[lammps-users] atoms missing in parallel job ?

LAMMPS LAMMPS Mailing List Mirror
1

Hi Everyone,

I am running a simple simulation where I have several bond, angles and LJ type VdW potentials.
It runs on a single processor without giving any errors, however when I want to use more than 1 cpu, It gives the following error.
Note: I start the simulation with minimization.

WARNING: Resetting reneighboring criteria during minimization
Setting up minimization ...
ERROR on proc 0: Angle atoms 786 787 788 missing on proc 0 at step 0
MPI_ABORT invoked on rank 0 in communicator MPI_COMM_WORLD with errorcode 1
ERROR on proc 1: Angle atoms 4902 4903 4904 missing on proc 1 at step 0
MPI_ABORT invoked on rank 1 in communicator MPI_COMM_WORLD with errorcode 1
ERROR on proc 2: Angle atoms 3586 3587 3588 missing on proc 2 at step 0
MPI_ABORT invoked on rank 2 in communicator MPI_COMM_WORLD with errorcode 1
ERROR on proc 3: Angle atoms 759 760 758 missing on proc 3 at step 0
MPI_ABORT invoked on rank 3 in communicator MPI_COMM_WORLD with errorcode 1
ERROR on proc 4: Angle atoms 1235 1236 1234 missing on proc 4 at step 0
MPI_ABORT invoked on rank 4 in communicator MPI_COMM_WORLD with errorcode 1
ERROR on proc 5: Angle atoms 4595 4596 4594 missing on proc 5 at step 0
MPI_ABORT invoked on rank 5 in communicator MPI_COMM_WORLD with errorcode 1
mpirun noticed that job rank 0 with PID 12686 on node snare.biology.gatech.edu exited on signal 1 (Hangup).

Do you have any idea why it doesn't work when I use multi processors ?
Thanks in advance,
-burak

2

This error occurs b/c atoms in a bond, angle, etc have moved
too far from a proc's sub-domain for the proc to find them.
Shouldn't happen if the dynamics of your system is reasonable
and you are rebuilding neighbor lists often enough. Have you
done viz on the system to see if it is behaving?

Steve

3

It is interesting that it behaves in single processor but not in multi processors.
I solved the problem by increasing LJ cut off distance. Now it is working.

Thanks,
-Burak

4

Again, the problem in parallel is an indication atoms
are moving too far between reneighborings. If lengthening
the cutoff fixes this, then good, but it doesn't mean your
simulation was running correctly before on 1 processor.

Steve