The rapid advancement of photorechargeable batteries is driven by the need for efficient solar energy utilization, with photoassisted lithium-sulfur batteries (PALSBs) emerging as strong contenders for next-generation high-energy-density systems due to excellent charge-discharge performance. The synergy of photo- and electrocatalysis suppresses the shuttle effect and accelerates the 16-electron sulfur reduction reaction, making reaction pathway identification a central challenge in photoelectrocatalysis. As reaction networks grow increasingly complex, conventional density functional theory faces computational constraints, struggling to capture key dynamic effects, such as catalytic surface reconstruction and excited-state dynamics influenced by photogenerated electrons, which critically shape polysulfide intermediate transformation pathways. To address these challenges, we introduce a multiscale approach─combining graph theory, machine learning, and non-adiabatic molecular dynamics to systematically uncover the photoelectrocatalytic mechanisms of PALSBs. A coupled thermodynamics and excited-state dynamics framework is established to quantitatively reveal how intermediate accumulation governs the rate-determining steps (RDS). The developed universal methodology efficiently constructs graph reaction networks and extracted reaction pathways at any stage. The bag-of-words model encodes the accumulation of lithium polysulfides within reaction pathways as feature vectors for the Transformer model. The approach highlights the impact of intermediate accumulation on the RDS and can be used with other complex reaction networks. Notably, it enables catalytic pathway selection based on light-induced carrier migration and identifies Li2S2 as the key intermediate that facilitates the fastest 16-electron reaction within PALSBs. The study provides theoretical support for optimizing reaction dynamics in PALSBs, elucidates photoelectrocatalysis synergy, and offers innovative strategies for kinetic control in high-efficiency energy conversion systems.
Meng et al. (Fri,) studied this question.