Date: Monday, October 10, 2016
Time: 11:00 am
Speaker: Paul Voyles, University of Wisconsin-Madison
Title: Solving Structurally Complex Materials Using Methods from Image Science and Optimization
Location: 67-3111 Chemla Room
Crystallography offers tremendously powerful approaches based on diffraction for solving structures with translation symmetry and a limited number of degrees of freedom. Nanostructures with a large fraction of surface atoms and glasses have much higher structural complexity – up to 3N degrees of freedom for a glass consisting of N atoms – and require different approaches, often including experimental data from microscopy. This seminar will discuss our recent efforts to adapt tools from data science, including image science and optimization, to solve the structure of complex materials. Image science has provided tools to improve the quality of microscopy data, including non-rigid image registration, which has resulted in the first sub-picometer precision high-resolution STEM images  and non-local PCA to remove noise from low-count hyperspectral data . Optimization provides a method to create structural models incorporating all the available information about a structure, including experimental data and the simulated total system energy. We have used genetic algorithm optimization to model the structure of nanoparticles, incorporating high-quality STEM image data , and hybrid reverse Monte Carlo optimization to model the structure of metallic glasses, incorporating fluctuation electron microscopy data . The resulting structural models can be interrogated to develop abstract structural characteristics of the material or used as the starting point for additional simulations to uncover structure-property relationships.