Neural network representability of fully ionized plasma fluid model closures
View/ Open
File version
Accepted Manuscript (AM)
Author(s)
Maulik, R
Garland, NA
Burby, JW
Tang, XZ
Balaprakash, P
Griffith University Author(s)
Year published
2020
Metadata
Show full item recordAbstract
The closure problem in fluid modeling is a well-known challenge to modelers aiming to accurately describe their systems of interest. Over many years, analytic formulations in a wide range of regimes have been presented but a practical, generalized fluid closure for magnetized plasmas remains an elusive goal. In this study, as a first step toward constructing a novel data-based approach to this problem, we apply ever-maturing machine learning methods to assess the capability of neural network architectures to reproduce crucial physics inherent in known magnetized plasma closures. We find encouraging results, indicating the ...
View more >The closure problem in fluid modeling is a well-known challenge to modelers aiming to accurately describe their systems of interest. Over many years, analytic formulations in a wide range of regimes have been presented but a practical, generalized fluid closure for magnetized plasmas remains an elusive goal. In this study, as a first step toward constructing a novel data-based approach to this problem, we apply ever-maturing machine learning methods to assess the capability of neural network architectures to reproduce crucial physics inherent in known magnetized plasma closures. We find encouraging results, indicating the applicability of neural networks to closure physics, but also arrive at recommendations on how one should choose appropriate network architectures for the given locality properties dictated by the underlying physics of the plasma.
View less >
View more >The closure problem in fluid modeling is a well-known challenge to modelers aiming to accurately describe their systems of interest. Over many years, analytic formulations in a wide range of regimes have been presented but a practical, generalized fluid closure for magnetized plasmas remains an elusive goal. In this study, as a first step toward constructing a novel data-based approach to this problem, we apply ever-maturing machine learning methods to assess the capability of neural network architectures to reproduce crucial physics inherent in known magnetized plasma closures. We find encouraging results, indicating the applicability of neural networks to closure physics, but also arrive at recommendations on how one should choose appropriate network architectures for the given locality properties dictated by the underlying physics of the plasma.
View less >
Journal Title
Physics of Plasmas
Volume
27
Issue
7
Publisher URI
Copyright Statement
© 2020 American Institute of Physics. This article may be downloaded for personal use only. Any other use requires prior permission of the author and AIP Publishing. This article appeared in Maulik, R; Garland, NA; Burby, JW; Tang, XZ; Balaprakash, P, Neural network representability of fully ionized plasma fluid model closures, Physics of Plasmas, 2020, 27 (7), Arcticle 072106, and may be found at https://doi.org/10.1063/5.0006457
Subject
Neural networks
Plasma physics; fusion plasmas; electrical discharges
Numerical and computational mathematics
physics.comp-ph
physics.plasm-ph