG, , ,
 
such that for  and  smooth tangent fields on the map =L ?I-@!AM,N0 is smooth.

2. ? is symmetric, i.e.
 
?I-PO$RQS0T1U?I-PQVOW0 XYO$MQZ*, $= 
 
3. For  a tangent field on and ![\,
  G-70D1  G-/#0^]_?I-7!,B,N0

83
 Proof
i) We shall see that 
 satisfies

 

 for !"
$#%&
In fact
'
(*),+.-/
$10243$5
6.1879
:),+;-<
$ 1= 0
*
:> 3 ) + -/
$, = 0 ?
@A
 B&
This proves the first statement. The smoothness of DCE 
(5BFF is obvious from
the definition.
ii) By i) is enough to prove this for G H NM  OP H MM  3AST UVLJWJWXWXUZY=
L
I K M
a local coordinate chart in # around & In this case we have IRQ
J
H J
H
for

(6.4) ),+[\)]^_&
In fact if `S[a>E S%bc# b d is the inverse of the coordinate chart and we consider
this as a function from S a b Y into d then (6.4) is just the formula
f f
(6.5)

ehU g ZU i  Ue i  Ujg &
e e e e
Now ii) just follows from (6.4) by taking normal components.
iii) is obvious from the definition.

We can now calculate the curvature tensor in terms of the above bilinear map :

Theorem 6.6 (Gaussa equations)

For _B^klBm tangent fields on # we have
n$o n n

pqB,AklmPrs '
OBktuu'
pqm\vrw 
qBkOx
pOBmyvrB&

Proof
Since both side of the equation are multilinear with respect to functions it suffices
to prove this for   H ,z H ,k{ H  where 3vSuU}VJJWXWJW~ UhY= is a local
J
H K
coordinate system. ThenI we have HJI
Q HJIL|
(6.7) ) +
$) ] k*) ]
$) + kt
similar to (6.4) above, and also
o
(6.8)
pqB k8 + ]
6kt€w ] +,
6k &
Now
),+
$),]uk),+,
u] k7‚),+
6
Okt
(6.9)
 + ]Oƒ k 7ƒ'
(q ]uk7ƒ),+„
6
$^ktv

84
and similarly

(6.10) 

   


subtracting (6.10) from (6.9) we then obtain

(6.11) !"#$%
 '&()*  +  )*"'&
,-

or

(6.12) ).#/01&2)
 +)*  34&(  ,*+
,-"5

Since 6 is a tangent field and 
* and )*   are normal fields we
therefore get from (6.12):

(6.13) 7 "896;: ;&<7 *"6;:=>7)
?)% 61:5

But again 7,) .6;:@ so that
073"61:AB7)  )*3"61:!7, ".  6;:
(6.14)
B7) )*3"61:!7, ""-"6CD:
and similarly

(6.15) C7)
- .61:=>7 .)61:"5

Substituting (6.14) and (6.15) in (6.13) now yields the desired equation.

In particular let us consider the case of a hypersurface, i.e. and let E!FHG FJI4K
L
us suppose that we have given a unit normal vector field , that is, a smooth vector
field along E Fsuch that MONPERQH SDTVUJWXYL9NOZ
for all []\^E
and such that the length
of LNis one. Notice that at least locally such a field always exists and it is unique up
to a sign _<`a5 
In this situation the bilinear function defines the “second fundamental
form” II by the equation

(6.16)  " cbb"#,L

for  and  vector fields on E!5 We then clearly obtain:

d
Corollary 6.17 (Gauss Theorema Egregium)
1. For %"*"96 tangent fields on a hypersurface E in FJI4K  we have
7)%"/96;:A!bDbe)*$bb4)%61'&Hbb4)% ,bbA61"5
2. In particular the sectional curvature for  RSTfUgWhXjihklZG>MONaE
 is given by

m Nl,?e bbihil$bb)kl"kn'&(bDb)io"kp$q 5
7)ihil:r7kl.kn:e&07ih"kP:8q

85
 Remark
Classically the metric 

 is called the “first fundamental form” and is denoted I,
so that the above formula can be memorized as “the second fundamental form divided
by the first”. For  a surface in this is called the Gaussian curvature.

The second fundamental form can also be expressed in terms of the derivative of
:

Proposition 6.18
Let   be a hypersurface with unit normal field .
1. For all   !#"%$'& (()

2. For * a tangent field in  and +,-. we have the Weingarten equations

/ /
" $'& 0*13254 0" $'& *613274#898;:< *=0)

3. In particular for   0*  >  

/ /
" $ & * 1 2 "%?@&   1

Proof
i) Since  BADC we have
E /
2(FG  H2JI "%$'&   1

this proves that " $ & is perpendicular to  hence is a tangent vector.

E
ii) Since 0*%KA ,
E / /
2LG 0*+M2 " $ & *1ON -0" $ &P*Q-1

which proves ii).

iii) Clearly follows from ii) and the symmetry of II (i.e. the symmetry of R ).
TS
The normal field considered as a map  U V is clearly a smooth map
TS
into the unit sphere R  called the Gauss map  U R . Then as noticed
above XW3Y [Z R\2]-. so that the differential
_^a`bc.d efbhgji c[kml\n%o b-c.d

b-c.d
maps to itself. In these terms ii) above becomes
o7z3{|{v} b-c.dL
(6.19) prqtsvu'wxGy u'wx~ _u'wxB€
o„ƒ
For ‚ this gives another characterization of the curvature:

86
 Corollary 6.20
Let

be a surface with normal field   Then the Gaussian curvature 
of at is given by det where       is the differential of
the Gauss map.

Proof
Let 
be an orthonormal basis for   and let "!$#&% be the matrix for 
with respect to '"(
that is, by (6.19)
!)#&%*,+- .#/''%102436575'89.#/'%1:
Then by corollary 6.17 the Gaussian curvature is given by
  ;5 @ 5'89".:9575'@ 89':<3=575'89':  !>7!$7?3=! /
@@ <! 7 !<A @@
@ !B' !B; @
which is the determinant of  .

Exercise 6.21
a. Let C 8ED>;FG:HJI #&% K #&%.D # F % LD<7FNM O be a symmetric bilinear form. Let
OQP 
RTS @@
be the hypersurface

D>VWU C 8XD>YD$:AZ @@ D M O\[

]^M_ ! O
OP  
so that the tangent space at is naturally identified with given by the first
coordinates in Show that the second fundamental form for `*aC
at 0 is II

b. For n=2 describe the 3 different cases N`bcJ] for the Gaussian curvature N` in
terms of the bilinear form Cd and draw pictures of the standard forms of for
each case.

Exercise 6.22
Let

be a surface with normal field 
and second fundamental form
II. Let e4 fhgGiYj
be a smooth curve without self-intersection . Also suppose
575kVlYlm npo lYlm nqstv
and
r
u w tvu
mxyn{z r
for all in | }h~ oY€X
a. Show that there is a smooth vector field ‚ along ƒ tangent to „ such that for all
|?†} ~ o€\o‡7‡ˆ ‚N‰E|YŠ o lYlm n q t‹u and also ‚Œ‰E|YŠ and lYlm n are linearly independent.
b. Now define Ž t
™›š u the set ‘}h~ oY€“’ ” by •‰X– o |YŠ –1‚N‰E|YŠB—˜ƒ<‰{|YŠ and show that for some
„›œ tsžŸt •‰-– o |7Š¡ – =‰;¢ ™ o ™ Š o |‘}h~ oY€¤£
87
 is a submanifold of which touches

along  in the sense of exercise 2.12 c).
Show that the Gaussian  curvature of

 is identically zero and conclude that


c.
is locally isometric to (

 is called an “osculating developable”).

88
Chapter 7 THE GAUSS-BONNET THEOREM
In this section we shall prove the classical Gauss-Bonnet theorem stating that the integral
of the curvature function over a compact surface is 2
times the Euler characteristic
  . First a few remarks about integration:
Let be a Riemannian manifold with Riemannian metric       Then for
a suitable class of “integrable functions” we shall define the integral as follows: For
convenience we assume the functions have compact support.
First suppose that the support of the function  is contained in a neighbourhood
such that      ! is a coordinate chart. Let "#%$'&%( be the matrix
defining the metric in this coordinate chart and let )+*-,/.0"##$1&2(3 Then we define  to be
integrable if 4576 8:9<;>= )+6  8:9 is (Lebegues or Riemann) integrable in    ! and
? ?
  4 76  8:9 ; = )D6  8:9
@ ACB
(7.1) ?E
 F4HG 9 0IJIJI/ G ! ;3K L7MONQPJRJS0SJS0RN-TVUXWYN P SJSJSZWYN T

Notice that if MH[\R^]\UR_]`a[cb [edFf

T is another coordinate chart and if gVi%h j'k%l is
the matrix defining i in this chart then

L
h m+n-o0p gV%i h j'k%lqmrn-o/pt€ s\iuv j Ryv k-z7„ {
vxw vVw
N € N „†ˆ‡
m+n-o|p}i~ZY€v jƒ‚ v R x„ v k v
vw v N vVw v N
N j € „
m+n-o|p s v kX{t‰ nŠo/p s i u v R‹v z7{
vw j N N
v v
m+n-o|p sŒv N kX{t‰ L
vw
m+n-o|paMŽˆ‘“’”\•–•5—|˜
Therefore if ™›š-œ-œžˆŸ¡£¢¥¤ we get from the transformation formula for integrals that

 ’”\ªQ«­¬›® + ¯˜ ’” ª:«°  ’²”tª:«<¬V³³µ´-¶|· “’²”tª:«<¬#¬V³³»º ˜D’” ª:«

§:¨ƒ© §:¨±© © ¸¹©
(7.2) ¦ ¦
°  ’F\ª:« ¬ º ˜+’F ª:«
¼ ½¨ ©
¦
where one integrand is integrable iff the other one is integrable. Hence the integral in
(7.1) does not depend on the choice of coordinate chart.

89
 In the general case we cover the support of the function by coordinate neigh-
bourhoods
   and since we have assumed supp to be compact we can take the
covering to be finite. Now choose a partition of unity
    subordinate to
  
Then is defined to be integrable iff   is integrable for all  and we define
 
(7.3)   
   
where the right hand side is defined by (7.1) since   has support in   . This
definition is independent of choice of covering and partition of unity. In fact if "!$# ! %
is another finite covering of supp and '& ! # ! % is a subordinate partition of unity then
clearly for (*)+,  is integrable iff ,   & ! is integrable for all -.(/ and
 
0    0  & ! 
 ! 
It thus follows that 12  is integrable 34(5 iff 16  & ! is integrable 3478)9-:./
and by symmetry iff 7 & ! is integrable 3;-<=/ 
Similarly in this case
  
 0    ;  & !?  & !
   >!  ! 
which shows that (7.3) is well-defined.

Now let us return to a 2-dimensional Riemannian manifold @ and let us define

some nice domains of integration in @ :

Definition 7.4
A polygonal domain ACBD@ is a compact connected subset such that the boundary
consists of finitely many piecewise geodesic curves with no double points.
We shall decompose such a polygonal domain into some particularly simple ones
namely small geodesic triangles defined as follows: Let E BF@ and GIHCJ be as in
theorem 3.8, that is, for K)LKNMOPE there is a unique minimal geodesic of length QRG
joining K and K M , and furthermore for all KSTE:) there is a normal neighbourhood around
K of radius G . Then we will define

Definition 7.5
Let E be as above and UVBWE any subset.
i) U is geodesically convex if whenever K)LK M XU also the minimal geodesic arc
connecting KY)LK M is contained in U1

ii) A subset UZB<E is called a small geodesic triangle with vertices [\])^["_`)^[ba if

a) U is geodesically convex and contains [\c)^["_`)^[ba ,

90
 b) whenever 
 is geodesically convex and contains    then 

c)     do not lie on one geodesic line.

We shall show that there are plenty of small geodesic triangles: For this let  
be any point and let   be chosen in such a way that

 ) the minimal geodesic arc  connecting  and  is contained in  !" $#
and is not part of a radial geodesic,
%
) the radial geodesics emanating from   and joining a point on  are also
contained in  .

p
2

p
1
p
2

Notice that if just &' and &( are chosen sufficiently close to & ) then * ) and + ) are
fulfilled if just & ) , & ', & ( are not on the same geodesic line. In fact take - ./1032 a
normal neighbourhood of radius 46587 and let &' , &( have distance less than :9 from & ) .
Then ;<=&' , &(?>@5A9' and any point on the minimal geodesic B) has distance less than
:( 4 from & ) hence lies in - .?/$C

Proposition 7.7
W and 7EDGF be as above and let & ) , &' , &(IHJ2 satisfy (7.6) * ) and + ) and let
K 0Let
2 be the K union of geodesic rays from & ) to the minimal geodesic arc B) joining &'
and &( . Then is a small geodesic triangle with vertices & ) , &' , &( C Conversely every
small geodesic triangle contained in 2 is the union of the geodesic rays joining one
K
vertex to its opposite side. Furthermore is a polygonal domain bounded by the 3
minimal geodesics joining the vertices.

91
 Proof
Let
 be constructed as in (7.6); then we shall show that is a small
geodesic triangle with vertices    .
First notice that is a polygonal domain bounded by the minimal geodesics
    
joining  and  ,  and  , respectively  and  . In fact clearly every
geodesic ray emanating from  intersects   precisely once so that the inverse image
of by the exponential map   is a compact set in    bounded by two line
segments and a curve.

−1
exp (Σ)

p
1

Next let us show that  is geodesically convex. For this let  and   be two points
in  and let ! be the minimal geodesic joining them and suppose ! does not stay inside
#" Then clearly some segment of ! will join two boundary points and otherwise stay
entirely outside  , so we can assume that %$&  lie on either ' ($)'+* or '+, and ! lies
entirely outside  except for the endpoints.
Case 1. +$& .- ' ( then ! is clearly part of ' ( , which is a contradiction.

p
3

q
p
1

p
2

Case 2. /1032 465&/ 78092+: . Then by replacing ; : by / we can suppose we are in

case 3 below.

92
 p
3

q
γ
1

p
1
q p
2

Case 3. 
 , 

It suffices to show that  stays in the same angular component as  . In fact it can
never intersect   , since in that case either  would be part of  or   , or we would
get new intersection points on   as in case 1.
Now  has only two possible angular components to stay in, and for  close to  
and close to  it must be the same component as for   . Hence by continuity the
component must be the same as the one containing   for all and  . This shows that
 is convex. That  satisfies b) in definition 7.5 ii) is obvious from the definition.
 a small geodesic triangle with vertices       .
Now conversely given
Then clearly by convexity the sides       are also contained in
 Furthermore

the set  consisting of all geodesic rays joining   to   is then also contained in
 and clearly satisfies (7.6). On the other hand we have just shown then that
  is geodesically convex. Hence by 7.5 ii) b)     that is,    This ends
the proof.

 !
Proposition 7.8
Every polygonal domain
 #"""$ &% 
can be triangulated into small geodesic triangles.

%
That is, there exist finitely many small geodesic triangles such that

1.
'
  *( )   (
2. Any two triangles intersect in at most one common vertex or one common side.

Proof

  
First observe that by proposition 7.7 any interior point in has a neighbourhood

 
  , + 

which is the interior of a geodesic triangle. In fact choose a neighbourhood
of as in 7.5 above, let be any point, and let be any point
opposite  
with respect to . Then for any geodesic through different from the-
line   .
and for any choice of     -
on sufficiently close to and on opposite sides,
the points      
will determine a geodesic triangle containing in the interior.

93
 P
p q
1

Next for each point on the boundary we have a neighbourhood as in either of

the figures

q q
q

It follows by compactness that we can cover
by finitely many geodesic triangles

such that i) the interiors cover the interior of
, ii) the boundaries cover the boundary
of
.
Now if two triangles  and  intersect then we subdivide the union   
into smaller triangles using all necessary new vertices, in order to get a triangulation.
Adding one triangle at a time we eventually obtain the desired triangulation.

With these preliminaries out of the way we can now prove the Gauss-Bonnet theorem
for a small triangle:

Theorem 7.9 (Gauss)

Let  be a small geodesic triangle with vertices  and opposite sides
   respectively. Let    
  be the angles between the geodesics
    "!   respectively. Let # be the curvature function. Then
$
#'&  )(  *(  ,+.-/
%

Note 1
Here % 0 # & 13 0 2 %54 # where 2 % is the characteristic function for  , that is,
% 7 6 8&:9 for 6<= ; > % 7 6 8
 &
6 =
?/
2 2 2 for

Note 2
The angle between two geodesics is just the angle between their tangents in the
tangent space at the point of intersection.

94
 C
β

A=p

γ
B

For the proof of theorem 7.9 we shall consider polar coordinates 
 with respect
to the point  Notice that since  is “small” it lies inside a normal neighbourhood
of and is the union of all geodesic rays joining to the opposite side  .

Recall that the metric is given by

    !
  "

where "# %$%&(& '  &

&('*) is a positive function. A priori " is not defined for
+, .
However, we have

Lemma 7.10
1. 1-/24
./0 36587 "9:
;6<= uniformly in  .

&? @
2. 1-/2>
./0 35 & 1 
;6A  uniformly in  .

&;B ? @
3. 1-/2>
./0 35 & 1 B 
;6A  uniformly in  .

Proof
ON TSVUXW*Y
Let CDE
GFIH*JKML
%J!.QPRML  then at 
;66Z "8
;6[ is the length of
the vector
d+e
$]\_^X` W
);acb fhgjilknmKoqp_rsutwvIxy]z{

g
where now y|} { are the polar coordinates in mX~*z . Now since ]€_ƒ‚ ~„† v depends
continuously on ‡ kˆm[~z and since  €X‚ ~*„ ‰‹ŠŒŽ‘ mX~*z ’ m[~z we get for given
“•”—–
that
™ ™
™ ™ “
™ ]€_X‚ ~ „( v y:š›{ ™œŸž š ž ž š ž ¡ šMknmX~*z¢}
(7.11) ˜ ˜
˜ ˜ –ƒ£
˜ ˜u ‡ close to

95
  
  
Now for


    !
#
so that " #$ "
%& Hence we conclude from (7.11) for '
#$)(
#
" " ** + ,.- 0/13245 ** 
(7.12) * *7698;: for  close to < &

  + ,
=> + ,
Clearly i) follows from (7.12). Next recall from proposition 5.25 that

(7.13) 
and since > is clearly continuous at zero (converging to >@? ) iii) follows.
Now for A CB A  we get from (7.13)
 + , - !/1324  + , - D/E32
 I FHG > + , K 0/
 A  A
FHJ
(7.14)

hence letting A ML < it follows for fixed that

NQEPRO S  +  , - 0/132
exists and by (7.12) the limit must be 1. Hence by (7.14) we conclude for A small

 + , -  /ET2
=UV I F G > + , K /

and arbitrary

 A
S
(7.15)

which clearly converges uniformly to 1 as A WL < . This proves the lemma.

