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2 机器学习方法辅助形核序参量设计^[75,77]

Fig.2 Machine learning to aid design of nucleation order parameters^[75] ( $d_{i}$ is the $i$ -th input descriptor. L is the loss function. $C (∙)$ is the time autocorrelation function. ${\tilde{λ}}_{i}$ and $α_{i j}$ are the solution of the genderized eigenvalue problem, corresponding to the eigenvalues and eigenvectors, respectively. $s$ is the collective variable established by the deep neural network. $h_{i}^{θ} (∙)$ is the $i$ -th component of the neural network output as a function of time, where $θ$ denotes the parameter of the neural network. $t$ and $τ$ are time. CV—collective variable. TICA—time-lagged independent component analysis)
(a) Deep-LDA^[75] (b) Deep-TICA^[77]