po
1.4 PROPERTIES OF MATRIX OPERATIONS
In this section we consider the algebraic properties of the matrix operations
just defined. Many of these properties are similar to familiar properties of
the real numbers. However, there will be striking differences between the
bers and the set of matrices in their algebraic behavior under
for example, under multiplication (as seen in Section 1.3).
Most of the properties will be stated as theorems, whose proofs will be left as
exercises.THEOREM 1
Proof
12 Seabee
Sec. 1.4 Properties of Matrix Operations 33,
(Properties of Matrix Addition) Let A, B, C, and D be m x n matrices.
(a) A+B=B+A.
(b) A+(B+C)=(A4 B)+C.
(c) There is a unique m x n matrix O such that
A+O=A aM
for any m ?
Wsc/,. Now
We leave the proofs of (a), (b),
prove only (c) here. Thus let A
The i, jth element of (AB)!
= FOW)(A) col, (By
= aybu + ayrby, +--+ ajpbps
wal
= at bh + al bh +. eal bl,
r
= bial +
2 tet bral, -
= POW BT) caly(Aly,
which is
We the sith element of BratSec. 1.4 Properties of Matrix Operations 39
Let
1 o 1
a-[} 1 3] and B=|2 2].
3-1
Then
r_fi2z
(AB) -("
and
ef Pape,
DEFINITION Ke matrix A = [a, is called symmetric if
EXAMPLE 12
AT =A.
‘That is, A is symmetric if itis a square matrix for which
=a, (Exercise 7.17).
If matrix A is symmetric, then the elements of A are symmetric with respect
to the main diagonal of A
3 1 0 0
5} and h=/0 1 0
6 oo 1
Section 2.1, Graph Theory, which can be covered at this time, uses mate-
rial from this section,
The matrices
1
A 2
3
are symmetric.
wanSec. 1
Properties of Matrix Operations 41
1, Verify Theorem 1.1 for
and
(QB Nerity (a) of Theorem 1.2 for
ef jhe 3]
and
Bag
P34
(Gvenity (a). (b), and (c) of Theorem 1.3 for r =
‘= ~2,and
aff Se [2 9
| 8, Verify (d) of Theorem 1.3 for r = ~3 and
=
Gyeity (b) and (d) of Theorem 1.4 for r =
} o-[E:
‘} o-[3 4]
possible, compute:
) (ABT. (b) BT AT () ABT.
AG, Determine a constant & stich that (kA)! (kA) =
(d) BT (ec) BTB.
9. If possible, compute:
(a) (GC 26)" B. (b) AN(D + F),
(c) BIC+A. (d) (2E)AT
(e) (BT + ANC.
10.10
~ A [3 3 and B ‘|
show that AB = 0.
1. It
aon telasal ae[2 al
e-[% 3):
show that AB = AC.
12. fA = ki a} stom that A? = 1s,
4=lt oo
42
13. Let A = [i H| Find
(a) A? +34.
(b) 249 +34? +4A + Shs
23.
(a) A= 24
(b) 34° = 24? +54 = 41
14. Let A= [: 4 Find
AK Determine a scalar r such that Ax = rx, where
vf dat)
[i
Is there more than one value of k that could be used?
17. Let
Zui
5 |
and a, = col,(A), j = 1,2, 3. Verify that
aysay ayo
yea) ayemy
ayy
ata, ofa; afwy
ala, aya
aya; ajay,1.5 SOLUTIONS OF LINEAR SYSTEMS OF EQUATIONS
In this section we shall systematize the familiar method of elimination of un-
knowns (discussed in Section 1.1) for the solution of linear systems and thus
obtain a useful method for solving such systems. This method starts with the
augmented matrix of the given linear system and obtains a matrix of a certain
form. This new matrix represents a linear system that has exactly the sameipter 1 Linear Equations and Matrices
DEFINITION
solutions as the given system. For example, if
1 0 0 2:
o 1 0-1
o 0 t 3
represents the augmented matrix of a linear system, then the solution is easily
found from the corresponding equations
at
The task of this section is to manipulate the augmented matrix representing a
given linear system into a form from which the solution can easily be found.
An m x n matrix A is said to be in reduced row echelon form if it satisfies
the following properties:
(a) All zero rows. if there are any, appear at the bottom of the matrix.
