[lammps-users] Losing atoms... update

A while ago I posted about LAMMPS losing two of my atoms. I have done some more investigation... it was suggested to me that the atoms were getting too close together, and gaining near infinite velocity. This seems extremely likely, but I am unsure how they are doing this... The timestep before the atoms are lost, I am dumping velocities, and finding two neighboring atoms with 10^9 range velocities; then they are lost. I calculated the distance between the atoms the timestep before (.501558541 Angstroms) they are lost, and the timestep before that (.501224 A). However, looking closely at the potential, it seems that they would need to get closer than .50032 A to reach the 'infinite' potential region. The potential isn't continuous in this region. If they don't get that close, they get a reasonable dose of potential (order 10^2, not 10^12). It's really really close to seeing the infinite potential though...

I then zoomed in on this area of time, by decreasing the timestep another order of magnitude. Then I get really really strange behavior. The atoms are still lost, but take a longer time to get lost. In the mean time, they are rocketing around the simulation box wildly. They pass on some energy to other atoms, but not in a uniform way, or in the way you would expect to see in a cascade. When I visualize it in xmovie, it's like 99% of the atoms are perfectly still, with about 10 atoms zooming all over the place at very high speed. They still only get as close as .501427 A though, before starting to go crazy. The potential is Fe_mm.eam.fs, in case anyone wants to check that I'm calculating the point at which they should see inifinite pair potential correctly.


All of the behaviors you describe sound bad. Atoms should
never be zooming or get close enough to sample the steep parts
of the potential. If your system is at high T you might ask the
author of the potentail (see top of file) if you are using is
appropriately for your system.