Pith. sign in

REVIEW 6 major objections 5 minor 56 references

A single Fourier Neural Operator closure, trained online inside a differentiable fluid solver, reproduces both linear and nonlinear electrostatic Landau damping and generalises across initial perturbation amplitudes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

An online-trained Fourier Neural Operator with a memory window reproduces linear and nonlinear Landau damping in a 1D electrostatic fluid model and interpolates across initial amplitudes.

T0 review reviewed 2026-08-03 challenge →

load-bearing objection Honest, well-scoped demonstration of online-trained non-Markovian FNO closure for 1D Landau damping; the core result holds for the single-mode setup, but the abstract oversells generalization and the closure is explicitly solver-coupled. the 6 major comments →

arxiv 2607.29364 v1 pith:CO2ZVXJT submitted 2026-07-31 physics.plasm-ph

A Neural Operator Closure for Landau Damping in Electrostatic Plasma

Samuel Burles , Enrico Camporeale , Oreste Pezzi This is my paper
classification physics.plasm-ph
keywords Landau dampingFourier Neural Operatorplasma fluid closuremoment hierarchynon-Markovian closureMori-Zwanzigeffective heat fluxa posteriori training
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper tries to show that a machine-learned closure can replace the missing heat-flux term in a fluid plasma model and thereby recover the full kinetic physics of one-dimensional electrostatic Landau damping, in both its linear and nonlinear (particle-trapping) regimes. The approach trains a Fourier Neural Operator online—inside a differentiable fluid solver, with the loss evaluated on the closed-fluid trajectories against kinetic moment data—rather than fitting it to kinetic snapshots. The central claim is that one such FNO, acting on a trailing memory window of resolved moments, reproduces both damping regimes, interpolates and slightly extrapolates across initial perturbation amplitudes, and remains numerically stable in independent fluid simulations. This matters because analytic closures such as Hammett-Perkins capture only the linear regime, while offline-trained neural closures can be inaccurate or unstable once coupled to a solver. The paper also establishes that the learned heat flux is an effective flux, not a pointwise match to the kinetic heat flux, and that the model's use of memory is physically structured.

Core claim

The paper's central claim is that a single Fourier Neural Operator, trained a posteriori by backpropagating through a differentiable fluid solver, supplies the heat-flux-gradient closure that the one-dimensional electrostatic moment hierarchy needs to reproduce Landau damping. Deployed in the same fluid solver, the FNO recovers the linear damping rate to within about 0.2 percent in the training window and reproduces the nonlinear saturation and bounce oscillations of the trapped state over hundreds of plasma periods. One FNO, trained on amplitudes from 10^-3 to 10^-1 and conditioned on a log-amplitude channel, tracks the damping rate through the linear-to-nonlinear transition and interpolate

What carries the argument

The load-bearing object is the Fourier Neural Operator closure for the heat-flux gradient, ∂xq, entering the pressure equation of a 1D electrostatic fluid model. The FNO has eight Fourier layers with 32 retained modes and 64 hidden channels; its 251 input channels consist of five resolved fields {n, u, p, E, ∂xq} over a trailing window of K=50 timesteps (a non-Markovian memory horizon of 5 omega_p^-1), plus a spatially constant log-amplitude conditioning channel that lets one model modulate between linear and nonlinear regimes. Inputs are normalised by the RMS density perturbation amplitude and passed through an inverse-hyperbolic-sine transform to compress dynamic range. Online training is

Load-bearing premise

The learned closure is defined only through its effect on one specific discretised fluid solver—SSPRK3 time stepping, second-order central differences, a 128-point grid, and fixed wavenumber k=0.35—so if the effective heat flux does not transfer to other solvers, grids, or wavenumbers, the claim of a broadly applicable learned plasma closure collapses.

What would settle it

Deploy the trained FNO in the same physical setup but with a different time integrator or grid (e.g., classical RK4 or 256 spatial points) and compare the closed-fluid trajectory to the Vlasov ground truth; if it becomes unstable or drifts beyond the error level reported in the paper, the portability claim is falsified. A sharper test: train at k=0.35 and run at k=0.5 with the same amplitudes; failure there would show the closure has learned the specific mode, not the damping mechanism.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A learned fluid closure can replace analytic linear-theory closures like Hammett-Perkins for problems where both linear and nonlinear regimes matter, using a single model rather than a regime switch.
  • Online-trained closures are stable in deployment by construction; the paper's matched comparison suggests that a-priori-optimal closures can be stable or accurate in pointwise flux but still fail to reproduce resolved-moment trajectories.
  • The heat flux in a truncated fluid model should be judged by its effect on the resolved moments, not by pointwise agreement with the kinetic flux; the learned effective flux is a legitimate closure even where it diverges from the kinetic one.
  • The amplitude-conditioning channel and input normalisation allow one model to span several decades of perturbation amplitude, so training sets need not cover every amplitude, only the regimes.
  • Memory-window dependence is physically structured (peak sensitivity at lags separated by the oscillation period), suggesting that finite-memory neural closures can encode genuine kinetic memory rather than being arbitrary fits.

