You are on page 1of 23

Unit 14: Numerical Methods for solving ODES and PDES

Session 35
Eigen Value Problem

Contents:
Introduction, p 496
35.1 Eigen Values or Latent Roots, p 497
35.2 Power method to find dominant eigen Values, p 499
35.3 Power method to find dominant eigen vector, p 500
35.4 Raleigh Quotient Method, p 502
35.5 Finding other eigen values, p 504
Solutions of Activities, p 514
Problems, p 515
Summary, p 517
Learning outcomes, p 518

Introduction

The subject of Eigen values and Eigen vectors is important because it turns
up in many branches of pure and applied sciences and in statistics. In this
session we shall discuss methods for determining the Eigen values of real
matrices.

496
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

35.1 Eigen Values or Latent Roots

If A is an (𝑛 × 𝑛) matrix, a vector 𝑥 and 𝐴𝑥 may be shown as follows

(a) (b)

Ax

Ax

Figure35.1

For 𝐴𝑥 to be a scalar multiple of 𝑥, there is special vector 𝑥, which are the


eigen vectors of x.

𝐴𝑥 = scalar multiple of 𝑥

That is 𝐴𝑥 = .

 is called the eigen value

𝑥 is called the eigen vector which corresponds to 

Since 𝐴𝑥 = 𝑥

= 𝐼𝑥

where 𝐼 – is the identify matrix Then (𝐴 − 𝐼) 𝑥 = 0

For x to be non – trivial,

𝑑𝑒𝑡 (𝐴 − 𝐼) = 0
𝑎11 𝑎12 . . 𝑎1𝑛
𝑎21 𝑎22 . . 𝑎2𝑛
If 𝐴 = . . . . .
. . . . .
[𝑎𝑛1 𝑎𝑛2 . . 𝑎𝑛𝑛 ]

497
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

𝑎11 − 𝜆 𝑎12 . . 𝑎1𝑛


𝑎21 𝑎22 − 𝜆 . . 𝑎2𝑛
Then 𝐴 − 𝜆𝐼 = . . . . . =0
. . . . .
[ 𝑎𝑛1 𝑎𝑛2 . . 𝑎𝑛𝑛 − 𝜆]
Expansion of the determinant produces the characteristic polynomial of
order n, whose roots are the required values of . There for  have
𝑛 different values for (𝑛 𝑥 𝑛) matrices.

As with linear equations, we have two types of method of determining eigen


values of  and the vectors 𝑥. They are direct process and indirect process.
But the decision which to use may depend not so much on the form of the
matrix as on what solutions are required.

For example, there are 𝑛 eigen values and generally 𝑛 eigen vectors for a
matrix of order 𝑛, and the practical problems may require us to find all the
solutions, or only a small number involving few of the values such as largest
or smallest modules and so on.

If only few solutions are needed, we apply iterative methods such as power
method discussed in this session.

Example 35.1

Find the eigen values and the eigen vectors of the matrix;

1 1 3
𝐴 = [1 −2 1]
3 1 3

We leave to the learner to show that the eigen values and the corresponding
eigen vectors are respectively

𝜆 = 1, 𝑥1 = (1 1 0)𝑇

𝜆 = 5, 𝑥2 = (−1 1 0)𝑇 , (0 0 1)𝑇

498
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

35.2 Power method to find dominant eigen Values

We assume the matrix 𝐴 is (𝑛 𝑥 𝑛) and it has 𝑛 real and distinct eigen


values. Then there are 𝑛 eigen vectors which form a basis for the
dimensional space.

Let , 𝑣1 𝑣2 , 𝑣3 , . . . 𝑣𝑛 be the eigen vectors. Take any arbitrary vectors 𝑣 (0) . It


can be expressed in the form

𝑣 (0) = 𝑐1 𝑣1 + 𝑐2 𝑣2 +. . . +𝑐𝑛 𝑣𝑛

Where 𝑐1 , 𝑐2 , . .. , 𝑐𝑛 are constants.

