Assessment of density functional theory for thermochemical approaches based on bond separation reactions

The recently proposed ATOMIC protocol is a fully ab initio thermochemical protocol that rests upon the concept of bond separation reactions (BSRs) to correct for systematic errors of composite wave function approaches. It achieves high accuracy for atomization energies and derived heats of formation if basis set requirements for all contributing components are balanced carefully. The present work explores the potential of density functionals as possible replacements of composite wave function approaches in the ATOMIC protocol. Twenty density functionals are examined for their accuracy in thermochemical predictions based on calculated bond-separation energies and precomputed high-level data for the small parent molecules entering BSRs. The best density functionals outperform CCSD (coupled cluster with singles and doubles excitations), but none reaches the accuracy of well-balanced composite wave function approaches that consider quasiperturbational connected triples excitations at least with small basis sets. Some functionals show unexpected problems with bond separation reactions and are analyzed further with a model of empirically calibrated bond additivity corrections. Finally, the benefit of adding empirical dispersion terms to common density functionals is analyzed in the context of BSR-corrected thermochemistry.