Next let us investigate-

+ , along the geodesic X ( For this we parametrize X by
arc length Y and we let Z Y 2 denote the angle between # - X 2 and the radial geodesic, i.e.
the angle between the tangent vectors []\ and Q at X Y
[]^ #
dα
ds

σ(s)

α
θ
p

96
 Notice that since a radial geodesic intersects the geodesic only once the polar



coordinate is increasing in  , so we can express the angle  as a function in along
. Then we have:

Lemma 7.15
Along the geodesic we have
   

  
          

 

Proof


Write  in polar coordinates , that is,



  "! $#&%'  )( *,+ $'-/.  0(&12 
*      1 3
 54
Then the equations (3.2) for the geodesic give 8 with :
 1 * 8;* :  
(7.16)  1 +7698;:<   >=
  
or by the identities (5.23):
 1   
1
(7.17)  1 @A? BDC 
 B E
The angle   is determined by
   

(7.18) #F%"'B HGJI  K ML  
 L 
L
LONQPSR9TVU
since both vectors have unit length. Differentiating this we obtain using (7.17):
 1  B  
1
 
(7.19) W'0-/.X ZY    1  A? K  C 
   E
but since is parametrized by arc length
  1 
1

C  + C  
 E  E ?
so that by (7.18)
  1 
1

(7.20) '9-[. 1   \#F%"' 1     C   C  
? ? KE KE
Hence we obtain from (7.19)

   
1
  
  Y Y   A? B C 
   E

97
    

or


(7.21)        
This gives the lemma.

We now prove the theorem:

C
σ( A )

σ(0)
A
B

Proof of theorem 7.9

   !#"%$
&('*)+,'.-*/1032546879 '*) :
As above let be parametrized by polar coordinates at the point

  ;$  
and choose the basis such that is the tangent vector of . Then

.<=  >.<= ?  @

for points in , varies between 0 and , and given , varies between 0 and
where are the polar coordinates of the unique point on the geodesic .
Now by (7.1) and proposition 5.25 we have

A AEG; 4 F AE F AEG; 4 F AE X - F
B C " I HKJMLONP CRQ S T" U VHKJWLONP X Q  - S 
D
AE ; 4 YZX F
" U XQ  S .<[\ ? ]U_^.`  by lemma 7.10
AE ; 4 acbed b f
" b fVgh i by lemma 7.15

jlk m g don k mqp]r don\s p

j k m g k t r n>u r k v p
j km g kt g kv r u
which was to be proven.

wyx{Next
z beweawant
polygonal domain. The boundary |
w by definition consists of finitely
to extend this theorem to more general polygonal domains. So let

many closed curves, which are broken geodesics. Each of the finitely many non-smooth

98
points of the boundary 
 we call a vertex and to each of these we can attach a
well-defined interior angle
 where  "!$# The quantities
(7.22) %&')(*+ (,-.%&/0-  1 !
are called the exterior angles.
We shall also need the notion of the Euler characteristic of 243

Definition 7.23
Let 2456 be a polygonal domain triangulated into finitely many triangles as in
proposition 7.8. Let 7 be the number of vertices, 8 the number of edges, 9 the number
of triangles in this triangulation. Then the Euler characteristic :; 2=< is
:; 2=<+>7?(*8@A9B#

Note
From Topology it is known that :; 2=< is a topological invariant, independent of
choice of triangulation. In particular for 2C6 an oriented 2-manifold
:;D6E<F;GH(JIK<
where I is the genus of 6 . This in turn is the number of handles, which should
be attached to a 2-sphere in order to obtain a manifold diffeomorphic to 6 (by the
classification of surfaces this is always possible).

Theorem 7.24 (Gauss-Bonnet)

Let 6 be a 2-dimensional Riemannian manifold with curvature function LM# Let
2C5?6 be a polygonal domain and let %+
N% be the exterior angles at the vertices
of 2 . Then
O W
LQ?RS:T;U2V<'( %&Z#
P YX

For the proof we need the following trivial lemma:

Lemma 7.25
[ Let [B 2 \ 5 6 be a polygonal domain triangulated into small geodesic triangles

 . Then for any integrable function ] on 2
\
O W O
]) ]a#
P ^X
_ `

99
 Proof
It is easily seen from the definition (7.3) of the integral
that it is a linear functional
on the vector space of integrable functions. Now the function

(7.26)  



is zero outside the boundaries of the triangles  . In a coordinate neighbourhood the

boundaries of triangles are clearly sets of measure zero, so the integral of the function
(7.26) is clearly zero by the definition. Hence
   
       



  

Proof of Theorem 7.24

Choose a triangulation of  as in proposition 7.8. Let  "!!!# %$ be the small
geodesic triangles in this and & #!!#!# &#' all the vertices. Let & !!!" &)( +*-,/. be all
the vertices on the boundary 0 . This includes the vertices 1 #!!#!# 12 of 03 (i.e. the
non-smooth points), but there may be more. For each +4 65 7 #!!!)98 we name the
3 vertices :;4 9< 4 9= 4 considered as vertices of +4 . Then by lemma 7.25 and theorem
7.9 we have
 $  $
(7.27)   :F4HG < 4IG = 4JKML 8 
?> 4  A@B> 4 CED D D

In this sum each vertex & ON   !!!"P. may occur several times as a vertex in
6R
different triangles.
For each interior vertex the different angles clearly add up to QL whereas for & 0
they add up to the interior angle S at & . Hence
$ (
: G < 4+G = 4JT S G . * JUQL
4 C D D D  C
(7.28) (
VQL . LWS J * L
 C

Notice that since the boundary consists of closed curves the number * is also the number
of edges in the boundary for the triangulation of  . Then a simple combinatorial
argument gives
(7.29) X 8 /QZYV * 

In fact counting the 3 vertices of all triangles gives a doubling of a vertex every time
two triangles have a common edge, i.e. at all edges, which are not part of the boundary.

100
Since there are 
 of these the formula (7.29) is proved. Inserting (7.29) in (7.28)

gives
    
 
$
(7.30)
       !#"  %&
(
*) for each + 10 65 
with
%Hence
' (7.27)  reduces to -,/. ,3242323,
(7.31) 78:9     "
 
 $ %& <; >= 
 $ %' 5
 
Notice that if + is a smooth point of the boundary then the exterior angle is 0 so (7.31)

is just the required formula in theorem 7.24.

; 
Remark
Notice that it follows from theorem 7.24 that the Euler characteristic
?= is
independent of choice of triangulation for .
=
Let us make a number of corollaries.

Corollary 7.32
For @
a compact Riemannian 2-manifold with curvature function
9
7A 9 B; @  5
In particular for @ oriented,

7A 9 DCE 0
F 
where F is the genus of @ .

Definition 7.33
A Polygon in @
is a polygonal domain homeomorphic to a disk.

Corollary 7.34
@ 9 H
= G 
 @  beP 0 a polygon with interior
) ,423242I, )J K ,3242423, KJ'LNM = 5 Let O>= 8 be the area of = . Then
Suppose has constant curvature and let
angles at the vertices

 J ) R
T> <V

Q >S UO?= 9 5
101
In particular for a triangle with vertices 


 


Proof
Immediate from theorem 7.24 and the fact that for a polygon !
Corollary 7.35
Suppose " has curvature function  satisfying  #%$'& then there cannot exist a
polygon with 2 vertices.

α2

α1

Proof
In fact if ( was a polygon with 2 vertices with exterior angles )+*-,.)0/ then 1324(65879
and so :
7?>A@CBD2)+*8EF)/G53HDI
;=<
contradicting I.
< J
Exercise 7.36
Let KML'N O P be a smooth map of Q8B dimensional oriented manifolds. Suppose
N is compact and P is connected. Then the degree of K is defined by
RTS-U 2K587 V
WYX[Z]\A^`_badcAegfihjk
where lnmpo is regular value (i.e. is non-singular for all rm l
xy depending on being orientation j]qgf preserving or reversing.jsuth kgk
and
egfvh4j6k8w
1. Suppose { and
j8qzf o have Riemannian metrics |[} and |Y~ respectively, and define
for rm€{ v‚dƒ i‚ ƒ+†
r | ~3‡ ŒŽ'
„h j6kh k+w j]q-ˆ-‰zŠ.ˆd‹
where  
• { andˆ •—Š-‘Ž– ‘Ž_ ‘dŠ.c ˆd’”o “ and ˆ ‹ -
Š Ž
‘ Ž
‘ -
‘ 
Š ˆ ‹ “
are positively oriented orthonormal bases for
’ i‚dƒ that fori‚-any function ˜ on o
t
™ t › integrable
respectively. Show
f f h k ™
˜š ˜œ
} h 
j k 4
h 6
j 8
k w 4
h 6
j k ~

102
2. Suppose

is a compact oriented submanifold with normal field  Show
that the Gauss-map   
satisfies
     
 
and
"! at # %$ 

103
 Chapter 8 LOCALLY AND GLOBALLY
SYMMETRIC SPACES

We have seen that in some sense a Riemannian manifold is determined locally by its
curvature. In particular manifolds of the same constant curvature are locally isometric.
In this section we shall show that this is even true globally provided the manifolds
are simply connected. This we shall show in the more general context of symmetric
spaces.

Definition 8.1
1. A manifold with connection

and associated torsion and curvature vector
 
fields and , is called locally affine symmetric if 1) 
, 2) is parallel along 
geodesics, i.e. if satisfies:
For every pair of points   let   denote the parallel transport
along a geodesic from  to . Then

 "!$#%&'(%)*!+#,&# -*!' &.0/

2. A Riemannian manifold is called a Riemannian locally symmetric space if
with the Riemannian connection is locally affine symmetric.

Remark
One can show that if 
is parallel along geodesics then  is parallel along any
curve (see Helgason [chapter I §7]).

Exercise 8.2
a. Show that a manifold with affine connection is an affine locally symmetric space
just 1) and 2) above are satisfied locally.

b. Show that a Riemannian manifold of constant sectional curvature is a locally

symmetric space.

The phrase “locally symmetric” is justified by the following:

Proposition 8.3
A manifold with connection

is locally affine symmetric iff for all points
 there is a normal neighbourhood 12
such that the map 3  412 
157698;:=<?>A@CBEDGF  )G#H6JIKBLDMF  JNOG#
is an affine transformation.

105
 
In particular a Riemannian manifold is Riemannian locally symmetric iff 




is an isometry for all  .
Proof
Clearly the second statement follows from the first and proposition 4.6 since

 
  
 is an isometry.
Let us prove the first statement:


Consider   !
  
#" Then clearly for $&% and ' parallel a)
and b) of Corollary 4.18 are satisfied. Therefore 
is an affine transformation.
For ( we shall use the following lemma:

Lemma 8.4
Let  be a manifold with connection ) and let * ,+.-/+/0  be a smooth curve.
For 132 a smooth vector field along * the covariant derivative along * at 40$% is given by
5 7 B
1 7 CEDGFIH
6 7 =?A<@ > ? 1 ? J1K:.L
4 298;: :
D
where ? KKMON :/P Q RMN ? P S is the parallel transport along * .

Proof
ForD FIgiven UTV% let WX2-%ZY[4\Y]C`_ be a parallel
_3c vector field along * such that
H H
W0:^ ? 1 ? " Choose local coordinates -ba.aba.- L around * % and put def g as
g.hji
usual. Then nqpsrt u psrt u r~#€O
nwvfx3yzS{ | v
Wk2lSm3n$o | for
|=}

By the mean value theorem

pE€ƒt p „t‡†ˆ€R‰ psr‹ŠŒt r‹ŠŽˆ „ €‘ 
z or v
oO‚ o‚ ‰ ‚ for some
u p t
nA• x xK›bœ
Therefore the ’G“ th component (with respect to the frame ” ) of – ˜E™G— šI–
—G — “ is
 pE„ t p „t  p`€£t € ‰ p¤r Š t pE„t ¥^
€Ÿž z €¢¡ or
o ‚ “ ‚ o ‚ “ ‰ ‚ “ ‚
} }

Since ¦ is parallel along § we have n

‰ † ‰ „ r~ª€
or { n §r | z v
‰ ‚ n ‚| ‰ o for all
|©¨
n n‡¬  €­ „ pE„t p „t
zQ« v œ
where § § Therefore, for —– ˜ o ‚ “ ‚ converges to
n }
‰ pE„ t pE„t‡† ‰ p „t
{ n §r | r
n ‚| ‰ ‰} ‚
| ¨ }

106


which is exactly the expression in local coordinates of     (see proof of lemma
2.8).

We can now finish the proof of proposition 8.3:

To show that is parallel amounts to show that for a geodesic and 
parallel fields along the vector field  is also parallel along . Now if

is any point of we can assume   "!# and we just have to prove
$ 
 
% &
' 
(8.5)  () *!,+

Now choose a normal neighbourhood - . around  such that /0.213- .54 -6. is an
affine transformation. By lemma 8.4
$  7 B 7 7
 
% &
' 
 ()  8 >@?9;:=< A
> K  NMPO
J  >  L > J >  KS TMVU
/DC"EGFIH KQ/PCERFIH
7 7 F F
 8 >W9;? :=< XA >  > K > J > J[\
/ZY E IF H E I F H
F F
so in order to prove (8.5) it is enough to show that for small /*] ! the parallel
transport from ^_`6Ka/\ to c b  "/d along takes  > to  > (because in that case
E
>  > e >  J > Xf > " >  ). Equivalently weF must show
E FIH E FIH E E I F H
F F F
(8.6)      +
E,g Eih Ekj E g h j lmg h jnporqts
Now since /0. is an affine transformation we have

a) J/0.fuvwD/0.fuxyz{/0.|u }vkxy
o o

b) J/0.fuvwD/N.fuXxcW/0.|u */0.|u}vRXxy
j j

for all vkx~ + For ^pb a) immediately gives €KvkNKxcD‚K vkxc ,
j_norq's o o
hence *! . Therefore in order to finish the proof we shall just prove that b) implies
o
(8.6) above. This obviously follows if we show

(8.7) /N. u ƒK 1 4 +

E orqts ok„Ns
& &
For this let  be a parallel vector field along  a Ka/'/0‡+ Then since /0. is an
n*† & &
affine transformation, the field /0. u    is a parallel field along the curve 4 € K  ,
& &
or equivalently Ka/0.fuˆ   is parallel along   , KQ'/ /0‡"+ But
n*†
F
/0. u  K :=‰ 1 . 4 .
o s o s
so  
Ka/0.|uT     )  KQ/0.|u  *   ( ) +
F
Hence by uniqueness of parallel transport
 z`  &
Ka/0.fuT" for all KQ/'/0‡"+
nŠ†
F
107
In particular 
 gives (8.7), which finishes the proof of proposition 8.3.

In order to study global properties of symmetric spaces we introduce the following

notion:

Definition 8.8
A connected and complete Riemannian manifold  is called a Riemannian globally
symmetric space if at each point  there is an isometry    such that
   
  !" for all  #%$ &('

Exercise 8.9
Show that the unit sphere )+*-, */.+0 and the hyperbolic 1  space 23. * (example
3.23) are Riemannian globally symmetric spaces.

Example 8.10. Siegel4 s upper half space

This is a generalization of example 3.20: Let Let    denote the set of complex
*
16571  matrices. For 8 #  *   let 89 denote the transpose of 8 and tr 8 is the trace.
Let
* , *   be the set of symmetric matrices 8:
;=<?>@ where ; and @ are
real and @ is a positive definite matrix.
* is clearly an openI subset of set ) *   of
all symmetric matrices, which is naturally identified with ABCA DFEHG ' The tangent space of
* at any point is thus naturally identified with ) *   and we define the Riemannian
metric at a point 8=
J;K<>L@ # by
*
(8.11) M/N OQPR
S!7TVUW&XY[Z]\^_Y`Z]\ abdcfe ^gea hjilklm n o

Exercise 8.12
1. Show that (8.11) defines a Riemannian metric in k"o
2. For prqtsLu=h k define vxw k3y k by v]mLz{n|}u3~ z€u3~ qpJe z‚h k"o
Show that v defines an isometry and observe that v]msƒ„n |}pq?sLuo Conclude that
the group of isometries of k acts transitively on k .

3. Show that k is complete. (Hint: this is a direct consequence of ii).)

4. Show that the map †_w k7y k defined by †‡mz{nˆ|Š‰€z Z]\ is an isometry keeping
the point sƒ fixed and with †Œ‹`| —id at that point.

5. Conclude that k is a Riemannian symmetric space.

108


   
6. Let be the set of matrices of the form

  


..
.

Show that
 in the induced metric is isometric to Euclidean  space via the

('*)+", 
isometry given by

"!  ! $#&%

'-) +/.
...

0
132 146587 ':9 9;<

0
7. Let be the set of matrices of the form where , .
'>= 
Show that in the induced metric is isometric to the hyperbolic plane of radius
(see Exercise 5.35).

8. One can show that the submanifolds and are geodesic, that is, whenever 0
two points are in the submanifold, the whole geodesic line joining them lies in the

point
'2
submanifold as well. Given this, show that the sectional curvature of
is not constant provided  @?3
at the

Remark
There is the following analogue of proposition 3.21 1) for , which we shall state

ACBED:FHGI J KML:D*FHGNI J
without proof:

O
The real symplectic group is the subgroup of consisting of
matrices satisfying

OQPSRTOVUWR
where R is the FXGZYFHGN[ matrix

a R\U^]`[ _
fCg D*FHGNI J acts as a group of isometries ona hc _ bed
by

O D:ijJkUWDmlMionopqJrDmstiunov8JxwzyI{i}| h O~U ] ls vp
bd
where

Notice that the isometries considered in exercise 8.11 ii) and iii) are of this form.

A Riemannian globally symmetric space is obviously locally symmetric by propo-

sition 8.3. For the other direction we shall prove the following:

109
 Theorem 8.13
Let be a connected and complete Riemannian locally symmetric space. If fur-
thermore is simply connected then is globally symmetric.

Before proving this theorem let us recall a few facts from the theory of covering
spaces (see e.g. Greenberg-Harper [1]).




 



Let be a connected topological space (we do not assume to be a Hausdorff
space!). A mapping
  
is called a covering of if each point in has


 

an open neighbourhood such that is a union of open sets of each mapping



 

homeomorphically onto under . Also recall that for a path connected space , the

 "! #$&% '"
 (# ) %*) *+

fundamental group a base point, is the set of homotopy


, %
classes of loops with is called simply connected
or 1-connected if .

