AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-22
Crystal-structure design by agentic AI in a language of motifs
Data-driven materials discovery interpolates more reliably than it extrapolates and seldom reveals new structure types. The authors present MatEvolve, an agentic-AI framework that designs crystals in an interpretable language of motifs: an AI agent edits a human-readable motif profile with a stated rationale, builds the crystal, and validates the most promising candidates with first-principles calculations. Built on Claude Fable 5 without fine-tuning and applied to rare-earth-lean permanent-magnet compounds, MatEvolve found novel (meta)stable crystals that are new structural prototypes and magnet compounds an order of magnitude more often than generative models under an equal validation budget.