1,335 33 10MB
Pages 303 Page size 516.739 x 788.924 pts Year 2009
Undergraduate Texts in Mathematics
Serge Lang
Introduction to Linear Algebra Second Edition
•
Springer
Undergraduate Texts In Mathematics Editors
s.
Axler
F. W. Gehring K. A. Ribet
Springer New York Berlin Heidelberg Hong Kong London Milan Paris Tokyo
Springer Books on Elementary Mathematics by Serge Lang
MATH! Encounters with High School Students 1985, ISBN 961291 The Beauty of Doing Mathematics 1985, ISBN 961496 Geometry: A High School Course (with G. Murrow), Second Edition 1988, ISBN 966544 Basic Mathematics 1988, ISBN 967877 A First Course in Calculus, Fifth Edition 1986, ISBN 962018 Calculus of Several Variables, Third Edition 1987, ISBN 964053 Introduction to Linear Algebra, Second Edition 1986, ISBN 962050 Linear Algebra, Third Edition 1987, ISBN 964126 Undergraduate Algebra, Second Edition 1990, ISBN 97279X Undergraduate Analysis, Second Edition 1997, ISBN 948414 Complex Analysis, Fourth Edition 1999, ISBN 985921 Real and Functional Analysis, Third Edition 1993, ISBN 940014
Serge Lang
Introduction to Linear Algebra Second Edition With 66 Illustrations
Springer
Serge Lang Department of Mathematics Yale University New Haven, CT 06520 U.S.A.
Editorial Board S. Axler Department of Mathematics Michigan State University East Lansing, MI 48824 U.S.A.
F. W. Gehring Department of Mathematics University of Michigan Ann Arbor. MI 48019 U.S.A.
K.A. Ribet Department of Mathelnatics University of California at Berkeley Berkeley, CA 947203840 U.S.A.
Mathematics Subjects Classifications (2000): 1501
Library of Congress Cataloging in Publication Data Lang, Serge, 1927Introduction to linear algebra. (Undergraduate texts in mathematics) Includes index. 1. Algebras, Linear. I. Title. II. Series. QA184.L37 1986 512'.5 8514758 Printed on acidfree paper. The first edition of this book was published by AddisonWesley Publishing Company, Inc., in 1970.
© 1970, 1986 by SpringerVerlag New York Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (SpringerVerlag, 175 Fifth Avenue, New York, New York 10010, U.S.A.), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the former are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone.
Printed in the United States of America 987 SpringerVerlag
(ASC/EB)
SPIN 10977149 IS
a part of Springer Science+ Busmess Media
springeronlin e. com
Preface
This book is meant as a short text in linear algebra for a oneterm course. Except for an occasional example or exercise the text is logically independent of calculus, and could be taught early. In practice, I expect it to be used mostly for students who have had two or three terms of calculus. The course could also be given simultaneously with, or immediately after, the first course in calculus. I have included some examples concerning vector spaces of functions, but these could be omitted throughout without impairing the understanding of the rest of the book, for those who wish to concentrate exclusively on euclidean space. Furthermore, the reader who does not like n = n can always assume that n = 1, 2, or 3 and omit other interpretations. However, such a reader should note that using n = n simplifies some formulas, say by making them shorter, and should get used to this as rapidly as possible. Furthermore, since one does want to cover both the case n = 2 and n = 3 at the very least, using n to denote either number avoids very tedious repetitions. The first chapter is designed to serve several purposes. First, and most basically, it establishes the fundamental connection between linear algebra and geometric intuition. There are indeed two aspects (at least) to linear algebra: the formal manipulative aspect of computations with matrices, and the geometric interpretation. I do not wish to prejudice one in favor of the other, and I believe that grounding formal manipulations in geometric contexts gives a very valuable background for those who use linear algebra. Second, this first chapter gives immediately concrete examples, with coordinates, for linear combinations, perpendicularity, and other notions developed later in the book. In addition to the geometric context, discussion of these notions provides examples for
VI
PREFACE
subspaces, and also gives a fundamental interpretation for linear equations. Thus the first chapter gives a quick overview of many topics in the book. The content of the first chapter is also the most fundamental part of what is used in calculus courses concerning functions of several variables, which can do a lot of things without the more general matrices. If students have covered the material of Chapter I in another course, or if the instructor wishes to emphasize matrices right away, then the first chapter can be skipped, or can be used selectively for examples and motivation. After this introductory chapter, we start with linear equations, matrices, and Gauss elimination. This chapter emphasizes computational aspects of linear algebra. Then we deal with vector spaces, linear maps and scalar products, and their relations to matrices. This mixes both the computational and theoretical aspects. Determinants are treated much more briefly than in the first edition, and several proofs are omitted. Students interested in theory can refer to a more complete treatment in theoretical books on linear algebra. I have included a chapter on eigenvalues and eigenvectors. This gives practice for notions studied previously, and leads into material which is used constantly in all parts of mathematics and its applications. I am much indebted to Toby Orloff and Daniel Horn for their useful comments and corrections as they were teaching the course from a preliminary version of this book. I thank Allen Altman and Gimli Khazad for lists of corrections.
Contents
CHAPTER I
1
Vectors . . . . . . . . . . . . . . . . . . . . . §1. §2. §3. §4. §5. §6.
Definition of Points in Space Located Vectors . . . . . . . . Scalar Prod uct . . . . . . . . . The Norm of a Vector. . . . Parametric Lines. . . . . . . . Planes . . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . .. .. ..
1 9 12 15 30 34
CHAPTER II
42
Matrices and Linear Equations §1. §2. §3. §4. §5 §6.
Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . Multiplication of Matrices. . . . . . . . . . . . . . . Homogeneous Linear Equations and Elimination. Row Operations and Gauss Elimination . . . . . . Row Operations and Elementary Matrices . . . . . Linear Combinations . . . . . . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
.. .. .. .. .. ..
43 47 64 70 77 85
CHAPTER III
Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . §1. Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
§2. §3. §4. §5. §6.
Linear Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. Convex Sets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. Linear Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dimension. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. The Rank of a Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
88 88 93 99 104 110 115
Vll1
CONTENTS
CHAPTER IV
Linear Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 123 §1. Mappings • • . • • • . • . . . . . . • . . • . . • §2. Linear Mappings. • . • . • • . • • • • . • . . • §3. The Kernel and Image of a Linear Map. . §4. The Rank and Linear Equations Again. . . §5. The Matrix Associated with a Linear Map. Appendix: Change of Bases . . . . . . . . . . . .
. • . . . .
. • . . . .
• • . . . .
• • . . . .
• • . . . .
. • . . . .
. . . . . .
. . . . . .
. . . • . .
. • . . . .
. . . . . .
. . . . . .
. • . . . .
. . . . . .
. . . . . .
.. .. .. .. .. ..
123 127 136 144 150 154
CHAPTER V
Composition and Inverse Mappings . . . . . . . . . . . . . . . . . . . . . . . 158 §1. Composition of Linear Maps . . . . . . . . . . . . . . . . . . . . . . . . .. §2. Inverses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
158 164
CHAPTER VI
Scalar Products and Orthogonality . . . . . . . . . . . . . . . . . . . . . ..
171
§1. Scalar Products. . . . • . . . . . . . . . • • . . . • • . . • • . . . . • . . . .. §2. Orthogonal Bases . . . . . . • . . . . . . . . • . . . . . . . . . . . . . . . .. §3. Bilinear Maps and Matrices. . . . . . . . . . . . . . . . . . . . . . . . . ..
171 180 190
CHAPTER VII
Determinants
195
§1. Determinants of Order 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . .. §2. 3 x 3 and n x n Determinants . . . . . . . . . . . . . . . . . . . . . . . . . §3. The Rank of a Matrix and Subdeterminants. . . . . . . . . . . . . . . .. §4. Cramer's Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . §5. Inverse of a Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. §6. Determinants as Area and Volume. . . . . . . . . . . . . . . . . . . . . ..
195 200 210 214 217 221
CHAPTER VIII
Eigenvectors and Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . ..
233
§1. Eigenvectors and Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . .. §2. The Characteristic Polynomial . . . . . . . . . . . . . . . . . . . . . . . . . §3. Eigenvalues and Eigenvectors of Symmetric Matrices . . . . . . . . . . . §4. Diagonalization of a Symmetric Linear Map. . . . . . . . . . . . . . . .. Appendix. Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . ..
233 238 250 255 260
Answers to Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 265 Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 291
CHAPTER
Vectors
The concept of a vector is basic for the study of functions of several variables. It provides geometric motivation for everything that follows. Hence the properties of vectors, both algebraic and geometric, will be discussed in full. One significant feature of all the statements and proofs of this part is that they are neither easier nor harder to prove in 3space than they are in 2space.
I, §1. Definition of Points in Space We know that a number can be used to represent a point on a line, once a unit length is selected. A pair of numbers (i.e. a couple of numbers) (x, y) can be used to represent a point in the plane. These can be pictured as follows:
y
    , (x, y) I
I I I
•o
•x
x
(a) Point on a line
(b) Point in a plane
Figure 1
We now observe that a triple of numbers (x, y, z) can be used to represent a point in space, that is 3dimensional space, or 3space. We simply introduce one more axis. Figure 2 illustrates this.
2
VECTORS
[I, §I]
zaXIS
""
""
" "
" " "
" " (x,y,z)
xaXIS
Figure 2
Instead of using x, y, z we could also use (Xl' X 2 , X3). 'The line could be called Ispace, and the plane could be called 2space. Thus we can say that a single number represents a point in Ispace. A couple represents a point in 2space. A triple represents a point in 3space. Although we cannot draw a picture to go further, there is nothing to prevent us from considering a quadruple of numbers.
and decreeing that this is a point in 4space. A quintuple would be a point in 5space, then would come a sextuple, septuple, octuple, .... We let ourselves be carried away and define a point in nspace to be an ntuple of numbers
if n is a posItIve integer. We shall denote such an ntuple by a capital letter X, and try to keep small letters for numbers and capital letters for points. We call the numbers Xl' ... ,x n the coordinates of the point X. For example, in 3space, 2 is the first coordinate of the point (2,3, 4), and 4 is its third coordinate. We denote nspace by Rn. Most of our examples will take place when n == 2 or n == 3. Thus the reader may visualize either of these two cases throughout the book. However, three comments must be made. First, we have to handle n == 2 and n == 3, so that in order to a void a lot of repetitions, it is useful to have a notation which covers both these cases simultaneously, even if we often repeat the formulation of certain results separately for both cases.
[I, §1]
3
DEFINITION OF POINTS IN SPACE
Second, no theorem or formula is simpler by making the assumption that n == 2 or 3. Third, the case n == 4 does occur in physics. Example 1. One classical example of 3space is of course the space we
live in. After we have selected an origin and a coordinate system, we can describe the position of a point (body, particle, etc.) by 3 coordinates. Furthermore, as was known long ago, it is convenient to extend this space to a 4dimensional space, with the fourth coordinate as time, the time origin being selected, say, as the birth of Christalthough this is purely arbitrary (it might be more convenient to select the birth of the solar system, or the birth of the earth as the origin, if we could determine these accurately). Then a point with negative time coordinate is a BC point, and a point with positive time coordinate is an AD point. Don't get the idea that "time is the fourth dimension ", however. The above 4dimensional space is only one possible example. In economics, for instance, one uses a very different space, taking for coordinates, say, the number of dollars expended in an industry. For instance, we could deal with a 7dimensional space with coordinates corresponding to the following industries: 1. Steel 5. Chemicals
2. Auto 6. Clothing
3. Farm products 7. Transportation.
4. Fish
We agree that a megabuck per year is the unit of measurement. Then a point (1,000, 800, 550, 300, 700, 200, 900) in this 7space would mean that the steel industry spent one billion dollars in the given year, and that the chemical industry spent 700 million dollars in that year. The idea of regarding time as a fourth dimension is an old one. Already in the Encyclopedie of Diderot, dating back to the eighteenth century, d'Alembert writes in his article on "dimension": Cette maniere de considerer les quantites de plus de trois dimensions est aussi exacte que l'autre, car les lettres peuvent toujours etre regardees com me representant des nombres rationnels ou non. J'ai dit plus haut qu'il n'etait pas possible de concevoir plus de trois dimensions. Un homme d'esprit de rna connaissance croit qu'on pourrait cependant regarder la duree comme une quatrieme dimension, et que Ie produit temps par la solidite serait en quelque maniere un produit de quatre dimensions; cette idee peut etre contestee, mais elle a, ce me semble, quelque merite, quand ce ne serait que celui de la nouveaute. Encyclopedie, Vol. 4 (1754), p. 1010
4
[I,
VECTORS
~ 1]
Translated, this means: This way of considering quantItIes having more than three dimensions is just as right as the other, because algebraic letters can always be viewed as representing numbers, whether rational or not. I said above that it was not possible to conceive more than three dimensions. A clever gentleman with whom I am acquainted believes that nevertheless, one could view duration as a fourth dimension, and that the product time by solidity would be somehow a product of four dimensions. This idea may be challenged, but it has, it seems to me, some merit, were it only that of being new.
Observe how d'Alembert refers to a "clever gentleman" when he apparently means himeself. He is being rather careful in proposing what must have been at the time a far out idea, which became more prevalent in the twentieth century. D' Alembert also visualized clearly higher dimensional spaces as "products" of lower dimensional spaces. For instance, we can view 3space as putting side by side the first two coordinates (x l' x 2 ) and then the third x 3 . Thus we write
We use the product sign, which should not be confused with other "products", like the product of numbers. The word "product" is used in two contexts. Similarly, we can write
There are other ways of expressing R4 as a product, namely
This means that we view separately the first two coordinates (x l' x 2 ) and the last two coordinates (X3' x 4 ). We shall come back to such products later. We shall now define how to add points. If A, B are two points, say in 3space, and then we define A
+B
to be the point whose coordinates are
Example 2. In the plane, if A == (1, 2) and B == (  3, 5), then A
+ B == ( 
2, 7).
[I, §1]
DEFINITION OF POINTS IN SPACE
In 3space, if
A ==
(
1, n, 3) and A
+
B ==
B == (j2,
(j2 
1, n
5
7,  2), then
+ 7, 1).
U sing a neutral n to cover both the cases of 2space and 3space, the points would be written
and we define A
+B
to be the point whose coordinates are
We observe that the following rules are satisfied: 1. 2. 3.
(A + B) + C == A A + B == B + A. If we let
+ (B + C).
o == (0, 0, ... ,0) be the point all of whose coordinates are 0, then
O+A==A+O==A 4.
for all A. Let A == (a l ,
...
,an) and let  A == (  at, ... , an). Then A
+ (A)
==
O.
All these properties are very simple, and are true because they are true for numbers, and addition of ntuples is defined in terms of addition of their components, which are numbers.
°
Note. Do not confuse the number and the ntuple (0, ... ,0). We usually denote this ntuple by 0, and also call it zero, because no difficuI ty can occur in practice. We shall now interpret addition and multiplication by numbers geometrically in the plane (you can visualize simultaneously what happens in 3space). Example 3. Let A
== (2,3) and B == (1, 1). Then A
+ B ==
(1, 4).
6
[I, §1]
VECTORS
The figure looks like a parallelogram (Fig. 3). (1,4) (2,3)
( 1,1)
Figure 3
Example 4. Let A
=
(3, 1) and B
=
+B
=
A
(1,2). Then
(4,3).
We see again that the geometric representation of our addition looks like a parallelogram (Fig. 4).
A+B
Figure 4
The reason why the figure looks like a parallelogram can be given in terms of plane geometry as follows. We obtain B = (1, 2) by starting from the origin 0 = (0, 0), and moving 1 unit to the right and 2 up. To get A + B, we start from A, and again move 1 unit to the right and 2 up. Thus the line segments between 0 and B, and between A and A + B are the hypotenuses of right triangles whose corresponding legs are of the same length, and parallel. The above segments are therefore parallel and of the same length, as illustrated in Fig. 5.
A+B B
LJ
Figure 5
[I, §1]
DEFINITION OF POINTS IN SPACE
7
Example 5. If A == (3, 1) again, then  A == (  3,  1). If we plot this point, we see that  A has opposite direction to A. We may view  A as the reflection of A through the origin.
A
A Figure 6 We shall now consider multiplication of A by a number. If c is any number, we define cA to be the point whose coordinates are
Example 6. If A == (2, 1,5) and c == 7, then cA == (14, 7,35). I t is easy to verify the rules: 5. 6.
c(A + B) == cA + cB. If Cl~ C 2 are numbers, then
and Also note that (I)A==A.
What is the geometric representation of multiplication by a number? Example 7. Let A == (1,2) and c == 3. Then cA == (3,6)
as in Fig. 7(a). Multiplication by 3 amounts to stretching A by 3. Similarly,!A amounts to stretching A by ~, i.e. shrinking A to half its size. In general, if t is a number, t > 0, we interpret tA as a point in the same direction as A from the origin, butt times the distance. In fact, we define A and
8
[I, §1]
VECTORS
B to have the same direction if there exists a number c > 0 such that A = cB. We emphasize that this means A and B have the same direction with respect to the origin. For simplicity of language, we omit the words "with respect to the origin". Mulitiplication by a negative number reverses the direction. Thus  3A would be represented as in Fig. 7(b).
3A = (3,6) 3A
3A
(a)
(b)
Figure 7
We define A, B (neither of which is zero) to have opposite directions if there is a number c < 0 such that cA = B. Thus when B =  A, then A, B have opposite direction.
Exercises I, § 1 Find A + B, A  B, 3A,  2B in each of the following cases. Draw the points of Exercises 1 and 2 on a sheet of graph paper. 1. A = (2,  1), B = (  1, 1)
2. A = ( 1, 3), B = (0, 4)
3. A = (2, 1,5), B = (1,1,1)
4. A
5. A = (n, 3, 1), B = (2n,  3,7)
6. A = (15,  2,4), B = (n, 3, 1)
7. Let A = (1,2) and B = (3,1). Draw A A  3B on a sheet of graph paper.
= (1, 2,3), B
+ B,
A
+ 2B,
8. Let A, B be as in Exercise 1. Draw the points A A  3B, A + tB on a sheet of graph paper.
A
(1,3, 4)
=
+ 3B,
+ 2B,
9. Let A and B be as drawn in Fig. 8. Draw the point AB.
A
A  B, A  2B,
+ 3B,
A  2B,
[I, §2]
9
LOCATED VECTORS
13
A
B
A
(b)
(a)
B A
A
B (d)
( C)
Figure 8
I, §2. Located Vectors We define a located vector to be an ordered pair of points which we . write AB. (This is not a product.) We visualize this as an arrow between A and B. We call A the beginning point and B the end point of the located vector (Fig. 9).
b2  a 2
{
r~B A~~~
Figure 9
We observe that in the plane,
Similarly,
10
[I,
VECTORS
~2J
This means that B == A
+ (B
 A)
+
+
Let AB and CD be two located vectors. We shall say that they are + equivalent if B  A == D  C. Every located vector AB is equivalent to + one whose beginning point is the origin, because AB is equivalent to I
Clearly this is the only located vector whose beginning point + is the origin and which is equivalent to AB. If you visualize the parallelogram law in the plane, then it is clear that equivalence of two located vectors can be interpreted geometrically by saying that the lengths of the line segments determined by the pair of points are equal, and that the ~~ directions" in which they point are the same. O(B  A).
I
In the next figures, we have drawn the located vectors O(B  A) , + I + AB , and O(A  B) , BA .
A~B
A~·B
BA
o
o AB Figure 11
Figure 10
+
Example 1. Let P == (1,  1, 3) and Q == (2, 4, 1). Then PQ is equiva+
lent to OC , where C == Q  P == (1, 5, 2). If A == (4, 2,5) +
and
B
== (5, 3, 3),
+
then PQ is equivalent to AB because
Q  P == B  A == (1,5, 2). +
Given a located vector OC whose beginning point is the OrIgIn, we + shall say that it is located at the origin. Given any located vector AB , we shall say that it is located at A. A located vector at the origin is entirely determined by its end point. In view of this, we shall call an ntuple either a point or a vector, depending on the interpretation which we have in mind. + + Two located vectors AB and PQ are said to be parallel if there is a number c =1= 0 such that B  A == c(Q  P). They are said to have the
[I, ~2J
11
LOCATED VECTORS
°
same direction if there is a number c > such that B  A = c(Q  P), and have opposite direction if there is a number c < such that B  A
= c( Q 
°
P).
In the next pictures, we illustrate parallel located vectors.
p
B A
Q
(b) Opposite direction
(a) Same direction
Figure 12
Example 2. Let P
= (3,7)
Q = (4,2).
and
Let A == (5, 1)
and
B = (  16,  14).
Then
Q P .
== (
7,  5)
B  A = (21, 15).
and
.
Hence PQ is parallel to AB, because B  A = 3(Q  P). Since 3 > 0, . . we even see that PQ and AB have the same direction. In a similar manner, any definition made concerning ntuples can be carried over to located vectors. For instance, in the next section, we shall define what it means for ntuples to be perpendicular.
B~ QP
BA
o Figure 13
Q
: /
12
[I, §3]
VECTORS +
+
Then we can say that two located vectors AB and PQ are perpendicular if B  A is perpendicular to Q  P. In Fig. 13, we have drawn a picture of such vectors in the plane.
Exercises I, §2 +
+
In each case, determine which located vectors PQ and AB are equivalent. 1. P = (1, 1), Q = (4, 3), A = (1, 5), B = (5, 2). 2. P
= (1,4), Q = (3,5),
A
= (5,7),
B
= (1, 8).
3. P = (1, 1,5), Q = (2,3, 4), A = (3,1,1), B = (0, 5,10). 4. P
= (2, 3,  4), Q = (  1, 3, 5),
A
= (  2, 3,  1),
B
+
= (  5, 3, 8). +
In each case, determine which located vectors PQ and AB are parallel. 5. P
= (1, 1), Q = (4, 3), A = (1, 5), B = (7, 1).
6. P = (1,4), Q = (3,5), A = (5,7), B = (9,6).
7. P = (1, 1, 5), Q = ( 2, 3, 4), A = (3, 1, 1), B = (  3,9, 17). 8. P
= (2,3, 4), Q = (1,3,5),
A
= (2,3, 1),
B
= (11,3, 28).
9. Draw the located vectors of Exercises 1, 2, 5, and 6 on a sheet of paper to illustrate these exercises. Also draw the located vectors QP and BA. Draw the points Q  P, B  A, P  Q, and AB.
I, §3. Scalar Product It is understood that throughout a discussion we select vectors always in the same ndimensional space. You may think of the cases n == 2 and n == 3 only. In 2space, let A == (aI' a 2) and B == (b l , b2). We define their scalar product to be
In 3space, let A == (aI' a 2 , a 3) and B == (b l , b2, b 3). scalar product to be
We define their
In nspace, covering both cases with one notation, let A == (aI' ... ,an) and B == (b l , ..• ,b n ) be two vectors. We define their scalar or dot product A·B to be
[I, §3]
13
SCALAR PRODUCT
This product is a number. For instance, if A
==
(1, 3,  2)
B == (  1, 4,  3),
and
then A . B ==  1 + 12
+ 6 ==
17.
For the moment, we do not give a geometric interpretation to this scalar product. We shall do this later. We derive first some important properties. The basic ones are: SP t. We have A· B
==
B· A.
SP 2. If A, B, C are three vectors, then A . (B
+ C) ==
A .B
+ A . C == (B + C)· A.
SP 3. If x is a number, then (xA)·B == x(A·B)
and
A· (xB)
== x(A . B).
SP 4. If A == 0 is the zero vector, then A· A == 0, and otherwise A·A > O.
We shall now prove these properties. Concerning the first, we have
because for any two numbers a, b, we have ab first property. For SP 2, let C == (c1, ... ,c n ). Then
and
Reordering the terms yields
==
ba. This proves the
14
[I,
VECTORS
~3J
which is none other than A· B + A . c. This proves what we wanted. We leave property SP 3 as an exercise. Finally, for SP 4, we observe that if one coordinate a i of A is not eq ual to 0, then there is a term af # and af > in the scalar prod uct
°
ai + ... + a;.
A .A =
Since every term shown.
IS
°
> 0, it follows that the sum
IS
> 0, as was to be
In much of the work which we shall do concerning vectors, we shall use only the ordinary properties of addition, multiplication by numbers, and the four properties of the scalar product. We shall give a formal discussion of these later. For the moment, observe that there are other objects with which you are familiar and which can be added, subtracted, and multiplied by numbers, for instance the continuous functions on an interval [a, bJ (cf. Example 2 of Chapter VI, §1). Instead of writing A· A for the scalar product of a vector with itself, it will be convenient to write also A 2. (This is the only instance when we allow ourselves such a notation. Thus A 3 has no meaning.) As an exercise, verify the following identities:
A dot product A· B may very well be equal to B being the zero vector. For instance, let A = (1,2,3)
and
°
without either A or
B = (2, 1, ~).
Then A·B =
°
We define two vectors A, B to be perpendicular (or as we shall also say, orthogonal), if A· B = 0. For the moment, it is not clear that in the plane, this definition coincides with our intuitive geometric notion of perpendicularity. We shall convince you that it does in the next section. Here we merely note an example. Say in R 3 , let E1 = (1,0,0),
E2
=
(0, 1,0),
E3 = (0,0,1)
be the three unit vectors, as shown on the diagram (Fig. t 4).
[I, §4]
15
THE NORM OF A VECTOR
z
~t~
Y
x
Figure 14
°
Then we see that E 1 • E2 == 0, and similarly E i • Ej == if i =1= j. And these vectors look perpendicular. If A == (a l ' a2' a 3), then we observe that the ith component of A, namely
is the dot product of A with the ith unit vector. We see that A is perpendlcular to Ei (according to our definition of perpendicularity with the dot product) if and only if its ith component is equal to 0.
Exercises I, §3 1. Find A· A for each of the following ntuples. (a) A =(2, 1), B=(I, 1) (b) A =(1,3), B=(0,4) (c) A =(2, 1,5), B=(I, 1, 1) (d) A =(1, 2,3), B=(1,3, 4) (e) A = (n, 3, 1), B = (2n, 3,7) (f) A = (15, 2,4), B = (n, 3, 1) 2. Find A· B for each of the above ntuples. 3. Using only the four properties of the scalar product, verify in detail the identities given in the text for (A + B)2 and (A  B)2. 4. Which of the following pairs of vectors are perpendicular? (b) (1, 1,1) and (2,3,1) (a) (1, 1,1) and (2,1,5) (c) (5,2,7) and (3, 1,2) (d) (n,2, 1) and (2, n,O) 5. Let A be a vector perpendicular to every vector X. Show that A =
o.
I, §4. The Norm of a Vector We define the norm of a vector A, and denote by
IIAII==~.
/lAII, the number
16
[I, §4]
VECTORS
Since A· A >0, we can take the square root. The norm times called the magnitude of A.
IS
also some
When n = 2 and A = (a, b), then
as in the following picture (Fig. 15).
b
)
y
a
Figure 15
Example 1. If A
=
(1, 2), then
I AI
IIAII
=
=
J1+4 = v'S. Jai + a~ + a~.
Example 2. If A = ( 1, 2, 3), then
IIAII = If n
Jl + 4 + 9 = fo·
= 3, then the picture looks like Fig. 16, with
A
A
z
,,
./
,
w',
,/ ./
"
// ,/./
~ (x, y)
Figure 16
7
= (x, y, z).
[I, §4]
17
THE NORM OF A VECTOR
If we first look at the two components (x, y), then the length of the x 2 + y2, as indicated. segment between (0, 0) and (x, y) is equal to w = Then again the norm of A by the Pythagoras theorem would be
J
Thus when n = 3, our definition of norm is compatible with the geometry of the Pythagoras theorem. In terms of coordinates, A = (aI' ... ,an) we see that
IIAII
=
Jai + ... + a;.
°
If A :f. 0, then I A I :f. because some coordinate and hence + ... + > 0, so I A I :f. 0. Observe that for any vector A we have
ai
a;
ai :f. 0,
so that
af
> 0,
IIAII = IIAll· This is due to the fact that
because ( 1)2
=
1. Of course, this is as it should be from the picture: A
A Figure 17
Recall that A and  A are said to have opposite direction. However, they have the same norm (magnitude, as is sometimes said when speaking of vectors). Let A, B be two points. We define the distance between A and B to be IIA  BII = J(A  B)·(A  B).
18
[I, §4]
VECTORS
This definition coincides with our geometric intuition when A, Bare points in the plane (Fig. 18). It is the same thing as the length of the + + located vector AB or the located vector BA .
B A
Length = IIABII = lIBAIl
BA
Figure 18
Example 3. Let A + located vector AB is
= (
liB
1, 2) and B = (3, 4). Then the length of the  All. But B  A = (4,2). Thus
liB  All
=
J16
+4=
fo·
In the picture, we see that the horizontal side has length 4 and the vertical side has length 2. Thus our definitions reflect our geometric intuition derived from Pythagoras.
B
4
3 A
2+' 1
3 2 1
123
Figure 19
Let P be a point in the plane, and let a be a number > O. The set of points X such that
IIX  PII < a will be called the open disc of radius a centered at P. The set of points X such that
IIX  PII < a
[I, §4]
19
THE NORM OF A VECTOR
will be called the closed disc of radius a and center P. The set of points X such that
IIX  PII == a is called the circle of radius a and center P. These are illustrated in Fig. 20.
p
a
Disc
Circle
Figure 20
In 3dimensional space, the set of points X such that
IIX  PII < a will be called the open ball of radius a and center P. The set of points X such that
IIX  PII
0, then IleA I = ellA II, i.e. if we stretch a vector A by multiplying by a positive number e, then the length stretches also by that amount. We verify this formally using our definition of the length. Theorem 4.1 Let x be a number. Then
IlxA11
=
Ixl IIAII
(absolute value of x times the norm of A). Proof By definition, we have
IIxAI12
=
(xA)·(xA),
which is equal to
by the properties of the scalar product. Taking the square root now yields what we want. Let S 1 be the sphere of radius 1, centered at the origin. Let a be a number > O. If X is a point of the sphere S 1, then aX is a point of the sphere of radius a, because
IlaX11 = a11X11 = a. In this manner, we get all points of the sphere of radius a. (Proof?) Thus the sphere of radius a is obtained by stretching the sphere of radius 1, through multiplication by a. A similar remark applies to the open and closed balls of radius a, they being obtained from the open and closed balls of radius 1 through multiplication by a.
Disc of radius 1
Disc of radius a
Figure 22
[I, §4]
21
THE NORM OF A VECTOR
We shall say that a vector E is a unit vector if vector A, let a = I A II. If a :f. 0, then
IIEII
=
1. Given any
1
A a
is a unit vector, because 1
1
=a=1.
A a
a
We say that two vectors A, B (neither of which is 0) have the same direction if there is a number c > such that cA = B. In view of this definition, we see that the vector
°
1
A
IIAII
is a unit vector in the direction of A (provided A :f. 0). A
1
E=A
IIAII
Figure 23
If E is the unit vector in the direction of A, and
I AI
=
a, then
A = aE. Example 4. Let A = (1,2, 3). Then vector in the direction of A is the vector
IIAII
=
Jl4.
Hence the unit
1 2 3) (Jl4' Jl4' Jl4 .
E=
Warning. There are as many unit vectors as there are directions. The three standard unit vectors in 3space, namely E1 = (1,0,0),
E2
=
(0, 1, 0),
E3 = (0,0, 1)
are merely the three unit vectors in the directions of the coordinate axes.
22
[I, §4]
VECTORS
We are also in the position to justify our definition of perpendicularity. Given A, B in the plane, the condition that
IIA + BII
=
IIA  BII
(illustrated in Fig. 24(b)) coincides with the geometric property that A should be perpendicular to B. A
IIABII
A 1
/
B
/ /
/
/
/
IIA+BII /1
/
/ /
1
/
/
1 1 1
B
;!
/ / /
I
B (a)
(b)
Figure 24
We shall prove:
IIA + BII Let
=
IIA  BII if and only if A·B = o.
denote "if and only if". Then
IIA + BII = IIA  BII
IIA + BI12 = IIA  BI12 A 2 + 2A . B + B2 = A 2
4A·B = 0
A· B
=

2A . B
+ B2
O.
This proves what we wanted. General Pythagoras theorem. If A and B are perpendicular, then
The theorem is illustrated on Fig. 25.
A+B
A
B
Figure 25
[I, §4]
23
THE NORM OF A VECTOR
To prove this, we use the definitions, namely
IIA + BI12 = (A + B)·(A + B) = A2 + 2A·B + B2 =
because A·B
IIAI12 + IIBI12,
0, and A·A
=
=
IIAI12, B·B = IIBI12 by definition.
Remark. If A is perpendicular to B, and x is any number, then A is also perpendicular to xB because A . xB = xA . B =
o.
We shall now use the notion of perpendicularity to derive the notion of projection. Let A, B be two vectors and B =I O. Let P be the point + + + on the line through DB such that PAis perpendicular to DB, as shown on Fig. 26(a). A
A
AcB
o
o (a)
(b)
Figure 26
We can write P = cB
for some number c. We want to find this number c explicitly in terms of + + A and B. The condition P A ~ DB means that A  P is perpendicular to B, and since P
=
cB this means that (A  cB)· B
=
0,
in other words, A . B  cB· B =
We can solve for c, and we find A· B
=
A·B
c=. B·B
o.
cB· B, so that
24
[I, §4]
VECTORS
Conversely, if we take this value for c, and then use distributivity, dotting A  cB with B yields 0, so that A  cB is perpendicular to B. Hence we have seen that there is a unique number c such that A  cB is perpendicular to B, and c is given by the above formula. A·B Definition. The component of A along B is the number c =   . B·B
.. fl' h A .B 0 A a ong B IS t e vector cB =   B. Th e projection B·B Example 5. Suppose
B = Ei = (0, ... ,0, 1, 0, ... ,0) is the ith unit vector, with 1 in the ith component and components.
°
in all other
Thus A· Ei is the ordinary ith component of A. M ore generally, we have simply
if
B is a unit vector, not necessarily one of the E i , then
c = A·B because B· B = 1 by definition of a unit vector. Example 6. Let A = (1,2, 3) and B of A along B is the number A·B B·B
=
(1,1,2). Then the component
3
1
6
2
c===
Hence the projection of A along B is the vector cB
= (  t,  t, 
1).
Our construction gives an immediate geometric interpretation for the scalar product. Namely, assume A =1= 0 and look at the angle f} between A and B (Fig. 27). Then from plane geometry we see that
c11B11  IIAII'
f} _
cos
or substituting the value for c obtained above. A·B=
IIAII IIBII cos
f}
and
A·B
cos (J =
IIA I IIBII'
[I, §4]
25
THE NORM OF A VECTOR
A
Figure 27
In some treatments of vectors, one takes the relation
A·B
=
IIAII IIBII
cos fJ
as definition of the scalar product. This is subject to the following disadvantages, not to say objections: (a) (b)
(c)
The four properties of the scalar product SP 1 through SP 4 are then by no means obvious. Even in 3space, one has to rely on geometric intuition to obtain the cosine of the angle between A and B, and this intuition is less clear than in the plane. In higher dimensional space, it fails even more. It is extremely hard to work with such a definition to obtain further properties of the scalar product.
Thus we prefer to lay obvious algebraic foundations, and then recover very simply all the properties. We used plane geometry to see the expreSSIon
A·B
=
IIAII IIBII
cos fJ.