Let 𝑣 (1) = 𝐴𝑣 (0)

Similarly 𝑣 (𝑘+1) = 𝐴𝑣 (𝑘)

|𝑣 (𝑘+1) |
Then we can prove lim = |1|
𝑘→∞ |𝑣 (𝑘) |

Where 𝑖 in the largest on dominant eigen value. This can be proved in the
following manner.

𝑣 (0) = 𝑐1 𝑣1 + 𝑐2 𝑣2 +. . . +𝑐𝑛 𝑣𝑛

𝑣 (1 = 𝐴𝑣 (0) = 𝐴( 𝑐1 𝑣1 + 𝑐2 𝑣2 +. . . +𝑐𝑛 𝑣𝑛 )

= 𝑐1 𝐴𝑣1 + 𝑐2 𝐴𝑣2 +. . . +𝑐𝑛 𝐴𝑣𝑛

By the definition; 𝐴𝑉1 = 𝜆1 𝑉1 , 𝐴𝑉2 = 𝜆2 𝑉2 of eigen vectors

By substituting

𝑣 (1) = 𝑐1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝑣2 + . . . + 𝑐𝑛 𝜆𝑛 𝑣𝑛

𝑣 (2) = 𝐴𝑣 (1)

= 𝐴(𝑐1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝑣2 + . . . + 𝑐𝑛 𝜆𝑛 𝑣𝑛 )

= 𝑐1 𝜆1 𝐴𝑣1 + 𝑐2 𝜆2 𝐴𝑣2 + . . . + 𝑐𝑛 𝜆𝑛 𝐴𝑣𝑛

= 𝑐1 𝜆1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝜆2 𝑣2 + . . . + 𝑐𝑛 𝜆𝑛 𝜆𝑛 𝑣𝑛

= 𝑐1 𝜆1 2 + 𝑐2 𝜆2 2 𝑣2 + .. . + 𝑐𝑛 𝜆𝑛 2 𝑣𝑛

499
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Similarly we can prove

𝑣 (𝑘+1) = 𝑐1 𝜆1 (𝑘+1) 𝑣1 + 𝑐2 𝜆2 (𝑘+1) 𝑣2 +. . . +𝑐𝑛 𝜆𝑛 (𝑘+1) 𝑣𝑛

  2 
k 1
 k 1 
= 𝜆1 (𝑘+1)
c1v1  c2   v2  ................ cn n vn 
  1  1 

  2 
k
 k 
𝑣(𝑘) = 1(𝑘) c1v1  c2   v2  ................ cn n vn 
  1  1 

k 

𝑣 (𝑘+1) |𝜆1 (𝑘+1) |[𝑐1 𝑣1 +0………… ]


| |= [ Since 𝜆1 > 𝜆2 … . . ]
𝑣 (𝑘) |𝜆1 (𝑘) |[𝑐1 𝑣1 +0………… ]

= |𝜆1 |

Suppose first that 1 is of the largest modules and |𝜆𝑟 | ≠ |𝜆1 | , 𝑟 =


1, 2, 3, … 𝑛.

𝜆 𝑘
If 𝜆1 = 𝜆2 as 𝑘 → ∞ (𝜆2 ) → 1
1

In such situations we can’t adopt this method

Therefore when the ratios of corresponding components of successive


vectors 𝑣 (𝑘+1) and 𝑣 (𝑘) will have the constant value 𝜆1 , the required
dominant eigen value is determined from this fact.

35.3 Power method to find dominant eigen vector

If the ratios of corresponding components of successive vectors converges


to a values 1,

|𝑣 (𝑘+1) |
that is |𝑣 (𝑘) |
→ |𝜆1 |

Then the eigen vector corresponding to 𝜆1 is 𝑣 (𝑘) . This can be proved in the
following manner

𝑣 (0) = 𝑐1 𝑣1 + 𝑐2 𝑣2 +. . . + 𝑐𝑛 𝑣𝑛

500
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

𝑣 (𝑘) = 𝑐1 𝜆1 (𝑘) 𝑣1 + 𝑐2 𝜆2 (𝑘) 𝑣2 + . . . +𝑐𝑛 𝜆𝑛 (𝑘) 𝑣𝑛

k 
  2 
k
 n  
k

1  c1v1  c2   v2  ............... cn   vn 
  1   1  

= 𝜆1 (𝑘) [𝑐1 𝑣1 + 0 .. . ] 𝑎𝑠 k  

= 𝜆1 (𝑘) 𝑐2 𝑣1

Therefore 𝑣 (𝑘 ) is the unnoramalized eigen vector corresponding to the eigen


value 𝜆1 .

