Graduation Year
2014
Document Type
Dissertation
Degree
Ph.D.
Degree Name
Doctor of Philosophy (Ph.D.)
Department
Physics
Degree Granting Department
Physics
Major Professor
Sagar A. Pandit, Ph.D.
Committee Member
Ananth Y. Grama, Ph.D.
Committee Member
Martin Muschol, Ph.D.
Committee Member
Jianjun Pan, Ph.D.
Committee Member
Lilia M. Woods, Ph.D.
Keywords
Molecular Modeling, Optimization, Simulation
Abstract
Molecular models allow computer simulations to predict the microscopic properties of macroscopic systems. Molecular modeling can also provide a fully understood test system for the application of theoretical methods. The power of a model lies in the accuracy of the parameter values which govern its mathematical behavior. In this work, a new software, called ParOpt, for general high dimensional non-linear optimization will be presented. The software provides a very general framework for the optimization of a wide variety of parameter sets. The software is especially powerful when applied to the difficult task of molecular model parameter optimization. Three applications of the ParOpt software, and the Nelder-Mead algorithm implemented within it, are presented: a coarse-grained (CG) water--ion model, a model for the determination of lipid bilayer structure via the interpretation of scattering data, and a reactive molecular dynamics (ReaxFF) model for oxygen and hydrogen. Each problem presents specific difficulties. The power and generality of the ParOpt software is illustrated by the successful optimization of such a diverse set of problems.
Scholar Commons Citation
Fogarty, Joseph C., "High Dimensional Non-Linear Optimization of Molecular Models" (2014). USF Tampa Graduate Theses and Dissertations.
https://digitalcommons.usf.edu/etd/5618