CVFF parameters and coul/long interactions

Dear LAMMPS users,

I am trying to use LAMMPS to run a MD simulation of the silicate-SDA
system. I have used the Discover form Accelrys to save the file as the
*.{car,mdf} then used msi2lmp.exe to create the *.lammps05 file using
CVFF parameters. Charges are calculated by Discover and saved into the
*.car file. Everything seems to work ok, simulation is progressing,
geometries are reasonable, I have only a problem with the electrostatic
interactions. Whereas in Discover electrostatics seems to stabilize the
whole system (approx. -2000 kcal/mol) in LAMMPS it is exactly opposite
(from +700 to +4000 kcal/mol, depending on how much I play with the
cutoffs, dielectric constant and so on). All other energies (bond,
angles, dihedrals, vdW) are similar in both runs. Could anybody help me
with this? This is my input file for the simulation:

echo screen

units real
atom_style full

boundary p p p

kspace_style ewald 0.01
dielectric 30
pair_style lj/cut/coul/long 10.0 15.0
bond_style morse
angle_style harmonic
dihedral_style harmonic
improper_style none

read_data 3c_TPA_3c.lammps05

neighbor 2.0 bin

timestep 1.0

thermo_style multi
thermo 50

fix 1 all nvt temp 275.0 275.0 100.0

dump trj all atom 100 trajectory

run 100000

Groeten - Regards - Pozdrowienia,

The only way to tell if you running the same simulation in LAMMPS
as you are with Accelrys, is to compare them, timestep by timestep
for a few timesteps. The energies and forces (pressures) should
match if you (and the conversion script) have done the conversion
correctly. We've done this in the past for some models, but the msi2lmp
tool is old and may not be working correctly for all kinds of models
in the current Accelrys.


Thank you Steve,

The thing is that even from the very first steps I have this discrepancy
and I suspect I must have made a mistake somewhere. The charges (which
obviously are troublesome) in the *lammps05 file match these in
Materials Studio, I checked and double-checked the input file (seems ok)
and I have run out of ideas where to look further...


Well, then it is a debugging problem. One good thing
to do is compare the breakdown of energies that LAMMPS
can give (thermo_style multi) with similar numbers from
Mat Studio. Then you should be able to tell if it is a dihedral
vs non-bond LJ vs whatever issue.


2011/3/25 Bartek Szyja <[email protected]...>: