xiv Preface

CHAPTER
• Acknowledgments

I would like to express my gratitude to the long list of reviewers that have contributed
so much to all eight editions of this book. Thanks also to the many users who have sent
In comments and suggestions.
A special thanks is due to the reviewers of the eighth edition:
Margaret Bayer, University of Kansas
Anton Betten, Colorado State University
Fernanda Botelho, University of Memphis
Pinyuen Chen, Syracuse University
Aimo Hinkkanen, University of Illinois, Urbana-Champaign
Naihuan Jing, North Carolina State University
Mohammad A. Kazemi, University of North Carolina, Charlotte
Terry A. Loring, University of New Mexico Matrices and Systems of Equations
Judi McDonald, Washington State University
Shashikant MUlay, University of Tennessee Probably the most important problem in mathematics is that of solving a system of
John C. Nardo, Oglethorpe University linear equations. Well over 75 percent of all mathematical problems encountered in
Thomas Riedel, University of Louisville scientific or industrial applications involve solving a linear system at some stage. By
Pavel Sikorskii, Michigan State University using the methods of modern mathematics, it is often possible to take a sophisticated
problem and reduce it to a single system of linear equations. Linear systems arise in
Tin- Yau Tam, AUburn University
applications to such areas as business, economics, sociology, ecology, demography, ge-
Alexander Tumanov, University of Illinois, Urbana-Champaign netics, electronics, engineering, and physics. Therefore, it seems appropriate to begin
Many of the revisions and new . .. .. this book with a section on linear systems.
comm en ts an d suggestions.
. exercise, in this latest edition are a direct result of their .
Special thanks to Mathe ti Edi .
the new edition and t P . rna ICS itor Carohne Celano for her help in planning
production of the bo ok ~~ec~ Manager Robert Merenoff for his work on coordinating III Systems of Linear Equations
editorial, production 0 . d ile s to accuracy checker Tom Wegleitner and to the entire
I would lik ' kr sa es staff at Pearson for all their efforts. A linear equation in n unknowns is an equation of the form
I e to ac nowledge the ibuti
Most of the first edition f th b k COntn ut,ons of Gene Golub and Jim Wilkinson. alxl +a2x2+···+al1xn =b
Scholar at Stanford Uni 0 . e 000 . was wruten m 1977-1978 while I was a Visiting
iversny unng that . dI where a" a2, ... ,an and b are real numbers and Xl, X2, ... , Xli are variables. A linear
numerical linear algebra . . b perm attended Courses and lectures on
have greatly influenced t~il:e;oo~ G;ne Golub and J. H. Wilkinson. Those lectures system of m equations in Il unknowns is then a system of the form
Gennund Dahlquist for h· sh If· mally,. I would Irke to express my gratitude to
though Gene Golub J. 'w ·lekP ul suggestIOns on earlier editions of this book. AI-
QIIXI + GI2X2 + + QlnXI/ = b,
h ' irn I mson and G
t ey Continue to live On in th '. errnun
d Dahlnu:
ahlqulst are no longer with us, Q21XI + Q22X2 + + Q21lXIl = b:
e memones of their friends. (1)

Steven J. Leon
[email protected] where the a;/s and the b;'s are all real numbers. We will refer to systems of the form (1)
as m x n linear systems. The following are examples of linear systems:
(a) XI + 2x, =5 (b) XI - X, + X3 =2 (c) XI + X, =2
2xI + 3X2 =8 2xI + X2 - X3 =4 Xl - X2 = I
X, =4
2 Chapter I Matrices and Systems of Equations

1.1 Systems of Linear Equations l

System (a) is a 2 x 2 system, (b) is a 2 x 3 system, and (c) is a 3 x 2 system.
By a solution of an m x n system, we mean an ordered n-tuple of numbers x,
(XI, X2, ... ,' x,,) that satisfies all the equations of the system. For example, the ordered
pair (I, 2) IS a solution of system (a), since

I . (I) + 2· (2) = 5
2 . (I) + 3 . (2) =8
The ordered triple (2, 0, 0) is a solution of system (b), since

I . (2) - I . (0) +I . (0) = 2 (i) (ii) (iii)

2 . (2) + I . (0) - I . (0) = 4
Figure 1.1.1.
~ctu~ly, system (b) has many solutions. If a is any real number it is easily seen
~ alItt efordered triple (2, a, a) is a solution. However system (c) has no solution It
'0 ows rom the third equatio th t h fi .' . The situation is the same for m x n systems. An m x n system mayor may not
b 4 U· . nat erst
e . smg XI = 4 in the first two ti coordmate of any solution would have to be consistent. If it is consistent, it must have either exactly one solution or infinitely
satisfy equa IOns, we see that the second coordinate must
many solutions. These are the only possibilities. We will see why this is so in Section 2
when we study the row echelon form. Of more immediate concern is the problem of
4 + X2 =2
finding all solutions of a given system. To tackle this problem, we introduce the notion
4 - X2 =1 of equivalent systems.
Since there is no real number that . fi b
solution. If a linear syst h sans .es oth of these equations, the system has no
m Equivalent Systems
the system has at Ieast: als no solution, We say that the system is inconsistent. If
. ne so ution we say that it is . Th ( ) . Consider the two systems
mconsistent, while system (a) d ('b . consistent. us, system c IS
The set of all I . s a an ) are both consIstent.
so uuons of a hnear syste . 11 d (a) 3xI + 2X2 - X3 = -2 + 2X2 = -2
If a system is inconsistent't
t I' .m IS ca e the solution set of the system.
nonempty solution set 1: ' I so uuon set IS empty. A consistent system will have a
. a so ve a consistent system, we must find its solution set.
X2 = 3
(b)
-3xl
3xI

- X2
-

+ X3
X3

= 5
2X3 = 4 3xI + 2X2 + X3 = 2
2 x 2 Systems
System (a) is easy to solve because it is clear from the last two equations that
Let us exarni ne . X2 = 3 and X3 = 2. Using these values in the first equation, we get
geometncally a system of the form
3xI + 2·3 - 2 =-2
allXI + al2X2 = b, x, =-2
a21XI + a22x2 = b:
Thus, the solution of the system is (-2,3,2). System (b) seems to be more difficult
Each equation can be represented ra hi '.
(XI, X2) will be a solution of th g p. cally as a line in the plane. The ordered pair to solve. Actually, system (b) has the same solution as system (a). To see this, add the
id e system If and a I if i I' first two equations of the system:
consi er the three systems n Y I It res on both lines. For example,

(i) XI + X2 = 2 3xI + 2X2 - X3 = -2

(ii) XI + X2 = 2 -3xl - X2 + X3 = 5
x, - X2 '" 2 (iii)
2
XI + X2 = I X2 = 3
-XI - X2 =-2
The two lines in system (i) int
If (XI, X2, X3) is any solution of (b), it must satisfy all the equations of the system.
set of (i). In system (ii) the two I' ersect at the point (2, 0). Thus, {(2 0)) is the solution
Thus, it must satisfy any new equation fonmed by adding two of its equations. There-
and hence its solution set is emptmeTshareparallel. Therefore, system' (ii) is inconsistent
same line A' y. e two equati' '" fore, X2 must equal 3. Similarly, (XI, X2, X3) must satisfy the new equation fonmed by
I . ny porm on this line will be I' ons in system (Ill) both represent the subtracting the first equation from the third:
allel : general, there are three POSSibiliti:s~o ullon of the system (see Figure 1.1.1).
, r both equatIOns represent th
one zero or' fi .
. the lines mtersect at a point they are par-
e same line Th I' ,
3xI + 2X2 + X3 = 2
, , m nItely many points. . e so unon set then contains either 3xj + 2X2 - X3 = -2
2X3 = 4
4 Chapter I Matrices and Systems of Equations
1.1 Systems of Linear Equations 5
Therefore, any solution of system (b) must also be a solution of system (a). By a
similar argument, It can be shown that any solution of (a) is also a solution of (b). This To summarize, there are three operations that can be used on a system to obtain an
equivalent system:
can be done by subtracting the first equation from the second:
I. The order in which any two equations are written may be interchanged.
X2 3
II. Both sides of an equation may be multiplied by the same nonzero real number.
3x] + 2X2 - X3 = -2
III. A multiple of one equation may be added to (or subtracted from) another.
-3x] - X2 + X3 = 5
Given a system of equations, we may use these operations to obtain an equivalent
Then add the first and third equations: system that is easier to solve.

33x] + 2X2 - X3 = -2
2X3 = 4 n X n Systems
3x] + 2X2 + X3 = 2
Let us restrict ourselves to n x n systems for the remainder of this section. We will
Thus, (x], X2, X3) is a solution of system (b) if and only if it is a solution of system (a). show that if an n x n system has exactly one solution, then operations I and III can be
Therefore, both systems have the same solution set, ((_ 2, 3, 2)}. used to obtain an equivalent "strictly triangular system."
Definition
TthwO
hsystemhsof equations involving the same variables are said to be equivalent if Definition A system is said to be in strict triangular form if, in the kth equation, the coef-
ey ave t e same solutJOnset.
ficients of the first k - I variables are all zero and the coefficient of Xk is nonzero
(k= I, ... ,n).
Clearly, if we interchange the ord . hi h
this will have no effe t he solun er 10 w 1C two equations of a system are written,
the original system .; on t 0. slounon set. The reordered system will be equivalent to EXAMPLE I The system
. or examp 0., the systems

x] + 2X2 =4 4xI + X2 =6 3x] + 2X2 + X3 = 1

X2 - X3 =2
3x] - X2 =2 and 3xI _ X2 =2
4x] + X2 =6 XI
+2 X2 =
4 2X3 =4
both involve the same three equations
is in strict triangular form, since in the second equation the coefficients are 0, I, -I,
solution set. and, consequently, they must have the same
respectively, and in the third equation the coefficients are 0, 0, 2, respectively. Because
If one equation of a system is multi r of the strict triangular form, the system is easy to solve. It follows from the third
will have no effect on the sotuu rp led through by a nonzero real number, this
. . JOnset and the new system WI'11 be equivalent to the equation that X3 = 2. Using this value in the second equation, we obtain
ongmal system. For example th'
, 0. systems

x] + X2 + X3 =3 X2 - 2 =2 or
-2x] - X2 + 4X3 =I and 2x] + 2X2 + 2X3 = 6
are equivalent. -2xI - X2 + 4X3 = I Using X2 = 4, X3 = 2 in the first equation, we end up with

If a multiple of one equation is added

be .equivalent to the original system Thi to another equation, the new system will
3x] + 2·4 + 2=
satisfy the two equations . IS fOllows Smce the n-tuple (x" ••• , xn) will x] =-3

. . .
anx]

ai]x]
+ ... + a·", x n
+"'+ainXn
-
-
=b.
b
i
Thus, the solution of the system is (-3,4,2).
•
If and only If It satisfies the equations ] Any n x n strictly triangular system can be solved in the same manner as the last
example. First, the nth equation is solved for the value of x". This value is used in the
(n - l)st equation to solve for Xn-]' The values Xn and Xn_] are used in the (n - 2)nd
anXI + ... + a· x - b
(ai] +aan)xI + ... + (a. + 'n n - i equation to solve for Xn-2, and so on. We will refer to this method of solving a strictly
]n aainhn = b, + ab, triangular system as back substitution.
6 Chapter I Matricesand Systems of Equations
1.1 Systems of Linear Equations 7
EXAMPLE 2 Solve the system
Using back substitution, we get
2xI - x, + 3X3 - 2X4 =I
x, - 2X3

4X3
+ 3X4
+ 3X4
=2
=3
X3 = 4, X2 = -2,

Let us look back at the system of equations in the last example. We can associate
XI =3
•
4X4 =4 with that system a 3 x 3 array of numbers whose entries are the coefficients of the Xi'S:

-n
2
Solution
-I
Using back substitution, we obtain 3
We will refer to this array as the coefficient matrix of the system. The term matrix
4X4 =4 X4 = I means simply a rectangular array of numbers. A matrix having m rows and n columns
4X3 + 3· I =3 X3 = 0 is said to be m x n. A matrix is said to be square if it has the same number of rows and
x, -
2 .0 +3 . I =2 x, =-1 columns-that is, if m = n.
2xI-(-I)+3·0-2·1 = I Xl = If we attach to the coefficient matrix an additional column whose entries are the
numbers on the right-hand side of the system, we obtain the new matrix
Thus, the solution is (I, -I, 0, I).
• 2 I
. In general, given a system of n linear equations in n unknowns, we will use opera-
tions I and III to try to obtain an equivalent system that is strictly triangular. (We will
see III the next section of the book that it is not possible to reduce the system to strictly
-I
3
-3
I -1]
tnanzular form III the cases wh th We will refer to this new matrix as the augmented matrix. In general, when an m x r
e» ere e system does not have a unique solution.) matrix B is attached to an m x n matrix A in this way, the augmented matrix is denoted
EXAMPLE 3 Solve the system by (A I B). Thus, if

XI

3xI
+ 2x, +
- X, - 3X3
X3 = 3
= -I A=
[ ",
a"
a12
an ". )
a2n
B=
[ bb,21
b12
b" b"
b" ]

Solution
2x, + 3X2 + X3 = 4
a/:/I am2 am" ' ., bm2 bmr
then
SUbtracting 3 times the first row from the second row yields
(A IB) = r all
:
alII bll
i; J
- ?X, - 6X3 =- 10 ami all/I! bml bmr
SUbtracting 2 times the first row from the third row yields With each system of equations, we may associate an augmented matrix of the form

-x, -X3 =-2 alII

If the second and third equations of Our s .

equations, we obtain the e . I ystem, respectIvely, are replaced by these new
qUlva enr system
The system can be solved by performing operations on the augmented matrix. The Xi'S
XI + 2x, + X3 = 3 are placeholders that can be omitted until the end of the computation. Corresponding
- 7x, - 6X3 = -10 to the three operations used to obtain equivalent systems, the following row operations
-X, - X3 = -2 may be applied to the augmented matrix:
If the third equation of thi .
. IS system IS repl db'
lImes the second equation we end . ahce y the sum of the third equation and - j Elementary Row Operations
, up Wit the f II· .
o OWlllg strictly triangular system:
I. Interchange two rows.
x, + 2x, + X3 = 3
II. Multiply a row by a nonzero real number.
-?x, - 6X3 = -10
III. Replace a row by its sum with a multiple of another row.
-~X3 == _~
7
8 Chapter I Matrices and Systems of Equations
1.1 Systems of Linear Equations 9
Returning to the example, we find that the first row is used to eliminate the ele-
ments in the first column of the remaining rows. We refer to the first row as the pivotal Row operation III is then used twice to eliminate the two nonzero entries in the first
column:
row. For emphasis, the entries in the pivotal row are all in bold type and the entire row
is color shaded. The first nonzero entry in the pivotal row is called the pivot.

(pivot al [ = I) [1 2 1 3] <-- pivotal row

o
I I I
-1 -1 I 06]
1
entries to be eliminated} --)- 3 -I - 3 ~I [o 2 -I -4 -13
o -2 -5 -I -15
a'i =3andaJI =2 2 3 I 4

By using row operation III, 3 times the first row is subtracted from the second row and Next, the second row is used as the pivotal row to eliminate the entries in the second
column below the pivot element -I:
2 times the first row is subtracted from the third. When this is done, we end up with
the matrix
I 2 I 3]
-j]
I I I I
[o
o
-7 -6 -10 <-- pivotal row
-I -I -2
o
o
-I -I
0 -3 -2
I
[
At this step .we choose the second row as our new pivotaJ row and apply row opera- o 0 -3 -3 -15
non III to ehmm atT the iast elementin the second column. This time, the pivot is -7
and the quotient -7 = '7 IS the multiple of the pivotal row that is subtracted from the Finally, the third row is used as the pivotal row to eliminate the last element in the third
third row. We end up with the matrix column:
I I J I

-l~]
I 2
o -7 I
-6 -10 3] o -I -I I
[
o 0 -'7 I 4 [o 0 -3 -2
~'7 o 0 o -I -2
This is the augmented matrix for the strictly triangular system, which is equivalent to
the original system. The solution of the system is easily obtained by back substitution. This augmented matrix represents a strictly triangular system. Solving by back substi-
tution, we obtain the solution (2, -1,3,2). •
EXAMPLE 4 Solve the system
In general, if an n x n linear system can be reduced to strictly triangular form, then
4 - x, - XJ + X4 = 0 it will have a unique solution that can be obtained by performing back substitution on
XI + X, + XJ + X4 = 6 the triangular system. We can think of the reduction process as an algorithm involving
2xI + 4x, + XJ - 2X4 = -1 n - I steps. At the first step, a pivot element is chosen from among the nonzero entries
3xI + X, - 2X3 + 2X4 = 3 in the first column of the matrix. The row containing the pivot element is called the
pivotal row. We interchange rows (if necessary) so that the pivotal row is the new first
Solution row. Multiples of the pivotal row are then subtracted from each of the remaining n - 1
The augmented matrix for this system is rows so as to obtain O's in the first entries of rows 2 through n. At the second step,

~-: -: : ~I
a pivot element is chosen from the nonzero entries in column 2, rows 2 through n, of
the matrix. The row containing the pivot is then interchanged with the second row of
the matrix and is used as the new pivotal row. Multiples of the pivotal row are then
subtracted from the remaining n - 2 rows so as to eliminate all entries below the pivot
2 4 I -2 I in the second column. The same procedure is repeated for columns 3 through n _ I.
. .. 3 I -2 2 3 Note that at the second step row 1 and column I remain unchanged, at the third step
Smce it is not possible to elirni
the first two rows and first two columns remain unchanged, and so on. At each step,
use row operation 1to int hi mate any entries by using 0 as a pivot element we will
fi ere ange the fi t ' the overall dimensions of the system are effectively reduced by 1 (see Figure 1.1.2).
rst row will be the pivotal d rs. two rows of the augmented matrix. The new
rowan the Pivot element will be I: If the elimination process can be carried out as described, we will arrive at an
equivalent strictly triangular system after n - I steps. However, the procedure will

-,,,.,) Ii -1 ~1-1-il ~p;.~,~

break down if, at any step, all possible choices for a pivot element are equal to O.
When this happens, the alternative is to reduce the system to certain special echelon,
or staircase-shaped, forms. These echelon forms will be studied in the next section.
They will also be used for m x n systems, where m t= n.
 10 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form II

7. The two systems 9. Given a system of the form

Step I

(iiii}(liiiiJ 2xI
4xI
+ X2 = 3
+ 3x, = 5
and
2XI+
4xI + 3X2
x2=-1
=
-mlXI

-m2XI
+ X2
+ X2
= b,
= b2

Step2

(1 iii i)~ (1 iii i) have the same coefficient matrix but different right-
hand sides. Solve both systems simultaneously

augmented matrix
by
eliminating the first entry in the second row of the
where m., m2, bl, and b2 are constants,
(a) Show that the system will have a unique solu-
tion if m 1

(b) Show that if

=f- m2.
= In" then the system will be

(1 i i i i)-' (1 iii
Inl

Step 3

Figure 1.1.2.
i) -and then performing back substitutions for each of
consistent
(c) Give a geometric
(b).
only if b, = b-:

10. Consider a system of the form

interpretation
WIt. SO L
of parts (a) and

the columns corresponding to the right-hand sides.

