Comprehensive Thermochemical Benchmark Set of Realistic Closed-Shell Metal Organic Reactions

• Published in: Journal of chemical theory and computation 2018-05, Vol.14 (5), p.2596-2608
• Main Authors: Dohm, Sebastian, Hansen, Andreas, Steinmetz, Marc, Grimme, Stefan, Checinski, Marek P
• Format: Article
• Language: English
• Subjects: Benchmarks; Catalysis; Chemical reactions; Metal compounds; Organic chemistry; Test sets; Thermochemistry; Transition metal compounds
• ISSN: 1549-9618; 1549-9626
• DOI: 10.1021/acs.jctc.7b01183

We introduce the new MOR41 benchmark set consisting of 41 closed-shell organometallic reactions resembling many important chemical transformations commonly used in transition metal chemistry and catalysis. It includes significantly larger molecules than presented in other transition metal test sets and covers a broad range of bonding motifs. Recent progress in linear-scaling coupled cluster theory allowed for the calculation of accurate DLPNO-CCSD­(T)/CBS­(def2-TZVPP/def2-QZVPP) reference energies for 3d,4d,5d-transition metal compounds with up to 120 atoms. Furthermore, 41 density functionals, including seven GGAs, three meta-GGAs, 14 hybrid functionals, and 17 double-hybrid functionals combined with two different London dispersion corrections, are benchmarked with respect to their performance for the newly compiled MOR41 reaction energies. A few wave function-based post-HF methods as, e.g., MP2 or RPA with similar computational demands are also tested and in total, 90 methods were considered. The double-hybrid functional PWPB95-D3­(BJ) outperformed all other assessed methods with an MAD of 1.9 kcal/mol, followed by the hybrids ωB97X-V (2.2 kcal/mol) and mPW1B95-D3­(BJ) (2.4 kcal/mol). The popular PBE0-D3­(BJ) hybrid also performs well (2.8 kcal/mol). Within the meta-GGA class, the recently published SCAN-D3­(BJ) functional as well as TPSS-D3­(BJ) perform best (MAD of 3.2 and 3.3 kcal/mol, respectively). Many popular methods like BP86-D3­(BJ) (4.9 kcal/mol) or B3LYP-D3­(BJ) (4.9 kcal/mol) provide significantly worse reaction energies and are not recommended for organometallic thermochemistry considering the availability of better methods with the same computational cost. The results regarding the performance of different functional approximations are consistent with conclusions from previous main-group thermochemistry benchmark studies.