Você está na página 1de 340

SCHAUM'S OUTLINE SERIES

THEORV AND PROBLEMS OF

SCHAVM'S OUTLINE OF

THEORY AXD PROBLEMS


OF
-v

ALGEBRA

LINEAR

ed

BY

SEYMOUR LIPSCHUTZ,
Temple University

Ph.D.

Associate Professor of Mathematics

SCHAIJM'S OIJTUl^E SERIES


McGRAW-HILL BOOK COMPANY
New
York, St. Louis, San Francisco, Toronto, Sydney

Copyright 1968 by McGraw-Hill, Inc. All Rights Reserved. Printed in the United States of America. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the
prior written permission of the publisher.

37989

8910 8HSH 754321

liX)mOM

^Q^fv^oiA

Preface
Linear algebra has in recent years become an essential part of the mathematical background required of mathematicians, engineers, physicists and other scientists. This requirement reflects the importance and wide applications of the subject matter.
This book is designed for use as a textbook for a formal course in linear algebra or as a supplement to all current standard texts. It aims to present an introduction to linear algebra which will be found helpful to all readers regardless of their fields of specialization. More material has been included than can be covered in most first
courses. This has been done to make the book more flexible, to provide a useful book of reference, and to stimulate further interest in the subject.

Each chapter begins with clear statements of pertinent definitions, principles and theorems together with illustrative and other descriptive material. This is followed by graded sets of solved and supplementary problems. The solved problems serve to illustrate and amplify the theory, bring into sharp focus those fine points without which the student continually feels himself on unsafe ground, and provide the repetition of basic principles so vital to effective learning. Numerous proofs of theorems are included among the solved problems. The supplementary problems serve as a complete review of the material of each chapter.
three chapters treat of vectors in Euclidean space, linear equations and These provide the motivation and basic computational tools for the abstract treatment of vector spaces and linear mappings which follow. A chapter on eigenvalues and eigenvectors, preceded by determinants, gives conditions for representing a linear operator by a diagonal matrix. This naturally leads to the study of various canonical forms, specifically the triangular, Jordan and rational canonical forms. In the last chapter, on inner product spaces, the spectral theorem for symmetric operators is obtained and is applied to the diagonalization of real quadratic forms. For completeness, the appendices include sections on sets and relations, algebraic structures and polynomials over a field.
first

The

matrices.

Dr.
I

wish to thank many friends and colleagues, especially Dr. Martin Silverstein and Tsang, for invaluable suggestions and critical review of the manuscript. also want to express my gratitude to Daniel Schaum and Nicola Monti for their very
I

Hwa

helpful cooperation.

Seymour Lipschutz
Temple University
January, 1968

CONTENTS
Page
Chapter

VECTORS IN
Introduction.

R"

AND

C"
Dot

Vectors in R.

product.

Norm

Vector addition and scalar multiplication. and distance in R". Complex numbers. Vectors in C.

Chapter

LINEAR EQUATIONS
System of linear equations. Solution of a system of linear equations. Solution of a homogeneous system of linear equations.
Introduction.

18

Linear equation.

Chapter

MATRICES
Matrices. Matrix addition and scalar multiplication. Matrix multiplication. Transpose. Matrices and systems of linear equations. Echelon matrices. Row equivalence and elementary row operations. Square matrices. Algebra of square matrices. Invertible matrices. Block matrices.
Introduction.

35

Chapter

VECTOR SPACES AND SUBSPACES


Introduction.
linear spans.

63

Examples of vector

Row

Linear combinations, space of a matrix. Sums and direct sums.


spaces.

Subspaces.

Chapter

BASIS

AND DIMENSION

86

Introduction. Linear dependence. Basis and dimension. Dimension and subspaces. Rank of a matrix. Applications to linear equations. Coordinates.

Chapter

LINEAR MAPPINGS
Mappings. Linear mappings. Kernel and image of a linear mapping. Singular and nonsingular mappings. Linear mappings and systems of linear equations. Operations with linear mappings. Algebra of linear operators. Invertible
operators.

121

Chapter

MATRICES AND LINEAR OPERATORS


Introduction.
Similarity.

150

Matrix representation of a linear operator. Matrices and linear mappings.

Change of

basis.

Chapter

DETERMINANTS
Determinant. Properties of determinants. Minors and cofactors. Classical adjoint. Applications to linear equations. Determinant of a linear operator. Multilinearity and determinants.
Introduction.

171

Permutations.

Chapter

EIGENVALUES AND EIGENVECTORS


Polynomials of matrices and linear operators. Eigenvalues and eigenvectors. Diagonalization and eigenvectors. Characteristio polynomial, Cayley-Hamilton theorem. Minimum polynomial. Characteristic and minimum polynomials of linear operators.
Introduction.

197

CONTENTS
Page
Chapter

10

CANONICAL FORMS
Invariance. Invariant direct-sum decomPrimary decomposition. Nilpotent operators, Jordan canonical positions. form. Cyclic subspaces. Rational canonical form. Quotient spaces.
Introduction.

222

Triangular form.

Chapter

11

LINEAR FUNCTION ALS AND THE DUAL SPACE


Introduction.
space.

249

Linear functionals and the dual space. Dual basis. Second dual Annihilators. Transpose of a linear mapping.

Chapter

12

BILINEAR, QUADRATIC
Law
of inertia.

AND HERMITIAN FORMS

261

Bilinear forms. Bilinear forms and matrices. Alternating bilinear forms. Symmetric bilinear forms, quadratic forms. Real symmetric bilinear forms.

Hermitian forms.

Chapter

IB

INNER PRODUCT SPACES


Cauchy-Schwarz inequality. OrthogoGram-Schmidt orthogonalization process. Linear nality. Orthonormal sets. functionals and adjoint operators. Analogy between A(V) and C, special operators. Orthogonal and unitary operators. Orthogonal and unitary matrices. Change of orthonormal basis. Positive operators. Diagonalization and canonical forms in Euclidean spaces. Diagonalization and canonical forms in
Introduction.

279

Inner product spaces.

unitary spaces.

Spectral theorem.

Appendix A

SETS AND RELATIONS


Sets,

315
Product
sets.

elements.

Set

operations.

Relations.

Equivalence

relations.

Appendix B

ALGEBRAIC STRUCTURES
Introduction.

320
fields.

Groups.

Rings, integral domains and

Modules.

AppendixC

POLYNOMIALS OVER A FIELD


Introduction.

327
Divisibility.

Ring of polynomials.

Notation.

Factorization.

INDEX

331

chapter

Vectors
INTRODUCTION

in R^

and

In various physical applications there appear certain quantities, such as temperature and speed, which possess only "magnitude". These can be represented by real numbers and are called scalars. On the other hand, there are also quantities, such as force and velocity, which possess both "magnitude" and "direction". These quantities can be represented by arrows (having appropriate lengths and directions and emanating from some given reference point O) and are called vectors. In this chapter we study the properties of such vectors in some detail.

We
(i)

begin by considering the following operations on vectors.

The resultant u + v of two vectors u obtained by the so-called parallelogram law, i.e. u + V is the diagonal of the parallelogram formed by u and v as shown on the right.
Addition:

and V

is

(ii)

Scalar multiplication: The product kn of a real number fc by a vector u is obtained by multiplying the magnitude of u by A; and retaining the same direction if or the opposite direction if k<0, as shown on the right.

k^O

familiar with the representation of the points in the plane If the origin of the axes is chosen at the reference point above, then every vector is uniquely determined by the coordinates of its endpoint. The relationship between the above operations and endpoints follows.

Now we assume the reader is

by ordered

pairs of real numbers.

(i)

Addition:

If

(a, &)

and

(c,

d) are the

will be the endpoint of

u + v, as shown
(a

endpoints of the vectors u and in Fig. (a) below.

v,

then (a +

c,

+ d)

+ c, b + d)

(ka, kb)

Fig. (a)

Fig. (6)

(ii)

Scalar multiplication: If (a, b) is the endpoint of the vector u, then {ka, kb) will be the endpoint of the vector kn, as shown in Fig. (6) above.

VECTORS IN

B"

AND

[CHAP.

Mathematically, we identify a vector with its endpoint; that is, we call the ordered pair of real numbers a vector. In fact, we shall generalize this notion and call an -tuple We shall again generalize and permit the coa) of real numbers a vector. {ai, C2, ordinates of the -tuple to be complex numbers and not just real numbers. Furthermore, in Chapter 4, we shall abstract properties of these %-tuples and formally define the mathematical system called a vector space.
(a, 6)
. . .

We
field

assume the reader is familiar with the elementary properties of the which we denote by R.

real

number

VECTORS IN
The

R"
numbers, denoted by R",
is called

set of all w-tuples of real

n-space.

particular

-tuple in R", say

(til,

Uz,

.,

Un)

a point or vector; the real numbers im are called the components (or: coordinates) of the vector u. Moreover, when discussing the space R" we use the term scalar for the elements of R, i.e. for the real numbers.
is called

Example

1.1:

Consider the following vectors:


(0,1),

(1,-3),

(1, 2,

VS,

4),

(-5,

-1,

0,ff)

The first two vectors have two components and so are points in B^; the last two vectors have four components and so are points in B*.

Two
ponents,
vectors

vectors
i.e.

u and v are eqtial, written u = v, if they have the same number of combelong to the same space, and if corresponding components are equal. The
and
1.2:

(1, 2, 3)

(2, 3, 1)

are not equal, since corresponding elements are not equal.

Example

Suppose (x-y, x

+ y, z-1) =

(4, 2, 3).

Then, by definition of equality of vectors,

X x

y
y

= =

4:

2
3

2-1 =
Solving the above system of equations gives

3,

1, and z

A.

VECTOR ADDITION AND SCALAR MULTIPLICATION


Let u and v be vectors in R":

u =

(Ml, U2,

Un)

and

{Vi, Vz,

.,

Vn)

The sum of u and v, written u + v,is the vector obtained by adding corresponding components:

U + V
The product
of a real

iUi

+ Vi,U2 + V2,

.,Un

+ Vn)
the vector obtained by multi-

number fc by the vector u, written ku, plying each component of u by k: kun) ku (kui, ku2,
. . .

is

Observe that u + v and ku are also vectors in R".

We

also define

-u = -1m
The sum
of vectors with different

and

u- v

m+

{-v)
is

numbers of components

not defined.

CHAP.

1]

VECTORS IN
1.3:

AND

C"

Example

Let u

(1,-3,2,4)

and v
(1

(3,5,-1,-2).

Then

5w

2m Example
1.4:

30)

= =
.

+ 3, -3 + 5, 2 - 1, 4 - 2) - (4, 2, 1, 2) ^ (5, -15, 10, 20) (5 1, 5 (-3), 5 2, 5 4) = (-7, -21, (2, -6, 4, 8) + (-9, -15, 3, 6)

7,

14)
It is similar

The vector

(0, 0,

.,

0) in P.",

to the scalar

in that, for

denoted by 0, any vector u =


,

is

called the zero vector.

(ztj,

%,

u),

(Ml

+ 0, M2 + 0,

+ 0) =

(Ml, 2*2

uj

Basic properties of the vectors in R" under the operations of vector addition and scalar multiplication are described in the following theorem.

Theorem

1.1:

For any vectors u,v,w G R" and any scalars


(i)

k, k'

(ii)

(iii)

(iv)

+ v) + w = u + u u+ u + {-u) = u +v V +u
(u
is

{v

+ w)

(v)
(vi)
(vii)
(viii)

G R: k(u + v) - ku + kv (ft + k')u = ku + k'u (kk')u = k{k'u) Vu, u

Remark:

Suppose u and v are vectors in R" for which u


A;

R.

Then u

for some nonzero scalar said to be in the same direction as v if fe > 0, and in the op-

kv

posite direction if

k <0.

DOT PRODUCT
Let u and v be vectors in R":

u =

(ui, Ui,

t()

and

(vi, Vz,

.,

Vn)

The dot or inner product of u and v, denoted hy wv, is the scalar obtained by multiplying corresponding components and adding the resulting products:

U'V =
The vectors u and v are zero: m v = 0.

UiVi

U^Vi,

UnVn

said to be orthogonal (or: perpendicular) if their dot product is

Example 15:

Let m

(1,

-2,

3,

-4),

-y

(6, 7, 1,

-2)

and

w=

(5,

-4,

5, 7).

Then
3

u-v = 1-6 + M'W = 1'5 +


Thus u and

(-2)'7

3-1 + (-4)'(-2)

(-2) (-4)

+ 3-5 +

(-4) -7

= 6-14 + 3 + 8 = = 5 + 8 + 15-28 =

are orthogonal.

Basic properties of the dot product in R" follow.

Theorem

1.2:

For any vectors u.v.w G R" and any scalar


(i)

fc

R:

{u

(ii)

+ v)"W {ku) V =
'

=^

u-w +

vw

(iii)

k{u v)

(iv)

= vu u-u^O, and

wv

wm =

iff

u-d

Remark:

The space R" with the above operations of vector addition, and dot product is usually called Euclidean n-space.

scalar multiplication

NORM AND DISTANCE

IN R

Let u and v be vectors in R": u = (uuUz, .. .,Vm) and v tance between the points m and v, written d{u, v), is defined by
d(U,V)

(vi,V2,

.,Vn).

The

dis-

\/(l

- '^i? + {U2-V2)^+

+(Un- Vn)'''

VECTORS IN
The norm
(or:

K"

AND
is

[CHAP.

length) of the vector u, written


.

||m||,

defined to be the nonnegative square


root ot U'u:
\\u\\

y/u'U

= yul + ul+

+ul
Observe that

By Theorem

1.2,

wu^O
Let

and so the square root exists.


d{u,v)

\\u

v\\

Example

1.6:

m = (1,-2, 4,1) and v =


d(u,v)

(3, 1,

-5,

0).

Then
(4

V(l
\\v\\

- 3)2 +

(-2

- 1)2 +
+

5)2

(1

- 0)2 = ^95

V32

12

(-5)2

02

= V35
in the plane R2, then
(&

Now
That

if

we

consider two points, say p


||p|l

(a, b)

and q

(c,

d)

Vo^TF

and

d{p,q)

= V(a - c)" +

- <i)'

corresponds to the usual Euclidean length of the arrow from the origin to the point p, and d{p, q) corresponds to the usual Euclidean distance between the points p and q, as shown below:
is, ||p||

(a, b)

1~
i-d|
9
1

(a, 6)

(c,

d)

e\

A similar result holds for points on the line R and in space R*. A vector e is called a unit vector if its norm is 1: Remark:
any nonzero vector m
direction as u.

||e||

R", the vector eu

u/\\u\\

is

= 1. Observe that, for a unit vector in the same

We now
Theorem
1.3

state a

fundamental relationship known as the Cauchy-Schwarz inequality.

(Cauchy-Schwarz):
inequality,

For any vectors u,v G

R",

\u-v\^

\\u\\ \\v\\.

vectors u,v

Using the above by

GW

we can now
cos
6

define the angle 6

between any two nonzero

^ U

Note that

if

u-v =

0,

then

90

(or:

ir/2).

This then agrees with our previous

definition of orthogonality.

COMPLEX NUMBERS
complex numbers is denoted by C. Formally, a complex number is an ordered pair (a, b) of real numbers; equality, addition and multiplication of complex num-

The

set of

bers are defined as follows:


(a, b)

=
+

(c,

d)

iff

a-

and

{a,b)

{c,d)

(a, b)(c, d)

+ c,b + d) = {ac - bd, ad + be)


{a

CHAP.

1]

VECTORS IN
number a with

AND

C"

We

identify the real

the complex

number

(a, 0):

<->

(a, 0)

This is possible since the operations of addition and multiplication of real numbers are preserved under the correspondence:
(a, 0)

(b, 0)

(a

+ b,

0)

and
(a, 0)

(a, 0)(6, 0)

{ab, 0)

Thus we view

as a subset of
(0, 1),

C and

replace
i,

by o whenever convenient and

possible.

The complex number


4-2

denoted by

has the important property that

= a =
(a, 6)

(0, 1)(0, 1)

(-1, 0)

= -1
(0,6)

or

= \/=l

Furthermore, using the fact

(a, 0)
(a, 6)

(0,b)
(a, 0)

and

=
a

(b, 0)(0, 1)

we have

(6, 0)(0, 1)

bi

The notation a + bi is more convenient than (a, b). For example, the sum and product of complex numbers can be obtained by simply using the commutative and distributive laws
9.71(1
7,^

1*

{a (a

+ bi) +

(c

+ di) ac

bi

di

{a

+ c) +

{b

+ d)i

+ bi){c + di) =

bci

+ adi +
z

bdi^

{ac

- bd) +
is

{be

+ ad)i

The conjugate

of the complex

number
z

=
a

(a,b)

= a + bi

denoted and defined by

0,

bi

(Notice that zz
z are given

a^

+ bK) =

If,

in addition,

#
+

then the inverse z-^ of z and division by


.

by
z_

a
a^

Z-1

zz

b^

+
.

b
a'
b^

and

w = wz
(-z)

where

w GC. We
Example
1.7:

also define

z = Iz
Suppose
z

and
and
(2

w-z = w+
5-2i.
(5

= 2
z

+ Si =
2

w=

Then
2
15t

+w =
(2

+ 3i) +

- 2t) =
10

3i

2t

=
16
5

zw
z

+ 3i)(5 - 2i) =
2
3i

4i

6t2

=
=

+ Hi
+
2i

+ Si = 5 2i 2 + 3t

and

w =

2t

(5 - 2i)(2 - 3i) _ 4-19t _ 13 (2 + 3i)(2-3t)

i.
13

31 13^

Just as the real numbers can be represented by the points on a line, the complex numbers can be represented by the points in the plane. Specifically, we let the point (a, b) in the plane represent the complex number z = a + bi,

The i.e. whose real part is a and whose imaginary part is b. absolute value of z, written |z|, is defined as the distance from z to the origin:
\z\

V^T&^
norm
of the vector
and
9
(a, 6).

Note that

|z|

is

equal to the
1.8:

Also,

\z

ZZ.

Example

Suppose

z-2 + 3i
1^1

= V4 +

w = 12 - 5i. Then = v'iS and |wl =

V144 + 25

13

VECTORS IN

AND

C"

[CHAP.

Remark:

In Appendix B we define the algebraic structure called a field. We emphasize that the set C of complex numbers with the above operations of addition and multiplication is a field.

VECTORS IN

C"

The set of all n-tuples of complex numbers, denoted by C", is called complex n-space. Just as in the real case, the elements of C" are called points or vectors, the elements of C are called scalars, and vector addition in C" and scalar multiplication on C" are given by
(Zl, Z2,
. .

.,

Zn)

(Wl, W2,
.

.,

Wn)

(^^l

+ Wi, Z2 + Wi,
. .

.,

+ W)

Z(2l, 22,

.,Zn)

{ZZi, 222,

. ,

ZZn)

where

Zi,

wi, z

C.
(2

Example

1.9:

+ 3i, 4-i, 3) + (3 -2i, 5i 4 - 6i) = = (-6 + 4i, 2 + 8i, 6i) 2i(2 + 3i, 4 - i, 3)
vectors in C":

(5

t,

+ 4i,

- 6t)

Now
The

let

u and v be arbitrary U
(2i, 22,
.
. .

Zn),

V
is

{Wi, Wi,

Wn),

2;,

Wi

G C

dot, or inner,

product of u and v

defined as follows:

U'V =
Note that
Wi
is real.

ZiWl

+ Z%W% +

ZnWn
Wi

this definition reduces to the previous one in the real case, since

Wi

when

The norm
||m||

of

is

defined
\/ZiZi

by
2222

= yJU'U =

22

V'|2ip

|22p

|2|2

Observe that

wu and so
1.10:

||||

are real and positive


(2

when u=0, and

when u =

0.

Example

Let

m =

+ 3i,

= {S -2i, 5, 4- 61). Then u-v = (2 + 3i)(3 - 2i) + (4 - iXS) + (2i)(4 - 6i) = (2 + 3i)(3 + 2t) + (4 - 1)(5) + (2i)(4 + 6t) = 13i + 20 - 5t - 12 + 8i = 8 + 16i
4-i,
2i)

and

u'u =

+ 3t)(2 + 3i) + (4 - i)(4 -i) + = (2 + 3i)(2 - 3i) + (4 - i)(4 + i) + = 13 + 17 + 4 = 34


(2

(2i)(2t)

(2i)(-2i)

||m||

- Vu'u =

\/34

The space C" with the above operations


product,
is called

of vector addition, scalar multiplication and dot

complex Euclidean n-space.


it

Remark:

were defined by u-v = ziWi + + ZnWn, then U'U-0 even though u-0, e.g. if u={l,i,0). In fact,
If

wv

is

possible for

w%

may

not even

be

real.

CHAP.

1]

VECTORS IN

AND

Solved Problems

VECTORS IN
1.1.

R"

Compute: (i) (3,-4,5) (iv) -(-6,7,-8).


(i)

(1,1,-2);
-4,

(ii)

(1,2,-3)

+
=

(4,-5);

(iii)

-3(4,-5,-6);
(4,

Add
The

corresponding components:

(3,

5)

(1, 1,

-2)

(3

+ 1,-4 + 1,5-2) =

-3,

3).

(ii) (iii)

stim is not defined since the vectors have different

numbers of components.

Multiply each component by the scalar:

3(4, 5, 6)

(12, 15, 18).


8).

(iv)

Multiply each component by 1:

(6, 7, 8)

(6,

7,

1.2.

Let

u^ (2, -7, 1),

(-3, 0,

4),

w=

(0, 5,

-8).

Find

(i)

3% - 4v,

(ii)

2u + Zv- 5w.

First perform the scalar multiplication and then the vector addition.
(i)

(ii)

3u-4v = 3(2, -7, 1) - 4(-3, 0, 4) = (6, -21, 3) + (12, 0, -16) = 2u + 3v-5w = 2(2, -7, 1) + 3(-3, 0, 4) - 5(0, 5, -8) = (4, -14, 2) + (-9, 0, 12) + (0, -25, 40) = (4 - 9 + 0, -14 + - 25, 2 + 12 + 40) = (-5, -39,
if
{x, 3)

(18,

-21, -13)

54)

1.3.

Find x and y

{2,x

+ y).
X

Since the two vectors are equal, the corresponding components are equal to each other:

2,

+
1.

Substitute x

2 into the second equation to obtain y

Thus x

and

j/

1.

1.4.

Find x and y

if

(4, y)

x(2, 3).
(4,

Multiply by the scalar x to obtain

y)

x{2, 3)

(2x, Zx).

Set the corresponding components equal to each other:


Solve the linear equations for x and
y:

2x,

3a;.

and y

6.

1.5.

Find

x,

y and

z if

(2,

-3, 4)

x{l, 1, 1)

+ y{l, 1, 0) + z(l, 0, 0).


+ j/(l, 1, 0) + (1, 0, + (y, y, 0) + (z, 0, 0)
0)

First multiply by the scalars x, y and z and then add:

(2,-3,4)

= = = =

a;(l, 1, 1)

{X, X, x)

(x-^y

+ z,x-\-y,x)
X
-\-

Now
To
or

set the

corresponding components equal to each other:


a;

j/

2,

3,

a;

solve the system of equations, substitute


3/

7. Then substitute into the

first

k = 4 into the second equation to obtain 4 + j/ = equation to find z = 5. Thus x 4, y = 7, z = 5.

1.6.

Prove Theorem
(i)

1.1:

For any vectors u,v,w


{v

GW
(v)
(vi)
(vii)

and any scalars


k{u

fc,

fc'SR,

(ii)

(iii)

(iv)

+ v) + w u-\ u u+ u + {-u) u+V = V +u


{u
ith

+ w)

(viii)

+ v) ku + kv (fc + k')u ku + k'u (kk')u - k{k'u) lu = u

Let

Wj, Vj

and Wj be the

components of u, v and w, respectively.

VECTORS IN
(i)

R"

AND

[CHAP.

definition, Mj + Vi is the ith component oi u + v and so (itj + Vj) + Wj is the tth component + (Vj + Wj) of (u + v) + w. On the other hand, Vi + Wj is the ith component oi v + w and so But Mj, Vj and Wj are real numbers for which the asis the tth component of u + (v + w). sociative law holds, that is,

By

(Ui

Accordingly,
(ii)

(m

+
.

1))

+ Vi) + Wi - Mj-l-(Di + Wj) + tf = u+ {v-^w) since their


0);

for

i-\,...,n

corresponding components are equal.

Here,

(0, 0,

.,

hence
(Ml, M2.

= =
. .

Mn)
.

+
.

(0, 0,
. ,

.,

0)
(Ml, M2,
. .

(Ml + 0,

Mg

+ 0,

M
. .

+ 0) =
. ,

M)

= M

(iii)

Since

m = 1(mi, M2,
M + (m)

M2, = (Ml, Mg, M) + = (Ml - Ml, M2 - M2,


.

m)

( Mi,
.
.

m), M2, M) - M) = (0, 0,


.
.

. ,

(Ml,
.

. ,

. ,

. ,

0)

(iv)

By

definition, n^
Mj

But

and

ijj

+ v^ is the ith component of u + v, and Vj + Mj is the ith. component of v are real numbers for which the commutative law holds, that is,
Wi

+ u.

ft

"(

Mi,

1,

Hence m
(v)

+v =

1;

+m

since their corresponding

components are equal.

fcMj

+ Vj is the ith component of u + v, k(Ui + Vi) is the ith component of k(u + v). Since and kvi are the ith components of ku and kv respectively, Ajm; + fcvj is the ith component of ku + kv. But k, Mj and v^ are real numbers; hence n i = 1, fc(Mj + Vi) = ftMj + fc^j,
Since Mj
. .
.

Thus k{u + v)
(vi)

ku

kv, as corresponding components are equal.

first plus sign refers to the addition of the two scalars k and k' whereas the second plus sign refers to the vector addition of the two vectors ku and k'u. By definition, (fe + fc')Mj is the ith component of the vector (k + k')u. Since fcMj and Aj'Mj is the ith component of ku + k'u. are the ith components of ku and k'u respectively, kUf +

Observe that the

k%
=

But

k, k'

and

Mj are real

numbers; hence
(fc

+ k')Ui

kUi

k'Ui,

1,

Thus
(vii)

(k

+ k')u =

ku

k'u,

as corresponding components are equal.


fe(fe'M).

Since
ith

fc'Mj is

component of

the ith component of k'u, k(k'u^ is the ith component of (kk')u and, since fc, k' and Mj are real numbers,
(fcfc')Mj

But

(fcfc')Mi is

the

fc(fc'Mj),

i=l,

...,n

Hence
(viii)

(kk')u

=
.

k(k'u),

as corresponding components are equal.

1(mi, M2,

M)

(1mi, lu^, .... 1m)

(mi, M2,

m)

u.

1.7.

Show that Ou =
a vector.
Method Method
1:

for any vector u, where clearly the

first

is

a scalar and the second

Om

0(mi, M2,

m)

(Omi, OM2,

Om)

(0, 0,

...,0)

2:

By Theorem
Adding

1.1,

Om

(0

+ 0)m =

Om

Om

Om

to both sides gives us the required result.

DOT PRODUCT
1.8.

Compute u v where:

(i)

u=
(4,

(iii)
(i)

tt

(3,-5,2,l),

v-

-3, 6), v 1,-2, 5).


(2,

(8, 2,

-3);

(ii)

u=
(3)

(1,

-8,

0, 5),

=
8.

(3, 6, 4);

Multiply corresponding components and add:

wv

+
1

(3)

=
+

(ii)
(iii)

The dot product

is

not defined between vectors with different

numbers of components.
(5)

Multiply corresponding components and add:

u-v =

(2)

8.

CHAP.

1]

VECTORS IN

R"

AND

C"

1.9.

Determine k so that the vectors u and v are orthogonal where u ^ (1, k, -3) and v = (2, -5, 4) (i)
(ii)

u =

(2, 3fc,

-4,

1, 5)

and v
v, set it

(6>

-1,
0,

3, 7, 2fc)

In each case, compute u


(i)

equal to

and solve for

k.
0,

(ii)

= 1 2 + 2 5fe (-5) + = 2-6 + 3fc'(-l) + (-4)'3 + 1*7 + U'V = 12 - 3fe - 12 + 7 + lO/c = 0,


u'v

fc

(-3) -4

12

-5k -

10

==

0,

fc

= -2

5'2fc

= -l

1.10.

Prove Theorem
(i)

1.2:

For any vectors u,v,w G R" and any scalar

kGK,
iff

(ii)

+ v)'W = U'W + (fcM)-^ = k{u'v)


{u

vw

(iii)

wv
M'M

i;

tt

(iv)

0,

and u-u

u =

Let
(i)

M =

Since

W). = (yi.'y2. '"n). W = (^1,^2. u + v = (mi + Vi, M2 + "2. .** + I'm). + (U + Vn)Wn (u + v)'W = (Ml + Vi)Wi + (% + '"2)^2 + = UiWi + ViWi + U2W2 + 1'2M'2 + + MW + VW = (MiWi + M2W2 + Mw) + (viWi + V2W2 + + yw) + = U'W + VW

(Mi,M2

O.

"

(ii)

Since

ku

(ku^, ku^,

.,

ku^),
fcM2'y2

(ku)

ku^Vi

ku^V^

HU^V^

+ M2'y2
1>n^n

-I

1-

'^n^n)

*=(* '

")

(iii)

U'V =

MjDi

M2''^2

Mn''^n

=
i,

'"l"!

''2'*2

'

"

'

= V'U
is

(iv)

Since wf is nonnegative for each negative,

and since the sum of nonnegative real numbers


2
I

non-

J. ,.2

=-

n
0.

Furthermore,

u'U =

iff

Mj

for each

i,

that

is, iff

DISTANCE AND NORM IN


1.11.

R
(i)
?;

Find the distance d{u, v) between the vectors u and v where: (iii) m = (5,3,-2,-4,-1), t; = (6,2,-1); (ii) =(3,-5,4),
In each case use the formula
(i)

u=

=
.

v = (6, -5); (2,-1,0,-7,2).


(1, 7),

d(u, v)

VW 144

v{)^

+ +( - vj^
=
13 25

d(u,v)
d(u,v) d(u,v)

= =

V(l

- 6)2 + - 6)2 +

(7

+ 5)2 = V25 + - 2)2 +


(4

\/l69

(ii)

V(3
V(5

(-5
(3

+ 1)2 = V9 +

49

a/83

(iii)

- 2)2 +

+ 1)2 +

(-2

+ 0)2 +

(-4

+ 7)2 +

(-1

- 2)2 =

\/47

1.12.

Find k such that

Now

solve

fc2

= (2, fc, 1, -4) and v = (3, -1, 6, -3). (d(u, i;))2 = (2 - 3)2 + (fe + 1)2 + (1 - 6)2 + (-4 + 3)2 = fe2 + 2fe + 28 = 2, -4. + 2fc + 28 = 62 to obtain
d{u, v)

6 where

fc

1.13.

Find the norm

||m||

of the vector

if

(i)

u= +
m2

(2,
4.

-7),
.
. .

(ii)

u=

(3,

-12, -4).

In each case use the formula


(i)

||m|1

y/u^

^2

IHI
11^11

= =

V22 V32

+
+

(-7)2

= V4 +

49

= ^53
144

(ii)

(-12)2

(_4)2

= V9 +

16

= V169 =

13

10

VECTORS IN
Determine & such that

R"

AND

[CHAP.

1.14.

||tt||

= VS^ where u =
=
12

{l,k,-2,5).
52

I|m|I2

fc2

+ =

(-2)2
3,

A;2

30

Now

solve

fc2

30

39

and obtain

fc

-3.

1.15.

Show

that

||m||

^ 0,

and

||m||

ifi

u=

0.

By Theorem

1.2,

wu

O,

and u'u

iff

0.

Since

||m||

yjii-u,

the result follows.

1.16.

Prove Theorem

1.3

(Cauchy-Schwarz):

For any vectors u

{u\,

m)

and v

(vi,

.,Vn)

in B",
n

\u-v\

]\u\\ \\v\\

We

shall prove the following stronger statement:

\u'v\

i.e.

|Mt'"tl

I|m||

HvH.

If M = or V = 0, then the inequality reduces to need only consider the case in which m # and v Furthermore,

j^
0,

and is therefore true. Hence we where ||m|| # and ||i;|| # 0.


\uvj

\U'V\

IMjI)!

+
G

MV|

\UiVi\

2 kiVil

Thus we need only prove the second

inequality.

Now

for any real

numbers

w,

3/

R,

2xy

(x

j/)2 =
a;2

x^

2xy

y^

or, equivalently,

3/2

(1)

Set X

|mj|/||m||

and y

Ifil/HvH

in (1) to obtain, for

any

i,

IHI IHI
But, by definition of the

IMP

IWP

^^'

norm
i

summing

{2)

with respect to

of a vector, ||m|| = 2m,^ = kiP and and using \u(Vi\ = ImjI |i;j|, we have

||i;||

= S^f =

2|v,-|2.

Thus

2M
IMI IHI
that
is,

2kP
IMP

2
2
II

It'iP

IMI^

IMP
IblP

IMP
ki^il
,1 ,1
11

IMP
1

IMI IHI
Multiplying both sides by
||m||

H'wH,

we

obtain the required inequality.

1.17.

Prove Minkowski's inequality:

For any vectors u-{ui,...,Un) and v =


If
IIm

{vi,

.,Vn)

in R",

||m

+ vH

=^

||tt||

+
+

||v||.

+ vjI =
JMj

0,

the inequality clearly holds.


jwjl

Thus we need only consider the case


mj, Vj

||m

i;1|

#0.

Now

+ V(|

|i)j|

for any real

numbers
i'<)='

R.
i'iP

Hence

\\u

v\\2

=
= =

2(i +

= + Vjj +

2k +
^

2 ki 2 ki +
(see

vil

\ui

2
Vil

ki

+ Vil

(kil

+ M)

Vil \ui\

ki +

Ivjj

But by the Cauchy-Schwarz inequality

preceding problem),

2M+fj|KI ^
Thus
Dividing by
||M

Ik+^IIIMI
|Im

and

2k +
||i;|l

'yilkl
||tt

^
(IMI

Ik

+ HIIHI
lbll)

+ f||2 ^
we

i;||

IHI

||m

+ H|

i;||

||m-|- v||,

obtain the required inequality.

CHAP.

1]

VECTORS IN
in

AND

11

1.18.

Prove that the norm


[Ni]:
[Nz]: [Na]:

R"

satisfies the

following laws:

For any vector w,

||m||^0;

and |H|
A;,

=
\\ku\\
\\u\\

iff

u=
|/cl

0.

For any vector M and any scalar


For any vectors u and
was proved
in

v,

\\u

v\\

^
=

= +

||m||.

||t;||.

[Ni]

Problem

1.15,

and

[Ng] in

Problem

1.17.

Hence we need only prove that

[Ni] holds.

Suppose u

(ui, ii2,

.,

u)

and so ku
(kui)^

(kui, ku^, ....

ku^. Then
khi\

||fcMl|2

(fcMi)2

(fcM)2

khi\

khil

The square

root of both sides of the equality gives us the required result.

COMPLEX NUMBERS
1.19.

Simplify:
(vi)

(i)

(5

+ 3i)(2-7i);

(ii)

(4-3*^;

("i)

g"^''

(i^)

|^;

(v)

',

i*.

,-31. *"

(l

+ 2i)''; (vii)(2^)'
31

(i)

(ii)

+ 3t)(2 - 7i) = 10 + 6i - 35i - 21i? = (4 - 3t)2 = 16 - 24t + 9t2 = 7 - 24t


(5

29i

r"\
^*"^

^'^'

(v)
(vi)

(3 + 4t) _ 3 + 4i ^ A + Aj - 4i)(3 + 4t) 25 25 25 - 7t)(5 - 3t) _ -11 - 41i _ _11_41. 2-7i _ (2 34 34 34 5 + 3t ^ (5 + 3i)(5-3t) ts = i^'i = (-l)t = -i; P^ = (i*)7't^ = i* = v^"P = 1; - 12 - 8i = -11 - 2i (1 + 2i)8 = 1 + 6i + 121* + 8i^ = 1 + 6i

4i

(3

1^

(-t)

-i

(vu)

\^2-3iJ

1 _ _ ~ -5-12i ~

(-5

(-5 + 12t) _ -5 + 12i _ __5_ - 12t)(-5 + 12t) 169 169

12

'

169*

1.20.

Let
(i)

= 2 - 3i and w = 4 + 5i Find: z + w and zw; (ii) z/w; (iii) and w;


2!

(iv)

\z\

and

\w\.

+ w = 2 - 3i + 4 + 5i = 6 + 2i zw = (2 - 3i)(4 + 5t) = 8 - 12i + lOi - 15t2 = 23 - 2i - 5i) _ -7 - 22i ^ _ 1. _ 22 f\ . - 2-3t _ (2 3i)(4 41 41 41 (4 + 5i)(4 - 5t) 4 + 6i w = 2 - 3t = 2 + 3t; w = 4 + 5i = 4 - 5t. (iii) Use a+U = a-bi:
(i)

(iv)

Use |a+6i|

= V^^TP:

\z\

|2

- 3t| = V4 + 9 =

Vl3;

\w\

[4

+ 5i|

=:

Vl6 + 25= Vil.

1.21.

Prove:
(i)

For any complex numbers z,w G.C,


(ii)

(i)

zw = zw, (iii) z = z. Suppose z = a+bi and w = e + di where a, b,c,d& R. z + w = (a+bi) + (e + di} = {a+c) + {b + d)i = {a + e) {b + d)i = a + c bi di = (a bi) + {e di) = z + w

+ w = z + w,

(ii)

zw

= (a+ bi)(c + di) = (ae bd) + (ad+ bc)i = {ac bd) (ad + bc)i = (a-bi)(c di) - zw
a

(iii)

a+bi

bi

= a

(b)i

= a+bi =

12

VECTORS IN
Prove:

AND

C"

[CHAP.

1.22.

For any complex numbers z,w GC,

\zw\

=
R.

\z\

\w\.

Suppose

= a + bi and w = c + di where ||2 = o2 + 62, |w|2 = c2 + d2,


z |zw|2

a,

b,c,dG
and

Then
{ac

zw =

bd)

{ad

bc)i

Thus

= (ac - 6d)2 + (ad + 6c)2 = a2c2 - 2abcd + b^d? + aU^ + 2abcd + 62c2 = a^C^ + d2) + 62(c2 + d2) = ((j2 + 62)(c2 + (2) =

|2|2

|^|2

The square root

of both sides gives us the desired result.

1.23.

Prove:

For any complex numbers z,wlC,


z

\z

w\
R.

\z\

\w\.

Suppose
V

(c,

d)

in R2.

a + bi and Note that


\z\

w = c + di
62

where

a, 6, c,

Consider the vectors

u=

(a, 6)

and

= Va2 +
(b

||m||,

jwl

Vc2

rf2

=
c,

|j^||

and

\z

w\

\{a

+ c) +

+ d)i\ = V( + c)2 +
||m

(6

+ d)2 =
||m||

||(a+

d)||

\\u

v\\

By Minkowski's

inequality (Problem 1.17),


\z

+ v|| ^
||m||

\\v\\

and
|z|

so
lw|

w\

\\u

v\\

||t;||

VECTORS IN
1.24.

C"
(3
2i, 4i, 1
(il)

Let M
(i)

=
v,

u+

+ 6i) and v = 4m, (iii) (1 + i)v,

+ i,2-3i, 5). (iv) (1 - 2i)u +


{5
(8

Find:
(3

+ i)v.

(i)

Add corresponding components: u + v

t,
4t:

+ i, 6 + 6t).
(8

(ii)

Multiply each component of u by the scalar Multiply each component of v by the scalar 1
(1

4m =
i:

12i,

16, 24

+ 4t).

(iii)

+ t)i;

(5

+ 6i + i2,

+ - 1 - 3i2, 6 + 5t) = 2

i
(4

+ 6i,

5 -

i,

+ 5i)

(iv)

First perform the scalar multiplication and then the vector addition:
(1

- 2i)u +

(3

+ i)v = =

(-1

- 8i 8 + 4i, 13 + 4i) + (13, 17 - 3i, 28 + 9t)

(14

8i,

- Ii,

15

+ 5i)

1.25.

Find u-v and


(3 - 2i,
4i, 1

vu
v

where:
(5

+ 6i),

u = {l 2i,S + i), v = + i,2-Si,l + 2i).


(i)

(4

+ 2i,

6i);

(ii)

Recall that the conjugates of the second vector appear in the dot product:
(2l,
. .

.,

Z)

(Wi,

...,WJ = iWi

2W

(i)

M v =

- 2t)(4 + 2i) + - 2i)(4 - 2i) + = (1


(1

+ t)(5 - 6t) (3 + j)(5 + 6t) =


(3

-lOi

23i

13i

vu
(ii)

= =

+ 2t)(l - 2i) + (4 + 2t)(l + 2i) +


(4 (3 (3

- 6i)(3 + 1) (5 - 6z)(3 - i) =
(5

lOi

23i

13i

u-v =

=
ij

- 2i)(5 + i) + - 2i)(5 - i) +

(4i)(2 (4i)(2

- 3i) + + 3i) +

(1 (1

+ 6i)(7 + 2i) + 6i)(7 - 2i) =

20

35i

tt

= =

In both

+ i)(3 - 2i) + (2 - 3t)(4i) + (7 + 2i)(l + 6t) + i)(3 + 2i) + (2 - 3i)(-4t) + (7 + 2i)(l - 6i) = 20 - 35i (5 examples, vu wv. This holds true in general, as seen
(5

in

Problem

1.27.

CHAP.

1]

VECTORS IN
where:
(i)

AND

13

1.26.

Find

||tt||

m=

(3

+ 4i,

- 2i,

- 3i);

(ii)

u={4s

i,

2i,

+ 2t, 1 - hi).
z)

Recall that zz
||m|P
(i)

w''+ 6^ when z = a+ hi. Use - u-u = ZiZi + Z2Z2 + + zZn


where
64, 60,

=
||m||

(z^, Z2

||m1P
||m||2

(3)2

(4)2

+
+

(5)2

(-2)2

+ +

(1)2

+ +

(-3)2

or

==

(ii)

42

(-1)2

22

32

22

12

(-5)2

or

||m||

= ^60 =

2\/l5

1.27.

Prove: For any vectors u,v EC" and any scalar z GC, (i) z{u'v), (iii) u-{zv) = z{u'v). (Compare with Theorem 1.2.)
Suppose u
(i)

wv = vu,

(ii)

(zu)

{zi, 2,

and v

(wj, W2,

w).
in

Using the properties of the conjugate established

Problem

1.21,

VU
(ii)

WiZi WiZ^
.
.

+ +

W2Z2 W2Z2

+ + +
.

+ +
+

WZ
WZ

= = =

WiZi Z^Wi

+ +

W2Z2 Z2W2

+ +

+ +

WZn zw

WV
^(m
* '")

Since

zu

(zz^, zz2,

.,

zz^),

(zu)
(iii)

ZZiWi

ZZ2W2

ZZWn

Z(ZiWi

+ 2M'2 +

"

"

"

+ ^nWn) =

Method

1.

Since zv

(zwi, zwg,

zwj,

(zv)

= =
(i)

ZiZWl

+ Z^Wl + (iWi + Z^2 +


(ii),

+ ZZW^ = ZiZWi + + Z^W^ = Z(U V)

Z2ZW2

2W

Method

2.

Using

and

(zv)

(zv)

= z(vu) = z(vu) =

z(u'v)

MISCELLANEOUS PROBLEMS
1.28.

Let u
(i)

=
v;

(3,

-2,

1, 4)

and v

(7, 1,

-3,

6).

Find:
(v)
||m||

u+

(ii)

4u;

(iii)

2u

Sv;

(iv)

u-v;

and

||i;||;

(vi)

d{u,v).

(i)

u + v
4m 2m

(3

+ 7,-2 + 1,1-3,4 + 6) = =

(10,-1,-2,10)

(ii)

(4'3, 4'(-2), 4-1, 4-4)


3i;

(12,-8,4,16)
9,

(iii)

(6,

-4,

2, 8)

(-21, -3,

-18)

(-15, -7, 11, -10)

(iv)

u-v =
1|m|1

21-2-3 +
4

24
16

= =

40

(v)

= V9 +

VSO,
(1

||v||

V'49
(4

36

\/95

(vi)

d(u,v)

V(3

- 7)2 +

(-2

- 1)2 +

+ 3)2 +

- 6)2 = V45 =

3\/5

1.29.

Let M
(i)

(7

- 2i,

+ 5i) and
(iii)

(1

+ i, -3 - 6i).
u-v;
(8

Find:
||mI|

u + v;
u + v

(ii)

2m;
(7

{S-i)v;
i,

(iv)

(v)

and

||v||.

(i)

= =

- 2i + 1 +

+ 5i - 3 - 6z) =
(4

i,

-1

- i)

(ii)

2m =
(3-i)v

(14i

- 4i2, 4t + 10t2) =
(3

+ 14i -10 + 4t)


(4

(iii)

+ 3i - i - 12, -9 - ISi + 3i + 6P) (2 (2

+ 2t, -15 - 15t)


36

(iv)

u-v =

(7 (7

=
(v)

- 2t)(TTt) + - 2t)(l - 1) +
+
(-2)2

+ 5i)(-3 - 6t) + 5i)(-3 + 6i) =


52

9i

3i

= -31 (-3)2

12i

IMI

V72

22

= V^,

\\v\\

Vl'^

1^

(-6)2

= V47

14

VECTORS IN

AND

C"

[CHAP.

1.30.

Any
PQ.

pair of points P = {ou) and Q fines the directed line segment from

We identify PQ with the PQ = V = {bi ai, 62 - a2,


identified

= (bi) in R deP to Q, written vector v = Q-P:


. .

.,bn- a)
where:

Find the vector v


(i)

with

PQ

P=
t)

(2,5),

Q=

(-3.4)

(ii)

P=(l,-2,4), Q = (6,0,-3)
V

(i)

(ii)

= =

Q-P Q-P

= (-3-2, 4-5) = (-5,-1) = (6 - 1, + 2, -3 - 4) = (6,


in

2,

-7)

1.31.

The
Xi,
.

set
.

H of elements
form

R" which are solutions of a linear equation in n unknowns


CiXl

.,Xn of the

C2X2

CnXn

&

(*)

with u
tion of

=
H.

(ci,

.,

c)

in R", is called a

(We frequently

identify

H
is
(fej,

kyperplane of R", and (*) is called an equawith (*).) Show that the directed line segment

PQ of any pair of points P,Q


u
is

GH
Q =

orthogonal to the coefficient vector u; the vector


.

said to be normal to the hyperplane H.


Suppose

P=

(!,

.,aj
C2a2

and

.,6).

Then the
c^bi

Oj

and the

64

are solutions of the


6

given equation:
Cjai

c^an

b,

C262

c&

Let

Then

t;

= PQ = = Ci(&i - aj) + 62(63 - Og) + = C161 - citti + C262 - C2a2 + = (Ci6i + C262 + + cbj v

Q-P^

{b,~ai,b2-a^, ...,b-a)

- aj + cfe - ca (CjOi + 02(12 + + co) = 6-6 =


+

c(6

Hence

v,

that

is,

PQ,

is

orthogonal to

u.

1.32.

Find an equation of the hyperplane H in R* if: (i) H passes through P = (3, -2, 1, -4) and is normal to m = (2,5,-6,-2); (ii) passes through P = (1,-2, 3, 5) and is parallel to the hyperplane H' determined by 4:X 5y + 2z + w = 11.

(i)

An

equation of

is

of the form

2x

5y

Qz

2w =

k since

it is

P
(ii)

into this equation to obtain

2.

Thus an equation of

is

normal to u. Substitute 2x + 5y 6z 2w = 2.

and H' are parallel iff corresponding normal vectors are in the same or opposite direction. Hence an equation of H is of the form 4x 5y + 2z + w = k. Substituting P into this equation, we find k = 25. Thus an equation of H is 4:X 5y + 2z + w = 25.

1,33.

The
and

line

in

R" passing through the point

P=

(a,)

in the direction of
t

u=

(Ui) =
is,

consists of the consists

points points

X P + tu, GK, that X= obtained from


(Xi)

of the

Xi
a;2

ai

-I-

Uit U2t

^2

-f-

(*)
n

an

Unt

where

takes on
I.

all

real values.
(*) is called

The

variable

is

called a parameter,

and

a parametric rep-

resentation of

CHAP.

1]

VECTORS IN

B,"

AND
line

C"

15

(i)

tion of

Find a parametric representation of the tt where: (a)P = (2,5) and m

passing through
(6)

and

in the direc1)

(-3,4);

P=

(4,

-2, 3,

and u

= Q

(2,5,-7,11).
(ii)

Find a parametric representation of the line passing through the points P and where: (a)P = (7,-2) and Q = (9,3); (6) P = (5, 4, -3) and Q = (l,-3,2).
In each case use the formula
(*).

(i)

'x

\y

= =

2
5

3t 4t

= 4 + 2t y = -2+ 5t z = 3 - 7t w = 1 + nt
X
line

(In

equation:
(ii)

K2 we usually eliminate Ax + Zy = 23.)

from the two equations and represent the

by a single

First compute

u =
(a)

PQ = Q

P.

Then use the formula


(2,5)
2t 5
(b)

(*).

u =
(x

Q-P

= = 1+ = -2 +
we

Q-P
X V
z

= (-4,-7,5)

= 5 -At = 4 -It = -3 + 5t
Q.)

(Note that in each case

could also write

QP P

Supplementary Problems
VECTORS IN R 1.34. Let u = (1,-2,5),
and
1.35.
||t;||;
i;

(3,l,-2).

Find:

(i)

v;

(ii)

-6m;

(iii)

2u

5v;

(iv)

u-v;

(v)

||m|1

(vi)

d(u,v).

Let
(iii)

w=

(2,

u +

-1, 0, -3), i; 2v 2w, (iv)

= (1, -1, -i; 3), w = u'v,u'W and vw;


=
(5,

(1,3,-2,2). Find: (i) 2u (v) d(u,v) and d(v,w).

3v;

(ii)

5u

3v

Aw;

1.36.

Let

M = (2,1,-3,0,4),
ll-ull;

t;

-3, -1,

2, 7).

Find:

(i)

u+

v;

(ii)

3u

2v;

(iii)

M'^y;

(iv)

||mI|

and
1.37.

(v)

d(u,v).

(5,

Determine & so that the vectors M and A;, -4, 2), i; = (1, -3, 2, 2fe). (iii) m Determine

1)

are orthogonal,
(1, 7,
fe

(i)

u = (3,k,2),

t;

(6,

4, 3).

(ii)

+ 2, -2),
(ii)

(3,k,-S,k).

1.38.

a;

and
and

2/

if:

(i)

(x,x

+ y) = =

(y

-2,6);

x(l,2)

-4(y,3).

1.39.

Determine Determine
(i)

a;

J/

if :

(i)

x(3,2)

2(y,-l);

(ii)

x(2,y)

y(l,-2).

1.40.

a;,

y and

if:

(3,-1,2) (-1,3,3)

(ii)

a:(l, 1, 1)
a;(l, 1, 0)

+ j/(l, -1, 0) + + j/(0, 0, -1) +


eg
ej

z(l, 0, 0) z(0, 1, 1)

1.41.

Let
(i)

(!, 0,0),

62

(0,1,0),
(ii)

=
a,

(0,0,1).

aei

6e2

ceg;

62

6,

Show that m 63 = c.

for

any vector

u=

(a,b,c)

in

&:

16

VECTORS IN

AND

[CHAP.

1.42.

Generalize the result in the preceding problem as follows. elsewhere: ith coordinate and
ei

Let

ej

e R"

be the vector with 1 in the

(1,0,0, ...,0,0),

62
.

=
.

(0,1,0

0,0),

...,

(0,

0,0,

.,0, 1)

Show

that for any vector


(i)

(aj, ag,
(1262

.,),

u =

aiCi

ae,

(ii)

ej

Oj

for

1,

..,n

1.43.

Suppose M

e K" has

the property that

1)

for every

R".

Show

that

it

0.

1.44.

Using d(u,v)

= \\u-v\\ and the norm properties [ATj], [iVj] and [N3] in Problem 1.18, show that the distance function satisfies the following properties for any vectors u,v,w G R":
d(u,v)^Q,
and
d{u, v)

(i)

itl

u=

v;

(ii)

d(u,v)

d(v,u);

(iii)

d{u,w)

d{u,v)

d(v,w).

COMPLEX NUMBERS
1.45.

Simplify:

(i)

(4

- 7t)(9 + 2i);

(ii)

(3-5i)2;

(iii)

j^^;
(iv)

(iv)|-3||;

(v)

(l-i)^.

1.46.

Simplify:

(i)

^;

(ii)

f^;

(iii)

i^s,

^2^

^^*;

.'

(^3^
zw;
(iii)

1.47.

Let z Let z

= 2-5i =
2

and and

w = 7 + 3i w = 6 - 5t.
=
1;
(ii)

Find: Find:

(i)

+ w;
(ii)

(ii)

/;
|w|.

(iv)

w;

(v)

1|,

lw|

1.48.

+i
(i)

(i)

z/w;

w;

(iii)

11,

1.49.

Show

that:

-i

\z\

\z\;

(iii)

real part of

^(z

z);

(iv)

imaginary part of

(z

z)/2i.

1.50.

Show

that

zw =

implies

or

w=

0.

VECTORS IN
1.51.

Let
(iv)

It

(1

+ 7i, 2 - 6i)
(v)

and
1|m1|

v-{5- 2i, 3 - 4t).


and
||t)|l.
1;

Find:

(i)

u+

v;

(ii)

(3

+ i)u;
m

(iii)

2m +

(4

- 7t)v;

wwandvw;
M
2i,

1.52.

= (3 - 7t, tfvandvM;
Let
Prove:
(i)

-1 + i)
1|m||

(iv)

and and Hiill

(4

- i,

11

+ 2i,

- Si).

Find:

(i)

v;

(ii)

(3

i)v;

(iii)

1.53.

For any vectors u,v,w G

C":

{u

+ v)-w =

WW

+ vw;

(ii)

w
and

(u

v)

wu + wv.
iff

(Compare with Theorem

1.2.)

1.54.

Prove that the norm in C"


[Ni]:
[A^a]:

satisfies the

following laws:
\\u\\

For any vector

u,

\\u\\

^
v,

0;

=0
|H|

0.

For any vector u and any complex number

z,

1|zm||
||i'l|.

\z\

\\u\\.

[Ns]:

For any vectors u and


1.18.)

\\u

v\\

(Compare with Problem

MISCELLANEOUS PROBLEMS
1.55.

Find an equation of the hyperplane


(i)

in

R3 which:
to
(3, 1,

passes through
contains contains
(1, (1,

(2,

7,

1)

and

is

normal
(0, 2,

11);

(ii) (iii)

-2, -5,

2), (0, 1, 3)

and

-1);

2)

and

is

parallel to

3x

ly
4z

iz

5.

1.56.

2w =

Determine the value of k such that 2x - ky 8. (Two hyperplanes are perpendicular

+
iff

- 5w = 11 is perpendicular to 7x + 2y -z corresponding normal vectors are orthogonal.)

CHAP.

1]

VECTORS IN

AND

17

1.57.

Find a parametric representation of the


(i)

line which:

passes through

(7,

1,

8) in the direction of (1, 3,

5)

(ii)

passes through (1,9,-4,5) and (2,-3,0,4)


passes through (4,-1,9) and
is

(iii)

perpendicular to the plane

Sx

2y

18

1.58.

Let P,

Q and R

be the points on the line determined by


*i

Oi

Mjt,

a;2

02
ti,

+
^2

Mgt,

x^
t.

a^-\- ut

which correspond respectively d(P,Q) d(Q,R) = d(P,R).

to the values

^^^

*3

^or

Show

that if

tj

<

(2

<

Ht

then

Answers
1.34.
(i)

to

Supplementary Problems
(-6, 12, -30);
d{u,v)
(iii)

u+v
||m||

(v)

= (4, -1, 3); (ii) -6m = = V30, = a/U (vi)


||i;||
;

2m

- 5i; =

(-13, -9, 20);

(iv)

-y

-9;

= ^62
(3,

1.35.

(i)

2u-Zv = (1, 1, 3, -15); (iv) wv 6, M w = 7,

(ii)

5u-3v-4w =
6;
(v)

-14,

11,
,

-32);
d(v,

(iii)

-m + 2i; - 2w =
3'\/2

(-2,-7,

2, 5);

vw =
(ii)

d(u, v)

\/38

w)

1.36.

(i)

M+r

(7,

-2, -4, 2,
(v)

11);

3u-2v =

(-4,

9,

-7, -4, -2);

(iii)

u-v =

38;

(iv)

||m1|

= VSO

||i;||=2V22;
1.37.
(i)

d(M,i>)

V^
k

k
x
x

=
=

6;

(ii)

fc

3;

(iii)

=
j/

3/2

1.38.

(i)

2,

J/

4;

(ii)

a;

=
(ii)

-6,

3/2

1.39.

(i)

=
=

-l, y
2,

-3/2;

x
(ii)

0,

0;

or

= =
0.

-2,
4

2/

= -4

1.40.

(i)

S, g

-2;

a;

-1,

j/

1,

1.43.

We
(i)

have that m m

which implies that m

1.45.

50-55i;
-^i;
z
(ii)

(ii)

-16

30i;

(iii)

(4

+ 7t)/65;
-1;
(iv)

(iv)

(H-3i)/2;

(v)

-2 -

2i.

1.46.

(i)

(5

+ 27t)/58;
(ii)

(iii)

-i,

i,

(4

+ 3i)/50.

1.47.

(i)

(v)

|z|

+ w = 9 - 2t; = = V29,

zw = 29

29i;

(iii)

z/w

(-1

- 41t)/58;
=
lw|

(iv)

5i,

w = 7 - 3f,

V5^(ii)

1.48.

(i)

z/w

(7

+ 16t)/61;
then
\zw\

= -

i,

w =
0.

5t;

(iii)

\z\

y/5

|m;|

= Vei
so z

1.50.

If

zw

0,

\z\

\w\

|0|

Hence

\z\

or

=0; and

=
21

or

w=

0.

1.51.

(i)

(ii)

(iii)

= (6 + 5i, 5 - lOi) - 16t) (3 + t)u = (-4 + 22t, 12 - 7i)v = (-8 - 41i, -4 - 33i) 2iu + (4
M+
t)

(iv)

u-v ||m|1

21

+
,

27i,
||i;||

vu
=

27i

(v)

= 3VlO

3a/6

1.52.

(i)

u-v =
(3

(-1
(13

- ei, + i,

-11,

-9 + 4t)

(iii)

u-v ||m||

12
8,

2i,

vu
v/215

12

2i

(ii)

+ t)i; =

31

+ 17i, 27-i)
(ii)

(iv)

JHI

=
7i/

1.55.

(i)

Sx

+ y-llz =

-12;

13x

4j/

7;

(iii)

4z

46.

1.56.

k-0
(i)

1.57.

(x

(ii)

= 1+ t y = 9 - 12* z = -4 + 4t w = 5t
x

(iii)

chapter 2
Linear Equations
INTRODUCTION
The theory of linear equations plays an important and motivating role in the subject of linear algebra. In fact, many problems in linear algebra are equivalent to studying a system of linear equations, e.g. finding the kernel of a linear mapping and characterizing the subspace spanned by a set of vectors. Thus the techniques introduced in this chapter will be applicable to the more abstract treatment given later. On the other hand, some of
the results of the abstract treatment will give us crete" systems of linear equations.

new

insights into the structure of "con-

For simplicity, we assume that all equations in this chapter are over the real field R. We emphasize that the results and techniques also hold for equations over the complex field C or over any arbitrary field K.

LINEAR EQUATION
By
a linear equation over the real
aiXi
field

R,

aix^

we mean an + anXn =

expression of the form


h
{!)

where the
ttt

ai, & G R and the Xi are indeterminants (or: unknowns are called the coefficients of the Xi respectively, and b is called the constant term or simply constant of the equation. A set of values for the unknowns, say

or variables).

The

scalars

Xl
is

k\,

Xi=

ki,

.,

Xn=

kn

a solution of

(i) if

the statement obtained by substituting


ttifci

for

Xi,

a2k2

afc

then said to satisfy the equation. If there is no ambiguity about the position of the unknowns in the equation, then we denote this solution by simply thew-tuple , ,, ,
is

true.

This set of values

is

U =

(fcl,

fe,

kn)

Example

2.1

Consider the equation

+ 2y 4z + w =
is

S.

The

4-tuple

u3

(3, 2, 1, 0)

a solution of the equation since

+ 22-4'l +
However, the 4-tuple

=
v
5

or
(1, 2, 4, 5)

=
is

is

a true statement. equation since


1
is

=
3

not a solution of the


3

+ 2.2-4-4 +

or

-6 =

not a true statement.


(1)

Solutions of the equation


cases:

can be easily described and obtained.


- 0.

There are three

Case

(i):

One

of the coefficients in (1) is not zero, say ai

Then we can rewrite the

equation as follows
axx-i

= b-

a2X2

anX

or

Xi

a^^b

a^^aiXi

aj"

ax

18

CHAP.

2]

LINEAR EQUATIONS

19

By arbitrarily assigning values to the unknowns X2, .,x, we obtain a value for Xi; these values form a solution of the equation. Furthermore, every solution of the equation can be obtained in this way. Note in particular that the linear equation in one unknown,
.
.

ax
has the unique solution x
Example
2.2:

h,

with a'Q

a^^b.

Consider the equation 2x

4y

8.

We

rewrite the equation as


2a;

4j/

or

2y

^z

Any value

for y and z will yield a value for x, and the three values will be a solution of the equation. For example, let 2/ = 3 and z = 2; then x = 4 + 2'3 ^-2 = 9. In other words, the 3-tuple u = (9, 3, 2) is a solution of the equation.

Case
equation

(ii):
is

All the coefficients in (1) are zero, but the constant is not zero.

That

is,

the

of the

form
Ojci

0*2

Oa;

6,

with 6

Then the equation has no


Case
(ill):

solution.

All the coefficients in (1) are zero, the equation is of the form

and the constant


Oa;

is

also zero.

That

is,

Oxi

Oa;2

Then every n-tuple

of scalars in

is

a solution of the equation.

SYSTEM OF LINEAR EQUATIONS We now consider a system of m linear


ttiiXi

equations in the

n unknowns
&i

Xi,

.,

x:

+ +

ai2a;2

+
+

ainXn
a2nXn

a2ia;2

a22X2

+ +

=62
(*)

OmlXl

+ am2X2 +

"

OmnXn

&m

belong to the real field R. The system is said to be homogeneous if the con.,kn) of real numbers is a solution (or: 6m are all 0. An ?i-tuple u = (fci, a particular solution) if it satisfies each of the equations; the set of all such solutions is termed the solution set or the general solution.
Oij, bi
. . ,

where the
stants
61,

The system of

linear equations

anXi
aziXi

+
4+

ai2X2 ai2a;2

+
+

aia; aiXn

... n^aVa 4- a22X2

-I-

nn-fa2nXn

= ^ -

n
(**)

amlXi
is called

+ 0^2X2 +
.
.

OmnXn

tion,

the homogeneous system associated with = (0, 0, namely the zero w-tuple 0)
.

(*).

The above system always has a

solu-

called the zero or trivial solution.

Any

other solution,

if it exists, is called

a nonzero or nontrivial solution.

The fundamental

relationship between the systems (*)

and

(**) follows.

20

LINEAR EQUATIONS
2.1:

[CHAP. 2

Theorem

Suppose m suppose

is is

a particular solution of the nonhomogeneous system (*) and the general solution of the associated homogeneous system (**).

u +
is

{u

+ w: w G W)
(*).

the general solution of the nonhomogeneous system

emphasize that the above theorem is of theoretical interest and does not help us to obtain explicit solutions of the system (*). This is done by the usual method of elimination described in the next section.

We

SOLUTION OF A SYSTEM OF LINEAR EQUATIONS Consider the above system (*) of linear equations. We reduce
follows:

it

to a simpler

system as

Step

1.

Interchange equations so that the


cient in the first equation, that
is,

first

unknown
an
-0.

xi

has a nonzero

coeffi-

so that

Step

2.

For each i>

1,

apply the operation


Li
-*

anLi

ttiiLi

That

is,

tiplying the first equation Li

replace the ith linear equation Li by the equation obtained by mulby -an, multiplying the ith equation L, by

an, and then adding.

We
the

then obtain the following system which (Problem 2.13)


solution set as
(*):

is

equivalent to

(*), i.e.

has

same

anaii

ai2X2

a'laXs

+ +
^"

a'mXn
0,2nXn

= = =

&i

ayja^jg

+
"I"

&2

amJ2^i2

dmnXn

Om

= 0. Here Xj^ denotes the first unknown with a nonzero coefficient in an equation other than the first; by Step 2, Xj^ ^ Xi. This process which eliminates an unknown from succeeding equations is known as (Gauss) elimination.

where an

Example

2.3:

Consider the following system of linear equations:

2x
3x

4x

+ + +

iy 6y 8y

+ +

z
z

2v
V

5v

+ 2w - 1 + iw = 7 w = 3

eliminate the unknown x from the second and third equations by applying the following operations:

We

L2

^ -3Li +
-SLj:
21,2:

2L2

and
3z 2z 5z

L3

->

-2Li

Lg

We

compute

-6x - 12y +
ex

12y

6v 2v 8v

- 6w = -3 + Sw - -14 + 2w = -17 - 'iw = -2 w = 3 - 5w =


1

-3Li +
and

2L2:

21,1:
I/g:

-4* 8y +
4x

2z
z

4v
5v

Sy

-2Li

L3:

3z

CHAP.

2]

LINEAR EQUATIONS
Thus the
original system has been reduced to the following equivalent system:

21

2x

iy

+ +

2v 8v V

+ 2w =

5z 32

+ 2w = -17

5w =
Xj

Observe that y has also been eliminated from the second and third equations.
the

Here

unknown

z plays the role of the

unknown

above.

We note that the above equations, excluding the first, form a subsystem which has fewer equations and fewer unknowns than the original system (*). We also note that:
(i)

if

sistent

an equation Oa;i + + and has no solution;

Oa;

= =

5,

b =0

occurs, then the system is incon-

(ii)

if

+ Oa; an equation Oaji + without affecting the solution.


occurs, then the equation can be deleted

Continuing the above process with each new "smaller" subsystem, we obtain by induction that the system (*) is either inconsistent or is reducible to an equivalent system in the
following form
aiiXi

ai2X2

ttisccs

+
a2.j,+ lXi +
1

(iinX

= =
=

bi

Cl'2uXj,

+ + +

a2nXn

&2

Ciri^Xj^

+ar,j,+ ia;j^+i

arnX

br

where

< ^2 <

< ^r and where


an

the leading coefficients are not zero:


a2i2

0,

^'

>

^^^r

in the system

(For notational convenience we use the same symbols an, bk in the system (***) as (*), but clearly they may denote different scalars.)

we

used

Definition:

(***) is said to be in echelon form; the unknowns Xi which are termed do not appear at the beginning of any equation {iy^lyjz, ., jr)

The above system


free variables.

The following theorem

applies.

Theorem

2.2:

The solution of the system two cases:


(i)

(***) in echelon

form

is

as follows.

There are

r n. That is, there are as many equations as unknowns. system has a unique solution. r

Then the

(ii)

there are fewer equations than unknowns. Then we can arbitrarily assign values to the n r free variables and obtain a solution of the system.

<n.

That

is,

Note in particular that the above theorem implies that the system (***) and any equivalent systems are consistent. Thus if the system (*) is consistent and reduces to case (ii) above, then we can assign many different values to the free variables and so obtain many solutions of the system. The following diagram illustrates this situation.

22

LINEAR EQUATIONS

fCHAP.

System of linear equations

Inconsistent

Consistent

No
solution

Unique
solution

More than one solution

In view of Theorem 2.1, the unique solution above can only occur when the associated homogeneous system has only the zero solution.
Example
2.4:

We
L3
2x

reduce the following system by applying the operations L2 3Li + 2L3, and then the operation L3 -* SL^ + Lgi

-*

ZL^ + 2L2 and

Sx
3a;

+ + +

y
2y
32/

2z
z

3z

+ 3w = + 2w = - 3w = =
-8,

2x

y
2/

+ +

22

4 5

4z 12z

Zy
that
is, Oa;

+ 3w = 5w = - 15w =

2x

y y

2z

5 7

Az

+ 3w = 1 5w = 5 = -8

The equation
system

+ Oi/ + Oz + Ow =
solution.

-8,

shows that the original

is inconsistent,

and so has no

Example

2.5:

L^ -> reduce the following system by applying the operations 2Lt + Lg and I/4 -* 2Li + L4, and then the operations L3 Lg and L4 - -21/2 + L^:
-

We

Lj + L2, L2 Ls
4
7

X
X

2x 2x

+ + + +

2y
Sy

5y
ey

+ +

Sz
z

4 11 13

a;

2]/

+ + +

32

y y
2y
a;

Az
2z
8z 3z

4z
2z

= = =

4 7 5

2y
2/

Sz 4z 2z

22

14

= = = =

2?/
2/

4z
2z

= = =

4
7 2

Observe first that the system is consistent since there is no equation of the form = 6, with 6 # 0. Furthermore, since in echelon form there are three equations By the third equation, in the three unknowns, the system has a unique solution. Substitut2 = 1. Substituting z = 1 into the second equation, we obtain y = Z. Thus a; = 1, y = Z ing z = 1 and y Z into the first equation, we find x = 1. and z = 1 or, in other words, the 3-tuple (1, 3, 1), is the unique solution of the
system.

Example

2.6:

We
L3
X
2x
5x

reduce th "following system by applying the operations L2 5Li + L3, and then the operation L3 -^ 2L2 + L^:

-2Li

L2 and

+ + +

2y
42/

2z 3z 8z

lOy

+ Zw = + 4w = + llw =

2
5

2y

2z
z

12

2z

+ Zw = - 2w = 4w =
2
1

2
1

2y

2z
z

+ Zw = - 2w = =

2
1

2y

-2z + Zw = z - 2w =

CHAP.

2]

LINEAR EQUATIONS

23

The system is consistent, and since there are more unknowns than equations in echelon form, the system has an infinite number of solutions. In fact, there are two free variables, y and w, and so a particular solution can be obtained by giving y and w any values. For example, let w = 1 and y = 2. Substituting w = 1 into the second equation, we obtain z = 3. Putting w = X, z = 3 and 2/ = into the first equation, we find a; = 9. Thus a; = 9, y 2, z = 3 and w = 1 or, in other words, the 4-tuple (9, -2, 3, 1) is a particular solution of the system.

Remark:

find the general solution of the system in the above example as follows. Let the free variables be assigned arbitrary values; say, y = a and w = b. Substituting into the second equation, we obtain 2 = 1 + 26. Putting y = a, z = l + 2b and w = h into the first equation, we find a; = 4 - 2a + &. Thus the general solution of the system is

We

w-h
a;

or, in

bers.

other words, Frequently, the general solution and w (instead of a and 6) as follows:

2a + b, = a, 2 = 1+ 26, w h (4 2a + 6, a, 1 + 26, 6), where a and 6 are arbitrary num4


i/

is left

in

terms of the free variables y

2y +

w,

+ 2w

or

(4

2y + w,y,l + 2w, w)

We
Example

will investigate further the representation of the general solution of a

system of linear equations in a later chapter.


2.7:

Consider two equations in two unknowns:


a^x OgX

6jj/

= =

Cj

622/

C2

According to our theory, exactly one of the following three cases must occur:
(i)

The system

is

inconsistent.

(ii)

The system The system

is is

equivalent to two equations in echelon form. equivalent to one equation in echelon form.

(iii)

When
(i)

linear equations in two unknowns with real coefficients can be represented as lines in the plane R^, the above cases can be interpreted geometrically as follows:

The two

lines are parallel.

(ii)

The two
The two

lines intersect in a unique point.


lines are coincident.

(iii)

SOLUTION OF A HOMOGENEOUS SYSTEM OF LINEAR EQUATIONS


begin with a homogeneous system of linear equations, then the system is clearly = (0, 0, ., 0). Thus it can always it has the zero solution be reduced to an equivalent homogeneous system in echelon form:
If

we

consistent since, for example,

aiiXi

ai2X2

aisXs

+
a2.J2+lXj2+l

a2i2Xj2

+
+

+ amXn = + a2nXn = +
drnXn

drj^Xj^

({r.i^+lXj^+l

Hence we have the two


(i)

possibilities:

r r

= w.
<n.

Then the system has only the zero


Then the system has a nonzero

solution.

(ii)

solution.

If we begin with fewer equations than unknowns then, in echelon form, hence the system has a nonzero solution. That is,

<n

and

24

LINEAR EQUATIONS

[CHAP.

Theorem

2.3:

A
2.8:

homogeneous system of linear equations with more unknowns than

equations has a nonzero solution.


Example

The homogeneous system

- 3z+ w = x-Sy + z -2w = 2x + y - Zz + 5w =


X

2y

has a nonzero solution since there are four unknowns but only three equations.

Example

2.9:

We

reduce the following system to echelon form: X

z
z

2x-3y + x-4v +

2z

= -

-5y +
-5y +
solution, since

3z 3z

= = =

-5j/

3z

= =

The system has a nonzero

we

three unknowns in echelon form. In other words, the 3-tuple (2, 3,

For example,
5) is

obtained only two equations in the let z = 5; then y = S and x 2.

a particular nonzero solution.

Example

2.10:

We

reduce the following system to echelon form:

2x
Sx

+ + +

4y

z z

2y

2z

= = =

z z

= = =

z z

2y

2y

= = =

-y +

5z

llz

Since in echelon form there are three equations in three unknowns, the system has only the zero solution (0, 0, 0).

Solved Problems SOLUTION OF LINEAR EQUATIONS 2x-Sy + 6z + 2v


2.1.

5w =

3
1
.

Solve the system:

^z

= Zw -

Since the equations begin with the unknowns x,y and v respectively, the other unknowns, z and w, are the free variables. into the third equation, find the general solution, let, say, z = a and w = 6. Substituting

The system

is

in echelon form.

To

36

or

-y

36

Substituting into the second equation,


J/

4a

36

01

4a

36

Substituting into the first equation,

2x

3(4a

- 36 - 1) +
56

6a

2(2

+ 36) 36

56

=
z

3a

56

Thus the general


a;

solution of the system is

3a

2,

j/

4a

1,

36,

w =

real numbers. Some texts or (3a-56-2, 4a-36-l, a, 2 + 36, 6), where a and 6 are arbitrary variables z and w instead of a and 6 as follows: leave the general solution in terms of the free

CHAP.

2]

LINEAR EQUATIONS
X

25

= = -

Sz 4z
2

y
V

5w 3w
+ 3w

2
1

or

(3

5w 2,

4z

- 3w -

1, z,

3w, w)

general solution.

After finding the general solution, we can find a particular solution by substituting into the For example, let a = 2 and 6 = 1; then
X

-1,

J/

4,

2,

5,

w=

or

(-1,

4, 2, 5, 1)

is

a particular solution of the given system.

X
2.2.

+ 2y-3z = -1 y+ + 3y 2z 4z

Solve the system:

3a;

7 2

5x

1,2

- 3Li +

Reduce to echelon form. Eliminate x from the second and third equations by the operations L2 and Lg -* 5Li + L3:
-3Li:
L2:

-3x

-6y+9z=
y

3 7

-SLj:
L3:

-5x - lOy +
5a:

15z 4z
llz

3x-

+
+

2z

= =
X

3^/

= = =

2
7

-3Li

Lg:

-7j/

llz

10

-5Li

L3:

-7j/

Thus we obtain the equivalent system

2y

3z

= -1

-7y + -7y

llz llz

= =
is

10
7
inconsistent, for if

The second and third equations show that the system


Ox

we

subtract

we

obtain

Oy

Oz

or

3.

2x+ y-2z =
2.3.

10
1
.

Solve the system:

3x
5a;

+ 2y + +
42/

2z 32

=
2L3:

Reduce
L2
-*

to echelon form.

Eliminate x from the second and third equations by the operations

-3Li

21,2

and

^3

"*

-5Li +
6z

-31,1:
21,2:

-6x -3y +
6x

= -30 =
2

-SLj:
2L3:

-lOx
10a;

5y
83/

4j/

+ +

4z lOz

+ +

lOz
6z

= -50 = 8
= -42

-3Li +

2L2:

J/

= -28

-5Li

2L3:

3y

16z

Thus we obtain the following system from which operation L^ - 3L2 + Lgi
2x

we

eliminate y from the third equation by the

y
J/

+ +

2z lOz 16z

10
to

2x

y
y

2z

10

= -28 = -42

lOz

= -28 =
42

3y

-142

In echelon form there are three equations in the three unknowns; hence the system has a unique By the third equation, z = 3. Substituting into the second equation, we find j/ = 2. Substituting into the first equation, we obtain a; = 1. Thus x = l, y = 2 and z = -3, i.e. the 3-tuple (1, 2, 3), is the unique solution of the system.
solution.

26

LINEAR EQUATIONS
x

[CHAP. 2

+ 2y-3z =
y

2.4.

Solve the system:

2x
4x

iz

2
14

+ 3y-2z =

operations

Reduce the system to echelon form. Eliminate x from the second and third equations by the Lz -* 2Li + L^ and L3 - 4In + L3:
-2Li.
L2-

-2x

4j/

6
4:Z

= -12

-4Li:
L3:

-ix - 8y +
4*

12z 2

= -24 =
14

2x- y+
-5j/

32/

or

+ 10 = -10 2 y - 2z -

-5y + lOz = -10


or

2z

Thus the system

is

equivalent to

2y

-Sz =

X
2z

2y

3z

=
=

2
2

or simply

2z

y-2z =

(Since the second and third equations are identical,

we can

disregard one of them.)

In echelon form there are only two equations in the three unknowns; hence the system has an infinite number of solutions and, in particular, 3 2 = 1 free variable which is z.

To obtain the general


J/

= 2 + 2a.

Thus

solution let, say, z - a. Substitute into the second equation to obtain or a; = 2 o. Substitute into the first equation to obtain a + 2(2 + 2a) - 3o = 6 the general solution is
a;

a,

2a,

or

(2

- a, 2 + 2o, o)
=
or

where a

is

any real number.


say,

The value,

yields the particular solution

a;

1,

3/

4,

(1,4, 1).

2.5.

Solve the system:

x-Sy + 4:Z 2w = 2y + 5z + w = y-Sz

=4

The system is not in echelon form since, for example, y appears as the first unknown in both unknown, the second and third equations. However, if we rewrite the system so that w is the second
then

we

obtain the following system which is in echelon form:


a;

2w

32/

4z
Bz 3z

= =

w+

2y
2/

+ -

2 4

Now if a 4-tuple (a, 6, c, d) is given as a solution, it is not clear if 6 should be substituted for Of course this or for y; hence for theoretical reasons we consider the two systems to be distinct. us from using the new system to obtain the solution of the original system. does not prohibit + Za. Substituting into the Let z = a. Substituting into the third equation, we find -6 - 11a. Substituting into the first second equation, we obtain w + 2(4 + 3a) + 5a = 2 or w = equation, = 5 - 17o or a; ^ _ ^^_^ _ ^^^^ _ 3(4 + Sa) + 4a = 5

y-A

Thus the general

solution of the original system is


a;

17o,

J/

3a,

a,

w = -6 -

11a

where o

is

any

real number.

CHAP.

2]

UNEAR EQUATIONS
x,

27

2.6.

Determine the values of a so that the following system in unknowns (i) no solution, (ii) more than one solution, (iii) a unique solution:
X

y and z has:

= =

2x
X

+ Zy + + ay +

az Sz

3
2

operations

Reduce the system to echelon form. Eliminate x from the second and third equations by the Lj ^ 2Li + L^ and I/3 - I/j + Lg:
-2Li.

-2x -2y +
2x

2z az
(a

+ 3y+
y

= -2 - 3
1

X
X

ay

+ + +

Sz

= -1 = 2 =
1

+ 2)z =
X

(a-l)y

4z

Thus the equivalent system

is

y
J/

+
(a

(a-l)y+

+ 2)z = 4z =
L3

1 1
-^

Now

eliminate y from the third equation by the operation

(a V^L^
a

L^,

-(a

- 1)1,2:
La:

-(a -

+ (a-l)y +
1)2/

(2

- a - a2)z = 14z = 1

or
to obtain the equivalent

- a - a^)z - 2(3 + a)(2 - a)z = 2 (6 z

a a

system

-V

y
J/

+
(a

1 1

+ 2)z = (3 + a)(2 - a)z =

which has a unique solution if the coefficient of z in the third equation is not zero, that is, if a # 2 and a = 3. In case a = 2, the third equation is = and the system has more than one solution. In case a = 3, the third equation is = 5 and the system has no solution.
Summarizing, we have:
(i)

3,

(ii)

2,

(iii)

a' 2 and o

3.

2.7.

Which
X,

condition
z

y and

must be placed on has a solution?

a,

and

c so that the

following system in unknowns

+ 2y Sz 2x + ey-llz X 2y + Iz
X

a
b
c

1/2 -^

Reduce 2Li

to echelon form.

L2 and

I/3 -

Lj +

Eliminating x from the second and third equation by the operations L3, we obtain the equivalent system

2y
2y


4-

Sz 5z

4j/

lOz

= = =

a
b c

2a a
Lg
-*

Eliminating y from the third equation by the operation equivalent system X + 2y 3z a

2L2

L3,

we

finally obtain the

2y

5z

= =

b
c

2a
26

5o

28
The system
if

LINEAR EQUATIONS
have no solution if the third equation is of the form Thus the system will have at least one solution
c

[CHAP.

will

=
if

fe,

with

0;

that

is,

2b

5a

= 0.

26

5a

or

5a

26

Note, in this case, that the system will have more than cannot have a unique solution.

one solution.

In other words, the system

HOMOGENEOUS SYSTEMS OF LINEAR EQUATIONS


2.8.

Determine whether each system has a nonzero

solution:
,

x-2y + Zz-2w = 3x-ly-2z + 4kw ^


Ax

+ Sy +

5z
(i)

+ 2w =

+ 2y-3z = 2x + 5y + 2z = Sx- y-4z =


x
(ii)

2x

+ 5y + 2z = x + Ay + 7z = x+3y + Sz =
("i)

(i)

The system must have a nonzero


Reduce to echelon form:

solution since there are

more unknowns than equations.

(ii)

+ 2y-Zz = 2x + 5y + 2z = 3x- y-4z =


x

x
to

+ 2y-Sz = y + 8z = -7y + 5z =

x
to

2y

3z 8z

61z

= = =

hence the system has In echelon form there are exactly three equations in the three unknowns; unique solution, the zero solution. a
(iii)

Reduce to echelon form:


X

2x
a;

x 3y the system has a In echelon form there are only two equations in the three unknowns; hence nonzero solution.
Sz
j/

+ + + +

2y 5y

+ + +

2z
7z

= = = =

2y y

2y

+ + +

4z 8z 4z

= = = =

2y y

Az

= =

2.9.

The vectors Ui,...,Um


dependent,

in,

say,

R" are said to be linearly dependent, or simply


such
that

ki,...,km, not all of them zero, Otherwise they are said to be independent. kmiim = 0. kiui are dependent or independent where: whether the vectors u, v and

+ +

if

there

exist

scalars

Determme

(i)

(ii)

(iii)

w = (8, -7, 1) u = (1, -2, -3), V = (2, 3, -1), w = (3, 2, 1) u = (ai, a2), v = (bi, 62), w = (ci, C2)
u=
(1, 1,

-1), V

(2,

-3,

1),

In each case:
(a) let
(6) find
(c)

XU

yv

+ zw =

where

x,

y and

z are

unknown

scalars;

the equivalent homogeneous system of equations;

then the vectors are determine whether the system has a nonzero solution. If the system does, system does not, then they are independent. dependent; if the

(i)

Let XU

yv

+ zw =

0:

x{l, 1,

-1)

j/(2,

-3,

1)

+ +

z{8,

-7,
-7z,

1)

(0, 0, 0)

or or

(a;, a;,

-x)

(2j/,

-3y,

y)

(8,

z)

= =

(0, 0, 0)

(x

+ 2y + 8z, x-Sy -Iz, -x + y + z)

(0, 0, 0)

CHAP.

2]

LINEAR EQUATIONS
Set corresponding components equal to each other and reduce the system to echelon form:

29

+ 8z = x-Zy -7z =
X

2y

+ ~5y 2y 3y

Sz

15z
9z

= = =

2y y

+ +
+

Sz 3z Sz

-X + y +

= = =

2y y

+ +

8z
3z

In echelon form there are only two equations in the three unknowns; hence the system has a nonzero solution. Accordingly, the vectors are dependent.

Remark:

We need
to

know

if

not solve the system to determine dependence or independence; a nonzero solution exists.
x(l,
(x,

we

only need

(ii)

-2, -3)

3/(2, 3,

-1)

z{3, 2, 1) (3z, 2z, z)

= =

(0, 0, 0)
(0, 0, 0) (0, 0, 0)

(x

+ (2y, 3y, -y) + + 2y + 3z, -2x + 3y + 2z, -3x - y + z) =


-2x, -3x)
3z 2z
z

2y

= = =

2y 7y
5y

+ + +

3z Sz

= =

2y 7y

-2x + 3y + -3x - y +

+ +

3z 8z

=
=

lOz

30z

In echelon form there are exactly three equations in the three unknowns; hence the system has only the zero solution. Accordingly, the vectors are independent.
(iii)

x(,ai, 02)
{ttix,

y{bi, 62)

+
+

z(ci, C2)

(0, 0)

a2x)
61J/

{byy, h^y)

(c^z, c^z)

(0, 0)
(0, 0)

and so

"''*'

'^

"'^

(dja;

+ CiZ,

a2X

b^y

+ C2Z) =

02*

62J/

C2Z

~ =

The system has a nonzero solution by Theorem 2.3, i.e. because there are more unknowns than equations; hence the vectors are dependent. In other words, we have proven that any three vectors in R2 are dependent.

2.10.

Suppose in a homogeneous system of linear equations the coefficients of one of the unknowns are all zero. Show that the system has a nonzero solution.
Suppose
are
all zero.

!, ..., are the unknowns of the system, and Then each equation of the system is of the form

Xj is the

unknown whose

coefficients

i^i
. .

+
.

ttj-i^j-i
.

Oajj

+
is

aj + i^j +

oa;
is

Then for example (0, .,0, 1, 0, .,0), where 1 equation and hence of the system.

the ;th component,

a nonzero solution of each

MISCELLANEOUS PROBLEMS
2.11.

Prove Theorem (*) and suppose

2.1:

is

Suppose m is a particular solution of the homogeneous system the general solution of the associated homogeneous system (**).

Then
is

u +

{u

+ w: w G W}
(*).

the general solution of the nonhomogeneous system


Let

u = (%

denote the general solution of the nonhomogeneous system (*). Un). Since m is a solution of (*), we have for t = 1, , m,
.
.

Suppose

uG U

and that

Now
(**),

suppose

w6
for
i

W
=

and that
1,
.
.

w=

(w^,

.,

w).

Since

is

a solution of the homogeneous system

we have

.,m,
OiiWi

aj2W2

fflin^^n

30

LINEAR EQUATIONS
Therefore, for

[CHAP.

l, ...,nt,

0,i(Mi

+ Wi) + =

Oi2(M2

+ W2) +
OjiWi

+ +

ai{Un
t2"'2

+ W) +
' " '

OjiMi
(OilMl
6i

aj2M2

linMn

inWn
'
"

=
That
is,

+ ai2M2 + + = 6j
(*).

+ tnO +

("il^l

+ i2W2 +

'

+ in'n)

M+w

is

a solution of

Thus u + w e^U, and hence

u +

U
i.e.

Now
t

suppose

(vi,

vj

is

any arbitrary element of U,

solution of

(*).

Then, for

1,

.,w,
ttji^i

ai2-U2

aiv

bj

Observe that v

= u+(v u). We claim that v-uGW. For i = 1, + m(^n ~ "n) ail(i;i Ml) + ai2(t'2 M2) + = (OjlVl + aj2'y2 + + ftin^n) ~ (il"l + t2"2 +
.

m,

"

'

+ {)

=
Thus V

6i

6i

=
(*), i.e.

- M is

a solution of the homogeneous system

u-mST^.

Then

vGu+W,

and hence

U Q
Both inclusion relations give us nonhomogeneous system (**).

u +

W
is,

U - u + W;

that

u+W

is

the general solution of the

2.12.

Consider the system (*) of linear equations (page by Ci, and adding, we obtain the equation
(CiOn

18).

Multiplying the ith equation

Cmaml)Xl

(Cittm

CmOmn)*

Ci5i

Cmbm

(1)

Such an equation is termed a linear combination of the equations any solution of () is also a solution of the linear combination (1).
Suppose M

in (*).

Show

that

(fci,

fcj

is

a solution of

(*).

Then

ffliifci

+ aah +
(1),

{*:

6i,

1,

. .

.,m

(2)

To show that
(Ciaii

tt

is

a solution of

we must

verify the equation

Cmd^Ofcl

("ifflln

emO'mn)K

!&!

"

"

"

C^fe^

But

this

can be rearranged into


Ci(anfel

0,lnK)
Ci6i

+
+

Cm(aml
cj)^

+ =

amn'n)

=
+

Ci^l

'

"

"

C^fem

or,

by

(2),

Cjbi

c^b^

which

is

clearly a true statement.

2.13.

of linear equations, suppose an = 0. Let (#) be the system ob^ -anLi + auU, i^l. Show that (*) and (#) tained from (*) by the operation same solution set. are equivalent systems, i.e. have the

In the system

(*)

equations In view of the above operation on (*), each equation in (#) is a linear combination of problem any solution of (*) is also a solution of (#). in (*); hence by the preceding

On

the other hand, applying the operation

!/{ -*

(-Oii^^i

i-O

to (#),

we

obtain the origi-

nal system (*). That is, solution of (#) is also a solution of

each equation in (*) is a linear combination of equations in (#); hence each


(*).

Both conditions show that

(*)

and (#) have the same solution

set.

CHAP.

2]

LINEAR EQUATIONS
2.2:
aiia;i

31

2.14.

Prove Theorem

Consider a system in echelon form:

ai^Xi

+ aizXz + a2J2^J2 + a2,J2+lJ2 + +


l

+ +

ainXn ainXn

= = =

bi

62

O'ri^^ir

<*r,]V+ia;j^+l

+
.

arnaJn
ari,

for

where
(i)

1<h<
= n.

< jr and where an ^


cases:

0, a2J2

^0,

^ 0,

The

solution is as

follows.

There are two

(ii)

r <n. obtain a solution of the system.

Then the system has a unique solution. Then we can arbitrarily assign values
number r of equations

to the

nr
If

free variables and


r

The proof is by induction on the the single linear equation


aiXi

in the system.

1,

then

we have

a^Xi

a^x^

aa;

6,

where

Oj #

The free variables are


2

x^,

., a;.

Iti,

xs

k^,

...,

fe.

Let us arbitrarily assign values to the free variables; say, Substituting into the equation and solving for Xi,

Xi

(6 "1

fflzfca

asks

ofc)

These values constitute a solution of the equation;


"i

for,

on substituting,

we
6

obtain or
6

(* "" 02*^2

ak)

ajt^

afc

which

is

a true statement.
if

Furthermore
tion since
a(b/a)

r
5

=%=

1,

then

we have ax =
if

b, is

where a
a solution,

# 0.
i.e.

is true.

Moreover

=
is

Note that x = b/a is a soluak = b, then k = b/a. Thus

the equation has a unique solution as claimed.

Now

assume r

>

and that the theorem

true for a system of r


"^ "2na;n

1 =

equations.
*2

r 1 equations
'*2J2*J2

We

view the

'*2,J2+1*J2 +

"^
1

as a system in the unknowns Xj^

Note that the system

is in

echelon form.

By

induction

we can

arbitrarily assign values to the

(n
. ,

J2

1)

As

(r

to obtain

a solution

(say,

Xj^
J2

fcj^,

&).

1) free variables in the reduced system in case r = 1, these values and arbitrary

values for the additional


of the first equation with

free variables (say,

a;2

^2,

a^j,-!

'fja-i)'

yield

a solution

Xi

Oil

(6j

012^2

aik)

(Note that there are (n J2 + 1) (r 1) + (jg 2) = n r free variables.) Furthermore, these values for Xi, .,x also satisfy the other equations since, in these equations, the coefficients of !,..., j-i are zero.
. .

Now if r = n, then 32 = 2. Thus by induction we obtain a unique solution of the subsystem and then a unique solution of the entire system. Accordingly, the theorem is proven.
2.15.

system

(*)

of linear equations
is

is

defined to be consistent

if

no linear combination
(I)

of its equations

the equation
Oa;i

0*2

Oa;
if

b,

where b
if it is

Show

that the system


(*) is

(*) is

consistent

and only

reducible to echelon form.

reducible to echelon form. Then it has a solution which, by Problem 2.12, is a solution of every linear combination of its equations. Since (1) has no solution, it cannot be a linear combination of the equations in (*). That is, (*) is consistent.

