SuperHEAT: Pushing the Boundaries of Accuracy in Computational Thermochemistry

Abstract The superHEAT model chemistry is presented and used to explore the limits of accuracy in the computational prediction of the thermochemistry of first-row molecules. This model chemistry extends every component of the HEAT protocol to their effective computational limits, including rigorous treatments of CCSD basis-set completeness, post-CCSD(T) correlation, and vibrational zero-point energies. These extensive changes enable superHEAT to achieve mean-absolute deviations of only 9 cm–1 for the atomization energies of small, first-row molecules as benchmarked against the Active Thermochemical Tables. Combined with new statistical analysis, we determine a nominal 95% confidence interval for superHEAT of 25 cm–1, almost 1/16th of the chemical-accuracy proposed as a goal for the quantum chemical community in Pople’s Nobel Lecture. This represents a significant step in the development of computational methods that can provide ground thermochemical truth in the absence of experimental data.