[gmx-developers] more on electrostatic decoupling

[Michael Shirts]

> I think I now understand where the confusion might come from.
> Some packages might do the pairs via a neighborlist.
>
> In Gromacs all pairs are stored explicitly in the topology.
> In mdrun we just loop over the list of pairs which are listed
> explicitly in the .top file and have been stored in the tpr
> file by grompp.

But HOW are the pairs interactions that are listed in the topology
calculated?  By the -nonbonded- part of the code, like the 1,4's?  So
that the PME calculations have nothing to do with it? I think that's
the real question.

> The current code would already do it, except that now one
> can not specify A and B state charges seperately for pair interactions
>  and thus they will be perturbed when the charges are perturbed.

So . . . what we want is the explicitly stated pair energies to remain
constant when the Coulombic interactions with the REST of the system
are changed, correct?

> The amount of water should be chosen large enough that the solute
> has no significant effect on the periodic interactions.
> Than one can take the limit to infinite water or not,
> but that will not make a difference for the solvation free energy.

I'm not entirely sure that we have a handle on this -- it would be
good to put some rigourous (i.e., not just handwaving) bound on this. 
 The ligand-ligand interactions are the same reguardless of the
solvent, but presumably, there is some effective screening by the
solvent molecules.  What this screening actually results in probably
involves the boundary conditions at infinity.

The other problem is, of course, the free energy of charging the vapor
states.  For ligand binding, the charging of the ligand in the two
simulations (decharging in presence of solvent, decharging in the
presence of solvent + protein); correct me if I'm wrong on this.  For
very small molecules with 1,2 and 1,3 exclusions, this is zero.  For
rigid molecules, its a trivial analytic calculation.

For desolvation of molecules with internal degrees of freedom, it's
more difficult, and we can't gloss over this step.

Best,
Michael