Suppose

On
it

the other hand, suppose


yield
(*) is

(*) is

must

an equation of the form


not consistent,

(1).

not reducible to echelon form. Then, in the reduction process, That is, (J) is a linear combination of the equations in (*).
inconsistent.

Accordingly

i.e. (*) is

32

LINEAR EQUATIONS

[CHAP. 2

Supplementary Problems
SOLUTION OF LINEAR EQUATIONS
2x
2.16.

Solve:

(i)

5x

+ +

Sy

7y

= =

1
(ii)

2x 3

+ +

4y
6y

= =

10
(iii)

Ax

-2y =

5
1

15

-6x + 3y =

2.17.

Solve:

2x
(i)

Sx

+ y - Sz = -2y + 2z = 5x -Sy - z =

5 5
(ii)

2x

16

+ 3y -2z = - 21/ + 3 = 4a; + 4z =


1/

5 2
1
(iii)

X
2x
3a;

+ 2y + + 3y+ + 2j/ +

3
8

17?

= = =

4
1

2.18.

Solve:

2x
(i)

+ +

3y

X- 2y
3x 2y

= =

x
(ii)

5 7

2x 3x

+ 2y-3z + 2w + 5y - 8z + Gw + Ay - 5z + 2w =
= 2 = 5 = -A

2 5 4
(iii)

2x
3x

+ + +

2y Ay
6j/

+ -

4z
2

+ 3w = + 3w = + 8w =

3 9

10

X 3x
2.19.

Solve:

(i)

2x
X

+ 2y + -2y - 5y + + Ay +

2z
z

X
(ii)

3a;

+ +

5y

y 2y

3z
6z

2x

+ + +

Az

2z
3z

= + 5w = - Aw =
13w

2
1

2.20.

(ii)

Determine the values of fc such that the system in unknowns no solution, (iii) more than one solution:

x,

y and

z has:

(i)

a unique solution,

kx
(o)

x X

+y+z = + ky + z = + y + kz =

X
l
1
(6)
2a;

+ +

2y
fc

+ +

kz
8

= =

2.21.

Determine the values of k such that the system (iii) more than one solution: (ii) no solution, X
(a)

in

unknowns
X
(6)

x,

y and

z has:

(i)

a unique solution,

3x

2x

+ + +

y Ay
3y

+ +

kz

2z-k
z

2x
X

+ ky + 2y +

3z
z

kz

= -3 = -2 - 1

2.22.

Determine the condition on

a, b

and

c so that the

system in unknowns

x,

y and a
b e

has a solution:

(i)

X + 2y 3x- y + X 5y +

3z 2z
Bz

= =

a
b e
(ii)

x 2y + Az = 2x + Sy - z 3x + y + 2z =

HOMOGENEOUS SYSTEMS
2.23.

Determine whether each system has a nonzero

solution:

(i)

-2z = x-8y + 8z 3x-2y + Az =


x

3y

(ii)

+ 2x 3x X

3y 3y 2y

-2z = + z = + 2z =

(iii)

+ 2y 2x - 3y + 4x 7j/ +
X

5z

+ Aw = 2z + 3w = z 6w

CHAP.

2]

LINEAR EQUATIONS

33

2.24.

Determine whether each system has a nonzero solution:

X2x+
(i)

2y

+ -

2z 2z
6z

2x 9x
(ii)

+ + -

4y 3y 2y
5y

+ + -

7z

= =
=

2z 3z

+ ^
+

4v
7v

- 5w -

+ w =
+ 3w 2w Q

3x+ 4y3x - lly +

5x 6x

v
3v

12z

+ 4z-

2.25.

Determine whether the vectors


(i)

u,

v and

w
(1,

are dependent or independent (see Problem 2.9) where:

u = u

(1, 3,

-1), V -1), V
3, 1),

= =
V

(2, 0, 1),

w =

-1,

1)

(ii)

(1, 1,

(2, 1, 0),

w =
-2),

(-1,

1, 2)

(iii)

u =

(1,

-2,

(3, 2, 1,

w =

(1, 6,

-5, -4)

MISCELLANEOUS PROBLEMS
2.26.

Consider two general linear equations in two unknowns x and y over the real

field

K:

ax
ex

+ +

by

= =

dy

Show
(i)

that:

it

-'2,
,

i.e.

if

ad

6c = 0,

then the system has the unique solution

^^

^
(ii)

_ af ee ~ ad-bc'
7 =
J
'^

ad

^f he

i*

7. tlien the system has no solution;

(iii)

ii

= 2 =

-f>

then the system has more than one solution.

2.27.

Consider the system

ax
ex

+ +

by

= =

dy

Show

that if

Also show that

ad-be'0, then the system has the imique solution x = d/(ad be), y = e/{ad if adbe = 0,e'0 or d ^ 0, then the system has no solution.

be).

2.28.

Show that an equation of the form Oki system without affecting the solution set.

Oa;2

Oa;

may

be added or deleted from a

2.29.

Consider a system of linear equations with the same number of equations as unknowns:
fflii*!

+ +
+

ai22
a222

+ +
+

+ + +

aix a2x

= =
=

61

a^xi

62
(i)

Onl*!
(i)

01.2*2

Suppose the associated homogeneous system has only the zero solution. unique solution for every choice of constants 6j. Suppose the associated homogeneous system has a nonzero solution. 64 for which {!) does not have a solution. Also show that if it has more than one.

Show that

(i)

has a

(ii)

Show
{1)

that there are

constants

has a solution, then

34

LINEAR EQUATIONS

[CHAP.

Answers
2.16.
(i)

to - 2a,

Supplementary Problems
j/

2,

-1;

(ii)

a;

a;

(iii)

no solution
rx

2.17.

(i)

(1,-3,-2);

(ii)

no solution;

(iii)

{-1

- 7a, 2 + 2a,

a)

or

|^ ^

= -1 - 7z g + 2z

2.18.

(i)

a;

3,

1/

= -1
ra;

(ii)

(-a

+ 26, 1 + 2a - 26, 56/2

a, 6)

o*"

(iii)

(7/2

2a, a, 1/2

6/2, 6)

= z + 2w ^ + 2z 2w = 7/2 or |^ ^ ^^^ +
a;

5w/2

2j/

w/2

2.19.

(i)

(2, 1,

-1);

(ii)

no solution

2.20.

(a)
(6)

(i)

k'l and
never has

fe

-2;

(ii)

fc

-2;
k

(iii)

fe

1
fe

(i)

a unique solution;

(ii)

4;

(iii)

t^

2.21.

(a)
(6)

(i)
(i)

fc

fc

# 3; ^2
6

(ii)

always has a solution;


A;

(iii)
(iii)

fe

and

# -5;
(ii)

(ii)

fc

-5;

fe

=3 =2
c yields

2.22.

(i)

2a

0.

Any

values for

a, b

and

a solution.

2.23.

(i)

yes;

(ii)

no;

(iii)

yes,

by Theorem

2.3.

2.24.

(i)

yes;

(ii)

yes,

by Theorem

2.3.

2.25.

(i)

dependent;

(ii)

independent;

(iii)

dependent

chapter 3

Matrices
INTRODUCTION
positions are important.

In working with a system of linear equations, only the coefficients and their respective Also, in reducing the system to echelon form, it is essential to keep the equations carefully aligned. Thus these coefficients can be efficiently arranged in a rectangular array called a "matrix". Moreover, certain abstract objects introduced in later chapters, such as "change of basis", "linear operator" and "bilinear form", can also be represented by these rectangular arrays, i.e. matrices. In this chapter, we will study these matrices and certain algebraic operations defined on The material introduced here is mainly computational. However, as with linear equations, the abstract treatment presented later on will give us new insight into the structure of these matrices.

them.

but

Unless otherwise stated, all the "entries" in our matrices shall come from some arbitrary, fixed, field K. (See Appendix B.) The elements of are called scalars. Nothing essential is lost if the reader assumes that is the real field R or the complex field C.

Lastly, we remark that the elements of R" or C" are conveniently represented by "row vectors" or "column vectors", which are special cases of matrices.

MATRICES
Let

K be an arbitrary field. A

rectangular array of the form


0,12 0,22
. .

din
0,2n

\Q,ml

Om2

...

fflr

where the Odi are scalars in K, is called a matrix over K, or simply a matrix if K is implicit. The above matrix is also denoted by (ohj), i = l, .,m, j = 1, .,n, or simply by (a). The m horizontal -tuples
.
. .

(ail, ai2,

ttln),

(tt21, 0^22,

a2n),

.,

{ami, am2,

Omn)

are the rows of the matrix, and the n vertical w-tuples


lai2X
a22
,

\am2l

are its columns. Note that the element ay, called the ij-entry or ij-component, appears in the ith row and the yth column. A matrix with rows and n columns is called an by matrix, or x n matrix; the pair of numbers (m, n) is called its size or shape.

35

36

MATRICES
/I -3
(

[CHAP.

Example

3.1:

The following
Its

is

a 2 X 3 matrix:
3,

4\
_c, )

rows are

(1,

4)

and

(0, 5,

2);

its

columns are

and

and the elements of the Matrices will usually be denoted by capital letters A,B, ., Two matrices A and B are equal, written A = B, if field by lower case letters a,b, they have the same shape and if corresponding elements are equal. Thus the equality of matrices is equivalent to a system of mn equalities, one for each pair of elements. two
. .

mxn

Example

3.2:

The statement
of equations:
.
..

^
(

"

\x-y

z-wj

'")=(,
VI
x

4/
y

.)

is

equivalent to the following system

X
2z
z

= = y +w = w = + =
1,

3
I

The

solution of the system is

2,

3,

w=

1.

Remark:

A matrix with one

row

is also

referred to as a

row

vector,
field

as a column vector. a 1 X 1 matrix.

In particular,

an element in the

and with one column K can be viewed as

MATRIX ADDITION AND SCALAR MULTIPLICATION Let A and B be two matrices with the same size, i.e. the
columns, say,

same number of rows and of


^u
^Hi

mxn matrices:
(an
a21

ai2 022

...
...

ain
CLin

\
.
I

^^2 ^22

... ...

bin
ban

Oml

ami

ffimn

&ml

6m2

&mti

The sum of

and B, written

A + J?,
1

is

the matrix obtained by adding corresponding entries:


ai2 a22

an + &n

A + B =

a21

+ +

&2I

+ + +

&12 622

...

am +
a2n

bin
?>2n

...

+ +

ami

bml

Omi

&m2

Omn

The product of a scalar k by the matrix A, written by multiplying each entry of A by k:


kai2

fc

or simply kA,

is

the matrix obtained

feain
.

Ckaii fca2i
kaml

ka22

kazn

fcOm2

kOmn
I

Observe that

A+B and kA are also mxn matrices. We


-A = -1-A
and
is

also define

A-B ^ A+

{-B)

The sum

of matrices with different sizes

not defined.

CHAP.

3]

MATRICES

37

Example

3.3:

Let

A =

(]

J\

and

B =

fj

^).

Then
4

A + B
3A

1 + 3-2 + 3 + 2 4-7 5 + 1-6 + 8

-2
6 9

-3

= =

3*1
3 '4

(-2)

3-6
12
"' -3 -24

3-5 ') +

3 '(-6)

15

-18

2A-SB
Example
3.4:

-c

r '8

-'
10

-12/

r
V2I
... ...
...

-7
29

-4
7

-36

The

mXn matrix

whose entries are


10

all zero,

,0
is

0,

called the zero matrix

that, for

any

mXn

and matrix

will be denoted

by

0.

It is similar to the scalar

A=

in

(a^),

A+ =

(a^

+ 0) =

(Oy)

A.

Basic properties of matrices under the operations of matrix addition and scalar multiplication follow.

Theorem

3.1 :

Let

F be the set of all m x n matrices over a field K. Then for any matrices

A,B,C
(i)

GV

and any scalars

ki, kz

K,
(v)
(vi)
(vii)
(viii)

(ii) (iii)

(iv)

(A+B) + C = A + A+ = A A + (-A) = A+B = B + A


we
also

{B + C)

+ B) = kiA + kiB {ki + fe)^ = kiA + k^A (kiki)A = kiik^A) 1- A = A and OA =


k,{A
...
.

Using

(vi)

and

(viii)

above,

have that

A + A = 2A,A + A + A = ZA,
(or
. .

Remark:

Suppose vectors
say,

in

R" are represented by row vectors


(ai, Oi,
.

by column
.

vectors);

. ,

ttn)

and

(bi, 62,

b)

Then viewed as matrices, the sum u + v and the

scalar product

ku are as

follows:

u+
But

{ai

+ bi,a2 + b2,...,an + b)

and

ku

(fcai,

kaz, ..., A;a)

this corresponds precisely to the sum and scalar product as defined in Chapter 1. In other words, the above operations on matrices may be viewed as a generalization of the corresponding operations defined in Chapter 1.

MATRIX MULTIPLICATION The product of matrices A


reason,
(i)

we

and B, written AB, is include the following introductory remarks.

somewhat complicated.

For

this

Let A = (Oi) and B = (bi) belong to R", and A represented by a row vector and B by a column vector. Then their dot product A B may be found by combining the matrices
as follows:

A-B
Accordingly, above.

lbl\
(tti,

02,

.,a)

aibi

a2b2

ttnbn

Wl
we
define the

matrix product of a row vector

by a column vector

as

38

MATRICES
bnXi

[CHAP.

+
+

biiXi b22X2

feisics

= =

y\
(1)
1/2

(ii)

Consider the equations


h2lXl

b23X3

This system

is

equivalent to the matrix equation

6n
&21

b.
&22

b.s\h\
&23/U3/
and

fyA

^^^.^pjy

^^ ^ ^
B
_
and the column

V^V
(yi),

where
vector

B (&), X = X as follows:
Dv-

(x,)

if

we combine

the matrix

""

"'-

""!(

/feiiaJi

\b2iXl

+ +

&i2a;2

b22X2

+ +

bisa^sN

b2SXs

fBi'X \B2-X

where Bi and B2 are the rows of B. vector yields another column vector.

Note that the product of a matrix and a column

(iii)

Now

auVi
consider the equations
a2iyi

+ +

ai22/2 (i22y2

zi

(2)
Z2

which we can represent, as above, by the matrix equation


^aii ^21

ai2\/yi\
,

( Zx
,
,

or simply

AY = Z
1/2

022/^2/2/

y22

where
of

A=

(Cij),

Y=

{yi)

as above, and
{2),

Z= + +

(z^.

Substituting the values of y\ and

(i) into

the equations of
aii(&iia;i a2i(&iia;i

we

obtain
ai2(62ia;i a22(&2ia;i

+ +

6i2a;2 &i2a;2

+ +

b\%x%) bisXs)

+ +

622332
&22a;2

+
+

&23a:;3)

=
=

btzx^)

22

or,

on rearranging terms,
(ttubii

+ ai2&2i)a;i + +
022&2i)a;i

(aii6i2

ai2&22)a;2
a22&22)a;2

+
+

(an&is
(021613

+ +

a\2b23)Xz
022623)033

= =

Zi

(azi&u

(2i&i2

(3)
22

On

AY = Z,

the other hand, using the matrix equation we obtain the expression

BX = Y

and substituting for

into

ABX = Z
This will represent the system
ftii

(3) if

we

define the product of


/aii6ii

and

as follows:
011613 021613

ffli2\/6n 612 bisX

021

022/1621

622

623/

+ 012621 YO21611 + 022621

011612
021612

+ 012622 + 022622

+ 012623 + O22623

Ai-B'
A2-jB'

ArB^ Ai-B^
Aa-B^ A2'B^

and J?S B^ and B^ are the columns of B. We emphasize that if these computations are done in general, then the main requirement is that the number of yi in (1) and (2) must be the same. This will then correspond to the fact that the number of columns of the matrix A must equal the number of rows of the matrix B.

where Ai and A2 are the rows of

CHAP.

3]

MATRICES
introduction,

39

With the above


Definition:

we now formally

define matrix multiplication.

Suppose
of

is

pxn

A = (a) and B = (&) are matrices such that the number of columns equal to the number of rows of B; say, A is an x p matrix and B is a matrix. Then the product is the matrix whose y-entry is

AB

mxn
. . .

obtained by multiplying the ith

row A,

of

by the yth column B' of B:

Ai-fii

AB =
That

A2-S1

Ai-B2 A2-B2

. .

Ai-5" A2-B"

,A-Bi
is,

Am-B^

Am'B"!

jail

/Cii

...

Cm

Cii

dml

Opn
P

\Cml

...

Ci

where

cy

aiiftij

ai2&23

+
is

+ avpbp. =
if

fc

2 1 =

Cifc&fci-

We emphasize that the qxn matrix, where p ^ q.


Example
3.5:

product

AB
(H
62

not defined

is

an
ra2
(02

mxp

matrix and

is

r
t

<i

"3
^3

raj
toi

u
2 4
1 1 1

6i

+ s6i + m6i

+ 562 + M^2

^'is

*"3

+ S63 + '^^s
1 3

Example

1
3.6:

3
1

1-1 + 2-0 3-1 + 4-0 1-1 + 1-3 0'l + 2-3

I'l + 2-2 3l + 4'2

11
6 8

2
4

1-2 + 1'4 0*2 + 2*4


is

4
6

The above example shows that matrix multiplication

not commutative,

i.e.

the products

AB and BA
Theorem

of matrices need not be equal.

Matrix multiplication does, however, satisfy the following properties:


3.2:
(i)

iAB)C = A{BC),

(associative law)

(ii)

(iii)

(iv)

A{B + C) = AB + AC, (left distributive law) (B + C)A = BA + CA, (right distributive law) k{AB) = {kA)B = A{kB), where is a scalar
A;

We assume that the sums and products in the above theorem are We remark that OA = and BO =^ where is the zero matrix.

defined.

TRANSPOSE
The transpose of a matrix A, written A*, A, in order, as columns:
/ttii
0,21

is

the matrix obtained by writing the rows of

0.12
ffi22

Oln 02n

'

\
/

/ttli

0.21
ffl22

aTOl\

O12

Om2
OmnJ

^Oml

Om2

Omni

\Oin
is

02n

Observe that

if

is

an

m x matrix,

then A'

an w x

m matrix.

40

MATRICES
/l
4^

[CHAP.

Example

3.7:

(J

_IJ

(2

-5^

The transpose operation on matrices

satisfies the following properties:

Theorem

3.3:

(i)

(A+B)* = A* + B*
(A')'

(ii)

(iii)

(iv)

= A {kAy kA\ {ABy = BA

for k a scalar

MATRICES AND SYSTEMS OF LINEAR EQUATIONS


The following system
of linear equations

anXi
a2iXi

+ ai2X2 + + a22X2 +
+
am2X2

aina;n

&i

+ +

annXn

=62

n\

OmlXi
is

OmnXn

equivalent to the matrix equation

/an
a2i

ai2
022

...

a2n

\IX2\

lb2

or simply

AX = B

(2)

lOml

fflm2

every solution of the system {1) is a solution of the matrix equation (2), and vice versa. Observe that the associated homogeneous = 0. system of (1) is then equivalent to the matrix equation

where

A=

(an),

X=

{Xi)

and

B=

(&i).

That

is,

AX

The above matrix

is called

the coefficient matrix of the system


/(111
tt21

(1),

and the matrix

O12
tt22

ttin (^2n

^ttml

(lm2

Otnn

is called

by

its

the augmented matrix of augmented matrix.


3.8:

(1).

Observe that the system

(1) is

completely determined

Example

The

coefficient

matrix and the augmented matrix of the system


2a;

3j/

4z
5z

x-2yare respectively the following matrices:

= /2 \l

7 3

/2

(1
Observe that the system

_2
is

-4\ -5;

3-4
-2
-5

7 3

^*^

equivalent to the matrix equation


'2
1

X\
3

,rj

-2

In studying linear equations it is usually simpler to use the language and theory of matrices, as indicated by the following theorems.

CHAP.

3]

MATRICES
3.4:

41

Theorem

are solutions of a homogeneous system of linear Then every linear combination of the m of the form kiUi + kiUz + + krOin where the fe are scalars, is also a solution of AX = 0. Thus, in particular, every multiple ku of any solution u of

Suppose

Ui,U2,

.,tin

equations

AX = 0.

AX =
Proof.

is also

a solution of

AX = 0.
. . . ,

We

are given that Aui

A{kui

kui

0, Au2 = 0, Aun = 0. Hence + knAun + fettn) = kiAui + kiAu^ + = fciO + ^20 + = fc0 +

Accordingly, kiUi

+ kiia

is

a solution of the homogeneous system

AX = 0.

Theorem

3.5:

Suppose the
field C).

field

K is

infinite (e.g. if

Then the system AX = an infinite number of solutions.

K is the real field R or the complex has no solution, a unique solution or


it

infinitely

Av

= B has more than one solution, then Proof. It suffices to show that if = B; that is, Au = many. Suppose u and v are distinct solutions of B. Then, for any k GK,

AX

has

AX

and

A{u + k{u-v)) =
tions

In other words, for are distinct


claimed.

Au + k{Au-Av) = B + k(B-B) = B solution of AX = B. Since each k e K, u + k(u-v) is (Problem 3,31), AX = B has an infinite number of
a.

all

such soluas

solutions

ECHELON MATRICES A matrix A = (an) is an echelon matrix, or is said to be


'

in echelon form, if the

number

of zeros preceding the first nonzero entry of a row increases rows remain; that is, if there exist nonzero entries
aiii, '^^h'

row by row

until only zero

"'^'r'

where
j

^i

< ^2 <

< jr

with the property that


aij

for

i^r,

< ji,

and for

i>r

We call

ttijj,

, ttrj,.

the distinguished elements of the echelon matrix A.

Example

3.9:

The following are echelon matrices where the distinguished elements have been
circled:
/(i)

-6\
2

1-3

0/

In particular, an echelon matrix tinguished elements are:


(i)

is called

a row reduced echelon matrix

if

the dis-

the only nonzero entries in their respective columns;

(ii)

each equal to

1.

third matrix above is an example of a row reduced echelon matrix, the other two are not. Note that the zero matrix 0, for any number of rows or of columns, is also a row reduced echelon matrix.

The

ROW EQUIVALENCE AND ELEMENTARY ROW OPERATIONS


A matrix A is said to be row
finite

equivalent to a matrix

if

can be obtained from


operations:

by a

sequence of the following operations called

elementary row

42

MATRICES
Interchange the ith row and the yth row: Rt <^
Multiply the ith row by a nonzero scalar
k:

[CHAP. 3

[Et]:

Rj.
-

[E2]:
[Es]:

Ri

kR,,

fc

v^ 0.
-*

Replace the ith row by k times the jth row plus the ith row: Ri

kRj

R,.

In actual practice
[E]:

we apply

[^2]
fe'

and then

["3]

in one step,

i.e.

the operation

Replace the ith row by Ri -* k'Rj + kRi, k^-O.

times the jth row plus k (nonzero) times the ith row:

and those used In fact, two systems with row equivalent augmented matrices have the same solution set (Problem 3.71). The following algorithm is also similar to the one used with linear equations (page 20).
similarity of the above operations
in solving systems of linear equations.

The reader no doubt recognizes the

Algorithm which row reduces a matrix


Step
1.

to echelon

form:

Suppose the ji column is the first column with a nonzero entry. Interchange the rows so that this nonzero entry appears in the first row, that is,
so that
ttijj

- 0.

Step

2.

For each

>

1,

apply the operation


Ri
-*

ttij^Ri

aijjiJt

Repeat Steps 1 and 2 with the submatrix formed by all the rows excluding the Continue the process until the matrix is in echelon form.

first.

Remark:

The term row reduce


3.10:

shall

mean

to

transform by elementary row operations.


to echelon

Example

The following matrix R2 ^ -2Ri + ^2 and

is

row reduced

i?3

^ -3fii + R3,
1

form by applying the operations and then the operation R3 - -SKj + 4^3:
2

A=2 4-22to0042to
4 2 3 5

-3

-3
4 2 2

Now
aijj,
. .
.

suppose
Orj^.

a matrix in echelon form with distinguished elements Apply the operations


(oij)

A=

is

Rk
for
i

-^

-ak^Ri

Oii-Rk,

fc

1,

.,

i-

then i = 3, ...,i = r. Thus A is replaced by an echelon matrix whose distinguished elements are the only nonzero entries in their respective columns. Next, multiply Ri by a~^, i~r. Thus, in addition, the distinguished elements are each 1. In other words, the above process row reduces an echelon matrix to one in row reduced echelon form.

2,

Example

3.11:

On the following echelon matrix A, apply the operation the operations fii - ^3 + Bi and R^ -> 5K3 + 2i22:

R^

-^

4^2 + 3i2i
/6 9
6

and then

A=0
matrix

/2

4
3

5toO
2/

6\

/6

9 3

-2\

5to0
2/

0^

4
2/

\0

\0
1/6

\0

Next multiply Ri by

1/6,

R2 by

and ^3 by 1/2
3/2 7/6

to obtain the

row reduced echelon

/l

0\

12/3
\0
1/

The above remarks show that any arbitrary matrix A is row equivalent to at least one row reduced echelon matrix. In the next chapter we prove, Theorem 4.8, that A is row equivalent to only one such matrix; we call it the row canonical form of A.

CHAP.

3]

MATRICES

43

SQUARE MATRICES A matrix with the same number of rows


matrix with n rows and n columns
is

as columns

is called

said to be of order n,

and

The diagonal
an,
a22,
. .

(or:

main

diagonal) of the n-square matrix


/l

A=
2 5 8

a square matrix. A square an n-square matrix. consists of the elements (Oij)


is called

. ,

ftjin.

3^

Example

3.12:

The following

is

a 3-square matrix:

6
9,

\7
Its diagonal elements are 1, 5

and

9.

triangular matrix or simply a triangular matrix below the main diagonal are all zero: entries

An upper

is

a square matrix whose

/an

ai2

ain\
ain

/ail

ai2 a22

am
a2n

O22

or

\0

...

ann/

Similarly, a lower triangular matrix is a square matrix

whose entries above the main


all zero:

diagonal are

all zero.

A diagonal matrix

is

a square matrix whose non-diagonal entries are


...
\

/a,
a2

'ai

...

or
a/

"-^
'

\o

...

an

In particular, the n-square matrix with I's on the diagonal and O's elsewhere, denoted by / or simply /, is called the unit or identity matrix; e.g., /l
0^

h =
\0
This matrix I
is

10
1,

similar to the scalar 1 in that, for any n-square matrix A,

AI = lA = A
The matrix
a scalar k diagonal entries are each k.
kl, for

G K,

is called

a scalar matrix;

it is

a diagonal matrix whose

ALGEBRA OF SQUARE MATRICES


two matrices can be added or multiplied. However, if we only consider square matrices of some given order n, then this inconvenience disappears. Specifically, the operations of addition, multiplication, scalar multiplication, and transpose can be performed on any nxn matrices and the result is again an n x n matrix.
Recall that not every

In particular,

if

is

any n-square matrix, we can form powers of A:


A^

= AA, A^ =
in the

A^A,

..

and A"

=/

We can also form polynomials

matrix A: for any polynomial

f{x)

ao

+ ai* +

UiX^

ttnX"

44

MATRICES
aj

[CHAP.

where the

are scalars,

we

define f(A) to be the matrix

/(A)

aol

aiA

a2A^

+ aA"

In the case that f{A)

is

the zero matrix, then

is called

a zero or root of the polynomial f{x).

Example

3.13:

Let

A = (J _l);
If f(x)

then

A^ =

(J

J)(J

_^2)

= (_^ "^

2a;2

- 3a; + 5,

then

If g{x)

x^

+ 3x- 10,

then

''^'

(J

^:) -

Ka

-I) -

<

:)

Thus

is

a zero of the polynomial g(x).

INVERTIBLE MATRICES A square matrix A is said


that

to be invertible if there exists a

matrix

with the property

AB = BA = I where / is the identity matrix. Such a matrix B is unique; for ABi - BiA = / and AB2 = B2A = I implies Bi = BJ - BiiABz) = iBiA)Bi = IB2 = B2 We call such a matrix B the inverse of A and denote it by A~*. Observe that the above relation is symmetric; that is, if B is the inverse of A, then A is the inverse of B.
Example
2
3.14: 1

5\/

-5\
2)
5\

3/1^-1
3

_ ~
_ "

/6-5
1^3-3
/

-10 + 10 -5 + 6

1
1

-1
,'2

-5\/2 2j\l
and
/
|

6-5
2

15-15
-5 +
6

s)

V-2 +

5\

3
^

Thus

-5\
2

3/"""\^-i

are invertible and are inverses of each other.

(Problem 3.37) that for square matrices, AB = I if and only if BA = /; hence it is necessary to test only one product to determine whether two given matrices are inverses, as in the next example.

We show

-11

+ + 12
48

2 4

Example

3.15:

(2-1

3)(-4

l|=|-22 + 4 + 18
-44-4 +
invertible

+ 0-2 2 + 0-2 + 0-3 4-1-3 8 + 0-8 8 + 1-8


other.

Thus the two matrices are

and are inverses of each

We now calculate X, y, z, w such that


a

the inverse of a general

2x2

matrix

fa A
{

b\
1

We

seek scalars

''

cd)\zwj~\0l)

^\( ^

y\ _ /l

0\
^^

fax + bz \cx + dz

ay
cy

+ bw\ + dwj

_ /l " \0

CHAP.

3]

MATRICES
to solving the following

45

which reduces

two systems of linear equations


1

in

two unknowns:

+ bz = \cx + d2 =
iax
If

jay + bw =
\cy

+ dw =

1
if

we

let

|A| if

= ad be,
\A\ = 0;

and only
ad

then by Problem 2.27, page 33, the above systems have solutions such solutions are unique and are as follows:

d
\A\'

_
"

b
ad

_
d/\A\

-be

-be
,

^
\A\'

_
ad

be

z.
\A\'

ad

"'

^
\A\

-be

Accordingly,

..

-b/\A\\
,i^i
i

_ -

'^-c/|A|

0'l\A\J

\ ( d i^ii \A\\^-c

-b"
a

Remark:

The reader no doubt recognizes \A\ = ad bc as the determinant of the matrix A; thus we see that a 2 x 2 matrix has an inverse if and only if its determinant
is

not zero. This relationship, which holds true in general, will be further investigated in Chapter 9 on determinants.

BLOCK MATRICES
Using a system of horizontal and vertical lines, we can partition a matrix A into smaller matrices called bloeks (or: eells) of A. The matrix A is then called a block matrix. Clearly, a given matrix may be divided into blocks in different ways; for example,
1

-2
3
1

3\
1',

-2

1
j

3\
=

/I

-2
3
1

"1
4

'

2 3

7-2
5
9/

=
\s

3
1

7 |-2
5
1

5]7
1

-2
9

4
is

9/

\3

The convenience

of the partition into blocks

can be obtained by carrying out the computation with the blocks, just as actual elements of the matrices. This is illustrated below.

that the result of operations on block matrices if they were the

Suppose

is

partitioned into blocks; say

Ain
Ain

Multiplying each block by a scalar

k, multiplies

each element of
.

by

k; thus

(kAii.
feAai

kAi2

kAm
kA2n
iCAmn j

&A22
rCAjn2

iCAjnl

Now suppose

a matrix

is

partitioned into the


j

same number of blocks as A; say


.
.

Bn
B21

B12 B22

Bin

B =

B2n
B,

\^Bml

Bm2

...

46

MATRICES

[CHAP. 3

Furthermore, suppose the corresponding blocks of A and B have the same size. Adding these corresponding blocks, adds the corresponding elements of A and B. Accordingly,
/All
A
I

I>

-^21
I

+ fill + -^21 + Bm\

Ai2 A22

+ Bi2 + B22

...
. .

Am + Em
Aln

Bin

\Am\

Am2 + Bm2

Amn + Bm
true.

The case of matrix multiplication

U and V are
^

is less obvious but partitioned into blocks as follows

still

That

is,

suppose matrices

U12
JJ

... ...

C/ip\

/Fa
^^^

7i2 V22

... ...

Vin\
V2:

-Zl
I

C/22

U2P

V2I

Vmi

...

Umpj
is

\Vj,l

F22

...

Fpn/

such that the number of columns of each block Uik block Vkj. Then

equal to the

number

of rows of each

'Wn
W-n

Wn
W22

...

Wm
Wzn

...

where

Wa = UnVn +

Ui2V2i

UipVpj
It

The proof
is left

of the above formula for UV is straightforward, but detailed and lengthy. as a supplementary problem (Problem 3.68).

Solved Problems

MATRIX ADDITION AND SCALAR MULTIPLICATION


3.1.

Compute:
2

-3
1

4^\

/3 -5

6
-

-1\
-3y
2
(iii)

-{I
^^
(i)

-5

-1

I2
5\

-2

[1 -1
-

1)

/3

-3
6

-{I
-

-5

Add

corresponding entries:

n 2\Q -5
I

-5

-0

(I
3

-2 -^)

/I

2-5
-5 +

-3

(0 +
(ii)

1-2

4-- l^ -1 --3y

4-333
2

-5 -1 -4

The sum

is

not defined since the matrices have different shapes.

(iii)

Multiply each entry in the matrix by the scalar 3:

0/1
^
'

2
-

-3N
'

'

_ -

/
'

-3
-12

-6

15-18

CHAP.

3]

MATRICES

47

3^.

Let

A = 72
.

-5

(3

0-4)'^ = (0-1
/6 -15

1\

/I

-2 -3\

5J'^

= (l-l-lj-F^"<i 3A + 4B-2C.
/

/O

-2\

First perform the scalar multiplication, and then the matrix addition:

^ . ^ 3A + 4B - 2C
,

3\

= (^

-12) + (0 -4

/4 -8 -12\ 2o}

-2
2

4\

/lO -25

-5\
lo)

+ (-2

2)

( 7

-2

3.3.

Fmda;,i/,zandwif

fxy\
^

/
\

1
/

2w
x

+y
3

\z

+w
+
3 6

First write each side as a single matrix:


/3a;

3y\

\3z

BwJ

_ ~

\z +

w-l
2a;

2w +

Set corresponding entries equal to each other to obtain the system of four equations,

The

solution

is:

2,

j/

= +4 3y = X + y + 6 3z = z + w 1 3w = 2w + 3 = 4, z = 1, w = 3.
3as

or

= 4 = 6+x 2y 2z = w 1 w = 3

3.4.

Prove Theorem

3.1(v):

Let

and

be

mxn

matrices and k a scalar.

Then

kiA+B) = kA + kB. Suppose A (Ojj) and B


is

the v-entry of k(A

tively

and so

fcay

Then Oy + 6jj is the y-entry of A + B, and so &(ajj + 6^) (bij). +B). On the other hand, ka^j and fcfty are the ij-entries of kA and kB respecfe6y is the ti-entry of kA + kB. But k, ay and & are scalars in a field; hence
k(aij

+ 6jj) =

fcfflij

kbij,

for every

i,

Thus k(A
Remark:

+ B)
7.

= kA +

kB, as corresponding entries are equal.

Observe the similarity of this proof and the proof of Theorem l.l(v) in Problem 1.6, page In fact, all other sections in the above theorem are proven in the same way as the corresponding sections of Theorem 1.1.

MATRIX MULTIPLICATION
3.5.

Let
if
(i)

(r

s)

denote a matrix with shape rxs.


is defined:
(iii)

Find the shape of the following products


(v)
(vi)

the product

(2x3)(3x4)

(1 (5

2)(3 2)(2

x x

1)

(3 (2

x
x

4)(3 2)(2

x
x

4) 4)

(ii)

(4xl)(lx2)

(iv)

3)

matrix are multipliable only when p = q, and then X p matrix and a Recall that an X n matrix. Thus each of the above products is defined if the "inner" numbers the product is an are equal, and then the product will have the shape of the "outer" numbers in the given order.

qXn

(i)

The product The product

is
is is is

a 2 X 4 matrix, a 4

(ii)

2 matrix.

(iii)

The product
The product The product The product

not defined since the inner numbers 2 and 3 are not equal. a 5

(iv)

3 matrix.

(v)
(vi)

is

not defined even though the matrices have the same shape.

is

a 2 X 4 matrix,

48

MATRICES

[CHAP. 3

3.6.

Let

^ =
Since

(2 _!)
is

and

^ ^
2

(3

-2

"

^^^^

^^^

^^'
is

^"^

^^'
To obtain the /2\ /

(i)

and

is

3,

the product

AB

is

defined and

a 2

3 matrix.

entries in the first

row

of

AB, multiply

the first

row

(1,

3)

of

by the columns

_4S
and
(

x3y'V-2

of B, respectively:

S\/2
-1
j(^ 3

0-4\
-2
6 y

_ ~

/1-2
V
2

+ 3-3 1-0 +
9

3- (-2)

(-4)

+
-6

0-6

-4 + 18\

/ll

14

To obtain the

entries in the second

row

of

AB, multiply
11

the second

row

(2,

1) of

by the

columns of B, respectively:
1

-1

y2 A
3

-4
-2

ey

^ ~

-6
2

14

V2-2 +

(-l)-3

(-1)

(-2)

(-4)

(-1)

6/

Thus
Note that

^^
J5 is

( 1

2-14

(ii)

and

is

2.

Since the inner numbers 3 and 2 are not equal, the product

BA

is

not defined.

3.7.

Given

A =

(2,1)

and

B =

/^

5-3)' *"^
AB

^^^

^^'

^"^

^^*

(i)

is defined and is a 1 X 3 matrix, i.e. a row Since A is 1 X 2 and B is 2 X 3, the product vector with 3 components. To obtain the components of AB, multiply the row of A by each column of B:

AB = (%,!)(
(ii)

\ 4

"!
5

)
/

=
2.

(2

4,

(-2)

5,

(-3))

(6, 1,

-3)

Note that

is

and

is 1

Since the inner numbers 3 and 1 are not equal, the product

BA

is

not defined.

3.8.

Given

A =
A
2
\ -3

and

B =

[^

^V
^

find

(i)

AB,

(ii)

BA.

(i)

Since
first

is 3

row

of

X 2 and B is AB, multiply


\
,

2 X 3, the product is defined and is a 3 X 3 matrix. the first row of A by each column of B, respectively:

AB

To obtain the

-1
"

/2-3 -4-4

-10

0\

1-1

-8

-10^

( 3

"o )
of

=
AB, multiply
/

To obtain the second row


respectively:
/

the second

row

of

by each column of B,

2
1

-1
i

\
1(

1-2-5 ^ ^
^

-1
1

-8 -2 +

-10 -5 +

\
I

/-I

11

-8 -2

-10^

-5

-6

3
/

To obtain the

third

row of AB, multiply the

third

row

of

by each column of B, respectively:

CHAP.

3]

MATRICES
-1 -8 -2
12
6

49

-10

1 1
i

"(s -:-:)
Thus

=
-3

-5
15

-8 -2
22

-10 -5
15

16

AB
5
is

(ii)

Since
first

row

2 of

3 and

i4 is

BA, multiply

3 X 2, the product BA is defined and is a 2 X 2 matrix. the first row of B by each column of A, respectively:

To obtain the

2-2 +

15

-1 +

0-20

15

-21

To obtain the second row of BA, multiply the second row of


-2
4

by each column of A, /15 (lO

respectively:

-5
)l

'

"
6

15

-21

-21

-3 +
-21 > "-1)

-3

Thus
Remark:

BA
Observe that in this case both AB and do not even have the same shape.

'15
.10

BA

are defined, but they are not equal; in fact they

3.9.

Let

A =

2
1

-1

/I

o\

and

B =

2
\4

-4 -1

-1
0,

-2

(i)

:;th
(i)

Determine the ahaite of AB. (ii) Let Ca denote the element column of the product matrix AJB, that is, AB = (co). Find:
Since

in the ith
c^a,

row and
C21.

Cu and

is

and

is

4,

the product

AB

is

a 2

4 matrix.
ith column of B.

(ii)

Now

Cy is defined as the product of the ith

row of A by the

Hence:

(-3)

(-2)

c,4

(2,-1,0)

2-1 + (-l)'(-l) + 0-0

C21

(1, 0,

-3)

(-3) -4

0-12 =

-11

1
3.10.

6\/4
5/(2 -1
6 5 2
(iii)

Compute:

(i)

-3
1
(ii)

-^(-l
(3,2)

(V)

(2,-1)

-6

-3
(i)

-7
and the second
is

(iv)

The

first

factor is 2

2,

so the product is defined

and

is

a 2

2 matrix:

50

MATRICES
1

[CHAP. 3 l-0 + 6'(-l) \ + 5-(-l)y

^-3
(ii)

6Y4 5A2
is

0\

-1/
X 2 and
1

l'4 + 6-2 V(-3)-4 + 5.2


/

(-3)-0

_ ~
and

/ 16

-6'
-5^

[-2
is

The

first

factor

the second is 2

1,

so the product is defined

a 2

matrix:

^V

2\

1-2

6- (-7)

-3
(iii)

5A-7;

\(-3)'2
is

+ 5'(-7))

_ ^

/-40'

[-41^

Now

the first factor

is is

distinct, the

product

2 X 1 and the second not defined. 2

2X2.

Since the inner numbers 1 and 2 are

(iv)

Here the first factor matrix:

is

and the second

is 1

2,

so the product is defined

and

is

a 2X 2

{>''
(v)

=
is

il--l
2

i:i)

2^

18

12

The first factor is 1 X 2 and the second which we frequently write as a scalar.
(2,-l)(^_g)

1,

so the product is defined

and

is

a 1

matrix

(2-1

(-1). (-6))

(8)

3.11.

Prove Theorem
Let

3.2(i):

(AB)C = A{BC).
and

A =

(oy),

B=

(bfl,)

C= +

(e^).

Furthermore,

let

AB = S =
=
n

(sj^)

and

BC = T =

(t,,).

Then
Sjfc

= =
i.e.

ajiftifc

at2b2k

+ +

at6mfc

2
3=1

Oyftjj.

hi

^ji'^ii

bjiCn

bjci

=
ith

2 lc=l
n

fcjfcCfci

Now

multiplying
is

S by
ilCll

C,

(AB) by C, the element in the

row and

Ith

column of the matrix

{AB)C

SJ2C21

+ A +

+
T,

SiCi

k=l
i.e.

StfcCfcl

=22
fc=l

{"'ifijkiOkl

j=l

On the other hand, multiplying of the matrix A{BC) is


il*ll

by

by BC, the element


tn

in the tth
n

row and

fth

column

i2*2!

aim*ml
is

ij*jl

=22

(6jfcCfci)

Since the above sums are equal, the theorem

proven.

3.12.

Prove Theorem
Let

3.2(ii):

A=

(tty),

A{B + C) = AB + AC. B = (6jfc) and C = (Cj^). Furthermore,


Then
djk

let

D = B + C=
m

(dj^),

E = AB =

(ej^)

and

F = AC =

(fik)-

=
= =

6jfc

Cjfc

e*

aii6ik

ai2*'2fc

+ +

ajm^mk

2 j=i
m

ij6jic

/ifc

Ojl^lfc

+
m

<i2<'2fc

"vm^mk

2 =

"ijCjfc

Hence the element

in the ith

row and
fik

feth

column of the matrix


tn

AB + AC
m
j=l

is

ik

i=l

ayftjfc

3=1
fcth

ttyCj-fc

i)(6jic

+ c^k)

On

the other hand, the element in the ith


Ojidifc

row and

column of the matrix

AD = A(B + C)

is

m
o-ad-jk

aisd^k

otmdmk

}=l

i=l

a.ij(bjk

+ Cjk)

Thus A{B + C)

AB + AC

since the corresponding elements are equal.

CHAP.

3]

MATRICES

51

TRANSPOSE
3.13.

Find the transpose A* of the matrix

A =

Rewrite the rows of

as the columns of A':

3.14.

Let

be an arbitrary matrix.

Under what

conditions is the product

AA*

defined?

Suppose A is an X n matrix; then A* is n X m. Observe that A*A is also defined. Here AA* is an X

Thus the product AA*


matrix, whereas

is

w m

A*A

is

always defined. an n X n matrix.

3.15.

Let

^ =
(

_!

4)

Find

(i)

AA\

(ii)

A*A.
'1

31

To obtain

A*, rewrite the rows of

A
3-

as columns:

A*

Then

^^

A*A

/I

-'

:)(: -:
1-3 + 2'(-l) + 0'4 \ 3'3 + (-1) (-!) + 4-4/

l'l + 2'2 + 0'0 3'1 + (-1)'2 + 4-0

_ ~

/5 \1

26

-i (3
1-1

-I

:) l'2 + 3'(-l)
2

3-3

l'0
2

(-1) '3

(-1)

(-1)

0l + 4'3

0-2 + 4' (-1)

+ 3-4 \ + (-1) 4 +4 4

3.16.

Prove Theorem
Let

8.3(iv):

{AB)*
(bj^).

= B*AK
Then the element
in the ith

A=

(oy)

and

B=

row and

jth column of the matrix


(1)

AB

is

anbij

ai^hzj

ai^h^j

Thus

(1) is

the element which appears in the jth

row and

ith

column of the transpose matrix (AB)*.

On

the other hand, the jth

row

of B* consists of the elements

from the jth column of B:


(2)

(6

bzj

...

6j)

Furthermore, the tth column of A* consists of the elements from the ith row of A:

(3)

Consequently, the element appearing in the ;th row and ith column of the matrix B*A* product of (2) by (S) which gives (1). Thus (AB)* = B*A*.

is

the

52

MATRICES

[CHAP.

ECHELON MATRICES AND ELEMENTARY ROW OPERATIONS


3.17.

Circle the distinguished elements in each of the following echelon matrices.

Which

are

row reduced echelon matrices?


/l

-3
5

l\

/O

-5

0\

2-4,
7
3/ \0
first

1,
0/
nonzero entries in the rows; hence

\0

The distinguished elements are the

[l)2-3

0l\
3/

/07-5
\0

0(1)2 -4,
An
echelon matrix
is

00,00
0/

0\

/0O5O2^
20
4 7,

\0

row reduced

if its

entries in their respective columns.


first is not.

distinguished elements are each 1 and are the only nonzero Thus the second and third matrices are row reduced, but the

/I
3.18.

Given

A =

2
\3

-2 -1
1
i.e.

3 -1\ 2 2 2
3/
.

(i)

Reduce

to echelon form,

(ii)

Reduce

to

row

canonical form,
(i)

to

row reduced echelon form.

Apply the operations ^2 -* -2i?i + Rz and R^ iJj -> -7B2 + 3B3 to reduce A to echelon form:

-ZRi
'1

^3.

and then the operation

-2

-1^
4

Ato|0 3-4 4|to|0 3-4


^0
7
ii)

-lOy
iSj -^

Method and Ri

1.

Apply the operation B,


4/?3

2i?2

+
(i)

7i?2

to the last

matrix in

ZRi, and then the operations to further reduce A: IZX 45^ 21


I

-^3 +

7i?i

to

3-4

4
I

to

0-12
7

-10
J

Finally, multiply Bj

by

1/21,

R^ by 1/21 and
'1

fig

by 1/7

to obtain the

row canonical form

of A:

10
1

15/7^

-4/7
-10/7
J

^0

Method 2. In the last matrix in (i), multiply R^ by 1/3 and matrix where the distinguished elements are each 1:

fig

by 1/7

to obtain

an echelon

'1-2
1

-1
4/3

-4/3
1

^0

-10/7/
R2, -

Now
jBj -^

apply the operation R^ - 2^2 + Ru and then the operations {1/3)R3 + jBi to obtain the above row canonical form of A.

(4/3)fi3

+ R^

and

Remark:

Observe that one advantage of the until the very last step.

first

method

is

that fractions did not appear

CHAP.

3]

MATRICES
/O
1

53 3 -2\
3

3.19.

Determine the row canonical form of

A =
I2
1

2 \2

1-4
3

2 -ll

1^

-4

Z\

to

Note that the third matrix

is

already in echelon form.

3 -4\
1

3.20.

Reduce
i.e.

A = -4
\
1

-6

to echelon form,

and then to row reduced echelon form,

2-5/
if

to its

row canonical form.


the "pivotal" element
is 1.

first

The computations are usually simpler and third rows:

Hence

first

interchange the

to

Note that the third matrix

is

already in echelon form.

3.21.

Show
of the

that each of the following elementary

row operations has an

inverse operation

same

type.
jth.

m
(i)

Interchange the ith row and the

row: Ri <^ Rj.


k:

Multiply the zth row by a nonzero scalar

Ri

kRi,

fc

^ 0.
^
is,

Replace the ith row by k times the jth row plus the ith row: Ri
Interchanging the same two rows twice,
is its
(ii)

kRj

+ Ru

we

obtain the original matrix; that

this operation

own

inverse.

Multiplying the ith row by k and then by fc-i, or by fc-i and then by k, we obtain the original matrix. In other words, the operations iJj - kRi and i?j -^ fe-iiJj are inverses.
flj - kRj + Ri and then the operation fij -^ kRj + fij, or applying the operation fij -* -kRj + i?j and then the operation fij - kRj + flj, we obtain the original matrix. In other words, the operations Ri - kRj + fij and iJj - kRj + flj are

(iii)

Applying the operation

inverses.

SQUARE MATRICES
3.22.

Let

A =
A^

^ -3^ ^4 _g

Find
1

(i)

A^

(ii)

A*,

(iii)

/(A),

where

fix)

2a^

4x

5.

(i)

= AA =
/

-3 )il
1-1

4)
1-2 + 2 -(-3) \ 4-2 + (-3) -(-3)/

+ 2-4

V4-l + (-3)-4

^ ~

-4\
17/

[-8

54

MATRICES

[CHAP. 3

(iii)

-'

c
+

-:)(-: ) / l-9 + 2-(-8) (^4-9 + (-3) -(-8)

l-(-4)
4 -(-4)

2-17

(-3) -17/

/-7 \eO

30\

-67 J

To
f(x)

find /(A), first substitute

for x and 57 for the constant 5 in the given polynomial

2x9

4a;

5:

/(A)

= 2A3-4A +

5/

2(-; -Z) -

{\

-s) + K'o

l)

Then multiply each matrix by

its

respective scalar:

/-14
1^120

60\

-4

-8\
12/

/5

0\ 5y

-134y

V-16

\0

Lastly, add the corresponding elements in the matrices:

_ "

/ -14 -4 + 1^120-16 +

60-8 +
-134

_
I'

-IS
104

52 \

12

+ 5/

-117/

3.23.

Referring to Problem 3.22, show that

is

a zero of the polynomial g{x)

a;^

+ 2a! - 11.
/(A),

is

a zero of g(x)

i.e. first

substitute

if the matrix g(A) is the zero matrix. for x and 11/ for the constant 11 in g(x)

Compute g(A) as was done for

x^

+ ^x- 11:

,.,

= .....

-n,

(-:.)-(! 4)
it:

-"GO
,

Then multiply each matrix by the

scalar preceding

g{A)

-4X _4x
17;

/2

4\
<

/-ll
V

V-8

V8

-( 6/

-11

Lastly, add the corresponding elements in the matrices:

g{A)
Since g{A)

/9 + 2-11 -4 + 4 + 0\ ^ ^_g^8^.Q i7_g_;^iy

/O
(^0

0,

is

a zero of the polynomial g(x).

3.24.

Given

A -

Find a nonzero column vector u

such that

Au -

3m.

First set

up the matrix equation

Au =
4

3u:

-3/U/
vector):

~ ^

^\y

Write each side as a single matrix (column


/

+ 3y\ \Ax-3yJ
x

/Sx^
V3j/y

equations (and reduce to Set corresponding elements equal to each other to obtain the system of echelon form):

+ 3J, = Ax-Zy a;

3a;

__

2x

Zy

Ax-

3y 6y

= =

2x

Sy

0-0

^^

2x

Sy

=
number
2
is

of solutions.

The system reduces to one homogeneous equation in two unknowns, and so has an To obtain a nonzero solution let, say, ?/ = 2; then = 3. That is, a;
solution of the system.

infinite

3,

i/

Thus the vector w

is nonzero g j

and has the property that

Au =

3m.

CHAP.

3]

MATRICES
/3
f

55

3.25.

Find the inverse of


Method
1.

l2

5^

We
,2

seek scalars

x, y, z

and

for which
*""

3/\2

w/

" \0 + +
5

l)

\2x +
(

Sz

2y

+ 3w)

GO
is
2/

or which satisfy

,.,..

r3a; <

l2a;

3
is

= =
a;

and
3, z

[31/

\2j/

+ 5w = + 3w =

The solution of the

first

system

=
1

2,

and of the second system

5,

w=

3.

Thus the inverse of the given matrix


Method
2.

/-3
is

5\
)

We

derived the general formula for the inverse

A-^

oi the

2X2

6c

matrix

A^i
Ttaslf

rTT
1^1

V-c
9-1(1

a'

where

lAI

ad

A=Q

ly

,h.

|A|

= -1 .d A-. =

-l(4

-^) =

(-J

4).

MISCELLANEOUS PROBLEMS 3.26, Compute AB using block multiplication,


2
I

where

1\

3
\0

and

2,

Hence

^1 GJ

and

S =

\0

TJ

where E, F, G, R,

S and T

are the given blocks.

//9

12 15N

/3N

/I

AB

ER ES + FT\ (^^^ ^^^+/^)


GT
J

= ~

jVl9 26 33/ Vl9 33y


\
(

V^yVoyj \0 V^/
(2)

ji9

26

33

0)

3.27.

Suppose
fined.

B=
A2,
.

{Ri, R2,

.,

i?),

i.e.

Show
i4i,

that
.

BA =
.
.

(RiA, RzA,
,

that Ri is the ith row of B. Suppose BA is de.,RnA), i.e. that RiA is the ith row of BA.
. . . .

A" denote the columns of A. By definition of matrix multiplication, the ith row of BA is {Ri A\Ri'A\ But by matrix multiplication, BjA = (Bj A^, i?i A2, i2i A"). Bj-A). Thus the ith row of BA is ftjA.

Let

.,

3.28.

Let Ci = (0, 1, where. Show that CjA


. . . , .
.

of

be the row vector with 1 in the tth position and the ith row of A. Observe that Cj is the ith row of /, the identity matrix. By the preceding problem, the IA is BjA. But lA = A. Accordingly, CjA = JBj, the ith row of A.
. ,

0)

else-

= Ri,

tth

row

3.29.

Show:

(i)

If If

(ii)

A has a zero row, then AB has a zero row. B has a zero column, then AB has a zero column.
a zero row or a zero column
. .

(iii)

Any matrix with

is

not invertible.
ith

(i)

Let

jBj

be the zero row of A, and B^,

.," the columns of B.


...,

Then the
..., 0)

row of

AB

is

{RrB\ Ri'B^

Ri-B^)

(0, 0,

56

MATRICES
Let Cj be the zero column of B, and Aj,

[CHAP. 3

(ii)

.,

the rows of A.

Then the

jth

column of

AB

is

/Ai-C/
A^'Cj

m'Cj
(iii)

matrix A is invertible means that there exists a matrix A~^ such that AA"^ = A~^A I. But the identity matrix / has no zero row or zero column; hence by (i) and (ii) A cannot have a zero row or a zero column. In other words, a matrix with a zero row or a zero column cannot
be invertible.

3.30.

Let

AB
An"*

is also

be invertible matrices (of the same order). Show that the product and (AB)-^ = B'^A'K Thus by induction, (AiA2- -An^^ = -Az^Ai^ where the Ai are invertible.

and

invertible

(AB)(B-iA-i)
and

= A(BB-i)A-i =

A/A-i

= AA

/ I

(B-iA-i)(AB)

= B-i(A-iA)B = B-^B = B^B =

Thus (AB)-i

= B-iA-i.

3.31.

Let u and v be distinct vectors.

Show

that, for each scalar

kGK,

the vectors

u + k{u v)

are distinct.
show that
(1)

It suffices to

if

u
then
fcj

ki{u

v) =
or
fci

k2(u

v)

(1)

k^.

Suppose

holds.

Then
k^iu

ki(u v) =

v)

{ki

k2)(u v) =
and
fci

Since

u and v are

distinct,

u v'0. Hence

fcg

/Cj.

ELEMENTARY MATRICES AND APPLICATIONS* 3.32. A matrix obtained from the identity matrix by a single
called

an elementary matrix. sponding to the operations Ri <^ R2, Ra

elementary row operation is Determine the 8-square elementary matrices corre-

^ IRs and
/I

722 -*

3i?i +
to obtain

R2.

o\
1

Apply the operations to the identity matrix /g

=
\o

1/

/I
^1

^2

=
\o

Eo

-7

3.33.

Prove: Let e be an elementary row operation and E the corresponding m-square elementary matrix, i.e. E-e(l-m). Then for any TO X % matrix A, e{A) = EA. That is, the result e(A) of applying the operation e on the matrix A can be obtained by multiplying A by the corresponding elementary matrix E.

is

Let iJj be the tth row of A; we denote this by writing A = (B^ a matrix for which AB is defined, then AB = (R^B, ..., R-^B).
ej

R^).

By Problem

3.27, if

We
i

also let

(0,

...,0,1,0
first

0),

A =

*This section is rather detailed and may be omitted in a results in Chapter 9 on determinants.

reading.

It is

not needed except for certain

CHAP.

3]

MATRICES
=
.
.

57

Here a
/

means that
e)
is

(cj,

.,

1 is the ith component. the identity matrix.


i^j ->

By Problem

3.28,

e^A

iJj.

We

also

remark that

(i)

Let

be the elementary row operation

Rj.

Then, for a
Bj
ej,
. .

=
.,

and A

j,

E and
e(A)

e(I)

=
(iBj,

(ej

ej

.,^,
.

.,

Rt

BJ
/s

Thus

^A =
(ii)

(fijA,

/\ ...,e^, ...,6iA, ...,e^A)

A
.
.

(fii,

., i?,-,

., ffj

fij

e(A)

Now

let e

be the elementary row operation

jB^ -> fcjBj,

fc

t^ 0.

Then, for

a =

i,

E =
Thus
(iii)

e(/)

(6i, ...,fcej, ...,

ej

and

e(A)
(fij,
. .

=
/\
,

(ftj,

., fcfij,

.,

BJ

^-A

/\
(fijA, ...,A;ejA,

...,e^A)

=
JBj

fefij,

=
/\

e(A)

Lastly, let e be the elementary

row operation

kRj

Kj.
=:

Then, for
(fij,
. . .

i,

E =
Using
(ftej

e(I)

(ei, ...,fcej

+ ej, ...,6j
BjA

and
Rf,

e(A)

fcfij

+ Bj,

i2

ej)A

fc(ej.A)

kRj

we have

EA = (M,

...,(fce^

+ ei)A,

...,eA)

(R^,

.,

kRj + Ri,

.,

RJ =

e(A)

Thus we have proven the theorem.

3^. Show

row equivalent to B if and only if there exist elementary matrices E2E1A = B. By definition, A is row equivalent to B if there exist elementary row operations ej, ..e, for which es(---(e2(ei(A)))- ) = B. But, by the preceding problem, the above holds if and only if Eg- -E^EiA = B where is the elementary matrix corresponding to e^.
that
is

El, ...,Es such that Es-

JS7j

3^5.

Show

that the elementary matrices are invertible and that their inverses are also elementary matrices.
Let E be the elementary matrix corresponding to the elementary row operation e: e(I) = E. Let e' be the inverse operation of e (see Problem 3.21) and E' its corresponding elementary matrix. Then, by Problem 3.33,
/

e'(e(/))

e'E

= E'E

and

e(e'(I))

eE'

= EE'

Therefore E'

is

the inverse of E.

3M. Prove
(i)

that the following are equivalent:

(ii)

(iii)

A is invertible. A is row equivalent to the identity matrix /. A is a product of elementary matrices.

Suppose A is invertible and suppose A is row equivalent to the row reduced echelon matrix B. Then there exist elementary matrices Ei,E2, -yE^ such that Eg- E2E1A = B. Since A is invertible and each elementary matrix E^ is invertible, the product is invertible. But if B ^ I, then B has a zero row (Problem 3.47); hence B is not invertible (Problem 3.29). Thus B = I. In other
words,
(i)

implies
if
(ii)

(ii).
.

Now

holds, then there exist elementary matrices E^, E^,

.,Eg such that

E.-'-E^EiA

= =

/,

and so

A =
then

(E,-

-E^Ei)-^

= E^^E^^-'-EJ^
Thus
(ii)

By

the preceding problem, the Ei

are also elementary matrices.


-E^),
(i)

implies

(iii).

Now
matrices

if

(iii)

holds

(A

EiE^-

must follow

since the product of invertible

is invertible.

58

MATRICES
Let

[CHAP. 3

3.37.

A and B be square B = A-K Thus AB = I


Suppose
is

is

matrices of the same order. if and only if BA = I. invertible. Then A is not row equivalent not

Show

that

if

AB = I,
/,

then

to the identity matrix

and so

row equivalent to a matrix with a zero row. In other words, there exist elementary matrices E^ such that E^- -E^E^A has a zero row. Hence E^- -EJE^AB has a zero row. Accordingly, El AB is row equivalent to a matrix with a zero row and so is not row equivalent to /. But this contradicts the fact that AB = /. Thus A is invertible. Consequently,

B = IB =
Suppose

(A'-->^A)B

= A-HAB) = A'^I = A"! row


.

1.38.

reducible to the identity matrix / by the (i) Show that this sequence of elemen., e. sequence of elementary operations ei, (ii) Use this result to obtain the inverse tary row operations applied to / yields A-K

is

invertible and, say,

it is

/I

2\

of

A =

2 \4

-1
1

3
8/

(i)

hypothesis and Let Ei be the elementary matrix corresponding to the operation ej. Then, by i = E^---EJEJ Problem 3.34, E- -E^EiA = I. Thus (E- EiEJ)A ^ I and hence A e -i can be obtained from / by applying the elementary row operations ej, In other words, A

(ii)

Form

the block matrix (A,

I)

and row reduce


2
I

it to

row canonical form:


/l
to 2

1
1

0\
I

(A, /)

[2-1
1

3
I

-1
I

8
1

2
to

-1 -1

-2 -6

to

Observe that the


inverse:

final block

matrix

is in

the form

(/,

B).

Hence

is

invertible

and

is its

A-i

=
B), then the given matrix is not

Remark:

In case the final block matrix is not of the form row equivalent to I and so is not invertible.

(/,

Supplementary Problems
MATRIX OPERATIONS
In Problems 3.39-3.41, let
4

-(o1
3.39.

0-3
2 3

!)
B,
(ii)

-"i-l
A+
C,
(iii)

Find:

(i)

A+
AB,
A*,

(ii)

3A

4B.
(y)

3.40.

Find:
Find:

(i)

AC,

(iii)

AZ),
IJtA',

(W) BC,
(iv)

BD,

(wi)

CD.

3.41.

(i)

(ii)

A'C,

(iii)

BA,

(y)

DW.

{wi)

DDK

CHAP.

3]

MATRICES

59

61
C,

&2
C2

h h],
C3

find

(i)

ejA,

C4/

3.43.

Let
(i)

Cj

(0,

... 0, 1, 0, .... 0)

where

1 is the ith

component.
3.28,

Show
Bj.)

the following:

Be*.

=
e^A

Cj,

the ith column of B.


for each
i,

(By Problem

ejA

(ii)

If

(iii)

If

= ejB = Be\ Ae\

then then

for each

i,

A = B. A = B.

ECHELON MATRICES AND ELEMENTARY ROW OPERATIONS 3.44. Reduce A to echelon form and then to its row canonical form,
/l
(i)

where
5
1^

A =

2 \3

2-1 2 1-2 4 2-6 6

l\
3
,

/2
(ii)

3-2
2 6

A =

5/

-1 \4 -5
3

-5

Ij

3.45.

Reduce

to echelon

form and then


2\
3
1

to its

/l
(i)

3
11

-1 -5
3
1

row canonical form, where /O 1 3 -2^

,..^

2-5
\4
1

(")

4-13 ^=0021
^

5/

\0

-3

4,

3.46.

Describe

all

the possible

2X2

matrices which are in row reduced echelon form.

3.47.

A
3.48.

Suppose A is a square row reduced echelon matrix. Show that has a zero row.

if

A # 7,

the identity matrix, then

Show

that every square echelon matrix

is

upper triangular, but not vice versa.

3.49.

Show
(i)

that
is

row equivalence
to

is

an equivalence

relation:

(ii)

(iii)

A A A

row equivalent
to to

A;
implies

row equivalent row equivalent

B B

row equivalent
equivalent to

to

A;

and

B row

Implies

row equivalent

to C.

SQUARE MATRICES
3.50.

"Let

g(x)

A = ( = a;2 - a;

\
8,

(i)

Find A^ and A3,

(ii)

If

/()

vfl

Zx^

2x

i,

find /(A),

(iii)

If

find g(A).

3.51.

Let
(iii)

B =

].

(i)U

f(x)

2x2

4x
(

+
]

Z,

find f{B).

(ii)

If

g{x)

x^

4x

12,

find g(B).

Find a nonzero column vector

u =

such that

Bu =

6m.

3.52.

Matrices A and /I 1\ , with


. (

are said to

commute

if

AB =

BA. Find

all

matrices

Vz

w/

which commute

VO

3.53.

Let

A =

<ll)
(
)

Find A".

60

MATRICES

[CHAP.

3.54.

Let

A =
(i)

\0

3/
B,

and
(ii)

B =

\0

,,

11
(iv)

Find:

A+

AB,

(iii)

A^ and A3,

A",

(v)

/(A) for a polynomial f{x).

3.56.

Suppose the 2-square matrix

commutes with every 2-square matrix A,


i.e.

i.e.

AB =

BA. Show that

B =
3.57.

\0

kj

for some scalar k,

is

a scalar matrix.

Let
(i)

Dfc

be the m-square scalar matrix with diagonal elements

k.

Show

that:

for any

mXn matrix A,

D^A =

kA;

(ii)

for any
of:

nXm matrix B,

BD^ =

kB.

3.58.

Show
(i)

that the sum, product and scalar multiple


is

upper triangular matrices


lower triangular matrices
diagonal matrices
scalar matrices
is

upper triangular;
lower triangular;

(ii)
(iii)

is

diagonal;

(iv)

is scalar.

INVERTIBLE MATRICES
3.59.

(i)

-'

'2
)
>

-3\
3/

Find the inverse of each matrix:

(")

-1
3.60.

2
1

-3\
,

2
(ii)

-V
1

Find the inverse of each matrix:

(i)

2 \5
2

,4-2
1
t.61.

5/

-3/

4\ 6 1/
is,

Find the inverse of

-1
5

l-l
3.62.

Show

particular,

that the operations of inverse and transpose commute; that A is invertible if and only if A* is invertible.
...

(A)-i

(A-i).

Thus,

(!
.

."!

///.

.M
if

invertible,

and what

is its

inverse?

3.64.

Show
Show

that

A A

is

row equivalent

to

if

and only

there exists an invertible matrix

such that

B = PA.
3.65.

that

is invertible if

and only

if

the system

AX =

has only the zero solution.

MISCELLANEOUS PROBLEMS
3.66.

Prove Theorem
(Parts
(i)

3.2:

(iii)

(B

+ QA = BA + CA;
in

(iv)

k(AB)

=
A;

(kA)B

A(kB),

where

fc

is

a scalar.

and

(ii)

were proven
(i)

Problem 3.11 and

3.12.)

3.67.

Prove Theorem 3.3: (Part (iv) was proven

(A

B)*

A*

+ BH

(ii)

(A')

(iii)

(feA)'

kA*,

for

k a scalar.

in

Problem

3.16.)

3.68.

defined and the number of Suppose A = (A^) and B = (B^,) are block matrices for which AB is of rows of each block B^j. Show that AB - (Gy) columns of each block Aj^ is equal to the number Ag^B^j. where Cy =

CHAP.

3]

MATRICES
called elementary

61

3.69.

The following operations are


[El]

column operations:

Interchange the tth column and the jth column.


Multiply the tth column by a nonzero scalar
A;.

[^3]

Replace the ith column by k times the jth column plus the ith column.
that each of the operations has an inverse operation of the same type.

Show
3.70.

matrix A is said to be equivalent to a matrix can be obtained from A by a finite sequence of operations, each being an elementary row or column operation. Show that matrix equivalence is an equivalence relation.

BUB

3.71.

Show that two consistent systems of linear equations have the same solution set if and only if their augmented matrices are row equivalent. (We assume that zero rows are added so that both augmented matrices have the same number of rows.)

Answers
3.39.

to

Supplementary Problems
-13 -3
(iii)

(i)

-1

i-l
Not

1)

(ii)

Not

defined.

f
4

18 \

17

0/

3.40.

defined.

<"

C)

<^'

CI
Not

/-5
(")

( 11

-2 4 -3 -12

')

<">

m
(iii)

":

1)
/

(vi)

1
3.41.
(i)
I

0\
3
I

-7
12

4\
(v)

-2

-1
2

(ii)

Not

defined.

(9, 9)

(iv)

0-6-8
\-3
6/

14

(vi)
\

-2
6

1-3
-3
9y

4/

3.42.

(i)

(ai, ag. 03. a*)

(ii)

(61, 62.

K K)
and

(iii)

(Ci. Cg, Cg, C4)

'1
3.44.
(i)
(

2
1

4/3^

3-6 3-2
-11
10
5

^0

-1/6^
5/11

/2
(ii)

l\
5

/l

4/11
1

13/11

-15

and

-10/11

15/11

-5/11
/

\0

0/

\0

/I

3 11

3.45.

(i)

r''-
^0

-1 -5

2 3 "

ll

4/11
1

13/11
3/11

and

-5/11

0/

/o

-2
11

o\
1 1

(ii)

-13
:

and

35
0/

0/

62

MATRICES

[CHAP.

'''

Co

o)'(2

D'Co

-^^'^

'^

'^

I) ''Co

J)

^"^

'=^'^"

^0
3.48.
(

1 1

V
1
)

Is

upper triangular but not an echelon matrix.

iO

1/

3.52.

Only matrices of the form

commute with (^

^J

/I
3.53.

2n\

^" = (o

i)
/9

3.54.

^+^

= (o
/14

14

('^^

- c =
2"

:)
3,

- c =

.;)

'" '<^'

(T

;,;

ON

()^^=(o

33;

^"=U

/3ci

3d,^

3.59.

(1)

(^
1-5

-2\
3)

(n)

,..,

/
f

1/3 ^/9

1/3

2/g

-3\
6
11

8
-5

-1 -3^

3.60.

(i)

10-7

12

8-6

5/

10

-1 -4y

/31/2 -17/2 -11^


3.61.

9/2

-5/2
4

-3
51

\-7

3.62.

Given

AA-i =

/.

Then

7'

(AA-i)'

= {A'^YAK
/a-i

That

is,

(A^^)'

(A*)"!.

3.63.

is

invertible

iff

each

aj 9^ 0.

Then

.0
'

a-i

...

\0

chapter 4

Vector Spaces and Subspaces


INTRODUCTION
In Chapter 1
ties.

Now
(i)

we studied the concrete structures B" and C" and derived various propercertain of these properties will play the role of axioms as we define abstract
it is

In particular, the conclu[Afi]-[M4] below. 1.1, 3, We will see that, in a certain sense, we get nothing new. In fact, we prove in Chapter 5 that every vector space over R which has "finite dimension" (defined there) can be identified with R" for some n.
called, "linear spaces".

"vector spaces" or, as

sometimes

sions

through

(viii)

of

Theorem

page

become axioms

[A]]-[A4],

The definition of a vector space involves an arbitrary field (see Appendix B) whose elements are called scalars. We adopt the following notation (unless otherwise stated or
implied):

K
a, &, c

the field of scalars, the elements of K, the given vector space, the elements of V.

or

A;

V
u, V,

We

remark that nothing essential or the complex field C.

is lost if

the reader assumes that

is

the real field

Lastly, we mention that the "dot product", and related notions such as orthogonality, not considered as part of the fundamental vector space structure, but as an additional structure which may or may not be introduced. Such spaces shall be investigated in the latter part of the text.
is

Definition :

scalar multiplication which assigns to

be a nonempty set with rules of addition and and to any u,v a sum u + v any uGV,kGK a product ku G V. Then V is called a vector space over K (and the elements of V are called vectors) if the following axioms hold:
Let
iiT

be a given

field

and

let

GV

GV

[Ai]: [A2]:

For any vectors u,v,w GV,

{u

+ v)

+w
and

= u+

{v-i-

w).

There is a vector in V, denoted by for any vector u GV.

called the zero vector, for

which u

u
0.

[A3]
[A4]:

For each vector

uGV
G

there

is

a vector in V, denoted by u, for which u


u.

+ {u) =

For any vectors u,v GV, For any


scalar k

u+v = v +

[Ml]:

[M2]

[Ms]:

[Mi]:

+ v) = ku + (a + b)u = au + For any scalars a,b GK and any vector {ab)u = a{bu). For any scalars a,b G K and any vector For the unit scalar 1 G K, lu = u for any vector u GV.
and any vectors u,v

GV, u GV, u GV,

k{u

kv. bu.

63

64

VECTOR SPACES AND SUBSPACES

[CHAP.

The above axioms naturally split into two sets. The first four are only concerned with the additive structure of V and can be summarized by saying that 7 is a commutative group (see Appendix B) under addition. It follows that any sum of vectors of the form
Vi

V2

+ Vm

requires no parenthesis and does not depend upon the order of the summands, the zero is unique, the negative u of u is unique, and the cancellation law holds: vector

u +
for any vectors

w =

+w

implies

u v
by

u,v,w G V.

Also, subtraction is defined

uV = u

{v)

the the other hand, the remaining four axioms are concerned with the "action" of of the axioms reflects this splitting. Using these on V. Observe that the labelling field space. additional axioms we prove (Problem 4.1) the following simple properties of a vector

On

Theorem

4.1:

Let
(i)

be a vector space over a

field

K.
7,
fcO

For any scalar

kGK

and

0.

(ii)

For

(iii)

(iv)

uGV, Ou = 0. = or m = 0. If ku ^ 0, where kGK and uGV, then For any scalar kGK and any vector uGV, {-k)u = k{-u) =
and any vector
A;

gK

-ku.

EXAMPLES OF VECTOR SPACES We now list a number of important


generalization of the space R".
Examplje
4.1:

examples of vector spaces.

The

first

example

is

of be an arbitrary field. The set of all n-tuples of elements Let addition and scalar multiplication defined by

with vector

(!,

a2

a)

(61,62, ...,6)
.

(01

+ 61,02+62

a+6)

and

fc(ai. <2.

n)

= =

C^^i- '2.

^O

where

<, 64,

k&K,

is

a vector space over K;

we

vector in K is the w-tuple of zeros, space is identical to the proof of Theorem 1.1, that R" with the operations defined there is a vector space over R.
,

(0, 0, ...

denote this space by X". The zero The proof that K" is a vector 0). which we may now regard as stating

Example

4.2:

Let

is

X n matrices with entries from an arbitrary field K. Then be the set of all with respect to the operations of matrix addition and vector space over a

scalar multiplication, by

Theorem

3.1.

Example

4.3:

+ at" with coefficienis oj polynomials Oo + a^t + Ogt^ + with respect to the usual operations is a vector space over from a of addition of polynomials and multiplication by a constant.
Let

V be

the set of

all

field

K.

Then

Example

4.4:

be any nonempty set. be an arbitrary field and let Let into K. The sum of any two functions functions from

f,g

Consider the set V of all is the function

eV

+ gGV

defined by
{f

+ g){x) =

f(x)

g(x)

and the product of a scalar defined by

kEK

and a function / e

is

the function

kfeV

(kf){x)

kf(x)

CHAP.

4]

VECTOR SPACES AND SUBSPACES


Then

65

a vector space over K (Problem 4.5). The zero which maps each x G X into S K: 0{x) = x G X. Furthermore, for any function f G V, f is that function in V for which (/)() = f(x), for every x G X.

with the above operations

is

vector in for every

is

the zero function

Example 45:

Suppose S is a field which contains a subfield K. Then E can be considered to be a vector space over K, taking the usual addition in to be the vector addition and defining the scalar product kv of and v S jF to be the product of k and v as element of the field E. Thus the complex field C is a vector space over the real field E, and the real field R is a vector space over the rational field Q.

kGK

SUBSPACES
Let TF be a subset of a vector space over a field K. is called a subspace of V if TF is a vector space over K with respect to the operations of vector addition and scalar multiplication on V. Simple criteria for identifying subspaces follow.
itself

Theorem

4.2:

(i)

(ii)

(iii)

W is& subspace of V and only W nonempty, W closed under vector addition: v,w G W implies v + w G W, W closed under scalar multiplication: v GW implies kv GW
if
if

is

is

is

for

every
Corollary

kGK.
V if and only
if
(i)

4.3:

W
4.6:

ia

a subspace of

implies av

+ bw G

GW (or W # 0),
GK.

and

(ii)

v,w

GW

for every a,b

Example

Let

V be

any vector space.

also the entire space

Then the set {0} consisting of the zero vector alone, and are subspaces of V.
Then the
set
:

Example

4.7:

(i)

Let

V V

be the vector space R^.


is

third component
(ii)

zero,
all

W consisting of those vectors whose


GR},
is

{{a,b,0)

a,b

a subspace of V.
4.2).

Let
set

be the space of

square

nX n

matrices (see Example

Then the
Ojj,

consisting of those matrices A symmetric matrices, is a subspace of V.


(iii)

(oy)

for which

ay

called

Let V be the space of polynomials (see Example 4.3). Then the set of polynomials with degree n, for a fixed n, is a subspace of V. Let

W consisting

(iy)

be the space of
set

all

Then the

consisting of

functions from a nonempty set all bounded functions in


if

(A function / every x G X.)

is

bounded

there exists

M GR

is a subspace of V. for such that |/(a;)| -

X V

into the real field R.

Example

4.8:

Consider any homogeneous system of linear equations in n unknowns with, say, real
coefficients:

aiiXi a2iXi

a-y^Xi

4-

+ +

aia; a2a;

a^sx^

be viewed as a point in R". a subspace of R" (Problem We comment that the solution set of a nonhomo4.16) called the solution space. geneous system of linear equations in n unknowns is not a subspace of R".
Recall that
solution of the system

The

set

W of

any particular
all

may

solutions of the

homogeneous system

is

66

VECTOR SPACES AND SUBSPACES


Example
4.9:

[CHAP. 4

be subspaces of a vector space V. We show that the intersection are subsince U and G C/ and S a subspace of V. Clearly and u,v e UdW. Now suppose m,v e.Ur\W. Then u,v spaces; whence are subspaces, and, since U and

Let

and

Vr\W

i& also

W
e

&U

&W

aw
for any scalars space of V.

6i)

?7

and

aw

6v

W
and so [7nTF
is

a,bK.

Accordingly, au

bv

&

UnW

a sub-

The

result in the preceding

example generalizes as follows.

Theorem

4.4:

The

intersection of

any number of subspaces of a vector space

is

subspace of V.

LINEAR COMBINATIONS, LINEAR SPANS Let F be a vector space over a field K and let
form
aiVi

vi,

...,VmGV.

Any

vector in

of the

a2V2 4-

+ amVm
The following theorem

where the OiGK,


applies.

is called

a linear combination of vi,...,Vm.

Theorem

4.5:

be a nonempty subset of V. The set of all linear combinations of vectors in S, denoted by L{S), is a subspace of V containing S. Furtheris any other subspace of V containing S, then L{S) CW. more, if

Let

In other words, L{S) is the smallest subspace of V containing S; hence it subspace spanned or generated by S. For convenience, we define L{0) = {0}.
Example
4.10:

is called

the

Let

be the vector space R3. The linear span of any nonzero vector u consists scalar multiples of u; geometrically, it is the line through the origin and the of point u. The linear space of any two vectors u and v which are not multiples of each other is the plane through the origin and the points u and v.

all

Example

4.11:

The vectors 6i = (1,0,0), eg = (0,1,0) and eg = (0,0,1) generate the vector space specifically, R3. For any vector (a, 6, c) G R^ is a linear combination of the ej;
(a, b. e)

= =

a(l, 0, 0)

aej

+ 6(0, 1, + ftej + 063

0)

c(0, 0, 1)

Example

4.12:

The polynomials
(in*):

1, t, t^, t^, ...


t,

y=
t.

L(l,

t^

.).

For any polynomial

generate the vector space V of all polynomials is a linear combination of 1 and

powers of

CHAP.

4]

VECTOR SPACES AND SUBSPACES


4.13:

67

Example

Determine whether or not the vector v - (3, 9, -4, -2) the vectors u^ = (1, -2, 0, 3), U2 == (2, 3, 0, -1) and Wg to the space spanned by the Mj.
V

is

(2,

a linear combination of -1, 2, 1), i.e. belongs

Set r as a linear combination of the Mj using unknowns XU^ + J/M2 + ZM3:


(3, 9,

x,

y and
-1,

z;

that

is,

set

-4, -2)

= =

!B(1,

-2,

0, 3)

i/(2, 3, 0,

-1)
2,

(2,
3/

2, 1)

(x

+ 2y + 2z, -2x + 3y-z,

3a; + z)

the equivalent system of equations by setting corresponding components equal to each other, and then reduce to echelon form:

Form
X

2y
3j/

-2x +

+ +

2z
2

=
=:=

3
9

X
or

2y
7y

+ +

2z
3z

22
3a;

= -4 = -2

2z

-7y
a;

+ +

5z

= = = =

x
or

2y 7y

15

+ +

2z 3z 22

-4
-11

-2z

- 3 - 15 = -4 = 4

2|/

2z
Bz 22

or

7y

= 3 - 15 = -4

Note that the above system is consistent and so has a solution; hence v is a linear combination of the Mj. Solving for the unknowns we obtain x = 1, y = 3, z 2. Thus V Ui + 3m2 2M3. Note that
tion,
if the system of linear equations were not consistent, i.e. had no soluthen the vector v would not be a linear combination of the Mj.

ROW SPACE OF A MATRIX Let A be an arbitrary mxn matrix over a field K:


(Hi
(I22

...
. .

ai
a,2n

\fflml

flm2

dmn/
dmn)

The rows

of A,

Rl viewed as vectors in

(ttll, 0,21,

.,

am),

Rm =

(Oml, am2,

.K",

span a subspace of K" called the row space of A.

That

is,

row space

of

L{Ri, R2,

Rm)

Analogously, the columns of A, viewed as vectors in K"", span a subspace of column space of A.

X"

called the

Now

suppose
(i)

we apply an elementary row


Ri <^ Rj,
(ii)

operation on A,
or
(iii)

Ri

kRi,

k'0,

Ri

->

kRj

+ Ri

and obtain a matrix B. Then each row of B is clearly a row of A or a linear combination of rows of A. Hence the row space of B is contained in the row space of A. On the other hand, we can apply the inverse elementary row operation on B and obtain A; hence the row space of A is contained in the row space of B. Accordingly, A and B have the same row
space.

This leads us to the following theorem.


4.6:

Theorem

Row

equivalent matrices have the

same row

space.

We shall prove (Problem 4.31), in particular, the following fundamental result concerning row reduced echelon matrices.

68

VECTOR SPACES AND SUBSPACES


4.7:

[CHAP. 4

Theorem

reduced echelon matrices have the same row space have the same nonzero rows.

Row

if

and only

if

they

Thus every matrix row canonical form.

is

row equivalent

to a unique

row reduced echelon matrix

called its

We

apply the above results in the next example.


4.14:

Example

Show

that the space


Ml

generated by the vectors


3),

(1, 2,

-1,

M2

(2, 4, 1,

-2),

and

wg

(3, 6, 3,

-7)

and the space

generated by the vectors


vi

(1, 2,

-4, 11)

and

v^

(2, 4,

-5, 14)

are equal; that

is,

U=

V.

Method

Show that each Mj is a linear combination of v^ and V2, and show that 1. each Vi is a linear combination of Mj, M2 and M3. Observe that we have to show that six systems of linear equations are consistent.
Method
2.

Form

the matrix

whose rows are the

Mj,

and row reduce

to

row

canonical form:
1

-1
3

3\

-1
3

2
I

-2
I

to

6
1

-8 -16/
1/3
"^

to

-8

\o

2
1

to

-8/3
/

Now

form the matrix

whose rows are Vi and 11\


/I
to

t>2,

and row reduce


/I
to
(j,

to

row canonical
1/3
1

form:

_ = ~

/I ,x
(

\2

2-4 4-5

-4 2 2-4 UN
3
o

14/

( VO

_o) -8/

_8/3

Since the nonzero rows of the reduced matrices are identical, the row spaces of and B are equal and so U = V.

SUMS AND DIRECT SUMS


Let

and

consists of all

Note that
u'

W be subspaces of a vector space V. The sum of U and W, written U + W, sums u + w where uGU and w &W: U + W = {u + w:uGU,wGW} = + eU + W, since OeU.OGW. Furthermore, suppose u + w and
+ W, with u,u' GU and w,w' e W. Then (u + w) + (u' + w') = {u + u') + {w + w') G U +
k,

+ w'

belong \joU

and, for any scalar

k{u

+ w) = ku + kw G U +

W
F
is also

Thus we have proven the following theorem.

Theorem

4.8:

The sum
4.15:

U+

of the subspaces

and TF of

a subspace of V.

Example

U consist of those Let V be the vector space of 2 by 2 matrices over R. Let V consist of those matrices matrices in V whose second row is zero, and let whose second column is zero:

- =

{(::)^' }

{(::)-'}

CHAP.

4]

VECTOR SPACES AND SUBSPACES


Now U
and

69

W are subspaces of V.
{(" o)

We

have:

U+W

"'^'"^A

*"d

VnW

[(I )

aeR
Ur\W

That is, W^ consists of those matrices whose lower right entry is 0, consists of those matrices whose second row and second column are zero.
Definition:

U+

and

The vector space denoted by


if

is

said to be the direct

sum

of its subspaces

and W,

V = V
GF

w
way
as v

every vector v

where

u&V

can be written in one and only one

= u+w

and

w gW.

The following theorem

applies.

Theorem

4.9:

The vector space


if:

is

the direct
(ii)

sum

of its subspaces
{0}.

{i)V

U+W,
U =

and
let

UnW =
6

and

if

and only

Example

4.16:

In the vector space R^,

U
:

be the xy plane and


a,

let

W be the yz plane:
{(0,

{{a, 6, 0)

S R}

and

b,c):

h,c& R}

Then R^ = since every vector in R3 is the sum in W. However, R* is not the direct sum of U and
unique; for example,
(3, 5, 7)

U+W
=

of a vector in

W
=

and a vector

since such

sums are not

(3, 1, 0)

(0, 4, 7)

and also

(3, 5, 7)

(3,

-4,

0)

(0, 9, 7)

Example

4,17:

In R3, let

be the xy plane and

U
Now any
vector in

=:

{(o, 6, 0):
(a, b, c)

W be the z axis: a,6GR} and W =


let

{(0, 0, c)

G R}
in

vector

in one

G R^ can be written as the sum of a vector and only one way:


{a, b, c)

and a

(a, 6, 0)

(0, 0, c)
is,

Accordingly, R3

is

the direct

sum

of

and W, that

R^

= U

W.

Solved Problems

VECTOR SPACES
4.1.

Prove Theorem
(i)

4.1:

Let

F be

a vector space over a

field

K.

For any scalar For


If

kGK

and

GV,

fcO

0.

(ii)

GK

0,

and any vector

(iii)

ku

where

kGK

(iv)

For any
By axiom
fee

kGK
[A^]

and any

uGV, Ou = 0. = or u = and uGV, then uGV, {-k)u = k{-u) = - ku.


fc

0.

(i)

fcO.

Adding

with m = 0, we have + = 0. Hence by axiom [Mi], kO to both sides gives the desired result.

fcO

fc(0

0)

(ii)

By

a property of K,

0.

Hence by axiom

[Mg],

Om

(0

+ 0)m =

Qu

Ou.

Adding

- Om

to both sides yields the required result.

70

VECTOR SPACES AND SUBSPACES


(iii)

[CHAP.

Suppose

fcw

and k

= 0.

Then there

exists a scalar fc^i such that


fe-iQ

fc~ifc

1;

hence

u = lu = {k-^k)u = k-Hku) =
(iv)

=
k{-u).

Using u

{-u)

0,

we
0,

obtain

kO

k{u

(-m))

few

Adding -ku
Adding -ku

to both

sides gives

ku

k(u).

Using k
sides yields

(-fe)

we

obtain

ku =

{k)u.

Thus (k)u

= Oit = + (-k))u = = k(u) = ku.


(fe

ku

(-k)u.

to

both

4.2.

Show
{k(v)

that for any scalar k and any vectors


definition

u and

v,

k{u-v) =

ku-

kv.
4.1(iv)

Using the

of subtraction

{u-v = u+ (-v))
k(-v)

and the result of Theorem

kv),
k(u

-v) =

k(u

+ (-v)) = ku +

ku

(-kv)

= ku -

kv

4.3.

In the statement of axiom [Mz], (a sign represent?


The

b)u

au

bu,

which operation does each plus

addition of the two scalars a and 6; hence it represents the addiin the field K. On the other hand, the + in au+ bu denotes the addition of the two tion operation represents a vectors au and bu; hence it represents the operation of vector addition. Thus each

in

(a+b)u denotes the

different operation.

4.4.

In the statement of axiom [Ma], represent?

(ab)u

a{bu),

which operation does each product


field

In (ab)u the product ab of the scalars a and 6 denotes multiplication in the product of the scalar ab and the vector u denotes scalar multiplication.

K, whereas the

multiplication; In a{bu) the product bu of the scalar 6 and the vector u denotes scalar product of the scalar a and the vector bu denotes scalar multiplication.

also,

the

4.5.

Let

V be the set of all functions from a nonempty set X into a field K. For any funcin V defined tions f.gGV and any scalar k G K, let f + g and kf be the functions
as follows:
{f

+ 9){x) -

fix)

g{x)

and

(kf){x)

=
is

kf(x),

yfx

(The symbol
Since space hold.
[All-

means "for

every".)

Prove that

a vector space over K.


all

X is nonempty, V is
+

also

nonempty.

We now

need to show that

the axioms of a vector

Let the function Now, a; e X.


((f

f.g.hGV. To show
(f

g)

that (f + g) + h and the function f + (g

= f + (g + h), it is necessary to show + h) both assign the same value to


(fix)

that each

+ g) + h)(x) = (f+(g + h))(x) =


But
f(x), g(x)

if

+ g){x) +
+
(g

h{x)

g{x))
(g(x)

+ +

h(x),
h(x)),
is

Vo;

f(x)

+ h)(x) =

f(x)

yfxGX
associative; hence

and

h(x) are scalars in the field


(f(x)

where addition of scalars

g(x))

h(x)

f(x)

(g(x)

h(x))

Accordingly,

(f

+ g) +

+(g + h).
0(a;)

[AJ:

Let

denote the zero function:


(/

= =

0, Va;

G X. Then

for any function


Vo;

V,

+ 0)(a;)
is

f(x)

0(a!)

f(x)

f(x),

Thus /

/,

and

the zero vector in V.

CHAP.

4]

VECTOR SPACES AND SUBSPACES


[A3]:

71

For any function / G V,


(/

let

-/ be

the function defined by


f(x)

(-/)()
Oix),

(-/))()

f(x)

(-/)(*)

f(x)

= - f(x). Then, yfx&X

Hence /
[AJ:

(-/)

0.

Let

f.g^V.
(/

Then
f(x)

+ ffKx) =
g

gix)
/(*)

g(x)

+
g(x)

f(x)

(g

+ f)(x),

y/x&X
/()

Hence f

f.

(Note that

g(x)

g(x) are scalars in the field


[Ml]:

K where

f{x)

follows

from the fact that

and

addition is commutative.)

Let

f,g&V and k & K. Then + 9))i.x) = k((f + g)(x)) =


=
(kf)(x)

k(f(x)

g(x))

kf(x)

(kg)(x)

(kf

+ kg)(x),

+ kg(x) ^fxeX

Hence k(f + g) = kf + kg. (Note that k(f(x) + g{x)) = kf(x) + kg(x) follows from the fact that k, f(x) and g(x) are scalars in the field K where multiplication is distributive over addition.)
[M2]:

Let

/ey

and

o, 6

6 X. Then
(a+h)f(x)
(af+hf)(x),

((a+6)/)(a;)

= =

af(x)

+ hfi^x) VaseX

(af)(x)

6/(a;)

Hence

(a

6)/

af

bf.

[Mg]:

Let

f&V
a(6/).

and

a, 6

X.

Then,
a(6/(a;))

({ab)f)(x)

(a6)/(x)

o(6/)(a;)

(a(6/))(a;),

Va;

;f

Hence

(ab)f

[AfJ:

Let /
all

y.

Then, for the unit

leK,

(!/)()

l/(a;)

f{x),

V G

X.

Hence 1/

/.

Since

the axioms are satisfied,

is

a vector space over K.

4.6.

