[petsc-users] More about SNESSetLagPreconditioner

[Matthew Knepley]

On Thu, Oct 4, 2012 at 12:07 AM, Gong Ding <gdiso at ustc.edu> wrote:

> call SNESSetLagPreconditioner before SNESSolve is useless.
> It seems SNESSetLagPreconditioner does not reset the internal information.
>
>

I am looking directly at the code:


http://www.mcs.anl.gov/petsc/petsc-dev/src/snes/interface/snes.c.html#SNESSetLagPreconditioner

It definitely changes the "internal information".

   Matt


> If you are calling SNESSolve(), just call SNESSetLagPreconditioner() again
> before each solve.
>
>
> On Wed, Oct 3, 2012 at 10:25 PM, Gong Ding <gdiso at ustc.edu> wrote:
>
>> Only implemented this strategy in TS is not enough, i.e. DC sweep (each
>> DC point requires nonlinear iteration) may also need it.
>> Will you implement it into the SNES and send me the patch?
>> For example, a flag -3 to SNESSetLagPreconditioner.
>> At the first ineration of each SENSSolve call, the PC will be built and
>> never change in the same SENSSolve iteration.
>> I'd like to try it asap since this strategy improves simulation speed
>> significantly.
>>
>> Gong Ding
>>
>>
>>
>>
>>
>> I've been meaning to implement this for a while. You can put custom SNES
>> control into a pre-stage or pre-step hook, but this particular case should,
>> in my opinion, be provided as a TS option.
>>
>> On Wed, Oct 3, 2012 at 11:00 AM, Gong Ding <gdiso at ustc.edu> wrote:
>>
>>> Hi,
>>> I am doing a nonlinear transient simulation. For each time step, a
>>> SNESSolve is called.
>>> I want to do a LU factorization at the first Newton iteration of EACH
>>> time step, and for the following Newton steps (in the same time step), no
>>> factorization any more. For the next time step, the LU factorization should
>>> be done again.
>>>
>>>
>>> SNESSetLagPreconditioner(-1) seems just skips all the following
>>> factorization, even in the different time step.
>>> How can I do?
>>>
>>>
>>> Gong Ding
>>>
>>>
>>>
>>
>


-- 
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/20121004/228b6cb9/attachment.html>