[gmx-developers] high-precision TIP3P parameters?
Michael R Shirts
Michael.Shirts at Colorado.EDU
Sat May 27 15:24:15 CEST 2017
➢ Another issue is that, when last I checked, all programs (AMBER, CHARMM, NAMD,
➢ GROMACS) use a different value of 1/(4*pi*eps0) for calculating electrostatics
➢ so there is always some error there, irrespective of the actual topology.
➢ Again, these are quantities that will not substantially change dynamics (if at
➢ all), but it seems as though we're chasing minutiae when there are other things
➢ that aren't ever going to be standard.
This, I’ve analyzed (though you may be referring to this study already):
https://link.springer.com/article/10.1007/s10822-016-9977-1
AMBER and CHARMM were the only codes examined where the effect of 1/(4*pi*eps0) was outside of rounding error. I think they were discussing changing it, but not sure the timing on that. GROMACS (though only after a recent change), LAMMPS and DESMOND used values that were close enough to experiment to be essentially indistinguishable after rounding error.
This study more or less says how close you can get after controlling for everything expect rounding error in the program itself (not rounding in the parameter files) which is about one part in 10^-5-10^-6. Note it used direct conversion from the AMBER parameters, not “native” TIP3P.
1) There’s a good argument that all the programs should probably bump up agreement in the native assigned parameters by 1 or 2 decimal places more. Beyond that is unlikely make anything more comparable.
2) Nonbonded are hardest to get right because of slightly differing LJ and Coulomb treatments. For periodic boxes of water with PME, relative errors of 10^-4 to 10^-5 might be the limit.
3) The importance of making this change is therefore pretty low.
Best,
~~~~~~~~~~~~~~~~
Michael Shirts
Associate Professor
michael.shirts at colorado.edu
http://www.colorado.edu/lab/shirtsgroup/
Phone: (303) 735-7860
Office: JSCBB C123
Department of Chemical and Biological Engineering
University of Colorado Boulder
On 5/27/17, 4:56 AM, "gromacs.org_gmx-developers-bounces at maillist.sys.kth.se on behalf of Justin Lemkul" <gromacs.org_gmx-developers-bounces at maillist.sys.kth.se on behalf of jalemkul at vt.edu> wrote:
On 5/27/17 4:47 AM, Mark Abraham wrote:
>
>
> On Sat, May 27, 2017 at 10:17 AM Igor Leontyev <ileontyev at ucdavis.edu
> <mailto:ileontyev at ucdavis.edu>> wrote:
>
> Hi,
> It is unlikely that inaccuracy in 4th digit of the parameters could
> introduce significant error in simulation results.
>
> However, for the energy comparison and debugging which is often made by
> curious people (including myself) when switching between Amber and
> Gromacs codes the inaccuracy is really bothering. Accepting suggested
> correction will match Amber and Gromacs code energies and make the
> comparison perfect (possibly for the cost of one time issue with
> regression test).
>
>
> True, but also makes GROMACS different from past versions of itself, which also
> would worry people doing such a comparison when e.g. contemplating upgrading
> their version. We're all for doing things correctly, but that should mean
> everybody actually implementing the published/accepted value, not agreeing with
> each other on how to be different.
>
> It sounds like we're happy enough to change the values in GROMACS to match the
> paper - can people please raise the issue/question with others who distribute
> parameter sets?
>
From the CHARMM perspective, we just have a standard format that we adhere to
in terms of the precision of values. So epsilon is -0.1521 instead of the more
precise -0.15207, but I don't see that changing on our end. That introduces the
possibility that GROMACS and AMBER match exactly but GROMACS and CHARMM do not,
but that already suggests CHARMM and AMBER do not match, a comparison I have
never done (though GROMACS and CHARMM do match quite nicely).
Another issue is that, when last I checked, all programs (AMBER, CHARMM, NAMD,
GROMACS) use a different value of 1/(4*pi*eps0) for calculating electrostatics
so there is always some error there, irrespective of the actual topology.
Again, these are quantities that will not substantially change dynamics (if at
all), but it seems as though we're chasing minutiae when there are other things
that aren't ever going to be standard.
-Justin
> Mark
>
> Igor
>
> >
> > Hi,
> >
> > These difference are irrelevant for any practical purposes, since the
> > model (and sampling) is not by far that accurate. But it would still be
> > nice to have all decimals that we have in files correct.
> > Is this somehow related to the differences in the kcal/mol to kJ/mol
> > conversion in different packages and/or at different times?
