Re: Reported any bugs in C-LAPACK routine DSPEVX?

From: Javier Almeida (javieralmeida_at_fis.ucm.es)
Date: 10/25/04 

Date: 25 Oct 2004 03:13:19 -0700

Andreas Krebs <krebs@math.tu-cottbus.de> wrote in message news:<4178B06B.4070108@math.tu-cottbus.de>...
> Hi,
> I do not know much about CLAPACK since I prefere to declare
> and call the latest Fortran LAPACK routines directly. The only problem
> with this is that Fortran stores matrices in columns and C stores
> matrices in rows. So, you have to call the matrix routines of
> LAPACK exchanging 'n' by 't' and 't' by 'n'. The first thing I would
> do in your case is one of the following.
>
> 1. Expand the SP-matrix to a GE-matrix, compute the eigenvalues with
> dgeevx.
>
> 2. Call the dspevx fortran-routine directly. Drop me a line if you
> need an example.
>
> Regards, Andreas
>

Hello, thanks for your suggestions.
    The fact that fortran stores matrices column-wise and C row-wise
is no longer a problem for me since CLAPACK (and LAPACK++) also works
like fortran and I've get used to store matrices this way. I've used
before routine dgeevx but as far as you involve generic matrices in
your problem you have to be aware of the possibility of getting
complex eigenvalues/eigenvectors (sometimes with null precision
machine imaginary parts, but imaginary after all). So I think the
solution may come from your second suggestion, but I'm not really sure
I know how to link fortran routines directly from C++, maybe I should
include a header in my programs (with correct type conversions) and
link against lapack_F77.a ?.

         I would be greatly thankful if you could help me with an
example.

Best Regards, javi