You are on page 1of 3

1.

4 Ill Conditioned Linear Systems



Practically, in some linear systems of equations we can see that small changes in the coefficients
of the system produce large changes in the solutions.

Such systems are called ill conditioned systems.

On the other hand, if the corresponding changes of the solutions are also small, then that system
is well conditioned.

Example 1.4.1

(i) The system


01 . 2 01 . 1 2
2 2
= +
= +
y x
y x


has the solution 2 / 1 = x and 1 = y

Whereas the system


05 . 2 01 . 2
2 2
= +
= +
y x
y x


has the solution 5 = x and 8 = y
Therefore the system is ill conditioned.

(ii) The system

31 10 9 5 7
33 9 10 6 8
23 5 6 5 7
32 7 8 7 10
4 3 2 1
4 3 2 1
4 3 2 1
4 3 2 1
= + + +
= + + +
= + + +
= + + +
x x x x
x x x x
x x x x
x x x x


Is ill conditioned since the system


1 . 31 10 9 5 7
9 . 32 9 10 6 8
9 . 22 5 6 5 7
1 . 32 7 8 7 10
4 3 2 1
4 3 2 1
4 3 2 1
4 3 2 1
= + + +
= + + +
= + + +
= + + +
x x x x
x x x x
x x x x
x x x x


Has the solution ( )
T
1 . 0 , 9 . 2 , 2 . 7 , 6 where as the original system has the solution ( )
T
1 , 1 , 1 , 1 .
Definition 1.4.2

The condition number of the non singular matrix A relative to a norm - is


1
) (

= A A A K

A matrix is well conditioned if ) (A K is closer to 1 and is ill conditioned if ) ( A K is significantly
greater than 1.

Note: - Matrix Norms

Let
ij
a be the elements of the matrix A.

(i) The Column Norm

=
i
ij
j
a A max
1


(ii) The Euclidean Norm

2 / 1
2
,
2
)
`

=

j i
ij
a A

(iii)The Raw Norm

j
ij
i
a A max

Exercise 1.4.3

(i) If
(
(
(

=
9 8 7
6 5 4
3 2 1
A then find
1
A ,
2
A and

A .

(ii) What is the condition number of the following system? Find whether the system is ill
conditioned or not.

(
(
(


=
0 0 1
2 1 1
1 0 1
A

1.5 Convergence Analysis of Iterative Methods

Definition 1.5.1

An n n matrix A is convergent if

( ) 0
lim
=
ij
k
A k

for n i ,... 2 , 1 = and n j ,... 2 , 1 =

Example 1.5.2

|
|
.
|

\
|
=
2 / 1 4 / 1
0 2 / 1
A Is this convergent or not?

Theorem 1.5.3

A is a convergent matrix iff 1 ) ( < A .

Theorem 1.5.4

For any
n
x R
) 0 (
e the sequence { }

0
) (k
x defined by,

C x T x
k k
+ =
) 1 ( ) (


for each 1 > k converges to unique solution


C x T x + =

iff

1 ) ( < T

You might also like