From: Arturas Ziemys (arturas.ziemys_at_uth.tmc.edu)
Date: Fri Jul 25 2008 - 09:26:53 CDT
Thanks, Axel, for nice answer. To some extend I understand that all
"tricks" in MD have side effects.
Getting back to your answer... Don't getting deep into PME guts, I
understand that covalent linkage over cell boundaries does not break
PME. However, I should be aware of system dimensions, especially in case
of pseudo 2d-systems, like membranes that you mentioned in your answer.
Then how are treated exclusions using PME: through the list of atom in
direct space (nearest neighborhood)? If the question seem silly -
disregard, as maybe I do not understand properly the implementations and
theory. I tried to go over PME through literature, but I failed get
clear understanding about that.
Axel Kohlmeyer wrote:
> On Thu, 24 Jul 2008, Arturas Ziemys wrote:
> AZ> Hi,
> AZ> I have general question. I created "infinite" polymer using binds
> AZ> across boundaries. I that kind of advice be Peter here in mailing
> AZ> list for nanotube, or Cruz-Chu paper of silica did that using NAMD.
> AZ> Q. How well PME is compatible with such a system ?
> that depends a lot on what your polymer depends of, if and how
> your are solvating it, what kind of properties you want to
> compute and what level accuracy you expect.
> when using PME or any equivalent periodic solver for electrostatics
> you are imposing an "infinite crystal" setup onto your system.
> this is found to be the best compromise between accuracy and
> system size for computing long range electrostatic interactions
> in bulk liquid water. now if you put something into your water
> box, that will be interacting with its periodic images in a
> 3d crystal-like fashion as well, which infers some artefacts,
> if your system has a significant dipole or higher residual
> multipole moments. to use such a 3d-ewald sum on a 2d-system
> people frequently enlarge the non-periodic dimension in the
> hope to minimize the coupling between the periodic images,
> similar for 1d-systems. without going too much into technical
> details (i wrote half a thesis about this and it can get pretty
> nasty) one has to understand that this does not help much if
> you have a resident multipole (particularly a dipole) along
> the "non-periodic" dimension, e.g. in a lipid bilayer. you
> will still have artificial changes (sometimes significant)
> of the electrostatic potential along that directions and
> even with a "proper" 2d-ewald summation, you may see some
> artefacts when using spherical cutoffs.
> that being said, if you add up the errors of using a force
> field, statistics and other simulation "tricks" (e.g.
> multi-timestepping) and you are not after sensitive properties
> or trying to compute ultra-accurate numbers, it may work
> just fine.
> to give a comparative example, in typical peptide simulations
> these days, the amount of solvating water between the periodic
> images of the peptides is predominantly chose quite a bit
> on the small side for the sake of computational efficiency
> and i sometimes wonder if people are aware how much their
> trajectories are affected by this. yet a lot of useful things
> can be still be learned from those runs and i suspect that
> generally the statistical errors outweigh the systematic ones,
> with the exception of a few pathological cases.
> so it all boils down to understanding the side effects
> of using PME and checking carefully how much your system
> would be affected.
> hope that helps,
> AZ> Arturas
-- Arturas Ziemys, PhD School of Health Information Sciences University of Texas Health Science Center at Houston 7000 Fannin, Suit 880 Houston, TX 77030 Phone: (713) 500-3975 Fax: (713) 500-3929
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:49:40 CST