Deep eutectic solvents (DESs) are mixtures of two compounds characterized by a melting point significantly below the predicted ideal eutectic melting point. Among the most used components for the preparation of DESs is choline chloride (ChCl), which is produced on the megaton scale and has applications as an animal food supplement. ChCl can form eutectic mixtures with a wide range of organic compounds close to room temperature.
We have employed a machine learning force field (MLFF) approach developed by Zhang et al. to study reline, a mixture of ChCl and urea in a 1:2 ratio. We found that a single molecular dynamics (MD) simulation trajectory is sufficient to reproduce the liquid structure compared to a first-principles MD reference. However, we recommend running at least five trajectories to investigate dynamical properties like diffusion coefficients, ionic conductivity, or hydrogen bond lifetimes, since these values slightly scatter around a mean value. Furthermore, the Arrhenius equation can be employed to determine energy barriers for molecular processes if dynamical properties are determined at several temperatures. In summary, the computational cost of MD simulations using our trained MLFF, including the generation of training datasets and model training, is considerably less expensive than a single first-principles MD simulation run. Additionally, large-size and long time-scale MD simulations are feasible with first-principles accuracy.