8. Solve the two systems
allXI + al2X2 = a
SECTION I, I EXERCISES a21xI + a22X2 = 0

r; ; I n
1. Use back SUbstit~tion to solve each of the followin Xl + 2X2 - 2X3 =1 Xl + 2X2 - 2X3 = 9 where all, a12, a21, and a22 are constants. Explain
systems of equations: g + 5X2 +
-2 I 2xI X, = 9 + + X, =
(a) XI - 3x, = 2
+ + =8
(a) (b) (;
3 -4 I ~J XI + 3x, + 4x, = 9
2xI
XI
5X2
+ 3X2 + 4x, = -2
9 why a system of this form must be consistent.
11. Give a geometrical interpretation of a linear equa-

-l)
(b) XI X, X3
2X2 = 6 tion in three unknowns. Give a geometrical de-
2x, + X3 =5 (c) [~-; ~ by doing elimination on a 3 x 5 augmented matrix scription of the possible solution sets for a 3 x 3
3X3 =9 5 2 6-1 and then performing two back substitutions. linear system.
(c) XI + 2x2 + 2x, + X, = 5
3X2 + x] - 2X4 = J
-i -~ ~ ~1
II
(d) (~
-X,+2x,=_1 J J 2 4 8
J
Row Echelon Form
= 4 4X4 5 3 -2 7
(d) XI + X, + X3 + X, + x, = 5 6. Solve each of the following systems:
In Section I we learned a method for reducing an n x n linear system to strict triangular
2x, + X, - 2x4 + X, = I (a) XI - 2X2 =5 (b) 2xI + X2 =8 form. However, this method will fail if, at any stage of the reduction process, all the
4x, + X, - 2x, = I 3x, + X2 = I 4xI - 3X2 = 6 possible choices for a pivot element in a given column are O.
X4 - 3xs :::::0 (c) 4xI + 3X2 = 4 (d) XI + 2X2 _ X3 =I
=2 2 EXAMPLE I Consider the system represented by the augmented matrix
2. Write out the coefficient rna .
2x, ,XI + 4X2 = 3 2x, - X2 + X, = 3
terns in Exerci~ tnx for each of the sys- -XI + 2x2 + 3x, = 7 1 1 1 1 1 1 <- pivotal row
3. In each of the following s . (e) 2x, =
+ X2 + 3X3 -I -I 0 0 I -I
equation as a line in the Iystems, mterpret each 4xI + 3X2 + 5x, = I -2 -2 0 0 3 I
graph the lines and det pane. For each system,
6x, + 5X2 + 5X3 = -3 0 0 I I 3 -I
number of solutions. ermme geometrically the
(I) I I 2 2 4 I
(a) x, + X2 = 4 3xI + 2x + = 0 2 X3

X, - X2 == 2
(b)
XI + 2x2 = 4 -2x, + x, - x, = 2 If row operation III is used to eliminate the nonzero entries in the last four rows of the

2x1
-2x1 - 4X2 =4 2xI - X2 + 2x, ""-I first column, the resulting matrix will be
(c)
-4xI
-

+ 2x, = -6
X2 "" 3
(d) XI + X2 "" 1 (g) "x
3 I
+ j'2X2+ 2X3=-J I I 1 I I I
Xl - X2 =J XI + 2x2 + tx] = 2. 0 0 1 1 2 0 <- pivotal row
4. W'nte an augmented matrix for -XI + 3X2 "" 3 !x
2 I
+ 2x 2 + T12 = x)
2
..L 0 0 2 2 5 3
10
in Exercise 3. each of the systems (h) 0 0 1 I 3 -I
X2 + X3 + X4 =0 0 0 I I
5. Write OUithe system of equations that co 3xI 3 0
to each of the follOWing augme ted rresponds
+ 3X3 - 4x, =7
n rnabices: XI + X, + X, + 2x, =6 At this stage, the reduction to strict triangular form breaks down. All four possible
2x, + 3X2 + X, + 3x, "" 6 choices for the pivot element in the second column are O. How do we proceed from
12 Chapter 1 Matrices and Systems of Equations
1.2 Row Echelon Form 13
here? Since our goal is to simplify the system as much as possible, it seems natural to
the solution set will be the set of all 5-tuples satisfying the first three equations,
move over to the third column and eliminate the last three entries:

I I I I I I
X, + x, + X3 + X4 + X, = I
o o I I X3 + X4 + 2x, =0 (I)
2 0
o o o o 1 3 x, = 3
o o o o I -I The variables corresponding to the first nonzero elements in each row of the reduced
o o o o I 0 matrix will be referred to as lead variables. Thus, XI, X3, and Xs are the lead variables.
The remaining variables corresponding to the columns skipped in the reduction process
In the fourth column, all the choices for a pivot element are 0; so again we move on to
will be referred to as free variables. Hence, X, and X4 are the free variables. If we
the next column. If we use the third row as the pivotal row, the last two entries in the
transfer the free variables over to the right-hand side in (I), we obtain the system
fifth column are eliminated and we end up with the matrix
XI + X3 + Xs = I - X, - X4
I I I X3 + 2x, = -X4 (2)
o o o
o o X, =3
3
o o -4 System (2) is strictly triangular in the unknowns XI, X3, and X,. Thus, for each pair
o o -3 of values assigned to x, and X4, there will be a unique solution. For example, if = x,
X4 = 0, then x, = 3, X3 = -6, and XI = 4, and hence (4,0, -6,0,3) is a solution to
The coefficient matrix that we end up with is not in strict triangular form; it is in the system.
staircase, or echelo?, form. The horizontal and vertical line segments in the array for
the coefficient matnx indicate the structure of the staircase form. Note that the vertical
Definition A matrix is said to be in row echelon form
drop IS I for each step, but the horizontal span for a step can be more than I.
The equations represented by the last two rows are (I) If the first nonzero entry in each nonzero row is I.
(ii) If row k does not consist entirely of zeros, the number of leading zero
OXI + Ox, + OX3 + OX4 + Ox, = -4 entries in row k + I is greater than the number of leading zero entries in
OXI + Ox, + OX3 + OX4 + Ox, = -3 row k.

Since there are no 5-tuples th t ld . . (iii) If there are rows whose entries are all zero, they are below the rows having
a cou satIsfy these equations, the system is inconSIstent. nonzero entries.
•
EXAMPLE 2 The following matrices are in row echelon form:
so a;~pp~se now that we change the right-hand side of the system in the last eX3iUple
00 tain a consIstent system. For example, if we start with
[Io 4I 32] , [Io 02 3]I , 3 I
-I
-2
I I
-1
-2
0
0
I I I
0
0
I-I
3
I
I
o 0 I 000 [~ 0
0 0
I
~] •
EXAMPLE 3 The following matrices are not in row echelon form:
001 133
I 1 2 2 4 4
[2035 4 6]
then the reduction process will yield the echelon form augmented matrix 004
, [0
o I
0 0)
0 ' [~ b)
The first matrix does not satisfy condition (i). The second matrix fails to satisfy condi-
I I I 1
tion (iii), and the third matrix fails to satisfy condition (ii). •
o 0 2 0
o 0 I 3
o 0 0 0 Definition The process of using row operations I, II, and ill to transform a 'linear system into
o 0 0 0 0 0 one whose augmented matrix is in row echelon form is called Gaussian elimina-
The last two equations of th red tion.
e uced system will be satisfied for any 5-tuple. Thus,
14 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form 15
Note that row operation II is necessary in order to scale the rows so that the leading
X2
coefficients are all I. If the row echelon form of the augmented matrix contains a row
of the form

[00"'011)

the system is inconsistent. OtherWise, the system will be consistent. If the system is
consistent and the nonzero rows of the row echelon form of the matrix form a strictly
triangular system, the system Willhave a unique solution.

Overdetermined Systems
A linear system is said to be d . .
.
len owns. Overdetennmed oven eterminen If there are more equations than un-
systems II (b . Figure 1.2.1.
are usua y ut not always) inconsistent. X'fZX~~X"",
EXAMPLE 4
Solve each of the following overdetermined systems:
X, -- ,
matrix form a strictly triangular system.
(a) XI + X, = I (b) XI + 2x, + X3 = I
XI - X, - 3 I
- 2xI - X, + X3 = 2 2 I I I 2 I
-XI + 2x, = -2 4 3
XI + X, + 3X3 =4 2 -I I 2 0 I I
0
System (c): -> '5
(e) 2xI - X, + 3X3 =5 4 3 3 4 0 0 0 0
XI + 2x, + XJ = I
3 I 2 3 0 0 0 0
2xI - X, + X3 =2
4xI + 3x, + 3xJ =4 Solving for X, and XI in terms of X3, we obtain
'Xl_ - 0 2. >J
3xI + X, + 2X3 = 3 X2 = -0.2X3
Solution XI = I - 2x, - X3 = I - 0.6X3

By now the reader should be famT l

. It follows that the solution set is the set of all ordered triples of the form
omit the intermediate step' d iar enough With the elimination process that we can (I - 0.00, -0.2a, a), where a is a real number. This system is consistent and has
s 10 re ucrng each of these systems. Thus, we may write infinitely many solutions because of the free variable X3. •

System (a): r : -: ~ 1-> r ~ -: 1 Underdetermined Systems

A system of m linear equations in n unknowns is said to be underdetermined if there are
-I 2 -2 0 0 I
fewer equations than unknowns (m < n). Although it is possible for underdetermined
It follows from the last row of the reduce .
three equations in system ( ) d matnx that the system is inconsistent. The systems to be inconsistent, they are usually consistent with infinitely many solutions.
at the point (2, -I) Howe~erePthresehnt lines in the plane. The first two lines intersect It is not possible for an underdetermined system to have a unique solution. The reason
there .' r, e t Ird line d . . Th for this is that any row echelon form of the coefficient matrix will involve r ::s m
are no POintSthat lie Onall th r oe~ not pass through this point. us,
nonzero rows. Thus, there will be r lead variables and n - r free variables, where
ree mes (see Figure 1.2.1).
n-r :::n-m > O. If the system is consistent, we can assign the free variables arbitrary
values and solve for the lead variables. Therefore, a consistent underdetermined system
System (b): 2I
3
-I 2 I 2I] II 0 2I I
I 0 I) will have infinitely many solutions.

4 34->00i3 EXAMPLE 5 Solve the following underdetermined systems:

2 -I 3 5 2
Using back substitution 0 0 0 0
(a) XI + 2x, + X3 = I (b) XI + X, + X3 + X4 + Xs = 2
(0. I. -0 3 I 5) Th ' we see that sy t . . 2xI + 4x, + 2X3 = 3 XI + X, + X3 + 2X4 + 2xs = 3
. , '. e solution is uni bees em (b) has exactly one solunan.
que ause the nOnzero rows of the reduced
XI + X, + X3 + 2X4 + 3xs = 2
16 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form 17
Solution
The following matrices are in reduced row echelon form:
System (a): [~; ~ Ii) -+ (~ ~ ~ I: J L-- L L r--
0 o 23) , 1o 20 0I 3I)
Clearly, system (a) is inconsistent. We can think of the two equations in system (a) and
(b) as representing planes in 3-space. Usually, two planes intersect in a line; however,
in this case the planes are parallel.
fNJ o
1 0
0 I I
0o 0I 02 OJI
[o 0

The process of using elementary row operations to transform a matrix into reduced row
0 0
,
[ 000 0

echelon form is called Gauss-Jordan reduction.

System (b): [: : 112] [1111
223 -+ 000 I
I I 232 0000 EXAMPLE 6 Use Gauss-Jordan reduction to solve the system
System (b) is consistent, and since there are two free variables, the system will have
mfimtely many solutions. In cases such as these it is convenient to continue the elimi- -Xl + X2 - X3 + 3X4 = 0
nation process and simplify the form of the reduced matrix even further. We continue 3x, + X2 - X3 - X4 =0
elintinating until all the terms above each leading 1 are eliminated. Thus, for sys- 2x] - X2 - 2X3 - X4 =0
tem (b), we will continue and eliminate the first two entries in the fifth column and
then the first element in the fourth column, as follows:
Solution
I I I
I I 1 0 (']1
[Io 0I
o
0 I I
0 0 0 1 -1 ij-+ [~ 0 0 1 0
0 0 0 I J) [ -~
1 -1 3
1 -I -I
-1 -2 -1
g) -+ [ -~
I -I
4 -4
1 -4
3
8
g)
° j]
I I 0 5

-+ [t 0 0 I 0
0 0 0 I
If we put the free variables over on the right-hand side, it follows that -+ [-t 1 -1
4 -4
0 -3
3
8
3
g) -+ [~
-I
1 -1
0
1 -3

1 -1
2
~) row
echelon
form
Xl = I- X2 - X3
-1 0 -2 0 0 -I reduced
=2
-+ [g g) -+ [t
X4

X5 =-1
1
0
0
1 -1
1 1
0
0
1 -I
1
gj row echelon
form
Thus, for any real numbers C/ and {J, the 5-tuple
If we set X4 equal to any real number a, then XI = C/, X2 = -C/, and X3 = a. Thus, all
(I - C/ - {J, C/, {J, 2, -1) ordered 4-tuples of the form (a, -C/, C/, C/) are solutions of the system. •
is a solution of the system. •

In the case where the row ech I APPLICATION I Traffic Flow

the standard procedure' t . e on form of a consistent system has free variables,
In the downtown section of a certain city, two sets of one-way streets intersect as shown
each leading I have b IS °lcontmue the elimination process until all the entries above
. een e 'mmated as in s t (b I The in Figure 1.2.2. The average hourly volume of traffic entering and leaving this section
resu Itmg reduced matrix is said bei ys em ) of the previous examp e. during rush hour is given in the diagram. Determine the amount of traffic between each
to e m reduced row echelon form.
of the four intersections.
Reduced Row Echelon Fonn
Solution
Definition
A matrix is said to be . At each intersection, the number of automobiles entering must be the same as the
• 10 reduced row echelon fonn .f
number leaving. For example, at intersection A, the number of automobiles entering is
(~~ ~e matrix is in row echelon form. I Xl + 450 and the number leaving is X2 + 6JO. Thus,
e first nonzero entry .
In each row is the only nonzero entry in its column.
Xl + 450 = X2 + 610 (intersection A)
18 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form 19

APPLICATION 2 Electrical Networks

In an electrical network, it is possible to determine the amount of current in each branch
in terms of the resistances and the voltages. An example of a typical circuit is given in
Figure 1.2.3.

8 volts

4 ohms
2 ohms
A ~---A,M-~---'
i,
B

3 ohms 2 ohms

9 volts

Figure 1.2.2. Figure 1.2.3.

