Thermochemical benchmarking of hydrocarbon bond separation reaction energies: Jacob's ladder is not reversed!

We reinvestigate the performance of Kohn-Sham density functional (DF) methods for a thermochemical test set of bond separation reactions of alkanes (BSR36) published recently by Steinmann et al. [J. Chem. Theory Comput. 5, 2950 (2009)]. According to our results, the tested approximations perform for this rather special benchmark as usual. We show that the choice of reference enthalpies plays a crucial role in the assessment. Due to the large stoichiometric factors involved, errors of various origin are strongly amplified. Inconsistent reference data are avoided by computing reference energies at the CCSD(T)/CBS level. These are compared to results for a variety of standard DFs. Two different versions of London dispersion corrections (DFT-D2 and DFT-D3) are applied and found to be very significant. The most accurate results are obtained with B2GPPLYP-D2 (MAD = 0.4 kcal mol-1) B2PLYP-D2 (MAD = 0.5 kcal mol-1) and B97-D2 (MAD = 0.9 kcal mol-1 methods. Dispersion corrections not only improve the computed BSR energies but also diminish the accuracy differences between the DFs. The previous DFT-D2 version performs better due to error compensation of medium-range correlation effects between the semi-classical and the density-based description. We strongly recommend not to overinterpret results regarding DF accuracy when based on a single set of chemical reactions and to use high-level theoretical data for benchmarking purposes.