Its

I)eterminants and Matrix Sol

Inversion
Th,
cal
exl
sys
Coyley's matrices (1857) have fourished
and today constirute a very imponant
and useful instrument in mathematics.
Eves (19'79)
-Ho*-ard
for
sp,(
the
pr(

Ex

4
This chapter continues the sody of linear algebra. The fust topic discussed is
determinanrs. Though some economists have claimed that determinants are almost
useless. we shall see that they do in fact play an imponant role in several areas of
mathematics that are of interest to economists- Nc
After introducing determinants, we consider the fundamentaliy important con- gn

cept of the inverse of a square matrix and its main properties. Cramer's rule for lA
the solution of a system of n linear equations and n unknowns is discussed nexr lin
Although it is not efficient for solving systems of equations with more than three
unknowns, Cramer's rule is often used in theoretical studies. AI

13.1 Determinants of Order 2

Consider the system of iinear equadons: TT

ml
41111 *aPb:b1 d1
[13:
a21x1 * a22x2: b1

438
 Sec. 13.1 / Deteminants ol Order 2 439
Its associated coefficient marix is

A: ( alt dt?
ll3.2l
A?t 421 )
Solving the equation sysrem [l3.ll in the usual way (see Section A-9) yields

b1a21- b2ap b2a11

- b1a21
4r --. xa
- [13.3]
atta?Z-Alan Attd2Z-a\An
The two fractions have a cornmon denominator. This number. a11an * a21ap- is
called the determinant of the marix A. Note that if this numbr is zero. then the
expressions for.r1 and 12 become meaningless-indeed, in this case, the equation
system [13.1] has eithcr no solutions or else infinirely many solurions.
I,irE The determinanr of A is denoted by either de(A) or lAl. Thus,
,iirr
GT, att an
i9Bl tAt : 4Zl Att = AlA!. - A2.tAn ll3.4l

for any 2 x 2 matix A. Such a determinant is said to have order 2. For the
special case of order 2 determinants, the rule for calculatin_e lAl is: (a) multipiy
the diagonal elements, @) muidply the off-diagonal elements. and (c) subtract the
product of the off-diagonal elements from the product of the diagoaal elements.

Example 13.1

Ea! fi I atz
rl :4-2-3-l = 5, lb, : A1 bt
frn J lb, Qtt
brazz.-bzan, : bzatt-brazt
A2 b2
rrd
Nota.' Geomerically, each of the rwo equations represents [13.1] represents the
graph of a straight line. If lAl + 0, rhe two lines intersect at a unique point. If
lAl :
0, either the two lines are parallel and there are no solutions, or the rwo
trd lines coincide and there is an infinite number of solutions.
From Exampie 13.1, we see that the numerarors of the expressions for -r1
and;r2 in [3.3] can also be wriuen as.determinants. If lAl 10, then

b1 An att br
b2 A1) a2t b2
fl :
[13.s]
tAt lAt

This is a special case of a result referred to as Cramer's rule (named after the Swiss
mathematician G. cramer, 1704-1752). It turns out rhar a similar rule applies to
tuee equations wirh three unknowns (see the next secrion). Later in Section 13.g,
'* e shall generalize the rule to z equarions in n
unknowns.
440 Chapter 13 / Determinants and Matrix lnversion

Example 13.2
(azt
Use [3.5] to find the solutions of

2x1 + 1x.: 7

2x1 - 2-4: -2
Solution

11 21 FIGU
al
-6 I 2-2 -18 3 a
a
Xl: 24 _11 1
24 _1) )
aa
deter

Check by substituuon that xl : l12' x2:312 teally is a solurion' rhe s

Example 133 to fir

ilse tl3.5l to find Q? *d Of in terms of the parameters when Tt=
elem
2(b+ilQ?+bQ?:a-dl
Prob
bQ?+2@+DQ?=a-d?
1.
Solution The determinant of the coefficient matrix is

A_ 2(b + Fr) b
-4(b+ErXb+ F)-b2
b 2(b + P2)
,,

Provided A t 0, by [13-5] the solutions are 3.

a-dt b

a?
a-a? 2(b+fr2) : 2(b+F)@-ar)-b(a-uz)
.A
A
4

2(b + p:) a-at

b a'- d2 2(b+fl)(a-oz)-b(a-qr)
Q?: A A
pa-
This example illustrates the convenience of cramer's rule when several 5
rameters are involved- 6

A Geometric lnterPretation
in Fig' I 3 ' I '
Determinants of order 2 have anice geometric interpretation. as shown
in Fig. 13.1. then the determinant is equal to tbe
If the two vectors are situated as
shaded area of the parallelogram. If we interchange the two row vectors in the
 Sec. 13. I/ Determinants of Order 2 441
a2t
a0+a1i t-,
T1
(at
422

T2

T7

lr. at:)
T1

a| I att
RGURE 13.1 Area = FIGURE 13.2
-lar
l4t
arz
?dl
I

determinant, then it becomes a negative numhr whose absolute value is equar to

the shaded area.
Figure l3-2 inustrates why the resurt craimed
Fig. I3.l is rue. we want
T. Note ttrat 2Tr * 2Tz * ZT3 1I : _rn
to find area (a11 * a2)(a12* a,). wherc
T1 : apat1, T.
= [a21a,,, ard T3 : lirrorr. Then I :;'rr;; _ ozrrn, by
elemenary algebra.

Problems

1. Compute the following determinants:

a.
30 b. la al a+b a-b 3', 2t
26 la bi c.
a-b a*b d.
3r-r Z,-l
2. Illustrate the geometric interpretation in Fig. l3.l for the determinant in
koblem 1(a).
3' use cramer's rure [13.5] to solve rhe fonowing sysrems of equations for .r
and y. Test the answers by substirution.
u 3x1- 12:8 b. r+3y:l c- ax_br.:l
xr-Zxz-5 3x-2y=l! bx*ay:2
4. l*t
att ap
A_ ( a2t and B- ( bl bn
Aaa
) bzt bzz )
Show thar lABl = lAl . lBl.
5. Find two 2 x 2 matrices A and B such that iA+Bl# lAl+lBi.
6. Use Cramer's ruie to find y and C when
y:C*/s*G6 C:a*by
:s- -i "

a.[ r: =c where y is national producr and C is private

consumpdon. The symbols 1s
5IIU (private investment), Ge (public consumption
and invesuneng, o,-ano a au
442 chapter 13 / Determinants and Matrix tnversion

represenr constants. with b < l. (Actually. this is a typical case in which

one should no, use Cramer's ruie. because Y and C can be found much more
simply. How?)
7. Consider the following linked macroeconomic model of two nations, i = 1.2,
that trade with each other:

)'r : Cr * Xt - Mr.
* Ar Cr: ctYt, M1= mtYr
Yz = Ct -l- A: -F X:. - Mt. Ct: czYz. M:.: mzYz.

Here, for i : 1.2. I, is income. C; is consumption, A; is (exogenous)

autonomous expenditure, X, denotes exports. and M; denotes imports of
counr.v i.
a- Interpret the two equations Xr : Mz and X2: ltll1.
b. Given the equations in part (a). calculate the corresponding equilibrium
values of )21 and Y2 as functions of the exogenous variables.
c- How does an increase in .A1 affect 12? Interpret your answer.

8. Let the matrix A(r) be defrned for every / by A(/)

t2 2r-1
: ( 2t2
Evaluate the determinant lA(r) I and find the values of r for which lA(r) I
)
= 0.
9. Let a(r) and b(t) be twice differenriable functions. Prove rhat

d I aUt btt) | _ 1 a(rr birr

- la"ttl b"Q)
l

a, lo'r,t b'u) | I

Verify this rule for the determinant lA(r)l in Problem 8.

13.2 Determinants of Order 3

Consider t}le system of three linear equations in three unknowns:

atrxt*apx2*a1ix3:b1
ayxl * arx2 * a*g - b2 tl3.6l
a31x1ia3?x2ia33x3=[3

Here the coefficient matrix A is 3 x 3. By applying the method of eliminatioo

along with some rather heavy algebraic computation, evenrually this system can
be solved for x1, x2. and .r3. The resulting expression for x1 is

:
b1a17a31 - b1a7a32 - b2apa3i * b2apa32* hanaZ: - b3a22a3
fl
at.crai. - a)tQ2:\a3Z* apa2;A31 - a2A2ta3j j a6A21A32 - a1a2?a)l
 Sec. 13.2 / Determinants of orcler 3 443

n ; hich We shall not triple the demands on the reader's patience and eyesight by grv-
cii more ing the corresponding expressions for x2 and x3. However, we do claim that
these expressions share the same denominator as that given for x1. This com-
mon denomioator is called the determinant of A, denoted by det(A) or lAl. It is
defined as

Iarr an att I

lAl =Ia, a)a a, I

J
"r, ";, ,;; I
tr3.7)
ar a2a13,
I - at ta23a32 * apaVa3l
rgenous) { - anq2taJ3 * aga21a37 - aga7ra3l
lrons of
Expansion by Cofactors
Consider the sum of the six terms in [3.7]. It looks quite Eessy, but the method
urlibrium
of expansion by cofactors makes it easy to write down all the terms. First, note
that each of the ttree elements a1, otz, and a13 in the first row of A appears in
exactly two tenns of u3.71. In fact, lAl can be wrinen as

lAl : arr (azz.aiz - aztazz) - a12(a21ay - aztay) * as(a21a32 - anaz)