The symbols in the figure have the following meanings:

Similarly,
A path along which current may flow
+ 520 = x3 + 480 (intersection B)
X2

X3 + 390 = x. + 600 (intersection C) ,

---II An electrical source

Th x. + 640 = x, + 310 (intersection D) ------wY"-- A resistor

e augmented matrix for the system is
The electrical source is usually a battery (with a voltage measured in volts) that drives

[ -:
~
o 0
-~
I-I
~
210
160]
-40
a charge and produces a current. The current will flow out from the terminal of the
battery that is represented by the longer vertical line. The resistances are measured in
ohms. The letters represent nodes and the i's represent the currents between the nodes.
-I 0 0 I The currents are measured in amperes. The arrows show the direction of the currents.
-330 If, however, one of the currents, say 12, turns out to be negative, this would mean that
The reduced row echelon form for this matrix is the current along that branch is in the direction opposite that of the arrow.
To determine the currents, the following rules are used:

[ o~ ~ ~ =: ~~~)
0 I -I 210
Kirchhoff's Laws
1. At every node, the sum of the incoming currents equals the sum of the outgoing
The. 00000
system ISconsistent and . currents.
l' , Smceth'
so utlOns.The traffic flow di ere IS a free variable there are many possible 2. Around every closed loop, the algebraic sum of the voltage gains must equal
X2 x ad· iagram does not . . ' .
, .3, n x. Umquely If th give enough mformation to deterrmne XI, the algebraic sum of the voltage drops.
secnons th . e amount of traffi . r
. ' e trafficon the remainin . c were known between any pair of inte.
If the amount of traffic between in g ane~es could easily be calculated. For example, The voltage drops E for each resistor are given by Ohm's law,
hour, then x. '= 200. Using this valtersectlOnsC and D averages 200 automobiles per E = iR
ue, we can the I c
n so ve lor x"
X2, and X3:
where i represents the current in amperes and R the resistance in ohms.
x, '= x. + 330 ee 530 Let us find the currents in the network pictured in Figure 1.2.3. From the first law,
X2 '= x. + 170,= 370 we have
X3 '= X. + 210 '= 410 I, ~ 12 + ts = 0 (node A)
-i, + i2 - i3 = 0 (node B)
-
 20 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form 21
By the second law,
carbon dioxide contains one carbon atom and glucose contains six, to balance the car-
4il + 2i2 =8 (top loop) bon atoms we require that
2i, + 5i3 = 9 (bottom loop)
The network can be represented by the augmented matrix
Similarly, to balance the oxygen, we need
-I I
I -1

° I]
2
and finally, to balance the hydrogen, we need
2 5
This matrix is easily reduced to row echelon form:

Ii -j1
If we move all the unknowns to the left-hand sides of the three equations, we end up
with the homogeneous linear system

fJ XI

2XI + X2 - 2X3 -
- 6X4

6X4
=a
=a
Solving by back substitution, we see that i, = I i = 2 and"
' 2 ,
_ I
13 - ,_
2x, - 12x4 = °
By Theorem 1.2.1, the system has nontrivial solutions. To balance the equation, we
Homogeneous Systems must find solutions (x" X2, X3, X4) whose entries are nonnegative integers. If we solve
A system of linear equations" id b the system in the usual way, we see that X4 is a free variable and
hand side are II IS Sal to e homogeneous if the constants on the right-
a zero Homogen 0 " ial
matter to find a sol t"". e us systems are always consistent. It is a mVI
lOn
mooeneous systemUh ;Just set all the variables equal to-zero. Thus, if an m x n ho-
o as a umque sol ti " 0)
The homogeneous s st . E u ron, It must be the trivial solution (0, 0, ... , . In particular, if we take X4 = I, then XI = X2 = X3 = 6 and the equation takes the form
knowns In the c
.
Yhem In xample 6 consIsted of m = 3 equations in n
ase t at n > m the ill I
4 un-
tl
=
additional nontrivial sol Uti Th' re WI a ways be free variables and, consequen y.
of underdetermined t IOns · ISresuit has essentially been proved in our discussion
sys ems but beca f. .
, , use 0 ItS Importance, we state it as a theorem.
Theol'em 1.2.1 APPLICATION 4 Economic Models for Exchange of Goods
An III x It homogeneous system of r . Suppose that in a primitive society the members of a tribe are engaged in three oc-
mear equatrong has a nontrivial solution if n > m.
Proof cupations: farming, manufacturing of tools and utensils, and weaving and sewing of
A homogeneous system is alwa s .
clothing. Assume that initially the tribe has no monetary system and that all goods and
have at most m nonzero ro Th consIstent. The row echelon form of the matrix can
services are bartered. Let us denote the three groups by F, M, and C, and suppose that
It variables altogether and:~ III ~~there are at most m lead variables. Since there are
the directed graph in Figure 1.2.4 indicates how the bartering system works in practice.
can be assigned arbitrary I ' ere must be Some free variables. The free vanables
h . va ues. For each . . bl s The figure indicates that the farmers keep half of their produce and give one-fourth
t ere IS a solution of the s assIgnment of values to the free vana e.
ystem. • of their produce to the manufacturers and one-fourth to the clothing producers. The
manufacturers divide the goods evenly among the three groups, one-third going to
APPLICATION 1 each group. The group producing clothes gives half of the clothes to the farmers and
Chemical Equations
divides the other half evenly between the manufacturers and themselves. The result is
In the process of photosynthesis I. _ summarized in the following table:
carbon dIoxide (CO) ad' p ants use radlem energy from sunlight to convert
h . 2 n water (H 0) . Th
c emlcal equation of the reaction' 2 f hlnto glucose (C6H1206) and oxygen (0,). e F M C
IS 0 t e form

x,C02 + x,H,O ~ x °+ F , I I
3 ,
I

To balance the equatJ' 3 2 X4C6HI206 M I I I

f 4 3 4
b on, we must ch
car on, hydrogen, and oxygen at oose XI, X" X3, and
so that the numbelS 0
X4
C I , I
oms are the s . . S'nce 4 3 4
arne on each Side of the equanon I
22 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form 23

Finally, using the third row of the table, we get

These equations can be rewritten as a homogeneous system:

The reduced row echelon form of the augmented matrix for this system is

o
1
Figure 1.2.4. o
There is one free variable: X3· Setting X3 '= 3, we obtain the solution (5, 3, 3), and the
The first column of the table' di general solution consists of all multiples of (5, 3, 3). It follows that the variables Xl,
fanners the second col ' di icates the distribution of the goods produced by the x2, and X3 should be assigned values in the ratio
, urnn In rcates th di ib '
the third column indicates th di ib ,e istn utlOn of the manufactured goods, and
e istn unon of th I thi
As the size of th t ib ' e c a mg. Xl : X2 : X3 '= 5 :3:3
en egrows the s t fb '
and, consequently the trib d 'd' ys em a anenng becomes too cumbersome
, , ' e eel es to rnstir t
this simple economic syst I u e a monetary system of exchange, For This simple system is an example of the closed Leontief input-output model.
or debt and that the pri o;m, we assume that there will be no accumulation of capital Leontief's models are fundamental to our understanding of economic systems. Mod-
, ces lor each of the th f
the eXisting bartering system. The u ' ree types of goods will reflect the values a ern applications would involve thousands of industries and lead to very large linear
of goods that fairly represent the q esnon IS how to assign values to the three types systems. The Leontief models will be studied in greater detail later, in Section 8 of
The problem can be tum d ,current banenng system. Chapter 6.
e mtoahnear t· .
rno e I th at was originally d I
d sys em of equations using an economic
Leontief. For this model eve'lolPed hy the Nobel Prize-winning economist Wassily
h h ' we WI let x he th d
y I e farmers, X2 be the val f hie monetary value of the goods produce
the clolhing produced A
.
ude.o t e manufactured goods and X3 be the value of all
. ccor 109 to th fi '
SECTION 1.2 EXERCISES
received by the farmers erst row of the table the value of the goods I. Which of the matrices that follow are in row eche-
. amounts to half th ' 2. The augmented matrices that follow are in row ech-
one-third the value of th e value of the farm goods produced pius Ion form? Which are in reduced row echelon form? elon form. For each case, indicate whether the cor-
od e manufactured d ' .
go s. Thus, the total value of oct pro ucts and half the value of the clothmg responding linear system is consistent. If the sys-
the system is fair, the total value ~~ s received by the farmer is ~Xl + j X2 + ~X3' If tern has a unique solution, find it.
total value of the farm goods prod goodds receIved by the farmers should equal XI, the
uce . Hence , w eave h t hee Ii
hnear equauon.
(a)
lb
2 3
0 I ~1 (b)
[~ooo ~ 00]
(a)
[~ 11 n (b) [~
!1-1]
3
~X
21
+3 +2'
x2
1
X3'=XI
I (e)
[~ !]0
0
(d)
[~~] (e)
[~
-2
0
0
4
I
0 I ~]
Using the second 1 4
received b row of the tahle and e .
y the manufacturers we b . quatmg the value of the goods produced an
d (e)
[~ ] I
0 .~
(0
[~ !J
0
I (d)
[~
-2
I
2
-I
n]
:]
, 0 tam a second equation:

n,
0 0 1 1 3 0 1
I
(g) [~
0
0
I
2
3
(h) [~ 0
0
1
0 (e) [~
3
014 21-2]
o 0 I
 24 Chapter I Matrices and Systems of Equations
1.2 Row Echelon Form 25
-I 3
1 2 (I) x, - x, + 2x] = 4 8. Consider a linear system whose augmented matrix (b) x, + 2x, + x, = -I
o + 3x, - x] = I
o
I
o 11 2x,
?X, + 3x, + 4x] = 7
is of the form
-Xl -X2 +

2x, + 3x,
2X3 = 2
=-2

[-l
3. The augmented matrices that follow are in reduced 2
(g) x, + x, + x] + .r. = 0 solve both systems by computing the row echelon
row echelon form. In each case, find the solution 4
2x, + 3x, - x, - x. = 2 form of an augmented matrix (A I B) and perform-

Trlo~rlg [7~~I;1 YS -2
3x, + 2x, + x, + x. = 5 ing back substitution twice.
;~; Of[~e line;b:
3x, + 6x, - x, - x. = 4 For what values of a will the system have a unique
0013 0001 13. Given a homogeneous system of linear equations,
(h) x, - 2x, = 3 solution?
if the system is overdetermined, what are the possi-
-3 2x, + = I x, 9. Consider a linear system whose augmented matrix bilities as to the number of solutions? Explain.
(c) [~ o
o r I-n (I)
-5x, =4 + 8x,
-x, + 2x, - x, = 2
is of the form
14. Given a nonhomogeneous system of linear equa-

I; J tions, if the system is underdetermined, what are

jl~1
2
(d) (~ ~ ~ ~ -2x, + 2x, + = x] 4
5 the possibilities as to the number of solutions? Ex-
3x, + 2x, + 2x, = 5 plain.

I-iii iI
I
-3x, + 8x, + 5x] = 17
('J Ii (j)
-XI
x, + 2x, - 3x] + x. = I
- X2 =6 + 4x] - x ..
(3) Is it possible for the system to be inconsistent?
Explain.
15. Determine the values of Xl, X2,
following traffic flow diagram:
X3, and X4 for the

~~I-~l
o 0 0
(k)
-2x, - 4x, + 7x] - x, = 1
x,
2x, - 2x,
+ 3x, + x] +
+ x] + 2r.
x. = 3
=8
(b) For what values of f3 will the system have in-
finitely many solutions?

10. Consider a linear system whose augmented matrix

4. :~::a~h of the systems in Exercise 3 make a list XI - 5X2 + X4 =5 , is of the form
. e ead variables and a second lis; of th f
variables. e ree (I) x, - 3x, + x, = I
S. For each of th 2x, + x, - x, = 2
use Gaussian el·sy~te~s of equations that follow
e lO11natlOnto obtain . ' x, + 4x, - 2x, = I
system whose coeffi . an eqUIvalent
form Indi crent matrix is in row echelon 5x, - 8x, + 2x] = 5
. Icate whether the .
the system is' s~stem IS consistent. If 6. Use GauSS-Jordan reduction to solve each of the (a) For what values of a and b will the system have
ables, Usebac~~~~::i:;' and Involves no free van- following systems; infinitely many solutions?
tion. If the system is 10~to find the unique solu.
variables, transform it ~ons~stentand there are free
(a) x, + x, = _I
4x, - 3x, = 3
(b) For what values of a and b will the system be
inconsistent?
- 540
- X3

-400

and find all solutions. 0 re uced row echelon form

(a) x, - 2x, =3 (b) + 3x, + x] + x. = 3
x, 11. Given the linear systems
2xI - X2 == 9
(b)
2x, - 3x,=5 + x] + 2x. = 8
2x, - 2x, (a) x, + 2x, = 2 (b) x, + 2x, = I
(c)
-4x, + 6x, = 8 3x, + x, + 2x, - x, = _I 3x, + 7x, = 8 3x, + 7x, = 7
XJ + X2:;:: 0
(c) x, + x, + x] = 0
2x, + 3x, = 0 solve both systems by incorporating the right-hand
XI - X2 - X3 =0 sides into a 2 x 2 matrix B and computing the re- 16. Consider the traffic flow diagram that follows,
3x, - 2x, = 0 where QI, G2, Q3, a4, hI, b2, b3, b4 are fixed positive
duced row echelon form of
(d) + 2x, - x, = 4
3x,
(d) x, + x, + x] + x, =0 integers. Set up a linear system in the unknowns Xl,

x, - 2x, + 2x, = 1
2x, x, + - x, + 3x, =0 X2, X3, X4 and show that the system will be consis-

l1x, + 2x, + x, = 14
7. Give a ge
x, - 2x,
.
+ x, + X4 =0
(AIB)= [13 27128
~1 tent if and only if

. Ornetnc explanation of why a homoge-

(e) 2x, + 3x, + x, = I neous linear sy t· . in 12. Given the linear systems
th s em consisting of two equauons a, + a, + a, + a. = b, + b, + b, + b
XJ + X2 + X3 == 3
4
. ree unknowns must have infinitely many solu-
t,ons What th . '"
(a) x, + 2x, + X3 =2
3x, + 4x, + 2x, = 4 f· are e possible numbers of soluuo -Xl - X2 + 2X3 =3
o a non~omogeneous 2 x 3 linear system? Givea
geometric expl .
3Oatlon of your answer.
2x, + 3x, =0 What can you conclude about the number of auto-
 26 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 27
mobiles entering and leaving the traffic network?
these reactions are measured in moles (a standard
unit of measurement for chemical reactions), Hew
many moles of nitrogen. hydrogen, and oxygenare
III Matrix Arithmetic
necessary to produce 8 mole of nitric acid? In this section, we introduce the standard notations used for matrices and vectors and
21. In Application 4. determine the relative valuesof define arithmetic operations (addition, subtraction, and multiplication) with matrices.
x" x" and x, if the di tribution of goods is as de. We will also introduce two additional operations: scalar multiplication and transposi-
scribed in the following table: tion. We will see how 10 represent linear systems as equations involving matrices and
vectors and then derive a theorem characterizing when a linear system is consistent.
FMC
The entries of a matrix are called scalars. They are usually either real or complex
F I I I
numbers. For the most part, we will be working with matrices whose entries are real
j 3 J
numbers. Throughout the first five chapters of the book, the reader may assume that
M ~ i ~ Ihe term scalar refers to a real number. However, in Chapter 6 Ihere will be occasions
ell I
3 6 i when we will use the set of complex numbers as our scalar field.
22. Determine the amount of each Current for the fol-
lowing networks: Matrix Notation
(a)
16 volts
If we wish to refer to matrices without specifically writing out all their entries, we will
17. Let (CI, c,) be a solution of the 2 x 2 system use capital leiters A, B, C, and so on. In general, a;j will denote the entry of the matrix
A that is in the ith row and the jth column. We will refer to this entry as the (i, j)
GIIXI + Gj2X2 = 0 2 ohms entry of A. Thus, if A is an m x n matrix, then
G21Xj + G22X2 = 0 i, 2 ohms
At----=---.J.,AIIr---..S
s(how that,for any real number c, the ordered pair
aCI, aC2) IS also a solution.

