Shallow recurrent decoder for reduced order modeling of <i>E</i> × <i>B</i> plasma dynamics

Abstract Reduced-order models (ROMs) are becoming increasingly important for rendering complex and multiscale spatiotemporal dynamics computationally tractable. Computationally efficient ROMs are especially essential for optimized design of technologies as well as for gaining physical understanding. Plasma simulations, in particular those applied to the study of E × B plasma discharges and technologies, such as Hall thrusters for spacecraft propulsion, require substantial computational resources in order to resolve the multidimensional dynamics that span across wide spatial and temporal scales. While high-fidelity computational tools are available, their applications are limited to simplified geometries and narrow conditions, making simulations of full-scale plasma systems or comprehensive parametric studies computationally prohibitive. In addition, experimental setups involve limitations such as the finite spatial resolution of diagnostics and constraints imposed by geometrical accessibility. Consequently, both scientific research and industrial development of plasma systems, including E × B technologies, can greatly benefit from advanced ROM techniques that enable estimating the distributions of plasma properties across the entire system. We develop a model reduction scheme based upon a shallow recurrent decoder (SHRED) architecture using as few measurements of the system as possible. This scheme employs a neural network to encode limited sensor measurements in time (of either local or global properties) and reconstruct full spatial state vector via a shallow decoder network. Leveraging the theory of separation of variables, the SHRED architecture demonstrates the ability to reconstruct complete spatial fields with as few as three-point sensors, including fields dynamically coupled to the measured variables but not directly observed. The effectiveness of the ROMs derived with SHRED is demonstrated across several plasma configurations representative of different geometries in typical E × B plasma discharges and Hall thrusters.

Authors

Institutions

Publication Details

Journal
Machine Learning Science and Technology
Published
2025-04-15
DOI
https://doi.org/10.1088/2632-2153/adcd20
Citations
1
Primary Topic
Plasma Diagnostics and Applications
Type
article
Field-Weighted Citation Impact
0.73

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Shallow recurrent decoder for reduced order modeling of E × B plasma dynamics

Maryam Reza, J. Nathan Kutz, Farbod Faraji
1 citations
Machine Learning Science and Technology
Plasma Diagnostics and Applications
0.73
article

Shallow recurrent decoder for reduced order modeling of E × B plasma dynamics

Maryam Reza, J. Nathan Kutz, Farbod Faraji
article en
1 citations

Abstract

Abstract Reduced-order models (ROMs) are becoming increasingly important for rendering complex and multiscale spatiotemporal dynamics computationally tractable. Computationally efficient ROMs are especially essential for optimized design of technologies as well as for gaining physical understanding. Plasma simulations, in particular those applied to the study of E × B plasma discharges and technologies, such as Hall thrusters for spacecraft propulsion, require substantial computational resources in order to resolve the multidimensional dynamics that span across wide spatial and temporal scales. While high-fidelity computational tools are available, their applications are limited to simplified geometries and narrow conditions, making simulations of full-scale plasma systems or comprehensive parametric studies computationally prohibitive. In addition, experimental setups involve limitations such as the finite spatial resolution of diagnostics and constraints imposed by geometrical accessibility. Consequently, both scientific research and industrial development of plasma systems, including E × B technologies, can greatly benefit from advanced ROM techniques that enable estimating the distributions of plasma properties across the entire system. We develop a model reduction scheme based upon a shallow recurrent decoder (SHRED) architecture using as few measurements of the system as possible. This scheme employs a neural network to encode limited sensor measurements in time (of either local or global properties) and reconstruct full spatial state vector via a shallow decoder network. Leveraging the theory of separation of variables, the SHRED architecture demonstrates the ability to reconstruct complete spatial fields with as few as three-point sensors, including fields dynamically coupled to the measured variables but not directly observed. The effectiveness of the ROMs derived with SHRED is demonstrated across several plasma configurations representative of different geometries in typical E × B plasma discharges and Hall thrusters.

Machine Learning Science and TechnologyVol. 6(2)
University of Washington (US), Imperial College London (GB)
National Science Foundation, European Space Agency
Openalex Percentile: Top 30%
Plasma Diagnostics and Applications
0.73
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.