Delete_atoms fails with MEAM potential

_Douglas_E_Spearot · August 31, 2014, 8:21pm

All-

I’m having a strange problem with delete_atoms in a simulation where I am running with the MEAM potential.

I believe that I have compiled the MEAM library and LAMMPS correctly, because if I comment out the “delete_atom overlap” command in my input file, the simulation runs fine.

Any ideas on why a delete_atoms command would fail after it writes the “System init for delete_atoms …” line?

Thanks!

-Doug Spearot

LAMMPS (1 Feb 2014)

Lattice spacing in x,y,z = 4.05 4.05 4.05

Created orthogonal box = (0 0 0) to (600 600 600)

2 by 3 by 2 MPI processor grid

Created 429664 atoms

Created 88209 atoms

429664 atoms in group core

88209 atoms in group base

Lattice spacing in x,y,z = 5.43 5.43 5.43

Created 1080298 atoms

1080298 atoms in group shell-structure-cr

Created 1080298 atoms

2160596 atoms in group shell-structure

Deleted 1080298 atoms, new total = 1598171

System init for delete_atoms …

lmp_razor:32484 terminated with signal 11 at PC=6c974c SP=7ffff47fe590. Backtrace:

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x6c974c]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x4ff84a]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x500b95]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x651fd3]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x64f607]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x6515fc]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x65fe49]

/lib64/libc.so.6(__libc_start_main+0xfd)[0x2b42116dbd1d]

/home/dspearot/lammps-1Feb14/src/lmp_razor[0x404bf9]

akohlmey · August 31, 2014, 8:50pm

All-

I’m having a strange problem with delete_atoms in a simulation where I am
running with the MEAM potential.

I believe that I have compiled the MEAM library and LAMMPS correctly,
because if I comment out the “delete_atom overlap” command in my input file,
the simulation runs fine.

Any ideas on why a delete_atoms command would fail after it writes the
“System init for delete_atoms …” line?

no idea, perhaps a bug that was fixed in a subsequent patch of LAMMPS.
please try again with the latest version.
i just made a simple test and it was not segfaulting. if you still
have a problem with that version, please provide a small test example.

axel.

_Douglas_E_Spearot · September 1, 2014, 2:50pm

Alex-

Thanks for your response. I've done some testing today on a "stripped down" version of my input file. The stripped down version builds a hemispherical Al nanodot on a fixed Al substrate. I have run both the Feb 1 and the Jun 28 versions of LAMMPS (last 2 stable releases). The input file is pasted below. The MEAM parameter files were downloaded from the NIST repository (according the student I'm working with). Based on these results, I'm concerned that I don't have the MEAM LAMMPS built properly. I've run these on a 2*6 core Intel Westemere system.

Feb 1 Lammps / Intel 14.0.3 / mvapich2 2.0:
1 Node, 12 cores: simulation runs to completion in 13.9 s
2 Nodes, 24 cores: simulation crashes with *.btr error files

Feb 1 Lammps / Intel 14.0.3 / openmpi 1.8.1:
1 Node, 12 cores: simulation runs to completion in 14.2 s
2 Nodes, 24 cores: simulation crashes without *.btr files (get a “lost atoms error in LAMMPS”?)

Jun 28 Lammps / GCC 4.7.2 / openmpi 1.8.1:
1 Node, 12 cores: simulation runs to completion in 269.2 s
2 Nodes, 24 cores: simulation crashes without *.btr files (get a “lost atoms error in LAMMPS”)

While the input file below doesn't test the exact same delete_atoms error as I mention below, I think there is a bigger issue with the way that I am building or implementing the MEAM library into LAMMPS (that is causing the delete_atoms error below). Can you reproduce any of these errors?

Thanks!

