Graduate Student Seminar

Abstract: Generative AI has been widely touted as a means to discover molecules and crystals, and now routinely produces samples conditioned on target properties. Many technologically relevant materials, however, are neither single molecules nor crystals: polymers, glasses, battery electrolytes, catalysts, and many electronic devices are examples of amorphous systems whose absence of long-range order makes them difficult to characterize, simulate, and design. In this talk, I will show how generative AI, simulations, and experiments can be combined to resolve amorphous structures from characterization data and to rationalize synthesis-structure-property relationships that have resisted conventional analysis. Across a range of inorganic amorphous materials, our generative models recover structures directly from experimental measurements more accurately than classical structure reconstruction methods. Our methods also sample configurations across processing conditions, compositions, and data sources orders of magnitude faster than molecular dynamics simulations, and are particularly useful for processing conditions that are not typically simulated (e.g., as in mechanochemistry). Using this approach, we resolve structures for long-standing problems in condensed matter physics, from paracrystallinity in silicon to a liquid-liquid phase transformation in high-pressure sulfur. Similar strategies extend to organic liquids probed by infrared spectroscopy, where the models disentangle mixture composition and local coordination despite strong non-ideality. Finally, I will show how our data-driven models rationalize synthesis and experimentation, allowing control of microstructure in battery materials and catalysts. Together, these results outline a roadmap for the design and structural elucidation of amorphous matter that closes the gap between computation and experiments. Daniel Schwalbe-Koda is an Assistant Professor of Materials Science and Engineering at UCLA. He obtained