Let

V be the set of ordered pairs of real numbers: V = {{a,b): a,bGR}. Show that V is not a vector space over R with respect to each of the following operations of addition in V and scalar multiplication on V:
(i)

(a, b)

(ii)

(a, 6)

(iii) (a,

6)

+ (c, d) = + (c, d) = + (c, d) =


= =

(a

+ c,b + d) and
and
6
k{a, b)

k{a, b)

{ka, b);

(a, 6)

(fee,

kb);

(o

+ c,

+ d) and
Then
(r

fe(a, 6) := (fc^a, fe^ft).

In each case show that one of the axioms of a vector space does not hold.
(i)

Let r

l, 8

2,

=
rv

(3, 4).

+ s)v = +

3(3,4)

(9,4)

+
sv,

sv

1(3, 4)

2(3, 4)

(3, 4)

(6, 4)

(9, 8)

Since

(r

+ s)v
(1,2),

rv

axiom [M^] does not

hold.

(ii)

Let

0)

w=

(3,4).

Then

+w w+v =
v

(1, 2) (3, 4)

+ +

(3, 4)

(1,2)

= =

(1, 2)

(3,4)

Since v

+w
1,

w+
2,
i;

v,

axiom [AJ does not

hold.

(iii)

Let r

=
+

(3, 4).

Then
(r

+ s)v = +

3(3, 4)

=
4)

(27, 36)

rv

SV
sv,

1(3, 4)

2(3, 4)

(3,

(12, 16)

(15, 20)

Thus

{r

+ s)v

rv

and so axiom [M2] does not

hold.

72

VECTOR SPACES AND SUBSPACES

[CHAP. 4

SUBSPACES
4.7.

Prove Theorem 4.2:


implies v
Suppose

Wis + wGW, and

a.

(iii)

subspace of V if and only implies kv v

if (i)

GW

GW

is

nonempty,
(ii)

(ii)

v,w

eW

for every scalar


and
(iii),

kGK.
the operations

Moreover, the axioms [A,], [AJ, of vector addition and scalar multiplication are well defined for W. belong to V. Hence we need only show since the vectors in [Ml], [Ma], [Mg] and [MJ hold in Then by (iii), Ou - S is nonempty, say that [A2] and [A3] also hold in W. By (i), then (-l)v = -v Lastly, it v G satisfies [Ag]. and v + = v for every v G W. Hence is a subspace of V. = 0; hence satisfies [A3]. Thus and V + (-v)

W satisfies

(i), (ii)

and

(iii).

By

(i),

is

nonempty; and by

uGW.
(iii)

W W

Conversely,

if

TF

is

a subspace of

then clearly

(i),

(ii)

and

hold.

4.8.

Prove Corollary
implies av

4.3:

+ bw
=
lv

GW
W
is

Suppose by (ii), v + w

W satisfies
if

e TF and (ii) a subspace of V if and only if (i) for all scalars a,b GK. Then, by (i), W is nonempty. Furthermore, if v,w G (i) and (ii).
ia

v,wGW
W
then,

+ lweW;

and

if

v&W

and

kGK
(i)

then,

by

(ii),

kv

kv

+ Ove W. Thus

by Theorem

4.2, Tf^ is

a subspace of V. a subspace of

Conversely,

then clearly

and

(ii)

hold in

W.

4.9.

Let
(i)

y = R^ Show

that

W
G

is

a subspace of
i.e.

where:

w = {(a, b,0):
W
=
(c,

a,b

R},

is

the xy plane consisting of those vectors whose

third component
(ii)

is 0;

{{a,b,c): a

+b+c =
sum

0},

i.e.

consists of those vectors each with the


is zero.

property that the


(i)

of its components

(0,

0,0)

d, 0) in

is 0. component of W, and any scalars (real numbers) k and k',

since the third

For any vectors

1;

(a, 6, 0),

w =

kv

k'w

= =

k(a, b, 0)
(ka, kb, 0)

+ k'(c, d, 0) + (fc'c, k'd,

0)

(ka

+ k'c,

kb

+ k'd,

0)

Thus kv
(ii)
tt

+ k'w e W,

and so

is

a subspace of V.

= (0, 0,0) +6+c =

GW

since
a' +

and

6'

+ + = 0. Suppose v = (a, b, c), w = (a', b', e') + C = 0. Then for any scalars k and k', kv + k'w = k(a, b, c) + k'(a', b', c') = (ka, kb, kc) + {k'a', k'b', k'c') = (ka + k'a', kb + k'b', kc + k'c')
(kb

belong to

W,

i.e.

and furthermore,
(ka

+ k'a') +

+ k'b') +

(kc

+ k'c')

= =

k(a+
fcO

+ c) + + fc'O =
b

k'{a'

b'

+ e')

Thus kv

+ k'w e W,

and so

is

a subspace of V.

4.10.

Let
(i)

V = R^ Show that W PF = {{a, b,c): a ^ 0},


nonnegative;
Pf

is

not a subspace of V where: consists of those vectors whose i.e.

first

component

is

(ii)

{(a, b, c): 1;
c)
:

d'

+ b^ + c^^ e Q},

1},

i.e.

W consists of those vectors whose length does

not exceed
(iii)

W = {(a, 6,
,,

a, b, c

i.e.

W consists of those vectors whose components are


of, say.

rational numbers.
In each case, show that one of the properties
(i)

Theorem

4.2 does not hold.

(1,2,3)

GW

and

fc

= -5 e
Hence

R.
is

But

W since -5

is negative.

-5(1,2,3) fc. not a subspace of V.

(-5, -10,-15)

does not belong to

CHAP.

4]

VECTOR SPACES AND SUBSPACES


V = (1, 0,0) belong to v

73

(ii)

eW GW

and
1^

since

w = (0, 1, 0)eW. 1^ + 0^ = 2 > 1.

But v Hence

+w =
Is

(1, 0, 0)

(0, 1, 0)

(1, 1, 0)

does not

not a subspace of V.

(iii)

(1,2,3)
its

W since
4.11.

and k = y/2GK. But fcr = \/2 (1,2,3) components are not rational numbers. Hence

(\/2,
is

2V2, 3\/2) does not belong to not a subspace of V.

Let V be the vector space of is a subspace of V where:


(i)

all

square

nxn

matrices over a

field

K.

Show
(otj)

that

consists of the

symmetric matrices,

i.e.

all

matrices

A=

for which

(ii)

W consists of matrices W= {AgV: AT = TA}.


all

which commute with a given matrix T; that

is,

(i)

OSW

since all entries of

belong to

W,

i.e.

whose ii-entry is and so TF is a subspace of V.


(ii)

are and hence equal. Now suppose A = (ay) and B = (6y) = ay and 5jj = 6y. For any scalars a, 6 G if aA + bB is the matrix aa^ + 66y. But aa^j + 66ji = aoy + 66y. Thus aA + 6B is also symmetric,
ftjj
,

OeW

since

or =

TO.

Now
(aA)T
T(aA)
T,

suppose

A,BgW;
= =
a(AT)

that

is,

AT - TA

and

BT =
b(TB)

TB. For

any scalars a,b G K,


{aA

+ bB)T = =

+ +
i.e.

{bB)T
T{hB)

r(aA

+ b(BT) = + 5B)

a{TA)

Thus aA

+ 6B

commutes with

belongs to

W; hence

is

a subspace of V.

4.12.

Let V be the vector space of is not a subspace of V where:


(i)

all

2 x 2 matrices over the real

field

R.

Show

that

(ii)

W consists of W consists of
(Recall that

all all

matrices with zero determinant;

matrices

for which A^
be.)

= A.
A =
f
)

(i)

det(

^ = ad
=
Hence
i

The matrices

and

B =

belong

to

W since

det(A)
1.

and det(B)

0.

But

A+B =

does not belong to

W since

det (A

+ B) =

W
(

is

not a subspace of V.

(ii)

The unit matrix

= <ll)

, )

belongs to

W since

But 2/

i'l

X
W

n
since

/1 (' Vo

")
1

/2
(

0\
I

does not belong to

4
^

2/

Hence

W is not a

subspace of V.

4.13.

Let
is
(i)

V be the vector space of a subspace of V where:

all

functions from the real

field

into R.

Show

that

(ii)

w = {f: W = {f:

/(3) /(7)

= 0}, i.e. W consists of = /(!)}, i.e. W consists


1;
i.e.

those functions which

map

3 into 0;

of those functions which assign the

same

value to 7 and
(iii)

W consists of the odd functions,

those functions / for which /(-)

= - /()

74

VECTOR SPACES AND SUBSPACES


Here
(i)

[CHAP.

denotes the zero function:


since
0(3)
b,

0(a;)

0,

for every
i.e.

&

R.

OeW

0.

Suppose
(af

f.gGW,
=
(1/(3)

/(3)

=
=
aO

and

sr(3)

0.

Then for any

real

numbers a and
Hence af + bg
(11)

bg)(3)
is

bg{3)

60

& W,

and so TF

a subspace of V.

OeW

since

0(7)

0(1).
b,

Suppose

f.g^W,
bg(l)

I.e.

/(7)

/(I)

and

r(7)

flr(l).

Then, for

any real numbers a and

(af+bg){7)

af(7)
is

a/(l)

6ff(l)

(a/+6fl-)(l)

Hence af + bg
(ill)

& W,

and so

a subspace of V.

OeW

since

0(-a;)

= -0 =
real

-0(a;).

Suppose
b,

f,g&W,
bg{x)

i.e.

/(-x)

-/()

and

g{-x)

g{x).

Then for any

numbers a and
6flr(-a;)

(a/+6f?)(-a!)

=
:

a/(-a;)

+
so

= - af(x) -

= - (a/(x) +

6flr(a;))

= -(af + bg)(x)

Hence af + bg

W, and

is

a subspace of V.

4.14.

Let V be the vector space of all functions from the real is not a subspace of V where:
(i)

field

into R.

Show

that

W={f:

/(7)

+ /(!)};
i.e.

(ii)

consists of all nonnegative functions,

all

function / for which

f{x)

^ 0,

yfxGR.
(i)

Suppose

f.geW, I.e. /(7) = 2 + /(l) and flr(7) = 2 + flr(l). Then = /(7) + fl-C?) = 2 + /(I) + 2 + flr(l) (f + g)i^) = 4 + /(I) + ir(l) = 4 + (/ + flr)(l) ^
Tl' is

(/

+ sf)(l)
/()
Is

Hence f + g^W, and so


(11)

not a subspace of V.
be defined by
/(a)
0.

Let

fc

= -2
But

and

let

GV =

V e

R.

(fc/)(5)

fe/(5)

(-2)(52)

= -50 <

since Then / G Hence kf W, and so


x^.

a;2 s= Q,

not a sub-

space of V.

4.15.

Let

the vector space of polynomials ao + ait ficients, i.e. Oi e R. Determine whether or not VF

V be

+
is

a^t^

af"

a subspace of

with real coefwhere:

(i)

(ii)

(iii)

W consists of W consists of W consists of

all
all all

polynomials with integral coefficients;

polynomials with degree


polynomials &o
t.

3;
+

&it^

h^t^

bnt^",

i.e.

polynomials with

only even powers of


(1)

do not always belong to W. For example, v = No, since scalar multiples of vectors in is "closed" under vector = f + | + 1*^ ^ W. (Observe that but ^i' 3 + 5t + 7(2 e belong to W.) elements in addition, i.e. sums of

(ii)

and

For, in each case, Yes. the scalar multiples of any element in


(iii).

W belong to

is

nonempty, the sum of elements

in

W belong to W, and
Xi,

W.

4.16.

Consider a homogeneous system of linear equations in n unknowns field K: ^ ainXn = n + anXi + ai2X2 +
,

..,Xn over a

aziXl

+ +

022*2

+ +

+ +

a2nX

OmlXl

am2X2

dmnXn

Show that the = (0, 0,


. .

solution set
0)

is

a subspace of the vector space K".

. ,

PF

since, clearly,

a0

ajjO

ttinO

0,

for

1,

.
.

.,m

CHAP.

4]

VECTOR SPACES AND SUBSPACES


=
(!, M2,

75

Suppose M

M)

and v

(vj, Vg,

v)

belong to W,
ttinMn

i.e.

for

I,

.,Tn

il"l OjiVi

+ +

ai2W2
ai2'y2

+ +

+ +

aiv

= =

Let a and 6 be scalars in K. Then


and, for
i

au
1,
. . .

m,

bv

(ciMx

6^1, au2

6^2.

>

<*w

6v)

aji(aMi

+ bvj) + ai2(au2 + 6f 2) + + ain(aMn + bv^) = o(ajiMi + OJ2M2 + + !*) + 6(ajii;i + (ij2i;2 + = aO + 60 =


"

aini'n)

Hence au

+ 6v

is

a solution of the system,

i.e.

belongs to W.

Accordingly,

is

a subspace of K".

LINEAR COMBINATIONS
4.17.

Write the vector v 62 = (1,2, 3) and 63

= =

(1, (2,

2,

5)

as a linear combination of the vectors

ei

(1, 1, 1),

-1,1).

We
we

wish to express v as v
(1,

=
5)

xei

3/62

ze^,

with

x,

y and

as yet

unknown

scalars.

Thus

require

-2,

= = =

+ j/(l, 2, 3) + z(2, -1, 1) (x, X, x) + (y, 2y, 3y) + (22, -z, z) + 2/ + 2z, + 2j/ z, + 32/ + 2)
x{l, 1, 1)
(a;
a;
a;

Form

the equivalent system of equations by setting corresponding components equal to each other,

and then reduce to echelon form:

x+
X
x

y
2y 3y

+ +

2z
z

= 1 = 2 = 5

x
or

y
j/

2y

+ 2z = 1 - 3z = -3 - z =
4,

x
or

+ y -

2z
Sz

= =

S
to obtain

5z=10
Solve for the

Note that the above system is consistent and so has a solution. X = 6, y = B, z 2. Hence v = 6ei + 862 + 263.

unknowns

4.18.

Write the vector v = (2, -5, 3) in R^ as a linear combination of the vectors (1,-3,2), 62 = (2, -4,-1) and 63 = (1,-5, 7).
Set V as a linear combination of the
(2,
Cj

Ci

using the unknowns

x,

y and

z:

xe^

+ j/eg + zeg.

-5,

3)

= =

x{\,

-3,

2)

{x

+ 2y + z,
x

+ z(l, -5, 7) -3x -4y-5z,2x-y + 7z)


y{2,

-4, -1)

Form

the equivalent system of equations and reduce to echelon form:

+ -3x x

2y+z=2
4y y

2y+z-2
2y

x
or

2y

5z Tz

2x

= -5 = 3

or

2z 5z

-5y +

= 1 = -1

2^

+ -

2z

= = =

2
1

The system is inconsistent and so has no solution. bination of the vectors Ci, e^ and 63.

Accordingly, v cannot be written as a linear com-

4.19.

For which value of k


the vectors v Set u = XV

will the vector

u=

=
+

(3, 0,

-2)

and w =
a(3, 0,

(2,

(1, -2, k) in R" be a linear combination of -1, -5) ?

yw:
fe)

(1,

-2,

=
3x

-2)

j/(2,

-1, -5)

(3a;

+ 2y,

-y, -2x

- 5y)

Form

the equivalent system of equations:

By

the first two equations,

+ 2y = 1, = x = 1,
j/

-y =
2.

-2,

-2x - 5y = k
k

Substitute into the last equation to obtain

8.

76

VECTOR SPACES AND SUBSPACES


Write the polynomial v
nomials
ei

[CHAP. 4

4^0.

t^

= t^-2t +
t2

5,

62

+ 4t 3 over R as = 2t^ - St and ca =


ej

a linear combination of the polyt

S.
x,

Set D as a linear combination of the

using the unknowns

y and

z:

xe^

ye^

263.

4t

= = -

a;(t2-2t
a;t2

+ 5) + 3/(22-3t) + 2(f + 3) + 5a; + 22/t2 - s^/t + zt + 3z {x + 2y)fi + {-2x-3y + z)t + (5a; + 3z) 2xt
t

Set coefficients of the same powers of

equal to each other, and reduce the system to echelon form:

+ 2x
x
hx

2y

X
or

2y

=
3z

x
or

2y
2/

=
+
13z

3j/+z=4
+
3z

2/+z=6
-IQy +

z=6
=
52

= -3
is

= -8

Note that the system


a;

-3,

2/

2, z

4.

Thus v

consistent and so has a solution. - -Ssj + ie^ + 463.

Solve for the unknowns to obtain

4.21.

Write the matrix

I
j

as a linear combination of the matrices

A =

;j).-a?)--(:-x
Set

as a linear combination of
3
1

A,B,C

using the unknowns x,y,z:

E xA + yB +

zC.

IN

/I

1\

/O

0\

/O

_iy

^'(i
X

oy +

\i
VJ

1) +

^o

-1
/
\a;

iKN,/0
0/
\y

0N/0 /O
x

2z\ 22 \0-/ \0 -s

X
2/

+ 2z z

Form

the equivalent system of equations by setting corresponding entries equal to each other:
a;

3,

1,

2z

1,

= 1 =
1.
Since these

Substitute a; = 3 in the second and third equations to obtain y = 2 and z values also satisfy the last equation, they form a solution of the system. Hence

E =

BA

2B

C.

4.22.

Suppose m is a linear combination of the vectors linear combination of the vectors Wi, Wn. .

Vi,

.,Vm and suppose each

Vi is

. ,

u =

aiVi

a2V2

+ OmVm

and

Vi

haWi

+ baWi +

biw

Show that u is also a linear combination of the wu Thus if ScL{T), then u = a^Vi + a^v^ + + a^v^ = ai(6iiWi + + b2nWn) + + a^(h.ml1l + + 6iW) + 02(62l"'l + = (diftji + (12621 + + aJi^^Wn + am6mi)wi + + (ai6i + a^h^n +
'

L{S) (ZL{T).

frmn'"')

m
or simply

m
^in

/
(

7n

"S,

i=l

2 i=l

^i

2 \3=1

h'^j

2 3=1

2 \i=l

cuba

Wj

LINEAR SPANS, GENERATORS


4.23.

Show

that the vectors

u=

(1, 2, 3),

(0, 1, 2)
(a, 6, c)

and
is

w=

(0, 0, 1)

generate

W.

We
Set

need to show that an arbitrary vector


(a, 5, c)

S R3

a linear combination of u, v and w.

xu

(a, b, c)

+ =

yv

zw:

x(l, 2, 3)

1/(0, 1, 2)

z(0, 0, 1)

{x,

2x

+ y,Sx + 2y + z)

CHAP.

4]

VECTOR SPACES AND SUBSPACES


of equations

77

Then form the system

= + +
y

a
or

2i/

2x
Zx

=6
+
z

2y

c
is

+ 3a;=:c + 2x = h X = a
x

is

The above system is in echelon form and a solution. Thus u, v and w generate R^.

consistent; in fact

a,

2a,

= e 2b + a

4.24.

Find conditions on a, b and c so that (a, b, c) G u = (2, 1, 0), V = (1, -1, 2) and w = (0, 3, -4).
Set
(a, 6, c)

W
w
-4)

belongs to the space generated by


using unknowns

as a linear combination of u, v and

x,

y and
2y

z:

(a, b, c)

xu

yv

zw.

{a, b, e)

x(2, 1, 0)

j/(l,

-1,

2)

+ = = =

z{0, 3,

=
it to

{2x

+ y,x-y + Sz,
2x

- 4z)

Form
2x
X

the equivalent system of linear equations and reduce

echelon form:

+y y+ 2y -

=a
Zz Az

2x
or

a a
c

=a
^z

b
c

3j/

2y

6z

26

or

Zy

iz

= =

2a

- 2b - 46 -

3c

The vector
consistent,

belongs to the space generated by u, v and w if and only if the above system is consistent if and only if 2a - 46 - 3c = 0. Note, in particular, that u, v and do not generate the whole space R^.
(a, b, c)

and

it is

4.25.

Show

that the xy plane

W = {(a,
(ii)

b, 0)} in
(2,

(1, 2, 0)

and v

=
=

(0, 1, 0);

u=

generated by -1, 0) and v = (1, 3, 0).

R^

is

u and v where:

(i)

u=
v.

In each case show that an arbitrary vector


(i)

(a,

b,0)eW

is

a linear combination of u and

Set

(a, b, 0)

xu

yv:
{a, b, 0)

=
X

x(l, 2, 0)

y(0, 1, 0)

= +
2x
x

(x,

2x

+ y,

0)

Then form the system of equations

2x

The system
(ii)

is consistent; in

fact

= a = b or = = a, y = b-2a

is

a solution.

Hence u and v generate W.

Set

(o, 6, 0)

xu

yv:

{a, 6, 0)

= + x +
2x

x{2,

-1,

0)
it

y(l, 3, 0)

=
+
y

(2x

+ y,-x + 3y,
a a

0)

Form

the following system and reduce

to echelon form:

y
Sy

= = =

a
b

2x
or

7y

= -

2b

The system
that

we

is generated by u and v. (Observe is consistent and so has a solution. Hence do not need to solve for x and y; it is only necessary to know that a solution exists.)

4.26.

Show
a

finite

that the vector space number of vectors.

V of polynomials

over any

field

K cannot be

generated by

L{S) of

Any finite set S of polynomials contains one S cannot contain polynomials of degree

of maximum degree, say m. Then the linear span greater than m. Accordingly, V f^ L{S), for any

finite set S.

78

VECTOR SPACES AND SUBSPACES


Prove Theorem
4.5:

[CHAP. 4

4.27.

Let S be a nonempty subset of V.

Then

L{S), the set of all

linear combinations of vectors in S, is a subspace of

containing S.

Furthermore,

if

is

any other subspace of

containing S, then L{S) C W.


is

If V G S, then Iv = v G L{S); hence S nonempty. Now suppose v,wGL(S); say,

a subset of L{S).

Also, L(S)

is

nonempty

since

is

fflj,

ai^Ui

a^nVm

and

b^Wi

6w

where

Vi,

w^

G S and

5j

are scalars.
a^vi

Then

V
and, for any scalar k,

+ w = is

+ a^v^ +

b^Wi

+ +

6w

kv
belong to L(S) since each
of V.

A;((iii;i

+ a^v^ -

ka^v^

+ ka^v^
is

a linear combination of vectors in S.

Accordingly, L(S)

a subspace

Now

suppose
all

is

a subspace of

V
Oj

containing

multiples a^Vi is, contains

Om'^'m

^ ^>

"where

K,

S and suppose v^, and hence the sum a^v^

.,v^E.

S cW. Then all + a^Vm ^ ^- That


claimed.

linear combinations of elements of S.

Consequently, L(S)

c T^ as

ROW SPACE OF A MATRIX


4.28.

Determine whether the following matrices have the same row space:

Row

reduce each matrix to row canonical form:

r. (^

1 z

z) 13
)

to

(J

\)

to

(;

-1 -2 ; r (s -2 -3
f

to

('

~
1

~
)

to

Vo

3>

/I -1 -l'

V
A
and

-3 -1 -1 3/

to

to

to

6/

Since the nonzero rows of the reduced form of A and of the reduced form of C are the same, C have the same row space. On the other hand, the nonzero rows of the reduced form of B are not the same as the others, and so B has a different row space.

4.29.

Consider an arbitrary matrix A = (a). Suppose u = (&i, ...,&) is a linear combination of the rows Ri, .. .,Rm of A; say u = kiRi + + kmRm. Show that, for each i, bi kiaii + feozi + + kmOmi where an, ., Omi are the entries of the ith column of A.

We

are g:iven

u =
(6i

ftjiJi

k^^R^;

hence
.

K)

= =

fci(an.
(feifflii

Om)

+
>

fem(ami.
^^l^ml

O-mn)

+ fc^Oml)

+ kmO'mn)

Setting corresponding components equal to each other,

we

obtain the desired result.

CHAP.

4]

VECTOR SPACES AND SUBSPACES

79

4.30.

Prove: Let

A=
(&)

(an)

and

let

be an echelon matrix with distinguished entries aij^, a^^, be an echelon matrix with distinguished entries bik^, &2fc2

., ttrj,,

bsk;.

Olj

****** \
a2i.

b-i

4i

:]c

^ ^

ifc

4i

9|i

Osfc

A =
a,v,

Suppose
of

A and B have the same row B are in the same position: ji


Clearly
i4

space.
32 =

fci,

Then the distinguished entries kz, ., jr = kr, and r = s.


. .

of

and

if

and
CiO

1.

We

first

and only show that

if
ij

B = 0, and so we need = k^. Suppose ii < k^.

Since the

first

row of

C2O

cji

A =

is in

the

for scalars
Ji

element a^

# 0.
B

Hence

k^,

row space of B, we Cj. But this contradicts the fact that and similarly fei ij. Thus j'l = fcj.

only prove the theorem when r 1 Then the j^th. column of B is zero. have by the preceding problem, Oi^^ =
the distinguished

Now
row

let

submatrix of
space.

A obtained by deleting the first row of A, and let B' be the obtained by deleting the first row of B. We prove that A' and B' have the same The theorem will then follow by induction since A' and B' are also echelon matrices.
A' be the submatrix of
. . . .

.,a) be any row of A' and let R^, ...,B,n ^^ the rows of B. Since R is in Let R = (!, 02, .,d^ such that R diRi + ^2^2 + + dmRm- Since the row space of B, there exist scalars d^, A is in echelon form and R is not the first row of A, the ^ith entry of R is zero: aj = for Furthermore, since B is in echelon form, all the entries in the fcjth column of B are i = ;j = fej.

except the

first:

61^.^ = 0,

but 62^1

^>

>

^rnkj

0.

Thus

=
Now
6ifc

Ofcj

difeifcj

dgO

d0
.

d,b.

and so

di

0.

Thus

is

a linear combination of R^,

.,Bm and so

is in

the

row

space of B'. Since R was any row of A', the row space of A' is contained in the row space of B'. Similarly, the row space of B' is contained in the row space of A'. Thus A' and B' have the same row space, and so the theorem is proved.

4.31.

Prove Theorem 4.7: Let A = {ckj) and B = (&) be row reduced echelon matrices. Then A and B have the same row space if and only if they have the same nonzero rows.
we
Obviously, if A and B have the same nonzero rows then they have the same row space. only have to prove the converse.

Thus

Suppose
exist scalars

and
.
.

B
c^

have the same row space, and suppose


such that

R
+

is

the ith

row

of A.

Then there

Cj,

. ,

R Cj

CiRi

C2R2

c^Rs
if

W
we show
that

where the Ri are the nonzero rows of B.

The theorem
the

is

proved

R^,

or

but

Cfc

for

A;

t.

Let ay be the distinguished entry in R,


i

i.e.

first

nonzero entry of R.

By

(1)

and Problem

4.29,
(2)

Cl&ljj

262Ji

+ B =

C,b,j.

But by the preceding problem

6y. is

a distinguished entry of

and, since

is

row reduced,

it is

the only nonzero entry in the ijth column of B. Thus from oy = 1 and 6y = 1 since A and B are row reduced; hence Cj

(2) 1.

we

obtain

Oy^

Cjfty..

However,

Now

suppose k

i,

and

b^j.

is

the distinguished entry in R^.


Cibij^

By

(i)

and Problem

4.29,
(S)

a.

"fc

Hb2j^

C,b,j^

80

VECTOR SPACES AND SUBSPACES


Since
"ijfc

[CHAP. 4

is

row reduced,

bj^j^ is

the only nonzero entry in the i^th column of B; hence by


a,.j

(3),

OkHj,^-

Furthermore, by the preceding problem

is

a distinguished entry of
b^j^

and, since

is

row reduced, a^^


is

0.

Thus

c^b^j^

and, since

1,

c,,

0.

Accordingly

R = R^

and the theorem

proved.

4.32.

Determine whether the following matrices have the same column space:
/l

5\

112
,

3^

A =
Observe that
the same

1
\1

4
1

3
9/

B =
\

-2 -3 -4
7

12

17y

row

space.

and B have the same column space if and only if the transposes A* and B* have Thus reduce A' and J5' to row reduced echelon form:
1
1
1

A'

to

-2

to

-2
0/

to

'l

-2
1

B*

-3 \s -4
2

12

to

17/

\0

-2 -4/

to

-2
0/

to

\0

Since A* and B* have the same row space,

and

have the same column space.

4.33.

Let

jR

be a row vector and

linear combination of the


defined,
Bi,

B a matrix for which RB is defined. Show that RB is a rows of B. Furthermore, if A is a matrix for which AB is

Suppose
. . . ,

show that the row space of AB is contained in the row space of B. i? = (aj, ttg, a^) and B = (6y). Let 5i, ...,B^ denote the rows
. . . ,

of

and

its

columns.

Then

RB =
= = =
Thus

{R'B^.R'B^, ...,R'B^)
(aibii

+ 02621 +
.

^-

ttm^ml. ai*12

+ 02*22 +
&2n)

ai(6ii, 612,

. ,

6i)

a^Bi

a252

+ a2(b2i, 622 + amBm

+ am&m2. + am(&ml.

l&ln
6m2.

+ 02*2n ^

1"

am&mn)

bmn)

fijB is

a linear combination of the rows of B, as claimed.


3.27, the

By Problem
result each

rows of
in the

row

of

AB

is

AB are RiB where i?j is the tth row row space of B. Thus the row space of

of A.

AB

is

Hence by the above contained in the row

space of B.

SUMS AND DIRECT SUMS 4.34. Let U and W be subspaces


(i)

of a vector space V.
JJ

Show

that:

(ii)

U and U+W
Let M

W are contained in
\&
[/.

+ W;

linear span of
(i)

the smallest subspace of V containing C/ and PF: C/ + W^ = L(C7, W).

and W, that

is,

U+W

is

the

Accordingly,
(ii)

a subspace of V and so d &W. Hence m = m + OS J7+W. U +W. Similarly, W is contained in U + W. Since 17 + W is a subspace of V (Theorem 4.8) containing both U and it must also contain the linear span of V and W: L{JJ,W) (ZU + W. On the other hand, if v GU +W then v u + w = lu+lw where uGU and w &W\

By

is

hypothesis TF contained in

is

Tl',

hence v

is
(Z

U+

a linear combination of elements in

UuW

and

so

belongs to L{'U,W).

Thus

L(U, W).
inclusion relations give us the required result.

The two

CHAP.

4]

VECTOR SPACES AND SUBSPACES

81

4.35.

Suppose
Let V

and

W are subspaces of a vector space


W.

{Wj) generates

Show
Then v
Mj's;

that {Ui,W}),

i.e.

{Vn}

{Wj),

V, and that {im} generates generates U + W.


?7,

U
tt

and
is

^U +W.

linear combination of

= u-\-w where u G U and w G W. Since {mJ generates and since {Wj} generates W, w is a linear combination of Wj's:
u

ffi^j

+ +

ciiUi

+ +

aMj

ttj

w =
Thus
and so
{mj,

b^Wj^

62WJ2

K'Wj^,

K bj e K
G

= u+w =
i7

ttiMj

a2Ui

a^Mj

biW,

b2Wj

fc^Wj

w^} generates

TF.

4.36.

Prove Theorem 4.9: if and only ii (i) V


Suppose V = where uG U and
(1)

The vector space = U+ and (ii)

is

the direct

sum

of its subspaces

and

W
w

UnW
and

{0}.

U W. w G W.
+

can be uniquely written in the form v = u + Thus, in particular, V = U + W. Now suppose v G UnW. Then:

Then any

1;

GV

where

vGU,
V - U

G W;

(2)

Since such a

sum

for

must be unique, v

0.

Accordingly,

+ v where OGU, UCiW = {0}.

GW

On

the other hand, suppose

there exist

uG U

and

Suppose also that v

w S W such that v = = u' + w' where u' G U u + w = u' + w' and

+W

= {0}. Let vGV. Since V = U + W, and u + w. We need to show that such a sum is unique. and w' G W. Then
so

UnW

u'

=
=

w'

But

u-u'GU

and w'

- w 6 W; hence by u u' 0, w' w = V


is

UnW
C7

{0},

and

so
TF.

u',

w =

w'

Thus such a sum for f G

unique and

V =

4.37.

Let

and PT be the subspaces of R^ defined by

U =
(Note that PF
Note
V
first

{{a,

b,c): a

c}

and

{(0, 6, c)}

is

the yz plane.)
JJnW'

that

=
c

{0},

for

a
i.e.

=
R^

and a

Show that R^ = U W. v = (a, b, c) G UnW implies that = a = 0, 6 = 0, which implies


1;

=
a)

(0, 0, 0).

We also claim that


where
(a, a, a)

C7

and

(,0,b

= U + W. For if = (a, 6, c) S RS, then v = (a, a, - a, c- a) GW. Both conditions, UnW = {0}

and R3

+ (0,b-a,c- a) = U + W,

imply R3

C7

TF.

4.38.

be the Let V be the vector space of ti-square matrices over a field R. Let U and Show that subspaces of symmetric and antisymmetric matrices, respectively. - M*, and anti-symmetric iff is symmetric ifi V = U W. (The matrix

M* = -M.)

We
We

first

show that

claim that

V = U + W. Let A be any arbitrary w-square matrix. A = ^(A + A*) + i(A - At) |(A + A') G 17 and that ^(A - A) G W. For {^{A+At)y = i(A+A)' = i(A + A) = ^{A+A')
is

Note that

that

is, -J^CA

+ A')

symmetric.

Furthermore,

(^(A-At))'
that
is,

= i(A-A')'
Suppose

:-

i(At-A) = -^(A-A'^)
Then

J(A

A') = -M

is

antisymmetric.

We
implies

next show that

or

UnW = {0}. M = Hence


0.

MGUnW.
{0}.

I7nW =

Accordingly,

V= UW.

M = M'

and M^

= -M

which

82

VECTOR SPACES AND SUBSPACES

[CHAP. 4

Supplementary Problems
VECTOR SPACES
4.39.

Let

y be

plication on

the set of infinite sequences V defined by


(!, a2. )

(a^, a^,

.)

in a field

K
+

with addition in

and scalar multi-

(6i, &2. )
.
. .

(ai

6i,
. .

02+
.

62,

.)

k(ai, 02,

(fcaj,

ka2,

where
4.40.

aj, bj,

G K. Show

that

is

a vector space over K.


of real

Let V be the set of ordered pairs on V defined by


(a, 6)

(a, 6)

numbers with addition

in

and scalar multiplication

(c,

d)

= {a+

c,b

+ d)

and

k{a, b)
:

=
lu

{ka, 0)

Show

that

satisfies all of the

axioms of a vector space except [AfJ

u.

Hence

[^4] is not a

consequence of the other axioms.

4.41.

Let V be the set of ordered pairs (a, b) of real numbers. Show that with addition in V and scalar multiplication on V defined by:
(i)

is

not a vector space over

{a,b)
(c, 6) (a, b) (a, 6)

(ii)

(iii)

(iv)

+ + + +

(c,d)
(c, (c, (c,

d) d) d)

= = = =

{a
(a

+ d,b + c) + c,b + d)
and
and

and and
k{a, b)

k(a, b) k(a, b)

= =

(ka, kb);
(a, 6);

(0, 0)

{ka, kb);

(ac, bd)

k{a, b)

(ka, kb).

4.42.

Let

V be

real field

the set of ordered pairs (zi, z^) of complex numbers. Show that R with addition in V and scalar multiplication on V defined by
(zj, Z2)

is

a vector space over the

(wi, W2)

(i

+ Wi,

+ '"'2)

and

k{zi, 22)

{kzi, kz2)

where
4.43.

z^, Z2,

Wi,

W2 ^ C and k GB,.

Let y be a vector space over K, and let F be a subfield of K. Show that V is also a vector space over F where vector addition with respect to F is the same as that with respect to K, and where scalar multiplication by an element k G F is the same as multiplication by k as an element of K.

4.44.

Show

that [A4], page 63, can be derived from the other axioms of a vector space.

4.45.

be vector spaces over a field K. Let V be the set of ordered pairs (u, w) where u Let U and uGU, w G W}. Show that y is a vector space over belongs to U and w to W: V = {{u, w) with addition in V and scalar multiplication on V defined by
:

K
U

(u,

w)

(u',

w')

=
k

(u

+ u',w + w')
K.
(This space

and

k(u,

w)

(ku,

kw)

where u, u' and W.)

U, w,w'

GW

and

is called

the external direct

sum

of

SUBSPACES
4.46.

Consider the vector space that TF is a subspace of V


(i)

V
if:

in

Problem

4.39, of infinite sequences (a^, 02,

.)

in a field K.

Show

(ii)

W consists of all sequences with as the first component; W consists of sequences with only a finite number of nonzero components.
all

4.47.

Determine whether or not (ii) a ^ (i) a = 26; where k^ S R.


which:

W
b

is

c;

a subspace of RS if (iv) a (iii) ab = 0;

W
=

consists of those vectors


b

(a, b, c)

c;

(y)

6^;

(vi)

kia + kib

G RS for + kgfi = 0,

4.48.

Let

W
(iii)

consists

be the vector space of w-square matrices over a field K. Show that T^ is a subspace of V if antisymmetric (A = -A), (ii) (upper) triangular, (i) of all matrices which are diagonal, (iv) scalar.

CHAP.

4]

VECTOR SPACES AND SUBSPACES

83

4.49.

Let

AX = B

Show
4.50.

be a nonhomogeneous system of linear equations in that the solution set of the system is not a subspace of K".
vector space of all functions from the real field of the following cases.
all

n unknowns over a

field

K.

Let
of
(i)

V be the V in each
that

into R.

Show
if

that

is

a subspace

W consists of
|/(a;)|

bounded functions. (Here /


Va;

R -^ R
is

is

bounded

there exists

Af

GR

such

^ M,
all all

G R.)
(Here /
:

(ii)

(iii)

(iv)

(v)

W consists of W consists of W consists of W consists of

even functions.

R -* R

even

if

/( )

f{x), Va;

R.)

continuous functions.
functions.
in,

all difTerentiable

all

integrable functions

say, the interval

a;

1.

(The last three cases require some knowledge of analysis.)


Discuss whether or not R^
a subspace of R^.

4.51.

is

4.52.

Prove Theorem
of y.

4.4:

The

intersection of

any number of subspaces of a vector space

is

a subspace

4.53.

Suppose

and

UcW

are subspaces of

for which

UuW

is

also a subspace.

Show

that either

or

WcU.

LINEAR COMBINATIONS
4.54.

Consider the vectors u


(i)

(1,

-3,

2)

and v

(2,

-1,

1) v.

in R3.

Write Write

(1, 7,

4) as a linear combination of u and


4)

(ii) (iii)

(2,

5,

as a linear combination of u and


is (1, k, 5)

v.

For which value of k


Find a condition on

a linear combination of u and vt a linear combination of u and v


v.

(iv)

a, b

and

c so that (a, 6, e) is

4.55.

Write m as a linear combination of the polynomials


(i)

2t^

+ 3t i

and

w = t^-2tZ

where

3*2

+ 8 - 5,

(ii)

4<;2

- 6t - 1. ^ -((._,) ^ = (_j
n) *"d

4.56.

Write

as a linear combination of

>

*^

where:

(i)

E =

Q "^

(ii)

E =

(J

_l^

LINEAR SPANS, GENERATORS


4.57.

Show that (1, 1, 1), (0, 1, 1) and bination of the given vectors. Show
that the yz plane

(0, 1,

1) generate R*,

i.e.

that any vector

(a, b, e) is

a linear com-

4.58.

W=

{(0, b, c)}

in

R*

is

generated by:

(i) (0, 1,

1)

and

(0, 2,

-1);

(ii) (0, 1,

2),

(0, 2, 3)

and

(0, 3, 1).

4.59.

Show that the complex numbers vector space over the real field R. Show
degree

w = 2 + 3t
- 1)^,

and

-l 2i
1

generate the complex

field

as a

4.60.

that the polynomials {l-t),

(1

and

generate the space of polynomials of

3.

4.61.

Find one vector

in

U =
4.62.

{{a, b, 0)},

and

R3 which generates the intersection of is the space generated by the vectors

and

(1, 2, 3)

where U is the and (1, 1, 1).

xj/ plane:

Prove:

L(S)

is

the intersection of

all

the subspaces of

containing S.

84

VECTOR SPACES AND SUBSPACES


Show that L{S) = L(S u{0}). That change the space generated by the Show
that
if
is,

[CHAP. 4

4.63.

by joining or deleting the zero vector from a

set,

we

do not

set.

4.64.

Sc

T,

then

L(S)

L(T).

4.65.

Show

that

LmS)) =

L(S).

ROW SPACE OF A MATRIX


4.66.

Determine which of the following matrices have the same row space:
/l -1

3\

.-<: (3
4.67.

-4

-.

:,, 5)'

-(.ra-O'
U2

'^

1:-'
-5)

Let

Ml
vi

= =

(1, 1,

-1),

= =

(2, 3,

-1),

M3

=
Vs

(3, 1,

(1,-1,-3),

V2

(3,-2,-8),
it;

(2,1,-3)
Vj.

Show
4.68.

that the subspace of R* generated by the

is

the

same as the subspace generated by the


is

Show

matrix
4.69.

that if any row of an echelon (row reduced echelon) matrix is still in echelon (row reduced echelon) form.

deleted, then the resulting

Prove the converse of Theorem

4.6:

Matrices with the same row space (and the same

size)

are

row
4.70.

equivalent.

Show
Let

that

and

have the same column space

iff

A and B* have the same row space.

4.71.

in the

and B be matrices for which column space of A.

AB

is defined.

Show

that the column space of

AB

is

contained

SUMS AND DIRECT SUMS


4.72.

We
(i)

extend the notion of

sum

to arbitrary

nonempty subsets

(not necessarily subspaces)


T}.

S and T

of

a vector space

by defining S

+ T =

{s

+t

sG
+

S,

tG

Show

that this operation satisfies:

(ii)

(iii)

(iv)

T = T + S; + S2) + S3 = Si + (Si S + {0} = {0} + S = S; S + V = V + S = V.


commutative law:
associative law:

S+

(S2

S3)

4.73.

Show

that for any subspace

W of a vector space V, W + W
S
of a vector space (properly contained).

= W.
is

4.74.

Give an example of a subset


(i)

which

not a subspace of

but for which

S+S =

S,

(ii)

S+SCS
of

4.75.

We extend the notion

sum

of subspaces to

more than two summands as

follows.

If T^i,

W^,

. ,

T^

are subspaces of V, then

Wi +
Show
(i)

W2+---+W

{wi

+ Wi+'-'+w^: WiGWi)

that:

L{Wi,
if Si

W2

W)

(ii)

generates Wi,

^ W, + W2+ + W; = 1, .,n, then Si U S2 U i


. .

U S generates W^ + W2 +

+ Wn-

4.76.

Suppose U,

aijd

W are subspaces of a vector space.


(UnV) + (UnW)
c

Prove that

Un(V+W)

Find subspaces of B? for which equality does not hold.

CHAP.

4]

VECTOR SPACES AND SUBSPACES


V
and

85

4.77.

Let U,

W be the following subspaces of R^:


{(a, b,c):

U =
Show that
4.78.
(i)

a+b + c =
V,
(ii) B,

0},
=^

V =

{(a, b,c):
(iii)

c},

{(0, 0, c)

K}

R3

= V+

V + W,

R^

= V + W. When

is

the

sum

direct?

Let V be the vector space of all functions from the real field B into B. even functions and the subspace of odd functions. Show that V = even iff f{-x) - f{x), and / is odd iff f(-x) = -f(x).)

Let U be the subspace of U W. (Recall that / is

4.79.

Show
4.80.

Let Wi, W^, ... be subspaces of a vector space that is a subspace of Y.

for which Wy<zW.i,<Z-

Let PF

= Wj U
S =

TFj

In the preceding problem, suppose Sj generates W^, generates W.

1, 2,

Show

that

Sj U Sa

4.81.

Let V be the vector space of w-square matrices over a field K. Let U be the subspace of upper triangular matrices and the subspace of lower triangular matrices. Find (i) U + W, (ii) UnW.

4.82.

Let Let

be the external direct

sum of

the vector spaces

U
=

and

W over a

field

K.

(See Problem 4.45.)

U =
that
(i)

{(m,0):

uGU},
V,
(ii)

ys.

{(0,w):

w & W}

Show

and

W are subspaces of

V = U

W.

Answers
4.47.
(i)

to

Supplementary Problems
(iv)

Yes.

Yes.

(ii) (iii)

No;e.g.

(1,2,3)

GW

but -2(1,

2, 3)

W'.

(v)
(vi)

No;
Yes.

e.g.

(9,3,0)eW' but 2(9,3,0)

^W.

No;e.g. (1, 0, 0), (0, but not their sum.

1, 0)

T7,

4.50.

(1)

Let f,g

GW
\a\Mf

\(af+bg)(x)\

That
(ii)

is,

M^ and Mg bounds for / and g respectively. Then = \af(x) + bg(x)\ ^ \af(x)\ + \bg(x)\ = |a| |/(*)| + |6| + \b\Mg is a bound for the function af + bg.
with
af(-x)

for any scalars a, 6


|ff(a;)|

G R, ^ \a\Mf+\b\Mg

(af

+ bg)(-x) =

bg(-x)

af(x)

bg(x)

(af

bg)(x}
(a, b, 0)

4.51.

No. Although one may "identify" the vector (a, b) G R2 with, say, they are distinct elements belonging to distinct, disjoint sets.
(i)

in the

xy plane

in R3,

4.54.

-3m + 2v.
u

(ii)

Impossible, Impossible.
(ii)

(iii)

-8.

(iv)

- 36 5c =

0.

4.55.

(i)

2v

w.

(ii)

4.56.

(i)

E = 2A- B +
-5,
0).

2C.

Impossible.

4.61.

(2,

4.66.

and C.
the matrix

4.67.

Form
that

and

A whose rows are the Mj and the matrix have the same row canonical forms.
S = S =
{(0,0), (0,1), (0,2), (0,3), {(0,5), (0,6), (0,7),
(ii)
. .

whose rows are the

Uj,

and then show

4.74.

(i)

InR2,let InR2,
is

.}.

(ii)

let

...}.

4.77.

The sum

direct in

and

(iii).

4.78.

Hint. f(x)
is

^(f(x)

+
(ii)

f(-x))

^f(x)

f(-x)),

where

^(f(x)

/(-))

is

even and ^(f(x)

- f(-x))

odd.

4.81.

(i)

V = U + W.

J7nW

is

the space of diagonal matrices.

chapter 5

Basis
INTRODUCTION
Some
(i)

and Dimension

of the fundamental results proven in this chapter are:

The "dimension"
If

of a vector space

is

well defined (Theorem 5.3).


is

(ii)

has dimension n over K, then

"isomorphic" to K" (Theorem

5.12).

(iii)

system of linear equations has a solution if and only if the augmented matrices have the same "rank" (Theorem 5.10).

coefficient

and

These concepts and results are nontrivial and answer certain questions raised and investigated by mathematicians of yesterday.

We will begin the chapter with the definition of linear dependence and independence. This concept plays an essential role in the theory of linear algebra and in mathematics in
general.

LINEAR DEPENDENCE Definition: Let F be a vector


. .

.,Vm&V are said space over a field Z. The vectors vi, if there exist scalars to be linearly dependent over K, or simply dependent, .,am&K, not all of them 0, such that ai,
. .

aiVi

aiVi

+ dmVm =

(*)

Otherwise, the vectors are said to be linearly independent over K, or simply independent.

Observe that the relation


only in this case, that
aiVi
is,

(*) will

always hold

if

the a's are

all 0.

If this relation holds

a^Vi

OmVm

only

if

ai

0,

Om

then the vectors are linearly independent. On the other hand, if the relation (*) also holds when one of the a's is not 0, then the vectors are linearly dependent. Observe that if is one of the vectors vi, ...,Vm, say vi = 0, then the vectors must be
dependent; for
Ivi

Ov2

+
^

Ot;m

=
is,

and the

coefficient of Vi is not 0.

On
0,

the other hand, any nonzero vector v

by

itself,

independent; for *^

kv

,a V =

t implies

A;

i.

Other examples of dependent and independent vectors follow.


Example
5.1:

The vectors m
for 3m

2v

= (1,-1,0), w = 0,
3(1,

1;

(1,3,-1)

and

w=

(5, 3,

-2)

are dependent since,

-1,

0)

2(1, 3,

-1)

(5, 3,

-2)

(0, 0, 0)

86

CHAP.

5]

BASIS

AND DIMENSION
= =
(0, 5,

87

Example

5.2:

We

show that the vectors u = (6, 2, 3, 4), v are independent. For suppose xu + yv + zw
scalars.

-3,

1)

where

x,

and y and

w=

(0, 0, 7,

-2)

z are

unknown

Then
(0, 0, 0, 0)

= =

x{6, 2, 3, 4)
(6a;,

2x

+ y{0, 5, -3, 1) + z{0, 0, 7, + 5y, 3x-Sy + Iz, Ax + y- 2z)

-2)

and

so,

by the equality of the corresponding components,


6a;

+ hy Zx-Zy + 4a; + y
2x

=0 =0
lz

2z

= =
0,

The first equation yields the third equation with


xu-'t yv

a;

a;

= 0; = 0,

the second equation with x

yields

0;

and

yields 2

=
x

0.

Thus
0,

+ zw =

implies

Accordingly

u,

v and

are independent.

Observe that the vectors in the preceding example form a matrix in echelon form:

Thus we have shown that the (nonzero) rows of the above echelon matrix are independent. This result holds true in general; we state it formally as a theorem since it will be frequently
used.

Theorem
follows:

5.1:

The nonzero rows

of a matrix in echelon

form are

linearly independent.

For more than one

vector, the concept of

dependence can be defined equivalently as

The vectors Vi, .,Vm are linearly dependent combination of the others.
. .

if

and only

if

one of them

is

a linear

For suppose,

say, Vi is a linear combination of the others:


Vi

aiVi

Ui-iVi-i

tti+iVi +

UrnVm

Then by adding Vi
aiVl

to both sides,

we

obtain

Oi-iVi-i

Vi

Ui + lVi +

+ amVm
Conversely,

where the

hence the vectors are linearly dependent. suppose the vectors are linearly dependent, say,
coefficient of Vi is not 0;
biVi

bjVj

+ bmVm =

where
bi^bj+iVj+i

bj -

Then
and so

Vj

bi^biVi

bf^bj-iVj-i

bi^bmVm
many
im-

Vj is

a linear combination of the other vectors.

We now make a slightly stronger statement than that above; this result has portant consequences.
Lemma
5.2:

The nonzero vectors


them, say
vt, is

Vi, .,Vm are linearly dependent if and only a linear combination of the preceding vectors:
.
.

if

one of

Vi

kiVi

kiVi

fci-iVi-i

88

BASIS

AND DIMENSION

[CHAP.

Remark

1.

The
set

set [vi,
vi,

vectors

.,Vm} is called a dependent or independent set according as the .,Vm are dependent or independent. We also define the emptyto be independent.
.

Remark

2.

If

two of the vectors dependent, For


coefficient of

Vi,

.,Vm are equal, say vi


,

a.,

n.,

- vz, - n

then the vectors are

and the

i;i

is

not

0.
if

Remark Remark Remark Remark

3.

Two

vectors Vi and v^ are dependent

and only

if

one of them

is

a multiple of

the other.
4.

which contains a dependent subset subset of an independent set is independent.


set
If the set {vu
{Vij, Vi^,
. .
. . . .

is

itself

dependent.

Hence any

5.

Vm}

Vi} is

independent, then any rearrangement of the vectors also independent.


is

6.

In the real space

R^ dependence

of vectors can be described geometrically as

follows: any two vectors u and v are dependent if and only if they lie on the same line through the origin; and any three vectors u, v and w are dependent if and only if they lie on the same plane through the origin:

u and V are dependent.

u,

V and

are dependent.

BASIS

AND DIMENSION
said to be of finite dimension n or to be n-dimensional, e , n, if there exists linearly independent vectors ei, e2, ... e) is then called a basis of V. The sequence {ei, 62,
is
. . . ,

We begin with a definition. A vector space V Definition: written dim V =


which span V.

The above

definition of dimension is

well defined in view of the following theorem.

Theorem

5.3:

Let

F be

finite

dimensional vector space.

Then every basis of

has the

same number of elements.

The vector space

{0} is defined to

the above definition since, vector space is not of finite


Example
5.3:

have dimension 0. (In a certain sense this agrees with When a is independent and generates {0}.) by definition, dimension, it is said to be of infinite dimension.
field.

Let

be any

Consider the vector space


ei

K" which
.

consists of n-tuples of ele-

ments of K.

The vectors
62

m 0, 0, = (1, n n = (0,1,0,

.,

n n\ 0,
0)

...,0,0)

(0,0,0

0,1)

form a

basis, called the usual basis, of K".

Thus K" has dimension

n.

CHAP.

5]

BASIS

AND DIMENSION
all

89

Example

5.4:

Let

be the vector space of

3 matrices over a field

K.

Then the matrices

/I

Vo

0\ 0/'

/O
i^o

ON

/O

0/'

Vo

1\ 0/'

( form a basis of U.

10

10

Thus dim C/ = 6. More generally, let V be the vector space of all m X % matrices over K and let E^ S y be the matrix with ly-entry 1 and elsewhere. Then the set {ffy} is a basis, called the usual basis, of V (Problem 5.32); consequently dim V mn.
Example
5.5:

W dim W = n+1.
Let
is

be the vector space of polynomials (in t) of degree linearly independent and generates W. Thus it

n.

The
a

set {1, t,t^,

is

basis

of

.,

t"}

and so

comment that the vector space of all polynomials is not finite dimensional since (Problem 4.26) no finite set of polynomials generates V.

We

The above fundamental theorem on dimension portant "replacement lemma":

is

a consequence of the following im-

Lemma

5,4:

Suppose the
is

set {vi, V2, If {wi, ., Vn} generates a vector space V. ., Wm} linearly independent, then n and V is generated by a set of the form
.

{Wi, ...,Wm,

Vij,

Vi^_J

Thus, in particular, any %

+1

or more vectors in

are linearly dependent.

set

by the

Observe in the above lemma that we have replaced of the vectors in the generating independent vectors and still retained a generating set.

Now suppose S is a subset of a vector space V. dependent subset of S if:


(i)

We

call {vi,

.,

Vm} a maximal in-

it is

an independent subset of S; and


. .

(ii)

{vi,

.,Vm,w}

is

dependent for any

w e S.

The following theorem

applies.

Theorem

5.5:

Suppose S generates
of S.

and
is

Then

(vi,

Vm}

{vi, ., Vm} a basis of V.


.
.

is

a maximal independent subset

The main
subsets
is

relationship between the dimension of a vector space and contained in the next theorem.

its

independent

Theorem

5.6:

Let
(i)

V be
Any Any

of finite dimension n.
set of

Then:
is

n + 1 or more

vectors

linearly dependent.
i.e.

(ii)

linearly independent set is part of a basis,

can be extended to

a basis,
(iii)

linearly independent set with


in

n elements

is

a basis.

Example

5.6:

The four vectors

K*
(1,1,1,1), (0,1,1,1),
(0,0,1,1), (0,0,0,1)

are linearly independent since they form a matrix in echelon form. since dim K* = 4, they form a basis of K*.

Furthermore,

Example

5.7:

The four vectors

in R3,

(257,-132,58), (43,0,-17),

(521,-317,94), (328,-512,-731)
3.

must be

linearly dependent since they

come from a vector space of dimension

90

BASIS

AND DIMENSION

[CHAP.

DIMENSION AND SUBSPACES


The following theorems give
basic relationships between the dimension of a vector space

and the dimension of a subspace.

Theorem

5.7:

Let

W he a subspace of an n-dimension vector space dim W = n, then W V. In particular


if

V.

Then dim

W -n.

Example

5.8:

be a- subspace of the real space B?. Let can only be ing theorem the dimension of

Now
0, 1, 2

dim R^
or
3.

3; hence by the precedThe following cases apply:

(i)

(ii)

(iii)

dim dim dim

(iv)

W W W dim W
has

= =

0, 1, 2, S,

then

then

W = {0}, a point; W a line through the origin;


is
is

then T^

a plane through the origin;


the entire space R^.

then

is

Theorem

5.8:

Let

and

U+
Note that
if

W
is

W be
finite

finite-dimensional subspaces of a vector space V.

Then

dimension and

dim(C7

+ F)
of

dim

f/

+ dim
i.e.

TF

- dim{UnW)
then

the direct
5.48).

sum

and W,

V = U W,

dim V =

dim

U + dim W
Example

(Problem

5.9:

Suppose

U
2.

and

W= dim W

W are the xy plane and yz plane, respectively, in R^:


Since RS = t/ + I^, the above theorem,

U=
17

{(a, 6,0)},

{(0, 6, c)}.

dim

{U+W) =
or

B.

Also,

dim
1

and

By
3

+ 2-dim(f/nTF)

dim{UnW) =

Observe that this agrees with the fact that {(0, 6, 0)}, and so has dimension 1.
z

UnW

is

the y axis,

i.e.

UnW

w
^Vr\W

^^0
^,/^ V
..^

^^

RANK OF A MATRIX
x w matrix over a field K. Recall that the row space of A is Let A be an arbitrary of R^ the subspace of K" generated by its rows, and the column space of A is the subspace columns. The dimensions of the row space and of the column space of A generated by its are called, respectively, the row rank and the column rank of A.

Theorem

5.9:

The row rank and the column rank

of the matrix
is

are equal.
value of
its

Definition:

The rank of the matrix A, written rank (A), rank and column rank.

the

common

row

Thus the rank of a matrix gives the maximum number of independent rows, and also of a matrix as the maximum number of independent columns. We can obtain the rank
follows.

/I

2 6

-1\

Suppose

A =
\

2 3

-3 -3
5/

We

reduce

to echelon

form using the elementary

row

operations:

106

CHAP.

5]

BASIS

AND DIMENSION
1

91

A
Recall that

to

-1 -3 -1 4 -6 -2
2

2 2

to

-1 -3 -1

\o

row equivalent matrices have the same row space. Thus the nonzero rows of the echelon matrix, which are independent by Theorem 5.1, form a basis of the row space of A. Hence that rank of A is 2.

APPLICATIONS TO LINEAR EQUATIONS


Consider a system of

m linear equations in
ttuXl
CziaJi

n unknowns

a;i,

over a field K:

+ ai2X2 + + a22X2 + +
am2X2

+ ainXn = + ChnXn =

bi

^2

dmlXi

OmnXn

&m

or the equivalent matrix equation

AX = B
where

(an)

is

the coefficient matrix, and


is

X=
ai2

(xi)

and

B=

(6i)

consisting of the

unknowns and of the

constants, respectively.

are the column vectors Recall that the augmented

matrix of the system

defined to be the matrix

an
{A,B)
a2i

ttln
.

bi

022

a2n

62

ttml

ftm2

dmn

hm

Remark

1.

The above

linear equations are said to be dependent or independent according


i.e.

as the corresponding vectors,

the rows of the augmented matrix, are

dependent or independent.

Remark

2.

Two systems of linear equations

are equivalent
i.e.

augmented matrices are row equivalent,

if and only if the corresponding have the same row space.

Remark

3.

We

equations, such as a system in echelon form.

can always replace a system of equations by a system of independent The number of independent equations will always be equal to the rank of the augmented matrix.
is

Observe that the above system

also equivalent to the vector equation

= B has a solution if and only if the column vector B is a linear combination of the columns of the matrix A, i.e. belongs to the column space of A. This gives us the following basic existence theorem.
Thus the system

AX

Theorem

5.10:

The system
coefficient

of linear equations

matrix

has a solution if and only if the and the augmented matrix (A, B) have the same rank.

AX B

92

BASIS

AND DIMENSION

[CHAP.

Recall (Theorem 2.1) that if the system = {v general solution is of the form v +

of the associated homogeneous system

AX = B does have a solution, say v, then its + w: wGW} where W is the general solution AX = 0. Now W is a subspace of K" and so has a
is

dimension.
applies.

The next theorem, whose proof

postponed until the next chapter (page 127),

Theorem

5.11:

The dimension of the

solution space

W of the homogeneous system of linear


is

is equations the rank of the coefficient

AX =

n-r

where n matrix A.

the

number

of

unknowns and r

is

In case the system


(see

AX =
. .

page
all

21),

say, Xi^.xi^,

and

other free variables

echelon form, then it has precisely n-r free variables Let Vj be the solution obtained by setting aji^. = 1, .,Xi^_^. Then the solutions Vi, 0. ., v^-r are linearly independent
is in
. .

(Problem 5.43) and so form


Example
5.10:

a basis for the solution space.

Find the dimension and a basis of the solution space equations x + 2y- Az + 3r- s =
X

of the system of linear

2x

+ +

2y iy

-2z + 2r+ -2z + 3r +

4s

Reduce the system to echelon form:


X

2y

+ 2z 6z Az

3r

r + 3r +

2s 68

= = =

2y

4z 2z

and then

+ -

3r
r

2s

= =
dim

There are 5 unknowns and 2 (nonzero) equations in echelon form; hence 2 = 3. Note that the free variables are y, r and s. Set:
(i)
J/

W
l

1,

0, s

0,

(ii)

3/

0,

1,

0,

(iii)

0,

0, s

to obtain the following respective solutions: vi

(-2,1,0,0,0),

V2

(-1,

0,

1, 0),

vs

= (-3,0,-1,0,1)

The

set {^i, V2, V3} is a basis of the solution space

W.

COORDINATES
be a basis of an n-dimensional vector space V over a field K, and be any vector in V. Since {ei} generates V, 1; is a linear combination of the d:
Let
{ei,
. . . ,

e}

let

ttiCi

aiCi

+ dnen,

OH

:K

i.e. the n are independent, such a representation is unique (Problem 5.7), completely determined by the vector v and the basis {ei}. We call scalars ai, ., a are coordinate ., a) the these scalars the coordinates of v in {ei}, and we call the -tuple (ai, denote it by [v]e or simply [v]: vector of v relative to {ei} and

Since the

d
.

[V]e

(tti,

ai,

.,

On)

Example

5.11

Let

be the vector space of polynomials with degree

^
R}

2:

V The polynomials
ei

{af2

5t

+c

a, 6, c

1,

02

form a basis for V.

= t-1 and = 2t2-5t + 6. Let v


e^, 63}.

63

= (t- 1)2 = i^ - 2t + 1 Find [v\ the coordinate vector of v

relative to the basis {cj,

CHAP.

5]

BASIS

AND DIMENSION
e;

93

Set V as a linear combination of the


yCi

using the unknowns

x,

+ zeg.
22

y and

z:

xe^

+ y(t - 1) + z(f2 - 2 + 1) = + 3/f - + t2 _ 2zt + z = zfi + (y- 2z)t + (x-y + z) =


x(\)
a;

3/

Then

set the coefficients of the

same powers of
X

equal to each other:


6

y
y

=
j/

-2z = -5
z

The

solution of the above system

is

a:

3,

1, z
[v]^

2.
(3,

Thus
-1,
2)

3ei

62

2e3,

and so

Example

5.12:

Consider the real space


the basis
/i

tt?.

Find the coordinate vector of r

(3, 1,

(1, 1. 1),

/a

(0, 1, 1),

/s
/;

-4) relative to
v

(0, 0, 1).

Set r as a linear combination of the

using the unknowns

x,

y and

z:

xfi

(3,1,-4)

= ^ =

a;(l, 1, 1)

(x, X, x)

+ j/(0, 1, 1) + z(0, 0, + (0, y, y) + (0, 0, 2)

1)

(x,x

+ y,x-\-y + z)

Then

set the corresponding components equal to each other to obtain the equivalent system of equations X

+
-\-

y y

=3 =1

X
having solution

= 4
Thus
[v]f

3,

-2, z

=
is

-5.

(3,

-2, -5).
ej

We
(0, 0, 1),

remark that

relative to the usual basis

e^

(1, 0, 0),

(0, 1, 0),

eg
v.

the coordinate vector of v

identical to v itself:

[v]^

(3, 1,

-4)

have shown above that to each vector there corresponds, relative to a given ., e}, an n-tuple [v]e in K\ On the other hand, if (ai, .,a) G j?, then there exists a vector in V of the form aiCi + + ae. Thus the basis {d} determines a one-toone correspondence between the vectors in V and the w-tuples in K". Observe also that if
basis
{ei,
.
.

We

vGV

an<i

= w =
V
t;

ttiei

biBi

+ +

+ ae + 6e + 6)e

corresponds to corresponds to

(ai, (&i,

a)
6)

then

and, for

+ &,)ei + any scalar k G K,


i
(ai

(a

corresponds to

(ai,

a)

(bi,

b)

kv

(A;ai)ei

{kan)e
[w]e

corresponds to

k{ai,

a)

That

is,

[v

+ w]e = M +

and

[kv]e

k[v]e

Thus the above one-to-one correspondence between V and K" preserves the vector space operations of vector addition and scalar multiplication; we then say that V and K" are
isomorphic, written

V ^ K"". We

state this result formally.

Theorem

5.12:

Let

be an -dimensional vector space over a

field

K.

Then

and K^ are

isomorphic.

94

BASIS

AND DIMENSION

[CHAP.

The next example gives a


Example
5.13:

practical application of the above result.

Determine whether the following matrices are dependent or independent:

^=a
5.4,

I'D[B]

"^G
=

3 5

-4\
AJ'

c = f

~"
9

U6

10

The coordinate vectors of the above matrices


page
89, are

relative to the basis in

Example

[A]

(1,2,-3,4,0,1),

(1,3,-4,6,5,4),

[C]

(3,8,-11,16,10,9)

Form

the matrix

M whose rows are the above coordinate vectors:


M
=

Row

reduce

M to echelon form:

40l\
M
and
to
I

/12-3
to

40
2 5

1-1

10

6/

\00000

1-1

Since the echelon matrix has only two nonzero rows, the coordinate vectors [A], [B] Accordingly, the [C] generate a space of dimension 2 and so are dependent. original matrices A, B and C are dependent.

Solved Problems

LINEAR DEPENDENCE
5.1.

Determine whether or not u and v are linearly dependent


(i)

if:

u = u = =

(3,4), V

(1,-3)

(iii)

u - (4,3,-2), v = (2,-6,7) u = (-4,6,-2), v = (2,-3,1)

(ii)

(2,-3), V

=
4N

(6,-9)

(iv)

/l-2

(V) '

[so
5*

-1/

/2-4 8\ 0-2/ [g
i3,

2-3N /l (,i), = -5 i) ^6
5t^
9'3

^/6-5
[l
2

-3

(vii)

M = 2 -

6*2

+ 2t-4t^ +

(viii)

u = l-St + 2t^- St\ V = -S + 9t-6t^ +


vectors u and v are dependent if and only
if

Two
(i)

one

is

a multiple of the other.

(vii)

3m. No. (ii) Yes; for v No. (viii) Yes; for v = 3m.

(iii)

No.

(iv)

Yes; for

u =

2v.

(v)

Yes; for

2m.

(vi)

No.

5.2.

Determine whether or not the following vectors in R^ are linearly dependent:


(i)

(1,-2,1),
(1,

(2,

1,-1),
-6),

(7, (3,

-4,1)
-1, -1),
(2, 4,

(iii)

(1,2,-3), (1,-3,2), (2,-1,5)


(2,

(ii)

-3,

7), (2, 0,
1.

-5)

(iv)

-3,

7), (0, 0, 0), (3,

-1, -4)

(i)

Method

Set a linear combination of the vectors equal to the zero vector using unknown
z:

scalars x, y and

x(l,

-2,

1)

2/(2, 1,

-1)

2(7,

-4,

1)

(0, 0, 0)

CHAP.

5]

BASIS
Then
0^

AND DIMENSION
(2v, y,

95

(x,

-2x,

x)

-y)

(Iz,

-4z,

z)

=
-

(0, 0, 0)

(x

+ 2y + 7z, -2x + y - 4:Z,

- y + z)

(0, 0, 0)

and reduce
X

Set corresponding components equal to each other to obtain the equivalent homogeneous system, to echelon form:

2y
2/

o, -2x +
X-

+ 7z - 4z = +
z

.
=

X
or

2y
5j/

7z

=
= =
X
or

-3j/

+ -

2y

7z

lOz
6z

y\-2z

The system, in echelon form, has only two nonzero equations in the three unknowns; hence the system has a nonzero solution. Thus the original vectors are linearly dependent.
Method
2.

Form

the elementary

row

the matrix whose rows are the given vectors, and reduce to echelon form using operations:

to

5-3

to

Since the echelon matrix has a zero row, the vectors are dependent. generate a space of dimension 2.)

(The three given vectors

(ii)

Yes, since any four (or more) vectors in R3 are dependent.

(iii)

Form

the matrix whose rows are the given vectors, and

row reduce the matrix

to echelon form:

to

Since the echelon matrix has no zero rows, the vectors are independent. (The three given vectors generate a space of dimension 3.) Since

(iv)

(0, 0, 0)

is

one of the vectors, the vectors are dependent.

5.3.

Let

V be

A,B,C

GV

the vector space of 2 x 2 matrices over R. are dependent where:

Determine whether the matrices

(i)

Set a linear combination of the matrices A, scalars x, y and z; that is, set xA + yB + zC

B =

and
0.

C equal to the zero matrix using unknown Thus:

:)

HID- c
z

:) _ ~

^
f^ \0

c
^\ 0)

:)

/x + y +
\

x x

+ z\ + yj

96

BASIS

AND DIMENSION

[CHAP.

Set corresponding entries equal to each other to obtain the equivalent homogeneous system of equations:

X X X X

+ + +

= =

=0 =0

Solving the above system we obtain only the zero solution, x = 0, y = 0, z = 0. We have shown that xA + yB + zC implies a; = 0, y = 0, z = 0; hence the matrices A,B and C are linearly independent.
(ii)

Set a linear combination of the matrices A,B and C equal to the zero vector using scalars x, y and z; that is, set xA + yB + zC = 0. Thus:
'1

unknown

2\

/3 -1\

-5\

/O

ON

2x\

/Zy
'^

-y\
2y)
'^

-5z\
/

\3x

x)
X
3x

\2y
z
4:Z

\-Az
5z\

_ ~
/O

/O
1,0

+ 3y + + 2y-

2x-y x

2y

\0

linear equations

Set corresponding entries equal to each other to obtain the equivalent homogeneous system of and reduce to echelon form:

2x
3x

x
or finally

+ + +

3y

y
2y 2y

= 5z -4z =
z

X
or

3y+z
7z 7z
z

=0
X

-ny -ly -y
z
z

= = = =

Sy

+ +

The system
X
j^

in echelon

= =

2,

0, z

= -1, z = 1. We have shown that xA + yB + = 0; hence the matrices are linearly dependent.

form has a free variable and hence a nonzero solution, for example, does not imply that x = 0, zC =

5.4.

Let

be the vector space of polynomials of degree are independent or dependent where: u,v,w

^3

over R.

Determine whether
9t St

gV
1^

(i)

u ^ i^-Sf^ + u =

(ii)

+ + 4t^-2t +
Bt

l,

V V

S,

= t^-t^ + St + 2, w = = t^ + 6t^-t + 4, w =
w
= +

2*8 B1^

- 4*2 + + St^ -

+5 +7

(i)

unknown

Set a linear combination of the polynomials u,v and scalars x, y and z; that is, set xu + yv + zw
x(t

0.

equal to the zero polynomial using Thus:

- 3*2 + 5t + l) +
+
5xt

y(t^

- t2 + 8t + 2) +
yt^

z{2fi

-U^ + 9t + 5) =
izt^
{x

or

xfi

3xt^

yt^

Syt

2y

2zt

9zt

5z

or

(x

+ y + 2z)t? +

{-3x

-y- iz)t^ +
t

(5*
0:

+ 8y + 9z)t +
= = =

+ 2y + 5z) =

The

coefficients of the

powers of

must each be
X

+ y 2z 4z 3a; 5x + 8y + 9z x + 2y + 5z
2/

Solving the above homogeneous system, hence u, v and w are independent.

we

obtain only the zero solution:

0,

0,

0;

CHAP.

5]

BASIS

AND DIMENSION
w
0.

97

(ii)

unknown

Set a linear combination of the polynomials u,v and scalars x, y and z\ that is, set xu + yv + zw =
x{ti

equal to the zero polynomial using Thus:

+ 42 - 2t + 3) +

3/(t3

6f2

- f + 4) +

2(3*3

+ (2 _ gt + +
izt^

7)

=
+
7z

or

xt^

4a;t2

2xt

Zx

yt

6yfi

yt

^y

3zt^

Szt

or

(x

+ y + 3z)i3 +

(4a;

+ 62/ + Sz)t^ +
t

(-2x

-y- 8z)t +
X

{3x

+ 4y + 7z) =
to echelon form:

Set the coefficients of the powers of

each equal to
3z

and reduce the system

+ + -2x X
4a;

y
62/

y Ay

+ + -

82 8z 7z
a;

3x
or finally

= = = =

2y
or

y y
3/

= 4:Z -2z = - 2z =
+ 3z

+ 82 = -2z =

that

The system in echelon form has a free variable and hence a nonzero solution. We have shown xu + yv + zw = does not imply that x = 0, y = 0, z - 0; hence the polynomials are

linearly dependent.

5.5.

Let V be the vector space of functions from R into R. Show that f,g,hGV are independent where: (i) f{t) = e^, g{t) = t\ h{t) = t; (ii) f{t) = sint, g{t) = cost,
h{t)

t.

In each case set a linear combination of the functions equal to the zero function scalars x, y and z: xf -\- yg + zh = 0; and then show that a; = 0, j/ = 0, z = 0. xf + yg -\- zh - {) means that, for every value of t, xf{t) + yg(t) + zh(t) = 0.
(i)

We

using unknown emphasize that

In the equation

xe^*

+ yt^ + zt = 0, substitute t = to obtain xe" + j/0 + zO = * = 1 to obtain xe^ + y =0 z t = 2 to obtain xe* + 4y + 2z =


-\-

or

a;

X
Solve the system
-

=0
+ +
y Ay

xe^ xe*

+ +

2z

= =

to obtain only the zero solution:

0,

0,

0.

Hence
(Ii)

/,

g and h are independent.


In the equation
*
t
a;

Method

1.

sin

3/

cos

zt

0,

substitute
Q

= =

to obtain
tt/2
TT

x-Q + y ! + z'Q =
a;

or
or or

y
a;

=
+

Q
7rzl2
Trz

to obtain to obtain

a;

+ +

j/

Zff/2

y(l)

jr

= =

y +

= = =

V
Solve the system

X
-y

= + jrz/2 = + vz =

to obtain only the zero solution:

0,

0, z

0.

Hence

/,

g and h are independent.


2.
t

Method

Take the
to get

first,

second and third derivatives of

x sin

2/

cos

with
(1)

respect to

X cos

2/

sin

+
t

= =
=

sin

y cos
2/

(2)
(5)

cos

sin

9g

BASIS

AND DIMENSION
(2)

[CHAP.

Add

(1)

and

(5) to

obtain
t

0.

Multiply
a;

by sin
t

and

(3)

by cos

t,

and then add:

sin

(2):
(3):

sin^

y sin

cost
t

cos

cos^

y sint cos
cos2
t;

=
=
or
to obtain
a;

-x(sin2
Lastly, multiply (2)

f)

by

cos t
+

and

(5)

by

sin

and then add

2/(cos2 e

+
zt

sin2

or
implies

y
a;

=
=
0,
j/

Since
/,
fir

x sint

y cost

0,

and h are independent.

5.6.

Let

u,

V and

be independent vectors.

Show
+ w) -

that

u + v,
where
x,

u-v

and

u-2v + w

are

also independent.

Suppose x(u + v) + y(u -v) + z(u-2v or xu + XV + yu yv + zu 2zv + w =


{x

y and z are scalars.

Then

+ y + z)u+
X

(x

y 2z)v + zw =
0:

But

u,

V and

are linearly independent; hence the coefficients in the above relation are each

x-y
The only
solution to the above system is

+ z = -2z =
2

= 0.

0,

0, z

Hence u

v,

u-v

and

u-2v + w

are

independent.

5.7.

Let

..,Vm be independent vectors, and suppose m is a linear combination of + OmVm where the ai are scalars. Show that the the Vi, say u = aiVi + onVz + above representation of u is unique.
vi, V2,
.

Suppose u

biVi

b2V2

+ bv

where the
(a2

6j

are scalars.

Subtracting,

= uu =
But the
Vi

(!

bi)vi +

62)^2 +

(m

bm)Vm
0:

each are linearly independent; hence the coefficients in the above relation are
ai

bi

0,

a2

b2

0,

.. .,

~"

*m

Hence
of the

a,

61,

02

62,

...,

a^

b^ and so the above representation of m as a linear combination

Vi is

unique.

5.8.

Show

= {l,l+i) in C^ are linearly dependent that the vectors v = (l+i, 2i) and R. over the complex field C but are linearly independent over the real field
of

Recall that 2 vectors are dependent iff v is 1, I* can be a multiple of

iff

one
(l

is

=
(l

+ i)w.
1

a multiple of the other. Since the first coordinate But 1 + i R; hence v and w are independent

overR. Since and


1

(i+i)w

+ i)(l,

+ i) =

(1

+ 1,

2i)

+tG

C,

they are dependent over C.

5.9.

Suppose S

{vi,

...,Vm}

is

also dependent.
Since
{v^, ...,v^} is

contains a dependent subset, say {vi, ...,Vr}. Show that Hence every subset of an independent set is independent.
a^, ...,a^,

dependent, there exist scalars


a^vi

not

all 0,

such that

a^v^

CHAP.

5]

BASIS

AND DIMENSION
.
.

99

Hence there

exist scalars aj

flr.

0,

0,

not

all 0,

such that

aiVi

a^Vr

O^r+j

+ Ov^ =

Accordingly,

is

dependent.

5.10.

Suppose
is

{vi, .,Vm} is independent, but a linear combination of the i;;.


.
.

{vi,

.,Vm,w}

is

dependent.

Show

that

Method
Oj^x

dependent, there exist scalars !,..., . ^. "o* *11 0, such that = 0, then one of the aj is not zero and Oi^yj + + a^v^ = 0. But this contradicts the hypothesis that {v^, .... i>} is independent. Accordingly, 6 # and so
1.
.
. .

Since {vj,
(im'"m

v^,

w}

is

+ 6w =

0.

If

w =
That
is,

6-i(-aiVi
Vi.

a0

-b-^a^Vi

- b-^a^Vm

is

a linear combination of the

Method
5.2,

2. If w = 0, then w = Ovi + one of the vectors in {dj, v^, w} vector cannot be one of the v's since {vj,
.
.

+
.

. ,

is
. . ,

Ov^. On the other hand, if w ^^ then, by Lemma a linear combination of the preceding vectors. This v^} is independent. Hence w is a linear combination

of the

v^.

PROOFS OF THEOREMS
5.11.

Prove Lemma 5.2: The nonzero vectors Vi, .,Vm are linearly dependent if and only if one of them, say vi, is a linear combination of the preceding vectors: Vi =
.

ait;i

+ <h-iVi-i.
Vj

Suppose

aj'Ui

aj-x-yj-i.

Then

aiVi

ai_iDi_i

Vj

Vi

Oi>j+i

Ov,n

=
a^, not
all

and the
0,

coefficient of vi is not 0.

Hence the

are linearly dependent.


. . . ,

Conversely, suppose the Vj are linearly dependent. Then there exist scalars Oi, such that ai^i + + a^Vm = 0. Let k be the largest integer such that o^ = 0.

Then

a^Vi

Ofc'Ufc

+
=

Ot)fc

+i

+ Ov^
v^

or

a^Vi

a^v^

Suppose fe > 1 and

1;

then

a^Vi

0, a^ =

and so

0.

But the

Vi

are nonzero vectors; hence

That

is, -y^ is

a linear combination of the preceding vectors.

5.12.

Prove Theorem

5.1:

The nonzero rows


,Ri}
is

Ri,

.,Rn of a matrix in echelon form are


a linear combination

linearly independent.

Suppose {RntRn-i,
of the preceding rows:

dependent. Then one of the rows, say R^,

is

(*) + <''n^n ^m "m+l^m + l + m + 2-'^m + 2 + Now suppose the fcth component of R^ is its first nonzero entry. Then, since the matrix is in echelon form, the fcth components of Rm+u -y^n ^^^ ^^ 0, and so the feth component of (*) is a^+i* + + Ore* = 0. But this contradicts the assumption that the feth component of B is "m+2* +

"

not

0.

Thus

i?i,

.,Rn are independent.

5.13.

Suppose
(i)

{vi,

.,

Vm} generates a vector space V.


vi,
.
.

Prove:

If

w GV,

then {w,

.,

Vm)

is

linearly dependent

and generates V.
[vi,
. .

(ii)

If V{ is a linear

combination of the preceding vectors, then

.,Vi-uVi+i,

.,

Vm]

generates V.
(i)

If

we
. .

y,
.,

then

{w, Vi,

Vj^} is linearly

selves generate V.

a linear combination of the Vj since {iij} generates V. Accordingly, dependent. Clearly, w with the Vj generate V since the Vf by themThat is, {w,Vi, .,v^} generates V.
is
.

100

BASIS

AND DIMENSION

[CHAP.

(ii)

Suppose

Vi

bination of the

Vi,

k^Vi + say,

fei_i'yj_i.

a^Vi

Let uGV. Since {Vii generates V, m is a linear com+ a^v^. Substituting for Vj, we obtain

u =

ttiVj

ai_ii;j_i

aj(fciVi

+ ki^^Vi-i) +
a^+iVi + i

aj+iVj+i

+ a^Vm

Thus
{vi,
.
.

{ai

+ aiki)vi +
1,

{ai^i

+ a^k^_l)v^_l +

+ a^Vm
delete Vj

..t'i-i, 1^1+
still

,^'ot}

generates V.

In other words,

we can

from the gen-

erating set and

retain a generating set.

5.14.

Prove
.

If [wi, .,Wm} generated by a set of the form is linearly independent, then .,Vi^_^}. Thus, in particular, any + l or more vectors in V are {wi, .,Wm,vi^, linearly dependent.
5.4:

Lemma

Suppose

{vi,

.,Vn} generates a vector space V.

m n

and

is

prove the theorem in the case that the generates V, we have by the preceding problem that
It suffices to

V;

are

all

not

0.

(Prove!)

Since the {Vj}

{wi, Vi, ..., -yj


is

{1)

linearly dependent and also generates V. By Lemma 5.2, one of the vectors in (1) is a linear combination of the preceding vectors. This vector cannot be Wj, so it must be one of the v's, say Vj. Thus by the preceding problem we can delete Vj from the generating set {!) and obtain the generating

Now we

repeat the argument with the vector Wj{Wi, W2, Vi,


.

That

is,

since

(2)

generates V, the set


(5)

..,Vj-i,Vj+i, ...,

vJ

is linearly

dependent and also generates V. Again by Lemma 5.2, one of the vectors in (3) is a linear combination of the preceding vectors. We emphasize that this vector cannot be Wj or Wj since w^} is independent; hence it must be one of the v's, say v^. Thus by the preceding problem {wi, we can delete v^ from the generating set (3) and obtain the generating set
.
. . ,

{Wi, W2, Vi,

.,

Uj_i, Hj+i,
forth.
set.

.,

Vfc_l, Vfc +

l,

.,

-vJ

We repeat the argument with Wg and so w's and delete one of the y's in the generating
set of the required form:

At each

step

If

m ^ n,

then

we are able to add one of the we finally obtain a generating

{Wl, ...,

Wm.

'"*n-m^

is not possible. Otherwise, after n of the above steps, we obtain ., w which w}. This implies that w+i is a linear combination of Wj, ., the generating set {wi, contradicts the hypothesis that {wj} is linearly independent.

Lastly,

we show

that
. .

m>n

5.15.

Prove Theorem 5.3: Let 7 be a finite dimensional vector space. V has the same number of vectors.

Then every

basis of

Since ) is another basis of V. Suppose {ei,e2, ...,e} is a basis of V, and suppose {fufi, dependent by the } must contain n or less vectors, or else it is generates V, the basis {fi.fz, } contains less than n vectors, then preceding problem. On the other hand, if the basis {fiJ^, ej is dependent by the preceding problem. Thus the basis {fiJz, } contains exactly n {ej

{e,}

vectors,

and so the theorem

is

true.

5.16.

Prove Theorem 5.5: Suppose {vi, ...,Vm} is a maximal independent subset of a S which generates a vector space V. Then [vi, ..,Vm} is a basis of V.
.

set

weS. Then, since {vj} is a maximal independent subset of S, {Ui, ...,, w} is G L(Vi). Hence dependent. By Problem 5.10, w is a linear combination of the i^j, that is, w linearly = L{S) C L(vi) C V. Accordingly, {vJ generates V and, since it is mS C L(Vi). This leads to V dependent, it is a basis of V.
Suppose

CHAP.

5]

BASIS

AND DIMENSION
Show
that

101

5.17.

Suppose

is

particular, a subset of

generated by a finite set S. S is a basis of V.

is

of finite dimension and, in

Method 1. Of all the independent subsets of S, and there is a finite number of them one of them is maximal. By the preceding problem this subset of S is a basis of V.

since

is finite,

Method 2. If S is independent, it is a basis of V. If S is dependent, one of the vectors is a linear combination of the preceding vectors. We may delete this vector and still retain a generating set. We continue this process until we obtain a subset which is independent and generates V, i.e. is a
basis of V.

5.18.

Prove Theorem
(i)

5,6:

Let

be of

finite

dimension

n.

Then:

(ii)

Any set of n + 1 or more vectors is linearly dependent. Any linearly independent set is part of a basis,

(iii)

A linearly independent set with n elements is a basis.


Suppose
{ei,
. . .

e} is a basis of V.

(i)

Since

{ei,

...,} generates V,
. .

any m

+1

or more vectors
5.4,

is

dependent by

Lemma

5.4.

(ii)

Suppose

{v^,

v,.} is

independent.

By Lemma

is

generated by a set of the form

S =
By
of
(iii)

{vi, ...,Vr,e,^, ...,ei^_^}

the preceding problem, a subset of S^ is a basis. But S contains contains n elements. Thus S is a basis of V and contains {vi,

n elements and every


. .

basis

.,v^} as a subset.

By
n

(ii),

an independent set T with n elements


Thus,

is

part of a basis.

But every basis of

contains

elements.

is

a basis.

5.19.

Prove Theorem

5.7:

Let

Whe a subspace
dim

of an 7i-dimensional vector space V.

dim W^n. In
Since

particular, if

W = n,

Then

then

W=V.

is

a basis of

of dimension n, any w + 1 or more vectors are linearly dependent. Furthermore, since consists of linearly independent vectors, it cannot contain more than n elements.

Accordingly, dim

W n.

it is also

In particular, if {w^, . , w} is a basis of W, then since = V when dim = n. a basis of V. Thus


. .

it is

an independent

set with

n elements

5.20.

Prove Theorem
Observe that

5.8:

dim(C7
is

+ W) = dim U + dim
. ,

W-

dim{Ur\W).

a subspace of both U and W. dim (Ur\W) = r. Suppose {v^, vj is a basis of UnW. to a basis of U and to a basis of W; say,
. .

U nW

Suppose dim

By Theorem

5.6(ii),

= n and m, dim we can extend {{}

{^1

Vr, Ml,

...,_,}
Let

and

{vi, ...,

vM)i, ..., w_,}

are bases of

and

W respectively.
B 7111,

{Vi,

...,V

Ml,

..,_ Wi,

.,

W-r}

Note that B has exactly basis of U-\-W. Since


generates

+nr
Uj}

elements.

{v^,

generates

Thus the theorem is proved if we can show that B is a and {v^, w^) generates W, the union B = {vj, Uj, w^}

U +W.
aiVi

Thus

it suffices to

show that

is

independent.

Suppose

a^v^

61M1

6_,.M_,.

Ci^i

c_,.m;_,.

(1)

where

Oj, bj,

c^ stte scalars.

Let

aiVi

a^Vr

61M1

+ bn-^u^_^

(2)

102

BASIS

AND DIMENSION

[CHAP.

By

(1),

we

also

have that
^n r *^n r
(S)

Since {vi,Uj} cU, v e Now {Vi) is a basis of

UnW
diVi

hy

(2);

and since

{w J c

TF,

:
. .

W
. ,

by
fo""

(5).

Accordingly,

vGUnW.
+

and

so there exist scalars di,

d,

which v

dj^i

d^v^.

Thus by

(^)

we have

d^V^

CiWi

C_rW_r

=
Cx

But

{vi, Wfc} is

c_, = 0.

and so is independent. a basis of Substituting this into (1), we obtain


aiVi

Hence the above equation forces

= 0,

+
is

a^Vr

bjUi

b^n-rUm-r

But {f
a,

j,

Uj} is a basis of

and so

independent.

Hence the above equation forces

= 0,

= 0,

61

= 0,

6to_^

= 0.
aj, 6^

and the theorem

Since the equation {1) implies that the is proved.

and

c^ are all 0,

B=

{vi, Uj,

w^}

is

independent

5.21.

Prove Theorem
Let

5.9:

The row rank and the column rank

of

any matrix are

equal.

be an arbitrary

m X w matrix:
jail
A

ai2
"'22

in ^2n

<*21
I

[O'ml

<'m2

'^r

Let Ri, B2

Bm

denote
i?j

its

rows:
(fflu,

O12,

<lin)>

>

^m

(<*ml> '*m2>

>

'*mn)

Suppose the row rank


Si

is

r and that the following r vectors form a basis for the


,

row
>

space:

(611, 612,

f>i),

S2

(621. 622.

>

^2n).

^r

i^rV ^r2.

*m)

Then each of the row vectors

is

a linear combination of the

S;:
/CjjOj.

R^
/?2

^^

^11*^1
'l'21'S'l

"" '^i2'^2

~r

' '

'

"T"

^" fe22'^2

+ +
' "

+
+

^2r"r

fin

"'ml"'l

"^"

"'m2"2

"

l^mr^r

where the

fcy

each other,

we

are scalars. Setting the ith components of each of the above vector equations equal to w: , obtain the following system of equations, each valid for i = 1,
. . .

li
ttgi

=
=

^^ll^li fe21*li

+ + +

^12621
*'22&2i

+ +

+
+ +

^Ir^ri
k2rbri

^mlbii

fcm2*2i

'

"

'

k^rbr

Thus for

1,

. ,

w:

In other words, each of the columns of

is

a linear combination of the r vectors

I'l

CHAP.

5]

BASIS

AND DIMENSION
has dimension at most
r, i.e.

103

rank
the

Thus the column space of the matrix row rank.

column rank

r.

Hence, column

Similarly (or considering the transpose matrix A*)

we

obtain

row rank column rank.

Thus

row rank and column rank are

equal.

BASIS
5.22.
(i)

AND DIMENSION
R^:
(1, 1, 1)

Determine whether or not the following form a basis for the vector space

and

(1,

-1,

5)
(3,

(iii)

(1, 1, 1),
(1, 1, 2),

(1, 2, 3)

and
and

(2,

-1,

1)

(ii)

(1, 2, 3), (1, 0,

-1),

-1,

0)

(iv)

(1, 2, 5)

(5, 3, 4)

and
(i)

(2, 1,
(ii).

-2)
is

and

No; for a basis of R3 must contain exactly 3 elements, since R^

of dimension

3.

(iii)

The vectors form a basis if and only if they are independent. rows are the given vectors, and row reduce to echelon form:

Thus form the matrix whose

to

The echelon matrix has no zero rows; hence the three vectors are independent and
basis for R^.
(iv)

so

form a

Form

the matrix whose rows are the given vectors, and

row reduce

to echelon form:

to

The echelon matrix has a zero row, i.e. only two nonzero rows; hence the three vectors are dependent and so do not form a basis for R3.

5.23.

Let

W be the subspace of R* generated by the vectors


Form
the matrix whose rows are the given vectors, and
/l 1

(1,

2,
(ii)

5,

3, 5). (i) Find a basis and the dimension of W. to a basis of the whole space R*.
(3, 8,
(i)

-3), (2, 3, 1, 4) and Extend the basis of

row reduce
1

to echelon form:

-2
7

3 3\ 2
to

-2
7

-3
2

to

-9 7-9
14

-9

-18
2)

The nonzero rows


space, that
(ii)

(1,

2,

5,

3) and

(0, 7,

9,

is,

of

W. Thus,

in particular,

dim

W=

of the echelon matrix


2.

form a basis of the row

We seek four
(0, 7,

independent vectors which include the above two vectors. The vectors (1, 2, 5, 3), 9, 2), (0, 0, 1, 0) and (0, 0, 0, 1) are independent (since they form an echelon matrix), and so they form a basis of R* which is an extension of the basis of W.

5.24.

Let

W be the space generated by the polynomials


V2

23 -st^'

+ Qt-l

Vi

2t^

~5P + 7t +

Find a basis and the dimension of W.


The coordinate vectors of the given polynomials
[vi]

relative to the basis {f3,

t^, t,

1} are respectively

[v^]

= =

(1,-2,4,1)

(2,-3,9,-1)

K] = K] =

(1,0,6,-5)
(2,-5,7,5)

104

BASIS

AND DIMENSION
row reduce

[CHAP.

Form

the matrix whose rows are the above coordinate vectors, and
1

to echelon form:

-2
1

4
1

-3

to

to

The nonzero rows (1, 2, 4, 1) and (0, 1, 1, 3) of the echelon matrix form a basis of the space generated by the coordinate vectors, and so the corresponding polynomials
3

2*2
2.

4t

and

t^

form a basis of W. Thus dim

W=

5.25.

Find the dimension and a basis of the solution space


x

W of the system
=
= =

+ 2y + 2z+ 2y +
3z
8z

s s s

+ + +

St
i

+
+

Sx

6y

5t

Reduce the system to echelon form:


X

2y ^

2z
z

2z

+ +

2s 4s

+ 3t = -2t = - 4 =

x
or

2y

+ 2zz +

2s

+ -

3t 2t

= =

The system

in echelon

solution space PT is
(i)

2 =

form has 2 (nonzero) equations in 5 unknowns; hence the dimension of the 3. The free variables are y, s and t. Set
(ii)
J/

= l,s = 0,t-0, =

= =

0, s

1,

0,

(iii)

0,

s^O, t-l

to obtain the respective solutions vi

(-2,

1, 0, 0, 0),

V2

(5, 0,

-2,

1, 0),

^3

(-7,

0, 2, 0, 1)

The

set {v^, V2, f 3} is a basis of the solution space

W.

5.26.

Find a homogeneous system whose solution


{(1,

set

is

generated by
-2, 5)}

-2,

0, 3), (1,

-1, -1,

4), (1, 0,

whose Method 1. Let v = (x, y, z, w). Form the matrix last row is v; and then row reduce to echelon form:
/I

first

rows are the given vectors and whose

1
I

IO-25II0
\x

-2 -1 -1
y

3\
4

/I

-2
1

/I

-2
1

-1 -2
y
z

\^\(iQ2x + y + z-hx-y + w^
\0

,.0

-1

\Q

2x

-3x + w/

tional

and only if the addihas dimension 2. Thus original first three rows show that row does not increase the dimension of the row space. Hence we set the last two entries to obtain the required homogeneous system in the third row on the right equal to

The

v&WH

2x 5x

+ +

=0
w =
and only
8(1,

y
if

Method

2.

We know
W:

that v

=
w)

(x,y,z,w)

P7
0, 3)

if

d
,

is

a linear combination of the gent(l, 0,

erators of

(x, y, z,

r(l,

-2,
r,

-1, -1,

4)

-2,

,.v

5)

The above vector equation

in

unknowns

and

t is

equivalent to the following system:

CHAP.

5]

BASIS

AND DIMENSION
= = 2x + y = z = w 3x
oe

105

= X = y -2r -s -s-2t = z 3r + 4s + 5t = w
r

r
^

+ t 8 + 2t -s-2t 8 + 2t
8

s 8

+ +

2t

= = = =

x 2x

2x
5x

+ + +

y y
y

+z w

^^^

Thus V

E.

W
is

ii

and only

if

the above system has a solution,

i.e. if

2x 5x

+ +

+ z =0 w =
(1) is

The above

the required homogeneous system. the transpose of the matrix

Remark: Observe that the augrmented matrix of the system

M used in the first method.


5.27.

Let

and

W be the following subspaces


{{a,b,c,d): b

of R*:

U =
We

+c+d =
(i)

0},

W
(ii)

{{a,b,c,d): a
(iii)

+b =

0, c

2d}

Find the dimension and a basis of


(i)

U,

W,

UnW.
b

seek a basis of the set of solutions


b

(a, 6, c, d)

of the equation

+
d

=
Set
(2)

or

0-a +

+
0,

=
a

The free variables are


(1)

a, c
0,

and

d. 0,

1,

= =

=
=

0, c

1,

(3)

0, c

0,

to obtain the respective solutions


^l

(1,0,0,0),

V2

(0,-1,1,0),
3.

1^3

(0,-1,0,1)

The
(ii)

set {v^, v^,

v^

is

a basis of U, and

dim U =
(a, 6, c, d)

We

seek a basis of the set of solutions

of the system

6
c

= =
1,

a
or

2d

+ 6 = - 2d = =
0,

The

free variables are 6 and d.


(i)

Set
6

0,

(2)

to obtain the respective solutions


Vy

(-1,

1, 0, 0),

V2

(0, 0, 2, 1)

The
(iii)

set {Ui,

v^

is

a basis of

W, and dim

W=

2.

U r^W

consists of those vectors

(a, b, c, d)

which satisfy the conditions defining

and the con-

ditions defining

W,

i.e.

the three equations

6+c+d=0
0+6
c

a+b
or
6

=0
+

=0
=
2d

c+d
c

2d

= =
is

The free variable is d. Set d = 1 of VnW, and Aim (V nW) = \.

to obtain the solution

(3, -3, 2, 1). Thus {v}

a basis

5.28.

Find the dimension of the vector space spanned by:


(i)

(1, (3,
t^

-2,3, -1) and


-6,
3,

(1, 1,

-2, 3)
4,

(v)

{\
/
1

and

^J
l\

(ii)

-9) and (-2,


3i

-2,

6)

/-3 -3

(iii)

2f2

+ +

and
i^

2t^

+
4

4*^

6f

^^^^

(^-i -i^
3 and

(iv)

i-2f2

5 and

34

(vii)

-3

106

BASIS

AND DIMENSION

[CHAP.

of dimension 2 if they are independent, and of dimension nonzero vectors span a space they are dependent. Recall that two vectors are dependent if and only if one is a multiple of the other. Hence: (i) 2, (ii) 1, (iii) 1, (iv) 2, (v) 2, (vi) 1, (vii) 1.

Two

1 if

5.29.

Let

V
An

dim 7 =

Show that be the vector space of 2 by 2 symmetric matrices over K. 3. (Recall that A = (ay) is symmetric iff A = A* or, equivalently, an = an.)
A =
(iii)

arbitrary 2 by 2 symmetric matrix is of the form (Note that there are three "variables".) Setting
(i)

I
(

h\
1

where

a, 6, c

G A.

1,

0, c

0,

(ii)

0,

1,

0,

0,

0, c

we

obtain the respective matrices

-.

= (;:)
is

-. =

(?;)
is,

- =
(1)

(::)
V
and
(2)

We

show that {E^, E^, E^)


(1)

a basis of V, that

that

it

generates

is

independent.

For the above arbitrary matrix

in V,

we have
aEi

A =
Thus
{^1, E^, Eg} generates V.

("'

^\

bE2

cEs

(2)

Suppose xEi

yE^

zE^

0,

where

x, y, z are

unknown

scalars.

That

is,

suppose

Ki

o)

+ ^(i +
is

J)

K2
=

1)

Oj

\y
x
0,

zj
0, z

\0

Setting corresponding entries equal to each other,

we

obtain

0,

0-

In other words,

xEi
Accordingly, {EijEz.Ea}

yE2

zEs

implies

0,

independent.

Thus

{El, E2, E3} is a basis of

and so the dimension of

is 3.

5.30.

Let V be the space of polynomials in is a basis of V:


(i)

of degree

^ n. Show that each


. . .

of the following

{1,

t^

"->, },
1.

(ii)

{1, 1

- 1,

(1

- tf,

(1

- )"-s

(1

- )"}.

Thus dim V = n +
(i)
. .

...,t"-i and t. Furthermore, Clearly each polynomial in y is a linear combination of l,t, none is a linear combination of the preceding poly1, t, ., t"-i and t are independent since t} is a basis of V. nomials. Thus {1, t

(ii)

form a basis of (Note that by (i), dim V = w+ 1; and so any m-l- 1 independent polynomials (1 - t) is of degree higher than the Now each polynomial in the sequence 1, 1 - t w + 1 polypreceding ones and so is not a linear combination of the preceding ones. Thus the (1 t) are independent and so form a basis of V. nomials 1, 1 t,
V.)
. . . ,

5.31.

Let

the vector space of ordered pairs of complex numbers over the real (see Problem 4.42), Show that V is of dimension 4.

V be

field

We

claim that the following

is

a basis of V:
{(1, 0),
z,
(i,

B =
Suppose u e V. Then v = (, w) where o, 6, c, d are real numbers. Then
V

0), (0, 1), (0, t)}

are complex numbers, and so v

= (a+bi,e + di)

where

ail, 0)

6(t, 0)

c(0, 1)

d(0,

t)

Thus

generates V.

CHAP.

5]

BASIS

AND DIMENSION
B
is

107

The proof

is

complete

if

we show

that

independent.
a;3(0, 1)

Suppose

xi(l, 0)

X2(i, 0)

0:4(0, i)

=
-'r

where

Xi, x^, x^,

X4

G R. Then
{xi

+ x^,
0,

Xs

+ x^x) =
0, a;4

Xx
(0, 0)

x^i x^i

and so
3
-V

Accordingly

x-^

0, a;2

as

and so

is

independent.

5.32.

Let Y be the vector space of with 1 as the i^-entry and

m x w matrices over a field K.


elsewhere.

Show
is

that {E-,^

Let E^ G V be the matrix Thus is a basis of 7.

dimV = mn.
We
Let

need to show that {By} generates

and

independent.
o,^E^^.

A=

(tty)

be any matrix in V. Then

A = 2
{.-

Hence {By} generates V.

Now

suppose that
is 0.

ajjii^ij

= =
0,

where the
i

are scalars.
j

The y-entry of

2 ij^ij

is .

and
are

the y-entry of independent.

Thus

asy

1,

.,

w,

m.

Accordingly the matrices

By

Thus {By}

is

a basis of V.

Remark: Viewing a vector in K as a 1 X w matrix, we have shown by the above result that the usual basis defined in Example 5.3, page 88, is a basis of X" and that dim K" = w.

SUMS AND INTERSECTIONS


5.33.

Suppose
sion 6.

are distinct 4-dimensional subspaces of a vector space and Find the possible dimensions of TJV^W.
TJ

of dimen-

are distinct, V -VW properly contains 17 and W; hence dim(f7+W)>4. Since V and But dim(?7+W) cannot be greater than 6, since dimV = 6. Hence we have two possibilities: = 5, or (ii) dim (U + PF) = 6. Using Theorem 5.8 that dim(f7+ T^) = dim U (i) dim(U+T7) dim (Un TF), we obtain dim

(i)

= =

4 4

+
+

4
4

-dim(f/nW) -dim(?7nW)
3.

or or

dim(t7nW)
dim(t/nTF)

=
=

3 2

(ii)

6
TJ

That

is,

the dimension of

r\'W must be either 2 or

5.34.

Let

J]

and

W be the subspaces of R* generated by


(1, 2, 8, 0), (2, 3, 3,

{(1, 1, 0,

-1),

-1)}
TF),

and
(ii)

{(1, 2, 2,

-2),

(2, 3, 2,

-3),

(1, 3, 4,

-3)}

respectively.
(i)

Find

(i)

dim (C/ +

dim(C7nW).

TJ-^W is the space spanned by all six vectors. Hence form the matrix whose rows are the given six vectors, and then row reduce to echelon form:

to

to

to

Since the echelon matrix has three nonzero rows, dim

iJJ

-VW)

Z.

108

BASIS

AND DIMENSION

[CHAP.

(ii)

First find dim U and dim W. Form the two matrices whose rows are the generators of respectively and then row reduce each to echelon form:

and

1 1

-1
3 3 2 2
to

1
1

-1
3
1
0.

2
3

to

-1

and
1

2
3

2
1

-2 -3 -3

'l

-2
1

to

to

-1 -2

Theorem

Since each of the echelon matrices has two nonzero rows, dim V - dim (UnW), 5.8 that dim (V +W) = dim U + dim

and dim

W=

2.

Using

we have
1

+2-dim(!7nW)

or

Aim{Ur\W) =

5.35.

Let

be the subspace of R^ generated by


{(1, 3,

-2,

2, 3), (1, 4,

-3,

4, 2), (2, 3,

-1, -2, 9)}

and

let

W be the subspace generated by


{(1, 3, 0, 2, 1), (1, 5,

-6,

6, 3), (2, 5, 3, 2, 1)}


(ii)

Find a basis and the dimension of


(i)

(i)

XJ

+ W,

f/n W.

U +W

is

six vectors

the space generated by all six vectors. and then row reduce to echelon form:
1
1

Hence form the matrix whose rows are the


/I

4
3 3
5

2
1 1

-2 2 -3 4 -1 -2
2

2
9
1

*
'

3-2 2 3\ 1-1 2-1 0-3 3-6 0-2 2

-6
3

6
2 2

3
1

2-440
7

2
1

5 3
1

\0
1

-1
3
1

-2 -5/
2 2 3

-2 -1
2

-1

-2 -1
2

-1 -2

to

-2
2

to

-2
6

-6

The

set of

nonzero rows of the echelon matrix,


{(1, 3,

-2,
(t/

2, 3), (0, 1,

-1,

2,

-1),

(0, 0, 2, 0,

-2)}

is

a basis

oiV+W;

thus dim

TF)

3.

(ii)

whose

First find homogeneous systems whose solution sets are U and first rows are the generators of U and whose last row is
to echelon form:

W respectively.
(, y, z, s, t)

Form the matrix and then row reduce


3

-2 -1
3
-3x

2 2

-1
3
s

-6
z

2x

-2x +

Sx +

t
j

4a;

CHAP.

5]

BASIS

AND DIMENSION
to obtain the

109

Set the entries of the third row equal to set is U:

homogeneous system whose solution


-6a;

-X + y +

Q,

4a;

22/

0,

-\-

=
last

Now

(x, y, z, 8, t)

form the matrix whose first rows are the generators of and then row reduce to echelon form:

and whose

row

is

to

-9aj

3y

'Ix

2y

2x

Set the entries of the third

row equal
z

to

to obtain the

homogeneous system whose solution


2x

9x + 3y +

0,

4x

2y

0,

=
set is

Combining both systems, we obtain the homogeneous system whose solution

U nW:

x+y + z
-2y -6a; + y -9x + 3y + 4a; 2j/ 2a; 4x
J/

+8
+
z
t

=0 =0
= -

x+y+z
+ -5y -6y 2y + ^ +
2y
-x
Az 6z

+
+
s

8z
iz 2z

=0
+
t

=0 =0
+
t

x +

y+z
+
4z
8z 4z

2y

+ + +

=0 =0
+
2t

y+z
+
Az
8z

=0
+ +
8

58
3s
s

2y

=0
2t

5s
8

+ -

2f 2t

= = =

There

is

solution

one free variable, which is t; hence dim(l7nT^ = 1. Setting t = 2, we obtain the Thus {(1, 4, -3, 4, 2)} is a basis of UnW. a; = 1, 2/ = 4, z = -3, 8 = 4, t = 2.

COORDINATE VECTORS
5^6.

Find the coordinate vector of v relative to the basis where (i) v = (4, -3, 2), (ii) v = (a, 6, c).

{(1, 1, 1), (1, 1, 0), (1, 0, 0)}

of R^

In each case set v aa a linear combination of the basis vectors using

unknown

scalars x, y and

z:

a;(l, 1, 1)

j/(l, 1, 0)

z(l, 0, 0)
is

and then solve for the solution vector


linearly independent.)

{x,y,z).

(The solution

unique since the basis vectors are

(i)

(4,-3,2)

= = =

a;(l, 1, 1)
(a;,

+ j/(l, 1, 0) + z(l, 0, X, x) + {y, y, 0) + (z, 0, 0) (x + y + z,x + y,x)


x

0)

Set corresponding components equal to each other to obtain the system

A,

X + y

3,

a;

Substitute = 2 into the second equation to obtain y = 5; then put x = 2, y = 5 into the first equation to obtain z = 7. Thus x = 2, y = -5, z = 7 is the unique solution to the system and so the coordinate vector of v relative to the given basis is [v] = (2, 5, 7).

110

BASIS

AND DIMENSION

[CHAP.

(ii)

{a, b, c)

(1, 1, 1)

+
y

Then
from which x {e, b c, a b).

c,

+ z{l, 0, 0) = (x + y + z,x + y,x) x + y + z = a, x + y = b, x c b c, z = ab. Thus [v] (c,b c,a


1/(1, 1, 0)

b),

that

is,

[(a, b, c)]

5JS7.

Let V be the vector space of 2 x 2 matrices over R. matrix A relative to the basis

Find the coordinate vector of the

GV

{{I
Set

iHri)^{i-iHi
-

I)}

w- -(!-?
unknown
scalars x, y,
z,

as a linear combination of the matrices in the basis using

w:

J)

"I i)*'(:-i)-'(i-i)*<i
w
\X
'

x)

'^

\y
X

0/

\0
z^

0/

+
X

+w
y

y
X

Set corresponding entries equal to each other to obtain the system

from which

ordinate vector

x 1 x + y =^ A, + w = 2, X y z = 3, = ll, z- -21, w = 30. Thus [A] - (-7, 11, -21, 30). (Note -l, y of A must be a vector in R* since dim V = 4.)

that the co-

5^8.

Let

W be the vector space of 2 x 2


1

symmetric matrices over R.

(See

Problem

5.29.)

Find the coordinate vector of the matrix

^
[

, ^

relative to the basis


)

-2\

/2

1\

4 -1\1

-2
Set

1/' vi

sy
=

\-i -5/j
-2\ ij +
,

as a linear combination of the matrices in the basis using unknown scalars


4

x,

y and

z:

-11\

^ = (_n

-7J

/I %-2

/2

1\

^l

3J

%-l

/x + 2y + 4:Z -2x + y - z -1\ -5) = (,-2x + y-. . + 3,-5.

Set corresponding entries equal to each other to obtain the equivalent system of linear equations and reduce to echelon form:

2y

Az
z z

2x + y -2x + y X

Sy

5z

= 4 -11 = -11 = -7

2y 5y
J/

+ + -

iz
7z

9z

= 4 - -3 = -11

X or

2y 5y

+ +

iz
Iz

52z

- 4 - -3 - 52

We
(4,

obtain
1).

from the
2,

first equation.

from the third equation, then y = 2 from the second equation, and then a = 4 Thus the solution of the system is x = 4, y 2, z 1; hence [A] = 3 by Problem 5.29, the coordinate vector of A must be a vector in K*.) (Since dim

5.39.

Let

{ei, 62, es}

and

{/i,

h,

fa}

be bases of a vector space


ei

(of

dimension

3).

Suppose

62
63

= =

Ci/i

+
+

ttaA

hifi+b^fi
Ci/i

C2/2

+ 0-3/3 + hifz + C3/3


ei, ea

(i)

Let

P be the matrix whose rows are the coordinate vectors of


{fi}:

and

es respectively,

relative to the basis

'

CHAP.

5]

BASIS

AND DIMENSION
{tti
61
Cl

111

02
&2
C2

as
63
C3

Show that, for any vector v GV, [v]eP = [vy. That is, multiplying the coordinate vector of v relative to the basis {ei} by the matrix P, we obtain the coordinate vector of V relative to the basis {/<}. (The matrix P is frequently called the change of basis
matrix,)
Suppose V
V

rei

seg

teg;

then

[v]^

(r,8,t).

Using

(i),

we have

+ O2/2 + agfa) + si^ifi + ^2/2 + ^3/3) + *(i/i + "2/2 + "3/3) = (roi + s6i + tci)/i + {ra2 + S62 + fc2)/2 + {ras + sb^ + tcs)^
r(aJi
[v]f

Hence

= =

{rai

+ sbi + tc^,
/

ra2

+ sb2+

tC2,

ra3

+ sb3+

tcg)

On

the other hand,


[v],P
(r, s, t)

ai
61

a2
62
C2

(I3

63
C3

Cl

(rtti

+ s6i + tCj,

ra2

+ 862 + tc2,

ra^

+ S63 + tcg)
column vectors rather than row
tCx

Accordingly, [v\eP

['"]/

vectors.

Remark: In Chapter 8 we Then, by above,

shall write coordinate vectors as

'ax

61

Ci\/r\
2l|S
Cg/
\ t

rai

QMe
where
is

a2
I

62

ra2
irfflg

&3

+ + +

+ 362 + S&3 +
s6i

*C2

tCgy

Q the matrix whose columns are the coordinate vectors of i, e^ ^^^d respectively, relative to the basis {/J. Note that Q is the transpose of P and that Q appears on the left of the column vector \v\g whereas P appears on the right of the row vector [1;]^.

RANK OF A MATRIX
5.40.

Find the rank of the matrix /I


(i)

where:

A =

u
l^ \3

-2 -1 4 3 3 -4 -7 8 1 -7
3
1

-3 -4 -3 -8

(ii)

A =

(iii)

(i)

Row

reduce to echelon form: /I


1

1-2

-3\
to to

4
3 8

3-1-4
-4 -7 -3 1-7 -8/

2
3

Since the echelon matrix has two nonzero rows, rank (A)
(ii)

Since row rank equals column rank, reduce to echelon form:

it is

easier to

form the transpose of

and then row

to

to

Thus rank (A)

3.

112

BASIS

AND DIMENSION
is

[CHAP.

(iii)

The two columns are linearly independent since one rank (A) = 2.

not a multiple of the other.

Hence

5.41.

Let A and B be arbitrary matrices for which the product rank (AB)^ rank (J?) and rank (AB) ^ rank (A).

AB

is

defined.

Show

that

By Problem 4.33, page 80, the row space of AB is contained in the row space of B; hence rank (AB) rank (B). Furthermore, by Problem 4.71, page 84, the column space of AB is contained in the column space of A; hence rank (AB) rank (A).

5.42.

Let

be an n-square matrix.

Show

that

is

invertible if

and only

if

rank (A)

= n.

Note that the rows of the w-square identity matrix / are linearly independent since / is in echelon form; hence rank (/) = n. Now if A is invertible then, by Problem 3.36, page 57, A is row equivalent to /; hence rank (A) = n. But if A is not invertible then A is row equivalent to a matrix with a zero row; hence rank (A) < n. That is, A is invertible if and only if rank (A) = n.

5.43.

Let JCij, Xi^, a;i^ be the free variables of a homogeneous system of linear equations with n unknowns. Let Vj be the solution for which: x^ 1, and all other free variables = 0. Show that the solutions Vi, V2, .,Vk are linearly independent.
. . . , .
.

ii,

Let A be the matrix whose rows are the Vi respectively. We interchange column 1 and column then column 2 and column ^2, ., and then column k and column i^l and obtain the kXn matrix
. .

"l.k +
''2,

Ci C2

(/,

C)

\o

...

Cfc,fc

+i

...

c/

The above matrix B is in echelon form and so its rows are independent; hence rank (B) = k. Since and B are column equivalent, they have the same rank, i.e. rank (A) = k. But A has k rows;
i.e.

hence these rows,

the

iij,

are linearly independent as claimed.

MISCELLANEOUS PROBLEMS
5.44.

The concept
as follows:
Vi^,
. .

of linear dependence is extended to every set of vectors, finite or infinite, the set of vectors A = {vi} is linearly dependent iflf there exist vectors a S Z, not all of them 0, such that .,Vi^G A and scalars ai,
. . .

aiVi^

aiVi^

ai7i

=
. . .

are linearly Otherwise A is said to be linearly independent. Suppose that Ai, A2, independent sets of vectors, and that AiCAzC---. Show that the union A = A1UA2U is also linearly independent.

Suppose A is linearly dependent. Then there exist vectors v^ K, not all of them 0, such that
a^Vi

v^& A and

scalars

!,...,

a2V2

a^Vn
. .

=
A;^ such that

(1)

Since

A = uAj

and the

Wj

S A,

there exist sets A,,

.,

v^eAi^, v^eAi^,
Let k be the

...,

veAi^
. . . ,

maximum
is

index of the sets

Aj.:

that each Aj.

contained in A^. Hence

Ui,'y2,

= max (ij, -.-yn 6 -^k

i).

It follows then, since


so,

Ai C Ag c

and

by

{1),

A^

is

linearly dependent,

which contradicts our hypothesis.

Thus

is

linearly independent.

CHAP.

5]

BASIS

AND DIMENSION
. .

113

5.45.

Consider a finite sequence of vectors S {vuV2, .,v}. Let T be the sequence of vectors obtained from S by one of the following "elementary operations": (i) interchange two vectors, (ii) multiply a vector by a nonzero scalar, (iii) add a multiple of one vector to another. Show that S and T generate the same space W. Also show that T is independent if and only if S is independent.
Observe that, for each operation, the vectors in T are linear combinations of vectors in S. On the other hand, each operation has an inverse of the same type (Prove!); hence the vectors in S are linear combinations of vectors in T. Thus S and T generate the same space W. Also, T is inde n, and this is true iff S is also independent. pendent if and only if dim

5.46.

Let
vi,
.

A
.

(ay)

.,Vn be Ml

and B = (by) be row equivalent mXn matrices over a any vectors in a vector space V over K. Let
ttiiVi

field

K, and

let

Ui

a^iVl

+ + +

anVi
a22V2

+ + +

-!- ainVn

+
"I"

a2nVn

= W2 =
Wi

6iit;i

-I-

bi2V2

+-!- binVn
-I

b2lVl

+ +

b22V2

-I-

&2nVn

Um =

OmlVl

am2V2

OmnVn

Wm =
space.

bmlVl

bm2V2

bmnVn

Show

that [Vd] and (w.} generate the

same

Applying an "elementary operation" of the preceding problem to {mJ is equivalent to applying an elementary row operation to the matrix A. Since A and B are row equivalent, B can be obtained from A by a sequence of elementary row operations; hence {tyj can be obtained from {mJ by the corresponding sequence of operations. Accordingly, {mj} and {wj generate the same space.

5.47.

Let

Vi,

.,Vn belong to a vector space

V
-I-

over a
ai2V2

field

K.

Let

= W2 =
Wi

aiiVi a2iVi

ainVn

+ +

a22V2

+-!- a2nVn

Wn
where
(i)

dnlVl

an2V2

"I"

"I"

annVn

Oij

S K. Let P

be the n-square matrix of

coefficients, i.e. let

P=

{an).

Suppose P is invertible. Show that {wi) and {vi} generate the same space; hence {wi) is independent if and only if {vi) is independent. Suppose

(ii)

is

not invertible.
is

Show

that {wi)
is

is

dependent.

(iii)

Suppose {wi)
Since
other

independent.

Show that P

invertible.

(i)

P
is.

is invertible, it is

problem {wj and

{vi^

row equivalent to the identity matrix /. Hence by the preceding generate the same space. Thus one is independent if and only if the

(ii)

P is not invertible, it is row equivalent to a matrix with a zero row. This means that generates a space which has a generating set of less than n elements. Thus {wj is {wit dependent.
Since

(iii)

This

is

the contrapositive of the statement of

(ii),

and so

it

follows

from

(ii).

5.48.

Suppose
(i)

V
.

is
.

the direct
.,tfot}cC/
(ii)

sum
and

of its subspaces
{wi,
. . . ,

and W,

i.e.

V = IIW. Show
is

that:

if

(Ml,

Wn)

CW
+

are independent, then {Vn,Wj)

also

independent;
(i)

dim V = dim

U + dim W.
6iWi +-!- bw

Suppose

a^Ui

(i,m,

+
-I-

0,

where

ttj,

bj

are scalars.

Then

(ajMi

-h ttrnWrn)

(^I'^l

+ bWn) =

114

BASIS

AND DIMENSION
and
0,

[CHAP.

where

0,

a^u^

+ a^Mm ^ U
axUi

h^Wi
0,

6n'"'n

'^^

Since such a

sum

for

is

unique, this leads to

+ aMm =

bjWi

h^w^^

The independence of the tij implies that the Oj are all 0, and the independence of the Wj implies that the 6j- are all 0. Consequently, {mj, Wj} is independent.
(ii)

Method 1. Since 5.8, page 90,

-U W,
+
diml?F

we have V
dim(f/nW')
. .
.

=U+W
=
dim

and

Ur\W -

{0}.

Thus, by Theorem

dimy
Method

dimC7

1/

+ dim W -

= dim U +

dim T^

w^ are bases of U and 2. Suppose {u^, ...,u^) and {ti, respectively, {mj, Wj) generates V = V +W. they generate TJ and Thus {u^, w^ is a basis of V; hence dim V (i), (mj, Wj) is independent.

W
On

respectively. Since the other hand, by

= dim ?7 +

dim W.

5.49.

Let [/ be a subspace of a vector space oiV such that F = a subspace

of finite dimension.

Show
it

that there exists

W
. .

UW.
is linearly

Let

{mj,

m,} be a basis of U.
.

Since {mJ

independent,

can be extended to a
. . . ,

basis of V, say, {mi, {Mj,Wi} generates V, V

m ^i = V + W.
.

Since w,}. be the space generated by {w^, wJ. Let On the other hand, C/ n H' = {0} (Problem 5.62). Accordingly,

y =

?7

T17.

5.50.

(or: ? is an extension of K), then is a subfield of a field Recall (page 65) that if may be viewed as a vector space over K. (i) Show that the complex field C is a vector space of dimension 2 over the real field R. (ii) Show that the real field R is a vector space of infinite dimension over the rational field Q.

(i)

We

where x + yi

claim that {l,i} is a basis of C over R. For if v&C, then v = a + hi = a'l-\-b'i that is, {1, i} generates C over R. Furthermore, if x'l + y'i = Q or a, 6 e K; - 0, where a;,j/ S R, then a; = and y = 0; that is, {l,i} is linearly independent over R. Thus {1, t} is a basis of C over R, and so C is of dimension 2 over R.
. .

(ii)

We

For suppose is linearly independent over Q. claim that, for any n, {1, Tr.tr^, ., tj-"} + a7r" = 0, where the aj G Q, and not all the aj are 0. Then ;r is a aol + ffliff + a^ifi + + ox". But it can be root of the following nonzero polynomial over Q: a^ + a^x + ajOi^ + is a transcendental number, i.e. that ir is not a root of any nonzero polynomial shown that .,jr" are linearly independent over Q. real numbers 1, jr, jt-^, over Q. Accordingly, the Thus for any finite n, R cannot be of dimension n over Q, i.e. R is of infinite dimension over Q.

tt-

w+1

5.51.

Let

a subfield of a field E: Suppose that E is of dimension n over L and subfield of E.) over K. Show that E is of dimension mn over K.
2:

be a subfield of a

field

L and L

KgLcE.
L
is

(Hence

K is

of dimension

Suppose
that {aiVj-. elements.

{vi,
i

.,i;} is

l,...,m,j

.,a} is a basis of L over K. We claim a basis of E over L and {a^, l,...,n} is a basis of E over K. Note that {aji^j} contains mn
. .

be any arbitrary element in E. bination of the Uj with coefficients in L:

Let

Since

{i;i,

...,v} generates

over L,

is

a linear com-

Since {oi,

a^)

W = 6i1)i + b2V2 + + generates L over K, each 64 S L


fen'Wn.

&{

S^
a,-

W
with co-

is

a linear combination of the

efficients in

K:
61

fciitti

fejafta

fcim^m

bn

'^nll

'^n22

'

"

'

Km"'n

CHAP.

5]

BASIS

AND DIMENSION
obtain

115

where k^

K. Substituting in

(1),

we

w = -

(kiitti H

^^U"!"!

+ ki^ajvi + (fcaitti + + A;2o,)v2 + + (Kia^ + + k^ajv + ki^amVl + 'C21l'"2 + + fe2mm''2 + + k^iaiV^ + + fcaV

"

'

'

"

"

i.}

where
erates

fc^j

K. Thus

is

a linear combination of the ajVj with coefficients in K; hence {ojVj} gen-

over X.
is

The proof
Xj,

complete

if

we show
is,

that {a^Vj} is linearly independent over K.

Suppose, for scalars

e X,

n(aiVj)

0;

that

{xiidiVi

+ Xi2a2Vi +
+ Kiattj +

+ Xi,a,Vi) +

+
+

{x^iaiV^
(a;ieii

+ x^^ch.i'n +

+ a;mav) =

or
.

(ajiitti

+ Xi,^a)vi +

+ x2a2 +

+ x^a)v =

Since {vi, ., v} is linearly independent over L, each coefficient must be 0:


.

L
.

and since the above


x^^Ui

coefficients of the Vj belong to

Kiiffli

Xi2a2

a;imm

= =

0,

.,

x2a2
ar^i

+ =
a;m

But

{ffli,

is linearly

independent over K; hence since the

K,
0,

ii

0,

Xi2

im

0,

...,

0,

a;2

...,

Accordingly,

{ttjVj} is

linearly independent over

and the theorem

is

proved.

Supplementary Problems
LINEAR DEPENDENCE
5.52.

Determine whether u and v are linearly dependent where:


(i)

(1, 2, 3, 4),

(4, 3, 2, 1)

(iii)

m=
u

(0, 1),

=
v

(0,

-3)
(0, 0,

(ii)

u=

(-1,

6,

-12), V

(|,

-3,

6)

(iv)

(1, 0, 0),

-3)

(vii)

u = -fi +

|t2

- 16,

^f3

- :^t2 + 8

(viii)

m=

t^

+ 3t + 4^

-y

t^

+ 4t + 3

5.53.

Determine whether the following vectors in R* are linearly dependent or independent:


(3,8,-5,7), (2,9,4,23);
(ii)

(i) (1, 3,

1,

4),

(1,-2,4,1),

(2,

1,0,-3),

(3,

-6,1,4).

5.54.

Let V be the vector space of 2 X 3 matrices over R. are linearly dependent or independent where:
1

Determine whether the matrices


/3 -8
7

A,B,C B V

-2

3\

_ /I -1
4
5

4\

-2 7

'

10

-1

<"'^=a4-:).
5.55.

-(4
3, 3,

I'D
1,

^=(r4-s
3 over R.

Let be the vector space of polynomials of degree linearly dependent or independent where:
(i)

Determine whether u,v,w

GV

are

u
u

=
-

fS

4*2
5^2

(ii)

t3

+ _

2t 2t

+
+

V
V

= -

ts

2*2

4t

t^

-4t^ -3t

A,

w w =

2fi 2fi

fi

3t

7fi

It

116

BASIS

AND DIMENSION
Show
/(*)

[CHAP.

5.56.

Let V be the vector space of functions from R into R. ent where: (i) f{t) = e, g(t) = sin t, h(t) = th (ii)
g{t)

that f,g,h
e,

eV

are linearly independ-

g(t)

e^*,

hit)

t;

(iii)

/(f)

e,

=
C

sin

t,

h(t)

cos

t.

5.57.

Show
field
(7, 1

that:

(i)

the vectors

(1

i, i)

and

(2,

1 + i)

in C^ are linearly dependent over the complex


field

but are linearly independent over the real


)

R;

(ii)

the vectors (3 -I-V2, 1

+ \/2)

and

+ 2'\/2
u,

in R2 are linearly dependent over the real field

but are linearly independent over the

rational field Q.
5.58.

Suppose
(i)

(ii)

v and w are linearly independent vectors. Show that: 2w, u v w and u + w are linearly independent; u+V u + V 3w, u + 3v w and v + w are linearly dependent.
u,

5.59.

is

Prove or show a counterexample: If the nonzero vectors a linear combination of u and v.


Vi, v^,
.

v and

are linearly dependent, then

5.60.

Suppose
(i)

. ,

f are linearly independent vectors.


>

Prove the following:


eij

{oi'Wi, a2>'2<

"n'^'n} is
.

linearly independent
.,

where each

(ii)

{vi,

.,v^-l,w,v^^l,
0.

v}

is

linearly independent

where

# 0. w = b^v^ +

b^Vi

bv

and
5.61.

5i #

Let V

ad
5.62.

be

= (a, 6) = 0.
{ui,
. .

and

(c,

d)

belong to K^.

Show

that {v,

w}

is linearly

dependent

if

and only

if

Suppose
L(Mi)

m^,

n L{Wj) =

{0}.

Show that is a linearly independent subset of a vector space V. , Wg} Wi, (Recall that L(Mj) is the linear span, i.e. the space generated by the Mj.)
. .

5.63.

Suppose
Ui,
. . .

, a^^) a^'e linearly independent vectors in Z", and suppose (a^, ..., ai), , (ttmi, v are linearly independent vectors in a vector space V over K. Show that the vectors
. .
.

Wi

au^'i

ffliB^n,

.,

Wm =

a^i^i

mn1'K

are also linearly independent.

BASIS
5.64.

AND DIMENSION
Determine whether or not each of the following forms a basis of
(i)

R^:

(1, 1)
(2, 1),

and
(1,

(3, 1)

(iii)

(0, 1) (2, 1)

and

(0,

-3)
87)

(ii)

-1) and

(0, 2)

(iv)

and (-3,

5.65.

Determine whether or not each of the following forms a basis of R^:


(i)

(1, 2, (2, 4, (1, 5,


(1, 3,

-1) and
-3),

(0, 3, 1)

(ii) (iii)

(0, 1, 1)

and
(3,

(0, 1,

-1)

-6),
-4),

(2, 1, 8), (1, 4,

-1,

4)

and

(2, 1, 1)

(iv)

-3) and

(2, 3,

-11)

5.66.

Find a basis and the dimension of the subspace


(i)

W of R* generated by:
7)

(1, 4,
(1,

-1,

3),

(2, 1,
(1,

-3, -1) and -3, -1,


2)

(0, 2, 1,
(3,

-5)

(ii)

-4, -2,

1),

and

-8, -2,

5.67.

Let

be the space of 2 X 2 matrices over

and

let

W be the

subspace generated by

ui).
5.68.

(-:

\y (-r:)

and

(4-;;

Find a basis and the dimension of W.


Let

W be the space generated by the polynomials


tt

(3

2*2

2t

1,

t)

t*

3*2

w =

2*3

t2

7t

Find a basis and the dimension of W.

CHAP.

5]

BASIS

AND DIMENSION

117

5.69.

Find a basis and the dimension of the solution space


X X 3x

W of each homogeneous system:


= = + 2x+
X
4:y

+ + +

Sy 5y 5y

+ + +
(i)

2z
z

8z

= =

+ -

2y
3y
y

2x 2x

+ -

7z

2z
z

+
(li)

+ +

2z 5z

= =

(iii)

5.70.

Find a basis and the dimension of the solution space


X

W of each homogeneous system:


x
2x 2x

X
2x

+ 2y ~2z + 2s - t = + 2y - z + 3s - 2t = + 4y ~ Iz + s + t =
(i)

+ + +

2y
4y iy

2z 2z

+ 3s-4t = - s + 5t = + 4:S -2t (ii)

5.71.

Find a homogeneous system whose solution


{(1,

set

is

generated by
-3),
(1,

-2,

0, 3,

-1),

(2,

-3,

2, 5,

-2,

1, 2,

-2)}

5.72.

Let

V and

W be the following subspaces of R*:


V =
{(a,b,c,d):

6-2c + d =
(i)

0},

W
(iii)

{{a,b,c,d):

d,

2c}

Find a basis and the dimension of


5.73.

V,

(ii)

W,

VnW.

n.

Let V be the vector space of polynomials in following is a basis of V:


(i)

of degree

Determine whether or not each of the

{1, 1

+ t,

(ii)

{1

+ t,

+ t+t2, l+t+t2 + t3, ..., l + t+t2+ t+ fi, t^ + t, ..., t"-2 + t"-i, t-i + t"}.
l

+t"-l +

t}

SUMS AND INTERSECTIONS 5.74. Suppose V and W are 2-dimensional


5.75.

subspaces of R3.

Show
4,

that

VnW

{0}.

Suppose

and

W are

subspaces of

and that dim

U=
U-

dim

W=5
W=2

and dim

V-

7.

Find the

possible dimensions of

U nW.
dim
1,

5.76.

Let U and P7 be subspaces of R3 for which Rs = r; w'.

dim

and

UlW.

Show

that

5.77.

Let

U be
let

the subspace of Rs generated by


{(1, 3,

-3, -1, -4),

(1, 4,

-1, -2, -2),

(2, 9, 0,

-5, -2)}

and

W be
dim

the subspace generated by


{(1, 6, 2,

-2,

3), (2, 8,

-1, -6, -5),

(1, 3,

-1, -5, -6)}

Find
5.78.

(i)

{U+W),

(ii)

dim (t/n VT).


Let

Let
{t3

be the vector space of polynomials over R.


t3

+ 4*2 - t + 3,

+ 5*2 + 5^
(i)

3*3

10(2

- 5t + 5}
(ii)

and

U and W be the subspaces generated by {t^ + U^ + &,t^ + 2t^-t + 5, 2*3 + 2*2 - 3* + 9}

respectively.

Find

dim

(f/

W'),

i\m.(VnW).

5.79.

Let

U
let

be the subspace of RS generated by


{(1,

-1, -1, -2,

0), (1,

-2, -2,

0,

-3),

(1,

-1, -2, -2,

1)}

and

W be the

subspace generated by
{(1,

-2, -3,

0,

-2),

(1,

-1, -3,

2,

-4),

(1,

-1, -2,

2,

-5)}

(i)

Find two homogeneous systems whose solution spaces are


Find a basis and the dimension of

and W, respectively,

(ii)

U r\W.

118

BASIS

AND DIMENSION

[CHAP.

COORDINATE VECTORS
5.80.

Consider the following basis of B^: {(2, 1), (1, -1)}. Find the coordinate vector of vSU^ v = (a,b). to the above basis where: (i) i; = (2,3); (ii) v = (4,-1), (iii) (3,-3); (iv)

relative

5.81.

- t, (1 - t)^, In the vector space V of polynomials in t of degree - 3, consider the following basis: {1, 1 basis if: (i) v = 2 - 3t + t* + 2t^; (1 _ t)3}. Find the coordinate vector of v S y relative to the above
(ii)
i;

= 3 - 2t - ^2;

(iii)

= a + bt + ct^ + dt^.
X
2 symmetric matrices over R, consider the following basis:

5 82

In the vector space PF of 2

{(-11).
Find the coordinate vector of the matrix

:)(-' 1)}
relative to the above basis
if:

AGW
I)

...

= =

(4

-I)

<"'

^<l
63

5.83.

Consider the following two bases of R*:


{ei
(i)

(1, 1, 1),

02

(0, 2, 3),

(0, 2,

-1)}

and

{/i

(1, 1, 0), /z

(1,

-1.

0), fs
[v]y. e,

(0, 0, 1)}

Find the coordinate vector ot v

(3,5,-2)

relative to each basis: [v]^

and

(ii)

Find the matrix


basis {/i, /a, /a)-

whose rows are respectively the coordinate vectors of the

relative to the

(iii)

Verify that [v]eP=[v]f.

5.84.

(of dimension n). Let P be the .,e} and {fu ../} are bases of a vector space V Suppose {e^, basis {fih Prove rows are respectively the coordinate vectors of the e's relative to the matrix whose = [v]f. (This result is proved in Problem 5.39 in the case n - 3.) that for any vector veV, [v]^P
. .

5.85.

Show

that the coordinate vector of

SF

relative to

any basis of

is

always the zero w-tuple

(0,0, ...,0).

RANK OF A MATRIX
5.86.

Find the rank of each matrix:

5.87.

Let

and

be arbitrary

mXn matrices.
A
and
{B)

Show

that

rank (A

B)

rank (A)

rank

(B).

5.88.

Give examples of 2
(i)

2 matrices

such that:
(ii)

rank (A

(iii)

rank (A

+ B)< + B) >

rank

(A), (A),

rank

rank (A

B)

rank (A)

rank (B)

rank

rank

(B)

MISCELLANEOUS PROBLEMS
5.89.

be the vector space of 3 X 3 symmetric matrices over K. Oy hibiting a basis of W. (Recall that A = (ay) is symmetric iff

Let

Show

that

dimW =
dim

by ex-

a^;.)

5.90.

Show that be the vector space of 3 X 3 antisymmetric matrices over K. W. (Recall that A = (a) is antisymmetric iff - -a^j.) exhibiting a basis of
Let

II'

by

5.91.

Suppose dim y = 5.6(iii), page 89).

n.

Show that a generating

set with

w elements

is

basis.

(Compare with Theorem

CHAP.

5]

BASIS

AND DIMENSION
Let

119

5.92.

Let

tj, 2,

<B

+ at

where

be symbols, and let be any field. aj G K. Define addition in V by


(ajfj

be the set of expressions

ajti

+ 03*2 +

+ 02*2 +

+ atj + (biti + 62*2 + + = (Oi + 6i)ti + (02 + 62)2 +



6)

K+
+

>)

Define scalar multiplication on


fc(i*i

by

+ 2*2 +
.
.

+ o<) =

ka^ti

ka^t^

feat
{t^,
. . .

that y is a vector space over basis of y where, for t = 1, , n,

Show

with the above operations.

Also show that

t} is

ti

Oti

Oti_i

li

Ofj +

ot

5.93.

Let y be a vector space of dimension n over a field K, and let he a, vector space of dimension over a subfield F. (Hence V may also be viewed as a vector space over the subfield F.) Prove that the dimension of V over F is ww.

5.94.

Let

U
=

and

W be vector spaces over the same


Let
t7

TF (see Problem 4.45).

field K, and let V be the external direct sum of U and and Jy be the subspaces of V defined by C7 = {(m, 0) m G 17} and
:

W
(i)

w G W). Show that U is isomorphic


{(0,

w):

to

under the correspondence

<r> (m, 0),

and that

is iso-

morphic to
(ii)

under the correspondence

*-> (0, w).

Show

that

dim

V =

dim

V+

dim W.

5.95.

Suppose
to

V = U

W. Let V be

the external direct product of

U and

W. Show that

V is isomorphic

under the correspondence v

= u+w

<r^

{u,w).

Answers
5.52.
(i)

to

Supplementary Problems
(vi)

no,

(ii)

yes,

(iii)

yes,

(iv)

no, (v) yes,

no,

(vii) yes,

(viii)

no.

5.53.

(i)

dependent, dependent,

(ii)

independent. independent.

5.54.

(i)

(ii)

5.55.

(i)

independent,

(ii)

dependent.
(ii)

5.57.

(i)

(2,-l

+ t) =

(l

+ i)(l-t,i);

(7, l

+ 2\/2) = (3-v^)(3 + -\/2, 1 + -/2).


u

5.59.

The statement
this case,

is false.

Counterexample:

(1, 0),

(2, 0)

and

w=

(1, 1)

in R*.

Lemma

5.2

requires that one of the nonzero vectors u,v,w is a linear combination of the preceding ones.

In

2m. no,
(iv) yes.

5.64.

(i)

yes,

(ii)

no,

(iii)

5.65.

(i)

no,

(ii)

yes,

(iii)

no,

(iv)

no.

5.66.

(i)

dim

W=
2.

3,

(ii)

dim

W=

2.

5.67.

dim

W=
=

5.68.

dim Ty
(i)

2.

5.69.

basis, {(7, -1, -2)};

dim ly

1.

(ii)

dim ly

0.

(iii)

basis, {(18, -1, -7)};

dim

W=

1.

5.70.

(i) (ii)

basis, {(2,-1,0,0,0), (4,0,1,-1,0), (3,0,1,0,1)};

dim Ty

3.

basis, {(2, -1, 0, 0, 0),

(1, 0, 1, 0, 0)};

dim ly

2.

120

BASIS

AND DIMENSION

[CHAP.

5x
5.71.

{
5.72.
(i)

+ +

y y

z z

s
t

= =
-1,
0, 1)};

basis, {(1, 0, 0, 0), (0, 2, 1, 0), (0,

dim

V=

3.

(ii)

basis, {(1,0,0,1), (0,2,1,0)}; basis, {(0,2,1,0)};

dim Tf

2.

(iii)

dim (VnH^)

1.

Hint.

VnW

must

satisfy all three conditions on a,b,c

and
yes,

d.

5.73.

(i)

(ii)

no.

For dim

V = n + 1,

but the set contains only n elements.

5.75.

dimd/nW') =
dim {U+W) dim

2,

3 or 4.

5.77.

3,

dim

(UnW)^
=

2.

5.78.

(U+W) =
+ \4x +
[30:
{(1,

3,

dim (t/n W)

1.

2y

+
s

5.79.

(i)

=0
=
[v]

(ii)

-2, -5,

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

-8 + 2y + + 2j/ + 2 X^x -1)}. dim {VnW) = 2


\Ax
(1, 2),
(iii)

= =

5.80.

(i)

= =
=

(5/3, -4/3),

(ii)

[v]

M=
0),

(0, 3),

(iv)

[v]

((a

6)/3,

(a-26)/3).

5.81.

(i)

M
[A]

(2,

-5,

7,

-2),

(ii)

(0, 4,

-1,

(iii)

[v]

(a

+ h + c + d, -b-2c-3d, c + 3d,

-d).

5.82.

(i)

(2,

-1,

1),

(ii)

[A]

(3, 1,

-2).

/I
5.83.
(i)

1\

[v],

(3,

-1,

2),

M; =

(4,

-1, -2);

(ii)

P =

-1

5.86.

(i)

3,

(ii) 2,

(iii) 3,

(iv) 2.

5.88.

(i)

B
(J (o

CI

-I)

<""^=a i>

"h:
0'
1 |, 1

(ii)

P'
o)'

c
1

:)
/o
l\
1

fl
5.89.

o\

/o

o\

<;io
\0

0,1
0/

0,0
0/

0,0
0/
\0

0\

/O

\o

\l

0/

,0

0;

l\
5.90.

0,0
-1

/O

0^
1

0/

\o -1

o;

5.93.

Hint. The proof is identical to that given in Problem 5.48, page 113, for a special case (when an extension field of K).

is

chapter 6

Linear
MAPPINGS Let A and 5
ment from

Mappings
aGA
there

be arbitrary

sets.

Suppose to each

is

assigned a unique ele(or-

of B; the collection, /, of such assignments A into B, and is written

is called

a function or mapping

map)

f:A-^B

or

A^B

We write f{a), read "/ of a", for the element of B that / assigns to a e A; it is called the value of fata or the image of a under /. If A' is any subset of A, then /(A') denotes the set of images of elements of A'; and if B' is any subset of B, then f-'{B') denotes the set of elements of A each of whose image lies in B':
f{A')

{/(a)

aGA'}

and

f'W) = {aGA:

f{a)

B'}

In particular, the set of all images, i.e. f{A), is called the image (or: ranflre) of /. Furthermore, A is called the domain of the mapping f:A^B, and B is called its co-domain.

We call f{A') the imxtge of A' and /-^(jB') the inverse image or preimage of B'.
To each mapping
{{a, f{a))
:

f:A^B

aGA}. We

call this set

there corresponds the subset of A x B given by the graph of /. Two mappings f:A-*B and g:A-^B

are defined to be equal, written f = g, if /(a) = 5r(a) for every aGA, that is, if they have the same graph. Thus we do not distinguish between a function and its graph. The negation of f = g is written f - g and is the statement: there exists an for which
f(a)

^ g{a).

aGA

Example

6.1

Let

f from

A = {a, b, e, d} A into B:

and

{x, y, z, w}.

The following diagram

defines a

mapping

Here

f(a)

y,

/(6)

x,

f{c)

z,

and

f(d)

y.

Also,

faa,b,d})

=
is

{f (a), fib), fid)}

{y,x,y}

{x,y}

The image
Example
6.2:

(or:

range) of /

the set [x,y,z}:

f(A)

{x,y,z}.

Let

/:R

be the mapping which assigns to each real number x

its

square

a;^:

V^

x^

or

Here the image of 3

is

9 so

we may

write

= a;2 /( 3) = 9.
f{x)

121

122

LINEAR MAPPINGS

[CHAP.

We use the arrow


/:A->5 bywriting

i^

to denote the

image of an arbitrary element x


x

eA

under a mapping

fix)

as illustrated in the preceding example. Example


6.3:

Consider the 2 X 3 matrix

A = A
that

'1

-3
4

5' If

,2 c
is,

we

write the vectors in R^ and


:

-1,

R2 as column vectors, then


V
\-*

determines the
T{v)

Av,

mappmg T RS -> R2 v& R3 - Av,


/I

defined

by

3\

-3

5\/

^^
^

Thus

if

then

T{v)

= Av =

4 -_i

-2/

\-2/

<

-10
12

Remark:

Every
defined

mxn
by

matrix

over a

field

determines the mapping

T X" - K
:

v v^ Av are written as column vectors. For convenience where the we shall usually denote the above mapping by A, the same symbol used for the vectors in if"

and

li:"'

matrix.
Example
6.4:

real field R. be the vector space of polynomials in the variable t over the polynomial / G V, the derivative defines a mapping D:V-^V where, for any Then we let D(f) = df/dt. For example, D(3t^ - 5t + 2) = 6t - 5.

Let

Example

6.5:

Let

Then

preceding example). be the vector space of polynomials in t over R (as in the V -* R where, for any to 1 defines a mapping the integral from, say,

polynomial

f&V, we

let

^(/)

= =

f(t) dt.

For example,

^(3(2-5* +
Note that this

2)

(3t^-5t

+ 2)dt

i
whereas the

map

is

map
Example
6.6:

in the preceding

from the vector space example is from V into

into the scalar field R,

itself.

Consider two mappings

f:A^B

and

g.B^C

illustrated below:

0.
a

-^-
Hence we can obtain the image of
/(a)

Let a G A; then /(a) G B, the domain of g. under the mapping g, that is, g(f{a))- This

map
is

g{f(a))

and from A into C is called the composition or product of / and g, In other words, (gf):A-^C is the mapping defined by gf.
{9f){a)

denoted by

g(f(o))

Our

first

theorem
Let

tells

us that composition of mappings satisfies the associative law.

Theorem

6.1:

f.A^B, g.B^C and h.C^D. Then


If a

ho{gof)

(hog)of.

We prove this theorem now.


and

G A,

then

{ho{gof)){a)
({hg)of){a)

=
=

h{igof){a))
{hog){f{a))

=
=

h{g{f{a)))

Hg{f{a)))
so ho{gof)

Thus iho{gof)){a) = {ihog)of){a) for every a G A, and

{hog)of.

Remark:

of a G A Let F:A-^B. Some texts write aF instead of F{(i) for the image functions F:A-*B and under F. With this notation, the composition of G-.B^C is denoted by F o G and not by G o F as used in this text.

CHAP.

6]

LINEAR MAPPINGS

123

We

next introduce some special types of mappings.

Definition:

A
if

mapping f:A-*B

is

said to be one-to-one (or one-one or 1-1) or injective

different elements of

A
if

have distinct images; that


a
v^ a'

is,

implies
f{a')

/(a) - f{a')

or, equivalently.

if /(a)

implies

a'

Definition:

mapping f:A-^B

is

said to be onto (or: /

maps

onto B) or surjective

if

every b

GB

is

the image of at least one a

G A.

A mapping which is both one-one and onto is


Example
6.7:

said to be bijective.
be defined by f(x)
follow:

Let
h(x)

/:R^R, g:B,-*R = x^. The graphs

and

fc

R - B

2'',

g{x)

x^

-x

and

of these

mappings

f(x)

2=^

g(x)

X'

h(x)

a;2

The mapping / is one-one; geometrically, this means that each horizontal line does not contain more than one point of /. The mapping g is onto; geometrically, this means that each horizontal line contains at least one point of g. The mapping h is neither one-one nor onto; for example, 2 and 2 have the same image 4, and 16 is not the image of any element of R.
Example
6.8:

Let A be any set. The mapping /:A-A defined by f{a) = to each element in A itself, is called the identity mapping on 1^ or 1 or /. Let

a,

i.e.

and

which assigns is denoted by

Example

6.9:

f:A-*B.

We

call

g-.B^A
fg =
Ifl

the inverse of

/,

written /~i,

if

and

gf =

1a

emphasize that / has an inverse if and only if / is both one-to-one and onto (Problem 6.9). Also, if then /"'(ft) = a where a is the unique element of A for which f(a) = 6.

We

6GB

mapping F:V -* U is called a spaces over the same field K. linear mapping (or linear transformation or vector space komomorphism) if it satisfies the following two conditions:
(1)
(2)

LINEAR MAPPINGS Let V and U be vector

For any
For any

v,wGV,

kGK

+ F{w). and any vGV, F{kv) = kF{v).


F{v
F{v)

+ w) =

In other words, F:V^ U is linear if it "preserves" the two basic operations of a vector space, that of vector addition and that of scalar multiplication.
Substituting k = into (2) we obtain F{0) the zero vector into the zero vector.

0.

That

is,

every linear mapping takes

124

LINEAR MAPPINGS

[CHAP.

Now

for any scalars a,b

gK

and any vectors v,w

eV

we

obtain,

by applying both

conditions of linearity,

F{av + bw)

F{av)

F{bw)

aF{v)

+ bF{w)

More

generally, for

any scalars

aiGK

and any vectors

viGV we

obtain the basic

property of linear mappings:


F{aiVi

+ aiVi +

We
linear

remark that the mappings and is sometimes used as


6.10:

+ anV) = aiF{vi) + a^FiVi) + condition F{av + hw) = aF{v) + bF{w)

anF(Vn)

completely characterizes

its definition.

Example

Let

be any

mapping

T:&^K^
T(v

mX%

matrix over a field K. by the assignment v

As noted
y->

Av.

previously, A determines a (Here the vectors in K" and K""

are written as columns.)

We

claim that

is linear.

For, by properties of matrices,


T{v)

+ w) - A(v + w) = Av + Aw =
T(kv)

T{w)

and

A(kv)

= kAv =

kT{v)

where v,w G K^ and kE. K.

In comment that the above type of linear mapping shall occur again and again. show that every linear mapping from one finite-dimensional fact, in the next chapter we of the above type. vector space into another can be represented as a linear mapping

We

Example 611-

Let

We

z) B be the "projection" mapping into the xy plane: F(x, y, Then Let v = (a, b, c) and w = {a', b', c'). show that F is linear.

i^

R^

-^

{x, y, 0).

F{v
and, for any

+ w) = F{a + a',b + 6', c + c') = {a + a', b + = (o, b, 0) + (a', b', 0) = F(v) + F(w)

6', 0)

fc

R,

F(kv)

F{,ka, kb, ke)

(ka, kb, 0)

k(a, b, 0)

kF(v)

That

is,

is

linear.

Example 612-

by F(x,y) Let F R2 ^ R2 be the "translation" mapping defined zero vector Observe that F(0) = F(0,0) = (1,2) ^ 0. That is, the not linear. onto the zero vector. Hence F is
:

=
is

(x

l,y

+ 2)

not mapped

Example

6.13:

Let

F.V^U
v,weV
+ w) =

any

be the mapping which assigns and any kG K, we have

G
and

[7

to every

veV.
=
fcO

Then, for

F{v

+
call

F(v)

F(w)

F(kv)

=
by
.

kF(v)
0.

Thus
Example
6.14:

is

linear.

We

the zero

mapping and

shall usually denote it

Consider the identity mapping


for any

I.V^V

v,wGV

and any
/(av

bw)

a,beK, = av + bw =

which maps each we have


al{v)

vGV

into itself

Then,

bl(w)

Thus
Example
6.15:

/ is linear.

defined in

t over the real field be the vector space of polynomials in the variable and the mtegral mappmg differential mapping D:V-^V Then the any and 6.5 are linear. For it is proven in calculus that for

Let

R JtV^B

u,v

SV

Examples 6.4 and fe e R,


d{u

dv du + v) dt- = dt+dl
,

djku)
^""^

j^du

dt
also,

"dt

that

is,

D(u + v)=D(u) + D(v)

and D(ku)

=
\

k D(u); and
u(t) dt

f
and
that
is,

{u(t)

v(t)) dt

= =

v(t) dt

f
J{u +
)

ku{t)dt

u{t) dt

SM + SM

and ^(few)

k^iu).

CHAP.

6]

LINEAR MAPPINGS
6.16:

125

Example

Let
ping

F:V ^ U
:

be a linear mapping which


C/ ->

is

mapping F"!
is

exists.

We

will

show (Problem

both one-one and onto. Then an inverse 6.17) that this inverse map-

also linear.

When we investigated the coordinates of a vector relative to a basis, we also introduced the notion of two spaces being isomorphic. We now give a formal definition.
Definition:

linear

mapping

F:V^U
U

is

called

an isomorphism
if

if it is
is

one-to-one.

The

vector spaces V, V onto U.


Example
6.17:

are said to be isomorphic

there

an isomorphism of

Let
its

be a vector space over

of dimension

n and
is

let {e^,

.,e} be a basis of V.

Then as noted previously the mapping v

[v]^,

coordinate vector relative to the basis {e},

which maps each v into an isomorphism of V onto K".


i.e.

eV

Our next theorem


it tells

gives us an abundance of examples of linear mappings; in particular, us that a linear mapping is completely determined by its values on the elements

of a basis.

Theorem

6.2:

V and U be vector spaces over a field K. Let {vi,'i;2, .,Vn} be a basis V and let Ui, Ui, .,Un be any vectors in V. Then there exists a unique linear mapping F:V-^U such that F{vi) = Ui, F{v2) = ..., F{vn) = th.
Let
of
. .

112,

We

bitrary; they

emphasize that the vectors Mi, zt in the preceding theorem are completely armay be linearly dependent or they may even be equal to each other.
.

KERNEL AND IMAGE OF A LINEAR MAPPING


We
begin by defining two concepts.
Definition:

Let
of

F:V-^U

be a linear mapping.
in U:

The image of F, written Im F,


F{v)

is

the set

image points

Im F =

{uGU

=u

for some v

G V}

The kernel of F, written KerF,

is

the set of elements in


F{v)

which map into

OGU:
Ker The following theorem
is easily

F = {vGV:

0}

proven (Problem

6.22).

Theorem

6.3:

Let
of

F:V->U

and the kernel of

be a linear mapping. Then the image of /?" is a subspace of V.

/^ is

a subspace

Example

6.18:

Let F:R3-^H^ be the projection mapping into the xy plane: F(x, y, z)


(x, y, 0).

Clearly the image of

is

the
l l

entire

xy plane:

||

||llill

(a, 6, c)

Im

F =

{(a, 6, 0)

o, b

G R}
the z axis:

Note that the kernel of

is

KerF =
map

{(0, 0, c):

G R}

since these points and only these points into the zero vector = (0, 0, 0).

128

LINEAR MAPPINGS
The dimension
of the solution space

[CHAP.

Theorem

5.11:

W of the homogeneous system of linear


is

equations rank of the coefficient matrix A.


is

AX

n-r where

the

number

of

unknowns and r

is

the

OPERATIONS WITH LINEAR MAPPINGS We are able to combine linear mappings in various ways to

obtain new linear mappings. These operations are very important and shall be used throughout the text. field K. Suppose F:V-*U and G:V-^U are linear mappings of vector spaces over a assigns F(v) + G{v) to We define the sum F + G to he the mapping from V into U which

^^^'

(F + G){v)

F{v)

Giv)

mapping from Furthermore, for any scalar kGK, we define the product kF to be the into U which assigns k F{v) to i; e F: ikF)iv) = kF{v)

We

show that if F and G are linear, then i^^ + G and kF are vectors v,w GV and any scalars a,h GK,
{F

also linear.

We

have, for any

+ G){av + bw)

= = = =
kF{av

F{av

+ bw) + Giav + bw) aF{v) + bF{w) + aG(v) + bG{w) a{Fiv) + G{v)) + b(F{w) + G{w)) a(F + G){v) + b{F + G){w)
k{aF{v)

and

{kF)(av

+ bw)

= =

+ bw) = akF{v) + bkF(w)

+ bF{w)) a(kF)(v) +

b(kF){w)

Thus

F+G

and kF are

linear.

The following theorem

applies.

Theorem

6.6:

Let

of all and U be vector spaces over a field K. Then the collection operations of addition and linear mappings from V into U with the above scalar multiplication form a vector space over K.

The space

in the

above theorem

is

usually denoted by

Hom(7,
Here

[/)

comes from the word homomorphism. dimension, we have the following theorem.

Hom

In the case that

and

are of finite

Theorem
and

6.7:

Suppose dim 7

and dim

Now suppose that

G:U^W

are vector V, U and are linear mappings:

= mn. spaces over the same field K, and that F:V-*U

U ^ n. Then dim Hom(V,

U)

mapping Recall that the composition function Goi?' is the is linear whenever = G{Fiv)). show that {GoF){v) and any scalars a,b GK, for any vectors v,w

from
and

into

We

GoF

are linear.

defined by have,

We

gV

{GoF)iav

+ bw)

= =

G{Fiav

+ bw)) = G{aF{v) + bFiw)) = aiGoF){v) + b{GoF)(w) aG{F{v)) + bGiF{w))

That

is,

G o F is linear.

CHAP.

6]

LINEAR MAPPINGS
of linear

129

The composition
related as follows:

mappings and that of addition and scalar multiplication are

Theorem

6.8:

Let V,

into

U and W be vector spaces over K. Let F, F' be linear mappings from U and G, G' linear mappings from U into W, and let k&K. Then:
(i)

Go(F +
(G

(ii)

= GoF + GoF' + G')oF = GoF + G'oF


F')

(iii)

k{GoF)

= {kG)oF =

Go(kF).

ALGEBRA OF LINEAR OPERATORS


Let F be a vector space over a field K. We novir consider the special case of linear mappings T:V^V, i.e. from V into itself. They are also called linear operators or linear transformations on V. We will write AiV), instead of Horn (V, V), for the space of all such

mappings.

By Theorem 6.6, A{V) is a vector space over K; it is of dimension n^ if V is of dimension Now if T,SgA{V), then the composition SoT exists and is also a linear mapping from V into itself, i.e. SoTgA(V). Thus we have a "multiplication" defined in A{V). (We shall write ST for SoT in the space A{V).)
n.

We

remark that an algebra


is

tion of multiplication
(i)

over a field is a vector space over in which an operadefined satisfying, for every F.G,H and every kGK,

GA

F{G + H) =
(G

(ii)

FG + FH + H)F = GF + HF
=
{kG)F

(iii)

k{GF)

G(kF).
i.e. if

If the associative
(iv)

law also holds for the multiplication,

for every

F,G,H gA,

{FG)H = F{GH)

then the algebra is said to be associative. Thus by Theorems 6.8 and 6.1, A{V) is an associative algebra over with respect to composition of mappings; hence it is frequently called the algebra of linear operators on V.

Observe that the identity mapping /

7 -> F

belongs to A{V).

we have TI - IT - T. We note that we can also form "powers" of T^^ToT,T^ = ToToT, .... Furthermore, for any polynomial
V(x)

Also, for

T;

we

any T G A{V), use the notation

tto

+ =

aix

+ +

a2X^

ax",

aiGK
ar"
if

we can form

the operator p{T) defined by p{T)


aol

aiT

+ a^T^ +

V{T)

(For a scalar kGK, the operator kl is frequently denoted by simply k.) In particular, = 0, the zero mapping, then T is said to be a zero of the polynomial p{x).
Example
6.21:

Let

T R3 ^ R3 be
:

defined

by

T(x,y,z)
(0, a, b)

=
+

(0,x,y).
(a, b, c)

Now

if

{a,b,c)

is

any element

of R3, then:

{T +

I)(a, b, c)

=
=

= (a,a+b,b + c)
=
(0, 0, 0)

and

T^(a, b, c)

T^0,

a, b)

T{0, 0, a)

Thus we

is

see that rs = 0, the zero mapping from a zero of the polynomial p{x) = v?.

into itself.

In other words,

130

LINEAR MAPPINGS

[CHAP.

INVERTIBLE OPERATORS

linear operator

T -.V^ V

is

said to be invertible
I.

if it

has an inverse,

i.e.

if

there

exists r-i

e A{V)
is

such that

TT-^ = T-^T =

Now T
suppose
over,

invertible if

and only
can

invertible then only

gV
finite

is

nonsingular,

i.e.

Thus in particular, if T is if it is one-one and onto. map into itself, i.e. T is nonsingular. On the other hand, Ker T = {0}. Recall (page 127) that T is also one-one. More-

assuming

has

dimension,

we

have, by
T)

Theorem

6.4,

dimF
Then

= =
i.e.

dim(Imr) + dim (Ker


dim (Im T) +

dim(Imr) + dim({0})

=
V; thus

dim (Im T)

ImT -V,
6.9:

the image of

is

is

onto.

Hence T

is

both one-one and onto

and so

is invertible.

We

have just proven

Theorem

A
6.22:

linear operator

T:V-*V
if it is

vertible if

and only

on a vector space of finite dimension nonsingular.

is

in-

Example

Let T be the operator on R2 defined by T(x, y) = (y, 2x-y). The kernel of T is hence T is nonsingular and, by the preceding theorem, invertible. We now Suppose (s, t) is the image of {x, y) under T\ hence (a;, y) find a formula for T-i. T(x,y) = (s,t) and T-'^(s, t) = (x, y). We have is the image of (s, ) under r-i;
{(0, 0)};

T(x, y)

(y,

2x

y) =
s
t)

(s, t)

and

so
a;

s,

2x-y =
+
li,
2/

is

Solving for x and y in terms of given by the formula T~^(s,

and

t,

we

obtain
s).

is

s.

Thus T^'

(|s

+ it,

The finiteness of the dimensionality of the next example. in


Example
6.23:

in the preceding

theorem

is

necessary as seen

Let

be the vector space of polynomials over K, and


T(ao

let

be the operator on
a< +

defined by

+ ait-\

+ ajn) =
t

a^t

Uit^

i.e.

T
is

increases the exponent of

in each
is

term by

1. is

and

nonsingular.

However, T

not onto and so

Now T is a linear mapping not invertible.

We now give an important application of the above theorem to systems of linear equations over K. Consider a system with the same number of equations as unknowns, say n. We can represent this system by the matrix equation
Ax =
where
b
(*)

we view as a linear operator on K". Suppose has only the zero solution. matrix equation Ax = the matrix A is nonsingular, i.e. the one-to-one and onto. This means that the Then, by Theorem 6.9, the linear mapping A is system (*) has a unique solution for any b G K". On the other hand, suppose the matrix A is singular, i.e. the matrix equation Ax = has a nonzero solution. Then the linear mapping A is not onto. This means that there exist b G K" for which (*) does not have a Furthermore, if a solution exists it is not unique. Thus we have proven the solution.

is

an n-square matrix over

K which

following fundamental result:

Theorem

6.10:

Consider the following system of linear equations:

anXi
a2lXl

+ ai2X2 4+ a22X2 + +
an2X2

+ amajn = + a2nXn =
+
annXn

bi b2

alXl

&

CHAP.

6]

LINEAR MAPPINGS
(i)

131

If the corresponding homogeneous system has only the zero solution, then the above system has a unique solution for any values of the bu If the corresponding
(i)

(ii)

there are values for the


(ii)

a solution; not unique.

homogeneous system has a nonzero solution, then: 6i for which the above system does not have whenever a solution of the above system exists, it is

Solved Problems

MAPPINGS
6.1.

State whether or not each diagram defines a {x,y,z}.

B=

mapping from

A=

{a, b, c)

into

(i)

(ii)

v(iii)

(i)

No. There No.


Yes.

is

nothing assigned to the element 6


z,

G A.
A.

(ii)

Two

elements, x and

are assigned to c

(iii)

6.2.

Use a formula
(i)

to define each of the following functions

from

into R.

(ii)

(iii)

To each number let / assign its cube. To each number let g assign the number To each positive number let h assign its let h assign the number 6.

5.

square,

and to each nonpositive number

Also, find the value of each function at 4,


(i)

2 and
we can

0.

Since / assigns to any


/(4)

number z
43

its

cube

a^,

define /

by

/()
/(O)

64,

/(-2)

(-2)3

-8,

= =

a:^.

Also:

0^

(ii)

Since g assigns 5 to any

number
flr(4)

x,

we can
5,

define

g by g{x)
5,
ff(0)

5.

Thus the value of g at each


5

number

4,

2 and

is 5:

fl-(-2)

(iii)

Two

different rules are used to define

h as

follows:
'a;2

h{x)

if if

X
a;

>
=s

Since 4

>

0,

h(4)

42

16.

On

the other hand,

-2,

and so ^-2)

6, fe(0)

6.

132

LINEAR MAPPINGS
Let A = {1,2,3,4,5} and let f:A-^A be the mapping defined by the diagram on the right, (i) Find the image of /. (ii) Find the graph of /.
(i)

[CHAP.

6^.

The image /(A) of the mapping / consists of all the points assigned to elements of A. Now only 2, 3 and 5 appear as the image of any elements of A; hence /(A) = {2,3,5}.

(ii)

The graph of / consists of the ordered pairs where a&A. Now /(I) = 3, /(2) = 5, /(3) = 5,
/(5)

(a, /(a)),

/(4)

2,

3;

hence the graph of


/

{(1,3),(2,5), (3,5), (4,2),(5,3)}

6.4.

Sketch the graph of

(i)

f{x)

^x^ + x-6,

(ii)

g{x)

= x^-^x^-x + Z.

Note that these are "polynomial functions". In each case set up a table of values for x and then find the corresponding values of f{x). Plot the points in a coordinate diagram and then draw a smooth continuous curve through the points.
(i)

m
6

(ii)

9{x)

-4
-3

-2 -1

-15

-2
-1

-4 -6 -6
-4
1

2
3 4

-3

2 3 6

15

6.5.

Let the mappings

f.A^B

and
f

g.B^C

be defined by the diagram

(i)

Find the composition mapping {gof):A-*C. ping: f,g and go f.

(ii)

Find the image of each map-

(i)

We

use the definition of the composition mapping to compute:

= g(f(a)) = = gim) = igf)ib) (9 me) = g(f(c)) =


(gofXa)
Observe that

g(y) g(x)

g{y)

= = =

s
t

we

arrive at the same answer if

we
X

"follow the arrows" in the diagram:


e

-*

-*

t,

b -*

-* s,

^ y -*

CHAP.

6]

LINEAR MAPPINGS
By
the diagram, the image values under the
r, s

I33

(ii)

mapping
and

/ are x

and

y,

and the image values under


{r, s, t}
t

g are

and

t\

hence

image of /

{x, y}

image oi g

By (i), 0f =

the image values under the composition mapping gof are {s, *} Note that the images of g and gf are different.

and

s;

hence

image of

6.6.

Let the mappings / and g be defined by f{x) = 2x + 1 and g{x) = x^-2. (i) Find (sro/)(4) and (/osr)(4). (ii) Find formulas defining the composition mappings gof and fog.
(i)

/(4)

g(A)
(ii)

= 2^4 + 1 = 9. = 42 - 2 =

Hence
14.

(ff

/)(4)

p(/(4))

^(9)

92

79.

Hence
g o f as

(/ofl')(4)

/(fir(4))

/<14)

14

29.

Compute the formula

for

follows:

(fff){x)

g(f{x))

fif(2a;

+ l) =

(2a;

+ 1)2 -

4a;2

4a;

Observe that the same answer can be found by writing y = f(x) = 2a! and then eliminating y: z - y^-2 = {2x- 1)^ -2 - 4x^ + 4a! - 1. j/2 - 2,
(fg)()

+1

and

g(y)

f{g(x))

/(a;2-2)

2(a;2-2)

2a!2

3.

Observe that

fog^gf.

6.7.

Let the mappings

f:A-^B,g:B-^C and h:C-*D


f

be defined by the diagram


h

D
a

Determine
(i)

if

each mapping
:

(i)

is

one-one,

(ii)

is

onto,

(iii)

has an inverse.

The mapping /

g.B->C
is
(ii)

is

one-one since each element of A has a different image. The mapping not one-one since x and z both map into the same element 4. The mapping h:C -* D
->
is

one-one.

The mapping f -.A^B is not onto since e B is not the image of any element of A. The mapping g:B-*C is onto since each element of C is the image of some element of B. The mapping h-.C^D is also onto.

(iii)

mapping has an inverse

if

and only

if it is

both one-one and onto.

Hence only h has an

inverse.

6.8.

Suppose f:A-*B and g.B^C; hence the composition mapping {gof):A^C exists. Prove the following, (i) If / and g are one-one, then gof is one-one. (ii) If / and g
are onto, then

gof

is,

onto,

(iii)

If

gof

is,

one-one, then /

is

one-one.

(iv) If

gof

is

onto, then g is onto.


(i)

Suppose (gf)(x)
is

(gf){y).

Then

g(f{x))

g{f(y)).

Since g
(g f){y)

is

one-one,

y.

We

have proven that

{g f){x)

one-one, f{x) = f(y). Since / implies x = y\ hence flro/ is

one-one.
(ii)

Suppose
exists

c&C. Since g ae.A for which

is onto,

there exists
b.

h
''

e.

f(a)

Thus

(g

f){a)

B for which g(h) = c. Since = g(f(a)) = g(b) = c; hence

/
o

is

onto, there
is onto.

flr

(iii)

Suppose /

Thus
is (iv)

is not one-one. Then there exists distinct elements x,y G A for which /(a;) {gf)(x) = g{f{x)) = g(f(y)) = (g f)(y); hence gf is not one-one. Accordingly one-one, then / must be one-one.

=
it

f{y).

gf

then (g f){a) = g(f{a)) G g{B); hence (g f){A) C g(B). Suppose g is not onto. properly contained in C and so (g f)(A) is properly contained in C; thus gf is not onto. Accordingly Hgofia onto, then g must be onto.
If

aGA,

Then g(B)

is

134

LINEAR MAPPINGS
Prove that a mapping
/0/-1

[CHAP. 6

6.9.

f:A-*B
i.e.

has an inverse

if

and only
:

if it is

one-to-one and onto.

Suppose / has an inverse,

there exists a function /-i

-^

for which

f~^f = 1a and
is

ig.

Since 1a
6.8(iv).
is
if

is

onto by Problem

one-to-one, / is one-to-one That is, / is both one-to-one

by Problem and onto.


6

6.8(iii);

and since 1b

onto, / is

Now
in

suppose /
6.

both one-to-one and onto.


f{a)
A.

Then each

SB

is

the image of a unique element


let

A, say
to

Thus

b,

then

=
6

b;

hence /(6)

b.

Now

g denote the mapping from

defined by
(i)
(?

6 1-^6.

We

have:

/)(<)

fir(/(o))

=
-

S(b)

= =

a, for

every a

G A;

hence s

1a-

(ii)

{fff)(b)

f{g{b))

fib)

6,

for every 6

B; hence
g.

fg -Ib-

Accordingly, / has an inverse.

Its inverse is the

mapping

6.10.

Let / R ^ R be defined by has an inverse mapping /"^


:

fix)

= 2x-S. Now

/ is one-to-one
f~^.

and onto; hence

Find a formula for

the Let y be the image of x under the mapping f: y - f(x) =2x-S. Consequently x will be for x in terms of y in the above equation: image of y under the inverse mapping f-K Thus solve X = {y + 3)/2. Then the formula defining the inverse function is f-Hy) = {y + 3)/2.

LINEAR MAPPINGS
6.11.

Show
(i)

that the following mappings


:

are linear:

R2

^ R2
=

defined

by F{x, y)=^{x

(ii)

F:R^-*R
Let v

defined

by F{x,y,z)

+ y, x). = 2x-Sy + 4z.


b')

(i)

(a,b)

and
V

w = (a',b'); hence + w - (a + a',b +

and

kv

{ka, kb),

k&R
a')

We

have F(v)

and
Since
(ii)

= (a' + b', a'). Thus + h+b' a+ F(v + w) = F(a + a',h + b') - (a.-\= (a + 6, a) + (a' + b', a') = F{v) + F(w) F(kv) - F(ka, kb) := (ka + kb, ka) = k(a + b,a) = =
(a
-t-

6, a)

and F(w)

a.'

kF{v)

v,

and

A;

were arbitrary,

is linear.

Let v-(a,b,c) and

w = (a',b',c'); hence kv - (ka, kb, and = (a + a',b + b',c + e') V + w We have F(v) = 2a - 36 4c and F(w) = 2a' - 36' 4c'. Thus F(v + w) = F(a + a',b + b',c + c') = 2(a a') - 3(6 + = (2a - 36 + 4c) + (2a' - 36' 4- 4c') = F(v) +
-I-

kc),

fc

eE

-t-

-1-

6') -h

4(c

-I-

c')

F(w)
kF(v)

and

F(kv)

F(ka, kb, kc)

2ka

3kb

4kc

k(2a

- 36 + 4c) =

Accordingly,

is linear.

6.12.

Show
(i)

that the following mappings


:

are not linear:

(ii)
(iii)

^ R defined by F(x, y) = xy. FrB?^B? defined by F{x, y) = {x + 1, 2y, F:W-^B? defined by F{x, y, z) = (\x\, 0).

R2

+ y).

(i)

Let w

(l,2)

and

We

have F(v)

w = (3,4); = 1*2 = 2

then v

and F(w)

+w = =3

(A,6).

12.

Hence

CHAP.

6]

LINEAR MAPPINGS
F(v
Accordingly,

135

+ w) =

F(4, 6)

24

F{v)

F{w)

F
=

is

not linear.

(ii)

Since F{0,

0)

(1, 0, 0)

(0, 0, 0),

cannot be linear.

(iii)

Let v

(1, 2, 3)

and k

-3; hence

kv

We

have F{v)

{1,0)

and so kF (v)

= (-3, -6, -9). = -S(1,0) = {-S,0). =


(3, 0)

Then
fcF('y)

Fikv)

F{-3, -6, -9)

and hence

F is

not linear.

6.13.

Let

V be

in F.

Let

the vector space of n-square matrices over K. Let be an arbitrary matrix r F -* 7 be defined by T{A) = + MA, where Show that
:

AM

AeV.

is linear.

For any

A,BGV

and any k G K, we have

T{A +B)

= (A + B)M + M{A + B) = AM + BM + MA + MB = {AM + MA) + {BM + MB) = T{A) + T{B)


kT{A)

and
Accordingly,

T{kA)

= {kA)M + M{kA) = k{AM) + k{MA) = k{AM + MA) ^

is linear.

6.14.

Prove Theorem 6.2: Let V and U be vector spaces over a field K. Let {^'i, ...,?;} be a basis of V and let Mi, be any arbitrary vectors in U. Then there exists a unique linear mapping F:V-^U such that F{Vi) = Ui, F{v2) = %2, ., F{Vn) = Un.
. . .

There are three steps

to the proof of the theorem:


(2)

(1) (3)

Define a

mapping

F{v^

F .V ^ U
a^,
-\
.

such that

Mi, i

= =

l,

...,n.

Show

that
.

is

linear.

Show

that

is

unique.
.

Step
for which
(Since the

{1).

Let V
a^v^

v
osj

eV. + a^v^

Since {v^,
-]

.,i;} is

a basis of V, there exist unique scalars

.,a

are unique, the

+ ai;. We define F:V ^ U by F{v) = a^Ui + a^u^ mapping F is well-defined.) Now, for i= \, ..., n, + Ivi + + Ov Vi - Ovi +

GX

h aM.

Hence

F{Vi)
first step
{2).

Omj

Imj

Om

m;

Thus the
Step

of the proof is complete.

Suppose

v
V

a^Vi

+ a^v^ +
(tti

+ a'U +
{a2

and
b2)v2

w=
+

b^Vi

h^Vi

h^v^.

Then

+w = =
ka^v^
a^u^

hi)Vi

(a

6)v

and, for any

kG K,
=

kv

+ kazVz +

+ ka^v^. By
and

definition of the

mapping F,

F{v)

OiMi

aM
{a^

F{w)

h^u^

h^Vi

6m

Hence

F{v

+ w) =
= =

{a^

+ h-i)ui +

62)^2

++( + 6)m
(6iMi

(!,

F{v)

+ ajMj + + F(w)
fc(oiMi

+ aM) +

+ 62M2 +

6M)

and

(fci;)

+ O2M2 +

+ omJ =

^^(1;)

Thus

f is
a^nVn,

linear.
(3).

Step
1-

Now
G(t))

suppose

G:V ^V

is

linear and

G{v^

M;, t

1,

.,

m.

If

Oi^i

+ a^v^ +

then

= =

G(aiVi
OiMj

+ a^v^ + + a2M2 +
i?

+ av) + * =
F.

a^G{v-^
ii'(t))

a2G{v^

aG(v)

Since

G(t))

F{v) for every

V,

G=

Thus

F is

unique and the theorem

is

proved.

136

LINEAR MAPPINGS
Let r R2
:

[CHAP. 6

6.15.

be the linear mapping for which


r(l, 1)

and

r(0, 1)

= -2
is

{1)

(Since {(1,1), (0,1)}

is

a basis of R^, such a linear mapping exists and


and

unique by

Theorem
First

6.2.)

Find
(a, 6)

T{a, b).
as a linear combination of
{a, b)
(1, 1)

we

write

(0, 1)

using unknown scalars x and

y:
(2)

x{l, 1)

j/(0, 1)

Then

(a, 6)

(x, x)

(0, y)

=
we
a

(x,x

+ y)
y

and so

x-a,
a

+ y=^b

Solving for x and y in terms of a and

b,

obtain

=
+
3x

and

(3)

Now

using

(1)

and

(2)

we have

T(a, b)

T{x(l, 1)
T(a, b)

3/(0,

D)

xT{l,

1)

yT(0,

1)

3x

2y

Finally, using (3)

we have

2y

3(a)

2(6

- a) =

5a

26.

6.16.

Let T -.V^U be linear, and suppose Vi, ...,Vn&V have the property that their .,Vn images T{vi), .... T{vn) are linearly independent. Show that the vectors vi, are also linearly independent.
. .

Suppose that, for scalars

ai,

. ,

o,

a^v^

r(0)

Tia^vi

+ a.2V2 +
all

+ ag^a + h ai') =
the
O;

+ aVn = 0. Then aiT(vi) + azTiv^) +

ar(v)
linearly

Since the T{Vi) are linearly independent, independent.

0.

Thus the vectors Vi,...,v are

6.17.

Suppose the linear mapping

F:V^U
is

is

one-to-one and onto.

Show

that the inverse

mapping F~^:U -*V

is also linear.

Suppose M.w'G U. which F{v) = u and F(v')

Since

F
u'.

one-to-one and onto, there exist unique vectors Since F is linear, we also have
F(v')

v,v'BV

for

F{v

+ v') -

F{v)

= u+
F-i(tt)

u'

and
v.

F{kv)

=
f',

kF(v)

ku

By

definition of the inverse


fc'U.

mapping,
"'

F-i(m')

F-i-{u-\-u')

+ v'

and

F-^ku) =

Then

F-Mm + m') =
and thus F"'
is linear.

F-i(m)

F-Mm')

and

F-HM

fei)

feF-Htt)

IMAGE AND KERNEL OF LINEAR MAPPINGS Let F R* - R5 be the linear mapping defined by 6.18. F{x,y,s,t) = {x-y + s + t,x + 2s-t,x + v + ?>s-Zt) image U of F, (ii) kernel W Find a basis and the dimension of the
:

(i)

of F.

(i)

The images of the following generators of R* generate the image


F(l, 0,0,0) F(0,
1, 0, 0)

V
(1,

of F:

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

F(0, 0,1,0)

F(0, 0,

0, 1)

= =

(1,2,3)

-1, -3)

Form

the matrix whose rows are the generators of

and row reduce to echelon form:

to

to

Thus

{(1, 1, 1), (0, 1, 2)} is

a basis of V; hence dim

C/

2.

CHAP.

6]

LINEAR MAPPINGS

137

(ii)

We

seek the set of

(x, y, s, t)

such that F{x,


t,x

y, s,

t)

(0, 0, 0), i.e.,


(0, 0, 0)

F(x, y,s,t)

= (x-y + s +

+ 2s-t,x + y + Bs-St) =
to

Set corresponding components equal to each other whose solution space is the kernel of F:

form the following homogeneous system

X X x

y+s+t
+
y
2s

=
or

y+s+t
y

=
or y

= + 3s-3t =
t

The free variables are


(a) (6)

and

t;

hence

+ s - 2t = 2y + 2s - 4t = dim W = 2. Set
(2, 1,

2t

s s

= =

1,
0, t

=
l

to obtain the solution

1,

0),

to obtain the solution (1, 2, 0, 1).

Thus
which

{(2, 1, is

-1, 0), (1, 2, 0, 1)} is a basis of W. the dimension of the domain R* of F.)

(Observe that

dim

C/

dim IT

4,

6.19.

Let

T:W^W

be the linear mapping defined by


T{x, y, z)

{x

+ 2y z,
(i)

+ z,

+ y 2z)
(ii)

Find a basis and the dimension of the


(i)

image

U
U

of T,
of T:

kernel

W of T.
i,

The images of generators of R^ generate the image


r(i,o,o)

(1,0,1),

r(o,i,o)

(2,i,i),

r(o, o, i)

(-1,

-2)

Form

the matrix whose rows are the generators of


1

U
1

and row reduce


1

to echelon form:
1

1
1 1

("
2
1

to

1 1

-1 1-1

'\
to

-1

-2

-1

Thus
(ii)

{(1, 0, 1), (0, 1,

-1)}

is

a basis of U, and so dim


T(x,y,z)

U=

2.

We

seek the set of {x,y,z) such that


T(x, y,z)

=
y

(0,0,0),

i.e.,

{x

+ 2y - z,

+ z,

-\-

y -2z)

(0, 0, 0)

Set corresponding components equal to each other to form the homogeneous system whose solution space is the kernel of T:

2y y

=
a a

x
or

2y

z z

= = =
z

a;

+ z = 2z =

+ y
y

X
or

2y
y

= =
a;

The only free variable


{(3,

= 1. is z; hence dim a basis of W. (Observe that dim sion of the domain R3 of T.)
1,
1)} is

Let

1;

U+

dim

then y = 2 + 1 =

1 and
3,

3.

Thus

which

is

the dimen-

6.20.

Find a linear map


Method
1.

R^

^ R* whose image is generated by (1, 2, 0, 4) and

(2, 0,

1, 3).

Consider the usual basis of R^: e^ = (1, 0, 0), eg = (0, 1. 0), eg = (0, 0, 1). Set F(ei) = (1, 2, 0, -4), F(e2) = (2, 0, 1, 3) and F{eg) = (0, 0, 0, 0). By Theorem 6.2, such a linear map F exists and is unique. Furthermore, the image of F is generated by the F(ej); hence F has the required property. We find a general formula for F(x, y, z):
F(x, y,
z)

= =
=z

F{xei

+ ye^ + zeg) =
-4)

xFie-^)

yF{e2)

2^(63)

x(\, 2, 0,
(x

2/(2, 0,

-1, -3)

2(0, 0, 0, 0)

2y, 2x,

y, 4x

3y)

'

138

LINEAR MAPPINGS
Method
2.

[CHAP.

Form

a 4 X 3 matrix

whose columns consist only of the given vectors; say,


1

2\

-1 -1 -4 -3 -3
is

Recall that

determines a linear

map A R3 ^ B^ whose image


:

generated by the columns of A.

Thus

satisfies the required condition.

6.21.

Let

be the vector space of 2 by 2 matrices over

and

let

Let

F:V^Y
We

be the linear dimension of the kernel


seek the set of

W of F.
^\
such that

map

defined

by F{A^ =

AM MA.
(p

Find a basis and the

(^

Fr'

q)

K:

:)

C
/x \s

X
2x
2s

I)
3y\
St

{I
/X

DC
+
3s 2s

+ +
2x

+
3(

2t

\
2t\

/-2s \-2s
Thus
2x

+ 2y2s

y x

_ ~
y

/ \0

2y

-2t = 2s =

or
s

= =

The free variables are y and


(a)
(6)

t;

hence dim

2.

To obtain a basis of

W set
t

1,
0,
t

=
1

to obtain the solution x


to obtain the solution

=
1,

1,

1, s

0,

0;

0, s

0, t

1.

^^"^{(o

o)'

G
6.3:

;)}

a basis of T^.

6.22.

Prove Theorem is a subspace of


(i)

Let
(ii)

F.V^U

and

the kernel of

be a linear mapping. Then F is a subspace of V.

(i)

the image of

and a,b& K. Since u and u' belong to G Im F. Now suppose u, u' Since F(Q) = 0, such that F(v) = u and F(v') = u'. Then the image of F, there exist vectors v,v'

GlmF

GV
+

F{av

+ bv') -

aF(v)

hF(v')

au

bu'

e Im F

Thus the image of


(ii)

is

a subspace of U.

G Ker F. Since F(0) = 0, to the kernel of F, F{v) =


F(av

Now

suppose

v,wG Ker F
0.

and a,b e K. Since v and


av

belong

and F(w)

=
==:

Thus

bw)

aF(v)

bF{w)

aO

60

and so

+ bw S KerF

Thus the kernel of

F is

a subspace of V.

6.23.

Prove Theorem 6.4: Let V be of finite dimension, and let ping with image U' and kernel W. Then dim U' + dim
Suppose dim V = n. Since Thus we need prove that dim U'

F:V-^U be W = dim V.
is finite;

a linear mapdim

ia

a.

subspace of V,

its

dimension

say,

W = r n.

= n r.

CHAP.

6]

LINEAR MAPPINGS
Let {wi,
Wr) be a basis of W.

139

We
{w'l

extend {wj to a basis of V:


Wr,Vi, ...,i;_J

Let

B =
is

{F{Vi),F(v2), ...,F(v^r)}

The theorem

proved

if

we show

that

Proof that

{Wj, Vj} generates

generates U'. Let and since v S V,

B is a basis of the image u S U'. Then there exists


U' of F. v

&V

such that F(v)

u.

Since

where the a

ftj

are scalars.

+ b^-r'^n-r since the Wj belong to the kernel = F(aiici + + a^Wf + biv^ + + b^^v-r) u = = aiF{wi) + + 6_^F(i;_,) + a^(Wr) + b^Fivi) + = OjO + + bn-rF(Vn-r) + a^O + biF(Vi) + = b,F(v,) 6_,FK_,)
V

OjWj

a,Wr

^l'"!

Note that F(Wi)

of F.

Thus

jF'(t')

++

Accordingly, the F{Vf) generate the image of F.

Proof that

is

linearly independent.

Suppose

a^Fivi)

a.2F(v2)

+
.

a_ri^K_,.)

=
belongs to the kernel

Then F(aiVi
of F.

+ 02^2 +

Since

{wj generates W,
a^Vi

and so a^Vi + + a_,T;_^ a^^^v^-r) there exist scalars 61, 6^ such that
.
.

a2^'2

an-r'Un-,

b^Wi

62^2

b^Wr
(*)

or

ail^i

an-r'Wn-r

b^Wi

fe^w^

=
of the W; and

Since {tWj, {} is a basis of V, it is linearly independent; hence the coefficients Accordingly, the F(v^ are linearly independent. are all 0. In particular, Oj = 0, ., a_r = 0. .

Vj in (*)

Thus

is

a basis of

V, and

so

dim

V nr

and the theorem

is

proved.

6.24.

= V + (v' v) . V + W and hence f~Hu) Cv + W. Now we prove (ii). Suppose v' G v+W. Then v' = + w where w G W. = kernel of /, f(w) = 0. Accordingly, f{v') = /(-u + w) = f(v) + f(w) = /(t)) + v' e /-i(m) and so v + Wc f-^(u).
v'
1;

f:V-*U is linear with kernel W, and that f{v) = u. Show that the "coset" + W = {v + w: w e W} is the preimage of u, that is, f~^{u) v + W. Suppose v + T^ c/-i(m). We first prove f~Hu)cv + W and We must prove that v'Gf-Hu). Then f(v') = u and so f(v' - v) = f(v') - f{v) = u-u = 0, that is, v'-vGW. Thus
Suppose
(i)

(ii)

(i).

Since
f(v)

W
m.

is

the

Thus

SINGULAR AND NONSINGULAR MAPPINGS 6.25. Suppose F:V ^U is linear and that V is of finite dimension. Show image of F have the same dimension if and only if F is nonsingular.
nonsingular mappings

that

V and the Determine all

R*

^ R^.
+ dim (Ker/i^). Hence V and ImF KerF = {0}, i.e. if and only if F is
have the same dinonsingular.

By Theorem
mension
T.
if

6.4, if

and only

dim F = dim (Im/f) dim (Ker F) = or

Since the dimension of R^ is less than the dimension of R*, so is the dimension of the image of Accordingly, no linear mapping T B* - R^ can be nonsingular.
:

6.26.

Prove that a linear mapping F:V-*U an independent set is independent.


Suppose

is

nonsingular

if

and only

if

the image of

F is nonsingular F

and suppose

{v^,

F(vJ are independent. the vectors F(vi) is linear, F(ajVi + a^v^, + a, e X. Since

We claim that ., v^} is an independent subset of V. Suppose aiF{Vi) + a<^{v^ +;+ aF{v) - 0, where + o^vj = 0; hence
. .

a^Vy

021^2

On^n

^ Ker F

140

LINEAR MAPPINGS

[CHAP.

But F is nonsingular, i.e. Ker F = {0}; hence a^v^ + a^v^ + av = 0. Since the i;; are linearly independent, all the a^ are 0. Accordingly, the F(v>i are linearly independent. In other words, the image of the independent set {v^, i)} is independent.
. . .

the other hand, suppose the image of any independent set is independent. If v G nonzero, then {v} is independent. Then {F{v)} is independent and so F(v) 0. Accordingly, nonsingular.

On

V F

is
is

OPERATIONS WITH LINEAR MAPPINGS


6.27.

Let (x -z,y).
(F

F:W^W
+ G)(x,

and G:W^ be defined by F{x, y, z) = {2x, y + z) and G{x, Find formulas defining the mappings F + G,ZF and 2F - 5G.
y, z)

y, z)

= =

F(x, y,
(2x,

z) + G(x, y, z) + z) + (x z,y) =

(3x

-z,2v + z)
By

(3F)(a;, y, z)

ZF(x,

y, z)

3(2*,
z)

z)

=
z)

{Qx,

+ 3z)
y

(2F

- 5G){x,

y, z)

= =

2F(x, y,
(Ax,

5G{x, y,
(-5a;

2(2a;,

z)

5{x

- z,
-2,y

y)

2y

+ 22) +

+ 5z,

-5y)

(-x

+ 5z,

+ 2z)

6.28.

Let

F:W-^W
.

{y, x)

and G/R'^W be defined by F(x,y,z) Derive formulas defining the mappings GF and
(GoF){x,y,z)

=
=

{2x,y

+ z) and
2x)

G{x,y)

FoG.
(y

G(F{x,y,z))

G{2x, y

z)

+ z,

The mapping

FG

is

not defined since the image of

is

not contained in the domain of F.

6.29.

Show:

(i)

the zero

mapping
(ii)

ment

of

Hom(F,
F G Hom
(F

U);

0, defined by 0{v) = for every v GV, is the zero elethe negative of F G Hom(7, U) is the mapping {-1)F, i.e.

-F =
(i)

(-l)F.
{V, U).

Let

Then, for every


{F

+ Q){v) =
v

Since
(ii)

+ 0)(v) =
[F

F(v)

for every

GV, + 0{v) = eV, F + =


v
F(,v)

F{v)

F(v)

F.

For every v G V,

Since
of F.

{F

+ {-l)F){v) = F{v) + {-l)F{v) = + {-l)F){v) = 0(v) for every vGV, F +

F{v)

F{v}

0{v)
is

(-l)F

0.

Thus (-l)F

the negative

6.30.

Show
By

that for Fi, ...,FG


{aiFi

Hom {V,
aiFj,

U) and

ai,

...,aGK, and for any

vGV,
n =
1.

+ a2F2 H

+ aF)(i;) =

aiFi{v)

+ aJFiiv) +

ajf'niv)

definition of the

mapping

(a^F^iv)

a^F^{v);

hence the theorem holds for

Thus by

induction,

(aiFi

+ (I2F2 +

+ aF)(i;) = =

(a^F^)(v)

aiFiCv)

+ {a^F^ + + a^F^iv) +

+ aF)(i;) + aF(D)

6.31.

Let /^:R3^R2,

G.W^B?
a,b,c

and

HrR^^R^
i?(a;, y, z)

be defined by
{2y, x).

i^'Cx, i/, 2)

G{x, y, z) = {2x + z,x + y) and are linearly independent.


Suppose, for scalars

Show

that

= {x + y + z,x + y), F,G,H G Hom (RS R2)


{1)

G K,

(Here

is

the zero mapping.)

aF + bG + cH = For e^ = (1, 0, 0) G R3, we have


aF(l,
0, 0)

(aF

+ bG + cH)(e{) = =

bG(l,

0, 0)

cH(l,

0, 0)

a(l, 1)

6(2, 1)

c(0, 1)

(a

+ 2b,a + b + c)

CHAP.

6]

LINEAR MAPPINGS
=
Thus by
(a

141

and

0(ei)

(0, 0).

{!),

+ 2b, a+b + e) =
26

(0, 0)

and so
6

a
Similarly for eg

and

(2)

(0, 1, 0)

R3,

we have
1, 0)

(aF

+ bG + cH){e2) = =

aF(0,

6G(0,

1, 0)

+
a
c

cH(0,

1, 0)

a(l, 1)

+
2e

6(0, 1)

c(2, 0)

(a+2c, a+6)
6

0(62)

(0,0)
(5)

Thus
Using
Since
(2)

a and
(5)

=
a

and
0,

we
(4),

obtain the mappings F,

0,

(*)

(1)

implies

and

are linearly independent.

6.32.

Prove Theorem
Suppose
{vi,
.

6.7:
.

Suppose dim y

and dim
. .

U = n. Then dim Hom {V, U) -

mn.

mapping
elements

in
Vj

Hom {V, V)
of V.

.,m} is a basis of V. By Theorem 6.2, a linear .,v} is a basis of V and {mj, is uniquely determined by arbitrarily assigning elements of t/ to the basis
define

We

F^ e Hom {V,U),

1,

m,

1,

...,n

Uj, and Fij(Vk) -0 for fe # i. That is, Fy maps Vi to be the linear mapping for which Fij{v^ theorem into Mj and the other v's into 0. Observe that {Fy} contains exactly mn elements; hence the is proved if we show that it is a basis of Hom {V, U).

Proof that {Fy} generates W2, ..., F(Vm) = Wm- Since w^

Hom (F,
G
U,

U).

Let

F G Hom {V, U).


fc

Suppose
u's; say,

F{vi)

w^, F(v2)

=
(i)

it is

a linear combination of the


Wk =

afclMl

fc2*2
TTi

+
n

fflfcMn>

=
is

1,

. ,

m,

Oy

Consider the linear mapping


proof that {Fy} generates

G =

2 2 ayFy i=l i=l


complete
Since
if

Since

a linear combination of the Fy, the

Hom (V,

t7) is

we show

that
for

F=

G.

We now

compute G(Vk), k
G(i;k)

=
m

l,

...,m.

Fy('Ufc)

k^i
t

and

^^((Vfc)

Mj,

= =

i=l

22 =
3

n
OiiF('yic)

=
3

2 =

OfciJ^)cj(vic)

=
3

2 =

Ofcij

a^iMj

ak2'"-2

fcnMn

Thus by

(1),

G{v^,)

w^.

for each k.

But

^(1;^)

= w^

for each

fe.

Accordingly, by Theorem

6.2,

F=

G; hence {Fy} generates

Hom (V, U).


independent.

Proof that {Fy}

is linearly

Suppose, for scalars ay

K,

i=l

2 2 =
3

For

i;^,

fc

1,

.,w,

0(v^)

=
i

22 =
l j

=l

ii^ij(^ic)

=
3

a^jF^j(v^)
fflfen^n

=
3

2
0,

aicjMi

=
But the
In other words,
all
is

afcl^l

ak2M2

Mj are linearly independent;

hence for k
is

+ =

+
.

1,

.,m,

we have
U)

a^i

0^2

0,

ajj

0.

the ay

and so {Fy}

linearly independent.

Thus {Fy}

a basis of

Hom (V,

17);

hence dim

Hom {V,

mn.

6.33.

Prove Theorem mappings from k&K. Then:


(iii)
(i)

6.8:

Let V,

and
let

V
(i)

into f7

and

be vector spaces over K. Let F, F' be linear G, G' be linear mappings from U into W; and let
Goi?'

Go(F + F') Go(fcF).

Goii'';

(ii)

{G

+ G')oF = GoF + G'oF;

fcCGoF)

= {kG)oF =
GV,

For every v

142

LINEAR MAPPINGS
(Go(F + F'mv) = G{(F + F'){v)) = G{F(v) +
F'(v))

[CHAP.

=
Since
(ii)

{G

(F

F'){v)

+ G{F'(v)) = {G'>F)(v) + {GoF')(v) = {G F + G o F'){v) = (G o F + G F'){v) for every vGV, Go {F + F') = GF + GF'.
G(F{v))

For every v

&V,
{(G

Since
(iii)

({G

+ G')F)(v) = {G + G')(F{v)) = G{F{v)) + G'{F(v)) = (Go F)(v) + {G' F){v) = (G F + G' F)(v) + G') F}(v) = {G F + G F')(v) for every v&V, (G + G') F = GF +
GV,
(k{GF))(v)

G' F.

For every v

k(GF){v)

k{G{F(v)))

{kG)(F{v))

(feGF)(i;)

and

{k{GFmv) = k(GoF){v) =

k(G{F(v)))

G{kF{v))

G{(kF){v))

{GkF){v)

Accordingly, k{GF) = (kG)oF = G(kF). (We emphasize that two mappings are shown to be equal by showing that they assign the same image to each point in the domain.)

6.34.

Let
(i)

F:V^V
Since

rank {GoF)

and ^ rank G,

G.U^W
(ii)

be linear.

rank (GoF)

Hence {GoF):V^W ^ rank F.


and so

is linear.

Show

that

(i)

F{V) c U, we also have


rank (GoF)

G(F{V)) c G(U)

dim G(F{V))

= dim =

((GoF)(y))

= dim =

(G(F(y)))

^ dim

^ dim G(V). Then G(?7) = rank G =


rank

(ii)

By Theorem

6.4,

dim (G(F(y))) ^ dim F(y). Hence


dim ((Go F)(y)) dim (G(F(y)))
=

rank (GoF)

dim F(y)

ALGEBRA OF LINEAR OPERATORS


6.35.

Let S and
(0, x).

T be the linear operators on R^ defined by S{x, y) = {y, x) and T{x, y) Find formulas defining the operators S + T,2S- ZT, ST, TS, S^ and T^.
=

S(x,y) + T(x,y) = {y,x) + (0,x) = {y,2x). = 2S(x,y)-3T{x,y) = 2{y,x) - Z((i,x) = (2y,-x). (ST)(x,y) = S{.T(x,y)) = S(f),x) - (a;,0). (TS)(x,y) = T(S(x,y)) = T(y,x) = {0,y). SHx,y) = S{S{x,y)) = S{y,x) = (x,y). Note S^ = I, the identity mapping. THx, y) = T(T(x, y)) = 7(0, x) - (0, 0). Note T^ = 0, the zero mapping.

{S+T){x,y)

(2S-ZT)(x,y)

6.36.

Let

be the linear operator on R^ defined by


r(3, 1)

(2,

-4)

and

T{1, 1)

=
is

(0, 2)

(i)

(By Theorem
First write

6.2,

such a linear operator exists and


as a linear combination of (3,1) and
(a, h)

unique.)

Find T{a,

b).

In

particular, find r(7, 4).


(a, 6)
(1, 1)

using unknown scalars x and

y:
(,2)

a;(3, 1)

y(l, 1)

Hence

(a, b)

{Sx, x)

(y, y)

=
^6

{Sx

+ y,
and

'Zx

+ y)
y

and so
[^

+ +

y
y

a
b

Solving for x and y in terms of a and

b,

= ^o
1)

= -|a + f 6

(5)

Now

using

(2), {1)

and

(3),

T(a, b)

= =

xT{3,
{2x,

Thus

m, 4)

(7 - 4,

20 - 21)

+ yT(l, 1) = -4x) + (0, 2y) = = (3, -1).

oo(2,
(2a;,

+ 2/(0, 2) -4a; + 2y) = (a-b,5b- 3a)


-4)

CHAP.

6]

LINEAR MAPPINGS

143

6.37.

Let

that
(i)

T be the operator on R^ defined by T{x, y, z) = T is invertible. (ii) Find a formula for T~^.
The kernel

{2x, 4a;

y,2x + 3y-z).
=
(0, 0, 0),
i.e.,

(i)

Show

W of T

is

the set of all


T(, y,
z)

(x, y, z)
(2a;,

such that T{x,

y, z)

4x

-y,2x + Sy-z) =

(0, 0, 0)

Thus

is

the solution space of the homogeneous system


2a;

0,

4x

0,

2x

Sy

=
is

which has only the trivial solution Theorem 6.9 is invertible.


(ii)

(0, 0, 0).

Thus

W {0};
(a;,

hence T

nonsingular and so by

Let

(r, s, t)

be the image of
(r, s, f)

(x, y, z)

under T; then

T(x, y, z) = of r, s and t,

and T-^r,

s, t)

(x, y, z).

We

y, z) is the image of (r, will find the values of x,

s, t) under T^k y and z in terms

and then substitute in the above formula for T~^. From


T{x, y,
z)

=
7r

(2a;, 4a;

-y,2x + 3y-z) =
Thus T~^
is

{r, s, t)

we

find

^r,

2r

s,

Ss
s, t)

t.

given by

r-i(r,

= (^r,2r-s,lr-3s-t)
is if

6.38.

Let

be of
if

finite

invertible if

and only
is

dimension and let T be a linear operator on V. Recall that T if T is nonsingular or one-to-one. Show that T is invertible

and only
equivalent:
(v)

onto.

By Theorem 6.4, dim V = dim (Im T) + dim (Ker T). Hence the following statements are (i) T is onto, (ii) Im T = V, (iii) dim (Im r) = dimV, (iv) dim (Ker T) = 0, Ker T = {0}, (vi) T is nonsingular, (vii) T is invertible.

6.39.

Let

be of

finite

for some operator


invertible.
(ii)

dimension and let T be a linear operator on V for which TS = I, S on V. (We call S a right inverse of T.) (i) Show that T is Show that S = T~^. (iii) Give an example showing that the above

need not hold


(i)

if

V is
n.

of infinite dimension.
the preceding problem,

Let
is

dim

V=

By

is

invertible if

and only if rank T = n. Hence rank T = n and T is invertible.


invertible if
/.

We

have

n = rank

and only if T is onto; hence T / = rank TS rank T n.

(ii)

rr-i = r-ir =
Let

Then s

/s

= (r-ir)s =
t

r-i(rs)
p(t)

= r-/= =
ao

r-i.

(iii)

be the space of polynomials in

over K; say,

+ Oji + Ojf^ +
a^t

+ at".

Let

and S be the operators on


T(p{t))

defined by

ai

+ a2t+ = =

a""i

and

S{p(t))

a^t^

at+i

We

have

(rS)(p())

T(S{p{t)))

r(aot

Oo

ajt

+ Oit2 + + o< = p{t)


ii

+ a<+i)

and so TS
S(T(k))

I,

S(0)

the identity mapping. 0'k. Accordingly,

On

the other hand,


= I.

and

fc

# 0,

then

(,ST){k)

ST

6.40.

Let S and
{x, 0).

T be the linear operators on R^ defined by S{x, y) = Show that TS = but ST # 0. Also show that T^ = T.
=
{0,0).

(0,

x)

and T{x, y)
(a;,j/)GR2,

it

is

(TS){x,y) = T(S{x,y)) = T{0,x) the zero mapping: TS = 0.

Since

TS

assigns

(0,0)

to every

(ST){x,y)=S(T{x,y))
it

does not assign

(0, 0)

S(x,0) = (0,x). For example, to every element of R^.

(Sr)(4, 2)

(0, 4).

Thus ST

- 0,

since

For any

{x,y)

R2,

T^x.y)

T(T(x,y))

T{x,0)

{x,Q)

T{x,y).

Hence T^

=- T.

144

LINEAR MAPPINGS

[CHAP. 6

MISCELLANEOUS PROBLEMS
6.41.

Let {ei, ei, 63} be a basis of Furthermore, suppose


T{ei)
2^(62)

and

{/i, /z)

a basis of

TJ.

Let

T-.V^U

be linear.

T(e3)

= = -

aifi
&i/i
Ci/i

+ 02/2 + b2/2 + C2/2

and

A = ('''/
^

''^

Show that,

for any v

GV,

A[v]e

[T{v)]f

where the vectors

in

K^ and K^ are written

as column vectors.
Suppose V

fejei

+ fc2e2 + ^363;
T(v)

then

[f]e

h\ ^2
+
ksTie^)

Also,

= =

kiT{ei)

k^T^e^)

kiiaJi
(Olfcl

+ 0^2) +
feifeg

kiibJi

62/2)

hi'ifi

"2/2)

+ Cifcg)/! +
{ ^,\

{a2ki

62^2

+ C2fc3)/2

Accordmgly,
Computing, we obtain

[Tiv)],

,[,1

:^,l

^Me

=(_._

)lfc2)

[a,k,

+ b,k2+c,kj

f^(^)l'

6.42.

Let
is

A;

be a nonzero scalar.

singular.

Hence T
T
is

is

that a linear singular if and only if


T{v)

Show

map T

is

singular

if

and only

if

kT
=

is

singular.
0.

Suppose

singular.

Then

for some vector v =

Hence

ikT){v)

kT(v)

&0

and so kT
kT(w)

is

singular.

Now suppose kT is singular. Then = (kT)(w) = 0. But # and w


fc

#^

{kT^(w) implies

for some vector

w#

0;

hence

^(fcw)

kw

= 0;

thus

is

also singular.

6.43.

be a linear operator on V for which E^ = E. (Such an operator is termed a the kernel. Show that: (i) if m G C/, Let C/ be the image of E and then '(m) - u, i.e. 7 is the identity map on U; (ii) if E ^I, then ^ is singular, i.e. E{v) = for some v^O; (iii) V =

Let

projection.)

UW.

(i)

If

u&TJ,

the image of S, then

E{v)

u ^ E(v)
(ii)

= u for some v GV. Hence = EHv) = E(E{v)) = E{u)


where v
v^ m.

using E^

E, we have

U E = I
We
By
first

then, for
E'(v

some v F,

m) =

S(v)

=u S(m) =
Bl'y)

By

(i),

E(u) v

u.

Thus

m =

where

u=0

(iii)

show that

V - U+
J7,

W. Let v
U

e y. Set m = E(v) and w = v- E{v). Then = (l>) + '('U) = M + Of


I)

definition,

m = E{v) S
E(w)

the image of E.

E{v

- E(v)) {0}.

We now show that w e TF, the kernel = E(v) - E^v) = E(v) - E(v) Let v

of E:

and thus

wG

W. Hence

V = U + W.

We

v&W,

next show that E{v) = 0. Thus V

UnW
=

eUnW.

Since
{0}.

E(v)

and

so

UnW

vGU,

E(v)

v by

(i).

Since

The above two properties imply that

V = U

W.

CHAP.

6]

LINEAR MAPPINGS
invertible if

145

6.44.

Show that a square matrix A with Theorem 6.9, page 130.)

is

and only

if it is

nonsingular.

(Compare

Recall that A is invertible if and only if A is row equivalent to the identity matrix 7. Thus the following statements are equivalent: (i) A is invertible. (ii) A and 1 are row equivalent, (iii) The = and IX = have the same solution space, (iv) = has only the zero soluequations tion, (v) A is nonsingular.

AX

AX

Supplementary Problems
MAPPINGS
6.45.

State whether each diagram defines a

mapping from

{1, 2, 3} into {4, 5, 6}.

6.46.

Define each of the following mappings /


(i)

R -> R by
3.

a formula:

To each number

let

/ assign its square plus

(ii)

To each number

let / assign its

cube plus twice the number.

(iii)

To each number 3 the number 2.

let / assign the

number squared, and

to each

number <

3 let / assign

6.47.

Let

/:R^R

be defined by f(x)

= x^-4x + 3.

Find

(i)

/(4),

(ii)

/(-3),

(iii) /(j/

- 2a;),

(iv)/(a!-2).

6.48.

Determine the number of different mappings from

{o, 6} into {1, 2, 3}.

6.49.

Let the mapping g assign to each name in the set {Betty, Martin, David, Alan, Rebecca} the of different letters needed to spell the name. Find (i) the graph of g, (ii) the image of g.
Sketch the graph of each mapping:
f(x)

number

6.50.

(i)

^x

1,

(ii)

g(x)

2x^

4x 3.
are illustrated in the

6.51.

The mappings f:A-^B, g:B-^A, h:C-*B, diagram below.

F-.B^C

and

GiA^C

Determine whether each of the following defines a composition mapping and, if it does, find domain and co-domain: {\)gf, {n)hf, (iii) Fo/, (iv)Gf, {y)gh, (vi) hGg.
6.52.

its

Let

/:R^R

and

fir

R -^ R

be defined by f(x)
(i)

x^

+ Sx + l
(iii)

and g(x)
(iv)

= 2x-3.

Find formulas

defining the composition

mappings

fg,

(ii)

gf,
f

gg,

ff.

6.53.

For any mapping f:A->B, show that 1b f

f'^A-

146

LINEAR MAPPINGS
For each of the following mappings /
Sx

[CHAP.

6.54.

R -> R

find a

formula for the inverse mapping:

(i)

f{x)

- 7,

(ii)

fix)

+ 2.

LINEAR MAPPINGS
6.55.

Show
(i)

that the following mappings


:

are linear:
{2x

(ii) (iii)

F F
jF

R2

^ R2
-

defined by F(x, y)

- y,

x).

R3

R2 defined by F{x,
defined defined

y, z)

{z,x
Zx).

+ y).

(iv)

R -> R2 F R2 ^ R2
:

by F(x)
by F(x,

=
y)

(2.x,

[ax

hy, ex

+ dy)

where

a, 6, c,

R.

6.56.

Show
(i)

that the following mappings


:

F
=

are not linear:


(x^, y^).

(ii)
(iii)

F R2 ^ R2 F R3 ^ R2 F R ^ R2
: :

defined by F(x, y)
defined

by Fix, y,z)

ix

+ l,y + z).

defined

by Fix)

ix, 1).

(iv)

F:R2->R

defined by Fix,y)

\x-y\.
over K.

6.57.

Let

V
->

S :V

be the vector space of polynomials in V defined below are linear:

Show that

the

mappings

T :V -*V and

+ ait + S(ao + ai +
Tiaa
6.58.

+ a^t") = + at") =

a^t

a^t^

at" + i

ax

a^t

aj"--^

Let V be the vector space ot that the first two mappings


(i)

nXn

T -.V

r(A)

= MA,
b)

(ii)

TiA)
:

- MA

be an arbitrary matrix in V. matrices over K; and let are linear, but the third is not linear (unless -AM, (iii) TiA) =^ + A.

Show
0):

^V

M
=

M=

6.59.

Find Tia,
Find Tia,

where T R2
:

R3

is

defined

by by

r(l, 2)

(3,

-1,

5)

and

r(0, 1)

(2, 1,

-1).

6.60.

b, c)

where T RS
Til, 1,

^R 1) =

is 3,

defined

r(0, 1,

-2)

and

^(O, 0, 1)

= -2

6.6L

Suppose
Let

F:V -*U

is linear.

Show

that, for

any

vGV,
map

Fi-v)
of

-Fiv).
V, denoted by

6.62.

W be

defined by

a subspace of V. Show that the inclusion t(w) = w, is linear.

W into

i:W cV

and

KERNEL AND IMAGE OF LINEAR MAPPINGS


6.63.

For each of the following linear mappings F, find a basis and the dimension of and (6) its kernel W: (i) F R3 -> R8 defined by F(x, y, z) = ix + 2y,y-z,x + 2z). F R2 ^ R2 defined by Fix,y) = ix + y,x + y). (ii)
: :

(a)

its

image

(iii)

F
V

R3

^ R2

defined

by Fix, y,z)

ix

+ y,y + z). R
and
let

6.64.

Let

be the vector space of 2 X 2 matrices over

Let

F V -* V
:

be the

linear
(ii)

map defined by FiA) the image U of F.

= MA. ^ RS ^ RS

Find a basis and the dimension of

(i)

the kernel TF of

and

6.65.

Find a linear mapping Find a linear mapping


Let
Dif)

R3 R*

whose image whose kernel

is

generated by generated by

(1, 2, 3)

and

(4, 5, 6).

6.66.

is

(1, 2, 3, 4)

and

(0, 1, 1, 1).

6.67.

V be the = df/dt.

vector space of polynomials in t over R. Find the kernel and image of D.

Let

D:V -*V

be the differential operator:

6.68.

Let
(ii)

F:V-^U be linear. Show that the preimage of any subspace of U

(i)

is

the image of any subspace of a subspace of V.

is

a subspace of

and

CHAP.

6]

LINEAR MAPPINGS
K,*

147

6.69.

Each of the following matrices determines a linear map from

into R^:

'12
(i)

1^

A =

2
^1

-1 -3

2
2

-1
I

(ii)

B =

-2/

Find a basis and the dimension of the image


6.70.

and the kernel

W of each map.

Let
or

r C
:

->
bi)

C be

the conjugate

mapping on the complex

T(a +

= a bi
(ii)

where

a, 6

space over

itself,

Show

that

field C. That is, T(z) = z where z G C, R. (i) Show that T is not linear if C is viewed as a vector is linear if C is viewed as a vector space over the real field R.

OPERATIONS WITH LINEAR MAPPINGS 6.71. Let R3 - R2 and G R^ ^ R2 be


iJ'
:

defined

Find formulas defining the mappings

F+G

by F{x, y, z) and SF 20.

(y,x

+ z)

and G(x,

y, z)

(2,

- y).

6.72.

Let R2 - R2 be defined by H(x, y) (y, 2x). Using the mappings F and problem, find formulas defining the mappings: (i) and G, (ii) (in) Ho(F + G) and + HG.
:

in the preceding

HF

FH

and

GH,

HoF

6.73.

Show
(i)

that the following mappings F,

and

H are linearly independent:

F,G,He

Horn

(R2, R2)

defined

by

(ii)

Fix, y) = {x, 2y), G{x, y) = {y,x + y), H{x, y) = (0, x). F,G,He Hom (R3, R) defined by F{x, y, z) = x + y + z, G(x, y,z) y + z, H(x, y, z) =

z.
rank G.
(Here

6.74.

For

F,G & Rom

{V, U),

show

that

rank (F

G)

^ rank

i^

has

finite

dimension.)

6.75.

Let

F :V -^ U

nonsingular.

and G:U-*V be linear. Give an example where GF


all

Show
is

that if

and

are nonsingular then

GF

is

nonsingular but

is not.

6.76.

Prove that

Hom (V,

Theorem

6.6,

U) does satisfy page 128.

the required axioms of a vector space.

That

is,

prove

ALGEBRA OP LINEAR OPERATORS


6.77.

Let S and T be the linear operators on R2 defined by S{x, y) {x + y, Find formulas defining the operators S + T, 5S - ST, ST, TS, S^ and T^. Let
p{t)

0)

and

T{x, y)

(y,

x).

6.78.

be the linear operator on R2 defined by


t2

T{x, y)

{x

+ 2y,

3x

+ Ay).

Find p(T) where

_ 5f _ 2.
T on
R^
is

6.79.

Show
(i)

that each of the following operators

invertible,

and
y).

find

a formula for

T~h

T{x, y, z)

= (x-3y- 2z,

- 4,

z),

(ii)

T{x, y,z)

{x

+ z,x- z,

6.80.

Suppose
sion.

S and T

Show

are linear operators on that rank (ST) = rank (TS)

V and that S = rank T.

is

nonsingular.

Assume

has

finite

dimen-

6.81.

E2(v)

Suppose V = U W. Let Ei and E2 be the linear operators on V defined by Ei(v) = u, = w, where v = u + w, ue.U,w&W. Show that: (i) Bj = E^ and eI = E2, i.e. that Ei and E2E1 = 0. and ?2 are "projections"; (ii) Ei + E2 I, the identity mapping; (iii) E1E2 =
Let El and E2 be linear operators on
is

6.82.

satisfying

(i),

(ii)

and

(iii)

of Problem 6.81.

Show

that

the direct

sum

of the image of

E^ and the image of 2-

^ Im ?i Im 2ST
is

6.83.

Show that

if

the linear operators

S and T

are invertible, then

invertible

and (ST)-^

T-^S-^.

148

LINEAR MAPPINGS
Let

[CHAP.

6.84.

have
that

finite

dimension, and let


{0}.

T be

a linear operator on

such that

rank

(T^)

rank

T.

Show

Ker

TnlmT =

MISCELLANEOUS PROBLEMS
6.85.

Suppose
the

T -.K^-^

X"

is

mXn matrix whose


G
R"-,

vector V
6.86.

T(v)

a linear mapping. Let {e^, e} be the usual basis of Z" and let A be columns are the vectors r(ei), Show that, for every ., r(e) respectively. Av, where v is written as a column vector.
. . .

Suppose F -.V -* U is linear and same kernel and the same image.

fc

is

a nonzero scalar.

Show

that the

maps

and kF have the

6.87.

Show

that

if

F:V -^ U

is

onto, then

dim

U-

dim V.

Determine

all

linear

maps

T:W-*R*

which are onto.


6.88.

Find those theorems of Chapter 3 which prove that the space of w-square matrices over associative algebra over K.
Let

K
the

is

an

6.89.

T :V ^ U

be linear and

let

he a subspace of V.

The

restriction of

to

is

map

Tt^-.W^U defined by r^(w) = T{w), for every (iii) Im T^r = T(W). (ii) Ker T^ = Ker T n W.
6.90.

wGW.

Prove the following,

(i)

T^^ is linear.

Two operators S, T G A(V) are said to be similar if there exists an invertible operator P G A{V) for which S = P-^TP. Prove the following, (i) Similarity of operators is an equivalence relation. (ii) Similar operators have the same rank (when V has finite dimension).

Answers
6.45.
(i)

to

Supplementary Problems
{"^
if if

No,
fix)

(ii)

Yes,
(ii)

(iii)

No.

6.46.

(i)

x2

+ 3,

/()

a;3

+ 2a;,

(iii)

fix)

a;

[-2

<

3 3

6.47.

(i)

3,

(ii)

24,

(iii) j/2

- 4xy + 4x2 ^^y + ^x + S,

(iv) a;^

- 8a! + 15.

6.48.

Nine.
{(Betty, 4), (Martin, 6), (David, 4), (Alan, 3), (Rebecca, 5)}.

6.49.

(i)

(ii)

Image of g
{g o f)
:

{3, 4, 5, 6}.

6.51.

(i)

-*

A,

(ii)

No,

(iii)

(F o /)

- C,

(iv)

No,

(v)

(goh) :C

^ A,

(yi)

{hGg) iB

^ B.

6.52.

(i)

(/ g){x)

(ii)

(f^/)(a;)

= 4a;2 - 6a; + 1 = 2a;2 + 6a;-l + 7)/3,


(ii)

(iii)

(g o g)(x)
(/

=Ax-9
=
a;*

(iv)

/)(a;)

+ Ga;* + 14a;2 + 15x + 5

6.54.

(i)

f-Hx)

(x

/-!()

= V^^^^.

6.59.

T{a, b)

=
c)

{-a

+ 26, -3a +

6,

7a

- 6).

6.60.

T{a, b,

8o

36

2c.

6.61.

F(v)

F{-v)

F(v

+ (-u)) =
-2)},

F(0)

=
2;

0;

hence F(-v)

-F{v).

6.63.

(i)

(a) {(1, 0, 1), (0, 1, (a) {(1, 1)},

dim

U=

(6) {(2,

-1, -1)}, dim

W=

\.

(ii)

dim

?7

1; t/

(6) {(1,

-1)},

dim T^
-1,
1)},

1.

(iii)

(a) {(1, 0), (0, 1)},

dim

2;

(6) {(1,

dim

W=

\.

CHAP.

6]

LINEAR MAPPINGS

149

6.64.

(i)

U^
{(_2
z)

")'

(o

i)|

I'asisof

KerF; dim(KerF) =

2.

(")

l)' (I _2)|

^^sisof

ImF; dim(ImF) =

2.

6.65.

F(x, y,

{x

+ 4y,
(x

2x

+ 5y,
2x

Sx

+ 6y).
0).

6.66.

F(x, y,z,w)

+ y - z,

+ y - w,

6.67.

The kernel of
(i)

is

the set of constant polynomials.

The image of

is

the entire space V.

6.69.

(a)
(6)

{(1,2,1), (0,1,1)}
{(4,

basis of

-2, -5,
R3;

0), (1,

-3,

0, 5)}

Im A; dim(ImA) = 2. basis of KerA; dim(KerA) =


basis of

2.

(ii)

(a)

ImB =

(6)

{(-1,2/3,1,1)}

KerB; dim(KerB)
z)

l.

6.71.

(F

G)(,x, y, z)

(y

+ 2z, 2x-y + z),


{x

(3F

2G)(x, y,

(3j/

-Az,x + 2y + Zz).

6.72.

(i)

(iii)

+ z,2y), (HG)(x,y,z) - {x-y.iz). (il) Not defined. (Ho(F + G)){x, y,z) = {HoF + Ho G)(x, y, z) = {2x-y + z, 2y + 4).
(HF){x,y,z)

6.77.

+ T)(x, y) = (x, x) (5S - 3r)(a;, y) = (5a; + 8y,


(S

(ST){x, y)

=z

{x- y,
(0,
a;

0)

-3x)

(TS){x, y)

+ y) +

SHx,

v)

Ti(x^ y)

= =
0.

(x

+ y,

0);

note that S^

S.

{-X, -y); note that

T^-\-I

Q,

hence

is

a zero of

x'^

1.

6.78.

v{T)

6.79.

(i)

T-Hr,

s, t)

(14*

+ 3s + r,

4t

+ s,

t),

(ii)

T-^r,

s, t)

(^r

+ ^s,

t,

^r

- |s).

6.87.

There are no linear maps from RS into R* which are onto.

chapter 7

Matrices and Linear Operators


INTRODUCTION
and, for v GV, suppose a basis of a vector space V over a field the coordinate vector of v relative to {ei}, which we write as a column vector unless otherwise specified or implied, is

Suppose
ttiei

{ei,

e} is

+ 0.262 +

+ Omen. Then

\an

Recall that the

mapping v

l^ [v]e,

determined by the basis

{Ci}, is

an isomorphism from

onto the space K".


{ei},

In this chapter we show that there is also an isomorphism, determined by the basis from the algebra A{V) of linear operators on V onto the algebra cA of n-square matrices

over K.

similar result also holds for linear

mappings F:V-^U, from one space

into another.

MATRIX REPRESENTATION OF A LINEAR OPERATOR


en} is and suppose (ei, a linear operator on a vector space V over a field r(e) are vectors in V and so each is a linear combination of Now T{ei), the elements of the basis {e,}:

Let

r be

a basis of V.

T(ei)
r(e2)

= = =

anCi
02161

+ +
+

+ 02262 +
01262

+ +

oi^en a2n6n

T{en)

Oniei

an2e2

oe

The following
Definition:

definition applies.
coefficients,

The transpose of the above matrix of


called the

denoted by

[T]e or [T], is

matrix representation of

relative to the basis {ei} or simply the


021

matrix of

in the basis

{et}:

(On
012

...
. .

Onl
fln2

'

022

Om
Example
7.1
:

a2n

...

0,

-* V Let V be the vector space of polynomials in t over R of degree ^ 3, and let D V be the differential operator defined by D{p(t)) = d{p(t))/dt. We compute the matrix of D in the basis {1, t, t^, fi}. We have:
:

D(l)
D(t)
D(fi) D{fi)

= = = =

= + Of + 0*2 + 0*3 1 = 1 + Ot + 0(2 + 0*3 + 2t + 0f2 + 03 It = + Ot + 3t2 + 0t3 3t2 =

150

CHAP.

7]

MATRICES AND LINEAR OPERATORS


Accordingly,
[D]

151

Example

7.2:

Let T be the linear operator on R2 defined by T(x, y) (ix 2y, 2x + y pute the matrix of T in the basis {/i = (1, 1), /a = (-1, 0)}. We have
T(Ji) Tifz)

We

com-

r(l, 1)

=
0)

(2,3)

3(1, 1)

(-1, 0)
1)

3/1
0)

/2

= n-1,

=
-2
2

(-4, -2)

-2(1,

2(-l,

-2/1

2/2

(3

Remark:

Recall that any n-square matrix

the
7.7)
if

map

A over defines a linear operator on K" by (where v is written as a column vector). We show (Problem that the matrix representation of this operator is precisely the matrix A
v
t^

Av

we

use the usual basis of K".


is

Our
by
its

first theorem tells us that the "action" of an operator T on a vector v matrix representation:
7.1:

preserved

Theorem
That
then

Let

(ei,

.,

e}

be a basis of
[T]e [v]e

and

let

be any operator on V.

Then, for

any vector

vGV,

[Tiv)]e.

we

is, if we multiply the coordinate vector of v by the matrix representation of T, obtain the coordinate vector of T{v).

Example

7.3:

Consider the differential operator


p{t)

D:V -^V
and so
t^,

in

Example
D{p{t))

7.1.

Let

= a+bt +

cfi

dt^

2ct

3dt^

Hence, relative to the basis

{1,

t, t^,

[p(t)]

and

[D(p{t))]

We

show that Theorem

7.1

does hold here:


'0
1

o\
2

[D][Pit)]

=
iO

3,,

= [D(pm

o/\
:

Example

7.4:

Consider the linear operator Let V = (5, 7). Then V


T{v)

T R2 ^ R2

in

Example

7.2:

T{x, y)

(4a;

2y,

2x

= =

(5,7)
(6,

=
=

7(1, 1)

2(-l, 0)

=
0)

7/1

2/2

17)

17(1, 1)

11(-1,

17/i

11/2
{/i, /a),

where

/j

(1, 1)

and

fz

(-1, 0).

Hence, relative to the basis

and

[T(v)]f

=
11

Using the matrix

[T]f in

Example

7.2,

we

verify that Theorem 7.1 holds here:

m,M,

(r^G)
<2y

(ID = i^<*

152

MATRICES AND LINEAR OPERATORS

[CHAP. 7

Now we have associated


on V.

By our

first

representation.

a matrix [T]e to each T in A{V), the algebra of linear operators theorem the action of an individual operator T is preserved by this The next two theorems tell us that the three basic operations with these
(i)

operators
addition,
(ii)

scalar multiplication,

(iii)

composition

are also preserved.

Theorem

7.2:

Let {ei, ...,e} be a basis of V over K, and let oA be the algebra of -square matrices over K. Then the mapping T h* [T]e is a vector space isomorphism from A{V) onto cA. That is, the mapping is one-one and onto
and, for any

S,T G A{V) and any


[T

keK,
and
[kT]e

+ S]e =

[T]e+[S]e
[ST]e

k[T]e

Theorem

7.3:

For any operators S,T G A{V),


are operators on
T{ei)
7(62)

[S]e [T]e.

We
V, and

illustrate the

above theorems in the case dim V

2.

Suppose

{ei, ez} is

a basis of

T and S

V for which
aiei
biei

= =

+ a^ez + 6262

S{ei)
'

8(62)

+ 0262 diCi + did


CiCi

[^i-Cy
Now we have
{T

S(ei)

i^i-

+ S){ei) + S){e2) =

T{ei)

aiCi
(tti

= (:;*; + 0262 + ciCi +


(a2

6262

=
(T
r(e2)

+ Ci)ei +
+
6262

+ 62)62 +
d^ez

+ 5(62) = =
/!

bid
(&i

did

+ di)ei +
/ci
,

(62

+ ^2)62

'^^"
'tti
^ J

l^ttz

+ +

ci

&i

ca

bz

+ di\ + dzj =

&i\
bzj

di

[az

\cz

dzj

Also, for k

EK, we

have
(A;r)(ei)

= =

fcr(ei)

{kT){ez)

kTiez)

+ azBz) = = k(biei + bzez) =


k{aiei

kaiCi

+ +

ka^ez

kbiei

kbzez

fkai

kbi\

/ai

bi\

.-m.

Finally,

we have
(Sr)(ei) == S(r(ei))

+ a2e2) = aiS(ei) + = ai{ciei + CzCz) + azidiBi + dzCz) = (ttiCi + a2di)ei + (aiCz + azdz)ez =
S(aiei

a2S(e2)

(Sr)(e2)

= = =

S{T{ez))
bi{ciei
{biCi

S(biei

+ bzCz) =

biS{ei)

b^Siez)

+ CzCz) +

bzidiBi
{biCz

+ dzBz)

+ bzdi)ei +
biCi

+ bzdz)ez
/ci

Accordingly,
^

J'

_ ~

/aiCi

[aicz

+ azdi + azdz

biCz

+ +

bzdi\

bzdz)

_ ~

dA

/ai

bi\

._

[cz

dz) \az

bzj

L>IJ

CHAP.

7]

MATRICES AND LINEAR OPERATORS

153

CHANGE OF BASIS
We have shown that we can represent vectors by n-tuples (column vectors) and linear operators by matrices once we have selected a basis. We ask the following natural question: How does our representation change if we select another basis? In order to answer this
question,

we

first

need a definition.
[ei,

Definition:

Let

.,e}

be a basis of
/i

and

let {/i, ...,/}

be another basis.

Suppose

= = =

anei
aziBi

+
+

ai2C2
022^2

+ +
+

aie
a2T.e

/2

+
+

fn

ttnlCi

a2e2

UnnCn
is

Then the transpose P of the above matrix of coeflScients tion matrix from the "old" basis {d} to the "new" basis
'

termed the transi-

{/{}:

fflll

ft21

ftnl
ffin2

*12

^'22

We comment
{/,}

invertible (Problem 5.47).

that since the vectors fi, .,fn are linearly independent, the matrix P is In fact, its inverse P^Ms the transition matrix from the basis back to the basis {Ci}.
. .

Example

7.5:

Consider the following two bases of R^:


{ei

(1, 0),

62

(0, 1)}

and

ih

=
=

(1,

D,

A=
+

(-1. 0)}

Then

A =
/2

(1,1)

=
=

(1,0)

(0,1)

e^

e^

(-1,0)

-(1,0)

0(0,1)

-ei

0e2
is

Hence the transition matrix

P from

the basis {ej} to the basis {/J


'1

-V
(-1, 0)

We

also

have

e^

= =

(1, 0)

= =

0(1, 1)

O/i

/g

62

(0,1)

(1,1)

(-1,0)

= /1+/2
to the basis {e^} is

Hence the transition matrix Q from the basis {/J back

Q
Observe that

IN
,

and Q are inverses:

We now show how coordinate vectors are affected by a change of basis.


Theorem
7.4:

Let P be the transition matrix from a basis space V. Then, for any vector v G V, P[v]f

{Ci}

to a basis {fi} in a vector

[v]e.

Hence

[v]f

P~'^[v]e.

We
{Ci} to

emphasize that even though


the

new

basis

{/i}, its effect is

is called the transition matrix from the old basis to transform the coordinates of a vector in the new
{ei}.

basis

{fi}

back to the coordinates in the old basis

154

MATRICES AND LINEAR OPERATORS

[CHAP. 7

We

illustrate the

matrix from a basis

above theorem in the case dim F = 3. Suppose {61,62,63} of F to a basis {fufzifa} of V; say,
ttiCi

P
ci\
C2

is

the transition

A =
/2

fa

= =

+ 0262 + 0363 biBi + 6262 + 6363 C161 + 6262 + 6363


v

1 0.1

bi

Hence

P =

02

&2
ba

\aa

Caj
/i

Now

suppose V

GF

and, say,

fei/i

+ /i;2/2 + fcs/s.

Then, substituting for the

from

above,

we

obtain

= =

+ a262 + a363) + fc2(biei + 6262 + 6363) + kaiciei + 0262 + 6363) (aifci + bife + Cika)ei. + (azki + bzki + C2ka)e2 + {aaki + bakz + 63^3)63
/i;i(aiei

Thus
[v],

/jcA

Ikz]
\lc3l

and

[v]e

+ a^ki + \a3ki +
jaiki
bik2
62^2

+ Cifc3^ 62^2 + dka b3k2 + cakal


bife

Accordingly,
P[v]f

^^ &2 ba

^,^\

jaikr

a2

62^2
Caj\kaj

ttzifci

\a3

\a3k1

+ + +

+ cM + 62^3 bakz + Cakaj

[V]e

Also, multiplying the above equation

by P~S we have

P-'[v]e Example
7.6:

P-'P[v]f

=
+

I[V],

=
=

[V],

Let v

(a, b)

e
V
V

R2.

Then, for the bases of R* in the preceding example,


(a, b)
(a, 6)

= ^

= =
=-

a(l, 0)

6(0, 1)

ae^

b{l,l)

+ (b-a)(-l,0)
and

+ =

be^
bfi

ib-a)/^

Hence

[v]^

Mf = P

(J^j

(ft

_ )
to {/J

By the preceding example, p-i are given by

the transition matrix

from {ej

and

its

inverse

I'D
We
verify the result of

'- =

(-:

Theorem

7.4:

-M.

= (_:

DC)

= (.!.) =

..

The next theorem shows how matrix representations of linear operators are affected by a change of basis.

Theorem

7.5:

Let P be the transition matrix from a basis {Ci} to a basis {/i} in a vector space V. Then for any linear operator T on F, [T]t = P-i[T]eP.
7.7:

Example

Let T be the linear operator on E^ defined by the bases of R^ in Example 7.5, we have
r(ei)

T(x, y)

(4a;

- 2j/,
4ei

2a;

j/).

Then for

ne^)
Accordingly,

= =

r(l, 0)

:=

(4, 2)

= =

4(1, 0)

2(0, 1)

2e2

r(0,l)

(-2,1)

-2(1,0)

(0,1)

-2ei

e2

/4 -2
[T]e

V2

CHAP.

7]

MATRICES AND LINEAR OPERATORS

155

We

compute

[T]f

using Theorem

7.5:

m, - p-i... = (_:
Note that
this agrees

ixi -ixi
[T]f in

-I)

7.2.

(i -I

with the derivation of

Example

Remark:

Suppose

P-

(Oij)

{ei, .... en} is

is any -square invertible matrix over a field K. a basis of a vector space V over K, then the n vectors
/i

Now

if

aiiCi

02162

aie,

i=l,

.,n

are linearly independent (Problem 5.47) and so form another basis of V. Furthermore, P is the transition matrix from the basis {{} to the basis {/{}. Accordingly, if A is any matrix representation of a linear operator T on V, then the matrix B = P~^AP is also a matrix representation of T.

SIMILARITY
Suppose A and B are square matrices for vs^hich there exists an invertible matrix P such that B = P~^AP. Then B is said to be similar to A or is said to be obtained from A by a similarity transformation. We show (Problem 7.22) that similarity of matrices is an equivalence relation. Thus by Theorem 7.5 and the above remark, we have the following
basic result.

Theorem
That

7.6:

Two

matrices A and B represent the same linear operator they are similar to each other.

if

and only

if

is, all

the matrix representations of the linear operator

T form an

equivalence

class of similar matrices.

linear operator

is

said to be diagonalizable
{{} is

by a diagonal matrix; the basis


gives us the following result.

if for some basis (Ci} it is represented then said to diagonalize T. The preceding theorem

Theorem

7.7:

be a matrix representation of a linear operator T. diagonalizable if and only if there exists an invertible matrix
Let

Then T

is

such that

P~^AP

is

a diagonal matrix.
if

That is, T is diagonalizable by a similarity transformation.

and only

if its

matrix representation can be diagonalized

We emphasize that not every operator is diagonalizable. However, we will show (Chapter 10) that every operator T can be represented by certain "standard" matrices called its normal or canonical forms. We comment now that that discussion will require some theory of fields, polynomials and determinants.
a function on square matrices which assigns the same value to similar matrices; that is, f{A) = f{B) whenever A is similar to B. Then / induces a function, also denoted by /, on linear operators T in the following natural way: f{T) = f{[T]e), where {d} The function is well-defined by the preceding theorem. is any basis.
suppose /
is

Now

perhaps the most important example of the above type of functions. Another important example follows.
is

The determinant
Example

7.8:

The
its

trace of a square matrix diagonal elements:

A =

(oy),

written tr (A),
022

is

defined to be the

sum

of

tr (A)

= an +

We

show (Problem 7.22) that similar matrices have the same trace. Thus we can speak of the trace of a linear operator T; it is the trace of any one of its matrix
representations:
tr {T)

tr ([T]g).

156

MATRICES AND LINEAR OPERATORS

[CHAP.

MATRICES AND LINEAR MAPPINGS We now consider the general case of linear mappings from one space into another. Let V and U be vector spaces over the same field K and, say, dim V = m and dim U = n. Furthermore, let {ei, em} and {/i, ...,/} be arbitrary but fixed bases of V and U
. . .

respectively.

U and

Suppose so each

F:V^U
is

Then the vectors F{ei), is a linear mapping. a linear combination of the fc


F{ei)
F{e2)

..,

F{em) belong to

= =

aii/i
ttzi/i

+ + +

ai2/2 022/2

+ + +

+ aifn + aznfn +
Ctmn/n

F{em)

ttml/l

dmifl

'

The transpose
and
{/i}:

representation of

of the above matrix of coefficients, denoted by [F]l is called the matrix relative to the bases {ei} and {ft}, or the matrix of F in the bases {ec}

/ ftll

ft21

ami
ttm2

rmf
L^Je

_ \

^^12

CI22

din

0/2n

dmn

The following theorems

apply.
[F]l [v]e

Theorem
That

7.8:
is,

For any vector v GV,

[F{v)],.
(ei}

multiplying the coordinate vector of v in the basis obtain the coordinate vector of F{v) in the basis {fi).

by the matrix

[F]l,

we

Theorem

7.9:

The mapping F ^ [F]f is an isomorphism from Hem (V, U) onto the vector space of % X m matrices over K. That is, the mapping is one-one and onto
and, for any F,

G G Horn {V, U) and any e K, [kF]f = and [F + G]f = [F]i + [G]/


fc

k[F]i

Remark:

over K has been identified with the linear mapZ" given by v M' Av. Now suppose V and U are vector and n respectively, and suppose {e;} is a basis spaces over K of dimensions of V and {fi} is a basis of U. Then in view of the preceding theorem, we shall given by [F{v)]f = A[v]e. We also identify A with the linear mapping are given, then A is identified with comment that if other bases of V and U
Recall that any ping from K'" into

nxm matrix A

F:V^U

another linear mapping from

into U.

Theorem

7.10:

Let

{ei}, {fi}

and

and

G:U-*W

be bases of V, U and be linear mappings. Then


{Qi}

W respectively.

Let

F:V-*U

[GoFYe

[GYfWVe

That is, relative to the appropriate bases, the matrix representation of the composition of the of two linear mappings is equal to the product of the matrix representations individual mappings.

We lastly show how the matrix


when new bases are
Theorem
7.11:

representation of a linear

mapping

F:V-*U

is

affected

selected.

P be the transition matrix from a basis {ei} to a basis (e,'} in V, and let Qbe the transition matrix from a basis {/i} to a basis {//} in [/. Then for any linear mapping F:V ^ U,
Let

[Ft =

Q-'inp

CHAP.

7]

MATRICES AND LINEAR OPERATORS


Thus
i.e.

157

in particular,

when

the change of basis only takes place in


[F]l.

JJ;

and

[F]iP

i.e.

when

the change of basis only takes place in V.


7.1, 7.2, 7.3

Note that Theorems and 7.11 respectively.

and

7.5 are special cases of

Theorems

7.8, 7.9, 7.10

The next theorem shows that every linear mapping from one space into another can be represented by a very simple matrix.

Theorem

7.12:

Let

F:V-*U be linear and, say, rankF = r. Then there V and of V such that the matrix representation of F has the

exist bases of

form

A =
where / is the r-square identity matrix. form of F.

We

call

the normal or canonical

WARNING
As noted previously, some texts write the operator symbol V on which it acts, that is, vT instead of T{v)

to the right of the vector

In such texts, vectors and operators are represented by n-tuples and matrices which are the transposes of those appearing here. That is, if
felCl

feez

knCn

then they write


[v]e

(A;i,

fe,

.,

kn)

instead of

[v]e

And

if

r(ei)
T{e2)

= = =

aiei 6iei

+ aid + +
+
6262

+ aen +
+
&ne

+
+

r(e)

ciei

0262

ce

then they write


'tti

Oi
b2

[T]e

bi

instead of

[T]e

lCi

C2

is also true for the transition matrix from one basis to another and for matrix repcomment that such texts have theorems resentations of linear mappings F:V ^ U. which are analogous to the ones appearing here.

This

We

158

MATRICES AND LINEAR OPERATORS

[CHAP.

Solved Problems

MATRIX REPRESENTATIONS OF LINEAR OPERATORS


7.1.

Find the matrix representation of each of the following operators


the usual
(i)

T on

R^ relative to

T{x, y) Note

basis {ei = (1, 0), 62 = (0, 1)}: = {2y, Sx - y), (ii) T{x, y) = (3x -4y,x + 5y).
that
if
(a, b)

first

R2,

then
0ei

(a, b)

ae^

be^.

(i)

r(ei) T{e^)

= =
= =

r(l,0)
T{Q,1)

= =
= =

(0,3)

(2,-1)
(3,1)

= = = =

3e2
e^
62

^ and

.^, rri.

= =

/O
(

2ei3ei+
-461

\S -1

(ii)

r(ei) Tie^)

r(l,0)
r(0,l)

/3 -4
(

(-4,5)

and
562

[rig

\1

7.2.

Find the matrix representation of each operator T in the preceding problem relative to the basis {A = (1,3), /a = (2, 5)}. We must first find the coordinates of an arbitrary vector (a, b) G K^ with respect to the basis
{/J.

We

have
(a, b)

=
+ =

x(l, 3)

+
a
5a

2/(2, 5)

(x

+ 2y,
5y 3a

Zx

+ 5y)
b

or or

X
a;

2y
26

and and

Sx

+ ~

=
6

=
= =

Thus
(i)

(a, 6)

(26

- 5a)/i +
+ +
I8/2

(3a

- 6)/2
30
18

We

have

T{x, y)

(2y,

Sx

- y).
(6,0)
(10, 1)

Hence
-3O/1

r(/i)

r(/2)

= =
y)

r(l,3)
r(2, 5)

= =

and
29/2

[T]f

-48
29

-48/i

(ii)

We

have T(x,
r(/i)

T(h)

4y,x + 5y). Hence = r(l,3) = (-9,16) = llfi-ASf^ = r(2,5) = (-14,27) = 124/1-69/2


=
(3x

and

77
[T]f

124

-43

-69

7.3.

Suppose that T

is

the linear operator on R^ defined by


z)

T{x, y,

=
T

(ttiic

+ a2.y + aaz,

bix

+ h^y + bsz,

cix

+ dy + Cs^)

Show

that the matrix of

in the usual basis (ei} is given


ai
[T]e

by

a2
&2
C2

as
bs
Cs

61
Ci

the rows of [T]e are obtained ponents of T{x, y, z).

That

is,

from the
(ai, 61, Ci) (02, 62, C2) (aa, 63, C3)

coefficients of x,

y and

z in the

com-

T{ei) 7(62) 7(63)

= = =

T{1, 0, 0) T(0, 1, 0)

r(0, 0, 1)

= = =

= = =

a^ei 0361 agei

+ + +

b^ez
6262

6363

+ + +

c^e^
6363 6363

Accordingly,
lT]e

/ai

03
62

aaX
63
1

(h

Remark:

This property holds for any space K'^ but only relative to the usual basis
{ei

(l, 0,

...,0), 62

(0,1,0, ...,0),

..., e

(0,

...,0,1)}

CHAP.

7]

MATRICES AND LINEAR OPERATORS

159

7.4.

Find the matrix representation of each of the following linear operators


relative to the usual basis
(i)

on R^

(ei

(1, 0, 0), 62

(0, 1, 0), 63

(0, 0, 1)}:

T{x,y,z) T{x,y,z)

(ii)

- {2x-Zy + Az,5x-y + 2z,Ax + ly), = {2y + z,x-4:y,Zx).


7.3:
(i)

By Problem

[T]^

-1

(ii)

[T]^

=
\

-4

7.5.

Let
(i)

T be

(ii)

= Find the matrix of T in the basis {/i = (1, 1, 1), /a = Verify that [T], [v]f = [T{v)]s for any vector v G R^.
the linear operator on R^ defined

by

T(x, y, z)

+ z,x Ay, 3a;). (1, 1, 0), fa = (1, 0, 0)}


{2y

must first find the coordinates of an arbitrary vector (a, h, c) G R^ with respect to the basis {fvfz'fai- Write {a,b,c) as a linear combination of the /j using unknown scalars x, y and z:
(a, b, c)

We

= =

x{l, 1, 1)
(x

y{l, 1, 0)

z(l, 0, 0)

+ y + z,x + y,x) =
=
c,

Set corresponding components equal to each other to obtain the system of equations

a,
a, b

b,

Solve the system for x, y and z in terms of


(a, 6, c)

and

c to find

~e,

= a b.

Thus

c/i

+ (6 - c)/2 +

(a - 6)/3

(i)

Since

T(x,y,z)

(2j/

r(/i)
r(/2)

T{fa)

= = =
~
V

r(l,l,l) r(l,l,0) r(l,0,0)


(a, b, c);

+ z, - 4j/, 3a;) = (3,-3,3) = 3/1-6/2 + 6/3 = (2,-3,3) = 3/1-6/2 + 5/3 = (0,1,3) = 3/1-2/2- /a
a;

and

[T]f

(ii)

Suppose v

then

(a,b,c)

c/i

(6

c)/2 +

(a

6)/3

and so

[v]

Also,
T{v)

= =

T(a,

b, c)

3a/i

(-2a

(26 + c, a - 46, 3(i) - 46)/2 + (-0 + 66 + c)/3

and so

[T{v)]f

Thus
[T]f[v]f

-6 -6 -2

6-c

-2a-45

lT(v)]f

7.6.

Let V
is

A {ei
{/i

(
j

and

let

T be

the linear operator on R^ defined by T{v)

= Av

(where

written as a column vector).

Find the matrix of


the usual basis;

in

each of the following bases:

(i)

(ii)

= (1, 0), = (l,3),

= (0, 1)}, /2 = (2,5)}.


62

i.e.

(i)

Tie,)

2yi\ =(l 4/\0/ ^ (I) = u, + \3 ^3/


-

3e,-

/I

^(^^)={3 4)(l)

=(:) =2ex + 4e2

160

MATRICES AND LINEAR OPERATORS

[CHAP.

Observe that the matrix of T in the usual basis is precisely the original matrix A which This is not unusual. In fact, we show in the next problem that this is true for any matrix A when using the usual basis.
defined T.
(ii)

By Problem

7.2,

(o, 5)

(26

- 5o)/i +

(3o

- h)!^.

Hence

^< = (a

X)

-'-'
/-5
and thus
[r]/

-8N
loy

7.7.

any -square matrix A = {an) may be viewed as the linear operator T on K" by T{v) = Av, where v is written as a column vector. Show that the matrix representation of T relative to the usual basis {et} of K" is the matrix A, that
Recall that
defined
is,

[T]e

= A.
I Oil

Ol2

lJl

\/ 1

fail

T(ei)

Aei

OiiBi

021^2

aie

<12

T(ei)

Aea

=
aji2j

n2

Oil

<*12

r(e)

= Ae =
I

2i

22

lnl

02n2

Onnen

(That

is,

r(e,)

Ae,

is

the ith column of A.)


0.11

Accordingly,

%2
*22

''itl

me

=
,

021

<*2n

= A
J

Onl

<'^n2

a.nn

7.8.

Each

of the sets (i) {l,t,e\te*) and (ii) {f^\t^*,t^e^*} is a basis of a vector space V be the differential operator on V, that is, D{f) df/dt. of functions / R - R, Let in the given basis. Find the matrix of
:

(i)

I>(1)

= = 1 I?(t) >(et) = e* i)(e) = e +

te*

= = = =

0(1)
1(1) 0(1) 0(1)

+ + + +

+ + 0(t) + 0(t) +
0(t)
0(t)

0(e) 0(et)

l(eO
l(e)

+ + + +

O(te')

O(te')

and
0(<et)
l(tet)

[D]

(ii)

= = 3e = D(<e30 = eSt + 3e8t 2)(t2e3t) = 2teS + Bt^e^* =


2)(e3)

3(e30 l(e30
0(e3)

+ + +

0(e3') 3(e3t)

2(e3')

+ 0(t263t) + 0(t2e3t) + 3(t2e3')

and

[D]

CHAP.

7]

MATRICES AND LINEAR OPERATORS


.,e} is a basis of

161

7.9.

Prove Theorem 7.1: Suppose {ei, on F. Then for any vGV, [T]e [v]e
. .

and T

is

a linear operator

[T{v)]e.

Suppose, for

1,

.,n,

n
Tiei)

Bjiei

ffljaea

+
is

Oie

%}

Then

[r]e is the n-square

matrix whose ith row

Now

suppose

k^ei

kzBz

fce

K^i

Writing a column vector as the transpose of a row vector,


[v],

(&i,

fcj,

...,fe)t

(2)

Furthermore, using the linearity of T,


T{v)

= =

T
n

(^

2 he^ = 2
/

hned =
n

fci (^

ae,-

j 1 \ i=l

2(2

*>

) j

2
i
is

(oij-fci

+ a2jfc2 ^

h a^fe)ej

Thus

[r(i;)]g is

the column vector whose jth entry


aijfci

a^jk^
is

aj&

(^)

On

i.e. (1)

the other hand, the ith entry of [r]e[^]e by (2). But the product of (1) and (2)
[T], [v],

is (3);

obtained by multiplying the ;th row of [T\g by [v]^, hence [r]c[v]e and [T(v)\g have the same entries.

Thus

[T(v%.

7.10.

V over X^, and let cA be the algebra Then the mapping T ^ [T]e is a vector space isomorphism from A{V) onto cA. That is, the mapping is one-one and onto and, for any S,T& A{V) and any kGK, [T + S\e = {T]e + [S\e and [kT]e = k[T]e.
Prove Theorem
7.2:

Let

{ei,

e}

be a basis of

of %-square matrices over K.

The mapping
its

is

values on a basis. operator

one-one since, by Theorem 8.1, a linear mapping is completely determined by The mapping is onto since each matrix & cA is the image of the linear

F(e^

2
i=l

e^

l,...,n

where (wy)

is

the transpose of the matrix


i

M.
n

Now

suppose, for

1,

. ,

w,

T{eO

2 i=i
(ay)

cmej

and

S{ei)

2 i=i
[r]^

SijCj

Let
for

A
i

and

be the matrices

A=

and

B=
+

(6y).

Then

A* and
6)ej

[5]^

= B*. We

have,

1,

...,%,

(r + SKej)

T(ei)

S{ei)

2K +
=
[r],+

Observe that

A + J?

is

the matrix (ay

+ 6y).

Accordingly,

[T

+ S], = (A+B)t =

A'

fit

[S]e

We

also have, for

1, ..

.,n,

n
(fcrXej)

n ayej

T(et)

fc

ikaij)ej

Observe that

kA

is

the matrix

(fcay).

Accordingly,

[kT],

(kA)t

kAt

k[T],

Thus the theorem

is

proved.

162

MATRICES AND LINEAR OPERATORS


Prove Theorem
7.3:

[CHAP.

7.11.

Let

{ei,

e}

be a basis of V.
n

Then for any

linear operators

S,Te A{V),
Suppose

[ST]e

=
n

[S]e [T]e.

r(ej)

2 1=1
=

"ij^j

and

S(ej)

=
B*.

2 fc=l

6jk/c.

Let

and

be the matrices

A=

(ay)

and

B=

(bjk).

Then

[T]^

A* and

[S]^

We have
= sCSoije,) =
\i 1
\
)

(ST)iei)

=
=

S(7'(ei))

2
i n
1

aS(e,)
\

2 =

n/
a
(

\fc

2 =

6ifc6fc

=
n

IC=1

2(2 \3 =

aijftjic

k
/

Recall that

AB

is

the matrix

AB =

(cjfc)

where
(AB)t

Cj^

=
J

2
=

"iibjk-

Accordingly,

[ST],

B*At

[S]AT]e

CHANGE OF
7.12.

BASIS, SIMILAR

MATRICES

Consider these bases of R^: {ei = (1,0), cz = (0,1)} and {/i ^ (1,3), /2 = (2,5)}. (ii) Find the transition matrix Q (i) Find the transition matrix P from {ei} to {/i}. that Q = P'K (iv) Show that [vy = P-^[v]e for any from {/i} to {ei}. (iii) Verify vector V eR^ (v) Show that [T]f = P-'[T]eP for the operator T on R^ defined by
T{x, y)
(i)

{2y,

Sx

- y).

(See Problems 7.1 and 7.2.)


/i

/2

= =

(1,3) (2,5)

= =

lei 261

+ +

362

^^^
562

p =

/^

^
5

\3 Thus
^^^

(ii)

By Problem

7.2,

(a, 6)

= = =

(26

- 5a)/i +
=
-5/1

(3a

- 6)/2.

61

(1,0) (0,1)

3/2

Q^/-5
V
3

62

=2/1-/2
(3
and

-1

(-)

^ =
If
i;

5)(~3 -1) = Co

1)

"

'

(iv)

/a\
(a,6),

then

Me=(j,)
P-'\vl

M/ =

/26-5a\
(

g^-^)3a

Hence

-5
3

2\/a\

_ /-5a +26
I

-1A6/ ~
2\
and
[T]f

-6
/-30 -48 \ 29 j" (^13
^""^

/O
(v)

By Problems

7.1

and

7.2;

['ne=(g_^)

7.13.

= (1,0,0)}. (i) Find the transition matrix P f rom {ei} {/i = (1,1,1), /2 = (1,1,0), /3 that Q = P \ to {/i}. (ii) Find the transition matrix Q from {A} to {ei}. (iii) Verify
(iv)

Consider the following bases of R:

{ei

(1,0,0), 62

(0,1,0), 63

(0,0,1)}

and

Show

for the
(i)

that [v]/ = P-^[v]e for any vector v G R^ (v) Show that [T]f = P ^[T]eP defined by T{x, y, z) = {2y + z,x- Ay, 3a;). (See Problems 7.4 and 7.5.)
/l

/a fs

= = =

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

(1,0,0)

= Iei+l62+l63 = lei + 1^2 + Oeg = lei + 062 + 063

and

P =

CHAP.

7]

MATRICES AND LINEAR OPERATORS


By Problem
7.5,
ei
(a, b, c)

163

(ii)

cf^

(b

- c)/2 +

(a

- b)fs.

Thus
/o
1^

62

63

= = =

(1,0,0)

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

= 0/1 + 0/2 + 1/3 = 0/1 + 1/2-1/3 = 1/1-1/2 + 0/3


1

and

Q =

1-1
\l -1
0;

'1
(iii)

iWo
1

1^

PQ ^

\l
,1

-1
0^

0/\l -1

(iv)

It

(a, 6, c),

then

[v],

= \b\

and

[v]f

U_
\a-

Thus

bi

/O
(v)

1\

By Problems

7.4(ii)

and

7.5,

[7]^

h
\3
2

3\
.

-4
0/

and

[T\f

-6 -6 -2 6 5 -1/ \
/

Thus

P-^[T\eP

=0 1-11-4
\l -1

/O

l\/o

l\/l
1

1
1

l\

3\

=
0/

0/\3

o/\l

-6 -6 -2 6 5 -1/ \

= m,

7.14.

Prove Theorem 7.4: in a vector space V.

Let

be the transition matrix from a basis

{cj}

Then for any

vGV,
+
04262

P[v]f

[v]e.

Also,
n

[v]f

= P-^[v]e.
P
is

to a basis {h)

Suppose, for i=l,...,n,

A =

ajiei

+
n

aje

matrix whose jth row

is
(oij, a2j, .... aj)

2 ^0^. j=i

Then

the -square
(i)

Also suppose V - kj^ + k2f2+ transpose of a row vector,

+ kj =
[V]f

*-i

Vj-

Then writing a column vector as the


(2)

(fci,

^2, ...,fc)t

Substituting for

/j

in the equation for v,

= =

2v,
n

2^i(ie,) = i(|iA)i

(aijfci

+ a2jk2 +

+ ajkn)ej
is

Accordingly,

[11]^

is

the column vector whose jth entry


aijfci

a2jfc2

ajk

(s)

On
(1)

the other hand, the yth entry of Plv]f is obtained by multiplying the ith by (2). But the product of (1) and (2) is (5); hence P[v]f and [v]^ have the

PMf =

row of P by [vh, i.e. same entries and thus

Me-

Furthermore, multiplying the above by P-i gives P~^[v]e

P-iP[v]f

[v]f.

7.15.

Prove Theorem 7.5: Let P be the transition matrix from a basis {d} to a basis a vector space F. Then, for any linear operator T on V, [T]t = P-i [T]eP.
For any vector

{/i}

in

vGV,

P-HT]^P[v]f

P-^[T],[v],

p-i[T(v)]^

[T(v)]f.

164

MATRICES AND LINEAR OPERATORS


But

[CHAP.

[T]f[v]f

[T{v)]f;

hence P-^[T],P[v]f
l-

[T],[v]f.

Since the

mapping v

[v]f

is

onto K,

P-i[T]^PX = [T]fX

for every

X iC.

Accordingly, P-i[r],P

[7]^.

7.16.

that similarity of matrices is an equivalence relation, that is: (i) A is similar to A; (ii) if A is similar to B, then B is similar to A; (iii) if A is similar to B and B is similar to C then A is similar to C.

Show

(i)

(11)

(iii)

A = I-^AI, A is similar to A. Since A is similar to B there exists an invertible matrix P such that A = P-^BP. Hence B = PAP-i = (P-i)-AP-i and P^^ is invertible. Thus B is similar to A. p-iPP, and since Since A is similar to B there exists an invertible matrix P such that A = = similar to C there exists an invertible matrix Q such that B = Q-^CQ. Hence A B is A is similar to C. p-iBP = P-^(Q-^CQ)P = (QP)->C(QP) and QP is invertible. Thus
The
identity matrix / is invertible

and /

/"i.

Since

TRACE
7.17.

The

trace of a square matrix

elements: tr (A) is similar to B then tr (A)


(1)

= an +
and

^+
=
B=

A=

(oij),

a.

written tr (A), is the sum of its diagonal Show that (i) tr (AB) = tr (BA), (ii) if A
n

tr (B).
(fty).

Suppose

A=

(a)

Then

AP =
n

(ci^)

where
n n

Cj^

aij&jfc-

Thus

tr(AP)

2 i=l
dj^

Cii

2 2 i=l }=1

ttyfeji

On

the other hand,

BA =
=

(d^^)

where

2 i=l

6jiHic-

Thus

tr(PA)

2 =

dji

3=1 i=l

2 2

6ii

>=1

2aa6;i
5

=1

tr(AP)

(ii)

If

is

similar to B, there exists an invertible matrix

such that

A = P-^BP.

Using

(i),

tr(A)

tr(P-iPP)

tr

(PPP-i)

tr (P)

7.18.

Find the trace of the following operator on


T{x, y,
z)

R^:

(aiflj

+ a2y + a^z,

bix

+ h^y + hsz,

Cix

+ Czy + csz)

We

first

must

find a

matrix representation of T. Choosing the usual basis {ej,

h
Cl

62
C2

&3
C3 /

and

tr (T)

tr ([T],)

di

63

C3-

7.19.

Let

V be the space of 2 x 2 matrices over R, and let


V
defined

Af

Let

T be the

linear

operator on

by T{A) = MA.

Find the trace of T.


Choose the usual basis of V:

We

must

first find

a matrix representation of T.

CHAP.

7]

MATRICES AND LINEAR OPERATORS

165

Then

nS,)

= ME, =
(^l

^)(J

=
(^^

= =

IE,

+ + + +

OE,

+
+

3^3

0^4

nE,)

= ME, =

^g ^^(^J J^

^J J)

0^1

IE,

OE,

+ + +

SE,

T(E,)

= ME, = (I ^Y I) ^ (^ I) V4 3 4/Vl 0/
= ME^ =
/I
(^3

::.

2E,

OE,

+
+

4E,

OE,

2\/0 J(^^

T(E,)

ON

J
=

=
/l

/O
(^^

2\

J
2 2 4

=
0\

OE,

2E,

OE,

4E^

Hence
[T]e

10
3
\0

4^

and

tr (T)

10.

MATRIX REPRESENTATIONS OF LINEAR MAPPINGS


7:20.

Let
(i)

if

R3

^ R2

be the linear mapping defined by F{x,

y, z)

{Sx

+ 2y-4z,x-5y + Sz).

(ii)

F in the following bases of R* and R^: {ft = (1, 1, 1), h = (1, 1, 0), fs = (1, 0, 0)}, {9i = (1, 3), g, = (2, 5)} Verify that the action of F is preserved by its matrix representation; that any vGR^ [F]nv]f = [F{v)],
Find the matrix of
By Problem
F(/i)
7.2,
(a, b)

is,

for

(i)

F(l,l,l)

- 5a)ffi + (3a - b)g2. = (1,-1) = -7g,+ 4g, =


(26

Hence
7 4
QQ
1

Q\

19

8/

F(fa)
If

F(1,0,0)
then,
(Sx

(3,1)

=
7.5,

-135^1+
v

ff2

(ii)

v-(x,y,z)
F(v)

by Problem

+ 2y-4z,x-5y + 3z)
and
,

- z)/, + {x - y)/,. Also, = (-13a; - 20y + 26z)gi + (8a; + lly - 15z)g2,


zf,

{y

r,

/-13a;

[i^(-)]a

( 8,

- 2O3/ + 26z \ + 11/- 15. )


/-13a;

T^^

-33 -13 \/
19

SJU-H

==

(8a;

- 20j/ + 26\ + ll.-15. ) =

,, ^, t^(^>^'

7.21.

Let

F:R^-^K^
.
.

be the linear mapping defined by

F{Xi, X2,

. ,

Xn)

{anXi

+ amXn,

021X1

+ aanXn,

OmlXi

+ amnXn)
of

Show
is

that the matrix representation of given by

F relative to the
ai2
CI22

usual bases of
'

K" and

K"

(ttu
0,21

...
...

ttln (l2n

flml

(tm2

dmnl

166

MATRICES AND LINEAR OPERATORS


That
is,

[CHAP.

the rows of [F] are obtained from the coefficients of the


.

Xi in

the components

of F{xi,

.,

x), respectively.
0)
.
. .

^(1,0
F{0,
1,
,

0)

= =

(ail, aai, ...,(ii) (ai2, a22'

/n
,

12 22

in

> "m2)

rpi

"21

a
tr

F{0,0,

.,

1)

(ai, (l2r

>

"rnn)

y^ml

m2

7.22.

Find the matrix representation of each of the following linear mappings relative to
the usual bases of R":
(i)

(ii)

(iii)

F F F

^ R3 R* ^ R2 R3 ^ R*
R2

defined
defined defined

by F{x, by F{x, by F{x,

y)

{Zx

-y,2x + 4y, 5x - ey)


(3a;

y, s, t)

y, z)

+ 2s-U,hx + ly-s- 2t) = {2x + Zy-%z,x + y + z. Ax - 5z, &y)


-4:y
. .

By Problem

7.21,

we need

only look at the coefficients of the unknowns in F{x, y,


3
g\

.).

Thus

(2
6

0/

7.23.

Let

T:R2^R2
{ei

be defined by T{x,y)

the bases

(1, 0), 62

(0, 1)}

= (2x-Zy,x + Ay). Find and {A = (1,3), ^ = (2,5)}

the matrix of

in

of R^ respectively.

(We can view T

as a linear

mapping from one space


Then
^'^ ^^^

into another, each having its

own

basis.)
7.2,

By Problem
r(ei) Tie^)

{a,b)

= =

r(l,0)

r(0,l)

= (26-5o)/i + (3a-6)/2. = (2,1) = -8/1+ 5/2 = (-3,4) ^ 23/1-13/2

f '

^ /-8
\
5

23

"13

7.24.

Let
fined
(i)

A =
Show
of

/
(

3
.

\1 -4

Recall that
is

determines a linear mapping

F-.W^B?

de-

by F(v)

7/ Av where v

written as a column vector.

that the matrix representation of is the matrix A itself: [F] = A.

relative to the usual basis of R^

and
R*.

(ii)

Find the matrix representation of


{/i

relative to the following bases of R^


{^1

and

(1, 1, 1),

U=
=

(1, 1, 0),

h = (1, 0, 0)},
^)

=
=

(1, 3),

g^

(2,

5)}

(i)

F(1,0,0)

(1 _4

= (1)

261

162

F(0,1,0)

(j _J

^)

(_J)

- 561-462

from which

W\ =
7.2,

2
-,

3\

_.

-A-

(Compare with Problem

7.7.)

(ii)

By Problem

(a, 6)

(26

- 5a)flri +

(Za-V^g^.

Then

CHAP.

7]

MATRICES AND LINEAR OPERATORS

167

F(h)

(I -4 -?)(;

--

-12flri

:)

8^2

5
F{f2)

il -4

^( 4)
I)

==

-41flri

24fir2

F(h)

il -4

'-

-SfiTi

5fr2

and

[F]

-12 -41 ~^^


8

24

7.25.

Prove Theorem
basis of
C7 is

7.12:

Let F:y-*J] be linear.

Then there

exists a basis of

Y
(

and a

where

such that the matrix representation A of i^ has the the r-square identity matrix and r is the rank ofF.

form

A =

\
the image of F.

Suppose dim

V-

m,

and dim
r\

are given that rank F = be a basis of the kernel of

JJ = n. Let be the kernel of F and hence the dimension of the kernel of F is and extend this to a basis of V:

m r.

We

Let {wi,

.,)_ J

Set

Ml that {mi,
. . . ,

F('Ui),
J7',

M2

^(va),

...,

Mr

F(t)^)

We note
of
J7.

mJ

is

a basis of

the image of F. Extend this to a basis


.
.

{%,
Observe that
F(t;i)

.,

Mr+l,

.,

M}

= = =

Ml

1mi
0*<i

0^2
1^2

^(va)

M2

+ + +

+ + + +

+ +
+

Om,
Om^

+ + +

Om^+i Om^+i

+ +

+ + + +
+

0m
0m

F(i;,)

M,

F(Wi)

=0
=

= =

Omi Omi

0^2
OM2

1m^

OMy+ 1

0m
Om

F(w^_r)

= 0% +

OM2

OWr

Om^+i

0m

Thus the matrix of

F in the

above bases has the required form.

Supplementary Problems
MATRIX REPRESENTATIONS OF LINEAR OPERATORS
7.26.

Find the matrix of each of the following linear operators


{ei

on R2 with respect to the usual basis

(1, 0), 62

(0, 1)}:

(i)

r(, y)

(2x

-3y,x + y),

(ii)

T{x, y)

(5x

+ y,Zx- 2y).
{/i

7.27.

Find the matrix of each operator


/2

(2, 3)}.

In each case, verify that

in the preceding problem with respect to the basis {T(v)]f for any v e R2. {T\f{v\f

(1, 2),

7.28.

Find the matrix of each operator

in

Problem 7.26

in the basis

{g^

(1, 3),

g^

(1, 4)}.

168

MATRICES AND LINEAR OPERATORS


Find the matrix representation of each of the following linear operators usual basis:
(i)

[CHAP. 7

7.29.

T on R3

relative to the

T(x,y,z)
T{x, y, z)
T(,x,

(ii)

(iii)

y,z)

= = =

(x,y,0)

-7y - Az, 3x + y + 4z, (z,y + z, x + y + z)


(2x
i.e.

6a;

- 83/ + z)

7.30.

Let D be the differential operator, vector space V of functions / E


:

D(f)

dfldt.

Each of the following

sets is
(i)

a basis of a
{e', e^t, te^'^},

^ R.

Find the matrix of


sin St, cos 3t}.

in

each basis:

(ii)

{sin

t,

cos

t},

(iii) {e5, te^*, t^e^t},

(iv) {1, t,

7.31.

Consider the complex field C as a vector space over the real field E. Let operator on C, i.e. T(z) = z. Find the matrix of T in each basis: (i) {1, i),

T
(ii)

be the conjugation
{1

+ i, 1 + 2i}.

7.32.

Let

V be
T{A)

the vector space of 2

2 matrices over

and

let

Af

Find the matrix of each


(i)

of the following linear operators


(ii)

T on V

in the usual basis (see

Problem

7.19) of V:

T{A)

= MA,

= AM,

(iii)

T(A)

=^MA- AM.

7.33.

Let ly and Oy denote the identity and zero operators, respectively, on a vector space V. Show that, for any basis {ej of V, (i) [1^]^ = I, the identity matrix, (ii) [Oy]^ = 0, the zero matrix.

CHANGE OF
7.34.
(i)

BASIS, SIMILAR

MATRICES
{e^

Consider the following bases of R^:

(1, 0), eg

(0, 1)}

and

{/i

(1, 2),

^=

(2, 3)}.

Find the transition matrices Verify Q = P-i.

and

Q from
v

{gj}

to

{/J and from

{/j} to {ej, respectively.

(ii) (iii)

Show Show

that that

[v]^

[T]f

= -

P[v]f

for any vector

fP.

P-'^[T]^P for each operator

in

Problem

7.26.

7.35.

Repeat Problem 7.34 for the bases

{/i

(1,2), /a

(2,3)}

and

{g^

(1,3),

g^

(1.4)}.

7.36.

Suppose {e^, e^} is a basis of V and T :V -^V is the linear operator for which T^e^) = Se^ 2e2 and T{e2) = ej + 4e2. Suppose {/i, /a} is the basis of V for which /i = ei + e^ and /z = 2ei + 3e2. Find the matrix of T in the basis {/i, /j}.
Consider the bases B {1, i} and B' = {1 + i, 1 + 2i} of the complex field C over the real field R. (i) Find the transition matrices P and Q from B to B' and from B' to B, respectively. Verify that Q = P-\ (ii) Show that [T]^, = P-'^[T]bP for the conjugation operator T in Problem 7.31.

7.37.

7.38.

to {/j}

Suppose {ej, {/J and {flrj} are bases of V, and that P and Q are the transition matrices from {ej and from {/J to {ffj, respectively. Show that PQ is the transition matrix from {ej to {fTj}.

7.39.

Let

be a 2 by 2 matrix such that only

is

similar to

itself.

Show

that

has the form

A =
Generalize to

wXw

matrices.

:)
More
generally,

7.40.

Show that all the matrices similar to an invertible matrix are invertible. similar matrices have the same rank.

show that

MATRIX REPRESENTATIONS OF LINEAR MAPPINGS


7.41.

Find the matrix representation of the linear mappings relative to the usual bases for R":
(i)

(ii)

F F
i^

R3
Ri!

-*

R2 defined by F{x,
defined by
defined

^ R*

(iii)

F:R*-*B,
:

by by

(iv)

R ^ R2

defined

+ Sy- 2z) F{x, y) = (3a; + 4j/, 5x -2y,x + ly, ix) F(x, y, s,t) = 2x + 3y-7s-t F(x) = (3x, 5x)
y, z)

(2x

- 4j/ + 9s,

5x

CHAP.

7]

MATRICES AND LINEAR OPERATORS


R3

169

7.42.

Let
(i)

i<'

^ R2

be the linear mapping defined by F{x,

y, z)

(2x

+ y z, =

Sx

2y + iz). =
(1,

Find the matrix of


{/i

in the following bases of RS and R^:

(l,l,l), /2

(1,1,0), /a

(1,0,0)}

and

{jti

(1, 3),

fir^

4)}

(ii)

Verify that, for any vector v

R3,

[F]^ [v]f

[F{v)]g.

7.43.

Let {ej and {/J be bases of V, and let ly be the identity mapping on V. Show that the matrix of ly in the bases {ej and {/;} is the inverse of the transition matrix P from {e^} to {f^}^, that is,

7.44.

Prove Theorem Prove Theorem Prove Theorem Prove Theorem

7.7,

page 155.
{Hint.

{Hint.

See Problem
7.10.)

7.9,

page

161.)

7.45.

7.8.

See Problem
See Problem

7.46.

7.9.

{Hint.
{Hint.

7.11.)

7.47.

7.10.

See Problem

7.15.)

MISCELLANEOUS PROBLEMS
7.48.

Let

be a linear operator on

and

let

W
T

be a subspace of

invariant under T, that


I

is,

m. T{W) C W. Suppose dim where A is an m X to submatrix.


7.49.

Show

that

has a matrix representation of the form

/A B\
.
)

Let

y =

1/

W, and

let

and

W each be invariant under a linear operator


respectively.

T V
:

->

V.

Suppose

dim U A and
7.50.

= m and dim V = n. Show that T B are mXm and nX n submatrices,


G

has a matrix representation of the form

/A
(
''

ON

where

Recall that two linear operators F and operator T on V such that G = T-^FT.
(i)

on

are said to be similar

if

there exists an invertible

Show that linear operators F and G are similar if and only matrix representations [F]^ and [G]g are similar matrices. Show
that
if

if,

for any basis {ej of V, the

(ii)

an operator

F is
over

diagonalizable, then

any similar operator


if

is also

diagonalizable.

7.51.

Two
(i)

mX n matrices A
Show Show

and

are said to be equivalent

there exists an m-square invertible

matrix

Q and an n-square invertible matrix P such that B QAP. Show that equivalence of matrices is an equivalence relation.
that
if

(ii)

and only
(iii)

A and B A and B

can be matrix representations of the same linear operator


are equivalent.
j '

F :V -> U
is

if

identity matrix

that every matrix A is equivalent to a matrix of the form V and r = rank A.


(

where /

the r-square

7.52.

Two
(iii)

algebras
/
:

mapping

A and B over a field K are said to be isomorphic (as algebras) if there exists a bijective A -* B such that for u,v S A and k G K, f{u + v) = f(u) + f(v), (ii) /(few) = fe/(w),
(i)

f{uv) f{u)f{v). (That is, / preserves the three operations of an algebra: vector addition, scalar onto multiplication, and vector multiplication.) The mapping / is then called an isomorphism of B. Show that the relation of algebra isomorphism is an equivalence relation.

7.53.

Let cA be the algebra of *i-square matrices over K, and let P be an invertible matrix in cA. that the map A \-^ P~^AP, where A G c/f, is an algebra isomorphism of a4 onto itself.

Show

170

MATRICES AND LINEAR OPERATORS

[CHAP.

Answers
/2 -3
7.26.
(i)

to

Supplementary Problems

6
(ii)

\1

-2

7.27.

Here

(a, 6)

(26

- 3a)/i +

(2a

- b)!^.

^.,

,18

25 \

^..^

/-23 -39
(

-11 -15 j

^"^

15

26

7.28.

Here

(a, 6)

(4a

- h)gi +

(6

- Za)g2.

-32 -45

35
(")

41

(25

35/

V-27 -32

7.29.

(i)

10 10
,0

'2
3

-7 -4I

14
1
(

(iii)

0,

6-8

1
7.30.
(i)

0'
2
1
(ii)

5
(iii)

101
5 2 5,
(iv)

'0

o\

0-3
iO
3

,0

2,

,0

0/

1
7.31.
(i)

-1

(ii)

-3 4 -2 -3
^c

h 6

-6
7.32.
(i)

a-d
da

(ii)

(iii)

7.34.

P =

-3

Q =
3 5

7.35.

P =

-1 -2

Q =

-1 -3

8
7.36.

11

-2 -1
2
-1

7.37.

P =

-1
1

-4
7.41.
(i)

-2

(iii)

(2,3,-7,-1)

(iv)

3
7.42.
(i)

11

-1 -8

chapter 8

Determinants
INTRODUCTION
To every square matrix A over a field determinant of A; it is usually denoted by
det(A)

there

is

assigned a specific scalar called the

or

|A|

This determinant function was first discovered in the investigation of systems of linear We shall see in the succeeding chapters that the determinant is an indispensable tool in investigating and obtaining properties of a linear operator.
equations.

We comment that the definition


in the case

of the determinant and most of its properties where the entries of a matrix come from a ring (see Appendix B).

also apply

We

shall begin the chapter

with a discussion of permutations, which

is

necessary for

the definition of the determinant.

PERMUTATIONS A one-to-one mapping


denote the permutation
1
.

<7

of the set {1,2,

.,} onto itself is called a permutation.

We

<r

by
...
.
.

2
.

n\
.

h H

or

<T

3i32

3n,

where
. . .

3i

.... =

cr(i)

JnJ

Observe that since o- is one-to-one and onto, the sequence /i J2 jn is simply a rearrange. We remark that the number of such permutations is n !, ment of the numbers 1, 2, and that the set of them is usually denoted by S. We also remark that if <7 /S, then the inverse mapping cr"^ G S; and if a.rGSn, then the composition mapping o-orGSn. In particular, the identity mapping
. . .

belongs to Sn.
Example Example

(In fact,
8.1
:

12...n.)
!

There are 2

2 permutations in Sg: 12

and

21.

8.2:

There are 3!

= 3'2'1 =

6 permutations in S3: 123, 132, 213, 231, 312, 321.


. .

Consider an arbitrary permutation a in Sn. a = ji jz jn. according as to whether there is an even or odd number of pairs
.

We
(i,

k)

say a is even or odd for which


(*)

i>

k
a,

but

precedes
<t,

A;

in

er

We then

define the sign or parity of

written sgn
r

by
is IS

sgna

if

<7

even

J.

[1
171

if

odd

172

DETERMINANTS
Example
8.3:

[CHAP. 8

Consider the permutation a


3
3,

35142

in S5.
1;

and
5

5 precede

and are greater than

hence
2;

(3, 1)

and

(5, 1)

satisfy
(4, 2)

(*).

and 4 precede and are greater than

hence

(3, 2), (5, 2)


(*).
<r

and

satisfy

(*).

5 precedes

and

is

greater than

4;
(*),

hence
a
is

(5, 4) satisfies

Since exactly six pairs satisfy

even and sgn

1.

Example Example

8.4:

The

identity permutation

= 12
odd.

is

even since no pair can satisfy

(*).

8.5

In S2, 12
In S3,

is

even, and 21
123, 231

is

and 312 are even, and 132, 213 and 321 are odd.
i

Example

8.6:

Let t he the permutation which interchanges two numbers other numbers fixed:
r{i)

and
j

and leaves the

j,

r(j)
i

=
j,

i,

T(fc)

k,

k^

i,

We

call T

a transposition.
T

If

<

then
...

= 12 ...{i-l)}{i+l)
l

ij-l)i{j+l) ...n

There are

2{j

i l) +

pairs satisfying
(h^)!
i^t
i),

(*):

(j,i),

where x

i+1,

.,jl

Thus the transposition

t is odd.

DETERMINANT Let A be
(Oij)

an -square matrix over a


jdll
A

field

K:
O-ln^
0'2n

(tl2 0.22

...
.

0-21

\ai

CLn2

Consider a product of n elements of A such that one and only one element comes from each row and one and only one element comes from each column. Such a product can be written in the form
ftlil 0.212

^^'n

where the factors come from successive rows and so the first subscripts are in the .,n. Now since the factors come from different columns, the sequence natural order 1,2, = ji 32 Conversely, each permutajn in Sn. of second subscripts form a permutation matrix A contains n\ such tion in Sn determines a product of the above form. Thus the
that
is,
. .

<t

products.
Definition:

The determinant of the w-square matrix


is

A=
. .

(Odi),

the following

sum which
\A\

is

summed over
o-)a'UiO,2j^

all

denoted by det(A) or permutations o- = ji jz


.

|A|,
. .

in Sn.

2/ (sgn

aj

That

is,

2
of the w-square matrix
ail
0-21

(sgn

C7)aicr(l) tt2(T(2)

Ona-in)

The determinant denoted by

is

said to be of order
ttln
0.2n

n and

is

frequently

ai2
0.22

Onl

0,n2

CHAP.

8]

DETERMINANTS

173

We

emphasize that a square array of scalars enclosed by straight lines is not a matrix but rather the scalar that the determinant assigns to the matrix formed by the array of scalars.
Example
8.7:

The determinant of a 1 X 1 matrix A (We note that the one permutation in Sj


In
1S2,

is

(an)
even.)

is

the scalar

on

itself:

\A\

On.

Example

8.8:

the permutation 12

is

even and the permutation 21


12
''ll'*22
'*12'*21

is

odd.

Hence

"H
0-1^

a^i

Thus

4 -5 -1 -2

4(--2)

(-5

-13

and

a
e

ad

be.

Example

8.9:

In 1S3, the permutations 123, 231 and 312 are even, and the permutations 321, 213 and Hence 132 are odd.

On
0-21

ai2
'*22

ai3
"23 <*33

<*ll'*22'*33

<*i2(l23a3l

di^a^ia^i
"" <*ll''23*32

131

*32

<Z'X3a220'3l

ttl2<*2l'>33

This

may

be written
Oll(<l22'*33

as:

~ ''23'*32)
a22 23
'''33

*12("21<'33

~ 23*3l) +
"^23

'*13('''2l'''32

~ <*22'*3l)

21

12
31

O32

021

022 032

13
O3I

"33

which

2X2

is a linear combination of three determinants of order two whose coefficients (with alternating signs) form the first row of the given matrix. Note that each matrix can be obtained by deleting, in the original matrix, the row and column

containing

its coefficient:

ttll

12

Oi3
"23

<11

I2
'h.i

Oi3
2:i '*33

ii
.l1

"12

"21

13
a31
Itj.,

<*33

031

"gi

033

4 7
1

Example

8.10:

(i)

6
9

7
1

2
9

7
1

3 8

+ +

6 9

4
8

2(6-63)
2
(ii)

3(5-56)

4(45-48)

27

-4
2
5

-4 -1

4
2

v 5

2 3
1

-1

(-4)
1

-4 -1

2(-20

+ 2)

As n increases, the number of terms in the determinant becomes astronomical. Accordwe use indirect methods to evaluate determinants rather than its definition. In fact we prove a number of properties about determinants which will permit us to shorten the computation considerably. In particular, we show that a determinant of order n is equal to a linear combination of determinants of order m. 1 as in case n = 3 above.
ingly,

PROPERTIES OF DETERMINANTS We now list basic properties of the


Theorem
8.1:

determinant.

The determinant

of a matrix

and

its

transpose A* are equal:

\A\

\A*\.

174

DETERMINANTS
By
this theorem,

[CHAP. 8

rows of

will

any theorem about the determinant of a matrix A which concerns the have an analogous theorem concerning the columns of A.

The next theorem gives certain cases for which the determinant can be obtained immediately.

Theorem

8.2:

Let
(i)

A
If
If

be a square matrix.

(ii)

(iii)

A has a row (column) of zeros, then \A\ = 0. A has two identical rows (columns), then |A| = 0. If A is triangular, i.e. A has zeros above or below
\A\

the diagonal, then


|/|

is

product of diagonal elements. the identity matrix.

Thus
is

in particular,

where

The next theorem shows how the determinant of a matrix


operations.

affected

by the "elementary"

Theorem

8.3:

Let
(i)

B be the matrix obtained from a matrix A by multiplying a row (column) of A by a scalar

fc;

then

|B|

fe|A|.

(ii)

(iii)

interchanging two rows (columns) of |A|; then |Z?j = |A|. adding a multiiJle of a row (column) of A to another; then

jB]

|A|.

We now state two


Theorem
8.4:

of the most important and useful theorems on determinants.

Let
(i)

(ii)

A be any n-square matrix. Then the following are equivalent: A is invertible, i.e. A has an inverse A~^. A is nonsingular, i.e. AX - has only the zero solution, or rank A = n, or the rows (columns) of A are linearly independent, |A| = 0. (iii) The determinant of A is not zero:
function.
is
is, the determinant of product of their deterequal to the

Theorem

8.5:

The determinant is a multiplicative a product of two matrices A and B


minants: \A B\

That

\A\ \B\

above two theorems using the theory of elementary matrices page 56) and the following lemma.

We

shall prove the

(see

Lemma

8.6:

Let

be an elementary matrix.

Then, for any matrix A,

\E A\

\E\\A\.

We

comment that one can

also prove the preceding

two theorems

directly without

resorting to the theory of elementary matrices.

MINORS AND COFACTORS


Consider an w-square matrix A = (ay). Let M denote the (w- l)-square submatrix of A obtained by deleting its ith row and .7th column. The determinant \Mij\ is called the mmor "signed" of the element ay of A, and we define the cofactor of Oni, denoted by A, to be the ^^""^=

A = i-iy^^m

Note that the "signs" (-1)*+^' accompanying the minors form a chessboard pattern with +'s on the main diagonal:

; : ;
+

::-\

We emphasize that My

denotes a matrix whereas

Ay denotes

a scalar.

CHAP.

8]

DETERMINANTS
2 3

175

4 2
3

Example

8.11:

Let

A =

Then

M23

=
2

5 8

6
9

7 8
1

and
9

=
The following theorem
applies

3 9

(-1)2
8

-(18-24)

Theorem

8.7

The determinant of the matrix A = (Odj) is equal to the sum of the products obtained by multiplying the elements of any row (column) by their respective cofactors:
\A\

OiiAii

+ +

ai2Ai2

+ +

ttinAin

j=i

o^a-^a

and

CLljAij

Ct2i-'4.2j

anjAnj

OijAij

The above formulas, called the Laplace expansions of the determinant of A by the ith. row and the yth column respectively, offer a method of simplifying the computation of \A\. That is, by adding a multiple of a row (column) to another row (column) we can reduce A to a matrix containing a row or column with one entry 1 and the others 0. Expanding by this row or column reduces the computation of \A\ to the computation of a determinant of
order one less than that of
\A\.
5

4
3

2
1

l\

Example

8.12:

Compute the determinant of

A =

'

-5 -7 -3 1 -2 -1
Perform the following

Note that a 1 appears in the second row, third column, operations on A, where fij denotes the ith row:
(i)

add -2R2 to
8.3(iii),

jRi,

(ii)

add 3^2 to Ra,

(iii)

add

li?2 to R^.

By Theorem
tions; that
is.

the value of the determinant does not change by these opera5

4
3

-2
3 2
1 1

\A\

-5

1-2 -7-3 9
4

2
1

-2
3

1-2-1
Now
Thus
1
if

2
0.

we expand by

the third column,

we may
5

neglect all terms which contain

_2
3
1

\A\

2
(-1)2
1

-2
3

-123
3
1

1-2

2
1

12
= }
38

2
1

3 2

(-2)
3

2
1

5
3

CLASSICAL ADJOINT
Consider an n-square matrix

A=

(an)
'

over a
ai2
ffl22

field

K:
ttin
tt2re

an
0,21

A =
1

fflnl

ffln2

176

DETERMINANTS
Oij

[CHAP. 8

The transpose of the matrix of cofactors of the elements


the classical adjoint of A:

of A, denoted by adj A,

is

called

'An
adj

A21

...

Ai

'^^^

^^^

^"^

^nn

We say "classical adjoint"


in

instead of simply "adjoint" because the term adjoint will be used Chapter 13 for an entirely different concept.

Example

8.13:

Let

The cofactors

of the nine elements of

are

An = +
^21

-4 -1
3

2 5

=
= =

-18,

A12

2
5

2,

Ai3

= +
1

-4
-l|
3

= =

-4
5

-1
3

-11,

A22

- +

2
1

-4
5

= =

14,

A23

= -

2
1

-1
3

A31

= + -4

-4
2

-10,

A32

-4
2

-4,

A33

= +

-4

= -8

We form the transpose of the above matrix of cofactors to obtain the classical adjoint
of A:
I

-18 -11 -10


2

adj

=
\

14-4 5-8/
=

Theorem

8.8:

For any square matrix A,

A -(adj A) =
where
/ is the identity matrix.

(adj
if

A) -A

\A\I

Thus,

\A\' 0,

Observe that the above theorem gives us an important method of obtaining the inverse
of a given matrix.

Example

8.14:

Consider the matrix

of the preceding example for which

\A\

46.

We
/l

have
0^
1

/2

A (adj A) =

-4 \l -1

-4\ /-18 -11 -10\ = 2 14 -4 2 4 5 -8/ 5/\


8.8,

/-46

0\

0-46
\

= -46
-46/
\0

1^

= -46/ =

|A|/

We

also have,

by Theorem

-18/-46 -11/-46 -10/-46\

9/23

11/46 5/23 \

A-i = r4T(adjA)

2/-46
4/-46

14/-46

5/-46

-4/-46 -8/-46/

-1/23 -7/23 2/23

\-2/23 -5/46 4/23/

APPLICATIONS TO LINEAR EQUATIONS


Consider a system of

linear equations in

n unknowns:

anX\
a2\X\

+ ai2a;2 + + a22a;2 +

+ aia; = + ainXn
"T

bi

&2

anliCi

an%X2

annXn

bn

CHAP.

8]

DETERMINANTS

177

Let A denote the determinant of the matrix A (oij) of coefficients: A = \A\. Also, let As denote the determinant of the matrix obtained by replacing the ith column of A by the column of constant terms. The fundamental relationship between determinants and the solution of the above system follows.

Theorem

8.9:

the unique solution

The above system has a unique is given by

solution if

and only

if

A
An

?^ 0.

In this case

_
The above theorem

Al

is known as "Cramer's rule" for solving systems of linear equations. emphasize that the theorem only refers to a system with the same number of equations as unknowns, and that it only gives the solution when A ^ 0. In fact, if A = the theorem does not tell whether or not the system has a solution. However, in the case of a homogeneous system we have the following useful result.

We

Theorem

8.10:

The homogeneous system Ax A = |A| = 0.


\2x
8.15:

has a nonzero solution

if

and only

if

Example

Solve, using determinants:

<
3a;

Zy
52/

= =

1
1

First compute the determinant

of the matrix of coefficients:

A
Since

2
3

-3
5

10

19

^^ 0,

the system has a unique solution.

We
2
3
is

also

have
-19

A.

7
1

-3
5

7
1

38,

Accordingly, the unique solution of the system


*'

T"i9

^x

38

2,

^y

-19

T=l9-

= -i

We remark that the preceding theorem is of interest more for theoretical and historical reasons than for practical reasons. The previous method of solving systems of linear equations, i.e. by reducing a system to echelon form, is usually much more efficient than by using determinants.

DETERMINANT OF A LINEAR OPERATOR


Using the multiplicative property of the determinant (Theorem
8.5),
\B\.

we

obtain

Theorem

8.11:

Suppose

and

are similar matrices.

Then

|A|

Now suppose T is an arbitrary linear operator on a vector space V. determinant of T, written det (T), by
det(r)

We

define the

|[r]e|
is in-

where

By the above theorem this definition [T]e is the matrix of T in a basis {et}. dependent of the particular basis that is chosen.
The next theorem follows from the analogous theorems on matrices.

Theorem

8.12:

Let
(i)

T and
det (S

iS

be linear operators on a vector space V.

Then

T)

det {S) det {T),

(ii)

is

invertible if

and only

if

det

(7)^0.

178

DETERMINANTS

[CHAP. 8

We also remark that


det(r)-i
if

det (Iv)

where Iv

is

the identity mapping, and that det (T~^)

is

invertible.

Example

8.16:

Let

be the linear operator on R3 defined by


T(x, y, z)

(2x

4y + z, X 2y + 3z,
'2

5x

+ y z)
l\
3
.

The matrix

of

in the usual basis of

R3

is

[T]

-4 -2

Then

,5

1-1/
+
1(1

det(r)

2-4 1 1-2 3 1-1 5

2(2

- 3) +

4(-l

15)

+ 10) = -55

MULTILINEARITY AND DETERMINANTS


Let cA denote the set of
n-tuple consisting of
its
all

n-square matrices
Ai, A2,
.
.

A
.

over a

field

K.

We may

view

as an

row vectors

.,

A:
.

A =
Hence cA may be viewed as the

{Ai, A2,

.,

An)

set of n-tuples of w-tuples in K:

The following
Definition:

definitions apply.

A
(i)

function

D.cA

-^
is:

is

said to be multilinear if

it is

linear in each of the

components; that
if

row Ai = B + C, then
D{A)

= D(...,B + C,...) = =

D{...,B,

...)

D(...,C,

...);

(ii)

if

row Ai = kB where k G K, then


D{A)
D(...,kB, ...)

= kD{...,B,

...).

We also say n-linear for multilinear if there are n components.


Definition:

A function D-.cA^K
two
identical rows:

is

said to be alternating

if

D{A) =
A,

whenever

has

D{Ai, A2,

.,

An)

whenever

Aj, i^ j

We

have the following basic


8.13:

result;

here / denotes the identity matrix.


that:

Theorem

There exists a unique function D-.oA -*K such


(i)

is

multilinear,

(ii)

is

alternating,

(iii)

D{I)

1.
is,

This function D is none other than the determinant function; that any matrix A^cA, D(A) = jA|.

for

CHAP.

8]

DETERMINANTS

179

Solved Problems

COMPUTATION OF DETERMINANTS
8.1.

Evaluate the determinant of each matrix:


3
(i)

a
(i)

a
a

(ii)

^4

b
-62.

-2
5

= 3-5-

(-2) -4

a
23.
(ii)

a
a

= (ab)(a+b) a'a

8.2.

Determine those values of k for which


k
4
fe

k
4

0.

2k
Hence k

2k

2A;2

4fc

0,

or

2fc(fe

- 2) =

0.

Q;

and k

2.

That

is, if

fe

or

2,

the determinant

is zero.

8.3.

Compute the determinant


'1
(i)

of each matrix: l\
/
,

3\ 3
(ii)

/2

2 3 2

1^

0\
,
I

[4-2
^2
1

(4

2 -3

(iii)

-3

(iv)
I

2 -4

.
|

5 -ly
2
3 3

\-l -3

(i)

4
2

-2
5

-2
1

-1

-1

4 2
2

-1

4 3 2

-2
5

1(2-15)
2
(ii)

- 2(-4-6) + 4
5

3(20
|4
ll

+ 4) =
2
3

79

4
5 2

2
3

-3
1

2
3

-3
1

-3
1

24

(iii)

2-3
5

2(10-9)

l(-9

+ 2) = -5

-1 -3

(iv)

10 2-4 13
3

1(6

+ 4) =

10

8.4.

Consider the 3-square matrix

\a2
\a3

&2 bs

C2
cs

Show

that the diagrams below can

be used to obtain the determinant of A:

Form the product of each of the three numbers joined by an arrow in the diagram on the and precede each product by a plus sign as follows:

left,

180

DETERMINANTS
Now form
right,

[CHAP. 8

the product of each of the three numbers joined by an arrow and precede each product by a minus sign as follows:

in the

diagram on the

Then the determinant of


!
\A\

asftgCi

bgC^ai

Cgagfei

is

precisely the

sum

of the above

two expressions:

61

Ci

a2

62
&3

"2

"3
\A\

The above method of computing

does not hold for determinants of order greater than

3.

8.5.

Evaluate the determinant of each matrix:


/2
(i)

-l\
2
(ii)

/a

c>
(iii)

2 -

10-2
\-2
3
3/
0:

A
(i)

-3

7/

Expand the determinant by


2 3

the second column, neglecting terms containing a

0-1
2 7

-1
3(4

-(-3)
3

3)

21

4-3
(ii)

Use the method


a
c b b

of the preceding problem:


c

a
c

a^

63

c*

abc

abc

abc

63

c3

3a6c

(iii)

Add

twice the first column to the third column, and then expand by the second row:
3
1

2-4 + 2(3)
-2 +
3 3
2(1)

10
-2

-2

+ 2(-2)

-1

-1

fi
8.6.

-1

Evaluate the determinant of

A =
6

First multiply the first

row by

and the second row by

4.

Then
3
1

3-6-2
6-4|A|

24|A|

2-4 1-4 1
6

-2 - (3) + 4(3) -4 -(3) -4 + 4(1) 1~(1)

-6 + 4(3)
2

6-5
14

-7

10
=
7/6.

= +

28,

and

\A\

28/24

14

8.7.

Evaluate the determinant of

A
A

Note that a
(where
i2j

denotes the ith row):

appears in the third row, first column. Apply the following operations on Thus (iii) add IB3 to R^. (i) add 2R3 to iSj, (ii) add 2R3 to i?2,

CHAP.

8]

DETERMINANTS
2
\A\

181

2
1 1

-3
3

-3 -2 2 -5 -2 2
4
3

-1
3
1

1 -6 -2 -1 -2 2

-1

= +

1 -6 -2 -1

-6

-3

-3
1

-1

-6 + 6(1) 1 +1 3-2 -2 -1 + 6(-2)

1 1

-2 -13
2

-13
17

-3

+2

+ 6(2)

-1

= -4

17

8.8.

Evaluate the determinant of

A
row
to the

First reduce A to a matrix which has 1 as an entry, such as adding twice the second row, and then proceed as in the preceding problem.
3
|A|

first

-2 -5
2 4 8 7

3-2
2

-5
8

5
2
2
3
1

-5 -3
8

-5 + 2(3) -2

-3 -5 -2 -5 -2 -2
4 7

2-3-57-3
+ 2(-2)
4

+ 2(-5)

-5 + 2(4)
8

4
3

3
1

-2 + 2(3) -2 + 2(1)
4

-5 + 2(3) -2 + 2(1)
7

4-3(3) 3-3(1)

2
2 3
1

-3
8

-2
2

+ 2(-2)

+ 2(-2) -3-3(-2)
8-3(2)
1

-3 -5
4
1

-3 + 2(2)
1

-5 + 2(2)
4

-5
4 3 3
1

-5
3

-5-(l)
3

=
1

2 2 4
1

- (3)

-1

-1

-1

2-(-l)

1-6
4
3
1

-6
-3(12

1-13
/t
8.9.

+ 6) = -54

+
5 6

-1

1 1
t

Evaluate the determinant of

A =

t-B
-6 + +

+ 4/
1

Add
to obtain

the second column to the first column, and then add the third column to the second column
t

2 2

\A\

t-2 t-2

1
t

4
to get

Now

factor

+2

from the

first

column and

t 2

from the second column

|A|

(t

+ 2)(t-2)

10 11
1

1 1

Finally subtract the first column from the third column to obtain

\A\

(<

+ 2)(t-2)

10 11
1
t

= +
4

(t+2)(t-2)(t +

4)

182

DETERMINANTS

[CHAP. 8

COFACTORS
8.10.

Find the cofactor of the 7 in the matrix

2 5

-3

-4

7
6

-2

2
rz

(-1)2
4 3

^
-3
2

4
3

-3
-2
2

4 7

-3
10

/ \

4 7

-3
61 10

-2

The exponent 2 +

3 comes

from the fact that 7 appears in the second row, third column.

/I
8.11.

2 3 5

3\

Consider the matrix

A =
(iv)

(i)

Compute

|A|

(ii)

Find adj A.

(iii)

Verify

\l

7/

A
(i)

(adj

A)

=
1

\A\ I.

Find
2
1

A-\
+
2

\A\

3 5

4 7

4 7

3
sl

3
1

= 1-20 +

21

+
(ii)

3 5

4
7
3

2
1 1 1

4
7 3 7
3

adj

A
+

2
5

7
3

2 3

+
Observe that the "signs" in the

That

is

adj

is

the transpose of the matrix of cof actors.

matrix of cofactors form the chessboard pattern

- +\ + \+ - +/

(iii)

A '(adj A) =

|A|/

(iv)

A-i

|]4i(adjA)

8.12.

Consider an arbitrary 2 by 2 matrix


(i)

d
A.

Find adj A.
adj

(ii)

Show
^1^1 j

that adj (adj A)

=
-

(X)

A =
(^_|^|

-b
a

\^_^

;
/ +la|

(^_,

(ii)

adj (adj A)

d
adj

-l-ciy
+|di;

^ /a
v''

-e

\^_|_6|

= CD-

CHAP.

8]

DETERMINANTS

183

DETERMINANTS AND SYSTEMS OF LINEAR EQUATIONS


8.13.

Solve for x and y, using determinants:


(i)

+ y = 3x 5y
2a;

__

(ii)

4
-13,
A:,

2hy Sax 5by


ax

c
,

where ab

0.

2c

2 3

-5

7 4

-5

-39, Ay

2
3

-4

-13.

Then x

A^/A

3,

A/A
a 3a

1.

(ii)

A =

-26 -56

a6,

Ax

2c

-26 -56

-be,

Aj,

a.

= ac.

Then

a;

A^-M

3a

2c

c/a, y

Aj,/A

c/b

Sy
8.14.

Solve using determinants:

3a;

3z

+ 2x = + 22; -1 =
z

S
a;

+1 - 5y

2i/ in columns:

First arrange the system in standard

form with the unknowns appearing


2z 3z

2x 3x

+ 3y + 5y + -2y -

= 1 - 8 = -1

Compute the determinant A


2

of the matrix

of coefficients:

3-1
5

A
Since A
the

2(-15

+ 4) - 3(-9-2) - l(-6-5)

22

-2
t^ 0,

-3

unknown

in the

the system has a unique solution. To obtain A^., A,, and A^, replace the coefficients of matrix A by the column of constants. Thus
2

13-1
A,

1-1
8
2

3 5

=
-

=
Aj,/A

66,

= =
A^/A

3
1

-22,

A,

3
1

44

-1 -2 -3
and X
Aj./A

-1 -3
2.

-2 -1

3,

2/

-1, z

PROOF OF THEOREMS
8.15.

Prove Theorem
Suppose

8.1:
(tty).

|A*|

|A|.

A=

Then A*
l^'l

= = =

(6jj)

where 6y
(Sgn

a^;.

Hence
. .
.

aes

2 2

a) 6io.(i) 62,^(2)

6o-(n)

(sgn ff)a(i).ia<r(2),2 o-(n),n

Let T

<r~i.

By Problem

8.36,

sgn r

sgn

a,

and

Hence
However, as a runs through
S.

|A|

<Tes
all

(sgn

t)

ai^d) a2T(2)
t

nT(n)

the elements of S,

<r~i

also runs through all the elements of

Thus

\At\

|A|.

8.16.

Prove Theorem 8.3(ii): Let B be obtained from a square matrix two rows (columns) of A. Then |B| = |A|.

by interchanging

prove the theorem for the case that two columns are interchanged. Let t be the transtwo numbers corresponding to the two columns of A that are interchanged. If A = (oy) and B (6jj), then 6y Ojtcj)- Hence, for any permutation a.
position which interchanges the

We

184

DETERMINANTS

[CHAP.

Thus

\B\

2
2

(sgn (sgn

ff)6io.(i) 620.(2) ... &no-(n)

=
Since the transposition t
is

a) OiTirCl) "2to-(2)

"-nraCn)

an odd permutation, sgn ra

sgn r

sgn

<r

sgn

a.

Thus sgn a

sgn

TO,

and so
\B\

(sgn

TOr)

a.iTO-(l)a2T<r(2)

"^nrcrCn)

But as
\B\

a runs through -\A\.

all

the elements of S, to also runs through

all

the elements of S^; hence

8.17.

Prove Theorem 8.2: (i) If A has a row (column) of zeros, then \A\ = 0. (ii) If A has two identical rows (columns), then \A\ - 0. (iii) If A is triangular, then \A\ = product of diagonal elements. Thus in particular, |/| = 1 where / is the identity matrix.
(i)

Each term in |A| contains a factor from every row and so from the row of term of |A| is zero and so \A\ = 0.

zeros.

Thus each
obtain the

(ii)

in K. If we interchange the two identical rows of A, we Suppose 1 + 1 # Hence by the preceding problem, 1A| = |A| and so \A\ = 0. matrix A.

still

in K. Then sgn <r = 1 for every a e S. Since A has two idensuppose 1 + 1 = rows, we can arrange the terms of A into pairs of equal terms. Since each pair is 0, the tical determinant of A is zero.

Now

(iii)

Suppose
a^j

A =

(ay)
i

is

whenever

<
t

j.

lower triangular, that is, the entries above the diagonal are Consider a term t of the determinant of A:
(sgn
<r)

all zero:

aiij a2i2

where
'*"*n'

<t

i^H ...in

Suppose
ii 7^ 1

ii - 1.

Then

1<
1

ii

and so a^^

0;

hence

0.

That

is,

each term for which

is zero.

Now

suppose
ij

ti

=
1

but
12

iz - 2.

Then 2 <

ig

and

so

a^ =

0;

hence

0.

Thus each

term for which

or

^
.

is zero.

Similarly we obtain that each term for which ij 7^ 1 or % # 2 or a^n = product of diagonal elements. Accordingly, 1A| = a^^a^^
.

...

or

t 9^

is zero.

8.18.

Prove Theorem
(i)

8.3:

Let

be obtained from

by
fe

(ii)

(iii)
(i)

by a scalar fe; then |B| = interchanging two rows (columns) of A; then |B| = - |A|. adding a multiple of a row (column) of A to another; then
multiplying a row (column) of

|A|

|B1

\A\.

If the jth
\B\

row

of

A
is,

is

multiplied
|B|

by

fc,

then every term in

|A|

is

multiplied

by

fc

and so

k\A\. That

=
=

2
fc

(sgn o)

an

a2t2

C^^jiP
.

ni

2 (sgn a

a) aii

a2i2

Oni

1^1

(ii)

Proved in Problem

8.16.

(iii)

Suppose c times the feth row is added to the jth row of A. Using the symbol yth position in a determinant term, we have
\B\

/\

to denote the

=
The
|B|

2 (sgn c 2 (sgn
is
8.2(ii)

or)

aii aji^

{ca^i^

+ ajj.)

ai^

<r)

aj

agi^

fci^

2 (sgn

c) a^i^ a^i^.

a^.

. .

ai^

first

sum
1A|

hence by Theorem

the determinant of a matrix whose feth and ;th rows are identical; the sum is zero. The second sum is the determinant of A. Thus

c'0

A.

CHAP.

8]

DETERMINANTS

185

8.19.

Prove

Lemma

8.6:

For any elementary matrix

",

l^'A]

IE"!

|A|,

Consider the following elementary row operations: (i) multiply a row by a constant A; # 0; interchange two rows; (iii) add a multiple of one row to another. Let E^, JS?2 and E^ be the corresponding elementary matrices. That is, Sj, E^ and E^ are obtained by applying the above
(ii)

operations, respectively, to the identity matrix


l^il

/.

By
-\I\

the preceding problem,

k\I\

k,

\E^\

-1,

\E,\

|/|

Recall (page 56) that SjA is identical to the matrix obtained by applying the corresponding operation to A. Thus by the preceding problem,

\E^A\

k\A\
is

\Ei\\A\,

lE^A]

-\A\

l^^l lA],

\E,A\

\A\

1|A|

I^g] 1A|

and the lemma

proved.

8.20.

Suppose B is row equivalent to A; say elementary matrices. Show that:


(i)

B = EnEn-i

E2E1A where
if

the E, are

\B\

\En\ \Er,-i\

\E2\ \Ei\ \A\,

(ii)

\B\

^
=

if

and only
fey

\A\

^ 0.

(i)

By

the preceding problem,


\B\

|J7iA|

|Bi| 1A|.

Hence

induction,

\E\\E_,...E2E,A\

\E^\\E,_,\...\E2\\E,\\A\
\B\ =

(ii)

By

the preceding problem, ^i

for each

i.

Hence

if

and only

if

\A\ - 0.

8.21.

Prove Theorem
(i)

8.4:

Let

A be an w-square matrix.
is

Then the following are


it suffices to

equivalent:

is invertible,
6.44,

(ii)

nonsingular,
(ii)

(iii)

|A| 9^ 0.

By Problem
equivalent.

(i)

and

are equivalent.

Hence

show that

(i)

and

(iii)

are

Suppose A is invertible. Then A is row equivalent to the identity matrix /. But |/| ?* 0; hence by the preceding problem, |A| ^ 0. On the other hand, suppose A is not invertible. Then A is row equivalent to a matrix B which has a zero row. By Theorem 8.2(i), \B\ = 0; then by the preceding problem, \A\ = 0. Thus (i) and (iii) are equivalent.

8.22.

Prove Theorem
If
is

8.5:

\AB\

=
is
. . .

\A\\B\.
also singular and so \AB\ = = |A| |B|. On the other hand E2E1, a product of elementary matrices. Thus, by Problem
if

is

singular, then

AB
E^

nonsingular, then
|A|

A
=

=^

8.20,

\E^...E^E,I\

\E\...\E2\\E,\\I\

\EJ...\E2\\E,\

and so

|AJ5|

\E...E2E,B\

\EJ

lE^WE^WB]

|A| |B|

8.23.

Prove Theorem

8.7:

Let

A=

(a);

then

\A\

= anAn +

ttizAia

+ aiAi, where
.,

Aij is the cofactor of an.

Each term in \A\ contains one and we can write \A in the form
\

only one entry of the ith row

(aij.Ojg,

a,)

of A.

Hence

|A|

ajiAfi

ai2A*2

aiAf

(Note Ay is a sum of terms involving no entry of the ith row of A.) we can show that At. = A;,. = (-l)+i|MI

Thus the theorem

is

proved

if

the matrix obtained by deleting the row and column containing the entry ay. (HisAy was defined as the cofactor of Oy, and so the theorem reduces to showing that the two definitions of the cofactor are equivalent.)
is

where Afy

torically, the expression

186

DETERMINANTS
First

[CHAP. 8

we

consider the case that


Orm'^nn

n, j

n.
a)

Then the sum of terms


02<t(2)

in \A\ containing a is

nn

2 a

(sgn

ffli<r(i)

<n-l,cr(n-l)

where we sum over all permutations aSS for which ain) = n. However, this is equivalent (Prob.,n-l}. Thus A* = |M| = (-!)+ 1M| lem 8.63) to summing over all permutations of {1,
. .

last,

consider any i and }. We interchange the ith. row with each succeeding row until it is and we interchange the jth column with each succeeding column until it is last. Note that the determinant |Afy| is not affected since the relative positions of the other rows and columns are not affected by these interchanges. However, the "sign" of |A| and of Ay is changed n i and then

Now we

nj

times.

Accordingly,

A% =

(-l)-i + -i |M|

(-l)* + MMy|

8.24.

Let

A = (an) and by the row vector

let

B
.

(bn,

be the matrix obtained from &m). Show that


. . ,

A
4-

by replacing the

ith

row

of

|B|

biiAn
i,

bt2Aa

&iAi

Furthermore, show that, for

j =

ajiAn

+ +

(lizAti

+
4-

ttjnAin

and
Let

aijAii

023^2!

aiAi

B=

(6y).

By

the preceding problem,


\B\

ftiiBa

6i2^i2

6iBi
j

Since

By

does not depend upon the ith


\B\

row of B, By
ftjiAji

= Ay +

for

n.

Hence

6i2Ai2

6iAi

Now let A' be obtained from A by replacing the ith row of has two identical rows, |A'| = 0. Thus by the above result,
|A'|

by the

jth

row

of A.

Since A'

ajiAji

aj2-^i2

+
"

O-jn^ln

= =
0.

Using

|A*|

\A\,

we

also obtain that

Oti-^ii

2j-^2i

ni-Ani

8.25.

Prove Theorem
(l/|Al)(ad3A).
Let

8.8:

A-(adjA) = (adjA)-A =

|A|/.

Thus

if

|AI

0,

A-* =

A=

(ay)

and

let

(adj

A)

(fty).

The
ffliz.

ith

row
<*tl)

of

is
'^)

(ii.

Since adj A is the transpose of the matrix of cofactors, the ith column of adj the cofactors of the jth row of A: (A,i,Aj2, ...,A,)

is

the transpose of
(2)

Now

by,

the ij-entry in

(adj A), is obtained

by multiplying

(1)

and

(2):

6y

a^Aji

Uj^Ajz

ttin'Ajn

Thus by Theorem

8.7

and the preceding problem,


"

\
\A\
1.

if

^i
In other words,

Accordingly,

A -(adj A)
\A]
I.

is

the diagonal matrix with each diagonal element \A\.


(adj

(adj

A)

Similarly,

A)-A =

CHAP.

8]

DETERMINANTS

187

8.26.

Prove Theorem 8.9: The system of linear equations Ax = b has a unique solution if and only if A = ]A| ^ 0. In this case the unique solution is given by Xi = Ai/A,
X2

A2/A,

Xn

An/A.

By

invertible if

preceding results, and only if A

Ax = b = |A| #

has a unique solution


0.

if

and only

if

is invertible,

and

is

Now
obtain

suppose

A#0. By

Problem
X

8.25,

A-i =

(1/A)(adj A).
(1/A)(adj
.

Multiplying

Ax =

by

A -1, we
(1)

= A-^Ax =
is

A)b
If
6

Notethattheithrowof (l/A)(adjA)
Xi

(l/A)(Aii,

A^,

(1/A)(6i Aii

b^A^i

++
=

.,

Aj).

(61, 63,

.,

&)'

then,

by

(i),

6A

J
A

However, as in Problem

8.24;

6iAii

62^21

6Aj

A;

the determinant of the matrix obtained by replacing the ith column of Thus Xi = (l/A)Aj, as required.

by the column vector

6.

8.27.

Suppose

is
I.

invertible.

Show
|/|

that |P-i]

|P|-i.

P-^P =

Hence

|p-ip|

|p-i| |p|,

and so

|P-i|

|P|-i.

8.28.

Prove Theorem

8.11:

Suppose

and

are similar matrices.

Since A and B are similar, there exists an invertible matrix P such that by the preceding problem, |P| = |P-iAP| = |P-i| \A\ \P\ = \A\ lP-i| \P\ - \A\

Then |A| = |B|, B = P-^AP. Then

We
and
\A\

remark that although the matrices P-i and A may not commute, their determinants |P-i| do commute since they are scalars in the field K.

8.29.

Prove Theorem
multilinear,
(ii)

8.13:

is

alternating,

function,
Let

and

i.e. D{A) = |A|. D be the determinant function: D(A) = \A\. We must that D is the only function satisfying (ii) and
(1),

There exists a unique function D.cA^K such that (i) D is (iii) D(/) = 1. This function D is the determinant
show that

satisfies

(i), (ii)

and

(iii),

(iii).

By

A=

()

preceding results, D satisfies (ii) and (iii); hence we need show that it is multilinear. Suppose = (Ai, Ag, ., A) where A^ is the fcth row of A. Furthermore, suppose for a fixed i,
. .

Aj
Accordingly,

Bi

+ =

Cj, 61

where Bi

(b^,

6)

and
;

Q=
=

(ci,

. ,

c)

a^

Cj,

ajj

=63 +
A)

02,

...,

Expanding D(A)

\A\

by the
.

ith row,
.

D(A)

= D(A =
=
(61

.,

Bi

+
+

Ci,

.,

=
. . .

aaA^,

at^A^^

Ui^A^,

ci)Aii

(62

+ C2)A;2 +

(6

+ c)Ai

(fciAii

62Ai2

+ 6Ai) +

(ciA(i

+ C2A,.2 +

+ c^Ai^)

However, by Problem 8.24, the two sums above are the determinants of the matrices obtained from by replacing the ith row by Bj and respectively. That is,

D(A)

I>(A...,Bi

Ci,

...,A)

=
Furthermore, by Theorem

I>(Ai, ...,Bi, ...,A)

Z)(Ai, ...,Ci, ...,A)

8.3(i),

Z)(Ai, ...,fcAj

A)

fcD(A...,Ai, ...,A)

Thus

is

multilinear,

i.e.

satisfies

(iii).

188

DETERMINANTS
We next must prove the uniqueness of D. the usual basis of K", then by (iii), Z>(ei, eg, 8.73) that
is

[CHAP. 8

Suppose
. .

satisfies

(i),

(ii)

and

(iii).

If {e^,

e}

.,

e)

D{I)

1.

Using
in^
. . .

(ii)

we

also

have (Problem

Z)(ejj,ei2.

'%) =

sgna,

where a
A^^ of
fcii

i^

(D

Now

suppose

A =

(ay).

Observe that the


(ki. fc2.

fcth

row

A
+

is

^k =
Thus
I'(A)

>

"fcn)

"k22
.

+
.
.

afce

I>(aiiei

+ aie,

02161

a2e,

oiei

ae)

Using the multilinearity of D, we can write D(A) as a sum of terms of the form

= 2 ("Hj a2i2
where the sum
are equal, say
is
ij

'^"iJ
in

^'\'

%
i^

' K^
G
{1,
. . . ,

summed

over

all

i^

but

j ' k,

sequences 11^2 then by (ii),

where

n}.

If

two of the indices

^(%. %,
Accordingly, the sum in we finally have that
(2)

'\) =
all

need only be summed over

permutations

i^i^

J.

Using

(1),

D{A)

= =

2 "212 ni) 2 (sgn a^ a2i^


(iii

D{ei^,

Bj^,

ejj

a)

ai^,

where a
is

i^i2

...in

Hence

is

the determinant function and so the theorem

proved.

PERMUTATIONS
8.30.

Determine the parity of a


Method
1.

542163.

We need

to obtain the
(5,

number

of pairs

(i, j)

for which

i>
1,

and

precedes

in

a.

There

are:

3 numbers

4 and 2) greater than and preceding

2 numbers
3
1

(5
(5,

and

4)

greater than and preceding


6)

2,
3,

numbers

4 and

greater than and preceding


4,

number

(5)

greater than and preceding


5,
6.

numbers greater than and preceding


numbers greater than and preceding
Since 3

+ 2+3 +

is

odd, a is an odd permutation and so

sgn a

-1.

Method

2.

Transpose

1 to the first position as follows:

^42163
Transpose 2 to the second position:

to

154263 125463 123546 123456

154263
Transpose 3 to the third position:

to

125463
Transpose 4 to the fourth position:
1

to

sT^

to

Note that 5 and 6 are in the "correct" positions. Count the number of numbers "jumped": 3 + 2 + 3 + 1 = 9. Since 9 is odd, a is an odd permutation. (Remark: This method is essentially the same as the preceding method.)

CHAP.

8]

DETERMINANTS
3.

189

Method

interchange of two. numbers in a permutation is equivalent to multiplying the permutation by a transposition. Hence transform a to the identity permutation using transpositions; such as.
5

An

14
1

X
4

2^1
5

X4
3

6/3 6^4
5

S^

12 12
Since an odd number,
5,

4 4

X
6 5

of transpositions

was

used, a is

an odd permutation.

8.31.

Let a = 24513 and t = 41352 be permutations in mutations to and oroT, (ii) a~^.
Recall that a

Ss.

Find

(i)

the composition per-

24513 and t

=
3

41352 are short ways of writing


4

""(^24513/
which means
<,(1)

/I

5\

and

_ ~ =
=

/I

2
1

4 5

5N

V4
1

2/

= =

2,

<t(2)

4,

<r(3)

=
=

5,

(4)

and

<r(5)

and
r(l)
4,

r(2)

=
3
J,

1,

r(3)

3,

T (4)

and

t(5)

=2
4
^^

(i)

12

12
^/
^/

3
^^

5
^r

1
2

4,

j 1
4

13

and
..

13
i

15
Thus
to<t

1
2

1
3

12
2
1

i
5

i
3

i
4

15243

and aT

12534.

1
(ii)

3\

/I

3
5

4
2

5\

Vl
That
is,

5/

V4

3/

ff-i

41523.

8.32.

Consider any permutation

<7

jiji

jn.

Show
i

that for each pair

(i,

k)

such that

i>
there
is

and

precedes kin a

a pair

(i*,

k*) such that


i*

<

k*

and

cr(i*)

>

<r(fc*)

(i)
is

and vice versa. Thus cr is even or odd according as to whether there odd number of pairs satisfying (1).
Choose

an even or

and

i* and A;* so that a(i*) = i and precedes k in a it and only if i* < k*.

a{k*)

fc.

Then

>

fe

if

and only

if

<r(i*)

>

a(fe*),

190

DETERMINANTS
Consider the polynomial g - g{xi, ...,Xn) polynomial g g{xi, X2, Xs, xt).

[CHAP.

8.33.

y[{Xi-Xj).
'"^^

Write out

explicitly the

sum

The symbol 11 is used for a product of terms in the same way that the symbol 2 of terms. That is, Yl (a'i ~ ""j) means the product of all terms (Kj Xj) for which
i<i

is
i

used for a

<

j.

Hence

g{Xl,...,Xi)

iXi

X2){Xi Xa)(Xi Xi)(X2-X3)(X2-Xi)(X3 X^)

8.34.

Let

(T

be an arbitrary permutation.
<T{g)

For the polynomial g


that
{

in the preceding problem,

define

= Yl (^-^c" 'i<3

^(tw)-

Show

if it

o-

is IS

even

[g
Accordingly, aig)
Since a
is

odd

(sgn

<j)g.

one-one and onto,


a{g)

to"

(a'.ra)

a'.ru))

.11. KjorOj
.

,{xi-Xj)

Thus
of the

a{g)

g or a(g)

= g
i

form

(;

- Xj)

where

>

according as to whether there is an even or an odd number of terms Note that for each pair (i, j) for which j.
i

<

and

17(1)

>

<r{j)

(1)

is a term (ajo-fj, - x^^j-^) in a(g) for which a(i) > a(j). Since a is even even number of pairs satisfying (1), we have a(g) = g it and only if a if and only if a is odd.

there

if
is

and only

if

there
<r{g)

is

an

even; hence

= -g

8.35.

Let u,tG Sn. Show that sgn (to a) = (sgn T)(sgn <t). Thus the product of two even or two odd permutations is even, and the product of an odd and an even permutation
is

odd.

Using the preceding problem, we have


sgXi-{r

a)g

{Tc)(g)

T(CT(flr))

T((sgn

<r)sr)

(sgn T)(sgn

<T)sf

Accordingly, sgn(TO<r)

(sgn T)(sgn

<r).

8.36.

Consider the permutation a


"
ttj^i

J1J2

jn.

Show that sgn


Omk^

(7-1

=
=

sgn <t and, for scalars


kiki
.
.

aj^2

ttj^n

=
a.

aik^chk^

where
Since
e

<7~^

kn

We have a-^<T both odd. Hence sgn


Since
a

c,

the identity permutation.

is

even, cri and a are both even or

<r-i

=
is

sgn

JiJz

in

a permutation,
ff(fci)

aj^i Oj^a

a,j^n

"iki "zkz

nk-

Then

fej,

k^,

..,k

have the property that

=
.

1,

^(fea)

2,

.,

"(K)

=n

Let T

^1^2

kn-

Then for

1,

. ,

w,

(<jT)(i)

a{T(i))

a(fcj)

Thus aT

e,

the identity permutation; hence t

ct~i.

CHAP.

8]

DETERMINANTS

191

MISCELLANEOUS PROBLEMS
8.37.

Find det (T) for each linear operator


(i)

T:

T
T

is

the operator on R^ defined by


T{x, y,
z)

{2x

z,x + 2y- 4:Z, Sx-3y + z)

(ii)

is

the operator on the vector space

of 2-square matrices over

defined

by

T(A)

MA

where

M
c

d,
1

2
1

0-1'

(i)

Find the matrix representation of T relative

to,

say, the usual basis:

[r]

24
-3
1/

Then
det (T)

0-1
2(2-12)

12-4
3-3
1

- l(-3-6) =

-11

(ii)

Find a matrix representation of T


1

in

some basis of V,
1

say,

E. =
a
c

, =
^1

0/'

E^ "*

=
VO
1,

Then

T(Ei)

= = = =
c

b\/l dj\0

a
c

=
a
c

aEi

0^2

+ + + +

cE^

+ + + +

OE4

a
c

T(E^)

6\/0 dj\0
b\/0 d)\i b\/0 djio
0\
c

= =

0iS7i

+ + +

aE^

OS3

cE^

a
c

'b

nE^)

^d
b
1

bEi

OE2

dEs

OEi

a
e

T{Ei)

0^1

bE2

OE3

dEi

'a

Thus

[T]e

=
b ^0

a d
b

and

dj
c

a
det (T)

=
6

a d
6

a
c

a
b

d d

a^d^

bH"^

2abcd

/111
8.38.

Find the inverse of

A =
of the

1 .0

1
1,

The inverse of
Set

is

form (Problem

8.53):

A- =
*

z
'

AA-'^

l,

the identity matrix:

Ai4-i

Set corresponding entries equal to each other to obtain the system


a;-|-l

0,

2/-t-z-|-l

0,

192

DETERMINANTS

[CHAP. 8

/l-l
The
solution of the system
is
a;

0\

1, y

0,

1.

Hence

1
1/

\0

A~^

could also be found by the formula

A-> =

(adj i4.)/|A|

8.39.

Let

be a 2-linear, alternating function.

generally,

show that
is

if

is

Show that D{A, B) multilinear and alternating, then


...)

= - D{B,

A).

More

D{...,A, ...,B,
that
is,

= -D{...,B,
0.

...,A, ...)

the sign

changed whenever two components are interchanged.


D{A + B,A+B) =
Furthermore, since

Since

is

alternating,

is

multilinear,

= D{A+B,A+B) = D(A,A + B) + D(B,A + B) = D{A, A) + D(A, B) + D(B, A) +


But D(A, A)

D{B, B)

and

I>(B,

B)

0.

Hence

=
Similarly,

D(A, B)
D(.
.

D(B, A)
.
.

or
. .

D{A, B)
.)

= -

D(B, A)

=
=

.,

A + B,

.,

A+ B,
+ +

D(,...,A

A,...)

D(....,A, ...,B, ...)

+ =
and thus

D(...,B,...,A,
...)

...)

+ D{...,B....,B,
...)

...)

D(...,A, ...,B,

D(...,B, ...,A,
A,...).

D(...,A

B,...)

= -D(...,B

8.40.

Let

be the vector space of 2 by 2 matrices

over R.
if
(i)

Determine

whether or not
(ii)

D.V^R

is 2-linear

(with respect to the rows)

D{M) = a + d,

D{M) =
No.

ad.

(i)

For example, suppose

A=

(1, 1)

and and

B-

(3, 3).

Then

D(A,B)

= = d(\ 3)
a^,

D(2A.B)

- C^^

3)

2D(A, B)

(ii)

Yes.

Let

A=

(tti,

B=

(61, 63)

and

C=

(cj, Ca);

then

D(A,C)

dT'
s, <

"^)
R,

01C2

and

D(B,C)

= d(^'

^')

= V2

Hence for any scalars

D(sA

+ tB,C) =
=

i?/i
s(aiC2)

+ +

**^

sa,+

tb,^^

^sa,

th,)c,

t(6iC2)

8D(A, C)

tD{B, C)

That

is,

is

linear with respect to the first row.

Furthermore,

D(C,A) = d(^;
I'J

=
R,
tB)

c,a,

and

D(C, B)

= D^^^

^=

c,b.

Hence for any scalars

s, t

D(C,sA +

d(
s(cia2)

"J-.,

?..)=
=
sI>(C,A)

Ci(sa2+t62)

=
That
is,

t(ci62)

+ D(C,B)

is

linear with respect to the second row.

Both linearity conditions imply that

is 2-linear.

CHAP.

8]

DETERMINANTS

193

Supplementary Problems
COMPUTATION OF DETERMINANTS
8.41.

Evaluate the determinant of each matrix:

2
(i)

5
(ii)

6 3

-2
....

8.42.

Compute the determinant

of each matrix:

t-2
(i)

-3

/t-5
-1
(

t-l)'

<">

8.43.

For each matrix

in the preceding problem, find those values of t for

which the determinant

is zero.

8.44.

Compute the determinant


/2
(i)

of each matrix:

l\

/3
,

-2
4/

(ii)

-2 -4\ 5 -1
6

/-2 -1
,

(iii)

\1

-3

\0

1/

8.45.

Evaluate the determinant of each matrix:


f
(i)

-3 -6

1
t

(ii)
I

+
6

-3 -3
t-4,1
for which the determinant

8.46.

For each matrix


is zero.

in the preceding problem,

determine those values of

/l
8.47.

Evaluate the determinant of each matrix:

10-20
3-1

2
3
^"^
I

2\
3

1-2
-1
4

(i)

1-21'
2)

\4 -3

l2

2 -1

1/

COFACTORS, CLASSICAL ADJOINTS, INVERSES


'l
8.48.

2-2

3^
I,

For the matrix

4
,1
(i)

find the cofactor of:

2 -3^
5,
(iii)

the entry

4,

(ii)

the entry

the entry

7.

8.49.

Let

Find

(i)

adj A,

(ii)

A-i.

'1
8.50.

2
1

2'

Let

A =

Find
1,

(i)

adj A,

(ii)

A-i.

a
8.51.

Find the

classical adjoint of each

matrix in Problem

8.47.

8.52.

Determine the general 2 by 2 matrix


Suppose

for which

A =

adj A.

8.53.

is

diagonal and
'

B
di

is

triangular; say,
...

A =
,0

02

...

and

B
lO

62

194

DETERMINANTS
(i)
(ii)

[CHAP.

Show Show Show

that adj
that

is

diagonal and adj

is triangular.

is invertible iff all

6; = 0;

hence

is

invertible

iff all

^ 0.

...

(iii)

that the inverses of

and

(if

either exists) are of the


\

form
di2

t-i

...

/&r*
I

dm
d2n

A-1

|0
lO

ar^

...

5_,

^
\o

62

a-

'

That is, the diagonal elements of elements of A and B.

A-i and B-^

are the inverses of the corresponding diagonal

DETERMINANT OF A LINEAR OPERATOR


8.54.

Let

be the linear operator on R^ defined by


T{x, y,
z)

(3a;

- 2z,

5y

+ 7z,x + y + z)

Find det
8.55.

(T).

Let

DiV^V
(i)

be the differential operator,


(ii)

i.e.
(iii)

D{v)
{sin

=
t,

dv/dt.

Find det (D)

if

is

the space gen-

erated by

{l,t, .... t"},

{e*, e^*, e^t},

cos

t}.

8.56.

Prove Theorem
(i)

8.12:

Let

T and S

be linear operators on V.

Then:
if

det (ST)
that:

det(S)'det(r);

(ii)

is

invertible if

and only

det (7) 9^

0.

8.57.

Show

(i)det(lv)

where ly

is

the identity operator;

(ii)

det (T-i)

det(r)-i

if

is

invertible.

DETERMINANTS AND LINEAR EQUATIONS


8.58.

Solve by determinants:

,..

(1)

+ [ix
(Sx
,

5y

2y

= =

8
,

,..,

(")

f2x-Sy = -1 I4x + 7y = 1
.
,

8.59.

Solve by determinants:

(i)

= = [sx-iy-Gz = (2x-5y + x + 2y 2z

7 3
(ii)

Az

<

+3 = - Sz = [sy + z =
(2z

2y
2

+ Sx + 1. - 2x
if

8.60.

Prove Theorem

8.10:

The homogeneous system

Ax =

has a nonzero solution

and only

if

A =

IA| == 0.

PERMUTATIONS
8.61.

Determine the parity of these permutations

in Sg:

(i)

=
(i)

3 2 1 5 4,

(ii)

1 3 5

4,

(iii)

4253

1.

8.62.

For the permutations


Let T Let

<r,

and v

in

Problem

8.61, find

rc,

(ii) Tr<r,

(iii)

cr-i,

(iv)

t-i.

8.63.

5.

Show

that t<t runs through S as a runs through S; that

is,

S
<f*(x)

= =

{t a

iSf}.

8.64.

have the property that <t(w) = n. Let a* e S_i be defined by that sgn <r* = sgn a. (ii) Show that as a runs through S, where cr{n) <r e S, <T(m) = n}. S_i; that is, S_i = {a*

aeS

<r(x).

(i)

Show

n,

c* runs through

MULTILINEARITY Let V = (K"")"", 8.65. D:V-^K.


(i)

i.e.

is

the space of m-square matrices viewed as m-tuples of

row

vectors.

Let

Show

that the following weaker statement is equivalent to

being alternating:

i?(Ai,A2, ...,A)

=
A^.Az,

whenever Aj
(ii)

Ai+i for some

i.

Suppose D is m-linear and alternating. then D(Ai,...,AJ = 0.

Show

that

if

,A^

are linearly dependent,

CHAP.

8]

DETERMINANTS

195

8.66.

Let

be the space of 2 by 2 matrices


is 0,

j
(i)

over R.

Determine whether or not


(ii)

D:V->^K
(iii)

2-linear (with respect to the rows) if


(iv)

D{M) = ac-hd,

D{M)

D{M) =

D{M) =

= ah- ed,

1.

8.67.

Let V be the space of M-square matrices over K. Suppose B B V is invertible and so Define Z> V -* by Z?(A) = det (Afi)/det (B) where A G V. Hence
:

det (B)

= 0.

fl(Ai, Ag,

. ,

A)

=
is

det (AiB, A^B, .... AB)/det (B)


the ith

where Aj

is

the tth

row

and that det (A) det (B). This method


alternating,

A and !>(/) = 1.
of
is

so

A^B

row

of

AB. Show that

Z> is

multilinear and det (AB)

(Thus by Theorem 8.13, D{A) used by some texts to prove Theorem

det (A)

and so

8.5, i.e.

|Ai5|

\A\ |B|.)

MISCELLANEOUS PROBLEMS
8.68.

Let

be an w-square matrix.

Prove

IfeAl

fc" \A\

8.69.

Prove:

1
1

x^

x\
xf '2

...

X2

x\~^ ~"-l X^
-
of order n.

a:

xl

...

The above

is

called the

Vandermonde determinant

8.70.

Consider the block matrix

where

and

are square matrices. Prove \M\


.

=
.

\A\ \C\.

More

generally, prove that if is a triangular block matrix with square matrices Aj, the diagonal, then \M\ = |Ai| [Agl "l^ml

.,

A^

on

8.71.

Let A, B,

C and D

be

commuting m-square matrices.

Consider

the

2-square

block

matrix

^
8.72.

"

(c d)-

P'^^^^tl'**

W
is,

\A\\D\-\B\\C\.

Suppose

is ortfeoflroMttZ,

that

A^A =

I.

Show
{e<}

that

|A|

1.

8.73.

Consider a permutation

Jiij

in. .

Let
.,

whose

tth

row

is

e^., i.e.

A=

(e,-^, e^-^,

e^^).

Show

be the usual basis of X, and that |A| = sgn a.

let

be the matrix

8.74.

Let A be an M-square matrix. The determinantal rank of A is the order of the largest submatrix of A (obtained by deleting rows and columns of A) whose determinant is not zero. Show that the determinantal rank of A is equal to its rank, i.e. the maximum number of linearly independent rows (or columns).

Answers
8.41.
(i)

to

Supplementary Problems

-18,

(ii)

-15.

8.42.

(i)

t2

3t

t-

10,

(ii)

t^

2t

8.

8.43.

(i)

5,

-2;

(ii)

i, t

0.

-2.

8.44.

(i)

21,

(ii)

-11,

(iii)

100,

(iv)

196

DETERMINANTS
(i)

[CHAP. 8

8.45.

(<

+ 2)(t-3)(t-4),
-2;
(ii)

(li)

(t

+ 2)2(t-4),
-2.

(iii)

(t

+ 2)2(f-4).

8.46.

(i)

3, 4,

4,

-2;

(iii)

4,

8.47.

(i)

-131,

(ii)

-55.

8.48.

(i)

-135,

(ii)

-103,

(iii)

-31.

8.49.

adj

A =
\

-1
2

-1
0/

A-i

(adj A)/|A1

-2

i -^ \-l 1

|
0,

1
8.50.

-2\
6
,

adj

A =

-3 -1
2
1

A-i

=31
/-I

2\

-5/

\-2 -1

5/

-29 -26 -2\ (-16 -38 -16 -30 29 -8 51 -13 -1 -13 1 28 -18/

/
,...
I

21

-14 -17 -19^

-44 -29
\

H
1

33 13

11

(")

21

17

-19 -18;

8.52.

A =

/k
(

8.54.

det(r)

=
(ii)

4.

8.55.

(i)

0,

6,

(iii)

1.

8.58.

(i)

= =

21/26,

29/26;

(ii)

-5/13, y

1/13.

8.59.

(i)

5,

1, z

1.

(ii)

Since

A = 0,

the system cannot be solved by determinants.

8.61.

agn a

1,

agn t

=
(ii)

1, sgn v

1.

8.62.

(i)

Tv =
Yes,

53142,

ir(r

52413,

(iii)

<r-i

32154,

(iv)

t-i

14253.

8.66.

(i)

(ii)

No,

(iii)

Yes,

(iv)

No.

chapter 9

Eigenvalues and Eigenvectors


INTRODUCTION
In this chapter we investigate the theory of a single linear operator T on a vector space V of finite dimension. In particular, we find conditions under which T is diagonalizable. As was seen in Chapter 7, this question is closely related to the theory of similarity transformations for matrices.

polynomial and

with an operator T: its characteristic These polynomials and their roots play a major role in the investigation of T. We comment that the particular field K also plays an important part in the theory since the existence of roots of a polynomial depends on K.
shall also associate certain polynomials
its

We

minimum

polynomial.

POLYNOMIALS OF MATRICES AND LINEAR OPERATORS Consider a polynomial f{t) over a field K: f(t) = Ont" + + ait + Oo.

If

is

a square

matrix over K, then we define


/(A)

aA"

ttiA

ao/

where
nomial

/ is the identity matrix.


fit)

In particular,

we say

that

is

a root or zero of the poly-

if /(A)

0.

Example

9.1:

Let

A =
(

V
=

and

let

/(t)

2t2

- 3t + 7,

g(t)

t^

- 5t - 2.
=

Then

^)-i:!r-<3:)-a;) ^<l
and
1

a^;::
/O ^^
^

.(A)

2V

-/I

^3
g(t).

-5^3 ^^-2^^

2\

/l

0\ ^^

Thus

is

a zero of

The following theorem

applies.
let

Theorem

9.1:

Let / and g be polynomials over K, and

be an w-square matrix over K.

Then
(i)

(ii)

+ flr)(A) = /(A)+flr(A) {fg)(A) = f(A)giA)


(/

and, for any scalar k


(iii)

G K,
kf{A)
for any polynomials
f{t)

{kf){A)

Furthermore, since

f{t) g{t)

g{t) f{t)

and

g{t),

f{A)g{A)

giA)f{A)

That

is,

any two polynomials

in the

matrix

commute.

197

198

EIGENVALUES AND EIGENVECTORS

[CHAP.

Now
cut"

suppose

ait

do,

T :V -^ V is a linear operator on a vector space V over K. If then we define f{T) in the same way as we did for matrices:
f{T)

f{t)

a^T"

+aiT +

aoI

where / is now the identity mapping. We also say that T is a zero or root of f(t) if f{T) = 0. remark that the relations in Theorem 9.1 hold for operators as they do for matrices; hence any two polynomials in T commute.

We

Furthermore,
of f(T).

if

is

In particular,

a matrix representation of T, then /(A) if and only if /(A) = 0. f{T) =

is

the matrix representation

EIGENVALUES AND EIGENVECTORS


Let
is called

T -.V -*V

an eigenvalue of T

be a linear operator on a vector space V over a if there exists a nonzero vector v


T{v)

field

K.

scalar X

eV

for which

\v

Every vector satisfying this relation is then called an eigenvector of T belonging to the eigenvalue A. Note that each scalar multiple kv is such an eigenvector:
T{kv)

kT{v)

k(\v)

\{kv)

The

set of all such vectors is a subspace of

(Problem

9.6) called the

eigenspace of

\.

The terms

characteristic value

vector) are frequently used instead of eigenvalue

and characteristic vector (or: proper value and proper and eigenvector.

Example

9.2:

Let

I.V^V
1 is

Hence Example
9.3

an eigenvalue of

Then, for every vGV, I{v) = v = Iv. be the identity mapping. /, and every vector in V is an eigenvector belonging to 1.

Let 7" R^ ^ R2 be the linear operator which v GB? by an angle rotates each vector = 90. Note that no nonzero vector is a Hence T has no eigenmultiple of itself.
:

values and so no eigenvectors.

Example

9.4:

Let D be the differential operator on the vector space V of diflferentiable functions. We have Hence 5 is an eigenvalue of D )(e*') = 5e^*. with eigenvector e^'.

If A is an w-square matrix over K, then an eigenvalue of A means an eigenvalue of A is an eigenvalue of A if, for some nonzero viewed as an operator on K". That is,

\gK

(column) vector v

G X",

Av =

\v
A.

In this case v
Example

is

an eigenvector of

belonging to

9.5:

Find eigenvalues and associated nonzero eigenvectors of the matrix

We

seek a scalar

and a nonzero vector

X =
=

such that

A = AX = tX:

2 2

DO
or

(:

The above matrix equation


X
,

is

equivalent to the homogeneous system


tx ty
(
-s

[3x

+ +

2y
2y

= =

{t-l)x- 2y =

\-Zx + (t-2)y

(i)

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


Recall that the homogeneous system has a nonzero solution terminant of the matrix of coefficients is 0:
if

199

and only

if

the de-

t-1
-3
Thus
t is

-2

t2

3t

if

= {t-4){t+l) =
=
4

an eigenvalue of
t

if

and only

or

1.

Setting

4 in

(1),

3x -Sx

~
+
is

2y 2y

= =

or simply

3x

2y

= =

Thus V =

\V/
t

\3/
to
<

a nonzero eigenvector belonging to the eigenvalue

4,

and every other eigenvector belonging


Setting

is

a multiple of

v.

= -l

in (1),

-2x -3x

2y
3y
is

= =

or simply

Thus
t

w =

a nonzero eigenvector belonging to the eigenvalue


t

1, and every other eigenvector belonging to

= 1

is

a multiple of w.
is

The next theorem gives an important characterization of eigenvalues which quently used as its definition.

fre-

Theorem

9.2:

Let
is

be a linear operator on a vector space over K. Then if and only if the operator Xl T is singular. eigenspace of A is then the kernel of XI T.

T:V -^V

XGK
The

an eigenvalue of T

Proof. X

is

an eigenvalue of

if

and only

if

there exists a nonzero vector v such that

T{v)
i.e.

XV

or
also

{Xl){v)-T{v)

or

{Xl-T){v)

=
if

Xl

is

singular.

We

have that v

is

in the eigenspace of X if

and only

the above

relations hold; hence v is in the kernel of XI

T.
9.14)

We now
Theorem

sta+2 a very useful

theorem which we prove (Problem


belonging
to distinct

by induction:
are
linearly

9.3:

Nonzero

eigenvectors independent.
Consider the functions
If

eigenvalues

Example

9.6:

ei', 6^',

., e''n'

where

aj,

.,a are distinct real numbers.

Accordingly, e^i', ..., e' and so, by are eigenvectors of D belonging to the distinct eigenvalues ai, , a, Theorem 9.3, are linearly independent.
is

the differential operator then

D(ei'')

a^e'^''*.

We remark that independent eigenvectors can belong Problem 9.7).


DIAGONALIZATION AND EIGENVECTORS Let T:V -^ V be a linear operator on a vector
that

to the

same eigenvalue

(see

space

with

finite

dimension

n.

Note

T can be represented by a

diagonal matrix
'A;i

...

k2

...

,0

...

kni

200

EIGENVALUES AND EIGENVECTORS


if

[CHAP.

if

and only

there exists a basis

[vi,

.,v} of
kivi

for which

T{vi) T{v2)

kiVi

T{V)

knVn

that

is,

values

ki,

such that the vectors vi, .,Vn are eigenvectors of In other words: ., fen.
. .

T belonging

respectively to eigen-

Theorem

9.4:

if

T V - V can be represented by a diagonal matrix B has a basis consisting of eigenvectors of T. In this case the diagonal elements of B are the corresponding eigenvalues.
linear operator
if
:

and only

We

have the following equivalent statement.

Alternate

Form

of

Theorem

9.4:

w-square matrix A is similar to a diagonal matrix and only if A has n linearly independent eigenvectors. In this case the diagonal elements of B are the corresponding eigenvalues.

An

if

In the above theorem, eigenvectors of A, then B


Example

if

we

let

be the matrix whose columns are the

n independent

= P~^AP.
A =
/
(
j
.

9.7:

Consider the matrix

By Example
/2
{^^

9.5,

has two independent

eigenvectors

/2\
(^^j

and

1\
Set

P =

1\

(^_J.

_^j

and so

_, P i -

/1/5
(^^^^

1/5

_^^^

Then

is

similar to the diagonal matrix

B = P-^AP =

/1/5
1^3/5

l/5\/l

2\/2
2/1^3

1\

the diagonal As expected, the diagonal elements 4 values corresponding to the given eigenvectors.

-2/5/^3 and 1 of

-1/

^ /4 0\ ~ \0 -l) matrix B are the

eigen-

CHARACTERISTIC POLYNOMIAL, CAYLEY-HAMILTON THEOREM


Consider an %-square matrix

over a

field

K:
.
. .

(0,11
Chl

ai2
ffl22

ain
fflZn

ttnl

ttre2

flnn

The matrix

tin

A, where

called the characteristic

/ is the n-square identity matrix and matrix of A:

* is

an indeterminant,

is

(t an
~tt21
t

a22
ffli2

ain a2n
aim

ftnl
Its

ttn2
{tin

...

determinant
is

AA{t)
t,

det

A)

which

a polynomial in

is

called the characteristic polynomial of A.


AA{t)

We

also call

det

(tin

-A) =

the characteristic equation of A.

CHAP.

9]

EIGENVALUES AND EIGENVECTORS

201

Now

from each column; hence the above


AA{t)

each term in the determinant contains one and only one entry from each row and characteristic polynomial is of the form

{t- an){t

- 022) (*- ann)


terms with at most n 2 factors of the form
t

+
Accordingly,
AaC*)

an

t"

(au

+ a22+

+aTO)t"~^

terms of lower degree

Recall that the trace of

A is the sum of its diagonal elements. Thus the characteristic polynomial Aa(*) = det (i/ A) of A is a monic polynomial of degree n, and the coefficient of i"~^ is the negative of the trace of A. (A polynomial is monic if its leading coefficient is 1.)
Furthermore,
if

we

set

in Aa(<),

we

obtain

Aa(0)

\-A\

(-1)"[A|

But

Aa(0) is the constant

acteristic polynomial of the

term of the polynomial AaC*). Thus the constant term of the charmatrix A is (1)" \A\ where n is the order of A.

13
Example
9.8:

0^

The

characteristic polynomial of the matrix

A =

2
4

21
-21

-3

^(t)

\tI-A\

t-2
t

=
2
3.

<2

28

-4
As
expected, A(t)
is

a monic polynomial of degree

We now

state one of the

most important theorems

in linear algebra.
its

Cayley-Hamilton Theorem
Example

9.5:

Every matrix

is

a zero of

characteristic polynomial.
/I
I

9.9:

The

characteristic polynomial of the matrix

A =
-2

2^
Ji

A(t)

\tI-A\

t-3

t-2
A
is

t2

3t

As

expected from the Cayley-Hamilton theorem,

a zero of

A{t):

'<^'^G iJ-'Q i)-'Q


The next theorem shows the intimate
and eigenvalues.

= (:

:)

relationship between characteristic polynomials

Theorem

9.6:

Let

value of
Proof.

is an eigenbe an n-square matrix over a field K. A scalar A if and only if A is a root of the characteristic polynomial A(t) of A.

\GK

By Theorem

9.2,

Furthermore, by Theorem 8.4, of A(t). Thus the theorem is proved.

an eigenvalue of A if and only if XI A is singular. XI A is singular if and only if |a7 A| = 0, i.e. A is a root
is

Using Theorems
Corollary
9.7:

9.3, 9.4

and

9.6,

we

obtain
A(t) of

If the characteristic

polynomial

an ^-square matrix

is

a product

of distinct linear factors:

202

EIGENVALUES AND EIGENVECTORS


A{t)
i.e. if ai,
. .

[CHAP.

{t- ai){t -

aa)

(t

- an)

On are distinct roots of A{t),

then

is

similar to a diagonal

matrix whose diagonal elements are the

ok.

Furthermore, using the Fundamental Theorem of Algebra (every polynomial over has a root) and the above theorem, we obtain
Corollary
9.8:

Let

least

be an w-square matrix over the complex one eigenvalue.

field C.

Then

has at

Example

9.10:

Let

A =

Its characteristic

polynomial

is

A(t)

-1

2
t

=
2

(-3)(t2

l)

We
(i)

consider two cases:

A
A

Then A has only the one eigenvalue is a matrix over the real field R. Since 3 has only one independent eigenvector, A is not diagonalizable.
is

3.

(ii)

a matrix over the complex field C. Then A has three distinct eigenvalues: and i. Thus there exists an invertible matrix P over the complex field C for which
3, i

/3

0^
i

P-iAP =
\0
i.e.

-i,

is

diagonalizable.

Now

suppose

show that

and

A and B are similar matrices, say B = P^^AP where P is invertible. We B have the same characteristic polynomial. Using tl P~^tIP,
\tI-B\

=
=

\tI-P-'AP\

\P-HIP - P-'AP\
\P-^\\tI-A\\P\
|P~i| |P|

\P-mi-A)P\

Since determinants are scalars and commute, and since

1,

we

finally obtain

\tI-B\

= \tI-A\

Thus we have proved

Theorem

9.9:

Similar matrices have the same characteristic polynomial.

MINIMUM POLYNOMIAL Let A be an w-square matrix over a field K.


f{t)

Observe that there are nonzero polynomials

for which

polynomials
9.25);

we

f{A) 0; for example, the characteristic polynomial of A. consider those of lowest degree and from them we select one
i.e.

Among

these

whose leading

coefficient is 1,

which

is

monic.

Such a polynomial m{t) exists and


of A.

is

unique (Problem

we

call it

the

minimum polynomial

Theorem
There

9.10:

The minimum polynomial m{f) of


as a zero.

A divides every polynomial which has A In particular, m{t) divides the characteristic polynomial A(t) of A.
A(i).

is

an even stronger relationship between m{t) and

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


9.11:

203

Theorem

The

characteristic

and minimum polynomials of a matrix

have the same

irreducible factors.

divide the other.

This theorem does not say that w(i) = A(t); only that any irreducible factor of one must In particular, since a linear factor is irreducible, m(t) and A(t) have the same linear factors; hence they have the same roots. Thus from Theorem 9.6 we obtain

Theorem

9.12:

scalar A

is

an eigenvalue for a matrix

if

and only
2

if

is

a root of the

minimum

poljTiomial of A.
1

Example

9.11:

Find the minimum polynomial

>(*)

of the matrix

2
2
5

The characteristic polynomial of A is A(t) = |t/- A] = (t- 2)(t- 5). By Theorem 9.11, both t 2 and t 5 must be factors of m(t). But by Theorem 9.10, vnif) must divide A(t); hence m(i) must be one of the following three polynomials:
TOi(t)

(t-2)(t-5),

We know from the

verify that ni(A) minimum polynomial of A.

= (t-2)3(t-5) Cayley-Hamilton theorem that in^iA) A(A) = 0. The reader # but WgCA) = 0. Accordingly, m^it) = {t 2)^ (t 5) is
W2(t)

(t-2)2(t-5),

m^fy

can
the

Example

9.12:

Let

be a 3 by 3 matrix over the real

field R.

We

show that

of the polynomial f{t) = t^ + 1. its characteristic poljmomial A(t).

By

the Cayley-Hamilton theorem, Note that A(t) is of degree 3; hence

cannot be a zero A is a zero of it has at least

one real root.

Now suppose A is a zero of f{t). Since f(t) is irreducible over R, f{t) must be the minimal polynomial of A. But /(t) has no real root. This contradicts the fact that the characteristic and minimal polynomials have the same roots. Thus A is not a zero of f{t).
The reader can verify that the following
field

by 3 matrix over the complex

is

a zero of

f(t):

fo

-1

0^

10
lO

CHARACTERISTIC AND MINIMUM POLYNOMIALS OF LINEAR OPERATORS Now suppose T:V^V is a linear operator on a vector space V with finite dimension. We define the characteristic polynomial A(<) of T to be the characteristic polynomial of any
matrix representation of T. By Theorem 9.9, A{t) is independent of the particular basis in which the matrix representation is computed. Note that the degree of A{t) is equal to the dimension of V. We have theorems for T which are similar to the ones we had for matrices:

Theorem Theorem

9.5':

is

a zero of

its

characteristic polynomial.
is

9.6':

The scalar

\GK

an eigenvalue of T

if

and only

if

is

a root of the

characteristic polynomial of T.

of T is defined to be the multiplicity The algebraic multiplicity of an eigenvalue of A as a root of the characteristic polynomial of T. The geometric multiplicity of the eigenvalue A is defined to be the dimension of its eigenspace.

\G K

Theorem

9.13:

The geometric
multiplicity.

multiplicity of

an eigenvalue A does not exceed

its

algebraic

204
Example

EIGENVALUES AND EIGENVECTORS


9.13:

[CHAP.

Let V be the vector space of functions which has {sin be the differential operator on V. Then

e,

cos 9} as a basis,

and

let

D{sme)
Z>(cos e)

cosfl

=
=

0(sin e)

l(cos9)
O(cos e)

= sin 9
is

l(sin e) +

The matrix

of

in the

above basis

therefore
t

A =
=
t2

[D]

=
(

Thus

det (tl

-A)

-1
t

and the characteristic polynomial of

is

A(t)

t^

l.

On the other hand, the minimum polynomial m{t) of the operator T is defined independently of the theory of matrices, as the polynomial of lowest degree and leading coefficient 1 which has T as a zero. However, for any polynomial f{t),
f{T)

if

and only

if

f{A)

A is any matrix representation of T. Accordingly, T and A have the same minimum polynomial. We remark that all the theorems in this chapter on the minimum polynomial of a matrix also hold for the minimum polynomial of the operator T.
where

Solved Problems

POLYNOMIALS OF MATRICES AND LINEAR OPERATORS


9.1.

Find f{A) where

A =

~z\ and

f{t)

f-M +

1.

9.2.

Show

that

^ =
=

^^

^ ^^^^ ^^

f(^)

*^

4t

5.

'-'

--" a;)'-G :)-<;:)


= =
(Z>2

a:
let

9.3.

Let V be the vector space of functions which has {sin 6, cos ^} as a basis, and D be the differential operator on V. Show that D is a zero of f{t) = t^ + 1.
Apply f(D)
to each basis vector:

/(D)(sin9)
/(D)(cos
e)

{D^
is

+ 7)(sin e) + /)(cos e)

= r2(sin e) + = DHcos e) +

/(sine)
/(cos e)
i;

= =

-sin

-cos

+ +

sin 9

cos e
into

= =
by
f(D).

Since each basis vector

mapped

into 0, every vector

SV

is also

mapped

Thus

fm =

0.

This result

is

expected since, by Example 9.13,

/(<) is

the characteristic polynomial of D.

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


Show
T

205

9.4.

Let

A be a matrix representation of an operator T. representation of f{T), for any polynomial f{t).


Let be the mapping T h* A, i.e. which sends the operator need to prove that <i>(f(T)) - f(A). Suppose fit) = at" -\duction on n, the degree of fit).
<t>

that /(A)

is

the matrix

into its

We

a^t

matrix representation A. a^. The proof is by in-

Suppose
matrix.

TO

0.

Recall that
<t>{f(T))

0(/')

where

/' is

the identity

mapping and

/ is the identity

Thus

=
0.

<t>(Hr)

ao0(/')

V
aiT

/(A)

and so the theorem holds for n

Now assume the theorem holds for polynomials of degree less than n. algebra isomorphism,
^if(T))

Then

since ^ is

an

= = =

0(ttr"

a_irn-i

a^')

a0(r) 0(r-i)

0(a_ir"-i

+ +

aiT
a<,/)

a^')
/(A)

oAA-i + (a_jA"-i +

aiA

and the theorem

is

proved.

9.5.

Prove Theorem 9.1: Let / and g be polynomials over K. Let A be a square matrix over K. Then: (i) (/ + fir)(A) = /(A) + flr(A); (ii) {fg){A) = /(A) g{A); and (iii) (fe/)(A) = kf(A) where fc G K.
Suppose /
f(A)

at"

+ Oi* + Oq

and g

b^t^

+ bit + =

bf,.

Then by

definition,

= aA +
and

+
ftj

ttiA

+
if

Oq/

and
m. Then

^(A)

ft^A"

biA

bol

(i)

Suppose

mn

let

i>

sr

= (a+6)t"+

+K +
+ =
6iA
n+

6i)t

(ao+6o)

Hence
(/

+ g)(A) = =
fg
fc

{a

+ 6)A" ++(! + 6i)A + +


6A

(a<,

6o)^

aA

ajA

tto^

60/

/(A)

flr(A)

m
c^t''

(ii)

By

definition,

c + t" +>"++ Ci*

Co

n+m
aj6fc_i.

fc

2 =

where

c^

eio^fc

"'i^fc-i

Ojcfro

1=0

Hence

(/ff)(A)

= 2
=
fcoo.

CfcA^

and
n+

/n
/(A)f,(A)

=
kf

2 \i=o

OiAMf 2 6jAi) /\j=o /

N/
+ +

2 2 2M*+^ = fc=0 t=0 j=0


and so

nm

m
CfcA"

(/ff)(A)

(iii)

By

definition,
(fe/)(A)

A;at"

fcajt

fcaA

fettiA

fcoo/

A;(aA

+ ajA + o,/) =

k /(A)

EIGENVALUES AND EIGENVECTORS


9.6.

Let

A,

vectors of

Vx

is

be an eigenvalue of an operator T:V^V. Let Vx denote the set of all eigenT belonging to the eigenvalue X (called the eigenspace of A.). Show that a subspace of V.

Suppose v,w

&

Vx,

that

is,

T(v)

\v

and

r(w)

\w.

Then for any scalars a,b

& K,

T(av

+ 6w) =

r(i;)

6 T(w)
to X,

i.e.

aiw)
av

b{\w)
V^.

- Mav + bw)
Hence Vx
is

Thus av

+ bw

is

an eigenvector belonging

+ bw G

a subspace of V.

206

EIGENVALUES AND EIGENVECTORS


1

[CHAP.

9.7.

Let
(ii)

A =

4\
g
I

(i)

Find

all

eigenvalues of
that

and the corresponding eigenvectors.


is

Find an invertible matrix


Form

P such
*/

P-^AP

diagonal.

(i)

the characteristic matrix

-AoiA:
\0
tj

\2

3J

-2

t-3

U)

The characteristic polynomial


A(t)

A(t) of

A
1

is its

determinant:

\tI-A\

-2

-4

=
3

4t

{t-5){t+l)

The roots of

A(t) are 5

and 1, and so these numbers are the eigenvalues of A.


5.

We

obtain the eigenvectors belonging to the eigenvalue


4
(1) to

First substitute

into
to

-4-

the characteristic matrix


5

obtain the matrix

The eigenvectors belonging


i.e.,

form the

solution of the

homogeneous system determined by the above matrix,

.-2

-4\/x\ 2)\y)

_ /0\ ~ \o)

f
""^

4x-4y =
2y

\^2x +

2/

form the kernel of the operator tl A for = 5.) The above system has only one independent solution; for example, x = 1, y = 1. Thus " = (1, 1) is an eigenvector which generates the eigenspace of 5, i.e. every eigenvector belong(In other words, the eigenvectors belonging to 5
t

ing to 5

is

a multiple of

v.

We

obtain the eigenvectors belonging to the eigenvalue 1.

Substitute

= 1 =

into {1)

to obtain the

homogeneous system

-2
-2

-4 -4

~2x -2x -

4.y 4i/

= =

or

2j/

The system has only one independent


is

solution; for example, x

2,

3/

-1.

Thus

w=
2

(2,

-1)

an eigenvector which generates the eigenspace of 1.

(ii)

Let

be the matrix whose columns are the above eigenvectors:


is

P
/5

1
1

-1

Then

B P~^AP

the diagonal matrix whose diagonal entries are the respective eigenvalues:

B = P-^AP =
{Remark. Here
vectors {v, w}.

'1/3

2/3\/l
-1/3 JV 2

1/3

4\/l 2\ _ sjil -1/ "

-1

is

Hence

the transition matrix from the usual basis of R2 to the basis of eigenB is the matrix representation of the operator A in this new basis.)

9.8.

For each matrix,

find all eigenvalues


1

and a basis of each eigenspace:


3\
3
,

(i)

A =
I

-3 -5

/-3
(ii)

-l'

5 =

-7
-6

5-1
6

6-6
Which matrix can be
(1)

-2

diagonalized,

and why?

Form
istic

the characteristic matrix tl polynomial A(t) of A:

and compute

its

determinant to obtain the character-

-3 -6

1
t

A(t)

\tI-A\

+
6

5
f

-3 -3

=
4

(t

+ 2)2(t-4)

The roots of

A(t) are

2 and

4;

hence these numbers are the eigenvalues of A.

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


We
find a basis of the eigenspace of the eigenvalue 2.

207

Substitute

= 2

into the char-

acteristic

matrix

tl

to obtain the

homogeneous system
f-Sa;

or

3a;

[6a;

+ + +

32/ 3j/ 62/

32
3z

= =
==

or

2/

6z

The system has two independent solutions, e.g. a; = l, 2/ = l, z = and a; = 1, j/ = 0, z = 1. Thus u = (1, 1, 0) and v = (1, 0, 1) are independent eigenvectors which generate the eigenspace of 2. That is, u and v form a basis of the eigenspace of 2. This means that every eigenvector belonging to 2 is a linear combination of u and v.

We

find a basis of the eigenspace of the eigenvalue 4.

Substitute

into the char-

acteristic

matrix

tl

to obtain the

homogeneous system
3x
Sx

6x

+ + +

By 9y 6y

--

3z

-- 3z

= = =

2y

z
z

= =

variable; hence any particular nonzero solution, e.g. x = 1, y = 1, generates its solution space. Thus w = (1, 1, 2) is an eigenvector which generates, and so forms a basis, of the eigenspace of 4.

The system has only one free

2,

Since A has three linearly independent eigenvectors, A is diagonalizable. be the matrix whose columns are the three independent eigenvectors:

In fact, let

/-2
Then P-^AP =
\

As

expected, the diagonal elements of

P~^AP
t

are the eigenvalues of

corresponding to the

columns of P.

+
7
6

-1

1 1
t

(ii)

A(t)

\tI-B\

t-5
-6
4.

=
2

(t

2)2(t-4)

The eigenvalues

of

are therefore

2 and

We

find

to obtain the

a basis of the eigenspace of the eigenvalue 2. homogeneous system X


--

Substitute

= -2

into tl

-B

Ix--ly + 6a; -6y


The system has only one independent solution, forms a basis of the eigenspace of 2.
e.g.

= = =
1,
2/

=0
u
(1, 1, 0)

a;

1,

0.

Thus

We find a basis of the eigenspace of the eigenvalue obtain the homogeneous system
7a; 7a;

4.

Substitute

into tl

to

2/ 2/

6a;-

62/

+ + +

2
z

6Z
a;

= = = =
0,
2/

7a;

2/

z
a;

= =
v
(0.1,1)

The system has only one independent solution, forms a basis of the eigenspace of 4.

e.g.

1, z

1.

Thus

Observe that B is not similar to a diagonal matrix since B has only two independent Furthermore, since A can be diagonalized but B cannot, A and B are not similar matrices, even though they have the same characteristic polynomial.
eigenvectors.

208

EIGENVALUES AND EIGENVECTORS

[CHAP.

9.9.

Let

A = A
and

Find

all

eigenvalues and the corresponding


(i)

eigenvectors of
field C.
(i)

viewed as matrices over

the real field R,

(ii)

the complex

Mt) =
Hence only 2
is

\tI-A\

=
Put
t

t-1

= A

(2

t-1
into tl

4i

(t

- 2)2

an eigenvalue.

and obtain the homogeneous system


X

X + y = X + y =

The system has only one independent solution, e.g. a; = 1, y = 1. Thus i) = (1, 1) is an eigenvector which generates the eigenspace of 2, i.e. every eigenvector belonging to 2 is a multiple
of
V.

We
Since
(ii)

also have
^B(t)

\tl-B\

t-2

1
t

=
1

t2

t2

has no solution in R,

has no eigenvalue as a matrix over R.

Since A^(t) = (t 2)2 has only the real root 2, the results are the same as in (i). That is, 2 is an eigenvalue of A, and v = (1, 1) is an eigenvector which generates the eigenspace of 2, i.e. every eigenvector of 2 is a (complex) multiple of v.

The

characteristic matrix of

is

Ajj(t)

\tl

B\ =
i.

t^

+ i.

Hence
t

and i are the eigen-

values of B.

We

find the eigenvectors associated with

Substitute

in tl

to obtain the

homogeneous system

'i-1
-2

i+lAy) ~
=

\/x\ _ /0\ [oj

/
""^

{i-l)x

+ =

=
(i

\-2x + {i+l)y =
x
1,

-l)x + y =

The system has only one independent solution, e.g. an eigenvector which generates the eigenspace of i.

1 -

Thus

w=

(l,l

i)

is

Now
'-i

substitute
1

into tl

to obtain the
{

homogeneous system
Q

\/x\

/0\

(il)x + y =
-2x

{-i

- l)y =
1,

(-i

-l)x + y

=
i)

The system has only one Independent solution, e.g. x an eigenvector which generates the eigenspace of i.

i.

Thus w'

(1,1-1-

is

9.10.

Find all eigenvalues and a basis of each eigenspace of the operator by T{x, y, z) = {2x + y,y- z, 2y + 4).

R^

-*

R^ defined

First find a matrix representation of T, say relative to the usual basis of R^:

/2

0^

A =
The characteristic polynomial
A(*) of

m
then

is

t-2
A(t)

-1

\tI-A\

=
T.

t-1
-2

=
4

(t-2)2(t-3)

t-

Thus

and 3 are the eigenvalues of

We

find a basis of the eigenspace of the eigenvalue 2.

Substitute

into tl

to obtain

the homogeneous system


/

/a\

y y

=0
+
z

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


e.g.

209

The system has only one Independent solution, a basis of the eigenspace of 2.

1,

0, z

0.

Thus u

(1, 0, 0)

forms

We
the

find a basis of the eigenspace of the eigenvalue 3.

Substitute

into tl

to obtain

homogeneous system

The system has only one independent solution, forms a basis of the eigenspace of 3.
Observe that T
is

e.g.

1,

1,

2.

Thus

(1, 1,

2)

not diagonalizable, since

has only two linearly independent eigenvectors.

9.11.

Show
T{v)

that

is

an eigenvalue of T
is

if

and only
if

if

is

singular.

We
=

have that

an eigenvalue of T

and only

if

there exists a nonzero vector v such that

Ov

0,

i.e.

that

is

singular.

9.12.

Let

and

B be w-square matrices. Show that AB


9.11

and

BA

have the same eigenvalues.

By Problem
singular,
(iv)

lowing statements are equivalent:

BA

is

and the fact that the product of nonsingular matrices is nonsingular, the fol(i) is an eigenvalue of AB, (ii) AB is singular, (iii) A or B is singular, (v) is an eigenvalue of BA.
there exists a nonzero vector v such that

Now suppose X is a nonzero eigenvalue of AB. Then ABv = Xv. Set w = Bv. Since \ # and v = 0,

Aw = ABv =
But

\v

and so

w #

is

an eigenvector of

BA

belonging to the eigenvalue X since

BAw - BABv =
Hence X
of
is

B\v = \Bv = \w

an eigenvalue of BA.

Similarly,

any nonzero eigenvalue of

BA

is also

an eigenvalue

AB.
Thus

AB

and

BA

have the same eigenvalues.

9.13.

Suppose
of

A.

is

an eigenvalue of an invertible operator T.

Show that A~*


X

is

an eigenvalue

T-K
Since

is invertible, it is also

nonsingular; hence by Problem 9.11,

# 0.

By

definition of

ing r-i to both sides, we obtain is an eigenvalue of r~i.

an eigenvalue, there exists a nonzero vector i; for which T(v) Xv. Applyv = T-i(\v) = xr-i(i;). Hence r-i(v) = X-iv; that is, X"!

9.14.

Prove Theorem 9.3: Let Vi, .,Vn be nonzero eigenvectors of an operator T:V ->V Then vi, belonging to distinct eigenvalues Ai, A. .,Vn are linearly independent.
.
.

. ,

The proof Assume > 1.


where the
Oj

is

by induction on Suppose

n.

Ii
-I-

1,

then Vj

is

linearly independent since

Vi

0.

OiV,

02^2

av

=
we
obtain by linearity
T{0)

{1)

are scalars.

Applying T
aiT(Vi)

to the

above relation,

<hT(v.,)

+ +

aT(v)

But by hypothesis

r(i;j)

XjVj;

hence
ajXiVi

02X2^2

ttn^n^n

(2)

210

EIGENVALUES AND EIGENVECTORS


On
the other hand, multiplying
(1)

[CHAP. 9

by

X,

Now
By

subtracting

(5)

from

(2),

ai(Xi

- Xj^i +
=
0.

a2{\2~K)'"2

+
0.

+
(1)

a_i(\_i
Xj

- Xjiin-i =
Xj

induction, each of the above coefficients is


aj

Since the

are distinct,
ai;

X 9^

for

# w.
Thus

Hence

a^-i

Substituting this into

we

get

0,

and hence a

0.

the Vi are linearly independent.

CHARACTERISTIC POLYNOMIAL, CAYLEY-HAMILTON THEOREM


9.15.

Consider a triangular matrix


'an
ai2

a22

...

CUnj

Find

its characteristic

polynomial A{t) and


is

its
is

eigenvalues.
t

Since

is

triangular and tl

diagonal, tl

A
t

also triangular with diagonal elements

/*-!!
tl

-<12
0.22

-a In
"-in

- A

^
\

t-anni
t :

Then

A(t)

|t/

A|

is

the product of the diagonal elements


A(t)

=
.

(t
.

- aii)(t - a22) (*- nn)


,
i.e.

Hence the eigenvalues of

are an,

tt22,

. ,

its

diagonal elements

1
9.16.

3\
3
.

Let

A =

Is

similar to a diagonal matrix?

If so, find

one such matrix.

3/
Since A are distinct,
is

is

are the diagonal elements 1, 2 and 3. Since they triangular, the eigenvalues of similar to a diagonal matrix whose diagonal elements are 1, 2 and 3; for example,

/i
2

o^

\0

3;

9.17.

For each matrix

find a polsoiomial

having the matrix as a root


/l

4-3

= (i -3> <) = (' :4)'


By

<'")

^ =

\\I
its

Therefore
(i)

the Cayley-Hamilton theorem every matrix is a root of we find the characteristic polynomial A(t) in each case.

characteristic polynomial.

A(t)

\tI-A\

= =

t-2
-1
f

-B

+
3

+t3

11

(ii)

A()

==

\tI-B\

t-2
-7
t

=
4

(2

+
-4

2t

13

t-1
(iii)

3
3

A(t)

\tI-C\

-2

-1

(t

- l)(t2 -2t-5)

t+1

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


9.5:

211

9.18.

Prove the Cayley-Hamilton Theorem


polynomial.
Let

Every matrix

is

a zero of

its

characteristic

be an arbitrary w-square matrix and let A(i) be


A(t)

its characteristic

polynomial; say,

\tI-A\

= <+

a_it"-i

++!* +

do

Now

let B(t)

of the matrix tl

denote the classical adjoint of the matrix The elements of B{t) are cofactors A and hence are polynomials in t of degree not exceeding n 1. Thus
B(t)

tl A.

B_it"-i

B^t

Bo
t.

where the Bj are re-square matrices over of the classical adjoint (Theorem 8.8),

K which

are independent of

By

the fundamental property

{tI-A)B{t)
or

\tI-A\I
(<"

(i/-i4)(B_it-i+

+Bi + Bo) =
coefficients of

+ a_it-i+

l-ai

+ aoK
t,

Removing parentheses and equating the

corresponding powers of
1

Bn-l =

Bn-2~AB_i

a-il

Bo

- ABi = -ABo =
.

a,/

agl
.,

Multiplying the above matrix equations by A", A"-^,

A, I respectively,

A"Bn-i = A" An-iB_2 - AJ?_i 3^ a-iA~i

A-2B_3-A-iB_2 = a_2A"-2
ABo - A^Bi - a^A

ABo =
Adding the above matrix equations,

o^

= A +
In other words, A(A)

a_iA-i
a zero of

aiA

oo/

0.

That

is,

is

its characteristic

polynomial.

9.19.

Show
By

that a matrix

and
\tI

its

transpose A' have the same characteristic polynomial.

the transpose operation,

(t/-A)'

tl*

A* :=

tl

- A*.

have the same determinant,


acteristic polynomial.

A\ -

|(t7-A)*|

\tI-A*\.

Hence

Since a matrix and its transpose A and A* have the same char-

9.20.

Suppose

/Ai
\

^1

B\

where Ai and Az are square matrices.


is

Show

that the char-

acteristic polynomial of

M
~B

the product of the characteristic poljmomials of Ai and


\
^"*=^ ^y Problem 8.70,

Ai.

Generalize.

tl-M = |t/ A| \tl B\,


I

/tl-A,

tl-A/'
as required.

\tI-M\

tl

Ai
tl

-B - A^

By

induction, the characteristic poljmomial of the triangular block matrix

212

EIGENVALUES AND EIGENVECTORS


lA,

[CHAP. 9

B
A2

M
where the Aj are square matrices,
is

=
\

D
...

A^l

the product of the characteristic polynomials of the Aj.

MINIMUM POLYNOMIAL
9.21.

Find the minimum polynomial m{t) of

A =

2
1 1

-2
The characteristic polynomial of

is

t-2
A(f)

-1

t-2
t1

t-2.

-1

t-1
2

-1

t-2

i-4

(t-3)(t-2)3

and

<

The minimum polynomial m{t) must divide A(i). Also, each irreducible factor 3, must be a factor of m{t). Thus m(t) is exactly one of the following:
/(t)

of A(t),

i.e.

(t-3)(t-2),

flr(t)

(i-3)(-2)2,

/i()

(t-3)(t-2)3

We

have
-1

/(A)

= (A-37)(A-2/)

-1

o\ 1' '0

1
-

-2 -2
1

-1

1/ \:
o\ lo
1

-2

ff(A)

(A

- 37)(A - 27)2 =

-1

-2

-1
,0

=
1

0-2
Thus
g{t)

(t

3)(t 2)2

is

the

minimum polynomial

of A.

Remark.

We know that

h{A)

A(A)

=;

of g{t) is less than the degn:ee of h(t); hence g(f), and not h(t),

by the Cayley-Hamilton theorem. However, the degree is the minimum poljmomial of A.

9.22.

Find the minimal polynomial m{t) of each matrix (where a^Oy.

\o

*/

\o
A(t)

A,
find

(i)

The
A(t)

characteristic polynomial of

A
B

is

{t

X)2. We

A \/ #

0;

hence

m{t)

(t-\)K

(ii)

The
or
{t

is A(i) = (t characteristic polynomial of find (B X/)2 - 0; thus OT(t) X)3.)

We

(iii)

The
A(t)

characteristic polynomial of

is

A(t)

X)^. (Note m(t) is exactly one of t \, = A(t) = (t - X)S. = (t - X)*. We find (C - X/)3 # 0; hence

{t

X)2 =

m{t)

(t-X)'.

CHAP.

9]

EIGENVALUES AND EIGENVECTORS

213

9.23.

Let

M
fe(t)

=
of

(A
I

0\

where

and

are square matrices.


multiple of the

Show

that the

minimum
g(f)

polynomial m(f) of

is

the least

common

minimum
=

polynomials

and

and
is

B respectively.

Generalize.
of

Since m(V)

the

minimum polynomial
is

M, m(Af) =

"m,(A.)

m(B)^

and hence m(A)


Similarly,

=
h{t)

and

mifi) = divides m{t).

0.

Since g{t)
is

the

minimum polynomial
g(t) g{t)

of A, g{t) divides m{t).

Thus m(t)

a multiple of

and and

h(t).

Now

let f{t)

be another multiple of

h(t);

then /(M)

/f(A)
V

\
/(B),
is

/O

0^

0.

But m{t) is the minimum polynomial of multiple of g{t) and /i(t).

M; hence

m(t) divides

/(t).

Thus m{t)

the least

common

We

then have, by induction, that the

minimum polynomial

of

M
where the Aj are square matrices,
the A,.
is

\o
the least

common

multiple of the

minimum polynomials

of

9.24.

Find the minimum polynomial m(i) of


'2

Let

A
are

A,

C and D

(t

2)^,

t2

and

5 t-1

respectively.
4:

c The
t2

D =

(5).

The minimum polynomials

of

characteristic polynomial of

is

\tI-B\

-2

t-3
of B.

7t

10

= (t-2)(t-5)

and so

it is

also the

minimum polynomial
'A '\

Observe that

=
,0

B
C
2?/

Thus m(t)

is

the least

common

multiple of the

minimum

polynomials of A, B,

C and

D. Accordingly, m(t)

= tm - 2)2(t 5)

9.25.

Show

that the

minimum polynomial

of a matrix (operator)

exists

and

is

unique.

By the Cayley-Hamilton theorem, A is a zero of some nonzero polynomial (see also Problem 9.31). Let n be the lowest degree for which a polynomial f(t) exists such that /(A) = 0. Dividing f(t) by its leading coefficient, we obtain a monic polynomial m(t) of degree n which has A as a zero. Suppose m'(t) is another monic polynomial of degree n for which m'{A) = 0. Then the difference m{t) m'(t) is a nonzero polynomial of degree less than n which has A as a zero. This contradicts the original assumption on n; hence m(t) is a unique minimum polynomial.

214

EIGENVALUES AND EIGENVECTORS


Prove Theorem
9.10:

[CHAP.

9.26.

The minimum polynomial m(t) of a matrix

(operator)

divides every polynomial which has


acteristic polynomial of A.

as a zero.

In particular, m{t) divides the char-

nomials

f{A) = 0. By the division algorithm there exist polyor deg r(t) < deg m(t). Suband r(t) = for which f{t) = m{t) q(t) + r(t) and 'm{A) = 0, we obtain r(A) = 0. If stituting t = A in this equation, and using that f(A) = r{t) = 0, then r(t) is a polynomial of degree less than m(t) which has A as a zero; this contradicts and so f{t) = m{t) q(t), i.e. m(t) divides /(*) the definition of the minimum polynomial. Thus r(t) =

Suppose
q{t)

f(t) is

a polynomial for which

and

r{t)

9.27.

Let m{t) be the minimum polynomial of an %-square matrix A. acteristic polynomial of A divides (m(i))''.
Suppose
m(t)

Show

that the char-

= f+

Cjf-i

-|-

-I-

c^-it

c^.

Consider the following matrices:

= / = A + cj Bo = A^ + CjA +
Bo
Bi

Cg/

B^_i

= =

A"--!
I

CiA'-2

4-

c^_i7

Then
Bi

B^

- ABg = cj B^-AB^ = C2I

Also,

-AB^^i = C^ - {Ar+CiAr-i+

+Cr-iA+CrI)

= c^ =
Set
B{t)
c^I

i(A)

= f-i^o + f-^Bi +

tBr-2

Br-1

= (t'-Bo+t'-'^Bi+ +tBr-i) - (t'-^ABo + tr-^ABi+ +ABr-i) = t^Bo+ tr-i{Bi-ABo)+ t'-2(B2-ABi)+ -f t(B,._i - AB^-a) - AB^-i + Cr-itl + C^ = f/ Cif-l/ + C^f-^I + = m(t)I The determinant of both sides gives \tl A\ \B{t)\ = \m(t) I\ = (TO(t))". Since \B(t)\ is a polynomial, 1*7 A divides (m(t))"; that is, the characteristic polynomial of A divides (n(t))".

Then {tI-A)-B(t)

-1-

9.28.

Prove Theorem 9.11: The characteristic polynomial A{t) and the minimum polynomial m{t) of a matrix A have the same irreducible factors.
Suppose
divides A(t).
(m(t))^.
f{t) is

On
f(t)

But

an irreducible polsmomial. If f{t) divides m{t) then, since m(t) divides A(t), f(t) the other hand, if f(t) divides A(t) then, by the preceding problem, /() divides Thus m{t) and A(t) have the same is irreducible; hence f(t) also divides m{t).

irreducible factors.

9.29.

be a linear operator on a vector space V of finite dimension. Show that T is and only if the constant term of the minimal (characteristic) polynomial of T is not zero. + a^t + a^f. Suppose the minimal (characteristic) polynomial of T is f(t) = f + a_it' 1 Each of the following statements is equivalent to the succeeding one by preceding results: (1) T is is not a root of m(t); (v) the is not an eigenvalue of T; (iv) invertible; (ii) T is nonsingular; (iii) constant term Uf, is not zero. Thus the theorem is proved.

Let

invertible if

-I-

CHAP.

9]

EIGENVALUES AND EIGENVECTORS

215

9.30.

Suppose

dimF =

Let

T:V ^V

he an invertible operator.

Show

that T-*

is

equal to a polynomial in
r

of degree not exceeding n.

n.

Let m(t) be the minimal polynomial of T. Then m{t) Since T is invertible, Kq # 0. We have

= f + a^-if i +
OiT

a^t

a^,

where

m(r)

y + a^_ir'-i +
I

do/
1
tto

Hence

+ aJ)T =

and

r-i

(Tr-i

+ ar-i r'-2 +

+ aj)

MISCELLANEOUS PROBLEMS
9.31.

Let T be a linear operator on a vector space V of dimension n. Without using the Cayley-Hamilton theorem, show that T is a zero of a nonzero polynomial.

- rfi. Consider the following N-\-l operators on V: I, T,T^, .... T^. Recall that the Let vector space A(V) of operators on V has dimension rfi. Thus the above iV + 1 operators are linearly dependent. Hence there exist scalars a^, Oj, ., aj, for which a^^T^ + + a^T + a^ = 0. Accordingly, T is a zero of the polynomial f{t) = a^^t^ + + a^t + Oq.

9.32.

Prove Theorem

9.13: Let A be an eigenvalue of an operator multiplicity of X does not exceed its algebraic multiplicity.

T:V ^V. The geometric

Vi, ...,Vr.

Suppose the geometric multiplicity of X is r. Then X contains r linearly independent eigenvectors Extend the set {dJ to a basis of V: {v^, ...,v^, w^, .,Wg}. We have
.

T{vi)
1'(^2)

= = = = =
=

XVi \V2

1\Vr)

W,
aiiVi a2iVi

T(w,)
I'(^2)

+ + +

... ...

+ + +

airVr a2rVr

+ 6nWi + + b2tWi +
+
&SIW1

+ + +

bi,w.
b2s'Ws

7'(W.)

O'sl'Vl

...

O'sr'Or

bss^s

The matrix of T in the above basis

is

/^
X

ii
{
1

21

0-sl\

'12

22

s2

r.
1

=
".

l<Hr
1

ttgr

O'sr

=
621 ...

(^'l

A^

*"
&12

6,1

Vo
/

5/

622

6r2

\.
where

hs

hs

6ss

A =

(0.;^)'

and

B=

()

By Problem
at least
r,

9.20 the characteristic polynomial of X/,,

acteristic polynomial of

M and

hence T.

which is (t \Y, must divide the charThus the algebraic multiplicity of X for the operator T is

as required.

9.33.

Show

that

A =

is

not diagonalizable.

The

characteristic polynomial of

is

A(t)

(t

1)^;

hence 1
t

is the

find a basis of the eigenspace of the eigenvalue 1.

Substitute

into the

only eigenvalue of A. We matrix tl A to obtain

the homogeneous system

216

EIGENVALUES AND EIGENVECTORS

[CHAP.

o)(^) = (I)
The system has only one independent
of the eigenspace of
1.

'

{
x

= =
1,

solution, e.g.

0.

Hence u

(1, 0)

forms a basis

Since

has at most one independent eigenvector,

cannot be diagonalized.

9.34.

Let

F be

an extension of a

field

K.

Let

be an w-square matrix over K.


\tl

Note that
that
is,

A may also be viewed as a matrix A over F. Clearly and A have the same characteristic polynomial. Show
same minimum polynomial.

~ A\ =

\tl

- A\,

that

and

also

have the

Let m{t) and m'(t) be the minimum polynomials of A and A respectively. Now m'{t) divides every polynomial over F vifhich has A as a zero. Since m(t) has A as a zero and since m{t) may be viewed as a polynomial over F, m'{t) divides m(t). We show now that m(t) divides m'(t).
Since m'(t)
is

a polynomial over
m'(t)

which

is

an extension of K, we may write


/aft) 62

fi(t)hi
61,
. .

+
.

+ +

fnit)b

where /i() are polynomials over K, and We have


m'{A)

6 belong to

F
+

and are linearly independent over K.


/(A)6

/i(A)bi

/2(A)62

=
that, for each pair {i,j),

(1)

Let alP denote the y-entry of

/fc(A).

The above matrix equation implies

a</'6i

aif 62

a^f &

=
tty'''

Since the

64

are linearly independent over


/i(A)

and since the a\P S K, every

0.

Then

0,

/2(A)

0,

...,

/(A)

which have A as a zero and since m(t) is the minimum polyare polynomials over as a matrix over K, m(t) divides each of the /;() Accordingly, by (1), m(t) must also That is, divide m'(t). But monic polynomials which divide each other are necessarily equal. m(t) = tn'{t), as required.
Since the
/i(t)

nomial of

9.35.

Let

{vi,

v} be a basis of 7.
T{vs)

Let
.
.

T{V2)

=
0.

chiVi,

aaivi

asiVi,

.,

T F -* 7 Tivn) =
:

be an operator for which


aiVi

T{vi)

0,

an,n-iv-i.

Show

that

T"-

It suffices to

show that
Ti{Vj)

= =
is

(*)

for

1,

.,n.

For then
THvj)

it

follows that
T-i(Ti(Vj))

r"-5(0)

0,

for ^

and, since {v^, ...,} is a basis,

T"
j.

0.

We

prove

{*)

follows (for j

2,

by induction on .,n} from


. .

The case

true by hypothesis.

The inductive step

Ti{Vj)

= = =

Ti-HT(Vj))

= TS-HajiVi+ +aj^j.iVj^i) +

aj^Ti-Hvi)
ajiO

+aj.,_irJ-i(Vj_i)

+ +

aj,j_iO

=
is

Remark. Observe that the matrix representation of T in the above basis


diagonal elements
0:

triangular with

/O

a^i

ttsi

...
..
.

a-ni

032

a2

...

a,_i

\0

...

CHAP.

9]

EIGENVALUES AND EIGENVECTORS

217

Supplementary Problems
POLYNOMIALS OF MATRICES AND LINEAR OPERATORS 9.36. Let f(t) = 2*2 - 5t + 6 and g(t) = t^ - 21^ + t +
i.

Find f(A), g{A), f(B)

and g(B) where

9.37.

Let

r:E2^R2

be defined by

T(x,y)

{x

y,2.x).

Let

/(i)

=
c.

t2

2t

3.

\n& f(T)(x,y).
be the differential

9.38.

Let

operator.

be the vector space of polynomials v(x) = ax^ Let f{t) = fi + 2t ~ 5. Find f(D){v{x)).

hx

Let

D:V ^V

9.39.

Let

A
.0 '8

Find A2, A3, A".

12
8

0^

9.40.

Let

B =

12
8/

Find a real matrix

such that

B=

As.

9.41.

Consider a diagonal matrix


'

M and
a. "-2

a triangular matrix N:
\

...

a^

...

M
Show
that, for

"

and
a

AT

02

...

...

any polynomial
'/(tti)

f(t),

f(M) and f(N) are of the form


\

...
f^"'^^

//(ai)

f(M)

=
I

^^

and
...

/(AT)

f{aj

9.42.

Consider a block diagonal matrix

M and
... ...

a block triangular matrix N:


\

/Ai

/Ai
and

B
A,

=
\

A2

N =

...

A,
that, for
\

where the Aj are square matrices. Show


'/(Ai)
...

any polynomial

/(*),

f(M) and f{N) are of the form

//(Ai)

X
/(A2)

...
...

f{M)

=
I

/(A2)

...

^^^
/(A)/

^(^)

...

...

/(Aj/
More

9.43.

Show

generally,

that for any square matrix (or operator) A, (P^iAP)" = P-iA"F where show that f{P-^AP) = p-if{A)P for any polynomial f(t).

is invertible.

9.44.

Let f{t) be any polynomial. then /(A) is symmetric.

Show

that:

(i)

/(Af)

(/(A))*;

(ii)

if

is

symmetric,

i.e.

A*

A,

EIGENVALUES AND EIGENVECTORS


9.45.

For each matrix,

find all eigenvalues

and linearly independent eigenvectors:


(")

^ =(1

/22\
3)'

=(3

/42\
3)'
(-)

^=(13
are diagonal.

/5-1

Find invertible matrices Pj, Pg and P3 such that P~iAPi,

P-^BP^ and P~^CPs

218

EIGENVALUES AND EIGENVECTORS


For each matrix,
find all eigenvalues

[CHAP.

9.46.

and a basis for each eigenspace:


/
1 1

/3
(i)

l\
2
,

2\

/l
,

1 1

0^

:=

4
1

(ii)

B =

-1

(iii)

C =
\0

\1

3/

\-l

14/

ly

When

possible, find invertible matrices Pj,

P2 and P3 such that P~^APi, F^'BPj ^"d Pj^CPs are

diagonal.

9.47.

Consider

A =

Vl

4y

and

B =

\1Z

-3/

as matrices over the real field R.

Find

all

eigen-

values and linearly independent eigenvectors.

9.48.

Consider A and B in the preceding problem as matrices over the complex values and linearly independent eigenvectors.

field C.

Find

all

eigen-

9.49.

space:

For each of the following operators T -.K^ -* R^, find all eigenvalues and a basis for each eigen(i) T{x,y) = (3x + 3y,x + 5y); (ii) T{x,y) = (y,x); (iii) T(x,y) = (y,-x).
For each of the following operators T{x, y,z) (x + y + (i)
2)

9.50.

T R^
:

-> R3,

find

all
(ii)

eigenvalues
T(,x,

eigenspace: (iii) t\x, y,

z,

2y

+ z,

2y

+ 3);

y,z)

and a basis for each (x + y,y + z, 2y z);

= (x-y,2x + 3y + 2z, x-^y + 2z).


field

9.51.

For each of the following matrices over the complex independent eigenvectors:

C, find all

eigenvalues and linearly

<"(:
9.52.

;)

""(J

D-

G:r). (;:?;
T.

Suppose v

is

aS + bT where a and
9.53.

an eigenvector of operators S and 6 are any scalars.

Show

that v

is

also

an eigenvector of the operator

Suppose v is an eigenvector of an operator T belonging to the eigenvalue is also an eigenvector of T" belonging to \".

X.

Show that

for

n >

0,

9.54.

Suppose X

is

an eigenvalue of an operator T. Show that

/(X) is

an eigenvalue of

f(T).

9.55.

Show Show

that similar matrices have the

same eigenvalues.
Give an example where

9.56.

that matrices

and A* have the same eigenvalues.

and A' have

different eigenvectors.

9.57.

Let
its

S and T

eigenspace.

be linear operators such that ST TS. Let X be an eigenvalue of Show that is invariant under S, i.e. S(W) C W.

T and

let

be

9.58.

Let
of

be a vector space of finite dimension over the complex invariant under a linear operator T V -* V. Show that
:

C. Let W # {0} be a subspace W contains a nonzero eigenvector of T.


field

9.59.

.yV^G K" be linearly independent eigenvectors of Let A be an ii-square matrix over K. Let v^, belonging to the eigenvalues Xj, X respectively. Let P be the matrix whose columns are the vectors Vi,...,v. Show that P~^AP is the diagonal matrix whose diagonal elements are the eigenvalues Xj, X.
. . . .
.

CHARACTERISTIC AND MINIMUM POLYNOMIALS


9.60.

For each matrix,

find a polynomial for

which the matrix

is

a root:
'2
3

-2^
4
1.

^^

^ = (4

D'

(")

^ = (3

3)'

^"^

'^='[1

i_i

CHAP.

9]

EIGENVALUES AND EIGENVECTORS

219

9.61.

Consider the w-square matrix

A =

Show
9.62.

that

/()

(t-

X)

is

both the characteristic and

minimum polynomial

of A.

Find the characteristic and minimum polynomials of each matrix:


/2 5 2 0\ 3
1

IX

A =
\0

4
3

2 5 7/

c =
1 3

0X000 0X00
0X0

0\

VO

x/

1
9.63.

0\

/2
and

0\
2 2
.

Let

A =
\0

B =
\0

Show

that

and

have different characteristic

1/

1/

polynomials (and so are not similar), but have the same matrices may have the same minimum polynomial.

minimum

polynomial.

Thus nonsimilar

9.64.

The mapping T:V -*V defined by T{v) = kv Show that T is the scalar mapping belonging to T is m(t) = t-k.
Let

is

called the scalar


if

k&K
for some

mapping belonging
if

to

&

K.

and only

the minimal polynomial of

9.65.

be an n-square matrix for which

A^ =

k>

n.

Show that A" =


polynomial.

0.

9.66.

Show

that a matrix

and

its

transpose A* have the same

minimum

9.67.

Suppose

f(t) is

T:V ^V. Show

an irreducible monic polynomial for which f(T) that f(f) is the minimal polynomial of T.

where

Is

a linear operator

9.68.

Consider a block matrix acteristic matrix of Af

\^

"/

Show

that

tl

=
(

fj

_ n

is

the char-

9.69.

Let r be a linear operator on a vector space of finite dimension. Let Tf' be a subspace of V invariant under T, i.e. T(W) cW. Let Tyf-.W-^W be the restriction of T to W. (i) Show that the characteristic polynomial of Tt divides the characteristic polynomial of T. (ii) Show that the minimum polynomial of Tyf divides the minimum polynomial of T.

ail
9.70.

i2
"^22

<*i3
"'23

Let

A =
,

'*21

Show

that the characteristic polynomial of

A
t

is

a^i

a^2
(ail

<^33

A()

+ 022 + a33)f2 +

"ll
"^21

''12 '*22

"11

"'13

"22 "32

Hi
"23
"21

^13

"31

"33

"22 032

"23

"33
"31

"S3

9.71.

Let be an m-square matrix. The determinant of the matrix of order n obtained by deleting the rows and columns passing through diagonal elements of A is called a principal minor of degree n~m. Show that the coefficient of t in the characteristic polynomial A(t) = \tI A\ is the sum of all principal minors of A of degree n multiplied by ( l)"-. (Observe that the preceding problem is a special case of this result.)

220

EIGENVALUES AND EIGENVECTORS


Consider an arbitrary monic polynomial f(t) = t" + a_if"^i n-square matrix A is called the companion matrix of f{t):
. .

[CHAP.

9.72.

a^t

a^.

The following

1 1

do -1 -
1

-ftn-l

Show
9.73.

that

f{t) is

the

minimum polynomial

of A.

Find a matrix

whose minimum polynomial

is

(i)

t^

St^

6t

8,

(ii)

t^

51^

2t

7t

4.

DIAGONALIZATION
9.74.

Let

A =

\c

dj

be a matrix over the real

field R.

Find necessary and

sufficient conditions

on

a, b, c

and d so that

is

diagonalizable,

i.e.

has two linearly independent eigenvectors.

9.75.

Repeat the preceding problem for the case that

A
if

is

a matrix over the complex


if its

field C.

9.76.

Show

that a matrix (operator)

is

diagonalizable

and only

minimal polynomial

is

a product

of distinct linear factors.

9.77.

Let

diagonalizable.

K^
9.78.

such that (i) AB = BA and (ii) A and B are both be Ji-square matrices over Show that A and B can be simultaneously diagonalized, i.e. there exists a basis of in which both A and B are represented by diagonal matrices. (See Problem 9.57.)

and

Let

E iV ^V

be a projection operator,

i.e.

E^

=
i

E.

Show
j

that

E
is

is

diagonalizable and, in fact,

can be represented by the diagonal matrix

A =

where r

the rank of E.

Answers
-26
9.36.

to

Supplementary Problems
-40

f(A)

-3 _27
(4a;

39\

/3

6\

/3
^(^)

12

9{A)

-65

-27)' /(^)=(o

,)'

= (o

15

9.37.

f(T)(x, y)

= =

- y, -2x + 5y).
+
(4a

9.38.

f(D){v(x))

-5aa;2

- 5b)x +

(2a

+ 26 - 5c).

"'

--':
Hint.

I)'

^'-(i
'2

t)'
61
c
)

^-ii:
Set

a
2

9.40.

Let

A =

B = A^

and then obtain conditions on

a, 6

and

c.

.0

2j
(/(A))

9.44.

(ii)

Using
Xi

(i),

we have

=
4,
6,

/(A)

/(A).

9.45.

(i)

(ii)

(iii)

= 1, M = (2, -1); = 1, M = (2, -3); Xi \ = 4, u= (1,1).

Xg

X2

= =

-u

1;

= =
/
(

(1, 1).

(1, 1).

Let Pi

2
(

1\
J
)

_^

and P2

1\
1.

P3 does not exist since

has only one independent

eigenvector, and so cannot be diagonalized.

CHAP.

9]

EIGENVALUES AND EIGENVECTORS


= 2, M = (1, -1, 0), V = (1, 0, -1); Xj = 6, w = (1, 2, 1). = 3, M = (1, 1, 0), V = (1, 0, 1); X2 = 1, w = (2, -1, 1). = 1, tt = (1, 0, 0), V = (0, 0, 1).
/

221

9.46.

(i)

Xi

(ii)

Xi

(iii)

1\
2

/I

2\

Let Pj

=
\

1
-1

and P2

1
1

P3 does not

exist since

C has

at most two

1/

\0
and

1/

linearly independent eigenvectors,

so cannot be diagonalized.

9.47.

(i)

= = =

3,

(1,-1);

(ii)

B
Xj

has no eigenvalues

(in R).

9.48.

(i)

3,

M= M=

(1,

-1).

(ii)

=
1)

2i,

(1,

- 2i);
Xi

Xg

=
m

-2i,

i;

(1,

+ 2i>
-1, v

9.49.

(i)

Xi

2,

(3,

-1); X2
(in B).

6,

=: (1, 1).

(ii)

1,

(1, 1);

X2

(1,

-1).

(iii)

There

are no eigenvalues

9.50.

(i)

(ii) (iii)

= 1, M = (1, 0, 0); Xj = 4, = (1, 1, 2). = 1, M = (1, 0, 0). There are no other eigenvalues X = (2, -2, -1); X3 = Xi = 1, M = (1, 0, -1); X2 = 2,
Xi
1;

(in R).
3,

!;

w=

(1,

-2, -1).

9.51.

(i)

Xi

1,

M=
(iv)

(1,0);
Xi

X2
t,

i, 1)

(1,1

(1,

-t).

M =

(2, 1 - i);

+ i). Xg =

(ii)

i,

X v

= =

1,
(2,

m=
1

(1,0).

(iii)

Xj

2,

u=

(3,i);

Xj

-2,

i).

9.56.

A = ( ^ ) Then X = 1 is the only eigenvalue and v = (1, 0) \^ /I 0\ X = 1 is still the only of X = 1. On the other hand, for A* = j ( ^ the eigenspace of X = 1. generates
Let
.
''

generates the eigenspace


eigenvalue, but

w=

(0, 1)

9.57.

Let v G W, and so T{v) = Xv. Then T(Sv) = S(Tv) - S(\v) = \(Sv), that is, Sv is an eigenvector and thus S(W) C W. of T belonging to the eigenvalue X. In other words, Sv e

9.58.

Let T:W-*W be the restriction of T to W. The characteristic polynomial of T is a polynomial an over the complex field C which, by the fundamental theorem of algebra, has a root X. Then X is which is also an eigenvector of T. T has a nonzero eigenvector in eigenvalue of T, and so

9.59.

Suppose T(v)

Xv.

Then

{kT){v)

kT{v)

k(\v)

(k\)v.

9.60.

(i)

/(<)

= t^-St + 43,

(ii)

g(t)

^t^-8t + 23,

(iii)

h(t)

t^

- 6*2 + 5f - 12.
(<-3); m(<)

9.62.

(i)

A(t)

(t-2)3(t-7)2; m(f)
m()

(t-2)2(t-7).

(ii)

Ht)

(t-3).

(iii)

A()

(t-X)5;

= -X.
/O

9.73.

Use the

result of

Problem

9.72.

(i)

A =

h
\0
1

-8\ -6
5/

(ii)

A =

9.77.

Hint.

Use the

result of Problem 9.57.

chapter 10

Canonical Forms
INTRODUCTION
r be a linear operator on a vector space of finite dimension. As seen in the preceding chapter, T may not have a diagonal matrix representation. However, it is still possible
Let
of ways. This is the main topic decomposition theorem, and the In particular, we obtain the primary of this chapter. triangular, Jordan and rational canonical forms.
to "simplify" the

matrix representation of

in a

number

We
true
if

comment that the triangular and Jordan canonical forms

exist for

if

and only
is

if

the characteristic polynomial A{t) of

K is the complex field

T has all its roots in the base field K. C but may not be true if K is the real field R.

This

always

introduce the idea of a quotient space. This is a very powerful tool and will be used in the proof of the existence of the triangular and rational canonical forms.

We also

TRIANGULAR FORM
be a linear operator on an n-dimensional vector space V. resented by the triangular matrix

Let

Suppose T can be rep-

(an

ai2
(122

...
. . .

ttin

0,2n

ann

Then the

characteristic polynomial of T,
A{t)

|*/-A|

{t

- an){t - a^i)
is

[t

- ann)
is

a product of linear factors. namely,


is

The converse

also true

and

an important theorem;

Theorem

10.1:

Let

be a linear operator whose characteristic poljmomial factors Then there exists a basis of V in which T is into linear polynomials. represented by a triangular matrix.

T:V-^V

Alternate

Form

of

Theorem

10.1:

Let A be a square matrix whose characteristic polynomial factors into linear polynomials. Then A is similar to a triangular matrix, i.e. there exists an invertible matrix P such that P'^AP is triangular.

say that an operator T can be brought into triangular form if it can be represented by a triangular matrix. Note that in this case the eigenvalues of T are precisely those entries appearing on the main diagonal. We give an application of this remark.

We

222

CHAP.

10]

CANONICAL FORMS
10.1
:

223

Example

Let

Show

is

be a square matrix over the complex field C. Suppose X is an eigenvalue of A2. that a/x or -Vx is an eigenvalue of A. We know by the above theorem that similar to a triangular matrix
/Ml

B
Hence A^

1^2

is

similar to the matrix

/.?

52

/jj

Since similar matrices have the same eigenvalues, X = ya? for some = VX or ^j = -y/x; that is, Vx or - Vx is an eigenvalue of A.

i.

Hence

INVARIANCE Let T:V-*V


T-invariant
restricted
if

be linear.

T maps

subspace IF of
i.e.

into itself,

if

vGW

is

implies

said to be invariant under T or T{v) G W. In this case T

defined

by T{w)

toW defines a linear operator on W; that is, T induces a linear operator f:W-*W = T{w) for every w GW.
10.2:

Example

Let

T K^ ^ R3
:

be the linear operator which rotates each vector about the z axis
T(x, y,
z)

by an angle

e:

{x cose

sin

e,

x sine
-

y cos

e,

z)

Observe that each vector w = {a, b, 0) in the xy plane remains in under the mapping T, i.e. is r-invariant. Observe also that the z axis U is invariant under T. Furthermore, the restriction of T to rotates each vector about the origin O, and the restriction of T to TJ is the identity mapping on U.

T(v)

.fj

^^

'

Example

10.3:

Nonzero eigenvectors of a linear operator T V ^ V may be characterized as generators of T-invariant 1-dimensional subspaces. For suppose T{^v) = \v, v 0. Then = {kv, k e K}, the 1-dimensional subspace generated by v, is invariant
:

under

because
T{kv)

k T(v)

k(\v)

= kWEiW

Conversely, suppose dim 17 = 1 and m generates U, and U is invariant under T. Then T{u) e V and so T(u) is a multiple of u, i.e. T(u) = /lu. Hence u is an eigenvector of T.

The next theorem gives us an important

class of invariant subspaces.


let f{t)

Theorem

10.2:

Let

T:V^V
is is

be linear, and

be any polynomial.

Then the kernel

of f{T) is invariant under T.

The notion

of invariance

related to matrix representations as follows.

Theorem

10.3:

an invariant subspace of T:V-^V. Then T has a block A B' matrix representation where A is a matrix representation of q r^
]

Suppose

the restriction of

to

W.

224

CANONICAL FORMS

[CHAP. 10

INVARIANT DIRECT-SUM DECOMPOSITIONS A vector space V is termed the direct sum of its
if

subspaces Wi,

.,Wr, written

every vector v

GV

can be written uniquely in the form


1)

= wi + W2+

iVr

with Wi

G Wi

The following theorem

applies.

Theorem

10.4:

Suppose Wi, ...,Wr are subspaces of V, and suppose


{Wn,
.

Wm^},

.,

{WrU

tVrn^}

are bases of Wi,...,Wr respectively. Then .,wii, Wi if and only if the union {wn,
. .

V
.
.

is

the direct
. .

sum
is

of the

..wn,

.,wJ

a basis

of V.

suppose T-.V-^V .,Wr: subspaces Wi,


.

Now

is

linear

and

is

the direct

sum

of (nonzero) T-invariant

V = Wi
Let
Ti denote the restriction of

Wr

and

T{Wi)

cWi,

i^l,...,r

Ti or

is

said to be the

T to Wi. Then T is said to be decomposable into the operators Also, the subTr. direct sum of the Ti, written T = Ti

spaces Wi, ...,Wr

are said to redvxe

Tor to form a T-invariant direct-sum decomposition of F.

reduce an operator T:V-^V; Consider the special case where two subspaces U and = 2 and dim = S and suppose {ui, u^} and (wi, W2, ws) are bases of [/ and say, dim C/ respectively, then respectively. If Ti and T2 denote the restrictions of T to C7 and

T2{wi)
Ti
(ui) '
'^

Tl (U2) ^
-^

= =

anUi
0.21^1

ai2U2

T2{W2)

rr,

a22U2

T2(W3)

= bnWi + = &21W1 + = bsiWl +


,
,

h\2W2

+ +

bi3W3 &23W3

&22W2
U

r,

,1,

I.

hz2W2

+
I

J> . O33W3

Hence
'&n
&21 ^31

A =

an ""
f
ttl2

a2i "''
0^22

and

B =

612
I

^22 &23

b
^"33
1

613

{mi, M2, wi, W2, wz) are matrix representations of Ti and Ta respectively. By the above theorem = T,{Ui) and r(Wi) = r2(Wj), the matrix of T in this basis is Since r(tti) is a basis of V.

the block diagonal matrix

A generalization
Theorem
10.5:

of the above

argument gives us the following theorem.


is

and V is the direct sum of T-invariant suba matrix representation of the restriction of spaces Wu , Wr. If Ai is T to Wi, then T can be represented by the block diagonal matrix
Suppose

T:V^V

linear

[Ai

...

M
The block diagonal matrix
direct

A2

...

...

A,
.,

M with
. .
.

sum

of the matrices Ai,

diagonal entries Ai, = Ai Ar and denoted by


.
.

A.

is

sometimes called the

Ar.

CHAP.

10]

CANONICAL FORMS

225

PRIMARY DECOMPOSITION
The following theorem shows that any operator is decomposable into operators whose minimal polynomials are powers of irreducible pols^omials. This is the first step in obtaining a canonical form for T,

T:V^V

Primary Decomposition Theorem


polynomial

10.6:

Let
m{t)

T:V^V

be a linear operator with minimal

/i(f)">/2(t)"^... /.(*)"'

where the
direct
fi{T)"'.

fi{f)

sum

are distinct monic irreducible polynomials. of T-invariant subspaces Wi, .,Wr where Wi
. .

Moreover,

/i(;t)i

is

V is the the kernel of the minimal polynomial of the restriction of


Then
is

to Wi.

Since the polynomials /i(i)"* are relatively prime, the above fundamental result follows (Problem 10.11) from the next two theorems.

Theorem

10.7:

Suppose T:V^V is linear, and suppose f{t) = git)h(t) are polynomials such that f{T) = and g{t) and h(t) are relatively prime. Then V is the direct sum of the T-invariant subspaces U and W, where U = Ker g{T) = Ker h{T). and

Theorem

10.8:

In Theorem 10.7, if f{t) is the minimal polynomial of T [and g{t) and h{t) are monic], then g{t) and h{t) are the minimal polynomials of the restrictions of T to U and respectively.

We

will also use the

primary decomposition theorem

to prove the following useful

characterization of diagonalizable operators.

Theorem

10.9:

linear operator T -.V has a diagonal matrix representation if and only if its minimal polynomial m{t) is a product of distinct linear polynomials.

^V

Alternate

Form

of

Theorem
Suppose

if its

10.9: A matrix A is similar to a diagonal matrix if and only minimal polynomial is a product of distinct linear polynomials.

Example

10.4:

A#

is

a square matrix for which


if

A^

I.

Determine whether or not


(i)

is

similar to a diagonal matrix


field C.

is

a matrix over

the real field R,

(ii)

the

complex

Since A^ - I, A is a zero of the polynomial f(t) = t^-1 = {t- l){t^ +t + The minimal polynomial m(t) of A cannot be t 1, since A ' I. Hence
m{t)

l).

t2

or

m(t)

t^

Since neither polynomial is a product of linear polynomials over R, A is not diagonalizable over R. On the other hand, each of the polynomials is a product of distinct linear polynomials over C. Hence A is diagonalizable over C.

is termed nilpotent if T" = for some positive integer n; k the index of nilpotency of T if T'' but T''-^ = 0. Analogously, a square matrix A is termed nilpotent if A" = for some positive integer n, and of index fc if A'' = but yj^k-i ^ Clearly the minimum polynomial of a nilpotent operator (matrix) of index k is m{t) f"; hence is its only eigenvalue.
:

NILPOTENT OPERATORS A linear operator T F - V


call

we

The fundamental

result

on nilpotent operators follows.

Theorem

10.10:

Let T:V-^V be a nilpotent operator of index k. Then T has a block diagonal matrix representation whose diagonal entries are of the form

226

CANONICAL FORMS
1
1

[CHAP. 10

N
.
.

they are

except those just above the main diagonal where are of orders of order k and all other ^ k. The number of of each possible order is uniquely determined by of all orders is equal to the nullity T. Moreover, the total number of
(i.e. all

entries of A^ are
1).

There

is

at least one

of T. of order i is In the proof of the above theorem, we shall show that the number of mi+i Mi- 1, where mj is the nullity of T\ We remark that the above matrix is itself nilpotent and that its index of nilpotency is of order 1 is just the 1 x 1 zero equal to its order (Problem 10.13). Note that the matrix

2mi

matrix

(0).

JORDAN CANONICAL FORM


An operator T can be put into Jordan canonical form if its characteristic and minimal polynomials factor into linear polynomials. This is always true if K is the complex field C. In any case, we can always extend the base field Z to a field in which the characteristic and minimum polynomials do factor into linear factors; thus in a broad sense every operator has a Jordan canonical form. Analogously, every matrix is similar to a matrix in Jordan canonical form.
Theorem
10.11:

Let T:V -> be a linear operator whose characteristic and minimum polynomials are respectively m{t) = (i - Ai)"' ...{t- Xr)and A{t) = (t- Ai)"' ...(*- XrY'

where the

Then T has a block diagonal matrix Ai are distinct scalars. representation / whose diagonal entries are of the form
/A;

1
Ai

... 1 ...

0\

/ ij

Ai

Ai/

For each
(i)

A.

the corresponding blocks Ja have the following properties:


at least one Ja of order mi;
is

There

is

all

other Ja are of order

^ mi.

(ii)

The sum of the orders of the Ja

m.
Ai.

(iii)

The number

of Ja equals the geometric multiplicity of

(iv)

The number of Ja of each

possible order is uniquely determined

by

T.

The matrix J appearing


operator T. Observe that
Ai

in the above theorem is called the Jordan canonical form of the block Ja is called a Jordan block belonging to the eigenvalue Ai. diagonal
...
Ai

1
Ai

^\

Ai

..

...

..

+
.
.

Ai

...
.
.

Ai Ai

'

.. ..

..

CHAP.

10]

CANONICAL FORMS

227

That

is, Jtj

Xil

where is the nilpotent block appearing in Theorem 10.10. In fact, we prove the above theorem (Problem 10.18) by showing that T can be decomposed into operators, each the sum of a scalar and a nilpotent operator.
Example 105:
Suppose the characteristic and minimum polynomials of an operator T are respectively
A()

(f-2)4(t-3)3

and
is

m{t)

(-2)2(t-3)2

Then the Jordan canonical form of T

one of the following matrices:

or

The first matrix occurs if T has two independent eigenvectors belonging to its eigenvalue 2; and the second matrix occurs if T has three independent eigenvectors belonging to 2.

on a vector space V of finite dimension over K. Suppose V and v ^ 0. The set of all vectors of the form f{T){v), where f{t) ranges over all polynomials over K, is a T-invariant subspace of V called the T-cyclic subspace of V generated by v;we denote it by Z{v, T) and denote the restriction of T to Z{v, T) by r. We could equivalently define Z{v,T) as the intersection of all T-invariant subspaces of V containing v.

CYCLIC SUBSPACES Let r be a linear operator

GV

Now

consider the sequence


V, T{v),

T\v), T\v),

of powers of T acting on v. Let k be the lowest integer such that T''{v) bination of those vectors which precede it in the sequence; say,
T^iv)

is

a linear com-

-ttfc-i

T'^-^v)

...

aiT{v)

aov
ao
(v)

Then
is

m(i)

t"

ak-it^-^

ait

the unique monic polynomial of lowest degree for which mv(T) T-annihilator of v and Z{v, T).

0.

We

call m(i) the

The following theorem

applies.

Theorem

10.12:

Let Z(v,
(i)

T),

T^ and m(i) be defined as above.


{v, T{v), ..., r'=-i (v)} is

Then:

The set

a basis of Z{v, T); hence dim Z{v, T)


is

fe.

(ii)

The minimal polynomial of T The matrix representation

m(i).

(iii)

of Tv in the above basis is

228

CANONICAL FORMS

[CHAP. 10

tto
-ai
-tti

1 1

aic-2 aic-i

The above matrix C

is called

the companion matrix of the polynomial m(t).

RATIONAL CANONICAL FORM

We

In this section we present the rational canonical form for a linear operator T:V^V. emphasize that this form exists even when the minimal polynomial cannot be factored into linear polynomials. (Recall that this is not the case for the Jordan canonical form.)

Lemma

10.13:

Let

T:V-*V

be a linear operator whose minimal polynomial

is /(*)"

where

f{t) is

a monic irreducible polynomial.

Then V

is

the direct

sum

V =
/(*)"', /()"^

Z{vi, T)

Zivr, T)

of T-cyclic subspaces Z{Vi, T) with corresponding T-annihilators

-, fit)"',
V

n =

Ml

^ %2 -

- Wr
same number

Any

other decomposition of

into jT-cyclic subspaces has the

of components and the same set of T-annihilators.

above lemma does not say that the vectors vi or the T-cyclic subuniquely determined by T; but it does say that the set of T-annihilators spaces Zivi, T) are are uniquely determined by T. Thus T has a unique matrix representation

We emphasize that the

Cr

where the

are companion matrices.

In fact, the Ci are the companion matrices to the

polynomials

/(*)"*.

Using the primary decomposition theorem and the above lemma, we obtain the following fundamental result.

Theorem

10.14:

Let

T:V^V
/{()

be a linear operator with minimal polynomial


m{t)

fi{tr^f2{tr- ... fsitr-

where the

are distinct monic irreducible polynomials. unique block diagonal matrix representation

Then T has a

'Cn
Clrj

Cs

where the C are companion matrices.


panion matrices of the polynomials

In particular, the C are the com-

/i(t)"
rris

where

mi =

nil

ni2

ni: \'

TCsl

^52

Msr.

CHAP.

10]

CANONICAL FORMS
is

229

The above matrix representation of T


nomials
/i(i)"

called its rational canonical form.

The

poly-

are called the elementary divisors of T.


10.6:

Example

Let V be a vector space of dimension 6 over R, and let T be a linear operator whose minimal polynomial is m{t) = (t^-t + 3)( - 2)2. Then the rational canonical form of T is one of the following direct sums of companion matrices:
(i)

C(t2-( +
C{f2-t + C(t2-t +

3)

(ii)

3) 3)

(iii)

C(2-t

+ 3)

C((<-2)2)
2)2)

C((t-2)2)
C((t-2)2)

C((t-

C(f-2)

C(t-2)
is,

where

C(f(t)) is the

companion matrix of

/(t);

that

/O

-3

.L__

12
(i)

(ii)

(iii)

QUOTIENT SPACES Let F be a vector space


in V,

we

write v

+W

over a field and let T7 be a subspace of V. for the set of sums v + w with w GW: V

If

is

any vector

These sets are called the cosets in V. partition V into mutually disjoint subsets.
Example
10.7:

W of W

{V

+ w:

wGW)
10.22) that these cosets

We

show (Problem

W be the subspace of R2 defined W = b): a=b} That W the line given by the equation x y = We can view V + W as translation of the
Let

by

{(a,

is,

is

0.

a,

line,

obtained by adding the vector v to each point in W. As noted in the diagram on the right, v is also a line and is parallel to W. Thus the cosets of in R2 are precisely all the lines parallel to W.

+W

new

In the next theorem we use the cosets of a subspace vector space; it is called the quotient space
10.15:

W of a vector space

to define

ofVhyW and is denoted by

V/W.
cosets of

Theorem

Let
tion
(i)

W he a subspace of a vector space over a field K.


and scalar multiplication:
{u

Then the

WinV form a vector space over K with the following operations


+ W) +
{v

of addi-

+ W) =

{u

+ v) +

W
it is first

(ii)

kiu

+ W) = ku + W, where kGK.

We note that, in the proof of the above theorem, operations are well defined; that is, whenever u +
(i)

W = u' + W
-

necessary to show that the = v' + W, then and v +

{u

+ v) +

(u'

+ V') +

and

(ii)

ku+W

ku'

+ W,

for any

k&K

230

CANONICAL FORMS
In the case of an invariant subspace,

[CHAP. 10

we have

the following useful result.

Theorem

10.16:

Suppose

is

a subspace invariantunder a linear operator^

T V -
:

V.

Then T induces a linear operator f on V/W defined by T{v -\-W) = T{v) + W. Moreover, if T is a zero of any polynomial, then so is T. Thus the minimum polynomial of T divides the minimum polynomial of T.

Solved Problems

INVARIANT SUBSPACES 10.1. Suppose T:V -^V is linear. Show


(i)

{0},

(ii)

V,

(iii)

kernel of T,

(iv)

that each of the following image of T.


invariant under T. invariant under T.
since the kernel of

is

invariant under T:

(i)

We
Let

have

T(Q)
G.

{0};

hence {0}
V; hence

is is

(ii)

For every v

V, T(v)

(iii)

u e Ker
is

T.

Then

^(m)

=
v

S Ker T

T
v

is

a subspace of V.

Thus

Ker T
(iv)

invariant under T.

Since T{v) G Im T for every T is invariant under T.

eV,

it

is

certainly true if

G Im

T.

Hence the image of

10.2.

Suppose {Wi}
Suppose V

is

a collection of T-invariant subspaces of a vector space V.

Show

that

the intersection
Thus tIv)

W=

eW;

then

HiWi is v e Wi

also T-invariant.
for every i. Since is T-invariant.

Wj

is

T-invariant,

T(v)

G Wj

for every

i.

:W=

riiWi and so

10.3.

Prove Theorem
nomial.
of /(r),

10.2:

Let

T:V-^V
is

Then the kernel

of f{T)
f(T)(v)

be any linear operator and invariant under T.


0.

let f{t)

be any poly-

Suppose V
i.e.

Ker/(r),

i.e.

We

need to show that

^(i;)

also belongs to the kernel

f(T)(T(v))

0.

Since f{t)t=tf(t),
f(T)T(v)

we have f(T)T=Tf(T). Thus

Tf(T){v)

T(0)

as required.

10.4.

Find

all

invariant subspaces of

A -

(
J

viewed as an operator on

R^.

First of all, subspaces, then

we have
it

that R^ and {0} are invariant under A. Now if A has any other invariant must be 1-dimensional. However, the characteristic polynomial of A is
A(t)

\tI-A\

t-2
-1
t

=
2

t2

Hence

A has no eigenvalues (in R) and so A has no eigenvectors. But the 1-dimensional invariant subspaces correspond to the eigenvectors; thus R2 and {0} are the only subspaces invariant under A.
10.3:

10.5.

Prove Theorem

Suppose

W
W.

is

an invariant subspace of T:V-^V.

Then T

has a block diagonal matrix representation


tion of the restriction

where
^
J

is

a matrix representa-

of

to

^
it

We We

choose a basis {wi, ....wj of

and extend

to a basis {w^,

.,Wr,Vi

v^}

of V.

have

CHAP.

10]

CANONICAL FORMS
A. A.

231

T{W2)
T{Wr)

T(W2)

= = = = =

a2iWi

+
+ + + +

Oar^r

r(Wr)
^(i;!) T{V<;^

a^iWi
b^jWi
621'l

+ + +
+

ftrr^r

bi^w^
hr'^r

+ +

Cij-Wj

C21^1

+ +
+

+ +
+

c^^v^ Cas^j

^^(ys)

6slWl

ftsr^r

CjiVi

Css^s

But the matrix of T


of equations.

in this basis is the transpose of the

matrix of
(

coefficients in the
)

above system

(See page 150.)

Therefore

it

has the form

the matrix of coefficients for the obvious subsystem.

where A C^ By the same argument,

is

the transpose of
is

the matrix of

relative to the basis {Wj} of

W.

10.6.

Let

T denote the restriction of an operator T T{w) T{w) for every w GW. Prove:
For any polynomial
f(t),

to

an invariant subspace W,

i.e.

(i)

/(f)(w)

= fiT)(w).

(ii)
(i)

The minimum polynomial of T divides the minimum polynomial of T. If /(*) = or if f{t) is a constant, i.e. of degree 1, then the result clearly holds. Assume deg/ = n > 1 and that the result holds for polynomials of degree less than n. Suppose that + ajt + oo /{*) = at" + a_i f-i +

Then

f(T){w)

= = = =

(ar

+ o_ir"-i + + +

(ah-i)(T(w))
(ar-i)(r(w))

+ ao/)(w) + + aoI)(w) (o_ir-i + + oo/)(w)


(a_i rn-i

fiTHw)

(ii)

Let m(t) denote the


for every
of

w S W;

that

minimum polynomial of T. Then by (i), m(T)(w) = m{T)(w) = 0(to) = is, T is a zero of the polynomial m(t). Hence the minimum polynomial

divides m{t).

INVARIANT DIRECT-SUM DECOMPOSITIONS


10.7.

Prove Theorem
i

10.4:
. .

Suppose
Wi^}
is

l,

.,r,
if

{wii,

.,

.,Wr are subspaces of V and suppose, for Wi, a basis of Wu Then V is the direct sum of the Wi if
. . .
. .

and only
,
.

the union
,,

B =

{Wu,

Win,,

Wrl,

.,

Wm.)

IS

a basis of V.
Suppose
V

where

B is a basis of V. Then, for any v &V, = duWii + + aijWij + + a^iW^i + Wj = a-ai^n + + ai.Wi. G PTj. We next show

+ a^w^^ =
sum
where W;

Wi
is

+ W2+
unique.

+ w^
Suppose

that such a

w'l

W2

Since {wji,

. ,

Win.} is a basis of Wi,

611W11

+ + w'r = b^Wn + w[ + 6ijWi^ +


S Wi
and so

6i.W{.

ftrl^rl
j.

is

Since B is a basis of V, Oy = 6y, for each i and each unique. Accordingly, V is the direct sum of the PFj.

Hence

+ Kn^^m^ w^ = w,' and

so the

sum

for v

where Wj

+ w, Conversely, suppose V is the direct sum of the W^. Then for any v GV, v = Wj + G PFj. Since {Wy.} is a basis of Wi, each w^ is a linear combination of the Wy. and so v
Thus

is a linear combination of the elements of B. independent. Suppose

spans V.

We now

show that

is

linearly

"llWli

! Win,

an^ri

232

CANONICAL FORMS
Note that
aa'ii

[CHAP. 10

+
is

+ ain.Wm. G
aaWti

W'j.

We
+

also have that

=
for
i

+ 0+---+0 where

Wi-

Since

such a sum

for

unique,

ai.Wi.

=
all

=
0.

1,

The independence of the bases {wy.} imply that and hence is a basis of V.

the a's are

Thus

is

linearly independent

10.8.

Suppose
(i)

T:V^V
is

is

linear

direct-sum decomposition
m{t)

and suppose T = Ti ^2 V = U W. Show that:

with respect to a T-invariant


m{t), mi{t)

the least

are the
(ii)

common multiple of mi{t) and m2{t) where minimum polynomials of T, Ti and T2 respectively;
Ai(t)

and

in2{t)

A{t)

T, Ti
(i)

Ai{t) A2{t), where A{t), and T2 respectively.


10.6,

and

A2(t)

are the characteristic polynomials of

By Problem
Wi(()

each of

TOi(t)

and m^it) divides m(t).

Now
0.

suppose

f{t) is

Me

and m2(t); then f{Ti){U) C/ and w G W. Now


/(T) V

and /(r2)(W)

=
M

Let

vGV;
w =
is

then

a multiple of both v - u + w with

/(r)
f(t).

w +

f(T)

w =

/(Ti)
f{t),

/(T2)

=
common
multiple of

That

is,

is

a zero of

Hence m(t) divides

and so m{t)

the least

Wi(t) and m2{t).


(ii)

By Theorem

10.5,

T has

a matrix representation

M
=

where

and

are matrix

representations of T^ and T2 respectively.

Then, by Problem

9.66,

A(t)

\tI-M\

tl

-A
tl

-B

\tI-A\\tI-B\

Ai(t)A2(t)

as required.

10.9.

Prove Theorem

10.7:

Suppose

T:V-*V

is linear,

and suppose

f{t)

= g{t) h{t)

are

and g{t) and polynomials such that /(T) is the direct sum of the T-invariant subspaces = Kerh{T).

h{t)

are relatively prime.

and

where
since g(t)

C7

Then V Kerflr(r) and


are relatively

are T-invariant by Theorem 10.2. Note first that U and prime, there exist polynomials r(t) and s(t) such that
r(t) sr(t)

Now
1

and

h(t)

s{t) h(t)

= =

Hence for the operator


Let

T,

r(T) g(T)

s{T) h(T)

(*)

veV;
first

then by

(*),

r(T) g(T) v

8(T) h(T) v since

But the

term

in this

sum

belongs to

W KerhCT)
=
is

h(T) r{T) g(T) V

r(T) g{T) h(T) v

r(T)f(T)v
the

r(T)(iv

Similarly, the second term belongs to U.

Hence

sum

of

To prove that

V = U W, we
1;.

uniquely determined by

must show that a sum v Applying the operator r{T)g(T) to

U and W. u + w with u&V, w &W, v = m + w and using g(T)u =


=
r(r)s'(r)w

is 0,

we

obtain
r(r)ff(r)i>

r(T)g(T)u

__

r(r)fl'(r)w
0,

Also, applying

(*) to

alone and using h{T)

w =

we

obtain
r(2')flr(T)w
v.

w =
ilarly

r(r)flr{r)w

+ 8(T)h(T)w =

Both of the above formulas give us w = t(T) g(T) v and so w is uniquely determined by M is uniquely determined by v. Hence V V @W, as required.

Sim-

CHAP.

10]

CANONICAL FORMS

233

10.10.

Prove Theorem 10.8: In Theorem 10.7 (Problem 10.9), if f{t) is the minimal polynomial of T (and g{t) and h{t) are monic), then g{t) is the minimal polynomial of the restriction Ti of T to U and h(t) is the minimal polynomial of the restriction Tz of

rto W.
and h(T2)
Let mi(t) and mgCf) be the minimal polynomials of T^ and T2 respectively. = because U = Ker g(T) and = Kerh(T). Thus

Note that 9(Tj)

=
(1)

mi(t) divides g(t)

and

m2{t) divides h{t)

common multiple of mi(t) and nizit). But mi{t) and m2(t) are and h{t) are relatively prime. Accordingly, f(t) = mj(t) m,2(t). We also have that f{t) g(t) h(t). These two equations together with (1) and the fact that all the polynomials are monic, imply that g(t) mi(t) and h{t) = m^^t), as required.
10.9, f{t) is the least
g{t)

By Problem

relatively

prime since

10.11.

Prove the Primary Decomposition Theorem with minimal polynomial


-mit)

10.6:

Let

T 7 - F
:

be a linear operator

/l(i)"i/2(i^.../r(<)"'

where the

Then V is the direct sum fi{t) are distinct monic irreducible polynomials. of T-invariant subspaces Wi, ...,Wr where Wi is the kernel of fi{TY\ Moreover, /i(i)"' is the minimal polynomial of the restriction of T to Wu
is
1.

The proof proved for r


and Fi where

by induction on

r.

The case r

By Theorem

10.7

we can

write

= 1 is trivial. Suppose that the theorem has been V as the direct sum of T-invariant subspaces W^
.
. .

W^

Theorem

10.8, the

is the kernel of /i(r)"i and where V^ is the kernel of fziT)"^ By /r(r)"'. minimal polynomial of the restrictions of T to TFj and Vi are respectively /i(f)"i

and/2()"2 ... /r ()"'.

Denote the restriction of

to

V^ by

T^.

By

the inductive hypothesis, V^ is the direct

sum

of

subspaces W2, .,'W^ such that 'W^ is the kernel of /{(Ti)". and such that /((<)! is the minimal polynomial for the restriction of T^ to PT,-. But the kernel of fi{T)"i, for i = 2, .,r is necessarily contained in V^ since /;(*)"' divides /2(t)"2 /r(*)"''- Thus the kernel of /i(r)i is the same as the kernel of fi{T^^i, which is W^. Also, the restriction of T to W^ is the same as the restriction of T^ to Wi (for i = 2, .,r); hence /;(*)" is also the minimal polynomial for the restriction of T to WiThus V = is the desired decomposition of T.
. .
. .

WiW2

Wr

10.12.

Prove Theorem 10.9: tion if and only if


polynomials.
Suppose m{t)
is

A linear operator T.V^V


its

minimal polynomial m{t)

has a diagonal matrix representais a product of distinct linear

a product of distinct linear polynomials; say,


m{t)

(t-Xi){t-X2)

... (t-X,.)

where the

Xj are distinct scalars. By the primary decomposition theorem, V is the direct sum of subspaces Wi,...,Wr where Wj = Ker(7'-Xi/). Thus ii v e Wi, then (T-\iI){v) = or T(v) Xj-y. In other words, every vector in TFj is an eigenvector belonging to the eigenvalue Xj. By Theorem 10.4, the union of bases for Wi, This basis consists of eigenvectors ., W^ is a basis of V.
.
.

and so T

is

diagonalizable.

Xj,

Conversely, suppose T is diagonalizable, i.e. V has a basis consisting of eigenvectors of T. Then the operator ., Xj be the distinct eigenvalues of T.
f(T)

Let

maps each

basis vector into the polynomial

0.

= {T-\J)(T-X2l)...(T-Kl) Thus f{T) = and hence the minimum

polynomial m(() of

divides

(t-Xi)(i-X2)...(t-X,/)

Accordingly,

m,(t) is

a product of distinct linear polynomials.

234

CANONICAL FORMS

[CHAP. 10

NILPOTENT OPERATORS, JORDAN CANONICAL FORM 10.13. Let T:V^V be linear. Suppose, for vGV, T''{v) =
(i)

but f'-^v)

- 0.

Prove:

The

set

S =

{v, T{v), ..., T'^-^iv)}


-S is

is

linearly independent.

(ii)

The subspace

(iii)

W generated by T-invariant. The restriction T of T to W nilpotent of index k.


is

(iv)

Relative to the basis {T''-^{v),

.,T{v),v} of
1
..
.

W,

the matrix of

is

of the

form

..

.. ..

Hence the above


(i)

/c-square

matrix

is

nilpotent of index k.

Suppose
av

di T{v)

02

T^v)

+ a^.^n-Hv)

(*)

and using r'=(i;) = 0, we obtain aT'<^-i(v) = 0; since Ti'-'^(v) ^ 0, a - 0. Now applying T^-z to (*) and using P'iv) = and a = 0, we find a^ r'=-i(i;) = 0; hence Next applying T''-^ to (*) and using T<^(v) = ! = 0. and a = ai = 0, we obtain a2T^~^{v) = 0; hence Og = 0. Continuing this process, we find that all the a's are 0; hence
Applying
T'^-i to (*)

S
(ii)

is

independent.

Let

veW.

Then
V

= =

bv

biT(v)

biT^v)

+ +

b^_iT'^-Hv)

Using THv)

0,

we have that
T{v)

bT{v)

biT2(v)

+ +

b^.^^T'^-H'")

Thus
(iii)

is

T-invariant.

By

hypothesis

T''{v)

0.

Hence, for

= =

k1,
r''

Tk{Ti(v))

+ (i;)

=
T'^

That

is,

applying

T'^ to
fc.

each generator of
the other hand,

W, we

obtain 0; hence

=
T

and so T
is

is

nilpotent

of index at most

On

Tf^-^v)

T''-^v)

= 0;

hence

nilpotent of index

exactly

fc.

(iv)

For the basis

{T'<'-^v), Ti'-^v),

.,T{v),v} of
rk(i;)

W,

T(T^-^(v))

= =

r(rfc-3(^))

r'=-2(-u)

T(T{y))

T'^(v)

T(v)

T(v)

Hence the matrix of T

in this basis is
1 1

CHAP.

10]

CANONICAL FORMS
Let

235

10.14.

Let T-.V-^V be linear.


(ii)

U=

KerT' and

W = KerT+\
T'+Mm)
hence

Show

that

(i)

UcW,
0.

T{W) C

U.

(i)

Suppose ueU = Kern Then THu) = and so MGKerr* + = W. But this is true for every m G f/;
i

T(,Ti(u))

T(0)

Thus

UcW.
r'+Mw) = r*(r(w)) =
r(0)

(ii)

Similarly, if

wG

W'

= Ker

r*+i,

then

T'+Mw) =

0.

Thus

and so r(W') c U.

10.15.

Let r F ^ F be linear. preceding problem,


:

Let

XcY cZ.
{Ml,
. .
.

X = Ker r*-^ Y = Ker 7*-^


Suppose
.

and

Z = Ker T*.
Wi,
. . .

By

the

{Ml, .... Mr},

Mr, Vi,

Vs},

{Mi,

,Ur, Vi,

Vs,

Wt}

are bases of X,

and

Z
is

respectively.

Show
Mr,

that
. .

s =
is

{Ml,

.,

r(wi),

.,

r(M;t)}

contained in

and

linearly independent.
T(Z)

By

the preceding problem,


exists a relation

c Y and hence S CY. Now suppose S


a^Mr

is linearly

dependent.

Then there

aiUi

6i

T(wi)

b^

T{wt)

where at least one coefficient is not zero. Furthermore, since must be nonzero. Transposing, we find 6fc
bi T{wi)

{u^} is

independent, at least one of the

6t

T(wt)

- aiUi +

5iWi

a^u^

X =

Ker P'^

Hence

Ti-^(biT(wi)

btT(wt))

Thus
Since

r->(6iWi

+ +

6tWt)

and so

6,Wt

G r = KerT*-!

{mj, Vj} generates Y, we obtain a relation among the Mj, ij and Wj; where one of the coefficients, one of the 6^, is not zero. This contradicts the fact that {Mj, Vj, w^} is independent. Hence S must also be independent.
i.e.

10.16.

Prove Theorem 10.10: Let T.V^V be a nilpotent operator of index k. Then T has a block diagonal matrix representation whose diagonal entries are of the form
1
. .

N
. .

There is at least one N of order k and all other N are N of each possible order is uniquely determined by T. N of all orders is the nullity of T.
Suppose
for
i

of orders

^ k.

The number

of

Moreover, the total number of

dimy =

n.

Let
is

Wi = Ker
of index k,

T,

1,.. .,k.
10.17,

Since

W2 = Ker ra W^ = Ker T". Set m^ = dim W^j, W^ = V and Wj^-i # V and so m^_i <m^ n. By

Problem

WiCW^C
Thus, by induction,

CW^ = V
V
such that
{u^,
. .

we can

choose a basis {mj,

.,m} of

.,

> is

a basis of

PFj.

We now choose a new basis for V with respect to which T has the desired form. It will be convenient to label the members of this new basis by pairs of indices. We begin by setting
v{l,k)

u^^_^ +

i,

w(2,

fc)

=M^_j

2,

...,

y(mfc-tfc_i,

fc)

=Mj^

236

CANONICAL FORMS
and setting

[CHAP. 10

v{l,k-l) = Tv{l,k), v{2,k-l) = Tv(2,k),

...,

vim^-m^.-i, k-1)

^ Tv{m^-m^^i,k)

By

the preceding problem,

Si
is

{Ml

...,

u^^_^, v{l,k-l),

...,

vCmfc-mfc^i, fe-1)}

ments

a linearly independent subset of W^-i(if necessary) which we denote by

We

extend Sj to a basis of Wfc-i by adjoining

new

ele-

y(mfe-m;,_i

l,

fc-1),

v(m^-mk-i +

2,

k~l),

...,

v(m^_i-- m^^^tk-V)

Next we

set
v(l,

k-2) =

Tv(l, k

v(in^_i

- 1), v(2, k-2) = Tv{2, k-1), ..., - mfc_2. k-2) = Tv(m^_i - m^.g, - 1)
fc

Again by the preceding problem,


Si
is

{Ml,

...,

u^^_^, v{l,k-2),

...,

u(mfc_i-w^^2. ^^-2)}
to a basis of TFfc_2

a linearly independent subset of W|c-2 which elements

we can extend

by adjoining

vim^-.i-

711^-2

1,

k-2), y(mfc_i-mfc_2+2, fc-2),

...,

vim^^^-ink-s, ^-Z)

Continuing in this manner we get a new basis for


as follows:
v{l, k), ...,
...,
'y(mfc

which for convenient reference we arrange

TOfc_i,

A;)

v{l,k-l),

v{m^-mk_i,k-l),
v(mfc-mfe_i,
i;(mfc-mfc_i,
2), 1),

...,

u(mfc_i

- Wfc-a,

fe

- 1)
..., ...,

i;(l, 2),

..., ...,

..., ...,

u(mfc_i u(mfe_i

- mfc_2,
-mfe_2,

2),
1),

^(ma-mi,
i;(m2-mi,

2)
1),

v(l, 1),

...,

v(mi,

1)

The bottom row forms a basis of Wi, the bottom two rows form a basis of W2, etc. But what is important for us is that T maps each vector into the vector immediately below it in the table or into if the vector is in the bottom row. That is,
Tv(i,i)

=
T

(v{i, j
^

1)

for
.

}
)

for
will

> =

1
.

1
if

Now

it is

clear (see Problem 10.13(iv)) that

have the desired form

the v{i,]) are ordered

lexicographically: beginning with v(l, 1) and moving up the first column to ^(l, k), then v{2, 1) and moving up the second column as far as possible, etc.

jumping

to

Moreover, there will be exactly


m^,

mfc_i

diagonal entries of order k


Wfc_2

(mfc_i

m;;_2)

(m^

mfc_i) =

2mfc_i

m^

diagonal entries of order

fc

2m2
2mi

mi m^ 7^2

diagonal entries of order 2 diagonal entries of order


.
. . ,

m^ are uniquely as can be read off directly from the table. In particular, since the numbers mj, determined by T, the number of diagonal entries of each order is uniquely determined by T. Finally,
the identity

mi = (mfc-mfc_i) +
shows that the nullity mj of T

(2mfc_i
is

-m^ -TOfc-a) +
number

+ (2m2-mi-m3) + (2mi-m2)

the total

of diagonal entries of T.

10.17.

Let

A =

^00000/
hence

0011l\ 00000/
2.

1\

Then

A^

/ooooo looooo
\0

/O

and

A3

0;

0/

A
is

is

nilpotent of index

Find the nilpotent matrix

in canonical

form

which

similar to A.

CHAP.

10]

CANONICAL FORMS
Since

237

is

2.

Note that rank

Accordingly

nilpotent of index 2, contains a diagonal block of order 2 and none greater than A - 2; hence nullity of = 5 - 2 = 3. Thus contains 3 diagonal blocks. must contain 2 diagonal blocks of order 2 and 1 of order 1; that is,

M=

_0_^j^_0^
|_0_^

OOlOllo
'_0
I

10.18.

Prove Theorem
By
the

10.11,

page 226, on the Jordan canonical form for an operator T.


.

r =

Ti

primary decomposition theorem, T is decomposable into operators T-^, ., r^, where (t Xj)"i is the minimal polynomial of Tj. Thus in particular,
.

T^,

i.e.

(Ti

- Xi/)", =
l
r,

0,

...,

(T^-\J)r-r

(i

Set Ni

= Ti- Xj7.

Then for

Ti

= Ni+

\I,

where

Nr't

=
iV{,

That
since

is,

Tj is the
is

sum

of the scalar operator Xj/ and a nilpotent operator


Tf.

which

is

of index mj

(t

Xj)*"!

the minimal polynomial of

by Theorem 10.10 on nilpotent operators, we can choose a basis so that iVj is in canonical In this basis, Ti = N^ + \I is represented by a block diagonal matrix Mj whose diagonal entries are the matrices J^. The direct sum J of the matrices Mj is in Jordan canonical form and, by Theorem 10.5, is a matrix representation of T,
form.
Lastly we must show that the blocks Jy satisfy the required properties. Property (i) follows from the fact that A^j is of index mj. Property (ii) is true since T and J have the same characteristic polynomial. Property (iii) is true since the nullity of Ni= Tj \I is equal to the geometric multiplicity of the eigenvalue Xj. Property (iv) follows from the fact that the Tj and hence the N^ are uniquely determined by T.

Now

10.19.

Determine

all

possible Jordan canonical

forms for a linear operator

T:V ->V

whose

characteristic polynomial is A{t)


Since
5
t

(t

2)^(t 5)^.
forms are

has exponent 3 in
twice.

A(t),

must appear three times on the main diagonal. Similarly

must appear
2

Thus the

possible Jordan canonical

11
2
1

2
5
1

!_2
5

5
(ii) (iii)

(i)

2
1

2,'

L?.i

5
I

-(

5
(vi)

(iv)

(V)

238

CANONICAL FORMS

[CHAP. 10

10.20.

Determine all possible Jordan canonical forms / for a matrix of order 5 whose minimal polynomial is in(t) {t 2y.
J must have one Jordan block of order 2 and the others must be of order 2 or are only two possibilities:
2
1
I

1.

Thus there

.i^_.
I

1
I
i

'

!__ + __
I

Note that

all

the diagonal entries must be 2 since 2

is

the only eigenvalue.

QUOTIENT SPACE AND TRIANGULAR FORM


10.21.

Let
(i)

W he a subspace of
(ii)

uGv + W,
Suppose

u-v GW,
vGW.

a vector space V. Show that the following are equivalent: (iii) v + W.

Gu

uG v + W.
u

Conversely, suppose u Thus (i) and (ii) are equivalent.


have: also equivalent.
also

Then there exists w^eW such that u = v + Wq. Hence u v = WoSW. Then u v = Wq where Wq S 1^. Hence u = v + WgSv + W.
iff

We

vGW

(m v)=v uGW

iflf

& u+ W.

Thus

(ii)

and

(iii)

are

10.22.

Prove:
(i)

The

cosets of PF in

partition

into mutually disjoint sets.

That

is:

any two cosets u +


each v

W and
if

+W

are either identical or disjoint; and

(ii)

gV

belongs to a coset; in fact, v

Gv + W.
so {v

Furthermore, u
for any
Let

+W

+W

and only
v

if

u vGW, and

+ w) +

w GW.
1)

= v + E v + W which proves Now suppose the cosets u+W and v + W are not disjoint; say, the vector belongs to both u+W and v + W. Then u xGW and x vGW. The proof of is complete if we show that u + W = v + W. Let M + Wq be any element in the coset u+W. Since u x, x v and Wq belong
e
V.

Since

G W, we have

(ii).

(i)

to

W,
(u

+ Wq)

(u

x) +
is

{x

v) +

Wo S

W
+ W.
if

Thus u + W(,Gv
is

contained in

u+ W

and hence the coset = v + W. and so u+

u+W

contained in the coset v

Similarly v

The last statement follows from the fact that by the preceding problem this is equivalent to u v G W.

u+W - v + W

if

and only

uGv + W,

and

10.23.

Let be the solution space of the homoDegeneous equation 2x + By + 4:Z = 0.


scribe the cosets of
TF
is

W in R^.
W

a plane through the origin O = (0, 0, 0), are the planes parallel to W. Equivalently, the cosets of are the solution sets of the family of equations

and the cosets of

2x

Sy

4z

k,

kGR
=
(a, b, c),

In particular the coset v + W, where v is the solution set of the linear equation

2x
or
2(x

Sy

Az

2a

36

4c

-a) +

3(y

- 6) +

4(2

- c) =

CHAP.

10]

CANONICAL FORMS

239

10.24.

Suppose is a subspace of a vector space V. Show that the operations in Theorem = u' + 10.15, page 229, are well defined; namely, show that if u + and v +

v'

+ W,
(i)

then
{u

+ v) +

{u'

+ V') +

and

(ii)

ku

ku'

+ W,

for any

k&K

(i)

(ii)

u + W ^ u' + W and v + W = v' + W, both u u' and v v' belong to W. But then + v) - (u' + v') - {u- u') + {v- v') e W. Hence (u + v) + W = (m' + v') + W. Also, since u u' S W implies k(u u') G W, then ku ku' = k(u u') G W; hence ku+W = ku' + W.
Since
(u

10.25.

Let

F be a vector space and W a subspace of V. Show that the natural map


by
rj{v)

ij

F - V/W,

defined

+ W,

is linear.

For any u,v ^

and any k G K, we have

v{u

+ v) =

+
kv

W
+

W+V+W
+ W) =
k

v{u)

v{v)

and
Accordingly,
r)

v{kv)
is

k(v

ri(v)

linear.

10.26.

and he a subspace of a vector space V. Suppose {wi, Wr} is a basis of the set of cosets {vi, Vs}, where Vj = Vj + W, is a basis of the quotient space. + Show that B = {vi, .,Vs, Wi, Wr} is a basis of V. Thus dim V = dim
Let
.

W W

dim (7/TF).
Suppose M

y.

Since {5^}

is

a basis of

V/W,
di'i'i

u = u
Hence u

a.2'U2

ttj^s

aiVy +

+ a^v^ + w u B

where
a^Vi

w G W. Since {w;} + + a^Vg + bjWi +

is

a basis of

W,

b^w^

Accordingly,

generates V.
is linearly

We now
Then
Since {Vj} Since {wj of y.
is
is

show that

independent.

Suppose

e^Vi

CgVs

djWi
c^Vs

+ -

+
=

dfWr

(1)

Cj'Di

independent, the c's are all 0. Substituting into (1), we find djWi + + d^w^ = 0. independent, the d's are all 0. Thus B is linearly independent and therefore a basis

10.27.

Prove Theorem 10.16:

a subspace invariant under a linear operator f on V/W defined by f{v + PF) = T{v) + W. Moreover, if T is a zero of any polynomial, then so is T. Thus the minimum polynomial of T divides the minimum polynomial of T.

Suppose

is

T:V^V.
We

Then T induces a

linear operator

first

show that f

is

well defined,

i.e.

if

u+W

v + Accordingly,

then

u-vGW
T{u+W) =
is linear.

and, since
T{u)

W
=

u+W = v + W
is

then

T-invariant,

T(u -v)

t(u+W) = f(v + W). If = T(u) - T{v) G W.

T(v)

T(v

+ W)

as required.

We

next show that


t{{u

t
(v

We
f(u

have

+W) +
=

W))

= =

+ v + W) = T(u + v) + W = T(u) + Tiv) + T(u) + W+ T(v) + W = f{u + W) + T(v + W)


T{ku)

and
f{k{u

W))

f(ku

+ W) =

W=

kT(u)

k{T(u)

+ W) = kf(u+ W)

Thus f

is

linear.

240

CANONICAL FORMS
Now, for any
coset

[CHAP. 10

IF in

VIW,
T(T(u))

f2(u+W) = THu) +
Hence T^

= =

W
+

t{T{u)

+ W) ^ t(f{u+W)) = t^u+W)

T^.

Similarly T"

T" for any

n.

Thus for any polynomial

/()

= =

at"

+
+

ao

= =

2 afi,
"2 diiTKiA

HT)(u+W)

= =

f(T)(u)

W
if

'^.a^Tiiu)

'^)

^ajFCw+W)
Accordingly,
is

=
T
is

^aifi(u+W)
a root of
f{t)

and so 7(r)
root of
f(t).

/(r).

= {^ a.ifi)(u + W) = f{f)(u+W) = W = /(f), i.e. f is also then 7(f) =

Thus the theorem

proved.

10.28.

Prove Theorem 10.1: Let T .V -^V be a linear operator whose characteristic polynomial factors into linear polynomials. Then V has a basis in which T is represented by a triangular matrix.
The proof
tion of
is

by induction on the dimension of V.


is

If

dim

V=

1,

then every matrix representa-

is

a 1 by 1 matrix which

triangular.

Now suppose dim V 1 and that the theorem holds for spaces of dimension less than n. Since the characteristic polynomial of T factors into linear polynomials, T has at least one eigenbe the 1-dimensional subvalue and so at least one nonzero eigenvector v, say T(v) a^^v. Let

n>

space spanned by
also that

v.

Set

V = VIW.

is

invariant under T.

= to - 1. Note Then (Problem 10.26) dim V = dim V dim T induces a linear operator f on V whose By Theorem 10.16,

minimum polynomial divides the minimum polynomial of T. Since the characteristic polynomial of r is a product of linear polynomials, so is its minimum polynomial; hence so are the minimum and characteristic polynomials of f. Thus V and f satisfy the hypothesis of the theorem. Hence,
by induction, there exists a basis
{v^,
. . . ,

0} of

such that

f(vs)

as2.V2

assVs

Now
{v,V2,

let

V2,

-tVn
is

be elements of

V =

which belong
Since

to the

cosets
a22'V2,

1)2,

respectively.

Then

...,vj

a basis of

(Problem 10.26).
a'22^2

f(v2)

we have ^

f (2)
But

and so
is

2X^2) "" 22'"2


v,

^
22'"2

is

spanned by

v;

hence T{v2)

a22'''2

a2i'"

a multiple of

say

T(V2)

a'22V2

and so

T^kv^)

"21^

Similarly, for
T{Vi)

n,
Ojs-ys

ai2V2

a^Vi

W
+ +

and so

T{Vi)

a^v

0(2^2

-!-+

a^iV^

Thus

T(v)
T{V2)

= = =

a,iv
a2iV
0.22^2

T(Vn)

aiV
is

a22

ann-yn

and hence the matrix of T in this basis

triangular.

CYCLIC SUBSPACES, RATIONAL CANONICAL FORM


10.29.

Prove Theorem
Z{v,T), and
(i)

10.12:

m()

t^

Let Z{v, T) be a T-cyclic subspace, T^ the restriction of + Oo the T-annihilator of v. Then: + 0.^-1*"-' +
dimZ(t;,r)

to

The

set {v, T{v), ..., r'=-i(v)} is a basis of Z{v,T); hence

k.

(ii)

The minimal polynomial


The matrix

of Tv

is TO(f).

(iii)

of T in the above basis is

CHAP.

10]

CANONICAL FORMS

241

fto
-ai

afc-2 Ctlc-l
.

(i)

By

definition of m^Ct), T''{v) is the first vector in the sequence v, T{v), T^v),

which

is

vectors which precede it in the sequence; hence the set B {v, T{v), ., r''-i(i;)} is linearly independent. now only have to show that Z(v, T) = L(B), the linear span of B. By the above, T^v) e L{B). prove by induction that T^{v) &L(B) for every n. Suppose and T^-^(v) E. L(B), i.e. !r"-i(v) is a linear combination of V, ..,T^-i{v). Then r"(v) = r(r-i(v)) is a linear combination of T{v), -tT^v).
linear

combination
. .

of

those

We

We

n>k

THv) G L{B); hence T^{v) L(B) for every n. Consequently f(T)(v) polynomial /(<). Thus Z{v, T) = L(B) and so B is a basis as claimed. But
(ii)

L(B)

for any

Suppose

m(t)
),

=
^

i*

6j_i<~i

we

&o

is

the minimal

polynomial of r.

Then, since

(v,

m(T^){v)

m(T){v)

T^(v)
. .

h^^iT^--^(v)

h^v

Thus m(r)

T^{v) is a linear combination of v,T{v),

and so m^(T^)

0.

Then

T^-i{v), and therefore k ^ s. m(t) divides m(t) and so s k. Accordingly


.,

fc

However, = s and

hence w(t)
(iii)

tn{t).

T(v)

= =

T(v)

T{T{v))

THv)

T{T''-Hv))

T^v)

-oov

a^T{v)

a^n(v)

a^^^n'Hv)

By definition, the matrix of T in this basis is the transpose of the matrix of coefficients of the above system of equations; hence it is C, as required.

10.30.

Let r y - y be linear. Let TF be a T-invariant subspace of V and T the induced operator on VIW. Prove: (i) The T-annihilator of v G V divides the minimal polynomial of T. (ii) The T-annihilator of v G VIW divides the minimal polynomial of T.
:

(i)

therefore,
(ii)

The r-annihilator of v by Problem 10.6,

GV
it

is

the minimal polynomial of the restriction of

to Z(v, T)

and

divides the minimal polynomial of T.


f,

polynomial of

The f-annihilator of p S VIW divides the minimal polynomial of T by Theorem 10.16.

which divides the minimal


a monic irreducible polyare of the form /(t)

Remark. In case the minimal polynomial of T is /(t)" where G V and the T-annihilator of where m n.
nomial, then the T-annihilator of v

f(t) is
i)

G VIW

10.31.

Prove

Let T F -* V be a linear operator whose minimal polynomial is /(t)" where f{t) is a monic irreducible polynomial. Then V is the direct sum of T-cyclic subspaces Zi = Z{vi, T), i = l, ., r, with corresponding T-annihilators

Lemma

10.13:

/(f)"i,

/{f)%

. ,

/(i)"',

n =

ni

^ 2 -

- r
same
and

Any

other decomposition of

into the direct

sum
If

of T-cyclic subspaces has the

number
the

of components and the

same

set of T-annihilators.

The proof is by induction on the dimension of V. lemma holds. Now suppose dim V > 1 and that

dim

V=

1,

then

is itself T-cyclic

the

lemma

holds for those vector spaces of

dimension less than that of V.

242

CANONICAL FORMS
Since the minimal polynomial of hence the r-annihilator of Vi is /()".

[CHAP. 10

V=

V/Zi and let f nomial of f divides /(<)"; hence the hypothesis holds for
the direct

there exists v^GV such that f{T)^~i(vi) = 0; Zi = Z{vi,T) and recall that Zi is T-invariant. Let be the linear operator on V induced by T. By Theorem 10.16, the minimal poly-

is /(<)",

Let

and

f.

Consequently, by induction,

is

sum

of f-cyclic subspaces; say,

V =

Z(%, r)

Z(v f)
ti

where the corresponding f-annihilators are

f{t)"2,

.,

/(*)">,

n2

n^

We
of V2.

claim that there

Let

is a vector V2 in the coset V2 whose T-annihilator is /(<)"2, the T-annihilator be any vector in Dg- Then f{T)"i (w) e Z^. Hence there exists a polynomial g(t) for

which
f(T)n.{w)

9{T){Vi)
(1),

(1)

Since /(<)"

is

the minimal polynomial of T,

we have by

=
But /()" is the r-annihilator of some polynomial h(t). We set
Since
1^1
;

f{T)^{w)

f(T)-n2g{T)(vi)
f(t)-''2 g(t)
,
>

hence
U2

/(i)" divides

and so

g{t)

f(t)2 h(t)

for

= w-

h{T)

,r^

(vi)

w Uj =

'i(^) (^1)

^1.

^2 also belongs to the coset 82-

Thus the T-annihilator

of V2 is a

multiple of the if-annihilator of V2f{T)^{v2)

On

the other hand, by

(1),

f(T)'H(w-h(T)(vi))
is /(t)"2
. .

f(T)^(w)

g{T){vi)

Consequently the T-annihilator of v^


Similarly, there exist vectors
Vj is /(t)"i,

as claimed.

v^,

.,Vf&V

such that ViGvl and that the T-annihilator of

the f-annihilator of ^.

We
Z2

set

Z{V2,T),

...,

Z,

Z(i;r)

Let d denote the degree of


of
Vi

/(t)

so that /(*)" has degree dwj.


v^,

Then

since /(t)"! is both the r-annihilator

and the f-annihilator of

we know

that

{Vi, T(v,), ...,

T*"i- 1 (Vi)}

and

{%
i

f(v^,

f d^ii- 1 (iTj)}
But
,,

are bases for Z(Vi, T) and Z(v^, f) respectively, for


Z(vZ, t); hence
.

2,...,r.
.s^
.

V=

Z(v^

f)

is

a basis for V. Therefore by Problem 10.26 and the relation


{Vi, ..., rd"i-l(Vi), V2, .... Tdr^^-HV2),

fi(v)

THv)

(see
(i;^)}

Problem

10.27),

...,V

..., Td-r-l

is

a basis for V.
It

Thus by Theorem

10.4,

V=
Wj,
.

Z(vi, T)
. .

Z(v^, T),

as required.

remains to show that the exponents


fit),

w^ are uniquely determined by T.

Since d denotes

the degree of

dimy =

d{nx^

h n^)

and

dimZj

drii,

=
-

l,...,r

Also, if s is any positive integer then (Problem 10.59) f(T)^(Z>i f{T)s(Vi) and it has dimension d(Mj-s) if i > s and dimension

is

a cyclic subspace generated by


if

Wi

s.

can be written uniquely in the form v Now any vector v Hence any vector in f(T)^{V) can be written uniquely in the form
f(T)Hv)

&V

w-^+

+ Wr where

w^

Zj.

= f{Ty(Wi)+
>
S,

+f(T)s(w,)
s,

where

/(r)(Wj)

f(Ty{Zi).

Let

be the integer, dependent on


%1
...,
Jlt

for which

>

S,

Wt + i

Then
and so

f(Ty{V)

f(T)HZi)

/(DM^t)

dim(/(r)'(V))

d[{ni

- s) +

{n^

- s)]

(*)

The numbers on the left of (*) are uniquely determined by T. Set s = re 1 and (*) determines the number of TOj equal to re. Next set s = re 2 and (*) determines the number of re, (if any) equal to and determine the number of % equal to 1. Thus n-1. We repeat the process until we set s = by T and V, and the lemma is proved. the Wi are uniquely determined

CHAP.

10]

CANONICAL FORMS
let

243

10.32.

Let F be a vector space of dimension 7 over R, and with minimal polynomial m{t) = {t^ + 2){t + 3f.
canonical forms for T.

T.V^V
all

be a linear operator
possible

Find

the

rational

matrix must be

must add up to 7. Also, one companion and one must be (t + 3)3. Thus the rational canonical form of T is exactly one of the following direct sums of companion matrices:
t^

The sum

of the degrees of the companion matrices

(i)

C(t2
C(2 C(t2
is,

(ii)
(iii)

+ 2) + 2)
+ 2)

+ 2) C((t + 3)3) + 3)3) C((t + 3)2) C(( + 3)3) C(t + 3)


C(t2 C((t

C(t

S)

That
^0

-2

-2
-27 -27 -9

\
\
1 1

A
V

/ -2
-27 -27 -9

1 1

-9
1
(i)

-3
-3/
(iii)

-6/

(ii)

PROJECTIONS 10.33. Suppose V = Wi Wr. The projection of V into its ping E:V ^ V defined by E{v) = Wk where v wi+ that (i) E is linear, (ii) E^ = E.

subspace Wk is the map+ Wr, Wi e Wi. Show

(i)

Since the
defined.

sum v = Wi+ Suppose, for m


(wi

V, u

+ w^, WiG W ~ w^i

is

uniquely determined by

v,

the

mapping
kwf, Wj

is

well

+ w^,
A:t;

w[ S W^.

Then

-{-

u =

+ w() +

(Wr

+ w'r)

and

fcwj

kw^,

+ w,' G

PFj

are the unique sums corresponding to v

E(v
and therefore
(ii)

+ u) = Wk +

wl^

+ u and kv. Hence = E{v) + E(u) and

E(kv)

kw^

kE(v)

7 is linear.

We
is

have that

the unique

+ 0+---+0 sum corresponding to w^ G Wk'-, hence E(w^) = w^. Then EHv) = E(E(v)) = E(Wk) = w^ = E(v)
w^ =
TOfc

0+---+0 +

for any v

V,

Thus E^

E, as required.

10.34.

E:V-*V is linear and E^ - E. Show that: (i) E(u) = u for any uGlmE, the restriction of to its image is the identity mapping; (ii) V is the direct sum of the image and kernel of E: V ImE KerE; (iii) is the projection of V into Im E, its image. Thus, by the preceding problem, a linear mapping T V -> V is a projection if and only if T^ = T; this characterization of a projection is frequently
Suppose
i.e.

used as
(i)

its definition.

If

M G Im E, then

there exists

E(u)
as required.
(ii)

e V for which E(v) = u; hence = E{E{y)) = EHv) = E{v) = u


E(v)

Let V

SV. We

can write v in the form v

E(v) G

Ker E.

= E{v) + v E(v). Now E(v - ^(1;)) = E(v) - E^(v) = E{v) - E(v) = Accordingly, V = Im + Ker E.
jE?

e.ImE

and, since

244

CANONICAL FORMS

[CHAP. 10

n Ker E. By (i), E{w) = w because w G Im /. On the other Now suppose w and so ImE n Ker E = {0}. These two because w G Ker E. Thus w = hand, E{w) = conditions imply that V is the direct sum of the image and kernel of E.

GImE

(iii)

Let v

by

(i),

and suppose v = u + w where uGlmE and w e Ker E. because w G Kerfi". Hence and E{w) = E{u + w) = E(u) + E{w) = u + = m ?(-!;)

eV
E

Note that E(u)

=u

That

is,

is

the projection of

into its image.

10.35.

Suppose

V=
if

UW
E{v)

and suppose
if

T:V-^V
t>

is linear.

r-invariant

and only

TE - ET where E
for every

is

that U and TF are both the projection of V into U.

Show
=

Observe that
iff

G U

Y,

and that

(i)

E(v)

iff

v:U,

(ii)

K(i;)

v&W.
Suppose

ET =

TE. Let u
r(M)

U.

Since E{u)

=
Now

T{E(u))
let

(TE)(u)

= w, = {ET){u) = = 0, = r(0) =
Let hence

E(T(u))

G U G

Hence

is

T-invariant.

w GW.
(TE){w)

Since E{w)

E{T(w))

(ET){w)

T{E{w))

and so

T(w)

W
where

Hence

is also

T-invariant.

Conversely, suppose

and

W are both
G
i7

T-invariant.

G Y and
S(r{w))

suppose

= w+w
T(u)

G r and w G

T^.

Then

?()

and

r(w)

G W;

TM

and

E(T(w))

0.

'^^'"^

(ET){v)

(Br)(M

and

+ TO) = (Er)(M) + (Br)(w) = E(T(u)) + E{T{w)) = {TE)(v) = (r7)(M + w) = r(S(M + ;)) = T{u)
for every
-y

That

is,

(ET){v)

(TE)(v)

F; therefore

ET = TE

as required.

Supplementary Problems
INVARIANT SUBSPACES
10.36.

Suppose

W W

is

invariant under

T:V -^V. Show


is

that

is

invariant under f{T) for any polynomial

10.37.
10.38.

Show

that every subspace of

invariant under I and

0,

the identity and zero operators.

is invariant under Suppose S + r and ST.

S:V^V

and

V -^

F.

Show

that

is

also invariant

under

10.39.

Let

r y
:

->

be linear and

let

W be the eigenspace belonging to an eigenvalue X of


1)

T.

Show

that

W
10.40.

is

r-invariant.

Let

linear operator on

be a vector space of odd dimension (greater than V has an invariant subspace other than

over the real field E. or {0}.

Show

that any

10.41.

Determine the invariant subspaces of


that there exist T-invariant subspaces

A =

_A

viewed as a linear operator on

(i)

R2,

(ii)

C^.

10.42.

Suppose dim

V=

n.

