| This thesis deals with semiempirical quantum chemistry, i.e., mathematical models based in derived equations of quantum chemistry but modified and measured for value by their ability to reproduce experimental results. Two of the branches of quantum chemistry that can be considered semiempirical are discussed in this thesis.;Traditionally, semiempirical quantum chemistry refers to the NDDO-HF methods such as MNDO, AM1, and PM3. These methods have the advantage of being extremely fast, reasonably accurate, and fairly widespread in the chemistry community. These methods are heavily parameterized, and they require 18 parameters per atom type to model the chemistry of that atom type. The first four chapters of this thesis detail the theory behind these methods, the optimization of atomic parameters using a genetic algorithm to produce a new atom type, and the results of new parameterizations of sodium, calcium, potassium and zinc for use with the standard AM1 and PM3 methods.;The other branch of quantum chemistry to be considered semiempirical in this thesis is density functional theory. All current density functional chemistry calculations require the specification of a mathematical functional to calculate the energy of exchange and correlation. These functionals are designed through theoretical considerations, but most are optimized using experimental data and evaluated by their ability to reproduce experimental data. They are therefore semiempirical. The fifth chapter of this thesis details the theoretical details of density functional theory, the form of the density functionals considered in the thesis, and the performance of the functionals in reproducing experimental heats of formation of a 372-compound data set at the basis sets 3-21G*, 3-21+G*, and MIDI!. |
