Developing Efficient and Accurate Polarizable Molecular Simulation Models for Studying Refrigerants and Their Binary Mixtures.

Hydrofluoroolefins (HFOs) are promising low-global-warming-potential refrigerants, but the scarcity of mixture data limits reliable prediction of their phase behavior and thermodynamic properties under practical operating conditions. Here, we develop a data-efficient polarizable molecular modeling strategy for HFO-1234yf, HFO-1234ze(E), and their binary mixtures by combining high-level quantum-mechanical potential-energy data with a limited set of condensed-phase experimental properties, including liquid density, heat capacity, and enthalpy of vaporization. The resulting force fields reproduce key thermodynamic properties of the pure fluids, including isobaric heat capacities and enthalpies of vaporization, over broad temperature ranges, with most deviations within 5% of experimental values. The models also accurately describe vapor-liquid equilibrium behavior across wide temperature and composition ranges, predicting mixture saturation pressures with a mean absolute error of 0.43 bar and surface tensions of the pure fluids and equimolar blend with a mean absolute error of 1.36 mN/m. Structural, energetic, and free-energy analyses reveal that fluorinated-group nonbonded interactions govern volatility and nonideal mixing, leading to preferential enrichment of HFO-1234yf in the vapor phase. Overall, this work demonstrates that polarizable force fields refined using high-level quantum data and minimal experimental input can provide accurate, transferable predictions for refrigerant blends in data-sparse regimes, offering a practical framework for molecular design and screening of next-generation low-GWP refrigerants.