AMBER Archive (2007)Subject: Re: AMBER: about coordinates ****** and diffusion
From: Thomas Cheatham (tec3_at_utah.edu)
Date: Sun Jan 28 2007  15:09:00 CST
> The simulations I was running were a 5 nanoseconds
> ones. Fortunately, in the end, I was able to calculate
> diffusion constants without any problem (what was
> really my goal).
>
> But now you have mentioned the problem related to
> rapid diffusion on short time scale, I realised about
> a "strange" behaviour of the meansquaredisplacement
> (MSD) paremeter on friday evening: The MSD plot versus
> time was really amazing. As time advances the MSD
> value grows even fastly. I expected a "reasonably"
> linear behavior of MSD versus time as the slope is the
> Diffusion Coefficient. Instead, I obtained an almost
> "quasiperfect" parabolic shape with no traces of
> fluctuations at all!!!
...
> > > The original calculation is a NPT Molecular Dynamics
> > > in CHCL3 solvent and box is a truncated octahedron
One possibility is that you are seeing growth in center of mass motion of
the overall periodic box; if energy is not conserved fully, and
temperature coupling is on (as it was since your simulations were NPT), as
the temperature drops, it is scaled back up. This scales all the
velocities up, including the net translation of the entire unit cell (that
is if there was a finite initial velocity of the whole system upon the
initial velocity assignment which is highly probable). If the center of
mass translation is not removed (NSCM=500) periodically, and energy is
draining, this will grow and grow and get faster and faster... This could
explain the MSD plot.
Check to see that NSCM is set to some finite value.
Also, check for energy conservation in your liquid with NVE simulation.
Potential sources of energy drain are:
(1) SHAKE tolerances are too big (TOL=0.0000001 is a good value)
(2) Berendsen pressure coupling (set coupling time to be large to minimize)
(3) Time steps that are too large (maybe DT=0.002 is too big for your liquid?)
(4) Inaccuracies at the cutoff (run PME w/ appropriate tolerances; maybe
run bigger cutoff)
If indeed the motion was gaining that rapidly in a MSD plot (i.e. over 5
ns) likely the energy is not being well conserved and you should see this
rather quickly in NVE simulation.
Good sampling,
tec3 @ utah.edu

The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
