[2602.13312] PeroMAS: A Multi-agent System of Perovskite Materials Discovery

View a PDF of the paper titled PeroMAS: A Multi-agent System of Perovskite Materials Discovery, by Yishu Wang and 10 different authors

View PDF
HTML (experimental)

Summary:As a pioneer of the third-generation photovoltaic revolution, Perovskite Photo voltaic Cells (PSCs) are famend for his or her superior optoelectronic efficiency and value potential. The event means of PSCs is exact and sophisticated, involving a collection of closed-loop workflows corresponding to literature retrieval, knowledge integration, experimental design, and synthesis. Nonetheless, current AI perovskite approaches focus predominantly on discrete fashions, together with materials design, course of optimization,and property prediction. These fashions fail to propagate bodily constraints throughout the workflow, hindering end-to-end optimization. On this paper, we suggest a multi-agent system for perovskite materials discovery, named PeroMAS. We first encapsulated a collection of perovskite-specific instruments into Mannequin Context Protocols (MCPs). By planning and invoking these instruments, PeroMAS can design perovskite supplies beneath multi-objective constraints, protecting the complete course of from literature retrieval and knowledge extraction to property prediction and mechanism evaluation. Moreover, we assemble an analysis benchmark by perovskite human consultants to evaluate this multi-agent system. Outcomes reveal that, in comparison with single Massive Language Mannequin (LLM) or conventional search methods, our system considerably enhances discovery effectivity. It efficiently recognized candidate supplies satisfying multi-objective constraints. Notably, we confirm PeroMAS’s effectiveness within the bodily world by actual synthesis experiments.