Accurate reaction barrier heights of pericyclic reactions: Surprisingly large deviations for the CBS‐QB3 composite method and their consequences in DFT benchmark studies. Issue 9 (3rd February 2015)

• Record Type: Journal Article
• Authors: Karton, Amir; Goerigk, Lars
• Is Part Of: Journal of computational chemistry. Volume 36:Issue 9(2015)
• Page Start: 622
• Page End: 632
• Publication Date: 2015-02-03
• Subjects: Chemistry -- Data processing -- Periodicals; 542.85
• DOI: 10.1002/jcc.23837
• Languages: English
• ISSNs: 0192-8651

Abstract

<abstract abstract-type="main"> <title> <x xml:space="preserve">Abstract</x> </title> <p>Accurate barrier heights are obtained for the 26 pericyclic reactions in the BHPERI dataset by means of the high‐level W<italic>n</italic>‐F12 thermochemical protocols. Very often, the complete basis set (CBS)‐type composite methods are used in similar situations, but herein it is shown that they in fact result in surprisingly large errors with root mean square deviations (RMSDs) of about 2.5 kcal mol<sup>−1</sup>. In comparison, other composite methods, particularly G4‐type and estimated coupled cluster with singles, doubles, and quasiperturbative triple excitations [CCSD(T)/CBS] approaches, show deviations well below the chemical‐accuracy threshold of 1 kcal mol<sup>−1</sup>. With the exception of SCS‐MP2 and the herein newly introduced MP3.5 approach, all other tested Møller‐Plesset perturbative procedures give poor performance with RMSDs of up to 8.0 kcal mol<sup>−1</sup>. The finding that CBS‐type methods fail for barrier heights of these reactions is unexpected and it is particularly troublesome given that they are often used to obtain reference values for benchmark studies. Significant differences are identified in the interpretation and final ranking of density functional theory (DFT) methods when using the original CBS‐QB3 rather than the new W<italic>n</italic>‐F12 reference values for BHPERI. In particular, it is observed that the more accurate W<italic>n</italic>‐F12 benchmark results in lower statistical errors for those methods that are generally considered to be robust and accurate. Two examples are the PW6B95‐D3(BJ) hybrid‐meta‐general‐gradient approximation and the PWPB95‐D3(BJ) double‐hybrid functionals, which result in the lowest RMSDs of the entire DFT study (1.3 and 1.0 kcal mol<sup>−1</sup>, respectively). These results indicate that CBS‐QB3 should be applied with caution in computational modeling and benchmark studies involving related systems. © 2015 Wiley Periodicals, Inc.</p> </abstract>