797 publications from this institution
Errors in kinetic and exchange contributions to the molecular bonding energy are assessed for approximate density functionals by reference to near-exact Hartree-Fock values. From the molecular calculations of Allan et al.and of Lee and Ghosh, it is demonstrated that the density-gradient expansion does not accurately describe the noninteracting kinetic contribution to the bonding energy, even when this expansion is carried to fourth order and applied in its spin-density-functional form to accurate Hartree-Fock densities. In a related study, it is demonstrated that the overbinding of molecules such as ${\mathrm{N}}_{2}$ and ${\mathrm{F}}_{2}$, which occurs in the local-spin-density (LSD) approximation for the exchange-correlation energy, is not attributable to errors in the self-consistent LSD densities. Contrary to expectations based upon the Gunnarsson-Jones nodality argument, it is found that the LSD approximation for the exchange energy can seriously overbind a molecule even when bonding does not create additional nodes in the occupied valence orbitals. LSD and exact values for the exchange contribution to the bonding energy are displayed and discussed for several molecules.
The negative correlation energy ${\ensuremath{\epsilon}}_{c}{(r}_{s},\ensuremath{\zeta})$ per particle of a uniform electron gas of density parameter ${r}_{s}$ and spin polarization $\ensuremath{\zeta}$ is well known, but its spin resolution into $\ensuremath{\uparrow}\ensuremath{\downarrow},$ $\ensuremath{\uparrow}\ensuremath{\uparrow},$ and $\ensuremath{\downarrow}\ensuremath{\downarrow}$ contributions is not. Widely used estimates are incorrect, and hamper the development of reliable density functionals and pair distribution functions. For the spin resolution, we present interpolations between high- and low-density limits that agree with available quantum Monte Carlo data. In the low-density limit for $\ensuremath{\zeta}=0,$ we find that the same-spin correlation energy is unexpectedly positive, and we explain why. We also estimate the $\ensuremath{\uparrow}$ and $\ensuremath{\downarrow}$ contributions to the kinetic energy of correlation.