Random-phase-approximation-based correlation energy functionals: Benchmark results for atoms

TL;DR

This paper evaluates and benchmarks various extensions of the random phase approximation (RPA) for atomic correlation energies in density functional theory, demonstrating improved accuracy over the unmodified RPA.

Contribution

It introduces and assesses three simple extensions of RPA, providing benchmark results and numerical techniques for more accurate correlation energy calculations.

Findings

Extensions (a) and (c) improve correlation and ionization energy predictions.

Highly converged RPA data serve as benchmarks for future studies.

Numerical techniques developed are useful for complex systems.

Abstract

The random phase approximation (RPA) for the correlation energy functional of density functional theory has recently attracted renewed interest. Formulated in terms of the Kohn-Sham (KS) orbitals and eigenvalues, it promises to resolve some of the fundamental limitations of the local density and generalized gradient approximations, as for instance their inability to account for dispersion forces. First results for atoms, however, indicate that the RPA overestimates correlation effects as much as the orbital-dependent functional obtained by a second order perturbation expansion on the basis of the KS Hamiltonian. In this contribution, three simple extensions of the RPA are examined, (a) its augmentation by an LDA for short-range correlation, (b) its combination with the second order exchange term, and (c) its combination with a partial resummation of the perturbation series including the second order exchange. It is found that the ground state and correlation energies as well as the ionization potentials resulting from the extensions (a) and (c) for closed sub-shell atoms are clearly superior to those obtained with the unmodified RPA. Quite some effort is made to ensure highly converged RPA data, so that the results may serve as benchmark data. The numerical techniques developed in this context, in particular for the inherent frequency integration, should also be useful for applications of RPA-type functionals to more complex systems.