> >
> > Cheers,
> >
> > Berk
> >
> > On 24/05/17 21:18 , Igor Leontyev wrote:
> >> Hi all. The issue was discussed in past but nothing was corrected
> >> since then.
> >>
> >>
> https://mailman-1.sys.kth.se/pipermail/gromacs.org_gmx-developers/2011-April/005178.html
> >>
> >>
> >> To make gromacs TIP3P energies matching Amber code energies:
> >>> Suggested correction in amberXX.ff/ffnonbonded.itp:
> >>> ;OW 8 16.00 0.0000 A 3.15061e-01 6.36386e-01
> >>> OW 8 16.00 0.0000 A 3.15076e-01 6.35968e-01
> >>
> >> Igor
> >>
> >>> So, going back to the original TIP3P paper, the parameters are listed
> >> as A and C parameters (in the formula a/r^12 ? b/r^6).
> >>>
> >>> The parameters listed there are A = 582000 A^12 kcal/mol and C = 595
> >>> A^6 kcal/mol.
> >>>
> >>> Converting to sigma/epsilon form, this gives:
> >>> Sigma = 0.31506561105 nm
> >>> Epsilon = 0.6362717354 kJ/mol
> >>>
> >>> This compares with the values in the most of the distributed ITP
> >>> files of
> >>>
> >>> Sigma = 0.315061 nm (a difference of 4.6 x 10^-6 nm, or a relative
> >>> difference of of 1.4 x 10^-5)
> >>> Epsilon = 0.636386 kJ/mol (a difference of 1.1 x 10^4 kJ/mol, or a
> >>> relative difference of 1.8 x 10^-4)
> >>> (note the GROMACS one is also 0.152100 kcal/mol)
> >>> I don?t know how much of a difference this would make (likely a
> >>> difference of the scale of).
> >>>
> >>> Now, that?s in the original paper. What values are ACTUALLY
> >>> specified by other codes?
> >>>
> >>> For AMBER, (as produced by tleap), it outputs.
> >>>
> >>> A = 581935.564 A^12 kcal/mol
> >>> C = 594.825035 A^6 kcal/mol
> >>>
> >>> Which leads to:
> >>>
> >>> Sigma = 0.315075 (a difference of 3.0 x 10^-5 nm, or a relative
> >>> difference of 3.0 x 10^-5)
> >>> Epsilon = 0.635968 kJ/mol (a difference of 9.4 x 10^4 kJ/mol or a
> >>> relative difference of 4.7 x 10^-4)
> >>> (which is also 0.152000 kcal/mol)
> >>>
> >>> So ? I don?t know. Nobody seems to be getting it _exactly_ right,
> >>> and it?s also not so clear how much of a difference it makes. One
> >>> can estimate that enthalpies would have a relative error of around
> >>> 10^-4, and relative errors in the density of 3*dsigma = 10^-4 to 10^-5.
> >>>
> >>> So I don?t know that there is anything actionable to take. Just
> >>> thought it was interesting to bring out. Something to ponder for now.
> >>
>
> >
> > I?d say ?probably? irrelevant for practical purposes. It wouldn?t take
> too much simulation time to determine the density sensitively enough to tell
> the difference. But, pretty much irrelevant.
> >
> > ? Is this somehow related to the differences in the kcal/mol to kJ/mol
> > ? conversion in different packages and/or at different times?
> >
> > I think that it?s due to differences in rounding from the original A and
> C values vs. rounding from conversion of them to epsilon and sigma values.
> The epsilons in AMBER and GROMACS are clearly equal to using either 0.152100
> or 0.15200 kcal/mol as the value (the parameters in the original paper
> correspond to 0.151207).
> >
> > I think there might have been a kJ/kcal issue at some point, but those
> were fixed a decade or so ago (I seem to recall that popping up when I was a
> graduate student, but not since then).
> >
> > ? These difference are irrelevant for any practical purposes, since the
> > ? model (and sampling) is not by far that accurate. But it would still be
> > ? nice to have all decimals that we have in files correct.
> >
> > At this stage, it probably involves getting some general set everyone
> together and deciding in what the ?right? values are. I could potentially do
> that . . . but as Berk points out, not that big a deal. Unless one is
> trying to match energy snapshots, and getting frustrated.
> >
> > For what it?s worth, the Desmond viparr utility uses for TIP3P: {"sigma":
> 3.15061, "epsilon": 0.1521}. Both match GROMACS .itp libraries, but not
> tleap or the original paper.
