I. The electron structure of the Criegee intermediates. II. The electronic structure of pyrazine. III. Approximate integral methods and correlated wavefunctions

Part 1: The Electronic Structure of the Criegee Intermediate: Ramifications for the Mechanism of Ozonolysis. Generalized valence bond (GVB) and configuration interaction (CI) calculations using a double zeta basis set have been carried out on methylene peroxide (H_2COOO), the reactive intermediate in the Criegee mechanism for ozonolysis of olefins. The ground state of methylene peroxide (using an open geometry) is shown to be a singlet biradical rather than a zwitterion. A strong analogy between methylene peroxide and its isoelectronic counterpart, ozone, is developed. The calculations also show that the ring state of methylene peroxide is 1 eV lower than the open form. Moreover, the ring state may reopen to give the dioxy-methane biradical. The ab initio results are combined with thermochemical data in order to analyze the stability of the Criegee intermediate as well as the possible modes of reaction in ozonolysis. With regard to ozonolysis in solution, the mechanism for epoxide formation is elucidated and the possible role of methylene peroxide rearrangement to dioxymethane is considered in interpreting the ^(13)O isotope experiments. With regard to ozonolysis in the gas phase, the production of many of the chemiluminescent species observed by Pitts and co-workers is explained. The production of reactive radicals such as OH and HO_2 in the course of ozonolysis, which may have important consequences for understanding the generation of photochemical air pollution, is also delineated. Part 2: The Electronic Structure of Pyrazine: A Valence Bond Model for Lone Pair Interaction. A valence bond (VB) model is developed to describe the interaction of the lone pair excitations in pyrazine. Extensive ab initio minimal basis set (MBS) configuration interaction (CI) calculations show that the description of the n cations and nπ* states of pyrazine afforded by the VB model is more accurate than that afforded by the molecular orbital (MO) model proffered by Hoffmann. The VB picture of the n cations and nπ* states involves the interaction (resonance) of two equivalent, localized excitations. The resultant splitting is large (1 to 2 eV) because of a light delocalization of the n orbitals induced by the Pauli principle. (The n orbitals are still 90% localized on the nitrogens.) The splitting of the nπ* states is comparable to that of the n cations because the π* orbital is delocalized, even though the excitation process is localized on one nitrogen. The MBS CI calculations indicate that the lowest ionization potential of pyrazine corresponds to the ^2A_g (n) state. Calculations on the lowest Rydberg states indicate that they involve excitations out of an n orbital rather than a π orbital, in opposition to earlier spectroscopic assignments. Finally, the calculations show that the forbidden 1 ^1B_2g (nπ*) states is 1 eV higher than the allowed 1 ^1B_3u (nπ*) state, so that the perturbations observed in the absorption spectrum must be ascribed to another source. Part 3: Comparison of INDO and Ab Initio Methods for the Correlated Wavefunctions of the Ground and Excited States of Ozone. The validity of using integral approximation schemes in conjunction with correlated wavefunctions has been tested by performing generalized valence bond (GVB) and extensive configuration interaction (CI) calculations with INDO approximate integrals on the ground and excited states of ozone. High quality ab initio calculations have previously shown correlation effects to be extremely important for describing ozone. We find that for the CI wavefunctions the INDO approximation leads to vertical excitation energies- within about 30% [from 0.8 eV too low to 0. 6 eV too high with an RMS error of 0.5 eV], as compared with comparable ab initio calculations. We also found that the INDO GVB wavefunctions lead to bond angles in good agreement with experimental and ab initio calculations, but produced bond lengths that were too short. Most important was the discovery that INDO grossly favors closed geometries as opposed to open geometries, predicting the ground state of ozone to be an equilateral triangle state (even for correlated wavefunctions) with an energy 6 eV below the correct open state! Part 4: Comparison of INDO and Ab initio Methods for Correlated Wavefunctions of the Ground and Excited States of Methylene and Ethylene. The usefulness of the INDO integral approximation for correlated wavefunctions was tested by carrying out GVB calculations on (1) the three lowest states of methylene as a function of bond angle and (2) the three lowest states of ethylene as a function of the dihedral twist angle. The methylene potential curves obtained with INDO were in good agreement (0.2 eV errors) with ab initio results, while the ethylene curves were very poor (2 to 4 eV errors). Comparison with ab initio calculations revealed two major problems in the INDO method (1) the use of empirical values from atomic spectra for the one-center exchange integrals and (2) the use of only one resonance or β parameter per atom. Part 5: Approximate Integral Methods and Correlated Wavefunctions Attempts to develop an approximate integral method that produces reliable results in conjunction with correlated wavefunctions are reported. Two basic lines of approach are pursued: (1) modification (or reparametrization) of INDO and (2) general investigation of truncated integral sets. The integral approximations were tested on the low-lying states of H_2, H_3, C_2, O_2, C_2H_4, O_3, and C_6H_6 using consistently correlated wavefunctions. No approximate method investigated gave satisfactory results on all the systems tested, with the O_3 and C_2 molecules presenting the greatest problems. However, good results, using approximate methods comparable in complexity with INDO, were obtained for describing the potential curves of the low-lying states of H_2, H_3, and C_2H_4. Finally, the calculations revealed two important guidelines for approximate integral methods. (1) Transformation to orthogonal atomic orbitals (especially the one-electron integrals) is necessary prior to any truncation of the integral set. Truncation of the standard nonorthogonal integral set leads to poor results, biasing the calculations toward short bond lengths and closed geometries, (2) Replacement of core electrons with local potentials rather than 2J-K potentials leads to better results, because the need to orthogonalize the valence orbitals to the core is obviated