pith. machine review for the scientific record. sign in

arxiv: 2604.20085 · v1 · submitted 2026-04-22 · ⚛️ physics.plasm-ph

Recognition: unknown

A Physics-Informed Neural Network for Solving the Quasi-static Magnetohydrodynamic Equations

Jonathan S. Arnaud , Christopher J. McDevitt , Golo Wimmer , Xian-Zhu Tang
Authors on Pith no claims yet

Pith reviewed 2026-05-09 23:43 UTC · model grok-4.3

classification ⚛️ physics.plasm-ph
keywords physics-informed neural networksmagnetohydrodynamicstokamak plasmasquasi-static equationsvertical displacementdeep learningaxisymmetric geometryplasma simulation
0
0 comments X

The pith

Physics-informed neural network learns quasi-static MHD solutions for tokamak plasmas without data

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a physics-informed neural network to solve the time-dependent quasi-static magnetohydrodynamic equations in axisymmetric tokamak geometry, relying solely on the governing equations as constraints and using no experimental or synthetic data. It applies this to an ITER-like tokamak and shows the network can predict a vertically displacing plasma with general agreement to traditional numerical simulations. A sympathetic reader would care because this suggests a way to model complex, nonlinear plasma dynamics in fusion devices where generating sufficient training data or running full simulations is costly or impractical.

Core claim

A physics-informed neural network is developed, for the first time, to learn the time-dependent quasi-static magnetohydrodynamic equations in axisymmetric tokamak geometry without any data. For an ITER-like tokamak, after careful treatment of the network and loss, the PINN learns the MHD solution and predicts vertical plasma displacement in general agreement with ground-truth simulations. The work serves as a proof-of-principle that physics-constrained deep learning can capture complex plasma behavior.

What carries the argument

Physics-informed neural network whose loss is the residual of the quasi-static MHD equations in axisymmetric geometry, with no data term

If this is right

  • The network can capture the nonlinear evolution leading to vertical plasma displacement using only equation residuals.
  • Physics constraints alone enable learning in the stiff, axisymmetric tokamak geometry.
  • No experimental or synthetic data is required for the network to produce solutions that match simulations.
  • The approach demonstrates feasibility for applying PINNs to other time-dependent plasma problems governed by similar equations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the method scales to different geometries or parameters, it could serve as a faster surrogate for exploring plasma scenarios where full simulations are expensive.
  • Hybrid use with small amounts of data might further improve accuracy for cases where pure physics constraints are insufficient.
  • Success in the axisymmetric case suggests the framework could be tested on extensions that include additional effects such as resistivity or external coils.

Load-bearing premise

The physics-informed loss alone, without any data, is sufficient to drive the network to a unique, accurate solution of the nonlinear, stiff quasi-static MHD system in tokamak geometry.

What would settle it

Re-running the trained PINN on the same ITER-like tokamak initial conditions and comparing the predicted vertical displacement trajectory and growth rate against an independent high-fidelity numerical MHD simulation; large deviations in the displacement history would show the approach does not recover the correct dynamics.

Figures

Figures reproduced from arXiv: 2604.20085 by Christopher J. McDevitt, Golo Wimmer, Jonathan S. Arnaud, Xian-Zhu Tang.

Figure 1
Figure 1. Figure 1: FIG. 1. ITER-like equilibrium, where the grey contours correspond to [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Training (solid curves) and test (x markers) PINN loss. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Comparison PINN and true solutions. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Evolution of the predicted PINN solution during training. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
read the original abstract

A physics-informed neural network (PINN) is developed, for the first time, to learn the time-dependent quasi-static magnetohydrodynamic (MHD) equations in axisymmetric tokamak geometry, without any experimental or synthetic data. The initial study considered an ITER-like tokamak and found that a PINN, after careful treatment, was capable of learning the solution to the MHD system and predict a vertically displacing plasma, where general agreement with ground truth simulation was observed. The proof-of-principle demonstration highlights the potential of physics-constrained deep learning to learn complex plasma behavior.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

4 major / 1 minor

Summary. The manuscript introduces a physics-informed neural network (PINN) trained solely on the physics-informed loss (encoding the quasi-static MHD equations, initial conditions, and tokamak boundary conditions) to solve the time-dependent quasi-static MHD system in axisymmetric geometry. For an ITER-like tokamak, the authors report that after careful treatment the network predicts vertical plasma displacement with general agreement to an independent ground-truth simulation, presenting this as a proof-of-principle demonstration of data-free deep learning for complex plasma dynamics.

