[gmx-users] Compressibility issue - real concern

[Berk Hess]

>From: "Jones de Andrade" <johannesrs at gmail.com>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: "Discussion list for GROMACS users" <gmx-users at gromacs.org>
>Subject: [gmx-users] Compressibility issue - real concern
>Date: Thu, 3 May 2007 06:33:12 -0300
>
>Hi all.
>
>Well, I'm having a great problem in some NpT simulations I'm running here,
>which leaded me to some concerns. Let me try to explain the situation:
>
>Ok, I know: everywhere, including this list, amber list and any
>computational chemistry list informs that the compressibility of the liquid
>doesn't really matters for the NpT simulations, it will just change the
>"response time" of the system.
>
>Knowing that, I've started to make some simulations on a new system here:
>CF4. NVT equilibration, box in the proper volume, 500 molecules. After long
>time, change to Parrinello-Rahman barostat. Almost blowed, but came back as
>fast as almost were lost.
>
>I was using tau_p of 0.400 and compressibility of 116 e-6 (not even I
>remember where that came from). The target volume is 45.66nm**3 at 123K and
>1 atm. The potential is from AMBER, and is properly written in the topology
>files (reverified that about a thousand times).
>
>The system kept going to a average volume around 53.4 nm**3, didn't matter
>whatever I did with the simulation parameters. The reproducibility, from 
>the
>original Kollman's paper, show's that there should be about 1-2% deviation.
>I kept trying for about 1ns of simulation, couldn't be more stable (it 
>would
>be great if it was in the correct value!).
>
>After all reverifications possible, I decided to change exactly what is
>supposed to not alter this property: compressibility. I lowered it from
>11.6e-5 to 4.5e-5.
>
>The volume reacted with an immediate drop to 50nm**3.
>
>I did not yet tried to change parrinello for berendsen. By now, I'm dealing
>with the tau_p increase, from 0.400 to 0.700, which also yelded me an
>reduction from 50nm**3 to 49-49.5nm**3. This means that it's not big deal 
>to
>change the tau_p, what is also interesting since amber official and 
>original
>guidelines uses the 0.400 ps coupling for pressure (only that with the nosè
>barosthat).
>
>The points I'm questioning here now is:
>
>1 - Is the compressibility of the system a value that basically should "not
>matter" what I put in there?
>
>2 - to what range is this "not matter" related?
>
>3 - I'll keep trying the compressibility, and see if will reach at some
>value the expected volume. The question now is, will it reach any point
>acceptable? And after this point, will it be still stable, start to grow 
>the
>volumes again, or will just keep falling (worst scenario)?
>
>4 - since it should not affect the simulated property... what's happening
>here, in such a way that it IS affecting the simulated property?
>
>I'm really sorry if this is any kind of obvious subject, but I really tried
>to find any different reason for these observations. Could not find nothing
>related in any list, including this one. So I thought about taking the 
>facts
>to the discussion list and see what appears.
>
>Thanks in advance for anyone and any answer!
>
>Sincerally yours,
>
>Jones

It's all in the manual.

The compressibility beta does matter.
With Berendsen the scaling formula included beta/tau,
with parrinello-rahman beta/tau^2.
Only these ratio matter.
So changing beta changes tau.

I don't know how AMBER sets the mass, but for Gromacs tau=0.4
gives oscillations with a period of 0.4 ps which is far to fast
and will lead to instabilities. I would use something like tau=4 ps.

Berk.

_________________________________________________________________
Play online games with your friends with Messenger 
http://www.join.msn.com/messenger/overview