物理化学学报 >> 2026, Vol. 42 >> Issue (5): 100213.doi: 10.1016/j.actphy.2025.100213

论文 上一篇    下一篇

T2MAT (文本到材料)：基于文本生成目标性能材料结构的通用智能体

宋志龙1,2, 陆帅华1, 周跫桦1,2,*(), 王金兰1,2,*()   

  1. 1 东南大学物理学院量子材料与器件教育部重点实验室, 江苏 南京 211189
    2 苏州实验室, 江苏 苏州 215004
  • 收稿日期:2025-08-25 修回日期:2025-10-27 录用日期:2025-10-27 发布日期:2026-01-23
  • 通讯作者: Email: qh.zhou@seu.edu.cn (周跫桦)jlwang@seu.edu.cn (王金兰)

T2MAT (text-to-material): a universal agent for generating material structures with goal properties from a single sentence

Zhilong Song1,2, Shuaihua Lu1, Qionghua Zhou1,2,*(), Jinlan Wang1,2,*()   

  1. 1 Key Laboratory of Quantum Materials and Devices of Ministry of Education, School of Physics, Southeast University, Nanjing 211189, Jiangsu Province, China
    2 Suzhou Laboratory, Suzhou 215004, Jiangsu Province, China
  • Received:2025-08-25 Revised:2025-10-27 Accepted:2025-10-27 Published:2026-01-23
  • Contact: Email: qh.zhou@seu.edu.cn (Qionghua Zhou)jlwang@seu.edu.cn (Jinlan Wang)

RichHTML

26

PDF

753

赞

补充材料

3

摘要：

人工智能生成内容(AIGC)——由AI系统无需人工干预自主生成的内容——已显著提升多个领域的效率。然而，材料科学中的AIGC技术面临双重挑战：既要高效发现超越现有数据库范围的新型材料，又需确保晶体结构的对称性与稳定性。为应对该挑战，我们开发了T2MAT(文本到材料)，这是一种端到端智能体，能够通过全面探索化学空间并结合全自动第一性原理验证，将用户输入的文本转化为超越现有数据库范围、具有目标性能的新型材料结构的逆向设计。此外，我们提出CGTNet (晶体图变换器网络)，这是一种专门用于捕捉长程相互作用的图神经网络，可显著提高性质预测的准确性和数据效率，从而增强逆向设计的可靠性。通过这些创新，T2MAT降低了对人类专业知识的依赖，加速了高性能功能材料的发现，为真正全自动的材料设计铺平了道路。

关键词: 智能体, 材料设计, 大语言模型, 生成式模型, 图神经网络

Abstract:

Artificial Intelligence-Generated Content (AIGC)—content autonomously produced by AI systems without human intervention—has significantly boosted efficiency across various fields. However, AIGC in material science faces challenges in efficiently discovering novel materials that surpass existing databases, while ensuring the invariance and stability of crystal structures. To address these challenges, we develop T2MAT (text-to-material), an end-to-end agent that transforms user-input text into the inverse design of novel material structures with target properties beyond existing database, enabled by comprehensive exploration of chemical space and fully automated first-principles validation. Furthermore, we propose CGTNet (Crystal Graph Transformer NETwork), a graph neural network specifically designed to capture long-range interactions, which dramatically improves the accuracy and data efficiency of property predictions and thereby strengthens the reliability of inverse design. Through these contributions, T2MAT reduces the reliance on human expertise and accelerates the discovery of high-performance functional materials, paving the way for truly autonomous material design.

Key words: Agent, Material design, Large language model, Generative model, Graph neural network