[Octopus-devel] ARPACK

[Umberto De Giovannini]

Thank you Alberto,

I new that the biconjugate gradiet linear solver is already implemented.
Although, and probably because of deep ignorance, to me linear solver
and eigensolver are somehow disconnected.
Could you give me a hint on how to use the linear solver to diagonalize?


I feel that although slower that the other eigensolvers ARPACK seems
to be the fastest track toward a working implementation of complex
scaling.
Anybody against the return of ARPACK?


umberto




On Tue, Apr 24, 2012 at 12:48 PM, Alberto Castro <acastro at bifi.es> wrote:
> Hi Umberto,
>
> In fact, ARPACK was interfaced to octopus for some time. It was slower
> than the current methods, and therefore no attempt was made to
> maintain the interface. If you look in the svn tree far enough, it
> should be there. I don't remember any other reason why the inteface
> was removed, but maybe someone remembers better.
>
> In any case, I believe that biconjugate gradient is now implemented as
> one of the linear solvers in the linear solver module.
>
> Cheers, Alberto.
>
> 2012/4/24 Umberto De Giovannini <umberto.degiovannini at gmail.com>:
>> Hello everybody,
>>
>> as you probably remember from discussion related to some of my recent
>> commit I am implementing some exotic complex scaled DFT in octopus.
>> One of the main problem here is that the Hamiltonian turns out to be
>> non-Hermitian.
>> In this case the eigensolves already implemented are unsuited and
>> something more general is needed.
>> The choice is narrowed down to either biconjugate gradient or Arnoldi schemes.
>> I was thinking to implement an Arnoldi solver with ARPACK.
>> ARPACK has been out for a while now (and pretty much without any
>> update until recently http://forge.scilab.org/index.php/p/arpack-ng/)
>> but appears to be kind of a standard for what Arnoldi scheme is
>> concerned..
>> However there has been no attempt to use it with octopus.
>> So here is the question..
>> Is there a reason why I should not implement it in octopus?
>> Are there better solutions to my problem?
>>
>> ciao
>>
>> Umberto
>> _______________________________________________
>> Octopus-devel mailing list
>> Octopus-devel at tddft.org
>> http://www.tddft.org/mailman/listinfo/octopus-devel
>
>
>
> --
> ===============================
> Dr. Alberto Castro
>
> Instituto de Biocomputación y Física de Sistemas Complejos (BIFI),
> Edificio I+D,
> Mariano Esquillor Gómez s/n,
> 50018 Zaragoza, España.
>
> Tel:   +34 976761000 (extension 5354)
> Fax:   +34 976762990
> skype: alberto_c_barrigon
> email: acastro at bifi.es
> ===============================