[gmx-users] why gromacs cannot repeat out the results?

[David van der Spoel]

On 11/06/15 03:59, 范聪 wrote:
> Hello, everyone!
>   I've run two molecular dynamics using gromacs with exactly the same inputs, in the very same computer, and there's no radom process such as velocity distribution.( I've used "mdrun -t nvt.cpt" to stop a new velocity distribution)
>   Now I'm puzzled that gromacs have output two results of great differences, both in the trajectory and in the energy calculation results. Now I don't know which one to trust.
> I've learned that Molecular Dynamics software solve Newton equitations to decide the movement of atoms, and calculate the potential energy according to the location of each atom.
> So how could it happen that the very same inputs generate so different results?
> Could there be some way to  avoid all above?

The simulations are still stochastic to some extent due to 
parallellization and dynamic load balancing which is dependent on timing 
the routines. There is a flag to maximize reproducability of the 
simulation (mdrun -reprod IIRC) but it is probably not needed.

You say nothing about the system and about the difference in energy. If 
the systems are in equilibrium the average energies should be the same 
in both simulations, within error bars of course.

>
>
>
>
>
>


-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
spoel at xray.bmc.uu.se    http://folding.bmc.uu.se