Error Resuming NPT Tempering Calculations

When resuming npt tempering calculations, I sometimes get a variety of errors including:

./log.lammps:ERROR on proc 0: Too many neighbor bins
./log.lammps.1:ERROR: Non-numeric pressure - simulation unstable
./log.lammps.10:ERROR on proc 0: Bond atoms 145 146 missing on proc 0 at step 20068
./log.lammps.12:ERROR: Domain too large for neighbor bins

Which suggest that the simulation box has become unstable.

What I don’t understand is that I have run the same calculation continuously for ~7ns without issue.

I defined the same pair_style, kspace, restraints and all the same fixes in the restart file (run.in.restart.temper) as I did with the initial run (run.in.temper).
The only differences are:
Minimization step is skipped
NVT equilibration skipped
World variable indices replaced with those of previous run

I am restraining the y-values of certain atoms according to a ratio of the box y-dimension 0.25 and 0.75, could there be some sort of unexpected interaction here?

Example temper and restart files:
run.in.temper (1.7 KB)
run.in.restart.temper (1.8 KB)

In this attatched example 15_0_original was run for ~7ns without issue. To test loading of tempering calculations I ran 15_1 from scratch and tried to resume calculation with 15_2 and then 15_3 giving the above errors suggesting box explosion.

Thanks in advance and please let me know if any more details are needed from me.

@zanper I didn’t look to closely but appears as if you are not handling the restart process correctly. Since you are using a multi-partition simulation, you must write and read a different restart file for each replica and those need custom file names. This would typically be done using a world style variable. It is not worth looking at anything else until you have corrected that step. Please note that you also seem to be overwriting the trajectory files from all replica for the same reason.

Naming the restart files according to the world variable seems to have done the trick.

Thank you,
-ZP