I’m running simulations of W-H using a Tersoff potential with the latest lammps version. Because this is computationally very expensive I thought of using an EAM potential for the W-W interaction and Tersoff only for W-H and H-H. I understand that this a rough approximation but it might compare OK with an expensive full Tersoff calculation. I’ve tried the following:
pair_style hybrid tersoff eam/alloy
pair_coeff * * tersoff WH.tersoff W H
pair_coeff * * eam/alloy W_zhou.eam.alloy W NULL
This is not giving correct solution energies for H.
The reason seems to be that neighbors for the triplets WHW, etc. are not being included in Tersoff, given that WW pairs are calculated by EAM.
Any workarounds to get this to work?
Any suggesions are welcomed.
Thanks in advance.
I'm running simulations of W-H using a Tersoff potential with the latest
lammps version. Because this is computationally very expensive I thought of
using an EAM potential for the W-W interaction and Tersoff only for W-H and
H-H. I understand that this a rough approximation but it might compare OK
with an expensive full Tersoff calculation. I've tried the following:
pair_style hybrid tersoff eam/alloy
pair_coeff * * tersoff WH.tersoff W H
pair_coeff * * eam/alloy W_zhou.eam.alloy W NULL
This is not giving correct solution energies for H.
The reason seems to be that neighbors for the triplets WHW, etc. are not
being included in Tersoff, given that WW pairs are calculated by EAM.
Any workarounds to get this to work?
this is a quite frequently asked question and the answer is simple: it
*cannot* work since you do not have pairwise additive potentials.
if you want to enforce this model despite the inherent inconsistency,
you would have to "doctor" up the tersoff parameter file to explicitly
turn off all terms that you want to override with EAM and then use
hybrid/overlay, but i would worry that there is little speedup. i
haven't done any benchmarking recently, but you could check, if
tersoff/table is suitable for your case, or install the USER-OMP
package (it works w/o OpenMP), which includes optimizations similar to
what the OPT package offers for many pair styles. there are also quite
a few general tuning settings in LAMMPS that can be used to speed up
calculations that are very specific to the input deck.