SCHAVM'S OUTLINE OF
ALGEBRA
LINEAR
ed
BY
SEYMOUR LIPSCHUTZ,
Temple University
Ph.D.
Copyright 1968 by McGrawHill, 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
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
63
Examples of vector
Row
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
150
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
197
CONTENTS
Page
Chapter
10
CANONICAL FORMS
Invariance. Invariant directsum decomPrimary decomposition. Nilpotent operators, Jordan canonical positions. form. Cyclic subspaces. Rational canonical form. Quotient spaces.
Introduction.
222
Triangular form.
Chapter
11
249
Linear functionals and the dual space. Dual basis. Second dual Annihilators. Transpose of a linear mapping.
Chapter
12
BILINEAR, QUADRATIC
Law
of inertia.
261
Bilinear forms. Bilinear forms and matrices. Alternating bilinear forms. Symmetric bilinear forms, quadratic forms. Real symmetric bilinear forms.
Hermitian forms.
Chapter
IB
279
unitary spaces.
Spectral theorem.
Appendix A
315
Product
sets.
elements.
Set
operations.
Relations.
Equivalence
relations.
Appendix B
ALGEBRAIC STRUCTURES
Introduction.
320
fields.
Groups.
Modules.
AppendixC
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)
The resultant u + v of two vectors u obtained by the socalled 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.
by ordered
(i)
Addition:
If
(a, &)
and
(c,
d) are the
u + v, as shown
(a
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
nspace.
particular
(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:
(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)
Example
Suppose (xy, x
+ y, z1) =
(4, 2, 3).
X x
y
y
= =
4:
2
3
21 =
Solving the above system of equations gives
3,
1, and z
A.
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
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
is
(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:
k, k'
(ii)
(iii)
(iv)
+ v) + w = u + u u+ u + {u) = u +v V +u
(u
is
{v
+ w)
(v)
(vi)
(vii)
(viii)
Remark:
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
Example 15:
Let m
(1,
2,
3,
4),
y
(6, 7, 1,
2)
and
w=
(5,
4,
5, 7).
Then
3
(2)'7
31 + (4)'(2)
(2) (4)
+ 35 +
(4) 7
= 614 + 3 + 8 = = 5 + 8 + 1528 =
are orthogonal.
Theorem
1.2:
fc
R:
{u
(ii)
+ v)"W {ku) V =
'
=^
uw +
vw
(iii)
k{u v)
(iv)
= vu uu^O, and
wv
wm =
iff
ud
Remark:
The space R" with the above operations of vector addition, and dot product is usually called Euclidean nspace.
scalar multiplication
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? + {U2V2)^+
+(Un Vn)'''
VECTORS IN
The norm
(or:
K"
AND
is
[CHAP.
m,
root ot U'u:
\\u\\
y/u'U
= yul + ul+
+ul
Observe that
By Theorem
1.2,
wu^O
Let
\\u
v\\
Example
1.6:
(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
(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~
id
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
u/\\u\\
is
We now
Theorem
1.3
state a
(CauchySchwarz):
inequality,
R",
\uv\^
\\u\\ \\v\\.
vectors u,v
GW
we can now
cos
6
^ U
Note that
if
uv =
0,
then
90
(or:
ir/2).
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
=
+
(c,
d)
iff
a
and
{a,b)
{c,d)
(a, b)(c, d)
CHAP.
1]
VECTORS IN
number a with
AND
C"
We
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,
possible.
denoted by
= a =
(a, 6)
(0, 1)(0, 1)
(1, 0)
= 1
(0,6)
or
= \/=l
(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
0,
bi
(Notice that zz
z are given
a^
+ bK) =
If,
in addition,
#
+
by
z_
a
a^
Z1
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
wz = w+
52i.
(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
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
z2 + 3i
1^1
= V4 +
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 ntuples of complex numbers, denoted by C", is called complex nspace. 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, 4i, 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
22
Observe that
wu and so
1.10:

when u =
0.
Example
Let
m =
+ 3i,
= {S 2i, 5, 4 61). Then uv = (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
4i,
2i)
and
u'u =
(2i)(2t)
(2i)(2i)
m
 Vu'u =
\/34
Remark:
were defined by uv = ziWi + + ZnWn, then U'U0 even though u0, 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"
(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 + 1,52) =
3,
3).
(ii) (iii)
numbers of components.
3(4, 5, 6)
(iv)
(6, 7, 8)
(6,
7,
1.2.
Let
(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)
3u4v = 3(2, 7, 1)  4(3, 0, 4) = (6, 21, 3) + (12, 0, 16) = 2u + 3v5w = 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
Thus x
and
j/
1.
1.4.
Find x and y
if
(4, y)
x(2, 3).
(4,
y)
x{2, 3)
(2x, Zx).
2x,
3a;.
and y
6.
1.5.
Find
x,
y and
z if
(2,
3, 4)
x{l, 1, 1)
(2,3,4)
= = = =
a;(l, 1, 1)
{X, X, x)
(x^y
+ z,x\y,x)
X
\
Now
To
or
set the
j/
2,
3,
a;
first
1.6.
Prove Theorem
(i)
1.1:
GW
(v)
(vi)
(vii)
fc,
fc'SR,
(ii)
(iii)
(iv)
+ w)
(viii)
Let
Wj, Vj
and Wj be the
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))
for
i\,...,n
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)
m)
( Mi,
.
.
. ,
(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
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
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 +
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,
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),
1(mi, M2,
M)
(mi, M2,
m)
u.
1.7.
Show that Ou =
a vector.
Method Method
1:
first
is
Om
0(mi, M2,
m)
(Omi, OM2,
Om)
(0, 0,
...,0)
2:
By Theorem
Adding
1.1,
Om
(0
+ 0)m =
Om
Om
Om
DOT PRODUCT
1.8.
Compute u v where:
(i)
u=
(4,
(iii)
(i)
tt
(3,5,2,l),
v
(8, 2,
3);
(ii)
u=
(3)
(1,
8,
0, 5),
=
8.
(3, 6, 4);
wv
+
1
(3)
=
+
(ii)
(iii)
is
numbers of components.
(5)
uv =
(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)
equal to
k.
0,
(ii)
fc
(3) 4
12
5k 
10
==
0,
fc
= 2
5'2fc
= l
1.10.
Prove Theorem
(i)
1.2:
kGK,
iff
(ii)
vw
(iii)
wv
M'M
i;
tt
(iv)
0,
and uu
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)
non
J. ,.2
=
n
0.
Furthermore,
u'U =
iff
Mj
for each
i,
that
is, iff
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=
=
.
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
\/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.
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.
m
of the vector
if
(i)
u= +
m2
(2,
4.
7),
.
. .
(ii)
u=
(3,
12, 4).
m1
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
ImI2
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
yjiiu,
1.16.
Prove Theorem
1.3
(CauchySchwarz):
{u\,
m)
and v
(vi,
.,Vn)
in B",
n
\uv\
]\u\\ \\v\\
We
\u'v\
i.e.
Mt'"tl
Im
HvH.
If M = or V = 0, then the inequality reduces to need only consider the case in which m # and v Furthermore,
j^
0,
\U'V\
IMjI)!
+
G
MV
\UiVi\
2 kiVil
inequality.
Now
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
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 =
2v,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
1.17.
{vi,
.,Vn)
in R",
m
+ vH
=^
tt
+
+
v.
+ vjI =
JMj
0,
m
i;1
#0.
Now
+ V(
i)j
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
preceding problem),
2M+fjKI ^
Thus
Dividing by
M
Ik+^IIIMI
Im
and
2k +
i;l
'yilkl
tt
^
(IMI
Ik
+ HIIHI
lbll)
+ f2 ^
we
i;
IHI
m
+ H
i;
m v,
CHAP.
1]
VECTORS IN
in
AND
11
1.18.
R"
satisfies the
following laws:
m^0;
and H
A;,
=
\\ku\\
\\u\\
iff
u=
/cl
0.
v,
\\u
v\\
^
=
= +
m.
t;.
[Ni]
Problem
1.15,
and
[Ng] in
Problem
1.17.
[Ni] holds.
Suppose u
(ui, ii2,
.,
u)
and so ku
(kui)^
ku^. Then
khi\
fcMl2
(fcMi)2
(fcM)2
khi\
khil
The square
COMPLEX NUMBERS
1.19.
Simplify:
(vi)
(i)
(5
+ 3i)(27i);
(ii)
(43*^;
("i)
g"^''
(i^)
^;
(v)
',
i*.
,31. *"
(l
+ 2i)''; (vii)(2^)'
31
(i)
(ii)
29i
r"\
^*"^
^'^'
(v)
(vi)
(3 + 4t) _ 3 + 4i ^ A + Aj  4i)(3 + 4t) 25 25 25  7t)(5  3t) _ 11  41i _ _11_41. 27i _ (2 34 34 34 5 + 3t ^ (5 + 3i)(53t) 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)
\^23iJ
1 _ _ ~ 512i ~
(5
12
'
169*
1.20.
Let
(i)
(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\ .  23t _ (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 = abi:
(i)
(iv)
Use a+6i
= V^^TP:
\z\
2
 3t = V4 + 9 =
Vl3;
\w\
[4
+ 5i
=:
1.21.
Prove:
(i)
(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 = (abi)(c di)  zw
a
(iii)
a+bi
bi
= a
(b)i
= a+bi =
12
VECTORS IN
Prove:
AND
C"
[CHAP.
1.22.
\zw\
=
R.
\z\
\w\.
Suppose
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) =
22
^2
1.23.
Prove:
\z
w\
R.
\z\
\w\.
Suppose
V
(c,
d)
in R2.
w = c + di
62
where
a, 6, c,
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
+ 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+
Find:
(3
+ i)v.
(i)
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
(14
8i,
 Ii,
15
+ 5i)
1.25.
vu
v
where:
(5
+ 6i),
(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 =
lOi
23i
13i
vu
(ii)
= =
 6i)(3 + 1) (5  6z)(3  i) =
(5
lOi
23i
13i
uv =
=
ij
 2i)(5 + i) +  2i)(5  i) +
(4i)(2 (4i)(2
 3i) + + 3i) +
(1 (1
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
mP
(i)
where
64, 60,
=
m
(z^, Z2
m1P
m2
(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
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
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)
uv;
and
i;;
(vi)
d{u,v).
(i)
u + v
4m 2m
(3
+ 7,2 + 1,13,4 + 6) = =
(10,1,2,10)
(ii)
(12,8,4,16)
9,
(iii)
(6,
4,
2, 8)
(21, 3,
18)
(iv)
uv =
1m1
2123 +
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).
uv;
(8
Find:
mI
u + v;
u + v
(ii)
2m;
(7
{Si)v;
i,
(iv)
(v)
and
v.
(i)
= =
 2i + 1 +
+ 5i  3  6z) =
(4
i,
1
 i)
(ii)
2m =
(3i)v
(14i
 4i2, 4t + 10t2) =
(3
(iii)
(iv)
uv =
(7 (7
=
(v)
 2t)(TTt) +  2t)(l  1) +
+
(2)2
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
.,bn a)
where:
with
PQ
P=
t)
(2,5),
Q=
(3.4)
(ii)
P=(l,2,4), Q = (6,0,3)
V
(i)
(ii)
= =
QP QP
2,
7)
1.31.
The
Xi,
.
set
.
H of elements
form
.,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
GH
Q =
P=
(!,
.,aj
C2a2
and
.,6).
Then the
c^bi
Oj
and the
64
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
QP^
{b,~ai,b2a^, ...,ba)
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)
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
P=
(a,)
in the direction of
t
u=
(Ui) =
is,
points points
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
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)
by a single
First compute
u =
(a)
PQ = Q
P.
(*).
u =
(x
QP
= = 1+ = 2 +
we
QP
X V
z
= (4,7,5)
= 5 At = 4 It = 3 + 5t
Q.)
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)
uv;
(v)
m1
(vi)
d(u,v).
Let
(iii)
w=
(2,
u +
3v;
(ii)
5u
3v
Aw;
1.36.
Let
M = (2,1,3,0,4),
llull;
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)
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"
(1,0,0, ...,0,0),
62
.
=
.
(0,1,0
0,0),
...,
(0,
0,0,
.,0, 1)
Show
(aj, ag,
(1262
.,),
u =
aiCi
ae,
(ii)
ej
Oj
for
1,
..,n
1.43.
Suppose M
e K" has
1)
for every
R".
Show
that
it
0.
1.44.
Using d(u,v)
= \\uv\\ 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)
(35i)2;
(iii)
j^^;
(iv)
(iv)3;
(v)
(li)^.
1.46.
Simplify:
(i)
^;
(ii)
f^;
(iii)
i^s,
^2^
^^*;
.'
(^3^
zw;
(iii)
1.47.
Let z Let z
= 25i =
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
1m1
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)
1m
(iv)
(4
 i,
11
+ 2i,
 Si).
Find:
(i)
v;
(ii)
(3
i)v;
(iii)
1.53.
C":
{u
+ v)w =
WW
+ vw;
(ii)
w
and
(u
v)
wu + wv.
iff
1.2.)
1.54.
satisfies the
following laws:
\\u\\
u,
\\u\\
^
v,
0;
=0
H
0.
z,
1zm
i'l.
\z\
\\u\\.
[Ns]:
\\u
v\\
MISCELLANEOUS PROBLEMS
1.55.
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 =
+
iff
CHAP.
1]
VECTORS IN
AND
17
1.57.
line which:
passes through
(7,
1,
5)
(ii)
(iii)
Sx
2y
18
1.58.
Let P,
Q and R
Oi
Mjt,
a;2
02
ti,
+
^2
Mgt,
x^
t.
a^\ ut
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)
2m
 5i; =
(iv)
y
9;
= ^62
(3,
1.35.
(i)
(ii)
5u3v4w =
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);
3u2v =
(4,
9,
(iii)
uv =
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
1.45.
5055i;
^i;
z
(ii)
(ii)
16
30i;
(iii)
(4
+ 7t)/65;
1;
(iv)
(iv)
(H3i)/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)
uv m1
21
+
,
27i,
i;
vu
=
27i
(v)
= 3VlO
3a/6
1.52.
(i)
uv =
(3
(1
(13
 ei, + i,
11,
9 + 4t)
(iii)
uv m
12
8,
2i,
vu
v/215
12
2i
(ii)
+ t)i; =
31
+ 17i, 27i)
(ii)
(iv)
JHI
=
7i/
1.55.
(i)
Sx
+ yllz =
12;
13x
4j/
7;
(iii)
4z
46.
1.56.
k0
(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
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 =
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
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 thewtuple , ,, ,
is
true.
is
U =
(fcl,
fe,
kn)
Example
2.1
+ 2y 4z + w =
is
S.
The
4tuple
u3
(3, 2, 1, 0)
+ 224'l +
However, the 4tuple
=
v
5
or
(1, 2, 4, 5)
=
is
is
=
3
+ 2.244 +
or
6 =
Case
(i):
One
equation as follows
axxi
= 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.
4y
8.
We
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 3tuple 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
solution.
All the coefficients in (1) are zero, the equation is of the form
is
also zero.
That
is,
Oxi
Oa;2
of scalars in
is
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 ?ituple 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
I
nnfa2nXn
= ^ 
n
(**)
amlXi
is called
+ 0^2X2 +
.
.
OmnXn
tion,
the homogeneous system associated with = (0, 0, namely the zero wtuple 0)
.
(*).
solu
Any
other solution,
if it exists, is called
The fundamental
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)
(*).
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.
first
unknown
an
0.
xi
has a nonzero
coeffi
so that
Step
2.
1,
anLi
ttiiLi
That
is,
replace the ith linear equation Li by the equation obtained by mulby an, multiplying the ith equation L, by
We
the
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:
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
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
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
Oa;
= =
5,
b =0
(ii)
if
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 <
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)
applies.
Theorem
2.2:
(***) 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.
Inconsistent
Consistent
No
solution
Unique
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
*
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,
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 3tuple (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 4tuple (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
wh
a;
or, in
bers.
other words, Frequently, the general solution and w (instead of a and 6) as follows:
is left
in
2y +
w,
+ 2w
or
(4
2y + w,y,l + 2w, w)
We
Example
6jj/
= =
Cj
622/
C2
According to our theory, exactly one of the following three cases must occur:
(i)
The system
is
inconsistent.
(ii)
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
(ii)
The two
The two
(iii)
we
aiiXi
ai2X2
aisXs
+
a2.J2+lXj2+l
a2i2Xj2
+
+
+ amXn = + a2nXn = +
drnXn
drj^Xj^
({r.i^+lXj^+l
possibilities:
r r
= w.
<n.
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:
2y
has a nonzero solution since there are four unknowns but only three equations.
Example
2.9:
We
z
z
2x3y + x4v +
2z
= 
5y +
5y +
solution, since
3z 3z
= = =
5j/
3z
= =
we
For example,
5) is
Example
2.10:
We
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).
5w =
3
1
.
^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
4a
36
01
4a
36
2x
3(4a
 36  1) +
56
6a
2(2
+ 36) 36
56
=
z
3a
56
3a
2,
j/
4a
1,
36,
w =
real numbers. Some texts or (3a562, 4a36l, 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
X
2.2.
+ 2y3z = 1 y+ + 3y 2z 4z
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/
2y
3z
= 1
7y + 7y
llz llz
= =
is
10
7
inconsistent, for if
we
subtract
we
obtain
Oy
Oz
or
3.
2x+ y2z =
2.3.
10
1
.
3x
5a;
+ 2y + +
42/
2z 32
=
2L3:
Reduce
L2
*
to echelon form.
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
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 3tuple (1, 2, 3), is the unique solution of the system.
solution.
26
LINEAR EQUATIONS
x
[CHAP. 2
+ 2y3z =
y
2.4.
2x
4x
iz
2
14
+ 3y2z =
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 
2z
is
equivalent to
2y
Sz =
X
2z
2y
3z
=
=
2
2
or simply
2z
y2z =
we can
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.
= 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
The value,
a;
1,
3/
4,
(1,4, 1).
2.5.
=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
2w
32/
4z
Bz 3z
= =
w+
2y
2/
+ 
2 4
Now if a 4tuple (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
yA
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
(al)y
4z
is
y
J/
+
(a
(al)y+
+ 2)z = 4z =
L3
1 1
^
Now
(a V^L^
a
L^,
(a
 1)1,2:
La:
(a 
+ (al)y +
1)2/
(2
 a  a2)z = 14z = 1
or
to obtain the equivalent
a a
system
V
y
J/
+
(a
1 1
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
a,
and
c so that the
+ 2y Sz 2x + eyllz 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
*
2L2
L3,
we
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.
solution:
,
+ Sy +
5z
(i)
+ 2w =
2x
+ 5y + 2z = x + Ay + 7z = x+3y + Sz =
("i)
(i)
(ii)
x
to
+ 2ySz = 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)
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.
in,
say,
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;
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
(0, 0, 0)
CHAP.
2]
LINEAR EQUATIONS
Set corresponding components equal to each other and reduce the system to echelon form:
29
+ 8z = xZy 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)
= =
(0, 0, 0)
(0, 0, 0) (0, 0, 0)
(x
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
. .
+
.
ttji^ji
.
Oajj
+
is
aj + i^j +
oa;
is
Then for example (0, .,0, 1, 0, .,0), where 1 equation and hence of the system.
MISCELLANEOUS PROBLEMS
2.11.
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}
(*).
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
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
u +
U
i.e.
Now
t
suppose
(vi,
vj
is
solution of
(*).
Then, for
1,
.,w,
ttji^i
ai2U2
aiv
bj
Observe that v
= u+(v u). We claim that vuGW. 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
umST^.
Then
vGu+W,
and hence
U Q
Both inclusion relations give us nonhomogeneous system (**).
u +
W
is,
U  u + W;
that
u+W
is
2.12.
Consider the system (*) of linear equations (page by Ci, and adding, we obtain the equation
(CiOn
18).
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
Cmd^Ofcl
("ifflln
emO'mn)K
!&!
"
"
"
C^fe^
But
this
0,lnK)
Ci6i
+
+
Cm(aml
cj)^
+ =
amn'n)
=
+
Ci^l
'
"
"
C^fem
or,
by
(2),
Cjbi
c^b^
which
is
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
!/{ *
(Oii^^i
iO
to (#),
we
(*)
set.
CHAP.
2]
LINEAR EQUATIONS
2.2:
aiia;i
31
2.14.
Prove Theorem
ai^Xi
+ +
ainXn ainXn
= = =
bi
62
O'ri^^ir
<*r,]V+ia;j^+l
+
.
arnaJn
ari,
for
where
(i)
1<h<
= n.
0, a2J2
^0,
^ 0,
The
solution is as
follows.
(ii)
Then the system has a unique solution. Then we can arbitrarily assign values
number r of equations
to the
nr
If
in the system.
1,
then
we have
a^Xi
a^x^
aa;
6,
where
Oj #
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)
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
Now
assume r
>
1 =
equations.
*2
r 1 equations
'*2J2*J2
We
view the
'*2,J2+1*J2 +
"^
1
is in
echelon form.
By
induction
we can
(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
a;2
^2,
a^j,!
'fjai)'
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 !,..., ji 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
(*) is
consistent
and only
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
(*) is
must
(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
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
+ 2y3z + 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
2zk
z
2x
X
+ ky + 2y +
3z
z
kz
= 3 = 2  1
2.22.
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.
solution:
(i)
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.
X2x+
(i)
2y
+ 
2z 2z
6z
2x 9x
(ii)
+ + 
4y 3y 2y
5y
+ + 
7z
= =
=
2z 3z
+ ^
+
4v
7v
 5w 
+ w =
+ 3w 2w Q
5x 6x
v
3v
12z
+ 4z
2.25.
u,
v and
w
(1,
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,
^^
^
(ii)
_ af ee ~ adbc'
7 =
J
'^
ad
^f he
i*
(iii)
ii
= 2 =
f>
2.27.
ax
ex
+ +
by
= =
dy
Show
that if
adbe'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
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)
constants
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
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)
(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
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)
\am2l
are its columns. Note that the element ay, called the ijentry or ijcomponent, 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:
.
