Dear Lammaps Coders:
We are trying to implement a EDIP carbon model out of Nigel Marks’ group in Australia. This is a bit different from Si EDIP in USER-MISC of lammps.
For force on atom ‘I’ due to its primary neighbor ‘J’ we need to consider a 4 loops over neighbors of ‘I’ and ‘J’ (see figure/loop below). I have a problem if neighbor ‘J’ in an ‘IJ’ pair is a ghost atom from neighboring processor (domain) because we don’t have all its neighbor information in this processor.
We basically need the dR_JN vector information. How do I tackle this problem?
Thanks
Srini
Dear Lammaps Coders:
We are trying to implement a EDIP carbon model out of Nigel Marks' group in
Australia. This is a bit different from Si EDIP in USER-MISC of lammps.
For force on atom 'I' due to its primary neighbor 'J' we need to consider a
4 loops over neighbors of 'I' and 'J' (see figure/loop below). I have a
problem if neighbor 'J' in an 'IJ' pair is a ghost atom from neighboring
processor (domain) because we don't have all its neighbor information in
this processor.
We basically need the dR_JN vector information. How do I tackle this
problem?
are you sure you are looking at the current LAMMPS code?
i think quite some time ago, changes were implemented, to make sure
that the neighborlist includes neighbors of ghosts. there are several
variants for that programmed and also several manybody pair styles
request this. please note, that those pair styles usually maintain
their own "special" neighbor lists for neighbors of neighbors, since
those lists are usually based on a very short cutoff and thus going
over the entire list of neighbors would be extremely wasteful. as
examples airebo, bop, comb3, lcbop come to mind...
please have a closer look.
axel.