Hi Ethan I just wondered why you used eigs instead of eigfact (or eig). The symmetric eigensolver in LAPACK can find a subset of the values and vectors. When the matrix is dense, I think this is faster than the iterative method used by ARPACK. Unfortunately, eigfact didn't had this possibility enabled, but I have opened a pull request with this functionality. Right now you cannot use eigs with a Symmetric type, but that should also be fixed soon.
2014-04-26 20:58 GMT+02:00 Steven G. Johnson <[email protected]>: > > > On Saturday, April 26, 2014 10:34:52 AM UTC-4, Ethan Anderes wrote: > >> My problem is sparse in the sense that the 2500x2500 matrices I'm working >> with are low rank. >> > Sparsity means that the matrix is mostly zeros (in which case you > sometimes use special algorithms). This has nothing to do with whether the > matrix is low rank. eye(n,n) is sparse but full rank, and ones(n,n) is > non-sparse but rank 1. > -- Med venlig hilsen Andreas Noack Jensen
