Student Inorganic Chemistry Seminar with Eric Riesel (Freedman Group)

, 4-370

A generative AI approach to crystal structure solutionfrom powder X-ray diffractionAbstract: X-ray diffraction (XRD) is a cornerstone technique in chemistry, offering an angstrom-scale window into the arrangement of atoms in a material. However, structure determination from powder X-ray diffraction (PXRD) patterns remains a significant challenge due to its lower information content than single-crystal XRD. This challenge poses a significant barrier to full structural characterization of compounds which are not amenable to single crystal growth and in situ measurements which rely on PXRD. In this talk, we introduce a pioneering generative machine learning model designed to solve crystal structures from PXRD data. Our model is the first to solve crystal structures using only the PXRD pattern and a list of elements which may be in the sample. The architecture accommodates additional input for enhanced accuracy, such as composition from ICP-OES, and offers rapid feedback on the quality of the structure solution. Our validation studies on 150 experimental PXRD patterns and thousands of simulated patterns demonstrate that our model successfully solves 45% and 60% of structures, respectively including intermetallic compounds, oxides, and compounds with molecular fragments. We proceed to solve many previously unsolved PXRD patterns, including those acquired from high-pressure experiments in our own laboratory. This model opens avenues towards materials discovery under conditions which preclude single crystal growth and towards automated materials discovery pipelines, opening the door to new domains of chemistry.Refreshments will be served outside of 4-370 at 4PM.