MgAlSi alloy materials have the main properties of light weight and high strength, good electrical and thermal conductivity and corrosion resistance, and have various applications in the industrial field, making an important contribution to the realization of lightweight and high performance needs. In order to be able to predict the material properties of MgAlSi alloys with a high degree of accuracy, this paper develops for the first time an interatomic potential function for MgAlSi alloys based on a neural network machine learning approach. The effectiveness of the developed machine learning potentials is verified by analyzing the problems encountered during the training process and the errors of the finally obtained potential functions, and comparing some of the radial distribution functions, coordination numbers, and predictions of properties such as the equation of state, lattice constants, shear modulus and bulk modulus with those of AIMD. It is found that the performance error of the deep potential model is basically kept in the same order of magnitude as that of DFT calculations, the computational speed can be up to nearly a thousand times that of DFT, and the computational cost is linearly related to the atomic number, which is well suited for large-scale molecular dynamics simulations, and it will provide a promising solution for accurate large-scale molecular dynamics simulations.

中文翻译：

---

MgAlSi合金的深电位拟合及力学性能研究

MgAlSi合金材料具有轻质高强、良好的导电导热性和耐腐蚀性等主要性能，在工业领域有着多种应用，为实现轻量化和高性能需求做出了重要贡献。为了能够高精度预测 MgAlSi 合金的材料性能，本文首次基于神经网络机器学习方法开发了 MgAlSi 合金的原子间势函数。通过分析训练过程中遇到的问题和最终获得的势函数的误差，并比较一些径向分布函数、配位数以及方程等性质的预测，验证了所开发的机器学习势的有效性。状态、晶格常数、剪切模量和体积模量与 AIMD 的结果相同。发现深势模型的性能误差基本与DFT计算保持在同一数量级，计算速度可达DFT的近千倍，计算成本与原子序数非常适合大规模分子动力学模拟，它将为精确的大规模分子动力学模拟提供有前途的解决方案。