Time-Embedded Convolutional Neural Networks for Modeling Plasma Heat Transport

Charles Heaton; Daniel Plummer; Francesco Miniati; Gianluca Gregori; Mila Fitzgerald; Mufei Luo; Sam M. Vinko; Yizhen Wang

Time-Embedded Convolutional Neural Networks for Modeling Plasma Heat Transport

T0 review · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The time-embedded convolutional network (TCNN) learns the coupled spatiotemporal evolution of normalized heat flux and the nonlocality parameter from fully kinetic PIC simulations, reproducing strongly nonlocal plasma heat transport across

desk verdict TCNN is a solid ML surrogate for nonlocal heat transport within its training family, but the paper never defines the per-case saturation time t_sa it uses to normalize time, which is a real generalization gap for anyone wanting to query the model on a new scenario; the ablation study is the best part. read the letter →

arxiv 2509.06088 v1 pith:DB5B3G27 submitted 2025-09-07 physics.plasm-ph

Mufei Luo , Charles Heaton , Yizhen Wang , Daniel Plummer , Mila Fitzgerald , Francesco Miniati , Sam M. Vinko , Gianluca Gregori This is my paper · ORCID

classification physics.plasm-ph PACS 52.25.Fi52.65.Rr07.05.Mh

keywords time-embeddedconvolutionalneuralnetworknonlocalheattransportplasmaparticle-in-cellsimulationsmachinelearningclosureLMVmodelfluxnonlocalityparameter

verification ladder T0 review T1 audit T2 compute T3 formal

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 introduces TCNN, a time-embedded convolutional neural network for modeling electron heat transport in plasmas, with emphasis on strongly nonlocal conditions. The authors are trying to establish that a single neural architecture can jointly predict the spatiotemporal evolution of both the normalized heat flux and the characteristic nonlocality parameter, rather than relying on the quasi-stationary assumptions of kernel-based models like LMV. Trained on fully kinetic particle-in-cell simulation data, TCNN is claimed to reproduce nonlocal dynamics, including flux suppression and preheating, across a broad range of collisionalities. If correct, it offers a fast, physically consistent surrogate for kinetic simulations in regimes where classical local theory and existing nonlocal closures break down. The paper also shows that time modulation, coupled prediction heads, and convolutional depth each contribute to the model's predictive performance.

What carries the argument

The machinery is a time-embedded convolutional stack. The input profile λ_free/L_T is lifted through a fully connected projection into a 64-dimensional latent space; a trainable time matrix supplies a different 64-dimensional vector for each of 44 normalized time steps, projected and added to the spatial embedding. Six 1D convolutional layers with GELU activations then aggregate information across the whole spatial domain, mirroring the LMV idea that heat flux at one point receives contributions from the entire domain. Two linear heads decode the shared representation into λ_free/L_T(x,t) and Q_e/Q_free(x,t). The work of this machinery is to keep the two output quantities coupled through sha

What would settle it

Take a temperature profile with a parameter combination absent from training, such as λ_free/L_T ≈ 0.02 with R=7 and L_T=100 or a non-isobaric density profile, run a fresh 1D PIC simulation to saturation, and compare TCNN's predicted Q_e/Q_free(x,t) and λ_free/L_T(x,t) to the PIC field; agreement within test-set error would support generalization, while large discrepancies would show the model only interpolates the training manifold.

Watch

Extended reading notes

Core claim

TCNN claims that the coupled spatiotemporal evolution of the nonlocality parameter λ_free/L_T and normalized heat flux Q_e/Q_free can be predicted together by one time-embedded convolutional network trained on fully kinetic PIC data. For 1D planar, isobaric, fully ionized Z=16 plasmas, it maps the initial λ_free/L_T profile to both quantities at 44 normalized times. This replaces the quasi-stationary LMV kernel assumption that made the earlier LINN model produce inconsistent kernels when temperature profiles evolve. The paper reports accurate nonlocal dynamics, including flux suppression and preheating, across λ_free/L_T ≈ 4×10^-4 to 0.1, with some degradation in the local, quasi-static regi

Load-bearing premise

The learned map is normalized in time by a per-case saturation time t_sa and is trained and tested on the same family of isobaric, 1D initial profiles, so a new scenario's t_sa and profile shape must be known or produced before the model can be applied.