\li) :0.
which is &e same as

422 a.^
* I 1 a., a", I
-' " I o.,
I -'
a". I

lAl = all 432 anl1 -a't '-la;t aj3ll-i-a,r '-lay -

a3zl
I
ll3.8l

In this way, the computation of a determinant of order 3 can be reduced to cal-

culating three deterrrinants of order 2. Note that all is multiplied by the second-
order determinant obained by deleting thefrsr row and thefrsr column of lAl.
Likewise, a12, with a minus sign anac-hed to it, is muitiplied by the determinant
obtaiued by deleting the frsr row and tlre second column of lAl- Finally, a13
is multiplied by the determinant obtained by deleting the frsr row and rhe third
column of lAl.
Example 13.4
Use [3.8] to compute
02
tAr :l-l13 l0
I5 23
:-:Tufietrlfi
Solution
:l:i:L- t4

IAI :3 Ir
l,
ol ^ I
-1 ol-r.i -l 1l
"l-u'l I
5 il'-
-tt I 5 ?l
-3 3-0+2-(-2-5):-5
444 Chapler 13 / Determinants and Matrix lnversion
Exa:

ExamPle 135
Use [13'8] to Prove that
laa?
lAl = tb6z
\cc2
: (b-a)(c-a)k-b)

Solution

:1 \u b'\ - \r ,:1*,, ll ilcl

1ll
lc 1-, ll .. I' ll
.,
: bc? _ bzc _ acz + abz + o2, - azb

that these six t"'""t :3 be wrinen

as
You are not expected to see at once
(b - a)(c - a)\c-;'-il;"t' vou should expand (b - a)t(c - a){c - b)J

and verifY the equaliry in that

waY'

determinant can be expanded in

terms of the elemens of any row or
A
of this in Section l3'5'
column- We shall give a full discussion
expression for xf in the beginning
A carefuLtuOy oi the numerator r'n &e as a determinant' The same
of this secdon reveals t'L ii c* aso
be wrinen Ar
for x2 and xr. ln fact" if lAl # 0' then
is true of the correspoJirg formulas Lik
one has that

4tt br dre all an bt

bt a0 aB I

An bz
b7 o.'= azt
azt
,;l
|
bz azz a?3
bt a3t a\z bt (azt
bt 432 433
x?: x3
rt: lal lAl
la t13.91

of t13'61 (see Section 13'8)'

This is Crarner's rule for the solution

in-th^e of 11' 12' and -r; of [13'9]

Note: lnthe determinants appearing l3merators
;;*; how the right-hand column in [13'6]'
F}G
/br \
(ai
\4,/
S;
when solving
wben solving for x1, to rhe second column
shifts from the first column makes it very eas] Ha
wien solviog for x3' This
for xz, and then to tUe tfrirO column coD

to remember Cramer's rule'

 Sec. 13.2 / Determinants ol Order 3 445
Example 13.6
Solve the following system of equations by using [13.9]:

2xt * 2xz X3 -J
4xt + 2x3 8
6xz 3xt 2

Solution In this case, the determinant lAl in [3.9] is

t- 2-r :
t')

IAI :14 0z -24

i0 6-3
The numerators in [13.9] are

3€ ,*T1EeD 1 -32 -l : 2 -3 -1 2? -3
- ; llc -
r 80 2 -12, 482 -
t, 40 8
1a

-r2 6 -3 0 -12 -3 06 -12

Hence, [3.9] yields the solution x1 : (12)/(-24) - I/2, xz: l2le24)
an\, ro\i :,f : -l/2, ilrd x3 = e1D/ezq: 3. Inserting this into &e original system
of equations confrms that this is a correct answer.
he begmru-ng
I The sa:ce
\ = 0. th--a A Geometric lnterpretation
Like determinants of order 2, those of order 3 also have a geometric interpretation
that is shown and explained in Fig- l3-3.
A.
-- A^

(ay, atz,
.{
r1i Q'
a3)
- t. azzi,

)
r-. -r I-iiCl (a11, a12, ar1)
r,\. _a Lr-,/

4l 4z ara the Yolurne of the

FIGURE 13.3 * An ?A il6 'box" spanned by
agr 42 4"3 the three vectons

,: len solvug Sarrus'Rule

:S Ii \ er\ 33-ir Here is an altcrnadve way of evaluating determinans of order 3 that rnany pcople 6nd
coovenienl W-rite dowu the determinant twice, except tbar the secoud time the last column
446 Chapter 13 / Determinan$ and Matnx lnversion

of the second determinant should be omitted:

at) ot:. 9n| -ot' :rri -

u3.101
"rr)orrlor:]ur-12
a 1/ a 32n a33]ai r
|

3
++-+
--'rz.

First.multiplyalongthethreelinesfaliingtotheright,givingalltheseproductsaplussign:

afi a??a31 + a\2a?3a31 * atlazt a3? [*]

Thenmultiplyalongthethreelinesrisiogtotheright'givingalltheseproducsaminussign:

- a3lazlal? [**]
-a3t anaB - a3za2iarl

Thesumofthetermsin[*]and[**lisexactlyequaltoiA|.(Itisimponanttonotethatthis
rule.knownasSarnrs,rule,doesnotgeneralizetodeterminantsoforderhigherthan3.)

Problems
13.3 l

1. Compute the following

1-1 0 -1 0r abc a0b
a. 132 b. J c. Od e d 0 e 0
100 2 ?l 00 f c0d
2. Let
1 -l 0\ /r ?3
{: ( J and B: I 2 34
2 i) \o 1 -1 )
lAl, lBl, lAl' lBl, and lABl'
Compute AB,
3.UseCramer,sru]etosolvethefollowingSys[emsofequations.Checkyo,u
answers.
a. xt-x2*x3: ? b'Xt- Xz -0
x1lx7-x3= 9 'r*3xz*2:4=0
x1 *2x2* xr = 0
-xt - x7- x3: -6
c. x*3Y-Zz=l
3x-2Y*52:14
2x'5Y-F3z:l
4. Show that
I +a 1 1

1 1+D I :abc*ab+ac+bc
1 1 1*c
 Sec. 13.3 / Determinants of Oder n 447

5. Consider the simple macro model described by the three equarions

Y:C*Ao. C:a*b(Y-T). T=d+ty

1_: _:
where r is income. c is consumption, r is tax revenue, 46 is the constant
(exogenous) autonomous expenditure, and a. b, d, and, r are all positive
-i,::'-
parameters- Find the equilibrium values of the endogenous variables y, c,
and ?" by
(a) successive elimination or substitution
O) writing the equations in matrix form and applying Cramer,s rqle-
6. Prove that
att al? at3 dtt an at3
a3t a3t a)s azt 422 a23

xe I:l: 3'i
a2t Aa) AB a3t alz 433

: th-- l (lhus, interchangrng rows 2 and 3 of the determinant changes its sign.)

13.3 Determinants of Order n

This section gives an alternarive definition of a determinant that is particularly
useful when proving general results. If you are not so interested in these proofs,
you might skip this section and reiy instead on expansions by cofactors in all your
work on determinants, as explairred in Section 13.5.
For a 3 x 3 matrix A = (a;;)3*3, [13.7] defined the determinant lAl as

AtP22aT - ana?3a32 * a12aVa31 - anaZtAT * a6a21A32

- aperA3l [*]
A closer examination of this sum reveals a definite paftem. Each term is the prod-
- e--i" \-u uct of three different elements of the matrix. Each product contains one element
from each row of A and one element from each column of A. In fact, the elements
in the six tenns are chosen from the marix A according to the pattern shown in
Fi-e. 13.4 (disregard rhe lines for a moment).
ln a 3 x 3 marix, there are precisely six different ways of picking three
elements with one element from each row arrd one element from each column. All
the six corresponding products appear in [x]. How do we determine the sign of

q lp. .4. t- J.
7- f
D
Y1
d. .l / 'a{
l+
o J /V
\JI c ld ( l
I

FIGURE 13.4
 lnversion
448 ChaPter 13 / Determinants and Matrix

in each of the six boxes' ttre following

ttre terms in [*]? By using the lines drawn
rule emerges:l

The Sign Rule

in mark in the array all the
e:.t:*'
To determine rhe sign of any term join all possible pairs of tlrese
we could
elemens appearing t ;;tl;;' [13.] 1]
will then either fall or rise' Now mark only
elemenc with lines' Thl.J lu";
righl If the number of these rising linesit is
is
those lines that rise to*tA 'f'"
is assigned a plus sign: if it is odd'
even, rhen ,t
" "orr"rporlin;-L#
assigned a minus sign'

box' no lines rise'

boxes in Fig' 13'4: In the fint
Let us apply this rule to the six exactly one line rises' so ana7ta33
so ayara33has a pius ti;;-' i; the fourth box'
has a minus sign' And so on' Suppose we pick n
Suppose A - (or)ixa is an arbiuary n x n marix'one
one eliment from each ro*
elemens from A, -a'#og""*""try n -To-:l""tly
elemenrs, giving an ex-
Take the product of these
element from each "oto*rr] (or
pression of the form a1,r'av'r"'anra'
Herlr' rz' "''r' rePresents a reshuffling
l';"" '' " n' The numbers f i' ' ' ' n can n choices:
be permuted
permutauoo) or tnt
'ot#"'
in n! : 1'2"'(' - I);;k;'"*
*ay't For rhe first element' there are
second element; and
therc are n I choices for the -
for each of these A"t "ioi*''
to oo'*o* or follows:
we define the determinant of A' detA lAl' as

