Development and applications of traditional and non-traditional modeling methods for transition metal chemistry

This dissertation describes four projects, in which different techniques (traditional and non-traditional) were developed and applied to different problems that relate to transition metal systems.; The first project (Chapter 3) utilizes the newly developed semi-empirical quantum mechanics (SEQM) method—PM3(tm)—a traditional method. This project reports the first systematic evaluation of the accuracy of PM3(tm) by comparison of 100 calculated TM complexes with 1200 experimental structures. We found that in general PM3(tm)—predicted geometries agreed very well with experimental data; the average absolute difference (Δ%) in bond lengths versus experiment for all tested complexes is only 3%.; The second project (Chapter 4) combines traditional and non-traditional methods, using genetic algorithms (GAs) to develop parameters for technetium (Tc), whose parameters were not available in PM3(tm). It is, to our knowledge, the first report on the methodology of parameter development for PM3(tm). The GA methodology is not only fast and accurate but also flexible enough to be extended to other parameter development areas, e.g., molecular mechanics.; The third project (Chapter 5) employs different artificial intelligence (AI) techniques (including fuzzy logic and neural networks) to classify intramolecular interactions between transition metals (M) and β-X substituents in the following structural motif (L_nMC_α(A₁)(A ₂)-C_β(B₁)(B₂)X). These interactions are relevant to the direct polymerization of functionalized olefins by Ziegler-Natta (ZN) catalysis. The efficiency and effectiveness of different soft computing techniques are compared. These AI methods not only give encouraging results with respect to general data mining issues, but also insight into the factors that effect interactions between transition metals and β-X substituents.; The fourth project (Chapter 6) combines artificial neural networks (ANNs) and genetic algorithms (GAs) to design an optimal catalyst for propane ammoxidation. The mole percentages of six components of a catalyst (P, K, Cr, Mo, Al ₂O₃/SiO₂, and VSb₅WSn) were used as inputs and the activity and the acrylonitrile selectivity served as the two outputs. Optimal linear combinations (OLC) of different ANNs greatly improves the simulation of the catalyst system versus a simple, single-network architecture. This trained OLC network is used to evaluate the yield of new catalyst compositions generated during GA optimization. The best yield of acrylonitrile found after GA optimization is 79%, which is higher than the previously reported highest yield (64%). (Abstract shortened by UMI.)...