LAMMPS minimum image convention

Hello,

I’m performing NPT simulations in LAMMPS with periodic boundary conditions and investigating the dependence of my results on the two-body potential cutoff radius. Initially, I ensured the cutoff was always smaller than half the smallest box dimension to comply with the minimum image convention. However, I accidentally ran simulations with a cutoff larger than half the box length, and surprisingly for my LAMMPS did not throw any errors. Also, the results appeared physically reasonable (not nonsense).

Upon researching this, I found some old discussions (2009–2010) with conflicting comments on whether cutoffs can exceed half the box size (e.g., this thread). Additionally, the LAMMPS Developer Communication Guide states:

Force computation in LAMMPS is not subject to minimum image conventions and thus cutoffs may be larger than half the simulation domain.

As a beginner, I’m confused by this and would appreciate clarification on:

  1. Can the cutoff radius indeed be arbitrarily large without causing calculation errors (e.g., due to unphysical self-interactions)?
  2. How does LAMMPS avoid infinite self-interactions with periodic images when the cutoff exceeds half the box size?

Thank you for your insights!

The cutoff cannot be arbitrarily large, but minimum image conventions do not apply with a few exceptions (e.g. pair style list). You are wrong about “self interactions” being unphysical. This is how periodic boundaries work.

That is an ill-poised question. See my comment above. The reason for other MD codes applying minimum image conventions is that they only have the “local” atoms present and thus when the cutoff is too large, they cannot find the corresponding periodic images of those local atoms. LAMMPS on the other hand maintains for each sub-domain a set of “ghost” atoms which are real copies and thus pairs in the neighbor list can be constructed without having to find the closest image of the “local” atom.

For more details, please carefully study the latest LAMMPS paper and the parallel algorithms section of the manual.