EE263 Autumn 2007-08 Stephen Boyd

Lecture 12
Jordan canonical form

Jordan canonical form

generalized modes

Cayley-Hamilton theorem

121
 Jordan canonical form

what if A cannot be diagonalized?

any matrix A Rnn can be put in Jordan canonical form by a similarity

transformation, i.e.

J1
T 1AT = J = ...
Jq

where
i 1
i ...
Ji = Cnini
... 1
i
Pq
is called a Jordan block of size ni with eigenvalue i (so n = i=1 ni )

Jordan canonical form 122

 J is upper bidiagonal

J diagonal is the special case of n Jordan blocks of size ni = 1

Jordan form is unique (up to permutations of the blocks)

can have multiple blocks with same eigenvalue

Jordan canonical form 123

note: JCF is a conceptual tool, never used in numerical computations!

X (s) = det(sI A) = (s 1)n1 (s q )nq

hence distinct eigenvalues ni = 1 A diagonalizable

dim N (I A) is the number of Jordan blocks with eigenvalue

more generally,
X
k
dim N (I A) = min{k, ni}
i =

so from dim N (I A)k for k = 1, 2, . . . we can determine the sizes of

the Jordan blocks associated with

Jordan canonical form 124

 factor out T and T 1, I A = T (I J)T 1

for, say, a block of size 3:

0 1 0 0 0 1
iIJi = 0 0 1 (iIJi)2 = 0 0 0 (iIJi)3 = 0
0 0 0 0 0 0

for other blocks (say, size 3, for k 2)

k k1 k2

(i j ) k(i j ) (k(k 1)/2)(i j )
(iIJj )k = 0 (j i)k k(j i)k1
0 0 (j i)k

Jordan canonical form 125

 Generalized eigenvectors

suppose T 1AT = J = diag(J1, . . . , Jq )

express T as
T = [T1 T2 Tq ]
where Ti Cnni are the columns of T associated with ith Jordan block Ji
we have ATi = TiJi
let Ti = [vi1 vi2 vini ]
then we have:
Avi1 = ivi1,
i.e., the first column of each Ti is an eigenvector associated with e.v. i
for j = 2, . . . , ni,
Avij = vi j1 + ivij

the vectors vi1, . . . vini are sometimes called generalized eigenvectors

Jordan canonical form 126

 Jordan form LDS

consider LDS x = Ax

by change of coordinates x = T x = J x
, can put into form x

i = Jix
system is decomposed into independent Jordan block systems x i

x
ni x
ni1 x
1
1/s 1/s 1/s

Jordan blocks are sometimes called Jordan chains

(block diagram shows why)

Jordan canonical form 127

 Resolvent, exponential of Jordan block

resolvent of k k Jordan block with eigenvalue :

1
s 1
s ...
(sI J) 1
= ...

1
s

(s ) 1
(s )
2
(s )k
(s )1 (s )k+1
=
... ..

(s )1
= (s )1I + (s )2F1 + + (s )k Fk1

where Fi is the matrix with ones on the ith upper diagonal

Jordan canonical form 128

by inverse Laplace transform, exponential is:

tJ t k1


e = e I + tF1 + + (t /(k 1)!)Fk1
k1

1 t t /(k 1)!

t 1 tk2/(k 2)!
= e ... ..

1

Jordan blocks yield:

repeated poles in resolvent

terms of form tpet in etA

Jordan canonical form 129

 Generalized modes

consider x = Ax, with

x(0) = a1vi1 + + ani vini = Tia

then x(t) = T eJtx

(0) = TieJita

trajectory stays in span of generalized eigenvectors

coefficients have form p(t)et, where p is polynomial

such solutions are called generalized modes of the system

Jordan canonical form 1210

with general x(0) we can write

q
X
x(t) = etAx(0) = T etJ T 1x(0) = TietJi (SiT x(0))
i=1

where
S1T
T 1 = ..
SqT

hence: all solutions of x = Ax are linear combinations of (generalized)

modes

Jordan canonical form 1211

 Cayley-Hamilton theorem

if p(s) = a0 + a1s + + ak sk is a polynomial and A Rnn, we define

p(A) = a0I + a1A + + ak Ak

Cayley-Hamilton theorem: for any A Rnn we have X (A) = 0, where

X (s) = det(sI A)
 
1 2
example: with A = we have X (s) = s2 5s 2, so
3 4

X (A) = A2 5A 2I
   
7 10 1 2
= 5 2I
15 22 3 4
= 0

Jordan canonical form 1212

corollary: for every p Z+, we have

p
 2 n1

A span I, A, A , . . . , A

(and if A is invertible, also for p Z)

i.e., every power of A can be expressed as linear combination of

I, A, . . . , An1

proof: divide X (s) into sp to get sp = q(s)X (s) + r(s)

r = 0 + 1s + + n1sn1 is remainder polynomial

then

Ap = q(A)X (A) + r(A) = r(A) = 0I + 1A + + n1An1

Jordan canonical form 1213

for p = 1: rewrite C-H theorem

X (A) = An + an1An1 + + a0I = 0

as

n1
I = A (a1/a0)I (a2/a0)A (1/a0)A
(A is invertible a0 6= 0) so

A1 = (a1/a0)I (a2/a0)A (1/a0)An1

i.e., inverse is linear combination of Ak , k = 0, . . . , n 1

Jordan canonical form 1214

 Proof of C-H theorem

first assume A is diagonalizable: T 1AT =

X (s) = (s 1) (s n)

since
X (A) = X (T T 1) = T X ()T 1
it suffices to show X () = 0

X () = ( 1I) ( nI)
= diag(0, 2 1, . . . , n 1) diag(1 n, . . . , n1 n, 0)
= 0

Jordan canonical form 1215

now lets do general case: T 1AT = J

X (s) = (s 1)n1 (s q )nq

suffices to show X (Ji) = 0

ni
0 1 0
X (Ji) = (Ji 1I)n1 0 0 1 (Ji q I)nq = 0
...
| {z }
(Ji i I)ni

Jordan canonical form 1216