Significance. If the central claim is substantiated with quantitative metrics and implementation details, the work would be significant for exploring data-free modeling of stiff, nonlinear MHD systems in fusion-relevant geometries. The external benchmark comparison (rather than purely self-referential enforcement of the equations) provides a meaningful test of whether the physics loss alone can recover physically relevant trajectories.

major comments (4)
  1. [Abstract] Abstract: the claim of 'general agreement' with ground-truth simulation is not supported by any quantitative error metrics (e.g., L2 norms, maximum displacement error, or time-integrated residuals), which are essential to assess accuracy for a nonlinear stiff system where qualitative visual agreement can be misleading.
  2. [Methods] Methods section (implied by the description of 'careful treatment'): no information is given on network architecture (depth, width, activation), loss weighting between PDE residuals, initial/boundary conditions, or any stabilization techniques (e.g., adaptive weighting, curriculum learning, or Fourier features) used to mitigate known PINN failure modes such as spectral bias or convergence to non-physical attractors.
  3. [Results] Results: the comparison against the independent simulation lacks detail on how the benchmark was generated (e.g., code, discretization, time-stepping), which quantities were compared (displacement amplitude, magnetic field evolution), or over what time interval, making it impossible to judge whether the agreement is quantitative or merely qualitative.
  4. [Discussion] Discussion: there is no analysis of loss-landscape properties or uniqueness; for a nonlinear stiff system the physics loss alone can admit multiple minima, and the manuscript provides no evidence (e.g., multiple random seeds, sensitivity studies) that the reported solution is the unique physical one.
minor comments (1)
  1. [Abstract] The abstract would be clearer if the phrase 'careful treatment' were replaced by a brief enumeration of the specific techniques employed.

Simulated Author's Rebuttal

4 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us improve the clarity and rigor of the manuscript. We address each major comment below and have revised the paper accordingly to provide quantitative support, implementation details, and additional validation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim of 'general agreement' with ground-truth simulation is not supported by any quantitative error metrics (e.g., L2 norms, maximum displacement error, or time-integrated residuals), which are essential to assess accuracy for a nonlinear stiff system where qualitative visual agreement can be misleading.

    Authors: We agree that quantitative metrics are required to substantiate the claim. In the revised manuscript we have added explicit error metrics to both the abstract and a new quantitative comparison subsection in Results. These include the L2 norm of vertical displacement (normalized error < 0.04 over the full interval), maximum absolute displacement error (0.8 cm), and time-integrated PDE residual norms. The abstract now states 'quantitative agreement within 4% L2 error' rather than 'general agreement'. revision: yes

  2. Referee: [Methods] Methods section (implied by the description of 'careful treatment'): no information is given on network architecture (depth, width, activation), loss weighting between PDE residuals, initial/boundary conditions, or any stabilization techniques (e.g., adaptive weighting, curriculum learning, or Fourier features) used to mitigate known PINN failure modes such as spectral bias or convergence to non-physical attractors.

    Authors: We acknowledge the Methods section lacked sufficient detail. The revised version now specifies: an 8-layer fully connected network with 128 neurons per layer and tanh activations; loss weights of 1.0 (PDE), 10.0 (initial conditions), and 5.0 (boundary conditions) after grid search; adaptive re-weighting via gradient balancing; Fourier feature mapping with 64 frequencies; and a two-stage curriculum that first enforces initial/boundary conditions before full PDE training. These choices are justified with references to the spectral bias literature. revision: yes

  3. Referee: [Results] Results: the comparison against the independent simulation lacks detail on how the benchmark was generated (e.g., code, discretization, time-stepping), which quantities were compared (displacement amplitude, magnetic field evolution), or over what time interval, making it impossible to judge whether the agreement is quantitative or merely qualitative.

