AMBER Archive (2008)
Subject: Re: AMBER: Sources of disruption during NVE
From: Gustavo Seabra (gustavo.seabra_at_gmail.com)
Date: Wed Oct 01 2008 - 14:32:06 CDT
There are a number of points of points to consider there:
> I was seeing variability of my absorption coefficients based on: simulation
> time, translation removal interval, integration step and basically any other
> parameter I changed.
It's not clear what you mean by variability. How large is the difference?
> So I performed the following production runs (as described in the tutorial I
> have performed 20ps of NVT to reach 300K, and 100ps NTP to reach ~1gm/cm^3):
> -100ps simulation time, 2fs integration step, NPT, translation removed every 2fs
> -100ps simulation time, 2fs integration step, NPT, translation removed every 10fs
Those simulation times are *very* small, and it's likely that your
system isn't even close to equilibrating yet. How large is it? Did you
look at any convergence properties to make sure it's OK? It is
possible that the differences you see are just the consequence of not
having an equilibrated ensemble.
> The result of every experiment was different, however when each set was
> average they produced almost the same result, and this absorption spectra
> was very close to the experimental data. The NVE results seemed to represent
> the experimental data better (I can sent the plots).
That's confusing... How different is "different"? In principle, to
extract any observable from an md calculation you *must* average it
over the simulation time. So, since you mention that the average
gives about the same result, maybe they aren't as different... BTW,
how did you do this averaging?
> There seems to be some source of noise in the system. The noise is large
> enough to mask the results until averaging cancels out the random noise.
This looks just like you are plotting the instantaneous value of your
property, which cannot be directly compared to experiment: It is
*expected* to fluctuate. It is the time-average that can be related to
> Therefore, I tried simulating in NVE regime w/ no translation
> removal (also in this amber archive discussion
> http://archive.ambermd.org/200803/0158.html). [...]
The discussions you mention are about a completely different thing (I
believe). There, what's being discussed is energy conservation, which
doesn't seem to be your issue.
> The only other disturbance I could think in NVE is the SHAKE procedure used
> to constrain hydrogen vibrations. However, there seems to be multiple views
> about what happens when SHAKE is turned off. In these discussions a very
> short time step ~1fs or .25fs is recommended
> However at the same time in this discussion
> (http://archive.ambermd.org/200508/0125.html) it is mentioned that "Amber
> force fields are not really designed to give proper dynamics for bonds
> involving hydrogen. In fact one of the reasons to constrain them is that
> they really are not described well by classical potentials." Does this mean
> that using TIP3P or SPC model w/o SHAKE would be meaningless anyways?
> Because the large van der Waal centered on oxygen atom facilitates hydrogen
> bonds only between water molecules and not between water and solute i.e.
> hydrogens from one water molecules would fly off and cluster around the
> oxygen atom of another water molecule?
It is true that the TIP3P water model was *designed* to be a rigid
water, thus it *must* be used with some sort of SHAKE algorithm (see
DOI:10.1063/1.445869). Additionally, as you mention, hydrogen atoms
are not well described by classical potentials, so it's always a good
idea to use shake. You can, however, try to turn off SHAKE only for
your solute, see the manual for "NOSHAKEMASK". If you do this, then
you certainly must change your time step to something smaller than the
vibrations of the bonds containing Hydrogens in your molecule.
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)