Zhengyu Sun, Weiwei Sun, Shaohan Li, Zening Yang, Mutian Zhang, Yang Yang, Huayun Geng, Jin Yu
{"title":"CrysGraphFormer: An Equivariant Graph Transformer for Prediction of Lattice Thermal Conductivity with Interpretability","authors":"Zhengyu Sun, Weiwei Sun, Shaohan Li, Zening Yang, Mutian Zhang, Yang Yang, Huayun Geng, Jin Yu","doi":"10.1039/d4ta04495a","DOIUrl":null,"url":null,"abstract":"To address the challenges of high error rates and poor generalization in current deep learning models for predicting lattice thermal conductivity (LTC), we introduce CrysGraphFormer, an innovative equivariant crystal graph Transformer model tailored for this task. The model incorporates an improved multi-head self-attention mechanism and human-designed feature descriptors. Utilizing a message-passing mechanism to update node information, it introduces relative coordinate differences to represent crystal symmetry, avoiding the complex and computationally expensive higher-order representations traditionally used. We constructed a comprehensive dataset containing 5729 LTC data points (300K), including 5477 materials from AFLOW, 112 MAX and MAB phase materials calculated using VASP, and 140 for half-Heusler alloys. Experimental results demonstrate that the CrysGraphFormer model achieves state-of-the-art performance in LTC prediction tasks and excels in predicting fundamental properties. The model offers good interpretability, providing insights from chemical and materials science perspectives. Furthermore, we validated the model's application potential in the field of thermoelectric materials by predicting the LTC of 59 thermoelectric materials and 55 ternary semiconductor materials, with results consistent with DFT calculations. Finally, the uncertainty of CrysGraphFormer was assessed using Monte Carlo Dropout method.","PeriodicalId":82,"journal":{"name":"Journal of Materials Chemistry A","volume":null,"pages":null},"PeriodicalIF":10.7000,"publicationDate":"2024-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Materials Chemistry A","FirstCategoryId":"88","ListUrlMain":"https://doi.org/10.1039/d4ta04495a","RegionNum":2,"RegionCategory":"材料科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"CHEMISTRY, PHYSICAL","Score":null,"Total":0}
引用次数: 0
Abstract
To address the challenges of high error rates and poor generalization in current deep learning models for predicting lattice thermal conductivity (LTC), we introduce CrysGraphFormer, an innovative equivariant crystal graph Transformer model tailored for this task. The model incorporates an improved multi-head self-attention mechanism and human-designed feature descriptors. Utilizing a message-passing mechanism to update node information, it introduces relative coordinate differences to represent crystal symmetry, avoiding the complex and computationally expensive higher-order representations traditionally used. We constructed a comprehensive dataset containing 5729 LTC data points (300K), including 5477 materials from AFLOW, 112 MAX and MAB phase materials calculated using VASP, and 140 for half-Heusler alloys. Experimental results demonstrate that the CrysGraphFormer model achieves state-of-the-art performance in LTC prediction tasks and excels in predicting fundamental properties. The model offers good interpretability, providing insights from chemical and materials science perspectives. Furthermore, we validated the model's application potential in the field of thermoelectric materials by predicting the LTC of 59 thermoelectric materials and 55 ternary semiconductor materials, with results consistent with DFT calculations. Finally, the uncertainty of CrysGraphFormer was assessed using Monte Carlo Dropout method.
期刊介绍:
The Journal of Materials Chemistry A, B & C covers a wide range of high-quality studies in the field of materials chemistry, with each section focusing on specific applications of the materials studied. Journal of Materials Chemistry A emphasizes applications in energy and sustainability, including topics such as artificial photosynthesis, batteries, and fuel cells. Journal of Materials Chemistry B focuses on applications in biology and medicine, while Journal of Materials Chemistry C covers applications in optical, magnetic, and electronic devices. Example topic areas within the scope of Journal of Materials Chemistry A include catalysis, green/sustainable materials, sensors, and water treatment, among others.