18. In Application 3, the solution (6 6 6 I)

taidb' '"
ne y settmg the free variable X4 = l.
was ob-
b
is Lohms
A=
["'
a2l
al2
an
""I
a211

al~l Qm2 all/II

(a) Detenni~e the solution corresponding to X4 = (b)

0: What mformation, if any, does this solution t, We will sometimes shorten this to A = (aU)' Similarly, a matrix B may be referred to
~I~e,a~oUlt~e chemical reaction? Is the term as (b;j), a matrix C as (cU)' and so on.
(b) ~~vla solutIOn"appropriate in this case?
II oose SOmeother valuesof X4 such as 2 4
5 and deternj] h ' , .or Vectors
How ne t ~ ~orresponding sOlutions,
" are these nOntnvlalsolutions related?
19. LiqUidbenzene burns in the at . Matrices that have only one row or one column are of special interest, since they are
object is placed di I mosphere. If a cold used to represent solutions of linear systems. A solution of a system of m linear equa-
will condense on t,:ct bY'Over the benzene, Water tions in n unknowns is an n-tuple of real numbers. We will refer to an n-tuple of real
(carbon) will also foe01ectandadepo'
h su 0 f soot (c) 'v'
numbers as a vector. If an n-tuple is represented in terms of a I x n malrix, then we
eaI·equation for thi nnont'. eob;ect Th e cherru-.
J •
8 volts
,I i, will refer to it as a row vector. Alternatively, if the n-tuple is represented by an II x I
s reacuon IS of the form
'I matrix, then we will refer to it as a column vector. For example, the solution of the
XlC6H6 + X202 -? x3c +x H 04 2 linear system
Detennine values of x ~4 ohms
the equation, I, X2, X3, and X4 to balance
2.ohms i,
~> Xl + X2 =3
20, Nitric acid is prepared .
commercially bY a series
. of
A or-----'\..A..IIr-~ _
'VVv _e s Xl - X2 = I
h hemi
tree c emrcal reaction I h
s. n t e firstr ti
gen (N2) is combined with h d eae Ion, nitro-
ammonia (NH,). Next th y rogen (H,) to form
can be represented by the row vector (2, I) or the column vector [ ~ J.
e 4.ohms i,
with oxygen (0,) to <o'nn ~mmol1laIS combined C In working with matrix equations, it is generally more convenient to represent the
-
and Water.Finally the NO
nurogen d' ,
II
)OXlde(NO,) t-----'J'NV·V'v--2....- ...V solutions in terms of column vectors (n x I matrices). The set of all n x I matrices of
, 2 reacts with s
water to form nitric acid (HNO ) orne of the real numbers is called Euclidean n-space and is usually denoted by IR". Since we will
(NO). The amounts of each Of:h and nitric oxide ~50hms
~ be working almost exclusively with column vectors in the future, we will generally
e components of
omil the word "column" and refer 10 the elements of IR" as simply vectors, ralher than
I'
28 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 29
as column vectors. The standard notation for a column vector is a boldface lowercase
letter, as in EXAMPLE I If

(I)
A=
[-i ~J 2
8

then

For row vectors, there is no universal standard notation. In this book, we will
a, = [ -il ' 3Z = [n, 33 = [~)
represent both row and column vectors with boldface lowercase letters and to distin. and
guish a row vector from a column vector we will place a horizontal arrow above the
letter.. Thus, the horizontal arrow indicates a horizontal array (row vector) rather than 1\1 =
a vertical array (column vector).
For example,
(3, 2, 5), 3z = (-I, 8,4)
•
Equality
i, ("'"",) and y, fEI For two matrices to be e ual, they must have the same dimensions and their corre-
sponding entries must agree.
are row and column vectors with four entries each.
Given an m x n matri A it i f
I h x ,I IS 0 ten necessary to refer to a particular row or Definition Two m x n matrices A and B are said to be equal if au = bU for each i and j.
coumn·IIT e standard notation for the jth column vector of A is 3 i, There is no
umversa Y accepted standard tati C ) tho
b00 k , Since
. .
we Use honzontal no anon lor. the
. ith row vector of a matrix A. In IS
vector of A by 3;. arrows to indicate row vectors, we denote the ith row
If A is an m x n matn th h Scalar Multiplication
nx, en t e ~ of A are given by
If A is a matrix and" is a scalar, then "A is the matrix formed by multiplying each of
3j = (ail, ai2, ... , aill) i-I
the entries of A by a,
- , ... ,m
and the column vectors are given hy

Definition If A is an m X n matrix and" is a scalar, then "A is the m x n matrix whose (i, j)

I
entry is aaij.
alj
a2} J
= .
3j
j = 1, ... ,n
For example, if
am}
8
Themt·A A = [:
a nx can be represented in te .. . 8 I~ )
rms of either Us column vectors or its row vectors.
then
1 4 24
A -- (a l,a2,· .. ,a
-A=
2 [; 4 ~J and 3A _ [12
- 18 24 3g J
ll
)

Or A=ll:j
B is an 3m
Similarly, if . Matrix Addition
n x r matrix, then
Two matrices with the same dimensions can be added by adding their corresponding
entries.

Definition If A = (aU) and B = (bu) are both m x n matrices, then the sum A +B is the
m x n matrix whose (i, j) entry is au + b., for each ordered pair (i, j).
30 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 31
For example,
Case 1. One Equation in Several Unknowns
Let us begin by examining the case of one equation in several variables. Consider, for
example, the equation
3x, + 2X2 + 5X3 = 4
If we set

If we define A - B to be A + (- I) B, then it turns out that A _ B is formedby

subtracting the corresponding entry of B from each entry of A. Thus,
A = (3 2 5] and x = Ud
and define the product Ax by
U ~J-(~ ~J=(~ ~J+(-I)(~ ;J
= (23 4I J + (-4-2-3-5 J
= (2-4
3-2
4-5
1-3
J then the equation 3x, + 2X2 + 5X3 = 4 can be written as the matrix equation

If 0 represents the matri .th h

=
.
(-i =~J
For a linear equation with n unknowns of the form
Ax =4

then x, WI t e same dlmensions as A, whose entries are all 0,

A+O=O+A=A
We will refer to 0 as the zero' '
m x n matrices F rth matrix. It acts as an additive identity on the set of all if we let
. u ermore each m x . d
' n matnx A has an additive inverse. Indee ,
A+(-I)A,=O=(-l)A+A A= (a, a2 , .. au] and x=
It is Customaryto de t h ",
no e t e addItIve inverse by -A, Thus,

-A = (-I)A and define the product Ax by

Matrix Multiplication and Linear Systems

We have yet to define the most im on ' . then the system can be written in the form Ax = b.
Much of the motivation behind t'he ;nt operatIon: the multiplication of two maWces, For example, if
systems of equations If w h efimtlOn comes from the applications to lweal
be w-: . e ave a system f ' It
can e written in the form 0 one linear equation in one unknown,

aX=b (2) A = (2
We generally think of a x and b bei
as I x I matrices. Our ~O~In as emg scalars; however, they could also be treated
m n is
an x linear system by a s~nw to generalize equation (2) so that we can represenl then
g e malnx equatIOnof the form
Ax = 2·3 + I· 2 + (-3)· I +4· (-2) =-3
Ax= b
where A is an m x n matri ' Note that the result of multiplying a row vector on the left by a column vector on the
first the case of one equati~' x ISan unknown vector in lRn, and b is in lRm. We consider right is a scalar. Consequently, this type of multiplication is often referred to as a scalar
n in several unknowns. product.
32 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 33

Case 2. M Equations in N Unknowns EXAMPLE 3

Consider now an In x n linear system

QljX! + QI2X2 + + Qj"Xn = hi A = [-~ ~], X = [~]

Q21XI + a22X, + + Q'nXn = b2

(3)
Ax = r -~ ~ : ~ :: ]
4·2+2·4
= [~~6] •
It is desirable to write the system (3) in a form similar to (2)-that is, as a matrix
equation EXAMPLE 4 Write the following system of equations as a matrix equation of the form Ax = b:
Ax= b (4)
3x, + 2x, + X3 = 5
where A = (au) is known, x is an n x I matrix of unknowns, and b is an In x I matrix
representing the right-hand side of the system. Thus, if we set XI - 2X2 + 5X3 = -2
2x, + X, - 3x] = l
a" aln XI

I"'
a21 a22 a2n Solution
X,
A= x= b=
2
Q'~I Qm2

and define the product Ax by

... Qmn Xn
[! -2
I •
An alternative way to represent the linear system (3) as a matrix equation is to
QljXj + Ql2X2 + + GlnX" express the product Ax as a sum of column vectors:

Ax=
a21xI + a22X, + + a'nxn
(5l allXj + GI2X2 + + QlnXn

Q21Xl + a22X2 + + a2nXn

Ax=
QmjXI + Qm2X2 + ... + GmnXn
[

then the linear system of equations (3) is equivalent to the matrix equation (4). a",IX' + a"'2X2:+'" + a",nxn
Given an In x n rnatnx A and a vector x in JRn, it is possible to compute a product all
Ax by (5) '. The product Ax will be an In x I matrix-that is, a vector in JRm. The rule a21
for determmmg the i th entry of Ax is =Xl .

QilXl + Qi2X2 + ... + QinXIl

[
.;
which is equal to aiX, the scalar product of the ith row vector of A and the column Thus, we have
vector x. Thus,
(6)
Using this formula, we can represent the system of equations (3) as a matrix equation
of the form
Ax =
(7)

EXAMPLE 2 EXAMPLE 5 The linear system

2xI + 3x, - 2x] =5

2l
5xI - 4X2 + 2x] =6
A=[4 J5 3 7 ' can be written as a matrix equation

• •
34 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 35

Definition If a" a2, ... , a" are vectors in IRm and c" c-; ... , Cn are scalars, then a sum of the Matrix Multiplication
form
More generally, it is possible to multiply a matrix A times a matrix B if the number
Cia, + czaz + ... + cnan
of columns of A equals the number of rows of B. The first column of the product is
is said to be a linear combination of the vectors a.. az, ... , an' determined by the first column of B; that is, the first column of A B is Ab, the second
column of A B is A bz, and so on. Thus the product A B is the matrix whose columns
are Ab" Abz, ... , Abn:
It follows from equation (6) that the product Ax is a linear combination of the
column vectors of A. Some books even use this linear combination representation as
AB = (Ab], Ab2, ...• Ab,,)
the definition of matrix vector multiplication.

The (i, j) entry of AB is the ith entry of the column vector Abj. It is determined
If A is an In x n matrix and x is a vector in IR", then by multiplying the ith row vector of A times the jth column vector of B.

Definition If A = (G;J) is an In x n matrix and B = (bU) is an n x r matrix, then the product

AB = C = (cU) is the In x r matrix whose entries are defined by
EXAMPLE 6
If we choose x] = 2, X2 = 3, and X3 = 4 in Example 5, then
"
Cij = 3ib = L
j aikbkj
(~J =2 (; J +3 ( -~ J +4 [ -; J k=[

Thus, the vector (5 J"

6 IS
'.
a linear combination of the three column vectors of the coef·
EXAMPLE 8 If
ficient matrix. It follows that the linear system in Example 5 is consistent and
3 -21
A = 2 4
[
I -3
x = [~)
is a solution of the system. then
•
te
The matrix equation (7)

of (7).
f . .
id . .
. provi es a ruce way of characterizing whether a linear sys·
m a equations IS consistent I d d h f .
. n ee , t e ollowlllg theorem is a direct consequence
,
AB = [~ =~1 [-~
3. (-2) -2·4
: ~J
3·]-2·1 3'3-2.61
Theorem 1.3.1 2· (-2) + 4·4 2·1+4·] 2·3+4·6
Consistency Theorem for Linear Systems [ 1·(-2)-3·4
I . I - 3 . I 1·3 - 3·6

=[-:~ ~ ~1
A linear system Ax = b is co". .'.
nation of the column nsrstenr If and only If b can be written as a linear combl
vectors of A.
EXAMPLE 7 The linear system
-14 -2 -15

The shading indicates how the (2,3) entry of the product A B is computed as a scalar
x] + 2xz =I
product of the second row vector of A and the third column vector of B. It is also
2x] + 4X2 =I possible to multiply B times A, however, the resulting matrix BA is not equal to AB.
is inconsistent, since the vector [ I J. . e In fact, AB and B A do not even have the same dimensions, as the following multipli-
I cannot be wntten as a linear combination of Ih cation shows:
column vectors ( I J and (21
-2·3 + I ·2 + 3 . I
w ld b 2 4 . Note that any linear combination of these veclOts BA = ( -2.(-2)+1.4+3.(-3)1
au e of the form 4·3+1·2+6·1 4· (-2) + I ·4 + 6 . (-3)
J
dh
x] UJ + X2 (24 J = ( 2x, ++ 4xz J
XI 2X2
-I
( 20
-I
-22 •
an ence the second entry of th
e Vector must be double the first entry. •
36 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 37
EXAMPLE 9 If Table 2 Amount Produced per Quarter
Season
A= [~~] and B= [~ ~] Product Summer Fall Winter Spring
A 4000 4500 4500 4000
then it is impossible to multiply A times B, since the number of column of A doesnot B 2000 2600 2400 2200
equal the number of rows of B. However, it is possible to multiply B times A. C 5800 6200 6000 6000

BA = [~ ;] [~ ~] = [ 1~
3 6
2~
15 24
1 • Solution
Let us consider the problem in terms of matrices. Each of the two tables can be repre-
sented by a matrix, namely,
If A and B are both II x II matrices, then A Band BA wiII also be n x II matrices.
but, in general, they will not be equal. Multiplication of matrices is not commutative. 0.10 0.30 0.15]
M = 0.30 0.40 0.25
EXAMPLE 10 If r 0.10 0.20 0.15

A=[OIOI]
and
and B = [I2 2I] 4000 4500 4500 4000]
P = 2000 2600 2400 2200
then r 5800 6200 6000 6000
If we form the product MP, the first column of M P will represent the costs for the
AB = [6 6] [i i] summer quarter:
and Raw materials: (0.10)(4000) + (0.30)(2000) + (0.15)(5800) = 1870
Labor: (0.30)(4000) + (0.40)(2000) + (0.25)(5800) = 3450
BA = [i i] [6 6] [i i) =
Overhead and
miscellaneous: (0.10)(4000) + (0.20)(2000) + (0.15)(5800) = 1670
Hence, AB of BA.
• The costs for the fall quarter are given in the second column of MP:

APPLICATION I Raw materials: (0. I0) (4500) + (0.30)(2600) + (0.15)(6200) = 2160

Production Costs Labor: (0.30)(4500) + (0.40)(2600) + (0.25)(6200) = 3940
A Companymanufactures three d. ;- Overhead and
categories In each
.
pro ucts. Its productIOn expenses are divided intot ret
category an ti '. . Ie
miscellaneous: (0.10)(4500) + (0.20)(2600) + (0.15)(6200) = 1900
item of each product A '. esumars IS given for the cost of producing a slOg
produced per quarter' Thn estimate IS also made of the amount of each product to~ Columns 3 and 4 of MP represent the costs for the winter and spring quarters, respec-
meeting, the compa~ w~~~;stlmates are given in Tables I and 2. At its stockholders tively. Thus, we have
each quarter in each of th h like to present a single table showing the total costs for
e tree categones: raw materials, labor, and overhead. 1870 2160 2070 1960]
MP = 3450 3940 3810 3580
Table I Pd'
ro uctron COSts per Item (dollars) r 1670 1900 1830 1740

Expe ..... Product The entries in row I of MP represent the total cost of raw materials for each of the four
A B quarters. The entries in rows 2 and 3 represent the total cost for labor and overhead,
Raw materials c respectively, for each of the four quarters. The yearly expenses in each category may
Labor 0.10 0.30 0.15 be obtained by adding the entries in each row. The numbers in each of the columns
0.30 0.40 may be added to obtain the total production costs for each quarter. Table 3 summarizes
Overhead and miscellaneous 0.25
0.10 0.20 the total production costs. •
0.15
 38 Chapter I Matrices and Systems of Equations

Table 3 1.3 Matrix Arithmetic 39

Season
Summer Fall Winter Spring Year
Raw materials
1,870 2,160 2.070 The matrix C in Example II is its own transpose. This frequently happens with
Labor 1,960 8,060
3,450 3,940 matrices that arise in applications.
3,810 3,580 14,780
Overhead and miscellaneous 1,670 1,900 1,830
Total production costs
6,990 8,000 -
7,710
1,740
7,280
7,140
29,980
Definition An n X n matrix A is said to be symmetric if AT = A.

Notational Rules The following are some examples of symmetric matrices:

Just as in ordinary algebra if '. 2 3 1

and there are no p th'
carried out before additi
an expressIOn Involves both multiplication
aren eses to mdlcate the a d
Thi
I' h
r er ate
.
and addition
. . .
operations, multiplications are
[6 -~ J [
3
4
I
5 -2
I

For example, if I Ions. IS IS true for both scalar and matrix multiplications.

APPLICATION 2 Information Retrieval

A = [3 4J The growth of digital libraries on the Internet has led to dramatic improvements in the
I 2 '
storage and retrieval of information. Modern retrieval methods are based on matrix
then theory and linear algebra.
In a typical situation, a database consists of a collection of documents and we wish
A+BC=[3 4)+[ 77J_[10
and
I 2 -I 4 - 0 J
l~
to search the collection and find the documents that best match some particular search
conditions. Depending on the type of database, we could search for such items as
research articles in journals, Web pages on the Internet, books in a library, or movies
in a film collection.
To see how the searches are done, let us assume that our database consists of m
documents and that there are n dictionary words that can be used as keywords for
The Transpose of a Matrix searches. Not all words are allowable, since it would not be practical to search for
Given an m x n matrix A it i < common words such as articles or prepositions. If the key dictionary words are or-
' I IS etten useful t c
are the rows of A. a rorrn a new n x m matrix whose columns dered alphabetically, then we can represent the database by an m x n matrix A. Each
document is represented by a column of the matrix. The first entry in the jth column
Definition of A would be a number representing the relative frequency of the first key dictionary
The transpose of an m x n matrix .
word in the jth document. The entry ai, represents the relative frequency of the sec-
A IS the n x m matrix B defined by
ond word in the jth document, and so on. The Jist of keywords to be used in the search
is represented by a vector x in 1R'". The dh entry of x is taken to be I if the ith word
(8) ,
for j = I, ... , nand i = I I in the Jist of keywords is on our search Jist; otherwise, we set X; = O. To carry out the
'--------.:--=:..:.:.' :":"':".:.'
m~. ~T~h~e
~tr~a~nspose
of A is denoted by AT ..J search, we simply multiply A T times x.

