Machine Learning-Enabled Development of Accurate Force Fields for Refrigerants

Hydrofluorocarbon (HFC) refrigerants with zero ozone-depletingpotential have replaced chlorofluorocarbons and are now ubiquitous.However, some HFCs have high global warming potential, which has ledto calls by governments to phase out these HFCs. Technologies to recycleand repurpose these HFCs need to be developed. Therefore, thermophysicalproperties of HFCs are needed over a wide range of conditions. Molecularsimulations can help understand and predict the thermophysical propertiesof HFCs. The prediction capability of a molecular simulation is directlytied to the accuracy of the force field. In this work, we appliedand refined a machine learning-based workflow to optimize the Lennard-Jonesparameters of classical HFC force fields for HFC-143a (CF3CH3), HFC-134a (CH2FCF3), R-50 (CH4), R-170 (C2H6), and R-14 (CF4). Our workflow involves liquid density iterations with moleculardynamics simulations and vapor-liquid equilibrium (VLE) iterationswith Gibbs ensemble Monte Carlo simulations. Support vector machineclassifiers and Gaussian process surrogate models save months of simulationtime and can efficiently select optimal parameters from half a milliondistinct parameter sets. Excellent agreement as evidenced by low meanabsolute percent errors (MAPEs) of simulated liquid density (rangingfrom 0.3% to 3.4%), vapor density (ranging from 1.4% to 2.6%), vaporpressure (ranging from 1.3% to 2.8%), and enthalpy of vaporization(ranging from 0.5% to 2.7%) relative to experiments was obtained forthe recommended parameter set of each refrigerant. The performanceof each new parameter set was superior or similar to the best forcefield in the literature.