[petsc-users] DIVERGED_NANORINF for CG/Bjacobi for transient diffusion

[Barry Smith]

> On Mar 30, 2016, at 5:18 PM, Justin Chang <jychang48 at gmail.com> wrote:
> 
> Hi all,
> 
> I am getting this error:
> 
> Linear solve did not converge due to DIVERGED_NANORINF
> 
> when solving an FEM for transient diffusion using hexahedron elements. The problem is highly heterogeneous (i.e., dispersion tensor with varying cell-centered velocity) and normally CG/Bjacobi has done well for me. If I did CG/Jacobi, my solver would require nearly 7000 iterations, but would get the "expected" solution.
> 
> What does the above error mean? And what could I do to address it?

   It is sometimes useful in this case to run with -ksp_error_if_not_converged either with or without -start_in_debugger this gives you an exact stack trace of where the problem was originally detected. For example if Matt is right then it might show an error in MatLUFactorNumeric_SeqAIJ

   Barry

> 
> Thanks,
> Justin