Title: Data Driven Macroscopic Modeling across Knudsen Numbers for Rarefied Gas Dynamics and Application to Rayleigh Scattering

Abstract: Macroscopic modeling of the gas dynamics across Knudsen numbers from dense gas region to rarefied gas region remains a great challenge. The reason is macroscopic models lack accurate constitutive relations valid across different Knudsen numbers. To address this problem, we proposed a Data-driven, KnUdsen number Adaptive Linear constitutive relation model named DUAL. The DUAL model is accurate across a range of Knudsen numbers, from dense to rarefied, through learning to adapt Knudsen number change from observed data. It is consistent with the Navier-Stokes equation under the hydrodynamic limit, by utilizing a constrained neural network. In addition, it naturally satisfies the second law of thermodynamics and is robust to noisy data. We test the DUAL model on the calculation of Rayleigh scattering spectra. The DUAL model gives accurate spectra for various Knudsen numbers and is superior to traditional perturbation and moment expansion methods.