Morton L.

Curtis

Matrix Groups

I
Springer-Verlag
New York Heidelberg Berlin
Dr. Morton L. Curtis
Department of Mathematics
Rice University
Houston, Texas 77001
USA

AMS Classifications: 20-01, 20G99, 22E20

L1brary of Congress Cataloging in PubUcation Data

Curtis, Morton Landers, 1921-

Matrix groups.

(Universitext)
Includes index.
I. Matrix groups. I. Title.
QAI71.C87 512'.2 79-23523

All rights reserved.

No part of this book may be translated or reproduced in any form

without written permission from Springer-Verlag.

© 1979 by Springer-Verlag New York Inc.

987654321

ISBN-13: 978-0-387-90462-7 e-ISBN-13: 978-1-4684-0093-9

DOI: 10.1007/978-1-4684-0093-9
 To my teacher and friend
Raymond Louis Wilder
this book is affectionately dedicated.
Introduction

These notes were developed from a course taught at Rice Univer-

sity in the spring of 1976 and again at the University of Hawaii in

the spring of 1977. It is assumed that the students know some linear

algebra and a little about differentiation of vector-valued functions.

The idea is to introduce students to some of the concepts of Lie

group theory-- all done at the concrete level of matrix groups. As

much as we could, we motivated developments as a means of deciding

when two matrix groups (with different definitions) are isomorphie.

In Chapter I "group" is defined and examples are given; homo-

morphism and isomorphism are defined. For a field k

denotes the algebra of n x n matrices over k We recall that

A E Mn(k) has an inverse if and only if det A # 0 , and define the

general linear group GL(n,k) We construct the skew-field E of

quaternions and note that for A E Mn(E) to operate linearlyon Rn

we must operate on the right (since we multiply a vector by a scalar

on the left). So we use row vectors for Rn, cn , E n and write xA

for the row vector obtained by matrix multiplication. We get a

complex-valued determinant function on Mn (E) such that det A # 0

guarantees that A has an inverse.

Chapter 11 introduces conjugation on R C Cc E and then an

inner product (,). Basic properties of (, are given and then

for k E [R, C, E} we define the orthogonal group

~(n,k) [A E Mn (k) I (xA,yA) (x,y) for all x,y E kn }

 viii

~(n,R) is written ~(n) and called the orthogonal group.

~(n,c) is written U(n) and called the unitary group. ~(n,H) is

written Sp(n) and called the symplectic group. If A E ~(n) then

det A E (l,-l} and the subgroup with det = I is denoted by SO(n)

and called the special orthogonal group. If A E U(n) then det A

is a complex number of unit length. The subgroup with det = I is

denoted by SU(n) and called the special unitary group. As a first

example of a matrix group isomorphism we show that Sp(l) ~ SU(2) .

In Chapter 111 we define the first invariant (i.e., something

unchanged by an isomorphism) of a matrix group; namely, its dimension.

A tangent vector to a matrix group G is y'(O) for some differen-

tiable curve y in G with y(O) = I The set TG of all tangent

vectors is shown to be a vector space, areal subspace of Mn(k)

(k E (R,C,H} ). The dimension of TG (as areal vector space) is

the dimension of G. Smooth homomorphisms are defined and shown

to induce linear maps of tangent spaces. Then dimension is seen to

be an invariant.

In order, in Chapter IV, to calculate the dimensions of our matrix

groups we develop the exponential map exp: Mn(k) ~ GL(n,k) and

the logarithm, log: U ~ Mn(k) where U in some neighborhood of I

in GL(n,k) We have that exp and log are inverses. exp: V ~ U

and log: U ~ V where V is a neighborhood of 0 on Mn(k) and

U is a neighborhood of I in M (k) (actually U c GL(n,k)). One-

n
parameter subgroups are defined and proved to be determined by their

derivatives at O. It follows that TG can be taken to be all

derivatives of one-parameter subgroups. Lie algebras are defined and

 ix

we see that each TG is a Lie algebra. Finally, we then calculate

the dimensions of SO(n) , U(n), SU(n) and Sp(n)

In Chapter V we consider the very specific question of whether

Sp(l) and SO(3) are isomorphie. We get a surjective homomorphism

0: Sp(l) 7 SO(3) with kernel = [l,-l} Then we define the center

of a group, show it is an invariant and then calculate Center Sp(l)

= [l,-l} and Center SO(3) = [I} , proving that Sp(l) ~ SO(3)

We define quotient groups and then note that we get new groups
G whenever a matrix group has nontrivial center.
center

In Chapter VI we do some topology which is needed in other parts

of the text. All of our matrix groups are in some euclidean space

and we just do topology of subsets of euclidean spaces. We give some

basic results about continuity of functions, connected sets and compact

sets. The proof that continuous functions preserve compactness

relegated to an appendix. We consider countable bases for open sets

since this is needed later in our study of maximal tori in matrix

groups. Finally, there is a short section on manifolds.

Chapters VII, VIII, and IX are devoted to studying maximal tori

in our matrix groups. We describe certain specific maximal tori.

We prove that any two maximal tori are conjugate and that, if G is

connected, then these conjugates cover G. At this stage we then

know the dimension, center and rank of all of our matrix groups and

these suffice to settle our original question as to which of these

groups are isomorphie. At the end of Chapter IX we discuss simple

groups and covering groups. The only new groups which arise are the

double covers of SO(n) (n = 3,4, ... ). This leads to the question:

 x

Is the double cover of SO (2n+l) isomorphic with Sp(n) ?

In Chapter X we construct the double cover Spin(n) of SO(n)

using Clifford algebras. We show that Spin(l) ~ S

o , Spin(2) ~ S
1
,

Spin(3) ~ Sp(l) (= S3) and Spin(4) ~ Sp(l) x Sp(l). Finally we

show that

Spin(5) ~ Sp(2) and Spin(6) ~ SU(4) .

In Chapter XI we finish our job by showing that

Sp(n) ~ Spin(2n+l) for any n> 2. This is done by looking at

normalizers of maximal tori and resulting Weyl groups. If the normal-

izer is the semidirect product of the torus and the Weyl group, we

say that the normalizer splits. If Spin(2n+l) and Sp(n) were

isomorphie we would have

Spin(2n+l) ~ ~
center - center

So our result is a consequence of the following three results.

(*) The normalizer in Sp(n) does not split for any n.

The normalizer in ~ splits ~ n E [1,2} .

center

Spin(2n+l)
The normalizer in SO(2n+l) ) splits
center
for n = 1,2,3, ...

Finally, in Chapter XII we give abrief introduction to abstract

Lie groups.
CONTENTS

Chapter 1 General Linear Groups 1

A. Groups 1
B. Fields, Quaternions 7
C. Vectors and Matrices 12
D. General Linear Groups 15
E. Exercises 19

Chapter 2 Orthogonal Groups 23

A. Inner Products 23
B. Orthogonal Groups 25
C. The Isomorphism Question 29
D. Reflections in Rn 31
E. Exercises 33

Chapter 3 Homomorphisms 35
A. Curves in a Vector Space 35
B. Smooth Homomorphisms 41
C. Exercises 43

Chapter 4 Exponential and Logarithm 45

A. Exponential of a Matrix 45
B. Logarithm 49
C. One-parameter Subgroups 51
D. Lie Algebras 56
E. Exercises 59

Chapter 5 SO(3) and Sp(l) 61

A. The Homomorphism 3
p : S ~ SO(3) 61
B. Centers 64
C. Quotient Groups 67
D. Exercises 71

Chapter 6 Topology 73
A. Introduction 73
B. Continuity of Functions, Open Sets, Closed Sets 74
C. Connected Sets, Compact Sets 79
D. Subspace Topology, Countable Bases 82
E. Manifolds 86
F. Exercises 89
Chapter 7 Maximal Tori 92
A. Cartesian Products of Groups 92
B. Maximal Tori in Groups 95
C. Centers Again 100
D. Exercises 104
 xii

Chapter 8 Covering ~ Maximal ~ 106

A. General Remarks 106
B. (t) for U(n) and SU(n) 108
C. (t) for SO(n) 111
D. (t) for Sp(n) 116
E. Ref1eetions in Rn (again) 119
F. Exereises 122
Chapter 9 Conjugaey of Maximal Tori 124
A. Monogenie Groups 124
B. Conjugaey of Maximal Tori 126
C. The Isomorphism Question Again 127
D. Simple Groups, Simp1Y-Conneeted Groups 129
E. Exereises 132

Chapter 10 Spin(k) 133

A. C1ifford A1gebras 133
B. Pin(k) and Spin(k) 137
C. The Isomorphisms 142
D. Exereises 144
Chapter 11 Normalizers, Wey1 Groups 145
A. Normalizers 145
B. Wey1 Groups 149
C. Spin(2n+1) and Sp(n) 151
D. SO(n) Sp1its 156
E. Exereises 162

Chapter 12 Lie Groups 163

A. Differentiab1e Manifo1ds 163
B. Tangent Veetors, Veetor Fie1ds 164
C. Lie Groups 172
D. Conneeted Groups 177
E. Abe1ian Groups 182
Appendix 184
Index 186
Chapter 1
General Linear Groups

A. Groups

Before we can discuss matrix groups we need to talk a little

about groups in general. If X and Y are sets, their Cartesian

product X x Y is defined to be the set of all ordered pairs (x,y)

with x E X and y E Y. A convenient notation for describing this

set of all ordered pairs is

X x Y (x,y)lx EX and y E Y} ,

the curly brackets being read as "the set of all" and the vertical

bar as "such that."

By a binary operation p on a set S we mean a function

p SxS~S

i.e., for an ordered pair (5 1 ,5 2) of elements of S p assigns

another element of S which we write as P(sl,s2) For example, the

set N = (1,2,3, ... } of natural numbers has two well-known binary

operations on it. Addition sends the ordered pair (a,b) of natural

numbers to the natural number a + b. Multiplication sends the

ordered pair (a,b) to ab.

 2

Definition: A ~ G is a set G along with a binary

operation

~:GxG-+G

satisfying certain properties. To state these properties it is con-

venient to adopt a simple notation--for ~(a,b) we just write ab.

Required properties of the operation:

(i) The operation is associative. This means that for any

a,b,c E G we have

(ab)c a(bc) .

(If we had maintained the ~(a,b) notation this would read

~(~(a,b),c) = ~(a,~(b,c))

(ii) There exists an identity element e of G. This means

that for any a E G we have ea = ae = a .

(iii) Inverses exist. This means that for any a E G there is

-1 -1
an element a- l E G such that aa =a a =e .

Note that properties (ii) and (iii) leave open the possibilities

that there may be more than one identity element and that an element

may have more than one inverse. But neither of these can happen.

proposition 1: ~ group G has exactly ~ identity element and

each a E G has exactly ~ inverse.

Proof: Suppose e and f are identity elements of G. Then

 3

fe = e since f is an identity element, and

fe f since e is an identity element .

Suppose both band c are inverses of a. Then

b eb (ca)b c(ab) ce c •

Exaroples

(1) The set Z = ( ••• ,-2,-1,O,1,2, ... } of integers is a group under

addition. ° is the identity and the inverse of a is -a.

(2) z is not a group under multiplication. The operation is asso-

ciative and 1 is the identity. But, for exarople, there is no in-

verse for 2

(3) The set ~ of rational numbers is a group under addition.

(4) The set ~ - (0) (i.e., all nonzero rationals) is a group under
multiplication.

( 5) R+ = (x € Rlx > O} is the set of all positive real numbers. It

forms a group under multiplication.

(6) Rn the set of all ordered n-tuples of real numbers is a group

under the following operation: if

x + Y

The identity is
 4

I!> (0,0, ... ,0)

(7) Let S = [a,b,e} i.e., S is a set with three elements whieh

we denote by a,b,e. Let G be the set of all one-to-one maps

(funetions) of S onto S For example f: S -? S given by

f(a) = b, f(b) e , fee) = a is one element of G. We define an

operation on G as follows: if f,g E G we let

fO g : S -? S

be defined by (fog)(a) = f(g(a», (fog(b) = f(g(b», (fog)(e)

= f(g(e» , i. e. , fo g means first apply g to S and then apply f.

Let i : S -? S be the identity element (i(a) = a, i(b) = b, i(e) = e).

Then this is the identity element for G for this operation. Then

the usual inverse of f E G is the inverse for f relative to this

operation. Thus G is a group. It is ealled the s~etrie group on

[a,b,e} (or just the symmetrie group on three elements).

Definition: A group G is abelian if for every a,b E G we

have ab = ba •

In the examples above, (1), (3), (4), (5), and (6) are abelian

groups, but the symmetrie group on three elements is not abelian.

(Exercise. )

The kind of funetions (mapping one group to another ) of interest

to us are those whieh "preserve" the operations--these are ealled

homomorphisms.

Definition: Let G and H be groups. A funetion cr: G -? H is

 5

a homomorphism if for every a,b in G we have

er (ab) er(a) er(b) .

What this means is that we can first multiply a and b (using the

operation in G) and then map the result by er, or we can map a

and b into H by er and multiply there--with the same result.

proposition 2: ~ homomorphism er: G ...,. H sends identity ~

identity and inverses to inverses.

Proof: Let e,e' be the identities in G,H. We have

er(e) = er(ee) = er(e)er(e) and er(e) has an inverse, call it h, in

H. So

e' her ( e) her(e) er(e) er(e) .

For a € G we have

-1 -1
er(a)er(a ) er(aa ) rr (e) e' ,

-1 -1
showing that er(a ) (er(a))

A homomorphism is surjective (or onto) if er(G) = H. If we

2
define (R additive group of reals, R as in example

(6)) by er(x) = (x,x) then er is a homomorphism but is not

2
surjective because er( R) is just the diagonal line in R But
2
p : R ...,. R defined by p (x, y) = x is a surjective homomorphism.

A homomorphism er: G ...,. H is injective i f er(a) = er(b) always

implies a = b; i.e., no two elements go to the same place. 5ome-

times this is called one-to-one-into, but we won't do that. For

example, the map er: R ...,. R (er(x) = (x,x)) is injective, and the map
 6

2
p R ~ R (o(x,y) = x) is not injective.

A homomorphism ~hich is both injective and surjective is called

an isomorphism. From an abstract point of vie~, t~o groups ~hich are

isomorphic are "really" the same group--even if they ~ere defined in

strikingly different manners. There is a classic example of this.

Let R be the additive group of all real numbers and let R+

(see Example 5) be the multiplicative group of all positive real

numbers. Let a be any real number greater than 1. Define

by

x
cr (x) a

Then cr is a homomorphism

cr(x + y) a(x)a(y) .

Also, a is injective. For, suppose a(x) = a(Y) This means

1 and a X- y = 1 ~hich implies

x - y = 0 or x = y Also, a is surjective. For, if y is any

x
positive real number x = 10gaY has the property that a y

Thus these t~o groups are isomorphic--not only that, but there are lots

of isomorphisms.

We conclude this section ~ith a simple, but important,remark. A

priori it looks difficult to see if a homomorphism cr: G ~ H is

injective. Do ~e really have to check all pairs a,b in G to see

if cr(a) = cr(b)? Fortunately not.

cr is injective ~ cr-l(e / ) e .
 7

-1 -1
cr(a) cr(b) .. cr(a) cr(b) e' .. o(ab ) e'

and

ab- 1 e .. a = b .

B. Fie1ds, Quaternions

Definition: A fie1d k is a set that has operations of addition

and mu1tip1ication satisfying certain requirements:

(i) mu1tip1ication distributes over addition;

a(b + c) ab + ac

(ii) k is an abe1ian group, with identity written as 0,

under addition.

(iii) k - (0) is an abe1ian group under mu1tip1ication.

Examp1es. The rationals ~ and the rea1s Rare fie1ds. We can

make R2 into a fie1d C (the comp1ex numbers) as fo11ows. If

(x 1 ,x 2) and (Y1'Y2) are two ordered pairs of real numbers, we

define (x 1 ,x 2) + (Y1'Y2) = (xl + Y1,x 2 + Y2) and we have seen that

this operation makes R2 into an abe1ian group. Suppose for mu1ti-

p1ication we try

(sure1y the most obvious thing). Then we wou1d have

(1,0)(0,1) = (0,0) .
 8

Now (0,0) is the additive identity or "zero" and we would have two

nonzero elements of p2 with a zero product. The result could not be

a field because:

Proposition 3: In ! fie ld k if a '" 0 and b '" 0 , then

ab '" 0 •

Proof: If a '" 0 then a t k - (0) which by (iii) is required

-1
to be a group under multiplication. Thus there is an a in k - (0)

such that a-la = 1 (the multiplicative identity). Thus if ab =0

we have

-1 -1
but a (ab) (a a)b lb o so b o.

The statement of Proposition 3 is equivalent to the statement

that a field has "no divisors of zero."

2
So how do we make p into a field? Our most naive attempt
failed flat. Well, what turns out to work is

(a,b)(c,d) = (ac - bd,ad + bc)

We must first verify that this distributes over addition.

(a,b)«c,d) + (e,f)) = (a,b)«c + e,d + f))

= (a(c + e) - b(d + f),a(d + f) + b(c + e) .

This should equal (a,b)(c,d) + (a,b)(e,f) . This latter equals

(ac - bd,ad + bf) + (ad - bf,af + be) and we easily check that these

are equal. Next we need to see that if (a,b) '" (0,0) then it has a
 9

multiplieative inverse. WeIl, (a,b) ~ (0,0) ~ a 2 + b 2 ~ 0; in

whieh ease, we need to find a multiplieative inverse for (a,b) . The

multiplieative identity elearly is (1,0) and

a -b
(a, b) (2 2' 2 2) (1,0) ,
a +b a +b

as you ean readily verify. Thus we have made R2 into a field whieh
we denote by C and eall the eomplex numbers.

You may know that there is a simple mnemonie deviee for remem-

bering multiplieation in r.. Write (a,b) =a + ib or a + bi and

treat these as polynomials in i with the side eondition that

Thus

(a + ib)(e + id) = ae + aid + ibe + ibid

= ae + iad + ibe + i 2bd

(ac - bd) + i(ad + be)

We ean eonsider R to be a subfield of C (i.e., a subset

whieh beeomes a field using the operations in the larger set) by

letting

X € ~ be x + iO .

Then if x,y € R we have

x + y = x + iO + Y + iO (x + y) + iO

xy (x + iO)(Y + iO) = (xy) + iO

2
S~we have taken the field R as all (x,O) in ~ and extended
 10

the operations in R to R2 to get a field.

2
This strongly suggests that we try to extend the field on ~

3
to a field on R Now for the bad news.

Proposition 4: The operations on r. cannot be extended to make

R3 into a field.

~i
Proof: Take basis vectors l,i,j so
3
that any element of ~ can be written

uniquelyas a+ib+jc with a,b,cE R 1

So if we are to have a multiplication extending that of c we must

have ij = a+ib+jc for some three real numbers a,b,c. But then

i(ij) = ia+i 2b+ijC so

-j ia - b + ijc

-j ia - b + (a + ib + jc)c

-j (a - b) + i(a + b) + jC 2 .

2
This implies c -1 , contradicting CER.

The main thrust of this proof is that if we insist that the

3
product ij be in R we get into trouble. Maybe if we had one more

dimension it would work. This is almost true; we can define a multi-

plcation on R4 which satisfies conditions (i) and (ii) for a

field but (iii) must be replaced by (iii)' k - (0) is a group under

multiplication--it is not ~ abelian~. We will just describe how

this can be done. You may be interested in reading "Hamilton's dis-

covery of the quatemions" by B. L. van der Waerden in the Mathematics

Magazine (vol. 49, #5, (1976)). We take a basis l,i,j,k for R4

 11

and define 1 i j k
1 1 i j k

i i -1 k -j

j j -k -1 i

k k j -i -1

Thus 1 aets as identity, ij = k, ji = -k , ete.

This te11s us how to mu1tip1y quadrup1es of real numbers:

(a + ib + je + kd)(x + iy + jz + kw) (ax - by - ez - dw)

+ i(ay + bx + ew - dz) + j(az + ex + dy - bw)

+ k(aw + bx + bz - ey)

4
R with this mu1tip1ieation is ea11ed the quaternions. It is easy

to verify that this does extend the mu1tip1ieation in C by taking

e = 0 = d and z o = w in the formu1a above. The modified fie1d

axioms (i), (ii), (iii)' are readi1y verified exeept for showing

that every nonzero quaternion has an inverse. But if

q = a + ib + je + bd

is not the zero (0 + iO + jO + kO) then a 2 + b 2 + e 2 + d2 # 0 and

we set

q
-1 a- ib -
2 2 2
kd ;e -
a +b +e +d

-1 -1
and readi1y verify that qq = 1 = q q

There are eertain eonstruetions we want to make for Rand C

and the quaternions (whieh we denote by 1H), so we will write

 12

k E {R,r:, ll.} •

c. Vectors and Matrices

For k E t F, C, E} let kn be the set of all ordered n-tuples of

elements of k. Define addition on kn by

This makes kn into an abelian group with identity ~ (0, ... ,0) .

For c E k we define

n
and this makes k into a vector space over k (for k = ll. we must

relax the usual definition which insists that k be a field).

Definition: A map kn $ kn is linear if it respects linear

combinations; i.e., if c,d E k and x,y E k n then

(*) ~(cx + dy)

In particular, ~(x + y) = 0(X) + 0(Y) so that a linear map is a

homomorphism of the additive group of kn Also

0(CX)

and these two conditions together are equivalent to (*)

Proposition 5: If kn ~ kn ! kn ~ both linear, then

 13

so is 1\10 $ .

Proof: (I\iO$)(cx+dy)

Definition: Mn(k) is the set of all n x n matrices with

elements from k

If M E Mn (K) M (m ij E k), we can define a linear

map $(M) by

$(M)(X 1 ,···,x) = (x 1 ' ... 'x )(m .. )

n n ~J

where matrix mu1tip1ication is indicated on the right; i.e., we are

mu1tip1ying a 1 xn matrix with an n xn matrix to give a 1 xn

matrix. This is easi1y seen to be linear.

$(M)(cx + dy) = (cx + dy) (m .. )

~J

c(x 1 ,··.,x )(m .. ) + d(Y1' ... 'y )(m .. )

n ~J n ~J

We use row vectors instead of co1umn vectors because we ~ 10nger

have a choice when k = E. We made En into a vector space by

defining sca1ar mu1tip1ication ~ the 1eft,

and this is not the same as (x 1 c, ... ,xn c) in general. If we use

co1umn vectors and mu1tip1y by matrices on the 1eft we do not a1ways

get linear maps. For q,c,d E E and x,y E En consider

 14

(~ .~) (CX ~ l Yl
d )

q cx + dy
n n

and we certainly can't expect this to equal

c (T)
q n
+ d (q~l) qYn

(Take n 1, x 1, y 1, d 0, c = i and q j.)

Conversely, given a linear map f1): k n .... k n it is easy to find

an n x n matrix M such that f1) = ~(M) (and it will clearly be

unique). The first row of M is the n-tuple ~(1,0, ... ,0), the

second row of M is ~(0,1,0, ... ,0), etc.

Note that if the matrix A gives the linear map ~ and the

matrix B gives the linear map W then AB gives ~o~ A linear

map ~ is an isomorphism if it is injective and surjective (same

-1
definitions as for group homomorphisms). Then f1) is also a linear

isomorphism and ~o~-l identity map = 0-lo~ . For the corresponding

-1 -1
matrices this means that M(0 )M(~) = I = M(0)M(0 ) so that

is a 2-sided inverse for M(~) . So if A- l is a left inverse for

A, then it is also ~ right inverse for A.

We make the set Mn(k) into a vector space in a fairly obvious

way:

(i) If A and then

A + B (a .. + b .. )
~J ~J
 lS

(11) If A (a ij ) and c E k, then

This is really no different from the way we made kn into a vector

2
space, but we are now working with n -tuples. However, there is
2
nothing to be gained by writing n elements in a line instead of an

n x n array.

