Machine Learning-Driven Neuroevolution Potentials for Imidazole Ionic Liquids: Capturing Ionic Correlations and Structure–Property Relationships

Ionic liquids (ILs) possess exceptional physicochemical properties, such as negligible vapor pressure, wide electrochemical stability windows, and tunable molecular architectures, making them attractive candidates for advanced electrolyte applications. These properties arise from intricate ionic correlations and variable ionicity between cations and anions, which govern thermal stability, electrical conductivity, and other key behaviors. In this work, we employ the neuroevolution potential (NEP) framework to develop machine learning-based force fields for a series of imidazole ILs with four distinct anions. The NEP models are rigorously validated against density functional theory (DFT) calculations, showing excellent agreement in energies and forces, and outperforming classical force fields in reproducing structural properties. Molecular dynamics simulations based on the NEP accurately predicted physical properties, including density, self-diffusion coefficients, and viscosity, in close agreement with experimental measurements. Importantly, the ML-driven NEP framework enables simulations of much larger system sizes and longer time scales than are feasible with DFT, while achieving significant speedups over classical force fields. Analysis of ionicity revealed values near 1/3, consistent with previous studies, and provided molecular-level insights into the interplay between ionicity and structural features. This study highlights the potential of machine learning-driven force fields for efficient, accurate, and scalable modeling of complex IL systems, providing practical guidance for designing IL-based electrolytes with tailored properties.