Minimizing Density Functional Failures for Non-covalent Interactions beyond van der Waals Complexes
Kohn-Sham density functional theory offers a powerful and robust formalism for investigating the electronic structure of many-body systems while providing a practical balance of accuracy and computational cost unmatched by other methods. Despite this success, the commonly used semi-local approximations have difficulties in properly describing attractive dispersion interactions that decay with R-6 at large intermolecular distances. Even in the short to medium range, most semi-local density functionals fail to give an accurate description of weak interactions. The omnipresence of dispersion interactions, which are neglected in the most popular electronic structure framework, has stimulated intense developments during the last decade. In this account, we summarize our effort to develop and implement dispersion corrections that dramatically reduce the failures of both inter- and intramolecular interactions energies. The proposed schemes range from improved variants of “classical” atom pairwise dispersion correction (e.g., dD10) to robust formulations dependent upon the electron density. Emphasis has been placed on introducing more physics into a modified Tang and Toennies damping function and deriving accurate dispersion coefficients. Our most sophisticated and established density-dependent correction, dDsC, is based on a simple GGA-like reformulation of the exchange hole dipole moment introduced by Becke and Johnson. Akin to its classical precursor, dDsC, dramatically improves the interaction energy of a variety of standard density functionals simultaneously for typical intermolecular complexes and shorter-range interactions occurring within molecules. The broad applicability and robustness of the dDsC scheme is demonstrated on various representative reaction energies, geometries and molecular dynamic simulations. The suitability of the a posteriori correction is also established through comparisons with the more computationally demanding self-consistent implementation. The proposed correction is then exploited to identify the key factors at the origin of the errors in thermochemistry beyond van der Waals complexes. We especially focus on charge transfer and mixed valence complexes, which are relevant to the field of organic electronics. These types of complexes represent insightful examples for which the delocalization error may partially counterbalance the missing dispersion. Our devised methodology reveals the true performance of standard density functional approximations and the subtle interplay between the two types of errors. The presented analysis provides guidance for future functional development that could further improve the modeling of the structures and properties of molecular materials. Overall, the proposed state-of-the-art approaches have contributed to stress the crucial role of dispersion and improve their description in both straightforward van der Waals complexes and more challenging chemical situations. For the treatment of the latter, we have also provided relevant insights into which type of density functionals to favor.