predicting thermochemical data of molecules. In the second section of this paper we summarize current state-of-the-art methods in computational thermochemistry. In the third section we describe in more detail the Gaussian-n approach and then discuss a recent development in this area, Gaussian-3 (G3) theory, which achieves a new level of accuracy. Finally, in the fourth section we assess prospects for the future of computational thermochemistry.

Current State of the Art in Computational Thermochemistry

The available methods for computing thermochemical data range from empirical schemes to ab initio molecular orbital theory.2 In this paper we restrict our discussion to the ab initio based methods. They can be roughly divided into three types: (1) very high level quantum chemical calculations with no experimental input; (2) composite techniques that combine moderate level ab initio quantum chemical calculations with some form of molecule-independent empirical parameters; and (3) techniques that use molecule-dependent empirical parameters obtained from accurate experimental data in combination with moderate level ab initio quantum chemical calculations. In this section we discuss some examples of these different approaches.

In principle it is known how to compute the thermochemical properties of most molecules to very high accuracy (0.5 kcal/mol). This can be achieved by using very high levels of correlation, such as are obtained with coupled clustered [CCSD(T)] or quadratic configuration [QCISD(T)] methods, and very large basis sets. The results of these calculations are then extrapolated to the complete basis set limit and corrected for some smaller effects such as core-valence effects and atomic spin-orbit effects. Unfortunately, this approach is limited to small molecules because of the ˜n7 scaling (with respect to the number of basis functions) of the correlation methods and the large basis sets used. This methodology has been used by Dunning, Feller, and coworkers3,4 at Pacific Northwest National Laboratory to systematically study a large number of small molecules having one and two non-hydrogen atoms. They have applied these very high level calculations to a diverse enough set of molecules to show that the methodology does perform to a very high level of accuracy.

Other groups have also used this type of approach to computational thermochemistry. Grey, Janssen, and Schaefer5 have used CCSD(T) with large basis sets to study the thermochemistry of CHn and SiHn hydrides, and some of their cations. They achieved bond energies accurate to 0.5 kcal/mol without any empirical corrections for these small molecules. Petersson and coworkers6 have used QCISD(T) with very large basis sets and have obtained a mean absolute deviation of 0.53 for a subset of the G2 test set of reaction energies. Bauschlicher, Langhoff, Taylor, and coworkers7 have used an approach based on converging to the one-particle limit through the use of atomic natural orbitals at a moderate level of correlation treatment. The correlation treatment is calibrated against full configuration interaction calculations on smaller systems or against accurate experimental data, in some cases. They have achieved accuracies of 1 kcal/mol or better using these methods.

Since the very high level calculations are difficult to extend to larger molecules, an alternative


For a recent review see: Computational Thermochemistry, K.K. Irikura and D.J. Frurip, eds., ACS Symposium Series 677, American Chemical Society, Washington D.C. (1998).


K.A. Peterson and T.H. Dunning, Jr., J. Chem. Phys. 106, 4119 (1997).


D. Feller and K.A. Peterson, J. Chem. Phys. 108, 154 (1998).


R.S. Grev, C.L. Janssen, and H.F. Schaefer III, J. Chem. Phys. 97, 8389 (1992).


J.A. Montgomery, Jr., J.W. Ochterski, and G.A. Petersson, J. Chem. Phys. 101, 5900 (1994).


C.W. Bauschlicher, Jr., and S.R. Langhoff, Science 254, 394 (1991).

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement