Closed kevinstratford closed 1 month ago
I've relaxed the constraint that two cells are required in a narrow channel. There is an example at
tests/regression/d3q27/serial-elip-s10.inp
This should fix the issue in the case of two narrow directions with walls. However, if one narrow direction is open, something different will be required (pending...).
I am going to leave the issue of the periodic dimension alone for the time being, as it should always be possible to extend the system to a size sufficient to satisfy the constraints of the cell list.
Summary
The cell list restrictions related to the particle extent cf the local domain size are too restrictive for some ellipsoid cases with high aspect ratio.
For example, an ellipsoid with long semi-major axis aligned down a long narrow channel currently fails owing to the constraints in the cross-channel direction.
Solution
In the case of a non-periodic dimension (walls) and no decomposition (
mpisz = 1
) it should be possible to relax the constraint on at least two cells safely.For periodic directions, the safety requirement remains live. However, some ability for the user to override this for 'safe' configurations can be considered.