Hello Amber Devs,
Hoping that someone can help me understand why energy drift is going in
different directions for multiple copies of the same system. The only
thing I can think of is that minute changes in the box volume between the
starting positions for my NVE dynamics are causing it. But, here's what I
see:
- I have taken the JAC system from our benchmarks and re-cast it with
SPC-Fw water so that I can run without SHAKE (or even SETTLE). I am
running with a 0.25fs time step (always), a 96 x 96 x 96 grid (the default
would be 64), and a direct sum tolerance of 1.0e-5 (this is deliberate--I
could dial this down but I want to see what happens with it on a higher
setting).
- I re-equilibrated the system at 300K in NPT for 1ns, four different times
with different random seeds. Changing the water model and nixing SHAKE
changed the system, so I had to bleed off that excess energy (and the
volume had to change half a percent or so), even though it was evenly
distributed and did not require restrained minimization.
- When I run the resulting systems, they all have nearly the same densities
but the box volume does differ by perhaps a hundredth of a percent. The
total energy of each system can differ by as much as 300 kcal/mol out of
54000 kcal/mol, starting temperatures by 2-3 degrees. All within
expectations.
- When I run with a 12A cutoff, I can stamp out the energy drift. Flat as
a pancake. Well, 0.2 +/- 0.2 kcal/mol-ns and temperature drift being a few
degrees per microsecond.
- When I run with a 9A cutoff, I see energy drift, a very steady drift over
tens of ns, but the slope of the drift I see is different for each copy of
the system. For each of the four systems, I get -7.89, 7.85, 0.35, or 0.28
kcal/mol-ns drift. I can go back to 12A cutoff and stamp out the drift,
then run more with a 9A cutoff and see the drift pop back up, nearly the
same value for each system as it was the first time. There really seems to
be something that makes each system drift at its own rate.
- What I'm really looking to study is the effect of allowing particles to
travel beyond the point at which the pair list could be violated. If I let
things travel to the maximum of my new setting, this changes the inherent
energy drift of each copy. In this case, the effect always seems to be
positive, by about 2-3 kcal/mol-ns, in other words about half of what the
drift might be if we just run with a 9A cutoff. It's a consistent effect
that heats the system as we would expect.
It seems like I can see the signal I want in spite of the noise, but I want
to understand why the noise is there in the first place.
Thanks,
Dave
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Aug 02 2018 - 23:00:02 PDT
