Re: question about equilibration step

From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Fri Aug 24 2007 - 10:09:39 CDT

Hi Audrey,
as I frequently do on this list, I'd highly recommend Leach's /Molecular
Modeling: Principles and Applications/ as a basic text on these
simulation methods; I can almost guarantee that reading through the MD
relevant chapters will teach more than you can get from a mailing list.
>
> ** In other words, I should have a .conf file that I use with the
> "minimize" command and a separate .conf file that I run with the "run"
> command? At this point let's say that I want to run the equilibration
> in NVT, I would specify the temperature using the "temperature"
> command. Then, if I want to run the MD in NVE then I would remove the
> "temperature" command in a separate conf file?
There's no need to have separate config files for minimize and run. The
minimizer is fundamentally different from any type of run; all it does
is follow a specific algorithm for changing the atomic coordinates to
search for a local minimum in potential energy, so there's no real
correlation between "minimization steps" and the time steps that you
have during an actual run. It's perfectly acceptable to have

minimize 5000
reinitvels 300
run 5000

in a config file (you see this sort of thing in the namd tutorial, for
example).

Also, you have to be careful. The namd temperature argument only
specifies the initial temperature for the system. It *does not* turn on
a thermostat (which is necessary for an NVT simulation). To do that, you
need to use one of the actual temperature control methods in NAMD
(http://www.ks.uiuc.edu/Research/namd/2.6/ug/node30.html). Again,
examples are included in the tutorial, which uses Langevin dynamics for
temperature control. If all you do is include a temperature command,
you'll still be running NVE. Also, although temperature sets the initial
velocity distribution, minimization by its nature reduces all velocities
to zero, so you need to reinitialize the temperature distribution
(reinitvels) prior to dynamics if you expect your system to start out at
a nonzero temperature.

> I think that in order to have an equilibrated system, I am looking for
> the energy to be equally partitioned because it is possible to have a
> constant total energy, but have only equilibrated bond stretching
> energies, but not bond bending, for example, because the time period
> required for equilibration is longer for bond bending. It is possible
> for the total energy in the system to remain constant while the bond
> bending equilibrates because equal loss and gain of energy from bond
> stretching to bond bending would result in no total energy change.
>
> ** Am I correct in my understanding? If so, how can I be sure that the
> simulation is, in fact, equilibrated? Are there any references that I
> should look at?
The Leach reference above, or any other textbook covering MD, is a good
place to start. The purpose of the initial equilibration period is just
to ensure that your system is properly coupled to any temperature and
pressure controls imposed on it, and that any fast, initial
conformational changes due to the input coordinates not being terribly
stable in the forcefield that you're using are ironed out. If you're
going to run an NVE simulation, it is still a good idea to first run an
NVT or (preferably) NPT equilibration (or both) so that your system has
a realistic volume and temperature when you start your production run.
Aside from energy, I'd suggest looking at the following measures for
equilibration:
-Temperature (if in a constant temp. ensemble) should be stable and correct
-Volume (if in a constant pressure ensemble) should be stable
-RMSD of your solute to its initial structure should have leveled off
(this is a good indicator for when any of those rapid, initial
conformational changes that I mentioned have finished)

You can, of course, plot your BOND and ANGLE energies (for example)
individually and make sure that they, too, have stabilized, but you
should find that equillibration between different energy types occurs
much faster than equilibration of system-wide observables. There will,
of course, be significant fluctuations in all of these energies,
including your total energy if you're running in a constant temperature
ensemble.

Peter
>
> I hope that I have been clear. Thank you for your help.
>
> Thank you in advance.
>
> Audrey
>
> --
> Audrey L. Salazar
> Amaral Research Group
> Dept. of Chemical and Biological Engineering Phone: 847.491.2188
> Northwestern University
> Evanston, IL USA

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:45:08 CST