A note on the accuracy of KS-DFT densities

The accuracy of the density of wave function methods and Kohn-Sham (KS) density functionals is studied using moments of the density, < r(n)> = integral rho(r) r(n)d tau = integral(infinity)(0) 4 pi r(2) rho(r)r(n)dr, where n = -1, -2, 0, 1, 2, and 3 provides information about the short-and long-range behavior of the density. Coupled cluster (CC) singles, doubles, and perturbative triples (CCSD(T)) is considered as the reference density. Three test sets are considered: boron through neon neutral atoms, two and four electron cations, and 3d transition metals. The total density and valence only density are distinguished by dropping appropriate core orbitals. Among density functionals tested, CAMQTP00 and omega B97x show the least deviation for boron through neon neutral atoms. They also show accurate eigenvalues for the HOMO indicating that they should have a more correct long-range behavior for the density. For transition metals, some density functional approximations outperform some wave function methods, suggesting that the KS determinant could be a better starting point for some kinds of correlated calculations. By using generalized many-body perturbation theory (MBPT), the convergence of second-, third-, and fourth-order KS-MBPT for the density is addressed as it converges to the infinite-order coupled cluster result. For the transition metal test set, the deviations in the KS density functional theory methods depend on the amount of exact exchange the functional uses. Functionals with exact exchange close to 25% show smaller deviations from the CCSD(T) density. Published by AIP Publishing.