[gmx-developers] Re: Coulomb decoupling?

Tue May 7 09:02:19 CEST 2013

Hi David,

I can answer the question with regards to A and B states. With free energy calculations switched on, Gromacs does the reciprocal space PME calculation twice (two PME grids), corresponding to the A and B states, with respective energies and forces mixed linearly according your lambda. If your solute is coupled in the A state and decoupled in B, the A state grid represents solvent and solute, the B state grid solvent only. All periodic interactions are represented in the A state. The B state PME grid only represents solvent-solvent electrostatics.

Then as the manual says, extra explicit interactions are added between all solute atom pairs in the B state. These are straight Lennard-Jones and Coulomb potentials and are calculated without cutoff, so the nominal equivalent for the B state would be a simulation of the solute in vacuo with infinite cutoff and no periodicity.

I'm not as clear on how this all interacts with the dispersion correction. If the only issue is what Berk mentions, that the average C6 will be marginally off, I would expect an error that's not zero but certainly tiny. If on the other hand there are differences in the way the solute is decoupled from the long range correction going from A to B this could be significant.

Best,

Floris

________________________________
From: David Mobley <dmobley at gmail.com>
To: Discussion list for GROMACS development <gmx-developers at gromacs.org> 
Sent: Friday, 3 May 2013, 2:24
Subject: Re: [gmx-developers] Re: Coulomb decoupling?

Berk (and all),

On Thu, May 2, 2013 at 12:54 PM, Berk Hess <hess at kth.se> wrote:

This is all describes in the manual, AFAIK.
>
>
Sorry there was some overlap. I hadn't found the discussion I'll cite below yet. 

On 05/02/2013 09:44 PM, David Mobley wrote:
>
>Right, what I'm asking about is 
>>A) how exactly is this end result achieved? (The system is periodic, so how is the periodicity removed for the end state?)
The periodicity of intra-molecular interactions is always removed.
>These interactions are excluded from PME and added directly as
    listed pairs.

The manual says this: "All intra-molecular non-bonded interactions for moleculetype couple-moltype are replaced by exclusions and explicit pair interactions. In this manner the decou-
pled state of the molecule corresponds to the proper vacuum state without periodicity
effects.  "

Does this apply to BOTH the A and B states? Your 
answer "the periodicity of intra-molecular interactions is always 
removed" suggests you're saying that the solute is never allowed (in 
either A or B state) to interact with copies of itself. Doesn't this 
mean that that (considering the case of a small molecule in solution 
being decoupled) the A state has periodic interactions between all of 
the solvent molecules, and between solvent and solute, but no periodic 
interactions between solute and solute? (If so, won't this tend to leave the solvent box with a net dipole moment for PME purposes?) 

>
>B) how was it validated that it is working as it should be?
I checked this and it works.

What I'm asking is, "checked how", and "works for what"? Specifically I'm 
trying to figure out whether this can be expected to always yield the 
same results as annihilation (assuming simulations are converged), even 
for larger/flexible/more polar molecules. To that end I'm trying to 
understand whether there are any limitations in the formalism, and 
exactly how it's been tested.

>
>C) when you say, "without cutoffs", is this referring to just Coulomb cutoffs or also LJ? I'm assuming just coulomb. If so, then there are internal LJ interactions in the gas phase which are missing outside the LJ cutoff (assuming the molecule is larger than the cutoff). While these are also missing in solution, they are generally captured well by the dispersion correction. In vacuum that is not the case, so neglect of these could adversely affect solvation estimates, it seems to me. Has this been tested? How?
LJ is treated as Coulomb, plain LJ, no cut-off.
>
>
OK, thanks. 

D) how will the use of decoupling affect dispersion corrections to the energy and pressure? (Will the dispersion corrections still give the correct free energy contribution in decoupling?) how has this been tested, if at all?
This is the only complicating factor.
>There is no correct way of using dispersion correction with
    decoupling.
>As the intra-molecular interactions are excluded, these do not end
    up the average C6
>and they do not end up in the pair count for dispersion correction.
>

So, are you saying dispersion corrections should be turned off when using 
decoupling? Dispersion corrections tend to contribute substantially to 
free energies unless one runs with a large cutoff, which would suggest 
this is a bad idea. It seems like I probably need to know exactly how 
the dispersion correction contribution to the free energy is computed in the case of decoupling, so I can estimate how wrong this will be due to the pair count/average C6 being wrong. 

Probably this also raises the question of whether GROMACS should not allow the user to run decoupling with the dispersion 
correction (since it's not correct) or whether it instead should issue a warning and provide some guidance as to how to fix things (if we have 
any such guidance to offer).

