Entries by Academic Web Pages

Machine-learning for finding the optimal configuration of configurationally complex materials

The increasing demand for innovative designs often outpaces the rate at which new designs are developed, in part, due to the time needed to search within a vast set of alternatives, especially when evaluating alternatives requires a time-consuming experiment. Helping to accelerate the search for designs, modern sequential experimental design (SED) methods iteratively use machine learning to predict the best points to sample. This project seeks to develop new machine learning methods for optimizing configurationally complex materials, accounting for the complexity associated with conducting physical experiments.

Accelerating the Search for Material Transition States Using Machine Learning

Understanding the movement of ions through a material is crucial to understanding important properties such as radiation, stress cracking, and ion conductivity. Fundamentally, these dynamics are determined by the relevant energy barriers to ion motion. As such, in order to advance our understanding of the mechanisms of materials processes for complex systems, there is a need to determine the energy landscapes for such ionic motion. However, determining complete maps of high-dimensional landscapes is currently not computationally feasible for systems with a large number of atoms or defects, mainly because of the challenge of finding all of the energy transition states (TSs). Without all of the TSs for a system, the simulator may not be able to uncover important mechanisms that govern the dynamics of material processes.