Box boundary and rcb tiling method

Hi,

I am using LAMMPS 16 Feb 2016 version with rcb tiling method to accelerate my simulation of non-homogeneous distribution of atoms. I found that, using this tiling method if atoms go out of the box of fixed boundary (by command “boundary f f f”), the atoms will not be lost, which are supposed to. These atoms then continue to go even farther until the whole simulation stops with error “step88000-ran out of space on far_nbrs: top=1117525, max=1027468application called MPI_Abort(MPI_COMM_WORLD, -14)” (I was using reax/c potential). Would you please tell me if this is a bug and going to be fixed or I should turn to use “grid” method?

Thanks for your help.

Best regards,

Qifang

Hi,

I am using LAMMPS 16 Feb 2016 version with rcb tiling method to accelerate
my simulation of non-homogeneous distribution of atoms. I found that, using
this tiling method if atoms go out of the box of fixed boundary (by command
"boundary f f f"), the atoms will not be lost, which are supposed to. These
atoms then continue to go even farther until the whole simulation stops with
error "step88000-ran out of space on far_nbrs: top=1117525,
max=1027468application called MPI_Abort(MPI_COMM_WORLD, -14)" (I was using
reax/c potential). Would you please tell me if this is a bug and going to be
fixed or I should turn to use "grid" method?

please first upgrade to the very latest LAMMPS version and verify
whether the problem persists or has been resolved.

axel.

Hi Axel,

Thanks for your reply. I tried 30 Jul 2016 version and recompiled the program (with Intel compiler 16.0.2 and Intel mpi). The problem still exists. Lammps exited after several hundreds of steps when the first atom went outside the box. Here is the error information:

Wait, LAMMPS should crash with fixed boundary conditions if atoms leave the box. It typically does this when rebuilding the neighbour list. Is there any more specific error to go by? In the log or something?

Wait, LAMMPS should crash with fixed boundary conditions if atoms leave the
box. It typically does this when rebuilding the neighbour list. Is there any
more specific error to go by? In the log or something?

LAMMPS doesn't crash, if you set

thermo_modify lost ignore

the situation the qifang reports is a true LAMMPS bug. i recalled that
it was reported before and i had hoped, that it was addressed, but it
wasn't.
this was before we were using the issue tracker to collect bug
reports. i've dug out the previous bug report and created a
simple/fast test example demonstrating the issue based on that more
detailed report.

https://github.com/lammps/lammps/issues/176

qifang,

thanks for reporting. no promises when this is going to be resolved.
this affects a part of LAMMPS that requires very deep knowledge of the
internals to sort out. for now, you will have to stick with the brick
decomposition.

axel.

Alright. Good find, Qifang!

I think I fixed this - will be in next patch …

Thanks,

Steve

Dear Steve,

That will be great! Thank you for your help!

Best regards,

Qifang

Oh great!
Indeed, I reported that bug month ago, but just did a (dirty) turn
around to run my simulations...

I will also check the patch ASAP and give you my feedback!

Thx!