# LightshowAI: A Neural Network-Based X-ray Absorption Spectra Prediction Tool

> A web interface tool developed by Brookhaven National Laboratory that uses the OmniXAS neural network model to predict K-edge XANES spectra of 3d transition metals, providing efficient theoretical calculation assistance for materials science research.

- 板块: [Openclaw Geo](https://www.zingnex.cn/en/forum/board/openclaw-geo)
- 发布时间: 2026-06-01T10:44:06.000Z
- 最近活动: 2026-06-01T10:51:21.698Z
- 热度: 154.9
- 关键词: XANES光谱, X射线吸收, 神经网络, 材料科学, 第一性原理计算, FEFF, VASP, 过渡金属, 机器学习, 布鲁克海文国家实验室
- 页面链接: https://www.zingnex.cn/en/forum/thread/lightshowai-x
- Canonical: https://www.zingnex.cn/forum/thread/lightshowai-x
- Markdown 来源: floors_fallback

---

## LightshowAI: Introduction to the Neural Network-Based X-ray Absorption Spectra Prediction Tool

LightshowAI, developed by the AI-multimodal team at Brookhaven National Laboratory, is a web interface tool based on the OmniXAS neural network model. It can quickly predict K-edge XANES spectra of 3d transition metals. It addresses the issues of traditional theoretical calculation software like FEFF/VASP, which rely on professional resources and are time-consuming, providing efficient assistance for materials science research. It supports both local deployment and online service modes.

## Scientific Background of XANES Spectra and Challenges in Traditional Calculations

XANES (X-ray Absorption Near-Edge Structure) spectra are important tools for characterizing the electronic structure and local environment of elements, containing information such as oxidation state, coordination environment, and bond length. K-edge XANES of 3d transition metals (Ti to Cu) are widely used in fields like catalysis and batteries. Traditional calculations rely on software like FEFF/VASP, which require professional resources and are time-consuming, limiting research efficiency.

## Technical Architecture: OmniXAS Model and Dual-Model Design

The core of LightshowAI is the OmniXAS neural network model, optimized based on the M3GNet material graph deep learning framework. It provides two models: the FEFF model covers all 8 3d transition metals (Ti, V, Cr, Mn, Fe, Co, Ni, Cu) and is suitable for quick estimation; the VASP model currently supports only Ti and Cu with higher accuracy. The dual-model design meets different needs.

## Usage: Local Deployment and Online Service

Local deployment requires creating a Conda environment with Python 3.11, cloning the GitHub repository (https://github.com/AI-multimodal/LightshowAI), installing it in editable mode via pip (dependencies like cmake and other compilation tools are needed), and running `xas_ui` to start the web interface. If you don't want to deploy locally, you can use the online service: https://lightshowai.bnl.gov/.

## Application Scenarios and Value: Accelerating Materials Science Research

LightshowAI can complete predictions in seconds (traditional calculations take hours/days). Its value lies in: 1. High-throughput screening: Pre-screen candidate materials to reduce the workload of high-precision calculations; 2. Experimental design: Optimize synchrotron radiation experiment parameters; 3. Teaching and training: Intuitively understand the relationship between structure and spectra; 4. Cross-scale connection: Mapping from atomic structure to macroscopic observations.

## Limitations and Future Outlook

Current limitations: Only supports K-edge of 3d transition metals; the VASP model only supports Ti/Cu; accuracy depends on training data. Future directions: Expand elements/edges (e.g., L-edge), support EXAFS spectra, improve accuracy, integrate material databases and calculation workflows. The project is under active development and may have breaking changes.

## Academic Compliance and Institutional Support

Usage requires citing relevant papers: Benchmark (Phys. Rev. Materials 8, 013801, 2024), Lightshow software (JOSS 8(87), 5182, 2023), OmniXAS model (Phys. Rev. Materials 9, 043803, 2025). The project is supported by the U.S. Department of Energy Office of Basic Energy Sciences and uses resources from the Center for Functional Nanomaterials (CFN) at Brookhaven National Laboratory. The software is provided 'as is' without warranty; the U.S. government holds a non-exclusive global license to ensure openness.