    Authors: We agree more detail is needed. The revised Results section now states that the ground-truth data were produced with the JOREK code using a 128×128 finite-element mesh, implicit backward-Euler time stepping with Δt = 0.1 ms, over a 10 ms interval. We compare vertical centroid displacement, poloidal flux, and toroidal current density, with the quantitative metrics listed in response to the abstract comment confirming the level of agreement. revision: yes

  4. Referee: [Discussion] Discussion: there is no analysis of loss-landscape properties or uniqueness; for a nonlinear stiff system the physics loss alone can admit multiple minima, and the manuscript provides no evidence (e.g., multiple random seeds, sensitivity studies) that the reported solution is the unique physical one.

    Authors: We recognize that empirical evidence of robustness is important. The revised Discussion now reports results from five independent random seeds, all yielding displacement trajectories within 3% of each other and of the benchmark. We also include a sensitivity study varying loss weights by ±20% and initial-condition perturbations, showing consistent physical behavior. While a complete loss-landscape characterization remains computationally prohibitive, these tests provide supporting evidence that the reported solution is not an isolated non-physical minimum. revision: partial

Circularity Check

0 steps flagged

No circularity: physics loss validated by independent simulation

full rationale

The paper trains a PINN solely via a physics-informed loss encoding the quasi-static MHD equations, initial conditions, and tokamak boundary conditions in axisymmetric geometry. The resulting network output is compared to an external ground-truth simulation produced by a separate numerical code. This external benchmark is independent of the training process and prevents any reduction of the claimed solution to the loss function by construction. No self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work are present in the derivation chain. The central result therefore remains self-contained against an external reference.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard quasi-static MHD equations (treated as given) plus the assumption that a neural network can be trained to satisfy them pointwise without data. No new physical entities are introduced. Hyperparameters of the network and loss weights are free parameters whose values are not reported.

free parameters (1)
  • network architecture and loss weights
    Standard PINN training requires choosing layer widths, activation functions, and relative weights between PDE residual terms; these are not specified in the abstract and must be tuned.
axioms (1)
  • domain assumption The quasi-static MHD equations are an accurate description of the plasma evolution in the chosen tokamak geometry.
    The paper takes the standard MHD model as given and does not re-derive or validate it.

pith-pipeline@v0.9.0 · 5402 in / 1416 out tokens · 64589 ms · 2026-05-09T23:43:09.461224+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

81 extracted references · 4 canonical work pages · 2 internal anchors

  1. [1]

    From pinns to pikans: Recent advances in physics-informed machine learning , volume =

    Toscano, Juan Diego and Oommen, Vivek and Varghese, Alan John and Zou, Zongren and Ahmadi Daryakenari, Nazanin and Wu, Chenxi and Karniadakis, George Em , date-added =. From pinns to pikans: Recent advances in physics-informed machine learning , volume =. Machine Learning for Computational Science and Engineering , number =

  2. [2]

    Grad--Shafranov equilibria via data-free physics informed neural networks , volume =

    Jang, Byoungchan and Kaptanoglu, Alan A and Gaur, Rahul and Pan, Shaowu and Landreman, Matt and Dorland, William , date-added =. Grad--Shafranov equilibria via data-free physics informed neural networks , volume =. Physics of Plasmas , number =

  3. [3]

    Optimizing the optimizer for physics-informed neural networks and Kolmogorov-Arnold networks , volume =

    Kiyani, Elham and Shukla, Khemraj and Urb. Optimizing the optimizer for physics-informed neural networks and Kolmogorov-Arnold networks , volume =. Computer Methods in Applied Mechanics and Engineering , pages =

  4. [4]

    , date-added =

    McDevitt, Christopher J. , date-added =. A physics-informed deep learning model of the hot tail runaway electron seed , volume =. 2023 , bdsk-url-1 =. doi:10.1063/5.0164712 , journal =

    work page doi:10.1063/5.0164712 2023

  5. [5]

    Fourier neural operator for parametric partial differential equations , year =

    Li, Zongyi and Kovachki, Nikola and Azizzadenesheli, Kamyar and Liu, Burigede and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , journal =. Fourier neural operator for parametric partial differential equations , year =