After working out some examples, we shall prove the inequality which allows us to justify this in nspace. Example 7. Let A = (1, 2,  3) and B = (2, 1, 5). Find the COSIne of the angle () between A and B. By definition,
A·B
cos ()
=
2
+2
15
11
foJ30 foG·
IIAII IIBII
Example 8. Find the COSIne of the angle between the two located +
+
vectors PQ and P R where p = (1, 2,  3),
Q=
(
2, 1, 5),
R = (1, 1,  4).
26
[I, §4]
VECTORS
The picture looks like this: Q
p
Figure 28
We let A
= Q  P = (  3,  1, 8)
and
+
B = R  P = (0,  1,  1).
+
Then the angle between PQ and P R is the same as that between A and B. Hence its cosine is equal to
cos () =
A·B
0+18
IIAII IIBII
fiJi
We shall prove further properties of the norm and scalar product using our results on perpendicularity. First note a special case. If
Ei = (0, ... ,0,1,0, ... ,0) is the ith unit vector of R n, and
then A·E·I = a·I
is the ith component of A, i.e. the component of A along E i • We have
so that the absolute value of each component of A is at most equal to the length of A. We don't have to deal only with the special unit vector as above. Let E be any unit vector, that is a vector of norm 1. Let c be the component of A along E. We saw that c
=
A·E.
[I, §4]
27
THE NORM OF A VECTOR
Then A  cE is perpendicular to E, and
A
=
A  cE
+ cEo
Then A  cE is also perpendicular to cE, and by the Pythagoras theorem, we find
Thus we have the inequality
c2 < IIAI12, and Icl < IIAII.
In the next theorem, we generalize this inequality to a dot product A· B when B is not necessarily a unit vector. Theorem 4.2. Let A, B be two vectors in Rn. Then
IA· BI < IIAII IIBII· Proof If B
0, then both sides of the inequality are equal to 0, and so our assertion is obvious. Suppose that B i= O. Let c be the component of A along B, so c = (A· B)/(B· B). We write =
A = A  cB
+ cB.
By Pythagoras,
Hence
c211BI12 < IIAI12. But 2 C
2
IIBII
(A·B)2 = (B B)2
2_IA.BI 2
2_IA.BI2
IIBII  IIBI14 IIBII  IIBII2 
Therefore
MUltiply by
I BII2 and take the square root to conclude the proof.
In view of Theorem 4.2, we see that for vectors A, B in nspace, the number A·B
IIAIIIIBIl has absolute value < 1. Consequently, 1
0, and such that the sum of all the elements in each column is equal to 1. Such a matrix is called a Markov matrix. If A is a square matrix, then we can form the product AA, which will be a square matrix of the same size as A. It is denoted by A2. Similarly, we can form A 3 , A 4 , and in general, An for any positive integer n. Thus An is the product of A with itself n times. We can define the unit n x n matrix to be the matrix having diagonal components all equal to 1, and all other components equal to O. Thus the unit n x n matrix, denoted by In' looks like this:
In
==
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1 0
0
0
0
We can then define AO == I (the unit matrix of the same size as A). Note that for any two integers r, s > 0 we have the usual relation
For example, in the Markov process described above, we may express the population vector in the (n + 1)th year as
where X 1 is the population vector in the first year. Warning. It is not always true that AB == BA. For instance, compute AB and BA in the following cases:
A =
(~
~)
B=
(~
1)5 .
You will find two different values. This is expressed by saying that multiplication of matrices is not necessarily commutative. Of course, in some special cases, we do have AB == BA. For instance, powers of A commute, i.e. we have A rAS == AS Ar as already pointed out above. We now prove other basic properties of multiplication.
[II, §2]
53
MULTIPLICATION OF MATRICES
Distributive law. Let A, B, C be matrices. Assume that A, B can be
multiplied, and A, C can be multiplied, and B, C can be added. Then A, B + C can be multiplied, and we have A(B
+ C)
=
AB
+ AC.
If x is a number, then A(xB) = x(AB). Proof Let Ai be the ith row of A and let Bk, Ck be the kth column of Band C, respectively.... Then Bk + Ck is the kth column of B + C. By definition, the ikcomponent of A(B + C) is Ai· (Bk + C k). Since
our first assertion follows. As for the second, observe that the kth column of xB is XBk. Since
our second assertion follows. Associative law. Let A, B, C be matrices such that A, B can be multi
plied and B, C can be multiplied. Then A, BC can be multiplied. So can AB, C, and we have (AB)C Proof Let A
=
=
A(BC).
(a ij ) be an m x n matrix, let B
(b jk ) be an n x r x s matrix. The product AB is an m x r =
matrix, and let C = (C kl ) be an r matrix, whose ikcomponent is equal to the sum
We shall abbreviate this sum using our
L
notation by writing
n
L aijbjk ·
j= 1
By definition, the iicomponent of (AB)C is equal to
54
MATRICES AND LINEAR EQUATIONS
[II, §2]
The sum on the right can also be described as the sum of all terms
where j, k range over all integers 1
o.
r cos cp) . , ( r SIn cp
Since R(8)rX
=
rR(8)X,
we see that multiplication by R(8) also has the effect of rotating r X by an angle 8. Thus rotation by an angle 8 can be represented by the matrix R(8).
R(8)X = t(cos(8
+ qJ), sin(8 + qJ»
x
= t(cos qJ, sin qJ)
Figure 1
Note that for typographical reasons, we have written the vector t X horizontally, but have put a little t on the upper left superscript, to denote transpose, so X is a column vector.
58
MATRICES AND LINEAR EQUATIONS
[II, §2]
Example. The matrix corresponding to rotation by an angle of given by COS
R(n / 3) ==
. ( SIn
n/3 n
 ( fi/2
IS
sin n/3) cos n/3
/3
1/2
n/3
fi/2). 1/2
Example. Let X == t(2, 5). If you rotate X by an angle of n/3, find the coordinates of the rotated vector. These coordinates are: R(n/3)X ==
==
1/2 J3/2)(2) 5 ( y ;;3/2 1/2
(1  5fi /2). fi + 5/2
Warning. Note how we multiply the column vector on the left with the matrix R(8). If you want to work with row vectors, then take the transpose and verify directly that
(2, 5)
1/2
;;
( y3/2
fi/2) 1/2
==
(1 
5fi/2,
fi + 5/2).
So the matrix R(8) gets transposed. The minus sign is now in the lower lefthand corner.
Exercises II, §2 The following exercises give mostly routine practice in the multiplication of matrices. However, they also illustrate some more theoretical aspects of this multiplication. Therefore they should be all worked out. Specifically: Exercises 7 through 12 illustrate multiplication by the standard unit vectors. Exercises 14 through 19 illustrate multiplication of triangular matrices. Exercises 24 through 27 illustrate how addition of numbers is transformed into multiplication of matrices. Exercises 27 through 32 illustrate rotations. Exercises 33 through 37 illustrate elementary matrices, and should be worked out before studying §5. 1. Let I be the unit n x n matrix. Let A be an n x r matrix. What is I A? If A is an m x n matrix, what is AI?
2. Let D be the matrix all of whose coordinates are O. Let A be a matrix of a size such that the product AD is defined. What is AD?
[II, §2]
59
MULTIPLICATION OF MATRICES
3. In each one of the following cases, find (AB)C and A(BC). (a) A
(b) A
(c) A
=G ') B=(' ~} c=G ~) =G ~} B= (~ ~} =G) 1 '
1
C
=
G
4
°
3
1
_:} B= (~
1
~}
C
5
=(
~ :)
1
4. Let A, B be square matrices of the same SIze, and assume that AB = BA. Show that (A
and
+ B)(A
 B) = A2  B2,
using the distributive law.
5. Let
B=G Find AB and BA.
6. Let 0) 7 . Let A, B be as in Exercise 5. Find CA, AC, CB, and BC. State the general rule including this exercise as a special case. 7. Let X = (1,0,0) and let
° What is XA? 8. Let X = (0, 1, 0), and let A be an arbitrary 3 x 3 matrix. How would you describe X A? What if X = (0, 0, I)? Generalize to similar statements concerning n x n matrices, and their products with unit vectors. 9. Let A
=
(!
3)5 .
Find AX for each of the following values of X.
(a) X
=(~)
(b) X
=(:)
(c) X
=(D
60
MATRICES AND LINEAR EQUATIONS
[II, §2]
10. Let
7
1 1 Find AX for each of the values of X given in Exercise 9. 11. Let
and
What is AX? 12. Let X be a column vector having all its components equal to 0 except the jth component which is equal to 1. Let A be an arbitrary matrix, whose size is such that we can form the product AX. What is AX?
13. Let X be the indicated column vector, and A the indicated matrix. Find AX as a column vector. (a)
=~} A=(~
0 0
~c) X=(=} A=(~ 14. Let
A= ( :
!}
:)
1
X
~)
1 0
(=) A=G
X=
(d)
Find the product
~)
X=(i} A=G
(b)
0 0
~)
AS for each one of the following ma
trices S. Describe in words the effect on A of this product.
G ~) A G !)
(a) S
15. Let
=
=
(b) S
=
C 0)
1 .
again. Find the product
SA for each one of the following
matrices S. Describe in words the effect of this product on A. (a) S =
G
~)
(b) S =
C
16. (a) Let A be the matrix
(~
1
o o
Find A 2 , A 3. Generalize to 4 x 4 matrices.
[II, §2]
61
MULTIPLICATION OF MATRICES
(b) Let A be the matrix
(~
Compute A 2, A 3 , A4.
1 1 0
17. Let 0
A=G
2 0
Find A 2, A 3 , A4. 18. Let A be a diagonal matrix, with diagonal elements a 1 , ••• ,an. What IS A 2, A 3, Ak for any positive integer k? 19. Let
A
=(~
o o
Find A 3 20. (a) Find a 2 x 2 matrix A such that A2 = J =
1 (
(b) Determine all 2 x 2 matrices A such that A 2 =
0
o.
21. Let (a) (b) (c)
A be a square matrix. If A 2 = 0 show that J  A is invertible. If A 3 = 0, show that J  A is invertible. In general, if An = 0 for some positive integer n, show that J  A IS invertible. [Hint: Think of the geometric series.] (d) Suppose that A2 + 2A + J = O. Show that A is invertible. (e) Suppose that A 3  A + J = O. Show that A is invertible.
22. Let A, B be two square matrices of the same size. We say that A is similar to B if there exists an invertible matrix T such that B = T A T 1. Suppose this is the case. Prove: (a) B is similar to A. (b) A is invertible if and only if B is invertible. (c) t A is similar to t B. (d) Suppose An = 0 and B is an invertible matrix of the same sIze as A. Show that (BAB 1 )" = O. 23. Let A be a square matrix which is of the form
all
*
0
a 22
0
* *
........
...
* *
0
a""
62
MATRICES AND LINEAR EQUATIONS
[II, §2]
The notation means that all elements below the diagonal are equal to 0, and the elements above the diagonal are arbitrary. One may express this property by saying that i > j.
if
Such a matrix is called upper triangular. If A, B are upper triangular matrices (of the same size) what can you say about the diagonal elements of AB?
Exercises 24 through 27 give examples where addition of numbers IS transformed into multiplication of matrices. 24. Let a, b be numbers, and let
A=G ~)
and
What is AB? What is A 2, A 3? What is An where n is a positive integer? 25. Show that the matrix A in Exercise 24 has an inverse. What is this inverse? 26. Show that if A, Bare n x n matrices which have inverses, then AB has an Inverse. 27. Rotations. Let R(O) be the matrix given by
R(O)
=
COS (
0
. 0
SIn
0).
sin cos 0
(a) Show that for any two numbers 0 1 , O2 we have
[You will have to use the addition formulas for sine and cosine.] (b) Show that the matrix R(O) has an inverse, and write down this inverse. (c) Let A = R(O). Show that
A2 =
2fJ sin 20
(COS
20).
sin cos 20
(d) Determine An for any positive integer n. Use induction. 28. Find the matrix R( 0) associated with the rotation for each of the following values of O. (a) n/2 (b) n/4 (c) n (d) n (e) n/3 (f) n/6 (g) 5n/4 29. In general, let 0 > O. What is the matrix associated with the rotation by an angle  0 (i.e. clockwise rotation by O)?
[II, §2]
63
MULTIPLICATION OF MATRICES
30. Let X = l( 1, 2) be a point of the plane. If you rotate X by an angle of n/4, what are the coordinates of the new point?
31. Same question when X
t( 1,3) and the rotation is by an angle of n/2.
=
32. For any vector X in R2 let Y = R(O)X be its rotation by an angle that I YII = IIXII·
o.
Show
The following exercises on elementary matrices should be done before studying §5.
33. Elementary matrices. Let
A = (
~
3
1
4
2
1
2
3 3
1
~)
5
4
Let U be the matrix as shown. In each case find U A.
o o o o o o o o o o
o o o o o o o o o o o
o o o o
o o o o
~) ~)
o o o o
o
o o o o
o o o o
o o
o o
o o o o 1
o
34. Let E be the matrix as shown. Find EA where A is the same matrix as in the preceding exercise.
(a)
(c)
(~
(~
o o o o
o o
(b)
o o o
o 5
o o o
o
~)
(d)
o
(~
o o
(~
o
o o
2
1
o
o
o
o o
~) ~)
64
[II, §3]
MATRICES AND LINEAR EQUATIONS
35. Let E be the matrix as shown. Find EA where A is the same matrix as in the preceding exercise and Exercise 33.
(a)
(~ ( 21
(c)
~
0
0
0
1
0
0
0
1
0
0
0 0
0
1
0
~)
0 0
(b)
0
(~
(d)
3 0
0
0 0 0
0 0
(~
t
0 0
0
1
0
2 0
0
~)
36. Let A = (aij) be an m x n matrix,
e;1 am!
...
a~"). a mn
Let 1 ~ r ~ m and 1 ~ s ~ m. Let Irs be the matrix whose rscomponent is t and such that all other components are equal to o. (a) What is Irs A ? (b) Suppose r =1= s. What is (Irs + Isr)A? (c) Suppose r =1= s. Let I jj be the matrix whose jjcomponent is 1 and such that all other components are O. Let E rs
=
Irs + Isr + sum of all I jj for j
=1=
r,
j =1= s.
37. Again let r =1= s. (a) Let E = I + 3I rs . What is EA? (b) Let c be any number. Let E = I + cI rs . What is EA?
The rest of the chapter will be mostly concerned with linear equations, and especially homogeneous ones. We shall find three ways of interpreting such equations, illustrating three dfferent ways of thinking about matrices and vectors.
II, §3. Homogeneous Linear Equations and Elimination In this section, we look at linear equations by one method of elimination. In the next section, we shall discuss another method. We shall be interested in the case when the number of unknowns is greater than the number of equations, and we shall see that in that case, there always exists a nontrivial solution. Before dealing with the general case, we shall study examples.
[II, ~3J
HOMOGENEOUS LINEAR EQUATIONS AND ELIMINATION
65
Example t. Suppose that we have a single equation, like 2x
+y
 4z
= o.
We wish to find a solution with not all of x, y, z equal to O. equivalent equation is 2x
= 
y
An
+ 4z.
To find a nontrivial solution, we give all the variables except the first a special value i= 0, say y = 1, z = 1. We than solve for x. We find 2x =  y
+ 4z =
3,
whence x =~.
Example 2. Consider a pair of equations, say (1 )
2x
(2)
x
+ 3y  z = 0, +
y
+z=
O.
We redute the problem of solving these simultaneous equations to the preceding case of one equation, by eliminating one variable. Thus we multiply the second equation by 2 and subtract it from the first equation, getting
y  3z =
(3)
o.
Now we meet one equation in more than one variable. We give zany value i= 0, say z = 1, and solve for y, namely y = 3. We then solve for x from the second equation, namely x =  y  z, and obtain x =  4. The values which we have obtained for x, y, z are also solutions of the first equation, because the first equation is (in an obvious sense) the sum of equation (2) multiplied by 2, and equation (3).
Example 3. We wish to find a solution for the system of equations
3x  2y x
+ z + 2w = 0,
+y
 z  w = 0,
2x  2y
+ 3z = o.
Again we use the elimination method. Multiply the second equation by 2 and subtract it from the third. We find
4y+5z+2w=0.
66
MATRICES AND LINEAR EQUATIONS
[II, §3]
Multiply the second equation by 3 and subtract it from the first. We find
 5y + 4z + 5w == O. We have now eliminate~ x from our equations, and find two equations in three unknowns, y, z, w. We eliminate y from these two equations as follows: Multiply the top one by 5, multiply the bottom one by 4, and subtract them. We get
9z  lOw == O. Now give an arbitrary value =I 0 to w, say w == 1. Then we can solve for z, namely
z == 10/9. Going back to the equations before that, we solve for y, using
4y == 5z
+ 2w.
This yields y
== 17/9.
Finally we solve for x USIng say the second of the original set of three equations, so that x ==  y
+ z + w,
or numerically, x
== 49/9.
Thus we have found: w == 1,
z == 10/9,
y == 68/9,
x
== 49/9.
Note that we had three equations in four unknowns. By a successive elimination of variables, we reduced these equations to two equations in three unknowns, and then one equation in two unknowns. Using precisely the same method, suppose that we start with three equations in five unknowns. Eliminating one variable will yield two equations in four unknowns. Eliminating another variable will yield one equation in three unknowns. We can then solve this equation, and proceed backwards to get values for the previous variables just as we have shown in the examples.
[II, §3]
HOMOGENEOUS LINEAR EQUATIONS AND ELIMINATION
67
In general, suppose that we start with m equations with n unknowns, and n > m. We eliminate one of the variables, say Xl' and obtain a system of m  1 equations in n  1 unknowns. We eliminate a second variable, say X 2 , and obtain a system of m  2 equations in n  2 unknowns. Proceeding stepwise, we eliminate m  1 variables, ending up with 1 equation in n  m + 1 unknowns. We then give nontrivial arbitrary values to all the remaining variables but one, solve for this last variable, and then proceed backwards to solve successively for each one of the eliminated variables as we did in our examples. Thus we have an effective way of finding a nontrivial solution for the original system. We shall phrase this in terms of induction in a precise manner. Let A == (aij), i == 1, ... ,m and j == 1, ... ,n be a matrix. Let bb ... ,b m be numbers. Equations like
. a m1 x 1
..