(b) The first nonzero entry from the left of a nonzero row is a 1. This entry is
called a leading one of its row.
(c) For each nonzero row, the leading one appears to the right and below any
leading one’s in preceding rows.
(d) If a column contains a leading one, then all other entries in that column
are zero.
A matrix in reduced row echelon form appears as a staircase (“echelon”)
patter of leading ones descending from the upper left comer of the matrix.
‘An m x n matrix satisfying properties (a), (b). and (c) is said to be in row
echelon form.
‘The following are matrices in reduced row echelon form since they satisfy
properties (a), (b), (c). and (4);
100% 10 0 0-2 4
oe o 1004 8
=\0 po ol. s=]o 0 0 1 7
Genk 0 0 0 0 0 Oo
09 0 0 0 0 Oo
and
12001
c=|00123
00000
The following matrices are not in reduced row echelon form. (Why not?)
1200 4 10 3
p=|0 0 0 o|. e=lo 2-2 $
oo 1-3 oo 1 2
10 3 4 1203 ¢
-|0 1-2 5 o 1-2 5
Felo 1 2 2] Glo o 4 2
9 0 0 06 0 0 0 0. sgn
DEFINITION
Sec. 1.5 Solutions of Linear Systems of Equations 45
The following are matrices in row echelon form:
1 5 0 2-2 4
0 103 4 8 ae:
n=|0 0 0 1 7-2), 1219 479 9
00000 0 Co Od
09000 0
and
0013 57 9
0 0 0 0 1-2 3
y=|o 0 0 001 2
0000001
000000 0
A useful property of matrices in reduced row echelon form (see Exer-
cise T9) is that if A is ann x n matrix in reduced row echelon form # In,
then A has a row consisting entirely of zeros.
We shall now turn to the discussion of how to transform a given matrix to
a matrix in reduced row echelon form.
An elementary row operation on an m x n matrix A = [ay] is any of the
following operations:
(a) Interchange rows r and s of A. That is, replace a,1, 4,2, .--5 dyn BY si.
4,25 +65 yy AN Gy 525 «25 ag DY Mp Dy =
(b) Multiply row r of A by c #0. Thatis, replace ari, d-2,
C025 -+ 5 Cra
(c) Add d times row r of Ato row s of A,r # s. That is, replace a,1, a,2,
25 Wan BY pt + dpi Gy + dra, 0-4 Aan + den.
irn BY Cap),
Observe that when a matrix is viewed as the augmented matrix of a linear
system, the elementary row operations are equivalent, respectively, to inter-
anging two equations, multiplying an equation by a nonzero constant, and
adding a multiple of one equation to another equation.
Let
0 0 1 2
A=|2 3 0 —
3.3 6 -!
Interchanging rows 1 and 3 of A, we obtain
oss
B=|2 3 0 -2].
00 STS
Multiplying the third row of A by }, we obtain
o Oo 1 2
c=|2 3 0 -2Thapter 1 Linear Equations and Matrices
DEFINITION
P EXAMPLE 4 |
THEOREM 1.5
Adding (~2) times row 2 of A to row 3 of A, we obtain
o o 1 2
D=| 2 3 0 -2
af S30 6 8.
Observe that in obtaining D from A, row 2 of A did not ¢ hange. .
An m xn matrix A is said to be row equivalent to an m xn matrix Bit B
can be obtained by applying a finite sequence of elementary row operations to
the matrix A
Let
12 4 3
A=|]2 1 3 2
1-2 2 3
If we add 2 times row 3 of A to its second row, we obtain
12 4 3
B=|4 -3. 7 8
1-202 3
so B is row equivalent to A.
Interchanging rows 2 and 3 of B, we obtain
12 4 3
c=]! -2 2 3
4-397 8
so C is row equivalent to B
Multiplying row 1 of C by 2, we obtain
4.8 6
D=|t -2 2 3
4-3 7 8
so Dis me equivalent to C. It then follows that D is row equivalent to A,
since we obtained D by a g cee: s
since we obtained D by applying three suecessive elementary row operations
Iti not difficult to show (Exercise T.2) that
1, every matrix is row equivalent to itself;
2. if A is row equivalent to B, then B is row equivalent to A; and
3. if A is row equivalent to B and B is row equivalent to C.
tt
equivalent to C, " mie pin Ait os
In view of 2, both statements, “A is row equivalent to B” and B is row equi¥-
alent to A.” can be replaced by “A and B are row equivalent.”