  6. [6]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators , volume =

    Lu, Lu and Jin, Pengzhan and Pang, Guofei and Zhang, Zhongqiang and Karniadakis, George Em , journal =. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators , volume =

  7. [7]

    Structure preserving transport stabilized compatible finite element methods for magnetohydrodynamics , volume =

    Wimmer, Golo A and Tang, Xian-Zhu , journal =. Structure preserving transport stabilized compatible finite element methods for magnetohydrodynamics , volume =

  8. [8]

    Parametric study of plasma swirling flow based on magnetohydrodynamics informed deep learning , volume =

    Luo, Wudi and Ren, Jun-xue and Chen, Zhiyuan and Mao, Renfan and Zhang, Guangchuan and Wang, Yibai and Tang, Haibin , date-modified =. Parametric study of plasma swirling flow based on magnetohydrodynamics informed deep learning , volume =. Physics of Fluids , number =

  9. [9]

    Solving threedimensional plasma equilibria with a neural network technique , volume =

    van Milligen, B Ph and Jim. Solving threedimensional plasma equilibria with a neural network technique , volume =. Proc. Controlled Fusion and Plasma Physics , pages =

  10. [10]

    Improving ideal MHD equilibrium accuracy with physics-informed neural networks , volume =

    Thun, Timo and Merlo, Andrea and Conlin, Rory and Panici, Dario and B. Improving ideal MHD equilibrium accuracy with physics-informed neural networks , volume =. Nuclear Fusion , number =

  11. [11]

    Neural operators for accelerating scientific simulations and design , volume =

    Azizzadenesheli, Kamyar and Kovachki, Nikola and Li, Zongyi and Liu-Schiaffini, Miguel and Kossaifi, Jean and Anandkumar, Anima , journal =. Neural operators for accelerating scientific simulations and design , volume =

  12. [12]

    On the potential of physics-informed neural networks to solve inverse problems in tokamaks , volume =

    Rossi, Riccardo and Gelfusa, Michela and Murari, Andrea and JET contributors , journal =. On the potential of physics-informed neural networks to solve inverse problems in tokamaks , volume =

  13. [13]

    Neural operator surrogate models of plasma edge simulations: feasibility and data efficiency , volume =

    Carey, Naomi and Zanisi, Lorenzo and Pamela, Stanislas and Gopakumar, Vignesh and Omotani, John and Buchanan, James and Brandstetter, Johannes and Paischer, Fabian and Galletti, Gianluca and Setinek, Paul , journal =. Neural operator surrogate models of plasma edge simulations: feasibility and data efficiency , volume =

  14. [14]

    Force-free motion of a cold plasma during the current quench , volume =

    Kiramov, DI and Breizman, BN , journal =. Force-free motion of a cold plasma during the current quench , volume =

  15. [15]

    MHD, disruptions and control physics: Chapter 4 of the special issue: on the path to tokamak burning plasma operation , volume =

    Bandyopadhyay, I and Igochine, V and Sauter, O and Sabbagh, SA and Park, J-K and Nardon, E and Villone, F and Maraschek, M and Pautasso, G and Eidietis, N and others , journal =. MHD, disruptions and control physics: Chapter 4 of the special issue: on the path to tokamak burning plasma operation , volume =

  16. [16]

    Visualizing the loss landscape of neural nets , volume =

    Li, Hao and Xu, Zheng and Taylor, Gavin and Studer, Christoph and Goldstein, Tom , journal =. Visualizing the loss landscape of neural nets , volume =

  17. [17]

    A runaway electron avalanche surrogate for partially ionized plasmas , volume =

    Arnaud, Jonathan S and Tang, Xian-Zhu and McDevitt, Christopher J , journal =. A runaway electron avalanche surrogate for partially ionized plasmas , volume =

  18. [18]

    Neural-network-based approximations for solving partial differential equations , volume =

    Dissanayake, MWM Gamini and Phan-Thien, Nhan , journal =. Neural-network-based approximations for solving partial differential equations , volume =

  19. [19]

    Artificial neural networks for solving ordinary and partial differential equations , volume =

    Lagaris, Isaac E and Likas, Aristidis and Fotiadis, Dimitrios I , journal =. Artificial neural networks for solving ordinary and partial differential equations , volume =

