Molecular modelling and simulation have gained increasing interest across various scientific communities. In the field of batteries, the accurate description of electrolytes and solvents is crucial for overall performance. Based on molecular simulations, thermophysical properties - such as densities, pressures, phase equilibria, and transport properties - can be predicted. Simulation and modelling are closely connected: reliable simulation methods and accurate molecular models are both essential for predicting thermophysical properties.
In the past, molecular models were adjusted to match selected thermophysical properties (e.g., bulk densities, phase equilibria, or self-diffusion coefficients). The process of model adjustment often depended on the “taste” or subjective judgement of the model developer. During this procedure, deviations between predicted and experimental properties had to be balanced. In recent years, this process has been improved by applying multi-criteria optimization based on the Pareto approach. This method does not yield a single molecular model for a given species but instead produces a set of models that describe the species with minimal overall deviation. The “end user” can then decide which properties are most relevant for their application and select models with the lowest deviation in those respective properties.
Within the BatCAT project, we combined the approach of multi-criteria optimization with machine learning. As powerful as multi-criteria optimization is, it still relies on experimental data, which is not always available for all fluids of potential interest. All fluids are represented in terms of their molecular structure, and we build a toolbox capable of generating molecular models based solely on this information.