+ ... + amnxn == bm
are called linear equations. We also say that (*) is a system of linear equations. The system is said to be homogeneous if all the numbers b1 , ••• ,bm are equal to o. The number n is called the number of unknowns, and m is the number of equations. The system of equations
will be called the homogeneous system associated with (*). In this section, we study the homogeneous system (**). The system (**) always has a solution, namely the solution obtained by letting all Xi == o. This solution will be called the trivial solution. A solution (x l ' ... ,xn ) such that some Xi is i= 0 is called nontrivial. Consider our system of homogeneous equations (**). Let AI' ... ,Am be the row vectors of the matrix (aij). Then we can rewrite our equations (**) in the form
Am·X ==
o.
Therefore a solution of the system of linear equations can be interpreted as the set of all ntuples X which are perpendicular to the row vectors of the matrix A. Geometrically, to find a solution of (**) amounts to finding a vector X which is perpendicular to AI' ... ,Am. Using the notation of the dot product will make it easier to formulate the proof of our main theorem, namely:
68
MATRICES AND LINEAR EQUATIONS
[II, §3]
Theorem 3.1. Let
be a system of m linear equations in n unknowns, and assume that n > m. Then the system has a nontrivial solution. Proof The proof will be carried out by induction. Consider first the case of one equation in n unknowns, n > 1:
If all coefficients a l , ... ,an are equal to 0, then any value of the variables will be a solution, and a nontrivial solution certainly exists. Suppose that some coefficient a i is =I o. After renumbering the variables and the coefficients, we may assume that it is a l . Then we give X 2 , ... 'X n arbitrary values, for instance we let X 2 == ... == Xn == 1, and solve for Xl' letting
In that manner, we obtain a nontrivial solution for our system of equations. Let us now assume that our theorem is true for a system of m  1 equations in more than m  1 unknowns. We shall prove that it is true for m equations in n unknowns when n > m. We consider the system (**). If all coefficients (aij) are equal to 0, we can give any nonzero value to our variables to get a solution. If some coefficient is not equal to 0, then after renumbering the equations and the variables, we may assume that it is all. We shall subtract a multiple of the first equation from the others to eliminate Xl. Namely, we consider the system of equations
which can also be written in the form
[II, §3]
69
HOMOGENEOUS LINEAR EQUATIONS AND ELIMINATION
In this system, the coefficient of Xl is equal to O. Hence we may VIew (***) as a system of m  1 equations in n  1 unknowns, and we have nl>m1. According to our assumption, we can find a nontrivial solution (x 2 , ... ,xn ) for this system. We can then solve for Xl in the first equation, namely
In that way, we find a solution of A 1 . X == O. But according to (***), we have
for i == 2, ... ,me Hence Ai· X == 0 for i == 2, ... ,m, and therefore we have found a nontrivial solution to our original system (**). The argument we have just given allows us to proceed stepwise from one equation to two equations, then from two to three, and so forth. This concludes the proof.
Exercises II, §3 1. Let E 1 = (1, 0, ... ,0),
E2 = (0, 1, 0, ... ,0),
... ,
En = (0, ... ,0, 1)
be the standard unit vectors of Rn. Let X be an ntuple. If X . Ei = show that X = o.
°for all
i,
2. Let AI' ... ,Am be vectors in Rn. Let X, Y be solutions of the system of equations
X·Ai=O Show that X solution.
+
and
for
i = 1, ... ,me
Y is also a solution. If c is a number, show that cX
IS
a
3. In Exercise 2, suppose that X is perpendicular to each one of the vectors A l ,.·· ,Am· Let c l , ... ,em be numbers. A vector
is called a linear combination of AI' ... ,Am. Show that X is perpendicular to such a vector.
70
MATRICES AND LINEAR EQUATIONS
[II, §4]
4. Consider the inhomogeneous system (*) consisting of all X such that X . Ai = bi for i = 1, ... ,m. If X and X' are two solutions of this system, show that there exists a solution Y of the homogeneous system (**) such that X' = X + Y. Conversely, if X is any solution of (*), and Y a solution of (**), show that X + Y is a solution of (*). 5. Find at least one nontrivial solution for each one of the following systems of equations. Since there are many choices involved, we don't give answers. (a) 3x
+y +z =0
+y+z=
(b) 3x
0
x+y+z=O (c) 2x  3y + 4z = 0 3x + y + z = 0
(d)
2x + y + 4z + w = 0  3x + 2y  3z + w = 0
x+y+z=O (e) x + 2y  4z + x + 3y + z 
W W
=0 =0
(f) 2x+3y+z+4w=O x + y + 2z + 3w = 0
2x + y + z  2w = 0 6. Show that the only solutions of the following systems of equations are trivial.
+ 3y = 0 xy=O
(b)
3x + 4y  2z = 0
(d) 4x  7y + 3z = 0
(a) 2x
(c)
4x + 5y = 0 6x + 7y = 0
x+y+z=O
x+y=O
 x  3y + 5z = 0
Y  6z = 0
(e) 7x  2y + 5z + w = 0
(f)
xy+z=O + w= 0 x+z+w=O
Y  2z
3x+y+z=O x  Y + z  2w = 0 xz+w=O x + y  3w = 0
II, §4. Row Operations and Gauss Elimination Consider the system of linear equations
3x  2y x
+
2x 
+ z + 2w == 1, z
y y
+ 3z
w
==  2, == 4.
The matrix of coefficients is
2 1
1
1 1 3
D·
[II, §4]
71
ROW OPERATIONS AND GAUSS ELIMINATION
By the augmented matrix we shall mean the matrix obtained by inserting the column
(!) as a last column, so the augmented matrix
2
G
1
1 1 3
2 1 0
IS
i).
In general, let AX = B be a system of m linear equations In n unknowns, which we write in full: aIIx l a 2I x i
+ ... + aInX n = b l , + ... + a 2n X n = b2 ,
Then we define the augmented matrix to be the m by n
+1
matrix:
In the examples of homogeneous linear equations of the preceding section, you will notice that we performed the following operations, called elementary row operations: Multiply one equation by a nonzero number. Add one equation to another. Interchange two equations. These operations are reflected in operations on the augmented matrix of coefficients, which are also called elementary row operations: Multiply one row by a nonzero number. Add one row to another. Interchange two rows. Suppose that a system of linear equations is changed by an elementary row operation. Then the solutions of the new system are exactly the
72
MATRICES AND LINEAR EQUATIONS
[II, §4]
same as the solutions of the old system. By making row operations, we can hope to simplify the shape of the system so that it is easier to find the solutions. Let us define two matrices to be row equivalent if one can be obtained from the other by a succession of elementary row operations. If A is the matrix of coefficients of a system of linear equations, and B the column vector as above, so that (A, B)
is the augmented matrix, and if (A', B') is rowequivalent to (A, B) then the solutions of the system AX == B
are the same as the solutions of the system A'X == B'.
To obtain an equivalent system (A', B') as simple as possible we use a method which we first illustrate in a concrete case. Example. Consider the augmented matrix in the above example. We have the following row equivalences:
2
1
2
1 1
1 3
1 0
G
!)
Subtract 3 times second row from first row
5
G
1 1
4 1
3
5 1 0
!)
Subtract 2 times second row from third row
5
(!
1
4 1
1
3
5
2
5
Interchange first and second row; multiply second row by  1.
1
(~
1 5
 1 4
 1 5
3
5
2
 2) 7
8
[II, §4]
ROW OPERATIONS AND GAUSS ELIMINATION
73
Multiply second row by 3; mUltiply third row by 5.
1 1 1 1 2) (oo 15 12 15 21 15 25 10 40
Add second row to third row.
(~
1 12 13
1
15
o
1 15 5
2)
21 19
What we have achieved is to make each successive row start with a nonzero entry at least one step further than the preceding row. This makes it very simple to solve the equations. The new system whose augmented matrix is the matrix obtained last can be written in the form: x
+y
z

w =  2,
15y  12z  15w 13z 
5w
=
21,
=
19.
This is now in a form where we can solve by giving w an arbitrary value in the third equation, and solve for z from the third equation. Then we solve for y from the second, and x from the first. With the formulas, this gIves:
z=
19
+ 5w 13
21 y
+ 12z + 15w
=
x =  1 y
15
+ z + w.
We can give w any value to start with, and then determine values for x, y, z. Thus we see that the solutions depend on one free parameter. Later we shall express this property by saying that the set of solutions has dimension 1. For the moment, we give a general name to the above procedure. Let M be a matrix. We shall say that M is in row echelon form if it has the following property:
Whenever two successive rows do not consist entirely of zeros, then the second row starts with a nonzero entry at least one step further to the right than the first row. All the rows consisting entirely of zeros are at the bottom of the matrix. In the previous example we transformed a matrix into another which IS in row echelon form. The nonzero coefficients occurring furthest to
74
MATRICES AND LINEAR EQUATIONS
[II, §4]
the left in each row are called the leading coefficients. In the above example, the leading coefficients are 1, 15, 13. One may perform one more change by dividing each row by the leading coefficient. Then the above matrix is row equivalent to
(~
1 1
1
s
o
1
;)
1 1
4
5
13
5 19
.
13
In this last matrix, the leading coefficient of each row is equal to 1. One could make further row operations to insert further zeros, for instance subtract the second row from the first, and then subtract % times the third row from the second. This yields:
o
7
6
S
S 1
0 1
1
o
S
2 )
1  ~~ .
+ 132
19
5
13
13
Unless the matrix is rigged so that the fractions do not look too horrible, it is usually a pain to do this further row equivalence by hand, but a machine would not care. Example. The following matrix is in row echelon form.
o o o o
2
o o o
3 0 0 0
417
5
24 o 3 1 000
Suppose that this matrix is the augmented matrix of a system of linear equations, then we can solve the linear equations by giving some varIables an arbitrary value as we did. Indeed, the equations are:
2y  3z
+ 4w + t = 5w + 2t =  3t
=
7, 4, 1.
Then the solutions are t = 1/3, W=
4  2t 5
z
any arbitrarily given value,
=
y= x
7
+ 3z 
4w  t
2
= any arbitrarily given value.
[II, §4]
ROW OPERATIONS AND GAUSS ELIMINATION
75
The method of changing a matrix by row equivalences to put it In row echelon form works in general. Theorem 4.1. Every matrix is row equivalent to a matrix in row echelon form.
Proof Select a nonzero entry furthest to the left in the matrix. If this entry is not in the first column, this means that the matrix consists entirely of zeros to the left of this entry, and we can forget about them. So suppose this nonzero entry is in the first column. After an interchange of rows, we can find an equivalent matrix such that the upper lefthand corner is not O. Say the matrix is all
a 2l
a 12 a 22
a ln a 2n
and all =1= O. We multiply the first row by a 2l /a ll and subtract from the second row. Similarly, we multiply the first row by ail/all and subtract it from the ith row. Then we obtain a matrix which has zeros in the first column except for all. Thus the original matrix is row equivalent to a matrix of the form
We then repeat the procedure with the smaller matrix
We can continue until the matrix is In row echelon form (formally by induction). This concludes the proof. Observe that the proof is just another way of formulating the elimination argument of §3. We give another proof of the fundamental theorem: Theorem 4.2. Let
76
MA TRICES AND LINEAR EQUATIONS
[II, §4]
be a system of m homogeneous linear equations in n unknowns with n > m. T hen there exists a nontrivial solution. Proof Let A = (a ij ) be the matrix of coefficients. Then A IS equIvalent to A' in row echelon form: ak1x k1
+ Sk/X) ak2 x k2 + Sk2(X)
= 0, =
0,
where a k1 =1= 0, ... ,akr =1= 0 are the nonzero coefficients of the variables occurring furthest to the left in each successive row, and Ski (X), ... ,Skr(X) indicate sums of variables with certain coefficients, but such that if a variable Xj occurs in Sk/X), then j > kl and similarly for the other sums. If Xj occurs in Ski then j > k i • Since by assumption the total number of variables n is strictly greater than the number of equations, we must have r < n. Hence there are n  r variables other than Xk1 ' ... ,xkr and n  r > O. We give these variables arbitrary values, which we can of course select not all equal to O. Then we solve for the variables x kr' Xkr _ l ' ••• ,Xk1 starting with the bottom equation and working back up, for instance Xkr
= 
Skr(x)/a kr ,
Xkr _ 1
= 
Sk r _l(x)/a kr _1,
and so forth.
This gives us the nontrivial solution, and proves the theorem. Observe that the pattern follows exactly that of the examples, but with a notation dealing with the general case.
Exercises II, §4 In each of the following cases find a row equivalent matrix in row echelon form.
1. (a) (
2. (a)
4
3 1
1
2
6
~ 2
1 1
4)
6 5
~
i)
1 1
4
3
2
3
2)
3 . 1
[II, §5]
3. (a)
77
ROW OPERATIONS AND ELEMENTARY MATRICES
(i
2
1
2
3
1
4
1
11
6
2
2 6
5 3
5
4. Write down the coefficient matrix of the linear equations of Exercise 5 in §3, and in each case give a row equivalent matrix in echelon form. Solve the linear equations in each case by this method.
II, §5. Row Operations and Elementary Matrices Before reading this section, work out the numerical examples given in Exercises 33 through 37 of §2. The row operations which we used to solve linear equations can be represented by matrix operations. Let 1 < r < m and 1 < s < m. Let Irs be the square m x m matrix which has component 1 in the rs place, and o elsewhere:
Irs =
0·········0 . . 0···1 rs ... 0 0······ ···0
Let A = (aij) be any m x n matrix.
What
IS
the effect of mUltiplying
IrsA?
r{
0···· ·····0 0···1 rs ···0
a 11 ... a In
s
as1 ... asn
S
=
as1 ... asn
Jr.
0 ···0
0······ ···0 ~
l
0 ···0
amI'" amn
The definition of mUltiplication of matrices shows that I rsA is the matrix obtained by putting the sth row of A in the rth row, and zeros elsewhere. If r = s then Irr has a component 1 on the diagonal place, and 0 elsewhere. Multiplication by I rr then leaves the rth row fixed, and replaces all the other rows by zeros.
78
[II, §5]
MATRICES AND LINEAR EQUATIONS
If r # s let
Then
Then IrsA puts the sth row of A in the rth place, and IsrA puts the rth row of A in the sth place. All other rows are replaced by zero. Th us J rs interchanges the rth row and the sth row, and replaces all other rows by zero. Example. Let
1
o o
~)
and
A=
3 2 1)
(2
1
4
2.
5
1
If you perform the matrix multiplication, you will see directly that J A interchanges the first and second row of A, and replaces the third row by zero. On the other hand, let
E=(!
1
o o
Then EA is the matrix obtained from A by interchanging the first and second row, and leaving the third row fixed. We can express E as a sum:
where Irs is the matrix which has rscomponent 1, and all other components 0 as before. Observe that E is obtained from the unit matrix by interchanging the first two rows, and leaving the third row unchanged. Thus the operation of interchanging the first two rows of A is carried out by mUltiplication with the matrix E obtained by doing this operation on the unit matrix. This is a special case of the following general fact. Theorem 5.1. Let E be the matrix obtained from the unit n x n matrix by interchanging two rows. Let A be an n x n matrix. Then EA is the matrix obtained from A by interchanging these two rows.
[II, §5]
ROW OPERATIONS AND ELEMENTARY MATRICES
79
Proof The proof is carried out according to the pattern of the example, it is only a question of which symbols are used. Suppose that we interchange the rth and sth row. Then we can write
E
=
Irs
+ Isr + sum
of the matrices I jj with j 1= r, j 1= s.
Thus E differs from the unit matrix by interchanging the rth and sth rows. Then
with j 1= r, j 1= s. By the previous discussion, this is precisely the matrix obtained by interchanging the rth and sth rows of A, and leaving all the other rows unchanged. The same type of discussion also yields the next result. Theorem 5.2. Let E be the matrix obtained from the unit n x n matrix by multiplying the rth row with a number c and adding it to the sth row, r =1= s. Let A be an n x n matrix. Then EA is obtained from A by multiplying the rth row of A by c and adding it to the sth row of A. Proof We can write
E= I
+ cI sr .
Then EA = A + cIsrA. We know that IsrA puts the rth row of A in the sth place, and multiplication by c mUltiplies this row by c. All other rows besides the sth row in cI srA are equal to O. Adding A + cIsrA therefore has the effect of adding c times the rth row of A to the sth row of A, as was to be shown. Example. Let 1 0
E=
0
1
0 0
4 0 0
0
0 1
0
0
1
0
Then E is obtained from the unit matrix by adding 4 times the third row to the first row. Take any 4 x n matrix A and compute EA. You will find that EA is obtained by multiplying the third row of A by 4 and adding it to the first row of A.
80
MATRICES AND LINEAR EQUATIONS
[II, §5]
More generally, we can let Ers{c) for r =I s be the elementary matrix. Ers{c) r
= I + cIrs.
~
s
1
0
0
0
0
1
0
0
0
c
1
0
0
0
0
1
It differs from the unit matrix by having rscomponent equal to c. The effect of multiplication on the left by Ers{c) is to add c times the sth row to the rth row. By an elementary matrix, we shall mean anyone of the following three types: (a) A matrix obtained from the unit matrix by multiplying the rth diagonal component with a, number c =I O. (b) A matrix obtained from the unit matrix by interchanging two rows (say the rth and sth row, r =I s). (c) A matrix Ers{c) = I + cIrs with r =I shaving rscomponent c for r =I s, and all other components 0 except the diagonal components which are equal to 1. These three types reflect the row operations discussed in the preceding section. Multiplication by a matrix of type (a) mUltiplies the rth row by the number c. Multiplication by a matrix of type (b) interchanges the rth and sth row. Multiplication by a matrix of type (c) adds c times the sth row to the rth row.
