[gmx-users] error estimates of free energy calculations

[Berk Hess]

>From: "David Mobley" <dmobley 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: Re: [gmx-users] error estimates of free energy calculations
>Date: Fri, 11 May 2007 08:19:13 -0700
>
>Hi,
>
>Berk may weigh in on this, but...
>
>>To test my protocol, I have been trying to reproduce some of Berk
>>Hess's hydration free energy results from his 2006 paper (J. Phys.
>>Chem. B 110, 17616-17626, 2006). Although my obtained average free
>>energies are quite comparable, my error estimates are much higher
>>than the errors reported in that paper. I'd like to figure out what's
>>causing this difference in accuracy, because I really need optimal
>>accuracy for my future calculations.
>>
>>I'm using GROMACS 3.3.1, the G53a6 force field and 3.0 nm
>>dodecahedron simulation boxes with ~631 SPC waters. After
>>mimimization and 500 ps of equilibration for each of the 21 equally
>>spaced lambda-values, I simulated 5 ns for the propane, phenylalanine
>>and asparagine analogues. Theoretically, this should give me a
>>significantly better accuracy than the paper's 500 ps production runs
>>(approx. 1/sqrt(10) times better).
>
>Looks to me like Berk is using 42 lambda values in solvent, as he
>spaces them more tightly around the feature in the curve, and 29 in
>vacuum. This should give lower uncertainties.

I think this could already explain the difference.

When decoupling Coulomb and LJ at the same time you get nasty
behavior especially around lambda=0. I would guess nearly all
your error comes from this point and maybe some from the middle peak.
As David says, it is much easier to decouple separately.
If you want to do it at the same time, you need to use an optimized spacing.

Also you said you integrated the error directly?
This is not correct, you should integrate the variance.
That would make your errors even worse.

Berk.

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