where:
lAl is a sum of n'! terms
element
elements of the mauix' with one
1. Each term is &e product of n column' Moreover' every u3.l2l
from each .o* *J onc clement from each
which each row and each column is
oroduct of exac-Uf z factors' in
;;;;;; once' must aPPcar in.this sum'
"**tivt",* is found by apptying the sign rule [13.11].
2. The sign of """t,

m
unusual way Howevcr' ir is cquivaleo:
lThe sigo nrlc in [13'i I] is dcscribcd in a llcwhat
thcsignrulcbascdono""oo."poofcvenandoddpcrmuationsrharisusedinmostbooksonlrg
algcbra-
 Sec. 13.3 / Dcterminants of Order n 449
rEg In symbols, one has

att aD aln
a2t 4n
tAt : a2!r

[13. l3]
Onl an2 ann

=L(*)at,,a2^ anrn
I
Example 13.7
Consider the determinant of an arbirary 4 x 4 matrix A = (o4)r*,

art ay7-.@1 aru

ME
I

lAl: aul
"rt @ ort oto
fis.. aq atz at3 @ I
I

:: d=
: 4 .3 .2. I = 24 terms. One of these terms is aBa2ta3?a41,
It consists of 4!
ctt and the co*espoDding facrors are the boxed erements in the array--what
sign
@ should this term have? According to the sign rule
[13.11], thaterm shourd
have a plus sign because there are exactly two lines that rise to the right.
E([ check that rhe four indicated terms in the following sum have been given
the correct sign:
tr,EE
.@ lAl : a,arraSax - a72a21ayd4a+ . .. + a!aZta32a$_ ... + a1aa2ia31a41

Note that there are 20 other terms that we have left out.

The determinant of an n x n matrix is called a determinant of order

n. In
general, it is difficult ro evaluate determinants by using definition
tl3.l2l, even if
n is only 4 or 5. lf n > 5, ttre work is usually For example, if n = g,
then z! :720, and so therc are 720 tcrms in "oo.ro*.
the sum defining the-dercrminant.
Fomnately, other me&ods are available that reduce the work considerabry.
Most
i -T compurcrs have standard programs for evaluating dercrminants.
-Jq
In a few special cases, it is easy to evaluate a detenninant even if the order
is high. For instance. ir is easy to see that

a\ an tlln
oan Ar1
= AllAZ2...Ann ll3.l4I
00 dnn
lilD
ri.l'' Here all the elements below rhe main d.iagonal are 0, and rhe determinant
is
the product of all the elemenrs on rhe main diagonal. Tb see why, note
that
450 chapter 13 / Determinants and Matrix lnversion

all from column 13.4

in order to have a tenn that is not 0, we have to choose 1'

Fromcolumn2,wecannotchoose@l2,becausewehavealreadypickedanei-
emenr from the fust row (a11). Hence, from column 2'
we have to pick a22
in order to have a tenn different from 0. From the third column, we have to
pick a33, and so on. Thus, only the t€rrrl 41141 "'ann c?trt be * 0- The sign
of this term is plus because none of the lines between the elements in the
term
rises.
triangular' The<
The matrix whose determinant is given in t13.14] is called upper
If a marix is a transpose of an upper triangular matrix, so tbat all elements above

the diagonal are 0, then the marix is lower riangular. The determinant
of a I
lower triangular matrix is also equal to the product of the elemenu on its main
.,
diagonal.

3.
Problems
4.
1. Use the definirion of determinant to compute the following:
1000 0 0 0 ll
0200 0 0 r 0l
a.
0030
b.
0 1 0 0l 6.
0004 I 0 0 0l
1001 I 0 02
0100 d.
0 I 0-3 8.
C-
0010 U 0 14
abcd 2 3 4 1l
2.Thedeterminantofthefoliowing5x5matrixconsistsof5l:l20terms.
with its
one of &em is the product of the boxed elements. write this term
correct sign. 9,

atr lr,;l an atq ats

a! ah 'jf] a)t dy
a3r a32 c.zi a3a E
Wt a$ a$ a+t a45
ast as: an ijC a5s

rhe term indicated by ttre boxes with ia colrect sign. (See the previous
3. write
problem.)

arr an aB aA @
azt aaa a73 @ a2s
a3r lZrl a33 ala a3s
aqr Z; @ au a4s
E asz 4$ a54 ass
 Sec. 13.4 / Easic Rutes for Determinants 451
LIi.i-. - 13.4 Basic Rutes for Determinants
:.ii 3--
Based on definirion [13.12] of the determinanr
of an n x r matnx A, one can
"a.' : !: prove a number of imponant propertiel.
certainiy, many of these are m"inry or
theoredcal interesq but they also make it simprer
to determinanr.
"rutrar.

i!t-.1' Theorem 13.1 (Rutes for Determinants)

3:': :
"
!. .- : l' If each of the entries in one or more rows or corumns of A is 0, then
IAI = 0.
2. The determinant of the transpose of A is the same
as oe dererminant
of the marix irselt lAl : lA,l_
3' If B is the marix obtained by muldprying each enrry
of one row (or
column) of A by the same number ,n.n lBl = alAl.
4. If two rows (or columns) of A are ",
interchanged, then the determinant
changes its sign, but keeps its absolute value.
5. If two rows (orcolumns) of A are equal, then
iAl = 0.
6. If two rows (or columns) of A are proponional, rhen
lAl : Q.
7
' rf a scalar muJtiple of one row (or column) is added to another- row
- - " (or
\
column), then the value of the determinant does
nor cirung..
8. The determinaat of the product of rwo n x z matrices
A and B is the
product of the dererminants of each of the factors:

lABl = lAl.lBl u3.i5l

9. If A is an n x n marrix and a is a real number.
*ten

laAf - 6v'141
[13.16]

3., 4, aln,d 7 implicitly assume that all

\utes other rows (or column5) remain un_
changed. Ir should also be recaired thar the
determinant of a sum i,
the sum of &e determinans: t*,iiil ,r,
E?*'1lflSi

lA+Bl +lAl+lBl
An exampiie of this geaeral inequaliry was asked
for iu hoblem 5 of Section r3.1.
hoofs for most of these propenies are given at
the end of this secdon- First,
however, let us illustrate them in iome ,p""ill
"u."r,
Rute t: l"d, ? = ari .o - arz .o :o
l
452 Chapter 13 / Determinants and Matrix lnversion

all atz @tr azt

:
Rule 2: lAl : a1r Aaa
: ano>-at2a21 ' lA'l = Atz Qtt
altazz-a\za?l

We see that lA'l has exactly the same tentrs as lAl' ln particular' lA'l : lAl
atl atz : afi(aaz) - an(aau)
Rule 3:
da2t da??

att at7
: d(attaz. - 417a21) = s A?t All
SU
AD
Rule 4:
a2l : A2lAlZ - Att,!: -(A1ya-t - AnAZt): - l::j Z';\
A1 atz Ex
A1 a : altat? - :
Rule 5 a12411 Q
A1 a 1

[,et us a]so see how rule 5 helps to confirm the result in Example 13.5
of sec-
tion 13.2. Notethattheproduct (b-a)(c -o)(c-b) is0if b: a'c: a'or
c : b, and in each of these tlree cases, two rows of the matrix are equal.
l1
Rule !'= \: at(flan) - an(Fat) - F(arap -
6: I l" Parz 411a12) = 0
I Fart I

Rule7: Multiply each entry in the first row of a determinant of order 2 by cr and
add the results pairwise to the entries in &e second row. Then the determinant Rr,
does not cuange its value. (Note careftrlly how we indicate
this operation.)
tic
in'
ati 412 d. att alT
a2t aD J aT+dar1 a2*aap Er
: att(al2 * aap) - an(azt * aa11)

: aoaz * uattatz - anaT - d.a12a11 -- attaL? - antt?t

att atz
a?t aD

This rule is panicularly useful for evaluadng determinants.

we give two examplet

Example 13.8
I 5 -l 1 I 5 -1 ll s -l2 -3
-1 13 .l_
-l+l l+5 3 + (-l) :lo 6
j 2l 2 I l: 2 | --l
15 -l r 5 -l
06 2 J,/6 0 6 2l I
Rr
0 -13 4 <-J 0 0 25/31
:r.6.T=to
 Sec. 13.4 / Basic Rules for Determinants 453

Here, I times the first row has been added to the second in orcier ,. ,arn a
zero in the fint column. Then (-3) times &e firsr row has been addcd ro the
third, which gives a second zero in the firsr column. Thereafter. l3i6 times
the second row has been added to the third, which adds a zero ro the sec-
ond column. Note the way in which we have indicated these operations. [n
the end, they produce an upper riangular matrix that we evaluate by means
of U3.l4l.

ln the next example, the first two steps involve more than one operation
simultaneously.

Example 13.9
a +b
a
a
a*b
a
a
-i'l : 3a* b 3a*b 3a*b
a+b a
icc-
a a a*b I a a a +b
;6 1 I I -a -a 1l I
:(3a*b) a a*b a :(3a*b) 0b 0
a a a*b 1__l 00 b

d : (3a * b). 1.b -b = b2(3a + b)

Rule 8: This is also an important properry. If you have done hoblem 4 of Sec-
tion l3-1, you have already proved it for 2 x 2 matrices. Here is an example
involving 3 x 3 matrices.

