Lecture 14:

Recall :
Solve AI=F ⇐ NIK
"
= PIK + I
p
- -
.

µ -
P u

III
'
P ) I
-

s
N
Converge pc
.

'

if
-

What N P is Not diagonalizable ?

valued matrix
theorem A Mnxnic ) be complex
-

Let C- nxn
.

: a

I
if pl A) s
Then : him Ak =
D= zero matrix if and only
.

k→o
" '
MIhtb iff punk 't
Corollary :
The iterative scheme . I =
converges

Ik MIK tb
"
Ci )
PI Consider
-

: =

I
"
= MI
't
tb -
12 )
-

Ci ) -14 I
k "
=
MEK ⇒ e-
K
= Mk @ °

Ek J iff Mk O iff ) at
→ →
PCM
kiss
 Theorem
→
Proof of :
°

#
I
of A with eigenvector
X be eigenvalue
( ) Let an
⇒

ftkv=(
I
Then :
Iii
Air
-

.
⇒ bing.tk )
VO
T
⇒ I =
Chainz
Ah ) .

171 at
i .
to
at
PLA ) o
i.

ha An be eigenvalues ( can be repeated )

Xi
⇐ I Let ,
. .

, ,

Jordan Canonical Form

the idea
.

We apply
Useful fact ( be used without
proof )
:

can

A Mnxn ( E ) there exists invertible Q E Mnxn ( G )

For
an
E
any ,

"
J is the Jordan canonical
Such that A = QJQ where

form of A . Actually ,

g
s
-

Qin
me "

.
.

,i
"
" -

"
.

(
Jordan block with
Jmj I Xij ) = is called
mj
0
I Xij
,
Now Ak ( QJQ
"
)k= QJKQ "
and

n⇒
, =

clan
it
=

TEEJ
a Jm "
,

"
g-

cmpiiij
cminiij
Jmdtid
=

(
-

Gm

cnn.i.in?jmi
,

;¥Iii
'
knit
m

For ke mo
-
I ' ,
.
-
-

Cmjitij
" '
"

.
.

s
.

"

(
mi -

O K

iff High I
Xij
Since for
liijl
) all
put at
, a
I
j .

Jk if k
Jnf
' so
for O
-

=
O all
j : →
feign
c ' . .

"
QJKQ
fig
" o
A
=

fi;y
i -
=
 relaxation method
Splitting choice 3 :
Successive over ( Sor
)
Dt U
iterative scheme :
A =
Lt
Consider the following
' '
Dijk
' '
UI
's
I LA )
LIK
.

+ + =

{ y-ktt.tw/xkti-cw-igxk )
- C * * I

where WEIR

Put C * I into # I :

) tb (
It SOR )
'

" '

( Lt D)
'T I =

( WID -
c Dtu ,

- -

N
p
'
Gauss Seidel method if =L
, .
SOR = -
w .
 remark the need to check
Important discuss we
convergence
:
:
,

))
' at
'

PIN
'
P )
Pf ¢two ) ( ID total
-
= -

G - S .

'
)
)
-

=p ( I LTD
U
-
'
P ) ) C -
ez
PCN