Proposition 5.3. An elementary matrix is invertible.
Proof For type (a), the inverse matrix has rth diagonal component c  1, because multiplying a row first by c and then by c  1 leaves the row unchanged. For type (b), we note that by interchanging the rth and sth row twice we return to the same matrix we started with. F'or type (c), as in Theorem 5.2, let E be the matrix which adds c times the sth row to the rth row of the unit matrix. Let D be the matrix which adds  c times the sth row to the rth row of the unit
[II, §5]
81
ROW OPERATIONS AND ELEMENTARY MATRICES
s). Then DE IS the unit matrix, and so IS ED, so E IS
matrix (for r invertible.
=1=
Example. other:
The following elementary matrices are Inverse to each
E=
1 0 0 0
0 1 0 0
4 0
0 0 0
E 1 =
0
1 0 0 0
0 1 0 0
4 0 1
0 0 0
0
We shall find an effective way of finding the inverse of a square matrix if it has one. This is based on the following properties.
If A, B are square matrices of the same size and have inverses, then so does the product AB, and
This is immediate, because
Similarly, for any number of factors: Proposition 5.4. If A 1, ... ,Ak are invertible matrices of the same size, then their product has an inverse, and
Note that in the righthand side, we take the product of the Inverses In reverse order. Then
A 1 ···AA1···A1I k k 1because we can collapse AkA;1 to I, then A k_ 1A;_11 to I and so forth. Since an elementary matrix has an inverse, we conclude that any product of elementary matrices has an inverse. Proposition 5.5. Let A be a square matrix, and let A' be row equivalent to A. Then A has an inverse if and only if A' has an inverse.
Proof There exist elementary matrices E b
...
,Ek such that
Suppose that A has an inverse. Then the righthand side has an inverse by Proposition 5.4 since the righthand side is a product of invertible matrices. Hence A' has an inverse. This proves the proposition.
82
MATRICES AND LINEAR EQUATIONS
[II, §5]
We are now in a position to find an inverse for a square matrix A if it has one. By Theorem 4.1 we know that A is row equivalent to a matrix A' in echelon form. If one row of A' is zero, then by the definition of echelon form, the last row must be zero, and A' is not invertible, hence A is not invertible. If all the rows of A' are nonzero, then A' is a triangular matrix with nonzero diagonal components. It now suffices to find an inverse for such a matrix. In fact, we prove: and only if A is row equivalent to the unit matrix. Any upper triangular matrix with nonzero diagonal elements is invertible.
Theorem 5.6. A square matrix A is invertible
if
Proof. Suppose that A is row equivalent to the unit matrix. Then A is invertible by Proposition 5.5. Suppose that A is invertible. We have just seen that A is row equivalent to an upper triangular matrix with nonzero elements on the diagonal. Suppose A is such a matrix: all
a 12
a l"
o a22
a 2n
By assumption we have all··· ann i= O. We multiply the ith row with aii 1. We obtain a triangular matrix such that all the diagonal components are equal to 1. Thus to prove the theorem, it suffices to do it in this case, and we may assume that A has the form
o
0
1
We multiply the last row by ain and subtract it from the ith row for i = 1, ... ,n  1. This makes all the elements of the last column equal to o except for the lower righthand corner, which is 1. We repeat this procedure with the next to the last row, and continue upward. This means that by row equivalences, we can replace all the components which lie strictly above the diagonal by o. We then terminate with the unit matrix, which is therefore row equivalent with the original matrix. This proves the theorem. Corollary 5.7. Let A be an invertible matrix. Then A can be expressed as a product of elementary matrices.
[II, §5]
ROW OPERATIONS AND ELEMENTARY MATRICES
83
Proof This is because A is row equivalent to the unit matrix, and row operations are represented by multiplication with elementary matrices, so there exist E l' ... ,Ek such that
Then A = Ell ... Ei: 1, thus proving the corollary. When A is so expressed, we also get an expression for the inverse of A, namely
The elementary matrices E 1 , ••• ,Ek are those which are used to change A to the unit matrix.
Example. Let
A
=(~
3
o
1
1
o
o
We want to find an inverse for A. We perform the following row operations, corresponding to the multiplication by elementary matrices as shown. Interchange first two rows.
1 1)
3
o
1, 1
(O~
1
0 0
Subtract 2 times first row from second row. Subtract 2 times first row from third row.
(~
1
5 2
1)
3 , 3
~}
1
G
2 2
Subtract 2/5 times second row from third row.
11)
2
o
6/5
5
3, 9/5
1
84
[II, §5]
MATRICES AND LINEAR EQUATIONS
Subtract 5/3 of third row from second row. Add 5/9 of third row to first row.
G
0
(2/9 5/3
0)o ,
1 5
2/5
9/5
1/3 0 6/5
5/9)
~/3
.
Add 1/5 of second row to first row.
(~
( 5/3 1/9
0)o ,
0
5 0
2/5
9/5
1/3 0 6/5
2/9)
~/3
.
Multiply second row by 1/5. Multiply third row by 5/9.
(~ Then A 
1
(1/3 1/9
~}
0
1 0
2/9
1/3 0
2/3
2/9) 1/3 . 5/9
is the matrix on the right, that is
A 1
1/9 1/3
=
( 2/9
1/3
o 2/3
2/9) 1/3 . 5/9
You can check this by direct multiplication with A to find the unit matrix. If A is a square matrix and we consider an inhomogeneous system of linear equations AX=B,
then we can use the inverse to solve the system, if A is invertible. Indeed, in this case, we multiply both sides on the left by A  1 and we find
This also proves:
Proposition 5.S. Let AX
B be a system of n linear equations in n unknowns. Assume that the matrix of coefficients A is invertible. Then there is a unique solution X to the system, and =
[II, §6]
85
LINEAR COMBINATIONS
Exercises II, §5 1. Using elementary row operations, find inverses for the following matrices. (a)
(~
(c) (
2 1 0
(e)
:)
3
(~
4
3 2 5
0 7
~)
(b) (_ ~
1
2
4
(d)
(~
2 2
n
:)
2
(f) (_:
~) 5
2
:)
Note: For another way of finding inverses, see the chapter on determinants. 2. Let r =F s. Show that
I;s =
O.
3. Let r =F s. Let Ers(c) = I + cl rs ' Show that E rs (c )Ers (c') = Ers( c + c').
II, §6. Linear Combinations Let AI, ... ,An be mtuples in Rm. Let
XI""'Xn
be numbers. Then we call
a linear combination of A I, ... ,An; and we call Xl"" ,xn the coefficients of the linear combination. A similar definition applies to a linear combination of row vectors. The linear combination is called nontrivial if not all the coefficients Xl' ... ,xn are equal to O. Consider once more a system of linear homogeneous equations
Our system of homogeneous equations can also be written in the form a 12
all a 2l Xl amI
+ X2
a2 2
+ ... + Xn

0 0 0
86
MATRICES AND LINEAR EQUATIONS
[II, §6]
or more concisely:
where A 1, ... ,A n are the column vectors of the matrix of coefficients, which is A = (aij)' Thus the problem of finding a nontrivial solution for the system of homogeneous linear equations is equivalent to finding a nontrivial linear combination of A 1, ... ,An which is equal to O. Vectors A 1, ... ,An are called linearly dependent if there exist numbers x l' ... ,x n not all equal to 0 such that
Thus a nontrivial solution (x l' ... ,x n ) is an ntuple which gives a linear combination of A 1, ... ,An equal to 0, i.e. a relation of linear dependence between the columns of A. We may thus summarize the description of the set of solutions of the system of homogeneous linear equations in a table. (a) It consists of those vectors X giving linear relations
between the columns of A. (b) It consists of those vectors X perpendicular to the rows of A, that is X· Ai = 0 for all i. (c) It consists of those vectors X such that AX = O. Vectors A 1, ... ,A n are called linearly independent if, gIven any linear combination of them which is equal to 0, i.e.
then we must necessarily have x j = 0 for all j = 1, ... ,no This means that there is no nontrivial relation of linear dependence among the vectors A 1, ... ,An. Example. The standard unit vectors E1 = (1, 0, ... ,0), ... ,En = (0, ... ,0, 1) of Rn are linearly independent. Indeed, let x l ' ... 'X n be numbers such that
[II, §6]
LINEAR COMBINATIONS
87
The lefthand side is just the ntuple (Xl' ... ,X n ). If this ntuple is 0, then all components are 0, so Xi = 0 for all i. This proves that E b ... ,En are linearly independent. We shall study the notions of linear dependence and independence more systematically in the next chapter. They were mentioned here just to have a complete table for the three basic interpretations of a system of linear equations, and to introduce the notion in a concrete special case before giving the general definitions in vector spaces.
Exercise II, §6 1. (a) Let A = (a ij ), B = (b jk ) and let AB = C with C = (C ik ). Let C k be the kth column of C. Express Ck as a linear combination of the columns of A. Describe precisely which are the coefficients, coming from the matrix B. (b) Let AX = C k where X is some column of B. Which column is it?
CHAPTER III
Vector Spaces
As usual, a collection of objects will be called a set. A member of the collection is also called an element of the set. It is useful in practice to use short symbols to denote certain sets. For instance we denote by R the set of all numbers. To say that "x is a number" or that "x is an element of R" amounts to the same thing. The set of ntuples of numbers will be denoted by Rn. Thus" X is an element of R n" and" X is an ntuple" mean the same thing. Instead of saying that u is an element of a set S, we shall also frequently say that u lies in S and we write u E S. If Sand S' are two sets, and if every element of S' is an element of S, then we say that S' is a subset of S. Thus the set of rational numbers is a subset of the set of (real) numbers. To say that S is a subset of S' is to say that S is part of S'. To denote the fact that S is a subset of S', we write S c S'. If S b S 2 are sets, then the intersection of S 1 and S 2, denoted by S1 n S2, is the set of elements which lie in both S1 and S2. The union of S 1 and S 2, denoted by S 1 U S 2, is the set of elements which lie in S 1 or S2·
III, §1. Definitions In mathematics, we meet several types of objects which can be added and multiplied by numbers. Among these are vectors (of the same dimension) and functions. It is now convenient to define in general a notion which includes these as a special case. A vector space V is a set of objects which can be added and multiplied by numbers, in such a way that the sum of two elements of V is
[III, §l]
89
DEFINITIONS
again an element of V, the product of an element of V by a number is an element of V, and the following properties are satisfied: VS 1. Given the elements u, v, w of V, we have
(u
+ v) + w =
u
+ (v + w).
VS 2. There is an element of V, denoted by 0, such that
for all elements u of V. VS 3. Given an element u of V, the element { 1)u is such that
u
+ (l)u
=
o.
VS 4. F or all elements u, v of V, we have
u+v VS 5. If c is a number, then c{u
=
v + u.
+ v) =
VS 6. If a, b are two numbers, then {a
cu
+ cv.
+ b)v =
av
+ bv.
VS 7. If a, b are two numbers, then (ab)v = a{bv). VS 8. For all elements u of V, we have 1· u = u (1 here is the number one).
We have used all these rules when dealing with vectors, or with functions but we wish to be more systematic from now on, and hence have made a list of them. Further properties which can be easily deduced from these are given in the exercises and will be assumed from now on. The algebraic properties of elements of an arbitrary vector space are very similar to those of elements of R2, R 3 , or Rn. Consequently it is customary to call elements of an arbitrary vector space also vectors. If u, v are vectors (i.e. elements of the arbitrary vector space V), then the sum u + {1)v is usually written u  v. We also write  v instead of (  1)v. Example 1. Fix two positive integers m, n. Let V be the set of all m x n matrices. We also denote V by Mat{m x n). Then V is a vector
90
[III, §1]
VECTOR SPACES
space. It is easy to verify that all properties VS 1 through VS 8 are satisfied by our rules for addition of matrices and multiplication of matrices by numbers. The main thing to observe here is that addition of matrices is defined in terms of the components, and for the addition of components, the conditions analogous to VS 1 through VS 4 are satisfied. They are standard properties of numbers. Similarly, VS 5 through VS 8 are true for multiplication of matrices by numbers, because the corresponding properties for the multiplication of numbers are true. Example 2. Let V be the set of all functions defined for all numbers. If f, g are two functions, then we know how to form their sum f + g. It is the function whose value at a number t is f(t) + g(t). We also know how to multiply f by a number c. It is the function cf whose values at a number t is cf(t). In dealing with functions, we have used properties VS 1 through VS 8 many times. We now realize that the set of functions is a vector space. The function f such that f(t) = 0 for all t is the zero function. We emphasize the condition for all t. If a function has some of its values equal to zero, but other values not equal to 0, then it is not the zero function. In practice, a number of elementary properties concerning addition of elements in a vector space are obvious because of the concrete way the vector space is given in terms of numbers, for instance as in the previous two examples. We shall now see briefly how to prove such properties just from the axioms. It is possible to add several elements of a vector space. Suppose we wish to add four elements, say u, v, w, z. We first add any two of them, then a third, and finally a fourth. Using the rules VS 1 and VS 4, we see that it does not matter in which order we perform the additions. This is exactly the same situation as we had with vectors. For example, we have (u
+ v) + w) + z = = =
+ (v + w») + z (v + w) + u) + z (v + w) + (u + z), (u
etc.
Thus it is customary to leave out the parentheses, and write simply
u
+ v + w + z.
The same remark applies to the sum of any number n of elements of V. We shall use 0 to denote the number zero, and 0 to denote the element of any vector space V satisfying property VS 2. We also call it
[III, §1]
91
DEFINITIONS
zero, but there is never any possibility of confusion. We observe that this zero element 0 is uniquely determined by condition VS 2. Indeed, if v
+ w=
v
then adding  v to both sides yields v
+v+
w
=
v
+ v = 0,
and the lefthand side is just 0 + w = w, so w = Observe that for any element v in V we have Ov =
o.
o.
Proof
o = v + (
l)v = (1  l)v
=
Ov.
Similarly, if c is a number, then cO =
Proof We have cO to get cO =
=
c(O
+ 0) =
o. cO
+ cO.
Add  cO to both sides
o.
Subspaces Let V be a vector space, and let W be a subset of V. Assume that W satisfies the following conditions. (i)
If v, ware elements of W, their sum v
+
w is also an element of
W (ii)
If v is an element of Wand c a number, then cv is an element of
W (iii)
The element 0 of V is also an element of W
Then W itself is a vector space. Indeed, properties VS 1 through VS 8, being satisfied for all elements of V, are satisfied also for the elements of W. We shall call W a subspace of V. Example 3. Let V = Rn and let W be the set of vectors in V whose last coordinate is equal to O. Then W is a subspace of V, which we could identify with Rn  1. Example 4. Let A be a vector in R3. Let W be the set of all elements B in R3 such that B· A = 0, i.e. such that B is perpendicular to A. Then W is a subspace of R3. To see this, note that O· A = 0, so that 0 is in W Next, suppose that B, C are perpendicular to A. Then (B
+ C)· A
= B· A
+ C· A
=
0,
92 so that B
VECTOR SPACES
+C
is also perpendicular to A. Finally, if x is a number, then (xB)·A
so that xB R3.
IS
[III, §1]
= x(B·A) = 0,
perpendicular to A. This proves that W is a subspace of
More generally, if A is a vector in Rn, then the set of all elements B in Rn such that B· A = 0 is a subspace of Rn. The proof is the same as when n = 3. Example 5. Let Sym(n x n) be the set of all symmetric n x n matrices. Then Sym(n x n) is a subspace of the space of all n x n matrices. Indeed, if A, B are symmetric and c is a number, then A + B and cA are symmetric. Also the zero matrix is symmetric. Example 6. If f, g are two continuous functions, then f + g is continuous. If c is a number, then cf is continuous. The zero function is continuous. Hence the continuous functions form a subspace of the vector space of all functions. If f, g are two differentiable functions, then their sum f + g is differentiable. If c is a number, then cf is differentiable. The zero function is differentiable. Hence the differentiable functions form a subspace of the vector space of all functions. Furthermore, every differentiable function is continuous. Hence the differentiable functions form a subspace of the vector space of continuous functions. Example 7. Let V be a vector space and let U, W be subspaces. We denote by U n W the intersection of U and W, i.e. the set of elements which lie both in U and W. Then U n W is a subspace. For instance, if U, Ware two planes in 3space passing through the origin, then in general, their intersection will be a straight line passing through the origin, as shown in Fig. 1.
[III, §2]
LINEAR COMBINATIONS
93
Example 8. Let U, W be subspaces of a vector space V. By
U+W we denote the set of all elements u + w with U E U and WE W. Then we leave it to the reader to verify that U + W is a subspace of V, said to be generated by U and W, and called the sum of U and W.
Exercises III, §1 1. Let A l ' ... ,A, be vectors in Rn. Let W be the set of vectors B in Rn such that B· Ai = 0 for every i = 1, ... ,r. Show that W is a subspace of Rn. of elements in R2 form subspaces. that x = y. that x  y = o. that x + 4y = O.
2. Show that the (a) The set of (b) The set of (c) The set of
following sets all (x, y) such all (x, y) such all (x, y) such
3. Show that the (a) The set of (b) The set of (c) The set of
following sets of elements in R 3 form subspaces. all (x, y, z) such that x + y + z = O. all (x, y, z) such that x = y and 2y = z. all (x, y, z) such that x + y = 3z.
4. If U, Ware subspaces of a vector space V, show that U n Wand U subspaces.
+ Ware
5. Let V be a subspace of Rn. Let W be the set of elements of R n which are perpendicular to every element of V. Show that W is a subspace of Rn. This subspace W is often denoted by V 1., and is called V perp, or also the orthogonal complement of V.