Example 35.2

Find the dominant eigen value and the eigen vector of the matrix

1 1 3
𝐴 = [1 −2 1]
3 1 3

Solution

Selecting an initial vector 𝑣 (0) = [1 1 1]𝑇

1 1 3 1 5
𝑉 (1) = 𝐴𝑣 (0) = [1 −2 1] [1] [0]
3 1 3 1 7
𝑣 (1)
|𝑣(0)| = 7

1 1 3 5 26
𝑣 (2) = 𝐴𝑣 (1) = [1 −2 1] [0] = [12]
3 1 3 7 36
𝑣 (2) 36
[𝑣(1)] = = 5.14
7

𝑣 (3) = [146, 38, 197]

𝑣 (3) 197
|𝑣(2)| = = 3.49
36

𝑣 (4) = [774 268 1066 ] 𝑇

501
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

𝑣 (4) 1066
| (3)
|= = 5.41
𝑣 197

𝑣 (5) = [4243 1312 5799 ]

𝑣 (5) 5799
|𝑣(4)| = 1066 = 5.44

The results obtain after five step are clearly close to the regard value.

 Dominant eigen value = 5.44

Dominant eigen vector = 𝑣 (4) = [774 268 1066]𝑇

35.4 Raleigh Quotient Method

This method is applicable only for symmetric matrices. If 𝐴 is symmetric


with distinct eigen values, eigen vectors are orthogonal.

That is 𝑣 ′ ≈ 𝑣1 = 1, 𝑣1 ′ ≈ 𝑣2 = 0 and so on.

Where 𝑣1 , 𝑣2 are eigen vectors of 𝜆1 , 𝜆2 .

With our basic assumption that the vectors are linearly independent, and
starting with an arbitrary vector

𝑥 (0) , We can construct 𝑥 (1) , 𝑥 (2) ,…….., 𝑥 (𝑘) such that 𝑥 (𝑘+1) = 𝐴𝑥 (𝑘)

It can be shown that, the quantity 𝑅𝑘 which is called the Raleigh quotient
given by

(𝑥 (𝑘) ) 𝐴𝑥 (𝑘)
𝑅𝑘 = ′
(𝑥 (𝑘) ) 𝑥 (𝑘)

tends to the dominant eigen value 𝜆1 as 𝑘 → ∞ This is called the Raleigh’s


principle.

502
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES


(𝑥 (𝑘) ) 𝐴𝑥 (𝑘)
That is 𝑅𝑘 = ′ → 𝜆1 as 𝑘 → ∞
(𝑥 (𝑘) ) 𝑥 (𝑘)

When |𝑅𝑘+1 − 𝑅𝑘 | ≤ ∈ (where ∈ is a very small number), take 𝑅𝑘 as


𝜆1 and 𝑥 (𝑘) as corresponding eigen vector.

This formula can be proved in the following manner.

Let 𝑥 (0) by any arbitrary vector,

With our basic assumption that the vectors are linearly independent, we can
write

𝑥 (0) = 𝑐1 𝑣1 + 𝑐2 𝑣2 + .. . + 𝑐𝑛 𝑣𝑛

𝑥 (1) = 𝐴𝑥 (0) = 𝑐1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝑣2 + 𝑐𝑛 𝜆𝑛 𝑣𝑛

(𝑥 (1) ) .𝐴(𝑥 (1) )
𝑅1 = ′
(𝑥 (1) ) (𝑥 (1) )

(𝑐1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝑣2 +. . . ) (𝑐1 𝜆1 2 𝑣1 + 𝑐2 𝜆2 2 𝑣2 + ⋯ )
= ×
(𝑐1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝑣2 + ⋯ ) (𝑐1 𝜆1 𝑣1 + 𝑐2 𝜆2 𝑣2 + ⋯ )

