(711C) Computational evaluation of liquid mixtures using physics-based models and machine learning

Complex fluid mixtures are critical components in many industrial processes and formulations. Understanding the physical properties of such fluid mixtures is vital in optimizing the process parameters, conditions, and compositions. An interesting feature of such complex mixtures is that there are numerous combinations, however it is impractical to test a large number of candidates in lab. Predictive models are therefore beneficial to obtain insights into the behavior of liquid mixtures. Traditional mixing rules often fail to capture the mixture properties as they do not account for inter-molecular interactions. One approach to overcome this shortcoming is to use molecular dynamics (MD) simulations. Typical MD simulations consider constant partial charges of atoms, and furthermore, do not consider polarization leading to inaccurate property estimations. To capture the correct partial changes, polarization, and dynamics, we use machine-learned (ML) potentials trained using thousands of ab-initio data. The resulting MD using such ML potential is highly accurate and can be extended to highly complex fluids, including ionic liquids. In this presentation, I will demonstrate the application of ML potential for liquid mixtures of electrolytes and ions and compare it with experimental results and classical MD.