[floquation]

Problem:
While simulating a case with dynamic mesh refinement, I have come across something peculiar. The solver was interDyMFoam.

The problem is that directly after the first dynamic mesh refinement, in the first PIMPLE iteration, the alpha field is calculated, U is calculated, p_rgh is calculated, phi is corrected and then when p_rgh is calculated in the second innerIteration, its initial residual is NaN. (The case crashes after this iteration output is shown.)
This is not a convergence issue.

Reproducibility:
The problem, in its most basic form, occurs with >1 outerIteration (at least 2 PIMPLE loops), presumably >1 innerIteration, and with relaxationFactors specified, e.g.:

relaxationFactors
{
    fields
    {
        p_rgh   1e-1;
        p_rghFinal 1;
    }
    equations
    {
        ".*" 1;
    }
}

The precise values do not matter: even "without underRelaxation" with urf=1, the problem occurs.
EDIT: under-relaxation for "equations" does not matter either. Only "fields" matter.

(I have only tried the simulation in parallel, but that cannot matter, right?)

Solution:
I have tried changing the solver/preconditioner for p_rgh: PCG+DIC, PCG+DILU, PCG+none, GAMG, ... It doesn't matter.
The problem disappears when nOuterIterations=1, and/or when the entry "relaxationFactors" is removed from fvSolution.
Therefore, the problem must be a combination between under-relaxation and OpenFoam treating the nonfinal PIMPLE loop differently than the final one.


Question: why?
I already "solved" my problem by eliminating the cause (i.e. by removing the relaxationFactors entry), but I'd like to understand why?
I know that "no under-relexation" (urf=1) still does something in OpenFoam; According to the user guide:

• : guaranteed matrix diagonal equality/dominance.

Does anyone know how this might affect a case with dynamic mesh refinement? Might it be a bug?

I should mention that dynamicRefineFvMesh cuts hexahedrons up in 8 smaller hexahedrons, which causes some hexahedrons to have more than 6 neighbours. This might affect the symmetry of the matrices?


Thanks in advance.