III, §2. Linear Combinations Let V be a vector space, and let v 1 , ••• ,V n be elements of V. We shall say that V 1 , ••. ,Vn generate V if given an element v E V there exist numbers Xl' ... ,xn such that
Example 1. Let E 1 , ••• ,En be the standard unit vectors in R n , so Ei has component 1 in the ith place, and component 0 in all other places.
94
[III, §2]
VECTOR SPACES
Then E l ,
...
,En generate Rn. Proof: gIven X
= (Xl' ... ,xn) ERn.
Then
n
X =
L xiE i, i= 1
so there exist numbers satisfying the condition of the definition. Let V be an arbitrary vector space, and let V l , ... ,V n be elements of V. Let Xl' ... ,xn be numbers. An expression of type
is called a linear combination of v l , ... ,V n • The numbers then called the coefficients of the linear combination.
Xl'." ,X n
are
The set of all linear combinations of Vl' ... ,Vn is a subspace of V. Proof Let W be the set of all such linear combinations. Let Yl'··· ,Yn be numbers. Then
(X 1 V1
+ ... + Xn vn) + (y 1 V1 + . .. + Y nvn) = (Xl + Yl)V l + ... + (xn + Yn)v n·
Thus the sum of two elements of W is again an element of W, i.e. a linear combination of Vl' ... ,Vn. Furthermore, if c is a number, then
is a linear combination of Finally,
V l , ... ,V n ,
and hence is an element of W
o == OV l + ... + OV n is an element of W This proves that W is a subspace of V. The subspace W consisting of all linear combinations of called the subspace generated by V l ,.·. ,Vn •
V l , ... ,V n
IS
Example 2. Let V l be a nonzero element of a vector space V, and let w be any element of V. The set of elements with
tER
[III, §2]
95
LINEAR COMBINATIONS
is called the line passing through w in the direction of V 1 • We have already met such lines in Chapter I, §5. If w = 0, then the line consisting of all scalar multiples tV 1 with t E R is a subspace, generated by V1 • Let VI' v2 be elements of a vector space V, and assume that neither is a scalar multiple of the other. The subspace generated by V 1 , V 2 is called the plane generated by VI' V 2 • It consists of all linear combinations with
t 1, t2
arbitrary numbers.
This plane passes through the origin, as one sees by putting
t1
=
t2
=
o.
Plane passing through the origin Figure 2
We obtain the most general notion of a plane by the following operation. Let S be an arbitrary subset of V. Let P be an element of V. If we add P to all elements of S, then we obtain what is called the translation of S by P. It consists of all elements P + V with V in S. Example 3. Let V 1 , V 2 be elements of a vector space V such that neither is a scalar multiple of the other. Let P be an element of V. We define the plane passing through P, parallel to V 1 , V 2 to be the set of all elements
where t 1 , t2 are arbitrary numbers. This notion of plane is the analogue, with two elements VI' v2 , of the notion of parametrized line considered in Chapter I. Warning. Usually such a plane does not pass through the orIgIn, as shown on Fig. 3. Thus such a plane is not a subspace of V. If we take P = 0, however, then the plane is a subspace.
96
[III, §2]
VECTOR SPACES
o~
________________
Plane not passing through the origin Figure 3
Sometimes it is interesting to restrict the coefficients of a linear combination. We give a number of examples below.
Example 4. Let V be a vector space and let v, u be elements of V. We define the line segment between v and v + u to be the set of all points v + tu,
0 < t < 1.
This line segment is illustrated in the following picture.
v+u v+tu
Figure 4
v
For instance, if t = !, then v + !u is the point midway between v and v + u. Similarly, if t = t, then v + tu is the point one third of the way between v and v + u (Fig. 5).
v+u
v+u v+!u v+iu v+tu v
v
(b)
(a)
Figure 5
[III, §2]
97
LINEAR COMBINATIONS
If v, ware elements of V, let u = w  v. Then the line segment between v and w is the set of all points v + tu, or
v + t(w  v),
o< t
O. Show that the following pairs of functions are linearly independent. (a) t, lit (b) et, log t 9. What are the coordinates of the function 3 sin t to the basis {sin t, cos t}?
+5
cos t = f(t) with respect
10. Let D be the derivative dldt. Let J(t) be as in Exercise 9. What are the coordinates of the function DJ(t) with respect to the basis of Exercise 9? In each of the following cases, exhibit a basis for the given space, and prove that it is a basis. 11. The space of 2 x 2 matrices. 12. The space of m x n matrices. 13. The space of n x n matrices all of whose components are 0 except possibly the diagonal components. 14. The upper triangular matrices, i.e. matrices of the following type: all
o
a l2
...
al )
a22
...
a 2n
..
o
0
.
.
ann
15. (a) The space of symmetric 2 x 2 matrices. (b) The space of symmetric 3 x 3 matrices. 16. The space of symmetric n x n matrices.
III, §5. Dimension We ask the question: Can we find three linearly independent elements in R 2 ? For instance, are the elements A = (1, 2),
B
= (
5, 7),
C = (10,4)
linearly independent? If you write down the linear equations expressing the relation
xA
+ yB + zC = 0,
[III, §5]
111
DIMENSION
you will find that you can solve them for x, y, z not equal to O. Namely, these equations are: x  5y + 10z = 0,
2x
+ 7y +
4z
=
O.
This is a system of two homogeneous equations in three unknowns, and we know by Theorem 2.1 of Chapter II that we can find a nontrivial solution (x, y, z) not all equal to zero. Hence A, B, C are linearly dependent. We shall see in a moment that this is a general phenomenon. In Rn, we cannot find more than n linearly independent vectors. Furthermore, we shall see that any n linearly independent elements of Rn must generate Rn, and hence form a basis. Finally, we shall also see that if one basis of a vector space has n elements, and another basis has m elements, then m = n. In short, two bases must have the same number of elements. This property will allow us to define the dimension of a vector space as the number of elements in any basis. We now develop these ideas systematically. Theorem 5.1. Let V be a vector space, and let {VI' ... ,Vm} generate V. Let WI' ... ,wn be elements of V and assume that n > m. Then WI'· .. 'W n
are linearly dependent. Proof Since {VI' ... ,Vm} generate V, there exist numbers (a ij ) such that we can write
If
Xl' ... ,X n XIW I
are numbers, then
+ ... + XnWn = (Xl all + ... + xnaln)V I + ... + (Xl amI + ... + Xnamn)V m
(just add up the coefficients of VI' ... ,Vm vertically downward). According to Theorem 2.1 of Chapter II, the system of equations xla ll
.
+ ... + xna ln =.
0
has a nontrivial solution, because n > m. In VIew of the preceding remark, such a solution (Xl' ... ,X n ) is such that
as desired.
112
[III, §5]
VECTOR SPACES
Theorem 5.2. Let V be a vector space and suppose that one basis has n elements, and another basis has m elements. Then m
= n.
Proof We apply Theorem 5.1 to the two bases. Theorem 5.1 implies that both alternatives n > m and m > n are impossible, and hence m = n.
Let V be a vector space having a basis consisting of n elements. We shall say that n is the dimension of V. If V consists of 0 alone, then V does not have a basis, and we shall say that V has dimension O. We may now reformulate the definitions of a line and a plane in an arbitrary vector space V. A line passing through the origin is simply a onedimensional subspace. A plane passing through the origin is simply a twodimensional subspace. An arbitrary line is obtained as the translation of a onedimensional subspace. An arbitrary plane is obtained as the translation of a twodimensional subspace. When a basis {VI} has been selected for a onedimensional space, then the points on a line are expressed in the usual form P
+ t1v 1 with
all possible numbers t 1 •
When a basis {v 1 , v2 } has been selected for a twodimensional space, then the points on a plane are expressed in the form
Let {VI' ... ,vn } be a set of elements of a vector space V. Let r be a positive integer < n. We shall say that {v 1 , ••• ,vr } is a maximal subset of linearly independent elements if V 1 , ••• ,Vr are linearly independent, and if in addition, given any Vi with i > r, the elements V 1 , ... ,Vr , Vi are linearly dependent. The next theorem gives us a useful criterion to determine when a set of elements of a vector space is a basis. Theorem 5.3. Let {V l' ... ,vn } be a set of generators of a vector space V. Let {v 1 , ••• ,vr } be a maximal subset of linearly independent elements. Then {v 1 , .•• ,vr } is a basis of v. Proof We must prove that v 1, ••.
that each sis, given
Vi
Vb
generate V. We shall first prove (for i > r) is a linear combination of V 1 , .•• ,Vr • By hypothethere exists numbers Xl' ... ,Xr , Y not all 0 such that ,V r
[III, §5]
113
DIMENSION
Furthermore, y =1= 0, because otherwise, we would have a relation of linear dependence for V 1 , ••• ,Vr • Hence we can solve for Vi' namely
thereby showing that Vi is a linear combination of V 1 , ••• ,Vr • Next, let V be any element of V. There exist numbers c 1, ... 'C n such that In this relation, we can replace each Vi (i > r) by a linear combination of V 1 , ••• ,Vr • If we do this, and then collect terms, we find that we have expressed V as a linear combination of Vb ... ,vr • This proves that v1 , ... ,vr generate V, and hence form a basis of V. We shall now give criteria which allow us to tell when elements of a vector space constitute a basis. Let v 1 , ... ,V n be linearly independent elements of a vector space V. We shall say that they form a maximal set of linearly independent elements of V if given any element w of V, the elements w, V 1 , ••• ,V n are linearly dependent. Theorem 5.4. Let V be a vector space, and {v 1, ... ,vn } a maximal set of linearly independent elements of V. Then {v 1 , ••• ,vn } is a basis of v.
Proof We must now show that v 1 , ••• ,Vn generate V, i.e. that every element of V can be expressed as a linear combination of V 1 , ••• ,vn • Let w be an element of V. The elements w, V 1 , ••• ,Vn of V must be linearly dependent by hypothesis, and hence there exist numbers xo, X b ... ,Xn not all such that
°
We cannot have Xo = 0, because if that were the case, we would obtain a relation of linear dependence among Vb". ,vn • Therefore we can solve for w in terms of V 1 , ••• ,Vn , namely W
=  Xl Xo
V1 
••• 
Xn

Xo
This proves that w is a linear combination of {V1' ... ,vn } is a basis.
Vn •
V 1 , ••• ,Vn ,
and hence that
Theorem 5.5. Let V be a vector space of dimension n, and let V 1, ... ,Vn be linearly independent elements of V. Then V 1 , ••• ,vn constitute a basis of v.
114
[III, §5]
VECTOR SPACES
Proof According to Theorem 5.1., {VI' ... ,V n} is a maximal set of linearly independent elements of V. Hence it is a basis by Theorem 5.4. Theorem 5.6. Let V be a vector space of dimension n and let W be a subspace, also of dimension n. Then W = V.
Proof. A basis for W must also be a basis for V. Theorem 5.7. Let V be a vector space of dimension n. Let r be a positive integer with r < n, and let VI' ... ,vr be linearly independent elements of V. Then one can find elements Vr + I' ... ,V n such that
is a basis of
v.
Proof Since r < n we know that {VI' ... ,Vr } cannot form a basis of V, and thus cannot be a maximal set of linearly independent elements of V. In particular, we can find vr + I in V such that
are linearly independent. If r + 1 < n, we can repeat the argument. We can thus proceed stepwise (by induction) until we obtain n linearly independent elements {VI' ... 'Vn}. These must be a basis by Theorem 5.4, and our corollary is proved. Theorem 5.S. Let V be a vector space having a basis conslstlng of n elements. Let W be a subspace which does not consist of 0 alone. Then W has a basis, and the dimension of W is < n.
Proof Let WI be a nonzero element of W. If {w I} is not a maximal set of linearly independent elements of W, we can find an element W2 of W such that WI' W2 are linearly independent. Proceeding in this manner, one element at a time, there must be an integer m n such that we can find linearly independent elements WI' W 2 , ... 'Wm' and suc!'l that
:s
is a maximal set of linearly independent elements of W (by Theorem 5.1 we cannot go on indefinitely finding linearly independent elements, and the number of such elements is at most n). If we now use Theorem 5.4, we conclude that {w1, ... ,w m } is a basis for W.
[III, §6]
THE RANK OF A MATRIX
115
Exercises III, §5 1. What is the dimension of the following spaces (refer to Exercises 11 through 16 of the preceding section): (a) 2 x 2 matrices (b) m x n matrices (c) n x n matrices all of whose components are 0 expect possibly on the diagonal. (d) Upper triangular n x n matrices. (e) Symmetric 2 x 2 matrices. (f) Symmetric 3 x 3 matrices. (g) Symmetric n x n matrices. 2. Let V be a subspace of R2. What are the possible dimensions for V? Show that if V:# R 2 , then either V = {O}, or V is a straight line passing through the origin. 3. Let V be a subspace of R 3 . What are the possible dimensions for V? Show that if V:# R 3 , then either V = {O}, or V is a straight line passing through the origin, or V is a plane passing through the origin.
III, §6. The Rank of a Matrix Let
be an m x n matrix. The columns of A generate a vector space, which is a subspace of Rm. The dimension of that subspace is called the column rank of A. In light of Theorem 5.4, the column rank is equal to the maximum number of linearly independent columns. Similarly, the rows of A generate a subspace of Rn, and the dimension of this subspace is called the row rank. Again by Theorem 5.4, the row rank is equal to the maximum number of linearly independent rows. We shall prove below that these two ranks are equal to each other. We shall give two proofs. The first in this section depends on certain operations on the rows and columns of a matrix. Later we shall give a more geometric proof using the notion of perpendicularity. We define the row space of A to be the subspace generated by the rows of A. We define the column space of A to be the subspace generated by the columns. Consider the following operations on the rows of a matrix.
Row 1. Adding a scalar multiple of one row to another. Row 2. Interchanging rows. Row 3. Multiplying one row by a nonzero scalar.
116
VECTOR SPACES
[III, §6]
These are called the row operations (sometimes, the elementary row operations). We have similar operations for columns, which will be denoted by Coil, Col 2, Col3 respectively. We shall study the effect of these operations on the ranks. First observe that each one of the above operations has an inverse operation in the sense that by performing similar operations we can revert to the original matrix. For instance, let us change a matrix A by adding c times the second row to the first. We obtain a new matrix B whose rows are
If we now add cA 2 to the first row of B, we get back A l . A similar argument can be applied to any two rows. If we interchange two rows, then interchange them again, we revert to the original matrix. If we multiply a row by a number c =1= 0, then multiplying again by c  1 yields the original row. Theorem 6.1. Rowand column operations do not change the row rank of a matrix, nor do they change the column rank. Proof First we note that interchanging rows of a matrix does not affect the row rank since the subspace generated by the rows is the same, no matter in what order we take the rows. Next, suppose we add a scalar multiple of one row to another. We keep the notation before the theorem, so the new rows are
Any linear combination of the rows of B, namely any linear combination of
is also a linear combination of A l , A 2 , ••• ,Am. Consequently the row space of B is contained in the row space of A. Hence by Theorem 5.6, we have row rank of B < row rank of A. Since A is also obtained from B by a similar operation, we get the reverse inequality row rank of A < row rank of B. Hence these two row ranks are equal. Third, if we multiply a row Ai by c =1= 0, we get the new row cA i . But Ai = cl(cA i), so the row spaces of the matrix A and the new matrix
[III, §6]
THE RANK OF A MATRIX
117
obtained by mUltiplying the row by c are the same. Hence the third operation also does not change the row rank. We could have given the above argument with any pair of rows Ai' A j (i =1= j), so we have seen that row operations do not change the row rank.
We now prove that they do not change the column rank. Again consider the matrix obtained by adding a scalar multiple of the second row to the first:
B=
Let B 1 , ••• ,Bn be the columns of this new matrix B. We shall see that the relation of linear dependence between the columns of B are precisely the same as the relations of linear dependence between the columns of A. In other words: A vector X = (x l' ... ,x n ) gives a relation of linear dependence
between the columns of B dependence
if and only if X gives a relation of linear
between the columns of A. Proof We know from Chapter II, §2 that a relation of linear dependence among the columns can be written in terms of the dot product with the rows of the matrix. So suppose we have a relation
This is equivalent with the fact that
X·B·=O I
for
i = 1, ... ,me
Therefore
X· A2 = 0, The first equation can be written
... ,
118
[III, §6]
VECTOR SPACES
Since X . A2 = 0 we conclude that X· A 1 = O. Hence X is perpendicular to the rows of A. Hence X gives a linear relation among the columns of A. The converse is proved similarly. The above statement proves that if r among the columns of Bare linearly independent, then r among the columns of A are also linearly independent, and conversely. Therefore A and B have the same column rank. We leave the verification that the other row operations do not change the column ranks to the reader. Similarly, one proves that the column operations do not change the row rank. The situation is symmetric between rows and columns. This concludes the proof of the theorem. Theorem 6.2. Let A be a matrix of row rank r. By a succession of row and column operations, the matrix can be transformed to the matrix having components equal to 1 on the diagonal of the first r rows and columns, and 0 everywhere else. r
r
1 0 o 1
o
0
o
0
o o
o o
0 0
1 0 o 0
o o
o
0
o
o
0
In particular, the row rank is equal to the column rank. Proof Suppose r =1= 0 so the matrix is not the zero matrix. Some component is not zero. After interchanging rows and columns, we may assume that this component is in the upper lefthand corner, that is this component is equal to all =1= O. Now we go down the first column. We multiply the first row by a 21 /a 11 and subtract it from the second row. We then obtain a new matrix with 0 in the first place of the second row. Next we multiply the first row by a 31 /a 11 and subtract it from the third row. Then our new matrix has first component equal to 0 in the third row. Proceeding in the same way, we can transform the matrix so that it is of the form all
o
a 12 a 22
a 1n a 2n
[III, §6]
THE RANK OF A MATRIX
119
Next, we subtract appropriate multiples of the first column from the second, third, ... , nth column to get zeros in the first row. This transforms the matrix to a matrix of type
Now we have an (m  1) x (n  1) matrix in the lower right. If we perform row and column operations on all but the first row and column, then first we do not disturb the first component all; and second we can repeat the argument, in order to obtain a matrix of the form 0
all
0 0
0 0
o a22 o 0 a33
a 3n
Proceeding stepwise by induction we reach a matrix of the form
o
o
o
0
o
0
with diagonal elements all' ... ,ass which are =1= O. We divide the first row by all' the second row by a 22 , etc. We then obtain a matrix 1 0
0
0
o
1
0
o o
0
1 0
0
0
0
Thus we have the unit s x s matrix in the upper lefthand corner, and zeros everywhere else. Since row and column operations do not change the row or column rank, it follows that r = s, and also that the row rank is equal to the column rank. This proves the theorem.