..
^
(
"
\xy
zwj
'")=(,
VI
x
4/
y
.)
is
X
2z
z
= = y +w = w = + =
1,
3
I
The
2,
3,
w=
1.
Remark:
row
is also
referred to as a
row
vector,
field
In particular,
an element in the
MATRIX ADDITION AND SCALAR MULTIPLICATION Let A and B be two matrices with the same size, i.e. the
columns, say,
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
an + &n
A + B =
a21
+ +
&2I
+ + +
&12 622
...
am +
a2n
bin
?>2n
...
+ +
ami
bml
Omi
&m2
Omn
fc
or simply kA,
is
feain
.
Ckaii fca2i
kaml
ka22
kazn
fcOm2
kOmn
I
Observe that
also define
AB ^ A+
{B)
The sum
not defined.
CHAP.
3]
MATRICES
37
Example
3.3:
Let
A =
(]
J\
and
B =
fj
^).
Then
4
A + B
3A
2
6 9
3
= =
3*1
3 '4
(2)
36
12
"' 3 24
35 ') +
3 '(6)
15
18
2ASB
Example
3.4:
c
r '8
'
10
12/
r
V2I
... ...
...
7
29
4
7
36
The
mXn matrix
all zero,
,0
is
0,
that, for
any
mXn
and matrix
will be denoted
by
0.
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
ki, kz
K,
(v)
(vi)
(vii)
(viii)
(ii) (iii)
(iv)
{B + C)
Using
(vi)
and
(viii)
above,
have that
A + A = 2A,A + A + A = ZA,
(or
. .
Remark:
Suppose vectors
say,
in
by column
.
vectors);
. ,
ttn)
and
(bi, 62,
b)
scalar product
ku are as
follows:
u+
But
{ai
+ bi,a2 + b2,...,an + b)
and
ku
(fcai,
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.
we
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:
AB
Accordingly, above.
lbl\
(tti,
02,
.,a)
aibi
a2b2
ttnbn
Wl
we
define the
by a column vector
as
38
MATRICES
bnXi
[CHAP.
+
+
biiXi b22X2
feisics
= =
y\
(1)
1/2
(ii)
b23X3
This system
is
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 \B2X
where Bi and B2 are the rows of B. vector yields another column vector.
(iii)
Now
auVi
consider the equations
a2iyi
+ +
ai22/2 (i22y2
zi
(2)
Z2
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^.
(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
into
ABX = Z
This will represent the system
ftii
(3) if
we
and
as follows:
011613 021613
021
022/1621
622
623/
011612
021612
+ 012622 + 022622
+ 012623 + O22623
AiB'
A2jB'
ArB^ AiB^
AaB^ 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.
CHAP.
3]
MATRICES
introduction,
39
we now formally
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 yentry is
AB
mxn
. . .
row A,
of
Aifii
AB =
That
A2S1
AiB2 A2B2
. .
Ai5" A2B"
,ABi
is,
AmB^
Am'B"!
jail
/Cii
...
Cm
Cii
dml
Opn
P
\Cml
...
Ci
where
cy
aiiftij
ai2&23
+
is
+ avpbp. =
if
fc
2 1 =
Cifc&fci
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
11
6 8
2
4
4
6
not commutative,
i.e.
the products
AB and BA
Theorem
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
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^
Theorem
3.3:
(i)
(A+B)* = A* + B*
(A')'
(ii)
(iii)
(iv)
for k a scalar
anXi
a2iXi
+ ai2X2 + + a22X2 +
+
am2X2
aina;n
&i
+ +
annXn
=62
n\
OmlXi
is
OmnXn
/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
is called
(1),
O12
tt22
ttin (^2n
^ttml
(lm2
Otnn
is called
by
its
(1).
(1) is
completely determined
Example
The
coefficient
3j/
4z
5z
= /2 \l
7 3
/2
(1
Observe that the system
_2
is
4\ 5;
34
2
5
7 3
^*^
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
A{kui
kui
0, Au2 = 0, Aun = 0. Hence + knAun + fettn) = kiAui + kiAu^ + = fciO + ^20 + = fc0 +
Accordingly, kiUi
+ kiia
is
AX = 0.
Theorem
3.5:
Suppose the
field C).
field
K is
infinite (e.g. if
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{uv)) =
tions
Au + k{AuAv) = B + k(BB) = B solution of AX = B. Since each k e K, u + k(uv) is (Problem 3,31), AX = B has an infinite number of
a.
all
such soluas
solutions
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
"'^'r'
where
j
^i
< ^2 <
< jr
for
i^r,
< ji,
and for
i>r
We call
ttijj,
, ttrj,.
Example
3.9:
The following are echelon matrices where the distinguished elements have been
circled:
/(i)
6\
2
13
0/
is called
if
the dis
(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
equivalent to a matrix
if
by a
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
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.
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,
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:
shall
mean
to
Example
is
row reduced
i?3
^ 3fii + R3,
1
form by applying the operations and then the operation R3  SKj + 4^3:
2
A=2 422to0042to
4 2 3 5
3
3
4 2 2
Now
aijj,
. .
.
suppose
Orj^.
A=
is
Rk
for
i
^
ak^Ri
OiiRk,
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
/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
as columns
is called
said to be of order n,
and
The diagonal
an,
a22,
. .
(or:
main
A=
2 5 8
. ,
ftjin.
3^
Example
3.12:
The following
is
a 3square 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
/an
ai2
ain\
ain
/ail
ai2 a22
am
a2n
O22
or
\0
...
ann/
diagonal are
all zero.
A diagonal matrix
is
/a,
a2
'ai
...
or
a/
"^
'
\o
...
an
In particular, the nsquare 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,
AI = lA = A
The matrix
a scalar k diagonal entries are each k.
kl, for
G K,
is called
a scalar matrix;
it is
In particular,
if
is
= AA, A^ =
in the
A^A,
..
and A"
=/
f{x)
ao
+ ai* +
UiX^
ttnX"
44
MATRICES
aj
[CHAP.
where the
are scalars,
we
/(A)
aol
aiA
a2A^
+ aA"
is
is called
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
matrix
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
_ ~
_ "
/65
1^33
/
10 + 10 5 + 6
1
1
1
,'2
5\/2 2j\l
and
/

65
2
1515
5 +
6
s)
V2 +
5\
3
^
Thus
5\
2
3/"""\^i
(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:
(21
3)(4
l=22 + 4 + 18
444 +
invertible
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
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
72
5
9/
=
\s
3
1
7 2
5
1
5]7
1
2
9
4
is
9/
\3
The convenience
can be obtained by carrying out the computation with the blocks, just as actual elements of the matrices. This is illustrated below.
Suppose
is
Ain
Ain
k, multiplies
each element of
.
by
k; thus
(kAii.
feAai
kAi2
kAm
kA2n
iCAmn j
&A22
rCAjn2
iCAjnl
Now suppose
a matrix
is
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
Ai2 A22
+ Bi2 + B22
...
. .
Am + Em
Aln
Bin
\Am\
Am2 + Bm2
Amn + Bm
true.
U and V are
^
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
Wn
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
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
25
5 +
3
(0 +
(ii)
12
4 l^ 1 3y
4333
2
5 1 4
The sum
is
(iii)
0/1
^
'
2

3N
'
'
_ 
/
'
3
12
6
1518
CHAP.
3]
MATRICES
47
3^.
Let
A = 72
.
5
(3
04)'^ = (01
/6 15
1\
/I
2 3\
5J'^
= (llljF^"<i 3A + 4B2C.
/
/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
3y\
\3z
BwJ
_ ~
\z +
wl
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
Then
tively
and so
fcay
Then Oy + 6jj is the yentry of A + B, and so &(ajj + 6^) (bij). +B). On the other hand, ka^j and fcfty are the ijentries of kA and kB respecfe6y is the tientry 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 +
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)
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)
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.
row
of
AB, multiply
the first
row
(1,
3)
of
by the columns
_4S
and
(
x3y'V2
of B, respectively:
S\/2
1
j(^ 3
04\
2
6 y
_ ~
/12
V
2
+ 33 10 +
9
3 (2)
(4)
+
6
06
4 + 18\
/ll
14
To obtain the
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
V22 +
(l)3
(1)
(2)
(4)
(1)
6/
Thus
Note that
^^
J5 is
( 1
214
(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 =
/^
53)' *"^
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
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
"
/23 44
10
0\
11
8
10^
( 3
"o )
of
=
AB, multiply
/
the second
row
of
by each column of B,
2
1
1
i
\
1(
125 ^ ^
^
1
1
8 2 +
10 5 +
\
I
/I
11
8 2
10^
5
6
3
/
To obtain the
third
third
row
of
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
22 +
15
1 +
020
15
21
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
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
row of A by the
Hence:
(3)
(2)
c,4
(2,1,0)
C21
(1, 0,
3)
(3) 4
012 =
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,
and
is
a 2
2 matrix:
50
MATRICES
1
^3
(ii)
6Y4 5A2
is
0\
1/
X 2 and
1
(3)0
_ ~
and
/ 16
6'
5^
[2
is
The
first
factor
the second is 2
1,
a 2
matrix:
^V
2\
12
6 (7)
3
(iii)
5A7;
\(3)'2
is
+ 5'(7))
_ ^
/40'
[41^
Now
is is
distinct, the
product
2X2.
(iv)
is
is 1
2,
and
is
a 2X 2
{>''
(v)
=
is
ill
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,
and
is
a 1
matrix
(21
(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,
row and
Ith
{AB)C
SJ2C21
+ A +
+
T,
SiCi
k=l
i.e.
StfcCfcl
=22
fc=l
{"'ifijkiOkl
j=l
by
in the tth
n
row and
fth
column
i2*2!
aim*ml
is
ij*jl
=22
(6jfcCfci)
proven.
3.12.
Prove Theorem
Let
3.2(ii):
A=
(tty),
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
in the ith
row and
fik
feth
AB + AC
m
j=l
is
ik
i=l
ayftjfc
3=1
fcth
ttyCjfc
i)(6jic
+ c^k)
On
row and
AD = A(B + C)
is
m
oadjk
aisd^k
otmdmk
}=l
i=l
a.ij(bjk
+ Cjk)
Thus A{B + C)
AB + AC
CHAP.
3]
MATRICES
51
TRANSPOSE
3.13.
A =
3.14.
Let
be an arbitrary matrix.
Under what
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
is
w m
A*A
is
3.15.
Let
^ =
(
_!
4)
Find
(i)
AA\
(ii)
A*A.
'1
31
To obtain
A
3
as columns:
A*
Then
^^
A*A
/I
'
:)(: :
13 + 2'(l) + 0'4 \ 3'3 + (1) (!) + 44/
_ ~
/5 \1
26
i (3
11
I
:) l'2 + 3'(l)
2
33
l'0
2
(1) '3
(1)
(1)
0l + 4'3
+ 34 \ + (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
AB
is
anbij
ai^hzj
ai^h^j
Thus
(1) is
row and
ith
On
row
(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.
Which
are
3
5
l\
/O
5
0\
24,
7
3/ \0
first
1,
0/
nonzero entries in the rows; hence
\0
[l)23
0l\
3/
/075
\0
0(1)2 4,
An
echelon matrix
is
00,00
0/
0\
/0O5O2^
20
4 7,
\0
row reduced
if its
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
Apply the operations ^2 * 2i?i + Rz and R^ iJj > 7B2 + 3B3 to reduce A to echelon form:
ZRi
'1
^3.
2
1^
4
lOy
iSj ^
Method and Ri
1.
2i?2
+
(i)
7i?2
to the last
matrix in
^3 +
7i?i
to
34
4
I
to
012
7
10
J
Finally, multiply Bj
by
1/21,
R^ by 1/21 and
'1
fig
by 1/7
to obtain the
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
'12
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
CHAP.
3]
MATRICES
/O
1
53 3 2\
3
3.19.
A =
I2
1
2 \2
14
3
2 ll
1^
4
Z\
to
is
3 4\
1
3.20.
Reduce
i.e.
A = 4
\
1
6
to echelon form,
25/
if
to its
first
Hence
first
interchange the
to
is
3.21.
Show
of the
inverse operation
same
type.
jth.
m
(i)
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
this operation
own
inverse.
Multiplying the ith row by k and then by fci, or by fci and then by k, we obtain the original matrix. In other words, the operations iJj  kRi and i?j ^ feiiJj 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)
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
11
4)
12 + 2 (3) \ 42 + (3) (3)/
+ 24
V4l + (3)4
^ ~
4\
17/
[8
54
MATRICES
[CHAP. 3
(iii)
'
c
+
l(4)
4 (4)
217
(3) 17/
/7 \eO
30\
67 J
To
f(x)
2x9
4a;
5:
/(A)
= 2A34A +
5/
2(; Z) 
{\
s) + K'o
l)
its
respective scalar:
/14
1^120
60\
4
8\
12/
/5
0\ 5y
134y
V16
\0
_ "
/ 14 4 + 1^12016 +
608 +
134
_
I'
IS
104
52 \
12
+ 5/
117/
3.23.
is
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)
x^
+ ^x 11:
,.,
= .....
n,
(:.)(! 4)
it:
"GO
,
scalar preceding
g{A)
4X _4x
17;
/2
4\
<
/ll
V
V8
V8
( 6/
11
g{A)
Since g{A)
/O
(^0
0,
is
3.24.
Given
A 
such that
Au 
3m.
First set
Au =
4
3u:
3/U/
vector):
~ ^
^\y
+ 3y\ \Ax3yJ
x
/Sx^
V3j/y
equations (and reduce to Set corresponding elements equal to each other to obtain the system of echelon form):
+ 3J, = AxZy a;
3a;
__
2x
Zy
Ax
3y 6y
= =
2x
Sy
00
^^
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/
is nonzero g j
Au =
3m.
CHAP.
3]
MATRICES
/3
f
55
3.25.
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 =
first
system
=
1
2,
5,
w=
3.
/3
is
5\
)
We
A^
oi the
2X2
6c
matrix
A^i
Ttaslf
rTT
1^1
Vc
91(1
a'
where
lAI
ad
A=Q
ly
,h.
A
= 1 .d A. =
l(4
^) =
(J
4).
where
1\
3
\0
and
2,
Hence
^1 GJ
and
S =
\0
TJ
where E, F, G, R,
S and T
//9
12 15N
/3N
/I
AB
= ~
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"). BjA). Thus the ith row of BA is ftjA.
Let
.,
3.28.
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)
is
not invertible.
ith
(i)
Let
jBj
Then the
..., 0)
row of
AB
is
{RrB\ Ri'B^
RiB^)
(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
/AiC/
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)(BiAi)
and
= A(BBi)Ai =
A/Ai
= AA
/ I
(BiAi)(AB)
Thus (AB)i
= BiAi.
3.31.
Show
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
^ IRs and
/I
722 *
3i?i +
to obtain
R2.
o\
1
=
\o
1/
/I
^1
^2
=
\o
Eo
7
3.33.
Prove: Let e be an elementary row operation and E the corresponding msquare elementary matrix, i.e. Ee(lm). 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
CHAP.
3]
MATRICES
=
.
.
57
Here a
/
means that
e)
is
(cj,
.,
By Problem
3.28,
e^A
iJj.
We
also
remark that
(i)
Let
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,
A
.
.
(fii,
., i?,,
., ffj
fij
e(A)
Now
let e
fc
t^ 0.
Then, for
a =
i,
E =
Thus
(iii)
e(/)
ej
and
e(A)
(fij,
. .
=
/\
,
(ftj,
., fcfij,
.,
BJ
^A
/\
(fijA, ...,A;ejA,
...,e^A)
=
JBj
fefij,
=
/\
e(A)
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)
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
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)
(ii)
(iii)
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
E.'E^EiA
= =
/,
and so
A =
then
(E,
E^Ei)^
= E^^E^^'EJ^
Thus
(ii)
By
implies
(iii).
Now
matrices
if
(iii)
holds
(A
EiE^
must follow
is invertible.
58
MATRICES
Let
[CHAP. 3
3.37.
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
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
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 AK
is
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
I)
it to
1
1
0\
I
(A, /)
[21
1
3
I
1
I
8
1
2
to
1 1
2 6
to
final block
matrix
is in
the form
(/,
B).
Hence
is
invertible
and
is its
Ai
=
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 3.393.41, let
4
(o1
3.39.
03
2 3
!)
B,
(ii)
"il
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,
(By Problem
ejA
(ii)
If
(iii)
If
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
l\
3
,
/2
(ii)
32
2 6
A =
5/
1 \4 5
3
5
Ij
3.45.
Reduce
to echelon
to its
/l
(i)
3
11
1 5
3
1
,..^
25
\4
1
(")
413 ^=0021
^
5/
\0
3
4,
3.46.
Describe
all
the possible
2X2
3.47.
A
3.48.
Suppose A is a square row reduced echelon matrix. Show that has a zero row.
if
A # 7,
Show
is
3.49.
Show
(i)
that
is
row equivalence
to
is
an equivalence
relation:
(ii)
(iii)
A A A
row equivalent
to to
A;
implies
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)
(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).
u =
such that
Bu =
6m.
3.52.
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)
3.56.
i.e.
AB =
B =
3.57.
\0
kj
is
a scalar matrix.
Let
(i)
Dfc
k.
Show
that:
for any
mXn matrix A,
D^A =
kA;
(ii)
for any
of:
nXm matrix B,
BD^ =
kB.
3.58.
Show
(i)
upper triangular;
lower triangular;
(ii)
(iii)
is
diagonal;
(iv)
is scalar.
INVERTIBLE MATRICES
3.59.
(i)
'
'2
)
>
3\
3/
(")
1
3.60.
2
1
3\
,
2
(ii)
V
1
(i)
2 \5
2
,42
1
t.61.
5/
3/
4\ 6 1/
is,
1
5
ll
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
(Ai).
Thus,
(!
.
."!
///.
.M
if
invertible,
and what
is its
inverse?
3.64.
Show
Show
that
A A
is
row equivalent
to
if
and only
such that
B = PA.
3.65.
that
is invertible if
and only
if
the system
AX =
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)
3.12.)
3.67.
(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.
column operations:
[^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
il
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)
068
\3
6/
14
(vi)
\
2
6
13
3
9y
4/
3.42.
(i)
(ii)
(61, 62.
K K)
and
(iii)
'1
3.44.
(i)
(
2
1
4/3^
36 32
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
iO
1/
3.52.
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)
(^
15
2\
3)
(n)
,..,
/
f
1/3 ^/9
1/3
2/g
3\
6
11
8
5
1 3^
3.60.
(i)
107
12
86
5/
10
1 4y
9/2
5/2
4
3
51
\7
3.62.
Given
AAi =
/.
Then
7'
(AAi)'
= {A'^YAK
/ai
That
is,
(A^^)'
(A*)"!.
3.63.
is
invertible
iff
each
aj 9^ 0.
Then
.0
'
ai
...
\0
chapter 4
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".
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
is lost if
is
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 :
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]:
{u
+ v)
+w
and
= u+
{vi
w).
which u
u
0.
[A3]
[A4]:
uGV
G
there
is
+ {u) =
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
k{u
kv. bu.
63
64
[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.
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)
field
K.
7,
fcO
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.
The
first
example
is
of be an arbitrary field. The set of all ntuples 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
we
vector in K is the wtuple 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
eV
+ gGV
defined by
{f
+ g){x) =
f(x)
g(x)
kEK
and a function / e
is
the function
kfeV
(kf){x)
kf(x)
CHAP.
4]
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.
is
is
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
Example
Let
V be
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
third component
(ii)
zero,
all
{{a,b,0)
a,b
a subspace of V.
4.2).
Let
set
be the space of
square
nX n
Then the
Ojj,
(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
is
bounded
there exists
M GR
X V
Example
4.8:
Consider any homogeneous system of linear equations in n unknowns with, say, real
coefficients:
aiiXi a2iXi
ay^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
[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
Theorem
4.4:
The
intersection of
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
is called
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
0)
c(0, 0, 1)
Example
4.12:
The polynomials
(in*):
y=
t.
L(l,
t^
.).
powers of
CHAP.
4]
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,
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
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.
...
. .
ai
a,2n
\fflml
flm2
dmn/
dmn)
The rows
of A,
Rl viewed as vectors in
(ttll, 0,21,
.,
am),
Rm =
(Oml, am2,
.K",
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)
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.
Theorem
Row
same row
space.
We shall prove (Problem 4.31), in particular, the following fundamental result concerning row reduced echelon matrices.
68
[CHAP. 4
Theorem
reduced echelon matrices have the same row space have the same nonzero rows.
Row
if
and only
if
they
is
row equivalent
to a unique
called its
We
Example
Show
(1, 2,
1,
M2
(2, 4, 1,
2),
and
wg
(3, 6, 3,
7)
(1, 2,
4, 11)
and
v^
(2, 4,
5, 14)
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
Mj,
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
t>2,
to
row canonical
1/3
1
form:
_ = ~
/I ,x
(
\2
24 45
4 2 24 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.
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
k{u
+ w) = ku + kw G U +
W
F
is also
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]
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
is
sum
of its subspaces
and W,
V = V
GF
w
way
as v
every vector v
where
u&V
= u+w
and
w gW.
applies.
Theorem
4.9:
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:
U
:
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
(3, 1, 0)
(0, 4, 7)
and also
(3, 5, 7)
(3,
4,
0)
(0, 9, 7)
Example
4,17:
In R3, let
U
Now any
vector in
=:
{(o, 6, 0):
(a, b, c)
{(0, 0, c)
G R}
in
vector
in one
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
field
K.
kGK
and
GV,
fcO
0.