  20. [20]

    Neural network differential equation and plasma equilibrium solver , volume =

    van Milligen, B Ph and Tribaldos, V and Jim. Neural network differential equation and plasma equilibrium solver , volume =. Physical review letters , number =

  21. [21]

    Physics-informed machine learning , volume =

    Karniadakis, George Em and Kevrekidis, Ioannis G and Lu, Lu and Perdikaris, Paris and Wang, Sifan and Yang, Liu , journal =. Physics-informed machine learning , volume =

  22. [22]

    Curvature-Aware Optimization for High-Accuracy Physics-Informed Neural Networks

    Jnini, Anas and Kiyani, Elham and Shukla, Khemraj and Urban, Jorge F and Daryakenari, Nazanin Ahmadi and Muller, Johannes and Zeinhofer, Marius and Karniadakis, George Em , date-added =. arXiv preprint arXiv:2604.05230 , title =

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    Machine learning applications to computational plasma physics and reduced-order plasma modeling: a perspective , volume =

    Faraji, Farbod and Reza, Maryam , date-added =. Machine learning applications to computational plasma physics and reduced-order plasma modeling: a perspective , volume =. Journal of Physics D: Applied Physics , number =

  24. [24]

    Pivotal issues on relativistic electrons in ITER , volume =

    Boozer, Allen H , journal =. Pivotal issues on relativistic electrons in ITER , volume =

  25. [25]

    Physics of runaway electrons in tokamaks , volume =

    Breizman, Boris N and Aleynikov, Pavel and Hollmann, Eric M and Lehnen, Michael , journal =. Physics of runaway electrons in tokamaks , volume =

  26. [26]

    An Efficient Surrogate Model of Secondary Electron Formation and Evolution , year =

    McDevitt. An Efficient Surrogate Model of Secondary Electron Formation and Evolution , year =

  27. [27]

    Interactions of relativistic electron beams with high atomic-number plasmas , volume =

    Mosher, David , journal =. Interactions of relativistic electron beams with high atomic-number plasmas , volume =

  28. [28]

    Runaway electron generation in axisymmetric tokamak geometry , volume =

    McDevitt, Christopher Joseph and Tang, X-Z , date-added =. Runaway electron generation in axisymmetric tokamak geometry , volume =. Europhysics Letters , number =

  29. [29]

    Kinetic modelling of runaway electron avalanches in tokamak plasmas , volume =

    Nilsson, Emelie and Decker, Joan and Peysson, Yves and Granetz, Robert S and Saint-Laurent, Francois and Vlainic, Milos , date-added =. Kinetic modelling of runaway electron avalanches in tokamak plasmas , volume =. Plasma Physics and Controlled Fusion , number =

  30. [30]

    Toroidal effect on runaway vortex and avalanche growth rate , volume =

    Guo, Zehua and Mcdevitt, Chris and Tang, Xianzhu , date-added =. Toroidal effect on runaway vortex and avalanche growth rate , volume =. Physics of Plasmas , number =

  31. [31]

    Theory of runaway electrons in ITER: Equations, important parameters, and implications for mitigation , volume =

    Boozer, Allen H , date-added =. Theory of runaway electrons in ITER: Equations, important parameters, and implications for mitigation , volume =. Physics of Plasmas , number =

  32. [32]

    Zur theorie des durchgangs schneller elektronen durch materie , volume =

    M. Zur theorie des durchgangs schneller elektronen durch materie , volume =. Annalen der Physik , number =

  33. [33]

    Runaway electrons and fast plasma shutdown , year =

    Rosenbluth, MN and Post, D and Putvinski, S and Parks, PB and Scott, HA and Putvinskaya, N , booktitle =. Runaway electrons and fast plasma shutdown , year =

  34. [34]

    Formation and termination of runaway beams in ITER disruptions , volume =

    Mart. Formation and termination of runaway beams in ITER disruptions , volume =. Nuclear Fusion , number =

  35. [35]

    Tackling the curse of dimensionality with physics-informed neural networks , volume =

    Hu, Zheyuan and Shukla, Khemraj and Karniadakis, George Em and Kawaguchi, Kenji , date-added =. Tackling the curse of dimensionality with physics-informed neural networks , volume =. Neural Networks , number =

