The DBH24/08 database and its use to assess electronic structure model chemistries for chemical reaction barrier heights

The diverse barrier height database DBH24 is updated by using W4 and W3.2 data (Karton, A.; Tarnopolsky, A.; Lame re, J.-F.; Schatz, G. C.; Martin, J. M. L. J. Phys. Chem. A 2008, 112, 12868) to replace previous W1 values; we call the new database DBH24/08. We used the new database to assess 348 model chemistries, each consisting of a combination of a wave function theory level or a density functional approximation with a one-electron basis set. All assessments are made by simultaneous consideration of accuracy and cost. The assessment includes several electronic structure methods and basis sets that have not previously been systematically tested for barrier heights. Some conclusions drawn in our previous work (Zheng, J.; Zhao, Y.; Truhlar, D. G. J. Chem. Theory Comput. 2007, 3, 569) are still valid when using this improved database and including more model chemistries. For example, BMC-CCSD is again found to be the best method whose cost scales as N6, and its cost is an order of magnitude smaller than the N7 method with best performance-to-cost ratio, G3SX(MP3), although the mean unsigned error is only marginally higher, namely 0.70 kcal/mol vs 0.57 kcal/mol. Other conclusions are now broader in scope. For example, among single-reference N5 methods (that is, excluding MRMP2), we now conclude not only that doubly hybrid density functionals and multicoefficient extrapolated density functional methods perform better than second-order Møller-Plesset-type perturbation theory (MP2) but also that they perform better than any correlation-energy-scaled MP2 method. The most recommended hybrid density functionals, if functionals are judged only on the basis of barrier heights, are M08-SO, M06-2X, M08-HX, BB1K, BMK, PWB6K, MPW1K, BHandHLYP, and TPSS25B95. MOHLYP and HCTH are found to be the best performing local density functionals for barrier heights. The basis set cc-pVTZ+ is more efficient than aug-cc-pVTZ with similar accuracy, especially for density functional theory. The basis sets cc-pVDZ+, 6-31+G(d,p), 6-31B(d,p), 6-31B(d), MIDIY+, MIDIX+, and MIDI! are recommended for double-ζ-quality density functional calculations on large systems for their good balance between accuracy and cost, and the basis sets cc-pVTZ+, MG3S, MG3SXP, and aug-cc-pVDZ are recommended for density functional calculations when larger basis sets are affordable. The best performance of any methods tested is attained by CCSD(T)(full)/augcc-pCV(T+d)Z with a mean unsigned error of 0.46 kcal/mol; however, this is several orders of magnitude more expensive than M08-SO/cc-pVTZ+, which has a mean unsigned error of only 0.90 kcal/mol.