15.1. THE POWER METHOD FOR EIGENVALUES 413

and so( n

r1

)/(n+1

r1

)≊ λ 1

sn+1. But limn→∞

( nr1

)/(n+1

r1

)= 1 and so, for large n it must be the case

that λ 1 ≊ sn+1.This has proved the following theorem which justifies the power method.

Theorem 15.1.1 Let A be a complex p× p matrix such that the eigenvalues are

{λ 1,λ 2, · · · ,λ r}

with |λ 1|>∣∣λ j∣∣ for all j ̸= 1. Then for x a given vector, let

y1 =Axs1

where s1 is an entry of Ax which has the largest absolute value. If the scalars {s1, · · · ,sn−1}and vectors {y1, · · · ,yn−1} have been obtained, let

yn ≡Ayn−1

sn

where sn is the entry of Ayn−1 which has largest absolute value. Then it is probably thecase that {sn} will converge to λ 1 and {yn} will converge to an eigenvector associatedwith λ 1. If it doesn’t, you picked an incredibly inauspicious initial vector x.

In summary, here is the procedure.

Finding the largest eigenvalue with its eigenvector.

1. Start with a vector, u1 which you hope is not unlucky.

2. If uk is known, uk+1 =Auksk+1

where sk+1 is the entry of Auk which has largest absolutevalue.

3. When the scaling factors sk are not changing much, sk+1 will be close to the eigen-value and uk+1 will be close to an eigenvector.

4. Check your answer to see if it worked well. If things don’t work well, try anotheru1. You were miraculously unlucky in your choice.

Example 15.1.2 Find the largest eigenvalue of A =

 5 −14 11−4 4 −43 6 −3

 .

You can begin with u1 = (1, · · · ,1)T and apply the above procedure. However, you canaccelerate the process if you begin with Anu1 and then divide by the largest entry to get thefirst approximate eigenvector. Thus 5 −14 11

−4 4 −43 6 −3

20 1

11

=

 2.5558×1021

−1.2779×1021

−3.6562×1015



15.1. THE POWER METHOD FOR EIGENVALUES 413and so (P/E) & +. But lim, <0 (Y/R) = | and so, for large n it must be the casethat A, = Sntl-This has proved the following theorem which justifies the power method.Theorem 15.1.1 Let A be a complex p x p matrix such that the eigenvalues are{A1,A2,°°° Ar}with |A,| > |A,| for all j A 1. Then for x a given vector, letAXy= _S|where s, is an entry of Ax which has the largest absolute value. If the scalars {s1,--+ ,5,—1}and vectors {Y1,°-* ,Yn—1} have been obtained, let_— AYn-1Yn = Snwhere S, is the entry of Ay,_, which has largest absolute value. Then it is probably thecase that {s,} will converge to 4, and {y,} will converge to an eigenvector associatedwith A. If it doesn’t, you picked an incredibly inauspicious initial vector x.In summary, here is the procedure.Finding the largest eigenvalue with its eigenvector.1. Start with a vector, w; which you hope is not unlucky.2. If ug is known, ugi1 = oan where 5,1 is the entry of Au, which has largest absolutevalue.3. When the scaling factors s, are not changing much, s;; will be close to the eigen-value and u;z+; will be close to an eigenvector.4. Check your answer to see if it worked well. If things don’t work well, try anotheru,. You were miraculously unlucky in your choice.5 —-14 11Example 15.1.2 Find the largest eigenvalue of A= —4 4 —43 6 -3You can begin with u; = (1,---, 1)" and apply the above procedure. However, you canaccelerate the process if you begin with A” wu; and then divide by the largest entry to get thefirst approximate eigenvector. Thus5 -14 ll 1 2.5558 x 10?!4 4 -4 1 |=] —1.2779 x 10?!3 6 3 1 —3.6562 x 10!5