Atomic charges for classical simulations of polar systems

Structure and reactivity often are dependent on the polarity of chemical bonds. This relationship is reflected by atomic charges in classical (semiempirical) atomistic simulations; however, disagreement between atomic charges from accurate experimental investigations, ab initio methods, and serniempirical methods has not been resolved. Our aim is to improve the basic understanding of the polarity Of compounds with a view to make force-field parametrizations more consistent and physically realistic. The concept is based on the relationship between the atomization energies of the elements and the possible strength of covalent bonding and the relationship between the ionization energies/electron affinities of the elements and the possible strength of ionic bonding. Both quantities, energetically, are of the same order of magnitude and influence atomic charges in a compound, which we illustrate by trends across the periodic table. The relationship between the pure elements and a given compound is shown in an extended Born model. We note that the extended Born model can be used to obtain physically justified charge estimates, relative to available reference compounds. This semiempirical concept has a stronger foundation than electronegativity equalization [Rappe, A. K.; Goddard, W. A., III. J. Phys. Chem. 1991, 95, 3358-3363], which is based on isolated gas-phase atoms and does not include covalent bonding contributions. We demonstrate the assignment of atomic charges for SiO2, the aluminosilicates mica and montmorillonite, and tetraalkylammonium ions, including local charge defects by Si --> Al- ... K+ and Al --> Mg- ... Na+ substitution. Our estimates of atomic charges correlate well with experimental data. Classical force fields based on these charges exhibit up to 1 order of magnitude less error in reproducing crystal geometries (only similar to0.5% deviation in unit-cell parameters), phase diagrams, and interfacial energies.