ece sso | LecTURE 3
We are now prepared te discuss the important
subject of eigenvalues and eigenvectors, which
will lead te one of the canonical Forma ive- the
Terdan form
Let A be a linear eperater that maps Cc%\c)
inte itself. Then a scolar Ain C is called an
eigenvalue of A if there exists anen-tere vector
1
wh C™ such that Ax 22x. Any nen-tere vector
x sobre h ying Ax AX a on eigenvector of
A associated with the ergenvaloe A. In order to
£
vad an eigenvalue of A, we need to solve.
CA-alyx =O OO
where Tie an Ndentiby, matin of order n. We
know thet @ has a non-trivial solution iff
det CA-AT)=0. T+ follows that Ala) &
det CA=AZ)=0, where ACAI Ga polynomial
of degree mith A and sd called the charactenishe
Belunomial of A, The roots of ACK) comespend
to the eigenvalues of A. Since Ala) 5 of ender
ny the mate A has n cigenvaloes Cnet
necessarily. all distinct DeEck §50 |
LecTuRE 3 rs
Lets fet consider the case where allof the
engenveloes of Aare distnet. Theos, let %,,%25
Aq be the distinct eigenvalues of A and let
vi be on ewgenvector of A associated with Ac,
vi) Then the set
for bthy2,
te
qualifies as a basis. Fromearlier, we recall
ny hes Ay
nae” Die Uncorly independent and
that A ve the representation of A with respect
fo anew bars, If we cheese the exgenvecters os
Glee bere een ee eeclomn e A vs the!
representation of Ay with reapect to
Evtiyt yy} sce Lovonzor 0), where 2,
stands at the ithcamponent. Thos,
Oo Or: ay -®
Note Asia diagonal matrix whose diagonal ele-
ments are the n distinct eigenvalues of &
We can alse ebtain ® bya similarity trans fenmLEcTORE 3
Thes,
A= &aQ 3)
If we now consider the case where the eigenvalues
of Aare not all distinct, then we must introduce.
the generalised eigenvector. A vector y ue said
te bea geveralined engenvector of rank k assocs-
ated with » fF
Cae eo
and
Garand y #0 @®
Note that if kal wehave thet CA-AT)Y =o
and V#O , which ia the definition ofan
eigenvector. Hence, the term “generalined
eigenvector "& appropriatee
ECE 550 | i LECTURE 3
Let y bea generolized eigenvector of rank k
associated with the ergenvalue 2, we define
the following
|]
¥
y"® Ca-at)y 2 Canandy®
= 2 Pa
fee Cao aay CA xT)
Pa
vi & Caran) ’y = Cara) y”
.®
Then foreach Lim 12 ik , v's a generalized
eigenvector of rank 2. The set of generalined
poy ve defined 4® obeve
1 linearly Independent. Furthermore, the
eigenvectors v!
generclined eigenvectors of A asssciated with
different eigenvalues are linearly independent
Let us consider a numerical example
Ge
Aso 3)
eo 2
det (Ana d= Cant)*Ca-2) 20
4ECE 550 | Lecture 3
Thon, the eigenvalues are a,
Finst consider Ay,
chon
on 3
eo 0
Se thet
[3]
Now consider >, ,
eae
eo 8
oot
Note thet the renk of CA-%,D 018 2, andte
we can only Find one ergenveeter associated
with A,. Thos, we must use. grreraliaed exgen-
vectors. Let y bea non-acre vector such thet
ce
Ca-aty = foo3 lyte
Bas oo| LECTURE 3
Ios cosy te see that
cs [ ° ]
1
°
13 thot vector. 14 tba generalined eigenvector
of rank 2. Thos
secon (S23 ]2) [8]
We now have our base je. EvivSy*}. Hence,
be coe deleiae AO bee stan| Eck 550 | LEcTORE 3
ee et A ed tobe 7 the oe
canonical form. We can also check this resul+
veing the similarity transform
fea
—
oo
coe
Poo
oo-
o-0
con
(es
it
ou
eo
Or, te averd taking @ mater inverse, simply evaloate
gangform, if te principal diagonal consists of Jomlan
hing elements are tron, "The mats A in 2-3)
fh Hnee Jordon blocks awacated with by and one Jordan bloe ss
fiat with Xs A dagonal mata sles a special ease of the Jordan
Form: al of ts Jontan bloeks are of dor
We dhaim that every liner tratsformation whieh maps (
rm repetition by «proper ehoice of basis. ‘The
ite has Jordan
bss vectors xonerilyeomirt of eigenvectors an
ertors Before giving 6 eneral procedure for computing the required
bss vectors, we need he following haze
Letw and v be the generalized eigenvetor of rank and i, respectively
sscciaed with the tne eigenvalue A. Define ut = (A — AD? for
rr rand v= (A— MD), forj = 1,2, 7° yk Ut the
foro veel wand wwe Iineely independent, then the generalized
Seenvectors a Vico are linearly independent Bl
The set [ow 1] fa liner independent st, x0 Ue st
Inv =, ¥) Gee Theorem 2-10). Hence what islet to be proved
That each wis linearly independent om the st fe, w= => and
teach vis linearly independent on the set [w= =, ut. "This ean
be proved by the arguments used inthe proof of Theorees 2-11; the proof
[A procedure for computing a Jordan-form representation. We give in
the folowing a procedure for computing wae of asx vectors for alinene
operator A that maps (2, C) int elf such thatthe ness representation
Cf Aisin the Jordan eanoneal
A. Compute the cgenvalues of A by solving det (A— AD) = 0
Let ns te the dstnet eigenvalues of A with mult
plies, reepective
2. Compu linetty independent generalized cigenvectons of
1B acuociated ith y as follns: compute (A AY, for t=
sy until the tank of (A — Ag} 6 equal to the rank of
fa — ips Find e genera eigenvector of rank say
Define a3 (AMD, for t= 2 rer
proeed to step 8. If F< my wo try to nd ‘another finest
Independent generalized eigenvector (sce Theorem 2-12) with the
largest posible rank; that iy try to Bind. another generalized
igenvector with rank fifths i not posible, try E~, and s0
forth until mliwealy independent generalized eigenvectors ar
found. "Note that ptA — il) = a then tow ae totally (1 — a)
chains of generated eigenvectors sociated withon A has Jordan bloks on the diagonal nd ero
throwers Each chsin of generalized eigenvectors generates a
Tondan bloc whose order is equa t the Fg ofthe ein
‘8 ao desc, we may check the eeult by using the sina
traneforeation A= AQ, oF one easly, QA = AQ, where
The repent
ent .
ranaorm she following mats intothe Jordan form
B-1 1 1 0 0
tol 1-1 0 0
rr) 1
1
1. Compute the eigenvnlies of A
a= yt =
dot (A = a) = (8 == 9) HO —
Hence A has eigenvalue 2 with malliplicty 5 and eigenvalue 0
with malty 1
2 Compute (A — a) for = 1 os fallow
syns tees teshiseg Ben)
yf 2 9 8 1 1) yaaeg
@-m=15 0 09 0-1 1p Am*®
0 0 2 20 0
EEE | otek oe oMEORH0 | BEM Cilag cr
@-m-lh 6 0 0 0 of
00 0 0 2