Catalyst-Agent: Autonomous heterogeneous catalyst screening with an LLM Agent

2026-03-01 · Source: cs.CL updates on arXiv.org · Field: Science & Research — Artificial Intelligence & Machine Learning, Physical Sciences & Chemistry, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

Catalyst-Agent is an LLM-powered agent that autonomously coordinates closed-loop heterogeneous catalyst screening for electrochemical applications like the oxygen reduction reaction (ORR), nitrogen reduction reaction (NRR), and CO2 reduction reaction (CO2RR). This Model Context Protocol (MCP) server-based system integrates various tools, including OPTIMADE for database searches, Meta FAIRchem's UMA MLIP within AdsorbML for adsorption energy calculations, and reaction-specific descriptor evaluations. Catalyst-Agent refines near-miss candidates through structural modifications. In ORR, NRR, and CO2RR campaigns, it achieved high performance, converging in 1.40-3.41 trials per successful material on average. It identified previously unreported CO2RR candidates such as Sn3Sc, Sn3Y, Tl3La, Pb3Y, and In3Y, with DFT single-point checks confirming outcomes. Ablation studies showed its chemically informed candidate selection and feedback-directed modification significantly outperformed randomized screening, which yielded only 13.3%, 16.7%, and 0% success for ORR, NRR, and CO2RR, respectively.

Key takeaway

For research scientists focused on electrochemical catalyst discovery, Catalyst-Agent demonstrates a paradigm shift from manual trial-and-error. You should consider integrating LLM-powered agents into your screening workflows to automate candidate selection, structural modification, and performance evaluation. This approach can significantly reduce the number of trials needed for successful material identification, accelerating the discovery of novel catalysts for ORR, NRR, and CO2RR applications.

Key insights

LLM agents can autonomously coordinate complex scientific workflows, significantly accelerating catalyst discovery.

Principles

• Chemically informed selection outperforms brute-force screening.
• Feedback-directed modification refines near-miss candidates.
• Tool-grounded LLMs enable adaptive, reproducible workflows.

Method

Catalyst-Agent searches materials databases, constructs slabs, computes adsorption energies, evaluates reaction-specific descriptors, and applies structural modifications in a closed loop.

In practice

• Use LLM agents to automate multi-step material discovery.
• Integrate MLIPs for faster adsorption energy calculations.
• Apply feedback loops for iterative candidate refinement.

Topics

• LLM Agents
• Catalyst Screening
• Materials Discovery
• Machine Learning Interatomic Potentials
• Electrocatalysis
• Autonomous Workflows

Best for: AI Scientist, Research Scientist

Related on AIssential

• From Data to Theory: Autonomous Large Language Model Agents for Materials Science — Autonomous LLM agents can perform end-to-end scientific theory development, from equation selection to code execution and validation.
• AdsMind: A Physics-Grounded Multi-Agent System for Self-Correcting Discovery of Adsorption Configurations on Heterogeneous Catalyst Surfaces — AdsMind is a physics-grounded multi-agent system using MLFF feedback for self-correction in heterogeneous catalysis configuration discovery.
• ChemAmp: Amplified Chemistry Tools via Composable Agents — ChemHAS uses hierarchical agent stacking to reduce prediction errors in chemistry tools, outperforming existing LLM and specialist models.
• Battery-Sim-Agent: Leveraging LLM-Agent for Inverse Battery Parameter Estimation — LLM agents can reframe inverse scientific problems as reasoning tasks, outperforming traditional black-box optimization.
• PolyJarvis: An LLM-Orchestrated Agent for Automated All-Atom Molecular Dynamics of Amorphous Homopolymers — LLM agents can autonomously execute complex scientific workflows, adapting protocols and recovering from errors.
• Benchmarking LLM Agents on Meta-Analysis Articles from Nature Portfolio — LLM agents critically fail at criterion-based screening in scientific meta-analysis, despite high retrieval recall.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.