An empirical charge transfer potential with correct dissociation limits

The empirical valence bond (EVB) method [J. Chem. Phys. 52, 1262 (1970)] has always embodied charge transfer processes. The mechanism of that behavior is examined here and recast for use as a new empirical potential energy surface for large-scale simulations. A two-state model is explored. The main features of the model are: (1) explicit decomposition of the total system electron density is invoked; (2) the charge is defined through the density decomposition into constituent contributions; (3) the charge transfer behavior is controlled through the resonance energy matrix elements which cannot be ignored; and (4) a reference-state approach, similar in spirit to the EVB method, is used to define the resonance state energy contributions in terms of 'knowable' quantities. With equal validity, the new potential energy can be expressed as a nonthermal ensemble average with a nonlinear but analytical charge dependence in the occupation number. Dissociation to neutral species for a gas-phase process is preserved. A variant of constrained search density functional theory is advocated as the preferred way to define an energy for a given charge