ARIA: A Causal-Aware Framework for Rescuing LLM Reasoning in Trustworthy Materials Discovery

The article introduces ARIA, a causal-aware framework designed to enhance reasoning in LLMs for materials discovery by addressing the issue of contextual tunneling. ARIA employs a three-tier cascade approach: direct causal reasoning, physics-informed analogical transfer, and explicit parametric fallback, utilizing a Knowledge Graph with 2,839 Process-Structure-Property (PSP) relations. This framework not only improves performance on forward prediction and inverse design tasks for two-dimensional materials but also ensures auditable causal traces, making AI-assisted materials discovery more reliable and grounded in physical principles.