(ii)
GK
0,
(iii)
ku
where
kGK
(iv)
For any
By axiom
fee
kGK
[A^]
and any
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
70
[CHAP.
Suppose
fcw
and k
= 0.
Then there
fc~ifc
1;
hence
u = lu = {k^k)u = kHku) =
(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
ku
(k)u.
to
both
4.2.
Show
{k(v)
u and
v,
k{uv) =
ku
kv.
4.1(iv)
Using the
of subtraction
{uv = u+ (v))
k(v)
kv),
k(u
v) =
k(u
+ (v)) = ku +
ku
(kv)
= ku 
kv
4.3.
b)u
au
bu,
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
different operation.
4.4.
(ab)u
a{bu),
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
X is nonempty, V is
+
also
nonempty.
We now
f.g.hGV. To show
(f
g)
that each
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
g(x))
h(x)
f(x)
(g(x)
h(x))
Accordingly,
(f
+ g) +
+(g + h).
0(a;)
[AJ:
Let
= =
0, Va;
G X. Then
V,
+ 0)(a;)
is
f(x)
0(a!)
f(x)
f(x),
Thus /
/,
and
CHAP.
4]
71
let
/ be
(/)()
Oix),
(/))()
f(x)
(/)(*)
f(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)
K where
f{x)
follows
and
addition is commutative.)
Let
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.
leK,
(!/)()
l/(a;)
f{x),
V G
X.
Hence 1/
/.
Since
is
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)
(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
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
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,
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
hold.
72
[CHAP. 4
SUBSPACES
4.7.
a.
(iii)
if (i)
GW
GW
is
nonempty,
(ii)
(ii)
v,w
eW
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
third component
(ii)
is 0;
{{a,b,c): a
+b+c =
sum
0},
i.e.
of its components
(0,
0,0)
d, 0) in
1;
(a, 6, 0),
w =
kv
k'w
= =
k(a, b, 0)
(ka, kb, 0)
0)
(ka
+ k'c,
kb
+ k'd,
0)
Thus kv
(ii)
tt
+ k'w e W,
and so
is
a subspace of V.
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)
is
first
component
is
(ii)
{(a, b, c): 1;
c)
:
d'
+ b^ + c^^ e Q},
1},
i.e.
not exceed
(iii)
W = {(a, 6,
,,
a, b, c
i.e.
rational numbers.
In each case, show that one of the properties
(i)
Theorem
(1,2,3)
GW
and
fc
= 5 e
Hence
R.
is
But
W since 5
is negative.
(5, 10,15)
CHAP.
4]
73
(ii)
eW GW
and
1^
since
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
all
square
nxn
matrices over a
field
K.
Show
(otj)
that
consists of the
symmetric matrices,
i.e.
all
matrices
A=
for which
(ii)
is,
(i)
OSW
belong to
W,
i.e.
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
+ 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.
all
field
R.
Show
that
(ii)
W consists of W consists of
(Recall that
all all
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 =
W since
det (A
+ B) =
W
(
is
not a subspace of V.
(ii)
= <ll)
, )
belongs to
W since
But 2/
i'l
X
W
n
since
/1 (' Vo
")
1
/2
(
0\
I
4
^
2/
Hence
W is not a
subspace of V.
4.13.
Let
is
(i)
all
field
into R.
Show
that
(ii)
w = {f: W = {f:
/(3) /(7)
map
3 into 0;
same
value to 7 and
(iii)
=  /()
74
[CHAP.
0(a;)
0,
for every
i.e.
&
R.
OeW
0.
Suppose
(af
f.gGW,
=
(1/(3)
/(3)
=
=
aO
and
sr(3)
0.
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
(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).
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)
all
f{x)
^ 0,
yfxGR.
(i)
Suppose
f.geW, I.e. /(7) = 2 + /(l) and flr(7) = 2 + flr(l). Then = /(7) + flC?) = 2 + /(I) + 2 + flr(l) (f + g)i^) = 4 + /(I) + ir(l) = 4 + (/ + flr)(l) ^
Tl' is
(/
+ sf)(l)
/()
Is
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 <
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
(i)
(ii)
(iii)
all
all all
3;
+
&it^
h^t^
bnt^",
i.e.
polynomials with
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
W belong to
is
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
solution set
0)
is
. ,
PF
since, clearly,
a0
ajjO
ttinO
0,
for
1,
.
.
.,m
CHAP.
4]
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
= =
au
1,
. . .
m,
bv
(ciMx
6^1, au2
6^2.
>
<*w
6v)
aji(aMi
aini'n)
Hence au
+ 6v
is
i.e.
belongs to W.
Accordingly,
is
a subspace of K".
LINEAR COMBINATIONS
4.17.
= =
(1, (2,
2,
5)
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,
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
x,
y and
z:
xe^
+ j/eg + zeg.
5,
3)
= =
x{\,
3,
2)
{x
+ 2y + z,
x
4, 1)
Form
+ 3x x
2y+z=2
4y y
2y+z2
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.
4.19.
u=
=
+
(3, 0,
2)
and w =
a(3, 0,
(2,
yw:
fe)
(1,
2,
=
3x
2)
j/(2,
1, 5)
(3a;
+ 2y,
y, 2x
 5y)
Form
By
+ 2y = 1, = x = 1,
j/
y =
2.
2,
2x  5y = k
k
8.
76
[CHAP. 4
4^0.
t^
= t^2t +
t2
5,
62
S.
x,
y and
z:
xe^
ye^
263.
4t
= = 
a;(t22t
a;t2
+ 5) + 3/(223t) + 2(f + 3) + 5a; + 22/t2  s^/t + zt + 3z {x + 2y)fi + {2x3y + z)t + (5a; + 3z) 2xt
t
+ 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
3,
2/
2, z
4.
Thus v
4.21.
I
j
A =
;j).a?)(:x
Set
as a linear combination of
3
1
A,B,C
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,
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
Show
u=
(1, 2, 3),
(0, 1, 2)
(a, 6, c)
and
is
w=
(0, 0, 1)
generate
W.
We
Set
S R3
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]
77
= + +
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)
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,xy + Sz,
2x
 4z)
Form
2x
X
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
W = {(a,
(ii)
b, 0)} in
(2,
(1, 2, 0)
and v
=
=
(0, 1, 0);
u=
R^
is
u and v where:
(i)
u=
v.
(a,
b,0)eW
is
Set
(a, b, 0)
xu
yv:
{a, b, 0)
=
X
x(l, 2, 0)
y(0, 1, 0)
= +
2x
x
(x,
2x
+ y,
0)
2x
The system
(ii)
is consistent; in
fact
= a = b or = = a, y = b2a
is
a solution.
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
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
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
[CHAP. 4
4.27.
Then
containing S.
Furthermore,
if
is
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
Accordingly, L(S)
a subspace
Now
suppose
all
is
a subspace of
V
Oj
containing
Om'^'m
^ ^>
"where
K,
.,v^E.
Consequently, L(S)
c T^ as
Determine whether the following matrices have the same row space:
Row
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
Omn)
+ fc^Oml)
+ kmO'mn)
we
CHAP.
4]
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.
bi
4i
:]c
^ ^
ifc
4i
9i
Osfc
A =
a,v,
Suppose
of
space.
32 =
fci,
of
and
if
and
CiO
1.
We
first
if
ij
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
.,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^
R
+
is
the ith
row
of A.
Then there
Cj,
. ,
R Cj
CiRi
C2R2
c^Rs
if
W
we show
that
The theorem
the
is
proved
R^,
or
but
Cfc
for
A;
t.
i.e.
first
nonzero entry of R.
By
(1)
and Problem
4.29,
(2)
Cl&ljj
262Ji
+ B =
C,b,j.
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
By
(i)
and Problem
4.29,
(S)
a.
"fc
Hb2j^
C,b,j^
80
[CHAP. 4
is
row reduced,
bj^j^ is
(3),
OkHj,^
is
a distinguished entry of
b^j^
and, since
is
0.
Thus
c^b^j^
and, since
1,
c,,
0.
Accordingly
R = R^
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
and
4.33.
Let
jR
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
+ am&m2. + am(&ml.
l&ln
6m2.
+ 02*2n ^
1"
am&mn)
bmn)
fijB is
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
space of B.
of a vector space V.
JJ
Show
that:
(ii)
U and U+W
Let M
W are contained in
\&
[/.
+ W;
linear span of
(i)
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+
UuW
and
so
belongs to L{'U,W).
Thus
L(U, W).
inclusion relations give us the required result.
The two
CHAP.
4]
81
4.35.
Suppose
Let V
and
{Wj) generates
Show
Then v
Mj's;
that {Ui,W}),
i.e.
{Vn}
{Wj),
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.
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,
GW
On
there exist
uG U
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
uu'GU
and w'
UnW
C7
{0},
and
so
TF.
u',
w =
w'
unique and
V =
4.37.
Let
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
=
a)
(0, 0, 0).
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,ba,c a) = U + W,
imply R3
C7
TF.
4.38.
be the Let V be the vector space of tisquare matrices over a field R. Let U and Show that subspaces of symmetric and antisymmetric matrices, respectively.  M*, and antisymmetric 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 wsquare 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,
(^(AAt))'
that
is,
= i(AA')'
Suppose
:
i(AtA) = ^(AA'^)
Then
J(A
A') = M
is
antisymmetric.
We
implies
or
MGUnW.
{0}.
I7nW =
Accordingly,
V= UW.
M = M'
and M^
= M
which
82
[CHAP. 4
Supplementary Problems
VECTOR SPACES
4.39.
Let
y be
plication on
(a^, a^,
.)
in a field
K
+
with addition in
(6i, &2. )
.
. .
(ai
6i,
. .
02+
.
62,
.)
k(ai, 02,
(fcaj,
ka2,
where
4.40.
aj, bj,
G K. Show
that
is
(a, 6)
in
(c,
d)
= {a+
c,b
+ d)
and
k{a, b)
:
=
lu
{ka, 0)
Show
that
u.
Hence
[^4] is not a
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
{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
(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)
U, w,w'
GW
and
is called
sum
of
SUBSPACES
4.46.
V
if:
in
Problem
.)
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.
W
b
is
c;
W
=
(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 wsquare 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]
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;)
R ^ R
is
is
bounded
there exists
Af
GR
such
^ M,
all all
G R.)
(Here /
:
(ii)
(iii)
(iv)
(v)
even functions.
R * R
even
if
/( )
f{x), Va;
R.)
continuous functions.
functions.
in,
all difTerentiable
all
integrable functions
a;
1.
4.51.
is
4.52.
Prove Theorem
of y.
4.4:
The
intersection of
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.
(1,
3,
2)
and v
(2,
1,
1) v.
in R3.
Write Write
(1, 7,
(ii) (iii)
(2,
5,
v.
(iv)
a, b
and
c so that (a, 6, e) is
4.55.
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^
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.
(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
field
as a
4.60.
(1
and
3.
4.61.
in
U =
4.62.
{{a, b, 0)},
and
and
(1, 2, 3)
xj/ plane:
Prove:
L(S)
is
the intersection of
all
the subspaces of
containing S.
84
[CHAP. 4
4.63.
set,
we
do not
set.
4.64.
Sc
T,
then
L(S)
L(T).
4.65.
Show
that
LmS)) =
L(S).
Determine which of the following matrices have the same row space:
/l 1
3\
.<: (3
4.67.
4
.
:,, 5)'
(.raO'
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.
is
the
Show
matrix
4.69.
that if any row of an echelon (row reduced echelon) matrix is still in echelon (row reduced echelon) form.
4.6:
size)
are
row
4.70.
equivalent.
Show
Let
that
and
iff
4.71.
in the
AB
is defined.
Show
AB
is
contained
We
(i)
sum
to arbitrary
nonempty subsets
S and T
of
a vector space
by defining S
+ T =
{s
+t
sG
+
S,
tG
Show
(ii)
(iii)
(iv)
S+
(S2
S3)
4.73.
Show
W of a vector space V, W + W
S
of a vector space (properly contained).
= W.
is
4.74.
which
not a subspace of
S+S =
S,
(ii)
S+SCS
of
4.75.
sum
of subspaces to
follows.
If T^i,
W^,
. ,
T^
Wi +
Show
(i)
W2++W
{wi
+ Wi+''+w^: WiGWi)
that:
L{Wi,
if Si
W2
W)
(ii)
generates Wi,
U S generates W^ + W2 +
+ Wn
4.76.
Suppose U,
aijd
Prove that
Un(V+W)
CHAP.
4]
85
4.77.
Let U,
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).)
4.79.
Show
4.80.
Let PF
= Wj U
S =
TFj
1, 2,
Show
that
Sj U Sa
4.81.
Let V be the vector space of wsquare 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
sum of
U
=
and
W over a
field
K.
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.
^W.
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)
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)
. .
Uj,
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
 f(x))
odd.
4.81.
(i)
V = U + W.
J7nW
is
chapter 5
Basis
INTRODUCTION
Some
(i)
and Dimension
The "dimension"
If
of a vector space
is
(ii)
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.
.,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.
(*) will
always hold
if
all 0.
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,
by
itself,
independent; for *^
kv
,a V =
t implies
A;
i.
The vectors m
for 3m
2v
= (1,1,0), w = 0,
3(1,
1;
(1,3,1)
and
w=
(5, 3,
2)
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
2)
and
so,
+ hy ZxZy + 4a; + y
2x
=0 =0
lz
2z
= =
0,
a;
a;
= 0; = 0,
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:
of a matrix in echelon
form are
linearly independent.
The vectors Vi, .,Vm are linearly dependent combination of the others.
. .
if
and only
if
one of them
is
a linear
For suppose,
aiVi
UiiVii
tti+iVi +
UrnVm
Then by adding Vi
aiVl
to both sides,
we
obtain
OiiVii
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^bjiVji
bi^bmVm
many
im
Vj is
We now make a slightly stronger statement than that above; this result has portant consequences.
Lemma
5.2:
Vi, .,Vm are linearly dependent if and only a linear combination of the preceding vectors:
.
.
if
one of
Vi
kiVi
kiVi
fciiVii
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
Vi,
a.,
n.,
 vz,  n
and the
i;i
is
not
0.
if
3.
Two
and only
if
one of them
is
a multiple of
the other.
4.
is
itself
dependent.
Hence any
5.
Vm}
Vi} is
6.
R^ dependence
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,
V and
are dependent.
BASIS
AND DIMENSION
said to be of finite dimension n or to be ndimensional, e , n, if there exists linearly independent vectors ei, e2, ... e) is then called a basis of V. The sequence {ei, 62,
is
. . . ,
The above
definition of dimension is
Theorem
5.3:
Let
F be
finite
has the
{0} is defined to
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
K" which
.
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
n.
CHAP.
5]
BASIS
AND DIMENSION
all
89
Example
5.4:
Let
K.
/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 lyentry 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
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
is
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
+1
or more vectors in
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.
We
call {vi,
.,
it is
(ii)
{vi,
.,Vm,w}
is
w e S.
applies.
Theorem
5.5:
Suppose S generates
of S.
and
is
Then
(vi,
Vm}
is
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)
can be extended to
a basis,
(iii)
n elements
is
a basis.
Example
5.6:
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:
in R3,
(257,132,58), (43,0,17),
(521,317,94), (328,512,731)
3.
must be
90
BASIS
AND DIMENSION
[CHAP.
Theorem
5.7:
Let
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.
(i)
(ii)
(iii)
(iv)
W W W dim W
has
= =
0, 1, 2, S,
then
then
then T^
then
is
Theorem
5.8:
Let
and
U+
Note that
if
W
is
W be
finite
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
U=
17
{(a, 6,0)},
{(0, 6, c)}.
dim
{U+W) =
or
B.
Also,
dim
1
and
By
3
+ 2dim(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:
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
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.
m linear equations in
ttuXl
CziaJi
n unknowns
a;i,
over a field K:
+ ai2X2 + + a22X2 + +
am2X2
+ ainXn = + ChnXn =
bi
^2
dmlXi
OmnXn
&m
AX = B
where
(an)
is
X=
ai2
(xi)
and
B=
(6i)
consisting of the
constants, respectively.
an
{A,B)
a2i
ttln
.
bi
022
a2n
62
ttml
ftm2
dmn
hm
Remark
1.
The above
dependent or independent.
Remark
2.
are equivalent
i.e.
Remark
3.
We
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
= 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 +
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.
Theorem
5.11:
solution space
AX =
nr
where n matrix A.
the
number
of
unknowns and r
is
AX =
. .
page
all
21),
say, Xi^.xi^,
and
echelon form, then it has precisely nr 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
. .
Find the dimension and a basis of the solution space equations x + 2y Az + 3r s =
X
2x
+ +
2y iy
4s
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
(2,1,0,0,0),
V2
(1,
0,
1, 0),
vs
= (3,0,1,0,1)
The
W.
COORDINATES
be a basis of an ndimensional 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
^
R}
2:
V The polynomials
ei
{af2
5t
+c
a, 6, c
1,
02
63
CHAP.
5]
BASIS
AND DIMENSION
e;
93
x,
+ zeg.
22
y and
z:
xe^
3/
Then
same powers of
X
y
y
=
j/
2z = 5
z
The
is
a:
3,
1, z
[v]^
2.
(3,
Thus
1,
2)
3ei
62
2e3,
and so
Example
5.12:
tt?.
(3, 1,
(1, 1. 1),
/a
(0, 1, 1),
/s
/;
4) relative to
v
(0, 0, 1).
x,
y and
z:
xfi
(3,1,4)
= ^ =
a;(l, 1, 1)
(x, X, x)
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
e^
(1, 0, 0),
(0, 1, 0),
eg
v.
identical to v itself:
[v]^
(3, 1,
4)
have shown above that to each vector there corresponds, relative to a given ., e}, an ntuple [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 onetoone correspondence between the vectors in V and the wtuples 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
(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 onetoone 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
Theorem
5.12:
Let
field
K.
Then
and K^ are
isomorphic.
94
BASIS
AND DIMENSION
[CHAP.
^=a
5.4,
I'D[B]
"^G
=
3 5
4\
AJ'
c = f
~"
9
U6
10
Example
[A]
(1,2,3,4,0,1),
(1,3,4,6,5,4),
[C]
(3,8,11,16,10,9)
Form
the matrix
Row
reduce
M to echelon form:
40l\
M
and
to
I
/123
to
40
2 5
11
10
6/
\00000
11
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.
if:
u = u = =
(3,4), V
(1,3)
(iii)
(ii)
(2,3), V
=
4N
(6,9)
(iv)
/l2
(V) '
[so
5*
1/
/24 8\ 02/ [g
i3,
23N /l (,i), = 5 i) ^6
5t^
9'3
^/65
[l
2
3
(vii)
M = 2 
6*2
+ 2t4t^ +
(viii)
Two
(i)
one
is
(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.
(1,2,1),
(1,
(2,
1,1),
6),
(7, (3,
4,1)
1, 1),
(2, 4,
(iii)
(ii)
3,
7), (2, 0,
1.
5)
(iv)
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
 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
53
to
Since the echelon matrix has a zero row, the vectors are dependent. generate a space of dimension 2.)
(ii)
(iii)
Form
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
5.3.
Let
V be
A,B,C
GV
(i)
Set a linear combination of the matrices A, scalars x, y and z; that is, set xA + yB + zC
B =
and
0.
:)
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
2xy 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.
 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/
we
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
+ 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:
each equal to
3z
+ + 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
In the equation
xe^*
or
a;
X
Solve the system

=0
+ +
y Ay
xe^ xe*
+ +
2z
= =
0,
0,
0.
Hence
(Ii)
/,
Method
1.
sin
3/
cos
zt
0,
substitute
Q
= =
to obtain
tt/2
TT
xQ + 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 =
0,
0, z
0.
Hence
/,
Method
Take the
to get
first,
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,
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
2/(cos2 e
+
zt
sin2
or
implies
y
a;
=
=
0,
j/
Since
/,
fir
x sint
y cost
0,
5.6.
Let
u,
V and
be independent vectors.
Show
+ w) 
that
u + v,
where
x,
uv
and
u2v + w
are
also independent.
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
xy
The only
solution to the above system is
+ z = 2z =
2
= 0.
0,
0, z
Hence u
v,
uv
and
u2v + 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^
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
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
(i+i)w
+ i)(l,
+ i) =
(1
+ 1,
2i)
+tG
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^,
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,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,
6i(aiVi
Vi.
a0
b^a^Vi
 b^a^Vm
is
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
+ <hiVii.
Vj
Suppose
aj'Ui
ajxyji.
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
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^
1;
then
a^Vi
0, a^ =
and so
0.
But the
Vi
That
is, y^ is
5.12.
Prove Theorem
5.1:
Ri,
linearly independent.
Suppose {RntRni,
of the preceding rows:
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,
5.13.
Suppose
(i)
{vi,
.,
Prove:
If
w GV,
then {w,
.,
Vm)
is
linearly dependent
and generates V.
[vi,
. .
(ii)
If V{ is a linear
.,ViuVi+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^^Vii) +
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'ii, 1^1+
still
,^'ot}
generates V.
In other words,
we can
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,
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!)
{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
That
is,
since
(2)
..,Vji,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
.
. . ,
.,
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
{Wl, ...,
Wm.
'"*nm^
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,
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
is
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)
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
S =
By
of
(iii)
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,
basis
.,v^} as a subset.
By
n
(ii),
is
part of a basis.
contains
elements.
is
a basis.
5.19.
Prove Theorem
5.7:
Let
Whe a subspace
dim
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
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),
{^1
Vr, Ml,
...,_,}
Let
and
{vi, ...,
are bases of
and
W respectively.
B 7111,
{Vi,
...,V
Ml,
..,_ Wi,
.,
Wr}
+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)
UnW
diVi
hy
(2);
and since
{w J c
TF,
:
. .
