Dimitrios-Paraskevas Gerakinis, Eleonora Ricci, George Giannakopoulos, Vangelis Karkaletsis, Doros N. Theodorou, Niki Vergadou,
Abstract
Hierarchical multiscale methods are essential for molecular simulations of complex chemical systems, such as organic fluids and soft matter systems in order to reach longer length and time scales. Coarse-Graining (CG) is often the basis of multiscale schemes. Machine Learning (ML) techniques have been recently explored for the development of atomistic force fields based on quantum mechanical calculations. However, integrating ML methods into CG force fields for bulk molecular systems is still rare. In this work, Graph Convolutional Neural Network (GCNN) architectures were adopted to develop CG Machine Learned potentials for bulk amorphous systems, implementing a strategy that includes a force-matching scheme using benzene liquid as a test system. The ML-based CG force fields developed were evaluated by conducting molecular dynamics simulations at the CG level, and the extracted properties were compared with the atomistic simulations to assess the effectiveness of the ML CG interaction potentials. The impact of hyperparameters, loss function construction, and GCNN architecture size have been examined, providing valuable insights for ML-based CG approaches in bulk soft matter systems.