Loading [MathJax]/jax/output/HTML-CSS/jax.js

机器学习分子动力学辅助材料凝固形核研究进展
陈名毅, 胡俊伟, 余耀辰, 牛海洋

Advances in Machine Learning Molecular Dynamics to Assist Materials Nucleation and Solidification Research
CHEN Mingyi, HU Junwei, YU Yaochen, NIU Haiyang
图1 机器学习势函数原理示意图
Fig.1 Principles of machine learning potential
(a) Behler-Parrinello neural network (BPNN) (The subscript i denotes the serial number of the atom, Ri is the environmental matrix, Gi is the embedding net, Di is the descriptor, N and N' are two fitting nets for different elements, Ei is the i-th atom's energy, E is the calculated total energy)
(b) machine learning potential based on the kernel method (x is the environmental matrix; xi and Ei are the i-th reference configuration in the training set and the corresponding energy, respectively; αi is the i-th fitting coefficient; k(∙) is the kernel function to measure the difference between x and xi )