Minisymposium Presentation

Amorphous materials, ubiquitous among natural and manufactured solids, exhibit topological disorder and increased surface entropy. Atomic vacancies, in the form of undercoordinated defects, profoundly alter fluid/surface interactions and can be exploited to design sorbents for direct air capture (DAC) of carbon dioxide and sustainable catalysts.Standard atomistic modeling and analysis techniques, however, are not tailored to amorphous solids, and new tools and approaches need to be devised to characterize surface heterogeneity. Machine learning (ML) provides powerful methods to address the complexity of amorphous structures, enabling the identification of intricate surface regions and relevant adsorption sites.We apply segmentation techniques to identify adsorptive structures with complex geometries that cannot be effectively analyzed using classical radial distribution functions. Additionally, we employ local atomic-environment descriptors to classify surface structures and develop proxies and surrogate models to accelerate the assessment and design of materials for energy and environmental applications.