[lammps-users] cannot compute PPPM

LAMMPS LAMMPS Mailing List Mirror
1

Hello,

I am using a supercomputer to run the LAMMPS code. In my case, there are water molecules (SPCE model) and cu particles (EAM potential). Two modules are available on this computer:

  1. LAMMPS/16Mar18-foss-2019b-Python-2.7.16

  2. LAMMPS/3Mar2020-intel-2019b-Python-3.7.4-kokkos

I load and run module ‘LAMMPS/3Mar2020-intel-2019b-Python-3.7.4-kokkos’ by a script file including following command:

mpirun -np 512 lmp -in input.txt>out.log 2>&1

After some steps the simulation stops and I receive the following error:

ERROR on proc 1: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940)
Last command: run {r1} ERROR on proc 2: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940) Last command: run {r1}
ERROR on proc 385: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940)
Last command: run {r1} ERROR on proc 257: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940) Last command: run {r1}
ERROR on proc 129: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940)
Last command: run {r1} ERROR on proc 3: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940) Last command: run {r1}

ERROR on proc 386: Out of range atoms - cannot compute PPPM (src/KSPACE/pppm.cpp:1940)
.
.
.
.
.
.

I know that, theoretically, atoms are moving way too far after certain time steps. I am sure that the time step and initial geometry are suitable. Actually when I use lammps/11Aug17 on another computers the case is run without any problem.

Best,
Alireza

2

do you run on the other machine also with 512 processors?

3

it would be helpful to see the settings of the executable (output of running LAMMPS with -help) and the actual thermo output up to the run terminates.

4

I reduced the 512 processors to 128 and 64 but it didn’t work.

AlsoI run the case on another machine with 96 processors without any problem.

do you run on the other machine also with 512 processors?

5

please note that the number of MPI ranks is not the only source of possible problems. and also the absolute number may not be of significance, but the way how the simulation cell is divided into subdomains.

but without having any other useful information available it is not possible to make any other suggestions.
in my experience there can be many serious flaws in simulations that do not show under some circumstances but do under others.
and when a persons says “but the calculation runs elsewhere” that is of no significance unless the two cases differ only by a single detail, which is obviously not the case here.

axel.