A Machine-Learning Approach to Simulate Confined Liquid Lithium Polysulfides for Battery Applications

Lithium-sulfur batteries have great potential to be a high-capacity and environmentally friendly alternative to existing lithium-ion batteries but are heavily limited by their inability to retain charge after cycling. There is a polysulfide shuttle effect where the battery loses effective mass during cycling due to transport of liquid polysulfides resulting in non-conducting Li₂S gradually coating the cathode during use. It has also been shown that embedding a metal-organic framework (MOF), a porous material, is able to dramatically improve the battery’s cycling capabilities by reducing liquid polysulfide diffusion into the electrolyte. However, the underlying mechanisms are not yet well understood. Here, we investigate how lithium and sulfur interact as a confined liquid. Our simulations use ORCA, a quantum chemistry software, and the extended Tight Binding (xTB) semi-empirical algorithm. We also use Parinello meta-dynamics to ensure sampling of a wider area in chemical configuration space. The energy, force, and coordinate data generated from these configuration searches are then used to train a machine learning model using DeePMD. This model is then used to simulate a lithium-sulfur-MOF system with an order of magnitude more atoms than can be conventionally done using strictly ab-initio techniques. We also intend to integrate other compounds, such as solvents, and battery anodes and cathodes, into our simulations. We hope to understand why lithium and sulfur atoms interact differently in the presence of a MOF, what features of these MOFs best suit our needs, and how may we best utilize them.