Toward improved force fields. 1. Multipole-derived atomic charges

The electrostatic energy component of classical force fields often includes some of the polarization energy component implicitly through the choice of atomic charges. In this and the subsequent articles we describe progress toward separating and accurately calculating both electrostatic and polarization energies. In the present contribution the distributed point charge representation of electrostatics is retained. Charges derived from several quantum chemical models including electron correlation at various levels are compared. We found that ignoring electron correlation in deriving charges for our force field can result in an error of several kcal mol in free energy difference simulations, and that this error can be comparable to the effect of ignoring polarization. We conclude that the accurate treatment of polarization in force fields also requires an accurate treatment of electron correlation. The work is based on the relatively new MPFIT charge fitting procedure (Ferenczy, G. G. J. Comput. Chem. 1991, 12, 913; Chipot, C.; et. al. J. Phys. Chem. 1993, 97, 6628), which produces point charges comparable to conventional molecular electrostatic potential-derived charges. These new charges are slightly less polar and more transferable and contain more chemical sense, but they are still conformationally dependent. The significance of different levels of electron correlation in these charges was examined through regression analysis, to determine scaling relationships between the charges, and through free energy difference simulations, to determine the effect of using alternative charge sets. The free energy calculations indicate that the Becke-Lee, Yang, and Parr nonlocal density functional method gives charges similar to second-order Møller-Plessett perturbation theory. The charges are shown to be insensitive to the precision of the quadrature used in the density functional calculations. For polar molecules, these methods generally gave free energies of hydration which were significantly smaller than those computed using Hartree-Fock charges. When the Hartree-Fock charges are scaled to reproduce the higher quality charges, the error is usually reduced, but is still significant in some cases. Since many force fields effectively exploit the polarity of the Hartree-Fock charges to mimic the effects of polarization in an ad hoc way, this result has important implications for force field design, as mentioned above. It is suggested that the electron density calculated by the density functional method is a suitable starting point to derive distributed multipole sets for use in force fields which include explicit polarization.