力学与实践 ›› 2020, Vol. 42 ›› Issue (1): 13-16.DOI: 10.6052/1000-0879-19-412

• 应用研究 • 上一篇    下一篇

运用机器学习方法设计原子链人工边界条件 1)

张慊*, 乔丹, 唐少强*,2)()   

  1. * 北京大学工学院力学与工程科学系,北京 100871
    † 北京大学数学科学学院概率统计系,北京 100871
  • 收稿日期:2019-11-11 出版日期:2020-02-20 发布日期:2020-03-13
  • 通讯作者: 唐少强
  • 作者简介:2) 唐少强,教授。E-mail: maotang@pku.edu.cn
  • 基金资助:
    1) 国家自然科学基金资助项目(11832001)

DESIGNING ARTIFICIAL BOUNDARY CONDITIONS FOR ATOMIC CHAINS BY MACHINE LEARNING 1)

ZHANG Qian*, QIAO Dan, TANG Shaoqiang*,2)()   

  1. * Department of Mechanics and Engineering Science, College of Engineering, Peking University, Beijing 100871, China
    † Department of Probability and Statistics, School of Mathematical Sciences, Peking University, Beijing 100871, China
  • Received:2019-11-11 Online:2020-02-20 Published:2020-03-13
  • Contact: TANG Shaoqiang

摘要:

本文运用机器学习方法设计一维线性原子链的人工边界条件。该方法基于前馈神经网络,通过对一小部分数值解进行训练后得到人工边界条件。应用该法不需要较多先验知识,编程简单,实现速度快,算例表明数值反射较小。

关键词: 分子动力学模拟, 人工边界条件, 机器学习, 前馈神经网络

Abstract:

In this paper, we adopt machine learning techniques to design artificial boundary conditions for one-dimensional linear atomic chain. Training a feedforward neural network with a small amount of numerical solutions, we obtain artificial boundary conditions. This approach requires little prior information, and programming and computation are fast. Numerical examples illustrate a relatively small reflection.

Key words: molecular dynamics simulation, artificial boundary conditions, machine learning, feedforward neural network

中图分类号: