Beruflich Dokumente
Kultur Dokumente
ALGE BRA
modelsbasicallyplay the same rolein
economicsas in
fdeed,
fashion. Theyprovide an articulatedframe
on which to show off
your matenalto advantage,...
a useful role,but fraught
with the dangersthat the designermay get
carriedaiay by his
personalinclinattonfor the model,
while the customermay forget
that the model is more streamlined
than reality
-j. H. Drze(1984)
ost mathematicar
modelsusedby economists
urtimatery
involvea systemof severar
equa_
trons.lf theseequations
areail rinear,
the studyof ,*r-, ,vr,u^ o"a;
;"
an
area
of
ma th e ma tc
s i l e dri n e arargebra.E venj f
ca
the equati ons
arenonri near,
much
may
be
rearned
from linearapproximations
aroundthe sorutionwe are interested
in_for exampre,how the
solutionchangesin response
to smailshocksto the exogenous
variables
or parameters
rnput-outputanalysrs
rsone prominentareaof
ss
iIv'"".",',
:i,T".,]
.-:.
Ji;ffi:il};l',iif,1=,
:"i,l::: ffi l'^'1"':'-r
:donwa
American
Econom.
lgtg_1g3ghavesystems*itf,
f,rnJr.Ol";.;;;;:o.l.;.]1:.,
,-.,.
_"j:::1":::^li]:Tr:1,andsimiraronesdeveropedintheformersovietuo_.
-:_^-_.:-_
to ro v i c hw, e rei n te n d e d
to
pl an
e
-: -
--'
hel p
the producti on
of mi l i taryequrpmer::-t r_.-= ., :_.:
::
d u ri n gth e S e c o n W
d o rl dW ar(1g3g-1g45\
tn o rd e rto c o mp re h end
suchenormous
systems
of equati ons,
i s
a n u m b e ro f ma th e ma ti cal
concepi such
s
as matl ces,vectors,
a^.
rn tro d u c eidn th i s c h a p te rand i n the
next,w berew e de.no.s' .:.:
A c tu a l l yth, e u s e fu l n e so sf li nearai gebra
extends
far beyondt: c ..
e q u a ti o n sF. o ri n s ta n c ern
, the theoryof di fferent
al and c -:_:-:=
n o n l i n e aorp ti mi z a ti oth
n e ory,i n sta.l rstin0
cs eco,.orne:,
as -_: --:.usedextensively.
-_
15.1 Systems
o f Linear Equat ions
Section l.-i n ll.eadr inrrociuced
s\.stemsof two slmultaneous
linear equationsin trvo
larrabls. Hee is iln erantple
of such a s),stem.wherenow the
two unknownsare denoted
br . r . , t a . r - .
5s0
C H A PT E R1 5 /
M AT R IXA N D V E C TOR
A LGE B R A
,:,:,;::
(x
A solution of th.
ailknow*"-J,li'1':"'Tiffi:::il,*T*;1'J*"#:tisnesbothequarionsre
;:::::;:.;':i.rl;::*:Ti.Texpressionror.rlintoillilll#lf,i;l:ru
t' :
0' and then r = -) .0
"'ut
+ z - ,. ,ir\))n
{a,, *r :"i,'rl
A generalnotation ro,
a ,yrtem *rn ,ro linear
equationsin two unknowns
fl and r:? i\
anxt* crnxz_bt
solution is therefore
a2t+ azzxz-bz
Note that azt, wjl
2 and,,is rhecoerficienr
in rhesecond
equarion
murtiplying",,
Since we ofter
n^eedto consider a large
-'l;l;i:H:";Tbscript
number of equations
a suitable notatio] tor
and unknowns, we need
an arbitrary linear system
of equations with m
unknowns. w" l
equations and r;
to be larger than' equal
to, or iess thann.If
denotedby *t, .
the
unknowns are
xn,
,
we usually write such
a systemin the form
allxr + atzxz i...
l
c t 2 1 x t+ ( t z z x zI . . . *
alrxn _
121
a2rxn _ S2
(l
qmlxl * am2x2
I
... I a.nx, _
S_
of,hesvs,em,andb1,
l:'i;;":;;;;;;,::;,i'::i::i:::::{cien,s
, bnareca'ed
Notecarefulrt
or tne,uscrrpts.In-genera
1,a sthecoefficient
'h" :.g:.
in theth
;;;;
;H::;:lj:;
"'
#*l'*l;fl
fii:[t"il;JrH,.;,l
of,hese
e
"""ih;;;;ay be0_indeed
"*
(1)is anordered
setorlisrornumbers
rr, ,r2,. . ., s, rhatsarisfies
i:::'ffi
*il:Tl:;ill1;l"iJ:,i::r!".*#r*i;*;:.*H,H
sr),say,uiii
(0,2),,,, ;'r:il;jf,lot", ,",, x) : (2,0)i, arorutio" sysrem
(x),bur
"],n"
If sysrem(l) has ar I
it
nosolurion,
Whenthesystem
i, t, ,"ti" i3r,,"ffi.ri:ff.n, is saidrobeconsisrenr.
ha:
(xt,x):
I X A M P IE
3r+5xr-6
3- . - i*5 - r : : 7
(b)
t' i xz-2xs:
-xz* 2xz-_l
551
Solution:
(a) Thissystemisinconsistent.
Nomatterwhatvaluesvalueswechoosefor,rl
and x2,the
expression3xt I 5xz cannotbe both 6 andl. The systemhas no solutions.
(b) This is a systemin the form (1) with 2 equationsand 3 unknowns. The system is
consistentand has infinitely many solutions.Choosingan arbitrary value for
-x3,say
,v3: ,t, the secondequationyields x2 : 2s * I, while the first gives .r1 : l. Hence
for all value.sof s, the list (xt, xz,.r3) : (1, 2s * r, s) is a solution of the system.The
systemhas one degreeoffreedom, becauseone variable (s) can be changedfreely. (See
Section12.10.)
c
{ i- ; : 1,:nS . \ \ ;
;..
, g u J--,
L \ i ;--".
[ 1 , -_
-_
f{.-;iitrl-l 91r"3r
= - f i1 1 , r 1 . '
"i
.r.'- . i
:rd
fr
There are computer programs that make it easy to check whether a system like (1) is consistent,and if it is, to find possiblesolutions,evenif there are thousandsof equationsand
unknowns.Still, economistsneedto understandthe generaltheoryof suchequationsysrems
so that they can follow theoretical argumentsand conclusions related to linear models of
this kind.
3qui: 1 , :
I,!. ",1i
n-:
ltl, Il!
IiC
A S i mpl eL e o n t ie fM o d e l
lk::.'' n. :.
'
!t
--
. - : i:--1
:1,i:,:"
:-
1-.a::i
Solution: Let xr denotethe total number of tons of fish to be produced..r: ihe iotal
numberof tons of timber, and x3 the total numberof fishin-eboats.
Considerfirst the demandfor fish. Becausep.r2 tons of fish are neecledr(r prrrduce.rl
units of timber,and becausethe final demandfor fish is r1l. \\'e musrha\e,1.:: .i.rt - i..
(Producingfishingboatsdoesnot requirefish as an input. so thereis no remr s ith ,r.. r In the
caseof timber,a similar argumentshous that the equation,r. : './,r-i- 1-must be ,atisfied.
Finally,forboatbuildin-s.onl the fishineindustrl need,c
boars:thereis no hnal den.rand
in
this case,and so ,t3 : cy,rt.Thus.the lbllo$ ine threeequatrons
mustbe satisfreci:
, -p.r.-rl 1
'
',u,-11
fj
a,i-l
1.
( *l
552
CHAPTER 15 /
M ATRI X AND V E C T O R A L G E B R A
expressions
for rhe rwo other variabresare easily
found,
."Jlr";:l"sponding
dr'+ Flz
1-aFY ,
*, :
;tffi:lffT}:'ffil
producrion
in tr,"
x.
' -
qYdt * dz
l-a7y
r
-^ -
and the
od, -f aBct2
(*x l
(xt,xz,x3)
onty
makes
sense
when
aBy< I . rn fact,
if uBy> 1.
ll#:::anv
""o'olffil]
positivefinaldemands
ror fishandtimber-
P RO B LE M SF OR S EC T IO N 1 5 .1
..
r. . . . i : . . _ :
-, . . : : . . r. . -:
.: ::: :..:
..:::
.:.:i.:i
it
:!j
::
(a) 3x - y - z - u:50
@)J ix * 8 -ry - z * w: o
y )2 + 3 2 -w
: -3
a x l+ b q y 1 + c y la : g
o * ] + b x 2 y 2 + c y 2 r+ a : g
Is this a linear system of equations
in a, b, c, andd?
3 . W ri te s y s te m(1 )o u ti n ful l
w henn: m:4an!aj
: l foral l i
I j ,w hl e' :0
for I - 1,2,3' 4. sum the four
equationsto derive a simpre equation
for the ne*
variable s :
I:t x, thensolve the whole system.
t
xi l :j i l :T " " " -o f
equations( l) when
a j:i* jfor i, j:1,2,3andbj: j
T
SEC T I O N 1 5 , 2
553
(i i ) s :0.158(c -l x) - 34.30
(i v) x:93.53
(i i i ) y :c* l -s
and MatrixOperations
15.2 Matrices
A matrix is simply a rectangularanay of numbersconsideredas an entity. When there are
la rows and n columns in the array, we have an m-by-n matrix (written as m x n). We
usually denote a matrix with bold capital letters such as A, B, and so on. In general, an
m x n matrix is of the form
t:l:
(l)
^:(':_'_,:
::)
In this book any array like (l) will be enclosedin parentheses;other mathematicrtn>,r:
economistsmay prefer to use squarebrackets.
The matrix A in (1) is saidto haveorder m x n. The mn numbersthat ccnstitu:.\ :ri
called its elementsor entries. In particular,a; denotesthe elementin the l tir ii' ;: J :::
. ,r: ::r-rjth column. For brevity, fhe m x n matrix in (1) is often expresssls trr
simply as (a), if the order m x n is eitherobviousor ttnimportaut.
A matrix with only one row is also called a row vector. and a ntatnr '' -.:. :' . :.column is called a column vector. We refer to both types as vectors. Ii i: u>L-;--,- -::-.- lrow or columnvectorsby smallbold lettersratherthancapitalleter..
EXAMPLE 1
A :{l
/7 -:)
-)r
\)
6t
B- r - t.
t.
' .1.
l' .
c:l
i -1
I r
.l
ol
r 1)
EXAM PLE 2
C, ' r r Lr u; t t he - l
i nlatrix A :
l a j ) +. r w i t h a : 2 i
- j
554
c H A p rER 1 5 /
M AT R tx A N D vE C ToR A LGE B R A
So l u ti o n : M a tri x Ahas4 . :
3
l 2entri es.B ecause
aj : 2i _ j ,att :2.
_ I
a tz : 2 . 1 _ 2 : 0 , a ,n :
2. l _ 3 : _1, unro on.i i recompl etemai xi s 1_ I
(? r _t
2.t_2 2.t_3\
/t 0
2 .2 - 2 2 .2 - 3 ]l t;2
o : l : ? -l
u i_ "3 i_"
il_ :):l:
t
l)
sysrem
au x t I
az t x t I
anxz *
azzxz i
a m l xt * am2x2*
... t
... I
...i
alrxn _
b1
a2nxn :
b2
a_rxn -
S*
,1J",iTi?iH:xr*:HT:1Til:t"
n:l;kJi:?ltilffi
3 x t-2 x z* 6-13:
5xt I
"1i"""T:":*tr"Tent
iig'--*A#'dg#'ffi
xz* 24:
5
-)
is
- ?,
(;
the'sontheright-hand
sidebv anm xr matrix,orcorumn
vecror.
I
Considerachainofstoreswithfourourrers,81,
Bz,Bz,andBa,eachsellingeighrdifferent commodities' v1 v2
'
' ' ' ' , vs ' Let a denoteirr""il* varue of the salesof commodin.
J"lll*lr"xilrJ,.*"in
monrh.
suitable
wayorrecording
rhisd;;;;. the8 x i
atz
at3
a+
o:I"i'":'":'":I
/ot'
\ r;,
*il:Tr-'":"j
ror dre mtrnrh *
whereas
the4 columns
rererrorhe4 outlers.
For
!'.li:.l ::mmodities'
meansthat the salesof commodity
n"..u""*
7 at outlet 3 were;;i;a
SECTION 15.2
555
Matrix Operations
So far matrices have been regardedasjust rectangulararraysof numbersthat can be useful
for storing information. The real motivation for introducing matrices,however,is that there
are useful rules for manipulating them that correspond (to some extent) with the familiar
rules of ordinary algebra.
First, let us agreehow to define equality betweenmatrices.If A : (ai)*,, and B :
(.bj)*,, are two m x n mafrices,then A and B are said to be equal, and we write A : B,
j :l ,2,...,n.Thus,tw omatri cesA
p ro v i d e d th a ta :bj foral l i :1,2,...,m,and
s --lled
cn!titute
and B are equal if they have the sameorder and if all their correspondingentries are equal.
If A and B arenot equal, then we write A + B.
L,tl J il! 9 .
E X A M PL E s
w rr.n i r(..,3
-\2 t .' -' ) :(
u
\a* l
'
2u
)t
r+ u,f'
/
Solution: Both are 2 x 2 mafrtces with four elementseach. So equality requires the
four equations3 : /, t - 1 : 2u, 2t : u + l, and u : t I w tobe satisfied.By
solving these simultaneousequations,it follows that the two matrices are equal if and only
i ft :3 ,u
w :2.Thenbothmatri cesur..quutto l ]
\o
?)
f /
Let us return to Example 4, where the 8 x 4 matrix A representsthe dollar valuesof total
salesof the 8 commodities at the 4 outlets in a cerlain month. Supposethat the dollar values
of salesfor the next month are given by a corresponding8 x 4 matrix B : (b ) s, +. The total
salesrevenuesfrom each commodity in each of the outlets in the course of these 2 months
combinedwould then be given by a new 8 x 4 matrix C : (.;j)sr+, where cij : a.- Lt
fo r i : 1 , ...,8 a n d for j : 1,...,4. Matri x C i s cal l edthe " sum" of A andB . a nd, , r e
w ri te C :A+ 8 .
In general,if A : (a.)-rn and B : (bj)*rn,we definethe sum of A and B as the
m x n matix (.aI b)*r,. Thus,
i: I r'r
i. : 1
u.a! ! i
: I,u :5 ,and
A*B:
l bj ),,,,,
-:
It ;rff--:G::r -';ll"
uA :
l e!
t - ll
Thus, to multiply a matrix b- a scalar.multipll' each enfn in the matrix br that scalar.
Returning to the chain of stores.the matrix equationB : lA u'ould mean that all the
entriesin B are twice the correspondingelementsin A-that is. the salesrevenuefor each
commodity in each of the outlets has exacti;' doubled from one month to the next. (Of
course.this is a ratherunlitel\ event.) Equivalently,2A : A * A.
gI ]
f.:
E X A M PL E 6
{ - B.-r.{..
