Bridging Machine Learning and Glassy Dynamics Theory for Predictive Polymer Modeling

Abstract

Understanding and predicting the glassy dynamics of polymers remain fundamental challenges in soft matter physics. While the Elastically Collective Nonlinear Langevin Equation (ECNLE) theory has been successful in describing relaxation dynamics, its practical application to polymers depends on a thermal mapping to connect theory with experiment, which in turn requires detailed thermodynamic data. Such data may not be available for chemically complex or newly designed polymers. In this work, we propose a simple approach that integrates machine learning-predicted glass transition temperatures (Tg) with a simplified thermal mapping based on an effective thermal expansion coefficient to overcome these limitations. This approach can provide quantitatively accurate predictions of relaxation dynamics across a broad range of polymers. Rather than replacing the original thermal mapping, our method complements it by trading formal rigor for computational efficiency and broader applicability in high-throughput screening and materials with limited available data. Moreover, we introduce a physically motivated modification to the thermal mapping that resolves discrepancies in the description of low-Tg polymers. Our results establish a generalizable approach for predictive modeling of glassy polymer dynamics and point toward new directions for theory-guided materials discovery.