Bond / Improper atoms %d %d %d %d missing

LAMMPS LAMMPS General Discussion
1

I am running tensile deformation simulations on epoxy polymers in LAMMPS using the fix bond/break command. However, I am encountering errors such as ‘bond atoms missing’ and ‘improper atoms missing’ shortly after bond breakage.

To investigate the cause of these errors, I visualized the simulation and observed that the two atoms involved in the bond break move apart abruptly after breaking. 1) I have tried restarting the simulation with geometry optimization immediately after bond breaking and 2) lowering the timestep to as small as 0.005 fs, but these approaches did not resolve the issue.

I noticed that when running the simulation using 2 nodes (48 cores in my setup), 4 nodes, and 6 nodes, the errors occur at different bonds. This led me to believe that the issue is related to communication between processors. I have reviewed the documentation for comm_modify , neigh_modify , and neighbor (skin) and tested various parameter adjustments, but the errors persist.

Finally, I ran the simulation on a single core and found that it completed without any errors. However, running the simulation on a single core is not feasible due to the significant computational cost.

Could you please provide guidance on how to resolve these errors while running the simulation across multiple cores?

Thank you for your assistance.

2

These kinds of “atom missing” errors result from atoms moving too fast between subdomains. This usually means that atoms experience very large forces and then get accelerated.

In your case, that is most likely a consequence of your fix bond/break settings, i.e. when you allow bonds to break, there is very large amount of potential energy in the local structure that is then released when you break the bond. This cannot be “fixed” by changing the neigborlist settings. Instead, you either have to revise your fix bond/break settings or you have to use a smaller timestep or both. With a smaller timestep your atoms will move less far per timestep and then using suitable neighbor list settings can lead to bonded atoms being communicated fast enough to avoid the error.

1 Like
3

@akohlmey would fix nve/limit be appropriate in this application to dissipate the energy or would the lack of energy conservation be a problem?