[thibault_pringuey]

Hello,

I am using OpenFOAM 1.5.x

In order to check the order of a numerical scheme, I am translating a sine field ( sin(pi * (pos().x+pos().y)) ) in a 2D cyclic domain. Such that at time = 1s the calculated field should overlap the initial field.

While this case runs fine on a single processor, in parallel (simple method; 4 processors: 2x2; original 0-halo decomposition) I notice a shift of the sine (~2 cell diagonals) at the end of the calculation. The shift of the field is forward, i.e. it presents the solution as it should look at a later time.

When using the n-halo parallel processing I developed, the shift is divided by 2. It is worth mentioning that for this 2x2 decomposition, using the n-halo decomposition effectively divide by 2 the number of processorXtoY boundaries per sub domain.

I would really appreciate your thoughts on this. I have attached a contour plot of the initial field in 2D and the plot over the diagonal for serial, 0-halo and n-halo calculations.

Many thanks in advance for your help.

With best wishes,


Thibault

[thibault_pringuey]

Solved it!

For some reason the boundaryField stored for processor patches (version 1,5) is the neighbourField and not the average of the internalField and the boundaryField as for the cyclic patches.