Assessment of Gaussian-3 and density-functional theories on the G3/05 test set of experimental energies

Abstract

The G3/99 test set [L. A. Curtiss, K. Raghavachari, P. C. Redfern, and J. A. Pople, J. Chem. Phys. 112, 7374 (2000)] of thermochemical data for validation of quantum chemical methods is expanded to include 78 additional energies including 14 enthalpies of formation of the first- and second-row nonhydrogen molecules, 58 energies of molecules containing the third-row elements K, Ca, and Ga-Kr, and 6 hydrogen-bonded complexes. The criterion used for selecting the additional systems is the same as before, i.e., experimental uncertainties less than +/- 1 kcal/mol. This new set, referred to as the G3/05 test set, has a total of 454 energies. The G3 and G3X theories are found to have mean absolute deviations of 1.13 and 1.01 kcal/mol, respectively, when applied to the G3/05 test set. Both methods have larger errors for the nonhydrogen subset of 79 species for which they have mean absolute deviations of 2.10 and 1.64 kcal/mol, respectively. On all of the other types of energies the G3 and G3X methods are very reliable. The G3/05 test set is also used to assess density-functional methods including a series of new functionals. The most accurate functional for the G3/05 test set is B98 with a mean absolute deviation of 3.33 kcal/mol, compared to 4.14 kcal/mol for B3LYP. The latter functional has especially large errors for larger molecules with a mean absolute deviation of 9 kcal/mol for molecules having 28 or more valence electrons. For smaller molecules B3LYP does as well or better than B98 and the other functionals. It is found that many of the density-functional methods have significant errors for the larger molecules in the test set.

A reevaluation of computed proton affinities for the common α-amino acids

“…Second, G3MP2 is much more accurate than B3LYP in general, wideranging surveys. For example, with the extended G2 test set, G3 methods have average absolute errors in the 1 kcal/mol range whereas B3LYP is nearly 4 kcal/mol [23]. This is not a truly fair comparison in the present context because B3LYP is known to perform very well for PAs, but not so well for some heats of formation (the largest component of the G2 test set).…”

Long-range corrected hybrid density functionals with damped atom–atom dispersion corrections

“…To test the performance of ωB97X-D outside its training set, we also evaluate its performance on various test sets involving the additional 48 atomization energies in the G3/05 test set [56] (other than the 223 atomization energies in the G3/99 test set [45][46][47]), 30 chemical reaction energies taken from the NHTBH38/04 and HTBH38/04 databases [49,50], 29 non-covalent interactions [51,57], 166 optimized geometry properties of covalent systems [58], 12 intermolecular bond lengths [51] and one long-range charge transfer excitation curve of two well-separated molecules. There are a total of 286 pieces of data in the test sets.…”

Assessment of density functional methods with correct asymptotic behavior

“…For a comprehensive comparison of various density functional methods, we examine the performance of the local spin density approximation (LSDA) [48], generalized gradient approximations (GGAs) (PBE [49] and BLYP [50]), meta-generalized gradient approximations (MGGAs) (M06L [51] and VS98 [52]), global hybrid functionals (B3LYP [53] and M06-2X [54]), LC hybrid functionals (ωB97 [27], ωB97X [27], and ωB97X-D [28]), and AC model potentials (LB94 [33] and LBα [35]), on various test sets involving the 223 atomization energies (AEs) of the G3/99 set [55], the 40 ionization potentials (IPs), 25 electron affinities (EAs), and 8 proton affinities (PAs) of the G2-1 set [56], the 76 barrier heights (BHs) of the NHTBH38/04 and HTBH38/04 sets [57], the 22 noncovalent interactions of the S22 set [58], the additional 48 AEs of the G3/05 set [59], the 113 AEs of the AE113 database [60], the 131 vertical IPs of the IP131 database [31], the 131 vertical EAs of the EA131 database [31,60], the 131 fundamental gaps (FGs) of the FG131 database [31,60], two dissociation curves of symmetric radical cations, 19 valence excitation energies, 23 Rydberg excitation energies, and one long-range charge-transfer excitation curve of two well-separated molecules. As discussed in Ref.…”

“…Here, we examine the performance of these functionals on the additional 48 AEs of the G3/05 set [59], using the G3LargeXP basis set [66] and the extra fine grid EML (99,590). As can be seen in Table II, ωB97, ωB97X, and ωB97X-D remain the best in performance, even if they are not particularly parametrized using this set.…”