[gmx-users] OpenMM

[Mark Abraham]

On 14/04/2010 7:10 PM, PACIELLO GIULIA wrote:
> Well,
> I have used the option -rerun to 'rerun' on CPU the simulation from the
> parallel code. As you can see in the following tables, results are very
> similar for non bonded terms, but not for the other energetic terms
> (these are avereges obtained from g_energy):

You haven't done a simulation. You've done an energy minimization. 
There's no reason to expect the same result for EM on different machines 
because there are so many different minima that might be found, and 
small numerical differences might mean you find different ones.

> PARALLEL CODE
> Angle: 4256.27
> Prop-di: 322.302
> R-B: 3101.14
> LJ-14: 2107.6
> Coul-14: 15212
> LJ(SR): -3410.22
> Coul(SR): -29368.3
> RF(excl): -9425.23
> Pot.En: -17204.5
> Kin.En: 5243.87
> Tot.En: -11960.6
>
> SERIAL CODE:
> Angle: 6089.56
> Prop-di: 466.121
> R-B: 3677.95
> LJ-14: 2234.85
> Coul-14: 15149.8
> LJ(SR): -3144
> Coul(SR): -29240.7
> RF(excl): -9425.25
> Pot.En: -14191.7
> Kin.En: 8262.96
> Tot.En: -5929.78
>
> I can't use the -reprod option, but I have noticed that the .gro files
> after the energy minimization are very similar (I have performed this
> minimization on the .gro files obtained running the parallel and serial
> codes)...systems are converging to the same equilibrium ensamble?

No, they've just done an EM. There's no velocities yet, so they're not 
in any ensemble. Equilibrate, and then look at ensemble averages of the 
equilibrated systems.

Mark