Where Pith is reading between the lines

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

  • Editorial extension: the fixed wavenumber k=0.35 is the true test boundary; the same recipe should be tried across wavenumbers, since the closure's memory of phase mixing is likely k-dependent, and the paper does not claim transfer across k.
  • Editorial extension: the sensitivity diagnostics suggest a testable refinement—if the memory window were shortened below the bounce period in the nonlinear regime, accuracy should degrade; scanning K would map the memory requirement quantitatively.
  • Editorial extension: because the effective flux is scheme-dependent, portability between fluid codes may require either retraining per discretisation or a parametrisation that conditions on discretisation metadata—an open problem the paper flags as future work.
  • Editorial extension: the same online-training-with-memory recipe may apply to other unresolved kinetic effects (e.g., pressure anisotropy in 2D/3D or electromagnetic fluctuations), where the Mori-Zwanzig memory integral is similarly non-trivial; the 1D electrostatic case is a proof-of-concept.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

6 major / 5 minor

Summary. The paper proposes a data-driven closure for the one-dimensional electrostatic fluid equations, replacing the heat-flux-gradient term in the pressure equation with a Fourier Neural Operator (FNO) trained online: the loss is computed on trajectories of a differentiable SSPRK3 solver closed by the FNO, rather than on pointwise fits to kinetic heat-flux data. The closure is non-Markovian, acting on a trailing K=50-step memory window of the resolved moments, with an amplitude-conditioning channel logA_scale. The authors train single-amplitude closures for linear (A=1e-3) and nonlinear (A=1e-1) Landau damping, and a multi-amplitude closure trained on A=1e-3 to 1e-1 at four points per decade. They report that the online-trained closures reproduce the kinetic electric-field energy evolution in both regimes, that the multi-amplitude closure interpolates across held-out amplitudes and extrapolates to amplitudes below the training range, and that the learned flux should be interpreted as an effective, scheme-dependent closure rather than the true kinetic heat flux. Comparison is made against an offline-trained FNO (which fails in the matched configuration) and against a calibrated Hammett-Perkins closure (which fails in the nonlinear regime). A sensitivity/ablation analysis is used to argue that the closure computes a genuine moment-to-flux relation with physically structured memory use.

Significance. If the central claims hold, the paper would be a useful demonstration that an online-trained, non-Markovian neural closure can reproduce both linear and nonlinear Landau damping with a single model, and that a posteriori training yields stable deployment in the same solver. The memory-window design, the amplitude-conditioning channel, and the explicit ablation/HSIC analysis are valuable contributions, and the authors are appropriately candid about the scheme dependence of the learned object. However, the paper's headline claims currently outrun the evidence: the generalization claim is only true below the training envelope, all reported errors come from single runs without seed variability, and the closure is only tested in the exact solver/grid/wavenumber used for training. The Hammett-Perkins benchmark is calibrated to the kinetic damping rate, so its linear-rate agreement is by construction. These issues are load-bearing for the abstract's wording about generalization and broad applicability, though the underlying method remains plausible and worthy of revision.