The basic theorem about covering spaces is, that for any connected, locally con-


-.


nected and “semi-locally 1–connected” space , there is a universal covering space

 
  /

015 
4 
, that is, a covering with simply connected. Furthermore is unique in
 
the following sense: Let

be a base point and for each covering
4 
234   
   5 
let


be a point over . Given a different universal covering
lying over there is a unique homeomorphism

67

such that
with
6   8 

 9 4 
and such that the diagram

;: <12


  between the points of  = 
and the group of homeomorphisms 6>

covering the identity. This group is
commutes. In particular there is a 1–1 correspondence

called the group of deck-transformations of  and is, in a natural way, isomorphic to

?   + @

 is clearly an open mapping, so we can identify
with the
quotient space 
?A
+ On the other hand suppose B is a group of homeomorphisms
 locally connected etc.  ...) space
such that B acts
properly discontinuously on
, that is, for each 7
there is a neighbourhood  
of some simply connected (and

of  such that C6 DEGF for all 6IK H JMLN,6NOB + Then the quotient  space B A

has fundamental group isomorphic to B and the projection
. BPA
is the universal
covering.

QR 
R ST 
Now suppose is a connected Riemannian manifold and let be
a covering. Then clearly can be given a structure such that is a local

R
diffeomorphism, and the Riemannian metric pulls back to a Riemannian metric on
.

Proposition 8.14
a. A curve U ! VWYXZ'P R is a geodesic iff \[] is a geodesic in ,

110
b. is complete iff
is complete,

c. is locally symmetric iff
is locally symmetric.

d. If  
is the universal covering then the deck transformations act as

isometrics, and the action is properly discontinuous.

Exercise 8.15
Prove proposition 8.14.

Remark
In particular the universal covering of a complete Riemannian locally symmetric
space is also a complete Riemannian locally symmetric space, hence by theorem 8.13
a Riemannian globally symmetric space.


 
On the other hand, suppose we have given a group of isometries of a connected
Riemannian manifold The action is called free if id, implies that has
no fixed point, and it is called discontinuous if no orbit has an accumulation point.


Proposition 8.16
The action of on is properly discontinuous iff it is free and discontinuous.


Proof

  
If acts properly discontinuously on then clearly the action is free. Suppose

   !"" # !$

it is not discontinuous. Then and a sequence of different elements of

%&(' *)" +,C -.+/#0 1

such that That is, for any open neighbourhood of we can find 2


2)243- 65  8 :
7 9
and 
1;<-.2= )>#
different (actually infinitely many) such that
so that 1?#A@B# 
Hence
which contradicts the

DE
proper discontinuity.

F G7IHKJ
Conversely suppose the action is free and discontinuous and let be arbitrary.

*L;MNO6=QP RDRNSG7T9 VUGW F$Y[Z

Then

and we consider a normal neighourhood # of  of radius X

^
 C
\\ such that #G@ 4#] that is, there is a !+S# such that ! S#F 0 Then
. Suppose there is a

L;_N`4=0abL;_N`!?=&cBF L;<!`&=  L;MN`!?=&cBL;M!dOd=eX

hence  id by the definition of , which shows that the action is properly discon-
tinuous.

It is now straightforward to prove the following:

111
 Proposition 8.17

a group of isometries
 
 is in a natural
a. Suppose is a simply connected Riemannian manifold and

  is the universal
of , such that the action is free and discontinuous. Then
way a Riemannian manifold such that the projection


covering with as the group of deck transformations.





b. If is the universal covering of a Riemannian manifold and is the group of
deck transformations then is isometric to

Exercise 8.18
Prove proposition 8.17. (Warning: Remember to show that  
 is a
Hausdorff space.).

Once we have proved theorem 8.13 we conclude from propositions 8.14 and 8.17:

Corollary 8.19

 

Every connected and complete Riemannian locally symmetric space is isometric


to a quotient of a simply connected Riemannian globally symmetric space
by a group of isometries acting properly discontinuously on

Remark
Thus if one wants to classify all complete locally symmetric spaces, this should
be done in two steps:
I. Find all simply connected Riemannian globally symmetric spaces.

II. Given a globally symmetric space find all isometry groups acting properly discon-
tinuously.
The first problem was completely solved by Élie Cartan. For details see e.g. Helgason
[2] or Wolf [8]. The second problem is quite hard and in general unsolved (see e.g.
Wolf, op.cit.). For spaces of constant curvature this problem is called the “Clifford-
Klein space form” problem.

We now turn to the proof of theorem 8.13, which follows from the more general

Theorem 8.20
Let  and be connected and simply connected complete Riemannian locally
   !#"$&%'( )%*',+-
./0 13254768.:9;'<"
symmetric spaces. Let a linear isometry. Then
there is an isometry iff
= ?> "!@AB"DCFEG"IHJK" > = > @ALCFEMHE N,OQPRS5ST@ALCUVHW$%*'X
= =
where and  are the curvature tensor fields.

112
 Furthermore is uniquely determined and is given by

      "!  #$

Remark 1 #  % &(')
'*
Clearly
+
theorem
,.-0/ $
8.13 follows from theorem 8.20 by taking
and

Remark 2   
' '*
Let 13254 and suppose that and has normal neighbourhoods  6 and
 * 6 of
radius 1 (i.e. the exponential
# 
mappings
 are diffeomorphisms
%7  * onto 6 and 6 ). Then
theorem 8.20 is true for 6 and 6 by corollary 4.18.

The second remark is essential for the proof of theorem 8.20, which will be divided
into a few lemmas. First a trivial one:

Lemma
#
8.21 # ;
% &98 : %
Let and be connected Riemannian
'< # manifolds and two
isometries. Suppose that at some point
0'=>?8@'=  ?8 $
(*) and A A
 8 $
Then

Proof # '
It suffices to show that the set BDC of points , for which (*) '
holds, is both open
and closed. 8 By continuity it is clearly closed, and if (*) ' holds
$ for then by proposition
4.3, and agree in a normal neighbourhood around Hence the set B is also open
which proves the lemma.
# %
Now let and be connected Riemannian 'E #
manifolds and let% $
be ' an * 
isometry
F'=
of
an open neighbourhood 6 of a point # onto an open set in Let .
:IH &KJKL ;M#
We
now want to extend to all of by extending
N+' $
along curves. So let G 4
be a piecewise smooth curve with G 4
QPRSH &JTLU
We say that is extendable along UPV
G if
there is a family
%E&
O 4 of isometries of open neighbourhoodsR_
P>,3PX* ] 6O ^ of G onto
open sets of such that if W W is sufficiently small then 6OZY[6\O and
*    &
`OKa 6OZYb6 O `O a 6OIY"6O
 &  a $ &  a $
andd8 such
&e 
that 6>c 6 Rc `O 6\O is called a continuation of along G
If O O is another continuation of
PfgH &JTL
along G then, by the argument of lemma 8.20
above the set of 4 for which
QPVh>?8 QPVV  +8 
`O G O G and `O AjilkmOUn O AXi kmOUn

113
is all of 
 . Hence for all  
  and  agree on the connected component of

containing  . So if continuations exist they are in some sense unique.
Now let  and be complete locally symmetric spaces and let  be an isometry
"! &!('
of a normal neighbourhood of a point #$% onto a normal neighbourhood 
'*)
# +,#- .

Lemma 8.22
)
Let /.0
102 3
4 # , be a piecewise smooth curve. Then  is extendable

along  . Furthermore there is a real number 5*67
such that for all  we can choose 
to be the normal neighbourhood around  of radius 598

Proof
) ' ' '
Let : length ; . The sets <>=@?(ABC=DE#&GF3HI:EJLK%  resp. <>= ?>AB = E# F%HI:EJMK

N are compact. Hence by theorem 3.8 these sets can be covered by a finite number
' R
of open sets O (resp. O ) such that for som PQ63
the normal neighbourhoods of
'
radius HIP exist and contain O for all =STO (similarly for O ). Let 5U6%
be smaller

than any of these P . We shall now extend  along  such that VWXY
1 is the
normal neighbourhood of radius 5 and center Z[ .
Let \^] be the supremum of the \L_
1 such that  is extendable along `?B
a\ .
) )
Put #  bcMd
 and =e bf# g*hMij\^](8 By completeness there is a point

=(]/k such that

V,#lm2n= ]  for G2 \ ] 8

' '
Then clearly AB = ]  # so we can choose neighbourhoods
XioHI: O and O as above
around #prq and =>] respectively. Now choose \1stiu\^] such that
' w

# mNO  =amNO
for all x\^syF$xiz\ ] 8
A-v# mc#p q i_5M AB=em= ] i_5

By remark 2 following theorem 8.20,  pg{ has an extension to all of the normal
neighbourhood | of radius 2 } around ~g€ . But if ‚ is the normal neighbourhood around
~ ‚ of radius }9ƒ then clearly ‚y„k| for 1†X‡jˆ*‡k^‰ and the extension of Š r€ agrees
with Š‚ on Z‚+ƒ‹^†t‡jˆ*Œ ‰(Ž Hence Š is extendable along ‘f’“ ƒb ‰b” and if ‰ ŒY•–Š
is extendable along  beyond ‰ by the same argument. This shows that ‰*— • and
ends the proof of the lemma.

Now let cƒ˜–™Z’ “ ƒ1• ”Zš › be two piecewise smooth curves such that

œ “D — ˜Dœ“4 — ~ and œW•( — ˜4œr•( —3ž

and let Š‚ and Ÿ‚ be the continuations of Š along  and ˜ respectively.

114
 Lemma 8.23
If and
are homotopic then  and  agree in a neighbourhood of 
Proof
That and
are homotopic means that there is a family   of piecewise

smooth curves with !#"$%&'()+*,-. (
/ such that the map
0&1243 15 is continuous. Let 67 be a continuation of 8 along % so that
 7 9 7* 4 7 9 7 
We shall show that the set of  :;< such that
)
(*)   and = agree in a neighbourhood of  
is open and closed. By the argument of lemma 8.21, this is clearly closed. So now
suppose (*) holds for >@? and we shall find ACBD such that (*) holds for EFHGI?JE KLAM
By lemma 8.22 we can find NOBL such that we can suppose that J7P is defined on the
normal neighbourhood Q 7 of RPST15 of radius UVN . By uniform continuity we can find
AWBX such that
Y  P 15J6 15&JKN OZD16Z[ and E ?\GI<E KLAM
for

For such  the normal neighbourhoods ]!7 of %15 of radius N is contained in Q 7 so

JP7 JQ^7  is a continuation of  along % . By uniqueness of continuations J7P and 67
agree in a neighbourhood of 15 . In particular 6 and 6 agree near  .
This proves the lemma.

Proof of theorem 8.20

As above choose normal neighbourhoods ] and ]=_ around " and "<_ and let
 *a` ]b3 ] _ be the isometry with  * dcfe)g8 according to the remark 2 following
the statement of the theorem. For any point h$i choose a piecewise smooth curve
7
` ;<J3 i with f[" , &'j , and let  be the continuation of  along
. Define

>0k@=02
Since i is simply connected this is well-defined by lemma 8.23. Clearly El]:-
*
and in general if  7 m] 7  is a continuation of  along then En]o>p= It follows
*
that  is locally an isometry.
Similarly using 8Sq
 esut 3 e i we get a map  t 3 i extending  q 
`+r r ` *
and also  is a local isometry. Now clearly Dv6 is defined by extending the identity
of ] along curves in i , so by uniqueness wv!#yx{zR Similarly v!-yx|zR which
ends the proof of the theorem 8.20 and hence also theorem 8.13. Notice that we have
implicitly used the following:

115
Exercise 8.24
Let 

  be piecewise smooth curves in a Riemannian manifold and
suppose  and  . Show that if and  are homotopic through
continuous curves then they are homotopic through piecewise smooth curves.

Corollary 8.25
Let  !
#"%$&'
be a complete connected Riemannian manifold of constant curvature
(*)
Then the universal covering is isometric to
( (
a. The sphere +,-/ .10 2 3/465798':<;9= ?>@ if $A .
(
b. Euclidean space  if B .
( H )
c. The hyperbolic space CD-/ . 0 E 2GF if

Proof
Let  2 denote the above mentioned standard model for a complete manifold of
(6)
constant curvature Notice that  2 is simply connected. Also by proposition 8.14
we can assume  simply connected. Now choose any points IKJA
LI MNJA 2
and normal neighbourhoods OQPR
SOTPVU of the same radius. By corollary 5.36 there is an
isometry WXYOQPZ OQPVU and in particular the condition of theorem 8.20 is fulfilled for
[ )
\W^] P Hence W extends to an isometry of  to  2 .

Exercise 8.26
a. Show that lemma 8.22 remains valid even if  is not complete.

b. Now let  be a locally symmetric space and _ a simply connected globally

symmetric space and let `a% b   be the universal covering. Choose basepoints
[
IK
g c
J 
g X
I b J  b with ` I&
b d
 I and I MeJf_ and suppose a linear isometry
[ji [lk [!mnco!pLqrls1tu/v6s rls1tNsvDwyxz9{}| 
P9  P U _ satisfies hL
 { ‚„ƒ for all
q u†n ‡ s z n&o‹ŠShow Œ z |
that there is a unique local isometry ~€a with ~  ~^ˆ‰ ˆ‰
{
c. Show that ~ in b) is an isometry iff is complete.
{ ƒŽu|
(Note: ~ in b) is called the developing map for in

Exercise 8.27
{
Let be a complete locally symmetric space.
{
a. Show that if is simply connected then the group of isometries acts transitively
{
on .

116
 
 every linear isometry      satisfies
  !    "#$ for
b. Show that if for some point
all " %
& ' then  has constant curvature

Exercise 8.28
For (a real number and )+*-,
let /.002. 1
be the standard model for the
)43 (
dimensional simply connected manifold of constant curvature (see corollary 8.25).
Let 5denote the group of isometries given in the 3 cases by:
a. 5 7  6  )8:9; .
5< >=  )? , the group of Euclidean motions of 1  that is, for @
1 and A

6 )BDC
E= )B is given by C GF HI 8JA F  F
1K
b.

c. 5>L6  9 M)B$N (see proposition 3.24).

1. Show that in all 3 cases 5 is the full group of isometries of  .OK

2. Show that 5 has a natural topology as a subset of some Euclidean space such that
P ) 5 is a topological group, i.e., the map 5RQS5T 5 given by  CMUVH C UVWYX
Z ) isThecontinuous of 5 on  . is continuous, i.e., the map 5RQ[  given by
 C F action F
\ C is continuous.
3. Show that if ]:^L5 is a subgroup acting without fixed points on  . then the
action is discontinuous iff ] is a discrete subset of 5 .

4. Conclude that the complete Riemannian manifolds of constant curvature ( are, up

to isometry, all manifolds  of the form  !]4_  . , where ][^`5 is a discrete
subgroup acting without fixed points on /. . In particular for (<aJb , ]/^26
dc 8e9;
is a finite group acting without fixed points on f 1 . K
Xg%h

117
Chapter 9 LIE GROUPS AND LIE ALGEBRAS

For the further study of symmetric spaces we need the basic properties of Lie groups
and Lie algebras.

Definition 9.1
A 
manifold  with a multiplication   is called a Lie group if
a.  is a group,

b. The map      given by   is a 
map.

Examples 9.2
a.  is a Lie group.

b. The quotient group  is a Lie group.

c. Let "!#  be the set of !$! real matrices and let %'&"!# )(*!+  be the
open subset of non-singular matrices with the usual matrix multiplication. Then
%,&!#  the general linear group of order ! over  is a Lie group.
c- ) Similarly the general linear groups %,&./!+  over and %,&./!+  over are Lie
groups ( and being the complex numbers and the quarternions respectively).

d) 0 2!34(5%,&"2!# 6
1 the subgroup of orthogonal matrices, i.e., matrices 7 satisfying
78"7597:78;9< makes a Lie group, the orthogonal group.

d- ) = !>?( ,% &.2!+ @ the subgroup of unitary matrices, i.e.

 matrices 7 satisfying
7:AB9 7 8  9 7CD makes a Lie group, the unitary group.
e) EF&.2!+ :(5%,&./!+  or EG&"!+ :(5%,&"2!#  the special linear groups of matrices
7 with det 7H9JI , are Lie groups.
e- ) EK0L2!3M9N01!33OPEG&"!# ;EK=B!>M9N=!>#OPEG&"!#  are Lie groups.

Exercise 9.3
Show that examples a) - e- ) are Lie groups.

Definition 9.4

119
 and 
are called isomorphic if there is an isomorphism
 
Two Lie groups

, i.e., a group isomorphism, which is also a diffeomorphism.
Exercise 9.5
Show that there are natural isomorphisms of Lie groups:  
Exercise 9.6
Show that if is a Lie group and if a subgroup  satisfies

1.  lies in the center of "! that is, #%$'&($#*)+$-,./!0)+#1,.!
2.  is discrete, that is, every #2,3 has a neighbourhood not meeting 54768#:9;
Then the quotient group  is in a natural way a Lie group.

Exercise 9.7
a. Show that if =< and ?> are Lie groups then also <A@2B> is a Lie group.

b. Show that  @  > >

Definition 9.8
be a Lie group. A subgroup C  is called a Lie subgroup if C
Let

smooth structure, such that i) C is a Lie group and ii) The inclusion D C E
has a
is
an immersion of manifolds.

Warning
If a subset F  G of an HI4 dimensional manifold has a smooth structure of
dimension JLKMH , such that the inclusion D F
 E  G is an immersion, then F is
not necessarily a submanifold, but what is called an immersed submanifold, i.e., every
point in F has a coordinate neighbourhood ONP!RQS in G and a neighbourhood T in F
such that i) UWVYX < [ZW\&^]_]_]`&aU%bc[Zc\& d for Ze,1Tf! and ii) Q
 TM V @3dg b is a
coordinate chart on F . We do not know that T can be chosen as NihjF . Furthermore
a subset F ^G may have different differentiable structures such that the inclusions
are submanifolds as the following drawing shows:

Example 9.9
Let kml npo n and qsr k the image of a line in n with irrational slope. Then
qut is a Lie subgroup, but q is not a submanifold in k . In fact q is dense in k .

120
 Proposition 9.10
Let be a Lie group. Let 
denote the connected component containing the identity
 . Then 
is an open submanifold and is a Lie subgroup.

Proof

is closed since components are always closed. 
is open because every point
of 
has an open neighbourhood diffeomorphic with an open ball in a Euclidean
space, i.e., a connected open neighbourhood in . Hence 
is an open submanifold.

is also a subgroup: In fact, fix  
; then 
is connected since  
is a diffeomorphism. Since    
  
we obtain  



Exercise 9.11
a. Let be a Lie group and
the identity component. Show that
is normal in .

b. For  "!$#&% show that 
' ()'!*#+% and -,.
'/ ,102

c. For 334657!*#98 % find 
and :,;
.
d. For <='!?>;8#&% find 
and -,;
.
Notation
For @A we shall use B B
C-D AEF8 C !GH%9I1
J -C D AEF8 J C $! H%9KLM
B
Notice that C and
J C are diffeomorphisms of .

