RE: Is it possible to perform NVT FEP on a fluid system?

From: Radak, Brian K (bradak_at_anl.gov)
Date: Tue Aug 09 2016 - 11:51:05 CDT

Unfortunately, overlapping results in both directions does not mean that the result has converged (it is a necessary, but not sufficient, condition). That test can only be used to obtain a negative result. It's possible that the NpT result only looks converged? You can imagine running a very short simulation in both directions and trivially getting very similar results.

I would strongly recommend a soft-core potential in this case (this is on be default in NAMD). I suppose the problem might become worse for large epsilon since the repulsive wall becomes steeper.

It's possible that pressure control is doing weird things here, especially if the density for NVT was not chosen according to a long NpT equilibration (or at least the results could/should be massively different in that case).

25 K sounds very low - I'd imagine convergence for anything is very slow at such a temperature, but again, that's just my intuition. I've never simulated anything below 290K.

HTH,
Brian

Brian Radak
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory

9700 South Cass Avenue, Bldg. 240
Argonne, IL 60439-4854
(630) 252-8643
brian.radak_at_anl.gov

________________________________________
From: B.W.J. Irwin [bwji2_at_cam.ac.uk]
Sent: Tuesday, August 09, 2016 11:23 AM
To: Radak, Brian K; namd-l_at_ks.uiuc.edu
Subject: RE: namd-l: Is it possible to perform NVT FEP on a fluid system?

Thanks Brian,

So, I used the ParseFEP plugin with the BAR estimator for the FEP
analysis.

The first plot to consider is change in free energy as a function of
lambda window.
When I carried out the entire experiment in the NpT ensemble, the
forward and backward functions overlapped near perfectly.

For the NVT ensemble there is a large mismatch between forward and
backward functions when the fixed neon is mostly present,
this mismatch closes in lambda windows where the fixed neon is mostly
annihilated.
I ran 64 lambda windows each for 24 ns (with all the usual endpoint
controls), and the mismatch did not close substantially.

The second plot is the overlapping probability distributions for each
lambda window.
For NpT, the overlaps are almost perfect.
For NVT, some distributions have multiple peaks (bimodal), and the shape
of the forward backward distributions do not match well, but get better
as the central atom becomes annihilated.

Standard forcefield for the solvent neons. I have increased the vdw
epsilon parameter on the fixed neon to 2.5, 5.0, 7.5 and 10 times the
original value as part of my experiment but kept the rmin parameter the
same. The NpT FEP experiment 'converged' (as above) for each of those
settings. The fixed particle was also a key aspect in the experiment I'm
trying to perform.

I didn't explicitly equilibrate the density, so the volume of the box is
slightly larger than the equilibrated volume at 1 atm.

I should probably mention I reduced the temperature to 25K, might this
have an impact?

Many thanks,

Ben

On 2016-08-09 16:22, Radak, Brian K wrote:
> Hi Ben,
>
> Can you elaborate on "does not converge well"? Is the convergence just
> slower? If so, how are you judging this? Is the system unstable? Are
> your force field settings standard or is anything unusual going on
> (besides the fixed particle, which is likely unnecessary)?
>
> I exclusively use NVT for free energy computations, but don't do
> annihilation that often.
>
> In general I would expect the NVT protocols to be simpler and more
> stable than the NpT protocols, but that's just a gut feeling. Also,
> there is some fudginess about how the kinetic virial is computed when
> using the alchemy modules (and probably Cartesian constraints also),
> so I try to avoid this when possible. Your idea to equilibrate the
> density first and then turn on alchemy is, in my opinion, the right
> one.
>
> Brian
>
>
> Brian Radak
> Postdoctoral Appointee
> Leadership Computing Facility
> Argonne National Laboratory
>
> 9700 South Cass Avenue, Bldg. 240
> Argonne, IL 60439-4854
> (630) 252-8643
> brian.radak_at_anl.gov
>
> ________________________________________
> From: owner-namd-l_at_ks.uiuc.edu [owner-namd-l_at_ks.uiuc.edu] on behalf of
> B.W.J. Irwin [bwji2_at_cam.ac.uk]
> Sent: Tuesday, August 09, 2016 5:22 AM
> To: namd-l_at_ks.uiuc.edu
> Subject: namd-l: Is it possible to perform NVT FEP on a fluid system?
>
> Dear all,
>
> I am trying to perform an FEP calculation on a box of neon with a fixed
> central neon atom which annihilates.
> It works well in the NPT ensemble and convergence is reached, but when
> I
> remove the pressure regulation and set the volume, the corresponding
> NVT
> calculation does not converge well. I would rather keep the volume of
> the box constant.
>
> I read in "Re: Query Regarding FEP tutorial: NPT or NVT?", "use NPT for
> solvated system and just constant temperature for in vacou
> simulations".
>
> Is it possible to perform NVT FEP on a fluid system and reach
> convergence with NAMD?
>
> Thanks in advance,
>
> Ben

This archive was generated by hypermail 2.1.6 : Sun Dec 31 2017 - 23:20:37 CST