II fallows from (8) h - - Simple Matching Searches

jth colum t at the jth row of A T h . .
ith n of A, and the ith column I' AT as the same entnes, respectively, as the
The simplest type of search determines how many of the key search words are in each
I row of A 0 has th . . I the
. e same entnes, respective y, as
document; it does not take into account the relative frequencies of the words. Suppose,
for example, that our database consists of these book titles:
EXAMPLE II
(a) If A = [4
1
2
56
3) ,thenAT = (I
25.
4) B'l, Applied Linear Algebra

(b) IfB= [-;

I
;
2
~1
5 ,then BT =
3[_~ 4
2 3
B2, Elementary Linear Algebra
B3, Elementary Linear Algebra with Applications
B4, Linear Algebra and Its Applications
B5, Linear Algebra with Applications
1 2
 40 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 41
B6. Matrix Algebra with Applications
B7. Matrix TheOlY )'1 ==)'3 ==)'4 ==)'5 == 3, the titles of books BI, B3, B4, and 85 must contain all three
search words. If the search is set up to find titles matching all search words, then the
The collection of keywords is given by the fOllowing alphabetical list: search engine will report the titles of the first, third, fourth, and fifth books.

algebra, application, elementary, linear, matrix. theory Relative-Frequency Searches

For a simple matching search' 0' , Searches of noncommercial databases generally find all documents containing the key
. for the enr-: , we Just use s and I s, rather than relative frequee-
~;~e ~~~:;~tnes of the d~tabase matrix. Thus, the (i, j) entry of the matrix will be I search words and then order the documents based on the relative frequencies of the
that Our search "tn~ears 10 t ehtltle of the jth book and 0 if it doe not. We will assume keywords. In this case, the entries of the database matrix should represent the relative
for example in o:/nletls sOPjlstlchated enough to equate various forms of a word. So. frequencies of the keywords in the documents. For example, suppose that in the dic-
, IS f0 tit es t e words appli d dr' ed tionary of all key words of the database the 6th word is algebra and the 8th word is
as forms of the word li . T te an app tcanons are both count
defined by Table 4. app IcatlOn. he database matrix for Our list of books is the array applied, where all words are listed alphabetically. If, say, document 9 in the database
contains a total of 200 occurrences of keywords from the dictionary, and if the word al-
gebra occurred 10 times in the document and the word applied occurred 6 times, then
Table 4 the relative frequencies for these words would be and 2\?o 2~'
and the corresponding
Array Representation for Database of linear Algebra Books
entries in the database matrix would be
Key WOrds Books
Bl B2 B3 a69 == 0.05 and ag9 == 0.03
algebra
B4 B5 B6
I B7
I I
application I I J To search for these two words, we take our search vector x to be the vector whose
I 0 0
1 I entries X6 and Xg are both equal to I and whose remaining entries are all O. We then
elementary I I 0
0 I compute
I 0
linear 0 0 0
I I Y == ATx
I I
matrix I 0 0
0 0 The entry of y corresponding to document 9 is
theory 0 0
0 0 I J
0 0 0 0 a69 . 1 + aS9 . I 0.08
0 I )'9 == ==

If the words Weare searchin f

matrix and search vector ar g Or are applied, linear, and algebra then the database Note that 16 of the 200 words (8% of the words) in document 9 match the key search
!/I/U e respectIvely given by , words. If y, is the largest entry of y, this would indicate that the jth document in the
database is the one that contains the keywords with the greatest relative frequencies.
I I I I
I 1 0
1 0 I I I
0 I J 0
I I 0 Advanced Search Methods
A== I
0 0 0
I I I I 0 A search for the keywords linear and algebra could easily tum up hundreds of docu-
I 0 0 x==
0 0 0 0 I ments, some of which may not even be about linear algebra. If we were to increase the
0 I I
0 0 0 0 0 number of search words and require that all search words be matched, then we would
0 0 I
If we set Y = AT x, then 0 run the risk of excluding some crucial linear algebra documents. Rather than match
all words of the expanded search list. our database search should give priority to those
documents which match most of the keywords with high relative frequencies. To ac-
1]0100 complish this, we need to find the columns of the database matrix A that are "closest"
101100 I 3 to the search vector x. One way to measure how close two vectors are is to define the
1//100 I 2 angle between the vectors. We will do this in Section I of Chapter 5.
y'O I J 0 J 0 0 0 3 We will also revisit the information retrieval application after we have learned
1101001==3 about the singular value decomposition (Chapter 6, Section 5). This decomposition
1100100 3 can be used to find a simpler approximation to the database matrix, which will speed up
0000110 2 the searches dramatically. Often it has the added advantage of filtering out noise; that
The value of)'1 is the nu 0 is, using the approximate version of the database matrix may automatically have the
the value of)' . h mber of search w d k effect of eliminating documents that use keywords in unwanted contexts. For example,
2 IS t e nUmber of mat h . Or matches in the title of the first bOO.
c.m~ S.~ a dental student and a mathematics student could both use calculus as one of their
seCond book title, and so on. '"
 42 Chapter I Matrices and Systems of Equations
1.3 Matrix Arithmetic 43
search words. Since the list of mathematics search words does not contain any oller
dental terms, a mathematics search using an approximate database matrix is likely10 (c) [143][32]
o I 4 I I 8. If
eliminate all documents relating to dentistry. Similarly, the mathematics documeas o 0 2 4 5 = [i ~ 1 ' = [-~ ~], = [~ :]
142 6) [34
would be filtered out in the dental student's search. A B C

Web Searches and Page Ranking

(d)
t II 65] verify that
(e) [461][3
2 I I 4 ~] (a) (A + B) + C = A + (B + C)
Modem Web searches could easily involve billions of documents with hundreds of (b) (AB)C = A(BC)
thousands of keywords. Indeed, as of July 2008, there were more than I trillion Web
pages on the Internet, and it is not uncommon for search engines to acquire or update
as many as 10 million Web pages in a single day. Although the database matrix fm
(I)
[-n [3 2 4 5] (c) A(B+C)=AB+AC
(d) (A
9. Let
+ B)C = AC + BC

pages on the Internet is extremely large, searches can be simplified dramatically, since 3. For which of the pairs in Exercise 2 is it possible to
the matrices and search vectors are sparse; that is, most of the entries in any column
are O's.
multiply the second matrix times the first, and what
would the dimension of the product matrix be?
A = [: _; I· b = [~I, c= [ =;]
For Internet searches, the better search engines will do simple matching searche 4. Write each of the following systems of equations as (a) Write b as a linear combination of the column
a matrix equation. vectors 31 and 32.
to find all pages matching the keywords, but they will not order them on the basis oflle
(b) Use the result from part (a) to determine a so.
relative frequencies of the keywords. Because of the commercial nature of the InterneL (8) 3Xl + 2x, = I ~ (I lution of the linear system Ax = b. Does the
people who want to sell P~oducts may deliberately make repeated use of keywords ro 2x, - 3x, = 5
system have any other solutions? Explain.
ensure that their \\feb srte IS highly ranked in any relative-frequency search. In fact,II
(b) x,+x, =5 (c) Write c as a linear combination of the column
ISeasy to surreptitIously list a keyword hundreds of times. If the font color of the won!
matches the background color of the page, then the viewer will not be aware thai the
2Xl + X2 - X3 =6 vectors 3 I and 32.
word is listed repeatedly. 3Xl - 2x, + 2x] = 7 10. For each of the choices of A and b that follow. de-
termine whether the system Ax = b is consistent
For Web searches, a more sophisticated algorithm is necessary for ranking Ile (c) 2x, + x, + X3 =4 by examining how b relates to the column vectors
pages that contain all of the key search Words. In Chapter 6, we will study a special Xl - X2 + 2X3 =2 of A. Explain your answers in each case.
type of matnx model for assigning probabilities in certain random processes. Th~
tYt~hofmodel IS referred to as a Markov process or a Markov chain. In Section J 5. If
3x] - 2X2 - X3 =0
(a) A = [ _; _:], b= U]
[1 !l I
arankmgs
. apterof 6,Web
we pages.
will see how to use Markov chains to model Web surfing and obtain
(b) A = [~ ~] b = [;
A=
References
"IU I verify thai

1. Berry, M,ichaelw.,
and Murray Browne, Understanding Search Engines: Math,
(a) SA = 3A + 2A (b) 6A = 3(2A)
(c) A = [;
321
; :],
b=[_n
elna/lca Modeling d T; R . (c) (AT)T = A

-----------
1 J. Let A be a 5 x 3 matrix. If
an ext etnevat, SIAM, Philadelphia, 1999.
6. If
b=31 +32 = 32 +33

A = [i ~ ~) and B = [-~ ~ -~ I then what can you conclude about the number of

SECT/ON
I. If
'.3 EXERCISES
- verify
(a) A
that
+B
+ B)
=B +A
+ 3B
solutions of the linear system Ax = b? Explain.
12. Let A be a 3 x 4 matrix. If

b=31 +32+33+34

i Jl
(b) 3(A = 3A
2. For each of the pairs of matrices that folloW,de'
Iermine
. whether it is possible to mulnp . Iy rhe fi~1 (c) (A + B)T = AT + BT then what can you conclude about the number of
A = [-~ 7. If solutions ofthe linear system Ax = b? Explain.
matrix times the second. If it is possible, perftJl1l1
the mUltiplication. 13. Let Ax = b be a linear system whose augmented

n
compute
matrix (A [h) has reduced row echelon form

1-~ ~ ~) fr
(a) 2A A=[ ~~] and B=li:]
(b) A + B
-21
I
(c) 2A -3B -2 4
(e) A B
(d) (2A)T _ (3B)T
(a)
oI 2
0
0
I
3
2
I
4 5

(g) ATBT
(I) BA ~"1? 7,n verify that
(a) 3(AB) = (3A)B = A(3B)
o
o
0 000
0 0 000
0
(h) (BA)T
(b) (i =:1 (I 2 3]
(b) (AB)' = BTAT
(a) Find all solutions to the system.
 44 Chapter I Matrices and Systems of Equations

(b) If 1.4 Matrix Algebra 45

16. In Application 2. uppose that we are searchi"!
the database of seven linear algebra books for the
7. a(AB) = (aA)B = A(aB)
search words elementary, matrix, algebra. Fonn 8. (a + ,IJ)A = aA + ,lJA
a search Vector x, and then compute a vector y thai 9. a(A + B) = aA + aB
represents the results of the search. Explain thesil'
mficance of the entries of the vector y.
determine b.
17. Let A be a 2 x 2 matrix with a"
f 0 and I. We will prove two of the rules and leave the rest for the reader to verify.
&et A be an In x II matrix. Explain why the matrix ex ~ a21 / a". Show that A can be factored inter
multiplicatIOns ATA and AAT are possible. product of the form Proof of Assume that A = (aij) is an m x n matrix and B = (bij) and C = (cij) are both n x r
15. A matrix A is said to be skew symmetric if AT =
Rule 4 matrices. Let D = A (B + C) and E = A B + A C. It follows that
,- A" Show that if a matrix is skew symmetric then n
Its diagonal entries must all be O. '
What is the value of b?
dij = L aik(bkj + Ckj)
k=!

and

III Matrix Algebra eij = L

k=!
n

Qikbkj +L
n

k=!
aikCkj

But
The algebraic rules used for real nurn n n n
used. For example if a a db
• n
bers mayor may not work when matrices are
are reaJ numbers then L aik (bkj + Ckj) = L aikbkj + L aikCkj
k=! k=l k=1

+b =b +a and = ba so that dij = eij and hence A(B + C) = AB + AC.

a

For reaJ numbers, the operations of add' . . .

ab

. •
The first of these algebra I ttion and multiplication are both commutaUve. Proof of Let A be an m x n matrix, B an n x r matrix, and C an r x s matrix. Let D = A Band
ICru es works when I . A
an d B ; that is, we rep ace a and b by square mamces Rule 3 E = Be. We must show that DC =
AE. By the definition of matrix multiplication,

However A+B=B+A n r

f •
we hav
I
e a ready seen that matri
act deserves special emphasis.
....
IX multIplicatIon IS not commutative. This
dil = L aut» and ekj = L bklClj
IJI/II/III
i k=! 1=1

The (i, j) entry of DC is

Warning: In general A B -s: BA M .
, r- . atrix m It' I' . .
U ip rcanon IS not commutative.
In this sect' .
Ion we exam me who hI'
IC a gebralc rul k I
es war for matrices and which do no.
Algebraic Rules and the (i, j) entry of AE is

The following theorem provides some usefuJ .

Theorem 1.4.1 rules for dOIng matrix algebra:
~ac; of the following statements is vali .
' .and C for Which the indicat d d for any scalars a and ,IJ and for any matrlces
1 A +B e operatIOns are defined Since
. ~B+A .
2. (A + B) + C A = + (B + C
3. (AB)C ~ A(BC) )
4. A(8+C) = AB+Ae
S. (A + B)C ~ AC + Be it follows that
6. (a,IJ)A ~ a(,lJA)
= = AE = A(BC)
(AB)C DC
•
 46 Chapter I Matrices and Systems of Equations
1.4 Matrix Algebra 47
The algebraic rules given in Theorem 1.4.1 seem quite natural, since they are simi.
then
lar to the rules that we use with real numbers. However, there are important differenc~
between the rules formatrix algebra and the algebraic rules for real numbers. Someof

EXAMPLE I
these differences are Illustrated in Exercises I through 5 at the end of this section.
If
A2 _
-
[ 1
I :) r: :)=[~ ~)
3

A = [1 2 J, and B = [ 2 I J'
A = AAA = AA2 = [ :
:)[~ ~)=[::)
3 4 -3 2 and
and, in general,
verify that A(BC) = (AB)C and A(B + C) = AB + AC.
Solution 2,,-1
An = [ 2n-1
•
A (BC) = [I3 42 J [41 21J = [616 II5 J
(AB)C = [=46 lJ5 J [I2 0)1 = [616 115 J APPLICATION I A Simple Model for Marital Status Computations

Thus, In a certain town, 30 percent of the married women get divorced each year and 20
percent of the single women get married each year. There are 8000 married women and
2000 single women. Assuming that the total population of women remains constant,
A(BC) = [I~ Ii) == (AB)C
how many married women and how many single women will there be after 1 year?
After 2 years?

Solution
A(B + C) = [I
3 4
2) [ -13 3I) == [ I
5
7
15
J Form a matrix A as follows: The entries in the first row of A will be the percentages
of married and single women, respectively, who are married after I year. The entries

!Iillil' AB + AC = [-4-6 II5 J + [5II 42 J == [ 5I in the second row will be the percentages of women who are single after I year. Thus,

Therefore, A _ [0.70 0.20)

- 0.30 0.80

+ C) == + AC
Notation
A(B AB
• If we let x == [~~~~), the number of married and single women after I year can be
computed by multiplying A times x.
Since (AB)c == A(BC) .
s . ,~m~m~y .
. arne IStrue for a prOduct of four omit the parentheses and write A BC. The 0.70 0.20) [8000) [6000)
ISmultipli ed b . or more matrices I h . Ax == [ 0.30 0.80 2000 - 4000
T hus If. I Y Itself a number of ti '. . n t e case where an n x II matnX
k is .. Imes It IS conv . .
' a POsitiveinteger, then' ement to use exponential notation. After I year, there will be 6000 married women and 4000 single women. To find the
number of married and single women after 2 years, compute
AT
= ~AA ".A
k times 2
A x == A (Ax) ==
[0.70 0.20) [6000
0.30 0.80 4000
J ==
[5000)
5000
EXAMPLE 2 If
After 2 years, half of the women will be married and half will be single. In general, the
number of married and single women after n years can be determined by computing
Aflx. •
 48 Chapter I Matrices and Systems of Equations
1.4 Matrix Algebra 49
APPLICATION 2 Ecology: Demographics of the Loggerhead Sea Turtle
If we let e, denote the average number of eggs laid by a member of stage i (i = 2, 3, 4)
The management and preservation of many wildlife species depends on our ability 00 in I year and form the matrix
model population dynamics. A standard modeling technique i to divide the life cycle

~I
of a species into a number of stages. The models assume that the population sizes fm PI ei e3
each stage depend only on the female population and that the probability of survival P2 0
L= (3)
of an individual female from one year to the next depends only on the stage of Ibe [ ~ qi P3
life cycle and not on the actual age of an individual. For example, let us consider a o q3 P4
four-stage model for analyzing the population dynamics of the loggerhead sea rurtle
(see Figure 1.4.1). then L can be used to predict the turtle populations at each stage in future years. A
matrix of the form (3) is called a Leslie matrix, and the corresponding population
model is sometimes referred to as a Leslie population model. Using the figures from
Table I, the Leslie matrix for our model is

L _
-
0.67
[ 0
0

o
0
0.7394
0.0006
0
127
o
o
0.81
79

0.8~77
I
Suppose that the initial populations at each stage were 200,000, 300,000, 500,
and 1500, respectively. If we represent these initial populations by a vector xo, the

I I[ I
Figure 1.4.1. Loggerhead Sea Turtle populations at each stage after 1 year are determined with the matrix equation

0 0 127
At each stage, we estimate the probability of survival over a l-year period. We
also estimata the ability to reproduce in terms of the expected number of eggs laidin
0.67 0.7394 0 o
79 [200,000
300,000 _ 182,000
355,820
x, = Lxo = 0 0.0006 0 o 500 - 180
[
a g~ven year. The results are summarized in Table I. The approximate ages for eacb o 0 0.81 0.8077 1,500 1,617
stage are listed In parentheses next to the stage description.

Table I (The computations have been rounded to the nearest integer.) To determine the popu-
Four-Stage Model for Loggerhead Sea Turtle Demographics lation vector after 2 years, we multiply again by the matrix L:
!I!~II~I Sta~e
Number
Description
.
A
nnua
I
E~laid X2 = Lx, = L2xo
~m~~ SUrviVOrs
. h'
rp per year
I
Eggs, hatchlings ( < I) In general, the population after k years is determined by computing Xk = Lkxo. To
2 0.67 o see longer range trends, we compute x,o, X25, and Xso. The results are summarized in
Juveniles and subadults (1-21)
3
Novice breeders (22)
0.74 o Table 2. The model predicts that the total number of breeding-age turtles will decrease
4 M 0.81 127 by 80 percent over a 50-year period.
ature breeders (23-54)
0.81 79
Table 2 Loggerhead Sea Turtle Population Projections
If d, represents the duration of h . h . for Stage Initial 10 25
that stage, then it can be h .t e I t stage and s, is the annual survivorshif rate. 50
Number population years years years
year will be s own that the proportion remaining in stage i the folloWIng
I 200,000 114,264 74,039 35,966

Pi:::::
(I-,d'-')
I

J -s./ d· s,
(11

4
2
3
300,000
500
1,500
329,212
214
1,061
213,669
139
103,795
68
and the proportion of the po ulatio . ' . !be 687 334
followmg year will be p n that will survive and move into stage I +I
A seven-stage model describing the population dynamics is presented in refer-
ence [1] to follow. We will use the seven-stage model in the computer exercises at the
(21 end of this chapter. Reference [2] is the original paper by Leslie.
 50 Chapter I Matrices and Systems of Equations
1.4 Matrix Algebra 51
References
Matrix Inversion
1. Crouse, Deborah T., Larry B. Crowder, and Hal Caswell, "A Stage-Based Po~
ulation Model for Loggerhead Sea Turtles and Implication for Conservation: A real number a is said to have a multiplicative inverse if there exists a number b such
Ecology, 68(5), 1987. that ab = I. Any nonzero number a has a multiplicative inverse b = ~. We generalize
the concept of multiplicative inverses to matrices with the following definition:
2. Leslie, P. H., "On the Use of Matrices in Certain Population Mathematics
Biometrika, 33, 1945.
Definition An n x n matrix A is said to be nonsingular or invertible if there exists a matrix
B such that AB = B A = I. The matrix B is said to be a multiplicative inverse of
A.
The Identity Matrix
If Band C are both multiplicative inverses of A, then
Just as the number I acts as an identity for the multiplication of real numbers, theres
a special matnx I that acts as an identity for matrix multiplication; that is, B = BI = B(AC) = (BA)C = IC = C

Thus, a matrix can have at most one multiplicative inverse. We will refer to the multi-
IA=AI=A (41 plicative inverse of a nonsingular matrix A as simply the Inverse of A and denote it by
A-I.
forhaT n x n matrix A. It is easy to verify that, if we define I to be an n x 11 maW
Wit's on the mai di I d '
. n Iagona an 0 s elsewhere then I satisfies equation (4) foran) EXAMPLE 3 The matrices
n x n matnx A More fo II h ' .

Definition
. rrna y, we ave the follOWingdefinition:

The 11 x n identity matrix is the matrix I - (8.)

- I),
h
were
[ ; i) and [-~-u
10
are inverses of each other, since

8;j = ( I
if I = j
0 if I ¥c j
-'
[ ; i) [-~-:J
10
[~ n
As an example, let us verify equation (4) in the case n = 3. We have and
111m/I., I
[~
0
I
0 n(~ t1 (~t 14
6
I
-
4
6
[-~-:J [; 1) [~ n
10
•
I
and EXAMPLE 4 The 3 x 3 matrices

n
2

-n
0 -2

U (~ o
[32 46
o I I 001 = (34IJ
I
2 6
0 I
3
8
[~ I
0
and
[~ I
0
In general, if B is any m x n matrix . are inverses, since
and C ISany n X r matrix, then

BI = B

The column vectors of the n x . .

and IC =C

to define a Coordinatesystem in E n Identuy matrix I are the standard vectors US

eli
and
H
2
I
0 n [~ -n [~ n
-2

0
I
0
I
0

column vector of I is e. rath huchdean n-Space. The standard notation for theJ th

-n [~ n [~ ~)
be written er t an the uSual ij. Thus, the n x n identity matrixclJI -2 2 0
[~
j,

I
1- 0
I
0
= I
0 •
- (e l,e2,· .. ,en)
 52 Chapter I Matrices and Systems of Equations
1.4 Matrix Algebra 53

EXAMPLE 5 The matrix The first three rules are straightforward. We leave it to the reader to verify that they are

A-I -
IO
]0 0
valid. To prove the fourth rule, we need only show that the (I, j) entries of (AB)T and
T
B AT are equal. If A is an m x n matrix, then, for the multiplications to be possible, B
has no inverse. Indeed, if B is any 2 x 2 matrix, then must have n rows. The (I, j) entry of (AB)T is the (j, I) entry of AB. It is computed
by multiplying the jth row vector of A times the zth column vector of B:
BA = ~]
Thus, BA cannot equal I. (5)

Definition An n x n matrix is said to be singular if it does not have a multiplicative inverse.

The (I, j) entry of BT AT is computed by multiplying the Ith row of BT times the jth
column of A T Since the Ith row of BT is the transpose of the Ith column of B and the
Note jth column of AT is the transpose of the jth row of A, it follows that the (I, j) entry
of BT AT is given by
Only square matrices have multiplicative inverses. One should not use the termssin·
gular and nonsingular when referring to nonsquare matrices.
Often we will be working with products of nonsingular matrices. It turns out~
(6)
any product of nonsingular matrices is nonsingular. The following theorem character'
rzes how the inverse of the product of a pair of nonsingular matrices A and 8 is relalfJ
to the inverses of A and 8:
It follows from (5) and (6) that the (I, j) entries of (AB)T and BT AT are equal.
Theorem 1.4.2 The next example illustrates the idea behind the last proof.
If A and 8 are nonsingular n x n matrices, then A B is also nonsingular and
(A8)-1 =
8-1 A-I
EXAMPLE 6 Let
Proof
(8-
1
A-I)AB = B-I(A-I A)B = B-1 B = 1
, A=
I 2 I]
3 3 5 , B =
lO2 I
2]1
(AB)(B-IA-I) = AA- =
= A(B8-I)A-I 1
f5
I~'II
I
f 2 4 I 4 I
tiIIDII~1
It follows by induction th t if A . . s th~ Note that, on the one hand, the (3, 2) entry of A B is computed taking the scalar product
'III h . a , I I, ... , Ak are all nonsingular fl x n matncc , of the third row of A and the second column of B:
t e product A I A2 .•. Ak IS nonsingular and

(AIAz···Ak)-1 = Ak-I .. ·A-IA-I

2 I AB =
J
3
23 51] f 21 01 2]I =
flO34 236
In the next section we '111
· . .
h .
' . WI earn ow to determine whether a matrix has a .
multi· f2 4 1 5 4 I 15 8 l~]
p Iicanve mverse We WIll I I nsUl'
gular matrix.' a so earn a method for computing the inverse of a no When the product is transposed, the (3,2) entry of A B becomes the (2,3) entry
of (AB)T:
10 34
Algebraic Rules for Transposes (AB)T = 6 23 l~]
There are four basic algebraic rules involvl'ng tra
nsposes:
5 14 f
T
On the other hand, the (2,3) entry of B A T is computed taking the scalar product of
Algebraic Rules for Transposes the second row of B T and the third column of AT:
1. (ATl' = A
2. ("Al' = "AT BT AT = 0 I 21 45] f 2I 3
3 2]
4 =
f 106 34
23 15]8
3. (A + B)T = AT + BT f 2 I 1 1 5 1 5 14 9
4. (ABl' = BTAT In both cases, the arithmetic for computing the (3,2) entry is the same.
•
S4 Chapter I Matrices and Systems of Equations
1.4 Matrix Algebra 55
Symmetric Matrices and Networks
Note that the matrix A is symmetric. Indeed, any adjacency matrix must be symmetric,
Recall that a matrix A is symmetric if AT =
A. One type of application that leads to for if {Vi, Vj} is an edge of the graph, then aij = aji = 1 and aij = aji = 0 if there is
symmetnc matnces IS problems involving networks. These problems are often solv~ no edge joining Vi and Vj. In either case, aij = aji.
with the techmques of an area of mathematics called graph theory. We can think of a walk in a graph as a sequence of edges linking one vertex to
APPLICATION 3 Networks and Graphs another. For example, in Figure 1.4.2 the edges {V" V2J, {V2, Vs} represent a walk
from vertex V, to vertex Vs. The length of the wolk is said to be 2, since it consists
Graph th . . of two edges. A simple way to describe the walk is to indicate the movement between
,eory
Iems m IS an Important areas of applied mathematics. It is used to model preb
virtually all the applied sci G h . vertices by arrows. Thus, V, --+ V2 --+ Vs denotes a walk of length 2 from V, to
, " rences. rap theory IS particularly useful in appli.
canons Involvmg communication networks. Vs. Similarly, V, --+ Vs --+ V2 --+ V, represents a walk of length 3 from V, to
A graph is defined to be t f . V,. It is possible to traverse the same edges more than once in a walk. For example,
unordered . f verri a se 0 pomts called vertices, together with a set of
pairs 0 vertIces which D d Vs --+ V3 --+ V5 --+ V3 is a walk oflength 3 from V5 to V3. In general, by taking powers
metrical representar' f' h are re erre to as edges. Figure 1.4.2 gives a geo- of the adjacency matrix, we can determine the number of walks of any specified length
Ion 0 a grap . We can think of th . V V V V d~
as correspondin" to th d . '. e veruces I, 2, 3, "an l between two vertices.
eo e no es in a commUOlcatlon network.
V,
.....---- -..1v, Theorem 1.4.3 If A is an n X n adjacency matrix of a graph and ai~) represents the (i, j) entry of Ak,
then agJ
is equal to the number of walks of length k from Vi to Vj.

Proof The proof is by mathematical induction. In the case k = I, it follows from the defini-
tion of the adjacency matrix that aij represents the number of walks of length I from
Vi to Vj. Assume for some III that each entry of Am is equal to the number of walks
of length III between the corresponding vertices. Thus, a~") is the number of walks
of length III from Vi to V,. Now, on the one hand, if there is an edge (V" Vj l. then
v,
ai~"Ja'j = a'<;")is the number of walks of length III + 1 from Vi to Vj of the form
Figure 1.4.2.
Vi --+ ... --+ V, --+ Vj

The line segments joining the vert' On the other hand, if {V" Vj J is not an edge, then there are no walks of length III +1
Ices correspond to the edges: of this form from Vi to Vj and

(V" V2), {V2, Vs}, (V3, V,), (V3, V ), (V" V ) ail

(III)
a'i -
_ (m)
ail .
0- 0
-
s s
Each edge represents a direct communicar' .
An actual communication t IOn lin], between two nodes of the network,
It follows that the total number of walks of length III + I from Vi to Vj is given by
eds I deed ! ne Work could i I . d (m)
~'.
ges. need, If there are mill'
I~O~~.
f . nvo ve a large number of vertices an
Id ail ali + ai2(Ill) a2i + ... + Gill(m) ani
quue confusIng. An altemat" , a graphIcal picture of the network wou
the h' rve IS to use a . If But this is just the (i, j) entry of Am+I,
grap ContaInS a total of n' matrix representation for the network.
VertIces we ca d fi
' n e ne an n x n matrix A by
•
EXAMPLE 7 To determine the number of walks of length 3 between any two vertices of the graph
= {I if {Vi, Vj) is an edge of the graph

~]
aij in Figure 1.4.2, we need only compute
o If there is no d '"
e ge JOInIng Vi and V,
The matrix A is called the d'
h' , a ~acencYIll , ,
grap m FIgure 1.4.2 is given by
J

a fIX of the graph. The adjacency matrix for the A3 =

~ ~
I I 2
:: 3 4
I 1 3 2 4
o 4 4 4 2

Thus, the number of walks of length 3 from V3 to V5 is aj~) = 4. Note that the matrix

IIH ilJ
A=
A 3 is symmetric. This reflects the fact that there are the same number of walks of
length 3 from Vi to Vj as there are from Vj to Vi' •
 56 Chapter I Matrices and Systems of Equations
1.4 Matrix Algebra 57

SECTION 1.4 EXERCISES 18. Let A be a nonsingular n x 11 matrix. Use math- 28. Let A be an m x 11 matrix. Show that ATA and AAT
1. Explain why each of the following algebraic rule ematical induction to prove that Am is nonsingular are both symmetric.
9. Let and
'---- will not work in general when the real numbers a

l~~g fl
(Am)-I = (A -I)m 29. Let A and B be symmetric n x n matrices. Prove
and b are replaced by n x 11 matrices A and B. that A B = B A if and only if A B is also symmetric.
(a) (a + b)' = a' + 2ab + b' /form = 1,2,3, ....
(b) (a + b)(a - b) = a' _ b' h 19. Let A be an 11 x II matrix. Show that if A' = 0, 30. Let A be an n x Il matrix and let
then I - A is nonsingular and (I - ,1)-1 = 1+,1.
(rill the rules in Exercise I work if a is replaced by t- thol =0 (or " B=A+AT and C=A-AT
20 Let A be an II x II matrix. Show that if AHI = 0,
...J~n n .x n matrix A and b is replaced by the" x n 10. lei A and 8 be n "1IIIIIicn. then I - A is nonsingular and
identity matrix l? (a) Show that B is symmetric and C is skew sym .
each of tho: 101..... d<tcnnine ",iJ<fh<t
metric.
3. Find nonzero 2 x 2 matrices A and B such that matn\ mu t be )1tlIlICInC or (I - ,1)-1 = I + A + A' + ... + A'
ric: (b) Show that every n x It matrix can be repre-
JAB O. = 21. Given sented as a sum of a symmetric matrix and a
14. Find nonzero matrices A, B, and C such that
(a, C =A 8 bl 0 = R= [cose -Sine) skew-symmetric matrix.
J reI £ = A8 Idl f = ~8 4 smB cosO 31. In Application 1, how many married women and
AC = BC and A oF B (e) G ... AB 8), 'H... 8 8A show that R is nonsingular and R-1 = RT. how many single women will there be after 3 years?

5. The matrix II. Let C be a non )1IIJnCIn("" n ItIlIln For rJ 22. An n x n matrix A is said to be an involution if
Ihe foil"" Iftg. dctemune bctba the P''''' lIIOIll1 A' = I. Show that if G is any matrix of the form 32. Consider the matrix

A = I: =: I mu I be )rnmctnc
A ... C C'
or COUld be ""'" )1I1IlICI/1<
II) B _ C _ C' G= Icose sin e I 1 0 J I

,~ [1
(a) 0 I I 0
smB -cose
has the property that A' = O. Is it possible for a I 0 0
nonzero symmetnc 2 x 2 matrix to have this pro
(e) D = C' rdl f. = C' C - ce I
then G is an involution. I
erty? Prove your answer P- (e) F = (/ C (I C'
23. Let u be a unit vector in lR/l (i.e., u" u = I) and let 0
0
I
0
I
1
0
6. Prov~ the asso~iative law of multiplication for 2 x 2 !fI G = (/ CH/-C') H = / - 2uuT. Show that H is an involution. (a) Draw a graph that has A as its adjacency ma-
matnces; that IS, let 12. Let

A = Iall
b12
, ... [a,a, al
a
I 24. A matrix A is said to be idempotent if A2 = A.
Show that each of the following matrices are idem-
potent:
trix. Be sure to label the vertices of the graph.
(b) By inspecting the graph, determine the number
of walks of length 2 from V, to V, and from V,

!i tl
a21 )
bn ' Sho" thai" d ... alia - n ,au'" O.1ho:n to Vs.
(a) [: ~) (c) Compute the second row of A3, and use it to

and show that

A '= d
I [n -0
"'u
QII
I I I
(b)
determine the number of walks of length 3
from V2 to V3 and from V2 to Vs.
4 4
33. Consider the graph
13. Use the result from crcise 12 to find the un.... I I

= A (BC) (c) [: 4: 4
(AB)C of each of the follOWing m;lln"", v, v,
7. Let
(a) E ;I (b) [~n II [~ n (25./Let
1
, 2
I

A be an idempotent matrix.
Al

A=( __ :
,
-!} I
2
14. Let A and B be 11 )( 11 maine . Show thai ,I (a) Show thaO
(b) Show mat I
(I + ,1)-1 = / -
- A) is
+
!A
also idempotent.
A is nonsing<Jlar and
AB=A and Bid
Compute A' and ,13 . What will A' t
urn out to be?. 26. Let D be an n x n diagonal matrix whose diagonal
8. Let then A mUM be ,"gular. VI V4
entries are either 0 or I.
I I 15. Let A be a non"ngular matm Show mal A I 6 (a) Show that D is idempotent.
2 -2 -~ -! al,o nOn"ngular and (A ,) I = A
(b) Show that if X is a nonsingular matrix and
(a) Determine
graph.
the adjacency matrix A of lhe
I I
A = -2 2 -1 -i \J 16. Prove that If A " non Ingular. then A' IS _,iP- A = X DX-I, then A is idempotent. (b) Compute ,12 What do the entries in the first
J _1 I gular and
-2 2 2-t 27. Let A be an involution matrix, and let row of A' tell you about walks of length 2 that
-~ __ ,I _.!
,
I
2
(AT) I = (A 'I' I t
start from VI?
(Him: (AB)' : BTA'.) B=2:(/+A) and C = -(/- A) (e) Compute ,13 How many walks of length 3
Compute A' and A 3
What will A'" and A"'+I 2
1~. Let A be :tOl' are there from V2 to V4? How many walks of
out to be? tum . an n x n maln~ and let 1: and y be ,,('I; Show that Band C are both idempotent and BC = length less than or equal to 3 are there from V,
:::aiR:. Show that If Ax _ Ayand f y. the1l.tb' O. to V4?
trix A mu't be \Ingular.
58 Chapter I Matrices and Systems of Equations
1.5 Elementary Matrices 59

For each of the conditional statements that follow, an- matrices A and 8 must be equal. Type I An elementary matrix of type I is a matrix obtained by interchanging two rows
swer true if the statement is always true and answer
35. If A and B are singular n x n matrices. then A + B of!.
false otherwise. In the case of a true statement, explain
is also singular.
or prove your answer. In the case of Q false statement,
give all example to show that the statement is not always 36. If A and 8 are nonsingular matrices. then (AS)'" EXAMPLE I The matrix
true. nonsingular and

34. If Ax = Bx for some nonzero vector x, then the

III Elementary Matrices is an elementary matrix of type I since it was obtained

rows of J. If A is a 3 x 3 matrix, then
by interchanging the first two

In this section, we view the process of solving a linear system in terms of matrix
multi ph cations rather than row operations. Given a linear system Ax = b, we can
multiply both sides by a sequence of special matrices to obtain an equivalent syslem [
01 01 0 0r[a ll
aZI

in row echelon form, The special matrices we will use are called elementary matrices.
o 0 I a31
We WIll use them to see how to compute the inverse of a nonsingular matrix and also
to obtain an Important matnx factorization. We begin by considering
multlplymg both SIdes of a linear system by a nonsingular matrix.