But Mn(k) is not just a vector space. It also has a multi-

plication which distributes over addition (on either side).

A(B + C) AB + AC

(B + C)A BA+CA •

Such a system is called an algebra. When we use the word algebra we

will always mean one with a two-sided multiplicative identity. For

is the multiplicative identity.

D. General Linear Groups

Definition: If u is an algebra, x E U is a unit if there

exists seme y E u such that xy = 1 = yx , i.e., if it has a multi-

plicative inverse.

proposition 6: If u is an algebra with ~ associative

 16

mu1tip1ication and U c Cl is the set of units in Cl , then U is a

group under mu1tip1ication.

proof: The operation is associative, there is an identity

element 1 and every element has an inverse.

Definition: The group of units in the algebra Mn(R) is denoted

by GL(n,R), in Mn(C) by GL(n,r) and in Mn('H) by GL(n,1H)

These are the general linear groups.

Note that: A E Mn(k) is a unit ~ A represents an isomorphism

of kn .

Definition: If G is a group and H is a subset of G, then

H is a subgroup of G if the operation on G makes H into a

group.

proposition 7: H is a subgroup of the ~ G if (H c G and)

(i) x,y E H ~ xy EH,

(ii) id. el. is in H,

(iii)

proof: (Exercise.) The subject of this course is the study of

subgroups of these general linear groups.

A 1 x 1 matrix over k is just an element of k and matrix

multip1ication of two is just multip1ication in k. So we see that

GL(l,R) = R - (0)
GL(l,c) = C - (0)
GL(l,'H) = 1H - (0)
 17

because all nonzero elements are units. GL(2,p) is the set of units

in the vector space M2 (P) of dimension 4 So

a,b,c,d E 1'l.}
GL(2,p) = t(~ ~)
ad - bc " 0

i.e., all points in 4-space not on the set where ad = bc .

For p and C we have determinants defined on Mn(P) and

Mn(C) and from linear algebra we know that

GL(n, p) det A " O}

GL(n,r.) = tA E Mn (c) I det A " O}

Suppose we define a "determinant" on M2 (JH) by

a.ö-Sy.

Then det(i
i
1)
.
= k - (-k) = 2k " 0 , but this matrix cannot be a unit
or the corresponding linear map would be an isomorphism, whereas

(j,-j)(~ J~) (0,0)

and the map is not injective. Similar definitions give similar

problems, but we can define a complex-valued determinant with the

desired property: naroely, A E Mn (H) has an inverse if and only i f

this determinant is nonzero.

Proposition 8: Let dJ: G ~ H be a homomorphism of groups.

Then dJ(G) is a subgroup of H.

Proof: dJ(id) = id so that dJ(G) contains the identity element

of H. If x,y E dJ(G) there exist a,b E G such that dJ(a) = x ,

dJ(b) = y. Then
 18

(/)(ab) E (/)(G) .

-1 -1
Finally, suppose x E (/)(G) . Then x = (/)(a) and so x (/)(a )

E 0(G) . So (/)(G) is a subgroup of H

If 0: G ~ H is an injective homomorphism, then 0 is an

isomorphism of G onto the subgroup (/)(G) of H, so we can then

consider G as a subgroup of H. We are going to construct an

injective homomorphism

'f : GL(n,ll) ~ GL(2n,c)

and then for A E GL(n, lH) we will assign as the determinant of A

the determinant of 'f(A) •

We begin with

defined by

I\i(x+iy+jz+kd) = (
X+i Y
-Z-iW)
z-iw x-iy

Lemma 9:

(11) 1\1 (et8)

(iii) I\i is injective

Proof: (i) is trivial and (ii) is a routine, but somewhat tedious,

computation, and (iii) is trivial.

Next, for A E Mn(ll) we set

 19

Le. 'I'(A) is the cOUlp1ex 2n x 2n matrix whose 2x2 block in the

Lemma 10:

Proof: Let A = (a. ) B (e ) Then

U\I U\I

By LeUlUla 9

and this is just the ij entry in 'I'(A)'I'(B) •

Now let A E GL(n, H) so that there exists A-1 E GL(n, H) with

-1 1
Then 'I' (A) has 'I'(A ) = ('I'(A))- so that 'I'(A)

is nonsingular and thus det 'I'(A) # 0 •

Converse1y, suppose det 'I'(A) # o. Then ('I'(A))-l exists and

since 'I'(GL(n,E)) is a subgroup of GL(2n,C) we have that

-1 -1
('I'(A))-l E 'I'(GL(n,H)). Thus :i[ A E GL(n, E) such that 'I'(A )
-1
= ('I'(A)) • Then

and 'I' is injective so AA- 1 = I. Thus A is nonsingular.

E. Exercises

1. Let rJ>: G ..,. H be a homOUlorphism of groups. The kerne1 of

rJ> is defined to be
 20

ker ~ (x E G I 0(X) identity of H}

Show that ker ~ is a subgroup of G.

2. A subgroup W of a group G is normal if for each x E G

we have

Show that ker ~ (Exercise 1) is anormal subgroup of G.

3. The center C of a group G is defined by

C = (y E G I xy = yx for all x E G)

Show that C is anormal subgroup of G.

4. Let S be a noneropty subset in a group G. Define the

centralizer C(S) of S by

C(S) (x E G I xs = sx for all x E S)

Show that C(S) is a subgroup of G .

s. Let S be a noneropty set in a group G. Define

N(S) = (x E G I xSx- l = S1

and call N(S) the normalizer of S. Show that C(S) c N(S) and

that N(S) is a subgroup of G Show that if S is a subgroup of

G, then S c N(S) and S is anormal subgroup of N(S)

6. Show that if (H I~ E A} is any collection of subgroups of

~

G, then their intersection is also a subgroup of G. If W is any

 21

sub set of G, by the subgroup generated ~ W we mean the inter-

section of all subgroups of Gwhich contain W Show that this is

the smallest subgroup of Gwhich contains W

7. Consider two specific elements of G GL(n,2)

A B

Let H be the subgroup of G generated by A and K be the sub-

-2 -1 2
group of G generated by B Prove that H = { ••• ,A ,A ,I ,A,A , ••• },

and similarly for K.

8. Continuing with exercise 7, show that the product set

HK {hk I h E H, k E K}

is not a subgroup of G. (Show that ABAB is not of the form

A rBs .)

9. We say that a subgroup K of G normalizes a subgroup H of G

i f for each k E K we have kHk- l H Prove that if K normalizes

H, then KH is a subgroup of G

10. We can define an injective map

iq
as follows: represent ~ E C as ~ = pe with p ~ 0 and set

cos e sin e)
JP (
-sin e cos e

Show that <tl (a.S)

 22

11. Let G be the multiplicative group of complex numbers of

th
unit length. We say that 0. € G is a primitive E-- !2!?! of unity
2 n-l
if a.n =1 , but none of 0.,0. , ••• ,0. are equal to one. Show
th
that an isomorphism of G onto itself must send pr~itive n roots
th
of unity to pr~itive n roots of unity for each n For each n,
th
how many primitive n roots of unity are there in G?

12. Let 0. = (a l ,a2 ,a 3) and e = (b l ,b 2 ,b 3) be two elements in

,? Take the two "purely imaginary" quaternions

Show that if 0.' and e' are multiplied as quaternions, then

a.'e' - real part (a.'e')

is just the usual cross product of vectors in R3

Chapter 2
Orthogonal Groups

A. Inner products

We have a consistent notion of conjugation for R C C cE.

Namely,

for x ER, x =x .

For a x + iy E C, a x - iy .

For q x + iy + jz + kYI E E, q x - iy jz-kYI.

We clearly have a a in all cases and

~ + B

It is an exercise to prove that

aB = Ba.

Of course for R or e this is the same as

a.B = a.B

Let k E {R,e,E} and define an inner product ( , ) on kn by

 24

Proposition 1: (,) ~ ~ following properties:

(i) (x,y+Z)· (x,y) + (X,Z)

(ii) (x+y,z~· (x,Z) + (y,z)

(i11) a(x,y)· (ax,n, (x,ay). (x,Y)ä

(iv) (X,y)· (y,X)

(v) (x,X) is always .! real ~ ~ 0 and

(x,X)· o. X· (0, ... ,0) .

(vi) is the standard basis for kn

(ei - (0, ... ,0,1,0, ... ,0)), then

(vii) ~ ~ product!! nondegenerate; !.~.,

If (x,y) o for all y then x = (0, ... ,0)

If (x,y) ° for a11 x then y = (0, ... ,0)

proof: Exercise.

Definition: The length lxi of x € kn is

Ixl=~.

Reca11 that 1f A € Mn(k) , tts conju~ate A is obtained by

replacing each a ij by a ij tA
its transpose is obtained by
replacing each a ij by a ji These two operations commute so that
 25

the symbol

(the conjugate transpose of A) is unambiguous •

Recall that for Hn we must operate on the right (since we
defined (sca1ar)(vector) on the 1eft). So we do the same for ~n

and cn • Thus we use row vectors.

E2! any x,y kn

proposition 2: E and A E Mn(k)
---
we have

t-
(xA,y) - (x,y A) •

Thus the 1eft hand side (xA,y) equa1s

and the right hand side (X,ytÄ) equa1s

It is easy to see that these contain exact1y the same terms.

B. Orthogonal groups

Again let k E (R,e,E} •

 26

Definition:

(t}(n,k) = [A E M (k)
n
I (xA,yA) = (x,y) for all x,y E kn } .

proposition 3: (t}(n,k) is ~~

proof: If A,B E (t}(n,k), then

(xAB,yAB) (xA,yA) (x,y)

so that

AB E (t}(n, k)

C1ear1y the identity matrix I is in (t}(n,k)

If A E (t}(n,k) we have

(eiA,ejA) (ei,e j ) ö ij
- {: if

if
i j

i ,f j

Now eiA is J·ust the i th row of A and we see that (e A e A)

i ' j
is just the ij entry in the product

t- t ÄA
Thus AA = I But then is also the identity since
t(tÄA ) t(ÄtA) AtÄ t Ä A- 1
= = I Thus a 1eft hand and right

inverse for A (More generally, we saw in section C of chapter I

that for matrices a 1eft inverse was automatica11y a right inverse.)

Finally,

-1 -1 -1 -1
(xA ,yA ) (xA ~,yA A) (x,y) ,

-1
showing that A E tt)(n,k) q.e.d.
 27

Definition: For k = R we write ~(n,k) as ~(n) and ca11 it

the orthogonal group. For k = C we write it as U(n) and ca11 it

the unitary group. For k = E we write it as Sp(n) and ca11 it

the symp1ectic group.

proposition 4: Let A € Mn(k) . Then the fo110wing conditons

are equiva1ent:

(i) A € ~(n,k)

(iii) Asends orthonorma1 bases to orthonorma1 bases

(iv) The rows of A form an orthonorma1 basis

(v) The co1umns of A form an orthonorma1 basis

t- -1
(vi) A=A

Proof: Exercise.

proposition 5: Let A € Mn(F) . Then A € ~(n) ~ A preserves

1engths.

Proof: Apreserves 1engths ~ (xA,xA) = (x, X) for all

n
x € ~ So ~ is trivial. Conversely, we have

«x+y)A, (x+y)A) (x+y,x+y) (x,X) + (x,y) + (y,X) + (y,y)

= (xA,xA) + (xA,yA) + (yA,xA) + (yA,yA)

This gives (x,y) + (y,X) = (xA,yA) + (yA,xA) and since (,) over
F is symmetric, this proves
 28

(xA,yA) = (x,y) ,i.e., A E ~(n)

Proposition 5 bis: proposition 5 also holds for c and E.

Proof: Calculate «e.+e.)A,(ei+e.)A) just as above to get

~ J J

(e.A,e.A) + (e.A,e.A) = 0 .
~ J J ~

Then consider x = x.ei + x.e. and calculate (xA,xA) We get

~ J J
xi~(e.A,e.A) + x.xi(e.A,e.A) = 0 and thus
J ~ J J J ~

<eiA,e.A)(x.X.
J ~ J
o

and this forces o q.e.d.

Let us look at ~(n) , U(n) and Sp(n) for small n. ~(l)

is the set of all real numbers of length one, so ~(l) (l,-l) .

U(l) is just the set of all comp1ex numbers of length one. This is

the circ1e group sI. Sp(l) is the group of all quaternions of

unit length. If we define

I}

to be the unit (k-1)-sphere we see that

~(l) = SO, U(l) = SI ,Sp(l) S3.

It is an interesting fact that these are the on1y spheres which can

be groups.

proposition 6: If k E (P,cl and A E ~(n,k) , then

(det A)(det A) = 1 .
 29

Proof: At Ä I = (det A)(det t Ä) = 1, and elearly

det Ä det A q.e.d.

Thus if A E ~(n) (= ~(n,R)), then det A E {l,-l}. We

define

SO(n) {A E ~(n) I det A I}

and eall this the special orthogonal group (also ealled the

rotation group). Similarly, we define

SU(n) (A E U(n) I det A I}

and eall this the special unitary group.

An example of an element of ~(2) - SO(2) is (1 0) This

o -1
sends e l = (1,0) to el and sends e 2 = (0,1) to It is

just the refleetion in the first axis, and has determinant equal to
-1 .

C. The isomorphism question

At the end of Chapter I we showed that two groups whieh were

defined quite differently were isomorphie. We have now defined several

series of groups (GL(n,k) for n = 1,2, ... and k E {R,C,E} and

!!l(n,k) for n = 1,2, ... and k E (R,r,E} ) and our major goal is to

find out whieh of these are isomorphie. The basic idea will be to

develop invariants of matrix groups (dimension, rank, ete.), i.e., two

groups whieh are isomorphie must have the same invariants. This will

make it possible to say that eertain groups are not isomorphie. But

when two differently defined groups are indeed isomorphie, an

 30

isomorphism may be hard to find. This is why we will work so hard to

deve10p invariants--to reduee as mueh as possib1e the eases where we

must look for isomorphisms. In this seetion we will give one iso-

morphism.
Suppose you suspeet that Sp(l) and SU(2) are isomorphie.

How wou1d you try to find an isomorphism? Sp(l) is the set of all

quaternions of unit 1ength and SU(2) is the set of all eomp1ex

t-
2 x 2 matriees A such that AA =I and det A = 1 The operation

in Sp(l) is mu1tipl1eation of quaternions, in SU(2) it 1s matrix

mu1tipl1cation.

Propos i tion 7: The map ~: Mn (ll) ~ M2n (C) defined in §D of

Chapter I induees an isomorphism

~ : Sp(l) ~ SU(2) .

proof: We have seen that ~ induees an 1njeetive homomorphism

of GL(n,ll) into GL(2n,c), so restrietion of ~ to Sp(l) is

still an injeetive homomorphism. So we just need to show that

(i) A E Sp(l) ~ ~(A) E SU(2) and

(ii) every B E SU(2) is some ~(A) with A E Sp(l) .

If A a + ib + je + kd then ~(A) = (a+ib -e-id) so that

e-id a+ib

-e-id) (a-ib
a-ib e+id

Also det ~ (A) 1 so ~(A) E SU(2)

Let B = (~ ~) E SU(2) Using det B = 1 and the fact that

the rows are orthogonal un1t veetors, we find that
 31

ö a. and y e

So, if a. = a + ib and e = c-id, we may take A a+ ib+ jc+ kd

and have ~(A) = B (and a 2 + b2 + c2 + d2 = 1).

D. Reflect~ons in pn

Let u be a unit vector in pn and let

uJ. 01

be its orthogonal comp1ement.

The projection of a vector v into

uJ. is to be v-ru where r E R is

to be chosen so that v-ru is in uJ.

So 0 = (v-ru,u) = (v,U) - r(u,u) and

thus

r = (v,U) .

C1ear1y then the ref1ection of v in uJ. is to be

v - 2ru v - 2(v,u)u .

Choose an orthonorma1 basis with Then,

(0\ .~).
using this basis, the ref1ection ~ is given by the matrix

Lot A be the linear ~p of ." ,iven by ,"nding

to By Proposition 4 A is orthogonal. So
 32

relative to our standard basis the reflection is

given by

Conversely we see that such a matrix represents a reflection in the

orthogonal complement of the vector elA.

2
In let the unit vector u be written as
"
u (cos~, sin~) .

Then (-sin~, cos ~) . is a unit vector in u~. The matrix A

sending el to u and e2 to (-sin ~, cos Cl) must satisfy

(1,0) ( a ll a 12 ) (cos ~, sin rr)

a 2l a 22

(0,1) (all a 12 ) (-sin ~, cos ~)

a 2l a 22

so

COS 0. sin ~)
A (
-sin ~ COS 0.

Thus the matrix giving reflection in u~ is

~ = ( COS 0. sin 0.) (-1 0) ( cos ~ -sin ~\

-sin ~ cos 0. °1 sin 0. cos 0. )

-cos 2rt sin

( 2<1 )
sin 20. cos 2~
 33

The matrix A is easily seen to be a rotation of through an

angle r1.

Later in this course we will prove that ~(n) is generated by

reflections--that iS, any element of ~(n) may be obtained by a

finite sequence of reflections.

E. Exercises

1. Prove Proposition 1.

2. Prove Proposition 4.

3. Let A be any element of ~(n) with det A -1. Show

that

~(n) - SO(n) = [BA I B E SO(n)} .

4. Show that any element of SO(2) can be written as

(cos 8 sin 8)
-sin 8 cos 8

5. If A E U(n) and ).. E C has length one, show that

)..A E U(n)

2
6. Let LI and L2 be lines through the origin in P Show

that reflection in LI followed by reflection in L2 equals a

rotation through twice the angle between LI and L2

7. A matrix A E Mn(F) is said to be idempotent if AA = A .

Show that the image of ~n under A is precisely the fixed-point

set of A Such a map is called a projection of Rn onto its image.

 34

What can you say about the determinant of a idempotent matrix? What

is the image of R2 under A = (~ ~)

8. A matrix A is nilpotent if some power of it is the zero

matrix. For example A = 0

0 0a b)
(000 c has its third power zero. Prove

that a nilpotent matrix is singular. Prove that any A with

whenever i > j

is nilpotent. Find two nilpotent matrices A and B whose product

AB is not nilpotent.

9. Let U be the set of all matrices A withall

diagonal elements equal to one and

whenever i > j

Prove that U is a group under matrix multiplication (the group of

unipotent matrices in Mn(R)) .

Chapter 3
Homomorphisms

We are going to define our first invariant of a matrix group, its

dimension. Matrix $roups whose dimensions are different ean't be

isomorphie. The dimension of a matrix group is going to be the di-

mension of its spaee of tangent veetors (a veetor spaee), so we first

define these.

Let V be a finite-dimensional real veetor spaee. By a eurve

y in V we mean a eontinuous funetion y:(a,b) ~ V where (a,b)

is an open interval in R.

y "",,-y_(e_)~_~ y' (e)

y' (e)

a
•
e b

For e € (a,b) we say y is differentiable at e if

lim y(e+h) - y(e)

~O h

exists. When this limit exists, it is a veetor in V We denote it

by y'(e) and eall it the tangent ~ to Y at y(e).
 36

It is a standard result from ealeulus that if we ehoose a basis

for V and thus represent V as being real

valued), then v' (e) exists if and only if eaeh v: (e)

L
exists and

v'(e) = (vi(c), ... ,y~(e)) .

Now ,
Mn (e) , Mn (H)
Mn (") can all be considered to be real
2 2
vector spaees (of dimensions n , 2n 2 and 4n ). I f G is a matrix

group in then a curve in is a curve in withall

Mn(k) ---- G Mn(k)
values v(u) for u E (a,b) lying in G

zero 0 I rI r E R
matrix

Suppose we have eurves v,a: (a,b) ~ G. Then we ean define a new

curve, the produet ~, by

(va)(u) = v(u)a(u)

proposition 1: Let v,a: (a,b) ..,. G be~, both of ~ ~

differentiable at c E (a,b) . Then ~ product ~ va is dif-

ferentiable at c and

(va)'(c) = v(c)a'(c) + v'(e)a(e) .

Proof: Let v(u) = (vij (u)) , a(u) = (a ij (u)) Then

(va) (u)

so that
 37

(vo) , (u)

v' (u) o(u) + v(u) 0' (u) .

Proposition 2: Let G be ~ matrix group in Mn(k). Let T be

the set of all tangent vectors v' (0) to curves v:(a,b) .... G,

v(O) = I (0 € (a,b)) . Then T is ~ subspace of Mn(k)

Proof: If v'(O) and 0'(0) are in T, then

(vo)(O) = v(O)o(O) = 11 = land

(vo)'(O) = v'(O)o(O) + v(O)o'(O) = v'(O) + 0'(0) .

Thus T is closed under vector addition.

T i8 also closed under scalar multiplication, for if

v' (0) € T and r € R, let

o(u) = V(ru) .

Then 0(0) = v(O) = I o is differentiable and 0'(0) = rv'(O)

Since Mn(k) is a finite dimensional vector space, so is T

Definition: If G is a matrix group, its dimension is the

dimension of the vector space T (of tangent vectors to G at I) .

Example 1: U(l) has dimension 1

i

v
-l.......---+-----<~ 1

U(l)

-i
 38

Example 2: dim Sp(l) = 3

Let v: (a,b) ~ Sp(l) be a smooth curve with v(O) = 1 Then

v' (0) will be an element of E = p.4. We first show v' (0) is in

the span of i,j,k; i.e. it is a quaternion with zero real part.

Let

v(t) = x(t) + i y(t) + j z(t) + k w(t)

with x(O) = 1 and y(O) = 0, z(O) = 0, w(O) = 0 We note

that x(O) is a maximum for the function x so that y'(O)

0+ i y'(O) + j z'(O) + k w'(O) , as asserted.

Conversely, let q = iu + jv + k A be any quaternion with zero

real part. We claim that there exists a smooth curve y in Sp(l)

such that y'(O) = q Indeed,

y(t) =
V 1 - sin 2 \,1t - sin 2 vt - sin 2 ).t + i sin ut + j sin vt + k sin kt

can be readily verified to be such a curve (which is defined on some

interval [O,E) , i.e. for t small).

2
Example 3: dim GL(n, R) n

The determinant function det: Mn (~) ~ R is continuous and

det(I) = 1. So there is some E-ball about I in Mn(R) such that

for each A in this ball
det A # 0 i.e.,

A E GL(n, R) If v is any

vector in Mn(R) define a

curve cr in Mn(R) by
 39

o(t) tv + I .

Then 0(0) = land 0'(0) = v and for sma11 t, o(t) is in GL(n,R).

Hence the tangent space T is all of Mn(~) which has dimension n2

2
A simi1ar argument shows that dim GL(n,C) = 2n .

We will now get upper bounds for the dimensions of O(n), U(n)

and Sp(n) after a few pre1iminaries.

Definition: A E Mn(R) is said to be skew-symmetric if

A + tA = 0 ;

Le. if for each i,j . In particu1ar, the diagonal

terms must all be zero.

Proposition 3: Let so(n) denote the ~ of all skew-symmetric

matrices in Mn(R). Then so(n) ~ ~ linear subspace of Mn(R) ,

and its dimension is

proof: The zero matrix is in so(n) , and i f A,B be10ng to

so(n) , then

(A + B) + t(A + B) o,

so that so(n) is c10sed under vector addition. It is also c10sed

under sca1ar mu1tip1ication, for is A E so(n) and r ER, then

t t t t
(rA) r A so that rA + (rA) = r(A + A) = 0

To check the dimension of so(n) we get a basis. Let

denote the matrix whose entries are all zero except the ij entry,

which is 1, and the ji entry, which is -1. If we define these

 40

E ij only for i < j, it is easy to see that they form a basis for

so(n), and it is easy to count that there are

n(n-l)
(n-l) + (n-2) + ... + 1 = 2 of them .

