[gmx-users] g_energy reporting 0.0000 interaction energy between energy groups in GPU run?

[Leandro Bortot]

Dear users,

     I did some simulations of organic molecules in solution to study how
they interact, but I'm facing some problems and your help would be greatly
appreciated.
     Consider the following case: I simulated two copies of molecule "AcO"
starting from a structure in which they are non-covalently interacting. I'm
using the AMBER99 forcefield with GAFF. I've confirmed by visual inspection
that the molecules are interacting with each other, i.e. they didn't
separate during the simulation. Additionally, I've specified the energy
groups "aco_1", "aco_2", "SOL", "Ion" in the original MD tpr. The problem
is that g_energy reports all "aco_1-aco_2" energy terms as exactly
0.000000. The energy terms are OK.

     This was only an example. I actually simulated 2, 3 and 4 copies of
interacting molecules in different ways and this happens in all cases.

     These results were obtained using GPUs. I used the same .tpr to rerun
the .xtc on CPUs only and now the "aco_1-aco_2" energy terms are OK. In
both cases I used GROMACS 4.6.6.

     I've searched the list but found nothing. Is this a problem/bug? Am I
doing something wrong with the GPUs?


Thank you in advance,
Leandro