𝑐1 2 𝜆1 3 +𝑐2 2 𝜆2 3 +⋯
= (𝑐 2 2
1 𝜆1 ) +(𝑐2 𝜆2 ) +⋯

2 𝜆 3
𝜆1 3 [𝑐1 2 +𝑐2 ( 2 ) +⋯ ]
𝜆1
= 2 𝜆 2
𝜆1 [𝑐1 2 +𝑐2 2 ( 2 ) +⋯ ]
𝜆1

Similarly

𝜆 𝑘+2
𝑐1 2 + 𝑐2 2 ( 2 ) +⋯
𝜆1
𝑅𝑘 = 𝜆1 𝑘+1
𝑐 2 + 𝑐 2 (𝜆2 ) + ⋯]
[ 1 2 𝜆1

𝑅𝑘 → 1 as 𝑘 → ∞

503
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Activity 35. 1

Shoe how to find the eigen values of A Raleigh quotient method

10 7 8 7
7 5 6 5
𝐴=[ ]
8 6 10 9
7 5 9 10

35.5 Finding other eigen values

By the method outlined in the previous chapter, we can find only the
dominant eigen value of a matrix. No other eigen values being calculable
easily by the power method. There are various possibilities for modifying
the basic method to obtain other eigen values as well.

Theorem

If  is the dominant eigen value of 𝐴, 𝑓(𝐴) is the dominant eigen values


of 𝑓(𝐴)

For example (i) if 𝜆 → 𝐴

then 𝜆3 + 3𝜆 → 𝐴3 + 3𝐴

(ii) 1⁄𝜆 → 𝐴−1

Therefore the dominant eigen value of the matrix 𝐴−1 is the reciprocal of the
smallest eigen values of 𝐴.

Example 35. 3

2 1
Find the smallest eigen value of matrix 𝐴 = [ ]
3 2

Solution
1
Inverse of the matrix 𝐴 = det 𝐴 𝐴𝑑𝑗(𝐴)

504
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

2 −3 𝑇 2 −1
𝐴𝑑𝑗 (𝐴) = [ ] =[ ]
−1 2 −3 2

𝑑𝑒𝑡 𝐴 = 4 – 3 = 1

2 −1
𝐵 = 𝐴−1 = [ ]
−1 2

Using the power method to evaluate the dominant eigen values of 𝐴−1 . Let
1
𝑥 (0) = [ ]
1

2 -1 1 1 
𝑥 (1) = 𝐵 𝑥 (0) = 
2 1   1
- 3   

𝑥 (1) 1
= =1
𝑥 (0) 1
2 −1 1 3
𝑥 (2) = 𝐵𝑥 (1) = [ ][ ] = [ ]
−3 2 −1 −5

𝑥 (2) 5
= =5
𝑥 (1) 1
2 −1 3 11
𝑥 (3) = 𝑥 (2) = [ ][ ] = [ ]
−3 2 −5 −19

𝑥 (3) 19
= = 3.8
𝑥 (1) 5
2 −1 11 30
𝑥 (4) = [ ][ ]=[ ]
−3 2 −19 −71
𝑥 (4) 71
= 19 = 3.74
𝑥 (1)

2 −1 30 131 𝑥 (5) 232


𝑥 (5) = [ ][ ]=[ ], = = 3.27
−3 2 −71 −232 𝑥 (4) 71

2 −1 131 494 𝑥 (6) 857


𝑥 (6) = [ ][ ]=[ ], = 232 = 3.69
−3 2 −232 857 𝑥 (5)

2 −1 494 1845 𝑥 (7)


𝑥 (7) = [ ][ ]=[ ], = 3.73
−3 2 −857 3196 𝑥 (6)

Giving 𝜆1 = 3.7 (to the first decimal point)

505
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Example 35.4

Find the dominant eigen value of the following.

