基于机器学习的硼稀体系热输运中非谐效应研究

分子动力学模拟能够考虑所有阶的非谐效应，是研究材料物性的常用方法，但其准确性通常受制于模拟中所用经验势函数的精度。传统的经验势函数通过拟合特定结构的现有实验测量物性得到，在预测其他结构和物性时可转移性差。并且新材料体系由于缺乏实验数据无法拟合势函数参数，严重限制了经验势函数的应用。本项目基于机器学习势函数，结合分子动力学模拟和玻尔兹曼输运理论，主要研究内容包括：（1）构建基于机器学习势函数的材料多目标物性预测平台；（2）通过算法优化和GPU架构提高运算效率，发展适用于大规模分子动力学模拟的机器学习势函数；（3）系统研究新型二维材料硼稀中非谐效应对热传导的影响及其物理机制；（4）研究非谐相互作用调控由硼稀构成的异质结界面热输运的物理机制。本项目旨在研究机器学习势函数相关的基础科学问题，并将其应用于硼稀体系热输运中非谐效应的研究，为新材料体系和具有强非谐效应体系的热物性预测提供有效手段。

Molecular dynamics simulation can take into account the anharmonicity to all orders, and is a widely used method to study the material's physical properties, but its accuracy is often limited by the accuracy of the empirical potential used in the simulation. The conventional empirical potential is obtained by fitting the existing experimental measurement results of certain physical property in a given structure, but its transferability is very bad when predicting other structures or physical properties not used in the fitting. In addition, the application of empirical potential is seriously limited in the studies of new materials, due to the lack of experimental data that can be used to fit the empirical potential. Based on the machine learning potential, this project will combine the molecular dynamics simulation and Boltzmann transport theory to predict the material's macroscopic physical properties. The main research topic includes: (1) Build a computational platform that can predict the material's multiple-target physical properties based on the machine learning potential. (2) Enhance the computational efficiency via the optimization of the algorithms and the GPU architecture, and develop the machine learning potential suitable for the large-scale molecular dynamics simulations. (3) Based on the machine learning potential, systematically study the effect of ahnarmonicity on heat conduction in the novel two-dimensional material borophene and the underlying physical mechanisms. (4) By using the machine learning potential and molecular dynamics simulations, investigate the physical mechanism of the anharmonic interaction for regulating the interfacial thermal transport across the hetero-junctions constructed by borophene. This project aims to study the fundamental scientific aspects related to the machine learning potential, and further applies it to study the effect of anharmonicity on the thermal transport in borophene system, which can provide effective tools for the prediction of thermal transport properties of new materials and systems with strong anharmonicity.