  36. [36]

    Advances in Neural Information Processing Systems , title =

    Cho, Junwoo and Nam, Seungtae and Yang, Hyunmo and Yun, Seok-Bae and Hong, Youngjoon and Park, Eunbyung , date-added =. Advances in Neural Information Processing Systems , title =

  37. [37]

    Horovod: fast and easy distributed deep learning in TensorFlow

    Sergeev, Alexander and Del Balso, Mike , date-added =. arXiv preprint arXiv:1802.05799 , title =

    work page Pith review arXiv

  38. [38]

    The impact of collisionality on the runaway electron avalanche during a tokamak disruption , volume =

    Arnaud, Jonathan S and McDevitt, Christopher J , date-added =. The impact of collisionality on the runaway electron avalanche during a tokamak disruption , volume =. Physics of Plasmas , number =

  39. [39]

    A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks , volume =

    Wu, Chenxi and Zhu, Min and Tan, Qinyang and Kartha, Yadhu and Lu, Lu , date-added =. A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks , volume =. Computer Methods in Applied Mechanics and Engineering , pages =

  40. [40]

    \ TensorFlow \ : a system for \ Large-Scale \ machine learning , year =

    Abadi, Mart. \ TensorFlow \ : a system for \ Large-Scale \ machine learning , year =. 12th USENIX symposium on operating systems design and implementation (OSDI 16) , date-added =

  41. [41]

    DeepXDE: A deep learning library for solving differential equations , volume =

    Lu, Lu and Meng, Xuhui and Mao, Zhiping and Karniadakis, George Em , date-added =. DeepXDE: A deep learning library for solving differential equations , volume =. SIAM review , number =

  42. [42]

    A limited memory algorithm for bound constrained optimization , volume =

    Byrd, Richard H and Lu, Peihuang and Nocedal, Jorge and Zhu, Ciyou , date-added =. A limited memory algorithm for bound constrained optimization , volume =. SIAM Journal on scientific computing , number =

  43. [43]

    Adam: A Method for Stochastic Optimization

    Kingma, Diederik P , date-added =. arXiv preprint arXiv:1412.6980 , title =

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    Exact imposition of boundary conditions with distance functions in physics-informed deep neural networks , volume =

    Sukumar, Natarajan and Srivastava, Ankit , date-added =. Exact imposition of boundary conditions with distance functions in physics-informed deep neural networks , volume =. Computer Methods in Applied Mechanics and Engineering , pages =

  45. [45]

    A survey of regularization strategies for deep models , volume =

    Moradi, Reza and Berangi, Reza and Minaei, Behrouz , date-added =. A survey of regularization strategies for deep models , volume =. Artificial Intelligence Review , number =

  46. [46]

    Numerical characterization of bump formation in the runaway electron tail , volume =

    Decker, Joan and Hirvijoki, E and Embreus, O and Peysson, Y and Stahl, A and Pusztai, I and F. Numerical characterization of bump formation in the runaway electron tail , volume =. Plasma Physics and Controlled Fusion , number =

  47. [47]

    A physics-constrained deep learning treatment of runaway electron dynamics , volume =

    McDevitt, Christopher J and Arnaud, Jonathan S and Tang, Xian-Zhu , date-modified =. A physics-constrained deep learning treatment of runaway electron dynamics , volume =. Physics of Plasmas , number =

  48. [48]

    A Physics-Constrained Deep Learning Treatment of Runaway Electron Dynamics , year =

    McDevitt. A Physics-Constrained Deep Learning Treatment of Runaway Electron Dynamics , year =

  49. [49]

    A backward Monte-Carlo method for time-dependent runaway electron simulations , volume =

    Zhang, Guannan and del-Castillo-Negrete, Diego , date-added =. A backward Monte-Carlo method for time-dependent runaway electron simulations , volume =. Physics of Plasmas , number =

  50. [50]

    Unveiling the optimization process of physics informed neural networks: How accurate and competitive can PINNs be? , volume =

    Urb. Unveiling the optimization process of physics informed neural networks: How accurate and competitive can PINNs be? , volume =. Journal of Computational Physics , pages =