Every m x n matrix is row equivalent to a matrix in row echelon form. sSee 1.5 Solutions of Linear Systems of Equations 47
We shall illustrate the proof of the theorem by giving the steps that must
be carried out on a Specific matrix A to obtain a matrix in row echelon form
that is row equivalent to A. We use the following example to illustrate the
steps involved.
Let
0 2 —4 '
o Oo 3 4
Am lo 49 2 4
2 0 9 7.
The procedure for transforming a matrix to row echelon form follows.
Procedure Example
Step 1. Find the first (count-
23-407
ing from left to right) column co 2 4 4
in A not all of whose entries are 5
zero. This column is called the oe 8 ; |
pivotal column, -
tL Pivotal column of A
Step 2. Identify the first fo 23-40
(counting from top to bottom) 0023 4
nonzero entry in the pivotal col- oS ee
umn, This element is called the @ 2-5 2 4
pivot, which we circle in A 2 0-6 9 WJ
Pivot
Step 3. Interchange, if neces- M252 4
‘sry, the first row with the row 0023 4
where the pivot occurs so that fret es Sk 4
the pivot is now in the first row _-*
Call the new matrix Ay 2 0-6 9 7
‘The first and third rows of A were interchanged.
Step 4. Multiply the first row r
ror -go4
‘of A) by the reciprocal of the 0023 4
pivot. Thus the entry in the first 0 afate |
row and pivotal column (where a. oe
the pivot was located) is now a 2 0 -6 9 7]
1, Call the new matrix Ap. The first row of A, was multiplied by $
Step 5. Add multiples of the tor-boag
first row of A) to all other rows [i xe
to make all entries in the pivotal w=fo 5 5
column, except the entry where 203
the pivot was located, equal to Lo -2 -1
zero. Thus all entries in the piv-(—2) times the first row of
otal column and rows 2,3, ..., its fourth row.
mare zero, Call the new matrix
Ay11 Linear Equations and Matrices
Step 6. Identify B as the
(m —1) x0 submatrix of Ax 0b-
tained by deleting the first row
of As; do not erase the first row
of Ax. Repeat Steps | through
Son B.
Step 7. Identify C as the
(m2) xn submatrix of By ob-
tained by deleting the first row
of Bs: do not erase this first row
of Bs. Repeat Steps I through §
on C.
1 2
ro .
B=|0 a
Lo 3
Pivotal
column of B ——~
1 2
41
3.4
F a,
The first and second rows of B were inter-
changed
1 1-$ § 2
fo 1 3-2 3
p=]0 0 2 3 4
0-2-1 7 3
‘The first row of B, was multiplied by 4
14 4 2
o 1 3-2 4
oo 23 4
0 60 2 3 4
2 times the first row of By was added to its
third row.
3 2
o 1 3-2 &
°@3 4
o 23 1]
Pivotal | Pivot
column af C ——___|
1 wig Tr 3
9° ts} 2S
re z
9 0 2 4
:
3
No rows of C had to be interchanges. The fist
tow of C was multiplied by £Remark
Bese) age
ae Wueekd
Solution
Sec. 1.5. Solutions of Linear Systems of Equations 49
eV sy Ts
o 1 §$-2 4
3
q-[) o 1 3 2
0 o 0 0 0.
(2) times the first row of C> was added to its
second row,
Step 8. \demtify D as the Poe Pas
(m3) xn submatrix of C3 ob- : ;
tained by deleting the first row oO 1 3-2 3
of Cx: do not erase this first row O-O..-"t 9 ca
of Cy. We now try to repeat ph
Steps 1-5 on D. However, be- D=[0 0 0 0 9}
cause there is no pivotal row in
D, we are finished, The matrix, 11 -$ 12
denoted by H, consisting of D or feat
and the shaded rows above D is H= 2 a
in row echelon form. 0 1 32
0 0 00
When doing hand computations, it is sometimes possible to avoid fractions by
suitably modifying the steps in the procedure. A row echelon form of a matrix
is not unique. Note that if we perform the following operation on the matrix
H—add (—1) times row 2 of H to its first row—then we obtain the matrix
1 0-4 3
o 1 3-2 3
0 0 1 3 2
0 0 0 0 06
Observe that W is row equivalent to H and hence is row equivalent to A.
