Exploring the relationship between excess entropy and machine learned softness in glass-forming systems

Characterizing the structure of disordered systems and connecting it to dynamic properties is a major challenge. A vast number of observables have accumulated in the toolkits of those studying glassy and supercooled systems. One of these familiar quantities is excess entropy, which has long been known to correlate well with the transport-coefficients of liquids at equilibrium. Additionally in recent years, data driven methods have profoundly improved our understanding of the structure-dynamics relationship in glassy system. Notable among these examples is the concept of softness, a machine learning model associated with the particles’ susceptability to rearrange based solely upon its static structure. Are these two quantities, softness and excess entropy, correlated? To answer this question, we analyze standard Lennard-Jones style glass formers in the supercooled regime. We find a strong relationship between the excess entropy calculated both locally and globally and the empirical entropic barriers to rearrangement extracted from softness.