tersoff potential parameter problem

We have some incompatibilities between Lammps and parcas (code
developped by Kai.nordlund in university of Helsinki) concerning the
W-H potential especially for the solution energy of H.
It seems that lammps has left out some of the terms which should be
in the Tersoff potential. The omitted terms seem to come from they
doing a sum like sum_(i) sum_(j>i) but only including the three-body
term bij and not the bji term.
The reason the Lammps version still works for pure bcc W might be due
to the high symmetry, giving bji = bij so that omission of bji goes unnoticed.
A calculation result for solution energy of H at interstitial site just come out
huge difference from another code PARCAS.

so, is that possible that the simplification of bij cause all these differences?

Aidan can reply to this.

Steve

Hi,

People have been using Nordlund/Albe/Ehrhart style alloy Tersoff
potentials extensively in LAMMPS for quite a while, and we have never
noticed a problem of this type. There may be something special about your
W-H potential that LAMMPS is handling incorrectly. Have you tried testing
Kai's code against similar LAMMPS Tersoff potentials?

[[email protected] potentials]$ grep -i albe *.tersoff
GaN.tersoff:# The following GaN potential is from J. Nord, K. Albe, P.
Erhart
SiC_Erhart-Albe.tersoff:# values are from Erhart/Albe - Phys Rev B, 71,
035211 (2005)

Also, if there is really a logical error in LAMMPS of the type you
suggest, it should cause the energy to change when the neighbor list is
rearranged. For example, the energy would depend on the number of MPI
processes used and/or the location of the origin for the periodic cell
(see LAMMPS command origin).

Let us know what you find out.

Aidan