[Skaiwalker]

Hello forum,

Problem: I want to model the melting process of magnesium ingots in a melting furnaces. Thus I've decided to use OF8 with "buoyantPimpleFoam" and fvOptions "solidificationMeltingSource". After a few timesteps either for Ux, or p_rgh the residuals won't fall under 1e-6 but will oscillate around a constant value in the magnitude of e-5 to e-4 . This happens at first for one timestep, then it converges well again for several time steps. As time progresses the the number of non converging timesteps increases gradually but there are always converging timesteps in between.

The results look not that bad as far as I can tell, nothing explodes or unexpected happens, but from a mathematical standpoint this introduces several sources of errors, if I understand the numerics correctly:

1. Not well converged results are physically more or less inaccurate
2. going on from an inaccurate result results in even more inaccurate results
3. the last pimple step is not computed with an underrelaxation of 1 introducing time continuity errors

Questions: (actually a lot )

Is it possible to converge in every timestep in this case, or am I hitting rock bottom on the residuals? How bad is the error growing through the above mentioned error sources? How can one ensure that pimple doesn't produce time step incontinuities?

Also my underrelaxation seems pretty odd to me (U 0.6, p 0.05) but it works to some extent and I've run a ton of cases to figure this out. Is it adequate to underrelax one parameter that much if it makes trouble? In the end it converges in the "not broken" timesteps.

Additional Info:
I've attached a screenshot on an arbitrary timestep, the x axis is lying. Also the case is attached. Frist run it with Allrun, then runSimulation.sh is enough after the mesh is decomposed. Post processing is done by "gnuplot utilities/plotResiduals_onTheFly.gnu".

Thank you for your help, I appreciate every single bit of info I can gather on OpenFOAM, its so amazing (when it doesn't drive you crazy )

PS: Be careful with ctrl-c on the run-scripts, because the mpi initialized solver processes will not be terminated. I didn't find a workaround up until now.