[gmx-users] gromacs with v-rescale thermostat

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

Dear users,
recently I have observed a strange behavior of my simulations. The total
potential energy of presumably independent simulations showed undesired
correlation (R=0.6) even when started with different velocities
(gen-seed=-1) and solute coordinates.

I found out that the reason is the v-rescale thermostat, which uses random
numbers seeded by the ld-seed option in the mdp file. Unfortunately in
gromacs 4.5 and 4.6 series, the default value is 1993, which causes the
interdependence of the trajectories.

The solution is to explicitly set ld-seed = -1 everytime the v-rescale
thermostat is used. This is also the default value in gromacs 5+ versions.

Hope this helps.


with best regards


Michal