Ready-to-Use Polymerization Simulations Combining Universal Machine Learning Interatomic Potential with Time-Dependent Bond Boosting for Polymer and Interface Design

Abstract

Although polymerization and curing reactions govern the performance of advanced materials, their simulation remains challenging owing to the need for accurate, transferable potentials and rarity of chemical events. Conventional reactive force fields such as ReaxFF require system-specific parametrization, while universal machine learning interatomic potentials (uMLIPs) exhibit limited sampling efficiency. This paper introduces a novel simulation framework integrating a uMLIP with a time-dependent bond-boost scheme. The bias potential increases monotonically with time, and the use of a unified parameter set across reaction classes enables consistent acceleration without system-specific tuning. For radical polymerization of vinyl monomers, the proposed framework reproduces characteristic trends, such as linear molecular-weight growth with conversion, initiator-concentration scaling, and relative monomer reactivity trends. For step-growth polycondensation of nylon-6,6, it captures the characteristic sharp increase in molecular weight at high conversion rates, consistent with experimental behavior. For epoxy curing at a copper substrate, it reveals interfacial ring-opening and cross-linking events, consistent with spectroscopic evidence of Cu-O-C bond formation. Overall, coupling uMLIPs with time-dependent bond boost enables practical and transferable simulations of polymerization and curing processes. The proposed framework reliably resolves mechanistic pathways and relative reactivity, offering molecular-level insights into polymer growth and interfacial adhesion.