SNAP potential cutoff


I parameterized my own SNAP + ZBL potential for silicon, see the attached files and following pairstyle commands:

pair_style hybrid/overlay zbl 3.8 5.03 snap
pair_coeff 1 1 zbl 14 14
pair_coeff * * snap Si.snapcoeffs Si Si.snapparams Si

This potential works for my purposes, except I’d like to extend SNAP to a longer cutoff.

During parameterization, I let the cutoff vary from 1st nearest neighbors to 4th nearest neighbors. Every optimized solution preferred a cutoff of 1st nearest neighbors. When I try to fix the cutoff at a longer distance, the optimization isn’t successful and a good solution cannot be found.

Possible issues are that I’m fixing the number of SNAP coefficients…

Do I need more SNAP coefficients when using a longer cutoff? Does anyone know how can I successfully use the SNAP potential for longer cutoff distances?

Thanks for your time.

Si.snapcoeffs (405 Bytes)

Si.snapparams (97 Bytes)

Very interesting. The particular form of the SNAP potential that we have implemented in LAMMPS is not the only choice one could make. It is possible that for certain physical systems, other choices might work better.