Anybody here use Lapack?

[Stimpson J. Cat]

I have been trying to make some C++ class wrappers for various functions from the Lapack Fortran library, and I am having a problem. The sgeev() and dgeev() functions for computing eigenvalues and eigenvectors of a matrix seem to be giving incorrect results for matrices with both real and complex eigenvalues.

Anybody have any experience with Lapack?


Dr. Stupid

[69dodge]

no experience with LAPACK. But Google found this:

http://www.netlib.org/lapack/lug/node135.html

Maybe that will help?

Also, I know that FORTRAN and C store two-dimensional arrays differently, so the routines you're calling might think of as columns what you think of as rows, and vice versa.

[Stimpson J. Cat]

Yeah, I checked all of those things. And the issue with C and Fortran handling 2 dimensional arrays differently isn't even an issue for me. I'm not actually using 2 dimensional arrays. Instead my Matrix class uses a 1 dimensional array which can then be indexed in either row-major or column-major format. One view simply gives the transpose of the other. And I already made sure I wasn't giving the function the transpose by mistake.

Based on some things I have found on the web, I am inclined to think that it is a bug in my Lapack library, due to it being compiled with compilor optimizations turned on. Unfortunately I don't have any way to compile the library myself.

Dr. Stupid