Spectral Learning of Magnetized Plasma Dynamics: A Neural Operator Application

2025/07/02 by Roberta Duarte, Duarte, Roberta, Rodrigo Nemmen +3 · 1 citation
Physics and Astronomy · Materials Science · #Model Reduction and Neural Networks #Quantum many-body systems #Machine Learning in Materials Science

paper · pdf · doi:10.48550/arxiv.2507.01388

Abstract

Fourier neural operators (FNOs) provide a mesh-independent way to learn solution operators for partial differential equations, yet their efficacy for magnetized turbulence is largely unexplored. Here we train an FNO surrogate for the 2-D Orszag-Tang vortex, a canonical non-ideal magnetohydrodynamic (MHD) benchmark, across an ensemble of viscosities and magnetic diffusivities. On unseen parameter settings the model achieves a mean-squared error of ≈ 6 × 10^-3 in velocity and ≈ 10^-3 in magnetic field, reproduces energy spectra and dissipation rates within 96% accuracy, and retains temporal coherence over long timescales. Spectral analysis shows accurate recovery of large- and intermediate-scale structures, with degradation at the smallest resolved scales due to Fourier-mode truncation. Relative to a UNet baseline the FNO cuts error by 97%, and compared with a high-order finite-volume solver it delivers a 25× inference speed-up, offering a practical path to rapid parameter sweeps in MHD simulations.

Citations

• Physics-informed deep-learning applications to experimental fluid mechanics
• Black hole weather forecasting with deep learning: a pilot study
• Machine learning–accelerated computational fluid dynamics
• Fourier Neural Operator for Parametric Partial Differential Equations
• NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations
• Lagrangian Neural Networks
• Hamiltonian Neural Networks
• Machine Learning for Fluid Mechanics
• Compressed Convolutional LSTM: An Efficient Deep Learning framework to\n Model High Fidelity 3D Turbulence
• Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• Deep Fluids: A Generative Network for Parameterized Fluid Simulations
• Attention Is All You Need
• Accelerating Eulerian Fluid Simulation With Convolutional Networks
• Adam: A Method for Stochastic Optimization

Cited by

• Fill the gaps: continuous in time interpolation of fluid dynamical simulations

Related

• Neural Operators Learn the Local Physics of Magnetohydrodynamics
• Resolving Turbulent Magnetohydrodynamics: A Hybrid Operator-Diffusion Framework
• LOGLO-FNO: Efficient Learning of Local and Global Features in Fourier Neural Operators
• Fourier neural operators for spatiotemporal dynamics in two-dimensional turbulence
• Incremental Spatial and Spectral Learning of Neural Operators for Solving Large-Scale PDEs