Moreover, W is also in row echelon form.
Every m xn matrix A = [a,j] is row equivalent to a unique matrix in reduced
row echelon form. .
We illustrate the proof of this theorem by also giving the steps that must be
carried out on a specific matrix A to obtain a matrix in reduced row echelon
form that is row equivalent to A. We omit the proof that the matrix thus
obtained is unique. The following example will be used to illustrate the steps
involved.
Find a matrix in reduced row echelon form that is row equivalent to the matrix
A of Example 5
We start with the matrix H obtained in Example 5 in reduced echelon form.
that is row equivalent to A. We add suitable multiples of each nonzero row
of H 10 zero out all entries above a leading 1. Thus, we start by adding (~$)50 Chapter |
Linear Equations and Matrices
es the third row of # to its second row:
” a ee
o 1 0-2 -§
Melo o a 32)
0 0 0 0 0
Next, we add § times the third row of Jy to its first row:
Remark
Solving Linear Systems
Asean
Proof
Proof
io 0 9 &
x o 1 0-2 -3
o o 1 3 2
0 0 0 0 0
Which is in reduced row echelon form and is row equivalent to A. .
‘The procedure given here for finding a matrix in reduced row echelon form
that is row equivalent to a given matrix is not the only one possible. As an
alternate procedure, we could first zero out the entries below a leading 1 and
then immediately zero out the entries above the leading 1. This procedure is
not as efficient as the one described above. In actual practice, we do not take
the time to identify the matrices Ay, A2,..., By. Bre... C1, Ca... and so
on. We merely start with the given matrix and transform it to reduced row
echelon form,
We now apply these results to the solution of linear systems,
Let Ax = b and Cx = d be two linear systems each of m equations inn
unknowns. If the augmented matrices [A b] and [c A] of these systems
are row equivatent, then both linear systems have exactly the same solutions
This follows from the definition of row equivalence and from the fact that the
three elementary row operations on the augmented matrix turn out to be the
three manipulations on a linear system, discussed in Section 1.1, yielding &
Innear system having the same solutions as the given system, We also note
that if one system has no solution, then the other system has no solution.
IFA and C are row equivalent m x n matrice
es, then the linear system Ax =
and Cx = Whave exactly the same sc Smear See
ations,
Exercise 1.3Sec. 1.5 Solutions of Linear Systems of Equations $1
The results established thus far provide us with two methods for solving
Hinear systems. ‘The key idea is to start with the linear system Ax = b, then
obtain a partitioned matrix [Cd] in either reduced row echelon form or
row echelon form that is row equivalent to the augmented matrix [A | b].
Now [C | d] represents the linear system C d, which is easier to solve
because of the simpler structure of [C | d]., and the set of all solutions to this
system gives precisely the set of all solutions to the given system Ax = b
‘The method where [C * d] is in reduced row echelon form is called Gauss’—
Jordan” reduction, the method where [C | d] is in row echelon form ts
called Gaussian elimination. Strictly speaking, the alternate Gauss~Jordan
reduction method described in the Remark above is not as efficient as that
neither Gauss—Jordan reduction
nor Gaussian elimination is used as much as the method involving the LU:
factorization of A that is presented in Section 1.7. However, Gauss—Jordan
reduction and Gaussian climination are fine for small problems, and we use
the former frequently in this book
GaussJordan reduction procedure for si
AX = bis.as follows
ing the linear system
Step 1. Form the augmented matrix [Ab]
Step 2. Transtorm [A | b] to reduced row echelon form [Cd] by using
elementary row operations.
