[petsc-dev] Algebraic Multigrid

[Matthew Knepley]

On Mon, Jul 25, 2016 at 3:17 PM, Ari Rappaport <arir at vosssci.com> wrote:

> Hi there!
> I am a student intern at Voss Scientific, a contractor for national labs.
> I have been researching multigrid methods this summer. My office has an
> established particle code which uses petsc libraries for matrix solutions.
> There is an interest in using multigrid methods to precondition our
> matrices. We have attempted to use the Algebraic Multigrid Preconditioner
> on our field solver and it was quite effective with one processor. However,
> when we moved to four processors the PCAMG took about 1000 times as many
> iterations. I am wondering if there are any special considerations that
> need to be made for using AMG on multiple domains which in turn means
> multiple processors? Also, I understand there is also a PCMG option that
> takes advantage of the actual grid structure of the problem. I am also
> wondering if this could improve preconditioning.
>

1) For all convergence questions, we need to see the output of

  -ksp_monitor_true_residual -ksp_view -ksp_converged_reason

2) Are you solving an elliptic equation?

3) GAMG should give about the same convergence in parallel, but I do not
know what you used until I see -ksp_view output

   Matt


> Thanks,
> Ari
>



-- 
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-dev/attachments/20160725/5a62bd30/attachment.html>