Use machine learning to link atomic structure with glass properties and behaviors

Bu Wang, Dane Moran, Izabela Szlufarska, University of Wisconsin-Madison

Materials Research Science and Engineering Centers, DMR-1720415

Glasses have disordered arrangements of atomswithout the repeating patterns that crystals have.However, there are small-scale patterns of atomsthat touch each other which strongly affect theenergy of the glass, how the atoms move whenthey get hot, and other properties like strengthand response to an electric field. Unfortunately,there are many possible patterns and many slightvariations of each one, so studying them is likesorting the grains of sand on a beach by size andcolor by hand–it’s an impossible task.Wisconsin MRSEC IRG 1 uses machine learningto sort the sand. They have developed algorithmsto find small-scale atomic patterns in largesimulations of glasses and link them to the glass’energy. Ongoing studies have connectedpatterns to atomic motions, which provides a pathto simulations of glasses over long times and lowtemperatures that are currently impossible.

IRG 1, 2020

Yu, Z., Liu, Q., Szlufarska, I., Wang, B., “Structural Signatures for Thermodynamic Stability in Vitreous Silica: Insight from Machine Learning and Molecular Dynamics Simulations.” Phys. Rev. Materials 2021, 5 (1), 015602. https://doi.org/10.1103/PhysRevMaterials.5.015602.