120
VECTOR SPACES
[III, §6]
Since we have proved that the row rank is equal to the column rank, we can now omit "row" or "column" and just speak of the rank of a matrix. Thus by definition the rank of a matrix is equal to the dimension of the space generated by the rows. Remark. Although the systematic procedure provides an effective method to find the rank, in practice one can usually take shortcuts to get as many zeros as possible by making row and column operations, so that at some point it becomes obvious what the rank of the matrix is. Of course, one can also use the simple mechanism of linear equations to find the rank. Example. Find the rank of the matrix 1 1
There are only two rows, so the rank is at most 2. On the other hand, the two columns
(~)
G)
and
are linearly independent, for if a, b are numbers such that
then 2a
+b=
0,
b = 0, so that a = O. Therefore the two columns are linearly independent, and the rank is equal to 2. Later we shall also see that determinants give a computation way of determining when vectors are linearly independent, and thus can be used to determine the rank. Example. Find the rank of the matrix. 1
23
210
2 1
1 3 42
[III, §6]
121
THE RANK OF A MATRIX
We subtract twice the first column from the second and add 3 times the first column to the third. This gives
1
2 2
0 3 3
1
6
0 6
3 5
We add 2 times the second column to the third. This gives
100 2 3 0 2 3 3 1 6 7 This matrix is in column echelon form, and it is immediate that the first three rows or columns are linearly independent. Since there are only three columns, it follows that the rank is 3.
Exercises III, §6 1. Find the rank of the following matrices. (a)
(c)
G ~) G ;) 1 2
(b) (  ~
2
(d)
4
1 1 4
0 (e)
Co
(g) (
0) 5
2 5
0 1 8
3
(i)
(:
1 2 4
D 0 2 2
4
(~
0 2 0
(h)
1 1
2
1
2) 5
2 3 2 3 8 12 0 0
(f)
4
:)
2
~)
3 3 2 8 12 1 5
122
VECTOR SPACES
[III, §6]
2. Let A be a triangular matrix
l l2 a::. a ... al") 1
(
. . r
a
:~~
Assume that none of the diagonal elements is equal to O. What is the rank of
A? 3. Let A be an m x n matrix and let B be an n x r matrix, so we can form the product AB. (a) Show that the columns of AB are linear combinations of the columns of A. Thus prove that rank AB (b) Prove that rank AB
~
~
rank A.
rank B. [Hint: Use the fact that rank AB
=
rank tCAB)
rank B
=
rank tB.]
and
CHAPTER IV
Li near Mappi ngs
We shall first define the general notion of a mapping, which generalizes the notion of a function. Among mappings, the linear mappings are the most important. A good deal of mathematics is devoted to reducing questions concerning arbitrary mappings to linear mappings. For one thing, they are interesting in themselves, and many mappings are linear. On the other hand, it is often possible to approximate an arbitrary mapping by a linear one, whose study is much easier than the study of the original mapping. This is done in the calculus of several variables.
IV, §1. Mappings Let S, S' be two sets. A mapping from S to S' is an association which to every element of S associates an element of S'. Instead of saying that F is a mapping from S into S', we shall often write the symbols F: S ~ S'. A mapping will also be called a map, for the sake of brevity. A function is a special type of mapping, namely it is a mapping from a set into the set of numbers, i.e. into R. We extend to mappings some of the terminology we have used for functions. For instance, if T: S ~ S' is a mapping, and if u is an element of S, then we denote by T(u), or Tu, the element of S' associated to u by T. We call T(u) the value of T at u, or also the image of u under T. The symbols T(u) are read" T of u". The set of all elements T(u), when u ranges over all elements of S, is called the image of T. If W is a subset of S, then the set of elements T(w), when w ranges over all elements of W, is called the image of Wunder T, and is denoted by T(W).
124
LINEAR MAPPINGS
[IV, §1]
Let F: S ~ S' be a map from a set S into a set S'. If x is an element of S, we often write x~F(x)
with a special arrow ~ to denote the image of x under F. Thus, for instance, we would speak of the map F such that F(x) = x 2 as the map X~X2.
Example 1. For any set S we have the identity mapping I: S ~ S. It is defined by I(x) = x for all x. Example 2. Let Sand S' be both equal to R. Let J: R ~ R be the function f(x) = x 2 (i.e. the function whose value at a number x is x 2 ). Then f is a mapping from R into R. Its image is the set of numbers
>
o.
Example 3. Let S be the set of numbers > 0, and let S' = R. Let g: S ~ S' be the function such that g(x) = X 1 / 2 • Then g is a mapping from S into R. Example 4. Let S be the set of functions having derivatives of all orders on the interval 0 < t < 1, and let S' = S. Then the derivative D = d/dt is a mapping from S into S. Indeed, our map D associates the function df/dt = Df to the function f According to our terminology, Df is the value of the mapping D at f Example 5. Let S be the set R 3 , i.e. the set of 3tuples. Let A = (2,3, 1). Let L: R3 ~ R be the mapping whose value at a vector X = (x,y,z) is A·X. Then L(X) = A·X. If X = (1, 1, 1), then the value of L at X is 6. Just as we did with functions, we describe a mapping by gIvIng its values. Thus, instead of making the statement in Example 5 describing the mapping L, we would also say: Let L: R3 ~ R be the mapping L(X) = A . X. This is somewhat incorrect, but is briefer, and does not usually give rise to confusion. More correctly, we can write X ~ L(X) or X ~ A . X with the special arrow ~ to denote the effect of the map L on the element X. Example 6. Let F: R2
~ R2
be the mapping given by
F(x, y) = (2x, 2y).
Describe the image under F of the points lying on the circle x 2
+ y2
=
1.
[IV, § IJ
MAPPINGS
125
Let (x, y) be a point on the circle of radius 1. Let u = 2x and v = 2y. Then u, v satisfy the relation
(U/2)2 + (V/2)2 = 1 or in other words,
Hence (u, v) is a point on the circle of radius 2. Therefore the image under F of the circle of radius 1 is a subset of the circle of radius 2. Conversely, given a point (u, v) such that
let x = u/2 and y = v/2. Then the point (x, y) satisfies the equation
and hence is a point on the circle of radius 1. Furthermore, F(x, y) = (u, v).
Hence every point on the circle of radius 2 is the image of some point on the circle of radius 1. We conclude finally that the image of the circle of radius 1 under F is precisely the circle of radius 2. Note. In general, let S, Sf be two sets. To prove that S = Sf, one frequently proves that S is a subset of Sf and that Sf is a subset of S. This is what we did in the preceding argument. Example 7. This example is particularly important in geometric applications. Let V be a vector space, and let u be a fixed element of V. We let
be the map such that Tu(v) = v + u. We call Tu the translation by u. If S is any subset of V, then Tu(S) is called the translation of S by u, and consists of all vectors v + u, with v E S. We often denote it by S + u. In the next picture, we draw a set S and its translation by a vector u.
126
[IV, §1]
LINEAR MAPPINGS
s u
o Figure 1
Example 8. Rotation counterclockwise around the origin by an angle f) is a mapping, which we may denote by Ro. Let f) = n/2. The image of the point (1, 0) under the rotation Rn/2 is the point (0, 1). We may write this as R n / 2 (1, 0) = (0, 1). Example 9. Let S be a set. A mapping from S into R will be called a function, and the set of such functions will be called the set of functions defined on S. Let f, g be two functions defined on S. We can define their sum just as we did for functions of numbers, namely J + g is the function whose value at an element t of S is f(t) + g(t). We can also define the product of f by a number c. It is the function whose value at t is cf(t). Then the set of mappings from S into R is a vector space. Example 10. Let S be a set and let V be a vector space. Let F, G be two mappings from S into V. We can define their sum in the same way as we defined the sum of functions, namely the sum F + G is the mapping whose value at an element t of S is F(t) + G(t). We also define the product of F by a number c to be the mapping whose value at an element t of S is cF(t). It is easy to verify that conditions VS 1 through VS 8 are satisfied.
Exercises IV, §1 1. In Example 4, give Df as a function of x when f is the function: (a) f(x) = sin x (b) f(x) = eX (c) f(x) = log x 2. Let P = (0, 1). Let R be rotation by n/4. Give the coordinates of the image of P under R, i.e. give R(P). 3. In Example 5, give L(X) when X is the vector: (a) (1, 2,  3) (b) ( 1, 5, 0 ) (c) (2, 1, 1) 4. Let F: R ~ R2 be the mapping such that F(t) = (e t , t). What is F(I), F(O), F( I)?
[IV, §2]
127
LINEAR MAPPINGS
5. Let G: R ~ R2 be the mapping such that G(t) = (t, 2t). Let F be as in Exercise 4. What is (F + G) (1), (F + G) (2), (F + G) (O)? 6. Let F be as in Exercise 4. What is (2F) (0), (nF) (I)? 7. Let A = (1, 1, 1, 3). Let F: R4 ~ R be the mapping such that for any vector X = (Xl' x 2 , x 3 , x 4 ) we have F(X) = X· A + 2. What is the value of F(X) when (a) X = (1, 1,0, 1) and (b) X = (2, 3, 1, I)? In Exercises 8 through 12, refer to Example 6. In each case, to prove that the image is equal to a certain set S, you must prove that the image is contained in S, and also that every element of S is in the image. 8. Let F: R2 ~ R2 be the mapping defined by F(x, y) = (2x, 3y). Describe the image of the points lying on the circle x 2 + y2 = 1. 9. Let F: R2 ~ R2 be the mapping defined by F(x, y) = (xy, y). image under F of the straight line X = 2.
Describe the
10. Let F be the mapping defined by F(x, y) = (eX cos y, eX sin y). Describe the image under F of the line X = 1. Describe more generally the image under F of a line x = c, where c is a constant. 11. Let F be the mapping defined by F( t, u) = (cos t, sin t, u). Describe geometrically the image of the (t, u)plane under F. 12. Let F be the mapping defined by F(x, y) = (xI3, yI4). under F of the ellipse
What is the image
IV, §2. Linear Mappings Let V, W be two vector spaces. A linear mapping L:V~W
is a mapping which satisfies the following two properties. First, for any elements u, v in V, and any scalar c, we have:
LM 1. LM2.
L(u
+ v)
= L(u)
+ L(v).
L(cu) = cL(u).
Example 1. The most important linear mapping of this course scribed as follows. Let A be a given m x n matrix. Define
by the formula
IS
de
128
[IV, §2]
LINEAR MAPPINGS
Then LA is linear. Indeed, this is nothing but a summary way of expressing the properties A(X
+
Y)
=
AX
+ AY
and
A(cX)
=
cAX
for any vertical X, Y in Rn and any number c. Example 2. The dot product is essentially a special case of the first example. Let A = (a l , ... ,an) be a fixed vector, and define
Then LA is a linear map from Rn into R, because A . (X
+
Y)
=
A .X
+ A· Y
and
A·(cX)
=
c(A·X).
Note that the dot product can also be viewed as multiplication of matrices if we view A as a row vector, and X as a column vector. Example 3. Let V be any vector space. The mapping which associates to any element u of V this element itself is obviously a linear mapping, which is called the identity mapping. We denote it by I. Thus I(u) = u. Example 4. Let V, W be any vector spaces. The mapping which associates the element 0 in W to any element u of V is called the zero mapping and is obviously linear. Example 5. Let V be the set of functions which have derivatives of all orders. Then the derivative D: V ~ V is a linear mapping. This is simply a brief way of summarizing standard properties of the derivative, namely.
D(f + g) = Df + Dg, D( cf)
=
cD(f)·
Example 6. Let V = R 3 be the vector space of vectors in 3space. Let V' = R 2 be the vector space of vectors in 2space. We can define a mappIng.
by the projection, namely F(x, y, z) = (x, y). We leave it to you to check that the conditions LM 1 and LM 2 are satisfied. More generally, suppose n = r + s is expressed as a sum of two positive integers. We can separate the coordinates (Xl' ... ,xn ) into two
[IV, §2]
129
LINEAR MAPPINGS
bunches (Xl' ... ,X r , X r + 1' ... ,Xr + s ), namely the first r coordinates, and the last s coordinates. Let
be the map easily that F Similarly, we of the linear
such that F(x l' ... ,X n ) = (X l' ... ,xr ). Then you can verify is linear. We call F the projection on the first r coordinates. would have a projection on the last s coordinates, by means map L such that
Example 7. In the calculus of several variables, one defines the gradient of a function f to be
grad f(X) =
of ( ox
"'"
1
Of)
oXn .
Then for two functions f, g, we have grad(f + g) = grad
f + grad
g
and for any number c, grad( cf)
=
c· grad
f
Thus grad is a linear map. Let L: V ~ W be a linear mapping. Let u, v, w be elements of V. Then L(u
+ v + w) =
L(u)
+ L(v) + L(w).
This can be seen stepwise, using the definition of linear mappings. Thus L(u
+ v + w) =
L(u
+ v) + L(w) =
L(u)
+ L(v) + L(w).
Similarly, given a sum of more than three elements, an analogous property is satisfied. For instance, let u 1 , ••• ,Un be elements of V. Then
The sum on the right can be taken in any order. A formal proof can easily be given by induction, and we omit it. If aI' ... ,an are numbers, then
130
[IV, §2]
LINEAR MAPPINGS
We show this for three elements.
+ a 2 v + a 3 w) =
+ L(a 2 v) + L(a 3 w) = a1L(u) + a 2 L(v) + a 3 L(w).
L(a1u
L(a1u)
With the notation of summation signs, we would write
In practice, the following properties will be obviously satisfied, but it turns out they can be proved from the axioms of linear maps and vector spaces. LM 3. Let L: V ~ W be a linear map. Then L(O) = O. Proof We have L(O)
= L(O + 0) = L(O) + L(O).
Subtracting L(O) from both sides yields 0
= L(O),
as desired.
LM 4. Let L: V ~ W be a linear map. Then L( v) = L(v). Proof We have
o=
L(O) = L(v  v) = L(v)
+ L( v).
Add  L( v) to both sides to get the desired assertion. We observe that the values of a linear map are determined by knowing the values on the elements of a basis. Example 8. Let L: R2 ~ R2 be a linear map. Suppose that L( 1, 1)
= (1, 4)
L( 2,  1)
and
= (  2, 3).
Find L(3,  1). To do this, we write (3, 1) as a linear combination of (1,1) and (2, 1). Thus we have to solve (3, 1)
= x(l,
1)
+ y(2,
1).
[IV, §2]
131
LINEAR MAPPINGS
This amounts to solving x
+ 2y = 3,
x y=1. The solution is x
=
!,
L(3, 1) = xL(1, 1)
y =
1.
Hence
1 (1,4) + 3 4 (2,3) + yL(2, 1) = 3
(7 3
3' 16).
=
Example 9. Let V be a vector space, and let L: V ~ R be a linear map. We contend that the set S of all elements v in V such that L(v) < 0 IS convex.
Proof Let L(v) < 0 and L(w) < O. Let 0 < t < 1. Then L(tv
+ (1
 t)w)
tL(v)
=
+ (1
 t)L(w).
Then tL(v) < 0 and (1  t)L(w) < 0 so tL(v) + (1  t)L(w) < 0, whence tv + (1  t)w lies in S. If t = 0 or t = 1, then tv + (1  t)w is equal to v or wand this also lies in S. This proves our assertion. For a generalization of this example, see Exercise 14. The coordinates of a linear map
Let first
be any mapping. Then each value F(v) is an element of R n, and so has coordinates. Thus we can write or Each Fi is a function of V into R, which we write
Example 10. Let F: R2 ~ R3 be the mapping
F(x, y)
=
(2x  y, 3x
+ 4y, x
 5y).
Then
F 1 (x, y)
=
2x  y,
F 2 (x, y)
=
3x
+ 4y,
F 3(X, y)
= x 
5y.
132
[IV, §2]
LINEAR MAPPINGS
Observe that each coordinate function can be expressed in terms of a dot product. For instance, let
Ai = (2, 1),
A2
=
A3=(I,5).
(3,4),
Then
i = 1, 2, 3.
for Each function is linear. Quite generally:
Proposition 2.1. Let F: V ~ R n be a mapping of a vector space V into Rn. Then F is linear if and only if each coordinate function F i: V ~ R is linear, for i = 1, ... ,no
Proof For v,
WE
V we have
F(v
+ w) = (F l(V + w), ... ,Fn(v + w)), F(v)
=
(F l(V), ... ,Fn(v)),
F(w)
=
(F l(W), ... ,Fn(w)).
Thus F(v + w) = F(v) + F(w) if and only if Fi (v + w) = Fi (v) + Fi (w) for all i = 1, ... ,n by the definition of addition of ntuples. The same argument shows that if c E R, then F( cv) = cF( v) if and only if for all
i
=
1, ... ,no
This proves the proposition. Example 10 (continued). The mapping of Example 10 is linear because each coordinate function is linear. Actually, if you write the vector (x, Y) vertically, you should realize that the mapping F is in fact equal to LA for some matrix A. What is this matrix A? The vector space of linear maps Let V, W be two vector spaces. We consider the set of all linear mappings from V into W, and denote this set by 2(V, W), or simply 2 if the reference to V and W is clear. We shall define the addition of linear mappings and their multiplication by numbers in such a way as to make 2 into a vector space. Let L: V ~ Wand let F: V ~ W be two linear mappings. We define their sum L + F to be the map whose value at an element u of V is L(u) + F(u). Thus we may write (L
+ F)(u)
=
L(u)
+ F(u).