-Doug

akohlmey · September 1, 2014, 2:55pm

Alex-

Thanks for your response. I've done some testing today on a "stripped down" version of my input file. The stripped down version builds a hemispherical Al nanodot on a fixed Al substrate. I have run both the Feb 1 and the Jun 28 versions of LAMMPS (last 2 stable releases). The input file is pasted below. The MEAM parameter files were downloaded from the NIST repository (according the student I'm working with). Based on these results, I'm concerned that I don't have the MEAM LAMMPS built properly. I've run these on a 2*6 core Intel Westemere system.

Feb 1 Lammps / Intel 14.0.3 / mvapich2 2.0:
1 Node, 12 cores: simulation runs to completion in 13.9 s
2 Nodes, 24 cores: simulation crashes with *.btr error files

Feb 1 Lammps / Intel 14.0.3 / openmpi 1.8.1:
1 Node, 12 cores: simulation runs to completion in 14.2 s
2 Nodes, 24 cores: simulation crashes without *.btr files (get a “lost atoms error in LAMMPS”?)

Jun 28 Lammps / GCC 4.7.2 / openmpi 1.8.1:
1 Node, 12 cores: simulation runs to completion in 269.2 s
2 Nodes, 24 cores: simulation crashes without *.btr files (get a “lost atoms error in LAMMPS”)

While the input file below doesn't test the exact same delete_atoms error as I mention below, I think there is a bigger issue with the way that I am building or implementing the MEAM library into LAMMPS (that is causing the delete_atoms error below). Can you reproduce any of these errors?

i cannot reproduce anything, since i cannot run this input without the
files meam.meamf and meam.alsimgcufe

axel.

akohlmey · September 1, 2014, 3:53pm

Alex-

Thanks for your response. I've done some testing today on a "stripped down" version of my input file. The stripped down version builds a hemispherical Al nanodot on a fixed Al substrate. I have run both the Feb 1 and the Jun 28 versions of LAMMPS (last 2 stable releases). The input file is pasted below. The MEAM parameter files were downloaded from the NIST repository (according the student I'm working with). Based on these results, I'm concerned that I don't have the MEAM LAMMPS built properly. I've run these on a 2*6 core Intel Westemere system.

Feb 1 Lammps / Intel 14.0.3 / mvapich2 2.0:
1 Node, 12 cores: simulation runs to completion in 13.9 s
2 Nodes, 24 cores: simulation crashes with *.btr error files

Feb 1 Lammps / Intel 14.0.3 / openmpi 1.8.1:
1 Node, 12 cores: simulation runs to completion in 14.2 s
2 Nodes, 24 cores: simulation crashes without *.btr files (get a “lost atoms error in LAMMPS”?)

Jun 28 Lammps / GCC 4.7.2 / openmpi 1.8.1:
1 Node, 12 cores: simulation runs to completion in 269.2 s
2 Nodes, 24 cores: simulation crashes without *.btr files (get a “lost atoms error in LAMMPS”)

While the input file below doesn't test the exact same delete_atoms error as I mention below, I think there is a bigger issue with the way that I am building or implementing the MEAM library into LAMMPS (that is causing the delete_atoms error below). Can you reproduce any of these errors?

the "lost atoms" error has nothing to do with the MEAM parameters.
it is due to an incorrect use of shrink-wrap boundary conditions.
if you replace "s" with "m" and it should work. for better
performance, you probably also want to add:

processors * 1 *

and later after you created the system:

balance 1.1 shift xz 5 1.0

to resolve the massive load imbalance that your system setup causes.

don't use GCC for executables to run this input. apparently the
gfortran compiler generates extremely inefficient code when compiler
the meam library.

so what delete_atoms command do you place where to trigger a segfault?

_Douglas_E_Spearot · September 2, 2014, 10:13pm

Alex-

Thank you very much for your help. Switching from "s" to "m" in the "boundary" command has fixed my problem.

The input file that I copied in my post below was a simplified version of a larger input file. I was trying to make sure that everything was correct up to the point that I (1) added a second element to the system (Si) and (2) performed a "delete_atoms overlap" between groups containing each atom type. Essentially, I was trying to debug my own code. The "delete_atoms" command works fine now with the boundary condition change.

Best,

-Doug