Definition 9.12
B A vector field N on
CPO N"QRN C Q (respectively J isPC O "called left invariant (respectively
N QRN"Q C ) for all S8TUVU
right invariant) if

Proposition 9.13
There is a natural 1–1 correspondence between WYXZ and the set of left invariant
smooth vector fields on given by: N a left invariant vector field corresponds to
NUX[WYXZ"
Proof B
Since a left invariant vector field N satisfies N C N C X\ CPO !*N"XP%8
it is clearly
determined by N"X . It suffices to show that every vector N"X6]WLXZ extends to a smooth
vector field N on by B
N C  CPO !*N"X^%
121

For this it is enough to show that thus defined is smooth in a neighbourhood
of
, since for , 

is a neighbourhood of and 
   

)
Choose a coordinate system 
 ! around   Then it is enough to consider "# $ ) " 
* ,+ .-&-&-.0/  Now also let 1324
such that 5 671 implies  89
 We shall$&%(show
'
that :<;5=  is smooth in 1 for any >?+ &-&--.0/A@

B ; =DC  E FD . "  BD; =C G "HB ; =JI  C K ;5L BD; =JI  C 
K
Here ;5= I  D  M;5=  8N   and since

1POQ14R

given by  5  TSR N  is smooth and since ;5= is smooth also $ 'U BD;5=V 8N   C is smooth
$&%
where ;XYW -&-- ;5ZY are the coordinates ;[W -&-- ;5Z used on the second copy of 1 in 1PO\1 .
This ends the proof.

Proposition 9.14
Y 
]  Y_^ W
Let and be left invariant vector fields on . Then also the Lie bracket
W is left invariant.

This follows from the following more general situation: Let ` @ a b c be a

gih d m0x e f on c are called
d:jDkl gnmio p
f jDk m l g
smooth map between two manifolds. Two vector fields on and
related if for all krqtsvuwj c

gAy e b c
Lemma 9.15
Let
c ig h
on . Suppose d€
as above and let
and f5€
are related,
gih dpz&{|d} ‚ o ƒ {0„ x
be vector fields on
Then also
eG{~f[z&{~f}
dpz{0d†}_‡
vector fields
and fXz{~f}(‡
are related.

ˆŠ‰
Proof

d‹z{0d†}~‡jDkŒl ng moo dpzŽj<d†}jkm l ggnm|m‘m‘h h d†}’j dpz“jDk†m l gng m m|m

o df[zŽz j j fJf }’} jjk k m m l l g:h fd }j} fXj f z&jz k jk m m l l g
o fXz{~f}(‡Djk m l gix

Proof of proposition 9.14.

122
 Take 

 
 A vector field  on is left invariant iff 
and
 
!"#$% &')(+*,-./
i.e. iff  is %0 related to itself. Therefore if 21 and 3 are left invariant vector fields,
&4 is 0 related to itself, 56
87/9 , and so by the lemma also : 21;<3>= is %?0 related
to itself, i.e. left invariant.

Definition 9.16
The Lie algebra of is the vector space @ABC
DFEG with the Lie product
@B.IH@A-. J @B.
given as follows: For K/LM!@ABC let N and L N be the corresponding left invariant
vector fields. Then : O/L+= is the unique vector in DPEG such that the corresponding left
invariant vector field is Q R N  LTN SU That is, : O/L+=WVX
Q YN  LTN SU
Definition 9.17
A Lie algebra (over ) is a finite dimensional real vector space with a bilinear

:Z0[\0]=_^ H J
operation

satisfying
1. : K`K=
ba c  
:d: K/Le=-Gfg=ihj:d:kLlGfm=B<O=nhj:o:kf]<c=-<LT=%
pa O/LlGfq
2. (Jacobi identity)

Proposition 9.18
@
For a Lie group, (G) is a Lie algebra as in definition 9.17
Proof
This is immediate from proposition 1.20.

Notation
We shall often use the notation
r@A-.<
@-. s t v
u
 s  ^#swJ v
Let us now determine in the examples above. First notice that if

 and L
is an open set and if we identify a vector field on with a function
(cf. remark following definition 1.15), then given two smooth vector fields
the Lie bracket is given by
(9.19) : O/L+=P
xCyz-L{0xC|mB}
x y{~ €
D
where denotes the directional derivative, that is, the differential evaluated at .
(9.19) follows by direct calculation and is just the fact that the torsion in Euclidean

123
space is zero (exercise 2.22).

Examples 9.20
a. Consider

   and let us calculate the Lie Bracket in the Lie algebra
 
  For   
  
 so !"
$#&%(' Hence a left invariant
vector field is just a constant function )  +* ' Hence by (9.19)
,
)-.0/1
3254 6.58792":;)<
>=?'
,
That is, @A"B
 with C ) ./D
= EF)-.G' This Lie algebra is called
commutative.

b. Consider H
>IKJ6LM ON3PQ6LR S'T is an open set in PQ6LM GU WV  so we shall
identify the tangent space of at every point with PQ LR ' In particular @A"R

X
H
YPQ6LM  and )ZFP[LR  corresponds to the left invariant vector field
\
)^]F
`_0)-'

(Note that G]a )+
b_) so G] )
_)C' ) Given )CS. cP[LR  we must
compute by (9.19)
d \ \ \ \
)e .00
f
>2H4-
g h .jki 7l29:5
g h )m<i '
\
Here the differential of . is given by
n \ n n
.[
 _0.5R
Q _j8op.  for . fixed 

so
tju-v ^
w xzy!t{v`|`y}w~ypt
2Y4<g qar<s s
Similarly 9€
p‚ r0z
w uCv>|>yptYy}wmƒ
s
Hence „
w† 0
t ‡ v>|`y}w~yptYˆl|`ypt‰y!wŠv`|5‹ wzt‰ˆ9t wŒ
s s x
v[‹wmt‰ˆ9t wŒŽx ƒ

It follows that the Lie bracket is just given by

wCSt v>wzt‰ˆ9t0w
(9.21) 

using the usual matrix multiplication.

124
 


Notation


   with the Lie product (9.21).

Next let us notice the functorial properties of :

Definition 9.22
A homomorphism of Lie groups is a smooth map  of the Lie groups 
and  , such that  is a group homomorphism.

Proposition 9.23


of Lie groups   induces a Lie algebra homomorphism
a. A homomorphism
    given by the differential of  at "! That is, 

#$ %'&(%*),+ $-#.%'& ,%*).+
/0%'& 1
% )2  !



b. In particular if 4
35 is a Lie subgroup then
 3  is a Lie subalgebra,
i.e., if %'& %*)'2  then also $ %6& %*).+2  !

Proof


%72  and 8 #,%92  and let % : respectively :8 be the corresponding


left invariant vector fields on  , respectively  . Then % : and :8 are <; related, in fact
for =>2>?A@  and BC2D


H

H

%: E =*F G K
HPEIORIQ % =*FJ
 ORQ EL,
% =
H NM H ELOST Q #.%U = V :8 ELS H = H ORQ
where we have used that 
WF EI ELS FX# since 'F E ELS FY<!
Now let %6& %*)Z2  and 8[& ,%'& 8) #.%*) and let %: \ ]:8 \ be the
corresponding left invariant vector fields. Then since %: \ and ]:8 \ are ^; related it

lemma 9.15 that _^%6: & %*: ).` and _a[:8 & :8 ).` are <; related. In particular for
follows from
=b2b?A@ 



$ %'& %*)c+ = V $ %'& %*)c+ =dFJ
G $-8e& 8#)c+ =
$-#.%'& ,%*)c+ =
so indeed

$ %'& %*)c+ $-#I%'& #.%*),+f!

b) is immediate from a).

The basic theorem about the interplay between Lie groups and Lie algebras is the
following:

125
 Theorem 9.24

   be a subalgebra. Then there is a
  such that 

Let be a Lie group,
unique connected Lie subgroup

For the proof we need the Frobenius theorem about integrable distributions:
   
  " !#%$&!')( *,+-.
In general let be a manifold of dimension . A distribution of dimension


on is a collection of dimensional subspaces one for each

/102(4343534(6/87 /902:*;(5353434(6/87<=*   ! *
Furthermore is assumed to be smooth in the following sense: Locally we can find
smooth vector fields such that span for each in a
neighbourhood.
/ >
 ? @ /A(;>CB 
The distribution is called integrable if whenever and are smooth vector fields
belonging to (i.e. takes values in then also belongs to . The following
is straightforward:

Lemma 9.25

D  
FE D
a) is integrable iff is locally integrable, i.e., iff every point has a neighbourhood
such that

is integrable.
D
/904(5353434(6/ 7
b) is locally integrable iff every point has a neighbourhood with smooth vector

D
fields satisfying
1. /105=* 6(535353G(6/ 7 =*  span !?H5IKJL M&*N+
D
2. there are smooth function O;QSP R +UTCVW (XY(6Z[(]\?
^'(534353G(; ( such that

_ 7
@ / Q (6/ R B
0 O QaP R / P 
P]`
The notion of integrability is relevant for finding integral manifolds for  . By
this we mean a connected  dimensional immersed submanifold b % (that is, b
has a differentiable structure such that the inclusion Z"cdb e f g is an immersion cf.
the remarks following 9.8 above) such that h&i j kml"i for all npoqjr We state the
following theorem without proof (see e.g. Spivak [6 Vol. I chap. 6] or Warner [7
§§ 1.60 – 1.64]):

Theorem 9.26 (Frobenius)

Let be a l st
dimensional distribution in g l
and suppose is integrable.
u g w l
g k wyv x<z u w u
Then there is a unique foliation of with all leaves as integral manifolds for .
That is, a disjoint union of leaves , each of which is a maximal integral
manifold for l . Furthermore the leaves u w are unique. Also, there are coordinate
charts {}|~€ƒ‚ for g such that

€Y{}|„‚ƒkq{tG~;‚†ˆ‡5‡5‡‰†Š{t~‹‚Œ Ž?~ ‘ ’~

126
and such that each 
 is a union of sets of the form
  
   !#"#"#"$%'&  &)(
with * ,+ .-0/21 #"#"#"$%34 satisfying  *!6587!9
Remark
 we obtain coordinate charts for :
of the form ;< ;   !"#"#"!< >=   =
Notice that
where  is a coordinate chart for @ as above and BA2C
DE one of the sets
 <?
defined there. This determines the differential structure of :
uniquely.

Proof of theorem 9.24

O Let O F be the distribution in G given by FHI JK$L
span and P
O O A MN>G 9 Clearly if
P are the corresponding left invariant vector fields
 #"
then P
#O !
" !
"   O   #" # #
" $
"
P span FEQ 9 Hence F is smooth and it is also integrable by lemma
9.25 since  #" ! #
" !
"  R
O OHSTNU W V ^`X _ba X
* XZY[]\
for some constants `^X _dc so that also
\
egf f U W V X f
a ^ c a _h ^i_ a kX j
XZY[ \
Now let l be the corresponding foliation and let m be the leaf of l through nj
tvu6wyx{z$|H} u~w z uK} z
For any oqpsr we clearly have |w }
U U u~w
u w<‚>ƒ u w ‚>ƒ
permutes the leaves of l . In
so by the uniqueness of the foliation it follows that
c [ U m U m j It
particular for o€pm u~m wŠis‰ a leaf containing o#„ o n c hence
t{Ž x that m is a subgroup. It remains to showt x that the multiplication
follows Ž in m is †ˆ‡ .
Ž notice that c c
mŒ‹ m for opm is smooth. Also choose a chart
c< around n in r as in theorem 9.26 with  n U and let ‘“’ be chosen such
For this first
[ ” c o•
that ” „ o•p Ž– for all Ž“ ™ p2‘ jt Put x ™Zš t x šNŸ – Ž–£¢¥¤~¦
U—˜ p [ ˜ U2##žU ˜ U ¡ c ‘ U
¤'§ˆ¨ª©§ Vœ› ¦ © ¤ © ¯
¤±¯ ©§>¯
so that are neighbourhoods ¤½§ of « in ¬ If¤'§Šweêchoose ©ÅÄ and ¤'such§ that ­6®
and ­°® are ²³ balls
¤'§Ãs©§>¦
in ´ then °
­ ® are ¶
µ
¤½§ ³ balls in D
· º
¸
©Ì§ ¹ so in ©Ì§
particular ©Ì§they
¯
are connected. Now given »¼ we have ¯¾KÐ ¿ÁÀÁ ¬ and ¾K¤½¿§ À> ¤'§contains
¦ :
« Æ
»ÈÇ)É,» so that ¾ ¿ À> That is, if »'Êyˈ¼ then »]Ç)Éyˈ¼ . Since ® Ê°­ÎÍ is
a chart for ¬ it follows that the map ®Ï»'ÊyË » Ç)É Ë is smooth in ¸ Hence the

127
theorem follows from the following:

Lemma 9.27
Let be a group with a
 structure such that left translation is
 and is
 Suppose
connected.  is a neighbourhood of  such that the map  
 is smooth. given by
Then is a Lie group.

Exercise 9.28
a. Prove lemma 9.27.

b. Prove the uniqueness of  in theorem 9.24.

Theorem 9.29
Let and  be Lie groups with Lie algebras and respectively. Let 
be a Lie algebra homomorphism. Then there is a neighbourhood  of  and a

map !"#  such that
1.
$%&' (  )* +-,/.1012 % + 

2. 4365 ' 87

Moreover'if9
:9 ;  are Lie group homomorphisms with <365
'=9 365>
then @?BAC ?BAC where D is the connected component containing E7

H ' 
Proof
First notice that the Lie group FG has Lie algebra with Lie product
I J  H JLKNM  H M4KPO<'QIBJ  NM  O H I JRKSM4KPO 7
Let T H be the set
'VUWJ HX YJZ\[[ J  ]
and notice that is a subalgebra: Indeed, if
J^M  then
I J H_ J`aM HX MO4'bI J^MO H I  Jc  MO<'bI J^dMO HX I J^MO 7
Hence by theorem 9.24 there is a unique connected Lie subgroup ecT\f> with
Lie algebra 7 Let gihj4kblZm n k be the projection. Then clearly
oFprq hEs t!jWuvnwk
is a Lie group homomorphism and clearly o@x j n is given by

(9.30) oyxEzY{}|_~{Z€@pf{ ‚{`ƒ „

128
   
 
Hence 
  is an isomorphism so there is an open neighbourhood
 of
  such that   "!$#% # is a diffeomorphism onto the
# open neighbourhood
of . Now let   be the projection onto and put
&(' 01#32
*),+.-/
Then since is a homomorphism clearly
& 54768 '9& :4; & :6< 4=>6?  476? @2
for with
&  ' 
Also
A
)? 
+.- so by (9.30)
&  CBD '  EBGFH*BD ' H*BIKJLBI 2

 A

This proves the first part.
& AM 1# &  ' M  ' H
As for the second part, let  with

 and suppose
is connected. Let N N
N O  P1Q!R# N O EST ' ESU & VS=W
X  Y1Z![# X EST ' ESU\M]5ST^
N N
Then clearly O and X areN immersions and N
N N O
5BD ' EB_ *H BD ' X
5BD 2
O ' X &`' M
Hence by theorem 9.24 and so .

Corollary 9.31
#
If and are Lie groups with isomorphic Lie algebras, then they are locally
%a a#
isomorphic. I.e., there are neighbourhoods and a diffeomorphism
& 
 with
& E476b 'c& E4; & :6< J4U\6? "dKegfCh 476i @2

Proof
Obvious from theorem 9.29.

As another application of theorem 9.29 one can show (for a proof see Spivak [6
chap. 10 problem 8] or Warner [7 theorem 3.27]):

Theorem 9.32
#
a. Let H
be a connected  simply
and
#  connected Lie group and let be another Lie
group. Suppose 7j j & is G & ' H 2
a Lie algebra homomorphism,
# then there is a
unique Lie group homomorphism  with

#  # 
b. In particular if and are simply connected Lie groups and
# j and j are
isomorphic, then and are isomorphic.

129
 With this the following proposition is straightforward.

Proposition 9.33
a. The universal covering of a connected Lie group is in a canonical way a Lie
group.





b. Two connected Lie groups and have isomorphic Lie algebras iff the universal
covering groups and are isomorphic.

Exercise 9.34
Prove proposition 9.33.

We shall prove theorem 9.32 in the following special case:

Corollary 9.35
Let be a Lie group and  )+ * Then there is a unique Lie group homomor-
phism     !#"%$'& & ( +


,   ./ be given by ,102 * 034 Then

Proof
Let us identify 
with and let -
by theorem 9.29 there is a real number 57698 and a differentiable map :%<;.5>=?5?@A
3CBEDGFH * 3 CBIJ3 FH whenever KLBKM=K FNK3=OKPBQDGFNKSRT5=
with
1.

2. && U ( VV (W " *  <" $ & & ( ) * 4

In order to extend  to we write every FX in the form

F *QY Z '[5 \ D^] = Y  =_K ]K`R [5 =

and we define
b
* a  3 1
 H
5 c [ Jdfeg3/]h Yji 8
3 FH 3< ;.5Hc [ <kldmeno ]` if
if
Y Rp8q
The uniqueness of  is contained in theorem 9.29.

Notation

The homomorphism above is called a one-parameter group with infinitesimal

generator . We also define

(9.36) r?s%tuvw * 3<xIH

130
 Clearly 
 
 for all   The map


 "!#
is called the exponential map for  .

Proposition 9.37
For  a Lie group there is a left invariant connection $ on  (that is, %'&(
*),+-. ) such that the geodesics through / are precisely the one-parameter groups.
In particular $ is complete.

Proof
Let 1024353536798 be a basis for 
:; and let 1 < 065353534=;< 8 be the corresponding
moving frame on  of left invariant vector fields. Let $ be the connection given by
L M L
the equation (4.11) with >@? ACBDFEG that is, IK H J < NE for O< P< any left invariant
vector fields. $ is obviously left invariant. Let us show that a one-parameter group
Q !  is a geodesic: Let RSUT7VWUX and let  < be the associated left invariant
XY[Z
vector field. Since 
:\^]G_`a
b\c"de
 we get
f f
f 
\g*ihkjlnmo f
Ee*ph"jlnmo:q r< jlsmo
 
so clearly t
f
f f  *I H w_xzy|{~},€. ƒ‚ „†
 vuu m
† † u
which shows that ‡ˆ‰Š ‰9‹ is au geodesic.

Remark
If Œ is compact there is a left and right invariant Riemannian metric (that is, *Ž
and Ž are isometries for all ‘-‹.Œ , see theorem 9.50 below). One can then show that
the one-parameter groups are geodesics with respect to the corresponding Riemannian
connection. But for Πnon-compact one cannot in general find a Riemannian metric
‚ ’”“ †
with this property. In fact for Œ ˆ• Š the exponential map is not surjective
in contrast to Hopf-Rinow– s theorem for a Riemannian connection (theorem 3.17), cf.
exercise 9.39 below.

