[lammps-users] bond or angle or dihedral atoms missing on some processors

Hi everyone,
I got some errors when I used LAMMPS to do the large-scale MD simulation. The error is : ERROR on proc 6: Bond atoms 10011 10017 missing on proc 6 at step 130192. Sometimes, the missing atoms are angle or dihedral atoms.
I checked the last configuration of the atoms that will be missing on processors in the next steps as mentioned in the ERROR message before the error occurred and found that their motions were smooth. There would not be violent motion in the next advancing step according to the given potential. The most surprise to me is that, the errors disappeared automatically when I restarted the simulation by using the last phase point data before the errors occurred. I don’t know why these errors happened and how to avoid these errors. Is there someone having this experience? Your any suggestions would be highly appreciated.

Helen
University of Missouri, Columbia

2010/6/14 Helen <[email protected]>:

Hi everyone,

dear helen,

I got some errors when I used LAMMPS to do the large-scale MD simulation.
The error is : ERROR on proc 6: Bond atoms 10011 10017 missing on proc 6 at
step 130192. Sometimes, the missing atoms are angle or dihedral atoms.
I checked the last configuration of the atoms that will be missing on
processors in the next steps as mentioned in the ERROR message before the
error occurred and found that their motions were smooth. There would not be
violent motion in the next advancing step according to the given potential.

there doesn't have to be a violent motion. atoms just have to move
too far before the next rebuild of the neighbor list happens.

The most surprise to me is that, the errors disappeared automatically when I
restarted the simulation by using the last phase point data before the
errors occurred. I don't know why these errors happened and how to avoid
these errors. Is there someone having this experience? Your any suggestions
would be highly appreciated.

while the most common reason for this kind of error message is high
potential energy, too aggressive settings for rebuilding the neighbor lists
or too large a time step can also trigger this issue.

cheers,
   axel.

To add to Axel's comment, a proc only knows
about ghost atoms out to the neighbor cutoff (pair cutoff + neigh skin).
If you are using a short pair cutoff and waiting too long
to reneighbor, then your error can happen. Try neigh_modify delay 0 check yes
and see if your problem goes away. It would help to know how far
away the missing atoms are on the error step, which you may be
able to infer from looking at a dump file on the proper timestep.
If it's quite far, then something is probably wrong with your model.

Steve

1 Like