Approaching closed-shell accuracy for radicals using coupled cluster theory with perturbative triple substitutions

Abstract

The erratic performance of CCSD(T) for radicals is analyzed using non-Hartree–Fock references as a starting point for correlations and by testing the (2) approach as an alternative to (T) for including higher-order correlation effects. Though CCSD(2) improves upon CCSD(T), correlating from a better-behaved reference makes both theories robust. Comparisons of calculated harmonic frequencies against experiment in a set of diatomic radicals from Brueckner-like orbitals demonstrate improvement approaching closed-shell accuracy. Additionally, we find that using BLYP Kohn–Sham orbitals yields similar improvements, and they are therefore a useful, inexpensive reference for high-level correlation methods in difficult systems. Root-mean-square errors of 1.0–1.2% are found in the cc-pVQZ basis for predicted harmonic frequencies in the test set using OD(T) and KS-CCSD(T), making these approaches quite competitive with CCSD(T) for closed-shell molecules. Finally, these improvements are correlated with spin contamination and the rate of change of the electron density with nuclear displacement.