Example 9.38
‚˜—™“ † † › ‚ † ‚
Consider Œ ˆ:š Š ˆŒ Š ˆš Š the set of švœOš matrices. The
exponential map is in this case given by

Ÿž ‚p£¤ ¡ ¥¦ ¤ ¡§¦ ¤©¨5¨5¨ª¤ ¡¬« ¦ ¤i¨5¨5¨­† † †

ˆ¡¢Š ¡®‹ ˆš Š
• š

131
where this series is absolutely convergent in the operator norm. To see that this is the
exponential map consider for fixed

         #"
     !


which is an absolutely convergent power series in . Hence
$ 

%&       
$        ('*) +

&,
 -
,
%& 
so . / is the unique solution to the linear differential equation
$
, , 2
$  . 01. / . 43
"
Now given 5
$
, , :2 
$  . 657 . 859/ . 65;<. 57
and also
$ A $
     ,  
$  . >=?. 57@ $  . CBD=?. 57 1E. F. 57G
H2  
. F. 75 . 57G3
Hence by uniqueness of the solutions we obtain
,  
. 859. I. 95

so that indeed .KJ L M1N  is a one-parameter group.

Exercise 9.39
Let O A4T U \a
 T $` U V
N QPRS V $ XB WW R _
WZW Y\[^]
a. Show that the Lie algebra is
A T U T U V " a

 0 P b V ` T B WW
O WW

%  
b. Show that J L N is given by
cddd gih+jkmln `
dd bEp rqtsvu^} w?~€ ‚iy^ƒ+z:„ {| ‡†
x '# if ˆ‰+Š ‡‹DŒ†
e  Y\[o]

%
b ddd  Ž † if ˆ‰^Š KŒ†
ddf
‘’+“”–•  Ž ™tšœ›} } o:^ƒ:„ ƒ„ †
ˆ\‰^Š ˜— if ˆ‰^Š ‡žDŒŸ

132
c. Let us consider 1-parameter groups the same if they have proprotional infinitesimal
generators.
Show that 




 






 if

lies in exactly one 1-parameter subgroup in infinitely many if 
and in no 1-parameter subgroup if ,  .

Let us list a few properties of the exponential map:

Proposition 9.40 
a. ! "$#&%('*),+
' is smooth and maps a neigbourhood of diffeomorphically onto
a neighbourhood of -/.

b. If 01"23+ ' is a homomorphism of Lie groups then 4567098  0:6;< i.e.,

the diagram below commutes

#=%(2>) + ?A@ #=%B'C)

D D
5 5

2 +? 'E.
c. In particular for 2 FG' a Lie subgroup
45 HJI #=%B2>)  5 K "L#&%(2>)M+32N.
I

Proof
a) follows from theorem 3.8, theorem C.1 in appendix C and proposition 9.37.
b) if OP" +32 is a 1-parameter group with OQ8SR!T9U JY then clearly 06ZO[" +\'
UWVX
_^
is a 1-parameter group with infinitesimal generatord^
^a`   ^/` = 
%(076]O) 8SR 0 8cb 6 O 8SR 0 8Y

so 45e%(08 Y )  0N6fOg%  )  0g%h45 Y )

c) clearly follows from b).

Remark
It follows in particular that the differentiable structure on a Lie subgroup 2 F'
is uniquely determined by the requirement that

45 H "L#&%(2>)M+32

133
is a diffeomorphism in a neighbourhood of 

Exercise 9.41

For  a Lie group let  be the left invariant connection given by  

 any left invariant vector fields (cf. proposition 9.37).
 
a. Show that the torsion tensor field  is left invariant and is given by 
  !"#%$'&   . Show also that the curvature tensor field is zero.

b. Show that if  is connected then a vector field is left invariant iff it is parallel along
all 1-parameter groups and their left cosets.

c. For  and ( two connected Lie groups and )+*,.- ( a diffeomorphism with
0/ 1 / 
) show that ) is an affine transformation iff it is a Lie group isomorphism.
(Hint: First observe that if ) is an affine transformation then it preserves left
invariance of vector fields.)

As an application of the exponential map we shall study homogeneous spaces. Let

 be a Lie group and (324 a closed Lie subgroup (in particular ( is an embedded
submanifold). Consider the set !56( of cosets 78( with the quotient topology.

Exercise 9.42
Show that 95:( is a Hausdorff space and the map <;=!56(>- 95:( defined by
 
7 76?@( - 7 7A?B( is continuous.

Theorem 9.43
For  a Lie group and ( 2C a closed Lie subgroup, the quotient space 95:(
is in a natural way a smooth manifold and the map D;E!56( - 956( defined by
 
7 7A?B( - 7 7A?F( is GIHJ

Proof
Let K be the Lie algebras of  and ( respectively and let 2 be a
 L
complementary subspace, i.e.
Let MN*O - 95:( be the canonical
projection. Since the map P%* -  defined by
SROTVUXWZY "Q \[]U^WZY R_ Q $ _R`$ 
P "Q

has differential equal to the identity at / , we can find neighbourhoods a 2 a 2

of in and such that
PV*a ;'a -b
U^WZY
is a diffeomorphism onto an open neighbourhood of / in  . In particular a 2V(
is an open subset, so we can choose c a neighbourhood of / in  such that
eUXWZY
cedf( a

134
By continuity we can find a compact neighbourhood 
 of  such that
    "!#$%& ('

Claim. )+* , .- /1032 is a homeomorphism. (The set , 4
5/ is called
a local cross section).

 *&-   6
7/ is continuous and one-to-one, hence a homeomor-

) is one-to-one on ,  6* Suppose
Clearly
phism, so we shall just see that

) ,  8 )   %9 !   !# %  ('

Then there exists :  2 such that

: 8      % ;!

hence : 8 , =<>!?<> ' It follows that
,    ,A@<B 8   %  '
But since C is one-to-one, < 8  and   8  % ' This proves the claim above.
Now put D 8 )FE  *?G- /1032' Then we shall use (DH ! DJILK ) as a coordinate
chart around 2  /A0M2 , and in general

D ILK EHNPORQTS*VUWD @ B -
shall be a coordinate chart around UX2  /A0M2' It is now straightforward to check that
this gives a smooth structure on /1032 and that the map /6YZ/1032[- /A0M2 given by
@ U ! U  2 B - UXU  2 is smooth.

/\Y]/A032 - /A0M2 / /A032

Remark

/ /A032
Notice that the above map is a left action of on , so
that acts smoothly on .

^`_ @bac! B d7e a the subgroup f @ d !#ag d B
h^i_ @"aP! B of

Examples 9.44
a. In consider for given
matrices of the form

U 8kjmlo prn qts

where l is the u$vZu identity matrix, nxwzy|{ u~}#€uƒ‚ an arbitrary u$v { €„u ‚ -
pMq w‡†iˆ‰{ g€Šu~} ‚ ‹ Then †iˆ‰{ P} ‚Œ nŽ{ u~}#g€Šu ‚ is in a natural way
diffeomorphic to the open set ‘“’ ”• yG{ P},um‚=–
matrix, and
“— ” of u linearly independent
135
  
column-vctors in 
and the left action of
corresponds to the natural
   
left matrix multiplication on such columns. In particular




     #$
b. Let 
"!
be the subgroup of matrices
%+*
%'&)( 
%,.-
#        
where %*0/
'
1% ,2/
 3
4 /6;$5 

&  Then
   by theorem
9.43 we get a manifold structure on 798 :
<
Notice that
    & ; => ;


@?
and hence by a) we have a bijection
 #
798 BA6C +D 81

where on the right we have used the usual right matrix multiplication of
 5 
  Now the set C 8 ;


on C
+D 8 !  +D  is in a natural way iden-
tified with the set of -dimensional linear subspaces of Therefore the manifold

7 8  : is usually identified
 &  with * this
 set and is called the Grassmann manifold
 FG
of -planes in For
7 E &  E is called the -dimensional
(real) projective space.

Exercise 9.45
9HJIKL2M I 
a. Let N! be the subgroup of orthogonal matrices of the form

%4& (O %1,/ I'   
1% ,P-

IQ$ MHRI' 2M
Show that is diffeomorphic to the submanifold S 8 !
5  
 +D
& +T 8 of matrices U satisfying
UWV U &
O
 
that is, the set of orthonormal vectors in
 S +D 8 is called the Stiefel manifold
of orthogonal k-frames in So in particular
I'  HJI'$2Y *
X Z[S +D *\&^] _

b. Show that
 I'$ I' HJIQ$  I'
7 8 `  BA6S +D 8
I'; HaI'   I $
'
where b! is the subgroup of matrices of the form
%+* 
%4& (  %*c/ I' d
%1,/ I' efM 
%<,P-

 * I'  I'
In particular  E is identified with ] +_ where &
 9  h3 & h 

acts by g g

136


We end this chapter with a few remarks on the adjoint representation of a Lie
group with Lie algebra

   
Definition 9.46

  For define to be the differential at of the inner automorphism

Proposition 9.47
a.  is invertible and

 !" #$% 

is a Lie group homomorphism.

b. The differential of Ad is given by

& ' )(* 
where & +, -./
)0 +213-54617+ 1 -8

c. 9;:<5=+,>/
?(@:A<B & +,1C+ 1 where (D:A<EF(G*B 5 #H$  is given in

example 9.38.

d. 9;:A<I+,
J9K:A<L=+,M16N + 1OPQ.
RS TUDV be given by
Proof
a) For let

TAUW  @
X   1  Y.1

so that

=ZD
[TAU\M]_^` 
Clearly TAUaT Uabdc
eT ^
gfhI1 so ji%;kl
gfh Similarly

ji%Wnmhk6
Sopji%qmrk51 s13qmtF.
Also the map uvBxwIy given by uz=s1  6
J   is clearly differentiable so
that u ] B{5xw|l5 {` is differentiable. Now

= =+,@
}u ]t~ €+Y3‚ U„ƒ ^>†,‡2ˆŠ‰Œ‹lŽ ‘‡2ˆŠ‰;ej’ ‡“

so the map •” – given by ‹=’s3PŽ˜—– ™š‹’Ž ‹=,Ž is differentiable from which
a) easily follows.

137
 b) For
put  
 and notice that

 
   
  "!$#

so that
 
 % )
&'(   ) +, - '(./1 032 
*
  $ ! #76 "!$#98
54
That is, for :
<;=>@?A

DFE
   )
(9.48) &'(7B:C D D ) :G  & '(B   
 0  $ ! #&6 "!$# 8
) H4
Now, for fixed

D D
) )
D :GI, 'JB    D :GL& '( 
 0  K!$# D   K!$#
(9.49)  ) 
M D :G+, '(/ 
  K!$#

On the other hand the left invariant vector field ' N corresponding to ' clearly satisfies

)
'PN OQ@:CR'SB:STUGOL  ) S : TUGOQ, ' 
V
  !$#
) 4
  ) :GWX&1 '(   YVZ
[?(
 !$#
H4
so that
D%E 
)
D D ) :\I, 'J+    'PN O.]+@:C  RR^_'Z
N B:3`
  K!$#,6 $! #   K!$# 8
 4 
Similarly D E
)
D ) D :G., '(&, +  a' ^ b
N B:C `
  K!$#,6 $! # 8
 4
Hence by (9.48)
 
I@'J,cd:eFR N B:C ` M
^ 'S ' ^ f
N B:3 ` gc h7'ieB@:C

so that indeed
  
W&'(gc h7'ieFRj 7@'(
8
 lk
c) now follows from b) and proposition 9.40 applied to ?m n9o@ 
8
d) is obvious from the definitions. In fact, the one parameter group  pm
X&1W/ 032 A
 has infinitesimal generator

 
 r, -WBP032&s   P&
Fq  K!$# 8


138
 As an application of the adjoint representation let us prove:

Theorem 9.50

If  isa compact
 Lie group then it has a bi-invariant Riemannian metric, that is,
are isometries.

Proof
First choose an inner product   in  and extend it to a Riemannian metric
on by
  "!
$#&% ' (
)#&% ' +*,.- /0  (1
     ,2

.54
As in chapter 7 this enables us to integrate functions on 32 By construction 768
:9
is an isometry for every . Hence clearly
; ;

. -@1
and integrable on 32
<>= <>=1? =
We claim that also
; ;
  -BA
(9.51) and integrable on 32
< = < =A? =
For this notice that (cf. exercise 7.36)
; ;
 FE 
HG IHJLK G
<8= < CD=1?
 
where
LR 4 G M
I N
J K G is the function O6
 whose value at P is Q the determinant of
TS& 6 U E TS with respect to an orthonormal basis. Since left translation is an
R
isometry S has the same determinant up to sign) for all P and
C VNR  NRW4  E
IHJNK IHJNK +76O
 QXIHC JNKVY&Z.[)\)]_^`ba [Lc dUef/g7hOeifLgXj
k lXmHnporqWmtsDuHvTwx
c kzy {D|}~ 
However, mHnLo ` qWm dg h defines a Lie group homomorphism, and
c c
since g is compact the image in is a compact subgroup of and hence is l€} It
follows that ‚ mHnNo a [ ‚ kƒ} and hence (9.51) is proved.
Now define another inner product in eifg by
„/„† ~ˆ‡H‰‰ŠkŒ‹ „ qWmŽ|‘ ~Šq’m“| ‡ ‰•”–—~ ~/‡0˜ eifg 

Then
„/„ qWm“|u‘ ~ŠqWmT|™uH‡‘‰‰šk „„ ~/‡H‰/‰(~ ›Bu1˜ g ~ ~/‡A˜ e+fg 
(9.52)

139
In fact
        "!#
  

   $  %!# &  
    


follows using (9.51). It is
now

' '-,/. checked
easily 0 0 211 the metric
' ,/. that 43652defined
798 by
 )((+*  *   
:<
is ;9=9
bi-invariant: ' ; before it
:>;?=?
 As :D;?=9invariant
C ,/. 0left
' ;?is ' ;EC ,F =E 11
. 0 :D;9also
and
  (@( *BA  *BA
: ;9= ; ,/.H0 ' ,/. 0 : ;?= ; ,/.H0 ' ,/. 0 I11
G(@( * *  * *
KJ"$LM9N ' ,/. 0 OJP$LM9N ' ,F. 0 21@1 ' ,F. 0 ' ,/. 0 211
G(@( *  * Q((+*  *
   '
 
by (9.52). This proves the theorem.

Exercise 9.53
Let R be a Lie group and S TGR a closed Lie subgroup. Let U be the Lie
algebras of R and S respectively.
0
a. Show that the projection VXW GR Y R[Z\S 527F rise to an exact sequence
gives
] ]
3 Y Y Y^ R_Z\S Y
5 7F
2
and that for ` S there is a commutative diagram
] ]
Y Y Y R_Z\S Y

abcdegf hbcdegf hiOjk

l m m m n2o dp_q\rsf m l

b. Show that there is a one-one correspondence between left invariant Riemannian

p[qtr q
metrics on and inner products u&vwxvzy in satisfying
bcdegf{ bcdeIf"| { | { |‚ q eƒ„rƒ
u w y~}u w y € w w

p
c. Show that if † ‡ w with † compact, then there exist left invariant metrics on
p_q
† .

Remark
Riemannian manifolds of the form considered in b) above are called homogeneous
Riemannian manifolds. In the next section we shall see that every simply connected
Riemannian symmetric space is a homegenous Riemannian manifold with compact
isotropy subgroup.

140
Chapter 10 THE ISOMETRY GROUP OF A
SYMMETRIC SPACE

From now on we shall call a connected Riemannian globally symmetric space just a
symmetric space. In this section we shall study the group of isometries of a symmetric
space, and we shall show that it is in a natural way a Lie group. As a first step we
make it into a topological group. For convenience we assume simply connected.

Definition 10.1
For a connected Riemannian manifold let
 denote the group of isometries
of .
 is given the compact open topology, that is, a basis for the topology is
given by the finite intersections of sets of the form

 
  !

where "# is compact and $% is open.

Proposition 10.2

 is a Hausdorff space with a countable basis for the topology.

Proof
Since is a connected manifold it has a countable basis &(' )+* and since is
locally compact we can assume that ',) is compact for all -. Then it is easily seen
0/
that the set of finite intersections of the sets ',)1'243 constitutes a countable basis
for the topology of
5.

Remark
The usefulness of this proposition lies in the fact that if 687+9;: < is a function
and 9 has a countable basis then 6 is continuous iff 6 maps convergent sequences to
convergent sequences.

Proposition 10.3
Let &>=?)+*@
 and 6AB
C. Then the following are equivalent:
a) =+)D: 6 pointwise on E.

b) =+)F: 6 uniformly on compact sets.

c) =+)8: 6 in the compact open topology.

141
 Proof
b) c) a) is easy. a) b): Given
compact and  Find finitely many
points 
such that any point 
has distance less than  from some point  .
Let  be so large that for all 
   ! for all "#$
Then if %#
has distance less than  from  we have
&' ()  * &' ( ' ,+ '-' )  ./+
+ '   )  /
0  1,+ '-' )  .,+  2-#345
for "678 This ends the proof.

More generally let 9 : ; be an open set and let < 9=;> denote the set of


isometries of 9 onto an open set of ; . Again < 9)?;> is given the compact open
topology.

