I am studying the pyrolysis of AlH3 (Aluminium Hydride) using ReaxFF (pair_style reax/c) and have some problems.
I built up an AlH3 crystal cell (5 × 5 × 2 supercell) and placed it in the centre of a 200 × 200 × 200 angstrom simulation box. I equilibrated the system for 50 ps and then run the simulations at temperatures of 1500 K, 2000 K, and 2500 K, respectively. However, for all the chosen temperatures, the simulation stopped after about 10 - 20 hours (timestep 0.1 fs; several millions of steps’ run) and reported the error ‘Non-numeric atom coords - simulation unstable’ after some nan values.
I know that this error usually results from bad dynamics of the system or close contacts between atoms. But my simulations only encountered this error after a certain runtime rather than at the beginning (which means there is no problem with my initial configuration). I also tried a smaller timestep 0.05 fs but the results were the same. The ReaxFF force field file I used is extracted from ‘J. Phys. Chem. C, 2016, 120 (17), pp 9464–9474’. I checked this force field file and it should be suitable for my system. The LAMMPS version I used is 11Aug17.
I would appreciate it very much if anyone could help me with it or provide any suggestion. Many thanks in advance.
If the error is repoducible, I suggest you output thermo info
every step in the time window when the crash occurs.
E.g. if the crash occurs around step 1200650:
run 1200600
run 100
You an also write out a restart file shortly before
the crash and see if you can restart from
that and still see the crash. That helps you
(or us if its a LAMMPS problem) to debug.
Thank you very much for your reply and suggestion.
I did write out the restart files. As you suggested, I restarted the simulation from the restart file before the crash (i.e. the crash occurs around 434000; I restarted from 400000 and run another 100000 for test). I kept all the settings same as before (input commands, running nodes and cores, etc.) and outputted thermo info every step. However, the simulation run without any problem for this 100000 timesteps and ended normally.
Do you have any idea why this happened? Many thanks.
There are several reasons why a restart might not produce identical behavior.
First, are you restarting on the same # of procs? If not, 34000 steps later
trajectories can diverge. The QEq in ReaxFF may also not restart
identically, not sure if that is the case.
Aidan (Ccd) may have other reasons why ReaxFF does not restart
identically. Can you also post your input/restart scripts in case
there is some other command that can trigger this?
Also, if you can afford to run the full 450K steps, I suggest you
just rerun from step 0 with thermo output every step
to see what happens around 434000. Again use the same
of procs or the trajectories may not be identical.
Yes, I restarted on the same # of procs. Please find attached my input and restart scripts. I would be very grateful if you would help me check the scripts if there are some other commands that make different restart behaviors. Many thanks.
I don’t see any obvious issue with the 2 scripts that would prevent an exact restart.
You are restarting at step 400,000. If you look at thermo output
on a fine timescale (e.g. every step) from 400,000 onward in both
the original and restart scripts, how quickly and by how much do they differ?