[gmx-developers] the "dipole" energy in a periodic system

[Berk Hess]

I have found references for the electrostatic energy
associated with the shape of the unit cell:
J. E. Roberts and J. Schnitker, J. Chem. Phys. 101, 5024 ~1994
J. E. Roberts and J. Schnitker, J. Phys. Chem. 99, 1322 ~1995

These papers show and apply the surface potential corrections for
ions solvated in a cubic box.
They refer to older work for non-cubic unit cells.
 From this it seems that Gromacs incorrectly determines
the dipole surface energy for non-cubic unit cells.
This affects Ewald with finite surface epsilon, as well as
dielectric constants from g_dipoles.
The error is only an incorrect weighting of the x, y and z components
of the system dipole. For an isotropic system this will probably
not result in significant errors for the dielectric constant.

I already noted before that the dipole terms are incorrect
in Gromacs when charge groups with net charge are present.
For ions consisting of one atom it is still correct,
but for charged molecules an appropriate location for the net
charge has to be used, instead of the current way, where
molecules are put into the box based on the location of the first atom.

It is now clear how to determine the electrostatic energy associated with
the shape of the unit cell. The problem here is that molecules
need to be made whole, which I don't know how to do efficiently
with the new domain decomposition.

Berk.