> >
> > Best,
> > ~~~~~~~~~~~~~~~~
> > Michael Shirts
> > Associate Professor
> > michael.shirts at colorado.edu <mailto:michael.shirts at colorado.edu>
> > http://www.colorado.edu/lab/shirtsgroup/
> > Phone: (303) 735-7860
> > Office: JSCBB C123
> > Department of Chemical and Biological Engineering
> > University of Colorado Boulder
> >
> > On 5/25/17, 1:51 PM,
> "gromacs.org_gmx-developers-bounces at maillist.sys.kth.se
> <mailto:gromacs.org_gmx-developers-bounces at maillist.sys.kth.se> on behalf of
> Berk Hess" <gromacs.org_gmx-developers-bounces at maillist.sys.kth.se
> <mailto:gromacs.org_gmx-developers-bounces at maillist.sys.kth.se> on behalf of
> hess at kth.se <mailto:hess at kth.se>> wrote:
> >
> > Hi,
> >
> > These difference are irrelevant for any practical purposes, since the
> > model (and sampling) is not by far that accurate. But it would still be
> > nice to have all decimals that we have in files correct.
> > Is this somehow related to the differences in the kcal/mol to kJ/mol
> > conversion in different packages and/or at different times?
> >
> > Cheers,
> >
> > Berk
> >
> > On 24/05/17 21:18 , Igor Leontyev wrote:
> > > Hi all. The issue was discussed in past but nothing was corrected
> > > since then.
> > >
> > >
> https://mailman-1.sys.kth.se/pipermail/gromacs.org_gmx-developers/2011-April/005178.html
> > >
> > >
> > > To make gromacs TIP3P energies matching Amber code energies:
> > > > Suggested correction in amberXX.ff/ffnonbonded.itp:
> > > > ;OW 8 16.00 0.0000 A 3.15061e-01 6.36386e-01
> > > > OW 8 16.00 0.0000 A 3.15076e-01 6.35968e-01
> > >
> > > Igor
> > >
> > >> So, going back to the original TIP3P paper, the parameters are listed
> > > as A and C parameters (in the formula a/r^12 ? b/r^6).
> > >>
> > >> The parameters listed there are A = 582000 A^12 kcal/mol and C = 595
> > >> A^6 kcal/mol.
> > >>
> > >> Converting to sigma/epsilon form, this gives:
> > >> Sigma = 0.31506561105 nm
> > >> Epsilon = 0.6362717354 kJ/mol
> > >>
> > >> This compares with the values in the most of the distributed ITP
> > >> files of
> > >>
> > >> Sigma = 0.315061 nm (a difference of 4.6 x 10^-6 nm, or a relative
> > >> difference of of 1.4 x 10^-5)
> > >> Epsilon = 0.636386 kJ/mol (a difference of 1.1 x 10^4 kJ/mol, or a
> > >> relative difference of 1.8 x 10^-4)
> > >> (note the GROMACS one is also 0.152100
> kcal/mol)
> > >> I don?t know how much of a difference this would make (likely a
> > >> difference of the scale of).
> > >>
> > >> Now, that?s in the original paper. What values are ACTUALLY
> > >> specified by other codes?
> > >>
> > >> For AMBER, (as produced by tleap), it outputs.
> > >>
> > >> A = 581935.564 A^12 kcal/mol
> > >> C = 594.825035 A^6 kcal/mol
> > >>
> > >> Which leads to:
> > >>
> > >> Sigma = 0.315075 (a difference of 3.0 x 10^-5 nm, or a relative
> > >> difference of 3.0 x 10^-5)
> > >> Epsilon = 0.635968 kJ/mol (a difference of 9.4 x 10^4 kJ/mol or a
> > >> relative difference of 4.7 x 10^-4)
> > >> (which is also 0.152000 kcal/mol)
> > >>
> > >> So ? I don?t know. Nobody seems to be getting it _exactly_ right,
> > >> and it?s also not so clear how much of a difference it makes. One
> > >> can estimate that enthalpies would have a relative error of around
> > >> 10^-4, and relative errors in the density of 3*dsigma = 10^-4 to
> 10^-5.
> > >>
> > >> So I don?t know that there is anything actionable to take. Just
> > >> thought it was interesting to bring out. Something to ponder for now.
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or
> send a mail to gmx-developers-request at gromacs.org
> <mailto:gmx-developers-request at gromacs.org>.
>
>
>
--
==================================================
Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow
Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201
jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul
==================================================
--
Gromacs Developers mailing list
* Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
* For (un)subscribe requests visit
https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or send a mail to gmx-developers-request at gromacs.org.
More information about the gromacs.org_gmx-developers
mailing list