Re: question about equilibration step

From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Fri Aug 24 2007 - 20:36:54 CDT

Hi Audrey,

> Langevin temperature control is performed by randomly selecting a
> particle and changing its velocity (both magnitude and direction)
> leads to problems with momentum conservation.
Just a clarification -- there's no "random" selection of particles here.
*All* particles in the system move according to the langevin equation,
and thus they all face a frictional force opposed and proportional to
their velocity, and a random (fluctuating) force that will be applied to
each particle. You are correct that this means that momentum is not
conserved, though.
> This is a problem always in MD, but it is accentuated more in Langevin
> and Berensen (which i think is an off-shoot of Langevin) than in
> Nose-Hoover thermostats. I am not disputing that its use still gives
> applicable, physical results, just that relaxation times could be off
> and this is something that I will care about in my simulations. This
> leads to a few questions that I will mark with **
Berendsen really should be considered distinct from Langevin dynamics,
as it involves rescaling the velocities to push the temperatures in the
desired direction, rather than applying random forces. The presence of a
random force causes Langevin dynamics to sample phase space at different
rates because the random force can affect barrier crossing rates (see,
eg, Adcock and McCammon, Chem. Rev. 1589 (2006). With *any* of these
thermostats, you will (given sufficient time) properly sample the
canonical ensemble, but as Neema pointed out, rates will be affected.
>
> ** I have had much harder time coming up with good references to
> understand the Nose-Hoover Langevin piston. Any recommendations? Can
> I draw analogies from the Nose-Hoover thermostat?
For this, I can't really recommend anything better than the NAMD manual
entry (http://www.ks.uiuc.edu/Research/namd/2.6/ug/node32.html; scroll
down to Nose-Hoover Langevin piston pressure control) and the references
cited therein.
>
> ** I know that the type of temperature control used in equilibration
> doesn't appreciably affect the finding equilibrium. Am I right in
> assuming that they barostat will also not affect equilibration?
>
> ** Is there a way to employ a Nose-Hoover thermostat in NAMD? Is the
> same as the temperature coupling mentioned here:
> http://www.ks.uiuc.edu/Research/namd/2.6/ug/node30.html
> <http://www.ks.uiuc.edu/Research/namd/2.6/ug/node30.html>
I don't believe the Nose-Hoover thermostat is implemented in NAMD. NAMD
temperature coupling is equivalent to Berendsen's method.

Peter
> ?
>
> Thank you much.
>
> Audrey
> Thanks,
>
> Audrey
>
> On 8/24/07, *Peter Freddolino* <petefred_at_ks.uiuc.edu
> <mailto:petefred_at_ks.uiuc.edu>> wrote:
>
> Hi Richard,
> >
> > I don't know why one would want to run an equilibration in NVT and
> > then do your production run in NVE. I would think one would want to
> > be consistent and do the same the whole way through, either use NVE
> > entirely or NVT entirely. The way I look at is is what if one gets
> > unexpected results, then one could ask was it because I used two
> > different ensembles? If one does only use one ensemble, then
> it's not
> > likely that would be the cause of unexpected results.
> There is at least one very good reason to do this: If you equilibrate
> entirely in NVE, then the only influx of kinetic energy that the
> system
> gets will be from your velocity initialization. Almost invariably (and
> please do try this), a minimized system will then begin to
> fluctuate and
> in the process some of this kinetic energy is converted to potential
> energy, causing your temperature to drop. If you equilibrate in this
> way, your energy should be stable almost immediately, but the
> temperature that the simulation is occurring that is much lower than
> physiologically relevant temperatures. If, on the other hand, you
> equilibrate first in NVT, and do so until the temperature is
> stable, you
> end up with the system at an appropriate temperature *and* with an
> amount of total energy that is appropriate for a system in equilibrium
> with a bath at that temperature. You can then sever the connection to
> the bath and simulate in NVE if you wish, but you will have a more
> realistic initial combination of conformation and velocity
> distribution.
> Similar logic is why equilibrating in NPT is a good idea even if
> you're
> going to to production runs in NVT (this is what I frequently do):
> the
> volume that you initially choose for your system is almost
> certainly not
> the true volume that it should have at reasonable pressures, and you
> risk having bubbles or regions of abnormal density form in your
> simulation if you don't somehow allow the volume of the system to
> adjust.
> >
> > As I stated earlier, one generally plots out say the energy of the
> > system as a function of time, and if the curve is relatively flat,
> > then one can say the system is at an equilibrium. I'm not aware of
> > any other way to quantitate whether or not one's system is at
> equilibrium.
> >
> Please see my email from earlier this morning for some other criteria
> worth looking at.
>
> Best,
> Peter
>
>
>
>
> --
> 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 - 15:45:09 CST