1 2 1 1
1 1 3
2 3 1 4
𝐴 = [1 −2 3] 𝐵=[ ]
1 1 2 2
3 1 3
1 4 2 3

Solution

Selecting an initial vector

𝑉 (0) = [1,1,1]𝑇

1 1 3 1 5
(1) (0)
𝑉 = 𝐴𝑉 = [1 −2 3] [1] = [2]
3 1 3 1 7

𝑉 (1) 7
| (0)
|= =7
𝑉 1

1 1 3 5 28
(2) (1)
𝑉 = 𝐴𝑉 = [1 −2 3] [2] = [22]
3 1 3 7 38

𝑉 (2) 38
| (1)
|= = 5.43
𝑉 7

1 1 3 28 164
(3) (2)
𝑉 = 𝐴𝑉 = [1 −2 3] [22] = [ 98 ]
3 1 3 38 220
𝑉 (3) 220
|𝑉 (2)| = = 5.78
38

1 1 3 164 1184
(4) (3)
𝑉 = 𝐴𝑉 = [1 −2 3] [ 98 ] = [ 628 ]
3 1 3 220 1250
𝑉 (4) 1250
|𝑉 (3)| = = 5.68
220

1 1 3 1184 5562
(5) (4)
𝑉 = 𝐴𝑉 = [1 −2 3] [ 628 ] = [3678]
3 1 3 1250 7930

506
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

𝑉 (5) 7930
|𝑉 (4)| = 1250 = 6.34

Difference between 𝑉 (3) and 𝑉 (4) in dose to the required value.

Hence Dominant Eigen value is 5.68

(b) Selecting 𝑋 (1) = [1,1,1,1]𝑇

1 2 1 1 1 5
2 3 1 4 1 10
𝐴𝑋 (1) =[ ][ ] = [ ]
1 1 2 2 1 6
1 4 2 3 1 10
5
(1) ′ 10
(𝑋 ) . 𝐴𝑋 (1) = [1 1 1 1] [ ] = 31
6
10
1
(1) ′ 1
(𝑋 ) 𝑋 (1) = [1 1 1 1] [ ] = 4
1
1
1
(𝑋 (1) ) . 𝐴𝑋 (1) 31
𝑅1 = = = 7.75
(𝑋 (1) )1 . 𝑋 (1) 4

𝑋 (2) = 𝐴𝑋 (1) = [5 10 6 10]𝑇

1 2 1 1 5 41
2 3 1 4 10 86
𝐴𝑋 (2) =[ ][ ] = [ ]
1 1 2 2 6 47
1 1 2 2 10 87
1
(𝑋 (2) ) .𝐴𝑋 (2)
𝑅2 = 1
(𝑋 (2) ) .𝑋 (2)

41
(2) ′ 86
(𝑋 ) . 𝐴𝑋 (2) = [5 10 6 10] [ ] = 2217
47
87
5
(2) ′ 10
(𝑋 ) 𝑋 (2) = [5 10 6 10] [ ] = 221
6
10

507
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

2217
𝑅2 = = 8.49
221

𝑋 (3) = 𝐴𝑋 (2) = [41 86 47 87]𝑇

X (3)  AX ( 2)  41 86 47 87


T

1 2 1 1 347
2 3 1 4 735
𝐴𝑋 (3) =[ ][ ]
1 1 2 2 395
1 4 2 3 740
1
(𝑋 (3) ) .𝐴𝑋 (3)
𝑅3 = 1
(𝑋 (3) ) .𝑋 (3)

347
(3) ′ 735
(𝑋 ) . 𝐴𝑋 (3) = [41 86 47 87] [ ] = 160382
395
740

41
(3) ′ (3) 86
(𝑥 )𝑥 = [41 86 47 87] [ ] = 18855
47
87
160382
𝑅3 = = 8.52
18855

Required value is closer to 𝑅3

Eigen Value = 10.03

Example 35.5

Find the dominant eigen value and the eigen vector of the following
symmetric matrix 𝐴 using Raleigh Quotient method or any other numerical
method.