Proposition 10.4
Let ; be a simply connected symmetric space. For any open set 9@:A;
 
restriction map BDC1< ;>FE < 9=;> a homeomorphism.
is the

Proof
By lemma 8.21 B is injective and by theorem 8.20 it is onto. Clearly B is continuous.
In order to see that BG1H is continuous let I
&&J K<  ;> be a sequence and  <  ;L
  E  on 9 In view of proposition 10.3 it is enough to prove that
'-M NE  O M 
such that
P M 6;Q
For this we shall need normal neighbourhoods R , which are regular in the following
sense: For every 
M MTS UR there is a unique V%XW'Y  ;> (of small length) such that
M S 0QZ\[]  V^ and the mapping RL_`R6EaWb; sending cM  M S  to V is a diffeomorphism
Y
S
onto an open neighbourhood. By theorem 3.8 every point has a regular normal
S
neighbourhood. Furthermore notice that for any two points and there is an isometry
of ; taking to (in fact the symmetry in the midpoint of a geodesic from to
S
does this). Therefore, we can find an d8> such that any point has a regular normal
neighbourhood of radius de With this in mind we shall prove

Lemma 10.5
Let ;XfI
&J and  be as above and suppose ; has regular normal neighbourhoods
of radius ghde Suppose
'-M )E  cM  for M in a neighbourhood of ij; . Then
OM kEl - M  monqp   1 M rdq

Given this lemma we can end the proof of the proposition as follows: Suppose
& El

142
on and choose
 . Let  be arbitrary and choose a curve 
from
 to   . Then by an argument similar to the proof of lemma 8.22 we have
for all   that !"$#% &'"(# for  in a neighbourhood of )*+#+, In particular
-."/# &'01#2, It remains to prove lemma 10.5:

Proof of lemma 10.5

Suppose ."(#3 &'"(# for  in a neighbourhood of
4,

Case 1. -!5
!#76&'8
!# 9;:<, Since -'=">?A@(BC0D#FEG6H>?A@IBJKML$0D#N# for small D 

O
B KP# and since expB is a homeomorphism on a small neighbourhood we conclude
O
that $LM0D#Q &ML/0D# for small D and hence by linearity for all DR B KP# . Hence
-<=K>?A@ B *D#SE converges to &T=F>U?A@ B 0D#NE for all D , that is, -.0I#V &'*I#W9XYZ[,

General case. Clearly it is enough to consider &\6 ]5^, So -!*I#_  in a

neighbourhood of
. Let
 be the midpoint of the geodesic from
to -.5
.# and
let ` be the symmetry in
 . Hence `.5
.#J6a.5
!# or equivalently

`bC.5
!#'6_
4,

also let ` be the symmetry in
. We claim that it is enough to show that ` `c
on the normal neighbourhood d of radius e . Because then by proposition 10.3 this
convergence is uniform on compact subsets of d and it follows easily that -P
]f^ on d iff `gbh-i ` on d which reduces the general case to case 1.
It therefore remains to prove that if
AGj
then the corresponding symmetries `
converges to ` in the neighbourhood d . But by assumption the normal neighbourhood
of radius kMe is regular. Therefore the mapping

=Sl2m5E_n >?A@oprq7D#p6[` o =Fcm5E =K1m6s>?A@op0D#NE

is differentiable on dutvd and is in particular continuous as a function in  . This ends

the proof of the lemma and hence of proposition 10.4.

As a corollary we get:

Corollary 10.6
For a simply connected symmetric space the mapping xw."P# given by sending
 to the symmetry ` o is continuous.

Proof
As seen in the proof of lemma 10.5 the mapping yn ` o of d into w."dzUy# (for
d a regular normal neighbourhood of a point
) is continuous. The statement therefore
follows from proposition 10.4.

143
 Corollary 10.7
Let be a simply connected symmetric space and choose
 . Let   


be the isotropy subgroup at , that is,

  

!#"


Then the homomorphism  $% '& ( is a homeomorphism of  onto a compact subgroup
of the group of linear isometries of )*(+ .

Proof
Put ,  )-(./ and let 01,2
be the group of isometries of On ,3"all 01 ,4
possible topologies agree: the matrix topology induced from a choice of a basis of , ,
the norm topology with respect to the inner product, the compact open topology and
the topology of pointwise convergence. From case 1 in the proof of lemma 10.5 it
follows that a sequence 6578
converges in a neighbourhood of iff
95+ &:(
converges
in 01 ,4. It therefore follows from proposition 10.4. that the topology on induced 
from  / is the same as the one induced from 0;,4<"
Now by theorem 8.20 the
@ =ACBD=EFG= HIJ01= @ ,2:A B<EFKHLB
image of in
@
is the set of linear isometries =?>!)-(./4% )*(+
satisfying


where is the curvature tensor field at . This is clearly
a closed subset of the compact group . 01,2
Corollary 10.8
 / is locally compact.

Proof
Let us find a compact neighbourhood of OM NP /Q" R S
Again for and subsets of
we put TURVBWSV!/XZY1U/\[]YUR2^ S`_a"
Choose
P8
and an open neighbourhood
S S
such that is compact and is contained in a normal neighbourhood of . Then

MbNcdTeO
BWSV^?Tgfh
B Si
and we claim that T f
B S i is compact. So let = 5jT f
B S i " Replacing X]= 5 _ with
a subsequence we can suppose that =C5O
 converges to some point k S . Let lnm be
the symmetry in the midpoint between
and k and also let lo5 be the symmetry in the
midpoint between
and = 5
 . By corollary 10.6 lo5p% l\m]" Also ln5rq= 5
st
so
since  is compact we can find a convergent subsequence of X]lo5Iqu= 5C_ and hence a
convergent subsequence of X6= 5C_a" This ends the proof.

We shall now study the action of / on ?"

Proposition 10.9
Let be a simply connected symmetric space.
a.  / is a topological group.

144
b. The evaluation map 
   sending
 to 
 is continuous.

c. Given  let  be the isotropy group at  . Then the mapping   
 
sending to 
 induces a homeomorphism 
"!# $  where the left
hand side is the set of cosets %&'( with the quotient topology.

Proof
a) and b) are trivial.
c) Consider the diagram of maps


,.)- !* / + 


0
As remarked above 
 acts transitively on  hence  is clearly bijective. Also  is
obviously continuous. We shall prove that  is open. It is clearly enough to find an open
-45 - -
neighbourhood 1 of id 2
 such that 1 3
1 and such that 76 1 is open.
For this choose  and 8 as in the proof of corollary 10.8, and put 1 391:
;<8= . Clearly
45 - 45 -
1 3
)8>?3 -
1@BA Also 1DCE; 8GF is compact.- Since  is continuous 
)!*
is Hausdorff hence C 1 C ; 8 F&F is compact and 6 C 1 C ; 8 F&F is a homeomorphism
-
onto 8 . Hence 76
1@ is a homeomorphism onto 8HA

In order to make 
 into a Lie group we first consider the isotropy subgroup
 at :I 

Proposition 10.10
The isotropy group  at JK is a compact Lie group.

Proof
As in the proof of corollary 10.7 we put LM3ONQPR and we identify  with the
subgroup of the orthogonal group ST
L' given by the set of linear isometries UQL@ L
satisfying VW
RXYZR[=Q\]3^VW
X_[>\W for all XYZ[`B\2aLb where V is the curvature
tensor field at  . Putting cd
ef_gB\Z[Gh3 iVW
ef_gj\kZ[>l we can identify  with the
subgroup
m3:nRT2SU
LWJo d c
pef_QgZQ\kZp[qr3cd
efZgB\k_[>sutWef_gB\Z[9 Lwvk
o
and we shall show that this is a Lie subgroup of the Lie group ST
LWZA For this let
y1 denote the vector space of multilinear functions LOJLIJLIL x; and let
fSU
Lz> 1 be the differentiable map given by
y

{|3}c~G
d€d]U]{BA
Also let
L‚9ƒ?„
Lz denote the Lie algebra of SU
LW , i.e. the set of skew-adjoint
linear endomorphisms of L . Then by differentiation it is easily checked that
3†.‡‰ˆ‹Š|ŒŽ ‘’W“

145
can be identified with the subspace of skew-adjoint linear maps satisfying

      
      
     
     
    "  !$#

Now by the Implicit Function Theorem there is a neighbourhood of in and % & '  !(
an (embedded) submanifold ) *+%
such that and , -/%0*1)
. On the other 2 34)5 #
!6
,87*5'  !9 >=@? A CB
hand it is easily checked that is a Lie subalgebra of so by theorem 9.24 there
is a corresponding connected Lie subgroup and we claim that In ,87*,:B #
fact for
that
=@? A; DE,GF =@? A  D H,:#
it is easy to check by differentiation that
that is, It follows that
<
is constant in so 

,87I-J%1*5,K-:%;*5)
,87 ) &
'  !6
and since and have the same tangent space at we get, by possibly making
% smaller that ,87D-L% M ,N-H%MM)O#
Hence is a submanifold of , in a
& PQR , P
'  !6
neighbourhood of . Since for left-multiplication by is a diffeomorphism of
it follows easily that ,S*5' !6
is a submanifold and hence a Lie subgroup. We
have already seen that ,
is compact.

Remark
,
'  2 TVU;#
Notice that it follows from the proof that is actually an embedded submanifold
of In particular the manifold structure is uniquely determined.

We shall now prove:

Theorem 10.11
a. W  U; has the structure of a Lie group such that the action

W  U;IX"U Y U
isZD[ .
b. For \]7QU ,
and the isotropy group at \ the homeomorphism ^Q_`W  Uacbd, e U
is a diffeomorphism.

The idea of the proof is simply to construct the local cross section as in theorem
9.43 and use the differentiable structure on U
together with the differentiable structure
,
on . Recall that in theorem 9.43 the local cross section is given by local 1-parameter
subgroups with infinitesimal generator in the complement of the Lie algebra of the
isotropy group. So we first construct these local 1-parameter subgroups of isometries.

146
 Definition 10.12
Let be a geodesic of
. A transvection along is an isometry  of
, which
preserves and induces parallel translation along on the tangent spaces at points of
. If goes through  then  is said to be a transvection at  .

Example 10.13
In Euclidean space with the flat metric the transvections are of course the parallel
translations.

Exercise 10.14
Describe the transvections on a sphere.

Lemma 10.15
a. Given a geodesic through  . Then there is a unique family    of
transvections along such that   .

b.    !"$#% &

c. Let ' be a normal neighbourhood of  . Then the map (*)' + ,-
. given by
associating to /1032546879 the transvection ;: along /"03254<=79 is continuous.

Proof
a) is obvious from theorem 8.20, explicitly >#@?5A CBD$E 6# 4 as shown in the proof
of proposition 8.3.
b) is obvious from the uniqueness of a) and the fact that FGHIJ-#LK M
which is easily proved.
c) By a) ( is given explicitly by

(ON-/1032546G7P=QR>#TSVU1W$X ZY[\  6# 4

so c) follows from corollary 10.6

Proof of theorem 10.11

Let ]I^(_V'` as in Lemma 10.15 above. Then clearly

(a)b]^+c'edf

is a homeomorphism, so choosing local coordinates on ' we get a gih structure on

] . Clearly the map jk"l nm+ (RjoLpql

'ertsvuxw.yFz|{~}V€‚

147
is 1–1 onto an open set, and since is a Lie group it follows that
 has a 
structure. Let

 
be the inverse map. For any  we give 
 !" the  structure
induced by
$#&%')(+* #-,/. 0
 112
*3# denotes left-multiplication by $4 We must show that these structures
where as usual
on their overlaps. It is enough to show that if 5
6 7'
0
%8
9 thencompatible
are
$:;(<>#=@? is  . Now if BADCDEC % AFE2<AGC2HAI
J2<EC2<E1K12 then &A ?ML % A L
and it follows that  L  is near to L so if we replace
above by some smaller
0N then
 can be written in the form  % A ? E ? for A ? '
O2PE ? Q14 So it is clearly enough
to show

A ? AGCO
R2BE ? EC) runs through all elements such that

Claim
If

A ? E ? AFCSEC % AFET2 AUV
W2XEYV12

then A and E are Z functions of A ? 2[AGC/2GE ? 2GEC/4

This will prove that \M" is a   manifold. But at the same time it will
prove that \M" is a Lie-group. In fact suppose we have proved the claim and put
] %
^_`!ab"G4 Then ] is a neighbourhood of c and by the claim the multiplication
]
restricted to is   . Also since is a Lie group and since clearly AKd A
=@? is  
on
(via e this map is equivalent to fhg ) it follows easily that id
] . Hence given the claim \M  is a Lie group due to the following lemma,  =@? is   on
the proof
of which is left to the reader:

Lemma 10.16
j]  ]  ] 
c j
Let be a topological group with a structure such that left translation is .

 d  =@? ] j  j
Suppose is a neighbourhood of such that the multiplication and the
map of into are . Then is a Lie group.

Proof of claim

A =@? ? E ? % AGCSEkC % AFE2XA ? A ? 2[AGCB
l2;E ? E ? 2GECm14

Suppose

notice that E ? AGCSE ? AGn is a transvection

@
= ? at L . In fact if AGC is the transvection from
LFirst
to AFC  L Q
% F
o r
p q g b
 t
s  then clearly E AGCSE ? is the transvection from L to
? oFprq g uME ?  st %
E ? AGC L [4 Since acts  on v5gt" (by definition of the  structure on ) via the
n
mapping E ? d E ?  and since
F
o r
p q g is a diffeomorphism onto the neighbourhood 
n
148
it follows that 
  is a  function of  and  , so 
is a  function
of  and  .
Now 
 where is the transvection from  to 
 . So
we must show that 
 depends  on  and 
 . Here 
 is constructed as
follows from  and 
 : Take !" #$ such that %&(' " )*+,
-/. parallel translate
 along the geodesic to  and apply %& '10 2 "43 Each of these are   operations as
functions of
 and  . Similarly the matrix coordinates of the differential at  of
the map

5
678    5

depends  on  and 
- so the coordinates of 98:; 5 depends  on
<=><<; 3 Hence since ? is a Lie group  depends  on  <=@<<; 3 This proves
the claim and hence that A #$ is a Lie group.
Notice that since B6CEDF G is a diffeomorphism we have also proved that the
action - H<=IJLFK NMJI is  as long as NMOIPQG . It is easy from this to prove
the remaining statements of theorem 10.11.

149
Chapter 11 SYMMETRIC SPACES AND
ORTHOGONAL INVOLUTIVE ALGEBRAS

In chapter 9 we associated to a Lie group its Lie algebra. Similarly we shall construct
to a symmetric space an associated algebraic object called an orthogonal involutive
algebra.

   


First a few remarks on the adjoint representation af a Lie Group . Recall that for
  
  
 
     
the mapping defined by has differential and

   
the mapping is called the adjoint representation of . The differential
of this is the map  defined by
   ! #"%$ '&(!*)+& ,&-!  .
For an arbitrary Lie algebra we still have    /
  0
   defined as above and
the image  21 / 0  is a Lie subalgebra of 0  which is the Lie algebra of
3 -. Let 45768 1 3  be the connected Lie subgroup corresponding to  (. If
is the Lie algebra of a Lie group then it follows from 9.47 that 9 :  ;< is
a Lie group homomorphism onto 45=6> (
 . Notice that the topology on 45=68
be induced from the topology on 3 -
 . need not

Now let 1 be a Lie subalgebra. Then   91   is a Lie subalgebra and
hence corresponds to a Lie subgroup ?A@ 1 45=68 B1 3 -. We now have:

Proposition 11.1
The following are equivalent
1. ?C@ is a compact Lie group.

2. The set ?C@ is compact in the induced topology from 3 -.
3. ? @ is closed in 3 EDFHG has a ? @ invariant positive definite symmetric bilinear
form.

4. ?C@ is closed in ; IDFJG has an  (K invariant positive definite symmetric
bilinear form , i.e. L
LM $ ,&-!*) &ON Q P RL !& $ ,&ON2) S"UT V   &W!W&(N  .

Proof
a) X
b) is trivial. On the other hand if ?C@ 1 ;< 
is compact then by proposition
7.6 it is a closed Lie subgroup in a canonical way which proves b) a). X
b) X
c) by integrating an arbitrary positive definite symmetric bilinear form over
?C@ (cf. theorem 9.50).

151
 c) b) is obvious since the orthogonal group is compact.
c) d) by differentiating the equation

     !#"$
%&'('
*) &+,-) 

/.  0 
132546  be the orthogonal78group

with respect to .
d) . Let
  corresponding to the positive

0   9;: 1 0   which is compact.

definite bilinear form . It follows from d) that is contained in the Lie algebra
of hence

Definition 11.2
If either of a) - d) above holds we say that is compactly embedded in <
=>@?  
If is compactly embedded in then is called a compact Lie-algebra. I.e. is
compact iff is a compact group.

Notice that 9
compact clearly implies that the corresponding Lie algebra is
compact. However, the other direction is in general not true (e.g. for Abelian Lie
algebras).

A H BC ) A
!F C " G C  A  J
Now return to a simply connected symmetric space and let with
IG  A  DEC
9C " G C  A & K 9  RS"TR9
symmetry . Let be the connected component of the isometry group
A
of . Furthermore let where is the isotropy group for
BC
. Finally let LNMOF!CQP A
be the mapping L
. With this notation we BC
now have:

Theorem 11.3
1. 9 C is connected and the mapping L3MUFVCXW79 CP A induced by L is a diffeomor-
phism.

2. The mapping
3
" c
b YZ M[FP F given by R]\P DEC_^ R ^`DEC is an involutive automorphism
(that is Y a ) such that 9 C is the connected component of the group 9 d of fixed
points for . Y
3. Let be the Lie algebra of FVC . Then
" e  where
"gfh*) i Y j"kQlm "nfho) i Y H"npOQl
: :
and is the Lie algebra of 9 C . Furthermore L :
 `
 r
" t
q s7uwv L M P x yXz|{6}n~ is
an isomorphism.
:
4. If  € ‚ƒ „‡†ˆ‰ ŠŒ‹ Ž{†~‘h’“ is the geodesic with tangent vector ”w•  and
,Š ‹ Ž {†~ is the family of transvections along that curve.
152
5. For 
 we have     and
    !"
#
Proof
a) For every %$
'&
there is a transvection taking to ; hence acts transitively$( $ )*
&  5 )*,+-/. &0#
on , so clearly Now let us show that is connected. 1324)*
Consider )687 & ;:<
and notice that around any point of we can find a &
neighbourhood 9
such that is homeomorphic to =96 (compare the proof
5B9/A CE>?D,FHG  7 )
IKJ 5;ALCMD,FHG 7 & D N
ALCMD,FHGD
of theorem 10.11. Using this it is easy to see that if
PO
is a curve and @
I   QO @ I J C B I  C D I J F R I  F @ N
A CED,FHG
is a homotopy of keeping the endpoints fixed (that
is, and
I JS5 A CED,FTG 7 & DRN
ALCMD,FHGD
), then we can find

UO I;J  I;J I  @ #

a lifted homotopy (keeping the endpoints fixed)
A CED,FVG > ALCEDVFVG
such that

5YA CED,FHG 7 )6 Z-#

and (To show this subdivide into small squares).
Now let U

PO @ 4 W
 X
2
and join by a curve 6
)  to Since @ is simply
IJ &
connected
as above, then clearly I <U5 A CED,FVG 7 )*
is homotopic to the constant loop and if is a lift of the homotopy
is a curve lying inside and joining to 
Z-# [
^]_`
acb  de?fg#
Hence is connected.
b) [\
is clearly a closed subgroup of with Lie algebra )
It is therefore enough to prove that is actually the Lie algebra of or equivalently 
that for 
hTiEj**
kl#
But for we have 

hiMjmon=p$EBbqhiMjmon=rs$(B N trhTiMjun=p$Ec
nv
D
so since $( is the only fixed point in a neighbourhood of $( , it follows that hiMjwonxp$(B
$( ynX
D hence hiMjwpn=?
0 yn#
c) Clearly

   z C and we have just proved that is the Lie algebra of
 . Obviously  # Now let {
|M}V~, & and let |€ be the family of
transvections along hTiMj },~ on={l |€r‚$(_# Now |€ is a one-parameter group so that
|€mƒhTiMjwon= for some 
# Then
  „ „ |€r‚$( † €s‡ˆ  „ „ hTiMj },~ n={( † €s‡ˆ ‰{
n †† n ††
so P5 7 |M}V~, & is onto. Hence by dimension reasons P5 7 |M},~V &^ is an
isomorphism.
d) Notice that  in the above argument lies in # In fact | : € is the one parameter
group of transvections along the geodesic
hTiMj },~ ‹ŠRn={ N ˆhiMj },~ n={(qŒ N 6hiEj*on= N cs$(B
bhTiMjmn=xp$E‰hTiMjmnEb  p$E D
so that b   ƒŠB , that is, 
# Since Ž5 7 |M} ~ &^ is an isomorphism an
arbitrary 

transvections along the geodesic hTiMj } ~ n={( where {l  P#
occurs as the infinitesimal generator of the one parameter group of

153
 e) For 
 and 
we have
  !"$#%& ')(*+
'
or equivalently

,!- .,/0 1,2&34,!" /50 1,6#7$ 3' (*+
8

Hence 9./50 1,  # /50 :9,;< so that /50 :;5
8 Also
;, /=0 19,;<&?>@A B!C  >@=4 !E DGFHI .  
J! D FK@ . I . ;6L' (ML
'
so clearly

I .N/=0 :;34 . I . O8

This ends the proof.