major comments (6)
  1. [§4.3, Fig. 5(a), Fig. 6] The abstract states the closure 'generalises to initial perturbation amplitudes outside the training set,' but the evidence shows this is only true for extrapolation below the training minimum. At A=0.2 and A=0.3, which lie above A_max=0.1, the damping rate departs sharply from the kinetic value and the WE(t) envelope is over-predicted (Fig. 5a and the last two panels of Fig. 6). The paper itself describes this as a 'failure to extrapolate above the training envelope.' The claim in the abstract and Section 1 should be qualified to specify interpolation and downward extrapolation, and the failure mode should be acknowledged in the abstract.
  2. [§3.2 and §4.4] The closure outputs ∂xq directly with no zero-mean constraint, so it can inject a spatially uniform source/sink into the pressure equation. On a periodic domain, no true heat-flux gradient can have a nonzero spatial mean, since ∫∂xq dx=0. The authors justify this as allowing net heating from Landau damping, but it means the learned object is not a physical closure in the standard sense; it is a solver-embedded effective term. This is consistent with the paper's own admission that the flux is 'not a unique or scheme-independent object' (§4.4) and that the FNO is 'tightly coupled to this particular numerical method' (§3.2). This combination undercuts the 'broadly applicable learned plasma closure' framing. The authors should either impose the zero-mean constraint and assess the consequences, or explicitly reframe the contribution as a solver-specific correction and provide a portability te
  3. [§4.1, Table 4, Table 6] The linear-regime claim of reproducing Landau damping is stated without the important test-window caveat. The fitted damping rate in the test window (t>75) has relative error 1.82e-1 (18%), and the electric-field error crosses 10% only 1.4 ω_p^{-1} after leaving the training window, reaching 50% at 45 ω_p^{-1} and 100% at 55 ω_p^{-1}. The paper does acknowledge this drift as 'expected,' but the abstract's unqualified 'reproduces both linear and nonlinear Landau damping' masks a substantial generalization gap in the linear case. The authors should either report the test-window degradation in the abstract/claims or define the reproduction claim only over the trained window and the nonlinear long-time behavior, where the performance is genuinely better.
  4. [§3.6, Table 4] The Hammett-Perkins benchmark coefficient is calibrated by matching the kinetic linear damping rate (χ=√(4.8/π)≈1.24, 22% below the canonical √(8/π)). Therefore, the statement that 'the calibrated closure reproduces the damping rate' (Table 4 caption) is true by construction and cannot serve as independent validation. The meaningful comparison is the oscillation frequency, where HP is shown to have a ~5% error. The paper should clearly label the HP linear-rate agreement as a calibration check, not a predictive success, and should avoid implying that HP validation strengthens the case for the FNO.
  5. [§4.1, §4.2, §4.3; Tables 4-6] All quantitative results are based on a single trained model for each configuration, with no repeated training seeds, no error bars, and no measure of training variability. Since the FNO training is stochastic (AdamW, random initialization, data sampling), the reported accuracy could be seed-dependent. The authors should either train multiple seeds and report mean/standard deviation for the key metrics (damping rates, errors, t_div), or at least state that all results are single-run and cannot be used to infer statistical robustness. This is particularly important for the claimed visual agreement in Figs. 1 and 3, which may not be representative.
  6. [§3.2, §5] The portability limitation is acknowledged in §3.2, but the paper still frames the contribution as a step toward closures 'suitable for large-scale fluid simulation.' All online tests use the same SSPRK3 integrator, second-order central differences, 128-point grid, and fixed k=0.35; no test is performed with a different spatial resolution, timestep, integrator, or wavenumber. As the reader's report notes, the learned effective flux is only defined through its action on this specific discretized dynamics. A minimal transfer test (e.g., varying Nx or Δt) would materially strengthen the portability claim. Without it, the conclusions should be restricted to the exact numerical configuration used.
minor comments (5)
  1. [Figure 3(b) caption] The panel title reads 'Linear (A=10^-1) (Offline Training)' but the case is the nonlinear amplitude A=10^-1. This is a typographical inconsistency with the surrounding text and should be corrected to 'Nonlinear'.
  2. [Figure 5(a) legend] The legend entry 'Landau' is ambiguous; it likely denotes the theoretical linear Landau rate. Please label it explicitly, e.g., 'linear Landau rate γ_Landau = ...'.
  3. [§3.2, Eq. (3.7)] The notation M_t is used for the window of K slots, and later M^(1), M^(2) for substage windows. The distinction between the accepted-state buffer and the provisional overwritten slot is described in prose but would be clearer if the buffer update rule were written as an explicit algorithm or pseudocode.
  4. [§4.5, Figure 10] The HSIC lag analysis is carefully caveated regarding autocorrelation, but the same caveat applies to Figure 12's correlation analysis where inputs are asinh-normalized; please state the normalization explicitly in the text or caption to avoid confusion.
  5. [§3.4, Table 3] The statement that 'the batch size is selected dynamically to make use of available GPU memory' and that 'the value of 50 corresponds to the 40GB A100' is helpful, but it would be useful to report the total number of training iterations per epoch (or total snapshot count) for reproducibility.

Circularity Check

1 steps flagged

Minor by-construction agreement in the calibrated Hammett-Perkins baseline; the FNO closure itself is an honest supervised-learning claim with genuine held-out-amplitude tests.

specific steps
  1. fitted input called prediction [§3.6 (Hammett-Perkins closure benchmark) and §4.1 / Table 4]
    "At our k = 0.35 the canonical coefficient sqrt(8/pi) ≈ 1.60 over-damps the mode; we therefore calibrate the coefficient to match the kinetic linear damping rate, obtaining chi = sqrt(4.8/pi) ≈ 1.24, 22% below the canonical value. ... With the calibrated coefficient the Hammett-Perkins fluid WE(t) tracks the kinetic decay envelope and rate across the full run, with a fitted damping rate gamma_HP = -3.52e-2 (Table 4), matching the kinetic rate by construction."