W
. ,
by
fo""
(5).
Accordingly,
vGUnW.
+
and
d,
which v
dj^i
d^v^.
Thus by
(^)
we have
d^V^
CiWi
C_rW_r
=
Cx
But
{vi, Wfc} is
c_, = 0.
= 0,
+
is
a^Vr
bjUi
b^nrUmr
But {f
a,
j,
Uj} is a basis of
and so
independent.
= 0,
= 0,
61
= 0,
6to_^
= 0.
aj, 6^
and
c^ are all 0,
B=
{vi, Uj,
w^}
is
independent
5.21.
Prove Theorem
Let
5.9:
of
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)
is
row
>
space:
(611, 612,
f>i),
S2
(621. 622.
>
^2n).
^r
i^rV ^r2.
*m)
is
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:
is
I'l
CHAP.
5]
BASIS
AND DIMENSION
has dimension at most
r, i.e.
103
rank
the
column rank
r.
Hence, column
we
obtain
Thus
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),
1,
0)
(iv)
(1, 2, 5)
(5, 3, 4)
and
(i)
(2, 1,
(ii).
2)
is
and
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:
to
The echelon matrix has no zero rows; hence the three vectors are independent and
basis for R^.
(iv)
so
form a
Form
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
(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)
row reduce
1
to echelon form:
2
7
3 3\ 2
to
2
7
3
2
to
9 79
14
9
18
2)
(1,
2,
5,
3) and
(0, 7,
9,
is,
of
W. Thus,
in particular,
dim
W=
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
23 st^'
+ Qtl
Vi
2t^
~5P + 7t +
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^
W=
5.25.
W of the system
=
= =
+ 2y + 2z+ 2y +
3z
8z
s s s
+ + +
St
i
+
+
Sx
6y
5t
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,t0, =
= =
0, s
1,
0,
(iii)
0,
s^O, tl
(2,
1, 0, 0, 0),
V2
(5, 0,
2,
1, 0),
^3
(7,
0, 2, 0, 1)
The
W.
5.26.
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
1
I
IO25II0
\x
2 1 1
y
3\
4
/I
2
1
/I
2
1
1 2
y
z
\^\(iQ2x + y + zhxy + 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
erators of
(x, y, z,
r(l,
2,
r,
1, 1,
4)
2,
,.v
5)
in
unknowns
and
t is
CHAP.
5]
BASIS
AND DIMENSION
= = 2x + y = z = w 3x
oe
105
= X = y 2r s s2t = z 3r + 4s + 5t = w
r
r
^
+ t 8 + 2t s2t 8 + 2t
8
s 8
+ +
2t
= = = =
x 2x
2x
5x
+ + +
y y
y
+z w
^^^
Thus V
E.
W
is
ii
and only
if
i.e. if
2x 5x
+ +
+ z =0 w =
(1) is
The above
Let
and
of R*:
U =
We
+c+d =
(i)
0},
W
(ii)
{{a,b,c,d): a
(iii)
+b =
0, c
2d}
U,
W,
UnW.
b
(a, 6, c, d)
of the equation
+
d
=
Set
(2)
or
0a +
+
0,
=
a
a, c
0,
and
d. 0,
1,
= =
=
=
0, c
1,
(3)
0, c
0,
(1,0,0,0),
V2
(0,1,1,0),
3.
1^3
(0,1,0,1)
The
(ii)
v^
is
a basis of U, and
dim U =
(a, 6, c, d)
We
of the system
6
c
= =
1,
a
or
2d
+ 6 =  2d = =
0,
The
Set
6
0,
(2)
(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
(a, b, c, d)
ditions defining
W,
i.e.
6+c+d=0
0+6
c
a+b
or
6
=0
+
=0
=
2d
c+d
c
2d
= =
is
a basis
5.28.
(1, (3,
t^
(1, 1,
2, 3)
4,
(v)
{\
/
1
and
^J
l\
(ii)
2,
6)
/3 3
(iii)
2f2
+ +
and
i^
2t^
+
4
4*^
6f
^^^^
(^i i^
3 and
(iv)
i2f2
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
.
= (;:)
is
. =
(?;)
is,
 =
(1)
(::)
V
and
(2)
We
a basis of V, that
that
it
generates
is
independent.
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
we
obtain
0,
0
In other words,
xEi
Accordingly, {EijEz.Ea}
yE2
zEs
implies
0,
independent.
Thus
is 3.
5.30.
of degree
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 ml 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
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
= (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
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.
Show
is
that {E,^
dimV = mn.
We
Let
and
independent.
o,^E^^.
A=
(tty)
A = 2
{.
Now
suppose that
is 0.
ajjii^ij
= =
0,
where the
i
are scalars.
j
The yentry of
2 ij^ij
is .
and
are
Thus
asy
1,
.,
w,
m.
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.
Suppose
sion 6.
are distinct 4dimensional 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
5.34.
Let
J]
and
{(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
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
+2dim(!7nW)
or
Aim{Ur\W) =
5.35.
Let
2,
2, 3), (1, 4,
3,
4, 2), (2, 3,
and
let
6,
(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
4
3 3
5
2
1 1
2 2 3 4 1 2
2
2
9
1
*
'
6
3
6
2 2
3
1
2440
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
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)
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
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
row equal
z
to
to obtain the
9x + 3y +
0,
4x
2y
0,
=
set is
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).
of R^
unknown
scalars x, y and
z:
a;(l, 1, 1)
j/(l, 1, 0)
z(l, 0, 0)
is
{x,y,z).
(The solution
(i)
(4,3,2)
= = =
a;(l, 1, 1)
(a;,
0)
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,
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
GV
{{I
Set
iHri)^{iiHi

I)}
w (!?
unknown
scalars x, y,
z,
w:
J)
"I i)*'(:i)'(ii)*<i
w
\X
'
x)
'^
\y
X
0/
\0
z^
0/
+
X
+w
y
y
X
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.)
5^8.
Let
(See
Problem
5.29.)
^
[
, ^
2\
/2
1\
4 1\1
2
Set
1/' vi
sy
=
\i 5/j
2\ ij +
,
x,
y and
z:
11\
^ = (_n
7J
/I %2
/2
1\
^l
3J
%l
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
and
{/i,
h,
fa}
(of
dimension
3).
Suppose
62
63
= =
Ci/i
+
+
ttaA
hifi+b^fi
Ci/i
C2/2
(i)
Let
and
es respectively,
'
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
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.
'ax
61
Ci\/r\
2lS
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.
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
12
3\
to to
4
3 8
314
4 7 3 17 8/
2
3
Since the echelon matrix has two nonzero rows, rank (A)
(ii)
it is
easier to
to
to
3.
112
BASIS
AND DIMENSION
is
[CHAP.
(iii)
The two columns are linearly independent since one rank (A) = 2.
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 nsquare matrix.
Show
that
is
invertible if
and only
if
rank (A)
= n.
Note that the rows of the wsquare 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.
the
iij,
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,
.,
v^eAi^, v^eAi^,
Let k be the
...,
veAi^
. . . ,
maximum
is
Aj.:
Ui,'y2,
i).
Ai C Ag c
and
by
{1),
A^
is
linearly dependent,
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
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,
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
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
{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
(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
+
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
+ 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)
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
= dim ?7 +
dim W.
5.49.
of finite dimension.
Show
it
W
. .
UW.
is linearly
Let
{mj,
m,} be a basis of U.
.
Since {mJ
independent,
can be extended to a
. . . ,
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
. .
Let
Since
{i;i,
...,v} generates
over L,
is
a linear com
Since {oi,
a^)
fen'Wn.
&{
S^
a,
W
with co
is
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
over X.
is
The proof
Xj,
complete
if
we show
is,
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 =
L
.
Kiiffli
Xi2a2
a;imm
= =
0,
.,
x2a2
ar^i
+ =
a;m
But
{ffli,
is linearly
K,
0,
ii
0,
Xi2
im
0,
...,
0,
a;2
...,
Accordingly,
{ttjVj} is
is
proved.
Supplementary Problems
LINEAR DEPENDENCE
5.52.
(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.
(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
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,
^=(r4s
3 over R.
Let be the vector space of polynomials of degree linearly dependent or independent where:
(i)
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
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)
R;
(ii)
+ \/2)
and
+ 2'\/2
u,
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
v and
5.60.
Suppose
(i)
. ,
{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
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.
1) and
3),
(0, 3, 1)
(ii) (iii)
(0, 1, 1)
and
(3,
(0, 1,
1)
6),
4),
1,
4)
and
(2, 1, 1)
(iv)
3) and
(2, 3,
11)
5.66.
W of R* generated by:
7)
(1, 4,
(1,
1,
3),
(2, 1,
(1,
(0, 2, 1,
(3,
5)
(ii)
4, 2,
1),
and
8, 2,
5.67.
Let
and
let
W be the
subspace generated by
ui).
5.68.
(:
\y (r:)
and
(4;;
(3
2*2
2t
1,
t)
t*
3*2
w =
2*3
t2
7t
CHAP.
5]
BASIS
AND DIMENSION
117
5.69.
+ + +
Sy 5y 5y
+ + +
(i)
2z
z
8z
= =
+ 
2y
3y
y
2x 2x
+ 
7z
2z
z
+
(li)
+ +
2z 5z
= =
(iii)
5.70.
X
2x
+ 2y ~2z + 2s  t = + 2y  z + 3s  2t = + 4y ~ Iz + s + t =
(i)
+ + +
2y
4y iy
2z 2z
5.71.
set
is
generated by
3),
(1,
2,
0, 3,
1),
(2,
3,
2, 5,
2,
1, 2,
2)}
5.72.
Let
V and
62c + d =
(i)
0},
W
(iii)
{{a,b,c,d):
d,
2c}
V,
(ii)
W,
VnW.
n.
of degree
{1, 1
+ t,
(ii)
{1
+ t,
+ t+t2, l+t+t2 + t3, ..., l + t+t2+ t+ fi, t^ + t, ..., t"2 + t"i, ti + t"}.
l
+t"l +
t}
subspaces of R3.
Show
4,
that
VnW
{0}.
Suppose
and
W are
subspaces of
U=
U
dim
W=5
W=2
and dim
V
7.
Find the
possible dimensions of
U nW.
dim
1,
5.76.
dim
and
UlW.
Show
that
5.77.
Let
U be
let
(1, 4,
(2, 9, 0,
5, 2)}
and
W be
dim
2,
3), (2, 8,
(1, 3,
Find
5.78.
(i)
{U+W),
(ii)
Let
{t3
+ 4*2  t + 3,
+ 5*2 + 5^
(i)
3*3
10(2
 5t + 5}
(ii)
and
respectively.
Find
dim
(f/
W'),
i\m.(VnW).
5.79.
Let
U
let
0), (1,
2, 2,
0,
3),
(1,
1)}
and
W be the
subspace generated by
{(1,
2, 3,
0,
2),
(1,
1, 3,
2,
4),
(1,
1, 2,
2,
5)}
(i)
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
{(11).
Find the coordinate vector of the matrix
:)(' 1)}
relative to the above basis
if:
AGW
I)
...
= =
(4
I)
<"'
^<l
63
5.83.
(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)}
(3,5,2)
and
(ii)
relative to the
(iii)
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
SF
relative to
any basis of
is
(0,0, ...,0).
RANK OF A MATRIX
5.86.
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.
n.
set with
w elements
is
basis.
CHAP.
5]
BASIS
AND DIMENSION
Let
119
5.92.
Let
tj, 2,
<B
+ at
where
ajti
+ 03*2 +
+ 02*2 +
6)
K+
+
>)
by
+ 2*2 +
.
.
+ o<) =
ka^ti
ka^t^
feat
{t^,
. . .
Show
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
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):
to
and that
is iso
morphic to
(ii)
Show
that
dim
V =
dim
V+
dim W.
5.95.
Suppose
to
V = U
W. Let V be
U and
W. Show that
V is isomorphic
= u+w
<r^
{u,w).
Answers
5.52.
(i)
to
Supplementary Problems
(vi)
no,
(ii)
yes,
(iii)
yes,
(iv)
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)(lt,i);
(7, l
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.
dim ly
1.
(ii)
dim ly
0.
(iii)
dim
W=
1.
5.70.
(i) (ii)
dim Ty
3.
(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)};
dim
V=
3.
(ii)
dim Tf
2.
(iii)
dim (VnH^)
1.
Hint.
VnW
must
and
yes,
d.
5.73.
(i)
(ii)
no.
For dim
V = n + 1,
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,
= =
5.80.
(i)
= =
=
(5/3, 4/3),
(ii)
[v]
M=
0),
(0, 3),
(iv)
[v]
((a
6)/3,
(a26)/3).
5.81.
(i)
M
[A]
(2,
5,
7,
2),
(ii)
(0, 4,
1,
(iii)
[v]
(a
+ h + c + d, b2c3d, 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
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 codomain.
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
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}.
defines a
mapping
Here
f(a)
y,
/(6)
x,
f{c)
z,
and
f(d)
y.
Also,
faa,b,d})
=
is
{y,x,y}
{x,y}
The image
Example
6.2:
(or:
range) of /
f(A)
{x,y,z}.
Let
/:R
its
square
a;^:
V^
x^
or
is
9 so
we may
write
= a;2 /( 3) = 9.
f{x)
121
122
LINEAR MAPPINGS
[CHAP.
i^
to denote the
eA
under a mapping
fix)
A = A
that
'1
3
4
5' If
,2 c
is,
we
1,
determines the
T{v)
Av,
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
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(25* +
Note that this
2)
(3t^5t
+ 2)dt
i
whereas the
map
is
map
Example
6.6:
in the preceding
itself.
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
Theorem
6.1:
ho{gof)
(hog)of.
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)
{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
Definition:
A
if
mapping f:A*B
is
different elements of
A
if
is,
implies
f{a')
/(a)  f{a')
or, equivalently.
if /(a)
implies
a'
Definition:
mapping f:A^B
is
maps
onto B) or surjective
if
every b
GB
is
G A.
said to be bijective.
be defined by f(x)
follow:
Let
h(x)
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 oneone; 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 oneone 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 /:AA 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
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 onetoone 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)
For any
For any
v,wGV,
kGK
+ 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,
124
LINEAR MAPPINGS
[CHAP.
Now
gK
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
viGV we
+ aiVi +
We
linear
anF(Vn)
completely characterizes
its definition.
Example
Let
be any
mapping
T:&^K^
T(v
mX%
As noted
y>
Av.
We
claim that
is linear.
+ w)  A(v + w) = Av + Aw =
T(kv)
T{w)
and
A(kv)
= kAv =
kT{v)
In comment that the above type of linear mapping shall occur again and again. show that every linear mapping from one finitedimensional 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)
(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
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
I.V^V
v,wGV
and any
/(av
bw)
a,beK, = av + bw =
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
dv du + v) dt = dt+dl
,
djku)
^""^
j^du
dt
also,
"dt
that
is,
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
:
is
mapping F"!
is
exists.
We
will
show (Problem
both oneone 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
onetoone.
The
there
an isomorphism of
Let
its
of dimension
n and
is
let {e^,
.,e} be a basis of V.
[v]^,
eV
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.
.
Let
of
F:V^U
be a linear mapping.
in U:
is
the set
image points
Im F =
{uGU
=u
for some v
G V}
is
OGU:
Ker The following theorem
is easily
F = {vGV:
0}
proven (Problem
6.22).
Theorem
6.3:
Let
of
F:V>U
/^ is
a subspace
Example
6.18:
is
the
l l
entire
xy plane:

llill
(a, 6, c)
Im
F =
{(a, 6, 0)
o, b
G R}
the z axis:
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:
AX
nr 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
+ 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)
= =
+ bF{w)) a(kF)(v) +
b(kF){w)
Thus
F+G
and kF are
linear.
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
and
are of finite
Theorem
and
6.7:
Suppose dim 7
and dim
G:U^W
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
That
is,
G o F is linear.
CHAP.
6]
LINEAR MAPPINGS
of linear
129
The composition
related as follows:
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)
(iii)
k{GoF)
= {kG)oF =
Go(kF).
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
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)
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.
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
tto
+ =
aix
+ +
a2X^
ax",
aiGK
ar"
if
we can form
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 ri
e A{V)
is
such that
TT^ = T^T =
Now T
suppose
over,
invertible if
and only
can
gV
finite
is
nonsingular,
i.e.
Thus in particular, if T is if it is oneone and onto. map into itself, i.e. T is nonsingular. On the other hand, Ker T = {0}. Recall (page 127) that T is also oneone. More
assuming
has
dimension,
we
have, by
T)
Theorem
6.4,
dimF
Then
= =
i.e.
dim(Imr) + dim({0})
=
V; thus
dim (Im T)
ImT V,
6.9:
the image of
is
is
onto.
Hence T
is
and so
is invertible.
We
Theorem
A
6.22:
linear operator
T:V*V
if it is
vertible if
and only
is
in
Example
Let T be the operator on R2 defined by T(x, y) = (y, 2xy). 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 Ti. T(x,y) = (s,t) and T'^(s, t) = (x, y). We have is the image of (s, ) under ri;
{(0, 0)};
T(x, y)
(y,
2x
y) =
s
t)
(s, t)
and
so
a;
s,
2xy =
+
li,
2/
is
and
t,
we
obtain
s).
is
s.
Thus T^'
(s
+ it,
in the preceding
theorem
is
necessary as seen
Let
let
be the operator on
a< +
defined by
+ ait\
+ ajn) =
t
a^t
Uit^
i.e.
T
is
in each
is
term by
1. is
and
nonsingular.
However, T
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 onetoone 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
K which
Theorem
6.10:
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)
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.
B=
mapping from
A=
{a, b, c)
into
(i)
(ii)
v(iii)
(i)
is
G A.
A.
(ii)
Two
elements, x and
are assigned to c
(iii)
6.2.
Use a formula
(i)
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,
2 and
we can
0.
number z
43
its
cube
a^,
define /
by
/()
/(O)
64,
/(2)
(2)3
8,
= =
a:^.
Also:
0^
(ii)
number
flr(4)
x,
we can
5,
define
g by g{x)
5,
ff(0)
5.
number
4,
2 and
is 5:
fl(2)
(iii)
Two
h as
follows:
'a;2
h{x)
if if
X
a;
>
=s
Since 4
>
0,
h(4)
42
16.
On
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;
6.4.
(i)
f{x)
^x^ + x6,
(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.
f.A^B
and
f
g.B^C
(i)
(ii)
(i)
We
g(y) g(x)
g{y)
= = =
s
t
we
we
X
*
*
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,
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.
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;22)
2(a;22)
2a!2
3.
Observe that
fog^gf.
6.7.
D
a
Determine
(i)
if
each mapping
:
(i)
is
oneone,
(ii)
is
onto,
(iii)
has an inverse.
The mapping /
g.B>C
is
(ii)
is
oneone since each element of A has a different image. The mapping not oneone since x and z both map into the same element 4. The mapping h:C * D
>
is
oneone.
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)
if
and only
if it is
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 oneone, then gof is oneone. (ii) If / and g
are onto, then
gof
is,
onto,
(iii)
If
gof
is,
oneone, then /
is
oneone.
(iv) If
gof
is
Suppose (gf)(x)
is
(gf){y).
Then
g(f{x))
g{f(y)).
Since g
(g f){y)
is
oneone,
y.
We
{g f){x)
oneone.
(ii)
Suppose
exists
is onto,
there exists
b.
h
''
e.
f(a)
Thus
(g
f){a)
/
o
is
onto, there
is onto.
flr
(iii)
Suppose /
Thus
is (iv)
is not oneone. 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 oneone. Accordingly oneone, then / must be oneone.
=
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
^
for which
f~^f = 1a and
is
ig.
Since 1a
6.8(iv).
is
if
is
onto by Problem
6.8(iii);
and since 1b
onto, / is
Now
in
suppose /
6.
Then each
SB
is
A, say
to
Thus
b,
then
=
6
b;
hence /(6)
b.
Now
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
mapping
6.10.
fix)
= 2xS. Now
/ is onetoone
f~^.
the Let y be the image of x under the mapping f: y  f(x) =2xS. Consequently x will be for x in terms of y in the above equation: image of y under the inverse mapping fK Thus solve X = {y + 3)/2. Then the formula defining the inverse function is fHy) = {y + 3)/2.
LINEAR MAPPINGS
6.11.
Show
(i)
are linear:
R2
^ R2
=
defined
by F{x, y)=^{x
(ii)
F:R^*R
Let v
defined
by F{x,y,z)
(i)
(a,b)
and
V
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.
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)
2ka
3kb
4kc
k(2a
 36 + 4c) =
Accordingly,
is linear.
6.12.
Show
(i)
(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)
Then
fcF('y)
Fikv)
and hence
F is
not linear.
6.13.
Let
V be
in F.
Let
the vector space of nsquare 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
T{A +B)
and
Accordingly,
T{kA)
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.
. . .
(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
+ ai;. We define F:V ^ U by F{v) = a^Ui + a^u^ mapping F is welldefined.) Now, for i= \, ..., n, + Ivi + + Ov Vi  Ovi +
GX
h aM.
Hence
F{Vi)
first step
{2).
Omj
Imj
Om
m;
Thus the
Step
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
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^
+ hi)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))
V,
G=
Thus
F is
is
proved.
136
LINEAR MAPPINGS
Let r R2
:
[CHAP. 6
6.15.
and
r(0, 1)
= 2
is
{1)
is
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)
y:
(2)
x{l, 1)
j/(0, 1)
Then
(a, 6)
(x, x)
(0, y)
=
we
a
(x,x
+ y)
y
and so
xa,
a
+ y=^b
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
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.
. .
ai,
. ,
o,
a^v^
r(0)
Tia^vi
+ a.2V2 +
all
+ ag^a + h ai') =
the
O;
ar(v)
linearly
0.
