[gmx-developers] PME-Switch vs. PME

[Shirts, Michael (mrs5pt)]

Echoing Berk's comments, yes, PME-switch is only intended for very short
switches (0.05 or shorter).  There is no way to propagate the effect of the
switch into the reciprocal space (without large amounts of extra
computation).

Berk, yes, we should synchronize best practices for 4.6.  It's still not
100% clear to me whether these exact cutoffs properly interact with free
energy code.  For example, it's not clear to me how the dispersion term is
calculated here, which is important to get right for free energy
calculations. I don't think any (non-buggy) functionality should be changed
for 4.6.2, but we can definitely figure out better defaults and warnings for
4.6.2 to make things clearer and errors like this are clearly warned
against.  You shouldn't be able to have a PME-switch bigger than 0.05
without seeing a lot of warnings.

Best,
~~~~~~~~~~~~
Michael Shirts
Assistant Professor
Department of Chemical Engineering
University of Virginia
michael.shirts at virginia.edu
(434)-243-1821


> From: Berk Hess <hess at kth.se>
> Reply-To: Discussion list for GROMACS development <gmx-developers at gromacs.org>
> Date: Wed, 6 Mar 2013 19:29:19 +0100
> To: Discussion list for GROMACS development <gmx-developers at gromacs.org>
> Subject: Re: [gmx-developers] PME-Switch vs. PME
> 
> Hi,
> 
> You don't mention your, critical, cut-off parameters.
> I guess you used a long switching range. The switch is only intended to
> avoid
> cut-off artifacts (and therefore does not act on the reciprocal).
> 
> Any switching function should only be applied over a very short range.
> As with PME ewald_rtol is usually 10^-5, you don't really need the switch,
> and if you do want to use it, you should switch over 0.05 nm or less.
> 
> But in 4.6 you can use exact cut-off's, either with the Verlet cut-off
> scheme,
> or at a high cost also with the group cut-off scheme. Thus PME-switch is
> no longer needed
> and we should deprecate it in 4.6.
> 
> Cheers,
> 
> Berk
> 
> On 03/06/2013 07:13 PM, Michel Cuendet wrote:
>> 
>> Hi,
>> 
>> I tried to compare PME and PME-switch for a small peptide in solution
>> with Amber96. I was surprised to see that when introducing the Switch
>> function, only the real-space energy is affected, whereas the
>> reciprocal-space part is not. The difference in total energy is huge,
>> about 4919 kJ/mol for my system. Below are the numbers. Shouldn't the
>> reciprocal energy be "switched" as well, in the spirit of what the
>> error function does in Eq. (4.160) and (4.161) of the 4.6.1 manual?
>> 
>> Unfortunately, the energy is not just shifted by a constant. Comparing
>> two different conformations with PME and PME-Switch gives me a
>> DeltaDeltaE of the order of 58 kJ/mol. I also compared the energies
>> with Amber12, and found that PME reproduced the Amber electrostatics
>> within 21 kJ/mol, with a DeltaDeltaE below 2 kJ/mol comparing the two
>> conformations in gromacs and Amber12.
>> 
>> I ran extensive simulations of this peptide with gromacs 4.5.5 and
>> PME-Switch (using various enhanced sampling methods). The peptide
>> would never stay folded, whereas it has been reported to have a
>> significant folded population in simulations with the same force field
>> and cutoffs done with the Amber8 package. Do you think this could be
>> due to the altered electrostatics? I can't see any other explanation
>> at the moment.
>> 
>> 
>> Gromacs 4.5.5 PME :
>> Coulomb-142898.323974
>> Coulomb (SR)-221730.3613
>> Coul. recip.-24396.81872
>> Potential-204582.3431
>> 
>> Gromacs 4.5.5 PME-Switch :
>> Coulomb-142898.323974
>> Coulomb (SR)-226649.5443difference = -4919.1830
>> Coul. recip.-24396.81872
>> potential-209501.5261
>> 
>> Gromacs 4.6.1 PME, coulomb-modifier=none
>> Coulomb-142898.32373
>> Coulomb (SR)-221730.4063
>> Coul. recip.-24396.46875
>> potential-204592.1719
>> 
>> Gromacs 4.6.1 PME-switch, coulomb-modifier=none
>> Coulomb-142898.32373
>> Coulomb (SR)-226649.625difference = -4919.2188
>> Coul. recip.-24396.46875
>> potential-209511.3906
>> 
>> 
>> I can send files and more elaborate data if required.
>> 
>> Thanks and kind regards,
>> Michel
>> 
>> 
>> ==========================================================
>> Michel Cuendet, PhD
>> Molecular Modeling Group
>> Swiss Institute of Bioinformatics
>> CH-1015 Lausanne
>> Switzerland
>> http://lausanne.isb-sib.ch/~mcuendet/
>> <http://lausanne.isb-sib.ch/%7Emcuendet/>
>> ==========================================================
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
> 
> -- 
> 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.