Compute
and(- j)n,oo:
(1 _: _?)
and * : (?
;)
556
C H A PT E Ri 5 /
Solution:
A +s :/1
\)
=(-! -e--i
'o: (,1-t -i) (-i)s
)
3 2\
_3 t/.
I ;'Ti':i;:'1J.":T#:lillli'lt,
the
difference
berween
rhe
rwo
marrice
.,.ft.
::ff,rll?*;t;:TiT
jlh*;jil,1#il:i,tst:*ffi
*x1
omponents
represent
increases
an
negativecomponents
representdecreases.
thedefinitionsgiven
it is easyto derivesomeusefur
*
"T1t",
rules.Let A, B, andc
* x z matrices'andlet
a
andBt"."Jnu-uers. Also,
-*#
matrix:*:onsistingonly of
ret0denotethem x t
zeros,called the zeromatrix.
Then:
R'U:1,E
5.$OR,.MAF
EtX,,ADDi TION,,AN,D
: M Ut_ltp Lte AT,'ON $y, SCA,,A.RS,,
(a )(A +B )+c:A+ ( B+ c)
(b) A +B =B *A
(c) A +o:A
(d)A +(_A ):o
(e) (a+f)A :aA + p t
(0 cv(A+ B) - cvA
+ aB
r:
'
1 5 .2
the matrix A :
,:f;*"t
(aj)2,2, where a =
2. Fo what values of
u and, d"",
3. Er.alrtrrgA - B and.iA
u,hen
(1 - u)2
uzu
(
6y
u2
o: (l l)
Tfor i -
r,2,3,
and a: 0 for
):1
l
_t /
\o
and n:(1
-3 u
u+5
-t \
1\"5
, l'
u
:;),
SECTION 15.3
4. EvaluareA+8, A
MATRIX MULTIPLICATION
o :(t l
u: (
and
;)
15.3 MatrixMultipfication
Ei
(i )
tt : attl t' l
l t : b 1 1 x 7 *b p x 2
012Y 2
+ anY 3
(ii)
yz : b21x1 I b22x2
ls : bttxt -f b32x2
e: ( o t t
ot2
'''\
, o, o'r, ;r: )
and
n:
/bt'
r:\
I ar, ur; I
\ .,, tt,, /
system (i) expresses
the variabresin termsof the y,s,
whereasin (ii), the ),s areexpressed
in terms of the 's' So the z variables
must be relatedto the variables.
Indeed,take the
expressionsfor y1, y2,and y3 in (ii)
and insert them into (i). rrris giv"s
F-
The matrix A
matrix C is 2
* ct1b21-1-a1-rD-31attbn
* ctnbzz I otzbzz\
! ctyb:l -i a-:-:1
ct21bp { azzbzz* orrUr,
)
B is i .' L Thus. B has as
many rows as A has columns.
The
t ri rr e let C : (c.i)trz, then
the number
r
: i/ .b:t * anbZt
I apb31
558
c H A p rER 1 s /
M AT R tx A N D vE croR A LGE B R A
(:?;)(;;):f;[?Sri;;r?
iti:):(:z
:,),
In order to generarizethe argument,assumethat as in (i) the variables
21,...,2n are e\pressedlinearly in terms of
]t, . .., yn, andthat as in (ii), the variables.)r, ..., , are
expressed
linearly in terms of x1, ..., xo. Then zr, ..., Zm carrbe expressed
linearly in
tetms of x1, . . - xp. The result we get leads directly to the following
definition:
'
MA,T,Rtfi.
1ry4,6d+ilr,Ptj:i.S:p,e
.l\|.
Supposethat A : (aj)*r, and that B : (bj)nrp. Then the product :
C
AB
is the m x p matrix c : (cj)**p, whose elementin the lth
row and the lth
column is the inner product
c j :
.A
L a b ,:
r|
atj * azbzj I...l ab
1...* a,br
r- |
l)11,:
T:)
SECTION 15 3 /
EX A M'IE2
L e t^ :(i
1) ,' ou:li
\+ -r
o/
MATRIX N4ULTIPLICATION
559
\- r
tJ
:(
^":(Hi)
)
ffi ,i)
(We haveindicatedhow the elementin the secondrow and first column of AB is found.
It is
th e i n n e rp ro d u c tothf esecondrow i nA andthefi rstcol umni nB ;thi 2.3+
si s 3.1+ 1.(-1) :
8.) The matrix productBA is not definedbecausethe numberof columnsin B (: 2) is not
equalto the numberof rows in A (: 3).
i
NOTE 1 In the previousexample,AB was definedbut BA was not. Even in casesin
which
AB and BA are both defined,they are usuallynot equal. SeeProblem I and the subsection
"Erors to Avoid" in Section 15.4. When we write AB, we say that we premultiply
B by
A, whereasin BA we postmultiply B by A.
EXAMPLE 3
In i ti a l l y ,th re e fl rm A
s ,B ,andC (numbered 1,2,anrl 3)sharethemarkettbrace r r ain
commodity. Firm A has 20Vaof the market, B has 60a/c,and.C has20Vo.In rhe courseot
the next year,the following changcsoccur:
A keeps85% of its customers,while losing 5Voto B ancll\Vc fo C
B keeps557oof its customers,while losing l0o/oto A and 35% to C
C keeps857oof its customers,while losing l07o to A and.5c/c
fo B
We c a n re p re s e n tmarketsharesofthethreefi rmsbymeansof
a mcrrket-l tre\' ((-tt)t . . Jer ni.
as a column vectors whosecomponentsareall nonne-eatir-e
and sun'ito i. Denne the rl,rrii:.
T and the initial market sharevectors bv
,.i
':(i]:)
':(:,l;;,:i
:,:i)
Notice that t is the percentage of l's cLrsttrmerstlho becrrn.ie's customers in
the nert
per iod. So T ir c alleJ t h c t t ', n t ; i t i , , t 7t ', , : , . i . , . .
Com put et her ec t or T s . s h o i r h a t t t i s 1 \ o a r n a r k e t s h a r e r e c t o r . a n d g i r e a n i n t e r p r e tat ion. W hat is r he int er p r e i a r i o no i T i T s r . T r T r T s l r . . . . 1
Solut t a. '
Ts:
/'r-< o lo
0.5,5
It1.5
\ 0.10 0.35
:(t,i)
::iX;:)
560
C H A PT E R1 5 /
vear-that
is,arter2ve,n,s,un
roon.(inprou",,,6,
youareasked
,::ffiffJ;.il'"
Systems
of Equations
in MatrixForm
The definition of matrix multiplication
was introducedin order to ailow
systemsof equations
to be manipulated' Indeed,
iiturns out that we can write linear
systems
of equationsvery
compactly by meansof matrix
l
multiplication. For instance,
the system
"onrlo",
3h I 4xz:5
7x1-2x2:)
NowdenneA :
(l
-)
":
o": (2
\7
(: , ), * o o :
(; ). * , . n we s e e ria r
So theoriginalsystemis equivalent
to thematrixequation
Ax_b
considerthegeneralrinearsystem(i)
in Section15.1.Suppose
we define
.:(i,)
,:(?.)
;"",o
tr m x n and,xis n x 1. The
marrix producr Ax is then defined
and,ismx 1. It follows
a ttx tl c tl z x 2 + ... + d1rxr, :
$1
c tz tx tl a z z x z I...*
A 2rX r:122
can be written as
.r.. _
,,. a-fr
_:
-..