5 −1 −2
𝐴 = [−1 3 −2]
−2 −2 5

508
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Solution

5 −1 −2
Given 𝐴 = [−1 3 −2]
−2 −2 5

1 1
 1,1,1  1
(1) T
Choosing 𝑥 = [1 1 1 ]𝑇 = [1 ] x (1)

1 1

2
𝐴(𝑥 (1) ) = [0]
1
𝑇
(𝑥 (1) ) . 𝐴𝑥 (1) = 3
𝑇
(𝑥 (1) ) . 𝑥 (1) = 3
𝑇
(𝑥 (1) ) . 𝐴(𝑥)(1) 3
𝑅1 = = =1
(𝑥 (1) )𝑇 . 𝑥 (1) 3

Similarly 𝑅2 = 3.4

𝑅3 = 5.197 𝑅4 = 5.507

Dominant eigen value 𝜆1 = 5.197

42
Eigen vector = [−22]
−3

Example 35.6

Find the eigen vector of 𝐴 using

a) Power method

b) Raleigh quotient method

3 2 4
𝐴 = [2 0 2]
4 2 3

509
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Solution

a). Power Method

3 2 4
𝐴 = [2 0 2]
4 2 3

Initial vector 𝑣 (0) = (1 1 1)𝑇

3 2 4 1 9
(1) (0)
𝑣 = 𝐴𝑣 = [2 0 2] [1] = [4]
4 2 3 1 9

𝑣 (1) 9
= =9
𝑣 (0) 1
3 2 4 71
𝑣 (2) = 𝐴𝑣 (1) = [2 0 2] [36]
4 2 3 71

𝑣 (2) 71
= = 7.89
𝑣 (1) 9

3 2 4 71 569
𝑣 (3) = 𝐴𝑣 (2) = [2 0 2] [36] = [284] =
4 2 3 71 569
𝑣 (3) 569
= = 8.0141
𝑣 (2) 71

3 2 4 569 4551
𝑣 (4) = 𝐴𝑣 (3) = [2 0 2] [284] = [2276]
4 2 3 569 4551
𝑣 (4) 4551
= = 7.998
𝑣 (3) 569

3 2 4 4551
𝑣 (5) = 𝐴𝑣 (4) = [2 0 2] = [2276]
4 2 3 4551
36409
(5)
𝐴𝑣 = [18204]
36409

𝑣 (5) 36409
= =8
𝑣 (4) 4551

510
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Eigen vector = 𝑣 (4) = [4551 2276 4551]𝑇

= [1 0.5 1]𝑇

b). Raleigh quotient

3 2 4 1
𝐴 = [2 0 2] choosing 𝑥 (1) = [1 1 1]𝑇 = [1]
4 2 3 1

Step 1

3 2 4 1 9
𝐴(𝑥1 ) = [2 0 2] [1] = [4]
4 2 3 1 9

9
(𝑥1 )𝑇 = 𝐴𝑥 (1)
= (1 1 1) [4]
9

= (9 + 4 + 9)

= 22

1
(𝑥 (1) )(𝑥 (1) ) = [1 1 1] [1]
1

= 3
𝑇
(𝑥 (1) ) 𝐴𝑥 (1) 22
𝑅1 = = = 7.333
(𝑥 (1) )𝑇 𝑥 (1) 3

Step 2

9
(2) (1)
𝑥 = 𝐴𝑥 = [4]
9
3 2 4 9
(2)
𝐴𝑥 = [2 0 2] [4]
4 2 3 9
71
𝐴𝑥 (2) = [36]
71

511
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

71
(2) 𝑇[ ]
[𝑥 ] = 9 4 9 [36]
71

= 1422

9
[𝑥 (2) ]𝑇 𝑥 (2) = [9 4 9] [4]
9

= 81 + 16 + 81 = 178
𝑇
[𝑥 (2) ] 𝐴𝑥 (2)
𝑅2 =
[𝑥 (2) ]𝑇 𝑥 (2)
1422
= = 7.989
178

Step 3

71
𝑥 (3) = 𝐴𝑥 (2) = [36]
71
3 2 4 71
𝐴𝑥 (3) = [2 0 2] [36]
4 2 2 71
564
(3)
𝐴𝑥 = [284]
569
569
(3) 𝑇 (3)
[𝑥 ] 𝐴𝑥 = [71 36 71] [284]
569