    The HP baseline's single free coefficient χ is fitted to exactly the quantity that Table 4 then reports as agreement: the kinetic linear damping rate at k = 0.35. The matching rate is therefore tautological, as the paper itself states ('matching the kinetic rate by construction'). This is a benchmark calibration, not a prediction, and it does not bear on the FNO's central claim; the FNO's in-sample agreement is likewise a fit, while the paper's independent content lies in the held-out amplitude tests (§4.3), which are genuine out-of-sample evaluations.

full rationale

The central claim is a machine-learning closure trained online against kinetic moment trajectories and tested against the same kind of ground truth; that is supervised learning, not circular derivation. The FNO is not constructed to output the kinetic heat flux or to match any fitted damping rate; it is optimized on trajectory loss and then evaluated on held-out data, including amplitudes below the training minimum (where it succeeds) and above it (where it fails). The paper's own limitations sections are explicit that the learned flux is solver-dependent and not unique, which is an honest scope restriction rather than a circular step. Self-citations (the Burles & Camporeale review, Pezzi et al. solver papers) are used for context and numerical method, not to justify a conclusion. The only by-construction agreement in the paper is the Hammett-Perkins baseline coefficient, which is calibrated to the kinetic damping rate and then reported as matching it 'by construction'; this is a minor, admitted artifact of a benchmark and does not affect the FNO result. Overall circularity is therefore low: score 2.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

The central claim rests on the assumption that the kinetic Vlasov-Poisson simulations provide reliable ground truth and that the specific discretized fluid solver used in training is a valid deployment environment. The FNO parameters are fitted to data; the only explicitly fitted physical coefficient is the calibrated Hammett-Perkins χ. The closure is an effective, solver-dependent quantity, not a derived physical law. No new physical entities are introduced.

free parameters (3)
  • FNO network weights = ~millions of parameters (8 Fourier layers, 64 hidden channels)
    Fitted to kinetic moment trajectories during online training; the effective closure is not unique and is expected to depend on the numerical scheme and training data (Sec 4.4).
  • Hammett-Perkins coefficient χ = sqrt(4.8/π) ≈ 1.24
    Calibrated to match the kinetic linear damping rate at k=0.35 (Sec 3.6). This is a benchmark parameter, not part of the FNO, but it is fitted to the target data.
  • Memory window length K = 50 steps (5 ω_p^{-1})
    Hand-chosen hyperparameter that defines the non-Markovian input; central to the closure's design (Sec 3.2, Table 2).
axioms (6)
  • domain assumption Vlasov-Poisson kinetic simulation provides a faithful ground truth for collisionless Landau damping.
    All training and testing uses an Eulerian Vlasov-Poisson solver as ground truth (Sec 3.1). If this solver is flawed, the learned closure learns the wrong target.
  • domain assumption The moment hierarchy can be truncated at pressure if a suitable effective heat flux is learned (Mori-Zwanzig projection).
    The closure framework assumes the unresolved higher moments can be represented by a history-dependent functional of resolved moments (Sec 2, 3.2).
  • domain assumption The FNO architecture (8 Fourier layers, 32 modes, 64 channels) is expressive enough for the closure.
    No comparison is made to alternative architectures under the same online procedure; the choice is motivated but unverified (Sec 5).
  • domain assumption The specific fluid discretization (SSPRK3, 2nd-order central differences) is representative for the closure's deployment.
    The closure is trained and tested on this single discretization; the paper acknowledges it is tightly coupled to the numerical method (Sec 3.2).
  • domain assumption Kinetic simulations are collisionless and noise-free; no collisional operators are used.
    The study focuses on collisionless dynamics; the Fokker-Planck operators in the Vlasov solver are disabled (Sec 3.1).
  • domain assumption The single-wavenumber perturbation (k=0.35) is representative of the linear-to-nonlinear transition.
    All results are at fixed k=0.35; no test at other wavenumbers, so the closure may not generalize across k (Sec 3.2).

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of A Neural Operator Closure for Landau Damping in Electrostatic Plasma." pith.science (2026). https://pith.science/paper/CO2ZVXJT

@misc{pith2026260729364,
  author       = {Pith},
  title        = {Pith review of: A Neural Operator Closure for Landau Damping in Electrostatic Plasma},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CO2ZVXJT}},
  note         = {Machine review of arXiv:2607.29364}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We present a data-driven plasma fluid closure for both linear and nonlinear electrostatic Landau damping in one dimension. A Fourier Neural Operator (FNO) is trained online within a differentiable fluid solver, with the loss computed on trajectories produced by the closed fluid simulation rather than on individual kinetic snapshots. The closure is non-Markovian, acting on a trailing window of the resolved moment history so as to represent the memory of the unresolved dynamics. We demonstrate that a single FNO trained in this way reproduces both linear and nonlinear Landau damping, generalises to initial perturbation amplitudes outside the training set, and remains numerically stable when deployed in independent fluid simulations. In the nonlinear regime the learned heat flux reproduces the resolved-moment dynamics without matching the kinetic heat flux pointwise, behaving as an effective closure that compensates for the truncated higher moments, though the learned specific flux is expected to depend on the numerical scheme and training data. A sensitivity analysis of the trained model shows that it computes a genuine moment-to-flux relation whose reliance on the memory window is physically structured.