4_ Ot,ttt-\.n
b,tt
Ax : 5
S E C TION15.4 /
:: . The firsl
561
P R OB L EMSF O R S EC T ]O N 1 5.3
1' compure the productsAB and BA, if possible,
for the fb'owing:
r -r ,:e Share
' ::. . -l ' s tot al
.:r::3tat10ns,
:::ri3l Share
, - .re asked
3)
\ +/
(d) A:
:)
(i_)
B :(
)
:: :luations
:::,lnsvefy
:
o :( ?) u :(-?_ ) .: (z D :(l
;)
i)
calculate(i)
3A+ 28 -2C +D
3Le, ^: ( : ;-i )
\, _r
(ii)AB (iii)C(AB).
* :(' ^ -:;)
tl
\;
;)
,:(t
' - \,
I
3
-2 )
xt I
x z:3
3 t * 5 z :5
(b)
@)2rt-3rz+r3:0
xt* xz--r3:0
/ 1 , \
(,
I Ji nrhesenserhatB A : . 1g,
6. In Example3, computeT(Ts).
Ir fbllows
15 .4 Ru lesfo r M at r ix M uf t pf ic at ion
The algebrarc
rr.rle
s in section1-<.1
concemin-g
matrixadditionandmultiplicationby a scalar
are ali natulnd ea:r Lr eni\.
multiplication
-\latrir
is a more complicatedoperationand
"
\\'emu:r
i\;-:,r:i
i.
hlr
rure:
appr1,.
we
havealreadynoticedthatthecommutative
--;;.-r
:--i_t-.r :: \oT hotd rn senerai.
The following threeimportanrrulesare generatTy
if i_ti
562
cHAPTER
15 /
a"#
rf;j,"d
c arematriceswhosedimensionsare
suchtharthe given operationsare
;ieiq+S*:if$rB;les$i
(AB)C : A(BC)
(associativelaw)
(l)
A(B + C) : AB + AC
(2)
(A + B)C : AC + BC
(right disrriburivelaw)
(3)
o:(
u:(3-)
" : ( ;i )
AB:(
)
BC:(-l-l)
( AB) c(:
) ( : i) : ( , i ; )
aru"r:
( )C: -!):(,1;)
u *":( 1l)
and
o':(; i)
A(B+c):
(; ) (: 3): (,1 ,)
on*oc:(3
)).(l i):('j )
:t(f
t(aB)cti,
k : \ j - r
:f
u,r,r,\:IA(BC)J
" ,0^),r,
,
",,(t
. k =1
/
l =,
s'here
riom
t . ";fr;Tt*:trLT;l?J:l,ii::
7runs
lfi '"-
oraIthepnrerms
abr,c:,
"
!'
S E C IION
563
(A + B)(A + B) : AA + AB + BA + BB
(*)
According to (2),
(A + B )(A + B ): (A + B )A + (A + B )B
By (3),onehas(A + B)A : AA * BA and(A ts)B:
+
AB BB, from
+
(x) follows.
which we seethat
Powersof Matrices
If A is a squarematrix, the associativelaw (1)
alows us to write AA as 42, and AAA as
.A3,and so on. In general,
(A is repeatedn times)
-l \
/l
f.rm
.fA"and
c.nnrm,l,J,l.,.,!#Til::;
il#:l.T::::::,j;;;T.'
L e rA
A 2:A A :/l
\o
-2\
t l'
-l
r - : '2--\
t :oll
n + : lr a : ll
t /)
- i)
,u
t/
A reasonabre
guess,therefore,is that fbr ar naturalnumbers
n,
L " :( l
\o
- l)
r x)
t)
{=/I
\o
-f r\
t)
-A \/t
-t\
1. Supposeitis correct
Then
/t
(.un
,ll;
l /\U
,l:f
l /- \0
/l
' -k-
1\
t/
564
CHAPTER
15 /
EXA*P'LE,3
MATRIXAND VECToRALGEBRA
Suppose
P andQ aren xn matricessuchthatpe :
e2p. provethat(pe)2 : e6p2.
sorution: The proofis simpleif we use
the associative
law (1) andpQ - Qzp repearedly:
The ldentityMatrix
The identity matrix of order n, denoted
by r, (or often just by I), is the n x n
matrixhaving
ones along the main diagonal and zeros
elsewhere:
/1
...
r'"-l:
,,:1 0 1
\o o ..
0l
.
(identitYmatrix)
, |
t/
(4 i
Errorsto Avoid
The rules of matrix algebra make
many argumentsvery easy,but one has
to be extremerr
careful to use only valid rules when
matrices are being multiplied. Look
(-) ;ilil,
at
the proof of ( 1), for instance.It is tempring
ro simprify tir" expr"srion AA aB
*
* BA + nn
on the right-hand side to aA
* 2aB + Bt. This is rrongt Even when aB
and
BA
are both
defined' AB is not necessarilyequal
to BA. Matrix multiplication is ol commutative.
EXAMPLE 4
sarLto-
0\
/0
u:(;
3)'
: )a n o n a : (0 \t
^ u:l: 1 .\ ut
j)
l\
;)
H. n " . , A B t B A .
0l --_..
S E C TION15.4 /
)F
-p-
555
ji3#J;f,1il:tril:gi:
::lH,
;IHi";#ffi:[::*";:,*']*":;ff
--
E X A M PL E 5
t"o:
( :)'":
Solution:
(-i -u')
"o,,ou,"o"
nu:(
)(_ j) :(3
3)
AB + 8.4
AB : 1 doesnot intply fhat
either
AB :49
(5)
A or B is ti
(6)
(7)
o o rT u s e sra w m a te ri al s
R t, R z,..., R * to produce
the commodi ti es
F o ri - 7 ' " ' ' n a n d j v1, vz, ... , v, .
r,...,n,w e,.retau" rn" l uun,i tyof
raw materi arR w hi chis
neededto produce each unit
of commodity v;. tt
quuntrti", form the mafix
"se
o,,
A _ (aijr*,,:
i
MI
at2
or., o:,
|r ^' om2;
'oml
dl n
"r: I
i
a*n /
ii?"llxin:ffi
;#;*:"k;
#*r#i*Tffi,""J,"JX3il"1T,"".fi
(x'
":l':'l
Since
a.in partie
ura.
566
CHAPTER 15 /
M ATRI X AN D V E C T O R
ALGEBRA
to producer1unitsof commodity
vr. Similaly aijuj is the amount
neededfor z7 unitsof
v (j :
' ' ' ' n)' Thetotalmonthly,"quir"*"nt of
raw materialR is therefore
''
qtutI ctzuz
* ...1anu, :f
o,ru,
j =1
wedennerheprice*,n
: (pt,p2,...,
r.,,fi!i:i:tT,"il"r#l,"";,r,1-,1"J,#
:lT:'#,TJiT"ffiX1',?"[:i"'L*,.-i"
bewritten
" 1, o,",il; ,;;;anarso
K : pr:
p(Au) _ pAu
how
manv
flighrs
rhere
*" u"t*""nrherwoairporrs.
