[petsc-users] KSP diverging

[Mohamad M. Nasr-Azadani]

Hello,

I am using GMRES + Hypre (BoomerAMG) as the preconditioner to solve the
Poisson equation for pressure.
Pressure linear system has a nullspace (Neumann boundary condition
everywhere).
After a long time of time integration of my CFD simulations, all of the
sudden, my code starts diverging, i.e. delta_t = 0.
I am still trying to find what exactly the problem is, but what I have
faced is when I solve the Poisson equation with the given KSP and PC, the
solution does not converge due to
KSP_DIVERGED_DTOL

Any hints? I have tried to fix the pressure at one node and also use the
MatNullSpace to handle that, no luck and both cases diverge.
The interesting thing is that when I use GMRES and other preconditioners,
e.g. PCASM or PCJACOBI, the linear system does not converge at all. It
reaches the maximum number of iterations (iter=10000) which technically
should not since the linear system is not that big. There is only 1.6
million grid points in the domain.

Any hints on what is happening here or how I could narrow it down? I know
that we had an early discussion on the convergnce for the incompressible
CFD problems and boomerAMG preconditioning, but it seems that for this
case, that is the only one that does not blow up for most of the time
integration I have.

I am running it on 24 processors and 1.6 million grid points.

Thanks in advance,
Best,
Mohamad
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20111222/1588efaa/attachment.htm>