[gmx-users] Re: How does fourierspacing correspond to fourier_nx, fourier_ny, and fourier_nz?

[Andrew DeYoung]

Hi Mark and Florian,

Thank you kindly!  

My colleagues and I are actually using going to use the plain Ewald method
in Gromacs.  We have a slab geometry and two capacitor plates.  In all cases
we use the Berkowitz dipole correction (ewald_geometry = 3dc) to obtain
pseudo 2-D summation.  We have tried both PME and plain Ewald methods.
Strangely, when we use PME, we observe a large drift in the electric
potential (which we calculate using g_potential) between the capacitor
plates.  When, however, we use the plain Ewald method, we do not observe the
drift in the electric potential between the plates; the potential is fairly
flat.

We have tried for several weeks to play with the PME parameters, but no
matter what PME parameters we use, the drift in the electric potential
persists.  When we use the plain Ewald method, however, the electric
potential is always flat, consistant with the results from another MD
package.  

Do you have any experience with significant differences between PME and
Ewald results?  It certainly seems from reading the mailing list archives
that there is really never any disadvantage to using PME.  In fact, I read
that Dr. Lindahl said that this is why the plain Ewald has never been
implemented in parallel; it just would not be worth the development effort.
PME is so much faster, that there is really never any reason to use anything
else.  So my colleagues are just stumped by how the PME and Ewald results
could be so much different for us.

Thank You!

Andrew DeYoung
Carnegie Mellon University

> The former for PME. The latter for plain Ewald (which you should
> probably not want to use). 
> 
> Mark