.":,*"n:,?*"ffi:iTriT:iJ,';
B there are 4 flights to airport
cJ in country c-"*n
Figure1
The relevant data can also
b 1 b 2fubactc2ca
P:?,(3
i l"il e:r(I )
Each.elementpj of thematrix p
representsthe I
o\o
o/
b,whtle".h
;i;;;;|7r
ore,"p.e,",,,
,n"""l,1Tli;ff::',Tflr1r'Ti:rJT,:"?#,
SECTI O N 1 5 , 4
567
How many ways are there of getting from a to c using two flights, with one connection
i n c o u n tryB ? B e tw e e na 2 andca,forexampl e,thereare3. 2+ 0. 0 + 2. 4* I . 0 : 14
possibilities.This is the inner product of the secondrow vectorin P and the third column
vector in Q. The samereasoningappliesfbr eacha andc. So the total number of flight
connections between the different airpofis in countries A anciC is given bv the matrix
product
* :"a :(l :
0 2\
0l
lt
r \lt0
r /lr
(3 I
o + l:\s
:^)
t o/
\o
P RO B LE MSF O R S EC T IO N 1 5 ,4
)\
\.r
+/
A:(;;)
_l
I 0\
B:( 12
; _ ; ;;)
,./
\J
2.Computelhemarrix p ro
r* d. yu.c: rr
/,,
ld
\r.
c:( -/_t
l
/I.
\_z
r,\
t\
i)
/x\
/ lf
,/
I
0-
,l
\:/
(a)(A+B)(A-B)
+ Al -B:
(b) (A-BrrA -B
_\B- ts
: \: -
5. Lom put e: f a)
/t
I
|
{)
0
|
\0 o
? )(;
i:)
-l
--i
t
-]
-r
-1
(b) Show that if AB : -\ and B-\ : B. then A and B are both idempotent.
(c ) Show that if A is idempotenr.rhen.\', : A for all positiveintegersn.
CHAPTER 15 /
HARDERPROBLEMS
7' (a) Provethatif :
A
(:
uo)''n*
A2 : (a + d)t. - (ad
-
bc)t2.
(b) Use(a) to find
an exampleof amarrix
A such that A2= 0, bur
A * 0.
(c) Usepart (a) to
showthat,{3 _ r
= 0' (Hint:Multiplytheequality'a
part(a)byA,andusethe"ouurrr9.tlfties42
o : 0 to derivean equadon,
thenmultipryor o;;;;;H;l't
whichyou ,r,ir
;;;".;:JTy;iffit;f
::::lj:#:i:T:iJ",",ff
lff :T;fi,,,,f
ff J;
att
atz
al -
o _[i,,,
| :
;::
:
I,,:\
: | :,
\o,,
an2
( o,, a2'
a22
t,:lo!,
(,;,
,;, )
So we can write
otr,
an ,\
o* r l
tl
,_,)
l::il.".,,,1:iT}'",n,:ffi,,;i;Tiill,ii,ji,#:.:"
:"h:i
E X A M'LE
1 Ler^:
(l
u :(; -i I
sotution:
n,: (-
_i)
B,:
+\
(-i i)
I I i)
(b) (A +B ),- A ,
+8,
(c) (uA), :
qA,
(d) (aB),:
B'A'
sECTION 15 5 /
THE TRANsPOSE
569
EXAM PLE 2
SymmetricMatrices
Squarematriceswith the property that they are symmetric
about the main diagonal arecalled
symmetric. For example,
c:')(i
_:;)
(:,)
A : A,
rhisproves
rhat
XX',, ,r--",,,11;,:-:,:::J..lo-.t"t',,
: ; : .,..:: ,
P RO B L E M SF O R SE C T ION 1 5 .5
l\
( _;
: ;,1" :(i ,J
C : r 1.,i .0. -1 ) .
t0 lr
anou = -2. Compute
A', B', (A -l B)', (aA)',
(
, ' i'
\- 1
AB, (AB)'.B'A'. andA'B' Thenverifr.all therulesin (2)
for theseparlicularvalues
of A, B, anda.
2.LerA :{
l:).e:
) /
CHAPTER 15 /
M ATRI X AN D V E C T O R
ALGEBRA
3. Show thar A _
/3
3\
lz -r
\3
tfandB:
0/
4 8\
o t 3 f a re s y mme rric .
13 0/
/0
l+
\8
o'
4 . F o rw h a tva.r
rues
",, ( , i t
;'
?
"lio)
_r ,r ,o,n.*.
4a
\-:
show that
AiAiA\
(a )F o x:rt/O^
r
howr har ,:( :
3 _i) ir onnogonur .
I
o/
\o
(b ) Sh o w rh a trh eZ xzmanx
( P
\ 4
-q\
.
is orrhosonalif andonly il
P2 + q2 - 1
o)
15 .6 Gaussian
Efimination
This section explains a general
method for finding possibresolutions
to a linear system of
equations' The method involves
systematicelimination of the unknowns
from eachequation
in turn' Becauseit is very efficienl
it is the starting point for computer
programs
designed
to solve large systemsof linear
equations.
Consider first the following
example.
E X A M P LE . l j
xt-|
X 3:
xz* 34-
_7
2
-10
571
Solution: We start by interchanging the first two equations, which certainly will not
alter the set of solutions. We obtain
xtl
xzl 34:
' X 2- X 3:
2
-1
xtl xz*
3r3:
x3 : -7
2xz -
li)
5* z+ l l 4-* 4
The last two equationsnow haveonly xz andh asunknowns. The next stepin the
systematic
procedureis to divide the secondequationin (i) by 2, so that the coefflcientofr
becomes1.
Thus,
xtl xzl
342
x2-
ir r : - l
5* z+ 114--4
_,,t
x-2 -
3x2-
I rt : -l
,1..
27
Tt3:
'sr stemof
equation
s designed
13:
2
( ii )
:- :
I
).
our eliminationprocedureled to a "staircase"in system(ii), with x1 x2, &nd as
leading
,
4
entries.In matrix notation.u'e have
I
3\/.,
_+f \ / 2\
I
lo
l. , l: l_ll
\0 0
t/\*r/
\ t/
lt
Thematriroi coelicients
ontheleft-handsideis uppertriangularbecause
all entriesbelow
theniaindi"s,.nlare0. \,Ioreover,
thediagonalelementsareal7l.
572
CHAPTER 15 /
M ATRI X A N D V E C T O R
ALGEBRA
orequations
inorder
toarrive
arsysrem
(ii)ae
ffi::Tffi*l#il::jilJ,"stem
t"i::"T::.8::ll*"
follows:
theelimination
procedure
unrilwealsoobrain
zerosaboverhe
:mentaryoperationsrequired
to do this are frequently represented
as
xt* xz* 3x3_
2 < __
*r-l*.:_tr _1
ff"ffi:.T1?:,l"fr"T"J,are
incruded;,;';", were
adding
-r rimes
thesecond
xl
+ lu :
'r-i*.:-l 1
'3: 1 + _4
At the r
ffl;T:;;3,"1;#l#:J*_
fl*?Tf
:il#li'l:#ffi"J;3,il':J,iTffi
x1
:2
x2
--3
N 3:
example.
xz*
x3:
4:
2x1-4x2t44-
3h+ 4xz
: ll
x3:
-2
-2
2 x t * x z r x 3 - 7 < -J
I
I
2 x t -4 x z * 4 4 I
3xt + 4xz
: 11 <-----____l
The result is
xt l 3xz -
x3:
-5*z + 3x3 : -1
-10x2 164:
_)
-5-t:+ 3x3:-1
-l
5 73
.._
j3- 3- 5jJ:
+
l
5
-10z*64:-2
xt I 3xz
105
< | I
x2
and
- i : :
?11
+----]
+ -3
0:0
0:0
t5-f.t:1
ta
z- ix::*
From this systemwe concludethat the solutions
ofthe original systemcan be expressedas
follows:
j-.411
l :--.d3+l
(*)
x2: 3r,+ I
+ !,
| t+ i , r)
(/ i s anyrealnumber)
(l )
C H A PT E R1 5 /
Bef,oreshor .
reducestr,"umlllsr;i:.T,,"#TJ"."::fl
:_1;:;:?THTffiT:?ffi
:3:iHl:
that we only needto know the
coefficienT o,{thesystemor"quuiion,
l;l?ffi,:11:,.T:lt"
2 canberepresented
bvaugmented
matrices
(each
wirhanexa
-z
<-ril
_J
<I
(:
-r
4\
/t
ro r
9 -ro
I -3/s_zl*_L
lo
-f o
'/'l
e
f
I -f o'
\o -5 3 -r I *___1
\; o
i'{)='
(i; { 'f)
^-,
the
last matrir
representsthe system xt
* txz : !, *, _
i", : *.
