Method for solving Vlasov equation based on physical informed Fourier neural operator

Fu Wei; Wang Chuan; Zhang Tianjue; Zhou Hongji

doi:10.11884/HPLPB202537.250071

Article Contents

Article Navigation > High Power Laser and Particle Beams > 2025 > Accepted Manuscript

Fu Wei, Wang Chuan, Zhang Tianjue, et al. Method for solving Vlasov equation based on physical informed Fourier neural operator[J]. High Power Laser and Particle Beams. doi: 10.11884/HPLPB202537.250071

Citation:

Fu Wei, Wang Chuan, Zhang Tianjue, et al. Method for solving Vlasov equation based on physical informed Fourier neural operator[J]. High Power Laser and Particle Beams. doi: 10.11884/HPLPB202537.250071

Citation:

PDF( 6783 KB)

Method for solving Vlasov equation based on physical informed Fourier neural operator

doi: 10.11884/HPLPB202537.250071

Department of Nuclear Technology and Application, China Institute of Atomic Energy, Beijing 102413, China

Received Date: 2025-04-10
Accepted Date: 2025-08-02
Rev Recd Date: 2025-08-21

Available Online: 2025-09-01

Abstract

Abstract

Background
The Vlasov equation is a cornerstone in plasma physics, governing the evolution of distribution functions in high-temperature, collisionless plasmas. Conventional numerical methods, including Eulerian and Lagrangian approaches, often encounter severe computational challenges due to the rapid increase in cost with fine grid resolutions and the curse of dimensionality. These limitations restrict their effectiveness in large-scale kinetic plasma simulations needed in fusion research and space plasma studies.
Purpose
This work aims to develop an efficient and scalable computational framework for solving the Vlasov equation that mitigates the drawbacks of traditional methods. The study particularly addresses the need for maintaining accuracy and physical consistency while significantly reducing computational costs in high-dimensional simulations. An approach based on the Physics-Informed Fourier Neural Operator (PFNO) is introduced.
Methods
The method integrates the high-dimensional function mapping ability of the Fourier Neural Operator with the physical constraints of the Vlasov equation. A physics-informed loss function is constructed to enforce mass, momentum, and energy conservation laws. The framework was evaluated through benchmark tests against finite element and spectral solvers, and its parallel performance was assessed on large-scale computing platforms.
Results
The PFNO approach demonstrates accuracy comparable to conventional solvers while achieving computational efficiency improvements of one to two orders of magnitude. The method shows strong generalization under sparse-data conditions, exhibits grid independence, and scales effectively in parallel computing environments, enabling efficient treatment of high-dimensional plasma dynamics. The study presents a novel paradigm for solving high-dimensional Vlasov equations by combining deep learning operators with physical principles.
Conclusions
The PFNO framework enhances efficiency without sacrificing physical accuracy, making it a promising tool for applications in inertial confinement fusion, astrophysical plasma modeling, and space plasma simulations. Future research directions include extension to multi-species and relativistic plasma systems.
- neural operator,
- Vlasov equation,
- deep learning,
- physics informed,
- plasma

FullText(HTML)

References(16)

References

[1]	Chen, Francis F. Introduction to plasma physics and controlled fusion [M]. Vol. 1. New York: Plenum Press, 1984.
[2]	Bardos, C, Anne N. A Vlasov equation with Dirac potential used in fusion plasmas [J]. Journal of Mathematical Physics 53, 2012, no. 11.
[3]	Palmroth, M. , Ganse, U. , et al. Vlasov methods in space physics and astrophysics [J]. Living Reviews in Computational Astrophysics, 2018, 4(1), p. 1.
[4]	Bertrand P, Albrecht-Marc M, et al. Vlasov models for laser-plasma interaction [J]. Transport Theory and Statistical Physics. 2005 Jan 1;34(1-2): 103-26.
[5]	Fijalkow, Eric. A numerical solution to the Vlasov equation [J]. Computer Physics Communications, 1999, 116. 2-3: 319-328.
[6]	Birdsall, C. K, et al. Plasma physics via computer simulation [M]. 2018.
[7]	Murawski K, Numerical solutions of magnetohydrodynamic equations [J]. Bulletin of the Polish Academy of Sciences-Technical Sciences, 2011, 59(2): 219-226.
[8]	Sonnendrücker, Eric, et al. The semi-Lagrangian method for the numerical resolution of the Vlasov equation[J]. Journal of Computational Physics, 1999, 149(2): 201-220. doi: 10.1006/jcph.1998.6148
[9]	Roska, Tamás, Leon O. Chua, et al. Simulating nonlinear waves and partial differential equations via CNN. I. Basic techniques[J]. IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications, 2002(10): 807-815.
[10]	Kovachki, Nikola, Zongyi Li, et al. Neural operator: learning maps between function spaces with applications to PDES[J]. Journal of Machine Learning Research, 2023, 24(89): 1-97.
[11]	Raissi, M. , Paris P. , et al. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations [J]. Journal of Computational Physics, 2019, 378 686-707.
[12]	Zhang, Pan, et al. A Maxwell's equations based deep learning method for time domain electromagnetic simulations[J]. IEEE Journal on Multiscale and Multiphysics Computational Techniques, 2021, 6: 35-40. doi: 10.1109/JMMCT.2021.3057793
[13]	Eivazi, Hamidreza, Mojtaba Tahani, et al. Physics-informed neural networks for solving Reynolds-averaged Navier–Stokes equations[J]. Physics of Fluids, 2022(7): 34.
[14]	Raissi, Maziar. Deep hidden physics models: deep learning of nonlinear partial differential equations[J]. Journal of Machine Learning Research, 2018(25): 1-24.
[15]	Qin, Yilan, Jiayu Ma, et al. Data-driven modeling of Landau damping by physics-informed neural networks. Physical Review Research, 2023, no. 3: 033079.
[16]	Juno, James, Ammar Hakim, et al. Discontinuous Galerkin algorithms for fully kinetic plasmas. Journal of Computational Physics 2018, 353: 110-147.