Figures

Figures reproduced from arXiv: 2607.29364 by Enrico Camporeale, Oreste Pezzi, Samuel Burles.

Figure 1
Figure 1. Figure 1: Electric-field energy WE(t) for the linear case (A = 10−3 ): kinetic ground truth against the fluid simulation closed with (a) the online-trained FNO, (b) the offline-trained FNO and (c) the analytic Hammett-Perkins closure (χ = p 4.8/π). All panels share the same axes and initial conditions; shaded regions in (a) and (b) mark the FNO training window and do not apply to the untrained Hammett-Perkins closur… view at source ↗
Figure 2
Figure 2. Figure 2: Space-time evolution of the evolved moments for the linear single-amplitude case [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Electric-field energy WE(t) for the nonlinear case (A = 10−1 ): kinetic ground truth against the fluid simulation closed with (a) the online-trained FNO, (b) the offline￾trained FNO and (c) the analytic Hammett-Perkins closure (χ = p 4.8/π). All panels share the initial conditions; shaded regions in (a) and (b) mark the FNO training window and do not apply to the untrained Hammett-Perkins closure in (c). P… view at source ↗
Figure 4
Figure 4. Figure 4: Space-time evolution of the evolved moments for the nonlinear single-amplitude [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Damping rate of the initial linear phase of each amplitude run, comparing [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Electric-field energy WE(t) for every amplitude in the test grid, comparing the kinetic simulation (blue) with the online-tested generalised FNO closure (red); each panel is labelled by its initial amplitude and by whether that amplitude was included in training. All FNO curves are produced by online testing of the single trained closure from the same initial conditions as the corresponding kinetic run. Ag… view at source ↗
Figure 7
Figure 7. Figure 7: Saturation electric-field energy as a function of initial amplitude, comparing [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Increase of the total energy over time for the kinetic and fluid+FNO simulations [PITH_FULL_IMAGE:figures/full_fig_p024_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Relative L 2 error of each evolved moment’s trajectory between the kinetic and fluid+FNO simulations, as a function of initial perturbation amplitude, for the single generalised FNO closure. One panel is shown per field (E, n, u, p, ∂xq); filled markers denote amplitudes included in the training set and open markers the held-out test amplitudes. Errors increase monotonically with amplitude as the dynamics … view at source ↗
Figure 10
Figure 10. Figure 10: Strength of the closure’s dependence on its memory inputs, as a function [PITH_FULL_IMAGE:figures/full_fig_p027_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Test of whether the closure reads the plasma state or merely continues its own [PITH_FULL_IMAGE:figures/full_fig_p028_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: How closely the input fields track one another, as a function of initial amplitude [PITH_FULL_IMAGE:figures/full_fig_p029_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Dependence of the closure on groups of related inputs, measured as the median [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Which of density and pressure the closure actually uses. Curves show the [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗

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

56 extracted references · 25 canonical work pages · 1 internal anchor

  1. [1]

    Physics of Plasmas , keywords =

    Eulerian simulations of collisional effects on electrostatic plasma waves. Physics of Plasmas , keywords =. doi:10.1063/1.4821613 , archivePrefix =. 1307.0683 , primaryClass =

    Pith/arXiv arXiv

  2. [2]

    Physics of Plasmas , keywords =

    Phase space dynamics of unmagnetized plasmas: Collisionless and collisional regimes. Physics of Plasmas , keywords =. doi:10.1063/5.0160549 , archivePrefix =. 2306.03567 , primaryClass =

    Pith/arXiv arXiv

  3. [3]

    Journal of Plasma Physics , keywords =

    ViDA: a Vlasov-DArwin solver for plasma physics at electron scales. Journal of Plasma Physics , keywords =. doi:10.1017/S0022377819000631 , archivePrefix =. 1905.02953 , primaryClass =

    Pith/arXiv arXiv 1905

  4. [4]

    A Numerical Method for the Accurate Solution of the Fokker–Planck–Landau Equation in the Nonhomogeneous Case , journal =

    Francis Filbet and Lorenzo Pareschi , abstract =. A Numerical Method for the Accurate Solution of the Fokker–Planck–Landau Equation in the Nonhomogeneous Case , journal =. 2002 , issn =. doi:https://doi.org/10.1006/jcph.2002.7010 , url =

    arXiv 2002

  5. [5]

    The integration of the vlasov equation in configuration space , journal =

    C.Z Cheng and Georg Knorr , abstract =. The integration of the vlasov equation in configuration space , journal =. 1976 , issn =. doi:https://doi.org/10.1016/0021-9991(76)90053-X , url =

    work page doi:10.1016/0021-9991(76)90053-x 1976

  6. [6]

    Mangeney and F

    A. Mangeney and F. Califano and C. Cavazzoni and P. Travnicek , abstract =. A Numerical Scheme for the Integration of the Vlasov–Maxwell System of Equations , journal =. 2002 , issn =. doi:https://doi.org/10.1006/jcph.2002.7071 , url =

    arXiv 2002

  7. [7]