For what values of the numbers
a, b, andc does the forl0wing system
of equationsha'e
solutions? Find the solutions
when they exist.
xt-2xz*
:+ 2x4:6
xtl
xzl
xz-
xtl Jxz-5r:_
x:b
x1:c
solution: we representthe
systemof equationsby its augmented
matrix, then perform
elementaryrow operations
as required by the Gaussianmethod:
-i _:_i
(i i -j ; )-' -r'_(
-2
t-s
-r,/
-\
t\
_u_3
: -:)
-(j; z i,",-;;.,)
-'
575
-tl t
a/
\ta + 2b;
I 0
-2 / t -t / t
\o
lt t - " t I
/
t' l
- t , + t ra : \ (a + 2 b )
x z -3 x z - j" o : \ f u -" ' ;
x1:\h2b)+lr-1 . ,
(s andt arbitrary
realnumbers
' 2a - 3b* c : 0)
x2 : !b - o + ]s + Jr
P RO BL E M SF O R S EC T IO N 1 5 .6
1. Solve the following systemsby Gaussianelimination
.
( a)
xI *
t:3
3xt+5xz-5
(b)
x\-
x2+ 13-5
2xtl 3xz
-,r3:
,f-
i -1.,
1l
'ni
=.
576
C H A PT E R.i 5 /
!* (a*
(a) x t-2y *
3 x* 4yj _
r ) z: bt
z:bz
7z: bz
1, + y r Jz: b1
G) xl2y*
z:bz
3x i4y *72: bz
15.7 Vectors
Recallthata matrix with only
onerow is arsocalreda row vector
, and,amatrixwith only
onecorumnis calreda corumn
vector. we refer to both,r0", u, o""rors.
As remarkedin
arervpicarlvdenoredbv sma' bold
retters.Thus,if a is a I x row
iJi[], * ;lJctors
a:(a1,a2,...,an)
Here' the numbersat, a2, .
' ., an afecalledthe components (or coordinates)
of the vector.
and a is its i th component or
i th coordinate. (Recail that when
we consider a as a matrix,
ar' ' ' ', (tn are calred entries
or erements.)If we want a
that a vector has
to it asan
"-fn"rrrl
;Ti:lffj:rlHar
n_vector.Alrernarively,
if a is ann_yector,rhen
we sayrhat
i..T,,f,:."Ti:::#:
:lTJtr:*'j:*,:*LtnHi:l:ffi:il,Hffi
dimensionarspu."m:.enoanvn-v"ctori;;"n*H,1#t*":Jf#'*H"':ff;iii:::h:;
Operations
on Vectors
since vectors are just special
types of matrices, the algebraic
operations introduced for
matrices are equally valid for
vectors. So:
(A) Two n -vectors a and
b are equar if and onry if a, their
correspondingcomponents
equal:we lhen write fl : b.
are
(B) If a and b are two _vectors,
their sum, denotedby a + b, is
the n_vectorobtained by
adding each component of
a to the correspondingcomponent
of b.r
(C) Ifaisan-vectorandisarearnumber,*"o"on"
taasthen-vectorwhosecomponents
are / dmes the corresponding
componentsin a.
(D) Thedifferencebetweentwon_vectors
a andbis definedas a _ b: a (-1 )b .
+
.,,"r, sum
sumrs
issrmply
simprv
nordefined,
not
difference. No should one add
denned,
noris
nor
istheir
rheir
a row vecror
yvLLUrto;;;j;,
LUa coruln vector, even if they
-':il:",J;::?:##jH*,::::jiT"::l:"'their
have the samenumber
of elements.
SECTION 15.7
V E C T OR S
577
Ifa and b are two n-vectors and t and s are real numbers,the -vector /a + sb is said to be
a linear combination of a and b. In symbols, using column vectors,
\')).,(':):(:'
I:u.ns t..
Here is an interpretation: Supposea and b are commodity vectors, whose jth components
arequantitiesof commodity number j . Now, if personsall buy the samecommodity vector
a and s persons all buy commodity vector b, then the vector /a * sb representsthe total
commodity vector bought by all / + ^tpersonscombined.
n:th onlr
c"ked in
[ ' l tort
r r,airir.
r,:: has n
i:e .r
that
Of course,the rules for matrix addition and multiplication by scalarsin (5) in Section 15.2
apply to vectors also.
The InnerProduct
Let us consider4 different commodities-say, apples,bananas,cherriesand dates. Suppose
youbuythecommodityvectorx: (5, 3,6,7). Thismeans,ofcourse,thatyoubuy5unitssay,kilos-of the first commodity, 3 kilos of the secondcommodity, etc. Supposethe prices
per kilo of these four different commodities are given by the price vector p : (4, 5, 3, 8),
meaning that the price per kilo of the first good is 4, the price per kilo of the secondgood
is 5, etc. Then the total value of the commodity vector you buy is
4.5+5.3+3.6+8.7:109
in er.-tlr
Ee:nentof
r:pe"'iaih
t-:nr in -1l,-e
-r-".
The resultof this operationon the two vectorsp andx is often written asp.x andis calledthe
innerproduct or scalarproduct ot dot product of p and x. In general,we havethe tblio$ ine
definition (formulated for row vectors):
d:-;ed ibr
a' b : atl
}enS are
rued
r lr
t1
azbz1... * a,,b, :l
lJ
;-t
ab
t:rrnenIS
.1 b.
l*-: i. their
re number
Note that the inner (scalar)productof two vectorsis not a vectorbut a nutnber (or scalar).
It is obtainedby simplymultiplyingall pairs(a, b), j : 1,2,. . . . 2,of the corresponding
componentsin the two vectorsa and b, and then finally addingthe results.Note that a ' b
is defned onlt i.f a and b areboth of the samedimension.
In the casen.henp is a price vectorwhosecomponentsaremeasuredin dollarsper kilo,
and x is a commoditl' r ector whosecomponentsare measuredin kilos, then eachproduct
p,;,r.,is an amLruntof money measuredin dollars,as is the inner productp .x : Li:t pr.
578
C H A PT E R1 5 /
iliqtr$ffi:&#,]i
If a : (1 , -2 ,3 ) andb : (_3,2,5),
computea .b.
S o l u ti o n : We gera. b : 1 . (_3)
+ (_2).2 + 2.5 : g.
vector.
rhenp'x is rheI x 1marrix,
u ,"ui* thatis equar
to
;iT#X1,t:L:ii::t"
b .a
(b ) a .(b + c ):a .b * a .c
(2)
(c ) (u a ).b : a . (a b ) :
cv(a.b)
(d ) a .a > 0
1. Computea-r-b.a
_b.2a*3b.and_5a*
2bwhena: (
\ -l
?) unn: (;)
/
S E C TION15 7 / V E C TOR s
2 ' L e ta : (r, 2 ,2 ),b:
a*b*c,
5 7g
_a_b_c
componentsof x?
componentsofx?