Editorial extensions

If this is right

In the tested 1D planar, isobaric, Z=16 settings, TCNN reproduces nonlocal heat transport—including flux suppression and preheating—that LMV and Spitzer–Härm fail to capture, from a single initial profile.
Because the network is a forward pass rather than a PIC run, it offers a fast surrogate for the expensive kinetic computation within the covered parameter range.
The joint prediction heads keep λ_free/L_T and Q_e/Q_free mutually consistent, so the model avoids the physically inconsistent kernels produced by the earlier quasi-stationary LINN approach in strongly time-dependent cases.
The ablation study pins the performance gain to specific design choices: convolutional depth, latent-space projection, and time embedding all contribute, with the full combination best.
A hybrid TCNN+LINN scheme is suggested by the authors to recover accuracy in the local, quasi-static regime where TCNN's temporal signal is weak.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next test is to condition the network on predicted t_sa rather than the known PIC saturation time; without that, the architecture cannot be applied to a new simulation where saturation has not been measured.
The same time-embedding plus coupled-prediction recipe could be transferred to other nonlocal closures, such as radiation diffusion or multispecies fluxes, where a characteristic length and a flux evolve together.
The paper's restriction to λ_free/L_T ≤ 0.1, imposed because more collisionless runs never saturate within the simulation window, suggests the model is a surrogate for the saturable transport regime and will need a different formulation for truly collisionless heat flux.
If TCNN generalizes beyond the training profile family, it could be embedded as a learned closure in radiation–hydrodynamics codes; the authors state integration is not their immediate goal, so that step remains unvalidated.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick read, in order of importance: the paper does what it says inside its own test family. TCNN jointly predicts Q_e/Q_free and λ_free/L_T over time from the initial profile, trained on 1D PIC data, and the Fig. 7-8 matches to PIC look genuinely good, especially in the strongly nonlocal regime where their earlier LINN produced inconsistent kernels. The architecture is straightforward—FC embedding, six 1D convs, time embedding, two heads—but the ablation in Fig. 6 is the strongest part: it shows you need both depth and a latent-space projection, and that the full model outperforms simpler variants. That is real evidence, not just a claim. The loss construction with trend-consistency and differential weighting is sensible and not overcomplicated.

The soft spots are real but mostly about generalization and verification, not about the core method. First, the t_sa issue: the model conditions on t_i/t_sa ∈ {1/44,...,1}, and t_sa is defined only as "the saturation time specific to each case." The paper never says how t_sa is computed or estimated for a new initial condition. If you don't know t_sa in advance, you can't even query the model at a physical time. The paper restricts to λ_free/L_T ≤ 0.1 because beyond that no saturation is reached, which confirms t_sa is not universal. This doesn't invalidate the reproduction results, but it makes the abstract's "broad range of collisionalities" read as more usable than it is. The authors say integration into hydro codes is not their immediate goal, so this is a forward-looking gap rather than a flaw in the training/eval protocol—still, a referee should push them to define t_sa and describe how to obtain it without running PIC.

Second, the quantitative evidence is thinner than the text implies. Fig. 5(b) has no error bars, no multiple seeds, and there is no quantitative comparison to LINN or SNB on the same test cases. The paper acknowledges degraded accuracy in the local regime, but the abstract's "broad range" glosses over that. Third, no code or data are provided, and the dataset is the authors' own PIC runs, so independent verification is impossible.

Who benefits: anyone working on ML closures for nonlocal transport, or plasma surrogate models more broadly. The ablation alone is worth a look. It's a decent paper for a specialist venue, not a breakthrough. I'd send it to peer review with a request to clarify t_sa, add uncertainty quantification, and release the data/code. It's not desk-reject material.

Circularity Check

1 steps flagged · score 4.0 of 10

Time normalization by target-derived t_sa embeds information about the predicted heat flux into the conditioning variable.

self definitional [Section III, Time-Embedded Convolutional Neural Networks Architecture; Section II, Fig. 2 caption]
"t_sa denotes the saturation time specific to each case. ... It is conditioned on a discrete time index t_i/tsa ∈ {1/44,2/44,...,1}, corresponding to one of 44 time steps (see the time matrix within the green box, the elements of this time matrix are trainable)."