    Proceedings of the Royal Society of London

    The. Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences , author =. 1956 , pages =. doi:10.1098/rspa.1956.0116 , abstract =

    arXiv 1956

  8. [8]

    Physical Review Letters , author =

    Fluid moment models for. Physical Review Letters , author =. 1990 , pages =. doi:10.1103/PhysRevLett.64.3019 , language =

    doi:10.1103/physrevlett.64.3019 1990

  9. [9]

    Journal of Geophysical Research: Space Physics , author =

    Do. Journal of Geophysical Research: Space Physics , author =. 2022 , note =. doi:10.1029/2021JA030241 , abstract =

    doi:10.1029/2021ja030241 2022

  10. [10]

    Multiphysics

    Lautenbach, Simon and Grauer, Rainer , month = oct, year =. Multiphysics. Frontiers in Physics , publisher =. doi:10.3389/fphy.2018.00113 , abstract =

    arXiv 2018

  11. [11]

    Burles, Samuel and Camporeale, Enrico , month = apr, year =. The. doi:10.48550/arXiv.2511.22486 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2511.22486

  12. [12]

    Physics of Plasmas , author =

    Hybrid simulations of magnetic reconnection with kinetic ions and fluid electron pressure anisotropy , volume =. Physics of Plasmas , author =. 2016 , pages =. doi:10.1063/1.4943893 , abstract =

    work page doi:10.1063/1.4943893 2016

  13. [13]

    Physical Review Letters , author =

    Equations of. Physical Review Letters , author =. 2009 , pages =. doi:10.1103/PhysRevLett.102.085001 , language =

    work page doi:10.1103/physrevlett.102.085001 2009

  14. [14]

    and Egedal, J

    Ohia, O. and Egedal, J. and Lukin, V. S. and Daughton, W. and Le, A. , month = sep, year =. Demonstration of. Physical Review Letters , publisher =. doi:10.1103/PhysRevLett.109.115004 , abstract =

    work page doi:10.1103/physrevlett.109.115004

  15. [15]

    Physics of Plasmas , author =

    Landau fluid models of collisionless magnetohydrodynamics , volume =. Physics of Plasmas , author =. 1997 , pages =. doi:10.1063/1.872517 , abstract =

    work page doi:10.1063/1.872517 1997

  16. [16]

    Physics of Plasmas , author =

    An improved ten-moment closure for reconnection and instabilities , volume =. Physics of Plasmas , author =. 2020 , pages =. doi:10.1063/5.0012067 , abstract =

    work page doi:10.1063/5.0012067 2020

  17. [17]

    Physics of Plasmas , author =

    Comparison of multi-fluid moment models with particle-in-cell simulations of collisionless magnetic reconnection , volume =. Physics of Plasmas , author =. 2015 , pages =. doi:10.1063/1.4906063 , abstract =

    work page doi:10.1063/1.4906063 2015

  18. [18]

    Physics of Plasmas , author =

    A. Physics of Plasmas , author =. 2005 , pages =. doi:10.1063/1.2096582 , abstract =

    work page doi:10.1063/1.2096582 2005

  19. [19]

    Journal of Geophysical Research: Space Physics , author =

    A fluid model with finite. Journal of Geophysical Research: Space Physics , author =. 2006 , note =. doi:10.1029/2005JA011425 , abstract =

    work page doi:10.1029/2005ja011425 2006

  20. [20]

    Journal of Plasma Physics , author =

    Landau fluid closures with nonlinear large-scale finite. Journal of Plasma Physics , author =. 2015 , pages =. doi:10.1017/S0022377814000671 , abstract =

    work page doi:10.1017/s0022377814000671 2015

  21. [21]

    Physics of Plasmas , author =

    Machine learning surrogate models for. Physics of Plasmas , author =. 2020 , pages =. doi:10.1063/1.5129158 , abstract =

    work page doi:10.1063/1.5129158 2020

  22. [22]

    Physics of Plasmas , author =

    Neural network representability of fully ionized plasma fluid model closures , volume =. Physics of Plasmas , author =. 2020 , pages =. doi:10.1063/5.0006457 , abstract =

    work page doi:10.1063/5.0006457 2020

  23. [23]

    Identification of high order closure terms from fully kinetic simulations using machine learning , url =

    Laperre, Brecht and Amaya, Jorge and Jamal, Sara and Lapenta, Giovanni , month = mar, year =. Identification of high order closure terms from fully kinetic simulations using machine learning , url =

  24. [24]

    Electron neural closure for turbulent magnetosheath simulations: energy channels , shorttitle =