[IV, §2]
133
LINEAR MAPPINGS
The map L + F is then a linear map. Indeed, it is easy to verify that the two conditions which define a linear map are satisfied. For any elements U, v of V, we have (L
+ F)(u + v) =
L(u
+ v) + F(u + v)
= L(u) + L(v) + F(u) + F(v) = L(u)
+ F(u) + L«v) + F(v)
= (L + F)(u) + (L + F)(v). Furthermore, if c is a number, then (L
+ F)(cu) =
L(cu)
+ F(cu)
= cL(u) + cF(u) = c[L(u) + F(u)] = c[(L
+ F)(u)].
Hence L + F is a linear map. If a is a number, and L: V ~ W is a linear map, we define a map aL from V into W by giving its value at an element u of V, namely (aL)(u) = aL(u). Then it is easily verified that aL is a linear map. We leave this as an exercise. We have just defined operations of addition and multiplication by numbers in our set.P. Furthermore, if L: V ~ W is a linear map, i.e. an element of .If, then we can define  L to be ( l)L, i.e. the product of the number  1 by L. Finally, we have the zeromap, which to every element of V associates the element 0 of W. Then.If is a vector space. In other words, the set of linear maps from V into W is itself a vector space. The verification that the rules VS t through VS 8 for a vector space are satisfied is easy and is left to the reader. Example t 1. Let V = W be the vector space of functions which have derivatives of all orders. Let D be the derivative, and let 1 be the identity. If f is in V, then
(D
+ l)f =
Df + f
Thus, when f(x) = eX, then (D + l)f is the function whose value at x is eX + eX = 2e x . If f(x) = sin x, then (D + 31)f is the function such that
«D + 3I)f)(x)
=
(Df)(x) + 3If(x)
=
cos x + 3 sin x.
134
[IV, §2]
LINEAR MAPPINGS
We note that 3·/ is a linear map, whose value at f is 3f. Thus (D + 3 ·/)f = Df + 3f. At any number x, the value of (D + 3· 1)f is Df(x) + 3f(x). We can also write (D + 31)f = Df + 3f.
Exercises IV, §2 1. Determine which of the following mappings F are linear. (a) F: R3 + R2 defined by F(x, y, z) = (x, z). (b) F: R4 + R4 defined by F(X) = x. (c) F: R3 + R3 defined by F(X) = X + (0, 1, 0). (d) F: R2 + R2 defined by F(x, y) = (2x + y, y). (e) F: R2 + R2 defined by F(x, y) = (2x, Y  x). (f) F: R2 + R2 defined by F(x, y) = (y, x). (g) F: R2 + R defined by F(x, y) = xy.
2. Which of the mappings in Exercises 4, 7, 8, 9, of §1 are linear? 3. Let V, W be two vector spaces and let F: V + W be a linear map. Let U be the subset of V consisting of all elements v such that F(v) = O. Prove that U is a subspace of V. 4. Let L: V + W be a linear map. Prove that the image of L is a subspace of W. [This will be done in the next section, but try it now to give you practice.] 5. Let A, B be two m x n matrices. Assume that AX=BX
for all ntuples X. Show that A = B. This can also be stated in the form: If LA = LB then A = B. 6. Let Tu; V + V be the translation by a vector u. For which vectors u is Tu a linear map? 7. Let L: V + W be a linear map. (a) If S is a line in V, show that the image L(S) is either a line in W or a point. (b) If S is a line segment in V, between the points P and Q, show that the image L(S) is either a point or a line segment in W. Between which points in W? (c) Let Vi' v2 be linearly independent elements of V. Assume that L(v i ) and L(v 2 ) are linearly independent in W. Let P be an element of V, and let S be the parallelogram with
0
~ ti ~
1 for
i = 1, 2.
Show that the image L(S) is a parallelogram in W. (d) Let v, w be linearly independent elements of a vector space V. Let F: V + W be a linear map. Assume that F(v), F(w) are linearly depen
[IV, §2]
LINEAR MAPPINGS
135
dent. Show that the image under F of the parallelogram spanned by v and W is either a point or a line segment. 8. Let E1 = (1,0) and E2 = (0,1) as usual. Let F be a linear map from R2 into itself such that and
F(E 2 ) = (1, 2).
Let S be the square whose corners are at (0,0), (1, 0), (1, 1), and (0, 1). Show that the image of this square under F is a parallelogram. 9. Let A, B be two nonzero vectors in the plane such that there is no constant c i= 0 such that B = cA. Let L be a linear mapping of the plane into itself such that L(E I ) = A and L(E 2 ) = B. Describe the image under L of the rectangle whose corners are (0, 1), (3, 0), (0, 0), and (3, 1). 10. Let L: R2 ~ R2 be a linear map, having the following effect on the indicated vectors: (a) L(3, 1) = (1, 2) and L( 1, 0) = (1, 1) (b) L(4, 1) = (1, 1) and L(l, 1) = (3, 2) (c) L(l, 1) = (2, 1) and L( 1, 1) = (6, 3). In each case compute L(l, 0). 11. Let L be as in (a), (b), (c), of Exercise 10. Find L(O, 1). 12. Let V, W be two vector spaces, and F: V ~ W a linear: map. Let W 1 , ... ,Wn be elements of W which are linearly independent, and let v l ' ... 'V n be elements of V such that F(vJ = Wi for i = 1, ... ,no Show that v 1' ••• ,vn are linearly independent. 13. (a) Let V be a vector space and F: V ~ R a linear map. Let W be the subset of V consisting of all elements v such that F(v) = o. Assume that W i= V, and let V o be an element of V which does not lie in W. Show that every element of V can be written as a sum W + ev o, with some W in Wand some number c. (b) Show that W is a subspace of V. Let {v 1 , ••• ,vn } be a basis of W. Show that {v o, VI' ... ,vn } is a basis of V. Convex sets 14. Show that the image of a convex set under a linear map is convex. 15. Let L: V ~ W be a linear map. Let T be a convex set in Wand let S be the set of elements VE V such that L(V)E T. Show that S is convex. Remark. Why do these exercises give a more general proof of what you should already have worked out previously? For instance: Let A ERn and let e be a number. Then the set of all X E Rn such that X· A ~ c is convex. Also if S is a convex set and c is a number, then cS is convex. How do these statements fit as special cases of Exercises 14 and 15? 16. Let S be a convex set in V and let U E V. Let 7;,: V ~ V be the translation by u. Show that the image 7;,(S) is convex.
136
[IV, §3]
LINEAR MAPPINGS
Eigenvectors and eigenvalues. Let V be a vector space, and let L: V ~ V be a linear map. An eigenvector v for L is an element of V such that there exists a scalar C with the property. L(v) = CV. The scalar c is called an eigenvalue of v with respect to L. If v =1= 0 then c is uniquely determined. When V is a vector space whose elements are functions, then an eigenvector is also called an eigenfunction. 17. (a) Let V be the space of differentiable functions on R. Let f(t) = eet, where c is some number. Let L be the derivative d/dt. Show that f is an eigenfunction for L. What is the eigenvalue? (b) Let L be the second derivative, that is
for any function f. Show that the functions sin tions of L. What are the eigenvalues?
t
and cos
t
are eigenfunc
18. Let L: V ~ V be a linear map, and let W be the subset of elements of V consisting of all eigenvectors of L with a given eigenvalue c. Show that W is a subspace. 19. Let L: V ~ V be a linear map. Let VI"" ,vn be nonzero eigenvectors for L, with eigenvalues c l , ... 'Cn respectively. Assume that c l' ... ,cn are distinct. Prove that VI"" 'V n are linearly independent. [Hint: Use induction.]
IV, §3. The Kernel and Image of a Linear Map Let F: V ~ W be a linear map. The image of F is the set of elements w In W such that there exists an element v of V such that F(v) = w. The image of F is a subspace of W
Observe first that F(O) = 0, and hence 0 is in the image. Next, suppose that w l' w2 are in the image. Then there exist elements v 1 , V 2 of V such that F(v 1 ) = W 1 and F(v 2 ) = w2 • Hence Proof.
thereby proving that w 1
+ W2
is in the image. If c is a number, then
F(cv 1 ) = cF(v 1 ) = cw 1 • Hence of W
cW 1
is in the image. This proves that the Image is a subspace
Let V, W be vector spaces, and let F: V ~ W be a linear map. The set of elements v E V such that F(v) = 0 is called the kernel of F.
[IV, §3]
THE KERNEL AND IMAGE OF A LINEAR MAP
137
The kernel of F is a subspace of V. Proof Since F(O) = 0, we see that 0 is in the kernel. Let v, w be in the kernel. Then F(v + w) = F(v) + F(w) = 0 + 0 = 0, so that v + w is in the kernel. If c is a number, then F(cv) = cF(v) = 0 so that cv is also in the kernel. Hence the kernel is a subspace. Example 1. Let L: R3 ~ R be the map such that L(x, y, z) = 3x  2y
Thus if A
=
+ z.
(3,  2, 1), then we can write L(X)
=
X·A
=
A·X.
Then the kernel of L is the set of solutions of the equation. 3x  2y
+z=
0.
Of course, this generalizes to nspace. If A is an arbitrary vector in Rn, we can define the linear map
such that LA(X) = A· X. Its kernel can be interpreted as the set of all X which are perpendicular to A. Example 2. Let P: R3 ~ R2 be the projection, such that P(x, y, z)
= (x,
y).
Then P is a linear map whose kernel consists of all vectors in R3 whose first two coordinates are equal to 0, i.e. all vectors (0, 0, z) with arbitrary component z. Example 3. Let A be an m x n matrix, and let
be the linear map such that LA(X) = AX. Then the kernel of LA is precisely the subspace of solutions X of the linear equations AX=O.
138
[IV, §3]
LINEAR MAPPINGS
Example 4. Differential equations. Let D be the derivative. If the real variable is denoted by x, then we may also write D = d/dx. The derivative may be iterated, so the second derivative is denoted by D2 (or (d/dx)2). When applied to a function, we write D2f, so that
Similarly for D 3 , D4 , ... ,D n for the nth derivative. Now let V be the vector space of functions which admit derivatives of all orders. Let al, ... ,am be numbers, and let 9 be an element of V, that is an infinitely differentiable function. Consider the problem of finding a solution f to the differential equation
We may rewrite this equation without the variable x, in the form
Each derivative Dk is a linear map from V to itself. Let
Then L is a sum of linear maps, and is itself a linear map. Thus the differential equation may be rewritten in the form L(f) = g. This is now in a similar notation to that used for solving linear equations. Furthermore, this equation is in "nonhomogeneous" form. The associated homogeneous equation is the equation L(f) = 0, where the righthand side is the zero function. Let W be the kernel of L. Then W is the set (space) of solutions of the homogeneous equation
If there exists one solution fo for the nonhomogeneous equation Lef) = g, then all solutions are obtained by the translation
fo See Exercise 5.
+
W = set of all functions fo
+f
with f in W
[IV, §3]
THE KERNEL AND IMAGE OF A LINEAR MAP
139
In several previous exercises we looked at the image of lines, planes, parallelograms under a linear map. For example, if we consider the plane spanned by two linearly independent vectors Vi' V2 in V, and L:V~W
is a linear map, then the image of that plane will be a plane provided L(v l ), L(v 2 ) are also linearly independent. We can give a criterion for this in terms of the kernel, and the criterion is valid quite generally as follows. Theorem 3.1. Let F: V ~ W be a linear map whose kernel is {O}. If Vi'··. ,Vn are linearly independent elements of V, then F(v l ), ... ,F(vn) are linearly independent elements of W Proof Let
Xl' ...
,xn be numbers such that
By linearity, we get
Hence XlVi + ... + xnvn = O. Since Vl, ... ,Vn are linearly independent it follows that Xi = 0 for i = 1, ... ,no This proves our theorem. We often abbreviate kernel and image by writing Ker and 1m respectively. The next theorem relates the dimensions of the kernel and image of a linear map, with the dimension of the space on which the map is defined. Theorem 3.2 Let V be a vector space. Let L: V ~ W be a linear map of V into another space W Let n be the dimension of V, q the dimension of the kernel of L, and s the dimension of the image of L. Then n = q + s. In other words, dim V = dim Ker L
+ dim 1m L.
Proof If the image of L consists of 0 only, then our assertion is trivial. We may therefore assume that s > o. Let {Wi' ... ,Ws } be a basis of the image of L. Let Vi' ... ,vs be elements of V such that L(v i ) = Wi for i = 1, ... ,so If the kernel is not {O}, let {u l , ... ,u q } be a basis of the kernel. If the kernel is {O}, it is understood that all reference to {u l , ... ,uq } is to be omitted in what follows. We contend that
140
LINEAR MAPPINGS
[IV, §3]
is a basis of V. This will suffice to prove our assertion. Let v be any element of V. Then there exist numbers Xl' ... ,xs such that
because {Wl' ... ,ws } is a basis of the image of L. By linearity,
and again by linearity, subtracting the righthand side from the lefthand side, it follows that
Hence v  X 1 V 1  •••  Xs vs lies in the kernel of L, and there exist numbers Y l' ... ,Yq such that
Hence
is a linear combination of v l , ... 'V s ' U l , ... ,uq • This proves that these s + q elements of V generate V. We now show that they are linearly independent, and hence that they constitute a basis. Suppose that there exists a linear relation:
Applying L to this relation, and using the fact that L(u j ) = 0 for j = 1, ... ,q, we obtain
But L(v l ), ... ,L(vs) are none other than W l , ... 'w s' which have been assumed linearly independent. Hence Xi = 0 for i = 1, ... ,so Hence
But U l , ... ,uq constitute a basis of the kernel of L, and in particular, are linearly independent. Hence all Yj = 0 for j = 1, ... ,q. This concludes the proof of our assertion. Example 1 (continued). The linear map L: R3 ~ R of Example 1 given by the formula L(x, y, z) = 3x  2y + z.
IS
[IV, §3]
THE KERNEL AND IMAGE OF A LINEAR MAP
141
Its kernel consists of all solutions of the equation
3x  y
+ z = O.
Its image is a subspace of R, is not {O}, and hence consists of all of R. Thus its image has dimension 1. Hence its kernel has dimension 2. Example 2 (continued). The image of the projection
in Example 2 is all of R2, and the kernel has dimension 1.
Exercises IV, §3 Let L: V ~ W be a linear map. 1. (a) If S is a onedimensional subspace of V, show that the image L(S) either a point or a line. (b) If S is a twodimensional subspace of V, show that the image L(S) either a plane, a line or a point.
IS
IS
2. (a) If S is an arbitrary line in V (cf. Chapter III, §2) show that the image of S is either a point or a line. (b) If S is an arbitrary plane in V, show that the image of S is either a plane, a line or a point. 3. (a) Let F: V ~ W be a linear map, whose kernel is {O}. Assume that V and W have both the same dimension n. Show that the image of F is all of W (b) Let F: V ~ W be a linear map and assume that the image of F is all of W Assume that V and W have the same dimension n. Show that the kernel of F is {O}. 4. Let L: V ~ W be a linear map. Assume dim V > dim W. kernel of L is not O.
Show that the
5. Let L: V ~ W be a linear map. Let w be an element of W. Let Vo be an element of V such that L(vo) = w. Show that any solution of the equation L(X) = w is of type Vo + u, where U is an element of the kernel of L. 6. Let V be the vector space of functions which have derivatives of all orders, and let D: V ~ V be the derivative. What is the kernel of D? 7. Let D2 be the second derivative (i.e. the iteration of D taken twice). What is the kernel of D2? In general, what is the kernel of Dn (nth derivative)? 8. (a) Let V, D be as in Exercise 6. Let L = D  I, where I is the identity mapping of V. What is the kernel of L? (b) Same question of L = D  aI, where a is a number.
142
LINEAR MAPPINGS
[IV, §3]
9. (a) What is the dimension of the subspace of Rn consisting of those vectors A = (a 1 , ••• ,an) such that a 1 + ... + an = O? (b) What is the dimension of the subspace of the space of n x n matrices (aij) such that n
a 11 +···+ann = "a i..J n.. =O?• i= 1
10. An n x n matrix A is called skewsymmetric if tA = A. Show that any n x n matrix A can be written as a sum A
= B + C,
where B is symmetric and C is skewsymmetric. [Hint: Let B = (A + tA)/2.J Show that if A = B 1 + C l' where B 1 is symmetric and C 1 is skewsymmetric, then B = Bland C = C 1. 11. Let M be the space of all n x n matrices. Let P: M
+
M
be the map such that A+tA
2
P(A) =
.
(a) Show that P is linear. (b) Show that the kernel of P consists of the space of skewsymmetric matrices. (c) Show that the image of P consists of all symmetric matrices. [Watch out. You have to prove two things: For any matrix A, P(A) is symmetric. Conversely, given a symmetric matrix B, there exists a matrix A such that B = P(A). What is the simplest possibility for such A?J (d) You should have determined the dimension of the space of symmetric matrices previously, and found n(n + 1)/2. What then is the dimension of the space of skewsymmetric matrices? (e) Exhibit a basis for the space of skewsymmetric matrices.
12. Let M be the space of all n x n matrices. Let Q: M
+
M
be the map such that AtA Q(A) =
2
(a) Show that Q is linear. (b) Describe the kernel of Q, and determine its dimension. (c) What is the image of Q?
[IV, §3]
143
THE KERNEL AND IMAGE OF A LINEAR MAP
13. A function (real valued, of a real variable) is called even if f( x) = f(x). It is called odd if f(  x) =  f(x). (a) Verify that sin x is an odd function, and cos x is an even function. (b) Let V be the vector space of all functions. Define the map P: V + V
by (Pf)(x) =