[lammps-users] Question on Bin size for Neighbor Searching


I have a question on the optimum bin size used for searching neighbors. As far as I understand the code, the optimum bin size is chosen to be a half of the cutoff (for short-range interactions). And the stencil of each bin contains 14 bins for half_bin_newton() algorithm. Is that correct?

If yes, that means there may be some beads that lying inside the cutoff of a atom are missed while building the neighbor list. Does it make the result inaccurate?

Any explanation will be appreciated.



See: http://lammps.sandia.gov/doc/neighbor.html

LAMMPS uses a "skin" or buffer region to store neighbors that are
beyond the cutoff so that the neighbor list need not be updated every
timestep. So as long as you are seeing the "Dangerous builds = 0"
message at the end of your run, you're probably OK.

With Newton turned on, neighbors are only stored if they are "above
and to the right" of the given atom, so only 14 of the 27 bins in the
3x3x3 cube of bins that hold potential neighbors need to be examined.