To the symmetric space P 9.TS U

we can associate the Lie algebra of QR
together with
8
V S U 8
the involutive automorphism In general consider a Lie algebra with an
involution Then again  W where

YXEZ
[[ V\]4_^`' a  XEZ
[[ V\  #=b^@'

and it is easy to see thatc c c
' dfe ' ' dKe ' ' dKe 8
In the following whenever is a Lie algebra with involution Vg' we shall denote the hji
and #"i eigenspaces by and respectively.

Definition 11.4
An orthogonal involutive Lie algebra is a triple 'LV"'lkT
where is a Lie

which is0m  # c V k c
algebra with involution and is a positive definite symmetric bilinear form on
invariant, that is,

k n'o-d;',pRqhrks:ol' n',p7d2t u (jZ
vlol'pw
8

Furthermore we require that e is compactly embedded and that does not contain
any non-zero ideal of .

Remark
k
Since is compactly embedded we can extend to a positive definite symmetric
V 
form on , which is -invariant and ad -invariant. However, only is part of the kyx
structure.

154
 Proposition 11.5

For a simply connected symmetric space let be the associated Lie algebra with
involution
 as in theorem 11.3, and let  be the bilinear form on induced from the
Riemannian metric on    via the isomorphism     
Then 
 is an orthogonal involutive Lie algebra.

Proof
!#" Most of this is already contained in theorem 11.3. Since   is compact clearly
$%'&(*) is compact so is compactly embedded. It remains to see that does
not contain an ideal of . So suppose & is an ideal of . Then +  ,-& . /0
which implies /102 In fact suppose 3 4 and +536 ,/70 then by theorem 11.3
e) we have
!#"
98;:2<>=23?  @
 A9B>C/@ 98;:<D=23?$B>
"
/@FEC:<G$H 9=23?@9B>IJ/0
KLBM4  =O4 
N
"
so since   acts effectively on 
   P we have 8;:<Q=R3S/T K=U4  hence 3V/W02
which ends the proof.

We will now prove that the orthogonal involutive Lie algebra determines the
symmetric space. In fact the curvature tensor field of is determined by the Lie-
product of 

Theorem 11.6
Let be a symmetric space as above and QXYZ the associated orthogonal
involutive Lie algebra. Then the curvature tensor field of at the point [R is given by
\
  ]@^B_@F3?`@bac/ de@f+g+ BD3G,$hae,b BD3'aP4 

Proof
i An element B 4 determines a 1-parameter group of isometries of namely
8;:<D=B76=U4 j and hence a vector field B  on , that is,
k

B  / k l$8h:<_=mB>n[@po 
= o
orqns
Now let B t be the right invariant vector field on uv corresponding to B . Then B t and
B  are Nd related, that is ,

@xw BLt y{z_|}€f~ ƒ‚ „ †‡%ˆ‰ Š

In fact both sides are equal to ‹ }>“ ‡ R” • G

“ – • Š It follows that for }_™›š ˆ the
‹ŒI9Ž;2’‘ – Œ˜—
vector fields œb}ž
 ™^v š Ÿ and } ~ ™›š ~;¡ are also ¢N£ related. Now it is easy to see that the

155
Lie multiplication on defined by right-invariant vector fields and the one defined by
left-invariant vector fields differ by a – sign. It follows that for 

 
   

 
 
so we can identify the Lie-algebra (except for a minus) with the Lie algebra of vector
fields on  of the form 
as above. For 
we call an infinitesimal isometry.
Similarly for   
 is called an infinitesimal transvection (rotation).
In order to prove the theorem we must therefore prove
   ( 
  )
'#  *
(*)
 !
 "

  $ # %
 &
 '
   + 
,
-#. /
    
Notice that the left hand side only depends on &
0 & and # & / Also
1 since
)
so the right hand side does not change if we replace # with
 another *  
vector field with the same value at 243 . Instead of (*) we shall therefore prove
   
(**) 56!
 7#98;:    =<
 )
'#98?>  

where #  8 @# & and # 8 is parallel along the geodesics through 2 3 .
In the following we write ACBD EF
 for and  vector fields on G/
By definition
 
!
 $#H8I BHJ KLJ-#984 KLJ BMJN#984 O BMJP KLJ Q #98R/
 
Since
 *   1 the last term vanishes at 2S3 . Notice also that  KLJ  # 8 TU   1
since # 8 is parallel in any direction at 2S3 . Since the torsion is zero we therefore have

B J 5 K J #98;: & 5VA B J K J #98;: &

Now any isometry W is an affine transformation, i.e. for arbitrary vector fields  and
# on  with W  transformed vector fields YX and #HX we have
K)Z[V# X L\ K9V#]^ X /
Letting W!G_-`Sacb7
C
Cbc
and differentiating with respect to b , we therefore
have
AFBMJ K9 #]L O BMJTP KdQ V#]fe K9 AFBMJg #]T
for arbitrary vector fields  and # .
It follows that
B J 5 K J 9# 8;: &  O B J P K J Q 57#98;: & e\5 K J A B J 75 #98;:h: &
 5 KiJ A cB J45 # 8;:g: &
   1
since again
   /
156
 Finally 
  since  is parallel and exp  preserves parallel trans-
lation. Hence again    
  
       
   
so we conclude that   

        
  ! "
It follows $#
that    
%'&)( *+&,      
     -
    ! 
 /.  10
/$2  !   -43 5 '&!( *6&879(: <; 
by the Jacobi identity. This proves (**) and ends the proof of the theorem.

Corollary 11.7
Let = and > be simply connected symmetric spaces such that the associated
orthogonal involutive Lie algebras are isomorphic. Then = and > are isometric.

Proof
Obvious from theorem 11.6 and theorem 8.20.

Example 11.8
Euclidean space ? with the usual metric is a symmetric space. The group of
isometries @ABC is the Euclidean group generated by the orthogonal group O %BC and
the translations DFEG DIHKJ for JML ? " The Lie algebra of O %BN is denoted %BC and
is the Lie subalgebra of OPRQ9 ?S of skew-symmetric endomorphisms. The orthogonal
involutive Lie-algebra associated to ? is then  BC9T ?U(WV6(YXZY( where 5 ([7Y\ ?Z
 and 5 (]7N^ %BC_ ?AG ? is given by the usual action of %BNW`baYcSdfe gih on\ gkj
The involution l is given by lnm e%oNhIprqd and lRm sgtp uqd9j Finally v is the usual
inner product on g j
Notice that if w eoCh is a Lie subalgebra, then also e x gUy l y vIh is an
orthogonal involutive Lie algebra.

Definition 11.9
e z y lz y v{z)h is an orthogonal involutive Lie subalgebra of e y l y vIh if
z|w y lA} z p~lz y
} z p and v|zp€vj

Definition 11.10
An orthogonal involutive Lie subalgebra of e eoCh9x g y ly v‚h in the above example
11.8 is called a Euclidean orthogonal involutive Lie algebra.
Equivalently e y l y v‚h is Euclidean if ƒ y „ p†Rj

157
 Now let

 be any orthogonal involutive Lie algebra. We want to construct
an associated symmetric space. For this we need the following proposition which will
be proved in section 13 (see remark following theorem 13.5).

Proposition 11.11
Let

  be an orthogonal involutive Lie algebra. Then there is a Lie group
with Lie algebra 
Taking the universal covering we can assume to be simply connected. Then there
is also an involutive automorphism  of with differential  and clearly the fixed

point set of  is a closed subgroup. The identity component we denote by  Clearly
  
is , the fixed points of the involution  . Now let  

the Lie algebra of
 and
let   be the natural projection. Since is closed  is a manifold and
!#" +*-,
 $%&
')( .

is an isomorphism. We give  a Riemannian metric as follows: clearly acts on 
 
by left translations and fixes the point ' . Also notice that the adjoint action of on
the Lie algebra keeps the subspace invariant. In fact
  0/213046587:9;<>=?@A./B13>4<5C7D FE G$  =?A HI/2130465C79J<>=K
, ,
for 9
= 
-O
Since the bilinear form  on is LM13 GH invariant it is also /21&N H invariant. We now
#"
identify $% with via the map  , and it is easy to see that under this identification
  "
the adjoint action of on corresponds to the induced action by on P% . We can
therefore extend the bilinear form to a Riemannian metric on all of  by the formula
, W"
?>9Q
F=KRQSTSVUXWZY\[G]  9Q
S^UAWTY\[G]  =_]`
9a
F= P%
,
where U W denotes left-multiplication by b . By definition  is H invariant.

Theorem 11.12
The Riemannian manifold  constructed above is a simply connected symmetric
space and the associated orthogonal involutive Lie algebra contains
K
K  as an
orthogonal involutive Lie subalgebra.

Proof

 is simply connected
 since is simply connected and is connected. In fact
suppose cdfehg8
jijk  is a curve with cl>gmnocpqir Xs't Using local cross sections of
v _ 
u one can lift c to a curve cJle g8
Zijk
,   such that cp>gmwyxK
& cpqir
'z
that is, cpqir  Since is connected we can join clqir to x by a curve inside
 Since c is clearly homotopic to a curve which is constant near 1 we can therefore
assume that cl{ir|x . Now since c is homotopic to the constant loop the same is
true for c%}~ c .

158
 Now let us show that is symmetric. In fact the symmetry
 at  is induced
by  that is,
       
 

 

Clearly     +

!"$#&%(') *



hence  
), % - ".#&%/' ) 0 1



It follows easily from this and the definition of the metric that
23 is an
isometry. To see that 
 is the geodesic symmetry at  it is enough to observe that
 687:9 

54 <;>=?@ ;A=BC  obvious
which is  since by definition
ED DF and
 6I7:9     % ) )L %MN 
since G4GH The symmetry at is clearly
KJ
 It
H
follows that is at least locally symmetric and it remains to see that is complete.
However, this follows  simply because the symmetries
KJ are globally
 
defined. In fact
TS TS S
suppose  OPQCR  is a maximal geodesic and choose UV WR near Then

with OXYUV the geodesic
5
6 6 S\[ [ 6

J COXWZ?U  U ZU  U Z?U  R
6 [^S 6 
agrees with the geodesic O on the interval ] U UV.] UV  Therefore O can be prolonged
6 KS
to the interval CR ZUV R which is larger than CR e c contrary to the
for UV`_ aEbd
maximality. This proves that is complete, and hence is a globally symmetric space.
Now let  as before denote the connected component of the isometry group of
. The involution f of  is given by
 
 ) ) 
fgWh
 h
 h 

Let  be  the corresponding Lie algebra

%
with involution fi . clearly there is a natural
j
map   sending to and as shown above
"$#k%/'  )Q % )  %  l


 fg

so and hence 4m  preserve the involutions. Also if Dno  is

 )X

evaluation at  clearly D Dn and since DA4 H and DnK4 H  are both isomorphism onto
H H
;=?@ it follows that 4p  is an isomorphism.
6
Now let q be the kernel of 4 . Then is an
invariant ideal of , hence
 r s r  r ut
But since 4   is an isomorphism so q and
vt
hence by assumption. Hence 4 is injective, which proves the theorem.

Remark 1
z z
It follows that the kernel wvx y of { is a discrete invariant subgroup of y contained
| z } zL~ | } | ~ } zL~ |
in y , so if we put y wI y €w  then  and here the map
z„ƒ (†g‡
{j‚ ‰ˆ is injective.

159
 Remark 2
In general we cannot expect to be onto. In fact any Euclidean orthogonal
involutive Lie algebra gives
  as the associated symmetric space and here


is the whole Euclidean group.
Definition 11.13.
 
An orthogonal involutive Lie algebra is called maximal if it is not a
subalgebra (in the sense of definition 11.9) of any bigger one.

Corollary 11.14
There is a 1–1 correspondence between simply connected symmetric spaces and
maximal orthogonal involutive Lie algebras.
Proof
  
Suppose is maximal then by definition   in the proof of
theorem 11.11 is onto.
Now suppose
is a simply connected symmetric space. We must show that
 
the associated orthogonal involutive Lie algebra is maximal. But suppose
  
is any bigger one and let
be the associated symmetric space. Then
 
since  the spaces
and
are isometric by theorem 11.6 and theorem 8.20
 
and hence by theorem 11.12,

160
Chapter 12 SEMI-SIMPLE LIE ALGEBRAS
AND LIE GROUPS

For the study of orthogonal involutive Lie algebras we need a few facts from the theory
of semi-simple Lie algebras.
Let be a Lie algebra. The Killing form on is the symmetric bilinear form
defined by


 


 !"  $&% '

#
Clearly is invariant under automorphism of , that is, if (*) + is an automorphism
then ,
(-.$(/!01
#$ 2.$3% ' Also is  
4 invariant, that is,

$5768.9$:-;<
.576=$9>?A@  #$0B6C% '
Definition 12.1
is called semi-simple if is non-degenerate, i.e., if


.B6DEA@ F G% H 6IA@J'

is called simple if it is semi-simple and has no proper ideals.
A Lie group is called semi-simple (simple) if its Lie algebra is semi-simple (simple).

Remark
If K is an ideal then it is easy to see that the Killing form of is the restriction
of the Killing form of .

Lemma 12.2
If is semi-simple then it has no non-zero Abelian ideal.

Proof
Suppose K is an Abelian ideal, that is, 5  9LA@J' Then for 6M% and N% 
 
6DO 
>P
EQ@:  
6DR 
SL
TK U
hence 
68S0  VW
 D
6DO D
STC@O' So since is semi-simple X@ .

Lemma 12.3
Let be semi-simple and K an ideal. Let
Y
[ZLG% \\ 
#B6DEI@ F!6X% ]^'
\
161
Then is an ideal and
 as Lie algebras. Furthermore and are
semi-simple.

Proof
Since
  
    

   
!
is an ideal. Hence also " is an ideal. Also
 #
 $ %
&   $ 
! '
so " is Abelian; hence "
( )by lemma 12.2. This proves
 as
Lie algebras. It follows that )

 )  or )
 is non-degenerate, hence by the remark
above is semi-simple.

Corollary 12.4
If is semi-simple then

*+-,.,.,/ 0
as Lie algebras, where * ,1,., 0 are all the simple ideals of .

Proof

3 54

Clearly by lemma 12.3 we can decompose into a direct sum of simple ideals.
Now suppose
" 39
: ; 4=< 2   2 
is a simple ideal different from
but then so

> Then

?@7
,.,1, 687

For a Lie algebra the center is defined as

A  ))  
 ;   B7
We then obviously have

Corollary 12.5
Let be semi-simple. Then
a) The center of is zero.

C D F E   7 In particular
b)
GIHKJL  ' 2 MON   7
is isomorphic to the Lie algebra of the Lie group

162
 Recall that in general 
  does not have the topology induced from   
However, for semi-simple 
  is a closed subgroup as we shall now see:
Let    be the group of automorphisms of . This is clearly a closed
subgroup, and so is a Lie group. The Lie algebra is easily seen to be the set of
derivations,  
  That is, ! is a derivation if

!#" $&%('*),+-".!/$0% '*)213" $0%4!5'6)% $0% '87

By the Jacobi identity 9:6 ;<= >

Proposition 12.6
If is semi-simple then 9: ?+ = ( In particular 
4  is the identity
component of @ (

Proof
The second statement clearly follows from the first. Now

" !&%A9:B$0)C+D!FEG9:H$IJK9:L$M=EN!8+O9:L!5$I(%

hence +9:6  is an ideal of P ( Since +Q the Killing form is non-degenerate

on Let RKSP=  be the ideal of P=  “orthogonal” to under T . Then clearly
U
RV+XW It follows that

" !&%9:B$&)Y+O9:Z![$I\+DW ]^$_7 %`!a7 R

Hence !5$_+W ];$ 7 % so R +bW which proves the proposition.

Proposition 12.7
U
Let be a Lie algebra with center and let  be a subalgebra such that +DW
If is compactly embedded then Tdc is negative definite.
c
In particular a semi-simple Lie algebra is compact iff T is negative definite.

Proof
By assumption we have a positive definite symmetric bilinear form e on invariant
under 9:=  For fg7 %9:Hf is therefore given by a skew-symmetric matrix with
respect to an orthogonal basis for e . Hence 9:Bf has imaginary eigenvalues hjilkP%nmnmPmP%(hiCo
and it follows that

TSHfN%>f \+p((9rqPs>L9:Nf=E L9:Bf@>\+ J tuYi u

v w W

U
unless 9:Bf8+aW% that is, unless f<7 +xW

Theorem 12.8

163
 Let be a semi-simple Lie group with Lie algebra . Then is compact iff is
compact.

Sketch proof
As remarked before
is trivial.
 . It is clearly enough to consider simply connected. Then the adjoint
homomorphism 
  
is the universal covering and the kernel is the center  of  So we must prove that
if is semi-simple and  is compact then  is finite. We can clearly give  a
bi-invariant metric, and it follows that the induced metric is bi-invariant, that  and
hence is complete and that the geodesics of through  are exactly the one-parameter
subgroups. Now suppose  is infinite and choose   such that
! " $#%&' ( # )* (+
Let ,-. such that /%,01/3254 and
687:9<; =,0-2+ for some ; > 
By going to a subsequence we can assume that ?@,0BA converges, i.e. let ,C2 IDFHK EFG J ,-B
N
9 M
One can then prove that L O?P,QAR is in the center of contradicting the semi-
simplicity of . We refer to Helgason [chapter II, § 6] for details.

