Toward Accurate Quantum Mechanical Thermochemistry: (2) Optimal Methods for Enthalpy Calculations from Comprehensive Benchmarks of 284 Model Chemistries.

Abstract

Accurate and efficient computations of the standard enthalpies of formation (ΔH_f^°) for small organic molecules are crucial for diverse chemical engineering and scientific applications. Building on part 1 of this work [J. Phys. Chem. A 2024, 128, 21, 4335-4352], we systematically benchmark 284 model chemistries for ΔH_f^° computations. These methods span semiempirical approaches, density functional theory (DFT), wave function theory, and composite schemes. We derive Petersson- and Melius-type bond-additivity corrections (BACs) for each method using a curated database of 421 reference species. We further validate the top-performing methods using an independent test set of 500 species, including ions, radicals, and other challenging cases. Across nearly all methods and species, BACs significantly improve accuracy, especially for neutral singlet species. Composite schemes coupling moderate-level DFT geometries with local coupled-cluster single-point energies strike an excellent balance between cost and accuracy, often approaching chemical accuracy (≤1 kcal/mol). Notably, DLPNO-CCSD(T)-F12d/cc-pVTZ-F12//ωB97X-D/def2-TZVPD with Petersson BAC attains the benchmark-best mean absolute error (MAE) of 0.57 kcal/mol. Switching to DLPNO-CCSD(T)-F12d/cc-pVDZ-F12//GFN2-xTB reduces the computational cost by an order of magnitude, with only a modest increase in MAE (0.96 kcal/mol). Although carefully tuned model chemistries can also benefit charged and open-shell species, the scarcity of robust reference data in these areas highlights the need for broader, high-accuracy thermochemistry datasets. Overall, this benchmark provides practical guidance on selecting optimal model chemistries to efficiently compute accurate ΔH_f^° under varied computational constraints and molecular complexities, laying a foundation for large-scale, high-throughput thermochemical calculations that will support data-driven discovery and industrial applications.