    Miloshevich, George and Vranckx, Luka and Lopes, Felipe Nathan de Oliveira and Dazzi, Pietro and Arrò, Giuseppe and Lapenta, Giovanni , month = sep, year =. Electron neural closure for turbulent magnetosheath simulations: energy channels , shorttitle =. doi:10.48550/arXiv.2510.00282 , abstract =

    work page doi:10.48550/arxiv.2510.00282

  25. [25]

    Proceedings of the National Academy of Sciences , author =

    Machine-learning heat flux closure for multi-moment fluid modeling of nonlinear. Proceedings of the National Academy of Sciences , author =. 2025 , note =. doi:10.1073/pnas.2419073122 , abstract =

    doi:10.1073/pnas.2419073122 2025

  26. [26]

    Data-driven modeling of

    Qin, Yilan and Ma, Jiayu and Jiang, Mingle and Dong, Chuanfei and Fu, Haiyang and Wang, Liang and Cheng, Wenjie and Jin, Yaqiu , month = aug, year =. Data-driven modeling of. Physical Review Research , publisher =. doi:10.1103/PhysRevResearch.5.033079 , abstract =

    work page doi:10.1103/physrevresearch.5.033079

  27. [27]

    Data-driven

    Liu, Kexin and Cheng, Wenjie and Fu, Haiyang and Jiang, Mingle and Ma, Jiayu and Qin, Yilan , month = dec, year =. Data-driven. 2022. doi:10.1109/ACES-China56081.2022.10065169 , abstract =

    arXiv 2022

  28. [28]

    Machine Learning: Science and Technology , author =

    Machine learning of hidden variables in multiscale fluid simulation , volume =. Machine Learning: Science and Technology , author =. 2023 , pages =. doi:10.1088/2632-2153/acf81a , abstract =

    doi:10.1088/2632-2153/acf81a 2023

  29. [29]

    Physical Review Research , author =

    Embedding physical symmetries into machine-learned reduced plasma physics models via data augmentation , volume =. Physical Review Research , author =. 2026 , note =. doi:10.1103/9fq1-vpqz , abstract =

    work page doi:10.1103/9fq1-vpqz 2026

  30. [30]

    Scientific Reports , author =

    Generalized neural closure models with interpretability , volume =. Scientific Reports , author =. 2023 , pmid =. doi:10.1038/s41598-023-35319-w , abstract =

    work page doi:10.1038/s41598-023-35319-w 2023

  31. [31]

    and Biferale, L

    Li, T. and Biferale, L. and Bonaccorso, F. and Scarpolini, M. A. and Buzzicotti, M. , month = apr, year =. Synthetic. Nature Machine Intelligence , publisher =. doi:10.1038/s42256-024-00810-0 , abstract =

    doi:10.1038/s42256-024-00810-0

  32. [32]

    and Noack, Bernd R

    Brunton, Steven L. and Noack, Bernd R. and Koumoutsakos, Petros , month = jan, year =. Machine. Annual Review of Fluid Mechanics , publisher =. doi:10.1146/annurev-fluid-010719-060214 , abstract =

    doi:10.1146/annurev-fluid-010719-060214

  33. [33]

    Plasma Physics and Controlled Fusion , author =

    Plasma. Plasma Physics and Controlled Fusion , author =. 2003 , pages =. doi:10.1088/0741-3335/45/6/701 , abstract =

    work page doi:10.1088/0741-3335/45/6/701 2003

  34. [34]

    Physics of Plasmas , author =

    Collisional effects on the numerical recurrence in. Physics of Plasmas , author =. 2016 , pages =. doi:10.1063/1.4940963 , abstract =

    work page doi:10.1063/1.4940963 2016

  35. [35]

    Wei, Shichen and Liu, Yuhong and Fu, Haiyang and Dong, Chuanfei and Wang, Liang , month = aug, year =. Data-. 2023. doi:10.23919/ACES-China60289.2023.10249492 , abstract =

    arXiv 2023

  36. [36]

    Li, Zongyi and Kovachki, Nikola and Azizzadenesheli, Kamyar and Liu, Burigede and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , month = may, year =. Fourier. doi:10.48550/arXiv.2010.08895 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2010.08895 2010

  37. [37]

    The Physics of Fluids , author =

    Collisionless. The Physics of Fluids , author =. 1965 , pages =. doi:10.1063/1.1761193 , abstract =

    doi:10.1063/1.1761193 1965

  38. [38]

    and Trivelpiece, Alvin W

    Krall, Nicholas A. and Trivelpiece, Alvin W. , year =. Principles of

  39. [39]

    Journal of Advances in Modeling Earth Systems , author =

    A Posteriori Learning for Quasi-Geostrophic Turbulence Parametrization , volume =. Journal of Advances in Modeling Earth Systems , author =. 2022 , pages =. doi:10.1029/2022MS003124 , number =

    doi:10.1029/2022ms003124 2022

  40. [40]

