Combinations of coupled cluster, density functionals, and the random phase approximation for describing static and dynamic correlation, and van der Waals interactions

TL;DR

This paper introduces a novel combination of coupled cluster methods, density functionals, and the random phase approximation to accurately describe static and dynamic correlation, including van der Waals interactions, without double counting or increased computational cost.

Contribution

It develops CCD0 and BD0 methods enhanced with meta-GGA density functionals and range separation, improving the treatment of static and dynamic correlations and long-range van der Waals forces.

Findings

High accuracy for weakly correlated systems

Reasonable performance on strongly correlated problems

Effective long-range van der Waals interaction modeling

Abstract

Contrary to standard coupled cluster doubles (CCD) and Brueckner doubles (BD), singlet-paired analogues of CCD and BD (denoted here as CCD0 and BD0) do not break down when static correlation is present, but neglect substantial amounts of dynamic correlation. In fact, CCD0 and BD0 do not account for any contributions from multielectron excitations involving only same-spin electrons at all. We exploit this feature to add---without introducing double counting, self-interaction, or increase in cost---the missing correlation to these methods via meta-GGA density functionals (TPSS and SCAN). Furthermore, we improve upon these CCD0+DFT blends by invoking range separation: the short- and long-range correlations absent in CCD0/BD0 are evaluated with DFT and the direct random phase approximation (dRPA), respectively. This corrects the description of long-range van der Waals forces. Comprehensive benchmarking shows that the combinations presented here are very accurate for weakly correlated systems, while also providing a reasonable description of strongly correlated problems without resorting to symmetry breaking.