AMBER Archive (2009)
Subject: Re: [AMBER] nanoseconds of simulation with nscm=0
From: Naser Alijabbari (na3m_at_virginia.edu)
Date: Sun May 24 2009 - 14:56:28 CDT
Dr. Cheatham, if DSUM_TOL and TOL are set to 10^-8 for example, and a 1 step
MD simulation is used with NSCM=1 with a binary rst file (which doesn't
exist in AMBER 10), before a production run in NVE with a 1fs integration
step and ntt=0, do you think one can run a simulation for nanoseconds
without having the molecule leave the periodic boundary? Or would the energy
somehow couple to an overall translation motion of the molecule no matter
what? Also what if the molecule crosses main box in PME simulation. Does it
violate a conservation of angular momentum? I am not sure what "... in the
case of Langevin, shift the center of mass to zero (or the center of the
box)" would mean.
On Sun, May 24, 2009 at 2:27 PM, Thomas Cheatham <tec3_at_utah.edu> wrote:
> > I am trying to understand where center of mass translation and rotation
> > form during a simulation. I came across the following threads
> > http://archive.ambermd.org/200705/0176.html
> > http://archive.ambermd.org/200601/0154.html. I have never have a
> > start in NVE without any COM. Usually there is always some COM, and if I
> By COM I assume you mean center of mass motion (not just center of mass).
> > understand the thread correctly this is due to accumulation of error when
> > integrating equations of motion. So in a nanosecond simulation the
> > would move away until coordinates would be too big to write to file. Have
> > others been able to start a simulation without any COM?
> No, only the growth in center of mass motion is due to lack of complete
> energy conservation (energy drains). The initial center of mass motion in
> a simulation is simply due to random temperature generation; it is highly
> unlikely if you pick a set of velocities at random to match a particular
> temperature distribution that the net-translation (or also rotation in a
> non-periodic system) will be exactly zero. This will also be true at
> every step in Langevin dynamics since random forces are added (however
> since they are random, they should not as easily grow if energy
> conservation is reasonable).
> The net-translation is likely small, but can grow due to lack of complete
> energy conservation. Where do we lose energy? Berendsen pressure
> coupling; SHAKE tolerances that are not stringent enough; larger time
> steps; PME that is not sufficiently accurate; cutoffs in GB; etc.
> You can get rid of initial COM motion only at periodic intervals through
> the NSCM variable (although I note that it used to be that it was also
> removed when not restarting a run however this behavior has changed). If
> you want it removed at the initial step, run 1 step of MD with NSCM=1 and
> then use the restrt file (IREST=1) for subsequent MD. Note that due to
> finite precision in the formatted restrt file, this could still lead to a
> (likely smaller) non-zero COM motion. To avoid that would require binary
> Note that keeping NCSM set to some interval is helpful as it will report
> energy problems or in the case of Langevin, shift the center of mass to
> zero (or the center of the box).
> Let me know if this doesn't answer your question sufficiently...
> AMBER mailing list
AMBER mailing list