From: walter_at_[hidden]
Date: 2001-11-21 05:51:49
--- In boost_at_y <mailto:boost_at_y>..., Peter Schmitteckert (boost) <
boost_at_s <mailto:boost_at_s>...> wrote:
> Dear Jörg Walter,
>
> On Tuesday 20 November 2001 16:31, walter_at_g <mailto:walter_at_g>...
wrote:
>
> > Thanks for the clarification. LAPACK functionality is out of scope
> > for ublas.
>
> In case you're intersted I could port my Jacobi eigen value
> routine to ublas.
Yes, I think it's important to verify the suitability of ublas for
implementing numerical methods.
> The method is not the most efficient but
> can be implemented in quite generic (templated) way.
> However, In order to do this I would need some extension to
> ubls, e.g. 'norm_type' , since the routine should honor the fact,
> that eigenvalues of hermitean matrices are real.
I already mentioned type_traits::norm_type in traits.h?
> In addition I could offer a sparse matrix diagonalization (Davidson
> algorithm) which is at least as efficient as Lanczos, but has
(almost) no
> problems with degeneries.
As your time permits.
> The most important question for me is: is ublas mature enough
> to use is in a new large scale number crunching project ?
I'm unsure about this. Mathias recently guessed, that ublas is in an
early beta test phase.
Regards
Joerg