When spectral solvers, the number of guard cells can be quite large.
When using domain decomposition, we can sometimes end up in a situation for which the number of physical cells in individual boxes is lower than the number of guard cells.
In this case, my understanding is that the code is incorrect, because it only considers guard cells communications with nearest-neighbors boxes, while in the above-mentioned situation the guard cells can also overlap with next-nearest-neighbors.
@atmyers @WeiqunZhang Is that also your understanding?
If yes, is there an easy way to add a safe-guard against this, within amrex?
One way that I can think of is to impose that the bocking_factor should always be larger than the number of guard cells? Does that seem reasonable?
When spectral solvers, the number of guard cells can be quite large. When using domain decomposition, we can sometimes end up in a situation for which the number of physical cells in individual boxes is lower than the number of guard cells.
In this case, my understanding is that the code is incorrect, because it only considers guard cells communications with nearest-neighbors boxes, while in the above-mentioned situation the guard cells can also overlap with next-nearest-neighbors.
@atmyers @WeiqunZhang Is that also your understanding? If yes, is there an easy way to add a safe-guard against this, within
amrex
? One way that I can think of is to impose that thebocking_factor
should always be larger than the number of guard cells? Does that seem reasonable?