Definition: A matrix B € Mn(C) is skew-Hermitian if

Thus if b jk = c + id then bkj -b jk = -c id and b kj

= -c + id. In particular if j = k we have c + id = -c + id so

that the diagonal terms of a akew-Hermitian matrix are purely imaginar~

Let su(n) be the set of skew-Hermitian matrices in Mn(C) . By

the observation just made we see that su(n) is not a vector space

over C .

Proposition 4: su(n) C Mn(C) ~ ~ real ~ space of dimension

n(n-l) 2
n + 2
=
2 n.

proof: Exercise.

We make a similar definition for matrices in Mn(E), and call

C € Mn (E) skew-symplectic i f

C + t c o.

In the exercises one shows that the set sp(n) of such matrices is a

real vector space of dimension

3n + 4 n(n-l) n(2n+l) .
2

proposition 5: If e ~ ~ ~ through ~ identity (e (0) = I)

 41

in O(n) then e' (0) is skew-symmetrie

in U(n) then e' (0) is skew-Hermitian

in Sp(n) then e'(O) !! skew-sympleetie

Proof: In eaeh ease we have that the produet eurve is eonstant

t-
B (u) e (u) I.

Thus its derivative is zero, and the result follows from Proposition 1.

Corollary:

Dim O(n) ~ n(n;l)

2
Dim U(n) ~ n

Dim Sp(n) ~ n(2n+l) .

Later we will show that these are equalities.

B. ~ horooroorphisms

Let rtJ: G ~ H be a horooroorphisro of matrix groups. Sinee G and

H are in veetor spaees, it is elear what it means for rtJ to be eon-

tinuous. Froro now on horoomorphism always ~ eontinuous horooroor-

phism. This being so, a eurve

p : (a,b) ~ G

givesaeurve ltl0p:(a,b)~H by (rtJ0p)(u)=rtJ(~(u)) in H.

Definition: A homoroorphism rtJ: G .... H of matrix groups is smooth

 42

if for every differentiable curve p in G, ~op is differentiable.

Definition: Let ~: G -+ H be a smooth homomorphism of matrix

groups. If y'(O) is a tangent vector to G at I we define a tan-

gent vector d~(y'(O)) to H at I by

The resulting map drl>: TG -+ TH is called the differential of ri>.

Proposition 6: dri>: TG -'> TH is ~ linear map .

Proof: If ~'(O) and 0'(0) are in TG , consider

d~(ap'(O) + ba'(O»

with a,b ER. By definition this equals

a(~op)'(O) + b(~oo)'(O) = a d~(p'(O» + b d~(o'(O»

proving that d~ is linear.

proposition 7: If G ~ H 1 K ~ smooth homomorphisms, then ~

is $o~ and

Proof: The first part 1s obv1ous. For the second, let y'(O)

be a tangent vector of G. Then

d(liio~) (v' (0» ('1J0~oy)'(O) = dlii(ri>ov)'(O) = d'1J0d~(v'(O»

Corollary: If rl>: G -'> H 1s ~ smooth 1somorphism, then

 43

d~ : TG ~ TH is ~ linear isomorphism and dim G dim H .

-1
Proof: I/J 0 I/J is the identity, so is the
-1
identity. Thus dI/J is injective and is surjective. I/J I/J is
-1
the identity, so is the identity.. Thus dI/J is

injective and dI/J is surjective. q.e.d.

c. Exercises

l. Let y:(-l,l) ~ M3 (F) be given by

CO"
sin t
y(t) -sion t cos t
0 :)
Show that y is a curve in SO(3) and find y' (0) Show that
2
(y )'(0) = 2y' (0)

2. Let 0: (-1,1) ~ M3 (R) be given by

o
cos t
o(t) - (:
-sin t

Ca1cu1ate 0' (0) Write the matrix for y(t)o(t) and verify that

(yo)' (0) = y' (0) + 0' (0) .

 44

3. Let p (-1,1) ~ M3 (C) be given by

iTTt
e 0 0

i TTt
2
P (t) 0 e 0
i nt
2
0 0 e

Show that p is a curve in U(3) . Calculate p'(O) •

4. Let Cl.: (-1, 1) ~ lH be defined by

Cl.(t) (cos t)j + (sin t)k .

Show that Cl. is in Sp(l) and calculate Cl.'(t) .

5. Let H be a subgroup of a matrix group G Show that TH

is a linear subspace of TG so that dim H ~ dim G

6. Show that the set sp(n) of n xn skew-symplectic matrices

is areal vector space and calculate its dimension.

7. Let T be the set of upper triangular matrices in Mn(R)

That is, A= (a .. ) E T i f and only i f a ij = 0 whenever i :> j
~J

Show that T is a linear subspace of Mn (R) and calculate its di-

mension. Show that T is a subalgebra of Mn (l<) (Le. show that T

is closed under multiplication). Show tbat A E T is nonsingular

(i.e. is a unit) if and only if each a ii ! O. (Note that the group

U defined in Exercise 9 of Chapter 2 is a subgroup of the group of

units in the algebra T .)

Chapter 4
Exponential and Logarithm

A. Exponential of ~ matrix

Given a matrix group G we have defined a vector space T -- the

tangent space to G at I. In this chapter we develop maps to send

T to G and G to T and study their properties. We will work

with real matrices -- developments for c and H are quire analogous.

(We need these maps to determine dimensions of some of our matrix

groups.)

Definition: Let A be areal n)( n matrix and set

where A2 means the matrix product AA, etc. We say that this
2
sequence converges if each of the n real-number sequences

converges.

Proposition 1: For any real n)( n matrix A, the sequence

 46

converges.

Proof: Let m be the largest la ij l in A Then:

The biggest element in the first term is 1

The biggest element in the second term is m.

2
< nm
The biggest element in the third term is -y
2 3
< n m
The biggest element in the fourth term is - 3 : , etc.
2 2 3
nm n m
Any ij sequence is domina ted by l , m ' 2 : ' 3 : ' ... '
k-2 k-l
n m
(k-l) ! , ...

Applying the ratio test to this maximal sequence gives

(k-l)! nm
n
k-2 k-l
m
T

Since n and mare fixed, the ratio goes to 0 as k -"> co ,

proving (absolute) convergence.

x
This exponential behaves somewhat like the familiar e (x € R) •

For if 0 is the zero matrix, we have

e
o I.

Also:

Proposition 2: If the matrices A and B commute, then

A+B AB
e e e

Proof: We will just indicate a proof by looking at the first

few terms.
 47

A2 2 3 2 2 3
A+B
e I+A+B+-2 +AB+~+~+AB+AB
2 6 2 2
+~+
6
2 3 2 3
AB
e e (I+A+ A2 +~ + ... )(I+B+ B2 +B6 + ... )
2 2 3 2 2 3
I+A+B+~+AB+~+~+AB+AB +~+
2 2 6 2 2 6

Corollar;)::: For an;):: real nl<n matrix A e A is nonsingular.

e
0 A-A
proof: A and -A eOUlUlute, so I e

thus 1 = (det e )(det e

A -A ) and det eA 0
'"
FrOUl this eorollary we see that the mBp exp: Mn (R) ~ Mn (R) ,
A
exp (A) = e ,aetually maps Mn(~) into GL(n,R) .

proposition 3: If A ~ ~ real skew-symmetrie matrix, then

A
e is orthogonal.

Proof: We have I = e
o proving
A
that e is orthogonal.

So, if so(n) c Mn(~) is the subspaee of skew-sYUlUletrie matriees,

we see that

exp: so(n) ~ ~(n) .

It is important to note two things whieh Proposition 3 does not

say: (i) it does not say that every orthogonal matrix is some eA

with A skew-sYUlUletrie (Le. it does not say exp: so(n) ~ ~(n) is

A
surjeetive), and (U) it does not say that e orthogonal implies

A is skew-sYUlUletrie. It is instruetive to examine the ease n = 2

in some detail.

The general 2 x2 real skew-sYUlUletrie matrix is of the form

 48

o x
0. - (-x 0)' x e: R •
2
To ca1cu1ate e"' , we ca1cu1ate the powers of 0. • 0.
2
- (-: _:2)
0.
3 -(:3 _:3) CI.
4
- (': X:) CI.
, (0 5 X')
-
-x 0
etc. Then

eCl. _
(10)+(0
o 1 -x
0) 1 ( 0
2 + 3T 3 X)
0
1
+2T (-:
2
_:3 )
-x x
1 x4 0 )
~
+ . ( 0 x 4 +sr
-x 0
S
1 (0 5 x ) + ...

From the 1,1 position we get

x2 x 4 x6
1 - 2T + 4T - GY + .•• - cos x , etc.

We find that

eCl. _ ( cos x Bin x)

-sin x cos x

which is a plane rotation of x radians . Thus for any real 2)( 2

skew-symmetric matrix CI. we have

det eCl. - 1 , i.e., eCl. E S~(2) .

Thus, for examp1e, the ref1ection cou1d never be

obtained this way.
Note also that eCl. =I does not imp1y that 0. is the zero matrix

(0. - (-~n 2;) has ea. = I)

We will see later that these results also hold for larger n
We conclude this section with a simple observation which is
sometimes quite useful in computations.

proposition 4: If A,B are n x n matrices over k E (R, C, JH}

and B ~ nonsingular , ~
 49

BAB- l _ Be~-l
e

B(C+D)B- l - BCB- l + BDB- l these and the definition of the expo-

nential of a matrix yield the result.

B. Logarithm

x
Just as e is defined for all x € R and log x is defined
only for x > O. the logarithm of a matrix will be defined only for
matrices near to the identity matrix I.
Let X be areal n xn matrix and set

(X-I) 2 (X-l) 3 (Xi) 4

log X - (X-I) - 2 + - +

Proposition 5: For X near I ~ ~ converges.

~: Let Y - X - land Y - (Yij) • and suppose each

Iyijl < e. Then
2
< !!L
- 2 •

The ratio test gives

n k k+l k k
~ T-'I'""K = k+l ne ~ ne .
n €

So the series converges for any X such that each entry of X- I

is < n
1
in magnitude.

proposition 6: In Mu(R) let U be ~ neighborhood of I on

 so

which log is defined and let V ~ ~ neighborhood of 0 such that

exp(V) c U Then

(i) for X E U e log X = X

A
(ii) for A E V log e = A

Proof: We do (ii) first. U log e A E V = eA E = A is defined

A A2 A3
(i.e., the series converges). e - I = A + 2T + 3T + So

A A 2 1 A2 2 1 A2 3
log e (A + 2T + ... ) - '2(A + 2T + ... ) + '3(A + 2T + ... ) +

A2 A2 A3 A3 A3
= A + [2T - T] + [6 - T + 3] + ... = A
2 3
(i) is similar. log X = (X-I) - ~
2
+ ~ -
3
1 X (X_I)2 1 (X_I)2 2
e og =[I+(X-I)- 2 + ... )+2T(X-I)- 2 + ... )

1 (X_I)2 3
+ TI [(X-I) - 2 + ... ) +
2 2 333
= X - ~
2 + ~
2 + lr~
3 - ~
2 + Q.:l.L.)
6 + ...

= X •

Proposition 7: If X and Y are near land log X and

log Y commute, then

log (XY) = log X + log Y

So if X is near I and orthogonal, log X is skew-symmetric.

Proof: e log XY = XY = e log X e log Y = elog X + log Y , and e

is one-to-one near 0

Next X and ~ commute so that log X and log ~ commute.

If X is orthogonal
 51

I = xSc
so 0 = log xSc = log X + log Sc = log X + t(log X) showing log X

is skew-symmetric.

We will be ab1e to say more after we have done a 1itt1e topo10gy.

The picture will be

so(n)

t!l(n)-SO(n)

log
o
-
exp
log defined
here

here

C. One-parameter subgroups

Definition: A ~-parameter subgroup y in a matrix group G

is a smooth homomorphism

Note that it suffices to know y on some open neighborhood U of 0

1 n
in F • For X E F , some and y(x) = (y(ti x))
 52

Example: Let k E [R, C, E} and A E Mn (k) Then

uA 2 A2
y(u) = e I+uA+u 2T+

is a one-parameter subgroup of GL(n,k) and y'(O) = A .

proposition 8: Let y be! ~-parameter subgroup of GL(n,k) .

such that

uA
y(u) e

Proof: Let cr(u) log y(u) . Then cr is a curve in Mn(k)

with

y(u) = ecr(u) •

Let cr' (0) =A We just need to show that cr(u) is a line through

o in Mn(k) , for then cr(u) = uA. Hold u fixed.

cr'(u) = lim cr(u+v) - cr(u) = lim log y(u+v) - log y(u)

V70 v V70 v

lim log(y(u)y(v)) - log y(u)

V70
v

Now u + v v + u and y is a one-parameter subgroup so that y(u)

and y(v) commute. Thus

log(y(u)y(v)) log y(u) + log y(v) .

So

o'(u) = lim log y(v) = 0'(0)

V70 V

This proves that cr'(u) is independent of u so o(u) is indeed a

line through 0 in Mu(k) •
 53

So any tangent vector to GL(n,k) is the derivative at 0 of

some one-parameter subgroup. We will see now that this is also true

for the orthogonal groups ~(n,k) .

Proposition 9: Let A be ~ tangent vector to ~(n,k) Then

there exists ~ unique ~-parameter subgroup y in ~(n, k) such that

A y' (0) •

Proof: By definition A pICO) where p is a curve in ~(n,k).

Thus

t --
p(u) p(u) = I

so that

p'(O)+to'(O) 0, Le.,

o.

Now y(u) = e uA is a one-parameter subgroup of GL(n,k) , but it

lies in ~(n,k) because

y(u) t y(u)
--
I .

This proves the proposition. So we have (for GL(n,k) and ~(n,k) )

a one-to-one correspondence between tangent vectors and one-parameter

subgroups.

Taking k = R we have that the tangent space to ~(n) = ~(n,R)

is so(n), the vector space of all skew-symmetric n xn matrices.

Thus dim ~(n) dim so(n) = n(n;l)

Taking k C we have that the tangent space to U(n) ~(n,c)

 54

is su(n), the vector space of all skew-Hermitian n)( n comp1ex

matrices. Thus

2
dim U(n) dim su(n) = n

Taking k '" E we get

dim Sp(n) = n(2n+1) .

What about the dimensions of SO(n) and SU(n)? We will see

in Chapter VI (proposition 3) that the tangent space to SO(n) is

again so(n) , so the dimension of SO(n) is also n(n2-1) . But the

dimension of SU(n) is one 1ess than the dimension of U(n) The

proof of this must also be deferred to a 1ater chapter, but we will

indicate here the resu1t on which it is based.

Definition: The trace of a matrix A = (a ij ) is the sum of the

diagonal terms;

Tr(A) = all + a 22 + .,. + a nn

We c1ear1y have

(i) Tr(A+B) = Tr(A) + Tr(B) and Tr(aA) a Tr(A), (so

Tr is linear) .

Now suppose A '" (a ij ) is real or comp1ex. Then

(ii) Tr(AB) = Tr(BA) .

To prove (ii) we just write it out. The sum of the diagonal terms in
AB is
 55

and the sum of the diagonal terms in BA is

Since ~ and C are commutative one easily checks that these are

equal.

Clearly

(iii) Tr(I) n.

Also

(iv) If B is nonsingular, then

Tr(A)

-1
Proof: By (ii) , Tr(B(AB- l )) Tr( (AB )B) Tr(AI) TrA .

Now we come to the crucial relation.

Theorem: If A ~ ~ real ~ complex matrix, then

Tr(A) A
e = det (e ) .

We will prove this later, but a few comments are in order here.

First off, (f) looks wrong because the left hand side depends only

on the diagonal elements of A and it is not immediately clear that

this is true for the right-hand side. The point is that det and

e are also invariant under conjugation just as (iv) for Tr; so if

B is non singular

-1
A -1
det (e BAB ) det (Be-~ )

We will prove (t) once we have found how to put matrices in simpler

forms by conjugation (in Chapter VIII).

 56

Suppose we know (t). The linear map

Tr : u(n) -> C

actually maps into iR C C since all diagonal terms in a skew-Hermi-

tian matrix are purely imaginary. It is easy to see that Tr(u(n» is

all of iR. From the rank theorem in linear algebra we know that

(all vector spaces being over ~ now)

dim u(n) = dim Tr(u(n» + dim Tr- l (0)

Thus the dimension of Tr-l(O) is just one less than dim u(n) i.e.

But su (n) = Tr -1 (0) is just the tangent space of SU(n) ,

since

Tr (C) o .. 1 = eTr(C)

C
.. e E SU(n) .

D. Lie algebras

It is easy to see that so(n), su(n) and sp(n) ~e not

closed under matrix multiplication. For example, if

2 _x 2 O )
then o. ( 2
o -x

which is not skew-syrornetric.

Proposition 10: For k E [?,C,H} and A,B E Mn (k) we define

[A,B] AB - BA .
 57

Then so(n) , su(n) and sp(n) are closed under [ , ] .

Proof: We need to show that

(AB - BA) + t(AB - BA) O.

The left-hand side is

o .

Thus so(n) , su(n) and sp(n) become algebras (over R) with this

bracket multiplication. This product has some obvious properties.

(i) [A,B] = [B,A]

(H) [A,B+C] [A,B) + [A,C]

[A+B,C] [A,C] + [B,C]

(Hi) For r ER, r[A,B] = [rA,B] = [A,rB]

Finally, , ] has one nonobvious property.

(iv) [A,[B,C)] + [B, [A,Cll + [C,[A,B)] = 0 .

Property (iv) is called the Jacobi identity and its proof is a routine

verification.

Definition: Areal vector space with a product satisfying

(i) ... (iv) is called a Lie algebra. (One could clearly consider com-

plex Lie algebras, but we will have no occasion to do so.)

Let us consider low dimensional Lie algebras. For dim 1 the

vector space is just Rand if x,y E R we have

 58

[x,y] = x[l,y] = xy[l,l] = 0 (by (i)) .

So we have the trivial product (which obviously satisfies (i) ... (iv)).

Consider ,,2 with basis

We must have

Then, for example,

By the Jacobi identity

so

which is true with no conditions on a,b If we take a = 0 = b

we get the trivial Lie algebra. For any other choice we get a non-

trivial Lie algebra. In the exercises one shows that these non trivial

2-dimensional Lie algebras are all "essentially the same."

We will not try to find out all nontrivial 3-dimensional Lie

algebras, but will simply look at two which arise quite naturally.

a
o
-c

clearly has dimension three. Also using basis

 59

[i,j] k
and defining [j ,k] i
[k, i] j

gives a 3-dimensional Lie algebra.

E. Exereises

1. Let A be a 3)( 3 skew-symmetrie matrix. Show that A2 is

symmetrie, but show by example that A3 eould be neither symmetrie

nor skew-symmetrie.

2. Let B € ~(3) - SO(3). Show that the series for log B

does not eonverge.

3. Prove the Jaeobi identity for [A,B] AB - BA

2
4. Prove that any two nontrivial Lie algebras on Rare iso-

morphie as Lie algebras.

5. Show that the two 3-dimensional Lie algebras defined above

(§D) are isomorphie.

Chapter 5
SO(3) and Sp(1)

3
A. The homomorphism p : 5 ~ 50(3)

We have seen that Sp(l) , whieh is all quaternions of unit

length, is just the unit 3-sphere in F,4 ( = E) . Also we have seen

3·2 53
that dim 50(3) =-2-=3 So dimension won't distinguish from

50(3) , and, for all we know now, they might be isomorphie. In this

seetion we define and study an "almost isomorphism" between them.

proposition 1: If q E 53 then the "left translation"

given ~ Lq (q') = qq' is an orthogonal ~ of ,,4 to ,,4.

4
Proof: As veetor spaees over R , JH and R are the same.
4
So L is surely a linear map of F- for i f a,b E R and
q
o.,a E E we have

L (ao.
q
+ ba) aqo. + bqa = aL (0.) + bL (a) .
q q

To see that L is orthogonal, it suffiees to show that L

q q
preserves the perpendieularity (using (, for F,4) of the four

unit veetors l,i,j,k. For example, let q = a + ib + je + kd and

 61

4
calculate (L (i) ,L (j)
q q
(using ( , ) for ~ ) . We get

ad + bc - bc - da = o. For (L q (l),L q (i) we get

(a + ib + je + kd, ai - b - kc + jd) -ab + ab + dc - dc o

The computations for other pairs of basis vectors are simiLar.

Definition of p: For q E S3 and a. E E we define

p(q)(a.) qa.q

That is, we do a left translation by q and a right translation by

4
q • By Proposition 1 this is an orthogonal map of R to 1. e. ,

p (q) E tt; (4) .

Since real quaternions commute with all other quaternions, if

x is a real quaternion

p (q)x qxq xqq x.

Note also that p(q) is the inverse of p(q) in the group tt;(4)

since and similarly for p(q)p(q) •

Together these two observations imply that p(q) maps the 3-space

spanned by i,j,k to itself (Exercise #3). Thus p(q) can be con-

sidered as an element of tt;(3) (Exercise #4).

Fact (to be proved after Chapter VI): p(q) is in S~(3) •

3
Proposition 2· p : S -> Stt;( 3) is a surjective homomorphism and

Ker(p) [1,-1) c S3 .

3
Proof: If Ql,q2 E Sand a. E Span(i,j,k) , then
 62

Tnus p is a nomomorpnism.

Clearly p(l) and p(-l) are tne identity in SO(3) so tnat

1 and -1 are in Ker p . Conversely, suppose p(q) is tne iden-

tity witn q =a + ib + je + kd. Tnen p(q)(i) i gives

(a + ib + je + kd)(i)(a - ib - je - kd) = i. And from tnis we get

But 1 and we eonelude

tnat e = 0 = d From p(q)j = j we get b = 0 Tnen a2 = 1 so

a E (l,-l) •
Finally we need to snow tnat p is surjeetive. This will be

quite easy onee we know some topology (Chapter VI) --otnerwise it is an

almost hopelessly eomplieated eomputation. Here we will just show

tnat we ean find a q E S3 such that p(q) is the element of SO(3)

whieh leaves k fixed, sends i to j and sends j to -i

Let q =a + ib + je + kd We want

(a + ib + je + kd)(k)(a - ib - je - kd) =k , or

(a + ib + je + kd)(ka - jb + ie + d) = k , so

ad - be + be - ad 0 (autom~ieally) ,

ae + bd + ae + bd =0 or 2(ae + bd) 0,

-ab + cd + de - ab or 2(ed - ab) O.

1 .

Now

2 2 2 2
a+d+b+e 1

so
 63

o or b o e .

80 the on1y eondition on q sueh that p(q)k = k is that

q = a + dk (with a2 + d2 1)

Next we want p(q)i j

(a + kd)(i)(a - kd) j

(a + kd)(ia + jd) j

o, a ±d

2
ad + ad 1 if a = d , 2a 1 ,

and we ean't have a = -d. Fina11y we insist that p(q)j -i. 80

(a + ka)(j)(a - ka) -i

(a + ka)(ja - ia) -1

2 2
a a o.
1 1 1
kl:-o
80 q = - + k - or q = Both will give the desired
.fi /2 J'I .fi
element of 80(3) (This shou1d be enough to eonviee us that we

shou1d not try the general proof of surjeetivity at this stage.)

Note that this does ~ prove that 83 and 80(3) are not iso-

morphie. p is not an isomorphism, but one might exist. In the next

seetion we give a fair1y easy proof that 83 t- 80(3)

 64

B. Centers

In Exercise #4 of Chapter I the ~ C of a group G is de-

fined as

C [x E G I xy yx for all y E G} ,

and was shown to be an abelian and normal subgroup of G. We leave

it as an exercise here to show that any isomorphisms of groups induces

an isomorphism of their centers. We will show that s3 ~ SO(3) by

5howing that their centers are not isomorphie.

