Ultra-fast and accurate binding energy prediction of shuttle effect-suppressive sulfur hosts for lithium-sulfur batteries using machine learning

Abstract The shuttle effect of lithium polysulfides (LiPS) leads to fast capacity loss in lithium-sulfur batteries, which hinders the practical applications and makes the discovery of shuttle effect-suppressive sulfur host materials highly significant. Here, we proposed a machine learning (ML) method to rapidly and accurately predict the binding energies towards LiPS including Li2S4, Li2S6, and Li2S8 adsorbed on the surface of sulfur hosts with arbitrary configurations and active sites. As a case study, MoSe2 was selected as a sulfur host to predict the binding energy when absorbing the LiPS. The ML method shows six orders of magnitude faster than the conventional density functional theory (DFT), with a low predicted mean absolute error (MAE) of 0.1 eV. Based on the transfer learning (TL), we demonstrated that the presented ML method can be transferred to other layered compounds with a similar AB2 structure to MoSe2, and can efficiently predict their binding strengths with hosts. WSe2 was employed as a case to validate the TL method, with the results showing that MoSe2 had a stronger binding strength than WSe2 when adsorbing the LiPS, and only one-seventh of the ML training data was required. The impacts of different adsorption sites, configurations and distances on the binding energy were analyzed when LiPS is absorbed, which is of great significance to understand the adsorption mechanism of LiPS with hosts. The proposed work provides an efficient ML method to screen and discover new AB2 typed two-dimensional layered materials for suppressing the shuttle effect in lithium-sulfur battery.