High Dimensional Non-Linear Optimization of Molecular Models

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