Abstract

In the process of modeling multilayer semiconductor nanostructures, an important role is played by the rapid acquisition of accurate values of the characteristics of the structure under consideration. One of these characteristics is the value of the interaction energy of atoms within the structure. The energy value is also important for obtaining other quantities, such as the bulk modulus of elasticity of the structure and the shear modulus. The paper discusses a method for obtaining the energy of the interaction between two atoms, based on machine learning methods. A model built based on the Gaussian Approximation Potential (GAP) is trained on a previously prepared sample and allows predicting the energy values of pairs of atoms for the test data. The values of the coordinates of interacting atoms, the distance between the atoms, the value of the lattice constant of the structure, an indication of the type of interacting atoms, and a value describing the environment of the atoms are used as attributes. The computational experiment is carried out with the participation of one-component compounds, such as Si, Ge, and C. The rate of obtaining the energy of interacting atoms and the accuracy of the obtained value are estimated. The speed and accuracy characteristics are compared with the values obtained using the multiparticle interatomic potential, the Tersoff potential.

中文翻译：

---

用于研究晶体结构性质的基于机器学习的原子间势

摘要

在对多层半导体纳米结构建模的过程中，快速获取所考虑结构特征的准确值起着重要作用。这些特征之一是结构内原子的相互作用能的值。能量值对于获得其他量也很重要，例如结构的体积弹性模量和剪切模量。本文讨论了一种基于机器学习方法获取两个原子之间相互作用能量的方法。基于高斯近似势 (GAP) 构建的模型在先前准备的样本上进行训练，并允许预测测试数据的原子对的能量值。相互作用原子的坐标值，原子之间的距离，结构的晶格常数的值、相互作用原子类型的指示以及描述原子环境的值被用作属性。计算实验在Si、Ge、C等单组分化合物的参与下进行，估计了相互作用原子的能量获取率和所得值的准确性。将速度和精度特性与使用多粒子原子间势（Tersoff 势）获得的值进行比较。估计获得相互作用原子能量的速率和获得值的准确性。将速度和精度特性与使用多粒子原子间势（Tersoff 势）获得的值进行比较。估计获得相互作用原子能量的速率和获得值的准确性。将速度和精度特性与使用多粒子原子间势（Tersoff 势）获得的值进行比较。