Data-driven classification of elementary rearrangement events in silica glass

Abstract

Reliable classification and prediction of elementary rearrangement events is a crucial step towards a profound understanding of the mechanical behavior of silica glass. Using various cyclic athermal quasistatic shear deformation protocols, we detect angle-changing and bond-breaking events in silica glass, both of which do or do not recover within a defined deformation protocol. We show in this contribution that data-driven approaches using rigorous statistical analyses and polynomial regression provide valuable insights into the mechanics of the marginally stable strain state of silica glass. We classify if a particular event recovers within a defined deformation protocol with up to 90% accuracy and predict the recovering strain with up to 95% accuracy.