[lammps-users] size segregation at processor boundaries

Hi all.

I was wondering if anyone had seen something like this before. I'm running on a 2x4x1 MPI grid. I use the "soft" potential with an energy scale of 1.0 a binary size distribution (radius=.3, radius=.5) with about 108,000 atoms. The initial configuration is completely random (generated by hand). I have a zero temperature langevin thermostat with timeconstant 1.0.

By the end of 5000 timesteps the large particles seem to have segregated to the processor boundaries (see image below --- big guys are white... little guys are green). Worse yet at about 9000 timesteps, I get an "ERROR: lost atom" message and LAMMPS terminates. Temperature and potential energy look very sane at that point expect for a NaN in the temperature column at the very last step (divide by zero somewhere?).

I have not run a system of this size on a single CPU, but I *have* run a virtually identical system with 1/4 the particles on a single CPU (not using MPI, but using the included "STUBS" library) and have not had these kinds of problems.

Anyone have any ideas as to what I might be doing wrong?

I am not using the most current version of LAMMPS (which Steve posted last week). I will upgrade at some point today and try again.

Thanks much,

input3 (466 Bytes)

output3.gz (176 KB)

dump3.tiff (298 KB)

Can you reproduce the problem with a smaller N and # of procs? Either way,
email me your
input config, preferably from a data file, not a restart file. I presume you
meant you ran the initial phase (up to step 1000) with a soft potential, since
this input file (1000-10000) is using lj/cut? Can you send me the entire input
script (equilibration with soft and dyanmics with lj/cut)?

I don’t think the new version of LAMMPS will make any difference.

Dumb question, but are you sure your initial config is not segragated, either initially
or after the soft equilibration? I don’t think things will move very far since
you have Langevin T = 0.