3
Proposition 3: The center of S = Sp(l) is [l,-l} , whereas

the center of SO(3) i5 (I} .

Proof: Since real quaternions commute with all quaternions, it

is clear that (l,-l} c Center s3 Conversely, suppose

q =a + ib + je + kd E s3 i5 in the center. Then qi = iq gives

ai - b - ck + dj ai - b + ck - dj

so that c = 0 d. Then qj jq gives

(a + ib)j j(a + ib)

2 3
and this implies b o. So q = a and a 1 . Thus Center S

= P,-l}

Suppose A E SO(3) is in the center. Since A commutes with

all elements of SO(3) it surely commutes with all elements of

 6S

T = ('0'"-s~ 9
sin 8
cos 8

° ~)
since T c: 50(3) Consider the standard basis e1 (1,0,0) ,
e 2 = (0,1,0) , e 3 = (0,0,1) for R3 ".

Claim: A 1eaves e3 fixed (or sends it to

Choose B E T which sends to and to and (auto-

matica11y) 1eaves e3 fixed. Then set Ae 3 = ae 1 + be 2 + ce 3 Then

this imp1ies a = 0= b

and since Apreserves 1ength, we must have c = 1, 01 C = -1 .

Thus A induces an orthogonal map of the e 1e 2 plane. Actua11y,

it is a rotation because:

5ub1emma: Any element of &(2) which commutes with all rotations,

is itse1f a rotation.

2 2
Let I/J R ~ R denote such an element of ~(2) . For any

rotation

cos 8 sin 8)
t = (
-sin A cos 8

we roust have ~t t~ Let ~

(0. B) We get
y Ö

0. cos e B cos 8 a. cos 8 + Y sin 8

0. sin 8 + B cos 8 B cos 8 + Ö sin 8

holding for all 8. 50 y -B and 0. &. Thus

8) and de t I/J
2 2
( 0. a. + B
-B a.
 66

Since this cannot equal -1 (and must be in [1,-1)), this proves

the sub lemma.

This also proves that c =1 (not -1) (since A € SO(3))

and we conclude that

A € T •

We can now finish the proof.

(co,
~)
e sin e
A = -Si: e cos A
0

and we let

R = (~ ~ ~)
-1 0 0
€ SO ( 3) .

Since A must commute with, 1\ we get

sin e
co,,) ( 0

C
0
AR cos e -sin e = -sin e cos e
:)- M
0 0 -cos e sin 8

Thus we must have cos e 1 and sin 8 = o. Thus A land

Proposition 3 is proved.

We will calculate the centers of all of the groups SO(n) ,

U(n) , SU(n) ,Sp(n) in a later chapter, after we know about maximal

tori. We conclude this chapter with a bit more abstract theory which

we will need later.

 67

c. Quotient groups

If H is a subgroup of G we define an equivalence relation

_ on G by
-1
x-y if xy € H
-1
This relation is reflexive, x - x since xx =e € H It is
s}'!l!!letric, x _ y => Y _ x, since xy-l € H => .(xy-l)-l - yx- l € H •

It is transitive , x _ y and since xy -1 € Hand

-1
yz € H imply that
into equivalence classes.

Let C(x) denote the class containing x. Then

C(x) Hx - (hx I h € H} .

-1
Also Hx -= Hy .. xy € H .. Y € C(x) .. X € C(y) • These equivalence
classes are called right ~ of H.

Example: Let G = s3 (= Sp(l» and H = {l,-l}. Then

Hq - {q,-q} = H(-q) so each equivalence class contains exactly two

points of S3.

Example: In G = U(3) let

H {AI I A a complex number of unit length}

Then H is a circle subgroup of G and the right cosets are circles.

Thus U(3) can be divided into disjoint circles filling up U(3)
3
Similarly, let G = S - Sp(l) and let H be the circle
223
{a + ib I a + b = l}. Thus S can be divided up into circles.

One defines left cosets in a similar manner

 68

xH {xh I h EH} .

Recall (Exercise #3, Chapter I) that a subgroup H is normal if

-1
xHx = H for all x in G

Observation: A subgroup H of G is normal (in G) = xH Hx

for every x E G .

Let G/H denote the set whose elements are the rightcosets of
H in G

Proposition 4: If H is ~ normal subgroup of G, then the

operation ~ G/H defined ~

(HX) (Hy) H(xy)

makes G/H into ~ group.

Proof: We need H normal to show the operation on G/H is well

defined. 5uppose Hx = Hz and Hy = Hw We must show that Hxy

= Hzw Well, xy(zw) -1 xyw -1 z -1 and yw

-1
= hl E H Also,
-1 -1
z x h2 50

and, since H is normal, h3 E H so that

and we have proved that the operation is well defined.

-1
The rest is easy. H = He E G/H is the identity and Hx is thE

inverse of Hx. (Associativity is inherited from G - (Hx)(HyHz)

= (HxHy)Hz since x(yz) = (xy)z ) .

 69

Example: G = 8p(1) and H = [l,-l} H is the center of G

and thus is anormal subgroup. Thus G/H is a group. We know it

is 80(3) .

There is a natural map Tl: G -> G/H given by Tl(x) = Hx. In the

exercises it is shown that Tl is a surjective homomorphism with ker-

nel H.

Let G be a group and x,y E G Then the element

-1 -1
xyx y

-1 -1
is called the commutator of x and y (because (xyx y )(yx) = xy).
Now the product of two commutators is not necessarily a commutator,

but we set [G,G] = [all finite products of commutators} .

proposition 5: [G,G] g ~ normal subgroup of G and

g ~ abelian group.

Proof: Closure and identity are clear and

-1 -1 -1 -1
(xyx y )(yxy x ) e,

-1 -1
showing [G,G] is a subgroup. Let Z E G and xyx y E [G,G] .

Then

-1 -1 -1 -1 -1 -1 -1 -1 -1
z(xyx y )z = zxy(z (xy) (xy)z)x «yz) (yz))y z

-1 -1 -1 -1 -1 -1
= [z(xy)z (xy) }[x(yz)x (yz) }(yzy z } E [G,G]

This easily extends to products of commutators, so that [G,G] is a

normal subgroup.

Finally, [G,G] x [G,G]y [G,G]xy [G,G]yx [G,G]y[G,G]x since

 70

-1 -1 -1
xy(yx) xyx y E [G,G] .
q.e.d.

In most instances we will encounter, if G is a matrix group and

C is its center, then G/C will have trivial center. But this need

not a1ways be the case.

proposition 6: For x E G define

w(x) : G ..,. G

-1 -1
Ez w(x) (y) xyx y Then G/C has nontrivial center ~ ~x E G-C

such that

w(x)(G) c: C .

Proof: ~

x i C = Cx # C so Cx is not the identity in G/C But for any

-1 -1
Y E G we have xyx y E C so that CxCy = Cxy = Cyx = CyCx and

Cx E center G/C

Converse1y, CX # C with Cx in the center imp1ies

-1 -1
CxCy Cxy Cyx CyCx so that xyx y E C for a11 Y E G •

Once we have done a 1itt1e topo10gy (Chapter VI) we easi1y have:

Coro11ary: If G is connected and C is discrete (in particu1al

II C ~ finite), then G/C has ~ ~.

 71

D. Exercises

1. Let G be a group and x € G. Show that left translation

LX : G -> G by x (Lx(g) = xg) is a one-to-one map of G onto G.

Let Rx be right translation so that

-1
R -1 0 Lx(g) = xgx
x

Show that R -1 0 Lx is an isomorphism of G onto G.

x

2. Do one more step in the proof of Proposition 1 by showing

(L (i),L (k») = 0 .
q q

3,4. These are listed by number in the text.

5. Show that o(i) , p(j) , p(k) are a11 in SO(3)

6. Show that the set T defined in the proof of Proposition 3

is an abelian subgroup of SO(3) .

7. Let rf,: G ... K be a surjective homomorphism of groups and

H Ker rf,. Then we have

G ~ K

I "
G/H

-1
Show that rf, 0 "
is weIl defined and gives an isomorphism of G/H

onto K.

8. Show that (see Exercise #6) the abelian subgroup T of SO(3)

is not anormal subgroup.

9. Show that the subgroup H I} of Sp(l)

 72

is not anormal subgroup.

10. Show an isomorphism of g~ups induces an isomorphism of

their centers.
Chapter 6
Topology

A. Introduetion

Our matrix groups are all subsets of euelidean spaees, beeause

they are all subsets of

There are eertain topologieal properties, notably eonneetedness and

eompaetness, whieh some of our groups have and others do not. These

properties are preserved by eontinuous maps and so are surely invar-

iants under isomorphisms of groups. So a eonneeted matrix group eould

not be isomorphie with a noneonneeted matrix group, and a similar

statement holds for eompaetness. We will define these properties and

deeide whieh of our groups have them. This will be done in seetions

Band C.

In seetion D we define and diseuss the not ion of a eountable basis

for open sets, a eoneept we will need in our study of maximal tori in

matrix groups. Finally, in seetion E we define manifold and show that

all of our matrix groups are manifolds. Then we prove a theorem

about manifolds whieh gives an easy proof that the homomorphism

p : Sp(l) ~ 50(3) (defined in Chapter V) is surjeetive.

 74

B. Continuity of functions, c10sed sets

open~,
- ----

Definition: A metric d on a set S is a way of assigning to

each x,y t S areal number d(x,y) (the distance from x to y)

in such a way that:

(i) d(x, y) ~ 0 and d(x,y) o .. x y ,

(ii) d(x,y) d(y,x)

(iii) d(x,y) + d(y,z) ~ d(x, z)

Condition (iii) is ca11ed the triangle inequa1ity.

,,~U,
~'-i-,1>A.(:,Z)
d(x,y)

We will define such a metric d on Rn and then for any S c Rn

n
d will also c1ear1y be ametrie on S • Reca11 that for x,y t R

we defined an inner product

Set d(x,y) = J(x-y,x-y) (ThUS we define d(x,y) to be the 1ength

of the vector x - y .)

n
proposition 1: This is a metric on R

Proof: Properties (i) and (ii) fo11ow from

(x,X) ~ 0 and (x,x) o .. x o

 75

and symmetry of the inner product. To prove the triangle inequality

we will prove the corresponding property of (, called the

Schwarz inequality.

For any x,y E Rn and t E R we have

(x+ty,x+ty) ~ 0 .

Using the bilinearity and symmetry of (, this gives

2
(x,X) + 2(x,y)t + (y,y)t ~ 0 .

This quadratic polynomial in t with real coefficients is always

~ 0 and thus it cannot have two distinct real roots. (A quadratic

polynomial can have only one minimum.) Thus the discriminant cannot

be positive; i.e.,

2
(2(x,y») - 4(y,y)(x,X\ ~ 0 .

So

2
(*) (x,y) ~ (x,X)(y,y\ .

The inequality (*) is the Schwarz inequality.

We apply (*) to the vectors x - y and y - z to get

(**) (x-y,y-z) ~ J(X-y,x-y) J(y-z,y-Z)

If we square both sides of (iii), write it in terms of (, ) and use

basic properties of (, ), we see that (iii) is equivalent to (**).

n n
We use this metric d on R to define open balls. Let x E R

and r > 0 be areal number. Set

B(x,r) = (y E Rn d(x,y) < r}

 76

and call this the open ball with center x and radius r. Open balls

in euclidean spaces allow us to give a fairly direct generalization of

the not ion of continuity of a function on R to functions defined on

spaces of dimension greater than one.

Let A be a subset of Rn and

be a function defined on A and taking values in some euclidean space

Rm

Definition: To say f is continuous ~ ~ point a E A means:

Given any open ball B(f(a),e) in Rm there exists an open

n
ball B(a,o) in R such that any point x E A n B(a,o) satisfies

f(x) E B(f(a),e) .

Another way of saying this is: Given € > 0 there ex1sts 0 > 0

such that if x E A satisfiee d(a,x) < 0, then f(x) satisfies

d(f(a),f(x)) < e .

Both ways are just precise ways of saying that f 1s continuous

if it sends "nearby points" of A to "nearby points" in

It is important to not1ce that the ~ontinuity of f depends on

the domain A of definition. For example define

by

f(xl " ~: if
if
x < 0
x:::>: 0

Then f is not continuous at o . But suppose we take A C f< to be

 77

all x ~ 0 and restrict f to A

f:A~".

Then this restricted f is continuous at O.

A function f: A ~ Rm (A eRn) is said to be continuous if it

is continuous at each a E A .

Example: If A c 1<n is a finite set then any f: A -'> "m is

continuous.

Proof: For any a E A let bl, ... ,b k be all of the other

points of A Let

i, ... , k

and let ö be the smallest of these. Then for any e > 0 any ele-

ment of A in B(a, ö) goes into B(f(a), e) (because a is the only

such element of A) .

Proposition 2: If A C Rn and

are continuous, then gof is continuous.

Proof: Let a E A and € > 0 Since g is continuous, there

exists n> 0 such that every element of f(A) in B(f(a),n) is

sent by g into B(g(f(a»,e) Since f is continuous there exists

ö> 0 such that every element of A in B(a,ö) is sent by f into

B(f(a),n) and is then sent by g into B(g(f(a»,€) .

q.e.d.
 78

Seme exercises on continuity are given at the end of this chapter.

Definition: A set U C Rn is an open set if each x E U lies

in some B(x,r) c U (where r will depend on x)

Clearly Rn is an open set. It is not quite so clear that the

anpty set ~ is open; but since there is no x E ~, there is no

requirement that some B(x,r) be contained in ~

Example: Any open ball B(y,s) is an open set. Let x E B(y,s)

(Le., d(x,y)<s) We must find r> ° such that B(x,r) c B(y,s).

Well, let r = s - d(x,y). If z E B(x,r) then d(z,x) < s - d(x,y)

and thus

d(z,x) + d(z,y) < s .

Using the triangle inequality we have

d(z,y). s d(z,x) + d(x,y) < s

showing z E B(y,s) .

Example: (0,1) = (x E Rio< x < 11 is an open set in F

(because it is the open ball B(~,~) ) . But

 79

(0,1] = (x E RIO < x ~ 1} is not open in R because 1 E (0,1]

but no B(l,r) lies in (0,1] since every such ball contains num-

bers greater than 1.

Definition: A subset C C Rn is defined to be c10sed if its

comp1ement Rn C is open.

(0,1] C R is neither open nor c10sed. We have seen it is not

open. Let T = R - (0,1] . Then °T E but no B(O,r) can 1ie in

T since each will contain points of (0,1]. Thus T is not open so

that (0,1] is not c10sed.

[0,1] = (x E Rio ~ x ~ 1} is a c10sed set.

Examp1e: Let KeRn be a finite set. Then K is c10sed.

For i f X E there will be a minimum distance ö from

x to points of K. Then

B(x,ö) c Rn - K ,

proving K is c10sed.

C. Connected sets, compact sets

Definition: A set D in Rn is connected if: Given x,y E D

there exists a continuous function

y:[O,l]-+D

(Le., y: [0,1] -+ Rn with y([O,ll) cD)

with y(O) x and y(l) y •

 80

Such a function may be ca11ed a path from X to y in D •

Examp1es: Fn is connected because

y(t) = (x + t(y-x))

is a path in Rn from x to y

D = [x E R x + o} is not connected. For examp1e, no path

from -1 to 1 in R can 1ie in D.

D = [(x 1 'x 2) E R2 (x 1 'x 2) + (O,O)} is connected.
2
Important examp1e: ~(n) C Rn is not connected. The matrices

and are in ~(n). If y: [0,11

~ ~(n) is a path from A to I, then the composite function

y det
[0,11 ~ ~(n) ~ R

wou1d be continuous (Proposition 2) • But it wou1d be a path in

(-1,1) C R from -1 to 1, contradicting the existence of y.

Reca11 that so(n) c Mn(R) consists of skew-symmetric matrices

and that if A E so(n), then exp A c ~(n)

proposition 3: exp maps so(n) into SO(n) .

Proof: For B E so(n) the path

tB
y(t) e

is a path from eO = I to As seen above, this implies that

det e B = +1 so eB E SO(n)
 81

proposition 4: Let D C Fn be connected and

be continuous, then f(D) is connected.

Proof: Given a,b € f(D) choose x,y € D such that f(x) = a

and f (y) b. Choose a path y from x to y in D. Then

f 0 y is a path from a to b in f(D)

Definition: A subset W of Rn is bounded if W lies in some

open ball. This is clearly equivalent to: W lies in some B(O,r) .

Now boundedness, unlike connectedness, is not preserved by con-

tinuous functions. For example, if W = (0,1) C Rand f: W .... F is

defined by f(x) = ~, then W is bounded but f(W) is not bounded.

Ne1ther 1s the property of being closed preserved by continuous

functions. For example, F is closed in Rand f: R .... F defined

by f(x) eX is continuous, but f(R) = [y € R I y > O} is not

closed.

However, when we put closed and bounded together, they are then
both preserved.

Definition: C C Rn is compact if it is closed and bounded.

proposition 5: If C C Rn is compact and

is continuous, then f(C) is compact.

The proof is relegated to an appendix.

 82

D. Subspace topo1ogy, countab1e bases

Sornetimes we will have a subset W of Rn and will want to

know which sub sets of W we shou1d ca11 open ~ in W.

Definition: If U eWe Rn we say U is an open set in W

if there exists an open set V in Rn such that

U V n W •

For examp1e, i f W = [0,1] eR, then

U = (~,11 = l:x E R I ~ < x S 1} is an open set in W, but not an

open set in R U' = [~,1] is not open in W

n
Note that i f W is an open set in R , then UcW is open in
n
W if and on1y i f it is open in R

n
For Wc R the co11ection of all open sets of W is the

subspace topo1o~l of W.

Reca11 that V c Rn is defined to be open if any x E V has

n
sorne B(x,r) c V This is equiva1ent to saying that VeR is open

if it is either the empty set or is a union of open balls. (Exercise.)

Of course, not every open set is an open ball, but open balls suffice

to give all open sets by taking unions (the "empty" union being

allowed) .

n
Definition: A co11ection of open sets in R is a
n
basis for open sets if every open set in R is a union of sorne of

the V 's
n
 83

Examples: The set of all open squares in R2 is a basis for the

2
open sets in R

The set of all open intervals (a,b) c ~ with a and b

rational is a basis for the open sets in R

The set of all open balls in Rn iS, of course, a basis for open

sets. But so is

(B(x,r) I x (xl' ... 'x n ) with each xi rational,}

and r is rational

(See Proposition 7.)

For a subset W of Rn we know which are the open sets in W

and we can give the same definition as above for the notion of a

basis for the open sets in W. Indeed, it is clear that if u = (V~}

is a basis for the open sets in Rn then (V n W} is a basis for

0.

the open sets of W

We want to get bases for open sets which are "minimial" in the

sense that they have no more sets than needed to do the job. The

notion that comes up is countability.

Definition: A set S is countable if its elements can all be

arranged in a finite or infinite sequence sl,s2,s3' ... that is,

every element of S will be somewhere in the sequence.

Examples: The set ~ of all positive rational numbers is

countable; for example

13 1245781
1, 2, 2' 2' 3, 3' 3' 3' 3' 3' 3' 4, 4'

is a sequence containing all positive rationals. Similarly,

 84

1
1, -1, 2, -2, 2' -2' 2'
1 1. -2"3 3, -3, •.. contains all of 0).

The set I = [0,1] = [x E RIO ~ x ~ l} is not countable. We

prove this contrapositively. Suppose

is a list of all elements of I It suffices to give an element of

I which cannot be in the list. Express the r 's as decimals

i

ri .xi/i2xi3·· .

r = 'Y1Y2 Y3··· Yj = 5
"
Let where if x .. 5 and Yj 1 if
JJ
x .. = 5 Then
JJ

"
rl because Yl xn
r "

r " r2
because , etc.
Y2 " x 22

But r EI.

Proposition 6: If A and Bare countable sets, then ~ is

their cartesian product A xB .

Proof: Let A

we can write

because following the path shown below will include all ofAx B .
 85

n
Proposition 7: (and hence any WeR ) has a countable

proof: The set

can be put in 1-1 correspondence with (n+l)-tuples (xl, ... ,xn,r) of

rational numbers (with r > 0 ) . By Proposition 6 this is a countable

set of balls.

Let V be any open set in ~n To show that V is a union of

elements of C it suffices to show that for y E V some B(x,r) E C

contains y and lies in V (For then V is the union of such

B(x,r) -- one for each y E V.) Since V is open seme

B(y,s) c V .

Choose x with all coordinates rational such that

s
d(x,y)-<:"3

s s
and let r be a rational number satisfying "3 -<: r -<: 2
 86

Then y € B(x,r) and B(x,r) c B(y,s) c V . q.e.d.

This proposition will be used in an essential way in our study of

maxUnal tori in matrix groups.

E. Manifolds

Definition: By a space we mean some subset of some Rn with the

subspace topology. A map

of spaces is a homeomorphism if it is one-to-one, f is continuous,

and f- l is continuous.

Example: f(x) = e ix is a one-to-one continuous map of

1 2
[O,2rr) c R to the unit circle S eR. But f is not a homeo-

morphism because f- l is not continuous. (rt "tears" the circle

open. )

Example: Let G be·a matrix group and x € G. Then left

translation L by x (Lx (g) = xg) is a homeomorphism L

x
: G ~ G •
x
(Exercise. )

n
A manifold is aspace which "locally" looks like some R
 87

Definition: Aspace X is an n-manifold if each x E X lies in

some open set homeomorphic to some B(o,r) C Rn. An n-manifold is

said to have dimension n.

Proposition 8: ~ matrix ~ of dimension n is an n-manifold.

Proof: The exponential map from the n-dimensional tangent space

T to G is continuous. It is one-to-one on some neighborhorhood V

of 0 in T because it has an inverse (log). Also this inverse is

continuous. Take B(o,r) c V and we have that the identity matrix

I has the right kind of neighborhood. For any x E G we have

Lx 0 exp: B (0, r) -7 G

being a homeomorphism (composition of homeomorphisms is a homeomor-

phis im) onto a neighborhood of x. Thus G is an n-manifold.

Definition: A manifold is called closed if it is compact

closed and bounded).

Proposition 9: GL(n,k) is ~ closed, but ~(n,k) is closed.

Proof: Clearly GL(n,k) is not bounded, because for every non-

zero real number r rI E GL(n,k). (This also shows that GL(n,k)

is not closed. Because 0 E Mn(k) - GL(n,k) but every ball with cen-

ter 0 will contain some rI E GL(n,k) .)

If A E ~(n,k) then the rows are unit vectors so that as a vector

in Mn(k) the length of A is ~ n. Thus ~(n,k) is a bounded set.

To see that Mn(k) ~(n,k) is open, suppose B E Mn(k) - ~(n,k).

Then there exists x,y E kn such that

 88

(xB,yB) # (x,y) .

Since <, ) is continuous, there is same open ball B(B,s) in

Mn(k) such that for B' E B(B,s) we have (xB',yB') I (x,y) Thus

B' " CtI(n,k) q.e.d.

We finish this chapter with a result which will be of substantial

use to us later on.

Proposition 10: Let N and M be closed n-manifolds with

N c M. If M is connected, then N =M .

proof: We want to show that M - N is empty. If it isn't, choose

y E M - N and x E N Since M is connected, there exists a path

p [0,1] -->M

-1
with p (0) = x and p (1) = y Then o (M - N) is an open set in

[0,1] (see Exercise 3) and it contains 1 but not 0 Let to

be the largest element of the closed set I - p-l(M - N) Then

(i) every B(to's) contains points of p-l(M - N) , but

(ii) since N is a manifold there is some open neighborhood U

By continuity of p same B(to'~) maps by p

into N. This contradiction shows N = M .

