[gmx-users] dipole correction in 3.1-beta

[Erik Lindahl]

Hi,


>
>   else if(epsilon_surface > 0)
>     dipole_coeff=2*M_PI*ONE_4PI_EPS0 / (2*epsilon_surface + 1) / vol;
>   else if(epsilon_surface < 0)
>     dipole_coeff=2*M_PI*ONE_4PI_EPS0 / vol;
>
> this way I can do a simulation in slab geometry if I choose
> epsilon_surface < 0... say -1.0.  I think there may be an extra 2 in the
> numerator rather than only  M_PI*ONE_4PI_EPS0 that you mention.  It seems
> that all the literature I find gives the formulae in Gaussian.. so to
> convert, I guess we have to multiply by ONE_4PI_EPS0.  Check on this..
>

Not necessary - we've already arrived at the same formula; the reason I didn't
have the 2 in the numerator was that we I divided by 1.5 instead of 3.
(2*epsilon+1 when epsilon==1).


>
> See:
> J. Chem. Phys., Vol. 109, No. 18, 8 November 1998
> and
> J. Chem. Phys., Vol. 111, No. 7, 15 August 1999
>
> The second one has the slab geometry correction in it by Berkowitz and Yeh.
> You may want to include this in the 3.1 release in case somebody wants to
> use it, but thats up to I guess you and others to decide.

No problem, the only reason we haven't included more references is laziness -
it's better the more we have since you avoid the hassle of looking it up in a
previous paper... I'll add it for 3.1.

>
>
> Please, If you can take a look at this and tell me what you think.  Is
> there anything I can change?
>

No, everything looks really good, apart from the slab thing that I haven't used
myself. I could of course add it, but since neither David or I work a lot with
dipole corrections it might be better if you tried it on a test system (where you
know what to expect :-), and just send me a diff (or the entire file) to avoid
more typos, and finally get the corrections right?

Actually, if this means slab geometries are working (is there any change needed
in the main pme routines?) it's probably better to add a real ewald_geometry
option that is either 3d or 2d, or maybe you have a better name for it.... The
sign thing works, but it's not that intuitive :-)

Cheers,

Erik