ChemGraph-XANES Framework Automates XANES Simulation Using LLM Agents
A novel framework known as ChemGraph-XANES has been introduced to streamline the simulation and analysis of X-ray absorption near-edge structure. This system overcomes the complexities of workflows that have previously hindered large-scale computational XANES applications. It integrates various technical elements, such as ASE, FDMNES, Parsl, and a tool interface based on LangGraph/LangChain, presenting XANES operations as typed Python tools. These tools can be managed by agents utilizing large language models, with a retrieval-augmented expert agent referencing the FDMNES manual for parameter selection in a multi-agent setup. The framework consolidates natural-language task definitions, structure acquisition, FDMNES input creation, parallel task execution, spectral normalization, and data curation aware of provenance. Computational XANES is extensively employed to investigate local coordination environments, oxidation states, and electronic structures in complex chemical systems. This framework was detailed in arXiv preprint 2604.16205v1, marked by a cross announcement type.
Key facts
- ChemGraph-XANES is an agentic framework for automated XANES simulation and analysis
- The framework addresses workflow complexity constraints limiting computational XANES use at scale
- Built on ASE, FDMNES, Parsl, and LangGraph/LangChain-based tool interface
- Exposes XANES workflow operations as typed Python tools orchestrated by LLM agents
- Includes retrieval-augmented expert agent that consults FDMNES manual for parameter selection
- Unifies natural-language task specification, structure acquisition, and FDMNES input generation
- Computational XANES probes local coordination environments, oxidation states, and electronic structure
- Announced in arXiv preprint 2604.16205v1 with cross announcement type
Entities
—