Development of Coarse Grained Polymer Models by Iterative Boltzmann Inversion and Cross-Entropy Optimization Methods

Bottom-up computational modeling has been widely used to investigate the material properties of polymer matrix in present years. Molecular dynamics (MD) simulations is the powerful simulation method which could accurately describes the molecular interactions between molecules and thus provide an accurate prediction on the performance of materials. Furthermore, by trainning coarse grained (CG) models based on full atomistic information, an accurate CG can be developmed and used for the study of larger molecular weights to bridge the scales between nanoscale and mesoscale. The CG simulations reduce the degree of freedom of MD and thus increase the time scale and length scale in simulations. To develop a CG force field of a polymer system, the Iterative Boltzmann inversion method (IBIm) is widely used. Recently, many studies have shown that machine learning and deep learning methods (ML/DL) are useful for the development of force fields. In this study, we develop a novel workflow for trainning CG force field by the integration of IBIm and Cross-entropy optimizer. Our results show that the structural distribution and properties of a full atomistic system are both converged in optimization. The cross-entropy optimizer is able to optimize the loss of polymer conformation and molecular properties and thus provide a suitable CG force field for polymer materials.