Problem with rectangular matrices

[Lisandro Dalcin]

On Tue, May 26, 2009 at 2:31 PM, Matthew Knepley <knepley at gmail.com> wrote:
> On Tue, May 26, 2009 at 12:29 PM, Jed Brown <jed at 59a2.org> wrote:
>>
>> Matthew Knepley wrote:
>>
>> > It not computing Ab. Its computing Ab, A^2b, A^3b, etc.
>>
>> Of course, but that's all possible if A*b is computable.  Lisandro
>> demonstrated that A*x was fine, which it better be because of how x was
>> obtained.  Had he tried A*b, which is the first thing the KSP needs, the
>> problem would have been obvious.
>
> This is the whole point. b and x cannot have different local sizes in a
> Krylov
> method. It makes no sense.
>

It seems I should go to for some coffee. You are right. The actual
limitations are vec operations involving vecs with different local
sizes. Perhaps making KSP work with different parallel x-b layouts
could be enhanced to work (for example, by scatter x <-> x_aux on
entry (for init_guess!=0) and on exit), but likely this is not worth
the effort..



-- 
Lisandro Dalcín
---------------
Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
PTLC - Güemes 3450, (3000) Santa Fe, Argentina
Tel/Fax: +54-(0)342-451.1594