= 91022

71
[𝑥 (3) ]𝑇 𝐴𝑥 (3) = [71 36 71] [36]
71
91022
𝑅3 = 11378

= 7.999

512
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Step 4

569
𝑥 (4) = 𝐴𝑥 (3) = [284]
569
3 2 4 569
𝐴𝑥 (4) = [2 0 2] [284]
4 2 3 569

𝐴𝑥 (4) = [4551 2276 4551]

4551
(4) 𝑇 (4)
[𝑥 ] 𝐴𝑥 = [569 284 569] [2276] = 58235422
4551
569
[𝑥 (4) ]𝑇 𝑥 (4) = [569 284 ]
569 284] = 7218178
[
569
𝑇
[𝑋 (4) ] 𝐴𝑥 (4) 5825422
𝑅4 = = = 7.999
[𝑥 (4) ]𝑇 𝑥 (4) 728178

Then 𝑅4 − 𝑅3 = 7.999 – 7.999

= 0

So the eigen vector = [𝐴𝑋 (𝑥) ]𝑇

= [4551 2276 4551]𝑇

= [1 0.5 1]𝑇

513
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Solutions of Activities

Activity 35.1

Selecting 𝑥 (1) = [1 1 1 1]𝑇

1st Step

10 7 8 7 1 32
(1) 7 5 6 5 1 23
𝐴𝑥 =[ ][ ] = [ ]
8 6 10 9 1 33
7 5 9 10 1 31

32
(1) 𝑇 (1) 23
[𝑥 ] 𝐴𝑥 = [1 1 1 1 ] = [ ] = 119
33
31

1
(1) 𝑇 (1) 1
[𝑥 ] 𝐴𝑥 = [1 1 1 1 ] [ ] = 4
1
1

(𝑥 (1) )1 𝐴𝑥 (1) 119


∴ 𝑅1 = = = 29.75
(𝑥 (1) )1 𝑥 (1) 4

2nd Step

𝑥 (2) = 𝐴𝑥 (1) = [32 23 33 31 ]𝑇

10 7 8 7 32 962
(2) 7 5 6 5 23 692
𝐴𝑥 =[ ][ ] = [ ]
8 6 10 9 33 1003
7 5 9 10 31 946


(𝑥 (2) ) (𝐴𝑥 (2) )
∴ 𝑅2 =
(𝑥 (2) ) (𝑥 (2) )

962
692
(𝑥 (2) )′ (𝐴𝑥 (2) ) = (32 23 33 31) [ ]
1003
946

= 109125

514
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

32
23
(𝑥 (2) )′ (𝑥 (2) ) = (32 23 33 31) [ ]
33
31

= 3603

109125
∴ 𝑅2 = = 30.287
3603

Proceeding in similarly way, we obtain the dominant eigen value 𝜆1 = 30. 289

0.958
0.689
Normalized eigen vector 𝑣1 = [ ]
1.0
0.944

Problems

1) Find the dominant eigen value of the following matrices

1 1 1 1
1 −3 2
1 2 3 4
(a) [4 4 −1] (b) [ ]
1 3 6 10
6 3 5
1 4 10 20

2) If 4 discuss equal in all respect are mounted on a central shaft which


is free to rotate in its bearings following equations, describe the
angular displacement of the disk.
𝐼𝜃 = −𝐶(𝜃1 − 𝜃2 )
𝐼𝜃2 = 𝐶(𝜃1 − 𝜃2 ) − 𝐶(𝜃2 − 𝜃3 )

𝐼𝜃3 = 𝐶(𝜃2 − 𝜃3 ) − 𝐶(𝜃3 − 𝜃4 )

𝐼𝜃4 = 𝐶(𝜃3 − 𝜃4 )

Assuming cuscillatory motion

𝜃1 + 𝑊 2 𝜃𝑖 = 0 𝑖 = 1,2, … .4

Show that equations can be put in the form

515
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

[𝐴][𝜃] = 𝜆[𝜃] where

= 𝜆 = 𝐼𝑊 2 /𝐶 [𝐴] is a 4 × 4 matrix and

[𝜃] = [𝜃1 𝜃2 𝜃3 𝜃4 ]𝑇

Starting with [1 − 1 1 1 − 1]𝑇 . Use power method to find eigen


value of the largest modulus, and the given vector to two decimal place.

