[lammps-users] temp/rescale causing energy to drift and velocity problems

Hello all,

I'm running a small scale (800 atom) simulation and attempting to control
temperature with the temp/rescale fix:

fix 1 all temp/rescale 1 0.5 0.5 0.0 1.0
fix 2 all nve

The trouble is, after about 10,000,000 timesteps, the potential energy
will very rapidly (in about 100,000 timesteps) decay to a much lower value
(not zero, just down) and also be much less noisy. The velocities also
approach a single value after this point. Running with different starting
data files has the same problem, and running at a higher temperature
causes it to happen even sooner (8,000,000 or so timesteps in)

My first thought was that tightly controlling the temperature that way was
causing the problem, but using:
fix 1 all temp/rescale 10 0.5 0.5 0.0 1.0
fix 1 all temp/rescale 1 0.5 0.5 0.005 0.99
both caused the energy drift to happen even sooner.

Any ideas? Is this a common problem with that thermostat? I've used the
nvt thermostat on the same start up data and it works fine, but I need a
very tight control on the temperature.

Jon Brown

Have you visualized the simulation? Is it doing what
you expect. Sometimes you can get a flying-icecube
effect where the atoms crystallize and the whole chunk
flies off in some direction. The thermostat will think
this has a good temperature, so it can sometimes
actually induce this effect.