On Sat, Feb 25, 2012 at 9:26 AM, case <case.biomaps.rutgers.edu> wrote:
> On Sat, Feb 25, 2012, Dilraj LAMA wrote:
>
> > I was running a simulation which got stalled after a
> > period due to some reason. I was saving the restart file (.rst) every
> > one nanosecond. Hence I restarted the job with the latest restart file
> > as input. Then I compared the energy values in the beginning of the
> > new run and the value at the same time step in the old simulation.
> >
> > I was expecting the energy values to be identical. However they are
> > not. Below are the values from the two simulation run.
> >
> > NEW SIMULATION:
> >
> > NSTEP = 500 TIME(PS) = 32531.000 TEMP(K) = 301.10 PRESS =
> 39.9
> > Etot = -94460.3737 EKtot = 23209.8266 EPtot =
> -117670.2004
> >
> > OLD SIMULATION:
> >
> > NSTEP = 16000500 TIME(PS) = 32531.000 TEMP(K) = 302.19 PRESS =
> -128.7
> > Etot = -94254.9707 EKtot = 23294.4062 EPtot =
> -117549.3769
> >
> > I would expect the values to be identical in both cases since I am
> > restarting my simulation with the same coordinate and velocity values
> > at this particular time step supplied through restart file.
> >
>
> I think Dan's reply is wrong in this case. The more likely culprit is that
> saving to a restart file, then restarting loses some precision. When
> running
> MD, small differences in starting coordinates (or velocities) quickly
> produces
> divergent trajectories. This is discussed in the "testing" section of the
> Users' Manual.
>
You'll also get wildly different results if you're using a Langevin
thermostat (or some other stochastic control method, if there are any),
since the random number stream will be different (regardless of what value
you set for "ig"). However, the effect Prof. Case mentioned is more than
enough to cause divergence very quickly (well under 500 steps) even if you
are not using a stochastic thermostat.
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Feb 25 2012 - 15:00:01 PST