Example 13.10
Check that lABl : lAl . lBl when
I 5 -l 30
A- -1 I 3 B_ -1 I
3Z I 52 3)
Solution Here lAl = 50 (as in Example 13.8) and lBl : -5. But

-t J -l
AB: 11 7 7
-T 124 9

So we find that lABl : -250 = lAl - lBl

dAtt dAO :
Rule 9. aanqan. - d,ct12d.a21 : d2(anar - anazt)
AAZ1 dAtc
1 4tt atz
= cl- a2t az

t
454 Chapter 13 / Delermnants and Matrix lnversion

Theorem 13.1 exhibits some of the most imponant rules for determinants' Confi-
dence in dealtng with them comes only from doing many problems'

On the Proof of Theorem l3.l:

Rule l: Each of the n I rerms in the determrnant is a product that includes an
elemenr from ar least one row (or column) in which every enEy is 0. Hence,
each term in the determinant is 0- and so the whole deerminant is 0'
RuLe 2: Each term in lAl is the product of enries chosen from A
to include
exactly one element from each row and one element from each column. Exactiy
dle same lerms. rherefore. must appar in lA'l dso'
Whataboutthesigns?Considerthefollowingspecialcase'withn=5:

att AB att dt5

A?t Att ld";'l
l-J
a?c aa5

lAl: a3t
444 aqs
a5t as2 asl as5

dlt a)t a
a1 ac2 asz

ir(l : AB 433 a4t a53

att 4zt dy
at5 Ay ass

The term at?a73a35a4'1a51 appears in both lAl and in lA'l' Of rhose lines
joining pairs of enries in lAl, the number thar rise to the right is 4, as is the
conespooding number of lines in lA'1. This is true in general. Transposing
means reflection about rhe main diagonal afiaa1 ...ano. fr the line I
jorning
any rwo enEies in one of the terms of lAl rises to the right (as it does bcrween
a4l and a35 in the iliustration)' then the mirror image L' of this line, which
joins the same rwo entries in the corresponding telTn in lA'1, will also rise to
the nght and vice versa.
Rule 3: F-achrerm in rhe sum defining lBl is a mulciplied by &e corresponding
term in the sum defining lAl- Hence, lBl : alAl'
Ruk 4: Ftsc consider the case in which two adjacent rows of lAl are inter-
changed. namely, rows i andi + I' in order to obain lA.l as indicated in dre
following displaY:

att atz

ait arn
lAl: at+r.l ai+t2 ai+ln

anl an! a@
 Sec. 13.4 / Basic Rules for Determinants 455

r:ir Confi-

dlt atz

ies ar
ai*t.t ai*l ai*ln
lA.l= ail ain

ra;i r
Anl An2 aM

Using the definitions of lAl and lA"l, it is clear that any rerm appearing in
lAl
also appears in lA"l. and vice versa_ Whar about the sign? Think of the pos!
tions of the individual entries of a term in lAl and of the corresponding term
in lA.l. All the lines joining enries that have nor changed their positioni have
the same slope as before. The same is rue for lines joining enuies whose posi-
tions do not change to entries in the two rows that are interchanged.
6t_ooi, for
example, at the line joining arn to a;2 in the rwo previously displayed determi-
nans-) It remains to consider the lines joining wo entries in the rows rhat are
interchanged. They must clearly change from rising to falling as one gocs to the
right. Because each rcrm in lAl and in lA.l contains exacdy one pair of enries
from these rwo rows, and because the line joining these rwo entries has changed
from rising to fallin-s as one soes to the righg the number of lines that rise to the
right in these two determinants must differ by exactly l. Hence, lAl
= _l{1.
Suppose next that rwo non-adjacent rows are to be interchanged. Smn
by interchanging the upper one of these rows with the row immed.iately below
it, then with the row next down. and so on. Suppose that s interchanges are
needed to bring the previous upper row into its new position. The second of
(::ng the two rows to be interchanged is now immediately above its original position.
Interchange it successively with rhe row directly above it exactly (s 1) times
)-lL: i -
to bring it to irs final new position. In all, we bave perforrned s +s _ I : 2s _ I
,'i €i:
, L. ^- interchanges of neighboring rows. Each of these interchanges changes the sign
15€ t:. of the determinant, and in all. we have an odd number of sign changes- Hence,
the determinant /ras changed its sign by interchanging two rows.
Rule 5: If we inrcrchange the two rows that are equal, the dercrminant will
be exactly the same. But according to rule 4. the dercnninant has changed its
sign. Hence, lAl = -lAl, which means thar lAl 0.
=
m fie Rule 6: This follows immediately from nrles 3 and 5.
Rule 7: Symbolically. the proof of this rule is as follows. for the case when
the scalar multiple a of row i is added to row j:

f {+)o,,, ... airi * aa;,)... an,,

-- (ai,,
s- -.
: l{+)ot,.. -Q*,... aj,t...an,, * a,f{*)r,,, ... airi . -ai,j.. -an,n

: lAl +a.0 = lAl

crhe last sum is zero bccause it is equar to a determinant with rows i and 7
equal.)
456 Chapter 13 / Determinants and Matrix lnversion
Harde
RaIeS:Theproofofthisruleforrhecasen=2istheobjectofProblem4in
l3'l' Here Problem I 1' suggests a proof of the-general case'
Section
multiplying each ettuy in A by o' By
9
Rule 9: T"be matrix oA is obtaineJby
there are n rows each of which
rufe E. tcet is then equal to a"lAl' bot'*
has o as a factor in each entrY'

Problems
10.
2\ - I
t.LetA=\l/t 4) B:( 1)
\) ol_ .

a- Compute AIi, BA, A'B', and B'A''

b.ShowthatlAl:lA'landlABl:lAl'lBl'IslA'B'l:lA'l'lB'l?
/2 I 3\
z.:.,,te'=(i; ,)
\r 2 5t
Compute A', and show next that IAI = lA'1. 11.

Evaluate the following determinants as simply

as Possible:
3.
30 t 2 I 1 2 3 2
4l
4l
l 0 -1 .b'loo
I -l
a.
2A 5
8
6 o 3 -rl
-10-ll2 -r -6 -9 -l2l
allZa
q
l -a 3 0
a 2a I
3a l0 0
4. Prove that each of the following determinants is zero:
II 2 J I a b+c x-y x-y ,2_y?
a. lz 4 5 b. I b c+a c. I 1 r+)
l^ 1 c a*b v 1 x
IJ 6 8
/l
Ir
lrtX= lr
5.

Compute
\r
X'X and lX'Xl''
ii)
6. I-et A and B be 3 x 3 matrices with lAl
:3 and lBl : -4' Where possibir*
I - 2Bl' lAl + lBl, and lA *
3lAl, Bl'
f,nd the numerical values of lABl,
T.Showthatanorthogonalmamx(seeProblemTofSectionl2.9)musrhatt
determinant I or -1.
13.5 Expar
8. A square matrix A of order n is called involutive if Al = I''
iUo* that the determilant of an involutive matrix is I or -l'
". / o\ ,oo (i , - o') are involurive (for aII a Accor
b.Showrhar(-16 _,/*-\r _a )--- sum (
- A)(I' + A) :0'
^
c- Show that A is involufrve c+ (I, from
 Sec. 13.5 / Expansion by Cofaaors 457

Harder Problems

9- Without computing the determinants, show that

b2+c2 ab ac u c ba
ab a2+c? bc c 0 ol
ac bc a2+b2 b a 0l
10. Prove that
a*b a a
a a*b a
D,: = bn-t (na + b)
:

a a a +h
(.Fllzt; Sody Example i3.9.)
11. a. l*t

lAt : dtt
a2t
at2
a2 tBt : brr
bzt
br.
b==1
I

Prove that
afi an 00 +-
tAt .tBl: A?t 412 00
-r0 brr be a aTl
0-l hr bZ An 412

I0 0 atrbn *anht anbe* anbn

:l
t0 0 atbrt * anb2l a71bp * a72by
t-t 0 b,, bp
lo -l bt bzz

: lABl

flffe have indicated the operations you should use to obtain the second
equality.)
b. Try to generalize the method in part (a) so that it applies *'hen A and
iiil B are arbitrary z x z matrices, and &us give a proof of rule 8 in Theo-
rem 13.1.
TG

13.5 Expansion by Cofactors

According to definition [13.12], the determinant of an n x n matix [ : (a;;) is a
I
sum of z! terms. Each term contains one element from each row and one element
from each column. Consider in panicular row i and pick out all ttre terms that
458 Chapter 13 / Deteminants and Matrix tnversion

have a;1 as a factor, then atl terms that have a;2 as a factor, and so on. Because
all the terms have one and only one element from row i. in this way we get all the
terms of lAl. So we can wrire

IAI :airCir *a;2C;2+'.-+ aijCij *...*ainC;n tl3.l7]

