Performance of B3LYP density functional methods for a large set of organic molecules

Testing of the commonly used hybrid density functional B3LYP with the 6-31G(d), 6-31 G(d,p), and 6-31+G(d,p) basis sets has been carried out for 622 neutral, closed-shell organic compounds containing the elements C, H, N, and O. The focus is comparison of computed and experimental heats of formation and isomerization energies. In addition, the effect of an empirical dispersion correction term has been evaluated and found to improve agreement with the experimental data. For the 622 compounds, the mean absolute errors (MAE) in the heats of formation are 3.1, 2.6, 2.7, and 2.4 kcal/mol for B3LYP/6-31G(d), B3LYP/6-31G(d,p), B3LYP/6-31+G(d,p), and B3LYP/6-31+G(d,p) with the dispersion correction. A diverse set of 34 isomerizations highlights specific issues of general interest, such as performance on differences in steric effects, conjugation, and bonding. The corresponding MAEs for the isomerizations are 2.7, 2.4, 2.2, and 1.9 kcal/mol. Improvement is obtained for isomerizations of amines and alcohols when both polarization and diffuse functions are used, but the overstabilization of linear alkanes compared to branched isomers can be relieved only with the dispersion correction. Besides the insights on DFT methods, the study also aimed to quantify the gains in accuracy that can be achieved by replacing energetics from NDO-based semiempirical methods with DFT results. Since the MAEs obtained with the PDDG/PM3 method for the 622 heats of formation and 34 isomerizations are 2.8 and 2.3 kcal/mol, negligible advantage in accuracy for the B3LYP-based methods emerged in the absence of the dispersion corrections.