Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Machine Learning Approach to Moment Closure Relations for Plasma: A Review

Samuel Burles, Enrico Camporeale

TL;DR

The paper surveys machine learning strategies for plasma moment closures, detailing neural network surrogates, physics-informed networks, neural operators, and equation-discovery methods like SINDy and PDE-Net. It highlights how data-driven closures can capture kinetic effects and improve large-scale fluid simulations, while noting challenges in accuracy of off-diagonal terms, generalization across regimes, and integration into 3D solvers. The review emphasizes trade-offs between interpretability and predictive power, data noise, and the computational costs of training, pointing to hybrid approaches and shared databases as paths forward. Overall, the work underscores the potential of data-driven closures to enable computationally feasible, kinetically faithful plasma simulations, while outlining concrete research directions to achieve robust, scalable implementations.

Abstract

The requirement for large-scale global simulations of plasma is an ongoing challenge in both space and laboratory plasma physics. Any simulation based on a fluid model inherently requires a closure relation for the high order plasma moments. This review compiles and analyses the recent surge of machine learning approaches developing improved plasma closure models capable of capturing kinetic phenomena within plasma fluid models. The purpose of this review is both to collect and analyse the various methods employed on the plasma closure problem, including both equation discovery methods and neural network surrogate approaches, as well as to provide a general overview of the state of the problem. In particular, we highlight the challenges of developing a data-driven closure as well as the direction future work should take toward addressing these challenges, in the pursuit of a computationally viable large-scale global simulation.

The Machine Learning Approach to Moment Closure Relations for Plasma: A Review

TL;DR

The paper surveys machine learning strategies for plasma moment closures, detailing neural network surrogates, physics-informed networks, neural operators, and equation-discovery methods like SINDy and PDE-Net. It highlights how data-driven closures can capture kinetic effects and improve large-scale fluid simulations, while noting challenges in accuracy of off-diagonal terms, generalization across regimes, and integration into 3D solvers. The review emphasizes trade-offs between interpretability and predictive power, data noise, and the computational costs of training, pointing to hybrid approaches and shared databases as paths forward. Overall, the work underscores the potential of data-driven closures to enable computationally feasible, kinetically faithful plasma simulations, while outlining concrete research directions to achieve robust, scalable implementations.

Abstract

The requirement for large-scale global simulations of plasma is an ongoing challenge in both space and laboratory plasma physics. Any simulation based on a fluid model inherently requires a closure relation for the high order plasma moments. This review compiles and analyses the recent surge of machine learning approaches developing improved plasma closure models capable of capturing kinetic phenomena within plasma fluid models. The purpose of this review is both to collect and analyse the various methods employed on the plasma closure problem, including both equation discovery methods and neural network surrogate approaches, as well as to provide a general overview of the state of the problem. In particular, we highlight the challenges of developing a data-driven closure as well as the direction future work should take toward addressing these challenges, in the pursuit of a computationally viable large-scale global simulation.

Paper Structure

This paper contains 28 sections, 18 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Background
  3. Machine Learning
  4. Why ML?
  5. ML Models
  6. Neural Networks
  7. Sparse Regression
  8. Plasma Closure Review
  9. Analytic Closure Training Data
  10. Kinetic Simulation Training Data
  11. Data Driven Equation Discovery
  12. SINDy Approach
  13. Neural Network Approach
  14. Discussion
  15. Overview of Prior Work
  16. ...and 13 more sections

Figures (2)

  • Figure 1: Schematic diagram from the paper by Qin et al. (a) Describes the data generated from the kinetic simulation. (b) The domains from which the training data were sourced from the kinetic simulation data. (c) The architecture for the physics-informed neural network. (d) The predictions were produced using the PINN. Credit to qin_data-driven_2023
  • Figure 2: Graph of model complexity (represented by the number of non-zero terms) versus the error of the discovered model. The hierarchy of physical assumptions leading to each model is highlighted. Credit: alves_data-driven_2022