This is called rhe expansion of l{l in terms of the elements of the ith row. The
coefficients C i t, . . . . C i n Na the cofactors of the elements ar r, . . ., ain.2
In the same way, we can expand lAl in terms of the erements of the 7th
column:

lal: a1iC1i * a21Cyi + .. .+ a;1C;1 *--- * a4Cni tl3.l8l

what makes expansions [13.]71 and [13.18] extremely useful is r]rat.in general each
cofactor c;1 can be found by applying the following procedure to the deerminanr
lAl. Firsr delete row j and column 7 to arrive at a determinant A;; of order n - 1.
which is called a minor. Mulripl.v rhe minor by the factor (_l),*j. This gives rhe
cofactor.
ln symbols. ir turns out thar the cofactor C;; must be gven by

alt at. j-t a at.j+t aln

dzt At ,-r dt ;-t a?rl

C;; : (-!)i+r [ 13.19

anl an. j 4n. -i+t ann

-1

where lines have been drawn through row i and column j, which are to be deletec
from the matrix.
A proof of [13.19] will be given ar the end of this section. If we look back
at [3-8] in Section 13.2, however, it confirms tl3.l9l in a special case. Indesn
Put lAl : anCtt * apCp -f c13C13. Then [13.8] implies that

: (-t)'*r
l-tr
' iI |u
a:r a't I c',: (-l)'- ,=.lo', a-
ai2 1'
afi l.
C,, C,= = 1-1;r+z I "
I I a:t ,3, l' ,;_ lo;;
precisely In 3eqelrlanss with [13.19].
Generally, formula [13.19] is rather complicated. Test your understanding cr
it by studying the following sxample.

lThe expansions in 3. 17] and

[1 t I 3. I 8] are also called cofactor etpnsions of fie dcrmri.*
which is a specia.l casc of thc general l-aplacc expansion.
 Sec. 73.S / Expansion by Cofaclors 4Sg
ln Beaauss
e ser all '&3
Example t3.ll
Check that the cofactor of
the element c in the determinant
30 0 2
1_'.1
tAt :
6I E 2 3 a2
*
-l I 0
52 0
0
IS C23 1-1;:+a -l IO
i:
-r,:'.,.
. T':: 3 5 /-)
Find the vaiue of lAl by
. -i.L^
: -- Uti
_* using Example 13.4 in Sec.
I3.2.
solution Because the erement c is in row
has been written corrcctty-. 2 and column 3, its cofactor
To find,h"
of iAi. rve use [13.18]
lll i i_ and expand in terms of rhe.elemen" ";;;;jvatue
;;"rA; cotunnof lAl (because
has so many zeros). This yields "f it
ien:ra-I ec:
ciet:rm:,ra-c:
lAl:azzCu:6(-112+3
J 02
crder r - -. -l IO = c(-1)(-5) = 5c
':s :rr:: ix 5 L5
Example 13.r 1 shows a simple
case in which we caa evaluare
by using cofacror exnanslol. a determinant
Th"--"-"-;'t, i*"rrrry simpre because
many zeros in the third coru*n-
tr.n" ze.os'ar",not there inidalry, we there are
to rure 7 in nreo.em i:.r caa often
fi::Lt1.tl:[1;lj,, ors""tio;r3r];;; exampres

Exampte 13.12

3 -l 2 3 -l 2
0-l -l 0-l -l 1,,,i -l -l
6l 2 03 _.> j ='J 3-2 =3(2* 3) = t5
To derive the equetiry labelled (l),
I it 3rr+fl expand by column I

Example 13.13
a ,:.:(. :U,i&,
Lia tl*T 2 0 3 -1
0 400 (l)
: (-l)i+- .4
:5 -l -7t')
0 1-l 2 0-l 2
J 25-? -3
ta
l1 3 -I
4 lo -l 2 o l-r 1 r

i<:r:ru m
lo l/2 -?/) 'l ,,, -i,rl:8 (3 - . ) : +

For equaliry (l), expand by row 2. For equaliry

(2), expana by column
l.
3To computc
E JSE@II". a gencfial
d€tcrminant usiog defnition
362rT.9f) gopcradons;f [13.12] rquircs no fcwer rhan
d.lo-i-t1
rcducerhe'.i,"*-il;;:H#l}:fiL]lolr"'*" use orrure z r" n*"*, r3.t can
460 Chapter 13 / Determinants and Matrix lnversion

Expansion by Alien Cofactors

According to [13.17l and [13.18], if the elements of any row (or column) of a
determinant are multiplied by the cofactors of the elements in that row (or coiumn)
and the products are added, the result is the value of the determinant. What happens
if we multiply the elements of a row (or column) by the cofactors of a different
(alien) row (or column)? Consider the following example-

Example 13.14
If A : (aij):,t, then the expansion of lAl in terms of the elements in the
second row is

lL : a21C21 * at.Cr * axCzz

Suppose we change the elemens a2t, ar, and a5 to a, b, and c. Then C21.
C::, and Cr Ne unchanged. So, the Dew determinant is

A1 At2 AD
abc - aCu * bCr, * cCv [*]
a3t a33 a-33

tn particular.if we rcplace a, b, and c by all, 412, and 413 or b) a3t, a32'

and a33, then the determinant in [,r,] is 0 because two rows are equal. Hence,

a11Cy*ypCn*arcCx=0
a31Cy*432Cn*atsCt-0
Thus, the sum of the products of the elements in either row I or row 3

mulnplied by the cofactors of the elements in row 2 is zero'

Obviously, the argument used in this example can be generalized: lf \r'

multiply the elements of a row (or a column) by the cofacto6 of a dif[erent ro*
(or column) and then add the producs, the result is 0.
We summarize all the results in this section in the foilowing theorem:

Theorem 132 (Cofactor Eryransion of a Determinant) Let'A

(aij)n,,. Suppose thar the cofactors C;1 Ne defined as in [13.19]- Then:

anC;1 *a;2Ci2*' '*ainC;n:lLl

[13.::
a;1Cy1 * a;2Cp*' '*a;nC1rn=$ (k+i)
atjCtj*a21C21 {' '* aniC4 : lAl --
[13
at jCr* * a2lCy; * ' -+a|jcnk-0 (k+i)
 Sec. 13.5 / Expansion by Cotactors 461

Theorem 13.2 says that an expansion of a determinant by row i in terms of

the cofactors of row k vanishes when t I i, and is equal to lAl if ,t : i. Likewise,
Ir an expansion by column 7 in terms of the cofactors of column & vanishes when
Ed k t' j. al.d, is equal to lAl if t : j.
lGE
AE Proof of formula [f3.19]: The definition of cofacror Ci7 (see ti3.l7l)
implies that C;7 is a sum of terms, each equal to the producr of n
- I el-
ements from the marix A, chosen in such a way that there is one element
from each row except row i. and one element from each column except
L column j. k follows thal except for a possible change in sign. C;y is the
determinant of the submatrix formed from A by crossing out row i and col-
urnn -1.
It remains only to derermine rhe correct sign. Look first at C1 1. The
determinant obtained by crossing out row I and column I is
11-
422 a23 arl
432 433 a3n
[*j
i4 an? an3 ann

The signs of the terms in Cy1 are the same as the signs of the corrcsponding
terms in attcl.The signs of the terms in a11C11 are determincd by tle oum-
EE" ber of lines joining pairs of eiements that rise as they go to the righr" including
all those that rise up to a1i. However. there cannot be any lines drat rise up
to all, so C11 is equal to the deterrn:nant [*J.
Consider next C,j. Move row to row I by i i I interchanges of two
-
adjacent rows. Tbe i I prcvious rows, while rctaining their old order, arc
-
moved one position down. By pcrforming j I interchanges, the new column
n$ j is moved left to the position of column l.- Because we have inrcrchanged
i - 1+J - I = i+"r -2 rows and columns in all, the determinanthas
changed sign i * j - 2 times. Because (-l)i+i-2 : (-l)i+j, lAl must bc
G given by
:il
ai1 ai. j
-l ai. j+t ain
At; atl at.j-l at. j+t aln

: :

lAl = (-l)i+i ai -t.l ai-t. j -l ai-t- j+l ai-l [**]

ai+t. j ai+t -t ai+1. j-l ai+t-j+t ai+l
: : :
anj anl an.j-l an.j+l dan
ifi
As with [*], the cofactor of the elcmenr a;; in [*,r.] is equal to the determi-
nant in u3.l9l, bccause by crossing our row I and column I iq [**] we ger
!-u@M
the same rcsult as by crossing out row i and column j in the original dacr-
minant. Hence. the complement of a;7 in the expansion of lAl is givcn by
113.191.
462 Chapter 13 / Deteminants and Matix tnversion

Problems
t
rl
1. Use the method in Examples 13.12 and 13.13 to compute the following
a
determinants:
t2 J 4 213 3 :
12 4
a. 13 9 b.
0-1 0 1l 32 r 6 n
t416 2-l 0 3
c.
r 30 9
-20 I 241t2 E
2. Compute the following determinants:

000a 0000 i
00a 00b0 00051
a- 0b0 b.
0c00 c. 00312
c00 d000 04 0 34
62 312
Harder Problems

3. One can prove that for all n :2,3,

xy xia ,i-'
X? x; *i-' *
(xi
fl
lSj<f Sa
xj)
xn x; ,:-'
This is the vandermonde determinant- The product symbol II means thar
we uke the product ofall factors ofform xi-xi for i > j, where i and varr
7
from 1 to n. Qf n : 3, then.IIrsj.i.l(xi - x) : (x2 -.r1)(x3
-xr)(x: -x:r.
See Example 13.5 in Section l3-2.) Prove the equality for n
= 4. (Hin;:
Multiply each of the third, secon{ and first columns successively by -r,-
before adding the results to the following column in each case. Then use tbc
result for z :3.)

