5 .

Tri diagonal matrix

algorithm
(TDMA)
Algorithm
#Thomas
i

I <*** S
A
O

Tridiagonal Matrix
**
i
L

L
f
I

Algorithm :

04
-

f =
(2 04
.

,
2 04, 2
.
.

04 ,
2 .

e= (1 -
). -
1)
g
=
( - 1. -
1 ,
-
1)

Step Decompose
ALU
1 .
·

Lower
per
Triang Trian
same method as before
Note : no
operations needed
forg

= II I
2 04 +
A [1](V) -0
.

49 %545
1 5550
395
.
-

1323
1 .

-
0 .

717 I
 AX = b
un UX = D
Lux = b
~ ↓
b To
LD
=
=

get X
2 .
Forward Solve to
get D
LD = b

I
40 8

I
·

D = 20 8 .

14 .
221

L 210 .
996

Sep
3. Back Substitution

I I
X = 65 970 .

93 778
.

124 538
I
.

159 480 .

Remark
Noj of operations
:
 Problem :
Transfe
#eat

in 23 : 7 Th
x= 0 x= L

T(x t) ,
:
temp as a func ? Of &
time t

Heat Conduction
equ in a

PDE
involving
In the
"steady State"
: =
0

=
XX =
Xi -

Yi=1 =
Xi+ 1
-

Xi

~TiTi) =

(Tit
-

2Ti +
Tr ) =
0

[x =
1
Ti + 1
-

2T + Ti x = 0

Mathematical model for for

system after numerical

method
 (*) Ti +1 2Ti +
Ti 0
-
=

i= 2 Tz-2Tz + T
,
=
0

Tg-2Tz =
Th1
i = 3:
Ty -2Tz + Tz =
0

E 4 :
Tg-2Ty + Tz =
0

i =
S: Tj-2Tg + Ty =
0

i =
6:
Tz-2Tj +
Tg =
0

-
=

25 + Tz =
-

Tbz
2100 0
-
T2 -

Tb,
+ 2100
Tz
-

O
01-210 O
Ty
00 1-21 O
Tj

~ Tbz
01-2 To -

Matrix
Tridiagonal
 Decomposition
Lolesky
A symmetric matrix A can be
T

decomposed as A = LL -

Multiply the terms in the RHS of (#) and

other . This will
set
equal to each
give following
recurrence relation :

i 1
-

=U
ne-Elijlis
In
, for E ,
2. .

-
lik =

ann-i
.
3
Computational Advantage Only : half the

Storage is neede d.
 -xamps I 55225 979

( = m = 5 =
2 .

4495

21 = 2 = 6 .

1237
lin
22 = =16 15572 .

=
4 1833
:

For third now 17 =

3

= =
i =
1 :
(3) 22

5 - 6 1237(22454)
i 132
= zl = 20 917
-

=
2 :
-
.

4- 1833

13 :
-1 - 13 =
T79-122 45
(20 917)3 .
·

= 6 1101

I I
.

L = 2 .

4495

6 -

12374 1833 .

122 .

45420917 6 1101 .

The evaluation of luk involves

taking
square
root of matrix
. For
positive definite
matrix ,
this will not be a problem .
 Definite Mara re-definite
Positive
if it is
symmetric and if Ax >o

for every n-dimensional rector (0 ·

that the Latrix
S .

Show is
positive definite
-

I I
A =
2 -

I O

-
12 -
I

10 -

12

symmetric because AT =
A

xAx
(xx27/
=

=>
2x -
26x + 2x2 -

2x33 22
+

= x+ x -
-

2x , 2 + x + x= 24 + 13+ 3
~
2
=
x + (x -) + (x43)2 + x3
70

unless u, =
x =
Xz =
0
 A matrix A
Therem: symmetric is

positive definite iff each of its

leading principal submatrices has a

positive
determinant.

-1

I I
2 O
A =

+ 2 -
I

10 + 2

det (A1) = 2 >O

det (A) = 3 > 0

det (A3) = 2 (4-1) + 1 ( 2) -

= 6 -
2 =
470
 The condition no .

=
+(a) ( x+ Sx
=
,

Relative -
error inf

f() f(x))
Eac
-

F(x)
Relative Error in K

#
condition no .
= Ratio of relativecrras

)
-.
-
= en

=
1 1) =
 Ax = b

/ 1 0001]
A = 0

(2) 1)
b =

· x=

with small in
Now , a
change b

Let
,
5
/ 20001]
=

we
get ,
ci =

/)
?

whichis different fromcondition a

 Evaluation of condition no .
requires use of Norms

Vector an Mai
norms :

IkIIo = max Wil

i

Il Allo = max
ii [lay) - Maximum
Row
sum

-operties: AFAIKA
11 Ac 11 = 11 All I all

of Mahix
Example Norms :

llAlly = max laij) "Maximum

1
jen i column Sum"

11 Ally = max

1 imag , "Maximum
Row Sum"

llAlly = la "The Frobenius Norm

(also called as Euclidean

llAll =
max(FA) "The 1-2
"
norm
also called spectral
norm
 Peturbed Right-hand side

(1) -

AC =
b >
- IIb 11 = 11 A xll
A(x + Sx) b + Sb 111 AllIIsl)
()
x - =

Subtract (*) from original egu

ASx = Sb

Su = Asb
(*) II sall =
I'll 115 b11
II Sull
llA-IllbI : All All
=

>
HAIA
-

Isal All All

a
Peturbed Coefficient matrix implies
- -

(A + SA)(x + Su) = b
subtract unperturped equation Ax = .
b

Asx + SAx + SASx = 0

Ask + SA(x +
Sx) = 0

I- Neglecting as
Say
Su = -
A SA(x + 5x)
= -
A SAC
 See ASAC
Ilsall = NA"11sAll 11

AllAAA
Condition No ·
K(A) = 11 All 11 A +11

Let

I 0001)
A =
1 .
0 1 .

21 .
0 1 .

Al =
?

11 Allo = max
&2 , 2 .

0001
=
2 .

000 1

k(A) = 40 , 000

Ill-conditioned
system .
 Method
Iterative :

Ax =
b
an( + 92(2 + -- -
+ dinCn = b ,

:
anix , + Anz(2 + ---- + Ann (n =
bu

For
3x3 system :
allx , + 112x + Aiskz = b ,

& 21 % + 922x2 + 92343 = b2

a 31x + dzz + AzzMz

=
by

d , 2x 9 , 34
x =
-

All
all
( =
-azx
-

- a b
=
3
-azx
Will study in next lecture