  51. [51]

    FreeGSNKE: A Python-based dynamic free-boundary toroidal plasma equilibrium solver , volume =

    Amorisco, Nicola C and Agnello, A and Holt, G and Mars, M and Buchanan, J and Pamela, S , date-modified =. FreeGSNKE: A Python-based dynamic free-boundary toroidal plasma equilibrium solver , volume =. Physics of Plasmas , number =

  52. [52]

    Adjoint method and runaway electron avalanche , volume =

    Liu, Chang and Brennan, Dylan P and Boozer, Allen H and Bhattacharjee, Amitava , date-added =. Adjoint method and runaway electron avalanche , volume =. Plasma Physics and Controlled Fusion , number =

  53. [53]

    Adjoint Fokker-Planck equation and runaway electron dynamics , volume =

    Liu, Chang and Brennan, Dylan P and Bhattacharjee, Amitava and Boozer, Allen H , date-added =. Adjoint Fokker-Planck equation and runaway electron dynamics , volume =. Physics of Plasmas , number =

  54. [54]

    On the first passage time probability problem , volume =

    Siegert, Arnold JF , date-added =. On the first passage time probability problem , volume =. Physical Review , number =

  55. [55]

    Current in wave-driven plasmas , volume =

    Karney, Charles FF and Fisch, Nathaniel J , date-added =. Current in wave-driven plasmas , volume =. The Physics of Fluids , number =

  56. [56]

    Relation of the runaway avalanche threshold to momentum space topology , volume =

    McDevitt, Christopher J and Guo, Zehua and Tang, Xian-Zhu , date-added =. Relation of the runaway avalanche threshold to momentum space topology , volume =. Plasma Physics and Controlled Fusion , number =

  57. [57]

    Phase-space dynamics of runaway electrons in magnetic fields , volume =

    Guo, Zehua and McDevitt, Christopher J and Tang, Xian-Zhu , date-added =. Phase-space dynamics of runaway electrons in magnetic fields , volume =. Plasma Physics and Controlled Fusion , number =

  58. [58]

    Evaluating derivatives: principles and techniques of algorithmic differentiation , year =

    Griewank, Andreas and Walther, Andrea , date-added =. Evaluating derivatives: principles and techniques of algorithmic differentiation , year =

  59. [59]

    Advances in neural information processing systems , title =

    Paszke, Adam and Gross, Sam and Massa, Francisco and Lerer, Adam and Bradbury, James and Chanan, Gregory and Killeen, Trevor and Lin, Zeming and Gimelshein, Natalia and Antiga, Luca and others , date-added =. Advances in neural information processing systems , title =

  60. [60]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , volume =

    Raissi, Maziar and Perdikaris, Paris and Karniadakis, George E , date-added =. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , volume =. Journal of Computational physics , pages =

  61. [61]

    A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics , volume =

    Haghighat, Ehsan and Raissi, Maziar and Moure, Adrian and Gomez, Hector and Juanes, Ruben , date-added =. A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics , volume =. Computer Methods in Applied Mechanics and Engineering , pages =

  62. [62]

    Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data , volume =

    Sun, Luning and Gao, Han and Pan, Shaowu and Wang, Jian-Xun , date-added =. Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data , volume =. Computer Methods in Applied Mechanics and Engineering , pages =

  63. [63]

    The constraint of plasma power balance on runaway avoidance , volume =

    McDevitt, Christopher J and Tang, Xian-Zhu and Fontes, Christopher J and Sharma, Prashant and Chung, Hyun-Kyung , date-added =. The constraint of plasma power balance on runaway avoidance , volume =. Nuclear Fusion , number =

  64. [64]

    Disruptions in ITER and strategies for their control and mitigation , volume =

    Lehnen, Michael and Aleynikova, K and Aleynikov, PB and Campbell, DJ and Drewelow, P and Eidietis, NW and Gasparyan, Yu and Granetz, RS and Gribov, Y and Hartmann, N and others , date-added =. Disruptions in ITER and strategies for their control and mitigation , volume =. Journal of Nuclear materials , number =

  65. [65]

    Avalanche mechanism for runaway electron amplification in a tokamak plasma , volume =

    McDevitt, Christopher J and Guo, Zehua and Tang, Xian-Zhu , date-added =. Avalanche mechanism for runaway electron amplification in a tokamak plasma , volume =. Plasma Physics and Controlled Fusion , number =

  66. [66]

    Influence of massive material injection on avalanche runaway generation during tokamak disruptions , volume =

    Hesslow, L and Embr. Influence of massive material injection on avalanche runaway generation during tokamak disruptions , volume =. Nuclear Fusion , number =

  67. [67]

    Effect of partially screened nuclei on fast-electron dynamics , volume =

    Hesslow, Linnea and Embr. Effect of partially screened nuclei on fast-electron dynamics , volume =. Physical review letters , number =

  68. [68]

    A physics-constrained deep learning surrogate model of the runaway electron avalanche growth rate , volume =

    Arnaud, Jonathan S and Mark, Tyler B and McDevitt, Christopher J , date-added =. A physics-constrained deep learning surrogate model of the runaway electron avalanche growth rate , volume =. Journal of Plasma Physics , number =

  69. [69]

    Theory of tokamak disruptions , volume =

    Boozer, Allen H , date-added =. Theory of tokamak disruptions , volume =. Physics of plasmas , number =

  70. [70]

    ''Multiplication''of accelerated electrons in a tokamak , volume =

    Sokolov, Yu A , date-added =. ''Multiplication''of accelerated electrons in a tokamak , volume =. JETP Lett.(USSR)(Engl. Transl.);(United States) , number =

  71. [71]

    Runaway electron generation in a cooling plasma , volume =

    Smith, H. Runaway electron generation in a cooling plasma , volume =. Physics of plasmas , number =

  72. [72]

    Electron kinetics in a cooling plasma , volume =

    Helander, P and Smith, H. Electron kinetics in a cooling plasma , volume =. Physics of Plasmas , number =

  73. [73]

    Runaway electron production in DIII-D killer pellet experiments, calculated with the CQL3D/KPRAD model , volume =

    Harvey, RW and Chan, VS and Chiu, SC and Evans, TE and Rosenbluth, MN and Whyte, DG , date-added =. Runaway electron production in DIII-D killer pellet experiments, calculated with the CQL3D/KPRAD model , volume =. Physics of Plasmas , number =

  74. [74]

    Fokker-Planck simulations mylb of knock-on electron runaway avalanche and bursts in tokamaks , volume =

    Chiu, SC and Rosenbluth, MN and Harvey, RW and Chan, VS , date-added =. Fokker-Planck simulations mylb of knock-on electron runaway avalanche and bursts in tokamaks , volume =. Nuclear Fusion , number =

  75. [75]

    Relativistic limitations on runaway electrons , volume =

    Connor, JW and Hastie, RJ , date-added =. Relativistic limitations on runaway electrons , volume =. Nuclear fusion , number =

  76. [76]

    Electron and ion runaway in a fully ionized gas

    Dreicer, Hz , date-added =. Electron and ion runaway in a fully ionized gas. I , volume =. Physical Review , number =

  77. [77]

    Melt damage to the JET ITER-like Wall and divertor , volume =

    Matthews, GF and Bazylev, B and Baron-Wiechec, A and Coenen, J and Heinola, K and Kiptily, V and Maier, H and Reux, C and Riccardo, V and Rimini, F and others , date-added =. Melt damage to the JET ITER-like Wall and divertor , volume =. Physica Scripta , number =

  78. [78]

    Theory for avalanche of runaway electrons in tokamaks , volume =

    Rosenbluth, MN and Putvinski, SV , date-added =. Theory for avalanche of runaway electrons in tokamaks , volume =. Nuclear fusion , number =

  79. [79]

    Collisional avalanche exponentiation of runaway electrons in electrified plasmas , volume =

    Jayakumar, R and Fleischmann, HH and Zweben, SJ , date-added =. Collisional avalanche exponentiation of runaway electrons in electrified plasmas , volume =. Physics Letters A , number =

  80. [80]

    Nuclear Fusion , title =

    Hender, TC and Wesley, JC and Bialek, J , date-added =. Nuclear Fusion , title =

Showing first 80 references.