The model's time coordinate is t_i/t_sa, where t_sa is the saturation time of the heat flux—the very quantity (Q_e/Q_free) the network is trained to predict. The paper never defines t_sa independently of the target dynamics or supplies an estimator for it, and it explicitly excludes cases with no saturation. Thus the test accuracy reported in Figs. 5(b), 7, and 8 is obtained by conditioning on a time axis that is constructed from the ground-truth saturation behavior. In a forward application the user would need to know t_sa in advance (e.g., from a PIC run) to select the correct time embedding, so the central claim of reproducing spatiotemporal dynamics is not a self-contained prediction; it is a post-hoc interpolation in a target-defined coordinate.

full rationale

The paper is otherwise a standard supervised regression on external PIC data: the model is trained and tested on held-out OSIRIS simulations from the same family of initial profiles, and the hand-tuned loss weights in Eq. (7) are ordinary hyperparameters rather than fitted predictions. There is no load-bearing self-citation chain: the previous LINN work is cited for dataset generation and motivation, but the TCNN architecture and its convolutional/time-embedding machinery are evaluated on their own. The main circularity is the t_sa normalization: the conditioning time index is defined via the saturation time of the target heat flux, so the reported 'accurate reproduction' implicitly assumes knowledge of the very quantity being predicted. This is a partial, definitional circularity rather than a full collapse of the derivation, because the spatial profiles at each normalized time are still learned nontrivially. Score 4 reflects that the central claim of a usable spatiotemporal surrogate is compromised, while the machine-learning content retains independent empirical content on the training family.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical entities, forces, or particles. Its free parameters are the neural network weights and hand-tuned loss hyperparameters. The load-bearing domain assumptions are the fidelity of the PIC data, the representativeness of the 1D isobaric profile family, and the availability of a per-case saturation time t_sa.

free parameters (5)

TCNN network weights = not listed (919,648 conv + 65,664 output parameters)
Learned from PIC data through the loss function in Eq. (7); the central result is a function of these fitted weights.
Loss weighting exponents p1 and p2 = p1=1.0 for λ_free/L_T, p1=0.75 for Q_e/Q_free, p2=0.8
Hand-chosen to balance local and nonlocal regime magnitudes; they directly shape the fitted model.
Loss combination coefficients = 0.5, 0.5, 0.05 (Eq. 7), trend terms weighted 1.0
Hand-chosen in Eq. (7); not derived from a physical principle.
Saturation time t_sa per case = case-dependent, not tabulated
Normalizes the time axis in training and deployment; required as an input but the paper does not explain how it is obtained a priori.
Time embedding matrix = dimensions (64, 44) according to Fig. 4
Trainable embedding that replaces explicit time as an input feature; part of the fitted capacity.

assumptions (6)

domain assumption OSIRIS PIC with relativistic binary collision models [19,20] correctly captures electron heat transport for Z=16, 1D isobaric conditions.
The training data are assumed to be faithful kinetic ground truth; this is inherited from prior code validation, not demonstrated here.
domain assumption The isobaric density n_e ∝ 1/T_e and the temperature profile family of Eq. (1) are representative of conduction-zone plasmas.
The setup follows ICF conduction-zone practice [21], but the model is only validated on this family.
domain assumption Coarse-graining 38,400 PIC cells to 64 points preserves the heat flux and λ_free/L_T profiles.
Stated without a quantitative convergence check; the training and test data all pass through this reduction.
ad hoc to paper The LMV-inspired nonlocal convolution ansatz of Eq. (6) is a valid inductive bias for the network.
The heat flux is written as a convolution of λ_free/L_T with a learned kernel ã; this is a modeling choice, not a derivation.
ad hoc to paper The loss functions of Eq. (7) enforce physical consistency between flux and nonlocality.
The trend and differential weighting losses are heuristic regularizers, not derived from energy conservation or kinetic theory.
ad hoc to paper The saturation time t_sa is known for each case, and the time index t/t_sa ∈ {1/44, ..., 1} carries the dynamics.
Needed to construct the input time embedding; the paper does not specify how t_sa is computed or supplied in a new simulation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Time-Embedded Convolutional Neural Networks for Modeling Plasma Heat Transport." pith.science (2026). https://pith.science/paper/DB5B3G27

@misc{pith2026250906088,
  author       = {Pith},
  title        = {Pith review of: Time-Embedded Convolutional Neural Networks for Modeling Plasma Heat Transport},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DB5B3G27}},
  note         = {Machine review of arXiv:2509.06088}
}

read the original abstract

We introduce a time-embedded convolutional neural network (TCNN) for modeling spatiotemporal heat transport in plasmas, particularly under strongly nonlocal conditions. In our earlier work, the LMV-Informed Neural Network (LINN) (Luo et al., arXiv:2506.16619) combined prior knowledge from the LMV model with kinetic Particle-in-Cell (PIC) data to improve kernel-based heat-flux predictions. While effective under moderately nonlocal conditions, LINN produced physically inconsistent kernels in strongly time-dependent regimes due to its reliance on the quasi-stationary LMV formulation. To overcome this limitation, TCNN is designed to capture the coupled evolution of both the normalized heat flux and the characteristic nonlocality parameter using a unified neural architecture informed by underlying physical principles. Trained on fully kinetic PIC simulations, TCNN accurately reproduces nonlocal dynamics across a broad range of collisionalities. Our results demonstrate that the combination of time modulation, coupled prediction, and convolutional depth significantly enhances predictive performance, offering a data-driven yet physically consistent framework for multiscale plasma transport problems.

