[petsc-users] PCKSP

[Matthew Knepley]

On Tue, Jun 19, 2012 at 12:13 PM, Margarita Satraki <
margarita.satraki at gmail.com> wrote:

> For my problem (incompressible nonlinear elasticity) pcksp seemed to work
> ok with gmres and even better with fgmres.
> The best option was to use LU as PC but this requires a lot of memory.
>

You can just use

  -ksp_type preonly
  -pc_type lu

to get a direct solution. Its not clear why you would want PCKSP.


> Why do you think that would not be a good option?
>

Incompressible elasticity has a constraint, so it is a saddle point system.
We would recommend using
PC_FIELDSPLIT, splitting into blocks for the displacement and pressure. You
can do this automatically
using -pc_fieldsplit_detect_saddle_point.

We would also proba bly recommend using AMG (ML, or GAMG, etc.) for the
elastic block. You have to provide
the near null modes (6 rot+trans), but there are examples of this from Mark
Adams and me.

    Matt


> Margarita
>
>
> On 19 June 2012 19:00, Matthew Knepley <knepley at gmail.com> wrote:
>
>> On Tue, Jun 19, 2012 at 11:54 AM, Margarita Satraki <
>> margarita.satraki at gmail.com> wrote:
>>
>>> Thank you very much Matt.
>>>
>>
>> I will just point out that this is unlikely to ever be what you want. Do
>> you have a particular solver
>> configuration in mind?
>>
>>     Matt
>>
>>
>>> Margarita
>>>
>>> On 19 June 2012 18:44, Matthew Knepley <knepley at gmail.com> wrote:
>>>
>>>> On Tue, Jun 19, 2012 at 11:41 AM, Margarita Satraki <
>>>> margarita.satraki at gmail.com> wrote:
>>>>
>>>>> If I understand correctly:
>>>>>
>>>>> For a system:
>>>>> M^{-1} A x = M^{-1} b
>>>>> we don't need to multiply M^{-1} A explicitly, but we solve M w = v
>>>>> whenever needed.
>>>>>
>>>>> So the Krylov method is used in order to solve that system, or
>>>>> equivalently to compute the vector M^{-1} v?
>>>>>
>>>>
>>>> Yes
>>>>
>>>>    Matt
>>>>
>>>>
>>>>>
>>>>> On 19 June 2012 17:54, Matthew Knepley <knepley at gmail.com> wrote:
>>>>>
>>>>>> On Tue, Jun 19, 2012 at 10:33 AM, Margarita Satraki <
>>>>>> margarita.satraki at gmail.com> wrote:
>>>>>>
>>>>>>> Hello,
>>>>>>>
>>>>>>> I have difficulty understanding how PCKSP works.
>>>>>>>
>>>>>>> From:
>>>>>>>
>>>>>>> http://www.mcs.anl.gov/petsc/documentation/tutorials/Columbia04/DDandMultigrid.pdf
>>>>>>> I understand that instead of using preconditioners, it uses Krylov
>>>>>>> methods for the ''inner solvers''.
>>>>>>>
>>>>>>> What are the ''inner solvers''? Is there some kind of a subsystem
>>>>>>> that is solved instead of applying a preconditioner?
>>>>>>>
>>>>>>
>>>>>> Nope, its jsut like a PC:
>>>>>>
>>>>>>   M^{-1} A x = M^{-1} b
>>>>>>
>>>>>> where now M^{-1} instead of being an LU solve, for instance, is a
>>>>>> Krylov solve.
>>>>>>
>>>>>>    Matt
>>>>>>
>>>>>>
>>>>>>> Many thanks,
>>>>>>>
>>>>>>> Margarita
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> What most experimenters take for granted before they begin their
>>>>>> experiments is infinitely more interesting than any results to which their
>>>>>> experiments lead.
>>>>>> -- Norbert Wiener
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> What most experimenters take for granted before they begin their
>>>> experiments is infinitely more interesting than any results to which their
>>>> experiments lead.
>>>> -- Norbert Wiener
>>>>
>>>
>>>
>>
>>
>> --
>> What most experimenters take for granted before they begin their
>> experiments is infinitely more interesting than any results to which their
>> experiments lead.
>> -- Norbert Wiener
>>
>
>


-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120619/04a53e8b/attachment.html>