| This dissertation reports what is, to our knowledge, the first direct comparison of neural networks with quantum mechanical techniques for the prediction of molecular properties for inorganic systems. The molecular properties investigated include prototypical structural (equilibrium bond length, r{dollar}sb{lcub}rm e{rcub}{dollar}), vibrational (equilibrium stretching frequency, v{dollar}sb{lcub}rm e{rcub}{dollar}), and energetic (bond dissociation energy, D{dollar}sb{lcub}rm e{rcub}{dollar}) properties for diatomics. Bond dissociation energies (BDEs) are extremely important in chemistry. However, they are difficult to calculate accurately using quantum mechanical (QM) techniques. Therefore, an alternative that gives similar accuracy to correlated QM techniques, but utilizes reduced computer time and resources, would be very useful. Neural Networks and QM methods are compared for EO diatomics for the prediction of BDE. Neural network, multiple linear regression, and multiconfiguration self consistent field calculations are reported for a diverse set of chemical problems (e.g., correction of harmonic vibration frequencies, prediction of relaxivity, prediction of Drago's E and C parameters from HOMO and LUMO energies, prediction of {dollar}Delta{dollar}H{dollar}sb{lcub}rm act{rcub}{dollar} for methane activation, and prediction of classical or nonclassical H{dollar}sb2{dollar} coordination). |
