Modeling Polarization through Induced Atomic Charges

Abstract

A distributed atomic charge model to account for intermolecular polarization is presented. The model is an approximation to the method of calculating induced dipoles from atomic polarizabilities. In this model, induced atomic dipoles are represented by charges on the atom itself plus neighboring atomic sites. The model therefore avoids the evaluation of charge−dipole and dipole−dipole interactions. Formulas for calculating the induced atomic charges are presented and the approximations involved are discussed. Numerical examples show that our model recovers a substantial part of the polarization energy while the gain in computational efficiency with respect to the induced dipole model is 2−4-fold, being in the upper range when gradients are also evaluated. The use of moments induced by permanent charges only, i.e., calculated without iteration, was found to give an effective approximation with a gain in computational efficiency of up to 7-fold. Analysis of the numerical discrepancies between the induced charge and the induced dipole model showed that the induced atomic charge model of polarization has a 10−40% error for close molecular interactions with the error at the lower end for water−water interactions at equilibrium geometries and it gives an improving approximation with increasing intermolecular separations. The ability of the induced charge model to improve the nonpolarizable TIP3P water model was investigated by comparing the properties of liquid water (e.g., density, diffusion coefficient, radial distribution function) predicted by molecular dynamics simulations. A noniterative polarization model combined with parameters consistent with experimental data (geometry, vacuum dipole moment, polarizability) and with adjusted Lennard-Jones parameters was shown to give properties in good agreement with experimental data. This result suggests that an effective polarizable force field can be built by combining the induced charge model with further energy components. It is argued that the induced charge model represents a significant step toward the best available distributed charge approximation to the induced dipole model and so conclusions drawn for the performance of the model bear significance to other distributed charge models.