Reax/c - simulation stops without error message

Stephane,

Your box shrunk quite a bit in the last stage so the neighbor list might need more frequent reconstructions. Could you try:

1) increase the safezone and mincap values, and

2) increase neighbor list construction frequency (via neigh_modify command).

Stephane,

Your box shrunk quite a bit in the last stage so the neighbor list might
need more frequent reconstructions. Could you try:

1) increase the safezone and mincap values, and

2) increase neighbor list construction frequency (via neigh_modify
command).

​...and perhaps run the initial ​equilibration as a separate input, writing
out a restart (or data) file at the end, and continue with the same
settings after reading back the restart (or data) file in a second input.
that avoids any issues due to overflowing those heuristic array dimensions.

axel.

p.s.: ray, couldn't we have some simple check, that is run infrequently and
would print a warning, when any of those heuristically pre-allocated arrays
are close to overflowing? there seems to be a recurring theme here, and
this kind of warning would be extremely helpful to many users but also
developers. there is a similar thing for neighbor lists...

I suspect that this problem is caused by the Nose-Hoover barostat, which is prone to numerical instabilities, especially for small systems.

fix fxnpt all npt temp 283 283 10 iso 1 1 100

100 fs is a rather fast time constant and not necessary given how slowly evolving the simulation is. I suspect if you increase that to 1 ps or 10 ps, your problems will go away. You could also try switching to a more stable but less rigorous form of barostatting, such as using the drag keyword, or using the Berendsen barostat.

Aidan

p.s.: ray, couldn’t we have some simple check, that is run infrequently and would print a warning, when any of those heuristically pre-allocated arrays are close to overflowing? there seems to be a recurring theme here, and this kind of warning would be extremely helpful to many users but also developers. there is a similar thing for neighbor lists…

FYI, we are working on a new version of ReaxFF that doesn’t rely on the heuristics and should be more memory robust.

Stan