6.17.
F:V^U
is
is
Show
is also linear.
Since
F
u'.
onetoone 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+
Fi(tt)
u'
and
v.
F{kv)
=
f',
kF(v)
ku
By
mapping,
"'
Fi(m')
Fi{u\u')
+ v'
and
F^ku) =
Then
FMm + m') =
and thus F"'
is linear.
Fi(m)
FMm')
and
FHM
fei)
feFHtt)
IMAGE AND KERNEL OF LINEAR MAPPINGS Let F R*  R5 be the linear mapping defined by 6.18. F{x,y,s,t) = {xy + s + t,x + 2st,x + v + ?>sZt) image U of F, (ii) kernel W Find a basis and the dimension of the
:
(i)
of F.
(i)
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
to
to
Thus
C/
2.
CHAP.
6]
LINEAR MAPPINGS
137
(ii)
We
(x, y, s, t)
y, s,
t)
F(x, y,s,t)
= (xy + s +
+ 2st,x + y + BsSt) =
to
Set corresponding components equal to each other whose solution space is the kernel of F:
X X x
y+s+t
+
y
2s
=
or
y+s+t
y
=
or y
= + 3s3t =
t
and
t;
hence
+ s  2t = 2y + 2s  4t = dim W = 2. Set
(2, 1,
2t
s s
= =
1,
0, t
=
l
1,
0),
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
{x
+ 2y z,
(i)
+ z,
+ y 2z)
(ii)
image
U
U
of T,
of T:
kernel
W of T.
i,
(1,0,1),
r(o,i,o)
(2,i,i),
r(o, o, i)
(1,
2)
Form
U
1
to echelon form:
1
1
1 1
("
2
1
to
1 1
1 11
'\
to
1
2
1
Thus
(ii)
1)}
is
U=
2.
We
=
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;
= 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.
R^
(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
2\
1 1 4 3 3
is
Recall that
determines a linear
Thus
6.21.
Let
and
let
Let
F:V^Y
We
W of F.
^\
such that
map
defined
by F{A^ =
AM MA.
(p
(^
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
= =
t;
hence dim
2.
To obtain a basis of
W set
t
1,
0,
t
=
1
=
1,
1,
1, s
0,
0;
0, s
0, t
1.
^^"^{(o
o)'
G
6.3:
;)}
a basis of T^.
6.22.
Let
(ii)
F.V^U
and
the kernel of
(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
is
a subspace of U.
Now
suppose
v,wG Ker F
0.
belong
and F(w)
=
==:
Thus
bw)
aF(v)
bF{w)
aO
60
and so
+ bw S KerF
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
Let
B =
is
{F{Vi),F(v2), ...,F(v^r)}
The theorem
proved
if
we show
that
Proof that
U' of F. v
&V
u.
Since
where the a
ftj
are scalars.
+ b^r'^nr since the Wj belong to the kernel = F(aiici + + a^Wf + biv^ + + b^^vr) u = = aiF{wi) + + 6_^F(i;_,) + a^(Wr) + b^Fivi) + = OjO + + bnrF(Vnr) + a^O + biF(Vi) + = b,F(v,) 6_,FK_,)
V
OjWj
a,Wr
^l'"!
of F.
Thus
jF'(t')
++
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
anr'Un,
b^Wi
62^2
b^Wr
(*)
or
ail^i
anr'Wnr
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
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'GfHu). Then f(v') = u and so f(v'  v) = f(v')  f{v) = uu = 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
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
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.
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
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
is
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
+ Q){v) =
v
Since
(ii)
+ 0)(v) =
[F
F(v)
for every
F{v)
F(v)
F.
For every v G V,
Since
of F.
{F
F{v)
F{v}
0{v)
is
(l)F
0.
Thus (l)F
the negative
6.30.
Show
By
Hom {V,
aiFj,
U) and
ai,
vGV,
n =
1.
+ a2F2 H
+ aF)(i;) =
aiFi{v)
+ aJFiiv) +
ajf'niv)
definition of the
mapping
(a^F^iv)
a^F^{v);
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)
Show
that
G K,
(Here
is
(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),
0,
(*)
(1)
implies
and
6.32.
Prove Theorem
Suppose
{vi,
.
6.7:
.
Suppose dim y
and dim
. .
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).
Hom (F,
G
U,
U).
Let
Suppose
u's; say,
F{vi)
w^, F(v2)
=
(i)
it is
Wk =
afclMl
fc2*2
TTi
+
n
fflfcMn>
=
is
1,
. ,
m,
Oy
G =
Since
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=
is linearly
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
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.
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?'
Goii'';
(ii)
{G
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)
is linear.
Show
that
(i)
G(F{V)) c G(U)
dim G(F{V))
= dim =
((GoF)(y))
= dim =
(G(F(y)))
^ dim
(ii)
By Theorem
6.4,
rank (GoF)
dim F(y)
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)
(2SZT)(x,y)
6.36.
Let
(2,
4)
and
T{1, 1)
=
is
(0, 2)
(i)
(By Theorem
First write
6.2,
unique.)
Find T{a,
b).
In
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
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)
oo(2,
(2a;,
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 + 3yz).
=
(0, 0, 0),
i.e.,
(i)
Show
W of T
is
(x, y, z)
(2a;,
y, z)
4x
y,2x + Syz) =
(0, 0, 0)
Thus
is
0,
4x
0,
2x
Sy
=
is
(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
=
7r
(2a;, 4a;
y,2x + 3yz) =
Thus T~^
is
{r, s, t)
we
find
^r,
2r
s,
Ss
s, t)
t.
given by
ri(r,
= (^r,2rs,lr3st)
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 onetoone. 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
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
if
V is
n.
of infinite dimension.
the preceding problem,
Let
is
dim
V=
By
is
invertible if
We
have
n = rank
(ii)
rri = rir =
Let
Then s
/s
= (rir)s =
t
ri(rs)
p(t)
= r/= =
ao
ri.
(iii)
over K; say,
+ Oji + Ojf^ +
a^t
+ at".
Let
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
+ a<+i)
and so TS
S(T(k))
I,
S(0)
On
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
Since
TS
assigns
(0,0)
to every
(ST){x,y)=S(T{x,y))
it
(0, 0)
(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.
and
{/i, /z)
a basis of
TJ.
Let
T.V^U
be linear.
T(e3)
= = 
aifi
&i/i
Ci/i
and
A = ('''/
^
''^
Show that,
for any v
GV,
A[v]e
[T{v)]f
in
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
Show
map T
is
singular
if
and only
if
kT
=
is
singular.
0.
Suppose
singular.
Then
Hence
ikT){v)
kT(v)
&0
and so kT
kT(w)
is
singular.
#^
{kT^(w) implies
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,
E{v)
u ^ E(v)
(ii)
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
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,
UnW
=
eUnW.
Since
{0}.
E(v)
and
so
UnW
vGU,
E(v)
v by
(i).
Since
V = U
W.
CHAP.
6]
LINEAR MAPPINGS
invertible if
145
6.44.
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.
mapping from
6.46.
R > R by
3.
a formula:
To each number
let
(ii)
To each number
(iii)
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.
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.
F.B^C
and
GiA^C
Determine whether each of the following defines a composition mapping and, if it does, find domain and codomain: {\)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)
= 2x3.
Find formulas
mappings
fg,
(ii)
gf,
f
gg,
ff.
6.53.
f'^A
146
LINEAR MAPPINGS
For each of the following mappings /
Sx
[CHAP.
6.54.
R > R
find a
(i)
f{x)
 7,
(ii)
fix)
+ 2.
LINEAR MAPPINGS
6.55.
Show
(i)
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)
F
=
(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)
\xy\.
over K.
6.57.
Let
V
>
S :V
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"^
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
Fiv)
of
Fiv).
V, denoted by
6.62.
W be
defined by
W into
i:W cV
and
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,yz,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
Let
F V * V
:
be the
linear
(ii)
= MA. ^ RS ^ RS
(i)
the kernel TF of
and
6.65.
R3 R*
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.
Let
D:V *V
6.68.
Let
(ii)
(i)
is
is
a subspace of
and
CHAP.
6]
LINEAR MAPPINGS
K,*
147
6.69.
into R^:
'12
(i)
1^
A =
2
^1
1 3
2
2
1
I
(ii)
B =
2/
W of each map.
Let
or
r C
:
>
bi)
C be
the conjugate
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.
defined
F+G
(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)
and
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
{V, U),
show
that
rank (F
G)
^ rank
i^
has
finite
dimension.)
6.75.
Let
F :V ^ U
nonsingular.
Show
is
that if
and
GF
is
nonsingular but
is not.
6.76.
Prove that
Hom (V,
Theorem
6.6,
That
is,
prove
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.
T{x, y)
{x
+ 2y,
3x
+ Ay).
_ 5f _ 2.
T on
R^
is
6.79.
Show
(i)
invertible,
and
y).
find
a formula for
T~h
T{x, y, z)
= (x3y 2z,
 4,
z),
(ii)
T{x, y,z)
{x
+ z,x z,
6.80.
Suppose
sion.
S and T
Show
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
^ Im ?i Im 2ST
is
6.83.
Show that
if
S and T
invertible
and (ST)^
T^S^.
148
LINEAR MAPPINGS
Let
[CHAP.
6.84.
have
that
finite
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
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
6.87.
Show
that
if
F:V ^ U
is
onto, then
dim
U
dim V.
Determine
all
linear
maps
T:W*R*
Find those theorems of Chapter 3 which prove that the space of wsquare 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.
(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
(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;)
(iii)
(g o g)(x)
(/
=Ax9
=
a;*
(iv)
/)(a;)
6.54.
(i)
fHx)
(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)
dim
U=
(6) {(2,
W=
\.
(ii)
dim
?7
1; t/
(6) {(1,
1)},
dim T^
1,
1)},
1.
(iii)
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 image of
is
6.69.
(a)
(6)
{(1,2,1), (0,1,1)}
{(4,
basis of
2, 5,
R3;
0), (1,
3,
0, 5)}
2.
(ii)
(a)
ImB =
(6)
{(1,2/3,1,1)}
KerB; dim(KerB)
z)
l.
6.71.
(F
G)(,x, y, z)
(y
(3F
2G)(x, y,
(3j/
Az,x + 2y + Zz).
6.72.
(i)
(iii)
+ z,2y), (HG)(x,y,z)  {xy.iz). (il) Not defined. (Ho(F + G)){x, y,z) = {HoF + Ho G)(x, y, z) = {2xy + z, 2y + 4).
(HF){x,y,z)
6.77.
(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.
T^\I
Q,
hence
is
a zero of
x'^
1.
6.78.
v{T)
6.79.
(i)
THr,
s, t)
(14*
+ 3s + r,
4t
+ s,
t),
(ii)
T^r,
s, t)
(^r
+ ^s,
t,
^r
 s).
6.87.
chapter 7
Suppose
ttiei
{ei,
e} is
+ 0.262 +
+ Omen. Then
\an
mapping v
l^ [v]e,
{Ci}, is
an isomorphism from
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 nsquare matrices
over K.
into another.
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,
denoted by
[T]e or [T], is
matrix representation of
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)
= = = =
150
CHAP.
7]
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
= n1,
=
2
2
(4, 2)
2(1,
2(l,
2/1
2/2
(3
Remark:
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
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:
D:V ^V
and so
t^,
in
Example
D{p{t))
7.1.
Let
= a+bt +
cfi
dt^
2ct
3dt^
{1,
t, t^,
[p(t)]
and
[D(p{t))]
We
7.1
o\
2
[D][Pit)]
=
iO
3,,
= [D(pm
o/\
:
Example
7.4:
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).
and
[T(v)]f
=
11
[T]f in
Example
7.2,
we
m,M,
(r^G)
<2y
(ID = i^<*
152
[CHAP. 7
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
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 oneone and onto
and, for any
keK,
and
[kT]e
+ S]e =
[T]e+[S]e
[ST]e
k[T]e
Theorem
7.3:
[S]e [T]e.
We
V, and
illustrate the
2.
Suppose
{ei, ez} is
a basis of
T and S
V for which
aiei
biei
= =
+ a^ez + 6262
S{ei)
'
8(62)
[^iCy
Now we have
{T
S(ei)
i^i
+ S){ei) + S){e2) =
T{ei)
aiCi
(tti
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)
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]
153
CHANGE OF BASIS
We have shown that we can represent vectors by ntuples (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
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
'
{/{}:
fflll
ft21
ftnl
ffin2
*12
^'22
We comment
{/,}
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:
(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
P from
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
Q
Observe that
IN
,
Let P be the transition matrix from a basis space V. Then, for any vector v G V, P[v]f
{Ci}
[v]e.
Hence
[v]f
P~'^[v]e.
We
{Ci} to
new
basis
is called the transition matrix from the old basis to transform the coordinates of a vector in the new
{ei}.
basis
{fi}
154
[CHAP. 7
We
illustrate the
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
= =
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.
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
Accordingly,
P[v]f
^^ &2 ba
^,^\
jaikr
a2
62^2
Caj\kaj
ttzifci
\a3
\a3k1
+ + +
[V]e
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.
= ^
= =
=
a(l, 0)
6(0, 1)
ae^
b{l,l)
+ (ba)(l,0)
and
+ =
be^
bfi
iba)/^
Hence
[v]^
Mf = P
(J^j
(ft
_ )
to {/J
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 = Pi[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]
155
We
compute
[T]f
using Theorem
7.5:
m,  pi... = (_:
Note that
this agrees
ixi ixi
[T]f in
I)
7.2.
(i I
Example
Remark:
Suppose
P
(Oij)
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
T form an
equivalence
linear operator
is
said to be diagonalizable
{{} is
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
and only
if its
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 welldefined 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
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
[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
..,
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
apply.
[F]l [v]e
Theorem
That
7.8:
is,
[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 oneone and onto
and, for any F,
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
into U.
Theorem
7.10:
Let
{ei}, {fi}
and
and
G:U*W
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.
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]
157
in particular,
when
JJ;
and
[F]iP
i.e.
when
and
Theorems
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 rsquare identity matrix. form of F.
We
call
WARNING
As noted previously, some texts write the operator symbol V on which it acts, that is, vT instead of T{v)
In such texts, vectors and operators are represented by ntuples and matrices which are the transposes of those appearing here. That is, if
felCl
feez
knCn
(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
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
[CHAP.
Solved Problems
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)
and
77
[T]f
124
43
69
7.3.
Suppose that T
is
T{x, y,
=
T
(ttiic
+ a2.y + aaz,
bix
+ h^y + bsz,
cix
+ dy + Cs^)
Show
by
a2
&2
C2
as
bs
Cs
61
Ci
That
is,
from the
(ai, 61, Ci) (02, 62, C2) (aa, 63, C3)
coefficients of x,
y and
z in the
com
= = =
T{1, 0, 0) T(0, 1, 0)
r(0, 0, 1)
= = =
= = =
+ + +
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]
159
7.4.
on R^
(ei
(1, 0, 0), 62
(0, 1, 0), 63
(0, 0, 1)}:
T{x,y,z) T{x,y,z)
(ii)
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)
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,
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
+ z,  4j/, 3a;) = (3,3,3) = 3/16/2 + 6/3 = (2,3,3) = 3/16/2 + 5/3 = (0,1,3) = 3/12/2 /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
and so
[T{v)]f
Thus
[T]f[v]f
6 6 2
6c
2a45
lT(v)]f
7.6.
Let V
is
A {ei
{/i
(
j
and
let
T be
= Av
(where
in
(i)
(ii)
i.e.
(i)
Tie,)
3e,
/I
^(^^)={3 4)(l)
160
[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
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)
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)
3(e30 l(e30
0(e3)
+ + +
0(e3') 3(e3t)
2(e3')
and
[D]
CHAP.
7]
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
Now
suppose
k^ei
kzBz
fce
K^i
(&i,
fcj,
...,fe)t
(2)
= =
T
n
(^
2 he^ = 2
/
hned =
n
fci (^
ae,
j 1 \ i=l
2(2
*>
) j
2
i
is
(oijfci
+ a2jfc2 ^
h a^fe)ej
Thus
[r(i;)]g is
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 oneone 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
The mapping
its
is
oneone 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
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
+ 6y).
Accordingly,
[T
+ S], = (A+B)t =
A'
fit
[S]e
We
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],
is
proved.
162
[CHAP.
7.11.
Let
{ei,
e}
be a basis of V.
n
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).
/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/ =
/265a\
(
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)
{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)
and
P =
CHAP.
7]
163
(ii)
cf^
(b
 c)/2 +
(a
 b)fs.
Thus
/o
1^
62
63
= = =
(1,0,0)
(0,1,0) (0,0,1)
and
Q =
11
\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 1114
\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.
Let
{cj}
vGV,
+
04262
P[v]f
[v]e.
Also,
n
[v]f
= P^[v]e.
P
is
to a basis {h)
A =
ajiei
+
n
aje
is
(oij, a2j, .... aj)
2 ^0^. j=i
Then
the square
(i)
+ kj =
[V]f
*i
Vj
(fci,
^2, ...,fc)t
Substituting for
/j
= =
2v,
n
2^i(ie,) = i(iA)i
(aijfci
+ a2jk2 +
+ ajkn)ej
is
Accordingly,
[11]^
is
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 =
Me
PiP[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 = Pi [T]eP.
For any vector
{/i}
in
vGV,
PHT]^P[v]f
P^[T],[v],
pi[T(v)]^
[T(v)]f.
164
[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,
Pi[T]^PX = [T]fX
for every
X iC.
Accordingly, Pi[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 = PAPi = (Pi)APi and P^^ is invertible. Thus B is similar to A. piPP, 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. piBP = P^(Q^CQ)P = (QP)>C(QP) and QP is invertible. Thus
The
identity matrix / is invertible
and /
/"i.
Since
TRACE
7.17.
The
= 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
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
such that
A = P^BP.
Using
(i),
tr(A)
tr(PiPP)
tr
(PPPi)
tr (P)
7.18.
R^:
(aiflj
+ a2y + a^z,
bix
+ h^y + hsz,
Cix
+ Czy + csz)
We
first
must
find a
h
Cl
62
C2
&3
C3 /
and
tr (T)
tr ([T],)
di
63
C3
7.19.
Let
Af
Let
T be the
linear
operator on
by T{A) = MA.
We
must
first find
a matrix representation of T.
CHAP.
7]
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.
Let
(i)
if
R3
^ R2
y, z)
{Sx
+ 2y4z,x5y + 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)
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
+ 2y4z,x5y + 3z)
and
,
{y
r,
/13a;
[i^()]a
( 8,
T^^
33 13 \/
19
SJUH
==
(8a;
,, ^, t^(^>^'
7.21.
Let
F:R^^K^
.
.
F{Xi, X2,
. ,
Xn)
{anXi
+ amXn,
021X1
+ aanXn,
OmlXi
+ amnXn)
of
Show
is
F relative to the
ai2
CI22
usual bases of
'
K" and
K"
(ttu
0,21
...
...
ttln (l2n
flml
(tm2
dmnl
166
[CHAP.
Xi in
the components
of F{xi,
.,
x), respectively.
0)
.
. .
^(1,0
F{0,
1,
,
0)
= =
/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
y)
{Zx
y, s, t)
y, z)
By Problem
7.21,
we need
.).
Thus
(2
6
0/
7.23.
Let
T:R2^R2
{ei
be defined by T{x,y)
the bases
(1, 0), 62
(0, 1)}
the matrix of
in
of R^ respectively.
as a linear
own
basis.)
7.2,
By Problem
r(ei) Tie^)
{a,b)
= =
r(l,0)
r(0,l)
f '
^ /8
\
5
23
"13
7.24.
Let
fined
(i)
A =
Show
of
/
(
3
.
\1 4
Recall that
is
F.W^B?
de
by F(v)
7/ Av where v
and
R*.
(ii)
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)
 561462
from which
W\ =
7.2,
2
,
3\
_.
A
7.7.)
(ii)
By Problem
(a, 6)
(26
 5a)flri +
(ZaV^g^.
Then
CHAP.
7]
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]
24
7.25.
Prove Theorem
basis of
C7 is
7.12:
Then there
exists a basis of
Y
(
and a
where
such that the matrix representation A of i^ has the the rsquare identity matrix and r is the rank ofF.
form
A =
\
the image of F.
Suppose dim
V
m,
and dim
r\
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
{%,
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
F in the
Supplementary Problems
MATRIX REPRESENTATIONS OF LINEAR OPERATORS
7.26.
(1, 0), 62
(0, 1)}:
(i)
r(, y)
(2x
3y,x + y),
(ii)
T{x, y)
(5x
+ y,Zx 2y).
{/i
7.27.
(2, 3)}.
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.
in
Problem 7.26
in the basis
{g^
(1, 3),
g^
(1, 4)}.
168
[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)
6a;
 83/ + z)
7.30.
D(f)
dfldt.
sets is
(i)
a basis of a
{e', e^t, te^'^},
^ R.
in
each basis:
(ii)
{sin
t,
cos
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)
2 matrices over
and
let
Af
T on V
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^
(1, 0), eg
(0, 1)}
and
{/i
(1, 2),
^=
(2, 3)}.
and
Q from
v
{gj}
to
(ii) (iii)
Show Show
that that
[v]^
[T]f
= 
P[v]f
fP.
in
Problem
7.26.
7.35.
{/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
is
similar to
itself.
Show
that
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
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 + 3y7st F(x) = (3x, 5x)
y, z)
(2x
 4j/ + 9s,
5x
CHAP.
7]
169
7.42.
Let
(i)
i<'
^ R2
y, z)
(2x
+ y z, =
Sx
2y + iz). =
(1,
(l,l,l), /2
(1,1,0), /a
(1,0,0)}
and
{jti
(1, 3),
fir^
4)}
(ii)
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.
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
is,
Show
that
/A B\
.