Repeating with starting vector [1 0 0 1]𝑇 . why is eigen value of the


largest modules is not obtained. Notice symmetry and find remaining eigen
values.

3) Find the first and second eigen values and eigen vectors of the
matrix,
5 −1 −2
𝐵 = [−1 3 −2]
−2 −2 5

4) Find the dominant eigen value and the eigen vector of the following
symmetric matrix. Using Rayleigh quatient method.

5 −1 −2
𝐴 = [−1 3 −2]
−2 −2 5

5) Find the dominant and smallest eigen values of the following


matrices.
8 −1 −5 5 −1 2
(a) 𝐴 = [−4 4 −2] (b) 𝐵 = [−1 3 −2]
−2 −2 −7 −2 −2 5

516
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

Summary

In this session we have seen,

Power method

Let 𝑣1 , 𝑣2 , 𝑣3 , . . . , 𝑣4 be the eigen vectors. Take any arbitrary vectors 𝑣 (0) . It


can be expressed in the form

𝑣 (0) = 𝑐1 𝑣1 + 𝑐2 𝑣2 + . . . + 𝑐𝑛 𝑣𝑛 where 𝑐1 , 𝑐2 , 𝑐𝑛 are constants.

Let 𝑣 (1) = 𝐴𝑣 (0)

Similarly 𝑣 (𝑘+1) = 𝐴𝑉 (𝑘)

|𝑉 (𝑘+1) |
Then we can prove lim = |𝜆|
𝐾→∞ |𝑣 (𝑘) |

Where  in the largest on dominant eigen value.

If the ratios of corresponding components of successive vectors converges


to a values 𝜆1

|𝑣 (𝑘+1) |
that is |𝑣 (𝑘) |
→ |𝜆1 |

Then the eigen vector corresponding to 𝜆1 is 𝑣 (𝑘) .

Raleigh Quatient Method

This method is applicable only for symmetric matrices.

If 𝐴 is symmetric with distinct eigen values, eigen vectors are orthogonal.

That is 𝑣 ′ ≈ 𝑣1 = 1, 𝑣1 ′ ≈ 𝑣2 = 0 and so on.

Where 𝑣1 , 𝑣2 are eigen vectors of 𝜆1 , 𝜆2 .

517
Copyright © 2018, The Open University of Sri Lanka (OUSL)
Unit 14: Numerical Methods for solving ODES and PDES

With our basic assumption that the vectors are linearly independent, and
starting with an arbitrary vector 𝑥 (0) , we can construct 𝑥 (1) , 𝑥 (2) , … ,
𝑥 (𝑘) such that 𝑥 (𝑘+1) = 𝐴 𝑥 (𝑘)

It can be shown that, the quantity 𝑅𝑘 which is called the Rayleigh quatient
given by

(𝑥 (𝑘) ) 𝐴𝑥 (𝑘)
𝑅𝑘 = ′
(𝑥 (𝑘) ) 𝑥 (𝑘)

Tends to the dominant eigen value 𝜆1 as → ∞ . This is called the Rayleigh’s


principle.

(𝑥 (𝑘) ) 𝐴𝑥 (𝑘)
That is 𝑅𝑘 = ′ → 𝜆1 as 𝑘 → ∞
(𝑥 (𝑘) ) 𝑥 (𝑘)

When |𝑅𝑘+1 − 𝑅𝑘 | ≤ ∈ (where  is a very small number), take 𝑅𝑘 as


𝜆1 and 𝑥 (𝑘) as corresponding eigen vector

Learning outcomes

 At the end of this session, the learner will be able to perform the
following activities.
 derive and use power method to find dominant eigen value
 Use Raleigh quotient method to find dominant eigen value of a
symmetric matrix.
 find other eigen values of matrices

518
Copyright © 2018, The Open University of Sri Lanka (OUSL)

You might also like