Density Relaxation in Time-Dependent Density Functional Theory: Combining Relaxed Density Natural Orbitals and Multireference Perturbation Theories for an Improved Description of Excited States

摘要

Making use of the recently developed excited state charge displacement analysis [E. Ronca et al., J. Chem. Phys. 140, 054110 (2014)], suited to quantitatively characterize the charge fluxes coming along an electronic excitation, we investigate the role of the density relaxation effects in the overall description of electronically excited states of different nature, namely, valence, ionic, and charge transfer (CT), considering a large set of prototypical small and medium-sized molecular systems. By comparing the response densities provided by time-dependent density functional theory (TDDFT) and the corresponding relaxed densities obtained by applying the Z-vector postlinear-response approach [N. C. Handy and H. F. Schaefer, /. Chem. Phys. 81, 5031 (1984)] with those obtained by highly correlated state-of-the-art wave function calculations, we show that the inclusion of the relaxation effects is imperative to get an accurate description of the considered excited states. We also examine what happens at the quality of the response function when an increasing amount of Hartree—Fock (HF) exchange is included in the functional, showing that the usually improved excitation energies in the case of CT states are not always the consequence of an improved description of their overall properties. Remarkably, we find that the relaxation of the response densities is always able to reproduce, independently of the extent of HF exchange in the functional, the benchmark wave function densities. Finally, we propose a novel and computationally convenient strategy, based on the use of the natural orbitals derived from the relaxed TDDFT density to build zero-order wave function for multireference perturbation theory calculations. For a significant set of different excited states, the proposed approach provided accurate excitation energies, comparable to those obtained by computationally demanding ab initio calculations.