Until now we have studied simple and semi-simple Lie algebras over  These
notions, of course, also make sense over  Now let be a real Lie algebra and let
2 S be the complexification, i.e. as a real vector space 2 T U and
V V V V V
UW,X#YU ZK[\2 ] X , #^Z[# UW,_#YZ[$2 ,_#YUWZ[$2`U ,X#^Z[1# ,X#^Za 

Proposition 12.9
Let be a real Lie algebra.
a) is semi-simple iff is semi-simple.

b) Suppose is simple. Then is not simple iff is the underlying real Lie algebra
of a complex Lie algebra.

Proof
a) is straightforward.
b) First assume that is not simple and write

2 bcT+ddd&T ef# ghi4j#

164
as in corollary 12.4, where 

 are complex simple ideals in . For each
 




we consider the homomorphism of real Lie algebras     defined

as the composition of the inclusion  and the projection onto the -th factor.

Clearly  is non-zero and since is simple  is injective. In particular, for each ,
   !" $# 
 &%
   ' " )"  "
It follows that  and  ( % Hence and *+
 is an isomorphism over the reals. Thus is isomorphic to the real Lie algebra
underlying .
On the other hand let be a complex Lie algebra and let be the underlying real Lie
algebra. Also let ,-  be the -linear map given by multiplication by . /0 on %
 12
Then the complexification has a non-trivial ideal consisting of all

elements of the form 304 ,53-
63+7
so indeed is not simple.

Definition 12.10
Let be a complex Lie algebra. A real form of is a subalgebra of the underlying
 1 
real Lie algebra of such that %

Equivalently a real form is the set of fixed points for an anti-linear involution
8 of the underlying real Lie algebra of . Here 8   is anti-linear if
8:9<;=?>  ;@8:9A=?>
= 7
; 7

and ; is the complex conjugate of ; . 8 is called the conjugation of with respect to

. We state without proof (see Helgason [chapter 3 § 6]) the following:

Theorem 12.11
Every complex semi-simple Lie algebra has a compact real form.

We prove that the compact real form is unique in some sense. More generally
we have:

Proposition 12.12
Let be a semi-simple real Lie algebra and let be the complexification. Let
be a compact real form of and let 8 and B be the conjugations of with respect to
0 and respectively. Then there is 27DCFEHG 9 > such that the compact real form 
is invariant under 8 .

Proof
 '
Notice that the conjugation with respect to  is B ' BJI and that  is
8 / invariant iff 8 B '  B ' 8 %

165
 Since is compact we can define a Hermitian inner product 

 on by

        

Now let  !" and it is easy to see that
  # $%&'    
Observe also that ()*,+.-/ Clearly the self-adjoint operator 0#12 has positive
eigenvalues so that the one-parameter group of automorphisms

04357698:4;=<?>A@0# ;' 
is well-defined. It follows that 0 3 CBEDGF/   Again (0 3 H#0I+ 3
We want to use 0KJ as L . So let  M0 3 0 + 3  Then
-
" - NO0 3 (0 + 3 N"(0 "+ 2 3 70 +"2 3
 - N0P3(0 + 3 N0 2 3Q R+ - N +R- 0 2 3570 2 3 +R-
Hence for ;  T - 4" - # - and the proposition is proved.
S

Corollary 12.13
Let be a complex semi-simple Lie algebra. If U and
- are compact real forms
then there is an automorphism LVWBEDXFY  such that L UZ -
Proof
Let [U and  be the conjugations with respect to U
- and
- respectively. By the
proposition we can assume that is invariant under 
- U and so

-  U'\ -] ^ U/R\ -

But since
_  is negative definite on
- and positive definite on ^ U we have ^ U\ - 
so UC
-
Remark
It follows from this and the theorem 12.8 and 12.11 that the classification of compact
semi-simple simply connected Lie groups is equivalent to the classification of complex
semi-simple Lie algebras.

We now turn to the study of non-compact semi-simple Lie algebras and show that
this is closely related to the study of orthogonal involutive Lie algebras.
Let  U be a real semi-simple Lie algebra, let  U be the complexification
and let  be the conjugation with respect to U . Now let be a compact real form with
corresponding conjugation of . By proposition 12.12 we can assume that and 
commute or equivalently that U is 9 invariant. Then
U` U ] UP U` U'\  U` U'\a^
166
is the eigenspace decomposition of


with respect to . This is called a Cartan
decomposition of . Notice that is positive definite. It follows that the bilinear
form  defined on by

          
is positive definite and   invariant. Therefore is compactly embedded in
if just   is closed in  ! #" But this is obvious since $  is the identity
component of the subgroup of $%&  fixed by the involution ')(*
,+-'.+/
0"
Therefore the triple
 1    has all the properties of an orthogonal invo-
lutive Lie algebra except that may contain a non-zero ideal of . This may very
well happen if in the decomposition of corollary 12.4 contains a compact ideal. We
therefore restrict to simple Lie-algebras:

Corollary 12.14
If is a non-compact simple real Lie algebra then there is a corresponding orthogo-
nal involutive Lie algebra
24365879:;9=<>:  8?
is a Cartan decomposition


of . The bilinear form is the restriction to of the Killing form .

Exercise 12.15

@BA $ #"

Show that a Cartan decomposition is unique up to isomorphism with an element

Example 12.16
Suppose is a complex semi-simple Lie algebra and let C
be a compact real
form. Then it is easily seen that the underlying real Lie algebra of is also semi-simple
and that  ?ED is a Cartan decomposition  D "
F
Notice also that if is simple then
the underlying real algebra is also simple so this gives an example of corollary 12.14.
(In fact suppose C is a real ideal, then
K CLD G G,
and D H D K
I
are complex ideals so by
K N
simplicity  ?ED " But then JD so also  J =M
hence J
which contradicts the semi-simplicity of .)

Exercise 12.17
Describe the Lie algebra and its Cartan decomposition for the following non-
compact simple groups:
a) O%PRQ ES The group of QUTVQ matrices of determinant one.

b) SO XW
Y :The group of matrices of OZ!XW[H Y% which leaves invariant the sym-
metric bilinear form:

]\_^#`Z^4a\0b6`cbdfege&eca\%h>`ih/HI\%h&jk^`gh&jk^lHNege&eHI\%h&jnmg`ihjom"

167
c) 
 The group of matrices of  leaving invariant the alternating
bilinear form
 "!#!$%!&'!)(*(+(, -.!/0%1!/-32

168
Chapter 13 THE STRUCTURE OF ORTHOGONAL
INVOLUTIVE LIE ALGEBRAS

In this chapter we shall reduce the classification of orthogonal involutive Lie algebras
(and equivalently of simply connected symmetric spaces) to that of compact simple
Lie algebras and their involutions. In particular we shall establish a duality between
symmetric spaces of compact and non-compact type generalizing the classical “duality”
between spherical and hyberbolic geometry.
Recall that an orthogonal involutive Lie algebra is a triple

 where  is
an involution of the Lie algebra , and if and denote the  and eigenspaces
of  , then  is a positive definite symmetric bilinear form on . Furthermore  is
  -invariant, is compactly embedded in and finally no non-zero ideal of is
contained in . Also  denote the Killing form on .
We need the following 3 lemmas:

Lemma 13.1
a. 

b.  is negative definite.

c. If
    invariant subspaces such that 
 !#"%$&('*)
+,

-

Proof
a) Clearly  is  invariant so /.0
123 4/.0
51# for .6
7186 
b) follows from proposition 12.7 provided does not contain any element in the
center of . But suppose . is such an element then 9;: <= .?> is a one-dimensional
ideal of contained in , which is a contradiction.
c) Let .*6
7186
A@BC) .0
1D+ 6  Then E F)G1H
I@J+K6  Hence
L@4
@MNB) .O
1 +P
I@M Q/.0
ERSQ
and hence @TU by b).

Lemma 13.2
Let  = @!6 &)G@4
+,Q->V
Then is an Abelian ideal contained in . Furthermore is semi-simple iff B
Proof

169
 is an ideal: In fact
 by definition, and if   and  then


  

  

 !
for "
so 
#$ .
Also %
is an ideal so %  !& On the other hand is '( invariant so
 % ) & Hence
is Abelian.
It follows that if is semi-simple then * by lemma 12.2. On the other hand
suppose is not semi-simple. Then
,+-. /10$2
345768" 9
is an ': invariant ideal and by lemma 13.1, b) %   so ) & Hence by lemma
13.1, c),
;* so ) & Therefore non-zero gives *
< =&

Lemma 13.3
Suppose is semi-simple. Then >
@?BADCE2
F':
G3 is a maximal orthogonal
involutive Lie algebra.

Proof
Notice that if K!L
':L
GNL M and is semi-simple then also L is semi-
2
H':
GI3J)
simple by lemma 13.2 because L  . It is therefore enough to prove O

since
then also L  P L
L Q 
R &
So let .
TS ) & By the Jacobi identity
#) and clearly
)
so is an ideal of . Let U be the orthogonal complement with respect to 0 . So
 V U and P
U Q W!& In particular UX) Y!& Hence  &

Definition 13.4
2
H':
GZ3 is called irreducible if it is not Euclidean and if does not contain a
proper [B\2 3 invariant subspace.

We can now prove:

Theorem 13.5
Let 2
]':
FGZ3 be an orthogonal involutive Lie algebra. Then there is a decomposition
 ^FV _4V5`a`a`bV c
into a direct sum of ': invariant ideals, such that there are corresponding decompositions
 ^FV _;V7`d`a`1V c
and
 ^]V _4V5`a`a`bV c 2G82 ef
g-3@!
ih#k
< j!3
of and , and such that

170
1.

 
is Euclidean,

2.     are irreducible and  are semi-simple.

Furthermore the decomposition is unique up to a permutation of the irreducible
factors.

Proof
Define   by

!#"$&%'()*+#", +#".- /

Then clearly  has real eigenvalues 01
2%4350768:98989:#0<;>=?@3/ Let
be the eigenspace

belonging to 01
A%B3 and let C be the eigenspaces belonging to 0D!E%B3F HGI3/
Clearly
and C JK%I89:989LM= are mutually orthogonal (with respect to ). Each
C we split into an orthogonal sum of irreducible NOP #Q invariant subspaces. So let
6 8989:9L RKKST=U be all these NO7 VQ invariant subspaces and let 0W6:898989L50DR be the
corresponding eigenvalues for  . By construction

%
X 6YX 98989 X R


is an orthogonal direct sum with respect to and since  C is a multiple of we have

F Z[&%\3 2%^
E ]_M)]`@3/

It follows by lemma 13.1, c) that

a
7 Zcbd%43 e%^
E ]f]H@3/

Now define
a
*% d gbih & >j/

By construction and by the Jacobi identity are NkOP VQ invariant. Similarly

a
d Zcbd%\3l 2%^
E ]Jmjno]H@3/

Hence pqr%f:98989cMS5 are ideals and

a
7 Zcbd%\3 2%s
E ]T*M]Hj/
  a
For t $  U% 0Du with 0DvE% 3 and  W wb is negative definite. Hence

 x&ly is non-degenerate, i.e., is semi-simple. Let

C % 6 X 989:9 X RM/

Then clearly C is a semi-simple ideal so if
+%fz[m-  {+ C &%\3i| then %

X Cg/

171
 Again is an
 invariant ideal and   by definition. Also let   
Then
  and since   
  we conclude from lemma 13.1, c) that
   so in particular 
    is Euclidean. This proves the existence
of a decomposition.
To prove the uniqueness first notice that    must necessarily be
eigenspaces for the operator , so in particular the decomposition
 !" where !  #$% &
is unique. It follows that the decomposition of 
'( into a Euclidean and a semi-
simple part is unique. We can therefore assume semi-simple. Notice also that

)( is irreducible iff there is no decomposition  * + into
, invariant
ideals. It follows using corollary 12.4 that if 
)( is irreducible and is semi-
simple then either is simple, or -  with a simple ideal of and

interchanges the factors. Therefore if is semi-simple with involution
it follows
easily from corollary 12.4 that has a unique decomposition into
 invariant ideals
(up to a permutation) which proves the uniqueness of the decomposition in the theorem.

Remark
It follows from theorem 13.5 that if 
./ is an orthogonal involutive Lie
algebra then  ! with ! semi-simple and a subalgebra of the Lie algebra of
the Euclidean group. Hence by corollary 12.5 is the Lie algebra of some Lie group.
This proves proposition 11.11.

Using the 1–1 correspondence between orthogonal involutive Lie algebras and
simply connected symmetric spaces we can also give a geometric formulation of theorem
13.5: We say that a simply connected symmetric space is irreducible if the corresponding
orthogonal involutive Lie algebra is irreducible.

Theorem 13.6
A simply connected symmetric space 0 has a unique decomposition
0 1024350  3  350 6 7.8  
such that 02 is some Euclidean space and 0:9 ';<   is irreducible.
In particular 0 is irreducible iff 0 is not Euclidean and does not factorize into a
product of symmetric spaces.

We shall now investigate the irreducible orthogonal involutive Lie algebras. For
this we introduce the notion of “duality”:
Suppose 
)/ is an orthogonal involutive Lie algebra. Let be the complex-
ification of , and let
also denote the extension of the involution to  As usual 
and consider the real vector space
=
 >; ? 

172
 
is clearly an
 invariant subalgebra of  Define on  by
          
 


and let
 Then it is straightforward to see that

   
  

 
is
 again an orthogonal involutive Lie algebra. is called the dual of

   Obviously

   
 

Lemma 13.7
! ! !
1. If

  "! 
 

as in theorem 13.5, then ! !

 #
  $ %! 
 


  & 
 (' ) 
 
2. is Euclidean.

  
 
3. is irreducible iff is irreducible.

 
4. If is irreducible then precisely one of or is compact.

Proof
1) and 3) are obvious.
  +  + , -  
4) By the proof of theorem 13.5, *  Hence *
 +  so either *  or *   is negative definite (and not both are). Now by lemma
13.1 *  is negative definite so 4) follows from proposition 12.7.

  is isomorphic to its dual, then it cannot have any irreducible factors
2) If
by 4).

Definition 13.8

 
If is compact then and the associated
 symmetric space . is said to

 
be of the compact type. If is not compact then and . are said to be of
the non-compact type.

Remark

173
 By theorem 12.8 a symmetric space of the compact type is in fact a compact space.

The compact and non-compact types are also distinguished by the sectional cur-
vature:

Proposition 13.9
Suppose is an irreducible symmetric space with associated orthogonal involutive
Lie algebra
  . Let be identified with the tangent space at the point  .
Then the sectional curvature for a two plane  is given by



 !"# $%&
where '(*) is an orthonormal basis for  and ,.
+ -0/21435/7698;:=<?>@A$B C*D

In particular
FEG- or HG- according to whether is of the compact or
non-compact type.

Proof
As noticed above there is IJ
+ - such that A$B JKD Then


 LAK
M
N$**AO
PQ !RSRP*
AO
P $%NT !UPD

Now  $%WV and $B is negative definite. Hence
 and  have opposite sign
and since is of the compact type iff YX- this proves the proposition.

Remark
One can prove that if is simply connected and has non-positive sectional
curvature then Z;[ \ ]
C_^ is a diffeomorphism (see e.g. Helgason [chapter I,
§ 13] or Milnor, [theorem 19.2]). In particular a simply connected symmetric space of
the non-compact type is diffeomorphic to Euclidean space.

We now have the following classification of the irreducible orthogonal involutive

Lie algebras:

Theorem 13.10
The irreducible orthogonal involutive Lie algebras
5* fall into the following
4 disjoint classes:
I. is a compact simple Lie algebra,  is an automorphism of order 2, C`$B a$X
-D
II.  d b cebefb is a compact simple Lie algebra, 
U$Kg
RhPiAj V
b k9:ml noA$B #.Xp- D
174


   
III. is a non-compact simple Lie algebra and is simple. is a Cartan
decomposition and

 is a Cartan decomposition with

IV. is the underlying real Lie algebra of a complex simple Lie algebra, and is
 and   
a compact real form. Again

Classes I and II are the ones of compact type and classes III and IV are the ones
of non-compact type. Furthermore duality interchanges class I and III and interchanges
class II and IV.


Proof
  
Since in all these classes cannot be decomposed into smaller
!"
invariant ideals,

$# %&
is irreducible in all cases. Also it is clear from the above that
with
Now suppose
 '( 
in the compact cases and in the non-compact cases.
is irreducible and is compact semi-simple. Then as

*)
) )
remarked in the proof of the uniqueness of the decomposition in theorem 13.5, either

 )
is simple in which case we are in class I, or where is a simple ideal and
interchanges the factors. Since is compact also is compact and we are in class II.

+-,/.0+1+2+ and +1+ , .0+-3

Also it follows using proposition 12.9 that

and that III and IV are disjoint.

By duality every irreducible orthogonal involutive Lie algebra with non-compact
is either in I* or II* hence either in III or IV and it follows that I* = III and II* =
IV. This ends the proof.

An irreducible symmetric space is of course said to be of class I, II, III or IV

according to the class of the associated orthogonal involutive Lie algebra. Thus the
symmetric spaces of class II are exactly the simply connected simple compact Lie groups
with a bi-invariant metric. By duality the classification of these is equivalent to the
classification of complex simple Lie algebras, which is given by the so-called Dynkin-
diagrams (see e.g. Humphreys [3]). For a complete classification of the symmetric
spaces of class I (or by duality of class III) see e.g. Helgason [chapter 10, § 6] or Wolf
[chapter 8, § 11]. Notice that this also gives a classification of the non-compact simple
Lie-algebras by corollary 13.13 and the exercise 12.15.

Exercise 13.11
Each Cartan decomposition of the Lie algebras considered in exercise 12.17 gives
rise to an irreducible symmetric space of type III. Describe in each case the correspond-
ing dual symmetric space of the compact type I.

175
References
The following are the references used i the text.
1. M.J. Greenberg and J.R. Harper: Algebraic Topology, A First Course, Benjamin-
Cummings Publ. Co., London, 1981.

2. S. Helgason: Differential Geometry, Lie Groups, and Symmetric spaces, Pure and
Applied Math. 80, Academic Press, New York, 1978.

3. J.E. Humphreys: Introduction to Lie Algebras and Representation Theory, Graduate

Texts in Math. 9, Springer-Verlag, Berlin-Heidelberg-New York, 1972.

4. S. Lang: Introduction to Differentiable Manifolds, Interscience Publ., London, 1962.

5. J. Milnor: Morse Theory, Annals of Math. Studies 51, Princeton University Press,
Princeton, 1963.

6. M. Spivak: A Comprehensive Introduction to Differential Geometry, I-V, Publish or

Perish, Boston 1970–75.

7. F. Warner: Foundations of Differentiable Manifolds and Lie Groups, Graduate Texts

in Math. 94, Springer-Verlag, Berlin-Heidelberg-New York-Tokyo, 1983.

8. J.A. Wolf: Spaces of Constant Curvature, Publish or Perish, Boston, 1974.

177