[petsc-users] Composite shell preconditiner

[Jed Brown]

On Fri, Aug 17, 2012 at 6:48 AM, Alexander Grayver
<agrayver at gfz-potsdam.de>wrote:

> Hello petsc team!
>
> I would like to apply some sort of correction during Krylov process every
> N iterations.
> So I decided to use composite preconditioner like that:
>
> -pc_type composite -pc_composite_pcs shell,sor/icc/gamg
>
> I've started to test with the simplest shell preconditioner that does
> nothing:
>
> subroutine ShellPCApply(pc,x,y,ierr)
>   implicit none
>
> #include "finclude/petscsys.h"
> #include "finclude/petscvec.h"
>
>   PC,intent(in)        :: pc
>   Vec,intent(in)       :: x
>   Vec,intent(inout)    :: y
>
>   call VecCopy(x,y,ierr);CHKERRQ(**ierr)
> endsubroutine ShellPCApply
>
> And it turned out that this influences convergence by itself. In case:
> -pc_type composite -pc_composite_pcs shell,sor
> ---->
> Linear solve converged due to CONVERGED_RTOL iterations 125
>
> whereas
> -pc_type composite -pc_composite_pcs sor
> ----->
> Linear solve converged due to CONVERGED_RTOL iterations 35
>
> Why does it happen? What is the right way to go then?
> Thanks in advance.
>
> Just in case, here is -ksp_view for -pc_composite_pcs shell,sor :
>
> PC Object:() 1 MPI processes
>   type: composite
>   Composite PC type - ADDITIVE
>

See this *ADDITIVE*? You are preconditioning with I+P_{SOR}. You probably
want multiplicative.


>   PCs on composite preconditioner follow
>   ------------------------------**---
>     PC Object:    (sub_0_)     1 MPI processes
>       type: shell
>         Shell: no name
>       linear system matrix = precond matrix:
>       Matrix Object:       1 MPI processes
>         type: seqaij
>         rows=59049, cols=59049
>         total: nonzeros=741555, allocated nonzeros=767637
>         total number of mallocs used during MatSetValues calls =0
>           not using I-node routines
>     PC Object:    (sub_1_)     1 MPI processes
>       type: sor
>         SOR: type = local_symmetric, iterations = 1, local iterations = 1,
> omega = 1
>       linear system matrix = precond matrix:
>       Matrix Object:       1 MPI processes
>         type: seqaij
>         rows=59049, cols=59049
>         total: nonzeros=741555, allocated nonzeros=767637
>         total number of mallocs used during MatSetValues calls =0
>           not using I-node routines
>   ------------------------------**---
>   linear system matrix = precond matrix:
>   Matrix Object:   1 MPI processes
>     type: seqaij
>     rows=59049, cols=59049
>     total: nonzeros=741555, allocated nonzeros=767637
>     total number of mallocs used during MatSetValues calls =0
>       not using I-node routines
>
> --
> Regards,
> Alexander
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120817/61c65250/attachment.html>