RE: [AMBER] skinnb and energy conservation

From: Scott Le Grand <SLeGrand.nvidia.com>
Date: Wed, 12 May 2010 17:27:47 -0700

If you don't update the nonbond list, ever, then you will only ever see atoms that were in range at step 0. I'm not sure exactly how this proves anything except that you really ought to update your nonbond list!

-----Original Message-----
From: amber-bounces.ambermd.org [mailto:amber-bounces.ambermd.org] On Behalf Of Jason Swails
Sent: Wednesday, May 12, 2010 13:24
To: AMBER Mailing List
Subject: Re: [AMBER] skinnb and energy conservation

On Wed, May 12, 2010 at 4:11 PM, liu junjun <ljjlp03.gmail.com> wrote:

> Thanks a lot to all of you. Now I understand more. As Bob and Bill pointed
> out, the ele is calculated for those outside the cut when PME is enabled, I
> agree with what Mike mentioned that vdw is quite small at the cut. However,
> in case of vacuum simulation, is there any different algorithms to deal
> with
> the electrostatics that moves out of the cutoff? If there isn't, there
> seems
> to be a convergence problem in vacuum simulation when cut isn't large.
>

I think cut is generally recommended to be infinite in this circumstance (or
large enough to encompass everybody), so the pairlist is complete.


>
> In order to understand this, I made a simple test simulation in vacuum, but
> don't see any problem. I don't know whether the tested simulation indicates
> all nonbond interactions were within the cutoff during the simulated time
> or
> not, so I made another test by fixing the nonbond list.
> 1). set skinnb=0 in the input
> 2). set do_update_list = .false. at the end of check_skin subroutine, so
> that sander stops rebuilding the nonbond list during the whole simulation.
>
> What I found surprises me. The vacuum simulation ran 100 MD steps without
> updating the nblist. The potential energy was going all the way to more
> negative values and the temperature was going all the way to higher values.
> The temperature was 600 K at the 100 step.
>
> Can anybody please explain this?
>
> Thanks!
>
> Junjun
>
> On Wed, May 12, 2010 at 1:11 PM, Robert Duke <rduke.email.unc.edu> wrote:
>
> > Ah, my prejudices emerge. And as pointed out by Bill and Dave and Mike,
> > pme does not help with vdw. But the vdw long-range energy term, the
> > dispersion term, is a function of distance of the form 1 / r**6 - that
> drops
> > pretty quickly. So at about 8 angstrom, you are in pretty good shape on
> vdw
> > truncation. Some folks will go out to 9 angstrom, and there is a slight
> > difference; longer cutoffs do have a performance cost though. Darden
> also
> > put an analytic correction to the vdw long range calcs in the pme code,
> > which is routinely used (see vdwmeth). This effects energies and
> pressures
> > but not individual atom forces; the correction for atom forces would be
> > pretty darn small... All this can be played with fairly easily to gain
> some
> > sense of the real impact. For PMEMD, you can set the vdw cutoff to
> really
> > long lengths if you want without increasing the electrostatic cutoff -
> you
> > use vdw_cutoff and es_cutoff in place of cut in the &cntrl namelist, and
> it
> > must be true that vdw_cutoff >= es_cutoff. The "skin" is really only a
> > mechanism to reduce the frequency of pairlist building, since it is an
> > expensive process; as I think Mike said, using the skin just gives you
> the
> > logical equivalent of rebuilding the list at each step (there are various
> > performance tradeoffs not mentioned...).
> > Regards - Bob Duke
> > ----- Original Message ----- From: "case" <case.biomaps.rutgers.edu>
> >
> > To: "AMBER Mailing List" <amber.ambermd.org>
> > Sent: Wednesday, May 12, 2010 12:57 PM
> > Subject: Re: [AMBER] skinnb and energy conservation
> >
> >
> >
> > On Wed, May 12, 2010, liu junjun wrote:
> >>
> >>>
> >>> I have a question about how SKINNB works in energy conservation. The
> >>> nonbond
> >>> list is extended but vdw and ele interactions truncated at cut. As I
> >>> understand, the atom pairs in the nblist can be considered as some kind
> >>> of
> >>> "candicates" for calculating vdw and ele interactions, and only those
> >>> with
> >>> distances less than cutoff are selected to calculate the vdw and ele
> >>> interactions. If my understanding is correct, then there seems to be a
> >>> problem in energy conservation. Lets assume there is an atom pair whose
> >>> initial distance is 9.99A in a system where cut=10A. The nonbond
> >>> interactions are calculated for this atom pair because 9.99A < 10A.
> After
> >>> one MD step, the distance might changes to 10.01A, then the nonbond
> >>> interactions are not calculated because 10.01 > 10A, which would cause
> a
> >>> sudden energy change.
> >>>
> >>
> >> You are correct: the use of skinnb does not remove problems arising from
> >> discontinuities in the Lennard-Jones terms at the cutoff. Its purpose
> is
> >> just
> >> to reduce the frequency at which the non-bonded pair list needs to be
> >> constructed.
> >>
> >> ....dac
> >>
> >>
> >> _______________________________________________
> >> AMBER mailing list
> >> AMBER.ambermd.org
> >> http://lists.ambermd.org/mailman/listinfo/amber
> >>
> >>
> >>
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>



-- 
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information.  Any unauthorized review, use, disclosure or distribution
is prohibited.  If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 12 2010 - 17:30:03 PDT
Custom Search