LLM-Guided Ligand Design for Controlling Nanoscopic Morphology of CsPbBr3 Perovskite Nanocrystals

Traditional nanocrystal ligand design often relies on trial-and-error, limiting systematic exploration. Here, we present a framework integrating a Large Language Model (LLM) agent into experimental planning, iteratively refining ligand design for nanocrystal synthesis. Through structured human-AI dialogue, we generated testable hypotheses and uncovered key design principles: planar aromatics promote anisotropic growth via π-stacking, while steric hindrance and chirality influence surface coverage and introduce directional biases. This iterative human-AI approach led to selecting (R)- and (S)-1-(1-naphthyl)ethylammonium (R/S-NEA) as promising ligand candidates. High-throughput experiments on CsPbBr3 perovskite nanocrystal growth revealed striking enantiomer-specific effects. S-NEA enhanced photoluminescence and yielded larger morphologies with increasing concentrations, while R-NEA promoted anisotropic growth with increased lattice strain. Density Functional Theory provided mechanistic insight: R-NEA preferentially binds CsPbBr3 (110) surfaces, while S-NEA shows enhanced stability on (100) and (111) facets. These findings demonstrate how ligand aromaticity and stereochemistry induce facet-specific binding and shape control. This study establishes a systematic, iterative framework for AI-assisted materials design, validating LLM-predicted structure-property relationships and offering a scalable methodology for accelerating discovery across diverse material systems through continuous refinement.