eigs with sparse matrices

Hi all,

I'm working with very sparse positive-definite symmetric matrices, and I would like to solve for the eigenvectors/values (I use eigs of course) . Unfortunately, I cannot solve for the eigenvectors for many of the matrices I construct. I get an ARPACK error from the routine dneupd saying that no eigenvalues could be found to sufficient accuracy.

I was wondering two things:

1. is there a more reliable way (i.e. doesn't exit with errors) to estimate eigenvalues/eigenvectors with very sparse matrices?

2. is there a faster way to tell if you can solve for a matrix's eigenvalues? It takes a long time for eigs to decide that it cannot find any eigenvalues.

I would appreciate any pointers,tips, and/or advice. Thanks in advance.

Answer 1

[the cyclist]

It would be really helpful if you could provide an example that gives the error.

If the error is related to the condition number (singular matrix), then you may be able to use the cond() function.

Answer 2

[Andrew Knyazev]

try my http://www.mathworks.com/matlabcentral/fileexchange/48-lobpcg-m instead