LEAP: A closed-loop framework for perovskite precursor additive discovery

In three linesLEAP couples a domain-specialized LLM with active learning to discover perovskite precursor additives for solar cell optimization. The model extracts mechanistic knowledge from literature and generates interpretable descriptors integrated into Bayesian optimization. Experimental validation: PCE 20.13–20.87% vs 19.25% control, champion 21.32%.

Summary generated by Claude — human-verified