Electron affinities of the first- and second-row atoms: Benchmarkab initioand density-functional calculations

A benchmark ab initio and density-functional theory (DFT) study has been carried out on the electron affinities of the first- and second-row atoms. The ab initio study involves basis sets of $\mathrm{spdfgh}$ and $\mathrm{spdfghi}$ quality, extrapolations to the one-particle basis set limit, and a combination of the coupled cluster with all single, double (and triple) excitations [CCSD(T)], CCSDT, and full configuration-interaction electron correlation methods. Scalar relativistic and spin-orbit coupling effects were taken into account. On average, the best ab initio results agree to better than 0.001 eV with the most recent experimental results. Correcting for imperfections in the CCSD(T) method improves the mean absolute error by an order of magnitude, while for accurate results on the second-row atoms inclusion of relativistic corrections is essential. The latter are significantly overestimated at the self-consistent-field level; for accurate spin-orbit splitting constants of second-row atoms, inclusion of $2s,2p$ correlation is essential. In the DFT calculations it is found that results for the first-row atoms are very sensitive to the exchange functional, while those for second-row atoms are rather more sensitive to the correlation functional. While the Lee-Yang-Parr (LYP) correlation functional works best for first-row atoms, its PW91 counterpart appears to be preferable for second-row atoms. Among “pure DFT” (nonhybrid) functionals, G96PW91 (Gill 1996 exchange combined with Perdew-Wang 1991 correlation) puts in the best overall performance, actually slightly better than the popular hybrid B3LYP functional. B3PW91 outperforms B3LYP, while the recently proposed one-parameter hybrid functionals such as B1LYP seem clearly superior to B3LYP and B3PW91 for first-row atoms. The best results overall are obtained with the one-parameter hybrid modified Perdew-Wang (mPW1) exchange functionals of Adamo and Barone [J. Chem. Phys. 108, 664 (1998)], with $\mathrm{mPW}1\mathrm{LY}p$ yielding the best results for first-row, and mPW1PW91 for second-row atoms. Indications are that a hybrid of the type a $\mathrm{mPW}1\mathrm{LYP}+\left(1\mathrm{}-a\right)$ mPW1PW91 yields better results than either of the constituent functionals.