Step 3. For each nonzero row of the matnix [C | d], solve the correspond-
ing equation for the unknown associated with the leading one in that row
‘The rows consisting entirely of zeros can be ignored, because the corre-
sponding equation will be satisfied for any values of the unknowns,
SCanl Friedrich Gauss (1777-1895) was born into # poor working-class family in Brunswick
Germany, and died in Gotugen, Germany, the most famous mathematician in the world, He
was a child prodigy with a genius that did not impress his father, who called him 4 “stargazer”™
However, hus teachers were impressed enough to arrange for the Duke of Brunswick to provide
«scholarship for Gauss af the lacal secondary school. Av a teenager there. he made original di:
Coveries in number thcory and began 1 ypeculate about non-Euclidean geometry. His scientific
publications inclade important contnbuvions in aumber theory, mathematical astronomy, mathe-
Inatical geography, statistics, differential geometry, and magnetism, His diaries and private notes
Contain many othe discoveries that he never published.
"An austere, conservative man who had few frends and whose private life was generally
tunbappy, be was very concemed that proper credit be given for scientiie discovenes. When he
‘relied on the results of others, be was careful to acknowledge them; and when others indepen-
dently discovered results 1 his private motes, he was quick to claim prionty.
1a his research he used a method of cakculation that later generations generalized to 10
reduction of matrices and named in his honor although the method was used in China almost
2000 years carer
*- Wilhelm Jordan (1842-1899) was born in southern Germany. He attended college in Stuttgart
‘and in 16% became full professor of geodesy at the technical college in Karlsruhe, Germany.
He participated in surveying several regions of Germany. Jordan was a prolife writer whose ma-
jor work, Handbuch der Vermessungskunde (Handbook of Geodesy), was translated into French.
Italian, and Russian He was considered a superb writer and an excellent teacher Unfortunately
the Gauss—Jordan reduction method has been widely attributed 4 Carmille Jordan (1838-1922),
44 well-known French mathematician, Moreover, it seems that the method was also discovered
tndependenily atthe same time by B 1 Clasen, a priest who lived in Luxembourg. This biograph.
ical sketch is based on an excellent article: S.C. Althoen and R. McLavghlin, “Gauss-Jordan
reduction: A brief history.” MAA Monshly. 94 (1987), 130-142.cer | Linear Equations and Matrices
‘The Gaussian elimination procedure for solving the linear system Ax=b
is as follows,
Step 1. Form the augmented matrix [A } b].
Step 2. Transform [A { b] to (some) row echelon form [C } 4] by using
elementary row operations.
Step 3. Solve the linear system corresponding to [C ; d] by back substi-
| tution (illustrated in Example 10). The rows consisting entirely of zeros
can be ignored, because the corresponding equation will be satisfied for
any values of the unknowns.
The following examples illustrate the Gauss-Jordan reduction procedure.
HEA Soe the linear system
x 2y +
de yt
Bro =
by Gauss—Jordan reduction.
Solution Step 1. The augmented matrix of this linear system is
Po a4
2-1 1
3.0 -1 3
) Step 2. We now trat ix is
Step 2. We nsform the matrix in Step 1 to reduced row echelon form
1 2 3:9
2-1 1:8
3 ot Fares
Pty
(2) times the first row was added to its
f
0
second row. (—3) time:
0 aed tts third ow a
V2 3: yan ry
Ion 2
oof ii? fs at row was multiplied by (—1).
12 3:9 btw ah u
oi] | Mins ct
oa a S the second row was added to its
bf iy] \wkas
112
\e
oo fie third row ‘as multiplied by (
f oO t:s TEMAS. oy w
11a (—2) times
a hist row DE SECON row was added to its
f oo: 3 -bin aa
1- ot (=1) times added
the
9 0 153 Second row and te ens wnLace
Solution
See. 1.5 Solutions of Linear Systems of Equations 53
‘Thus, the augmented matrix is row equivalent to the matrix
1 0 0; 2
OY Of-t Q)
O 0 1:3
in reduced row echelon form
‘Step 3. The linear system represented by (2) is
s
Solve the linear system
x+ y+2z7-Sw= 3
2x +Sy- 2-9w= -3 7
e+ y- z+3w=-ll 9
x—3y+22+7w= -5
by Gauss—Jordan reduction
Step 1. The augmented matrix of this linear system is,
1 ot 2-5:
2 5 -1 -9:
2 1 1 3
1-3 2 7
Step 2. The augmented matrix is row equivalent to the matrix (verify)
1 0 0 2
o 1 0-3
0 0 1 -2 @
o 0 0 0
which is in reduced row echelon form
Step 3. The linear system represented by (4) is
x +2w=-5
y -3w= 2
z-lw= 3.
