[gmx-users] neighbour search problem with simulation restarts

[michal.kolar at marge.uochb.cas.cz]

Dear users,
I run an NVT Hamiltonian replica exchange simulation with Plumed patch.
When restarting the simulation using cpt file I sometimes experience
sudden jumps in some energy components. The Coulomb energies are affected
most notably.

I've found out that these changes occur when a different neighbor-search
grid is used after the restart, possibly due to particular charge
distribution (mdrun guess). Also, the domain decomposition -rdd and -rcon
distances are slightly different. Thus I feel it is Gromacs rather then
Plumed related problem.

Is there any reason for such behavior?

May it be related to any of known bugs?

I use PME eletrostatics with group cut-off scheme, gmx 4.6.7 compiled with
Intel 12, run with MPI-only parallelization.


with best

Michal