Sequential and Intelligent Dissipative Particle Dynamics: A Multiscale Method Based on Time-Sequence Matching and Machine Learning

Abstract. We present a bottom-up multiscale model, termed sequential and intelligent dissipative particle dynamics (SIDPD), which bridges the gap between micro- and meso-scales with high accuracy and computational efficiency. This model employs machine learning techniques to predict the mesoscopic force fields directly from microscopic data, thus maintaining microscopic accuracy. Moreover, it matches force and momentum sequences rather than using memory kernels, an essential but computationally expensive component for bridging micro- and mesoscales, further enhancing its efficiency. The resulting force fields are applied in dissipative particle dynamics simulations to analyze static and dynamic properties. We consider three distinct microscopic systems with different densities, star polymer, methane and water, to evaluate the model performance. It is shown that SIDPD accurately preserve both static and dynamic properties with high efficiency, even for systems with high density and strong long-range interactions.