Accurate and Interpretable Dipole Interaction Model-Based Machine Learning for Molecular Polarizability

• DOI: 10.1021/acs.jctc.2c01094
• Publication Date: 2023-02-08T22:17:06Z
• Authors: Chaoqiang Feng; Jin Xi; Yaolong Zhang; Bin Jiang; Yong Zhou
• Keywords: Training set

Polarizabilities play significant roles in describing dispersive and inductive interactions of the atom molecular systems. However, an accurate prediction polarizabilities from first principles is computationally prohibitive. Although physical models or statistical machine learning have been proposed, either a lack description local chemical environments demanding large number samples for training has limited their practical applications. In this study, we combine physically inspired dipole interaction model neural network method predicting polarizability tensors molecules. With environment precisely described requirement rotational covariance naturally fulfilled, hybrid proven to give prediction, essentially reducing samples. The atomic are interpretable transferable larger molecules unseen set. This promising may find its wide range applications, such as spectroscopic simulations construction polarizable force fields.