Error running on multiple pprocessors

LAMMPS LAMMPS Mailing List Mirror
1

Dear lammps users,

I have a problem in running my file on multiple processors. When I try “mpirun -np 1 ./lmp_mpi -in in.myfile” I get no error. When I try using 2 processors, again I have no problem, but when increasing the np to 8 or 16, I get the following error:

ERROR on proc 0: Restrain atoms 36 25 28 30 missing on proc 0 at step 0 (…/fix_restrain.cpp:459)
application called MPI_Abort(MPI_COMM_WORLD, 1) - process 0
ERROR on proc 1: Restrain atoms 8356 8345 8348 8350 missing on proc 1 at step 0 (…/fix_restrain.cpp:459)
application called MPI_Abort(MPI_COMM_WORLD, 1) - process 1

I searched the error section of the manual and I found that it is relaed to dihedrals. But I don’t know why it happens only when I use more than 2 processors???
Anyone can help with finding where the problem could be?

Thanks.

Neda

2

Dear lammps users,
I have a problem in running my file on multiple processors. When I try
"mpirun -np 1 ./lmp_mpi -in in.myfile" I get no error. When I try using 2
processors, again I have no problem, but when increasing the np to 8 or 16,
I get the following error:

ERROR on proc 0: Restrain atoms 36 25 28 30 missing on proc 0 at step 0
(../fix_restrain.cpp:459)
application called MPI_Abort(MPI_COMM_WORLD, 1) - process 0
ERROR on proc 1: Restrain atoms 8356 8345 8348 8350 missing on proc 1 at
step 0 (../fix_restrain.cpp:459)
application called MPI_Abort(MPI_COMM_WORLD, 1) - process 1

I searched the error section of the manual and I found that it is relaed
to dihedrals. But I don't know why it happens only when I use more than 2
processors???

​because your domain decomposition subdomains are ​too small to contain all
atoms listed in the restraint. you may have to increase your communication
cutoff or check whether the restrain selection is correct.

axel.

3

Thanks Axel,

But when I increase my lj and coloumbic cut offs from 10 and 12 angstrom to 12 and 14 , I get another error “neighbour list overflow”

How can I solve this problem? How can I get a sense of values that should be defined in neighbour and neighbour_modify commands?

Thanks

Neda

4

Thanks Axel,
But when I increase my lj and coloumbic cut offs from 10 and 12 angstrom
to 12 and 14 , I get another error "neighbour list overflow"

​communication cutoff != interaction cutoff.​

http://lammps.sandia.gov/doc/comm_modify.html

How can I solve this problem? How can I get a sense of values that should

be defined in neighbour and neighbour_modify commands?

​you need to spend more time reading the documentation and also need to
spend more attention to detail.​

​it is quite suspicious, that you get a neighbor list overflow with 14
angstrom cutoff. you either have an unusual system or a bad geometry.

axel.​