[gmx-users] Re: Free Energy Calculation
floris_buelens at yahoo.com
Sun Aug 9 15:31:36 CEST 2009
> The question is:
> Do you trust you simulations when you/others do not find significant problems
> or do you only trust your simulations when you are sure the integrators etc
> are implemented exactly correctly or at least the magnitude of the (small)
> errors are known.
> Personally I (and the majority of people) fall in the first category.
> David is clearly in the second category.
I think you can recognise free energy calculation specialists like David and myself by a heightened level of paranoia, along with other signs like staring blankly at the wall wondering which part of your protocol could be causing that 5 kcal/mol deviation... so while I'm a big supporter of the first category I see where David's coming from, having spent a lot of time in the past running controls for almost certainly inconsequential variables like this.
If I can offer my two eurocents-worth, I would sleep easier at night with a barostat that's both well-behaved and produces a rigorous NPT ensemble... I'm in the process of migrating to GROMACS from NAMD and have never heard any complaints about their Langevin piston thermostat (http://dx.doi.org/10.1063/1.470648), would there be any downsides to having this method in GROMACS?
From: Berk Hess <gmx3 at hotmail.com>
To: Discussion list for GROMACS users <gmx-users at gromacs.org>
Sent: Friday, 7 August, 2009 16:16:19
Subject: RE: [gmx-users] Re: Free Energy Calculation
All the discussion is mainly a matter about how paranoid you are.
A Berendsen barostat does not generate a proper ensemble.
But it does get the average pressure right. The questions is how large
the effects of the "ensemble error" is. I would say that in practice
this would be negligible compared to fluctuations, force field errors,
insufficient sampling etc. But strictly speaking it is there.
Parrinello-Rahman in principle gives you a proper NPT ensemble
(combined with a proper thermostat). But to have a real symplectic
or even only reversible integrator is difficult and might be computationally
expensive, especially when constraints are present, as will nearly always
be the case with most water models.
Again errors of a non-reversible integrator are, in most cases, extremely small
and smaller than the other errors you make.
In addition PR-coupling might produce nasty volume oscillations.
The question is:
Do you trust you simulations when you/others do not find significant problems
or do you only trust your simulations when you are sure the integrators etc
are implemented exactly correctly or at least the magnitude of the (small)
errors are known.
Personally I (and the majority of people) fall in the first category.
David is clearly in the second category.
In the past decade it has been shown that several algorithms that had been
used a lot caused serious artifacts. But I would say that we are currently
in a state where most things have been checked and although we know
there are some errors (BTW descretizing analytic differential equations also
produces errors), these errors have no effect on MOST of the things we are
interested in. Of course you should not derive compressibilities using
a Berendsen barostat, as a big name in the field has done even not to long ago.
Date: Fri, 7 Aug 2009 07:32:33 -0700
From: floris_buelens at yahoo.com
Subject: Re: [gmx-users] Re: Free Energy Calculation
To: gmx-users at gromacs.org
Hi David and Berk,
I guess the best proper solution would be to use a Parrinello-Rahman barostat.
>But in Gromacs this is currently not implemented in a reversible way.
> Yes, I would love to see a barostat that samples the correct distribution.
Can you clarify? The wording in the manual implies that Parrinello-Rahman coupling samples the correct distribution... When you say it's not reversible do I understand right that means you can't run your simulation backwards and get an identical result, if yes, is this ever relevant in practice for e.g. solvation free energy calculation? David, why don't you use Parrinello-Rahman coupling?
>In practice I think the issues with the Berendsen thermostat are not relevant.
>It does not generate the correct pressure and volume fluctuations,
>but the average pressure and volume are correct. The effects of incorrect
>>fluctuations is negligible and is surely much smaller than simulating at the wrong
>volume, which can have a large effect on solvation free energies.
It seems fairly like you are right (assuming your first line is meant to refer to the Berendsen barostat), but I think no one knows for sure at this point whether it matters and when it might matter. I agree that simulating at an incorrect volume can be a big problem, however. This is unlikely, but can happen, which is why we now do the affine transformation I just mentioned when looking at solvation free energies.
Anyway -- I'd just like to see a proper barostat implemented at some point, then the paranoid (like me) can run with a proper barostat instead of worrying about whether or not this matters, AND someone could test whether observables they are interested in come out different depending on the choice of barostat.
>I also think that issues with Berendsen pressure coupling are much less severe
>>than the issues with Berendsen temperature coupling. Berendsen tcoupl can have
>significant effects on your ensemble. But in Gromacs 4.0 this can easily be avoided
>by using the v-rescale thermostat.
Subject: RE: [gmx-users] Re: Free Energy Calculation
>Date: Thu, 6 Aug 2009 16:01:04 -0400
>From: PNAGY at UTNet.UToledo.Edu
>To: gmx-users at gromacs.org
>A question to the answer below:
>> If the Berendsen barostat does not sample correctly the distribution of pressures
>>then what guarantees that the the volume/density is correct
>>at the end of the equilibration phase when the user is advised to switch to
>>The University of Toledo
>>From: gmx-users-bounces at gromacs.org on behalf of David Mobley
>>Sent: Thu 8/6/2009 2:38 PM
>>To: Mauricio Carrillo Tripp
>>Cc: Discussion list for GROMACS users
>>Subject: [gmx-users] Re: Free Energy Calculation
>>Sorry for the delay answering. These questions are better put on the
>>GROMACS users list.
>>1) Yes -- the Berendsen barostat does not sample the correct
>>distribution of pressures.
>>2) Regenerating velocities is fine for a couple of reasons: (a) we are
>>after thermodynamics, not dynamics, and (b) regenerating velocities
>>does not de-equilibrate the system (see the Andersen thermostat, for
>>On Jun 8, 2009, at 3:02 PM, Mauricio Carrillo Tripp wrote:
>>> I've been following directions at http://www.dillgroup.ucsf.edu/group/wiki/index.php/Free_Energy:_Tutorial
>> to calculate hydration free energies of a few small molecules, and I
>>> have a couple of questions:
>>> 1) Is there a reason for using NVT instead of NPT for the production
>>> 2) I noticed that between the equilibration and the production runs
>>> you generate velocities at each step,
>>> instead of using the ones from the previous step. Is there a
>>> reason for doing this? wouldn't this
>>> 'de-equilibrate' an already equilibrated system?
>>> Thank you.
>>> 0 | Mauricio Carrillo Tripp, PhD
>>> / | Department of Molecular Biology, TPC6
>>> 0 | The Scripps Research Institute
>>> \ | 10550 North Torrey Pines Road
>>> 0 | La Jolla, California 92037
>>> / | trippm at scripps.edu
>>> 0 | http://www.scripps.edu/~trippm
>>> ** Aut tace aut loquere meliora silentio **
What can you do with the new Windows Live? Find out
>>gmx-users mailing list gmx-users at gromacs.org
>>Please search the archive at http://www.gromacs.org/search before posting!
>>Please don't post (un)subscribe requests to the list. Use the
>>www interface or send it to gmx-users-request at gromacs.org.
>>Can't post? Read http://www.gromacs.org/mailing_lists/users.php
See all the ways you can stay connected to friends and family
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the gromacs.org_gmx-users