Modeling polymer degradation at the intersection of molecular simulations and machine learning

Molecular simulations are a key tool to predict and understand how the molecular-level features of a macromolecule (e.g., chemistry, size, topology, composition) combine with processing approaches (e.g., heat, flow, solvents, surfaces/interfaces) to control the structure and properties of polymeric materials. Choosing the appropriate model resolution is particularly challenging for polymers, as capturing both angstrom-scale chemical detail and nm-to-micron-scale collective processes is often necessary for a complete description of a polymer's structure and properties. Particularly in the context of polymer sustainability, methods that capture chemical specificity (including possible reactivity) while reaching length scales commensurate with multichain collective behavior could be transformative. In this talk, I will discuss how molecular simulations driven by machine learning (ML)-based interaction potentials may enable the direct modeling of physio-chemical processes relevant to polymer degradation, recycling, and disposal. I will discuss our progress towards establishing practical guidelines for training ML interaction potentials for macromolecules. Then, I will present our application of ML potentials to model the bulk thermal degradation of polyolefins and their hydrogenation at a catalytic surface as representative model systems.