AdsMind: A Physics-Grounded Multi-Agent System for Self-Correcting Discovery of Adsorption Configurations on Heterogeneous Catalyst Surfaces
Researchers have developed AdsMind, a novel multi-agent system designed to accelerate the discovery of adsorption configurations on heterogeneous catalyst surfaces. This closed-loop framework integrates machine learning force fields (MLFFs) with large language models (LLMs) to enable autonomous error correction and improve search reliability. AdsMind significantly reduces the number of MLFF relaxations required compared to heuristic methods, achieving high success rates and providing more accurate results than open-loop LLM agents, thereby supporting more efficient autonomous chemistry workflows. AI
IMPACT This system could significantly speed up materials science research by automating complex configuration discovery processes.