Re: Coulomb-Energies and MMM2D

[Rudolf Weeber]

Hi, 

On Mon, Dec 16, 2019 at 01:40:33PM +0100, Fabian Glatzel wrote:
> > I'm still heaving some issues with the Coulomb energy using ELC with the
> > pot_diff parameter as well as using MMM2D with pot_diff.
> > The used system is a 2d plate capacitor (periodic boundary conditions
> > along x&y) with charged hard spheres (Weeks-Chandler-Andersen) and with
> > an external potential of 1V.
> > I simulated this system using three different setups:
> > 1) ELC and the pot_diff argument
> > 2) MMM2D and the pot_diff argument
> > 3) ELC and a linear electric potential (corresponding to a constant
> > surface charge density at the electrodes and hence neglecting the effect
> > of image charges)
To put in my two cents:

The advanced 2D electrostatics features such as pot_diff are currently not 
actively used by members of the core team.
Therefore, we can only help to a limited extent, here.

The MMM2D method was kept as "reference" method. ELC should be able to do 
anything that MMM2D does, at much better performance, since MMM2D only scales 
at N^(5/3) with a high prefactor.
While MMM2D will remain available in the 4.1 series, it is likely that future 
versions of Espresso will not ship it anymore. Rather, we would focus our 
efforts on the ELC method.

Do I understand correlty, that your profiles obtained with ELC + a potential 
difference seem reasonable.
Based on Konrad's explanations, I assume, they won't match those for ELC and a 
constant external electric field.

In case you can provide a test which compares the forces for a certain ion 
configuration against an analytical solution or a numeric results that can be 
re-computed in a reasonable time by a simpler method (e.g., direct evaluation 
of the sums, eventhough they converge slowly), that would be much appreciated.

ELC with potential difference and non-neutral systems are known to have issues 
and the functionallity is currently deactivated.

The current tests for the ELC+ pot_diff combination are in 
testsuite/python/elc.py

Regarsd, Rudolf