Re: Re:Re: Re: Re: Re: Dielectric Constant

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Sat May 01 2010 - 13:22:10 CDT

On Sat, 2010-05-01 at 21:03 +0800, ydhuang2727 wrote:

> Dear Axel,
> Execuse me for my late reply!
> Thanks a lot for you suggestion.
>
> >ok. the difficulty is that i don't know your code and i know from personal
> >experience, how easy it is to miss a factor of two or similar somewhere.
> If you are interested in my code, i would glad to show you the part of
> calculating the total electrostatic energy.

no, i am not interested. the molecular simulation world has already
too many MD codes. hell, i wrote one myself last fall, even if it was
only for teaching purposes. yet, in my opinion, more often than not,
the time is better spent on improving and extending existing MD
packages rather than writing new ones.

> >i am not so sure about "same structure"
> I'm sorry for my poor description in english. More detailed, it is to

it is not the english. i have seen plenty worse.

it is a matter of taking into account that when
using e-mail, you have to explain and describe
details that may otherwise be obvious, and that you
have to give a complete description, even if it may
feel redundant. the people you are communicating
with may not have the same opinion of how things are
to be done and thus would make different choices for
what you omit to describe.

> say i download a X-ray or NMR protein 3D structure and calculate the
> electrostatic energy of the structure respectively with my code and
> NAMD.

yes, i understood that. you also mentioned that you
would minimize the structures, but didn't say whether
you compare the energies before or after the minimization.

i also think, like i mentioned before, that you should
first compare an artificial system with only a few charges
and where you can compute the energy manually. only if
that would be consistent, i would move on to larger systems.

> >i am not so sure about "same structure". you mention minimization, you
> >also mentioned polarizability at some point. are you _really_ comparing
> >apples to apples here?
>
> Yes, I like apples! The reason i mentioned polarizabiltiy is that I
> just tried to explain, in my opinion, that whether different
> dielectric constant be used.

no. it must not be changed. the dielectric constant is a continuum
property _generated_ by the atoms and molecules. it represents how
a medium gets polarized by its environment. so unless you remove
some atoms from that environment, e.g through coarse graining or
using an implicit solvent, the dielectric constant has to be kept
at 1.0.
  
> Hopefully, i feed you curiosity.

well, if you ask about my curiosity, i am most curious about why
you want to write another code and what would be different from
other packages that are already around.

> Conversely, Will you please briefly explain how
> namd get the electrostatic energy of a system, so that i can see
> what's wrong with my code. Are you okay?

this is explained in great detail in the NAMD paper, and the
references in it. that is a much better source than me ramblings.