Show

T:V^V

has a triangular matrix representation if and only if cW^ = V for which dim TFfc = fc, k = l,...,n. WiCWzC

INVARIANT DIRECT-SUMS
10 43

The subspaces Wi,...,Wr are said

to be independent if

each Wi = 0. Show that LiWi) denotes the linear span of the Wi-)
10.44.

L(Wd =

Wi

@Wr
if

if

Wi + and only

-|-

w^

0,

if

the

Wi
(ii)

Wj G Wi, implies that are independent. (Here

that V = W^ + i,...,Wr) =

Show Show

Wi
{0},
fe

Wr
=
l,

if

and only

(i)

V=

L{Wi)

and

W^nL{Wi,

.,Wk-i.

...,r.
if

10.45.

that

L(Wi)

= W,---Wr

and only

if

dimLd^i)

dim Wi

-t-

dim W^.

CHAP.

10]

CANONICAL FORMS
is

245

10.46.

fi(t)

Suppose the characteristic polynomial of T V - V are distinct monic irreducible polynomials. Let
:

A(t)

V =

position of

WiQ

into r-invariant subspaces.


7"

Show that

/((t)! is

fAt)' where the be the primary decomthe characteristic polynomial of the

Wr

/i(f)"i /2(f)2

restriction of

to PFj.

NILPOTENT OPERATORS
10.47.

Suppose

S and T

are nilpotent operators which commute,

i.e.

ST =

TS.

Show

that

S+T

and

ST

are also nilpotent.


10.48.

Suppose
that

is

a supertriangular matrix,

i.e. all

entries on

and below the main diagonal are

0.

Show

is nilpotent.

10.49.

Let
is

be the vector space of polynomials of degree nilpotent of index n + 1.

n.

Show

that the differential operator on

10.50.

Show

that the following nilpotent matrices of order n are similar:


1 1
. . . .

and
.

\0

...

0/

10.51.

Show that two nilpotent matrices of order 3 are similar if and only if they have the same index of nilpotency. Show by example that the statement is not true for nilpotent matrices of order 4.

JORDAN CANONICAL FORM


10.52.

Find all possible Jordan canonical forms for those matrices whose characteristic polynomial and minimal polynomial m{t) are as follows:
(i)

A(t)

A(t)

(ii)

A{t)
A(t)

(iii)

= = =

(t-2)4(f-3)2,
(t-7)5, (t-2)7,
m(t) m(t)

m{t)

(t-2)2(t-3)2

= =

(t-7)2 (t-2)3
m(t) =: (-3)2(i-5)2
is

(iv)

A(t)

(t-3)*(t-5)\

10.53.

Show Show

that every complex matrix


10.50.)
all

similar to its transpose.

{Hint.

Use Jordan canonical form and

Problem
10.54. 10.55.

that

complex matrices

of order

for which

A" =

are similar.

Suppose

is

a complex matrix with only real eigenvalues. Show that

is

similar to a matrix with

only real entries.

CYCLIC SUBSPACES
10.56.

Suppose

T:V -*V
v.

is linear.

Prove that Z{v, T)

is

the intersection of all T-invariant subspaces

containing
10.57.

Let fit) and g{t) be the T-annihilators of u and v respectively. atively prime, then f(t)g(t) is the T-annihilator of u + v.

Show

that if

f(t)

and

g{,t)

are rel-

10.58.

Prove that

Z(m, T) r-annihilator of u.

Z(v, T)

if

and only

if

g(T)(u)

where

g(f)

is

relatively

prime to the

10.59.

= Z{v, T), and suppose the T-annihilator of v is /(*)" where f(t) is a monic irreducible polyLet nomial of degree d. Show that f{T)^{W) is a cyclic subspace generated by f(Ty(v) and it has dimension d{n s) if n > s and dimension if n s.
Find
(i)

RATIONAL CANONICAL FORM


10.60.
all

possible rational canonical forms for:

(ii) (iii)

6X6 matrices with minimum 6X6 matrices with minimum


8

X 8 matrices with minimum

polynomial m(t) polynomial mit) polynomial m(t)

= = =

(t^
(t

+ 3){t + 1)2 + 1)3 {t^ + 2)^(t + Z)^

10.61.

Let be a 4 X 4 matrix with minimum polynomial m(t) = (t^ + \){fi 3). Find the rational canonical form for A if A is a matrix over (i) the rational field Q, (ii) the real field B, (iii) the complex field C.

246

CANONICAL FORMS

[CHAP. 10

10.62.

Find the rational canonical form for the Jordan block

10.63.

Prove that the characteristic polynomial of an operator


divisors.

T :V

-^

is

a product of

its

elementary

10.64. 10.65.

Prove that two


Let
C(f(t))

3X3

matrices with the same

minimum and

characteristic polynomials are similar.


f(t).

denote the companion matrix to an arbitrary polynomial

Show that

f(t) is

the char-

acteristic polynomial of C{f(t)).

PROJECTIONS
10.66.

Suppose

i^j;
10.67.

(ii)

V = Wi I = E^+
.,Er
0, i

Wr +E^.
(iii)

Let Ei denote the projection of

into Wi-

Prove:

(i)

EiE^

0,

Let El,
(ii)

..

be linear operators on

such that:
that

(i)

EiEj

^ i;

Bj

+E^. Prove
i.e.

Ef = /;, i.e. the Ei are V = Im Ej Im B^

projections;

10.68.

Suppose

E -.V ^V
^
)

is

a projection,
is

E^ = E. Prove
and

that

has a matrix representation of the

form
10.69.

where r

the rank of

/^ is the r-square identity matrix.

Prove that any two projections of the same rank are similar.
10.68.)

{Hint.

Use the result of Problem

10.70.

Suppose
(i)

E -.V -^V is a projection. Prove: I-E isa projection and V = ImE

Im (I-E);

(ii)

+E

is

invertible

(if

+1 #

0).

QUOTIENT SPACES
10.71.

Let IF be a subspace of V.
linearly independent.

Show

Suppose the set of cosets {vi + W,V2 + that the set of vectors {v^, V2, ..., vj in V

W, ...,+
is

IF} in

V/W

is

also linearly independent.

10.72.

Let IF be a subspace of V.

Suppose the set of vectors

and that

L(Ui)

nW

{0}.

Show

{mi, Wg that the set of cosets {mi

m} in

V
,

is

linearly independent,

IF,

IF}

in

V/W
. .

is

also

linearly independent.
10.73.
ii + IF} is Show that {ui + W, tt} is a basis of U. and that {mj, Suppose V = U a basis of the quotient space V/W. (Observe that no condition is placed on the dimensionality of
. . .

V
10.74.

or IF.)

Let IF be the solution space of the linear equation


ajXi

+
,

020:2

+
,

an^n

0.

O'i^

K
is

and

let

(5i, 63

6)

X".

Prove that the coset v


IF of IF in K"

the solution set of the

linear equation
aiXi
10.75.

a2X2

+
,

a^Xn

where
and
let IF

i._j,j^ 6 - ajfti +

.a.

h a6

Let

be the vector space of polynomials over


of the

by
10.76.

t*, i.e.

form

Ugt*

+ a^t^

-\

h a_^t^.

Show

be the subspace of polynomials divisible that the quotient space V/W is of dimension 4.

Let

and IF be subspaces of
(i)

such that

also be viewed as a coset of IF in

since

u&U
(ii)

WcUcV.

Note that any coset

implies

w e V; hence U/W

is

IF of IF in ?7 may a subset of V/W.

Prove that
10.77.

?7/IF is a subspace of

V/W,

dim {V/W)

dim(i7/IF)

= dim{V/U}.

Let

and IF be subspaces of V. secting each of the cosets of t/ in

Show that

the cosets of

UnW
v,v'

in

can be obtained by inter-

by each of the cosets of IF

in V:

V/{UnW)
10.78.

{{v+U)n{v'+W):

eV}

be linear with kernel IF and image U. Show that T:V under the mapping quotient space V/W is isomorphic to the Furthermore, show that e V/W -> U defined by e{v -I- IF) = T{v). T = io 0OTJ where i; F -> V/W is the natural mapping of V into C is the inclusion mapping, V/W, i.e. r,{v) = -y -t- IF, and t Let
:

^V

yjy^

i.e.

i(u)

u.

(See diagram.)

CHAP.

10]

CANONICAL FORMS

247

Answers
10.41.
(i)

to

Supplementary Problems
=
L((2, 1

R2 and {0}
2
1

(ii)

C\

{0}, PTj

- 2i)), W^ =
2
1
I

L{{2, 1

2i))

10.52.

(i)

2
4_.

1
I

L2_.
1

^2^
i
I

r3
(ii)

i_lL
3
1

7
i I

1
;

7
7

1'
I

'

L_i4-I

111. 7 "[7I

(iii)

2
2
1

2
1

-I

H
I

1
I

2 2
1
I

2
I

111I

2 /

TiL.
3
1
I

(iv)

3 3
1
I

r3
1

Wi^
L.
\
I

1
I

5
I

rL-

f-

[5
'

n
5
I

3 L"-' _ r5""i"i
I

|5
I

^-in_
I

248

CANONICAL FORMS
(i)

[CHAP. 10

10.60.

'0
1

-3
1

-3
1

-3 -1 -2
-1
1

-1
1

-2
-1

-2/

(ii)

'0

0-1

'0
1

10-3
1-3
1

0-1 0-3 1-3


1

1 1

-1 -3 -3

-1 -2
-1,

-1
-1,

(iii)

-4
1 1

-4
-9 -6

-9
1

-6/

2
1
1

-4
1

-9 -6
-3.

10.61.

(i)

(ii)

/O

-1
y/^

(iii)

V^

-W
10.62.

-Vsl

^0

-x*\
4\3

10 10
\0
1

-6\2
4X

chapter

11

Linear Functionals
INTRODUCTION

and the Dual Space


K

In this chapter we study linear mappings from a vector space V into its field of scalars. (Unless otherwise stated or implied, we view as a vector space over itself.) Naturally all the theorems and results for arbitrary linear mappings on V hold for this special case.

because the special relationship of apply in the general case.

However, we treat these mappings separately because of their fundamental importance and 7 to Z gives rise to new notions and results which do not

LINEAR FUNCTIONALS AND THE DUAL SPACE Let F be a vector space over a field K. A mapping
tional (or linear form)
if,

<i>:V -*

for every u,v

GV

is

is

termed a linear func-

and every a,b G K,


a
4,{u)

4,{au -^hv)

<j>{v)

In other words, a linear functional on


Example
11.1:

a linear mapping from


i.e.

into K.
...,a)

Let
is

jt-j

K"

linear

and so

it is

be the ith projection mapping, a linear functional on X".


t

7rj(ai, aj,

a^

Then

ttj

Example

11.2:

Let

be the vector space of polynomials in

over R.

Let

^ iV -^ R
is

be the integral

operator defined by ^{p(t)) a linear functional on y.

j
o

p{t) dt.

Recall that

linear;

and hence

it is

Example

11.3:

Let

be the vector space of n-square matrices over K.

Let

T iV ^
(ay)

be the trace

mapping
T{A)

- +

a22

where

A =

That is, T assigns to a matrix the sum of its diagonal elements. linear (Problem 11.27) and so it is a linear functional on V.

This

map

is

By Theorem

6.6,

a vector space over

K with
{(t>

the set of linear functionals on a vector space V over a addition and scalar multiplication defined by

field

is also

+ (T){v) =

(j>{v)

(t{v)

and

{!i<j>)iv)

k<j){v)

where ^ and a are linear functionals on V and is denoted by V*.


Example
11.4:

and k G K.

This space

is

called the diuil space of

Let V = K", the vector space of ti-tuples which we write as column vectors. Then the dual space V* can be identified with the space of row vectors. In particular, = {a^, any linear functional a) in y* has the representation
<f>
. . .

<p{xi,

.,Xn)

(tti.aa,

.,a)

249

250

LINEAR FUNCTIONALS AND THE DUAL SPACE


or simply
0(a;i

[CHAP.

11

a^Xi

02*2

In^n

Historically, the above formal expression

was termed a

linear form.

DUAL BASIS
Suppose y is a vector space of dimension n over K. By Theorem 6.7, the dimension of In fact, each basis of V is of dimension 1 over itself.) the dual space V* is also n (since determines a basis of V* as follows:

Theorem

11.1:

Suppose

{Vi,

.,v} is a basis of

over K.

Let

^j,

.,^

G V*

be the

linear functionals defined

by

^'(^^^

jo if^^i

Then

{^i

^} is a basis of V*.
basis.

The above basis {<|>^) is termed the basis dtial to (Vi) or the dval mula which uses the Kronecker delta Si, is a short way of writing
4>^(vJ

The above

for-

1,

<l>,{v^)
</,,(t;,)

= =

0,
1,

<i>^(v^)

0,
0,

.,

.I>,ivj <l>,ivj

^^{v,)

0,

<l>,{v,)

.,

By Theorem

6.2,

these linear mappings ^, are unique and well defined.


Consider the following basis of R^:
{^i> ^2}'

Example Hii:

{v^

(2,1), Uj

(3,1)}.

Find the dual basis

We

seek linear functionals


^i('i;i)

^i(a;,

y)

=
0,

ax

by and 02(*. y)

=
1

ex

+ dy

such that

1,

01(^2)

= + + + +

02(^1)

0,

9*2(^2)

Thus

= = 0iK)
0,(vi)

0i(2, 1)

0j(3,l)

= = = =

2a
3a
2c
3c

= =

11

^^

-1, 6

Oj

02K) =
02(^2)

02(2,1)
02(3, 1)
is

= Oj = ij
4>2(.x,

^^

= l,d = -i - 2y}.

Hence the dual basis

{4>i(x, y)

= -x + Sy,

y)

The next theorems give

relationships between bases


.

and their
{<l>^,

duals.

Theorem

11.2:

Let

{vi,

v} be a basis of

V and let
4,^{u)v^

...,4>Jhe the dual basis of V*.

Then

for any vector

uGV,
+ +

u =

4,^{u)v^

<t>Su)v^

and, for any linear functional a

V*,

Theorem

11.3:

Let
{ffj,

{vi,
.
.

...,Vn} and {wi, ...,Wn} be bases of V and let {<^,, .. o-} be the bases of 7* dual to {Vi} and {Wi} respectively.
[Vi) to {Wi}.

.,<#>}

and

Suppose

is

the transition matrix from


{4>^}

Then

(P-i)* is the transition

matrix from

to

(o-.j.

CHAP.

11]

LINEAR FUNCTIONALS AND THE DUAL SPACE

251

SECOND DUAL SPACE We repeat: every vector


functionals on V.
consists of all

space V has a dual space F* which consists of all the linear has a dual space V**, called the second dual of V, which the linear functionals on V*.

Thus V*

itself

for any

We now show that each GV* we define


<j>

GV

determines a specific element v

GV**.

First of

all,

^
vi<j>)

=
-^

<l>iv)

It

remains to be shown that this


linear functionals
v(a(j>

any

^,

o-

V*,
(acf>

map v.V* we have

is linear.

For any scalars a,b

GK

and

ba)

+ b(T)(v)

a <i>(v)

b a(v)

av{<j>)

bv{a)

That

is,

is

linear
If

and so v GV**.

The following theorem

applies.

Theorem

11.4:

V has finite

dimension, then the mapping v

is

an isomorphism of

onto V**.

that this

The above mapping v i^ t; is called the natural mapping of V into V**. We emphasize mapping is never onto F** if 7 is not finite-dimensional. However, it is always
it is

linear and, moreover,

always one-to-one.

Now suppose V does have finite dimension. By the above theorem the natural mapping determines an isomorphism between V and V**. Unless otherwise stated we shall identify V with V** by this mapping. Accordingly we shall view V as the space of linear functionals on V* and shall write V = V**. We remark that if {^J is the basis of V* dual to a basis {Vi} of V, then {vi} is the basis of V = V** which is dual to (^J.

ANNIHILATORS
Let

^gV*
We
is

W he a subset (not necessarily a subspace) of a vector space V. A linear functional called an annihilator of W = {0}. 4>{w) = for every w GW,
is

if

show that the

set of all such

a subspace of V*.

Clearly

a,b

gK

mappings, denoted by G W^. Now suppose

W and
^,
<r

i.e.

if

<j>{W)

called the annihilator of

W,

G +

W^.

Then, for any scalars

and for any


(a<ji

w GW,
+ b<j){w) = a^(w) +
b (t{w)

aO

bO

Thus a^ + baG
its

and so
is

W
F

is

a subspace of V*.

In the case that IF


annihilator
11.5:

a subspace of F,

we have

the following relationship between

W and
Then
where

W.
Suppose
(i)

Theorem
Here
IF"" is

has

finite

dim

W + dim IF"
^(v)

dimension and IF
(ii)

is

a subspace of

F.

= dim F and

TF

= W.
IF""

W" = {vGV:
of

viewed as a subspace of

= F

for every ^ G W>} or, equivalently, under the identification of F and F**.

(TF")"

The concept

an annihilator enables us
anXi

to give another interpretation of a

homogeneous

system of linear equations,

ai2X2

ainXn

(*)

amlXi

+ am2X2 +

UmnXn

252

LINEAR FUNCTIONALS AND THE DUAL SPACE

[CHAP.

11

A = (an) is viewed as an element Here each row {an, oa, ., (kn) of the coefficient matrix ., Xn) is viewed as an element of the dual space. of K" and each solution vector ^ = {xi, X2, In this context, the solution space S of (*) is the annihilator of the rows of A and hence of the row space of A. Consequently, using Theorem 11.5, we again obtain the following fundamental result on the dimension of the solution space of a homogeneous system of
. . . .

linear equations:

dimS = dimK" dim (row

space of A)

= n - rank (A)

TRANSPOSE OF A LINEAR MAPPING


Let
U.

T :V -^ U

Now
into K:

be an arbitrary linear mapping from a vector space V into a vector space for any linear functional ^ G U*, the composition ^ o T is a linear mapping from

That

is,

(j)oT

GV*. Thus

the correspondence
</>

<f>oT
call it

is a mapping from U* into V*; we denote words, T*:TJ* -^ V* is defined by

it

by T' and

the transpose of T.

In other

r'(0)

4,oT

Thus

{T\4,)){v)

^{T{v)) for every v

&V.
T' defined above is linear.
<^,

Theorem

11.6:

The transpose mapping

Proof. For any scalars a,b


T*{a<j>

and any linear functionals

f/*,

+ ba-) = =
T
is

+ 6tr)or = a T\4,) + h T*{a)


(a^

a{ci>oT)

b(aoT)

That

is,

T' is linear as claimed.

We

emphasize that
into V*:

if

a linear mapping from


^
j,,

into U, then T* is a linear

mapping

from U*

The name "transpose" for the mapping

T* no doubt derives

from the following theorem.

Theorem

11.7:

Let T-.V-^V be linear, and let A be the matrix representation of T relative to bases {Vi} of V and {Ui} of V. Then the transpose matrix A* is the matrix representation of T*:U*-* V* relative to the bases dual to
{Mi}

and

{Vi}.

CHAP.

11]

LINEAR FUNCTIONALS AND THE DUAL SPACE

253

Solved Problems

DUAL SPACES AND BASES


11.1.

Let

R2

-*

and
(i)

(t{x,

y)

R and a - Sx- y.
J/)

R^

-*

Find

(i)

be the linear functionals defined by ^{x, y) ^ + a, (ii) 4^, (iii) 2^ - 5<r.

+ 2y

(^

)(,

=
4

i>(x,y)
0(a;,2/)

(ii)

(40)(aj,2/)

+ =

a(x,y)

2y

3x

=
5(3x

4x

+
j/)

i(x

+ 2y) =
ha(x,y)

ix

8y

(iii)

{2<p~5a){x,y)

2<p{,x,y)

2(a;

+ 2j/) -

-13a;

9j/

11.2.

Consider the following basis of R^: Find the dual basis {^j, 4)^, ^g}.

{vi

(1,-1,3), Vi

(0,1,-1), Vs

(0,3,-2)}.

We

seek linear functionals


^i(, y, )

Oi*

+ azv + aaz,
0i(i'i) 02(i;i)

4>^(x, y, z)

h-^x

+ h^y + bgz,
1

03(3;,

j/,

z)

e^x

+ CjW + CgZ

such that

=
= =

0i(v2)
02('y2)

03(^1)

03(^2)

= = =

01(^3)
?*2(''3)

03(1^3)

= = =
1

We

And

^^ as follows:
4>\ko\)
0l(^'2)

01(^3)

= = =

0i(l.

-1,

3)

0i(0, 1, 0i(0, 3,

-1) -2)

= = =
1,

a2
fflj

3a2
02

+ 303 = - tts = - 2a3 = =


0.

Solving the system of equations,

we

obtain

a,^

=
3)

0, 03

Thus

0i(a;, V, 2)

We

next find

02-

02(^1) 02(^2)
02(1^3)

= = = =
7,

02(1.

-1.

02(0,1,-1) 02(0,3,-2)

= = = = = = = =

61

= 62- 63 = 362-263 =
62

363

Solving the system,


Finally,

we

obtain

61

h^
03(1,

2, 63 -1,

3.

Hence
C2

02(a', y, z)

7x

2y

3.

we

find 03:
03(^1)
03('"2)

03(^3)

= = = =

3)

Ci

03(0,1,-1) 03(0,3,-2)
C2

= = 3c2-2c3 =
3C3

C2-

C3

Solving the system,

we

obtain

Cj

2,

1,

Cg

1.

Thus

03(x, y, z)

= 2x +

z.

11.3.

Let
{a

be the vector space of polynomials over R of degree bt: a,b GR}. Let ^j:F-R and ^2 ^"*'' be defined by
=

1,

i.e.

V =

<^i(/(i))

= S^'mdt

and

Um)

==

S^'fi*)^^

(We remark that


basis {vi,
Let Vi
t;2}

<j>^

of

and ^^ are linear and so belong which is dual to {<^j, ^g}.


i^a

to the dual space V*.)

Find the

+ bt

and

= c + dt. By =
1,
02('yi)

definition of the dual basis,

0i(^i)

= = =
a

and

0i(V2)

0,

02(^2)

Thus
0j(vj)

= =

(a+6t)df

+ +

^b

= =

or
02(''i)

2, 6

= -2

(a

6t)

dt

2a

26

254

LINEAR FUNCTIONALS AND THE DUAL SPACE

[CHAP. 11

01(^2)

{c

+ dt)dt + dt)dt -

+ ^d or
c

= 1,

MV2) =
In other words, {2

{c

2c

+ V

2d

2t, ^ +

t} is

the basis of

which

is

dual to

{961,

02)-

11.4.

Prove Theorem 11.1: Suppose {vi, .,Vn} isa basis of be the linear functionals defined by
. .

V over K.
i

Let

<l>^,

<j>^

& V*

S.(v.)

8..

fl
i

if

Then

{^j,

.,^} is a basis of V*.


{^i,
. .

We

first

show that

.,<i6}

spans V*. Let

be an arbitrary element of V*, and suppose

Set

<r

ki<f>i

fc0.

Then

a(Vi)

= = =

(fci0i
fei

+ +

+ fe0)(lJi)
k2<p2{Vi)

01(1^1)

+
^2

fe0(Vl)

fci

&

fci

Similarly, for

2,

w,

<T(Di)

(fci0i

+ /i:0)(i;i) +

=
Thus
&101
<p{Vi)

A;i0i('!;i)

ki^iivi)

+ +

fe0n(-i'i)

^i

a{Vi)

+
It

fe0.

and a agree on the basis vectors, for i = 1, ..,n. Since Accordingly, {0i 0} spans V*.
.

cr

remains to be shown that

{0i,

is linearly

independent.

Suppose

ai0i

a202

a0

Applying both sides

to v^,

we

obtain

= = =
Similarly, for
i

0(vj)

{aj0i

+ a0n)(i'i) + + O'nSi'nC'yi) = tti + a

ai0i(l'i)

+
a2

ttz

02(^1)

2,

,
(-Ui)

=
=
That
is,

=
+

(ai0i ^

+ a^)(Vi)
i S6i(^i)

a-l 0l('yt)

S^nC^'i)

"i

0,

.,a.

0.

Hence

{0i, ...,</>n} is

linearly independent and so it is a basis of V*.

11.5.

v} be a basis of Prove Theorem 11.2: Let {vi, vector uGV, basis of V*. Then, for any
. . .

and

let

{<j>^,

..

.,<f,Jhe the dual

and, for any linear functional a


a Suppose

GV*,

=
M

cr(i;>j

.7(1;,)^,

<7(i;J<^

(2)
(5)

aiVi

a^Vi

+ =

Then
0i(m)

di 0i(i;i)

^2 01(^2)

SilC'^n)

O2

==

CHAP.

11]

LINEAR FUNCTIONALS AND THE DUAL SPACE


i

255

Similarly, for

2,

.,n,

</>i{u)

!
a2'

,j,i(vi)

+
n-

Ui

<f>i{Vi)

0i(i;)

;
(3),

That

is,

ipiiu)

Oj, 02(w)
(2).

0nW =

Substituting these results into


to both sides of (1),

we

obtain

(1).

Next we prove

Applying the linear functional a


a{u)

= = =

Since the above holds for every

+ ^aM^K) + + </>n(u) <r(v) o{Vi) ^i(w) + o(V2) 02(m) + + a{vj <f,{u) + <r(t))0)(M) {<'{Vi)'f>l + ('(^2)02 + m G V, a a{vi)<f,i + <r(i^2)02 + + "{vj^^
0i(M)<r(vi)

as claimed.

11.6.

Prove Theorem 11.3: Let {vu...,Vn} and {wi,...,Wn} be bases of V and let CT} be the bases of V* dual to {vi} and {Wt} respectively. {^1, ^} and (<7j, Suppose P is the transition matrix from {Vi} to {Wi}. Then (P~)' is the transition matrix from {(j>J to {<tJ.
.
.

Suppose

Wi

OuUi
(121^1

W2 =

+ +

ai2V2
a22''2

+ +

+ +

aii;

<ri

a2nVn

02

= =

ftn^i

62101

+ +

6i202 ^2202

"

"

"

+ +

6i0
hn'f'n

w =
where

aii)i

P=

()

and

Let Ri denote the

+ a2i'2 + a - b^i,pi + 6202 + + av + Q = (6y). We seek to prove that Q = (P-i). tth row of Q and let Cj denote the ith column of P. Then Rj = (6ti. 6i2. and Cj = (dji, aj2, a^^Y 6i)
'

6n0

By

definition of the dual basis,


<fi(Wj)

= =

(6(101
6jiaji

+ 6i202 + + 6j2aj2 +
Thus
iJ,C2

+ 6j^)(ajii)i + aj2V2 + + ajv) = Rfij = Sjj + 6ia.j


where Sy

is

the Kronecker delta.

/KiCi

...

QPt

^2^1
\RnCl

R2C2

KC\ R2C
^rflnl

!"'

"1=7

Rvp2
as claimed.

and hence

Q=

(P)-i

(P-i)

11.7.

Suppose V has finite dimension. <j>GV* such that <^(v) # 0.

Show

that

if

GV, v'0,
6.1,
.
.

then there exists

We
mapping

extend {v} to a basis {i), ^2. such that 0(1;) <f>:V ^

>

^n) of V-

By Theorem

and

^(i;^)

0, i

2,

. ,

n.

there exists a unique linear Hence ^ has the desired

property.

11.8.

Prove Theorem isomorphism of

11.4:

If

onto V**.

We
a,b

first

prove that the

has finite dimension, then the mapping v ^ v (Here v V* -* K is defined by v{<j>) = ^(v).) map v \-^ v is linear, i.e. for any vectors v,w eV and any

is

an

scalars

& K, av + bw = av + bw. For any


av

linear functional
<f,{av

<fi

V*,

+ bw (0)
+

= =

bw)

a ^{v)

(f>{w)

av(</,)

bw(4,)

(av

bw)(ip)

Since

av
v

+ bw
At.

(<i>)

(at)

6M))(0)

for every

<p&V*, we have SaT+Vw

= 0?+

6w.

Thus the

map

1-^ 1;

is

linear.

256

LINEAR FUNCTIONALS AND THE DUAL SPACE


Now
suppose

[CHAP.

11

G V* for which V GV, v = 0. Then, by the preceding problem, there exists implies v=Q, the map v H- r and thus v ^ Q. Since v # Hence v (<j,) - <t,(v) # <f,{v) is nonsingular and hence an isomorphism (Theorem 6.5). Now dim V = dim V* = dim V** because V has finite dimension. Accordingly, mapping v ^ v is an isomorphism of V onto V**.
<f>

# 0.

ANNIHILATORS
11.9.

Show

that if <j>GV* annihilates a subset S of V, then ^ annihilates the linear span L{S) of S. Hence S = (L(S)). .yW^G S for which v = a^w^ + a^w^ + + a^w^. Suppose V e L{S). Then there exist Wj,
.

<f>{v)

Ui 0(Wi)

(12

0(W2)

+
<f>

('('^r)

"l"

^^20

'

afi

Since v

was an arbitrary element

of L(S),

annihilates L(S) as claimed.

11.10.

be the subspace of R* spanned by vi = (1, 2, -3, 4) and v^ = (0, 1, 4, -1). Find Let a basis of the annihilator of W. By the preceding problem, it suffices to find a basis of the set of linear functionals <f>{x, y, z, w) = and <p(v2) 0: ax + hy + CZ + dw for which 4>{vi) =
0(1,2,-3,4)

0(0,1,4,-1)

= =
d o

2&
6

3c

4d

4c-d

= =

The system
Set
tional
c

of equations in

unknowns

a, h, c,

is in echelon

= 1, d =
= 0,
2/.

to obtain the solution


11a;

11, 6

= -4,

form with free variables c and d. and hence the linear funcc = 1, d =
and hence the linear func-

0i(a;, y, z,

Set c
tional

02(^.

= = -1 . *") =
w)

4j/

+
w.

z.

to obtain the solution

= 6, 6 = -1, c = 0, d = -1

&x

The

set of linear functionals {0i, 02} is a basis of

W>, the annihilator of W.

11.11.

Show
(i)

that:

(i)

for any subset

of V,

S^S"";

(ii)

if

S1CS2, then S^cS?.


v(0) = 0(v) = Accordingly,
0.

V e S. Then for every linear functional Therefore, under the identification of F and V**,

Let

e S*, S So.

Hence

(SO)*.

S C S"*.
annihilates every ele-

(ii)

Let

ment of

0GS2. Then G Si, i.e.


11.5:

(!;)

for every

Sg.

But S1CS2; hence

Si.

Therefore S^ cSj.

11.12.

Prove Theorem

Then
(i)

(i)

dim

W + dim W

Suppose

has

finite
(ii)

dimension and

is

& subspace of V.

= dim V and

W^ = W.
want
to

=r Suppose dim V = w and dim a basis {wi, ...,w^}oiW and extend


Consider the dual basis
,

it

n.

We

show that dim


,

W'>
.

= n-r. We
.

choose
v_r}-

to the following basis of V:


{wi,

.,w

Vi,

.,

\01i

>

0r> "it

>

<'ti-r/

By
!

definition

so it

dual basis, each of the above a's annihilates each Wj-, hence G W*. We claim that {ctj} is a basis of W'". Now {<tj} is part of a basis of V* and ff_, is linearly independent.
of
the

We next

show that
<7

{a^}

spans
4

a(Wi)0i

001

4-

G W<>. By Theorem 11.2, + <T{Wr)</,r + a(Vi)ai + + a{v^-r)<'n-r + CT(v_r)<r-r + 00, tr(Vi)(Ti +


Let
<r

W.

-I-

<f(''l)<'l

+ aiv^-rW-r
it is

Thus {ffi dim V dim


(ii)

ffn-r)

spans PF" and so

a basis of W^.

Accordingly,

dim

W^"

= m

as required.

Suppose dimV^^w and diraW = r. Then dim F* by (i), dim TVO" = n- (n-r) = r; therefore dim

=m

and, by

(i),

WC

TFOO.

Accordingly,

W = WO".

W = dim W".

By

dimTF<' = TO-r. Thus the preceding problem,

CHAP.

11]

LINEAR PUNCTIONALS AND THE DUAL SPACE

257

11.13.

Let
That
then

U
Let

and e

W be subspaces of V.
Then
t/o

Prove:

(C7

0G {U+W)0.
and

annihilates

U+W
n
PFO.

Wf =

n W^.

is,

TF";

hence

C/o

On
-y

the other hand, suppose

= M+w

where
i.e.

annihilates

U+W,

a e n W. Then a annihilates U and also W. m [7 and w e W. Hence ctCv) = (r{u) + a{w) = + e. (U + TF)". Accordingly, U'>+W'>c(U+ W)'>.

and so, in particular, ,p annihilates Thus {U + W)" C U" n W.


If

and y.

ve^U+W,
0.

Thus

Both inclusion relations give us the desired equality.

Remark: Observe that no dimension argument


for spaces of finite or infinite dimension.

is

employed in the proof; hence the result holds

TRANSPOSE OF A LINEAR MAPPING


11.14.

Let

be the linear functional on R^ defined by ^(a;, y) - x - 2y. For each of the following linear operators T on R^, find iT%4,)){x, y): (i) T{x, y) = {x, 0); (ii) T{x,y) = (y,x + y); (iii) T{x,y) = (2x~Zy,hx + 2y).
<j>

By
vector
(i)

definition

of the transpose mapping,

rf(0)

.pof,

i.e.

(Tt{4,)){v)

<p{T(v))

for

every

V.

Hence

{Tt{,p)){x, y)

(ii)

(rt(0))(x,
{Tt(<f,))(x,

2/)

= =

</,(T{x,y))

4>{T{x,y))
<f.(T(x, y))

(iii)

y)

= = =

0(x, 0)
<f>{y,x

+ y) = y - 2{x + y) = -2x - y - 3y, 5x + 2y) = i2x - 3y) - 2i5x + 2y) = ~8x-ly. ,f,(2x

11.15.

Let T-.V-^U be linear and


of T'
is

let

T*:U*^ V*
r'(0)

the annihilator of the image of T,

Suppose
y

e Ker

T*;

that
0(m)

is,

its transpose. Show that the kernel Ker T* = (Im T)". o y = o. If m G Im T, then m = T(v) for some

be

i.e.

V; hence

=
m

0(7(1;))

(4>T){v)

=
(Im

0{v)
2^.

=
Thus Ker T* C (Im
r)*.

We

have that

<p{u)

for every

G Im
(Im

T;

hence
that
is,

On

the other hand, suppose o

T)";

^(Im T)

{0}.

Then, for every v


0(v)

&V,
and so

(THamv)

{aoT)(v)
!;

a(T{v))

= =

=
0.

We

have that

(r'(a))('y)

Q(v)

for every

Y;

hence

T*(a)

Therefore

S Ker T*

(Im r)o c Ker TK

Both inclusion relations give us the required equality.

11.16.

Suppose rank(r)

V and U have = rank(rOV=n


T)0)

finite

dimension and suppose

T:V ^ U

is

linear.

Prove:

Suppose dim

and dim

dim ((Im

By

the preceding problem, claimed,

U = m. Also suppose rank (T) = r. Then, by Theorem = dim 17 - dim (Im T) = m - rank (T) = m - r Ker Tt = (Im T)'>. Hence nullity (T') = m r. It then follows
C/*

11.5,

that, as

rank(r)

dim

nullity (T')

m ~

(m-r)

rank

(T)

11.17.

Prove Theorem
of

relative to bases {vi,

matrix A* is [Ui] and {Vj}.


Suppose

be linear and let A be the matrix representation Then the transpose it} of U. Vm} of V and {ui, the matrix representation of T:JJ* ^ V* relative to the bases dual to
11.7:

Let

T:V -^ U
. .

.,

^(^i) T{V2)

= =

a-iiU-i

a2iUi

+ a^^u-i + + a22M2 +

+ ain^n + a2nU

.^.

258

LINEAR FUNCTIONALS AND THE DUAL SPACE

[CHAP.

11

We

want

to

prove that
r'(o'2)

1201

+ +

<*2202

+
+

ttm2'^m
{2)

r'(of)

a-in't'l

'''2ti02

0'mn</>m

where

{(tJ

and

{^j} are the bases dual to

{mJ and

{vj} respectively.

Let
T(v)

+ fejVa + + fcv. Then, by (1), = + km TivJ r(-i;i) + ^2 r(i'2) + + C-mn'^'n) = &i(aiiMi + + 2n**n) + + fcm(a'mlMl + + ajM) + fc2(a2lMl + + ka'mn)'>^n = (fciaii + fc2a21 + + fcmaml)^! + + (fcitlln + fc2*2n +
e.V and suppose
fci

k^v^

'

'

'

"

'

=
Hence for
;/

i=l

2
1,

(fciOii

+ fc22i +

+ kmOmdUi

n,

(Tt(aj)(v))

=
=

ffjCrCv))

=
fe2a.2j

0-j

(feiaii

fc2(l2i

1-

fcmmi)t
)
(^)

fciOij

k^amj

On

the other hand, for j


(aij0i

+ a2j02 +

= 1, .,n, + a^j^J(v) = =
.
.

(ajj^i

k^aij

+ a2j4'2 + k2a2j +
+

++

amj0m)(fei'^i

A:2'U2

" '

+ km^m)
{i)

k^a^j

Since v

e.V was

arbitrary,
r'(CTj)

(3)

and
aij0i
is

(4)

imply that
a2j*2

a.mj't'm'

1,

.,

which

is (2).

Thus the theorem

proved.

11.18.

Let

matrix over a be an arbitrary A are equal. column rank of

mxn

field

K.

Prove that the row rank and the

and K^ Let T:K-> K'n be the linear map defined by T{v) = Av, where the elements of X" to the usual bases are written as column vectors. Then A is the matrix representation of T relative Hence of if" and K", and the image of T is the column space of A.
rank
(T)

column rank of

A
Hence
of

By Theorem

11.7,

is

the matrix representation of T* relative to the dual bases.

rank

(T')

= =

column rank of A*

row rank

hence the row rank and the column rank of A are But by Problem 11.16, rank(r) proved in a direct way (This result was stated earlier as Theorem 5.9, page 90, and was equal.
rank(r');
in

Problem

5.21.)

Supplementary Problems
DUAL SPACES AND DUAL BASES
11.19.

Let

R3

a(x, y, z)

- R R3 -> R be the and = Ax-2y + 3z. Find (i) +


<t
:

linear functionals defined


(ii)

by

^(x, y,z)

= 2x-By + z

and

<!>

<',

3^,

(iii)

20

- 5<r.
15

11.20.

Let

be the linear functional on R2 defined by

0(2,1)

and 0(1,-2)

-10.

Find ^(x,y) and,

in particular, find

0( 2,

7).

11.21.

Find the dual basis of each of the following bases of (ii) {(1, -2, 3), (1, -1, (i) {(1, 0, 0), (0, 1, 0), (0, 0, 1)},

R^:
1), (2,

-4,

7)}.

CHAP.

11]

LINEAR FUNCTIONALS AND THE DUAL SPACE

259

11.22.

Let

be the vector space of polynomials over

of degree

2.

Let

functionals on

0i,

<f>2

and ^3 be the linear

defined by

Him) = f mat,
Here
of
11.23.
f{t)

<j>^{m)

= m),

.p^mm =
f(t).

m)
{/i(),

= a+bt + ct^eV
is

and
03}.

/'(f)

denotes the derivative of

Find the basis

fzd), ^(t)}

which

dual to {^i,

<f>2,

Suppose u,vGV some scalar k. Suppose 0,ffGy* some scalar k.


Let

and that

<f>{u)

implies

<p{v)

= =

for

all

<f,

V*.

Show that v = ku

for

11.24.

and that

<f>[v)

implies

<t(v)

for all

vGV. Show
^K

that

k^,

for

11.25.

V V

Show
11.26.

that:

be the vector space of polynomials over K. For a^K, define <pa'-V (i) 0 is linear; (ii) if a = b, then 0 7^ 0b-

by

0a(/(*))

/('')

Let

be the vector space of polynomials of degree

2.

Let
/(a),

a,b,c.K be
0b (/())

distinct scalars.

Let

0a, 0b and 0e be the linear functionals defined by 0a (/()) = that {0a, 0b, 0c} is linearly independent, and find the basis

/(6), 0c (/(*))

/(c). is its

Show
dual.

{/i(t), /2(t), fait)}

of

which

11.27.

Let V be the vector space of square matrices of order n. Let T iV T{A) = ail + 22 + + "rm> where A = (o.^). Show that T is linear.
Let a on Let
ith

^K

be the trace mapping:

11.28.

W he
V
.

a, subspace of V. For any linear functional such that a(w) = <p(w) for any w &W, i.e.

on W, show that there


is

is

a linear functional

the restriction of a to TF.

11.29.

, e} be the usual basis of K". Show that the dual basis {ci, projection mapping: viia^, .,a) = a.;.
.
.

is

Wi,

., ir}

where

vi is

the

11.30.

Let

y be

also belongs to V*.

a vector space over B. Let 0j, 02 S V* and suppose Show that either 0i = or 02 = 0.

<r

y -> B

defined by a(v)

01(1;) 02('y)

ANNIHILATORS
11.31.

Let be the subspace of B* spanned by (1,2,-3,4), (1,3,-2,6) and (1,4,-1,8). the annihilator of W.

Find a basis of

11.32.

Let

W be the subspace of B^ spanned by


that, for

(1,

1,0)

and

(0, 1, 1).

Find a basis of the annihilator of W.


the linear span of S.
Prove:

11.33.

Show
Let

any subset S of V, L{S)

S""

where L{S)

is

11.34.

and

W be

subspaces of a vector space

of finite dimension.

{U n W)"

U"

+ W.

11.35.

Suppose

V - U

W. Prove that V* =

W WO.
<l>(x,y)

TRANSPOSE OF A LINEAR MAPPING


11.36.

Let r B3
:

be the linear functional on B^ defined by

Zx

2y.

For each linear mapping

^ B2,

find (rK0))(a;, y,

z):
(ii)

(i)

T{x,y,z)

(x

+ y,y + z);
and

T(x,y,z)

(x

+ y + z,2x-y).

11.37.

Suppose
Suppose Suppose
rt(0)

S:U-^V
T:V -*U
r y
:

TiV^W
and

are linear.

Prove that (ToS)t

= StoTK
=
(Ker
T)".

11.38.

is linear

has

finite

dimension.

Prove that Im T'

11.39.

- [7

is linear

and u

U.

Prove that u

GlmT
h> Tt
is

or there exists

/>

GV*

such that

and 0(m)

1.

11.40.

Let y be of finite dimension. onto Hom {V*, V*). (Here T

Show that
is

the

mapping T

an isomorphism from

Hom (V,

V)

any linear operator on

V.)

260

LINEAR PUNCTIONALS AND THE DUAL SPACE

[CHAP.

11

MISCELLANEOUS PROBLEMS 11.41. Let y be a vector space over R. The line wv {tu + (1 t)v: a t 1). A subset S
Let
<6

segment uv joining points m, v S V is defined by of y is termed convex if u,v GS implies uvcS.

e y* and

let 0},

Prove

W+ = {vGV: 4,(v) > W and W~ are that W +


,

= {vGV:

<t>(v)

0},

W" =

{vi=,V:

<t>(v)

<

0}

convex.

11.42.

Let y be a vector space of nonzero linear functional

finite

</>

on V.

hyperplane i/ of y is defined to be the kernel of a dimension. Show that every subspace of V is the intersection of a finite

number of hyperplanes.

Answers
11.19.
(i)

to

Supplementary Problems
(iii)

6x-5y + 4z,
y)

(ii)

ex-9y + Sz,

-16x + iy -

ISz

11.20.

<p(x,

Ax

ly, 0(-2, 7)

41

11.21.

(i)

{^i{x,y,z)
{>t,i(x,

(ii)

y, z)

= X, <t,2(x,y,z) - y, <t>z{x,y,z) = z} = -3x -hy- 2z, ^ii'^, y, z) = 2x + y,


t.

</>3(x,

y,z)

2y

+ z]
0b-

11.25.

(ii)

Let

f{t)

Then

^a(/(*))

0b(/(O).

and therefore ^

11.26.

|/i()

(_6)(_e)

h(t)

(6_)(5_^)

/3W

(^_)(,_6)

11.31.

{0i(a;,

2/,

z, t)

5x

- + 2,
2/

02(.

1/. . *)

22/

- t}

11.32.

{<j>(x,

y,z)

= x-y + z)

11.36.

(i)

(Tt(<t>))(x,y,z)=Zx

+ y-2z,

(ii)

(r'(0))(a;,2/,z)

= -x +

Sj/

+ 3x.

chapter 12

Bilinear,

Quadratic and Hermitian Forms


A
V

BILINEAR FORMS Let F be a vector


mapping
(i)

space of finite dimension over a

field

K.

bilinear

form on

is

f.VxV^K

which

satisfies

f{aui
f{u,

(ii)

+ bu2, v) = avi + bvi) =

af{ui, v)

af{u, Vi)

+ +

bf{u2, v)

bf{u, Vi)
is

for

all

first

and all im, Vi e V. We express condition (i) by saying / variable, and condition (ii) by saying / is linear in the second variable.
Example
12.1:

a,b&K

linear in the

Let

<j>

and a be arbitrary linear f unctionals on V.

f(u,v)

4>(u)<t(v).

bilinear

written /

form = ^ (g)

X - be defined by Let / (Such a / is bilinear because (p and a are each linear. and <t and so is sometimes / turns out to be the "tensor product" of
:

Then

<f>

a.)

Example

12.2:

Let / be the dot product on R"; that


f{u, v)

is,

= u'v =
(h^).

a^bi
is

a^h^

a6

where u
Example
12.3:

(a^)

and v

Then /

a bilinear form on R".

Let

A =
/ on

(ffly)

be any

nXn

matrix over K.

Then
'

A may
ai2
0.22

be viewed as a bilinear
...

form

X" by

defining

an
a^i

a.i

I
I I

Vi
2/2

f(X,Y)

= XtAY =
=

{xi,X2

a;

J +

"2n

*nl

'n2

Vnl

U-

^ij^iVj

lll2/l

12'l2/2

The above formal expression in variables Xi,yi is termed the bilinear polynomial corresponding to the matrix A. Formula (i) below shows that, in a certain sense, every bilinear form is of this type.

B{V) denote the set of bilinear forms on V. placed on B(V) by defining f + g and kf by:
will let
{f

We

vector space structure

is

+ 9){u,v) = {kf){u,v) =
In fact,

f{u,v)

g{u,v)

kf{u,v)

for any f,g

& B{V) and any kE^K.


Let

Theorem

12.1:

be a vector space of dimension n over K. Let {4,^, ...,</>} be a basis of the dual space V*. Then {fij:i,j = 1,. .,w} is a basis of B{V) Thus, in particular, where fa is defined by fi}{u,v) = <j>^{u) <i>.{v).

dimB{V) =

w'.

261

262

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

[CHAP. 12

BILINEAR FORMS AND MATRICES


Let / be a bilinear form on V, and
let {ei,
. . .

e^}

be a basis of V.

Suppose u,v

eV

and suppose

u =

aiCi

ae,

biCi

bnen

Then
f{u, v)

f{aiei

+ anCn, +

biCi

bne)
n

ai&i/(ei, ei)

0162/(61,62)

ab/(en, e^)

i,3

^ =

aib}f{ei,ej)
1

Thus

is

completely determined by the n^ values

/(ei, e,).

= {an) where an f{ei, e,) is called the matrix representation of f relThe matrix ative to the basis {ei} or, simply, the matrix of f in {ei}. It "represents" / in the sense that
'&i^

f{u,v)

aibjfiei, ej)

(ai,

a)A

^|
^bnj

Me^Me
u

(1)

for

all

u,v

GV.

(As usual,

[u]e

denotes the coordinate (column) vector of

GV

in the

basis

{ei}.)

We
basis
is

next ask,
selected ?

the transition
for every

does a matrix representing a bilinear form transform when a new The answer is given in the following theorem. (Recall Theorem 7.4 that matrix P from one basis {e^} to another {el} has the property that [u]e = P[u]e'

how

uG
12.2:

V.)

Theorem

Let P be the transition matrix from one basis to another. matrix of / in the original basis, then

If

is

the

B = P'AP
is

the matrix of / in the

new

basis.
definition.
if

The above theorem motivates the following


Definition:

A matrix B
(or:

is said to be congruent to a matrix A nonsingular) matrix P such that B P^AP.

there exists an invertible

Thus by the above theorem matrices representing the same bilinear form are congruent. We remark that congruent matrices have the same rank because P and P' are nonsingular;
hence the following definition
Definition:
is

well defined.

The rank of a bilinear form / on V, written rank(/), is defined to be the rank of any matrix representation. We say that / is degenerate or nondegenerate according as to whether rank (/) < dim V or rank (/) = dim V.

ALTERNATING BILINEAR FORMS A bilinear form / on 7 is said to be alternating


(i)

if

f(v,v)

for every v

GV.

If / is alternating,

then
f{u, u)

=
and so

f{u

+ v,u + v) =
(ii)

f{u, v)

f{v, u)

f{v, v)

f{u,v)

-f{v,u)

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

263

for every u,v GV. bilinear form which satisfies condition (ii) is said to be skew symmetric (or: anti-symmetric). If 1 + 1 v^ in K, then condition (ii) implies f{v, v) = f{v, v) which implies condition (i). In other words, alternating and skew symmetric are equivalent

when

+1

9^ 0.

The main structure theorem of alternating

bilinear

forms follows.

Theorem

12.3:

Let / be an alternating bilinear form on V. Then there exists a basis of V in which / is represented by a matrix of the form

Oil
-1_04
I

l1

LzLjPJ,
I

1-1

-T-4
lOj
ro-

Moreover, the number of


it is

-1

,,

is

uniquely determined by / (because

equal to ^ rank

(/)).

In particular, the above theorem shows that an alternating bilinear form must have even rank.

SYMMETRIC BILINEAR FORMS, QUADRATIC FORMS A bilinear form / on V is said to be symmetric if


f{u,v)

f{v,u)
/,

for every u,v .V.

If

is

a matrix representation of

we can
Y'A'^X
its

write

f{X,Y)

= X'AY = {X'AYY =

(We use the fact that symmetric,


and since
versely
if

X*AY

is

a scalar and therefore equals


=.

transpose.)

Thus

if

/ is

Y*A*X = f(X,Y) = f{Y,X)


this is true for all vectors

Y'AX

is

symmetric, then /

is

X, Y it follows that symmetric.


is

A = A*

or

is

sjTnmetric.

Con-

The main

result for

symmetric bilinear forms

given in

Theorem

12.4:

(in which 1 + 1^0). Let / be a symmetric bilinear form on V over .,v} in which / is represented by a diagonal Then V has a basis {vi,
. .

matrix,

i.e.

f{vi, Vj)

for

i - j.

Alternate

Form

Let A be a symmetric matrix over X (in which 1 + 1 ^^^O). of Theorem 12.4: Then there exists an invertible (or: nonsingular) matrix P such that P*AP That is, A is congruent to a diagonal matrix. is diagonal.

264

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

[CHAP. 12

of obtaining the diagonal

a product of elementary matrices (Problem 3.36), one way is by a sequence of elementary row operations and the same sequence of elementary column operations. These same elementary row operations on / will yield PK This method is illustrated in the next example.
Since an invertible matrix
is

form P*AP

2-3
5

Example

12.4:

Let

4
8

|,

a symmetric matrix.

It is

convenient to form the block

-3

-4

matrix (A,

I) 1

2 5

(A.I)

2 3

-3 -4
8

1
1

-4

apply the operations R^ - 2/Ji + R^ and R^ -^ SR^ + R3 to {A, I), and then the corresponding operations C^ -^ 2Ci + C^ and Cg -* 3Ci + C3 to A to obtain

We

2
1

-3
2

-2
3

1 1

and then

-1

We next apply the operation C3 - 2C2 + C3 to obtain


/l
1 1

^3

->

2R2 + R3 and then

the corresponding operation

-2
7

and then

-5

-2

1 0! -2 -5 7 -2
1

Now A

has been diagonalized.

We

set

P -

and then

Pt A

Definition:

mapping q:V-*K

is called

a quadratic form

if

q{v)

= f{v,v)

for some

symmetric bilinear form / on V.

We
1

call

1 7^

in if, then / is obtainable

q the quadratic form associated with the symmetric bilinear form from q according to the identity
f(u,v)

/.

If

= Uq{u + v) /.

q{u)

q{v))

The above formula

is called

the polar form of

Now
form

if / is

represented by a symmetric matrix

A=

(an),

then q
ai2
^22

is

represented in the
ain\ /xi\
azn \l X2

(an
a21

... ...

...
ttnl

a2

...

annj \Xnj
atiXiXj

y
u

ttiiXiXj

aiiccf

022X2

annxl

+22

*<^

The above formal expression

in variables Xi is

ing to the symmetric matrix A. diagonal representation


q{X) that
is,

termed the quadratic polynomial correspondObserve that if the matrix A is diagonal, then q has the
a22xl

- X*AX - anxl +

annxl

Theorem

the quadratic polynomial representing q will contain no "cross product" terms. 12.4, every quadratic form has such a representation (when 1 + 1^0).

By

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

265

Example

12.5:

Consider the following quadratic form on R^:


<l(^>

v)

2x2

\2,xy

5j/2

One way
which
X
is

of diagonalizing q
fully described in

= s + Zt,

to

is by the method known as "completing the square" Problem 12.35. In this case, we make the substitution obtain the diagonal form

q(x, y)

2(s

+ 3t)2 -

12(s

+ U)t +

5(2

2s^

13*2

REAL SYMMETRIC BILINEAR FORMS. LAW OF INERTIA


In this section we treat symmetric bilinear forms and quadratic forms on vector spaces over the real field R. These forms appear in many branches of mathematics and physics. The special nature of R permits an independent theory. The main result follows.

Theorem

12.5:

Let / be a symmetric bilinear form on V over R. Then there is a basis of V in which / is represented by a diagonal matrix; every other diagonal representation has the same number P of positive entries and the same number of negative entries. The difference S = P is called the signature of /.

real

symmetric bilinear form /

is

said to be nonnegative semidefinite

if

q{v)

f{v,v)

for every vector v; and

is

said to be positive definite if


q{v)

f{v,v)

>

for every vector v =0.


(i)

By

the above theorem,

is

nonnegative semidefinite
if
/.

(ii)

/ is positive definite
<S is

and only if S and only if S = dim V


if

rank (/)

where

the signature of
12.6:

Example

Let / be the dot product on R"; that


f(u, v)

is,

wV
=

eii6i

+
is

a2&2

"

"n^n

where u

(ttj)

and v

(6j).

Note that /

symmetric since

f(u, v)

WV

vu

f{v, u)

Furthermore, /

is

positive definite because


f(u, u)

= a^+ al+

al

>

when M

v^ 0.

In the next chapter we will see how a real quadratic form q transforms when the transiP is "orthogonal". If no condition is placed on P, then q can be represented in diagonal form with only I's and I's as nonzero coefficients. Specifically,
tion matrix

Corollary

12.6:

Any

real quadratic
qytVif
.

form q has a unique representation


. .

in the

form

iXyn)

is

2
vCi
I

'

~T'

i^g

<jO

+1

The above result for real quadratic forms or Sylvester's Theorem.

sometimes referred to as the

Law

of Inertia

266

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

[CHAP. 12

HERMITIAN FORMS Let y be a vector space


be such that
(i)

of finite dimension over the complex field C.


f{aui

Let /

V xV -* C

+ bu2,v) = ^
fiv,u)

af{ui,v)

bf{u2,v)

(ii)

f{u,v)

where

a, 6 G C and im, v &V. Then / is called a Hermitian form on V. denotes the complex conjugate of A; G C.) By (i) and (ii),

(As usual, k

f{u, avi

+ bv2) = =
(iii)

f{avi

+ bv2, u) af{vi, u) + b f{v2, u) df{vi,u) + bf{v2,u) = df{u,Vi) + bf{u,V2) + bv2) a f{u, Vi)

That

is,

f{u, avi

b f{u, Vi)

express condition (i) by saying / is linear in the first variable. On the other hand, we express cond ition ( iii) by saying / is conjugate linear in the second variable. Note that, by (ii), f{v, v) = f{v, v) and so f{v, v) is real for every v GV.

As

before,

we

Example

12.7:

Let

A=

(dy)

be an

Ji

X w matrix over

C.

We

write

for the matrix obtained

by

taking the complex conjugate of every entry of A, that A* for A = AJ. The matrix A is said to be Hermitian
If

is,

if

A = (a^). We also write A* = A, i.e. if ay = a]^.

is

Hermitian, then f{X, Y)

^X'^AY

defines a

Hermitian form on C" (Prob-

lem

12.16).

The mapping q:V->-B, defined by q{v) = f{v, v) is called the Hermitian quadratic form or complex quadratic form associated with the Hermitian form /. We can obtain / from q according to the following identity called the polar form of /:
f{u, v)

=
.

l{q{u

+ v)-

q{u

- v)) +

j(q{u

+ iv) -

q{u

- iv))

= (fo) where feij = /(ei, e,) is suppose {ei, e} is a basis of V. The matrix By (ii), f{ei,ej) = f(ej,ei); hence {ej}. called the matrix representation of / in the basis are real. Thus any diagonal repis Hermitian and, in particular, the diagonal entries of theorem is the complex analog of resentation of / contains only real entries. The next Theorem 12.5 on real symmetric bilinear forms.

Now

Theorem

12.7:

., en} of Let / be a Hermitian form on V. Then there exists a basis {ei, for represented by a diagonal matrix, i.e. f{ei, ei) = V in which / is = j. every diagonal representation of / has the same number Moreover, i
.
.

of positive entries, and the

same number

of negative entries.
/.

The

difference

S = P-N
is

is called the signature of

Analogously, a Hermitian form /

said to be nonnegative semidefinite if

q{v)

f{v,v)

^
>

for every v

eV, and

is

said to be positive definite if


q{v)

f{v,v)

for every v =0.


Example
12.8:

Let / be the dot product on C"; that


f(u, V)

is,

U'V =
(Wj).

ZiWi
is

Z2W2

zw
Moreover, /
is

where u

({)

and v

= +

positive definite since, for


f(u, u)

Then / any v # 0,

a Hermitian form on C".

zii

Z2,Z2+

zz

l^iP

I22P

+ K\^ >

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

267

Solved Problems
BILINEAR FORMS 12.1. Let u = {xi, X2, xs)
f{u, v)

and v

(t/i, 1/2,

ya),

and
5x22/1

let

Sxiyi

2xiyi

7*21/2

8a;2i/3

4x33/2

Xzys

Express
Let

/ in

matrix notation.

be the

3X3

matrix whose i;-entry

is

the coefficient of XiVy

Then

/3 -2
f{u,v)

0\/j/i^
:

= XtAY =

(^i.xa.ajg)

7-8

\0

4 -l/\2/3i

12.2.

Let A be an X matrix over K. Show that the following mapping / form on K": f{X,Y) = X*AY. For any a,bGK and any Xj, Fj e K", = (aZi + 6X2)*^ Y = (aXj + 6X^) .4 F /(aATi + 6Z2, F) = aXlAY + bXlAY = a/(Zi, F) + 6/(X2, F)
Hence /
is linear in

is

a bilinear

the first variable.

Also,

/(X, aFi

+ ftFa)

XtA(aFi

6F2)

= aX^AYi + bX^AY^ =
is

a /(X, Fj)

6 /(X, F2)

Hence /

is linear in

the second variable, and so /

a bilinear form on K".

12.3.

Let / be the bilinear form on R^ defined by


f{{xu
(i)

X2), (yi, 2/2))

2xi2/i

3xi2/2

X22/2

(ii)

(iii)

Find the matrix A of / in the basis {Ui (1,0), 112 = (1, 1)}. Find the matrix B of / in the basis {vi = (2, 1), V2 = (1, -1)}. Find the transition matrix P from the basis {mi} to the basis
that

{Vi},

and verify

B^P*AP.
(ay)

(i)

Set

A =

where ay

= /(ttj, Uj): = f(ui,ui) = an ai2 = /(mi,M2) = 021 = /(M2.M1) = 022 = /(M2.W2) =


*^

/((1,0), (1,0)) /((1,0), (1,1)) /((1,1), (1,0)) /((l.l), (1,1))

=2-0 + 0=
= 2-3 + = 2-0 + = 2-3 +

= -1

0=
1

Thus

A =

=(r;)'
(

ft

*^ matrix of / in the basis {u^, n^}.

(ii)

Set

B=

(6y)

where 6y
611

612
621

622

= = = = =

/(Vj,-!;,):

/(^i.i'i)
/(^i.'y2)
f(->'2,'>'i)

/(^2.^2)

= = = =

/((2,1), (2,1)) /((2.1), (1,-1))

/((l.-l), (2,1))
/((I, -1), (1,-1))

= 8-6 + 1 = 4 + 6-1 = 4-3-1 = 2+3+1

= = = =

3 9

Thus

B =

/3
I

9\
ft

's

^^^ matrix of / in the basis {vi, v^}.

(iii)

We

must write

Vi

and V2

in

terms of the

Mj:

ri

V2

= =

(2,1)

(1,0)

(1,1)

(1,-1)

2(1,0)

-(1,1)

= M1 + M2 = 2Mi-M2

268

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

[CHAP. 12

Then

^ =

_j)

and so

P'

Q _M

Thus

12.4.

Prove Theorem 12.1: Let F be a vector space of dimension n over K. Let {^j, ^} be a basis of the dual space V*. Then {/: ij" = 1,. .,%} is a basis of B{V) where / is defined by f^.{u,v) = ^.(m)^.(v). Thus, in particular, dimB(F) = n\
.
. .

Let

{e^,

.,

e} be the basis of

and suppose
(2aij/ij)(es,

/(ej, e^)
ej)

=
s,t

ay.

We

V dual to {^J. claim that /

We first show that {/} = ^ay/y. It suffices


ej)

spans B(y). Let /


to

show that

/(e^.e,)

G B(y) =

for

l,...,n.

We

have

(2 ay/y)(es, et)
as required.
It
s,t

2ay/(es,

=
"st

2 Oy ^i(es) ^^(et)
=
f(es,et)

=
Hence
{/y} spans B{V).
is

2ayisSjt

remains to show that {/y}


.

linearly

independent.

Suppose
a^s
is

Soy/y =

0.

Then

for

1,.

.,n,

=
The
last step follows as above.

0(es, et)

= (2 ao/y)(es, Bf) =
and hence

Thus

{/y} is independent

a basis of B(V).

12.5.

denote the matrix representation of a bilinear form / on F relative to a basis Show that the mapping / i- [/] is an isomorphism of B{V) onto the vector space of w-square matrices.

Let
{ei,

[/]
.

.,e) of V.

Since /
onto.

is

It suffices to

completely determined by the scalars show that the mapping / l-> [/]

f(e^, ej),

the

mapping

/*"*[/]
is,

is

one-to-one and

is

a homomorphism; that

that
(*)

[af+bg]
However, for
i,j

= =

a[f]

b[g]

1,.

.,n,

(af

bg){ei, e^)

/({, e^)

bg(ei, Bj)

which

is

a restatement of

(*).

Thus the

result is proved.

12.6.

Prove Theorem

12.2:

Let

basis {gj}. If is the matrix of / in the original basis matrix of / in the new basis {e\}.
Let u,v
P[v]e-

be the transition matrix from one basis {e,} to another {Ci}, then B = P'^AP is the
have
P[u]g,

eV.
hence

Since
[u]l

[v]e-

P is the transition matrix from {e^} to {e,-}, we = [u]l, PK Thus f{u,v) = [u]lA[v], = [u]l.PtAP[v],.
of V, P^ A

[u]g

and

Since

u and v are arbitrary elements

is

the matrix of / in the basis {e^}.

SYMMETRIC BILINEAR FORMS. QUADRATIC FORMS


12.7.

Find the symmetric matrix which corresponds to each of the following quadratic
polynomials:
(i)

q{x, y)
q{x, y)

(ii)

= -

4x^

6xy

7y^

(iii)

q{x, y, z) q(x, y, z)
. .

xy

+ y^
A
(uy)

(iv)

= =

3x^
x^

+ 4xy - y' + - 2yz + xz

8xz

Qyz

z^

The symmetric matrix

the coefficient of xf and has the entries ay and

representing q(xi, .,x) has the diagonal entry a^ equal to ajj each equal to half the coefficient of ajjajj-. Thus

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

269

(ii)

12.8.

For each of the following


such that P*

real

symmetric matrices A,
its

find

a nonsingular matrix

AP

is

diagonal and also find

signature:

(i)

A -

-:

(ii)

A =

(i)

First form the block matrix (A,

I):

{A, I)

Apply the row operations

sponding column operations C^


1

R2^^Ri + R2 and iZa - 2i2i + iJj -> SCj + C^, and C3 - 2Ci + C3
1

to (A,/)
to

and then the corre-

to obtain
1

-3 -2
1

2
1

0\
1

/I

and then

-2
\o
1

1 1

-2
B3

1/
- i?2

-2

Next apply the row operation


C3-* C2

+ ^^3

^^^ then the corresponding column operation

+ 2C3

to obtain

and then

Now A

has been diagonalized.

Set

then

P^AP -

The signature S of

is

5 =
I):

2-1 =

(ii)

First form the block matrix (A,

1 1 1

(A,/)

1 1

-2
2

-1

In order to bring the nonzero diagonal entry 1 into the first diagonal position, apply the row operation Ri <-> R^ and then the corresponding column operation Ci -> C3 to obtain
1 1

-1
2
1 1 1

1\

-1
and then
2
\

1 1 1 1

-2
1

-2
1

0/
i?2 ~*

Apply the row operations


column operations C^
1
-*

2Ci

+ C^

2Bi + -^2 ^nd JB3 -> iJj + R^ and C3 ^ Ci + C3 to obtain


1\

and then the corresponding

2
2
3

/-I
and then
\

3
1 1

2
3

3
1 1

2
1

1/

270

BILINEAR, QUADRATIC
Apply the row operation
C3
-

AND HERMITIAN FORMS

[CHAP. 12

iJg ->

3^2 + 2R3
l\

and then the corresponding column operation

-3C2 + 2C3
/-I

to obtain

2
\

12
Set

/-I
and then
\

1^

12
-14
2

-7

2-3-4/
/O
2\
1

-3 -4,

Now A

has been diagonalized.

P =
\l

2
1

-3 -4/

then

P'AP =

The signature S of

is

the difference

S =

1.

12.9.

'

Suppose 1 + 1 v^ in K. Give a formal algorithm to diagonalize (under congruence) a symmetric matrix A = (an) over K.
Case
I:

aii=0.

Apply the row operations


Cj -*

fij ->

ajxi?x + OxxiJj, A

2,

.,n,

and then the


0"

corresponding column operations

Oji Cj + an C;

to reduce

to the

form

/ill
(

^0

Case II: a^ = but a^ ^ 0, for some i > 1. Apply the row operation Ri ^i2j and then the corresponding column operation Cj <-> to bring ctjj into the first diagonal position. This reduces the matrix to Case I.

All diagonal entries Oji = 0. Choose i, j such that ay = 0, and apply the row operainto the Rj + Rf and the corresponding column operation Ci-* Cj + Cj to bring 2ay # ith diagonal position. This reduces the matrix to Case II.

Case

III:

tion

i?i ->

In each of the cases,

we can

finally

reduce

to the

form

/ail
(

0\
_
)

where

is

a symmetric

matrix of order

less

than A.

By

induction
1

we can
in

finally bring
is

into diagonal form.

Remark: The hypothesis that

+1

#^

K,

used in Case III where

we

state that 2ay = 0.

12.10.

Let q be the quadratic form associated with the symmetric bilinear form /. Verify (Assume that the following polar form of /: fiu,v) - ^q{u + v) - q{u) - q{v)).
1

+ 1^0.)
q(u

+ v)

q{u)

qiv)

= = =

f(u

f(u, u)

+ v,u + v) f{u, u) f(v, v) + f{u, v) + f{v, u) + fiy, v) -

f{u, u)

f(v, v)

2f{u,v)

If

1 7^ 0,

we can

divide by 2 to obtain the required identity.

12.11.

Prove Theorem
1

12.4:

matrix,
Method
If

+ 1^0). Then V i.e. f{Vi, v,) =


1.

(in which Let / be a symmetric bilinear form on V over represented by a diagonal has a basis {vi, .,Vn) in which / is
. .

for

i - j.

dim V

then the theorem clearly holds. Hence we can suppose f = Q and - for every v&V, then the polar form of / (see Problem 12.10) implies that / = 0. Hence we can assume there is a vector t?i e V such that f(Vi, v^ = 0. Let consist of those vectors 1; G y for which /(^i, v) = 0. U be the subspace spanned by Vi and let
/ or
if

= = n>

dim V

1,

1.

If

q(v)

f(v, v)

We
(i)

claim that

V = U W.

Proof that
Since fore u

UnW =
=
=
0.

{0}:

Suppose

uG U nW.
=

Since

ue^U, u

kv^ for some scalar

k&K.

uGW,
=
kvi

f{u,u)

Thus

UnW = {0}.

f(kvi,kvi)

k^ f(Vi,Vi).

But

/(^i.-Wi) ?^ 0;

hence

k-0

and there-

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

271

(il)

Proof that

V=

U + W:

Let

vG

V.

Set

Then Thus and

f{v^,w)

f(vv)

- ;^^/(^i'^i) =

"

By

(i)

w G W. By (1), v is (ii), V = U W.
/ restricted to

the

sum

of an element of

and an element of W. Thus

V = U + W.

Now

is

induction there is a basis {^2. . v} of by the very definition of W, fiv^, Vj) = has the required property that /(-Uj, Vj) =

n. 1; hence by a symmetric bilinear form on W. But dim such that f(v^, Vj) = for i = j and 2 i, j n. But v} of V for j = 2, .,n. Therefore the basis {v-^,

for

i = j.

Method

2.

is congruent to a The algorithm in Problem 12.9 shows that every symmetric matrix over diagonal matrix. This is equivalent to the statement that / has a diagonal matrix representation.

12.12.

Let

A =

^
I 1

a diagonal matrix over K.

Show

that:

(i)

for any nonzero scalars with diagonal entries aifcf


if

fci,

fc

e
is

If ,

is

congruent to a diagonal matrix

(ii)

K K

is

I's
(iii)

and
is

the complex field C, then O's as diagonal entries;

congruent to a diagonal matrix with only

if

I's,

the real field K, then A is congruent to a diagonal matrix with only I's and O's as diagonal entries.
be the diagonal matrix with diagonal entries
fc^.

(i)

Let

Then

ptAP =

^"2
I

11

02

02^2

O-nKl

(ii)

Let

be the diagonal matrix with diagonal entries

f>i

fl/Voi
"]
-.

if { '^

r,

Then P^AP has

the required form.

(iii)

be the diagonal matrix with diagonal entries the required form.

Let

6j

<

fl/A/hl
^

if

Oi^O _
'

Then P^AP has

Remark. We emphasize that (ii) is no longer true gruence (see Problems 12.40 and 12.41).

if

congruence

is

replaced by Hermitian con-

12.13.

Let / be a symmetric bilinear form on V over R. Then there is a basis of V / is represented by a diagonal matrix, and every other diagonal representation of / has the same number of positive entries and the same number of negative entries.

Prove Theorem

12.5:

in

which

u^} of V in which / is represented by a diagonal By Theorem 12.4, there is a basis {itj, ., w} is another basis of matrix, say, with P positive and negative entries. Now suppose {wi, V in which / is represented by a diagonal matrix, say, with P' positive and N' negative entries. We can assume without loss in generality that the positive entries in each matrix appear first. Since = P' + N', it suffices to prove that P = P'. rank (f) - P +
. . . ,

272

BILINEAR, QUADRATIC
Let
f{v,v)

AND HERMITIAN FORMS

[CHAP. 12

[7

Note that dimU = P and dimW = n- P'. Thus dim{U+W) = dimU + dimW - dim(UnW) = P + {n-P')-0 = P But dim(U+W) ^ dim V = n; hence P-P' + n^n or P ^ P'. Similarly, P' ^ P fore P = P', as required.

UnW = {0}.

>

be the linear span of u^, .., up and let be the linear span of Wp, + 1, for every nonzero v e U, and f{v,v) ^ for every nonzero v
.

m).

& W.
P'

Then Hence

+ n

and there-

Remark. The above theorem and proof depend only on the concept of theorem is true for any subfield K of the real field R.

positivity.

Thus the

12.14.

An nxn real symmetric matrix A is said to be positive definite if X*AX > for every nonzero (column) vector G R", i.e. if A is positive definite viewed as a bilinear form. Let B be any real nonsingular matrix. Show that (i) B*B is symmetric and (ii) B*B is positive definite. {Btpy = ptP" = B<-B\ hence B'JS is symmetric. (i)

(ii)

Since
itself,

is

nonsingular,

BX #

for any nonzero


is positive.

X S R".

Hence the dot product of

BX

with
as

BX-BX =

(BXY(BX),

Thus XHBtB)X

{XtBt){BX)

= (BX)HBX) >

required.

HERMITIAN FORMS
12.15.

Determine which of the following matrices are Hermitian:

2-i
6
i
(ii)

+
i

i\

A
(i)

matrix

A =
is

(Oj^)

is

Hermitian

iff

A=

A*,

i.e.

iflf

o.^

= 'ajl.

(ii)

(iii)

The matrix The matrix The matrix

Hermitian, since

it is

equal to

its

conjugate transpose, symmetric,


is

is

not Hermitian, even though

it is

is

Hermitian.

In fact, a real matrix

Hermitian

if

and only

if it is

symmetric.

12.16.

Let

A
For

be a Hermitian matrix.

Show
Ye
C",

that /

is

a Hermitian form on C where /

is

defined

all

by f(X, a, 6 e C and

Y)^X<^A
all

Y.
X^, X^,

/(aXi + 6X2,

y)

= =

+ hX^YAY = (aX\+hXl)AY aX\AY + bXlAY = af(Xi,Y) +


(aX^
Also,

bf(X2,Y)

Hence

/ is linear in the first variable.

f{X, Y)

= xTaY = (XTaYY =

Wa^X = YtA*X =
We

Yt

AX =
is
a.

f(Y,X)
scalar and so
it is

Hence /
equal to

is

a Hermitian form on C". (Remark.


transpose.)

use the fact that X*

AY

its

12.17.

Let
V.
(i)

be a Hermitian form on V.
that:

Let

H be the matrix
{ei}

of / in a basis {d,

. ,

e} of

Show
f{u,v)
if

[u]lH\v]e tor

al\

u,vGV;
from
is

(ii)

is

the transition matrix

(or:

B-Q*HQ
(ii)

where Q = P)

new basis {e,'} of V, then B = matrix of / in the new basis {e,'}. the
to a

P*HP

Note that
(i)

is

the complex analog of


and suppose
ajCi

Theorem
and
v

12.2.

Let u,v

GV
u =

0362

ae

b^ei

62^2

&

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS


6161

273

Then

f{u, v)

fia^Ci

+ ae,

6e)

as required.
(ii)

"

Since

is

the transition matrix from {ej to {e^}, then


P[u],.

= =

[u] P[v],,

[v],

and so
[v]^,.

[m]^

[m]*, P',

[^= P ['^

Thus by
hence P*

(i),

f(u,v)
is

{u]\H

[v]^

[u\l,P^HP

But u and v are arbitrary elements of V;

HP
1

the matrix of / in the basis {el}.

1
i

+
4

2i

12.18.

Let

H =
P

3i,a

Hermitian matrix.

Find a nonsingular ma-

-2i
trix

such that

1 + Si j P*HP is diagonal. 2
I):

First

form the block matrix {H,


1 1

1
j

+
4

2i

2
3i

3i

-2i

Apply the row operations

R2 -* (1 + 'i)Ri + R2 ^^^ ^3 ^ 2i/2i + i?3 to (A, /) and then the corresponding "Hermitian column operations" (see Problem 12.42) C2 -> (1 t)Ci + C2 and C3 - 2iCi + Cg to A to obtain

and then

Next apply the row operation R^ tion C3 -* hiC^ + 2C3 to obtain

->

SiBg + Zi^a

and the corresponding Hermitian column opera-

and then

Now

has been diagonalized.


'1

Set
i

-1 +
1

+
2

9i^

P =

-hi

and then

P^HP

,0

Observe that the signature

of

is

jS

1 =

1.

MISCELLANEOUS PROBLEMS
12.19.

Show

that any bilinear form / on F is the sum of a symmetric bilinear form and a skew symmetric bilinear form. Set g(v,,v) = ^[f{u,v) + f(v,u)\ and h{u,v) = ^[f(u,v) f(v,u)\. Then g is symmetric because
g{u,v)

^[f(u,v)

f(v,u)]

^[f{v,u)

f{u,v)]

=
=

g(v,u)

and

/i

is

skew symmetric because


h{u,v)

^[f{u,v)

f{v,u)]

-^[/(v.m)

-/(,-!))]

-h(v,u)

Furthermore, f

+ h.

274

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

[CHAP. 12

12.20.

Prove Theorem
a basis of

Let / be an alternating bilinear form on V. in which / is represented by a matrix of the form


12.3:
1
I

Then there

exists

-1
I

1-10

L_oj

T,
To"
1

Moreover, the number of ( \'^ to i[rank (/)]).

"/

M^

uniquely determined by / (because

it is

equal

= fcifcj f(u, m) =
If

0,

then the theorem is obviously true. Also, if dim y = and so / = 0. Accordingly we can assume that dim F > 1
0,

1,

then

fik^u, fcaw)

and f

- 0.

G 7 such that /(mj, u^) # 0. In fact, multiplying that f{ui, u^ = 1 and so /(m2, "i) = 1- Now Ui and 0. M2 are linearly independent; because if, say, u^ = fewj, then /(mj, u^) = /(mi, ku^) = k f(ui, u^) = Let C7 be the subspace spanned by Ml and M2, i.e. U = L{ux,u^. Note: n 1
Since
/

there exist (nonzero)


factor,

Wj, Wg

Ml by an appropriate

we can assume

(i)

the matrix representation of the restriction of / to


if

?7 in

the basis {mi, Mg}

is

_^

(ii)

uG

U, say u

aui

6M2,

then

/(m. Ml) f(u, M2)

= =

/(aMi f(aui

+ +

6m2, Ml) bu2, M2)

= 6
=
"'

Let TF consist of those vectors

wGV
is

such that /(w,Mi)


f(w, m)

and /(w,M2)

0.

Equivalently,

We claim that V = U+W. Let veV.


Since

W V=UW.
Set

= {wGV:
It

for every

uGU}
and
v
so
it

clear that

UnW = {0},
and

remains to show that

M = M
is

f{v,

u^uy

/(, Mi)m2
Mj,

w =
show

f(u,U])

f(v,Ui);

a linear combination of Mi and hence


/(W, Ml)

uGU. We
=
fiv, Ml)

-u that weW.
=

U)

By

(i)

and

(ii),

f(v

- U, - M,

Ml)

/{m. Ml)

Similarly, /(m, M2)

f(v, Mg)

and so

f{w, M2)

=
m

/(f

M2)

/(v, U2)
f/

/(m, M2)

=
This shows that

Then

wG

TF and

so,
?7

by

(i),

+ w where m G

and

w&W.

V^U+W\

and therefore

y=

TF.

there exists is an alternating bilinear form on W. By induction, the restriction ot f to has the desired form. in which the matrix representing / restricted to .,m of IF a basis M3, desired form. Thus Mi,M2,M3,. .,M is a basis of V in which the matrix representing / has the

Now

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

275

Supplementary Problems
BILINEAR FORMS
12.21.

Let M
(i)

(xi,

x^ and
2a;i2/2

(i/x,

y^. Determine which of the following are bilinear forms on R^:


(iv)

/(m, V)

(ii)

f(u, v)

(iii)

/(m,v)

= = =

3/2

SajjJ/i

f(u, v) f{u, v)
/(w, v)

xi

(v)
(vi)

3a;2V2

= = =

X1X2
1
0.

j/iJ/2

12.22.

Let / be the bilinear form on R2 defined by


/((!, X2), (vi, 2/2))
(i)

SxiVi

2x12/2

4x2yi

a;2?/2

Find the matrix


Find the matrix Find the

(ii) (iii)

A of / in the basis B of / in the basis transition matrix P from

{u^
{v^

= (1, 1), Mj = (1, 2)}. = (1, 1), V2 = (3, 1)}. B=


P* A P.

{mJ to {vj and verify that

12.23.

Let

be the vector space of 2

X 2 matrices over

B.

Let
(i)

Af

^ j,
is

and

let

f{A,B)

tr (At MB)
(ii)

where

A,B&V

and "tr" denotes


'1
-''

trace,

Show

that /

a bilinear form on V.

Find the matrix of / in the basis

0\ ^

/O (^

o)'\o

1\ 1^ ^^ /o oj' [1
Prove:

0\ ^

/O 1^

0^
1,

o)'[q

12.24.

Let B{V) be the set of bilinear forms on


(i)

over K.

if

f,gGB(V), then

+ g and

kf, for

of the vector space of functions


(ii)

G K, also from V X V into K;


k
f{u, v)

belong to BiV), and so B(V)

is

a subspace

if

and a are linear functionals on V, then

,p{u) a(v)

belongs to B{V).

12.25.

Let / be a bilinear form on V. For any subset


S-*-

of V,
S"""
(ii)

we write

{ve^V: f{u,v)

for every

mSS},

= {vGV
SjCSg

f{v,u)

=
S2 c

foreverywGS}

Show
(iii)

that: (i) S"*" {0}-L ={0}T

and S^ are subspaces of V;

implies

st

and

sj C s}

=y.
form on V, then
rank
(/)

12.26.

Prove:

If / is a bilinear
y-*"

= dim y -

dimy-*"

= dim y - dimy^

and

hence dim

dim y^. For each


Prove:

12.27.

Let / be a bilinear form on V.


u(x)
(1)

uGV,

let

u .V

-^

and

u iV

-*

be defined by

=
As.

f(x,u)
>^

and u{x)=f(u,x).
linear,
it

u and u are each


u
h*

i.e.

^ ^ u,u G V*;

(ii) (iii)

u and m
(/)

are each linear mappings from


m)

into V*;

rank

= rank (m K

= rank {u K

tt).

12.28.

is

that congruence of matrices is an equivalence relation, i.e. (i) A congruent to B, then B is congruent to A; (iii) if A is congruent to then A is congruent to C.

Show

is

congruent to A; (ii) if A and B is congruent to C,

SYMMETRIC BILINEAR FORMS. QUADRATIC FORMS


12.29.

Find the symmetric matrix belonging to each of the following quadratic polynomials:
(i)
(ii)

q(x, y, z) q(x, y, z) q(x, y, z)

(iii)

(iv)

q(x, y, z)

= = =

- Sxy + y^ - IGxz + xz + y^ xy + y^ + 4xz + z^ xy + yz.


2x^
x^

liyz

52

276

BILINEAR, QUADRATIC

AND HERMITIAN FORMS


matrix

[CHAP. 12

12.30.

For each of the following matrices A,

find a nonsingular

such that P'^AP


1

is

diagonal:

1-2-3^
g 6 9
9

(i)

^ = (3

3\
4)'
()

A =

-2

! 6

-9

\
.

(iii)

A =

(12-5-1
,

2 _g -2-5

-3 -1
In each case find the rank and signature.

11

12.31.

Let q be the quadratic form associated with a symmetric bilinear form


alternate polar

/.

Verify the following

form of

/:

f(u,v)

^[q(u

+ v)

q(u

v)].
Show
that:

12.32.

Let S(V) be the set of symmetric bilinear forms on V.


(i)

S{V)

is

a subspace of B(V);

(ii)

if

dim

V=

n,

then

dim S(y)
with
the

^n(n

+ 1).
form
qix,y)

12.33.

Let / be the symmetric bilinear form ax^ + bxy + cy^. Show that:
(i)

associated

real

quadratic

is

nondegenerate

if

and only

if
if

b^

(ii)

/ is positive definite if

and only

4ac = 0; and b^ 4ac < a >


Show
n

0.

12.34.

Suppose A is a real symmetric positive P such that A = PP.


Consider a real quadratic polynomial
(i)

definite matrix.

that there exists a nonsingular matrix

12.35.

qix^,

a;)

ijj

2 =

Oij^i^j,
1

where

ay

=
=

ttjj.

If

an

'^ 0,

show that the substitution


Xi

Vi

(ai22/2

+ -ln2/n).
q'iVz,

2/2.

'

q'

2/n

aji

yields

the

equation

q{xi

x^)

= a^yl +

-yVn),

where

is

also

quadratic

polynomial.
(ii)

If

an =

but, say,

ajg

^ 0,

show that the substitution


X2

xi-yi + 2/2.
yields the equation

= yi- V2>

2/3.

^n
i.e.

Vn
(i).

q{xi,...,x)

2 Mi^/jis

"^^^^^

^n

^^ 0.

reduces this case to case

This method of diagonalizing q


in the preceding

known

as "completing the square".

12.36.

Use steps of the type


12.29 to diagonal form.

problem to reduce each quadratic polynomial Find the rank and signature in each case.

in

Problem

HERMITIAN FORMS
12.37.

For any complex matrices A, B and any k e C, show


(i)

that:
(iv)

ATB = A+B,

(ii)

M^JcA,

(iii)

AB = AB,

A*

A^.

12.38.

For each of the following Hermitian matrices H,


diagonal:

find a nonsingular

matrix
,

such that P*
.

HP

is

^^.,

Find the rank and signature in each


Let

case.

12.39.

be any complex nonsingular matrix.

Show that

H = A*A

is

Hermitian and positive

definite.

12.40.

We say that B is Hermitian congruent to A B = Q*AQ. Show that Hermitian congruence

if is

there exists a nonsingular matrix an equivalence relation.

such that

CHAP.

12]

BILINEAR, QUADRATIC

AND HERMITIAN FORMS


. .

277

12.41.

Prove Theorem 12.7: Let / be a Hermitian form on V. Then there exists a basis {e^, .,e} of V which / is represented by a diagonal matrix, i.e. /(ej, ej) = for t # j. Moreover, every diagonal representation of / has the same number P of positive entries and the same number N of negative entries. (Note that the second part of the theorem does not hold for complex symmetric bilinear forms, as seen by Problem 12.12(ii). However, the proof of Theorem 12.5 in Problem 12.13 does carry over to the Hermitian case.)
in

MISCELLANEOUS PROBLEMS
12.42.

Consider the following elementary row operations:


[ai]

Bi <^ Rj,

[oj

Ri

-*

kRi, k - 0,

[og]

jB;

kRj

+ fij

The corresponding elementary column operations


Ih]
If

are, respectively,
[bs]

Ci^Cj,

[bzl

Ci-^kCi,ky^O,

d^kCj + Ct
Ci-*kCj

is

the complex field C, then the corresponding Hermitian column operations are, respectively,
[ci]

Ci'^Ci,

[cj

Cj^feCj, fc^O,
[6J

[cs]

+ Ci

(i)

Show that the elementary matrix corresponding to matrix corresponding to [a^].


Show
that the elementary matrix corresponding to
[aj.

is

the transpose of the elementary

(ii)

[cj

is

the conjugate transpose of the ele-

mentary matrix corresponding to

12.43.

Let

and
if:

and

W be vector spaces over K.


+
bv2,

mapping /

VxW-^ K

is called

a bilinear form on

(i)

f(avi
f(v,

w)

(ii)

owj

+ hWi)
V;

= =

af(vi,

w)

af(v, Wj)

+ bfiv^, w) + bf^v, w^
the following:

for every
(i)

a,bQK,

V, w^

W. Prove

The set B(V, W) of bilinear forms on from into K.

VXW

and

W
/y

is

a subspace of the vector space of functions


is

(ii)

If {^1, ...,0^} is a basis of


j

V* and

{aj,

...,}
is

= l,...,n}

is

a basis of

B(V,W) where
then

dim B(V, W)

= dim F
if

a basis of W*, then defined by fij(v,w) -

{/y
4>i(v)

i=l,...,m,

aAw).

Thus

dim W.

{Remark. Observe that

V = W,

we

obtain the space B{V) investigated in this chapter.)

m times
12.44.

Let

m-linear)

be a vector space over K. A mapping f: form on V if / is linear in each variable,

VXVX---xV-*K
i.e.

is

called

a multilinear

(or:

for

l,

...,m,

f{...,au+bv,
where ^ denotes the
ith component,

...)

af(...,u, ...)

bf(...,v, ...)
fixed.

and other components are held

An

i-linear

form /

is

said to be alternating if
f{vi,
. . . ,

Vf^)

whenever

Vj

v^,

i^^ k

Prove:
(i)

The The

set

VXVX
(ii)

B^(V)

XV

of m-linear forms on into K.

is

a subspace of the vector space of functions from

set A,(y) of alternating m-linear


1.

forms on

is

a subspace of B^{V).

Remark Remark

If
If.

m = 2,
V=

then

we

obtain the space B(F) investigated in this chapter.

2.

K'", then the determinant function is a particular alternating m-linear

form on V.

278

BILINEAR, QUADRATIC

AND HERMITIAN FORMS

[CHAP. 12

Answers
12.21.
(i)

to
(iv)

Supplementary Problems
(v)

Yes

(ii)

No

(iii)

Yes

No

No

(vi)

Yes

12.22.

(i)

A =

4 7

1
(ii)

-4
20 32
(iii)

P =

-2 -2

12.23.

(ii)

2
12.29.
(i)

-4 -8'
(ii)

1-417
1-8
7 5,

10
-i

-i^
(iii)

'0

i
1

2'
(iv)

'o

i
2

U
.0

o'

0;

1/

0,

12.30.

(i)

P =

-3
2

PtAP
0\
3
,

-2

S =

0.

'1
(ii)

2
1

P -

PtAP

=02
loo

/^

S =
-38 y

1.

2/
/l
(iii)

-1 -1
1

26 13

P =
\o

19
1

PtAP =

S =

2.

12.38.

(i)

P = P =

A
1

ptHP =
3i

^1

S =

2.

-2 +
1

(ii)

PtHP =
i

-14
'1

1
(iii)

-3 +
i

0'

P =

PtHP
lO

10
0-4,

S =

1.

chapter 13

Inner Product Spaces


INTRODUCTION
involves an arbitrary field K. In this chapter we either the real field or the complex field C. In the first case we call V a real vector space, and in the second case a complex vector space.
restrict

The

definition of a vector space

K to be

V R

Recall that the concepts of "length" and "orthogonality" did not appear in the investigation of arbitrary vector spaces (although they did appear in Chapter 1 on the spaces R" and C"). In this chapter we place an additional structure on a vector space V to obtain an

inner product space, and in this context these concepts are defined.

V shall denote a vector space of finite dimension unless otherwise In fact, many of the theorems in this chapter are not valid for spaces of infinite dimension. This is illustrated by some of the examples and problems.
emphasize that
stated or implied.

We

INNER PRODUCT SPACES We begin with a definition.


Definition:

vectors u,v

be a (real or complex) vector space over K. Suppose to each pair of there is assigned a scalar {u, v) G K. This mapping is called an inner product in V if it satisfies the following axioms:

Let

GV

[/i]

[h]
[la]

+ hu2, V) = {u,v) = (v/u)


{aui,

a(ui, v)

h{U2, v)

{u,

u) s= 0;

and

{u, m)

if

and only

if

u=

0.

The vector space


Observe that {u,u)
sense.
is

with an inner product


[72],

is called

an inner product space.


relation in [h]

always real by
||m||

and so the inequality


'\/{u,

makes

We

also use the notation

u)

This nonnegative real number ||m|| is called the norm or length of u. [/2] we obtain (Problem 13.1) the relation
{u,

Also, using

[/i]

and

avi

bv2)

d{u, Vi)

b{u, V2)
[/2]

If the base field

K is

real,

the conjugate signs appearing above and in

may

be ignored.

In the language of the preceding chapter, an inner product is a positive definite symmetric bilinear form if the base field is real, and is a positive definite Hermitian form if the base field is complex.

A real inner product space is sometimes called a Euclidean space, and a complex inner product space is sometimes called a unitary space.
279

280
Example

INNER PRODUCT SPACES


13.1

[CHAP. 13

Consider the dot product in R":

Mv
where u

ajfei

a262

Ok6

= (6;). This is an inner product on R", and R" with this Although there are usually referred to as Euclidean n-spa-ce. many different ways to define an inner product on R" (see Problem 13.2), we shall assume this inner product on R" unless otherwise stated or implied.
(aj)

and v

inner product

is

Example

13.2:

Consider the dot product on C":

where u (Zj) and v = (Wj). As in the real case, this is an inner product on C" and we shall assume this inner product on C" unless otherwise stated or implied.

Example

13.3:

Let V denote the vector space oi product in V:

mXn matrices
(A,B)

over R.

The following

is

an inner

tr (B*A)

where

tr stands for trace, the

sum

of the diagonal elements.

Analogously, if U denotes the vector space of following is an inner product in U:

mXn

matrices over C, then the

{A,B}

tr(B*A)
of the matrix B.

As
Example
13.4:

usual,

B* denotes the conjugate transpose

Let

Then the following

be the vector space of real continuous functions on the interval is an inner product on V