An error occurs when running dynamics for a charged model using Kspace with PPPM method.

Dear LAMMPS users,

I meet a problem when running dynamics for a charged model using Kspace method of PPPM. A NVT dynamics simulation at 300K was submitted on cluster using MPIs for the models of ~210000 atoms (domain: 120 x120 x 310 Angstrom^3) in which all atoms have charge.

However, an error like : “ERROR on proc 15: Bond atoms 205827 205826 missing on proc 15 at step 1470 (…/neigh_bond.cpp:65)” occurs. I have viewed the discussions in LAMMPS users forum regarding this matter. It may be the bad dynamics run.

But I can’t fix this problem for months although I have successfully run the similar simulations of aqueous copolymer lubricants confined between metal oxide surfaces (NOTED this model also have charges) several times.

I have reduced the temperature to 10 K, or reduced the timestep to 0.1 fs, or combined of both, but it did not work. The models had also been relaxed before running dynamics in Material Studio as well as in LAMMPS. The Nose-Hoove NVT was used and updated after 100 timesteps.

Finally, when I ignored the Coulombic interactions and kspace solver in the input script, the model run well without any error. I realized that the possible reason is somewhere in using kspace solver.

In my simulation, the confined shear models are used, so that they are not periodic in z direction. Therefore, the PPPM solver with slab option is used.

Further description is shown as following:

clear

dimension 3

boundary p p f ## long coulombic

units metal

atom_style full

neighbor 2.0 bin

neigh_modify delay 0 every 1 check yes

kspace_style pppm 1.0e-4

kspace_modify slab 3.0

############ PAIRE ############

pair_style hybrid buck/coul/long 9.5 9.5 lj/cut/coul/long 9.5 9.5 lj/class2/coul/long 9.5 9.5

pair_modify mix arithmetic

bond_style harmonic

angle_style harmonic

dihedral_style opls

.

.

.

############# MINIMIZE NOT FOR RESTART ####################

minimize 1.0e-5 1.0e-6 1000 10000

reset_timestep 0

############## INITIAL TEMP NOT FOR RESTART #################

velocity all create 300 4928459 dist gaussian loop geom

velocity gBd set 0.0 0.0 0.0 units box

timestep 0.001

.

.

.

fix fNvtFlow all nvt temp 300 300 0.1 # fix fNvtFlow all nvt temp 10 10 0.1

fix_modify fNvtFlow temp cT3dProAll

.

.

.

The setting for PPPM solver is set as default.

Have you guys ever met the same problem like this ? Should I change to another solver or set more options for PPPM solver in this case ?

I will really appreciate your helps !

Best regards,

Thi Dinh TA

Dear LAMMPS users,

I meet a problem when running dynamics for a charged model using Kspace
method of PPPM. A NVT dynamics simulation at 300K was submitted on cluster
using MPIs for the models of ~210000 atoms (domain: 120 x120 x 310
Angstrom^3) in which all atoms have charge.

However, an error like : “ERROR on proc 15: Bond atoms 205827 205826
missing on proc 15 at step 1470 (../neigh_bond.cpp:65)” occurs. I have
viewed the discussions in LAMMPS users forum regarding this matter. It may
be the bad dynamics run.

But I can’t fix this problem for months although I have successfully run
the similar simulations of aqueous copolymer lubricants confined between
metal oxide surfaces (NOTED this model also have charges) several times.

I have reduced the temperature to 10 K, or reduced the timestep to 0.1 fs,
or combined of both, but it did not work. The models had also been relaxed
before running dynamics in Material Studio as well as in LAMMPS. The
Nose-Hoove NVT was used and updated after 100 timesteps.

Finally, when I ignored the Coulombic interactions and kspace solver in
the input script, the model run well without any error. I realized that the
possible reason is somewhere in using kspace solver.

​not sure, i agree with the assessment.
the error message you quote happens when a bond becomes very long (or its
coordinates invalid, i.e. NaN)​. since it happens after a small number of
steps, it is likely caused by something in your original geometry or
topology. just dump your coordinates frequently and visualize the results.
you should be able to see where the long bond happens. if it is always the
same atoms, you can label and monitor them.

axel.

Dear Axel, and Arun,

Thanks for your advices !

The problem was not from Kspace solver. When I dumped the results and geometry data for each time step, and found that there was a problem in using Buckingham potential to describe the interactions between surface’s atoms and oxygen of water molecule (Noted there was no interaction between surface’s atoms and hydrogen). The water’s bond was broken when the water molecules moved toward metal oxide surface. The total energy and temperature increased dramatically, and the losing of hydrogen atoms occurred. This problem has been overcame using simpler LJ 9-6 potential for the interactions between surface’s atoms and oxygen.

Best regards,

Thi Dinh TA

Associate Research Fellow

School of Mechanical, Materials & Mechatronic Engineering

University of Wollongong

NSW 2522, Australia

Phone : (+61) 405 234 812

Email: [email protected]…3974…