Boost logo

Ublas :

Subject: Re: [ublas] lapack eigenvalues sparse matrix
From: Kraus Philipp (philipp.kraus_at_[hidden])
Date: 2010-07-16 16:45:23

Hi Rob,

thanks for the hint with the "s". I've found it also.
I change my matrix data to full dense matrix, it is a little bit
clumsy, but it works
I hope the support for ARPACK comes into the bindings.



Am 16.07.2010 um 22:34 schrieb Rob Patro:

> Phil,
> LAPACK does not handle sparse matrices. As far as I remember, the
> 's' in sgeev has to do with the precision of the matrix scalar
> components; it does not have to do with sparsity. To perform an
> efficient eigendecomposition on sparse matrices, one needs to
> consider completely different algorithms than those used in LAPACK.
> For example, the ARPACK library provides for an efficient
> eigendecomposition of sparse matrices of different varieties. There
> was talk recently on this list of wrapping uBLAS matrices for
> consumption by ARPACK.
> Cheers,
> Rob
> On Fri, Jul 16, 2010 at 10:19:47PM +0200, Kraus Philipp wrote:
>> Hello,
>> I use the boost/numeric/bindings/lapack/driver/geev.hpp for
>> calculate eigenvalues and -vectors for dense matrix data. How can I
>> use it for sparse matrix data? I think the LAPACK call should be
>> sgeev.
>> Thanks
>> Phil
>> _______________________________________________
>> ublas mailing list
>> ublas_at_[hidden]
>> Sent to: rob.patro_at_[hidden]
> _______________________________________________
> ublas mailing list
> ublas_at_[hidden]
> Sent to: philipp.kraus_at_[hidden]