Lost atoms: original 1792 current 1786 (src/thermo.cpp:425) when using coul/long

Dear LAMMPS User,
I am writing this mail regarding the simulation of nitrogen-nitrogen model with five pseudo charges on it to estimate shear viscosity. I am currently using LAMMPS (27 May 2021) version. When I use pair_style hybrid/overlay table linear 1301 coul/long 100, I happen to get lost atoms. But, when using pair_style hybrid table linear 1301 it works perfectly fine. But, I have to use coul/long to take into account the long range interaction as well.

But, when using coul/long in the pair_style for the calculation of long range interaction it creates atoms are missing or bond atoms missing in proc. I have attached the input and output script for the above mentioned simulation in the following drive link.


Any help will be highly appreciated.

Thanking You,

Unfortunately, your input file is a hard to decipher chaotic mess and thus very “debug unfriendly”

  • It uses variables excessively, so it is difficult to figure out what commands have what arguments
  • It has loads of commented out commands, so it is difficult to identify which commands are actually in use
  • it diverts output to different log files, so one has to figure out what output is printed where and when

From what I can see you are using fix rigid/nvt/small for time integration and a timestep of 1fs for a system of molecules in gas phase. So my first guess would be that your timestep is too large. Your molecules may begin to rotate rather quickly when they bump into each other and the rotational time integration for rigid bodies is known for being less stable than the time integration for the center of mass of the molecules or for individual atoms.

When you add the coulomb interaction via overlay you are modifying your forces. Typically parameterizations are set up to balance coulomb and non-coulomb where the non-coulomb is designed to keep the coulomb suitably separated. Thus atoms will be able to come closer and the repulsive interaction will be “harder”, leading to larger forces and thus larger translational or rotational velocities and thus requiring a shorter timestep.

1fs is quite aggressive for small rigid bodies even under favorable circumstances, so my suggestion would be to try shorter timesteps, perhaps 0.25fs or 0.1fs.

Thanks Axel. I’ll try to reduce the timesteps.

Warm regards,