High-Throughput Screening of Gaseous Molecule Space for Developing SF₆ Alternatives

Sulfur hexafluoride (SF6) is a powerful insulating gas, yet its global warming potential (GWP) of 25,200 compels us to urgently seek eco-friendly alternatives. Various experiments and computations have been conducted to develop SF6 alternatives that fulfill requirements for dielectric breakdown strength, GWP, and boiling point. However, the efforts has been hindered by limited experimental results as well as inadequate computational performance to extrapolatively predict gas properties of unknown molecular structures. Thus, this study conducts high-throughput screening of gaseous molecule space using quantum mechanics-assisted machine learning model. This exploration has expanded the existing small-molecule space to encompass 3 × 108 molecules, a 1500-fold increase over databases, equivalent to registration efforts spanning 4.5 × 104 years. The computational results identify potential candidates that pose high breakdown strength at low-temperature and high-pressure conditions while maintaining low GWP. These findings lead to crucial insights into the next steps of computational design and experimental verification for SF6 alternatives.