Corollary: The map p Sp(l) --> SO(3) (see Chapter V) is sur-

jective.

proof: Since p is a hameomorphism on same neighborhood of each

point, we see that the image p(Sp(l) is a 3-manifold. Since p i

continuous, this image is a closed 3-manifold (Proposition 5). It

 89

remains to prove that SO(3) is connected (so that we may apply

Proposition 10). It suffices to show that any A E SO(3) may be

joined to the identity matrix I by a path in SO(3) .

is an orthonormal

basis for ",3 Let B be a rotation sending el to Ae l and

leaving the direction perpendicular to the (el,Ae l ) plane fixed.

(lf Ae l = e l proceed directly to the next step. lf el and Ae l

are antipodal on S2 , then we have two choices for B .) Clearly,

there is a path w from I to B in SO(3) Now Bel = Ae l so

Be 2 and Be 3 are an orthonormal basis for the plane perpendicular

to Ae l . Let C be a rotation of this plane sending Be 2 to Ae 2

and (If we can't do this, we would have detA -1.)

There is a path a from I to C in SO(3) Since A = BC, we

can multiply the paths IU and to get a path from I to A in

SO(3) .
q.e.d.

Note: In Chapter VIII we will prove that SO(n) is connected

for all n

F. Exercises

1. Show that the definition of continuity reduces to the usual

~ - 0 definition for f: (a,b) ... '"

2. Suppose we have A c Rn and have functions

 90

We have seen that fand g continuous imp1ies that gof is con-

tinuous. Give examp1es to show that:

f continuous and g 0 f continuous /. g continuous

g continuous and g 0 f continuous ,f. f continuous

n
3. Show that for Ac R and f: A ~ Rm then f is continuous

.. for each open set U in Rm , f- 1 (U) is an open set in A.

Rn
4. Show that i f A,B are connected sets in and A n B '" r/l ,
then A UB is connected.

5. Let H be any connected subgroup of a matrix group G. Show

that

-1
S U xHx
XEG

is connected.

6. Show that matrix mu1tip1ication is continuous (with one matrix

is continuous) .

7. Show that an arbitrary union of open sets is an open set.

n
8. Let A C Rn and XE" We say that x is a limit point

of A (x 1p A) if every

B(x,r) n A

is an infinite set. Show that C C Rn is a c10sed set ..

( x 1p C ~ X E C ) .
 91

9. Let D c Rn be open and c1osed. Show t hat if D is not

empty, then D = Rn (See the proof of Proposition 10.)

Chapter 7
Maxi mal Tori

A. Cartesian products of groups

If G and H are groups, we make G x H into a group by defining

(g,h) (g' ,h') (gg' ,hh') .

This works and if G,H are abelian so is G x H. (Exercises.)

Example: If G is a group of n x n matrices and H is a

group of m x m matrices, we can represent elements of G x H as

(n + m) x (n + m) matrices by

(g,h) I 0)
n[ ( g

~ 1m
n m

Then matrix multiplication gives the operation described above on

G xH (Exercise.) Let us look at an important special ca se of

this.

Let Then
 93

Both G and H are circle groups, and the (abelian) group G x H

= SI x SI is called a 2-torus.

Definition: A ~-~ is the Cartesian product of k circle

groups.

We have seen that a k-torus can be represented by a "block

diagonal" 2k x 2k real matrix. But it is easy to see that

T = ( ' "1 ,i~2 0 )

o .. e
iek

is a k-torus, so we can represent a k-torus as diagonal complex

k x k matrices.

Proposition 1: If G is ~ abelian matrix ~ and y,o are

~-parameter subgroups, then yo is ~ ~ - parameter subgroup.

Proof: (yo) (s+t) y(s+t)o(s+t)

y(s)v(t)o(s)o(t)

v(s)o(s) y(t)o(t)

(yo)(s)(yo)(t) .

Corollary: If G is an abelian matrix group then exp: (TG)e ~ G

is a homomorphism from the vector group of the tangent space

(TG)e to G at e.

Proof: Let ~ = y'(O) , ~ = 0'(0) with y,O being one-parameter

subgroups. Then from Chapter IV we know that exp(~) = y(l) and

 94

exp(~) = cr(l) . We have that ycr is a one-parameter subgroup and

(ycr) '(0) = y'(O) + cr'(O) ~ + ~ (Chapter 111), so exp(~+~)

= (ycr)(l) = y(l)cr(l) = exp ~ exp ~ .

Now we know that exp is 1-1 on some neighborhood V of 0

in (TG)e' So for G abe1ian we know that ker(exp) = L is a

discrete subgroup of the vector group (TG)e (i.e. some neighborhood

of 0 contains no point of L except 0). Next we consider when

exp: (TG)e ~ G is surjective.

proposition 2: Let G be ~ connected matrix ~ and let H be

any subgroup of G containing ~ open neighborhood U of e. Then

H = G •

Proof: Since U c Hand H is a subgroup, we must have

H .

Thus 2 3
W UUU uU U ... cH.

Each Uk is open, so W is an open set. (Exercise, Chapter VI.)

But W is also c10sed. For, let x be a limit point of W. Then

xU is an open set containing x (e E U) and hence must contain some

point of W. Thus

for some

But then x=u •.. uU -1 EW. In a connected space G only and

1 m
Gare both open and closed (Exercise, Chapter VI), W 1 ~ so W= G

so H =G .

Corollary: If G is a connected abelian matrix group, then

 95

exp: (TG)e ~ G is a surjective homomorphism with a discrete kerneI.

Proof: exp is a homomorphism so exp«TG)e) is a subgroup of

G and it eontains some neighborhood U of e. Thus exp(TG)e = G

In the exercises it is proved that Chis implies that exp(TG)e

Tk x Rn - k for some k. So we have

Theorem 1: Any compact eonnected abelian matrix ~ G is a

torus.

B. Maximal tori in groups

Definition: A subgroup H of a matrix group G is a torus if it

is isomorphie with a k-torus for some k. It is a maximal torus if it

is not contained in any larger torus subgroup of G.

Proposition 3:

e
I
fI(
~)J
e
eos sin
T = 1-Si~ e
cos e
o

is a maximal torus in 50(3) .

Proof: Clearly T is isomorphic with « e~s

-s~n
e
e
sin
eos
which

is a eircle group and thus is al-torus.

Suppose there is a larger torus subgroup T' of 50(3), i.e.,

T:,t:T'cSO(3) .

5ince T' would be abelian, we would have: mE 50(3) such that

 96

~ J T but ~ commutes with every element of T. 50 it suffices to

prove that:

If ~ E 50(3) commutes with each t E T, then ~ E T. Refer

back to our proof that Center 50(3) {I) in Chapter V, and you will

see that we have already proved this fact.

Proposition 4:

J}
81 sin 81 0 0
{( co, \)1 cos 81 0 0

T" -": 0 cos 82 sin

0 -sin 82 cos

is a maximal torus in 50(4) .

Proof: This clearly 1s a 2-torus and is a subgroup of 50(4)

As before, it suffices to prove that if rb E 50(4) commutes with all

elements of T then rb E T

Let V be the 2-plane in R4 spanned by e l ,e 2 and W be the

4 We see that
2-plane in p spanned by e 3 ,e 4 T consists of all

pairs

(rotation of V , rotation of W) •

Claim:

Take ~ E T such that ~ is the identity on V but is not the iden-

tity on W. Then

~(el) = ae l + be 2 + ce 3 + de 4

~rb(el) = ae l + be 2 + c / e 3 + d / e 4
 97

This shows c = 0 = d so rp( e l ) E V The same kind of proof shows

m(e 2 ) E V Dually, ct>(e 3 ) and m(e 4 ) are in W

So we know that rp is orthogonal on V and is orthogonal on W.

A priori, it could be a reflection in each and we would still have

rp E SO(4) But ct> commutes with all rotations on V and is thus a

rotation on V. (See the proof of Proposition 3, Chapter V.) Simi-

larly, m is a rotation on W, so ct> E T .

q.e.d.

From Propositions 1 and 2 the general result about maximal tori

in SO(n) should be clear. We have n/2 of the 2 x 2 rotation

matrices for n even and have a 1 in the n,n position for n

odd. The proof of the general case is an obvious extension of the

above proofs.

Proposition 5:

is a maximal torus in U(n) .

Proof: Let ct> E U(n) commute with each n E T. Consider any

n of the form

Cl. = E T •

such Cl.'S can move any vector which is not a multiple of el . Hence
 98

Similar arguments give

1, ... ,n .

Thus

and, since rh E U (n) , each is of unit length. Thus rh E T

and T is maximal.

proposition 6:

T o

is a maximal torus in SU(n)

Proof:

i el
A matrix

ie
n
0- of the form

i(9 l +·· .+9 n )

('6' ?an) has

det 0- e e e so that 0. E SU(n) ~ det ~

o So the T described here is just the

intersection of SU(n) with the maximal torus given for U(n) .

First we must check that this is an (n-l)-torus. To do this,

use
 99

It is an exercise to show that this works.

Now for n > 2 the same proof as used for U(n) will work, but

for n = 2

T - (,:, e~i')
and we do not have matrices er _
,-
(1 0)0 e- 19
to use. But for n =2

we give a direct simple proof. If

(a b (i 0)
rb d) E SU(2) and er. E T
c 0 -i

then

(ai -bi ( ai bi)

~er. -di) 0, rb
ci -ci -di

Thus b 0 = c and ~ E T .

Proposition 7: The maximal torus given for U(n) is also a maxi-

mal torus for Sp(n) .

Proof: Just as for U(n) we can show that any element of Sp(n)
 100

iS10
which conunutes with all
(
o
e . . i
e Sn
) must be diagonal with the

diagonal elements having length 1. But now these elements are

quaternions. However (Exercise), any quaternion which conunutes with

i must be a complex number.

q.e.d.

C. Centers again

Now that we know maximal tori in our matrix groups, we are able

to calculate the centers.

Proposition 8: Center (Sp(n)) (1,-1}

Proof: We have seen that if any element conunutes with all ele-

ments of the maximal torus we have described, then it must lie in

that maximal torus. Hence, in every ca se

Center c T .

If A E Center Sp(n) , then Since A must

conunute with the matrix j1, it follows that the diagonal elements

must be real (and since they are of unit length) they are ±l 1t

is an exercise to show that they all have the same sign. Now I and

-I are in the center.

q.e.d.
 101

proEosition 9: Center U(n) {eieI} ~ SI

Center SU(n) (wI I wn I}

proof: If B E Center U(n) we get that B is diagonal with

diagonal elements complex numbers of unit length. Let

B (o", "'..~) Iln

with each 1 . Let

(: 0... 0)
1
0 0 ... 0
A = 0 1. 0
o '.1

Then AB = BA shows II I = 112 ' etc. , so all Il are equal. Clearly

any is in the center, so the center of U(n) is as asserted.

For SU(n) we note that the same argument will show that an

element must be of the form eieI to be in the center. But

ie ine
det(e I) e

and since this reust be 1, eie must be an n th root of unity.

q.e.d.

So Center SU(n) SU(n) n Center U(n) .

Finally, we want to calculate the center of SO(n). It turns

out that it depends on whether n is even or odd·. The groups SO(2n)

and the groups SO (2n+l) are different in some important ways.

Now SO(2) = SI is abelian so its center is the group itself.

We have al ready proved that the center of SO(3) is just (I}.

 102

Claim: For k ~ 3 any element in the center of SO(k) must be

a diagonal matrix.

As before, if A E SO(k) is in the center, it must be in our

standard maximal torus. So suppose A E Center SO(k) is of the form

Co,
)
91 sin 91 0
A -Si~ 9 1 cos 9 1 0
0 *

(J 0 ~ 0)
Let
0
1
0 01 SO(k)
P E
. 0
o.
1

Then PA has zero in the 1,2 position, whereas AP has sin 91

in the 1,2 position. Thus sin 91 = O. Simi1ar arguments show all

off diagonal terms are zero.

It fo110ws also (since each sin 0i 0) that each diagonal term

(cos 8i ) is 1 or -1 . So each 2 x 2 block is (1 0) or

0 1
-1 0)
(0 Arguments like we used for U(n) show that all diagonal
-1
terms must be ecp al. So we finally conc1ude that

Center SO(2n+1) = tI}

Center SO(2n) = tI,-I}

0 ') 0)
°
-1_1 -1
F or examp 1e , ( is in the center of SO(4) , but ( -1
O -1
-1 -1
is not even in SO(3) •
 103

We now tabulate the information we hBve generated about our

groups.

Dimensions, Centers, Maximal tori

Dimension Center Standard Maximal torus

2
U(n) n

SU(n)

SO(2n+l) (2n+l) (2n) = 2n2+n (I}

2

2n(2n-l)
SO(2n)
2 (I,-I}

Sp(n) (I,-I}

Note that we have nothing to distinguish SO(2n+l) and ~

Center
A good part of the remainder of this book is devoted to deeiding for

whieh n these are isomorphie.

 104

D. Exercises

1. Show that the operation defined on G x H does make it into

a group. prove that if G and H are abelian, so is G x H .

2. Do the exercise in the first example of §A.

3. Show directly that the product of the matrices

( c~s 9 sin 9) ( cos ~ sin rt»

and
-s~n ~ cos 9 -sin rt> cos rt>

is the matrix for a rotation through angle 9 + rt>

4. Let T be a maximal torus in a matrix group G and let

-1
prove that xTx is also a maximal torus in G.

5. Prove that if q is a quaternion such that qi iq then

q is a complex number.

6. Show that ~ ~ SU(n)

center - center

7. A lattice subgroup K of Rn consists of all integar linear

combinations of some set of linearly independent vectors. More

explicitely, let

be linearly independent vectors in Rn

we form

n
K is a subgroup of II
 105

A subgroup H of Rn is diserete if some neighborhood of 0 in

Rn eontains no point of Hother than O. Prove that: A diserete

subgroup of ~n is a lattiee subgroup. (Choose a nonzero veetor

vl E H such that no element of H lies in the interval [O,vll in

Show that ~l eontains all integral multiples of but no

other elements of H • Choose V z in H with not an integral

n
multiple of Show that the span of and V z in R eontains

all integral linear eombinations of but no other ele-

ments of H. ete.)

8. Show that if L is a lattiee group in Rn generated by

vl,···,v k Then Rn/L is isomorphie with the produet of k-torus

n-k
and R
Chapter 8
Covering by Maximal Tori

A. General remarks

In Exercise 4 of Chapter VII one showed that if T is a maximal

torus in a matrix group G, then for any x E G, XTx- l is also a

maximal torus. What we prove in this chapter is that if T is our

standard maximal torus in one of our connected matrix groups G, then

-1
G xTx

showing that every element of G lies in at least one maximal torus.

To say that G U XTx- l is to say that given y E G there

exists x E G such that

-1
Y E xTx

This is equivalent to: Given y E G , there exists Z E G such that

-1 -1
zyz E T (take x = Z ) • So we want to show that Y E G can be

put in diagonal or 2 x 2 block-diagonal form by conjugation.

We begin by reviewing a little linear algebra. Let V be a

vector space over a field k and let

be a linear map.
 107

Definition: A subspace W of V is ~ - stable if

In this case, we can restrict the domain of 0 to W

to get a linear map

3
Examples: Let V Rand W be the 2-plane spanned by el

and If

~)
sin
( cos 8
o= -Si; 8 cos
0

then W is o- stable.
2
Let V = R and 0 = (0
1
1)
0
Then W [(x,x)} is o- stable

and so is W' = [(x,-x)}

An important special ca se occurs when dirn W 1.

Definition: A nonzero vector v E V is an eigenvector for

o if there exists a AE k such that

A is the corresponding eigenvalue.

Now if 0 has one eigenvector v then it has a "line" of them;

namely, if r E k with r # 0, then

0(rv) A(rv)

(We will have to be a little more careful when using our "skew" field

E .) Thus rv is also an eigenvector with the same eigenvalue A.

This suggests that eigenvalues may be more fundamental.

 108

Definition: For any A c k , set

VO.) (v c V I tb(v)

Thus V(A) includes the zero vector 0 and all eigenvectors having

eigenvalue A' It is easy to see that V(A) is a subspace of V ,

the eigenspace belonging to A

For example, taking A o , we see that

V(O) = (v c V I mev) = O}

is just the kernel (or null space) of the linear map tb.

Now take dim V to be finite. We mayas well take V = kn .

Then linear maps tb: V -> V correspond 1-1 with elements of Mn(k) ,

so we can think of tb c Mn(k) .

Proposition 1: V(A) ~ {o} ~ det(tb - AI) O.

Proof: =>

If 0 ~ v satisfies rb(v) = AV, then (rb - AI)v = 0

showing that tb - AI is singular. Thus det(m - AI) = 0

If det(tb - AI) = 0, then rb - AI must send some nonzero

vector v to 0 ; i.e., mev) = AV, so V(A) ~ (01 .

B. (t) for U(n) and SU(n)

We are concerned with C - linear maps cn -> cn which we

 109

represent by elements of Mn(C) We begin with two easy observations:

(a) If A is unitary and Wc cn is A - stable, then Alw

is unitary.

If x,y E W then xA and yA E Wand

(xA,yA) (X,y)

(b) If A is unitary and W c Cn is A - stable, then the

orthogonal complement

W.1. = (x E Cn I (x,y) o for all y E W}

is also A - stable.

Let x E W.1. For any y E W we have

t- -1
(xA,y) (x,y A) = (x,yA )

-1
Now A is an isomorphism of W onto W so yA E W Thus

(xA,y) o for all y E W, showing xA E W.1.

Proposition 2· For A E U(n) there exists an orthonormal basis

n
of eigenvectors vl , ... , v
n
E C for A.

Proof: By (a) and (b) above, it suffices to show that a unitary

matrix C always has an eigenvector v For we can make v unit

length and restrict C to etc.

But let A E Mn(C) Then

p(~) det(A - ~I)

is a polynomial in ~ and r. is algebraically closed so it has a

 llO

root. Thus A has an eigenvector.

Proposition 3: The conjugates of T

cover U(n) .

Proof: Given A E U(n) we will find B EU(n) such that

Let , ... , v be an orthonormal basis consisting of eigenvectors

n
of Let B send

e. to v.
J J

Then BAB- 1 sends e. to v. to A.V. to A.e. So

J J J J J J

BAB- 1
{~'..~)
Since BAB- 1 is unitary, each L e is a unit vector; 1. e. ,
J j
I Aj I = 1 Thus BAB -1 E T
q.e.d.

Corol1ary: The conjugates of

T = {( ~1 ~8nl Z8 j o}
cover SU(n) .
 111

Proof: Given A E SU(n) we want B E SU(n) such that

-1
BAB E T. If we take B' E U(n) such that

B/A(B,)-l is in the maximal torus for U(n)

it is actually in the maximal torus for SU(n) since

det(B/A(B/)-l) = det A = 1

1
If we choose U E C such that det B' and set
n
u

BuB',

then B/A(B/)-l and B E SU(n) .

c. (t) for SO(n)

For any matrix group G and maximal torus T we have that

-1
U xTx
XEG

is a connected set, so if G is not connected, (t) could not hold.

Since ~(n) is not connected, we know that

-1
(!j(n)" U xTx
XE (!) (n)

But we will prove (t) for SO(n) and this will, incidentally,

prove that

SO(n) is connected
 112

Clearly we ean have no result like proposition 2 for 50(n) .

For example A = ( eos e sin 8) € 50(2) with A '" I has no nonzero

-sin e eos e
eigenveetor. But this is about all that ean happen--we ean find

stable 2-planes. To see this we eonsider symmetrie linear maps.

n n
Definition: A linear map 5 ~ ~ ~ is symmetrie if

n
(x5,Y) (x,y5) for a11 x,y € R

Note that for A € Mn(R), 5 A + t A is symmetrie. For we

have

t t
(x5,y) = (xA,y) + (x A,y) = (x,y A) + (x,yA) (x,5y)

Corresponding to (a) and (b) for U(n) we have: If 5 is

symmetrie and W€ ~n is 5-stable, then 51w is symmetrie and W~

is 5-stable. Thus, just as for U(n) , it suffiees to show that a

symmetrie matrix always has an eigenveetor.

Proposition 4: A (real) symmetrie matrix 5 has ~ eigenveetor.

Proof: Define t: ~n _ (0) ~ R by

t(x) = ~
(x,X)

Note that for nonzero r € R we have t(rx) t(x). Thus if Sn-l

is the unit sphere, we have

t (~n (0» t (Sn-l)

Now is eontinuous and sn-l is eompaet. Thus ~ (Sn-l) is eom-

t
paet and we ean find v €
Sn-l such that
 113

,(v) is an absolute maximum

We claim that v is then an eigenvector for S.

For any nonzero y € Rn if we set

f(t) = ~(v + ty) ,

we must have f'(O) = 0 This will allow us to show v is an

eigenvec'tor and to find its eigenvalue.

We have

(v+ty,vSHyS) 0. (t)
f (t)
(v+ty,v+ty) a (t)

f '(0)
~/(O)a(O) - a'(O)~(O)

(a (0)) 2

Since v is of unit length, a(O) 1 so that

f I (0) I'! , (0) - a ' (0)0.(0) •

We easily calculate that a'(O) = 2(v,y) and 0.(0) = (v,vS) and

0.'(0) = (y,vS) + (v,yS) = 2(vS,y) (S and , ) are symmetric).

This gives f/(O) = 2(vS,y) - 2(v,vS)(v,y) = 0 , or

«vS - (V,vS)v),y) 0

Now (,) is nondegenerate and y is arbitrary, so

vS (v,vS)v q.e.d.

We first use ~roposition 4 to study ~(n) and then prove (t)

for SO(n) .
 114

Proposition 5: Let A E ~(n)

subspaee W of Rn with

dim W E (1,2)

proof: S A + tA is symmetrie. Let w be an eigenveetor for

S. Consider wand wA

ease (i) If wand wA are 1inear1y dependent, wA ~w,

then W = (rw IrE R} is A-stab1e and dim W 1.

ease (ii) If 1inear1y independent, let W Span(w,wA) We have

t t-1
wS = ~w , 1. e. , wA + w A = ~w and A A so that

-1
wA + wA or ~wA .

This imp1ies that W is A-stab1e. For, let

/lW + BwA E W .

Then

2
(/lW + BwA)A /lwA + BwA

n.wA + B üwA - w)

(-B)w + (~+B~)wA E W

Proposition 6: The eonjugates of our standard maximal torus in

SO(n) eover SO(n)

proof: We ehoose an orthonorma1 basis

A E SO(n) as fo110ws.

If A has a stab1e one-dimensiona1 subspaee, ehoose one and take

 l1S

to be a unit vector in it. Next do the same for and con-

tinue as long as possible, generating an orthonormal set vl"",v k

of eigenvectors for A. Let A is ortho-

gonal on W Choose a stable 2-space Wl and let be an

orthonormal basis for Wl ' etc. This gives an orthonormal basis

v l ,v 2 '··· ,vk',vk+l.:.,vk+2:

Wn _k
-2-

We use this to find B E SO(n) such that

BAB- l E T our standard maximal torus.

Let C map e. to v. Then C E (!)(n) and

~ ~

CAC- l E T .

(Note that det CAC- l det A = 1.) If it happens that det C = 1

let B = C If det C -1 we seek D E (!)(n) - SO(n) such that

DTD- l = T Then B = DC will satisfy BAB- l E T and det B = 1

Such a D is easy to find; e. g.,

D =(~:,0)
o '1

This proves CI') for SO(n) .

 116

D. ( t) for Sp (n)

Our proof that conjugates of T __ (e

o
0)i Sl
.
.. iS
e n
cover Sp(n)

will relyy on the following property of JH •

Proposition 7: JH has square roots.

2
Proof: Given q E JH we must find h E JH such that h = q

We mayas well take Iql 1 (if q = 0 take h = 0 and may then

assume Ihl = 1. If q x + iy + jz + kd set h ~ + i8 + jy + k~.