)
Let
y =
1/
W, and
let
and
T V
:
>
V.
Suppose
dim U A and
7.50.
/A
(
''
ON
where
Recall that two linear operators F and operator T on V such that G = T^FT.
(i)
on
if
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,
(ii)
an operator
F is
over
diagonalizable, then
is also
diagonalizable.
7.51.
Two
(i)
mX n matrices A
Show Show
and
matrix
Q and an nsquare 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
F :V > U
is
if
identity matrix
where /
the rsquare
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 *isquare 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
[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/
V27 32
7.29.
(i)
10 10
,0
'2
3
7 4I
14
1
(
(iii)
0,
68
1
7.30.
(i)
0'
2
1
(ii)
5
(iii)
101
5 2 5,
(iv)
'0
o\
03
iO
3
,0
2,
,0
0/
1
7.31.
(i)
1
(ii)
3 4 2 3
^c
h 6
6
7.32.
(i)
ad
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
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.
of the determinant and most of its properties where the entries of a matrix come from a ring (see Appendix B).
also apply
We
is
necessary for
<7
We
<r
by
...
.
.
2
.
n\
.
h H
or
<T
3i32
3n,
where
. . .
3i
.... =
cr(i)
JnJ
Observe that since o is onetoone 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 oorGSn. 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 =
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)
i>
k
a,
but
precedes
<t,
A;
in
er
We then
written sgn
r
by
is IS
sgna
if
<7
even
J.
[1
171
if
odd
172
DETERMINANTS
Example
8.3:
[CHAP. 8
35142
in S5.
1;
and
5
5 precede
hence
2;
(3, 1)
and
(5, 1)
satisfy
(4, 2)
(*).
hence
and
satisfy
(*).
5 precedes
and
is
greater than
4;
(*),
hence
a
is
(5, 4) satisfies
1.
Example Example
8.4:
The
identity permutation
= 12
odd.
is
(*).
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
j,
r(j)
i
=
j,
i,
T(fc)
k,
k^
i,
We
call T
a transposition.
T
If
<
then
...
= 12 ...{il)}{i+l)
l
ijl)i{j+l) ...n
There are
2{j
i l) +
pairs satisfying
(h^)!
i^t
i),
(*):
(j,i),
where x
i+1,
.,jl
t is odd.
DETERMINANT Let A be
(Oij)
field
K:
Oln^
0'2n
(tl2 0.22
...
.
021
\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:
A=
. .
(Odi),
the following
sum which
\A\
is
summed over
o)a'UiO,2j^
all
A,
. .
in Sn.
2/ (sgn
aj
That
is,
2
of the wsquare matrix
ail
021
(sgn
C7)aicr(l) tt2(T(2)
Onain)
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:
is
(an)
even.)
is
the scalar
on
itself:
\A\
On.
Example
8.8:
the permutation 12
is
is
odd.
Hence
"H
01^
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
021
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(663)
2
(ii)
3(556)
4(4548)
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,
determinant.
The determinant
of a matrix
and
its
\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\
is
Thus
is
in particular,
where
affected
by the "elementary"
Theorem
8.3:
Let
(i)
fc;
then
B
feA.
(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.
Let
(i)
(ii)
A be any nsquare 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:
That
\A\ \B\
above two theorems using the theory of elementary matrices page 56) and the following lemma.
We
(see
Lemma
8.6:
Let
be an elementary matrix.
\E A\
\E\\A\.
We
two theorems
directly without
A = iiy^^m
Note that the "signs" (1)*+^' accompanying the minors form a chessboard pattern with +'s on the main diagonal:
; : ;
+
::\
We emphasize that My
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
(1824)
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:
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)
(iii)
add
li?2 to R^.
By Theorem
tions; that
is.
4
3
2
3 2
1 1
\A\
5
12 73 9
4
2
1
2
3
121
Now
Thus
1
if
2
0.
we expand by
we may
5
_2
3
1
\A\
2
(1)2
1
2
3
123
3
1
12
2
1
12
= }
38
2
1
3 2
(2)
3
2
1
5
3
CLASSICAL ADJOINT
Consider an nsquare matrix
A=
(an)
'
over a
ai2
ffl22
field
K:
ttin
tt2re
an
0,21
A =
1
fflnl
ffln2
176
DETERMINANTS
Oij
[CHAP. 8
of A, denoted by adj A,
is
called
'An
adj
A21
...
Ai
'^^^
^^^
^"^
^nn
instead of simply "adjoint" because the term adjoint will be used Chapter 13 for an entirely different concept.
Example
8.13:
Let
The cofactors
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
adj
=
\
144 58/
=
Theorem
8.8:
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:
\A\
46.
We
/l
have
0^
1
/2
A (adj A) =
4 \l 1
/46
0\
046
\
= 46
46/
\0
1^
= 46/ =
A/
We
also have,
by Theorem
9/23
11/46 5/23 \
Ai = r4T(adjA)
2/46
4/46
14/46
5/46
4/46 8/46/
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:
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:
if
and only
if
Example
<
3a;
Zy
52/
= =
1
1
A
Since
2
3
3
5
10
19
^^ 0,
We
2
3
is
also
have
19
A.
7
1
3
5
7
1
38,
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.
we
obtain
Theorem
8.11:
Suppose
and
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
Then
T)
(ii)
is
invertible if
and only
if
det
(7)^0.
178
DETERMINANTS
[CHAP. 8
det (Iv)
where Iv
is
is
invertible.
Example
8.16:
Let
(2x
4y + z, X 2y + 3z,
'2
5x
+ y z)
l\
3
.
The matrix
of
R3
is
[T]
4 2
Then
,5
11/
+
1(1
det(r)
2(2
 3) +
4(l
15)
+ 10) = 55
nsquare 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)
The following
Definition:
definitions apply.
A
(i)
function
D.cA
^
is:
is
said to be multilinear if
it is
components; that
if
row Ai = B + C, then
D{A)
= D(...,B + C,...) = =
D{...,B,
...)
D(...,C,
...);
(ii)
if
= kD{...,B,
...).
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
result;
Theorem
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.
a
(i)
a
a
(ii)
^4
b
62.
2
5
= 35
(2) 4
a
23.
(ii)
a
a
= (ab)(a+b) a'a
8.2.
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.
of each matrix: l\
/
,
3\ 3
(ii)
/2
2 3 2
1^
0\
,
I
[42
^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(215)
2
(ii)
 2(46) + 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)
23
5
2(109)
l(9
+ 2) = 5
1 3
(iv)
10 24 13
3
1(6
+ 4) =
10
8.4.
\a2
\a3
&2 bs
C2
cs
Show
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
asftgCi
bgC^ai
Cgagfei
is
precisely the
sum
of the above
two expressions:
61
Ci
a2
62
&3
"2
"3
\A\
3.
8.5.
l\
2
(ii)
/a
c>
(iii)
2 
102
\2
3
3/
0:
A
(i)
3
7/
01
2 7
1
3(4
(3)
3
3)
21
43
(ii)
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
24 + 2(3)
2 +
3 3
2(1)
10
2
2
+ 2(2)
1
1
fi
8.6.
1
A =
6
row by
4.
Then
3
1
362
64A
24A
24 14 1
6
6 + 4(3)
2
65
14
7
10
=
7/6.
= +
28,
and
\A\
28/24
14
8.7.
A
A
Note that a
(where
i2j
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
1 1
2 13
2
13
17
3
+2
+ 6(2)
1
= 4
17
8.8.
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
32
2
5
8
5
2
2
3
1
5 3
8
5 + 2(3) 2
3 5 2 5 2 2
4 7
23573
+ 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
43(3) 33(1)
2
2 3
1
3
8
2
2
+ 2(2)
+ 2(2) 33(2)
83(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)
16
4
3
1
6
3(12
113
/t
8.9.
+ 6) = 54
+
5 6
1
1 1
t
A =
tB
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\
t2 t2
1
t
4
to get
Now
factor
+2
from the
first
column and
t 2
A
(t
+ 2)(t2)
10 11
1
1 1
Finally subtract the first column from the third column to obtain
\A\
(<
+ 2)(t2)
10 11
1
t
= +
4
(t+2)(t2)(t +
4)
182
DETERMINANTS
[CHAP. 8
COFACTORS
8.10.
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\
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
= 120 +
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
 +\ + \+  +/
(iii)
A '(adj A) =
A/
(iv)
Ai
]4i(adjA)
8.12.
d
A.
Find adj A.
adj
(ii)
Show
^1^1 j
=

(X)
A =
(^_^
b
a
\^_^
;
/ +la
(^_,
(ii)
adj (adj A)
d
adj
lciy
+di;
^ /a
v''
e
\^__6
= CD
CHAP.
8]
DETERMINANTS
183
+ y = 3x 5y
2a;
__
(ii)
4
13,
A:,
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.
3a;
3z
+ 2x = + 22; 1 =
z
S
a;
+1  5y
2i/ in columns:
2x 3x
+ 3y + 5y + 2y 
= 1  8 = 1
of the matrix
of coefficients:
31
5
A
Since A
the
2(15
+ 4)  3(92)  l(65)
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
131
A,
11
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)
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
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
=
Since the transposition t
is
a) OiTirCl) "2to(2)
"nraCn)
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\
all
all
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,
is zero.
Now
suppose
ij
ti
=
1
but
12
iz  2.
Then 2 <
ig
and
so
a^ =
0;
hence
0.
Thus each
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,
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:
",
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)
/.
By
\I\
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\
1A
I^g] 1A
proved.
8.20.
B = EnEni
E2E1A where
if
the E, are
\B\
\En\ \Er,i\
(ii)
\B\
^
=
if
and only
fey
\A\
^ 0.
(i)
By
J7iA
Bi 1A.
Hence
induction,
\E\\E_,...E2E,A\
\E^\\E,_,\...\E2\\E,\\A\
\B\ =
(ii)
By
for each
i.
Hence
if
and only
if
\A\  0.
8.21.
Prove Theorem
(i)
8.4:
Let
A be an wsquare matrix.
is
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
.,
Each term in \A\ contains one and we can write \A in the form
\
(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)+iMI
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
186
DETERMINANTS
First
[CHAP. 8
we
n, j
n.
a)
in \A\ containing a is
nn
2 a
(sgn
ffli<r(i)
<nl,cr(nl)
where we sum over all permutations aSS for which ain) = n. However, this is equivalent (Prob.,nl}. 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
let
B
.
(bn,
A
4
by replacing the
ith
row
of
B
biiAn
i,
bt2Aa
&iAi
j =
ajiAn
+ +
(lizAti
+
4
ttjnAin
and
Let
aijAii
023^2!
aiAi
B=
(6y).
By
ftiiBa
6i2^i2
6iBi
j
Since
By
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
+
"
Ojn^ln
= =
0.
Using
A*
\A\,
we
Oti^ii
2j^2i
niAni
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 ijentry in
by multiplying
(1)
and
(2):
6y
a^Aji
Uj^Ajz
ttin'Ajn
Thus by Theorem
8.7
\
\A\
1.
if
^i
In other words,
Accordingly,
A (adj A)
\A]
I.
is
(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
Ax = b = A #
if
and only
if
is invertible,
and
is
Now
obtain
suppose
A#0. By
Problem
X
8.25,
Ai =
(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.
6.
8.27.
Suppose
is
I.
invertible.
Show
/
that Pi]
Pi.
P^P =
Hence
pip
pi p,
and so
Pi
Pi.
8.28.
Prove Theorem
8.11:
Suppose
and
Since A and B are similar, there exists an invertible matrix P such that by the preceding problem, P = PiAP = Pi \A\ \P\ = \A\ lPi \P\  \A\
We
and
\A\
remark that although the matrices Pi and A may not commute, their determinants Pi 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
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).
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,
where
n}.
If
^(%. %,
Accordingly, the sum in we finally have that
(2)
'\) =
all
permutations
i^i^
J.
Using
(1),
D{A)
= =
D{ei^,
Bj^,
ejj
a)
ai^,
where a
is
i^i2
...in
Hence
is
proved.
PERMUTATIONS
8.30.
542163.
We need
to obtain the
(5,
number
of pairs
(i, j)
for which
i>
1,
and
precedes
in
a.
There
are:
3 numbers
2 numbers
3
1
(5
(5,
and
4)
2,
3,
numbers
4 and
number
(5)
+ 2+3 +
is
sgn a
1.
Method
2.
Transpose
^42163
Transpose 2 to the second position:
to
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)
24513 and t
=
3
""(^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/
ffi
41523.
8.32.
<7
jiji
jn.
Show
i
(i,
k)
such that
i>
there
is
and
precedes kin a
a pair
(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[{XiXj).
'"^^
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
8.34.
Let
(T
be an arbitrary permutation.
<T{g)
define
= Yl (^^c" 'i<3
^(tw)
Show
if it
o
is IS
even
[g
Accordingly, aig)
Since a
is
odd
(sgn
<j)g.
to"
(a'.ra)
a'.ru))
.11. KjorOj
.
,{xiXj)
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 (ajofj,  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.
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.
J1J2
jn.
(71
=
=
aj^2
ttj^n
=
a.
aik^chk^
where
Since
e
<7~^
kn
c,
is
<ri
=
is
sgn
JiJz
in
a permutation,
ff(fci)
aj^i Oj^a
a,j^n
"iki "zkz
nk
Then
fej,
k^,
..,k
=
.
1,
^(fea)
2,
.,
"(K)
=n
Let T
^1^2
kn
Then for
1,
. ,
w,
(<jT)(i)
a{T(i))
a(fcj)
Thus aT
e,
ct~i.
CHAP.
8]
DETERMINANTS
191
MISCELLANEOUS PROBLEMS
8.37.
T:
T
T
is
{2x
(ii)
is
defined
by
T(A)
MA
where
M
c
d,
1
2
1
01'
(i)
to,
[r]
24
3
1/
Then
det (T)
01
2(212)
124
33
1
 l(36) =
11
(ii)
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.
A =
of the
1 .0
1
1,
The inverse of
Set
is
form (Problem
8.53):
A =
*
z
'
AA'^
l,
Ai4i
0,
2/tzl
0,
192
DETERMINANTS
[CHAP. 8
/ll
The
solution of the system
is
a;
0\
1, y
0,
1.
Hence
1
1/
\0
A~^
A> =
(adj i4.)/A
8.39.
Let
generally,
show that
is
if
is
=  D{B,
A).
More
D{...,A, ...,B,
that
is,
= D{...,B,
0.
...,A, ...)
the sign
Since
is
alternating,
is
multilinear,
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,...)
+ =
and thus
D(...,B,...,A,
...)
...)
+ D{...,B....,B,
...)
...)
D(...,A, ...,B,
D(...,B, ...,A,
A,...).
D(...,A
B,...)
= D(...,B
8.40.
Let
over R.
if
(i)
Determine
whether or not
(ii)
D.V^R
is 2linear
D{M) = a + d,
D{M) =
No.
ad.
(i)
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
D(sA
+ tB,C) =
=
i?/i
s(aiC2)
+ +
**^
sa,+
tb,^^
^sa,
th,)c,
t(6iC2)
8D(A, C)
tD{B, C)
That
is,
is
Furthermore,
D(C,A) = d(^;
I'J
=
R,
tB)
c,a,
and
D(C, B)
= D^^^
^=
c,b.
s, t
D(C,sA +
d(
s(cia2)
"J.,
?..)=
=
sI>(C,A)
Ci(sa2+t62)
=
That
is,
t(ci62)
+ D(C,B)
is
is 2linear.
CHAP.
8]
DETERMINANTS
193
Supplementary Problems
COMPUTATION OF DETERMINANTS
8.41.
2
(i)
5
(ii)
6 3
2
....
8.42.
of each matrix:
t2
(i)
3
/t5
1
(
tl)'
<">
8.43.
is zero.
8.44.
of each matrix:
l\
/3
,
2
4/
(ii)
2 4\ 5 1
6
/2 1
,
(iii)
\1
3
\0
1/
8.45.
3 6
1
t
(ii)
I
+
6
3 3
t4,1
for which the determinant
8.46.
/l
8.47.
1020
31
2
3
^"^
I
2\
3
12
1
4
(i)
121'
2)
\4 3
l2
2 1
1/
22
3^
I,
4
,1
(i)
2 3^
5,
(iii)
the entry
4,
(ii)
the entry
the entry
7.
8.49.
Let
Find
(i)
adj A,
(ii)
Ai.
'1
8.50.
2
1
2'
Let
A =
Find
1,
(i)
adj A,
(ii)
Ai.
a
8.51.
Find the
matrix in Problem
8.47.
8.52.
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.
that adj
that
is
is triangular.
6; = 0;
hence
is
invertible
iff all
^ 0.
...
(iii)
and
(if
form
di2
ti
...
/&r*
I
dm
d2n
A1
0
lO
ar^
...
5_,
^
\o
62
a
'
Let
(3a;
 2z,
5y
+ 7z,x + y + z)
Find det
8.55.
(T).
Let
DiV^V
(i)
i.e.
(iii)
D{v)
{sin
=
t,
dv/dt.
if
is
erated by
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
(ii)
det (Ti)
det(r)i
if
is
invertible.
Solve by determinants:
,..
(1)
+ [ix
(Sx
,
5y
2y
= =
8
,
,..,
(")
f2xSy = 1 I4x + 7y = 1
.
,
8.59.
Solve by determinants:
(i)
= = [sxiyGz = (2x5y + x + 2y 2z
7 3
(ii)
Az
<
+3 =  Sz = [sy + z =
(2z
2y
2
+ Sx + 1.  2x
if
8.60.
Prove Theorem
8.10:
Ax =
and only
if
A =
IA == 0.
PERMUTATIONS
8.61.
in Sg:
(i)
=
(i)
3 2 1 5 4,
(ii)
1 3 5
4,
(iii)
4253
1.
8.62.
<r,
and v
in
Problem
8.61, find
rc,
(ii) Tr<r,
(iii)
cri,
(iv)
ti.
8.63.
5.
Show
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
i.e.
is
row
vectors.
Let
Show
being alternating:
i?(Ai,A2, ...,A)
=
A^.Az,
whenever Aj
(ii)
i.
Show
that
if
,A^
CHAP.
8]
DETERMINANTS
195
8.66.
Let
j
(i)
over R.
D:V>^K
(iii)
D{M) = achd,
D{M)
D{M) =
D{M) =
= ah ed,
1.
8.67.
Let V be the space of Msquare 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
where Aj
is
the tth
row
A and !>(/) = 1.
of
is
so
A^B
row
of
Z> is
det (A)
and so
8.5, i.e.
Ai5
\A\ B.)
MISCELLANEOUS PROBLEMS
8.68.
Let
be an wsquare 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.
where
and
=
.
\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
Consider
the
2square
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
let
be the matrix
8.74.
Let A be an Msquare 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)(t3)(t4),
2;
(ii)
(li)
(t
+ 2)2(t4),
2.
(iii)
(t
+ 2)2(f4).
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/
Ai
(adj A)/A1
2
i ^ \l 1

0,
1
8.50.
2\
6
,
adj
A =
3 1
2
1
Ai
=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
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,
8.61.
agn a
1,
agn t
=
(ii)
1, sgn v
1.
8.62.
(i)
Tv =
Yes,
53142,
ir(r
52413,
(iii)
<ri
32154,
(iv)
ti
14253.
8.66.
(i)
(ii)
No,
(iii)
Yes,
(iv)
No.
chapter 9
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
aA"
ttiA
ao/
where
nomial
In particular,
we say
that
is
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
applies.
let
Theorem
9.1:
Then
(i)
(ii)
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,
in the
matrix
commute.
197
198
[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,
is
T .V *V
an eigenvalue of T
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
(Problem
eigenspace of
\.
The terms
characteristic value
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.
:
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 wsquare 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:
We
seek a scalar
X =
=
such that
A = AX = tX:
2 2
DO
or
(:
is
[3x
+ +
2y
2y
= =
{tl)x 2y =
\Zx + (t2)y
(i)
CHAP.
9]
199
and only
if
the de
t1
3
Thus
t is
2
t2
3t
if
= {t4){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
<
4,
is
a multiple of
v.
= l
in (1),
2x 3x
2y
3y
is
= =
or simply
Thus
t
w =
= 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
T{v)
i.e.
XV
or
also
{Xl){v)T{v)
or
{XlT){v)
=
if
Xl
is
singular.
We
have that v
is
in the eigenspace of X if
and only
the above
T.
9.14)
We now
Theorem
by induction:
are
linearly
9.3:
Nonzero
eigenvectors independent.
Consider the functions
If
eigenvalues
Example
9.6:
ei', 6^',
., e''n'
where
aj,
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
D(ei'')
a^e'^''*.
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
[CHAP.
if
and only
[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
Alternate
Form
of
Theorem
9.4:
wsquare 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
if
we
let
n independent
= P~^AP.
A =
/
(
j
.
9.7:
By Example
/2
{^^
9.5,
eigenvectors
/2\
(^^j
and
1\
Set
P =
1\
(^_J.
_^j
and so
_, P i 
/1/5
(^^^^
1/5
_^^^
Then
is
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/
eigen
over a
field
K:
.
. .
(0,11
Chl
ai2
ffl22
ain
fflZn
ttnl
ttre2
flnn
The matrix
tin
A, where
* 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
We
also call
det
(tin
A) =
CHAP.
9]
201
Now
each term in the determinant contains one and only one entry from each row and characteristic polynomial is of the form
{t an){t
+
Accordingly,
AaC*)
an
t"
(au
+ a22+
+aTO)t"~^
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
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
A =
2
4
21
21
3
^(t)
\tIA\
t2
t
=
2
3.
<2
28
4
As
expected, A(t)
is
We now
in linear algebra.
its
CayleyHamilton Theorem
Example
9.5:
Every matrix
is
a zero of
characteristic polynomial.