The row in (4) consisting entirely of zeros has been ignored.
Solving each equation for the unknown that corresponds to the leading
entry in each row of (4), we obtain
-$ —2w
2+3w
3+2w
nek
wudpapter 1
Linear Equations and Matrices
Solution
=-5-2
243r
= 3+2 (5)
w= +
Because r can be assigned any real number in (5), the given linear system (3)
has infinitely many solutions. 2
Solve the linear system
my +2m 3x + ts =2
ny + ey tary — Bayt ts + 2 6)
xy +2m — Beat ty y=
3xy + Ory +45 — 9x4 + 4s + 346 = 9
by GaussJordan reduction,
Step 1. The augmented matrix of this linear system is
12 0-3 1 0:2
12 1-3 1 2:3
12 0-3 2 1:4
3.6 1-9 4°39
Step 2, The augmented matrix is row equivalent to the matrix (verify)
1 2 0-3 0 -1:0
0 0 1 0 0 2:1
0 0 0 oO 1 4)2 a
0.
09 9 0 0 0 0
Step 3. ‘The linear system represented by (7) is
x42 -3y — xe =0
% + 2X6
Ast X%
Solving each equation for the unknown that corresponds to the leading entry
in each row of (7), we obtain
X1 = 6 +344 — 20
xy = 1-26
ts = 2-16.
Letting x6 =r, x4 = s, and x) = f, a solution to the linear system (6) is
xp=rt+3s—2r
t
nade 8)
where r, s, and r are any real numbers, Thus (8) is the solution to the give?
linear system (6). Since r, s. andr can be assigned any real numbers, the “
linear system (6) has infinitely many solutions.Solution
1S
Solution
See. 1.5 Solutions of Linear Systems of Equations 55
‘The following example illustrates the Gaussian elimination procedure and
back substitution.
Solve the linear system given in Example 7 by Gaussian elimination.
Step 1. The augmented matrix of this linear system is,
162 3:9
2-1 158
300 -1:3
Step 2. A matrix in row echelon form that is row equivalent (o this augmented
matrix is (verify)
x+2y $3259
yt
Step 3. The process of back substitution starts with the equation
next substitute this value of z in the preceding equation y + z =
for y, obtaining y = 2—z = 2-3 = —I. Finally, we substitute the values
just obtained for y and z in the first equation x + 2y +32 = 9 and solve for
“x, obtaining x = 9 — 2y — 3c = 9 +2—9 = 2. Thus, the solution is x = 2,
y=—land: .
Solve the linear system
x+2yt3ct4w= 5
x+3yt5e+7w= Ul co)
x ~ 2-2w=-6
by Gauss-Jordan reduction.
Step 1. ‘The augmented matrix of this linear system is
102 3 4:5
13°59 7:0
10 -1 2-6
Step 2. The augmented matrix is row equivalent to the matrix (verify)
1 0-1 2:0
o 1 2 3:0). (10)
0 0 0 OF1
‘Step 3. The last equation of the linear system represented by (10) is
Ox + Oy + 02 + Ow
which has no solution for any x, y, z, and w. Consequently. the given linear
system (9) has no solution. .See, 1,5. Solutions of Linear Systems of Equations 57
is row equivalent to (verify)
1 0 0:0
o 1 0:0
0 0 1:0
which is in reduced row echelon form. Hence the solution to (13) is
y=z=0,
which means that the given homogeneous system (13) has only the trivial
solution. .
Consider the homogeneous system
4)
‘The augmented matrix of this system,
11 ob 10)
1 0 0 1: 0],
2 1 010
is row equivalent to (verify)
10 0 1:0
0 1 0-10
9 0 1 1 0
which is in reduced row echelon form. Hence the solution to (14) is
where r is any real number. For example, if we let r = 2, then
x w=2
js « nontrivial solution (o this homogeneous system. That is.
roid 7 0
0 0 1 |= 0
1 2 1 off 3} Lo
(Verify by computing the matrix product on the left side.) .
Example 13 shows that a homogeneous system may have a nontrivial so-
lution. The following theorem tells of one case when this occurs.
A-homogeneous system of m equations in n unknowns always has a nontrivial
‘olution if m