[gmx-users] position restraints & domain decomposition & Some interactions seem to be assigned multiple times : deadly combination

[Dr. Vitaly Chaban]

Dear All -

Did anyone observe the crash of domain decomposition while using
position restraints? The formal error is

"Some interactions seem to be assigned multiple times"

Sotimes it also gives me:

"Not all bonded interactions have been properly assigned to the domain
decomposition cells"

The error happens during the first 1,000 steps of MD. It does not
depend on the number of cores exploited, either 8 or 128. There are no
problems with the same system, but without restraints. There are no
problems with "-nt X". PBC are present. The system contains more than
200,000 interaction centers. The force constant for position
restraints are notably small, i.e. for any particle in the starting
configuration the corresponding restraint force does not exceed kT.

If I use particle decomposition instead, no crash occurs, but the
simulation is parallelized very poorly, e.g. 2 times speedup by
increasing number of cores from 8 to 64.

Any ideas and solutions, please? Note, changing the force constants
for restraints does not change the situation, so I would guess the
problem is not just the system blowing up, as suggested at

http://www.gromacs.org/Documentation/Errors#Software_inconsistency_error.3a_Some_interactions_seem_to_be_assigned_multiple_times.

-- 
Dr. Vitaly V. Chaban
MEMPHYS - Center for Biomembrane Physics
Department of Physics, Chemistry and Pharmacy
University of Southern Denmark
Campusvej 55, 5230 Odense M, Denmark