13.6 The lnverse of a Matrix

Suppose tbat a is a real n .mber satisfying d * O. Then there is a unique number
o-r with the properry that ocr-l : a.-td: l. We call c-l the (muitiplicaur: l.
inverse of a. we saw in Section 12.8 that the identiry matrix I (with I's along tir 1

main diagonal and 0's elsewhere) is the matrix equivalent of I in the real numb< 3.
system.a This makes the following terminology seem narural.
Given matrix A. if rhere exiss a matrix X such that As f(
an lr
AX:;1L -1 trAI lAxt
matri
4From no* on. we writc I instead of r, whencver the dimcnsion n of the idcntiry ,.r- I
obvious.
 Sec. tg.6 / The lnverse of a Matrix 463
then we say that x is an inverse of A. Moreover,
A is said to be invertible in
this case. Because XA AX I, the marix A is also
= = an inverse of X_that is, A
and X are inverses of each other. Note &at both the
two matrix products AX and
XA are defined and can also be equar onry if A and X are square
marices of the
same order- Thus, onry square marices can have inverses.
Bui not even alr square
matrices have inverses, as the following example shows.

Example 13.15

(a) Show that

56
A= ( 510 and X= ( r/2 -3/r0
) -r /4 r/4 )
are inverses of each otber,
(b) l0
Show that A = ( 00 ) has no inverse

Solution

(a) (; ,3)(_1ft -{):):(i,i:fli_ -ts/t0 + 6/4

-15lt0 + r0/4 )
=(; ?)
and likewise we verify that XA: I.
O) Observe that for all real numbers x, y, z, and u,
e#
tu (;
0
(x
-qk 0 ) J)=(; d)
rtr So there is no way of choosing x,
!, Z: and rp to make the product of
these two marices equar to, **, (i
3) * no inverse.

The foltowiag questions arise:

1. Which matrices have inverses?

2. Can a given matrix have morc than one inverse?
3. How do we find the inverse if it exists?
As for question r, it is easy to find a nzcessary condition
for a maEix A to have
an inverse. In facr, from [13.22] and rure g in Theorem
t3.t, ii fottows thar
r!i lAxi : lal . lxl : lll. Using tl3.l4l of Section 13.3, we sce that the identity
matrix of any dimension has determinant r. Thus. if
X is an inverse of A, then
I
lA['iXl: t I*]
464 Chapter 13 / Determinants and Matix lnversion

We conclude from this equation that lAl I 0 is a necessary condition for A to have
an inverse, because IAI : 0 would violate [*]' As we shall see in the next section,
the condition lAl + 0 is also szffcient fot A to have an inverse- Hence:

A square matrix A has an inverse <+ lAl l0 t 13.231

A square matrix A is said to be singular if lAl : 0 and nonsingular if lAl # 0

According to Ii3.23], a matrix has an inverse iff it is nonsingular-
Concerning question 2, the answer is no-a matrix cannot have more than
one inverse. Indeed, suppose that X satisfies [13.22] and moreover AY : I for
some other square matrix Y. Then
Sc
Y: fY : 6A)Y : X(AD : )C[: X If.
shows that if YA : I, then Y : X- Thus, the inverse of A u
e4r
A similar argunent
unique { ir exisrs.
The inverse is usually wrinen A-1. Whercas for numbers, we can wri'e
a-t : lla, the symbol I/A has no meaningl there are no rules for dividing mam-
ces. Note also that even if the product A-'B is defined, it is usually quite differea:
from BA-l because matrix multiplication does not corlmute- To
The full answer to question 3 is given in lhe next section. Here we on-r [13
consider the case of 2 x 2 matrices. The

Example 13.16
Nor
Find the inverse of A : (a b
(when it exists)
Ev
d )
Solution We shall find a 2 x 2 matrix X such that AX - XA: I. lh:
requires looking for numbers x, ), a, and u; such that

: (l 0\
(: :) (: l,) t) Prt
Matrix multiplication implies that W-e

ax*bz:1. ay*bw:0
cx*dz:0, cy*dw:7 Tbeorem 13
- lEut_
Note that we have two systems of equarions here. one given b.v &e
equations on the left, and the other given by the two equations on the s .a -:
Both systems have A as a co[lmon coefficient matrix. If lAl = ad - i: = ':, .{B rs :
solving them using Cramers' rule ([3.5] of Section 13.1) or other*'i-
- ..,-
-i
) -c -b - --{
*-
ad-bc' '- ad - bc'
t- ad-bc' U=-
 Sec. 13.6 / The lnverse ol a Matix 465

ir i,rr A to ha\. Hence, we have proved that if lAl = ad - bc 10, tben

:e next secoor'-
ei'!ce:

A: ( a b d-b
d) ==:+ A-l ad-bc ( -ca 113.24)
C
)
tr -.- -

For square matrices of order 3, one can use Cramers' rule [I3.9] to derive
rlar l: -{ = a formula for the inverse. Again, the requirement for the inverse to exist is
..r o 6^d _af that the deterrninant of the coefficient matrix is not 0. Full details wiil be
grven rn Secuon lJ. /.
;er_{Y=I:cr
Some Useful lmplications
If A-! is the inverse of A. then A-lA : I and A,A-t = I. Actua'ly, each of these
e4uations implies the other, in the sense that
r.ve.se o.l A :

;. 11 9 q41 *'n'i
I :::+ x-A-r
AX = Ir 3.25]

.jjriding matn- YA=I -+ Y: A-l [ 13.26]

\ erute ditferee:
To prove [13.25], suppose AX: I. Then lAl . lXl : 1, so lAl I 0, and &en by
Here we tui U3.231, A-r exisrs. Multiplying AX: I from the left by A-r yields X: A-r.
The proof of [3.26] is almost the same.
These implications are used repeatedly in proving properties of the inverse.
Note also how we use [3,25] in the next example.
Exampte 13.17
la:r
Find the inverse of the n x n matrix A if A - A? : I.
-l{-I
Solution The matrix e4uation A - A2 : I yields A(I -A) = I. But then
it follows from 83.25J that A has the inverse A-l : I - A.

Properties of the lnverse

We shall now prove some usefi.rl rules for the inverse.

Theorem 133 (Properties ofthe Inverse) Let A and B be invenible

,:xnroatrices.Then:
T*11(
ti3i. l:
=:
:3a-s :: :'€ :g!u; (a) l-t is invertible, and (A-r)-r = 6.
-- - i[
O) AA is invenible, and (AB)-r - B-rA-r-
r a:::;--:< '-drNi (c) Ttre transpose A'is invertibie, and (A')-l : (A-r)'.
(d) (cA)-t :6'-ra-l whenever c is a rntmber 10.
466 Chapter 13 / Oeterminants and Matrix lnversion

Proof !n each case, we use [13.25]: Solvi

Let A
(a) We have A-lA - I. so A : (A-lr-1. are ma
(b) To prove that X: B-rA-l is the inverse of AB' we just have to venfy
that (AB)X is equal to I. But

(AB)X = (AB)@-rA-r): A68B-|)A-r

= [lr1-l = fu{-1 = J In case
Provrdr
(c) Transposing the equation A-lA = I in accordance with [12'a4](d) in
Section 12.9 gives (A-lA)'= A'(A-1)' :l'= L Hence' (A')-l =
(A-'x. Theorem 13.4
(d) Here (cA)(c-lA-l) :66-t6a-l = I 'I: I, so c-lA-l = (cA)-l

Nofes
1. It is imponant ro *unk carefully through the implications of the four rul*
in Theorem 13.3 and to understand their uses. A somewhat dramatic stot-:
might help you to appreciate nrle (c)' Some years ago a team of
(huma'o t
P
calculators was working in a cenu'al bureau of statistics. After 3 weeks b.

of hard work, they finally found the inverse A-r of a 20 x 20 matrix 'l c{

Then the boss came along and said: "Sorry, I was really rnterested in ti* dr
tc
inverse of the transpose of A-" Panic-until they realized that propen-v-
s€

would save them from having to redo all the calculations. They simpi;u
n
transposed the inverse matrix that it had taken 3 wecks to fiod, becau-*
to (c), the inverse of the transpose is the transPose of the tr- Examp
"".oiding
verse. S

2.SupposethatAisinvertibleandalsosymmetrlc-{hatis'A':A'Thenru€
(ciimplies that (A-r)': (A')-t : A-1, so A-l is symmetric' The inver:e
of a symmetic nwtrix is sytrunetric'
3. Rule (b) can be extended to products of several matric€s. For instance' t- '{
B, and C are all invenible n x n matrices' then I
(ABC)-r : linrlc)-r : c-' (AB)-r
: C-r (B-rA-t ; : 6-t3-ta-t
T]

where rule (b) has been used twice. Note the assumption in @) that A
2"tI m
In and ecooometrics. we often cons,js 1I
B are both n x n matrices. statistics
products of the form XX" where X is n x rn- Then )o(' is n x n. lf the
o*
ie.minant IXX'I is not 0, then G(X'1-r exists, but (b) does not apply be'er;a
X-r and X'-l are onlY defined rf n : m'
4. It is a collrmon fallacy to misinrcrpret (d). For instance, a colrect applicas:u T1

of (d) is ttris: (*a)-r :zA-t - th

 Sec. 15.6 / The lnverse of a Matix 462
Solving Equations by Matrix lnversion
Let A be aD n x n matrix. If B is an arbitrary matrix, we consider whether trere
are matrices X and Y of suitable dimensions such that

AX=B tU
YA: B p.l
In case [I], the rnarix B must hav: n rows, and in case
[2], B must have z columns.
Provided these condirions are satisfied, we have the following result:

