Database optimization algorithm for empirical potentials

The development for accurate and efficient empirical potential models requires years of efforts and is highly intuitional. This study provides an automated, quantitative algorithm to find the optimal empirical potential model for a pre-determined testing set of desired structure properties. We employ Bayesian sampling technique to estimate the errors for the structural property functions in the testing set. We provide the first analytical derivations of how modifications in the fitting database affect the testing set errors. A new binary modified embedded-atom method functional form is developed for Ti-O interactions where O is in the dilute limit. The optimal Ti-O potential are tested against a variety of structure properties to verify the transferability of the potential. We propose and optimize two types of objective functions which measures the transferability in the testing set. One aims to minimize the relative errors of different fitting databases for the testing set, and the other uses the logistic function in classification regression analysis to categorize the prediction errors in the testing set into good and bad ones. We develop a parallelized genetic algorithm to efficiently evaluate the objective function and perform global search for the optimal empirical potential model.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Materials Science & Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2016

Author dc:creator

• Zhang, Pinchao

Contributors dc:contributor

• Trinkle, Dallas R.
• Bellon, Pascal
• Chen, Yuguo
• Ferguson, Andrew

• Empirical potential models
• Optimization algorithm
• Bayesian statistics
• Monte Carlo
• Genetic algorithm
• Titanium oxygen interaction

Statement dc:rights

• Copyright 2016 Pinchao Zhang

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/90714

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/90714