[petsc-users] Infinite or not-a-number generated in norm!

[Barry Smith]

   I have added to Petsc-dev VecValidValues() that will check that input and output vector values for many commonly used routines such as PCApply() have only valid entries to simplify tracking down these kinds of problems.

    Barry

On Dec 30, 2011, at 8:52 AM, Jed Brown wrote:

> On Fri, Dec 30, 2011 at 08:39, Dominik Szczerba <dominik at itis.ethz.ch> wrote:
> I am getting the below error with fgmres. I was wondering how to start
> debugging the problem. All the values in the matrix and rhs seem fine.
> Any suggestions are very welcome.
> 
> I suspect your custom preconditioner is computing NaN for the call to the inner solver. Run with -fp_trap -start_in_debugger to see where NaN was first generated.