2
h =q gives

2 2
8 - y

y 2~y = z 2".~ w

Since Iql = 1 = Ihl , these give 2n 2 1 + x. Choose ~ satis-

fying this and then solve for 8, v, 5

Next we find stable subspaces for A E Sp(n) much as we did

for orthogonal matrices.

n
proposition 8: For A E Sp(n) there exists a nonzero v E JH

such that

W Span(v,vA) is A stable.

Proof: DeHne t :JHn - (0) .... JH by

t (x) ~
(x,x)
 117

For any nonzero real number r we have ~(rx) = ~(x) so that ~ is

constant on real rays through 0 in En - (0) So we can find

V E

making a; an absolute maximum. For arbitrary y E E n we set

f(t) ~(v+ty),tER,

and we have f' (0) O. This gives

o (yA,v) + (vA,y) - «y,V) + (v,y))(vA,v)

or

-1
o «vA - (vA,v»)v,y) + (y,(vA - (vA,v)v) .

Choose y to be an arbitrary but real n-tuple and we see that

-1
(vA + vA 2(vA,v)v,y) o

for all real n-tuples y. This implies that'

-1
vA + vA = 2(vA,v)v

or

(*) 2(vA,v)vA .

Just as in proposition 5, such a formula proves W Span(v,vA) is

A - stable.

proposition 9: Any A E Sp(n) has an eigenvector.

Proof: Choose v such that Span(v,vA) is A - stable. We

 118

claim that A has an eigenvector of the form

'" a.v+S(vA) , a.,a E 1I .

We have
2
a.vA + S(vA )

a.vA + s(2(vA,v)vA - v) from (*)

(0. + 2s (vA, v»vA - Sv

We seek ), E]H such that ",A )'''', or

(0. + 2S (vA,v»vA - Sv ).,a.v + ).,s(vA)

(0. + 2S(vA,v»(vA) ),S(vA)

This gives (putting -),0. in for B )

2
a.()., - 2(vA,v) + 1) O.

Since ]H has square roots, ",e can find )., such that
2
)., - 2(vA,v) + 1 = 0 q.e.d.

Proposition 10: ho1ds for Sp(n)

Proof: Just as for U(n) ",e no", find an orthonorma1 basis of

eigenvectors and thus B E Sp(n) such that

BAB- 1
 119

with the Al""'A n quarternions of unit length. If we conjugate

this by a diagonal matrix

E Sp(n) , 1 ,

we get

-1
QBA(QB)

This will be in the standard maximal torus if each

qj~jqj is a complex nunber. It is proved in Exercise 5

that we can choose the q. '5 to do this.

J

E. Re1fections in Rn (again)

Let x E ~n be nonzero and let x~ be the (n-l)-hyperp1ane

through 0 perpendicu1ar to x.

Proposition 11: The reflection ~: Rn ~ Rn in x~ ~ given ~

the formula

~(y) y - 2~x
(x,x')

Proof: ,,
,,
,
,
t-~-- p(y)
.k------'~ Y

x
 120

Let p(y) be the projection of y in x~ Then y - p(y) is some

multiple of x , say rx, and (p(y),X) = 0 This gives

r = ß.n So
(X,x)

p(y) ~ x
y - (x ,X)
and ~(y) - 2 ~ x
(X,X)

We can check directly that ~ is an orthogonal map.

(dl(y), ~(z) ( -2 ~ x z-2 ~ x)

y (x,x) , (X,x)

(y z)-2~(y x)-2~(x z)+4(x,y)(X,Z)(X X)

, (x,X) , (x,x) , (X,X)(X,X)'

and this is just (y,z)

Also it is easy to see that det ~ = -1 because we can choose

a basis with x as first vector and the other vectors forming a

basis for x~ Relative to this basis ~ is given by

proposition 12: If A E ~(n) is reflection in a hyperplane W

in Rn and B E ~(n) then

is reflection in the hyperplane WB-I

proof: Let w' Then

w' ,

so that BAB- I is the identity on WB-I

 121

Let ~l,···,CLn be an orthonorma1 basis with ~1 perpendicu1ar

to W and CL2' •.. '~n a basis fpr W. Then BAB- 1 is the identity
-1 -1
on CL 2B , •.. ,CLnB and

proving the proposition.

proposition 13: ~(n) is generated ~ ref1ections.

Proof: We want to show that any element of ~(n) may be

written as the product of a finite number of ref1ections.

First we prove this for elements of our standard torus T

Let m be the biggest integer ~ 2n so that T has m blocks

sin 8 1 sin 8
cos 8 ) , ... , B
sin 8
m) arranged a10ng
1 m
m

the diagonal. Let '~1 E T be the identity except in block Bi

Then any element of T is sure1y a finite product of elements

~1'··· '~m
Let ~i be ref1ection in the hyperplane formed by all coordinates

except those for B.~ and the first co ordinate for

2

Let be ref1ection in
1

the hyperplane formed by all coordinates except those for B. and

~
8i
the coordinate at angle :r in the plane. Then

Thus every element of T is the product of finite1y many ref1ections.

Given A E SQ(n) , choose B such that

 122

BAB- 1 E T

Writing BAB- 1 as a finite product of ref1ections and using

Proposition 12 gives A as such a product. Fina11y, let C E ~(n)

Then C = AD ~here A E SQ(n) and D is a ref1ection (Exercise).

q.e.d.

F. Exercises

'::'*i)
cos !!.
4
1. Show that A (
,50(2) h., no A - ,"b1.
-S1n
•
'4'1i
2
subspace in ~ other than (o} Find all A - stab1e subspaces in
2 1)
R for A = (0
1
°
2. Find all eigenvectors and eigenva1ues of

(0 1) , (1 0) , (1 ~) , (0 1)
1
° °° ° ° 1

3. In llin sho~ that i f (x,y) has zero real part for every

y , then x = °
4. Let D Show that ~(n) - SQ(n) SQ(n)D .

Sho~ that D may be rep1aced by any other element of G(n) - SQ(n) .

5. Prove that if ).. E lli has 1ength 1 , there exists q E lli of

1ength 1 such that q)..q is a comp1ex number.

 123

2
6. Let Ll ,L 2 be two lines through 0 in R with angle e
between them. Let $1 be reflection in Ll and $2 be reflection

in L2 . Let W be the subgroup of ~(2) generated by \11 1 and

\112 For what values of e will W be a finite group?

7. Recall that in order to prove that

1 + dim SU(n) = d1m U(n)

in Chapter IV we used the fact that

Prove th1s now for A E U(n). (We can wr1te

A = BCB- l

where B 1s un1tary and C is an element of our standard maximal

torus in U(n) . Then

BCB- l C -1
Be B

0)
A
e e

o .0) (~
and with I Aj I 1 so eC
C = (" •
·"n An
e
Al An A +· .. +>.
Thus det e A = det (BeCB- l ) det e
C
e e e l n

= e TrC = e TrA .)
Chapter 9
Conjugacy of Maxi mal Tori

A. Monogenie groups

Definition: A subset Y of a spaee X is said to be dense in

X if every nonempty open set in X eontains at least one point in

Examples: Both ~ and ~ - ~ are dense in R.

is dense in ,,2

Lemma: If Y is dense in X and C is a elosed set in X

wi th Y c ethen C = X .

proof: X - C is open, so if it is nonempty it must eontain some

Y E Y. But Y E X - C eontradiets Y ce.

Definition: A matrix group G is monogenie (i.e., one gen-

erator) if there exists x E G sueh that

2 3
f x ,x,x,···1
 125

is dense in G. Then any such x is called a generator.

Note that the additive group F is not monogenie, nor are any of

the vector groups Fn Of course, the identity (matrix) cannot

ever be a generator. Consider the circle group SI thought of as

the additive group of reals modulo 1 . If x = E

q
E sI is rational,

then the powers

x,2x,3x,4x, ...

will not form a dense set in SI because they will all lie in the

set
1 2 q-l q
[q,q' ... '-q-,q O}

But if we choose x to be irrational we should get a generator.

However, we will use a more topological way of finding a generator,

and it will generalize immediately to higher dimensional tori.

proposition 1: SI is monogenie

Proof: Let ul ,u 2 ,U 3 , ... be a countable basis for the open

sets in SI Then to prove the proposition it suffices to find

X E SI such that: Given Uk , there exists an nE N = [1,2,3, ... }

such that

We proceed as folIows.

Choose a nondegenerate closed interval 11 = [al,b l ] c Ul .

Choose an n l E N such that length [nlal,nlbl] ~ 2 Then

nll l is all of sI, and thus we can

 126

choose 12 (nondeg. ) such that n l 1 2 c U2 '

choose n2 such that n 21 2 is all of 51.

Then we can

choose 13 (non deg. ) such that n 31 3 c U3 '

choose n3 such that n 31 3 is a11 of 51 , etc.

Because Ul ,U 2 ,U 3 , ... must contain arbitrarily small sets, we have

that 11 n 1 2 n 1 3 n ... is a single element x.

But x is a generator for 51. For, given Uk , we have

q.e.d.

Now the r-dimensional torus T is just all r-tuples of real

numbers, each taken modulo 1. 50 by using cubes instead of intervals,

we get

proposition 2: The r-torus T ~ monogenic.

B. Conjugacy of maximal tori

We are now ready to prove that for a connected matrix group,

any two maximal tori are conjugate. Let G E t 5O(n),U(n),5U(n),5p(n)}

and let T be our standard maximal torus in G We know that

-1
G = U xTx
XEG

Proposition 3: Let T' be any maximal torus!!! G . Then for

some x E G, T' = XTx- l

Proof: 5ince T' is a torus it is monogenic and we choose a

generator y for T' . Then for some x

 127

-1
Y E xTx

-1
Now xTx is a group, so

y 2 3 -1
(y,y ,y , ... ) c xTx ,

and, of course, Y c T' Thus Y is contained in the c10sa d set

-1
T' n xTx in T' and Y is dense in T' So by our lemma we

have

T' n xTx -1 -1
T' or xTx ::> T'

-1
But xTx is a maximal torus in G so xTx -1 T'

Definition: The rank of a matrix group G is the dimension

of a maximal torus in G.

This is c1ear1y an invariant for isomorphisms.

c. The isomorphism question aga in

For a matrix group G we now have two numerica1 invariants, its

dimension n and its rank r , and one subgroup invariant, its

center C Furthermore, we have ca1cu1ated these and we now

tabu1ate our resu1ts according to rank.

 128

Rank 1: Group Dimension Center

U(l) 1

SU(2) 3

SO(2) 1

SO(3) 3 {T}
Sp(l) 3 {T,-I}=71/2

Rank 2:

U(2) 4

SU(3) 8 'll/3

SO(4) 6 (I,-I}=z/2
SO(5) 10 (I}
Sp(2) 10 {I,-Il
Rank 3:

U(3) 9 Sl

SU(4) 15 71./4

SO(6) 15 (I,-Il=Z/2
SO(7) 21 {Tl
Sp(3) 21 (I,-Il=71./2

For rank 4 and greater we have

dirn U < dirn SU < dirn SO(even) < dirn SO(odd) = dirn Sp .

Sinee SO(odd) and Sp have different centers, they are not

isomorphie. So we just need to look at ranks 1, 2, 3.

We know that U(l) ~ SO(2) (eire1e group). SO(3) ~ Sp(l)

and we proved in Chapter 11 that SP(l) ~ SU(2) So we know all

about rank 1.
 129

For rank 2 on1y 50(5) and Sp(2) have the same dimension

and they have nonisomorphie centers.

For rank 3 SU(4) and SO(6) have the same dimension and

different centers, and the same app1ies to SO(5) and Sp(2) .

So we have solved our isomorhism problem for these groups. But

we can generate some other groups and the isomorphism question

for these needs to be reso1ved. In the next section we consider ways

of getting new groups.

If H is any proper subgroup of the center (Sl) of U(n) ,

then we have a group U(n)/H, and H is finite, so

dim U(n) '" n 2 dim U (n)

H

so these groups are not isomorphie to any of our others.

If H is a finite proper subgroup of the center ~/n of

SU(n) (n not a prime), then

dim Su~n) = n2 _ 1 dim SU(n)

and again these are not isomorphie with any of our other groups except

we might have

S~}i) ;" SO(6) .

Indeed, we will see 1ater that this is the case.

D. Simple groups, simp1y-connected groups

Definition: A matrix group G is simple if it has no nontrivial

 130

normal subgroup. This is the same as saying it has no quotient

groups other than itself and the trivial group.

If G is not simple, then by choosing a nontrivial normal sub-

group H of G we get a new group G/H So we want to see which

of our groups are simple, and, when they are not, what normal sub-

groups they have. We state the result, but a proof would take us

too far afield.

Theorem: ~ ~ SU(n)
center - center

SO(2n+l)

SO(2n)
center

~
center

So we only get the groups we already know about.

Another way of generating new groups is somewhat more sophis-

ticated. A path w in G is a smooth curve w: [0,1] ~ G with

w(O) = e Let P(G) be the set of all paths in G. Then P(G)

becomes a group if we define

(wa)(t) w(t)a(t) .

proposition 4: If G is a connected group, the map

p : P(G) ~ G, defined by

p(<U) w(l)

is a surjective homomorphism.
 131

Proof: G connected implies that for any x € G there is a

path from e to x in G, so p is surjective. No~

o(wo) (wO) (1) w(l)o(l)

so p is a homomorphism.

We denote the kernel of p by n(G) and call it the loops

in G.

Definition: G is simply-connected means that n(G) is

connected.

If G is not simply-connected, let nO(G) be the identity

component of n(G) Then nO(G) is also anormal subgroup of

P(G) and we set

and call this group the universal covering group of G

So ~e get ne~ groups this way whenever G is not simply-

connected. Clearly p: p(G) ~ Ginduces a homomorphism

The kernel of p is denoted by nl(G) and is called the fundamental

~ of G. Clearly G is simply-connected ~ nl(G) =1 (i.e., the

trivial group). We state but do not prove the following.

 132

Theorem: .rl(Sp(n)) 1 for n 1,2, ...

.r l (U (n)) 1 for n = 2,3, ...

.rl(SU(n)) 1 for n = 2,3, ...

Z
TTl(SO(n)) ="2 for n 3,4, ...

So the only new groups we can generate as universal covering

groups are

r--!
SO(n)

r--J
We see that p SO(n) -'> SO(n) is a 2-1 homomorphism. We construct

and study these new groups in the next chapter.

E. Exercises

1. Let Sl be the additive group of reals mod 1. Show

that i f X E [0,1] is irrational, then x is a generator for

(Hint: For y = ~ rational, [ny n=1,2, ... } gets within

q
of each point in Sl. Show that i f x is irrational we can write
Pi
x = ~im Yi where each y. = is rational and = ... )
~-'>tII
~ qi

2. Let G be a matrix group which is not necessarily connected.

Let GO Jx G I :[ pa th w : [0, 1] G
l
E -'>

w(O) = 0 w(l) = x

Call this the identity component of G. Show that GO is a sub-

group of G.

3. Let G be a matrix group and H anormal subgroup of G.

Show that the identity component HO of H is also anormal subgroup of

G .
Chapter 10
Spin(k)

A. Clifford Algebras

One way of constructing groups which are subsets of seme Rn

is: Let G be a finite-dimensional real algebra and let G be the

group of units in G. We get more groups as subgroups of G For

exarople, we have used the algebra Mn(R) in which the group of units

is GL(n,R) and we have the important subgroup SO(n) Our groups

Spin(k) are subgroups of the group of units in the Clifford

algebra Ck .

For k = 0,1,2, ... we will define a real algebra Ck of

dimension 2k First we set

Next, Cl is to be two dimensional: we take a basis l,e; let

1 act as a multiplicative identity and define e2 = -1. This

specifies the multiplication

(a + be)(c + de) (ac - bd) + (ad + bc)e

and we see that

 134

as 2-dimensional real algebras.

Next C2 is to be 4-dimensional. We take a basis

2
and set el = -1 , = -1 and This specifies

the multiplication and we see that the assignroent

1 f') 1

e l f') i

e 2 f') j

e l e 2 f')k

gives an isoroorphism

Definition: Ck is the algebra which is generated, as an

algebra, by e l ,e 2 , ..• ,e k subject to

2
e. = -1 and e.e. -e e if i.f j
~ J ~ i j

To explain what is meant by "generated as an algebra," consider

To we must add 1 and products of the e 's to get

i
a vector-space basis for C3 , and we get

which gives dimension 2 3 = 8. The same argument shows that

(e. . .. e.
~l ~r

is a vector-space basis for Ck (r Odenotes the element 1 of the

 135

basis).

Note that sending ei to imbeds

as a subalgebra of Ck '

Now CO,C l are fields and C2 is a skew field but we cannot

expect this for k > 2 .

Proposition 1: For all has divisors of zero.

Proof: It suffices to find a divisor of zero in C3 ; i.e.,

an x # 0 such that there exists a y # 0 with xy = O. Well,

There are two notable ways of getting algebras which have

divisors of zero.

Definition: If G and a are real algebras, their direct sum

G e ß is the set of all pairs (x,y) with x € a and y € ß

with operations

(x,y) + (z,w) (x+z,y+w)

r(x,y) (rx,ry)

(x,y) (z,w) (xz, yw)

proposition 2: a e ß has divisors of ~, and for any field

has divisors of zero if n ~ 2 .

 136

Proof: (1,0)(0,1) = (0,0)

For our purposes we need to know the Clifford algebras up

through Cs . There is a uniform way of getting up through C4

(due to Alan Wiederhold). We consider

as follows: ~ t C is represented by

and (ql,q2) E 1HEll1H by

ql
(0
° ). Then the assignment
q2

(i ?) --> el
° ~

(j ?) --> e2
° J

(k 0) --> e3
°-k
(0 k) --> e4
k 0

gives isomorphisms

c ! Cl

E! C2

1HEllE 1 C3
M2 (lH) 1 C4

Finally, the assignment

 137

(:-~ ~ ~) (~-~ L:) (; : ~ -~)

o 0 0 -~ 0 0 1 0 0 0 -i 0

1
:
( ~ -~ -~) (~-~ -; -:)
0 0, 0 -~ 0 0 0

induces an isoroorphisro M4 (C) 1 Cs

k
Let R denote the k-space in Ck spanned by e 1 , ... ,e k and

let Sk-1 be the unit sphere in

Proposition 3: If C*k denotes the group of units in

then
k-1 *
SeC k .!: .~. , each x E Sk-1 is a unit.

Proof: Let 1 .

Then

Definition: Pin(k) i5 the subgroup of C*k generated by Sk-1

 138

Thus each element of Pin(k) is a finite product of elements

k-l
of S .

We define a conjugation in Ck . It suffices to define it on

basis elements and we set

For example,

1* -e.
~
-eie j

Clearly **
(x) x. Also, we have (xy) * y *x * because

«e·l···e.
~ ~r
)(e·l···e.
J Js
)) * (-e. ) ... (-e·l)(-e. ) ... (-e. l )
Js J ~r ~

J Js ~
*
(e. l ·· .e. ) (e. l ... e. )
~r
*

We also define an automorphism ~ of Ck by ~(ei) = -ei .

(Note that conjugation is not an automorphism, because it agrees

on each ei ' but ~(e.e.) = (-e.)(-e.) = e.e. whereas

~J ~ J ~J

= -e. e.) .)
~ J

Definition: For U E Pin(k) and we set

o (u) (x) 1l.XU "

It is not clear that p(u)(x) E Rk but it is a consequence of

the next two propositions.

k-l
Proposition 4: If u E S C Pin(k) , then 0 (u) is
k
reflection in R in the hyperplane perpendicular to u.
 139

Proof: Pick an orthonorma1 basis of "k with

Consider

If i ~ 1 this equa1s ui and if i 1 it equa1s -u 1 , proving

the p;,oposition.

proposition 5: is a homomorphism of Pin(k) onto ~(k)

and ker 0 = (1,-1}

Proof: Every element of Pin(k) is a finite product of elements

k-1
of S , so to prove p is a homomorphism it suffices to take
k-1 k
u,v E S and x E R Then

p(uv)(x) = ~(uv)x(uv) * o.(u) (t;,(v)xv*)u* p(u)(p(v)x)

k
It now fo110ws that for each u E Pin(k) , p(u) does map R into
k
Rand it is orthogonal since it is a product of ref1ections. In

Chapter VII we showed that ~(k) is generated by ref1ections so

that p is surjective.

It is c1ear that 1 and -1 are in the kerne1 of p • If we

have in the kerne1 of p with r > 1 (we cannot have

r = 1 ) we get a contradiction as fo110ws: If p(e i1 ... e ir ) is the

k
identity, then for every x E R we must have

Taking x = eil gives and this

 140

gives e 2" ,e,

i l.r
= 0 since r must be even (the product of an odd

number of reflections cannot be the identity),

-1
Definition: Spin(k) = p (SQ(k)),

Let us calculate Spin(l) , Spin(2) and Spin(3), Cl = c

so Cl* = c - (0) and Pin(l) is the subgroup of Cl* genera ted

by S
o= tel,-el} This is just tel,e l
2
= -1, e l
3
= -e l , e l
4
",.1)
Z
"''4'
Now p(e l ) = p(-e l ) is a reflection and

Spin(l) = tl,-l} '" Z/2

JH - (0) and Pin(2) is the subgroup generated by

1 2 2
S = tael + be 2 I a + b = 1)

Now (ae l + be 2 ) (ce l + de 2 ) -(ac + bd) + (ad - bc)e l e 2 ' so that

Pin(2) must contain

S l} ,

Actually it is easy to see that

1
Pin(2) = S \J Sand Spin(2) S

Similar computations easily give

Spin(3) l}

and we see that

Spin(3) '" Sp(l) ,

We conclude this section by calculating the centers of the

 141

spin groups,

First off we a1ways have f1,-1} in the center, This is all

of the center in Spin(l) (being all of Spin(l)) and Spin(2)

equa1s its center, So we now assume

k ~ 3 ,

fai1s to contain e, Consider

J

r-1
(e. 1 .,.e. )(e. e.) (-1) e. (e. 1 " oe. )e.
1 1r 1r J 1r 1 1r J

r-1 r
(-1) (-1) (e. e i .) (e 1" ,ei)
1r J i r

-Ce. e .. )(e. 1 " ,ei)

1r 1J 1 r

proving e i1 ",e ir is not in the center, Thus e 1 e 2 " ,e k and

-e 1 e 2 ",e k are the on1y candidates (besides 1 and -1) , Now

for k odd e 1 " ,e k is not in Spin(k) For k = 2n we see that

(-1) (2n-1)r 1

so that is in the center,

Proposition 6: (i) If k is odd Center Spin(k) f1,-1}

(ii) Center Spin(2n) = z/4 if n is odd;

(iii) Center Spin(2n) = Z/2 ffi z/2 if n is~.

Proof: We have proved (i) and we h~ve seen that

Center Spin(2n) = f1,-1,e1",e2n,-e1",e2n} Note that

n(2n+l)
(e1",e2n)(e1",e2n) = (-1) wh ich is 1 if n is even and
 142

is -1 if n is odd. This proves the proposition.

C. The isomorphisms

We have already noted that if eil ... e ir t Spin(k) then r is

even. In this ca se we have

So that if x t Spin(k) , x is a linear combination of "even-

graded" basis elements and xx* 1 .

Definition: Define !/J: Ck _l ~ Ck by

1, ... , k-l .

Then !/J extends to an algebra homomorphism since

and

we see that !/J is an isomorphism of Ck _ l onto the even-graded

part of Ck

Proposition 7: The isomorphisms ~ (§A) and !/J satisfy:

!/J(x* )

1!i (S1) (1\1 (M) *

Proof: = (-ek)(-e.) = -e.e k = !/J(e. * )

1. 1. 1.

For Cl ,C 2 ,C 3 ,C 4 we check the second formula on the basic

matrices
 143

1M = (-~ _~) ! -eI

11
(iJ(M) ) *

The others work the same way, and the verification for

M4 (C) ! eS is equa11y routine.

