Estimation of the dielectric constant of polymer melts by machine learning incorporating higher-order structural descriptors based on persistent homology

The polymer MI (Materials Informatics) method for exploring the properties of polymer melts has not yet been established because no descriptor that can describe higher-order structures has been found. From past studies, we have confirmed that the higher-order structure of polymer melt can be expressed by interpreting the coordinate data of all-atom MD using persistent homology, which is one of TDA (Topological Data Analysis). In this study, a learning model has been constructed by machine learning using a persistent diagram vectorized from a comprehensive data set of various types of the polymer melt as a descriptor and a dielectric constant as the objective variable. As a result, it has been shown that a model with high prediction accuracy based on a descriptor with physical meaning was constructed, a search space was defined by using SOM, and an unknown region could be estimated by Gaussian process regression.