Figures

Figures reproduced from arXiv: 2509.06088 by the authors.

**Figure 2.** FIG. 2. Heat transport in the local and strongly nonlocal regimes, both initialized with a temperature ratio of [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** FIG. 3. Normalized heat flux [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: FIG. 5. Training and evaluation performance of the TCNN [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 7.** Figure 7: FIG. 7. Spatiotemporal prediction results by TCNN in the [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

**Figure 6.** Figure 6: FIG. 6. Performance comparison of four architectural vari [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 8.** Figure 8: FIG. 8. Spatiotemporal prediction results by TCNN in the [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 30 canonical work pages

[1]

Operating directly in real space by applying a single convolutional layer to convolve parameterλ free/LT across the domain

work page
[2]

Lifting the input vector (λ free/LT ) into a latent space, and applying a single convolutional layer to convolve the lifted vector across the latent domain

work page
[3]

Operating directly in real space, but applying mul- tiple convolutional layers to convolve parameter λfree/LT layer by layer across the domain

work page
[4]

Furthermore, in any of the above strategies, the temporal information must be introduced to capture the dynamics of heat transport

Lifting the input vector (λ free/LT ) into a latent space, before applying multiple convolutional lay- ers to convolve the lifted vector across the latent domain. Furthermore, in any of the above strategies, the temporal information must be introduced to capture the dynamics of heat transport.. The first strategy directly adheres to the formulation of Eq....

work page
[5]

Komarov, A

S. Komarov, A. A. Schekochihin, E. Churazov, and A. Spitkovsky, Self-inhibiting thermal conduction in a high-β, whistler-unstable plasma, J. Plasma Phys.84, 905840305 (2018)

work page 2018
[6]

Gregori, S

G. Gregori, S. H. Glenzer, J. Knight, C. Niemann, D. Price, D. H. Froula, M. J. Edwards, R. P. J. Town, A. Brantov, W. Rozmus, and V. Y. Bychenkov, Effect of nonlocal transport on heat-wave propagation, Phys. Rev. Lett.92, 205006 (2004)

2004
[7]

Arran, M

C. Arran, M. Holec, M. Sherlock, M. Marinak, and C. P. Ridgers, The accuracy of multi-group models for nonlocal electron transport in magnetized plasmas, Phys. Plasmas 32, 043908 (2025)

work page 2025
[8]

O. A. Hurricane, P. K. Patel, R. Betti, D. H. Froula, S. P. Regan, S. A. Slutz, M. R. Gomez, and M. A. Sweeney, Physics principles of inertial confinement fusion and u.s. program overview, Rev. Mod. Phys.95, 025005 (2023)

2023

Show all 40 references

[9]

E. L. Yerger, M. W. Kunz, A. F. Bott, and A. Spitkovsky, Collisionless conduction in a high-beta plasma: a colli- sion operator for whistler turbulence, J. Plasma Phys. 91, E20 (2025)

2025
[10]

G. P. Schurtz, P. D. Nicola ¨ ı, and M. Busquet, A nonlocal electron conduction model for multidimensional radiation hydrodynamics codes, Phys. Plasmas7, 4238 (2000)

2000
[11]

J. D. Huba,NRL Plasma Formulary(Naval Research Laboratory, Washington, DC, 2023) nRL/PU/6790–23- 528

2023
[12]

Spitzer and R

L. Spitzer and R. H¨ arm, Transport phenomena in a com- pletely ionized gas, Phys. Rev.89, 977 (1953)

1953
[13]

A. V. Brantov, V. Y. Bychenkov, V. T. Tikhonchuk, and W. Rozmus, Nonlocal electron transport in laser heated plasmas, Phys. Plasmas5, 2742 (1998)

1998
[14]

J. F. Luciani, P. Mora, and J. Virmont, Nonlocal heat transport due to steep temperature gradients, Phys. Rev. Lett.51, 1664 (1983)

1983
[15]