_ a forxi n terms
of a anclb.
a .a + a .b :a .(a fb ).
8. For what valuesof x is the inner
productof (.v,x _ l, 3) and (x,
r, 3l) equalto 0?
9' A construction company has
an order for severalhouses
of three diff.erenttypes: 5 of
type A'7 0f type
and
12 0f type c. write down a
-B'
3-dimensionarvector x ri,hose
coordinatesgive the number of
housesof eachtype. Suppose
that eachhouseof tr p:
A requires20 units of timber, type
B requires1g units, and type
c requiresr-i ur:::.
Write down a vector u that gives
the differ
ofeach
ofthethee
different
rypes
A,,,
compudng
. ..
rvlqr Lrtrl
":TriJHJ,'iff;HI.:.IT,:,__
tt'ilr"J,J:lt thefirm'sprofitis
br thesctarFr(riuJ:
p \ \\':": iip i is
-siren
s80
CHAPTER 15 /
M ATRI X A N D V E C T O R
ALGEBRA
5.8 Geometricfnterpretation
of Vectors
Vectors,in contrast to general
matrices
word
"vector"
isoriginarv
ru,*u,i,uffi','"t;l|T#,T1jtr[:Tly;*,l"l ,li
particular, the word is rerated
to the act of moving a person
or object from one prace to
ffi'ffi*
Hiffjjlogist
islikelvtothinkofavecror
asacarierofdisease,
such
as
same
shift,because
the anly *-pon"ntsaresriilu1 and
ilTJil,Tfi:sent
Figure I
"tffl"l1J; ; :'
@t, q) : @t * at, p2 -r uz )
Figure2
on the otherhand' ifthepair
(at,az) is given, the corresponding
shiftis obtained br
moving dl units in the direction
oi rn"
u, well as a2 unrtsin the direction
of the
"-uiir,
-r'-anis'If we start at the point p : (pt , p2),
thenwe arrive at the point
withcoordinares
tq\.q:l : (pt L at. p: *
Q
:). alsoshowni Fie. 2.
5 81
VectorOperations
Ifwe representvectorsby directed rine segments,
the vector operationsa * b, a _ b, and a
can be given interestinggeometricinterpretations.
Let a : (at, az) and b : (b1, b) both
start at the origin (0, 0) of the coordinate system.
* by, a2 122)
o
Figure3
Figure4
,/^
u _ a,/
,/
//
,a,'
Figur e5
,/
./
b
''
b +( a - b ) : a
Figure 6
582
c H A p rER 1 s /
M AT R tx AN D vE C ToR A LGE B R A
3-Spaceand n-Space
The plane is often also called 2-spaceand denoted
R2. we representa point or a vector in a
plane by a pair of real numbersusing two mutually
orthogonal coordinaterines.
In a similar way, any point or vector in 3-spaceR3 can
be representedby a triple of rear
numbersusing threemutually orthogonalcoordinate
lines, asexplainedin Section I 1.3. Anr.
3-vector(at, az, a3) canbe consideredin an obvious
way asa geometric
in 3-spaceR3' As with orderedpairs in the plane, there
""";";;;;;";;;
is a natural correspondence
between
ordered triples (at , az, az) and geometric vectors regarded
as directed line segments.The
parallelogram law of addition remains valid in
R3, as does the geometric interpretation of
the multiplication of a vector by a scalar.
The setRn of all n -vectorswas introducedin Section
11.5.when > 4, it hasno natural
spatial interpretation. Nevertheless,geometric language
is sometimesstill used to discuss
properties of R', becausemany properties of R2
and R3 carry over to R/,. In particular, the
rules for addition, subtraction,and scalarmultiplication
of vectorsremain exactly the same.
Lengthsof Vectors
andthe cauchy-schwarz
Inequarity
If a : (r, az, . ' . , a,), we definethe length (or norm)
of the vectora, denotedby llall, as
:
llall .fi'a. or
llall:
" l+ a l+ . . .
(1r
(Cauchy-Schwarz
inequatity)
et
SECI I O N 1 5 . 8
G E O I V E T R I C I N T E R P R E T A T I O NO F V E C T O R S
Orthogonal i ty
ConsiderFig. 7, which exhibitsthreevectors,a, b, and a _ b in R2 or R3.
t:
o.,r/t"
Figure7
According to Pythagoras'theorem,the angle d betweenthe two vectors
a and b is a risht
a n g l e(= 9 0 " ) i f a n do n l y i f (Oi l 2 + @ D 2 : (A B )2,or
:
l l al l 2+ l l bl l 2 l l a _ nl l 2.rtr is
implies that 0 :90' if and only if
a .a * b .
b : (a -b).
(a-b)
: a. a_a.b
_b.a+ b.b
(*)
Because
a ' b : b ' a , equal i ty(x) requi res2a.b :0, andso a.b : 0. w hen the an gle
betweentwo vectorsa and b is 90', they are said to be orthogonal, and we
write a r b.
Thus, we haveprovedthat two vectorsin R2 or R3 are orthogonalif and only if
their inner
p ro d u c ti s 0 . In s y m b o l s:
aIb
++
a.b:0
C O50:
r, -.
a.b
,i -
llai u
"
;-'\-..
' /- ''
I-
'\-.i
/ .,
tI l:l
s84
CHAPTER 15 /
M ATRI X A N D V E C T O R A L G E B R A
In statistics, cos6, defined by (4), is called the correlation coefficient, often denoted by p. It is a
measure of the degree of "correlation" between the data. When p : 1, there is a positive constant
a > 0suchthatd - d:a(p
-l),implyingthatdemandandprice areperfectlycorrelated.
ltis
more plausible, however, that p : - 1 becausethis relationship holds for some a < 0. Generally, if
p > 0 the variables arc positively correlated, if p < 0 the variables arenegatively correlated, andif
p : 0 they are uncorrelated.
EXAMPLE
(Orthogonalityineconometrics)Intheearlierexampleoflinearregression(Example
the regressioncoefficients a and fl were chosen to minimize the loss function
1T
L(u,B) = I J-r?
T-"
13.3.4),
: *r-,o'-u- nx')2
This required choosing : Fy (o,y /o*,) lt, and.p : 6"! /o,,,where r, and i., denotethe means
ofxr and respectively, whereas o"" is the variance ofxr, and o' is the covariance ofx, with yr.
Ther es ult inger r or s b e c o m e ' : l t - & - f l *, : j t - & y - ( o , r f o r , ) ( x t - & r ) . B y d e f i n i t i o n o f
t, and p,n, one has
IT
(* J
ID ' :O
In addition,
rT
a \:
/- ,",
T
it Ir,_,
,r ,
^
- tr xLr yo
/t
\t
- r ;) : o*,- ffo^, :o
Ii
(**)
(1, 1...,1),x:
Def inet hev ec t or s l:
( x 1 , . . . , x r ) , a n d : ( , , , . . . , 2 7 ) . T h e n e q u a t i o n ( x)
( ;I",' equation (xx) shows that the inner product
shows that the inner product of 6 and I is 0.- Moreover,
of andx is 0.
Note that L(a, il : Q/DllV - ul - Bxll2. Geometrically,the scalars and,p arechosenso
that the point i : Al + px in the hyperplane2containing the points 1 and x is as close as possible
to y in the Z-dimensional space R7. This involves having the vector y - i : be orthogonal to 1
and x, and to every other vector al * Bx in this hyperplane. Accordingly, i is catled the orthogonal
projection ofy onto this hyperplane.
=
p*O.gl:fii$rQ,n, e:1j+x'ls
and illustratewith
SECTION
3 . L e ta : (1 , 2 ,2 ),b:
(0,0, _3), andc:
and verify that (2) holds for a and b.
