Improvements to the pseudospectral electronic structure method and experimental protein model initiation.

摘要

This dissertation comprises two works in the field of molecular modeling. First, it presents improvements and a significant reformulation of Friesner's pseudospectral method for Hartree-Fock electronic structure calculations into a hybrid method using both analytic and grid-based integration schemes. The improvements are applicable to other ab initio electronic structure methodologies as well. It describes the use of and improvements to a new approach to generating integrals required for the core grid-based method. Additionally, a complex reformulation relying on fast analytic methods and an efficient selection and control structure allows the inclusion of selected analytic correction terms, greatly reducing the required grid density. Absolute energies agree with conventional basis set codes to within 0.25 kcal/mole, and relative energies agree to better than 0.1 kcal/mole for a wide variety of test molecules. Accelerations of CPU times of as large as a factor of 6.5 are obtained as compared to GAUSSIAN 92, with the actual timing advantage increasing for larger basis sets and larger molecules. The method is shown to be highly reliable and capable of handling extended basis sets.;Second, it presents an investigation into a novel method of initial model building through fragment placement. Despite continuing efforts, protein structure determination from X-ray crystallography data at lower resolutions usually requires manual intervention. Placement of atoms in the partial structure at the start of model building can be particularly critical since errors can be self-reinforcing in ensuing work. The approach uses a coarse six-dimensional real-space search followed by a constrained minimization. Results are given in comparison to a tool based on an exhaustive six-dimensional search alone from a popular crystallography package using ten sets of experimental data. Placement of a standard set of fragments shows equal or often significantly improved agreement with final independently solved models.