Accurate energies calculated by empirical corrections to the local spin density approximation

We tested the efficacy of three empirical correction schemes on atomization energies calculated by the following theories: Kohn–Sham density functional theory (KS-DFT) with local spin density approximation (LSDA), two KS-DFT gradient-corrected methods, one hybrid Hartree–Fock/KS-DFT method similar to B3LYP, and the ab initio extrapolation procedures G1 and G2. Empirical corrections improved the LSDA results greatly, while the other theories were improved slightly or not at all. The best procedure for correcting LSDA atomization energies brings the mean absolute deviation from experiment from 38.3 to 4.0 kcal/mol on a subset of 44 molecules in the G1 dataset that were not used in deriving the empirical parameters. This corrected LSDA is interesting for three reasons: it could be a useful computational tool in some cases, it implies that the LSDA itself gives accurate energies for reactions where atomic coordinations stay unchanged, and it gives insight into the search of better functionals.