Successes and failures of DFT functionals in acid/base and redox reactions of organic and biochemical interest

Abstract

The performance of 18 different DFT functionals in the prediction of absolute and relative energies of organic and biochemical acid/base and redox reactions was evaluated, using MP2 extrapolated to the complete basis set limit and CCSD(T)/aug-cc-pVTZ energies as benchmark. Absolute reduction energies were predicted with relatively large average errors (2–4 kcal mol −1) except for the best functional, PBE0 (1.3 ± 1.2 kcal mol −1). The DFT predictions of relative reduction energies afforded mean unsigned errors (2.1 kcal mol −1 for the best functional, PBE0) which, although relatively large, are smaller than those obtained with MP2 with comparable basis sets (3.2 kcal mol −1). This relatively poor performance was mostly due to significant underprediction of the reduction energy of a model disulfide bridge, and overprediction of that of a model quinone, as eliminating these reaction from the test set enables most of the tested hybrid-GGA functionals (B3LYP, B3PW91, B97-1, BHHLYP, PBE0, PBE1PW91, X3LYP), though not meta-GGA or meta-hybrid-GGA functionals, to achieve small mean unsigned errors (<1.5 kcal mol −1). In the acid/base reaction test set, several hybrid-GGA and meta-hybrid-GGA functionals (BHHLYP, B97-1, B97-2, X3LYP, M06-2X, as well as the popular B3LYP) yielded small errors (0.8–1.5 kcal mol −1) in the computation of relative energies across the tested model reactions. For most of the proton transfers between different acid/base pairs (or electron transfer between redox pairs) tested we could find at least one functional with very high accuracy (error < 0.2 kcal mol −1). The reactions involving electron transfer between quinone (or disulfide) and other redox groups stand out as the clearest example of the shortcomings of DFT methods, as the best functionals are most often wrong by 1–3 kcal mol −1.