Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

78 matrices and system of linear equations

3.5 finding dominating eigenvalue

The dominating eigenvalue of a square matrix is the eigen-


value with the largest absolute value. In other words, if A is
a square n ⇥ n matrix, and li (i = 1, 2 . . . n) are eigenvalues
of A such that,

| l1 | > | l2 | | l3 | ··· |ln |


then l1 is the dominating eigenvalue.

3.5.1 Power method

In this section we will discuss the power method for finding


the dominating eigenvalue of a given matrix. Then we will
prove the convergence of the method for diagonalizable
matrices.
Let v(0) be any random normalized vector in R n . Then the
power method is defined by
8
( k 1)
<v(k) = Av
>
k Av(k 1) k 2 k = 1, 2, . . . (29)
>
:l(k) = (v(k) ) T Av(k)

Pseudocode is available in Algorithm 15.

3.5.2 Convergence of power method

We assume that the given matrix A is diagonalizable, with


eigenvalues li (i = 1, 2 . . . n) satisfying

| l1 | > | l2 | | l3 | ··· | l n |.
Associated eigenvectors are xi (i = 1, 2 . . . n). Thus l1 is the
dominating eigenvalue of A with associated eigenvector x1 .
3.5 finding dominating eigenvalue 79

Algorithm 15 Power Method with Scaling


1: procedure PowerMethod(A, tol, Nmax )
2: Choose a non-zero initial vector v(0) 2 R n
3: k=1
4: l (0) = 0
5: while k  Nmax do
6: Compute w(k) = Av(k 1)
7: Set l(k) = (v(k 1) ) T w(k)
(k)
8: Normalize v(k) = kw w(k) k
9: if |l(k) l(k 1) | < tol then
10: return l(k) , v(k)
11: end if
12: k = k+1
13: end while
14: return "Power Method did not converge after Nmax
iterations"
15: end procedure
80 matrices and system of linear equations

Suppose x0 is the initial guess. Then v0 can be written as

v0 = a1 x1 + a2 x2 + · · · + a n x n .

We assume, a1 6= 0. Now note that,

Ax0
v1 = ,
k Ax0 k
Ax1 AAx0 A2 x 0
v2 = = =
k Ax1 k k Ax0 kk kAAx
0
k k A2 x 0 k
Ax0 k

...
Ak x0
vk =
k Ak x0 k
Now

A k v0 = A k ( a1 x1 + a2 x2 + · · · + a n x n )
= l1k a1 x1 + l2k a2 x2 + · · · + lkn an xn
✓ ◆k ✓ ◆k !
l 2 l n
= l1k a1 x1 + a2 x2 + · · · + an xn
l1 l1

Therefore,
⇣ ⌘k ⇣ ⌘k
l2 ln
A k v0 a1 x1 + l1 a2 x2 + · · · + l1 an xn
vk = =± ⇣ ⌘k ⇣ ⌘k
k A k v0 k l2 ln
k a1 x1 + l1 a2 x2 + · · · + l1 an xn k

Taking limit we get,


x1
lim vk = ± .
k!• k x1 k

Thus, vk converges to the dominating eigenvector. Once we


know the eigenvector, it is easy to get the corresponding
3.5 finding dominating eigenvalue 81

eigenvalue. If v is an eigenvector of A, then Av = lv. Then


we have,

v T Av
l= ,
vT v

which is also known as Rayleigh quotient.

Therefore, in power method the approximated dominating


eigenvalue can be calculated as

T
(k) (vk ) Avk T
l = T
= (vk ) Avk .
(vk ) vk

You might also like