[gmx-users] changing ewald_rtol
mark.j.abraham at gmail.com
Tue Nov 18 20:03:15 CET 2014
On Tue, Nov 18, 2014 at 7:38 PM, Johnny Lu <johnny.lu128 at gmail.com> wrote:
> from online manual:
> "*ewald-rtol (1e-5)* The relative strength of the Ewald-shifted direct
> potential at *rcoulomb* is given by *ewald-rtol*. Decreasing this will give
> a more accurate direct sum, but then you need more wave vectors for the
> reciprocal sum."
> Is it ok to use ewald-rtol=1e-9 for Amber99SB-ILDN forcefield?
That's not a question that has a useful single answer.
1) It wasn't parameterized with any Ewald method, so any subsquent use of
that method is valid only upon empirical evidence.
2) Different observables in different systems have different dependencies
on the quality of the electrostatics model, and this one parameter is only
part of any story.
3) A tiny magnitude of forces at the cutoff matters little if they're being
added to other forces at shorter distances - and if there's more than about
5 orders of magnitude difference, then the small ones get lost in the
accumulation in floating-point representation
4) Making one part of the electrostatics approximation super accurate while
ignoring the other doesn't do any good when you add them together, because
the largest error dominates.
5) Some other error might be more dominant still (e.g. incomplete sampling,
interior residues have the same charges as exterior residues, no
Will the sum in reciprocal space become very inaccurate?
Yes, like the docs say, because ewald-rtol and rcoulomb are used to compute
the Ewald splitting parameter beta, which is then used in reciprocal space
Or this will just make the simulation run slower?
By itself, it's just multiplication by a different scaling factor. Fixing
the issues in reciprocal space (increasing number of grid points and/or
spline interpolation order) will be expensive.
> Thanks again.
> Gromacs Users mailing list
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> * 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-users or
> send a mail to gmx-users-request at gromacs.org.
More information about the gromacs.org_gmx-users