[gmx-developers] (no subject)

[H.J.Risselada]

Dear gmx-developers,

I discovered a bug in GROMACS-3.3 in mdlib/update.c 
concerning the usage of constraints.

I discovered that the temperature on the slaves during a 
parallel run was 30 K lower at ref_t 300 K then on the 
master at all times.

I used a LJ-particles only test system without constaints:

Constaints where put on "none" in the mdp file.

However,

  /* if edyn->bEdsam == FALSE --- this has no effect */
     bHaveConstr = bHaveConstr || ed_constraints(edyn);

returns FALSE on the master, but TRUE on the slaves.

This means that the constrainst loops become active on the 
slaves, which will interfere with the dynamics.

When T-coupling is used, one temperature group can be both 
devided on the master as on the slaves. The total mean 
temperature of the T-group will be the desired 
temperature.
However the desired maxwell distribution will be 
disrupted.

This Bug may have a significant effect on the users of 
GC-models.

I tested it for the version 3.3, but it may also be 
present in the older versions.

greetings

Jelger.