(169ai) A Synergy of Molecular Simulation, Mathematical Programming, and Machine Learning for the Phaseout of Harmful Refrigerants

In 2016, 197 countries signed the Kigali agreement to phase out refrigerant fluids with high global warming potential (GWP) that can cause adverse climate change effects. These refrigerant fluids are often azeotropic or near-azeotropic mixtures of two or more hydrofluorocarbons (HFCs). One of the major technical hurdles that must be surmounted to enable phaseout requirements is the separation of these high GWP refrigerant mixtures, for reuse and recycling, to prevent them from being vented or incinerated. The other critical need is discovering and designing new, greener refrigerant fluids. We have applied several computational molecular science tools synergistically to help address these challenges.

Addressing the first challenge requires the design of advanced separation technologies. Extractive distillation using ionic liquids (ILs) as entrainers has been proposed as a viable solution. However, there is limited understanding of IL/HFC systems, including transport and interfacial data. We applied molecular dynamics (MD) simulations using several advanced methods and techniques for reliable studies of thermal conductivity, viscosity, self-diffusivity, and interfacial tension for several IL/HFC systems. Our results were able to excellently capture experimentally observed qualitative trends for these properties for all IL/HFC systems studied. We also investigated and successfully elucidated the molecular origins of some of these properties and the link between the molecular liquid structure and these properties. Our data and insights from MD simulations will support secondary-level screening of ILs to select or design the ‘optimal’ IL for use in separating HFCs.

To address the second challenge, we applied mathematical programming in a molecule generation tool we developed called FineSMILES to exhaustively generate all or almost all acyclic molecules that could serve as refrigerants in vapor-compression refrigeration cycles, based on thermodynamic arguments. We generated hundreds of thousands of molecules, more than 50 % of which were not reported in the PubChem database. We then applied machine learning for property prediction to rapidly screen these generated molecules based on technical, safety, and environmental performance considerations. We used the retrosynthetic accessibility scores for the preliminary assessment of synthesizability. We obtained several tens of computationally discovered molecules that have high potential for use as refrigerants and may be likely synthesizable. Finally, we further applied high throughput molecular simulations to assess the technical performance of the discovered fluids and to design and assess azeotropic or near-azeotropic green refrigerant mixtures using the discovered molecules as components. Five novel, high-performing green refrigerant mixtures were designed and recommended for further experimental testing.