[lammps-users] Error when running on more than 1 core

LAMMPS LAMMPS Mailing List Mirror
1

I am having trouble running a LAMMPS job (using lmp_openmpi) on multiple cores.

I can run the job on a single core (using mpirun -np 1 ...) and the job runs fine.

However when I try to run the same job over multiple cores fails at the initialisation stage, with the error massage at the bottom. I have no idea what could be going wrong, as both the data and input files are exactly the same in both cases.

I can run other jobs in LAMMPS on multiple cores so I don't think there is anything wrong with my mpi setup.

Thanks,
Michael

-------------------------SINGLE CORE---------------------------
LAMMPS (13 Aug 2010)
Scanning data file ...
   4 = max bonds/atom
   6 = max angles/atom
   9 = max dihedrals/atom
Reading data file ...
   triclinic box = (-1 -1 -1) to (49 41.07 66.958) with tilt (23.72 0 0)
   1 by 1 by 1 processor grid
   11550 atoms
   8350 bonds
   16200 angles
   22550 dihedrals
Finding 1-2 1-3 1-4 neighbors ...
   4 = max # of 1-2 neighbors
   6 = max # of 1-3 neighbors
   12 = max # of 1-4 neighbors
   16 = max # of special neighbors
2850 atoms in group surfactant
8550 atoms in group mobile
3000 atoms in group surface
EwaldN initialization ...
   G vector = 0.341444
Setting up run ...
   vectors: nbox = 24, nkvec = 10757
Memory usage per processor = 125.778 Mbytes
Step Temp E_pair E_mol E_vdwl E_coul E_long TotEng Press Lx Ly Lz Volume
        0 220.70174 4223139.2 9618735.1 5596790.6 -745643.72 -628007.65 13849472 7683583.4 50 42.07 67.958 142949.65
      100 0.74796963 4320913.9 5500529.9 5694561.8 -745639.08 -628008.78 9821469.6 9054039.6 50 42.07 67.958 142949.65
...

2

michael,

the initial potential energy of this input seems to be
very high. so it could very well be that you have some
very close contacts and that will create problems when
running in parallel.

there are multiple ways to deal with this.
first you should try to identify those atoms
and see if this is caused through incorrect
input (geometry, topology, or box size) or
parameters (force field, time step, mass etc.).

other options to deal with it would be to first
run a minimization to remove potential energy
and/or use the nve/limit integrator. have a look
at the manual. there some more comments on
what to do when you get the lost atoms errors.

cheers,
    axel.