    Physical Review Fluids , author =

    Embedded training of neural-network subgrid-scale turbulence models , volume =. Physical Review Fluids , author =. 2021 , pages =. doi:10.1103/PhysRevFluids.6.050502 , number =

    work page doi:10.1103/physrevfluids.6.050502 2021

  41. [41]

    Differentiable Turbulence: Closure as a partial differential equation constrained optimization , url =

    Shankar, Varun and Chakraborty, Dibyajyoti and Viswanathan, Venkatasubramanian and Maulik, Romit , year =. Differentiable Turbulence: Closure as a partial differential equation constrained optimization , url =. doi:10.48550/arXiv.2307.03683 , publisher =

    Pith/arXiv arXiv doi:10.48550/arxiv.2307.03683

  42. [42]

    Foundations of Data Science , author =

    Scientific machine learning for closure models in multiscale problems: a review , volume =. Foundations of Data Science , author =. 2024 , pages =. doi:10.3934/fods.2024043 , number =

    doi:10.3934/fods.2024043 2024

  43. [43]

    Proceedings of the Royal Society A , author =

    A priori estimation of memory effects in reduced-order models of nonlinear systems using the. Proceedings of the Royal Society A , author =. 2017 , pages =. doi:10.1098/rspa.2017.0385 , number =

    arXiv 2017

  44. [44]

    Physical Review Fluids , author =

    Non-. Physical Review Fluids , author =. 2017 , pages =. doi:10.1103/PhysRevFluids.2.014604 , number =

    doi:10.1103/physrevfluids.2.014604 2017

  45. [45]

    Recurrent neural network closure of parametric

    Wang, Qian and Ripamonti, Nicol\`. Recurrent neural network closure of parametric. 2020 , pages =. doi:10.1016/j.jcp.2020.109402 , journal =

    arXiv 2020

  46. [46]

    Nature Machine Intelligence , author =

    Learning nonlinear operators via. Nature Machine Intelligence , author =. 2021 , pages =. doi:10.1038/s42256-021-00302-5 , number =

    doi:10.1038/s42256-021-00302-5 2021

  47. [47]

    Measuring Statistical Dependence with

    Gretton, Arthur and Bousquet, Olivier and Smola, Alex and Sch\". Measuring Statistical Dependence with. 2005 , pages =. doi:10.1007/11564089_7 , booktitle =

    doi:10.1007/11564089_7 2005

  48. [48]

    SIAM/ASA Journal on Uncertainty Quantification , author =

    Shapley Effects for Global Sensitivity Analysis: Theory and Computation , volume =. SIAM/ASA Journal on Uncertainty Quantification , author =. 2016 , pages =. doi:10.1137/15M1048070 , number =

    work page doi:10.1137/15m1048070 2016

  49. [49]

    and Lee, Su-In , year =

    Lundberg, Scott M. and Lee, Su-In , year =. A Unified Approach to Interpreting Model Predictions , booktitle =

  50. [50]

    SIAM/ASA Journal on Uncertainty Quantification , author =

    On Shapley Value for Measuring Importance of Dependent Inputs , volume =. SIAM/ASA Journal on Uncertainty Quantification , author =. 2017 , pages =. doi:10.1137/16M1097717 , number =

    doi:10.1137/16m1097717 2017

  51. [51]

    Journal of Plasma Physics , author =

    A learned closure method applied to phase mixing in a turbulent gradient-driven gyrokinetic system in simple geometry , volume =. Journal of Plasma Physics , author =. 2022 , keywords =. doi:10.1017/S0022377822000095 , abstract =

    work page doi:10.1017/s0022377822000095 2022

  52. [52]

    Advances in Neural Information Processing Systems , volume =

    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 Desmaison, Alban and K. Advances in Neural Information Processing Systems , volume =

  53. [53]

    arXiv preprint arXiv:2412.10354 , year =

    A Library for Learning Neural Operators , author =. arXiv preprint arXiv:2412.10354 , year =

    arXiv

  54. [54]

    Journal of Plasma Physics , author =

    An introductory guide to fluid models with anisotropic temperatures. Journal of Plasma Physics , author =. 2019 , keywords =. doi:10.1017/S0022377819000801 , abstract =

    work page doi:10.1017/s0022377819000801 2019

  55. [55]

    and Zank, G.P

    Hunana, P. and Zank, G.P. and Laurenza, M. and Tenerani, A. and Webb, G.M. and Goldstein, M.L. and Velli, M. and Adhikari, L. , month = sep, year =. New. Physical Review Letters , publisher =. doi:10.1103/PhysRevLett.121.135101 , abstract =

    work page doi:10.1103/physrevlett.121.135101

  56. [56]

    Journal of Plasma Physics , author =

    An introductory guide to fluid models with anisotropic temperatures. Journal of Plasma Physics , author =. 2019 , keywords =. doi:10.1017/S0022377819000850 , abstract =

    work page doi:10.1017/s0022377819000850 2019

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.