To achieve chemical accuracy in predicting the thermochemical properties of metal-containing molecules, one typically requires CCSDTQP—a highly intractable model that includes up to pentuple excitations. By initiating CCSD(T) calculations from DFT densities rather than Hartree-Fock, we found that chemically accurate bond dissociation energies for metal diatomics can be achieved just by including approximate triple excitations. Our results show that quantitative accuracy for metal complexes is within reach through only a light modification to quantum chemistry's gold-standard model. These results have broad implications for catalysis research and for the training of machine learning models on high-quality yet affordable benchmark data.
