Why does Matlab give different eigenvalues for the same matrix?

Dear all,

I am computing the eigenvalues for a generalized problem with shift inverted using the function eigs. The problem is that if I repeat the calculation with the same matrices I obtain different eigenvalues. For example: -5.87e-4 + i 1.677622 and 1.105e-3 + i 1.679422. What does it depend on? Have you ever experienced such a behaviour?

Thaks,

Nora

Answer 1

[John D'Errico]

You are using eigs. For some eigenvalues, those estimates can vary, because eigs uses a random start.

If you insist on getting a deterministic result, you will need to either

1. Use eig, which does not use a random start.

2. Use eigs, but set v0. From the help, we see:

 eigs(A,K,SIGMA,OPTS) and eigs(A,B,K,SIGMA,OPTS) specify options:

    ...
    OPTS.v0: starting vector [N-by-1 vector | {randomly generated}]
    ...

[Nora]

Yes, I was thinking about that. I'll try to assign a starting vector v0.

Answer 2

[Thorsten]

Are you 100% sure that you used the same matrix for your computations?

[Nora]

Yes, I am.

[Thorsten]

Could you save the matrix as matfile and post, including the code that reproduces the different eigenvalues?

[Nora]

It is a very big matrix. I'll try to post it.