W. Q. Yuan, Z. H. Zhao, S. P. Zhu, X. T. He, and B. Qiao, Space-time dependent non-local thermal transport effects on laser ablation dynamics in inertial confinement fusion, Plasma Physics and Controlled Fusion66, 045023 (2024)

2024
[16]

E. M. Epperlein and R. W. Short, A practical nonlocal model for electron heat transport in laser plasmas, Phys. Fluids B3, 3092 (1991)

1991
[17]

O. V. Batishchev, V. Y. Bychenkov, F. Detering, W. Roz- mus, R. Sydora, C. E. Capjack, and V. N. Novikov, Heat transport and electron distribution function in laser pro- duced plasmas with hot spots, Phys. Plasmas9, 2302 (2002)

2002
[18]

Michel, R

O. Michel, R. Duclous, P.-E. Masson-Laborde, C. Enaux, and P. Lafitte, A nonlocal electron transport model in the diffusion scaling of hydrodynamics, Phys. Plasmas 30, 022712 (2023)

2023
[19]

Chrisment, P

A. Chrisment, P. Loiseau, J.-L. Feugeas, P.-E. Masson- Laborde, J. Mathiaud, V. Tikhonchuk, and P. Nicola ¨ ı, Analysis of a kinetic model for electron heat transport in inertial confinement fusion plasmas, Phys. Plasmas29, 062301 (2022)

2022
[20]

Sentoku and A

Y. Sentoku and A. Kemp, Numerical methods for parti- cle simulations at extreme densities and temperatures: Weighted particles, relativistic collisions and reduced currents, J. Comput. Phys.227, 6846 (2008)

2008
[21]

C. Lamy, B. Dubroca, P. Nicola ¨ ı, V. Tikhonchuk, and J.-L. Feugeas, Modeling of electron nonlocal transport in plasmas using artificial neural networks, Phys. Rev. E 105, 055201 (2022)

2022
[22]

M. Luo, C. Heaton, Y. Wang, D. Plummer, M. Fitzger- ald, F. Miniati, S. M. Vinko, and G. Gregori, Learning heat transport kernels using a nonlocal heat transport theory-informed neural network (2025), arXiv:2506.16619 [physics.plasm-ph]

2025 arXiv
[23]

R. A. Fonseca, L. O. Silva, F. S. Tsung, V. K. Decyk, W. Lu, C. Ren, W. B. Mori, S. Deng, S. Lee, T. Kat- souleas, and J. C. Adam, Osiris: A three-dimensional, fully relativistic particle in cell code for modeling plasma based accelerators, inComputational Science — ICCS 2002,...

2002
[24]

Sentoku, K

Y. Sentoku, K. Mima, Y. Kishimoto, and M. Honda, Ef- fects of relativistic binary collisions on pic simulation of laser plasmas, J. Phys. Soc. Jpn.67, 4084 (1998)

1998
[25]

H¨ uller, A

S. H¨ uller, A. Porzio, A. H´ eron, and P. Mora, Laser-driven ion acceleration in the presence of increasing heating of relativistic electrons at steep overdense plasma interfaces, Phys. Plasmas31, 103103 (2024)

2024
[26]

S. X. Hu, K. A. Nichols, N. R. Shaffer, B. Arnold, A. J. White, L. A. Collins, V. V. Karasiev, S. Zhang, V. N. Goncharov, R. C. Shah, D. I. Mihaylov, S. Jiang, and Y. Ping, A review on charged-particle transport modeling for laser direct-drive fusion, Phys. Plasmas31, 040501 (2024)

2024
[27]

A. R. Bell, R. G. Evans, and D. J. Nicholas, Elecron energy transport in steep temperature gradients in laser- 11 produced plasmas, Phys. Rev. Lett.46, 243 (1981)

1981
[28]

T. D. Arber, K. Bennett, C. S. Brady, A. Lawrence- Douglas, M. G. Ramsay, N. J. Sircombe, P. Gillies, R. G. Evans, H. Schmitz, A. R. Bell, and C. P. Ridgers, Contemporary particle-in-cell approach to laser-plasma modelling, Plasma Phys. Controlled Fusion57, 113001 (2015)

2015
[29]

Kovachki, Z

N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhat- tacharya, A. Stuart, and A. Anandkumar, Neural opera- tor: Learning maps between function spaces with appli- cations to pdes, J. Mach. Learn. Res.24, 1 (2023)

2023
[30]

E. P. Alves and F. Fiuza, Data-driven discovery of re- duced plasma physics models from fully kinetic simula- tions, Phys. Rev. Res.4, 033192 (2022)

