Intelligent Dissipative Particle Dynamics

Dissipative particle dynamics is a mesoscopic simulation method based on particles,which enables the investigation of large spatial and temporal scales.It has become a common tool for studying complex fluids such as red blood cells and polymers,and holds significant research value in both theory and application.We have developed a bottom-up coarse-graining method called Intelligent Dissipative Particle Dynamics(IDPD)using machine learning.This method can directly construct an accurate mesoscale force field from microscale molecular simulations,embedding the consistency between microscale and mesoscale into deep neural networks,including configuration,phase,static and dynamic properties.The deep neural network is trained with molecular simulation results,and microscale dynamics are mapped to mesoscale dynamics through Mori-Zwanzig projection.It learns to generate conservative and non-conservative force fields for mesoscale systems and applies them to dissipative particle dynamics simulations.To apply IDPD,we started with molecular dynamics simulations and coarsegrained polymers and methane fluids.We quantitatively compared the results of IDPD and molecular dynamics and found that IDPD can not only preserve the static properties but also the dynamic properties.In addition,we used the Constrained K-Means clustering algorithm to achieve coarse-graining of multiple molecules,i.e.,dynamically packing multiple molecules.In fact,preserving the dynamic properties of microscale systems and coarse-graining of multiple molecules are two open problems faced by many current coarse-graining methods.