Abstract:We introduce a hyperbolic closure for the Grad moment expansion of the Bhatnagar-Gross-Krook's (BGK) kinetic model using a neural network (NN) trained on BGK's moment data. This closure is motivated by the exact closure for the free streaming limit that we derived in our paper on closures in transport \cite{Huang2022-RTE1}. The exact closure relates the gradient of the highest moment to the gradient of four lower moments. As with our past work, the model presented here learns the gradient of the highest moment in terms of the coefficients of gradients for all lower ones. By necessity, this means that the resulting hyperbolic system is not conservative in the highest moment. For stability, the output layers of the NN are designed to enforce hyperbolicity and Galilean invariance. This ensures the model can be run outside of the training window of the NN. Unlike our previous work on radiation transport that dealt with linear models, the BGK model's nonlinearity demanded advanced training tools. These comprised an optimal learning rate discovery, one cycle training, batch normalization in each neural layer, and the use of the \texttt{AdamW} optimizer. To address the non-conservative structure of the hyperbolic model, we adopt the FORCE numerical method to achieve robust solutions. This results in a comprehensive computing model combining learned closures with methods for solving hyperbolic models. The proposed model can capture accurate moment solutions across a broad spectrum of Knudsen numbers. Our paper details the multi-scale model construction and is run on a range of test problems.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is to provide a hyperbolic closure method for the Grad moment expansion of the Bhatnagar - Gross - Krook (BGK) kinetic model. Specifically, the author proposes a moment - closure method based on neural network (NN) training, which can handle the data generated from the kinetic model and maintain the hyperbolicity and Galilean invariance of the system. ### Main problems and challenges: 1. **Moment - closure problem**: In the moment method, in order to reduce the dimension of the kinetic equations, the moments are usually truncated to a certain fixed order. However, this will lead to an unclosed system because the time evolution of the highest - order moment depends on higher - order moments. 2. **Non - conservative structure**: For nonlinear problems such as the BGK model, traditional moment - closure methods often cause the system to lose hyperbolicity and are non - conservative in the highest - order moment. 3. **Computational complexity**: The computational complexity of directly solving the BGK equation in a high - dimensional phase space is very high, so effective dimension - reduction methods are required. ### Solutions: - **Hyperbolic closure**: The author introduces a new closure method, which ensures the hyperbolicity of the system by learning the relationship between moment gradients through a neural network. - **Machine - learning method**: Use a neural network to learn the moment - closure relationship from the kinetic data of the BGK model, so as to better capture the kinetic effects. - **Numerical method**: Adopt the FORCE numerical method to solve the stability problems caused by the non - conservative structure. ### Main contributions of the paper: 1. **Hyperbolicity and Galilean invariance**: By designing the output layer of the neural network, the hyperbolicity and Galilean invariance of the closed system are ensured, making the model run stably outside the training window. 2. **Adaptation to different Knudsen numbers**: This method is applicable to Knudsen numbers in the low - to - medium range and shows good predictive ability. 3. **Verification and application**: The correctness and effectiveness of the algorithm are verified through two types of training data (HME model and kinetic model), and good performance under different initial conditions is demonstrated. In conclusion, this paper aims to provide an efficient, accurate and stable hyperbolic moment - closure method for the BGK equation by combining machine learning and traditional moment - closure methods.

Hyperbolic Machine Learning Moment Closures for the BGK Equations

Learning Invariance Preserving Moment Closure Model for Boltzmann-BGK Equation

Direct Flux Gradient Approximation to Moment Closure of Kinetic Equations

Machine learning-based moment closure model for the semiconductor Boltzmann equation with uncertainties

Learning fast, accurate, and stable closures of a kinetic theory of an active fluid

Numerical Regularized Moment Method of Arbitrary Order for Boltzmann-BGK Equation

Machine Learning Moment Closure Models for the Radiative Transfer Equation I: Directly Learning a Gradient Based Closure

Closure equation and higher-order moment relations in the Gauss-Hermite lattice Boltzmann method

Uniformly Accurate Machine Learning Based Hydrodynamic Models for Kinetic Equations

Globally hyperbolic regularized moment method with applications to microflow simulation

Uniformly accurate machine learning-based hydrodynamic models for kinetic equations

A unified fourth-order Bhatnagar-Gross-Krook lattice Boltzmann model for high-dimensional linear hyperbolic equations

Generalized neural closure models with interpretability

Structure-preserving neural networks for the regularized entropy-based closure of the Boltzmann moment system

Globally Hyperbolic Regularization of Grad's Moment System

High order moment closure for Vlasov-Maxwell equations

Generalizable data-driven turbulence closure modeling on unstructured grids with differentiable physics

A Framework on Moment Model Reduction for Kinetic Equation

Globally Hyperbolic Moment System by Generalized Hermite Expansion

Quantum Hydrodynamic Model by Moment Closure of Wigner Equation