Equivalent Systems
the effects of AE1 =
[
all
a21
a31
al2
a22
a32
al3]
a23
a33 [~~n V
Multiplying A on the left by EI interchanges the first and second rows of A. Right
multiplication of A by EI is equivalent to the elementary column operation of inter-
Given an m x n linear system A b '. lu'
I '. x = , we can obtain an equivalent system by IOU • changing the first and second columns. •
P ying both Sides of the equation by a nonsingular m x m matrix M:

Ax = b (I) Type II An elementary matrix of type II is a matrix obtained by multiplying a row of

MAx = Mb (21 I by a nonzero constant.
Clearly, any solution of (I) will also be a solution of (2) On the other hand, ifi isa
solution of (2), then . EXAMPLE 2
M-I(MAi) = M-I(Mb)
1 0I 0]0
and it follows that the two systems are e . I
Ax= b E2 =
[ 0030

l: quiva ent
a transform the system A - b . '. an
apply a sequence of n . x - t? a simple- form that is easier to solve, we.c is an elementary matrix of type II. If A is a 3 x 3 matrix, then
The new system will thonSlbnguflarmatnces E I, ... , Ek to both sides of the equation.
en e a the form

Ux =c E2A = rOO]
o 1 0 [all
a21
a,2
a22
al3]
a23
[ all
a21
al2
an
al3 ]
a23
where U = Ei . " EI A and c - E .. o 0 3 a31 a32 a33 3a31 3a32 3a33
alent to the original id - k'" E2EI b. The transfonmed system will be equlV
. , ProVI ed that M - E E" M IS al2
nOnsmgular, since it is a prod f :-- k . " I IS nonsmgular. However, [all al2 al3] [1 0 0] [ all 3al3 ]
. We will show next that anvor nonsmgular matrices. .
AE2 = a21 a22a23 010 = aZI a22 3a23
plished by mUltiplying A : the three elementary row operations can be aceoll!
a31 a32 a33 0 0 3 a31 a32 3a33
On t e left by a nonsingular matrix.
Elementary Matrices Multiplication on the left by Ez perfonms the elementary row operation of multiplying
the third row by 3, while multiplication on the right by E2 perfonms the elementary
If we stan with th id . column operation of multiplying the third column by 3. •
. e I entrty matrix I roW
operatIon, the resulting m t" and then perform exactly one elementary
There are three type: nfx]1Scalled an elementary matrix. f
I 0 e ementary m tri pes 0 Type III An elementary matrix of type III is a matrix obtained from J by adding a
e ementary row Operations. a ces corresponding to the three ty
multiple of one row to another row.
 60 Chapter I Matrices and Systems of Equations
1.5 Elementary Matrices 61
EXAMPLE 3
ith row to the jth row, that is,

IaaI a3} I
£3 =
[a a I 0
a I ith row
is an elementary matrix of type III. If A is a 3 x 3 matrix, then £=
a m I jth row
al2 + 3a32
a22
a32
a a a
3all + a13} then E can be transformed back into 1 either by subtracting m times the Ith row from
3a21 + a23 the jth row or by subtracting m times the jth column from the ith column. Thus,
3a31 + a33

Multiplication on the left b E add 3 '. .

tion on the right adds 3 ti Y h3 s lImes the third row to the first row. Multiplie;
unes t e first column to the third column. I

. In general, suppose that E is an '.

bemg obtained from 1 bye'th n x n elementary rnatnx. We can think of E ~
matrix,premultiplying A b I ;~ a ro: operatIOn Ora column operation. If A is ann XI
Y
A. If B is an m x n matri as t eeffect of performing that same row operation on
same column operation a xB'postmultlplymg B by E is equivalent to performing thai
n .
Definition A matrix B is row equivalent to a matrix A if there exists a finite sequence
Theorem 1.5.1
If E is an elementary matrix the £. . . E, £2, ... , £k of elementary matrices such that
of the same type. ' n ISnOnsmgular and E: I is an elementary matIU

Proof
If £ is the elementary matrix of t If . .
rows, then E can be transfo d~ ormed from 1 by Interchanging the ith andJib In other words, B is row equivalent to A if B can be obtained from A by a finite
Therefore, E E = 1 and h rme E ac~ Into 1 by interchanging these same rows agaiJI. number of row operations. In particular, if two augmented matrices (A I b) and (B I c)
type II formed by multipl ence IS Its own inverse. If E is the elementary matrix of are row equivalent, then Ax = band Bx = c are equivalent systems.
~ansformed into the identit~n~~~:t~ row of 1 by a nonzero scalar a, then e canbe The following properties of row equivalent matrices are easily established:
Y I/a. Thus, Y multIplYing eJlher its Ith row or its Ith coluDIII I. If A is row equivalent to B, then B is row equivalent to A.
II. If A is row equivalent to B, and B is row equivalent to C, then A is row
equivalent to C.
Property (I) can be proved using Theorem 1.5.1. The details of the proofs of (I) and
o (II) are left as an exercise for the reader.
£-1 _ I
l/a Theorem 1.5.2 Equivalent Co~ditions for Nonsingularity
Ith row
I Let A be an n x n matrix. The following are equivalent:
o
(a) A is nonsingular.
(b) Ax = 0 has only the trivial solution O.
Finally, if E is the elementary .
(c) A is row equivalent to I.
matrix of type III • . ....
IOrmed from 1 by adding m urnes'"
 62 Chapter I Matrices and Systems of Equations
1.5 Elementary Matrices 63
Proof We prove first that statement (a) implies statement (b). If A is nonsingular and xi"
EXAMPLE 4
solution of Ax = 0, then Compute A -I if
4
x = IX = (A-IA)x = A-I(Ax) = A-IO =0
Thus, Ax = 0 has only the trivial solution. Next, we show that statement (b) imple
A = {-l -2
2

statement (c). If we use elementary row operations, the system can be rransformr
Solution
into the form Ux = 0, where U is in row echelon form. If one of the diagonal e1~
ments of U were 0, the last row of U would consist entirely of O's. But then Ax =1
would be equivalent to a system with more unknowns than equations and, hence, ~
[-l -2
4
a
3 I
0
0
I
~J~ [~-}'-;
4 3 I
I
0
I
~J
-,I
Theorem 1.2.1, would have a nontrivial solution. Thus, U must be a strictly triangulr 2 3 0 0 -2 a
matrix with diagonal elements all equal to 1. It then follows that 1 is the reduced rO'l
I ;
echelon form of A and hence A is row equivalent to I. a 3

Ii
4 - \

n
4 3 I 0 2: -2: 2 -?
Finally, we will Show that statement (c) implies statement (a). If A is row equia
lent to I, there exist elementary matrices E I , £2, ... , £k such that
I 6 o 1.\ ~3
~
[~ 2
0
3
6
I
I
I
3
~ 2

0
0
6
I
2: -2:
3
I
-2:
I

I
A = Ek£k_1 '" Ell = Ek£k_1 ... £1
I I I
I 0 0 a I I

-. -!]
-2: -2: 2: I 0 -2: -2:
But since E, is invertible, i = I, ... , k, the product £k£k-I ... £1 is also invenible
Hence, A is nonsingular and ~ I I I I
0 2 0 ~ a
•
I
2: -2: -2: 0
I
A-I = (£kEk_1 '" £1)-1 = Ell £:;1 ... e;' I 0 0 6 3 0 0 I 6
1
2

Corollary 1.5.3 The system Ax = b of n linear equations in n unknowns has a unique solution if and Thus,
only if A is nonsingular. I

Proof If A is nonsingular and x is any solution of A x = b. then

Ax = b
A-I =
-.
n -1]
-2:
I

I
2:
•
g
MUI:!IYin bothsides of this equation by A -I, we see that x must be equal to A -lb. EXAMPLE 5 Solve the system
. onversely, If Ax = b has a unIque solution x then we claim that A cannot be
sl~~lar. Indeed, If A were singular, then the equati~n Ax = 0 would have a solution
XI + 4X2 + 3X3 = 12
z . But this would Imply that y =
x + z is a second solution of Ax b, = since -X, - 2X2 = -12
2xI + 2X2 + 3X3 = 8
Ay = A(x + z) = Ax + Az = b + 0 = b
Therefore, if Ax = b has au' . I
Solution
mque SOIUt1On,then A must be nonsingular.
If A is nonsingular then A . . tat)' The coefficient matrix of this system is the matrix A of the last example. The solution
matrices E I, ... , E'k SUch that ts row equIvalent to I and hence there exist elemen of the system is then 5x'3 '3 x\

Thus, the same series of element.

EkEk_I··· E,A

EkEk_1 '" Ell = A-I

=I
Multiplying both sides of this equation on the right by A -', we obtain

. rna'
.eA'be -: ~!-11 HI- [il •
lJix A into I will transfonn I int:r ~~woperatIons that transforms a nonsmgul.u:, If
we augment A by I and perf :. ThIS gIves us a method for computmg A . '10 Diagonal and Triangular Matrices
I on the augmented matrix ~nn ~ e ~Iementary row operations that transform A I~
An n x n matrix A is said to be upper triangular if au = 0 for i > j and lower
row echelon form of the au~ en Will ~ transformed into A -'. That is, the redu
tnangu
. ar I aij - 0 I'cor i < J' . Also , A is said to be triangular If 11IS either upper
I'f
mented matrix (A II) will be (IIA-I).
 64 Chapter I Matrices and Systems of Equations
1.5 Elementary Matrices 65
triangular or lower triangular. For example, the 3 x 3 matrices
then it is easily verified that

3o 22 II] OJ o
[ 005 and
(
I
600
0
[1 4
I 4 3 LU -- 2

2 -3
I 5
-I
~] = A •
are both triangular, The first is upper triangular and the second is lower triangular.
A tnangular matnx may have O's on the diagonal. However, for a linear systeJl
Ax = b to be In stnct tnangular form, the coefficient matrix A must be upper triangulM The matrix L in the previous example is lower triangular with I's on the diagonal.
with nonzero diagonal entries. We say that L is unit lower triangular. The factorization of the matrix A into a product
An n x n matrix A is diagonal if a;j = 0 whenever i i= j. The matrices of a unit lower triangular matrix L times a strictly upper triangular matrix U is often
referred to as an LU factorization.

(6~~j (~~~j
To see why the factorization in Example 6 works, let us view the reduction process

(6~J 001 000

in terms of elementary matrices. The three row operations that were applied to the
matrix A can be represented in terms of multiplications by elementary matrices

are all diagonal. A diagonal matrix is both upper triangular and lower triangular. (3)

Triangular Factorization where

If an n x n matrix A can b d d .
operation III th '. .e re uce to strict upper triangular form using only fO\\
factorizatio ' ~n!: IS possIble to .represent the reduction process in terms of a matriJ
n. e I ustrate how this IS done in the next example.
EXAMPLE 6 Let
correspond to the row operations in the reduction process. Since each of the elementary
matrices is nonsingular, we can multiply equation (3) by their inverses.

A=(i 4 -I~ ~J
9
A = EI'E:;'E:;'U
and let us use only row operation III
[We multiply in reverse order because (E3E2E,)-' = EI' E:;' E:;'.J However, when
, i' step, We subtract 1 tim th fi to carry out the reduction process. At the firSI
, 2
first row from the third.
es erst row from th d . etbe
e secon and then we subtract tWIC
the inverses are multiplied in this order, the multipliers I", 131,132 fill in below the
diagonal in the product:

i 4
(4 5
-I
To keep track of the multiples of t
2j
2
9
~ (~
0 -9
~ ij
5
,
E-1E-'E-'
1 23-
-
[~
_
I,

~~][~~~]r~
o I 2 0 I 0 -3
o
I

and. 1 ,
3 =
2. We complete the eli he first row that were subtracted, we set 12, '" l) In general, if an n x n matrix A can be reduced to strict upper triangular form using
POSitIOn: nunatlOn process by eliminating the -9 in the (3. only row operation III, then A has an LU factorization. The matrix L is unit lower
triangular, and if i > j, then IU is the multiple of the jth row subtracted from the ith

Let I _
[g j ij ~ (~ ~ ij
5 0 0 8
row during the reduction process.
The LU factorization is a very useful way of viewing the elimination process. We
will find it particularly useful in Chapter 7 when we study computer methods for solv-
32 - -3, the mUltiple of th all
the resulting matrix U d e second row Subtracted from the third row If we C ing linear systems. Many of the major topics in linear algebra can be viewed in terms
an set .
of matrix factorizations. We will study other interesting and important factorizations
in Chapters 5 tluough 7.

o
1
-3
66 Chapter I Matrices and Systems of Equations
I.5 Elementary Matrices 67

SECTION 1.5 EXERCISES

1. Which of the matrices that follow are elementary an el~menW) IIIalIl F we/llhal
(b) Find
(g) I-~ -3
6
8
-31
I
3
(b) I-I 1
-I -2
0
I
-l1
21. In general, matrix multiplication is not commuta-
tive (i.e., AB f. SA). However, in certain special
cases the commutative property does hold. Show
matrices? Classify each elementary matrix by type. FB =C. 11. Given that
(c) I C 1'01' equl\aIcm 10 ? plain,
(a) 1~ ~ I (b) 1~ ~] (a) if D1 and D2 are n x n diagonal matrices, then
6. Le1
A= 1~ ;] and B = 1~ ~] D,D, = D,D,.

(c) I 5~ 0~ ~1
I
2. Find the inverse of each matrix in Exercise J. For
(d) r ~0 0~ ~I 1
(a) Find elemen
= r~
mamc
i J
£,. £,. E, uchir
compute A -I and use it to
(a) find a 2 x 2 matrix X such that A X = B.
(b) if A is an n x n matrix and

(b) find a 2 x 2 matrix Y such that YA = B. where Qo, Gl, ... ,ak are scalars, then
each elementary matrix, verify that its inverse i an
E,E,£,A =U 12. Let AB = BA. (
elementary matrix of the same type.
3. For each of the following pairs of matrices, find an where U, an upper lruIngular rnatn.<. 22. Show that if A is a symmetric nonsingular matrix,
elementary matrix E such that EA = B: (b) Determme tile Ime or £1. E,. £J aod~ then A -I is also symmetric.
=
(a) A = r; -~], 1-1 ;] B =
7. Let
L E,' E, I E
Verify thaI A =L
I \\1w l)-pe of mamx"1
Solve each of the following matrix equations:
23. Prove that if A is row equivalent
row equivalent to A.
to B, then B is

(a) AX +B =C (b) XA +B =C 24. (a) Prove that iJ( A is row equivalent to Band B is
(b) A= 1-; ~ ~I'B= r ;
F-2 : ~51 A=(~ I] (c) AX +B = X (d) XA +C = X row equivalent to C, then A is row equivalent
to C.

I ~-~ ;
3 1 4 4 13. Is the transpose of an elementary matrix an elemen-
(8) Expre A • product r cl~menlo1l)matll'" (b) Prove that any two nonsingular n x n matrices
(b) Exp A I tary matrix of the same type? Is the product of two
prodLKl or elcmcnlal) ""'"
(c) A = 1, ces. elementary matrices an elementary matrix? are row equivalent.

