Date: Thu, 1 May 2003 17:30:48 -0400 (EDT)
From: darden
Subject: Re: Ewald: coeff, error estimate

Dear Paula
If I understand your question (i.e. is ee your units conversion
coefficient??)kappa would be equivalent to the ewald_coefficient used in
regular ewald or pme in amber. So that for example at the cutoff distance
the argument of erfc is approximately 3: for a 9 angstrom cutoff I
typically use a coefficient of ~1/3

The ewald error estimate which I cooked up is probably a bit oversold.
Maybe I should have stuck to an analytic variant. Besides it only works
for charges (if you use Ewald with induced dipoles the error estimate is
garbage). It should not be thought of in kcals, but rather as a relative
error. Literally it is the approximate virial plus approx energy over the
approx energy. Since the numerator should be zero and in practice the
pme energy is obtained more accurately than the virial--this actually
estimates the relative error in the virial which in turn is an ok
estimator for the relative force error (compared to converged ewald)

hope this helps
tom d
On Thu, 1 May 2003, Paula Petrone wrote:

>
> Hi,
>
> I am calculating Ewald energies and would like to compare with AMBER's
> result.
>
> The Ewald coefficient kappa that appears in Ewald real term:
> Ewald real= ee*qq/2*sum[erfc(kappa*rij)/r]
> Which is the relation of kappa to the coefficient that appears in the
> OUTPUTFILE? Is the Ew. coefficient=kappa/box?
>
> The error estimate in the output file is the ewald error estimate in
> kcals, per atom. Correct?
>
> Thanks a lot!
>
> Paula.
>