585
15 .9 L ine sa nd planes
Let a :
'f:l
a'
.r)
z.t-'
ta"'_,
F i g u r e1
Let f be a real nuntber and pui \ : a - .- b _
a, : r 1 _ )a f Ib. Then / : 0 gives x :
a
and / : I gives x : b. In
sen:r;i. br the seometric rure for adding vectors, we see that as
/ runs through arr the rer r:ul.nber:.o x clescribes
the whole straight line t.
s86
CHAPTER 15 /
M ATRI X AND V E C T O R A L G E B R A
(1)
for somereal numberr
(2)
3t
x3:(1 -t).2+t.4
:2+2t
Thislineintersects
thexlx2-plane
when/3 :0. Then2*2t :0, so / _ _1,
implying
that11 - 3 andxz: 5. It folrows
thattheline meets
-- the
-"- xlx2-pTaneat
t
'v \'z v'a"w aL the
point (3, 5, 0).
as shown inFig.
2.
Figure3
Suppose
p : (p t,..., p ) e R " . The
strai ghtl i ne Z passi ngthrough(pt,
..., p,) i nfhe
samedirection as the vector a : (at,
. . . , a) isgiven by
x : p + ta
."Tt
}jt
( 3r
f a, D is p _
ja.
sECTION 15.g /
587
Hyperplanes
Considerlirst a plane.p
in Rl passingrhrough
r
0," i"""?-;;:;:;;"'"Jffi;
i,;^:l;::';*J?''
'$;:;:tlli-;,T,;Jff:;
rrrus,irx:r,,,i,,l.iT,'ffiTfl
Jr::f.".",X1":'i.,T."Jf
f:1:lJ,Tyffi
X1;3;il;
orthogonalto p. Therefore,
the innerproau.t oip
p .(x -
a ):0
or
and x _ a must be 0, so
that
f+
Figure4
So (4) is the general
planein R3 passingthrough
"*:,]::r":"
a : (at. 02. ej).\ote t h a r
(pt' pz'
pi of xt,x2'x3represenr
a nonzerovecrorthar rs
ili#"tott"nts
nor.'ai r, ri_E X A M P LE 2
f<
-,,rrrthp =
normar
Doesihe'"",l';1lift"il"[1JJJ:tr1J;.!r'
Solution: Using(4),the
equation
is
-l .(.rr -2)+1.(z_
1)+3(..--._r_lr_rl
If.the-linegivenby thethree
equations
,1. : I
thisplane,thenwe must
ha'e
-
St-rlrn
r g th i sc q u ri ,,l r,,.
r
_::
r tl q L \
::- Lri nt
of inerseciion
is givenby
:-1.//\
following general
588
cH Ap rE R .ts /
M AT R tx AN D vE C ToR A LGE B R A
.JH,YF,tR'.PA$,,8*N:.ltl*:.F:i.
T h e h y p e rp l a n e
H i n R u througha:
(a1,...,a)w hi chi s
orthogonal
-4,;. tothe
p - @t'"', p,) isthesetof an
poinrs
:
.., x,)
;11-r?,;'""tor
"
(s)
:
:
P'(x-a):O
l
Note that if the normal
precisery
sf 0,then
rhe
same,", i'Ji1ff3r;1ffjiffi#ru;l where
"r"i!lj
using the coordinaterepresentation
ofthe
vectors,the hypelplane has
the equation
p t@ t - a) + pz@ z- a)
+ ... I p,(xn _ an) :0
(O)
pnxn < m
(7)
when (7) is satisfied with
equality, it describes thebudget
(hyper) prane whosenormar
the price vector(p1, pz,
is
. . . , pr).
Usually,it is implicitly assumed
thatx1 > 0, x2 > 0, . . ., xn>
0. See Fig. 11.4.rfbrthe
c&setr : 3. Note that in
this figure the vector(,p,-q rl
, i, nor_ to the plane.
:
P RO E LE M SF OR SE C T ION
I5 .9
and(10,2, I).
(b) thatpassesthroughpoint
(1,3,2)andhas
2 . T h e l i n e Z i s g i v e n b y : _t
xl
* 2,x2:2t
_1, andx3 :t
* 3.
(a) Verify that the poinr :
a
(2, _1,3; lies on L,butthat(l,
l, 1) doesnor.
(b) Find the equation
for the plane through a that is
orthogonal to Z.
(c) Find the point where
L intersectsthe plane 3xt
+ 5xz _ x3 : 6.
l).
R E V IE WP R OB LE MS
FOR C H A P TER15
589
t 2!*
3z : l .
pafi (a).
t ) . then
R E VIE W P R OB L EMSF O R
. -:
- --,,tr1t--(6 r
C H A P TE R 15
t/
^:( \-r? 9)
calculate(wherepossible),
:::ae ) p er
(a )A -n
:. :': \ :
2\
/) 1\
\ r -t ) .-: ( i) ": (l j .l )
B:/-r
(b)A + B _2C
(c)A B
(d)C (A B )
(e)A D
(f)DC
(a) 2A.- 38
:-:ralis
(b) (A _ B),
(c) (C,A)B,
(f)D,D
. rr the
,
(a.l
2 x t-5 x z-3
(b)
5 l * 8z :5
t:u
-y* zl
x* 3y+ 22i _4r-t
5 .L e rA :(
(c)
,t * 4v -i- 8z
+
-'\
-2\
j t5l
\_6::51
B:f
f :d
J-
axi
_5
5
\_+
I.
- - A.
3x-Jr-
7 - - l ,-
j\
2_; )
2 ol
/()
c:(
- ,
r:f .r i iq f .s:
p
q/
(a ) Pro rerh a rT S:
S. Tr = +T* 4S.andf3: jf
+ lS.
tb
Cr tn ,j3 .- . .:- :
I
:_
-- r -
t;
\0
_l
o
-5
;)
0\
la,z
\o
+I ) ::b t
x *2 r {
Findrhemarrices
A + B, A _ B. AB. BA..qfnt.
."] 'ou,a.
6. Define the two marricesT and
S br
). I(a
( ,{#)
590
CHAPTER 15 /
M ATRI X A N D V E C T O R
ALGEBRA
(t
b o\
|
-7 ' (a )
L e L a ' : I -t
uI.w here aandbarearbi rraryconsranrs.Fi ndA A: 4u.
i
\ o -b
"/
(b) A squarematrix is
called skew-symmetricif B - _8,,
where B, is the transpose
of B' show that if.Bc is an arbitrary
matrix such that c,BC is defined,
then c,BC is
skew-symmetricif B is. when
is ihe matrix a defined in (a)
skew-symmeffic?
(c) If A is any square
matrix, prove that a] :
1(A * a,) is symmetnc and that
a2 : j(A - A,) is sk"w-sy-m"r.i".
V"rir, that'i : A, + Ar,"'unaexplain
in your
own words what you have proved.
8. solve the following equation
systemsby Gaussianelimination.
(a )
xt I 4xz:
2 q * 2 x z :8
xt-3xz* x3:Q
(c)
xt* 3xz!_44:Q
5xt*
3xt+ 4xz-J3:1
xz*
-:vj :Q
ay l 2z
:0
-2x-
al l
z:4
2ax* 3a2y* 92:4
prices(p1,0,,,:i:JTn:1ffi
j,:,TJ:;i,Aff
::i:'"T:,1'""J*
ilill,";ff
prices if both plants are to
;il::
remain
in use?
12.
'ii:lliiii:i.iii,iir
..'.
ep2 :2p2
and