Theorem 13.4 If lAl * 0, then:

AX: B e=a )( = -4-lB u3.27)

.
YA:B +=+ Y=BA-l r1
3.291
--i--
s::{-.
JI:,3i Proof . Mulriply each.side of the equation Ax - B in [13.27] on the ]eft
- 3=Ir by A-r. This yields : A-rB. Because tA-ta)i = trX: i. *e
l:. -{ ^-t1aX)
conclude that X = A-lB is the only possible solution of the equ^tion. On
the orher hand by substituting X
= 6-tg into AX B, we ,o th"t it really
=
1.. satisfies the equacion.
The proof of [i3.28] is similar-rnuidpry each side of yA = B on the
right by A-l
L - --
E-_ Example 13.1E
Solve the following system of equations by using Theorem 13.4:
:c ri:!E
r:.€1e 2x * y:J
2x *2y :4
t1l
:- : -.{
Solution Suppose we define

A: ( 22 t) X: ( x
) ,: (i)
Then [l] is equivalent to &e matrix equation AX = B. Because lAi : 2 * 0,
-{ ,rilr matrix A has an inverse, and according to Theorem 13.4, X : A-rB. The
Esfu inverse of A is found from tl3.Z4),66d r,tls efnin
EiE c-
E:3@E
(;)=^-'(i) :(_l -'n (i) =(l)
iic@--@ The solution of [l] is, t]rerefore, x : l, ] = 1. (Check by substitution that
this really is the correct solurion.)
468 Chapter 13 / Determinants and Matix lnversion

Problems

1. Prove that the inverse of (; ?) ,'( r/3

213
0
-1 )
/l I -l I 0
2. Prove that the inverse of (;t -i)''( -zn 0
8r7 -1
rfi
3. Find uumbers a atd b that make A the inverse of B when

2-l -1 /l 2 4\
a 114 ,=(?
[:
r18 t/8 -r18 )
b and
\r)
4. Solve the following systems of equations by using Theorem l3'4' (Sec
Example 13.18.)
a.2x:31:3 b.7x-3y= 8 c 7x-3Y:0
3x-4y=J 3x*4Y:ll 3x-4Y=Q
| /-l - 6r Show ttrat A3 : I' use this to find A-r'
s. r.et A :
; (; l')-
10 l0\ 1
6.GivenmatrixA:(O I 1)
\l o t)
a. Compute lAl, A2, and A3. Show that A3 -ZAz +A -I =0, where I is
the identiry marix of order 3, and 0 is the zero matrix'
b. Show rhat A has an inverse and A-l : (A - D2' 13.7 A G
c- Find a matrix P such that 1l2 : A. Are therc other matrices with thit
properry? Th,
an(
7. .. *, n: (3 -l 1) .o*n"* AA', IAA'1, and (AA')-r' imr
(AA';-t
b. The matices AA' and (a) are both symmeric' Is this a
in Part of
coincidencc?
E. Suppose that A, P, and D are square matrices such that A: PDFI' nor
cor
a. Show thar A2 : PilFt. tlus
b. Show by induction that A' = PDUFI for any positive inrcger m' coi
_1ttL t) (
9. Given B = ( t14 , compute Bz + B, 83 - 28 * I, and then B-r
) -r/2 (the

10. a- Let C be a square matrix of order n that satisfies C2 + C : I. Show tba

Q haq 2s inversc aod C-l : I * C.
b. Show that C3 - -I + 2C and C = 2l - 3C.
 Sec. 13.7 / A General Formula tor the lnverse 469

11. a. Suppose that X is an m x n matrix and that lX'Xl I 0' Show that the
mabax

A=I.-x(x'x)-tx'
is idempotent--+hat is, A2 = A. (See Problem 7 of Section 12.8.)
ll
Check the result in part (a) for the particuiar matrix X = 2
h.
I )
12. lathe matrices A and T be defined by
) 3 '1 t3
T=fr
^: (i
1 I J and (? -83
_1 1) 2 r-3
where s and t are real numbers'
rem l3'4' (s€r a. Prove that T : A-l for suitable values of s and r'
b. Suppose that the matrix X satisfies tbe equation 31 : 2X * C' wbere
_ =0
.: (lr2
_1."-
/5 2 J 3 0

-ir =0 s:lz 3 5
)
and 0 J I
\s 2 3 5 4
) nnd A-l
Use the result in Part (a) to find X'
13. kt D be an n x n marrix such that DP = 2D+3I' Prove tloat D3
: aD*bl
for suitable values of a ad b. lipd 5imil21 exprcssions for D6 and D-t (that
is, expressed in the form qD+ 0l)'
-l=0,whereI*
h.
'3,7 A General Formula for the lnverse
r raarrices wltb =u: facts about the i-nverse
The prcvious section certainly Presents tle most important
ana its properties, and as such is "what every economist should
know." It is less
to the inverses
important io, *ort economists to know a lot about how
compute
r.$'r-1.
are available'
of iarge marices, because powerful comPuter ryograms
srmmetric. ls us o the inYerse of any
Nevertbeless, this secdon Presents an explicit formula for
inefficient for
nonsingular n x n matrix A. Though this formula is extremely
ft dT: theoretical inrcrest. The key to
r-{ = PDF .olnpr"uog inverses of large marices, have
ll-Z rt Section 13.5 on the expansion of dercrminans by
rh.is formula is Theorem
i.gve rntege; 'r: cofactors.
Let Crr, .... Cnndenote the cofactors of the elemena in A' Then by U3'20I
B - I. and
-ia D-' the rule for expansion by cofactors), one has

..2- C=1 Ssrs tAl ifi=I: [*]

airCu * aizCxz* "'* oiaC6: lf.i*k
{ 0
47O Chapter 13 / Determinants and Matrix lnversion

The sums on rhe left-hand side look very rnuch iike those appearing
in rnan-ix
products. ln fact, the n2 different equations in [*] reduce to the single mauix
equadon

all at? aln

Crr Ctr Cnt 0 0
Cn Crz c,2 lAt 0 \
ait ai? 4in

:
Crn Ctn Cnn ):(: 0 lAl
atl anT ann

Here the right-hand side matrix is equal to lAl 'I' Let C+ =

(Ci;) denote the
matrix in the product on the left-hand side
matrix of .ofu"torr. Then the second
is the matrix c+ with its row and column indices interchanged. Thus, it is tlre
transpose (C+)' of that matrix, which is called the adjoint of A' and denoted b)
adj (A). Thus,
Crr Ctr Cnr
Cv C*z Cnz
adj (A) : (C+)' = tt3.291

Ctn Cxn Cnn

The previous equation, therefore' can be wriuen as Aadj(A)

: lAl 'I' In case
lAl ; O, this evidently implies that A-r : (l/lAl) 'adj (A)' We have proved:

Theorem 135 (General pslnule for the Inverse) Any square mamx
A : (a,j) with determinant lAl I 0 has a unique inverse A-t satisfuing
^*n
AA-r : A-lA = I, which is given by

A-r :l.ritnl [1_1Ii

iAt

ff lAl :0, then there is no matrix X such ttrat AX = XA: I'

Example 13.19
Let
2 )
l: 4 J
I 2 i)
Show that A has arr inverse aod flnd the inverse-
 Sec. 13.7 / A Geneal Formula lor the tnverse 411
: in lnamx Solution Accordins ro Theorem 13.5. A
has an mverse itr lAi
,gie matrix 0.
X;:t,:H."alculation
shows thar lAi-:'-j,"ro the
inverse exists. =
'The

t-
cr = |lJ" 4l
I
. l: l0
I

C
4 II ^ lt rl
l- 2
I . l: -15 Ln=l- 5
lr 1t
-l
i: 4l 2 4l la
:) 12 4i- -4 Czz = 1 4l 4 L:r=-l- J I

lr 2 - -l

CJ
J
)
4 _ _9. C;:: _ lz 4 l't ?l
denote tre
I
l+ r =14 I I:-6
rl
:-hand side
Hence, the inverse of A is
s. rt is tbe
ienoed t'r
A-'= a
tAl
Ctt
Cn
Ctt
Cz:.
Ctr
Cy
(
l0 -1 *9
*15 4 t4
'l:
- tJ C3 Cst 5
5 -r -6 )
':
(Check the resuh by showing
that AA-r _ L)
Finding lnverses by Elementary
Row Operations
I. ir --"a-r
Theorem l3-5 presented a general
rci li: formula for tl
Although.r,irio.,,ur"isimportant,il;;d""d:,I,::tJr'rlH:x11*LllT
matrices much larger than 2 x 2. w.
,ha,;;;y discuss ;*;;;;; inding tire
is very efficient. Versions or i,
iil:::r:- *. i"n"rally used by computers to find

The fo'owing operations on a matrix

are called erementary row
operations:
(a) interchan$ng any rwo rows
(b) multiplying each elemear of a
row by any scalar a
G) adding element of the ith row, q times
I 0
the corresponding element of
::f*
Now' in order to inven an z x n
matrix A, firsr form the n x 2n
by writing down the n columns matrix (A : r)
of .q, rolloweil, *" ,
elementary row operations to this oi i. i:,"n +pry
"otu-r*
matrix in orde. to tansform it to an n x 2n marix
(I : B) whose first z corumns are
alr the corumns of I. It w,r
If it is impossible ,o p"troT-:r.t;;;;;"s. fonow that B 4-r
-
-
then A has no inverse.s The
method is illustrated by the following
"**pf.
5It is not
difficult to prove thar rhis proccdure work.
bur wc omit the proof.

