Calibration and comparison of the Gaussian-2, complete basis set, and density functional methods for computational thermochemistry

We have reexamined several high-accuracy Gaussian-2, complete basis set and density functional methods for computational thermochemistry tin order of increasing speed): G2, G2(MP2), CBS-Q, G2(MP2,SVP), CBS-q, CBS-4, and B3LYP/6-311 + G(3 df,2p). We have employed Delta(f)H(298)(0) for the "extended G2 neutral test set" for this comparison. Several errors in previous studies have been corrected and experimental spin-orbit interactions have been included in all calculated atomic energies. The mean absolute deviations from experiment are 1.43, 1.76, 1.19, 1.64, 2.34, 2.66, and 3.43 kcal/mol, respectively. The maximum deviations from experiment are 10.6, 8.8, 8.1, 9.4, 11.4, 12.9, and 24.1 kcal/mol respectively. The species responsible for these maximum errors are in order: SiF4, SiF4, Cl2C=CCl2, F2C=CF2, ClF3, ClF3, and SiCl4. All seven methods have relatively large errors for bonds to halogens, but these errors are sufficiently systematic to benefit from empirical corrections. After a discussion of ill conditioning in the "bond separation reaction" implementation of isodesmic reactions, we determine "isodesmic bond additivity corrections" (BACs) for several types of bonds by least-squares fits to the heats of formation for 76 organic species with up to ten carbons and a variety of heteroatoms. The mean absolute deviations are reduced from 1.49, 1.93, 1.22, 1.53, 2.28, 3.09, and 3.45 kcal/mol to 0.55, 0.57, 0.77, 0.63, 1.03, 0.98, and 1.16 kcal/mol. The maximum errors are reduced to about 3 kcal/mol for all but the DFT method (4.2 kcal/mol). The BACs are especially useful for larger molecules with many similar bonds. For example, the CBS-Q error for Cl2C=CCl2 is reduced from 8.1 to 3.0 kcal/mol and the CBS-4 errors for benzene and naphthalene are reduced from 10.5 and 17.5 to 2.1 and 1.6 kcal/mol, respectively. (C) 1998 American Institute of Physics. [S0021-9606(98)3 1048-X].