axel.
> Appreciate your time very much!!
>
>
> All the best,
>
> Yandong,
>
>
>
>
>
>
>
>
>
>
> 在2010-04-30 10:34:42,"Axel Kohlmeyer" <akohlmey_at_gmail.com> 写道:
> >On Thu, Apr 29, 2010 at 9:59 PM, ydhuang2727 <ydhuang2727_at_163.com> wrote:
> >> Dear Axel,
> >> Well, let me put others aside, and what i'm comparing is the
> >> electrostatic energy obtain by my code and by namd with the same force
> >> field.
> >
> >ok. the difficulty is that i don't know your code and i know from personal
> >experience, how easy it is to miss a factor of two or similar somewhere.
> >
> >> And my problem is why they are so different for the same structure.
> >
> >i am not so sure about "same structure". you mention minimization, you
> >also mentioned polarizability at some point. are you _really_ comparing
> >apples to apples here? i would only compare energies at the zeroth
> >step when you feed NAMD and your code the _exact_ same coordinates,
> >the same box size etc. then and only then each of the components should
> >be the same. minimization in a highly dimensional coordinate system,
> >can very easily lead to different minima.
> >
> >just setting the dielectric constant to a different value
> >
> >> My Simulation goes well and reach equilibrium, energy constant, in
> >> NVE ensemble. But i still feel worried about such a clear and unreasonable
> >> difference when dielectric constant is 1.0.
> >
> >well, have you tried some much simpler systems?
> >like two water molecules without periodic boundaries?
> >where you can compute and compare energies and forces
> >even with a pencil and a piece of paper?
> >
> >a simulation that runs and conserves energy proves that
> >your integrator is correct and that your time step, you cutoff
> >and other parameters are reasonable, but it does not at all
> >prove that your interactions are computed correctly. i have
> >had this happen to me just about a year ago.
> >
> >> In other words, I think the so-called "set velocity zero" is somewhat
> >> like damped MD as you mentioned. And my approachment is simply and crudely
> >> make the velocity damped directly down to zero but with a relax time.
> >
> >ok. the "relax time" part is what was missing and what was tickling
> >my curiosity. thanks a lot for providing this information. it may occasionally
> >come over as a bit rough, but one of the major forms of "payback"
> >that one can get from answering questions on mailing lists is that
> >people satisfy somebody's curiosity. so i always feel like being
> >cheated out of my share, if other people get their questions answered,
> >come back with more questions, but don't answer mine.
> >
> >cheers,
> > axel.
> >
> >
> >> Thanks for your time!
> >>
> >> Best wishes,
> >>
> >> Yandong,
> >>
> >> 在2010-04-29 21:34:36,"Axel Kohlmeyer" <akohlmey_at_gmail.com> 写道:
> >>>On Thu, Apr 29, 2010 at 8:39 AM, ydhuang2727 <ydhuang2727_at_163.com> wrote:
> >>>> Dear Jerome,
> >>>>
> >>>> Thanks for your kind analysis.
> >>>> The dielectric constant of a protein is about 2 to 4, thus when the protein
> >>>> is polarizable, a proper dielectric constant bigger than 1.0 should be
> >>>> adopted to reduce the intensity bwteen two charged atoms. Should we?
> >>>
> >>>no. why?? the dielectric constant is a _result_ of the charges in your system
> >>>reacting to its environment. if you use explicit charges, your
> >>>dielectric constant
> >>>should _always_ be 1.0. setting it to a larger value is a (crude) approximation
> >>>to an implicit solvent. in this case you (wrongly) would assume a continuum
> >>>solvent. but inside a protein you are making a large error, since you would
> >>>not have any solvent there.
> >>>
> >>>> If this is the case, i can abtain the same values as what namd got.
> >>>
> >>>perhaps you should _first_ explain to us what you are actually comparing.
> >>>it looks to me as if you are doing an apples to oranges comparison and
> >>>trying to fix up stuff with an empirical scaling factor is not going
> >>>to help much.
> >>>
> >>>also, you still owe us an explanation for how your atoms can move when
> >>>you set the velocities to zero.
> >>>
> >>>thanks,
> >>> axel.
> >>>
> >>>
> >>>> Appreciate your attention!
> >>>>
> >>>>
> >>>> Good day!
> >>>>
> >>>> Yandong,
> >>>>
> >>>>
> >>>> 在2010-04-29 15:59:42,"Jérôme Hénin" <jhenin_at_ifr88.cnrs-mrs.fr> 写道:
> >>>>>Dear Yangdong,
> >>>>>
> >>>>>> Based on the CHARMM22 force field, i use my own code to roughly minimize
> >>>>>> a peptide and compare each term of interactions with the results of
> >>>>>> minimization of namd2.6. My approachment is simply to set velocity zero
> >>>>>> everay step, which is inferior to annealing algorithm that namd may adpot.
> >>>>>>
> >>>>>> All interactions are comparable and consistent with the results of namd
> >>>>>> except for electrostatic interaction when dielectric constant is 1.0, the
> >>>>>> absolute value of which by my code is nearly twice as big as that by namd.
> >>>>>> As to VDW and Electrostatic interactions, all pairs of atoms without bond
> >>>>>> interaction( bond stress, angle bend, dihedral torsion, improper dihedral)
> >>>>>> are considered in my code. I did not use cut-off method. Can you find
> >>>>>> something wrong?
> >>>>>
> >>>>>Maybe. In the CHARMM ff, the rule is to include nonbonded terms for
> >>>>>all atoms except those linked through one or two bonds (which will
> >>>>>typically be coupled by bond and angle terms). Atoms connected by a
> >>>>>torsion term (3 bonds) have their full nonbonded interactions (the
> >>>>>so-called 1,4-interactions). There are a few exceptions: for some atom
> >>>>>types, the CHARMM parameter files list modified Lennard-Jones
> >>>>>parameters for 1,4 interactions. Electrostatics is never modified,
> >>>>>though.
> >>>>>
> >>>>>Cheers,
> >>>>>Jerome
> >>>>
> >>>>
> >>>> ________________________________
> >>>> 网易为中小企业免费提供企业邮箱(自主域名)
> >>>
> >>>
> >>>
> >>>--
> >>>Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> >>>http://sites.google.com/site/akohlmey/
> >>>
> >>>Institute for Computational Molecular Science
> >>>Temple University, Philadelphia PA, USA.
> >>>
> >>
> >>
> >> ________________________________
> >> 网易为中小企业免费提供企业邮箱(自主域名)
> >
> >
> >
> >--
> >Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> >http://sites.google.com/site/akohlmey/
> >
> >Institute for Computational Molecular Science
> >Temple University, Philadelphia PA, USA.
> >
>
>
>
> ______________________________________________________________________
>
> 网易为中小企业免费提供企业邮箱(自主域名)

-- 
Dr. Axel Kohlmeyer  akohlmey_at_gmail.com
http://sites.google.com/site/akohlmey/
Institute for Computational Molecular Science
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:55:44 CST