[petsc-users] 2 level schur

[Matthew Knepley]

On Thu, Mar 20, 2014 at 6:20 PM, Luc Berger-Vergiat <lb2653 at columbia.edu>wrote:

>  Hi all,
> I am solving a four field problem using two Schur complements. Here are
> the arguments that I usually pass to PETSc to do it:
>
> -ksp_type gmres -pc_type fieldsplit -pc_fieldsplit_type schur
> -pc_fieldsplit_schur_factorization_type full
> -pc_fieldsplit_schur_precondition selfp -pc_fieldsplit_0_fields 2,3
> -pc_fieldsplit_1_fields 0,1 -fieldsplit_0_ksp_type preonly
> -fieldsplit_0_pc_type fieldsplit -fieldsplit_0_pc_fieldsplit_type schur
> -fieldsplit_0_pc_fieldsplit_schur_factorization_type full
> -fieldsplit_0_pc_fieldsplit_schur_precondition selfp
> -fieldsplit_0_fieldsplit_Field_2_fields 2
> -fieldsplit_0_fieldsplit_Field_3_fields 3
> -fieldsplit_0_fieldsplit_Field_2_ksp_type preonly
> -fieldsplit_0_fieldsplit_Field_2_pc_type ilu
> -fieldsplit_0_fieldsplit_Field_3_ksp_type preonly
> -fieldsplit_0_fieldsplit_Field_3_pc_type jacobi -fieldsplit_1_ksp_type
> preonly -fieldsplit_1_pc_type lu -malloc_log mlog -log_summary time.log
>
> One issue with this is that when I change
> -fieldsplit_0_fieldsplit_Field_2_fields 2 to
> -fieldsplit_0_fieldsplit_Field_2_fields 3 it is ineffective, as if PETSc
> automatically assign IS 2 to Field 2 even though it is not what I want.
> Is there a way to pass the arguments correctly so that PETSc goes about
> switching the IS set of -fieldsplit_0_fieldsplit_Field_2 and
> -fieldsplit_0_fieldsplit_Field_3?
> This is crucial to me since I am using the selfp option and the matrix
> associated to IS 3 is diagonal. By assigning the fields correctly I can get
> an exact Schur preconditioner and hence very fast convergence. Right now my
> convergence is not optimal because of this.
>

I believe the inner Schur field statements should not be using the original
numbering, but the inner numbering, after they have been reordered.

   Matt


> Thanks!
>
> Best,
> Luc
>



-- 
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/20140320/bda27f02/attachment.html>