Towards the intrinsic error of the correlation consistent Composite Approach (ccCA)

Abstract

The correlation consistent Composite Approach (ccCA) has been made more robust by (a) modifying the basis set used in computing B3LYP equilibrium geometries and harmonic vibrational frequencies so that the correlation consistent basis sets are used throughout ccCA; (b) separately extrapolating the MP2 and Hartree–Fock complete basis set limit energies; (c) uniformly treating unrestricted open shell wave functions; (d) utilizing newly recommended enthalpies of formation for C, B, Al, and Si atoms; and (e) using theoretically derived vibrational scale factors. This modified ccCA formulation has been used to compute the 454 energetic properties (enthalpies of formation, dissociation energies, ionization potentials, electron affinities, and proton affinities) in the G3/05 test set. This new formulation, which does not contain any optimized parameters, has a small systematic statistical bias (mean signed deviation of −0.20 kcal mol⁻¹), and has a mean absolute deviation of 1.01 kcal mol⁻¹ with the incorporation of modification d) or 0.99 kcal mol⁻¹ without. This is compared to a G4(MP2) MAD of 1.04 kcal mol⁻¹ and a G3(MP2) MAD of 1.39 kcal mol⁻¹. These modifications result in minimal change with respect to the computational requirements of the current ccCA methodology. The ccCA model chemistry is the first MP2-based model chemistry to achieve an accuracy of ± 1.00 kcal mol⁻¹ for the G3/05 training set without any optimized parameters, and it is the only MP2-based model chemistry uniformly applicable to systems comprised of elements from H to Kr.

Keywords:

• model chemistries
• composite methods
• correlation consistent
• ccCA
• thermochemistry

Acknowledgements

This research is partially supported by the National Science Foundation via a CAREER Award CHE-0239555 and via CHE-0809762 (to A.K.W.), and by a grant from the United States Department of Energy, Office of Basic Energy Sciences (to T.R.C.), Grant No. DE-FG02-03ER15387. Computational resources were provided via the National Science Foundation (CHE-0342824 and CHE-0741936), and by the Teragrid Alliance under TG-CHE010021 (to A.K.W.) and utilized the NCSA IBM p690 and the SGI Altix. Additional support was provided by the University of North Texas Academic Computing Services for the use of the UNT Research Cluster. CASCaM is supported by grant from the U.S. Department of Education. The authors would like to thank Henry F. Schaefer III for his extraordinarily scientific leadership and mentorship.