Fig. 1. Benchmark study of DFT methods.

a Trade-off between accuracy (left axis, blue) and computational cost (right axis, orange) for a selection of common QM methods. M06-2X/def2-TZVP was selected for subsequent calculations. MAE and CPU time were averaged over 368 different bonds. b Effect of conformer sampling. Molecules were optimized with MMFF94s, and the lowest-energy conformers were used to initialize DFT calculations. The plot shows the difference between the Boltzmann average enthalpy for the entire ensemble and the DFT-calculated enthalpy of the first conformer as a function of the number of optimizations performed. Exhaustive conformer sampling only changes the median resulting enthalpies by <0.5 kcal mol⁻¹, with a relatively narrow inner-quartile range (IQR).