Re: unsymmetric eigenvalue problem

[Patrick Alken <patrick dot alken at colorado dot edu>]

Hello all,

  I have implemented some convergence tricks suggested by
Numerical Recipes and have attached the latest patch.
This should be the last patch for a while, since I have now
tested it on randomly generated 1000x1000 matrices and compared
the results with octave and find perfect agreement.

  In terms of speed, my code runs slower than octave's eigenvalue
solver (which I think is essentially the fortran code from LAPACK)
but I only notice the slowdown at around N = 500 matrices. There
is certainly room for improvement here but I believe the priority
for now is to get an eigenvalue solver into GSL. I will continue
trying to optimize the code but it will involve studying the
LAPACK source for a while.

  I haven't gotten much feedback so far so I hope others will find
this useful..

Patrick Alken

Attachment: gsl.unsymm.patch
Description: Text document