Can DFT and ab initiomethods describe all aspects of the potential energy surface of cycloreversion reactions?

We introduce a representative benchmark database of 20 cycloreversion reaction energies obtained by means of the high-level W1 thermochemical protocol. We use these benchmark values to assess the performance of a variety of contemporary DFT, double-hybrid DFT (DHDFT), standard ab initio, and compound thermochemistry methods. We show that this set of reaction energies provides an extremely challenging test for nearly all of the considered DFT and DHDFT methods. For example, about 80% of the considered functionals result in root-mean-square deviations (RMSDs) above 10kJ mol(-1). The best DFT and DHDFT procedures are B97X and DSD-PBEP86-D3, with RMSDs of 4.7 and 7.9kJ mol(-1), respectively. Coupled with the fact that the barrier heights for these reactions also pose a significant challenge for many DFT methods, this work shows that only a handful of functionals can quantitatively describe all aspects of the potential energy surface of this important class of reactions. In addition, this work shows that London dispersion effects are particularly large for this class of reactions. For example, empirical D3 dispersion corrections reduce the RMSDs for the DFT and DHDFT procedures by amounts ranging from 3.5 (PBE and B2K-PLYP) to 22.0 (BLYP) kJ mol(-1).