Effects of London dispersion on the isomerization reactions of large organic molecules: a density functional benchmark study

A benchmark set of 24 isomerization reactions of large organic molecules (consisting of 24 to 81 atoms) is presented (termed ISOL). The molecules are much larger than what is typically considered in thermochemical tests. To obtain reference isomerization energies, complete basis set (CBS) extrapolations at the (SCS)-MP2 level have been computed that are augmented by perturbative third-order corrections (SCS-MP3 and MP2.5 methods). Based on these carefully examined reference data, a diverse set of common density functionals varying from GGA to double-hybrid functional level with and without dispersion correction (DFT-D) is tested. Double-hybrid and the PBE0 hybrid functionals are found to be the methods of choice for the type of main group thermochemistry examined here. For all isomerizations with an average reaction energy of 22.7 kcal mol(-1) (in a range between 0.5 and 74.5 kcal mol(-1)), PBE0-D, B2PLYP-D and B2GP-PLYP-D yield mean absolute deviations of 2.5, 4.1 and 2.9 kcal mol(-1). Most importantly it is found that the use of a dispersion correction is essential if such large molecules are considered. For all DFT methods the MAD is lowered very significantly by 1.4-5.0 kcal mol(-1) when DFT-D is used. Intramolecular (mainly medium-range) London dispersion interactions account in some cases for more than 50% (41 kcal mol(-1)) of the isomerization energy even though the size of the systems remains unchanged. This study also demonstrates for the first time clearly that typical DFT errors are larger than expected (about 5 kcal mol(-1)) and that chemical accuracy (about 1 kcal mol(-1)) even for these electronically well-behaved molecules is currently not reached by DFT. We propose this new test set as a difficult challenge for electronic structure methods that claim to be routinely applicable to large molecules. We also suggest to use a distance range resolved dispersion energy as a diagnostic for problematic cases in DFT.