Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Matrix Computations

(16B1NMA533)
Semester: Vth

Department of Mathematics
Jaypee Institute of Information Technology
Noida

1
Contents of the lecture

Module No. Title of Module CO


5 Householder’s Method 5

 Householder’s Method

2
Householder’s method
 Householder’s transformation is to find a symmetric
tridiagonal matrix T that is similar to a given symmetric
matrix A.

Definition: Let w  Rn with wT w  1 . Then the nn matrix


P  I  2 wwT

is called a Householder transformation (or Householder


matrix).
Theorem: If P  I  2wwT is a householder’s matrix, then P is
symmetric and orthogonal. So P-1= PT = P.
Proof: The symmetry is from
PT  ( I  2wwT )T  I T  (2wwT )T
 I  2( wT )T ( w)T  I  2 wwT
P

The orthogonality is from


PPT  ( I  2 wwT )( I  2 wwT )
 I  4 wwT  4 wwT wwT
 I  4 wwT  4 w( wT w) wT
 I  4 ww2  4 wwT  I
 PPT  PT P  I .
Some basics about the Householder matrix:
(a) P has eigenvalues 1 (multiplicity n-1) and -1.

Proof: Here, Pw  w
As, Pw  ( I  2ww )w  w  2ww w  w  2w.1  w.
T T

For any x orthogonal to w


 wT x  0
 Px  ( I  2 wwT ) x
 Px  x  2wwT x  x  0  x.
 Px  x.
These imply that P has eigenvalues 1 (multiplicity n-1) and -1.

Point to remember:
 Euclidean norm: The Euclidean-norm of vector 𝑥 is given by
1/2
 
|| x ||2    | xi |2  as x  R n .
 i 
|| x ||2 || y ||2 .
 Two vectors x and y of Rn are of same length if
(b) For any two distinct vectors x and y in Rn with same length, the Householder matrix

( x  y)
P  I  2 wwT , w
|| x  y ||2
satisfies Px  y.
Proof: To prove: Px  y,
n
Since, two vectors x and y of R are of same length if

|| x ||2 || y ||2 .


and xT y  y T x and xT x  y T y .

x y
 ( xT x  xT y  y T x  y T y  2 xT x  2 y T x )  0.
|| x  y ||2
2

x y
 (( x  y )T ( x  y )  2( x  y )T x)  0
|| x  y ||2
2

x y
 (|| x  y ||22 2( x  y )T x)  0
|| x  y ||2
2

( x  y )( x  y )T x
 ( x  y)  2  0.
|| x  y ||22
x  y ( x  y )T
 x2 x y
|| x  y ||2 || x  y ||2
 Px  y.
Example: If x = (3, 0, 4)T, choose y  (|| x ||2 ,0,0)T  (5,0,0)T . Verify that

Px  y.

Solution: Let w  ( x  y )/ || x  y ||2 .


T
(2, 0, 4)T 2 4 
w  , 0, 
4  16  20 20 

 Px  Ix  2 wwT x
T
 2 4   2 4 
 Px   3, 0, 4   2  
T T
, 0,   , 0,  3, 0, 4
 20 20   20 20 
T
 2 4  10
 Px   3, 0, 4   2. 
T
, 0,  .
 20 20  20
 40
T
80 
 Px   3, 0, 4  
T
, 0, 
 20 20 
 Px   5, 0, 0   y
T

 Px  y.
Example: If x = (1, 1, 3, 0, 4)T, find a Householder matrix to change x to the form y
= (1, 1, *, 0, 0).
Solution: To make || x ||2 || y ||2 , y  (1,1,5,0,0).

|| x ||2  12  12  32  02  42  27
and
|| y ||2  12  12  *2  02  02  2  *2

Here,
for || x ||2  || y ||2
 y  (1,1,5, 0, 0)T .

There exist more possible values for y.

Choose w  ( x  y )/ || x  y ||2 and P  I  2 ww as before.


T

You can easily verify that Px  y.


Important Point:
 For a Householder matrix P, if Px  y, then xT P  yT since P is symmetric.

x  R n , let y  ( || x ||2 , 0,...., 0)  Rn , then || x ||2 || y ||2 .


T
 For any
There is a Householder matrix P such that Px  y . This is the key step of
Householder method. You may pick “+” or “-” in y.
Practice Questions:
Ques 1: If x = (2, 4, 3, 0, -4)T, find a Householder matrix to change x to the
form y = (1, 1, *, 0, 0) and verify that Px  y.

Ques 2: If x = (1, 4, 2, 0)T, find a Householder matrix to change x to the


form y = (2, 0, *, 0) and verify that Px  y.
References
1. Bronson, R., Matrix Methods an Introduction, Academic Press, 1991.
2. Wilkinson, J. H., Householder's method for symmetric matrices, Numerische
Mathematik, 354-361, 4(1), 1962.

You might also like