[gmx-users] Verlet and Group lists produce difference charge/voltage distributions

[Matt Thompson]

I am encountering an error in the electric potential distribution when I
use Verlet lists, but not Group lists. I'm using gromacs 5.1.

My system is a channel simulation: two parallel surfaces are separated by a
few nanometers of electrolyte molecules (ionic liquids in this case). The
separation is 6.7 nm, which is long enough to observe bulk-like behavior in
the center and consider the two interfaces as not interacting with each
other.

The potential phi (V) distribution is a double integral over the charge
distribution using Poisson's Equation. The boundary conditions are phi = on
a surface and d(phi)/dz = 0 in the bulk. This is done over the width of the
channel (i.e. z = 0 to z = L) and even though there is no boundary
condition at
z = L, we expect phi = 0 to hold true because the system is symmetric. This
happens just fine with Group cutoff-scheme, but with Verlet lists, the
integral results in phi ~ 0.15 V at the other surface. This is not an
acceptable error and the only difference in these two runs was the
cutoff-scheme.

The log files don't throw any obvious errors and I'm not using anything
that is currently unsupported by Verlet (
http://www.gromacs.org/Documentation/Cut-off_schemes#Group_vs_Verlet_pair-lists
). I don't think the rest of my .mdp parameters are exotic but I'm
including it.

I'd obviously like to be able to use Verlet lists for performance, but I'm
also curious if this may point to an issue with or improper use of Verlet
lists.

Thanks,
Matt Thompson