Machine Learning Guided Enhanced Sampling for Catalyzed Plastic Degradation

Chemical recycling of polymers is a promising method for robust and selective cleavage of polymer waste to upcycled products such as liquid fuel. The hydrogenolysis reaction has favorable cleaving pathways for upcycling, but the underlying mechanisms of the polymer cleavage are unknown. Computational methods such as Density Functional Theory can reveal electronic structure, kinetic data, and thermodynamic criteria for optimal carbon-carbon cleavage. This method, though, lacks thermal stability and entropic contributions of larger polymer chains. To overcome this, we demonstrate a computational framework based on machine learning-based enhanced sampling and ab initio molecular dynamics (AIMD) to efficiently simulate polymer cleavage via hydrogenolysis. From AIMD simulations of dodecane on the Ru(0001) surface, we observe a dependence on chain adsorption position and conformation before carbon cleavage, indicating a strong relationship between chain flexibility, polymer-surface binding motifs, and cleaving rate. From machine learning-based enhanced sampling simulations, we compute the free energy surface of dodecane hydrogenolysis on Ru(0001), revealing the free energetic landscape of polyethylene decomposition for the first time. Insights gained from these calculations reveal a more comprehensive study of the mechanisms behind polymer cleavage and recycling.