Reasoning-to-Simulation: An Agentic Framework for Discovery of Electrolyte Materials

Jurģis Ruža, Rafael Gomez-Bombarelli

Keywords: LLM, large language models, agentic AI, materials discovery, materials optimization, agents, battery materials

TL;DR: We propose an agentic framework that leverages Large Language Models (LLMs) as autonomous orchestrators for closed-loop materials discovery.

Abstract: The discovery of novel metal-ion battery electrolytes is hindered by the vast combinatorial space of solvent-salt formulations and the complexity of physics-based validation. We propose an agentic framework that leverages Large Language Models (LLMs) as autonomous orchestrators for closed-loop materials discovery. Our system features a Discovery Agent that performs multi-generational composition optimization and a Simulation Assistant that translates natural language prompts into executable liquid electrolyte molecular dynamics simulations. By integrating generative reasoning with automated simulation verification, our framework enables an autonomous "reasoning-to-simulation" cycle that identifies high-performance electrolytes. This work demonstrates the ability of agentic LLMs to bridge the gap between chemical intuition and rigorous computational materials science by discovering a state-of-the-art liquid electrolyte solvent formulation with 67 mS/cm ionic conductivity.

Submission Track: Full Paper

Submission Category: AI-Guided Design

Submission Number: 37