Deep Learning Potential-Based Molecular Dynamics Simulation of Polyethylene Pyrolysis

A fundamental understanding of polyethylene (PE) pyrolysis is crucial for the recycling of plastic solid wastes. However, quantum mechanics (QM) methods are restricted to small-scale systems, while reactive molecular dynamics parametrized using DFT pathways lacks the required accuracy. To bridge this gap, we developed a deep learning potential model (DP model) with the accuracy of the QM method and the efficiency of the reactive force field to study the pyrolysis products and mechanism of PE. The accuracy of the DP model was verified by benchmarking the model’s performance on the data set, while the transferability was validated by kinetic analysis of the pyrolysis of PE with a larger degree of polymerization. In the kinetic analysis, the single-step behavior of the mass loss curve and activation energy (284.0–340.8 kJ/mol) are in agreement with experimental results. Based on the evolution of products in PE pyrolysis, the pyrolysis of PE can generally be divided into three stages: structural activation, initial pyrolysis, and deep pyrolysis. Crucially, the DP model accurately captures the observed product evolutions. As the reaction progresses, however, gas represented by ethylene becomes the main pyrolysis product in the deep pyrolysis stage. We further elucidate the dominant reaction mechanisms, including C–C random scission, β-scission leading to ethylene formation, and intramolecular hydrogen transfer governing the selectivity of α-olefins. Furthermore, we analyzed C–C random scission and β-scission, both of which jointly promote the fragmentation of PE chains and dictate the relative product distribution, particularly during the initial stages of pyrolysis. Overall, the present work provides atomic-level insights into the pyrolysis of PE and indicates that machine learning force fields could serve as a useful and reasonably accurate tool for studying and guiding the development of plastic waste upcycling processes.