Thanks,
David

On Thu, May 2, 2013 at 12:54 PM, Berk Hess <hess at kth.se> wrote:

This is all describes in the manual, AFAIK.
>
>
>On 05/02/2013 09:44 PM, David Mobley wrote:
>
>Right, what I'm asking about is 
>>A) how exactly is this end result achieved? (The system is periodic, so how is the periodicity removed for the end state?)
The periodicity of intra-molecular interactions is always removed.
>These interactions are excluded from PME and added directly as
    listed pairs.
>
>
>B) how was it validated that it is working as it should be?
I checked this and it works.
>
>
>C) when you say, "without cutoffs", is this referring to just Coulomb cutoffs or also LJ? I'm assuming just coulomb. If so, then there are internal LJ interactions in the gas phase which are missing outside the LJ cutoff (assuming the molecule is larger than the cutoff). While these are also missing in solution, they are generally captured well by the dispersion correction. In vacuum that is not the case, so neglect of these could adversely affect solvation estimates, it seems to me. Has this been tested? How?
LJ is treated as Coulomb, plain LJ, no cut-off.
>
>
>D) how will the use of decoupling affect dispersion corrections to the energy and pressure? (Will the dispersion corrections still give the correct free energy contribution in decoupling?) how has this been tested, if at all?
This is the only complicating factor.
>There is no correct way of using dispersion correction with
    decoupling.
>As the intra-molecular interactions are excluded, these do not end
    up the average C6
>and they do not end up in the pair count for dispersion correction.
>
>Cheers.
>
>Berk
>
>
>
>>
>>Thanks!
>>
>>On Thursday, May 2, 2013, Berk Hess wrote:
>>
>>Hi,
>>>
>>>You didn't explain exactly what you are doing.
>>>The decouple mdp options decouple the molecule to a vacuum
              state, i.e. pure Coulomb without cut-off's.
>>>
>>>Cheers,
>>>
>>>Berk
>>>
>>>On 05/02/2013 07:10 PM, David Mobley wrote:
>>>
>>>Could I get some input on this?
>>>>
>>>>I have a couple of cases for rather polar molecules
                where decoupling and annihilation give me statistical
                significant differences in hydration free energies. The
                differences are not that large, but significant. I'm
                trying to find out what's already been done to validate
                so I know how much time/effort to spend testing to try
                and figure out if there is a problem here.
>>>>
>>>>Thanks.
>>>>
>>>>
>>>>
>>>>On Tue, Apr 30, 2013 at 1:25 PM, David van der Spoel <spoel at xray.bmc.uu.se> wrote:
>>>>
>>>>On 2013-04-30 18:02, David Mobley wrote:
>>>>>
>>>>>Hi,
>>>>>>
>>>>>>In GROMACS 4.6 and later, there's now a new
                        feature available to allow
>>>>>>decoupling of solute molecules in free energy
                        calculations. I wanted to
>>>>>>inquire as to how Coulomb decoupling works, as
                        I'm not clear.
>>>>>>
>>>>>>Specifically, imagine I'm running a calculation
                        of the hydration free
>>>>>>energy of a small molecule in water, and I
                        decouple it (LJ and Coulomb)
>>>>>>from its surroundings. What is the final
                        reference state for the small
>>>>>>molecule? Is it the small molecule interacting
                        with periodic copies of
>>>>>>itself in the gas phase (bad)? Or is it not
                        interacting with periodic
>>>>>>copies of itself either? If the latter, how is
                        this achieved?
>>>>>>
Good question, also one would like to be able to decouple a molecule only in the central box and not in the surrounding boxes. This does not make a difference for liquids but it does for crystals.
>>>>>
>>>>>
>>>>>>Since I'm not familiar with the Coulomb
                        decoupling aspect and it is
>>>>>>conceptually more complicated than LJ
                        decoupling, I want to make sure I
>>>>>>understand how it's supposed to be working.
>>>>>>
>>>>>>Thanks!
>>>>>>David
>>>>>>
>>>>>>
>>>>>>--
>>>>>>David Mobley
>>>>>>dmobley at gmail.com <mailto:dmobley at gmail.com>
>>>>>>949-385-2436
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>>-- 
>>>>>David van der Spoel, Ph.D., Professor of Biology
>>>>>Dept. of Cell & Molec. Biol., Uppsala
                        University.
>>>>>Box 596, 75124 Uppsala, Sweden. Phone:  +46184714205.
>>>>>spoel at xray.bmc.uu.se    http://folding.bmc.uu.se
>>>>>-- 
>>>>>gmx-developers mailing list
>>>>>gmx-developers at gromacs.org
>>>>>http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>>>Please don't post (un)subscribe requests to the
                        list. Use the www interface or send it to gmx-developers-request at gromacs.org.
>>>>>
>>>>
>>>>
>>>>-- 
>>>>David Mobley
>>>>
>>
>>-- 
>>Sent from my mobile device. Please pardon any unusual brevity or
      typos. 
>>
>>
>>
>
>--
>gmx-developers mailing list
>gmx-developers at gromacs.org
>http://lists.gromacs.org/mailman/listinfo/gmx-developers
>Please don't post (un)subscribe requests to the list. Use the
>www interface or send it to gmx-developers-request at gromacs.org.
>

-- 
David Mobley
dmobley at gmail.com
949-385-2436

-- 
gmx-developers mailing list
gmx-developers at gromacs.org
http://lists.gromacs.org/mailman/listinfo/gmx-developers
Please don't post (un)subscribe requests to the list. Use the 
www interface or send it to gmx-developers-request at gromacs.org. 