[lammps-users] shake atoms missing after 0.6 ns

LAMMPS LAMMPS Mailing List Mirror
1

I am simulating a protein (~3500 atoms) on a surface in TIP3P water. After about 0.6 ns, I get the following errors:

ERROR on proc 3: Shake atoms 98373 98374 98375 missing on proc 3 at step 300632
ERROR on proc 0: Shake atoms 81729 81730 81731 missing on proc 0 at step 300632
ERROR on proc 2: Shake atoms 23205 23206 23207 missing on proc 2 at step 300632
ERROR on proc 11: Shake atoms 68871 68872 68873 missing on proc 11 at step 300632
ERROR on proc 7: Shake atoms 60075 60076 60077 missing on proc 7 at step 300632
ERROR on proc 10: Shake atoms 32601 32602 32603 missing on proc 10 at step 300632
ERROR on proc 1: Shake atoms 19869 19870 19871 missing on proc 1 at step 300632
ERROR on proc 15: Shake atoms 11112 11113 11114 missing on proc 15 at step 300632
ERROR on proc 4: Shake atoms 71460 71461 71462 missing on proc 4 at step 300632
ERROR on proc 6: Shake atoms 60702 60703 60704 missing on proc 6 at step 300632
ERROR on proc 5: Shake atoms 56676 56677 56678 missing on proc 5 at step 300632
ERROR on proc 8: Shake atoms 82116 82117 82118 missing on proc 8 at step 300632
ERROR on proc 12: Shake atoms 47937 47938 47939 missing on proc 12 at step 300632
ERROR on proc 14: Shake atoms 70740 70741 70742 missing on proc 14 at step 300632
ERROR on proc 9: Shake atoms 10239 10240 10241 missing on proc 9 at step 300632
ERROR on proc 13: Shake atoms 55431 55432 55433 missing on proc 13 at step 300632

There does not appear to be anything wrong with the system. The only property that I output that seems to change before this happened is the molecular energy, which starts to go up slightly. From a visualization, it looks like that part of the protein may leave the simulation box at this time. That should not cause this error, should it? All of the atoms above are in water molecules, but I use SHAKE on the hydrogens in the protein as well.

Thanks,
Brian

2

The error means that a processor who needs all 3 atoms to
compute the SHAKE constraint can't find one or more of
the atoms. This is b/c it is too far away from that processor
so it isn't in its ghost atom list. Normally shake distances
are much less than neighbor cutoffs - is that the case
for your model? If so, you need to find out where
those atoms are on that timstep and why they are so far away.

Writing out a restart file write before the bad timestep
and re-running to failure should help you debug.

Steve