[gmx-users] Coredump when using PME

[Robert Bjornson]

Hi,

I'm experiencing a coredump when running gromacs 3.3 on a system of
72000 atoms.  The coredump only occurs when I use pme for
electrostatics.

Here is the stacktrace (no symbols compiled in unfortunately)
(gdb) where
#0  0x0000000000457877 in spread_q_bsplines ()
#1  0x000000000045b4f8 in do_pme ()
#2  0x0000000000437d81 in force ()
#3  0x000000000046bb6a in do_force ()
#4  0x000000000041d329 in do_md ()
#5  0x000000000041ba14 in mdrunner ()
#6  0x0000000000420057 in main ()

I also noticed some odd behavior in stderr output.  No idea if they
are related, but the
non-ewald run doesn't show this behavior:

==================
Step 20, time 0.04 (ps)  LINCS WARNING
relative constraint deviation after LINCS:
max 0.096362 (between atoms 523 and 524) rms 0.004632
bonds that rotated more than 30 degrees:
 atom 1 atom 2  angle  previous, current, constraint length
    524    525   66.5    0.1431   0.1395      0.1430
    524    526   37.7    0.1502   0.1430      0.1500

Back Off! I just backed up step19.pdb to ./#step19.pdb.1#
Wrote pdb files with previous and current coordinates

Step 21, time 0.042 (ps)  LINCS WARNING
relative constraint deviation after LINCS:
max 0.120046 (between atoms 524 and 526) rms 0.005158
bonds that rotated more than 30 degrees:
 atom 1 atom 2  angle  previous, current, constraint length
    524    525   64.7    0.1398   0.1573      0.1430
    524    526   55.8    0.1433   0.1680      0.1500

Back Off! I just backed up step20.pdb to ./#step20.pdb.1#
Sorry couldn't backup step20.pdb to ./#step20.pdb.1#

Back Off! I just backed up step20.pdb to ./#step20.pdb.1#
Sorry couldn't backup step20.pdb to ./#step20.pdb.2#

Back Off! I just backed up step21.pdb to ./#step21.pdb.1#
Wrote pdb files with previous and current coordinates
Wrote pdb files with previous and current coordinates

Back Off! I just backed up step21.pdb to ./#step21.pdb.2#
Sorry couldn't backup step21.pdb to ./#step21.pdb.2#
Wrote pdb files with previous and current coordinates
Wrote pdb files with previous and current coordinates
-----------------------------------------------------------------------------


Some background data:
I'm running on 8 processors using LAM-MPI and pbs-pro (although the
problem also occurs for sequential runs)

The only difference between the two runs is the use of pme in the run
that fails.  PME failed in a similar way on a different input set, as
well.

Does anyone have any idea why this might be happening?  I can provide
the input/output files or corefile, but didn't want to broadcast them
to everyone.

Thanks very much for any assistance,

Rob Bjornson