| Literature DB >> 16965071 |
Geoffrey P F Wood1, Leo Radom, George A Petersson, Ericka C Barnes, Michael J Frisch, John A Montgomery.
Abstract
A restricted-open-shell model chemistry based on the complete basis set-quadratic Becke3 (CBS-QB3) model is formulated and denoted ROCBS-QB3. As the name implies, this method uses spin-restricted wave functions, both for the direct calculations of the various components of the electronic energy and for extrapolating the correlation energy to the complete-basis-set limit. These modifications eliminate the need for empirical corrections that are incorporated in standard CBS-QB3 to compensate for spin contamination when spin-unrestricted wave functions are used. We employ an initial test set of 19 severely spin-contaminated species including doublet radicals and both singlet and triplet biradicals. The mean absolute deviation (MAD) from experiment for the new ROCBS-QB3 model (3.6+/-1.5 kJ mol(-1)) is slightly smaller than that of the standard unrestricted CBS-QB3 version (4.8+/-1.5 kJ mol(-1)) and substantially smaller than the MAD for the unrestricted CBS-QB3 before inclusion of the spin correction (16.1+/-1.5 kJ mol(-1)). However, when applied to calculate the heats of formation at 298 K for the moderately spin-contaminated radicals in the G2/97 test set, ROCBS-QB3 does not perform quite as well as the standard unrestricted CBS-QB3, with a MAD from experiment of 3.8+/-1.6 kJ mol(-1) (compared with 2.9+/-1.6 kJ mol(-1) for standard CBS-QB3). ROCBS-QB3 performs marginally better than standard CBS-QB3 for the G2/97 set of ionization energies with a MAD of 4.1+/-0.1 kJ mol(-1) (compared with 4.4+/-0.1 kJ mol(-1)) and electron affinities with a MAD of 3.9+/-0.2 kJ mol(-1) (compared with 4.3+/-0.2 kJ mol(-1)), but the differences in MAD values are comparable to the experimental uncertainties. Our overall conclusion is that ROCBS-QB3 eliminates the spin correction in standard CBS-QB3 with no loss in accuracy.Entities:
Year: 2006 PMID: 16965071 DOI: 10.1063/1.2335438
Source DB: PubMed Journal: J Chem Phys ISSN: 0021-9606 Impact factor: 3.488