123 papers
This paper introduces GET-SEI, a general framework combining graph contrastive learning, extended dynamic mode decomposition, and transition path theory to automatically characterize local atomic environments and quantify lithium transport kinetics and pathways across diverse solid-state electrolyte/lithium metal interfaces for targeted SEI engineering.
This study utilizes machine learning to reveal that charged domain walls in two-dimensional bismuth possess lower energy than uncharged ones and host topological interfacial states with accidental band crossings at the Fermi level, highlighting their potential for next-generation ferroelectric devices.
The paper introduces ChemNavigator, an agentic AI system that autonomously integrates large language model reasoning with computational chemistry to discover six statistically significant, chemically grounded design rules for organic photocatalysts, outperforming previous machine learning approaches by independently deriving interpretable structure-property relationships through a hypothesis-driven scientific workflow.