[lammps-users] ch2lmp dihedral/improper inconsistency

Dear LAMMPS users,

I ran into a problem with the ch2lmp tool, and thought I would post here before contacting the authors of the tool. Following the example folder provided by the authors, one builds 1ac7.data from a psf file. I noticed, however, that the number of dihedrals listed in the psf file (1497) is inconsistent with the number of dihedrals in the lammps data file (1659). Despite this inconsistency, the data file still seems to work well with LAMMPS, although I’m unsure if it is the same structure as in the psf (perhaps it is the same, and the number was such that some dihedral ids were skipped?).

Second, I tried using the charmm2lammps tool on 1d3z pdb. I generated the psf using psfgen, as instructed in the readme file, vmd commands listed below. This psf contained 1231 atoms, 1242 bonds, 2257 angles, 3293 dihedrals, 205 impropers and 74 cross terms. When I used the charmm2lammps tool, however, a LAMMPS data file was created with 1231 atoms, 1242 bonds, 2257 angles, 3428 dihedrals, 202 impropers (not sure about cross terms). When I attempt to use this data file in a lammps simulation, I receive an error: “1-3 bond count inconsistent”

So my first question is, can the example provided in ch2lmp be trusted? and second, having understood the dihedral numbering, what is wrong with 1d3z, perhaps a missing improper?

Please let me know if I haven’t provided enough information.

Thank you,
Harold Hatch

I would contact the ch2lmp authors:

Paul Crozier - pscrozi at sandia.gov
Pieter in't Veld - pieter.veld at basf.com


Dear LAMMPS community,

Pieter Veld explained to me the different in the number of dihedrals:

"The reason you see more dihedrals in the conversion finds its origin in how
LAMMPS uses weights in interpreting its dihedrals. To get the same
representation as is needed by a program like CHARMM, LAMMPS needs a
different amount of dihedrals in its dihedral list than CHARMM does. The
inconsistency in dihedral count, however, is offset by the weight that has
been assigned to each dihedral entry. I will by no means state, that this
tool is a 100% proof solution. However, as Paul mentioned, during its
development we made sure that all nonbonded and bonded energies produced by
CHARMM correspond to the ones generated by LAMMPS. Bottom line is that the
CHARMM and LAMMPS way of interpreting the dihedral lists are not identical,
which then leads to not being able to compare content of the lists on a
one-to-one basis. Energetically the outcome should be identical."

And he is certainly correct, as I am getting the same interaction energies
from building in namd 1UBQ.pdb, and converting to lmp. I still don't quite
understand the bound count consistency error from 1d3z, but that concern is
less pressing and probably due to a mistake I made.

Harold Hatch