Learning-based approach to plasticity in athermal sheared amorphous packings: Improving softness

The plasticity of amorphous solids undergoing shear is characterized by quasi-localized rearrangements particles. While many models exist, the precise relationship between plastic dynamics and structure a particle’s local environment remains an open question. Previously, machine learning was used to identify structural predictor called “softness.” Although softness has been shown predict which particles will rearrange with high accuracy, method can be difficult implement in experiments where data are limited combinations descriptors it identifies often interpret physically. Here, we address both these weaknesses, presenting two major improvements standard method. First, present natural representation each observed mobility, allowing for use statistical that simpler provide greater accuracy datasets. Second, employ persistent homology as systematic means identifying simple, topologically informed, quantities easy measure experimentally. We test our methods on two-dimensional athermal packings soft spheres under quasi-static shear. find same information predicts small variations response also predictive events localize. excellent achieved sheared using simply species number nearest neighbor contacts.