[PanPeter]

Hi,

I am the facing the following problem: I use a modified simpleFoam solver to have a scalar field convected with the flow field. A membrane boundary condition is defined to allow some flux between two channels separated by this membrane boundary. The flux is dependent on the scalar fields on both sides of that membrane. The solver works fine and I see my flux assume a steady-state value as the two flow domains converge.

Decomposing the mesh and running it in parallel works fine too. But I found out that the steady-state values differ for different decompositions, see attached figure. The membrane boundary is flat and defined in the xy plane. I compare a serial case with three parallel cases with decompositions n = (2, 2, 1), n = (2, 1, 1) and n = (1, 2, 1).

Non of the cases is completely converged, due to my conservative choice of final residuals, so results after one hour of iteration time. A pattern can be recognized from the graph. The cases cutting through the xz plane (JwN221 and JwN211) are performing a lot different compared to the cases JwSerial and JwN121.

Now my question: I don't understand why the decomposition makes such a severe difference in the results. Can somebody shed some light on this?

cheers,
Fynn