/I
I
9.9:
The
A =
2
2^
Ji
A(t)
\tIA\
t3
t2
A
is
t2
3t
As
a zero of
A{t):
= (:
:)
Theorem
9.6:
Let
value of
Proof.
is an eigenbe an nsquare 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,
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
202
[CHAP.
{t ai){t 
aa)
(t
 an)
then
is
similar to a diagonal
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
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)
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
PiAP =
\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,
\tIB\
=
=
\tIP'AP\
\PHIP  P'AP\
\P^\\tIA\\P\
P~i P
\PmiA)P\
1,
we
finally obtain
\tIB\
= \tIA\
Theorem
9.9:
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.
is
unique (Problem
we
call it
the
minimum polynomial
Theorem
There
9.10:
A divides every polynomial which has A In particular, m{t) divides the characteristic polynomial A(t) of A.
A(i).
is
CHAP.
9]
203
Theorem
The
characteristic
irreducible factors.
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
if
and only
2
if
is
a root of the
minimum
poljTiomial of A.
1
Example
9.11:
>(*)
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)
(t2)(t5),
= (t2)3(t5) CayleyHamilton theorem that in^iA) A(A) = 0. The reader # but WgCA) = 0. Accordingly, m^it) = {t 2)^ (t 5) is
W2(t)
(t2)2(t5),
m^fy
can
the
Example
9.12:
Let
field R.
We
show that
By
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
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
its
algebraic
204
Example
[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
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
A =
~z\ and
f{t)
fM +
1.
9.2.
Show
that
^ =
=
^^
^ ^^^^ ^^
f(^)
*^
4t
5.
''
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).
mapped
SV
is also
mapped
Thus
fm =
0.
This result
is
/(<) is
CHAP.
9]
205
9.4.
Let
that /(A)
is
the matrix
into its
We
a^t
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)
Now assume the theorem holds for polynomials of degree less than n. algebra isomorphism,
^if(T))
Then
since ^ is
an
= = =
0(ttr"
a_irni
a^')
a0(r) 0(ri)
0(a_ir"i
+ +
aiT
a<,/)
a^')
/(A)
oAAi + (a_jA"i +
aiA
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
(a<,
6o)^
aA
ajA
tto^
60/
/(A)
flr(A)
m
c^t''
(ii)
By
definition,
Co
n+m
aj6fc_i.
fc
2 =
where
c^
eio^fc
"'i^fci
Ojcfro
1=0
Hence
(/ff)(A)
= 2
=
fcoo.
CfcA^
and
n+
/n
/(A)f,(A)
=
kf
2 \i=o
N/
+ +
nm
m
CfcA"
(/ff)(A)
(iii)
By
definition,
(fe/)(A)
A;at"
fcajt
fcaA
fettiA
fcoo/
A;(aA
+ ajA + o,/) =
k /(A)
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.
& 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
[CHAP.
9.7.
Let
(ii)
A =
4\
g
I
(i)
Find
all
eigenvalues of
that
P such
*/
P^AP
diagonal.
(i)
AoiA:
\0
tj
\2
3J
2
t3
U)
A(t) of
A
1
is its
determinant:
\tIA\
2
4
=
3
4t
{t5){t+l)
The roots of
A(t) are 5
We
First substitute
into
to
4
form the
solution of the
.2
4\/x\ 2)\y)
_ /0\ ~ \o)
f
""^
4x4y =
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
Substitute
= 1 =
into {1)
to obtain the
homogeneous system
2
2
4 4
~2x 2x 
4.y 4i/
= =
or
2j/
2,
3/
1.
Thus
w=
2
(2,
1)
(ii)
Let
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
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.
(i)
A =
I
3 5
/3
(ii)
l'
5 =
7
6
51
6
66
Which matrix can be
(1)
2
diagonalized,
and why?
Form
istic
and compute
its
3 6
1
t
A(t)
\tIA\
+
6
5
f
3 3
=
4
(t
+ 2)2(t4)
The roots of
A(t) are
2 and
4;
CHAP.
9]
207
Substitute
= 2
acteristic
matrix
tl
to obtain the
homogeneous system
fSa;
or
3a;
[6a;
+ + +
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
Substitute
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.
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
P~^AP
t
corresponding to the
columns of P.
+
7
6
1
1 1
t
(ii)
A(t)
\tIB\
t5
6
4.
=
2
(t
2)2(t4)
The eigenvalues
of
are therefore
2 and
We
find
to obtain the
Substitute
= 2
into tl
B
= = =
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
[CHAP.
9.9.
Let
A = A
and
Find
all
eigenvectors of
field C.
(i)
(ii)
the complex
Mt) =
Hence only 2
is
\tIA\
=
Put
t
t1
= A
(2
t1
into tl
4i
(t
 2)2
an eigenvalue.
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)
\tlB\
t2
1
t
=
1
t2
t2
has no solution in 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
values of B.
We
Substitute
in tl
to obtain the
homogeneous system
'i1
2
i+lAy) ~
=
/
""^
{il)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'
(1,11
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
t2
A(t)
1
\tIA\
=
T.
t1
2
=
4
(t2)2(t3)
t
Thus
We
Substitute
into tl
to obtain
/a\
y y
=0
+
z
CHAP.
9]
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
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)
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
Ov
0,
i.e.
that
is
singular.
9.12.
Let
and
and
BA
By Problem
singular,
(iv)
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
BAw  BABv =
Hence X
of
is
B\v = \Bv = \w
an eigenvalue of BA.
Similarly,
BA
is also
an eigenvalue
AB.
Thus
AB
and
BA
9.13.
Suppose
of
A.
is
is
an eigenvalue
TK
Since
is invertible, it is also
# 0.
By
definition of
an eigenvalue, there exists a nonzero vector i; for which T(v) Xv. Applyv = Ti(\v) = xri(i;). Hence ri(v) = Xiv; 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.
.
.
. ,
is
by induction on Suppose
n.
Ii
I
1,
then Vj
is
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
[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
 Xjiini =
Xj
Since the
are distinct,
ai;
X 9^
for
# w.
Thus
Hence
a^i
we
get
0,
and hence a
0.
a22
...
CUnj
Find
its characteristic
its
is
eigenvalues.
t
Since
is
triangular and tl
diagonal, tl
A
t
/*!!
tl
<12
0.22
a In
"in
 A
^
\
tanni
t :
Then
A(t)
t/
A
is
=
.
(t
.
are an,
tt22,
. ,
its
diagonal elements
1
9.16.
3\
3
.
Let
A =
Is
If so, find
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.
find a polsoiomial
43
<'")
^ =
\\I
its
Therefore
(i)
the CayleyHamilton theorem every matrix is a root of we find the characteristic polynomial A(t) in each case.
characteristic polynomial.
A(t)
\tIA\
= =
t2
1
f
B
+
3
+t3
11
(ii)
A()
==
\tIB\
t2
7
t
=
4
(2
+
4
2t
13
t1
(iii)
3
3
A(t)
\tIC\
2
1
(t
 l)(t2 2t5)
t+1
CHAP.
9]
211
9.18.
Every matrix
is
a zero of
its
characteristic
its characteristic
polynomial; say,
\tIA\
= <+
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 resquare matrices over of the classical adjoint (Theorem 8.8),
K which
are independent of
By
{tIA)B{t)
or
\tIA\I
(<"
(i/i4)(B_iti+
+Bi + Bo) =
coefficients of
+ a_iti+
lai
+ aoK
t,
corresponding powers of
1
Bnl =
Bn2~AB_i
ail
Bo
 ABi = ABo =
.
a,/
agl
.,
A, I respectively,
A2B_3AiB_2 = a_2A"2
ABo  A^Bi  a^A
ABo =
Adding the above matrix equations,
o^
= A +
In other words, A(A)
a_iAi
a zero of
aiA
oo/
0.
That
is,
is
its characteristic
polynomial.
9.19.
Show
By
that a matrix
and
\tI
its
(t/A)'
tl*
A* :=
tl
 A*.
A\ 
(t7A)*
\tIA*\.
Hence
Since a matrix and its transpose A and A* have the same char
9.20.
Suppose
/Ai
\
^1
B\
Show
acteristic polynomial of
M
~B
Ai.
Generalize.
/tlA,
tlA/'
as required.
\tIM\
tl
Ai
tl
B  A^
By
212
[CHAP. 9
B
A2
M
where the Aj are square matrices,
is
=
\
D
...
A^l
MINIMUM POLYNOMIAL
9.21.
A =
2
1 1
2
The characteristic polynomial of
is
t2
A(f)
1
t2
t1
t2.
1
t1
2
1
t2
i4
(t3)(t2)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.
(t3)(t2),
flr(t)
(i3)(2)2,
/i()
(t3)(t2)3
We
have
1
/(A)
= (A37)(A2/)
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
02
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),
9.22.
\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
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)
(tX)'.
CHAP.
9]
213
9.23.
Let
M
fe(t)
=
of
(A
I
0\
where
and
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)^
=
h{t)
and
0.
Since g{t)
is
the
minimum polynomial
g(t) g{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.
M; hence
m(t) divides
/(t).
Thus m{t)
the least
common
We
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.
Let
A
are
A,
C and D
(t
2)^,
t2
and
5 t1
respectively.
4:
c The
t2
D =
(5).
of
characteristic polynomial of
is
\tIB\
2
t3
of B.
7t
10
= (t2)(t5)
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 CayleyHamilton 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
[CHAP.
9.26.
(operator)
as a zero.
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
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
= f+
Cjfi

I
c^it
c^.
= / = A + cj Bo = A^ + CjA +
Bo
Bi
Cg/
B^_i
= =
A"!
I
CiA'2
4
c^_i7
Then
Bi
B^
Also,
AB^^i = C^  {Ar+CiAri+
+CriA+CrI)
= c^ =
Set
B{t)
c^I
i(A)
= fi^o + f^Bi +
tBr2
Br1
= (t'Bo+t''^Bi+ +tBri)  (t'^ABo + tr^ABi+ +ABri) = t^Bo+ tri{BiABo)+ t'2(B2ABi)+ f t(B,._i  AB^a)  AB^i + Critl + C^ = f/ Cifl/ + 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 {tIA)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]
215
9.30.
Suppose
dimF =
Let
T:V ^V
he an invertible operator.
Show
that T*
is
equal to a polynomial in
r
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
ri
(Tri
+ ari r'2 +
+ aj)
MISCELLANEOUS PROBLEMS
9.31.
Let T be a linear operator on a vector space V of dimension n. Without using the CayleyHamilton 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.
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
is
/^
X
ii
{
1
21
0sl\
'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,
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
Substitute
into the
216
[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
cannot be diagonalized.
9.34.
Let
F be
an extension of a
field
K.
Let
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
fi(t)hi
61,
. .
+
.
+ +
fnit)b
6 belong to
F
+
/i(A)bi
/2(A)62
=
that, for each pair {i,j),
(1)
/fc(A).
a</'6i
aif 62
a^f &
=
tty'''
Since the
64
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) =
:
T{vi)
0,
an,nivi.
Show
that
T"
It suffices to
show that
Ti{Vj)
= =
is
(*)
for
1,
.,n.
For then
THvj)
it
follows that
Ti(Ti(Vj))
r"5(0)
0,
for ^
T"
j.
0.
We
prove
{*)
follows (for j
2,
The case
true by hypothesis.
Ti{Vj)
= = =
TiHT(Vj))
= TSHajiVi+ +aj^j.iVj^i) +
aj^TiHvi)
ajiO
+aj.,_irJi(Vj_i)
+ +
aj,j_iO
=
is
triangular with
/O
a^i
ttsi
...
..
.
ani
032
a2
...
a,_i
\0
...
CHAP.
9]
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.
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
12
8
0^
9.40.
Let
B =
12
8/
such that
B=
As.
9.41.
M and
a. "2
a triangular matrix N:
\
...
a^
...
M
Show
that, for
"
and
a
AT
02
...
...
any polynomial
'/(tti)
f(t),
...
f^"'^^
//(ai)
f(M)
=
I
^^
and
...
/(AT)
f{aj
9.42.
M and
... ...
/Ai
/Ai
and
B
A,
=
\
A2
N =
...
A,
that, for
\
any polynomial
/(*),
//(Ai)
X
/(A2)
...
...
f{M)
=
I
/(A2)
...
^^^
/(A)/
^(^)
...
...
/(Aj/
More
9.43.
Show
generally,
that for any square matrix (or operator) A, (P^iAP)" = PiA"F where show that f{P^AP) = pif{A)P for any polynomial f(t).
is invertible.
9.44.
Show
that:
(i)
/(Af)
(/(A))*;
(ii)
if
is
symmetric,
i.e.
A*
A,
^ =(1
/22\
3)'
=(3
/42\
3)'
()
^=(13
are diagonal.
/51
218
[CHAP.
9.46.
/3
(i)
l\
2
,
2\
/l
,
1 1
0^
:=
4
1
(ii)
B =
1
(iii)
C =
\0
\1
3/
\l
14/
ly
When
diagonal.
9.47.
Consider
A =
Vl
4y
and
B =
\1Z
3/
Find
all
eigen
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,
z,
2y
+ z,
2y
+ 3);
y,z)
9.51.
For each of the following matrices over the complex independent eigenvectors:
C, find all
<"(:
9.52.
;)
""(J
D
G:r). (;:?;
T.
Suppose v
is
aS + bT where a and
9.53.
Show
that v
is
also
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
/(X) is
an eigenvalue of
f(T).
9.55.
Show Show
same eigenvalues.
Give an example where
9.56.
that matrices
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
:
9.59.
.yV^G K" be linearly independent eigenvectors of Let A be an iisquare 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.
. . . .
.
is
a root:
'2
3
2^
4
1.
^^
^ = (4
D'
(")
^ = (3
3)'
^"^
'^='[1
i_i
CHAP.
9]
219
9.61.
A =
Show
9.62.
that
/()
(t
X)
is
minimum polynomial
of A.
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
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) = tk.
Let
is
k&K
for some
mapping belonging
if
to
&
K.
and only
9.65.
A^ =
k>
n.
0.
9.66.
Show
that a matrix
and
its
minimum
9.67.
Suppose
f(t) is
an irreducible monic polynomial for which f(T) that f(f) is the minimal polynomial of T.
where
Is
a linear operator
9.68.
\^
"/
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
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 msquare 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
[CHAP.
9.72.
a^t
a^.
The following
1 1
do 1 
1
ftnl
Show
9.73.
that
f{t) is
the
minimum polynomial
of A.
Find a matrix
is
(i)
t^
St^
6t
8,
(ii)
t^
51^
2t
7t
4.
DIAGONALIZATION
9.74.
Let
A =
\c
dj
field R.
sufficient conditions
on
a, b, c
and d so that
is
diagonalizable,
i.e.
9.75.
A
if
is
field C.
9.76.
Show
is
diagonalizable
and only
minimal polynomial
is
a product
9.77.
Let
diagonalizable.
K^
9.78.
such that (i) AB = BA and (ii) A and B are both be Jisquare 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
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^
a, 6
and
c.
.0
2j
(/(A))
9.44.
(ii)
Using
Xi
(i),
we have
=
4,
6,
/(A)
/(A).
9.45.
(i)
(ii)
(iii)
Xg
X2
= =
u
1;
= =
/
(
(1, 1).
(1, 1).
Let Pi
2
(
1\
J
)
_^
and P2
1\
1.
CHAP.
9]
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/
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
.
''
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)
(t2)3(t7)2; m(f)
m()
(t2)2(t7).
(ii)
Ht)
(t3).
(iii)
A()
(tX)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
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
exist for
if
and only
is
if
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 ndimensional vector space V. resented by the triangular matrix
Let
(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
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
/.?
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
be linear.
T maps
subspace IF of
i.e.
into itself,
if
vGW
is
implies
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 rinvariant. 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 Tinvariant 1dimensional subspaces. For suppose T{^v) = \v, v 0. Then = {kv, k e K}, the 1dimensional 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.
Theorem
10.2:
Let
T:V^V
is is
be linear, and
be any polynomial.
The notion
of invariance
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 DIRECTSUM DECOMPOSITIONS A vector space V is termed the direct sum of its
if
subspaces Wi,
.,Wr, written
every vector v
GV
= wi + W2+
iVr
with Wi
G Wi
applies.
Theorem
10.4:
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.
Now
is
linear
and
is
the direct
sum
of (nonzero) Tinvariant
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
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)
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.
A generalization
Theorem
10.5:
of the above
and V is the direct sum of Tinvariant 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
A.
is
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
10.6:
Let
m{t)
T:V^V
/i(f)">/2(t)"^... /.(*)"'
where the
direct
fi{T)"'.
fi{f)
sum
are distinct monic irreducible polynomials. of Tinvariant subspaces Wi, .,Wr where Wi
. .
Moreover,
/i(;t)i
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 Tinvariant 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
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^
I.
is
is
a matrix over
(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^ki ^ Clearly the minimum polynomial of a nilpotent operator (matrix) of index k is m{t) f"; hence is its only eigenvalue.
:
we
The fundamental
result
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).
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.
There
is
all
^ mi.
(ii)
m.
Ai.
(iii)
The number
(iv)
by
T.
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()
(f2)4(t3)3
and
is
m{t)
(2)2(t3)2
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 Tinvariant subspace of V called the Tcyclic 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 Tinvariant subspaces of V containing v.
GV
Now
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
ttfci
T'^^v)
...
aiT{v)
aov
ao
(v)
Then
is
m(i)
t"
akit^^
ait
the unique monic polynomial of lowest degree for which mv(T) Tannihilator of v and Z{v, T).
0.
We
applies.
Theorem
10.12:
Let Z(v,
(i)
T),
Then:
The set
fe.
(ii)
m(i).
(iii)
228
CANONICAL FORMS
[CHAP. 10
tto
ai
tti
1 1
aic2 aici
is called
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
is /(*)"
where
f{t) is
Then V
is
the direct
sum
V =
/(*)"', /()"^
Z{vi, T)
Zivr, T)
, fit)"',
V
n =
Ml
^ %2 
 Wr
same number
Any
other decomposition of
above lemma does not say that the vectors vi or the Tcyclic subuniquely determined by T; but it does say that the set of Tannihilators spaces Zivi, T) are are uniquely determined by T. Thus T has a unique matrix representation
Cr
where the
polynomials
/(*)"*.
Using the primary decomposition theorem and the above lemma, we obtain the following fundamental result.
Theorem
10.14:
Let
T:V^V
/{()
where the
are distinct monic irreducible polynomials. unique block diagonal matrix representation
Then T has a
'Cn
Clrj
Cs
/i(t)"
rris
where
mi =
nil
ni2
ni: \'
TCsl
^52
Msr.
CHAP.
10]
CANONICAL FORMS
is
229
The
poly
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{f2t + C(t2t +
3)
(ii)
3) 3)
(iii)
C(2t
+ 3)
C((<2)2)
2)2)
C((t2)2)
C((t2)2)
C((t
C(f2)
C(t2)
is,
where
C(f(t)) is the
companion matrix of
/(t);
that
/O
3
.L__
12
(i)
(ii)
(iii)
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
V/W.
cosets of
Theorem
Let
tion
(i)
Then the
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

{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
Theorem
10.16:
Suppose
is
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
{0},
(ii)
V,
(iii)
kernel of T,
(iv)
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.
eV,
it
is
certainly true if
G Im
T.
10.2.
Suppose {Wi}
Suppose V
is
Show
that
the intersection
Thus tIv)
W=
eW;
then
HiWi is v e Wi
also Tinvariant.
for every i. Since is Tinvariant.
Wj
is
Tinvariant,
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
of f{T)
f(T)(v)
let f{t)
be any poly
Suppose V
i.e.
Ker/(r),
i.e.
We
^(i;)
f(T)(T(v))
0.
Since f{t)t=tf(t),
f(T)T(v)
Tf(T){v)
T(0)
as required.
10.4.
Find
all
invariant subspaces of
A 
(
J
viewed as an operator on
R^.
we have
it
that R^ and {0} are invariant under A. Now if A has any other invariant must be 1dimensional. However, the characteristic polynomial of A is
A(t)
\tIA\
t2
1
t
=
2
t2
Hence
A has no eigenvalues (in R) and so A has no eigenvectors. But the 1dimensional 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
Then T
where
^
J
is
a matrix representa
of
to
^
it
We We
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
+ +
CijWj
C21^1
+ +
+
+ +
+
c^^v^ Cas^j
^^(ys)
6slWl
ftsr^r
CjiVi
Css^s
matrix of
(
coefficients in the
)
above system
Therefore
it
is
the transpose of
is
the matrix 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 fi +
Then
f(T){w)
= = = =
(ar
+ o_ir"i + + +
(ahi)(T(w))
(ari)(r(w))
(a_i rni
fiTHw)
(ii)
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).
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 = aai^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,
. ,
611W11
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
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
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
+
=
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.
Thus
is
linearly independent
10.8.
Suppose
(i)
T:V^V
is
is
linear
directsum decomposition
m{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)
and
A2(t)
By Problem
Wi(()
each of
TOi(t)
Now
0.
suppose
f{t) is
Me
and /(r2)(W)
=
M
Let
vGV;
w =
is
then
/(r)
f(t).
w +
f(T)
w =
/(Ti)
f{t),
/(T2)
=
common
multiple of
That
is,
is
a zero of
and so m{t)
the least
By Theorem
10.5,
T has
a matrix representation
M
=
where
and
are matrix
Then, by Problem
9.66,
A(t)
\tIM\
tl
A
tl
B
\tIA\\tIB\
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 Tinvariant subspaces = Kerh{T).
h{t)
and
where
since g(t)
C7
are Tinvariant 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)
= =
T,
r(T) g(T)
s{T) h(T)
(*)
veV;
first
then by
(*),
r(T) g(T) v
But the
term
in this
sum
belongs to
W KerhCT)
=
is
r(T)f(T)v
the
r(T)(iv
Hence
sum
of
To prove that
V = U W, we
1;.
uniquely determined by
is 0,
we
obtain
r(r)ff(r)i>
r(T)g(T)u
__
r(r)fl'(r)w
0,
Also, applying
(*) to
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
=
(1)
and
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.
