I'm trying to examine the optimal way to charge parameterize my molecules for solvation free energy calculations in MDYNAMIX. I've been able to reproduce the experimental solvation free energies of some molecules, e.g. highly polar ones but am having difficulty with molecules with mixed polar/hydrophobic character. Currently, I'm using the GAMESS binary with the CHELPG scheme, and while this seems better than, e.g. using Mulliken and Lowden charges, the resultant CHELPG charges are sometime unintuitive and don't reproduce accurate solvation free energies (e.g. the experimental DG for methanol is ~5kcal/mol and my calculation using CHELPG is ~3kcal/mol). I'd like to tighten up these to at minimum within 1kcal/mol.
Currently, I'm fitting CHELPG to the electrostatic potential at the MP2/cc-pVDZ level of theory, and I'm constraining the fitted partial charges only to reproduce the total charge of the system. For flexible molecules--with many degrees of freedom--I'm averaging the results over multiple conformation in an attempt to better reproduce the dynamical character of the dipole in solution. With this being said, I'm still not achieving the tight results I need.
Some things that I would like to mention.
Mulliken and Lowden charges exclude immediately.
In any electrostatics calculations basis sets with sufficiently diffuse functions should be used, not cc-pVDZ but aug-cc-pVDZ, not Def2-svp but Def2-svpd.
In fact Def2-svpd and aug-cc-pVDZ are the minimum for electrostatic calculations.
In the original AMBER 6-31G(d) used, however, note that it is not taken into account electron correlation. 6-31G(d) at the Hartree-Fock level overestimates electrostatics. This is a fortunate circumstance since in the liquid the dipole moments of the molecules greater than in vacuo, and this circumstance simulated by using 6-31G(d). If we calculate in MP2/aug-cc-pVDZ or PBE0/Def2-svpd, we must somehow take into account that we will use our model in a liquid, that is in a polar environment. This can be done using the COSMO model, for example.
As a first attempt, I would recommend those settings that are in the NWChem Electrostatics section, plus COSMO.
I've scoured the literature, and this is by far the most informative and concise answer I've come crossed. Thank you. I never would have considered diffusion function unless working with anions, but it makes since when I think about it.