Theorem: The composed maps

give isomorphism of Sp(2) with SpineS) and SU(4) with Spin(6) .

Proof: Sp(2) and SU(4) are those matrices such that

MtM = I so that

1 = ~(M
L-_
M) = ~(M)$(M)
* and

We get all even graded units x with xx* 1 in this manner.

Thus

SpineS) c ~~(Sp(2»

Spin(6)'c ~w(SU(4» ,

and ~$(Sp(2» is a c10sed connected manifo1d of dimension 10 and

SpineS) is a c10sed manifo1d of dimension 10, so

SpineS) = ~~(Sp(2». Simi1ar1y, the two 1S-dimensiona1 manifo1ds

Spin(6) and ~w(SU(4» are equa1.

 144

D. Exercises

1. Show that

gives an isoroorphism U(l) ~ Spin(2) .

2. Show that

gives an isoroorphism Sp(l) ~ Spin(3) .

3. Show that

Sp(l) x Sp(l) ~ lHEI>lH! C3 ~ C4

gives an isoroorphism Sp(l) x Sp(l) ~ Spin(4) .

4. Let G be an abelian group with exactly four elements.

Show that G ~ z/4 if and only if G has an element of order 4

and G ~ Z/2 EI> z/2 otherwise.

Chapter 11
Normalizers, Weyl Groups

A. Normalizers

We saw in Chapter X that Sp(l) ~ Spin(3) and Sp(2) ~ SpineS)

(also that Spin(6) ~ SU(4)). The only remaining question in

our isomorphism puzzle is whether Sp(k) and Spin(2k+l) are

isomorphie for k ~ 3. In this ehapter, we will use normalizers

and Weyl groups to show that they are not.

Definition: Let S be a nonempty subset of a group G. Set

N(S) = (x E G I XSx- l = S}

and eall N(S) the normalizer (of S in G). N(S) is a subgroup

of G (Exereise S, Chapter I).

Lemma 1: If S is a subgroup of G, then S is anormal

subgroup of N(S)

Proof: (Exereise S, Chapter I).

We ealeulate now the normalizers of our standard maximal tori in

S3 (= Sp(l)) and SO(3)

Let x be an element of a group G, and define

 146

A : G -;. G
x

by (conjugation by x ). Then A is an isomorphism

x
of G onto G.

.
Our standard maximal torus in Sp(l) is T = (a+~b I a 2+b 2 I) .
Let N be the normalizer of T in Sp(l) If q E N then

A : T -;. T is an isomorphism of the circ1e group T onto itse1f.

q

Lemma 2: There ~ exact1y ~ isomorphisms of the circ1e ~.

Proof: Let (as T above) be the set of comp1ex numbers

of unit 1ength. Then F (1,-1,i,-i1 is the set of a l l 4 th

roots of unity in SI If ~: SI -;. SI is an isomorphism, it

must have ~(1) 1 , and therefore ~(-1) = -1. Thus

0(i) E (i,-i1. Both possibi1ities give isomorphisms and knowing

o on F determines ~. (See Exercise #5.) q.e.d.

Let q E N If A is the identity then q commutes with

q
each x E T so q E T So suppose A is the other isomorphism.
q
-1
Since I ql = 1 , q q , so we have

qiq -i

If we write q x + iy + jz + kw this gives

2 2 2 2
x- + Y z w -1 ,
and since
2 2 2 2
x + Y + z + w 1

2 2
we get x = 0 y Furthermore, i f q jz +kw with z + w 1
 147

-i Hence

N TUT* (disjoint union)

where T* = (jz + kw I z2 + w2 = 1} Thus the quotient group

N/T = Z/2 (T is normal in N), and we have

T -> N -> 7l./2 •

Next we ca1cu1ate the normalizer N' of the standard maximal

r"
torus T' in SO(3) Reca11 that

m
sin
T' -s~n 8 cos e
0

Now T' c N' and we seek the elements of N' - T'

Lemma 3: N' - T' {( c~s

Sl.n
o
r/J
r/J
sin
-cos
o
r/J
r/J
J)}
Proof: Each such matrix is symmetric and is its own inverse

and has determinant +1 , so it is in SO(3). It is a routine,

but instructive, exercise to show that

V (cos
J)( c~s
cos 8
(c~s
Sl.n
r/J
r/J
sin r/J
-cos r/J -sin 8
sin 8
cos
0)(
o Sl.n
r/J
r/J
sin r/J
-cos r/J o sin 88 ~in
= cos 88 0)
0
0 0 0 0 1 0 0 - 0 0 1

I t is 1ess routine (but also instructive) to show that these are the

on1y matrices in SO(3) which give the nontrivial isomorphism

of T'

So we have:
 148

T ~ N ~ Z/2 for 5p(1)

T' ~ N' ~ Z/2 for 50(3)

and both T and T' are eirele groups.

proposition 1: N' ~ N .

proof: If rJ>: N ~ NI 'Illere an isomorIilism it would have to

send T to T' and N- T to N' - T'. If x E N - T we would

have rJ>(x 2) = (rJ>(x))2 But it is easy to eheek that if
2 2
X E N - T , then x -1 and i f YEN I - T ' , then y = I

50 we see that N and N' are groups having eirele subgroups

and isomorphie quotient groups (namely Z/2), but N and N' are

not isomorphie. They are different extensions of 51 by z/2

Definition: Let A be an abelian normal subgroup of Band C

be the quotient group,

~ being the inelusion and B being the quotient map. We say say

this extension splits if there is a homomorphism y: C ~ B sueh

that BOy is the identity on C. (y is called a homomorphic

cross section.)

Then T' ~ N' ~ Z/2 splits (send identity to identity and send

the non identity element of z/2 to (~-~ ~)), but T ~ N~ Z/2

o 0-1
does not split (beeause we must send identity to identity and the

non identity element of Z/2 into N - T and the result is not

 149

a homomorphism).

This is the way we are going to distinguish Sp(k) and

Spin (2k+1) for k ~ 3 .

Let T be a maximal torus in a matrix group G and N be

the normalizer of T. We have

where W N/T, ~ is the ine1usion and B is the quotient map.

Definition: W is ea11ed the Wey1 group of G. We note that

if we used another maximal torus T' , then T' = XTx- 1 and

Ax : G -+ G wou1d induee an isomorphism of T to T' , N to the

normalizer N' of T' and henee induee an isomorphism of Wey1 groups

(Exereise). Simi1ar1y, isomorphie groups will have isomorphie Wey1

groups.

p,roposition 2: Let x, YEN • Then A and A are the same

x Y
isomorphism ~ T .. x and y be10ng to the same ~ of T in N

proof: ~

-1 -1
x = Y'r for some 'r E T So Ax(t) = xtx -1 Y'rt(y'r)
-1
Y'rt'r Y
-1
and sinee T is abelian this is just yty A (t)
Y

-1 -1 -1 -1-1
If xtx yty for all t E T , then (y x)t(y x) = t so
-1 -1
that y x eommutes with all t E T aud T is maximal so y x E T .
 150

Corollary: W acts on T .

-1
Proof: For w E W choose x E 8 (w) and define w(t)

By Proposition 2, this is independent of the choice of x.

Proposition 3: W is ~ finite group.

proof: By the results above it suffices to show that a torus

T has only finitely-many distinct isomorphisms. Now

T = Sl x ••. x Sl and an isomorphism ~: T ~ T is determined by its

restrietions to the co ordinate circle groups. But ~Isl is deter-

mined by knowing it on the set of fourth roots of unity (l,i,-l,-i)

in Sl. It follows that an isomorphism m of T is determined by

its values on the 4r (dim T = r) fourth roots of unity. But an

isomorphism sends a fourth root of unity to a fourth root of unity.

So each m is determined by its permutation of 4r elements and

so the automorphism group of T is finite.

From this it follows that N is an extension of T by a finite

group. In particular then, T is the connected component of N

containing the identity. We have

GG T

G XT~
~yT

81 G G
W
.
id
 151

c. Spin(2n+l) and Sp(n)

An isomorphism of Spin(2n+l) onto Sp(n) would induce an

isomorphism of their centers and would induce

Spin(2n+l) ~ ~
center - center

Theorem: For n = 3,4,5, ... the normalizer of a maximal torus

Spin(2n+l)
in splits, but the normalizer of a maximal torus in
center

Sp(n)
center does ~ split. So for n = 3,4,5, ...

Spin(2n+l) ~ Sp(n)

We will prove slightly more; namely,

(*) The normalizer in Sp(n) does not split for any n

The normalizer in ~ splits n E (1,2)

center

The normalizer in Spin(2n+l)

SO(2n+l) ) splits for
center
n = 1,2,3, . . . .

Proof of (*): We want to show that

does not split in Sp(n) ; i.e. that no homomorphic cross-section y

can exist. We will suppose y exists and obtain a contradiction.

~ = (jl.~) ~
Let
o ·1
E Sp(n) . We claim that E N. Take any
 152

t €
(1o 0) eien
in T and we ca1cu1ate

'I1 t 'l1
-1

o ("1
(lI.?) 0 )("
~""ei9n ""I
'1
0) ;"

iA 1
je (-j)
0
i8 2
e
0 i9 n
e

i8 1
. -i8 -1
and je (-j) j(cos 81 + i sin 81) (- J) = e 1 so that 'I1 t 'l1

.0)
is aga in in T
2 (-11 •
Let w = B (T'\) Then, since '11
o '1
€ T we have that

2 2
w = B(T'\ ) = 1 E W. On the other hand, since y is assumed to be a

cross section we have

'1')' y(w)

must be in the same coset of T as 'I') iS, and since y is to be a

homomorphism we must have

2 2
('rI' ) y(w ) y(l) I

So we can comp1ete our proof of (*) by showing that no such T'\'

exists; i.e. no element in the coset of 'rI square to the identity.

Any such '11' can be written 'I') 'I'1t

 153

l
T1 '
. ) ( i8 )
( J" " , ." .iB
n

o
,,'=
i8
n
e

Thus
. i81. i 8l

2
Je Je
2i8 2
o
(" ,) e

o e
2i8
n

-1
o
" I

o e
2i8
n

and (*) is proved.

Proof of (+): We want to prove that the normalizer in

~ splits .. n E (1,2} . For n = 1 , ~ = 50(3) and we
center center
have seen that N splits for 50(3) (§A) For n = 2 we have

5p(2) "" 5pin(5) and ~

center
"" 50(5) and we will show in (T) that

50(odd) always splits. So it suffices to show here that ~

center
 154

does not split for n ~ 3 .

For notational convenience we will prove this for n = 3 and

the general argument will be clear. The idea is like for the proof

of (*). We need to find a relation in W which cannot lift (via y)

to a relation in N

o
~) (~ ~)
1
Let 1 and cr o and one checks easily
o o
that ~,cr are in N Let

wl B (TI) and w2 = B(cr)

2 2
Also since ~ and cr are in T we have

2 2
wl 1 and W z 1

Also

(~ ~)
j
~cr 0
0

I t is left as an exercise to show tha t the fourth power of 'l"\cr lies

in T So we have

If we had a homomorphic cross section then we would have to have

(y(w l ))
zE [I,-I} (= center)
Z
(y(w Z)) E (I,-I}

4
(y(wl)y(w Z)) E [I,-I}

So we will assume that we have ~' in the coset of '1"\, cr' in the
 ISS

coset of a such that (~,)2, (a,)2 and (~'a,)4 are all in

(1,-11 .

Lemma a: (~,)2 -I.

(" '",
Proof: We have
Je 0

)A,)
i 82
~ , ~t 0 e

0 0

and (~' ) 2 cannot be I since it has -1 in the 1,1 position.

-I and Tl'

Lemma b: For some positive integer m, a' may be written as

~:'
(_l)m/dJ

a' 0

0
i~)
Proof: We have

a' at

a' o
o
 156

2
Then (0' )

and this must be I or -I This shows that is 1 or -1

i dJ 3 m
so the e = i for sorne m. 1f m is odd (0,)2 must be -I
i0 1 i 02
and then e e -1 shows we can take dJ = -dlt. 1f m is even

we can take °
= 1/)1 and e
i0 2 m il/)
(-1) e ,and Lemma b is proved.

We finish (t) by

4
Lemma c: (Ti'a' )

Proof: Frorn Lemmas a and b we have

. i</>( - l)me il/)

Je

Tl'a' o
o

and a routine (if sornewhat tedious) ca1cu1ation proves the lemma

and hence (+)

D. SO(n) sp1its

We showed that the normalizer in SO(3) sp1it, but that was easy

because we knew the Wey1 group was just Z/2, so we simp1y had to

find an element of N - T whose square equa11ed I. (lt turned out

 157

that any element of N - T would do.) For n> 3 we will have

bigger Weyl groups. For notational simplicity we will work with

50(5) first and then consider the general case.

Let

cos 8 1 sin 9 1 0 0 0

91 cos 9 1 0 0 0

T(8 l ,8 2 ) 0 0 cos 8 2 sin 8 2 0

0 0 -sin 8 2 cos 8 2 0

0 0 0 0 0

be an element in our maximal torus. Let pl,P 2 be the two 2-planes

5
in p spanned by coordinates 1,2 and 3,4 respectively. Let ~

denote the line of the 5 th coordinate.

proposition 4: Pl,P2'P ~ the only subspaces stable for each

T(8 l ,8 2 ) in T. If N is the normalizer of T and A E N , then

PA P and

proof: The first statement is clear. Now PA is sorne line,

-1
say AA = 1<' 5ince A E N for every T E T we have ATA E T ,

so that Then

-1
p'TA R or P'T PA p' ,

proving that ~'= ~ .

P. For any 'T E T we

have
 158

So P,. = P 1A = P Thus P E fP1,P2). üf course the same proof works

for P2 .

Consider two specific elements of N

(~ j) ~)
0 0 0 0 0 1
1 0 0 0 1 0

A -
0
0
0
0
1
0
1
0
0
B =

(1 1
0
0
0
0
0
0
0
0

We have A- 1 A and we easily find that

A'r(81'A 2 )A ,.( 91'-9 2 ) .

Also B B- l and

If we think of these as actions on the tangent space !(T) of

T (i.e., take differentials) we see that these are reflections

"
"""
"
""
"
", BA
",
----------~--------------~~----_7 91
""
""
""- "-
" SB = w2

X
 159

We have T ~ N ~ W. Let w1 = S(A) and w2 = S(B). It is not

hard to see that w1 and generate a subgroup W' of W with

W' containing all permutations and sign changes of (Sl,S2}.

Reca11 that no two different elements of W have the same action

on T (Proposition 2), so if W' contains all possib1e isomorphisms

of T, it will fo110w that W' = W.

Proposition 5: The isomorphisms of T = sl x sl are in 1-1

correspondence with the isomorphisms of Z x Z .

proof: We can represent T as a quotient group T = .!L.2:....]

Z x Z
Thus any linear isomorphism of R x R which sends Z x Z onto

z x Z gives an isomorphism of T But any isomorphism of Z x Z

is extended easi1y to such a linear isomorphism of R x R •

Converse1y, any isomorphism of T has a differential wh ich is a iso-

morphism of q x R sending z x Z onto Z x Z .

Now isomorphisms of ~ x z are easy to determine. They cor-

respond 1-1 with the set of eight permutations-and-sign-changes on

(l,O),(O,l)}. So the Wey1 group of SO(5) is the set of permu-

tations-and-sign-changes on (A 1 ,9 2 } I t is genera ted by the two

2 2 4
ref1ections w1 ,w 2 and we have w = 1 w2 1 and (w 1w2 ) = 1
1
(This last since w1w2 is a rotation of %.)
We have now that SO(5) splits. For if we let y(l) = I ,
y(w 1 ) = A, y(w 2 ) = B we see that it on1y remains to check that
 160

We11,

(~
0 0 1

J)
0 1 0
AB 0 0 0
1 0 0
0 0 0

Then

~)
1 0 0

(l
Z 0 0 0
(AB) = 0 0 1
0 1 0
0 0 0

and so

~)
0 0 0

(i
1 0 0
(AB) 4 = 0 1 0 I
0 0 1 ,
0 0 0

We have noticed before that the groups SO(odd) and SO(even)

are somewhat different. Let us ana1yze SO(4) just as we have

,(l ~)
1 0
0 0 c- 1
so (5) . Let C
0 0
Then = C and
0 1

C1'(9 1 ,9 Z)C ,.(-9 1 , -9 Z)

~)
0 0
0 1

(1
Let B Then B E SO(4) , B- 1 = B and
1 0
0 0

B1'(8 1 ,8 Z)B 1'(-8 Z ,-8 1 )

 161

So we can permute and and

we can change both signs. We claim that no element of W can

change just one sign. Any DEN must be of the form

or of the form
~~0
~

(see Proposition 4). And each shaded area is an orthogonal Z x Z

matrix. If we had aDE N such that changed just

one sign but also permuted we cou1d combine with B to get

one changing one sign but not permuting. Such a matrix cannot have

the form But i f with being Z x Z

orthogonal matrices and

that (R(8) = ( cos 8 sin 8) ). This imp1ies that

-sin A cos 8

p commutes with all rotations. Thus p is a rotation, so

det p = +1. But then cr is a rotation and we cannot have

as required.

Otherwise the homomorphic cross section proof goes just as for

The elements w1 = s(C) and Wz = S(B) generate Wand

C, y(w Z) = B generates a homomorphic cross section. The

general case is more difficu1t on1y because of notation. The Wey1

group SO(n) is generated by all permutations of [81, ... ,8n1 and

all even sign changes.

 162

E. Exercises

1. Show that an isomorphism of Lie gtDUPS induces an isomorphism

of their Wey1 groups.

2. Prove (t) for ;ePn~n)r (n 2 3) (i.e., genera1ize the

proof given for ~)
center

3. Show that reflection in L

x
is given by

sin
21/J)
-cos 21/J

4. Show that, with r as in 3 ,

r
( cos e
-sin e
sin

cos :) r
( c~s
s~n
e

e
-sin

cos :)
S. Show that the on1y isomorphism of s1 = [eie} which

1eaves i fixed is the identity. Show that there is exact1y one

isomorphism of S1 which sends i to -i .

Chapter 12
Lie Groups

A. Differentiable manifolds

If U is an open set in Rn and we have a function f: U -7 rr ,

we say that f is smooth (or C=) if all higher partial derivatives

of fexist and are continuous. The composition of smooth functions

is smooth. In the case m = n if f: U -> Rn is one-to-one onto

f(U) ,with f(U) open in and both f and are smooth

then f is a diffeomorphism (from U to f(U)) .

Let M be a n-manifold (§E, Chapter VI). By definition, for any

p E M we have U open in Rn and a homeomorphism

into M with ~(U) being an open neighborhood of p. Such a pair

(U,~) will be ca lIed achart. A collection of charts such that the

~(U)'S cover M will be called an atlas.

!Wo charts dJ: U -> M and 1iJ: V -> Mare said to overlap smoothly
 164

-1
i f (either ~(U) n ~(V) ~, or) 1\1 0 t1J is a diffeomorphism.

Definition: M is a differentiab1e manifo1d if it has an atlas of

smooth1y over1apping charts.

Given such an atlas we can maxima1ize it. We simp1y add in all

charts which over1ap smooth1y with the given ones in the atlas. Since

compositions of diffeomorphisms are again diffeomorphisms we see that:

If two new charts over1ap smooth1y with the given charts in the atlas,

then they over1ap smooth1y with each other. We ca 11 a maximal atlas

a differentiab1e structure.

Suppose we have differentiab1e manifo1ds M and N and a function

f : U .... N with U an open set in M. What shou1d it mean for f to

be smooth? For x € U choose charts t1J: V .... M with t1J(V) an open

neighborhood of x and 1\1: W .... N with \iJ (W) an open neighborhood

-1
of f(x) We say f is smooth at x if 1\1 0 f 0 t1J is smooth at
-1
t1J (x). It is easy to check that this definition is independent of

the choices of charts (within the maximal atlases).

A smooth one-to-one map f: M .... N with a smooth inverse is a

diffeomorphism of the differentiab1e manifo1ds. If M= ~ =N and f

3
is given by f(x) = x then f is smooth and one-to-one, but

is not smooth (since its derivative fai1s to exist at 0) so that f

is not a diffeomorphism.

B. Tangent vectors, vector fie1ds

n
If we have a smooth curve or surface in some ~ then the concept
 165

of a vector tangent to the curve or surface is not hard to define. But

our differentiable manifolds are to be thought of as spaces in their

own right - not necessarily sitting in sorne euclidean space. How then

do we define "tangent vector?" The idea is that if we have a vector

in Rn we can use it to differentiate functions - essentially taking

directional derivatives. So we will call a tangent vector a thing

which differentiates functions.

Let M be a differentiable m-manifold and p E M. We set

A(p) {(U,f) I p E U open, f: U ~ R smooth}

We used A(p) to denote this set of pairs (U,f) since we are going

to make it into an algebra. It is routine to verify that if we define

(U ,f) + (V, g) (U n V, f + g) and

r (U, f) (U, rf)

that these operations make A(p) into areal vector space. Then we

define

(U, f) (V, g) (U n V, fg)

and A(p) becornes an algebra.

Definition: A tan~ent vector Tl to M at p is a linear map

" : A(p) ...,. ~ such that:

(i) if f g on some neighborhood of p , then Tl(f) Tl( g)

(11) ,,(fg) f(p),,(g) + Tl(f)g(p)

Suppose f is a constant function, f(x) = r for all x in sorne

 166

neighborhood of p. Then fg = rf in that neighborhood, and if ~

is any tangent vector we have T1(rg) = r,,(g) since is linear.

On the other hand, by (ii) we have

~(fg) f(p)~(g) + ~(f)g(p)

So we see that T1(f)g(p) = 0 for all g. Thus T1(f) = o. So a

tangent vector sends any constant function to zero. Also if we have

two functions f,g such that f(p) = 0 g(p) , then for any tangent

vector T1 to M at p we have T1(fg) o (by (H) ) .

Any chart tJ;: U ..,. M (with P E tJ; (U)) in the differentiable struc-

ture gives some tangent vectors as follows. If f is in A(p) , then

fO tJ; is a map of U C Rn into Rand we let

~i (p) (f)

-1
be the i th partial derivative of f 0 rb at tJ; (p) Thus

3 i (P) : A(p) ..,. R is a tangent vector. So the chart (U,tJ;) gives n

tangent vectors 0l(P) , ... , ~n(p) at p.

Let T M be the set of all tangent vectors to M at p. If we

p
define operations by

~(f) + 'lief)

(r~) (f) r~ (f) ,

then it is routine to verl fy that T M becomes areal vector space.

p

proposition 1: dim T M = n (= dim M)

p
 167

Proof: We have seen that if we take a chart ~: U -+ M in the

differentiable structure with p E ~(U) , we get n tangent vectors

al(p) , ..• , !n(p) in TM We will show that these are a basis for
P
TM
p
We may assume ~(O) = P Then for any f E A(p) we can write

n
f(x) f(p) + ~
i=l

for suitable smooth functions This is just the multivariate

version of the following result about functions of one variable. Let

f be a smooth real-valued function defined on seme neighborhood of

o in R. We note that

f(x)-f(O) _ f'(O)
x
Hx)
x

is smooth (and defined for x # 0). Solving for f(x) gives

2
f(x) f(O) + f'(O)x + $(x)x

Now let ~ be any element of TM and apply to f . is

P
linear, so ~(f(p)) =0 since f (p) is constant, and ~(~i. (x)x.x.)
J ~ J
= 0 since both ~i.(x)x. and x. vanish at p. Thus
J ~ J

and we conclude that

~(xl)! (p) + ... + ~(x )a (p)

1 n n

showing that spans TM.

P
Now these tangent vectors are linearly independent, for if
 168

is the zero tangent veetor and we apply it to the funetion xi we get

q.e.d.