2022
[31]

H¨ uller, A

S. H¨ uller, A. Porzio, J.-C. Adam, and A. H´ eron, On the non-thermal nature of distributions of electrons acceler- ated by high intensity lasers at the vacuum-plasma inter- face, Phys. Plasmas26, 083107 (2019)

2019
[32]

M. Luo, C. Riconda, I. Pusztai, A. Grassi, J. S. Wurtele, and T. F¨ ul¨ op, Control of autoresonant plasma beat-wave wakefield excitation, Phys. Rev. Res.6, 013338 (2024)

2024
[33]

M. Luo, C. Riconda, A. Grassi, N. Wang, J. Wurtele, T. F¨ ul¨ op, and I. Pusztai, Evolution of autoresonant plasma wave excitation in two-dimensional particle-in- cell simulations, J. Plasma Phys.91, E31 (2025)

2025
[34]

Miniati and G

F. Miniati and G. Gregori, Learning transport processes with machine intelligence, Sci. Rep.12, 11709 (2022)

2022
[35]

E. R. Ingelsten, M. C. McGrae-Menge, E. P. Alves, and I. Pusztai, Data-driven discovery of a heat flux closure for electrostatic plasma phenomena, J. Plasma Phys.91, E64 (2025)

2025
[36]

Huang, C

Z. Huang, C. Dong, and L. Wang, Machine-learning heat flux closure for multi-moment fluid modeling of nonlin- ear landau damping, Proc. Natl. Acad. Sci. USA122, e2419073122 (2025)

2025
[37]

D. D. Carvalho, D. R. Ferreira, and L. O. Silva, Learning the dynamics of a one-dimensional plasma model with graph neural networks, Machine Learning: Science and Technology5, 025048 (2024)

2024
[38]

C. J. McDevitt and X. Tang, A physics-informed deep learning description of knudsen layer reactivity reduc- tion, Phys. Plasmas31, 062701 (2024)

2024
[39]

D¨ opp, C

A. D¨ opp, C. Eberle, S. Howard, F. Irshad, J. Lin, and M. Streeter, Data-driven science and machine learning methods in laser–plasma physics, High Power Laser Sci. Eng11, e55 (2023)

2023
[192]

The electron mean free pathλ free is changed by adjusting the background electron densityn e0, thereby modifying the characteristic ratioλ free/LT . To examine the influence of hotter electrons on heat transport, we increase the temperature ratioR=T h/Tc while main- taining a ...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Operating directly in real space by applying a single convolutional layer to convolve parameterλ free/LT across the domain

work page

[2] [2]

Lifting the input vector (λ free/LT ) into a latent space, and applying a single convolutional layer to convolve the lifted vector across the latent domain

work page

[3] [3]

Operating directly in real space, but applying mul- tiple convolutional layers to convolve parameter λfree/LT layer by layer across the domain

work page

[4] [4]

Furthermore, in any of the above strategies, the temporal information must be introduced to capture the dynamics of heat transport

Lifting the input vector (λ free/LT ) into a latent space, before applying multiple convolutional lay- ers to convolve the lifted vector across the latent domain. Furthermore, in any of the above strategies, the temporal information must be introduced to capture the dynamics of heat transport.. The first strategy directly adheres to the formulation of Eq....

work page

[5] [5]

Komarov, A

S. Komarov, A. A. Schekochihin, E. Churazov, and A. Spitkovsky, Self-inhibiting thermal conduction in a high-β, whistler-unstable plasma, J. Plasma Phys.84, 905840305 (2018)

work page 2018

[6] [6]

Gregori, S

G. Gregori, S. H. Glenzer, J. Knight, C. Niemann, D. Price, D. H. Froula, M. J. Edwards, R. P. J. Town, A. Brantov, W. Rozmus, and V. Y. Bychenkov, Effect of nonlocal transport on heat-wave propagation, Phys. Rev. Lett.92, 205006 (2004)

2004

[7] [7]

Arran, M

C. Arran, M. Holec, M. Sherlock, M. Marinak, and C. P. Ridgers, The accuracy of multi-group models for nonlocal electron transport in magnetized plasmas, Phys. Plasmas 32, 043908 (2025)

work page 2025

[8] [8]

O. A. Hurricane, P. K. Patel, R. Betti, D. H. Froula, S. P. Regan, S. A. Slutz, M. R. Gomez, and M. A. Sweeney, Physics principles of inertial confinement fusion and u.s. program overview, Rev. Mod. Phys.95, 025005 (2023)

2023