Computationally Feasible Phase Diagrams and Properties of Materials under Experimentally Challenging Conditions
CME Department Seminar
March 19, 2024
11:00 AM - 12:00 PM America/Chicago
Presenter: Keith Phuthi, PhD Candidate, University of Michigan
Location: ERF 1047
Abstract: Materials are the key to unlocking technologies that will enable a more sustainable future. It is therefore important to discover new functional materials and to characterize existing materials as accurately as possible. While this can sometimes be done experimentally, some materials are difficult to work with and experiments can rarely be performed in a high-throughput manner. Computational methods emerge as powerful tools which have exponentially grown in use due to more powerful computers, improved algorithms and use of data-based Machine Learning methods. For a material found computationally to exist requires that it be thermodynamically (meta-)stable, thus it is vitally important that there exist reliable computational methods for determining stability phase diagrams at arbitrary external conditions. In this talk, I demonstrate new methods for determining phase diagrams and material properties under conditions that are more realistic or challenging to explore experimentally such as high pressures and temperatures and have up until recently been difficult to determine reliably computationally. I consider metal hydrides and lithium metal, both classes of materials important in energy storage and with other exciting properties such as superconductivity. We present methods for determining the stability of metal hydrides under the barely explored conditions of combined pressure and electrochemical potential. We also predict the stability of different phases of lithium and tabulate entropy and free energy data at pressures for which no such experimental data exists. This work combines a number of atomic simulation tools including first-principles calculations, molecular dynamics, machine learning and statistics in novel ways to achieve these goals efficiently, accurately and reproducibly.
Speaker Bio: Keith Phuthi is a 5th year PhD student in the mechanical engineering department at the University of Michigan under the supervision of Prof. Venkat Viswanathan. His research currently focuses on building tools for efficient and accurate prediction of macroscopic material properties using atomic scale simulations and machine learning. He obtained his bachelor's degree in physics at the Massachusetts Institute of Technology where his research focused on experimental studies of extremely low temperature electron transport and particle detection as well as simulation and design of particle detectors under the supervision of Prof. Joseph Formaggio. He obtained his master's degree at Carnegie Mellon University in mechanical engineering also under the supervision of Prof. Venkat Viswanathan where he was awarded Google Collab and Oracle Research awards to work on the development of Machine Learning Interatomic Potentials. He has taught classes on topics in thermodynamics, machine learning and introductory engineering and is always active in teaching for K-12 outreach programs.
Date posted
Mar 18, 2024
Date updated
Mar 18, 2024