I~-~ ;j 14. Let U and R be n x 25. Let A and B be m x n matrices. Prove that if B is
~ 8. Compute the LU f8C10llUUOOof each of !II< I> IT upper triangular matrices and
set T = UR. Show that T is also upper triangular row equivalent to A and U is any row echelon form
lowing matriee, A, then B is row equivalent to U.
B = and that '» = ujjrjj for j = 1, ... , n.
o 3 5 (8)
I~~] Ib) I-2 I2 4] 15. Let A be a 3 x 3 matrix and suppose that 26. Prove that B is row equivalent to A if and only
if there exists a nonsingular matrix M such that
4. For each of the following pairs of matrices, find an
elementary matrix E such that A E B: = r 3I 5I
[-2 I -1J 2a, + a, - 4a, =0 B=MA.
i ~1' I ~ i n
(c) Id) 4 I

(a) A = I~ B = 9. Let
iJ
-2 2 -6 -3 How many solutions will the system Ax = 0 have?
Explain. Is A nonsinguJar? Explain.
27. Is it possible for a singular matrix B to be row
equivalent to a nonsingular matrix A? Explain.
28. Given a vector x E lR'+I, the (II + I) X (11 + I)
0 16. Let A be a 3 x 3 matrix and suppose that matrix V defined by
(b) A = 1~ :], B = 1~ -;] A_
U 3
2 ~ )
(c) A= I-~ -~ ;1, (8) Verify that

2
-3)
Will the system Ax = 0 have a nontrivial solution?
ViJ={lj_1
Xi
if j = I
for j = 2, ... , 11 + 1

-1
6 1-2 Is A nonsingular? Explain your answers.

5. Let
B=I_! -~ jl A 1= [

(b) Use A-I 10 '<JIve Ax = b for tlH:

I
-2
-I
3
folJoOitC
It7\ Let A and B be n x n matrices and let C = A -
~ Show that if Ax" = Bx" and Xo
be singular.
l' 0, then C must
B.
is called the Vandermonde
(a) Show that if

and
Vc
matrix.

ee y

choices of b 18. Let A and B be n. x 11 matrices and let C = A B.

A = I~ 2I
o
41
3 ,
2 B = I~ 2
I 4)
3 ,
(I) b =
II. l.I)r
(lU) b"'I-2,1.0Ir
10. Find the inverse of each of the following maJ!i<tl'
(ill b .,dl.l.J Prove that if B is singular, then C must be singular.
[Him: Use Theorem 1.5.2.]
19. Let U be an n. x n upper triangular matrix with then
/
II -: ~)
2 6 nonzero diagonal entries.

C", 0 -I2 -34/ (8)

( (bl (~;] (a) Explain why U must be nonsingular.
P(Xi) = Yi, i=I,2, ... ,1l+1

Ii n
(b) Explain why V-I must be upper triangular.

(a) Find an elementary matrix E

EA = B.
h
2 2 6 (c)
Id) I~~] 20. Let A be a nonsingular n x n matrix and let B be an
n x r matrix. Show that the reduced row echelon
(b) Suppose that Xl, X2, •.. ,Xn+l are all distinct.
Show that if c is a solution to Vx = 0, then the
coefficients c., C2, ... , c; must all be zero and

i)
sue that
(e) [I o II (0 [2o 03 0 51 form of (A IB) is (lIC), where C = A -I B. hence V must be nonsingular,
o 0 I 0 3
68 Chapter I Matrices and Systems of Equations
1.6 Partitioned Matrices 69
For each of the following, answer true if the statement
30. If E and F are elementary matrices and G = EF. then
is always true and allswer false otherwise. In the case
then G is nonsinguiar.
of a true statement, explain or prove your answer. In the
case of a false statement, give an example to show that
the statement is not always true.
31. If A is a 4 x 4 matrix and a, + a, = a, + 2"" th<n

I n
A must be singular. and hence,
29. If A is row equivalent to I and AB = AC. then B
must equal C. 32. If A is row equivalent to both B and C, then A j,
row equivalent to B + C.
A(h b" b3) = [ _~ I ~i
In general, if A is an m x n matrix and B is an n x r matrix that has been partitioned
into columns (b, ... , b,.), then the block multiplication of A times B is given by
III Partitioned Matrices
AB = (Ab, Ab" ... , Ab,)
Often it is useful to think of a matrix as being composed of a number of submatrices.
A rnatnx C can be partitioned into smaller matrices by drawing horizontal lines be- In particular,
tween the rows and vertical lines between the columns. The smaller matrices are often
referred to as blocks. For example, let (a" ... , a,,) = A = Al = (Ae" ... , Ae,,)
Let A be an m x n matrix. If we partition A into rows, then
-2 4 I
I I 1

If lines are drawn betwee th·

3
622
2-1
1]
rth
I . n e second and third rows and between the third and fou
co umns, then C wIll be divided into four submatrices, C , C", C", and C,,: If B is an n x r matrix, the Ith row of the product AB is determined by multiplying the
II ith row of A times B. Thus the Ith row of AB is a(i, :)B. In general, the product AB

~-~ ~ : ~j
can be partitioned into rows as follows:

I
a,B ]
a,B
3 3 2 -] 2 AB= .
46224
One useful way of partition" . . . 31/:B
JOg a matnx is mto columns. For example, if
To illustrate this result, let us look at an example. If

B =
{-l 2 1J
2
1 4
3 1
I A = U ~J and B = [ -13 2
I
then we can partition B into three column sUbmatrices:
then
a,B = [ 1 9 -1 )
-I2 32 IJ J
B = (b, b" b3) =
{
1 4 I
a,B = [5 10 -5)
Suppose that we are given a rna . . AD a3B = [ -4 9 4)
can be viewed as a block multi li . ~nx A with three columns; then the product
result is a matrix with three b10~~~atlOn. Each block of B is multiplied by A, and the These are the row vectors of the product A B:
. Ab" Ab" and Ab3; that is,

For example, if
AB = A(b" b" b3) = (Ab, Ab" Ab
3
) AB = [;~~] = [---,;-~_1",,~_'-r-:~]':
a3B -4 9 4
Next, we consider how to compute the product A B in terms of more general par-
titions of A and B.
70 Chapter I Matrices and Systems of Equations

1.6 Partitioned Matrices 71

Block Multiplication
is an It! x (n - s) matrix, B, is an s x r matrix, and B2 is an (n - s) x r matrix. If
Let A be an m x n matrix and B an n X r matrix. It is often useful to partition C = AB, then
and B and express the product in terms of the submatrice of A and B. Considern II S n
following four cases:
cij = L ai/bl) = L ai/bl} + L ailblj
Case 1. If B = B2 J ' where B, is an n x 1 matrix and 82 is an n x (r-I
1=1 1=1 l=s+1

Thus, cij is the sum of the (i, j) entry of A, 8, and the (i, j) entry of A2B2. Therefore,
matrix, then

AB = A(b" , b., bt+" ... , b,)

and it follows that
= (Ab
"
, Abt, Abt +" , Ab.)
= (A(b" , bt), A(bt+" , b,))
== [AB, A82 J
Thus,
Case 4. Let A and 8 both be partitioned as follows:

A= B= _B...:,,-,
[ B2, B22
J
+-B",:'",,-2S
II - s

S II - S t r - t
Case 2. If A =
then
(1;), where A, is a k x n matrix and A
2
is an (m _ k) x n matril
Let

a, A,= A2 =
a,B
B, = [B" B2 =

[~~
I B=
ak+'
ak
B==
s,»
ak+,B
It follows from case 3 that

AB= [A, A21 (~;) =A,BJ+A2B2

am 3mB It follows from cases I and 2 that

[;~ 1 8
== f A2B
A,B J

Thus,
[at ]B
Therefore,

In general, if the blocks have the proper dimensions, the block multiplication can
be carried out in the same manner as ordinary matrix multiplication. That is, if

Case 3. Let A _ [ I A = r A" . .. A't 1

and B =
8"
.
- A, A2 and B == (B,)
B2 - where A,
.
IS an It! x S
matrix,Al As1 Ast r s.,
72 Chapter I Matrices and Systems of Equations
1.6 Partitioned Matrices 73
then
in which case

;I[!
I I
I I 6 4
2 I I
where
U
2 I
I I I f 18I~ 159 6
•
I
3 2
2 1 2
10 1
1~

c., = L AikBkj
k==! EXAMPLE 2 Let A be an n x n matrix of the form

The multiplication can be carried out in this manner only if the number of columasr
A;k equals the number of rows of Bkj for each k.
[ ~I
A~2]
EXAMPLE I Let
where A II is a k: x k matrix (k < n). Show that A is nonsingular if and only if A II and
A22 are nonsingular.

Solution
and If A II and A22 are nonsingular, then

Partition A into four blOCks

B = f :;; :;:]
d
=
I~ : j
2~1 1:1 2:1 and

o ]
An
[Aii0
[~ o ]_
[ho
In-k

0
In_k
-

]-1
-
I

an perform the block multiplication.

Solution so A is nonsingular and

Since each Bk· has two ro h f

two Possibilities: ws, t e A;k 's must have two columns. Thus, we have one0

Conversely, if A is nonsingular, then let B = A -I and partition B in the same manner

~l
I I as A. Since
(i) (All A12]

in Which case
A21 A22 =
U 2
3
I
2 it follows that
BA = I = AB

I I I

Or
ItttHl
2 2
3 3 2 2
I I I 2 I
3 I I
3 2 I jJ
6
= [ I~ 9
18 15
4
6
10 In Thus,

BIIAIi = h = AIIBII
(ii)
B22A22 = I"_k = AnB22
Hence, A II and A22 are both nonsingular with inverses B'I and B22, respectively .
•
 74 Chapter I Matrices and Systems of Equations
1.6 Partitioned Matrices 75
Outer Product Expansions
This representation is referred to as an outer product expansion. These types of ex-
Given two vectors x and y in IR", it is possible to perform a matrix multiplication of til: pansions play an important role in many applications. In Section 5 of Chapter 6, we
vectors if we transpose one of the vectors first. The matrix product xT y is the prodei will see how outer product expansions are used in digital imaging and in information
of a row vector (a I x n matrix) and a column vector (an n x I matrix). The resultwf retrieval applications.
be a I x I matrix, or simply a scalar:
EXAMPLE 3 Given

.', ~ «",>,. ,<.) If I~'>" + <'" + . +'.,. X =

compute the outer product expansion of X yT.

u nand y = u ~)
This type of product is referred to as a scalar product or an inner product. The scab
Solution
product IS one of the most commonly performed operations. For example, when we
multiply two matrices, each entry of the product is computed as a scalar product (a fOIl
vector times a column vector).

xy
It is also useful to mUltiply a column vector times a row vector. The matrix produd
T

IS the product of an n x I matrix times a I x n matrix. The result is a full n x n

XyT =
U ~)[; n 2
4

matnx.
[n [I
;~~~;~~:
I 31+[~)
2 4 I)
[2

[~ 46
2
6+
9)
3 4
[28 4 4l)
16
•
8 2
x"Y2 x"y"
The product xy" is ret d h .n,
. erre to as t e outer product of x and y. The outer product ma"~
has special structure In that each of its rows is a multiple of yT and each of its column SECTION 1.6 EXERCISES
vectors ISa mUltiple of x. For example, if
1. Let A be a nonsingular n x n matrix. Perform the 4. Let
Hi,

(rJ (~J
following multiplications:

(a) A-I [ A I 1 (b) [~I A-I 1= [ 6 n, E _

-
[0 6] , 0= [~
I ~]
then
x= and y =
(e) [A 11' [A I 1 C = [_: n, D -_[2 0
~I
xy T = [4JI [3 5 21 = (123 205 28J (d) [A I 1 [A 11' and

3
Note that each row is a multi I f
9 15 6 (e) [ A;' 1 [A I I
I
I
I
2
I
I
8 = [811 8l2]
We are now d p eo (3,5,2) and each column is a multiple ofx.
rea y to general" he i to ma' 2. Let 8 = ATA. Show that bij = aT aj_ 8'1 822 3 I I
trices Suppose th IZe t e Idea of an outer product from vectors 3. Let
. at we stan with . . y Wecan 3 2 2
then form a matrix pr d X T an m x n matrix X and a k x n matrix.. "5
If w e panltlOn
".
and perform the blo k o uctI": y X Into columns an d yT Into 10"of
Outer products of ve~to~~ lIphcatlon, we see that X yT can be represented as a sum A= U -: 1 and 8 = [i ~] Perform each of the following block multipIica-
tions:

(a)
(b)
Calculate Abl and Ab"
Calculate 31 Band a2B.
(a)
[~~ 1 [811
Bll
Bl2]
822

y; )
yi (e) Multiply AB, and verify that its c.olumn vee- (b) [g ~] [811
8"
81,]
8"
tors are the vectors in part (a) and Its row vee-

y~
tors are the vectors in part (b). (e) [g ~] [ B'I
B2l
812]
822
 76 Chapter I Matrices and Systems of Equations
Chapter One Exercises 77
(d) [E
o
0
E
JIB',B" B"
B 12 ]
(a) Show that D = (dlle" d"e, , d••e.).
15. Let 0 be the k x k matrix whose entries are all O. 19. Let A be an n x n matrix and x E JRIl .
5. Perform each of the following block multiplica- (b) Show that AD = (dllal. d"a" , d.. a.J. I be the k x k identity matrix, and B be a k x k
tions: (3) A scalar c can also be considered as a I x I
10. Lei U be an m x m matrix, Ier V be an n x n ma- matrix with the property that B2 = O. If
trix, and let matrix C = (c), and a vector b E JRIl can be
considered as an n x I matrix B. Although the
(a) [; matrix multiplication CB is not defined, show
that the matrix product BC is equal to cb, the
where Ej is an n x It diagonal matrix with diago- determine the block form of A-' + A' + A3 scalar multiplication of c times b.
nal entries ar, U2, ... , an and 0 is the (m -n) xn 16. Let A and B be It x n matrices and define 2n x 2n (b) Partition A into columns and x into rows and
zero matrix. matrices Sand M by
1
2
I-I] -[
(a) Show that if V = (V,. V,), where V, ha"
columns. then
(c)
perform the block multiplication of A times x.
Show that

, o3 i V'£ = V, '£,
Determine the block form of S-I and use it to com-
-, , o
5
20. ]f A isannxn matrix with the property that Ax = 0
4 3 pute the block form of the product S-' MS.
(b) Show that if A = V'£VT, then A can be e- for all x E JR.', show that A = O. [Hint: Let x = e
17. Let for j = I, ... ,n.]
J
o 0 pressed as an outer product expansion of the
form A=[AII A21
A12]
Al2 21. Let Band C be n x n matrices with the property
000
that Bx = Cx for all x E JR.". Show that B = C.
00100 where Al I is a k x k nonsingular matrix. Show that
I -I 22. Consider a system of the form
A can be factored into a product of the form
o I 000 2 -2 11. Let
(d) I 0 0 0 0
3 -3
A = [All AI,]
o 0 0 0 I 4 -4 A" o
o 0 0 I 0 5 -5 where all four blocks are n x n matrices. where where A is a nonsingular n x n matrix and a, b, and
6. Given c are vectors in JRIl .
(a) If A II and A" are nonsingular, show that A
B = AliA-III and C = A22 - A2IA-II1A12 (a) Multiply both sides of the system by
must also be nonsingular and that A-I muslbe
of the form (Note that this problem gives a block matrix version
Y=1124J
2 3 j of the factorization in Exercise 17 of Section 3.)

til, (a) Compute the outer product expansio-, of X yT ~"

[~ o
C
A-'
J 18. Let A, B, L, M, S, and T be n x n matrices with
A, B. and M nonsingular and L, S, and T singular.
(b) Compute the outer product expansion of Y XT '2 to obtain an equivalent triangular system.
How IS the Outer product expansion of YXT . Determine whether it is possible to find matrices X
lated to th re (b) Determine C. and Y such that (b) Sety = A-'a and z = A-'b. Show that if
e outer product expansion of XyT? T
7. Let . 12. Let A and B be n x n matrices and let M be a block fJ - c Y oF 0, then the solution of the system
o I o 0 0 0 M A can be determined by letting
matrix of the form o 0
l
I 0 0 0 A T
o 0 o I 0 0 T L
= [A L AIl
Is it Possible to perfo th bl
and

. .
AT
A [2 AI2 M=l~ ~J o
o
0
0 000
o 0 I 0
X
L
A
=
A
S
y 0 000 0 B T and then setting
of AAT and AT A? E rm . e ock mUltJphcations Use condition (b) of Theorem 1.5.2 to show thati!
l
. xpam.
either A or B is singular, then M must be singular. If so, show how; if not, explain why .
8. Let A beanm x n matrix X . X=Z-Xn+IY
13. Let
B an m x r matrix Sh • h an n x r matrix, and
. OWL at

AX= B A=[~ ~] . ), Chapter One Exercises

if and only if Where all four submatrices are k x k. Deternlloe
and A4
MATLAB EXERCISES
= bj• j ::;: I, ... , r
14. LId 'dblock
AXj et enote the n x n identity matrix. Fin a 2n The exercises that follow are to be solved computation- mathematical principles illustrated in the computations.
form for the inverse of each of the following 2n x ally with the software package MATLAB, which is de-
9. ~et A be an n x n matrix and let D be matrices: Save a record of your session in a file. After editing and
dlagonaJ matrix. an n X n scribed in the appendix of this book. The exercises printing out the file, you can fill in the answers to the
(a) I~ ~] (b) U ~J also Contain questions that are related to the underlying questions directly on the printout.