Mass Lumping HPC

Question

Hi everyone,

I am running simulations on HPC and I noticed that when I run with mass_lumping=1 everything is fine, but when I run with mass_lumping=0 the simulation diverges on the very first step (L5 : parabolic PDE (PETSc) solver diverged. Reason: indefinite preconditioning.). When I run the same code on my PC, it works with both mass_lumping on and off.

I installed openCARP on HPC by using Spack.

Do you know where I should look for the issue? I guess it could relate to some solver settings, but I don't know where to start investigating this problem.

Best,

Jakub

Answer 1

Hi,

In my experience these kind of issues originate from bad mesh quality, i.e. collapsed elements.

The parabolic system is more resistant, as it is diagonal dominant (and even more so with mass lumping), but then solvability problems pop up when changing parallel layout or region parameters. One thinks that the issue was introduced by the change in parameters, while it has been present all the time.

To test my hypothesis, I would suggest running "meshtool query quality" on your input mesh, to sanity check it. The max. quality metric should not be over 0.98.

Best wishes, Aurel