Definition: A veetor field X on an open set W in M is an

assignment of X E TM for eaeh p E W .

P P

We are interested in veetor fields whieh are eontinuous in the sense

that if points p and q are elose then X and X are elose.

p q
But sinee X and X are in different veetor spaees, this looks
p q
diffieult to formulate. Aetually there is a neat way of even defining

a smooth veetor field.

Let X be a veetor field on Wand let f: W -7 'R be smooth.

We get a new funetion

Xf : W -7 R

defined by (Xf)(p) X (f) . This makes sense beeause f E A(p) for

p
eaeh p E W

Definition: The veetor field X is smooth if for eaeh smooth f,

we have Xf smooth.

It is easy to verify that if X and Y are smooth veetor fields

then so is X + Y ,

(X+Y) (f) X(f) + Y(f)

Also if X is a smooth veetor f'ield and r E ", then rX is smooth

(rX(f) = r(X(f») . Thus the smooth veetor fields form a veetor spaee.

For a eoordinate ehart ~: U -7 M in the differentiable strueture

 169

we have smooth vector fields (11'···' an So i f are

smooth real-valued functions on U we have

as a smooth vector field. Conversely, if X is smooth on U, we

have

and the functions are smooth because the

function obtained by applying the smooth vector field X to the

smooth function Xi
If we have two smooth vector fields X and Y we can get a real-

valued operator on real-valued smooth functions as follows:

f ~ X (Yf)
P

Since Y and f are smooth, so is Yf and therefore X assigns

p
the real number X (Yf) to this function. But this operator from
p
A(p) to ~ need not be a tangent vector. It is linear and depends

onlyon f near p (condition (i) ), but it may fail to satisfy con-

dition (ii) in the definition of a tangent vector. Indeed, if X 0i

and then (ii) is false because the operation is just the

mixed second partial derivative.

proposition 2: For smooth vector fields, X,Y the operator

f ~ X (Yf) - Y (Xf)
P P

is a tangent vector. ("The mixed second partials cancel out") .

 170

Proof: Let X be a smooth field and f,g be smooth functions.

We assert that

(f) X(fg) = (Xf)g + f(Xg)

At a point p the left hand side is

X (fg) = X (f)g(p) + f(p)X (g)

P P P

since X satisfies (ii), and this is just the right hand side
p
evaluated at p Of course, we have a similar formula for Y(fg)

Then

X (Y(fg)) = X «Yf)g + f(yg))

P P

= X (Yf)g(p) + (Yf) (p)X (g) + X (f)(Yg)(p) + f(p)X (Yg)

P P P P

and similarly

Y (X(fg)) Y (Xf)g(p) + (Xf)(p)Y (g) + Y (f)(Xg)(p) + f(p)Y (Xg)

P P p P P

Thus

(X Y - Y X) (fg) = (X Y - Y X) (f)g(p) + f(p)(X Y - Y X)(g)

P P P P P P

proving th~ condition (ii) is satisfied. The linearity and condition

(i) are true since they are true for both terms. So

XY - YX is a tangent vector at p
p p

Let [X,Y] = XY - YX be the vector field defined by

[X,Y] = XY - YX
P P P
 171

Proposition 3: The set !(W)

set W in a differentiab1e manifo1d M forms.~ Lie algebra under

[ ,

Proof: We have seen that !(W) is a vector space. From the

definition of [X,Y] it is obvious that [Y,X] = -[X,Y] and it is

easy to show that this mu1tip1ication distributes over addition

([X+Y,Z] = [X,Z] + [Y,Z] ) Proof of the Jacobi identity is forma11y

just the same as for matrix mu1tip1ication with [A,B] = AB - BA

This Lie algebra !(W) is usua11y infinite dimensional, but

when our differentiab1e manifo1d is a Lie group we will get an impor-

tant finite-dimensional subalgebra. (See section C).

Let M,N be differentiab1e manifo1ds and

be a smooth map. If w(p) q we get a map

ca11ed the differential of W as fo11ows:

If ~ E T M and f E A(q) we set

p

~(f 0 1\1)

Proposition 4: defines ~ map of TM tnto and it is

P
~ linear~.

Proof: We need to show that d1\1~ is a linear derivation from

A(q) to Rand assigns the same real number to two functions which

agree on any neighborhood of q This final condition is inherited

 172

from the same property for ~ and linearity is easy to prove. So

suppose f,g E A(q) We have

dg(fg) ~(fgo 1\1) = ~«fo '11) (go 1\1»

(fo 1\1) (pH(go 1\1) + ~(fo '11) (go .JJ)(p)

= f(q)d1\l~(g) + d.JJ~(f)g(q)

showing that d,u~ is a derivation.

Finally d$: T M ~ T N is linear. For we have
p q

c. Lie groups

Let G be a differentiable n-manifold which is also a group and

the operations

G ~ G

-1
(a,b) I-i> ab a I-i> a

are smooth functions. Then G is called a Lie ~.

Let G be a Lie group with identity element e and suppose

X is a tangent vector at e (Xe E TeG) Then we can get a vector

e
field defined on all of G as follows. For any g E G let

Lg : G ~ G be the diffeomorphism given by L (x) = gx for each

g
x E G . This is called left-translation by g . We set

dL gXe
 173

(dL : T G -'> T G). Such vector fields are called left-invariant; Le.
g e g
a vector field X on G is left-invariant if it satisfies (t) .

Proposition 5: If X,Y are left-invariant vector fields on G ,

so is [X,Y] .

Proof: Let g E G and f E A(g) and we calculate

dL [X,Y] (f) = [X,y]. (f 0 L )

g e e g

= X (Y (f 0 L )) - Y (X (f 0 L ))
e g e g

= dL gXe (Yf) - dL gYe (Xf)

Xg (Yf) - Yg (Xf)

= [X,Y]g(f) , proving (t)

Now one can see easily that X,Y left invariant implies X+Y

is also and so is rX (r E R) . Thus the set of left-invariant

fields on G becomes a subalgebra of the Lie algebra of all sm06th

vector fields. Since left-invariant vector fields correspond one-to-

one with elements of TeG, this Lie algebra is n-dimensional. We

denote it by X(G) and call it the Lie algebra of G. It is con-

venient to use the language of categories and functors to discuss

passing from G to X(G)

Categories and Functors

A category consists of objects A,B,C, ... , and for each pair

A,B of objects a set Hom(A,B) called the morphisms from A to B .

Each Hom(A,A) is required to contain the identity morphism iA ,

 174

and there must be a law of composition so that ~ E Hom(A,B) and

B E Hom(B,C) give a unique B0 Cl. E Hom(A,C) .

Examples:

(i) objects - sets

morphisms - functions

composition is ordinary composition of functions

(ii) objects - groups

morphisms - homomorphisms

ordinary composition

(iii) objects - vector spaces

morphisms - linear maps

ordinary composition

(iv) objects - differentiable manifolds

morphisms - smooth maps

Definition: Y E Hom(A,B) is an isomorphism if there exists

o E Hom(B,A) such that

1\ 0 y

In (i) an isomorphism is a one-to-one function, in (ii) it is a group

isomorphism, in (iii) it is a linear isomorphism and in (iv) it is a

diffeomorphism.

If Cl and C2 are categories a functor

F : Cl -> C2

must send objects to objects and morphisms to morphisms such that:

 175

we have either

A ----=. F (A)

Cl1 1 F(rJ)

(I) B ----=. F (B) with F (8 0 Cl)

s1 1 F(S)

C -F(C)

or

(II) B ~ F(B) with F(S 0 Cl) F(a.) 0 F(B)

8i r F(S)

C ------,. F ( C)

In case (I) F is called a covariant functor and in case (11)

it is a contravariant functor.

If Cl and C2 are both example (ii) above, then the functor

which assigns to V its dual v* gives a contravariant functor.

We have a category with Lie groups as objects and smooth homo-

morphisms as morphisms and we have a category of Lie algebras and Lie

algebra homomorphisms.

Proposition 6: The assignment G ~ !(G) gives! covariant func-

tor from the Lie ~ category ~ the Lie algebra category.

proof: We must first say how a homomorphism of Lie groups induces

a homomorphism of their Lie algebras. Let

 176

be a homomorphism of Lie groups. We get a linear map

by taking the differential of 0. Since these can be identified with

~(G) and t(H) we have a linear map of the Lie algebras. We must

just check that it is a homomorphism of Lie algebras; i.e. that it

preserves the produet [X,Y].

Let X,Y E t(G) (i.e. left-invariant vector fields on G) and

we calculate

d0[X,Y]f [X,Y](fo 0) X(Y(f 0 0)) - Y(X(f 0 dl))

d0X(Yf) - d~Y(Xf) [d0X,ddlY]f

Sinee this holds for all f, d~[X,Y] = [d0X,d~Y] as required.

The functor ~ does not map objects in a one-to-one manner

two nonisomorphie Lie groups ean have isomorphie Lie algebras. In

fact, if ~: G -'> H is a Lie group homomorphism which is a diffeo-

morphism on some neighborhood of e in G, the Lie algebra homo-

morphism d0: t (G) -'> t (H) will be an isomorphism of Lie algebras.

Recall the two-to-one homomorphism p of Sp(l) onto SO(3) . We

see that dp is an isomorphism of their Lie algebras (we proved

these are isomorphie in Exereise 5 Chapter IV).

It turns out, however, that every real finite-dimensional Lie

algebra is the Lie algebra of some Lie group. Indeed there is the

following theorem (which we are not prepared to prove here - see page

133 of Hoehsehild, The Strueture of Lie Groups, Holden-Day, 1965).

 177

Theorem: Let t be ~ finite dimensional real Lie algebra. Let

G be the group of self-isomorphisms of !. Then !(G) "" ! .

We conclude this section with some remarks about subgroups and

subalgebras. We would like somehow to have a subgroup H of a Lie

group G have its Lie algebra !(H) be a subalgebra of the Lie

algebra t(G). The catch is that H may not be a Lie group at all.

The simple classical example is: G Sl x Sl (a 2-torus).

2
G = exp(R) where exp(x,y) (e
~~
,e
~~
). Let L
.
be a l1ne

through (0,0) in R2 making angle e with the x-axis where e is

2TT
'chosen so that is irrational. Let H = exp(L) . This is a sub-
T
group of R2 but is not a Lie group since it is not a manifold.

It is dense in G (every point of G is a limit point of H) .

If, on the other hand, ~TT is rational then H will be a circle sub-

group of G. The result (which we will not prove here) is that a

closed subgroup of a Lie group is itself a Lie group. For G and H

as first described (H bad) we have that !(G) R2 is the trivial

Lie algebra (always [X,Y] = 0) and thus L is a Lie subalgebra of

!(G) . exp(L) is a subgroup H of G, but the subalgebra L cannot

be !(H) because !(H) does not exist.

D. Connected groups

Proposition 7: Let X be ~ pathwise connected space and let Y

be a subset of X

Y = X •
 178

Proof: We assume Y is not empty and choose y E Y. For any

x E X take a path p: [0,1] "* X from y to x.

Let C be the points t E [0,1] which map into Y. We have

oE C (since p(O) = y E Y), and we will show 1 E C so that

x E Y

Now C is closed. For any limit point of C maps to a limit

point of Y (by continuity of p) . This limit point of Y belongs

to Y so, by definition of C, the limit point of C belongs to C

If 1 'C let t o be the least upper bound of C Since C is

closed t o E C, and thus p(t O) E Y. Since Y is open and p is

continuous, some interval (t o - g, t o+ €) maps into Y, contra-

dicting that to is the least upper bound of c. Thus 1 E C and

x E Y •

Proposition 8: If G is ~ pathwise connected Lie ~ and H

is a subgroup which contains ~ open neighborhood U of e in G

then H = G .

Proof: Let

U2 {XY!XEU,YEU}

U3 {xyz! x E U, Y E U, Z E U}, etc.

Since H is a subgroup we see that

lies in H. As a union of open sets, W is an open set. But W is

also closed, for suppose x is a limit point of W. Then the open

set xU contains some

 179

That iS, there is u E U such that

u
m

but then

By Proposition 7 we see that W being both open and closed and

nonempty implies that W= G Since WeH we have H = G .

Corollary: Let f: K ~ G be ~ homomorph1sm of L1e groups w1th

G pathwise connected. If f(K) contains ~ open ne1ghborhood of

e in G, then f is surject1ve.

If G is a Lie group which is not connected, let GO denote all

x E Gwhich can be connected to the identity element e by a path

in G We call GO the identity component of G.

Proposition 9: The identity component GO of G is ~ pathw1se

connected subgroup of G

Proof: GO is pathw1se connected, for i f x,y E GO we can choose

-1
paths p,a from e to x,y and then p a 1s a path from x to y

and it lies in GO since each p (t) and a(t) belong to GO

GO is a subgroup because it contains e and i f x,y E GO w1th

paths p,a then the path

(pa) (t) p(t)a(t)

is a path in G from e to xy. Thus xy E GO Similarly, i f

is a path from e to x, then

 180

-1
,. (t) (p (t»

-1
is a path from e to x

Go is open. For if x E GO we can take an open neighborhood

n
U of x in G homeomorphic to an open ball in l< (G is an

n-manifold). Since x can be connected by a path 0 from e and

each point in U can be connected by a path from x in U, we see

that U c GO .

GO is closed. For if x is a limit point of GO we can take

U as above. Some point y in U is in GO and so we have a path

cr from e to y. Since we have in U a path from y to x we

see that x E GO
We are going to need an exponential ~ from T G to G and
e
we cannot use the one from Chapter IV because the elements of T G
e
and G cannot be assumed to be matrices. If y: (-€,€) ~ G is a

smooth curve with y(O) = e we define the derivative y' (0) of y

at e to be

y' (0) dy(~)

where ~ is the unit vector 1 in R (= TO(R) ) . (In the exercises

one shows that for matrix groups this agrees with our definition in

Chapter II) .

We will need the following uniqueness theorem from differential-

equation theory (see Milnor, Morse Theory, Annals of Math Studies

#51, 1963, Lemma 2.4).

Theorem: ~ any Tl E TeG, there exists exactly ~ ~-

parameter subgroup y in G such that

 181

y' (0) (see Chapter IV )

Definition: Given any ~ E TeG , take y to be the one-parame-

ter subgroup in G such that y' (0) and set

exp(,,) y(l)

Let and be as in this definition. Set

"
y

o (t) exp(t,,)

We claim that this curve in G is just the one-parameter subgroup y.

To see this hold t fixed and consider the curve cr(u) = y(tu)

Then cr is easily seen to be a one-parameter subgroup and

cr ' (0) ty'(O)

Thus, by definition of exp, exp(t,,) cr (1) y(l) y(t) So we

have proved

Proposition 10: t ~ exp(t~) ~ the unique one-parameter sub-

~ whose derivative at 0 is ".

This implies that exp when restricted to a one-dimensional

subspace of T G is a homomorphism into G .

e
For matrix groups we saw that the exponential map was a diffeo-

morphism of some neighborhood of I in the group (having log as

inverse). This is still true in our more general setting and is based

on the

Inverse Function Theorem: Let M,N be differentiable n-manifold

and 0: M ~ N be ~ smooth map. If ddl: T M ~ T ()N is an

p dl P --
 182

isomorphism, then ~ is a diffeomorphism of ~ open neighborhood

of p .

(A proof of this can be found in most good advanced ca1cu1us

books).

We now app1y this theorem to

Let ~ E TeG and let y be the unique

one-parameter subgroup such that y'(O) = n Then (as we saw in

the proof of Proposition 10) exp(tn) y(t) so that y is the image

under exp of the line t ~ t~ So d(exp) sends the tangent

vector to this line (name1y n) to y' (0) (which is also ~) . Thus

Proposition 11: d(exp) : T G ~ T G is the identity map

e e

Corollary: If G is ~ pathwise connected Lie group, then

exp : TeG ~ TeG is surjective. If G is not pathwise connected,

exp TeG is the identity component GO of G

E. Abelian groups

Just as for abe1ian matrix groups we have the fo110wing result

(and the proof is the same).

proposition 12: Let G be an abelian Lie group. If y,p are

~-parameter subgroups, then so is their product.

Again, just as for abelian matrix groups we have

 183

proposition 13: Let G be an abe1ian Lie~. Then for any

exp(X+Y) exp(X) exp(Y)

Proof: Let y,p be the one-parameter subgroups satisfying

y'(O) = X, p'(O) = Y By proposition 12, yp is a one-parameter

subgroup and (yp) '(0) y'(O)p(O) + y(O)p'(O) = y'(O) + p'(O) = X+Y

Thus exp(X+Y) = (yp)(l) = y(l)p(l) exp(X) exp(Y)

Thus for an abe1ian Lie group G, exp is a homomorphism from

the additive veetor group of TeG to the group G Taking note of

the eoro11ary to proposition 11, we see that if G is pathwise eon-

neeted, then exp is surjeetive. Thus a eonneeted abe1ian ~ ~

~ quotient ~ of the veetor ~ TeG

Again, just as for matrix groups, exp is a diffeomorphism of

some neighborhood of 0 in T G so that the kerne1 is a diserete

e
subgroup of the veetor group Te G So just as for matrix groups

Theorem: A eonneeted abe1ian Lie ~ G is isomorphie to ~

eartesian produet of ~ torus and ~ veetor ~. If G is also eom-

Appendix

In Exercise 8 of Chapter VI the concept of limit point is

defined and it is proved that C is closed ~ (x lp C = X E C) 1. e. ,

a set is closed if and only if it contains all of its limit points.

n
Lemma: If CeR is bounded and infinite then C has a limit

point.

Proof: We must find x E Rn such that each B(x,e) n C is an

infinite set.

Since C is bounded we can find an n - dimensional cube Kl

containing C Let ~ be the length of the sides of Kl . Divide

each edge into two equal parts and cut Kl into zn equal cubes of

side length ~ Call those cubes Kll,KlZ, ... ,KlZn At least one

of those must contain infinitely-many points of C. Choose one such

and call it Subdivide KZ into cubes of side and

choose one of these, K3 , containing infinitely-many points of C.

Continue this process to get Kl ~ KZ ~ K3 ~ ... Clearly

Kl n KZ n K3 n ... is a single point x E Rn

Then x lp C. For, consider any B(x,€). Take m such that

Then

and Km contains infinitely-many points of C. q.e.d.

Theorem: (proposition 5 of Chapter VI) If C c pn is closed

and bounded and

 185

is continuous, then f(C) is closed and bounded.

Proof: Suppose f(C) is not bounded. Choose xl E C such

that Yl = f(x l ) i B(o,l)

Choose X2 E C such that Y2

Choose Xk E C such that Yk = f(x k ) i B(o,k). It is easy to prove

that Y = tY l 'Y2' ... } has no limit point and that X = tXl'x2, ... }

is an ifinite set. Also X c C is bounded. So by the lemma X has

some limit point x. Then also x lp C. Since C is closed x E C.

But then y = f(x) must be a limit point of Y = f(X) by continuity.

Thus f(C) is bounded.

m
Suppose f(C) is not closed. Then there exists y E R such

that y lp f(C) but Y i f(C) . Choose xl E C such that

Yl = f(x l ) E B(y,l) .

Choose x2 E C such that Y2

Choose Xk E C such that Yk = f(x k ) E B(y,~). Clearly y is the

only limit point of Y = tYl' ... 'Yk' ... }. Just as before

X = tXl' ... 'xk, ... } is infinite and bounded. Let x lp X Then

x lp C so X E C and f(x) lp Y = f(X) This implies f(x) = y ,

but f(x) E f(C) . So f(C) is closed. q.e.d.

Index

Abe1ian group 4

Algebra 15

Algebra of matrices 15

Atlas 163

Basis for open sets 82

Binary operation 1

Bounded set 81

Cartesian product of sets 1

Cartesian product of groups 92

Category 173

Center of a group 20

Centers of Sp(l) and SO(3) 64

Center of Sp(n) 100

Centers of U(n) and SU(n) 101

Center of Spin(n) 141

Centra1izer (of a set in a group) 20

Chart 163

C1ifford algebra 134

C10sed manifo1d 87

C10sed set 79
Commutator 69

Commutator subgroup 69

Compact set 81

Comp1ex numbers 9

Conjugation in ~, C, E 23
 187

Conjugate of a matrix 24

Conjugacy of maximal tori 126

Conjugate of a reflection 120

Conjugates of max T cover U(n), SU(n) 110

Conjugates of max T cover SO(n) 114

Conjugates of max T cover Sp(n) 118

Connected set 79

Continuity of a function 76

Cosets of a subgroup 67

Countability 83

Countable basis for open sets 85

Cross section 148

Curve in a vector space 35

Curve in a matrix group 36

Dense subset 124

Diffeoroorphism 163

Differentiable curve 164

Differentiable manifo1d 164

Differentiable structure 164

Differential of a smooth horooroorphism 42, 171

Dimension of a matrix group 37

Dimension of GL(n,R) and GL(n,C) 38, 39

Dimensions of some matrix groups 41

Direct suro of algebras 135

Discrete subgroup 94

Divisors of zero 8, 135

Eigenvector, eigenvalue 107

 188

Eigen spaee 108

Exponentia1 of a matrix 45

Fie1d 7

Funetor 174

Fundamental group of a matrix group 131

General linear groups 16

Generator (of a monogenie group) 125

Group 2

Group extensions 148

Groups of rank 1,2,3 128

Homeomorphism (of spaees) 86

Homomorphism (of groups) 4

Homomorphism of Sp(l) onto SO(3) 61

Homomorphism of Pin(k) onto O(k) 138

Indempotent matrix 33

Identity eomponent of a group 132

Injeetive homomorphism 5

Inner produet 23

Isomorphism (linea r) 14

Isomorphism of groups 6

Isomorphism of Sp(l) and SU(2) 30

Isomorphism of Sp(2) and Spin(5) 143

Isomorphism of SU(4) and Spin(6) 143

Jaeobi identity 57

Kerne1 of a homomorphism 19

Lattiee subgroup of Rn 104

Left translation 60, 71

 189

Length of a veetor 24

Lie algebra 57, 171

Lie group 172

Lie algebras of Sp(l) and SO(3) 58

Linear map 12

Logarithm of a matrix (ne ar I) 49

Loop group n(G) 131

Manifold 87

Maximal torus 95
Maximal torus in SO(n) 97
Maximal tori in U(n) and SU(n) 97, 98
Maximal torus in Sp(n) 99
Metrie 74

Monogenie group 124

Nilpotent matrix 34

Normal subgroup 20
Normalizer ( of a set in a group ) 20
Normalizers (of max. tori) in Sp(l) and SO(3) 147
One-parameter subgroup 51
Open ball 75

Open set 78
Orthogonal groups 27
Path 80
Path in a group 130
Pin(k) 137
Primitive root of unity 22
Projeetion 33
 190

Quaternions 1H 11

Quaternions have square roots 116

Quotient group 68

Rank of a matrix group 127

Reflections 31, 119

Reflections generate ~(n) 121

Rotation group 27

Schwarz inequality 75

Simple group 129

Simply-connected group 131

Skew-Hermetian matrix 40

Skew-Symmetric matrix 39

Skew-symplectic matrix 40

Smooth homomorphism 41
Split group extension 148

Spin(k) 140

Special orthogonal group 29

Special unitary group 29

Stable subspace for a linear map 107

Subgroup 16
Subspace topology 82

Surjective homomorphism 5

Symmetric group 4

Symmetric linear map 112

Sumplectic group 27
Tangent space 37
Tangent vector 35, 165
 191

Table of dimensions, centers, maximal tori 103

Torus 93
Trace of a matrix 54

Transpose of a matrix 24

Triangle inequality 74

Unipotent matrix 34

Uniqueness of one-parameter subgroups 53

Unit (in an algebra) 15

Unitary group 27

Universal covering group 131

Vector field 168

Weyl group 149