[petsc-users] questions about KSP with-pc_type hypre

Tue Oct 8 09:26:08 CDT 2013

> Subject: Re: [petsc-users] questions about KSP with-pc_type hypre
> From: bsmith at mcs.anl.gov
> Date: Mon, 7 Oct 2013 16:10:10 -0500
> CC: mfadams at lbl.gov; petsc-users at mcs.anl.gov
> To: pengxwang at hotmail.com
> 
> 
> On Oct 7, 2013, at 3:28 PM, Roc Wang <pengxwang at hotmail.com> wrote:
> 
> > Sorry for the mis-info. 
> > 
> > Yes, I have Dirichlet BCs. The smaller round plate on the left side and the rectangular plane on the right size are the Dirichlet BCs. And the inside the hole of the rectangular plane is Neumann BCs.  I am just wondering the Neumann BCs of the hole have some effect which lead to divergence.  
> 
>    "Some" Neumann boundary conditions shouldn't have much effect. Are you sure the Neumann BCs are applied correctly and that Neumann BCs are not being applied everywhere?  If Neumann BCs are applied almost everywhere on the boundary that can affect convergence.

Yes, the Neumann BCs are applied correctly, since the solution is correct if the KSP solver runs without -pc_type hypre option. On the right side of the domain, the Neumann BCs are only within a hole in the center of the plane.  
> 
>    What happens with -pc_type lu?  Does it solve the linear system and give back answers you expect?

No, it doesn't work if I use -pc_type lu. The error information shows:

[0]PETSC ERROR: --------------------- Error Message ------------------------------------
[0]PETSC ERROR: No support for this operation for this object type!
[0]PETSC ERROR: Matrix format mpiaij does not have a built-in PETSc LU!
[0]PETSC ERROR: ------------------------------------------------------------------------
[0]PETSC ERROR: Petsc Release Version 3.3.0, Patch 6, Mon Feb 11 12:26:34 CST 2013 
[0]PETSC ERROR: See docs/changes/index.html for recent updates.
[0]PETSC ERROR: See docs/faq.html for hints about trouble shooting.
[0]PETSC ERROR: See docs/index.html for manual pages.
[0]PETSC ERROR: ------------------------------------------------------------------------
[0]PETSC ERROR: ./x.r on a arch-linu named athena.coeaccess.psu.edu by pwang Tue Oct  8 10:23:50 2013
[0]PETSC ERROR: Libraries linked from /home/pwang/soft/petsc-3.3-p6/arch-linux2-c-debug/lib
[0]PETSC ERROR: Configure run at Fri Oct  4 15:18:39 2013
[0]PETSC ERROR: Configure options --with-cc=gcc --with-fc=gfortran --download-f-blas-lapack --dtime elap.:0     0.0000  1000000  
ownload-mpich --with-cxx=g++ --download-hypre=1 --download-ml=1 --download-hdf5=1
[0]PETSC ERROR: ------------------------------------------------------------------------
[0]PETSC ERROR: MatGetFactor() line 3876 in src/mat/interface/matrix.c
[0]PETSC ERROR: PCSetUp_LU() line 133 in src/ksp/pc/impls/factor/lu/lu.c
[0]PETSC ERROR: PCSetUp() line 832 in src/ksp/pc/interface/precon.c
[0]PETSC ERROR: KSPSetUp() line 278 in src/ksp/ksp/interface/itfunc.c
[0]PETSC ERROR: KSPSolve() line 402 in src/ksp/ksp/interface/itfunc.c
[0]PETSC ERROR: LinearSolver() line 264 in "unknowndirectory/"src/solver.cpp
rank 2 in job 959  athena.coeaccess.psu.edu_42121   caused collective abort of all ranks
  exit status of rank 2: killed by signal 6 
rank 1 in job 959  athena.coeaccess.psu.edu_42121   caused collective abort of all ranks
  exit status of rank 1: killed by signal 6 

How can I make -pc_type lu work? Thanks a lot

> 
> 
>    Barry
> 
> > 
> > I will try gamg as your suggestion. Thanks.
> > 
> > 
> > Subject: Re: [petsc-users] questions about KSP with-pc_type hypre
> > From: mfadams at lbl.gov
> > Date: Mon, 7 Oct 2013 16:16:30 -0400
> > CC: petsc-users at mcs.anl.gov
> > To: pengxwang at hotmail.com
> > 
> > I assume this is a Lapacian and its not a pure Neumann problem.  I did not quite follow you description.  Do you have some Dirichlet BCs?
> > 
> > Hypre is a very robust solver, for a scalable solver.  ILU, the default PC, is very robust but slow (that's why its the default).  You could try gamg, another AMG solver to get another data point.
> > 
> > On Oct 7, 2013, at 3:08 PM, Roc Wang <pengxwang at hotmail.com> wrote:
> > 
> > Hello,
> > 
> >    I am trying to solve a linear system with KSP for a 3-D domain.  The computational domain is a cubic space with Neumann condition on the lateral sides( y, and z directions).  A smaller round plate is on the left side in x direction.  The right side is a rectangular plane either with or without a hole.  In testing studies, the KSP solver works fine both with and without -pc_type hypre runtime option for the domain with a whole plane(without a hole) on the right, but diverges for the case of a plane with a hole with -ps_type hypre option. 
> > 
> >   My question is whether the hypre pre-conditioner is sensitive to the geometry boundary?  If the answer is yes, is there any other pre-conditioner available for my particular case(the domain with a plane which with a hole on the right side) to reduce the number of iterations? Thanks in advance.
> > 
> > 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20131008/0e9c63c6/attachment.html>