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 Tinvariant 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.
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 Tinvariant 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
to
V^ by
T^.
By
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.
(tXi){tX2)
... (tX,.)
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) Xjy. 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
0.
polynomial m(() of
divides
(tXi)(iX2)...(tX,/)
Accordingly,
m,(t) is
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 =
is
linearly independent.
(ii)
The subspace
(iii)
(iv)
.,T{v),v} of
1
..
.
W,
the matrix of
is
of the
form
..
.. ..
/csquare
matrix
is
nilpotent of index k.
Suppose
av
di T{v)
02
T^v)
+ a^.^nHv)
(*)
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
Tinvariant.
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)
{T'<'^v), Ti'^v),
.,T{v),v} of
rk(i;)
W,
T(T^^(v))
= =
r(rfc3(^))
r'=2(u)
T(T{y))
T'^(v)
T(v)
T(v)
in this basis is
1 1
CHAP.
10]
CANONICAL FORMS
Let
235
10.14.
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
XcY cZ.
{Ml,
. .
.
and
Z = Ker T*.
Wi,
. . .
By
the
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
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
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
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
.,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(mfctfc_i,
fc)
=Mj^
236
CANONICAL FORMS
and setting
[CHAP. 10
...,
vim^m^.i, k1)
^ Tv{m^m^^i,k)
By
Si
is
{Ml
...,
u^^_^, v{l,kl),
...,
vCmfcmfc^i, fe1)}
ments
We
new
ele
y(mfem;,_i
l,
fc1),
v(m^mki +
2,
k~l),
...,
v(m^_i m^^^tkV)
Next we
set
v(l,
k2) =
Tv(l, k
v(in^_i
 1), v(2, k2) = Tv{2, k1), ...,  mfc_2. k2) = Tv(m^_i  m^.g,  1)
fc
{Ml,
...,
u^^_^, v{l,k2),
...,
u(mfc_iw^^2. ^^2)}
to a basis of TFfc_2
we can extend
by adjoining
vim^.i
711^2
1,
...,
vim^^^inks, ^Z)
TOfc_i,
A;)
v{l,kl),
v{m^mk_i,kl),
v(mfcmfe_i,
i;(mfcmfc_i,
2), 1),
...,
u(mfc_i
 Wfca,
fe
 1)
..., ...,
i;(l, 2),
..., ...,
..., ...,
u(mfc_i u(mfe_i
 mfc_2,
mfe_2,
2),
1),
^(mami,
i;(m2mi,
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
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
mfc_i
(mfc_i
m;;_2)
(m^
mfc_i) =
2mfc_i
m^
fc
2m2
2mi
mi m^ 7^2
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 = (mfcmfc_i) +
shows that the nullity mj of T
(2mfc_i
is
m^ TOfca) +
number
+ (2m2mim3) + (2mim2)
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
in canonical
form
which
similar to A.
CHAP.
10]
CANONICAL FORMS
Since
237
is
2.
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,
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)rr
(i
Set Ni
= Ti Xj7.
Then for
Ti
= Ni+
\I,
where
Nr't
=
iV{,
That
since
is,
Tj is the
is
sum
which
is
of index mj
(t
Xj)*"!
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
T:V >V
whose
(t
2)^(t 5)^.
forms are
has exponent 3 in
twice.
A(t),
must appear
2
Thus the
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
is
Let
(i)
W he a subspace of
(ii)
uGv + W,
Suppose
uv GW,
vGW.
Gu
uG v + W.
u
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
That
is:
W and
if
+W
(ii)
gV
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
u+W
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.
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
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
ij
F  V/W,
defined
+ W,
is linear.
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
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.
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
uvGW
T{u+W) =
is linear.
and, since
T{u)
W
=
u+W = v + W
is
then
Tinvariant,
T(u v)
T(v)
T(v
+ W)
as required.
We
t
(v
We
f(u
have
+W) +
=
W))
= =
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"
n.
/()
= =
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).
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
If
dim
V=
1,
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 1dimensional 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
^
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,
Ojsys
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
annyn
triangular.
Prove Theorem
Z{v,T), and
(i)
10.12:
m()
t^
Let Z{v, T) be a Tcyclic subspace, T^ the restriction of + Oo the Tannihilator of v. Then: + 0.^1*"' +
dimZ(t;,r)
to
The
k.
(ii)
of Tv
is TO(f).
(iii)
CHAP.
10]
CANONICAL FORMS
241
fto
ai
afc2 Ctlcl
.
(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(ri(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)
and so m^(T^)
0.
Then
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 Tinvariant subspace of V and T the induced operator on VIW. Prove: (i) The Tannihilator of v G V divides the minimal polynomial of T. (ii) The Tannihilator of v G VIW divides the minimal polynomial of T.
:
(i)
therefore,
(ii)
GV
it
is
to Z(v, T)
and
polynomial of
Remark. In case the minimal polynomial of T is /(t)" where G V and the Tannihilator of where m n.
nomial, then the Tannihilator 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 Tcyclic subspaces Zi = Z{vi, T), i = l, ., r, with corresponding Tannihilators
Lemma
10.13:
/(f)"i,
/{f)%
. ,
/(i)"',
n =
ni
^ 2 
 r
same
and
Any
other decomposition of
sum
If
number
the
same
set of Tannihilators.
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 Tcyclic
the
lemma
242
CANONICAL FORMS
Since the minimal polynomial of hence the rannihilator 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 Tinvariant. 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
V =
Z(%, r)
Z(v f)
ti
f{t)"2,
.,
/(*)">,
n2
n^
We
of V2.
Let
is a vector V2 in the coset V2 whose Tannihilator is /(<)"2, the Tannihilator 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
we have by
=
But /()" is the rannihilator 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.
of V2 is a
On
(1),
f(T)'H(wh(T)(vi))
is /(t)"2
. .
f(T)^(w)
g{T){vi)
as claimed.
v^,
.,Vf&V
the fannihilator of ^.
We
Z2
set
Z{V2,T),
...,
Z,
Z(i;r)
/(t)
Then
we know
that
T*"i 1 (Vi)}
and
{%
i
f(v^,
f d^ii 1 (iTj)}
But
,,
2,...,r.
.s^
.
V=
Z(v^
f)
is
fi(v)
THv)
(see
(i;^)}
Problem
10.27),
...,V
..., Tdrl
is
a basis for V.
It
Thus by Theorem
10.4,
V=
Wj,
.
Z(vi, T)
. .
Z(v^, T),
as required.
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(Mjs) if i > s and dimension
is
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
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 n1. 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
(i)
C(t2
C(2 C(t2
is,
(ii)
(iii)
+ 2) + 2)
+ 2)
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.
(i)
Since the
defined.
V, u
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
E(v
and therefore
(ii)
+ u) = Wk +
wl^
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)
Let V
SV. We
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
=u
That
is,
is
the projection of
10.35.
Suppose
V=
if
UW
E{v)
and suppose
if
T:V^V
t>
is linear.
rinvariant
and only
TE  ET where E
for every
is
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
Tinvariant.
w GW.
(TE){w)
Since E{w)
E{T(w))
(ET){w)
T{E{w))
and so
T(w)
W
where
Hence
is also
Tinvariant.
Conversely, suppose
and
W are both
G
i7
Tinvariant.
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
that
is
10.37.
10.38.
Show
0,
S:V^V
and
V ^
F.
Show
that
is
also invariant
under
10.39.
Let
r y
:
>
be linear and
let
T.
Show
that
W
10.40.
is
rinvariant.
Let
linear operator on
be a vector space of odd dimension (greater than V has an invariant subspace other than
Show
that any
10.41.
A =
_A
(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 DIRECTSUMS
10 43
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)
that V = W^ + i,...,Wr) =
Show Show
Wi
{0},
fe
Wr
=
l,
if
and only
(i)
V=
L{Wi)
and
W^nL{Wi,
.,Wki.
...,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
Show that
/((t)! is
Wr
/i(f)"i /2(f)2
restriction of
to PFj.
NILPOTENT OPERATORS
10.47.
Suppose
S and T
i.e.
ST =
TS.
Show
that
S+T
and
ST
Suppose
that
is
a supertriangular matrix,
i.e. all
entries on
0.
Show
is nilpotent.
10.49.
Let
is
n.
Show
10.50.
Show
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.
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)
= = =
(t2)4(f3)2,
(t7)5, (t2)7,
m(t) m(t)
m{t)
(t2)2(t3)2
= =
(t7)2 (t2)3
m(t) =: (3)2(i5)2
is
(iv)
A(t)
(t3)*(t5)\
10.53.
Show Show
{Hint.
Problem
10.54. 10.55.
that
complex matrices
of order
for which
A" =
are similar.
Suppose
is
is
CYCLIC SUBSPACES
10.56.
Suppose
T:V *V
v.
is linear.
is
containing
10.57.
Let fit) and g{t) be the Tannihilators of u and v respectively. atively prime, then f(t)g(t) is the Tannihilator of u + v.
Show
that if
f(t)
and
g{,t)
are rel
10.58.
Prove that
Z(m, T) rannihilator 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 Tannihilator 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)
(ii) (iii)
= = =
(t^
(t
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.
10.63.
T :V
^
is
a product of
its
elementary
10.64. 10.65.
3X3
minimum and
Show that
f(t) is
the char
PROJECTIONS
10.66.
Suppose
i^j;
10.67.
(ii)
V = Wi I = E^+
.,Er
0, i
Wr +E^.
(iii)
into Wi
Prove:
(i)
EiE^
0,
Let El,
(ii)
..
be linear operators on
such that:
that
(i)
EiEj
^ i;
Bj
+E^. Prove
i.e.
projections;
10.68.
Suppose
E .V ^V
^
)
is
a projection,
is
E^ = E. Prove
and
that
form
10.69.
where r
the rank of
Prove that any two projections of the same rank are similar.
10.68.)
{Hint.
10.70.
Suppose
(i)
Im (IE);
(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
10.72.
Let IF be a subspace of V.
and that
L(Ui)
nW
{0}.
Show
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.)
+
,
020:2
+
,
an^n
0.
O'i^
K
is
and
let
(5i, 63
6)
X".
IF of IF in K"
linear equation
aiXi
10.75.
a2X2
+
,
a^Xn
where
and
let IF
i._j,j^ 6  ajfti +
.a.
h a6
Let
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
since
u&U
(ii)
WcUcV.
implies
w e V; hence U/W
is
Prove that
10.77.
?7/IF is a subspace of
V/W,
dim {V/W)
dim(i7/IF)
= dim{V/U}.
Let
Show that
the cosets of
UnW
v,v'
in
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_i4I
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
01
'0
1
103
13
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
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.
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 *
GV
is
is
4,{au ^hv)
<j>{v)
into K.
...,a)
Let
is
jtj
K"
linear
and so
it is
7rj(ai, aj,
a^
Then
ttj
Example
11.2:
Let
over R.
Let
^ iV ^ R
is
be the integral
j
o
p{t) dt.
Recall that
linear;
and hence
it is
Example
11.3:
Let
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,
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)
and k G K.
This space
is
Let V = K", the vector space of tituples 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
[CHAP.
11
a^Xi
02*2
In^n
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
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,
Example Hii:
{v^
(2,1), Uj
(3,1)}.
We
^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}.
{4>i(x, y)
= x + Sy,
y)
and their
{<l>^,
duals.
Theorem
11.2:
Let
{vi,
v} be a basis of
V and let
4,^{u)v^
Then
uGV,
+ +
u =
4,^{u)v^
<t>Su)v^
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
Then
matrix from
to
(o.j.
CHAP.
11]
251
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
GV
GV**.
First of
all,
^
vi<j>)
=
^
<l>iv)
It
any
^,
o
V*,
(acf>
is linear.
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**.
applies.
Theorem
11.4:
V has finite
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 finitedimensional. However, it is always
it is
always onetoone.
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
a subspace of V*.
Clearly
a,b
gK
W and
^,
<r
i.e.
if
<j>{W)
W,
G +
W^.
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*.
a subspace of F,
we have
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
homogeneous
ai2X2
ainXn
(*)
amlXi
+ am2X2 +
UmnXn
252
[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:
space of A)
= n  rank (A)
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
it
by T' and
the transpose of T.
In other
r'(0)
4,oT
Thus
{T\4,)){v)
&V.
T' defined above is linear.
<^,
Theorem
11.6:
f/*,
+ ba) = =
T
is
a{ci>oT)
b(aoT)
That
is,
We
emphasize that
into V*:
if
mapping
from U*
T* no doubt derives
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]
253
Solved Problems
Let
R2
*
and
(i)
(t{x,
y)
R and a  Sx y.
J/)
R^
*
Find
(i)
+ 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
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
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 = 362263 =
62
363
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
= = 3c22c3 =
3C3
C2
C3
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:FR and ^2 ^"*'' be defined by
=
1,
i.e.
V =
<^i(/(i))
= S^'mdt
and
Um)
==
S^'fi*)^^
<j>^
of
Find the
+ bt
and
= c + dt. By =
1,
02('yi)
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
[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,
We
first
show that
.,<i6}
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
{0i,
is linearly
independent.
Suppose
ai0i
a202
a0
to v^,
we
obtain
= = =
Similarly, for
i
0(vj)
{aj0i
ai0i(l'i)
+
a2
ttz
02(^1)
2,
,
(Ui)
=
=
That
is,
=
+
(ai0i ^
+ a^)(Vi)
i S6i(^i)
al 0l('yt)
S^nC^'i)
"i
0,
.,a.
0.
Hence
{0i, ...,</>n} is
11.5.
v} be a basis of Prove Theorem 11.2: Let {vi, vector uGV, basis of V*. Then, for any
. . .
and
let
{<j>^,
..
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]
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 =
we
obtain
(1).
Next we prove
= = =
+ ^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
+ a2i'2 + a  b^i,pi + 6202 + + av + Q = (6y). We seek to prove that Q = (Pi). 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
= =
(6(101
6jiaji
+ 6i202 + + 6j2aj2 +
Thus
iJ,C2
where Sy
is
/KiCi
...
QPt
^2^1
\RnCl
R2C2
KC\ R2C
^rflnl
!"'
"1=7
Rvp2
as claimed.
and hence
Q=
(P)i
(Pi)
11.7.
Show
that
if
GV, v'0,
6.1,
.
.
We
mapping
>
^n) of V
By Theorem
and
^(i;^)
0, i
2,
. ,
n.
property.
11.8.
11.4:
If
onto V**.
We
a,b
first
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
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
= 0?+
6w.
Thus the
map
1^ 1;
is
linear.
256
[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
of L(S),
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
4cd
= =
The system
Set
tional
c
of equations in
unknowns
a, h, c,
is in echelon
= 1, d =
= 0,
2/.
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.
= 6, 6 = 1, c = 0, d = 1
&x
The
11.11.
Show
(i)
that:
(i)
of V,
S^S"";
(ii)
if
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
(!;)
for every
Sg.
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
it
n.
We
W'>
.
= nr. We
.
choose
v_r}
{wi,
.,w
Vi,
.,
\01i
>
0r> "it
>
<'tir/
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
W.
I
<f(''l)<'l
+ aiv^rWr
it is
ffnr)
a basis of W^.
Accordingly,
dim
W^"
= m
as required.
Suppose dimV^^w and diraW = r. Then dim F* by (i), dim TVO" = n (nr) = r; therefore dim
=m
and, by
(i),
WC
TFOO.
Accordingly,
W = WO".
W = dim W".
By
CHAP.
11]
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
= 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 y.
ve^U+W,
0.
Thus
is
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
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) = ~8xly. ,f,(2x
11.15.
let
T*:U*^ V*
r'(0)
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
T)";
^(Im T)
{0}.
&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*
11.16.
Suppose rank(r)
finite
T:V ^ U
is
linear.
Prove:
Suppose dim
and dim
dim ((Im
By
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 ~
(mr)
rank
(T)
11.17.
Prove Theorem
of
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)
= =
aiiUi
a2iUi
+ a^^ui + + a22M2 +
+ ain^n + a2nU
.^.
258
[CHAP.
11
We
want
to
prove that
r'(o'2)
1201
+ +
<*2202
+
+
ttm2'^m
{2)
r'(of)
ain't'l
'''2ti02
0'mn</>m
where
{(tJ
and
{mJ and
{vj} respectively.
Let
T(v)
+ fejVa + + fcv. Then, by (1), = + km TivJ r(i;i) + ^2 r(i'2) + + Cmn'^'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
0j
(feiaii
fc2(l2i
1
fcmmi)t
)
(^)
fciOij
k^amj
On
+ 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).
proved.
11.18.
Let
mxn
field
K.
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
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)
by
^(x, y,z)
= 2xBy + z
and
<!>
<',
3^,
(iii)
20
 5<r.
15
11.20.
Let
0(2,1)
and 0(1,2)
10.
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]
259
11.22.
Let
of degree
2.
Let
functionals on
0i,
<f>2
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)
fzd), ^(t)}
which
dual to {^i,
<f>2,
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
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.
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
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.
is
a linear functional
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
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
(1,
1,0)
and
(0, 1, 1).
11.33.
Show
Let
S""
where L{S)
is
11.34.
and
W be
of finite dimension.
{U n W)"
U"
+ W.
11.35.
Suppose
V  U
W. Prove that V* =
W WO.
<l>(x,y)
Let r B3
:
Zx
2y.
^ B2,
find (rK0))(a;, y,
z):
(ii)
(i)
T{x,y,z)
(x
+ y,y + z);
and
T(x,y,z)
(x
+ y + z,2xy).
11.37.
Suppose
Suppose Suppose
rt(0)
S:U^V
T:V *U
r y
:
TiV^W
and
are linear.
= StoTK
=
(Ker
T)".
11.38.
is linear
has
finite
dimension.
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.
Show that
is
the
mapping T
an isomorphism from
Hom (V,
V)
V.)
260
[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
e y* and
let 0},
Prove
= {vGV:
<t>(v)
0},
W" =
{vi=,V:
<t>(v)
<
0}
convex.
11.42.
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)
6x5y + 4z,
y)
(ii)
ex9y + 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)
</>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)
= xy + z)
11.36.
(i)
(Tt(<t>))(x,y,z)=Zx
+ y2z,
(ii)
(r'(0))(a;,2/,z)
= x +
Sj/
+ 3x.
chapter 12
Bilinear,
field
K.
bilinear
form on
is
f.VxV^K
which
satisfies
f{aui
f{u,
(ii)
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>
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:
is,
= u'v =
(h^).
a^bi
is
a^h^
a6
where u
Example
12.3:
(a^)
and v
Then /
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
is
+ 9){u,v) = {kf){u,v) =
In fact,
f{u,v)
g{u,v)
kf{u,v)
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
[CHAP. 12
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
/(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
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
new
basis.
definition.
if
A matrix B
(or:
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.
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
263
for every u,v GV. bilinear form which satisfies condition (ii) is said to be skew symmetric (or: antisymmetric). 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.
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
11
T4
lOj
ro
1
,,
is
equal to ^ rank
(/)).
In particular, the above theorem shows that an alternating bilinear form must have even rank.
f{v,u)
/,
If
is
a matrix representation of
we can
Y'A'^X
its
write
f{X,Y)
= X'AY = {X'AYY =
X*AY
is
transpose.)
Thus
if
/ is
Y'AX
is
symmetric, then /
is
A = A*
or
is
sjTnmetric.
Con
The main
result for
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
[CHAP. 12
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
23
5
Example
12.4:
Let
4
8
,
a symmetric matrix.
It is
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
^3
>
2
7
and then
5
2
1 0! 2 5 7 2
1
Now A
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
We
1
call
1 7^
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))
is called
Now
form
if / is
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
*<^
in variables Xi 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
265
Example
12.5:
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
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
is
if
q{v)
f{v,v)
is
f{v,v)
>
By
is
nonnegative semidefinite
if
/.
(ii)
/ is positive definite
<S is
rank (/)
where
the signature of
12.6:
Example
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
= 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
.
in the
form
iXyn)
is
2
vCi
I
'
~T'
i^g
<jO
+1
Law
of Inertia
266
BILINEAR, QUADRATIC
[CHAP. 12
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
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
is
^X'^AY
defines a
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
.
.
same number
of negative entries.
/.
The
difference
S = PN
is
q{v)
f{v,v)
^
>
for every v
eV, and
is
f{v,v)
is,
U'V =
(Wj).
ZiWi
is
Z2W2
zw
Moreover, /
is
where u
({)
and v
= +
Then / any v # 0,
zii
Z2,Z2+
zz
l^iP
I22P
+ K\^ >
CHAP.
12]
BILINEAR, QUADRATIC
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
is
Then
/3 2
f{u,v)
0\/j/i^
:
= XtAY =
(^i.xa.ajg)
78
\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
Also,
/(X, aFi
+ ftFa)
XtA(aFi
6F2)
= aX^AYi + bX^AY^ =
is
a /(X, Fj)
6 /(X, F2)
Hence /
is linear in
12.3.
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
=20 + 0=
= 23 + = 20 + = 23 +
= 1
0=
1
Thus
A =
=(r;)'
(
ft
(ii)
Set
B=
(6y)
where 6y
611
612
621
622
= = = = =
/(Vj,!;,):
/(^i.i'i)
/(^i.'y2)
f(>'2,'>'i)
/(^2.^2)
= = = =
/((l.l), (2,1))
/((I, 1), (1,1))
= = = =
3 9
Thus
B =
/3
I
9\
ft
's
(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 = 2MiM2
268
BILINEAR, QUADRATIC
[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
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
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 wsquare 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
onetoone 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
is