I
I
472 Chapter 13 / Detenninants and Matrix lnversion

ExamPte 13-20
)
/r 3 3\
Find the inverse of A = (l '^ \) 3
6 matrix whose first three columns
Solution Firsq write down the 3 xtlree columns are the columns of the
are the columns of I and wtrose next 4,

3x3identitYmatrix:
/l 3 3 l0 0\
(r 3 4 o I o)
\t 4 3 0 0 rl
use elementary operadons on this matnx
The idea is now sysrematically to
sothat,intheend,thethreefirstcolumnsconstituteanidentitymatrix.Then
three columns constitute the inverse
of A'
the last
the result to the second row'
First, multiPly tn" first row by -1 and {d
column' You should be able
This gives a zero in th" ;;; row and the first
used and why they are chosen'
then to understand the other operations

/l 3 3 I 0 0\ -l /t 331 10 -1
(r 3 4 o I o) --J (o 0ll-l
\l 431 00 ?)
I
.-l
\r 4 3 0 0 r/
/l 3 3 1 00 0 3" 4 0

oI
-lo\oro -1 10 I (s 0lr-l
I 0 I -l
1

i)
-l 0t )
0
-3
/r 0 0 r 'l
-3 31

-
rl 0
[o o
0
I -1
-3 -3
1 0 <-l [o I o ", -l 0 ll
\o I o -l 0 I ) *l \oori-l 1 o)

We conclude that
1 -?
-7, 13.8 Cram
A-r -r01 1 0 )
-1 Cram
(Check that AA-r = I') of rhe
the sr

Problems

l.UseTheoreml3.5tocomputetheinversesofthefollowingmatrices'ifd*:t
exist: rl
)a 102 0 0tt
r)
a. ( 45 ) b. ( 2-r 0
o 2 -t ) ' (-i a

-; 8/
 *c. 13.8 / Camels Rute 473
/-2 3 2r
2. Find the inverse of le o 3)
\ + t -t)
0.2 0.6 0.2 r

e
3. Let A: 0 0.2 0.4 ). Find (r-A)-l
aoiuITInS
rES of &e
0.2 0.2 o)
4. consider the following problem. Repeated observations of a phenomenon
lead to a number of equarions with the same square coefficient marix (ai;),
but with different right-hand sides:

A11X1 *...*A1nXn=[i,

'irr. mlrr (i = l, .... p) [*]

arnx. Ttc: 4a1x1 *..-*arnxn:[,n

eaa'nd :o'* Explain how to find the solutions of all the systems by using row operadons
,r.iC be ab'L: to get

atl dln btt bpt 0 Di,

'i')
I
I (:
ant ann btn brn I bi" b;, /
What then is the solution of the system of equations [*] for i
U
i
= k?
1 5. use the method in Example 13.20 to compute the inverses of the following
maEices (check each result by verifying that AA-r
= I).
I 2
2 J 32 -l
a. (J 4 )
b.
(i 4 5 c.
( -l 5 8
5 6 -9 -6 J )

"rts Cramer'sRule
cramer's rulc for solving n linear equations in n unknowns is a direct generalization
of the same rule for systcms of equations with rwo or three unknowns. consider
the system Ax: b, or

_-ffi,
a11X1 * apx2 * * atoxn - bt
aztxt * atzx2 * * a4xn - b2
[13.31]

aa1x1 * ar2x2 *... * aorxn : $n

474 Chapter 13 / Determinants and Matix lnversion

l*t D1 denote the determinaat obtained from lAl by replacing the

jth column
vecror with the column vector whose components are b1, b2, ''', '
bn Thus'

4tt at j-t bt at j+t Aln

42t a2j-t bZ azj+t A?rl
Dj= (j : 1, . .. , n) [13.32i

anl anj-t bn anj+l ann

Note that expanding Dr' by its jth column gives

Dj : Crjbr * C4b2* "'* Cn.1bn [*]

where the cofactors c;i ue given by ti3.l9l of Section 13.5. Now we can proE
the following result:

Theorem 13.6 (cramer,s Rule) The general linear system u3.3tl

of z equations in n unknowns has a unique solution if A is nonsingular
(lAl + 0). The solution is
Dt 12: Do

lfi' "'
D2
x, = x"=lri [ 13.3i-
lli'
where Dr, Dz, ..., Dn ate defined by [13.32]'

Proof System [13.31] can be written in the form

at I al2 aln ,rl bt

d)t A1) av1 x2 b2
tu
anl an2 ann )(, )(
Xa bn

Using formula [3.30] for thc invene of the cefficient matrix A yields

Ctt Czt C,T

1
Cn Cn Cn2
t2l
(;) lAt
)(;)
where dre cofactors C;; are given by t13.l9l of Section l3'5' From [2]' wc
have in panicular that

I.^ Di (j=l-2.....n)
x1=fr;Gtibrr...*C^ib)=-tl
.
t3l
 Sec. ,3.8 / Cramels Rule 475

whcrc the last egualiry follows from the previous equadon [*]' This proves
: 7th column Ir 3.33].
Thus,

Example 1321
has solu-
ixamine for which values of p the following system of equations
[ 13.32] tions, and find these solutions.

px+ y -l
x- y*::0
,,.-,-?
[*l -J

, \\e can ProlE Solution The coefficient maeix has determinant

pl 0
A I -l I =l-p
02 .l
]J

According to Theorem 13.6' the sysrcm has a unique solution lf | - p* 0-

that is, if p t' l.In this case, the determinants in [13'32] are
ti3.--i-
1l 0 pr 0 p 1i
Dt: 0-l I Dz l0 I Dl= 1-1 0
)L -l 03-l 0 23
We find &at Dr : 2, D2: I - 3p' and D3 : -l - 3p' so U3'331
yields
!
iu Dt 2 Dz D3
-1-]! -- lAl --]-3P
^-l.ll -r-p' ' -t.,rt - l-p' l-P

r r:elds when p I 1- When P :1, however' adding the last two of ttre original
which
equations impties that x * y = 3- Tbis conEadicts the first eguation'
becomes x * ! = l- So there is no solution in rhis casc-

trl
Homogeneous S4stems of Equations
we end this chaprcr by studying thc special case in which the right-band side of
From [2]. wc rhe system of equations [i3-i1] consists only of zeros' The sysrcm
is then called
homogeneous. A homogeneous sysem will always have the so-called trivid so-
lu6on x1 = x? : ' ' ' :-x^ = 0' In many problems' one is intercsted in loowing
t3_ when a homogeneous system has nonfivial solutions'
476 Chapter 13 / Determinants and Matix lnversion

Theorem 13.7 (Nontrivial Solutions of Homogeneous Systems) The

homogeneous linear system of equations with z equations and n unknowns
Prob
a11x1 * A12x2 *''- alrrn :
-F Q

a21x'1 * a72x2 *' " * aV"xn : Q I

ll3.34l

ar1xl * ao2x2 * "' * aonxn : S

has nonuivial solutions if agd only if the coefficient matrix A : (ai;),*n is

siDgular (that is, iff lAl : 0).
1

Partial Argumenr: Suppose that lAl # 0- Then' by Cramer's rule, x1' ' - ', ru are
given by t13.331. But the numerator in each ofthese fractions is 0, because each of
tbe determinants Dt, . . . , D^ contains a colunn consistitrg endrely of zeros. Then
the system only has the rivial solution. ln other words: System [13.34] can onlv
luve nontrivial solutions if thc determinant lAl vanishes. One can also Prove tbat
if lAl :0, then system tf3.34l does have nontrivial solutions'6

Example 1322 3.
Examine for which values of .tr the following system of equations has norr'
trivial solutions:
5x*2Y*z=l-x
2x* y :ly tl:
Y ! r 1-
-

Solution The variables x, y, and z appear on both sides of the equatioas

so we start by puaing the system into standard form:

(5-I)x* 2Y* ::0

k+(l-r)) :0 I::
r *(l-l)z:0
According to Theorem 13.7, system [2] has a nontrivial solutioD iff the c*
efficient matrix is singular:
5 _I ?l
2 1-] 0 :0
I 0 l-).
5Sc. Theorcm 145 of Secdon 14.3: If lAl = 0. then &e rank of A is less than z. and 1.g
ti33a) bas at least I dcgree of frccdom, so it has an infinie number of solutions'
 Sec. 19.8 / Cramels RuF- 477
The value of the determinant is ),(l
nonrivial solutions iff I : 0, l, or 6.
- fxl - 6). Hence, system tll has

Problems

l. Use Cramer's rule to solve the following two systems of equarions:

[13.34] a. x+zy- z:-5 b.;r*) =3
2x- y+ z: 6 x +z :2
x- y-32=-3 !*z*u=6
I ) *z:l
2. Use Theorem 13.7 to prove that the following sysrem of equations hes a
unique solution for all values of br, h, bs, and find the solution-
r.. . ... xn are
eaause each of
rf zeros. Then 3x1* x2 -br
3.311 can oilY xr- x2*24=[1
rlso prove tbat
2x1*3x2- xt:h
3. Prove that the homogeneous system of equations
nons has ne
ax*by*cz:0
bx*cy*az=0
cx+ay*bz:0
has a nonrivial solution if and only if a3 + b3 + I - 3abc = A.

i r.De equatroB!

uoon iff rh: ;---

:x: :. ao: }u