[petsc-users] SNESSetUp

[Matthew Knepley]

On Wed, Oct 9, 2013 at 6:08 PM, Mark F. Adams <mfadams at lbl.gov> wrote:

> >
> > Your Fortran is hurting my eyes :)
> >
> > Is solver%rVec2 == 0?
>
> Should be OK.  Here is a fuller listing appended.
>
> SNESSetUp seems wrong in that I don't see how SNESSetUp can work unless a
> user drills though the interface (or I'm missing some SMESSetXXX method or
> a side effect).  This code is in SNESSetUp:
>
>   ierr = SNESGetFunction(snes,&snes->vec_func,NULL,NULL);CHKERRQ(ierr);
>   if (snes->vec_func == snes->vec_sol)
> SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_IDN,"Solution vector cannot be
> function vector");
>   if (snes->vec_rhs  == snes->vec_sol)
> SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_IDN,"Solution vector cannot be right
> hand side vector");
>
>   if (!snes->vec_sol_update /* && snes->vec_sol */) {
>     ierr = VecDuplicate(snes->vec_sol,&snes->vec_sol_update);CHKERRQ(ierr);
>     ierr =
> PetscLogObjectParent((PetscObject)snes,(PetscObject)snes->vec_sol_update);CHKERRQ(ierr);
>   }
>
>
> It seems like first, the (snes->vec_rhs  == snes->vec_sol) test should be
> moved to SNESSolve where these are apparently set.
>

This is not good enough since the next lines will fail if snes->vec_sol is
NULL. I think we do not to substitute vec_func for vec_sol, but
we also need to move creation of the update vector since it must be the
size of the solution. Will do it now.

   Matt


> Second can we use snes->vec_func instead of snes->vec_sol here?  This
> would remove all dependance on vec_rhs and vec_sol in SNESSetup.
>
> Mark
>
>   !  Extract global and local vectors from DM; then duplicate for remaining
>   !     vectors that are the same types
>   call MatGetVecs(solver%KKTmat,solver%xVec2,solver%bVec2,ierr)
>   call VecDuplicate(solver%bVec2,solver%rVec2,ierr)
>
>   call SNESSetDM(solver%snes,solver%da,ierr)
>
>   ! call SNESSetApplicationContext(solver%snes,solver,ierr)
>   ! call SNESSetApplicationContext(solver%snes,grid,ierr)
>
>   !  Set function evaluation routine and vector
>   call SNESSetFunction(solver%snes,solver%rVec2,FormFunction,solver,ierr)
>
>   call
> SNESSetJacobian(solver%snes,solver%KKTmat,solver%KKTmat,FormJacobian,solver,ierr)
>
>   ! - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
>   !  Customize nonlinear solver; set runtime options
>   ! - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
>   !  Set runtime options (e.g., -snes_monitor -snes_rtol <rtol> -ksp_type
> <type>)
>   call SNESSetFromOptions(solver%snes,ierr)
>
>   call SNESSetUp(solver%snes,ierr) ! pre setup, same as (old) KSP
>
>
>


-- 
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/20131010/39388046/attachment.html>